Nonlinear Analysis 54 (2003) 405 – 415

www.elsevier.com/locate/na

Initial and boundary value problems for fuzzy

di#erential equations
Donal O’Regana;∗ , V. Lakshmikanthamb , Juan J. Nietoc
a Department of Mathematics, National University of Ireland, Galway, Ireland
b Department of Mathematical Sciences, Florida Institute of Technology,
Melbourne, FL 32901-6975, USA
c Departamento de An+alisis Matem+atico, Facultad de Matem+aticas,
Universidad de Santiago de Compostela, Spain

Received 13 December 2002; accepted 10 January 2003

Abstract
We prove a Wintner-type result for fuzzy IVPs and a superlinear-type result for fuzzy BVPs.
Both results rely on a generalized Schauder theorem in metric spaces.
? 2003 Elsevier Science Ltd. All rights reserved.

Keywords: Fuzzy di#erential equations; Generalized Schauder theorem

1. Introduction

This paper presents a new existence result of Wintner-type for the fuzzy initial value
problem



x (t) = f(t; x(t)) for t ∈ [0; T ] (here T ¿ 0 is >xed);
(1.1)
x(0) = x0 ∈ E n
and a new existence result of superlinear type for the fuzzy boundary value problem




x (t) = f(t; x(t)) for t ∈ [0; 1];
(1.2)
x(0) = x(1) = 0̂ ∈ E n ;

∗ Corresponding author. Department of Mathematics, University College Galway, Galway, Ireland.

Fax: +44-353-91-525700.
E-mail address: [email protected] (D. O’Regan).

0362-546X/03/$ - see front matter ? 2003 Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0362-546X(03)00097-X
406 D. O’Regan et al. / Nonlinear Analysis 54 (2003) 405 – 415

where F : [0; a]×E n → E n is continuous (here a=T for (1.1) and a=1 for (1.2)). Our
results in particular improve those in [7,8,10] since a more general growth condition
is placed on the nonlinearity f (in [8,10] f was assumed to be bounded). We also
note here that an extra assumption (see Theorems 2.3 and 2.5) is needed to guarantee
existence in [8,10].
For the remainder of this section we gather together some notation and results from
the literature. Let C(Rn ) denote the family of nonempty, convex, compact subsets of
Rn . For A and B two nonempty bounded subsets of a Banach space (Y; :), we de>ne
the Hausdor# distance between A and B as
 
dH (A; B) = max sup d(a; B); sup d(b; A) ;
a∈A b∈B

where d(a; B) = inf b∈B a − b.

A fuzzy set u ∈ E n is a function u: Rn → [0; 1] for which
(i) u is normal;
(ii) u is fuzzy convex;
(iii) u is upper semicontinuous;
and
(iv) the closure of {x ∈ Rn : u(x) ¿ 0}, denoted [u]0 , is compact.
For 0 ¡  6 1 the -level set [u] is de>ned by

[u] = {x ∈ Rn : u(x) ¿ }:

Clearly, [u] ∈ C(Rn ) for  ∈ [0; 1]. De>ne D : E n × E n → [0; ∞) by

D(u; v) = sup{dH ([u] ; [v] ):  ∈ [0; 1]}:

We know (E n ; D) is a complete metric space and for all u; v; w ∈ E n and  ∈ R we have

D(u + w; v + w) = D(u; v) and D(u; v) = ||D(u; v): (1.3)

We de>ne 0̂ ∈ E n as 0̂(x) = 1 if x = 0 and 0̂(x) = 0 if x = 0. It is well known [11] that

(E n ; D) can be embedded isometrically as a cone C in a Banach space X , i.e. there
exists an embedding j : E n → X (see also [5, Theorem 2.1]) de>ned by

j(u) = u; 0̂ where u ∈ E n ; (1.4)

here :; : is de>ned in [11, p. 555]. Notice also that

 u; v X = D(u; v) for u; v ∈ E n ; (1.5)

so in particular

juX = D(u; 0̂) for u ∈ E n : (1.6)

 D. O’Regan et al. / Nonlinear Analysis 54 (2003) 405 – 415 407

Let T ¿ 0 be >xed. By C([0; T ]; E n ) we mean the set of continuous functions from

[0; T ] to E n . We metricize C([0; T ]; E n ) by setting
H (u; v) = sup D(u(t); v(t))
t∈[0;T ]

for u; v ∈ C([0; T ]; E n ). It is well known that C([0; T ]; E n ) is a complete metric space.

