1

FUZZY SETS

1.1
BASIC NOTIONS

A classical (crisp) set is defined as a collection of elements

x  X . An element can either belong to or not belong to a set A, A  X .
Such a classical set can be described in different ways: one can enumerate
its elements, one can use the analytically representation (for instance,
A  10 ) or the membership (characteristic) function. The characteristic
function  A of a subset A  X is a mapping
 A : X  0,1 ,
where the value zero is used to represent non-membership and the value
one is used to represent membership. The truth and falsity of the
statement “ X is in A ” is determined by the ordered pair  x ,  A  x  : the
statement is true if the second element of the ordered pair is 1 and the
statement is false if it is 0.
Fuzzy sets were introduced by Zadeh (1965) in order to represent
and manipulate data that was not precise, but rather fuzzy. Similarly with
the crisp case, a fuzzy subset A of a set X is defined as a ordered pairs
with the first element from X and the second element from the interval
0,1 ; the set X is referred to as the universe of discourse for the fuzzy
subset A .
Definition 1.1. (Zadeh, 1965) If X is a nonempty set then a fuzzy set A
in X is defined by its membership function  A : X  0, 1 and  A x 
represents the degree of membership of element x in fuzzy set A ; then A
is represented as A   x ,  A  x  / x  X  .

Example 1.1. We can define the set of natural numbers “close to 1” by

A   2 , 0.25 ,  1, 0.5 , 0 , 0.75 , 1, 1 , 2 , 0.75 , 3, 0.5 , 4 , 0.25
or by

 A x  
1
.
1   x  1
2

If A is a fuzzy set in X then we often use the notation

A   A  x1  / x1   A  x 2  / x 2     A  x n  / x n    A  xi  / xi (discrete
n

i 1

case)
and
A   A  x1  / x1   A  x 2  / x 2   X  A  x  / x (continuous case).

Definition 1.2. A fuzzy subset A of a classical set X is called normal if

there exists an x  X such that  A x   1 ; otherwise A is subnormal.

A nonempty fuzzy set A can always be normalized by dividing  A x  by

sup  A  x  .
x

Definition 1.3. Let A be a fuzzy subset of X ; the support of A , denoted

supp  A , is the crisp subset of X given by
supp  A  x  X /  A  x   0.

Definition 1.4. An  -level set or  -cut of a fuzzy set A of X is a non-

fuzzy set A defined by

10
 x  X /  A  x     if   0
A  
cl supp  A if   0
where cl supp  A is the closure of the support of A ; the 1-level set of
A is named core of A .

Example 1.2. (Fuller, 1995, 1998) For X   2 ,  1, 0, 1, 2 , 3, 4 and

A  0.0 / 2  0.3 /  1  0.6 / 0  1.0 / 1  0.3 / 2  0.0 / 4
we have
 1, 0, 1, 2, 3 if 0    0.3

A  0, 1, 2 if 0.3    0.6
1 if 0.6    1


1.2
OPERATIONS ON FUZZY SETS

The classical set theoretic operations from ordinary set theory can
be extended, in different ways, to fuzzy sets via their membership
functions. The basic operations where suggested by Zadeh (1965) . Let A
and B are fuzzy subsets of a nonempty (crisp) set X .

Definition 1.5. The intersection of A and B is defined as

 A B x   min A x ,  B x , x  X
Definition 1.6. The union of A and B is defined as
 A B x   max A x ,  B x , x  X
Definition 1.7. The complement A of a fuzzy set A is defined as
 A x   1   A  x 

11
Example 1.3. (Zimmermann, 1991) Let A and B be fuzzy subsets of
X  1, 2,  , 10
A  1, 0.2 , 2 , 0.5 , 3, 0.8 , 4 , 1 , 5, 0.7  , 6 , 0.3
B  3, 0.2 , 4, 0.4  , 5, 0.6  , 6 , 0.8 , 7 , 1 , 8, 1.
Then
C  A  B  3, 0.2 , 4 , 0.4  , 5, 0.6  , 6 , 0.3
D  A  B  1, 0.2 , 2 , 0.5 , 3, 0.8 , 4 , 1 , 5, 0.7  , 6 , 0.8 , 7 , 1 , 8, 1
B  1, 1 , 2, 1 , 3, 0.8 , 4 , 0.6  , 5, 0.4  , 6 , 0.2  , 9 , 1 , 10 , 1 .

