See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/260016640

A new definition of fractional derivative

Article  in  Journal of Computational and Applied Mathematics · July 2014

DOI: 10.1016/j.cam.2014.01.002

CITATIONS READS

439 5,829

4 authors:

Roshdi R Khalil M. Al Horani H Horani

University of Jordan University of Jordan
232 PUBLICATIONS   1,051 CITATIONS    27 PUBLICATIONS   579 CITATIONS   

SEE PROFILE SEE PROFILE

Abdelrahman Yousef M. Sababheh

University of Jordan Princess Sumaya University for Technology
9 PUBLICATIONS   457 CITATIONS    51 PUBLICATIONS   608 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

conformable differential equations View project

Analyticalapproximations of partial di_erential equations of fractional order with multistep approach,Journal of Computational and Theoretical Nanoscience View
project

All content following this page was uploaded by M. Sababheh on 03 November 2017.

The user has requested enhancement of the downloaded file.

 Journal of Computational and Applied Mathematics 264 (2014) 65–70

Contents lists available at ScienceDirect

Journal of Computational and Applied

Mathematics
journal homepage: www.elsevier.com/locate/cam

A new definition of fractional derivative

R. Khalil a,∗ , M. Al Horani a , A. Yousef a , M. Sababheh b,c
a
Department of Mathematics, The University of Jordan, Al Jubaiha, Amman 11942, Jordan
b
Department of Basic Sciences, Princess Sumaya University For Technology, Al Jubaiha, Amman 11941, Jordan
c
Department of Mathematics, Sharjah University, Sharjah, United Arab Emirates

article info abstract

Article history: We give a new definition of fractional derivative and fractional integral. The form of the
Received 28 May 2013 definition shows that it is the most natural definition, and the most fruitful one. The
Received in revised form 31 October 2013 definition for 0 ≤ α < 1 coincides with the classical definitions on polynomials (up to
a constant). Further, if α = 1, the definition coincides with the classical definition of first
MSC: derivative. We give some applications to fractional differential equations.
26A33
© 2014 Elsevier B.V. All rights reserved.
Keywords:
Fractional derivative
Fractional integral

1. Introduction
dn f
Fractional derivative is as old as calculus. L’Hospital in 1695 asked what does it mean dxn if n = 12 . Since then, many
researchers tried to put a definition of a fractional derivative. Most of them used an integral form for the fractional derivative.
Two of which are the most popular ones.
(i) Riemann–Liouville definition. For α ∈ [n − 1, n), the α derivative of f is
dn f (x) t

1
Dαa (f )(t ) = dx.
Γ (n − α) dt(t − x)α−n+1
n
a

(ii) Caputo definition. For α ∈ [n − 1, n), the α derivative of f is

f (n) (x)
 t
1
Dαa (f )(t ) = dx.
Γ (n − α) a (t − x)α−n+1

Now, all definitions including (i) and (ii) above satisfy the property that the fractional derivative is linear. This is the only
property inherited from the first derivative by all of the definitions. However, the following are some of the setbacks of the
other definitions:
(i) The Riemann–Liouville derivative does not satisfy Dαa (1) = 0 (Dαa (1) = 0 for the Caputo derivative), if α is not a natural
number.
(ii) All fractional derivatives do not satisfy the known formula of the derivative of the product of two functions:
Dαa (fg ) = fDαa (g ) + gDαa (f ).

∗
Corresponding author.
E-mail addresses: [email protected] (R. Khalil), [email protected] (M. Al Horani), [email protected] (A. Yousef), [email protected],
[email protected] (M. Sababheh).

0377-0427/$ – see front matter © 2014 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.cam.2014.01.002
66 R. Khalil et al. / Journal of Computational and Applied Mathematics 264 (2014) 65–70

(iii) All fractional derivatives do not satisfy the known formula of the derivative of the quotient of two functions:
gDαa (f ) − fDαa (g )
Dαa (f /g ) = .
g2
(iv) All fractional derivatives do not satisfy the chain rule:
Dαa (f ◦ g )(t ) = f (α) g (t ) g (α) (t ).
 

(v) All fractional derivatives do not satisfy: Dα Dβ f = Dα+β f in general.

