FUZZY

sets and systems

ELSEVIER Fuzzy Sets and Systems 86 (1997) 213-222

Integrals of fuzzy-number-valued functions

Yun Kyong Kim a'*, Byung Moon Ghilb
aDepartment of Mathematics, Dongshin University, Naju, Chonnam 520-714 South Korea
bDepartment of Mathematics, Sun Moon University, Asan, Chungnam, South Korea

Received June 1995; revised November 1995

Abstract
Using the representation theorem of fuzzy numbers, we give an alternative approach for the integrals of fuzzy-number-
valued functions, and for the Radon-Nikodym theorem of the fuzzy-number-valued measures. @ 1997 Elsevier Science
B.V.
Keywords: Fuzzy numbers; Fuzzy-number-valued functions; Fuzzy-number-valued measures

1. Introduction

Integrals of set-valued functions were studied by Aumann [1], Debreu [6], Hiai and Umegaki [10], Papa-
georgiou [12], and others. Integrals of fuzzy set valued functions as a natural generalization of set-valued
functions have been established by Kaleva [11], Puri and Ralescu [13], Stojakovic [16], and so on. Also,
Fubini theorem for fuzzy-set-valued integrals was obtained by Zhang and Guo [18], and the Radon-Nikodym
theorem for fuzzy-set-valued measures by Ban [2], Purl and Ralescu [14], Stojakovic [17]. But they are all
based on the corresponding theory of set-valued functions.
The purpose of this paper is to give an alternative approach slightly different from those in the above works
without the aid of the corresponding results for multivalued function. This will be accomplished by using the
representation theorem of fuzzy numbers discussed in [8].
According to the representation theorem for fuzzy numbers [8], a fuzzy number is completely determined
by the end-points of the intervals. Since then, integrals of fuzzy-number-valued functions can be changed
into non-fuzzy forms, we only need to discuss the corresponding classical results for point-valued functions.
This offers us a more concise methodology than in former works and assists us in obtaining some extended
results.
Unfortunately, our work is not applicable to the case of 2- or higher-dimensional fuzzy-number-valued
functions. For, the representation theorem of fuzzy numbers is meaningful only in the one-dimensional
case.

*Corresponding author.

0165-0114/97/$17.00 @ 1997 Elsevier Science B.V. All rights reserved

SSDI 01 65-0114(95 )00400-9
214 EK. Kim, B.M. Ghil/Fuzzy Sets and Systems 86 (1997) 213 222

2. Preliminaries

A fuzzy number is a fuzzy set u : R ~ [0, 1] with the properties:

(1) u is normal, i.e., there exists x ~ R such that u(x) = 1;
(2) u is upper semicontinuous;
(3) u is a convex fuzzy set, i.e., u(Zx + (1 - 2)y)~> m i n { u ( x ) , u ( y ) } for x , y E R and Z c [0, 1];
(4) suppu = cl{x E R; u ( x ) > 0 } is compact.
The family of all fuzzy numbers is denoted by E. For a fuzzy set u, we define

[u]~ = f { x u ( x ) > e } , o < e~l,

{ supp u, e = 0.
Then, it is easily established that u is a fuzzy number if and only if
(1) [u]~ is a closed and bounded interval for each e E [0, 1],
(2) [u]l ¢ O,
From this characterization of fuzzy numbers, it follows that a fuzzy number u is completely determined by
the end points of the intervals [u]~ [u~-, u +]

Theorem 2.1 (Goetschel and Voxman [8]). For u c E, denote u-(:O = u~,u+(~) = u +. Then
(1) u - ( c 0 is" a bounded increasing function on [0, 1];
(2) u+(e) is a bounded decreasing function on [0, 1];
(3) u-(1)~< u+(l);
(4) u - ( e ) and u+(e) are left continuous on (0, 1] and right continuous at 0;
(5) I f u (~) and u+(a) satisJ.), above (1) (4), then there exists a unique v E E such that v~ = u - ( ~ ) , G =
u+(e).

