Fuzzy Sets and Systems 122 (2001) 315–326

www.elsevier.com/locate/fss

On possibilistic mean value and variance of fuzzy numbers 

Christer Carlsson ∗ , Robert Full'er 1
 Akademi University, Lemminkainengatan 14B, FIN-20520 Abo,
IAMSR, Abo  Finland
Received 22 September 1999; received in revised form 25 January 2000; accepted 13 March 2000

Abstract
Dubois and Prade introduced the mean value of a fuzzy number as a closed interval bounded by the expectations
calculated from its upper and lower distribution functions. In this paper introducing the notations of lower possibilistic
and upper possibilistic mean values we de1ne the interval-valued possibilistic mean and investigate its relationship to the
interval-valued probabilistic mean. We also introduce the notation of crisp possibilistic mean value and crisp possibilistic
variance of continuous possibility distributions, which are consistent with the extension principle. We also show that the
variance of linear combination of fuzzy numbers can be computed in a similar manner as in probability theory.
c 2001
Elsevier Science B.V. All rights reserved.

Keywords: Fuzzy numbers; Possibility theory

1. Introduction

In 1987 Dubois and Prade [2] de1ned an interval-valued expectation of fuzzy numbers, viewing them as
consonant random sets. They also showed that this expectation remains additive in the sense of addition of
fuzzy numbers.
In this paper introducing the notations of lower possibilistic and upper possibilistic mean values we de1ne
the interval-valued possibilistic mean, crisp possibilistic mean value and crisp ( possibilistic) variance of a
continuous possibility distribution, which are consistent with the extension principle and with the well-known
de1nitions of expectation and variance in probability theory. The theory developed in this paper is fully
motivated by the principles introduced in [2] and by the possibilistic interpretation of the ordering introduced
in [3].
 Supported by the E-MS Bullwhip project TEKES 40965=98.
∗ Corresponding author. Fax: + 358-2-2154809.
E-mail address: christer.carlsson@abo.1 (C. Carlsson).
1 On leave from Department of Operations Research, ED otvDos Lor'and University, Kecskem'eti utca 10-12, H-1053 Budapest, Hungary.
Partially supported by FKFP-0157/2000 and OTKA T032412.

0165-0114/01/$ - see front matter  c 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 5 - 0 1 1 4 ( 0 0 ) 0 0 0 4 3 - 9
316 C. Carlsson, R. Full-er / Fuzzy Sets and Systems 122 (2001) 315–326

A fuzzy number A is a fuzzy set of the real line R with a normal, fuzzy convex and continuous membership
function of bounded support. The family of fuzzy numbers will be denoted by F. A -level set of a fuzzy
number A is de1ned by [A] = {t ∈ R | A(t)¿} if ¿0 and [A] = cl{t ∈ R | A(t)¿0} (the closure of the
support of A) if  = 0. It is well known that if A is a fuzzy number then [A] is a compact subset of R for
all  ∈ [0; 1].
Let A and B ∈ F be fuzzy numbers with [A] = [a1 (); a2 ()] and [B] = [b1 (); b2 ()],  ∈ [0; 1]. In 1986
Goetschel and Voxman introduced a method for ranking fuzzy numbers as [3]

1
1
A6B ⇔ (a1 () + a2 ()) d6 (b1 () + b2 ()) d: (1)
0 0

As was pointed out by Goetschel and Voxman this de1nition of ordering given in (1) was motivated in part
by the desire to give less importance to the lower levels of fuzzy numbers.

2. Possibilistic mean value of fuzzy numbers

We explain now the way of thinking that has led us to the introduction of notations of lower and upper
possibilitistic mean values. First, we note that from the equality

1
1
(a1 () + a2 ())=2 d
ML (A) := (a1 () + a2 ()) d = 0
1 ; (2)
0
0
 d

it follows that ML (A) is nothing else but the level-weighted average of the arithmetic means of all -level sets,
that is, the weight of the arithmetic mean of a1 () and a2 () is just .
Second, we can rewrite ML (A) as

