JOURNAL OF MATHEMATICAL ANALYSIS AND APPLICATIONS 204, 609]625 Ž1996.

ARTICLE NO. 0456

Existence and Uniqueness for a Nonlinear Fractional

Differential Equation
Domenico Delbosco

and

Luigi RodinoU
View metadata, citation and similar papers at core.ac.uk brought to you by CORE
provided by Elsevier - Publisher Connector
Dipartimento di Matematica, Uni¨ ersita
´ di Torino, ¨ ia Carlo Alberto, 10,
10123 Torino, Italy

Submitted by Mimmo Iannelli

Received March 20, 1995

We prove existence and uniqueness theorems for some classes of nonlinear

fractional differential equations. Q 1996 Academic Press, Inc.

1. INTRODUCTION

Many papers and books on fractional calculus have appeared recently.

Most of them are devoted to the solvability of linear fractional equations
in terms of special functions Žsee for example Miller and Ross w7x and
Campos w1x. and to problems of analyticity in the complex domain Žsee for
example Ling and Ding w6x.. No contribution exists, as far as we know,
concerning nonlinear fractional equations of the form

D s u s f Ž x, u .

where 0 - s - 1 and D s is the standard Riemann]Liouville fractional

derivative, considered in Rq or in an interval Ž0, h. with h ) 0.
In principle, one may reduce such an equation to an integral equation
with weak singularity and apply to it basic techniques of nonlinear analysis
Žfixed points theorems, Leray]Schauder theory.. In practice, to obtain

* Work partially supported by 40% Funds of the MURST of Italy.

609
0022-247Xr96 $18.00
Copyright Q 1996 by Academic Press, Inc.
All rights of reproduction in any form reserved.
610 DELBOSCO AND RODINO

explicit results, one has to take into account the peculiarity of the kernel.
We proceed in this direction here, presenting a list of theorems of
existence and uniqueness.
We shall not give applications in this paper, but we remark that the
previous fractional differential equation is easily shown to be equivalent to
a nonlinear heat conduction problem by using the Heaviside calculus Žcf.
Courant and Hilbert w2, Appendix to Chapter Vx.; this allows an extension
to a nonvariational setting of some results of w4x. We finally mention that
similar results, for a different fractional differential equation, have been
obtained by the first author w3x.
The paper is organized as follows. In Section 2 we recall the definitions
of fractional integral and derivative and related basic properties used in
the text. Section 3 contains results for solutions which are continuous at
the origin. Section 4 concerns initial value problems of the type

D s u s f Ž x, u .
½ u Ž a. s b

with a g Rq, b g R.

2. DEFINITIONS AND PRELIMINARY RESULTS

Let us denote by C 0 ŽRq. the space of all continuous real functions

defined on Rqs  x g R, x ) 04 and by L1loc ŽRq. the space of all real
functions defined on Rq which are Lebesgue integrable on every bounded
subinterval of Rq. Consider also the space C 0 ŽRq 0 of all continuous real
.
functions on Rq 0 s  x g R, x G 04 , which later on we shall identify, by
abuse of notation, with the class of all f g C 0 ŽRq. such that f Ž0 q . s
lim x ª 0q f Ž x . g R.
The definitions and the results of the fractional calculus reported below
are not exhaustive but rather oriented to the subject of this paper. For the
proofs, which are omitted, we refer the reader to Miller and Ross w7x or
other texts on basic fractional calculus.
DEFINITION 2.1. The fractional primitive of order s ) 0 of a function
f : Rqª R is given by

1 x sy 1
Isf Ž x. s H Ž x y t. f Ž t . dt
GŽ s. 0

provided the right side is pointwise defined on Rq.

 NONLINEAR FRACTIONAL DIFFERENTIAL EQUATION 611

For instance, I s f exists for all s ) 0, when f g C 0 ŽRq. l L1loc ŽRq. ;

note also that when f g C 0 ŽRq . s 0 Ž q.
0 then I f g C R 0 and moreover I f 0
s Ž .

s 0.
EXAMPLE 2.1.1.

G Ž m q 1.
Isx m s x m qs , s ) 0, m ) y1.
G Ž m q s q 1.