The above theorem implies that we can identify a fuzzy number u with the parameterized representa-
tion {(u~-, u+)[ 0~< e ~<1}. Suppose that u, v E E are fuzzy numbers represented by {(u2, u + ) [ 0 ~ < ~ < 1} and
{(v~-, v+) I 0 ~<:~ ~< 1}, respectively. If we define

(u + v)(z) = sup m i n ( u ( x ) , v ( y ) ) , (2.1)

X+)'=Z

{u(z/;.), ;. ¢ o,
(2.2)
(),u)(z) = o, 2 = o, where 0 = Z{o},

then
,+

+)lO<.e<-..1}, 2~>0,

~u--
~{(;,u2,),u
k{(;,u~+,;-u;)lo~<=~<l}, ;. < o.
We define a metric d on E by
d(u,v)= sup dH([U]~,[V]~), (2.3)
0~<:~< I

where dH is the Hausdorff metric defined as

+ .+
d.([u]~,[v]~) -- m a x ( I u ; - v~- I, l u~ - ~ I). (2.4)
Also, d(u, 0) will be denoted by Ilull.
 EK. Kim, B.M. Ghil/Fuzzy Sets and Systems 86 (1997) 213~22 215

3. Integrals of fuzzy-number-valued functions

Throughout this paper, (Q, 27,/~) denotes a complete a-finite measure space. I f F : f2 ---+E is a fuzzy-number-
valued function and B is a subset of R, then F - I ( B ) denotes the fuzzy subset of f2 defined by F - l ( B ) ( c o ) =
SUPxeBF(eJ)(x) for every co E f2. The fuzzy-number-valued function F : f2 --+ E is called measurable if for
every closed subset B of R the fuzzy set F - I ( B ) is measurable when considered as a function from (2 to
[0, 1]. This concept of measurability for fuzzy-set-valued functions was introduced by Butnariu [3] as a natural
generalization of measurable multifunctions. Kaleva [1 1] defined F : f2 ---+E to be strongly measurable if for
each 7 E [0, l] the set-valued function F= : (2 --+ I(R) defined by F=(eo) = [F(o))]~ is measurable, where I(R)
is the set of all closed bounded intervals on R endowed with the topology generated by the Hausdorff metric
dH defined as in (2.4). The next theorem shows that the above definitions are equivalent.

Theorem 3.1. For F : f2 --* E, F(o)) = {(F~-(co),F+(co)) I 0~<c~< 1}, theJbllowing conditions are equivalent.
(1) F is" measurable.
(2) F is strongly measurable.
(3) For each ct E [0, 1], F 2 and F + are measurable.

Proof. The equivalence of (1) and (2) follows from Butnariu [3] and that of (2) and (3) is due to Congxin
and Ming [5]. []

Let F : f2 ---+ E,F(oJ) = ((f~(~),f~(o))10~l}, be measurable. If for each e E [0,1],Fg- and

F + are integrable, it follows from Theorem 2.1 and the Lebesgue-dominated convergence theorem that the
parameterized representation {(fA F~- dl~, fA F~+ d#) [ 0 ~<:~ ~< 1} is a fuzzy number for each A E 27. This enables
us to define the integral of F without using the integral of set-valued function.

Definition 3.2. A measurable function F : f 2 - - + E , F ( m ) = {(F£-(~o),F+(o)I 0~<c~< 1} is called integrable if

for each ~ E [0, 1], F~- and F + are integrable, or equivalently, if F o and F0+ are integrable. In this case,
the integral of F over A E 2; is defined by

Fd ={(fF;d ,fF+du) (3.1)

It follows immediately that F is integrable if and only if the real-valued function ~o F-~ [IF(co)ll is integrable.
Note that in [2, 5, 11, 13, 16] the terminology integrably bounded was used instead.
The following theorem was obtained by Kaleva [11]. We use a slightly different approach from that of
[111.

Theorem 3.3. Let F, G : ~2 ~ E be integrable. Then

(1) i r a and b are real numbers, then aF + bG is integrable and for each A E Z, fa(aF + b G ) d # =
a fA F d # + b fA Gd/~;
(2) if d is the metric on E which defined as in (2.3), then d(F, G) is a real-valued integrable Junction
and Jot each A E S,

d (~Fdu, fAGdl,) <.~Ad(F,G)d#. (3.2)

In particular, I] • Fd~ll ~ J~ IIFII dlz.

216 Y.K. Kim, B.M. Ghil/Fuzzy Sets and Systems 86 (1997) 213-222

Proof. The assertion (1) is trivial. To prove the assertion (2), We first note that d(F, G) is measurable, for,
if {rk} is a countable dense subset of [0, 1], then

d(F(co), G(co)) = sup max(lFr~ (co) - G~(co)l , IF+(co) - G + ( o 0 l )

k
which follows from the fact that F~- and F~+ considered as functions of ~ are left continuous on (0, 1] and
right continuous at 0. Now the integrability of d(F, G) follows from

d(F, G)<~ IIFII + IlCll,

where IIFll = max(IFol, rF~ I) and j[Cll = max([ Go [, i t ? I) are integrable functions. Finally, it follows that
for each ~ E [0, I],

Similarly,

tF+dl~-~'G+dp ~ ~d(F,G)dp.

