Available online at www.sciencedirect.

com

Information Sciences 178 (2008) 1756–1779

www.elsevier.com/locate/ins

q
Comparison results for fuzzy differential equations
Rosana Rodrı́guez-López
Departamento de Análisis Matemático, Facultad de Matemáticas, Universidad de Santiago de Compostela,
15782 Santiago de Compostela, Spain

Received 28 May 2007; received in revised form 8 October 2007; accepted 11 October 2007

Abstract

In this paper, we present several comparison results for the solutions of fuzzy differential equations, by using the Huku-
hara derivative. These results constitute the extension to the fuzzy context of some comparison results for the solutions of
linear differential equations. Besides, some side-results about the Hukuhara difference and the partial orderings are given.
We illustrate the applicability of the new results by showing several examples.
 2007 Elsevier Inc. All rights reserved.

Keywords: Fuzzy differential equation; Fuzzy maximum principle; Non-negative fuzzy functions

1. Introduction

Monotone iterative technique constitutes a useful method to prove the existence of solution and to approx-
imate solutions of differential equations of different types (ordinary, functional, integro-differential equations,
etc.).
In the development of the method of upper and lower solutions and iterative techniques of monotone type,
maximum principles play an important role, since they allow to compare functions on the basis of differential
inequalities. The description of such techniques can be found in [23] for the case of ordinary differential equa-
tions. In this reference, we find several comparison results valid in the classical case and their application to
obtain monotone sequences which approximate the extremal solutions to problems relative to ordinary differ-
ential equations between a lower and an upper solution. In this sense, some useful comparison results are the
following.
Lemma 1. Let u 2 C 1 ð½0; T ; RÞ and M 2 R, M > 0, be such that
u0 ðtÞ þ MuðtÞ 6 0; t 2 ½0; T ;

q
Research supported in part by Ministerio de Educación y Ciencia and FEDER, projects MTM2004-06652-C03-01 and MTM2007-
61724, and by Xunta de Galicia and FEDER, project PGIDIT05PXIC20702PN.
E-mail address: [email protected]

0020-0255/$ - see front matter  2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.ins.2007.10.011
 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779 1757

and
uð0Þ 6 0 or uð0Þ 6 uðT Þ:
Then u(t) 6 0, t 2 [0, T].

Lemma 2. Let u 2 C 1 ð½0; T ; RÞ and M 2 R, M > 0, be such that

u0 ðtÞ þ MuðtÞ P 0; t 2 ½0; T ;
and
uð0Þ P 0 or uð0Þ P uðT Þ:
Then u(t) P 0, t 2 [0, T].
To the best of our knowledge, comparison results or maximum principles have not been previously inves-
tigated in detail in the fuzzy context. In this paper, considering two different orderings in the space of fuzzy
numbers, and following different approaches, we present several results which are analogue to Lemmas 1 and 2
in the fuzzy case and allow to compare the solutions to different ‘linear’ fuzzy differential equations relative to
different fuzzy initial value conditions. The study is more complex than the corresponding analysis in the clas-
sical case, due to the nature of fuzzy numbers.
Definition 1. We consider E1 the space of fuzzy numbers
x : R ! ½0; 1
satisfying the following properties:

(i) x is normal: there exists t0 2 R with x(t0) = 1.

(ii) x is fuzzy convex: for all t; s 2 R and k 2 [0, 1],
xðkt þ ð1  kÞsÞ P minfxðtÞ; xðsÞg:
(iii) x is upper-semicontinuous.
(iv) ½x0 ¼ ft 2 R : xðtÞ > 0g is a compact set.

The level sets of x,

½xa ¼ ft 2 R : xðtÞ P ag; a 2 ð0; 1;
0
and [x] are non-empty compact convex sets in R (see [14]).
Thus, for x 2 E1, we denote the level sets of x by
a
½x ¼ ½xal ; xar  for all a 2 ½0; 1:
We define two orderings in E1, as follows.
Definition 2. Let x, y 2 E1. We say that x 6 y if and only if
xal 6 y al and xar 6 y ar for every a 2 ½0; 1:

Definition 3. Let x, y 2 E1. We say that x  y if and only if

xal P y al and xar 6 y ar for every a 2 ½0; 1;

that is,
a a
½x  ½y 8a 2 ½0; 1:
An important difference between these orderings is that there are no fuzzy numbers x with x  v{0}, since
x  v{0} means that the non-empty sets [x]a  {0}, for all a 2 [0, 1], then [x]a = {0}, for all a, and x = v{0}.
However, there are fuzzy numbers with x < v{0}, for instance, v{1}.
1758 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779

The addition of fuzzy numbers and the product of a fuzzy number by an scalar are given by the Zadeh’s
Extension Principle:
a a a
½x þ y ¼ ½x þ ½y ; a 2 ½0; 1;

and, for k 2 R, k 5 0,
a a
½kx ¼ k½x ; a 2 ½0; 1:

See also [14, p. 4], or [24, p. 3].

In Section 2, we prove some comparison results for the solutions of ‘linear’ fuzzy differential equations. The
peculiarity of fuzzy numbers produces, as stated in [32], different behavior for the solutions of equations
u 0 (t) + Mu(t) = r(t) and u 0 (t) = Mu(t) + r(t), where M 2 R, M 5 0 and u(t) 2 E1, for all t. In fact, an initial
value problem relative to the second equation is always solvable but, considering the first equation, we have to
impose certain assumptions to obtain the existence of solution.
First, we present some properties concerning order preservation under operations with fuzzy numbers, in
relation with the two different orderings 6 and . Then, we establish results which allow to compare a fuzzy
function with the characteristic function of the set {0}, by checking that it satisfies a certain fuzzy differential
inequality. The use of the Hukuhara difference can be useful to compare two fuzzy functions in some cases.
More generally, we prove some results which allow to establish a comparison between two fuzzy functions
which are solutions to different ‘linear’ fuzzy differential equations. Note that the study of the existence of
solutions for the corresponding initial value problems (some of them are always solvable but other are solvable
under appropriate conditions) is included in [32]. Here, we consider that the functions we want to compare are,
in fact, solutions to certain ‘linear’ problems. Our study is complemented with several examples which illus-
trate the applicability of the results obtained.
The basic concepts relative to metric spaces of fuzzy sets which are needed in this work can be found in [14].
Some works relative to the study of initial value problems for fuzzy differential equations are [2,3,6–9,11–
14,16,17,20,21,24,25,29–33]. Some numerical methods for fuzzy differential equations are developed in
[4,5,19,26]. Fuzzy functional differential equations with delay and some impulsive fuzzy differential equations
are studied in [24,34]. On the other hand, [1,27,28] are related to the process of defuzzification and
[10,18,22,35,40] include some applications of fuzzy differential equations and fuzzy difference equations. In
relation with new concepts of differentiability of fuzzy valued-mappings, see, for instance, [6,36]. Ref. [37]
is devoted to the study of the characterization of relative compactness in spaces of fuzzy sets and its applica-
tion to the resolution of fuzzy differential equations. For the establishment of fixed point theorems for a cer-
tain class of fuzzy mappings, see [39], and for applications of the parametric representation of fuzzy numbers,
we refer to [38]. We also cite the monograph [15] on fuzzy sets and their applications.

2. Maximum principles

First, we establish some properties of the partial orderings 6 and  which can be proved easily by using the
definition of the fuzzy addition and multiplication by an scalar.
Lemma 3. The following properties are valid:

• Given x, y 2 E1, x = y if and only if x 6 y and x P y.

• If x, y, z 2 E1 are such that x 6 y, then x + z 6 y + z.
• If x, y, z, w 2 E1 are such that x 6 y, and z 6 w, then x + z 6 y + w.
• If x, y 2 E1 are such that x 6 y, and M 2 R, M > 0, then Mx 6 My.

Analogously, for the partial ordering .

Lemma 4. The following properties are valid:

• Given x, y 2 E1, x = y if and only if x  y and x  y.

 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779 1759

• If x, y, z 2 E1 are such that x  y, then x + z  y + z.

• If x, y, z, w 2 E1 are such that x  y, and z  w, then x + z  y + w.
• If x, y 2 E1 are such that x  y, and M 2 R, M > 0, then Mx  My.
Some properties of fuzzy functions with respect to these partial orderings which are useful to our procedure
are the following:
Lemma 5. If T > 0, J = [0, T], and f, g 2 C(J, E1)
Z t Z t
f 6g) f ðsÞds 6 gðsÞds for t 2 J ;
0 0
Z t Z t
f g) f ðsÞds  gðsÞds for t 2 J :
0 0

The proof of the properties in Lemma 5 follows directly from the definition of the integral of a fuzzy func-
tion (see, for instance, [14]).

Theorem 1. Let u 2 C1([0, T], E1), and M 2 R, M > 0.

(i) If u 0 (t) + Mu(t) P v{0}, t 2 [0, T], and u(0) P v{0}, then u(t) P v{0} for every t 2 [0, T].
(ii) If u 0 (t) + Mu(t) 6 v{0}, t 2 [0, T], and u(0) 6 v{0}, then u(t) 6 v{0}, for every t 2 [0, T].

An analogous result holds replacing P by  in (i) (respectively 6 by  in (ii)).

