Application of Fuzzy Logic To Social Choice Theory

i i
K23798 2015/2/2
i i
MONOGRAPHS AND RESEARCH NOTES IN MATHEMATICS
Application of
Fuzzy Logic to
Social Choice
Theory
John N. Mordeson
Creighton University
Omaha, Nebraska, USA
Davender S. Malik
Terry D. Clark
i i
i i
i i
K23798 2015/2/2
i i
MONOGRAPHS AND RESEARCH NOTES IN MATHEMATICS
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
Forthcoming Titles (continued)

Practical Guide to Geometric Regulation for Distributed Parameter Systems,
Eugenio Aulisa and David S. Gilliam
Reconstructions from the Data of Integrals, Victor Palamodov
Special Integrals of Gradshetyn and Ryzhik: the Proofs Volume ll, Victor H. Moll
Stochastic Cauchy Problems in Infinite Dimensions: Generalized and Regularized
Solutions, Irina V. Melnikova and Alexei Filinkov
Symmetry and Quantum Mechanics, Scott Corry
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
No claim to original U.S. Government works

Version Date: 20141217
International Standard Book Number-13: 978-1-4822-5099-2 (eBook - PDF)
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
1 Fuzzy Maximal Subsets 1

1.1 Fuzzy Set Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Fuzzy Maximal Sets . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2 Fuzzy Choice Functions 21

2.1 Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2 Consistency Conditions . . . . . . . . . . . . . . . . . . . . . . 26
2.3 M-Rationality and G-Rationality . . . . . . . . . . . . . . . . . 37
2.4 Full Rationality of Fuzzy Choice Functions on Base Domain . . 46
2.5 Quasi-Transitive Rationality of Fuzzy Choice Functions . . . . 53
2.6 Full Rationality and Congruence Axioms of Fuzzy Choice
Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.8 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3 Factorization of Fuzzy Preference Relations 71

3.1 Basic Definitions and Results . . . . . . . . . . . . . . . . . . . 71
3.2 Quasi-Subtraction . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.3 Factorizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.4 Intuitionistic Fuzzy Relations . . . . . . . . . . . . . . . . . . . 93
3.5 Intuitionistic Fuzzy Preference Relations and Their
Factorization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4 Fuzzy Non-Arrow Results 115

4.1 Nondictatorial Fuzzy Aggregation Rules . . . . . . . . . . . . . 115
4.2 Auxiliary Functions . . . . . . . . . . . . . . . . . . . . . . . . 123
4.3 Arrows Theorem and Max-Star Transitivity . . . . . . . . . . . 126
i i
i i
i i
K23798 2015/2/2
i i
vi CONTENTS
4.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

4.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5 Fuzzy Arrows Theorem 137

5.1 Dictatorial Fuzzy Preference Aggregation Rules . . . . . . . . . 137
5.2 Representation Rules, Veto Players, Oligarchies,
and Collegiums . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
5.3 Decisive Sets, Filters, and Fuzzy Arrows Theorem . . . . . . . 153
5.4 Fuzzy Preferences and Social Choice . . . . . . . . . . . . . . . 156
5.5 Fuzzy Preferences and Arrow-Type Problems . . . . . . . . . . 162
5.6 The Structure of Fuzzy Preferences: Social Choice Implications 167
5.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
5.8 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
6 Single Peaked Fuzzy Preferences: Blacks Median Voter

Theorem 175
6.1 Fuzzy Aggregation Preference Rules . . . . . . . . . . . . . . . 176
6.2 Fuzzy Voting Rules . . . . . . . . . . . . . . . . . . . . . . . . . 180
6.3 Single-Peaked Fuzzy Profiles . . . . . . . . . . . . . . . . . . . . 186
6.4 The Core in Fuzzy Spatial Models . . . . . . . . . . . . . . . . 191
6.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
6.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
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
8 Arrow-Type Results under Intuitionistic Fuzzy Preferences 239

8.1 Fuzzy Preference Profiles and Fuzzy Aggregation Rules . . . . 239
8.2 IIA3 and Nondictatorial Fuzzy Aggregation Rules . . . . . . . . 246
8.3 IIA2, IIA4, and Fuzzy Arrows Theorem . . . . . . . . . . . . . 251
8.4 Representation Rules, Veto Players, and Oligarchies . . . . . . 258
8.5 Fuzzy Preference and Arrowian Results . . . . . . . . . . . . . 264
8.6 Intuitionistic Arrows Theorem and Gibbards Oligarchy
Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
8.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
8.8 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
9 Manipulability of Fuzzy Social Choice Functions 289

9.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
9.2 Fuzzy Social Choice Functions . . . . . . . . . . . . . . . . . . 292
9.3 Impossibility Results . . . . . . . . . . . . . . . . . . . . . . . 294
9.4 Non-Manipulable Partitioning . . . . . . . . . . . . . . . . . . . 297
i i
i i
i i
K23798 2015/2/2
i i
CONTENTS vii
9.5 Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300

9.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
9.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
10 Similarity of Fuzzy Choice Functions 305

10.1 Fuzzy Choice Functions . . . . . . . . . . . . . . . . . . . . . . 306
10.2 Similarity of Fuzzy Choice Functions . . . . . . . . . . . . . . . 309
10.3 Arrow Index of a Fuzzy Choice Function . . . . . . . . . . . . . 316
10.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
10.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
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
[24] introduced the concepts of positive transitivity and negative transitivity

of a strict component of a fuzzy relation. They allow for the establishment of
Arrowian results for crisp and fuzzy preferences.
In social decision-making contexts, a manipulator often has incentive to
misrepresent his/her preference. The GibbardSatterthwaite theorem states
that a social choice function over three or more alternatives that does not
provide incentive to individuals to misrepresent their sincere preferences must
be dictatorial. In Chapter 9, we extend their result to the case of fuzzy weak
preference relations. The results in this chapter are based on those of Ab-
delaziz, Figueria, and Meddeb [1] and also Duddy, Perote-Pena, and Piggins
[13].
In Chapter 10, we present Georgescus degree of similarity of two fuzzy
choice functions [19]. It induces a similarity relation on the set of all choice
functions defined on a collection of fuzzy choice functions. The degree of simi-
larity of two fuzzy choice functions allows for the evaluation of how rational a
fuzzy choice function is. Thus a ranking of a collection of fuzzy choice functions
with respect to their rationality can be obtained. We also consider Georgescus
Arrow index of a fuzzy choice function [22]. The Arrow index provides a de-
gree to which a fuzzy choice function satisfies the fuzzy Arrow axiom. We also
present a result that provides a degree to which a fuzzy choice function is fully
rational.
John Mordeson dedicates the book to his grandchildren, John, Michael,
Matthew, Marc, Jack, Jessica, David, Josh, Emily, Emma, Jenna, Elizabeth,
and Isabelle. Davender Malik dedicates the book to his wife Sadhana Malik
and daughter Shelly Malik. Terry Clark thanks John Mordeson, mathematics,
Creighton University and Mark Wierman, computer science, Creighton Uni-
versity for eight years of intellectually stimulating and amazingly productive
collaborative work.
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
Fuzzy Maximal Subsets
One paradigmatic change in science and mathematics this century concerns

the concept of uncertainty. It was once thought science should strive for cer-
tainty and that uncertainty was regarded as unscientific. An alternate view
now believes that uncertainty is essential to science. The publication of Lotfi
A. Zadehs seminal paper, [23], introducing the concept of fuzzy set theory
is an important work in the modern concept of uncertainty. Zadehs paper
challenges both Aristotelian two-valued logic and probability theory as being
the sole agent for uncertainty. Given a universal set X, a fuzzy subset of X
is a function of X into the closed interval [0, 1]. For an element x in X, the
statement that x is in is not necessarily true or false, but may be true to
some degree. This degree is given by (x).
It is stated by Zadeh in [23] that in the realm of soft sciences, sciences in
which imprecision, uncertainty, incompleteness of information and partiality of
truth lie at the center rather than the periphery, crisp-set-based mathematics
is not adequate as a modeling language. It is the inadequacy of crisp-set-
based mathematics as a modeling language that underlies the paucity of solid
theories in soft sciences. See also [4]. To overcome these inadequacies, we make
use of fuzzy set theory.
By Zadehs argument [23], we would expect to see considerable interest
in fuzzy logic in the social sciences. However, we do not. Why is that so?
One reason relates to the division in these disciplines between empirical and
theoretical research. The two are often disconnected, with theoretical findings
either being ignored by empiricists or those same conclusions having no rea-
sonable empirical referent. This is particularly evident in political science and
is the major impetus behind the National Science Foundations (NSF) initia-
tive to bring the two into more meaningful dialog [9]. Labeled the Empirical
Implications of Theoretical Models (EITM) program, the effort has arguably
yet to have made any notable impact in the discipline. Thus explicitly formal
approaches, to which fuzzy logic might contribute, have a minimal impact on
the empirical work that dominates the disciplines journals.
i i
i i
i i
K23798 2015/2/2
i i
2 1. Fuzzy Maximal Subsets
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.
1.1 Fuzzy Set Theory

We begin with a short review of some basic concepts from fuzzy set theory.
Let X be a nonempty set. If A and B are subsets of X, we let A\B
denote the set {x X | x A, x / B}. We use the symbol for the
union of two sets and for the intersection. A binary relation on X is a
subset R of the set of ordered pairs X X. Let R be a binary relation on
X. Then R is called reflexive if (x, x) R for all x X, transitive if for
all x, y, z X, (x, y), (y, z) R implies (x, z) R and complete if for all
x, y X, either (x, y) R or (y, x) R. R is called symmetric if for all
x, y X, (x, y) R implies (y, x) R. R is called antisymmetric if for all
distinct x, y X, (x, y) R implies (y, x) / R and irreflexive if (x, x)
/R
for all x X. We let P(X) denote the power set of X, i.e., the set of all
subsets of X. We let X = {(x, x) | x X}. If (x, y) R, we sometimes
i i
i i
i i
K23798 2015/2/2
i i
1.1. Fuzzy Set Theory 3
write xRy. Given R, we define a binary relation P on X in terms of R by

for all x, y X, (x, y) P if and only if (x, y) R, (y, x)
/ R. We call P the
strict binary relation associated with R. If f is a function from a set X
into a set Y and A is a subset of X, we use the notation f |A to denote the
restriction of f to A.
A fuzzy subset of X is a function of X into the closed interval [0, 1]. We
use the notation : X [0, 1] to denote a fuzzy subset of X. We let
FP(X) denote the fuzzy power set of X, i.e., the set of all fuzzy subsets
of X. If FP(X), we define the support of , written Supp(), to be
{x X | (x) > 0}. The cosupport of , written Cosupp(), is defined to
be the set {x X | (x) < 1}. Let P (X) = P(X)\{} and FP (X) =
{ F P(X) |Supp() 6= }. If S is a subset of X, we let 1S denote the
characteristic function of S in X. That is, 1S is the function of X into
{0, 1} such that 1S (x) = 1 if x S and 1S (x) = 0 otherwise. Let t [0, 1]
and x X. Define the fuzzy subset t{x} of X by t{x} (x) = t and t{x} (y) = 0
if y 6= x. Then t{x} is called a fuzzy singleton. Let , FP(X). Then we
write if (x) (x) x X and we write if and there exists
x X such that (x) < (x). Let t [0, 1]. We let t = {x X | (x) t}.
Then t is called the t-cut or the t-level set of . We let denote the fuzzy
subset of X that maps all elements of X to 0. We use the notation to
denote the minimum or infimum of a set of real numbers and to denote
the maximum or supremum of a set of real numbers. Let R(X) denote the
set of all relations on X. A fuzzy subset of X X is called a fuzzy binary
relation on X or simply a fuzzy relation on X. Let FR(X) denote the set
of all fuzzy relations on X. We let FR (X) = FP (X X).
We next define the notion of a t-norm, a concept central to our study. The
notion of a t-norm first appeared in the context of probabilistic metric spaces
Schweizer and Sklar [6]. They were then used as a natural interpretation of
the conjunction in semantics of mathematical fuzzy logics, Hajek [4].
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
The bounded difference t-norm is also known as the Lukasiewicz t-norm.

Let be a t-norm and let a, b [0, 1]. Then a and b are called zero
divisors with respect to if a 6= 0 6= b and a b = 0.
Let denote a t-norm on [0, 1]. Let and be fuzzy subsets of X. Define
the fuzzy subset of X by x X, ( )(x) = (x) (x). (Even though
the t-norm can vary, we use the notation since the context of its use will
be clear.)
Definition 1.1.3 Let F R (X). Define the fuzzy relation on X by

x, y X, (x, y) = (x, y) (y, x). Then is called the indifference oper-
ator with respect to .
A fuzzy binary relation is used to indicate the degree to which each al-
ternative is preferred to every other alternative. The indifference operator
indicates the degree to which a player is indifferent with respect to a pair of
alternatives.
Definition 1.1.4 Let FR (X) and let be a t-norm.

(1) If (x, x) = 1 for all x X, then is called reflexive.
(2) If (x, y) = (y, x) for all x, y X, then is called symmetric.
(3) If (x, y) > 0 implies (y, x) = 0 for all x, y X, then is called
asymmetric.
(4) If (x, y) (y, z) (x, z) for all x, y, z X, then is called max-
transitive.
(5) If (x, y) > 0 or (y, x) > 0 for all x, y X, then is called complete.
(6) If (x, y) + (y, x) 1 for all x, y X, then is called strongly
complete;
(7) If (x, y) (y, x) = 1 for all x, y X, then is called strongly
connected.
Completeness for a fuzzy relation says that for any two alternatives, the
degree of preference of at least one over the other must be positive. This may
seem implausible, but an entirely different theory would be developed without
the assumption of completeness.
A binary relation R on X is called acyclic if for all x1 , . . . , xn X, x1 P x2 ,
x2 P x3 , . . . , xn1 P xn implies x1 Rxn .
For FR (X), we often associate with an asymmetric fuzzy relation
.
Definition 1.1.5 Let FR (X) and let be a t-norm. Let be an asym-

metric fuzzy relation associated with .
(1) If (x1 , x2 )(x2 , x3 ). . .(xn1 , xn ) (x1 , xn ) for all x1 , . . . , xn
X, then is called acyclic (with respect to )
(2) If (x1 , x2 ) (x2 , x3 ) . . . (xn1 , xn ) > 0 implies (x1 , xn ) > 0 for
all x1 , . . . , xn X, then is called partially acyclic (with respect to ).
i i
i i
i i
K23798 2015/2/2
i i
1.2. Fuzzy Maximal Sets 5
Various types of asymmetric preference relations are used in fuzzy social

choice. In the following definition, we list some of the ones that are used the
most.
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.
Definition 1.1.7 Let F R(X) and let be an asymmetric fuzzy prefer-

ence relation associated with . Then
(1) is called simple if x, y X, (x, y) = (y, x) implies (x, y) =
(y, x);
(2) is called regular if x, y X, (x, y) > (y, x) if and only if
(x, y) > 0.
If a strict preference relation associated with is simple, then (x, y) =

(y, x) implies (x, y) = 0 = (y, x) for all x, y X. The strict preference
relations (0) , (1) , and (3) are simple. However, (2) in general is not. The
strict preference relations (1) and (3) are in fact regular.
1.2 Fuzzy Maximal Sets

Let X be a finite set and R be a relation on X. In the crisp case, a neces-
sary condition for the maximal set M (R, S) = {x S | y S, xRy} to be
nonempty for all subsets S of X is that R be reflexive and complete. Transitiv-
ity of R together with completeness and reflexivity are sufficient for M (R, S)
to be nonempty for all subsets S of X. However, transitivity is not necessary
for M (R, S) to be nonempty.
The maximal set comprises the subset of alternatives that are at least
weakly preferred to all others. If there is more than one alternative in the set,
then the player is indifferent between them. An important finding of social
choice theory is that any choice made by a collective will not necessarily be
contained in the collectives maximal set.
i i
i i
i i
K23798 2015/2/2
i i
Definition 1.2.1 Let C be a function from X X into [0, 1]. Define M :

FR (X) F P (X) FP (X) by (, ) FR (X) F P (X),
M (, )(x) = (x) (~{{t [0, 1] | C(x, w) t (x, w)} | w Supp()})
for all x X, where ~ is a t-norm. Then M (, ) is called a maximal fuzzy
subset with respect to C associated with (, ).
The point of the t-norm ~ different from is that in [5], ~ is and
consequently we want our results to correspond to those in [5], but yet have
the freedom of letting ~ = . If C(x, w) = (w) (w, x) x, w X and
~ = , then M is the maximal fuzzy subset defined in [5]. We denote this
maximal fuzzy subset by MG , but allow ~ to replace . Whenever we use MG ,
we assume ~ has no zero divisors. If C(x, w) = 1 x, w X and = = ~,
then M is the maximal fuzzy subset defined in [8]. We denote this maximal
fuzzy subset by MM .
The definition of a fuzzy maximal subset associated with (, ) gives the
degree to which the elements of X are maximal with respect to and . The
degree of maximality of an element x in X can neither be larger than the
degree that it is a member of nor larger than the smallest degree for which
x is preferred to those y which have positive membership in .
The above definition allows for alternative notions of a fuzzy maximal set
that are not available in crisp logic. The fuzzy subset provides a measure
of the degree to which each element in the set of alternatives is in the set
of most preferred alternatives by each individual. The fuzzy relation is a
measure of the degree to which some alternative is preferred to another. Since
is asymmetric, it provides the degree to which x is strictly preferred to y.
The fuzzy relation addresses the symmetrical relationship between x and y.
As the following example demonstrates, a political actor may prefer x to y at
some relatively high level, nevertheless the choice for y is not unacceptable.
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
In the next several results, we determine the relationship between fuzzy

maximal subsets and crisp maximal subsets.
Proposition 1.2.3 Let R be a relation on X and let S be a nonempty subset

of X. Then

1 if x S and w S, (x, w) R
M (1R , 1S )(x) = 0 if x
/ S or w X such that w S, C(w, x) 6= 0,
(x, w)
/ R.

Proof. We have that

1 if (x, w) R,
{t [0, 1] | C(x, w)t 1R (x, w)} = 1 if C(x, w) = 0,
0 if w S, C(w, x) 6= 0, (x, w)
/ R.

In Proposition 1.2.3, we have that if C(x, w) = 1S (x) (w, x), then

MG (, )(x) = 1 if x S and w S, [(x, w) R or w
/ S or (w, x)
/ R].
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.
Proposition 1.2.5 Suppose has no zero divisors. Let R be a relation on X

and let S be a nonempty subset of X. Suppose that x, w X, 1R (x, w) = 0
implies C(x, w) 6= 0. Then

1 if x S and (x, w) R w S,
M (1R , 1S )(x) =
0 otherwise.
Proof. Clearly, M (1R , 1S )(x) = 1 if x S and (x, w) R w S.
Suppose x / S or (x, w) / R for some w S. If x / S, then clearly
M (1R , 1S )(x) = 0. Suppose (x, w) / R for some w S. Then 1R (x, w) = 0
and so C(x, w) 6= 0. In this case, t = 0 is the only possible value such that
C(x, w) t = 0 since has no zero divisors. Thus M (1R , 1S )(x) = 0.
If R is complete and C(x, w) = 1S (x) 1R (w, x), then C(x, w) 6= 0 if
1R (x, w) = 0 and x S. By Proposition 1.2.5, MG (, ) contains the crisp
case if R is complete. We state this formally in the Corollary 1.2.6.
i i
i i
i i
K23798 2015/2/2
i i
Corollary 1.2.6 Let R be a relation on X and let S be a nonempty subset

of X. Suppose R is complete. Then

1 if x S and (x, w) R w S,
M (1R , 1S )(x) =
0 otherwise.
Proposition 1.2.7 Let (, ) FR (X) F P (X) and x X. Then
MM (, )(x) = (x) ({(x, w) | w Supp()}).
Proof. Since X is finite, we have that MM (, )(x) = {t [0, 1] | (x)

(x, w) t w Supp()} = {(x) (x, w) | w Supp()} = (x)
({(x, y) | w Supp()}).
Proposition 1.2.8 Let ~ = . Then (, ) FR (X)FP (X), MM (, )

MG (, ).
Proof. Let (, ) FR (X) FP (X) and x X. Let w0 X be such
that (x, w0 ) = {(x, w) | w Supp()}. Then MM (, )(x) = (x) w0 .
Now MG (, )(x) = (x) ({{t [0, 1] | (w) (w, x) t (x, w)} | w
Supp()})

(x) if (w) (w, x) (x, w) w Supp(),
=
(x) ({(x, w) | (w) (w, x) > (x, w), w Supp()} otherwise.
Let w1 = {(x, w) | (w) (w, x) > (x, w), w Supp()} if w Supp()

such that (w) (w, x) > (x, w). Since w0 w1 , the desired result holds.
Proposition 1.2.9 Let (, ) FR (X) F P (X) and x X. Then

(x) if C(x, w) (x, w) w Supp(),
M (, )(x) =
(x) tx otherwise,
where tx = {t [0, 1] | C(x, w) t (x, w), w Supp()} if w Supp()

such that C(x, w) > (x, w).
In Proposition 1.2.9, tx = {(x, w) | C(x, w) > (x, w), w Supp()} if

= and tx = {(x, w)/C(x, w) | C(x, w) > (x, w), w Supp()} if is
the product t-norm.
The concept of a maximal set is of considerable importance to spatial
models. If the maximal set is empty, then every alternative can be defeated
by at least one other alternative. When this situation occurs, spatial models
can not predict the outcome. The crisp characterization demonstrates that
a maximal set is assured when R is reflexive, complete, and acyclic [2]. The
next result gives a corresponding result for maximal fuzzy subsets.
i i
i i
i i
K23798 2015/2/2
i i
Theorem 1.2.10 Let = (0) . Suppose has no zero divisors. Let

FR (X) be reflexive and complete. Then M (, ) 6= for all FP (X) if
and only if is partially acyclic.
Proof. Suppose is partially acyclic. Let FP (X). Then Supp() 6= .
Let x1 Supp(). If (x1 , w) > 0 w Supp(), then by Proposition 1.2.9,
M (, )(x1 ) > 0 and so M (, ) 6= . Suppose there exists x2 Supp()\{x1 }
such that (x1 , x2 ) = 0. Then (x2 , x1 ) > 0 since is complete. Thus
(x2 , x1 ) > 0. Suppose there exists x1 , . . . , xk Supp() such that xi
Supp()\{x1 , . . . , xi1 } and (xi , xi1 ) > 0 for i = 2, . . . , k. If (xk , w) > 0
for all w Supp(), then M (, ) 6= as above. Suppose this is not the
case and there exists xk+1 Supp()\{x1 , . . . , xk } such that (xk , xk+1 ) = 0.
Then (xk+1 , xk ) > 0 since is complete. Hence by induction, either there
exists x Supp() such that (x, w) > 0 for all w Supp(), in which case
M (, ) 6= , or since Supp() is finite, Supp() = {x1 , . . . , xn } is such that
(xi , xi1 ) > 0 for i = 2, . . . , n. Since is partially acyclic and is reflexive,
(xn , xi ) > 0 for i = 1, . . . , n. That is, (xn , w) > 0 for all w Supp(). Thus
M (, )(xn ) > 0 and so M (, ) 6= .
Conversely, suppose M (, ) 6= for all F P (X). Suppose
x1 , . . . , xn X are such that (x1 , x2 ) > 0, (x2 , x3 ) > 0, . . . , (xn1 , xn ) >
0. We must show (x1 , xn ) > 0. Let S = {x1 , . . . , xn }. Then M (, 1S ) 6= .
Since (xi1 , xi ) > 0 and so (xi , xi1 ) = 0, i = 2, . . . , n, we have that
M (, 1S )(xi ) = 0, i = 2, . . . , n. Thus it must be the case that M (, 1S )(x1 ) 6= 0
and so (x1 , xi ) > 0 for i = 1, . . . , n. Thus (x1 , xn ) > 0 and so is partially
acyclic.
Definition 1.2.11 Let (, ) FR (X) F P (X). Define the fuzzy subset

nd of X by x X, nd(x) = 1 {(y, x) | y Supp()}, where is the
strict preference relation associated with . Then nd is called the degree of
nondominance of x with respect to (, ). Let N D(, ) = {x X | nd(x) =
1}.
Clearly x N D(, ) if and only if (y, x) = 0 y X.
Definition 1.2.12 Let (, ) F R (X) F P (X). Suppose is regular.

Define the fuzzy subset Md (, ) of X as follows: x X,

M (, )(x) if x N D(, ),
Md (, )(x) =
0 otherwise.
Definition 1.2.13 Let FR (X). If (x1 , x2 ) (x2 , x3 ) . . .

(xn1 , xn ) > 0 implies not (xn , x1 ) > 0 for all x1 , . . . , xn X, then
is called strictly acyclic.
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.
Theorem 1.2.14 Suppose has no zero divisors. Let FR (X) be reflex-

ive and complete. Suppose is regular. Then is strictly acyclic if and only
if Md (, ) 6= FP (X).
Proof. Suppose is strictly acyclic. Let FP (X). Then Supp() 6= .
Let x1 Supp(). If (x1 , w) (w, x1 ) w Supp(), then (x1 , w) > 0
w Supp(). Thus Md (, )(x1 ) > 0 and so Md (, ) 6= . Suppose there
exists x2 Supp()\{x1 } such that (x2 , x1 ) > (x1 , x2 ) Then (x2 , x1 ) > 0.
Suppose there exists x1 , . . . , xk Supp() such that xi Supp()\{x1 , . . . ,
xi1 } and (xi , xi1 ) > (xi1 , xi ) and so (xi , xi1 ) > 0 for i = 2, . . . , k. If
(xk , w) (w, xk ) for all w Supp(), then Md (, ) 6= as above. Suppose
this is not the case and there exists xk+1 Supp()\{x1 , . . . , xk } such that
(xk+1 , xk ) > (xk , xk+1 ). Then (xk+1 , xk ) > 0. Hence by induction, either
there exists x Supp() such that (x, w) (w, x) for all w Supp(), in
which case Md (, ) 6= , or since Supp() is finite, Supp() = {x1 , . . . , xn }
is such that (xi , xi1 ) > 0 for i = 2, . . . , n. Since is strictly acyclic and
is reflexive, not (xi , xn ) > 0 for i = 1, . . . , n Thus (xn , xi ) (xi , xn )
and so (xn , xi ) > 0 for i = 1, . . . , n. That is, (xn , w) (w, xn ) for all w
Supp(). Thus Md (, )(xn ) > 0 and so Md (, ) 6= .
Suppose Md (, ) 6= FP (X). Suppose x1 , . . . , xn X are such
that (x1 , x2 ) > 0, (x2 , x3 ) > 0, . . . , (xn1 , xn ) > 0. We must show not
(xn , x1 ) > 0. Let S = {x1 , . . . , xn }. Then Md (, 1S ) 6= . Since (xi1 , xi ) >
0, Md (, 1S )(xi ) = 0 for i = 2, . . . , n. Thus Md (, 1S )(x1 ) > 0. Hence (x1 , xi )
(xi , x1 ) for i = 1, . . . , n. Thus not (xn , x1 ) > 0. Hence is strictly acyclic.
The preceding result and the following corollary give corresponding results
to Theorem 1.2.10 for regular fuzzy preference relations.
Corollary 1.2.15 Suppose has no zero divisors. Let FR (X) be reflex-

ive and complete. Suppose is regular. If is strictly acyclic, then M (, ) 6=
FP (X).
i i
i i
i i
K23798 2015/2/2
i i
Example 1.2.16 For regular strict fuzzy preference relations, M (, ) 6=

FP (X) ; is strictly acyclic: Let X = {x, y, z}. Define FR (X)
by for all w X, (w, w) = 1 and
(x, y) = (y, z) = (x, z) = 1/2,

(y, x) = (z, y) = 1/4, (z, x) = 3/4.
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 (, ) = .
Proposition 1.2.17 M (, ) = Md (, ) (, ) F R (X) F P (X) if

and only if is of type (0) .
Proof. Suppose is of type (0) . Let (, ) FR (X)FP (X). Suppose
M (, )(x) > 0 for x Supp(). Then (x, w) > 0 w Supp(). If (w, x) >
(x, w) > 0 for some w Supp(), then since is of type (0) , (x, w) = 0 =
(w, x). Thus Md (, )(x) > 0. Hence Md (, )(x) = M (, )(x).
Conversely, suppose M (, ) = Md (, ) (, ) FR (X) FP (X).
Suppose is not of type (0) . Then there exists x, y X such that (y, x) > 0
and (x, y) > 0. Let S = {x, y}. Then Md (, 1S )(y) = 0, but M (, 1S )(y) > 0
since (y, x) > 0. Hence Md (, 1S ) 6= M (, 1S ).
We next consider strict preferences of type (2) .
Theorem 1.2.18 Suppose is of type (2) . Suppose has no zero divisors.

Let F R (X) be reflexive and complete. If is strictly acyclic, then
M (, ) 6= FP (X).
Proof. Suppose is partially acyclic. Let FP (X). Then Supp() 6= .
Let x1 Supp(). If (x1 , w) (w, x1 ) w Supp(), then (x1 , w) > 0
w Supp(). Thus M (, )(x1 ) > 0 and so M (, ) 6= . Suppose there
exists x2 Supp()\{x1 } such that (x2 , x1 ) > (x1 , x2 ) Then (x2 , x1 ) > 0.
Suppose there exists x1 , . . . , xk Supp() such that xi Supp()\{x1 , . . . ,
xi1 } and (xi , xi1 ) > (xi1 , xi ) and so (xi , xi1 ) > 0 for i = 2, . . . , k. If
(xk , w) (w, xk ) for all w Supp(), then M (, ) 6= as above. Suppose
this is not the case and there exists xk+1 Supp()\{x1 , . . . , xk } such that
(xk+1 , xk ) > (xk , xk+1 ). Then (xk+1 , xk ) > 0. Hence by induction, either
there exists x Supp() such that (x, w) (w, x) for all w Supp(), in
which case M (, ) 6= , or since Supp() is finite, Supp() = {x1 , . . . , xn }
is such that (xi , xi1 ) > 0 for i = 2, . . . , n. Since is strictly acyclic and
is reflexive, not (xi , xn ) > 0 for i = 1, . . . , n Thus (xn , xi ) = 1 for
i = 1, . . . , n. That is, (xn , w) > 0 for all w Supp(). Thus M (, )(xn ) > 0
and so M (, ) 6= .
i i
i i
i i
K23798 2015/2/2
i i
Example 1.2.19 Consider of Example 1.2.16. Then M (, ) 6=

FP (X). Clearly is not strictly acyclic when is of type (2) since (u, v) >
0 u, v X such that u 6= v.
Definition 1.2.20 Let (, ) F R (X) F P (X). Define the fuzzy subset

M2 (, ) of X by x X,

M (, )(x) if (x, w) = 1 w Supp(),
M2 (, )(x) =
0 otherwise.
Proposition 1.2.21 Suppose is of type (2) . Suppose has no zero divisors.

Let F R (X) be reflexive and complete. If M2 (, ) 6= F P (X),
then is strictly acyclic.
Proof. Suppose M2 (, ) 6= F P (X). Suppose x1 , . . . , xn X
are such that (x1 , x2 ) > 0, (x2 , x3 ) > 0, . . . , (xn1 , xn ) > 0. We must
show not (xn , x1 ) > 0. Let S = {x1 , . . . , xn }. Then M2 (, 1S ) 6= . Since
(xi1 , xi ) > 0, (xi , xi1 ) < 1 and so M2 (, 1S )(xi ) = 0 for i = 2, . . . , n.
Thus M2 (, 1S )(x1 ) > 0. Hence (x1 , xi ) = 1 for i = 1, . . . , n. Thus not
(xn , x1 ) > 0. Hence is strictly acyclic.
Suppose we change the definition of strictly acyclic when is of type (2)
to the following:
Let FR (X). If (x1 , x2 ) (x2 , x3 ) . . . (xn1 , xn ) > 0 implies

not (xn , x1 ) > (x1 , xn ) for all x1 , . . . , xn X.
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= .
Example 1.2.22 Consider of Example 1.2.19. Then M (, ) 6=

FP (X). Clearly is not acyclic when is of type (2) and with the altered
definition of acyclic given immediately above since (x, y) > 0, (y, z) > 0
and (z, x) > (x, z).
The following result gives a necessary and sufficient condition for an ele-
ment to belong to a level set of a fuzzy maximal subset.
Proposition 1.2.23 Let (, ) FR (X) F P (X). Let = = ~. Then

x X, t [0, 1], x M (, )t if and only if x t and w X, C(x, w)
t (x, w).
i i
i i
i i
K23798 2015/2/2
i i
Proof. We have that
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.
Proposition 1.2.24 Let S be a nonempty subset of X and let be a fuzzy

relation on X. Let x X and Sx = {w X | C(x, w) > (x, w)} and
tx,w = {t [0, 1] | C(x, w) t (x, w)}, where w Sx . Then x X,

1 if x S and Sx = ,
M (, 1S )(x) = ~{tx,w | w Sx } if x S and Sx 6= ,
0 if x / S.

Proof. Let x S. Then
M (, 1S (x)) = 1S (x) (~wX ({t [0, 1] | C(x, w) t (x, w)}))

= ~wX ({t [0, 1] | C(x, w) t (x, w)})
from which the desired result follows easily.

We next provide some examples involving fuzzy maximal subsets for vari-
ous t-norms.
Example 1.2.25 Let X = {x, y, z}. Define the fuzzy relation on X as

follows:
(x, x) = (y, y) = (z, z) = 1,

(x, y) = 3/4, (y, z) = 1/2, (x, z) = 1/4,
(y, x) = (z, y) = (z, x) = 0.
Let = = ~. Then MG (, 1X )(x) = 1, MG (, 1X )(y) = 0, MG (, 1X )(z) = 0

by Proposition 1.2.24. In this example, MG (, 1X ) = 1T , where T = {x}.
Example 1.2.26 Now let = ~ = product. Then
MG (, 1X )(x) = wX {{t [0, 1] | (w, x) t (x, w)}}

= 1 ({t [0, 1] | (y, x) t (x, y)})
({t [0, 1] | (z, x) t (x, z)})
= ({t [0, 1] | 0 t 3/4})
({t [0, 1] | 0 t 1/4})
= 1,
i i
i i
i i
K23798 2015/2/2
i i
MG (, 1X )(y) = ({t [0, 1] | (x, y) t (y, x)})

1 ({t [0, 1] | (z, y) t (y, z)})
= ({t [0, 1] | 3/4 t 0})
1 ({t [0, 1] | 0 t 1/2})
= 0,
MG (, 1X )(z) = ({t [0, 1] | (x, z) t (z, x)})
1 ({t [0, 1] | (y, z) t (z, y)})
= ({t [0, 1] | 1/4 t 0})
({t [0, 1] | 1/2 t 0}) 1
= 0.
Let and ~ denote the Lukasiewicaz t-norm. Then
MG (, 1X )(x) = 1,
MG (, 1X )(y) = ({t [0, 1] | 3/4 t 0})
1 ({t [0, 1] | 0 t 1/2})
= 1/4 1 1/2
= 0,
MG (, 1X )(z) = ({t [0, 1] | 1/4 t 0})
({t [0, 1] | 1/2 t 0}) 1
= 3/4 1/2 1
= 1/4.

follows:
(x, x) = (y, y) = (z, z) = 1,
(x, y) = 3/4, (y, z) = 1/2, (x, z) = 1/4,
(y, x) = 7/8, (z, y) = (z, x) = 0.
Let = = ~. Then MG (, 1X )(x) = 3/4, MG (, 1X )(y) =
1, MG (, 1X )(z) = 0. In this example, MG (, 1X ) 6= 1T for any subset T
of X.
Let and ~ denote product. Then
MG (, 1X )(x) = wX {{t [0, 1] | (w, x) t (x, w)}}
= 1 ({t [0, 1] | (y, x) t (x, y)})
({t [0, 1] | (z, x) t (x, z)})
= ({t [0, 1] | 7/8 t 3/4})
({t [0, 1] | 0 t 1/4})
= 6/7,
MG (, 1X )(y) = ({t [0, 1] | (x, y) t (y, x)}) 1
({t [0, 1] | (z, y) t (y, z)})
= ({t [0, 1] | 3/4 t 7/8}) 1
({t [0, 1] | 0 t 1/2})
= 1,
i i
i i
i i
K23798 2015/2/2
i i
MG (, 1X )(z) = ({t [0, 1] | (x, z) t (z, x)}) 1

({t [0, 1] | (y, z) t (z, y)})
= ({t [0, 1] | 1/4 t 0})
({t [0, 1] | 1/2 t 0}) 1
= 0.
Let and ~ denote the Lukasiewicaz t-norm. Then
MG (, 1X )(x) = ({t [0, 1] | 7/8 t 3/4})

({t [0, 1] | 0 t 1/4})
= 7/8 1
= 7/8,
MG (, 1X )(y) = ({t [0, 1] | 3/4 t 7/8}) 1

({t [0, 1] | 0 t 1/2})
= 111
= 1,
MG (, 1X )(z) = ({t [0, 1] | 1/4 t 0})
({t [0, 1] | 1/2 t 0}) 1
= 3/4 1/2 1
= 1/4.

follows:
(x, x) = (y, y) = (z, z) = 1,

(x, y) = 3/4, (y, z) = 1/2, (z, x) = 1/4,
(y, x) = (z, y) = (x, z) = 0.
For any t-norm without zero divisors, MG (, 1X )(x) = ({t [0, 1] | 0 t

(x, y)})~({t [0, 1] | 1/4t 0}) = 3/4~0 = 0. Similarly, MG (, 1X )(y) =
0, M (, 1X )(z) = 0. In this example, MG (, 1X ) = 1T , where T = . In this
example, we have a cycle.
We next consider when M (, 1S ) is a characteristic function even though
may not be. That is, we are interested in determining when the fuzzy maximal
subset is crisp given that the set S is crisp and when the preference relation
is fuzzy, i.e., when the decision maker states this degree of preference.
Proposition 1.2.29 Let FR (X) and let S be a subset of X. Suppose

is a t-norm without zero divisors. Then M (, 1S ) = 1T for some T S if
and only if x, y S, (x, y) C(y, x) > 0 implies (x, y) C(y, x).
Proof. Suppose M (, 1S ) = 1T for some T S. Let y S. w Sy , let

ty,w = {t [0, 1] | C(w, y) t (y, w)}. Then M (, 1S )(y) = ~{ty,w | w
Sy }, where Sy = {w X | C(w, y) > (y, w)} and if Sy 6= . Suppose Sy 6= .
i i
i i
i i
K23798 2015/2/2
i i
Then it follows that 0 < {(y, w) | w Sy } < 1. However, this contradicts

the hypothesis. Hence Sy = . Thus C(w, y) (y, w) w S, ty,w = 1
w S.
Conversely, suppose that x, y S, (x, y) C(y, x) > 0 implies (x, y)
C(y, x). Let T = {x S | @w S, C(w, x) > (x, w)}. Suppose x S\T. Then
w S such that C(w, x) > (x, w). By our hypothesis, (x, w) = 0. Thus
M (, 1S )(x) = 0. Hence M (, 1S ) = 1T .
We next consider the relationship between fuzzy maximal subsets and level
sets.
Proposition 1.2.30 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= , {x X | x t , (x, y) t y Supp()} =
6 .
Proof. By [22], {x X | x Supp(), (x, y) t y Supp()} =
6 .
The desired result follows since t Supp().
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= .
Proposition 1.2.32 Let (, ) FR (X) F P (X). Then x X, t

[0, 1], x MM (, )t if and only if x t and (x, y) t y Supp().
Proof. We have that x MM (, )t MM (, )(x) t {r [0, 1] |
(x) r and (x, y) r y Supp()} t r [0, 1] such that r t
and (x) r, (x, y) r y Supp() r [0, 1] such that r t and
x r , (x, y) r y Supp() x t and (x, y) t y Supp().
Corollary 1.2.33 Let (, ) FR (X) FP (X). If t (0, s ], where s =

{s Im() | s = Supp()}, then MM (, )t = MM (t ,Supp()).
Proof. MM (, )t = {x t | (x, y) t y Supp()} =
MM (t ,Supp()) since t = Supp().
Proposition 1.2.34 Let FR (X). Then S FP (X)T S such that

MM (, 1S ) = 1T if and only if @x, y X (not necessarily distinct) such that
0 < (x, x) (x, y) < 1.
Proof. Suppose there exists x, y X such that 0 < (x, x) (x, y) < 1.
Let S = {x, y}. Then MM (, 1S )(x) = (x, x) (x, y). Thus @T S such
that MM (, 1S ) = 1T .
Conversely, suppose @x, y X such that 0 < (x, x) (x, y) < 1. Let
S FP (X). Then x S, (x, x) (x, y) = 0 or (x, x) (x, y) = 1 for
i i
i i
i i
K23798 2015/2/2
i i
1.3. Exercises 17
all y S. Hence x S, {t [0, 1] | x (1S )t , (x, y) t y (1S )t } = 0 or

1, i.e., MM (, 1S )(x) = 0 or 1. Thus the desired conclusion holds.
Let X = {x, y}. Define : X X [0, 1] as follows: (x, x) = 0, (x, y) =
1/2, (y, y) = (y, x) = 1. Then M (, 1X )(x) = 0 and M (, 1X )(y) = 1.
Thus M (, 1X ) = 1T , where T = {y} even though 0 < (x, y) < 1. Also,
MM (, 1{y} ) = 1{y} and MM (, 1{x} ) = 1 . If we require in Proposition 1.2.34
that T 6= , then we get the following result.
Proposition 1.2.35 Let F R (X). Then S P (X) T S, T 6= ,

such that MM (, 1S ) = 1T if and only if @x, y X (not necessarily distinct)
such that 0 < (x, y) < 1.
Proof. The result follows from Proposition 1.2.34 and the fact that if T 6=
, then (x, x) > 0 x X else if (x, x) = 0 for some x X, MM (, 1{x} ) =
1T , where T = . Note also that the assumption @x, y X (not necessarily
distinct) such that 0 < (x, y) < 1 implies (x, x) > 0 x X.
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.
2. D. Austen-Smith and J. S. Banks, Positive Political Theory I: Collective

Preference, University of Michigan Press 2000.
3. C. A. Cioffi-Revilla, Fuzzy sets and models of international relations,
American Journal of Political Science, 25 (1981) 129159.
i i
i i
i i
K23798 2015/2/2
i i
4. T. D. Clark, J. M. Larson, J. N. Mordeson, J. D. Potter, M. J. Wierman,

Applying Fuzzy Mathematics to Formal Models in Comparative Politics,
Springer-Verlag Berlin Heidelberg, Studies in Fuzziness and Soft Com-
puting 225, 2008.
5. Georgescu, Fuzzy Choice Functions: A Revealed Preference Approach,
Studies in Fuzziness and Soft Computing, Springer-Verlag Berlin Hei-
delberg 2007.
6. P. Hajek, Metamathematics of Fuzzy Logic, Kluwer, Dordrecht, 1998.

7. M. Koenig-Archibugi, Explaining government preferences for institu-
tional change in EU foreign and security policy, International Organi-
zations, 58 (2004) 137174.
8. J. N. Mordeson, K. R. Bhutani, and T. D. Clark, The rationality of

fuzzy choice functions, New Mathematics and Natural Computation, 4
(2008) 309327.
9. R. B. Morton, Methods and models: A guide to the empirical analysis
of formal models in political science, Cambridge, Cambridge University
Press, 1999.
10. P. Pennings, Beyond dichotomous explanations: Explaining constitu-
tional control of the executive with fuzzy set-sets, European Journal of
Political Research, 42 (2003) 541567.
11. C. C. Ragin, Fuzzy-set social choice, Chicago, University of Chicago
Press, 2000.
12. G. S. Sanjian, Fuzzy set theory and US arms transfers: Modeling
the decision-making process, American Journal of Political Science, 32
(1988) 10181046.
13. G. S. Sanjian, Great power arms transfers: Modeling the decision-

making processes of hegemonic, industrial, and restrictive exporters,
International Studies Quarterly, 35 (1991) 173193.
14. G. S. Sanjian, A fuzzy set model of NATO decision-making: The case
of short-range nuclear forces in Europe, Journal of Peace Research, 29
(1992) 271285.
15. G. S. Sanjian, Cold War imperatives and quarrelsome clients: Modeling
US and USSR arms transfers to India and Pakistan, The Journal of
Conflict Resolution, 42 (1998) 97127.
16. G. S. Sanjian, Promoting stability or instability? Arms transfers and
regional rivalries, 1950-1991, International Studies Quarterly, 43 (1999)
641670.
i i
i i
i i
K23798 2015/2/2
i i
1.4. References 19
17. G. S. Sanjian, Arms and arguments: Modeling the effects of weapons

transfers on subsystem relationships, Political Research Quarterly, 54
(2001) 285309.
18. B Schweizer and A. Sklar, Probabilistic Metric Spaces, North-Holland,
New York, 1983.
19. S. T. Seitz, Apollos oracle: Strategizing for peace, Syntese, 100 (1994)
461495.
20. M. Smithson and J. Verkuilen, Fuzzy set theory: Applications in the

social sciences, Thousand Oaks, Sage Publications, 2006.
21. C. S. Taber, POLI: An expert system model of US foreign policy belief
systems, American Political Science Review, 86 (1992) 888904.
22. J. V. Verkuilen, Assigning membership in a fuzzy set analysis, Sociolog-

ical Methods and Research, 33 (2005) 462496.
23. L. A. Zadeh, Fuzzy sets, Information and Control, 8 (1965) 338353.
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
Fuzzy Choice Functions
Classical revealed preference theory postulates a connection between choices

and revealed preferences. Political scientists have employed this theory in a
wide variety of approaches and issues. For instance, comparativists predict
parliamentary vote outcomes based on the preferences for legislators revealed
in party manifestos. Social choice theorists initially expected on the basis of
revealed preference theory that collective choice was based on collective pref-
erences. In the previous chapter, we demonstrated that a maximal set may not
exist under all conditions. This conclusion has led scholars to question the re-
lationship between revealed social preferences and social choice. In particular,
they have asked whether the choices made by collective actors are consistent
with their collective social preference. In other words, if we assume that the
preferences of a set of individuals are not cyclic, we would like to know if
their collective choices are rationalizable. Formally, do their collective choices
comprise a proper subset of the maximal set when one exists? The question
is not a trivial one. For example, scholars often estimate the preferences of
legislators based on roll-call votes [39]. If we do not assume that legislators
roll-call votes are related to their preferences, then we must estimate their
preferences by some other means.
Considerable work has been done on the rationalizability of fuzzy social
preference and fuzzy choice functions [5, 6, 7, 19, 20, 33]. Orlovsky first consid-
ered human preferences as fuzzy relations for drawing conclusions in decision
making problems [37]. Banerjee [4] introduced the concept of a fuzzy choice
function whose domain is crisp and range is fuzzy. Banerjee also introduced
three congruence axioms in order to characterize rationality of choice func-
tions. Wang in [50] later proved that these axioms are not independent. (We
examine these results in Chapter 7.) Georgescu [21] considers the fuzzy choice
functions defined on non-empty subsets of non-zero fuzzy subsets of the set of
alternatives. Georgescus work is a natural extension of [4]. In subsequent pa-
pers [21, 22, 23, 25] and the monograph [24], fuzzy forms of revealed preference
axioms and congruence axioms are developed. In [25], Georgescu introduced
21
i i
i i
i i
K23798 2015/2/2
i i
22 2. Fuzzy Choice Functions
a fuzzy Arrow axiom, a weak fuzzy congruence axiom, and a strong fuzzy
congruence axiom.
2.1 Basic Properties

The first two sections of this chapter address sufficient conditions for a par-
tially acyclic fuzzy choice function to be rationalizable. We find that certain
fuzzy choice functions that satisfy conditions and are rationalizable. Fur-
thermore, any fuzzy choice function that satisfies these two conditions satisfies
Arrow and WARP
Formal models have found increasing application in comparative politics
in recent years. Spatial models have been particularly popular. The individ-
ual and collective preferences of humans are inherently vague and ambiguous
no matter how they are aggregated. The conventional crisp approach forces
upon spatial models a degree of precision in human thinking that is not re-
alistic. Moreover, as was demonstrated in [14], a crisp approach requires a
larger number of highly restrictive assumptions in order to generate stable
predictions.
In the absence of a maximal set, spatial models predict cycling. Cycling
is a condition under which every alternative can be majority defeated by at
least one other alternative. Austen-Smith and Banks [2, pp. 6-21] determine
under what conditions observable political outcomes (i.e., the choices made
by political actors) are consistent with the preferences of players. If the choice
function used to select an alternative is rationalizable, then choices should be
a proper subset of a maximal set, when one exists.
Unless otherwise stated, we assume in this chapter that strict preferences
are of type (0) .
Definition 2.1.1 Let C : FP (X) FP (X) be such that C()

FP (X). Then C is called a fuzzy choice function on X. Define C :
P (X) P (X) by S P (X), C (S) = Supp(C(1S )).
A fuzzy choice function C assigns to every element in X the degree to

which it is chosen with respect to every fuzzy subset of X. The degree to
which it is selected cannot be larger than the degree for which it is a member
of .
Proposition 2.1.2 Let C be a fuzzy choice function on X. Then C is a

choice function on X.
Proof. Let S P (X). Let x C (S). Then x Supp(C(1S )). Now
C(1S ) 1S and so Supp(C(1S )) Supp(1S ) = S. Thus x S. Hence
C (S) S. Thus C is a choice function on X.
i i
i i
i i
K23798 2015/2/2
i i
2.1. Basic Properties 23
Suppose C is a fuzzy choice function on X and F P (X). Then it

follows that C()t t t Im(). The condition (not assumed here) that
S P (X), C(1S ) = 1T for some nonempty T S assures that C maps char-
acteristic functions onto characteristic functions and so C can be considered
to be a choice function on X. Consequently, if this condition is assumed, then
we can abuse the notation and write C(t ) t t Im() FP (X).
Lemma 2.1.3 Suppose , FR (X). Then is the strict preference rela-

tion associated with if and only if Supp() is the strict preference relation
associated with Supp() and = on Supp().
Proof. is the strict preference relation associated with (x, y X,
(x, y) = (x, y) if (x, y) > 0 and (y, x) = 0; (x, y) = 0 otherwise)
x, y X, (x, y) Supp() if (x, y) Supp() and (y, x) / Supp() and
= on Supp() Supp() is the strict preference relation associated with
Supp() and = on Supp().
If the preferences of players are known, reflexivity, completeness, and
acyclicity are easily determined. However, they are not always known. For a va-
riety of reasons ranging from electoral strategy to national security, politicians
and political actors often hide or misrepresent their preferences. Therefore,
scholars must deduce their preferences from their observed choices. Among
the conventional approaches is to estimate the preferences of parties and pres-
idents based on roll-call votes and the executive veto. In so doing, individuals
assume that observable choices comprise a subset of the maximal set. That
is, the choice function used by political actors is rationalizable, by which we
mean that observable political outcomes (choices over a set of alternatives)
reflect preference.
Assuming that a fuzzy maximal set has nonempty support, we next con-
sider the conditions under which fuzzy choice functions are rationalizable.
Definition 2.1.4 Let C be a fuzzy choice function on X. Let F R (X).

Then C is called rationalizable with respect to if C() = M (, ) for all
FP (X).
Example 2.1.5 Let X = {x, y, z}. Define : X X [0, 1] as follows:
(x, x) = (y, y) = (z, z) = 1, (x, y) = (x, z) = 1/2,

(y, z) = (z, y) = 1/4, (y, x) = (z, x) = 0.
Clearly is reflexive and complete. Now u, v X, (u, v) = 1/2 if and only

if u = x and v = y or z, and (u, v) = 0 otherwise. Thus it follows that
is acyclic. Define C : FP (X) FP (X) by FP (X), C()(w) =
i i
i i
i i
K23798 2015/2/2
i i
MM (, )(w) w X. Then rationalizes C by Theorem 1.2.10. By Proposi-

tion 1.2.9, we have the following table.
S {x} {y} {z} {x, y} {x, z} {y, z} {x, y, z}

C(1S )(x)
1 0 0 1/2 1/2 0 1/2
= M (, 1S )(x)
C(1S )(y)
0 1 0 0 0 1/4 0
= M (, 1S )(y)
C(1S )(z)
0 0 1 0 0 1/4 0
= M (, 1S )(z)
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.
Proposition 2.1.6 Let FR (X) be such that is complete, partially

acyclic, and (x, x) Supp() x X. Suppose has no zero divisors. Define
C : FP (X) FP (X) by FP (X), C()(x) = M (, )(x) x X.
Then rationalizes C.
Proof. Let x X. Then

M (, )(x) > 0
(x) (~wX ({t [0, 1] | C(w, x) t (x, w)})) > 0
(x) > 0 and {t [0, 1] | C(w, x) t (x, w)})) > 0 w X
(x) > 0 and w X, tw (0, 1] such that C(w, x) tw (x, w).
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
2.1. Basic Properties 25
preference relation associated with Supp() by Lemma 2.1.3, it follows that

Supp() = t is acyclic for all t (0, t ]. Since (x, x) t t [0, t ] and
x X, t is reflexive t [0, t ]. Let s = {(x) | x X}. Then it fol-
lows that {x X | x t and (x, y) t y t } 6= if t [0, t s ]
by [2, Theorem 1.1, p. 5]. Thus {x X | x t and (x, y) t y
Supp()} =6 if t [0, t s ], where s = {(x) | x Supp()}. Thus
{x t | (x, w) t w Supp()} = 6 t (0, t s ]. Hence x
Supp() such that w X, (x, w) tw , where tw (0, t s ]. (Actually,
tw can be taken independent of w.) Thus (x, w) C(w, x) t w X,
where t = {tw | w X}. Since X is finite, t > 0. Thus x X such that
M (, )(x) > 0.
Example 2.1.7 Let X = {x, y, z}. Suppose has no zero divisors. Define the
fuzzy relation on X as follows:
(x, x) = (y, y) = (z, z) = 1, (x, y) = 1/2, (x, z) = 1/2,

(y, z) = (z, y) = 1/4, (y, x) = (z, x) = 0.
Clearly, is reflexive and complete. Now u, v X, (u, v) = 1/2 if u = x

and v = y or z, and (u, v) = 0 otherwise. Thus it follows that is acyclic.
Define C : FP (X) FP (X) by FP (X), C()(w) = M (, )(w)
w X. Then rationalizes C by Proposition 2.1.6. By Proposition 1.2.9, we
have the following table:
S {x} {y} {z} {x, y} {x, z} {y, z} {x, y, z}

C(1S )(x)
1 0 0 1 1 0 1
= MG (, 1S )(x)
C(1S )(y)
0 1 0 0 0 1 0
= MG (, 1S )(y)
C(1S )(z)
0 0 1 0 0 1 0
= MG (, 1S )(z)
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.
Definition 2.1.8 Let C be a fuzzy choice function on X. We say that C is

crisply rationalized if there exists R R(X) such that C() = M (1R , )
FP (X).
Proposition 2.1.9 Let C be a fuzzy choice function on X such that S

P (X), C(1S ) = 1T for some T S. Then C is crisply rationalized if and
only if C is crisply rationalized by RC , where RC is defined by x, y X,
xRC y if and only if x C ({x, y}).
Proof. If C is crisply rationalized by RC , then C is crisply rationalized.
Conversely, suppose C is crisply rationalized by R. Then since S P (X),
i i
i i
i i
K23798 2015/2/2
i i
C(1S ) = 1T for some T S, M (1R , 1S ) = C(1S ) = 1T . Hence C (S) =

Supp(C(1S ) = Supp(1T ) = T. Thus
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).
Hence C (S) = M (R, S), i.e., R rationalizes C and in fact R = RC by [2,

Lemma 1.1, p. 8]. Let FP (X). Then x X,
C()(x) = M (1R , )(x) = M (1RC , )(x).
That is, RC rationalizes C.

The following result states that if there exists a fuzzy binary relation on
X such that the level sets t of every fuzzy subset of X is rationalized by
t for every t, then must be crisp and in fact is the base relation for C .
Proposition 2.1.10 Let C be a fuzzy choice function on X. If there exists

FR(X) such that FP (X) and t Im(), C (t ) = M (t , t ),then
there exists a relation R on X such that = 1R .
Proof. Let S P (X) and let t (0, 1]. Define : X [0, 1] by (x) = t
if x S and (x) = 0 otherwise. Then C (t ) = M (t , t ) by hypothesis.
Since t = S and S is arbitrary, t rationalizes C . Hence t = RC . Since t is
arbitrary, t = RC t (0, 1]. Thus it follows that = 1R for some relation
R on X. (In fact, R = RC .)
2.2 Consistency Conditions

In this section, we consider the effect of placing restrictions on the behavior
of choice functions. We consider the fuzzification of conditions , , ,path
independence, Arrow axiom, and WARP. These conditions place restrictions
on the behavior of choice functions. The fuzzification follows without difficulty
because of our choice of the strict fuzzy preference relation to associate with
fuzzy preference relations, namely strict fuzzy preference relations of type (0) .
We use the notation M for crisp maximal sets at times as well as for the fuzzy
maximal set defined in Definition 1.2.1.
Rational behavior of consumers is an important concern in welfare eco-
nomics. To study the rationality of consumers, Samuelson in [42] introduced
the theory of revealed preferences through preference relations associated with
i i
i i
i i
K23798 2015/2/2
i i
2.2. Consistency Conditions 27
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}.
Recall that Sy = {w X | (w, y) > (y, w)}.
Proposition 2.2.2 Let FR (X) and let S be a subset of X. Suppose

is a t-norm without zero divisors. Then MG (, 1S ) = 1T for some T S if
and only if x, y S, (x, y) (y, x) > 0 implies (x, y) = (y, x).
Proof. Suppose MG (, 1S ) = 1T for some T S. Let x, y S. Suppose

(x, y) (y, x) > 0. Then (x, y) > 0 and (y, x) > 0. Suppose (x, y) >
(y, x) > 0. w Sy , let ty,w = {t [0, 1] | (w, y) t (y, w)}. Then
MG (, 1S )(y) = ~{ty,w | w Sy }. Since Sy 6= and (y, x) > 0 it follows
that 0 < ~{(y, w) | w Sy } < 1. However, this contradicts the hypothesis.
Thus (x, y) = (y, x).
Conversely, suppose that x, y S, (x, y) (y, x) > 0 implies (x, y) =
(y, x). Let T = {x S | @w S, (w, x) > (x, w)}. Suppose x S\T. Then
w S such that (w, x) > (x, w). By our hypothesis, (x, w) = 0. Thus
MG (, 1S )(x) = 0. Hence MG (, 1S ) = 1T .
The next result says that if a fuzzy choice function C when restricted to
crisp sets S yields a crisp subset of S and if C is rationalizable with respect
to in FR(X), then is also crisp.
Proposition 2.2.3 Let C be a fuzzy choice function on X such that S

P (X), C(1S ) = 1T for some nonempty T S. Suppose that there exists
FR(X) such that C(1S ) = MG (, 1S ) S P (X). Then there exists a
relation R on X such that = 1R .
Proof. Let S P (X) and let t (0, 1]. Then x MG (, 1S )t {{u
[0, 1] | (w, x) u (x, w)} | w S} t (x, w) (w, x) t w S
and x MG (t , S) MG (1t , 1S )(x) > 0 {{u [0, 1] | 1t (w, x)
u 1t (x, w)} | w S} > 0 w S, [(w, x) t (x, w) t or
(w, x) / t ]. Thus MG (, 1S )t = MG (t , S). Hence C (S) = Supp(C(1S )) =
i i
i i
i i
K23798 2015/2/2
i i
Supp(1T ) = T = (1T )t = C(1S )t = MG (, 1S )t = MG (t , (1S )t ) = MG (t , S).

Since S is arbitrary, t rationalizes C t (0, 1]. Thus t = RC t (0, 1].
Definition 2.2.4 Let C be a fuzzy choice function on X. Define C : XX

[0, 1] by x, y X, C (x, y) = C(1{x,y} )(x).
Proposition 2.2.5 Let C be a choice function on X and let FR(X).

Suppose rationalizes C. Then x, y X,

1 if (y, x) (x, y),
C (x, y) =
tx,y otherwise,
Furthermore, C rationalizes C for characteristic functions.
Proof. Let x, y X. Then
C (x, y) = C(1{x,y} )(x)

= 1{x,y} (x)
(~wX ({t [0, 1] | 1{x,y} (w) (w, x) t (x, w)))
= ({t [0, 1] | (x, x) t (x, x)})
~({t [0, 1] | (y, x) t (x, y)})
= {t [0, 1] | (y, x) t (x, y)}
from which the desired result is immediate. We now show C rationalizes C

for characteristic functions. It suffices to compare C (w, x) t C (x, w) and
(w, x) t (x, w). Now C (w, x) t C (x, w) implies t = 1 is the largest
possible t if C (w, x) C (x, w) and C (w, x) t C (x, w) implies tx,w is
the largest possible t if C (w, x) C (x, w). Also, (w, x)t (x, w) implies
t = 1 is the largest possible t if (w, x) (x, w) and (w, x) t (x, w)
implies tx,w is the largest possible t if (w, x) (x, w). The desired result
now follows since C (x, w) = tx,w if (w, x) > (x, w) and (x, w) > (w, x)
C (x, w) > C (w, x).
Corollary 2.2.6 Let C be a fuzzy choice function on X and let FR(X).

Suppose rationalizes C. Then x X, C (x, x) = 1.
Recall that in the crisp case, condition requires that if an alternative x

is chosen from a set T and S is a subset of T containing x, then x should still
be chosen.
Definition 2.2.7 Let C be a fuzzy choice function on X. Then C is said to

satisfy condition if , FP (X), implies C() C().
C is said to satisfy condition for characteristic functions if S, T
P (X), S T implies C(1T ) 1S C(1S ).
i i
i i
i i
K23798 2015/2/2
i i
Condition says that the degree of membership of an element in C() is

at least as large as the minimum of the degrees of membership of the elements
in C() and .
Lemma 2.2.8 Let C be a fuzzy choice function on X. Suppose has no

zero divisors. Let x1 , . . . , xn X. Suppose C satisfies condition . Then
C (x1 , x2 ) C (x2 , x3 ) . . . C (xn1 , xn ) > 0 implies C (x1 , x2 ) > 0.
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 ),
which contradicts C (xi1 , xi ) > 0. Thus C(1{x1 ,...,xn } )(xi ) = 0 for i =

2, . . . , n. Hence C(1{x1 ,...,xn } )(x1 ) > 0. Thus
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 ).
Proposition 2.2.9 Suppose C is a fuzzy choice function on X that satisfies

condition . Suppose has no zero divisors. If C is crisp, then C is acyclic.
Proof. Let x1 , . . . , xn X. Suppose that C (x1 , x2 ) C (x2 , x3 ) . . .
C (xn1 , xn ) > 0. Then C (x1 , xn ) > 0 by Lemma 2.2.8 and so C (x1 , xn ) =
1. Thus
C (x1 , x2 ) C (x2 , x3 ) . . . C (xn1 , xC ) C (x1 , xn ).
In the crisp case, condition requires that if an alternative x was chosen

from a set S and also from a set T, then it should be chosen from the set
S T. In the next definition, we fuzzify this condition.

satisfy condition if , F P (X), C() C() C( ). C is said
to satisfy condition for characteristic functions if S, T P (X),
C(1S ) C(1T ) C(1S 1T ).
Condition says that the degree of membership of any alternative x in
C( ) should not be strictly less than the smallest degree of membership
of x in C() and C().
i i
i i
i i
K23798 2015/2/2
i i
Proposition 2.2.11 Suppose C is a fuzzy choice function on X. If C is par-

tially acyclic and C satisfies conditions and for characteristic functions,
then C rationalizes C for characteristic functions, i.e., C(1S ) = MG (C , 1S )
S P (X). Conversely, if C rationalizes C for characteristic functions,
then C satisfies condition and satisfies condition if = = ~.
Proof. Suppose that C is partially acyclic and C satisfies conditions
and for characteristic functions. Clearly C is complete and reflexive. Since
C is acyclic, x X such that MG (, )(x) > 0 FP (X) by the proof of
Proposition 2.1.6. By condition , 1{x,y} C(1S ) C(1{x,y} ), where x, y S.
By the definition of C , C (x, y) = C(1{x,y} )(x) 1 C(1S )(x) = C(1S )(x).
Now
MG (C , 1S )(x) = 1S (x)
(~wX ({t [0, 1] | 1S (w) C (w, x) t C (x, w)}))
= ~wX ({t [0, 1] | 1S (w) C (w, x) t C (x, w)}))
= ~wS ({t [0, 1] | C (w, x) t C (x, w)}))
C(1S )(x),
where the latter inequality holds since the largest t for which C (w, x) t
C (x, w) is 1 if C (w, x) < C (x, w) and is C (x, w) (which is C(1S )(x))
if C (w, x) > C (x, w). Thus C(1S ) MG (C , 1S ). Suppose S = {x}. Then
it follows easily that C(1{x} ) = 1{x} = MG (C , 1{x} ). Now suppose S =
{x, y1 , . . . , yn }. If n = 1, then
C(1{x,y1 } )(x)
= C (x, y1 ) {t [0, 1] | C (y1 , x) t C (x, y1 )}
= ~w{x,y1 } ({t [0, 1] | C (w, x) t C (x, w)})
= 1{x,y1 } (x) (~wX ({t [0, 1] | 1{x.y1 } (w) C (w, x) t C (x, w)}))
= MG (C , 1S )(x).
Hence C(1S ) MG (C , 1S ). Suppose n 1 and C(1{x,y1 } ) . . .
C1{x,yk } ) C(1{x,y1 ,...,yk } ) for 1 k < n, the induction hypothesis. By
condition , C(1{x,y1 ,...,yk } ) C(1{x,yk+1 } ) C(1{x,y1 ,...,yk+1 } ). Thus it follows
by induction that yS C(1{x,y} ) C(1S ). Thus C(1S ) MG (C , 1S ) and so
C(1S ) = MG (C , 1S ).
Conversely, suppose C rationalizes C for characteristic functions. Let x
X. Let S, T P (X) be such that S T. If (C(1T ) 1S )(x) = 0, then
(C(1T ) 1S )(x) C(1S )(x). Suppose that (C(1T ) 1S )(x) > 0. Then x S
and (C(1T ) 1S )(x) = C(1T )(x). Now
0 < C(1T )(x)
= MG (C , 1T )(x)
= 1T (x)
(~wX ({t [0, 1] | 1T (w) C (w, x) t C (x, w)}))
1 (~wX ({t [0, 1] | 1S (w) C (w, x) t C (x, w)}))
= MG (C , 1S )(x)
= C(1S )(x),
i i
i i
i i
K23798 2015/2/2
i i
where the inequality holds since x S T. Thus C(1T ) 1S C(1S ).

Hence condition holds for characteristic functions. Let S, T P (X) and
x X. Suppose that 0 = (C(1S ) C(1T ))(x). Then (C(1S ) C(1T ))(x)
C(1ST )(x). Suppose that (C(1S ) C(1T ))(x) > 0. Since S, T S T, it
follows that MG (C , 1S ) MG (C , 1T ) MG (C , 1ST ) (since = = ~) =
MG (C , 1S 1T ). Thus C(1S ) C(1T ) C(1S 1T ). Hence condition holds
for characteristic functions.
Example 2.2.12 Let X = {x, y}. Let C be the choice function on X such
that
C(1{x} )(x) = 1, C(1{x} )(y) = 0, C(1{y} )(x) = 0, C(1{y} )(y) = 1,

C(1{x,y} )(x) = 1/2, C(1{x,y} )(y) = 0.
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.

satisfy path independence (PI) if , F P (X), C( ) = C(C()
C()).
Definition 2.2.13 is a fuzzification of the crisp case which says that choices
from are the same as those arrived at by first choosing from and from
, and then choosing among these chosen alternatives.
Path independence yields the following result: An alternative x
Supp(C( )) if and only if x Supp(C(C() C())). Since x
Supp(C( )) implies x Supp( ) and also x Supp(C(C() C()))
implies x Supp(C() C()) similar comments concerning decentralization
as those in [2, p. 13] can be made under certain conditions.
There are several ways to define set difference, \, of two fuzzy subsets
of and of X. For example, consider the following definition.
Definition 2.2.14 Let , FP(X).

