Sunteți pe pagina 1din 16

on 05/11/17. For personal use only.

from www.worldscientific.com

PITTSBURGH

Downloaded

Fractals OF

by UNIVERSITY

Fractals, Vol. 25, No. 3 (2017) 1750033 ( 16 pages) c World Scientific Publishing Company DOI: 10.1142/S0218348X17500335

Publishing Company DOI: 10.1142/S0218348X17500335 ON HADAMARD FRACTIONAL CALCULUS LI MA ∗ and CHANGPIN

ON HADAMARD FRACTIONAL CALCULUS

LI MA and CHANGPIN LI Department of Mathematics, Shanghai University Shanghai 200444, P. R. China mali20062787@163.com lcp@shu.edu.cn

Received January 8, 2017 Accepted March 29, 2017 Published May 11, 2017

Abstract

This paper is devoted to the investigation of the Hadamard fractional calculus in three aspects. First, we study the semigroup and reciprocal properties of the Hadamard-type fractional oper- ators. Then, the definite conditions of certain class of Hadamard-type fractional differential equations (HTFDEs) are proposed through the Banach contraction mapping principle. Finally, we prove a novel Gronwall inequality with weak singularity and analyze the dependence of solu- tions of HTFDEs on the derivative order and the perturbation terms along with the proposed initial value conditions. The illustrative examples are presented as well.

Keywords: Hadamard-Type Fractional Operators; Semigroup; Well-Posedness; Banach Con- traction Mapping Principle; Perturbation.

1. INTRODUCTION

Recently, a large amount of interest in fractional calculus has been stimulated due to its various applications, especially in different realms of applied sciences like physics, mechanics, engineering, and biology, see Refs. 112 and references therein.

Corresponding author.

So far, there have existed several definitions of frac- tional integrals and fractional derivatives, such as Gr¨unwald–Letnikov, Riemann–Liouville, Caputo, Riesz, Erd´elyi–Kober, and Hadamard, integrals and/or derivatives. 2,1317 Generally speaking, they are not equivalent 2,4 and have their respective

1750033-1

on 05/11/17. For personal use only.

from www.worldscientific.com

PITTSBURGH

Downloaded

Fractals OF

by UNIVERSITY

L. Ma & C. P. Li

backgrounds. 1820 To the best of our knowledge, most of publications focus on the fractional dif- ferential equations involved Riemann–Liouville and Caputo derivatives. 2123 However, there are still some of the articles discussing the properties of Hadamard version. For example, in Ref. 24, the authors list the main two differences between the Hadamard fractional derivative and the Riemann–

Liouville fractional derivative. Overall, on one hand, the kernel of the Hadamard fractional integral takes the logarithmic form of log( x ) instead of the form of (x t). On the other hand, the Hadamard frac- tional derivative can be regarded as generalization

t

operator (x dx ) n , whose construction is well

suited to the case of the half-axis and is invariant relative to dilation, while the Riemann–Liouville fractional derivative can be viewed as an extension of the classic differential operator ( dx ) n cursorily.

In this paper, we investigate Hadamard fractional calculus, including Hadamard fractional integrals and Hadamard fractional derivatives, Hadamard- type fractional integrals, and Hadamard-type frac- tional derivatives, which have slight differences but are unitedly called Hadamard fractional calculus. How to determine the definite conditions for frac- tional differential equations seems to be a ticklish matter. 25 It is worth mentioning that the studies of the definite conditions for fractional differential equations have been more or less mentioned. 24,2631 Delbosco and Rodino show 32 the existence and uniqueness for a nonlinear Riemann–Liouville type fractional differential equation. In Ref. 33, the authors discuss the well-posed conditions for a cer- tain class of fractional order boundary value prob- lems with Caputo derivatives. There are few stud- ies on definite conditions of Hadamard fractional differential equations (HFDEs) and Hadamard-type fractional differential equations (HTFDEs). In this paper, we consider such questions. In addition, it is remarked that memory depen- dency of solutions is a desired property in frac- tional differential equations. So this motivates us to detect the parameter dependence of the solutions of HFDEs and HTFDEs. The rest of the paper is organized as fol- lows: In Sec. 2, we introduce some definitions of Hadamard-type fractional integral and derivative along with the associate function spaces. In Sec. 3, we discuss semigroup and reciprocal properties of Hadamard-type fractional integral and derivative operators. Section 4 is dedicated to investigating

of the

d

d

the well-posedness. In addition, Sec. 5 deals with the continuous dependence of parameters. The last section concludes this paper.