Consequently, we obtain

[]

The following theorem first obtained by Puff and Ralescu [13] for the case of nonatomic measure spaces
extends the classical Lebesgue-dominated convergence theorem.

Theorem 3.4. Let {Fn, n/> 1 } and F be measurable fuzzy-number-valued functions such that Fn(~o) d F(oJ)
for almost every 09 E (2. I f there exists an inteorable function h : (2 ---+R such that for all n
[[F,(~)[[ ~< h(~o),
then F is integrable and

f F,, dB ~ f F dlz.

Proof. The integrability of F is trivial. To prove the second statement, we first note that

d(Fn,F)<~ d(Fn, 6) + d( (),F)<~ h + IlFll.

Since d(Fn, F) ~ 0 a.e., it follows from (3.2) and the classical Lebesgue-dominated convergence theorem
that

d(f F. dp,jFdk 0 <~f d(F.,F)dtz~O. []

Let L(~2, Z, It) = L denote the Banach space of (equivalence classes of) integrable function f : Q ~ R with
the norm

Hfl[ = f If(c°)l dp.

Jo
 Y.K. Kim, B,M. GhillFuzzy Sets and Systems 86 (1997) 213-222 217

For the case o f fuzzy-number-valued functions, we denote £~a(E) the space o f all integrable functions F : f2 --~
E, where two functions F, G E 5°(E) are considered to be identical if F(~o) = G(~o) a.e. For F, G E &°(E),
we define

A(F, G) = L d(F, G) d#. (3.3)

Then, it is easy to show that A is a metric on ~ ( E ) .

In view o f Theorem 3.1 in [15], ~,e(E) is complete with respect to the metric

D(F, G) sup l m a x ( i F ~ - - G~-], IF+ - G + i ) d # . (3.4)

,/f2

The next theorem is slightly different from that in [15].

Theorem 3.5. Lf(E) is a complete metric space with respect to the metric A.

Proof. Let {Fn} be a Cauchy sequence in ~ ( E ) . Then, for each ~ E [0,1] the sequences {(Fn)~-} and
((Fn)~+ } are Cauchy in L. Since L is a Banach space, there exist G~-, G~+ E L such that (Fn)~ ~ G~ and
( F , ) + --~ G~+ in the L-norm. Now, let {rk} be a countable dense subset o f [0, 1] with r0 = 0 and

A = {~° : supG~(oo) > inf G+(~°)}

Then it follows that p(A) = 0. If we define

0 if co E A ,
F2(~o) = Go(~o) if 09 ~ A, ~ = 0,
suprk<~G~(~o ) i f e o ~ A , ~ > 0,

0 if ~o E A,
F+(co) = G~-(~o) if o9 ~ A, c~ = 0,
infrk<~ G+(og) if co q~ A,~ > 0,

then F = { ( F ~ - , F + ) [ 0 ~ < ~ < 1} is an integrable fuzzy-number-valued function. N o w we shall show that for
each c~ E (0, 1], F~- = G~- a.e. and F~+ = G~ a.e.
Let {~k} be an increasing sequence such that l i m k ~ c~k = c~ > 0. First, we note that as in the proof of
A(Fn,F) --~ 0 below, we can obtain

lim f [ (F~)~ - G~ ] d# = 0 uniformly in ~k.

n ~ j

Then, the inequality

shows that G~ converge to G~- in L-norm. Since G~ converges to F~- a.e., this implies that F~- = O~- a.e.
By similar arguments, we obtain that F + -----G + a.e. It remains to show that Fn A F. Let e( > 0) be given.
Then, since the sequence {F~} is Cauchy, there exists N such that f d(F,,,Fm)d# < c for n,m >~N. Let
n(>~N) be given. Then for any m>~N and any integer l,

max [(F,)~ - ( F m ) ~ i d # < e.

l~k<~l
218 K i d Kim, B.M. Ghil/Fuzzv Sets and Systems 86 ('1997) 2 1 3 ~ 2 2

Letting m -~ vo, we obtain

max ](Fn)~--F,~ld#~<c.
l<~k<~l

By the classical monotone convergence theorem, we conclude

sup I(F,)r~ - F~I d#~<c.