(vi) The Caputo definition assumes that the function f is differentiable.

Our interest in fractional derivatives started when Professor S. Momani showed us how to solve the differential equation
 
1 1 2 3
y 2
+ y = x2 + x2 ; y(0) = 0
Γ (2.5)
1
where y( 2 ) is the fractional derivative of y of order 12 . The known solution for this differential equation is not easy to be
obtained, so we thought that a new definition of fractional derivative may facilitate some computations.
The object of this paper is to present a new, yet an easy definition of fractional derivative. The new definition seems to
be a natural extension of the usual derivative, and it satisfies the first four properties mentioned above. Our definition coincides
with the known fractional derivatives on polynomials (up to a constant multiple).
For the history and main results on fractional derivatives and fractional differential equations, we refer the reader
to [1–4].

2. The definition

f (t +ε)−f (t )
Let f : [0, ∞) → R and t > 0. Then the definition of the derivative of f at t is
df
dt
= limε→0 ε
. According
dt n
to this, one has dt = nt . So the question is: Can one put a similar definition for the fractional derivative of order α ,
n −1

where 0 < α ≤ 1? Or in general for α ∈ (n, n + 1] where n ∈ N.

Let us write Tα to denote the operator which is called the fractional derivative of order α . For α = 1, T1 satisfies the
following properties:
(i) T1 (af + bg ) = aT1 (g ) + bT1 (f ), for all a, b ∈ R and f , g in the domain of T1 .
(ii) T1 (t p ) = pt p−1 for all p ∈ R.
(iii) T1 (fg ) = fT1 (g ) + gT1 (f ).
gT (f )−fT (g )
T1 ( g ) = 1 g 2 1 .
f
(iv)
(v) T1 (λ) = 0, for all constant functions f (t ) = λ.
Now, we present our new definition, which is the simplest and most natural and efficient definition of fractional
derivative of order α ∈ (0, 1]. We should remark that the definition can be generalized to include any α . However, the
case α ∈ (0, 1] is the most important one, and once it is established, the other cases are simple.

Definition 2.1. Given a function f : [0, ∞) −→ R. Then the ‘‘conformable fractional derivative’’ of f of order α is defined by

f (t + ε t 1−α ) − f (t )
Tα (f )(t ) = lim
ε→0 ε
for all t > 0, α ∈ (0, 1). If f is α -differentiable in some (0, a), a > 0, and limt →0+ f (α) (t ) exists, then define
f (α) (0) = lim f (α) (t ).
t →0+

We will, sometimes, write f (α) (t ) for Tα (f )(t ), to denote the conformable fractional derivatives of f of order α . In addition,
if the conformable fractional derivative of f of order α exists, then we simply say f is α -differentiable.
We should remark that Tα (t p ) = p t p−α . Further, our definition coincides with the classical definitions of R–L and of
Caputo on polynomials (up to a constant multiple).
As a consequence of the above definition, we obtain the following useful theorem.

Theorem 2.1. If a function f : [0, ∞) −→ R is α -differentiable at t0 > 0, α ∈ (0, 1], then f is continuous at t0 .
1−α
f (t0 +ε t0 )−f (t0 )
Proof. Since f (t0 + ε t01−α ) − f (t0 ) = ε
ε . Then,
f (t0 + ε t01−α ) − f (t0 )
lim [f (t0 + ε t01−α ) − f (t0 )] = lim · lim ε.
ε→0 ε→0 ε ε→0
 R. Khalil et al. / Journal of Computational and Applied Mathematics 264 (2014) 65–70 67

Let h = ε t01−α . Then,

lim [f (t0 + h) − f (t0 )] = f (α) (t0 ) · 0,

h→0

which implies that

lim f (t0 + h) = f (t0 ).
h→0

Hence, f is continuous at t0 . 
One can easily show that Tα satisfies all the properties in the following theorem.