2. PRELIMINARIES

In this section, we mainly present some basic defini- tions, function spaces, and properties related to the Hadamard-type fractional integral and fractional derivative. First of all, let f (x) be a function defined on (a, b), where 0 a<b ≤ ∞ and µ R.

Definition 1. The Hadamard-type fractional inte- gral with parameter µ R (Hadamard-type inte- gral for brevity) of the given f (x) with order α > 0 is defined as 14,15

H D

α

a

,

+

µ f(x) =

Γ(α) x

1

a

x µ

t

× log x α1 f(t) dt ,

t

t

(1)

where x (a, b), and 0 a<b ≤ ∞.

Definition 2. The Hadamard-type fractional derivative with parameter µ R (Hadamard-type derivative for brevity) of the given f (x) is defined

as 14,15

(2)

where δ = x dx , n 1 < α n Z + , x (a, b), and 0 a<b ≤ ∞.

If letting µ = 0, Definitions 1 and 2 are just the usual Hadamard integral and derivative, respec- tively. Hadamard fractional calculus always repre- sents Hadamard(-type) integral and derivative in this paper if no confusion arises.

Definition 3. The single-parameter Mittag–Leffler function and the two-parameter Mittag–Leffler function are defined as 2

H D

+ , µ f(x) = x µ δ n x µ ( H D

α

a

d

(nα) a + , µ

f (x)),

E α (x) =

k=0

x k

Γ(αk

+ 1) ,α> 0

and

E α, β (x) =

respectively.

k=0

x k

Γ(αk

+ β) ,α> 0,β> 0,

Obviously,

E α, 1 (x) = E α (x),

E 1 (x) = E 1, 1 (x) = e x .

In the sequel, for the Hadamard-type frac- tional integral operator H D a+, µ (·), we define space

α

1750033-2

on 05/11/17. For personal use only.

from www.worldscientific.com

PITTSBURGH

Downloaded

OF

Fractals

by UNIVERSITY

On Hadamard Fractional Calculus

X

Lebesgue measurable functions f on [a, b] for which

(a, b) (c R, 1 p ≤ ∞) of those real-valued

p

c

f X

p

c

< , where

f X

p

c

= b

a

|t c f(t)| p dt

t

1/p

,

(1 p < , c R)

(3)

and

f X

c

=

ess sup atb [t c |f (t)|], (c R).

(4)

(a, b)

In particular, if c = 1/p, then the space X coincides with the common L p (a, b) space.

Then, for the Hadamard-type fractional dif-

ferentiation operator H D a+, µ (·), we define space

AC

p

c

α

n

δ;

µ [a, b] of those functions satisfying

n

δ;

AC

µ [a, b] = g : [a, b]

R|δ n1 (x µ g(x))

AC[a, b], µ R; δ = x dx ,

d

(5)

where AC[a, b] is the set of absolutely continu-

ous functions on [a, b]. In particular, if µ = 0,

AC n 0 [a, b] AC n [a, b].

δ;

δ

Next, we construct the generalized weighted space C µ, γ, log [a, b] defined as

C µ, γ, log [a, b] = f(x)| log x γ x µ f(x)

a

C[a, b], f C µ, γ, log

=

log x γ x µ f(x) C .

a

(6)

Obviously, C 0, γ, log [a, b] C γ, log [a, b], C 0, 0, log [a, b] C[a, b]. In other words, the generalized weighted space C µ, γ, log [a, b] can be regarded as an extension of weighted space C γ, log [a, b], which can be found in Refs. 2 and 24. In the following, we give the fundamental proper- ties for the Hadamard(-type) fractional operators.

Lemma 1 (see Ref. 2 ). Let α > 0,β > 0, 1

p ≤ ∞, 0 a<b ≤ ∞ and let µ, c R with µ c.

(a, b) the semigroup property holds,

Then for f X i.e.

p

c

H D

α

a + , µ H D a +

β

,

µ f(x) = H D

(α+β) a + , µ

f (x).