(1) Define \ by x X,

(x) if (x) = 0,
(\)(x) =
0 if (x) > 0.
i i
i i
i i
K23798 2015/2/2
i i

(x) if (x) < (x),
(\)(x) =
0 if (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().
Proposition 2.2.15 Let C be a fuzzy choice function on X. Suppose C satis-

fies PI. Then (, ) S(\) such that , C() C(), i.e., C satisfies
condition for S(\).
Proof. Let x X. Then
(C() )(x) = C()(x) (x) = C( )(x) (x)
= C( (\))(x) (x) = (C(C() C(\))(x)) (x)
(C() C(\))(x)) (x)
= C()(x) (x)) (C(\)(x) (x))
(C()(x) (x)) ((\)(x) (x)) = C()(x) 0
= C()(x).
Thus C() C().
The notion of quasi-transitivity will play an important role in our consid-
eration of dictatorships and oligarchies later in the book. We use it here in
the study of rationalization.
Definition 2.2.16 Let FR (X). Then

(1) is said to be quasi-transitive if x, y, z X, (x, y) (y, z)
(x, z);
(2) is said to be partially quasi-transitive if x, y, z X, (x, y)
(y, z) > 0 implies (x, z) > 0.
Proposition 2.2.17 Let FR (X). If t is quasi-transitive t [0, 1],

then is quasi-transitive. Conversely, if = and is quasi-transitive, then
t is quasi-transitive t [0, 1]
Proof. Let x, y, z X. Suppose t is quasi-transitive t [0, 1]. Let
(x, y) (y, z) = t. Then (x, y), (y, z) t and so (x, z) t . Thus (x, z)
(x, y)(y, z). Hence is quasi-transitive. Conversely, suppose = and is
quasi-transitive. Let t [0, 1]. Then (x, y), (y, z) t (x, y) t, (y, z)
t (x, z) (x, y) (y, z) t. Hence (x, z) t and so t is quasi-
transitive.
i i
i i
i i
K23798 2015/2/2
i i

satisfy condition if x, y X, C(1{x,y} ) = t{x} for some t (0, 1] if and
only if s [0, 1], C(t{x} s{y} ) = t0{x} for some t0 (0, 1].
Proposition 2.2.19 Let C be a fuzzy choice function on X. Suppose C sat-

isfies condition . If C satisfied PI, then C is partially quasi-transitive.
Proof. Let x, y, z X. Suppose C (x, y) C (y, z) > 0. Then C (x, y) >
0 and C (y, z) > 0 and so C(1{x,y} )(x) = t0 > 0, C(1{x,y} )(y) = 0 and
C(1{y,z} )(y) = s0 > 0, C(1{y,z} )(z) = 0. Now C(1{x,y,z} ) = C(C(1{x} )
C(1{y,z} ) = C(t{x} s0{y} ) = t00{x} . Also, C(1{x,y,z} ) = C(C(1{x,y} C(1{z} )) =
C(t0{x} r{z} ). Thus C(t0{x} r{z} ) = t00{x} . Hence C(1{x,z} )(z) = 0 and so
C (z, x) = 0. Now c (x, z) > 0 since 1 6= C(1{x,z} ) 1{x,z} . Thus C (x, z) >
0.
Example 2.2.20 Let X = {x, y, z}. Suppose C is a fuzzy choice function on

X such that C(1{x} ) = t{x} , C(1{y} )(y) = s{y} , and C(1{z} )(z) = r{z}

for

some t , s , r (0, 1]. Suppose also that
C(t{x} s{y} ) = (t t){x} t, s (0, 1],

C(t{x} r{z} ) = (t t){x} t, r (0, 1],
C(s{y} r{z} ) = (t s){y} s, r (0, 1],
C(t{x} s{y} r{z} ) = (t t){x} t, s, r (0, 1].
Then
C(C1{x} ) C(1{y,z} )) = C(t{x} t{y} ) = (t t ){x} ,

C(C1{y} ) C(1{x,z} )) = C(t{y} t{x} ) = (t t ){x} ,
C(C1{z} ) C(1{x,y} )) = C(t{z} t{x} ) = (t t ){x} .
Hence it follows that C satisfies PI for characteristic functions. Since
C(1{x,y} 1{x} and C(t{x} s{y} ) 1{x} ,

C(1{x,z} 1{x} and C(t{x} r{z} ) 1{x} ,
C(1{y,z} 1{y} and C(s{y} r{z} ) 1{x} ,
C satisfies condition .
Proposition 2.2.21 Let C be a fuzzy choice function on X. If C is par-

tially acyclic and C satisfies PI, condition , and condition for characteris-
tic functions, then C is rationalized for characteristic functions by a partially
quasi-transitive fuzzy relation on X. Conversely, if C is rationalized for char-
acteristic functions by a quasi-transitive fuzzy relation on X, then C satisfies
PI and conditions and for characteristic functions.
i i
i i
i i
K23798 2015/2/2
i i
Proof. Suppose C is partially acyclic and C satisfies PI, condition ,

and condition for characteristic functions. Since C satisfies PI, C satisfies
condition for S(\) by Proposition 2.2.15. Hence C satisfies condition for
characteristic functions. By Proposition 2.2.11, C rationalizes C for charac-
teristic functions. By Proposition 2.2.19, C is partially quasi-transitive.
Conversely, suppose C is rationalized for characteristic functions by a
quasi-transitive fuzzy relation on X. By Proposition 2.2.11, C satisfies condi-
tions and for characteristic functions. Now
C(C(1S ) C(1T )) = MG (C , C(1S ) C(1T ))
= MG (C , M (C , 1S ) MG (C , 1T )).
Since C(1S 1T ) = MG (C , 1ST ) , we must show that
MG (C , MG (C , 1S ) MG (C , 1T )) = MG (C , 1ST ).
By [36, Corollary 2.15, p. 314], Proposition 2.2.11, and [2, Theorem 1.3, p.
14], we have t [0, 1] (and writing for C ) that
MG (C , 1ST )t = MG (t , S T ) = MG (t , M (t , S) MG (t , T ))
= MG (t , MG (t , (1S )t ) MG (t , (1T )t )
= MG (t , MG (t , (1S )t ) MG (t , (1T )t )
= MG (t , (MG (, 1S ) MG (, 1T ))t )
= MG (, MG (, S) MG (, T ))t
Thus MG (, 1ST ) = MG (, MG (, 1S ) MG (, 1T )).
Definition 2.2.22 Let FR(X). Then is called partially transitive if

x, y, z X, (x, y) (y, z) > 0 implies (x, z) > 0.
In the crisp case, condition requires that if an alternative is chosen from
a set S and is still chosen from a larger set T, then all previously chosen
alternatives from S must be chosen from T .

satisfy condition if for all , F P (X), and C() C() 6=
imply C() C().
Condition says that if extends and the degree of membership of an
alternative in both C() and C() is positive, then the degree of membership
of every alternative in C() is as large or larger than its degree of membership
in C().
Proposition 2.2.24 Let C be a fuzzy choice function on X. Suppose C sat-

isfies conditions and . Then the following properties hold.
(1) C satisfies condition for all (, ) SC , where SC = {(, ) | ,
FP (X), ( ) 6= 1 }.
(2) C is partially transitive.
i i
i i
i i
K23798 2015/2/2
i i
Proof. (1) Let , FP (X). Suppose C()C() = 1 . Then condition

clearly holds. Suppose C()C() 6= 1 . By condition , C() C().
Let (, ) SC . Then 1 6= C() = C()C(). Hence by condition
, C() C( ). Thus C() C() C( ).
(2) Let x, y, z X be such that C (x, y) > 0, C (y, z) > 0. By con-
dition , C(1{x,y,z} ) 1{x,z} C(1{x,z} ). Thus (i) if C(1{x,y,z} )(x) > 0,
then C(1{x,z} )(x) > 0 by condition . (ii) Suppose C(1{x,y,z} )(y) > 0.
Then C(1{x,y} )(y) > 0 by condition . Hence by condition , C(1{x,y} )
C(1{x,y,z} ). Since C(1{x,y} )(x) = C (x, y) > 0, C(1{x,y,z} )(x) > 0. Thus
by (i), C(1{y,z} )(x) > 0. Suppose C(1{x,y,z} )(z) > 0. Then by condi-
tion , C(1{y,z} )(z) > 0. By condition , C(1{y,z} ) C(1{x,y,z} ) and
so C(1{x,y,z} )(y) > 0 since C(1{y,z} )(y) = C (y, z) > 0. Thus by (ii),
C(1{x,z} )(x) > 0. Since C(1{x,y,z} ) 6= 1 , we have in all cases that C (x, z) =
C(1{x,z} )(x) > 0.
Definition 2.2.25 Let C be a fuzzy choice function in X. Then C is called

resolute if |Supp(C())| = 1 FP (X).
Let SC be as defined in Proposition 2.2.24.
Theorem 2.2.26 Let C be a fuzzy choice function on X. If C is rational-

ized by a partially transitive fuzzy relation, then Supp(C(1S ) Supp(C(1T ))
S, T P (X) such that S T. Conversely, if C satisfies conditions and
, then is C rationalizable for characteristic functions in SC .
Proof. Suppose C satisfies conditions and . Then C satisfies condition
for SC by Proposition 2.2.24. Since condition holds for C for characteristic
functions, C is rationalizable for characteristic functions in SC by Proposition
2.2.11.
Suppose C is rationalizable by a partially transitive fuzzy relation. Then C
satisfies conditions and for characteristic functions by Proposition 2.2.11.
Let S, T P (X) be such that S T. Suppose C(1S ) C(1T ) 6= 1 . Then
(C(1S ) C(1T ))(x) > 0 for some x X. Suppose C(1S )(y) > 0. Then
C (y, x) > 0 since C rationalizes C. Now C(1T )(x) > 0 and C (y, z) >
0z T. Thus C (y, z) > 0 z T since C is rationalized by a partially
transitive relation. Hence C(1T )(y) > 0. Thus Supp(C(1S ) Supp(C(1T )).
We note that if S, T P (X) are such that S T and if C is rationalized

for characteristic functions, then C(1T )|S C(1S ) on S. Hence if C(1S )
C(1T ), then C(1T )|S = C(1S ) on S.
Corollary 2.2.27 Let C be a fuzzy choice function on X that is resolute. If

C satisfies condition , then C is rationalizable for characteristic functions
in SC .
i i
i i
i i
K23798 2015/2/2
i i
Proof. Since C is resolute, C satisfies condition .

The next two conditions each identify choice functions admitting transitive
rationalizations. In the crisp case, the Arrow axiom says that for S T
that if there are alternatives in S that are chosen from T, then exactly these
alternatives must be chosen from S, i.e., S C(T ) = or S C(T ) = C(S).

satisfy the Arrow axiom if , FP (X) such that , C() = 1
or C() = C().
In the crisp case, condition WARP says that if an alternative x is chosen
from S and alternative y isnt and y is chosen from T, then x is not a member
of T.
Definition 2.2.29 Let C be a fuzzy choice function on X. Then C is said

to satisfy the weak axiom of revealed preference (WARP) if ,
FP (X), x Supp(), y Supp()\Supp(C()) and y Supp(C()) imply
x Supp().
/
Theorem 2.2.30 Let C be a fuzzy choice function on X. Then (1) (2)

(3), where
(1) C satisfies conditions and ;
(2) C satisfies Arrow;
(3) C satisfies WARP.
Proof. (1) (2) : Let , F P (X) be such that . Then
C() C() by condition . Suppose C() 6= 1 . Then C() 6= 1 . In
fact, ( C()(x) > 0 implies C()(x) > 0 and so (C() C())(x) > 0.
Thus C() C() 6= 1 . Hence by condition , C() C(). Since C()
, C() C(). Thus C() = C().
(2) (1) : Let , FP (X) be such that . Suppose that C() 6=
1 . Then C() () = C() by Arrow. Hence condition holds. That
condition holds is immediate.
(2) (3) : Let x Supp(C()), y Supp()\Supp(C()), y Supp().
Let = . Now Supp() = Supp()Supp() 6= since y Supp() and
y Supp(C()) Supp(). In fact, y Supp(). Since y Supp(C()), y
Supp()Supp(C()) = Supp( C()). Thus Supp( ()) 6= and so
C() 6= 1 . Now and so by Arrow, C() = C(). Hence
y Supp(C()). But y Supp(
/ C()) and so C() 6= C(). Thus
since , we have by Arrow that C() = 1 . Hence x Supp(C())
implies x Supp()
/ which in turn implies x Supp()
/ since x Supp() and
Supp() = Supp()Supp().
i i
i i
i i
K23798 2015/2/2
i i
2.3. M-Rationality and G-Rationality 37
2.3 M-Rationality and G-Rationality

In this section, we consider the situation in which choices are deduced from
vague preferences. We follow Georgescu [24]. We consider M -Rationality and
G-Rationality for choice functions, where M fuzzifies the notion of a -
maximal element and G fuzzifies the notion of a -greatest element. Let C
be a fuzzy choice function. Rationality means that C is recoverable from a
fuzzy preference relation and for rationality which is normal C is recoverable
from the fuzzy revealed preference relation associated with C.
Let be a continuous t-norm. Let a b = {t [0, 1] | a t b}. The
operation is called the residuum associated with , [24].
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.
Definition 2.3.2 Let FR(X). Then we call a fuzzy weak preference

relation (FWPR) if is reflexive and complete.
Let B be a nonempty family of nonzero fuzzy subsets of X.
Definition 2.3.3 Let FP(X) and be a FWPR on X. Define the fuzzy

subsets M (, ) and G(, ) of X as follows:x X,
M (, )(x) = (x) {(y) (y, x) (x, y) | y X},

G(, )(x) = (x) ((y) (x, y) | y X}.
Proposition 2.3.4 Let FP(X) and be a FWPR on X. Then
M (, )(x) = (x) {{t [0, 1] | (y) (y, x) t (x, y)} | y X},

G(, )(x) = (x) {{t [0, 1] | (y) t (x, y)} | y X}.
Proposition 2.3.5 Let FP(X) and be a FWPR on X. Let = .

Then

(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
Proposition 2.3.6 Let S be a subset of X and be a FWPR on X. Then

{(y, x) (x, y) | y S} if x S,
M (1S , )(x) =
0 if x
/ S.
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}.
Recall the negation associated with Lukasiewicz, Godel, and product t-

norms:
Lukasiewicz: a = 1 a;

1 if a = 0.
Godel and product t-norms: a =
0 if a > 0.
A fuzzy subset of a set X is called normal if x X such that (x) = 1.
Let B denote a set of non-zero fuzzy subsets of X. We often assume the
following two conditions, [24, p. 91].
(H1) Every B and C() are normal fuzzy subsets of X.
(H2) B includes all fuzzy subsets 1{x1 ,...,xn } for all x1 , . . . , xn X, n 1.
Definition 2.3.7 Let C : B F P(X) be a fuzzy choice function on (X, B).

Define the following revealed preference relations , , on X by x, y X,
(1) (x, y) = {C()(x) (y) | B};
(2) (x, y) = (x, y) (y, x);
(3) (x, y) = (x, y) (y, x).
Suppose has no zero divisors in Definition 2.3.7(2). Then
(x, y) (y, x) y) {t [0, 1] | (y, x) t = 0}

= (x,
(x, y) 1 if (y, x) = 0
=
(x, y) 0 if (y, x) > 0
Thus (x, y) = (x, y) if (y, x) = 0 and (x, y) = 0 if (y, x) > 0. (Suppose

is complete. Then here is our (0) because if (y, x) = 0, then (x, y) > 0.)
Definition 2.3.8 Assume hypothesis (H2) holds. Let C : B FP(X) be

a fuzzy choice function on (X, B). Define the following revealed preference
relations , , on X by x, y X,
i i
i i
i i
K23798 2015/2/2
i i
(1) (x, y) = C(1{x,y} )(x);

(2) (x, y) = (x, y) (y, x);
(3) (x, y) = (x, y) (y, x).
(x, y) is the degree of truth of the statement at least the alternative x
is chosen from the set {x, y}.

Define the following revealed preference relations e, on X by x, y X,
e, e
e(x, y) = {(C()(x) (y)) C()(y) | B};
(1)
(2) e(x, y) = e (y, x);
(3) i(x, y) = e(x, y) e(y, x).
e

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.
Definition 2.3.10 Let C : B FP(X) be a fuzzy choice function on (X, B).

Define the following revealed preference relations and 1 on X by B,
x, y X,
(1) (x, y) = {((x) C()(y)) C()(x) | B};
(2) 1 (x, y) = ((x) C()(y) C()(x) | B}.
Proposition 2.3.11 [24] Suppose is the Godel t-norm. Let be a reflexive

and max- transitive fuzzy preference relation on a finite set X. Then G( , )
and M ( , ) are fuzzy choice functions on (X, B).
Proof. Suppose that G( , ) is not a fuzzy choice function. Then there
exists B such that G(, )(x) = 0 for all x X. Let x1 X be such that
(x1 ) 6= 0. Then
(x1 ) ({(y) (x1 , y) | y X}) = G(, )(x1 ) = 0.
Since (x1 ) 6= 0 and X is finite, there exists x01 X such that
1 if (x01 ) (x1 , x01 )

0 0
0 = (x1 ) (x1 , x1 ) =
(x1 , x01 ) if (x01 ) > (x1 , x01 ).
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
y3 X such that (y3 ) > 0 and (y2 , y3 ) = 0. Continuing this process, we

obtain a sequence y1 , y2 , . . . , yk , . . . such that (yk ) > 0 and (yk , yk+1 ) = 0,
but yk+1 = yj some j < k + 1 since X is finite. Since is max- transitive, 0 =
(yk , yk+1 ) = (yk , yj ) (yk , k1 ) . . . (yj+1 , yj ) > 0, a contradiction.
Thus no such x exists and so G( , ) is a fuzzy choice function. Since G( , )
M ( , ), it follows that M ( , ) is a fuzzy choice function.
An interesting paper by Martinetti, de Baets, Diaz, and Montes [35 35]
concerning when G( , ) is a fuzzy choice function has recently been published.
We present some of its results in the exercises at the end of this chapter.
Definition 2.3.12 [24] A fuzzy choice function C on (X, B) is called G-

rational if there exists a fuzzy preference relation on X such that C() =
G(, ) for all B. A fuzzy choice function C on (X, B) is called M -
rational if there exists a fuzzy preference relation on X such that C() =
M (, ) for all B.
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,
G ()(x) = G(, )(x) = (x) {(y) (x, y) | y X}
and
M ()(x) = M (, )(x) = (x) {(y) (y, x) (x, y) | y X}.
I(, ) = {(x) (x) | x X}.

The transitive closure of a fuzzy binary relation on a set X is the
smallest transitive fuzzy binary relation on X containing .
Let tc denote the transitive closure of and the transitive closure of

e.
Definition 2.3.14 [24] A fuzzy choice function C is called G-normal if

C() = G () for all B. A fuzzy choice function C is called M -normal
if C() = M () for all B.
Social choice theory most often assumes a binary preference ordering is
complete, reflexive, and transitive. A particular case is consumer theory. If
preferences satisfy these axioms, then they can be characterized by utility
functions. Samuelson developed a theory of revealed preference in [40] based
on utility functions. He provided the following concepts.
Directly revealed preferred. An alternative x is directly revealed pre-
ferred to a different alternative y if y was available when x was chosen.
i i
i i
i i
K23798 2015/2/2
i i
Weak axiom of revealed preference (WARP): If alternative x is

directly revealed preferred to y, then y cannot be directly revealed preferred
to x.
Revealed preferred. The revealed preferred relation is the transitive
closure of the directly revealed preferred relation.
Strong axiom of revealed preference (SARP): If alternative x is
revealed preferred to y, then y will never be revealed preferred to x.
Generalized axiom of revealed preference (GARP): If an alterna-
tive x is revealed preferred to y, then y is never strictly revealed preferred to
x.
We next list axioms of fuzzy revealed preference. We lay the foundation
for their further study later in the chapter and in Chapter 7.
W AF RP weak axiom of fuzzy revealed preference: For all x, y X,
e(x, y) (y, x), i.e.,
e(x, y) 1 if (y, x) = 0 and
e(x, y) = 0 if (y, x) > 0.
SAF RP strong axiom of fuzzy revealed preference: For all x, y X,
(x, y) (y, x).
The next axioms are congruence axioms for fuzzy choice functions.
W F CA weak fuzzy congruence axiom: For all B and for all
x, y X, (x, y) C()(y) (x) C()(x).
SF CA strong fuzzy congruence axiom: For all B and for all
x, y X, tc (x, y) C()(y) (x) C()(x).
We now consider fuzzy versions F and F of conditions and , respec-
tively. The conditions F 0 and F 0 are weak forms of F and F , respectively.
F : I(, ) (x) C()(x) C()(x).
F 0 : , FP(X), x X, if (x) (x), then (x) C()(x)
C()(x). (, FP(X), if , then C() C().)
F : I(, ) C()(x) C()(y) C()(x) C()(y).
F 0 : , F P(X), x X, if (x) (x), then C()(x) C()(y)
C()(x) C()(y).
The following conditions are also fuzzy versions of known consistency con-
ditions. We list the corresponding crisp consistency conditions first for the
convenience of the reader.
: For S, T P (X) and x, y C(S), x 6= y if S T then C(T ) 6= {x}.
2 : For all S P (X) and x S, if x C(S), then x C({x, y}) for
all y S.
2 : For all S P (X) and for all x S, if x C({x, y}) for all y S,
then x C(S).
(+) : For all S, T P (X) such that S T and for all x C(S) and
y S, if y C(T ), then x C(T ).
i i
i i
i i
K23798 2015/2/2
i i
F : S, T P (X), x, y X, I(1S , 1T ) (C(1S )(x) C(1S )(y)

(C(1T )(x) ({C(1T )(z) | z X, z 6= x}.
F 2 : B and x, y X, C()(x) (y) C(1{x,y} )(x).
F 2 : B and x X, (x) ({(y) C(1{x,y} )(x)) | y X}
C()(x).
F (+) : , B and x, y X, I(, ) C()(x) (y) C()(y)
C()(x).
The choice theory developed by Uzawa [49], Arrow [1], and Sen [43-46] was
based on many of the consistency conditions just listed. Ritcher [40], Hansoon
[29] and Suzumura [47] extended these results.
For any choice function C, we consider the following statements:
(i) is a regular preference relation and C is G-normal;
(ii) is a regular preference relation and C is G-normal;
(iii) C satisfies W F CA;
(iv) C satisfies SF CA :
(v) C satisfies W AF RP ;
(vi) C satisfies SAF RP ;
(vii) = ;
(viii) = e and C is G-normal.
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.
Proposition 2.3.16 (Georgescu [24]) If C is a G-normal fuzzy choice func-

tion, then C satisfies F .
Proof. Since C = C, b it suffices to show for all , B and x X that
the following inequality holds
I(, ) (x) C()(x)

b C()(x).
b
i i
i i
i i
K23798 2015/2/2
i i
By Lemma 2.3.1(2), it follows for all u X that
(u) ((u) (u)) ((u) (x, u))

= (u) (u) ((u) (x, u)
= (u) (u) (x, u) (x, u).
By Lemma 2.3.1(1), with a = ((u) (u)) ((u) (x, u)), b = (u), c =

(x, u)), we have
((u) (u)) ((u) (x, u))

(u) (x, u).
Thus
I(, ) (x) C()(x)

b
= {(u) (u) | u X} (x) (x) {(u) (x, u) | u X}
(x) {((u) (u)) ((u) (x, u))} | u X}
(x) {(u) (x, u))} | u X}
= C()(x).
b
Let = 1S , where S is a nonempty subset of X. Let C be a G -normal

fuzzy choice function with respect to . Then x X,
C(1S )(x) = G (1S , )

= 1S (x) ({1S (y) (x, y) | y X})
{(x, y) | y S} if x S,
=
0 otherwise.
Hence if C is G -normal, then C(1S ) = MM (, 1S ). Thus rationalizes C

with respect to characteristic functions.
Corollary 2.3.17 Suppose C is G -normal. Then C satisfies condition F

for characteristic functions.
Proof. By the argument immediately preceding this corollary, we have for

x S that C(1S )(x) = M (, 1S )(x) = {(x, y) | y S} {(x, y) | y
T } = M (, 1T )(x) = C(1T )(x). Suppose x T \S. Then 1S (x) C(1T )(x) =
0 C(1S )(x).
Proposition 2.3.18 (Georgescu [24]) If a fuzzy choice function C satisfies

W F CA, then conditions F and F hold.
i i
i i
i i
K23798 2015/2/2
i i
Proof. Since C is W F CA, C is G-normal by Theorem 2.3.15. Hence by

Proposition 2.3.16, condition F holds. Suppose that condition F does not
hold. Then there exists , B and x, y X such that
I(, ) C()(x) C()(y) C()(x) C()(y) =
(C()(x) C()(y)) (C()(y) C()(x)).
Thus either
I(, ) C()(x) C()(y) C()(x) C()(y)
or
I(, ) C()(x C()(y) C()(y) C()(x).
Assume the first case. Then by Lemma 2.3.1(1),
(a) I(, ) C()(x) C()(y) C()(x) C()(y).
Now
I(, ) C()(x) C()(y) C()(x) C()(y) (x) C()(x)
= C()(y) (x) C()(x) (y)
since C()(y) (y) = C()(y). Since C()(y) (x) (y, x), we have that
(b) I(, ) C()(x) C()(y) C()(x) (y, x) C()(x) (y).
By (a) and (b), it follows that (y, x) C()(x) (y) C()(y). Hence
W F CA does not hold.
A similar argument can be used if the second case holds.
Proposition 2.3.19 (Georgescu [24]) If a fuzzy choice function C satisfies

F and F , then W F CA holds.
Proof. Let B and x, y X. Since I(1{x,y} , ) = (x) (y), it follows
that
(x) C()(y) (x, y)
= (x) (y) C()(y) (x, y)
= I(1{x,y} , ) (x) (y) C()(y) (x, y)
= I(1{x,y} , ) 1{x,y} (y) C()(y) (x) (y) (x, y)
C(1{x,y} )(y) (x) (y) C()(y) (x, y)
since I(1{x,y} , ) 1{x,y} (y) C()(y) C(1{x,y} )(y) by condition F . Re-
placing (x, y) with its expression in Definition 2.3.7, we obtain
(x) C()(y) (x, y)
C(1{x,y} )(y) (x) (y) C()(y) ({C()(x) (y) | B})
= {C(1{x,y} )(y) (x) (y) C()(y) C()(x) (y) | B}.
i i
i i
i i
K23798 2015/2/2
i i
By condition F ,
C()(x) (y) = C()(x) (x) (y)

= I(1{x,y} , ) 1{x,y} (x) C()(x)
C(1{x,y} )(x).
Thus by F and Lemma 2.3.1(2),
C(1{x,y} )(y) (x) (y) C()(y) C()(x) (y)

C(1{x,y} )(y) (x) (y) C()(y) C(1{x,y} )(x)
= I(1{x,y} , ) C(1{x,y} )(x) C(1{x,y} )(y) C()(y)
[C()(x) C()(y)] C()(y)
C()(y) [C()(y) C()(x)]
= C()(y) C()(x)
C()(x).
Since these inequalities hold for all B, (x) C()(y) (x, y) C()(x).
Thus W F CA holds for C.
Theorem 2.3.20 For a fuzzy choice function C, W F CA holds if and only if

F and F hold
Proof. The result holds from Propositions 2.3.18 and 2.3.19.
A fuzzy preference relation is said to be strongly total if x, y X, x 6=
y, either (x, y) = 1 or (y, x) = 1.
Definition 2.3.21 Let C : B FP(X) be a fuzzy choice function. Then C

is said to be full rational if C = G( , ) for some reflexive, max- transitive,
and strongly total fuzzy preference relation .
Definition 2.3.22 Let C be a fuzzy choice function on a fuzzy choice space

(X, B). Then C is said to satisfy the fuzzy Arrow axiom FAA if for all 1 , 2
B and for all x X,
I(1 , 2 ) 1 (x) C(2 )(x) E(1 C(2 ), C(1 )).
Theorem 2.3.23 (Georgescu [24]) If C : B FP(X) is a fuzzy choice

function, then the following statements are equivalent:
(1) C is full rational;
(2) C satisfies F AA.
i i
i i
i i
K23798 2015/2/2
i i
2.4 Full Rationality of Fuzzy Choice Functions

on Base Domain
We now examine full rationality of fuzzy choice functions on base domain.
The results are based on those of Desai and Chaudhari [17]. The concept
of rational choice as an optimizing choice is binary in nature in the view of
Bossert, Surmont, and Suzumura [9]. Consequently, they discussed classical
choice functions defined on a domain that contained all singleton and all
two-element subsets of the universal set of alternatives. They subsequently
characterize full rationality of choice functions in terms of the T-congruence
axiom. The choice of functions may or may not contain subsets of three or more
elements of the universal set of alternatives in the approach adopted in [9]. In
contrast, fuzzy choice functions are defined on the set domain consisting of all
characteristic functions of all single and two element subsets of the universal
set in [10, 17, 12]. Moreover, the effects of introducing various congruence
and revealed preference axioms on the rationality of fuzzy choice functions
are examined. Chaudhari and Desai [10] subsequently introduce the notion
of a fuzzy T-congruence axiom to characterize full rationality of fuzzy choice
functions.
In this section, we present the concept of a weak form of fuzzy T-
congruence axiom, which was introduced in [17]. We characterize the full
rationality of fuzzy choice functions in terms of this axiom and the fuzzy
Chernoff axiom. We show that this form is insufficient to imply full rational-
ity of fuzzy choice functions defined on base domain. However, the weak fuzzy
T-congruence axiom together with the fuzzy Chernoff axiom characterizes full
rationality. We also show that the fuzzy Arrow axiom alone does not imply full
rationality of fuzzy choice functions defined on a base domain and also that
G-rational fuzzy choice functions with transitive rationalizations characterize
their full rationality if the fuzzy Chernoff axiom holds.
Recall that the operation is a continuous t-norm called the G
odel t-
norm. The fuzzy implication operator associated with is the binary oper-
ation on [0, 1] defined as follows: a, b [0, 1],
(
1 if a b
ab=
b if a > b
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.
Recall that a fuzzy subset of X is called normal if (x) = 1 for some

x X and is called non-zero if (x) > 0 for some x X. For , F(X),
i i
i i
i i
K23798 2015/2/2
i i
2.4. Full Rationality of Fuzzy Choice Functions on Base Domain 47
we denote the degree of subsethood of in by I(, ) and the degree of

equality of and by E(, ), where I(, ) = {(z) (z) | z X} and
E(, ) = {(z) (z) | z X}. Clearly, if and only if I(, ) = 1
and = if and only if E(, ) = 1.
A fuzzy binary relation on X is also called fuzzy preference relation
on X. Recall that a fuzzy binary relation is said to be strongly connected
if (x, y) = 1 or (y, x) = 1 for all x, y X. We next recall the definitions of
fuzzy choice function and fuzzy revealed preference relation for base domains
and few basic results discussed in [10].
Let X be a non-empty set of alternatives and B be a non-empty family
of non-zero fuzzy subsets of X which contain characteristic functions of all
singletons and two-element subsets of X. Then B is called the base domain.
A fuzzy choice function on (X, B) is a function C : B FP(X) such that
for all B, C() 6= and C() .
In this approach the domain, B, of the fuzzy choice functions do not nec-
essarily contain characteristic functions of the set having three or more ele-
ments. In classical social choice theory, it is assumed that every choice set is
non-empty. Consequently, we assume in this section as was done in [17] that
for every B, C() is a normal fuzzy subset of X.
Let C : B FP(X) be a fuzzy choice function on (X, B). The fuzzy
revealed preference relation of C is defined as follows: B, x, y
X,
(x, y) = {C()(x) (y) | B}.
Define the fuzzy relation on X as follows: x, y X,
(x, y) = C(1{x,y} )(x).
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,
G(, )(x) = (x) ({(y) (x, y) | y X}).
Then G(, )(x) represents the greatest degree of an alternative x in relation

to the fuzzy set .
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 2.4.4 [10] If a fuzzy choice function C defined on base domain is

G-rational with rationalization , then .
The following is the fuzzy T-congruence axiom given in [10].
Fuzzy T-congruence axiom: For all x, y, z X and B,
(x, y) (y, z) C()(z) (x) C()(x)
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,
I(1 , 2 ) 1 (x) C(2 )(x) I(1 C(2 ), C(1 )).
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.
Theorem 2.4.6 (Desai and Chaudhari [17]) If a fuzzy choice function C

defined on base domain satisfies the fuzzy Chernoff axiom, then = .
Proof. Let x, y X and B. Then by definition of I, we have
I(1{x,y} , ) = (x) (y). For all x, y X, we have
(x, y) = {C()(x) (y) | B}

= {(x) (y) 1{x,y} (x) C()(x) C()(x) | B}
= {I(1{x,y} , ) 1{x,y} (x) C()(x) C()(x) | B}
{I(1{x,y} C(), C(1{x,y} )) C()(x) | B}

{C()(x) C()(x) C(1{x,y} )(x) | B}
= {C()(x) C(1{x,y} )(x) | B}
{C(1{x,y} )(x) | B}
= (x, y)
Thus . Since clearly , we have = .

In Theorem 2.4.5, we have that the fuzzy choice function C defined on the
base domain is full rational if and only if it satisfies the fuzzy T-congruence
axiom. In the remainder of this section, we present the work of Desai and
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,
(x, y) (y, z) C()(z) (x) C()(x).
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.
Lemma 2.4.7 If a fuzzy choice function C defined on a base domain satisfies

the weak fuzzy T-congruence axiom, then is max-min transitive.
Proof. Since the fuzzy choice function C is defined on base domain B and
C is normal, we have 1{x,z} B for all x, z X and also C(1{x,z} )(x) = 1
or C(1{x,z} )(z) = 1. Suppose C(1{x,z} )(x) = 1. Then (x, z) = 1. Thus is
max-min transitive. Suppose C(1{x,z} )(z) = 1. Then
(x, y) (y, z) = (x, y) (y, z) 1{x,z} (x) C(1{x,z} )(z)

C(1{x,z} )(x)
= (x, z).
Hence is max-min transitive.

Clearly, the fuzzy T-congruence axiom implies the weak fuzzy T-
congruence axiom. The following example shows the converse does not hold.
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
(x, y) (y, z) (x) C()(z)

= C(1{x,y} )(x) C(1{y,z} )(y) (x) C()(z)
= (x, y) (y, z) (x) C()(z)
(x, z) S(x) C()(z),
where the inequality holds by the max-min transitivity of . Thus
(x, y) (y, z) (x) C()(z) (x, z) (x) C()(z) (2.1)
By the max-min transitivity of and Theorem 2.4.4, we have (x, z)

C()(z) (t) (x, t) for all t X. Therefore, by Lemma 2.4.1(2), we
i i
i i
i i
K23798 2015/2/2
i i
have (x, z) C()(z) (t) (x, t) for all t X. Hence

(x, z) C()(z) (x) (x) ({(t) (x, t) | t X})
= G(, )(x)
= C()(x).
Thus by (2.1) we have that C satisfies the weak fuzzy T-congruence axiom.
Conversely, suppose that C satisfies the weak fuzzy T-congruence axiom.
Then 1{x,z} B for all x, z X. Since C is normal, we have C(1{x,z} )(x) = 1
or C(1{x,z} )(z) = 1. If C(1{x,z} )(x) = 1, then by Lemma 2.4.1(4), C(1{x,z} )(x)
1{x,z} (x) (x, z) implies C(1{x,z} )(x) (x, z) and hence (x, z) = 1
and is clearly max-min transitive. Suppose C(1{x,z} )(z) = 1. We also have
(x, y) = C(1{x,y} )(x) = (x, x) (x, y) = (x, y) for all y X. Thus
(x, y) (y, z) = (x, y) (y, z)
= (x, y) (y, z) C(1{x,z} )(z) 1{x,z} (x)
C(1{x,z} )(x)
= (x, z)
= (x, z),
where the inequality holds by the weak T-congruence axiom. Hence is max-
min transitive.
The following is a characterization theorem for full rationality of fuzzy
choice functions.
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
By Lemma 2.4.1(1), the above inequality reduces to

I(, ) (x) C()(x) (y) C()(y) (z) (y, z) for all z X.
Hence
I(, ) (x) C()(x) (y) C()(y)
(y) ({(z) (y, z) | z X})
= C()(y).
By Lemma 2.4.1(1), we have
I(, ) (x) C()(x) (y) C()(y) C()(y) for all y X
Thus I(, ) (x) C()(x) I( C(), C()).
In [25], Georgescu establishes that the fuzzy Arrow axiom and the property
of full rationality of fuzzy choice functions defined on general domain are
equivalent, [25, Theorem 4.5]. As remarked in [17], if this result is true for
fuzzy choice functions defined on the base domain, then the full rationality
of fuzzy choice functions implies FCA and the above theorem will be trivial.
But the following discussion rules out the possibility.
Fuzzy Arrow axiom (FAA): [21] For all 1 , 2 B and x X,
I(1 , 2 ) 1 (x) C(2 )(x) E(1 C(2 ), C(1 )).
In Example 2.4.8, if C(1X )(y) = s is replaced by C(1X )(y) = r, then C is
full rational with rationalization . For 1{x,w} and 1X , we have I(1{x,w} , X)
1{x,w} (x) C(1X )(x) = 1 and
E(C(1{x,w} ), 1{x,w} C(X)) 1{x,w} (w) C(1X )(w) C(1{x,w} )(w) = s.
Hence the fuzzy Arrow axiom does not hold on the base domain. Hence full
rationality does not imply the FAA on the base domain.
The following result follows from Theorems 2.4.10 and 2.4.11.
Theorem 2.4.12 Let C be a fuzzy choice function defined on a base domain.

Then C is full rational if and only if it satisfies the fuzzy Chernoff axiom and
is G-rational with transitive rationalization.
The following result follows easily from Theorems 2.4.5, 2.4.11, and 2.4.12.
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
2.5. Quasi-Transitive Rationality of Fuzzy Choice Functions 53
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.
2.5 Quasi-Transitive Rationality of Fuzzy

Choice Functions
In this section, we follow the lead of Desai [16] in a discussion of the rationality
of fuzzy choice functions. He characterized the rationality of fuzzy choice func-
tions with reflexive, strongly connected and quasi-transitive rationalization in
terms of the path independence property and the fuzzy Condorcet property.
The definition of quasi-transitivity in Definition 2.2.16 uses a t-norm . In this
section, we let = .
In order to study the rationality of a consumer, Samuelson [42] introduced
the concept of the revealed preference theory in terms of a preference rela-
tion associated with a demand function. Uzawa [49] states that the behavior
of a consumer is rational if he has a definite preference over all conceivable
commodity bundles and he chooses those commodity bundles that are opti-
mum with respect to his preference subject to the budgetary constraints. The
problem of rationality appears in many other disciplines. Consequently, the
problem of rationality has been examined in a more general framework, see
Georgescu-Rogen [28 28], Uzawa [49], Arrow [1], Sen [43, 44], Richter [40],
Hansson [29], and Suzumura [47, 48]. They introduced notions such as the
revealed preference axioms, the consistency conditions, and the congruence
axioms.
De Baets and Fodor [19] and Fodor [20] studied the case of vague prefer-
ences and exact choices. Barrett, Pattanaik, and Salles [6, 7] studied the case
when crisp choice functions are generated by the fuzzy preference relations,
see Section 2.4. In [4 4], Banerjee studied the revealed preference theory in
the context of the fuzzy set theory. In his approach, the domain of a choice
function is made only of crisp sets of alternatives and the range is made
of fuzzy sets of alternatives, Section 7.2. Georgescu [21, 22] examined fuzzy
choice functions whose domain and range are both fuzzy sets of alternatives.
She studied the rationality of fuzzy choice functions and various fuzzy revealed
preference axioms, fuzzy congruence axioms and fuzzy consistency conditions.
In the previous section, rationality of fuzzy choice functions defined on the
domain that contains characteristic functions of all single and two-element
subsets of the universal set was examined.
We recall the definition of the residuum operation on [0, 1], i.e.,
(
1 if a b
a b = {t [0, 1] | a t b} =
b if a > b
i i
i i
i i
K23798 2015/2/2
i i
for all a, b [0, 1]. The biresiduum operation on [0, 1] is defined by
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
Let be a fuzzy relation on X. In this section, the asymmetric part of

is the fuzzy relation on X defined by (x, y) = (x, y) (y, x) for all
x, y X. We think of (x, y) as showing the degree to which x is strictly
preferred to y. Recall that under this definition, is of type (0) .
Let X be a non-empty set of alternatives and B is a non-empty family of
non-zero fuzzy subsets of X. Then the pair (X, B) is called fuzzy choice space.
In this section, a fuzzy choice function on the fuzzy choice space (X, B) is a
function C : B F P(X) such that for each B, C() 6= and C() .
In the language of fuzzy consumers, X is called the set of bundles and B is
called the family of fuzzy budgets. For any fuzzy budget the real numbers
(x) and C()(x) denote the availability degree of the bundle x in the fuzzy
budget and the degree to which the bundle x is chosen from the fuzzy budget
respectively.
Georgescu extended the results of Uzawa-Arrow-Sen theory using the fol-
lowing conditions presented in Section 2.3:
(H1) Every B and C() are normal fuzzy subsets of X,
(H2) B includes all fuzzy sets 1S , where S X.
In this section, we also consider these conditions.
Let (X, B) be a fuzzy choice space and be a fuzzy relation on X. For any
B the fuzzy subsets M (, ) and G(, ) of X are defined as in Definition
2.3.3 with replaced by . Thus we have
M (, )(x) = (x) ({(z) (z, x) (x, z) | z X})

G(, )(x) = (x) ({(z) (x, z) | z X}).
We note that G(, ) M (, ) for all B and G(, ) = M (, ) if is

strongly connected. Note also that G(, ) C() for all B.
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.
Definition 2.5.3 Let C : B F(X) be a fuzzy choice function. Then C is

said to satisfy the fuzzy Condorcet property if for all B and x X,
(x) ({(z) (x, z) | z X}) C()(x).
The characterization of a choice function whose rationality is reflexive,

strongly connected, and max-min transitive has previously been determined.
In this section, transitivity is weakened to quasi-transitivity.
Definition 2.5.4 A fuzzy choice function C on (X, B) is said to be Q-

rational if there exists a reflexive, strongly connected and quasi-transitive
fuzzy preference relation on X such that C() = G(, ) for all B.
Since every max-min transitive fuzzy preference relation is quasi-transitive,
every full rational fuzzy choice function is Q-rational.
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,
(x, y) (y, z) (x, z)). (2.2)
i i
i i
i i
K23798 2015/2/2
i i
Let x, y, z X. If (y, x) > 0 or (z, y) > 0, then (y, x) = 0 and (z, y) =

0. In this case, inequality (2.2) holds. Suppose (y, x) = 0 and (z, y) = 0.
Since is strongly connected, (x, y) = 1 and (y, z) = 1. By the definition of
, it follows that C(1{x,y} ) = 1{x} and C(1{y,z} ) = 1{y} . Thus by fuzzy path
independence property, we have that
(x, y) (y, z) C(1{x,y} )(x)

= C(C(1{x} ) C(1{y,z} ))(x)
C(1{x} 1{y,z} )(x)
= C(1{x,y,z} )(x)
= C(1{x,y} 1{z} )(x)
C(C(1{x,y} ) C(1{z} ))(x)
= C(1{x} 1{z} )(x)
= C(1{x,z} )(x)
= (x, z).
Hence
(x, y) (y, z) (x, z) (2.3)
Now
(x, y) (y, z) (z, x) (z, x)

= C(1{x,z} )(z)
= C(C(1{x,y} ) C(1{z} ))(z)
C(1{x,y} 1{z} )(z)
= C(1{x,y,z} )(z)
= C(1{x,y} 1{y,z} )(z)
C(C(1{x,y} ) C(1{y,z} ))(z)
= C(1{x} 1{y} )(z)
= C(1{x,y} )(z)
= 0.
By Lemma 2.5.1(8), it follows that
(x, y) (y, z) (z, x). (2.4)
By the idempotent property of the Godel t-norm and inequalities (2.3) and
(2.4), inequality (2.2) holds.
Lemma 2.5.6 Let C be a fuzzy choice function. Then C is G-rational if and

only if it satisfies the fuzzy Condorcet property.
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
(y) (y, x) (y) (y, x)
for all B and x, y X. Thus
(x) ({(y) (y, x) | y X}) (x) ({(y) (y, x) | y X}).
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
G(, )(x) (z) (z) [(z) (x, z)]

= (z) (x, z)
(x, z)
(x, z)
(x, z).
G(, )(x) (z) (x, z) (2.5)
for all z X. Thus
G(, )(x) (x) ({(z) (x, z) | z X})

= G(, )(x)
= C()(x).
Hence C satisfies the fuzzy Condorcet property.

We now show that the fuzzy independence property and the fuzzy Con-
dorcet property are independent.
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
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.
Theorem 2.5.9 (Desai [16]) Let C be a fuzzy choice function. Then C is

Q-rational if and only if C satisfies the fuzzy path independence property and
the fuzzy Condorcet property.
Proof. Suppose that C satisfies the fuzzy path independence property
and the fuzzy Condorcet property. Then by Lemmas 2.5.5 and 2.5.6, C is
Q-rational with rationalization .
Conversely, suppose that C is Q-rational with rationalization . To prove
C satisfies the fuzzy path independence property, first we prove the following
inequality. Let , B and x X. Then for all y, z X, we have by Lemma
2.3.1(2) that
I(, ) (y) C()(y) (z) (z) [(z) (z)] C()(y)

= (z) (z) C()(y)
(z) C()(y)
= (z) G(, )(y)
(z) [(z) (y, z)]
= (z) (y, z)
(y, z).
Thus by Lemma 2.3.1(2), the above inequality reduces to
I(, ) (y) C()(y) (z) (t, z) for all z X.
Hence
I(, ) (y) C()(y) (y) ({(z) (t, z) | z X})

= C()(y).
I(, ) (y) C()(y) C()(y) for all y X.
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).
By Lemma 2.3.1(2) we have that
C( )(x) (x) C( )(x) C()(x).
By Lemma 2.3.1(2), the above inequality reduces to
C( )(x) (x) C()(x).
By the definition of , a > b, implies (a b) = b. Since C()(x) (x),we

have (x) C()(x) = C()(x). Thus C( )(x) C()(x). Similarly
we have that C( )(x) C()(x) for all x X. Consequently, C( )
C() C(). Since C() C() , it follows by inequality (2.6) and
above inclusion that
C( )(x) I(C() C(), )

I((C() C()) C( ), C(C() C()))
(C() C())(x) C( )(x) C(C() C())(x).
for all x X. By Lemma 2.3.1(1), we have
C( )(x) (C() C())(x) C( )(x) C(C() C())(x).
Since C( ) C() C(), we have that
C( )(x) C(C() C())(x) for all x X.
Therefore,
C( ) C(C() C()). (2.7)
Now for any , B and x X, we have
C(C() C())(x) = G(C() C(), )(x)

(C() C())(z) (x, z).
Since C() C() , we have by Lemma 2.5.1(2) that
(C() C())(x) (x, z) ( )(x) (x, z).
i i
i i
i i
K23798 2015/2/2
i i
Therefore the above inequality becomes

C(C() C())(x) ( )(z) (x, z)
for all z X. Therefore
C(C() C()) C( ). (2.8)
Hence inequality (2.7) and (2.8) prove that C satisfies the fuzzy path inde-
pendence property. To prove the fuzzy Condorcet property, consider B
and x X. Since C is Q-rational, we have . Also, and .
Thus . Hence by Lemma 2.5.1(2), we have
(z) (x, z) (z) (x, z)
for all z X. Thus
{(z) (x, z) | z X} {(z) (x, z) | z X}.
Hence
(x) ({(z) (x, z) | z X}) (x) ({(z) (x, z) | z X})
= G(, )(x)
= C()(x).
Consequently, C satisfies the fuzzy Condorcet property.
Georgescu [21, 22, 23, 25] has studied fuzzy choice theory in a very general
manner. She extended the results of the UzawaArrowSen theory and the
RichterHanssonSuzumura theory in the context of fuzzy set theory. The
results of the UzawaArrowSen theory were extended in the context of fuzzy
set theory under the hypotheses (H1) and (H2). The results of the Richter
HanssonSuzumura were extended without assuming (H1) and (H2).
2.6 Full Rationality and Congruence Axioms

of Fuzzy Choice Functions
We present two axioms, the fuzzy direct revelation axiom (FDRA) and the
fuzzy transitive closure coherence axiom (FTCCA). Connections between full
rationality, G-rationality, G-normality, the fuzzy congruence axiom, and the
weak fuzzy congruence axiom are also presented. This section is based on
Chaudhari and Desai [11].
Let be a fuzzy binary relation on X. Recall that is called strongly
connected if x, y X, (x, y) = 1 or (y, x) = 1. The max- transitive
closure tc of is defined as follows: x, y X,
tc (x, y) = (x, y)
{{{(x, z1 ) (z1 , z2 ) . . . (zk , y) | z1 , z2 , . . . , zk X}
| k N}.
i i
i i
i i
K23798 2015/2/2
i i
2.6. Full Rationality and Congruence Axioms of Fuzzy Choice Functions 61
Let B denote a non-empty family of non-zero fuzzy subsets of X. A fuzzy

choice function is a function C : B F P (X) such that C() for all
B. Let denote the strict fuzzy preference relation associated with .
Recall that is defined by x, y X, (x, y) = (x, y) (y, x) and that
if has no zero divisors, then is of type (0) . If tc (x, y) (y, x) for
all x, y X, then is said to be acyclic. Clearly is max- transitive is
equivalent to . Also tc is the smallest max- transitive relation
containing . If 1 2 , then tc tc
1 2 for any two fuzzy binary relations 1
and 2 on X.

Define the fuzzy binary operations on X as follows: x, y X,
(x, y) = {C()(x) (y) | B},

(x, y) = (x, y) (y, x),
(x, y) = (x, y) (y, x).
Then is called the fuzzy revealed preference relation generated by C,

is called the strict fuzzy revealed preference relation generated by
C, and is called the indifference fuzzy revealed preference relation
generated by C.

C is said to satisfy weak fuzzy congruence axiom (WFCA) if B
and x, y X, (x, y) C()(y) (x) C()(x).
Proposition 2.6.3 [24] Let C be a fuzzy choice function. If C satisfies

WFCA, then is max- transitive, reflexive, and strongly connected on X.
Proposition 2.6.4 [24] Let C be a fuzzy choice function. If C satisfies

WFCA, then condition F holds and x X,
(x) ({(z) (x, y) | z X}) C()(x).
Definition 2.6.5 Let C be a fuzzy choice function. Then C is said to satisfy

the fuzzy transitive closure coherence axiom (FTCCA) if B and
x X,
(x) ({(z) ( )tc (x, y) | z X}) C()(x),
where ( )tc is the transitive closure of .

the fuzzy congruence axiom (FCA) if , B and x, y X,
( )tc (x, y) C()(y) (x) C()(x).
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
C()(y) = G(, )(y) = (y) {(z) (y, z) | z X}

(z) (y, z) z X.
Thus C()(y)(z) (y, z) by Lemma 2.3.1(1). Hence (x, y)C()(y)

(z) (x, y) (y, z) (x, z) for all z X. By Lemma 2.3.1(1), we have
(x, y) C()(y) (z) (x, z).
Hence (x, y) C()(y) {(z) (x, z) | z X}. Thus
(x, y) C()(y) (x) (x) {(z) (x, z) | z X}

= G(, )(x).
Hence (x, y) C()(y) (x) C()(x) since C is G-rational. Thus the

desired result follows by Lemma 2.5.6.

C is said to satisfy fuzzy direct relation axiom (FDRA) if B and
x, y X, (x, y) C()(y) (x) C()(x).
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
(x) {(z) (x, z) | z X} (x, y) C()(y) (x).
Since C satisfies WFCA, we have that
(x, y) C()(y) (x) C()(x).
Therefore, (x) {(z) (x, z) | z X} C()(x).
Lemma 2.6.10 Let C be a fuzzy choice function. If C satisfies FTCCA, then

C is G-normal.
i i
i i
i i
K23798 2015/2/2
i i
2.6. Full Rationality and Congruence Axioms of Fuzzy Choice Functions 63
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.
Lemma 2.6.11 Let C be a fuzzy choice function and be a fuzzy preference

relation on X. If is a max- transitive G-rationalization of C, then ( )tc
.
Proof. Now
( )tc (x, y)
= (x, y) ({{ (x, z1 ) . . . (zk , y) | zi X, i = 1, . . . , k} | k N}).
Thus by Lemma 2.5.6,
( )tc (x, y)
= (x, y) ({{(x, z1 ) . . . (zk , y) | zi X, i = 1, . . . , k} | k N})
(x, y) ({{(x, y) | zi X, i = 1, . . . , k} | k N}) = (x, y).
Hence ( )tc .
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
Thus G(, ) C() and the desired result follows.
Theorem 2.6.13 Let C be a fuzzy choice function. If C satisfies WFCA and

FTCCA, then C is full rational.
Proof. By Proposition 2.6.3, is reflexive, strongly complete, and tran-
sitive. Since C satisfies FTCCA, C is G-rational by Theorem 2.6.12. Thus C
is full rational.
tion. If C satisfies FDRA and is max- transitive, then C satisfies WFCA.
Proof. Let B and x, y X. Then for all z X,
(x, y) C()(y) (z) (x, y) (y, z) (x, z).
Thus (x, y) C()(y) (z) (x, z). By Lemma 2.3.1(1),
(x, y) C()(y) (x) (x) ({(z) (x, z) | z X})
and so the desired result holds.
Theorem 2.6.15 Let C be a fuzzy choice function. If C satisfies FDRA and

is max- transitive, then C satisfies conditions F and F .
Proof. The proof follows from Theorem 2.6.14 and Proposition 2.6.4.
Theorem 2.6.16 Let C be a fuzzy choice function. Then C is G-rational if

and only if C satisfies FDRA.
Proof. Suppose that C is G-rational with rationalization . By Lemma

2.5.6, . Thus (y) (x, y) (y) (x, y) for all y X. Thus
(y) ({(y) (x, y) | y X}) (x) ({(y) (x, y) | y X})

= G(, )(x) = C()(x).
Conversely, suppose C satisfies FDRA. Then (y) ({(y) (x, y) |

y X}) C()(x) and so G(, )(x) C()(x) for all B and x X.
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
Proof. Suppose C satisfies FTCCA. Let B and x, y, z X. By the

definition of , (y, z) C()(y) (z). Thus for all z X,
( )tc (x, y) C()(y) (z) ( )tc (x, y) (y, z)
( )tc (x, y) ( )tc (y, z)
( )tc (x, z).
Hence ( )tc (x, y) C()(y) (z) ( )tc (x, z) by Lemma 2.5.1(3). Thus
( )tc (x, y) C()(y) {(z) ( )tc (x, z) | z X}. Hence
( )tc (x, y) C()(y) (x) (x) ({(z) ( )tc (x, z) | z X})
C()(x)
since C satisfies FTCCA.
Conversely, suppose C satisfies FCA. Let B and x X. Since C is
normal, there exists y X such that C()(y) = 1. Thus (y) = 1. Now
{(z) ( )tc (x, z) | z X} (y) ( )tc (x, y) ( )tc (x, y). Hence
(x)({(z) ( )tc (x, z) | z X}) ( )tc (x, t)C()(t)(x) C()(x)
since C satisfies FCA.
In this section, we introduced the notions of FDRA and FTCCA given in
[1]. They are related to each other in the sense that a fuzzy choice function
satisfying FDRA with max- transitive rationalization satisfies FTCCA and
conversely. Also, if a fuzzy choice function satisfies WFCA and FTCCA, then
it is full rational.
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.
6. [35] Let be a fuzzy preference relation on a finite set X. Prove that

{{(x, y) | y X} | x X} > 0 if either of the following two
conditions hold:
(1) is complete and -acyclic and has no zero divisors.
(2) is strongly complete and -acyclic.
7. [35] If a fuzzy preference relation on a finite set X is acyclic, prove

that {{(y, x) (x, y) | y X} | x X} > 0.
8. [35] Let be a fuzzy preference relation on a finite set X. Prove that
G( , ) is a fuzzy choice function if either of the following conditions
hold:
(1) is complete and -acyclic, where is continuous and without zero
divisors.
(2) is strongly complete and -acyclic, where is continuous.
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
5. C. R. P. Barrett, K. Pattanaik and M. Salles, On the structure of fuzzy

social welfare functions, Fuzzy Sets and Systems, 19 (1986) 111.
6. C. R. Barrett, P. K. Pattanaik and M. Salles, On choosing rationally

when preferences are fuzzy, Fuzzy Sets and Systems, 34 (1990) 197212.
7. C. R. Barrett, P. K. Pattanaik and M. Salles, Rationality and aggre-
gation of preferences in an ordinal fuzzy framework, Fuzzy Sets and
Systems, 49 (1992) 913.
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.
13. S. R. Chaudhari and S. S. Desai, On interrelations between fuzzy con-

gruence axioms, New Mathematics and Natural Computation, 8 (2012)
297-310.
14. T. D. Clark, J. M. Larson, J. N. Mordeson, and M. J. Wierman, Ap-
plying Fuzzy Mathematics to Formal Modeling in Comparative Politics,
(Springer-Verlag, Berlin 2008).
15. T. D. Clark, J. M. Larson, J. N. Mordeson, and M. J. Wierman, Exten-
sion of the portfolio allocation model to surplus majority governments:
A fuzzy approach, Public Choice, 134 (2008) 179199.
16. S. S. Desai, On quasi-transitive rational fuzzy choice functions, New

Mathematics and Natural Computation, 10 (2014) 91102.
17. S. S. Desai and S. R. Chaudhari, On the full rationality of fuzzy choice
functions on base domains, New Mathematics and Natural Computation,
8 (2012) 183193.
i i
i i
i i
K23798 2015/2/2
i i
18. S. S. Desai and S. R. Chaudhari, On interrelations between fuzzy con-

gruence axioms through indicators, Annals of Fuzzy Mathematics and
Informatics, to appear.
19. B. De Baets and J. Fodor, Twenty years of fuzzy preference relations
(19781997), Belgian Journal of Operations Research, Statistics and
Computer Science, 37 (1997) 6182.
20. J. Fodor and M. Roubens, Fuzzy Preference Modelling and Multicriteria
Decision Support, (Kluwer, 1994).
21. I. Georgescu, On the axioms of revealed preference in fuzzy consumer
theory, Journal of Systems Science and Systems Engineering, 13 (2004)
279296.
22. I. Georgescu, Consistency conditions in fuzzy consumer theory, Funda-

menta Informaticae, 61 (2004) 223245.
23. I. Georgescu, Revealed preference, congruence and rationality: A fuzzy
approach, Fundamenta Informaticae, 65 (2005) 307328.
24. I. Georgescu, Fuzzy Choice Functions: A Revealed Preference Approach,
Springer-Berlin, 2007.
25. I. Georgescu, Arrows axiom and full rationality for fuzzy choice func-
tions Social Choice and Welfare, 28 (2007) 303319.
26. I. Georgescu, Acyclic rationality indicators of fuzzy choice functions
27. I. Georgescu, Similarity of fuzzy choice functions, Fuzzy Sets and Sys-
tems, 158 (2007) 13141326.
28. I. Georgescu and N. Roegen, Choice and revealed preference Southern
Economic Journal, 21 (1954) 119130.
29. B. Hansson, Choice structures and preference relations Synthese, 18

(1968) 443458.
30. J. Kacprzyk, M. Fedrizzi, and H. Nurmi, Group decision making and
consensus under fuzzy preferences and fuzzy majority, Fuzzy Sets and
Systems 49 (1992) 2131.
31. K. P. Klement, R. Mesiar and E. Pap, Triangular Norms, (Kluwer,
2000).
32. G. J. Klir and B. Yuan, Fuzzy Sets and Fuzzy Logic: Theory and Appli-
cations, Prentice Hall, New York, 1995.
i i
i i
i i
K23798 2015/2/2
i i
2.8. References 69
33. P. Kulshreshtha and B. Shekar, Interrelationship among fuzzy prefer-

ence based choice functions and significance of rationality conditions: A
taxonomic and intuitive perspective, Fuzzy Sets and Systems, 109 (2000)
429445.
34. E. Laver (ed), Estimating the Policy Position of Political Actors, Lon-
don: Routledge, 2001.
35. D. Martinetti, B, de Baets, S. Diaz, and S. Montes, On the role of
acyclicity in the study of rationality of fuzzy choice functions, Fuzzy
Sets and Systems, 239 (2014) 3550.
36. J. N. Mordeson, K. R. Bhutani and T. D. Clark, The rationality of fuzzy
choice function, New Mathematics and Natural Computation, 4 (2008)
309327.
37. S. A. Orlovsky, Decision-making with a fuzzy preference relation, Fuzzy
38. C. R. Plott, Path Independence, Rationality, and Social Choice, Econo-
metrica, 41 (1973) 10751091.
39. K. T. Poole, Spatial Models of Parliamentary Voting, Cambridge: Cam-
bridge University Press 2005.
40. M. Richter, Revealed preference theory, Econometrica, 34 (1966) 635
645.
41. P. A. Samuelson, Consumption theory in terms of revealed preference,
Econometrica, 15 (1948) 243253.
42. P. A. Samuelson, A note on the pure theory of consumers behavior,
Economica, 5 (1938) 6171.
43. A. K. Sen, Quasi-transitivity, rational choice and collective decisions,
Review of Economic Studies, 36 (1969) 381393.
44. A. K. Sen, Choice functions and revealed preference, Review of Economic
Studies, 38 (1971) 307312.
45. A. K. Sen, Social choice theory: A re-examination, Econometrica, 45
(1977) 5389.
46. A. K. Sen, Internal consistency of choice, Econometrica ,61 (1993) 495
521.
47. K. Suzumura, Rational choice and revealed preference, Review of Eco-
nomic Studies, 43 (1976) 149159.
48. K. Suzumura, Rational Choice, Collective Decisions and Social Welfare,
Cambridge University Press, 1983.
i i
i i
i i
K23798 2015/2/2
i i
49. H. Uzawa, A note on preference and axioms of choice, Annals of the

Institute of Statistical Mathematics, 8 (1956) 3540.
50. X. Wang, A note on congruence conditions of fuzzy choice functions,

i i
i i
i i
K23798 2015/2/2
i i
Chapter 3
Factorization of Fuzzy
Preference Relations
We show that there is an inclusion reversing correspondence between conorms

and fuzzy strict preference relations in factorizations of fuzzy preference re-
lations into their strict preference and indifference components. We also as-
sociate various conorms with asymmetric components in a factorization of a
fuzzy preference relation.
3.1 Basic Definitions and Results

The factorization R = P I of a preference relation R into a strict prefer-
ence relation P and an indifference relation I is unique in the crisp case. In
the fuzzy setting, there are several factorizations of a fuzzy weak preference
relation that generalize the crisp case. In Fono and Andjiga [12], classical fac-
torization of a fuzzy relation into a symmetric component (indifference) and
an asymmetric component (regular fuzzy strict preference) were generalized.
The results were used to obtain fuzzy versions of Gibbards oligarchy theorem
and Arrows impossibility theorem. Previous fuzzy versions of Gibbarda oli-
garchy theorem and Arrows impossibility theorem were established in Dutta
[11] and Richardson [24]. This factorization is important in the examination of
Arrowian type results. In Banerjee [3], it was shown using factorization that
the fuzzy analog of the general possibility theorem is valid under relatively
weak transitivity restrictions. Other works can be found in [4, 5, 8, 9, 17, 18,
19, 26].
Proposition 3.1.1 Let , , FR(X). Suppose that (i) Supp() =

Supp() Supp(), (ii) is asymmetric, and (iii) is symmetric. Then the
following properties hold.
71
i i
i i
i i
K23798 2015/2/2
i i
72 3. Factorization of Fuzzy Preference Relations
(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.
Proposition 3.1.2 Let , , FR(X). Suppose that (i) Supp() =

Supp() Supp(), (ii) is asymmetric, and (iii) is symmetric. Then (1)
(2) (3), where
(1) Supp() Supp() = ;
(2) x, y X, (x, y) > 0 implies (x, y) > 0 and (y, x) = 0;
(3) x, y X, (x, y) = (y, x) implies (x, y) = (y, x) = 0.
Proof. (1) (2) : Let x, y X. Suppose (x, y) > 0. Since Supp()
Supp() = , (x, y) = (y, x) = 0. Suppose (x, y) > 0 and (y, x) > 0. Then
by (i), (y, x) > 0, but this is impossible since (x, y) > 0. Thus (y, x) = 0
and so (x, y) = (y, x) > 0 by (i), but this is also impossible. Hence not
both (x, y) > 0 and (y, x) > 0. By (i), (x, y) > 0 since (x, y) > 0 Thus
(x, y) > 0 and (y, x) = 0.
(2) (1) : Suppose Supp() Supp() 6= . Let (x, y) Supp()
Supp(). Then (x, y) > 0 and (y, x) = (x, y) > 0. By (2), (x, y) > 0
and (y, x) = 0. However, (y, x) = 0 and (y, x) > 0 is impossible by (i).
Thus Supp() Supp() = .
(2) (3) : Let x, y X. Suppose (x, y) = (y, x). If either (x, y) > 0
or (y, x) > 0, then by (2) (x, y) 6= (y, x). Hence (x, y) = (y, x) = 0.
Note: [(x, y) = (y, x) > 0 (x, y) = (y, x) = 0] Supp()
Supp() = .
Let FR(X) and be an asymmetric fuzzy binary preference relation
associated with . Recall that is called simple if x, y X, (x, y) = (y, x)
implies (x, y) = (y, x).
As previously mentioned, the first appearance of t-norms appeared in the
context of probabilistic metric spaces Schweizer and Sklar [25]. This was fol-
lowed by their use as an interpretation of the conjunction in the semantics of
i i
i i
i i
K23798 2015/2/2
i i
3.1. Basic Definitions and Results 73
fuzzy mathematical logics Hajek [14]. The notion of a t-conorm is a natural

companion to the concept of a t-norm in that a t-conorm is the interpretation
of the disjunction in fuzzy logic. These operators are applied in fuzzy control
to formulate assumptions of rules as the fuzzy intersection of fuzzy subsets.
There are many ways to model these connectives. The combining of these
connectives in logical statements, called data fusion (combining of evidence),
is crucial in building expert systems. The types of connectives used are appli-
cation dependent. For example, connectives used in data fusion are in medical
science and geophysics. We next present the definition of a t-conorm.
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).
Proposition 3.1.4 Let , , FR(X). Suppose is a conorm. Suppose

that (i) = , (ii) is asymmetric, and (iii) is symmetric. Then x, y
X, (x, y) = (x, y) (y, x).
Proof. Let x, y X. Since = , (x, y) (x, y) and (y, x) (y, x)
by the union axioms. Since is symmetric, (x, y) = (y, x) (x, y) (y, x).
Since is asymmetric, either (x, y) = 0 or (y, x) = 0, say (y, x) = 0. Then
(y, x) = (y, x). Thus (x, y) = (x, y) (y, x).
Proposition 3.1.5 Let , , FR(X). Suppose is a conorm. Suppose

that (i) = , (ii) is asymmetric, and (iii) is symmetric. If is
simple, then x, y X, (y, x) (x, y) if and only if (y, x) = 0.
Proof. Suppose (y, x) (x, y), but (y, x) > 0. Then since is
asymmetric, (x, y) = 0. Since (y, x) > (x, y) and (y, x) = (x, y),
(y, x) (x, y). Hence (y, x) = (x, y) and so (y, x) = (x, y) since
is simple, a contradiction. Thus (y, x) = 0. Conversely, suppose (y, x) = 0.
Then (y, x) = (y, x) = (x, y) (y, x) and so (y, x) (x, y).
Propositions 3.1.4 and 3.1.5 provide intuitively appealing results for and
when = .
Definition 3.1.6 Define the conorm 2 on [0, 1] by a, b [0, 1], a 2 b =

1 (a + b) for all a, b [0, 1].
The following factorization was given by Orlovsky [22] and characterized
by Richardson [24].
i i
i i
i i
K23798 2015/2/2
i i
Proposition 3.1.7 Let , , FR(X). Suppose that (i) = 2 , (ii)

is asymmetric, and (iii) is symmetric. Then x, y X, (x, y) + (x, y)
1 (x, y) = 0 ((x, y) (y, x)).
Proof. Let x, y X. Suppose (x, y) + (x, y) 1. Then (x, y) = ( 2
)(x, y) = 1((x, y)+(x, y)) by (i). By hypothesis, (x, y) = (x, y)+(x, y)
and so
(x, y) = (x, y) (x, y) = (x, y) (x, y) (y, x)

= ((x, y) (x, y)) ((x, y) (y, x))
= 0 ((x, y) (y, x)).
Conversely, suppose (x, y) = 0 ((x, y) (y, x)). Then by Proposition

3.1.4,
(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 .
Proposition 3.1.8 Suppose = max. Let , , FR(X). Suppose that (i)

= , (ii) is asymmetric, and (iii) is symmetric. Then x, y X,
x) (x, y) = (y, x)]

[(x, y) = (y,
(x, y) if (x, y) > (y, x),
(x, y) =
0 otherwise.
Proof. Let x, y X. Suppose (x, y) = (y, x) (x, y) = (y, x). By

(i), (x, y) (x, y). Suppose (x, y) > (y, x). Assume (x, y) > (x, y).
Then since = max, (x, y) = (x, y) = (y, x) = (y, x), where the last
equality holds by Proposition 3.1.4. However, this is a contradiction. Thus
(x, y) = (x, y). Now suppose (x, y) (y, x). If (x, y) = (y, x), then
(x, y) = (y, x) and so (x, y) = (y, x) = 0 by (iii). If (x, y) < (y, x),
then it has just been shown that (y, x) = (y, x) > 0. Hence (x, y) = 0 by
(iii).
Conversely, suppose

(x, y) if (x, y) > (y, x),
(x, y) =
0 otherwise.
i i
i i
i i
K23798 2015/2/2
i i
3.1. Basic Definitions and Results 75
Suppose (x, y) = (y, x). Then (x, y) = (y, x) = 0 by the hypothesis.

Let FR(X). Recall that is strongly complete if x, y X, (x, y) +
(y, x) 1.
Proposition 3.1.9 Let = max. Let , , F R(X). Suppose that (i) =

, (ii) is asymmetric, and (iii) is symmetric. If is strongly complete
and Supp() Supp() = , then
(1) x, y X, (x, y) > (x, y) (x, y), (y, x) {0, 1};
(2) x, y X, (x, y) > (x, y) (x, y) = (y, x) = (x, y) = (y, x);
(3) x, y X, (x, y) = (x, y) (x, y), (y, x) {0, 1}.
Proof. Let x, y X. Then (x, y) = (x, y)(x, y) and (y, x) = (y, x)

(y, x).
(1) Suppose (x, y) > (x, y). Then (x, y) = (x, y). Thus (x, y) = 0
by hypothesis. Hence (y, x) = 0. By (ii), (y, x) = 0. Thus (y, x) = 0.
Since is strongly complete, (x, y) + (y, x) 1. Hence (x, y) = 1. Thus
(x, y), (y, x) {0, 1}.
(2) Suppose (x, y) > (x, y). Then by hypothesis, (x, y) = 0. Also,
(y, x) = (x, y) > 0 and so (y, x) = 0 by hypothesis. Hence (x, y) =
(x, y) = (y, x) = (y, x).
(3) Finally, suppose (x, y) = (x, y). Then (x, y) = (x, y) = (x, y) = 0,
where the last equality holds since Supp() Supp() = . Since is strongly
complete, (y, x) = 1. Thus (x, y), (y, x) {0, 1}.
In the following example, we illustrate the previous proposition.
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() = .
Banerjee proved the next factorization result [3].
Proposition 3.1.11 Let , , FR(X). Suppose that (i) = 2 , (ii)

is asymmetric, and (iii) is symmetric. If (iv) x, y X, (x, y) < 1 implies
(y, x) > 0 and (v) (x, y) + i(x, y) 1, then x, y X, (x, y) = 1 (y, x)
and (x, y) = (x, y) (y, x).
Proof. Let x, y X. It follows that 2 satisfies the union axioms. Hence
i(x, y) = (x, y) (y, x) by Proposition 3.1.4. By (i) and (v), it follows that
(x, y) = (x, y) + (x, y). Suppose (y, x) = 1. Then (x, y) (y, x). Thus
by Proposition 3.1.5, (x, y) = 0. Hence (x, y) = (x, y). Thus (x, y) =
1 (y, x). Suppose (y, x) < 1. Then by (iv), (x, y) > 0. Thus (y, x) = 0.
By (iv), (x, y) = 1. Thus (x, y) > (y, x) and so (x, y) = (y, x). Hence
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}.
The following result is crucial to the factorization results.
Proposition 3.2.1 (Fono and Andjiga [12]) Let be a FWPR. Let x, y X.

Suppose (x, y) > (y, x). Then the following statements hold.
(1) The equation
(x, y) = (y, x) t, t [0, 1] (3.1)
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
Proposition 3.2.2 Let be FWPR on X and let and be fuzzy binary

relations on X. 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),
(ii) (x, y) (y, x) (x, y) = 0,
(iii) (x, y) > (y, x) (x, y) is a solution to Eq. (3.1).
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).
Proof. By Proposition 3.2.2, it suffices to show that (2) of Proposition

3.2.2 is equivalent to (2) of Corollary 3.2.3. By the hypothesis concerning
and Proposition 3.2.1, (y, x) (x, y) is the unique solution to Eq. (3.1).
Assume (2) of Corollary 3.2.3 holds. Then (x, y) = (y, x) (x, y). Now
(x, y) > 0 if and only if (x, y) > (y, x) by the definition of . Thus (2) of
Proposition 3.2.2 holds. Assume (2) of Proposition 3.2.2 holds. Then (x, y)
is a solution to Eq. (3.11). Hence (x, y) = (y, x) (x, y) since Eq. (3.1)
has a unique solution. Thus (2) of Corollary 3.2.3 holds.
Let d : [0, 1]2 [0, 1] be defined by a, b [0, 1],

a if b = 0,
d (a, b) = b if a = 0,
1 otherwise.

Then d is called the drastic t-conorm.
i i
i i
i i
K23798 2015/2/2
i i
Let A : [0, 1]2 [0, 1] be defined by a, b [0, 1], A (a, b) = a + b ab.

Then A is a t-conorm called the algebraic sum.
Let Y : [0, 1]2 [0, 1] be defined by a, b [0, 1], Y (a, b) = 1 (a +
1/
b ) , where > 0. Then Y is a t-conorm in the Yager class.
In general, suppose (x, y) = (y, x). Then (x, y) = (x, y) t and t = 0
is a solution. Thus if g(t) = (x, y) t is strictly increasing in t, then t = 0
is the only solution to Eq. (3.11). Hence we would define (x, y) = 0 in this
case.
Proposition 3.2.4 Let = A . Then the solution to Eq. (3.1) is given as

follows: x, y X,
(
(x,y)(y,x)
1(y,x) if (x, y) > (y, x),
(x, y) =
0 otherwise.
We let = (4) in this case.

Proof. Let x, y X. Suppose (x, y) > (y, x). Then (x, y) = (y, x)
t = (y, x) + t (y, x)t. Thus (x, y) (y, x) = t(1 (y, x)). Hence the
desired result now follows easily.
Proposition 3.2.5 Let = Y . Then the solution to Eq. (3.1) is given as

follows: x, y X,

((x, y) (y, x) )1/ if (x, y) > (y, x),
(x, y) =
0 otherwise.
We let = (5) in this case.

Proof. Let x, y X. Suppose (x, y) > (y, x). Then (x, y) = (y, x)
t = 1 ((y, x) + t )1/ . Suppose ((y, x) + t )1/ 1. Then (x, y) =
((y, x) + t )1/ . Hence (x, y) = (y, x) + t and so t = ((x, y)
(y, x)w )1/ . Thus the desired result follows easily.
Proposition 3.2.6 Let be a FWPR. Let d denote the drastic conorm.