Proof. Take
a
½uðtÞ ¼ ½ual ðtÞ; uar ðtÞ; a 2 ½0; T ;
then ual(Æ) and uar(Æ) are functions in C 1 ð½0; T ; RÞ.
Consider (i) and the relation P. Theorem 2.5.2 [24], the definition of the operations in E1 and the
hypotheses imply that, for each a 2 [0, 1],
u0al ðtÞ þ Mual ðtÞ P 0; u0ar ðtÞ þ Muar ðtÞ P 0; t 2 ½0; T ;
and
ual ð0Þ P 0; uar ð0Þ P 0;
then, by Lemma 2 for every a 2 [0, 1], and t 2 [0, T],
ual ðtÞ P 0; uar ðtÞ P 0;
which means that u(t) P v{0}, t 2 [0, T].
For the case  in (i), the behavior of the left endpoint of the level sets is different, that is, for every a 2 [0, 1],
u0al ðtÞ þ Mual ðtÞ 6 0; t 2 ½0; T ; ual ð0Þ 6 0;
so that, by Lemma 1,
ual ðtÞ 6 0 for every a 2 ½0; 1; and t 2 ½0; T ;
therefore u(t)  v{0}, t 2 [0, T].
The proof of (ii) can be obtained similarly. h

Theorem 2. Let u 2 C1([0, T], E1), and M 2 R, M > 0, be such that

u0 ðtÞ þ MuðtÞ P vf0g ; t 2 ½0; T ;
and
uð0Þ P uðT Þ:
Then u(t) P v{0}, t 2 [0, T] (analogously, for ).
1760 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779

Proof. Using the same notation of the proof of Theorem 1, functions ual(Æ) and uar(Æ) belong to C 1 ð½0; T ; RÞ,
for every a 2 [0, 1].
Consider the case P. Then, for each a 2 [0, 1], we have
u0al ðtÞ þ Mual ðtÞ P 0; u0ar ðtÞ þ Muar ðtÞ P 0; t 2 ½0; T ;
and
ual ð0Þ P ual ðT Þ; uar ð0Þ P uar ðT Þ;

in consequence, by Lemma 2, we obtain, for every a 2 [0, 1], and t 2 [0, T],
ual ðtÞ P 0; uar ðtÞ P 0;

that is, u(t) P v{0}, t 2 [0, T].

For , and the left endpoint of the level sets ual, we have,
u0al ðtÞ þ Mual ðtÞ 6 0; t 2 ½0; T ; ual ð0Þ 6 ual ðT Þ;

and, applying Lemma 1, we achieve

ual ðtÞ 6 0 for every a 2 ½0; 1; and t 2 ½0; T ;
hence, u(t)  v{0}, t 2 [0, T]. h

Remark 1. Note that, if u(0) P u(T), then ual(T) 6 ual(0) and uar(T) 6 uar(0), for every a 2 [0, 1], hence the
diameter of each level set diam([u(t)]a) = uar(t)  ual(t) could perfectly increase in the variable t. However,
u(0)  u(T) represents [u(T)]a  [u(0)]a "a 2 [0, 1], which joint to the differentiability of u in the sense of Huku-
hara provides [u(T)]a = [u(0)]a "a 2 [0, 1], and, thus, the diameter of the level sets of u is constant,
diam([u(t)]a) = diam([u(0)]a) "t 2 [0, T] "a 2 [0, 1].
It is also possible to prove the following result.
Theorem 3. Let u 2 C1([0, T], E1), and M 2 R, M > 0, be such that
u0 ðtÞ þ MuðtÞ 6 vf0g ; t 2 ½0; T ;
and
uð0Þ 6 uðT Þ:

Then u(t) 6 v{0}, t 2 [0, T].

An analogous result follows for the partial ordering . Note that u 0 (t) + Mu(t)  v{0}, for t 2 [0, T], implies
that u 0 (t) + Mu(t) = v{0}, for t 2 [0, T].
Once we have studied how to compare a function with v{0}, we consider the comparison between two dif-
ferent fuzzy functions. In the real case, we compare x and y just comparing the difference x  y with the null
function. In the fuzzy case, substraction is not allowed and we consider the Hukuhara difference.
Definition 4. If x, y 2 E1, we say that z 2 E1 is the Hukuhara difference of x and y, z = x  Hy, if x = y + z,
that is,

½xa ¼ ½ya þ ½za 8a 2 ½0; 1:

We present a first comparison result, considering the difference of Hukuhara of x and y, provided it exists.
To this purpose, we give a characterization of the order relations in terms of the Hukuhara difference.

Lemma 6. If x, y 2 E1 are such that the Hukuhara difference x  Hy exists, then

x 6 y () ðxH yÞ 6 vf0g :
The same conclusion can be deduced for the partial ordering .
 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779 1761

Proof. Taking into account that

a
½xH y ¼ ½xal  y al ; xar  y ar  for every a 2 ½0; 1;
we have that
 
xal 6 y al xal  y al 6 0
x 6 y () 8a () 8a () ðxH yÞ 6 vf0g :
xar 6 y ar xar  y ar 6 0
For partial ordering , we get
 
y al 6 xal xal  y al P 0
x  y () 8a () 8a:
xar 6 y ar xar  y ar 6 0
The existence of x  Hy implies that xal  yal 6 xar  yar, hence xal  yal = xar  yar = 0, that is,
[x  Hy]a = {0} for every a, or (x  Hy)  v{0}. Reciprocally, if (x  Hy)  v{0}, then (x  Hy) = v{0}, and
x = y + v{0} = y, thus x  y. h
To compare the Hukuhara difference of two fuzzy functions x and y, x  Hy, with the characteristic func-
tion v{0}, that is, to establish the ‘sign’ of x  Hy, we can use Theorems 1–3. To this aim, it is convenient to
obtain the expression of the Hukuhara derivative of the fuzzy function x  Hy, in case it exists. In the follow-
ing Lemma, we give the expression of the level sets of the Hukuhara derivative of x  Hy, under the appro-
priate conditions.
Lemma 7. Suppose that x, y : [0, T] ! E1 are differentiable and such that the Hukuhara difference function
x  Hy : [0, T] ! E1, given by (x  Hy)(t) = x(t)  Hy(t), t 2 [0, T], exists and is differentiable in the sense of
Hukuhara in [0, T). Then
0 a
½ðxH yÞ ðtÞ ¼ ½x0al ðtÞ  y 0al ðtÞ; x0ar ðtÞ  y 0ar ðtÞ 8a 2 ½0; 1; t 2 ½0; T Þ:

Proof. For each t 2 [0, T),

 
ðxH yÞðt þ hÞH ðxH yÞðtÞ 0
0 ¼ limþ d 1 ; ðxH yÞ ðtÞ
h!0 h
 
½ðxal  y al Þðt þ hÞ; ðxar  y ar Þðt þ hÞH ½ðxal  y al ÞðtÞ; ðxar  y ar ÞðtÞ 0
¼ limþ sup d H ; ðxH yÞ ðtÞ
h!0 a2½0;1 h
( 
xal ðt þ hÞ  y al ðt þ hÞ  xal ðtÞ þ y al ðtÞ 
¼ limþ sup max    ðxH yÞal ðtÞ;
0
h!0 a2½0;1 h
 )
xar ðt þ hÞ  y ar ðt þ hÞ  xar ðtÞ þ y ar ðtÞ 
  ðxH yÞar ðtÞ :
0
 h
This implies that
 
0 xal ðt þ hÞ  xal ðtÞ y al ðt þ hÞ  y al ðtÞ
ðxH yÞal ðtÞ ¼ limþ  ¼ x0al ðtÞ  y 0al ðtÞ;
h!0 h h
and
0
ðxH yÞar ðtÞ ¼ x0ar ðtÞ  y 0ar ðtÞ: 

Remark 2. To compare two elements x and y in E1, it is not necessary that the Hukuhara differences x  Hy or
y  Hx exist. If the Hukuhara differences [x]a  H[y]a or [y]a  H[x]a exist at each level a 2 [0, 1], then we can
compare the fuzzy numbers levelwise. For instance, if
a a a a
½x H ½y P 0 for a with diam½x P diam½y ;
and
a a a a
½y H ½x 6 0 for a with diam½x < diam½y ;
1762 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779

then x P y. Here, the partial ordering in the level sets is induced by 6, that is, an interval I = [c, d] is such that
I P 0 if d P c P 0.
Note that, if diam[x]a = diam[y]a, then
a a a a
½x H ½y P 0 () ½y H ½x 6 0:

Example 1. Consider the triangular fuzzy number x = (1; 1, 1), that is,
8
> t; t 2 ½0; 1;
<
xðtÞ ¼ 2  t; t 2 ½1; 2;
>
:
0; otherwise;
whose level sets are given by [x]a = [a, 2  a], for every a 2 [0, 1], and the trapezoidal fuzzy number y defined
by
8  
>
> 4t  7; t 2 74 ; 2 ;
>
>  
< 1; t 2 2; 92 ;
yðtÞ ¼  
> 10  4t; t 2 9 ; 5 ;
>
>
> 4 2
:
0; otherwise;
a aþ7 10a
with level sets ½y ¼ 4 ; 4 for every a 2 [0, 1]. The diameters of the respective level sets are
3  2a
diam½xa ¼ 2  2a; diam½ya ¼8a 2 ½0; 1;
4
thus the differences of Hukuhara x  Hy and y  Hx do not exist. However, the Hukuhara differences   exist
levelwise. Indeed, diam[x]a P diam[y]a for every a 2 0; 56 , therefore [x]a  H[y]a exists, for every a 2 0; 56 , and
 
a a a þ 7 10  a 3a  7 2  3a
½x H ½y ¼ ½a; 2  aH ; ¼ ; 6 0:
4 4 4 4

For a 2 5
6
;1 , diam[y]a > diam[x]a, thus [y]a  H[x]a exists, and
 
a a þ 7 10  a
a 3a þ 7 2 þ 3a
½y H ½x ¼ ; H ½a; 2  a ¼ ; P 0:
4 4 4 4
This shows that x 6 y. See Fig. 1.

