XXXVII SUMMER SCHOOL ON MATHEMATICAL PHYSICS

RAVELLO (ITALY), SEPTEMBER 17 -29, 2012

Gruppo Nazionale per la Fisica Matematica (GNFM-INdAM)

Exercises on Fractional Calculus in IR+

Francesco MAINARDI
Department of Physics, University of Bologna, GNFM and INFN
Via Irnerio 46, I–40126 Bologna, Italy.
[email protected] [email protected]
http://www.fracalmo.org/mainardi
http://scholar.google.com/scholar?hl=en&lr=&q=f+mainardi
Ex.1 Prove that
Z ∞ Z 1
−u n
n! = e u du = (−ln t)n dt .
0 0

Ex.2 Show that ∞

1
(−1)n
Z X
−u z−1
e u du =
0 n=0
(n + z)n!
for z 6= −n with n = 0, 1, 2, . . .
Hint: Use the series representation of the exponential function. This result
leads to the analytical continuation of the Gamma function to the left half
plane Re(z) ≤ 0 excluding the points z = 0, −1, −2, . . . These points turn
out to be simple poles with residue Rn = (−1)n /n! with n = 0, 1, 2, . . ..
Ex.3 Using integration by parts, show that
Γ(z + 1) = zΓ(z) , Re(z) > 0 .
where the Gamma function is defined as
Z ∞
Γ(z) := e−u uz−1 du , Re(z) > 0 .
0
This means that the Gamma function satisfies a finite difference equation as
above. Furthermore, it is sufficient to know the Gamma function in the strip
0 < Re(z) ≤ 1 in order to compute it in all the half-plane Re(z) > 0 by
iteration ahead:
Γ(z + n) = z(z + 1) · · · (z + n − 1) Γ(z) , n ∈ IN .
Iterating behind, Γ(z) = Γ(z + 1)/z, we can get the analytical continuation
of the Gamma function in the left half plane (except in the simple poles
z = 0, −1, −2, . . .).

1
Ex.4 Introduce the substitution st = u in the integral (Laplace transform)
below to show the result
Z ∞
Γ(α + 1)
α
L[t ; s] := e−st tα dt =
0 sα+1
with Re(α) > −1 and Re(s) > 0.
Ex.5 Considering Re(ν) > 0, Re(µ) > 0 and Re(z) > 0 show the result
Z ∞
µ 1 Γ(ν/µ) 1 Γ(1 + ν/µ)
e−zt tν−1 dt = ν/µ
= .
0 µ z ν z ν/µ

Ex.6 Verify the following inequalities

x α < ex < x x ,

for x >> 1 and any α > 0.

Furthermore, based on Stirling asymptotic formula
√
n! ' 2πe−n nn+1/2 , n → ∞ ,

estimate the order of magnitude of 10!

Ex.7 The convolution between two (sufficiently well-behaved) functions f (t)
and g(t) with t ∈ IR is defined by the integral
Z +∞
f (t) ? g(t) := f (τ ) g(t − τ ) dτ .
−∞

a) Show the validity of the commutative property

f (t) ? g(t) = g(t) ? f (t) ,

so that we can write

Z +∞ Z +∞
f (t) ? g(t) ≡ f (τ ) g(t − τ )dτ ≡ f (t − τ ) g(τ ) dτ .
−∞ −∞

b) If both functions are causal (namely vanishing for t < 0) show

Z t Z t
f (t) ? g(t) ≡ f (τ ) g(t − τ ) dτ ≡ f (t − τ ) g(τ )dτ .
0 0

In view of their applications in the framework of Fourier transforms (t ∈ IR)

and Laplace transforms (t ∈ IR+ ) the two convolutions are usually referred
to as Fourier convolution and Laplace convolution, respectively.