Then has a factorization = (0) d if and only if x, y X, x 6= y, either
(x, y) = (y, x) or not both (x, y) > 0, (y, x) > 0.
Proof. Suppose = d . Let x, y X be such that x 6= y. Since
is complete, either (x, y) > 0 or (y, x) > 0, say (x, y) > 0. Suppose
(y, x) > 0. Then (0) (x, y) = 0 = (0) (y, x). Thus (x, y) = (x, y) and
(y, x) = (y, x). Hence (x, y) = (y, x).
Conversely, suppose either (x, y) = (y, x) or not both (x, y) > 0 and
(y, x) > 0. Suppose (x, y) = (y, x). Then (0) (x, y) = 0 = (0) (y, x). Thus
0d (x, y) = (x, y) = (y, x). Hence = (0) d . Suppose not both (x, y) >
i i
i i
i i
K23798 2015/2/2
i i
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.
Proposition 3.2.7 Let be a FWPR. Let be a solution to Eq. (3.1). Then

the following statements hold.
(1) If = d , then = (0) .
(2) If = , then = (1) .
(3) If = 2 , then = (3) .
(4) If = 2 , and (iv) and (v) of Proposition 3.1.11 hold, then = (2) .
(5) If = A , then = (4).
(6) If = Y , then = (5) .
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,
(x, y) = {t [0, 1] | (y, x) 2 t (x, y)}

= {t [0, 1] | 1 ((y, x) + t) (x, y)} = t ,
where t = 0 ((x, y) (y, x)). Hence = (3) .

(4) The desired result follows from Proposition 3.1.11.
Let be a FWPR and an asymmetric fuzzy binary relation associated
with . Recall that is said to be regular if x, y X, (x, y) > 0 if and only
if (x, y) > (y, x).
It follows easily that (1) , (3) , (4) , and (5) are regular.
Let n : [0, 1]2 [0, 1] be defined by a, b [0, 1],

a b if a + b < 1,
n (a, b) = .
1 otherwise
i i
i i
i i
K23798 2015/2/2
i i
Then n is a t-conorm called the nilpotent maximum conorm.

Let H2 : [0, 1]2 [0, 1] be defined by a, b [0, 1],
2 (a, b) = (a + b)/(1 + ab).
Then H2 is a t-conorm called the Einstein sum.

Let p : [0, 1]2 [0, 1] be defined by a, b [0, 1],
1
p (a, b) = 1 0 ((1 a)p + (1 b)p 1) p , p 6= 0.
Then A is a t-conorm called the Schweizer and Sklar conorm.

Let : [0, 1]2 [0, 1] be defined by a, b [0, 1],
1
(a, b) = 1 (a + b ) , > 0.
Then A is a t-conorm called the Yager conorm.

Let s : [0, 1]2 [0, 1] be defined by a, b [0, 1],
(s1a 1)(s1b 1)

s (a, b) = 1 logs 1 + , s > 0, s 6= 1.
s1
Then s is a t-conorm called the Frank conorm.
Proposition 3.2.8 Let be a FWPR. Let n denote the nilpotent conorm.

Then has a factorization = n if and only if x, y X, either (1)
(x, y) = 1 or (y, x) = 1, or (2) (x, y) = (y, x), or (3) (x, y) + (y, x) < 1
and (x, y) 6= (y, x).
Proof. Suppose not (1) or (2) or (3). Then there exists x, y X such that
(x, y) 6= (y, x), (x, y) < 1, (y, x) < 1, and (x, y) + (y, x) 1. Suppose
(x, y) > (y, x). Then (x, y) = (y, x). Suppose there exists such that
= n . Then (x, y) = (x, y) n (x, y) 6= 1 and so (x, y) = (x, y).
However, this is impossible since (x, y) n (y, x) 1 and so (x, y) = 1.
Thus no such exists.
Conversely, suppose (x, y) = (y, x). Then let (x, y) = 0 and we have
(x, y) = 0 n (x, y) = (x, y) n (x, y). Suppose (x, y) = 1 or (y, x) = 1
and (x, y) 6= (y, x), say (x, y) = 1. Then let (x, y) = 1 and (y, x) = 0.
Then (x, y) = 1 = 1n (y, x) = (x, y)n (x, y). Suppose (x, y)+(y, x) <
1 and (x, y) 6= (y, x), say (x, y) > (y, x). Then let (x, y) = (x, y) and so
(x, y) = (x, y) n (y, x) = (x, y) n (x, y). We can let (y, x) [0, (y, x)]
here.
We next determine results for similar to those in Proposition 3.2.7 for
other conorms.
i i
i i
i i
K23798 2015/2/2
i i
Proposition 3.2.9 Let be a FWPR. Let be a solution to Eq. (3.1). Then

(1) If = H2 , then is defined by x, y X,
(
(x,y)(y,x)
(x, y) = 1+(x,y)((y,x) if (x, y) > (y, x),
0 otherwise.
(2) If = p , then is defined by x, y X,

1
1 [(1 (x, y))p (1 (y, x))p + 1] p if (x, y) > (y, x),
(x, y) =
0 otherwise.
(3) If = , then is defined by x, y X,

1
((x, y) (y, x) ) if (x, y) > (y, x),
(x, y) =
0 otherwise.
(4) If = s , then is defined by x, y X,
(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).
(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).
(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.
Proposition 3.2.10 Let C = { | is a continuous conorm} and let

A = { F R(X) | is asymmetric}. Then there exists a function f of
C FR(X) into A such that if f (, ) = f (1 , ) FR(X), then = 1 .
Furthermore, Im(f ) = { A | (, ) C FR(X), = }.
Proof. Let (, ) C F R(X) By Proposition 3.2.1, A such that
= . Define f : C FR(X) A by (, ) C F R(X), f ((, )) =
, where = . Clearly the domain of f is C FR(X). Suppose = 1 .
Then x, y X, (x, y) = {t [0, 1] | t (y, x) (x, y)} = 1 (x, y),
where the latter equality holds since = 1 . Hence f is well defined.
Suppose that f (, ) = f (1 , ) FR(X). Let a, b [0, 1]. Let x, y
X and F R(X) be such that (x, y) = a b and (y, x) = b. Let t =
{t [0, 1] | t b a b}. Then t = {t [0, 1] | t b = a b}. Hence t a.
By hypothesis, (x, y) = (x, y) (x, y) = (x, y) 1 (x, y). Thus since
(x, y) = t , (x, y) = a b = t b = t 1 b a 1 b, where the inequality
holds since t a. Hence a b a 1 b. By symmetry, a 1 b a b. Thus
a b = a 1 b.
Let 1 and 2 be fuzzy preference relations and let 1 and 2 be conorms.
We now determine under which two factorizations of , say = 1 1 =
i i
i i
i i
K23798 2015/2/2
i i
2 2 have the property that 1 2 for all such factorizations if and only
if 1 2 .
Proposition 3.2.11 Let 1 and 2 C. Let FR(X). Suppose that =

1 1 = 2 2 . Suppose further that either 1 or 2 is strict. If 1 2 ,
then 1 2.
Proof. Suppose that there exist x, y [0, 1] such that 1 (x, y) > 2 (x, y).
Since a 1 b a 2 b for all a, b [0, 1], we have
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
(x, y) = 1 (x, y) 1 (x, y) > 2 (x, y) 1 (x, y)

2 (x, y) 2 (x, y) = (x, y)
which is impossible. Thus no such x, y exist. Hence 1 2.

Suppose 2 strict. Then 1 (x, y) 2 (x, y) > 2 (x, y) 2 (x, y). Thus
(x, y) = 2 (x, y) 2 (x, y) < 1 (x, y) 2 (x, y)

1 (x, y) 1 (x, y) = (x, y)
which is impossible. Hence no such x, y exist. Thus 1 2.
Proposition 3.2.12 Let 1 and 2 C. Suppose for all FR(X) that

= 1 1 = 2 2 . If 1 2 , then 1 2 for all such factorizations.
Proof. Suppose that 1 2 . Suppose there exists FR(X) and
some x.y X such that 1 (x, y) > 2 (x, y). We have 2 (x, y) 1 (x, y)
2 (x, y)2 (x, y). Also 1 (x, y)1 (x, y) > 2 (x, y)1 (x, y), where the strict
inequality holds by the minimality of 1 . Hence (x, y) = 1 (x, y)1 (x, y) >
2 (x, y) 1 (x, y) 2 (x, y) 2 (x, y) = (x, y), which is a contradiction.
Thus 1 2 .

= 1 1 = 2 2 . Suppose further that either 1 or 2 is strict. If 1 2
for all such factorizations, then 1 2.
Proof. Suppose there exist a, b [0, 1] such that a 1 b < a 2 b.
Suppose 1 is strict. Let FR(X) be such that there exist x, y X such
that (x, y) = a 1 b and (y, x) = b. Then (x, y) (y, x). Thus (x, y) = b.
Since 1 is strict, 1 (x, y) = a. Thus (x, y) = 1 (x, y) 1 (x, y) = a 1 b <
a 2 b 2 (x, y) 2 (x, y) = (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
Suppose 2 is strict. Let FR(X) be such that there exist x, y X such

that (x, y) = a 2 b and (y, x) = b. Then (x, y) (y, x). Thus (x, y) = b.
Since 2 is strict, 2 (x, y) = a. Thus (x, y) = 1 (x, y) 1 (x, y) a 1 b <
a 2 b = 2 (x, y) 2 (x, y) = (x, y), a contradiction. Hence no such a, b exist.
Thus 1 2 .
Let be a conorm. Then is said to be nearly strictly increasing if
b [0, 1], there exists a0 (0, 1] such that a a0 , a b = 1 and g is strictly
increasing on [0, a0 ].

= 1 1 = 2 2 . Suppose further that either 1 or 2 is nearly strict.
If 1 2 for all such factorizations, then 1 2.
Proof. Suppose there exist a, b [0, 1] such that a 1 b < a 2 b.
Suppose 1 is nearly strict. Let FR(X) be such that there exist
x, y X such that (x, y) = a 1 b and (y, x) = b. Then (x, y) (y, x).
Thus (x, y) = b. Suppose a < a0 , where a0 = {t [0, 1] | t 1 b = 1}. Since
1 is nearly strict, 1 (x, y) = a. Thus (x, y) = 1 (x, y) 1 (x, y) = a 1 b <
a 2 b 2 (x, y) 2 (x, y) = (x, y), a contradiction. Hence no such a, b exist.
Thus 1 2 . Suppose a a0 . Then a 1 b = 1 and so a 1 b < a 2 b is
impossible. Once again no such a, b exist. Hence 1 2 .
Suppose 2 is nearly strict. Let FR(X) be such that there exist
x, y X such that (x, y) = a 2 b and (y, x) = b. Then (x, y) (y, x).
Thus (x, y) = b. Suppose a < a0 , where a0 = {t [0, 1] | t 2 b = 1}. Since
2 is nearly strict, 2 (x, y) = a. Thus (x, y) = 1 (x, y) 1 (x, y) a 1 b <
a 2 b = 2 (x, y) 2 (x, y) = (x, y), a contradiction. Hence no such a, b exist.
Thus 1 2 Suppose a a0 . Then a 2 b = 1 and 2 (x, y) a0 . Hence
1 = (x, y) = 2 (x, y) 2 b = a 2 b > a 1 b 2 (x, y) 1 b 1 (x, y) 1 b =
(x, y), a contradiction. Hence no such a, b exist. Thus 1 2 .
Proposition 3.2.15 Let be a FWPR. If is regular, then is simple.
Proof. Let x, y X. Suppose (x, y) = (y, x). Then since is regular,

(x, y) = 0 = (y, x).
Proposition 3.2.16 Let be a FWPR. If = is a factorization of

and is strictly increasing, then is regular.
Proof. Let x, y X. Suppose (x, y) > (y, x). Then (x, y) = (y, x).
Hence (x, y) > 0 else (x, y) = 0 (y, x) = (y, x) which is impossible.
Conversely suppose (x, y) > 0. Suppose (x, y) (y, x). Then (x, y) =
(x, y). Thus (x, y) = (x, y) (x, y) > 0 (x, y) = (x, y) which is
impossible. Thus (x, y) > (y, x).
i i
i i
i i
K23798 2015/2/2
i i
3.3. Factorizations 85
Proposition 3.2.17 Let is a FWPR. Suppose = = are

factorizations of , where denotes maximum. If is regular, then .
Proof. Let x, y X. By Proposition 3.1.8,

(x, y) if (x, y) > (y, x),
(x, y) =
0 otherwise.
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:
a p b = 1 0 ((1 a)p + (1 b)p 1)1/p , p 6= 0
Frank:
(s1a 1)(s1b 1)

a s b = 1 logs 1 + , s > 0, s 6= 1
s1
Dombi: a b = [1 + [(1/a 1) + (1/b 1) ]1/ ]1 , > 0

Hamacher: a b = [a + b + (r 2)ab]/[r + (r 1)ab, r > 0
Schweizer and Sklar 2: a b = [ap + bp ap bp ]1/p , p > 0
Schweizer and Sklar 3:
p
+| ln(1b)|p )1/p
a b = 1 e(| ln(1a)| , p>0
i i
i i
i i
K23798 2015/2/2
i i
Schweizer and Sklar 4:
a b = 1 [(1 a)(1 b)]/[(1 a)p + (1 b)p (1 a)p (1 b)p ]1/p , p > 0
Dubois and Prade:
a b = 1 [(1 a)(1 b)/[(1 a) (1 b) ], [0, 1]
Weber:

ab=1 a+b ab , > 1
1
Yu: a b = 1 (a + b + ab), > 1
Proposition 3.3.1 [15] Let = . Let F R(X). Suppose = . Then

x, y X,
(x, y) if (x, y) > (y, x)
(x, y) =
0 otherwise.
Proof. The result follows from Proposition 3.1.8.
Proposition 3.3.2 Let 2 be the t-conorm defined by a, b [0, 1], a 2 b =

1 (a + b). Let FR(X). Suppose = 2 . Then x, y X,

0 if (x, y) (y, x)
(x, y) =
(x, y) (y, x) if (x, y) > (y, x).
Proof. If (x, y) (y, x), then (x, y) = (x, y) and so (x, y) = 1

((x, y)+(x, y)). Thus (x, y) = 0. Suppose (x, y) > (y, x). Then (x, y) =
1 ((x, y) + (y, x)). Suppose (x, y) + (y, x) 1. Then (x, y) 1 (y, x)
and (x, y) = 1. Clearly, (x, y) = 1(y, x) = (x, y)(y, x) is the minimal
solution. Suppose (x, y)+(y, x) 1. Then (x, y) = (x, y)+(y, x). Hence
(x, y) = (x, y) (y, x).
Proposition 3.3.3 [15] Let = d . Let FR(X). Suppose = .

Then x, y X,

(x, y) if (x, y) > (y, x) = 0
(x, y) =
0 otherwise.
Proof. Proposition 3.2.7(1).
Theorem 3.3.4 [15] Let = A . Let F R(X). Suppose = . Then

x, y X,
i i
i i
i i
K23798 2015/2/2
i i
(
(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.
Proposition 3.3.5 [15] Let = Y . Let FR(X). Suppose = .

Then x, y X,

[(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).
Proposition 3.3.6 Let = n . Let FR(X). Then has a factorization

= if and only if x, y X, either (1) (x, y) = 1 or (y, x) = 1 or (2)
(x, y) = (y, x) or (3) (x, y) + (y, x) < 1 and (x, y) 6= (y, x).
Proof. Proposition 3.2.8.
Proposition 3.3.7 [15] Let = H2 . Let FR(X). Suppose = .

Then x, y X,
(
(x,y)(y,x)
(x, y) = 1(x,y)(y,x) if (x, y) > (y, x)
0 otherwise.
Proposition 3.3.8 [15] Let = p . Let FR(X). Suppose = .

Then x, y X,

1 [(1 (x, y))p (1 (y, x))p + 1]1/p if (x, y) > (y, x)
(x, y) =
0 otherwise.
Proposition 3.3.9 [15] Let = s . Let FR(X). Suppose = .

Then x, y X,
(s 1)(s1(x,y) 1)

(x, y) = 1 logs +1 .
s1(y,x) 1
Proof. The proof follows from Proposition 3.2.9(4).
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).
Proof. Let t [0, 1] be such that (x, y) = t (y, x). Then
(x, y) = 1 [(1 t)(1 (y, x))]/[(1 t) (1 (y, x)) ]
(1) Suppose 1t = (1t)(1(y, x)). Then (x, y) = 1(1(y, x)) =

(y, x), a contradiction.
(2) Suppose 1 (y, x) = (1 t) (1 (y, x)) . Then (x, y) =
1 (1 t) = t. (Note 1 t = 1 (x, y) < 1 (y, x).)
(3) Suppose = (1 t) (1 (y, x)) . Then (x, y) = 1 [(1
t)(1 (y, x)]/. Thus (1 t)(1 (y, x)) = (1 (x, y)). Hence t = 1
[(1 (x, y))/(1 (y, x))]. (Note (1 (x, y))/(1 (y, x)) 1 and so
1 t = (1 (x, y))/(1 (y, x)) .)
Theorem 3.3.11 Let be the Hamacher conorm. Then x, y X,

FR(X) such that (x, y) > (y, x), (x, y) = t (y, x) has a solution for t if
and only if r = 1. If r = 1, then (x, y) = ((x, y) (y, x))/(1 (y, x)).
Proof. Suppose r = 1. Then a, b [0, 1], a b = a + b ab, i.e., is
the algebraic sum and the desired result is known. Conversely, suppose r > 0
and (x, y) = t (y, x) has a solution FR(X) and x, y X such that
(x, y) > (y, x). Then
(x, y) = [t + (y, x) + (r 2)t(y, x)]/[r + (r 1)t(y, x)].
Thus
t = [(y, x) r(x, y)]/[(r 1)(y, x)(x, y) 1 (r 2)(y, x)]
and 0 t 1. Hence either
(1) 0 (y, x) r(x, y) (r 1)(y, x)(x, y) 1 (r 2)(y, x)
or
(2) 0 (y, x) r(x, y) (r 1)(y, x)(x, y) 1 (r 2)(y, x).
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
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
The desired result follows since 0 t 1.
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
(1 t)(1 (y, x))

(x, y) = 1
[(1 t)p + (1 (y, x))p (1 t)p (1 (y, x))p ]1/p
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.,
(1 t)p (1 (y, x))p = (1 (x, y))p (1 t)p

+(1 (x, y))p (1 (y, x))p
(1 t)p (1 (x, y))p (1 (y, x))p ,
which implies that
(1 (x, y))p (1 (y, x))p = (1 tp )[(1 (y, x))p +

(1 (x, y))p (1 (y, x))p
(1 (x, y))p ].
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
The desired result follows since 0 t 1.
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
Proposition 3.3.15 Let denote the Yu 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 (x, y) < 1
1 + (y, x)
and (x, y) = 1 if (x, y) = 1.

Proof. If (x, y) = 1, then (x, y) = 1 (y, x). Suppose (x, y) < 1.
Suppose there exists t [0, 1] such that (x, y) = t (y, x). Then (x, y) =
t + (y, x) + t(y, x) = t(1 + (y, x)) + (y, x). Thus
(x, y) (y, x)
t= .
1 + (y, x)
Suppose (y, x) > 0. Since > 1, > 1/(y, x). Thus 1 + (y, x) > 0.
Also, (x, y) (y, x) 1 (y, x) and so (x, y) (y, x) 1 + (y, x).
Thus 0 < t 1 and so the desired result follows.
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.
Proposition 3.3.17 Let denote the Weber conorm. Let x, y X. Suppose

FR(X), (x, y) > (y, x) and (x, y) = (x, y) (y, x). Then
(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.
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
3.4. Intuitionistic Fuzzy Relations 93
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 =
We must show that t 0. We show that

p 1/p
1 p
ln | ln(1 (y, x)| > 0.
1 (x, y)

1 1 1 1
Now 1(x,y) > 1(y,x) and so ln 1(x,y) > ln 1(y,x) . Consequently,

1 1
ln > ln(1(y, x)) and so ln 1(x,y) > | ln(1(y, x))|. Hence
1(x,y) p p
1 1
ln 1(x,y) > | ln(1(y, x))|p and ln 1(x,y) | ln(1(y, x))|p >
0. (Note ea 1 if a > 0.)
3.4 Intuitionistic Fuzzy Relations

The proofs of many factorization results for an intuitionistic fuzzy binary
relation h , i involve dual proofs, one for with respect to a t-conorm
and one for with respect to a t-norm . In this section, we show that one
proof can be obtained from the other by considering and dual under an
involutive fuzzy complement. We follow the lead in [2] in this and the next
section.
Intuitionistic fuzzy sets were introduced by George Gargov. Krassimir
Atanassov has been the leader in the field [1, 2]. An intuitionistic fuzzy set con-
sists of a set X and two fuzzy subsets and of X such that (x) + (x) 1.
i i
i i
i i
K23798 2015/2/2
i i
The fuzzy subset can be thought of as providing the degree of membership

of elements of X in a subset of X while can be thought of as providing the
degree of nonmembership of elements of X in the subset. Other interpreta-
tions of and are possible, e.g., validity and nonvalidity, respectively. The
fuzzy subset of X defined by (x) = 1 (x) (x) for all x X is the
uncertainty (indeterminacy) of x in the subset. These various interpretations
of and open the door for a variety of interpretations of concepts in political
science and other disciplines.
In Fono, Nana, Salles, and Gwet [13], a factorization of an intuitionistic
fuzzy binary relation into a unique indifference component and a family of reg-
ular strict components was provided. This result generalizes a result in [10]
with the (max, min) intuitionistic fuzzy t-conorm. In [13], a characterization
of C-transitivity for a continuous t-representable intuitionistic fuzzy t-conrm
C was established. This enabled the authors to determine necessary and suffi-
cient conditions on a C-transitive intuitionistic fuzzy binary relation under
which a component of satisfies pos-transitivity and negative transitivity.
In Cornelis, Deschrijver, and Kerre [6], some results on t-representable
intuitionistic fuzzy t-norms (T, S) were established, where T is a fuzzy t-norm
and S is a fuzzy t-conorm such that T (a, b) = 1 S(1 a, 1 b) for all
a, b [0, 1].
An intuitionistic fuzzy binary relation h , i on a set X is a
pair of fuzzy binary relations and on X such that for all x, y X,
(x, y) + (x, y) 1. In [21], Nana and Fono obtained some intuitionistic
versions of Arrows impossibility theorem. They also provide an example of
a nondictatorial intuitionistic fuzzy aggregation rule. They also establish an
intuitionistic version of Gibbards theorem. Many of their results rely on the
factorization of intuitionistic fuzzy reference relations. The proofs of many
factorization results for an intuitionistic fuzzy binary relation h , i involve
dual proofs, one for with respect to a t-conorm and one for with re-
spect to a t-norm . We show that one proof can be obtained from the other
by considering and dual under an involutive fuzzy complement. The key
ideas are presented just prior to Theorem 3.5.8.
We present preliminary results needed for the rest of the chapter. We
present basic properties concerning the factorization of intuitionistic fuzzy
preference relations. In the final section, we present factorization results in-
volving a wide variety of t-conorms and t-norms.
We say that a t-norm is strict if a, b, d [0, 1], b < d implies (a, b) <
(a, d). We say that a t-conorm is strict if a, b, d [0, 1], b < d implies
(a, b) < (a, d).
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
3.4. Intuitionistic Fuzzy Relations 95
(2) a, b [0, 1], a b implies c(a) c(b) (montonicity).

If c(c(a)) = a for all a [0, 1], then c is called involutive.
It is known that if c : [0, 1] [0, 1] satisfies conditions (2) and is convolu-
tive, then c is continuous and satisfies condition (1), [16]. It is also clear that
if c is involutive, then c is one-to-one and thus strictly decreasing.
The following result is certainly known, at least for the standard comple-
ment. We present its proof for the sake of completeness.
Proposition 3.4.2 Let c be an involutive fuzzy complement.

(1) Let be a t-conorm. Define : [0, 1]2 [0, 1] by a, b [0, 1],
a b = c(c(a) c(b)). Then is a t-norm.
(2) Let be a t-norm. Define : [0, 1]2 [0, 1] by a, b [0, 1], a b =
c(c(a) c(b)). Then is a t-conorm.
Proof. (1) Let a, b, d [0, 1]. Then 1 b = c(c(1) c(b)) = c(0 c(b) =
c(c(b)) = b. Also a b = c(c(a) c(b)) = c(c(b) c(a)) = b a. Now b d
c(a) c(b) c(a) c(d) = c(c(a) c(b)) c(c(a) c(d)) a b a d.
We next show the associative law.
a (b d) = a (c(c(b) c(d))) = c(c(a) (c(b) c(d))

= c((c(a) c(b)) c(d)) = c(c(a b) c(d))
= (a b) d.
(2) The proof of (2) is similar to that of (1).

Let c be an involutive fuzzy complement. Given . Define in terms of
and c as above. Now define b in terms of and c as above. Then b =
since c is one-to-one. Similarly, given , then b = , where is defined in
terms of and c and then b is defined in terms of and c.
Thus if is determined from by an involutive complement c, then
is determinable from by c and vice versa. In either case, we call and
dual under c.
Proposition 3.4.3 Let c be an involutive fuzzy complement. Let i and i

be dual t-conorms and t-norms under c, respectively, i = 1, 2. Then 1 2
if and only if 1 2 .
Proof. We have
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
Proposition 3.4.4 Let c be a involutive fuzzy complement. Let and be

dual t-conorms and t-norms under c, respectively. Then is strict if and only
if is strict.
Proof. Let a, b, d [0, 1]. Suppose is strict. Then
b < d a b < a d c(a) c(b) > c(a) c(d)

c(c(a) c(b)) < c(c(a) c(d)) a b < a d.
Thus is strict. That is strict implies is strict follows similarly.
Proposition 3.4.5 Let c be an involutive fuzzy complement. Let and

be dual t-conorms and t-norms under c, respectively. Then is continuous if
and only if is continuous.
Proof. Suppose is continuous. Let c c be the function of [0, 1]2 into
[0, 1]2 defined by (a, b) [0, 1]2 , (c c)(a, b) = (c(a), c(b)). Let (a, b) [0, 1]2 .
Then
(c (c c))(a, b) = c(((c c)(a, b)))

= c((c(a), c(b)))
= c(c(a) c(b))
= a b,
where denotes composition of functions. Thus = c c c. Hence

is continuous since c, , and c c are continuous. The converse follows in a
similar manner.
3.5 Intuitionistic Fuzzy Preference Relations

and Their Factorization
and one for with respect to a t-norm . We show that one proof can be
obtained from the other by considering and dual under an involutive fuzzy
complement. In this section, we present some basic properties concerning the
factorization of intuitionistic fuzzy preference relations.
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}.
Definition 3.5.2 Let FR(X). Then

(1) is called symmetric if x, y X, (x, y) = (y, x);
(2) is called c-asymmetric if x, y X, (x, y) < 1 implies (y, x) = 1.
i i
i i
i i
K23798 2015/2/2
i i
3.5. Intuitionistic Fuzzy Preference Relations and Their Factorization 97
For F R(X), we often associate an c-asymmetric fuzzy relation on

X.
Definition 3.5.3 Let FR(X) and be an c-asymmetric binary fuzzy

relation associated with . Then
(1) is called simple if x, y X, (x, y) = (y, x) implies (x, y) =
(y, x);
(2) is called c-regular if x, y X, (x, y) < (y, x) if and only if
(x, y) < 1.
Let and be a t-norm and a t-conorm, respectively. Let , , FR(X).
By = or = , we mean for all x, y X, (x, y) = (x, y (x, y)
or (x, y) = (x, y) (x, y). In either case, we refer to = or =
as a factorization of .
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
(x, y) = t (y, x), t [0, 1] (3.2)
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
t = {t [0, 1] | t (y, x) (x, y)}.
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
(y, x) (x, y) = {t [0, 1] | t (y, x) (x, y)} = (x, y)
and so Eq. (3.2) becomes (x, y) = t (y, x) and so t = (x, y).

Suppose is strict. Then g is one-to-one.
Proposition 3.5.5 Let T . Let , , F R(X). Then statements (1)

and (2) are equivalent.
(1) x, y X, (x, y) = (x, y)(x, y), is c-asymmetric, is symmetric,
and is simple.
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].
Proposition 3.5.6 Suppose is a t-norm Suppose , , FR(X). If (i)

= , (ii) is c-asymmetric, and (iii) is symmetric, then (x, y) =
(x, y) (y, x).
Let c be an involutive fuzzy complement and FR(X). Define the fuzzy

preference relation c on X by x, y X, c (x, y) = c((x, y)).
Proposition 3.5.7 Let c be a involutive fuzzy complement. Let and be

dual t-conorms and t-norms under c, respectively.
(1) If = , then c = c c ;
(2) If = , then c = c c .
i i
i i
i i
K23798 2015/2/2
i i
Proof. (1) Let x, y X. Then (x, y) = (x, y) (x, y) = c(c((x, y))

c((x, y))). Thus c((x, y)) = c((x, y)) c((x, y)). That is, c (x, y) =
c (x, y) c (x, y).
(2) The proof of (2) is similar to that of (1).
Consider (1) of Proposition 3.5.7, where is asymmetric and is sym-
metric. Then (x, y) = (x, y) (y, x) by [15, Proposition 2.6, p. 20] or [24,
Proposition 1.2, p. 364] and so
c (x, y) = c((x, y) (y, x)) = c((x, y)) c((y, x)) = c (x, y) c (y, x).
Hence we see that if c = c c = bb , where b is c-asymmetric and b

is symmetric, then c = b by Proposition 3.5.6. If is strict, the solution to
b = c .
(3.2) is unique so it follows that
Similar comments can be made concerning (2).
Key idea: Let c be an involutive complement and let and be a t-
conorm and t-norm, respectively, dual with respect to c. Let f be the function
of FR(X) into FR(X) defined by FR(X), f () = c . Then f is one-to-
one and onto. Hence when a result concerning c is determined from a result
concerning , then the result for c is a general result since f is one-to-one
and onto. Thus if h , i is an intuitionistic fuzzy preference relation, a result
for has an immediate corresponding result for .
Theorem 3.5.8 Let c be an involutive fuzzy complement. Let and be

dual t-conorms and t-norms under c, respectively. Let 1 , 2 C(T ). Let
FR(X). Suppose that = 1 1 = 2 2 . Suppose further that either 1
or 2 is strict. Then 1 2 if and only if 1 2 for all such factorizations.
Proof. 1. We know c = 1c 1 c = 2c 2 c . Also 1 2 if and only
if 1 2 and in fact i is strict if and only if i is strict, i = 1, 2. From
[14, Theorem 3.11, p. 32], we have that 1 2 if and only if 1c 2c and
so 1 2 if and only if 1 2 .
Proof. 2. Suppose 1 2 . Suppose x, y X such that 1 (x, y) >
2 (x, y). Now 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) 2 (x, y). Hence (x, y) = 1 (x, y) 1 (x, y) > 2 (x, y) 1 (x, y)
2 (x, y) 2 (x, y) = (x, y), a contradiction. Suppose 2 is strict. Then
1 (x, y) 2 (x, y) > 2 (x, y) 2 (x, y). Thus (x, y) = 2 (x, y) 2 (x, y) <
1 (x, y) 2 (x, y) 1 (x, y) 1 (x, y) = (x, y), a contradiction. Hence no
such x, y exist. Thus 1 2 .
Conversely, suppose 1 2 . Suppose a, b [0, 1] such that a1 b < a2 b.
Suppose 1 is strict. Let F R(X) be such that x, y X such that
(x, y) = a1 b and (y, x) = b. Then (x, y) (y, x). Thus (x, y) = b. Since
(x, y) = 1 (x, y) 1 (x, y) and 1 is strict, 1 (x, y) = a. Thus (x, y) = a 1
b < a 2 b 2 (x, y) 2 (x, y) = (x, y), a contradiction. Suppose 2 is strict.
i i
i i
i i
K23798 2015/2/2
i i
Let F R be such that x, y X such that (x, y) = a 2 b and (y, x) = b.

Then (x, y) (y, x). Thus (x, y) = b. Since = 2 2 and 2 is strict,
2 (x, y) = a. Thus (x, y) = 1 (x, y) 1 (x, y) a 1 b < a 2 b = (x, y), a
contradiction. Hence no such a and b exist. Thus 1 2 .
Proposition 3.5.9 Let T . Suppose = , (i) = , (ii) is

c-asymmetric, (iii) is symmetric. Then is simple if and only if x, y X,

(x, y) if (x, y) < (y, x)
(x, y) =
1 otherwise.
Proof. Let x, y X. Assume is simple. By (i), (x, y) (x, y). Suppose

(x, y) < (y, x). Assume (x, y) < (x, y). Then (x, y) = (x, y) = (y, x) =
(y, x), where the last equality holds by Proposition 3.5.5. However, this is
impossible. Thus (x, y) (x, y) and so (x, y) = (x, y). Now suppose
(x, y) (y, x). If (x, y) = (y, x), then (x, y) = (y, x) and so (x, y) =
(y, x) = 1 since is simple. If (x, y) > (y, x), then it has just been shown
that (y, x) = (y, x) < 1. Hence (x, y) = 1 by Proposition 3.5.5.
Conversely, suppose (x, y) = (x, y) if (x, y) < (y, x) and 1 otherwise.
Suppose (x, y) = (y, x). Then (x, y) = (y, x) = 1 by the hypothesis. It
follows that is simple.
Proposition 3.5.10 Let FR(X). Suppose = . Let c be an invo-

lutive fuzzy complement and let and be a dual t-conorm and a t-norm
with respect to c, respectively. Then
(1) c = , where c = c .
(2) c = , where c = c .
Proof. (1) Since = , c = c c . Suppose a [0, 1] and x, y X
such that c = a c , where a > c (x, y). Then (x, y) = c(a c (x, y)) =
c(c(c(a)) c (x, y)) = c(a) (x, y). However, this contradicts the minimality
of since c(a) < c( c (x, y)) = c(c( (x, y)) = (x, y).
(2) The proof for (2) is similar to that of (1).
Theorem 3.5.11 Let 1 , 2 C(T ). Let FR(X). Suppose that =

1 1 = 2 2 . Then 1 2 for all such factorizations if and only if
1 2 .
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
Conversely, suppose 1 2 . Suppose there exists FR(X) such that

1 (x, y) > 2 (x, y) for some x, y X. Now 1 (x, y) 1 (x, y) 1 (x, y) 2
(x, y). Also 2 (x, y)2 (x, y) < 1 (x, y)2 (x, y), where strict equality holds
by the maximality of 2 . Thus (x, y) = 1 (x, y) 1 (x, y) < 1 (x, y) 2
(x, y) 1 (x, y) 1 (x, y) = (x, y), a contradiction. Hence no such x, y
exist. Thus 1 2 .
Let T denote the set of all intuitionistic fuzzy preference relations h , i
on X. Then T n is the set of all such n-tuples.
Proposition 3.5.12 Let c be an involutive fuzzy complement. Let (h1 , 1 i,

. . . , hn , n i) T n be such that i = ci , i = 1, . . . , n. Let i N. Let
x, y X. Then
[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.
Corollary 3.5.13 Let c be an involutive fuzzy complement. Let (h1 , 1 i,

. . . , hn , n i) T n be such that i = ci , i = 1, . . . , n. If i is a dictator for
, then i is a dictator for = c and conversely.
Example 3.5.14 Let N = {1, 2} and X = {x, y}. Let (h1 , 1 i, h2 , 2 i)

T 2 be such that
1 1
1 (x, y) = , 1 (y, x) = 0, 2 (x, y) = 1, 2 (y, x) = ,
2 2
1 1 1
1 (x, y) = , 1 (y, x) = , 2 (x, y) = 0, 2 (y, x) = .
2 4 2
Suppose player 1 is a dictator for . Then 1 (x, y) > 1 (y, x) and so
1 (x, y) > 0. Hence (x, y) > 0 since 1 is a dictator for . Now
1 (x, y) = 1 and 1 (y, x) < 1. Thus if player 1 were a dictator for
, then (y, x) < 1. But we should have (x, y) < 1 if player 1 is to
be a dictator for (h1 , 1 i, h2 , 2 i). (Note that it is not the case that
1 (x, y) > 1 (y, x) 1 (x, y) < 1 (y, x).)
Let c be an involutive fuzzy complement. We have x, y X,
(x, y) > 0 c (x, y) < 1, (x, y) = 0 c (x, y) = 1, (x, y) = 1

c (x, y) = 0, (x, y) = (y, x) c (x, y) = c (y, x),
and Supp() = Cosupp(c ). Then is asymmetric c is c-asymmetric.
i i
i i
i i
K23798 2015/2/2
i i
Proposition 3.5.15 [15, Proposition 2.2, p.18] Let , , FR(X). Suppose

that (i) Supp() = Supp() Supp(), (ii) is asymmetric, (iii) is symmetric.
Then
(1) x, y X, (x, y) > 0 and (y, x) = 0 if and only if (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.
Then the following proposition follows from Proposition 3.5.15 and the
discussion immediately preceding it.
Proposition 3.5.16 Let , , FR(X). Suppose that

(i) Cosupp(c ) = Cosupp( c )Cosupp(c ), (ii) c is c-asymmetric, (iii) c
is symmetric. Then
(1) x, y X, c (x, y) < 1 and c (y, x) = 1 if and only if c (x, y) =
c
(y, x) = 1.
(2) x, y X, c (x, y) < 1 and c (y, x) < 1 if and only if c (x, y) =
c
(y, x) < 1.
Since is an arbitrary element of FR(X) and f is a one-to-one function
from FR(X) onto FR(X), we have that c is also an arbitrary element from
FR(X). Hence the following result is true.
Proposition 3.5.17 Let , , FR(X). Suppose that

(i) Cosupp() = Cosupp()Cosupp(), (ii) is c-asymmetric, (iii) is
symmetric. Then
(1) x, y X, (x, y) < 1 and (y, x) = 1 if and only if (x, y) = (y, x) =
1.
(2) x, y X, (x, y) < 1 and (y, x) < 1 if and only if (x, y) = (y, x) <
1.
Consider an intuitionistic fuzzy subset h , i of X. If Proposition 3.5.15
holds for , then Proposition 3.5.17 holds for (and conversely).
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
Einstein: a H2 b = ab/(2 a b + ab)

(
0 (ap + bp 1)1/p if 2 ap bp [0, 1]
Schweizer and Sklar: ap b = 0 otherwise
h a b
i
Frank: a s b = logs 1 + (s 1)(s
s1
1)
, s > 0, s 6= 1
h 1 i1
Dombi: a b = 1 + ( a1 1) + ( 1b 1)

,>0
Hamacher: a b = ab/[r + (1 r)(a + b ab], r > 0
Schweizer and Sklar 2: a b = 1 [(1 a)p + (1 b)p (1 a)p (1 b)p ]1/p ,
p>0
p
+| ln b|p )1/p
Schweizer and Sklar 3: a b = e(| ln a| ,p>0
p p p p 1/p
Schweizer and Sklar 4: a b = ab/[a + b a b ] ,p>0
Dubois and Prade: a b = ab/a b , [0, 1]
Weber: a b = 0 (a + b + ab 1)/(1 + ), > 1
Yu: a b = 0 [(1 + )(a + b 1) ab], > 1
We now derive factorizations for t-norms. We give direct proofs for several
results. Later we use factorization results for t-conorms to derive factorization
results for t-norms.
Proposition 3.5.19 Let = . Let F R(X). Suppose = . Then

x, y X,
(x, y) if (x, y) < (y, x)
(x, y) =
1 otherwise.
i i
i i
i i
K23798 2015/2/2
i i
Theorem 3.5.20 Let = A . Let FR(X). Suppose = . Then

x, y X, (
1 if (x, y) (y, x)
(x, y) = (x,y)
(y,x) if (x, y) < (y, x).
Proof. Let x, y X. Then (x, y) = (x, y) i(x, y). Since (x, y) =

(x, y) (y, x), we have (x, y) = 1 if (x, y) (y, x). Suppose (x, y) <
(y, x). Then (x, y) = (x, y) (y, x) = (x, y)(y, x). Thus (x, y) =
(x, y)/(y, x).
Let c be the standard fuzzy complement. Then A and A are dual with
respect to c. Suppose = A and (x, y) = (x, y) (x, y) as in The-
orem 3.5.4. Then c (x, y) = c (x, y) c (x, y), where = A . Suppose
(x, y) > (y, x). Then c (x, y) < c (y, x). By Theorem 3.5.4, c (x, y) =
(c (x, y) c (y, x))/(1 c (y, x)) = (1 (x, y) 1 + (y, x))/(y, x) =
((y, x) (x, y))/(y, x). Thus (x, y) = 1 ((y, x) (x, y))/(y, x) =
(x, y)/(y, x) which agrees with Theorem 3.5.20.
Proposition 3.5.21 Let 2 be the t-norm defined by a, b [0, 1], a 2 b =

0 (a + b 1). Let FR(X). Suppose = 2 . Then x, y X,

1 if (x, y) (y, x)
(x, y) =
1 + (x, y) (y, x) if (x, y) < (y, x).
Proof. If (x, y) (y, x), then (x, y) = (x, y) and so (x, y) = 1.

Suppose (x, y) < (y, x). Then (x, y) = (y, x). Thus (x, y) = 0 ((x, y) +
(y, x) 1). If (x, y) + (y, x) 1 0, then (x, y) = 1 + (x, y) (y, x). If
(x, y) + (y, x) 1 0, then (x, y) = 0 and so (x, y) 1 (y, x) + (x, y).
In this case, (x, y) = 1 (y, x) is the maximal solution.
Proposition 3.5.22 Let = p . Let FR(X). Suppose = . Then

x, y X,

1 if (x, y) (y, x)
(x, y) =
(1 + (x, y)p (y, x)p )1/p if (x, y) < (y, x)
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].)
Proposition 3.5.23 Let = w . Let FR(X). Suppose = . Then

x, y X,
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
(1 t)w = (1 (x, y))w (1 ((y, x))w ,

t = 1 [(1 (x, y))w (1 (y, x))w ]1/w .
7
(Note that if (x, y) = 0, (y, x) = 8 and w = 1, then any t [0, 81 ] is a
solution.)
Remark 3.5.24 Let = s . Let FR(X). Suppose = . Then

x, y X,
(x, y) = logs [1 + (s 1)(s(x,y) 1)/(s(y,x) 1)] if (x, y) < (y, x).
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.
Proposition 3.5.26 Let = n . Let FR(X). Suppose (x, y) < (y, x)

and (x, y) = (x, y) (y, x). Then (x, y) = (x, y).
Proof. Since (x, y) = (x, y) (y, x), there exists t [0, 1] such that
t + (y, x) > 1 or (x, y) = 0 and (x, y) = t (y, x) for t [0, 1 (y, x)].
In either case, (x, y) = (x, y).
The dual to Proposition 3.3.6 is the following result.
Proposition 3.5.27 Let = n . Let FR(X). Then has a factoriza-

tion = if and only if x, y X, either (1) (x, y) = 0 or (y, x) = 0
or (2) (x, y) = (y, x) or (3) (x, y) + (y, x) > 1 and (x, y) 6= (y, x).
Proof. (x, y) + (y, x) < 1 if and only if 1 (x, y) + 1 (y, x) > 1.
i i
i i
i i
K23798 2015/2/2
i i
Proposition 3.5.28 Let = H2 . Let FR(X). Suppose = . Then

x, y X,

(2(x, y) (x, y)(y, x))/((x, y) + (y, x) (x, y)(y, x))
(x, y) = if (x, y) (y, x)
1 if (x, y) < (y, x).

Proof. Let t be a solution to Eq. (3.2). Then (x, y) = t(x, y)/(2 t

(x, y) + t(x, y)) and so 2(x, y) t(x, y) (x, y)(x, y) + t(x, y)(x, y) =
t(x, y). Thus t(x, y) + t(x, y) t(x, y)(x, y) = 2(x, y) (x, y)(x, y).
Hence
t = (2(x, y) (x, y)(x, y))/((x, y) + (x, y) (x, y)(x, y)).
The desired result now follows easily.
Proposition 3.5.29 Let = d . Let FR(X). Suppose (x, y) < (y, x)

and (x, y) = (x, y) (y, x). Then

(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.
Proposition 3.5.30 Let = p . Let F R(X). Suppose (x, y) =

(x, y) (y, x). Then

[(x, y)p (y, x)p + 1]1/p if (x, y) < (y, x),
(x, y) =
1 otherwise.
Proof. For = p , we have for (x, y) = (x, y) (y, x),

1 [(1 (x, y))p (1 (y, x))p + 1]1/p if (x, y) > (y, x)
(x, y) =
0 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
Hence for c (x, y) > c (y, x)

1/p
(1 c (x, y))p (1 c (y, x))p

c
(x, y) = 1 .
1 (1 c (y, x))p
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

1/p
(1 (x, y))p (1 (y, x))p

(x, y) = 1 .
(1 (y, x)) + (1 (x, y))p (1 (y, x))p (1 (x, y))p
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).
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

(
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.
Thus for c (x, y) < c (y, x),

(
c
0 if c (x, y) = 0
(x, y) = p c p 1/p
e[(ln( c (x,y) )) | ln( (y,x)| ]
1
if c (x, y) > 0.
i i
i i
i i
K23798 2015/2/2
i i
Proposition 3.5.35 Let denote the Yu norm. Let F R(X). Suppose

(x, y) < (y, x) and (x, y) = (x, y) (y, x). Then x, y X,
(y, x) (x, y)
(x, y) = 1 if (x, y) > 0
1 + (1 (y, x))
and (x, y) = 0 if (x, y) = 0.
Proof. For the Yu conorm with FR(X), (x, y) > (y, x), and
(x, y) = (x, y) (y, x) we have x, y X,
(x, y) (y, x)
(x, y) = if (x, y) < 1
1 + (y, x)
and (x, y) = 1 if (x, y) = 1. Thus for c (x, y) < c (y, x),
(1 c (x, y) (1 c (y, x))
c (x, y) = 1 if c (x, y) > 0
1 + (1 c (y, x))
and c (x, y) = 0 if c (x, y) = 0.
The next result follows from Proposition 3.3.14 and a dual argument.
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.
Theorem 3.5.37 Let be the Hamacher norm. Then x, y X,

FR(X) such that (x, y) < (y, x), (x, y) = t (y, x) has a solution for
t if and only if r = 1.
Proof. For the Hamacher conorm , we have x, y X, FR(X)
such that (x, y) > (y, x), (x, y) = t (y, x) has a solution for t if and
only if r = 1. Now (x, y) > (y, x) and (x, y) = t (y, x) if and only if
c (x, y) < c (y, x) and c (x, y) = 1 (t (1 c (y, x)).
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))
Proof. The proof is immediate by Lemma 3.5.38.
Example 3.5.40 Let h , i be an intuitionistic fuzzy set. Let denote the

algebraic sum and the algebraic product. Then x, y X,
(
(x,y) (y,x)
1 (y,x) if (x, y) > (y, x)
(x, y) =
0 if (x, y) (y, x)
and (
1 if (x, y) (y, x)
(x, y) = (x,y)
(y,x) if (x, y) < (y, x).
Let x, y X be such that (x, y) = 3/4, (y, x) = 1/4, (x, y) = 1/4,

and (y, x) = 1/2. Then (x, y) + (x, y) 1 and (y, x) + (y, x) 1
and
(x, y) (y, x) (x, y)
(x, y) + (x, y) = +
1 (y, x) (y, x)
[ (x, y) (y, x)] (y, x) + [1 (y, x)] (x, y)
=
[1 (y, x)] (y, x)
[3/4 1/4]1/2 + [1 1/4]1/4 7/16 7
= = = > 1.
[1 1/4]1/2 3/8 6
i i
i i
i i
K23798 2015/2/2
i i
3.6. Exercises 111
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
(x, y) (y, x) (x, y)

(x, y) + (x, y) = +
1 (y, x) (y, x)
1/4 1/8 1/4 1/8 1
= + = +
1 1/8 3/4 7/8 3
1 1
= + < 1.
7 3
In this section, we provided factorization results for intuitionistic binary
relations h , i. We showed that the results for and are dual. A further
research project would be to apply the results to Arrowian-like conditions.
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
4. [17] Let be a fuzzy weak preference relation and (, ) an axiomatic

factorization of . If T is a continuous non-Archimedian t-norm with zero
divisors, then prove that there exists s (0, 1) such that (x, y) {0, 1}
or (x, y) > 1 s for all x, y X.
5. [17] Prove that a t-norm T is a continuous Archimedean t-norm with
zero divisors if and only if there exists an order isomorphism such that
T = W .

factorization of . If is a reciprocal order automorphism and T = W ,
then prove that for all x, y X,
(x, y) = 1 (((x, y)) ((x, y)))
and
((x, y)) + ((y, x)) 1)
1 ( ) (x, y) (x, y) (y, x).
2

factorization of . If is a reciprocal order automorphism, T = W , and
= (1 )c , then prove that for all x, y X,
(x, y) = 1 (((x, y)) ((y, x))) 0,

(x, y) = (x, y) (y, x),
where c (x, y) = 1 (x, y).

factorization of . If is a reciprocal order automorphism, T = W ,
then prove that = 1 is equivalent to = (1 )c and when one
of these relationships holds,
(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.
2. K. T. Atanassov, Intuitionistic Fuzzy Sets, Studies in Fuzziness and Soft

Computing, Physica-Verlag, Heidelberg, 1999.
i i
i i
i i
K23798 2015/2/2
i i
3.7. References 113
3. A. Banerjee, Fuzzy preferences and Arrow-type problems in social

choice, Soc. Choice Welfare, 11 (1994) 121130.
4. C. R. Barrett, P. K. Pattanaik, and M. Salles, On the structure of fuzzy

5. A. Bufardi, On the construction of fuzzy preference relations, J. Multi-
criteria Decision Anal., 7 (1998) 169175.
6. C. Cornelis, G. Deschrijver, and E. E. Kerre, Implication in intuition-

istic fuzzy intervalvalued fuzzy set theory: Construction, classification,
application, International Journal of Approximate Reasoning, 35 (2004)
5595.
7. M. Dasgupta and R. Deb, Factoring fuzzy transitivity, Fuzzy Sets and
Systems, 118 (2001) 489502.
8. B. De Baets and E. E. Kerre, Fuzzy preference relations and their
characterization, J. Fuzzy Math., 3 (1995) 373381.
9. B. De Baets, B. Van de Walle, and E. E. Kerre, Fuzzy preference struc-
tures without incomparability, Fuzzy Sets and Systems, 76 (1995) 333
348.
10. D. Dimitrov, Intuitionistic fuzzy preferences: A factorization, Advanced
Studies in Contemporary Mathematics, 5 (2002) 93104.
11. B. Dutta, Fuzzy preference relations, Math. Social Sci., 13 (1987) 215
229.
12. L. A. Fono and N. G. Andjiga, Fuzzy strict preference and social choice,
13. 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, Hindawi Publishing Corporation Volume 2009,
Article ID 580918, 38 pages dos:10.1155/2009/580918.
14. P. Hajek, Metamathematics of Fuzzy Logic, Kluwer, Dordrecht, 1998.
15. J. E. Herr and J. N. Mordeson, Factorization of fuzzy preference rela-

tions, Advances in Fuzzy Sets and Systems, 9 (2011) 1735.
16. G. J. Klir and Bo Yuan, Fuzzy Sets and Fuzzy Logic: Theory and Appli-
cation, Prentice Hall, Upper Saddle River, New Jersey, 1995.
17. B. Llamazares, Characterization of fuzzy preference structures through
Lukasiewicz triplets, Fuzzy Sets and Systems, 136 (2003) 217234.
i i
i i
i i
K23798 2015/2/2
i i
18. B. Llamazares, Factorization of fuzzy preferences, Social Choice Welfare,

24 (2005) 475496.
19. J. N. Mordeson and T. D. Clark, Fuzzy Arrows theorem, New Mathe-

matics and Natural Computation, 5 (2009) 371383.
20. J. N. Mordeson, T. D. Clark, and K. Albert, Factorization of intuition-
istic fuzzy preference relations, New Mathematics and Natural Compu-
tation, 10 (2014) 125.
21. G. N. Nana and L. A. Fono, Arrow-type results under intuitionistic fuzzy

preferences, New Mathematics and Natural Computation, 9 (2013) 97
123.
22. S. Orlovsky, Decision making with a fuzzy preference relation, Fuzzy
23. S. Ovchinnikov, Structure of fuzzy binary relations, Fuzzy Sets and Sys-
tems, 6 (1981) 169195.
24. G. Richardson, The structure of fuzzy preference: social choice implica-
tions, Soc. Choice Welf., 15 (1998) 359369.
25. B. Schweizer and A. Sklar, Probabilistic Metric Spaces, North-Holland,

New York, 1983.
26. F. F. Tang, Fuzzy preferences and social choice, Bull. Res., 46 (1994)
263269.
i i
i i
i i
K23798 2015/2/2
i i
Chapter 4
Fuzzy Non-Arrow Results
In this chapter, we focus on one of Arrows conditions, independence of ir-

relevant alternatives to obtain nondictatorial results. We give special consid-
eration to auxiliary functions considered by Garcia-Lapresta and Llamazares
[21]. It was shown in [21] that neutral aggregation rules based on quadratic
arithmetic means generalize simple majority when individuals have ordinary
preferences, and collective preferences are reciprocal and that a weighted av-
erage is a nondictatorial fuzzy aggregation rule. We also show that it is not
the automorphic image of the ordinary average.
4.1 Nondictatorial Fuzzy Aggregation Rules

The results in this section are based on Dutta [18], Richardson [39], Banerjee
[3], and Mordeson, Giblisco, Clark [28].
Let T be a subset of FR(X), where X is a finite set.
Definition 4.1.1 A function fe : T n T is called a fuzzy aggregation

rule.
In the crisp case, an independent of irrelevant alternatives aggregation rule

requires the social preference between alternatives x and y to depend only on
the individual preferences between x and y. It is known that the Borda rule
fails to satisfy the independence of irrelevant alternatives property.
Definition 4.1.2 Let fe : T n T be a fuzzy aggregation rule. Then fe satisfies

(1) Independence of irrelevant alternatives (IIA1) if (1 , . . . , n ),
(01 , . . . , 0n ) T n , x, y X, [i N, i (x, y) = 0i (x, y) (x, y) =
0 (x, y)].
(2) Pareto condition (PC) with respect to if (1 , . . . , n ) T n ,
x, y X, (x, y) {i (x, y) | i N }.
115
i i
i i
i i
K23798 2015/2/2
i i
116 4. Fuzzy Non-Arrow Results
(3) Positive responsiveness (PR) with respect to if (1 , . . . , n ),

(01 , . . . , 0n ) T n , x, y X, i = 0i i 6= j, (x, y) = (y, x), and
(j (x, y) = 0 and j0 (x, y) > 0 or j (y, x) > 0 and j0 (y, x) = 0) implies
0 (x, y) > 0.
An independent of irrelevant alternatives (IIA1) fuzzy aggregation rule
requires an alternative x to be socially preferred to an alternative y to the
same degree whenever all individuals prefer x to y to the same degree. The
Pareto condition requires x to be socially preferred to y to a degree at least as
large as the smallest degree of the strict preference of the individuals whenever
all individuals strictly prefer x to y. If a fuzzy aggregation rule is positive
responsive, then by changing at least one individuals, say j, preference from
j (x, y) = 0 to j0 (x, y) > 0 or j (y, x) > 0 to j0 (y, x) = 0, then x becomes
strictly preferred to y with the change. Thus with positive responsive fuzzy
aggregation rules, individuals can break ties ((x, y) = (y, x)).
Let = (1 , . . . , n ) T n . Let fe be a fuzzy aggregation rule. In the
following, when a strict preference of a particular type for = fe() is assumed,
then it is assumed that strict preference for i is of the same type, i = 1, . . . , n.
Proposition 4.1.3 Let T. Let , be two different types of strict prefer-

ence with respect to . Suppose x, y X, (x, y) > 0 if and only if (x, y) > 0.
Let fe : T n T be a fuzzy aggregation rule. Let T n . Then fe satisfies PR
with respect to if and only if fe satisfies PR with respect to .
Proof. Suppose fe satisfies PR with respect to . Suppose (1 , . . . , n ),
(01 , . . . , 0n ) T n , x, y X, i = 0i i 6= j, (x, y) = (y, x), and (j (x, y) =
0 and (x, y) > 0 or j (y, x) > 0 and j0 (y, x) = 0). Then (1 , . . . , n ),
0
(01 , . . . , 0n ) T n , x, y X, i = 0i i 6= j, (x, y) = (y, x), and (j (x, y) =

0 and j0 (x, y) > 0 or j (y, x) > 0 and j0 (y, x) = 0) since j (x, y) > 0 if and
only if j (x, y) > 0 and j0 (x, y) > 0 if and only if j0 (x, y) > 0 for all x, y X.
Thus (x, y) > 0 since fe satisfies PR with respect to . Thus (x, y) > 0 by
hypothesis and so fe satisfies PR with respect to .
Corollary 4.1.4 Let fe : T n T be a fuzzy aggregation rule. Then fe satisfies

PR with respect to (3) if and only if fe satisfies PR with respect to (1) .
Proof. It suffices to show that x, y X, (3) (x, y) > 0 if and only if

(1) (x, y) > 0. However, this is immediate from the definitions since (x, y) =
(y, x) > 0 if and only if (x, y) (y, x) > 0.
In the following definition, we define a fuzzy aggregation rule which we
show later is nondictatorial, but yet satisfies certain reasonable properties.
i i
i i
i i
K23798 2015/2/2
i i
4.1. Nondictatorial Fuzzy Aggregation Rules 117
Definition 4.1.5 Define the fuzzy aggregation rule fe : T n T as follows:

= (1 , . . . , ) T n , x, y X,
n
X
fe()(x, y) = wi i (x, y),
i=1
Pn
where i=1 wi = 1 and wi > 0, i = 1, . . . , n.
Proposition 4.1.6 Let fe be defined as in Definition 4.1.5. Then fe satisfies

PR with respect to (3) and (1) .
Proof. Let , 0 T n . Suppose i = 0i , i = 1, . . . , n, i 6= j. Let x, y X.

Suppose (x, y) = 0 (x, y). Suppose also that either (1) (j (x, y) = 0 and
j0 (x, y) > 0) or (2) (j (y, x) > 0 and j0 (y, x) = 0), where strict preference is
of type (3) . Then 0 (x, y) = 0 (0 (x, y) 0 (y, x)) and
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.
Proposition 4.1.7 Let fe be the fuzzyPn aggregation rule defined in Definition

4.1.5. Then x, y X, (2) (x, y) = i=1 wi i (x, y).
Pn
Proof.
Pn Let x, y X. Then (2) P(x, y) = 1(x,
Pn y) = 1 i=1 wi i (x, y) =
n
1 i=1 wi (1 i (x, y)) = 1 i=1 wi + i=1 wi i (x, y).
Example 4.1.8 Let fe be the fuzzy aggregation rule defined in Definition

4.1.5. Then fe does not satisfy PR with respect to (2) . Let X = {x, y} and
N = {1, 2}. Define 1 , 2 and 01 , 02 on X as follows: i (x, x) = i (y, y) = 1,
i = 1, 2, 1 (x, y) = 1/2, 1 (y, x) = 1, 2 (x, y) = 3/4, 2 (y, x) = 1/2 and
0i (x, x) = 0i (y, y) = 1, i = 1, 2, 01 (x, y) = 1/2, 01 (y, x) = 1, 02 (x, y) = 1,
02 (y, x) = 1. Let w1 = 1/3 and w2 = 2/3. Now 1 = 01 . Also,
(x, y) = 1/31 (x, y) + 2/32 (x, y) = 1/3 1/2 + 2/3 3/4 = 2/3
and
(y, x) = 1/31 (y, x) + 2/32 (y, x) = 1/3 1 + 2/3 1/2 = 2/3.
i i
i i
i i
K23798 2015/2/2
i i
Thus (x, y) = (y, x). However,
2 (y, x) = 1 2 (x, y) = 1 3/4 = 1/4 > 0
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.
Proposition 4.1.9 Let fe be the fuzzy aggregation rule defined in Definition

4.1.5. Then fe satisfies PC with respect to (3) .
Proof. Let x, y X and T n . Let m = mx,y = {i (x, y) | i =
1, . . . , n}. There is no loss in generality in assuming m = 1 (x, y). Suppose
m > 0. Then

2 (x, y) n (x, y)
1 w1 + w2 + . . . + wn + w2 1 + . . . + wn 1
m m
1 (x, y) n (x, y)
= w1 + . . . + wn .
m m
Hence
1 (x, y) = m w1 1 (x, y) + . . . + wn n (x).
Since
Pn m > 0, i (x, y) Pn> 0 i N. Thus P i (x, y) > i (y, x) i N. Hence
n
i=1 wi i (x, y) > i=1 wi i (y, x). Thus i=1 wi [i (x, y) i (y, x)] > 0.
However, i (x, y) i (y, x) = i (x, y), i = 1, . . . , n. Hence
n
X
0 < 1 (x, y) 1 (y, x) wi [i (x, y) i (y, x)]
i=1
n
X n
X
= wi i (x, y) wi i (y, x)],
i=1 i=1
or {i (x, y) | i N } (x, y) (y, x) = (x, y). Hence fe satisfies PC with

respect to (3) . If m = 0, then clearly fe satisfies PC with respect to (3) .
Definition 4.1.10 Let be a fuzzy binary relation on X. Then is said to

satisfy type T2 -transitivity if x, y, z X, (x, z) (x, y) + (y, z) 1.
Definition 4.1.11 Let febe a fuzzy aggregation rule. Then

(1) fe is said to be max-min transitive if T n , fe() is max-min
transitive;
(2) fe is said to be T2 -transitive if T n , fe() is T2 -transitive.

4.1.5. Let = (1 , . . . , n ) FRn (X). If i is T2 -transitive i N, then fe()
is T2 -transitive.
i i
i i
i i
K23798 2015/2/2
i i
Proof. Let x, y, z X. Then i (x, z) i (x, y) + i (y, z) 1 i N. Thus

n
X n
X n
X n
X
(x, z) = wi i (x, z) wi i (x, y) + wi i (y, z) wi
i=1 i=1 i=1 i=1
= (x, y) + (y, z) 1.
Let H2 = { F R(X) | is reflexive, strongly complete, and T2 -

transitive}.
Let C be a nonempty subset of N. Then C is called a coalition.
Let FRn and x, y X. Then
P (x, y; ) = {i N | i (x, y) > 0}.
Definition 4.1.13 Let fe be a fuzzy aggregation rule. Then

(1) fe is called nondictatorial if it is not the case that there exists i N
such that for all T n , x, y X, i (x, y) > 0 implies (x, y) > 0;
(2) fe is said to be weakly Paretian if T n , x, y X [i N,
i (x, y) > 0 implies (x, y) > 0];
(3) fe is called neutral if x, y, z, w X, P (x, y; ) = P (z, w; 0 ) and
P (y, x; ) = P (w, z; 0 ) imply fe()(x, y) > 0 if and only if fe(0 )(z, w) > 0.
A fuzzy aggregation rule is called dictatorial if it is not nondictatorial.
The weakly Paretian condition requires an alternative x to be strictly
preferred to an alternative y whenever all individuals strictly prefer x to y.
In the crisp case, the Pareto condition and the weakly Paretian condition are
equivalent. A neutral fuzzy aggregation rule treats all pairwise comparisons
the same in that the labeling of alternatives is immaterial. All that matters
are the individuals preferences. We see that a similar situation holds in the
fuzzy case.
Definition 4.1.14 Let fe be a fuzzy aggregation rule.

(1) A coalition C is called an oligarchy if for all distinct x, y X and
all = (1 , . . . , n ) T n , (i) i (x, y) > 0 i C (x, y) > 0 and (ii)
j C, [j (x, y) > 0 (y, x) = 0].
(2) An individual j N is called a dictator if for all distinct x, y X
and all = (1 , . . . , n ) T n , j (x, y) > 0 (x, y) > 0.
(3) An individual j N is called a vetoer if for all distinct x, y X and
all = (1 , . . . , n ) T n , j (x, y) > 0 (y, x) = 0.
A fuzzy aggregation rule is called oligarchic if there exists L N such
that every member of L has a veto and L is an oligarchy. It is known in the
crisp case that there exist aggregation rules that acyclic, but not oligarchic.
Such an example is the rule employed by the United Nations Security Council
before 1965.
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.
Proof. Let fe be the fuzzy aggregation rule defined in Definition 4.1.5.

Clearly, fe is not dictatorial. By Propositions 4.1.6, 4.1.9, and 4.1.12, it only
remains to show that fe satisfies IIA1. However, this is immediate.
Definition 4.1.16 Let , 0 T. Let Im() = {s1 , . . . , sk } and Im(0 ) =

{t1 , . . . , tm } be such that s1 < . . . < sk and t1 < . . . < tm . Then and 0
are said to be equivalent, written 0 , if s1 = 0 t1 = 0, k = m, and
si = 0ti for i = 1, . . . , k.
If the fuzzy binary relations and 0 in Definition 4.1.16 are reflexive, then
sk = 1 = tm .
Proposition 4.1.17 Let , 0 T. Let x, y X. Suppose 0 . Then

(x, y) > (y, x) if and only if 0 (x, y) > 0 (y, x).
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)}.
Definition 4.1.18 Let fe be a fuzzy aggregation rule. Then fe is said to be

independent of irrelevant alternatives (IIA3) if x, y X, [, 0 T n ,
i |x,y 0i |x,y i N fe()|x,y fe(0 ) |x,y ].
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.

= (1 , . . . , ) T n , x, y X,

1 if x = y,
fe()(x, y) = 1 if i (x, y) > 0 i N,
t() otherwise.

Pn
Proposition 4.1.20 Let strict preference be defined by (1) or (3) . Let fe be

the fuzzy aggregation rule defined in Definition 4.1.19. Suppose Im(t) (0, 1).
Then fe is independent of irrelevant alternatives IIA3.
i i
i i
i i
K23798 2015/2/2
i i
Proof. Let , 0 T n and x, y X. Suppose i |x,y 0i |x,y i N. Let

i N. Then i |x,y (x, y) > i |x,y (y, x) 0i |x,y (x, y) > 0i |x,y (y, x) by
Proposition 4.1.17. Thus i (x, y) > i (y, x) 0i (x, y) > 0i (y, x). Hence by
definition of fe, we have fe()(x, y) = 1 fe(0 )(x, y) = 1 and so fe()(x, y) =
t() fe(0 )(x, y) = t(0 ). Thus fe()|x,y fe(0 )|x,y .
If t is a constant function in Proposition 4.1.20, then we can conclude
fe()|x,y = fe(0 )|x,y in the proof. If t() 1/2, then we can conclude the
fe() is strongly complete.
Proposition 4.1.21 Let strict preference be defined by (1) or (3) . Suppose

Im(t) (0, 1). Let fe be the fuzzy aggregation rule defined in Definition 4.1.19.
Then fe is weakly Paretian.
Proof. Let x, y X. If iN i (x, y) = 0, then clearly (x, y) 0. Suppose
iN i (x, y) > 0. Then i (x, y) > 0 i N. Hence i (x, y) > i (y, x) i N.
Thus fe()(x, y) = 1 and fe()(y, x) = t() < 1. Hence (x, y) iN i (x, y)
for = 1 and (x, y) > 0 for = (3) .
Let T. Recall that is called max-min quasi-transitive or just quasi-
transitive if x, y, z X, (x, z) (x, y) (y, z), Definition 2.2.16.
Proposition 4.1.22 Let be defined by (1) . Let T. If is max-min

transitive, then is max-min quasi-transitive.
Proof. Suppose there exists x, y, z X such that
(x, z) (x, y) (y, z) (4.1)
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
(z, x) (x, z) (4.5)
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.
Theorem 4.1.23 Let be defined by (1) . Then there exists a nondictatorial

fuzzy aggregation rule that is max-min transitive, independent of irrelevant
alternatives IIA3, and weakly Paretian.
Proof. Let fe be defined as in Definition 4.1.19. Clearly, fe is reflexive and
complete. By Propositions 4.1.20 and 4.1.21, fe is independent of irrelevant
alternatives IIA3 and weakly Paretian. By definition, fe() is reflexive
T n . It remains to be shown that fe is max-min transitive. Let T n . Let
x, y, z X. Suppose fe()(x, y) fe()(y, z) = 1. Assume x 6= y 6= z. Then
i N, i (x, y) > 0 and i (y, z) > 0. By Proposition 4.1.22, we have i
N that i (x, z) i (x, y) i (y, z) since i is max-min transitive. Thus
i (x, z) > 0 i N. Hence i (x, z) > i (z, x) i N. Thus fe()(x, z) = 1.
If fe()(x, y) fe()(y, z) = t(), then clearly fe() fe()(x, y) fe()(y, z).
Assume x = y 6= z. Then i N, i (y, z) > 0. Now i N, i (x, z) =
i (y, z) > 0. Thus fe()(x, z) = 1. A similar argument holds for x 6= y = z.
If x = y = z, then the result is immediate. The result is also immediate if
fe()(x, y) fe()(y, z) = t().
Example 4.1.24 Let X = {x, y, z}. Define the fuzzy binary relation on X
as follows:
(x, x) = (y, y) = (z, z) = 1,

(x, y) = (y, x) = (y, z) = (z, y) = 1/2, (x, z) = 0, (z, x) = 1.
i i
i i
i i
K23798 2015/2/2
i i
4.2. Auxiliary Functions 123
Then
(x, z) = 0 1/2 + 1/2 1 = (x, y) + (y, z) 1,

(z, x) = 1 1/2 + 1/2 1 = (z, y) + (y, x) 1,
(x, y) = 1/2 0 + 1/2 1 = (x, z) + (z, y) 1,
(y, x) = 1/2 1/2 + 1 1 = (y, z) + (z, x) 1,
(y, z) = 1/2 1/2 + 0 1 = (y, x) + (x, z) 1,
(z, y) = 1/2 1 + 1/2 1 = (z, x) + (x, y) 1.
Also, a, b X,
(a, b) = (a, a) + (a, b) 1,

(a, b) = (a, b) + (b, b) 1,
(a, a) (a, b) + (b, a) 1.
Thus is T2 -transitive. However, (2) (x, y) = 1 (y, x) = 1 1/2 >

0, (2) (y, z) = 1 (z, y) = 1 1/2, but (2) (x, z) = 1 (z, x) = 1 1 = 0.
Hence is not max-min quasi-transitive with respect to (2) .
Note also that is partially transitive since (u, v) > 0 u, v X. However,
(2) (x, y) > 0, (2) (y, z) > 0, but (2) (x, z) = 0 so is not partially quasi-
transitive (Definition 2.2.16) with respect to (2) .
Proposition 4.1.25 There exists a weakly Paretian (with respect to (2) )

fuzzy aggregation rule fe : H2n H2 that satisfies IIA3 and is nondictatorial.
Proof. Define fe : H2n H2 by H2n , x, y X,

1 if x = y,
fe()(x, y) =
t() if x 6= y.
Then clearly fe() is reflexive and complete. Let x, y, z X. Then for x 6= z,
fe()(x, z) = t() and fe()(x, y) + fe()(y, z) 1 t()
since both fe()(x, y) = 1 and fe()(y, z) = 1 is impossible. For x = z,

fe()(x, z) = 1 fe()(x, y) + fe()(y, z) 1. Thus fe() is T2 -transitive. Let
= (2) . Suppose i (x, y) > 0i N. Then x 6= y. Hence (x, y) =
1 fe()(y, x) = 1 t() > 0. Thus fe is weakly Paretian with respect (2) .
Clearly, fe satisfies IIA3 by a similar argument as in Proposition 4.1.20 .
4.2 Auxiliary Functions

In this section, we show how neutrality can be used to classify certain fuzzy
aggregation rules.
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 ,
fn ((a1 , . . . , an )) = a1 fn ((1, 0, . . . , 0) + . . . + an fn ((0, . . . , 0, 1)).
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
4.2. Auxiliary Functions 125
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 .
Proof. For i = 1, . . . , n, let 1i = (u1 , . . . , un ), where uj = 0, j = 1, . . . , n;

j 6= i, and ui = 1. Then there exists a unique linear transformation fc n of
n
R onto R such that fn (1i ) = wi , where wi = fn (1i ). Since fn is addi-
c
Pn Pn c Pn ci 1i ) =
tive, i=1 wi = i=1 fn (1i ) = fn ((1, . . . , 1)) 1. Now fn ( i=1
Pn
i=1 ci fn (1i ), where
c
Pn ci R, i =
Pn 1, . . . , n. Thus if c i Pn for i =
[0, 1]
1, . . . , n, then fn ( i=1 ci 1i ) = i=1 ci fn (1i ) [0, 1] since i=1 wi 1.
Let (a1 , . . . , an ) [0, 1]n . Then fc |
n [0,1] n ((a 1 , . . . , a n )) = f
c n ((a1 . . . , an )) =
,
Pn Pn
i=1 ai fn (1i ) = i=1 ai fn (1i ) = fn ((a1 , . . . , an )) since f is linearly decom-
c e
posable.
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.
Definition 4.2.6 A fuzzy aggregation rule fe is called anonymous if for all

(1 , . . . , n ) FRn (X) and for all permutations of N,
fe((1) , . . . , (n) ) = fe(1 , . . . , n ).
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,
fe()(u, v) = w1 1 (u, v) + w2 2 (u, v),
i i
i i
i i
K23798 2015/2/2
i i
where = (1 , 2 ). Let f2 be the auxiliary function associated with fe.