Similarly, it follows that

f sup I(F,)r+~ - F ~ I d# ~<c.

Therefore,

A(F~,F) = / sup max( l ( f , )2 -/~;. I, I(Fn)r~-

+ fr~2 ])d#---<
e . + []
J k

Now we conclude this section by giving an example to show that the metric A defined as in (3.3) may be
strictly larger than the metric D defined as in (3.4).

E x a m p l e 3.6. Let 2 be the a-field o f Lebesgue measurable subsets o f f2 = (0, 1) and let # be the Lebesgue
measure. For co E (2, i f we define
X X

G(0))(x) = (1 - (1 - 0))x) l[o, 1](x) + (1 ÷ (1 - 0))x) l[-i,o)(x),

then F and G are fuzzy-number-valued functions. It follows immediately that

F~ (0)) = 0)(1 - :¢),F~(co) = - F ~ ( 0 ) ) ,

1-z¢
G~,~)(0)) = 1[~,j)(0)) + ~ 1(o,~)(0)), G2(0)) = -G+~(0)).

Thus, F and G are measurable. Now, d(F(0)), G(0))) = 0)2 - co + 1 so that

A(F,G) = /01(0) 2 - e ) + 5
1 ) d o ) = g.

On the other hand,

-- 0) if 0) < ~,
IF((0)) - G2((o) l = I F~-(0)) - a j ( 0 ) ) l = (1 ~) 1 - 0)
1 - (1 - ~)0) if 0)j>~.

Hence,

if:~ < 1,
max(IF;- - a 2 1 , If~* - G+l)d# = 0
if ~ = 1.

Therefore, D ( F , G) = e -1/2 < A(F, G).

 EK. Kim, B.M. Ghil/Fuzzy Sets and Systems 86 (1997) 213-222 219

4. R a d o n - N i k o d y m theorem for fuzzy-number-valued measures

Set-valued and fuzzy-set-valued measures have received much attention in the last few years because of
their usefulness in several applied fields such as mathematical economics and optimal control. The Radon-
N i k o d y m theorem for set-valued measures was established b y Debreu and Schmeilder [7] and Hiai [9]. The
case o f fuzzy-set-valued measures can be found in [2, 14, 17] and so on. In earlier works, in order to introduce
a notion o f countable additivity, the following concept o f sums for a series o f fuzzy numbers was used:

ui (x) = sup (xi) : x = ~ x i , ~ I xi I < ec .

i i i=1

In this section, we shall introduce a notion o f fuzzy-number-valued measures different from those in the
above works and prove a R a d o n - N i k o d y m theorem which generalizes the result for real-valued measures.

Definition 4.1. Let {u,} be a sequence o f fuzzy numbers in E and u E E. The series ~ f - - 1 u~ is said to be
convergent to u if

d b/i, u ~ 0 as n ~ o~.
\i= 1 /

In this case, we call u the sum o f the series y-~o~=,u. and write u = ~ . ~ _ , u..

It follows immediately that the series ~ f _ I u. converges to u if and only if

t/ n
~-]~(ui); --* u ; and ~( U /)~
+ ---+u~+
i=1 i=1

uniformly in ~ C [0, 1]. Furthermore, i f u n = {(u..~,Un.~)

- + [0~<~ ~< 1} are fuzzy numbers such t h a t ~ ° ~ l u n _ ~ =
u ; and ~ f _ , u~.~
+ = u + converge uniformly in ~, then u = { ( u ~ - , u + ) ] 0 ~ < ~ < l } is a fuzzy number and u =

Definition 4.2. A fuzzy-number-valued measure is a set function 9Jl : X --, E with properties
(1) ~ ( 0 ) = 6,
(2) 9J~(Uf_ , A . ) = ~--]~.~1 9J~(a.) for every sequence {A.} o f pairwise disjoint elements o f Z.

Note that a set function 9J~ " Z --~ E, ~ l ( A ) = { ( ~ ( A ) , g J ~ + ( A ) ) 1 0 ~ < ~ < 1 } is a fuzzy-number-valued
measure if and only if
(1) for each :~ E [0, 1], 9J~- and 9J~+ are real-valued measures,
(2) the families {gJt~-10~<2~<1} and {gJ~+ ] 0 ~ < ~ < 1 } o f measures are both uniformly countably addi-
tive, that is for any sequence {A.} o f pairwise-disjoint elements o f Z, the convergence o f 9Jl~-(Un~ i A . ) =
~ n ~ , sJJ~;(An) is uniform in ~ E [0, 1], and likewise for {gJ~+ 10~<c¢~< 1}.