Example 1.4. (Zimmermann, 1991) We consider the fuzzy sets

A  „ x is considerable larger than 10“
B  „ x is approximately 11“,
given by its membership functions
0 for x  10
 A x   

 1   x  10 2 
1
for x  10
and

 B  x   1  x  114 
1
.
Then

 A B  x   

min 1   x  10 2  , 1  x  11 
1 4 1
 for x  10
0 for x  10


 A B  x   max 1  x  102  , 1  x  11 
1 4 1
 for x  X .

12
 Figure 1.1 Intersection and union of two fuzzy sets

Another type of operations with fuzzy sets are algebraic operations. Some
of these are presented below.

Definition 1.8. Let A1 ,  , An be fuzzy sets in X 1 ,  , X n ; the Cartesian

product is a fuzzy set in the product space X 1    X n with the
membership function
 A  A x   min A xi  / x  x1 ,  , x n , xi  X i .
1 n i
i

Definition 1.9. The m -th power of a fuzzy set A is defined by

 A x    A x  m , x  X .
m

Definition 1.10. The algebraic (probabilistic) sum C  A  B is defined

as C   x , A B  x  / x  X  where
 A B  x    A  x    B  x    A  x    B  x 
.
Definition 1.11. The bounded sum C  A B is defined by
C   x , A B  x  / x  X 
 A B  x   min1,  A x    B x  .
Definition 1.12. The bounded difference C  AB is given by
C  x , AB  x  / x  X 
 AB  x   max0,  A  x    B  x   1 .

Definition 1.13. The algebraic product C  A  B is defined as

C  x , AB  x  / x  X 
 AB  x    A x    B  x  .

Example 1.5 (Zimmermann, 1991) Let A  3, 0.5, 5, 1, 7 , 0.6  and
B  3, 1, 5, 0.6  . Then, according to above definitions, we obtain

13
A  B  3, 3 , 0.5,5, 3 , 1, 7 , 3 , 0.6, 3, 5 , 0.5, 5, 5 , 0.6,7 , 5, 0.6 
A 2  3, 0.25 , 5, 1 , 7 , 0.36
A  B  3, 1 , 5, 1 , 7 , 0.6 
A  B  3, 1 , 5, 1 , 7 , 0.6 
AB  3, 0.5 , 5, 0.6 
A  B  3, 0.5 , 5, 0.6 .

Definition 1.1. The inclusion and equality operations are defined by

A  B  A x   B x  x  X
A  B  A x   B  x  x  X .

Another way to define intersection and union of two fuzzy sets

was addressed by Bellman and Giertz in 1973, interpreting the
intersection as “logical and” and union as “logical or” and using triangular
norms and triangular conorms in order to model logical connectives “and”
and “or”, respectively. Triangular norms and conorms were introduced by
Schweizer and Sklar (1960) to model the distances in probabilistic metric
spaces.

Definition 1.15. A function T : 0, 1 0, 1  0, 1 is a t-norm iff it is

commutative, associative, non-decreasing and T x , 1  x x  0, 1 . A
continuous t-norm T is called Archimedean if T  x, x   x x  0, 1 .

The most important t-norms are:

Minimum: Tm  x , y   minx , y
Lukasiewicz: TL  x , y   maxx  y  1, 0
Probabilistic: T p x , y   xy
minx , y if maxx , y  1
Weak: Tw x , y   
0 otherwise

14
Definition 1.16. A function S : 0, 1 0 , 1  0, 1 is a t-conorm iff it is
commutative, associative, non-decreasing and S x , 0  x x  0, 1 . A
continuous t-conorm S is called Archimedean if S  x , x   x x  0, 1 .

The basic t-conorms are:

Minimum: S m  x , y   maxx , y
Lukasiewicz: S L x , y   minx  y , 1
Probabilistic: S p x , y   x  y  xy
maxx , y if minx , y  0
Strong: S s  x , y   
1 otherwise

The names weak t-norm and strong t-conorm result from the following
inequalities:
Tw x , y   T  x , y   minx , y
maxx , y   S x , y   S s x , y

for every t-norm T and t-conorm S .