Then f2 (1 (u, v), 2 (u, v)) = fe()(u,
v). Let be an automorphism of [0, 1].
(a1 )+(a2 )
Suppose that f2 ((a1 , a2 )) = 1 2 for all a1 , a2 [0, 1]. Then
(a1 )+(a2 )
(f2 ((a1 , a2 ))) = 2 for all a1 , a2 [0, 1]. Thus
(f2 ((1 (x, y), 2 (x, y))))

(1 (x, y)) + (2 (x, y)) (2/3) + (1/3)
= =
2 2
(1/3) + (2/3) (1 (y, x)) + (2 (y, x))
= =
2 2
= (f2 ((1 (y, x), 2 (y, x)))).
Also,
(f2 ((1 (x, y), 2 (x, y))))

1 2 3 1 5
= (fe()(x, y)) = + =
4 3 4 3 12
and
(f2 ((1 (y, x), 2 (y, x))))

1 1 3 2 7
= (f ()(y, x)) =
e + = .
4 3 4 3 12
5 7

Hence 12 = 12 . However, this contradicts the fact that P
is one-to-one.
n
Thus P
no such automorphism exists. That is, we cannot obtain i=1 wi i (x, y)
n 1
from i=1 n i (x, y) by an automorphism of [0, 1].
4.3 Arrows Theorem and Max-Star

Transitivity
In this section, we consider the important work of Duddy, Perote-Pena, and
Piggins [17]. Max- transitivity plays a central roll in fuzzy preference theory.
Attention is restricted to fuzzy aggregation rules that satisfy counterparts of
unanimity and independence of irrelevant alternatives. In this section, the set
of triangular norms that permit preference aggregation to be non-dictatorial
is characterized. This set contains exactly those norms that contain a zero
divisor.
Many triangular norms do not possess zero divisors, e.g., max-min and
max-product. The Lukasiewicz triangular norm does possess zero divisors,
e.g., TL (x, y) = 0 when x = 12 , y = 12 .
A survey of the literature on social choice theory can be found in Salles
[40]. Various results can be found in [5, 12, 17, 31, 3, 9, 39, 18, 20, 34, 16,
i i
i i
i i
K23798 2015/2/2
i i
4.3. Arrows Theorem and Max-Star Transitivity 127
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.
Definition 4.3.1 Let fe be an FAR. Then fe is said to satisfy

(1) Independence (I) if for all (1 , . . . , n ), (01 , . . . , 0n ) S n and for all
x, y X, j (x, y) = 0j (x, y) for all j N implies (x, y) = 0 (x, y);
(2) Unanimity (U) if for all (1 , . . . , n ) S n , for all x, y X, and for
all v [0, 1], j (x, y) = v for all j N implies (x, y) = v;
(3) Neutrality if for all (1 , . . . , n ), (01 , . . . , 0n ) S n and for all x, y, z,
w X, j (x, y) = 0j (z, w) for all j N implies (x, y) = 0 (z, w).
An FAR fe is said to be dictatorial if there exists an individual i N

such that for all x, y X, and for every (1 , . . . , n ) S n , i (x, y) = (x, y).
The properties I, U and the dictatorship condition are stronger than the
conditions commonly used in the literature. The reader will easily observe this
in Chapter 5. There are two reasons for using them according to Duddy, Perot-
Pena, and Piggins [17]. In other work, I and U follow from the requirement
that a non-constant F AR cannot be manipulated, [34, 35, 36, 16]. Thus if
one accepts the formulation of strategy-proofness presented in these articles,
then I and U follow as logical consequences. Second, the weaker conditions
would require one to take a position on the factorization issue that was
mentioned earlier which is something that Duddy, Pena, and Piggins chose to
avoid. Statements of the standard conditions can be found in [40] and [4].
The purpose of this section is to prove the next theorem.
i i
i i
i i
K23798 2015/2/2
i i
Theorem 4.3.2 (Duddy, Perote-Pena, and Piggins [17]) If has no zero

divisors, then any FAR satisfying I and U is dictatorial. Moreover, if has a
zero divisor, then a non-dictatorial FAR exists that satisfies I and U.
We first prove sufficiency. The following lemma holds for any triangular
norm. The lemma generalizes Lemma 1 in [34]. See also [14, 15].
Lemma 4.3.3 ([17]) Let fe be an FAR. If fe satisfies I and U, then fe is

neutral under any triangular norm.
Proof. We consider five cases.
(1) If (a, b) = (c, d), then the result follows immediately from the fact that
fe satisfies Independence.
(2) Let (a, b), (a, c) X X. Let (1 , . . . , n ) S n be such that j (b, c) = 1
for all j N. Since fe satisfies Unanimity, (b, c) = 1, where = fe(). Since
is max- transitive, we have (a, c) (a, b). Since j (b, c) = 1 for all j N
and the j are max- transitive, it follows that j (a, c) j (a, b) for all j N.
Let (1 , . . . , n ) S n be such that j (b, c) = 1 and j (c, b) = 1 for all j N.
From the above argument, we have that (a, c) (a, b) and j (a, c) j (a, b)
for all j N. However, an identical argument shows that (a, b) (a, c) and
j (a, b) j (a, c) for all j N. Thus it must be the case that (a, b) = (a, c)
and j (a, b) = j (a, c) for all j N. Since (1 , . . . , n ) S n is arbitrary,
this condition holds for all profiles (1 , . . . , n ) S n such that j (b, c) = 1
and j (c, b) = 1 for all j N. Let F n denote the set of all such profiles. Let
1 , . . . , bn ) S n be such that bj (a, b) = bj (a, c) for all j N. Then there
(b
exists a profile (01 , . . . , 0n ) F n such that bj (a, b) = bj (a, c) = 0j (a, b) =
0j (a, c) for all j N. Since fe satisfies I, we have b(a, b) = b(a, c) = 0 (a, b) =
0 (a, c). Consider distinct profiles (001 , . . . , 00n ), (1 , . . . , n ) S n such that
00j (a, b) = j (a, c) for all j N. Then there exists a profile (
1 , . . . , n ) F
n
such that 00j (a, b) = j (a, c) =

j (a, b) = j (a, c) for all j N. Since f
e
satisfies I, we have 00 (a, b) = (a, c) = (a, b) = (a, c).
(3) (a, b), (c, b) X X. Let (1 , . . . , n ) S n be such that j (a, c) = 1
for all j N. Since fe satisfies U, we have (a, c) = 1. Since is max-
transitive, we have (a, b) (c, b). Since j (a, c) = 1 for all j N and
individual preferences are max- transitive, it follows that j (a, b) j (c, b)
for all j N. Let (1 , . . . , n ) S n be such that j (a, c) = 1 and j (c, a) = 1
for all j N. From the argument above, we have that (a, b) (c, b) and
j (a, b) j (c, b) for all j N. However, an identical argument shows that
(c, b) (a, b) and j (c, b) j (a, b) for all j N. Thus (a, b) = (c, b)
and j (a, b) = j (c, b) for all j N. Since (1 , . . . , n ) S n is arbitrary,
this condition holds for all profiles (1 , . . . , n ) S n such that j (a, c) = 1
and j (c, a) = 1 for all j N. Let G n denote the set of such profiles. Let
1 , . . . , bn ) S n be such that bj (a, b) = bj (c, b) for all j N. Then there
(b
exists (01 , . . . , 0n ) G n such that bj (a, b) = bj (c, b) = 0j (a, b) = 0j (c, b) for
i i
i i
i i
K23798 2015/2/2
i i
all j N. Since fe satisfies I, we have that b(a, b) = b(c, b) = 0 (a, b) =

0 (c, b). Consider distinct profiles (001 , . . . , 00n ), (1 , . . . , n ) S n such that
00j (a, b) = j (c, b) for all j N. Then there exists ( n
1 , . . . , n ) G such
that 00j (a, b) = j (c, b) =
j (a, b) = j (c, b) for all j N. Since f satisfies
e
00
I, we have that (a, b) = (c, b) = (a, b) = (c, b).
(4) (a, b), (c, d) X X with a, b, c, d distinct. Let (1 , . . . , n ) S n be
such that j (b, d) = j (d, b) = j (a, c) = j (c, a) = 1 for all j N. By U, we
have that (d, b) = 1. Since is max- transitive, we have (a, b) (a, d).
An identical argument shows that (a, d) (a, b) and so (a, b) = (a, d).
Since j (d, b) = j (b, d) = 1 for all j N and individual preferences are max-
transitive, it follows that j (a, b) = j (a, d) for all j N. This argument
can be repeated to show that (a, b) = (c, d) and j (a, b) = j (c, d) for all
j N. Since (1 , . . . , n ) S n is arbitrary, this condition holds for all profiles
(1 , . . . , n ) S n such that j (b, d) = j (d, b) = j (a, c) = j (c, a) = 1 for
all j N. Let G n be the set of all such profiles. Let (b 1 , . . . , bn ) S n be
such that bj (a, b) = bj (c, d) for all j N. Then there exists (01 , . . . , 0n ) G n
such that bj (a, b) = bj (c, d) = 0j (a, b) = 0j (c, d) for all j N. By I, we
have that b(a, b) = 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 00
1 , . . . , n ) G such that j (a, b) = j (c, d) = j (a, b) =

j (c, d) for all j N. By I, we have that (a, b) = (c, d) = (a, b) =

00
(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.
Proof. By Lemma 4.3.3, fe is neutral. Let (1 , . . . , n ) S n be such that

i (a, b) = 0 for all i N. Since fe satisfies U, we have that (a, b) = 0. Let
(01 , . . . , 0n ) S n be such that 0i (a, b) = 1 for all i N. By U, we have that
i i
i i
i i
K23798 2015/2/2
i i
0 (a, b) = 1. Consider the following sequence of profiles:
(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
We note that if is a triangular norm with a zero divisor, then there

exists a zero divisor x such that x x = 0 : There exists x, y (0, 1) such that
x y = 0. Suppose that x y. Then x x = 0 from the definition of t-norm.
Let M (a, b) = {x [0, 1] | at (1 , . . . , n ) S n there exists an i N such
that i (a, b) = x and i (a, b) j (a, b) for all j N }.
Let m(a, b) = {x [0, 1] | at (1 , . . . , n ) S n there exists an i N such
that i (a, b) = x and j (a, b) i (a, b) for all j N }.
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.
1. Let be a fuzzy preference relation. Show that if is not strongly

complete, then (2) is not asymmetric.
2. If is TM and strongly connected and is regular, prove that is
partially quasitransitive.
3. If is TM and strongly connected and is regular, prove that is
negatively transitive.
4.5 References
1. K. J. Arrow, Social Choice and Individual Values, Wiley, New York 1951.
2. A. Banerjee, Rational choice under fuzzy preferences: The Orlovsky

choice function, Fuzzy Sets and Systems, 53 (1993) 295299.
3. A. Banerjee, A Fuzzy preferences and arrow-type problems in social
choice, Soc. Choice Welf., 11 (1994) 121130.
i i
i i
i i
K23798 2015/2/2
i i
4.5. References 133
4. C. R. Barrett and M. Salles, Social choice with fuzzy preferences, Work-

ing paper. Centre for Research in Economics and Management, UMR
CNRS 6211, University of Cane, 2006.
6. C. R. Barrett, P. K. Pattanaik, and M. Salles, Rationality and aggre-
gation of preferences in an ordinally fuzzy framework, Fuzzy Sets and
Systems, 49 (1992) 913.
7. K. Basu, Fuzzy revealed preference, J. Econ. Theory, 32 (1984) 212227.
8. K. Basu, R. Deb, and P. K. Pattanaik, Soft sets: An ordinal reformula-
tion of vagueness with some applications to the theory of choice, Fuzzy
9. A. Billot, Economic Theory of Fuzzy Equilibria, Springer, Berlin 1995.
10. M. Dasgupta and R. Deb, Fuzzy choice functions, Soc. Choice Welf., 8
(1991) 171182.
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.
16. C. Duddy, J. Perote-Pena, and A. Piggins, Manipulating an aggregation

rule under ordinally fuzzy preferences, Soc. Choice Welf., 34 (2010)
411428.
17. C. Duddy, J. Perote-Pe na, and A. Piggins, Arrows theorem and max-
transitivity, Social Choice and Welfare, 36 (2011) 2534.
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.
19. B. Dutta, S. C. Panda, and P. K. Pattanaik, Exact choice and fuzzy

preferences, Math Soc. Sci., 11 (1986) 5368.
20. L. A. Fono and N. G. Andjiga, Fuzzy strict preferences and social choice,
21. J. L. Garcia-Lapresta and B. Llamazares, Aggregation of fuzzy prefer-

ences: Some results of the mean, Social Choice and Welfare, 17 (2009)
673690.
22. J. A. Goguen, L-fuzzy sets, J. Math. Anal. Appl., 18 (1967) 145174.
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.

30. S. V. Ovchinnikov, Structure of fuzzy binary relations, Fuzzy Sets and
Systems, 6 (1981) 169195.
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
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
4.5. References 135
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.
37. A. Piggins and M. Salles, Instances of indeterminacy, Analyse und Kri-

tik, 29 (2007) 311328.
38. C. Ponsard, Some dissenting views on the transitivity of individual pref-
erences, Ann. Oper. Res., 23 (1990) 279288.
39. G. Richardson, The structure of fuzzy preferences: Social choice impli-

cations, Soc. Choice Welf., 15 (1990) 359369.
40. M. Salles, Fuzzy utility. In: Barbara S., Hammond P.J., and Seidl C.
(eds), Handbook of utility theory, Vol. 1: Principles, Kluwer, Dordrecht,
1998.
41. A. K. Sen, Interpersonal aggregation and partial comparability, Econo-

metrica, 38 (1970) 393409.
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
Fuzzy Arrows Theorem
5.1 Dictatorial Fuzzy Preference Aggregation

Rules
Arrows Theorem [1] is one of the most important discoveries in social science
theory. Arrow asks if there are any methods for aggregating the preferences
of individuals that meet several reasonable conditions: universal admissibility,
transitivity, unanimity (weak Paretianism), and independence from irrelevant
alternatives. Through a series of formal proofs he concludes that the only
methods for achieving all four of these conditions are dictatorial, that is, the
social choice is perfectly aligned with the preferences of one individual.
A voting system is a function that maps the voting preferences of the voters
for the candidates to a ranking of the candidates. We list certain reasonable
assumptions one might think a perfect voting system should satisfy:
(i) The voting system preserves rationality, that is, the output of the voting
system is a total ordering.
(ii) The output of the voting system is determined only by the ranking
preferences of the voters; no other factors are allowed.
(iii) If all voters prefer one candidate to another, then the output of the
voting system is the favored candidate.
(iv) All candidates are treated equally.
(v) The ranking of two candidates in the output of the voting system is
independent of the voters preferences for a third candidate.
(vi) There is no dictator.
Arrows theorem says that these six axioms are inconsistent.
In the situation where fuzzy preference aggregation rules are allowed, there
are many types of transitivity, independence of irrelevant alternatives, strict
137
i i
i i
i i
K23798 2015/2/2
i i
138 5. Fuzzy Arrows Theorem
fuzzy preference relations, and so on. We examine how various combinations

of these concepts yield impossibility theorems.
We begin by considering a certain set of definitions for Arrows conditions.
This set of definitions does not lead to a different conclusion than that reached
by Arrow under crisp logic. We next consider applications of the fuzzy ver-
sion of Arrows Theorem involving representation rules, oligarchies, and veto
players. We also consider an approach that uses the notion of filters to further
strengthen the fuzzy version of Arrows Theorem. In the last three sections,
we consider approaches by Dutta, Banerjee, and Richardson.
Definition 5.1.1 Let be a fuzzy binary relation on X. Then is called a

fuzzy weak order on X if it is reflexive, complete, and partially transitive.
Definition 5.1.2 Let be a fuzzy binary relation on X. Define the fuzzy

subset of X X by x, y X, (x, y) = (x, y) (y, x).
Assume that each individual i has a fuzzy weak order i on X. Let FWR
denote the set of all fuzzy weak orders on X. When there is no indifference,
i.e., i (x, y) = 0 for all x, y X, x 6= y, individual is preferences i are
said to be strict. Let FWP denote the set of all strict orders on X. A fuzzy
preference profile on X is a n-tuple of fuzzy weak orders = (1 , . . . , n )
describing the fuzzy preferences of all individuals. Let FWRn denote the set
of all fuzzy preference profiles. Recall P(X) denotes the power set of X. For
any F WRn and for all S P(X), let eS = (1 eS , . . . , n eS ), where
i eS = i |SS , i = 1, . . . , n. For all FWRn and x, y X, let
P (x, y; ) = {i N | i (x, y) > 0} and R(x, y; ) = {i N | i (x, y) > 0}.
Let FB denote the set of all reflexive and complete fuzzy binary relations on
X.
Definition 5.1.3 A function fe : FWRn FB is called a fuzzy preference

aggregation rule.
We sometimes suppress the underlying fuzzy preference aggregation rule
and write (x, y) for fe()(x, y) and (x, y) > 0 for fe()(x, y) > 0 and
fe()(y, x) = 0.
Definition 5.1.4 Let fe be a fuzzy preference aggregation rule.

(1) fe is said to be a simple majority rule if FWRn , x, y X,
f ()(x, y) > 0 if and only if |P (x, y; )| > n/2.
e
(2) Let strict preferences be regular. fe is said to be a Pareto extension
rule if FWRn , x, y X, (x, y) > 0 if and only if R(x, b y; ) = N and
P (x, y; ) 6= , where R(x, y; ) = {i N | i (x, y) i (y, x)}.
b
i i
i i
i i
K23798 2015/2/2
i i
5.1. Dictatorial Fuzzy Preference Aggregation Rules 139
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.
Definition 5.1.5 Let fe be a fuzzy preference aggregation rule. Then

(1) fe is said to be nondictatorial if it is not the case that i N such
that FWRn , x, y X, i (x, y) > 0 implies (x, y) > 0;
(2) fe is said to be weakly Paretian if F WRn , x, y X [i N,
i (x, y) > 0 implies (x, y) > 0];
(3) fe is said to be independent of irrelevant alternatives IIA2 if
, 0 FRn , x, y X, Supp(i e{x,y} ) = Supp(0i e{x,y} ) i N implies
Supp(fe()e{x,y} ) = Supp(fe(0 )e{x,y} ).
In the 1992 U. S. election, Clinton won the election with approximately
43 percent of the vote while Bush had approximately 38 percent and Perot
19 percent. Had all Perot voters voted for Bush if Perot had not run, then
Bush would have won the election. This is a violation of the condition of
independence of irrelevant alternatives in the crisp case.
Definition 5.1.6 Let fe be a fuzzy preference aggregation rule. Let (x, y)

X X. Let be a fuzzy subset of N. Then
(1) is called semidecisive for x against y, written xD e y, if
n
FWR , [i (x, y) > 0 i Supp() and j (y, x) > 0 j / Supp()] implies
(x, y) > 0;
(2) is called decisive for x against y, written xD y, if FWRn ,
[i (x, y) > 0 i Supp()] implies (x, y) > 0.
Definition 5.1.7 Let fe be a fuzzy preference aggregation rule. Let be a

fuzzy subset of N. Then is called semidecisive (decisive) for fe if (x, y)
X X, is semidecisive (decisive) for x against y.
Proposition 5.1.8 Let fe be a fuzzy preference aggregation rule. Suppose fe

is dictatorial with dictator i. Let be a fuzzy subset of N. Suppose is a
coalition, i.e., |Supp()| 2. Then is decisive if and only if i Supp().
i i
i i
i i
K23798 2015/2/2
i i
Proof. Suppose is decisive. Suppose i

/ Supp(). Let x, y X be such
that x 6= y. Then FWRn such that j (x, y) > 0 j Supp() and
i (y, x) > 0. Thus (x, y) > 0 and (y, x) > 0, a contradiction. Hence i
Supp(). The converse is immediate.
In the next definition we restate Definition 2.2.16 since we are interested
in = in this chapter.
Definition 5.1.9 Let FR (X). Then is said to be partially quasi-

transitive if x, y, z X, (x, y) (y, z) > 0 implies (x, z) > 0, where
is a fuzzy strict preference relation associated with .

(1) fe is said to be (partially) transitive if FRn , fe() is (partially)
transitive;
(2) fe is said to be (partially) quasi-transitive if FRn , fe() is
(partially) quasi-transitive;
(3) fe is said to be (partially) acyclic if FRn , fe() is (partially)
acyclic.
In the remainder of the section, we assume that |X| 3 unless otherwise
specified.
Proposition 5.1.11 Let be a fuzzy binary relation on X. If is partially

transitive, then is partially quasi-transitive with respect to = (0) .
Proof. Let x, y, z X. Suppose (x, y) > 0 and (y, z) > 0. Then
(x, y) > 0, (y, x) = 0 and (y, z) > 0, (z, y) = 0. Hence (x, z) > 0.
Suppose (x, z) = 0. Then (z, x) > 0. Thus since (x, y) > 0, (z, y) > 0 by
partial transitivity, a contradiction. Thus (x, z) > 0.
Example 5.1.12 (Partial transitivity ; partial quasi-transitivity for =

(1) or = (3) ) Let X = {x, y, z}. Define the fuzzy relation on X as
follows:
(x, x) = (y, y) = (z, z) = 1, (x, y) = (y, z) = 2/3,

(y, x) = (z, y) = 1/3, (x, z) = (z, x) = 1/2.
Then is partially transitive since (u, v) > 0 u, v X. However, (x, y) >

0, (y, z) > 0, but (x, z) = 0.

independent of irrelevant alternatives IIA4 if x, y X, [, 0
FRn (X), Cosupp(i e{x,y} ) = Cosupp(0i e{x,y} ) i N Cosupp(f ()e{x,y} )
= Cosupp(f (0 )e{x,y} )].
i i
i i
i i
K23798 2015/2/2
i i
Proposition 5.1.14 Let , 0 FR(X). Let x, y X. Then Cosupp(e{x,y} )

= Cosupp(0 e{x,y} ) if and only if (u, v {x, y}, (u, v) < 1 0 (u, v) < 1).
Proof. Suppose Cosupp(e{x,y} ) = Cosupp(0 e{x,y} ). Let u, v {x, y}.

Then (u, v) < 1 (u, v) Cosupp(e{x,y} ) (u, v) Cosupp(0 e{x,y} )
0 (u, v) < 1. Conversely, suppose u, v {x, y}, (u, v) < 1 0 (u, v) <
1. Then (u, v) Cosupp(e{x,y} ) (u, v) < 1 0 (u, v) < 1
Cosupp(0 e{x,y} ).
For = (2) and x, y X, (x, y) > 0 if and only if (y, x) < 1. For
= (1) , or (3) , or (4) , or (5) , has the property that x, y X, (x, y) > 0
if and only if (x, y) > (y, x). This property plays a key role in the proof of
the next result.
Lemma 5.1.15 Let be a fuzzy subset of N. Let fe be a partially quasi-

transitive fuzzy preference aggregation rule that is weakly Paretian and either
independent of irrelevant alternatives IIA2 with strict preferences of type (0)
or independent of irrelevant alternatives IIA3 with regular strict preferences
or is independent of irrelevant alternatives IIA4 with strict preferences of type
(2) . If is semidecisive for x against y w.r.t , then (v, w) X X, is
decisive for v against w.
Proof. Suppose is semidecisive for x against y. Let be any fuzzy
preference profile such that i (x, z) > 0 i Supp(), where z / {x, y}. For
IIA3 and = (1) , or (3) , or (4) , or (5) , let 0 be a fuzzy preference profile
such that
0i (x, z) = i (x, z) and 0i (z, x) = i (z, x) i N,

i0 (x, y) > 0 i Supp() and j0 (y, x) > 0 j N \Supp(),
i0 (y, z) > 0 i N.
Since i (x, z) > 0 i Supp(), i0 (x, z) > 0 i Supp() by the definition

of 0 . Since xD e y, 0 (x, y) > 0. Since fe is weakly Paretian, 0 (y, z) > 0. Since
fe is partially quasi-transitive, 0 (x, z) > 0. Hence 0 (x, z) > 0 (z, x). Since
i e{x,z} = 0i e{x,z} i N and fe is IIA3, e{x,z} 0 e{x,z} . Thus (x, z) >
(z, x). (For strict preferences of type (0) and the crisp case, we write: Hence
0 (x, z) > 0 and 0 (z, x) = 0. Since Supp(i e{x,z} ) = Supp(0i e{x,z} ) i N
and fe is IIA2, Supp(e{x,z} ) = Supp(0 e{x,z} ). Thus (x, z) > 0 and (z, x) =
0.) (For strict preferences of type (2) , we write: Hence 0 (z, x) < 1. Since
Cosupp(i e{x,z} ) = Cosupp(0i e{x,z} ) i N and fe is IIA4, Cosupp(e{x,z} ) =
Cosupp(0 e{x,z} ). Thus (z, x) < 1.) Hence (x, z) > 0. Since was arbitrary,
xD z. Thus since z was arbitrary in X\{x, y},
z X\{x, y}, xD
e y xD z. (5.1)
i i
i i
i i
K23798 2015/2/2
i i
Since is decisive for x against z implies is semidecisive for x against z,

interchanging y and z in the above argument implies is decisive for x against
y.
For IIA3 and regular strict preferences, let 00 be any fuzzy preference
profile with i00 (y, z) > 0 i Supp() and let + be a fuzzy preference profile
such that
+
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().
Then i+ (y, z) > 0 i Supp(). Since xD z, + (x, z) > 0. Since fe is

weakly Paretian, + (y, x) > 0. Since fe is partially quasi-transitive, + (y, z) >
0. Since + 00 + 00
i e{y,z} = i e{y,z} i N and f is IIA3, e{y,z} e{y,z} . Since
e
+ + + 00 00
(y, z) > 0, (y, z) > (z, y). Hence (y, z) > (z, y). (For = (0)
and the crisp case, we write: Since + 00
i e{y,z} = i e{y,z} i N and f is
e
00
IIA2, Supp( e{y,z} ) = Supp( e{y,z} ). Since (y, z) > 0, + (y, z) > 0
+ +
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)
Now decisive for y against z implies is semidecisive for y against z. Thus

by (5.1), is decisive for y against x. We have (v, w) X X,
e y xD v (by (5.1)) xD
xD e v vD w (by (5.2)) with v replacing y.
We next present our first impossibility theorem.
Theorem 5.1.16 (Fuzzy Arrows Theorem) Let fe be a fuzzy aggregation rule.

Suppose strict preferences are of type (0) . Let fe be weakly Paretian, partially
transitive, and independent of irrelevant alternatives IIA2. Then fe is dictato-
rial.
Proof. Since fe is partially transitive, fe is partially quasi-transitive since
= (0) . By Lemma 5.1.15, 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 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.
i i
i i
i i
K23798 2015/2/2
i i
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
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.
Since is semidecisive for x against y and j (x, y) > 0 j Supp(),

(x, y) > 0. Since |Supp()| = m, it is not the case that (z, y) > 0 oth-
erwise 0 is semidecisive for z against y, where 0 (j) = (j) j Supp()\{i}
and 0 (i) = 0. (Suppose (z, y) > 0. Supp(i e{z,y} ) = Supp(0i e{z,y} ) i
N, 0 F Rn (X) Supp(fe()e{z,y} ) = Supp(fe(0 )e{z,y} ) 0 F Rn (X)
by independence of irrelevant alternatives (z, y) Supp(fe(0 )), (y, z) /
Supp(fe(0 )) 0 FRn (X) since this is the case for fe() and j (z, y) > 0
j Supp(0 ) and j (y, z) > 0 otherwise.) However, this contradicts the
minimality of m since |Supp(0 )| = m 1. Thus fe()(y, z) > 0 since fe() is
complete. Since fe is partially transitive, fe()(x, z) > 0. Suppose (x, z) = 0.
Then fe()(z, x) > 0. Since fe()(y, z) > 0, we have fe()(y, x) > 0 by par-
tial transitivity. However, this contradicts the fact that (x, y) > 0. Hence
(x, z) > 0. By independence of irrelevant alternatives, is semidecisive for
x against z, where (i) = (i) and (j) = 0 for j N \{i}. However, this
contradicts the fact that m > 1.
Definition 5.1.17 Let be a fuzzy relation on X. Then is said to be weakly

transitive if x, y, z X, (x, y) (y, x) and (y, z) (z, y) implies
(x, z) (z, x).
Example 5.1.18 Let X = {x, y, z}.

(a) Define the fuzzy relation on X as follows:
(x, x) = (y, y) = (z, z) = 1,

(x, y) = (y, z) = (x, z) = 1/2,
(y, x) = (z, y) = 1/2, (z, x) = 1.
Then is max-min transitive, but is not weakly transitive since (x, z)

(z, x).
(b) Define the fuzzy relation on X as follows:
(x, x) = (y, y) = (z, z) = 1,

(x, y) = (y, z) = (x, z) = 1/2,
(y, x) = (z, y) = 1/4, (z, x) = 1/8.
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:
(x, x) = (y, y) = (z, z) = 1,

(x, y) = (y, z) = 1, (x, z) = 0,
(y, x) = (z, y) = (z, x) = 0.
Then is weakly transitive, but not transitive in the crisp sense. Note that
is not complete.
Proposition 5.1.19 Suppose is exact. If is weakly transitive and com-

plete, then is transitive in the crisp sense.
Proof. Suppose (x, y) = 1 and (y, x) = 1. Since is weakly transitive,
(x, z) (z, x). Since is complete, (x, z) = 1.
Proposition 5.1.20 Let be a fuzzy relation on X. Then the following prop-

erties are equivalent:
(1) is weakly transitive.
(2) For all x, y, z X, (x, y) (y, x) and (y, z) (z, y) with strict
inequality holding at least once, then (x, z) > (z, x).
Proof. Suppose (1) holds. Assume that (x, y) (y, x) and (y, z) >
(z, y). Then (x, z) (z, x). Suppose (z, x) (x, z). Then (z, y)
(y, z) by (1), a contradiction. Thus (x, z) > (z, x). A similar argument
shows that (x, y) > (y, x) and (y, z) (z, y) implies (x, z) > (z, x).
Suppose (2) holds. Let x, y, z X. Suppose (x, y) (y, x) and (y, z)
(z, y). Suppose (z, x) > (x, z). Then by (2), (z, x) > (x, z) and (x, y)
(y, x), we have (z, y) > (y, z), a contradiction. Hence (x, z) (z, x).
Corollary 5.1.21 Let be a fuzzy relation on X. If is weakly transitive,

then is partially quasi-transitive.
We next consider an impossibility result when strict preferences are reg-

ular and with different types of transitivity and independence of irrelevant
alternative conditions.
Theorem 5.1.22 (Fuzzy Arrows Theorem) Let fe be a fuzzy aggregation rule.

Suppose strict preferences are regular. Let fe be weakly Paretian, weakly tran-
sitive, and independent of irrelevant alternatives IIA3. Then fe is dictatorial.
Proof. Since fe is weakly transitive, fe is partially quasi-transitive by the
Corollary 5.1.21. By Lemma 5.1.15, it suffices to show that i N, x, y X
such that {i} is semidecisive for x against y. (That is, with Supp() = {i}
i i
i i
i i
K23798 2015/2/2
i i
5.2. Representation Rules, Veto Players, Oligarchies, and Collegiums 145
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.
(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
i (x, y) > 0, i (y, z) > 0, i (x, z) > 0,

k
/ Supp(), k (y, z) > 0, k (z, x) > 0, k (y, x) > 0.
Since is semidecisive for x against y, i (x, y) > 0 i Supp(), and

j (y, x) > 0 j Supp(),
/ we have that (x, y) > 0. Since |Supp()| =
m, it is not the case that (z, y) > 0 otherwise 0 is semidecisive for z
against y, where 0 (j) = (j) j Supp()\{i} and 0 (i) = 0. (Suppose
(z, y) > 0. Then i e{z,y} 0i e{z,y} ) i N, 0 FRn (X) fe()e{z,y}
fe(0 )e{z,y} 0 FRn (X) by independence of irrelevant alternatives IIA3
fe(0 )(z, y) > fe(0 )(y, z) since fe()(z, y) > fe()(y, z). Thus 0 (z, y) > 0.)
However, this contradicts the minimality of m since |Supp(0 )| = m 1.
Since (z, y) = 0, fe()(y, z) fe()(z, y). Since fe()(x, y) > fe()(y, x),
fe()(x, z) > fe()(z, x) by weak transitivity. Hence (x, z) > 0. By indepen-
dence of irrelevant alternatives IIA3, is semidecisive for x against z, where
(i) = (i) and (j) = 0 for j N \{i}. However, this contradicts the fact
that m > 1.
5.2 Representation Rules, Veto Players,

Oligarchies, and Collegiums
Suppose the set of players must select a member to be its representative with
respect to certain decisions. For the fuzzy situation, we consider a represen-
tative rule to be a function from the set of fuzzy preference profiles into the
set of individual fuzzy singletons.
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
Definition 5.2.2 Let g be a representation rule. Then

(1) g is called dictatorial if and only if , 0 FRn , Supp(g()) =
Supp(g(0 ));
(2) g is called independent of irrelevant alternatives IIA2 if , 0
FRn , x, y X, Supp(i e{x,y} ) = Supp(0i e{x,y} ) i N implies
Supp(Supp(g()) e{x,y} ) = Supp(Supp(g(0 ) e{x,y} ).

FBRn , x, y X, i e{x,y} 0i e{x,y} i N implies
Supp(g()) e{x,y} Supp(g(0 )) e{x,y} .
The following theorem is a companion to Theorem 5.1.16.
Theorem 5.2.3 Let strict preferences be of type (0) . Let g be a representa-

tion rule. Then g is independent of irrelevant alternatives IIA2 if and only if
g is dictatorial.
Proof. Let fe be a fuzzy aggregation rule determined by g. Let FRn .
Then fe() = Supp(g()) . Since i is partially transitive i N, Supp(g())
is partially transitive. Thus fe() is partially transitive. Hence fe is partially
transitive. Let x, y X. Suppose i (x, y) > 0 i N. Then Supp(g()) (x, y) >
0, but Supp(g()) (x, y) = fe()(x, y). Thus fe is weakly Paretian. Since fe() =
Supp(g()) and fe(0 ) = 0Supp(g(0 )) , fe is independent of irrelevant alternatives
if and only if g is. Thus if g is independent of irrelevant alternatives, then
fe is dictatorial by fuzzy Arrows Theorem. Thus g is dictatorial. Conversely,
if g is dictatorial, then it follows easily that g is independent of irrelevant
alternatives.
Theorem 5.2.4 Let strict preferences be regular. Let g be a representation

rule. Then g is independent of irrelevant alternatives IIA2 if and only if g is
dictatorial.
Proof. Let fe be a fuzzy aggregation rule determined by g, i.e., fe()(x, y) =
Supp(g()) . Let FWRn . Then fe() = Supp(g()) . Since i is weakly
transitive i N, Supp(g()) = i0 is partially transitive. Thus fe() is
weakly transitive. Hence fe is weakly transitive. Let x, y X. Suppose
i (x, y) > 0 i N. Then Supp(g()) (x, y) > 0. Thus i0 (x, y) > i0 (y, x),
i.e., fe()(x, y) > fe()(y, x). Hence (x, y) > 0. Thus fe is weakly Paretian.
Since fe() = Supp(g()) and fe(0 ) = 0Supp(g(0 )) , fe is independent of irrele-
vant alternatives IIA2 if and only if g is. Thus if g is independent of irrelevant
alternatives, then fe is dictatorial by fuzzy Arrows Theorem. Thus g is dicta-
torial. Conversely, if g is dictatorial, then it follows easily that g is independent
of irrelevant alternatives.
i i
i i
i i
K23798 2015/2/2
i i
The Pareto extension rule is nondictatorial, weakly Paretian, indepen-

dent of irrelevant alternatives, and quasi-transitive. Such rules evade Arrows
theorem. However, if there are as many individuals as alternatives and each
alternative is preferred by some individual, then the Pareto extension rule
says that every alternative is as good as any other.
Definition 5.2.5 Let x, y X. An individual i N is said to have a veto

for x against y, written xVi y, if for every FRn , i (x, y) > 0 implies not
(y, x) > 0. An element i N is said to have a veto if for all x, y X, i has
a veto for x against y.
Definition 5.2.6 A fuzzy aggregation rule is called oligarchic if there exits

FN (called an oligarchy) such that
(1) every member of Supp() has a veto;
(2) is decisive.
Definition 5.2.7 Let x, y X. Let strict preferences be of type (0) . An

individual i N is said to have a semiveto for x against y, written xVei y, if
FRn , i (x, y) > 0 and j (y, x) > 0 j 6= i implies not (y, x) > 0.
Lemma 5.2.8 Let strict preferences be of type (0) . Suppose is partially

quasi-transitive. Then
(1) (x, y) > 0 and (y, z) > 0 implies (x, z) > 0;
(2) (x, y) > 0 and (y, z) > 0 implies (x, z) > 0.
Proof. (1) Suppose not (x, z) > 0. Since is complete, (z, x) > 0.
Since also (x, y) > 0, we have by partial quasi-transitivity that (z, y) > 0.
However, this is impossible since (y, z) > 0.
(2) The proof here is similar to that in (1).
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
i (x, y) > 0, i (x, z) > 0, i (y, z) > 0,

j 6= i, j (y, x) > 0, j (y, z) > 0.
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
i+ (y, x) > 0, i+ (y, z) > 0, i+ (x, z) > 0

j 6= i, j+ (z, x) > 0, j+ (y, x) > 0.
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.
The following theorem is a companion to Theorems 5.1.16 and 5.2.3.
Theorem 5.2.10 Let strict preferences be of type (0) . If a fuzzy aggregation

rule is partially quasi-transitive, weakly Paretian, and independent of irrele-
vant alternatives IIA2, then it is oligarchic.
Proof. Since fe is weakly Paretian, there exists FP(N ) such that
is semidecisive, namely = 1N . Let be such that Supp() is smallest for
which is semidecisive. Let m = |Supp()|. By Lemma 5.1.15, is decisive
for all ordered pairs (u, v) X 2 . If m = 1, then the proof is complete. Suppose
m > 1. Let x, y, z X be distinct. Let be a fuzzy preference profile such
that for some i Supp(),
i (x, y) > 0, i (x, z) > 0, i (y, z) > 0,
j Supp()\{i}, j (z, x) > 0, j (z, y) > 0, j (x, y) > 0, (5.5)

k
/ Supp(), k (y, z) > 0, k (y, x) > 0, k (z, x) > 0.
Since i Supp() and is decisive, (x, y) > 0. Since |Supp()| = m,
there is such a such that not (z, y) > 0 else Supp()\{i} would be
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.
Lemma 5.2.11 Let strict preferences be regular. Let fe be a fuzzy aggregation

rule which is partially transitive, weakly Paretian, and independent of irrel-
evant alternatives IIA3. 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
i (x, y) > 0, i (x, z) > 0, i (y, z) > 0,

j 6= i, j (y, x) > 0, j (y, z) > 0.
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)
y. Let 00 be any fuzzy preference profile such that i00 (y, z) > 0. Let + be a
i+ (y, x) > 0, i+ (y, z) > 0, i+ (x, z) > 0

j 6= i, j+ (z, x) > 0, j+ (y, x) > 0.
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)
(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.
Theorem 5.2.12 Suppose strict preferences are regular. If a fuzzy aggrega-

tion rule is partially quasi-transitive, weakly Paretian, and independent of
irrelevant alternatives IIA2, then it is oligarchic.
Proof. Since fe is weakly Paretian, there exists FP(N ) such that
is semidecisive, namely = 1N . Let be such that Supp() is smallest for
which is semidecisive. Let m = |Supp()|. By Lemma 5.1.15, is decisive
for all ordered pairs (u, v) X 2 . If m = 1, then the proof is complete. Suppose
m > 1. Let x, y, z X be distinct. Let be a fuzzy preference profile such
that for some i Supp(),
i (x, y) > 0, i (x, z) > 0, i (y, z) > 0,
j Supp()\{i}, j (z, x) > 0, j (z, y) > 0, j (x, y) > 0, (5.8)
k
/ Supp(), k (y, z) > 0, k (y, x) > 0, k (z, x) > 0.
there is such a such that not (z, y) > 0 else Supp()\{i} would be the
support of some fuzzy subset of N semidecisive for z over y and thus de-
cisive, contradicting the minimality of m. (Now for , 0 FRn satisfying
(5.8), Supp(i e{y,z} ) = Supp(0i e{y,z} ), i = 1, . . . , n.) Hence for any such ,
fe()(y, z) > fe()(z, y). Thus fe()(y, z) > 0 by completeness. Since fe is par-
tially 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 5.2.11. Hence fe is oligarchic since i was arbitrary.
Let L(fe) denote the set of all decisive coalitions associated with the ag-
gregation rule fe.
Definition 5.2.13 Let L FP(N ) be a family of coalitions.

(1) L is called monotonic if L, 0 FP(N ), 1N 0 implies
0
L;
(2) L is called proper if L, 0 FP(N )\L if Supp()Supp(0 ) =
.
i i
i i
i i
K23798 2015/2/2
i i
Lemma 5.2.14 Let fe be a fuzzy aggregation rule. Then L(fe) is monotonic

and proper.
Proof. Suppose L and 0 FP(N ) are such that 0 . Suppose
is not decisive for fe. Then there exists F Rn and x, y X such that
0
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 ) = .
Definition 5.2.15 Let fe be a fuzzy aggregation rule. If L (fe) L(fe) such

that 0 < |L (fe)| < , L (fe) 6= , then fe is called collegial.
Theorem 5.2.16 Suppose |X| n, where |N | = n. Let fe be a fuzzy aggrega-

tion rule. If fe is partially acyclic and weakly Paretian, then fe is collegial.
Proof. Since fe is weakly Paretian, 1N L(fe) so L(fe) 6= . Suppose fe is
not collegial. Then there exists L (fe) L(fe) such that 0 < |L (fe)| < and
(L (fe) )(i) = (i) = 0 i N and so L (fe) (i) = 0. Thus i N, there
exists L(fe) such that i / Supp(). By Lemma 5.2.14, L(fe) is monotonic
and so in particular i N, 1N \{i} L(fe) since 1N \{i} . Let the fuzzy
preference profile be such that
1 (xi , xi+1 ) > 0 for i = 1, . . . , n 1;
2 (xi , xi+1 ) > 0 for i = 2, . . . , n 1; 2 (xn , x1 ) > 0;
3 (xi , xi+1 ) > 0 for i = 3, . . . , n 1; 3 (xn , x1 ) > 0, 3 (x1 , x2 ) > 0;
...
n (xn , x1 ) > 0, n (xi , xi+1 ) > 0 for i = 1, . . . , n 2.
Thus for j = 2, . . . , n, P (xj1 , xj ; ) = N \{j} implying since 1N \{j} L(fe)

and Supp(1N \{j} ) = N \{j} that j (xj1 , xj ) > 0. Also P (xn , x1 ; ) = N \{1}
implying since 1N \{1} L(fe) and Supp(1N \{1} ) = N \{1}that (xn , x1 ) > 0.
Hence we have a cycle (xn , x1 ) > 0, (x1 , x2 ) > 0, . . . , (xn1 , xn ) > 0.
Thus fe is not partially acyclic. (Suppose = (0) . Then (x1 , x2 ) >
0, . . . , (xn1 , xn ) > 0 ; (x1 , xn ) > 0 since (xn , x1 ) > 0. For = (1) ,
(3) , (4) , or (5) , (x1 , x2 ) > 0, . . . , (xn1 , xn ) > 0 ; (x1 , xn ) (xn , x1 )
since (xn , x1 ) > 0.)
It is known that the converse of the previous result is not true. In Austen-
Smith and Banks [2], it is stated that there exist weakly Paretian and collegial
aggregation rules that are not acyclic, e.g., the post-1965 United Nations
Security Council.
i i
i i
i i
K23798 2015/2/2
i i
Lemma 5.2.17 Suppose |X| > n. Let fe be a fuzzy aggregation rule. If fe is

partially acyclic, weakly Paretian, and independent of irrelevant alternatives
IIA2 for strict preferences of type (0) and IIA3 for regular strict preferences,
then there exists i N and x, y X such that i has a veto for x against y.
Proof. Let X = {x1 , . . . , xm }, where m > n = |N |. Suppose it is not the
case that there exist i N and x, y X such that i has a veto for x against y.
Then i N, i has no veto for xi+1 against xi . Since fe is independent of irrel-
evant alternatives, there exists (i) that depends exclusively on (i) e{xi ,xi+1 }
(i)
such that i (xi+1 , xi ) > 0 and (i) (xi , xi+1 ) > 0. Now let (0) be such that
(0)
i N, i (xn+1 , x1 ) > 0. Pick any FRn such that e{x1 ,...,xn+1 } is given
as follows: for = (0) ,
Supp(e{xi ,xi+1 } ) = Supp((i) e{xi ,xi+1 } ) i = 1, . . . , n and

Supp(e{x1 ,xn+1 } ) = Supp((0) e{x1 ,xn+1 } ),
and for regular,
e{xi ,xi+1 } (i) e{xi ,xi+1 } , i = 1, . . . , n and e{x1 ,xn+1 } (0) e{x1 ,xn+1 } .
Since fe is independent of irrelevant alternatives IIA3 and by the construction

of , we have that
(x1 , x2 ) > 0, . . . , (xn1 , xn ) > 0, (xn , xn+1 ) > 0.

(0)
Since fe is weakly Paretian and i (xn+1 , x1 ) > 0 i N, we have that
(xn+1 , x1 ) > 0. However, for = (0) , this contradicts the fact that
(x1 , xn+1 ) > 0 since fe is partially acyclic. For regular strict preferences,
this contradicts the fact that (xn+1 , x1 ) (x1 , xn+1 ) since fe is partially
acyclic.
Definition 5.2.18 Let strict preferences be of type (0) . Let fe be a fuzzy

aggregation rule. Then fe is called neutral if , 0 FRn , x, y, u, v X,
P (x, y; ) = P (u, v; 0 ) and P (y, x; ) = P (v, u; 0 ) imply fe()(x, y) > 0 if and
only if fe(0 )(u, v) > 0.
Neutral fuzzy aggregation rules ignore the names of alternatives, whereas

anonymous fuzzy aggregation rules ignore the names of individuals.
Definition 5.2.19 Suppose strict preferences are regular. Let fe be a fuzzy

aggregation rule. Then fe is called regularly neutral if , 0 FRn ,
x, y, u, v X, P (x, y; ) = P (u, v; 0 ) and P (y, x; ) = P (v, u; 0 ) imply
fe()(x, y) > fe()(y, x) if and only if fe(0 )(u, v) > fe(0 )(v, u).
i i
i i
i i
K23798 2015/2/2
i i
5.3. Decisive Sets, Filters, and Fuzzy Arrows Theorem 153
Theorem 5.2.20 Suppose |X| > n. Let fe be a fuzzy aggregation rule.

(1) Let strict preferences be of type (0) . If fe is partially acyclic, weakly
Paretian, and neutral, then there exists i N such that i has a veto.
(2) Suppose strict preferences are regular. If fe is weakly transitive, weakly
Paretian, and regularly neutral, then there exists i N such that i has a veto.
Proof. By setting u = x and v = y in Definition 5.2.18, we see that fe
neutral implies fe is independent of irrelevant alternatives. Thus it follows by
Lemma 5.2.17 that some i has a veto for some pair of alternatives x, y. Hence
by Lemma 5.2.9 for (1) and Lemma 5.2.11 for (2), i has a veto for every pair
of alternatives u, v.
5.3 Decisive Sets, Filters, and Fuzzy Arrows

Theorem
We use the concepts developed in the previous section to further demonstrate
variants of Fuzzy Arrows Theorem. We begin by considering decisive sets and
filters. Let S be a set and F(S) be collection of fuzzy subsets of S. Recall
denotes the fuzzy subset of S defined by (i) = 0 for all i S. We consider
the following properties:
(P1) 1S F(S), / F(S);
(P2) 1 F(S) and 1 2 implies 2 F(S), where 2 is a fuzzy subset
of S;
(P3) 1 , 2 F(S) implies 1 2 F(S);
(P4) 1 , 2 , . . . , r F(T ) implies ri=1 i 6= ;
(P5) / F(S) implies 1S\Supp( ) F(S), where is a fuzzy subset of S.
Clearly, (P1) and (P3) imply (P4).
Note that if we restrict (P1) - (P5) to characteristic functions, then (P1)
- (P5) become the crisp case.
Definition 5.3.1 Let S be a set and F(S) be a collection of fuzzy subsets of

S. Then
(1) F(S) is called a filter if F(S) satisfies (P1), (P2), and (P3);
(2) F(S) is called a prefilter if F(S) satisfies (P1), (P2), and (P4);
(3) F(S) is called an ultrafilter if F(S) satisfies (P1), (P2), (P3), and
(P5).
Let fe be a fuzzy aggregation rule. If |X| n and fe is weakly Paretian,

then it follows that L(fe) is a prefilter.
Let L(fe) = { FP(N ) | is decisive for fe}.
Let FQ = { F B | is partially quasi-transitive}, FR = { F B | is
partially transitive}, and FR0 = { FB | is weakly transitive}.
i i
i i
i i
K23798 2015/2/2
i i
Theorem 5.3.2 Suppose fe is a fuzzy preference aggregation rule. Suppose

that fe is weakly Paretian and independent of irrelevant alternatives IIA2.
Then the following properties hold:
(1) If fe(FRn ) FQ, then L(fe) is a filter.
(2) If strict preferences are of type (0) and fe(FRn ) F R, then L(fe) is
an ultrafilter. If strict preferences are regular and fe(FRn ) FR0 , then L(fe)
is an ultrafilter.
Proof. Since fe is weakly Paretian, 1N L(fe). Now FRn such that
i N, i (x, y) > 0, we have that not (y, x) > 0 since fe is weakly Paretian.
Thus / L(fe). Hence (P1) holds. Now (P2) holds since L(fe) is decisive.
(1) Suppose fe(FRn ) F Q. Suppose , L(fe). Let x, y, z X. Let
be such that
i Supp() Supp(), i (x, z) > 0, i (z, y) > 0, i (x, y) > 0,
j Supp()\Supp(), j (x, z) > 0, j (y, z) > 0,
k N \Supp(), k (z, x) > 0, k (z, y) > 0.
Since L(fe) and i Supp(), i (x, z) > 0, we have (x, z) > 0. Since
L(fe) and i Supp(), i (z, y) > 0, we have (z, y) > 0. Since
fe(FRn ) FQ, (x, y) > 0 by partial quasi-transitivity. Since fe is inde-
pendent of irrelevant alternatives IIA2, is decisive for x against y. Since
x and y are arbitrary and Supp( ) = Supp() Supp(), L(fe).
Hence (P3) holds.
(2) Since FR FQ, (P3) holds by (1). Thus it remains only to show
that (P5) holds. Let FP(N ). Let x, y, z X and let be such that i
Supp(), i (x, z) > 0. Suppose 1N \L is not decisive for y against x, where L =
Supp(), i.e., not yD1N \L x. Since not yD1N \L x, there exists a fuzzy preference
profile 1 such that j N \L, j1 (y, x) > 0 and fe(1 )(x, y) > 0. Let 2 be a
fuzzy preference profile such that 2 e{x,y} = 1 e{x,y} and 2 e{x,z} = e{x,z}
and
i L, i2 (x, z) > 0, i2 (y, z) > 0,
j N \L, j2 (y, x) > 0, j2 (y, z) > 0.
Since fe is independent of irrelevant alternatives IIA2, fe(2 )(x, y) > 0. Since fe

is weakly Paretian, 2 (y, z) > 0. Since fe(FRn ) FR for = (0) , 2 (x, z) >
0. (For regular strict preferences, fe(FRn ) FR0 implies 2 (x, z) > 0 by
Proposition 5.1.20.) Since fe is independent of irrelevant alternatives IIA2, is
decisive for x against z, xD z. Let 0 be such that i Supp(), i0 (z, y) > 0.
Consider the fuzzy preference profile 3 such that 3 e{x,y} = 1 e{x,y} and
3 e{y,z} = 0 e{y,z} and also
i Supp(), i3 (z, y) > 0, i3 (z, x) > 0,
j N \Supp(), j3 (y, x) > 0, j3 (z, x) > 0.
i i
i i
i i
K23798 2015/2/2
i i
5.3. Decisive Sets, Filters, and Fuzzy Arrows Theorem 155
Since fe is independent of irrelevant alternatives IIA2 and 3 e{x,y} = 1 e{x,y} ,

fe(3 )(x, y) > 0. Since fe is weakly Paretian, 3 (z, x) > 0. Since fe(FRn ) FR
for = (0) , 3 (z, y) > 0. (For regular strict preferences,fe(FRn ) FR0
implies 3 (z, y) > 0 by Proposition 5.1.20.) Since fe is independent of irrelevant
alternatives IIA2, is decisive for z against y, zD y. Since z is arbitrary up to
z / {x, y}, we have by a similar argument as above that the following holds,
where L = Supp() :
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].
Theorem 5.3.3 (Fuzzy Arrows Theorem) Let fe be a fuzzy aggregation rule

that is weakly Paretian. If either strict preferences are of type (0) and fe is
IIA2 and partially transitive or strict preferences are regular and fe is IIA3
and weakly transitive, then fe is dictatorial.
Proof. By (2) of Theorem 5.3.2, L(fe) is an ultrafilter. Since N is finite
and Supp() is finite for all FP(N ), {Supp() | L(fe)} is finite. Thus
there exists a finite number 1 , . . . , k L(fe) such that ki=1 Supp(i ) =
L(fe) Supp(). Now ki=1 i 6= by (P3). Thus
L(fe) Supp() = ki=1 Supp(i ) 6= .
Let i L(fe) Supp(). Suppose that 1{i}

/ L(fe). Then by (P5), 1N \{i}
L(fe). Hence i e Supp() N \{i}, a contradiction. Thus 1{i} L(f ).
L(f )
e
Therefore fe is dictatorial.
We have shown that a fuzzy Arrows Theorem holds under an assump-
tion of partial transitivity. The only condition under which any method for
aggregating preferences is both weakly Paretian and independent of irrele-
vant alternatives is if it is dictatorial. This is true even when the definitions
of weakly Paretian, independence, and dictatorship incorporate fuzzy logic.
i i
i i
i i
K23798 2015/2/2
i i
These definitions are not terribly restrictive. For an aggregation function to

be dictatorial, the degree to which the group prefers a choice need not be at
the same intensity level as that of the individual whose choice is always pre-
ferred by the group. Hence the group might conceivably prefer a lower level
or even higher level of intensity. In analogous fashion, weakly Paretian merely
requires that every member of a group prefers a certain alternative over an-
other at various levels of intensity, the group need not prefer the alternative
at any fixed intensity level relative to the individual preferences as long as
the preference is positive. Similarly, independence from irrelevant alternatives
is released from any requirement that the degree of intensity with which one
alternative is preferred to another across two profiles be associated with the
degree of intensity of the social choice outcomes.
5.4 Fuzzy Preferences and Social Choice

We next turn our attention to some of the work of Dutta. In [11], Dutta
demonstrate that a fuzzy aggregation rule is oligarchic rather than dictatorial
when the definition of transitivity is relatively strong. However, if positive re-
sponsiveness is assumed, then the dictatorship result holds. Moreover, under a
weaker transitivity condition, these impossibility results, including oligarchics,
do not hold.
Let be an FWPR on X. Recall that is said to be T2 -transitive if for
all x, y, z X, (x, z) (x, y) + (y, z) 1. Recall also that
H1 = { FR(X) | is reflexive, strongly complete and

max-min transitive},
H2 = { FR(X) | is reflexive, strongly complete and T2 -transitive}.
A fuzzy weak preference relation on X is called exact (EWPR) if

Im() {0, 1}. Let R be an exact fuzzy preference relation on X. Let (x, y)
R. Then (x, y) P or (x, y) I and so R = P I, where P and I are
asymmetric and symmetric relations, respectively. Also, P I = .
Proposition 5.4.1 Let be a strongly complete FWPR on X. Suppose

satisfies the following conditions.
(1) = , where denotes maximum.
(2) is symmetric;
(3) is asymmetric;
(4) = .
Then either is an exact fuzzy preference relation on X or for all x, y X,
(x, y) = (y, x) = (x, y) = (y, x).
Proof. From (1) we have for all x, y X that
(x, y) = (x, y) (x, y)
(y, x) = (y, x) (y, x).
i i
i i
i i
K23798 2015/2/2
i i
5.4. Fuzzy Preferences and Social Choice 157
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)
Another possibility is for all x, y X,
(x, y) = 1 (y, x) and (x, y) = (x, y) (y, x). (5.10)
Both (5.9) and (5.10) fail to satisfy the requirements of = and =

. In view of Propositions 5.4.1 and 3.1.9, Dutta chose to use the derivation of
and outlined by (1) and (2) in Proposition 3.1.9. (See Propositions 3.1.8
and 3.1.11.) Thus we make the same choice in this section.
We next present some of these transitivity properties of fuzzy weak pref-
erence relations.
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:
(x, z) (x, y) (y, z) (5.11)
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
0 = (x, z) (x, z) (z, x). (5.15)
We consider two cases.

(1) Suppose
(x, y) (y, z) = (x, y). (5.16)
By (5.11),
(x, z) (x, y) (5.17)
By the maxmin transitivity of ,
(y, x) (y, z) (z, x). (5.18)
However, (5.16) and (y, x) (y, z) (y, x) (x, y) contradicting (5.13).

Similarly, (5.15), (5.19) and (y, x) (z, x) (y, x) (x, y). Hence
(y, z) (z, x) > (y, x), contradicting (5.18). Thus case (1) is not possible.
(2) Suppose
(x, y) (y, z) = (y, z). (5.19)
Then from (5.11),
(x, z) (y, z). (5.20)
By the max-min transitivity of ;
(z, y) (z, x) (5.21)
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 Let fe : H1n H1 be an FAR satisfying IIA1 and PC.

Then there exists a unique oligarchy C.
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,
(x, y) > 0 and (y, z) > 0 (x, z) > 0. (5.23)
By Proposition 5.4.4(2), it follows that maxmin transitivity implies that

(5.23) holds.
From Proposition 5.4.1, it follows that the above proposition is true when
the range of fe is expanded to include FWPRs which are max-star transitive
under any triangular t-norm. However, a corresponding expansion of the do-
main of fe may not be permissible because the proof requires PP-transitivity of
i , and this is not necessarily valid under every notion of max-star transitivity
as can be seen by Example 5.4.3.
If preferences are assumed to be exact, then the conditions imposed here
lead to Arrows [1] result. The fuzzy framework allows for the construction
of nondictatorial FARs as we saw in Chapter 4. Of course, fe is oligarchic, N
being the unique oligarchy.
One reason the Arrow Impossibility Theorem can be avoided in the fuzzy
framework is due to transitivity. Arrows result uses the fact that transitivity
of R implies PI and IP transitivity. However, as Example 5.4.3 shows, PI and
IP do not follow as corollaries of transitivity of the FWPR.
The following discussion is from Dutta [11]. In an intuitive sense, strict
preference under fe coincides with the Pareto dominance relation. This is clear
if it is assumed that individual preferences are exact, while allowing social
preferences to be fuzzy. The problem with the Pareto dominance relation is
that it is not strongly complete. Suppose that for some x, y X, and for some
i i
i i
i i
K23798 2015/2/2
i i
n-tuple of individual preferences, neither x nor y Pareto dominates the other.

Then fe()(x, y) = fe()(y, x) = 12 and is strongly complete. Thus fe leads
to a possibility result (see Theorem 4.1.23). The counterpart of fe in the crisp
case would not be strongly complete since (x, y) = (y, x) = 0 in this case.
However, there is no nondictatorial FAR satisfying positive responsiveness in
addition to the conditions like those imposed in Theorem 4.1.23.
Proposition 5.4.6 Let n 3, and fe : H1n H1 be a FAR satisfying IIA1,

PC and PR. Then fe is dictatorial.
Proof. By Proposition 5.4.5, there is a unique oligarchy C. If C consists
of a single individual, then that individual is a dictator. Thus let i, j C.
Let x, y be distinct alternatives in X. Consider (1 , . . . , n ) Hin such
that i (x, y) > i (y, x) and j (y, x) > j (x, y). Since i and j are vetoers,
(x, y) = (y, x) = 0.
Proposition 5.4.6 is the fuzzy counterpart of a result in Mas-Collel and
Sonnenschein [18].
The results proved so far in this section have assumed that individual
and social preferences are max-min transitive. T2 -transitivity allows one to
avoid impossibility results. The Gibbard oligarchy result is avoided in this
case because T2 -transitivity does not ensure that (5.23) is satisfied.
Theorem 4.1.23 and the previous results show that if we allow judgements
about social welfare to be fuzzy, Arrow-type impossibility theorems can be
avoided. Some scholars feel that the final social choice must be exact even
if social preferences are permitted to be fuzzy. Thus we next examine the
implications of these results here and in Chapter 7 for exact social choice. We
next consider the relationship between choice and the fuzzy social preferences.
Dutta et al. [12] discuss several alternative notions of rationalisability of exact
choice by fuzzy preferences in the context of revealed preference theory. We
present below a notion discussed by Dutta [11].
Definition 5.4.7 A function C : P (X) P (X) is called an exact choice

function (ECF) if for all A P (X), =
6 C(A) A.
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}.
Definition 5.4.9 An ECF C is H-rationalizable in terms of an FWPR

if for all A X, C(A) = B(A, ).
Under H-rationalizability, the player chooses those elements in A which
score the highest with respect to the function G(A, ).
i i
i i
i i
K23798 2015/2/2
i i
Definition 5.4.10 Let fe : T n T be an FAR and C an ECF. Then fe H-

generates C if and only if for all (1 , . . . , n ) T n , C is H-rationalizable in
terms of fe(1 , . . . , n ).
The following condition, first proposed by Bordes [8], is a well-know ratio-
nality condition in social choice.
Definition 5.4.11 An ECF C satisfies Property (+) if for all x, y X,

and all A, B X, (x, y A B and y C(A) and x C(B)) (y C(B)).
The following lemma, proved in Dutta et al. [12], shows the importance of
Property (+) in the present context.
Lemma 5.4.12 If C is an ECF which is H-rationalizable in terms of some

H1 , then C satisfies Property (+).
Proposition 5.4.13 (Dutta [11]) Let fe : H1n H1 be a nondictatorial FAR

satisfying IIA and PC and C an ECF which is H-generated by fe. Then there
exists A X and (1 , . . . , n ) H1n such that (i) x, y A and i (x, y) = 1
for all i N and (ii) y C(A).
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.
5.5 Fuzzy Preferences and Arrow-Type

Problems
Here we consider some interesting results of Banerjee [4]. In the special case,
where the individual preference relations are exact but the social preference
i i
i i
i i
K23798 2015/2/2
i i
5.5. Fuzzy Preferences and Arrow-Type Problems 163
relation is permitted to be fuzzy, it is possible to distinguish between different

degrees of power of the dictator, [4]. This power increases with the strength
of the transitivity requirement.
In the previous section, it is shown that if a relatively strong version of
transitivity is used, fuzzy aggregation rules which satisfy fuzzy analogs of Ar-
rows independence and Pareto conditions and whose range consists of fuzzy
binary relations would be oligarchic, but not necessarily dictatorial. If weaker
versions of transitivity are used, oligarchies can be avoided. However, if posi-
tive responsiveness is assumed, then dictatorship is obtained.
To get around this problem, Banerjee looked for an alternative way of
defining the union operator.
There are still other definitions of the union operator in the literature. The
choice between the alternative definitions should be decided with reference to
the context. In the present context it seems to Banerjee that 2 would be
pertinent in that it fulfills certain intuitive requirements.
However, if we take the 2 union operator, then Propositions 3.1.1 and
3.1.5 are unable to define and uniquely on the basis of a given . To get a
unique definition, we add two further conditions.
In this section, we assume that for all x, y X, (x, y) = 1 (y, x) and
(x, y) = (x, y) (y, x). Note that here is (2) defined in Definition 1.1.6.
Most of the theorems on social choice depend strongly on the transitivity
properties of the individual and the social preference relations. Applications
of fuzzy set theory to choice theory have made use of various definitions of
transitivity. Basu [7] introduced the following definition:
Definition 5.5.1 A fuzzy weak preference relation on X 2 is called T -

transitive if for all x, y, z X, (x, z) 12 (x, y) + 12 (y, z) for all
z X\{x, y} such that (x, y) 6= 0 and (y, z) 6= 0.
Proposition 5.5.2 If a weak preference relation R is T2 -transitive, then for

all x, y, z X, [(x, y) > 0 and (y, z) > 0] (x, z) > 0, where we recall
fuzzy preferences are of type (2) .
Let N be a finite set of individuals with cardinality not less than 2.
X is a set of alternatives. Let H, H1 ,and H2 be the set of reflexive and
strongly complete fuzzy weak preference relations satisfying T, max-min and
T2 -transitivity, respectively.
Definition 5.5.3 Let fe be a fuzzy aggregation rule. A subset M N is

almost decisive over x, y X if for all (1 , 2 , . . . , n ) T n , [j (x, y) > 0
for all j M and j (y, x) > 0 for all j
/ M ] (x, y) > 0.
In the propositions below, we consider H, H1 , or H2 .
i i
i i
i i
K23798 2015/2/2
i i
Proposition 5.5.4 (Banerjee [4]) Let fe : H2n H2 be a fuzzy aggregation

rule satisfying IIA1 and PC. Then there exists a dictator.
Proof. If a group of individuals G is almost decisive between a pair (x, y)
X, then it is decisive. The proof which uses Proposition 5.5.2 is similar to the
proof of the Field Expansion Lemma in the exact case (see Sen [28]).
We show that if G is a decisive group, a proper subset of G is a decisive
group. Even though the proof of this part is similar to that of the Group
Contraction Lemma in the exact case (see Sen 1985), we present it here for
later reference. Let G be decisive. Let G = G1 G2 , where G1 and G2 are
non-empty. Suppose that for all i G1 , i (x, y) > 0 and i (y, z) > 0 and for
all i G2 , i (y, z) > 0 and i (z, x) > 0. Suppose also for all i N \G that
i (z, x) > 0 and i (x, y) > 0, where x, y, z X. If (x, z) > 0, G1 is almost
decisive over (x, z). Hence, it is decisive. If (z, x) > 0, (x, z) = 0. Thus
(z, x) = 1 (x, z) = 1.
It follows that (y, x) > 0 since if (y, x) = 0, (x, y) = 1 and T2 -
transitivity would then imply that (z, y) = 1 and so (y, z) = 0. However,
this contradicts the fact that G is decisive. Thus in this case G2 is decisive.
(The roles of IIA and PC are similar to their roles in the exact case.)
As in the exact case, the proof of the proposition is completed by repeated
application of this group contraction result.
Corollary 5.5.5 Proposition 5.5.4 remains valid if H2 is replaced by either

H or H1 .
Proof. H H1 H2 .
It is very important to notice that the proof of Proposition 5.5.4 depends
on the fact that for all x, y X, (x, y) = 1 (y, x). This proof does not
apply if, given , we define by the relation: for all x, y X, (x, y) = (x, y)
if (x, y) > (y, x) and (x, y) = 0, otherwise. In fact, the counterexample
constructed in Dutta [11] for the case H1 , provides a counterexample to the
proposition since H1 H2 .
Consider the special case where the individual preference relations are
exact, but the social preference relation is fuzzy. Banerjee notes that it can
be considered to be the natural first step in extending the traditional exact
framework. For this case, the earlier definitions have to be restated. We assume
that individual exact preference relations are transitive. Denote the set of
transitive exact preference relations by U.
Definition 5.5.6 A fuzzy aggregation rule is a function fe : U n

FR(X).
Definition 5.5.7 Let fe be a fuzzy aggregation rule. Then fe is said to satisfy

the exact Pareto condition (EPC) if for all (1 , 2 , . . . , n ) U n and all
distinct x, y X, (x, y) = 1 if i (x, y) = 1 for all i N .
i i
i i
i i
K23798 2015/2/2
i i
5.5. Fuzzy Preferences and Arrow-Type Problems 165
In modifying the definition of a dictator, a complication arises. An individ-

ual i can be defined to be a dictator if for all x, y X, (x, y) > 0 whenever
i (x, y) = 1. It is easily seen that this definition of a dictator would reduce
to the traditional definition for U since in this case for all x, y X, 1 is the
only allowable positive value of (x, y). However, Banerjee raises an interest-
ing question concerning the power of a dictator. How powerful is individual i
as dictator if for all x, y X, (x, y) is positive but small, and (x, y) close to
1, whenever i (x, y) = 1?
This leads to the following definition which distinguishes between different
degrees of dictatorships. Definition 5.5.8(1) is a modification of Definition
4.1.13(1) and that following Definition 4.3.1.

(1) An individual j N is a dictator if for all distinct x, y X and for
all (1 , 2 , . . . , n ) U n , j (x, y) = 1 (x, y) > 0.
(2) An individual j N is an almost perfect dictator if for all distinct
x, y X and for all (1 , 2 , . . . , n ) U n and for any positive number < 1,
j (x, y) = 1 (x, y) .
(3) An individual j N is a perfect dictator if for all distinct x, y X
and for all (1 , 2 , . . . , n ) U n , j (x, y) = 1 (x, y) = 1.
Clearly, a dictator is not necessarily either an almost perfect dictator or a
perfect dictator. An almost perfect dictator is a dictator, but not necessarily
a perfect dictator.
The following result follows from Proposition 5.5.4.
Proposition 5.5.9 If fe : U n H2 is a fuzzy aggregation rule satisfying

IIA1 and EPC, there exists a dictator in the sense of Definition 5.5.8.
However, there does not necessarily exist a perfect dictator as the following
result shows.
Proposition 5.5.10 (Banerjee [4) There exists a fuzzy aggregation rule fe :

U n H2 which satisfies IIA1 and EPC and which is not perfectly dictatorial.
Proof. Define the fuzzy aggregation rule fe as follows:
For all x X and for all (1 , . . . , n ) U n , (x, x) = 1;
For all distinct x, y X and for all (1 , . . . , n ) U n ,
(i) (x, y) = 1 and (y, x) = 0 if i (x, y) = 1 and i (y, x) = 0 for all
i = 1, . . . , n;
(ii) (x, y) = and (y, x) = 1 , where (0, 1), if 1 (x, y) = 1 and
1 (y, x) = 0 and there exists j 6= 1 for which j (y, x) = 1;
(iii) (x, y) = 0 and (y, x) = 1 otherwise.
It follows that is strongly complete for all (1 , . . . , n ) U n . It is easily
seen that fe satisfies IIA1. Condition EPC follows from Case (i). Case (ii),
i i
i i
i i
K23798 2015/2/2
i i
where (x, y) = 1 (y, x) = < 1 shows that individual 1 is not a perfect

dictator.
It remains to be shown that for all (1 , . . . , n ) U n , H2 . However,
for later reference, we prove the stronger preposition that H1 , i.e., for
all x, y, z X, (x, z) (x, y) (y, z). Note that (x, y) can take any of
the three values 1, or 0. Similarly for (y, z). The inequality can be estab-
lished for all the combinations of values of (x, y) and (y, z). For instance,
let (x, y) = 1 and (y, z) = . Here (x, y) (y, z) = . Also i (x, y) = 1
for all i N and 1 (y, z) = 1 so that 1 (x, z) = 1. Thus (x, z) equals either
1 or . Hence the desired inequality follows. The other cases are similar.