1
1 1
2
a1 () d + 2 0 a2 () d
ML (A) = (a1 () + a2 ()) d = 0
0 2
 1 1   1 1 
1 0
a1 () d 0
a2 () d 1 0
a1 () d 0
a2 () d
= 1
+ 1
= 1 + 1 :
2 2 2
2  d  d
0 0

Third, let us take a closer look at the right-hand side of the equation for ML (A). The 1rst quantity, denoted
by M∗ (A) can be reformulated as

1
1
a1 () d 0
M∗ (A) = 2 a1 () d =
1
0
0
 d
1 1
0
Pos[A6a1 ()]a1 () d Pos[A6a1 ()] × min[A] d
= 1 = 0 1 ;
0
Pos[A6a1 ()] d 0
Pos[A6a1 ()] d

where Pos denotes possibility, i.e.

Pos[A6a1 ()] = ((− ∞; a1 ()]) = sup A(u) = :

u6a1 ()

(since A is continuous!) So M∗ (A) is nothing else but the lower possibility-weighted average of the minima
of the -sets, and it is why we call it the lower possibilistic mean value of A.
 C. Carlsson, R. Full-er / Fuzzy Sets and Systems 122 (2001) 315–326 317

In a similar manner, we introduce M ∗ (A), the upper possibilistic mean value of A, as

1 1
∗ a2 () d
M (A) = 2 a2 () d = 0  1
0
0
 d
1 1
0
Pos[A¿a2 ()]a2 () d Pos[A¿a2 ()] × max[A] d
= 1 = 0 1 ;
0
Pos[A¿a2 ()] d 0
Pos[A6a2 ()] d
where we have used the equality

Pos[A¿a2 ()] = ([a2 (; ∞))) = sup A(u) = :

u¿a2 ()

Let us introduce the notation

M (A) = [M∗ (A); M ∗ (A)]:

that is, M (A) is a closed interval bounded by the lower and upper possibilistic mean values of A.

Denition 2.1. We call M (A) the interval-valued possibilistic mean of A.

If A is the characteristic function of the crisp interval [a; b] then M ((a; b; 0; 0)) = [a; b], that is, an interval
is the possibilistic mean value of itself. We will now show that M is a linear function on F in the sense of
the extension principle.

Theorem 2.1. Let A and B be two non-interactive fuzzy numbers and let  ∈ R be a real number. Then

M (A + B) = M (A) + M (B); M (A) = M (A);

i.e.

M∗ (A + B) = M∗ (A) + M∗ (B); M ∗ (A + B) = M ∗ (A) + M ∗ (B);

and

[M∗ (A); M ∗ (A)] if ¿0;
[M∗ (A); M ∗ (A)] =
[M ∗ (A); M∗ (A)] if  ¡ 0;
where the addition and multiplication by a scalar of fuzzy numbers is de8ned by the sup-min extension
principle [4].

Proof. Really, from the equation

[A + B] = [a1 () + b1 (); a2 () + b2 ()];

we have

1
M∗ (A + B) = 2 (a1 () + b1 ()) d
0

1
1
=2 a1 () d + 2 b1 () d
0 0
= M∗ (A) + M∗ (B)
318 C. Carlsson, R. Full-er / Fuzzy Sets and Systems 122 (2001) 315–326

and

1
∗
M (A + B) = 2 (a2 () + b2 ()) d
0

1
1
=2 a2 () d + 2 b2 () d
0 0
= M ∗ (A) + M ∗ (B);
furthermore, from

  [a1 (); a2 ()] if ¿0;
[A] = [A] = [a1 (); a2 ()] =
[a2 (); a1 ()] if  ¡ 0
for ¿0 we get

1
1
M∗ (A) = 2 (a1 ()) d = 2 a1 () d = M∗ (A);
0 0

1
1
∗
M (A) = 2 (a2 ()) d = 2 a2 () d = M ∗ (A)
0 0

and for ¡0

1
1
M∗ (A) = 2 (a2 ()) d = 2 a2 () d = M ∗ (A);
0 0

1
1
M ∗ (A) = 2 (a1 ()) d = 2 a1 () d = M∗ (A);
0 0

which ends the proof.