Remark 3. For the order relation , y  x is equivalent to [y]a  [x]a for every a, thus diam[y]a 6 diam[x]a for
every a, and diam[x]a  diam[y]a has to have constant sign for x and y to be ordered.
In this case, an interval I = [c, d] is such that I  0 if c 6 0 6 d (equivalently, {0}  I).

x y

1 7/4 2 9/4 5/2

Fig. 1. x 6 y.
 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779 1763

If diam[x]a P diam[y]a for every a, and [x]a  H[y]a  0, that is, [x]a  H[y]a  {0}, then y  x. If
diam[x]a P diam[y]a, for every a, and [x]a  H[y]a  0, that is, [x]a  H[y]a  {0}, then [x]a  H[y]a = {0}, and
x = y.
Note that, if diam[x]a = diam[y]a, then
½xa H ½ya  0 () ½ya H ½xa  0 ð½ya H ½xa ¼ f0gÞ:

Example 2. If diam[x]a  diam[y]a has no constant sign for a 2 [0, 1], then x and y are not -comparable. How-
ever, even if [x]a  H[y]a exists, for every a 2 [0, 1], the Hukuhara difference of x and y, x  Hy, does not neces-
sarily exist, since the corresponding levelwise differences might not define a fuzzy number. In such a case, the
levelwise comparison is again useful.
For instance, consider the trapezoidal fuzzy number x defined by
8  
>
> 2ðt þ 1Þ; t 2 1;  12 ;
>
<  
1; t 2  12 ; 12 ;
xðtÞ ¼ 1 
>
> 2ðt  1Þ; t 2 2 ; 1 ;
>
:
0; otherwise;
a a2 2a
with level sets ½x ¼ 2 ; 2 for every a 2 [0, 1], and the triangular fuzzy number y = (0; 1, 1), that is,
8
< t þ 1; t 2 ½0; 1;
>
yðtÞ ¼ t þ 1; t 2 ½1; 0;
>
:
0; otherwise;
whose level sets are [y]a = [a  1, 1  a], for every a 2 [0, 1]. Note that diam[x]a = 2  a P 2  2a = diam[y]a
for every a 2 [0, 1], in consequence the differences [x]a  H[y]a exist, for every a 2 [0, 1]. However, the difference
x  Hy does not exist, since
 h a ai
a a a2 2a
½x H ½y ¼ ; H ½a  1; 1  a ¼  ; 8a;
2 2 2 2
and the diameter of these sets is increasing in a, hence they do not define a fuzzy number. However, although
x  Hy does not exist, using that [x]a  H[y]a  {0} "a 2 [0, 1], we deduce that x  y. See Fig. 2.

Example 3. On the other hand, for the partial ordering 6, if x represents the same trapezoidal fuzzy  number
 x
a
of Example 2, and y denotes the triangular fuzzy number y = (1; 1, 1), then ½x ¼ a2 2
; 2a
2
, and
[y]a = [a, 2  a], for every a 2 [0, 1]. Therefore, there exists [x]a  H[y]a, for every a 2 [0, 1],
 h a i
a2 2a a
½xa H ½ya ¼ ; H ½a; 2  a ¼   1;  1 ;
2 2 2 2
but x  Hy does not exist. Since [x]a  H[y]a 6 0 for every a 2 [0, 1], then x 6 y. See Fig. 3.

x
y

-1 -1/2 1/2 1

Fig. 2. x  y.
1764 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779

x y

-1 -1/2 1/2 1 2

Fig. 3. x 6 y.

Lemma 8. If x, y, z 2 E1 are such that the Hukuhara differences x  Hz and y  Hz exist, then
x 6 y () xH z 6 yH z:
The same characterization can be deduced for the partial ordering .

Proof. If x 6 y, then xal 6 yal, xar 6 yar for every a 2 [0, 1], and thus xal  zal 6 yal  zal, xar  zar 6 yar  zar,
obtaining x  Hz 6 y  Hz. Analogously, x  y implies that yal 6 xal, and xar 6 yar, thus xal  zal P yal  zal,
xar  zar 6 yar  zar.
Reciprocally, if x  Hz 6 y  Hz, using Lemma 3 we get
x ¼ ðxH zÞ þ z 6 ðyH zÞ þ z ¼ y:
For the partial ordering , we use Lemma 4. h
From the point of view of applications, it is more useful the following comparison result, which does not
require the existence of the Hukuhara differences of x and y. Here, C1([c, d], E1) represents the set of functions
u : [c, d] ! E1 which are continuous and differentiable in the sense of Hukuhara in [c, d] with Hukuhara deriv-
ative u 0 continuous in [c, d], where [c, d] is a real interval.
Theorem 4. Let x, y 2 C1([c, d], E1), and M 2 R, M > 0, be such that
x0 ðtÞ þ MxðtÞ 6 y 0 ðtÞ þ MyðtÞ; t 2 ½c; d;
and
xðcÞ 6 yðcÞ:
Then x(t) 6 y(t), t 2 [c, d]. An analogous result is valid for .

Proof. Let
½xðtÞa ¼ ½xal ðtÞ; xar ðtÞ; ½yðtÞa ¼ ½y al ðtÞ; y ar ðtÞ; a 2 ½0; 1; t 2 ½c; d:
For the partial ordering 6, and each a 2 [0, 1], xal, yal, xar, yar are functions in C 1 ð½c; d; RÞ and
x0al ðtÞ þ Mxal ðtÞ 6 y 0al ðtÞ þ My al ðtÞ; x0ar ðtÞ þ Mxar ðtÞ 6 y 0ar ðtÞ þ My ar ðtÞ; t 2 ½c; d;
xal ðcÞ 6 y al ðcÞ; xar ðcÞ 6 y ar ðcÞ;
that is,
0
ðxal  y al Þ ðtÞ þ Mðxal  y al ÞðtÞ 6 0; t 2 ½c; d; ðxal  y al ÞðcÞ 6 0;
0
ðxar  y ar Þ ðtÞ þ Mðxar  y ar ÞðtÞ 6 0; t 2 ½c; d; ðxar  y ar ÞðcÞ 6 0:
By Lemma 1 for a 2 [0, 1] and t 2 [c, d], we get
ðxal  y al ÞðtÞ 6 0; ðxar  y ar ÞðtÞ 6 0;
thus x(t) 6 y(t), t 2 [c, d].
For the partial ordering ,
x0al ðtÞ þ Mxal ðtÞ P y 0al ðtÞ þ My al ðtÞ; t 2 ½c; d; xal ðcÞ P y al ðcÞ;
 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779 1765

hence Lemma 2 applied to (xal  yal) yields

xal ðtÞ P y al ðtÞ; t 2 ½c; d;
and, consequently, x(t)  y(t), t 2 [c, d]. h
Theorem 4 could have been established with the following formulation, and its proof could have been
obtained through a reasoning similar to [32], as follows.
Theorem 5. Let x, y 2 C1([0, T], E1), T > 0, and M 2 R, M > 0, be such that
x0 ðtÞ þ MxðtÞ ¼ r1 ðtÞ; t 2 ½0; T ;
y 0 ðtÞ þ MyðtÞ ¼ r2 ðtÞ; t 2 ½0; T ;
with r1 6 r2, and x(0) 6 y(0). Then x(t) 6 y(t), t 2 [0, T]. An analogous result is valid for .

Proof. Since x and y are solutions to the corresponding fuzzy differential equations, its expressions can be
extracted from Theorem 1 [32],
Z t
xðtÞ ¼ xð0ÞvfeMt g þ r1 ðsÞvfeMðstÞ g ds; t 2 ½0; T ;
0
Z t
yðtÞ ¼ yð0ÞvfeMt g þ r2 ðsÞvfeMðstÞ g ds; t 2 ½0; T ;
0

and the conclusion follows from the hypotheses and properties in Lemmas 3–5. h
In Theorem 5 and the results included below, r1 and r2 are fuzzy functions, that is, r1(t), r2(t) 2 E1 for every
t 2 [0, T], and r1, r2 are continuous fuzzy functions, r1, r2 2 C([0, T], E1). In fact, it is easy to check that if
T > 0 and x 2 C1([0, T], E1) is a solution to one of the following fuzzy differential equations x 0 (t) +
Mx(t) = r1(t), t 2 [0, T], x 0 (t) = Mx(t) + r1(t), t 2 [0, T], x 0 (t) = Mx(t) + r1(t), t 2 [0, T], or x 0 (t)  Mx(t) =
r1(t), t 2 [0, T], where M 2 R, M > 0, and r1 is a fuzzy function, then r1 is necessarily a continuous fuzzy func-
tion, that is, r1 2 C([0, T], E1).
It is also possible to prove a comparison result where condition r1 6 r2 is not necessary to obtain x 6 y,
and x(0) 6 y(0) does not have to be imposed a priori.
Theorem 6. Let x, y 2 C1([0, T], E1), T > 0, u0, v0 2 E1, and M 2 R, M > 0, be such that
x0 ðtÞ þ MxðtÞ ¼ r1 ðtÞ; t 2 ½0; T ;
0
y ðtÞ þ MyðtÞ ¼ r2 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ; yð0Þ ¼ v0 ;
and suppose that the following conditions hold
Z t
ðu0 Þal  ðv0 Þal 6 ðr2 ðsÞal  r1 ðsÞal ÞeMs ds; t 2 ½0; T ; a 2 ½0; 1;
0
Z t
ðu0 Þar  ðv0 Þar 6 ðr2 ðsÞar  r1 ðsÞar ÞeMs ds; t 2 ½0; T ; a 2 ½0; 1:
0

Then x(t) 6 y(t), t 2 [0, T].