(1) A group of individuals M N is called almost decisive of degree
over (x, y), where x, y X if for all (1 , . . . , n ) U n , [i (x, y) = 1 for all
i M and j (y, x) = 1 for all j
/ M ] (x, y) .
(2) A group of individuals M N is called decisive of degree if
for all distinct x, y X and for all (1 , . . . , n ) U n , [i (x, y) = 1 for all
i M ] (x, y) .
The proof of the following result is similar to the crisp case.
Proposition 5.5.12 (Banerjee [4]) Let fe : U n H1 be a fuzzy aggregation

rule satisfying IIA1 and EPC. Then there exists a unique individual who is
an almost perfect dictator.
The example constructed in the proof of Proposition 5.5.10 can be used

to establish the following result.
Proposition 5.5.13 There exists a fuzzy aggregation rule fe : U n H1

which satisfies IIA1 and EPC and which is not perfectly dictatorial.
The existence of a perfect dictator follows for H, however.

(1) A group M N is almost decisive over (x, y), if for all x, y X
for all (1 , . . . , n ) U n , (i (x, y) = 1 for all i M and i (y, x) = 1 for all
i N \M ) (x, y) = 1.
(2) M is perfectly decisive if for all x, y X and for all (1 , . . . , n )
U n , (i (x, y) = 1 for all i M ) (x, y) = 1.
Proposition 5.5.15 (Banerjee [4]) Let fe : U n H be a fuzzy aggregation

rule satisfying IIA1 and EPC. Then there exists a perfect dictator.
i i
i i
i i
K23798 2015/2/2
i i
5.6. The Structure of Fuzzy Preferences: Social Choice Implications 167
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.
5.6 The Structure of Fuzzy Preferences: Social

Choice Implications
In this section we consider some results of Richardson [26] involving strong
connectedness, Definition 5.6.1. Fuzzy preference relations are assumed to be
reflexive.
Banerjee [4] objects to this version of strict preference, (1) , on the intuitive
grounds that if (x, y) > 0, (x, y) should be less than (x, y). Banerjee states
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
x, y X, (x, y) = 1 (y, x) = 0, (5.26)
which is similar in appearance to the contrapositive of Banerjees condition,
x, y X, (x, y) = 0 (y, x) = 1. (5.27)
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
5.6. The Structure of Fuzzy Preferences: Social Choice Implications 169
Definition 5.6.1 is minimally (TM) transitive if x, y, z X, (x, y) =

1 and (y, z) = 1 (x, z) = 1.
Definition 5.6.2 is negatively transitive if x, y, z X, (x, y) > 0 or

(z, y) > 0.
Note that is negatively transitive if and only if
x, y, z X, (x, z) = 0 and (z, y) = 0 (x, y) = 0. (5.28)
Proposition 5.6.3 If is TM and strongly connected, then any regular is

partially quasitransitive.
Proof. If is strongly connected, regularity becomes
x, y X, (y, x) = 1 (x, y) = 0. (5.29)
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) .
Proposition 5.6.4 If is TM and strongly connected, then any regular is

negatively transitive.
Proof. By (5.28) and (5.29), we have that x, y, z X, (x, z) = 0 and
(z, y) = 0 (z, x) = 1 and (y, z) = 1 (y, x) = 1 (x, y) = 0.
We next consider social choice implications. Given two sets of fuzzy pref-
erence relations W and V, a fuzzy aggregation rule (FAR) is a function
f : W n V. The Pareto Condition (PC) and decisiveness involve the use
of strict preference. Hence they are particularly relevant in this section, and
need to be restated.
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 }.
Definition 5.6.6 For a given FAR, a coalition of voters C N is decisive

if U n and x, y X, C (x, y) > 0 (x, y) > 0,where C (x, y) > 0
means i (x, y) > 0 for all i C.
Definition 5.6.7 Let x, y X. For a given FAR, a coalition C N is

almost decisive over (x, y) if W n such that C (x, y) > 0, N \C (y, x) >
0, and (x, y) > 0.
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 171
Proof of the Group Contraction Lemma. Let D be a decisive group

of at least two members, and let D1 and D2 be non-empty and disjoint subsets
of D such that D1 D2 = D. Define the preference profile = (1 , . . . , n ) as
follows:
i D2 i (x, y) > 0, i (z, y) > 0, i (z, x) > 0,
i D1 i (x, y) > 0, i (y, z) > 0, i (x, z) > 0,
i/ D i (y, x) > 0, i (y, z) > 0, i (z, x) > 0.
Since D is decisive, (x, y) > 0. By Proposition 5.6.4, is negatively
transitive. Thus we must have either (x, z) > 0 in which case D1 is decisive,
or (z, y) > 0 in which case D2 is decisive.
Corollary 5.6.9 Let strict preference be defined by any regular . If fe : E n

SM satisfies IIA1 and PC, there exists a dictator.
Proof. The proof follows the pattern of that in Theorem 5.6.8.
Since no additional condition such as PR is required, and since the set of
exact transitive, and connected weak preference relations is a proper subset
of SM , Corollary 5.6.9 is a generalization of Arrows Theorem.
Without strong connectedness, the possibilities for nondictatorial FARs
when the transitivity requirement is T2 (or weaker) reappear, even when (1)
is replaced by (3) .
5.7 Exercises
1. Prove Lemma 5.4.12.
2. Prove that T-transitivity max-min transitivity T2 -transitivity.
3. Prove Proposition 5.5.2.

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

and Structure, Michigan Studies in Political Science, The University
choice, Soc Choice Welfare, 11 (1994) 121130.
5. S. Barbera and H. Sonnenschein, Preference aggregation with random-
ized social orderings, Journal of Economic Theory, 18 (1978) 244254.
6. C. R. Barrett, P. K. Pattanaik and M. Salles, On the structure of fuzzy

7. K. Basu, Fuzzy revealed preference theory, Journal of Economic Theory,
32 (1984) 212227.
8. G. Bordes, Consistency, rationality and collective choice, Review of Eco-

nomic Studies, 43 (1976) 447457
9. M. Dasgupta and R. Deb, Fuzzy choice functions, Soc Choice Welfare,
8 (1991)171182.
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
5.8. References 173
19. A. McLennan, Randomized preference aggregation: Additivity of power

and strategy proofness, Journal of Economic Theory, 22 (1972) 185192.

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.
24. S. V. Ovchinnikov, Structure of fuzzy binary relations, Fuzzy Sets and

Systems, 6 (1981) 169195.
25. S. V. Ovchinnikov, Representations of transitive fuzzy relations, in: H.J.
Skala, S. Termini and E. Trillas, eds., Aspects of Vagueness, D. Reidel,
Dordrecht, 1984.

cations, Soc. Choice Welf., 15 (1990) 359369.
27. A. K. Sen, Collective Choice and Social Welfare, Holden-Day, San Fran-
cisco, 1970.
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
Single Peaked Fuzzy

Preferences: Blacks
Median Voter Theorem
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
176 6. Single Peaked Fuzzy Preferences: Blacks Median Voter Theorem
models predicting the outcomes of numerous political games, to include the

outcome in presidential vetoes and the likelihood that a legislative committee
will report a bill to the floor of the parent assembly. (See for example [19,
10].)
We show that the Median Voter Theorem holds for fuzzy preferences.
Our approach considers the degree to which players prefer options in binary
relations. We demonstrate that Blacks Median Voter Theorem holds for fuzzy
single-dimensional spatial models even though there are significant differences
between conventional models and their fuzzy counterparts. In contrast to crisp
spatial models, where preferences in space are most often represented by a
single ideal point, in the fuzzy case, a political actors ideal policy position
encompasses a range of alternatives. Moreover, contrary to the conventional
approach, preferences are not assumed to monotonically decrease from the
actors ideal point.
6.1 Fuzzy Aggregation Preference Rules

In this section, we give some basic properties of fuzzy aggregation rules. We
first recall some notation. A weak order on X is a fuzzy binary relation
on X that is reflexive, complete, and (max-min) transitive. Assume that
each individual i has a weak order i on X. Let FT R denote the set of all
weak orders on X. A fuzzy preference profile on X is a n-tuple of weak
orders = (1 , . . . , n ) describing the preferences of all individuals. Let FT Rn
denote the set of all fuzzy preference profiles. For all FT Rn and x, y X,
let P (x, y; ) = {i N | i (x, y) > 0} and R(x, y; ) = {i N | i (x, y) > 0}.
Let FB denote the set of all reflexive and complete fuzzy binary relations
on X. In this chapter, a fuzzy preference aggregation rule is a function
fe : FT Rn FB.
We assume throughout the chapter that fuzzy strict preferences are of type
(0) .
Definition 6.1.1 Let be a fuzzy subset of N. Then is called a coalition

if |Supp()| 2.
Definition 6.1.2 Let fe be a fuzzy preference aggregation rule. Let L(fe) =

{ FP(N ) | is decisive for fe}.
If L(fe), then provides the degree to which a set of players in N

constitute a winning coalition.

is dictatorial with dictator i. Let be a fuzzy subset of N. Suppose is a
coalition. Then is decisive if and only if i Supp().
i i
i i
i i
K23798 2015/2/2
i i
6.1. Fuzzy Aggregation Preference Rules 177
Proof. Suppose is decisive. Suppose i / Supp(). Let x, y X be

such that x 6= y. Then FRn such that j (x, y) > 0 j Supp() and
i (y, x) > 0. Thus (x, y) > 0 and (y, x) > 0, a contradiction. Hence i
Supp(). The converse is immediate.
Proposition 6.1.3 says that if making collective decisions over a set of
alternatives results is possible with an individual i being a dictator, then
coalition is decisive if and only if that individual is at least partially a
member of the coalition.
Definition 6.1.4 Let fe be a fuzzy preference aggregation rule on FT Rn .

Define f : T Rn B by = (R1 , . . . , Rn ) T Rn , where T R is the set of
all crisp reflexive, complete, and transitive relations on X and B is the set of
all reflexive and complete crisp relations on X.
f () = {(x, y) X X | fe(1R1 , . . . , 1Rn )(x, y) > 0}.
Then f is called the preference aggregation rule associated with fe.

We have that (x, y) X X,
(x, y) f (R1 , . . . , Rn ) (x, y) Supp(fe(1R1 , . . . , 1Rn )).
Since fe(1R1 , . . . , 1Rn ) FB, it follows that (x, x) Supp(fe(1R1 , . . . , 1Rn ))

and so (x, x) f () for all x X. Also, x, y X, either (x, y) f ()
or (y, x) f () since either (x, y) Supp(fe(1R1 , . . . , 1Rn )) or (y, x)
Supp(fe(1R1 , . . . , 1Rn )) because fe(1R1 , . . . , 1Rn ) FB. That is, f () B.
Let FT Rn . Let fe() be denoted by . Let T Rn and f ()
be denoted by R. Let e = (1R1 , . . . , 1Rn ), where = (R1 , . . . , Rn ). Then
P (x, y; e) = P (x, y, ) since 1Ri (x, y) > 0 if and only if (x, y) Ri , i =
1, . . . , n.
In the following result, we make a strong connection between the crisp and
fuzzy case.
Proposition 6.1.5 Let fe be a fuzzy preference aggregation rule and f be the

preference aggregation rule associated with fe. Let be a fuzzy subset of N. If
is decisive with respect to fe, then Supp() is decisive with respect to f.
Proof. Let = (R1 , . . . , Rn ) T Rn and (x, y) X X. Suppose
(x, y) Pi i Supp(). Then 1Pi (x, y) > 0 i Supp(). Since
is decisive with respect fe, (x, y) > 0. Thus fe(1R1 , . . . , 1Rn )(x, y) > 0
and fe(1R1 , . . . , 1Rn )(y, x) = 0. Hence (x, y) Supp(fe(1R1 , . . . , 1Rn ) and
(y, x) / Supp(fe(1R1 , . . . , 1Rn ). Thus (x, y) f (R1 , . . . , Rn ) and (y, x)
/
f (R1 , . . . , Rn ). Hence xP y.
i i
i i
i i
K23798 2015/2/2
i i

(1) fe is called a simple majority rule if FRn , x, y X, (x, y) >
0 if and only if |P (x, y; | > n/2;
(2) fe is called a Pareto extension rule if FRn , x, y X,
(x, y) > 0 if and only if R(x, y; ) = N and P (x, y, ) 6= .
A slightly different version of the Pareto extension rule for fuzzy aggrega-
tion rules was presented in Definition 5.1.4. It was discussed there that this
rule was based on two rules. For the first, no one cares which of the two alter-
natives is chosen. Also it is in no ones interest to select y rather than x and
it is of interest for someone to strictly prefer x to y. Further discussion of the
Pareto extension rule can be found in Sen [28].
Proposition 6.1.7 Suppose fe is a simple majority rule. Let be a fuzzy

subset of N. If is decisive with respect to fe, then |Supp()| > n/2.
Proof. Since fe is a simple majority rule, it follows by comments preceding

Proposition 6.1.5 that f is a simple majority rule. Since is decisive with
respect to fe, we have that Supp() is decisive with respect to f by Proposition
6.1.5, where f is the aggregation rule associated with fe. Hence |Supp()| >
n/2.
Proposition 6.1.8 Suppose fe is a Pareto extension rule. Let be a fuzzy

subset of N. If is decisive with respect to fe, then Supp() = N.
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.
Definition 6.1.9 Let L FP(N ) be such that L, is a coalition. Then

(1) L is said to be monotonic if L, 0 FP(N ), N Supp(0 )
Supp() implies 0 L;
(2) L is said to be proper if L, 0 FP(N ), Supp(0 )Supp() =
implies 0
/ L.
Suppose L = L(fe) for some fuzzy aggregation rule. By monotonicity, it

follows that adding individuals to a decisive fuzzy subset yields another de-
cisive fuzzy subset. If L is proper, then two decisive fuzzy subsets must have
an individual that belongs to their support.
Proposition 6.1.10 Let fe be a fuzzy preference aggregation rule. Then

(1) L(fe) is monotonic;
(2) L(fe) is proper.
i i
i i
i i
K23798 2015/2/2
i i
6.1. Fuzzy Aggregation Preference Rules 179
Proof. (1) Suppose , 0 FP(N ) are such that Supp() Supp(0 ).

Suppose 0 is not decisive for fe. Then there exists FT Rn and x, y
X such that 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.
(2) Let L(fe). Let x, y X and FT Rn be such that i (x, y) > 0
for all i Supp() and i (y, x) > 0 for all i N \Supp(). Then (x, y) > 0.
Let 0 FP(N ). Suppose that Supp()Supp(0 ) = . If 0 L(fe), then
(y, x) > 0, a contradiction. Hence 0 is not decisive for fe for all 0 FP(N )
such that Supp(0 )Supp() = . Thus L(fe) is proper.
Definition 6.1.11 Let fe be a fuzzy aggregation rule. Then fe is called colle-

gial if L (fe) L(fe) such that 0 < |L (fe)| < , L (fe) 6= .
Since the members of a fuzzy collegium are at least partially members of

every fuzzy decisive set for a given fuzzy preference aggregation rule, their
agreement is needed for a decision to be made. Thus we may concentrate on
the set of alternatives which the fuzzy collegium favors in order to predict a
likely outcome. We will show that the fuzzy collegium for a fuzzy aggregation
rule in single peaked, single-dimensional models is the individual whose most
preferred alternatives lie at the median.

(1) fe is called acyclic if fe() is acyclic FT Rn ;
(2) fe is called partially acyclic if fe() is partially acyclic FT Rn .
The following result allows to obtain the result that a fuzzy preference
aggregation rule which is partially acyclic weakly Paretian is also collegial,
Proposition 6.1.14.
Proposition 6.1.13 Let fe be a fuzzy aggregation rule and f the aggregation

rule associated with fe.
(1) If fe is partially acyclic, then f is acyclic.
(2) If fe is weakly Paretian, then f is weakly Paretian.
Proof. (1) Since fe is partially acyclic, it follows that x1 , . . . , xk X,

1P (x1 , x2 ) > 0, . . . , 1P (xk1 , xk ) > 0 implies 1R (x1 , xk ) > 0, where P
and R correspond to f. Thus x1 P x2 , . . . , xk1 P xk implies x1 Rxk since
1P (xi1 , xi ) > 0 (xi1 , xi ) Supp(1P ) = P, i = 2, . . . , n.
(2) Since fe is weakly Paretian, T Rn , we have that x, y X, (i N,
1Pi (x, y) > 0 implies 1P (x, y) > 0). Thus i N, xPi y implies xP y.
The following result states that any fuzzy preference aggregation rule that
is partially acyclic and weakly Paretian results in a fuzzy collegium.
i i
i i
i i
K23798 2015/2/2
i i
Proposition 6.1.14 If fe is partially acyclic and weakly Paretian, then fe is

collegial.
Proof. By Proposition 6.1.13, f is acyclic and weakly Paretian. Thus

L(fe) Supp() 6= by Proposition 6.1.5 and [2, Theorem 2.4, p. 43]. Hence
e e
e 6= L (f ) L(f ) such that 0 < |L (f )| < .
L (f )
e
6.2 Fuzzy Voting Rules

Recall that we let be of type (0) throughout the chapter. We consider the
case for regular in the exercises at the end of the chapter.
Simple rules form a subset of the set of voting rules. Voting rules in-
clude majority rules, super majority rules, and the unanimity rule. We de-
fine the characteristics of each class. We show that fuzzy simple rules pro-
duce a maximal set located at the median, which is generalizable to the core
in n-dimensional space, n > 1, Theorem 6.3.9. Consequently, fuzzy single-
dimensional models predict an outcome at the median, in a manner analo-
gous to conventional models, when fuzzy simple rules are used to aggregate
the preferences of individuals.
Let R X X. We let Symm(R) denote the symmetric closure of R, i.e.,
the smallest subset of X X that contains R and is symmetric. One reason
for the form the following definition takes is for it to correspond with the
fact that strict preferences are of type by (0) . In the exercises at the end of
this chapter, the form is different in order to accommodate strict preferences
which are regular.
Definition 6.2.1 Let fe be a fuzzy aggregation rule. Let L FP(N ). Let

FT Rn and set P() e = {(x, y) X X | L, i Supp(),
i (x, y) > 0}. Define feL : FT Rn FB by FRn , x, y X,

1 if (x, y)
/ Symm(P()),
e
{ (x, y) | L, i Supp(), (x, y) > 0}
i i
feL ()(x, y) =

if (x, y) P(),
e

0 if (x, y) Symm(P())\
e P().
e
In the following example, we illustrate the definition of feL . It becomes clear

how the definition corresponds to the crisp case in Austen-Smith and Banks
[2, p. 59]. For example, those elements not in Symm(P()) e must include all
elements of the form (x, x) and we wish feL to be reflexive.
i i
i i
i i
K23798 2015/2/2
i i
6.2. Fuzzy Voting Rules 181
Example 6.2.2 Let X = {x, y, z}, N = {1, 2, 3}, and = (1 , 2 , 3 ), where

i (x, x) = i (y, y) = i (z, z) = 1, i = 1, 2, 3 and
1 (x, y) = 0.7, 1 (x, z) = 0.2, 1 (y, z) = 0.5,

2 (x, y) = 0.5, 2 (x, z) = 0.3, 2 (z, y) = 0.6,
3 (y, x) = 0.7, 3 (z, x) = 0.5, 3 (y, z) = 0.2.
Each i is defined to be 0 otherwise, i = 1, 2, 3. Let L = {, 0 }, where (1) =

0.5, (2) = 0.2, (3) = 0 and 0 (1) = 0, 0 (2) = 0.7, 0 (3) = 0.5. Now
Supp() = {1, 2} and Supp(0 ) = {2, 3}. Hence P() e = {(x, y), (x, z)} and
Symm(P()) = {(x, y), (y, x), (x, z), (z, x)}. Thus it follows that
e
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.
Definition 6.2.3 Let fe be a fuzzy aggregation rule. Then fe is called a partial

fuzzy simple rule if F Rn , x, y X, fe()(x, y) > 0 if and only if
feL(fe) ()(x, y) > 0.
The following result supports the fact preceding the definition and also
helps explain our definition of a partial fuzzy simple rule.

rule associated with fe. If fe is a partial fuzzy simple rule, then f is a simple
rule.
Proof. Let FT Rn and let x, y X. Recall that P()
e = {(x, y)
X X | L(f ), i Supp(), i (x, y) > 0} and
e
L(fe)
= { FP(N ) | is decisive forfe}
= { FP(N ) | FRn , i (x, y) > 0 i Supp() (x, y) > 0}.
Let = (1R1 , . . . , 1Rn ), where Ri R, i = 1, . . . , n. Then

1 if (x, y) / Symm(P()),
e

{1Pi (x, y) | L(fe), i Supp(), 1Pi (x, y) = 1}

feL(fe) ()(x, y) =

if (x, y) P(),
e

0 if (x, y) Symm(P())\ P().
e e
i i
i i
i i
K23798 2015/2/2
i i
Thus for = (1R1 , . . . , 1Rn ),

(
1 if (x, y) P()
e (X X\Symm(P())),
e
fL(fe) ()(x, y) =
e
0 if (x, y) Symm(P())\P().
e e
Now f (R1 , . . . , Rn ) = {(x, y) | fe(1R1 , . . . , 1Rn )(x, y) > 0} = Supp(fe(1R1 ,

. . . , 1Rn )) = Supp(feL(fe) (1R1 , . . . , 1Rn )) = P()
e (X X\Symm(P())).
e By
definition,
fL(f ) (R1 , . . . , Rn ) = P() (X X\Symm(P())),
where P() = {(x, y) | L L(f ), i L, xPi y} and = (R1 , . . . , Rn ).

For = (1R1 , . . . , 1Rn ), L(fe), i Supp(), xPi y (Pi corresponds to
1Ri ) L L(f ), i L, xPi y (Pi corresponds to Ri ), i.e., given , let L =
Supp() and given L, let = 1L . Hence P((1 e R , . . . , 1R )) = P((R1 , . . . , Rn )).
1 n
Thus P(1R1 , . . . , 1Rn ) = P(R1 , . . . , Rn ). Therefore, f = fL(f ) .

e
Definition 6.2.5 Let fe be a fuzzy aggregation rule. Then

(1) fe is called decisive if , 0 F Rn and x, y X, [P (x, y; ) =
P (x, y; 0 ) and (x, y) > 0] imply 0 (x, y) > 0;
(2) fe is called monotonic if , 0 F Rn and x, y X, [P (x, y; )
P (x, y; 0 ), R(x, y; ) R(x, y; 0 ) and (x, y) > 0] imply 0 (x, y) > 0.
A fuzzy aggregation rule is decisive if the set of individuals strictly pre-

ferring x to y remains the same under a different preference profile, then the
social choice remains the same in that x is strictly preferred to y.
An aggregation rule is monotonic if x is preferred to y, then increasing the
support for x over y or decreasing the opposition does not change the social
preference.
We next characterize partial fuzzy simple rules.
Theorem 6.2.6 Let fe be a fuzzy aggregation rule. Then fe is a partial fuzzy

simple rule if and only if fe is decisive, neutral, and monotonic.
Proof. Let x, y X. Suppose L(fe) (x, y) > 0. Then L(fe) such that
i Supp(), i (x, y) > 0. Thus (x, y) > 0. Suppose fe is decisive, neutral,
and monotonic. Let F T Rn . Suppose (x, y) > 0. In order to show fe
is a partial fuzzy simple rule, it suffices to show P (x, y; ) L(fe) for then
L(fe) (x, y) > 0. Let u, v X. Let be any fuzzy preference profile such that
i P (x, y; ), i (u, v) > 0. Let L+ = P (u, v; )\P (x, y; ). Let 0 be a fuzzy
preference profile such that P (u, v; 0 ) = P (x, y; ) and P (v, u; 0 ) = P (y, x; ).
Since fe is neutral, we have that 0 (u, v) > 0. Let 00 F T Rn be defined by
00i e{u,v} = i e{u,v} if and only if i L+ P (x, y; ) and 00j e{u,v} = 0j e{u,v}
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 )
Definition 6.2.7 Let fe be a fuzzy aggregation rule. The decisive structure

of fe, denoted D(fe), is defined to be the set D(fe) = {(, ) FP(N )FP(N ) |
Supp() Supp() and x, y X, FT Rn , i (x, y) > 0 i Supp()
and j (x, y) > 0 j Supp() implies (x, y) > 0}.
It is worth recalling the crisp situation. An aggregation rule is a plurality
rule if x, y X, alternative x is chosen over alternative y when the number
of individuals strictly preferring x to y exceeds the number strictly preferring
y to x. The plurality rule is a voting rule, but not a simple rule. Majority rule
is both a simple rule and a voting rule.
Definition 6.2.8 Let D F P(N ) FP(N ). Then D is called monotonic

if (, ) D, Supp() Supp( 0 ) Supp( 0 ) and Supp() Supp()
Supp( 0 ) imply ( 0 , 0 ) D for all ( 0 , 0 ) FP(N ) F P(N ).
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.
Definition 6.2.9 Let fe be a fuzzy aggregation rule. Let D F P(N )F P(N )

be such that (, ) D implies Supp() Supp(). Let FT Rn and set
FP D () = {(x, y) X X | (, ) D, i Supp(), i (x, y) > 0 and j
Supp(), j (x, y) > 0}. Define feD : FT Rn FB by FT Rn , x, y X,

1 if (x, y) / Symm(FP D ())
{ (x, y) j (x, y) | (, ) D, i (x, y) > 0, i Supp(),

i
feD ()(x, y) =

j (x, y) > 0 j Supp(j )} if (x, y) FP D (),
0 if (x, y) Symm(FP D ())\FP D ().

Example 6.2.10 Let X = {x, y, z}, N = {1, 2, 3}, and = (1 , 2 , 3 ), where

i (x, x) = i (y, y) = i (z, z) = 1, i = 1, 2, 3 and
1 (x, y) = 0.7, 1 (x, z) = 0.2, 1 (y, z) = 0.5,

2 (x, y) = 0.5, 2 (x, z) = 0.3, 2 (z, y) = 0.6,
3 (x, y) = 0.1, 3 (y, x) = 0.7, 3 (z, x) = 0.5, 3 (y, z) = 0.2.
i i
i i
i i
K23798 2015/2/2
i i
Each i is defined to be 0 otherwise, i = 1, 2, 3. Let D = {(, ), ( 0 , 0 )},

where (1) = 0.5, (2) = 0.2, (3) = 0, (1) = 0.5, (2) = 0.2, (3) = 0.3,
0 (1) = 0, 0 (2) = 0.7, 0 (3) = 0.5, and 0 = 0 . Now Supp() = {1, 2},
Supp() = N, and Supp( 0 ) = {2, 3} = Supp( 0 ). Now i (x, y) > 0 i
Supp() and i (x, y) > 0 i Supp(). Thus (x, y) FP D (). Now i (x, z) >
0 i Supp(), but 3 (x, z) = 0 and 3 Supp(). Hence (x, z) / FP D ().
Note also that (y, z) / FP D () since it is not the case that i (y, z) > 0
i Supp() or i0 (y, z) > 0 i Supp( 0 ). Thus it follows that FP D () =
{(x, y)}, Symm(FP D ()) = {(x, y), (y, x)} and
fe()(x, y) = {i (x, y) j (x, y) | i = 1, 2; j = 1, 2, 3}

= {0.7 0.7, 0.7 0.5, 0.7 0.1, 0.5 0.7, 0.5 0.5, 0.5 0.1}
= 0.7.
We also have that feD ()(u, v) = 1 if (y, x) 6= (u, v) 6= (x, y) and that fe(y, x) =
0.
Definition 6.2.11 Let fe be a fuzzy aggregation rule. Then fe is called a par-

tial fuzzy voting rule if FT Rn and x, y X, fe()(x, y) > 0 if and
only if feD(fe) ()(x, y) > 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.

rule associated with fe. If fe is a partial fuzzy voting rule, then f is a voting
rule.
Proof. Let = (R1 , . . . , Rn ) Rn and let = (1R1 , . . . , 1Rn ). Then
(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 ,
where S = Supp(1Supp() ), W = Supp(1Supp() ) for the implication and

= 1S and = 1W for the implication . Thus
(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.
Theorem 6.2.13 Let fe be a fuzzy aggregation rule. Then fe is a partial fuzzy

voting rule if and only if it is neutral and monotonic.
Proof. Suppose fe is a partial fuzzy voting rule. We show that fe is
monotonic. We first note that D(fe) is monotonic. Let (, ) D(fe) and
0 , 0 FP(N ) be such that Supp() Supp( 0 ) Supp( 0 ) and Supp()
Supp() Supp( 0 ). Since Supp() Supp( 0 ) and Supp() Supp( 0 ),
clearly ( 0 , 0 ) D(fe). Let , 0 FT Rn and x, y X be such that
P (x, y; ) P (x, y; 0 ), R(x, y; ) R(x, y; 0 ), and (x, y) > 0. By Defini-
tions 6.2.9 6.2.9 and 6.2.11 6.2.11, we have that since (x, y) > 0, (, )
D(fe) such that Supp() Supp() and x, y X, FT Rn , i (x, y) > 0
i Supp() and j (x, y) > 0 j Supp(). By hypothesis,
{i N | i (x, y) > 0} {i N | i0 (x, y) > 0},

{j N | j (x, y) > 0} {j N | 0j (x, y) > 0}.
Since ( 0 , 0 ) D(fe), 0 (x, y) > 0. Thus fe is monotonic. We now show fe is

neutral. Let , 0 FRn and x, y, z, w X be such that
{i N | i (x, y) > 0} = {i N | i0 (z, w) > 0},
{i N | i (y, x) > 0} = {i N | i0 (w, z) > 0}.
Then it follows easily that i (x, y) > 0 i0 (z, w) > 0 i N. Suppose

fe()(x, y) > 0. Then (x, y) (X X\Symm(FP D(fe) ()) F P D(fe) (), say
(x, y) FP e (). Then (, ) D(fe) such that i Supp(), i (x, y) >
D(f )
0, j Supp(), i (x, y) > 0}. By hypothesis, it follows that (z, w) FP D(fe)
(0 ). Hence fe(0 )(z, w) > 0 and in fact 0 (z, w) > 0. Suppose (x, y) X
X\Symm(FP D(fe) ()). By the argument just given, (z, w) / FP D(fe) (0 ) else
(x, y) FP e (). Thus (z, w) X X\FP e (0 ). Hence fe(0 )(z, w) > 0.
D(f ) D(f )
Thus fe is neutral.
For the converse, we first show that D(fe) (x, y) > 0 implies (x, y) > 0.
Suppose feD(fe) ()(x, y) > 0 for F Rn and x, y X. Then (x, y) (X
X\Symm(FP D(fe) ()) FP D(fe) (), say (x, y) F P D(fe) (). Thus (, )
D(fe) such that i Supp(), i (x, y) > 0, j Supp(), i (x, y) > 0}. Since
(, ) D(fe), fe()(x, y) > 0. Suppose (x, y) (X X)\Symm(FP e ()). D(f )
Then (y, x) (X X)\Symm(FP D(fe) ()) and so feD(fe) ()(y, x) = feD(fe) ()(x,
y) = 1. However, this contradicts the assumption that D(fe) (x, y) > 0. Thus
(y, x) Symm(FP e ())\FP( e ()). Hence it follows that fe()(y, x) = 0.
D(f ) D(f )
Now suppose that fe is neutral and monotonic. Let x, y X and F Rn .
i i
i i
i i
K23798 2015/2/2
i i
Suppose that (x, y) > 0. Let (, ) F P(N )FP(N ), Supp() Supp(),

be such that Supp() = S and Supp() = W, where S = P (x, y; ) and
W = R(x, y; ). We wish to show (, ) D(fe). Let z, w X. Let 0
FT Rn be such that i0 (z, w) > 0 i S and 0i (z, w) > 0 i W,
i.e., P (z, w; 0 ) = S and R(z, w; 0 ) = W. Since fe is neutral, 0 (z, w) > 0.
Now let 00 F T Rn be such that P (z, w; 0 ) P (z, w; 00 ) and R(z, w; 0 )
R(z, w; 00 ). Since fe is monotonic, 00 (z, w) > 0. Thus we have that i (z, w) >
0 i S and i (z, w) > 0 i W implies (z, w) > 0. Hence (, ) D(fe)
and so D(fe) (x, y) > 0.
6.3 Single-Peaked Fuzzy Profiles

We first consider single-peakedness. We then develop the argument relating
fuzzy fe-medians to single-peakedness in order to introduce the notion of a core,
a maximal set (set of undominated alternatives) under some fuzzy preference
aggregation procedure. Our main result, Theorem 6.3.9, says that the core is
located at the median if preferences of all political actors are single-peaked in
single-dimensional policy space.
A strict ordering on a set X is a relation on X that is irreflexive, tran-
sitive, and asymmetric.
Definition 6.3.1 Suppose |X| = n 2. Suppose Q is a strict ordering of X.

Label X so that at X, at+1 Qat , t = 1, 2, . . . , n 1. Let FT R. Then
is called single-peaked on X with respect to Q if
(at , at+1 ) > 0, (at+1 , at+2 ) > 0, . . . , (an1 , an ) > 0 and

(at , at1 ) > 0, (at1 , at2 ) > 0, . . . , (a2 , a1 ) > 0.
Recall that in Definition 6.3.1 is max-min transitive.

An approach to the problem of cyclical majorities is that of single-peakedn-
ess. If the set of individual preferences satisfy a certain unimodal pattern,
called single-peaked preference, then majority decisions must be transitive
regardless of the number individuals holding any of the possible orderings, as
long as the total number of persons is odd.
Definition 6.3.2 Let FT Rn . Then is said to be single-peaked on X if

there exists a strict ordering Q of X such that for all i N, i is single-peaked
on X with respect to Q.
Let FS denote the set of all single-peaked fuzzy profiles.
i i
i i
i i
K23798 2015/2/2
i i
6.3. Single-Peaked Fuzzy Profiles 187
Definition 6.3.3 Let FT Rn . Suppose is single-peaked with respect to

a strict ordering Q. Define xi to be that element of X such that i (xi , y) > 0
e , L
y X (xi 6= y), i = 1, . . . , n. Define L e + : X FP(N ) by z X,
i N,

e (z)(i) = i (xi , z) if zQxi , e + (z)(i) = i (xi , z) if xi Qz,
L , L
0 otherwise, 0 otherwise.
For any FT Rn , let xi denote individual is most preferred alternative

from X and for all z X, Le (z)(i) is the degree to which xi is strictly preferred
to z when zQxi . Similar comments hold for L e (z) is the fuzzy
e + (z)(i). Then L
subset of X which gives the degree of the individuals whose ideal points lie to
e + (z) denotes the fuzzy subset of X which gives the degree
the left of z and L
of the individuals whose ideal points lie to the right of z.
Definition 6.3.4 Let fe be a fuzzy preference aggregation rule. Let

FT Rn . Suppose is single-peaked with respect to a strict ordering Q. Then an
element z of X is called an fe-median if Supp(L e (z))
/ {Supp() | L(fe)}
+
and Supp(L (z))
e / {Supp() | L(f )}.
e
Definition 6.3.5 Let F : [0, 1]2 [0, 1]. FR (X), FP (X),

x X, define the fuzzy subset MF (, ) of X by x X,
MF (, )(x) = {t [0, 1] | F ((x, y), (x)) t y Supp()}.
The fuzzy subset MF (, ) is called the maximal fuzzy subset of and

with respect to F.
Definition 6.3.6 Let fe be a fuzzy preference aggregation rule. Let F :

[0, 1]2 [0, 1]. Define FS the fuzzy subset F,fe(; Q) of X by z X,

{F (fe()(z, y), 1) | y X} if z is an fe-median
F,fe(, Q)(z) =
0 otherwise.
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.
Options Player 1 Votes Player 2 Votes Player 3 Votes Results

x vs y x y y y
x vs z x x or z z x or z
y vs z y y z y
We see that alternative y is chosen if the voting system is majoritarian. Thus

the median voter y votes in favor of the outcome that wins the election.
If F = , we use the notation Cfe() for CF,fe(), fe()(x) for F,fe()(x),
and M for MF .
Example 6.3.8 Let F = . Let X = {x, y, z} and N = {1, 2, 3}. Let =

(1 , 2 , 3 ) FR3 be such that
1 (x, y) = 0.1, 1 (y, z) = 0.7, 1 (x, z) = 0.9,

2 (y, x) = 0.2, 2 (x, z) = 0.6, 2 (y, z) = 0.9,
3 (z, y) = 0.5, 3 (y, x) = 0.7, 3 (z, x) = 1.
Let Q be such that a3 Qa2 Qa1 , where a3 = z, a2 = y, and a1 = x. It follows

that is single-peaked with respect to Q, where for 1 , t = 1, for 2 , t = 2, and
for 3 , t = 3. Now x1 = x, x2 = y, and x3 = z since 1 (x, w) > 0 w X,
w 6= x, 2 (y, w) > 0 w X, w 6= y, 3 (z, w) > 0 w X, w 6= z. Hence
e (x)(1) = 0 since not xQx, L

L e (x)(2) = 0 since not xQy,
L (x)(3) = 0 since not xQz, L (y)(1) = 1 (x, y) since yQx,
e e
e (y)(2) = 0 since not yQy, L
L e (y)(3) = 0 since not yQz,
e (z)(1) = 1 (x, z) since zQx, L
L e (z)(2) = 2 (y, z) since zQy,

L (z)(3) = 0 since not zQz,
e
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
6.3. Single-Peaked Fuzzy Profiles 189
{2, 3}, {1, 2, 3}}. Since

e (x)) = , Supp(L
Supp(L e (y)) = {1}, Supp(L e (z)) = {1, 2},
e + (x)) = {2, 3}, Supp(L
Supp(L e + (y)) = {3}, Supp(L e + (z)) = ,
it follows that Supp(L e (y))

/ {Supp() | L(fe)} and Supp(L e + (y))
/
{Supp() | L(f )}. Hence y is an f -median. Thus fe(; Q)(y) =
e e
fe()(y, x) fe()(y, z). Now M (fe(), 1X )(y) = {t [0, 1] | fe()(y, w) t
w X} = fe()(y, x) fe()(y, z). It follows that
fe()(y, x) = 2 (y, x) 3 (y, x) = 0.2 0.7 = 0.7,
fe()(y, z) = 1 (y, z) 2 (y, z) = 0.7 0.9 = 0.9.
Hence fe()(y, x) fe()(y, z) = 0.7.

The median voter theorem explains things that occur in majoritarian vot-
ing systems. In the United States, the Democratic and Republican candidates
usually move their platforms to the middle during election campaigns. Politi-
cians tend to adopt similar platforms and rhetoric during campaigns. That is,
they tailor platforms to the median voter. The median voter theorem reflects
that radical candidates usually are not elected.
The median voter theorem also explains the rise in government programs
over the past several decades. See for example Husted and Lawrence [17].
Voters with the median income take advantage of their status as voters by
electing politicians who propose taxing those who earn more than the median
voter and then redistributing the money. A detailed account can be found in
Rice [25] and Husted and Lawrence [17]. Also Holcombe analyzed the Bowen
equilibrium level of education expenditure in over 200 Michigan school dis-
tricts and found that actual expenditures were only about 3% away from the
estimated district average. Limitations of the median voter theorem can be
found in Krehbiel [18] and Buchanan and Tollisin [7]. A large problem for me-
dian voter theory is that government systems are composed of individuals who
are self-interested. To continue to obtain their own set of goals, individuals
focus on their re-election rather than their constituents wishes.
We now present our main conclusion.
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
Thus if z is an fe-median, CF,fe()(z) = F,fe()(z). That is, MF (fe(), 1X ) =

F,fe() on Supp(F,fe()). Hence it suffices to show Supp(MF (fe(), 1X )) =
Supp( e()). Let x Supp(MF (fe(), 1X )). Then F (fe()(x, y), 1X (x)) =
F,f
F (fe()(x, y), 1) > 0 for all y X. Suppose F,fe() = 0, i.e., x /
Supp(F,fe()). Let Q be a strict ordering of X with respect to which is
single-peaked. Since X is finite, we are free to label X so that at+1 Qat for all
t 1 (and t |X|). Then x = at for some t 1 and x is not an fe-median
since x / Supp(F,fe()). Thus either Supp(L e (x)) {Supp() | L(fe)} or
Supp(L e + (x)) {Supp() | L(fe)}. Assume the former. Then Supp(L e (x))

= Supp() for some L(f ). Hence u, v X, i (u, v) > 0 i L
e e (x) im-
plies (u, v) > 0. Since is single-peaked, i (at1 , at ) > 0 i Supp(Le (x)).
Thus since L(f ) is decisive, f ()(at1 , at ) > 0, i.e., f ()(at1 , x) > 0. But
e e e
then fe()(x, at1 ) = 0 and so from the assumption that F (0, 1) = 0, we
have F (fe()(x, at1 ), 1) = 0, a contradiction. It therefore follows that x
Supp(F,fe()). Hence
Supp(MF (fe(), 1X )) Supp(F,fe()).
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
Supp(MF (fe(), 1X )) Supp(F,fe()).
Since Supp(F,fe()) Supp(MF (fe(), 1X )), we have Supp(MF (fe(), 1X )) =

Supp( e()). Hence MF (fe(), 1X ) = e().
F,f F,f
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.
Corollary 6.3.10 Let F = . Let fe be a fuzzy aggregation rule. If fe is a

partial fuzzy simple rule, then FS, Cfe() = fe().
i i
i i
i i
K23798 2015/2/2
i i
6.4. The Core in Fuzzy Spatial Models 191
6.4 The Core in Fuzzy Spatial Models

We rely on the results in [23] for this section. In crisp spatial models, pre-
dictions concerning voting outcomes rely heavily on the existence of a core.
Individuals choosing among a set of alternatives by majority rule will not be
able to arrive at a stable choice in the absence of a core. Most voting rules will
permit another option to defeat the previously chosen one. Such problems are
particularly problematic to majority rule spatial models at dimensionalities
greater than one. We feel that fuzzy spatial models might offer a partial so-
lution to this problem. We present the notion of a fuzzy core. We find that a
fuzzy core is more likely in two or more dimensions as the number of players
increases.
We make use of fuzzy set theory due to the disconnect between predictions
and empirical reality. We consider the basic rationale underlying the notion of
a core. A core in conventional spatial models comprises the set of predictable
outcomes. Without a highly improbable set of conditions or the imposition of
severe restrictions on models, the core is empty under sincere voting whenever
the dimensionality of models exceeds one, [20].
We present conditions under which the fuzzy core equals the fuzzy subset
of fe-medians, Theorem 6.4.11, where fe is a fuzzy aggregation rule. We also
give conditions under which the fuzzy core is nontrivial, Theorem 6.4.32.
Recall that FR (X) = FP (X X) and N denotes the set of positive
integers. Let denote a t-norm on [0, 1]. A fuzzy subset of a set S is said to
have the inf property if for all subsets S 0 of S, there exists x S 0 such that
(x) = {(y) | y S 0 }. In this section, we define a class of fuzzy maximal
subsets which contain these previously defined fuzzy maximal subsets. The
strict ordering on X taken in the finite case, [20], is now taken to be the strict
inequality > of R, the set of real numbers.
As previously mentioned, strict preferences are of type (0) .
Recall that a fuzzy weak order on X is a fuzzy binary relation on X that
is reflexive, complete, and (max-min)-transitive and that FT R denotes the
set of all fuzzy weak orders on X.
Let F T Rn . x, y X, recall that R(x, y; ) = {i N | i (x, y) 0}
and P (x, y; ) = {i N | i (x, y) > 0}.
We use the following definition of a maximal fuzzy subset. Let C be a
function from X X into [0, 1]. Define M : FR (X) FP (X) FP (X)
by FR (X), FP (X),
M (, )(x) = (x) (({t [0, 1] | C(x, w) t (x, w) w Supp()}))
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
If C(x, w) = (w) (w, x) x, w X, then M is the maximal fuzzy subset

defined in [11] and Chapter 1. Recall that we denote this maximal fuzzy subset
by MG . If C(x, w) = 1 x, w X, then M is the maximal fuzzy subset defined
in [21] and Chapter 1. Recall that we denote this maximal fuzzy subset by
MM .
Proposition 6.4.1 Let C be a function from X X into [0, 1]. Let

FP (X) and F R (X). Then {{t [0, 1] | C(x, w) t (x, w)} | w
Supp()} = {t [0, 1] | C(x, w) t (x, w) w Supp()}.
Proof. We have that {t [0, 1] | C(x, w) t (x, w) w Supp()} =

t0 C(x, w) t0 (x, w) w Supp() and @t1 [0, 1] such that C(x, w)
t1 (x, w) w Supp() and t1 > t0 . Let w Supp() and let tw = {t
[0, 1] | C(x, w) t (x, w)}. Then
{tw | w Supp()} = {{t [0, 1] | C(x, w) t (x, w)} | w Supp()}.
Let t1 = {tw | w Supp()}. Then C(x, w) t1 (x, w) w Supp() and

so t1 t0 . Since tw t0 w Supp(), t1 t0 . Thus t0 = t1 .
Let x X and F R (X). Let Sx = {w X | C(x, w) > (x, w)}. If
Sx 6= , let tx,w = {t [0, 1] | C(x, w) t (x, w)}, w Sx .
Proposition 6.4.2 Let FR (X) and x X. Then

1 if Sx = ,
M (, 1X )(x) =
{tx,w | w Sx } if Sx 6= .
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.
Definition 6.4.3 Let X R, x X, and F T Rn . Then is said to be

single-peaked on X if i N, xi X such that
(1) y X\{xi }, i (xi , y) > 0;
(2) y, z X, y < z < xi implies i (z, y) > 0;
(3) y, z X, xi < z < y implies i (z, y) > 0.
Throughout this section, we assume X R unless otherwise specified. Let

FS denote the set of all single-peaked fuzzy preference profiles on X.
i i
i i
i i
K23798 2015/2/2
i i
Definition 6.4.4 Let FS. Let xi be that element of X such that

e , L
i (xi , y) > 0 y X\{xi }, i N. Define L e + : X FP(N ) by z X,
i N,

e (z)(i) = i (xi , z) if xi < z, e + (z)(i) = i (xi , z) if xi > z,
L L
0 otherwise, 0 otherwise.
e (z)(i) represents the degree to which player i prefers a position to the
L
left of alternative z and L e + (z)(i) represents the degree to which player i
prefers a position to the right of alternative z.
Definition 6.4.5 Let fe be a fuzzy preference aggregation rule. Let FS.

Then an element z of X is called an fe-median if
e (z))
Supp(L / {Supp() | L(fe)} and
e + (z))
Supp(L / {Supp() | L(fe)}.
An fe-median passes through any alternative for which there is no plausible
alternative either to its right or left that is even minimally preferred by some
majority coalition.
We now develop the concept of a fuzzy core and connect it to a nonempty
fuzzy maximal subset.
Definition 6.4.6 Let fe be a fuzzy aggregation rule. Define Cfe : FT Rn

FP(X) by FT Rn , Cfe() = M (fe(), 1X ). Then Cfe is called the fuzzy
core of .
In previous chapters, the set of alternatives X did not possess any structure
such as an ordering of its elements. With single peaked preferences this is no
longer true. The maximal outcomes are located with respect to the ordering Q
yielding preferences single-peaked. Thus the core of a preference aggregation
rule fe at a profile is the set of maximal elements in X under the binary
relation fe().
Definition 6.4.7 Let fe be a fuzzy aggregation rule. Define fe : FT Rn

FP(X) by FT Rn , z X,

1 if Sz = and z is an fe-median,
fe()(z) = {tz,w | w Sz } if Sz 6= and z is an fe- median,

0 otherwise.
If C(x, w) = 1 x, w X, then fe becomes

{fe()(z, y) | y X} if z is an fe-median,
fe()(z) =
0 otherwise.
The following examples illustrate these concepts.
i i
i i
i i
K23798 2015/2/2
i i
Example 6.4.8 Let C(x, w) = 1 x, w X, where X = [0, 1]. Let N =

{1, 2, 3}. Let fe be a partial fuzzy simple rule, i.e., fe()(x) > 0 feL(fe) ()(x) >
0. Let L(fe) = {{1, 2}, {1, 3}, {2, 3}, N }. Let xi X be player is preferred
alternative, i = 1, 2, 3, where x1 < x2 < x3 . For all i N, define i F Rn
by y, z X,

1 if |z xi | = |y xi |,
i (z, y) = 1 |z xi | if |z xi | < |y xi |,
0 if |z xi | > |y xi |.

It follows that i is reflexive and complete, i = 1, 2, 3. We next show i is max-

min transitive i N. Let u, v, w X and i N. If either i (u, v) = 0 or
i (v, w) = 0, then clearly i (u, v) i (v, w) i (u, w). Suppose i (u, v) > 0
and i (v, w) > 0. Then |u xi | |v xi | and |v xi | |w xi |. Thus
|u xi | |w xi |. Hence it follows that i (u, v) i (v, w) i (u, w).
Now
Supp(L e (x)) = if x x1 ,
Supp(L e (x)) = {1} if x1 < x x2 ,
Supp(L e (x)) = {1, 2} if x2 < x x3 ,
Supp(L e (x)) = {1, 2, 3} if x3 < x,
Supp(L e + (x)) = {1, 2, 3} if x < x1 ,
Supp(L e + (x)) = {2, 3} if x1 x < x2 ,
Supp(L e + (x)) = {3} if x2 x < x3 ,
Supp(L e + (x)) = if x3 x.
Also
e (x1 )) = , Supp(L
Supp(L e + (x1 )) = {2, 3},
e (x2 )) = {1}, Supp(L
Supp(L e + (x2 )) = {3},
e + (x3 )) = .
Supp(L (x3 )) = {1, 2}, Supp(L
e
Thus {x2 } is the set of all fe-medians.
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.

Then fe() is reflexive and complete. (Note that if we let fe()(x2 , 0) = 0

or fe()(x2 , 1) = 0, then fe() is not complete since fe()(0, x2 ) = 0 and
fe()(1, x2 ) = 0.) However, {fe()(x, y) | y X} = 0 for all x X. Thus
fe()(x) = 0 for all x X, where C(x, w) = 1 for all x, w X.
i i
i i
i i
K23798 2015/2/2
i i
Definition 6.4.10 Let FR(X). Then C and are called compatible if

x X there does not exist w X such that C(x, w) = (x, w) = 0.
If C(x, w) = 1 x, w X, then C and any are compatible. If C(x, w) =

(w, x) w, x X and is complete, then C and are compatible. Hence the
next result holds for MM and MG .
The following theorem is the main result of this section.
Theorem 6.4.11 Let fe be a partial fuzzy simple rule and X R. Then

FS, if fe()(z, ) is bounded away from 0 for each fe-median z and C and
fe() are compatible, then Cfe() 6= and Cfe() = fe(fe()).
Proof. Let FS and z X. Then by Proposition 6.4.2,

1 if Sz = ,
M (fe(), 1X )(z) =
{tz,w | w Sz } if Sz 6= .
Hence if z is an fe-median, Cfe()(z) = fe()(z). Thus it suffices to show

that Supp(M (fe(), 1X )) = Supp(fe(). Let x Supp(M (fe(), 1X )). Then
M (fe(), 1X )(x) > 0 and so {t [0, 1] | C(x, w) t fe()(x, w) w
X} > 0. Since C and fe() are compatible, fe()(x, y) > 0 y X. Suppose
e()(x) = 0, i.e., x
f / Supp( e()). Then x is not an fe-median. Hence either
f
Supp(L e (x)) {Supp() | L(fe)} or Supp(L e + (x)) {Supp() |

L(fe)}. Assume the former. Then Supp(L e (x)) = Supp() for some L(fe).
Thus u, v X, i (u, v) > 0 i Supp(L e (x)) implies (u, v) > 0. Since
is single-peaked, i (y, x) > 0 i Supp(L e (x)) for xi < y < x. Hence
since L(fe) is decisive, fe()(y, x) > 0 for xi < y < x. But then fe()(x, y) = 0
which contradicts the fact that fe()(x, y) > 0 y X. It thus follows that
x Supp(fe()). Hence Supp(M (fe(), 1X )) Supp(fe()). Now assume
the latter. Then Supp(L e + (x)) = Supp() for some L(fe). Since is
single-peaked, i (y, x) > 0 i Supp(L e (x)) for xi > y > x. Hence since
L(fe) is decisive, fe()(y, x) > 0 for xi > y > x. But then fe()(x, y) = 0
which contradicts the fact that fe()(x, y) > 0 y X. It thus follows that
x Supp( e()). Hence Supp(M (fe(), 1X )) Supp( e()). Since clearly
f f
Supp(fe() Supp(M (fe(), 1X )), Supp(M (fe(), 1X )) = Supp(fe()). Since
fe()(z, ) is bounded away from 0, C e() 6= .
f
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
Corollary 6.4.12 Let fe be a partial fuzzy simple rule and X R. Then

FS, if fe()(z, ) is bounded away from 0 for some fe-median z and C
and fe() are compatible, then Cfe() 6= .
We now consider the implications of the core for k-dimensional spatial
models, Rk . These models are widely used in political science. We use the
usual topology in the Euclidean space Rk . Suppose X is a subset of Rk . A
subset Z of X is called open relative to X (or simply open) if Z = Y X for
some subset Y that is open in Rk .
We assume throughout that is reflexive and a complete fuzzy binary
relation on X.
Definition 6.4.13 Let be a fuzzy binary relation on X and let x X.

Define 1 (x) = {y X | (x, y) > 0} and (x) = {y X | (y, x) > 0}.
Definition 6.4.14 Let be a fuzzy binary relation on X. Then is called

lower continuous at x X if 1 (x) is open and is called upper con-
tinuous if (x) is open for all x X.
Definition 6.4.15 Let be a fuzzy binary relation on X. The is said to

satisfy condition F if finite subsets Y of X, x X such that (x, y) > 0
y Y.
Note that condition F does not require that x be a member of Y. It follows

that if is acyclic on X, then condition F holds.
Lemma 6.4.16 Suppose X is compact and is lower continuous. Sup-

pose that C and are compatible. Suppose satisfies condition F and that
M (Supp(), X) 6= , implies there exists x M (Supp(), X) such that (x, )
is bounded away from 0. Then M (, 1X ) 6= . Conversely if M (, 1X ) 6= ,
then satisfies condition F and if M (Supp(), X) 6= , there exists x
M (Supp(), X) such that (x, w) > 0w X.
Proof. Suppose satisfies condition F and M (Supp(), X) 6= implies

x M (Supp(), X) such that (x, ) is bounded away from 0. Then for all
finite subsets Y of X, there exists x X such that (x, y) Supp() y Y.
Thus Supp() satisfies condition F, [2]. Since is lower continuous, {y X |
(x, y) Supp()} = {y X | (x, y) > 0} is open for all x X. Thus
Supp() is lower continuous since if R = Supp(), then P = Supp(). Hence
M (Supp(), X) = {x X | (x, y) Supp() y X} 6= by [2, Lemma 5.1,
p. 124]. Let x M (Supp(), X) be such that (x, ) is bounded away from
0. Then {t (0, 1] | (x, y) t y X} is not empty. Thus there exists
i i
i i
i i
K23798 2015/2/2
i i
t0 (0, 1] such that (x, y) t0 y X. Hence C(x, y) t0 (x, y) y X.

Thus M (, 1X )(x) > 0. Hence M (, 1X ) 6= .
Conversely, suppose M (, 1X ) 6= . Then there exists x X such that
M (, 1X )(x) = {t [0, 1] | C(x, w) t (x, w) w X} > 0. Since C and
are compatible, (x, w) > 0 w X. It follows that x M (Supp(), X). Let
Y be a finite subset of X. Then {t [0, 1] | (x, y) t y Y } > 0. Hence
there exists x X such that (x, y) > 0 y Y. That is, satisfies condition
F.
Corollary 6.4.17 Suppose X is compact and is lower continuous. Suppose

C and are compatible. Then Supp(MM (, 1X )) = M (Supp(), X) 6= if
and only if satisfies condition F and for all x M (Supp(), X), (x, ) is
bounded away from 0.
Example 6.4.18 Let be the fuzzy binary relation on X = [0, 1] defined as

follows: (x, y) X X,

x if x > y,
(x, y) = 1 if x = y = 1,
0 otherwise.

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].
Definition 6.4.19 Let X Rk be a convex set. Let be a fuzzy preference

ordering on X. Then is said to be strictly convex if x, y X, (x, y) > 0
and x 6= y imply (ax + (1 a)y, y) > 0 a (0, 1). If Rn , then is called
strictly convex if i is strictly convex for all i N.
i i
i i
i i
K23798 2015/2/2
i i
Definition 6.4.20 Let Y X. The convex hull of Y, written Con(Y ), is

defined to be the set
(m m
)
X X
Con(Y ) = ai yi | m N, ai [0, 1], yi Y, i = 1, . . . , m, ai = 1 .
i=1 i=1
Definition 6.4.21 Let be a fuzzy binary relation on X. Then is called

semi-convex if x X, x
/ Con((x)).
KKM [6] Let A = {a0 , a1 , . . . , am } Rk and let {S0 , S1 , . . . , Sm } be a
collection of closed sets. Let M = {0, 1, . . . , m}. If L M, Con({ai | i
L}) iL Si . Then jL Sj 6= .
Lemma 6.4.22 Suppose X is convex. Let be a fuzzy binary relation on X.

If is lower continuous and semi-convex on X, then satisfies condition F.
Proof. Let Y = {y0 , y1 , . . . , ym } be a finite subset of X. Let M =

{0, 1, . . . , m}. Since 1 (x) is open x X, X\1 (x) is closed x X.
Suppose there exists L M such that
Con({yi | i L}) * iL (X\1 (yi )) Con(Y ).
Then there exists z Con({yi | i L}) such that z /

iL (X\1 (yi ))Con(Y ). Hence z / iL (X\1 (yi )) since z Con(Y ).
Thus z / X\1 (yi ) i L. Hence z 1 (yi ) i L and so yi (z)
i L. Thus z Con({yi | i L}) Con((z)). However, this contradicts
the hypothesis that is semi-convex. Thus
Con({yi | i L}) iL (X\1 (yi )) Con(Y ).
Hence by KKM, iL (X\1 (yi ))Con(Y ) 6= . Let x iM (X\1 (yi ))

Con(Y ). Then x X\1 (yi )i M and so x / 1 (yi ) i M. That is,
i M, it is not the case that (yi , x) > 0 and (x, yi ) = 0. Hence i M,
either (yi , x) 0 and (x, yi ) > 0 or (yi , x) = 0 and (x, yi ) = 0. In the
former case, we have that satisfies condition F. The latter case is impossible
since is complete.
We now consider the characteristics of collective preferences over a set of
alternatives X when a maximal fuzzy subset is nonzero.
Theorem 6.4.23 Suppose X is compact and convex. Let be a fuzzy binary

relation on X. Suppose C and are compatible. If is lower continuous and
semi-convex and M (Supp(), X) 6= implies there exists x M (Supp(), X)
such that (x, ) is bounded away from 0, then M (, 1X ) 6= .
Proof. The result follows by Lemmas 6.4.16 and 6.4.22.
i i
i i
i i
K23798 2015/2/2
i i
We next consider the degree to which a fuzzy core is likely to be nonempty

as the dimensionality of spatial models increases.
Let FRn . Recall that x, y X, P (x, y; ) = {i N | i (x, y) > 0}.
Suppose X Rk . Then k is the dimension of the policy space. Assume in
what follows that X is convex and compact.
Theorem 6.4.24 Let fe be a partial fuzzy voting rule. If FT Rn is strictly

convex and fe()(x, ) satisfies the inf propertyx X and feL(fe) ()(x, )
is bounded away from 0 x X and C and fe() are compatible, then
Supp(Cfe()) = Supp(Cfe e ()).
L(f )
Proof. Let x Supp(Cfe()). Since C and fe() are compatible,

fe()(x, y) > 0 y X. Hence y X, either (x, y) > 0 or (fe()(x, y) > 0 and
fe()(y, x) > 0). Thus y X, either (x, y) P()
e or (x, y) / Symm (P()).
e
Hence fe e ()(x, y) > 0 y X. Since fe e ()(x, ) is bounded away
L(f ) L(f )
from 0, Cfe ()(x) > 0. Thus Supp(Cfe()) Supp(Cfe ()). Suppose
L(fe) L(fe)
x / Supp(Cfe ()). Then M (fe(), 1X )(x) = 0 and so {fe()(x, y) | y

X} = 0. Since fe()(x, ) has the inf property, there exists y X such
that fe()(x, y) = 0. Thus e(y, x) > 0 since fe() is complete. Since fe is
f
a partial fuzzy voting rule, fe is monotonic and neutral by Theorem 3.14,
[24]. Hence (P (y, x; ), R(y, x; )) D(fe). Since is strictly convex, we have
i R(y, x; ) that i (z, x) > 0, where z = ax + (1 a)y for some a (0, 1).
Since R(y, x; ) is a decisive coalition, i.e., R(y, x; ) L(fe), fe e (z, x) > 0
L(f )
and so feL(fe) ()(x, z) = 0. Thus x

/ Supp(Cfe ()).
L(fe)
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().
Definition 6.4.25 Let fe be a fuzzy aggregation rule. If K(L(fe)) 6= , then fe

is called collegial.
Definition 6.4.26 Let fe be a fuzzy aggregation rule. The Nakamura num-

ber associated with a simple rule fe, written s(fe), is defined as follows:
(
if fe is collegial,
s(fe) =
min{|LSupp | |L L(fe) and K(L) = } otherwise.
i i
i i
i i
K23798 2015/2/2
i i
Definition 6.4.27 Let L FP(N ). Then L is called

(1) monotonic if L, 0 FP(N ), Supp(0 ) Supp() implies
0
L;
(2) proper if L, 0 FP(N ), N \Supp() = Supp(0 ) implies 0
/
L.
Lemma 6.4.28 Let fe be a fuzzy aggregation rule. Then L(fe) is monotonic

and proper.
Proof. Let L(fe) and 0 FP(N ). Suppose Supp(0 ) Supp(). Let

x, y X. If i (x, y) > 0 i Supp(0 ), then i (x, y) > 0 i Supp(). Hence
(x, y) > 0 and so 0 L(fe). Thus L(fe) is monotonic.
Now suppose Supp(0 ) = N \Supp(). There exists Rn such that
i (x, y) > 0 i Supp() and i (y, x) > 0 i Supp(0 ). Thus if 0 L(fe),
then (y, x) > 0. However, this is impossible since L(fe) and so (x, y) > 0.
Thus 0 / L(fe).
Lemma 6.4.29 Suppose fe is a partial fuzzy simple rule. Then s(fe) 3. If fe

is noncollegial, then s(fe) n.
Proof. Suppose |LSupp (fe)| = 1. Then fe is collegial and so s(fe) = .

Suppose |LSupp (fe)| 2. Let 1 , 2 L(fe). By Lemma 6.4.28, L(fe) is proper.
Hence Supp(1 )Supp(2 ) 6= . Thus K(L) 6= if |LSupp | = 2 and L L(fe).
Hence s(fe) 3.
Suppose fe is noncollegial. Then i N, there exists L(fe) such that
i/ Supp(). Since L(fe) is monotonic, it follows i N that 1N \{i} L(fe).
Let L = {1N \{i} | i N }. Then L L(fe), K(L) = and |LSupp | = n. Thus
s(fe) n.
Let i N and x X. Define Pi (x) = {y X | i (y, x) > 0} and Pi1 (x) =
{y X | i (x, y) > 0}. Let FP(N ) be a coalition. Define P (x) =
iSupp() Pi (x) and P1 (x) = iSupp() Pi1 (x). Let fe be a fuzzy aggregation
rule. Define Pfe(x) = L(fe) P (x) and P e1 (x) = L(fe) P1 (x).
f
Lemma 6.4.30 Suppose Rn is lower continuous. Then for all partial

fuzzy simple rules fe, fe() is lower continuous.
Proof. Let x X and L(fe). Then P1 (x) is a finite intersection of

open sets and hence is open. Thus P e1 (x) is a union of open sets and hence
f
is open. Since P 1 (x) = eP
1
L(f ) (x) and L(fe) in the definition of
fe
P e1 (x), it follows that P e1 (x) = 1 (x) = {y X | (x, y) > 0}. That is,
f f
fe()1 (x) = 1 (x) is open.
i i
i i
i i
K23798 2015/2/2
i i
Lemma 6.4.31 Let fe be a partial fuzzy simple rule. Suppose Rn is semi-

convex. If k s(fe) 2, then fe() is semi-convex.
Proof. Suppose fe() is not semi-convex. Then there exists z X such

that z Con(Pfe(z)) = Con(L(fe) P (z)) = Con(L(fe) iSupp() Pi (z)) =
Con(L(fe) iSupp() {y X | i (y, z) > 0}). By Caratheodorys Theo-
rem [6], there exist y1 , . . . , yk+1 Pfe(z) such that z Con({y1 , . . . yk+1 }).
Since yj P e(z) for j = 1, . . . , k + 1 and fe is simple, there exist deci-
f
sive coalitions 1 , . . . , k+1 such that yj Pj (z) for j = 1, . . . , k + 1. Let
L = {1 , . . . , k+1 }. By hypothesis, k s(fe) 2. Hence |LSupp | |L|
k + 1 s(fe) 1. Thus s(fe) 6= min{|LSupp | |L L(fe) and K(L) = }. Hence
K(L) = L Supp() 6= by the definition of s. Thus there exists i N such
that i Supp(j ) for j = 1, . . . , k + 1. Hence yj Pi (z) for j = 1, . . . , k + 1.
Thus z Con({y1 , . . . , yk+1 }) Con(Pi (z) which contradicts the fact that i
is semi-convex.
We conclude with our main result concerning k-dimensional policy space.
Theorem 6.4.32 Let fe be a partial fuzzy simple rule. Let Rn . Suppose

C and fe() are compatible. Suppose is lower continuous and semi-convex.
Suppose M (Supp(fe()), X) 6= implies there exists x M (Supp(fe()), X)
such that fe()(x, ) is bounded away from 0. If k s(fe) 2, then Cfe() 6= .
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.
2. [14] Let L F(N ). Let FRn and set P() = {(x, y) X X |

L, (x, y) > 0, i Supp()}. Define gL : FRn F R by for all
FRn and all x, y X,

1 if (x, y)
/ Symm(P())

{{i (x, y) | i Supp()} | L and i (x, y) > 0,

gL ()(x, y) = i Supp()}
if (x, y) P()

{{i (x, y) | i Supp()} | L and i (y, x) > 0,

i Supp()} if (x, y) Symm(P())\ P().

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).
3. [14] Let f be a fuzzy aggregation rule. Then

(1) f is called a regular fuzzy simple rule if for all FRn and all
x, y X,
f()(x, y) > f()(y, x) gL(f) ()(x, y) > gL(f) ()(y, x).;
(2) f is called regularly neutral if , 0 FRn and x, y, z, w X,

[P (x, y; ) = P (z, w; 0 ) and P (y, x; ) = P (w, z; 0 )] imply f()(x, y)
> f()(y, x) f(0 )(z, w) > f(0 )(w, z).
Let f be an FPAR and suppose is regular. Then prove that f is a
regular fuzzy simple rule if and only if f is decisive, monotonic, and
regularly neutral.
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
6.5. Exercises 203
D, i (x, y) > 0 i Supp() and j (y, x) = 0 j Supp()}. Define

gD : FRn FR by for all FRn and all x, y X,

if (x, y)
/ Symm(R())
1

{{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().

