Defects of Properties in Mathematics

Defects of Properties in
Mathematics
Quantitative Characterizations
Copyright © 2002. World Scientific Publishing Company. All rights reserved.
Gal, Sorin G., and Adrian I. Ban. Defects of Properties in Mathematics, edited by Sorin G. Gal, and Adrian I. Ban,
World Scientific Publishing Company, 2002. ProQuest Ebook Central, .
Created from boglib on 2016-12-29 04:29:34.
SERIES ON CONCRETE AND APPLICABLE MATHEMATICS
Series Editor: Professor George A. Anastassiou

Department of Mathematical Sciences
The University of Memphis
Memphis, TN 38152, USA
Published
Vol. 1 Long Time Behaviour of Classical and Quantum Systems
edited by S. Graffi & A. Martinez
Vol. 2 Problems in Probability
by T M. Mills
Vol. 4 Stochastic Models with Applications to Genetics, Cancers, Aids and
Other Biomedical Systems
by Tan Wai-Yuan
Forthcoming:
Vol. 3 Introduction to Matrix Theory

by F. Szidarovszky & S. Molnar
Gal, Sorin G., and Adrian I. Ban. Defects of Properties in Mathematics, edited by Sorin G. Gal, and Adrian I. Ban, World
Scientific Publishing Company, 2002. ProQuest Ebook Central, .
Series on Concrete and Applicable Mathematics Vol. 5
Defects of Properties in
Mathematics
Quantitative Characterizations
Adrian I Ban & Sorin G Gal

University ofOradea, Romania
V f e World Scientific
wb New Jersey •London'Sine
London • Singapore • Hong Kong
Published by
World Scientific Publishing Co. Pte. Ltd.
P O Box 128, Farrer Road, Singapore 912805
USA office: Suite IB, 1060 Main Street, River Edge, NJ 07661
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.
DEFECTS OF PROPERTD2S IN MATHEMATICS

QUANTITATIVE CHARACTERIZATIONS
Copyright © 2002 by World Scientific Publishing Co. Pte. Ltd.
All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright
Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to
photocopy is not required from the publisher.
ISBN 981-02-4924-1
Printed in Singapore by Uto-Print

To our lovely wives Olimpia Ban and Rodica Gal
This page is intentionally left blank
Preface
The m a i n aim is to reveal a m e t h o d of research in m a t h e m a t i c s , called by

us quantitative study of the defect of property, which can be used in var-
ious fields of m a t h e m a t i c s . Our viewpoint over the m a t h e m a t i c a l entities
is t h a t of an analyst, even if they belong to algebra, geometry, topology
or logic. We examine in this monograph in a systematic way, the quanti-
tative characterizations of the "deviation from a (given) property", by our
terminology called "defect of property", in: Set Theory, Topology, Measure
Theory, Real Function Theory, Complex Analysis, Functional Analysis, Al-

gebra, Geometry, N u m b e r Theory and Fuzzy Mathematics. We also present
a great variety of applications and open problems. To our knowledge, it
is the first t i m e in literature t h a t a book has systematically studied this
direction of research. This book is interdisciplinary in m a t h e m a t i c s and
contains the research of both authors over the past six years in these sub-
jects. It also references most of the works of other main researchers in these
areas. Each chapter can be read independently. T h e introduced concepts
are simple and the proving m e t h o d s are rather elementary, t h a t is, mak-
ing this material accessible to undergraduate and graduate students and
researchers. T h e large spectrum covered by the topic, makes impossible to
have a complete bibliography, which is only introductory and depends on
the a u t h o r s ' preferences. T h e book is of wide audience and it is good for
researchers, undergraduate and graduate students, courses and seminars.
Oradea, August 1, 2001 The authors
Contents
Preface vii
Chapter 1 Introduction 1
1.1 General Description of the Topic 1
1.2 On Chapter 2: Defect of Property in Set Theory 3
1.3 On Chapter 3: Defect of Property in Topology 8
1.4 On Chapter 4: Defect of Property in Measure Theory 18
1.5 On Chapter 5: Defect of Property in Real Function Theory . . 24

1.6 On Chapter 6: Defect of Property in Functional Analysis . . . 31
1.7 On Chapter 7: Defect of Property in Algebra 45
1.8 On Chapter 8: Miscellaneous 50
Chapter 2 Defect of Property in Set Theory 57

2.1 Measures of Fuzziness 57
2.2 Intuitionistic Entropies 67
2.3 Applications 76
2.3.1 Application to determination of degree of interference 76
2.3.2 Application to description of the performance of systems 81
2.3.3 Application to digital image processing 84
2.4 Bibliographical Remarks 86
Chapter 3 Defect of Property in Topology 87

3.1 Measures of Noncompactness for Classical Sets 87
3.2 Random Measures of Noncompactness 95
3.3 Measures of Noncompactness for Fuzzy Subsets in Metric Space 98
ix
x Contents
3.4 Measures of Noncompactness for Fuzzy Subsets in Topological

Space Ill
3.5 Defects of Opening and of Closure for Subsets in Metric Space 125
3.6 Bibliographical Remarks and Open Problems 127
Chapter 4 Defect of Property in Measure Theory 131

4.1 Defect of Additivity: Basic Definitions and Properties 131
4.1.1 Application to calculation of fuzzy integral 142
4.1.2 Application to best approximation of a fuzzy measure . 145
4.1.3 A metric on the family of fuzzy measures 146
4.2 Defect of Complementarity 148
4.3 Defect of Monotonicity 164
4.4 Defect of Subadditivity and of Superadditivity 173
4.5 Defect of Measurability 181
Chapter 5 Defect of Property in Real Function Theory 185

5.1 Defect of Continuity, of Differentiability and of Integrability . 185
5.2 Defect of Monotonicity, of Convexity and of Linearity 193
5.3 Defect of Equality for Inequalities 200
Chapter 6 Defect of Property in Functional Analysis 211

6.1 Defect of Orthogonality in Real Normed Spaces 211
6.2 Defect of Property for Sets in Normed Spaces 234
6.3 Defect of Property for Functionals 246
6.4 Defect of Property for Linear Operators on Normed Spaces . . 252
6.5 Defect of Fixed Point 260
Chapter 7 Defect of Property in Algebra 277

7.1 Defects of Property for Binary Operations 277
7.2 Calculations of the Defect of Property 289
7.3 Defect of Idempotency and Distributivity of Triangular Norms 293
7.4 Applications 300
Chapter 8 Miscellaneous 305

8.1 Defect of Property in Complex Analysis 305
Contents xi
8.2 Defect of Property in Geometry 307

8.3 Defect of Property in Number Theory 320
8.4 Defect of Property in Fuzzy Logic 327
Bibliography 337
Index 349
Chapter 1
Introduction
To convey some of the essence of this monograph to the reader, in this

chapter we briefly present some i m p o r t a n t motivation for writing it and its
main results. For the convenience of the reader, the results are numbered
as they are in their respective chapters.
1.1 General D e s c r i p t i o n of t h e Topic

It is well-known t h a t the definition of a m a t h e m a t i c a l concept consists in

the s t a t e m e n t of one or several properties (axioms) t h a t must be verified
by some m a t h e m a t i c a l objects. More exactly, we can describe this by the
followings.
Let U be a given abstract set of elements and let us denote by P a
specific property of some elements in U. Obviously, P divides U into two
disjoint sets:
Up = {x 6 U; x has the property P}
and
Up- = {x £ U; x does not satisfy the property P} .
A powerful tool of study of Up and U-p might be the introduction (not

necessarily in an unique way) of a quantity E (x), defined for all i £ [ ' with
values in a normed space (Y, ||-||) (which in general is R or C ) , having the
property
x G Up if and only if E (x) = Oy.
1
2 Introduction
In this case, for x £ Up- we necessarily have E (x) ^ Oy and consequently

the quantity 11Ê7 (a?)11 can be considered t h a t it measures the "deviation" of
x from t h e property P.
We will call \\E (x)\\ as defect of property P for the element x £ U.
As a consequence, the following kind of application holds: given a family
of operators A\ : U -> Y, VA £ T, which satisfy A\ (x) = Oy, Vz £ Up, A £
T, another family of operators B\ : Y —> Y, A £ T, can be constructed such
t h a t Bx (Oy) = Oy,VA £ T and A A (z) + Bx (E (x)) = Oy.Vz £ U,X £ T
(or more general, P A ( X ) | | < C \\BX (E (x))\\,~ix £ [7, A £ I \ with C a
positive constant). In other words, a formula valid for all x £ £/p, can be
transformed by a "disturbance" factor (which depends on E(x)), into a
more general formula, valid for all x £ U.
Many well-known concepts in various fields of m a t h e m a t i c s can be de-
scribed by this scheme: the measures of noncompactness (defects of com-
pactness, in our terminology) t h a t measure the deviation of a classical set
in a topology from the property of compactness, the measure of nonconvex-
ity (defect of convexity, in our terminology) t h a t measures the deviation of
a set in normed linear spaces from the property of convexity, the moduli
of oscillation and of continuity (defects of continuity, in our terminology)
t h a t measure the deviation of a function from the property of continuity,

the minimal displacement of points under mappings (defect of fixed point,
in our terminology) t h a t measures the deviation of a m a p p i n g from the
property of fixed point, the areolar derivative (defect of holomorphy, in our
terminology) t h a t measures the deviation of a complex function from the
property of holomorphy, and so on.
Beside these, in this monograph we introduce and study m a n y other
defects of property, as follows: measures of noncompactness for fuzzy sets,
fuzzy and intuitionistic entropies, defect of additivity, subadditivity, su-
peradditivity, complementarity, monotonicity for set functions, defect of
monotonicity, convexity (concavity), differentiability, integrability for real
functions, defect of equality for inequalities, defect of balancing, absorption,
orthogonality for sets in normed linear spaces, defect of sublinearity for
functionals, defect of symmetry, permutability, derivation for linear opera-
tors, defect of commutativity, associativity, identity element, invertibility,
idempotency for binary operations, defect of orthogonality and parallel-
ness in Euclidean and non-Euclidean Geometries, defect of curvature and
of torsion in Geometry, defects of properties in Number Theory, defects of
On Chapter 2: Defect of Property in Set Theory 3
properties in Fuzzy Mathematics.

Many applications and open problems also are presented.
Because of the great variety, it would be impossible to make a very
deep study of all the above concepts by this book. This task is left in many
cases to the reader, the main goal of the book being only to put in evidence
a method of research called quantitative study of the defect of property,
which gives the opportunity to examine from the same viewpoint, basic
concepts in various fields of mathematics.
1.2 On Chapter 2: Defect of Property in Set Theory
In this chapter we consider the measures of fuzziness as measuring the " de-
viation" of a fuzzy set from the concept of classical set, that is as measuring
the defect of crisp (classical) set and the intuitionistic entropies as measur-
ing the "deviation" of an intuitionistic fuzzy set from the concept of fuzzy
set, that is as measuring the defect of fuzzy set. Then, some applications
to the determination of degree of interference (mainly in the geography of
population), to description of systems performance and to digital image
processing are given.
We denote by FS(X) - {A \A : X ->• [0,1]} the class of all fuzzy sets

onX.
A general definition of the measure of fuzziness is the following (see
Rudas-Kaynak [181]):
Definition 2.1 A measure of fuzziness is a positive real function dc defined
on T(X) C FS(X), that satisfies the following requirements:
(i) IfA<ET{X),A(x) £ {0, l},VxeX then dc(A) = 0.
(ii) If A -< B then dc (A) < dc(B), where A -< B means that A is
sharper than B.
(Hi) If A is maximally fuzzy then dc (A) assumes its maximum value.
Let us consider (X, A, p) a measure space and let us denote
TA (X) = {A G FS(X); A is ^-measurable} .
Definition 2.4 Let (X,A) be a measurable space and t £ (0,1). A t-

measure of fuzziness is a function rf' : TA P O —> R that satisfies the
following conditions:
4 Introduction
(t) If A(x) e {0,1} , Vz G X then d£ (A) = 0;

(ii) If A -<t B then <PC{A) < 4 ( 5 ) > where yl -<t B if and only if
A(x) < B{x) for B{x) < t and A(ar) > B(x) for B ( x ) > t;
(Hi) If A(«) = t,Vx £ X then <4(^4) i s the m a x i m u m value of d],.
D e f i n i t i o n 2.5 Let (X,A, /i) be a measure space and t G (0, 1) . A function

5* : .T7^ ( X ) —> R t h a t satisfies the following conditions:
(i) sÂ) = 0 if and only if A(x) e {0,1} , ji-a.e i £ X ;
(«) If ,4 -<t B then s« (A) < 4 ( 5 ) ;
( m ) «c(^4) is the m a x i m u m value of s* if and only if A(x) — t,/j,-a.e.
x £l,
is called strict t-measure of fuzziness with respect to fi.
D e f i n i t i o n 2.6 Let N be a fuzzy complement. A ^-measure of fuzziness

is called symmetrical with respect to N (or ./V-symmetrical) if dÂ) =
dtc[AN),\tAeTA{X).
A family of ^-measures of fuzziness is given by
Theorem 2 . 3 Let X be a finite set, t 6 (0,1) and h : R + - ^ R + increasing,

X
such that h(0) = 0, [gx)x^x <9 '• P , 1] —>• R + increasing on [0,t] and
decreasing on [t, 1], such that gx(0) = gx(f) = 0,Var 6 X and gx(t) is the
maximum value of all functions gx,x 6 X. The function d%c : FS(X) —> R
defined by
dtc(A) = h("£9AM^)))
is a t-measure of fuzziness. If, in addition,
gx(a) = gx (N(a)) ,\/a e [0,1],\/x £ X,
where N is a fuzzy complement, then d^ is a N-symmetrical t-measure of

fuzziness.
Under additional conditions, we obtain a family of strict ^-measures of

fuzziness.
sa
T h e o r e m 2 . 4 / / the functions h and [gx)xeX tisfy the hypothesis in
are
Theorem 2.3 and, in addition, h is strictly increasing, [gx)x^x strictly
increasing on [0,t] and strictly decreasing on [t, 1], then the function s* :
On Chapter 2: Defect o] Property in Set Theory 5
FS(X) -> R defined by
st(A) = h t ^ gx (A(x)))
\x€X )
is a strict t-measure of fuzziness which is N-symmetrical if ^(o) —

gw (N(a)), Va £ [0,1], Vz £ X, where N is a fuzzy complement.
Other families of f-measures of fuzziness can be obtained by using the

concept of i-norm function in the sense o f V i v o n a [217].
D e f i n i t i o n 2 . 7 Let t £ (0,1) and (X,A,n) be a measure space. An A-

measurable function with respect to the first variable, <pt : X x [0,1] —>• [0,1]
t h a t satisfies the properties
(i) (pt{x,0) = <pt(x,l)=0,\/x€X;
(ii) <pt(x, •) is increasing on [0,2] and decreasing on [t, 1];
(Hi) ipt(x, t) = 1, Var G X,
is called tf-norm function.
T h e o r e m 2.5 Let t G ( 0 , 1 ) , (X,A,fi) be a measure space and <pt : X x

[0, 1] —> [0, 1] a t-norm function. The function df1 : TA (X) —k [0, 1] defined
by
^(A)=Jlpt(x,A{x))d^
x
is a t-measure of fuzziness. If N : [0, 1] —> [0, 1] is a fuzzy complement and
the t-norm function ipt verifies
ipt {x, N(a)) = p , ( i , o ) , V i e X, Va G [0,1],
then the t-measure of fuzziness dfi(A), defined as above, is N -symmetrical.
An intuitionistic fuzzy set (see e.g. Atanassov [5]) A on X, is an object

having the form
A = {(x, fjtA(x),i/A(x)) : x G X) ,
where the functions \iA,vA : X —> [0,1] define the degree of membership
and the degree of non-membership of the element x G X to the set A C X,
respectively, and for every x £ X,
HA{x) + vA(x) < 1.

6 Introduction
We denote by IFS(X) the family of intuitionistic fuzzy sets on X.

Let us consider (X, A, m) a finite measure space and let us denote by
TA (X) the family of all ^-measurable intuitionistic fuzzy sets on X.
Definition 2.9 A real function df : IA (X) —> R+ is called an intuitionistic

entropy (on XA (X)) if the following properties are satisfied:
(i) df(A) = 0 if and only if PA(X) + Â(X) = 1, m-a.e. x G X;
(ii) df(A) is maximum if and only if HA(X) = VA(X) — 0, m-a.e. x G X;
(Hi) df(A) = df(A),\/AeIA(X);
(iv) If A, Be XA (X) and A df(B).
Definition 2.10 Let (X,A, m) be a measure space and let us denote

D = {(p, v) G [0,1] x [0,1] : /i + 1/ < 1} . An intuitionistic norm function is
an „4-measurable function with respect to the first variable, $ : X x D —>
[0,1] with the following properties, for every element x £ X :
(i) $ (x,p, v) = 0 if and only if p + v — 1;
(ii) $ (x, p, v) = 1 if and only if p = v = 0;
(Hi) $(x,n,v) = <b(x,v,n)\
(iv) If n < n' and v < v1 then $ (x, p,v)>3> (x,p', v1) .
Theorem 2.6 Let (X,A,m) be a finite measure space and <S> an intu-
itionistic norm function. Then, the function dj : XA (X) —>• [0,1] defined
by
d
1Â) = ~ilF\l ${x,HA(x),vA(x))&m
with A — {(x, PA(X), VA(X)) • x G X}, is an intuitionistic entropy.
The intuitionistic entropies can be characterized by using Iv-functions.
Definition 2.11 Let (X,A, m) be a measure space and ip : [0,1] ->• [0,1]
be a continuous function such that if a + (5 < 1 then <p(a) + <p(f3) < 1. The
function 1^ : XA (X) —\ [0,1] denned by
I v =
^ m(X)J i1 ~ • [0,1]
be a continuous function and dv : XA (X) —>• [0,1]. The function dv is an
On Chapter 2: Deject of Property in Set Theory 7
intuitionistic entropy and an Iv -function if and only if
d A
f( ) = T77JK J (1 - <P ( M * ) ) - <f Mx))) dm,
where <p satisfies the conditions:

(i) <p is increasing;
(ii) <p(a) = 0 if and only if a = 0;
(in) <p(a) + ip(/3) = 1 if and only if a + (3 = 1.
Between intuitionistic fuzzy measures and intuitionistic entropies we can
establish the following connection.
Theorem 2.9 Let (X,A,m) be a measure space. If d^ : 1A (X) ->• [0,1]
is an intuitionistic entropy and an I^-function then dv is an intuitionistic
fuzzy TM-tneasure on 1^ (X), where TM is the triangular norm given by
TM (x,y) = mm(x,y). Moreover, if ip is additive then dv is an intuition-
istic fuzzy h-measure on 1^ (X), for every continuous triangular norm h
which verifies h (x,y) + hc (x,y) = x + y,Vx,y £ [0,1].
At the end of this chapter we give some applications to the determina-

tion of degree of interference, applications to the geography of population
and to description of the systems performance. Also, a method that permits

to obtain a binary digital image from a fuzzy digital image is presented.
For example, let us consider V a country and let us denote by X =
{Si, ...,Sn} a partition of V, that is Si, i £ {1,..., n} represent all the dis-
tricts of V. We can introduce a normal indicator Tu,x (that is 0 < Tu,x <
1) which estimates the degree of homogeneity of the territorial distribu-
tion of population on V, with respect to the organization corresponding to
X. We notice that, the density of population, the usual indicator used in
this situation, is a global indicator which does not take into account the
pointwise aspects. For example, if there exists a great concentration of
population in certain zones, this indicator is not significant.
Let us denote by Tme,i, the density of population in V, that is
_ total population in V
'-meat — "FT} >
area of V
by TMAX, the maximum possible density of population in a district of V,
that is
total population in V
l-MAX =
area of the smaller district S;
8 Introduction
and by Tj, the density of population in S,,i € { 1 , . . . , n } , t h a t is
_ population in Si
area of Si
We define the fuzzy set P : X —>• [0,1] by
P(Si) = ¥ ^ - ,
1 MAX
?<J
and we denote t = J "* .
1 MAX
We p u t T ^ x — s^P), where s£ is a strict ^-measure of fuzziness in
normalized form. Following the axioms of strict ^-measures of fuzziness
(see Definition 2.5), we get properties of the indicator Tv,x which are in
concordance with our intuition:
( 0 1 ) Tv,x has the m i n i m u m value (equal to 0) if and only if the entire
population of V is completely situated in the district with the smallest area;
( 0 2 ) If Tj <Ti <t or t <Ti < Tj then the contribution of the district
Si to IV, x is greater t h a n the contribution of the district Sj;
( 0 3 ) Tx,u has the m a x i m u m value (equal to 1) if and only if in all
districts of V the density of population is equal to Yme<i.

Choosing a suitable strict t-measure of fuzziness we can get interesting
results concerning the territorial distribution of population.
1.3 O n C h a p t e r 3: D e f e c t o f P r o p e r t y i n T o p o l o g y
This chapter mainly discusses the concept of measure of noncompactness

(defect of compactness, in our terminology) in classical setting, in r a n d o m
setting and in fuzzy setting.
Let (X, p) be a metric space and let us denote
Vb (X) = {Y C X; Y jL 0, Y is bounded} .
D e f i n i t i o n 3.1 (i) (Kuratowski [129], [130]). T h e Kuratowski's measure

of noncompactness for Y EVb (X) is given by
a(Y) =inf {e > 0;3n <=N,Ai eX,i = T~n with

Y A and
C Ur=i i diam (A) < £} ,
where diam (Ai) = sup {p (x, y) ;x,y £ Ai} ,i = 1, n.
On Chapter 3: Deject of Property in Topology 9
(ii) (see e.g. Istratescu [l05]) The Hausdorff's measure of noncompact-

ness of Y C X is given by
ft(Y) = inf | e > 0 ; 3 n € N , j f c £ Y, i = l~n~ with Y C \jB(y{,e)

I »=i
T h e o r e m 3.1 (i) (see e.g. Rus [182], p.85-87 or Banas-Goebel [32]) Let
X,Y £Vb(X). We have:
0<a(A) < diam (A);
A C B implies a (A) < a(B);
a (A) — a (A) and moreover a (A) — 0 if and only if A is compact;
a {V£ (A)) < a (A)+2e, where Ve (A) = {x eX;p (x, A) < e} and p (x, A) =
vai{p{x,y) ;y€ A};
a(AUB) = max{a(A) ,a(B)};
a(AnB) < min{a{A),a{B)};
Let At C X, A'+i C Ai be with A{ closed and nonempty, i = 1,2,....
/ / linin-yoo a (An) = 0 then P ^ L i ^ n zs nonempty and compact.
If, in addition, X is a Banach space, then a (A + B) < a (A) + a (B) ,
a (cA) = |c|a(A) , c £ R, a (convA) = a (A), where convA is the convex
hull of A eVb{X).
(ii) (see e.g. Beer [36], Banas-Goebel [32])
h (A) = 0 if and only if A is totally bounded (that is, Ve > 0, 3x\,..., xp £ A
such that \fx £ X, 3a;,- with p (x, x{) < e);
AC B implies h (A) < 2ft ( 5 ) ;
h(A)=h{A);
h{Al>B) < max {ft (A), ft ( B ) } ;
ft is continuous on CL(X) = {Y C X; Y is closed, Y ^ 0} with respect to
Hausdorff topology, i.e. if Yn,Y £ CL(X), n = 1, 2,..., satisfy
DH (Yn,Y) —} 0 (where DH is the Hausdorff-Pompeiu distance), then
lim n ^ 0 O ft(y n ) = h (Y). Also, ft is upper semicontinuous on CL(X) with
respect to the so-called Vietoris topology.
In addition,
h(A) <a (A) < 2ft (A) ,VAePb (X)
and if moreover X is Hilbert space, then
V2h {A) <a{A)< 2ft (A), VA £ Vb (X).

10 Introduction
The properties in Theorem 3.1 suggest an axiomatic approach in Banach

spaces [X, ||-||), as follows. Firstly we need the notations:
RC {X) = {Y C X; Y ^ 0, Y is relatively compact} ,

CO{X) = { Y C X ; Y ^ 0 , Y is compact}.
Definition 3.2 (Banas-Goebel [32]). K C RC{X) is called kernel (of a

measure of noncompactness) if it satisfies:
(i) A G K, implies A G K,;
(«') AGIC,BCA,B ^) A £ K. implies conwA G /C;
(u) The set K,c = {A G /C; A is compact} is closed in CO(X) with re-
spect to Hausdorff topology (i.e. the topology induced on CO(X) by the
Hausdorff-Pompeiu distance).
Definition 3.3 (Banas-Goebel [32]). The function fi : Vb{X) -> [0,+oo)

is called measure of noncompactness (or defect of compactness, in our ter-
minology) with the kernel K (denoting ker/i = K-) if satisfies:
(i) fi(A)=0 if and only if A G K;
(ii) n(k)=n(A);
(in) /i (convA) — \i (A);

(iv) Ac B implies fj, (A) < \x (B);
(v) n {XA + {1~X)B)< Xn (A) + (1 - A) ii (B) , VA G [0,1];
(vi) If A„ eVb{X),An = An and An+1 cAn,n= 1,2,... and if
limôo// (An) = 0, then | X L i ^ " ^ 0-
If /C = i?C(X) then // will be called full (or complete) measure.
Many examples satisfying Definitions 3.2, 3.3 are given.
Let 7r be the set of all open coverings of a topological space X. Let us
consider the family C = 2W of all subsets of K ordered by the relation
x < y if and only if x D y, for x,y G C.
Then (C, <) is a complete lattice with the minimal element 0 = w. Now
we can define the C-measure of noncompactness as follows:
Definition 3.4 (Kupka-Toma [128]). If A C X then the measure of non-
compactness of A is the element of C defined by
m(A) - h> e 7r 3 finite 8 CV : A C \Js\ •
On Chapter 3: Defect oj Property in Topology 11
Theorem 3.2 (Kupka-Toma [128]). The mapping m : 2X —> C has the

following properties:
(i) If A is a compact subset of X then m (A) = 0 . If A is closed then
the converse is true;
(ii) If AC B then m(A) <m (B) ;
(tit) m (A U B) = m (A) n m (B) = sup {m (A), m {B)} ;
(iv) m(Ar\B) < m(A)Um(B) =inf{ra(A) ,m(B)}.
Theorem 3.3 (Kupka-Toma [128]). Let X be a complete uniform space.
Then we have the followings:
(i) For every closed subset A C X, m (A) = 0 if and only if $ (A) = 0;
(ii) For every decreasing net (A^ : 7 € T) of closed nonempty subsets of
X,
inf{$(A7) : 7 e r } = O ^ l i m m ( A 7 ) = 0.
•yev
(Here $ denotes Lechicki's measure in Lechicki [132]).
Theorem 3.4 (Kupka-Toma [128]). Let (A1 : 7 £ T) be a decreasing net
of closed nonempty subsets of a topological space X. Then the following
implication is true:
limm (Ay) = 0 => A = ( | Ay is nonempty and compact.
Let us denote by A the set of all nondecreasing and left continuous

functions / : R —> [0,1], such that / (0) = 0 and linx^+oo/ (x) — 1-
Definition 3.5 (Menger [145]). A probabilistic metric space (PM-space,
shortly) is an ordered pair (S,F), where S is an arbitrary set and F :
S x S —> A satisfies:
(i) FPtq (x) — 1, Vx > 0 if and only if p — q;
(ii) FPiq (x) = Fq>p (x) ,Vp,q£ S;
(in) If FPig (x) = 1 and F g , r (y) = 0, Vp, q,r£ S then Fp>q (x + y) = 1.
(RexeFPiq(x)=F(p,q)(x)).
Definition 3.6 (Egbert [69]). Let A C S,A ± 0. The function DA(-)
denned by
DA (X) = sup inf FPiq (x) , x £ R
t<xP,i£A
is called the probabilistic diameter of A. If sup {DA (X) ; x £ R } = 1 then
A is called bounded.
12 Introduction
Definition 3.7 (Boc§an-Constantin [41], see also Istratescu [105], Constan-

tin- Istratescu [58]). Let A be a bounded subset of S. The mapping
a A (x) = sup [s > 0;3n 6 N, A{,i = 1, n with

A
= U = i Ai and D
A , {x)>e> ,
where x G R, is called the random Kuratowski's measure of noncompact-

ness.
Theorem 3.6 (Boc§an-Constantin [41]). We have:

(i) aA e A;
(ii) a A (x) > DA (X) , Vx G R ;
(Hi) If 0 ^ AC B then aA (x) > aB (x) , Vx G R ;
(iv) aAuB (x) = min {aA (x), aB (x)} , Vx G R ;
(v) aA (x) = a-^-(x), where A is the closure of A in the (e, X)-topology of
S (where by (s, X)-neighborhood of p G S, we understand the set Vp (e, A) =
{qeS;FPtq> l-A},e>0,AG[0,l]).
Theorem 3.8 (Boc§an-Constantin Ul])- Let (S,p) be an usual metric
space and (S,F) the corresponding PM-space generated by (S,p).
(i) A C (S, F) is probabilistic precompact if and only if a A (X) —

H(x),\/x£ R;
(ii) A C (S, p) is precompact if and only if A is probabilistic precompact
subset of (S, F);
(Hi) For any bounded A C (S, p) we have
OLA (x)=H(x-a (A)) , V i e R ,
where a (A) is the usual Kuratowski's measure of noncompactness in (S,p)

and a A (x) is the random Kuratowski's measure of noncompactness in the
generated PM-space (S, F).
Definition 3.9 (Menger [145]). (S, F) is called PM-space of Menger-type

with the t-norm T, if F satisfies the first two properties in Definition 3.5
and the third one is replaced by
Fv,d (x + y)>T [FPir (x), Fr>(1 (x)), Vp, q,reS.
We denote it as the triplet (S, F,T).
On Chapter 3: Defect of Property in Topology 13
D e f i n i t i o n 3 . 1 0 (see e.g. Istratescu [105]). If (S,F,T) is a P M - s p a c e

of Menger-type and A, B C S, then t h e probabilistic (random) Hausdorff-
Pompeiu distance between A a n d B is given by
D^B ix) — s u p T ( inf s u p F p „ (t) , inf s u p F p „ (t) I , x G R .

t<x \P£Aq€B ' ItBpeA ' J
D e f i n i t i o n 3 . 1 1 (see e.g. Istratescu [105]). Let (S,F,T) be a P M - s p a c e
of Menger-type a n d A C S, bounded. T h e r a n d o m Hausdorff's measure of
noncompactness of A is given by
HA {x) = sup {s > 0; 3 finite Fe C A such t h a t D^pc (x) > e} .
Further, we extend t h e above concepts a n d results t o fuzzy subsets of usual

metric spaces.
D e f i n i t i o n 3 . 1 2 T h e sets
Go {<PA) = {(a;, y); 0 < y = ipA (x), x G X} = Graph (F)
and
HG0 (<pA) = {(x, y);0<y<<pA(x),x£X} = hypo (F) n (A x (0, 1])

are called t h e support graph a n d t h e support hypograph of (A, ^ A ) respec-

tively, where F : A ->• ( 0 , 1 ] , F (x) = <pA (x), Va; G A.
T h e diameter a n d t h e hypo-diameter of t h e fuzzy set (A, <PA) are given by
D(<pA) - sup {d* (a,b);a,b£Go {<PA)}
and
hD (<pA) = sup {d* (a, b);a,be HG0 (<PA)} ,
respectively. If D (<PA) < + ° ° {hD(<pA) < + o o ) we say t h a t (A,<PA) is

bounded (hypo-bounded). Here d* : X x [0,1] -¥ R + , where (X, d) is metric
space, is given by d* ( ( x i , r i ) , (x2,r2)) = m a x { d (xlt x2) , \ri - r2\} .
D e f i n i t i o n 3 . 1 3 Let [A,IPA] be bounded (or hypo-bounded, respectively).

T h e Kuratowski's measure and t h e Kuratowski's hypo-measure of noncom-
pactness of (A, ^ A ) are given by
K (<pA) - inf [e > 0;3n G N , (At, <pA,) with D(<pA, )<e,i= l,n,
such t h a t <PA {X) 0;3n eN,(Ai,<pAi) with hD(<pAt) <e,i = T~h~,

such that f A (z) 0;3n£ N,3Pi = (xi,ri) E G0 {<pA),i = T~n~,

such that VP = (x, r) £ G0 {<PA) , 3P* with d* (P, Pk) < e}
and by
hH (<pA) = inf{e>Q;3ne'N,3Pi = (xi,ri)eHGo(<PA),i = ^n,

such that VP = (x, r) £ HG0 (<pA) , 3Pk with d* (P, Pk) < s}
respectively.
T h e o r e m 3.9 (i) K (<pA) < hK (ipA) •

(ii) K {ipA) < K* (Go (ipA)), H {<pA) = H* (Go (<PA)) , hH (<pA)

= H* (HGo {'PA)), where K* and H* represent the usual Kuratowski's and
Hausdorff's measures of noncompactness (respectively) of the subsets in the
metric space (X x [0,1] ,d*). (Here (A, <PA) is considered hypo-bounded).
T h e o r e m 3.10 Let (A, <pA) C (B,ipB) be two I-bounded fuzzy subsets of

(X,d). We have:
(i) K {<pA) < D {<pA) .
(ii) (A, (pA) C (B,(pB) implies K (<pA) < K(<pB), where (A,<pA) C
(B, <ps) means ipA (x) < ipB (x), Vx 6 X.
(Hi) K (<pA V <PB) = ma,x{K (tpA), K (ips)}, where (<pA V tpB) (x)
= max{(y^)(a;) ,(<pB) (x)} ,*tix EX.
T h e o r e m 3.11 (i) hH (<PA) = + ° ° if and only if HGo (<PA) is unbounded.

