J. Korean Math. Soc. 38 (2001), No. 6, pp.

11911204

HADAMARD-TYPE FRACTIONAL CALCULUS

Anatoly A. Kilbas

Abstract. The paper is devoted to the study of fractional integra-

tion and differentiation on a finite interval [a, b] of the real axis in
the frame of Hadamard setting. The constructions
Rx under consider-
ation generalize the modified integration a (t/x) f (t)dt/t and the
modified differentiation + ( = xD, D = d/dx) with real , be-
ing taken n times. Conditions are given for such a Hadamard-type
fractional integration operator to be bounded in the space Xcp (a, b)
of Lebesgue measurable functions f on R+ = (0, ) such that
Z b
dt
|tc f (t)|p < (1 p < ),
a t
c
ess supatb [u |f (t)|] < (p = ),
for c R = (, ), in particular in the space Lp (0, ) (1
p ). The existence almost everywhere is established for the
corresponding Hadamard-type fractional derivative for a function
g(x) such that x g(x) have derivatives up to order n 1 on [a, b]
and n1 [x g(x)] is absolutely continuous on [a, b]. Semigroup and
reciprocal properties for the above operators are proved.

1. Introduction

The purpose of this paper is to develop fractional integration and

differentiation in the Hadamard setting. For natural n N = {1, 2, }
and real and a 0 such an aproach is based on the nth integral of the
form
Z x Z Z tn1
n dt1 t1 dt2 dtn
(J,a+ f )(x) = x tn f (tn )
a t1 a t2 a tn
(1.1) Z x
1 t x n1 dt
= log f (t) (x > a)
(n 1)! a x t t

Received June 7, 2001.

2000 Mathematics Subject Classification: 26A33, 47B38.
Key words and phrases: Hadamard-type fractional integration and differentiation,
weighted spaces of summable and absolutely continuous functions.
1192 Anatoly A. Kilbas

and the corresponding derivative

1 d
(D,a+ g)(x) = (( + )f ) (x) = xf 0 (x) + f (x), =x ,
(1.2) dx
n 1 n1
D,a+ g = D,a+ (D,a+ g) (n = 2, 3, ) (x > a).
The fractional versions of the integral (1.1) and the derivative (1.2)
are given by
Z x
1 t x 1 dt
(1.3) (Ja+, f )(x) = log f (t) ( > 0; x > a)
() a x t t
and
d
n
(Da+, g)(x) = x n x J0+, g (x), = x
(1.4) dx
( > 0; n = [] + 1, R),
respectively, [] being integral part of . When = 0, (1.3) and (1.4)
take the forms
Z x
1 x 1 du
(1.5) (Ja+ f )(x) = log f (u) ( > 0; x > a)
() a u u
and

n d
(1.6) (Da+ y)(x) = n Ja+, g (x), = x ( > 0; n = [] + 1).
dx
The integral (1.5) was introduced by Hadamard [5] in the case a = 0 and
therefore Ja+ f and D f are often referred to as Hadamard fractional
a+
integral and derivative of order [6, Section 18.3 and Section 23.1, notes
to Section 18.3]. Therefore we may call the more general constructions
in (1.3) and (1.4) Hadamard-type fractional integral and derivative of
order .
It is well developed an approach to fractional calculus by RRiemann
x
and Liouville based on the generalization of usual integration a f (t)dt
and differentiation D = d/dx, see for example [6, Chapters 2 and 3].
Hadamard fractional calculus approach is studied less. Some facts for
the Hadamard calculus operators (1.5) and (1.6) were presented in [6,
Section 18.3]. The Mellin approach was suggested in [1] to study the

properties of the operators J0+,
and D0+, defined on the the half-axis
R+ = (0, ). Some properties of the operator J0+, and three of its
modifications were invetsigated in [1]-[3].
The aim of this paper is to study the properties of the Hadamard-type
fractional operators (1.3) and (1.4) on a finite interval [a, b] of the real
line R = (, ) for a > 0. The paper is organized as follows. First
 Hadamard-type fractional calculus 1193