Prove the following statements. If (x, y) P(), then gD ()(x, y) >

gD ()(y, x). If strict preferences are partial (regular), then D is partial
(regular).
5. [14] Let f be a fuzzy aggregation rule. Then

(1) f is called a regular fuzzy voting rule if for all FRn and all
x, y X,
f()(x, y) > f()(y, x) gD(f) ()(x, y) > gD(f) ()(y, x),
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,
f()(x, y) > f()(y, x) gL(f) ()(x, y) > gL(f) ()(y, x).
(3) f is called regularly neutral if , 0 F Rn and x, y, z, w

X, [P (x, y; ) = P (z, w; 0 ) and P (y, x; ) = P (w, z; 0 )] imply f()(x, y)
> f()(y, x) f(0 )(z, w) > f(0 )(w, z).
Prove the following statements. Let f be an FPAR and suppose is
regular. Then f is a regular fuzzy simple rule if and only if f is decisive,
monotonic, and regularly neutral.
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
D, i (x, y) > 0 i Supp() and j (y, x) = 0 j Supp()}. Define

gD : FRn FR by for all FRn and all x, y X,

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().
Prove the following statements. If (x, y) P(), then gD ()(x, y) >

gD ()(y, x). If strict preferences are partial (regular), then D is partial
(regular).
7. Let f be a fuzzy aggregation rule. Then f is called a regular fuzzy

voting rule if for all FRn and all x, y X,
f()(x, y) > f()(y, x) gD(f) ()(x, y) > gD(f) ()(y, x),
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.
Let f be an FPAR and suppose is regular. Prove that f is a regular

fuzzy voting rule if and only if f is monotonic and regularly neutral.
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,

where 0 < < < 1.
i i
i i
i i
K23798 2015/2/2
i i
6.6. References 205
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,

where 0 < < < 1.
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.

where 0 < < < 1.
6.6 References
1. K. J. Arrow, Social Choice and Individual Values, John Wiley and Sons,
New York, 1951.
tive Preference, Michigan Studies in Political Science, The University of
Michigan Press, 2000.

and Structure, Michigan Studies in Political Science, The University of
Michigan Press, 2005.
4. D. Black, The Theory of Committees and Elections, London, Cambridge
University Press, 1958.
5. D. Black, On Arrows impossibility theorem, Journal of Law and Eco-

nomics, 12 (1969) 227248.
i i
i i
i i
K23798 2015/2/2
i i
6. K. C. Border, Fixed Point Theorems with Applications to Economics

and Game Theory, London Cambridge University Press 1985.
7. J. M. Buchanan and R. D. Tollison, The Theory of Public Choice-II,
University of Michigan Press, 1984.
8. T. D. Clark and J. N. Mordeson, Applying fuzzy mathematics to polit-
ical science: What is to be done? Critical Review 2 (2008) 1319.
9. A. Downs, An Economic Theory of Democracy, New York, Harper and
Row Publishers, 1957.
11. I. Georgescu, Fuzzy Choice Functions: A Revealed Preference Approach,
Springer, Studies in Fuzziness and Soft Computing Vol. 214 2007.
tems, 158 (2007) 13141326.
13. A. Gibbard, Manipulation of voting schemes: A general result, Econo-
metrica, 41 (1973) 587601.
14. M. B. Gibilisco, A. M. Gowen, K. E. Albert, J. N. Mordeson, M. J.
Wierman, and T. D. Clark, Fuzzy Social Choice, Springer, Studies in
Fuzziness and Soft Computing 315 Springer 2014.
15. M. B. Gibilisco, J. N. Mordeson, and T. D. Clark, Fuzzy Blacks median
voter theorem: Examining the structure of fuzzy preference rules and
strict preference, New Mathematics and Natural Computation, 8 (2012)
195217.
16. R. G. Holcomb, An empirical test of the median voter model, Economic
Inquiry, 18 (1980) 260275.
17. T. A. Husted and A. W. Kenny, The effect of the expansion of the voting
franchise on the size of government model, Journal of Political Economy,
105 (1997) 5482.
18. K. Krehbiel, Legislative organization, Journal of Economic Perspectives,
18 (2004) 113128.
19. D. R. Kiewit and M. D. McCubbins, Presidential influence on congres-
sional appropriations decisions, American Journal of Political Science,
32 (1988) 713736.
20. R. D. McKelvey, Intransitivities in multidimensional voting models and
some implications for agenda control, Journal of Economic Theory, 12
(1976) 472482.
i i
i i
i i
K23798 2015/2/2
i i
6.6. References 207
21. J. N. Mordeson, K. R. Bhutani, and T. D. Clark, The rationality of

fuzzy choice functions, New Mathematics and Natural Computation, 4
(2008) 309327.
22. J. N. Mordeson and T. D. Clark, Fuzzy Arrows Theorem, New Mathe-
23. J. N. Mordeson and T. D. Clark, The core in fuzzy spatial models, New
Mathematics and Natural Computation, 6 (2010) 1729.
24. J. N. Mordeson, L. Nielsen, and T. D. Clark, Single peaked fuzzy pref-

erences in one-dimensional models: Does Blacks median voter theorem
hold?, New Mathematics and Natural Computation 6 (2010) 116.
25. T. W. Rice, An examination of the median voter hypothesis, The West-
ern Political Quarterly, 38 (1985) 211223.
26. T. Romer and H. Rosenthal, The elusive median voter, Journal of Public
Economics, 12 (1979) 143170.
27. M. A. Satterthwaite, Strategy-proofness and Arrows conditions, exis-
tence and correspondence theorems for voting procedures and social
welfare functions, Journal of Economic Theory, 10 (1975) 187217.
28. A. K. Sen, Collective Choice and Social Welfare, North Holland Pub-
lishing Company, 1979.
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
7.1 Fuzzy Preference and Preference-Based

Choice Functions
The results in this section are based on Barrett, Pattanaik, and Salles [9].
We consider the question as to how unambiguous or exact choices are gen-
erated by fuzzy preferences and whether the exact choices induced by fuzzy
preferences satisfy certain plausible rationality properties. Also, see [7, 11, 12,
15, 21, 23, 30]). Several alternative rules for generating exact choice sets from
fuzzy weak preference relations are presented. To what extent the exact choice
sets generated by these alternative rules satisfy certain fairly weak rationality
conditions is also examined.
Given fuzzy weak preference relations, we present the nine rules given in
[9] for inducing exact choice sets. Different sets of rationality properties are
considered in terms of which the performance of the nine preference-based
choice functions is measured. The first set of properties gives sufficient con-
ditions for choosing an alternative, given a set of feasible alternatives, and
given a fuzzy weak preference relation. The second set of properties gives suf-
ficient conditions for rejecting an alternative, and the third set of properties
states that if an exact best alternative exists, then the choice set induced by
a preference based choice function should possess certain relations to the set
of exactly best alternatives. It is shown that if the fuzzy preferences are not
assumed to satisfy max-min transitivity, then all the preference-based choice
functions which are examined here perform rather badly in terms of some of
these plausible criteria. However, if the admissible fuzzy-preferences are as-
sumed to satisfy max-min transitivity, then one of the preference-based choice
functions satisfies all the rationality conditions under consideration. Let X be
a given finite set of alternatives such that |X| 4.
Definition 7.1.1 Let FR(X). Then
209
i i
i i
i i
K23798 2015/2/2
i i
210 7. Rationality
(1) is called a fuzzy ordering if is reflexive, strongly complete, and

max-min transitive;
(2) is called exact if (X X) {0, 1}.
We let FC(R) denote the set of all fuzzy relations on X that are reflex-
ive and strongly complete. The set of all fuzzy orderings will be denoted by
FT (R). Throughout, we let FC(R)0 denote a nonempty subset of FC(R).
Recall P (X) denotes P(X)\{}.
Definition 7.1.2 A function C : P (X) FC(R)0 P (X) is called a

preference-based choice function (PCF) if for all B P (X) and for
all FC(R)0 , =
6 C(B, ) B.
We think of FC(R)0 as those fuzzy relations which are admissible. Given
FC(R)0 and a crisp set B of available alternatives, then C(B, ) is the
crisp set of alternatives chosen from B with respect to . Often an individuals
preferences are fuzzy, but his actual choices are not fuzzy, i.e., they are crisp.
We next introduce nine preference-based choice functions.
Let B P (X) and F C(R). Using the notation of [9], we define the
following real-valued functions of B as follows: x B,
M F (B, )(x) = {(x, y) | y B\{x}},
mF (B, )(x) = {(x, y) | y B\{x}},

M A(B, )(x) = {(y, x) | y B\{x}},
mA(B, )(x) = {(y, x) | y B\{x}},
The notation can be interpreted as follows: M F (B, )(x) as max for x
in B, with respect to and mF (B, )(x) as min for x in B, with respect
to . Similarly M A and mA can be interpreted as max against and min
against, respectively. These four functions can be considered to be fuzzy
subsets of B with respect to . Define the remaining five functions as follows:
x B,
X
SF (B, )(x) = (x, y),
yB\{x}
X
SA(B, )(x) = (y, x),
yB\{x}
M D(B, )(x) = {(x, y) (y, x) | y B\{x}},

mD(B, )(x) = {(x, y) (y, x) | y B\{x}},
and X
M D(B, )(x) = [(x, y) (y, x)].
yB\{x}
i i
i i
i i
K23798 2015/2/2
i i
7.1. Fuzzy Preference and Preference-Based Choice Functions 211
The notation SF, SA, M D, mD, and SD can be interpreted as sum

for, sum against, max difference, min difference, and sum difference,
respectively.
We present some different possible types of preference-based choice func-
tions.
Definition 7.1.3 [9] Let C : P (X) FC(R)0 P (X) be a PCF. Then C

is called
(1) max-MF if for all B P (X) and for all FC(R)0 ,
C(B, ) = {x B | M F (B, )(x) M F (B, )(y) for all y B\{x}}.
(2) max-mF if for all B P (X) and for all R FC(R)0 ,
C(B, ) = {x B | mF (B, )(x) mF (B, )(y) for all y B\{x}}.
(3) max-SF if for all B P (X) and for all FC(R)0 ,
C(B, ) = {x B | SF (B, )(x) SF (B, )(y) for all y B\{x}}.
(4) min-MA if for all B P (X) and for all FC(R)0 ,
C(B, ) = {x B | M A(B, )(x) M A(B, )(y) for all y B\{x}}.
(5) min-mA if for all B P (X) and for all FC(R)0 ,
C(B, ) = {x B | mA(B, )(x) mA(B, )(y) for all y B\{x}}.
(6) min-SA if for all B P (X) and for all FC(R)0 ,
C(B, ) = {x B | SA(B, )(x) SA(B, )(y) for all y B\{x}}.
(7) max-MD if for all B P (X) and for all FC(R)0 ,
C(B, ) = {x B | M D(B, )(x) M D(B, )(y) for all y B\{x}}.
(8) max-mD if for all B P (X) and for all FC(R)0 ,
C(B, ) = {x B | mD(B, )(x) mD(B, )(y) for all y B\{x}}.
(9) max-SD if for all B P (X) and for all FC(R)0 ,
C(B, ) = {x B | SD(B, )(x) SD(B, )(y) for all y B\{x}}.
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],
C(B, ) = {x X | {1 (y, x) | y B\{x}}

{1 (y, x0 ) | y B\{x}} for all x0 B\{x}}.
If is strongly complete, then the max-mF rule here is the same as a choice
rule discussed by Switalski [29],
C(B, ) = {x B | {(x, y) (1 (y, x)) | y B\{x}}

{(x0 , y) (1 (y, x0 ) | y B\{x}} for all x0 B\{x}}.
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,
OV (B, )(x) = {(1 (y, x) + (x, y)) 1 | y B\{x}}.
Definition 7.1.4 Let C : P (X) F C(R)0 P (X) be a PCF. C is max-

OV if for all B P (X) and for all FC(R)0 ,
C(B, ) = {x B | OV (x, B, ) OV (y, B, ) for all y y B\{x}}.
max-OV PCFs were introduced by Orlovsky [21]. Related ideas can be

found in Ovchinnikov and Ozernoy [23].
For all B P (X) and for all F T (R), define the set T (B, ) as
follows:
T (B, ) = {x B | (x, y) (y, x) for all y B\{x}}.
Proposition 7.1.5 (Barrett, Pattanaik, and Salles [9]) Let C : P (X)

FC(R)0 P (X) be a PCF. Then C is max-mD if and only if it is max-
OV.
Proof. Let B P (X) and FC(R)0 .

Suppose T (B, ) 6= . Then for all x T (B, ), OV (B, )(x) = 1, and for
all x B\T (B, ), OV (B, )(x) < 1. Thus
T (B, ) = {x B | OV (B, )(x) OV (B, )(y) for all y B\{x}}. (7.1)
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
T (B, ) = {x B | mD(B, )(x) mD(B, )(y) for all y B\{x}}. (7.2)
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,
OV (B, )(x) = {1 (y, x) + (x, y) | y B\{x}}.
Thus for all x, z B, OV (x, B, ) OV (z, B, ) if and only if
{(x, y) (y, x) | y B\{x}} {(z, y) (y, z) | y B\{x}}.
Hence
{x B | OV (B, )(x) OV (B, )(y) for all y B\{x}}

= {x B | mD(B, )(x) mD(y, B, )(y) for all y B\{x}}.
Thus when T (B, ) = , a PCF is max-mD if and only if it is max-OV.

We now consider some basic restrictions on preference-based choice func-
tions, which represent different aspects of rationality. See [2, 28, 29], for
counterparts on rational choice and exact preference to some of these prop-
erties. We examine the performance of the different PCFs in terms of these
properties.
The following rationality conditions are considered: conditions for choosing
an alternative; conditions for rejecting an alternative; and conditions stipu-
lating certain relations between the set of chosen alternatives and the set of
alternatives which, in an exact sense, weakly dominate all feasible alternatives.
Each set contains two different, but related conditions.
Throughout the rest of this section, C denotes a PCF as defined in Defi-
nition 7.1.2.
The following definition provides some conditions for choosing alternatives.
Definition 7.1.6 [9] Let C be a preference-based choice function. Then

(1) C is said to satisfy the reward for pairwise weak dominance
(RPWD) if for all B P (X), all FC(R), and all x B, [(x, y)
(y, x) for all y B\{x}] implies x C(B, );
(2) C is said to satisfy the reward for pairwise strict dominance
(RPSD) if for all B P (X), all FC(R), and all x B, [(x, y) > (y, x)
for all y B\{x}] implies x C(B, ).
i i
i i
i i
K23798 2015/2/2
i i
214 7. Rationality
If an available alternative x strictly dominates every other available al-

ternative in a pairwise comparison, then RPSD requires that x should be
chosen. RPWD requires x to be chosen if it weakly dominates every other
alternative in a pairwise comparison. If C satisfies the property that for all
F C(R)0 and for all x, y X, (x, y) (y, x) (resp. (y, x) (x, y))
implies x C({x, y}), then RPWD (resp. RPSD) is implied by Sens [12] Con-
dition (or Aizerman and Malishevskis [1] Concordance). All of the PCFs
introduced in Definition 7.1.3 satisfy this property.
Example 7.1.7 We show (1) and (2) in this example.

(1) If FT (R) FC(R)0 and C C (1) C (3) C (4) C (6) C (7) C (9) , then
C satisfies neither RPWD nor RSPD.
(2) If FC(R)0 = FC(R) and C 7i=1 C (i) C (9) , then C satisfies neither
RPWD nor RPSD.
Consider (1). Let FT (R) FC(R)0 and let B = {x, y, z}. Let , 0
FT (R) be such that
(x, x) = (y, y) = (z, z) = 1,

(x, y) = s, (x, z) = s,
(y, x) = r, (y, z) = t,
(z, x) = r, (z, y) = r
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
(x, x) = (y, y) = (z, z) = 1,

(x, y) = t1 , (x, z) = t3 ,
(y, x) = t2 , (y, z) = t2 ,
(z, x) = t4 , (z, y) = t5
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
(x, y) (y, x) for all y B\{x}. (7.3)
Suppose C is max-mF. Then we must show that mF (B, )(x) mF (B,

)(z) for all z B\{x}. It is sufficient to show that for all distinct z, w
B\{x}, (x, w) mF (B, )(z). Let z, w B\{x}, z 6= w. Since is a fuzzy or-
dering, (x, w) (x, z) (z, w). If (x, z) = (x, z) (z, w), then (x, w)
(x, z), so that by (7.3), we have (x, w) (x, z) (z, x) mF (B, )(z).
If (z, x) = (x, z) (z, w), then (x, w) (z, w) mF (B, )(z). This
completes the case where C is max-mF.
Suppose that C is min-mA. Then we must prove that mA(B, )(x)
mA(B, )(z) for all z B\{x}. It is sufficient to show that for all distinct
z, w B\{x}, (w, z) mA(B, )(x). Let z, w B\{x}, z 6= w. Since is
max-min transitive, (w, z) (w, x) (x, z). If (w, x) = (w, x) (x, z),
then (w, z) (w, x) mA(B, )(x). If (x, z) = (w, x) (x, z), then by
(7.3), it follows that (w, z) (x, z) (z, x) mA(B, )(x).
Even though RPSD is a weaker condition than RPWD, the performance
of the nine types of PCFs does not improve when RPSD rather than RPWD
is assumed. Note also that only three (max-mF, min-mA, max-mD) of our
type of PCFs satisfy RPSD.
We now introduce rationality conditions which specify when an alterna-
tive should be rejected.

(1) C is said to satisfy strict rejection (SREJ) if B P (X),
FC(R)0 , and x B, ((y, x) (x, y) for all y B\{x} and (y, x) >
(x, y) for some y B\{x}) implies x / C(B, );
(2) C is said to satisfy weak rejection (WREJ) if for all B P (X),
all F C(R)0 , and all x B, (y, x) > (x, y) for all y B\{x} implies
x / C(B, ).
SREJ implies WREJ, but not conversely. SREJ says that if an alterna-
tive is weakly dominated in every relevant pairwise comparison in which it is
involved, and if it is strictly dominated in at least one of these pairwise com-
parisons, then it should not be chosen. WREJ states that if an alternative is
i i
i i
i i
K23798 2015/2/2
i i
216 7. Rationality
strictly dominated in every relevant pairwise comparison, then it should not

be chosen.
Suppose C satisfies the property that for all FC(R)0 and for all
x, y X, (x, y) > (y, x) implies C({x, y}, ) = {x}. Then both SREJ
and WREJ are weaker restrictions than Sens [27] Condition (or Aizerman
and Malishevskis [1]). Note that all the PCFs introduced in Definition 7.1.3
satisfy this property.
We next determine which of the nine types of PCFs satisfy SREJ and
WREJ.
Lemma 7.1.10 Suppose that is a fuzzy ordering and B P (X). Then

T (B, ) 6= .
Proof. Suppose that T (B, ) = . Then there exist x1 , x2 , . . . , xn B such
that for all i, 1 i n1, (xi , xi+1 ) > (xi+1 , xi ) and (xn , x1 ) > (x1 , xn ).
We assume without loss of generality that (rn , x1 ) = (x1 , x2 ) (x2 , x3 )
(xn , xn1 ) (xn , x1 ). Since is max-min transitive, we have that
(x1 , xn ) (x1 , x2 ) (x2 , xn ). Suppose (x1 , x2 ) (xn , x1 ) > (x1 , xn ).
Then it follows that
(x1 , xn ) (x2 , xn ). (7.4)
Since is transitive, (x2 , xn ) (x2 , x3 ) (x3 , xn ). Given (x2 , x3 )
(xn , x1 ) > (x1 , xn ), and (7.4), it follows that (x2 , xn ) (x3 , xn ). Hence
by (7.4), we have that
(x1 , xn ) (x3 , xn ). (7.5)
Continuing this process, it follows that
(x1 , xn ) (xn1 , xn ). (7.6)
Given (xn1 , xn ) (xn , x1 ) > (x1 , xn ), we have a contradiction.
Example 7.1.11 We show conditions (1) and (2) hold.

(1) If FT (R) FC(R)0 and C C (2) C (4) C (6) , then C does not satisfy
WREJ. Also, if FT (R) FC(R)0 and C 6i=1 C (i) , then C does not satisfy
SREJ.
(2) If FC(R)0 = FC(R) and C C (2) C (3) C (4) C (6) C (8) , then C
does not satisfy WREJ. Also, if FC(R)0 = FC(R) and C C (1) C (2) C (3)
C (4) C (5) C (6) C (8) , then C does not satisfy SREJ.
(1) Let B = {x, y, z}.
(a) Let FT (R) FC(R)0 be such that
(x, x) = (y, y) = (z, z) = 1,

(x, y) = s, (x, z) = s,
(y, x) = t, (y, z) = s,
(z, x) = t, (z, y) = s.
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
0 (x, x) = 0 (y, y) = 0 (z, z) = 1,

0 (x, y) = r, 0 (x, z) = r,
0 (y, x) = s, 0 (y, z) = t,
0 (z, x) = s, 0 (z, y) = t.
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
00 (x, x) = 00 (y, y) = 00 (z, z) = 1,

00 (x, y) = r, 00 (x, z) = t,
00 (y, x) = s, 00 (y, z) = s,
00 (z, x) = t, 00 (z, y) = r.
where 1 > t > s > r > 0. If C is max-MF or max-SF, then x C(B, 00 ).

Therefore, SREJ does not hold.
(d) Let FT (R) FC(R)0 be such that
(x, x) = (y, y) = (z, z) = 1,

(x, y) = s, (x, z) = s,
(y, x) = t, (y, z) = t,
(z, x) = s, (z, y) = s.
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.
(d12 +d13 + +d1n )+(d21 +d23 + +d2n )+ +(dn1 +dn2 + +dn,n1 ) = 0.

Since SD(B, )(y) < 0 for all y B, we have
(d12 + d13 + + d1n ) + (d21 + d23 + + d2n )+

+ (dn1 + dn2 + + dn,n1 )
= [SD(B, )(z1 ) + SD(B, )(z2 ) + + SD(B, )(zn )]
< 0,
which is a contradiction. Thus that if C is max-SD, then C satisfies SREJ.

Suppose C is max-MD or max-SD. Then C satisfies SREJ and it follows
that if C is max-MD or max-SD, then C satisfies WREJ. Also if C is max-MF
or min-mA, then C satisfies WREJ.
(2) By part (1), if FC(R)0 FT (R) and C is max-MD or max-SD, then
C satisfies SREJ. We show that if FC(R)0 FT (R) and C is max-mD, then
C satisfies SREJ.
Let B P (X), F C(R)0 F T (R), and x B be such that for all
y B\{x}, (x, y) (y, x), and for some y B\{x}, (x, y) < (y, x).
Suppose C is max-mD. By Lemma 7.1.10, T (B, ) 6= . Let x0 T (B, ).
Then mD(B, )(x0 ) 0 > mD(B, )(x). Hence x / C(B, ).
By part (1), and given that if FC(R)0 FT (R) and C is max-mD, then C
satisfies SREJ, it follows that if FC(R)0 FT (R) and C C (1) C (5) C (7)
C (8) C (9) then C satisfies WREJ. Let B P (X). Let FC(R)0 FT (R)
and x B be such that for all y B\{x, z}, (x, y) < (y, x). Let z B\{x}
be such that for all y B\{x}, (x, z) (x, y). Let w B\{x, z}. Then
i i
i i
i i
K23798 2015/2/2
i i
by the max-min transitivity of , (z, w) (z, x) (x, w). Since by our

assumptions (z, x) > (x, z) (z, w), it follows that (z, w) (x, w).
Thus for all w B\{x, z}, (z, w) (x, w). Given (z, x) > (x, z), it
follows that SF (B, )(z) > SF (B, )(x) and thus x C(B, ).
The properties in Definition 7.1.13 below require that if a best alternative
exists under the FBPR concerned in the exact sense, then the set of alterna-
tives actually chosen must be faithful to the set of exactly best alternatives.
For all B P (X) and for all FC(R), define the set D(B, ) as follows:
D(B, ) = {x B | for all y B, (x, y) = 1}.

(1) C is said to satisfy upper faithfulness (UF) if for all B P (X)
and for all FC(R)0 , D(B, ) 6= implies C(B, ) D(B, ); .
(2) C is said to satisfy lower faithfulness (LF) if for all B P (X)
and for all FC(R)0 , D(B, ) 6= implies D(B, ) C(B, ).
Example 7.1.14 We show that conditions (1) and (2) hold.

(1) If FT (R) FC(R)0 and C C (1) C (4) C (5) C (7) , then C does not
satisfy UF.
(2) If FC(R)0 = FC(R) and C C (1) C (4) C (5) C (6) C (7) C (8) , then C
does not satisfy UF. Also, if FC(R)0 = FC(R) and C K(min-mA, min-SA,
max-MD, max-SD), then C does not satisfy LF.
Consider (1). Let B = {x, y, z}. Let 0 FT (R) be such that
0 (x, x) = 0 (y, y) = 0 (z, z) = 0 (x, y) = 0 (x, z) = 0 (y, z) = 1,

0 (y, x) = 0 (z, x) = 0 (z, y) = 0.
If C is max-MF, or min-MA, or max-MD, then x C(B, 0 ) and so UF does

not hold.
Let 00 FT (R) FC(R)0 be such that 00 (y, x) = 1 and 00 (u, v) =
0 (u, v) for all (u, v) X X\{(y, x)}. If C is min-MA, then z C(B, 00 )
and so UF does not hold.
(2) Since FT (R) FC(R), from part (1), it follows that FC(R)0 = FC(R)
and if C C (1) C (4) C (5) C (7) , then UF does not hold for C. We now show
that if FC(R) = FC(R)0 and C C (6) C (8) C (9) , then UF does not hold for
C. Let 000 FC(R) be such that
000 (y, z) = t, 000 (z, y) = s, 000 (u, v) = 1 (u, v) X X\{(y, z), (z, y)}
where 1 > t > s > 0. If C is min-SA, or max-mD, or max-SD, then y

C(B, 000 ) and so UF does not hold.
We now consider LF. If C is min-mA, min-SA, or max-MD, or max-SD,
/ C(B, 000 ) and thus LF does not hold.
then x
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
(z, w) (w, z) (x, w) (w, x) for all w B. (7.13)
From (7.11) and since z C(B, ), it follows that if C is min-mA or min-SA,

then x C(B, ). From (7.13) and since z C(B, ), it follows that if C is
max-MD or max-SD, then x C(B, ).
In [9], other preference-based choice functions were considered. We provide
some of these in the exercises.
If the fuzzy preferences are not constrained to satisfy transitivity, then all
the preference-based choice functions discussed in this section violate some
properties which have often been considered to be basic properties of rational
choice. However, if the admissible fuzzy preferences are constrained to be
fuzzy orderings, then the preference-based choice function, max-mD, satisfies
all the rationality properties of Barrett, Pattanaik, and Salles that we have
presented in this section. We summarize Theorems 7.1.8, 7.1.12, and 7.1.15
and Examples 7.1.7, 7.1.11, and 7.1.14. The symbol indicates that the PCF
satisfies the property under consideration when F(R)0 = F(R) and where the
symbols and ~ indicate the PFC satisfies the property under consideration
when F(R)0 = FT (R).
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 ~ ~
In [16], the structure of the nine preference-based choice functions of this

section were examined. It was shown that for crisp total pre-orders, first and
last alternatives exist in a finite set of alternatives that has a strongly complete
fuzzy pre-order. This result is used to characterize each of those crisp choice
functions for crisp total pre-orders and strongly complete fuzzy pre-orders.
When preferences are strongly complete fuzzy pre-orders, the six consistency
conditions of Sen of those preference-based choice functions are determined
by means of those characterizations.
i i
i i
i i
K23798 2015/2/2
i i
222 7. Rationality
7.2 Fuzzy Choice Functions, Revealed

Preference and Rationality
This section is concerned with fuzzy revealed preference theory and with the
problem of rationalizing fuzzy choice functions. The work is based on that
presented in Banerjee [5]. We assume the domain of the choice function con-
sists of all (crisp) finite subsets of the universal set of alternatives. However,
we also assume that the choice is fuzzy in the sense that the decision maker
can state the degree of his choice of an alternative. This is in contrast to the
previous section, where the choice was crisp and the preferences were fuzzy.
Banerjees two types of weak and strong axioms of fuzzy revealed preferences
(WAFRP and SAFRP) are stated. We show that WAFRP is not equivalent
to SAFRP and that SAFRP does not characterize rationality. In the special
case where every set of available alternatives has at least one element which
is unambiguously chosen, we show that WAFRP is equivalent to SAFRP, but
SAFRP still does not characterize rationality. In [5], a fuzzy congruence condi-
tion stronger than SAFRP was proposed and is presented here. It is shown to
be necessary and sufficient for rationality. The interested reader may examine
[8-11, 14, 19, 22] for results concerning choice with fuzzy preferences.
In crisp revealed preference theory, the congruence condition was intro-
duced by Richter [26]. A weaker version was proposed by Sen [28]. In crisp
revealed preference theory, the case where the finite subsets of the universal
set are included in the domain of the choice function is discussed in Arrow
[2] and Sen [28]. In traditional theory, it is known that WARP and SARP are
equivalent conditions and that WARP is a characterization of rationality of
the choice functions. It is also known that in the case of the competitive con-
sumer, WARP is not equivalent to SARP, but SARP characterizes rationality.
(See [17] and [24].)
We now consider fuzzy choice functions and rationality.
Let X denote the (crisp) universal set of alternatives. Let E denote the set
of all non-empty, finite, and crisp subsets of X. Let FP f (X) denote the set
of all nonempty fuzzy subsets of X with finite supports.
The following definition is the third definition of a fuzzy choice function
in this book.
Definition 7.2.1 A fuzzy choice function is a function C : E FP f (X)

such that S E, Supp(C(S)) S.
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
7.2. Fuzzy Choice Functions, Revealed Preference and Rationality 223
set of alternatives has an element which is unambiguously chosen. We discuss

this question later.
Throughout the section, we assume that
C({x})(x) = C({y})(y) for all x, y X.
Definition 7.2.2 For all S E, an element x S is called dominant in S

if y S, C(S)(x) C(S)(y).
Recall that a fuzzy preference relation (a function from X 2 into [0, 1]) is
weakly transitive if x, y, z X, (x, y) (y, x) and (y, z) (z, y) implies
(x, z) (z, x).
Definition 7.2.3 [5] Let be a fuzzy preference relation on X. Then

(1) is called weakly reflexive if x X, 0 < (x, x) (x, y) y S,
(2) is called intensely transitive if it is max-min transitive and weakly
transitive,
(3) is called a weak fuzzy ordering if it is weakly reflexive, complete,
and intensely transitive.
The notion of what has been called weak transitivity appeared in Ponsard
[25].
Definition 7.2.4 A fuzzy choice function C is said to reveal a fuzzy pref-

erence relation if x, y X,
(x, y) = {C(S)(x) | S E, x, y S}.
It follows from Definition 7.2.4 that x, y X, if S E such that

C(S)(x) > 0, then (x, y) > 0 and also that S E, C(S)(x) > 0 im-
plies (x, y) > 0 y S. In the crisp case, Definition 7.2.4 reduces to the most
frequently used definition of a revealed (weak) preference relation.
The fuzzy strict preference relation and the fuzzy indifference relation
corresponding to for this section is defined as follows: x, y X,
(x, y) = 0 ((x, y) (y, x))
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,

e(x, y) = 0 ({C(S)(x) C(S)(y) | S E, x, y S}).

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 is called normal if S E and x S,
C(S)(x) = C(S)(x).
b
From the definition of the revealed preference relation , it can be shown

that normality implies that C({x, y})(x) = (x, y) for all x, y X. That is,
normality of a choice function means that it can be recovered by its own
revealed preference relation.
In the crisp case, Sen [28] introduced the concepts of image and normality.
In the crisp case, there is a natural way of constructing a choice function from
a weak reference relation. This is as follows: S E, let C(S)
b = {x S | xy
y S}, where x, y X and xy can be interpreted to be (x, y) = 1.
In the fuzzy case, however, there are many alternatives. Definition 7.2.6 is a
reasonable one for the situation when choice is fuzzy.
Definition 7.2.7 Let be a fuzzy preference relation on X. Let S E and

x S. Then x is said to be relation dominant in S in terms of if (x, y)
(y, x) y S.
The next two conditions are additional regularity conditions.
Relation dominant revealed preference (RDRP): A choice function
is said to satisfy RDRP if S E and x S, x is dominant in S if and
only if it is relation dominant in S in terms of the fuzzy revealed preference
relation .
Condition of revealed preference dominance (CRPD): A fuzzy
choice function is said to satisfy CRPD if the fuzzy revealed preference re-
lation is such that S E and for all distinct x, y, z S, if x is relation
dominant in S, then (x, z) (y, x).
Condition CRPD says that if x dominates all elements of S, then the
extent to which it is preferred to another element z S cannot be less than
i i
i i
i i
K23798 2015/2/2
i i
the extent to which a third element y of that set is preferred to x. It is easily

verified that in the crisp case RDRP is implied by normality and CRPD is
trivially satisfied.
Definition 7.2.8 A fuzzy choice function is called rational if it is normal,

satisfies RDRP and CRPD, and the fuzzy revealed preference relation is a
weak fuzzy ordering.
We next consider weak and strong axioms of fuzzy revealed preference and
congruence.
The weak axiom of revealed preference (WARP) was proposed by Samuel-
son [26] and the strong axiom of revealed preference (SARP) was proposed by
Houthakker [18], Ville [31], and von Neumann and Morgenstern [32]. In Ar-
row [2] and Sen [28], these axioms were adapted to the case of crisp set-valued
choice functions. We now present fuzzy versions of these axioms.
Definition 7.2.9 Let C be a fuzzy choice function. Define a fuzzy preference

e (x, y) on X so that for all x, y X,
relation e (x, y) > 0 if and only if
there exists x0 , x1 , . . . , xn X such that x0 = x, xn = y, and e(xi1 , xi ) > 0
for i = 1, 2, . . . , n.
e (x, y) >
The reader is asked in the exercises to show that for all x, y X,

0 implies e (y, x) = 0.
e can be thought of as providing a degree of indirect
strict preference.

the
(1) weak axiom of fuzzy revealed preference 1 (WAFRP1) if x, y
X, e(x, y) > 0 implies (y, x) < 1.
(2) strong axiom of fuzzy revealed preference 1 (SAFRP1) if
x, y X, e (x, y) > 0 implies (y, x) < 1.
WAFRP1 and SAFRP1 are consistency conditions. WAFRP1 says that
if x is directly strictly preferred to y to a positive extent, then y cannot be
definitely weakly preferred to x. SAFRP1 says that this conclusion is valid if
x is indirectly strictly preferred to y to a positive extent under the relation
e .

Definition 7.2.11 Let C be a fuzzy choice function. Define a fuzzy preference

relation on X so that for all x, y X and for all t (0, 1], (x, y) t
if and only if there exists x0 , . . . , xn X such that x0 = x, xn = y, and
e(xi1 , xi ) t for i = 1, 2, . . . , n.

Under the relation , x is indirectly strictly preferred to y at least to the
extent t if and only if there is a sequence of elements connecting x to y so that
each alternative in the sequence is directly strictly preferred to its immediate
successor to the extent t.
i i
i i
i i
K23798 2015/2/2
i i
226 7. Rationality

the
(1) weak axiom of fuzzy revealed preference 2 (WAFRP2) if for all
x, y X and for all t such that 0 < t 1,
e(x, y) t implies (y, x) 1 t;
(2) strong axiom of fuzzy revealed preference 2 (SAFRP2) if for all
x, y X and for all t such that 0 < t 1, (x, y) t implies (y, x) 1 t.
WAFRP2 and SAFRP2 are also consistency conditions. Clearly, SAFRP1
(SAFRP2) implies WAFRP1 (WAFRP2). Also, SAFRP2 (WAFRP2) implies
SAFRP1 (WAFRP1). Both WAFRP1 and WAFRP2 reduce to WARP in the
case crisp. Similarly, both SAFRP1 and SAFRP2 reduce to SARP. Thus both
WAFRP1 and WAFRP2 are valid fuzzifications of the crisp axiom WARP.
Also, SAFRP1 and SAFRP2 are valid fuzzifications of SARP.
Definition 7.2.13 (Banerjee [5]) Let C be a fuzzy choice function. Then

(1) C is called fuzzy congruent 1 (FC1) if S E and x S, y domi-
nant in S C(S)(x) = (x, y);
(2) C is called fuzzy congruent 2 (FC2) if S E and x S, y domi-
nant in S and (x, y) x is dominant in S;
(3) C is called fuzzy congruent 3 (FC3) if S E and x S, and all
t such that 0 < t 1, {C(S)(y) t and (x, y) t} C(S)(x) t;
(4) C is called fuzzy congruent if it satisfies FC1, FC2, and FC3.
It follows that in the crisp case each of the three conditions FC1, FC2, and
FC3 reduces to the weak congruence condition defined in Sen [27]. However,
in the fuzzy case these conditions are independent.
Banerjee states the following in [5]: A congruence condition was introduced
in a crisp revealed preference theory by Richter [26]. Attention was focussed
there on the case of the competitive consumer. It was based on the concept of
indirect revealed preference in the wide sense which used the idea of a tran-
sitive closure of the crisp revealed preference relation. Richters congruence
condition is, in general, independent of SARP. In the case of non-satiety of
the consumer, it is weaker. In the case studied by Sen (i.e., where the choice
function is set valued and its domain includes all finite subsets of the universal
set of alternatives), the weak congruence condition is equivalent to Richters
condition and both are equivalent to WARP as well as SARP.
The fuzzy congruence condition defined here is stronger than SAFRP2 and
hence is stronger than SAFRP1, WAFRP2 and WAFRP2.
We next characterize rationality.
It is known in crisp revealed preference theory that SARP characterizes
rationality. This provides the motivation for determining whether a fuzzy
version of this proposition is true.
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.
By Definition 7.2.4, we get the following values for : (x, x) = (y, y) =

(z, z) = 1 and (x, y) = 1, (y, x) = t, (y, z) = 1, (z, y) = s, (x, z) = 1,
(z, x) = 0. By Definition 7.2.5,

e(x, y) = 0 ((C({x, y})(x) C({x, y})(y))
(C({x, y, z})(x) C({x, y, z})(y)))
= (1 s) (1 r) = 1 r.
We also have

e(y, z) = s,
e(x, z) = 1,
e(y, x) = 0,
e(z, y) = 0,
e(z, x) = 0.
It follows for all x, y X such that
e(x, y) > 0 that (y, x) < 1. It also follows
that there does not exist (u, v) such that e(u, v) = 0, but e (u, v) > 0. Thus
SAFRP1 holds. However, C({x, b y, z})(y) = {(y, x), (y, y), (y, z)} = t 6=
r = C({x, y, z})(y). Hence C is not normal.
It follows that WAFRP1 does not imply normality. However, it can be

shown that WAFRP1 and hence SAFRP1 imply a weak form of normality
in that S E and x S, C(S)(x) > 0 C(S)(x) b > 0. However, this
is insufficient for generating a fuzzy choice function back from its own fuzzy
revealed preference relation.
We establish the following lemmas in order to study the rationality impli-
cations of the other fuzzy revealed preference conditions stated previously.
Lemma 7.2.15 If C is fuzzy congruent 1, then it is normal.

Proof. It follows that S E and x S, C(S)(x) C(S)(x).
b Suppose
C is not normal. Then S E and x S such that C(S)(x) < C(S)(x)b =
{(x, y) | S E, y S}. Thus C(S)(x) < (x, y) y S. This a contradic-
tion of fuzzy congruence 1 since the inequality does not hold for y which is
dominant in S.
The aspect of rationality provided by FC1 guarantees that a fuzzy choice
function can be generated back from the preference relation revealed by it.
Lemma 7.2.16 Suppose C is fuzzy congruence 1. Then S E and x S,

if x is dominant in S, then x is relation dominant in S in terms of .
i i
i i
i i
K23798 2015/2/2
i i
228 7. Rationality
Proof. Let C be FC1 and x be dominant in S E. Then C(S)(y) =

(y, x)y S. However, C(S)(x) C(S)(y). Hence if y S such that
(x, y) < (y, x), then (y, x) > (x, y) C(S)(x) C(S)(y) = (y, x), a
contradiction.
We ask the reader to prove the following result.
Lemma 7.2.17 Suppose C is fuzzy congruent 2. Then S E and x S,

if x is relation dominant in S in terms of , then x is dominant in S.
Lemmas 7.2.16 and 7.2.17 show that under FC1 and FC2 the two concepts
of dominance and relation dominance become equivalent.
Lemma 7.2.18 Suppose C is fuzzy congruent 1 and fuzzy congruent 2. Then

is weakly transitivity.
Proof. Suppose that there exist x, y, z X such that (x, y) (y, x)
and (y, z) (z, y). Let S = {x, y, z}. If y is dominant in S, (x, y) (y, x)
implies that x is dominant in S by FC2. If z is dominant in S, (y, z)
(z, y) implies that y is dominant in S. Hence by x is dominant in S. Thus
x is dominant and so relation dominant in S in all cases by FC1. Hence
(x, z) (z, x).
Lemma 7.2.19 Suppose C is fuzzy congruence 1 and fuzzy congruence 2.

Then C satisfies CRPD.
Proof. Assume FC1 and FC2. Suppose there exist S E and x, y S such
that x is relation dominant in S, but (x, z) < (y, x). Then x is dominant
in S since it is relation dominant by Lemma 7.2.18. Now (y, x) > (x, z)
C(S)(x) C(S)(y). However, this contradicts FC1 since x is dominant.
Lemma 7.2.20 If C is fuzzy congruent 1, then is weakly reflexive.

Proof. Since FC1 implies normality, we have if for any x X, (x, x) = 0,
then C({x})(x) = C({x})(x)
b = (x, x) = 0, a contradiction. Thus (x, x) > 0
x X. Suppose that there exist x, y X such that (x, x) < (x, y). Let S =
{x, y}. Since FC1 implies normality, C(S)(x) = (x, y). Similarly, C(S)(y) =
(y, x). If (x, y) (y, x), then C(S)(x) C(S)(y) and so x is dominant in
S. Hence C(S)(x) = (x, x) by FC1. Thus (x, x) = (x, y), a contradiction. If
(y, x) > (x, y), then C(S)(y) > C(S)(x) and so y is dominant in S. Hence
by FC1, C(S)(y) = (y, y). Thus (y, x) = (y, y). The assumption that
C({x})(x) = C({y})(y) implies (x, x) = (y, y). Hence (y, x) = (y, y) =
(x, x) < (x, y), a contradiction. Thus is weakly reflexive.
The above lemmas show that FC1 and FC2 imply a large part of rationality
as defined in Definition 7.2.8. However, they do not imply full rationality which
i i
i i
i i
K23798 2015/2/2
i i
includes Condition A as well as weak completeness and max-min transitivity

of the revealed preference relation.
Lemma 7.2.21 Suppose C is fuzzy congruent 1. Then S E and x, y, z

X, if x is dominant in S, then (x, z) (y, z).
Proof. Suppose there exist S E and x, y, z X such that x is dominant
in S, but (x, z) < (y, z). Then by normality and FC1, (y, z) > (x, z)
C(S)(x)
b = C(S)(x) = (x, x) = (y, y). This contradicts the fact that is
weakly reflexive by Lemma 7.2.20.
Theorem 7.2.22 (Banerjee [5]) Let C be a fuzzy choice function. Then C

satisfies FC if and only if C is rational.
Proof. Suppose C satisfies FC. By Lemmas 7.2.15-7.2.20, we have that C
is normal, C satisfies RDRP and CRPD and that the fuzzy revealed preference
relation is weakly reflexive and weakly transitive.
Suppose x, y X such that (x, y) = (y, x) = 0. Let S = {x, y}. By
normality, we have C(S) = , which is impossible. Thus is complete. Let
x, y, z X be such that (x, z) < (x, y) (y, z) = t, say. Then (x, y) t
and (y, z) t. Therefore,
(x, y) t > (x, z). (7.14)
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).
Hence C(S)(x) = (x, z) by normality.

Now if y is dominant in S, FC1 implies C(S)(x) = (x, y) t. Since
(x, y) t and by FC1, C(S)(x) = (x, z) < t and so FC3 is contradicted.
Thus x is dominant in S. This contradicts Lemma 7.2.21 by (7.15).
Conversely, assume C is rational. Assume that C is normal and is a
weak fuzzy ordering satisfying RDRP and CRPD. Suppose FC1 doesnt hold.
Then there exists S E and x, y S such that y is dominant in S, but
C(S)(x) < (x, y). Since C is normal, C(S)(x) = C(S)(x).
b Let C(S)(x)
b =
{(x, y) | y S} = (x, z) for some z S. Then C(S)(x) = (x, z) < (x, y).
If (x, y) (y, z), then max-min transitivity is violated. Hence (x, y) >
(y, z). But then CRPD doesnt hold since y is dominant in S by RDRP.
Let S E and x, y S be such that y is dominant in S and (x, y)
(y, x). Since y is dominant in S, rationality implies FC1 which in turn implies
(y, z) (z, y) by Lemma 7.2.16. Now weak transitivity implies (x, z)
i i
i i
i i
K23798 2015/2/2
i i
230 7. Rationality
(z, x) and so x is relation dominant in S. Hence x is dominant in S. Thus

rationality implies FC2.
Let S E and x, y S be such that
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.
Theorem 7.2.23 (Banerjee [5]) Let C be a fuzzy choice function. If C is

rational, then WAFRP2 holds.
Proof. Suppose WAFRP2 does not hold. Then x, y X and t, 0 < t 1,
such that
e(x, y) t,
(7.18)
(y, x) > 1 t. (7.19)
(7.18) implies S E such that C(S)(x)C(S)(y) t. Now S has a dominant
element, say z. Thus
C(S)(y) = C(S)(x) t R(x, y) t

1t
< (y, x) by (7.19).
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
C(S)(y) C(S)(x) t < C(S)(x) (x, y)

(y, z).
Since z is dominant in S, the strict inequality violates FC1 and thus rationality.
We see from Theorem 7.2.22 that FC WAFRP2.
i i
i i
i i
K23798 2015/2/2
i i
Lemma 7.2.24 Let C be a fuzzy choice function If C satisfies FC1,

thenx, y, z X and t, 0 < t 1, ((x, y) 1 t and (y, z) 1 t)
(x, z) 1 t.
Proof. Suppose that there exist x, y, z X such that (x, y) 1 t and
(y, z) 1 t, but (x, z) > 1 t.
Let S = {x, y, z}. If z is dominant in S, C(S)(x) = (x, z) since C satisfies
FC1. FC1 C(S)(x) = (x, z). But (x, z) > 1 t (x, y). Since FC1
implies normality, C(S)(x) = C(S)(x)
b = (x, y). Hence (x, z) = (x, y),
contradiction.
If x is dominant in S, C(S)(x) = (x, x) by FC1. Thus (x, x) = (x, y) <
(x, z) as above. However, this contradicts the weak reflexivity of which
holds by FC1.
Suppose y is dominant in S. Then by Lemma 7.2.21, (y, z) (x, z) >
1 t. This contradicts the hypothesis (y, z) 1 t. However, C(S) 6= .
Hence it must be the case that (x, z) 1 t.
Proposition 7.2.25 Let C be a fuzzy choice function. If C satisfies FC1,

then WAFRP2 and SAFRP2 are equivalent.
Proof. It suffices to show that SAFRP2 follows from FC1 and WAFRP2.
Suppose that x, y X are such that (x, y) t for some t, 0 < t 1. Then
there exists x0 , x1 , . . . , xn X, x0 = x, xn = y and
e(xi1 , xi ) t for all
i = 1, 2, . . . n. Since C satisfies FC1, repeated application of Lemma 7.2.24
7.2.24 yields (y, x) 1 t.
It follows that rationality implies SAFRP2 since rationality implies FC1.
However, WAFRP2 and SAFRP2 are not equivalent. We show in the following
example that neither of these axioms guarantees rationality.
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.
Definition 7.2.28 Let C be a choice function. Suppose S E, x S such

that C(S)(x) = 1. Then C is called rational if C is normal and the fuzzy
revealed preference relation is a fuzzy ordering.
Proposition 7.2.29 Let C be a choice function. Suppose S E, x S

such that C(S)(x) = 1. Then C satisfies FC if and only if C is rational.
There is now a significant change in the result regarding WAFRP2 and
SAFR2.
The proof of the next result is similar to the crisp case, Arrow [2] and Sen
[28].
Proposition 7.2.30 Let C be a choice function. Suppose S E, x S

such that C(S)(x) = 1. Then WAFRP2 and SAFRP2 are equivalent.
Theorem 7.2.31 (Banerjee [5]) Let C be a choice function. Suppose S E,

x S such that C(S)(x) = 1. Then WAFRP2 implies normality of the choice
function and weak transitivity of the fuzzy revealed preference relation.
i i
i i
i i
K23798 2015/2/2
i i
7.3. Exercises 233
Proof. By Lemma 7.2.15, it suffices to show for normality that WAFRP2

implies FC1. If FC1 does not hold for C, S E and x, y S such that x is
dominant in S and C(S)(y) < (x, y). Since C(S)(x) = 1,
C(S)(x) C(S)(y) > C(S)(x) (y, x) = 1 (y, x).
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 ,
C(B, ) = {x B | (x, y) for all y B};
(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
(J 1 is a slightly generalized version of some PCFs presented in [11] and

[15]. J 2 has been discussed by Dutta, Panda, Pattanaik [15].)
2. Let C : P (X) F(R)0 P (X) be a PCF. If C J 1 J 2 , then prove

that for all F(R)0 , there exists an exact 0 F(R) such that for all
B P (X), C(B, ) = D(B, 0 ).
3. Let C : P (X) F(R)0 P (X) be a max-mD PCF, where F(R)0
FT (R). Prove that C J 2 .
e (x, y) > 0 implies

4. Use Definition 7.2.5 to show that for all x, y X,

e (y, x) = 0.
5. [4] Use numerical examples to show that max-min transitivity and weak
transitivity are independent conditions in the set inclusion sense.
6. Show that conditions FC1, FC2, and FC3 are independent.

7. Prove Lemma 7.2.17.
8. Suppose S E, x S such that C(S)(x) = 1. Prove Proposition
7.2.29.
The following exercises are based on [33]. In the following exercises, a
fuzzy choice function is a function C : E FP (X)
Let C be a fuzzy choice function and the revealed preference. Consider
the statements:
(1) is a -regular revealed preference relation and C is normal;
(2) is a -regular generated preference relation and C is normal;
(3) C satisfies WFCA;
(4) C satisfies SFCA;
(5) C satisfies WAFRP;
(6) C satisfies SAFRP;
(7) = e;
(8) = e and C is normal.
9. Prove that
(i) (1), (2), (3), and (4) are equivalent.
(ii) (5), (6), (7), and (8) are equivalent.
(iii) If (T, S, N ) is a strong DeMorgan triple, then (3) (5).
10. Let N denote a strong negation.
Condition F : If S1 S2 , then C(S2 )(x) C(S1 )(x) for any x S1 .
Condition F 2 : For all x, y S, C(S)(x) C({x, y})(x).
i i
i i
i i
K23798 2015/2/2
i i
7.4. References 235
Condition F : If S1 S2 and x, y S1 , then C(S1 )(x) C(S1 )(y)

C(S2 )(x) C(S2 )(y).
Condition F + : If S1 S2 and x, y S1 , then C(S1 )(x) C(S2 )(y)
C(S2 )(y).
Condition F : If S1 S2 , x, y S1 , and x 6= y, then C(S1 )(x)
C(S1 )(y) N (C(S2 )(x) ({N (C(S2 )(z)) | z 6= x})).
Condition F 2 : For all x, y S, C({x, y})(x) C({x, y})(y) C(S)(x)
C(S)(y).
Condition F 0 : For all x, y S1 S2 , C(S1 )(x) C(S1 )(y) C(S2 )(x)
C(S2 )(y).
Condition F : For all x S1 S2 , C(S1 )(x)C(S2 )(x) C(S1 S2 )(y).
Prove that if a fuzzy choice function C satisfies F , then it satisfies F .
Prove that if C is a fuzzy choice function that salsifies F + and is
continuous, then C satisfies F , F , is -transitive.
11. If C is a fuzzy choice function that satisfies F 2 and F 2 , prove that C
satisfies F , F 0 , F , F , and F .
12. Prove that a fuzzy choice function C satisfies WFCA if and only if it
satisfies F and F + .
13. Prove that if a fuzzy choice function satisfies F and F , then C is
normal under the G odel t-norm.
14. Prove that if a fuzzy choice function C satisfies F and F , then is
-quasi-transitive.
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
6. C. R. Barrett, Fuzzy preferences and choice: A progress report, Fuzzy

7. C. R. Barratt and P. K. Pattanik, On vague preferences, in: H. von G.
Enderle, Ed., Ethik and Wirtschafwswissenschaftz, Dunker and Hum-
bolt, Berlin 1985.
9. C. R. Barrett, P. K. Pattanaik, and M. Salles, On choosing rationally
10. C. R. Barrett, P. K. Pattanaik, and M. Salles, Rationality and aggre-
gation of preferences in an ordinally fuzzy framework, Fuzzy Sets and
Systems, 49 (1992), 913.
11. K. Basu, Fuzzy revealed preference theory, Journal Econom. Theory, 32
(1984) 212227.
12. M. Dasgupta and R. Deb, Rational choice with fuzzy preferences,
Mimeograph (1987).
13. M. Dasgupta and R. Deb, Fuzzy Choice Functions, Mimeo, Dept. of
Economics, Southern Methodist Univ., 1988.
14. B. Dutta, Fuzzy preferences and social choices, Math. Social Sci., 13
(1987) 215229.
15. B. Dutta, S. Panda, and P. R. Pattanaik, Exact choice and fuzzy pref-
erences, Math. Social Sci., 11 (1986) 5368.
16. S. Fotso and L. A. Fono, On the consistency of some crisp choice func-
tions based on a strongly complete fuzzy pre-order, New Mathematics
17. D. G. Gale, A note on revealed preferences, Economica, 27 (1960) 971
978.
18. H. S. Houthakker, Revealed preference and utility function, Economica,
NS 17 (1950) 159174.
19. J. Kacprzyck and M. Roubens, Eds., Nonconventional Preference Rela-
tions in Decison-Making, Lecture notes in Economics and Mathematical
Systems, Springer, Berlin, 1988.
20. H. Nurumi, Approaches to collective decision-making with fuzzy prefer-
ence relations, Fuzzy Sets and Systems, 6 (1981) 249259.
i i
i i
i i
K23798 2015/2/2
i i
7.4. References 237
22. S. V. Ovchnnikov, Structure of fuzzy binary relations, Fuzzy Sets and

Systems, 6 (1981) 169195.
23. S. V. Ovchinnikov and V. R. Ozernoy, Using fuzzy binary relations for

identifying noninferior decision alternatives, Fuzzy Sets and Systems, 25
(1988) 2132.
24. H. Peters and P. Wakker, Independence of irrelevant alternatives and
revealed group preference, Econometrica, 59 (1992) 17871801.
25. C. Ponsard, An application of fuzzy subsets theory to the analysis of the

consumers spatial preference, Fuzzy Sets and Systems, 5 (1981) 235244.
645.
27. P. A. Samuelson, A note on the pure theory of consumers behavior,

Economica, 5 (1938) 6167, 353354.
28. A. K. Sen, Choice functions and revealed preference, Rev Economica
Studied, 38 (1971) 307317.
29. A. K. Sen, Social choice theory: A re-examination, Econometrica, 45

(1977) 5389.
30. Z. Switalski, Choice functions associated with fuzzy preference relations,
in [19] 106118.
31. J. Ville, Sur les conditions dexistence of dune ophelimite totale et dun
indice du niveau des pix, Annales de l Universite de Lyon, 9 (1946)
3229.
32. J. von Neumann and O. Morgenstern, Theory of Games and Economic
Behavior, Princeton University Press, Princeton 1957.
33. C. Wu, X. Wang, and Y. Hao, A further study on rationality conditions

of fuzzy choice functions, Fuzzy Sets and Systems, 176 (2011) 119.
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
Arrow-Type Results under

Intuitionistic Fuzzy
Preferences
8.1 Fuzzy Preference Profiles and Fuzzy

Aggregation Rules
and one for with respect to a t-norm . We show that one proof can
be obtained from the other by considering and dual under an involutive
fuzzy complement. This section follows the development of that in [23].
Recall that an intuitionistic fuzzy set consists of a set X and two fuzzy
subsets and of X such that (x) + (x) 1. In [16], a factorization of an
intuitionistic fuzzy binary relation into a unique indifference component and
a family of regular strict components was provided. This result generalizes
a result in [10] with the (max, min) intuitionistic fuzzy t-conorm. In [16], a
characterization of C-transitivity for a continuous t-representable intuitionistic
fuzzy t-conorm C was established. This enabled the authors to determine
necessary and sufficient conditions on a C-transitive intuitionistic fuzzy binary
relation under which a component of satisfies pos-transitivity and negative
transitivity.
In [9], some results on t-representable intuitionistic fuzzy t-norms (T, S)
were established, where T is a fuzzy t-norm and S is a fuzzy t-conorm such
that T (a, b) = 1 S(1 a, 1 b) for all a, b [0, 1].
We extend some known Arrowian results involving fuzzy set theory to
results involving intuitionistic fuzzy sets. Many of the proofs for intuitionistic
fuzzy sets can be obtained from existing results for fuzzy sets in a dual manner.
239
i i
i i
i i
K23798 2015/2/2
i i
240 8. Arrow-Type Results under Intuitionistic Fuzzy Preferences
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.
Definition 8.1.1 Let h , i be an intuitionistic fuzzy binary relation on X.

Then
(1) h , i is called reflexive if x X, (x, x) = 1 and (x, x) = 0;
(2) h , i is called complete if x, y X, either (x, y) > 0 or
(y, x) > 0 and (x, y) < 1 or (y, x) < 1;
(3) h , i is called min-max transitive or simply transitive if
x, y, z X, (x, y) (y, x) (x, z) and (x, y) (y, x) (x, z).
i i
i i
i i
K23798 2015/2/2
i i
8.1. Fuzzy Preference Profiles and Fuzzy Aggregation Rules 241
(4) h , i is called partially transitive if x, y, z X, (x, y) > 0 and

(y, z) > 0 implies (x, z) > 0 and (x, y) < 1 and (y, z) < 1 implies
(x, z) < 1.
Definition 8.1.2 Let h , i be an intuitionistic fuzzy binary relation on X.

Then
(1) h , i is called symmetric if x, y X, (x, y) = (y, x) and
(x, y) = (y, x).
(2) h , i is called asymmetric if x, y X, (x, y) > 0 implies
(y, x) = 0 and (x, y) < 1 implies (y, x) = 1.
Let T denote a set of intuitionistic fuzzy binary relations on X. An
intuitionistic fuzzy preference profile on X is a n-tuple of intu-
itionistic fuzzy binary relations = (h1 , 1 i, . . . , hn , n i) describing
the intuitionistic fuzzy preferences of all individuals. Let T n = { =
(h1 , 1 i, . . . , hn , n i) | hi , i i T, i = 1, . . . , n}. Let P(X) de-
note the power set of X. For any T n and for all S P(X), let
eS = (h1 eS , 1 eS i, . . . , hn eS , n eS i), where i eS = i |SS , the re-
striction of i to S S, i = 1, . . . , n and similarly for i eS . For h , i T,
we associate an intuitionistic asymmetric fuzzy relation h , i with h , i.
For all T n and x, y X, let
R(x, y; ) = {i N | i (x, y) > 0 and i (x, y) < 1}

and P (x, y; ) = {i N | i (x, y) > 0 and i (x, y) < 1},
where is an intuitionistic asymmetric fuzzy binary relation asso-

ciated with , i = 1, . . . , n.
Definition 8.1.3 Let h , i T and h , i be an asymmetric intuition-

istic fuzzy preference relation associated with h , i .
(1) h , i is called simple if x, y X, (x, y) = (y, x) implies
(x, y) = (y, x) and (x, y) = (y, x) implies (x, y) = (y, x).
(2) h , i is called regular if x, y X, (x, y) > (y, x)
(x, y) > 0 and (x, y) < (y, x) (x, y) < 1.
It follows easily that if is regular, then is simple as is the case for
.
Definition 8.1.4 Let h , i T. Define h(i) , (i) i T, i = 0, 1, 2, 3, 4, 5,

as follows: x, y X,
(1) (0) (x, y) = (x, y) if (y, x) = 0 and (0) (x, y) = 0 otherwise;
(0) (x, y) = (x, y) if (y, x) = 1 and (0) (x, y) = 1 otherwise;
(2) (1) (x, y) = (x, y) if (x, y) > (y, x) and (1) (x, y) = 0 other-
wise;
(1) (x, y) = (x, y) if (x, y) < (y, x) and (1) (x, y) = 1 other-
wise;
i i
i i
i i
K23798 2015/2/2
i i
(3) (2) (x, y) = 1 (y, x);

(2) (x, y) = (y, x);
(4) (3) (x, y) = 0 ( (x, y) (y, x)).
(3) (x, y) = 1 (1 + (x, y) (y, x)).
(5) (4) (x, y) = ( (x, y) (y, x))/(1 (y, x)) if (x, y) > (y, x)
and (4) (x, y) = 0 otherwise;
(4) (x, y) = (x, y)/ (y, x)) if (x, y) < (y, x) and (4) (x, y) =
1 otherwise;
(6)(5) (x, y) = (, (x, y)w (y, x) )1/ if (x, y) > (y, x) and
(5) (x, y) = 0 otherwise, > 0;
(5) (x, y) = 1 [(1 (x, y))w (1 (y, x)) ]1/ if (x, y) <
(y, x) and (5) (x, y) = 1 otherwise, > 0.
Definition 8.1.5 A function fe : T n T is called a fuzzy aggregation

rule.
Let fe be a fuzzy aggregation rule. Let = (h1 , 1 i, . . . , hn , n i) T n .
Let FRn (X) = {(1 , . . . , n ) | (h1 , 1 i, . . . , hn , n i) T n } and
FRn (X) = {(1 , . . . , n ) | (h1 , 1 i, . . . , hn , n i) T n }. Let ff :
FRn (X) F R(X) and fe : FRn (X) F R(X). Then ff and f induce a
e
fuzzy aggregation rule fe on T. That is fe() = hff ( ), f ( )i, where =
e
(1 , . . . , n ) and = (1 , . . . , n ). We sometimes write fe = (ff , f ). We
e
sometimes suppress the notation hf ( ), f ( )i and write h , i. In this
f e
case, we write h , i for the asymmetric fuzzy preference relation associated
with h , i.
Definition 8.1.6 Let fe : T n T be a fuzzy aggregation rule. Then fe satisfies

(1) Independence of irrelevant alternatives (IIA1) if (h1 , 1 i,
. . . , hn , n i), (h01 , 01 i, . . . , h0n , 0n i) T n , x, y X,
[i N, i (x, y) = 0i (x, y) (x, y) = 0 (x, y)],

[i N, i (x, y) = 0i (x, y) (x, y) = 0 (x, y)].
(2) Pareto condition (PC) with respect to h , i if (h1 , 1 i, . . . ,

hn , n i) T n , x, y X,
(x, y) {i (x, y) | i N },
(x, y) {i (x, y) | i N }.
(3) Positive responsiveness (PR) with respect to h , i if

(h1 , 1 i, . . . , hn , n i), (h01 , 01 i, . . . , h0n , 0n i) T n , x, y X,
i = 0i i 6= j, (x, y) = (y, x), and

0
(j (x, y) = 0 and j (x, y) > 0
0
or j (y, x) > 0 and j (y, x) = 0) 0 (x, y) > 0.
i i
i i
i i
K23798 2015/2/2
i i
i = 0i i 6= j, (x, y) = (y, x), and

0
(j (x, y) = 1 and j (x, y) < 1
0
or j (y, x) < 1 and j (y, x) = 1) 0 (x, y) < 1.
Let = (h1 , 1 i, . . . , hn , n i) T n . Let fe be a fuzzy aggregation

rule. In the following, when an asymmetric fuzzy preference of a particular
type for h , i = fe() is assumed, then it is assumed that the associated
asymmetric fuzzy preference for hi , i i is of the same type, i = 1, . . . , n.
The proof of the next result follows easily.
Proposition 8.1.7 Let h , i T. Let h , i, h , i be two different

types of asymmetric fuzzy preference relations associated with h , i. Sup-
pose x, y X, (x, y) > 0 if and only if (x, y) > 0 and (x, y) < 1 if and
only if (x, y) < 1. Let fe : T n T be a fuzzy aggregation rule. Let T n .
Then fe satisfies PR with respect to h , i if and only if fe satisfies PR with
respect to h , i.
Corollary 8.1.8 Let fe : T n T be a fuzzy aggregation rule. Then fe satisfies

PR with respect to h(3) , (3) i if and only if fe satisfies PR with respect to
h(1) (1) i.
Proof. It suffices to show that x, y X, (3) (x, y) > 0 if and only if
(1) (x, y) > 0 and (3) (x, y) < 1 if and only if (1) (x, y) < 1. However, this
follows immediately from the definitions since (x, y) > (y, x) if and only
if (x, y) (y, x) > 0 and (x, y) < (y, x) if and only if 1 + (x, y)
(y, x) < 1.

= (h1 , 1 i, . . . , hn , n i) T n , x, y X,
* n n
+
X X
f ()(x, y) =
e wi i (x, y), wi i (x, y) ,
i=1 i=1
Pn
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.
Proposition 8.1.10 Let fe be defined by Definition 8.1.9. Then fe satisfies

PR with respect to (3) and (1) .
Proof. Let , 0 T n . Suppose i = 0i , i = 1, . . . , n, i 6= j. Let x, y X.
Suppose (x, y) = (y, x). Suppose also that either (1) (j (x, y) = 0 and
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))].
For case (1), we write
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.
For case (2), we write
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.
Proposition 8.1.11 Let fe be the fuzzy aggregation

Pn rule defined in Defini-
tion
Pn 8.1.9. Then x, y X, (2) (x, y) = w
i=1 i i (x, y) and (2) (x, y) =
w
i=1 i i (x, y).
i i
i i
i i
K23798 2015/2/2
i i
Proof. Let x, y X. Then

n
X n
X
(2) (x, y) = 1 (y, x) = 1 wi i (y, x) = 1 wi (1 i (x, y))
i=1 i=1
n
X n
X n
X
= 1 wi + wi i (x, y) = wi i (x, y).
i=1 i=1 i=1
and
n
X n
X
(2) (x, y) = (y, x) = wi i (y, x) = wi i (y, x).
i=1 i=1
Example 8.1.12 Let fe be the fuzzy aggregation rule defined in Definition

8.1.9. That fe does not satisfy PR with respect to (2) is shown in Example
4.1.8.

8.1.9. Then fe satisfies PC with respect to h(3) , (3) i.
Proof. Let x, y X and T n . Let m = mx,y = {i (x, y) | i =
1, . . . , n}. There is no loss in generality in assuming m = 1 (x, y). Suppose
m > 0. Then

2 (x, y) n (x, y)
1 w1 + w2 + . . . + wn + w2 1 + . . . + wn 1
m m
1 (x, y) n (x, y)
= w1 + . . . + wn .
m m
Thus
1 (x, y) = m w1 1 (x, y) + . . . + wn n (x). (8.1)
Since
Pn m > 0, i (x, P y) > 0 i N. Thus i P(x, y) > i (y, x) i N. Hence
n n
i=1 wi i (x, y) > i=1 wi i (y, x). Thus i=1 wi [i (x, y) i (y, x)] > 0.
But i (x, y) i (y, x) = i (x, y), i = 1, . . . , n. Hence by (8.1),
n
X
0 < 1 (x, y) 1 (y, x) wi [i (x, y) i (y, x)]
i=1
n
X n
X
= wi i (x, y) wi i (y, x)],
i=1 i=1
or {i (x, y) | i N } (x, y) (y, x) = (3) (x, y). Hence fe satisfies

PC with respect to (3) . If m = 0, then clearly fe satisfies PC with respect
to (3) .
i i
i i
i i
K23798 2015/2/2
i i
We now show fe satisfies PC with respect to (3) . Let x, y X. Let

m = mx,y = {i (x, y) | i N }. Suppose m < 1. Then Thus (x, y)
{i (x, y) | i N }. If m = 1, then the result is immediate.
Definition 8.1.14 Let h , i T. Then h , i is said to satisfy type T2 -

transitivity if x, y, z X, (x, z) (x, y) + (y, z) 1 and (x, z)
(x, y) + (y, z).

T2 -transitive if T n , fe() is T2 -transitive.

8.1.9. Let = (h1 , 1 i, . . . , hn , n i) T n . If hi , i i is T2 -transitive
i N, then fe() is T2 -transitive.
Proof. Let x, y, z X. Then i (x, z) i (x, y) + i (y, z) 1 and
i (x, z) i (x, y) + i (y, z) i N. Thus
n
X n
X n
X n
X
(x, z) = wi i (x, z) wi i (x, y) + wi i (y, z) wi
i=1 i=1 i=1 i=1
= (x, y) + (y, z) 1
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).
8.2 IIA3 and Nondictatorial Fuzzy

Aggregation Rules
We now consider alternative definitions of IIA.
Let H2 = {h , i T | h , i is T2 -transitive}.
Let C be a nonempty subset of N. Then C is called a coalition.
Definition 8.2.1 Let fe be a fuzzy aggregation rule. An individual j N

is called a dictator if for all distinct x, y X and all = (h1 , 1 i,
. . . , hn , n i) T n , j (x, y) > 0 (x, y) > 0 and j (x, y) < 1
(x, y) < 1.
A fuzzy aggregation rule fe is said to be nondictatorial if it is not dicta-

torial.
i i
i i
i i
K23798 2015/2/2
i i
8.2. IIA3 and Nondictatorial Fuzzy Aggregation Rules 247
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.
Definition 8.2.3 Let h , i, h0 , 0 i T. Let Im( ) = {s1 , . . . , sk } and

Im(0 ) = {t1 , . . . , tm } be such that s1 < . . . < sk and t1 < . . . < tm . Then
and 0 are said to be equivalent, written 0 , if s1 = 0 t1 = 0, k = m,
and ( )si = (0 )ti (level sets) for i = 1, . . . , k.
Let Im( ) = {u1 , . . . , uh } and Im(0 ) = {v1 , . . . , vn } be such that u1 >

. . . > uh and v1 > . . . > vn . Then and 0 are said to be equivalent,
written 0 , if s1 = 1 t1 = 1, h = n, and ( )si = (0 )ti (level sets)
for i = 1, . . . , h.
Let h , i, h0 , 0 i T. Then h , i and h0 , 0 i are said to be equiv-
alent, written h , i h0 , 0 i, if and 0 are equivalent and and 0
are equivalent.
If the intuitionistic fuzzy binary relations in Definition 8.2.3 are reflexive,
then Im( ) = 1 = Im(0 ) and Im( ) = 0 = Im(0 ).
Proposition 8.2.4 Let h , i, h0 , 0 i T. Let x, y X. Suppose h , i

h0 , 0 i. Then (x, y) > (y, x) if and only if 0 (x, y) > 0 (y, x) and
(x, y) > (y, x) if and only if 0 (x, y) > 0 (y, x).
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 .
Definition 8.2.5 Let (ff , f ) be a fuzzy aggregation operator. Then (f , f )

e f e
is said to be independent of irrelevant alternatives (IIA3) if x, y
X, (h1 , 1 i, . . . , hn , n i), (h01 , 01 i, . . . , h0n , 0n i) T n , i e{x,y}
0i e{x,y} i N ff f 0
()e{x,y} f ( )e{x,y} ] and [i e{x,y} i e{x,y}
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
Definition 8.2.6 Let the fuzzy aggregation rule fe : T n T be induced by the

fuzzy aggregation rules ff and f as follows = (h1 , 1 i, . . . , hn , n i)
e
n
T , x, y X,

1 if x = y,
ff (
)(x, y) = 1 if i (x, y) > 0 i N,
t () otherwise.

0 if x = y,
fe ( )(x, y) = 0 if i (x, y) < 1 i N,
t () otherwise.

Clearly, fe() = (ff

, f )() in Definition 8.2.6 is reflexive and complete.
e
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
Proof. Let h , i, h0 , 0 i T n and x, y X. Suppose i e{x,y}

0i e{x,y} i N. Let i N. Then i e{x,y} (x, y) > i e{x,y} (y, x)
0i e{x,y} (x, y) > 0i e{x,y} (y, x) by Proposition 8.2.4. Thus i (x, y) >
i (y, x) 0 (x, y) > 0 (y, x). Hence by definition of ff
i i , f ( )(x, y) = 1
f
0 e 0 0
ff ( )(x, y) = 1 and so f ( )(x, y) = t ( ) f ( )(x, y) = t ( ). Hence
f
fe ( )e{x,y} fe (0 )e{x,y} . Similarly, fe ( )(x, y) = 0 fe (0 )(x, y) = 0.
Hence fe ( )e{x,y} fe (0 )e{x,y} .

If t is a constant function in Proposition 8.2.7, then we can conclude

fe()e{x,y} = fe(0 )e{x,y} in the proof. If t() 1/2, then we can conclude
that fe() is strongly complete, i.e., fe()(x, y) + fe()(y, x) 1.
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.
Proposition 8.2.9 Let strict preference be defined by (1) or (3) . Suppose

Im(t) (0, 1). Let (ff , f )be the fuzzy aggregation rule defined in Definition
e
8.2.6. Then fe satisfies PC with respect to (1) and is weakly Paretian with
respect to (3) .
Proof. Let x, y X and (h1 , 1 i, . . . , hn , n i) T n . If iN i (x, y)
= 0, then clearly (x, y) iN i (x, y). Suppose iN i (x, y) > 0. Then
i (x, y) > 0 i N. Hence i (x, y) > i (y, x) i N. Thus ff ( )(x, y) =
1 and ff (
)(y, x) = t (
) < 1. Hence (x, y)
iN i (x, y) for =
(1) and (x, y) > 0 for = (3) .
i i
i i
i i
K23798 2015/2/2
i i
8.2. IIA3 and Nondictatorial Fuzzy Aggregation Rules 249
If iN i (x, y) = 1, then clearly (x, y) iN i (x, y). Suppose

iN i (x, y) < 1. Then i (x, y) < 1 i N. Hence i (x, y) 0. Hence
(x, y) iN i (x, y) for = (1) and (x, y) < 1 for = (3) .
Definition 8.2.10 Let h , i T. Then h , i is called (max-min)

quasi-transitive if x, y, z X, (x, z) (x, y) (y, z) and (x, z)
(x, y) (y, z).
Proposition 8.2.11 Let be defined by (1) . Let h , i T. If h , i is

max-min transitive, then h , i is max-min quasi-transitive.
Proof. The case for is known, Proposition 5.4.4 or [12, Proposition 2.9
(2), p. 220]. Suppose satisfies (max-min) transitivity, but not (max-min)
quasi-transitivity. Then there exists x, y, z X such that
(x, z) (x, y) (y, z), (x, z) > (x, y) (y, z).
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 ,
(z, y) (z, x) (8.10)
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 ).
Theorem 8.2.12 Let be defined by (1) . Then there exists a nondictatorial

n
, f ) : FRT r FR that is max-min transitive,
fuzzy aggregation rule fe = (ff e
independent of irrelevant alternatives IIA3, and PC.
Proof. Let fe be defined as in Definition 8.2.6. Clearly, fe is reflexive and
complete. By Propositions 8.2.7 and 8.2.9, fe is independent of irrelevant al-
ternatives IIA3 and PC. By definition, fe((h1 , 1 i, . . . , hn , n i)) is reflex-
ive (h1 , 1 i, . . . , hn , n i) FRnT r . It remains to be shown that fe is
transitive. Let (h1 , 1 i, . . . , hn , n i) F RnT r . Let x, y, z X. Suppose
( )(x, y) f ( )(y, z) = 1. Assume x 6= y 6= z. Then i N, i (x, y) > 0
ff f
and i (y, z) > 0. By Proposition 8.2.11, we have i N that i (x, z)
i (x, y) i (y, z) since i is transitive. Thus i (x, z) > 0 i N. Hence
i (x, z) > i (z, x) i N. Thus ff ( )(x, z) = 1. Assume x = y 6= z.
Then i N, i (y, z) > 0. Now i N, i (x, z) = i (y, z) > 0. Thus
( )(x, z) = 1. A similar argument holds for x 6= y = z. The desired result
ff
when x = y = z is immediate. If ff ( )(x, y) f ( )(y, z) = t ( ), then
f
clearly ff ( )(x, z) f ( )(x, y) f ( )(y, z).
f f
Suppose f ( )(x, y) f ( )(y, z) = 0. Assume x 6= y 6= z. Then i
e e
N, i (x, y) < 1 and i (y, z) < 1. By Proposition 8.2.11, we have i N
that i (x, z) i (x, y) i (y, z) since i is transitive. Thus i (x, z) < 1
i N. Hence i (x, z) < i (z, x) i N. Thus fe ( )(x, z) = 0. Assume
x = y 6= z. Then i N, i (y, z) < 1. Now i N, i (x, z) = i (y, z) < 1.
Thus fe ( )(x, z) = 0. A similar argument holds for x 6= y = z. The desired
result when x = y = z is immediate. If fe ( )(x, y) fe ( )(y, z) = t ( ),
then clearly fe ( )(x, z) fe ( )(x, y) fe ( )(y, z).
Example 8.2.13 An example of (2) that is T2 -transitive and partially tran-

sitive, but not max-min quasi-transitive can be found in Example 4.1.24.
i i
i i
i i
K23798 2015/2/2
i i
8.3. IIA2, IIA4, and Fuzzy Arrows Theorem 251
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,
where 0 < t ( ) < 1. Let = (2) . Suppose i (x, y) > 0 i N. Then

x 6= y. Hence (x, y) = 1 ff ( )(y, x) = 1 t ( ) > 0. Thus f is weakly
f
Paretian with respect (2) . Clearly, ff satisfies IIA3 by a similar argument as
in Proposition 8.2.7. Now suppose i (x, y) < 1 i N. Then x 6= y. Hence
(x, y) = fe ( )(y, x) = t ( ) < 1. Thus fe is weakly Paretian with respect
to (2) . Clearly, fe satisfies IIA3 by a similar argument as in Proposition 8.2.7.
That every member of N is a dictator follows by definition since x, y N,
x 6= y, (x, y) > 0 and (x, y) < 1.
8.3 IIA2, IIA4, and Fuzzy Arrows Theorem

In this section, we give consideration to two new definitions of independence of
irrelevant alternatives, IIA. Recall that we sometimes suppress the underlying
fuzzy preference aggregation rule and write (x, y) for fe()(x, y) and (x, y) >
0 for fe()(x, y) > 0 and fe()(y, x) = 0.
Let (h1 , 1 i, . . . , hn , n i) T n . Recall that we sometimes write
h , i for (h1 , 1 i, . . . , hn , n i), where = (1 , . . . , n ) and =
(1 , . . . , n ).
Definition 8.3.1 Let fe = (ff , f ) be a fuzzy preference aggregation rule.

e
Then fe is said to be independent of irrelevant alternatives IIA2 if
(h1 , 1 i, . . . , hn , n i), (h01 , 01 i, . . . , h0n , 0n i) T n , x, y X,
Supp(i e{x,y} ) = Supp(0i e{x,y} ) i N implies Supp(ff ( )e{x,y} ) =
0 0
Supp(f ( )e{x,y} ) and Cosupp(i e{x,y} ) = Cosupp(i e{x,y} ) i N
f
implies Cosupp(fe ( )e{x,y} ) = Cosupp(fe (0 )e{x,y} ).