(7)

Lemma 2 (see Ref. 2 ). Let α β > 0, 1 p ≤ ∞, 0 a<b ≤ ∞ and let µ, c R with µ c.

Then for f X

p

c

(a, b) there holds

H

D

β

a + , µ H D a +

α

,

µ f(x) = H D

(αβ) a + , µ

f (x).

(8)

In particular, if β = m Z + , then

H

D

m

a + , µ H D a +

α

,

µ f(x) = H D

(αm) a + , µ

f (x).

(9)

Lemma 3 (see Ref. 2). If 0 <α< 1, f (x)

L(a, b), and H D

(1α)

a

1

f (x) AC

δ

[a, b], then

+

H

D α

a

+

H

+ f(x) = f(x) + c log x α1 ,

D α

a

a

(10)

where c is a constant associated with some initial value.

Lemma 4 (see Ref. 2). If 0 <α< 1, f (x) L(a, b), then the relation H D α + f (x)=0 valids, if and only if,

a

f(x) = c log x α1 ,

a

(11)

where c is an arbitrary real number.

Motivated by the above two lemmas for Hada- mard fractional differentiation operator H D α + (·), it is trivial to obtain analogous conclusions for Hadamard-type differentiation operator H D

(·) as follows:

a + , µ

a

α

Lemma 5.

H D

(1α)

a + , µ

f

If 0

<α< 1, f (x)

1

(x) AC δ; µ [a, b], then

L(a, b), and

H D

α

a + , µ H D a

α

+ , µ f (x) = f (x) + cx µ log x

a

α1 ,

(12)

where c is a constant associated with some initial value.

1 α and f (x)

C µ, γ, log [a, b], then the relation H D α + , µ f(x)=0

valids, if and only if,

Lemma 6. If 0 <α< 1, γ

a

f (x) = cx µ log x α1 ,

a

(13)

where c is an arbitrary real number.

Based on Ref. 34, we introduce the following two lemmas.

Lemma 7. If p 1

and a, b 0, then

(a + b) p 2 p1 (a p + b p ).

(14)

1750033-3

on 05/11/17. For personal use only.

from www.worldscientific.com

PITTSBURGH

Downloaded

Fractals OF

by UNIVERSITY

L. Ma & C. P. Li

Lemma 8 (H¨older inequality). Let p > 1 and q satisfies p + 1 q = 1. If u(t) L p (a, b) and v(t) L q (a, b), then u(t)v(t) L(a, b), and

1

t

a

|u(τ)v(τ)|dτ u p v q ,

(15)

where t (a, b].

Next, we present the inequality which has been proved in Ref. 30.

Lemma 9.

we have

log

If λ, υ, ω > 0, then for any t > a, a > 0,

a 1υ t

t

a

log s υ1 log

t

×

s ω ds

a

s

λ ,

a λ1

s

(16)

where C = max{1, 2 1υ }Γ(λ)(1 + λ(λ + 1)).

The following Beesack inequality can be found in Refs. 35 and 36.

Lemma 10 (Beesack inequality). Let a(t), q(t) L(a, b), u(t) and b(t) be real valued contin- uous functions on [a, b], and b(t), q(t) are both non- negative functions, satisfying

u(t) a(t) + q(t)

a

t

b(τ )u(τ )dτ,

t [a, b],

(17)

then we have

u(t) a(t) + q(t)

a

t

a(τ)b(τ)

× exp

τ

t

q(s)b(s)ds dτ.

(18)

Remark 1. Furthermore, if u(t) satisfies the same conditions as Lemma 10, and suppose that a(t), q(t) are both continuous on [a, b], then we have the fol- lowing inequality:

u(t) a(t)

+ q(t)

1 ρ t

1

a

a(s)b(s)ds ,

(19)

b

where ρ =

a

q(s)b(s)ds < 1 is demanded, which

has been proved in Ref. 37.

3. SEMIGROUP AND RECIPROCAL PROPERTIES OF HADAMARD-TYPE FRACTIONAL INTEGRAL AND DERIVATIVE

In this section, we address some basic properties, for instance, semigroup and reciprocal properties on Hadamard-type fractional operators as follows:

Lemma 11. (I) For Hadamard-type fractional inte- gral (1), if α 1, then we obtain

H D

α

a

,

+

µ f(x) =

a

x

x µ f(t) dt .

t

t

(20)

Moreover, as α 0 + , we have

(21)

(II) For Hadamard-type fractional derivative (2),

H

D

, µ f (x) = f (x).

