Fractional Derivatives and Fractional Mechanics

Danny Vance
June 2, 2014

Abstract
This paper provides a basic introduction to fractional calculus, a branch
of mathematical analysis that studies the possibility of taking any real
power of the differentiation operator. We introduce two different def-
initions of the fractional derivative, namely the Riemann-Liouville and
Caputo forms, and examine some basic properties of each. Later, we dis-
cuss fractional mechanics, where the time derivative is replaced with a
fractional derivative of order α. We then solve some simple fractional
differential equations leading up to the fractional damped harmonic oscil-
lator problem.

Contents

1 Basic Fractional Operators 1

1.1 The Fractional Integral . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Fractional Derivatives . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Examples and properties . . . . . . . . . . . . . . . . . . . . . . . 3

2 Fractional Mechanics 6
2.1 Basic Fractional Differential Equations . . . . . . . . . . . . . . . 7
2.2 The Fractional Damped Oscillator . . . . . . . . . . . . . . . . . 10

1. Basic Fractional Operators

1.1. The Fractional Integral
The first fractional operator we will introduce is the fractional integral, which is
a generalization of the n-tuple iterated integral to any real order. We start by
expressing any nth iterated integral as a single integral, using Cauchy’s formula
for repeated integration.
Theorem 1.1 (Cauchy formula for repeated integration). Let f be some con-
tinuous function on the interval [a, b]. The nth repeated integral of f based at
a,

1
 Z x Z σ1 Z σn−1
f (−n) (x) = ··· f (σn ) dσn dσn−1 · · · dσ2 dσ1 ,
a a a
is given by single integration:
Z x
1
f (−n) (x) = (x − t)n−1 f (t) dt
(n − 1)! a
.

Proof. A proof is given by induction. Consider the base case n = 1. We have

that (x − t)n−1 = (x − t)0 = 1, so

Z x Z x Z x
1 1
f (σ1 ) dσ1 = (x − t)0 f (t) dt = (x − t)n−1 f (t) dt
a (0!) a (n − 1)! a

and the statement holds for n = 1. Now suppose the statement holds for some
arbitrary n. We will prove it for n + 1 by switching the order of integration:
Z x Z σ1 Z σn
(−(n+1))
f (x) = ··· f (σn+1 ) dσn+1 dσn · · · dσ2 dσ1
a a a
Z x Z σ1
1
= (σ1 − t)n−1 f (t) dt dσ1
(n − 1)! a a
Z xZ x
1
= (σ1 − t)n−1 f (t) dσ1 dt
(n − 1)! a t
Z x
1
= ((x − t)n − (t − t)n )f (t) dt
(n)! a
Z x
1
= (x − t)n f (t) dt
(n)! a
The result follows by induction. [6]

This theorem gives us the framework we need to take an integral of any

real degree. First note that Γ(n) = (n − 1)! for positive integers n, where Γ is
the gamma function. We can define the Riemann-Liouville fractional integral
by replacing the power n in the integrand with some α ∈ R+ , and replacing
(n − 1)! with Γ(α).

Definition 1 (Riemann-Liouville Operator). Let f be a continuous function,

α ∈ R+ , and t ∈ R. The fractional integral of order α is defined as:
Z t
α 1
J f (t) = (t − u)α−1 f (u) du [1]
Γ(α) 0

2
1.2. Fractional Derivatives
Now that we’ve defined the fractional integral, we can easily define fractional
differentiation of any positive real power by combining the standard differential
operator with a fractional integral of order between 0 and 1. We just have to
choose which operator to apply first. For example, we can define the derivative
of order 1.5 of a function f (t) as either of the following:

D1.5 f (t) = D2 J .5 f (t)

D1.5 f (t) = J .5 D2 f (t)
These two approaches provide the basis for two different definitions of the
fractional derivative. The first definition, in which the fractional integral is
applied bef ore differentiating, is called the Riemann-Liouville fractional deriva-
tive. The second, in which the fractional integral is applied af terwards, is
called the Caputo derivative. These two forms of the fractional derivative each
behave a bit differently, as we will see. Here are their formal definitions:
Definition 2. Pick some α ∈ R+ , let n be the nearest integer greater than
α. The Riemann-Liouville fractional derivative of order α of a function f(t) is
given by:
Z t
α dn n−α 1 dn
D f (t) = n J f (t) = (t − u)n−α−1 f (u) du
dt Γ(n − α) dtn 0

Definition 3. Pick some α ∈ R+ , let n be the nearest integer greater than α.