Theorem 2.2. Let α ∈ (0, 1] and f , g be α -differentiable at a point t > 0. Then

(1) Tα (af + bg ) = aTα (f ) + bTα (g ), for all a, b ∈ R.
(2) Tα (t p ) = pt p−α for all p ∈ R.
(3) Tα (λ) = 0, for all constant functions f (t ) = λ.
(4) Tα (fg ) = fTα (g ) + gTα (f ).
gT (f )−fT (g )
(5) Tα ( g ) = α g 2 α .
f

(6) If, in addition, f is differentiable, then Tα (f )(t ) = t 1−α dt (t ).

df

Proof. Parts (1) through (3) follow directly from the definition. We choose to prove (4) and (6) only since they are crucial.
Now, for fixed t > 0,
f (t + ε t 1−α )g (t + ε t 1−α ) − f (t )g (t )
Tα (fg )(t ) = lim
ε→0 ε
f (t + ε t 1−α )g (t + ε t 1−α ) − f (t )g (t + ε t 1−α ) + f (t )g (t + ε t 1−α ) − f (t )g (t )
= lim
ε→0 ε
f (t + ε t 1−α ) − f (t ) g (t + ε t 1−α ) − g (t )
 
= lim .g (t + ε t 1−α ) + f (t ) lim
ε→0 ε ε→0 ε
= Tα (f )(t ) lim g (t + ε t 1−α ) + f (t )Tα (g )(t ).
ε→0

Since g is continuous at t, limε→0 g (t + ε t 1−α ) = g (t ). This completes the proof of part (4). (5) can be proved in a similar
way. To prove (6), let h = ε t 1−α in Definition 2.1, and then ε = t α−1 h. Therefore,
f (t + ε t 1−α ) − f (t )
Tα (f )(t ) = limε→0
ε
f (t + h) − f (t )
= lim
h→0 ht α−1
f (t + h) − f (t )
= t 1−α lim
h→0 h
1−α df
=t (t ). 
dt

Conformable fractional derivative of certain functions

(1) Tα (t p ) = pt p−α for all p ∈ R.

(2) Tα (1) = 0.
(3) Tα (ecx ) = cx1−α ecx , c ∈ R.
(4) Tα (sin bx) = bx1−α cos bx, b ∈ R.
(5) Tα (cos bx) = −bx1−α sin bx, b ∈ R.
(6) Tα ( α1 t α ) = 1.
However, it is worth noting the following conformable fractional derivatives of certain functions:
(i) Tα (sin α1 t α ) = cos α1 t α .
(ii) Tα (cos α1 t α ) = − sin α1 t α .
1 α 1 α
(iii) Tα (e α t ) = e α t .
√
One should notice that a function could be α -differentiable at a point but not differentiable, for example, take f (t ) = 2 t.
Then T 1 (f )(0) = limt →0+ T 1 (f )(t ) = 1, where T 1 (f )(t ) = 1, for t > 0. But T1 (f )(0) does not exist. This is not the case for
2 2 2
the known classical fractional derivatives.
Although the most important case for the range of α is (0, 1), but, what if α ∈ (n, n + 1] for some natural number n?
What would be the definition?
68 R. Khalil et al. / Journal of Computational and Applied Mathematics 264 (2014) 65–70

Definition 2.2. Let α ∈ (n, n + 1], and f be an n-differentiable at t, where t > 0. Then the conformable fractional derivative
of f of order α is defined as
f (⌈α⌉−1) (t + ε t (⌈α⌉−α) ) − f (⌈α⌉−1) (t )
Tα (f )(t ) = lim
ε→0 ε
where [α] is the smallest integer greater than or equal to α .

Remark 2.1. As a consequence of Definition 2.2, one can easily show that
Tα (f )(t ) = t (⌈α⌉−α) f ⌈α⌉ (t )
where α ∈ (n, n + 1], and f is (n + 1)-differentiable at t > 0.

Analysis of the definition

The previous definitions of fractional derivative do not enable us to study the analysis of α -differentiable functions.
However, our definition makes it possible to prove basic analysis theorems like the Rolle’s theorem and the mean value
theorem.