We introduce the crisp possibilistic mean value of A by (2) as the arithmetic mean of its lower possibilistic
and upper possibilistic mean values, i.e.
M∗ (A) + M ∗ (A)
ML (A) = :
2
The following theorem shows two very important properties of ML : F → R.

Theorem 2.2. Let [A] = [a1 (); a2 ()] and [B] = [b1 (); b2 ()] be fuzzy numbers and let  ∈ R be a real
number. Then
ML (A + B) = ML (A) + ML (B)
and
ML (A) = ML (A):

Proof. First we 1nd

1
ML (A + B) = (a1 () + b1 () + a2 () + b2 ()) d
0

1
1
= (a1 () + a2 ()) d + (b1 () + b2 ()) d = ML (A) + ML (B)
0 0
 C. Carlsson, R. Full-er / Fuzzy Sets and Systems 122 (2001) 315–326 319

and for ¿0 we get

1
1
ML (A) = (a1 () + a2 ()) d =  (a1 () + a2 ()) d = ML (A)
0 0

and, 1nally, for ¡0 we have

1
1
ML (A) = (a2 () + a1 ()) d =  (a1 () + a2 ()) d = ML (A);
0 0

which ends the proof.

Example 2.1. Let A = (a; ; ) be a triangular fuzzy number with center a, left-width ¿0 and right-width
¿0 then a -level of A is computed by

[A] = [a − (1 − ); a + (1 − )]; ∀ ∈ [0; 1];

that is,

1

M∗ (A) = 2 [a − (1 − )] d = a − ;
0 3

1

M ∗ (A) = 2 [a + (1 − )] d = a +
0 3

and therefore,
 
 
M (A) = a − ; a +
3 3

and, 1nally,

1
−
ML (A) = [a − (1 − ) + a + (1 − )] d = a + :
0 6

Specially, when A = (a; ) is a symmetric triangular fuzzy number we get ML (A) = a. If A is a symmetric
fuzzy number with peak [q− ; q+ ] then the equation

q− + q+
ML (A) =
2
always holds.

3. Relation to upper and lower probability mean values

We show now an important relationship between the interval-valued expectation E(A) = [E∗ (A); E ∗ (A)]
introduced in [2] and the interval-valued possibilistic mean M (A) = [M∗ (A); M ∗ (A)] for LR-fuzzy numbers
with strictly decreasing shape functions.
320 C. Carlsson, R. Full-er / Fuzzy Sets and Systems 122 (2001) 315–326

An LR-type fuzzy number A ∈ F can be described with the following membership function [1]:
 

 q− − u

 L if q− − 6u6q− ;

 



1 if u ∈ [q− ; q+ ];
A(u) = 

 u − q+

 R if q+ 6u6q+ + ;

 




0 otherwise;
where [q− ; q+ ] is the peak of A; q− and q+ are the lower and upper modal values; L; R : [0; 1] → [0; 1],
with L(0) = R(0) = 1 and L(1) = R(1) = 0 are non-increasing, continuous mappings. We shall use the notation
A = (q− ; q+ ; ; )LR . The closure of the support of A is exactly [q− − ; q+ + ].
If L and R are strictly decreasing functions then we can easily compute the -level sets of A. That is,
[A] = [q− − L−1 (); q+ + R−1 ()];  ∈ [0; 1]:
Following [2, p. 293] the lower and upper probability mean values of A ∈ F are computed by

1
E∗ (A) = q− −  L(u) du;
0

1
E ∗ (A) = q+ +  R(u) du
0

(note that the support of A is bounded) and the lower and upper possibilistic mean values are obtained as

1
1
−1
M∗ (A) = 2 (q− − L ()) d = q− −  L−1 () d;
0 0

1
1
M ∗ (A) = 2 (q+ + R−1 ()) d = q+ +  R−1 () d:
0 0

Therefore, we can state the following lemma.