(ii) If (A,<pA) C (B,ipB) then hH (tpA) < 2hH (>pB) .
(Hi) hH (<PA) = 0 if and only if HGo (<PA) is totally bounded.
(iv) hH (ipA V ipB) < ma,x{hH (ipA), hH (fB)}, where (<pA V tpB) (x)
= max {ifiA (x), <pB (x)} , Va: £ X.
By using the level sets method, we can introduce
On Chapter 3: Defect of Property in Topology 15
Definition 3.14 The (a)-Kuratowski measure of noncompactness of (A, <pA)

is given by
aK(pA) = sup{Ko(Ax);\€(0,l}},
where A\ = {x £ X; ipA (x) > A} a n d K$ is the usual Kuratowski's measure
of noncompactness of usual subsets in (X, d).
T h e (a)-Hausdorff measure of noncompactness of (A, >pA) is given by
ocE (<pA) = sup {H0 {Ax) ; A 6 (0,1]} ,
where HQ represents t h e usual Hausdorff's measure of noncompactness of

usual subsets in (X, d). Obviously, aH (ipA) can take the + o o value.
D e f i n i t i o n 3 . 1 5 (see e.g. Weiss [220]) If (X, d) is a metric space, then t h e

induced fuzzy topology on (X, d) is t h e collection of all fuzzy sets of (X, d)
with <PA '• X —> [0,1] lower semicontinuous on X.
A fuzzy set (A,<pA) is called (a)-bounded if for each A £ (0,1],-A* is
bounded in t h e metric space (X, d).
A fuzzy set (A, tpA) is called (a)-compact if for each A £ ( 0 , 1 ] , A\ is com-
pact in t h e metric space (X, d).
T h e o r e m 3 . 1 3 Let {A,tpA) ,{B,<PB) be (a)-bounded.

(i) otK (<PA) — 0 if and only if each Ax, A £ (0,1] is compact in (X, d),
where Ax denotes the closure of Ax in (X,d). Also, if (A,<PA) is in-
compact then aK (PA) = 0.
(it) If (A,ipA) C (B,<pB) (i-e. fA (x) < <PB (X) ,Var £ X) then
aK (ipA) < aK (ipB).
(Hi) aK (<pA) < Ko(A), where KQ represents the usual Kuratowski's
measure for subsets in (X,d), and in general we have no equality.
(iv) aK (ipA V ipB) = m a x { a A " (ipA) , aK (<PB)}, where (<pA V <pB) (x) —
m a x {<pA (x) , <pB (x)} , \/x £ X.
(v) aH (<pA) = 0 if and only if each Ax,X £ (0,1], is totally bounded.
(vi) If (A, <pA) C (B, <pB) then aH (<pA) <2aH {ipB).
(vii) aH (<pA V ipB) < max{aH (<pA) ,aH (tpB)} •
In what follows one consider measures of noncompactness of fuzzy sub-

sets in (fuzzy) topological spaces.
D e f i n i t i o n 3 . 2 1 Let (X, T) be a classical topological space a n d A a fuzzy

subset of X (i.e. A : X —> [0,1]). Then, t h e (a)-measure of noncompactness
16 Introduction
of the fuzzy set A will be
aM(A) = f] m(Aa),
oe(o,i]
where Aa = {x £ X;A (x) > a } and aM : Ix —> C, Ix denoting the class

of all fuzzy subsets of X.
Definition 3.22 Let Ai G TF,i G / and Al,...,An G 7>. The family
1
K = {(^4i)j e / , -A , •••, ^4.™} of fuzzy sets is an open J-cover of a fuzzy set A
if
A C {\>ieIAi)7Als...IAn,
where ( V , e / A ) (x) — sup {A2- (a;) ; i € 1} , x E X and i C B means A (x) <

B(x),\/x€X.
We say that < (Aj) eJ , Akl, ...,Ahp > is a finite open J-subcover in K, of
the fuzzy set A, if A C (VjejA,) s'J4 s'...3'_A's'', where J C J is finite and
fel
{ki,...,kp} C { l , . . . , n } .
Definition 3.23 A fuzzy set A G Ix is (C, s)-compact if each open iT-cover
of A has a finite J-subcover of A. A fuzzy set ^4 is (£, ?)-compact if for all
open s-cover of A and for all e > 0, there exists a finite ?-subcover of B,
where B is the fuzzy set defined by B (x) — max (0, A (x) — e) , V* G X.
Definition 3.24 Let A G Ix ,(X,Tp) be a quasi-fuzzy topological space
and s a triangular conorm. The (C, s)-measure of noncompactness of A is
the element of Cs defined by
m,Q (A) = {V G nS; 3S C V, finite open ?-subcover of A] .
The (L, s)-measure of noncompactness of A is the element of C 5 defined by
msL {A) = {v, G TTS;V£ > 0,3S* C P*, finite, 6, = {(Aj)jeJ,A\ ...,AP}
with max{0,^(a:) - e] < ({VjeJAj)sA1s...sAp) (x) ,Vx G X) .
T h e o r e m 3.15 Let (X,T) be a topological space. The mapping aM :

Ix —> C has the following properties:
(i) If A G Ix is compact in the sense of Weiss [220], Definiti on .i.o,
then aM (A) = 0c- If A is closed (in the sense of Proposition 3.3 in Weiss
On Chapter 3: Deject of Property in Topology 17
[220]) and aM (A) = Oc, then A is compact (in the sense of Definition 3.5
in Weiss [220]).
(ii) If ACB then aM (A) < aM (B).
(in) aM {A V B) = aM (A) D aM (B).
(iv) aM (A A B) < aM (A) U aM (B).
(Here {A V B) (x) = max {A (x), B (x)} ,{AAB) (X) = min {A (x) , B (x)} ,
Vz G X).
Theorem 3.16 Let (X,Tp) be a quasi-fuzzy topological space. The map-

ping m^ : Ix —• Cs has the following properties:
{i) If Ae Ix is (C,s)-compact then m^ (A) = 0 | .
(ii) If AC B then mg (A) < msc (B) .
(Hi) msc (AV B) = m£ (A) n mg (B) .
(iv) msc (AAB) < msc (A) U msc (B).

ping mf : Ix —> Cs has the following properties:
(i) If Ae Ix is (L,T) -compact then m£ (A) = 0 | .
(ii) If ACB then ml (A) < msL (B) .
(Hi) m[ (AV B) = mf (A) n mf (B) .
(iv) mf (AAB) < mf (A) U mf (B).
Applications to upper semicontinuity of fuzzy multifunctions also are

presented.
At the end of this chapter one present:
Definition 3.34 Let (X, p) be a metric space and D (A, B) be a certain

distance between the subsets A,BcX. For Y C X, we call:
(i) defect of opening of Y, the quantity dop (D) (Y) = D (Y, intY) ;
(ii) defect of closure of Y, the quantity dCL (D) (Y) = D (Y,Y) .
Theorem 3.23 1/ (X,p) is a metric space and Y € Vt, (X), then:

(i) dop (D*) (Y) = 0 ) / and only if Y is open;
(ii) dcL (-D*) (Y) = 0 if and only if Y is closed.
In addition,
dCL(D*)(\Y) = \dCL(D*)(Y),V\>0;
dOP(D*)(\Y) = Ad O p( J D*)(Y),VA>0.
(Here D* is a special distance).
18 Introduction
1.4 On Chapter 4: Defect of Property in M e a s u r e Theory
This chapter introduces and studies some defects of property for set function
(especially, for fuzzy measures): defect of additivity, defect of complemen-
tarity, defect of monotonicity, defect of subadditivity and of superadditivity.
Also, in Section 4.5, the defect of measurability for sets is discussed.
Let A be a cr-algebra of subsets of X.
Definition 4.1 (see e.g. Ralescu-Adams [168]) A set function fi : A —t

[0, oo) is said to be a fuzzy measure if the followings hold:
(»)M0) = O;
(it) A, B G A and AC B implies p (A) < /* (B) .
Definition 4.2 The defect of additivity of order n, n £ N, n > 2, for the
measure fi is given by
an M = sup {hj (Ui=1Ai) - YTi=i» {Ai)\ • A{ e A,Vi G N,

Ai nAj = V),i^ j}
The defect of countable additivity for the measure /J. is given by

aoo (/i) = sup { ^ ( U S i A - ) - E ^ i M A - ) l : At G A,Vi G N,

AiHAj = 0 , j ^ j }
Theorem 4.1 (i) an (/«) < an+\ (//), Vn > 2;
(n) If n is a continuous from below fuzzy measure then a^ (p)
= lirr^ôoan (/*) ;
(m) a„ (//) < a n _x (/i) + a 2 (//) , Vn > 3;
(iv) an (/J.) < (n — 1) a 2 (p) , Vn > 3;
(v) a n ( M ) < ( n - l ) A i ( A - ) , V n > 2 ;
(•ui) / / // is a superadditive fuzzy measure then an (/i) < /j. (X) , Vn > 2.
An important class of fuzzy measures are the A-additive fuzzy measures.
Definition 4.3 (see e.g. Kruse [126]) Let A G ( - 1 , oo), A ^ 0 and A be a
cr-algebra on the set X. A continuous fuzzy measure \x with fJt(X) — 1 is
called A-additive if, whenever A, B G A, A f) B = 0,
li{A UB)=n{A) + n(B) + X/i {A) n{B).
Theorem 4.2 (Wierzchon [222]) Let m be a classical finite measure,

m : A —> [0, oo). Then \i = t o m is a X-additive fuzzy measure if and only
On Chapter J^: Deject of Property in Measure Theory 19
if the function t : [0, oo) —>• [0, oo) has the form t(x) = £ - ^ , c > 0 , c / 1,
A e (-l,0)U(0,oo).
Concerning the defect a „ j (//) of the above measures, we can prove the
following.
Theorem 4.3 If fi = t om (with t and m as above) then we have
X + U
«n,A (ft) < " Af " 7 1 ^ , Vn > 2, VA G ( - 1 , 0) U (0, oo)
l l
and
a
ln(A+l)
oo,A(^) < 1 - ,VAe(-l,0)U(0,oo).
A
An useful result is
Theorem 4.4 (see Ghermdnescu [89], p.260, Th. 4.9,) The measurable
solutions of the functional equation f (x + y) — f (x) — f (y) = <p (x, y) are
given by
f(x) — B(x) + ax
<p{x,y) = B{x + y)-B{x)-B{y)
where B is an arbitrary measurable function and a is an arbitrary real

constant.
Theorem 4.4 is used to generate classes of fuzzy measures as follows.
Theorem 4.5 Let B(x) be such that B(0) = 0 and f(x) — B(x) + ax
is strictly increasing and f : [0,oo) —> [0,oo). Then n(A) — (/ o m) (A)
(where m is a classical finite measure) is a fuzzy measure which satisfies
the functional equation
1 1
n(AuB)=n(A) + »(B)+v (r (n (A)), r (/i (£))) ,
Â,B G A, AD B — 0, where if and f are those in Theorem 4-4 and f~l

represents the inverse function of f (here we assume f (0) = 0 and
l i m ^ + o o / (x) = +ooj.
We estimate the defects of additivity for fuzzy measures obtained by

Theorem 4.5 as follows.
20 Introduction
Theorem 4.6 / / fx : A —)• [0, oo) is the fuzzy measure defined by ft (A) =
Xm2(A), where m is a classical finite measure, then an (fi) < tL
^-fi(X),
for all n > 2 and A > 0.
Theorem 4.7 / / fi : A —> [0,oo) is the fuzzy measure defined by the

relation
ln((X + lf^ + (X+l)^B)-l)

"(^*> = - ^ hT(ATT) '
for every A, B E A,AC\ B = ®, then
In (n (A + l ) " ( x ) - n + l )
an,A (M) < „ „ ( * ) [ ^ X T ^ ,
for every n > 2, A £ (0, oo).

Next, one present three kinds of applications of the previous results: to
approximative calculation of the fuzzy integrals, to best approximation of
a fuzzy measure and to the definition of a metric on the family of fuzzy
measures. Thus, for example, an estimation of the quantity of best approx-
imation of a fuzzy measure /J, : A —¥ [0,1] by elements from
M = {m : A —> [0,1]; m additive on the er-algebra .4} ,
is given by
^4<E(^)<l,\/n>2,
Krj
n+1 - ~ ' -
where
E(fi) = mf{d(n,m);m€ M} < 1
and d (fi, v) = sup {|/i (A) — v (A) |; A £ A} is a metric on
T — {fi : A —> [0,1]; fi (0) = 0, [i nondecreasing on the <r-algebra .4} .
Another introduced concept is given by
Definition 4.6 The value

sup {\p(X)-»(A)-n(A°)\: A £ A}
W
MX)
On Chapter 4: Defect of Property in Measure Theory 21
where Ac = X \ A, is called defect of complementarity of the fuzzy measure

fj,.
Theorem 4.8 Let pt, fi' : A -> [0, oo) be fuzzy measures. We have:
(*') 0 < c ( A i ) < l ;
(Hi) c(n) = c(nc), where \f is the dual of \i, that is fic{A) = ft(X) —
[i(A°), Ac being the complementary of A;
(iv) c (a/i) = c (n), Va > 0;
We estimate the defects of complementarity for fuzzy measures obtained

by Theorem 4.5 as follows.
Theorem 4.11 / / fi is a X-additive fuzzy measure then
(i-VT+T) 2
CA (I*) < jjj •
Theorem 4.12 / / fj, : A —>• [0, oo) is £/*e /uzzy measure defined by /j, (A)
— \m2(A),\/A G A, where m is a finite measure, then c(fi) < |,VA G
(0,oo).
Theorem 4.13 / / fj, : A —> [0, oo) is the fuzzy measure defined by
n(A) = " infA+i) ' ^ G .4, where m is a finite measure and A > 0, then
In (k^p- + l)
C
<^2ln(A"(X) + l ) ~ L
Also, we prove that a cardinality of a convenient fuzzy set is a good

indicator of complementarity in pointwise sense.
At the end of Section 4.2, we give interpretations, applications and
numerical examples with respect to the proved results.
Concerning other properties of set functions, we consider
Definition 4.12 Let v : V (X) —» R be a set function. The defect of
monotonicity of v is given by
dMON {v) = - sup {\v(Ai)-v (A)| + \v (A2) - v (A)\ - \v {Ai) - v {A2)\;
A1,A,A2£V(X),A1 CACA2}.
22 Introduction
Theorem 4.15 The set function v : V (X) —¥ R is monotone if and only

if dMON (v) — 0.
Theorem 4.16 Let v : V (X) —> R be a set function. We have:
(i) 0 < dMON (v) < 2 sup {|^ (A) | ; A £ V (X)}. If v is non-negative
then 0 < dMON{v) < sup {v (A) ;AeV(X)}.
(ii) dMON (oiv) = \a\dMON (") , Va £ R, where (av) (A) = av(A),
\/AeV(X).
(Hi) dMON {VB) < dMON (v), where vB (A) = v (An B) ,VA £ V (X),
is the induced set function on B £ V (X).
(iv) If v is an additive set function then dMON (v) < o-i (\v\), where
\v\(A) = \v(A)\,AeV(X).
(v) If v is a signed fuzzy measure then dMON (v) = cfjvfOiv (vc), where
vc is the dual of v, that is i>c (A) = v (X) - v (Ac) , MA £ V (X) .
Theorem 4.19 For any v £ S = {v : V (X) —> R; v is bounded}, we have
dMON v
P
&MON t,A >->
\v) -z ( ) , .
where
EMON (v) — inf {d (v, m) ;m £ M} ,
M = {m : V (X) —>• R; m is monotone and bounded}
and
d{vuu2) = s u p { h (A) -v2{A)\;A€V (X)} , Wuv2 £ S.
Definition 4.15 Let v ; A -> R, u (0) = 0, where A C V (X) is a a-algebra.

The defect of subadditivity of order n, n > 2, of v is given by
d SU
SUB (y) = P Y I \JAi ) ~ YJV (Ai) 'AiÂ'
AinAj =9,i^j,i,j = T^n}
and the defect of superadditivity of order n, n > 2, of v is given by
4"JPH = s u p i £ V ( ^ ) - i , ( U ^ ) ; ^ € . 4 ,
.i'=l \i=l
On Chapter 4' Defect oj Property in Measure Theory 23
Ai C\Aj = 0 , i # j,i,j = l,n} .
Similarly, the defect of countable subadditivity of v is given by
{
/ 00 \ 00
A-DAj =9,i^j,i,j€'N}
and the defect of countable superadditivity of v is given by
{
CO / CO \
AiDAj =0,i# i,i,j6N}.
T h e m a i n properties of the above defined defects are the followings.

T h e o r e m 4 . 2 0 (i) 0 < 4 t / B (") < a n H > ° < d^p (v) < an (v),
Vn € N , n > 2, where an (v) is the defect of additivity of order n of u;
(ii) 0 < dfUB (v) < aTO (i/) , 0 < dfup (v) < a^ (v), where ax (v) is
the countable defect of additivity of v;

(iii) If v is subadditive or countable subadditive then dgyp (is) = an (v)
or dfUB (is) = aoo (is), respectively;
(iv) If is is superadditive or countable superadditive then dg^jB (is) =
an (is) or d'gjjp (is) = a^ (is), respectively;
(«) d^B (is) < d%+^ (is), d$>P (is) < d&V (is), Vn G N , n > 2;
(vi) a^B (is) — 0 if and only if is is subadditive; 4 1 / p (u) = ^ ' /
and only if is is superadditive; d^\jB (is) = 0 if and only if is is countable
subadditive; d^p (is) = 0 if and only if v is countable superadditive;
(vii) d$B(v) < d&V(v) + <f&B(V) and d^p(is) < 4nJ>1)(") +
{
d slp(v), VnGN,n>3.
d<
(viii) 4"C/B (") < ( « - ! ) suB (v) and d
Vn G N , n > 3 .
Let (X, /o) be a bounded metric space, CL (X) = {A C X; X \A £ Tp) ,

where Tp is t h e topology generated by t h e metric p and tp : Tp —• R + . We
define Tp : CL(X) - f R + by
Tp (A) = diam (X) -<p(X\A),VA£ CL(X),
24 Introduction
and <p*, <p* :V (X) -> R+ by
<p* (A) = sup {lp{F); F G CL(X),F C A} ,
y?» (A) = inf W (G); G G T„ A C G} .
Definition 4.17 Keeping the above notations and assumptions, we say

that A £V (X) is (^-measurable if <p* (A) — <p„ (A). The quantity
dMAS {A) = <p* (A) - <p* (A)
is called defect of ^-measurability of A.
T h e o r e m 4.17 (i) 0 < dMAS (A) <<p{X),VAeV (X) and dMAS (A) = 0
if and only if A is tp-measurable;
(11) dMAS (A) = dMAs(Ac),VA£V(X);
(Hi) dMAs ( i U B ) < dMAs (A)+dMAS (B) ,VA,BeV (X) , Af)B = 0;
(iv) Let T be the one to one transformation of the entire real line into
itself, defined by T (x) = ax + (3, where a and (3 are real numbers and
a ^ 0. If for every bounded A C R we denote T (A) = {ax | f t i £ A],
then
dMAs(T(A)) = \a\dMAs{A).
(Here p is the usual metric induced on A or T(A), respectively).
1.5 On Chapter 5: Defect of Property in Real Function

Theory
This chapter discusses various defects of properties of the real functions of

real variable.
Definition 5.1 (see e.g. Siretchi [199], p.151, p.165, [200], p.239). Let
/ : E —» R. For XQ G E, the quantity defined by
u (x0; f) = inf {6 [f (V n E)]; V e V (x0)} ,
is called oscillation of/ at XQ, where V (x0) denotes the class of all neighbor-
hoods of x0 and S [A] = sup {|ai — a2\;ai,a2 G A] represents the diameter
On Chapter 5: Defect of Property in Real Function Theory 25
of the set 4 c R .
For XQ limit point of E, the quantity defined by
Q (*„;/) = inf {S[f(VnE\ {x0})] ;VeV (x0)}

is called pointed oscillation of / a t XQ.
For e > 0, the quantity defined by
w ( / ; e ) B = s u p { | / ( a r i ) - f (x2)\;\xl - x2\ <e,xlyx2 G E]
is called modulus of continuity of / on E with step S > 0.

T h e o r e m 5.1 (see e. g. Sireichi [199], p. 166, p. 154, [200], p.211, p.239).
(i) f is continuous at XQ 6 E if and only if u> (XQ; f) = 0;
(ii) f has finite limit at XQ £ E' if and only if Q (XQ', / ) = 0. Also
lim / (x) - lim / (x) < Q (x0; f)

X-+XQ X—YXo
and if, in addition, f is bounded on E, then
2 (zo; / ) = lim / (z) - lim / (x),
where
l i m / (i) = sup {inf f (V n E\{x0}) ;V £V {x0)}
and
HnTf(x) = inf { s u p / ( K n E\ {x0}); V G V(x0)} .

t-Ho
(ill) / is uniformly continuous on E if and only if inf {u> (/; e) ; e > 0} —

0.
Definition 5.2 The quantities
rfc(/) (zo) =u{x0;f) ,
^lim (/) (»o) = v{x0;f)

and
dU6 ( / ) ( £ ) = inf {«;(/; e ) ; e > 0 } ,

can be called defect of continuity of f on XQ, defect of limit of / on XQ and
defect of uniform continuity of / on E, respectively.
26 Introduction
Definition 5.3 (see Burgin-Sostak [44], [45]). Let / : X -> Y be with

I , 7 C R and XQ £ X. The defect of continuity of / at XQ can be defined
by
fi (x0;f) = sup{|y - f (x0)\ ;y is limit point of f (x„) when xn -* a;0} .
The defect of continuity of / on X is defined by
H (f)x = sup {// (a:0; / ) ; a?0 G X } .
T h e o r e m 5.2 ('see Burgin-Sostak [44], U^])-

(i) f is continuous on Xo if and only if n(xa;f) — 0;
=
(ii) / is continuous on X if and only if /i (f)x 0>
(Hi)
IA*(*O;/) -n(x0;g)\ < n(xo;f + g)<M(xo;f) + v(x0;g),

\(j,(x0;f) - fi(x0;g)\ < n(x0;f-g),
A*(»O;—/) = t*(xo;f)
and similar inequalities for \i (f)x hold.

(iv)
/* (*o; fg)<v {"o; f) • \\g\\Xo + /* (*<>;5) • l l / I U 0 ,
êrellfifll^^supi^Whte^JJI/ll^^supll/WhiG^J,^ «sa
neighborhood of xo and f , g : X -lY, with X, Y C R .
(«) / / we c/e/me ^ (*„; / ) = i £ ^ , Mr ( / ) * = T K ^ e r e
•M = s u P r e x / (x) - infêx/ (a;), / : A" -> y , tfien fiH (f)x = n(to f)x,
where t (x) — kx, k > 0 fixed.
Definition 5.4 Let / : [a, 6] —> R and a?o £ [a, 6]. The quantity
ddif (f)(x0) =Q(x0;F),
where F : [a, b] \ {x0} -> R is given by F (a;) = ^ ^ 5 ^ ^ , is called defect

of differentiability of / on io-
Corollary 5.1 (i) f is difFerentiable on xo if and only if ddif (/) (#0) = 0.
(ii) If / is locally Lipschitz on XQ (i.e. Lipschitz in a neighborhood of
a,'o) then
ddif (/) (zo) = lim F (x) - lim F (x).

X-\XQ T-tXn
On Chapter 5: Defect oj Property in Real Function Theory 27
Theorem 5.3 Let f : [a,b] —^ R and Xo € [a,b). If f is locally Lipschitz

on xo, then
v (xo] f) <{b- a) ddi} (/) (x0).
Definition 5.6 Let / : [a, 6] -> R be bounded. The real number
dint(f)([a,b}) = f f(x)dx- f f(x)dx

Ja Ja
is called defect of integrability of / on [a, b].
Let us denote
B [a, b] = {/ : [a,b] ->• R ; / bounded on [a,b]},

C [a, b] — {g : [a, b] —> R; g continuous on [a, 6]}
and for / £ B [a, b] the quantity of best approximation
Ec(f)=M{\\f-g\\;geC[a,b]},
where ||/|| = sup {\f (t)\;t £[a,b]} .

T h e o r e m 5.4 Let f £ B[a,b] be with
Cf = {x £ [a, 6]; x is point of discontinuity of / } .
Then
su
E (f) > T>W(x;f);x£Cf}
Let us denote
L[a,b] = {/:[a,6]^R;||/||i<+ao}J
C\ [a, b] = {g : [a, b]—^R;g is differentiable on [a, b]} ,
where \\f\\L — sup < \lizlzllMl -tx,y £[a,b],x ^ y> is the so-called Lips-
chitz norm. If we denote Ei(f) = inf {||/ — g\\L ; g £ C\ [a, b]}, we can
present
Theorem 5.5 Let f £ L[a, b] be with
Df = { i £ [a, b]; x is point of non-differentiability of / } .
28 Introduction
Then
EL (/) > sup {ddif (/) (x);xeDf}.
Theorem 5.6 For each f £ B [a, b] we have
Ei(f)>dint(f)([a,b]),
where Ei (/) = inf {||/ — g\\ ; g is Riemann integrable on [a, b]} .

Definition 5.7 Let / : E ->• R. The quantity
dM(f)(E) = sup{|/(x1)-/(x)| + |/(x2)-/(x)|
- | / ( < E l ) - f{X2)\;Xi,X,X2 E E,X\ < X < X2}
is called defect of monotonicity of / on E.

For / 6 B [a, 6], let us define
EM (f) ([a, b]) = inf {\\f -g\\;g E M[a,b]} ,

where M [a, b] = {g : [a, b] —>• R; g is monotone on [a, b]} .
Theorem 5.8 For any f £ B[a, b] we have
EM{f){[a,b])>dM{f)^h]).
Definition 5.8 Let / : [a, b] —> R be bounded on [a,b]. The quantities
dcoNC (/) ([a, b]) = sup {A' (/) (a:) - f(x);x€ [a, b]}
and
dcoNV (/) ([a, b}) = sup {/ (x) - k (/) (x);xE [a, b]}
are called defect of concavity and of convexity of / on [a,b], respectively,

where K (/) : [a, b] —> R is the least concave majorant of / and k (/) :
[a, b] —y R is the greatest convex minorant of / on [a,b].
Theorem 5.9 Let f e B [a,b]. Then:
{i) dcoNC (/) ([a,fr]) = 0 if and only if f is concave on [a,b];
(ii) dcoNV (/) ([a, b]) = 0 if and only if f is convex on [a, b].
Definition 5.9 Let p £ N and / : [a, b] —> R be such that the derivative
of order p, f(p> (x), exists, Lebesgue measurable and bounded on [a,b] and
On Chapter 5: Defect of Property in Real Function Theory 29
la \f(p) (x)\dx ^ °- The

quantities
+ (/)(M1) = J^W
and
*P (/) (Ml) = {^ )(j:)dX

are called degree of convexity (or defect of concavity) of order p of / on
[a, 6], respectively. Here / | p ) (x) = f^(x) if /<*>) (x) > 0 , / | p ) ( x ) = 0
if /(?) (a;) < 0 and f _ {x) = - / ( " ) {x) if /(") (i) < 0 , / i p ) (x) = 0 if
( p)
/( p ) (*) > 0.
Theorem 5.10 Let f : [a, b] —> R be satisfying the conditions in Definition

5.9. We have:
(«) K+] (/) ([«. *]) . Al P ) (/) ([«. *]) > 0 and
)
A f ( / ) ( [ a , 6 ] ) + A'W(/)([a,6]) = l.
(«') tf / ( p ) (*) > 0, a-e. * G [a, b] then K^ (/) ([a, b}) = 1, t/ /(p> (i) <
0, a.e. s € [a, 6] ffcen A'i p) (/) ([a, 6]) = 1.
(in) Z/e£ us suppose, in addition, that /( p ) is continuous on [a,b]. Then
f is convex (concave) of order p on [a, 6] if and only if K+' (/) ([a, 6]) = 1
(K_ (/) ([a, 6]) = 1, respectively).
Further we study two concepts that measure the " quality" of an inequal-
ity: the absolute defect of equality and the averaged defect of equality.
Definition 5.10 Let L,R be two functions, L : D\ —> R, R : D2 —> R,
where D\, D2 C R", n 6 N such that
L {x) < R (x), \/x G D C D i f l D j .
The absolute defect of equality for this inequality is defined by
d # (L, # ) = sup {fl (a) - L (x);a: G D) .

If, in addition, D is Lebesgue measurable and L (x), R (x) are continuous on
D then the averaged defect of equality on [—r, r]n for the above inequality
30 Introduction
is defined by
/ (R{x)-L(x))diin
dD {L R) {r)
' = M D nH-.r]") '
n
where [—r,r] = [—r,r] x ... x [—r, r] and fj>n is the Lebesgue measure on
n
Rn.
T h e o r e m 5.11 (i) If L(x) and R(x) are given by Definition 5.10 then
we have:
d%b(L + Q,R+Q) = dab(L,R),

daDv(L + Q,R+Q){r) = daDv(L,R)(r),Vr>0
for any Q : D —> R and
daDb {kL, kR) = kdaDb (L, R),

d% (kL, kR) (r) = kdfi {L, R) ( r ) , Vr > 0
for any constant k > 0.

(«') J / J } C R " , 0 < I i ( a ; ) < Ri(x),\/x E D and 0 < L2 (x) <

R2 (x) , V x £ D then
daDb {LlL2,R1R2) < daDb (LltRi) supL 2 (x) + daDb (L2, R2) supiJi {x)
x£D x£D
and
daDv (LXL2, RiR2) (r) < d% (LUR1) (r) SU P J L 2 ( * ) + ^ ( I 2 ) R2) (r) sup/?! (a:)
xgD are!)
/or ei>en/ r > 0.

(m)dS'(^«)('-)<^6n[_rir].(I,iJ).
(iv) A{x) dab (B, C) and
d%(A,C)(r)>d%(B,C)(r),\/r>0.
Also, many concrete examples of inequalities are studied.
Open problem 5.2 A central problem in approximation theory is that

of shape preserving approximation by operators. One of the most known
result in this sense is that the Bernstein polynomials preserve the convexity
On Chapter 6: Defect of Property in Functional Analysis 31
of order p of / , for any p 6 N . If, for example, / : [0,1] —> R is monotone

nondecreasing or convex, etc., then the Bernstein polynomials
Bn{f){x) = j^Cknxk{l-xf-kf
are monotone nondecreasing, or convex, etc., respectively.

But if / is not, for example, monotone on [0,1], then it is n a t u r a l to ask
how much of its degree of monotonicity is preserved by the Bernstein poly-
nomials Bn ( / ) (x). More exactly, it is an open question if there exists a
constant r > 1 (independent of at least n, but possibly independent of /
too), such t h a t
dM (Bn ( / ) ) ([0,1]) < rdM (/) ([0,1]) ,Vn G N ,
where d\i is the defect of monotonicity in Definition 5.7.

Similarly, are open questions if there exist constants r, s,t,u,v > 1 (inde-
pendent of at least n, but possibly independent of / too), such t h a t
dcoNV {Bn ( / ) ) ([0,1]) < rdcoNV (/) ([0,1]), Vn G N ,

0) ( 0)
Al (Bn ( / ) ) ([0,1]) < sK _ (/) ([0,1]) ,Vn G N ,
:) (
Al (Bn ( / ) ) ([0,1]) < tK y (/) ([0,1]), Vn G N ,
(Bn ( / ) ) ([0,1]) < ud*CONV(f)([0,l]),\/ne-N,
and
dLiN (Bn ( / ) ) ([0,1]) < vdLIN (f) ([0,1]), Vn G N .
1.6 O n C h a p t e r 6: D e f e c t o f P r o p e r t y i n F u n c t i o n a l A n a l y s i s
This chapter introduces and studies some defects of property in functional

analysis: defect of orthogonality, defect of convexity, of linearity, of balanc-
ing for sets, defect of subadditivity (additivity), of convexity for functionals,
defect of symmetry, of normality, of idempotency, of permutability for linear
operators, defect of fixed point.
Let (E, ||-||) be a real normed space.
D e f i n i t i o n 6.1 Let X, Y C E. We say t h a t X is orthogonal to Y in the:
32 Introduction
(i) a-isosceles sense if
\\x - ay\\ = ||x + ay\\, Vx G X, Vy G Y,
where a G R \ {0} is fixed. We write X LjY (see James [107]).

(it) a-Pythagorean sense if
II* - ay\\2 = ||x|| 2 + a2 \\yf , Vx G X, Vy G Y,
where a G R \ {0} is fixed. We write X Lp Y (see James [107]).

(Hi) BirkhofF sense if
IWl < Ik + XV\\ , VA G R, Vx G X, Vy G Y.

We write X ± B Y (see BirkhofF [39]).
(iv) Diminnie-Freese-Andalafte sense if
(1 + a2) \\x + y\\2 = \\ax + yf + \\x + ayf , Vx G X, Vy G Y,
where a G R \ { 1 } is fixed. We write X 1-DFA Y (see Diminnie-Freese-

Andalafte [63]).
(v) Kapoor-Prasad sense if
\\ax + byW2 + \\x + yW2 = Wax + y|| 2 + ||x + 6y||2 , Vx G X, Vy G Y,
where a,b G (0,1) are fixed. We write X J-KP Y (see Kapoor-Prasad

[113]).
(vi) Singer sense if ||a;|| • ||y|| = 0 or
X V y , Vx G X, Vy G Y.
\\y\\
We write X _LS Y (see Singer [197]).
(vii) usual sense if
(x,y) = 0,VxGX,VyGY,
where, in addition, the norm ||-|| is generated by the inner product (•, •).
We write X _L Y.
Corresponding to the concepts of orthogonality in Definition 6.1, we
introduce the following
Definition 6.2 Let (E, ||-||) be a real normed space and X,Y C E. We
call defect of orthogonality of X with respect to Y, of:
(i) a-isosceles kind, a £ R \ {0} fixed, the quantity
df(X,Y;a) = sup \\x - aytf - ||z + ay\f

x6X,y€Y
(ii) a-Pythagorean kind, a £ R \ {0} fixed, the quantity
dj;(X,Y;a)= sup \\x\\2 + a2 \\yf - \\x - ay\\'

x€X,y£Y
(Hi) BirkhofF kind, the quantity
d$(X,Y)= sup sup \\\x\\2 - \\x + Xy\\2} .