Proof. The proof is completed using Theorem 1 [32]. h

Example 4. Let r(t) = (t; 1, 1) 2 E1 for every t, that is,

8
< r  t þ 1;
> r 2 ½t  1; t;
rðtÞðrÞ ¼ r þ t þ 1; r 2 ½t; t þ 1;
>
:
0; otherwise;
1766 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779

with level sets given by [r(t)]a = [a + t  1, a + t + 1] for every a 2 [0, 1], and consider T = 10, r1(t) = r(t),
t 2 [0, 10], and r2(t) = r(T  t) = r(10  t), t 2 [0, 10]. Note that r1 and r2 are not 6-comparable, since
r1 ð0Þ ¼ rð0Þ ¼ ð0; 1; 1Þ < r2 ð0Þ ¼ rð10Þ ¼ ð10; 1; 1Þ;
and
r1 ð10Þ ¼ rð10Þ ¼ ð10; 1; 1Þ > r2 ð10Þ ¼ rð0Þ ¼ ð0; 1; 1Þ:
On the other hand, take u0 = r(0), and v0 = r(50) = (50; 1, 1), with level sets [u0]a = [a  1, a + 1],
[v0]a = [a + 49, a + 51], for every a 2 [0, 1], and choose M ¼ 101 . We check that conditions in Theorem 6 hold.
Indeed, note that
r2 ðsÞal  r1 ðsÞal ¼ a þ 10  s  1  ða þ s  1Þ ¼ 10  2s ¼ r2 ðsÞar  r1 ðsÞar 8a 8s;
then
Z t Z t
eMt 10 2 Mt
ðr2 ðsÞal  r1 ðsÞal ÞeMs ds ¼ ð10  2sÞeMs ds ¼ ð10  2tÞ  þ ðe  1Þ; t 2 ½0; 10; a 2 ½0; 1:
0 0 M M M2

Consider, for M and a 2 [0, 1] fixed, the function

 
1 2
uðtÞ ¼ ð10  2tÞeMt  10 þ ðeMt  1Þ ; t 2 ½0; 10:
M M

It is easy to check that u(0) = 0, u is non-decreasing in (0, 5) and non-increasing in (5, 10). Replacing M by its
value 101 , we obtain u(10) = 100(e  3). Moreover,
Z t
1
ðu0 Þal  ðv0 Þal ¼ ða  1Þ  ða þ 49Þ ¼ 50 < 100ðe  3Þ 6 ðr2 ðsÞal  r1 ðsÞal Þe10s ds; t 2 ½0; 10; a 2 ½0; 1;
0

and
Z t
1
ðu0 Þar  ðv0 Þar ¼ ða þ 1Þ  ða þ 51Þ ¼ 50 < 100ðe  3Þ 6 ðr2 ðsÞar  r1 ðsÞar Þe10s ds
0
Z t
1
¼ 10s
ð10  2sÞe ds; t 2 ½0; 10; a 2 ½0; 1:
0

To justify the existence of a unique solution to problems

(
x0 ðtÞ þ MxðtÞ ¼ r1 ðtÞ; t 2 ½0; 10;
xð0Þ ¼ u0 ;
and

y 0 ðtÞ þ MyðtÞ ¼ r2 ðtÞ; t 2 ½0; 10;
yð0Þ ¼ v0 ;

we apply Theorem 1 [32]. Using that M 6 1, we deduce the validity of condition (3) in Remark 1 [32],
 Z t 
a a a
diamð½r1 ðtÞ Þ ¼ 2a þ 2 P MeMt diamð½u0  Þ þ diamð½r1 ðsÞ ÞeMs ds
0

¼ eMt ðð2a þ 2ÞðM þ eMt  1ÞÞ 8a 8t 2 ½0; 10;

and analogously for r2 and v0. It is also easy to prove that, since 0 < M < 1, condition (4) in Remark 1 [32]
holds.
By the comparison result Theorem 6, such solutions x and y satisfy the inequality x(t) 6 y(t), for every
t 2 [0, 10].
 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779 1767

Theorem 7. Let x, y 2 C1([0, T], E1), T > 0, u0, v0 2 E1, and M 2 R, M > 0, be such that
x0 ðtÞ þ MxðtÞ ¼ r1 ðtÞ; t 2 ½0; T ;
y 0 ðtÞ þ MyðtÞ ¼ r2 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ; yð0Þ ¼ v0 ;
and suppose that the following conditions hold
Z t
ðu0 Þal  ðv0 Þal P ðr2 ðsÞal  r1 ðsÞal ÞeMs ds; t 2 ½0; T ; a 2 ½0; 1;
0
Z t
ðu0 Þar  ðv0 Þar 6 ðr2 ðsÞar  r1 ðsÞar ÞeMs ds; t 2 ½0; T ; a 2 ½0; 1:
0

Then x(t)  y(t), t 2 [0, T].

Proof. The proof is finished applying Theorem 1 [32]. h

Example 5. Let T > 0, and consider the triangular fuzzy number r(t) = (0; 1 + t, 1 + t) 2 E1, for every
t 2 [0, T], whose level sets are given by [r(t)]a = [(a  1)(1 + t), (1  a)(1 + t)] for every a 2 [0, 1], and
t 2 [0, T]. Take r1(t) = r(t), t 2 [0, T], and r2(t) = r(T  t), t 2 [0, T]. It is clear that r1 and r2 are not
-comparable.
Consider M > 0 such that MTeMT  eMT + 1  M < 0 (this condition is trivially satisfied, for instance, if
T = 1 and 0 < M < 1), and take
T T 2
l ¼  eMT  þ 2 ðeMT  1Þ;
M M M
which is negative since d(x) = xex  x + 2(ex  1) < 0, for x > 0.
We choose the initial conditions u0 = v{0}, and the triangular fuzzy number v0 = (0; l, l), with level sets
[v0]a = [(a  1)(l), (1  a)(l)], for every a 2 [0, 1].
For this choice, conditions in Theorem 7 hold. Indeed,
r2 ðsÞal  r1 ðsÞal ¼ ða  1Þð1 þ T  sÞ  ða  1Þð1 þ sÞ ¼ ða  1ÞðT  2sÞ 8a 8s;
and
r2 ðsÞar  r1 ðsÞar ¼ ð1  aÞðT  2sÞ 8a 8s:
Therefore,
Z t  
Ms eMt T 2 Mt
ðr2 ðsÞal  r1 ðsÞal Þe ds ¼ ða  1Þ ðT  2tÞ  þ ðe  1Þ ¼ ða  1ÞaðtÞ;
0 M M M2
and
Z t
ðr2 ðsÞar  r1 ðsÞar ÞeMs ds ¼ ð1  aÞaðtÞ;
0

where
eMt T 2
aðtÞ ¼ ðT  2tÞ  þ 2 ðeMt  1Þ; t 2 ½0; T :
M M M
We check that
ðu0 Þal  ðv0 Þal P ða  1ÞaðtÞ; ðu0 Þar  ðv0 Þar 6 ð1  aÞaðtÞ; t 2 ½0; T ; a 2 ½0; 1:
It is easily obtained that a(0) = 0, a(T) = l < 0, and a is non-decreasing in 0; T2 and non-increasing in T
2
;T .
Hence
 
T
0 > l ¼ aðT Þ 6 aðtÞ 6 a 8t 2 ½0; T ;
2
and, in consequence,
1768 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779

Z t
ðu0 Þal  ðv0 Þal ¼ ða  1Þl ¼ ða  1ÞaðT Þ P ða  1ÞaðtÞ ¼ ðr2 ðsÞal  r1 ðsÞal ÞeMs ds; t 2 ½0; T ;
0
a 2 ½0; 1;
Z t
ðu0 Þar  ðv0 Þar ¼ ð1  aÞl ¼ ð1  aÞaðT Þ 6 ð1  aÞaðtÞ ¼ ðr2 ðsÞar  r1 ðsÞar ÞeMs ds; t 2 ½0; T ;
0
a 2 ½0; 1:
The choice
( of M and T allows to affirm, using Theorem 1 [32], that problems
x0 ðtÞ þ MxðtÞ ¼ r1 ðtÞ; t 2 ½0; 10;
xð0Þ ¼ u0 ;
and
(
y 0 ðtÞ þ MyðtÞ ¼ r2 ðtÞ; t 2 ½0; 10;
yð0Þ ¼ v0
have a unique solution. Indeed, it is obvious that
 Z t 
a Mt a a Ms
diamð½r1 ðtÞ Þ ¼ ð1 þ tÞð2  2aÞ P Me diamð½u0  Þ þ diamð½r1 ðsÞ Þe ds
0
Z t  
1
¼ MeMt ð1 þ sÞð2  2aÞeMs ds ¼ eMt ð2  2aÞ ð1 þ tÞeMt  1  ðeMt  1Þ
0 M
8a 8t 2 ½0; T :
For r2 and v0, we get, for every a 2 [0, 1], and t 2 [0, T],
 Z t 
diamð½r2 ðtÞa Þ ¼ ð1 þ T  tÞð2  2aÞ P MeMt diamð½v0 a Þ þ diamð½r2 ðsÞa ÞeMs ds
0
 Z t 
¼ MeMt ð2  2aÞðlÞ þ ð2  2aÞ ð1 þ T  sÞeMs ds
0
 
Mt 1
¼ e ð2  2aÞ Ml þ ð1 þ T  tÞeMt  ð1 þ T Þ þ ðeMt  1Þ ;
M
1 Mt
where we have used that f ðtÞ ¼ Ml  ð1 þ T Þ þ M ðe  1Þ 6 0 for every t 2 [0, T]. Note that f is non-
decreasing and f(T) 6 0, which is evident from the expression of l and the choice of T and M. In consequence,
condition (3) in Remark 1 [32] holds.
It can also be proved that (4) in Remark 1 [32] holds for both problems.
By the comparison result Theorem 7, x(t)  y(t), for every t 2 [0, T].
Taking into account Theorem 2 [32], we prove the following comparison result.
Theorem 8. Let x, y 2 C1([0, T], E1), u0, v0 2 E1, and M 2 R, M > 0, be such that
x0 ðtÞ ¼ MxðtÞ þ r1 ðtÞ; t 2 ½0; T ;
y 0 ðtÞ ¼ MyðtÞ þ r2 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ; yð0Þ ¼ v0 ;
where, for every t 2 [0, T], a 2 [0, 1],
Z
1 1 t 1
 diamð½u0 a ÞeMt  diamð½r1 ðsÞa ÞeMðtsÞ ds þ ððu0 Þal þ ðu0 Þar ÞeMt
2 2 0 2
Z t Z
1 1 1 t
þ r1 ðsÞal þ r1 ðsÞar eMðtsÞ ds 6  diamð½v0 a ÞeMt  diamð½r2 ðsÞa ÞeMðtsÞ ds
2 0 2 2 0
Z
1 1 t
þ ððv0 Þal þ ðv0 Þar ÞeMt þ ðr2 ðsÞal þ r2 ðsÞar ÞeMðtsÞ ds; ð1Þ
2 2 0
 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779 1769

and
Z
1 a 1 t a 1
diamð½u0  ÞeMt þ diamð½r1 ðsÞ ÞeMðtsÞ ds þ ððu0 Þal þ ðu0 Þar ÞeMt
2 2 0 2
Z Z
1 t 1 a 1 t a
þ ðr1 ðsÞal þ r1 ðsÞar ÞeMðtsÞ ds 6 diamð½v0  ÞeMt þ diamð½r2 ðsÞ ÞeMðtsÞ ds
2 0 2 2 0
Z
1 1 t
þ ððv0 Þal þ ðv0 Þar ÞeMt þ ðr2 ðsÞal þ r2 ðsÞar ÞeMðtsÞ ds: ð2Þ
2 2 0
Then x(t) 6 y(t), t 2 [0, T].