Definition 4.3. If m is a classical measure in Z, and 9J~ is a fuzzy-number-valued measure, then 9Jl is called
absolutely continuous with respect to m if re(A) = 0 implies 9JI(A) = 0.

Theorem 4.4. I J ' F " ~ --+ E is integrable, then the set function defined by

93~(A) = f F d # , A E Z,

is a fuzzy-number-valued measure which is absolutely continuous with respect to #.

220 EK. Kim. B.M. Ghil/Fuzzy Sets and Systems 86 (1997) 213-222

Proof. It is trivial that 9Jl(0) = 0 and 9Jl~- ,99l~

+ are real-valued measures. To prove that ~ is a fuzzy-
number-valued measure, it suffices to show that {gJl~-10~< ~ ~< 1} is uniformly countably additive. Let {An}
be a sequence o f pairwise-disjoint elements o f ~7. Then

IA~ - ~ l l ~ - ( i ) = F~- ~ E IIFll du --+ 0

i i= ~,,+lAi i=n+l i

uniformly in u, as n ---+ oo. This proves that {gJ~-I 0 ~ a ~< 1 ) is uniformly countably additive. The absolute
continuity with respect to p is trivial. []

The next theorem is the converse o f the above theorem which generalizes the classical Radon-Nikodym
theorem.

Theorem 4.5. I f a fuzzy-number-valued measure 931 : Z --+ E is absolutely continuous with respect to Ix, then
there exists a unique integrable function F : 12 ---+E such that for all A E Z,

9J~(A) = ~ F d#.

Proof. First we note that for each a E [0, 1], 9Jl~ and 931+ are real-valued measures which are absolutely con-
tinuous with respect to Ix. The classical Radon-Nikodym theorem implies that there exist integrable functions
f~- and f + such that for all A E 27,

?iJl; (A ) = f f ; d ix,

9JI+(A) = iA f + dIx.
Now, let {rk} be a countable dense subset o f [0,1] with r0 = 0 and

B = {°J : sup f~(°J) > inf f+(~°)}

Then it is trivial that Ix(B) = 0. If we define

0 if~EB,
F~- ( ~ ) = fo(OJ) if ~o ~ B,~ = 0,
SUprh< ~ f ~ ( o J ) if co g~ B,~ > 0,

0 if ~o E B,
F + (~s~) =
l f0v (¢o)
infr~ <~ f+(~o)
then F =- { ( F ~ - , F + ) I 0 ~ < ~ I }
if ~o q~ B,~ = 0,
if co q~B,~ > 0,
is an integrable fuzzy-number-valued function. Let { ~ } be an increasing
sequence such that limn-+~o an = ~ > 0. Then, by the monotone convergence theorem,

fAFSdp= limooSAf~dlx~
,, foreachAc27.

Also, left continuity o f 9Jl~- at ~ implies that for each A E 27

f f~- dix = 9JI~-(A)= lira ~I~,,(A)= lim f f~,, dix.

JA n ---+ o o n ---+ o o JA
 Y.K. Kim, B.M. Ghil/Fuzzy Sets and Systems 86 (1997) 213-222 221
Hence, Fg- = f~- a.e. and similar arguments show that F + = f + a.e., which together imply

~(A) = { F d~.
To prove the uniqueness of F, let G : f2 ---* E be any integrable function such that for all A E S

9~(A) = L G dp.

Then, by the uniqueness of the classical Radon-Nikodym derivative, for each ~, F~- = G~- a.e. and F + = G +
a.e. so that there exists B E S with # ( B ) = 0 such that

F7-(CO) = Gr(co ) and F+(co) = G+~(co)

for all co E O\B and rational r E [0, 1]. By the left continuity of F f and G~ at ~, we have

F~-(CO) = G~-(CO) and F+(co) = G+(co)

for all co E Q\B and ¢~ E [0, 1]. Hence, we conclude that F = G a.e. []

5. Fubini theorem for fuzzy-number-valued integrals

The Fubini theorem for set-valued integrals and fuzzy-set-valued integrals was established by Zhang and
Guo [18] using the properties of support function. In this section, the Fubini theorem for fuzzy-number-valued
integrals is given using only the classical Fubini theorem.