J
x£X,y£Y\eR "-
(iv) Diminnie-Freese-Andalafte kind, the quantity
diFA{X,Y;a)= sup (l + a2) \\x + y\\2 - (\\ax + y\\2 + \\x + ay\\2)

V
x£X,yeY '
where a £ R \ {1} is fixed.
(v) Kapoor-Prasad kind, the quantity
dKP (X, Y; a, b) = sup \ax + by\\ +\\x + y\\

x£X,yeY
2
+ k + Ml )
1
lax + ;
where a, 6 £ (0,1) are fixed.
{vi) Singer kind, the quantity dj (X,Y) defined by dj {X,Y) = 0 if
X = {0} or Y = {0} and
x y X y
d^ {X, Y) = sup
xEX\{0},yeY\{0}
contrariwise.
(vii) usual kind
dx(X,Y)= sup \{x,y)\,
x€X,yeY
if, in addition, the norm ||-|| is generated by the inner product {•, •) on E.
Theorem 6.1 X _L, V if and only if d^ {X, Y) = 0, where * represents
any kind of orthogonality in Definition 6.1.
Theorem 6.2 Let (E, (•, •)) be a real inner product space endowed with the
norm \\x\\ = \/{x, x), x £ E. We have:
34 Introduction
(i) df{X,Y;a) = 4\a\-dx(X,Y).

(ii) dji (X, Y; a) = 2 |a| • d x (X, Y).
(in) 4 (x, y) = o « / y = {0} an</ 4 (x, Y) = suPxeXiyeYK{0] (ff$-
2
<supxeX\\x\\ <(dmm(X)f zfY^{0}.
(iv) d^FA {X, Y; a) = 2 (a - l ) 2 • dL (X, Y).
(v) d^p (X,Y;a,b) = 2 ( 1 - a) (1 - b) • dL (X,Y) .
(vi) d$ (X, Y) = 0 if X = {0} or Y = {0} and d^ {X, Y) = d x ( X ' , Y ' ) ,
where X' = P (X) ,Y' = P (Y) , P (x) = nf^ (i. e. X' and Y' are the pro-
jections of X and Y on the unit sphere {u 6 E; \\u\\ = I}).
T h e o r e m 6.3 Let (E,\\-\\) be a real normed space. E is an inner product
space if and only if for all X, Y C E we have
dj-(X,Y;l) = 2dji(X,Y;l)
and
dj-(X,Y;-l) = 2dji(X,Y;-l).
R e m a r k . From the proof of Theorem 6.3 it easily follows t h a t (E, ||-||) is

an inner product space if and only if, for all X, Y C E, we have
2d±.(X,Y;l)<«f±(X,Y;l)
and
2d£(X,Y;-l) <dj(X,Y;-l).
Concerning the defect of orthogonality of Cartesian product we present

T h e o r e m 6 . 4 Let (E\, \\W{) , (E2, ||-|| 2 ) be two real normed linear spaces
w u
and let us introduce on E\ x E2 the norm | | ( u i , « 2 ) | | = v l l i | l i + ll 2||2>
V ( u i , « 2 ) e ^ i x E2. For all XX,YX C Ei,X2,Y2 C E2, we have
d x ( X i x X 3 , Yi x Y 2 ) < d x ( X i , Yi) + di (X2,Y2),
for any kind " * " of orthogonality in Definition 6.1, excepting Singer or-
thogonality.
Next we present some applications.
D e f i n i t i o n 6.3 Let (E, (•, •)) be a real inner product space and X C E.
We call defect of orthogonality of X , the quantity
d1 (X) = sup {\(x, y)\;x,yeX,x^y}.
On Chapter 6: Deject of Property in Functional Analysis 35
Remark. Obviously X is orthogonal system in E if and only if d1 (X) = 0.

Theorem 6.8 Let (E, {•, •)) be real inner product space and X = {x\,..., xn}
C E with {xk,xk) = l,Vfc E { l , . . . , n } . Then for any x E E the following
estimate
\x — sn\\ — <d±(X)- £ |c-

»=i
AoWs, where sn = X)£=1Cfcfffc, c* = (x, Xk) •
Corollary 6.1 In the hypothesis of Theorem 6.8, for any x E E we have
\x~sn\\ - <dL{X)-\\x\\2 -n(n-l).
Corollary 6.2 If X } is countable, such that (ar», a?,-) =

T , i € N, then
< rfx (X) • ||a:1

-
fc = l
}Ckxk
-E
;=I
N 36
Remark. If X is orthonormal then d 1 (X) = 0 and the inequality in The-

orem 6.8 becomes the well-known equality ||x - s n || 2 = ||a;||2 — X^=i lc>|2 •
Definition 6.4 Let (E, ||-||) be a real normed space. G C E and x E E
be fixed. We say that g0 E G is element of best approximation of x by
elements by G, with respect to the orthogonality JU if x — go _L* G, where
-L* can be any orthogonality in Definition 6.1.
Theorem 6.9 Let us suppose that G C E is a linear subspace of E and
x E E\G. If there exists go E G, such that for a given a e R \ { 0 } , we have
x ~ go -Lp G (i.e. if go is element of best approximation of x with respect
to \-p), then go is element of best approximation in usual sense (that is,
Ik - 2/o|{ = inf {\\x - g\\ ;g € G}), x - g0 ±R g0 (±R is orthogonality in
Roberts sense (see e.g Singer [198], p.86)) and \\go\\ < \\x\\, \\x — 2g 0 || =
INI-
Theorem 6.10 Let (E, ||-||) be a real normed space, f 6 E*,
H = {y E E;f(y) = 0}. If there exists z E E satisfying \f {z)\ = 1 and
z J_p H for a = 1, (that is, if 0 is best approximation of z by elements in
H, with respect to ± p for a = 1), then for all x E E\G with \f (x)\ < 1,
36 Introduction
there exists element of best approximation in H, in classical sense (i.e.

3g0 G H with \\x - g0\\ = inf {\\x - g\\ ;g G H}).
Definition 6.5 The quantity 0 < E^ (x) = inf {d± (x - g,G) ;g G G) will
be called almost best approximation of a; with respect to _L* and an element
go G G with E^ (x) = d^ (x — go, G) will be called element of almost best
approximation of x (by elements in G), with respect to _L*. If, in addition,
E^ (x) = d^ (x — go, G) = 0, then #0 is element of best approximation
defined by Definition 6.4.
Theorem 6.11 Let (E, ||-||) be a real normed space, G C E be compact and
x G E (x G E\G for ^-s)- Then for each _L» in Definition 6.1 (excepting
A-B), there exists g* G G such that E+ (x) = d^ {x — g*, G).
Let (X, ||-||) be a real normed space.
Definition 6.7 A subset Y C X is called:
(i) convex, if XVl + (1 - A) y2 G Y, Vj/i, y2 G Y, A G [0,1].
(ii) linear, if
ayi+/3y2 eY,\fyi,y2eY,a,l3eB..
(Hi) balanced, if
ay£Y,Vy€Y, \a\ < 1.

(iv) absorbent, if
Va; G X, 3A > 0 such that x G AY.

R e m a r k . The following characterizations also are well-known:
Y is convex if and only if Y = convY, where
n n ^
{ 'Yôtiyi;n G N,?/,- G Y,at >Q,iE {1, . . . , n } , ^ a , - = 1 >

i=l i'=l J
represents the convex hull of Y;
Y is linear if and only if Y ~ spanY, where
spanY = < ^ P a ^ n G N,j/j G V, a,- G R, i G {l,...,n}> ;
Y is balanced if and only if Y — balY, where

balYin Mathematics,
Gal, Sorin G., and Adrian I. Ban. Defects of Properties = {ay;y£Y, \a\ <
edited by Sorin 1}.and Adrian I. Ban,
G. Gal,
Definition 6.8 Let (X, ||-||) be a real normed space and D (A, B) be a
certain distance between the subsets A,B C X (D will be specified later).
For Y C X, we call:
(i) defect of convexity of Y (with respect to D), the quantity
dcoNV (D) (Y) = D (Y, convY);
(ii) defect of linearity of Y, the quantity
dLiN(D)(Y)=D{Y,spanY);
(Hi) defect of balancing of Y, the quantity
dBAL(D)(Y) = D(Y,balY);
(iv) defect of absorption of bounded Y, the quantity dABS (D) (Y) = 0,

if Y is absorbent and
dABs{D){Y) = D(Y,B{0,RY))
if Y is not absorbent, where

Ry = sup{||y||;j,eY},
B(0,RY) = {xeX;\\x\\<RY}.
A natural candidate for D might be the Hausdorff-Pompeiu distance
DH (YI, Y2) = max{d (y x ,Y 2 ) ,d(Y2,Y1)} ,
where d(Y1,Y2)- sup {d (y^, Y 2 ); yx G Yi} ,

d{yi, Y2) — inf {\\y\ — y2\\ ;y2 G ^2}- In this case we present
Theorem 6.13 Let (X, ||-||) be a real normed space and let us suppose that
y C X . y ^ i s closed. Then:
(i) Y is convex if and only if dcoNV (DH) (Y) = 0.
(ii) Y is linear if and only if rfx/jv (DJJ) (Y) = 0.
(Hi) Y is balanced if and only if dBAL (DH) (Y) = 0.
(iv) bounded Y is absorbent if and only if dABs (DH) (Y) = 0.
Theorem 6.14 Let (X, ||-||) be a real normed space. For all bounded sets
Y,Yi,Y2 C X, with Yi,Y2^ 0, we have
38 Introduction
(0
(DH)(Y) = dLIN (DH) (Y) ;
dLiN (DH) (Y) = 0 if and only if Y is linear;
dLiN(DH)(XY) = XdLIN(DH)(Y),\fX>0;
dLiN(DH)(Y1+Y2) < dLIN(DH)(Y1) + dLIN(DH)(Y2).
dBAL{DH){Y) = dBAL (DH) (Y) ;

dBAL {DH) (Y) = 0 if and only if Y is balanced;
dBAL(DH)(XY) = XdBAL(DH)(Y),VX>0;
dBAL{DH)(Yl+Y2) < dBAL(DH)(Yl) + dBAL(DH)(Y2);
DH (balYubalY2) < DH(YUY2);
\dBAL {DH) (YI) - dBAL (DH) (Y2)\ < <2DH (YUY2).

(Hi)
dABs(DH)(Y) = dABS (DH) (F) ;

dABS(DH)(XY) = XdABS(DH)(Y),WX>0;
\dABs{DH)(Y1)-dABS(DH)(Y2)\ < 2DH(YUY2)

+ DH(B(Q,RYI),B(Q,RY2)).
For Y eVb (Y) = {Y C X; Y is bounded}, let us denote
ECONV(Y) = inf{DH(Y,A);AeVb(X),Aconvex},
EBAL (Y) = M{DH (Y, A);AeVb(X),A balanced} ,
the best approximation of a bounded set by convex bounded and by bal-

anced bounded sets, respectively.
T h e o r e m 6.19 Let (X, ||-||) be a real normed space. Then for any Y 6
Vb (X) we get
ECONV{Y) > ldCoNv{DH){Y),
EBAL(Y) > \dBAL(DH)(Y).
Definition 6.9 Let (X, +, •) be a real linear space.

(i) Let Y C X be convex. Then / : Y —>• R is called convex if
f(am + a 2 y 2 ) < a i / ( j / i ) + a 2 / ( 2 / 2 ) , Vj/,- € Y,a,- > 0, i = l , 2 , a i + a 2 = 1
(It is well-known that this is equivalent with / (Â-iîVi) ^ S l L i ^ ' / (?/«) >
Vn e N, Vy,- € Y,a{ > 0,t € {1, ...,n},J^=lai = 1);
(M) / : X —> R is called subadditive if
f(x + y)<f(x) + f(y),\/x,yeX;
(in) f : X —> R is called positive homogeneous if
/ (Xx) = Xf (x) ,VA > O.Vz 6 X;
(iv) f : X —> R is called absolute homogeneous if
f{Xx) = \X\f{x),VX£R,\/xeX;
(v) f : X —> R is called sublinear if it is subadditive and positive

homogeneous;
(vi) f : X —> R is called quasi-seminorm if it is sublinear and / (x) > 0,
V I G I ;
(vii) f : X —> R is seminorm if it is subadditive and absolute homoge-
neous;
(viii) f : X —>• R is norm if it is seminorm and / (x) = 0 implies x = 0.
Concerning these properties we can introduce the following defects.
Definition 6.10 Let (X, +, •) be a real linear space.
(i) If Y C X is convex and / : Y —> R, then the n-defect of convexity
of / on Y, rc > 2, is denned by
d
coJVV (/) (Y) = SU
P "j / ( Ylaiyi ~ X/* 1 '^ (yi) >yi£Y>ai>°>
ie {l,...,n},£\*i = 1 > .
j=i J
(ii) Let Y C A" be a linear subspace of X and / : Y ->• R. The defect
of subadditivity of / on Y is defined by
dsADD (/) (Y) = sup {/ (yi + y2) - (/ (yx) + f (y2)); !/i, jfc G Y} .
40 Introduction
The defect of absolute homogeneity of / on Y is defined by
dAH if) (Y) = sup {|/ i\y) - |A| / (y) |; A e R, y € y } •
Theorem 6.20 Let iX, +, •) be a real linear space.

(?) Let Y C X be convex and f : Y —y R . Then for all n > 2 we have
d(c]oNV if) (Y) < 4 " c U l (/) (Y) < dÔNV if) (Y) + dfONV if) (Y).
Also, f is convex on Y if and only if for any fixed n > 2, we have

d(c]ONV if) (Y) = 0.
For f,g : Y —¥ R, a > 0 and n > 2 we have
dcoNV
(f + 9)(Y)< (9)(Y)
and
d(cn)ONvi«f)iY)=<*dPONVif)iY).
(it) Let Y C X be linear subspace of X and f : Y —> R . / / / (0) = 0

then f is subadditive (on Y) if and only if dsADD if) (Y) — 0. / is absolute
homogeneous (on Y) if and only if dAH if) (Y) = 0. Moreover, if f,g :

Y —y R then
dsADD if + g) (Y) < dSADD if) (Y) + dsADD (</) (Y) ,

dAHif + 9)iY) < dA„if)iY)+dAHi9)iY),
dSADDiaf)iY) = adSADDif)iY),Va>0,
dAH{af){Y) = \a\dAHtf){Y),VaeIl.
For Y C X linear subspace or convex set, let us denote
B0 (Y) = {/ : Y ->• R; / is bounded on Y and / (0) = 0} ,

ESADD if) (Y) = mf {||/ - g\\; g G B0 (Y), g is subadditive on Y} ,
/ G B0 (Y) and
B (Y) = {/ : Y -> R; / is bounded on Y} ,

^COJVV (/) ( y ) = inf {\\f-g\\;geB(Y),g is convex <m Y} ,
f € B (Y), respectively, where ||/|| = sup {|/(a;)|; x G Y}.
Theorem 6.21 (i) Let Y C X be linear subspace of X and f £ BQ {Y)-

Then
ESADD(f)(Y)>dsADD(f)iY).
(ii) Let Y <Z X be convex and f £ B (Y). Then
ECONV (/) (Y) > dcoNv(f){Y)^n > 2
Remark. By Theorems 6.20, (i) and 6.21, (ii), there exists the limit
l i m n ^ o o d ^ ^ y (/) (Y) and moreover
\ lim dÔNV (/) (Y) < ECONV (/) (Y).
Let (X, (•, •)) be a Hilbert space over R or C and
LC (X) = {A : X —>• X; A is linear and continuous on X} .
For any A £ LC (X), we define the norm |||-||| : LC (X) ->• R + by |||A||| =
sup {\\A (x)\\ ; ||a;|| < 1}, where \\x\\ = ^(x,x).
The following concepts are well-known in functional analysis (see e.g.
Muntean [155] and Ionescu-Tulcea [103]).

Definition 6.11 The operator A £ LC (X) is called:
(i) symmetric (or Hermitean) if (A(x) ,y) = (x,A(y)} ,Vx, y £ X;
(ii) normal, if AA* = A* A, where A* is the adjoint operator of A defined
by (A (x) ,y) = (x,A* (y)) ^x,y£X and AA* (x) = A (A* (x)) , Vx £ X;
(iii) idempotent, if A2 = A, where A2 (x) — A (A (x)) , Vz £ X;
(iv) isometry, if (A (x) , A (y)) = (x, y) , Vx, y £ X (or equivalently, if
\\A(x)\\ = \\x\\,Vx£X);
Also, two operators A, B £ LC (X) are called permutable if AB = BA.
Suggested by these properties, we can introduce the following

Definition 6.12 Let A, B £ LC (X).
(i) The defect of symmetry of A is given by
dsiM (A) = sup {\{A (x), y) - (x, A (y))\; ||*||, |jy|| < 1} ;
(ii) The defect of normality of A is given by
dNOR (A) = sup {\\A (A* (x)) - A* (A (x))\\; ||z|| < 1} = \\\AA* - A*A\\\;
42 Introduction
(Hi) The defect of idempotency of A is defined by
diDEM (A) = sup {|| A 2 (*) - A (x)\\ ; ||z|| < 1} = || \A2 - A\|| ;
(iv) The defect of isometry of A is given by
diso (A) = sup {\\A (x) - x\\; ||z|| < 1} ;
(i>) The defect of permutability of A and B is given by
dpERM [A, B) = sup {\\A (B (x)) -B(A (x))\\; |HI < 1} = \\\AB - BA\\\.
Also, if Y C LC(X), then we can introduce the corresponding defects for

Y, by
DSIM (Y) = sup {dsiM [A); A e Y} ;

DNOR(Y) = Sup{\\\AA*-A*A\\\;A£Y};
DIDEM(Y) = 8up{|||A2-A|||;^6y};
DISO(Y) = Bup{diSO{A);A£Y};
dpsRM {Y) = sup {\\\AB - BA\\\ ;A,BeY}.
T h e o r e m 6.22 Let A, B £ LC (X).

(i) A is symmetric if and only if dsiM {A) = 0;
(«') A is normal if and only if dôR [A) = 0;
(Hi) A is idempotent if and only if dm EM (A) = 0;
(iv) A is isometric if and only if diso {A) = 0;
(v) A and B are permutable if and only if dpERM (A, B) = 0.
T h e o r e m 6.23 Let A,B £ LC(X). We have

(i) dsiM (XA) = |A| dsiM (A), VA £ R;
dNOR (XA) = |A|2 dNOR (A),V\£RorC.
(ii) dSiM (A + B) < dsiM (A) + dsiM (B);
dsw(AB) < 2 | | | A B - / | | | , where I(x) = x, Vz £ X;
dNOR (A + B) < dNOR (A) + dNOR (B) + dPERM {A, B*)
+dPBRM {A*, B).
(Hi) dsiM (A 1 ) = dsiM (A) , if there exists A 1

and A is isometry.
(iv) \dSIM (A) - dSIM (A*)\ < 2 | p - 4*|||;
<1NOR(A*) = dNOR(A);
dNOR (A + A*) < dNOR (A) + dNOR (A*);
\diDBM (A) - dlDEM (A*)\ < | p - 4*||| + J \A2 - (4*) 2 || ;
dlDEM (I - A) = dlDEM (A) ;
dpBRM (A, A*) = dNOR (A) .
For given A 6 LC (X), let us introduce the following quantities of best
approximation:
ESIM (A) = inf {\\\A - B\\\; B G LC (X) , B is symmetric} ,
ENOR(A) = wf{\\\A-B\\\ + \\\A*-B*\\\;BeLC(X),

\\\B\\\ < 1, B is normal} ,
EIDEM(A) = inf{|||A-JB|||+|||42-52|||;BeLC(X),
B is idempotent} ,
EISO (A) = inf { | p -B\\\;B€LC(X),B is isometry} .
T h e o r e m 6.24 Let A G LC(X). We have:
d
ESIM (A) > -Âl,
EMOR(A) > ^ | M , if ||L4||| < i,
ElDEM (A) > diDEM (A) ,

EISo(A) > diSo(A).
Definition 6.13 Let (X, d) be a metric space and M C X. The defect of

fixed point of / : M -> X is defined by ed (/; M) = inf {d (x, f (x)) \x G M } .
If there exists XQ G M with e^ (/; M) = d (x0, f (XQ)), then xo will be called
best almost-fixed point for / on M.
T h e o r e m 6.25 Let (X, d) be a metric space and f : X —^ X be a non-

expansive mapping, i.e. d(f(x),f(y)) < d(x,y), for all x,y G X. Then
ed (/"; X) < ned (/; X) and inf {d ( / " (a:), / " + 1 (a;)) ; x G X) = ed (/; X),
Vn G {1,2,...}, where fn denotes the n-th iterate of f.
44 Introduction
Let us denote
T = {g : X —> X;g has fixed point in X} ,
where (X, d) is supposed to be compact. A natural question is to find the
best approximation of a function / : X —> X, f ^ T, by elements in T. In
this sense, we can define
Ef(f)=mf{D(f,g);gef},
where D (/, g) = sup {d (f (a?), g (x)) ;x€X} .
The following lower estimate for E? (/) holds.
Theorem 6.26 We have ed (f;X) < E? (/), for any f : X -> X.
Theorem 6.30 Let (X, {,)) be a real Hilbert space, \\x\\ = \/(x, x), x 6 X,
the metric generated by norm |||| denoted by d and M C X.
(i) Let f : M —¥ X be Gateaux derivable on xo G M, x0 interior point
of M, with ed (/; M) > 0. / / ||ar0 - / (x0)\\ - ed (/; M), x0 G M, then
< * o - / ( x o ) , ( l x - V / ( a r o ) ) ( A ) > = 0.
for all heX.
(ii) If F (x) = \\x — / (x)\\ , x G X is moreover Gateaux derivable on

M C X open convex, then \\xo — f (xo)\\ = td (/; M) if and only if
{x0 - / (x0), (ljf - V / («o)) (ft)> = 0, V/i G X.
Let us consider C[0,1] endowed with the uniform metric

d(f,g) = s u p { | / O e ) - 0 ( a : ) | : a ; € [ O , l ] } . Denoting^! : C[0,1] -»• C [0,1]
by
(v4tt) (x) = 1 + X u(s — x) u(s)ds,

Jx
where A > i , it is known that A has no fixed points in C [0,1]. However,

we can prove
Theorem 6.32 Let A : M —t C[0,1], where \~> \,A is defined as above
and
M = {u G C [0,1] : u is derivable, u'(x) > 0, Mx G [0,1],

0 < "(0) < «(1) < 1}.
On Chapter 7: Defect oj Property in Algebra 45
Then t& (A; M) — 4 4 A 1 and a best almost-fixed point for A on M is

u: [Q,l]-*R,u(x) = ^ , V z e [0,1].
Open problem 6.1 If Y C X is an absorbent subset of the linear space

(X, +, •) then the well-known Minkowski's functional attached to Y is de-
fined by
pY (X) = inf {a > 0; x G oiY} , x € X.
This functional characterizes the quasi-seminorms and the seminorms as

follows (see e.g. Muntean [154], p. 43-45):
(i) p : X —>• R is quasi-seminorm if and only if there exists Y C X,
absorbent and convex such that p = py;
(ii) p : X —> R is seminorm if and only if there exists Y C X, absorbent
convex and balanced such that p = pY;
In the proof of (?) , given an absorbent subset Y C X, the subadditivity
of py is essentially a consequence of the convexity of Y (because py is al-
ways positive homogeneous if Y is absorbent). Let us suppose, in addition,
that (X, H'll) is a real normed space. Then, would be natural to search
for a relationship between the defect of subadditivity dsADD (PY) (X) and
the defect of convexity dcoNV (D) (Y) (where D = DH or D = D*) in

such a way that for absorbent Y C X, dsADD {PY) {X) = 0 if and only
if dcoNV (D) (Y) — 0. Similarly, in the proof of (ii), given an absorbent
and convex subset Y C X, the absolute homogeneity of py is essentially
a consequence of the fact that Y is balanced. Therefore, in this case
would be natural to search for a relationship between the defect of abso-
lute homogeneity d^H {PY) (X) and the defect of balancing dsAL (D) (Y),
(where D — DH or D = D*) in such a way that for absorbent and convex
Y CX, dAH (PY) (X) = 0 if and only if dBAL (D) (Y) = 0.
1.7 On Chapter 7: Defect of Property in Algebra
Let (X, d) be a metric space and F : X x X —t X be a, binary operation

on X. If F is not commutative, or is not associative, or is not distributive
(with respect to another binary operation G : X x X —> X), or has no
identity element, or no every element has an inverse, so on, it is natural to
look for a concept of defect of F with respect to these properties.
It is the main aim of this chapter to introduce and study the concept of
46 Introduction
defect of F with respect to the above properties.
Definition 7.1 Let (X, d) be a metric space, Y C X and F : X x X ^r X

be a binary operation on X.
(i) The quantity
4OM (F) (Y) = sup {d (F (x, y), F (y, x)) ;x,yeY]
is called defect of commutativity of F on Y with respect to the metric d.

(n) The quantity
4s (F) (Y) = ^ P {d (F (F (x, y),z),F (x, F (y, z))) ;x,y,zeY}
is called defect of associativity of F on Y with respect to the metric d.

(Hi) The quantity sup {d (F (x, y), x); x 6 Y} is called defect of identity
element at right of y (with respect to F) and
4DR (F) (Y) = inf {sup {d (F (x, y) , x); x G Y} ; y e Y}
is called defect of identity element at right of F on Y (with respect to d).

Similarly, we can define the defect of identity element at left,
4DL (F) (Y) = inf {sup {d (F(y,x),x);x£Y};y£ Y} .
If there exists en 6 Y such that
e
IDR
(F) (Y) = sup {d (F [x, eR),x);x£Y}> 0,
then eR will be called best almost-identity element at right of F on Y.
Analogously, if there exists ej, 6 Y such that
4DL (F) (Y) = sup {d (F (eL, x),x);x€Y}>0,
then eL will be called best almost-identity element at left of F on Y.

(iv) Let us suppose that there exists a € X such that F(x,a) —
F(a,x) -x,^xe X. Then
inf {d (F(x, y), a ) ; y £ Y}
and
M{d(F(y,x),a);y£Y}
On Chapter 7: Defect of Property in Algebra 47
are called defects of invertibility of x at right and at left, respectively. The

quantities
ediNR (F) (Y) = sup {inf {d (F (x, y) , a); y G Y} ; x G Y}

and
4NL (F) (Y) = sup {inf {d (F(y,x),a);yeY};xe Y)

are called defects of invertibility of F on Y, at right and at left, respectively.
The quantity ejN(F)(Y) = max {ejNL (F) (Y) , edINR (F) (Y)} is called
defect of invertibility of F on Y (with respect to d).
If there exists x*R £Y such that
0 <inf{d(F (x,y) ,a);y eY} = d(F (x,xK) ,a),
then x*R will be called best almost-inverse at right of x with respect to F.

Similarly, if there is x*L 6 Y with
0 < inf {d {F {y,x),a);yEY} = d {F (x*L ,x),a),
then x*L will be called best almost-inverse at left of x with respect to F.

(v) The quantity
edREL(F)(Y)=sup{d(x,y);x,yeY,F(z,x) = F(z,y)}
is called defect of regularity at left of z on Y. Analogously,
eRER (F) (Y) = sup {d (x, y);x,yEY,F (x, z) = F (y, z)}

is called defect of regularity at right of z on Y.
(vi) The quantity
e<
IDEM
(F) (Y) = sup{d(F (x,x) ,x) ;x eY}
is called defect of idempotency of F on Y (with respect to d).
(vii) If G : X x X —> X is another binary operation on X then
4 / s x ( ^ G) (Y) = sup {d (F (x, G(y,z)),G (F (x, y), F (x, z))) ;
x,y,z € Y}
is called defect of left-distributivity of F with respect to G on Y. Similarly,
4ISR (F; G) (Y) = sup {d (F (G (y, z),x,),G(F {y, x), F (z, x))) ;
x,y, 2 e Y}
48 Introduction
is called defect of right-distributivity of JF with respect to G on Y . If

F = G, then edDISL (F; F) (Y) and edDISR (F; F) (Y) are called defects of
autodistributivity (left and right) of F on 7 . Also,
edAB (F; G) (Y) = m a x { s u p {d (F (a, G (a, b)) ,a);a,beY} ,
sup {d {G (a, F (a, 6)) , a ) ; a, 6 e Y}}
is called defect of absorption of (F, G) on Y.
L e m m a 7.1 With the notations in Definition 7.1 we have:

(i) F is commutative on Y if and only if edCOM (F) (Y) = 0.
(ii) F is associative on Y if and only if e ^ 5 (F) (Y) = 0.
(Hi) If F has identity element at right in Y, i.e. there is a 6 Y such
that F(a,x) = x,Vx G Y, then ejDR(F)(Y) = 0. Conversely, if (Y,d) is
compact and F is continuous on Y x Y (with respect to the box metric on
Y xY) then ejDR (F) (Y) = 0 implies that F has identity element at right
in Y. Similar results hold in the case of identity element at left.
(iv) If F has identity element in Y and each x £ Y has inverse at
right, then edNR (F) (Y) = 0. Conversely, if (Y, d) is compact and F is
continuous on Y x Y (with respect to the box metric on Y x Y), then

edNR (F) (Y) = 0 implies that each x E Y has inverse at right. Similar
results hold for invertibility at left.
(v) A set Y C X is called regular at left if each z £ Y is regular at
left (i.e. F(z,x) = F(z,y) implies x = y). Then Y is regular at left
with respect to F if and only if eREL (F) (Y) = 0. Similar results hold for
regularity at right.
(vi) A set Y C X is called idempotent if each x £ Y is idempotent (with
respect to F). Then Y is idempotent with respect to F if and only if
ediDEM(F)(Y) = 0.
(vii) F is left-distributive (right-distributive) with respect to G on Y if
d d
and only if e DISL (F;G)(Y) = 0 (e DISR (F;G)(Y) - 0). Also, the pair
(F, G) has the property of absorption on Y if and only if edAB (F, G) (Y) = 0.
D e f i n i t i o n 7.2 Let / , g : X —> X be with f o g = i, where i is the identity

function on X. If G is a binary operation on X, then F : X x X —> X
given by F = g o G(f,f) is called the (/, </)-dual of G (i.e. F(x,y) =
g(G(f(x)J(y))),V(x,y)eXxX).
D e f i n i t i o n 7.3 Let (X, d) be a metric space and F,G : X x X —>• X. We
On Chapter 7: Defect of Property in Algebra 49
say t h a t F is less t h a n G (we write F < G) if F is the (/, g)-dual of G,

where d (g (x) , g (y)) < kd(x,y) , Vx, y £ X and 0 < k < 1 is a constant.
T h e o r e m 7.2 PFii/j £ne notations in Definitions 7.1 and 7.3, the condition
F < G implies:
(»') 4OM (F) (Y) < kedCOM (G) ( / (Y)) ,W C X . /n parteuZar,
( F ) (X) < k
e
COM
(G)(X).
F y ke
("') 4 s ( ) ( ) < AS (G) ( / ( Y ) ) , VY C X . 7n particular,
e
AS
(Hi) Let Fi be' the (f,g)-duals of Gi,i £ {1,2}, such that F{ < Gi,i £
{ 1 , 2 } . Then
e
DiSL (Fi\F2) (Y) < kedDISL ( G i ; G 2 ) ( / ( Y ) ) ,VY C X , and in particular
e
DISL
(F1;F2)(X)<k
e<
biSL (G\\ G2) (X) . Similar results hold for efîsR-
T h e o r e m 7.5 Let (X,-,d) be a metrizable non-commutative group, N a
closed normal divisor of X and d a metric that is left invariant. If d
denotes the induced metric on X/N and 0 the induced operation on X/N,
then
e
coM (©) (X/N) < edCOM (•
)(*)•
For F,G : X x X —>• X, we define the distance between F and G by
D (F,G) = sup {d(F (x,y) ,G (x,y)) ;x,y e X} .
Also, we define
ECOM (F) = inf { D (F, G);G€ COM (X)}
and
EAS (F) = inf {D (F, G ) ; G £ A S (X)} ,
the best approximation of a binary operation on X by c o m m u t a t i v e oper-

ations and by associative operations, respectively.
T h e o r e m 7.6 Let (X,d) be a metric space with d bounded on X. We

have \tdCOM (F) < ECOM (F) and \edAS (F) < EAS (F), for any binary
operation F : X x X ->• X.
In Section 7.2 we consider examples of calculations for the defects stud-

ied in Section 7.1. Section 7.3 deals with triangular norms and conorms.
50 Introduction
T h e o r e m 7.7 Let T be a triangular norm and S be a triangular conorm.