Definition 8.3.2 Let fe be a fuzzy preference aggregation rule. Let (x, y)

X X. Let be a fuzzy subset of N. Then
i i
i i
i i
K23798 2015/2/2
i i
(1) is called semidecisive for x against y, written xD e y, if (h1 , 1 i,

. . . , hn , n i) T n , [(i (x, y) > 0 i Supp() and j (y, x) > 0
j / Supp()] implies (x, y) > 0) and (i (x, y) < 1 i Cosupp() and
j (y, x) < 1 j / Cosupp()) implies (x, y) > 0].
(2) is called decisive for x against y, written xD y, if h , i FRn ,
[(i (x, y) > 0 i Supp()] implies (x, y) > 0) and (i (x, y) < 1 i
Cosupp()] implies (x, y) < 1.]
Definition 8.3.3 Let fe be a fuzzy preference aggregation rule. Let be a fuzzy

subset of N. Then is called semidecisive (decisive) if (x, y) X X,
is semidecisive (decisive) for x against y.

is dictatorial with dictator i. Let be a fuzzy subset of N. Suppose Supp() is
a coalition. Then is decisive if and only if i Supp()Cosupp().
Proof. Suppose is decisive. Suppose i / Supp()Cosupp(). Let
x, y X be such that x 6= y. Then (h1 , 1 i, . . . , hn , n i) T n such
that j (x, y) > 0 j Supp() and i (y, x) > 0 and so (x, y) > 0
and (y, x) > 0, a contradiction, or j (x, y) < 1 j Cosupp() and
i (y, x) < 1. Thus (x, y) < 1 and (y, x) < 1, a contradiction. Hence i
Supp()Cosupp(). The converse is immediate.
Definition 8.3.5 Let h , i FR. Then

(1) h , i is said to be partially quasi-transitive if x, y, z X,
(x, y) (y, z) > 0 implies (x, z) > 0 and x, y, z X, (x, y)
(y, z) < 1 implies (x, z) < 1.
(2) h , i is said to be acyclic if x1 , . . . , xn X, (x1 , x2 ) . . .
(xn1 , xn ) (x1 , xn ) and x1 , . . . , xn X, (x1 , x2 ). . . (xn1 , xn )
(x1 , xn )
(3) h , i is said to be partially acyclic if x1 , . . . , xn X,
(x1 , x2 ). . . (xn1 , xn ) > 0 implies (x1 , xn ) > 0 and x1 , . . . , xn X,
(x1 , x2 ) . . . (xn1 , xn ) < 1 implies (x1 , xn ) < 1.

(1) fe is said to be (partially) transitive if FRn , fe() is (partially)
transitive;
(2) fe is said to be (partially) quasi-transitive if FRn , fe() is
(partially) quasi-transitive;
(3) fe is said to be (partially) acyclic if FRn , fe() is (partially)
acyclic.
In the remainder of this and the next section, we assume that |X| 3
unless otherwise specified.
i i
i i
i i
K23798 2015/2/2
i i
Proposition 8.3.7 Let h , i be an intuitionistic fuzzy binary relation on

X. If h , i is partially transitive, then h , i is partially quasi-transitive
with respect to = (0) .
Proof. Let x, y, z X. Suppose (x, y) > 0 and (y, z) > 0. Then
(x, y) > 0, (y, x) = 0 and (y, z) > 0, (z, y) = 0. Hence (x, z) > 0.
Suppose (x, z) = 0. Then (z, x) > 0. Thus since (x, y) > 0, (z, y) > 0
by partial transitivity, a contradiction. Thus (x, z) > 0.
Suppose (x, y) < 1 and (y, z) < 1. Then (x, y) < 1, (y, x) < 1
and (y, z) < 1, (z, y) = 1. Hence (x, z) < 1. Suppose (x, z) = 1. Then
(z, x) < 1. Thus since (x, y) < 1, (z, y) < 1 by partial transitivity, a
contradiction. Hence (x, z) < 1.
Example 8.3.8 An example such that partial transitivity ; partial quasi-

transitivity for = (1) or = (3) can be found in Example 5.1.12.
Definition 8.3.9 Let fe be a fuzzy aggregation rule. Then fe is said to be in-

dependent of irrelevant alternatives IIA4 if x, y X, (h1 , 1 i,
. . . , hn , n i), (h01 , 01 i, . . . , h0n , 0n i) T n [Cosupp(i e{x,y} ) =
Cosupp(0i e{x,y} ) i N Cosupp(fe( )e{x,y} ) = Cosupp(fe(0 )e{x,y} )]
and Supp(i e{x,y} ) = Supp(0i e{x,y} ) i N Supp(fe( )e{x,y} ) =
Supp(fe(0 )e{x,y} )].
Proposition 8.3.10 Let h , i, h0 , 0 i FR. Let x, y X. Then

(1) Cosupp( e{x,y} ) = Cosupp(0 e{x,y} ) if and only if u, v {x, y},
(u, v) < 1 0 (u, v) < 1;
(2) Supp( e{x,y} ) = Supp(0 e{x,y} ) if and only if (u, v {x, y}, (u, v)
> 0 0 (u, v) > 0.
Proof. (1) Suppose Cosupp( e{x,y} ) = Cosupp(0 e{x,y} ). Let u, v

{x, y}. Then (u, v) < 1 (u, v) Cosupp( e{x,y} ) (u, v) Cosupp(
0 e{x,y} ) 0 (u, v) < 1. Conversely, suppose u, v {x, y}, (u, v) < 1
0 (u, v) < 1. Then (u, v) Cosupp( e{x,y} ) (u, v) < 1 0 (u, v) <
1 Cosupp(0 e{x,y} ).
(2) Suppose Supp( e{x,y} ) = Supp(0 e{x,y} ). Let u, v {x, y}. Then
(u, v) > 0 (u, v) Supp( e{x,y} ) (u, v) Supp(0 e{x,y} )
0 (u, v) > 0. Conversely, suppose u, v {x, y}, (u, v) > 0 0 (u, v) >
0. Then (u, v) Supp( e{x,y} ) (u, v) > 0 0 (u, v) > 0
Supp(0 e{x,y} ).
For = (2) and x, y X, (x, y) > 0 if and only if (y, x) < 1 and
(x, y) < 1 if and only if (y, x) > 0. The strict preferences (1) , (3) , (4) ,
and (5) are regular. These properties play a key role in the proof of the next
result.
i i
i i
i i
K23798 2015/2/2
i i
Lemma 8.3.11 Let be a fuzzy subset of N. Let fe be a partially quasi-

transitive fuzzy preference aggregation rule that is weakly Paretian and either
independent of irrelevant alternatives IIA2 with strict preferences of type (0)
or independent of irrelevant alternatives IIA3 with regular strict preferences
or is independent of irrelevant alternatives IIA4 with strict preferences of type
(2) . If is semidecisive for x against y w.r.t , then (v, w) X X, is
decisive for v against w.
Proof. The proof for can be found in Lemma 5.1.15. Suppose is
semidecisive for x against y. Let be any fuzzy preference profile such that
i (x, z) < 1 i Supp(), where z / {x, y}. Let 0 be a fuzzy preference
profile such that
0i (x, z) = i (x, z) and 0i (z, x) = i (z, x) i N,

0 0
i (x, y) > 0 i Supp() and j (y, x) > 0 j N \Supp(),
0
i (y, z) > 0 i N.
0
Since i (x, z) < 1 i Supp(), vi (x, z) < 1 i Supp() by the
0 0
definition of . Since xD y, (x, y) < 1. Since fe is weakly Paretian,
e
0 (y, z) < 1. Since fe is partially quasi-transitive, 0 (x, z) < 1. For IIA3
and regular, 0 (x, z) < 0 (z, x). Since vi e{x,z} = 0vi e{x,z} i N
and fe is IIA3, e{x,z} 0 e{x,z} . Thus (x, z) < (z, x). (For strict
preferences of type (0) and the crisp case, we write: Hence 0 (x, z) < 1
and 0 (z, x) = 1. Since Cosupp(vi e{x,z} ) = Cosupp(0vi e{x,z} ) i N
and fe is IIA2, Cosupp( e{x,z} ) = Cosupp(0 e{x,z} ). Thus (x, z) < 1
and (z, x) = 1.) (For strict preferences of type (2) , we write: Hence
0 (z, x) > 0. Since Supp(ii e{x,z} ) = Supp(0i e{x,z} ) i N and fe is IIA4,
Supp( e{x,z} ) = Supp(0 e{x,z} ). Thus (z, x) > 0.) Hence (x, z) < 1.
Since was arbitrary, xD z. Thus since z was arbitrary in X\{x, y},
z X\{x, y}, xD
e y xD z. (8.12)
Since is decisive for x against z implies is semidecisive for x against z,

interchanging y and z in the above argument implies is decisive for x against
y.
Let 00 be any fuzzy preference profile with vi
00
(y, z) < 1 i Supp() and
+
let be a fuzzy preference profile such that
+
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
IIA3 and regular, we have since + 00

vi e{y,z} = vi e{y,z} i N and f is IIA3,
e
00 00
e{y,z} e{y,z} . Since (y, z) < 1, (y, z) < +
+ + +
(z, y). Hence (y, z) <
+
(z, y). (For = (0) and the crisp case, we write: Since vi e{y,z} = 00vi e{y,z}
00
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)
Now decisive for y against z implies is semidecisive for y against z. Thus

by (8.12), is decisive for y against x. We have (v, w) X X,
e y xD v (by (4.1)) xD
xD e v vD w (by (8.13)) with v replacing y.
Let FRT = { FR | is partially transitive}.
Theorem 8.3.12 (Fuzzy Arrows Theorem) Let fe be a fuzzy aggregation rule

from FRnT into FR. Suppose strict preferences are of type (0) . Let fe be weakly
Paretian, partially transitive, and independent of irrelevant alternatives IIA2.
Then fe is dictatorial.
Proof. The proof for can be found in Theorem 5.1.16. Since fe is par-
tially transitive, fe is partially quasi-transitive. 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 from Lemma 8.3.11 because is decisive for x against y for all
x, y X. Hence FRn (X), x, y X, i (x, y) < 1 implies (x, y) < 1.
namely = 1N . For all (u, v) X X, let m(u, v) = {m(u, v) | (u, v) X 2 }.
Without loss of generality suppose is semidecisive for x against y with
|Supp()| = m. If m = 1, the proof is complete. Suppose > 1. Let i
Supp(). Consider any fuzzy profile FRn such that
i (x, y) < 1, i (y, z) < 1, i (x, z) < 1,

j Supp()\{i}, vj (z, x) < 1, j (x, y) < 1, j (z, y) < 1,
k / Supp(), k (y, z) < 1, vk (z, x) < 1, k (y, x) < 1.
Since is semidecisive for x against y and j (x, y) < 1 j Supp(),

(x, y) < 1. Since |Supp()| = m, it is not the case that (z, y) < 1 other-
wise 0 is semidecisive for z against y, where 0 (j) = (j) j Supp()\{i}
and 0 (i) = 0. (Suppose (z, y) < 1. Supp(i e{x,y} ) = Supp(0vi e{x,y} )
i i
i i
i i
K23798 2015/2/2
i i
i N, 0 FRn (X) Supp(fe( e{x,y} ) = Supp(fe(0 e{x,y} ) 0

FRn (X) by IIA2 (z, y) Supp(fe(0 ), (y, z) / Supp(fe(0 ) (z, y)
0 0 n
Supp(fe( )) FR (X) since this is the case for fe( ) and j (z, y) < 1
and j (y, z) < 1 otherwise.) However, this contradicts the minimality of
m since |Supp(0 )| = m 1. Thus fe( )(y, z) < 1 since fe( ) is complete.
(Actually since (x, y) = 1, (y, z) < 1.) Since fe is partially transitive,
fe( )(x, z) < 1. (Actually (x, z) < 1 by partial quasi-transitivity.) By IIA2,
is semidecisive for x against z, where (i) = (i) and (j) = 0 for
j N \{i}. However, this contradicts the fact that m > 1.
Definition 8.3.13 Let h , i be an intuitionistic fuzzy relation on X. Then

h , i is said to be weakly transitive if x, y, z X, (1) (x, y) (y, x)
and (y, z) (z, y) implies (x, z) (z, x) and (2) (x, y) (y, x)
and (y, z) (z, y) implies (x, z) (z, x). A fuzzy aggregation rule fe
is said to be weakly transitive if fe() is weakly transitive for all FRn .
Example 8.3.14 Examples such that is max-min transitive, but is not

weakly transitive; is weakly transitive, but not max-min transitive; is weakly
transitive, but not transitive in the crisp sense (and not complete) can be found
in Example 5.1.18.
Remark 8.3.15 Suppose h , i is exact. If h , i is weakly transitive and

complete, then h , i is transitive in the crisp sense.
Proof. Suppose (x, y) = 1 and (y, x) = 1. Since is weakly transitive,
(x, z) (z, x). Since is complete, (x, z) = 1. Suppose (x, y) = 0
and (y, x) = 0. Since is weakly transitive, (x, z) (z, x). Since
is complete, (x, z) = 0.
Proposition 8.3.16 Let h , i be an intuitionistic fuzzy relation on X.

Then the following properties are equivalent:
(1) h , i is weakly transitive.
(2) For all x, y, z X, (i) (x, y) (y, x) and (y, z) (z, y)
with strict inequality holding at least once, then (x, z) > (z, x) and (ii)
(x, y) (y, x) and (y, z) (z, y) with strict inequality holding at
least once, then (x, z) < (z, x)
Proof. The proof for can be found in Proposition 5.1.20.
Suppose (1) holds. Assume that (x, y) (y, x) and (y, z) < (z, y).
Suppose (z, x) (x, z). Then (z, y) (y, z) by (1), a contradiction.
Thus (x, z) < (z, x). A similar argument shows that (x, y) < (y, x)
and (y, z) (z, y) implies (x, z) < (z, x).
Suppose (2) holds. Let x, y, z X. Suppose (x, y) (y, x) and
(y, z) (z, y). Suppose (z, x) < (x, z). Then by (2), (z, x) <
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).
Corollary 8.3.17 Let h , i be a fuzzy relation on X. Suppose strict pref-

erences are regular. If h , i is weakly transitive, then h , i is partially
quasi-transitive.
Let FCR denote the set of all reflexive and complete intuitionistic fuzzy
preference relations that are consistent.
Theorem 8.3.18 (Fuzzy Arrows Theorem) Let fe : FCRn FR be a fuzzy

aggregation rule. Suppose strict preferences are regular. Let fe be weakly tran-
sitive, weakly Paretian, and independent of irrelevant alternatives IIA3. Then
fe is dictatorial.
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.
(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,
k / Supp(), k (y, z) > 0, k (z, x) > 0, k (y, x) > 0.
i (x, y) < 1, i (y, z) < 1, i (x, z) < 1,

j Supp()\{i}, j (z, x) < 1, j (x, y) < 1, j (z, y) < 1,
k / Supp(), k (y, z) < 1, k (z, x) < 1, k (y, x) < 1.
Since is semidecisive for x against y, i (x, y) > 0 i Supp(), and
j (y, x) > 0 j
/ Supp(), we have that (x, y) > 0. Since |Supp()| = m,
it is not the case that (z, y) > 0, otherwise 0 is semidecisive for z against
y, where 0 (j) = (j) j Supp()\{i} and 0 (i) = 0. (Suppose (z, y) >
0. Then i e{z,y} 0i e{z,y} ) i N, 0 FRn (X) fe( )e{z,y}
i i
i i
i i
K23798 2015/2/2
i i
fe(0 )e{z,y} 0 F Rn (X) by independence of irrelevant alternatives IIA3

fe(0 )(z, y) > fe(0 )(y, z) since fe( )(z, y) > fe( )(y, z). Thus 0 (z, y) >
0.) However, this contradicts the minimality of m since |Supp(0 )| = m 1.
Since (z, y) = 0, fe( )(y, z) fe( )(z, y). Since fe( )(x, y) > fe( )(y, x),
fe( )(x, z) > fe( )(z, x) by consistency. Hence (x, z) > 0. By independence
of irrelevant alternatives IIA3, is semidecisive for x against z, where (i) =
(i) and (j) = 0 for j N \{i}. However, this contradicts the fact that
m > 1.
Since is semidecisive for x against y, i (x, y) < 1 i Supp(), and
j (y, x) < 1 j / Supp(), we have that (x, y) < 1. Since |Supp()| = m,
it is not the case that (z, y) < 1, otherwise 0 is semidecisive for z against
y, where 0 (j) = (j) j Supp()\{i} and 0 (i) = 0. (Suppose (z, y) <
1. Then i e{z,y} 0i e{z,y} ) i N, 0 FRn (X) fe( )e{z,y}
fe(0 )e{z,y} 0 FRn (X) by independence of irrelevant alternatives IIA3
fe(0 )(z, y) < fe(0 )(y, z) since fe( )(z, y) < fe( )(y, z). Thus 0 (z, y) <
1.) However, this contradicts the minimality of m since |Supp(0 )| = m 1.
Since (z, y) = 1, fe( )(y, z) fe( )(z, y). Since fe( )(x, y) < fe( )(y, x),
fe( )(x, z) < fe( )(z, x) by consistency. Hence (x, z) < 1. By independence
of irrelevant alternatives IIA3, is semidecisive for x against z, where (i) =
(i) and (j) = 0 for j N \{i}. However, this contradicts the fact that
m > 1.
8.4 Representation Rules, Veto Players, and

Oligarchies
In the previous section, we demonstrated that fuzzy intuitionistic versions
of Arrows Theorem hold under a certain set of definitions and various IIA
conditions. In this section, we lay out some concepts that give an application
of the fuzzy version of Arrows Theorem.
Recall that by , we mean h , i = (h1 , 1 i, . . . , hn , n 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.
Let g be a representation rule for T. Then = h , i T n ,

Supp(g()) = {i0 } for some i0 N. By the notation, Supp(g()) , we mean
i0 and by Supp(g()) , we mean i0 .
Definition 8.4.2 Let g be a representation rule for T. Then

(1) g is called dictatorial for T if and only if , 0 T n , Supp(g()) =
Supp(g(0 ));
i i
i i
i i
K23798 2015/2/2
i i
8.4. Representation Rules, Veto Players, and Oligarchies 259

T , x, y X, Supp(i e{x,y} ) = Supp(0i e{x,y} ) i N implies
n
Supp(Supp(g()) e{x,y} ) = Supp(Supp(g(0 ) e{x,y} )
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} ).
T , x, y X, i e{x,y} 0i e{x,y} i N implies Supp(g()) e{x,y}
n
Supp(g(0 ) e{x,y} and i e{x,y} 0i e{x,y} i N implies Supp(g()) e{x,y}

Supp(g(0 ) e{x,y} .
It follows that Cosupp(g( )) = {j} for some j N. By Cosupp(g( )) ,
we mean j .
Theorem 8.4.3 Let strict preferences be of type (0) . Let g be a representa-

tion rule for FRnT . Then g is independent of irrelevant alternatives IIA2 if
and only if g is dictatorial.
Proof. The proof for can be found in Theorem 5.2.3. Let fe be a
fuzzy aggregation rule determined by g. Let FRn . Then fe( ) =
Cosupp(g( )) . Since i is partially transitive i N, Cosupp(g( )) is par-
tially transitive. Thus fe( ) is partially transitive. Hence fe is partially transi-
tive, Let x, y X. Suppose i (x, y) < 1 i N. Then Cosupp(g( )) (x, y) <
1, but Cosupp(g( )) (x, y) = fe(x, y). Thus fe is weakly Paretian. Since
fe( ) = Cosupp(g( )) and fe(0 ) = 0Cosupp(g(0 )) , fe is IIA2 if and only

if g is. Thus if g is IIA2, the fe is dictatorial by Theorem 8.3.12. Conversely,

if g is dictatorial, then it follows easily that g is IIA2.
Theorem 8.4.4 Let strict preferences be regular. Let g be a representation

rule for FCR. Then g is independent of irrelevant alternatives IIA3 if and
only if g is dictatorial.
Proof. Let fe be a fuzzy aggregation rule determined by g. Let FCRn .
Then fe() = hSupp(g()) , Supp(g()) i. Since i and i are consistent i
N, hSupp(g()) , Supp(g()) i = hi0 , i0 i is consistent. Thus fe() is con-
sistent. Hence fe is consistent. Let x, y X. Suppose i (x, y) > 0 and
i (x, y) < 1 i N. Then Supp(g()) (x, y) > 0 and Supp(g()) (x, y) <
1.Thus i0 (x, y) > i0 (y, x) and i0 (x, y) < i0 (y, x). Hence (x, y) > 0
and (x, y) < 1. Thus fe is weakly Paretian. Since fe() = hSupp(g()) ,
Supp(g()) i and fe(0 ) = hSupp(g(0 )) , Supp(g(0 )) i, fe is independent of ir-
relevant alternatives IIA3 if and only if g is. Thus if g is independent of
irrelevant alternatives IIA3, then fe is dictatorial by Theorem 8.3.18. Thus g
i i
i i
i i
K23798 2015/2/2
i i
is dictatorial. Conversely, if g is dictatorial, then it follows easily that g is

independent of irrelevant alternatives IIA3.
Definition 8.4.5 Let x, y X. An individual i N is said to have a veto

for x against y, written xVi y, if for every FRn , i (x, y) > 0 implies not
(y, x) > 0 and i (x, y) < 1 implies not (y, x) < 1. An element i N is
said to have a veto if for all x, y X, i has a veto for x against y.
Definition 8.4.6 A fuzzy aggregation rule is called oligarchic if there exists

FP(N ) (called an oligarchy) such that
(1) every member of Supp() has a veto;
(2) is decisive.
Definition 8.4.7 Let x, y X. An individual i N is said to have a

semiveto for x against y, written xVei y, if FRn , i (x, y) > 0 and
j (y, x) > 0 j 6= i implies not (y, x) > 0 and i (x, y) < 1 and
j (y, x) < 1 j 6= i implies not (y, x) < 1.
Lemma 8.4.8 Let strict preferences be of type (0) . Suppose is partially

quasi-transitive. Then
(1) (x, y) > 0 and (y, z) > 0 implies (x, z) > 0,
(x, y) < 1 and (y, z) < 1 implies (x, z) < 1.
(2) (x, y) > 0 and (y, z) > 0 implies (x, z) > 0,
(x, y) < 1 and (y, z) < 1 implies (x, z) < 1.
Proof. (1) The case for can be found in Proposition 5.1.20 or [21,
Lemma 3.8, p. 378]. Suppose its not the case that (x, z) < 1. Since
is complete, (z, x) < 1. Since also (x, y) < 1, we have by partial quasi-
transitivity that (z, y) < 1. However, this is impossible since (y, z) < 1.
(2) Suppose its not the case that (x, z) < 1. Since is complete,
(z, x) < 1 and so (z, x) < 1. Since also (y, z) < 1, we have by par-
tial quasi-transitivity that (y, x) < 1. However, this is impossible since
(x, y) < 1.
Lemma 8.4.9 Let strict preferences be of type (0) . Let fe : FRnT FR be

a fuzzy aggregation 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. The proof for can be found in Lemma 5.2.9 or [21, Lemma 3.7,
p. 377]. Let be such that
i (x, y) < 1, i (x, z) < 1, i (y, z) < 1,

j 6= i, j (y, x) < 1, j (y, z) < 1.
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)
y. Let 00 be any fuzzy preference profile such that i00 (y, z) < 1. Let + be a
+ + +
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 x against z, + (x, z) < 1. Since fe is weakly Paretian,

+ (y, x) < 1. Since fe is partially quasi-transitive, + (y, z) < 1. Since only the
preferences for i are specified and Supp(+ 00
i e{y,z} ) = Supp(i e{y,z} ), IIA2
00
implies (y, z) < 1 and so i has a veto for y against z. Thus
z
/ {x, y}, xVei y yVi z. (8.15)
(8.14), i has a veto for y against x. We have (v, w) X X,
xVe y xVi v by (8.14) xVe v v Vei w by (8.15)
with v replacing y.
Theorem 8.4.10 Let strict preferences be of type (0) . If a fuzzy aggrega-

tion rule fe : FRnT FR is partially quasi-transitive, weakly Paretian, and
independent of irrelevant alternatives IIA2, then it is oligarchic.
Proof. The proof for can be found in Theorem 5.2.10. Since fe is weakly
Paretian, there exists FP(N ) such that is semidecisive, namely = 1N .
Let be such that Supp() is smallest for which is semidecisive. Let m =
|Supp()|. By Lemma 8.3.11, is decisive for all ordered pairs (u, v) X 2 . If
m = 1, then the proof is complete. Suppose m > 1. Let x, y, z X be distinct.
Let be a fuzzy preference profile such that for some i Supp(),
i (x, y) < 1, i (x, z) < 1, i (y, z) < 1,

j Supp()\{i}, j (z, x) < 1, j (z, y) < 1, j (x, y) < 1, (8.16)
k / Supp(), k (y, z) < 1, k (y, x) < 1, k (z, x) < 1.
i i
i i
i i
K23798 2015/2/2
i i

there is such a such that not (z, y) < 1 else Supp()\{i} would be
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 , satisfy-
ing (8.16), Supp(i e{x,y,z} ) = Supp(0i e{x,y,z} ), i = 1, . . . , n.) Hence for any
such , fe( )(y, z) < 1 by completeness. Since fe is partially quasi-transitive,
fe( )(x, z) < 1 and hence not (z, x) < 1. Thus i has a semiveto for x against
z. Hence i has a veto over all ordered pairs (u, v) X 2 by Lemma 8.4.9. Thus
fe is oligarchic since i was arbitrary.
Let FQR = {h , i F R | h , i is partially quasi-transitive}.
Lemma 8.4.11 Let strict preferences be regular. Let fe : FQRn F R be

a fuzzy aggregation rule which is partially quasi-transitive, weakly Paretian,
and independent of irrelevant alternatives IIA3. 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. The proof for can be found in Lemma 5.2.11. Let h , i
FQRn and z X\{x, y} be such that
i (x, y) > 0, i (x, z) > 0, i (y, z) > 0,

j 6= i, j (y, x) > 0, j (y, z) > 0.
i (x, y) < 1, i (x, z) < 1, i (y, z) < 1,

j 6= i, j (y, x) < 1, j (y, z) < 1.
Note that j e{x,z} and j e{x,z} for all j 6= i can be arbitrarily assigned. Since
i has a semiveto for x against y, (y, x) = 0 and (y, x) = 1. Thus (x, y)
(y, x) and (x, y) (y, x). Since fe is weakly Paretian, (y, z) > 0 and
(y, z) < 1. Now suppose (z, x) > 0 and (z, x) < 1. Since fe is partially
quasi-transitive by assumption, (y, z) > 0 and (z, x) > 0 imply (y, x) >
0 and (y, z) < 1 and (z, x) < 1 imply (y, x) < 1, a contradiction. Hence
(z, x) = 0 and (z, x) = 1. (We have (z, x) = 0 for this , but not an
arbitrary preference profile. Let h0 , 0 i FQRn be such that i 0
(x, z) > 0
0
and i (x, z) < 1. Because j e{x,z} for all j 6= i are arbitrarily assigned,
IIA3 implies 0 (z, x) = 0. Also, we have (z, x) = 1 for this , but not
an arbitrary preference profile. Because j e{x,z} for all j 6= i are arbitrarily
assigned, IIA3 implies 0 (z, x) = 1.). Thus we have
z X\{x, y}, xVei y xVi z. (8.17)
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.
+ + +
Since fe is partially quasi-transitive, + (z, x) > 0 and + (z, x) < 1, a contra-

diction. Thus + (z, y) = 0 and + (z, y) < 1. Since 00j e{y,z} and + j e{y,z} can
be arbitrarily assigned for all j 6= i so that + e
i {y,z} 00
e
i {y,z} and 00j e{y,z}
+ +
and j e{y,z} can be arbitrarily assigned for all j 6= i so that i e{y,z}
00i e{y,z} , IIA3 implies 00 (z, y) = 0 and 00 (z, y) = 1 and so i has a veto for y
against z. Thus
z X\{x, y}, xVei y yVi z. (8.18)
(8.17), i has veto for y against x. We have (v, w) X X, xVei y xVi v (by
(8.17)) xVei v vVi w (by (8.18)) with v replacing y.
Theorem 8.4.12 Suppose strict preferences are regular. If a fuzzy aggrega-

tion rule fe : FQRn FR is partially quasi-transitive, weakly Paretian, and
independent of irrelevant alternatives IIA3, then it is oligarchic.
Proof. The proof for can be found in Theorem 5.2.12. Since fe is weakly
Paretian, there exists FP(N ) such that is semidecisive, namely = 1N .
Let be such that Supp() is smallest for which is semidecisive. Let m =
|Supp()|. By Lemma 8.3.11, is decisive for all ordered pairs (u, v) X X.
If m = 1, then the proof is complete because a dictator is an oligarchy of size
1. Suppose m > 1. Let x, y, z X be distinct and let h , i FQRn be
such that for some i Supp(),
i (x, y) > 0, i (x, z) > 0, i (y, z) > 0,

j Supp()\{i}, j (z, x) > 0, j (z, y) > 0, j (x, y) > 0,
k / Supp(), k (y, z) > 0, k (y, x) > 0, k (z, x) > 0.
i (x, y) < 1, i (x, z) < 1, i (y, z) < 1,

j Supp()\{i}, j (z, x) < 1, j (z, y) < 1, j (x, y) < 1,
k / Supp(), k (y, z) < 1, k (y, x) < 1, k (z, x) < 1.
Since is decisive, (x, y) > 0 and so (x, y) > (y, x) because is
regular and (x, y) < 1 and so (x, y) < (y, x) because is regular. Since
|Supp()| = m, there is no h0 , 0 i FQRn such that 0 e{y,z} e{y,z} and
0 (z, y) > 0 and 0 e{y,z} e{y,z} and 0 (z, y) < 1 else (z, y) > 0 and
i i
i i
i i
K23798 2015/2/2
i i
(z, y) < 1 by IIA3 and so Supp()\{i} = Supp(0 ) FP(N ) would be such

that 0 is semidecisive and thus decisive by Lemma 8.3.11, a contradiction of
the minimality of m. Hence for any such h , i, (z, y) = 0 and (z, y) =
1, By the regularity of and , (y, z) (z, y) and (y, z) (z, y).
Now suppose (z, x) > 0 and (z, y) < 1. Since fe is partially quasi-
transitive, (z, x) > 0 and (x, y) > 0 imply (z, y) > 0; and (z, x) < 1
and (x, y) < 1 imply (z, y) < 1. Then by the regularity of and ,
(z, y) > (y, z) and (z, y) < (y, z). However, this is a contradiction.
Thus (z, x) = 0 and (z, x) = 1. Hence i has a veto for x against z. By
Lemma 8.4.11, i has a veto over all ordered pairs (u, v) X X. Because i
was an arbitrary member of Supp(), fe is oligarchic.
In this section, we extended some known Arrowian results involving fuzzy
set theory to results involving intuitionistic fuzzy sets. We pointed out how
the use of an involutive fuzzy complement can be used to obtain the results.
A future research project might be to consider more aggressively the use of an
involutive fuzzy complement in obtaining dual results for by using known
results for .
8.5 Fuzzy Preference and Arrowian Results

In this section, we use intuitionistic fuzzy preference relations to examine
Arrowian results in the intuitionistic case as developed by Nana and Fono
[24]. The concepts of pos-transitivity and the negative transitivity of a strict
component of a relation are important properties. They allow for the estab-
lishment of Arrowian results for crisp and fuzzy preferences, Arrow [2] and
Fono and Andjiga [13]. Motivated by Fono et al. [16], a factorization of an
intuitionistic fuzzy relation (IFR) is established in Chapter 3 and previous
sections in this chapter. A unique indifference component and a family of reg-
ular strict components was obtained. Necessary and sufficient conditions on a
(T, S)-transitive IFR were obtained such that each of its strict components is
pos-transitive and negative transitive. The purpose of this section is to char-
acterize, by means of the (min, max)-transitivity, the pos-transitivity and the
negative transitivity of a regular strict component obtained in [16] and also
to establish some new Arrowian type results with (T, S)-transitive IFRs.
In Chapter 3, we presented results on factorization of an IFR = h , i
into a unique indifference and a family of strict components. We present here
a result from [16] that provides the characterization of two properties of a
strict component of a (T, S)-transitive IFR which stipulates that a strict

component of is pos-transitive if and only if satisfies C1 or C1 and also

a strict component of R is negative transitive if and only if satisfies (C1
R R
and C2 ) or (C1 and C2 ). We show that when is a (min, max)-transitive
IFR, i.e., (T, S) = (min, max)), a strict component of is pos-transitive and
a strict component of is negative transitive if and only if satisfies one of
i i
i i
i i
K23798 2015/2/2
i i
8.5. Fuzzy Preference and Arrowian Results 265
the new conditions T R or T R which as stated in [24] can be interpreted as

transitivity of indifference in the pairwise comparisons of degrees of preference.
We present some properties of an IFAR which are intuitionistic fuzzy versions
of the four concepts of a crisp aggregation rule, namely, Pareto condition (PC),
independence of irrelevant alternatives (IIA), decisive coalition, veto, oligarchy

and dictator. By using the conditions (C1 and C2 ) or (C1R and C2R ), an
intuitionistic fuzzy version of Arrows impossibility theorem is established.
Without these conditions, a non-dictatorial IFAR and an intuitionistic fuzzy
version of Gibbards oligarchy theorem is obtained.
In Sections 8.1-8.4 and Chapter 3, we showed how Arrowian results could
be obtained for intuitionistic fuzzy preference relations using a dual argument.
In the remainder of the chapter, such results can not be so easily obtained.
We next explain the logic. In previous sections, definitions were of the type
(p q) (r s), where p q was an implication for and r s was
an implication for . Many definitions in the remainder of this chapter are
of the type (p r) (q s). These definitions are not in general equivalent.
Definitions of the type (p q) (r s) imply those of the type (p r)
(q s), but not conversely. However, in many situations p q and r s are
equivalent due to the duality involved between and . In this case, the
two types of definitions (p q) (r s) and (p r) (q s) are equivalent.
Consequently, arguments are dual in nature when this occurs. However, not
all definitions in the remainder of the chapter are of this nature and so full
proofs are needed in those spots.
Let X be a set of alternatives with |X| 3. Let L = {(a1 , a2 ) [0, 1]2 ,
a1 + a2 1}. Define the relation L on L defined by (a1 , a2 ), (b1 , b2 ) L ,
(a1 , a2 ) L (b1 , b2 ) (a1 b1 and a2 b2 ). (L , L ) is a complete lattice.
0L = (0, 1) and 1L = (1, 0) are the units of L .
In the following, we recall some useful definitions and results on fuzzy
operators and intuitionistic fuzzy operators (see Cornelis et al. [9], Fono et al.
[15, 13, 16], Fono and Salles [17], Fotso and Fono [18], Klement et al. [20] and
Njanpong [255]).
Throughout the remainder of the chapter, we assume that T is a left-
continuous t-norm and S is right-continuous t-conorm.
The fuzzy -implication IT associated with T is a binary operation on
[0, 1] defined by a, b [0, 1], IT (a, b) = {t [0, 1], T (a, t) b}. The fuzzy
coimplicator JS associated with S is a binary operation on [0, 1] defined by
a, b [0, 1], JS (a, b) = {t [0, 1], b S(a, t)}.
Some usual examples are: T = TM = min; S = SM = max; a, b [0; 1],

1 if a b
ITM (a, b) = Imin (a, b) = and
b if a > b

b if a < b
JSM (a, b) = Jmax (a, b) = .
0 if a b
i i
i i
i i
K23798 2015/2/2
i i
Definition 8.5.1 (Cornelis et al. [9]) A t-representable intuitionistic

fuzzy t-norm T (t-conorm J ) is an increasing, commutative, associa-
tive binary operation on L such that there exists a fuzzy t-norm T and
a fuzzy t-conorm S (a fuzzy t-conorm S and a fuzzy t-norm T ) satisfying
a = (a1 , a2 ), b = (b1 , b2 ) L , T (a, b) = (T (a1 , b1 ), S(a2 , b2 )) (J (a, b)) =
(S(a1 , b1 ), T (a2 , b2 )). T and S (S and T ) are called the representants of T
(J ).
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 ).
Definition 8.5.2 J = (S, T ) satisfies condition G if (a1 , a2 ), (b1 , b2 )

L , a1 > b1 , a2 < b2 , and a1 + a2 = b1 + b2 imply IT (b2 , a2 ) + JS (b1 , a1 ) 1.
It follows that J = (max, min) satisfies condition G. See also Fono et
al. [16] for some examples of t-representable intuitionistic fuzzy t-conorms
satisfying condition G and those which violate it. We ask the reader to provide
the proofs in the Exercises.
We next recall some basic notions on intuitionistic fuzzy binary relations
(see Atanassov [3], Bustince and Burillo [8], Dimitrov [11], Fono et al. [16]
and Njanpong [24]).
Definition 8.5.3 [16]

(1) An intuitionistic fuzzy set (IFS) in X is given {hx, (x), (x)i |
x X}, where the functions : X [0, 1] and : X [0, 1] satisfy the
condition x X, (x) + (x) 1.
(2) An intuitionistic fuzzy relation (IFR) on X is a function :
X X L where x, y X (x, y) = ( (x, y), (x, y)).
(3) Let h, i and h0 , 0 i be two IFSs and J = (S, T ). The union
h, i (S,T ) h0 , 0 i associated with J is an IFS on X defined by x
X, ((S,T ) 0 )(x) = S((x), 0 (x)) and ((S,T ) 0 )(x) = T ((x), 0 (x)).
(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
which x is at least as good as y and (x, y) is the degree to which x is not

at least as good as y.
Definition 8.5.4 Let be an IFR and T = (T, S).

(1) is strongly complete if x, y X, (x, y) + (y, x) 1 and
(x, y) + (y, x) 1.
(2) is -symmetric if x, y X, (x, y) = (y, x), i.e.,
(x, y) + (x, y) = (y, x) + (y, x).
(3) is perfect antisymmetric if (x, y) X X, x 6= y,
(x, y) > 0 or [ (x, y) = 0 and (x, y) < 1]

(y, x) = 0 and (y, x) = 1.
(4) is simple if (x, y) X X,

(x, y) = (y, x) (x, y) = (y, x)

(x, y) = (y, x) (x, y) = (y, x)
.
(5) is T -transitive if x, y, z X, (x, z) T ( (x, y), (y, z)) and
(x, z) S( (x, y), (y, z))
If T = (min, max), we simply say that is transitive instead of (min, max)-
transitive.
Throughout the remainder of this chapter, we assume that is reflexive,
strongly complete and -symmetric.
In the next paragraph, we recall a factorization of an IFR into a unique
indifference I and a family of strict components.
Definition 8.5.5 [16] Let J = (S, T ) satisfy condition G, be an IFR and,

and be two IFRs. Then and are the indifference of and the strict
component of associated with J respectively if the following conditions are
satisfied:
(1) = (S,T ) ,
(2) is simple,
(3) is perfect antisymmetric,
(4) is symmetric.
Consider the following system:

(i) a + b 1,
(ii) S( (y, x), a) = (x, y), (8.19)
(iii) T ( (y, x), b) = (x, y).

i i
i i
i i
K23798 2015/2/2
i i
Proposition 8.5.6 [16] Let J = (S, T ) satisfy condition G, be an IFR

and, and be two IFRs. Then the following statements are equivalent:
(1) and are the indifference of and the strict component of associated
with J , respectively,
(2) (i) x, y X, (x, y) = (y, x) = min( (x, y), (y, x)) and
(x, y) = (y, x) = max( (x, y), (y, x)).
(ii) x, y X, (cxy , gxy ) L such that cxy > 0, cxy is a solution
of (ii) of (8.19)
gxy < 1, gxy is a solution of (iii) of (8.19) and

0 if (x, y) (y, x)
(x, y) = and
cxy otherwise

1 if (x, y) (y, x)
(x, y) = .
gxy otherwise
It was established in [16, p. 14] that each strict component obtained

in the previous result satisfies the following property: is regular, that is,
x, y X,
(x, y) (y, x) (x, y) = 0
.
(x, y) (y, x) (x, y) = 1
Definition 8.5.7 [16] Let be an IFR and be a regular strict component

of .
(1) is pos-transitive if x, y, z X,

(x, y) > 0 (y, z) > 0 (x, z) > 0
and .
(x, y) < 1 (y, z) < 1 (x, z) < 1
(2) is negative transitive if x, y, z X,

(x, y) = 0 (y, z) = 0 (x, z) = 0
and .
(x, y) = 1 (y, z) = 1 (x, z) = 1
The pos-transitivity of says that for the strict preference relation

associated with and for all x, y, z X, if x is strictly preferred to y and
y is strictly preferred to z, then x is strictly preferred to z. Considering the
contrapositive, we see that the negative transitivity of says that if x is
strictly preferred to z, then either x is strictly preferred to y or y is strictly
preferred to z.
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
(2) is negative transitive if x, y, z X, (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)
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,
2 (x, y, z) = S( (x, y), (y, z)),

2 (x, y, z) = T ( (x, y), (y, z)),
3 (x, y, z) = JS ( (y, z), (y, x)) JS ( (x, y), (z, y)),
3 (x, y, z) = {IT ( (x, y), (z, y)), 1 JS ( (x, y), (z, y)),
IT ( (y, z), (y, x)), 1 JS ( (y, z), (y, x))}.
We next consider conditions important in the study of pos-transitivity and

negative transitivity.
Definition 8.5.9 [16] Let be an IFR.

(1) satisfies condition C1 if x, y, z X,

(y, x) < (x, y)

(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)]

(y, x) > (x, y)

(z, y) > (y, z)
(x, z) [3 (x, y, z), 2 (x, y, z)]
(z, x) > (x, z) .
(z, x) [3 (x, y, z), 2 (x, y, z)]
i i
i i
i i
K23798 2015/2/2
i i


(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)]

(y, x) = (x, y) (y, x) > (x, y)
or
(z, y) > (y, z) (z, y) = (y, z)
(x, z) [3 (x, y, z), 2 (x, y, z)]
(z, x) > (x, z) .
(z, x) [3 (x, y, z), 2 (x, y, z)]

Consider condition C1 . Assume that x is strictly preferred to y with
respect to and y is strictly preferred to z with respect to . Then in the
crisp case, we have is a crisp total pre-order and x is strictly preferred to
z with respect to . However, in the intuitionistic fuzzy case, this is not true
when the two degrees of comparisons of x and z belong to the particular
subset [2 (x, y, z), 3 (x, y, z)] of [0, 1], i.e., there is not a comparison between

the two degrees (x, z) and (z, x). Therefore, condition C1 alleviates this
difficulty.
In the following, we recall general results on characterizations of the pos-
transitivity and the negative transitivity of a given regular strict component
of an IFR.
Theorem 8.5.10 [16] Let T = (T, S), be a T -transitive IFR, and be a

regular strict component of . Then

(1) is pos-transitive if and only if satisfies C1 or C1 .

(2) is negative transitive if and only if satisfies (C1 and C2 ) or (C1

and C2 ).

In [16] the conditions C1 , C1 , C2 , and C2 were used to obtain char-
acterizations of the pos-transitivity and the negative transitivity of a regular
strict component of an IFR. These results are valid for each t-representable
intuitionistic fuzzy t-norm T = (T, S). For the sake of simplicity, we examine
these conditions in the case where T = TM = (min, max).
To obtain the particular cases of the previous characterizations when T =
(min, max), we need the two following conditions:
Definition 8.5.11 Let be an IFR on X.

(1) satisfies condition T if
x, y, z X, (x, y) = (y, x) = (y, z) = (z, y)

implies (x, z) = (z, x).
i i
i i
i i
K23798 2015/2/2
i i
(2) satisfies condition T if
x, y, z X, (x, y) = (y, x) = (y, z) = (z, y)

implies (x, z) = (z, x).
These two conditions can be viewed as a type of transitivity of indifference

in the pairwise comparisons of degrees of preference. Since is -symmetric,
T and T are equivalent.
The following result shows that (min, max)-transitive IFRs satisfy condi-

tions C1 and C1 . Furthermore, when is (min, max)-transitive, then con-

ditions C2 and C2 become T and T , respectively.
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
[ (z, y) < (y, x) = (x, y) < (y, z)] or

[ (z, y) < (y, z) (x, y) = (y, x)] or
[ (z, y) = (y, z) (y, x) < (x, y)] or
[ (y, x) = (x, y) (z, y) < (y, z)] or
[ (y, x) < (x, y) (z, y) = (y, z)] or
[ (y, x) < (z, y) = (y, z) < (x, y)].
i i
i i
i i
K23798 2015/2/2
i i
We consider the following three cases:

(1) If (z, y) < (y, x) = (x, y) < (y, z) or (z, y) < (y, z)
(x, y) = (y, x) or (y, x) < (x, y) (z, y) = (y, z) or (y, x) <
(z, y) = (y, z) < (x, y), then interval [2 (x, y, z), 3 (x, y, z)] is empty
since 2 (x, y, z) > 3 (x, y, z). Hence the result.
(2) If (z, y) = (y, z) (y, x) < (x, y), then 2 (x, y, z) = (y, z)
and 3 (x, y, z) = (z, y). The assertion (x, z), (z, x) [2 (x, y, z), 3 (x,
y, z)] becomes (x, z) = (z, x) = (z, y) = (y, z). These equalities and
the inequality (y, x) < (x, y) contradict condition T . Thus this case is
not possible.
(3) If (x, y) = (y, x) (z, y) < (y, z), then 2 (x, y, z) = (x, y)
and 3 (x, y, z) = (y, x). Conjunctions (x, z), (z, x) [2 (x, y, z), 3 (x,
y, z)] become (x, z) = (z, x) = (x, y) = (y, x). These equalities and
the inequality (z, y) < (y, z) contradict condition T R . Thus this case is
not possible.

(2ib) Suppose that satisfies C2 . We show that satisfies T . Assume
to the contrary that x, y, z X such that (x, y) = (y, x) = (y, z) =
(z, y) and (x, z) 6= (z, x). We show that it is impossible.
If (x, z) > (z, x), then the transitivity of implies (z, x)
min( (z, y), (y, x)). Then (x, y) = (y, x) = (y, z) = (z, y)
(z, x) < (x, z).
Then
3 (x, z, y) = min(Imin ( (z, y), (z, x)), 1 Jmax ( (z, y), (z, x)),
Imin ( (x, z), (y, z)), 1 Jmax ( (x, z), (y, z)))
= min(1, (y, z)) = (y, z) = (z, y)
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
(x, z) min( (x, y), (y, z)).
Thus
(x, y) = (y, x) = (y, z) = (z, y) (x, z) < (z, x).
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.
Proposition 8.5.13 (Nana and Fono [24]) Let be a (min, max)-transitive

IFR and be a regular strict component of . Then the following statements
hold:
(1) is pos-transitive.
(2) The following two statements are equivalent:
(i) satisfies T or T .
(ii) is negative transitive.
Proof. (1) Assume that is (min, max)-transitive. We show that is pos-

transitive. Since is (min, max)-transitive, the first result of Lemma 8.5.12

implies that satisfies conditions C1 and C1 . Using the first result of The-
orem 8.5.10, it follows that is pos-transitive.
(2) Suppose (i) holds. Assume that satisfies condition T . We show
that is negative transitive. Since is (min, max)-transitive, the first result of

Lemma 8.5.12 implies that satisfies condition C1 . Since satisfies condition
T and is (min, max)-transitive, the second result of Lemma 8.5.12 implies

that satisfies condition C2 . Since satisfies conditions C1 and C2 , the
second result of Theorem 8.5.10 implies that is negative transitive.
Assume that satisfies condition T . By the -symmetry of , it follows
that is negative transitive by an analogous prove.
Suppose that (ii) holds. Assume that is negative transitive. We show that
satisfies condition T or T . Since is (min, max)-transitive, it suffices

to show that satisfies condition C2 or C2 by the second result of Lemma
8.5.12. Since is negative transitive, the second result of Theorem 8.5.10

implies that satisfies condition C2 or C2 .
i i
i i
i i
K23798 2015/2/2
i i
Any strict component of a (min, max)-transitive is pos-transitive and only

the strict components of transitive IFRs satisfying conditions T R or T R are
negative transitive.
Consequently, we introduce the following notations for a t-norm T .
(1) W T be the set of reflexive, complete, -symmetric and T -transitive
IFRs.
(2) W1T be the set of reflexive, complete, -symmetric and T -transitive

IFRs satisfying condition C1 or condition C1 .
(3) W2T be the set of reflexive, complete, -symmetric and T -transitive

IFRs satisfying conditions (C1 and C2 ) or (C1 and C2 ).
(4) CO be the set of crisp total pre-orders (reflexive, complete and transi-
tive crisp binary relations)
We have CO W2T W1T W T .
In the case where T = (min, max), we have by Lemma 8.5.12(1) that
W T and W1T become the set W of reflexive, strongly complete, -symmetric
and transitive IFRs and by Lemma 8.5.12(2) W2T becomes the set W2 of
elements of W satisfying T or T R . Hence the previous inclusions become
CO W2 W.
G is the set of admissible IFRs and G W T in the general case. In the
particular case, G W.
8.6 Intuitionistic Arrows Theorem and Gib-

bards Oligarchy Theorem
In this section, we use Theorem 8.5.10 and Proposition 8.5.13 to establish an
intuitionistic fuzzy version of the crisp Arrow impossibility theorem and an
intuitionistic fuzzy version of the crisp Gibbards oligarchy theorem.
Let N = {1, 2, . . . , n} with n 2 be a finite set of individuals or voters.
A coalition C is a non empty subset of N and P (N ) is the set of coalitions
of N. We recall some notation: i = hi , i i, i = hi , i i, i N, and
= (h1 , 1 i, . . . , hn , n i). Let Gn = { | i G, i N }. Given C
P (N ) and x, y X, C (x, y) > 0 means that i C, i (x, y) > 0 and,
C (x, y) < 1 means i C, i (x, y) < 1.
A function fe : Gn G is called an intuitionistic fuzzy aggregation rule
(IFAR). For Gn , fe() = = ( , ) G is a social preference and
= ( , ) is its strict component.
We introduce intuitionistic fuzzy versions of some usual properties of a
crisp aggregation rule.
Definition 8.6.1 Let fe be an IFAR, u, v X, i N and C P (N ).
i i
i i
i i
K23798 2015/2/2
i i
8.6. Intuitionistic Arrows Theorem and Gibbards Oligarchy Theorem 275
(1) fe satisfies intuitionistic fuzzy Pareto condition (IFPC) if

Gn , x, y X,

i (x, y) > 0 (x, y) > 0
i N, .
i (x, y) < 1 (x, y) < 1
(2) fe satisfies independence of irrelevant alternatives (IIA) if

, 0 Gn ,
0
i (x, y) > 0 i (x, y) > 0

0

i (x, y) < 1 i (x, y) < 1

x, y X,
i N 0

i (y, x) > 0 i (y, x) > 0
0
i (y, x) < 1 i (y, x) < 1

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
(4) i is almost semi-decisive over (u, v) if Gn ,

i (u, v) > 0 N \{i} (v, u) > 0 (v, u) = 0
and .
i (u, v) < 1 N \{i} (v, u) < 1 (v, u) = 1
(5) C is almost decisive on (u, v) if Gn ,

C (u, v) > 0 N \C (v, u) > 0 (u, v) > 0
and .
C (u, v) < 1 N \C (v, u) < 1 (u, v) < 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
(7) i has a veto if Gn , x, y X,

i (x, y) > 0 (y, x) = 0
.
i (x, y) < 1 (y, x) = 1
(8) f is dictatorial if there exists i0 N such that i0 is a dictator.

(9) f is an oligarchy if there exists a unique decisive coalition S of N
such that i S, i has a veto.
i i
i i
i i
K23798 2015/2/2
i i
We next establish an analogue of the crisp Arrow impossibility theorem

for (T, S)-transitive IFARs.
Theorem 8.6.2 (Nana and Fono [24]) Let T = (T, S) be an intuitionistic

fuzzy t-norm and fe : (W1T )n W2T be an IFAR. If fe satisfies conditions IIA
and IFPC, then fe is dictatorial.
To establish this theorem, we need the two following lemmas. As in crisp
case and fuzzy case, we first establish an intuitionistic fuzzy version of the
crisp field expansion lemma.
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
and j0 and j00 coincide on {u, v}.

We show first that 00
(u, v) > 0
(8.24)
00 (u, v) < 1
By the first system of (8.22) and the first system of (8.23), we have that
00
N (u, x) > 0
00
N (u, x) < 1
i i
i i
i i
K23798 2015/2/2
i i
Since fe satisfies IFPC, we have

00
(u, x) > 0
(8.25)
00 (u, x) < 1
Analogously, by the second system of (8.22) and the second system of

(8.23), we have that 00
N (y, v) > 0
00
N (y, v) < 1
Since fe satisfies IFPC,
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
Analogously, by the pos-transitivity of 00 and (8.26) and (8.28), we have

(8.24).
We now show (8.21). The last system of (8.22) and the hypothesis yield
00
C (u, v) = 0 (u, v) > 0
00 0
C (u, v) = C (u, v) < 1
Since is perfect antisymmetric,

00 0
C (v, u) = C (v, u) = 0
00 0
C (v, u) = C (v, u) = 1
0 00
Since N \C and N \C coincide on {u, v}, we have
(
00 0
N \C (v, u) = N \C (v, u)
00 0
N \C (v, u) = N \C (v, u)
By the two previous systems, we have that

00 0
N (v, u) = N (v, u)
00 0 (8.29)
N (v, u) = N (v, u)
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
Since (W1T )n , the first result of Theorem 8.5.10 implies that i N, i is

pos-transitive. Hence, (8.32), (8.33) and (8.34) yield

D1 (x, z) > 0 D1 (z, y) > 0 N \D (y, x) > 0
, and . (8.35)
D1 (x, z) < 1 D1 (z, y) < 1 N \D (y, 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
We consider two cases:

(1) If (x, z) > 0 and (x, z) < 1, then this system, the first system of
(8.35), (8.36), the fact that D1 is almost decisive on (x, z) and Lemma 8.6.3
imply that D1 is decisive. Thus K = D1 .
(2) If (z, y) > 0 and (z, y) < 1, then this system, the second system
of (8.35), the fact that D2 is almost decisive on (y, z) and Lemma 8.6.3 imply
that D2 is decisive. Thus K = D2 .
We now prove the theorem.
Proof of Theorem 8.6.2 We find i0 N such that (W1T )n , x, y
X,
i0 (x, y) > 0 (x, y) > 0
.
i0 (x, y) < 1 (x, y) < 1
Since fe satisfies IFPC, N is decisive. Since |N | 2, Lemma 8.6.4 (the group

contraction lemma) implies that N admits a decisive sub coalition denoted
N1 . We consider two cases:
(1) If |N1 | = 1 and N1 is decisive, then i0 is the unique element of N1 .
(2) If |N1 | 2, then we can repeat the previous procedure to obtain a
dictator i0 by applying the group contraction lemma.
We have from Proposition 8.5.13 and Theorem 8.6.2, the intuitionistic
fuzzy Arrow Theorem when T = (min, max).
Corollary 8.6.5 Let fe : W n W2 be an IFAR. If fe satisfies conditions IIA

and IFPC, then fe is dictatorial.
We now show that in the case where T = (min, max) that if we do not
assume conditions T and T , we obtain a non-dictatorial IFAR.
Proposition 8.6.6 [24] There exists a non-dictatorial IFAR fe : W n W

satisfying simultaneously IIA and IFPC.
i i
i i
i i
K23798 2015/2/2
i i
Proof. Let 12 , 1 . Let fe be an IFAR defined by W n with

fe() = , x, y X (with x 6= y), (x, x) = 1, (x, x) = 0

1 if i N, i (x, y) > i (y, x)
(x, y) =
otherwise,
.
0 if i N, i (x, y) < i (y, x)
(x, y) =
1 otherwise.
We first show that fe is well-defined, i.e., W n , = fe() W.

Let W n . We show that = fe() W. is reflexive since x
X, (x, x) = 1 and (x, x) = 0.
We show that is strongly complete, i.e., x, y X, (x, y) + (y, x) 1
x) 1. Let x, y X. We consider three cases:
and (x, y) + (y,
i (x, y) > i (y, x)
(1) If i N, , then
i (x, y) 1 and (x, y)+ (y, x) = 0+(1) <
1.
i (y, x) > i (x, y)
(2) If i N, , then
i (y, x) 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.
We now show that is transitive. Assume to the contrary that there

x, y, z X such that
(x, z) < min( (x, y), (y, z))
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 N, i (x, y) > i (y, x) and i (y, z) > i (z, y). (8.37)
i N, i (x, y) > i (y, x) and i (y, z) > i (z, y).

Otherwise, i N, i is pos-transitive since i W. Thus (8.37) and the
pos-transitivity of i imply i N, i (x, z) > i (z, x). Hence (x, z) = 1.
This contradicts the fact that (x, z) = ( 21 , 1).
(2) If (x, z) > max( (x, y), (y, z)), then (x, z) = 1 and
(x, y) = (y, z) = 0. Thus by the definition of , it follows that
i N, i (x, y) 
0.
We show that is -symmetric, i.e., x, y X, (x, y) + (x, y) =
(y, x) + (y, x). We distinguish two cases:
(i) If (x, y) = 1, then (y, x) = and i N, i (x, y) > i (y, x). By the
-symmetry of i ,we get i N, i (x, y) < i (y, x), i.e., (x, y) = 0 and
(y, x) = 1 . We have (x, y) + (x, y) = 1 = (y, x) + (y, x).
(ii) If (x, y) = , then j N such that j (x, y) j (y, x). By the
-symmetry of j , we have j (x, y) j (y, x) and (x, y) = 1 . Thus,
(x, y) + (x, y) = 1. To compute (y, x) + (y, x), we consider three
subcases:
Suppose that j (x, y) = j (y, x). We show that (y, x) + (y, x) =
1. By the -symmetry of j , j (x, y) = j (y, x). Thus (y, x) = and
(y, x) = 1 . Hence (y, x) + (y, x) = + (1 ) = 1 = (x, y) +
(x, y).
Suppose that j (x, y) < j (y, x) and k N {j}, k (x, y) < k (y, x).
Let us show that (y, x) + (y, x) = 1. The -symmetry of k implies k
N, k (x, y) > k (y, x). Thus (y, x) = 1 and (y, x) = 0. We get (y, x)+
(y, x) = 1 + 0 = 1 = (x, y) + (x, y).
Suppose that j (x, y) < j (y, x) and k N, k (x, y) k (y, x). We
show that (y, x) + (y, x) = 1. As in the previous case, the -symmetry of
j and k imply j (x, y) > k (y, x) and k (x, y) k (y, x). The definition
of implies (y, x) = and (y, x) = 1 . Hence (y, x) + (y, x) =
+ (1 ) = 1 = (x, y) + (x, y). Now is -symmetric. We conclude
that W.
i i
i i
i i
K23798 2015/2/2
i i
We next show that fe satisfies condition IFPC. Suppose that i N,

i (x, y) > 0 and i (x, y) < 1. We show that (x, y) > 0 and (x, y) < 1.
By definition of i , i N, i (x, y) > i (y, x) and i (x, y) 0 = (y, x) and (x, y) = 0 <
1 = (y, x). By the definition of ,we have the desired result.
We now show that fe satisfies condition IIA. Let , 0 W n and x, y X

be such that
0
i (x, y) > 0 i (x, y) > 0

0

(x, y) < 1 i (x, y) < 1

i
i N 0 (8.39)
i (y, x) > 0 i (y, x) > 0

i (y, x) < 1 0
i (y, x) < 1

0
(x, y) > 0 (x, y) > 0 (y, x) > 0
We show that and
(x, y) < 1 0 (x, y) < 1 (y, x) < 1
0
(y, x) > 0
.
0 (y, x) < 1
By definition of 0 , (8.39) becomes
0
i (x, y) > 0i (y, x)

i (x, y) > i (y, x)

0
i (x, y) i (x, y) i (y, x) > 0i (x, y)
0

i (y, x) = 0 (y, x)

(x, y) = 1 > = (y, x)
0 0
(x, y) = 0 < 1 = (y, x) 0 (x, y) = 0 < 1 0= (y, x)
(y, x) = 1 > = (x, y) (y, x) = 1 > = (x, y)

(y, x) = 0 < 1 = (x, y) (y, x) = 0 < 1 = 0 (x, y)
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)
Then by definition of i0 , the previous system implies i0 (x, y) > 0 and

i0 (x, y) < 1. Thus we obtain (x, y) > 0 and (x, y) < 1. Otherwise, the
collective preference = fe() on {x, y} satisfies (x, y) = (y, x) = and
(x, y) = (y, x) = 1 . By the definition of ,the previous system gives
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
We now show that

00 (y, x) = 0

(8.44)
00 (y, x) = 1.
Assume to the contrary that 00 (y, x) > 0 and 00 (y, x) < 1. This system,
(8.41) and the pos-transitivity of PL imply 00 (v, u) > 0 and 00 (v, u) < 1.
This contradicts (8.43).
By the second and the fourth systems of (8.40) and the perfect antisymme-
try of j , it follows that 00 and 0 coincide on (x, y). This assertion, condition
IIA and (8.44) imply 0 (y, x) = 0 and 0 (y, x) = 1. Hence the desired result
holds.
The following Lemma is an intuitionistic fuzzy version of the crisp veto-
field expansion lemma for crisp and fuzzy aggregation rules.
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.
Proof. We first determine the unique decisive coalition.

Consider the set Dfe of decisive coalitions of N . Condition IFPC implies that
N is decisive, that is, N Dfe. Since N is finite and Dfe 6= , Dfe has a minimal
decisive coalition, say S. Then S is the unique decisive coalition.
We next show that i S, i has a veto. Let i S. By the Veto field
expansion lemma, it suffices to show that i is almost semi-decisive over a pair
of options. Let x, y, z X. Let (W1T )n satisfy the following conditions:

i (x, y) > 0 i (y, z) > 0
(i) and

i (x, y) < 1
i (y, z) < 1
S\i (x, y) > 0 S\i (z, x) > 0

(ii) and . (8.45)

S\i (x, y) < 1 S\i (z, x) < 1
N \S (y, z) > 0 N \S (z, x) > 0

(iii) and

S\i (y, z) < 1 N \S (z, x) < 1
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
Since k N, k is pos-transitive, we have by (i) of (8.45 8.45) that

i (x, z) > 0
(8.47)
i (x, z) < 1
We consider two cases:
i i
i i
i i
K23798 2015/2/2
i i
8.7. Exercises 285
(1) If (z, x) = 0 and (z, x) = 1, then (8.47), the second system of

(ii) of (8.45) and the second system of (iii) of (8.45) imply that i is almost
semi-decisive over (x, z).
(2) We show that
(z, x) > 0
. (8.48)
(z, x) < 1
is not possible.
By the second system of (i) of (8.45), the first system of (iii) of (8.45)
and the pos-transitivity of each j , it follows that N \(S\i) (y, z) > 0 and
N \(S\i) (y, z) < 1. By (ii) of (8.45) and the pos-transitivity of each j , we
have that S\i (z, y) > 0 and S\i (z, y) < 1. The pos-transitivity of , (8.48)
and (8.46) imply (z, y) > 0 and (z, y) < 1. The three previous systems
imply that S\{i} is almost decisive on (y, z). By Lemma 8.6.3, we have that
S\{i} is decisive, which contradicts the minimality of S. Thus this case is not
possible.
8.7 Exercises
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.
4. [22] Show by example that partial transitivity ; partial quasi-

transitivity for = (1) or = (3) .
5. [22] Construct examples that show is max-min transitive, but not
consistent; is consistent, but not max-min transitive; is consistent,
but not transitive in the crisp sense (and not complete).
6. (a) Show that J = (max, min) satisfies condition G.

(b) For all l [0, ] show that IlF = (SlF , TlF ) satisfies condition G,
where IlF is the class of Frank t-conorms, t-norms.
(c) If S and T are dual, then the restriction of I = (S, T ) on L1 =
{(x1 , x2 ) | x1 + x2 = 1} satisfies condition G.

8. Prove Theorem 8.5.10.
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.
4. K. T. Atanassov, Intuitionistic Fuzzy Sets, Fuzziness and Soft Comput-

ing Physica-Verlag, Heidelberg 1999.
tive Preference, Michigan Studies in Political Science, The University of
choice, Social Choice and Welfare, 11 (1994) 121130.
7. C. R. Barrett, P. K. Pattanaik, and Maurice Salles, On the structure of
fuzzy social welfare functions, Fuzzy Sets and Systems, 19 (1980) 110.
8. H. Bustince and P. Burillo. Structures on intuitionistic fuzzy relations.

9. C. Cornelis, E. Kerre and G. Deschrijver. Implication in intuitionistic
fuzzy and interval-valued fuzzy set theory: Construction, classification,
application. International Journal of Approximate Reasoning, 35 (2004)
5595.
10. D. Dimitrov. Intuitionistic fuzzy preferences: A factorization. Advanced
Studies Contemporary Mathematics, 5 (2002) 93104.
11. D. Dimitrov. The Paretian liberal with intuitionistic fuzzy preferences:
A result. Social Choice and Welfare, 23 (2004) 149156.
12. B. Dutta, Fuzzy preferences and social choice, Mathematical Social Sci-
ences, 13 (1987) 215229.
14. L. A. Fono, V. Donfack-Kommogne, and N. G. Andjiga, Fuzzy arrow-
type results without the Pareto principle based on fuzzy pre-orders,
i i
i i
i i
K23798 2015/2/2
i i
8.8. References 287
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
19. Garcia-Lapresesta, J. Luis, and B. Llamazares, Aggregation of fuzzy
preferences: Some rules of the mean, Social Choice and Welfare, 17
(2000) 673690.
20. E. Klement, R. Mesiar and E. Pap. Triangular norms, Kluwer Academic

Publishers, Dordrecht, 2000.

23. J. N. Mordeson, T. D. Clark. M. J. Wierman, K. K. Nelson-Pook, and
K. Albert, Intuitionistic fuzzy preference relations and independence of
irrelevant alternatives, Journal Fuzzy Mathematics, 22 (2014) to appear.
24. G. N. Nana and L. A. Fono, Arrow-type results under intuitionistic fuzzy
preferences, New Mathematics and Natural Computation, 9 (2013) 127.
25. G. N. Njanpong. Relations binaires floues intuitionistes et application au
choix social. These de Doctorat PhD, Departement de Mathematiques,
Faculte des Sciences, Universite de Yaounde I-Cameroun, Janvier 2011.
26. S. V. Ovchinnikov, Social choice and Lukasiewicz logic, Fuzzy Sets and
Systems, 43 (1991) 275289.
cations, Social Choice and Welfare, 15 (1998) 359369.
i i
i i
i i
K23798 2015/2/2
i i
28. M. Salles. Fuzzy utility. Volume 1 of Handbook of Utility Theory, pages

321344. Principles. Kluwer Academic Publishers, Boston, 1998.
29. M. Salles, S. Geslin and A. Ziad. Fuzzy aggregation in economic en-

vironments: Quantitative fuzziness, public goods and monotonicity as-
sumptions. Mathematical Social Sciences, 45 (2003) 155166.
30. A. K. Sen, Collective Choice and Social Welfare, Advanced Textbooks
in Economics, Vol. 11, Editors C. J. Bliss and M. D. Intriligator, North
Holland 1979.
31. 31. A. K. Sen, Social Choice Theory, Handbook of Mathematical Eco-
nomics, Vol. 3, Elsevier Sciences Publishers B. V., North Holland, 1986,
10731181, Chapter 22.
32. 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.
33. Zeshui Xu. Intuitionistic preference relations and their application in
group decision-making. Information Sciences, 177 (2007) 23632379.
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
290 9. Manipulability of Fuzzy Social Choice Functions
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.
Definition 9.1.1 A crisp binary relation (CBR) on X is a function

: X 2 {0, 1}. A CBR on X is called
(1) complete if for all x, y X, (x, y) = 1 or (y, x) = 1;
(2) transitive if for all x, y, z X, (x, y) = 1 and (y, x) = 1 implies
(x, z) = 1;
(3) antisymmetric if for all x, y X, (x, y) = 1 and (y, x) = 1 implies
x = y.
Definition 9.1.2 A fuzzy binary relation (FBR) on X is a function

: X 2 [0, 1]. A FBR on X is called frail max-min transitive if
for all x, y, z X, (x, y) (y, x) (y, z) (z, y) implies (x, z)
(x, y) (y, z).
Definition 9.1.3 A fuzzy weak preference relation (FWPR) on X is

a reflexive FBR.
Definition 9.1.4 Let be a CBR. Then is called a crisp linear order

(CLO) if it is complete, anti-symmetric, and transitive.
Let i be a CLO, where i N. The best alternative set for individual i
coincides with the most preferred element. It is given by
P0 (X, i ) = {x X | y X, i (x, y) = 1}.
We now present three different definitions of an alternative to be at least

as good as another alternative.
Definition 9.1.5 [1] Let i N. Let x, y X.

(1) Simple degree rule (SDR) : i (x, y) i (y, x).
i i
i i
i i
K23798 2015/2/2
i i
9.1. Preliminaries 291
The best alternative set is defined as follows:
P1 (X, i ) = {x X | y X, i (x, y) i (y, x)}.
(2) Confidence threshold rule (CTR) : i (x, y) t, where t (0, 1/2].

P2 (X, i ) = {x X | y X, i (x, y) t}.
(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.
P3 (X, i ) = {x X | d(i , X)(x) = {d(i , X)(y) | y X}}.
We use the following notation.

Let Dc denotes the set of all strongly complete FWPRs and let D0 denotes
the set of all CLOs on X.
Let D1 = { Dc | satisfies max-min transitivity} and let D2 = {
Dc | satisfies frail max-min transitivity}.
D1 is used in the context of SDR and D2 is used in the context of CTR.
D3 = D1 when adopting DDR.
It follows from [13] that D0 D1 = D3 D2 since any CLO is reflexive
and since max-min transitivity implies frail max-min transitivity trivially.
The following result concerning the best alternative sets can be found in
[13, 19, 25, 26].
Proposition 9.1.6 Let A be a nonempty subset of X. Let be a FWPR.

