JOURNAL OF THE

AMERICAN MATHEMATICAL SOCIETY

Volume 00, Number 0, Pages 000–000
S 0894-0347(XX)0000-0

A NEW FRACTIONAL DERIVATIVE WITH CLASSICAL PROPERTIES

UDITA N. KATUGAMPOLA
arXiv:1410.6535v2 [math.CA] 8 Nov 2014

Abstract. We introduce a new fractional derivative which obeys classical properties including: linearity,
product rule, quotient rule, power rule, chain rule, vanishing derivatives for constant functions, the Rolle’s
Theorem and the Mean Value Theorem. The definition,
−α
f (teǫt ) − f (t)
D α (f )(t) = lim ,
ǫ→0 ǫ
is the most natural generalization that uses the limit approach. For 0 ≤ α < 1, it generalizes the
classical calculus properties of polynomials. Furthermore, if α = 1, the definition is equivalent to the
classical definition of the first order derivative of the function f . Furthermore, it is noted that there are
α−differentiable functions which are not differentiable.

The paper is dedicated to my school teacher

T.H. Gunaratna Silva

1. Introduction
The derivative of non-integer order has been an interesting research topic for several centuries. The
n
idea was motivated by the question, “What does it mean by ddxnf , if n = 12 ?”, asked by L’Hospital in
1695 in his letters to Leibniz [5–7]. Since then, the mathematicians tried to answer this question for
centuries in several points of view. The outcomes are many folds. Various types of fractional derivatives
were introduced: Riemann-Liouville, Caputo, Hadamard, Erdélyi-Kober, Grünwald-Letnikov, Marchaud
and Riesz are just a few to name [1,2,4,9–11]. Most of the fractional derivatives are defined via fractional
integrals [11]. Due to the same reason, those fractional derivatives inherit some non-local behaviors,
which lead them to many interesting applications including memory effects and future dependence [8].
Among the inconsistencies of the existing fractional derivatives are:
(1) Most of the fractional derivatives except Caputo-type derivatives, do not satisfy Daα (1) = 0 , if α
is not a natural number.
(2) All fractional derivatives do not obey the familiar Product Rule for two functions:
Daα (f g) = f Daα (g) + gDaα (f ).
(3) All fractional derivatives do not obey the familiar Quotient Rule for two functions:
 f  gDα (f ) − f Dα (g)
a a
Daα = .
g g2
(4) All fractional derivatives do not obey the Chain Rule:
Daα (f ◦ g)(t) = f (α) g(t) g (α) (t).

2010 Mathematics Subject Classification. 23A33 .

Key words and phrases. Fractional Derivatives, Classical Calculus, Product Rule, Quotient Rule, Chain Rule, Rolle’s
theorem, Mean Value Theorem.
Department of Mathematical Sciences, University of Delaware, Newark DE 19716, USA.
Email: [email protected] Preprint submitted to J. Amer. Math. Soc.

c
XXXX American Mathematical Society

1
2 U.N. KATUGAMPOLA

(5) Fractional derivatives do not have a corresponding Rolle’s Theorem.

(6) Fractional derivatives do not have a corresponding Mean Value Theorem.
(7) All fractional derivatives do not obey: Daα Daβ f = Daα+β f , in general.
(8) The Caputo definition assumes that the function f is differentiable.
To overcome some of these and other difficulties, Khalil et al. [3], came up with an interesting idea
that extends the familiar limit definition of the derivative of a function given by the following.
Definition 1.1. [3] Let f : [0, ∞) → R and t > 0. Then the “conformable fractional derivative” of f of
order α is defined by,
f (t + ǫt1−α ) − f (t)
(1.1) f (α) (t) = lim ,
ǫ→0 ǫ
for 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+

As a consequence of the above definition, the authors in [3], showed that the α−derivative in (1.1),
obeys the Product rule, Quotient rule and has results similar to the Rolle’s Theorem and the Mean Value
Theorem in classical calculus. The purpose of this work is to further generalize the results obtained in [3]
and introduce a new fractional derivative as the most natural extension of the familiar limit definition of
the derivative of a function f at a point.

2. New Fractional Derivative

In this section, we give the main definition of the paper and obtain several results that are close
resemblance of the results found in classical calculus. We prove that the new fractional derivative satisfies
the Product rule, Quotient rule, Chain rule and have results which are natural extensions of the Rolle’s
theorem and the Mean Value Theorem. To this end, we start with the following definition, which is a
generalization of the limit definition of the derivative.
Definition 2.1. Let f : [0, ∞) → R and t > 0. Then the fractional derivative of f of order α is defined
by,
f (teǫt
−α
) − f (t)
(2.1) Dα (f )(t) = lim ,
ǫ ǫ→0

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

The first result is the generalization of Theorem 2.1 of [3].