α

a

+

if α 0 + , then we have

H

D

+ , µ f (x) = f (x).

α

a

If α (n 1) + , then

H D

a α + , µ f(x) =

x µ δ n1 (x µ f (x))

(22)

where δ = x

d

dx

. If α n , we have

(δ + µ) n1 f (x),

(23)

H D

α + , µ f(x)

a

= x µ δ n (x µ f (x)) (δ + µ) n f(x),

(24)

where δ = x d

dx . In particular, as α 1, then

H D

a α + , µ f (x) = µf (x) + xf (x)

(δ + µ)f(x).

(25)

In view of the definition of Hadamard-type

fractional integral (1), it is easy to formulate case α = 1. For (21), we can use integration by parts as

follows.

Proof.

lim

α0

+

H D

α

a

,

+

µ f(x)

=

=

lim

α0 +

lim

α0 +

Γ(α)

1

a

x

x µ log x α1 f(t) dt

t

t

t

Γ(1 + α)

1

t

x µ f(t) log x

t

α

x

a

x log x

t

a

α d t

x µ f(t) = f(x).

(26)

1750033-4

on 05/11/17. For personal use only.

from www.worldscientific.com

PITTSBURGH

Downloaded

Fractals OF

by UNIVERSITY

For case (II), combining (2), we can easily get

lim

α0

+

H D

+ , µ f(x)

α

a

=

+ x µ x

lim

α0

dx x µ Γ(1 α) x

d

1

a

×

log x α f(t) dt

t

t

= x µ+1 d(

x

t µ f(t) dt )

a

t

dx

= f (x).

x µ

t

(27)

The proofs of other cases for this lemma can be similarly obtained, so we omit them here. The proof is thus completed.

obtained, so we omit them here. The proof is thus completed. Lemma 12. If 0 <α<

Lemma 12. If 0 <α< 1 and β > 0, then the following relations hold.

H D

H D

a + , µ x µ log

α

x

a

β1

Γ(β)

α) x µ log x β+α1
=

Γ(β +

a

α

a

+

, µ x µ log

x

a

β1

Γ(β)

α) x µ log x βα1
=

Γ(β

a

,

.

(28)

(29)

Proof.

H D

First we prove (28). Based on (1), we have

, µ x µ log x

a

α

a

+

β1

=

Γ(α) x

x

µ

a

log x α1 log

t

a β1 dt

t

t

= x µ (log x Γ(α)

)

β+α1

a

1

0

(1 y) α1 y β1 dy

=

Γ(β)

α) x µ log x β+α1 ,

a

Γ(β +

(30)

where the inner integral term is evaluated by the

and by using

the relation between the Beta function and Gamma function. After a few calculations, we can obtain

formula (29) by utilizing (2) and (28). Thus, we finish this proof.

change of variable y = log a / log x

t

a

this proof. change of variable y = log a / log x t a Remark 2.

Remark 2. For Hadamard fractional integral and derivative, we have analogue conclusions which can be considered as µ = 0 in Lemmas 11 and 12, respectively.

On Hadamard Fractional Calculus

Moreover, motivated by Lemma 2, for 0 <α<β, we have the following two conclusions:

Let n 1 <α<β n Z + ,1 p

, 0 a<b ≤ ∞ and let µ, c R, such that µ c.

Then for f X there holds

b]

Theorem 1.

p

c

(a, b) and H D

α

a

+

,

µ f AC

n

δ;

µ [a,

H

D β

α a + , µ H D a +

,

µ f(x) = H D

(βα) a + , µ

f

(x).

In particular, if β = n, then

H

D

n

a + , µ H D a +

α

,

µ f(x) = H D

(nα) a + , µ

f

(x).

(31)

(32)

Proof. Combining definitions of Hadamard-type fractional integral (1) and derivative (2) with

Lemma 1, we have

H D

β

a + , µ H D a +

α

, µ f(x)

=

=

x µ δ n x µ ( H D

x µ δ n x µ ( H D

(nβ) a + , µ

(n+αβ) a + , µ

H D

α

a

,

+

µ f (x))

f (x))

(33)

The inner term of (33) can be formulated as follows.

x µ δ · δ n1 x µ ( H D

(n+αβ) a + , µ

=

f (x)).