Recall that the law of composition

I r I s s I rqs
holds for all r, s ) 0.
DEFINITION 2.2. The fractional derivative of order 0 - s - 1 of a
continuous function f : Rqª R is given by

1 d x ys
Dsf Ž x. s H Ž x y t. f Ž t . dt
GŽ1 y s. dx 0

provided that the right side is pointwise defined on Rq.

EXAMPLE 2.2.1. As a basic example, we quote for l ) y1

G Ž l q 1.
Dsxl s x lys ,
G Ž l y s q 1.

giving in particular D s x sy 1 s 0.
We have D s I s f s f for all f g C 0 ŽRq. l L1loc ŽR.. From definition 2.2
and Example 2.2.1 we then obtain:
LEMMA 2.3. Let 0 - s - 1. If we assume u g C 0 ŽRq. l L1loc ŽRq. , then
the fractional differential equation

D su s 0
has u s cx sy1 , c g R, as unique solutions.
From this lemma we deduce the following law of composition.
PROPOSITION 2.4. Assume that f is in C 0 ŽRq. l L1loc ŽRq. with a frac-
tional deri¨ ati¨ e of order 0 - s - 1 that belongs to C 0 ŽRq. l L1loc ŽRq. .
Then
I s D s f Ž x . s f Ž x . q cx sy 1
for some c g R.
612 DELBOSCO AND RODINO

When the function f is in C 0 ŽRq

0 , then c s 0.
.
In all the definitions and results of this section the set Rq can be
substituted by the intervals Ž0, a. or Ž0, ax, a ) 0. For simplicity, in the next
sections we shall often limit arguments to the choice a s 1.
A more precise analysis of the operators I s, D s can be given in the
frame of the spaces Cr0 ŽRq 0 , r G 0, of all functions f g C R
. 0 Ž q.
such that
0 Ž q. x.
x f g C R 0 . We define similarly Cr 0, a , which turns out to be a
r 0 Žw

Banach space when endowed with the norm

5 f 5 r s max x r < f Ž x . < .
xg w0, a x

We have C00 ŽRq 0 Ž q.

0 s C R 0 and C 0 0, a s C
. 0 Žw x. 0 Žw
0, ax., the Banach space
of all continuous functions on w0, ax with norm
5 f 5 s max < f Ž x . < .
xg w0, a x

Obviously Cr0 ŽRq 1 Ž q.

0 ; L loc R
. if r - 1.
Let 0 - s - 1; if f g Cr0 ŽRq0
. with r - s, then I s f g C 0 ŽRq
0 , with
.
0 Ž q.
I f 0 s 0. If f g C s R 0 then I f is bounded at the origin, whereas if
s Ž . s

f g Cr0 ŽRq
0
. with s - r - 1, then we may expect I s f to be unbounded at
the origin. Concerning Proposition 2.4, the last part can now be stated
more precisely:
If f g Cr0 ŽRq
0
. with r - 1 y s and D s f g C 0 ŽRq. l L1loc ŽRq. , then
I D f s f.
s s

3. DIFFERENTIAL EQUATIONS OF REAL ORDER:

CONTINUOUS SOLUTIONS ON w0, 1x

Consider the fractional differential equation

D s u s f Ž x, u . Ž 3.1.
where 0 - s - 1 and f : w0, ax = R ª R, 0 - a F q`, is a given function,
continuous in Ž0, a. = R.
We introduce the following definition of a solution for Eq. Ž3.1..
DEFINITION 3.1. A real valued function u g C 0 ŽŽ0, a.. l L1 ŽŽ0, a.., or
u g C 0 ŽRq. l L1loc ŽRq. in the case a s q`, with continuous fractional
derivative D s u on Ž0, a., is a solution of fractional differential equation
Ž3.1. if
D s u Ž x . s f Ž x, u Ž x . .
for all x g Ž0, a..
 NONLINEAR FRACTIONAL DIFFERENTIAL EQUATION 613