Theorem 2.2. If a function f : [0, ∞) → R is α−differentiable at a > 0, α ∈ (0, 1], then f is continuous
at a.
−α
f (aeǫa )−f (a)
Proof. Since f (aeǫa
−α
) − f (a) = ǫ ǫ, we have
f (aeǫa ) − f (a)
−α

lim f (aeǫa ) − f (a) = lim

 −α 
. lim ǫ.
ǫ→0 ǫ→0 ǫ ǫ→0

Let h = ǫa1−α + O(ǫ2 ). Then, limh→0 f (a + h) − f (a) = Dα (f )(a) · 0 = 0, which, in turn, implies that
 

limh→0 f (a + h) = f (a). This completes the proof. 

The following is the main result of this paper.
 A NEW FRACTIONAL DERIVATIVE WITH CLASSICAL PROPERTIES 3

Theorem 2.3. Let α ∈ (0, 1] and f, g be α−differentiable at a point t > 0. Then,

(1) Dα af
+ bg = aDα (f ) + bDα (g), for all a, b ∈ R.
 

(2) Dα tn
= ntn−α for all n ∈ R.

(3) Dα C = 0, for all constant functions, f (t) = C.
(4) Dα f g) = f Dα (g) + gDα (f ).
(5) Dα fg ) = gD (f )−f D α (g)
α

g2 .
(6) Dα (f ◦ g)(t) = f ′ g(t) Dα g(t), for f differentiable at g(t).

(7) If, in addition, f is differentiable, then Dα f )(t) = t1−α df

dt (t).

Proof. Part (1) and (3) follow directly from the definition. We shall prove (2), (4), (6) and (7), since the
proofs are different from what appear in [3]. Now, for fixed α ∈ (0, 1], n ∈ R and t > 0, we have

(teǫt )n − tn
−α

Dα tn = lim


ǫ→0 ǫ
nǫt−α
e −1
= tn lim
ǫ→0 ǫ
2 2 −2α

n ǫnt−α + ǫ n 2!t + ···

= t lim
ǫ→0 ǫ
ǫnt−α + O(ǫ2 )
(2.2) = tn lim
ǫ→0 ǫ
= ntn−α .

This completes the proof of (2). Then, we shall prove (4). To this end, since f, g are α−differentiable at
t > 0, note that,

f (teǫt )g(teǫt ) − f (t)g(t)

−α −α

Dα f g (t) = lim


ǫ→0 ǫ
f (teǫt )g(teǫt ) − f (t)g(teǫt ) + f (t)g(teǫt ) − f (t)g(t)
−α −α −α −α

= lim
ǫ→0 ǫ
h f (teǫt−α ) − f (t) i g(teǫt ) − g(t)
−α
ǫt−α
= lim · g(te ) + f (t) lim
ǫ→0 ǫ ǫ→0 ǫ
α ǫt −α
α



= D f (t) lim g(te ) + f (t)D g (t).
ǫ→0

Since g is continuous at t, limǫ→0 g(teǫt ) = g(t). This completes the proof of (4). The proof of (5) is
−α

similar and is left for the reader to verify. Next, we prove (6) in two different approaches. First, suppose
u = g(t) is α−differentiable at a point a > 0 and y = f (u) is differentiable at a point b = g(a) > 0. Let
ǫ > 0, and ∆y = f (beǫb ) − f (b). Since beǫb = b + ǫb1−α + O(ǫ2 ) = b + ∆u, where ∆u = ǫb1−α + O(ǫ2 ),
−α −α

we have

(2.3) ∆y = D1 f (b)∆u + ǫ1 ∆u

where ǫ1 → 0 as ∆u → 0. Thus, ǫ1 is a continuous function of ∆u if we define ǫ1 to be 0 when ∆u = 0.