2
Ex.8 Prove the following result for the Laplace convolution between power-
law functions
Z t
p−1 q−1
t ?t := τ p−1 (t − τ )q−1 dτ = tp+q−1 B(p, q)
0

with Re(p) > 0 and Re(q) > 0. Here B(p, q) denotes the Beta function
Z 1
Γ(p) Γ(q)
B(p, q) := up−1 (1 − u)q−1 du = .
0 Γ(p + q)

For which values of p and q such convolution integral is independent on t?

Verify the result Z t
−1/2 −1/2 dτ
t ?t = √ √ = π.
0 τ t−τ
Ex.9 Using Ex.8 and defining the so-called Gel’fand-Shilov function
 λ−1
 t
t > 0,
φλ (t) := Γ(λ)
0 t < 0,


with Re(λ) > 0, show the result for the Laplace convolution

φλ (t) ? φµ (t) = φλ+µ (t).

Remark: This result will lead to the semi-group property of the Riemann-
Liouville fractional integral.
Ex.10 Using the definition of the ascending Pochhammer’s symbols

(z)n := z (z + 1) (z + 2) · · · (z + n − 1) , n ∈ IN

and the descending Pochhammer’s symbols

(z)−n := z (z − 1) (z − 2) · · · (z − n + 1) , n ∈ IN

prove for n ∈ IN that

Γ(z + n) Γ(z + 1)
(z)n = , (z)−n = .
Γ(z) Γ(z − n + 1)

Furthermore verify that

(−z)−n = (−1)n (z)n , n ∈ IN ,

3
both starting from the definition of Pochhammer’s symbols and from their
expression in terms of the Gamma function. For the last purpose, take into
account the Reflection formula
π
Γ(z) Γ(1 − z) = .
sin πz
Remark 1: Pochhammer’s symbols are commonly defined in the ascending
way, so that when we speak about Pochhammer’s symbols without attribute
we mean in the ascending way. Note that descending Pochhammer’s symbols
are used in the definition of the binomial coefficients, see Ex.11.
Remark 2: Pochhammer’s symbols are defined for all z ∈ C but care is
required when the argument of the Gamma functions is a non-positive integer
number for which the functions exhibit simple poles so that the corresponding
reciprocal functions are vanishing.
Ex.11 Express the binomial coefficients in terms of Gamma functions
Solution: We have
α α(α − 1) · · · (α − n + 1) (α)−n Γ(α + 1)
:= = = .
n n! n! Γ(α − n + 1) Γ(n + 1)
For α = m ∈ IN, we find the classical result
m m!
= , m = 1, 2, . . . n .
n (m − n)! n!

Ex.12 Recalling the power series representations of the Mittag-Leffler func-

tions in one, two and three parameters with z ∈ C, show that they are entire
functions. ∞
X zn
Eα (z) := , Re(α) > 0 ; (E1)
n=0
Γ(αn + 1)
∞
X zn
Eα,β (z) := , Re(α) > 0 , β ∈ C ; (E2)
n=0
Γ(αn + β)
∞
γ
X (γ)n
Eα,β (z) := zn , Re(α) > 0 , β ∈ C , γ > 0 ; (E3)
n=0
n!Γ(αn + β)
Hint: Prove that the radius of convergence of the corresponding power series
is infinite by using the Stirling asymptotic formula for the (ratios between)
Gamma functions.
Ex.13 Using the series representation of the Mittag-Leffler functions in Ex.12
prove
ez − 1
E1,2 (z) = ,
z
4
 E2,1 (z 2 ) = cosh(z) , E2,1 (−z 2 ) = cos(z) ,
sinh(z) sin(z)
E2,2 (z 2 ) = , E2,2 (−z 2 ) =
z z
Ex.14 Using the series representation of the Mittag-Leffler functions in Ex.12
prove
Eα,β (z) + Eα,β (−z) = 2 E2α,β (z 2 ) ,
Eα,β (z) − Eα,β (−z) = 2z E2α,α+β (z 2 ) .
Specific examples are

E1,1 (z) + E1,1 (−z) = 2 E2,1 (z 2 ) ⇐⇒ ez + e−z = 2 cosh(z) ,

E1,1 (z) − E1,1 (−z) = 2z E2,2 (z 2 ) ⇐⇒ ez − e−z = 2 sinh(z) .