In addition, for every t-norm T and t-conorm S ,

T 0 , 0   S 0 , 0  0 , T 1, 1  S 1, 1  1

Definition 1.17. A strong negation is an involutive decreasing function

from [0, 1] into itself.
The relation between a t-norm T and a t-conorm S , via a strong
negation , is given by

Theorem 1.1. (Alsina, Trillas & Valverde, 1980) If T is a t-norm and

C is a strong negation then
 
S  x , y   C T C x  , C y  
is a t-conorm and reciprocally,

 
T  x , y   C S C x  , C y  ;

15
namely, T and S are C-dual.
Ling (1965) proved that any continuous Archimedean t-norm can
be written in the form
T  x, y  f
 1
 f  x  f  y 
where f :  0, 1   0,  is a strict decreasing continuous and f
 1
is the
pseudo-inverse of f defined by

1 if x  0 , f 1
f  1
x    f 1 x  if x   f 1, f 0 
0
 if x   f 0 ,  .

A representation for strong negations was given by Trillas (1979)

in the form

C  x   t 1 t 1  t x 

where t : 0,1  0,   is a continuous and strictly increasing (or

decreasing) function with t 0  0 (or t 0  1 ) and t 1 is a finite
number.
Instead of ''+'' and ''- '' from Ling’s and respectively Trillas’s
formula one can use general operations obtaining t-norms (respectively
negations) on a easier way.

Theorem 1.2. (Iancu, 1997) Let f : 0, 1  I  0,   be a continuous

strictly decreasing function and  : I  I  I with the following
properties:
(2.1)  x, y    y, x 
(2.2)  x,  y, z    x, y , z 
(2.3)  x, y    x, z  if y  z with equality iff y  z
(2.4)  is continuous
(2.5)  x , e  x
16
for all x , y , z  I and e  f 1 . Then
T  x, y   f 1  f  x , f  y  x, y  0, 1
is a t-norm where f ( 1 ) is the pseudo-inverse of f .

Example 1.6. For I  0,  ,  x , y   x  y  xy , f  x   1  x we obtain the

t-norm T  x , y   max2 x  2 y  xy  2, 0 .

Theorem 1.3. (Iancu, 1997) Let I  R and  : I  I  I an application

satisfying the following conditions, for all x , y , z  I :
(3.1) - (3.4) identical to (2.1) - (2.4)
(3.5) there is e  I such that  x , e   x x  I
(3.6) x  I there is x'  I such that  x , x'   e and
 : I  I ,  x   x' is a continuous strictly decreasing function
(3.7) let J  e,    I and t : 0, 1  J be a continuous strictly
increasing function with t 0  e and t 1 is a finite number.
Then C  x   t 1 t 1 , t  x  is a strong negation for x  0, 1 .

Example 1.7. For I  R , x , y   x  y  1, e  1 , J  1,   ,

t x   2 x  1  x  1 and   x   2  x we have C  x   1  x  1  3x  .

Observation 1.1. Simultaneously using of functions  and f

(respectively t ) in previous theorems allows obtaining of t-norms
(respectively negations) on a easier way than in the case    . For
instance, if   x  y we don't work with functions of the type
f ( x )  ax  b in order to obtain the t-norm from Example 1, being
necessary more complicated forms.

If the ordinary t-norms (t-conorms) are used for combining the

information with great (small) belief degrees we obtain , generally, results
that are contrary with the reality (Iancu, 1997). Starting from a given t-
17
norm (t-conorm) we can obtain a new t-norm (t-conorm) which gives
results in accordance with the reality if it is used for combining of
information with great (small) belief degrees. The first result with this
property is given in (Pacholczyk, 1987) , where are presented the
following operators with threshold a  0, 1 :

 1 a
1  a x if x  a
- negation C a  x   
 a 1  x  if x  a
1  a

 a 1 a 1 a 
 T x, y if x  a and y  a
- t-norm Ta  x , y   1  a  a a 
min x , y  x  a or y  a.
 if

corresponding to t-norm T  x , y  ;

S  x , y  if x  a and y  a
- t-conorm S a  x , y   
max x , y  if x  a or y  a.

corresponding to t-conorm S x , y  .