Lemma 3.1. If A ∈ F is a fuzzy number of LR-type with strictly decreasing (and continuous) shape functions
then its interval-valued possibilistic mean is a proper subset of its interval-valued probabilistic mean;
M (A) ⊂ E(A):

Proof. From the relationships

1
1
1
1
L(u) du = L−1 () d and R(u) du = R−1 () d;
0 0 0 0
we get

1
1
1
1
L−1 () d ¡ L(u) du and R−1 () d ¡ R(u) du;
0 0 0 0

which ends the proof.

Lemma 3.1 reNects on the fact that points with small membership degrees are considered to be less important
in the de1nition of lower and upper possibilistic mean values than in the de1nition of probabilistic ones.
 C. Carlsson, R. Full-er / Fuzzy Sets and Systems 122 (2001) 315–326 321

In the limit case, when A = (q− ; q+ ; 0; 0), the possibilistic and probabilistic mean values are equal, and the
equality

E(A) = M (A) = [q− ; q+ ]

holds.

Example 3.1. Let A = (a; ; ) be a triangular fuzzy number with center a; left-width ¿0 and right-width
¿0 then
   
   
M (A) = a − ; a + ⊂ E(A) = a − ; a +
3 3 2 2
and
− E∗ (A) + E ∗ (A) −
ML (A) = a + L
= E(A) = =a+ :
6 2 4
However, when A is a symmetric fuzzy number then the equation

ML (A) = E(A):
L

always holds.

4. Variance of fuzzy numbers

We introduce the (possibilistic) variance of A ∈ F as

1  2 
a1 () + a2 ()
Var(A) = Pos[A6a1 ()] − a1 () d
0 2

1  2 
a1 () + a2 ()
+ Pos[A¿a2 ()] − a2 () d
0 2

1  2  2 
a1 () + a2 () a1 () + a2 ()
=  − a1 () + − a2 () d
0 2 2


1 1
= (a2 () − a1 ())2 d:
2 0
The variance of A is de1ned as the expected value of the squared deviations between the arithmetic mean and
the endpoints of its level sets, i.e. the lower possibility-weighted average of the squared distance between the
left-hand endpoint and the arithmetic mean of the endpoints of its level sets plus the upper possibility-weighted
average of the squared distance between the right-hand endpoint and the arithmetic mean of the endpoints of
its level sets.

Remark 4.1. From a probabilistic viewpoint, the possibilistic mean and variance of a fuzzy number A can
be viewed as the expected values, when the level gamma is treated as a random variable having a beta
distribution Beta(2, 1) (for which the higher the level, the higher its weight or density), of the conditional
mean value and variance given gamma, respectively, of the random variable taking on values a1 () and a2 ()
322 C. Carlsson, R. Full-er / Fuzzy Sets and Systems 122 (2001) 315–326

with probabilities 0:5. This could be an additional argument for the “coherence” between the de1nitions of
E(A) and Var(A) introduced in this paper.

The standard deviation of A is de1ned by


A = Var(A):

For example, if A = (a; ; ) is a triangular fuzzy number then

1
1 ( + )2
Var(A) = (a + (1 − ) − (a − (1 − )))2 d = ;
2 0 24
especially, if A = (a; ) is a symmetric triangular fuzzy number then

2
Var(A) = :
6
In the limit case, when A = (a; 0) is a fuzzy point (i.e. a1 () = a2 () = a = const:; ∀) we get Var(A) = 0.
If A is the characteristic function of the crisp interval [a; b] then

1  2
1 b−a
Var(A) = (b − a)2 d = ;
2 0 2
that is,
b−a a+b
A = ; ML (A) = :
2 2
In probability theory, the corresponding result is: if the two possible outcomes of a probabilistic variable have
equal probabilities then the expected value is their arithmetic mean and the standard deviation is the half of
their distance.
We show now that the variance of a fuzzy number is invariant to shifting. Let A ∈ F and let  be a
real number. If A is shifted by value  then we get a fuzzy number, denoted by B, satisfying the property
B(x) = A(x − ) for all x ∈ R. Then from the relationship

[B] = [a1 () + ; a2 () + ];

we 1nd

1
1
1 1 2
Var(B) = ((a2 () + ) − (a1 () + )) d = (a2 () − a1 ())2 d = Var(A):
2 0 2 0