Proof. Using Theorem 2 [32], we get, for all t 2 [0, T], and a 2 [0, 1],
eMt eMt
xðtÞal ¼  U 1 ðt; aÞ þ U 2 ðt; aÞ; ð3Þ
2 2
eMt eMt
xðtÞar ¼ U 1 ðt; aÞ þ U 2 ðt; aÞ; ð4Þ
2 2
where
Z t
a a
U 1 ðt; aÞ ¼ diamð½u0  Þ þ diamð½r1 ðsÞ ÞeMs ds;
0
Z t
U 2 ðt; aÞ ¼ ðu0 Þal þ ðu0 Þar þ ðr1 ðsÞal þ r1 ðsÞar ÞeMs ds;
0

and analogously for y. By hypotheses (1) and (2), we obtain that x(t)al 6 y(t)al, x(t)ar 6 y(t)ar for every
t 2 [0, T]. h

Remark 4. Conditions (1) and (2) are equivalent to the following estimate
 Z t 
1  a a a a Ms
 Mt
e
 diamð½v 0  Þ  diamð½u 0  Þ þ ðdiamð½r2 ðsÞ Þ  diamð½r 1 ðsÞ ÞÞe ds 
2 0
 Z t
a a a a
6 mpð½v0  Þ  mpð½u0  Þ þ ðmpð½r2 ðsÞ Þ  mpð½r1 ðsÞ ÞÞeMs ds eMt
0
a
for every t 2 [0, T], and a 2 [0, 1], where mpð½u Þ ¼ 12 ðual þ uar Þ denotes the midpoint of [u]a.

Example 6. Consider T = 10, M = 1, and functions r, r1 and r2 in Example 4, that is, r(t) = (t; 1, 1) 2 E1, for
every t, [r(t)]a = [a + t  1, a + t + 1], for every a 2 [0, 1], r1(t) = r(t), t 2 [0, 10], and r2(t) = r(T  t) =
r(10  t), t 2 [0, 10], where r1 and r2 are not 6-comparable. As initial conditions, we choose the crisp numbers
u0 = v{0} and v0 ¼ vf8e10 þ12g . Then, we get
a a
diamð½r1 ðsÞ Þ ¼ 2  2a ¼ diamð½r2 ðsÞ Þ 8a; s;
a a
mpð½r1 ðsÞ Þ ¼ s; mpð½r2 ðsÞ Þ ¼ 10  s 8a; s:

Hence, the left-hand side of the estimate in Remark 4 is equal to zero, and its right-hand side is equal to
 Z t 
Mt eMt 10 2
10
8e þ 12 þ Ms
ð10  2sÞe ds e ¼ 8e10 þ 12 þ ð10  2tÞ  þ 2 ðeMt  1Þ eMt :
0 M M M
Using that M = 1, we can easily check that

8e10 þ 12 P ð10  2tÞet þ 10  2ðet  1Þ ¼ 12et þ 2tet þ 12 ¼ wðtÞ 8t 2 ½0; 10;

since w(0) = 0, w is non-increasing in (0, 5) and non-decreasing in (5, 10), with w(10) = 8e10 + 12. In conse-
quence, the solutions to
1770 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779
(
x0 ðtÞ ¼ xðtÞ þ r1 ðtÞ; t 2 ½0; 10;
xð0Þ ¼ vf0g ;
(
y 0 ðtÞ ¼ yðtÞ þ r2 ðtÞ; t 2 ½0; 10;
yð0Þ ¼ vf8e10 þ12g
satisfy, by Theorem 8 and Remark 4, that x(t) 6 y(t) for every t 2 [0, 10].
For an example with non-crisp initial conditions, take the triangular fuzzy numbers u0 = (0;1, 1), and
v0 = (8e10 + 12; 1, 1), which satisfy
a a
diamð½u0  Þ ¼ 2  2a ¼ diamð½v0  Þ 8a;
a a
mpð½u0  Þ ¼ 0; mpð½v0  Þ ¼ 8e10 þ 12 8a:

Corollary 1. If u0 6 v0 and r1 6 r2, then condition

Z t Z t
diamð½u0 a Þ þ diamð½r1 ðsÞa ÞeMs ds ¼ diamð½v0 a Þ þ diamð½r2 ðsÞa ÞeMs ds ð5Þ
0 0

for every t 2 [0, T], a 2 [0, 1], implies the validity of (1) and (2) in Theorem 8.

Proof. By hypothesis, (u0)al 6 (v0)al, (u0)ar 6 (v0)ar for every a 2 [0, 1], and (r1(s))al 6 (r2(s))al, (r1(s))ar 6
(r2(s))ar for every s 2 [0, T], and a 2 [0, 1], so that

ððu0 Þal þ ðu0 Þar ÞeMt 6 ððv0 Þal þ ðv0 Þar ÞeMt ;
ððr1 ðsÞÞal þ ðr1 ðsÞÞar ÞeMs 6 ððr2 ðsÞÞal þ ðr2 ðsÞÞar ÞeMs ; s 2 ½0; T ;

and (5) implies (1) and (2). This result can also be obtained by checking that inequality in Remark 4 trivially
holds under the hypotheses of the Corollary, since its right-hand side is greater or equal to zero and its left-
hand side is equal to zero. h

Corollary 2. If u0 6 v0, r1 6 r2, diam([u0]a) = diam([v0]a), and

a a
diamð½r1 ðtÞ Þ ¼ diamð½r2 ðtÞ Þ 8t 2 ½0; T 
for all a 2 [0, 1], then conditions (1) and (2) in Theorem 8 are valid.

Theorem 9. Let x, y 2 C1([0, T], E1), u0, v0 2 E1, and M 2 R, M > 0, be such that
x0 ðtÞ ¼ MxðtÞ þ r1 ðtÞ; t 2 ½0; T ;
y 0 ðtÞ ¼ MyðtÞ þ r2 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ; yð0Þ ¼ v0 ;
where (2) holds, and the reverse inequality of (1) holds. Then x(t)  y(t), t 2 [0, T].

Proof. In this case, we obtain that y(t)al 6 x(t)al, x(t)ar 6 y(t)ar for all t 2 [0, T], a 2 [0, 1]. h

Remark 5. Estimates in Theorem 9 are equivalent to the following estimate

 Z t 
  Mt
mpð½v0 a Þ  mpð½u0 a Þ þ ðmpð½r ðsÞ
a
Þ  mpð½r ðsÞ
a
ÞÞe Ms
dse
 2 1 
0
 Z t
1 a a a a
6 diamð½v0  Þ  diamð½u0  Þ þ ðdiamð½r2 ðsÞ Þ  diamð½r1 ðsÞ ÞÞeMs ds eMt
2 0

for every t 2 [0, T], a 2 [0, 1].

 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779 1771

Example 7. Let T > 0. Consider r1(t) = (t + 1; 1, 1) for every t 2 [0, T], whose level sets are [r1(t)]a =
[a + (t + 1)  1, a + (t + 1) + 1] = [a + t, a + t + 2], for every a 2 [0, 1]. Besides, take r2 given by
(
a ½2a þ t  1; 3 þ t  2a; if 34 6 a 6 1;
½r2 ðtÞ ¼ a 
3
þ t þ 14 ;  a3 þ t þ 74 ; if 0 6 a < 34 ;

which is obtained by shifting the fuzzy number r2(0) in Fig. 4. Note that r1 and r2 are not -comparable,
since, for t = 0, [r1(0)]a = [a, a + 2], and
(
a ½2a  1; 3  2a; if 34 6 a 6 1;
½r2 ð0Þ ¼ a 1 
3
þ 4 ;  a3 þ 74 ; if 0 6 a < 34 ;
a
hence, for a close to 1 (a < 1), [r1(0)]a = [a, a + 2] [2a  1, 3  2a] = [r2(0)]a, but, for a close to 0, ½r2 ð0Þ ¼
a 1  a
3
þ 4 ;  a3 þ 74 ½a; a þ 2 ¼ ½r1 ð0Þ . For the above-defined functions r1 and r2, diam([r1(s)]a) = 2  2a,
(
a 2ð2  2aÞ; if 34 6 a 6 1;
diamð½r2 ðsÞ Þ ¼
 2a3 þ 32 ; if 0 6 a < 34 ;

and mp([r1(s)]a) = s + 1 = mp([r2(s)]a), for every a 2 [0, 1], and s 2 [0, T].
Take M = 1, and choose two initial conditions u0 and v0 whose level sets have the same midpoint, that is,
mp([u0]a) = mp([v0]a), for each a, and such that diamð½u0 a Þ  diamð½v0 a Þ 6  12 ð1  eT Þ, for instance
u0 = v{0} and v0 ¼ v½1ð1eT Þ;1ð1eT Þ .
4 4
Note that
Z t
a a
ðdiamð½r2 ðsÞ Þ  diamð½r1 ðsÞ ÞÞes ds ¼ KðaÞð1  et Þ;
0

where
(
2  2a; if 34 6 a 6 1;
KðaÞ ¼
 2a3 þ 32  2 þ 2a ¼ 4a3  12 ; if 0 6 a < 34 :

To check that condition in Remark 5 holds, we prove that diam([v0]a)  diam([u0]a) + K(a)(1  et) is non-
negative, that is,
a a
diamð½u0  Þ  diamð½v0  Þ 6 KðaÞð1  et Þ for every t 2 ½0; T ; a 2 ½0; 1:
This is trivially satisfied for 6 a 6 1. Suppose that 0 6 a < 34, then KðaÞ ¼ 4a3  12 is non-decreasing in a, thus
3
4
KðaÞ P Kð0Þ ¼  12, and hence, for all t 2 [0, T], and a 2 [0, 1],
a a 1 1
diamð½u0  Þ  diamð½v0  Þ 6  ð1  eT Þ 6  ð1  et Þ 6 KðaÞð1  et Þ:
2 2

σ2(0)

3/4

σ1(0)=(1;1,1)

1/4 1 7/4 2

Fig. 4. r1(0), r2(0).