Now, if ∆t is an increment in t and ∆u and ∆y (with the possibility of both being equal to 0) are
corresponding increments in u and y, respectively. Then we may write, using Equation (2.3),

(2.4) ∆u = Dα g(a)∆t + ǫ2 ∆t
4 U.N. KATUGAMPOLA

where ǫ2 → 0 as ∆t → 0, and
h i
∆y = D1 f (b) + ǫ1 ∆u
h i h i
= D1 f (b) + ǫ1 · Dα g(a) + ǫ2 ∆t

where both ǫ1 → 0 and ǫ2 → 0 as ∆t → 0. Taking ∆t = ǫ, we now have,

∆y
Dα (f ◦ g)(t) = lim
ǫ→0 ǫh i h i
= lim D1 f (b) + ǫ1 · Dα g(a) + ǫ2
∆t→0
= D1 f (b)Dα g(a) = f ′ (g(a))Dα g(a).
It can be seen that Equation 2.4, can not be written in the form of Equation 2.3. This makes it necessitates
to use the α−derivative in Equation 2.4.
Now, we prove the result following a standard limit-approach.
To this end, in one hand, if the function
g is constant in a neighborhood containing a, then Dα f ◦ g (a) = 0. On the other hand, assume that
the function g is non-constant in the neighborhood of a. In this case, we can find an ǫ0 > 0 such that,
g(x1 ) 6= g(x2 ) for any x1 , x2 ∈ (a − ǫ0 , a + ǫ0 ). Now, since g is continuous at a, for ǫ sufficiently small, we
have
f g(aeǫa ) − f g(t)
−α


α


D f ◦ g (a) = lim
ǫ→0 ǫ
ǫa−α



f g(ae ) − f g(a) g(aeǫa ) − g(a)
−α

= lim ·
ǫ→0 g(aeǫa−α ) − g(a) ǫ



f g(a) + ǫ1 − f g(a) g(aeǫa ) − g(a)
−α

= lim · , where ǫ1 → 0 as ǫ → 0,
ǫ→0 ǫ1 ǫ



f g(a) + ǫ1 − f g(a) g(aeǫa ) − g(a)
−α

= lim · lim
ǫ1 →0 ǫ1 ǫ→0 ǫ
′ α


= f g(a) D g(a),
for a > 0. This establishes the Chain Rule.
To
prove part (7), we use a similar line of argument as in Equation (2.2). Thus, taking h = ǫt1−α 1 +
O(ǫ) , we have

f (teǫt
−α
) − f (t)
Dα (f )(t) = lim
ǫ→0 ǫ
f (t + ǫt1−α + O(ǫ2 )) − f (t)
= lim
ǫ→0 ǫ
f (t + h) − f (t)
= lim htα−1
ǫ→0
1+O(ǫ)
df
= t1−α (t),
dt
since, by the assumption, f is differentiable at t > 0. This completes the proof of the theorem. 

Remark 2.4. The result similar to (6) does not appear in [3]. The argument used in Equation (2.2)
was/will be utilized in several occasions in the paper.
 A NEW FRACTIONAL DERIVATIVE WITH CLASSICAL PROPERTIES 5

The following theorem lists α−fractional derivative of several familiar functions. The results are
identical to that of conformable fractional derivative discussed in [3], which will be shown to be a special
case of α−fractional derivative defined in (2.1).
Theorem 2.5. Let a, n ∈ R and α ∈ (0, 1]. Then, we have the following results.
(a) Dα (tn ) = ntn−α .
(b) Dα (1) = 0.
(c) Dα (eax ) = ax1−α eax .
(d) Dα (sin ax) = ax1−α cos ax.
(e) Dα (cos ax) = −ax1−α sin ax.
(f) Dα ( α1 tα ) = 1.
It is easy to see from part (6) of Theorem 2.3 that we have rather unusual results given in the next
theorem.
Theorem 2.6. Let α ∈ (0, 1] and t > 0. Then,
(i) Dα (sin α1 tα ) = cos α1 tα .
(ii) Dα (cos α1 tα ) = − sin α1 tα .
1 α 1 α
(iii) Dα (e α t ) = e α t .
Theorem 2.6 suggests that there is a pseudo-invariant space corresponding to the α−fractional de-
rivative on which the sine, cosine and the exponential function, ex behave as they are having familiar
classical derivatives. Due to the Fourier theory, this in turn, implies that any function which possesses a
Fourier expansion behaves as having classical derivatives. This would be an interesting topic for further
research.