In a set of papers (Iancu, 1997a; 1997b; 1998a; 1999a; 1999b;

2003; 2005) this result were generalized working in a ordered field
I ,  , , I  R and with an arbitrary number n  1 of threshold; in this
way, various classes of operators with threshold are obtained.
We remain in the conditions of Theorem 1.3 and note  x , y   x  y.
We take : I  I  I with the following properties:
i) x  y  x  z iff y  z x , y , z  I and x  e ,
ii) I , ,  is a field.

18
 1
We note x and the inverse element of x corresponding to  and 
x
1 x
respectively. For the simplification of writing we note x   and
y y
x  x  x2.

Theorem 1.4. (Iancu, 2003; 2005) If , ,  and t have the previous

meaning and n  N , n  1 , 0  a1  a2    a n  1 ,
t a n i t a n i 1  t a   t a n i 1 t ai   t a n i 
 i   and  i   i 1 then
t ai 1 t a i  t ai 1 t ai 

 1  t 1t a n  
t  t 1  t  x  if x  a1
  t a1  
 1
C a1 ,an  x   t t x    i    i  if ai  x  ai 1 and 1  i  n

t 1  t 1t  x   t a1   if x  a
  t 1t a n  
n


 1  t a1   t a n   t 1  t  x  
t   if x  a1
  t 1  t a1  t a n   t  x   t a1   t a 
n 
 1
C a1 ,an  x   t t  x    i    i  if ai  x  ai 1 and 1  i  n
*


t 1  t a1   t a n   t 1t  x  
  if x  a n
  t 1  t a1 t a n   t  x t a1   t a n  

and
 1  t a1   t a n   t 1 
t   if x  a1
  t a1   t a n   t 1t a n   t  x  

C a*1*,an  x   t 1 t x    i    i  if a i  x  a i 1 and 1  i  n

t 1  t a1   t a n   t 1t  x   if x  a
  t 1t a n   t x   n


19
     
 t a   t a 
   n     n 1   
are strong negations and they have t 1  
2    2    as fixed
 2 
 
 
 
point, where [ x] is the greatest integer which is smaller than or equal to
x.

Observation 1.2. It is easy to verify the folloving properties:

i  x  a1  N a ,....,a x   a n
1 n

ii  x  a n  N a ,....,a x   a1
1 n

ii  x  ai , ai 1   N a ,...,a x   a ni , a ni 1 

1 n
i  1, 2 , , n  1

 
where N a1 ,, an  C a1 ,, an , C * a1 ,, an , C ** a1 ,, an .

Observation 1.3. The relation (ii) says that if the confidence in a

proposition p is greater than or equal to the threshold a n then the
confidence in non p is smaller than or equal to the threshold a1 . This
observation can be used for handling the confidences associated with the
information from a knowledge base.
Example 1.8. (Iancu, 2003; 2005) For    ,    , n  2 and

t x  
2x
one obtain
x 1

 2a1 a 2 x  a 2 x  x  a1 a 2  a1
 if x  a1
 2 a1 x  a 2 x  x  a1 a 2  a1
 a a  1x  a1  a 2  2a1 a 2
C a1 ,a2 x    1 2 if a1  x  a 2
 xa1  a 2  2  1  a1 a 2
 1  x a1 1  a 2 
 if x  a 2
 x1  2a1  a 2   1  2a1 a 2  a 2

20
  a1 a 2 1  3x 
 if x  a1
 x1  3a1  a 2   a1 a 2
 a a  1x  a1  a 2  2a1 a 2
C a*1 ,a2 x    1 2 if a1  x  a 2
 x a 1  a 2  2   1  a1 a 2
 1  x a1a 2
 if x  a 2
 x1  3a1  a 2   3a1 a 2
and

 a1 a 2 1  x 
 if x  a1
 x1  a1  a 2   a1 a 2
 a a  1x  a1  a 2  2a1 a 2
C a*1*,a2 x    1 2 if a1  x  a 2
 x a 1  a 2  1  1  a1 a 2
 1  x a1 a2
 if x  a 2
 x1  a1  a 2   a1 a 2

Theorem 1.5. (Iancu, 2005) Let S be a t-conorm and S ' S M , S . For

0  a1  a 2    a n  1

max( x , y ) if x  a1 or y  a1

S a1 ,,an ;S '  x , y   S ( x , y ) if x  a n and y  a n
 S' ( x , y )
 otherwise
is a t-conorm.