1772 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779

By Theorem 9 and Remark 5, we get that x(t)  y(t), t 2 [0, T], where x and y are, respectively, the solutions to
problems
(
x0 ðtÞ ¼ xðtÞ þ r1 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ;
and
(
y 0 ðtÞ ¼ yðtÞ þ r2 ðtÞ; t 2 ½0; T ;
yð0Þ ¼ v0 :

Corollary 3. If u0  v0, r1  r2, and

Z t
1 1
ððu0 Þal þ ðu0 Þar ÞeMt þ ðr1 ðsÞal þ r1 ðsÞar ÞeMðtsÞ ds
2 2 0
Z t
1 1
¼ ððv0 Þal þ ðv0 Þar ÞeMt þ ðr2 ðsÞal þ r2 ðsÞar ÞeMðtsÞ ds ð6Þ
2 2 0

for every t 2 [0, T], a 2 [0, 1], then estimates in Theorem 9 are valid and, thus, the conclusion of Theorem 9
follows.

Proof. By hypothesis, [u0]a  [v0]a for all a 2 [0, 1], and [r1(s)]a  [r2(s)]a for all a 2 [0, 1], and s 2 [0, T], then
diamð½u0 a Þ 6 diamð½v0 a Þ for all a 2 ½0; 1;

and
a a
diamð½r1 ðsÞ Þ 6 diamð½r2 ðsÞ Þ for all a 2 ½0; 1; and all s 2 ½0; T :
These properties, joint to condition (6) easily provide the validity of estimates in Theorem 9 ((2) and the re-
verse of (1)). In other words, it is trivial to check that the estimate in Remark 5 holds (see Remark 6). h

Remark 6. Condition (6) is equivalent to

Z t Z t
a a a a
mpð½u0  Þ þ mpð½r1 ðsÞ ÞeMs ds ¼ mpð½v0  Þ þ mpð½r2 ðsÞ ÞeMs ds
0 0

for every t 2 [0, T], a 2 [0, 1].

Corollary 4. If u0  v0, r1  r2, and for all a 2 [0, 1],

1 1
mpð½u0 a Þ ¼ ððu0 Þal þ ðu0 Þar Þ ¼ ððv0 Þal þ ðv0 Þar Þ ¼ mpð½v0 a Þ;
2 2
and
a 1 1 a
mpð½r1 ðtÞ Þ ¼ ðr1 ðtÞal þ r1 ðtÞar Þ ¼ ðr2 ðtÞal þ r2 ðtÞar Þ ¼ mpð½r2 ðtÞ Þ
2 2
for every t 2 [0, T], a 2 [0, 1], then conditions in Theorem 9 are valid.

Theorem 10. Let x, y 2 C1([0, T], E1), T > 0, u0, v0 2 E1, and M 2 R, M > 0, be such that
x0 ðtÞ ¼ MxðtÞ þ r1 ðtÞ; t 2 ½0; T ;
0
y ðtÞ ¼ MyðtÞ þ r2 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ; yð0Þ ¼ v0 ;

with r1 6 r2, and u0 6 v0. Then x(t) 6 y(t), t 2 [0, T]. An analogous result is valid for .
 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779 1773

Proof. Applying Theorem 3 [32], we deduce that

Z t
xðtÞ ¼ u0 vfeMt g þ r1 ðsÞvfeM ðtsÞ g ds; t 2 ½0; T ;
0
Z t
yðtÞ ¼ v0 vfeMt g þ r2 ðsÞvfeMðtsÞ g ds; t 2 ½0; T ;
0
and the conclusion follows again from the hypotheses and properties in Lemmas 3–5. h
We show that hypothesis r1 6 r2 can be eliminated and that it is not necessary to impose a priori the esti-
mate u0 6 v0, obtaining the following comparison results.
Theorem 11. Let x, y 2 C1([0, T], E1), T > 0, u0, v0 2 E1, and M 2 R, M > 0, be such that
x0 ðtÞ ¼ MxðtÞ þ r1 ðtÞ; t 2 ½0; T ;
0
y ðtÞ ¼ MyðtÞ þ r2 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ; yð0Þ ¼ v0 ;
and suppose that
Z t
ðu0 Þal  ðv0 Þal 6 ðr2 ðsÞal  r1 ðsÞal ÞeMs ds;
Z0 t
ðu0 Þar  ðv0 Þar 6 ðr2 ðsÞar  r1 ðsÞar ÞeMs ds
0
for t 2 [0, T], a 2 [0, 1].
Then x(t) 6 y(t), t 2 [0, T].

Proof. It is deduced from the expressions of x and y, given in Theorem 3 [32]. h

Analogously for the partial ordering .
Theorem 12. Let x, y 2 C1([0, T], E1), T > 0, u0, v0 2 E1, and M 2 R, M > 0, be such that
x0 ðtÞ ¼ MxðtÞ þ r1 ðtÞ; t 2 ½0; T ;
y 0 ðtÞ ¼ MyðtÞ þ r2 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ; yð0Þ ¼ v0 ;
and suppose that
Z t
ðu0 Þal  ðv0 Þal P ðr2 ðsÞal  r1 ðsÞal ÞeMs ds;
Z 0t
ðu0 Þar  ðv0 Þar 6 ðr2 ðsÞar  r1 ðsÞar ÞeMs ds
0
for t 2 [0, T], a 2 [0, 1].
Then x(t)  y(t), t 2 [0, T].

Example 8. Consider T > 0, M > 0, and the same functions defined in Example 4, with level sets
½r1 ðtÞa ¼ ½a þ t  1; a þ t þ 1; ½r2 ðtÞa ¼ ½a þ ðT  tÞ  1; a þ ðT  tÞ þ 1
for every a 2 [0, 1], t 2 [0, T]. Take initial conditions u0 and v0 in E1 such that u0 6 v0. Then estimates in The-
orem 11 hold, since (u0)al  (v0)al 6 0, (u0)ar  (v0)ar 6 0 for every a, and
r2 ðsÞal  r1 ðsÞal ¼ r2 ðsÞar  r1 ðsÞar ¼ T  2s 8a 2 ½0; 1 8s 2 ½0; T ;
and, hence
Z t Z t
1
ðr2 ðsÞal  r1 ðsÞal ÞeMs ds ¼ ðr2 ðsÞar  r1 ðsÞar ÞeMs ds ¼ 2 ðMðT  2tÞeMt þ MT þ 2ðeMt  1ÞÞ:
0 0 M
1774 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779

Taking b(t) = M(T  2t)eMt + MT + 2(eMt  1), t 2 [0, T], we get that b(t) P min{b(0), b(T)} =
min{0, b(T)}, for t 2 [0, T]. Proving that b(T) > 0, we obtain that b(t) P 0 for every t 2 [0, T].
This implies that the unique solution to problem
 0
x ðtÞ ¼ MxðtÞ þ r1 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ;
and the unique solution to problem
 0
y ðtÞ ¼ MyðtÞ þ r2 ðtÞ; t 2 ½0; T ;
yð0Þ ¼ v0
satisfy that x(t) 6 y(t) for every t 2 [0, T].
Now, for M > 0, T > 0, and the same functions r1 and r2, consider the initial conditions u0, v0 2 E1 such
that (u0)ar 6 (v0)ar, and
  
1 T 1 T
ðu0 Þal  ðv0 Þal P 2 MT þ 2ðeM 2  1Þ ¼ 2 b for every a 2 ½0; 1:
M M 2
For instance, u0 = v{0} and v0 ¼ vh i are admissible initial conditions. The pair of fuzzy
M T
 1 MT þ2 e 2 1 ;0
M2
T
numbers shown in Fig. 5 is also admissible, where C P M12 MT þ 2 eM 2  1 . Applying Theorem 12, the
corresponding solutions satisfy that x(t)  y(t) for every t 2 [0, T].

Theorem 13. Let x, y 2 C1([0, T], E1), T > 0, u0, v0 2 E1, and M 2 R, M > 0, be such that
x0 ðtÞ  MxðtÞ ¼ r1 ðtÞ; t 2 ½0; T ;
y 0 ðtÞ  MyðtÞ ¼ r2 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ; yð0Þ ¼ v0 ;
with r1 6 r2, and u0 6 v0, and suppose that the following condition holds
Z t Z t
a a Ms a a
diamð½u0  Þ þ diamð½r1 ðsÞ Þe ds ¼ diamð½v0  Þ þ diamð½r2 ðsÞ ÞeMs ds
0 0

for every t 2 [0, T], a 2 [0, 1]. Then x(t) 6 y(t), t 2 [0, T].

Proof. Using Theorem 4 [32], since x and y are solutions to the corresponding fuzzy differential equations, we
obtain their expressions as
eMt
xðtÞal ¼  W 1 ðt; aÞ þ eMt W 2 ðt; aÞ;
2
eMt
xðtÞar ¼ W 1 ðt; aÞ þ eMt W 2 ðt; aÞ
2

v0

u0
P-C P

Fig. 5. Admissible initial conditions u0  v0.