The Caputo fractional derivative of order α of a function f(t) is given by:
Z t
dn 1
D∗α f (t) = J n−α n f (t) = (t − u)n−α−1 f (n) (u) du
dt Γ(n − α) 0
These definitions, taken from a paper by Boyadjiev, Ishteva, and Scherer, [1],
can be extended to all real powers. For α < 0, we use the definition Dα f (t) =
D∗α f (t) = J −α f (t). For α = 0, we set D0 f (t) = D∗0 f (t) = f (t). This extension
unifies integration and differentiation as one operator, sometimes referred to
as the differintegral [6]. Since there are many legitimate ways to define the
fractional derivative, of which the Caputo and Riemann-Liouville forms are
only two, there are many different legitimate ways to define the differintegral.

1.3. Examples and properties

Lemma 1.2. Let α > 0, C, K ∈ R, and let f and g be functions such that their
fractional derivatives and integrals exist. Then

J α (Cf (t) + Kg(t)) = CJ α f (t) + KJ α g(t)

Dα (Cf (t) + Kg(t)) = CDα f (t) + KDα g(t)

D∗α (Cf (t) + Kg(t)) = CD∗α f (t) + KD∗α g(t)

3
Proof. The linearity of these operators follows from the linearity of the integer-
order derivatives and integrals by which they are defined.
Z t
tα
Since (t − u)α−1 du = , the fractional integral of order α of 1 is given
0 α
by:
Z t
1 tα tα
J α1 = (t − u)α−1 du = =
Γ(α) 0 αΓ(α) Γ(α + 1)
The nth fractional integral of order α of 1 is then given by:
tnα
J nα 1 =
Γ(nα + 1)

Theorem 1.3. The Riemann-Liouville derivative of order α > 0 with n − 1 <

α < n of the power function f (t) = tp for p ≥ 0 is given by:

Γ(p + 1) p−α
D α tp = t
Γ(p − α + 1)

Proof. We compute the Riemann-Liouville Derivative of power α > 0 as:

Z t
1 dn
Dα tp = (t − u)n−α−1 up du
Γ(n − α) dtn 0

Set u = vt for 0 ≤ v ≤ 1, du = tdv and we get:

Z 1
1 dn
D α tp = ((1 − v)t)n−α−1 (vt)p tdv
Γ(n − α) dtn 0
1
dn n+p−α
Z
1
= (1 − v)n−α−1 v p dv t
Γ(n − α) 0 dtn
1 dn
= B(p + 1, n − α) n tn+p−α
Γ(n − α) dt
1 Γ(n + p − α + 1) p−α
= B(p + 1, n − α) t
Γ(n − α) Γ(p − α + 1)
1 Γ(p + 1)Γ(n − α) Γ(n + p − α + 1) p−α
= t
Γ(n − α) Γ(n + p − α + 1) Γ(p − α + 1)
Γ(p + 1) p−α
= t
Γ(p − α + 1)

4
 Z 1
Note: B is the Beta function B(x, y) = v x−1 (1−v)y−1 dv for x > 0, y > 0
0
Γ(x)Γ(y)
and we used the fact that B(x, y) = . [1]
Γ(x + y)
This formula allows us to quickly find the fractional derivative of any poly-
nomial, by simply taking fractional derivatives of each term separately. Figure
1 shows several graphs of the Riemann-Liouville fractional derivatives of various
orders of the function f(x) = x.
We would hope that the fractional derivative of a constant function is always
zero, but this is simply not always the case. If we use our formula for Dα tp with
t−α kt−α
p = 0, we get Dα 1 = , so Dα k = , which only evaluates to 0
Γ(1 − α) Γ(1 − α)
if k = 0.