f to be bounded
in Section 2 we give conditions for the operator Ja+,
p
in the space Xc (a, b) (c R, 1 p ) of those complex-valued
Lebesgue measurable functions f on [a, b] for which kf kXcp < , where
Z b 1/p
c p dt
(1.7) kf kXcp = |t f (t)| (1 p < , c R)
a t
and
(1.8) kf kXc = ess supatb [tc |f (t)|] (c R).
In particular, when c = 1/p (1 p ), the space Xcp (a, b) coincides
p
with the classical Lp (a, b)-space: Lp (a, b) X1/p (a, b) with
Z b 1/p
p
kf kp = |f (t)| dt (1 p < ),
(1.9) a
kf k = ess supatb |f (t)|.

Next in Section 3 we prove that the Hadamard-type fractional deriv-

ative Da+, g exists almost everywhere for a function g(x) AC;n [a, b]

such that x g(x) have = xD (D = d/dx) derivatives up to order n 1
on [a, b] and n1 [x g(x)] is absolutely continuous on [a, b]:
n
AC; [a, b] = {h : [a, b] C : n1 [x h(x)] AC[a, b],
(1.10) x
R; = x }.
dx
Here AC[a, b] is the set of absolutely continuous functions on [a, b] which
coincide with the space of primitives of Lebesgue measurable functions:
Z x
(1.11) h(x) AC[a, b] h(x) = c + (t)dt, (t) L( a, b),
a
see, for example, [6, (1.4)].
In conclusion in Section 4 we establish semigroup and reciprocal prop-

erties for the operators Ja+,
and Da+, .
We note that the corresponding results for the Hadamard fractional
calculus operators (1.5) and (1.6) are also presented in Sections 2-4.

2. Hadamard-type fractional integration in the space Xcp (a, b)

In this section we show that the Hadamard-type fractional integration

operator Ja+, is defined on Xcp (a, b) for c. To formulate the result
1194 Anatoly A. Kilbas

we need the incomplete gamma-function (, x) defined for > 0 and

x > 0 by [4, 6.9(2)]:
Z x
(2.1) (, x) = t1 et dt.
0

Theorem 2.1. Let > 0, 1 p , 0 < a < b < and let R

and c R be such that c. Then the operator Ja+, is bounded in
p
Xc (a, b) and

(2.2) kJa+, f kXcp Kkf kXcp ,
where

1 b
(2.3) K= log
( + 1) a
for = c, while

1 b
(2.4) K= ( c) , ( c) log
() a
for > c.
Proof. First consider the case 1 p < . Since f (t) Xcp (a, b),
then tc1/p f (t) Lp (a, b) and we can apply the generalized Minkowsky
inequality (see, for example, [6, (1.33)]. In accordance with (1.3) and
(1.7) we have
Z b Z x p 1/p


cp 1 t x 1 dt dx
kJa+, f kXc =
p x log f (t)
a () a x t t x
Z Z !1/p
b x/a x du p
1
= xc1/p u (log u)1 f dx
a () 1 u u
Z b/a Z b x p dx 1/p
1 1 cp
u (log u) x f dt
1 at u x
Z b/a Z !1/p
b/u
c1 1 c p dt
= u (log u) |t f (t)| du
1 a t
and hence

kJa+, f kXcp M kf kXcp ,
where Z b/a
M= uc1 (log u)1 du.
1
 Hadamard-type fractional calculus 1195

Direct calculation show that M coincides with K given in (2.3) and

(2.4), when = c and > c, respectively. Thus (2.2) is proved for
1 p < .
Let now p = . By (1.3) and (1.8) we have
Z x c
c 1 t x 1 c dt
|x (Ja+, f )(x)| log |t f (t)|
() a x t t
and thus
(2.5) |xc (Ja+,

f )(x)| K(x)kf kXc ,
where Z x/a
du
K(x) = u(c) (log u)1 .
1 u
When = c, then for any a x b

1 x 1 b
(2.6) K(x) = log log .
( + 1) a ( + 1) a
If > c, then making the change of variable ( c)u = y and taking
(2.1) into account we find
1 h x i
K(x) = ( c) , ( c) log .
() a
(, x) is increasing function and thus

1 b
(2.7) K(x) ( c) , ( c) log
() a
for any a x b. It follows from (2.5)-(2.7) that for any a x b
(2.8) |xc (Ja+,

f )(x)| Kkf kXc ,
where K is given by (2.3) and (2.4) when = c and > c, respectively.
Hence, in accordance with (1.8), from (2.8) we obtain the result in (2.2)
for p = . This completes the proof of theorem.