 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779 1775

for t 2 [0, T], a 2 [0, 1], where

Z t
a a
W 1 ðt; aÞ ¼ diamð½u0  Þ þ diamð½r1 ðsÞ ÞeMs ds;
0
Z t
a a
W 2 ðt; aÞ ¼ mpð½u0  Þ þ mpð½r1 ðsÞ ÞeMs ds;
0

and analogously for y. The hypotheses imply that mp([u0]a) 6 mp([v0]a), and mp([r1(s)]a) 6 mp([r2(s)]a), for
every s 2 [0, T], a 2 [0, 1], and the conclusion follows. h

Example 9. Take T > 0, M > 0, and r1, r2 given by r1(t) = (t; 1, 1), t 2 [0, T], and r2(t) = r1(t + 1) =
(t + 1; 1, 1), t 2 [0, T], that is, [r1(t)]a = [a + t  1, a + t + 1], a 2 [0, 1], t 2 [0, T], and [r2(t)]a = [a + t,
a + t + 2] for every a 2 [0, 1], t 2 [0, T]. Obviously, r1 6 r2. As initial conditions, we choose u0 = (0; 1, 1)
and v0 = (3; 1, 1), where [u0]a = [a  1, a + 1], [v0]a = [a + 2, a + 4] for every a 2 [0, 1], and u0 6 v0. Using
that
diamð½u0 a Þ ¼ 2  2a ¼ diamð½v0 a Þ 8a 2 ½0; 1;
and
a a
diamð½r1 ðsÞ Þ ¼ diamð½r2 ðsÞ Þ ¼ 2  2a 8a 2 ½0; 1; s 2 ½0; T ;
and, applying Theorem 13, we deduce that, if problems
 0
x ðtÞ  MxðtÞ ¼ r1 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ;
and
(
y 0 ðtÞ  MyðtÞ ¼ r2 ðtÞ; t 2 ½0; T ;
yð0Þ ¼ v0
are solvable (see section 2.4 [32]), then x(t) 6 y(t), for t 2 [0, T].

Theorem 14. Let x, y 2 C1([0, T], E1), T > 0, u0, v0 2 E1, and M 2 R, M > 0, be such that
x0 ðtÞ  MxðtÞ ¼ r1 ðtÞ; t 2 ½0; T ;
0
y ðtÞ  MyðtÞ ¼ r2 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ; yð0Þ ¼ v0 ;
with r1  r2, and u0  v0, and suppose that the following condition holds
Z t Z t
a a a a
mpð½u0  Þ þ mpð½r1 ðsÞ ÞeMs ds ¼ mpð½v0  Þ þ mpð½r2 ðsÞ ÞeMs ds
0 0

for every t 2 [0, T], a 2 [0, 1]. Then x(t)  y(t), t 2 [0, T].

Proof. The proof is finished similarly to the proof of Theorem 13, using the hypotheses and the fact that
diam([u0]a) 6 diam([v0]a), and diam([r1(s)]a) 6 diam([r2(s)]a), for every s 2 [0, T], a 2 [0, 1]. h

Example 10. Take T > 0, M > 0, r1 as in Example 9, and r2 given by r2(t) = (t; 2, 2) for t 2 [0, T], that is,
[r1(t)]a = [a + t  1, a + t + 1], a 2 [0, 1], t 2 [0, T], [r2(t)]a = [2(a  1) + t, 2(1  a) + t], for every a 2 [0, 1],
t 2 [0, T]. It is clear that r1  r2. Consider the initial conditions u0 = r1(0) = (0; 1, 1), with [u0]a = [a  1,
a + 1], for every a 2 [0, 1], and v0 = v[1, 1], hence u0  v0. Estimate in Theorem 14 also holds, since
mp([u0]a) = 0 = mp([v0]a) for all a, and mp([r1(s)]a) = mp([r2(s)]a) = s for all a and s. If problems
 0
x ðtÞ  MxðtÞ ¼ r1 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ;
1776 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779

and

y 0 ðtÞ  MyðtÞ ¼ r2 ðtÞ; t 2 ½0; T ;
yð0Þ ¼ v0
have a solution, then x(t)  y(t) for all t 2 [0, T], by application of Theorem 14.
The following Theorems allow to compare x and y without imposing a priori the conditions [r1 6 r2,
u0 6 v0] or [r1  r2, u0  v0].
Theorem 15. Let x, y 2 C1([0, T], E1), T > 0, u0, v0 2 E1, and M 2 R, M > 0, be such that
x0 ðtÞ  MxðtÞ ¼ r1 ðtÞ; t 2 ½0; T ;
0
y ðtÞ  MyðtÞ ¼ r2 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ; yð0Þ ¼ v0 ;
and suppose that the following condition holds
 Z t 
eMt  
ðdiamð½r2 ðsÞ Þ  diamð½r1 ðsÞ ÞÞe ds
a a a a Ms
 diamð½v0  Þ  diamð½u0  Þ þ
2 0
 Z t 
a a a a
6 eMt mpð½v0  Þ  mpð½u0  Þ þ ðmpð½r2 ðsÞ Þ  mpð½r1 ðsÞ ÞÞeMs ds
0

for every t 2 [0, T], a 2 [0, 1]. Then x(t) 6 y(t), t 2 [0, T].

Proof. It is deduced from the expressions for x, y given in Theorem 4 [32]. h

Example 11. Take T > 0, M > 0, and consider functions r1 and r2 similar to those in Examples 4, 6, and 8,
that is, r1(t) = (t; 1, 1) 2 E1, for t 2 [0, T], [r1(t)]a = [a + t  1, a + t + 1] for every a 2 [0, 1], t 2 [0, T], and
r2(t) = r(T  t), t 2 [0, T]. For this choice, diam([r1(s)]a) = 2  2a = diam([r2(s)]a) for all a and s, and
mp([r1(s)]a) = s, mp([r2(s)]a) = T  s for all a and s.
We choose initial conditions u0 6 v0 such that diam([u0]a) = diam([v0]a), and mp([u0]a) 6 mp([v0]a) for each
a. We can take, for instance, u0 = v{0} and v0 = v{1}. The fuzzy numbers u0 = v[1, 1] and v0 = v[0, 2] are also
admissible initial conditions. Moreover, estimate in Theorem 15 holds, since
Z t
a a
0 6 mpð½v0  Þ  mpð½u0  Þ þ ðT  2sÞeMs ds
0

a 1 a
¼ mpð½v0  Þ  mpð½u0  Þ þ 2 ðMðT  2tÞeMt þ MT þ 2ðeMt  1ÞÞ
M
a a 1
¼ mpð½v0  Þ  mpð½u0  Þ þ 2 bðtÞ;
M
where b is defined in Example 8. By Theorem 15, x(t) 6 y(t) "t 2 [0, T], if there exist x and y satisfying,
respectively,
 0
x ðtÞ  MxðtÞ ¼ r1 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ;
and

y 0 ðtÞ  MyðtÞ ¼ r2 ðtÞ; t 2 ½0; T ;
yð0Þ ¼ v0 :

Theorem 16. Let x, y 2 C1([0, T], E1), T > 0, u0, v0 2 E1, and M 2 R, M > 0, be such that
x0 ðtÞ  MxðtÞ ¼ r1 ðtÞ; t 2 ½0; T ;
y 0 ðtÞ  MyðtÞ ¼ r2 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ; yð0Þ ¼ v0 ;
 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779 1777

and suppose that the following condition holds

 Z t 

Mt 

ðmpð½r2 ðsÞ Þ  mpð½r1 ðsÞ ÞÞe ds
a a a a Ms
e mpð½v0  Þ  mpð½u0  Þ þ
0
Mt
 Z t 
e
6 diamð½v0 a Þ  diamð½u0 a Þ þ ðdiamð½r2 ðsÞa Þ  diamð½r1 ðsÞa ÞÞeMs ds
2 0

for every t 2 [0, T], a 2 [0, 1]. Then x(t)  y(t), t 2 [0, T].

Proof. Analogous to the proof of Theorem 15. h

Example 12. Take T > 0, M > 0, and r given by r(t) = (0; 1 + t, 1 + t) 2 E1 for every t, [r(t)]a = [(a  1)
(1 + t), (1  a)(1 + t)] for every a 2 [0, 1] and t. Define r1(t) = r(t), and r2(t) = r(T  t), t 2 [0, T]. Functions
r1 and r2 are not -comparable, since
a a a a
½r1 ð0Þ ¼ ½rð0Þ ¼ ½a  1; 1  a  ½r2 ð0Þ ¼ ½rðT Þ ¼ ½ða  1Þð1 þ T Þ; ð1  aÞð1 þ T Þ 8a;

with [r1(0)]a [r2(0)]a "a < 1, thus r1(0)  r2(0), but

a a
½r1 ðT Þ ¼ ½ða  1Þð1 þ T Þ; ð1  aÞð1 þ T Þ  ½r2 ðT Þ ¼ ½a  1; 1  a 8a;

where, for a < 1, the inclusion is strict, hence r2(T)  r1(T).

Besides, diam([r1(s)]a) = (1 + s)(2  2a), diam([r2(s)]a) = (1 + T  s)(2  2a), and mp([r1(s)]a) = 0 =
mp([r2(s)]a) for all a and s.
We choose initial conditions u0  v0 such that mp([u0]a) = mp([v0]a), for each a, and
1
diamð½u0 a Þ  diamð½v0 a Þ 6 ð2  2aÞ MTeMT  MT þ 2ðeMT  1Þ 8a: ð7Þ
M2
To check that estimate in Theorem 16 holds, we prove that
 Z t 
eMt a a a a
06 diamð½v0  Þ  diamð½u0  Þ þ ðdiamð½r2 ðsÞ Þ  diamð½r1 ðsÞ ÞÞeMs ds
2 0
Mt
 Z t 
e
¼ diamð½v0 a Þ  diamð½u0 a Þ þ ðT  2sÞð2  2aÞeMs ds
2 0

for every t 2 [0, T], a 2 [0, 1]. This is equivalent to

a a 1
diamð½u0  Þ  diamð½v0  Þ 6 ð2  2aÞ MðT  2tÞeMt  MT þ 2ðeMt  1Þ
M2
for every t 2 [0, T], a 2 [0, 1].
If c(t) = M(T  2t)eMt  MT + 2(eMt  1), t 2 [0, T], c is non-decreasing in 0; T2 and non-increasing in
T
2 ; T , hence c(t) P min{c(0), c(T)}, for t 2 [0, T]. Since c(T) < 0, then
1 1
ð2  2aÞ 2
MTeMT  MT þ 2ðeMT  1Þ 6 ð2  2aÞ 2 cðtÞ; t 2 ½0; T ;
M M
and hypothesis (7) on the initial values implies the above-mentioned estimate. By Theorem 16, x(t)  y(t)
"t 2 [0, T], if there exist x and y satisfying, respectively,

x0 ðtÞ  MxðtÞ ¼ r1 ðtÞ; t 2 ½0; T ;
xð0Þ ¼ u0 ;
and

y 0 ðtÞ  MyðtÞ ¼ r2 ðtÞ; t 2 ½0; T ;
yð0Þ ¼ v0 :
1778 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779

v0

-(1-μ) -1 1 (1-μ)

Fig. 6. Trapezoidal fuzzy number v0.