Now since j : E n → C ⊆ X we can de>ne a map J : C([0; T ]; E n ) → C([0; T ]; X ) by
[Jx](t) = j(x(t)) = jx(t) for t ∈ [0; T ];
here x ∈ C([0; T ]; E n ) (note if x ∈ C([0; T ]; E n ) and t0 ; t ∈ [0; T ] then (1.4) implies
[Jx](t) − [Jx](t0 )X = jx(t) − jx(t0 )X = D(x(t); x(t0 ))). Also, it is easy to check that
J : C([0; T ]; E n ) → J (C([0; T ]; E n )) is a homeomorphism. To see that J is continuous
let x n (n ∈ {1; 2; : : :}), x ∈ C([0; T ]; E n ) with H (x n ; x) = supt∈[0; T ] D(x n (t); x(t)) → 0 as
n → ∞. Then
Jx n − JxC([0; T ]; X ) = sup jx n (t) − jx(t)X = sup D(x n (t); x(t))
t∈[0;T ] t∈[0;T ]

→0 as n → ∞;
so J is continuous. To see that J −1 is continuous let yn (n ∈ {1; 2; : : :}), y ∈ J (C([0; T ];
E n )) with yn − yC([0; T ]; X ) → 0 as n → ∞. Then there exists x n (n ∈ {1; 2; : : :}), x ∈
C([0; T ]; E n ) with yn = Jx n and y = Jx. Thus,
H (J −1 yn ; J −1 y) = sup D(J −1 yn (t); J −1 y(t)) = sup D(x n (t); x(t))
t∈[0;T ] t∈[0;T ]

= sup jx n (t) − jx(t)X = sup yn (t) − y(t)X

t∈[0;T ] t∈[0;T ]

→0 as n → ∞
since yn (t) = jx n (t) and y(t) = jx(t). Thus J −1 is continuous.
Our next result is known as the Arzela–Ascoli theorem in metric spaces
(see [6,9]).

Theorem 1.1. Let X be a compact metric space and Y and metric space. A subset
 of C(X; Y ) (the space of continuous mappings of X into Y with the topology of
uniform convergence) has compact closure i9  is equicontinuous on X and the subset
(x) = {(x):  ∈ } of Y has compact closure for each x ∈ X .

Next we state the >xed point result we will need in Section 2. Its proof can be
found in [3] (in fact a more general version can be found in [1]).

Theorem 1.2. Let X ∈ AR and F : X → X a compact continuous map. Then F has a

;xed point.

Remark 1.1. Recall a space Z is called an absolute retract (written Z ∈ AR) if Z is

metrizable and for any metrizable space W and any embedding h : Z → W the set
h(Z) is a retract of W .
408 D. O’Regan et al. / Nonlinear Analysis 54 (2003) 405 – 415

Finally, we recall the concepts of di#erentiability and integrability (see [4,7] for
a more detailed discussion). Let x; y ∈ E n . If there exists z ∈ E n such that x = y + z
then z is called the H-di#erence of x and y and is denoted by x − y. A mapping
F : [0; T ] → E n is di#erentiable at t0 ∈ [0; T ] if there exists F
(t0 ) ∈ E n such that the
limits
F(t0 + h) − F(t0 ) F(t0 ) − F(t0 − h)
lim and lim
h→0+ h h→0+ h
exists and are equal to F
(t0 ) (here the limit is taken in the metric space (E n ; D) and
at t = 0; t = T we consider only one-sided derivatives). Let F : [0; T ] → E n and let
T
F (t)=[F(t)] . The integral of F over [0; T ] denoted by 0 F(t) dt is de>ned levelwise
by the equation
 T 
F(t) dt
0
 T
= F (t) dt
0
 T 
n
= f(t) dt : f : [0; T ] → R is a measurable selection of F
0

for all 0 6  6 1.

2. Existence theory

Our >rst result is a Wintner-type result for (1.1). Its proof is based on Theorems
1.1 and 1.2.

Theorem 2.1. Let f : [0; T ] × E n → E n be continuous and suppose the following

conditions hold:


 ∃ a continuous nondecreasing function : [0; ∞) → [0; ∞)

and a continuous function q : [0; T ] → [0; ∞) with (2.1)



D(f(s; u); 0̂) 6 q(s) (D(u; 0̂)) for s ∈ [0; T ] and u ∈ E n
 T  ∞
du
q(s) ds ¡ (2.2)
0 D(x0 ;0̂) (u)

and
   t 

 for each t ∈ [0; T ]; the set x0 + f(s; u(s)) ds: u ∈ A
0 (2.3)

 n
is a totally bounded subset of E ;
 D. O’Regan et al. / Nonlinear Analysis 54 (2003) 405 – 415 409

here

A = {y ∈ C([0; T ]; E n ): D(y(t); 0̂) 6 a(t) for t ∈ [0; T ]} (2.4)

with
 t  z
du
a(t) = I −1 q(s) ds and I (z) = : (2.5)
0 D(x0 ;0̂) (u)

Then (1.1) has a solution.