Remark 3.2. We may apply the results of Section 2, in particular

Proposition 2.4 and the subsequent remarks, to reduce Ž3.1. to an integral
equation. In fact, if u g C 0 Žw0, ax., or more generally, u g Cr0 Žw0, ax. with
r - 1 y s, and further assumptions guarantee f Ž x, uŽ x .. g C 0 ŽŽ0, a.. l
L1 ŽŽ0, a.., then the equation Ž3.1. is equivalent to the integral equation

u Ž x . s I s f Ž x, u Ž x . . . Ž 3.2.
Such a reduction will be systematically used in this section.
If we allow u g Cr0 Žw0, ax. with 1 y s F r - 1, then u is a solution of
Ž3.1. if and only if for some c g R

u Ž x . s I s f Ž x, u Ž x . . q cx sy 1 . Ž 3.2U .
It will be natural in this case to submit u to an initial condition; results in
this connection will be given in Section 4.
We first present a local existence theorem.
THEOREM 3.3. Let 0 F s - s - 1 and let f : w0, 1x = R ª R be a gi¨ en
function continuous in Ž0, 1x = R. Assume that t s f Ž t, y . is a continuous
function on w0, 1x = R. Then the fractional differential equation

D s u s f Ž x, u . Ž 3.3.
has at least a continuous solution defined on w0, d x for a suitable d F 1.
Proof. According to Remark 3.2, we are reduced to consider the
following nonlinear integral equation

1 x sy1
uŽ x . s H Ž x y t. f Ž t , u Ž t . . dt.
GŽ s. 0

Let T : C 0 Žw0, 1x. ª C 0 Žw0, 1x. be the operator defined as

1 x sy 1
Ž Tu . Ž x . s H Ž x y t. f Ž t , u Ž t . . dt.
GŽ s. 0

We claim that the operator T is compact. Indeed, the operator is the

composition of two simple operators in this way

T s A( N
where
Ž Nu . Ž x . s x s f Ž x, u Ž x . .
614 DELBOSCO AND RODINO

is a continuous and bounded operator ŽNemytskii operator. and

1 x sy 1 y s
Ž A¨ . Ž x . s H Ž x y t. t ¨ Ž t . dt
GŽ s. 0

is a compact operator, since s y s ) 0; since for example w5x.

Moreover, from Example 2.1.1 we have for 0 F x F d F 1

1 x sy 1 y s
A¨ Ž x . F ž sup ¨ Ž x. / GŽ s. H Ž x y t . t dt
xg w0, d x 0

GŽ1 y s .
F d sy s sup ¨ Ž x. ;
GŽ1 y s q s. xg w0, d x

therefore, taking the norms in C 0 Žw0, d x.,

5 A¨ 5 F e 5 ¨ 5 ,

where we may assume e ) 0 as small as we want by shrinking d ) 0.

Now fix Br as a domain of the operator T, where Br s  ¨ g
C 0 Žw0, d x. : 5 ¨ 5 F r 4 , which is a convex, bounded, and closed subset of the
Banach space C 0 Žw0, d x..
For d sufficiently small, we have

T Ž Br . : Br .

The Schauder fixed point theorem assures that operator T has at least
one fixed point and then Ž3.3. has at least one continuous solution u
defined on w0, d x, where d F 1.
EXAMPLE 3.3.1. From the preceding proof we also obtain that, under
the assumptions of Theorem 3.3, all solutions u g C 0 Žw0, d x., d ) 0, of
Ž3.3. vanish at the origin.
Observe also that we cannot expect uniqueness for such solutions, in
general. Consider for example the equation

D su s ur Ž 3.4.

with 0 - r, s - 1, which admits the two solutions u s 0 and

u Ž x . s kx srŽ1yr . ,
 NONLINEAR FRACTIONAL DIFFERENTIAL EQUATION 615

where
1r Ž ry1 .
GŽ m. s
with m s
ks
ž GŽ m y s. / 1yr
q1

as we have from Example 2.2.1

EXAMPLE 3.3.2. Theorem 3.3 is a local existence theorem and we
cannot expect all the solutions to extend to Rq or w0, 1.. Consider for
example the equation