We have:
( 0 eiDEM (T) = sup {x - T (x, x); x G [0,1]} ;
(ii) ejDEM (S) = sup {S (x, x) - x; x G [0,1]} ;
(Hi) If S is the dual of T, that is S (x, y) — 1 — T (1 - x, 1 — y) , Va:, y G
[0,1], then ejDEM (T) = eIDEM (S) ;
(iv) / / Ti < T2, that is Ti (a;, y) < T 2 (a;, j/) ,Vx,y£ [0, 1], tfien
eiDEM (T2) < eiDEM (Ti);
(v) If Si < S2, that is S\ (x, y) < S2 (x, y), Va;, y G [0,1], tften
eiDEM (Si) < eiDEM (S2);
(vi) / / T* zs f/je reverse of triangular norm T, that is
T* (x, y) = m a x ( 0 , x + y — 1 + T (1 — x, 1 — y)) (see Kimberling [117] or
Sabo [186]), then eIDEM (T*) < eIDEM (T) .
T h e o r e m 7 . 1 1 (i) 0 < eois (F',G) < 1 for every triangular norms or

conorms F and G;
(ii) eois (T; SM) = 0, for every triangular norm T;
(Hi) eois (S;TM) = 0, for every triangular conorm S;
(iv) eois {F; F) = eiDEM (F) for every triangular norm or conorm F;
(v) If S is the dual of T then eDIS (T; T) = eDIS (S; S) .
C h a p t e r 7 ends with some applications.
1.8 O n C h a p t e r 8: M i s c e l l a n e o u s
In this chapter we study some defects of property in Complex Analysis,

Geometry, N u m b e r Theory and Fuzzy Logic.
P o m p e i u [163] notes t h a t if Jc f (z) dz ^ 0, then \fcf(z)dz\ can be
considered as a measure of non-holomorphy of / inside of the d o m a i n
bounded by the closed curve C .
D e f i n i t i o n 8.1 Let D C C be a bounded domain and / : D —y C , inte-

grable on D. T h e number
duoh ( / ) (D) - sup

(X f(z)dz ; C C D, closed rectifiable curve >
is called defect of holomorphy of / on D.
On Chapter 8: Miscellaneous 51
The following properties are immediate:
dHOL(f + g)(D) < dHOL(f){D)+dHoL(g)(D),\/f,geC(D)

dHoL(Xf)(D) = \\\dHOL(f)(D),\/\eC,feC(D).
Continuing the ideas in Pompeiu [163], let us suppose that / £ C 1 (£)),

that is if / = u + iv then u, v are of C 1 class.
Definition 8.2 (Pompeiu [163]) Let / £ C1 (D), f {z) = u (x, y) + iv (x, y) ,

z = x + iy £ D and ZQ = XQ + Wo £ D. The areolar derivative of / at ZQ is
given by
dA(f)(z0)=lim*J°fWdZ,
C-t-zo m (AJ
where the limit is considered for all closed curves C (in D) surrounding ZQ,
that converge to ZQ by a continuous deformation (m (A) represents the area
of the domain A closed by C).
Remark. Because for / £ C 1 (D) and z0 £ D, it is obvious that (see

Pompeiu [164])
d\ (/) (zo) = 0 if and only if / is differentiable at zo,
we can call \d,A (/) (ZQ)\ as defect of differentiability of / £ Cl (D) on z0.

In Szu-Hoa Min [209], |dyi (/) (zo)\ is called deviation from analiticity.
The defect of differentiability of / on D given by
dA (f)(D)= sup {\dA(f)(z)\;z£D},
appears in the estimate of / (z) by the Cauchy's integral:
/(0
2niJc -d£ < rdA (/) (A)
if we choose C = {u £ C; \u — a\ — r} C D, A = int (C) C D and z £ D
such that |z — a| > 2r.
In the Euclidean and non-Euclidean geometries, the concept of curva-
ture and torsion of a curve C in a point M are introduced by (see e.g.
Mihaileanu [148], p. 99-100)
Aa
7 (M) = lim ——
52 Introduction
and
A0
r (M) = lim —-,
respectively, where A a is the angle of tangents in M and M', A6 is the

angle of binormals in M and M' and As is the length of the arc MM'
(M,M' E C) when M' tends to M.
Definition 8.3 Let C be a curve. The quantities
7(C) = sup { | 7 ( M ) | ; M G C }
and
T{C) = sup {\r(M)\;M EC}
are called the defect of right line and the defect of plane curve of C, respec-
tively.
Definition 8.5 Let ABC be a triangle in a geometry. The quantity
DA {ABC) = \A + B + C - TT|
is called defect of Euclidean triangle.

Let us assume that the absolute of a non-Euclidean space is given by
q2x\ + x\ + x\ + x\ = 0,
where g2 G R and let us denote e = ^.

The inner product of two points X and Y with coordinates Xi, i E
{0,1, 2,3} and yt,i E {0,1,2,3} is given by
X .v - - 92;coj/o + xiyi + x2yi + x3y3

\fq2xl + x\ + x\ + xl^/q2yl + y\ + y\ + y\
and the inner product of two planes a : aoô + a i ^ i + a2X2 + 031:3 = 0,
/? : PQXQ + /?!«! + f32x2 + 03x3 = 0 is given by
£ 2 a 0 /?0 + « l A +»2/?2 +Q3/?3

v / £ a + a? + a 2 + a 2 ^ 2 / ? 2 + /?? + /?! + Pi'
2 2
Definition 8.6 (i) The quantity
d0RTH(X,Y) = \X-Y\
is called defect of orthogonality of the points X and Y.

(ii) The quantity
doRTH («,/?) = \a- j3\

is called defect of orthogonality of the planes a and /?.
Theorem 8.2 (i) doRTH {X, Y) = 0 if and only if X and Y are orthogo-
nal; doRTH {a,/3) = 0 if and only if a and /? are orthogonal.
{ii) 0 < doRTH (X,Y) < 1 and 0 < dORTH (<*,/?) < 1.
(in) d0RTH (X, Y) - d0RTH (Y, X); dORTH (a, (3) = d0RTH (P, «)•
(iv) doRTH (X, Y) = cos I where d is the distance between X and Y;
doRTH {aiP) = |cos^|, where 6 is the angle of the planes a and /?.
Let us consider two right lines u and v with the common point X and
let U, V be the orthogonals of the point l o n u and v, respectively (that
is U £ u and V £ v,U • X = 0,V • X = 0). The angle A of the right lines
u and v is given by (see Mihaileanu [148], p.33)
cos A = U • V.
Definition 8.7 The quantity
doRTH (u,v) = |cosA|
is called defect of orthogonality of the right lines u and v.

Theorem 8.3 (?) doRTH (u,v) = 0 if and only if u and v are orthogonal,
(ii) 0 < d0RTH(u,v) < 1.
(in) d0RTH (u, v) = d0RTH (v, u).
[iv) doRTH («, V) = cos I where S is the distance between u and v.
(v) If u : ax + by + cz = 0 and v : a'x + b'y + c'z = 0 are two right lines
1 iL J I \ |aa'+66'+cc'|
in plane, then dORTH («,«) = M ^ W a ^ H , l i a .
Now, we consider u : ax + by + cz — 0 and v : a'x + b'y + c'z = 0 two
right lines in a non-Euclidean plane. By definition, u and v are parallel if
their angle is null (see Mihaileanu [148], p. 39).
dpAR(u,v) = |sinA|
is called defect of parallelness of the right lines u and v.
54 Introduction
Theorem 8.4 (i) 0 < dPAR (u,v) < 1.

(ii) dpAR (u,v) = dpAR (v,u).
(Hi) dPAR(u,u) = 0.
/• \ j I \ • <5 /(a6'-a'b) 2 +£ 2 (ac'-a'c) 2 +£ 2 (be'-6'e) 2
(iv) dPAR(u,v)= s i n - =yJi (a4b?+X*){a>4b"+k*) '•

In what follows we deal with Number Theory.
Definition 8.12 Let x 6 R. The quantity
dn.z (x) = min (x - [x], [x] — x + 1)
is called defect of integer number of x, where [x] is the integer part of x.
T h e o r e m 8.7 (i) dnz(x) = min ({a:} , 1 — {x}), where {x} denotes the
fractional part of x.
(H) <^RZ (x) = 0 if and only if x 6 Z and dnz (x) = d (x, Z) .
(Hi) 0 < d R Z (x) < | , V x € R.
(j«) C^RZ (x) = dRZ (~x), Vx e R.
Let us denote by X = {#1, ...,xn} a finite set of real numbers and let
us consider the fuzzy set Ax : R -»• [0,1] corresponding to X and defined
by Ax (x) — 0 if x G R \ X and Ax (a) = {x} if x e X, where {x} is the
fractional part of x. If dc is a normalized measure of fuzziness (that is has

values in [0,1]), then we give the following
dj (X) = dc (Ax)
is called defect of integer of the set X.
T h e o r e m 8.8 (i) dj (X) = 0 (i.e has the minimum value) if and only if
xk GZ,VA;€ { l , . . . , n } .
(ii) di (X) = 1 (i.e. has the maximum value) if and only if C/RZ (X/,) is
maximum, \/k 6 {1,..., n} .
(Hi) Let Y = {j/i, ...,yn} be a finite set of real numbers. If {xk} < {yk}
for{yk}< \ and {xk} > {yk} for {yk} > ±,Vfc 6 {1, ...,n}, then rf/ (X) <
di(Y).
Definition 8.14 Let p, q two natural numbers. The quantity
Ddiv (p, q) — min {r, q - 1 - r} ,

where r is t h e remainder of division of p by q, is called defect of divisibility

of p with respect to q.
T h e o r e m 8 . 9 Let p,pi,P2,q be natural numbers.

(i) Ddiv (p,q) — 0 if and only if q/p (i.e. q is divisor of p).
(ii) 0 < Diiv (p, q) < [§] + 1, where [x] is the integer part of x.
(Hi) If pi = p 2 (modg) then Ddiv (p\,q) - Ddiv (p2,q) and
Ddiv (pi -P2,q) = 0-
(iv) Ddiv (p1 + p2, q) = Ddiv ( n + r 2 , q) and Ddiv (pip2, q)
— Ddiv (rir2,q), where ry and r2 are the remainders of division of p\ and
P2 by q.
D e f i n i t i o n 8 . 1 7 Let m and n be positive integers. T h e quantities
^perfect ( n ) n - J^ k
and
âmicable (m-n )
i/m,i^m jln,j^n
are called defect of perfect number of n a n d defect of amicable numbers of

m, n, respectively.
A fuzzy logic (see e.g. Butnariu-Klement-Zafrany [5l]) can be described

as a [0, l]-valued logic, t h a t is one real number t (p) £ [0,1] is assigned t o
each proposition p.
D e f i n i t i o n 8 . 1 8 (see Hajek-Godel [98]) T h e propositional form A is a 1-

tautology (or s t a n d a r d tautology) if t (A) — 1, for each evaluation.
D e f i n i t i o n 8 . 1 9 Let A be a propositional form which is represented with

propositional forms A\, ...,A„ and connectives. T h e quantity
<*TAUT(A) = 1-inf {*(Ai,...,An,Vs,AT,-);

Ai, •••)An propositional forms}
is called defect of 1-tautology of propositional form A.
A m o n g all fuzzy logics, min — m a x logic is the most used in practice.
56 Introduction
Theorem 8.12 (see Butnariu-Klement-Zafrany [51]) A propositional form

A in min —max fuzzy logic is a tautology if and only if t (A) > 0.5, for
every evaluation.
Let A be a propositional form which depends on propositional forms
Ai,...,An and connectives V, A,—)•,->.
dTAUT (A) = max{0, 0.5 - inf {t (Al,...,An, V, A, ->, - . ) ;
Ax,...,A n are propositional forms}}
is called defect of tautology in min — max fuzzy logic (or TM-tautology) of

propositional form A.
In intuitionistic fuzzy logic, two real non-negative numbers, p (p) and
v (p), are assigned to each proposition p, with the following constraint:
A«(P) + " ( P ) < 1 -

Definition 8.21 (see e.g. Atanassov [12]) The propositional form A is an
intuitionistic fuzzy tautology if and only if
p(A)>v(A).
Definition 8.22 Let A be a propositional form which can be represented

with arbitrary propositional forms A\,..., An and connectives. The quantity
di-TAur{A) = ma,x{0,sup{i/(Ai,...,An) - p{Ai, ...,An);

A\, ...,An propositional forms}}
is called defect of intuitionistic fuzzy tautology of propositional form A.

Theorem 8.13 (i) 0 < dI^TAUT {A)<\.
(ii) di^TAUT [A) = 0 if and only if A is an intuitionistic fuzzy tautol-
ogy-
Chapter 2
Defect of Property in Set Theory
In this chapter we consider the measures of fuzziness as measuring the

"deviation" of a fuzzy set from the concept of crisp (classical) set, t h a t is
as measuring the defect of crisp (classical) set and the intuitionistic entropies
as measuring the "deviation" of an intuitionistic fuzzy set from the concept
of fuzzy set, t h a t is as measuring the defect of fuzzy set. In the last section,
some applications to the determination of the degree of interference (mainly
in the geography of population), to description of systems performance and
to digital image processing are given.
2.1 Measures of Fuzziness
Given a set X, a fuzzy subset A of X (or a fuzzy set A on X) is denned by

a function
liA : X -+ [0,1]
such t h a t HA (X) expresses the degree of membership of # to A, i.e., the

degree of compatibility of x with the concept represented by the fuzzy set
A. Therefore, HA (X) = 0 means t h a t x is definitely not a m e m b e r of A and
)XA (X) = 1 means t h a t x is definitely a member of A; if either fj,'A ix) — 0
or HA (X) — 1 for every x 6 X, then A is a crisp set. We denote by FS(X)
the class of all fuzzy sets on X and we identify the fuzzy sets with their
membership functions, t h a t is
FS(X) = {A\A:X->[0,1]}.
57
58 Deject of Property in Set Theory
T h e measures of fuzziness (see e.g. Ban-Fechete [20], Klir [120], Knopf-

macher [122], Roventa-Vivona [180], Vivona [217]) or, in other words, the
fuzzy entropies (see e.g. Rudas-Kaynak [181], R o m a n Flores-Bassanezi [34])
are real functions which attach to fuzzy sets values t h a t characterize their
degree of fuzzification, t h a t is measure the difference between fuzzy sets
and classical (or crisp) sets.
A general definition of the measures of fuzziness is the following (inspired
by Rudas-Kaynak [181]):
Definition 2.1 A measure of fuzziness is a positive real function dc de-

fined on J-(X) C FS(X), t h a t satisfies the following requirements:
(*') If A eT(X),A(x) G {0, l } , V a ; e X then dc{A) = 0.
(n) If A -<; B then dc (A) < dc(B), where A -< B means t h a t A is
sharper t h a n S(see e.g. relation (2.2) below).
(Hi) If A is maximally fuzzy (see e.g. relation (2.3) below) then dc (A)
assumes its m a x i m u m value.
R e m a r k s . 1) If (X,A,fi) is a measure space (all measure-theoretic terms

and results used in this chapter m a y be found in Halmos' book [99]), we de-
note TA {X) = {A £ FS(X); A is „4-measurable}. In m a n y papers (see e.g
R o m a n Flores-Bassanezi [34], Knopfmacher [122], Roventa-Vivona [180])

it is considered T(X) = TA (^0 and the condition (i) in Definition 2.1 is
replaced by
dc(A) = 0 if and only if A(x) G {0,1} ,fi-a.e. x G X. (2.1)
If the set X is finite, A = V(X) and the measure fi is defined by n(A) =

cardA, then TA (X) = FS(X) and we obtain the definition introduced by
Klir [120].
2) T h e best known acceptations for the sharpness relation -< and for
fuzzy maximality introduced by De Luca and Termini [138] and further
investigated by m a n y authors (see e.g Batle-Trillas [35], De Luca-Termini
[139], Knopfmacher [122], R o m a n Flores-Bassanezi [34], Ban-Fechete [20]),
are the followings:
A -B(x) for B (x) > ]-. (2.2)
Measures oj Fuzziness 59
and
A is maximally fuzzy if and only if A (x) = - , Va: £ X (2.3)
(or A(x) — -,/i-a.e. x 6 X if (X, .^(X),^) is a measure space).
In the case when X is finite, the best known example in this sense was
given by De Luca and Termini [138]:
Example 2.1 The function dc : FS(X) -> R defined by
dc{A) = - ] T (A(x) log2 A(x) + (1 - A(x)) log2 (1 - A(x)))

x£X
f by convention 01og20 = 0) is a measure of fuzziness.
Theorem 2.1 (Loo [135]) Let X be a finite set. If the function h : R + —>•
R+ is increasing, the functions {gx)x£x >9* '• $> ^\ ~^ ^ + are increasing on
[0, | ] , decreasing on [|, l] such that gx(0) = gx(l) = 0, Va; £ X and gx ( | )
is the unique maximum value of gx,\/x £ X, then dc : FS(X) —>• R defined
by
dc(A) = h(^^(A(^))
\x€X )
is a measure of fuzziness.
Remarks. 1) If
gx (A(x)) = -A(x) log2 A(x) - (1 - A{x)) log2 (1 - A{x)),
for every x 6 X and h is the identical function on R+, then we get the
measure of fuzziness in Example 2.1.
2) When for a given w £ [1, oo], we take h(b) = ft™ and
{A{X))W
a (A(x))-l > *^*)e[0,|]
gx(A(x))-^ (1 _i4(ar)r> if^ ) G [I,l]
for every x 6 X, the measures of fuzziness proposed by Kaufmann [114] are
obtained. In this way, a measure of fuzziness is introduced as the distance
60 Defect of Property in Set Theory
( H a m m i n g if w = 1, Euclid if w = 2, Minkowski if w £ [1, oo]) between the

fuzzy set A and its nearest crisp set CA, t h a t is
G A W
- \ 1, XA(z)>±.
Sometimes, additional conditions are required to the concept of measure

of fuzziness. T h u s , an usual one is the equality between the measure of
fuzziness of a fuzzy set and the measure of fuzziness of its complement. In
this sense, we introduce the concept of fuzzy complement.
Definition 2.2 (see Klir [120], Rudas-Kaynak [181]) A function N :

[0,1] —» [0,1] is a fuzzy complement, if for all a,b £ [0,1], the following
axioms are satisfied:
(i) N(0) = 1 and N(l) = 0, t h a t is N gives the same results as the
classical complement for crisp conditions;
(ii) If a N (6), t h a t is N is monotonically decreasing;
(Hi) N is a continuous function;
(iv) N (N (a)) = a, t h a t is iV is involutive.
E x a m p l e 2.2 If t G (0,1) then Nt : [0,1] -> [0,1] defined by

^ i , if a; < t
Nt(x) = t
•£i(l-x), if a? > *
is a fuzzy complement. Indeed, Nt(Q) = 1, Nt(l) = 0, the continuity and

the monotonicity are obvious. Because Nt(x) £ [t, 1] if and only if x £ [0,t]
and Nt(x) £ [0,t] if and only if a; £ [t, 1], we get
Nt(Nt(x)) = N t ( l - ^ - x
1 - 1 + ^—^-x ) =x,\/xe [0,t],

i-1
and
Nt(Nt(x)) = Nt[-L-(i-x)
l
~ V(rr7 ( 1 _ a ; ) ) = a;'Va;€^1]'
Measures of Fuzziness 61
Definition 2.3 Let N be a, fuzzy complement and A 6 FS(X). The

fuzzy set A~ G FS(X) defined by ~A~N(X) = N(A{x)), for every x G X, is
called the complement of A.
Remark. The conventional complement of a fuzzy set A, A(x) = 1 — A(x),

for every x G X, is obtained considering as fuzzy complement IV : [0,1] ->
[0,1] defined by N(a) = 1 - a.
Another axiom considered as natural in the definition of measures of
fuzziness on T(X) C FS(X) is
dc(A) = dc(AN), (2.4)
for every A G T{X), if ~AN G T(X) too.

Remark. In Knopfmacher [122] and Vivona [217] the conventional com-
plement is considered.
Also, in Knopfmacher [122] are added the following requirements for a
measure of fuzziness:
(i) dc (AVB) + dc(A A B) = dc(A) + dc(B), VA, B e TA (X), where
(A V 5 ) (x) = max(A(x),B{x)) and (A A B) {x) = min(A(z), B(x));

(ii) dc is a continuous function on TA (X) relative to uniform metric p
on TA (X) , p (A, B) = sup s G J r \A(x) - B(x)\;
(Hi) The restriction of dc to the family of constant fuzzy sets (A a ) a g r 0 n
(that is Aa(x) = a, Va; G X), is strictly increasing function of a.
In the same paper [122], a family of measures of fuzziness (in normalized
form) which verify all these conditions is given, as follows.
Theorem 2.2 Let (X,A,p) be a measure space with 0 < p. (A) < +oo
and TA (X) the set of all fuzzy sets A on X that are measurable as real-
valued functions. If A denotes an arbitrary real-valued function of a £ [0, 1],
such that A(0) = A(l) = 0, A(a) = A(l - a),Va £ [0,1] and A is strictly
increasing on [0, | ] , then dc : TA (X) —¥ R defined by
is a measure of fuzziness which verifies the conditions in Definition 2.1 (in

the sense of (2.2) and (2.3),), (2.4) (with the conventional complement) and
the above conditions (i) — (Hi), where J denotes the Lebesgue integral.
Replacing the conventional interpretations (where the remarkable value

is \) of the relation "sharper than" and of property "maximally fuzzy" in
Remark after Definition 2.1, we obtain the following more general and useful
definition of ^-measures of fuzziness, the fixed value t £ (0, 1) becoming
important.
Definition 2.4 Let (X, A) be a measurable space and t £ (0,1). A t-
measure of fuzziness is a function d^ : FA (X) —>• R that satisfies the
following conditions:
(t) If A[x) £ {0,1}, Vz G X then d*c (A) = 0;
(n) If A -<t B then dfc(A) < d'c(B), where A <t B if and only if
A{x) < B{x) for B(x) < t and A(x) > B(x) for B(x) > t.
(Hi) If A(x) = t,Vx £ X then dÂ) is the maximum value of d\.
Definition 2.5 Let (X, A, fi) be a measure space and ( £ (0,1). A func-
tion 4 : TA (X) -> R that satisfies the following conditions:
(i) s\{A) = 0 if and only if A(z) G {0, 1} ,/i-a.e ar £ X;
(ii) li A^t B then 4 ( A ) < 4 ( 5 ) ;
(m) 4(^4) i s * n e maximum value of 4 if a n d only if A(x) = t,fi-a,.e.
x£X,
is called strict ^-measure of fuzziness with respect to JX.
Remarks. 1) In general, if X is finite then we consider A = V(X) and

fj, (A) = cardA, VJ4 G ^(-X"), in this case p-a.e meaning everywhere.
2) It is obvious that any strict ^-measure of fuzziness with respect to a
measure fi, is a ^-measure of fuzziness.
Definition 2.6 Let TV be a fuzzy complement. A i-measure of fuzziness

is called symmetrical with respect to TV for TV-symmetrical^ if
dtc(A) = dtc(AN),VA£fA(X). (2.5)

Remark. The most natural fuzzy complement for ^-measures of fuzzi-
ness is that introduced in Example 2.2. If t = \ we get the conventional
complement.
Based on the idea in above Theorem 2.1, we can give a family of t-
measures of fuzziness. The result is more general than Theorem 2.1 in
Ban-Fechete [20].
Theorem 2.3 Let X be a finite set, t G (0,1) and h : R + -> R + increas-
ing, such that h(0) = 0, (9x)x^x i9* '• IP> 1] —^ **-+ increasing on [0,t] and
Measures of Fuzziness 63
decreasing on \t, 1], such that gx(Q) — gx(l) = 0,VJ; £ X and gx(t) is the
maximum value of all functions gx,x G X. The function dfc : FS(X) —> R
defined by
dte(A) = h('£is*(M*)))
\x£X I
is a t-measure of fuzziness. If, in addition,
gx(a) = gx (N(a)), Va G [0,1], Vz € X,
where N is a fuzzy complement, then d\ is a N -symmetrical t-measure of

fuzziness.
Proof. We verify the conditions in Definition 2.4.
(i) If A(x) E {0,1}, Va? G X then gx (A(x)) = 0, Va? G X and
T,*ex9x (M*)) = 0. therefore d*e{A) = h{0) = 0.
(u) If A -<t 5 then the monotonicity of functions <7X, a? 6 X implies
gx (A(x)) < gx (B{x)) ,Va? G X. This means J2xtx9x (Mx))
c
< XêJ<:#£ (•^(• ))i that is d£(A) < rf*(-S) because h is increasing.
(in) rf'(A) assumes the maximum value if J2xzx9x (^(• c )) assumes the
maximum value, that is A(x) = t, Va; G -X".

Relation (2.5) is also satisfied because
dte<Arf)=ht'£igs(N(A(x)))\ = h("£gx(A(x))) =d{{A),

\x£X / Wx /
therefore d£ is a ./V-symmetrical t-measure of fuzziness. D
Example 2.3 Let X be a finite set and t G (0,1). If h : R + ->•
7?+, h(x) = x and gx : [0,1] —> R+ are defined by
f a, ifae[0,<]
9
*W-{ ^(a)] ifae[M];
where TV is a fuzzy complement, then the conditions in Theorem 2.3 are
verified, therefore
dl(A)=J29x(A(x))
xex
is a N-symmetrical t-measure of fuzziness. Indeed, gx(0) = 0 and gx(l) =
gx(N(l)) — gx{0) = 0,Va; G X,gx are increasing on [0,t] and decreasing
on [t, 1], Va; G X, because TV is decreasing. Also, t is the maximum value

of gx if a £ [0,f] and N(t) is the maximum value of gx if a £ [t, 1],
Wx £ X, therefore ^ ( t ) is the maximum value, Va; 6 X. In addition, TV
being involutory implies
N(a), if N(a) G [0,t]

9s(N(a)) =
N{N(a)), i{N(a)e[t,l]
N(a), ifae[<,l] _ , .
ifa6[0,t] ~9â>-
Under additional conditions, we obtain a family of strict ^-measures of

fuzziness.
Theorem 2.4 / / the functions h and (gx)xcx satisfy the hypothesis in

Theorem 2.3 and, in addition, h is strictly increasing, (gx)xpx are strictly
increasing on [0,t] and strictly decreasing on [t, 1], then the function s^. :
FS(X) ->• R defined by
»tM) = h(^2gx(A(x))
\x6X /
is a strict t-measure of fuzziness which is TV-symmetrical ifgx{a) =

gx (N(a)), Va £ [0,1] , Va: G X, where N is a fuzzy complement.
Proof. We must prove only the necessity in conditions (i) and (in) of
Definition 2.5, the other requirements being proved by the proof of Theorem
2.3.
If 4 ( A ) = O t h e n ^ e x ^ ^ a ; ) ) = 0, which implies A(x) G {0,1}, Va; £
X, because the functions gx are strictly increasing on [0,t] and strictly de-
creasing on [t, 1].
If Sg(j4) is the maximum value of s', then Ylxcx9x (J^(*c)) assumes the
maximum value, that is gx (A(x)) is the maximum value of gx, Va; G X.
The strict monotonicity of gx on [0, t] and on [t, 1], for every x £ X, implies
A(x) =t,Vx e X. •
In what follows, we give an example of strict t-measure of fuzziness of

entropy type. In the particular case t = i , we obtain the ^-measure of
fuzziness introduced by De Luca and Termini (see Example 2.1).
Measures oj Fuzziness 65
Example 2.4 Let X be finite, t G (0,1), ut : [0,1] ->• [0,1] defined by
and H : [0,1] -» R + defined by if (y) = - y log2 y - (1 - y) log 2 (l - J/) ("by

convention, 0 log2 0 = 0). We denote gx = H o « t , Vx G X. The function u t
is strictly increasing on [0,1]. If a £ [0,t] then ut(a) £ [0, | ] and H being
strictly increasing on [0,1], we obtain that the functions gx are strictly
increasing on [Q,t],\/x G X. Analogously, if a G [t, 1] then wt(a) G [|, l]
and because the function H is strictly decreasing on [|, l ] , we obtain that
the functions gx are strictly increasing on [t, 1] , Vz G X. The monotonicity
of H implies that the function gx assumes the maximum value if and only if
the function ut has the value equal to \, that is, if and only if the argument
of ut and implicitely of the function gx, x G X, is t. Because
0,(0) = H (« t (0)) = H{0) = 0, Vz G X,
and
«M1) = if (u t (l)) = H(l) = 0,Vx G X,

we get that the functions (gx)X£x ver ify the hypothesis in Theorem 2.4.
Because h : R + —>• R + , /i(a;) = x, also verifies these hypothesis, we obtain
that the function defined by
Hl{A) = -J2xex (ut (A(x)) log2 ut (A(x))

+ (l-ut(A(x)))log2(l-ut(A(x))))
is a strict t-measure of fuzziness. We prove that this i-measure of fuzzi-

ness is Nf-symmetrical, where Nt is the fuzzy complement introduced in
Example 2.2. Indeed, if o G [0,<], then Nt(a) G [t, 1] and
ut(Nt(a)) = ——-r +
2(1 -t) 2(1 -t)
1 - i ^ a + 1 - 2* , a ,
= 2(TT^ = l - 2 i = l-«t(«).
If a G [*, 1], then iVt(a) G [0, *] and
«tW(a)) =
2t 2* 2(1 — t)
a
, ( l-2t \ , . ,
Because H(y) = H{1 — j/), Vj/ £ [0,1], we obtain
Hl{ANt) = "£H(ut(Nt(A(x))))='£H(l-ut(A(x)))
x£X x^X
t
= J2H(ut(A(x))) = H c(A),VAeFS(X).
xex
Other families of ^-measures of fuzziness can be obtained by using the

concept of t-norm function in the sense of Vivona [217].
Definition 2.7 Let t 6 (0,1) and (X,A,fi) be a measure space. An A-
measurable function with respect to the first variable, <ft '• X x [0,1] —>• [0,1]
that satisfies the properties
(*) <pt{x,Q) = tpt{x,l) = 0,VxeX;
(ii) <ft{x, •) is increasing on [0,t] and decreasing on [t, 1];
(Hi) <pt[x,t) = l,Va: E X,
is called i-norm function.
Example 2.5 The function ipt : X x [0,1] —> [0,1] defined by <pt(x, a) =
(H o ut) (a), where H and ut are defined as in Example 2.4, is a t-norm
function.
Corresponding to Theorem 2.3 we prove the following result.
Theorem 2.5 Let t 6 (0,1), (X,A,fi) be a measure space and <pt : X x
[0,1] —» [0,1] a t-norm function. The function d^ : Tj. (X) —» [0,1] defined
by
d£*(A)= I Vt{x,A{x))dn
Jx
is a t-measure of fuzziness. If N : [0,1] —> [0,1] is a fuzzy complement and
the t-norm function <pt verifies
(ft (x, N(a)) = (pt (x, a), Vx £ X, Va G [0,1],

then the t-measure of fuzziness d|?'(A), defined as above, is N-symmetrical.
Proof. It is immediate by using the properties of ipt and the properties
of Lebesgue integral. •
Intuitionistic Entropies 67
R e m a r k . If t h e i-norm function ipt is strictly increasing on [0,£] a n d

strictly decreasing on [t, 1], then the i-measures of fuzziness in T h e o r e m 2.5
are strict.
2.2 Intuitionistic Entropies
Let X be fixed. A n intuitionistic fuzzy sets (see e.g. Atanassov [5]) A on

X is an object having the form
A = {{x,nA(x),vA{x)) : x £X} ,
where t h e functions fJ,A,vA '• X —> [0,1] define the degree of membership
and t h e degree of non-membership of the element x £ X t o t h e set A C X,
respectively, a n d for every x £ X, HA{X) + VA{X) < l . W e denote by IFS(X)
the family of intuitionistic fuzzy sets on X.
We recall t h e following useful relations a n d operations on IFS(X) (see
Atanassov [5]-[10], Atanassov-Ban [14]):
AC B if a n d only if Â{X) < ^B{X) a n d vA(x) > VB{X), VX £ X;

A = B if and only if fiA(x) = HB{X) and VA{X) = vB(x),Vx £ X;
A A(x)) : x £ X} and
B = {(x, HB{X), VB{X)) : x £ X}. Also, if h is a t r i a n g u l a r n o r m ( t h a t is, an
associative a n d commutative binary operation h : [0,1] x [0,1] —> [0,1] such
t h a t h (x, 1) = x, Va; £ [0,1] and y < z implies h (x, y) < h [x, z) , \/x £ [0,1])
or a triangular conorm (that is, an associative a n d c o m m u t a t i v e binary
operation h : [0,1] x [0,1] -)• [0,1] such t h a t h(x,Q) = x,Vx £ [0,1] a n d
y < z implies h(x,y) < h (x,z) ,\/x £ [0,1]), we define (see Burillo-Bustince
[46])
AhB - {(x,h(/j.A{x),HB{x)),hc{i'A(x),UB(x))) : x £ X} ,
where hc (x,y) = 1 — h (1 — x, 1 — y) , Mx, y £ [0,1], is t h e triangular norm

(conorm) corresponding to triangular conorm (norm) h. We also notice
t h a t these operations can be extended to countable case (see B a n [17]).
T h e intuitionistic entropy, introduced by Burillo-Bustince [47], can be

considered as a measure of the level of intuitionism of an intuitionistic
fuzzy set, i.e. it is a quantitative expression of the difference between an
intuitionistic fuzzy set and a fuzzy set, similar to the fuzzy entropy (or
measure of fuzziness) which measures the difference between a fuzzy set
and a crisp set.
In the above cited paper, the following formal conditions are required
for an intuitionistic entropy:
(i) to be null when the set is a fuzzy set;
(ii) to be m a x i m u m if the set is totally intuitionistic;
(Hi) as in the case of fuzzy sets, the entropy of an intuitionistic fuzzy
set has to be equal to the entropy of its complement;
(iv) If the degree of membership and the degree of non-membership of
each element increase, the sum will do as well, and therefore, this set will
become more fuzzy, and the entropy will decrease.
Taking into account the previous considerations, in the case when X is
finite, we give the following definition.
D e f i n i t i o n 2.8 (Burillo-Bustince [47])A real function dj : IFS{X) -»