Proof. Let A be as in (2.4) and a as in (2.5). It is immediate from (1.6) that

A∼
= B ≡ {Jw ∈ C([0; T ]; X ): w ∈ C([0; T ]; E n ) and

jw(t)X 6 a(t) for t ∈ [0; T ]}:

Clearly, B is a convex subset of the Banach space C([0; T ]; X ), so in particular B

is an absolute retract.
 t As a result A is an absolute retract. If we show F : A → A
(here Fx(t) = x0 + 0 f(s; x(s)) ds for x ∈ C([0; T ]; E n )) is continuous and compact then
Theorem 1.2 guarantees that F has a >xed point, i.e. (1.1) has a solution.
We >rst show F : A → A. Let x ∈ A and t ∈ [0; T ]. Now (1.3) and [4, Theorem 4.3]
implies

D(Fx(t); 0̂) 6 D(Fx(t); Fx(0)) + D(x0 ; 0̂)

 t
=D f(s; x(s)) ds; 0̂) + D(x0 ; 0̂
0
 t
6 D(f(s; x(s)); 0̂) ds + D(x0 ; 0̂)
0
 t
6 q(s) (D(x(s); 0̂)) ds + D(x0 ; 0̂)
0
 t
6 q(s) (a(s)) ds + D(x0 ; 0̂)
0
 t
= a
(s) ds + D(x0 ; 0̂) = a(t);
0

since
 a(t)  t
du
= q(s) ds:
D(x0 ;0̂) (u) 0

As a result F : A → A. Next we show F : A → A is continuous. Let x n (n ∈ {0; 1; : : :})

and x ∈ A with H (x n ; x) → 0 as n → ∞. Then for t ∈ [0; T ] we have (use (1.3) and
410 D. O’Regan et al. / Nonlinear Analysis 54 (2003) 405 – 415

[4, Theorem 4.3])

 t  t
D(Fx n (t); Fx(t)) = D f(s; x n (s)) ds; f(s; x(s)) ds
0 0
 t
6 D(f(s; x n (s)); f(s; x(s))) ds
0
 T
6 D(f(s; x n (s)); f(s; x(s))) ds:
0
As a result
 T
H (Fx n ; Fx) 6 D(f(s; x n (s)); f(s; x(s))) ds: (2.6)
0
Let
.n (s) = D(f(s; x n (s)); f(s; x(s)))
and notice since f : [0; T ] × E n → E n is continuous that
.n (s) → 0 as n → ∞ for s ∈ [0; T ]:
In addition, (2.1) implies for s ∈ [0; T ] that
.n (s) 6 q(s)[ (D(x n (s); 0̂)) + (D(x(s); 0̂))]
6 2q(s) (a(s)):
As a result
 T  T
lim .n (s) ds = lim .n (s) ds = 0;
n→∞ 0 0 n→∞

so this with (2.6) yields

H (Fx n ; Fx) → 0 as n → ∞:
Thus F : A → A is continuous.
It remains to show F : A → A is compact. To see this we apply Theorem 1.1. Notice
{Fx(t): x ∈ A} is equicontinuous since if t0 ; t1 ∈ [0; T ] with t0 6 t1 and x ∈ A we have
(use (1.3) and [4, Theorems 4.2 and 4.3])
 t1
D(Fx(t0 ); Fx(t1 )) = D f(s; x(s)) ds; 0̂
t0
 t1
6 D(f(s; x(s)); 0̂) ds
t0
 t1
6 q(s) (D(x(s); 0̂)) ds
t0
 t1
6 q(s) (a(s)) ds = a(t1 ) − a(t0 ):
t0
Now Theorem 1.1 (together with (2.3)) guarantees that F : A → A is compact.
 D. O’Regan et al. / Nonlinear Analysis 54 (2003) 405 – 415 411

Next we discuss (2.3) using the compactness criterion in (E n ; D) due to Diamond and
Kloeden [2]. Let U be a nonempty subset of E n and let U ∗ = j0 (U ) be its embedded
image in X0 = L∞ (S n−1 ; BV [0; T ]) ∩ L∞ ([0; T ]; C(S n−1 )) (here S n−1 is the unit sphere
in Rn and j0 : E n → X0 and X0 are as described in [2] (note in [2] their j is our j0 )).
We say U is uniformly support bounded if ∃K ¿ 0 with
sup |a| 6 K
a∈[u]0

for all u ∈ U . We say U ∗ is equileftcontinuous in  ∈ [0; 1] uniformly in x ∈ S n−1 if

for each 4, ∃5 = 5(4) ¿ 0 with
u∗ (6; x) 6 u∗ (; x) ¡ u(6; x) + 4
for all 6 −  ¡  6 6, x ∈ S n−1 and u∗ ∈ U ∗ ; here u∗ (; x) = supa∈[u] a; x where :; :
is the inner product in Rn .