Mx 1r2 2
D 1r2 u s ŽuqM. , M ) 0, Ž 3.5.
p 1r2

which admits the solution

y1 r2
u Ž x . s Ž My2 y x . yM

Žuse formula IIc, page 354, in Miller and Ross w7x.. The domain of
existence of uŽ x . is w0, 1rM 2 ., which shrinks to zero as M ª q`.
The following theorem shows that uniqueness and global existence can
be obtained under an uniform Lipschitz-type assumption. The result
applies in particular to the linear case f Ž x, u. s g Ž x . q hŽ x . u with g, h g
Cs0 Žw0, 1x., s - s.
THEOREM 3.4. Let 0 F s - s - 1 and assume t s f Ž t, y . is continuous on
w0, 1x = R. Assume further

L
f Ž t , u. y f Ž t , ¨ . F uy¨ Ž 3.6.
ts

for some positi¨ e constant L independent of u, ¨ g R, t g Ž0, 1x. Then the

equation

D s u s f Ž x, u .

has a unique solution u g C 0 Žw0, 1x..

Proof. As in the proof of Theorem 3.3, we are reduced to studying the
operator

1 x sy 1
Tu Ž x . s H Ž x y t. f Ž t , u Ž t . . dt, Ž 3.7.
GŽ s. 0
616 DELBOSCO AND RODINO

which is well defined and continuous as a map T : C 0 Žw0, 1x. ª C 0 Žw0, 1x., in
view of the assumption of continuity on t s f Ž t, y .. Let us define the iterates
of operator T as is standard:

T1 s T T n s T (T ny 1 .
It will be sufficient to prove that T n is a contraction operator for n
sufficiently large. Actually, we have for u, ¨ g C 0 Žw0, 1x.
n
Ž KL .
T n uŽ x . y T n ¨ Ž x . F x nŽ sy s . 5 u y ¨ 5 Ž 3.8.
G Ž n Ž s y s . q 1.

where the constant K depends only on s and s . In fact,

1 x sy 1
Tu Ž x . y T¨ Ž x . F H Ž x y t. f Ž t , u Ž t . . y f Ž t , ¨ Ž t . . dt
GŽ s. 0

which we further estimate using Ž3.6. and Example 2.1.1 by

GŽ1 y s . L
x sy s 5 u y ¨ 5 . Ž 3.9.
G Ž s y s q 1.

Therefore Ž3.8. is proved for n s 1, if K G G Ž1 y s .. Assuming by

induction that Ž3.8. is valid for n, we obtain similarly

T nq 1 u Ž x . y T nq1 ¨ Ž x .
K n Lnq1 x
F 5u y ¨ 5H0 Ž x y t . sy1 nŽ sy s .y s
t dt
G Ž n Ž s y s . q 1. G Ž s .

G Ž n Ž s y s . y s q 1 . K n Lnq 1
s x Ž nq1.Ž sy s . 5 u y ¨ 5 ,
G Ž n Ž s y s . q 1. G Ž Ž n q 1. Ž s y s . q 1.

and then Ž3.8. follows for n q 1 if K is given by

G Ž n Ž s y s . y s q 1.
K s max K n , Kn s . Ž 3.10.
n G Ž n Ž s y s . q 1.

Note that Ž3.10. defines actually a finite K, since K n F 1 for n G Ž1 q

s .rŽ s y s .. Taking n sufficiently large in Ž3.8., we have, say,
Ž KL. nrG Ž nŽ s y s . q 1. F 1r2 and therefore

5 T n u y T n ¨ 5 F 12 5 u y ¨ 5 ,

which gives the proof.

 NONLINEAR FRACTIONAL DIFFERENTIAL EQUATION 617

Finally, we consider the limit case when in Theorem 3.4 we have s s s.