Theorem 5.1, Let (~'~l,Zl,]./l) and (~r~2,Z2, J/2) be two complete a-finite measure spaces, and let (~r'~1 × ~r~2,
Zt × S2,#l × P2) be their product measure space. I f a fuzzy-number-valued function F : f21 x Q2 --* E is
pl × p2-integrable, then
(1) the fuzzy-number-valued function F ( . , 092) : COl w-+ F(cot, (-02) is lq-integrable for 092 E 02, P2-a. e.
(2) the fuzzy-number-valued function co2 ~-+ fo, F(o0t, o02) dpl(o0t ) is #2-inteyrable, and

fQ Fd#l×p2=£ £ F(col,co2)d/-tl(col)dp2(co2).
I X(22 2 1

ProoL Let F(col, (-02) = {(F~-(col, co2),F~+(col, co2)) I 0 ~ ~ ~ 1}. Then, since F 0 and F + are Pl × p2-integrable,
it follows immediately from the classical Fubini theorem that, for co2 E (22 p2-a.e., col ~ Fo(col,co2) and
col ~-+ F~-(col,co2) are pl-integrable which implies the first assertion. Also, for the second assertion, an
appeal to the classical Fubini theorem shows that f~, Fo(col,co2)d~l(~01 ) and f~, F+(col,co2)d/~l(cot) are
#2-integrable. Therefore, fo, F(col, co2) d/~l (col) is p2-integrable and

£,×aFd#,×#2={I£~×%F:d#,×#2,I F+dp,×p2)0~<~<1}
. JaQl X Q 2

=£2£ Fd#ldp2" []
222 EK. Kiln, B.M. Ghil/Fuzzy Sets and Systems 86 (1997) 213 222

References

[1] R.J. Aumann, Integrals of set-valued functions, J. Math. Anal. Appl. 12 (1965) 1-12.
[2] J. Ban, Radon-Nikodym theorem and conditional expectation of fuzzy-valued measures and variables, Fuzz), Sets and Systems 34
(1990) 383-392.
[3] D. Butnariu, Measurability concepts for fuzzy mappings, Fuzzy Sets and Systems 31 (1989) 77 82.
[4] W. Congxin and M. Ming, Embedding problem of fuzzy number space: part I, Fuzzy Sets and Systems 44 (1991) 33 38.
[5] W. Congxin and M. Ming, Embedding problem of fuzzy number space: part II, Fuzzy Sets and Systems 45 (1992) 189 202.
[6] G. Debreu, Integration of correspondences, Proe. 5th Berkeley Symp. Math. Statist. Prob., Vol. 2 (1966) 351 372.
[7] G. Debreu and D. Schmeidler, The Radon+Nikodym derivative of a correspondence, Proe. 6th Berkeley Syrup. Math. Statist.
Prob., Vol. 2 (1972) 41-56.
[8] R. Goetschel and W. Voxman, Elementary fuzzy calculus, Fuzzy Sets and Systems 18 (1986) 31 43.
[9] F. Hiai, Radon-Nikodym theorems for set-valued measures, J. Multivariate Anal. 8 (1978) 96-118.
[10] F. Hiai and H. Umegaki, Integrals, conditional expectations and martingales of multivalued functions, J. Multivariate Anal. 7 (1977)
149-182.
[11] O. Kaleva, Fuzzy differential equations, Fuzz)' Sets and Systems 24 (1987) 301 317.
[12] N. Papageorgiou, On the theory of Banach space valued multifunctions part 1; Integration and conditional expectation, J. Multivariate
Anal. 17 (1985) 185-206.
[13] M. Puri and D. Ralescu, Fuzzy random variables J. Math. Anal. Appl. !14 (1986) 409-422.
[14] M. Puri and D. Ralescu, Convergence theorem for fuzzy martingales, J. Math. Anal. Appl. 160 (1991) 107-122.
[15] M. Stojakovic, Fuzzy conditional expectation, Fuzzy Sets and Systems 52 (1992) 53 60.
[16] M. Stojakovic, Fuzzy random variables, expectation and martingales, J. Math. Anal. Appl. 184 (1994) 594-606.
[17] M. Stojakovic, Fuzzy valued measure, Fuzzy Sets and Systems 65 (1994) 95-104.
[18] D. Zhang and C. Guo, Fubini theorem for F-valued integrals, Fuzz)' Sets and Systems 62 (1994) 355-358.