Theorem 2.2 (Diamond and Kloeden [2]). A closed set U of (E n ; D) is compact i9

U is uniformly support bounded and U ∗ is equileftcontinuous in  ∈ [0; 1] uniformly
in x ∈ S n−1 .

Combining Theorems 2.1 and 2.2 yields the following result.

Theorem 2.3. Let f : [0; T ] × E n → E n be continuous and assume (2.1) and (2.2)
hold. In addition suppose
   t 

 for each t ∈ [0; T ]; the set j0 x0 + f(s; x(s)) ds: x ∈ A
0 (2.7)

 n−1
is equileftcontinuous in  ∈ [0; 1] uniformly in x ∈ S ;
here A is as in (2.4). Then (1.1) has a solution.

Proof. From Theorem 2.1 it is enough to check that (2.3) holds. Fix t ∈ [0; T ]. Now
for x ∈ A we have (see the proof of Theorem 2.1)
D(Fx(t); 0̂) 6 a(t) 6 a(T ):
That is for each x ∈ A we have
sup dH ([Fx(t)] ; [0] ) 6 a(T ):
0661

This implies for each x ∈ A that

sup d(a; [0] ) 6 a(T ) for each  ∈ [0; 1];
a∈[Fx(t)]

so in particular for  = 0. Now since [0]0 = {0} we have immediately that

{Fx(t): x ∈ A}
is uniformly support bounded.
This together with (2.7) and Theorem 2.2 guarantees for each t ∈ [0; T ] that the set
{Fx(t): x ∈ A} is relatively compact in E n .
412 D. O’Regan et al. / Nonlinear Analysis 54 (2003) 405 – 415

Our next result is a superlinear type result for (1.2).

Theorem 2.4. Let f : [0; 1] × E n → E n be continuous and suppose the following

conditions hold:


 ∃ a continuous nondecreasing function : [0; ∞) → [0; ∞)

and a continuous function q : [0; 1] → [0; ∞) with (2.8)



D(f(s; u); 0̂) 6 q(s) (D(u; 0̂)) for s ∈ [0; 1] and u ∈ E n

M
∃M ¿ 0 with 1 ¿1 (2.9)
(M ) 0
supt∈[0; 1] |G(t; s)|q(s) ds

and

 


1
 for each t ∈ [0; 1]; the set G(t; s)f(s; u(s)) ds : u ∈ A
0 (2.10)



is a totally bounded subset of E n ;

here

A = {y ∈ C([0; 1]; E n ): D(y(t); 0̂) 6 M for t ∈ [0; 1]} (2.11)

and


−s(1 − t); 0 6 s 6 t;
G(t; s) =
−t(1 − s); t 6 s 6 1:

Then (1.2) has a solution.

Proof. Let F : A → C([0; 1]; E n ) be de>ned by

 1
Fx(t) = G(t; s)f(s; x(s)) ds for x ∈ A:
0

If we show F : A → A is continuous and compact then Theorem 1.2 guarantees that

F has a >xed point, i.e. (1.2) has a solution.
We >rst show F : A → A. Let x ∈ A and t ∈ [0; T ]. Now (1.3), [4, Theorem 4.3],
(2.8) and (2.9) imply
 
1
D(Fx(t); 0̂) = D G(t; s)f(s; x(s)) ds; 0̂
0

 1
6 D(G(t; s)f(s; x(s)); 0̂) ds
0
 D. O’Regan et al. / Nonlinear Analysis 54 (2003) 405 – 415 413

 1
= |G(t; s)|D(f(s; x(s)); 0̂) ds
0
 1
6 sup |G(t; s)|q(s) (D(x(s); 0̂)) ds
0 t∈[0;1]
 1
6 (M ) sup |G(t; s)|q(s) ds 6 M:
0 t∈[0;1]

As a result F : A → A.
Next we show F : A → A is continuous. Let x n (n ∈ {0; 1; : : :}) and x ∈ A with
H (x n ; x) → 0 as n → ∞. Then for t ∈ [0; 1] we have (use (1.3) and [4, Theorem 4.3])
  1 
1
D(Fx n (t); Fx(t)) = D G(t; s)f(s; x n (s)) ds; G(t; s)f(s; x(s)) ds
0 0