THEOREM 3.5. Let 0 - s - 1. Assume that t s f Ž t, y . is continuous on
w0, 1x = R and moreo¨ er, that
L
f Ž t , u. y f Ž t , ¨ . F uy¨ Ž 3.11.
ts
where
1
L- ,
GŽ1 y s.
for all u, ¨ g R and t g Ž0, 1x. Then
D s u Ž x . s f Ž x, u .
has a unique solution u g C 0 Žw0, 1x..
Proof. We shall prove that under the preceding assumptions T :
C 0 Žw0, 1x. ª C 0 Žw0, 1x., defined by Ž3.7. is a contraction operator.
Indeed, setting s s s in Ž3.9., we have the estimate
Tu Ž x . y T¨ Ž x . F LG Ž 1 y s . 5 u y ¨ 5 .
Thus we obtain that
5 Tu y T¨ 5 F k 5 u y ¨ 5
where
k s LG Ž 1 y s . - 1.
Our theorem is proved.
EXAMPLE 3.5.1. The assumption L - 1rŽ G Ž1 y s . in Theorem 3.5 is
not a technical accident, as the following simple example shows. The linear
equation
1
D su s xys u
GŽ1 y s.
admits the two solutions uŽ x . s 0, uŽ x . s 1.
Therefore, the result of uniqueness fails, in general, if L s 1rG Ž1 y s .
in Ž3.11..

4. INITIAL VALUE PROBLEM: CONTINUOUS

SOLUTIONS ON Ž0, 1x

We open this section with some basic examples, concerning the case
when the solutions in C 0 ŽRq. are submitted to an initial condition.
618 DELBOSCO AND RODINO

PROPOSITION 4.0. Let 0 - s - 1. For all Ž a, b . g Rq= R the initial

¨ alue problem
D su s 0
½ u Ž a. s b

admits u s ba1y s x sy1 as unique solution in C 0 ŽRq. l L1loc ŽRq. .

This proposition follows directly from Lemma 2.3.
COROLLARY 4.0.1. Let 0 - s - 1. Assume f Ž x . g C 0 ŽRq. l L1loc ŽRq. .
Then for all Ž a, b . g Rq= R the initial ¨ alue problem

D su s f Ž x .
½ u Ž a. s b

has a unique solution in C 0 ŽRq. l L1loc ŽRq. gi¨ en by

1 a sy1
x sy 1
uŽ x . s b y H Ž a y t.
ž GŽ s. 0
f Ž t . dt
/ a sy1
1 x sy 1
q H Ž x y t. f Ž t . dt.
GŽ s. 0

Now we introduce the function e s : Rqª Rq defined by

q` x k sy1
es Ž x . s Ý G Ž ks . .
ks1
0 Ž q.
The function e s Ž x . belongs to C1ys R 0 . Indeed, taking the norm in
0 Žw
C1y s 0,
x.
h , h ) 0, we have
q` hŽ ky1. s
5 e s Ž x . 5 1y s F Ý - `.
ks1 G Ž ks .

Remark that the function e s Ž x . can be expressed by means of the

classical Mittag]Leffler special function Žsee Miller and Ross w7, Chap. 5x..
THEOREM 4.1. Let 0 - s - 1. The initial ¨ alue problem

D su s u
½ u Ž a . s b,
Ž 4.1.

where a ) 0 and b g R, has a unique solution u g C1y x.

s 0, h , h ) a,
0 Žw

gi¨ en by
y1
u Ž x . s be s Ž a . es Ž x . .
 NONLINEAR FRACTIONAL DIFFERENTIAL EQUATION 619

Proof. We are reduced to consider the integral equation

u Ž x . s cx sy 1 q I s u Ž x . , c g R. Ž 4.1U .
Fix h ) a and find u g C1y
0 Žw x.
s 0, h .
U
From 4.1 we obtain, by iteration,
Ž .

x sy1 x 2 sy1 x sny1

u Ž x . s cG Ž s . q q ??? q q I n s uŽ x . .
GŽ s. GŽ2 s. G Ž sn .

Letting n ª q`, one has 5 I n s uŽ x .5 1y s ª 0 if u g C1ys0 Žw

0, h x., as we
deduce easily from Definition 2.1. On the other hand, the sum in the
0 Žw
right-hand side tends to e s Ž x . in the C1ys 0, h x.-norm. This implies that

u Ž x . s cG Ž s . e s Ž x . .

This equality enables us to solve the initial value problem uniquely.