(a) ν= 2.0 (b) ν= 1.0

1
Figure 1. 2 −fractional derivatives of power function f (x) = xν .

In [1, 2], the author introduced fractional integrals and derivatives, which generalize the Riemann-
Liouville and the Hadamard fractional integrals and derivatives to a single form. In the next subsection,
we show that the results obtained by Khalil et at. [3] are special cases of the more general extension
discussed in this paper.
6 U.N. KATUGAMPOLA

2.1. Further Generalizations. Let us first define a truncated exponential function given by,
k
X xi
exk =
i=0
i!
With the help of this definition, we will define another fractional derivative given below.
Definition 2.7. Let f : [0, ∞) → R and t > 0. Then the fractional derivative of f of order α is defined
by,
f (tekǫt
−α
) − f (t)
(2.5) Dkα (f )(t) = lim ,
ǫ→0 ǫ
for t > 0, α ∈ (0, 1). If f is α−differentiable in some (0, a), a > 0, and limt→0+ Dkα (f )(t) exists, then
define
Dkα (f )(0) = lim Dkα (f )(t).
ǫ→0

Now, It is easy to see that

f (t + ǫt1−α ) − f (t)
(2.6) D1α (f )(t) = lim ,
ǫ→0 ǫ
and
f (teǫt
−α
α ) − f (t)
(2.7) D∞ (f )(t) = lim ,
ǫ→0 ǫ
where the expression on the right-hand-side of (2.6) is the conformable fractional derivative defined in
(1.1) in [3], while the one on (2.7) is the α−fractional derivative defined in this paper. Furthermore, if
α = 1 the Equation (2.6) becomes the classical definition of the first derivative of a function f at a point
t. While, noting that teǫt = t + ǫ + O(ǫ2 ), we may also show that the Equation (2.7) is equivalent
−α

to the classical definition of the first derivative of a function f . These observations further suggest that
there are corresponding results similar to the Rolle’s theorem and the Mean Value theorem. Next, we
give those two results.
Theorem 2.8 (Rolle’s theorem for α−Fractional Differentiable Functions). Let a > 0 and f : [a, b] → R
be a function with the properties that
(1) f is continuous on [a, b],
(2) f is α−differentiable on (a, b) for some α ∈ (0, 1),
(3) f (a) = f (b).
Then, there exists c ∈ (a, b), such that Dα (f )(c) = 0.
Proof. We prove this using contradiction. Since f is continuous on [a, b] and f (a) = f (b), there is
c ∈ (a, b), at which the function has a local extrema. Then,
f (ceǫc f (ceǫc ) − f (c)
−α −α
α ) − f (c)
D f (c) = lim = lim .
ǫ→0− ǫ ǫ→0+ ǫ
But, the two limits have opposite signs. Hence, Dα f (c) = 0. 
Theorem 2.9 (Mean Value Theorem for α−Fractional Differentiable Functions). Let a > 0 and f :
[a, b] → R be a function with the properties that
(1) f is continuous on [a, b],
(2) f is α−differentiable on (a, b) for some α ∈ (0, 1),
f (b)−f (a)
Then, there exists c ∈ (a, b), such that Dα (f )(c) = 1 α 1 α.
αb −αa
 A NEW FRACTIONAL DERIVATIVE WITH CLASSICAL PROPERTIES 7

Proof. Consider the function,

f (b) − f (a)  1 α 1 α 
g(x) = f (x) − f (a) − 1 α 1 α αx − αa .
αb − αa
Then, the function g satisfies the conditions of the fractional Rolle’s theorem. Hence, there exists c ∈ (a, b),
such that Dα (g)(c) = 0. Using the fact that Dα ( α1 xα ) = 1, the result follows. 
As a result of the fractional Mean Value Theorem, we also have the following proposition.
Proposition 2.10. Let f : [a, b] → R be α−differentiable for some α ∈ (0, 1). Suppose that Dα f is
bounded on [a, b]. Then, if
(a) a > 0, or
(b) Dα f is continuous at either a or b,
the function f is uniformly continuous on [a, b], and hence f is bounded.
ν
Figure 1 depicts the graphs of the α−fractional derivatives of√ the power function f (x) = x for ν =11
and 2. It is also interesting to note that the function f (x) = x has a constant derivative equal to 2 ,
√
and not equal to π/2 as in the case of the Riemann-Liouville derivative. This also suggests that the
new derivative should have a rather different interpretation in the geometrical sense, which has yet to be
discovered.
Further, it can be seen that a function could be α−differential at a point but not differentiable in the
1
classical sense. (cf. page 67 of [3]). For example, the function f (t) = 3t 3 is everywhere 13 −differentiable
and has the derivative of 1 while it is not differentiable at t = 0. The similar results are there for the
Riemann-Liouville and the Caputo derivatives.
Next, we consider the possibility of α ∈ (n, n + 1], for some n ∈ N. We have the following definition.
Definition 2.11. Let α ∈ (n, n + 1], for some n ∈ N and f be an n−differentiable at t > 0. Then the
α−fractional derivative of f is defined by
f (n) (teǫt ) − f (n) (t)
n−α

Dα f (t) = lim ,
ǫ→0 ǫ
if the limit exists.
As a direct consequence of Definition 2.11 and part (7) of theorem 2.3, we can show that Dα f (t) =
n+1−α (n+1)
t f (t), where α ∈ (n, n + 1] and f is (n + 1)−differentiable at t > 0.
Now, the question whether the fractional derivative Dα f has a corresponding α−fractional integral
will be answered next. For simplicity, we shall only consider the class of continuous functions here. The
results can easily be extended to more general settings and will be discussed in forthcoming papers.