R + is called an intuitionistic entropy on IFS(X), if df has the following
properties:
(i) df(A) = 0 if and only if fiA(x) + vA{x) = 1, Vz £ X;
(ii) df(A) = cardX = TV if and only if fiA(x) — vA(x) — 0, \/x £ X;
(Hi) df(A) = df(A),VA £ IFS(X);
(iv) UA 1(B).
We reformulate this definition in a more general frame, even if in Ban-

Gal [23], a result of approximation of intuitionistic fuzzy sets by discrete
intuitionistic fuzzy sets ( t h a t is functions whose degrees of m e m b e r s h i p
and degrees of non-membership take a finite number of values) reduces, in
a certain sense, the infinite case to the finite case.
Let us consider (X,A,m) a finite measure space and let us denote by
1A (X) the family of all .4-measurable intuitionistic fuzzy sets on X, t h a t
is A = {(x,fiA(x), vA(x)) : x £ X} G XA(X) if and only if [iA and vA are
^4-measurable functions.
Definition 2.9 A real function dj : 2A (X) ->• R + is called an intuition-

istic entropy (on IA (X)) if the following properties are satisfied;
(i) df(A) = 0 if and only if fiA(x) + vA(x) = 1, ra-a.e. x £ X;
(it) dj(A) is maximum if and only if fiA(x) = VA(X) — 0, m-a.e. x £ X;

(Hi) df{A) = df(A),VAeiA(x);
(iv) If A, Be 1A (X) and A df(B).
Remark. The particular situation in Definition 2.8 is obtained if X is

finite, A = V (X) and m (A) = cardAÂ G A.
In what follows we consider only intuitionistic entropies in the normal-
ized form, that is their maximum values are equal to 1.
Definition 2.10 Let (X,A,m) be a measure space and let us denote
D = {(p., v) G [0,1] x [0,1] : n + v < 1} . An intuitionistic norm function is
an ,4-measurable function with respect to the first variable, $ : X x D —>
[0,1] , with the following properties, for every element x £ X :
(?) <3> (x, fi,i/) = 0 if and only if p. + 1/ = 1;
(ii) $ (x, p, v) = 1 if and only if p = v = 0;
(Hi) <&(x,p,v) = ®(x,i/,p);
(iv) If p < p' and v < v' then $ (a;, p, v) > $ (a;, //, i/) .
We obtain families of intuitionistic entropies by using intuitionistic norm

functions and Lebesgue integral, similar to Knopfmacher [122] for the case
of fuzzy entropy.
Theorem 2.6 Let (X,A,m) be a finite measure space and $ an intu-

itionistic norm function. Then, the function dj : IA (X) —> [0,1] defined
by
df(A) = ^(x,pA(x),vA(x))dm
m(X) Jx
with A = {(x, HA(X), VA{X)) '• x £ X}, is an intuitionistic entropy.
Proof. Let A,B e 1A(X),A — {{x, pA(x),i/A(x)) : x G X} and
B = {{X,HB(X),VB(X)) '• x G X}. We notice that A = B m-a.e., that is
PA(X) = VB(X) and VA(X) — VB(X) m-a.e. x G X, implies $ (x, PA(X), VA(X))
= ^(X,PB{X),IIB(X)) m-a.e. x G X, therefore dJ(A) = dJ(B).
We verify the conditions (i) — (iv) in Definition 2.9.
(i) HE = {{X,HE(X),UE(X)) : x £ X} £lA (X) verifies pE(x)+vE(x) =
1, Vx € X, then
^ £ » = ^ o X o d ™ = 0-
70 Deject oj Property in Set Theory
If A (ZIA (X) is an intuitionistic fuzzy set with dJ(A) = 0, then

$ (x,/iA(x), VA{X)) = 0 ra-a.e. x £ X and Definition 2.10, (i), implies
HA(X) + vA(x) = 1 m-a.e. a; 6 X .
(«') If i? verifies HE{X) = ^ ( E ) = 0, Va: £ X, then
^ B > = sno/x
.(X)
ldra = 1
If d^(A) = 1 then J „ $ (a;, ^ ( a : ) , ^,i(a:))dra = fx l d m , t h a t is

$ (ar,//^(a;), i^(a;)) = 1 ra-a.e. x £ X. By Definition 2.10, ( i i ) , we get
IÂ(X) = vA(x) = 0 m-a.e. x £ X .
(in) , (it;) Are immediate by t h e properties ( m ) and (iv) in Definition
2.10. •
R e m a r k . A m o n g all t h e possible intuitionistic entropies previously intro-

duced, t h e most n a t u r a l and simple is obtained by setting 3>(X,/J,,I/) =
1 — fi — v, t h a t is
d
*(^) = -r7T I (i-^W-^Wl^vieiifi).
m A
l J Jx
T h e o r e m 2 . 7 Let (X,A,m) be a finite measure space and <£ be an intu-

itionistic norm function. If $ is a continuous function with respect to the
second and third variables, then the intuitionistic entropy df : 1 A {X) —>
[0,1] defined by
tf(A) =
fv\ ®{x,(iA(x),vA(x))dm
rn(A) Jx
is a continuous function with respect to the metric d : 1 A (X) x 1 A (X) —>

[0,1] , defined by
d(A,B) = m a x ( sup \fiA(x) - HB{X)\ , sup \uA(x) - i/B(x)\) ,

\xex xex J
where A = {(x,fiA(x), vA{x)) : x 6 X} , B = {{X,/IB{X),I/B(X)) : x e X} .
Proof. Because D = {(/i, u) € [0,1] x [0,1] : \i + v < 1} is a compact set,

$ (x, •, •) is uniformly continuous function. Let e > 0. There is S > 0 such
t h a t x £ X with \fiA(x) - HB(X)\ < S and \vA(x) — vB{x)\ < S implies
\®{x,[iA{x),i>A(x)) -${X,HB(X),I>B{X))\ < e.
But d(A,B) < S implies \/J-A{X) — HB{%)\ < $ and \VA{%) — vB{x)\ < S,^x e
X, therefore
| $ (a;, HA(X), VA(X)) - $ (a;, nB{x),vB{x)) \ < e, Va; <E X.
We obtain the existence of a 6 > 0 such that d(A, B) < 6 implies
— - / $(i,^(i),i/A(i))dm —- / ${x,nB{x),vB{x))dr <e

m{X)
\A) Jx m(A) Jx
therefore \I*{A) - I*{B)\ <e. D
In Burillo-Bustince [47] for the finite theorem of characterization

of intuitionistic entropies by using /^-functions is given. The idea can
be extended to general case, by considering particular intuitionistic norm
functions.
Definition 2.11 Let (X,A, m) be a measure space and [0,1] defined by
Iv =
i1 ~ A{%)) '• % 6 X}, is called /^-function.

Theorem 2.8 Let (X,A,m) be a finite measure space, • [0,1]. The function dv is an
intuitionistic entropy and an Ip-function if and only if
d<
piA) = Z^lF\ I (l-<p(MA(x))-<p{vA(x)))dm,
m(A) Jx
where ip satisfies the conditions:
(i) <p is increasing;
(ii) <p(a) = 0 if and only if a = 0;
(Hi) tp(a) + <p(P) = 1 if and only if a + /? = 1.
Proof. (<=) Let us consider $ : X x D -»• [0,1], $ (a;, n, v) — 1 - <p (fi) -

tp (u), where D = {(/i, v) € [0,1] x [0,1] : n + v < 1} . The function $ is
constant with respect to x, therefore it is A-measurable with respect to the
first variable.
We verify (i) — (iv) in Definition 2.10.
(i) $ (x, /j,, v) = 1 — (<p(fj) + (//) + <p (v) = 1 if
and only if fi + v = 1.
(ii) $ (ar, //, z^) = 1 if and only if <p(/i) + ip (v) = 0 if and only if tp(/j.) =
<p(v) = 0 if and only if /z = ^ = 0.
(m) $ (a;, H,V)=1- <p(n) -<p(v) = $ (X, V, fi) .
(iv) If fj, < fj,' and v < v' then ip(fi) + tp(v) < tp(ft') + <p(v'), that is
<&(x,n,v)>$(x,n',u').
$ being an intuitionistic norm function, Theorem 2.6 implies that dv is
an intuitionistic entropy.
If a + /3 < 1 then
dv({(x,a,0):xeX}) = -}— [ (l-<p(a)-p(P))dm
= l-<p(a)-<p(l3)>0,
which implies ) We assume that d^ is an intuitionistic entropy and an /^-function.
Being /^-function it has the form
<1<P(A) = m—nF\
A
I ^-f{Â{x))-<p{vA{x)))Am
l ; Jx
if A = {{X,P,A(X), VA{X)) '• x G X], where [0,1] is a continuous
function such that a + /? < 1 implies <p(a) + <p((3) < 1. We prove that <p
satisfies the conditions (i) — (Hi) .
(i) If a < a1 and a, a' 6 [0,1], we construct the following intuitionistic
fuzzy sets: A — {(x, a, 0) : x e X} and B = {(x, a1, 0) : x 6 X } . Because
A^B,it follows /(>!) > 1(B), that is
/ (I - f {l-<p(a')-ip{0))dm

Jx Jx
and
l-ip(a)-ip(0)>l-<p (a1) -<p(Q).
We obtain cp (a) < ip (a1) .

(ii) dv ({(a;, 0,0) : x € X}) = 1 by Definition 2.10, (ii). But
dlp({{x,0,0):xeX})
= -Tv^l ( 1 - P ( 0 ) - P ( 0 ) ) d m = 1-2^(0),
therefore 1 = 1 - 2^(0) and y(0) = 0.

Conversely, if p (0)) dm

m(A) Jx
= 1 - p(a) = 1
and Definition 2.10, (n), implies a = 0.

(m) If a + /? = 1, we take A = {(x, a, 0) : x £ X} . We know by Defini-
tion 2.10, (i), that dv(A) = 0. But
d
«>W = T^ln
m(A) /"
Jx (1 - ^ («) - p(/3)) dm = 1 - ^(a) - p(/?).
We obtain 9? (a) + <f{/3) — 1.
Conversely, let y> (a) + <p(P) = 1. If we assume that a + /? 7^ 1, then two
cases may occur:
(a) a + p < 1
Then
d„({0c, a, /?) : x € X}) = — i — [ (l-<p(a)- !
Because a + (1 - a) = 1 and /?+ (1 — /?) = 1, we get y (a) + (ar) + y > ( / ? ) + p ( l - a ) + p ( l - / ? ) = 2 ,
therefore ^> (1 — a) + ^>(1 — /?) = 1.

By 1 - a + 1 - /? = 2 - (a + /?) < 1 we get
dv({<a:,l-a,l-/?> : i ^ } )
= - T W / ( l - v ( l - « ) - ¥ ) ( l - W d m = 0.
Then Definition 2.10, (i), implies 1 — a + 1 — / ? = 1, therefore a + /? = 1, in

contradiction with the hypothesis.
We obtain the unique possibility a + (3 = 1. D
Further we need to introduce the concept of intuitionistic fuzzy measure

with respect to a triangular norm.
Definition 2.12 Let h be a triangular norm or a triangular conorm and
X ± 0 . j \ family X C IFS{X) that satisfies;
(i) Ox e X, where Ox = {(x, 0,1) : x G X};
(ii) A G X implies A EX;
(Hi) (.A n ) n 6 N C X implies hn€NAn EX,
is called intuitionistic fuzzy /i-algebra and the pair (X,X) is called intu-
itionistic fuzzy ^-measurable space. (Here hn^jqAn extends the operation
h to countable case.)
Example 2.6 If (X, A) is a measurable space then XA (X) is an intu-
itionistic fuzzy h -algebra, for every continuous triangular norm or conorm
h.
Definition 2.13 Let (X, X) be an intuitionistic fuzzy /i-measurable space.
A function fh : X —> R + is called intuitionistic fuzzy /i-measure if it satisfies
the following conditions:
(i) m ( o x ) = 0;
(ii) A,B EX implies fh(AhB) + fh(AhcB) = rh(A) + fh(B);
(Hi) ( A i ) „ e N CX,AnC An+i,Vn G N and limn^Ân G 1 imply

lim n _ +0O m( J 4 n ) = m(limn_>ooA0-
Between intuitionistic fuzzy measures and intuitionistic entropies we can
establish the following connection.
Theorem 2.9 Let (X,A, m) be a measure space. If d^ : XA (X) —)• [0,1]
is an intuitionistic entropy and an I^-function , then dv is an intuitionistic
fuzzy TM -measure on XA (X), where TM is the triangular norm given by
TM (X, y) = min(x, y) . Moreover, ifip is additive then d^ is an intuitionistic
fuzzy h-measure on XA (X), for every continuous triangular norm h which
verifies h(x,y) + hc (x, y) = x + y,Vx,y E [0,1].
Proof. By Theorem 2.8 we have
d
f^ = ^TrY / {I ~ <P {»A(X)) - v (vA(x))) dm,
m(A) Jx
with ip verifying the conditions (i) — (Hi) in the same theorem. We obtain
d^(0x) = d^({(x,0,l):xEX})
=
^y^(1_(¥'(0)+¥'(1)))dm =0
'
Let (^4n)n£N ^ ^ (^0 be an increasing sequence of intuitionistic fuzzy
sets, An = {{x,fiAn(x),vAn(x)) : x EX}. Then
limd„(A„) = lim — — / ( l - ( p (PA,, (*)) + ¥> ( " A . (a:)))) dm

• CO 71 —J- OO / 1 1 1 , ^ \ 1 I Y
= du> \(x, lim fiAn(x), lim J/A„(«)) :xGl)=(i«, lim A„) ,
because <p is continuous (see Ban [17]).

If ip is additive and h (x, y) + hc (x, y) = x + y, Vx, y £ [0,1] then we
have
d^, {XhB\ + dv (AhcB\
=
^TTV / t 1 - (^ C1 (Â^), /ifl(^))) + *J C»c (^(a;), ^B(S))))) dm
+—— / (l-(<p(hc(iiA(x)iiiB(x)))+<p(h(vA(z),i/B(x)))))dm
= ~^Tx) J {1-{<p{HA{x)) + <p{nB{x)) + <p{vA{x))+<p{vB{x))))&m
=
m(x) J ( 1 -(^(^( a ; )) + ^(^( ; , ; )))) d m
= dv(A) + dv(B).
For / = Tjvf the proof is similar and the conclusion is true even if <p is not
additive. •
Another concept of entropy for intuitionistic fuzzy sets was introduced

and studied in Szmidt-Kacprzyk [206], [207]. It is similar to other consid-
erations for ordinary fuzzy sets (see Definitions 2.1 and 2.4). In this sense,
if X is a finite set, A,B ElFS{X),A = {(x,piA{x),i>A(x)) : x G X),
B = {{X,HB{X)>VB{X)) '• x G X}, then the entropy for intuitionistic fuzzy
sets is a function which satisfies
(i) E (A) = 0 if and only if A is crisp;

(ii) E (A) — 1 if and only if HA{%) = ^ f * ) , Vi 6 X;
(in) E [A) < E (B) HA is less fuzzy than B, i.e. if Â{X) < ^B(^) and
vA{x) > vB{x) for HB(X) < VB{X) or fiA{x) > HB{X) and VA(X) < VB{X)
for HB{X) > VB{X);
(iv) E(A) = E(A) .
The condition (i) shows that an entropy in this sense, for an intuition-
istic fuzzy set can be considered as its defect of crisp set.
2.3 Applications
2.3.1 Application to determination of degree of interference

Let U be a finite non-empty set and V(U) be the family of all subsets of
U. A family X = {C\, ..., Cn} C V(U) is a partition of [/ if are satisfied
(«•) c , i nc i = 0,Vi,ie{i,...,n},^j.
We name atoms the elements of X and we denote by T{U) the class of all
partitions of U.
In what follows, we introduce a normal indicator Ipq(X) (that is

0 < IPQ(X) < 1) of the degree of interference between the elements of U
which have a certain property " P " and those with another property "Q"
(in this order), with respect to the partition X G T{U).
Let us denote by pk, the number of elements in C'k which have the
property " P" and by q^, the number of elements in Ck which have property
" Q". The value that presents the relation between the number of elements
with the property " P" and the number of elements that have at least one
of the properties "P" or "Q" on U, is
n
l = Efe=iPfc
E * = i (P* + 9k)'
In general, by degree of interference between the elements with prop-
erty " P " and those with property "Q", we will understand a function
IPQ '• T{U) —> [0,1] which satisfies the following requirements:
(II) IPQ(X) = 0 if and only if pk = 0 or qk = 0, for all k £ {1, ...n}, that

is, the indicator Ipq assumes the minimum value if and only if each atom
of the partition has only elements with the property " P " or only elements
Applications 77
with the property " Q " ;

(72) Let k,j G {l,...,n},k * j . If p - % - < rffc < t or p - % - >
—*£— > t then the contribution of Cj t o IPQ(X) is smaller t h a n t h a t of
(73) J P Q ( X ) = 1 if and only if p * = 2, for every & £ { 1 , . . . , n } , t h a t

is the indicator 7 P Q assumes the m a x i m u m value if and only if in every
a t o m of the partition X is fulfilled the global situation on U, between the
elements which have the property " P " and those with property " Q".
These three requirements seem t o be reasonable and intuitively accept-

able for the characterization of degree of interference.
Now, for every X £ T(U) we define the fuzzy set A : X —>• [0,1],
A(Ck) = p/+qfc iVk G {lj •••>"}) where Ck,Pk,qk are as above and we p u t
Ipq (X) = S ' ( J 4 ) , where s£ is a strict normalized ^-measure of fuzziness.
Due t o conditions (i) — (Hi) in Definition 2.5, we obtain the following result:
T h e o r e m 2.10 The function IPQ : T(U) —> [0,1] defined as above, veri-
fies the requirements (II) — (13).
Proof. It is obvious. •
It is often n a t u r a l to admit t h a t the degree of interference between the

elements of a set with respect to two properties " P " and " Q", does not
depend on order, t h a t is IPQ = IQP-
T h e o r e m 2 . 1 1 If IPQ and IQP are evaluated with the strict t-measure

of fuzziness Hi in Example 2-4, given in normalized form, then IPQ(X) =
iQP(x)yxeT(u).
Proof. It is obvious t h a t the value which presents the global situation
between the number of elements with the property " Q" and the n u m b e r of
elements t h a t have at least one of the properties " P" or " Q" on U, is given
by
n
_ Ek = l1k =1—1
Y2=i{Pk + <}k)
We denote A(Ck) = ^ ^ = 1 - p ^ ^ V f c € { 1 , . . , » } . If t h e strict t-

measure of fuzziness i j ' is used to calculate IQP , then
IQP(X) = Hi'(A) = H1^ (A)
= -T^-lJ2u^(1-A(Ck))log2u^t(l-A(Ck))
where |X| is the cardinal of the set X.

Because (see Example 2.4)
I &, ifaG[0,<]
ut(a)
I 5 ^ + ^ . if«e[<,i],
l-o
Ul
-t(1"a) =
| ^ ? ifae[0,t],
we get wt(a) + «i_t(l — a) = l,Va £ [0,1], therefore
IQP(X) = - 4 T E (! - u< ( A ( C *))) loS2 (1 - «t (^(C*)))

R e m a r k s . 1) We can use the ^-measures of fuzziness to determinate the

indicator IPQ(X) even iff ^ \. Nevertheless, for t near to 0 or 1, we get
the value of the indicator near to 0, which is not eloquent.
Tn - !PIP(X)
2) We can use the formulaIp1...pn{X) = —' l, ~ 1, 'g ? 3 '- , to calculate
the degree of interference of elements in U with respect to the properties
"Pi", ...,"Pn", where Ip,pi is the degree of interference of elements in U
with the properties " F , " and "Pj".
In what follows, we present a numerical example with applications to
the geography of population.
Let U be the set of habitants of a country having n districts and let us
denote by Ck, the set of habitants of the district k, k £ {1,..., n} . Also, we
consider two nationalities of this country, denoted by " P" and " Q". We
desire to determinate the degree of interference of these, knowing the result
of census of population (see table below). We denote by pk and by q/,,
the number of persons (expressed in tens of thousands, for example) from
Ck belonging to the nationality " P" and "Q", respectively. Because it is
Applications 79
natural to consider equality between the degree of interference of nationality

" P " with "Q" and the degree of interference of nationality "Q" with " P " ,
the above Theorem 2.11 justifies the t- measure of fuzziness given in Example
2.4 to determinate the value of indicator IPQ (X), where X £ T(U).
District Pk Qk County Pkj Ikj

1 8 10
2 10 12
1 30 50 3 6 8
4 2 7
5 4 13
1 8 5
2 3 8
2 30 20
3 7 4
4 12 3
1 6 14
3 20 40 2 10 7
3 4 19
1 20 6
2 5 28
4 40 70
3 6 22
4 9 14
If X = {Ci, C2, C3, C4} with the values in above table, then we obtain
t = 0.4. For this value, by using the indicated ^-measure of fuzziness, we
get
lpq (X) = 0.9723.
Now, let us assume that each district has more counties and let us denote by
Ckj, the set of habitants of the county j from district k. Also, we denote
by Pkj and q^j, the number of persons (expressed in tens of thousands)
from Ckj belonging to the nationality " P " and " Q", respectively. For this
territorial structure, that is for the partition X' = {Cii,Ci2> ...,644} (see
table), we obtain the degree of interference of nationality " P " with the
nationality " Q",
IpQ (X') = 0.8727.
The degree of interference corresponding to the districts, IPQ (X), is close to
1 because the ratio of the number of habitants of nationality " P " and the to-
tal number of habitants, calculated for each district ( ^ , k G {1,2,3,4})
is relatively close to the same ratio calculated for the level of whole country,
that is t = 0.4. IPQ (X1) has a smaller value because in more counties the
ratio —^r*— have remote values from t = 0.4, for example „ PV„ = 0.8
and P" = 0.15 (see also (12)).
V
P42+942 \ II
Modifications of the indicator of degree of interference (regarding more

census of population or with respect to different territorial structures) leads
to interesting conclusions from the geography of population viewpoint.
Finally, we observe that other indicators in geography of population can
be introduced with the help of ^-measures of fuzziness.
Let V be a country and let us denote by X — {Si,..., Sn}, a partition of
V, that is Si, i G {1,..., n} represent all the districts of V. We can introduce
a normal indicator Tu,x (that is 0 < Tux < 1) which estimates the degree
of homogeneity of territorial distribution of population in V, with respect
to the organization corresponding to X. We notice that, the density of
population, the usual indicator used in this situation, is a global indicator
which does not take account the pointwise aspects. For example, if there
exists a great concentration of population in certain zones, this indicator is
not significant.
Let us denote by Tme({, the density of population in V, that is
total population in V
1 med — r~Tr '
area of V
by TMAX, the maximum possible density of population in a district of V,
that is
__ total population in V
area of the smaller district 5,
and by I \ , the density of population in Si,i G {1,..., n}, that is
population in 5,-
area of Si
We define the fuzzy set P : X -> [0,1] by
P(Si) = - ^ ,
l-MAX
and we denote t — r r"?d .
Applications 81
We put Tv,x — SC{P)J where s' is a strict t-measure of fuzziness in

normalized form. Following the axioms of strict t-measures of fuzziness
(see Definition 2.5), we get properties of the indicator Ty,x which are in
concordance with our intuition:
( 0 1 ) I Y x has the m i n i m u m value (equal to 0) if and only if the entire

population of V is completely situated in the district with the smallest area;
( 0 2 ) If Tj < Ti < t or t < Ti < Tj then the contribution of the district
S{ to Tv,x is greater t h a n the contribution of the district Sj;
( 0 3 ) Tx,u has the m a x i m u m value (equal to 1) if and only if in all
districts of V the density of population is equal to Tmeci.
Choosing a suitable strict ^-measure of fuzziness (for example, t h a t given

in Example 2.3) we can get interesting results concerning the territorial
distribution of population.
2.3.2 Application to description of the performance of systems
Fuzzy m e t h o d s in the study of the performance of systems are described

in m a n y papers (see, for example, Kaleva [ i l l ] and [112], Kaufmann [114],
Kaufmann-Grouchko-Cruon [115], Misra-Sharma [150], Soman-Krishna

[20l]), taking into account t h a t the performance level of a system is a value
included in [0,1]: the value 0 corresponds to non-working and the value 1
corresponds to o p t i m u m working.
In Kaufmann [114] and Kaleva [ i l l ] a m a p p i n g indicating the perfor-
mance level of a system as function of level of performance of its components
is discussed. But, it is often i m p o r t a n t to know the degree of homogene-
ity of a system, t h a t is how much the performance level of components is
different from the performance level of system. In this application, we in-
troduce the degree of homogeneity of a system with the help of ^-measures
of fuzziness, by using its performance function.
Let X = {xi, ..., xn} be the set of components of a system. If to each
component Xi we attach a value a; = A(xf) £ [0,1] indicating the perfor-
mance level of t h a t component, then the working of system is described
by the fuzzy set A : X —> [0,1]. For a coherent system (that is, a system
having a series-parallel network representation) we can consider a func-
tion indicating the performance level of the system, as a function of the
component performances. We denote this function by $ and we call it,
performance function (see Kaleva [112], Kaufmann [114]).
We introduce a normal index H<$tx (that is, 0 < H^x (A) < 1, for
every fuzzy set A) of the degree of homogeneity for a system with the set
of its components, X = {x\, ...,xn} and the performance function \P. We
assume t h a t t = ^ (a\,..., an) 6 [0,1] is the performance level of the system
X, where a,- is the performance level of component Xi, for i 6 {1, . . . , n } .
T h e following four requirements appear to be reasonable and intuitively
acceptable to characterize the degree of homogeneity H^x •
(HI) H$tx (A) = 0 if and only if a,- = 0 or a,- — 1, for all i £ { 1 , . . . , n},
t h a t is the index H%x assumes the m i n i m u m value if and only if all the
components of the system are in two states: operating or failed.
(HI) Let i,j £ { l , . . . , n } . If a; < a,j < \P ( a i , ...,a„) or a* > a,j >
vp ( a i , ...,a„) then the contribution of the component X{ to H^x (A) is
smaller t h a n t h a t of the component Xj, t h a t is the contribution of a com-
ponent is greater if its working is near to the working of the system.
(H3) H<ix (A) = 1 if and only if a,- = ^ ( a 1 ; . . . , an), for all i £ {1, ..., n},
t h a t is the index H^x assumes the m a x i m u m value if and only if the
performance level of every component is equal to the performance level of
the entire system.
(HA) T h e degree of homogeneity of the system is the same if the perfor-
mance level of components is at equal distance from the performance level

of system in any sense (from 0 or from 1).
If we compare the conditions (i) — (Hi) in Definition 2.5 and the addi-
tional condition (2.5) with the previous requirements, then we see t h a t the
degree of homogeneity Hytx(A) of a system with components {x\,..., xn},
can be considered as the symmetrical normalized strict \P (cti,..., a n )-measu-
re of fuzziness of the fuzzy set A : X —>• [0, 1], A(XJ) — a; which indicates
the performance level of components, if the performance level of system is
t = V(ai,...,an) €(0,1).
In concordance with the situation \P ( a i , . . . , a„) G (0,1) and with the
requirements (HI) — (HA), the following statements are acceptable for the
situation t — $ (a,i,..., an) 6 {0,1} .
(hi) If t — ^ (oti, . . . , a n ) = 0, then the degree of homogeneity of the

system is equal to 1 if and only if a,- = 0, for all i £ { 1 , . . . , n}, equal to 0 if
and only if a; = 1, for all i £ { 1 , . . . , n} and decreases when the performance
level of components increases.
(h2) If t = * ( a i , ...,a„) = 1, then the degree of homogeneity of the
system is equal to 1 if and only if a; = 1, for all i G { 1 , . . . , n } , equal t o 0 if
Applications 83
and only if a,- = 0,for all i £ {1, ..., n} and decreases when the performance
level of components decreases.
Even if any strict and symmetrical i-measure of fuzziness can be used
to determine the value of the index Hytx{A), a suitable choice is given
by Example 2.3, taking the fuzzy complement in Example 2.2. If X =
{x\,..., xn}, then we normalize and simplify and this /-measure of fuzziness
(t € (0,1)) becomes
where g(a) = a if a £ [Q,t] and g(a) = j 4 r ( l - a) if a £ [t, 1].

In fact, these t-measures of fuzziness are introduced (for t = \) by
Kaufmann [114] (see also Klir [120]).
Taking into account (hi) and (/i2), a formula of calculus for the degree
of homogeneity of a system is given by
f sitK(A), if 0 < < < 1
H*,x(A) = \ ££^(1-0,-), if< = 0
l Êr = 1 «i> if< = i,
where t = ty (oi, ...,a„).

In a concrete case, we consider a system with three components
X = {xi, X2, X3}, having the performance function
f (a,b, c) = max(min(a,6) , min(a,c)) ,

where a, 6, c represent the level performance of components x\, X2 and X3,
respectively. The table of values of this function, corresponding to different
ordering of variables is (see Kaleva [112]):
< < min(a,b) min(a,c) $

a b c a a a
a c b a a a
b a c b a a
b c a b c c
c a b a c a
c b a b c b
We obtain the below table with the results of degrees of homogeneity

corresponding to different ordering of variables (n = 3, a,\ — a, a 2 = 6, 03 =
a = 0
q(3—a — b — c)
a G (0,1) a 6 c a
3(l-a)
a = 1 1 1
1-b-c
a = 0 3
a(3 — a — 6 — c)
ae(0,l) a c 6 3(l-a)
a= 1 1
a = 0 3-e
3
b — ab+a — a2+a — ac
a G (0,1) b a c 3(1-a)
b+2
a= 1 3
a = 0 3-6
3
a G (0,1) c a b c— ac-\-a — a2-\-a — ab
3(1-a)
a= 1 e+2
3
6= 0
6G(0,1) C 6 a
6= 1
c= 0
-c -\-b — bc-j-c—ac
c G (0,1) 6 c a
3(1^1
c= 1 b+2
3
2.3.3 Application to digital image processing

Digital image processing is a discipline with many applications in docu-
ment reading, automated assembly and inspection, radiology, hematology,
meteorology, geology, land-use management, etc. (see Rosenfeld [175]).
One of methods used in the study of digital images and, more general, of
discrete arrays in two or more dimensions of whose elements which have
values 0 or 1, is the digital topology (for details see Herman [lOl], Kong-
Rosenfeld [123], Kong-Roscoe-Rosenfeld [125] and Kong-Kopperman-Meyer
[124]). Although some parts of digital topology can be generalized to fuzzy
digital topology, which deals with gray-scale image arrays whose elements
lie in the range [0, 1] (for details, see Rosenfeld [l76j-[l79]), nevertheless the
Applications 85
study of binary image is more simple.

In what follows, by using ^-measures of fuzziness, we give an acceptable
m e t h o d to obtain a binary digital image from a fuzzy digital image.
A digital image is typically obtained quantifying the brightness values
of a image in a discrete and bounded grid of points, n a m e d background and
denoted in what follows by X. If the brightness values are considered in
[0,1], then we obtain a fuzzy digital image or a gray-scale image array. If
the set of values is {0,1} we obtain a binary digital image.
For every i £ l w e denote by A(x) G [0,1] the gray-level of the point
x. So we obtain the fuzzy set A : X —» [0,1].
Let t 6 (0,1). By tresholding a fuzzy digital picture with value t, t h a t
is by classifying the points x G X according to the fact t h a t their gray-level
exceed t or not exceed t, we obtain a binary digital picture (if A{x) > t
then the new value in x G X is 1 and if A(x) < t then the new value in
x G X is 0), but we lose a part of the information contained in the initial
picture. Our purpose is to choose the value (or values) t such t h a t the lost
information to be m i n i m u m . We denote by Pf(A) the lost information if
the level of tresholding is t £ (0, 1).
Intuitively, we can accept the following natural conditions on the lost
information Pt(A):
(PI) Pl(A) = 0 if and only if A(x) G {0,1} , Vx G X, t h a t is the initial

digital image is binary.
(P2) a) If the gray-levels of two points are smaller t h a n t then the
contribution to lost information of the point with greater gray-level is more
i m p o r t a n t t h a n t h a t of the point with smaller gray-level;
b) If the gray-levels of two points are greater t h a n t, then the contribution
to lost information of the point with smaller gray-level is more i m p o r t a n t
t h a n t h a t of the point with greater gray-level.
( F 3 ) Pt(A) is m a x i m u m if and only if, for any point x G X, the lost
information is m a x i m u m , t h a t is A(x) = t, Vcc G X.
We notice t h a t these conditions are verified for every strict i-measure

of fuzziness s£. So, if A G FS(X) is the fuzzy set which represents the
gray-level of the points of background X (for a fuzzy digital image) and
isc}te(o i) ' s a f a r r , ily of strict ^-measures of fuzziness, then
MS,(A)= infP*(A)= inf sllA)

cV ; v cV ;
t€(o,i) te(o,i)
is the minimal lost information by transition from a gray-scale image array

to a binary image array, with respect to the family of t-measures of fuzziness
{Se}se(0,l)-
Remarks. 1) Because A is a finite fuzzy set, we can assume that
0 < A(x0) < A(Xl) < ... < A(xk) < A(xk+i) <1
given up on the common values. In this case we obtain
Msi{A)= min {stc"{A);A{xp) <tp < A(xp+i)\ .
" p6{0,...,fe}
and the values of i for which is obtained this minimum represent an interval.
2) In practice, the members of family {stc]t€,f) t \ are of the same kind,
for any t G (0,1).
2.4 Bibliographical Remarks
Definitions 2.4 and 2.7, Example 2.5, Theorem 2.5 are in Ban [16], Defini-
tion 2.6, Theorem 2.4, Examples 2.2 and 2.3 are in Ban [21], Definitions
2.9, 2.10 are in Ban [18] and Theorems 2.6-2.9 are proved in Ban [18]. Def-
initions 2.12, 2.13 are in Ban [17], Theorems 2.10, 2.11 and Example 2.4
are in Ban-Fechete [20]. The application to systems performance is in Ban
[21] and to the calculus of degree of interference is in Ban-Fechete [20].
Completely new are Definition 2.5 and Theorem 2.3.
Chapter 3
Defect of Property in Topology
A well-known concept in topology is that of measure of noncompactness,

which in our terminology can be called defect of compactness. The main
aim of this chapter is to present this concept in classical setting, in random
setting and in fuzzy setting.
Also, other defects of topological properties are considered.
3.1 Measures of Noncompactness for Classical Sets

Let (X, p) be a metric space and let us denote
Vb {X) = {Y C X; Y ^ 0, Y is bounded} .
Mainly, two functionals that measure the degree in which subsets of X

fail to be compact are well-known.
Definition 3.1 (i) (Kuratowski [129], [130]). The Kuratowski's measure
of noncompactness for Y EVb (X) is given by
a (Y) = inf {e > 0; 3n £ N, At £X,i = T/n with

Y C \J"=1At and diam (At) < e} ,
where diam (A,) = sup {p (x, y) ;x,y £ Ai] , i = 1, n.