δ n1 x µ ( H D

(n+αβ) a + , µ

f (x))

=

δ n1

Γ(n + α β) x

1

a

t µ

× log x

t

n+αβ1 f(t) dt

t

(n + α β 1)(n + α β 2)

··· (n + α β (n 1))

=

=

Γ(n + α β)

× x t µ log x αβ

a

t

f(t) dt

t

Γ(1 + α β) x

x µ

a

× x µ log x

t

t

(1+αβ)1 f(t) dt

t

(34)

Substituting the above term into (33) and taking (2) into account, we obtain

x µ H D

(1+αβ)

a + , µ

=

f (x).

H

D β

a + , µ H D a +

α

,

µ f(x)

=

x µ δx µ ( H D

(1+αβ)

a +

f

(x))

= H D

(βα) a + , µ

f (x).

(35)

1750033-5

on 05/11/17. For personal use only.

from www.worldscientific.com

PITTSBURGH

Downloaded

OF

Fractals

by UNIVERSITY

L. Ma & C. P. Li

15,

and

. Besides, the

µ [a, b] ensures that

a + , µ f ) exists almost everywhere on

[a, b], which has also been proved in Ref. 15. Thus, this theorem is proved.

If

µ (·),

the

µ

c,

then by Theorem

H D

α

a

+

,

H D

2.1

in Ref.

(·),

(1+αβ)

a + , µ

p

H

D

operators

(n+αβ)

(·)

a + , µ

condition H D

H

D β

a + , µ ( H D

are bounded in X

α

a

+

,

α

µ f

AC

n

δ;

c

bounded in X − α a + , − α µ f ∈ AC n δ

If the difference between α and β is bigger than one, then we have

Theorem 2. Let n 1 < α n m 1 < β

m, n, m Z + , 1 p ≤ ∞, 0 a<b ≤ ∞ and let

p (a, b) and

µ, c R with µ c. Then for f X

µ [a, b] there holds

c

H

D

α

a

+

,

µ f AC

m

δ;

H

D

β

a + , µ H D a +

α

,

µ f(x) = H D

(βα) a + , µ

f

(x).

(36)

In particular, if β = m, then

H

D

m

a + , µ H D

α

,

a

+

µ f(x) = H D

(mα) a + , µ

f

(x).

(37)

The main idea of this proof is identical

with that of Theorem 1. So we just present a gen-

eral sketch of the proof here. First of all, we should observe the variation of β α. From the condition n 1 < α n m 1 < β m, n, m Z + ,

we get m n 1

is [β α] = m n + 1, or [β α] = m n, or [β α] = m n 1, where [θ] is the smallest integer which is bigger than θ. Without loss of generality, we only discuss the case when [β α] = m n.

m n + 1, that

Proof.

<

β α

H

D β

a + , µ H D a +

α

,

µ f(x)

=

x µ δ m x µ ( H D

(m+αβ) a + , µ

f (x)),

f (x)). (38)

Employing the similar technique used in Theorem 1, we can obtain

= x µ δ mn · δ n x µ ( H D

(m+αβ)

a +

δ n x µ ( H D

(m+αβ) a + , µ

f (x))

= x µ H D

((mn)(βα)) a + , µ

f (x).

(39)

In the presence of the conditions f X

α

a

+

,

m

δ;

c

(a, b)

µ [a, b], we substitute (39) into

p

µ f AC

and H D

(38), then we finish this proof.

f ∈ AC and H D ( 38 ), then we finish this proof. By exchange

By exchange of the two operators, we can con- clude a more general theorem than Lemma 2.35 of Ref. 2.

Theorem 3. Let β α > 0, n 1 < α n

Z + , m 1

1 p ≤ ∞ and let µ, c R with µ c. Then

<

β

m

Z + ,0

a<b

,

p

for f AC δ; µ [a, b] and H D α + , µ f X c (a, b), there

n

a

holds

H D

β

a + , µ H D

a + , µ f(x) = H D

α

(βα) a + , µ

f (x).