Theorem. 1.6. (Iancu, 2005) Let T , S  be a pair (t-norm, t-conorm) dual

with respect to C  x   t 1 t 1t  x  , S ' S M , S  and T ' is the dual of S '
with respect to the same negation, i.e. T ' TM , T  . We use the following
notations
t a1 
k ,
t 1t a n 
  t (1)  t a1 t a n  ,

21
 1 
 z   t 1   t ( z )  ,
k 
 t z   t 1  t a1   t 1t a n  
 z   t 1   and
 t 1  t a1  t a n   t  z   t a1   t a 
n 

 z , i   t 1 t 1t z    i  i  ,
where  i  and  i  have the same signification from the last theorem.
For 0  a1  a 2    a n  1 we define

t 1 k  t T   x ,  y  if x  a1 and y  a1



t 1 k  t T '   x ,   y ,i 

 if x  a1 and y  ai , ai 1 , 1  i  n  1


t 1 k  t T '   x ,i ,  y 

 if y  a1 and x  ai , ai 1 , 1  i  n  1


Ta1 ,,an ;T '  x , y   t 1 t 1t T '   x ,i ,   y , j    k    k 



if x  ai , ai 1 , y  a j , a j 1 , l  maxi , j  and
 there is an int eger k  n  l , n  1 such that

 T '   x ,i ,   y , j   C a k 1 , C a k 

 1
t k  t T '   x ,i ,   y , j 

 
if x  ai , ai 1 , y  a j , a j 1 , T '   x ,i ,   y , j   C a n 


min( x , y ) if x  a n or y  a n

and

22
  1  t a1   t a n   t T  x ,  y  
t  
    t 1t T  x ,  y t a1   t a n  
 if x  a1 and y  a1

 1  t a1   t a n   t T '   x ,   y ,i  
t  
    t 1 t T '   x ,   y , j   t a 1   t a n  
 if x  a1 and y  ai , ai 1 , 1  i  n  1

 1  t a1   t a n   t T '   x ,i ,  y  
t  
    t 1 t T '   x , i ,  y    t a 1   t a n 
 if x  ai , ai 1 , 1  i  n  1 and y  a1
Ta*1 ,an ;T ' ( x , y )   1
t t 1t T '   x ,i ,   y , j    k    k 



if x  ai , ai 1 , y  a j , a j 1 , l  maxi , j  and
 there is an int eger k  n  l , n  1 such that

 T '   x ,i ,   y , j   C a k 1 , C a k 
  t a1   t a n   t T '   x ,i ,   y , j  
t 1  
    t 1t T '   x ,i ,   y , j t a1   t a n  


 if x  ai , ai 1 , y  a j , a j 1 and 
 T '   x ,i ,   y , j   C a n 

min x , y  if x  an or y  a n

Then
(i) Ta1 ,,an ;T ' is a t-norm C a1 ,,an - dual with t-conorm S a1 ,,an ;S'

(ii) Ta*1 ,,an ;T ' is a t-norm C a*1 ,,an - dual with t-conorm S a1 ,,an ;S' .

Example 1.9 (Iancu, 2005) By particularization, one can be obtained

new extension of another known t-operators. For instance, for n  2 ,
a  a1 , b  a 2 ,    ,    , t  x   x and T '  TM  min in Theorem
1.6 it results

23
  a 1 b 1 b 
 T x, y if x  a and yb
Ta / b  x , y   1  b  a a 
min x , y 
 otherwise

that is an extension of Pacholczyk’s (1987) t-norm and

 abT   x ,   y 
 if x  a and yb
T *
a/b x , y    1  a  b 1  T  x ,   y   ab
min x , y 
 otherwise

where   z  
1  a 1  b z , which is a t-norm with 1-threshold a and
1  a  b z  ab
parameter b .