The preceding example suggest we look for solutions u g C1y 0 Žw x.
s 0, h of
the general autonomous initial value problem

D s u s f Ž u.
½ u Ž a . s b,
Ž 4.2.

where 0 - s - 1, a ) 0, and b g R. As for the function f : R ª R, we

shall argue initially under the hypothesis:

f Ž 0. s 0
½ f Ž u. y f Ž ¨ . F L u y ¨
Ž 4.3.

for some positive constant L independent of u, ¨ g R.

LEMMA 4.2. Let 0 - s - 1. Assume that f : R ª R satisfies Ž4.3.. Then
for any fixed c g R the nonlinear integral equation

1 x sy1
u Ž x . s cx sy1 q H Ž x y t. f Ž u Ž t . . dt
GŽ s. 0

has a unique solution u g C1y

0 Žw
s 0, h
x. for all h ) 0.
Proof. Our problem is equivalent to the problem of determination of
fixed points of the continuous operator

1 x sy 1
Ž Tc u . Ž x . s cx sy1 q H Ž x y t. f Ž u Ž t . . dt.
GŽ s. 0
620 DELBOSCO AND RODINO

It is immediate to verify that

0 Žw
Ži. Tc : C1ys 0, h x. ª C1ys
0 Žw
0, h x. is well defined.
Indeed, we have

5 Ž Tc u . Ž x . 5 1y s s max x 1y s Tc u Ž x . F c q R 5 u 5 1ys
xg w0, h x

since < f Ž j .< F L < j < from assumption, the constant R being given by

x 1y s x sy 1 sy1
GŽ s.
max L
xg w0, h x
H Ž x y t.
GŽ s. 0
t dt s L
GŽ2 s.
hs

in view of Example 2.1.1.

Žii. Tcn is a contraction operator for n sufficiently large.
Indeed we have, computing as in Ži.,

LG Ž s .
x 1y s Tc u Ž x . y Tc¨ Ž x . F x s 5 u y ¨ 5 1y s
GŽ2 s.

for any x g Ž0, h. and for all u, ¨ g C1y

0 Žw x.
s 0, h and, by induction, arguing
as in the proof of Theorem 3.4 that

Ln G Ž s .
x 1y s
Tcn u Ž x. y Tcn ¨ Ž x. F x n s 5 u y ¨ 5 1y s
G Ž Ž n q 1. s .

for all u, ¨ g C1y

0 Žw x.
s 0, h .
Thus we get

Ln h n s G Ž s .
5 Tcn u y Tcn ¨ 5 1y s F 5 u y ¨ 5 1ys .
G Ž Ž n q 1. s .

Since
Ln h n s
ª0
G Ž Ž n q 1. s .

as n tends to q`, then for n sufficiently large the operator T n is a

contraction operator. Therefore for any c there exists u s uŽ x, c . defined
on Ž0, h. for all h ) 0, satisfying the required equation.
We shall begin by applying Lemma 4.2 to the case of an initial condition
at the origin and then pass to considering the problem Ž4.2..
 NONLINEAR FRACTIONAL DIFFERENTIAL EQUATION 621

THEOREM 4.3. Let 0 - s - 1. Assume that f : R ª R satisfies Ž4.3..

Then for all b g R, the initial ¨ alue problem

D s u s f Ž u.
½ x 1y s u Ž x . < xs0 s b
Ž 4.4.

has a unique solution u g C1y

0 Žw
s 0, h
x. for all h ) 0.
Proof. Since we seek a solution u g C1y 0 Žw
s 0, h
x. the initial value prob-
lem Ž4.4. is equivalent to the nonlinear integral equation
1 x sy1
u Ž x . s bx sy1 q H Ž x y t. f Ž u Ž t . . dt. Ž 4.5.
GŽ s. 0

In fact, in view of Remark 3.2, it suffices to verify that

lim x sy1 u Ž x . s b,
xª0

or, equivalently,

x 1y s x sy1
lim H Ž x y t.
xª0 G Ž s . 0
f Ž u Ž t . . dt s 0

which is evident from the first part of the proof of Lemma 4.2. Theorem
4.3 follows then by applying Lemma 4.2.
THEOREM 4.4. Let 0 - s - 1. Assume that f : R ª R satisfies Ž4.3..
Then the initial ¨ alue problem Ž4.2. has a unique solution u g C1y
0 Žw
s 0, h
x.
for all h ) a, pro¨ ided a - a0 , where a0 is a suitable positi¨ e constant
depending on s and L.
Proof. The initial value problem Ž4.2. will be solved in two steps:
Ž1. Local existence. The integral equation