(ii) (see e.g. Istratescu [105]) The Hausdorff's measure of noncompact-
ness of Y C X is given by
h (Y) = inf i e > 0; 3n £ N, Vi £Y,i = \~^, with Y C [JB {yi}s)

I i=i
87
88 Defect of Property in Topology
Concerning the quantities a (Y) and h (Y) the following results are
known.
T h e o r e m 3.1 (i) (see e.g. Rus [182], p.85-87 or Banas-Goebel [32]) Let
X,Y eTb(X). We have:
0 < a(A) < diam (A) ;
A C B implies a (A) < a (B) ;
a (A) = a (A) and moreover a (A) = 0 if and only if A is compact;
a (V£ (A)) < a {A)+2e, where V£ (A) = {x £ X; p (x, A) < s} and p (x, A) -
ini{p(x,y) ; y £ A} ;
a(Al)B) = max {a (A), a (B)} ;
a(AnB) < min {a (A), a (B)} ;
Let Ai C X,A{+i C Ai be with Ai closed and nonempty, i = 1,2,....
7/lim n _ ) . 00 a (An) — 0 then pl^LjAi is nonempty and compact.
If, in addition, X is a Banach space, then a (A + B) < a (A) + a (B),
a (cA) = \c\a(A),c G R, at (convA) — a (A), where convA is the convex
hullofAeVb{X).
(ii) (see e.g. Beer [36], Banas-Goebel [32])
h (A) = 0 if and only if A is totally bounded (that is, Ve > 0, 3a; i,..., xp 6 A
such that \jx £ X, 3xi with p (x, xi) < e);
AcB implies h (A) < 2h (B);

h(A)=h(A);
h(AUB) < m&x{h{A) ,h(B)};
h is continuous on CL(X) = {Y C X; Y is closed, Y ^ 0} with respect to
Hausdorff topology, i.e. ifY„,Y 6 CL(X),n = 1,2, ..., satisfy
DH (Yn,Y) "—>°° 0 (where DH is the Hausdorff-Pompeiu distance), then
lim„_).00/i (Yn) — h(Y); Also, h is upper semicontinuous on CL(X) with
respect to the so-called Vietoris topology.
In addition,
h{A)<a{A) < 2h (A) ,VAeVb {X)
and if moreover X is Hilbert space, then
V2h (A) <a(A)< 2h {A) ,\/AeVb (X).
R e m a r k s . 1) If (X, p) is a complete metric space, then for A £ CL(X),

we have
a (A) = 0 if and only if A is compact
Measures of Noncompactness for Classical Sets 89
and
h (A) = 0 if and only if A is compact.
Because of these properties, we also can call a (A) and h (A) as defects of
compactness.
2) The Kuratowski's measure of noncompactness has many applications
to fixed point theory (see e.g. Darbo [60], Sadowski [188]).
The properties in Theorem 3.1 suggest an axiomatic approach in Banach

spaces (X, | | | | ) , as follows. Firstly we need the notations:
RC(X) = {Y C X; Y ^ 0, Y is relatively compact} ,

CO (X) = {Y C X; Y £ 0, Y is compact} .
Definition 3.2 (Banas-Goebel [32]). K, C RC(X) is called kernel (of a

measure of noncompactness) if it satisfies:
(i) A £ K, implies A £ K.;
(ii) AeJC,B cA,B j:® implies B £ AC;
(Hi) A,B eK implies \A + (1 - A) B e /C, VA G [0,1];
(iv) A E K, implies convA £ /C;

(v) The set fCc = {A £ K; A is compact} is closed in CO(X) with re-
spect to Hausdorff topology (i.e. the topology induced on CO(X) by the
Hausdorff-Pompeiu distance).
Definition 3.3 (Banas-Goebel [32]). The function n : Vb{X) -> [0, +oo)
is called measure of noncompactness (or defect of compactness, in our ter-
minology) with the kernel K. (denoting ker fi = K.) if satisfies:
(i) n(A) = Q if and only if A £ K;
(ii) n(A) =fi(A);
(Hi) n (convA) = /j, (A);
(iv) Ac B implies p (A) < \i (B);
(v) fi (\A + (1 - A) B) < \fx (A) + (1 - \)/i{B),V\ £ [0,1];
(vi) If An eVb(X),An = An and An+1 C An,n= 1,2,... and if
lim„ôoA« iAn) - 0, then f f ^ l i ^ " # ®-
If K, = RC(X) then /i will be called full (or complete) measure.
The following examples are in Banas-Goebel [32].
Example 3.1 a and h are full measures of noncompactness.

E x a m p l e 3.2 fi {A) = diam (A) , A £ Vb{X), with kernel

K, = {{x} ; x £ X} .
E x a m p l e 3.3 Let F C X be closed and K = RC(X). Then /i(A) =

h(A) + dH(A,F),Ae Vb(X), where
dH (A, F) = sup {inf {\\a - y\\ ; a £ A} ; y £ F} .
E x a m p l e 3.4 Let us suppose that (X, ||-||) is a Banach space t h a t have

a Schauder basis { e , } i e N , t h a t is each x £ X has an unique representation
x = YmLafi {x)eit where tpi : X —>• R , i £ N . Denoting 7?n : X —^ X
by _R„ (x) = S t n + i V i (x) eit the function / / ( A ) = lim^ôo sup |||-Rn||| is
a regular measure of noncompactness on Vb(X) (that is full, and satisfies
H (A + B) < n (A) + p (B) , [J, (AL) B) < ma,x{fi(A) ,/«(J9)}), where \\\Rn\\\
denotes the norm of linear continuous operator Rn.
E x a m p l e 3.5 Let (K, p) be a compact metric space and X — C (K; R ) =

{/ : K —)• R ; / continuous on A'}, endowed with the uniform norm
ll/H = m a x { | / (x)\; x £ A'}. For any / £ X we can define the modulus of
continuity of / by
u(f;e)=Bup{\f(t)-f(s)\;t,8eK,p(t,8) < e} ,Ve > 0

and for any A C X, let us define
u (A; e) = sup {w (/; e); / £ A) , Ve > 0.
Then
u(A) = limw (-4;e)
is a measure of noncompactness on Vb(X), which moreover satisfies
n(A) = 2h(A),
where h (A) is the Hausdorff measure of noncompactness in Definition 3.1.
E x a m p l e 3.6 Let X = Lp[a,b],l < p < + o o , be endowed with the
norm | | / | | p = (j* \f {t)\p dtj " and let /3 : [0, +oo) -)• [0, +oo) be with the
property l i m ^ o / ? (2) = (3 (0) = 0. For / £ X, let us define the Kolmogorov
modulus of continuity of / with respect to /? by
upifc) =aup{\\fh- f\\p- 0(\h\);\h\<e} ,
(where fh (t) = f (t + h) ,Mt G [a, b]) and for A G Vb{X), let us define
up{A;e) = sup {o^ (/; e) ; / G A],

Up {A) = Kmuf3{A;e).
£-1-0
Then up : Vb(X) —> [0, + o o ) is a regular measure of noncompactness.
R e m a r k . For m a n y other details concerning the axiomatic approach of

measures of noncompactness, see the interesting book Banas-Goebel [32].
T h e end of this section is based on an idea used in K u p k a [127] which

is the following: W h e n we want to model "metric" or "uniform" notions
in pure topological case, it suffices sometimes to replace "e > 0" by "open
cover of X " . Then, we can obtain analogous results to the metric case. In
this sense, we will introduce a topological analogue of measure of noncom-
pactness and we will show some basic properties of this kind of measure.
Finally, as applications we prove the upper semicontinuity of the limit of a
decreasing net of upper semicontinuous multifunctions.
Let IT be the set of all open coverings of a topological space X. Let us
consider the family C = 2*" of all subsets of TT ordered by the relation
x < y if and only if x D y, for x,y G C.
T h e n (C, < ) is a complete lattice with the m i n i m a l element 0 = TT. NOW

we can define the C-measure of noncompactness as follows:
Definition 3.4 (Kupka-Toma [128]). If A C X then the measure of

noncompactness of A is the element of C defined by
m(A) = {V£n 3 finite 8 C V : A c | J < 5 } .
R e m a r k . Lechicki in his article [132] has introduced a measure of non-

compactness for uniform spaces, whose values are in a smaller system of
sets t h a n C. B u t in general topological spaces we have no notion of "uni-
formness" as in the case of uniform spaces. T h e above notion of measure
of noncompactness applied in the class of uniform spaces is stronger t h a n
Lechicki's one.
T h e o r e m 3.2 (Kupka-Toma [128]). The mapping m : 2X ->• C has the

following properties:
92 Deject of Property in Topology
(i) If A is a compact subset of X then m (A) = 0 . If A is closed then

the converse is true;
(«') IfAcB then m(A) <m (B) ;
(Hi) m (A U B) = m (A) l~l m (B) = sup {m (A) ,m(B)};
(iv) m(AC\B) <m(A)Um (B) = inf {m (A), m (B)} .
The next two examples prove that if A is not closed, then the equality
m (A) = 0 does not imply the compactness of A and that the inequality in
(iv) can be strict.
Example 3.7 Let X be a compact topological space, which contains a
noncompact subset A. Then m (X) = 0 and consequently m (A) = 0 too,
but A is not compact.
Example 3.8 Let X = R with the usual topology, A = (—oo,0],B =
[0,oo). A n B = {0} is compact and therefore m(Af\B) — Q but
inf {m (A), m (B)} ^ 0 because the open covering
P = {(n-l,B+l);n£Z}
of X does not contain a finite subcovering neither of the set A nor of the
set B.
In the class of complete uniform spaces, we can partially compare the

measure of noncompactness m, with Lechicki's measure which is denoted
by $ (see Lechicki [132]).
Theorem 3.3 (Kupka-Toma [128]). Let X be a complete uniform space.
Then we have the following:
(i) For every closed subset A C X, m (A) = 0 if and only if$(A) = 0;
(n) For every decreasing net (A1 : 7 6 T) of closed nonempty subsets of
X,
inf {$ (Ay) : 7 € T} = 0 =>• limm (Ay) = 0 .
The next example shows that if A is not closed, then the equivalence (i)
of Theorem 3.3 may not be true, because in general we can have m (A) ^
m (A), where A is the closure of A. This also shows that topological mea-
sure of noncompactness can distinguish such subtle differences which are in-
distinguishable by the classical (metric) measure of noncompactness. More-
over, one can easily see that m (A) = 0 is in topological case stronger than
m (A) = 0 in the metric case.
Example 3.9 Consider R with the a-topology (open sets are the sets of
the form 0\N where O is an usual open set and N is a nowhere-dense set
(see Njastad [160]). Let us take the following a-open set
A=(0,1)\ j i ; n € N J .
The a-closure of A is A — [0, 1]. The open covering V of X defined by
*=wu{(^^^^)i"eN}uM;!"£N}}
contains a finite subcovering of the set A, namely {A}, and therefore V G
m(A). But, V £ m(A) because the points ^ € A cannot be covered by
finite number of sets from V.
Let (]?, <) be a upward directed set. If (x 7 : 7 G Y) is a net in C, then

we shall write l i m ^ r ^ = © if and only if
v {v} e c, 37o e r, v 7 > r 70: ^ < {P} •

The next theorem is a topological analogue of the Theorem 2.2 in

Lechicki [132]. It works with a complete uniform space which in general, of
course, is not our case. Because of (n) in Theorem 3.3, we can state that
the theorem generalizes Theorem 2.2 of Lechicki, if the space X is supposed
to be uniformly complete.
Theorem 3.4 (Kupka-Toma [128]). Let (A1 : 7 G T) be a decreasing net

of closed nonempty subsets of a topological space X. Then the following
implication is true:
limm (A-f) — Q => A = ( | A1 is nonempty and compact.
The next theorem generalizes a theorem which is well-known in the

metric case. The notion of upper semicontinuity being a topological one
(see e.g. Michael [147]), it is natural to search an analogous result in
topological setting.
Theorem 3.5 (Kupka-Toma [128]). Let X,Y be topological spaces and

(F-f : X —>• Y) _ r be a decreasing net of u.s.c. multifunctions with closed
nonempty values. If for each x G X, limber m (F-y (x)) = Q, then the multi-
function
F :X ->Y,F{x) = f]F1 (x)
is upper semicontinuous with nonempty compact values.
Remark. The net (F 7 : 7 £ T) is decreasing and it is easy to see that this

net converges to F in the sense of the Vietoris topology (see e.g. Michael
[147]).
The condition in Theorem 3.5 that the measure of noncompactness of
Fn (rr) tends to zero, cannot be omitted as the next example shows.
Example 3.10 Let X = Y = [0,1) U (l,+oo) with the usual topology.
For each n £ N and x £ X we define the multifunctions
Fn(x) 0,l + z + - \ { 1 }
n
F{x) [0,1 + ar] \ {1} i f x # 0 , F ( 0 ) = [0,l)
Then Mx G X, Fn (x) D Fn+1 (x) and F (x) = f]n€NFn (x). The multifunc-
tions Fn are upper semicontinuous but F is not upper semicontinuous at

the point 0. To see this, let us remark that F (0) C [0,1) = U, U is open in
X and for each open neighborhood V of 0 we have F (V) <£ U.
The next example shows that in Theorem 3.5, the hypothesis that the
values of multifunctions are closed, cannot be omitted.
Example 3.11 Let us consider the complet metric spaces X — [0, +00),
Y C R with the usual metric. For each x G X, we define the multifunctions:
Fn(x) U{1}:
nx + 1
1
F(x) -00, U{1} ifx£Q,F{0) = {1}.
x
The
Vz e X, n G N -. Fn (1) D Fn+1 (x) and F (x) = f] Fn

n€N
The multifunctions Fn are all u.s.c. but their limit F is not, because it is
not u.s.c. at the point 0.
Random Measures oj Noncompactness 95
Finally, we give an example which shows why it is necessary to consider

the family n of all open coverings of X and not only a subfamily of cov-
erings of certain type, as for example ball coverings in metric space. We
will construct two nice families of coverings which generate very different
measures of noncompactness.
Example 3.12 Let X = (—§•, f) with the usual topology. We can con-
sider two families of open coverings of X:
7T! = {V€;e> 0},V£ = {{a-e,a + e)nX;a€ X} ,

""2 = {Q £ ;e > 0} ,Q E = { ( a r c t g ( a - e ) ,arctg(a + e)) ;a e R } .
For e tending to 0, the coverings Ve (resp. Q€) are getting finer in obvious
way. The space X with the family 7Ti evokes (—§,§) with usual metric, X
with 7T2 evokes R with the usual metric. The second space is complete, the
first one is not. Let us take the decreasing sequence of closed subsets of X
/ 7T 7T 1 \
An=(^--,-+-J,nGN.
Let C\ = 27ri,C2 = 2*2• If we consider mi (resp. m2) -the measures of
noncompactness with values in C\ (resp. C-z), then
limmi (An) = K\ (minimum of C\),

nEN
limm2(A„) = 0 (maximum of Co).
nraCTJ
£N
3.2 Random Measures of Noncompactness
In this section we present extensions of the Kuratowski's and of the Haus-

dorff 's measures of noncompactness to probabilistic metric spaces in Menger
sense.
Let us denote by A the set of all nondecreasing and left continuous
functions / : R —>• [0,1], such that / (0) = 0 and lirm^+oo/ (x) = 1.
Definition 3.5 (Menger [145]). A probabilistic metric space (PM-space,

shortly) is an ordered pair (S,F), where S is an arbitrary set and F :
5 x 5 - > A satisfies:
(i) Fp>q (x) = 1, Va; > 0 if and only if p = q;
(it) FPiq (x) = FgtP (x) ,Vp,q £ S;
96 Deject oj Property in Topology
(Hi) If FPtg (x) = 1 and F g>r (y) = 0, \fp, q,r G S then Fp>q (x + y) — 1.
(HereF P i ,( a ! ) = F(p,g)( a r).)
Definition 3.6 (Egbert [69]). Let ^ C S,A # 0. The function £>A (•)
defined by
DA (X) ~ sup inf Fp q (x), x G R
is called the probabilistic diameter of A. If sup {DA (X) ; x 6 R-} = 1 then

A is called bounded.
Definition 3.7 (Boc§an-Constantin [4l], see also Istratescu [105],
Constantin-Istratescu [58]). Let A be a bounded subset of S. The mapping
aA (x) = sup < £ > 0 ; 3 u £ N , At, i = 1, n with A = 1 ) A and DAt (x) > e\
where x 6 R, is called the random Kuratowski's measure of noncompact-

ness.
Remark. In our terminology, a A can be called random defect of compact-

ness of A (of Kuratowski type).
Theorem 3.6 (Bocsan-Constantin Ul])- We have:

(i) aA e A;
(n) aA (x) > DA (X) , Mx £ R;
(Hi) IftyÂcB then aA (x) > aB (x) , Vx G R ;
(iv) aAuB (x) = min {aA (x), aB (x)} , \tx G R ;
(v) aA (x) = a-^(x), where A is the closure of A in the (e, A) -topology of
S (where by (e, A)-neighborhood of p G 5, we understand the set Vp (e, A) =
{q G 5; FPtq > 1 - A} , e > 0, A G [0,1].
The function a^ (x) can be calculated by
Theorem 3.7 (Bocsan-Constantin fyl])- Let KA be the set of functions
/ G A such that there exists a finite cover of A, A = [J-£jAj, J-finite, with
DAj (x) >f(x), Vj e J,x£R. Then
a A (x) = inf {/ (x);fe KA} ,VX€R.
It is known that in usual metric spaces, the Kuratowski's measure is

used to characterize the compactness. In this sense, for the probabilistic
case, we need the following
Random Measures oj Noncompactness 97
Definition 3.8 (Boc§an-Constantin [41]). We say t h a t t h e P M - s p a c e

(S, F) is probabilistic precompact (or probabilistic totally bounded) if for
every e > 0, A £ (0,1), there exists a finite cover of S, S = [Ji€jAi, /-finite,
such t h a t DA, (e) > 1 — A.
Now, let us suppose t h a t (S, p) is an usual metric space. It is well-known

t h a t it generates the probabilistic metric space (S,F), with Fpq(x) —
H (x — p (p, q)), where H is the function given by H (x) = 0 if a; < 0, H (x) =
1 if x > 0. We have
T h e o r e m 3.8 (Bocsan-Constantin [\l]). Let (S,p) be an usual metric

space and (S,F) the corresponding PM-space generated by (S,p).
(i) A C (S, F) is probabilistic precompact if and only if a A (X) — H (x),
V*G R ;
(it) A C (S, p) is precompact if and only if A is probabilistic precompact
set of {S, F);
(Hi) For any bounded A C (S, p) we have
aA (x) = H (x - a (A)), Wx G R ,
where a (A) is the usual Kuratowski's measure of noncompactness in (S,p)

and a A (x) is the random Kuratowski's measure of noncompactness in the

generated PM-space (S, F) .
R e m a r k . For other details see e.g. Constantin-Istratescu [58].
In what follows we consider the r a n d o m Hausdorff's measure of non-

compactness. In this sense, firstly we need the followings.
Definition 3.9 (Menger [145]). (S, F) is called PM -space of Menger-

type with the <-norm T, if F satisfies the first two properties in Definition
3.5 and the third one is replaced by
Fp>q (x + y)>T (Fp<r (x) , Fr<q (x)) , Vp, q, r £ S.
We denote it as the triplet (S, F, T).
Definition 3.10 (see e.g. Istratescu [105]). If (S,F,T) is a P M - s p a c e

of Menger-type and A, B C S, then the probabilistic (random) Hausdorff-
Pompeiu distance between A and B is given by
DAB (X) = s u p T ( inf s u p i ^ ? (t), inf supFp>? (t) ) , x E R.

t<x \pe^ ? es ' q£BP£A ' J
Definition 3.11 (see e.g. Istratescu [105]). Let (S,F,T) be a P M - s p a c e

of Menger-type and A C S, bounded. T h e r a n d o m Hausdorff's measure of
noncompactness of A is given by
hA (x) = sup {e > 0; 3 finite F£ C A such t h a t DAFe (x) > e) .
R e m a r k . T h e function hA {X) has similar properties with a A (X) . Other

details concerning r a n d o m Hausdorff's measures of noncompactness can be
found in e.g. Constantin-Istratescu [58].
3.3 M e a s u r e s of N o n c o m p a c t n e s s for F u z z y S u b s e t s i n
Metric Space
In this section we will extend in various ways the concepts and results in
Section 3.2 to fuzzy subsets of classical metric spaces.
Let (X, d) be a metric space and the box metric d* : X x R —¥ R + de-
fined by d* (P, Q) = max{d(x,y) , \r — s\], for all P — (x,r) ,Q = (y, s) £
X xR.
For the concept of fuzzy subset of X introduced by Section 2.1, we

obviously can give an equivalent definition, as a couple (A,ifA), where
<PA '• X —> [0,1] is the membership function and A = {x £ X;tpA (x) > 0}
is the so-called support of (A, ipA) •
Let us recall t h a t in literature b o t h forms (notations) for the concept of
fuzzy set are used. Because for the purpose of this section seems to be more
convenient and t o accustom the reader with it too, everywhere in Section
3.3 this second form will be used.
Firstly, let us present some concepts.
Definition 3.12 T h e sets G0 {<PA) = {{x, y); 0 < y = ipA (x), x E X}

= Graph (F) and HG0 (<PA) = {{*, y) \ 0 < V < <PA ( * ) , x e X}
= hypo(F) f l ( A x (0,1]) are called the support graph and the support hy-
pograph of (A, ^>A) respectively, where F : A ->• ( 0 , 1 ] , F (x) = ipA (x), \/x £
A.
T h e diameter and the hypo-diameter of the fuzzy set (A,ipA) are given by
D {<pA) = sup{d* {a, b);a,beG0 (ipA)}
Measures of Noncompactness for Fuzzy Subsets in Metric Space 99
and
hD (ipA) = sup {d* (a, 6); a, b £ HG0 (<pA)} ,
respectively. If D (f>A) < + 0 0 {hD(<pA) < + 0 0 ) we say t h a t (A,ipA) is

bounded (hypo-bounded).
Definition 3.13 Let (A, <pA) be bounded (or hypo-bounded, respectively).

T h e Kuratowski's measure a n d t h e Kuratowski's hypo-measure of noncom-
pactness of (A, <PA) are given by
K (<pA) = M {e > Q;3n 6N,(Ai,(pAt) with D(<PA,) <e,i = T^n,

x
such t h a t ipA (x) pAi ( );« = 1, « } , Va; £ X]
and by
hK ((pA) = inf {e > Q;3n £ N,(Ai,<pAi) with hD(<pAi )<e,i= L/n,

X x
such t h a t <PA ( ) < s u p {<pAi ( ) ;i = l,n} ,Vx £ X} ,
respectively.
T h e Hausdorff's measure a n d t h e Hausdorff's hypo-measure of noncom-

pactness of (A, <pA) are given by
H{<pA) = inf {e>0;3ne'N,3Pi = (xi,ri) £ Go ((pA),i = '^n,

such t h a t \/P = (x, r) £ G0 (<pA) , 3Pk with d* (P, Pk) < e]
and by
hH(<pA) = mf{e>0;3n£N,3Pi = (xi,ri)£HGo{<pA),i = T~h~,

such t h a t V P = (x, r) £ HG0 {<pA), 3Pk ,d* (P, Pk) < e}
respectively.
Concerning t h e above concepts can be proved the following relations.
T h e o r e m 3 . 9 (?) K {<pA) < hK (ipA).

(ii) K (y>A) < K* (Go (<pA)), H (<pA) = H* (G0 (<pA)), hH {<pA)
= H* (HGQ (<pA)), where K* and H* represent the usual Kuratowski's and
Hausdorff's measures of noncompactness (respectively) of the subsets in the
metric space (X x [0,1] ,d*). (Here (A,tpA) is considered hypo-bounded).
Also, we have:
T h e o r e m 3.10 Let {A,ipA) C {B,ipB) be two bounded fuzzy subsets of
(X,d). We have:
(i) K{<pA) <D(<pA).
{ii) (A, (pA) C {B,<pB) implies K (<pA) < K(<pB), where {A,<pA) C
(B, ips) means <pA (x) < ipB {x), Va; G X.
{Hi) K (ipA V <ps) = max{A' (ipA), K (<PB)}, where (ipA V ipB) (x)
= m&x{(ipA) (x) , (<PB) {X)} ,VZ G X.
Proof, (i) and (ii) immediately follow from Definition 3.13.
{Hi) Since {A, (pA) C {C,<pc) and (B,tpB) C {C,<pc), where <pc {x) =
tpA {x) V ipB {x) ,\/x G X, by (ii) we immediately get
max {A' (ipA), K (<pB)} < K {(pA V ipB) •

Conversely, for any fixed S > 0, by the definitions of K {<pA) and K {fs) as
infimums, there exist £i,£2 > 0, n, m G N and {Ai, ipAi) ,i — l,n, (Bj,tpBj)
,j = Tjn with D {ipAt) <sui = T~n, D (<pBj) <e2,j = T~m,
fA {X) < sup {(pA, (x); i = T~n~) , <pB (x) < sup {<pB. {x); j = T~^} , x G X,
such that
£i < A {<pA) + S,e2< K {ipB) + 6.

This implies
max{£i,e 2 } < max {A' {<pA), K {<pB)} +6.

On the other hand
(<PA V <pB) {x) < sup {<pCk {x); k = l,n+m} ,
where Q = Ai, i = T~n, Cn+j = Bj, j = l , m and
D{<Pck) < m a x { £ i , e 2 } , for all k = l,n + m.

We get
A'(V?A V *pB) < max{£i,£ 2 } < ma,x{K (<pA) ,K(<pB)} + S.
Passing with S ->• 0+ we obtain
A {ipA V <pB) < max {A [ipA) , A (y>B)} ,

which proves (Hi) too. •

Theorem 3.11 (i) hH (IPA) — +00 if and only if HGo (<PA) is un-
bounded.
(ii) If (A, ifA) C (B, <pB) then hH (<pA) < 2hH (tpB).
(Hi) hH (<PA) = 0 if and only if HGo (<PA) 2S totally bounded.
(iv) hH (ipA V tpB) < max {hH (ifA) ,hH (<PB)}, where (<pA V <pB) (x)
= m.&x{<pA (x) ,<PB (x)} ,V;c G X.
Proof, (i) and (Hi) follow immediately by Lemma 1, (o), (b) in Beer [36].
Now, let (A,<PA) C (B,<PB), i.e. <pA (x) < fB (x) , V i £ l . It is easily
seen that HGo ('PA) C HGO (<PB), which combined with Lemma 1, (c) in
Beer [36] proves (ii) .
(iv) Firstly we have
HGo {<pA V <PB) = HGo (fA) U HGo (<pB) •

Indeed, let (x,r) G HG0(<PAV <PB), i.e., 0 < r < max{y>A (x), <PB (X)}.
This implies 0 < r < <pA (x) or 0 < r < ips(x), wherefrom (x,r) 6
HG0(<PA)UHG0(<PB).
Conversely, let (x,r) G HGO(<PA) U HGQ(IPB), which implies 0 < r <
ifiA (x) or 0 < r < ipB (x) and 0 < r < (pA (x) V <pB (x), i.e., (x,r) G
HGO(<PAV <PB)- By Lemma 1, (/) in Beer [36] and by Theorem 3.9, we
get
hH(<pAV<pB) = h*(HGo(vAV<pB))
= h*(HGo(<pA)UHGo(<pB))
< m<ix{h*(HG0(vA)),h*(HGo(<PB))}
= max{hH ((fA) ,hH (<pA)} ,
which proves (iv) too. •

Theorem 3.12 Let (A,<pA) be a bounded fuzzy subset of (X,d). Then
the inequality K (fA) > KQ (A) holds, where A'o represents the usual Ku-
ratowski's measure of noncompactness of subsets in (X,d).
Proof. By Definition 3.12 we can write
K (<pA) = inf {s > 0; 3n G N, {Ai,<pAi) with D* (G0 (<pAi)) < e,
i = 1, n, such that 0 < (fiA (x) 0 ; 3 n £ N,(AitipAi) with D* (G0{<pAi)) <£,

i = M , such that Go {<pA) C U" =1 tf G 0 (vu,)} • (3.1)
Indeed, let e > 0 be such that 3n £ N, (Ai,<pAi) with D* (G0 (<pAi)) < £ and
0 < <pA (x) < sup {ipA, (x) ;i = \,n) ,Mx £ .A. Then if (x,y) £ G0 (<fiA),
i.e., 0 < y = <pA(x), we get that there exists j £ {l,...,n} such that
0 < y = <pA {x) < <pAj{x), i.e., (x,y) £ HG0(tpAj) C U?=1HG0{<pAi),
which immediately proves that K (fA) > M.
Conversely, let e > 0 be such that 3n £ N, (Ai,tpAi) with £>* (GO (V?A,)) <
£, i = 1, n and Go ( ^ A ) C U"=1.ffGo {<pAi)- Suppose that x £ A. We have
0 < ipA (x) — y and by (x, y) £ Go (<pA) there exists j £ {1,..., n} such that
(x,y) £ # G 0 (^4,), i.e. 0 < <pA(x) < <pAj (x) K (<pA) and as a conclusion we get (3.1).
Further we will prove that
M > inf {e > 0; 3n £ N, 3A,- C X with Do ( A ) <£,i = T~n,

such that A C U^=1A,-} = A'o (.4) , (3.2)
where Do represents the usual diameter of subsets in (X, d). Indeed, let
£ > 0 be such that 3n £ N,(Ai,(pAi) with D* {Go(<fA,)) < e,i = T/n
and Go {(fA) C U" =1 ffGo {<pAi) • By £>* (Go {<pA,)) < £ we immediately get
d* [(xi,yi) ,(£2,2/2)] < £ for all (xk,yk) £ GO(<£>A,), Ar = 1,2. This implies
^(*i,*2) < e, 12/1 - 2/21 < £ , for all xi,x2 eA,0<yk= <pAi (xk) ,k = T^2.
Therefore, D* {Go{<pAi)) < s implies Do(Ai) < e. Then, let (x,y) £

Go (ipA), i.e., 0 < y = <pA (x). There exists j £ {1,..., n] such that (x, y) £
HGQ (<fAj), i.e., 0 < y < ipAj (x), which implies x £ Aj. As a conclusion,
A C U"=1 A', which immediately implies (3.2).
Now by (3.1) and (3.2) the theorem is proved. •
R e m a r k s . 1) Because in general the converse inequality in (3.2) does not

hold, this means that in general we have K (<pA) > A'o {A).
2) By Gal [84], Theorem 3.5, (ii),hH (<pA) = H* {HG0 {<pA)) = 0, if and
only if HGQ (<pA) is totally bounded. Now we want to prove that if Go (<pA)
cannot be closed in the metric space (X x [0,1], d*) .
Indeed,let x £ A, i.e. <PA{X) > 0. Obviously t h a t (x,0) £ HGO(<PA) •

On t h e other hand, (x, 0) £ HG0 (<PA), since for r „ = n * , n £ N , where
0 < ^ < ¥>A(a;),Vn > n0, we have (x,rn) £ HG0(<PA) ,Vra G N a n d
d* [(a:,r„) ,(x,0)] = m a x {d (x, x) ,rn) A 0.
As a conclusion, # G o (<£A) C HGQ {'PA), strictly. B u t because there exist
totally bounded sets which are not closed (even in a compact metric space),
this means t h a t in general can exist fuzzy sets [A, if A ) with hH (<PA) — 0-
By using t h e level sets m e t h o d , we can introduce
D e f i n i t i o n 3 . 1 4 T h e (a)-Kuratowski measure of noncompactness of

{A, ipA) is given by
aK {9A) = sup {A'o (Ax); A € (0,1]} ,
where A\ = {x £ X; <PA (X) > A} and A'o is the usual Kuratowski's measure
of noncompactness of usual subsets in (X, d).
T h e (a)-Hausdorff measure of noncompactness of (A, IPA) is given by
aH (<pA) = sup {H0 (Ax) ;\£ (0,1}},

where HQ represents t h e usual Hausdorff's measure of noncompactness of

usual subsets in (X, d). Obviously, aH (<PA) c a n take the + o o value.
We also need t h e following
D e f i n i t i o n 3 . 1 5 (see e.g. Weiss [220]) If (X, d) is a metric space, then

the induced fuzzy topology on (X, d) is the collection of all fuzzy subsets
of (X, d) with if A '• X —> [0,1] lower semicontinuous on X.
A fuzzy set (A, IPA) is called (a)-bounded if for each A £ ( 0 , 1 ] , Ax is
bounded in t h e metric space (X, d).
A fuzzy set (A, PA) is called (a)-compact if for each A 6 ( 0 , 1 ] , Ax is com-
pact in t h e metric space (X, d) .
Now we are in position to prove t h e
T h e o r e m 3.13 Let (A,<PA) , (B,<PB) be (a)-bounded.