Theorem. 1.7. (Iancu, 2003) Let T , S  be a pair (t-norm, t-conorm) dual

with respect to C  x   t 1 t 1t  x  , S' S M , S  and T ' is the dual of S '
with respect to the same negation, i.e. T ' TM , T  .
We use the following notations
t a1   t a n 
k ,
t 1t a n 

 t 1  t a1   t z   t 1 

  z   t 1   and
 t a1   t a n   t 1t a n   t  z  
 z , i   t 1 t 1t z    i  i  ,
where  i  and  i  have the same signification from the Theorem 1.4.
For 0  a1  a 2    a n  1 we define

24
  1  t T  x ,  y  
t  k  
  t 1t T  x ,  y  
 if x  a1 and y  a1

 1  t T '  x ,   y ,i  
t  k  
  t 1  t T '   x ,   y , i  
 if x  a1 and y  ai , ai 1 , 1  i  n  1

 1  t T '   x ,i ,  y  
t  k  t 1t T '   x ,i ,  y  
  
 if x  ai , ai 1 , 1  i  n  1 and y  a1
Ta*1*,an ;T ' ( x , y )   1
t t 1t T '   x ,i ,   y , j    k    k 



if x  ai , ai 1 , y  a j , a j 1 , l  maxi , j  and
 there is an int eger k  n  l , n  1 such that

 T '   x ,i ,   y , j   C a k 1 , C a k 
  t T '   x ,i ,   y , j  
t 1  k  
  t 1t T '   x ,i ,   y , j  

 
if x  ai , ai 1 , y  a j , a j 1 and
 T '   x ,i ,   y , j   C a n 

min x , y  if x  a n or y  a n

Then Ta*1*,,an ;T ' is a t-norm C a*1*,,an - dual with t-conorm S a1 ,,an ;S' .

Example 1.10. (Iancu, 2003) For n  2 , a  a1 , b  a 2 ,    ,

  , t  x   x and T '  TM  min in Theorem 1.7 we obtain a new
parametric extension for t-norms with 1-threshold

 ab
 T   x ,   y  if x  a and y  b
Ta / b  x , y   1  b
min x , y  otherwise

where   z  
1  b z , a is the threshold and b is the parameter.
1  b z  ab

25
1.3
FUZZY NUMBERS

An important concept in fuzzy set theory is called extension

principle and it is used to extend any point operations to operations
involving fuzzy sets.

Definition 1.18. Let X be the cartesian product X  X 1  X 2    X n

and A1 ,  , An be fuzzy sets in X 1 ,  , X n , respectively, f is a mapping
from X to a universe Y , y  f x1 ,  , x n  . Then the extension principle
allows us to define a fuzzy set B in Y by

B   y ,  B  y  / y  f  x1 ,  , x r ,  x1 ,  , x r   X 

where


 x , ,sup  
min  A1  x1 ,  ,  An  x n  if f 1  y   
 B  y    1 xn  f  y 
1

0 otherwise

For r  1 the extension principle reduces to

B  f  A   y ,  B  y  / y  f  x , x  X  where

 sup  A  x  if f 1  y   

 B  y    x f 1  y 
0 otherwise

Example 1.8. For A   1, .5, 0 , .8, 1, 1, 2 , . 4  and f ( x )  x 2 it

results B  0 , .8, 1, 1, 4 , .4 

Definition 1.19. A fuzzy number A is a fuzzy set of the real line with a
normal, fuzzy convex and continuous membership function of bounded
support. The set of fuzzy numbers will be notated by F (R).
26
Definition 1.20. A quasi fuzzy number A is a fuzzy set of the real line
with a normal, fuzzy convex and continuous membership function
satisfying the limit conditions lim  A t   0, lim  A t   0 .
t  t  

Definition 1.21. A fuzzy number A is called positive (negative) if its

membership function satisfies  A  x   0, x  0 x  0  .

Definition 1.22. A binary operation  is called increasing (decreasing) if

x1  y1 and x2  y 2 . Then x1  x 2  y1  y 2 x1  x2  y1  y 2  .

If the normal algebraic operations  ,  , , : are extended to

operations on fuzzy numbers they shall be denoted by  ~,  ~ and ~: ,
~, 
respectively.

Theorem 1.4. (Dubois & Prade, 1980). If A and B are fuzzy numbers
with continuous and surjective membership functions from R to 0, 1 and
 is a continuous increasing (decreasing) binary operation, then A ~ B is
a fuzzy number whose membership function is continuous and surjective
from R to 0, 1 .