(1) If Dj , then Pj (A, ) 6= for j {0, 1, 2, 3}.
(2) If D0 , then P0 (A, ) = P1 (A, ) = P2 (A, ) = P3 (A, ).
We let = (1 , . . . , n ) denote a profile of individuals preference relations.
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
Then i D1 , i = 1, 2, 3. We see that P1 (X, 1 ) = {y} since 1 (y, x) >

1 (x, y) and 1 (y, z) > 1 (z, y). Let a confidence threshold be given to be
t = .45. Then P2 (X, 2 ) = {x} and P2 (X, 1 ) = P2 (X, 3 ) = X. We also have
P3 (X, 3 ) = {z} since d(3 , X)(x) = d(3 , X)(y) = .6 and d(3 )(z) = .9.
i i
i i
i i
K23798 2015/2/2
i i
9.2 Fuzzy Social Choice Functions

We now present three types of fuzzy social choice functions and dictatorships
together with their corresponding manipulation procedures that were intro-
duced in [1].
Definition 9.2.1 A fuzzy social choice function (FSCF) is a function

of X into Rn ,where Rn is the set of all profiles of individuals fuzzy weak
preference relations.
Definition 9.2.2 A j-fuzzy social choice function (j-FSCF) is a func-

tion of Djn into Rn , j = 0, 1, 2, 3.
When each individual expresses an FWPR in Dj , a j-FSCF defines a social
choice. A 0-FSCF can be viewed as a voting rule.
Example 9.2.3 Let i , i = 1, 2, 3, be defined as in Example 9.1.7. Then i

Dj , i = 1, 2, 3; j = 1, 2, 3. Apply the arithmetic mean to obtain the fuzzy social
choice S and consider P1 (X, S ) = {x} as the social choice. Such an FSCF
can be viewed as an j-FSCF for j {1, 2, 3}, S is given by the following
table.
S x y z
x 1 .67 .71
y .58 1 .65
z .68 .63 1
We use the following notation:
(|0i ) is the profile of individuals fuzzy weak preference relations (1 , . . . ,
i1 , 0i , i+1 , . . . , n ).
(|0i , 0i+1 ) is the profile of individuals preference relations (1 , . . . , i1 ,
i , 0i+1 , . . . , n ).
0
(|01 , . . . , 0k ) for k {1, . . . , n}, is the profile of individuals fuzzy weak

preference relations (01 , . . . , 0k , k+1 , . . . , n ).
0 is the profile of individuals fuzzy weak preference relations (01 , . . . , 0n ).
Suppose a profile of individuals fuzzy weak preference relations is in Djn
and the social choice is obtained by the use of a j-FSCF, vj , j {0, 1, 2, 3}.
If every individual i N expresses a fuzzy weak preference relation i that
belongs to Dj , then Pj (X, i ) is the individuals best alternative set. Sup-
pose that an individual m N with sincere fuzzy weak preference relation
m knows the n 1 fuzzy weak preference relations of the remaining indi-
viduals and the j-FSCF, vj . Then individual m can anticipate the outcome
vj (). The individual then determines whether or not the outcome vj () can
be considered as a good social choice on the basis of his sincere preference,
m , that is, does outcome vj () belong to individual ms best alternative set
Pj (X, m )? If the outcome vj () does not belong to the set Pj (X, m ), then
i i
i i
i i
K23798 2015/2/2
i i
9.2. Fuzzy Social Choice Functions 293
individual m seeks a social choice at least as good as the outcome vj (n ) on

the basis of his fuzzy weak preference relation m . If there exists a binary
relation 0m in Dj such that the outcome vj (|0m ) is better than the outcome
vj (), then individual m can manipulate the j-FSCF, vj , by revealing 0m .
Thus the j-manipulability of a j-FSCF can be defined as follows:
Definition 9.2.4 [1] (j-manipulability) Let vj be a j-FSCF for j = 0, 1, 2, 3.

(1) v0 is said to be 0-manipulable if there exists m N, D0n and
m D0 such that v0 (|0m ) 6= v0 () and m (v0 (|0m ), v0 ()) = 1.
0

0
m D1 such that there exists x X such that m (x, v1 ()) > m (v1 (), x),
v1 (|0m ) 6= v1 (), and m (v1 (|0m ), v1 ()) (v1 (), v1 (|0m )).
0
m D2 such that for a fixed t [0, 1/2] there exists x X such that
m (v2 (), x) < t, v2 (|0m ) 6= v2 () and m (v2 (|0m ), v2 ()) t.
0
m D3 , there exists x X such that d(m , X)(x) > d(m , X) (v3 ()),
v3 (|0m )) 6= v3 () and d(m , X)(v3 (|0m )) d(m , X)(v3 ()).
The 0-manipulability of a 0-FSCF coincides with the one of Gibbard-
Satterthwaite. The j-manipulability of a j-FSCF implies its 0-manipulability
when its domain is restricted to the set D0n , j = 1, 2, 3.
Definition 9.2.5 (j-strategy proof ) A j-FSCF vj is said to be j-strategy

proof if it is not j-manipulable, j = 0, 1, 2, 3.
Example 9.2.6 [1] Consider Example 9.2.3.

(1) Individual 1 can 1-manipulate the FSCF. The alternative x is not a
member of his best alternative set P1 (X, 1 ) = {y}. Individual 1 can thus
reveal the non-sincere fuzzy relation 01 to obtain y as the social choice.
01 x y z
x 1 0 0
y .95 1 1
z .9 0 1
The arithmetic mean now yields
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
The arithmetic mean here is

x y z
x 1 .43 .51
y .67 1 .45
z .67 .67 1
The following definition generalizes the Gibbard-Satterthwaite impossibil-

ity result [19, 26] to the fuzzy case. For j = 0, 1, 2, 3, an individual m N
with a fuzzy weak preference relation m in Dj is considered to be a j-dictator
for the j-FSCF vj if the outcome vj () belongs to his best alternative set
Pj (X, m ) for all = (1 , . . . , m , . . . , n ). If a j-dictator exists for a j-FSCF
vj , then vj is considered to be j-dictatorial.
Definition 9.2.7 [1] (j-dictatorship) Let vj be a j-FSCF for j = 0, 1, 2, 3.

(1) v0 is said to be 0-dictatorial if there exists m N such that for all
D0n , x X, m (v0 (), x) = 1.
D1n , x X, m (v1 (), x) m (x, v1 ()).
D2n , for a fixed t (0, 1/2], x X, m (v2 (), x) t.
D0n , x X, d(m , X)(v3 ()) d(m , X)(x).
The 0-dictatorship of a 0-FSCF coincides with the G-S dictatorship. Also,
for j {1, 2, 3}, the j-dictatorship of a j-FSCF implies its 0-dictatorship when
its domain is restricted to D0n .
9.3 Impossibility Results

In this section, the strategy-proofness theorem of G-S is generalized to the
j-FSCFs for j = 1, 2, 3.
Theorem 9.3.1 (Gibbard [19], Satterthwaith [26]) Let v0 : D0n X be a 0-

FSCF such that v0 (D0n ) = X. If v0 is 0-strategy-proof, then it is 0-dictatorial.
i i
i i
i i
K23798 2015/2/2
i i
9.3. Impossibility Results 295
The GibbardSatterthwaite theorem states that for a set N of three or

more individuals and set A of at least three alternatives that if the individuals
can have any preference ordering of A for his or her preferences, then every
nondictatorial social choice procedure is manipulable for some distribution of
preferences. In legislative politics for example, there is something known as
the killer amendment. It is an amendment to a bill that will cause the bill to be
defeated. Here a legislator may not like the amendment, but knows the bill will
be defeated if the amendment is added. Another example involves plurality
voting. Here any number of candidates may compete for a particular office with
the candidate receiving the most votes being declared a winner. Individuals do
not wish to waste their votes on a candidate they prefer, but who they think
has no chance of winning. Consequently, they vote for a different candidate.
Lemma 9.3.2 Let A be a non-empty subset of X and be a CBR on X. If

satisfies the following conditions, then is a CLO on X and P0 (X, ) A.
(1) is a CLO on A;
(2) (x, y) = 1 and (y, x) = 0 for all x A and y X\A;
(3) is a CLO on X\A.
Proof. By (1), (x, x) = 1 for all x A and by (3) (x, x) = 1 for all
x X\A. Thus is reflexive. By (1), (x, y + (y, x) 1 for all x, y A; by
(2), (x, y + (y, x) 1 for all x A, y X\A; by (3), (x, y + (y, x) 1
for all x, y X\A. Thus is strongly complete. Let x, y, z X be such that
(x, y) = 1 and (y, z) = 1. Suppose that z A and x X\A. If y A, then
(x, y) = 0, a contradiction. If y X\A, then (y, z) = 0, a contradiction,
Thus we have either x, y A or x, z X\A or (x A and z X\A). In each
of the three cases (x, z) = 1. Hence is transitive.
Let x P0 (X, ). Suppose that x X\A. Then there exists y A such
that (x, y) = 0. However, this contradicts the assumption that x P0 (X, ).
Hence P0 (X, ) A.
We next show that the GS result can be extended to the fuzzy context.
We show that any j-strategy-proof j-FSCF is j-dictatorial. The proof follows
along the lines as in [29].
Theorem 9.3.3 (Abdelaziz, Figueira, and Meddeb [1]) Let vj be a j-FSCF

such that vj (D0n ) = X and j {1, 2, 3}. If vj is j-strategy-proof, then it is
j-dictatorial.
Proof. We prove the case for j = 1 since the cases for j = 2, 3 follow in an
entirely similar manner. Consider a 1-strategy-proof 1-FSCF, v1 : D1n X.
Let v0 : D0n X be a 0-FSCF such that for all D0n , v0 () = v1 (). The 0-
FSCF is 0-strategy proof since v1 is 1-strategy-proof. Thus by Theorem 9.3.1,
v0 is 0-dictatorial. Let individual 1 be the 0-dictator for v0 . We show that
individual 1 is also a 1-dictator for v0 .
i i
i i
i i
K23798 2015/2/2
i i
Let D1n . Let x0 be v1 (). Consider P1 (X, 1 ) = {x X | y X,

1 (x, y) 1 (y, x)} be the best alternative set of individual 1. We will show
that x0 P1 (X, 1 ).
Let 0 D0n be a profile of individuals CLOs such that:
For i = 1,
0
i is a CLO on P1 (X, 1 ),
0 (x, y) = 1 and 0i (y, x) = 0 if x P1 (X, 1 ), y X\P1 (X, 1 ),
i0
i is a CLO on X\P1 (X, 1 ).
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 ).
By Lemma 9.3.2, 0i is a CLO for all i N and P0 (X, 01 ) P1 (X, 1 ).

Suppose that xk = v1 (|01 , . . . , 0i , . . . , 0k ) is the social choice when the k
first individuals change their preference relations i to 0i in order to contradict
individual 1.
Note that k {0, 1, . . . , n}. If k = n, then xn = v1 (0 ) = v0 (0 ). Thus
xn belongs to P0 (X, 1 ) because of the dictatorship of 1 by v0 . Therefore
xn P1 (X, 1 ).
Let j denote the least k in {0, 1, . . . , n} such that xk P1 (X, 1 ). For x0
to be a member of P1 (x, 1 ), it must be shown that j = 0. The proof is by
contradiction. Suppose that j 1.
If j = 1, then
x1 = v1 (|01 ) P1 (X, 1 ) (9.1)
x0 = v1 ()
/ P1 (X, 1 ). (9.2)
Now (9.1) implies y X, 1 (x1 , y) 1 (y, x1 ). Thus 1 (x1 , x0 )
1 (x0 , x1 ). Also (9.2) implies x X such that 1 (x, x0 ) > (x0 , x). There-
fore, x1 is 1-manipulable by individual 1 at .
If j > 1, then
xj = v(|01 , 02 , . . . , 0j ) P1 (X, 1 ),
xj1 = v1 (|01 , 02 , . . . , 0j1 )
/ P1 (X, 1 ).
Thus 0j (xj1 , xj ) = 1 and 0j (xj , xj1 ) = 0. Hence
0j (xj1 , xj ) > 0j (xj , xj1 ) .
Suppose (|01 , . . . , 0j ) is the profile of individuals preference relations. If in-

dividual j declares a fuzzy preference j instead of a crisp relation 0j , then
v1 (|01 , . . . , 0j ) is changed in his favor. Therefore, v1 is 1-manipulable by in-
dividual j at (|01 , . . . , 0j ).
i i
i i
i i
K23798 2015/2/2
i i
9.4. Non-Manipulable Partitioning 297
Consequently, j must be equal to 0. Hence x0 = v1 () P1 (X, 1 ) for all

D1n . It follows that individual 1 is also a 1-dictator.
We have shown how an individual could manipulate a social choice even if
the preferences of the individuals are ambiguous. Three definitions of strategy-
proofness were proposed and it was shown in all the three cases that the G-Ss
theorem remains valid.
9.4 Non-Manipulable Partitioning

The final two sections are based on the results of Duddy, Perote-Pena, and
Piggins [13]. Consider a group of individuals that wishes to classify a set
of alternatives as belonging to one of two sets. Since the individuals may
disagree, an aggregation rule is employed to determine a compromise outcome.
If it is required that the social classification should not be imposed nor be
manipulable, then it is shown that the only such rules are dictatorial. A survey
concerning the potential for social choices to be manipulated can be found in
Barbera [3].
Consider a set N of individuals that wish to partition a finite set X into
two subsets, say A and B. Suppose the individual evaluators submit crisp,
non-empty, strict subsets of X. The subset submitted by an evaluator con-
tains the alternatives that this evaluator classifies as belonging to set A. The
complement of this set is the set of alternatives in X that the evaluator clas-
sifies as belonging to set B.
Let X be a finite set of alternatives containing at least three members. Let
N = {1, . . . , n}, n 2, denote the set of evaluators. Let be a function of X
into P(N ). A profile is a function of X into P(N ) such that for all i N,
there exists x, y X such that i (x)\(y). The set (x) can be interpreted
as the set of evaluators who have voted for x at profile . We require that
for every evaluator i there exists at least one alternative x that the evaluator
has voted for and at least one alternative y that the evaluator has not. Let P
denote the set of all profiles.
A social classification is a function v : X [0, 1] such that (i) if the
range of v contains 1 then it also contains a number < 1/2 and (ii) if the
range of v contains 0, then it also contains a number > 1/2. Let V denote
the set of all social classifications. In the remainder, of the chapter greater
than and less than mean in the strict sense.
An aggregation rule is a function f : P V. If and 0 are distinct
profiles and there exists i N such that (x)\{i} = 0 (x)\{i} for all x X,
then we say that and 0 differ at evaluator i only.
We now define some properties that aggregation rules may satisfy.
i i
i i
i i
K23798 2015/2/2
i i
Definition 9.4.1 [13 Let f be an aggregation rule. Then f is said to be (or

satisfy)
(1) Non-manipulable (NM) : x X, i N, and , 0 P, if and
0
/ (x) implies v(x) v 0 (x);
differ at evaluator i only, then i
(2) Non-imposition (NI) : x X, , 0 P such that v(x) = 0 and
0
v (x) = 1;
(3) Dictatorial: i N such that x X and P, i (x) implies
v(x) > 1/2 and i / P (x) implies v(x) < 1/2;
(4) Monotone independent (MI) : x X and , 0 P, if (x)
0 (x), then v(x) v 0 (x).
In other words, if every evaluator who votes for x at profile also votes
for x at profile 0 , then the social classification of x at must be no greater
that it is at 0 .
Lemma 9.4.2 If an aggregation rule is non-manipulable, then it satisfies

monotone independence.
Proof. Let f be an aggregation rule that is NM. Assume that f is not MI.
Then there exists , 0 P and x X such that (x) 0 (x) and v(x) > v 0 (x).
Let = (0) and 0 = (n) . Let the sequence of profiles (0) , (1) , . . . , (n)
be such that at (1) evaluator 1 votes as he/she does at while evaluators
2, . . . , n vote as they do at profile 0 ; at profile (k) evaluators 1, . . . , k vote
as they do at profile while evaluators k + 1, . . . , n vote as they do at profile
0 , k = 1, . . . , n. Since v (0) (x) > v (n) , there exist profiles (t1) and (t) such
that v (t1) (x) > v (t) (x). By construction t / (t1) (x) and (t1) and (t)
(t1)
differ only at evaluator t. However, t / (x) and v (t1) (x) > v (t) which
contradicts NM.
The following condition is used in the next result.
Unanimity: For all x X and for all v V, (x) = implies v(x) = 0
and (x) = N implies v(x) = 1.
Lemma 9.4.3 If a non-manipulable aggregation rule satisfies non-imposition,

then it satisfies unanimity.
Proof. Let f be an aggregation rule that is NI and NM. Let x X.
Then there exists P such that v(x) = 0. Let P be a profile such that
(x) = . By Lemma 9.4.2, f is MI since it is NM. Since (x) (x), v (x)
v(x) by MI. Hence v (x) = 0. Let P be a profile such that (x) = N.
By NI there exists 0 P with v 0 (x) = 1. Since 0 (x) (x), we have
v 0 (x) v (x) by MI. Thus v (x) = 1.
Neutrality: x, y X and , 0 P,
(1) if (x) = 0 (y), then v(x) and v 0 (y) are both less than 1/2 or both
greater than 1/2 or both equal to 1/2;
i i
i i
i i
K23798 2015/2/2
i i
9.4. Non-Manipulable Partitioning 299
(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.
Lemma 9.4.4 If a non-manipulable aggregation rule satisfies non-

imposition, then it satisfies neutrality.
Proof. Suppose f is an aggregation rule that is NM and NI. Let x, y X
and , 0 P be such that either (x) = N \0 (y) or (x) = 0 (y).
(1) Suppose x 6= y and (x) = N \0 (y). Let be a profile such that
(x) = (x), (y) = 0 (y), and 0 (x) = for all z X\{x, y}. By Unanimity,

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.
Theorem 9.4.5 (Duddy, Perote-Pena, and Piggins [13]) If f is a non-manip-

ulable aggregation rule that satisfies non-imposition, then f is dictatorial.
Proof. Let x X. Consider a sequence of profiles (0) , (1) , . . . , (n) such
that (0) (x) = , . . . , (i) (x) = {1, . . . , i} for i = 1, . . . , n. By Lemma 9.4.3,
f satisfies unanimity and hence v (0) (x) = 0 and v (n) (x) = 1. Thus there
exists j N such that v (j1) (x) 1/2 and v (j) (x) > 1/2. Let 0 be a profile
such that 0 (y) = N \(j) (x), 0 (x) = {j} and 0 (z) = (j1) for all z
X\{x, y}. By Lemma 9.4.4, f satisfies Neutrality which in turn implies that
since v (j1) 1/2, 0 (z) 1/2 for all z X\{x, y}. Since v (j) (x) > 1/2, we
have v 0 (y) < 1/2 by neutrality. Thus by the definition of social classification,
we have that v 0 (x) > 1/2. (This is true even though j is the only individual
who votes for x at profile 0 .) Let 00 P be such that 00 (x) = N \0 (x). By
Neutrality, v 00 (x) < 1/2. Let P. If j (x), then 0 (x) (x) and so
v(x) > 1/2 by MI. If j / (x), then (x) 00 (x) and so v(x) < 1/2 by MI.
Thus when evaluator j votes for x, the social classification of x must be greater
than 1/2. When evaluator j does not vote for x, the social classification of x
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].
Example 9.5.1 Let N = {1, 2, 3} and X = {w, x, y, z}.

For i = 1, 2, 3, define vi : X [0, 1] as follows:
v1 (w) = 1, v1 (x) = 0, v1 (y) = 0, v1 (z) = 0,
v2 (w) = 1, v2 (x) = 1, v2 (y) = 1, v2 (z) = 0,
v3 (w) = 1, v3 (x) = 1, v3 (y) = 1, v3 (z) = 0.
i i
i i
i i
K23798 2015/2/2
i i
9.6. Exercises 301
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
Define v : X [0, 1] by C X, v(C) = (sC m)/(M m). Then M = 3

and m = 0. Thus it follows that
v(w) = 1, v(x) = 2/3, v(y) = 2/3, v(z) = 0.
Example 9.5.2 Let N = {1, 2, 3} and X = {w, x, y, z}.

For i = 1, 2, 3, define vi : X [0, 1] as follows:
v1 (w) = 1, v1 (x) = 0, v1 (y) = 0, v1 (z) = 1,
v2 (w) = 1, v2 (x) = 1, v2 (y) = 1, v2 (z) = 0,
v3 (w) = 1, v3 (x) = 1, v3 (y) = 1, v3 (z) = 0.
Define sC , M, m, and v as in Example 9.5.1. Then M = 3 and m = 1.
Hence it follows that
v(w) = 1, v(x) = 1/2, v(y) = 1/2, v(z) = 0.
Thus evaluator 1 has successfully lowered the social classification of both x
and y. Hence the rule is manipulable.
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).
(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
2. If C is a strategy-proof fuzzy choice function, prove that it is weakly

Paretian.
3. If C is a strategy-proof fuzzy choice function, prove that it is monotonic.

4. Let M be a fuzzy choice function defined as follows: F(X)n x
X, M ()(x) = median{p1 , . . . , pn1 , 1 (x), . . . , n (x)}, where pi [0, 1],
i = 1, . . . , n 1. Prove that M is strategy-proof. (M is called a fuzzy
augmented median voter rule.)
5. Let i0 (x) = 1 and i00 (x) = 0 for all x X and some i N. Prove that a
fuzzy choice function C is strategy-proof if and only if for all F(X)n
and all x X and all i N, C()(x) = median{i (x), C(N \i , i00 )(x),
C(N \i , i0 )(x)}.
6. Prove that a fuzzy choice function C is strategy-proof if and only if it

is a fuzzy augmented median voter rule.
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.
2. K. J. Arrow, Social Choice and Individual Values, Wiley, New York,

1951.
3. S. Barbera, An introduction to strategy-proof social choice functions,
Soc. Choice Welf., 18 (2001) 619653.
4. J-P Barthelemy, B. Leclerc, and B. Monjardet, On the use of ordered

sets in problems of comparison and consensus of classifications, Journal
of Classification, 3 (1986) 187234.
5. C. R. Barrett, P. K. Pattianaik, and M. Salles, On choosing rationality
6. C. P. Chambers and A. D. Miller, Rules for aggregating information,

Social Choice and Welfare, 36 (2011) 7582.
7. M. Dasgupts and R. Deb, Transitivity and fuzzy preferences, Social
Choice Welfare, 13 (1996) 305318.
8. S. Diaz, B. De Baets, and S. Montes, Additive decomposition of fuzzy

pre-orders, Fuzzy Sets and Systems, 158 (2007) 830842.
9. S. Diaz, S. Montes, and B. De Baets, Transitive decompositions of fuzzy
preference relations: the case of nilpotent, Kybernetika, 40 (2004) 7188.
i i
i i
i i
K23798 2015/2/2
i i
9.7. References 303
10. S. Diaz, S. Montes, and B. De Baets, Transitive bounds in additive fuzzy

preference structures, IEEE Trans. Fuzzy Systems, 15 (2007) 275286.
11. F. Dietrich and C. List, Strategy-proof judgement aggregation, Eco-

nomics and Philosophy, 23 (2007) 269300.
12. D. Dimitrov, T. Marchant, and D. Mishra, Separability and aggrega-
tion of equivalence relations, Economic Theory, DOI 10.1007/s00199-
011-0601-2 (2011).
13. C. Duddy, J. Perote-Pena, and A. Piggins, Non-manipulable partition-

ing, New Mathematics and Natural Computation, 8 (2012) 273282.
14. B. Dutta, Fuzzy preference and social choice, Math. Social Sci., 13
(1987) 215229.
15. B. Dutta, S, C. Panda, and P. K. Pattanaik, Exact choices and fuzzy

preferences, Math. Soc. Sci., 11 (1986) 5368.
16. P. C. Fishburn and A. Rubinstein, Aggregation of equivalence relations,
Journal of Classification, 3 (1986) 6165.

Decision Support, Kluwer Academic Publishers, Dordrecht, 1994.
18. J. L. Garcia-Lapresta and B. Liamazares, Aggregation of fuzzy prefer-
ences: Some rules of the mean, Social Choice and Welfare, 17 9(2000)
673690.
19. A. Gibbard, Manipulation of voting schemes: A general result, Econo-

metrica, 41 (1973) 587601.
20. K. Inada, The simple majority decision rule, Econometrica, 37 (1969)
490506.
21. P. Kulshreshtha and B. Shecker, Interrelationships among fuzzy

preference-based choice functions and significance of rationality condi-
tions: a taxonomic and intuitive perspective, Fuzzy Sets and Systems,
109 (2000) 429445.
22. B. Leclerc, Efficient and binary consensus functions on transitively val-
ued relations, Mathematical Social Sciences, 8 (1984) 4561.
23. C. List, Which worlds are possible? A judgment aggregation problem,
Journal of Philosophical Logic, 37 (2008) 5765.
24. B. G. Mirkin, On the problem of reconciling partitions, in H. M. Blalock
(ed.) Quantitative Sociology: International Perspectives on Mathemati-
cal and Statistical Modeling, Academic Press New York, 1975.
i i
i i
i i
K23798 2015/2/2
i i
25. B. Peleg, Game Theoretic Analysis in Communities, Cambridge Univer-

sity Press, Cambridge, 1984.
26. M. A. Satterthwaite, Strategy-proofness and Arrows conditions: exis-

tence and corresponding theorems for voting procedures and social wel-
fare functions, J. Economic Theory, 10 (1975) 187217.
27. K. Sengupta, Fuzzy preferences and Orlovsky choice procedure, Fuzzy
28. K. Sengupta, Choice rules with fuzzy preferences: Some characteriza-

tions, Social Choice and Welfare, 16 (1999) 259272.
29. F. Tang, Fuzzy preferences and social choice, Bull. Economic Res., 46
(1994) 263269.
30. L. A. Zadeh, Quantitative fuzzy semantics, Information Sciences, 3

(1971) 159176.
i i
i i
i i
K23798 2015/2/2
i i
Chapter 10
Similarity of Fuzzy Choice

Functions
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
306 10. Similarity of Fuzzy Choice Functions
ally means indistinguishable. This consideration leads to the paradox that

objects A and B may be indistinguishable and B and C may be indistinguish-
able and yet A and C can be distinguished. Thus Poincare denies transitivity
in the real world. One way to model indistinguishability is by the use of a
t-norm in such a way that indistinguishability between objects is measured.
Let E be a fuzzy relation on a set X such that E is reflexive, symmetric
and max- transitive. If elements x, y in X are indistinguishable of degree t,
i.e., E(x, y) = t and y, z are indistinguishable of degree s, i.e., E(y, z) = s,
then x, z are indistinguishable of degree E(x, z) t s. Here we use the term
similar rather indistinguishable.
Much of fuzzy revealed preference theory focuses on the connection of the
fuzzy choice functions and fuzzy revealed preferences associated with them.
These connections are expressed by several partial functions between the set
of choice functions and the set of fuzzy preference relations. These functions
connect the similarity of fuzzy choice functions and the similarity of fuzzy pref-
erence relations. These functions translate the (, t)-equality of fuzzy choice
functions into the (, t)-equality of fuzzy preference relations [7, 15, 46].
The degree of similarity of two fuzzy choice functions allows for the eval-
uation of how rational a fuzzy choice function is. Consequently, a ranking of
a collection of fuzzy choice functions with respect to their rationality can be
obtained. This way of evaluating the rationality of a fuzzy choice function
with the consistency condition F studied in [17, 18] is also presented.
10.1 Fuzzy Choice Functions

We assume is a left-continuous t-norm throughout this chapter. The Fodor
nilpotent minimum t-norm is left continuous, but not continuous. Recall that
the residuum associated with a t-norm is a b = {t [0, 1] | a t b}.
The negation operation associated with is defined by a = a 0 = {t
[0, 1] | a t = 0}.
We next state some known and useful lemmas.
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
10.1. Fuzzy Choice Functions 307
Recall that the biresiduum associated with the left-continuous t-norm is

defined by

(a, b) = a b = (a b) (b a).
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).
Lemma 10.1.4 Let X be a non-empty set and let f, g : X [0, 1] be func-

tions. Then
(1)

({f (x) | x X}, {g(x) | x X}) {
(f (x), g(x)) | x X};

(2) ({f (x) | x X}, {g(x) | x X}) { (f (x), g(x) | x X}.
Recall that a fuzzy subset of X is called normal if (x) = 1 for some
x X. For , FP(X), let
I(, ) = {(x) (x) | x X} and E(, ) = {(x) (x) | x X}.
Clearly, I(, ) = 1 and = E(, ) = 1. For all x X, we

have
I (, ) (x) (x) and E (, ) (x) (x).
I(, ) is called the subsethood degree of and and E(, ) the
degree of equality (or degree of similarity) of and , [6, p. 82].
Recall that a fuzzy preference relation on X is called strongly total if
(x, y) = 1 or (y, x) = 1 for any distinct x, y X. is reflexive and strongly
total if and only if it is strongly connected.
Let 1 and 2 be two fuzzy preference relations. The degree of similarity
E(1 , 2 ) of 1 and 2 is defined as follows:
E(1 , 2 ) = {1 (x, y) 2 (x, y) | x, y X}.
A fuzzy relation E on X is said to be a similarity relation if it is

reflexive, symmetric and max- transitive.
Let 0 t 1. We say that two fuzzy subsets , of X are (, t)-equal,
written =(,t) , if E(, ) t. The (, t)-equality generalizes the Cai t-
equality [7]. See also [46].
We present some fuzzy choice functions introduced in [18]. We also present
some fuzzy preference relations associated with fuzzy choice functions.
i i
i i
i i
K23798 2015/2/2
i i
A fuzzy choice space is a pair (X, B) where X is a non-empty set of alter-

natives and B is a non-empty family of non-zero fuzzy subsets of X. A fuzzy
choice function on (X, B) is a function C : B F P (X) such that for all
B, C() .
The members of B can be given an interesting interpretation. They can be
considered as vague criteria or vague attributes. If x X and B, then (x)
can be thought of as the degree to which x is available with respect to . It
represents the degree to which x satisfies the criterion and C()(x) expresses
the degree to which the alternative x can be chosen with respect to . Thus
a fuzzy choice problem can be thought of as a multicriterial decision-making
problem.
Results of previous chapters are proved under the assumption that the
choice function C satisfies the following hypotheses:
(H1) For all B, C() is a normal fuzzy subset of X;
(H2) B includes the characteristic functions 1{x1 ,...,xn } for any n 1 and
x1 , . . . , xn X.
Since C() , is also normal in H1 for all B. From H2, it follows
immediately that B includes all non-empty finite subsets of X in the crisp
case.
Let (X, B) be a fuzzy choice space and a fuzzy preference relation on
X. For all B, we define the fuzzy subsets M (, ) and G(, ) of X as in
Chapter 2:
M (, )(x) = (x) ({((y) (y, x)) (x, y) | y X}),

G(, )(x) = (x) ({(y) (x, y) | y X}).
The functions M ( , ) : B F P(X) and G( , ) : B F P(X) are not in

general fuzzy choice functions. In the case of the Godel t-norm , a sufficient
condition for M ( , ) and G( , ) to be fuzzy choice functions is for to be a
reflexive and max- transitive fuzzy preference relation on a finite set X and
B FP(X), see Proposition 2.3.11.
The functions M ( , ) and G( , ) allow the notion of rationality of fuzzy
choice functions to be introduced. Let C be a fuzzy choice function on (X, B).
Then C is called M -rational if C = M ( , ) for some preference relation
on X and is called G-rational if C = G( , ) for some preference relation
on X. In case of the G odel t-norm , if is reflexive and strongly total, then
M ( , ) = G( , ) [18]. In this case, M -rationality and G-rationality coincide.
Let C be a fuzzy choice function on (X, B). Define the fuzzy revealed
preferences C, C and C on X by for all x, y X,
C (x, y) = {C()(x) (y) | B};

C (x, y) = C(1{x,y} )(x);
C (x, y) = {C()(x) (y) C()(y) | B}.
i i
i i
i i
K23798 2015/2/2
i i
10.2. Similarity of Fuzzy Choice Functions 309
Let x, y X. Then C (x, y) is the degree of truth of the statement that

there is a criterion such that the alternative x is chosen with respect to
criterion and alternative y satisfies . Also, C (x, y) represents the degree
of truth of the statement that the alternative x is chosen from {x, y} and
C (x, y) is the degree of truth of the statement that there is a criterion
with respect to which x is chosen and y is rejected.
The functions C , C and C are fuzzy versions of some preference rela-
tions studied in classical revealed preference theory [1, 32, 35].
Let C be a fuzzy choice function. Then C is said to be G-normal if C
= G( , C ) and is said to be M -normal if C = M ( , C }) [16]. G-normality
is a special case of G-rationality and M -normality is a special case of M -
rationality.
10.2 Similarity of Fuzzy Choice Functions

We present two of Georgescus notions, the degree of similarity of fuzzy choice
functions and the (, t)-equality of fuzzy choice functions and her results that
establish some correspondences between the similarity of fuzzy choice func-
tions and the similarity of fuzzy preference relations. We determine how these
correspondences translate the (, t)-equality of fuzzy choice functions into the
(, t)-equality of fuzzy preference relations.
Recall that denotes a left-continuous t-norm.
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.
Proposition 10.2.2 Let C1 , C2 , C3 be fuzzy choice functions on (X, B). Then

(1) C1 = C2 if and only if E(C1 , C2 ) = 1;
(2) E(C1 , C2 ) = E(C2 , C1 );
(3) E(C1 , C2 ) E(C2 , C3 ) E(C1 , C3 ).
Proof. Conditions (1) and (2) clearly hold.
(3) By Lemma 10.1.2(5) and Lemma 10.1.3(5), we have that
E(C1 , C2 ) E(C2 , C3 ) = ({(C1 () (x) C2 () (x)) | B, x X})
({(C2 () (x) C3 () (x)) | B, x X})
{(C1 () (x) C2 () (x))
(C2 () (x) C3 () (x)) | B, x X}
{C1 () (x) C3 () (x) | B, x X}
= E(C1 , C3 ).
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.
Theorem 10.2.3 (Georgescu [19]) (1) For all 1 , 2 Dom(1 ),

E (1 (1 ), 2 (2 )) E(1 , 2 ) E(1 , 2 );
(2) For all 1 , 2 Dom(2 ), E (2 (1 ), 2 (2 )) E(1 , 2 ).
Proof. (1) By Lemma 10.1.3(2), (7), (8) and Lemma 10.1.4(1), we have
that
E (1 (1 ), 1 (2 ))
= {
(1 (1 )()(x), 1 (2 )()(x)) | B, x X}
= { (M (, 1 )(x), M (, 2 )(x)) | B, x X}
= {
((x) {((y) (1 (y, x)) 1 (x, y)) | y X},
(x) {((y) 2 (y, x)) 2 (x, y) | y X} | B, x X}
{
(((y) 1 (y, x)) 1 (x, y) | y X},
{((y) 2 (y, x)) 2 (x, y) | y X} | B, x X}
{
(((y) 1 (y, x)) 1 (x, y),
(S(y) 2 (y, x)) 2 (x, y)) | B, x, y X}
{
((y) 1 (y, x), (y) 2 (y, x))

(1 (x, y), 2 (x, y)) | B, x X}
(1 (y, x), 2 (y, x))

(1 (x, y), 2 (x, y)) | x, y X}
E(1 , 2 ) E(1 , 2 ).
i i
i i
i i
K23798 2015/2/2
i i
(2) By Lemma 10.1.3(2), (7), (8) and Lemma 10.1.4(1), we have
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 ).
Corollary 10.2.4 (1) If 1 =(,t) 2 , then
1 (1 ) =(,tt) 1 (2 ) and 2 (1 ) =(,t) 2 (2 );
(2) Let be the G

odel t-norm. If 1 =(,t) 2 , then 1 (1 ) =(,t) 1 (2 ).
The next result establishes how the functions 1 , 2 and 3 transfer the
similarity of choice functions into the similarity of the corresponding fuzzy
preference relations.
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
E(1 (C1 ), 1 (C2 ))

= {
(1 (C1 )(x, y), 1 (C2 )(x, y)) | x, y X}
=
{ (C1 (x, y), C2 (x, y)) | x, y X}
=
({C1 ()(x) (y) | B},
{C2 ()(x) (y) | B}) | x, y X}
{
(C1 ()(x) (y), C2 ()(x) (y)) | x, y X, B}

{ (C1 ()(x), C2 ()(x)) | x X, B} = E(C1 , C2 ).
(2) By the definitions, we have that
E(2 (C1 ), 2 (C2 )) = {

(2 (C1 )(x, y), 2 (C2 )(x, y)) | x, y X}
= {
(C1 (x, y), C2 (x, y)) | x, y X}
= {
(C1 (1{x,y} )(x), C2 (1{x,y} )(x)) | x, y X}.
i i
i i
i i
K23798 2015/2/2
i i
Since {1{x,y} | x, y X} B, it follows that
{
(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
{ (3 (C1 )(x, y), 3 (C2 )(x, y)) | x, y X}

{
( C1 (x, y), C2 (x, y)) | x, y X}
{ ( C1 (x, y), C2 (x, y)) | x, y X}
{ ({C1 ()(x) (y) C1 ()(y) | B},
{C2 ()(x) (y) C2 ()(y) | B}) | x, y X}
{ (C1 ()(x), C2 ()(x))

(C1 ()(y), C2 ()(y)) | x, y X, B}
{ (C1 ()(x), C2 ()(x))

(C1 ()(y), C2 ()(y)) | x, y X, B}
{ (C1 ()(x), C2 ()(x)) | x, y X, B}
{
(C1 ()(y), C2 ()(y)) | x, y X, B}
= E (C1 , C2 ) E (C1 , C2 ) .
Corollary 10.2.6 (1) If C1 =(,t) C2 , then 1 (C1 ) =(,t) 1 (C2 ) , 2 (C1 )

=(,t) 2 (C2 ) and 3 (C1 ) =(,tt) 3 (C2 ) ;
(2) If is the Godel t-norm, then C1 =(,t) C2 implies 3 (C1 ) =(,t)
3 (C3 ) .
Let Cb denote G( , C ). Then a fuzzy choice function C is G-normal if and

only if C = C.
b That is, if and only if the rationality of C is ensured by the
fuzzy revealed preference relation C .
If the t-norm is the G odel t-norm , then it was shown in [16] that
C C.
b
The degree of similarity E(C, C)
b can be considered a measure of the degree
of rationality of C; the larger E(C, C)
b is, the more rational the choice function
C is.
Recall the following consistency condition for a fuzzy choice function C
defined in Chapter 2.
Condition F . For all , B and x X, I(, ) (x) C( )(x)
C()(x).
The next result connects the degree of similarity E(C, C)
b and the consis-
tency condition F .
i i
i i
i i
K23798 2015/2/2
i i
Theorem 10.2.7 (Georgescu [19]) Let C be a fuzzy choice function on

(X, B). For all , B and x X, E(C, C)
b (I(, ) (x) C( )(x))
C()(x).
b
Proof. In view of Lemma 2.3.1(1), it suffices to prove that E(C, C)
I(, ) (x) C( )(x) C()(x). We have
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
Thus I(, ) (x) C(

b )(x) C()(x).
b
Therefore,
b I(, ) (x) C( )(x)
E(C, C)
b I(, ) (x) C(
E(C, C) b )(x)
E(C, C)
b C()(x)
b
b )(y) C( 0 )(y) | x X, 0 B} C()(x)
= {C( 0 b
(C()(x)
b C()(x)) C()(x)
b
= C()(x)
b C()(x)
= C()(x),
where the next to last equality follows by Lemma 10.1.1(2).

We see that the more rational the fuzzy choice function C is, the larger
the degree to which C verifies the consistency condition F .
Theorem 10.2.8 (Georgescu [19]) Suppose the fuzzy choice function C on

(X, B). If C satisfies hypotheses H1 and H2, then E(C, C)
b E(C , C ).
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
By the definition of C , we have since C C that
E(C , C ) = {C (x, y) C (x, y) | x, y X}

= {C (x, y) C (x, y) | x, y X}.
Since C is reflexive [16] and 1{x,y} (x) = 1, 1{x,y} (z) = 0 for z

/ {x, y},
we have by Lemma 10.1.1(4) that
C(1{x,y} )(x) = 1{x,y} (x) ({1{x,y} (z) C (x, z) | z X})

= [1{x,y} (x) C (x, x)] [1{x,y} (y) C (x, y)]
= C (x, x) C (x, y)
= C (x, y).
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 ).
Example 10.2.9 We consider the degree of similarity of a particular col-

lection of fuzzy choice functions. Let X = {x, y} and B = {, }, where
, F P(X) are defined by = s1{x} + 1{y} and = 1{x} + t1{y} for
0 s, t 1.
Consider the function C : B FP(X) defined by C() = q1{x} + 1{y}
and C() = 1{x} + r1{y} , where 0 q s and 0 r t. Then C is a fuzzy
choice function on (X, B). The fuzzy preference relation C is as follows:
C (x, x) = C (y, y) = 1;
C (x, y) = (C()(x) (y)) (C()(x) (y)) = t q;

C (y, x) = (C()(y) (x)) (C()(y) (x)) = s r.
The values of C()

b and C()
b are as follows:
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
The degree of similarity E(C, C)

b of C and C
b is as follows:
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
Example 10.2.10 Let X = {x, y} and let and be fuzzy subsets of X

given by = 21 1{x} + 1{y} , = 1{x} + 12 1{y} . If we take and as available
sets, we obtain the choice space (X, B). Consider a decision-making problem
for which there are 5 fuzzy choice problems. The 5 fuzzy choice problems are
specified by the corresponding choice functions:
1 1
C1 () = 1{x} + 1{y} , C1 () = 1{x} + 1{y} ,
3 5
1 1
C2 () = 1{x} + 1{y} , C2 () = 1{x} + 1{y} ,
8 3
1 1
C3 () = 1{x} + 1{y} , C3 () = 1{x} + 1{y} ,
3 3
1 1
C4 () = 1{x} + 1{y} , C4 () = 1{x} + 1{y} ,
8 9
1 1
C5 () = 1{x} + 1{y} , C5 () = 1{x} + 1{y} .
7 8
i i
i i
i i
K23798 2015/2/2
i i
Using the degree of similarity determined in Example 10.2.9, we have that

1 1 1
E(C1 , C
c1 ) = = ,
3 5 5
1 1 1
E(C2 , C
c2 ) = = ,
8 3 8
1 1 1
E(C3 , C
c3 ) = = ,
3 3 3
1 1 1
E(C4 , C
c4 ) = = ,
8 9 9
1 1 1
E(C5 , C
c5 ) = = .
7 8 8
We now can obtain the following ranking of the five fuzzy choice functions:
C3 followed by C1 and then C2 , C5 and finally C4 .
It follows that the fuzzy choice problem given by the fuzzy choice function
C3 is the most rational.
Georgescu provides the following interesting discussion: Most of the results

from fuzzy revealed preference theory are achieved by following two methods:
the first one from fuzzy choice functions to fuzzy preference relations, and the
second one from fuzzy preference relations to fuzzy choice functions. This is
expressed in this section by the functions i and i .
Theorems 10.2.7 and 10.2.8 establish the behavior of the functions i and
j towards the similarity. Thus an evaluation of how the information is pre-
served by the functions i and j can be obtained.
The degree of similarity E(C, C) b of C and C b defines a measure of Cs
rationality. The more rational the act of choice represented by C is, the bigger
E(C, C)
b is. When E(C, C) b = 1, C is G-normal. Thus E(C, C) b is an indicator
that allows for a classification of the fuzzy choice functions with respect to
their rationality, Example 10.2.10.
Theorem 10.2.8 connects this indicator with the consistency condition F .
10.3 Arrow Index of a Fuzzy Choice Function

We now consider the Arrow index as presented by Georgescu [22]. The Arrow
index A(C) of a fuzzy choice function C is a fuzzy version of the classical
Arrow axiom. It is a measure of the degree to which C satisfies the Fuzzy
Arrow Axiom. We present Georgescus result that A(C) characterizes the
degree to which C is full rational. The Arrow index allows for the ranking of
fuzzy choice functions with respect to their rationality.
The Arrow axiom (AA) was introduced in [1] in order to characterize the
full rationality of choice functions. See also [39] for additional information. The
i i
i i
i i
K23798 2015/2/2
i i
10.3. Arrow Index of a Fuzzy Choice Function 317
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.
Definition 10.3.1 ([6]) Define the fuzzy subsets of FR(X) as follows:

FR(X),
Ref() = {(x, x) | x X};
Trans() = {((x, y) (y, z)) (x, z) | x, y, z X};
ST() = {(x, y) (y, x) | x, y X, x 6= y}.
i i
i i
i i
K23798 2015/2/2
i i
Then Ref() is called the degree of reflexivity of , Trans() is called the

degree of transitivity of , and ST() the degree of strong totality of .
Clearly, is reflexive if and only if Ref() = 1, is max- transitive if and
only if Trans() = 1, and is strongly total if and only if ST() = 1.
The indicators in Definition 10.3.1 allow for the comparison of two fuzzy
preference relations 1 and 2 with respect to these properties. For example, if
Trans(1 ) Trans(2 ), then the fuzzy preference relation 1 can be considered
to be more transitive than 2 .
Lemma 10.3.2 ([6]) Let be fuzzy preference relation on X and x, y, z X.

Then Trans() (x, y) (y, z) (x, z).
Proof. (x, y) (y, z) (x, z) = 1 if (x, y) (y, z) (x, z) and
the result is immediate in this case. Suppose (x, y) (y, z) > (x, z). Then
((x, y)(y, z)) (x, z) = (x, z). Hence Trans() (x, z) and the desired
result follows.
Often authors develop their results concerning fuzzy choice functions and
fuzzy preference relations on the notion that social choice is governed by fuzzy
preferences, but the act of choice is exact and so the choice functions are crisp,
[3, 4, 5, 26]). However, several real world situations require the consideration
of fuzzy choice functions [2, 14, 28, 45]. The first fuzzy choice function was
emphasized by Orlovsky [30].
In [16, 17, 19, 20], Georgescu considered fuzzy choice functions C for which
the domain and the range consist of fuzzy subsets of X. For this case, the avail-
able sets of alternatives are fuzzy subsets of X. This leads to the notion of
the degree of availability of an alternative x with respect to an available set.
The availability degree is useful when the decision maker possess partial infor-
mation concerning the alternative x or when a criterion limits the possibility
of choosing x. Hence the available sets can be considered criteria in decision
making.
We assume in this section that hypotheses H1 and H2 hold. We fix a
continuous t-norm and its residuum .
If C is a fuzzy choice function on (X, B), then recall that
C (x, y) = {C()(x) (y) | B},
C (x, y) = C(1{x,y} )(x)
for all x, y X.
Let x, y X. Then the real number C (x, y) is the degree of truth of the
statement that there exists a criterion such that alternative x is chosen with
respect to and alternative y verifies . The real number C (x, y) represents
the degree of truth of the statement that from the set {x, y} at least the
alternative x is chosen.
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.
Let (X, B) be a fuzzy choice space and be a fuzzy preference relation on

X. For all B, recall the fuzzy subset G(, ) of X is defined as follows:
G(, )(x) = (x) {(y) (x, y) | y X}
for all x X. Then G( , ) can be considered a function from B into FP(X).
Lemma 10.3.4 ([20]) Suppose C is a fuzzy choice function on (X, B). Then
C() G(, C ) for all B.
In general, G( , Q) is not a fuzzy choice function. There might exist a B

such that G(, ) is the zero function. However, by Lemma 10.3.4, G( , C ) is
a fuzzy choice function and is called the image of C.
A fuzzy choice function C on (X, B) is called rational if C = G( , ) for
some fuzzy preference relation on X. If is reflexive, max- transitive, and
total, then C is called full rational.
Let C1 , C2 be two fuzzy choice functions on (X, B). Recall from the pre-
vious section that the degree of similarity of C1 and C2 is defined by
E(C1 , C2 ) = {C1 ()(x) C2 ()(x) | x X, B}
By Proposition 10.2.2, the function (C1 , C2 ) E(C1 , C2 ) is a similarity
relation on the set of fuzzy choice functions defined on (X, B). Similarity
of fuzzy functions is a concept analogous to similarity of fuzzy preference
relations ([6, 14, 25, 27]).
The connection between the similarity of two fuzzy choice functions and
the similarity of fuzzy preference relations associated with them is given by
the next result.
Lemma 10.3.5 Let C1 , C2 be a fuzzy choice function on (X, B). Then

E(C1 , C2 ) C1 ()(x) C2 ()(x)
for all B and x X.
Proof. We have
E(C1 , C2 ) C1 ()(x) = C1 ()(x) {C1 ( )(y) C2 ( )(y) | y X, B}
C1 ()(x) [C1 ()(x) C2 ()(x)]
= C1 ()(x) C2 ()(x)
C2 ()(x).
i i
i i
i i
K23798 2015/2/2
i i
Lemma 10.3.6 Let C be a fuzzy choice function on (X, B) and be a fuzzy

preference relation on X. Then
E(C, G( , )) C()(x) (y) (x, y)
for all B and x, y X.

Proof. By Lemma 10.3.5, we have
E(C, G( , )) C()(x) G(, )(y) (y) (x, y).
Hence the desired inequality follows.

Arrows axiom (AA) is a condition introduced in [1] to characterize full
rationality of crisp choice functions. A crisp choice function C is said to satisfy
AA if for all available sets of alternatives S1 , S2 , the following implication
holds:
S1 S2 S1 C(S2 ) = or S1 C(S2 ) = C(S1 ).
As in Georgescu [20], we say that a fuzzy choice function C on (X, B)

satisfies the fuzzy Arrow axiom (FAA) if for all 1 , 2 B and all x X
the following statement holds:
I(1 , 2 ) 1 (x) C(2 )(x) E(1 C(2 ), C(1 )).
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) = {(I(1 , 2 ) 1 (x) C(2 )(x)) E(1 C(2 ), C(1 ))

| x X, 1 , 2 B}.
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.
Lemma 10.3.8 (Georgescu [22]) Suppose is the minimum t-norm . If C

is a fuzzy choice function (X, B), then A(C) C(1{x,y} )(x) C(1{y,z} )(y)
C(1{x,y,z} )(x).
Proof. Assume to the contrary that
A(C) C(1{x,y} )(x) C(1{y,z} )(y) C(1{x,y,z} )(x). (10.1)
We show that (10.1) implies the condition
A(C) C(1{x,y} )(x) C(1{y,z} )(y) C(1{x,y,z} )(y). (10.2)
Assume that (10.2) does not hold, i.e.,
A(C) C(1{x,y} )(x) C(1{y,z} )(y) C(1{x,y,z} )(y). (10.3)
Now I(1{x,y} , 1{x,y,z} ) = 1 and 1{x,y} (y) = 1. Hence
A(C) C(1{x,y,z} )(y)

= A(C) I(1{x,y} , 1{x,y,z} ) 1{x,y} (y) C(1{x,y,z} )(y)
I({x,y} , 1{x,y,z} ) 1{x,y} (y) C(1{x,y,z} )(y)
[I(1{x,y} , 1{x,y,z} ) 1{x,y} (y)
C(1{x,y,z} )(y) E(1{x,y} C(1{x,y,z} ), C(1{x,y} ))].
Therefore, by Lemma 2.3.1(2)
A(C) C(1{x,y,z} )(y) E(1{x,y} C(1{x,y,z} ), C(1{x,y} ))

I(C(1{x,y} ), 1{x,y} C(1{x,y,z} ))
C(1{x,y} )(x) (1{x,y} (x) C(1{x,y,z} (x))
C(1{x,y} )(x) C(1{x,y,z} (x)).
It now follows that
A(C) C(1{x,y,z} )(y) C(1{x,y} )(x) C(1{x,y,z} (x)).
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
contradicting (10.1). Thus we obtain (10.2). Similarly, we get

A(C) C(1{x,y} )(x) C(1{y,z} )(y) C(1{x,y,z} )(z). (10.4)
From (10.1), (10.2), and (10.4), we have that C(1{x,y,z} )(x) 6= 1,
C(1{x,y,z} ) (y) 6= 1 and C(1{x,y,z} )(z) 6= 1. However, since H1 holds, this
contradicts that C(1{x,y,z} ) is a normal fuzzy subset of X. This contradiction
yields the desired result.
The following proposition states that max- transitivity of the fuzzy pref-
erence relation C is ensured if the Arrow axiom holds.
Theorem 10.3.9 (Georgescu [22]) Suppose is the minimum t-norm . If

C is a fuzzy choice function (X, B), then A(C) Trans(C ).
Proof. Let x, y, z X. We show that
A(C) C(1{x,y} )(x) C(1{y,z} )(y) C(1{x,y} )(z).
By Lemma 10.3.8.
A(C) C(1{x,y} )(x) C(1{y,z} )(y) C(1{x,y,z} )(x). (10.5)
Now
A(C) C(1{x,y,z} )(x)
= A(C) I(1{x,y} , 1{x,y,z} ) 1{x,y} (x) C(1{x,y,z} )(x)
I(1{x,z} , 1{x,y,z} ) 1{x,z} (x) C(1{x,y,z} )(y)
[I(1{x,z} , 1{x,y,z} ) 1{x,z} (x)
C(1{x,y,z} )(x) E(1{x,z} C(1{x,y,z} ), C(1{x,z} ))]
E(1{x,z} C(1{x,y,z} ))
(1{x,y} C(1{x,y,z} ))(x) C(1{x,z} )(x)
C(1{x,y,z} )(x) C(1{x,z} )(x).
Hence we have
A(C) C(1{x,y,z} )(x) C(1{x,z} )(x).
By (10.5), it follows that
A(C) C(1{x,y} )(x) C(1{y,z} )(y) A(C) C(1{x,y,z} )(x) C(1{x,z} )(x).
Thus for all x, y, z X, we have
A(C) (C(1{x,y} )(x) C(1{y,z} )(y)) C(1{x,z} )(x)
= (C (x, y) C (y, z)) C (x, z).
Therefore,
A(C) {(C (x, y) C (y, z)) C (x, z) | x, y, x X} Trans(C ).
i i
i i
i i
K23798 2015/2/2
i i
Corollary 10.3.10 Suppose is the minimum t-norm . If C satisfies FAA,

then C is max- transitive.
Proof. By Theorem 10.3.9, A(C) = 1 implies Trans(C ) = 1.
Lemma 10.3.11 (Georgescu [22]) Suppose = . If C is a fuzzy choice

function (X, B), then
E(C, G( , Q)) Trans() Ref() ST() A(C). (10.6)
Proof. Let 1 , 2 B and x X. We next prove the inequality
E(C, G( , )) Trans() Ref() ST() (10.7)

(I(1 , 2 )) 1 (x) C(2 )(x)) E(1 C(2 )), C(1 )).
Let = E(C, G( , )Trans()Ref()ST() I(1 , 2 ) 1 (x) C(2 )(x).

Then the inequality in (10.7) is equivalent to the inequality
E(1 C(2 ), C(1 )). (10.8)
Let z, v X. By Lemmas 10.3.6 and 10.3.2, we have that
I(1 , 2 ) 1 (x) C(2 )(x) C(1 )(z) 2 (v)

[E(C, G( , )) C(1 )(z) 1 (x)]
[E(C, G( , )) C(1 )(x) 2 (v)]
(z, x) (z, v)
Trans() (z, x) (x, v)
(z, v).
By Lemma 2.5.1, we obtain
C(1 )(z) = I(1 , 2 ) 1 (x) C(2 )(x) C(1 )(z) 2 (v) (z, v)
for all v X. Hence
C(1 )(z) ({2 (v) (z, v) | v X}).
Also,
C(1 )(z) I(1 , 2 ) 1 (x) C(2 )(x) C(1 )(z)

C(1 )(z) I(1 , 2 )
1 (z) [1 (z) 2 (z)]
2 (z).
Therefore,
C(1 )(z) 2 (z) ({2 (v) (z, v) | v X}) = G(2 , )(z).
i i
i i
i i
K23798 2015/2/2
i i
Since C(1 )(z) E(C, G( , )) by Lemma 10.3.5, we have that
C(1 )(z) E(C, G( , )) G(2 , )(z) C(2 )(z).
Hence by Lemma 2.5.1(4) and (10.3), it follows that
C(1 )(z) C(2 )(z) (10.9)

= [C(1 )(z) 1 (z)] [C(1 )(z) C(2 )(z)]
= C(1 )(z) (1 (z) C(2 )(z)).
Let y, z X. By Lemma 2.3.1(2), Lemma 10.3.6, and Lemma 10.3.2, we have
C(1 )(z) C(2 )(z) 1 (y)

= I(1 , 2 ) 1 (x) C(2 )(x) 1 (z) C(2 )(z) 1 (y)
1 (y) 2 (y) C(2 )(x) C(2 )(z)
= 1 (y) 2 (y) C(2 )(x) C(2 )(z)
[E(C, G( , )) C(2 )(z) 2 (x)]
E(C, G( , )) C(2 )(x) 2 (y)]
(z, x) (x, y)
Trans() (z, x) (x, y)
(z, y).
Thus we have that 1 (z) C(2 )(z) 1 (y) (z, y) for all y X.
Hence
1 (z) C(2 )(z) 1 (z) ({1 (y) (z, y) | y X})

= G(1 , )(z).
Since E(C, G( , )), we obtain
1 (z) C(2 )(z) E(C, G( , )) G(1 , )(z) C(1 )(z).
Therefore, we have for all z X,
(1 (z) C(2 )(z)) C(1 )(z). (10.10)
By (10.9) and (10.10), it follows that
{(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).
Theorem 10.3.12 (Georgescu [22]) Suppose is the minimum t-norm . If

C is a fuzzy choice function (X, B), then
A(C) = {E(C, G( , )) | FR(X)} Trans() Ref() ST()]

= E(C, G( , C )) Trans(C ).
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
{E(C, G( , )) Trans() Ref() ST() | FR(X)} A(C). (10.11)
Now we shall prove the inequality
A(C) E(C, G( , C )). (10.12)
Let B and x X. We next prove the following two inequalities:
A(C) C()(x) G(, C ) (10.13)
A(C) G(, C )(x) C()(x). (10.14)

Let z X. Then
A(C) C()(x) C()(z)
= A(C) (I(1{x,y} , ) 1{x,y} (x) C()(x)
I(1{x,y} , ) 1{x,y} (x) C()(x)
[(I(1x,z} , ) 1{x,y} (x) C()(x) E(1{x,z} C(), C(1{x,z} )]
= I(1{x,y} , ) 1{x,y} (x) C()(x) E(1{x,z} C(), C(1{x,z} ))
E(1{x,z} C(), C(1{x,z} )
(1{x,z} (x) C()(x) C(1{x,z} )(x))
= C()(x) C (x, z)
C()(x) C (x, z).
Hence
A(C) C()(x) (z) C (x, z).
Thus A(C) C()(x) (z) C (x, z) for all z X. Hence
A(C) C()(x) (x) ({(z) C (x, z) | z X})

= G(, C )(x).
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
G(, C )(x) (y) C (x, y) = 1 C (x, y) = C (x, y) = C(1{x,z} )(x).
We note that
G(, C )(x) (x) = (x) (y) C()(y)

= I(1{x,y} , ) 1{x,y} (y) C()(y).
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
Since G(, C )(x) 1({x,y} )(x), it follows that
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).
Hence A(C) G(, C )(x) C()(x) and (10.14) holds.

The inequalities (10.13) and (10.14) hold for all B and x X. Thus
A(C) {C1 ()(x) G(, C )(x) | B, x X} = E(C, G( , C )).
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
A(C) E(C, G( , C )) Trans(C )

= E(C, G( , C )) Trans(C ) Ref(C ) ST(C )
{E(C, G( , )) Trans() Ref() ST() | FR(X)}.
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,
C (x, y) C()(y) (x) C()(x).
Strong fuzzy congruence axiom (SFCA)

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.
Definition 10.3.14 ([21]) Let C be a fuzzy choice function on a fuzzy choice

space (X, B). Define the following indicators:
(1) WFCA(C) = {((x) C()(y) C (x, y)) C()(x) | x, y X,
B}
(2) SFCA(C) = {((x) C()(y) W (x, y)) C()(x) | x, y X,
B}.
Suppose is the minimum t-norm . Then the following assertions are
equivalent:
(1) WFCA(C) = 1 if and only if C satisfies the axiom WFCA;
(2) SFCA(C) = 1 if and only if C satisfies the axiom SFCA.
The indicator WFCA(C) (resp. SFCA(C)) expresses the degree to which
the fuzzy choice function C satisfies the axiom WFCA(C) (resp. SFCA(C)).
Theorem 10.3.15 ([21]) Suppose is the minimum t-norm . If C is a

fuzzy choice function, then
WFCA(C) = SFCA(C)
= E(C, G( , C )) Trans(C ) = E(C, G( , C )) Trans(C ).
The next result follows from Theorems 10.3.12 and 10.3.15.
Theorem 10.3.16 Suppose is the minimum t-norm . If C is a fuzzy

choice function, then
A(C) = WFCA(C) = SFCA(C)

= E(C, G( , C )) Trans(C ) = E(C, G( , C )) Trans(C ).
By Theorems 10.3.15 and 10.3.16, one obtains the equality of WFCA(C)

and SFCA(C) with the Arrow index, and an evaluation of their value in terms
of the similarity and the transitivity indicator.
Suppose that is the minimum t-norm . The fuzzy Arrow-Sen theorem
established in [18, 21] is a particular case of Theorem 10.3.15. When C is a
crisp choice function, an important part of the classic Arrow-Sen theorem ([1,
35, 39]) is obtained.
The connection as developed by Georgescu between the fuzzy choice func-
tions and the fuzzy preference relations which rationalize them is realized by
conditions on fuzzy choice functions such as the axiom of revealed preference
and congruence, consistency properties. The fuzzy Arrow axiom (FAA) is one
such condition.
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 329
4. Prove that Trans() Ac ().
5. Let 1 , 2 be fuzzy preference relations. Prove that Ac (1 ) E(1 , 2 )

Ac (2 ).
Define the indicators Ac RatG (C) = {E(C, G( , )) Ac()) | FR}
and Ac RatM (C) = {E(C, M ( , )) Ac()) | FR}
6. If C is a fuzzy choice function on (X, B) that satisfies H1 and H2 , prove
that RatG (C) = Ac Rat(C).
7. If C is a fuzzy choice function on (X, B) that satisfies H1 and H2 ,
prove that RatG (C) = Ac RatG(C) = F (C) F Cond(C), where
F Cond(C) = {I(G(, C ), C()) | B}. (F Cond is the fuzzy version
of the Condorcet property.)
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,
3. C. R. Barrett, P. K. Pattanaik, and M. Salles, On choosing rationally

4. C. R. Barrett, P. K. Pattanaik, and M. Salles, Rationality and aggrega-
tion of preferences in ordinal fuzzy frameworks, Fuzzy Sets and Systems,
49 (1992) 913.

social welfare functions, Fuzzy Sets and Systems, 19, (1996) 110.
6. R. Belohl
avek, Fuzzy Relational Systems, Foundations and Principles,
Kluwer, Dordrecht, 2002.
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
11. B. De Baets, B. Van De Walle, and E. Kerre, Fuzzy preference structures

without incomparability, Fuzzy Sets and Systems, 76 (1995) 333348.
12. B. De Baets, B. Van De Walle, and E. Kerre, Characterizable fuzzy
preference structures, Ann. Oper. Res., 80 (1998) 105136.
13. D. Dubois and H. Prade, Fuzzy Sets and Systems, Theory and Applica-
tions, Academic Press, New York, 1980.
Decision Support, Kluwer, Dordrecht, 1994.
15. I. Georgescu, A generalization of the Cai -equality of fuzzy sets,
Proceedings of the Fuzzy Information Processing Conference, Beijing,
China, 1-4 March 2003, pp. 123127.
16. I. Georgescu, On the axioms of revealed preference in fuzzy consumer
theory, J. Systems Sci. Systems Eng., 13 (2004) 279296.
17. I. Georgescu, Consistency conditions in fuzzy consumers theory, Fund.
Inform., 61 (2004) 223245.
18. I. Georgescu, Rational choice and revealed preference: A fuzzy approach,
Ph.D. Dissertation 60, Turku Centre for Computer Science, 2005.
tems, 158 (2007) 13141326.
20. I. Georgescu, Arrows axiom and full rationality for fuzzy choice func-
tions, Social Choice and Welfare, 28 (2007) 303319.
21. I. Georgescu, Fuzzy choice functions: A revealed preference approach,
Berlin, Springer, 2007.
22. I. Georgescu, Arrow index of a fuzzy choice function, Fundamenta In-
formaticae, 99 (2010) 245261.
23. I. Georgescu, Acyclic rationality indicators of fuzzy choice functions,
24. I. Georgescu, Ranking fuzzy choice functions by their rationality indi-
cators, Fuzzy Optimization and Decision Making, 6 (2007) 367389.
25. P. H
ajek, Methamathematics of Fuzzy Logic, Kluwer, Dordrecht, 2000.
26. S. Klaue, K. Kurbel, and I. Loutschko, Automated negotiations on
agent-based e-marketplaces: An overview. Proceedings of the 14th Bled
Electronic Commerce Conference, Bled, Slovenia 508519 2001.
27. E. P. Klement, R. Mesiar, and E. Pap, Triangular Norms, Kluwer, Dor-
drecht, 2000.
i i
i i
i i
K23798 2015/2/2
i i
10.5. References 331
28. P. Kulshreshtha and B. Shekar, Interrelationship among fuzzy

preference-based choice function and significance of rationality condi-
tions: a taxonomic and intuitive perspective, Fuzzy Sets and Systems,
109 (2000) 429445.
29. K. Kurbel, and I. Loutschko, Towards multi-agent electronic market-
places: What is there and what is missing? The Knowledge Engineering
Review, 18 (2003) 3346.
30. S. Orlovsky, Decision making with a fuzzy preference relation, Fuzzy

31. S. Ovchinnikov, Decision-making with ternary binary relation, in Pro-
ceedings of the Conference on Information Processing Management of
Uncertainty IPMU 2004, Perugia, Italy 511516.

645.
33. M. Roubens, Some properties of choice functions based on valued binary
relations, European J. Oper. Res., 40 (1989) 309321.
34. P. A. Samuelson, A note of the pure theory of consumers behavior,

Economica, 5 (1938) 6171.
35. A. K. Sen, Choice functions and revealed preference, Rev. Econom.
Stud., 38 (1971) 307317.
36. A. K. Sen, Social choice theory: A re-examination, Ecometrica, 45 (1977)

5389.
37. A. K. Sen, Choice, Welfare and Measurement, Cambridge MA: MIT
Press and Oxford: Blackwell, 1982.
38. K. Suzumura, Rational choice and revealed preference, Rev. Econom.
Stud., 43 (1976) 149159.
39. K. Suzumura, Rational choice, collective decisions and social welfare,
Cambridge University Press, Cambridge, 1983.
40. E. Trillas and L. Valverde, An Inquiry into Indistinguishability Opera-
tors, in: H. J. Skala, S. Termini, E. Trillas (Eds.), Aspects of Vagueness,
Reidel, Dordrecht, 1983.
41. E. Turunen, Mathematics Behind Fuzzy Logic, Physica-Verlag, Wurzburg,
1999.
42. H. Uzawa, A note on preference and axioms of choice, Ann. Inst. Statist.
Math., 8 (1956) 3540.
i i
i i
i i
K23798 2015/2/2
i i
43. B, Van deWalle, S. Heitch, and P. Faratin, Coping with one-to-many

multicriteria negotiations in electronic marketplaces: Beyond price dis-
covery, Munich, Germany, 2001.
44. X. Wang, B. De Baets, E. E. Kerre, A comparative study of similarity
measures, Fuzzy Sets and Systems, 73 (1995) 259268.
45. Ph. de Wilde, Fuzzy utility and equilibria, IEEE Trans. Systems Man
Cybernet, 34 (2004) 17741785.
46. M. Ying, Perturbations of fuzzy reasoning, IEEE Trans. Fuzzy Systems,

7 (1999) 625629.
47. L.A. Zadeh, Similarity relations and fuzzy orderings, Inform. Sci., 3
(1971) 177200.
i i
i i

Application of Fuzzy Logic To Social Choice Theory

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Application of Fuzzy Logic To Social Choice Theory

Transféré par

Droits d'auteur :

Formats disponibles

i i

MONOGRAPHS AND RESEARCH NOTES IN MATHEMATICS

MONOGRAPHS AND RESEARCH NOTES IN MATHEMATICS

Forthcoming Titles (continued)

No claim to original U.S. Government works

International Standard Book Number-13: 978-1-4822-5099-2 (eBook - PDF)

1 Fuzzy Maximal Subsets 1

2 Fuzzy Choice Functions 21

3 Factorization of Fuzzy Preference Relations 71

4 Fuzzy Non-Arrow Results 115

4.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

5 Fuzzy Arrows Theorem 137

6 Single Peaked Fuzzy Preferences: Blacks Median Voter

8 Arrow-Type Results under Intuitionistic Fuzzy Preferences 239

9 Manipulability of Fuzzy Social Choice Functions 289

9.5 Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300

10 Similarity of Fuzzy Choice Functions 305

[24] introduced the concepts of positive transitivity and negative transitivity

Fuzzy Maximal Subsets

One paradigmatic change in science and mathematics this century concerns

2 1. Fuzzy Maximal Subsets

1.1 Fuzzy Set Theory

1.1. Fuzzy Set Theory 3

write xRy. Given R, we define a binary relation P on X in terms of R by

4 1. Fuzzy Maximal Subsets

The bounded difference t-norm is also known as the Lukasiewicz t-norm.

Definition 1.1.3 Let F R (X). Define the fuzzy relation on X by

Definition 1.1.4 Let FR (X) and let be a t-norm.

Definition 1.1.5 Let FR (X) and let be a t-norm. Let be an asym-

1.2. Fuzzy Maximal Sets 5

Various types of asymmetric preference relations are used in fuzzy social

Definition 1.1.7 Let F R(X) and let be an asymmetric fuzzy prefer-

If a strict preference relation associated with is simple, then (x, y) =

1.2 Fuzzy Maximal Sets

6 1. Fuzzy Maximal Subsets

Definition 1.2.1 Let C be a function from X X into [0, 1]. Define M :

1.2. Fuzzy Maximal Sets 7

In the next several results, we determine the relationship between fuzzy

Proposition 1.2.3 Let R be a relation on X and let S be a nonempty subset

Proof. We have that

In Proposition 1.2.3, we have that if C(x, w) = 1S (x) (w, x), then

Proposition 1.2.5 Suppose has no zero divisors. Let R be a relation on X

8 1. Fuzzy Maximal Subsets

Corollary 1.2.6 Let R be a relation on X and let S be a nonempty subset

Proposition 1.2.7 Let (, ) FR (X) F P (X) and x X. Then

MM (, )(x) = (x) ({(x, w) | w Supp()}).

Proof. Since X is finite, we have that MM (, )(x) = {t [0, 1] | (x)

Proposition 1.2.8 Let ~ = . Then (, ) FR (X)FP (X), MM (, )

Let w1 = {(x, w) | (w) (w, x) > (x, w), w Supp()} if w Supp()

Proposition 1.2.9 Let (, ) FR (X) F P (X) and x X. Then

where tx = {t [0, 1] | C(x, w) t (x, w), w Supp()} if w Supp()

In Proposition 1.2.9, tx = {(x, w) | C(x, w) > (x, w), w Supp()} if

1.2. Fuzzy Maximal Sets 9

Theorem 1.2.10 Let = (0) . Suppose has no zero divisors. Let

Definition 1.2.11 Let (, ) FR (X) F P (X). Define the fuzzy subset

Clearly x N D(, ) if and only if (y, x) = 0 y X.

Definition 1.2.12 Let (, ) F R (X) F P (X). Suppose is regular.

Definition 1.2.13 Let FR (X). If (x1 , x2 ) (x2 , x3 ) . . .

10 1. Fuzzy Maximal Subsets

Theorem 1.2.14 Suppose has no zero divisors. Let FR (X) be reflex-

Corollary 1.2.15 Suppose has no zero divisors. Let FR (X) be reflex-

1.2. Fuzzy Maximal Sets 11

Example 1.2.16 For regular strict fuzzy preference relations, M (, ) 6=