Theorem 2.3 (Rolle’s Theorem for Conformable Fractional Differentiable Functions). Let a > 0 and f : [a, b] → R be a given
function that satisfies
(i) f is continuous on [a, b],
(ii) f is α -differentiable for some α ∈ (0, 1),
(iii) f (a) = f (b).
Then, there exists c ∈ (a, b), such that f (α) (c ) = 0.
Proof. Since f is continuous on [a, b], and f (a) = f (b), there is c ∈ (a, b), which is a point of local extrema. With no loss of
generality, assume c is a point of local minimum. So
f (c + ε c 1−α ) − f (c ) f (c + ε c 1−α ) − f (c )
f (α) (c ) = lim = lim .
ε→0+ ε ε→0− ε
But, the first limit is non-negative, and the second limit is non-positive. Hence f (α) (c ) = 0. 

Theorem 2.4 (Mean Value Theorem for Conformable Fractional Differentiable Functions). Let a > 0 and f : [a, b] → R be a
given function that satisfies
(i) f is continuous on [a, b].
(ii) f is α -differentiable for some α ∈ (0, 1).
f (b)−f (a)
Then, there exists c ∈ (a, b), such that f (α) (c ) = 1 α 1 α .
αb −αa

Proof. Consider the function

f (b) − f (a)
 
1 α 1
g (x) = f (x) − f (a) − x − aα .
1
α
bα − α1 aα α α
Then the function g satisfies the conditions of Rolle’s theorem. Hence there exists c ∈ (a, b), such that g (α) (c ) = 0. Using
the fact that Tα ( α1 t α ) = 1, the result follows. 
Along the same lines in basic analysis, one can use the present mean value theorem to prove the following proposition.

Proposition 2.1. Let f : [a, b] → R be α -differentiable for some α ∈ (0, 1).

(a) If f (α) is bounded on [a, b] where a > 0. Then f is uniformly continuous on [a, b], and hence f is bounded.
(b) If f (α) is bounded on [a, b] and continuous at a. Then f is uniformly continuous on [a, b], and hence f is bounded.
It is well-known that if f ′ (t ) is bounded
√ on I = [a, b], then f is uniformly continuous on I. However, the converse need
not be true. To see this, consider f (t ) = 2 t on I = [0, 1]. Then f is uniformly continuous on [0, 1] but f ′ (t ) is not bounded
there. However, boundedness of f (α) (t ) for 0 < α < 1 and the continuity of f on I (continuity of f at 0 in the subspace
topology is equivalent to right continuity of f at 0), which implies, by the above proposition, the uniform continuity of f
on I.

3. Fractional integral

When it comes to integration, the most important class of functions to define the integral is the space of continuous
functions.
So, using the Weierstrass theorem, it is enough to define the fractional integral on polynomials. This suggests the
following.
 R. Khalil et al. / Journal of Computational and Applied Mathematics 264 (2014) 65–70 69

t p+α
Let α ∈ (0, ∞). Define Jα (t p ) = p+α for any p ∈ R, and α ̸= −p.
t k+α
If f (t ) = k=0 bk t , then we define Jα (f ) = k=0 bk Jα (t ) =
n k
n k
n
k=0 bk k+α .
k+α
If f (t ) = k=0 bk t , where the series is uniformly convergent, then we define Jα (f ) =
∞ ∞
k
k=0 bk tk+α .
Clearly, Jα is linear on its domain. Further, if α = 1, then Jα is the usual integral.
Now, according to our definition, if α = 21 , then
3
∞
 (−1)n t 2n+ 2
Jα (sin t ) = ,
n =0
(2n + 23 )(2n + 1)!
similarly for cos t and et , and for any α ∈ (0, 1).
These examples suggest the following definition for the α -fractional integral of a function f starting from a ≥ 0.
f (x)
Definition 3.1. Iαa (f )(t ) = I1a (t α−1 f ) =
t
a x1−α
dx, where the integral is the usual Riemann improper integral, and
α ∈ (0, 1).
t √ √
cos x dx = sin t , and I 01 cos 2 t = sin 2 t.
√  
So, I 01 t cos t = 0
2 2
One of the nice results is the following.

Theorem 3.1. Tα Iαa (f )(t ) = f (t ), for t ≥ a, where f is any continuous function in the domain of Iα .
Proof. Since f is continuous, then Iαa (f )(t ) is clearly differentiable. Hence,
d a
Tα (Iαa (f ))(t ) = t 1−α Iα (f )(t )
dt
f (x)
 t
1−α d
=t dx
dt a x1−α
1−α f (t )
=t
t 1−α
= f (t ). 