3. Fractional Integral
As in the works of [3], it is interesting to note that, in spite of the variation of the definitions of the
fractional derivatives, we can still adopt the same definition of the fractional integral here due to the fact
that we obtained similar results in Theorem 2.5 as of the results (1)–(6) and (i)–(iii) in [3]. So, we have
the following definition.
Definition 3.1 (Fractional Integral). Let a ≥ 0 and t ≥ a. Also, let f be a function defined on (a, t] and
α ∈ R. Then, the α−fractional integral of f is defined by,
Z t
f (x)
(3.1) Iaα (f )(t) = 1−α
dx
a x
if the Riemann improper integral exists.
8 U.N. KATUGAMPOLA

It is interesting to observe that the α−fractional derivative and the α−fractional integral are inverse
of each other as given in the next result.
Theorem 3.2 (Inverse property). Let a ≥ 0, and α ∈ (0, 1). Also, let f be a continuous function such
that Iaα f exists. Then  
Dα Iaα (f ) (t) = f (t), for t ≥ a.

Proof. The proof is a direct consequence of the fundamental theorem of calculus. Since f is continuous,
Iaα f is clearly differentiable. Therefore, using part (7) of theorem 2.3, we have
  d
Dα Iaα (f ) (t) = tα−1 Iaα (f )(t),
dt
d t f (x)
Z
= tα−1 dx,
dt a x1−α
f (t)
= tα−1 1−α ,
t
= f (t).
This completes the proof. 
The applications of this α−integral and derivative will be discussed in a forthcoming paper. Several
interesting applications of a similar fractional derivative defined in (1.1) and the corresponding fractional
integral appear in [3], the major reference of this paper.
We conclude the paper with the following questions, which have yet to be answered.
(i) What is the physical meaning or geometric interpretation of the α−derivative?
(ii) Is there a similarity between the α−derivative and the Gâteaux derivative?
(iii) One of the limitations of this version of the fractional derivative is that it assumes that the variable
t > 0. So the question is whether we can relax this condition on a special class of functions, if so,
what is it?

References
[1] U.N. Katugampola, New approach to a generalized fractional integral, Appl. Math. Comput., 218(3)(2011) 860–865.
[2] U.N. Katugampola, New approach to generalized fractional derivatives, B. Math. Anal. App., 6(4)(2014) 1–15.
arXiv:1106.0965.
[3] R. Khalil, M. A. Horani, A. Yousef, M. Sababheh. A new definition of fractional derivative, J. Comput. Appl. Math.
264(2014) 65–70.
[4] Kilbas, A. A., Srivastava, H.M., and Trujillo, J.J., Theory and Applications of Fractional Differential Equations,
Elsevier B.V., Amsterdam, Netherlands, 2006.
[5] G. W. Leibniz, “Letter from Hanover, Germany to G.F.A. L’Hospital, September 30, 1695”, Leibniz Mathematische
Schriften, Olms-Verlag, Hildesheim, Germany, 1962, p.301-302, First published in 1849.
[6] G. W. Leibniz, “Letter from Hanover, Germany to Johann Bernoulli, December 28, 1695”, Leibniz Mathematische
Schriften, Olms-Verlag, Hildesheim, Germany, 1962, p.226, First published in 1849.
[7] G. W. Leibniz, “Letter from Hanover, Germany to John Wallis, May 28, 1697”, Leibniz Mathematische Schriften,
Olms-Verlag, Hildesheim, Germany, 1962, p.25, First published in 1849.
[8] Machado, J. A., And I say to myself: “What a fractional world!”, Frac. Calc. Appl. Anal. 14(4)(2011) 635-654.
[9] Oldham, K. B. and Spanier, J., The fractional calculus, Academic Press, New York, 1974.
[10] Podlubny, I., Fractional Differential Equations, Academic Press, San Diego, California, U.S.A., 1999.
[11] Samko, S.G., Kilbas, A.A., and Marichev, O.I., Fractional Integrals and Derivatives: Theory and Applications, Gordon
and Breach, Yverdon et alibi, 1993.

Department of Mathematics, University of Delaware, Newark DE 19716, USA.

E-mail address: [email protected]