Putting c = 1/p in Theorem 2.1 and taking (1.9) into account, we

deduce the boundedness of the operator Ja+, in the space Lp (a, b).

Corollary 2.2. Let > 0, 1 p , 0 < a < b < and let

R be such taht 1/p. Then the operator Ja+, is bounded in
p
L (a, b) and

(2.9) kJa+, f kp K1 kf kp ,
1196 Anatoly A. Kilbas

where K1 is given by (2.3) for = 1/p, while

1 1 1 b
(2.10) K1 = , log
() p p a
for > 1/p.

Setting = 0 in Theorem 2.1 we obtain the corresponding statement

in (1.5).
for the Hadamard fractional operator Ja+

Theorem 2.3. Let > 0, 1 p , 0 < a < b < and let c 0.

is bounded in X p (a, b) and
Then the operator Ja+ c

(2.11) kJa+ f kXcp K2 kf kXcp ,
where

1 b
(2.12) K2 = log
( + 1) a
for c = 0, while

1 b
(2.13) K2 = (c) , c log
() a
for c < 0.

Remark 2.4. It follows from [1, Theorem 4(a)] that if > c, then

the Hadamard-type fractional operator J0+, is bounded in Xcp (R+ ) and

kJ0+, f kXcp K3 kf kXcp .
Such a result formally follow from (2.2) and (2.4) if we put a = 0, b =
and take into account the relation
(2.14) (, ) = ().

Remark 2.5. It follows from Corollary 2.2 that the operator Ja+, is
p
bounded in L (a, b) for 1/p. Similar result can not be obtained from
Theorem 2.3 for the Hadamard fractional operator Ja+ . This fact leads

to conjecture that the operator Ja+ is probably bounded from Lp (a, b)

into another space.

Remark 2.6. The results in Theorem 2.1 and Theorem 2.3 for
Hadamard-type and Hadamard fractional integration operators are ana-
logues of those for the classsical Riemann-Liouville fractional integrals
(see [6, Theorem 2.6]). We only note that the weighted space Xcp (a, b)
is suitable for the former, while the space Lp (a, b) for the latter.
 Hadamard-type fractional calculus 1197

3. Hadamard-type fractional differentiation in the space

n [a, b]
AC;

In this section we give sufficient conditions for the existence of the

Hadamard-type fractional derivative Da+, g in (1.4). Since the result
n
will be state in terms of the space AC; [a, b] defined in (1.10), we first
characterize this space.

Theorem 3.1. The space AC; n [a, b] consists of those and only those

functions g(x), which are represented in the form

" Z x #
1 x n1 X
n1
x k
(3.1) g(x) = x log (t)dt + ck log ,
(n 1)! a t a
k=0

where (t) L1 (a, b) and ck (k = 0, 1, , n1) are arbitrary constants.

Proof. First prove necessity. Let g(x) AC;n [a, b], where = xD

(D = d/dx). Then by (1.10) n1 [x g(x)] AC[a, b] and hence by

(1.11)
Z x
n1
(3.2) [x g(x)] = (t)dt + cn1 ,
a

where (t) L1 (a, b) and cn1 is an arbitrary constant. Rewrite (3.2)