Figure 1: Riemann-Liouville Derivatives of a linear function [3]

Taking the Caputo Derivative yields different results:

Theorem 1.4. The Caputo derivative of order α > 0 with n − 1 < α < n of
the power function f (t) = tp for p ≥ 0 satisfies:

 Γ(p + 1) tp−α (p > n − 1)



Dα tp = Γ(p − α + 1)
0 (p ≤ n − 1)


t
dn p
Z
1
Proof. In the second case, p ≤ n−1, so n
t = 0 and D∗α tp = 0=
dt Γ(n − α) 0

5
 dn p Γ(p + 1) p−n
0. In the first case, we have t = t , so our Caputo Derivative
dtn Γ(p − n + 1)
is given by:
Z t
1 Γ(p + 1)
D∗α tp = (t − u)n−α−1 up−n du
Γ(n − α) 0 Γ(p − n + 1)
Z 1
Γ(p + 1)
= ((1 − v)t)n−α−1 (vt)p−n t dv
Γ(n − α)Γ(p − n + 1) 0
Z 1
Γ(p + 1)
= tp−α (1 − v)n−α−1 v p−n dv
Γ(n − α)Γ(p − n + 1) 0

Γ(p + 1)B(p − n + 1, n − α) p−α Γ(p + 1)Γ(p − n + 1)Γ(n − α) p−α

= t = t
Γ(n − α)Γ(p − n + 1) Γ(n − α)Γ(p − n + 1)Γ(p − α + 1)
Γ(p + 1) p−α
= t
Γ(p − α + 1)

So Dα tp and D∗α tp coincide when p > n−1, but in all other cases D∗α tp evaluates
to 0 and Dα tp does not. This formula also gives us that D∗α k = kD∗α t0 = 0
∀α > 0 for all constants k. So the positive order Caputo fractional derivative
of a constant is always zero. This is one way in which Caputo derivatives are
considered to be more well-behaved than Riemann-Liouville derivatives [1].

2. Fractional Mechanics
For the sake of simplicity and consistency, from here on out we will use the
Caputo derivative of order α with 0 < α < 1. With fractional calculus, we
are now able to construct classical mechanics with fractional derivatives. Given
position x(t), we introduce the fractional velocity v(t) and fractional acceleration
a(t) as follows:

v(t) = D∗α x(t), a(t) = D∗α v(t)

We will also make reference to the Mittag-Leffler function, Eα,β , which is a very
special function which finds widespread use in the world of fractional calculus.
Just as the exponential function naturally arises in solutions of integer order
differential equations, the Mittag-Leffler function plays an analagous role in the
solution of non-integer order differential equations. The Mittag-Leffler function
in its most general form depends on two parameters α > 0 and β. It is defined
as:
∞
X zn
Eα,β (z) =
n=0
Γ(αn + β)

6
Alternatively, if the β parameter is not needed, the Mittag-Leffler function can
be defined as:
∞
X zn
Eα (z) = [5]
n=0
Γ(αn + 1)
The exponential function is equal to the Mittag-Leffler function for α = 1.
Figure 2 shows graphs of the Mittag-Leffler function for various parameters.

Figure 2: The Mittag-Leffler function, defined with several different parameters.

[7]

2.1. Basic Fractional Differential Equations

In fractional mechanics, Newton’s second law of motion becomes F = ma =
mD∗α v, where m is the mass of the body in motion. When the force is constant,
F
the body moves with a constant fractional acceleration of m . Now consider
the vertical motion of a body in a resisting medium in which there exists a
resisting force proportional to the fractional velocity, as is sometimes the case
with viscoelastic drag in certain types of materials like polymers [4]. We assume
the body is projected downward with zero initial velocity (v(0) = 0) in a uniform
gravitational field. For some constant k denoting the ratio of resistance to
fractional velocity, the equation of motion is given by:

F = mD∗α v = mg − kv
Applying a fractional integral of degree α and dividing both sides by m we get:
k α
v(t) = gJ α [1] − J [v(t)]
m
k n nα
Multiply both sides by (− m ) J and sum both sides for n from 0 to ∞ to get:

7
 ∞ ∞ ∞
X k n nα X k X k
(− ) J v(t) = g (− )n J (n+1)α [1] + (− )n+1 J (n+1)α v(t)
n=0
m n=0
m n=0
m