The covariance between fuzzy numbers A and B is de1ned as

1
1
Cov(A; B) = (a2 () − a1 ())(b2 () − b1 ()) d:
2 0

Taking into consideration that supp(A)=[a1 (0); a2 (0)] and supp(B) = [b1 (0); b2 (0)] we 1nd that the covariance
measures how much the products of -levels sets of two fuzzy numbers A and B are close to the product of
-levels sets of the universal fuzzy sets in supp(A) and in supp(B). From

(a2 (0) − a1 (0))(b2 (0) − b1 (0)) ¿ (a2 () − a1 ())(b2 () − b1 ()) ¿ 0;
 C. Carlsson, R. Full-er / Fuzzy Sets and Systems 122 (2001) 315–326 323

for any  ∈ (0; 1) it follows that

(a2 (0) − a1 (0))(b2 (0) − b1 (0))
¿ Cov(A; B) ¿ 0
4
holds for any pair of (continuous) fuzzy numbers A and B in supp(A) and in supp(B), respectively. In the
limit case, when A is constant (i.e. a1 () = a2 () = a = const:; ∀) we 1nd that Cov(A; B) = 0.
Let A = (a; ) and B = (b; ) be symmetric triangular fuzzy numbers. Then

Cov(A; B) = :
6
The following theorem shows that the variance of linear combinations of fuzzy numbers can easily be computed
(in the same manner as in probability theory).

Theorem 4.1. Let ;  ∈ R and let A and B be fuzzy numbers. Then

Var(A + B) = 2 Var(A) + 2 Var(B) + 2|| Cov(A; B);

where the addition and multiplication by a scalar of fuzzy numbers is de8ned by the sup–min extension
principle.

Proof. Suppose ¡0 and ¡0. Then we 1nd

1 1
Var(A + B) = (a1 () + b1 () − a2 () − b2 ())2 d
2 0


1 1
= ((a1 () − a2 ()) + (b1 () − b2 ()))2 d
2 0



1 1 1 1
= 2 × (a1 () − a2 ())2 d + 2 × (b1 () − b2 ())2 d
2 0 2 0


1 1
+ 2 × (a1 () − a2 ())(b1 () − b2 ()) d
2 0


1 1
= 2 Var(A) + 2 Var(B) + 2 × (a2 () − a1 ())(b2 () − b1 ()) d
2 0

= 2 Var(A) + 2 Var(B) + 2 Cov(A; B)

= 2 Var(A) + 2 Var(B) + 2|| Cov(A; B):

Similar reasoning holds for the case ¿0 and ¿0. Suppose now that ¿0 and ¡0. Then we get


1 1
Var(A + B) = (a1 () + b2 () − a2 () − b1 ())2 d
2 0


1 1
= ((a1 () − a2 ()) + (b2 () − b1 ()))2 d
2 0



2 1 1 2 2 1 1
= × (a1 () − a2 ()) d +  × (b2 () − b1 ())2 d
2 0 2 0
324 C. Carlsson, R. Full-er / Fuzzy Sets and Systems 122 (2001) 315–326

1
1
+ 2 × (a1 () − a2 ())(b2 () − b1 ()) d
2 0

1
2 2 1
=  Var(A) +  Var(B) − 2 × (a2 () − a1 ())(b2 () − b1 ()) d
2 0

= 2 Var(A) + 2 Var(B) + 2|| Cov(A; B);

which ends the proof.

As a special case of Theorem 4.1 we get Var(A)= 2 Var(A) for any  ∈ R and
Var(A + B) = Var(A) + Var(B) + 2 Cov(A; B):
Let A = (a; ) and B = (b; ) be symmetric triangular fuzzy numbers and   be real numbers. Then
2 2  (|| + ||)2
Var(A + B) = 2 + 2 + 2|| = ;
6 6 6 6
which coincides with the variance of the symmetric triangular fuzzy number
A + B = (a + b; || + ||):
Another important question is the relationship between the subsethood and the variance of fuzzy numbers.
One might expect that A ⊂ B (that is A(x)6B(x) for all x ∈ R) should imply the relationship Var(A)6Var(B)
because A is considered a “stronger restriction” than B.
The following theorem shows that subsethood does entail smaller variance.