4. Applications

Now we will solve fractional differential equations according to our definitions.

Example 4.1. y(1/2) + y = x2 + 2x3/2 , y(0) = 0. This is the equation that we mentioned in the Introduction, approximating
Γ (2.5) = 1.33 to 1.
√
(1/2)
Let us find a solution√yh of the
√ homogeneous equation y + y = 0. We look for a solution of√the form yh = er x .
y(1/2) + y = 0 H⇒ 2r er x + er x = 0 H⇒ 12 r + 1 = 0, the auxiliary equation. Hence, yh = e−2 x . It is easy to verify that
yp (x) = x2 is a particular solution of the nonhomogeneous equation.
√
Now, the general solution is y(x) = yh (x) + yp (x) = Ae−2 x
+ x2 , where A is constant. Finally, the initial condition
y(0) = 0 implies that A = 0. Hence, y(x) = x2 .
3
2
We should remark here that such an equation is a known one that people solve using the R–L derivative, with Γ (2.5)
x2
replacing 2x3/2 , and we get the same solution, but our method using our definition is much more easier.

Example 4.2. One can easily show that the auxiliary equation for y(α) + y = 0, 0 < α ≤ 1 is α r + 1 = 0, so that the
1 α
solution is given by y(x) = e− α x . More details including the method of undetermined coefficients will be discussed in a
forthcoming paper.
In the following example we will show the benefit of the fractional derivative product rule which allows us to use the
idea of the integrating factor.
√
Example 4.3. y(1/2) + x y = xe−x .
√
We solve this equation by multiplying it by eI1/2 ( x)
= ex . Then
√ 2
ex y(1/2) + xex y = x H⇒ (ex y)(1/2) = x H⇒ ex y(x) = x3/2 + C
3
H⇒ y(x) = 23 x3/2 e−x + Ce−x , C is constant, which can be easily verified to be a solution of the above equation.
70 R. Khalil et al. / Journal of Computational and Applied Mathematics 264 (2014) 65–70

3 √
1
Example 4.4. y( 2 ) =
x2 +y x
2x+3y
.
1 √
We assume that we are looking for a differentiable y. So, by (i) in Theorem 2.2, we have y( 2 ) = xdy/dx.
3 √
′
√ x 2 +y x ′ x+y
Hence, the fractional differential equation becomes y x = 2x+3y
. So, y = 2x+3y
. This is a homogeneous differential
equation of order one and can be solved easily.

Closing remarks
(i) We believe that the fractional derivative presented here, for 0 < α < 1, is local by nature. Here the following question
is raised: Can its physical meaning be interpreted easier than those previously presented in history?
(ii) Note that the derivative presented here does not have any delay effect but the other fractional ones have (there is a
kernel in the definition inside the integral)!.
(iii) Consider the very simple differential equation y(1/2) + y = 0. If one has to solve it using the Caputo or Riemann–Liouville
definition, then he/she must use either the Laplace transform or the fractional √
power series technique. However, using
1 α 1 α
our definition and the fact Tα (e α t ) = e α t , one can easily see that y = ce−2 t
is the general solution.
1 1
(iv) According to one of the classical definitions, D (sin ax) = a sin(ax + π /4). So what if a = −1? Now according to our
2 2

definition Tα (sin α1 t α ) = cos α1 t α . To evaluate Dα (sin α1 t α ) using the classical definitions, it is not an easy job.

Acknowledgments

The authors would like to thank Professor S. Momani for introducing them to the subject of fractional calculus.
Also, the authors are thankful to the referee for his invaluable suggestions, which put the article in its present shape.

References

[1] K.S. Miller, An Introduction to Fractional Calculus and Fractional Differential Equations, J. Wiley and Sons, New York, 1993.
[2] K. Oldham, J. Spanier, The Fractional Calculus, Theory and Applications of Differentiation and Integration of Arbitrary Order, Academic Press, USA, 1974.
[3] A. Kilbas, H. Srivastava, J. Trujillo, Theory and Applications of Fractional Differential Equations, in: Math. Studies., North-Holland, New York, 2006.
[4] I. Podlubny, Fractional Differential Equations, Academic Press, USA, 1999.

View publication stats