Ex.15 Prove the following Laplace transform pairs for the auxiliary functions
of Mittag-Leffler type defined below

α sα−1 s−1
eα (t; λ) := Eα (−λ t ) ÷ α = , (LT − E1)
s +λ 1 + λs−α
sα−β s−β
eα,β (t; λ) := tβ−1 Eα,β (−λ tα ) ÷ = , (LT − E2)
sα + λ 1 + λs−α
sαγ−β s−β
eγα,β (t; λ) := tβ−1 Eα,β
γ
(−λtα ) ÷ α = . (LT − E3)
(s + λ)γ (1 + λs−α )γ
where we have used the sign ÷ for the juxtaposition of a function depending
on t with its Laplace transform depending on s.
Hint: The above Laplace transform pairs can be formally obtained by trans-
forming term by term the corresponding power series representations in t (use
the result in Ex.4) and summing the resulting (geometric or binomial) series
in s. This is legal because the original functions are entire of exponential
type.
Solution: For convenience let us sketch the proof for (LT-E3) that contains
(LT-E2) for δ = 1 and (LT-E1) for β = γ = 1.
Z ∞ ∞  n
γ −st β−1 γ α 1 X n Γ(γ + n) λ
eα,β (t; λ) ÷ e t Eα,β (−λt ) dt = β (−1) .
0 s n=0 Γ(γ) sα

On the other hand, using the binomial series and Ex.10, Ex.11, we have for
|z| < 1:
∞   ∞ ∞
−γ
X −γ n
X Γ(1 − γ) n
X Γ(γ + n) n
(1 + z) = z = z = (−1)n z .
n=0
n n=0
Γ(1 − γ − n)n! n=0
Γ(γ)n!

5
Comparison between the two above equations yields the required Laplace
transform pair.
Remark: We outline that the above auxiliary functions (for restricted values
of the parameters) turn out to be completely monotonic (CM) functions so
that they enter in some types of relaxation phenomena of physical relevance.
We recall that a function f (t) is CM in IR+ if (−1)n f n (t) ≥ 0 similarly with
e−t . For a Bernstein theorem, more generally they are expressed in terms of
a (generalized) real Laplace transform with positive measure
Z ∞
f (t) = e−rt K(r) dr , K(r) ≥ 0 .
0

Limiting to the auxiliary function in three parameters, we can prove for λ > 0
(
0 < α, β, γ ≤ 1 ,
eγα,β (t; λ) := tβ−1 Eα,β
γ
(−λtα ) CM iff
0 < αγ ≤ β .

Ex.16 With α > 0 show that

d
tα−1 Eα,α (−tα ) = − Eα (−tα ) .
dt
Solution: By using the series expansions of the Mittag-Leffler functions en-
tering in both sides the proof is as follows:
∞
α−1 α α−1
X tαn
t Eα,α (−t ) = t (−1)n
n=0
Γ(αn + α)

and
∞ ∞
d α d X n tαn X αn tαn−1
− Eα (−t ) = − (−1) =− (−1)n
dt dt n=0 Γ(αn + 1) n=1
Γ(αn + 1)

∞
X tαn−1
=− (−1)n .
n=1
Γ(αn)

Then, setting n = m + 1 so that m = n − 1, we get

∞ α(m+1)−1 ∞
d α
X
m t α−1
X tαm
− Eα (−t ) = (−1) =t (−1)m
dt m=0
Γ(α(m + 1)) m=0
Γ(αm + α)

= tα−1 Eα,α (−tα ) .