∞ ∞ ∞
X k n nα X k X k
⇐⇒ (− ) J v(t) − (− )n+1 J (n+1)α v(t) = g (− )n J (n+1)α [1]
n=0
m n=0
m n=0
m

∞ ∞ ∞
X k n nα X k X k
⇐⇒ (− ) J v(t) − (− )n J (n+1)α v(t) = g (− )n J (n+1)α [1]
n=0
m n=1
m n=0
m

The two sums on the left cancel each other out except for the n = 0 case, giving
us:
∞
X k n (n+1)α
v(t) = g (− ) J [1]
n=0
m
nα
t
Recall that J nα [1] = . Replacing n with n + 1 and plugging this into
Γ(nα + 1)
our formula we get:

∞ ∞
X k n t(n+1)α mg X k tnα
v(t) = g (− ) = (−1)n+1 ( )n
n=0
m Γ((n + 1)α + 1) k n=1 m Γ(nα + 1)

∞ ∞
" #
k α n k α n
mg X (−1)n+1 ( m t ) mg X (− m t )
= = 1−
k n=1 Γ(nα + 1) k n=0
Γ(nα + 1)
 
mg k
= 1 − Eα,1 (− tα )
k m
and so we’ve found a solution for v(t). This problem was solved in a paper by
Min Jung and Won Sang Chung [2].

Now considerthe fractional harmonic oscillator problem:

mD∗2α x(t) = −mw2 x(t)

with the initial conditions:

x(0) = A, (D∗α x)(0) = v0

Applying a fractional integral of order α to the equation and dividiing both
sides by m, we get:

D∗α x(t) − v0 = −w2 J α x(t)

8
Move v0 to the other side and integrate once more and we get:

x(t) − A = v0 J α 1 − w2 J 2α x(t)
Now multiply both sides of the equation by (−w2 )m J 2mα to get:

(−w2 )m J 2mα x(t) − (−w2 )(−w2 )m J (m+2)α x(t) =

(−w2 )m v0 J (2m+1)α [1] + (−w2 )m AJ 2mα [1]
Now summing both sides for m from 0 to ∞ yields:
∞
X ∞
X
(−w2 )m J 2mα x(t) − (−w2 )m+1 J 2(m+1)α x(t) =
m=0 m=0
∞
X ∞
X
v0 (−w2 )m J (2m+1)α [1] + A (−w2 )m J 2mα [1]
m=0 m=0
∞
X ∞
X
⇐⇒ x(t) + (−w2 )m J 2mα x(t) − (−w2 )m J 2mα x(t) =
m=1 m=1
∞
X ∞
X
v0 (−w2 )m J (2m+1)α [1] + A (−w2 )m J 2mα [1]
m=0 m=0

∞
X ∞
X
⇐⇒ x(t) = v0 (−w2 )m J (2m+1)α [1] + A (−w2 )m J 2mα [1]
m=0 m=0

∞ ∞
X t(2m+1)α X t2mα
= v0 (−w2 )m +A (−w2 )m
m=0
Γ((2m + 1)α + 1) m=0
Γ(2mα + 1)
∞ ∞
X (−1)m (wtα )2m v0 X (−1)m (wtα )2m+1
=A +
m=0
Γ(2mα + 1) w m=0 Γ((2m + 1)α + 1)
v0
= ACα,1 (wtα ) + Sα,1 (wtα )
w
where Cα,1 and Sα,1 are the Mittag-Leffler cosine and sine functions:

∞ ∞
X (−1)m (x)2m X (−1)m (x)2m+1
Cα,1 (x) = , Sα,1 (x) =
m=0
Γ(2mα + 1) m=0
Γ((2m + 1)α + 1)

Mittag-Leffler cosine and sine functions can also be written as:

1 1
Cα,1 (x) = [Eα,1 (ix) + Eα,1 (−ix)] , Sα,1 (x) = [Eα,1 (ix) − Eα,1 (−ix)]
2 2i
These formulas are analagous to the formulas for cosine and sine in terms of eix
and e−ix . [2]