(i) aK (IPA) = 0 if and only if each Ax, A £ (0,1] is compact in (X, d),
where Ax denotes the closure of Ax in (X,d). Also, if (A,ipA) is (a)-
compact then aK (PA) — 0.
(ii) If (A,ip A ) C (B,ipB) (i.e. <pA(x) < fB(x),^x £X) then
aK (IPA) < aK (ips) •
(Hi) aK (<PA) < Ko(A), where A'o represents the usual Kuratowski's
measure for subsets in (X,d), and in general we have not equality.
(iv) aK (ipA V <ps) = max{o;K (<pA) , aK (<PB)}, where (<pA V <ps) (x) =
max {if A (x) ,<PB (X)} , V i € l .
(v) aH (<PA) — 0 if and only if each A\, A G (0,1], is totally bounded.
(vi) If (A, <pA) C (B, <pB) then aH (<pA) <2aH (<pB).
(vii) aH (ifA V <PB) < max {aH (ipA), aH (<PB)} •
Proof, (i) We have aK (<pA) = 0 if and only if K0 (Ax) = 0, VA G (0,1],

which immediately implies the desired conclusion.
(ii) Since (A, <pA) C (B, (pB) implies A\ = {x G X; ipA (x) > A} C B\ =
{x G X\ ifB (x) > A} , VA G (0,1], the proof is immediate.
(Hi) Firstly it is easy to check that
A = {xeX;pA(x)>0}= ( J AX= (J {xeX;<pA(x)>\}.

A€(0,1] A£(0,1]
Since A\ C A, we have K0(A\) < K0(A),\/X G (0,1], and passing to

supremum we get aK (<pA) < A'o (A).
Now if for example X = R, A = [0, +oo), A = U~ =1 [0, n], then
sup {A'o ([0, n]); n G N } = 0 but K0 (A) ^ 0, which means that in general
«A:(VA) <KQ(A).
(iv) By (ii) we easily get
max {aA' (<pA), aK (<PB)} < aK (<PA V <pB) •
Conversely, for all A G (0,1] we have
{x G X; maxjÂ (x) , <pB (x)} > A}

= {x€X;<pA(x)>\}\j{xEX;<pB(x)>\}.
Applying the usual Kuratowski's measure A'o, we get
K0({x G X;ma,x{<pA(x) ,<pB(x)} > A}) = max{A' 0 (>U) ,K0(BX)}
< max{atf (<pA), aK (ipB)} , VA G (0,1],
wherefrom passing to supremum with A G (0,1], we obtain aK (fA V fs) <

m&x{aK (<pA) ,aK (<pB)} .
(v), (vi) and (vii) are immediate consequences of Definition 3.14 and of
Beer [36], Lemma 1, (a), (c), (/). •
Corollary 3.1 For all (A,Â) hypo-bounded, we have
aK (PA) < K (<pA) < K* (Go (<pA)) < K* (HG0 (fA)) < hK (<pA).
l e
Proof. Firstly we will prove t h a t hK (<pA) > K* (Go (<PA)), - - that
hK(<pA) = M{e>Q-,3neN,{Ai,<pAi)mthir{HGo{<pAi))<e,i =
1, n, and 0 < ipA (x) inf {e > 0 ; 3 M j C l x(0,1] with D* (Mi) < e, i - Y~h,
n \
such t h a t HGo (tpA) C M M , - > .
<=i J
So, let e > 0 be for which there exist n G N , (Ai,(pAi) with D* (HGo (fA{))
< e, i = 1, n, such t h a t 0 < ipA (x) A (X) A'* (Go ( p ^ ) ) , which together with
the obvious inequality A'* (HGo (<PA)) > K* (Go (f>A)) and with t h e Theo-
rems 3.9, 3.12 a n d 3.13, prove the corollary. •
R e m a r k s . 1) In the axiomatic definition of the measure of noncompactness

(denoted by e.g., fi) for usual subsets of a Banach space, among others must
be checked t h a t fi (A) — \x (A) and fi (A) = /i (conv (A)), where A is the clo-
sure of A a n d conv (A) is t h e convex hull of A (see e.g., Banas [31]). From
this viewpoint, similar questions can be considered for t h e above measures
of noncompactness aK (VA) , K (<pA), hK (<PA) > aH (<PA) , H (<PA) , hH ( P A )
T h e difficulty consists in proper definitions for the closure a n d for t h e con-
vex hull of a fuzzy set.
For example, for (A, <pA) we can consider the closure (.4, IPA) in the induced
topology in Definition 3.15, i.e., (A,<PA) — (B,^PA)I where
PA ix) = inf
{f ix); fA P on X, p-fuzzy convex on X} , V i £ X .
2) By using the above measures of noncompactness, it would be inter-

esting to extend the known Darbo's [60] and Sadovski's [187] results of fixed
point to fuzzy-set-valued mappings.
3) By Corollary 3.1 it follows t h a t if (A,IPA) has the support A rela-
tively compact in (X, d) then aK (<PA) = 0 and t h a t if Go {<PA) ls relatively
compact in (X x [0,1], d*) then aK {<PA) = K (<PA) — 0.
4) By T h e o r e m 3.9, ( n ) , the measures of noncompactness of Hausdorff 's
type for fuzzy sets are completely characterized by means of the usual Haus-
dorff's measure of noncompactness for some usual sets. Then, from the
inequalities in Corollary 3.1 arise the question:
If we denote any measure of noncompactness of Kuratowski-type of a fuzzy
set (A, <PA) by K (<PA), then find an usual set (in X or in X x [0,1], respec-
tively) such t h a t K (IPA) = K* (M), where K* (M) represents the usual
Kuratowski's measure of noncompactness of M.
While the above question seems to be difficult, we will solve it by in-
troducing other measures of noncompactness of Kuratowski-type for fuzzy
sets.
These new measures are suggested by the following remark.
Let (X, d) be a metric space and A C X. As it is well-known, the
Kuratowski's measure of noncompactness of A denoted by A'o (A) is given

by
A'o (A) = inf {e > 0; 3n £ N , At G X with D0 (At) < s,
i — l,n such t h a t Ad I ) A, > .

»=i J
But if we denote B{ = At D A, then obviously get D0 (Bi) < Do (Ai) < e

and A = \J"=1B~i. Therefore we immediately obtain
A'o (.4) = inf {e > 0; 3n e N , At e X with D0 ( A ) < e,
i — 1, n such t h a t A — \\Ai \ .
»=i J
(Here Do {Ai) is the usual diameter of Ai in (X, d)).
As a consequence we can introduce
Definition 3.16 Let (A,(PA) be a bounded (or hypo-bounded, respec-
tively) fuzzy subset of (X, d). Then we define
A'i {<PA) = inf {e > 0; 3n G N, (Ai,<pA,) with D (<pAi) <e,i = T~n,

such that ip A (x) = sup {fAi (x) ; i = 1, n} , Var G X} ,
and
hKi{<pA) = inf {e > 0;3n G N, (Ai,<pAi) with hD{<pAi) < s,i = l,n,
such that ^ (x) = sup {^^ (I);J'= l,n} , V i G ^ } ,
respectively.
T h e o r e m 3.14 (i) hK (IPA) < hK\ {<pA) ,V(A,<pA) hypo-bounded.

(«) A (y>A) < A'i (<pA) = K* (Go (p A )) , V(A, ^ ) founded.
Proof, (i) It is immediate by definitions.
(if) The first inequality is immediate by definitions. On the other hand,
the converse inequality A'i {if A ) < K(ipA) seems that does not hold in
general, because if we follow the idea in the usual case by defining
ipA. (x) = min{(pA (x), <pAi (x)} , x G X, i = l,n, then although we have
<pA (x) - sup {ip*A. (x); i = 1, n] , x G X (3.3)
and <p*A. (x) < (fA (x),Vai G X, however G 0 (<p*Ai) £ G0 (<fA,) and D (<p*Ai) •£
D(fA,)<s.
In order to obtain K\ (<PA) = A* (Go (<PA)), firstly we will prove the
inequalities
KI{VA) > inf {e > 0; 3n G N, {At, <pAi) with D* (G 0 (<pAi)) <£,«' =
M , s u c h that G 0 (<pA) C ( J G 0 {<pAi) \ > K* (G 0 (<pA)).

»=i J
Indeed, the second inequality is obvious. Now, let e > 0 be for which there
exist n G N,(Ai,<pAi) with D* (Go{<fA,)) = D{<pAt) < e, i = l,n, such
that <fA (x) = sup {<pA, (x) ;i = l,n] ,Vx E X and let (x,y) G Go (<PA), 2-e.,
0 < y = y ^ (x). Then there exists j G {1,..., n] such that 0 < y — fA (x) =
VAj (a;), i-e., (x,y) £ G0 (<pAj). As a conclusion, G 0 ( ^ A ) C U L i ^ o (<PA.)
which proves the first inequality too.
Conversely, by the proof of Theorem 3.3 in Gal [84] we have
K*(G0{(pA)) = M{e>0]3neN,{Ai,<pA,) with D{<PA.) <£,i = T~n,
such t h a t Go (<PA) C | ^ J G O (fAt) > •

i=l J
Therefore, let e > 0 be for which there exist n £ N , (Ai, <pAi) with
D* (Go (<PAi)) <e,i = l,n a n d
•n
Go(VA) = \jG0(<pAl). (3.4)

i=i
Let x G X. There exist two possibilities:

a) ipA (x) > 0;
b) ipA (x) — 0.
Case a). If we denote y — <pA (x), we get (x, y) G Go {'PA) a n d by relation
(3.4) there exists j G { 1 , . . . , n } such t h a t y — ipA (x) = fAj (X). Let us
denote J = { j 6 { 1 , . . . , n } ; ^>A (^) = fAj {x)}• Obviously we have fA {x) —
sup {fAj {x) ;j G J } . If y>A (a;) / such t h a t 0 < <PA {x) < <pAk (x) = y', which implies
t h a t (x,y') G G0(fAk)- By (3.4) we get (x,y') G G 0 ( ^ ) , i.e., y' =
fA (x), which implies t h e contradiction k G J• As a conclusion, fA (x) —
snp{fAj {x);j G { l , . . . , n } } .
Case 6). Let fA (x) — 0 and suppose t h a t there exists j G { 1 , . . . , n } with 0 <
y = y ^ . (x). T h e n by (3.4) we get (x,y) G Go (fA) a n d t h e contradiction
<PA {X) > 0. Therefore, fAl (x) = 0, for all i = l , n and 0 = V?A (a;) =
sxvp{tpAi ( « ) ; i = l , n } .
As a conclusion, X * (GO(<£>A)) > K\ (<PA)> which combined with t h e
inequality K\ (<PA) > A'* (Go ( ^ A ) ) proves (ii). O
C o r o l l a r y 3 . 2 (i) K\ (<pA V <ps) < m a x { A ' i (<pA), A'I (fa)}-

(ii) hKi (fA V fB) < m a x {/i A'I (fA), hK\ (y?s)}.
Proof, (i) For any fixed 8 > 0, by t h e definition of A'I ( ^ 4 ) a n d K\ (fs)

as infimums, there exist £1, £2 > 0, n, m G N a n d (.Aj, y>Ai), J = 1, n,
(Bj, y s j , j = T~m with D (f^) < ex, i = 1, n, D (y>Bj.) < £ 2 , j = 1, m,
fA(x) = s u p { < ^ < ( » ) ; « ' = l , n } , ^ s ( a ; ) = sup {^ B j (a;); j = l , m } ,x G
X , such t h a t
£1 < A'i (<y?A) + S, £ 2 < A'i (V?B) + S.
This implies
m a x { e i , £ 2 } < m a x {A'i (fA),I<i (fB)} + 6.
Measures oj Noncompactness for Fuzzy Subsets in Metric Space 109
On the other h a n d , we easily have
(<PA V <pB) (X) = sup {pck (x); k = 1, n + m] ,
where C; = Aj,i = 1, n, Cn+j = Bj,j = 1, m and
D(<pCk) <ma,x{ei,£2} for all k = 1, n + m.
We have
A'i (ipA V <pB) < m a x { e i , £ 2 } < m a x { A ' I (ipA), A'i (<PB)} + $
and passing with S —> 0 + we obtain
I<i (<PA V <pB) < m a x {A'i (ipA) , A'i ( V B ) } •
(n) T h e proof is entirely analogous. •
R e m a r k . T h e converse inequalities in («') and (ii), in general are not valid

because it is easy to check t h a t A'i and hK\ lose the property of monotony,
i.e., (A,<pA) C (B,ipg) does not imply t h a t A'i (ipA) < K\ (<PB) and t h a t
hKi (<pA) < hKi(tpB).
T h e fuzzy metric concepts for fuzzy sets can be obtained by replacing
the usual metric space with a fuzzy metric space. At the end of this section
we introduce some concepts concerning this idea.
First, we need some known concepts, i.e., those of fuzzy real n u m b e r
and of fuzzy metric space (see e.g., Mashhour-Allam-El Saady [142]).
Definition 3.17 Let <p, tp : R —> [0,1] be non-ascending functions.

We say t h a t if is equivalent with ip if <£>- (f) = i>- ( t ) , Vf G R and we write
ip ~ ijj, where <p- (t) = l i m s ^ t ^ (s) .
A fuzzy real number is an equivalence class <p under the above relation ~ ,
such t h a t if f 6 (p then £ + (—oo) = 1 and £_ (-f-oo) = 0, where £+ (t) =
lim s v^^ (s) and ^_ (t) = l i m s ^ t ^ (s).
T h e set of all fuzzy real numbers will be denoted by R ([0,1]), while by

R* ([0,1]) one denotes the class of all fuzzy real numbers satisfying y?_ (0) =
1. If <p, V> G R ([0,1]) then () (s) = sup {ip (a) + i> (b); a + 6 = s} .
Definition 3.18 (see e.g., Mashhour-Allam-El Saady [142], p.59). (X,df)

is called fuzzy pseudo-metric space if dj : X x X —>• R* ([0,1]) satisfies:
(i) df (x, x) = 0,Vo: G X, where 0 : R -> [0,1] is given by 0 (s) = 1, if
s < 0 and 0 ( s ) = 0, if s > 0.
(ii) dj (x,y) = df (y, x) , Vz, y £ X.

(Hi) df (x, z) < df (x, y) + df (y, z), \/x, y,z £ X,
where if ip, $ £ R([0,1]) then (p < $ means <p' (t) < ip' (t) ,Vt G R, V €
Now, let (X, df) be a fuzzy pseudo-metric space and (A, <PA) be a fuzzy
subset of X. We can introduce the following
Definition 3.19 The fuzzy diameter of the fuzzy set (A, <PA) will be
Df (<pA)(s) = s u p { d } (a,b)(s) ;a,be G0(<PA)} , s £ R,

where d*f (a,b) (s) = max{d/ (ai,&i) (s), |a2 — 62|} , a = ( a i , a 2 ) ,
b = (h,h)e Go {<pA).
If we replace Go ('PA) by HGQ (<PA), then we obtain the concept of fuzzy
hypo-diameter denoted by hDj (<PA) («)• If Df ('PA) (S) (hDf (<PA) (S))<
+oo, Vs £ R, we say that (A, <PA) is of finite fuzzy diameter (hypo-diameter).
For simplicity, let us denote by D*. (<PA) an Y between Df (<PA) a n d
hDf (cpA) •
Definition 3.20 Let (A,<pA) be with D) (ipA) (s) < +co,Vs £ R. The
fuzzy Kuratowski's measure of noncompactness of (A, PA) will be
(XF(<PA)(S) = inf {e> 0;3n£N, (Ai,<pA,) ,i = T~n with D*f(tpA){s),

< e,such that p>A (x) 0; 3n £ N, 3P,- = (ar,-, r ; ) € M ( y A ) , i = T~H,
such that VP = (x, r) £ M (<pA), 3Pk with d* (P, Pk) (s) < e} , s £ R,
where for simplicity, M (<PA) denotes any between Go (<PA) and HGo ('PA) •
R e m a r k s . 1) In Definition 3.20, we therefore have defined four fuzzy

measures of noncompactness for fuzzy sets.
2) By analogy with the introduction of the so-called concept of abstract
fuzzy measure (for usual sets), it would be interesting to use the axiomatic
definition of the measure of noncompactness in Banach spaces in Banas
[3l], p. 133-134, in order to introduce a concept of abstract fuzzy measure
of noncompactness for (usual) sets.
Measures of Noncompactness for Fuzzy Subsets in Topological Space 111
3.4 M e a s u r e s of N o n c o m p a c t n e s s for F u z z y S u b s e t s i n
Topological Space
T h e m a i n aim of this section is to generalize the results in K u p k a - T o m a

[128] to fuzzy subsets in (fuzzy) topological spaces.
Let m and C be as in Definition 3.4.
By using the weak a-cuts of a fuzzy set, we can introduce the following
crisp-fuzzy kind of definition.
Definition 3.21 Let (X, T) be a classical topological space and A a fuzzy

subset of X (i.e. A : X —• [0, 1]). Then, the (a)-measure of noncompactness
of the fuzzy set A is given by
aM{A)= p| m(Aa),
o6(0,l]
where Aa = {x £ X;A (x) > a} and aM : Ix -+ C, Ix being the class of

all fuzzy subsets of X.
R e m a r k s . 1) If a,/3 G (0,1] satisfy a < /?, then obviously Ap C Aa and

by K u p k a - T o m a [128], Proposition 1, (6), we get m(Ap) < m(Aa) (i.e.
m(Aa) C m(Af})).
2) If A is a classical set, then Aa — A for all a G (0,1] and aM (A) =
m(A). Therefore, in this case aM one reduces to the classical m in Defi-
nition 3.4.
In what follows we introduce the concept of s'-compactness for fuzzy

sets. We recall t h a t a triangular norm is a function t : [0,1] x [0,1] —»
[0,1] which is commutative, associative, monotone in each component and
satisfies the condition t (x, 1) = x. Also, s : [0,1] x [0, 1] —> [0,1] defined by
s(x,y) — 1 — t (1 — x,l — y) is called the corresponding conorm of t. We
extend t to Ix pointwise, i.e. (AtB) (x) = t (A (x) , B (x)). Then t can
be considered as "intersection" of fuzzy subsets. Similarly, ? correspond to
the "union" of fuzzy subsets. Also, finite (or countable) "intersections" and
"unions" are defined in the obvious way (see e.g. Mesiar [146]).
Let (X, Tp) be a quasi-fuzzy topological space (Lowen [136]), i.e. a
fuzzy topological space in Chang's sense and let s be a triangular conorm.
We can introduce the following.
Definition 3.22 Let At G TFti G / and A1,...,A" G TF. T h e family

l n
K. = {(Ai)ieI ,A ,..., A } of fuzzy sets is an open s-cover of a fuzzy set A
if
A C (Vie/A-) sA1s...sAn,
where ( V f e / A ) {x) = sup {A{ (x) ; i £ 1} , x £ X and A C B means A (x) <

B{x),Vx£X.
We say t h a t < (Aj) • 3 , Akl, ...,Ak" > is a finite open s-subcover in K, of
the fuzzy set A, if A C (\/j€jAj)sA s~...'sAk>>,
kl
where J C / is finite and
{k1,...,kp} C {l,...,n}.
Definition 3.23 A fuzzy set A £ Ix is called (C, s)-compact if each

open s-cover of A has a finite s"-subcover of A. A fuzzy set A is called
(X, s)-compact if for all open s-cover of A and for all e > 0, there ex-
ists a finite ?-subcover of B, where B is the fuzzy set defined by B (x) —
m a x (0, A (x) - e), Vx G X.
Because B (x) < A (x) , V s ; £ X , it follows t h a t if A is (C, S)-compact,

then yt also is (L, s)-compact.
R e m a r k s . 1) If the triangular conorm is s (x, y) = so (x, y) = m a x (a;, y),

then (C, So)-compactness becomes the compactness in C h a n g ' s sense (see
Chang [54]) and (L, «o)-compactness becomes the compactness in Lowen's

sense (see Lowen [136] and Lowen-Lowen [137]).
2) If s\ and s 2 are triangular conorm and Si < s?, then any open
?i-cover is an open S2-cover. In particular, any classical open cover (i.e.
Jo-cover) is an open s"-cover because m a x (x, y) < s (x, y) , Va;, y 6 [0,1], for
every triangular conorm s.
3) Let s ^ so- If K, = {(A')jgj > ^ i •••) An} is an open s"-cover of a fuzzy
set, then in general ( V , g / A ) 7A1's...7An is not an open fuzzy set in (X, Tp).
Now, let ns be the set of all open s-covers of (X, T) ,CS - V (KS) the
family of all subsets of 7r 5 , ordered by the relation A < Q if and only if
Ci C A , VA, Q G C S , where C means the classical inclusion. Then, obviously
(Cs, < ) is a complete lattice with the minimal element 0^ = ns.
We can introduce the following.
Definition 3.24 Let A £ Ix, (X,Tp) be a quasi-fuzzy topological space

and s a triangular conorm. T h e (C, s)-measure of noncompactness of A is
the element of Cs given by
msc(A) = {V6ns;38cT>, finite open s"-subcover of A j .
T h e (L, s)-measure of noncompactness of A is the element of C given by
m £ (A) = { p . e 7r 5 ;Ve > 0,36f C £>*, finite, <S» = {(Aj)j€J,A\ ..., Ap}
with ma.x{Q,A(x)-e} < ((Vj€JAj)sA1s...s~Ap)(x),\/x<EX}.
R e m a r k s . 1) Obviously, m^ and mf are similarly defined if (X, T) is a

fuzzy topological space in the Lowen's sense (see e.g. Lowen [136], Defini-
tion 1.1).
2) T h e (L, s)-measure ra£ is suggested by Lowen [136], Definition 4.1
and means in fact s-measure of noncompactness in Lowen's sense.
3) By Definition 3.24 we obviously get
msL (A) < msc (A) (i.e. msc (A) C msL (A)).
4) If si and S2 are triangular conorms and Si < s2, then
mg 1 (A) C mf? (A) and mf 1 (A) C mf 2 (A) for all A £ IX •
In what follows we study the measure a M . T h u s , we present

T h e o r e m 3 . 1 5 Let (X, T) be a topological space. The mapping aM :

I —^ C has the following properties:
(i) If A G / is compact in the sense of Weiss [220], Definiti on 6.0,
then aM (A) = Oc- Conversely, if A is closed (in the sense of Proposition
3.3 in Weiss [220]) and aM (A) = Oc, then A is compact (in the sense of
Definition 3.5 in Weiss [220]).
(ii) IfAcB then aM (A) < aM (B).
(in) aM (AV B) = aM (A) D aM (B).
(iv) aM (AhB) < aM (A) U aM (B).
(where (A V B) (x) = m a x {A (a;) , B (x)} , (A A B) (x) = min{A(x) ,B(x)}
\fx £ X).
Proof, (i) Let A 6 Ix be compact as in Weiss [220], Definition 3.5. It

follows t h a t all Aa,a £ (0,1] are compact in the topological space (X,T)
and then by K u p k a - T o m a [128], Proposition I, (a), we get m (Aa) — Oc, for
all a E (0,1]. This immediately implies
aM(A)= p| m(Aa) = 0C-

<*e(o,i]
114 Defect oj Property in Topology
Conversely, suppose that aM (A) = n<*e(o,i]m (^«) = Oc- If a E (0,1] then

m(Aa) = 0c, since 0C = rLe(o,i] m (^<*) c m (^<*) C ®c- Because Aa
is closed, by Kupka-Toma [128], Proposition l,(a), we obtain that Aa is
compact. Hence A is compact (according to Definition 3.5 in Weiss [220]).
(ii) If A(x) < B(x) ,Vx £ X, then A„ C B«,Va G (0,1], where Aa,Ba
are the corresponding level sets of A and 5 , hence m (Aa) D m (Ba), Va E
(0,1] (see Kupka-Toma [128], Proposition 1, (6)).
We get
aM{A)= P | m{Aa)D f] m(Ba)=aM (B) .

a£(0,l] «€(0,l]
(Hi) Because AaUBa — (A V B)a, by using Proposition 1, (c), in Kupka-

Toma [128] we set
m (Aa) n m (Ba) - m (Aa U Ba) = m((AV B)a) .
Therefore,
aM(A)naM{B) = I Q m ( ^ a ) ) n ( f] m(Ba)
\«e(o,i] / \ae(o,i]
= f) (m(Aa)C\m(Ba)) = f] m((AV B)a)
»6(o,i] «e(o,i]
= aM(AVB).
(iv) The proof is immediate by (ii). •
The next example proves that the inequality in (iv) can be strict.
Example 3.13 Let X = R be endowed with the usual topology and
A (x) - - if x G [0, +oo) and A (x) = 0 if x G (-co, 0),
B(x) = - if x e (-co, 0] and B (x) = 0 if x G (0, + c o ) ,
two fuzzy subsets of X. Then (A A B)a = {0} for a G (0, \] and (A A 5 ) a

= 0for a G [ | , l ] , hence a i t f ( A A 5 ) = 0 c .
On the other hand,
aM(A) PI m(Aa) I D = m([0,+oo))

V«e(o,i] \«e(i,i]
and
t \
aM(B) D m(Ba) | fl •((-oo,0]),
\«6(0,i] \ae(i,i]
but a M (A) U aM (B) 7^'Oc, because the open cover

V = {(n — 1, n + 1) ; n 6 Z} of X does not contain a finite subcover neither
of the set [0, +oo), nor of the set (—oo, 0].
Remark. Theorem 3.15 is an extension of Proposition 1 in Kupka-Toma

[128] (see also Theorem 3.2).
Now, for the (C, s)-measure of noncompactness we get

ping m,Q : Ix —>• Cs has the properties:
(i) If A 6 Ix is (C, 7)-compact then rag {A) — 0%.
(ii) If Ac B then m% {A) < msc (B) .
(Hi) msc (A V B) = msc (A) n msc (B) .
(iv) msc (AAB) < rag (^4) U m£ (B).
Proof, (i) Let us suppose that A 6 Ix is (C, s)-compact and V is an

open J-cover of X. Then evidently V also is an open ?-cover of A, so that
there exists 6, a finite JT-subcover of A such that S <ZT>. Thus m^ (A) = 0§.
(ii) If V G 7r5 contains a finite ^-subcover of the fuzzy set B, then
the same J-subcover also covers A and this proves the inclusion m* (B) C
ITIQ (A), which is equivalent to rag (A) < rag (B) .
(Hi) Because of (ii) it is evident that rag (AV B) > mg (A) and
rag (,4 V 5 ) > m g ( 5 ) , hence rag (A V B) > rag (A) f~l rag (B). In or-
der to prove the converse inequality, which is equivalent to the inclu-
sion ITlr (A) n rag (B) C m%(AV B), let V E rag (A) n rag (B) , V =
{(Ai)iei , A1, ...,An}. Then there exists K C I,K finite and {k\, ...,kp} C
{l,...,n} such that K. = {(^/Ofceit ,Ak\ ...,Akp} is a finite S"-subcover of
A, that is
A C (Vk€KAk)sAklI...sAk".
Also, there exists a finite subset L C I and {h,...,lr} C {l,...,n} such

l1 lr
that C = {{Ai)t£L ,A , ...,A } is a finite s-subcover of B, that is B C
(Vi£LAi)sAhs...sAlr. Then
AVflC (Vj e jfuL^j)s^' Cl s.--s^ fe,, s^' 1 s---S-A' r ,
which implies that ICllC — i (Aj)-eKuL ,Akl,..., Ak», Ah, ...,Alr > is a finite
s-subcover (in V) of A V S. This proves that V £ m^ (A V B) .
(iv) The proof is evident by (ii). •
The next examples prove that for s = So and s = SQO the converse
statement in (i) can be false even if A is C-closed (i.e. in the Chang's
sense, see Chang [54]).
Example 3.14 Let X = R be endowed with the quasi-fuzzy topol-
ogy Tp = {D; D (x) = c £ [0,1], \/x £ R } . If s = s0, where s0 (x, y) —
max(x,y) is the triangular conorm corresponding to the usual union, then
the fuzzy set A (x) = | , Vz G X is a C-closed fuzzy set and obviously
mc° (A) = 0 C °. But A is not (C,?o)-compact (i.e. compact in Chang's

sense) because the open so-cover of A, {Pi;i= 1,2,3,...} where Pi (x) =
I — j , Mx G X, does not contain a finite so-subcover of A.
Example 3.15 For arbitrary X, let us consider
TF = {/„ : X - ^ [ 0 , l ] ; / „ ( a : ) = l - ^ ) n € N * ) n - e v e n | u
U {</„ : X ->• [0,1] ;gn (x) = 1 - ^ , n G N, n > 3 | U {0, X } ,
where 0,X G / x , 0 (a:) = 0 and X (x) = l,\/xEX.

It is easy to see that Tp is a quasi-fuzzy topology on X. We will prove that
the fuzzy set A (x) = h, Va: £ X is a C-closed set with
m c °° = 0C°°, but A is not (C, s'00)-compact.
Indeed, by \ = 1 - (l — | ) we get that ^4 is C-closed. Also, A is not (C, Soo)-
compact, because by A (x) = \ln>zgn (x), Va; £ X, we cannot choose a finite
number of g„ (x) with ^4 (x) = \/gn (x).
On the other hand, if X = (Vkeihk)ôoA1's00...s00A'n, we have the follow-
ing possibilities: (i) I is finite; (ii) I is infinite and hk = fk for an infinite
number of k; (Hi) I is infinite and h^ = gu, for any k > fc0-

The case (i) obviously implies that A is covered by a finite Soo-subcover.
In the case (ii), we can choose a ko G / with \ < hk0 (x) = fk0 (x), Vx G X
and A C /ifc 0 s oo A 1 s oo ...s 0O A m .
Finally, in the case (Hi) we get Vkzihk (x) = | , Va; 6 X, which implies
- = A1 (x) Soo-.sÂ"1 (x) ,\fx £ X, i.e. A C Al'Sx...scoAm.
As a conclusion, taking into account the definition of m^ , we get

m£«(A)=0|~.
Note that any fuzzy set A satisfying A (x) < c < \,*ix & X is (C, s<x>)-
compact.
However, a natural question arises: if A is C-closed and mc°° (A) = 0C°°,
then in what conditions A still has some properties of compactness ?
The following remarks give answers to this question.
R e m a r k s . 1) Firstly, we will prove that if (X,TF) is (C, Soo)-oompact

(which implies msc°° (X) = 0^°° and therefore msc°° (A) = 0^°°), then each
C-closed fuzzy set A C X also is (C, Soo)-compact.
Indeed, let A C (Vi£iAi)'s00A17oo...?sO0Arn be given with I finite. Since A

is C-closed and ASoo (1 — A) = X, we obtain that 1 — A is open and
X = (V,-6/Ai) s 0O A 1 s 0O ...s 0O A m s 0O (1 - A) .
Taking into account that X is (C,?oo)-compact, there is J C /, ./-finite

with
1 = max {At (x) ;i G JjsÂ1"1 (x) s00...scoAkp (x) ,Vx G X
or
1 = max {A- (x) ;i£ J} s00Ahl (x) Soo ...scaAk" (x) sTC (1 -A(x)),Vx G X,
where k\,..., kp G {1,2,..., m} , k{ ^ Aj.

In the first case, we obviously can cover A with a finite Soo-subcover. In the
second case, denoting B (x) — max {A; (x) ;i G J} ScoAkl ( xjs00...s00A. p \Xj
we therefore can write
l = B(x)8oo{l-A{x))tVxeX,
i.e.
• sat ^ i At \ U B(x)-A(x) +2 \B(x)-A(x)\
1 = mm{B(x) = 1-A(x),l}= ,
for every x G X, which means
0= (B(x)-A(x))-\B(x)-A(x)\,Vx£X.
This immediately implies A(x) < B (x) ,\/x £ X, where B (x) is a finite
Seo-subcover.
2) Let (X,Tp) be an arbitrary quasi-fuzzy topological space (not nec-
essarily (C,Soo)-compact) and A C X,A being C-closed and satisfying
m£~ (A) = 0c°°- If moreover ,4 (a;) > 1 - £ 0 ,Vx G X (where £ 0 G (0, § ) )
then for any e G [eo, 1] and any 5oo-cover of A, there is a finite Soo-subcover
of A - e.
Indeed, let A C (VieiAi) 's00Al's00...'s00Am = B. Reasoning exactly as
the above remark, we get X = BSco (1 — A), and taking into account t h a t
msc°° (A) = 0;?°°, we can write
A C (\JiejAi)'soaAkl's00...s00Ak<'
or
A C {ViejAi)s00Akls00...'s00Ak<>s00 (1 - A),
where J C / is finite and ki, ...,kp E { 1 , 2 , . . . , m } , k{ ^ &,-.