6
The equality, however, is more easily proved by using the Laplace transform
pairs of the left and right hand sides, as follows:
1
L.H.S. tα−1 Eα,α (−tα ) ÷ ,
sα +1
and, since Eα (0) = 1,

d sα−1 1
R.H.S. − Eα (−tα ) ÷ −s α +1= α .
dt s +1 s +1

Ex.17 Using the Laplace transform (LT-E1) in Ex.12, prove for 0 < α < 1:

tα t2α

1 − + · · · t → 0+ ,


Γ(α + 1) Γ(2α + 1)
Z ∞ 


Eα (−tα ) = e−rt Kα (r) dr '
0 
 t−α t−2α
− ··· t → +∞ ,


Γ(1 − α) Γ(1 − 2α)


with
1 rα−1 sin(απ) 1 sin(απ)
Kα (r) = = > 0 . (SE.1)
π r + 2 r cos(απ) + 1 π r r + 2 cos(απ) + r−α
2α α α

Hint: Inverting the Laplace transform according to the Bromwich formula

we need to compute only a proper integral along the branch cut on the real
semi-axis because the main branch of the Laplace transform does not exhibit
any other singularities in the half plane., see [Gorenflo and Mainardi (1997)].
We obtain
Z α−1 Z ∞
1 st s
α
Eα (−t ) = e α ds = e−rt Kα (r) dr
2πi Br s +1 0

where
sα−1 
  
1
Kα (r) = − Im .
π sα + 1 s=r eiπ
A trivial computation leads to the two equivalent expressions in (SE.1) for
Kα (r). The algebraic function Kα (r) turns to be positive for 0 < r < ∞:
in fact the numerator is positive being sin(απ) > 0 for 0 < α < 1 and the
denominator is, for any α not integer, always positive being (rα − 1)2 ≥ 0.
The positivity of such function (referred to as the spectral function) proves
that our Mittag-Leffler function is CM for 0 < α < 1. For α = 1 we recover
E1 (−t) = exp(−t) because K1 (r) = δ(r − 1).

7
The asympotic behaviour for t → 0 is of course simply derived by the Taylor
series, but it can be formally derived from the first few terms by inverting
the power series representation of the Laplace transform around s → ∞. In
fact
sα−1 1 1 1 1 1
α
= −α
' − α+1 + 2α+1 + · · · s → ∞ .
s +1 s1+s s s s
Similarly the asymptotic behaviour for t → +∞ can be formally derived by
inverting the power series representation of the Laplace transform as s → 0+ :

sα−1 1 1 1
α
' 1−α (1 − sα + · · · ) ' 1−α − 1−2α + · · · s → 0+ .
s +1 s s s

Ex.18 Using the integral representation in Ex.15 prove with λ > 0 that
Z ∞
α
eα (t; λ) := Eα (−λt ) = e−rt Kα (r; λ) dr ,
0

where
Kα (r; λ) = λ1/α Kα (λ1/α r) .
Ex.19 For sufficiently well-behaved causal functions f (t) with t ∈ IR+ , we
denote in (0, t) the integral operator of order n ∈ IN J n as the n-fold repeated
integral, namely
Z t Z τn−1 Z τ1
n
J f (t) := ... f (τ ) dτ dτ1 . . . dτn−1 .
0 0 0

The resulting function turns out to be the primitive of order n of f (t), i.e
fn (t) that is vanishing in t = 0+ along with its derivatives of order 1, . . . , n−1.
Check the Cauchy formula
Z t
n 1
J f (t) := fn (t) = f (τ )(t − τ )(n−1) dτ = Φn (t) ∗ f (t) ,
(n − 1)! 0

where Φn (t) := tn−1 /Γ(n) with t > 0 denotes the Gel’fand-Shilov function of
order n . For complementation we define J 0 := I, the identity operator.
Hint: The formula can easily be checked by using known properties of the
Laplace transforms because Φn (t) ÷ 1/sn
Ex.20 For sufficiently well-behaved causal functions f (t) with t ∈ IR+ , we
denote in (0, t) by Dn the derivative operator of order n ∈ IN, namely

n (n) dn f (t)
D f (t) := f (t) := .
dtn

8
For complementation we define D0 := I, the identity operator.
Extending (formally) the definition of the Gel’fand-Shilov functions for neg-
ative value of the index via generalized functions of Dirac type (impulsive
functions)