9
2.2. The Fractional Damped Oscillator
The damped harmonic oscillator is the simplest model in classical mechanics
of motion with dissipation which corresponds to frictional force proportional to
velocity. We will consider the fractional damped oscillator defined by:

mD∗α x(t) = −mw2 x(t) − 2mγD∗α x(t)

with the following initial condition:

x(0) = A, (D∗α x)(0) = v0

Now we will solve this fractional differential equation, once again using a
series approach. Let us assume the solution is given by:
∞
X
x(t) = cn tnα
n=0

Γ(p + 1) p−α
Recall that D∗α tp = t when D∗α tp 6= 0. Inserting our dif-
Γ(p − α + 1)
ferential equation into our series representation we get the following recurrence
relation for each integer n ≥ 0:

Γ((n + 2)α + 1) Γ((n + 1)α + 1) Γ(nα + 1)

cn+2 + 2γcn+1 + w 2 cn =0
Γ(nα + 1) Γ(nα + 1) Γ(nα + 1)

⇐⇒ cn+2 Γ((n + 2)α + 1) + 2γcn+1 Γ((n + 1)α + 1) + w2 cn Γ(nα + 1) = 0

Let an = Γ(nα + 1) and we have:

an+2 + 2γan+1 + w2 an = 0
We can represent this recurrence relation with the quadratic equation:

t2 + 2γt + w2 = 0
Let p, q, be solutions of the quadratic equation and we have:

0 = (t − p)(t − q) = t2 − pt − qt + pq
⇐⇒ t2 − pt = q(t − p)
which allows us to rewrite our recurrence relation as:

an+2 − pan+1 = q(an+1 − pan )

Now there are three possibilities:

10
 Case 1: Over-damped oscillator: In this case γ 2 > w2 and by the
quadratic formula we have:
p p
p = −γ + γ 2 − w2 , p = −γ − γ 2 − w2

Case 2: Critically damped oscillator: In this case γ 2 = w2 and we

have:
p = q = −γ
Case 3: Under-damped oscillator: In this case γ 2 < w2 and we have:
p p
p = −γ + i γ 2 − w2 , p = −γ − i γ 2 − w2

Iterating through the recurrence relation yields:

n−1
n n−1
X q
an − p a0 = (a1 − pa0 )p ( )k
p
k=0

which gives the following solution:

(
A1 pn − B1 q n (p 6= q)
an =
pn a0 + (a1 − pa0 )npn−1 (p = q)
with
a1 − pa0 a1 − pa0
A1 = a0 + , B1 =
p−q p−q
So the position has the following form:
(
v0 −qA α v0 −pA α
p−q Eα,1 (pt ) − p−q Eα,1 (qt ) (p 6= q)
x(t) =
AEα,1 (ptα ) + v0 −pA
α tα Eα,α (ptα ) (p = q)

According to Won Sang Chung and Min Jung [2], fractional differential equa-
tions expressed in terms of Caputo derivatives require the same boundary condi-
tions as their integer-order counterparts, due to the previously-established fact
that the Caputo derivative of a constant is always zero. This makes all of the
solutions explored in this paper unique.

References
[1] Boyadjiev, L.. Ishteva, M.. Scherer, R.. On The Caputo Operator of Frac-
tional Calculus And C-Laguerre Functions.
[2] Chung, Won Sang. Jung, Min. (2013, Feb 18). On the fractional damped
oscillators and fractional forced oscillators.
[3] Hale, Nick. (Oct 2010). Fractional Calculus in Chebfun.
http://www2.maths.ox.ac.uk/chebfun/examples/integro/html/FracCalc.shtml

11
[4] Kempfle, Nolte, and Schfer (2003). Does a Real Material Behave Fraction-
ally? Applications of Fractional Differential Operators to the Damped Struc-
ture Borne Sound in Viscoelastic Solids, Journal of Computational Acoustics
(JCA), Volume 11, Issue 3.
[5] Loverro, Adam. (8 May 2004). Fractional Calculus: History, Denitions and
Applications for the Engineer.
[6] McGonegal, Ben. (13 Jan 2014). Fractional Brownian Motion and the Frac-
tional Stochastic Calculus.
[7] Weisstein, Eric W. ”Mittag-Leffler Function.” From MathWorld–A Wolfram
Web Resource. http://mathworld.wolfram.com/Mittag-LefflerFunction.html

12