The membership function of A~ B can be determined from the

membership functions of A and B , according with the

Theorem 1.5. (Dubois & Prade, 1980). If A, B  F R  with continuous

membership functions then the extension principle for the binary function
: RR R gives the fuzzy number A~
 B:
 A~ B z   sup min A x ,  B  y  .
z  x y

Example 1.9. Let A  1, 0.3 , 2 , 1 , 3, 0.4  and

B  2 , 0.7  , 3, 1 , 4 , 0.2  be fuzzy numbers. Then
A~  B  2, 0.3 , 3, 0.3, 4 , 0.7  , 6, 1 , 8, 0.2  , 9 , 0.4  , 12 , 0.2 .

27
In applications one work with triangular or trapezoidal fuzzy numbers,
given by the following definitions.

Definition 1.23. A fuzzy set A is called triangular fuzzy number with

center a , left width   0 and right width   0 , denoted as
A  a , ,  , if its membership function is
1  a  x   if x  a   , a 

 A x   1   x  a   if x  a , a   
0
 otherwise
A triangular fuzzy number with center a is seen as a fuzzy quantity “ x is
approximately equal to a ”.

Definition 1.24. A fuzzy set A is called triangular fuzzy number with

tolerance interval a , b , left width   0 and right width   0 , denoted
as A  a , b ,  ,   , if its membership function is
1  a  x   if x  a   , a 
1 if x  a , b

 A x   
1   x  b   if x  b , b   
0 otherwise
A trapezoidal fuzzy number with tolerance interval a , b is seen as a
fuzzy quantity “ x is approximately in the interval a , b ”.

Computational efficiency is of particular importance when fuzzy

set are used to solve real problems. The LR-representation of fuzzy
numbers, suggested by Dubois and Prade (1979), increases computational
efficiency, without limiting the generality.

Definition 1.26. Any fuzzy number A can be represented as

28
  La  x    if x  a   ,  
1 if x  a , b

 A x   
 R x  b    if x  b , b   
0 otherwise
where a , b is the core of A , L , R : 0 , 1  0 , 1 are continuous and non-
increasing shape functions with L0   R0   1 and L1  R1  0 ; a
such representation is referred as A  a , b ,  , LR .

Definition 1.27. Any quasi fuzzy number A can be represented as

 La  x    if x  a
1 if x  a , b

 A x   
 R x  b    if x  b
0 otherwise
where a , b is the core of A , L , R : 0 ,   0 , 1 are continuous and
non-increasing shape functions with L0   R0   1 and
lim Lx   lim Rx   0 ; a such representation is referred as
x  x 

A  a , b ,  , LR .

For a  b one use the notation A  a ,  , LR and A is named quasi-

triangular fuzzy number. If Lx   R x   1  x then instead
A  a , b ,  , LR is used the notation A  a , b ,  ,   .

Example 1.10. (Zimmerman, 1991). Let Lx   , Rx  

1 1
,
1 x 2
1 2 x
  2,   3, a  5. Then A  5, 2 , 3LR is given by

29
  5 x 1
 L 2   if x 5
   5 x
2

1  
  2 
 A x   
 R x  5   1
x 5
  if
  3  2 x  5 
 1
 3

The operations with LR fuzzy numbers are considerably simplified:

Dubois and Prade gave exact formulas for ~  and approximate
 and ~
~ ~
expressions for  and : .

Theorem 1.6. (Dubois & Prade, 1980) Let LR fuzzy numbers

A  a ,  , LR and B  b ,  ,  LR . Then
1) a ,  ,  ~
LR  b ,  ,    a  b ,    ,    
LR LR

2)  a ,  , LR   b , ,  LR
3) a ,  , LR ~
 b ,  ,  LR  a  b ,    ,    LR .

L x   R x   A  1, 0.5, 0.8LR

1
Example 1.11. For , and
1 x2
B  2, 0.6, 0.2LR it results  B  3, 1.1, 1LR
A~ and
 B   1, 0.7 , 1.4 LR .
A~

Theorem 1.7. (Dubois & Prade, 1980) If A and B are LR fuzzy

numbers, then
a ,  , LR ~ b ,  , LR  ab , a  b , a  bLR if A and B are
positive;
a ,  , LR ~ b ,  , LR  ab , b  a, b  a LR if A  0 and B  0 ;
a ,  , LR ~ b ,  ,  LR  ab ,  b  a,  b  a LR if A and B are
negative.