1 a sy 1
uŽ x . s b y H Ž a y t.
ž GŽ s. 0 /
f Ž u Ž t . . dt a1ys x sy1

1 x sy 1
q H Ž x y t. f Ž u Ž t . . dt Ž 4.6.
GŽ s. 0

is equivalent to the initial value problem Ž4.2., cf. Corollary 4.0.1.

Note that solving Ž4.6. is equivalent to finding the fixed points of the
operator

s Ž w 0, h x . ª C1ys Ž w 0, h x .
0 0
A: C1y
622 DELBOSCO AND RODINO

defined as A¨ Ž x . s uŽ x . where uŽ x . is the unique solution of

1 a sy1
uŽ x . s b y H Ž a y t.
ž GŽ s. 0 /
f Ž ¨ Ž t . . dt a1ys x sy1

1 x sy 1
q H Ž x y t. f Ž u Ž t . . dt.
GŽ s. 0

Note also that operator A is well defined from Lemma 4.2.

We shall now prove that A is a contraction.
Indeed, setting A¨ i s u i , we have
5 u1 y u 2 5 1y s s 5 A¨ 1 y A¨ 2 5 1ys
LG Ž s . h s LG Ž s . a s
F 5 u1 y u 2 5 1y s q 5 ¨ 1 y ¨ 2 5 1ys
GŽ2 s. GŽ2 s.

for all ¨ 1 , ¨ 2 g C1ys

0 Žw
0, h x..
Let us assume
GŽ s. 1
L as - ,
GŽ2 s. 2
that is,
1rs
GŽ2 s.
a - a0 s
ž 2 LG Ž s . / .

Taking h y a ) 0 sufficiently small, we also have

LG Ž s . h s 1
-
GŽ2 s. 2
and then
5 A¨ 1 y A¨ 2 5 1y s F R 5 ¨ 1 y ¨ 2 5 1ys

with R s 2 LŽ G Ž s . a srG Ž2 s .. - 1.
Therefore A is a contraction operator.
This shows that initial problem Ž4.2. has a unique solution.
Ž2. Continuation of solution. Since we know the value of uŽ x . on
Ž0, ax, then we can compute

1 a sy1
c* s b y H Ž a y t.
ž GŽ s. 0 /
f Ž u Ž t . . dt a1ys .
 NONLINEAR FRACTIONAL DIFFERENTIAL EQUATION 623

Lemma 4.2 enables us to solve the integral equation

1 x sy 1
w Ž x . s c* x sy1 q H Ž x y t. f Ž w Ž t . . dt,
GŽ s. 0
obtaining a unique solution w g C1y 0 Žw
s 0, h
x. for all h ) 0.
Now uŽ x . and w Ž x . agree on Ž0, ax. Thus the solution uŽ x . admits w Ž x .
as its continuation.
The proof is complete.
By using somewhat more sophisticated arguments, we can enlarge the
bound a0 given in the preceding proof; however, we do not know whether
existence and uniqueness in Theorem 4.4 hold for all a g Rq, as we had in
Theorem 4.1.
Finally we want to study the not-globally-Lipschitzian case. For simplic-
ity, we limit ourselves to analyzing the model f Ž t . s t a , with a ) 0, under
the initial condition Ž4.4..
THEOREM 4.5. Let 0 - s - 1 and b g R. If we assume 0 - a -
1rŽ1 y s ., then the initial ¨ alue problem
D su s ua
½ x 1y s u Ž x . < xs0 s b
Ž 4.7.
0 Žw
has at least a solution in C1y s 0, h
x. for h ) 0 sufficiently small.
Proof. If u g C1y
0 Žw x. Ž . a
s 0, h and a s y 1 ) y1, then u g L 0, h . In
1 ŽŽ ..
view of Remark 3.2, we are therefore reduced again to the nonlinear
integral equation
1 x
u a Ž t . dt.
sy 1
u Ž x . s bx sy1 q H Ž x y t. Ž 4.7.i.
GŽ s. 0
The existence of a solution to the above problem can be formulated as a
fixed point equation Tu s u where
1 x
u a Ž t . dt
sy1
Ž Tu . Ž x . s bx sy1 q H Ž x y t. Ž 4.7.ii.
GŽ s. 0
0 Žw
in the space C1y x.
s 0, h . To prove existence we use Schauder’s fixed point
theorem.
The proof is divided in two parts:
A priori estimates. We seek an a priori estimate of fixed points of
operator T defined by Ž4.7.ii. by means of a closed sphere S r defined as