in the form Z
d n2 1 x cn1
[x g(x)] = (t)dt + .
dx x a x
Changing x to t and t to u and integration both sides of this relation we
have Z x
x x
n2 [x g(x)] = log (t)dt + cn2 + cn1 log ,
a t a
where cn2 and cn1 are arbitrary constants. Repeating this procedure
m (1 m n 1) times we obtain
Z x
nm x m1 (t)
[x g(x)] = log dt
a t (m 1)!
(3.3)
cn1+km x k
X m
+ log ,
k! a
k=0

where cnm1 , , cn1 are arbitrary constants. Now (3.3) with m = n

yields (3.1), and necessity is proved.
1198 Anatoly A. Kilbas

Let now g(x) is represented by (3.1), or

Z x
x n1 (t) X x k
n1
x g(x) = log dt + ck log .
a t (n 1)! a
k=0

Taking -derivative m (1 m n 1) times, we have

Z x
x nm1 (t)
m [x g(x)] = log dt
a t (n m 1)!
X k!ck
n1
x km
+ log .
(k m)! a
k=m

From here for m = n 1 we obtain (3.2) (with c = (n 1)!cn1 ) and

n [a, b]. This
hence, in accordance with (1.10) and (1.11), g(x) AC;
completes the proof of theorem.

Note that it follows from our proof that (t) and ck are given by
0 gk (a)
(3.4) (t) = gn1 (t), ck = (k = 0, 1, , n 1),
k!
where
(3.5) gk (x) = k [x g(x)] (k = 0, 1, , n 1), g0 (x) = x g(x).
Hence (3.1) can be rewritten in the form
"Z #
x n1 g 0 (t) n1
X k
x n1 gk (a) x
(3.6) g(x) = x log dt + log ,
a t (n 1)! k! a
k=0

Now we ready to prove the result giving sufficient conditions for the
existence of the Hadamard-type fractional derivative (1.4).

Theorem 3.2. Let > 0, n = [] + 1, R and g(x) AC; n [a, b].

Then the Hadamard-type fractional derivative Da+, g exists almost ev-
erywhere on [a, b] and may be represented in the form
h Z x
1 x n1 0
(Da+, g)(x) = x log gn1 (t)dt
(n ) a t
(3.7) x k i
n1
X gk (a)
+ log ,
(k + 1) a
k=0

where gk (a) (k = 0, 1, , n 1) are given by (3.5).

 Hadamard-type fractional calculus 1199

Proof. Since g(x) AC;n [a, b], we have representation (3.6). Substi-

tuting this relation into (1.4) we have

Z x
1 x n1
(Da+, g)(x) = x n log
(n ) a t
hZ t t
n1 0
gn1 (u)
(3.8) log du
a u (n 1)!
X gk (a)
n1
t k i dt
+ log .
k! a t
k=0
Interchanging the order of integration and applying the Dirichlet formula
(see, for example, [6, (1.33)]) we have
Z x Z
x n1 dt t t n1 0
log log gn1 (u)du
a t t a u
Z x Z x
0 x n1 t n1 dt
= gn1 (u)du log log .
a u t u t
The inner integral is evaluated by the change of variable y =
log(t/u)/ log(x/u) and using the formulas [4, 1.5(1) and 1.5(5)] for the
beta function:
Z x
x n1 t n1 dt (n )(n) x n1
log log = log
u t u t (2n ) u
and hence
Z x Z
x n1 dt t t n1 0
log log gn1 (u)du
a t t a u
Z
(n )(n) x x 2n1 0
= log gn1 (u)du.
(2n ) a u
Substituting this relation into (3.8) and taking n -differentiation, we
obtain (3.7). Thus theorem is proved.

Corollary 3.3. If 0 < < 1, R and g(x) AC; 1 [a, b], then

Da+, g
exists almost everywhere on [a, b] and
Z
x h x x dt
(Da+, g)(x) = log [t g(t)]0
(1 ) a t t
(3.9) i
x
+ lim [t g(t)] log .
ta+ a
1200 Anatoly A. Kilbas

When = 0, from Theorem 3.2 we deduce sufficient conditions for

the existence of the Hadamard fractional derivative (1.6).