30
Example 1.12. (Zimmerman, 1991). For A  2, 0.2, 0.1LR ,
B  3, 0.1, 0.3LR and

1 if -1  x  1
Lx   Rx   
0 otherwise
we obtain
 2 x
 L 0.2  if x2
   1 if 1.8  x  2.2
 A x    = 
 R x  2  if x2 0 otherwise
  0.1 

 3 x
 L 0.1  if x3
   1 if 2.7  x  3.3
 B x    = 
 R x  3  if x3 0 otherwise
  0.3 
therefore A and B are positive numbers.
According to the last theorem, it results
A~ B  2  3, 2  0.1  3  0.2 , 2  0.3  3  0.1  6 , 0.8, 0.9LR .
LR

1.4
FUZZY RELATIONS

Fuzzy relations are natural extensions of classical relations and

they are important because can describe interactions between variables.
We consider only binary relations, the extension to n -any relations being
straight forward.

31
Definition 1.28. Let X and Y be nonempty sets. A fuzzy relation R is a
fuzzy subset of X Y : R   x , y ,  R  x , y  / x , y   X  Y  where
 R  x , y  is the degree of membership of x , y  in R .

Example 1.13. Fie X  Y  R and R = „considerably larger than“. This

relation can be defined by
0 if x y
 R~ x , y   

 1   y  x  2 
1
if x  y.
The operations with fuzzy relations can be defined by analogy with the
operations of fuzzy sets.

Definition 1.29. Let R and S be two fuzzy relations in the same product
space; the their union/intersection is defined by

 R  S x , y   max R x , y  ,  S x , y ,
 R  S  x , y   min R x , y  ,  S  x , y , x ,y   X  Y

One can be use any t-norm T and t-conorm S instead of min and
max operations, respectively. Some other operations, such as the
projection and the cylindrical extension of fuzzy relations can be useful.

Definition 1.30. Let R be a fuzzy relation on X  Y . The projection of R

on X is a fuzzy subset of X , defined by

  
 X R    x , sup  R  x , y  /  x , y   X  Y 
 y  
The total projection is given by

 T R   sup sup R  x , y  /  x , y   X  Y .
x y

Definition 1.31. The largest relation in X the projection of which is R is

called the cylindrical extension of R .
32
Example 1.14. An example of fuzzy relation and its projections is

 y1 y2 y3 y4 y5 y6 X 
 
 x1 0.1 0.2 0.3 0.6 1 0.2 1 
 
 x2 0.3 0.5 0.4 1 0.5 0.6 1 
x 0.4 0.8 1 0.7 0.8 0.8 1 
 3 
 Y 0.4 0.8 1 1 1 0.8 1 
 1 
 T
The cylindrical extension of  Y is

 y1 y2 y3 y4 y5 
y6
 
 x1 0.4 0.8 1 1 1 0.8 
 
 x2 0.4 0.8 1 1 1 0.8 
 x 0.4 0.8 
 3 0.8 1 1 1

Fuzzy relations in different product spaces can be combined by the

operation “composition”. Different versions of “composition” have been
suggested but the sup-min composition is the best known and the most
frequently used.

Definition 1.32. Let R and S be two fuzzy relations on X  Y and

Y  Z , respectively. Their sup-min composition is defined as

 
R  S   x , z , sup minRx , y , S  y , z  / x  X , y  Y , z  Z 
 yY  

Example 1.15. (Fullér, 1995). The sup-min composition of the relations

33
  z1 z2 z3 
 y1 y 2 y3 y 4   
   y1 0.4 0.9 0.3 
 x1 0.5 0.1 0.1 0.7   
R  and S   y 2 0 0.4 0 
 x 2 0 0.8 0 0  y 0.9 0.5 0.8 
 x 0.9 1 0.7 0.8   3 
 3   y 4 0.6 0.7 0.5 
 

 z1 z2 z3 
 
 x1 0.6 0.8 0.5 
is RS  
 x2 0 0.4 0 
 x 0.7 0.9 0.7 
 3 

Changing the operation min from the last definition with a t-norm
T one obtain the sup  T composition. Following ( ? 1973) one can
define the sup-min composition of a fuzzy set and fuzzy relations as
follows

Definition 1.33. Let T be a t-norm. The membership function of the

composition of a fuzzy set A in X and a fuzzy relation R in X  Y is
defined by
 A R  y   sup T  A x ,  R x , y , y  Y
x X

34