S r s  u g C1ys
0
Ž w 0, h x . : 5 u y bx sy 1 5 1y s F r 4 .
624 DELBOSCO AND RODINO

To obtain r ) 0 such that T S r : S r , we note that

x 1y s x
5 Tu y bx sy1 5 1y s s max H0 Ž x y t . sy 1 a Ž sy1. a Ž1ys.
t t u a Ž t . dt
xg w0, h x GŽ a .
G Ž a Ž s y 1. q 1. a
F h a Ž sy1.q1 5 u 5 1y s.
G Ž a Ž s y 1. q 1 q s .

Since 5 u 5 1y s F r q < b <, it will be sufficient to impose

a
5 u y bx sy1 5 1y s F const.h a Ž sy1.q1 Ž r q b . F r.
In view of the assumption a Ž s y 1. q 1 ) 0, the second estimate is
satisfied if, say, r s < b < and h is chosen sufficiently small.
Proof of compactness of T. To prove the compactness of

s Ž w 0, h x . ª C1ys Ž w 0, h x .
0 0
T : C1y
defined by Ž4.7.ii., it will be sufficient to argue on the operator

T* : C 0 Ž w 0, h x . ª C 0 Ž w 0, h x .
defined in this way:

Ž T* u . Ž x . s x 1y s T Ž x sy1 u Ž x . . .
We have T#u s b q T U u where the operator
1 x
T U uŽ x . s
sy 1 a Ž sy1.
Hx 1ys
Ž x y t. t u a Ž t . dt
GŽ s. 0

turns out to be compact from classical sufficient conditions, since a Ž s y 1.

) y1. The proof is complete.
Remark 4.6. If in the preceding theorem one assumes a G 1rŽ1 y s .,
then u g C1y0 Žw x. Ž . s a
s 0, h and b / 0 in 4.7 give D u s u f L 0, h , so our
1 ŽŽ ..
very reduction to the integral equation fails in this case. Observe also that
under the assumptions of the previous theorem we cannot expect unique-
ness as shown by Example 3.3.1.

REFERENCES

1. L. M. C. M. Campos, On the solution of some simple fractional differential equations,

International J. Math. Math. Science. 13 Ž1990., 481]496.
2. R. Courant and D. Hilbert, ‘‘Methods of Mathematical Physics,’’ Interscience, London,
1962.
 NONLINEAR FRACTIONAL DIFFERENTIAL EQUATION 625

3. D. Delbosco, Fractional calculus and function spaces, J. Frac. Cal. 6 Ž1994., 45]53.
´
4. G. Duvaut and J. L. Lions, ‘‘Les inequations ´
en mecanique et en physique,’’ Dunod, Paris,
1972.
5. M. A. Krasnosel’skii, ‘‘Topological Methods in the Theory of Nonlinear Integral Equa-
tions,’’ Pergamon, Oxford, 1964.
6. Y. Ling and S. Ding, A class of analytic functions defined by fractional derivation, J. Math.
Anal. and Appl. 186 Ž1994., 504]513.
7. K. S. Miller and B. Ross, ‘‘An Introduction to the Fractional Calculus and Fractional
Differential Equations,’’ Wiley, New York, 1993.
8. M. Riesz, L’integrale de Riemann]Liouville et le probleme ` de Cauchy, Acta Math. 81
Ž1948., 1]222.