 1
6 D(G(t; s)f(s; x n (s)); G(t; s)f(s; x(s))) ds
0
 1
= |G(t; s)|D(f(s; x n (s)); f(s; x(s))) ds
0
 1
6K D(f(s; x n (s)); f(s; x(s))) ds;
0

where

K = sup{|G(t; s)|: (t; s) ∈ [0; 1] × [0; 1]}:

Thus,
 1
H (Fx n ; Fx) 6 K D(f(s; x n (s)); f(s; x(s))) ds
0

→0 as n → ∞;

by essentially the same reasoning as in Theorem 2.1. Thus F : A → A is continuous.

It remains to show F : A → A is compact. To see this we apply Theorem 1.1. We
need only notice that {Fx(t): x ∈ A} is equicontinuous since if t0 ; t1 ∈ [0; 1] with t0 6 t1
and x ∈ A we have (use (1.3), [4, Theorem 4.3], (2.8) and (2.9))
 
1
D(Fx(t0 ); Fx(t1 )) = D [G(t0 ; s) − G(t1 ; s)]f(s; x(s)) ds; 0̂
0

 1
6 D([G(t0 ; s) − G(t1 ; s)]f(s; x(s)); 0̂) ds
0
414 D. O’Regan et al. / Nonlinear Analysis 54 (2003) 405 – 415

 1
= |G(t0 ; s) − G(t1 ; s)|D(f(s; x(s)); 0̂) ds
0
 1
6 (M ) |G(t0 ; s) − G(t1 ; s)|q(s) ds:
0

Now Theorem 1.1 (together with (2.10)) guarantees that F : A → A is compact.

Combining Theorems 2.2 and 2.4 yields the following result.

Theorem 2.5. Let f : [0; 1] × E n → E n be continuous and assume (2.8) and (2.9)
hold. In addition suppose
 
 
 1

 for each t ∈ [0; 1]; the set j0 G(t; s)f(s; x(s)) ds: x ∈ A
0 (2.12)



is equileftcontinuous in  ∈ [0; 1] uniformly in x ∈ S n−1 ;
here A is as in (2.11). Then (1.2) has a solution.

Remark 2.1. The ideas in this section easily extend (we leave the details to the reader)
to the nonhomogeneous problem




x (t) = f(t; x(t)) for t ∈ [0; 1];
x(0) = x0 ∈ E n ; x(1) = x1 ∈ E n :

Acknowledgements

The research of Juan J. Nieto was partially supported by Ministerio de Ciencia y

TecnologNOa and FEDER, project BFM2001-3884-C02-01; and also partially supported
by Xunta de Galicia and FEDER, Project PGIDIT02PXIC20703PN.

References

[1] R.P. Agarwal, D. O’Regan, S. Park, Fixed point theory for multimaps in extension type spaces,
J. Korean Math. Soc. 39 (2002) 579–591.
[2] P. Diamond, P. Kloeden, Characterization of compact subsets of fuzzy sets, Fuzzy Sets and Systems
29 (1989) 341–348.
[3] J. Dugundji, A. Granas, Fixed point theory, in: Monogra>e Matematyczne, Vol. 16, Polish Scienti>c
Publishers, Warsaw, 1982.
[4] O. Kaleva, Fuzzy di#erential equations, Fuzzy Sets and Systems 24 (1987) 301–317.
[5] O. Kaleva, The Cauchy problem for fuzzy di#erential equations, Fuzzy Sets and Systems 35 (1990)
389–396.
[6] J.L. Kelley, General Topology, Van Nostrand Reinhold, New York, 1955.
[7] V. Lakshmikantham, R. Mohapatra, Theory of Fuzzy Di#erential Equations and Inclusions, Taylor &
Francis, London, to appear.
 D. O’Regan et al. / Nonlinear Analysis 54 (2003) 405 – 415 415

[8] V. Lakshmikantham, K.N. Murty, J. Turner, Two point boundary value problems associated with
nonlinear fuzzy di#erential equations, Math. Inequalities Appl. 4 (2003) 527–533.
[9] J. Munkres, Topology, Prentice-Hall, London, 1975.
[10] J.J. Nieto, The Cauchy problem for continuous fuzzy di#erential equations, Fuzzy Sets and Systems
102 (1999) 259–262.
[11] M.L. Puri, D.A. Ralescu, Di#erentials of fuzzy functions, J. Math. Anal. Appl. 91 (1983) 552–558.