Theorem 3.4. Let > 0, n = [] + 1 and g(x) AC;0 n [a, b]. Then

the Hadamard fractional derivative Da+ g exists almost everywhere on
[a, b] and may be represented in the form
Z x
1 x n1 n
(Da+ g)(x) = log ( g)(t)dt
(n ) a t
(3.10) X ( k g)(a) x k
n1
+ log .
(k + 1) a
k=0

Corollary 3.5. If 0 < < 1 and g(x) AC;0 1 [a, b], then D g
a+
exists almost everywhere on [a, b] and
Z x
1 x 0 dt
(Da+ g)(x) = log g (t)
(1 ) a t t
(3.11)
g(a) x
+ log .
(1 ) a

Remark 3.6. The results in Theorem 3.2 and Theorem 3.4 for
Hadamard-type and Hadamard fractional differentiation operators are
analogues of those for the classsical Riemann-Liouville fractional deriva-
tives (see [6, Theorem 2.2]). We only note that the weighted space
n [a, b] is suitable for the former, while the space AC n [a, b] for the
AC;
latter.

4. Semigroup and reciprocal properties of Hadamard-type

fractional calculus operators

First we prove the semigroup property for the Hadamard-type frac-

tional integration operator Ja+, in (1.3).

Theorem 4.1. Let > 0, > 0, 1 p , 0 < a < b < and

let R and c R be such that c. Then for f Xcp (a, b) the
semigroup property holds
+
(4.1) Ja+, Ja+, f = Ja+, f.
 Hadamard-type fractional calculus 1201

Proof. First we prove (4.1) for sufficiently good functions f . Ap-

plying Fubinis theorem we find

(Ja+, Ja+, f )(x)
Z x
1 u x 1 du
= log
() a x u u
Z u
(4.2) 1 t u 1 dt
log f (t)
() a u t t
Z x Z x
x x 1
u 1 du
= t1 f (t)dt log log .
()() a t u t u
The inner integral is evaluated by the change of variable = log (u/t)/
log(x/t):
Z x
x 1 u 1 du x +1 ()()
log log = log .
t u t u t ( + )
Substituting this relation into (4.2) and taking (1.3) into account we
have
Z x
1 u x +1 du
(Ja+, Ja+, f )(x) = log
( + ) a x u u
+
= (Ja+, f )(x),
and thus (4.1) is proved for sufficiently good functions f .
If c, then by Theorem 2.1 the operators Ja+, , J +
a+, and Ja+,
are bounded in Xcp (a, b), hence the relation (4.1) is true for f Xcp (a, b).
This completes the proof of theorem.

Corollary 4.2. Let > 0, > 0, 1 p , 0 < a < b < and

let R be such that 1/p. Then for f Lp (a, b) the semigroup
property (4.1) holds.

When m = 0, from Theorem 4.1 and Theorem 2.3 we obtain the

semigroup property for the Hadamard fractional integration operators
(1.5).

Theorem 4.3. Let > 0, > 0, 1 p , 0 < a < b < and

c 0. Then for f Xcp (a, b) the semigroup property holds
+
(4.3) Ja+ Ja+ f = Ja+ f.
1202 Anatoly A. Kilbas

Next we consider the composition between the operators of Hadamard-

type fractional differentiation (1.4) and fractional integration (1.3).

Theorem 4.4. Let > > 0, 1 p , 0 < a < b < and let
R and c R be such that c. Then for f Xcp (a, b) there holds

(4.4) Da+, Ja+, f = Ja+, f.
In particular, if = m N, then

m m
(4.5) Da+, Ja+, f = Ja+, f.