In the first case it is obvious t h a t A — s C (V,-£j>lj) 2' 00 J4' !I S' 0 O ...S 0 O J4 A : P , for
all e G [eo, 1].
In the second case, we easily obtain
2A-1C (Vi€jAi)scoAklsO0...s00Ak'>.
Taking now into account t h a t
0 < A(x)-e< A(x)-e0 < 2A (x) - l , V x G X, Ve G [ e 0 , l ] ,
we get our conclusion.
T h e o r e m 3 . 1 7 Let (X,Tp) be a quasi-fuzzy topological space. The map-

ping m £ : Ix —> Cs has the properties:
(i) If A G Ix is (L,Tj- compact then msL (A) = 0 | .
(ii) If AC B then mf (A) < msL (B).
(in) mf (AV B) = m£ (A) H m[ (B) .
(iv) msL (AAB) < ml (A) U ra£ (B).
Proof, (i) If A G Ix is (L, s)-compact and 25* is an open s-cover of X,

then Vt is an open 5T-cover of A and for all e > 0 there exists J C £>*, a
finite ?-subcover of B, where B (x) = max (0,-A (at) — e) , Var G X. Thus
m£(A)=0g.
(if) Let D* G TTS and e > 0. If X>* contains a finite s"-subcover of the
fuzzy set defined by max(0, B (x) — e), Mx G X, then the same s"-subcover
also covers the fuzzy set defined by ma,x(0,A(x) — e) ,Vx G X. The in-
equality max(0, A (x) - s) < max(0, B (x) — e), Va: G X implies the inclu-
sion ro£ (B) C »7i£ (A), which is equivalent to msL (A) < m£ (5) .
(Hi) The inequality m£ (AVB) > msL (A) n m£ (5) is evident because
of (ii). In order to prove the inclusion ra£ (A) fl m£ (S) C m£ (A V B), let
2>* G msL(A)f\msL(B),V)f = {(A,-) i6/ , A 1 ,..., An} and e > 0. Then there
exist K C I, K finite and {A?i, ..., kp} C {1,..., n} such that
max (A (a;) - e , 0 ) < (\ZkeKAk) (x) sAkl (x) s...sAk" (x) ,Va; G X.
Also, there exist a finite subset L C I and {/i, ...,/ r } C {1, ...,n} such that
max (B (x) - e , 0 ) < (V, e L ^ ( ) (a?) s.4'1 ( i ; ) s . . . s i ' ' ( i ; ) , V r £ l .

The equality
max (max (^4 (x) , B (x)) — s,0) = max (max (A (x) — e, 0) ,
max(B(x)-e,0)),Va:GX
implies
max((.4(a:) V B ( i ) ) - e , 0 )
< {VJZKULAJ) (X) sAkl (x) s...sAk" (x) sAh (x) s...sA1'- (x),
for every x G X. This proves that X>„ £msL(AV B) .

(iv) It is evident from (ii). D
In what follows we deal with the connections between upper semiconti-

nuity and measures of noncompactness.
Firstly we recall the following:
Definition 3.25 (Kupka-Toma [128], p.458). Let ( r , < r ) be an upward

directed set. If (x^\j G T) is a net in C, we will write l i m 7 g r ^ ^ 7 ' — Oc, if
V {B} G C, 3 7 o G r such t h a t x™ < {B} , V 7 > r 7o-
Let us introduce some new concepts for fuzzy sets.
Definition 3.26 Let (X, T) be a topological space and (T, < r ) an upward
directed set. Then (A^ G I X ; 7 G r ) is called decreasing net if fi < r 72
implies A ^ C A™. If (Y,TF) is a quasi-fuzzy (or fuzzy) (see e.g. Lowen
[136], p.622) topological space, then a m a p p i n g F : X —¥ IY will be called
fuzzy set-valued m a p p i n g (see e.g. Gal [85]).
Definition 3.27 We say t h a t F is u.s.c. (upper semicontinuous) at x$ G

X, if for any U G Tp with F (xo) C U, there exists a neighborhood V (XQ)
of Xo (in the classical topology T) such t h a t F (t) C U, W G V (XQ) .
On R we consider the topology r = { ( a , 0 0 ) ; a G R } U {0}. T h e topol-

ogy which one obtains by taking on / = [0,1] the induced topology will be
denoted by T• If (Y, Tp) is a fuzzy (or quasi-fuzzy) topological space, then
we denote by T1 the initial topology on Y for the family of "functions" Tp
and the topological space T . T1 is the topology associated in a n a t u r a l way
to the fuzzy topology Tp (see Lowen [136]).
Definition 3.28 Let (X,Tp) be a quasi-fuzzy topological space. T h e

fuzzy set-valued m a p p i n g F : X —>• IY is called (a)-u.s.c. at Jo £ I if
the usual set-valued mappings Fa : X —> V (Y) are u.s.c. at XQ for all
a G (0,1], where Fa (x) = A^ if F (x) = A^ G IY and the topology
considered on Y is T . We say t h a t (.f^ ) : X —y 7 y ) _ is a decreasing
1 7
net if (_F(T) (X^ £ jY^ ls a decreasing net for all x G X.
We can present the following two results which are extensions of the
Theorems 1 and 2 in K u p k a - T o m a [128].
T h e o r e m 3.18 Let (X,T) be a topological space and (T, < r ) an upward

directed set. If (A™; 7 G T) is a decreasing net of closed subsets in Ix (in
the sense of Proposition 3.3 in Weiss [220]) which satisfy the condition
V7 G T, 3a;o G X (depending on 7) with A^1' (XQ) = 1, (3-5)
then the following implication is true:
limaM = Oc =>• A = A 7 gr^4 is nonempty and compact
(in the sense of Definition 3.5 in Weiss [220]).

Proof. By the definition of a-measure of noncompactness, we have
limaM ( A ™ ) = 0C <* lim f] m (A^ = Qc.

7 7
«e(o,i]
Using Definition 3.25 we obtain
V{B}£C,3lo£T:BC f| m ( A ™ ) , V 7 > r To
"6(0,1]
or equivalently
V { 6 } 6 C , 3 T „ e r : 6 c m ( A ™ ) ,V 7 > r 7o,Va e (0,1],
that is
limm (i4W) = 0 c , V a G ( 0 , l ] .
For 7 G r,yl( 7 ' is closed and satisfies condition (3.5), therefore A„ is

closed (and nonempty by (3.5)), for every a £ (0, 1] .
Moreover, (Aa,-/£ Tj is a decreasing net of subsets in V{X), for

every a £ (0,1]. By Kupka-Toma [128], Theorem 1, we get that n 7 6 r A «
is nonempty and compact for every a £ (0,1], therefore A = A-y^rA^ is
nonempty and compact, because pLer-^" = ( ^ 7 € r ^ 7 0 a - ^
The next example proves that condition (3.5) in Theorem 3.18 is essen-
tial.
Example 3.16 Let X = R be endowed with the usual topology and
r = N \ {0,1}. Then, (A^ :k£T), where A™ : X -> [0,1], A^ (x) = ±
if x £ [0, £] and A^ (x) — 0 contrariwise, is a decreasing net of closed
fuzzy subsets, because Ah ' = 0 if a > £ and Ah,' = [0, £] if a < \.
Also
limaM UÂ = lim H m (AÂ

fcer \ / fcer
«e(o,i]
/
lim
fcer \»e(o,t]
n -(^n n m{A{"))
W* i
= limm
o-k nm(0) = Oc.
Moreover, A^ is a nonempty fuzzy set for all k £ T. However
( A * 6 r ^ ) (z) = inf A^ (x) = 0,Vx £ X.
T h e o r e m 3.19 £e£ (A", T) fee a topological space, (T, <p) an upward di-
rected set and (Y, TF) a quasi-fuzzy topological space. If (F(T) : X ->• J y ;
7 6 T) zs a decreasing net of (a)-u.s.c. fuzzy set-valued mappings on X
such that each F^"1' (x) = Ax is a closed fuzzy set (in the sense of Propo-
sition 3.3 in Weiss [220]) which satisfies condition (3.5) in Theorem 3.18
and if for each x £ XAimy^rotM [Ax') = Oc, then the fuzzy set-valued
mapping F : X —t IY given by F (x) = A 7 e r ^ ' 7 ^ (x) ,x £ X, is (a)-u.s.c.
on X, with nonempty and compact values (in the sense of the Definition
3.5 in Weiss [220]).
Proof. limygraM lij. 1 ') = 0C, Vx G X implies (see the proof of Theo-
rem 3.18),
limm ( ( ^ ) ) J = foam ( ( f W (*)) J = 0C, V, G X, Va G (0,1].
Let a G (0,1]. If (F^;j £ T) is a decreasing net (see Definition 3.28),

7
then (F^ ) (X) ;-f £ T) is a decreasing net for all x £ X. Because F^ is
(a)-u.s.c. V 7 G r , F i 7 ) : X -> V (Y) is u.s.c. V7 £ T,Va G (0,1], if we
consider on Y the topology T1 (see Definition 3.28). Moreover, (F^ (x))
is closed for every 7 G T,a £ (0,1] and a; G X, and the condition (3.5)
implies that [Ax1') = (F^ (X)) is nonempty for every 7 G I\ a £ (0,1]
and a; £ X. By Kupka-Toma [128], Theorem 2, applied for the net of
classical multifunctions [F^1' : 7 G Tj , we obtain that the multifunction
Fa : X^V(Y),
Fa(x) = f][F^(x))a
is u.s.c. with nonempty compact values of T1. But
n ( ^ (-))„=n ( ^ 1 = M > ) a = ( A 7 e r ^ > (x))

7€T 76T
Hence Fa is u.s.c. and ( O v e r ^ (x)) ^s c o m P a c t ar>d nonempty for every

a G (0,1], if we consider on Y the topology T1. Therefore, the fuzzy set
valued mapping F is (or)-u.s.c. on X with nonempty and compact values
(in the sense of Definition 3.5 in Weiss [220]). •
In what follows, we consider the measures of noncompactness m^. and

m£ introduced by Definition 3.24. Firstly, we need two concepts.
Definition 3.29 (Lowen [136]). Let {X,TF) be a quasi-fuzzy topological

space (or a fuzzy topological space in the sense of Lowen in [136]) and
A £ Ix. We say that A is C-closed (L-closed, respectively) if B £ Tp,
where B (x) = 1 - A (x), Vx £ X.
Definition 3.30 Let (r, < r ) be an upward directed set. If (x^) : 7 6 F)

is a net in C , we will write l i m 7 € r x ( 7 ) = 0 | , if V{#} £ Cs ,3f0 £ T such
that x™ < { £ } , V 7 > r 7 o -
We present
Theorem 3.20 Let (X,Tp) be a quasi-fuzzy (or fuzzy) topological space

and (r, < r ) an upward directed set. If (A^ : 7 £ T) is a decreasing net

of nonempty and C-closed fuzzy subsets in Ix which satisfy the condition
(3.5) in Theorem 3.18, then the following hold:
limm|°° (AÂ = 0^°° =>• A = A 7 6 r-4 ( 7 ) is nonempty,
C-closed and msc°° (A) - o|~.
Proof. Suppose that A - 0. Then X = X\A = X\ (Al€TA^)

= V 7 € r (X\A^). The hypothesis implies VP 6 T T S ~ , 3 7 0 G T : V7 > r
7o, V £ mc°° [A^>). If we apply this condition to the open Soo-cover
V = {X \ A^\~i G r } , we can state the existence of a finite Soo-subcover
of the fuzzy set A^, i.e. A^ C V? =1 (X\A^'*>). Let us choose 7 > r
7 i , i G { 0 , l , . . . , n } . Then AW C A^ ,Vi £ {0,1, ...,n}, therefore X\A™
D X \ A^ ,\/i £ {0,1, ...,n} and
A(~f) c A^°1 C V? =1 (x \ A^'A C V? =1 (X \ A ™ ) =X\ A™,
which is a contradiction with the condition (3.5). As a conclusion, we obtain

A ^ 0. Since (A^ : 7 G T) is a decreasing net, we get
0f~ = l i m m £ ~ ( A ( 7 ) ) = inf [ m ^ ° ( ^ h ) ) , 7 G r }
(here inf is in ( C 5 ~ , < ) ) = \J m £ ~ (^7>)
W
> m£~ ( A 7 6 M ) = ™c~ (A) > 0|~,
z.e. mf.~ (A) = 0 | ~ .

Obviously then A is C-closed. D
T h e o r e m 3.21 Let (X,Tp) be a fuzzy topological space and ( I \ < r ) an

upward directed set. If (A^ : 7 £ r ) is a decreasing net of nonempty and
L-closed fuzzy subsets in Ix which satisfy the condition (3.5) in Theorem
S. 18, then the following hold:
limmf 0 0 (A^ = 0f~ =» A = A 7 € r 4 ( 7 ) ts

nonempty, L-closed and m £ ~ (A) = 0C°° .
Proof. Similar to the proof of Theorem 3.20. •
R e m a r k s . 1) T h e condition (3.5) in the hypothesis of T h e o r e m s 3.20 and

3.21 can be replaced by the following:
V7 G T, 3x0 G X (depending on 7) with A™ (x0) > - . (3.6)
2) Because of the examples given after Theorem 3.16, in the conclusions

of Theorems 3.20 and 3.21 we cannot state t h a t A is (C, Soo)-compact and
(L,S 0 0 )-compact, respectively. In fact, the compactness cannot be derived
neither for other triangular norms, as can be seen in the following example:
Let X - R , TF = {a \a : X -> [0,1], a (x) = a G [0,1]} , s (x, y) = s0 (x, y)
= max(x,y). Choose T = { 3 , 4 , . . . } and A^ (x) - \ + ~,\/x G X. T h e n
(A^ : 7 G r ) is a decreasing net of nonempty and C-closed (or L-closed)
fuzzy subsets which satisfies the condition (3.6). Also, lirruygr^c ( ^ ^ ) =
Of, because m$ (A™) - 0£ for all 7 G T. Finally, A G Ix,A{x) =
Dejects of Opening and of Closure for Subsets in Metric Space 125
(AyzrA^') (x) = ^,Va; £ X and A is not (C, s)-compact (see also the
example after Theorem 3.16).
At the end of this section, we briefly present another kind of measure
of noncompactness in fuzzy topological spaces.
Let (X,Tp) be an arbitrary fuzzy topological space.
Definition 3.31 (Lowen-Lowen [137]). We say that X is (a, e)-compact
(where a £ (0,1] ,£ £ (0,a)) if each open a-cover has a finite sub-(a,e)-
cover.
We say that X is a + -compact (where a £ (0,1]) if it is (a, s)-compact for
all e £ (0,a).
We say that X is ^"-compact if it is (a, e)-compa,ct for all a £ (s, 1).
Remark. (Lowen-Lowen [137]) X is compact in the sense in Lowen [136]

if and only if it is (a, £)-compact for all a £ (0,1] and s £ (0, a), if and only
if it is a + -compact for all a £ (0,1], if and only if it is £~-compact for all
££(0,1).
Definition 3.32 (Lowen-Lowen [137]). The degree of compactness of X

is defined by
c (X) = sup {1 — e; X is £~-compact} .

Theorem 3.22 (Lowen-Lowen [137]). (X, Tp) ts compact (in Lowen
[136] sense) if and only if c (X) = 1.
Remarks. 1) The quantity dc (X) = 1 — c(X) can be called defect of

compactness (or measure of noncompactness) for X.
2) For other details see Lowen [136].
3.5 Defects of Opening and of Closure for Subsets in

Metric Space
Let (X,T) be a topological space, i.e. 1 ^ 1 and T a topology on X.

The following concepts are well-known.
Definition 3.33 A subset Y C X is called:
(i) open, if Y £7~;
(ii) closed, if X \ Y £ T, i.e. if Y = X \ A with A £ J.
Remark. Let (X, p) be a metric space and TP be the topology generated

by the metric p. The following characterizations are well-known:
Y C X is open if and only if Y = intY, where
intY = {y G Y; 3r > 0 such that B (y; r) C Y}
and
B(y;r) = {z G X;p(y, z) < r} .
Y C X is closed if and only if Y = Y, where
Y = iz G X; 3yn G V, n G N such that /> (j/„, z) " H ? ° O } .
If Y C X is not open or closed, it is natural to introduce the following
Definition 3.34 Let (X,p) be a metric space and D (A, B) be a certain

distance between the subsets A, B C X. For Y C X, we call:
(i) defect of opening of Y, the quantity dop (D) (Y) = D (Y, intY) ;
(«') defect of closure of Y, the quantity dCL (D) (Y) = D (Y, F ) .
Remark. A natural candidate for .D might be the Hausdorff-Pompeiu

distance
DH (YI , Y2) = max {p* (Y,, Y 2 ), p" (Y2, Yx)} ,
where p* (Yi,Y2) = sup {p (yx, Y 2 ); yx EY1},p(yi,Y2) = inf {p (yu y2) ;

J/2 G Y2} . However, Z?if seems to be not suitable for dci and dop intro-
duced by Definition 3.34. Indeed, we have
dCL (DH) (Y) = DH (Y, Y) - DH (Y, T) = DH (Y, Y) = 0,
for all Y C X and
dop (Z?) (Y) = 0,
for all convex Y C X, when (X, ||-||) is a normed space and p(x,y) =
Ik - 2/11 (by using the relation Y = intY for convex Y, see e.g. Popa [165],
p. 13).
Bibliographical Remarks and Open Problems 127
That is why instead of DJJ we have to use another distance, as for

example, D* : Vb (X) x Vb {X) -» R + given by
D* (Y Y\ = 1 °' if Yi = Y2
1 U V
\ sup {p(y1,y2)]yl£Yl,y2£Y2}, ifYx^Y2,
where Vb {X) = {Y C (X, p) ; Y ^ 0, Y is bounded} (we recall that by def-
inition Y is bounded if diam (Y) = sup {p (x, y) ;x,y £ Y] < +oo).
R e m a r k . It is not difficult to prove that D* is a metric on Vb (X) .

As an immediate consequence, it follows
Theorem 3.23 / / (X, p) is a metric space and Y £ Vb {X), then:
(i) dop (-D*) (Y) = 0 if and only ifY is open;
(ii) dcL {D*) (Y) = 0 if and only ifY is closed.
In addition,
dCL(D*)(\Y) = \dCL(D*)(Y),V\>0;
dOP{D*)(XY) = XdOp(D*)(Y),VX>0.
Proof. While the first two relations are obvious, the last two relations
follows from XY - \Y and int (AY) = Xint (Y), VA > 0. •
3.6 Bibliographical Remarks and Open Problems
Definitions 3.12-3.14, 3.16, 3.19, 3.20, Theorems 3.9, 3.12-3.14 and Corol-
laries 3.1, 3.2 are in Gal [84], Theorems 3.10, 3.11 are in Gal [85], Definitions
3.21-3.24, 3.26-3.28, 3.30 and Theorems 3.15-3.21 are in Ban-Gal [22]. New
are Definition 3.34 and Theorem 3.23.
O p e n problem 3.1 An open problem is to introduce and study concepts
of measures of noncompactness for fuzzy subsets of a PM-space. In what
follows we point out some helpful details. Firstly we need to recall some
concepts in random analysis.
It is known (see e.g. Istratescu [105], p.25) that to any usual metric space
(5, p) can be attached the so-called simple probabilistic metric space of
Menger-type, for any *-norm T, denoted by (S,FP,T), with Fp : S x
S -> A defined by Fft,{x) = G(x/p(r,s)), if r ^ 8,Fft,(x) = H (x),
if r = s, where G £ A is fixed and satisfies limn-KxjG (x) = 1 and
H(x) = 0, if x < 0,H{x) = 1, if x > 0. Also, if ( S i , F , T ) and (S2,G,T)
are two PM-spaces, by the T-product of them we mean the ordered triplet
(Si x S2,T(F,G),T), where T(F,G) : (Si x S2) x (Si x S2) - • A is de-
fined for all r,s £ Si x S2, r = (ri, r2), s — (s\, s2) by
T (F, G) (r, «) (*) = T (Frit$1 (x), Gr2,S2 (x)) .
(see e.g. Egbert [69], Xavier [226], Istratescu [105], p.81-82). Given (S, F, T)
a PM-space of Menger-type, by a fuzzy subset of it we mean a mapping
A : S —> [0,1]. Let us attach to the metric space ([0,1] ,p) ,p* (x,y) =
\x — y\, its PM-space ([0,1], Fp , T) and let us consider its T-product with
(S, F, T), that is (S x [0,1], T (F, F"') , T).
In Gal [83] were introduced the concept of probabilistic HausdorfF distance
between two fuzzy subsets A,B of (S, F, T), by
D{p (A, B) (x) = supT ( inf sup T (F, FP') (r, s) (t),
t<x \r6G0(A)s£Go(B) \ J
inf sup T(F,F"'\ (r,s)(t)) , x G R,

W
s€G0(B)reGo(A) V / y
where
Go (A) = {(x,y)£Sx[0,l];Q<y = A(x),x€S],

G0(B) = {(x,y)€Sx[0,l];0<y = B(x),xeS},
and the concept of probabilistic diameter of a fuzzy subset A of (S, F, T),

by
Dp (A) (x) = sup ( inf T (F, F"') (r, s)(t)),x£ R.

t<x \r,seG0(A) \ / y
If sup {Dp (A) (x); x 6 R } = 1 then we say that the fuzzy set A is proba-
bilistically bounded.
Then we can introduce the random probabilistic Kuratowski's and Haus-
dorff's measures of noncompactness of the fuzzy set A, by
ap (A) (x) — sup {e > 0; 3n G N, A,- : S -> [0, l],i = T^n,
with Dp (At) [x) > e and A (t) 0; 3Ae finite fuzzy subset of A
such that D{P {A, Ac) (x) > s\ ,
respectively, where by a finite fuzzy subset of A, denoted here Ac, we mean

that there exist n 6 N and a\,..., an G A, such that Ae : S —• [0,1], Ac (s) =
0 if s ^ a,-, Vi = 1, n and i4e (a;) = yl (a»), Vf = 1, n.
The above concepts that combine fuzziness with randomness, obviously
extend the corresponding random (probabilistic) concepts in Section 3.2.
Therefore, would be interesting to extend/study the results mentioned in
Section 3.2 for the above ap (A) (x) and hp (A) (x).
Chapter 4
Defect of Property in Measure Theory
If a set function (particularly, a fuzzy measure or a non-monotonic fuzzy

measure) is non-additive or non-monotonic, it is n a t u r a l to look for a con-
cept which measure the deviation of t h a t set function from additivity or
from monotonicity. In Section 4.1 and Section 4.3 we introduce and study
these deviations, called by us defects. T h e non-additivity implies t h a t the
s u m between the measure of a set and the measure of its complement is
not equal to the measure of entire space. As a consequence, in Section 4.2
we introduce and discuss the defect of complementarity. Defects of other

properties (subadditivity, superadditivity, submodularity, supermodularity,
k-monotonicity, k-alternativity)are discussed in Section 4.4. Finally, in Sec-
tion 4.5 we discuss two kinds of defects of measurability for sets.
4.1 Defect of Additivity: Basic Definitions and P r o p e r t i e s
T h e measure is one of the most i m p o r t a n t concepts in m a t h e m a t i c s and

so is the integral with respect to measure. In applications, the property
of additivity is very convenient, but it is considered too rigid. For exam-
ple, the experience of artificial intelligence researches shows t h a t the use of
probability functions for describing subjective judgements is unjustified and
t h a t it leads to m a n y erroneous results (see Szolovitz-Pauker [208], Wierz-
chon [22l]). As a solution, the concept of fuzzy measure was proposed by
Sugeno [204], generalizing the usual definition of a measure by replacing
the additivity property by a weaker requirement.
T h e concepts of fuzzy measure and their corresponding fuzzy integrals
constitute an i m p o r t a n t topic in fuzzy m a t h e m a t i c s (see e.g. Butnariu-
131
132 Deject of Property in Measure Theory
Klement [52], Congxin-Minghu [57], Dennenberg [61], [62], Dubois-Prade

[64]-[67], Kruse [126], Kwon-Sugeno [131], Lee-Leekwang [134], De Luca-
Termini [138], Murofushi-Sugeno [l56]-[l59], Ralescu-Adams [168], Schmei-
dler [189], Suarez Diaz-Suarez Garcia [202], bugeno [204], Sugeno-Murofushi
[205], Wang-Klir [219], Wierzchon [221], [222], Yoneda-Fukami-Grabisch
[228]). In this section we consider the following definition of fuzzy measure:
Definition 4.1 (see e.g. Ralescu-Adams [168]) Let X be a set and A be
a cr-algebra of subsets of X. A set function /i : A -» [0, oo) is said to be a
fuzzy measure if the followings hold:
(ii) A,B € A and AC B implies (i (A) < ft (B) .

If moreover we have
(Hi) {An} C A,A„ /~ A implies fx(An) / * ft (A) (continuity from
below);
(iv) {An} C A,An \ A implies n(An) \ n(A) (continuity from
above),
then n is called continuous fuzzy measure.
Comparing it with a classical measure, the main difference is that the
fuzzy measure is not additive. Consequently, appears as natural the fol-
lowing question: how can be evaluated the degree of additivity of a fuzzy

measure ?
In the sequel we give an answer to this question, by introducing the
concept of defect of additivity for a fuzzy measure. Also, for A-additive
fuzzy measures we are able to calculate and estimate this defect. Then, we
estimate the defect of additivity for some fuzzy measures generated by a
solution of the general functional equation
f(x + y)- f{x) - f(y) = <p{x, y).
Finally, we give applications to some questions like: the approximative

calculation of some fuzzy integrals, the introduction of a metric on the
family of fuzzy measures and the best approximation of a fuzzy measure.
Let fi : A —> [0, oo) be a fuzzy measure, where A C V{X) is a a-algebra.
Definition 4.2 The defect of additivity of order n, n G N, n > 2, for the
measure fi is given by
a„ (n) = sup •
** IM< -I>(^-)
Deject of Additivity: Basic Definitions and Properties 133
A{ £A,Air)Aj = 0,i ^ j}
T h e defect of countable additivity for the measure [i is given by
ctoo in) = sup • P \{JAi) -^2^(Ai

\i=i / »=i
At G A, Vi e N , A,- n A,- = 0, i # j }
We have
T h e o r e m 4.1 (i) o„ (^) < a n + i (//) , Vn > 2;

(M') / / // is a continuous from below fuzzy measure then a^ (//)
= lim„_ j . 0 0 a n (/i);
( m ) a n (//) < a „ _ i (/i) + a 2 (//), Vn > 3;
(it>) a n (//) < (n — 1) a 2 (//) , Vn > 3;
(v) an{fi)<(n-l)fi(X),\/n>2;
(vi) If fi is a superadditive fuzzy measure then an (/j,) < /j, (X), Vra > 2.
Proof, (i) It is immediate by taking An+i = 0 in the definition of

(ii) Using the continuity of fi on increasing sequences of sets (see Defi-

nition 4 . 1 , (Hi)), we have
** \UA': -X>M=
Vi=l
lim
n-+oo
^ IM -I>(^-) < lim a„ (//)
for every sequence ( A j ) i e N C .4, A{C\Aj = 0 if i ^ j . Passing to s u p r e m u m

we obtain a ^ (//) < lining ooa„ (//).
For the proof of the converse inequality, let (-*4.,-).eN C .4 be a disjoint
sequence such t h a t Am = 0, Vm > n + 1. We obtain
MIM<
Vi=l
"5> ( j 4 < ) ^ IM -&(^«-
^.»=1
< aoo (//) ,
therefore an (fi) < a ^ (/x) , Vn > 2. Passing to limit with n —> oo, we get
lim n ( / i ) < a T O (//) •
As a conclusion,
(//) = lim a n (//)
134 Defect of Property in Measure Theory
(Hi) We get
n \ n
( (J A- -X>(A) /n-1 n-1

;=i / 1=1
H ((jAi) - » (\jAi) - fi(An) + » [ [JAi] -^(Ai
\i = l V! = l / fcl
/n-1 /n-1 \
< M U^ - I > ( A ] »=i

f U ^ -M
\i = l
\jAA-^{An)
< a n _ i (yu) + a 2 (/i)
for every disjoint family {Ai,...,^4 n } C .4, which immediately implies

an (p) < a„_i (pi) + a2 (p).
(iv) From (Hi), by recurrence we obtain 0,3 (p) < a2 (pi) + a2 (p) , after-
wards a 4 (fi) < a 3 (n) + a2 (p) < 3a 2 (p), a 5 (pi) < a4 (pi) + a2 (pi) < 4a 2 (pt)
and finally an (pi) < (n — 1) a 2 (/L<) .
(v) By
A
| » M U S) - A* (i4) -p(B)\<p(X), VA, B & A,
we get the inequality a2 (p) < // ( X ) , that is (using (iv)) an (pt)

<(n-l)p(X).
(vi) If pi is superadditive, that is
pi (A U B) > p (A) + pt (B) , Wl, B e A, A n 5 = 0,
then we obtain
4
MCM* -i>(* = C U * -&(^)<A*W
<»'=! »'=1 U=l t= l
for every sequence {Ai,...,A„} C ^4,^4j fl Aj = 0 if i 7^ j . Passing to

supremum we get the desired inequality. •
Example 4.1 If pi is additive then an (pi) = 0, for all n > 2.
Example 4.2 If po is the fuzzy measure which models the total ignorance
(see Dubois-Prade [64], p.128), that is pi0 : V (X) -> [0,1], pi0(A) - 0 if
A 7^ X and /io(^4) = 1 if A = X, then an (po) — 1 for all n > 2. Indeed,
because / J 0 ( X ) - ^0 (A) - p0 (X \ A) = 1,VA € V(X),A £ X,A ± 0
we have 02(^0) > 1, that is (see Theorem 4.1, (i)) a n (//0) > l,Vn > 2.

Defects of Properties in Mathematics

Uploaded by

Copyright:

Available Formats

Defects of Properties in Mathematics

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Defects of Properties in Mathematics

Uploaded by

Copyright:

Available Formats

Defects of Properties in

Series Editor: Professor George A. Anastassiou

Vol. 3 Introduction to Matrix Theory

Adrian I Ban & Sorin G Gal

British Library Cataloguing-in-Publication Data

DEFECTS OF PROPERTD2S IN MATHEMATICS

Printed in Singapore by Uto-Print

The m a i n aim is to reveal a m e t h o d of research in m a t h e m a t i c s , called by

Theory, Real Function Theory, Complex Analysis, Functional Analysis, Al-

Oradea, August 1, 2001 The authors

1.5 On Chapter 5: Defect of Property in Real Function Theory . . 24

Chapter 2 Defect of Property in Set Theory 57

Chapter 3 Defect of Property in Topology 87

3.4 Measures of Noncompactness for Fuzzy Subsets in Topological

Chapter 4 Defect of Property in Measure Theory 131

Chapter 5 Defect of Property in Real Function Theory 185

5.4 Bibliographical Remarks and Open Problems 208

Chapter 6 Defect of Property in Functional Analysis 211

Chapter 7 Defect of Property in Algebra 277

Chapter 8 Miscellaneous 305

8.2 Defect of Property in Geometry 307

To convey some of the essence of this monograph to the reader, in this

1.1 General D e s c r i p t i o n of t h e Topic

It is well-known t h a t the definition of a m a t h e m a t i c a l concept consists in

Up = {x 6 U; x has the property P}

Up- = {x £ U; x does not satisfy the property P} .

A powerful tool of study of Up and U-p might be the introduction (not

x G Up if and only if E (x) = Oy.

In this case, for x £ Up- we necessarily have E (x) ^ Oy and consequently

t h a t measure the deviation of a function from the property of continuity,

properties in Fuzzy Mathematics.

1.2 On Chapter 2: Defect of Property in Set Theory

We denote by FS(X) - {A \A : X ->• [0,1]} the class of all fuzzy sets

Definition 2.4 Let (X,A) be a measurable space and t £ (0,1). A t-

(t) If A(x) e {0,1} , Vz G X then d£ (A) = 0;

D e f i n i t i o n 2.5 Let (X,A, /i) be a measure space and t G (0, 1) . A function

D e f i n i t i o n 2.6 Let N be a fuzzy complement. A ^-measure of fuzziness

A family of ^-measures of fuzziness is given by

Theorem 2 . 3 Let X be a finite set, t 6 (0,1) and h : R + - ^ R + increasing,

is a t-measure of fuzziness. If, in addition,

gx(a) = gx (N(a)) ,\/a e [0,1],\/x £ X,

where N is a fuzzy complement, then d^ is a N-symmetrical t-measure of

Under additional conditions, we obtain a family of strict ^-measures of

FS(X) -> R defined by

is a strict t-measure of fuzziness which is N-symmetrical if ^(o) —

Other families of f-measures of fuzziness can be obtained by using the

D e f i n i t i o n 2 . 7 Let t £ (0,1) and (X,A,n) be a measure space. An A-

T h e o r e m 2.5 Let t G ( 0 , 1 ) , (X,A,fi) be a measure space and <pt : X x

ipt {x, N(a)) = p , ( i , o ) , V i e X, Va G [0,1],

then the t-measure of fuzziness dfi(A), defined as above, is N -symmetrical.

An intuitionistic fuzzy set (see e.g. Atanassov [5]) A on X, is an object

HA{x) + vA(x) < 1.

We denote by IFS(X) the family of intuitionistic fuzzy sets on X.

Definition 2.9 A real function df : IA (X) —> R+ is called an intuitionistic

Definition 2.10 Let (X,A, m) be a measure space and let us denote

with A — {(x, PA(X), VA(X)) • x G X}, is an intuitionistic entropy.

The intuitionistic entropies can be characterized by using Iv-functions.

with A — {{x, PA(%),VA(X)) : x G X}, is called /^-function.

intuitionistic entropy and an Iv -function if and only if