Theorem 4.2. Let A; B ∈ F with A ⊂ B. Then Var(A)6Var(B).

Proof. From A ⊂ B it follows that b1 ()6a1 ()6a2 ()6b2 (), for all  ∈ [0; 1]. That is,
a2 () − a1 ()6b2 () − b1 ()
for all  ∈ [0; 1], and therefore,



1 1 1 1
Var(A) = (a2 () − a1 ())2 d6 (b2 () − b1 ())2 d = Var(B):
2 0 2 0
which ends the proof.

Remark 4.2. Alternatively, we could also introduce the variance of A ∈ F as

1
Var  (A) = ([ML (A) − a1 ()]2 + [ML (A) − a2 ()]2 ) d
0

1
= (a21 () + a22 ()) d − E 2 (A);
0

i.e. the possibility-weighted average of the squared distance between the expected value and the left- and
right-hand endpoints of its level sets; and the covariance as

1

Cov (A; B) = ([ML (A) − a1 ()][ML (B) − b1 ()] + [ML (A) − a2 ()][ML (B) − b2 ()]) d:
0

Then, the following theorem would hold

 C. Carlsson, R. Full-er / Fuzzy Sets and Systems 122 (2001) 315–326 325

Theorem 4.3. Let ;  ∈ R such that ¿0 and let A and B be fuzzy numbers. Then

Var  (A + B) = 2 Var  (A) + 2 Var  (B) + 2 Cov (A; B);

where the addition and multiplication by a scalar of fuzzy numbers is de8ned by the sup–min extension
principle.

Proof. Suppose ¡0 and ¡0. Using the linearity of the expected value we 1nd

Var  (A + B)

1
= ((E(A + B) − a2 () − b2 ())2 + (E(A + B) − a1 () − b1 ())2 ) d
0

1
= (ML (A) + ML (B) − a2 () − b2 ())2 d
0

1
+ (ML (A) + ML (B) − a1 () − b1 ())2 d
0

1
= ((ML (A) − a2 ()) + (ML (B) − b2 ()))2 d
0

1
+ ((ML (A) − a1 ()) + (ML (B) − b1 ()))2 d
0

1
= 2 ([ML (A) − a1 ()]2 + [ML (A) − a2 ()]2 ) d
0

1
+ 2 ([ML (B) − b1 ()]2 + [ML (B) − b2 ()]2 ) d
0

1
+ 2 ([ML (A) − a1 ()][ML (B) − b1 ()] + [ML (A) − a2 ()][ML (B) − b2 ()]) d
0

= 2 Var  (A) + 2 Var  (B) + 2 Cov (A; B):

Similar reasoning holds for the case ¿0 and ¿0, which ends the proof.

However, nothing can be said about Var  ( A + B) if ¡0; and it is not clear if subsethood entails
smaller variance. However, if A = (a; ; ) is a triangular fuzzy number then

2 + 2 + 
Var  (A) =
18
and, therefore, any triangular subset of A will have smaller variance.

5. Summary

We have introduced the notation of interval-valued possibilistic mean of fuzzy numbers and investigated
its relationship to interval-valued probabilistic mean. We have proved that the proposed concepts “behave
properly” (in a similar way as their probabilistic counterparts).
326 C. Carlsson, R. Full-er / Fuzzy Sets and Systems 122 (2001) 315–326

Acknowledgements

The authors are thankful to Didier Dubois and Maria Angeles Gil for their useful comments and suggestions
on the earlier versions of this paper.

References

[1] D. Dubois, H. Prade, Fuzzy Sets and Systems: Theory and Applications, Academic Press, New York, 1980.
[2] D. Dubois, H. Prade, The mean value of a fuzzy number, Fuzzy Sets and Systems 24 (1987) 279–300.
[3] R. Goetschel, W. Voxman, Elementary fuzzy calculus, Fuzzy Sets and Systems 18 (1986) 31–43.
[4] L.A. Zadeh, Fuzzy Sets, Inform. and Control 8 (1965) 338–353.