Proof. Let m 1 < m (m N). If = m, then by (1.4)

m d m
(4.6) (Da+, y)(x) = x x x y(x)
dx
and hence
Z x
m d m1 d 1 x 1 du
(Da+, Ja+, f )(x) =x x x u log f (u) .
dx dx () a u u
Applying the formula of differentaition under the integral sign and using
the relation for the gamma-function [4, 1.2(1)] and (1.3) we obtain
m
(Da+, Ja+, f )(x)
m1 Z x
d x x 1 du
=x x u log f (u)
dx () a x u u
m1 Z x
d 1 x 2 du
= x x u log f (u)
dx ( 1) a x u u
m1
d 1
= x x x Ja+, f )(x).
dx
Repeating this procedure k (1 k m) times we have

m d mk k
(Da+, Ja+, f )(x) = x x x (Ja+, f )(x),
dx
and (4.5) follows for k = m.
If m 1 < < m, then (4.4) follows from (4.1) and (4.5):
m m m m+
Da+, Ja+, f = Da+, Ja+, Ja+, f = Da+, Ja+, f = Ja+, f.
Thus theorem is proved.
 Hadamard-type fractional calculus 1203

Corollary 4.5. Let > > 0, 1 p , 0 < a < b < and

let R be such that 1/p. Then for f Lp (a, b) the relation (4.4)
holds. In particular, (4.5) is valid for = m N.

Theorem 4.6. Let > > 0, 1 p , 0 < a < b < and

c 0. Then for f Xcp (a, b) the relation holds

(4.7) Da+ Ja+ f = Ja+ f.
In particular, if = m N, then
m m
(4.8) Da+ Ja+ f = Ja+ f.

Theorem 4.4 is also true for = which means that the Hadamard-
type fractional differentiation (1.6) and integration (1.5) are reciprocal
operations if the former is applied first. The result below is proved
similarly to the proof of Theorem 4.4.

Theorem 4.7. Let > 0, 1 p , 0 < a < b < and let R

and c R be such that c. Then for f Xcp (a, b) there holds

(4.9) Da+, Ja+, f = f.
In particular, if 1/p, then (4.10) is valid for f Lp (a, b).

Theorem 4.8. Let > 0, 1 p , 0 < a < b < and let c 0.

Then for f Xcp (a, b) there holds

(4.10) Da+ Ja+ f = f.

Remark 4.9. It follows from [2, Theorem 1(a)] that if > 0, > 0,
1 p and > c, then for f Xcp (R+ ) the semigroup property
+
(4.11) J0+, J0+, f = J0+, f
holds. Such a result follows from (4.1) if we put a = 0 and b =
, J
and take into account that the operators J0+, +
0+, and J0+, are
p
bounded in Xc (R+ ) when > c.

Remark 4.10. The results presented in Theorems 4.7 and 4.8 show
that the Hadamard-type and Hadamard fractional differentiation (1.4)
and (1.6) and integration (1.3) and (1.5) are reciprocal operations if
the formers are applied first. It is the problem when the latters can
1204 Anatoly A. Kilbas

be applied first. Such a problem is solved for the Riemann-Liouville

fractional calculus operators (see [6, Theorem 2.4]).

Acknowledgement. The present investigation was partly sup-

ported by the Belarusian Fundamental Research Fund.

References
[1] P. L. Butzer, A. A. Kilbas, and J. Trujillo, Fractional calculus in the Mellin
setting and Hadamard-type fractional integrals, J. Math. Anal. Appl. (to appear)
[2] , Compositions of Hadamard-type fractional integration operators and the
semigroup property, J. Math. Anal. Appl. (to appear)
[3] , Mellin transform and integration by parts for Hadamard-type fractional
integrals, J. Math. Anal. Appl. (to appear)
[4] A. Erdelyi, W. Magnus, F. Oberhettinger, and F.G. Tricomi, Higher Transcen-
dental Functions, Vol. 1, McGraw-Hill Boo. Coop., New York, 1953; Reprinted
Krieger, Melbourne, Florida, 1981.
[5] J. Hadamard, Essai sur letude des fonctions donnees par leur developpment de
Taylor, J. Math. Pures et Appl., Ser. 4 8 (1892), 101186.
[6] S. G. Samko, A. A. Kilbas, and O. I. Marichev, Fractional Integrals and Deriva-
tives. Theory and Applications, Gordon and Breach, Yverdon et alibi, 1993.

Department of Mathematics and Mechanics

Belarusian State University
220050 Minsk, Belarus
E-mail : [email protected]