An admissible pair of initial conditions is u0 = v[1,1] and v0 the trapezoidal fuzzy number (see Fig. 6) given by
8 1
>  l ðr þ 1Þ þ 1; r 2 ½ð1  lÞ; 1Þ;
>
>
< 1; r 2 ½1; 1;
v0 ðrÞ ¼ 1
>
> l ðr  1Þ þ 1;
>
:
r 2 ð1; 1  l;
0; otherwise;
where l coincides with the constant given in Example 5, l ¼  MT eMT  MT þ M22 ðeMT  1Þ < 0.
The level sets of v0 are the following
a
½v0  ¼ ½ð1  ð1  aÞlÞ; 1  ð1  aÞl 8a 2 ½0; 1:
Note that, for this choice, (7) holds, since diam([v0]a) = 2  2(1  a)l, and diam([u0]a)  diam([v0]a) =
2(1  a)l "a.

3. Conclusions

Some classical comparison results for the solution to linear ordinary differential equations have been
extended to the fuzzy context in relation with two different partial orderings. It is also shown how the Huku-
hara difference can be used, under appropriate conditions, to compare two fuzzy functions.
We have obtained results which allow to compare the solutions to ‘linear’ first-order fuzzy differential equa-
tions, with non-necessarily comparable ‘non-homogeneous’ terms. Examples are provided to illustrate the
applicability of the new results, showing the qualitative difference between the establishment of comparison
results in the ordinary case and the corresponding study in the fuzzy case, which is more complex.

Acknowledgement

The author wishes to thank the editor-in-chief and the referees for their helpful and interesting comments.

References

[1] S. Abbasbandy, B. Asady, Note on ‘A new approach for defuzzification’, Fuzzy Sets Syst. 128 (2002) 131–132.
[2] R.P. Agarwal, D. O’Regan, V. Lakshmikantham, A stacking theorem approach for fuzzy differential equations, Nonlinear Anal. 55
(2003) 299–312.
[3] R.P. Agarwal, D. O’Regan, V. Lakshmikantham, Viability theory and fuzzy differential equations, Fuzzy Sets Syst. 151 (2005) 563–
580.
[4] T. Allahviranloo, N. Ahmady, E. Ahmady, Numerical solution of fuzzy differential equations by predictor–corrector method, Inform.
Sci. 177 (7) (2007) 1633–1647.
[5] E. Babolian, H. Sadeghi, S. Javadi, Numerically solution of fuzzy differential equations by Adomian method, Appl. Math. Comput.
149 (2004) 547–557.
[6] B. Bede, S.G. Gal, Generalizations of the differentiability of fuzzy-number-valued functions with applications to fuzzy differential
equations, Fuzzy Sets Syst. 151 (2005) 581–599.
[7] B. Bede, I.J. Rudas, A.L. Bencsik, First order linear fuzzy differential equations under generalized differentiability, Inform. Sci. 177 (7)
(2007) 1648–1662.
 R. Rodrı́guez-López / Information Sciences 178 (2008) 1756–1779 1779

[8] T.G. Bhaskar, V. Lakshmikantham, V. Devi, Revisiting fuzzy differential equations, Nonlinear Anal. 58 (2004) 351–358.
[9] W. Congxin, S. Shiji, Existence theorem to the Cauchy problem of fuzzy differential equations under compactness-type conditions,
Inform. Sci. 108 (1998) 123–134.
[10] E.Y. Deeba, A. de Korvin, Analysis by fuzzy difference equations of a model of CO2 level in the blood, Appl. Math. Lett. 12 (1999)
33–40.
[11] P. Diamond, Time-dependent differential inclusions, cocycle attractors and fuzzy differential equations, IEEE Trans. Fuzzy Syst. 7
(1999) 734–740.
[12] P. Diamond, Stability and periodicity in fuzzy differential equations, IEEE Trans. Fuzzy Syst. 8 (2000) 583–590.
[13] P. Diamond, Brief note on the variation of constants formula for fuzzy differential equations, Fuzzy Sets Syst. 129 (2002) 65–71.
[14] P. Diamond, P.E. Kloeden, Metric Spaces of Fuzzy Sets: Theory and Applications, World Scientific, Singapore, 1994.
[15] D. Dubois, H. Prade, Fuzzy Sets and Systems: Theory and Applications, Academic Press, New York, 1980.
[16] D.N. Georgiou, I.E. Kougias, On Cauchy problems for fuzzy differential equations, Int. J. Math. Math. Sci. 15 (2004) 799–805.
[17] D.N. Georgiou, J.J. Nieto, R. Rodrı́guez-López, Initial value problems for higher-order fuzzy differential equations, Nonlinear Anal.
63 (2005) 587–600.
[18] B. Hlebcar, Forecasting closing overvoltages in high-voltage networks using a fuzzy model, Fuzzy Sets Syst. 118 (2001) 1–8.
[19] E. Hüllermeier, Numerical methods for fuzzy initial value problems, Int. J. Uncertain. Fuzziness Knowledge-Based Syst. 7 (1999)
439–461.
[20] O. Kaleva, Fuzzy differential equations, Fuzzy Sets Syst. 24 (1987) 301–317.
[21] O. Kaleva, The Cauchy problem for fuzzy differential equations, Fuzzy Sets Syst. 35 (1990) 389–396.
[22] V. Křivan, G. Colombo, A non-stochastic approach for modeling uncertainty in population dynamics, Bull. Math. Biol. 60 (1998)
721–751.
[23] G.S. Ladde, V. Lakshmikantham, A.S. Vatsala, Monotone Iterative Techniques for Nonlinear Differential Equations, Pitman,
Boston, 1985.
[24] V. Lakshmikantham, R.N. Mohapatra, Theory of fuzzy differential equations and inclusions, Taylor & Francis, London, 2003.
[25] V. Lakshmikantham, J.J. Nieto, Differential equations in metric spaces: an introduction and an application to fuzzy differential
equations, Dyn. Contin. Discrete Impuls. Syst. Ser. A Math. Anal. 10 (2003) 991–1000.
[26] M. Ma, M. Friedman, A. Kandel, Numerical solutions of fuzzy differential equations, Fuzzy Sets Syst. 105 (1999) 133–138.
[27] M. Ma, A. Kandel, M. Friedman, A new approach for defuzzification, Fuzzy Sets Syst. 111 (2000) 351–356.
[28] M. Ma, A. Kandel, M. Friedman, Correction to ‘A new approach for defuzzification’, Fuzzy Sets Syst. 128 (2002) 133–134.
[29] M.T. Mizukoshi, L.C. Barros, Y. Chalco-Cano, H. Román-Flores, R.C. Bassanezi, Fuzzy differential equations and the extension
principle, Information Sciences, in press, doi: 10.1016/j.ins.2007.02.039.
[30] J.J. Nieto, The Cauchy problem for continuous fuzzy differential equations, Fuzzy Sets Syst. 102 (1999) 259–262.
[31] J.J. Nieto, R. Rodrı́guez-López, Bounded solutions for fuzzy differential and integral equations, Chaos Solitons Fract. 27 (2006)
1376–1386.
[32] J.J. Nieto, R. Rodrı́guez-López, D. Franco, Linear first-order fuzzy differential equations, Int. J. Uncertainty, Fuzziness Knowledge-
Based Syst. 14 (2006) 687–709.
[33] D. O’Regan, V. Lakshmikantham, J.J. Nieto, Initial and boundary value problems for fuzzy differential equations, Nonlinear Anal.
54 (2003) 405–415.
[34] G. Papaschinopoulos, G. Stefanidou, P. Efraimidis, Existence, uniqueness and asymptotic behavior of the solutions of a fuzzy
differential equation with piecewise constant argument, Inform. Sci. 177 (2007) 3855–3870.
[35] C.-W. Park, C.-H. Lee, M. Park, Design of an adaptive fuzzy model based controller for chaotic dynamics in Lorenz systems with
uncertainty, Inform. Sci. 147 (2002) 245–266.
[36] L.J. Rodrı́guez-Muñiz, M. López-Dı́az, M.A. Gil, D.A. Ralescu, The s-differentiability of a fuzzy-valued mapping, Inform. Sci. 151
(2003) 283–299.
[37] H. Román-Flores, M. Rojas-Medar, Embedding of level-continuous fuzzy sets on Banach spaces, Inform. Sci. 144 (2002) 227–247.
[38] L. Stefanini, L. Sorini, M.L. Guerra, Parametric representation of fuzzy numbers and application to fuzzy calculus, Fuzzy Sets Syst.
157 (2006) 2423–2455.
[39] C. Wu, G. Wang, Convergence of sequences of fuzzy numbers and fixed point theorems for increasing fuzzy mappings and
application, Fuzzy Sets Syst. 130 (2002) 383–390.
[40] D. Zhang, X.-L. Bai, K.-Y. Cai, Extended neuro-fuzzy models of multilayer perceptrons, Fuzzy Sets Syst. 142 (2004) 221–242.