Φn (t) := t−n−1 /Γ(−n) = δ (n) (t) ÷ sn , n = 0, 1, . . . ,

prove that
Z t+
n
D f (t) = Φ−n (t) ∗ f (t); = f (τ ) δ (n) (t − τ ) dτ ,
0−

where the Laplace convolution, intended for generalized functions of Dirac

type, includes the extremes 0, t as internal points.
Hint: We recall for the Dirac type functions the known property
Z b
δ (n) (τ − τ0 ) dτ = (−1)n f n (τ0 ) , a < τ0 < b .
a

Since δ (n) (τ − τ0 ) = (1)n δ (n) (τ0 − τ ) we have

Z t+ Z t+
(n) n
f (τ ) δ (t − τ ) dτ = (−1) f (τ ) δ (n) (τ − t) dτ ,
0− 0−

the required identity is (formally) proved.

Ex.21 For sufficiently well-behaved causal functions f (t) with t ∈ IR+ , prove
that
n−1
n n n n
X tk
D J f (t) = f (t) , J D f (t) = f (t) − f (k) (0+ ) , n ∈ IN .
k=0
k!

This proves that in general Dn J n = I and J n Dn 6= I, so the derivative

of order n is the left inverse (but not necessarily the right inverse) of the
corresponding integral operator. Of course for both operators the semigroup
property holds

Dm Dn = Dn Dm = Dm+n , J m J n = J n J m = J m+n , m, n ∈ IN .

Hint: It is sufficient to consider the case n = 1 and then iterate the process
by induction. In fact

d t
Z Z t
f (τ ) dτ = f (t) , f 0 (τ ) dτ = f (t) − f (0+ ) .
dt 0 0

9
Ex.22 Prove that for the power-law (integrable in IR+ ) functions tγ with
γ > −1 we have

Γ(γ + 1) Γ(γ + 1)
J n tγ = tγ+n , Dn tγ = tγ−n , γ > −1 , t > 0 .
Γ(γ + 1 + n) Γ(γ + 1 − n)

Hint: To prove this is sufficient to recall from any course on calculus that

J n tγ = tγ+n /(γ + 1)n , Dn tγ = (γ)−n tγ−n ,

and then express Pchhammer’s symbols in terms of Gamma functions ac-

cording to Ex.10.
Ex.23 The Riemann-Liouville integral of order α > 0 for a sufficiently well-
behaved causal function f (t) is defined as straightforward generalization of
the n-fold repeated integral of Ex.19, namely
Z t
α 1
J f (t) = Φα (t) ∗ f (t) = (t − τ )α−1 f (τ ) dτ , t > 0 , α > 0 .
Γ(α) 0

In the above convolution integral, we refer to Φα (t − τ ) as the Abelian kernel

(so named after the Abel integral equations). We note that the Abelian
kernel is weakly singular for 0 < α < 1.
For complementation we define J 0 := I (Identity operator), i.e. we mean
J 0 f (t) = f (t).
Prove the following Laplace transform pair for the Riemann-Liouville integral

fe(s)
J α f (t) ÷ , α ≥ 0,
sα
which is the straightforward generalization of the corresponding formula for
the n-fold repeated integral.
Ex.24 Prove
Γ(γ + 1)
J α tγ = tγ+α , α ≥ 0, γ > −1 , t > 0,
Γ(γ + 1 + α)

which is the straightforward generalization of the corresponding formula for

the n-fold repeated integral.
Ex.25 Prove the semi-group property for the the Riemann-Liouville integral

J α J β = J α+β , α, β ≥ 0,

which implies the commutative property J β J α = J α J β .

10
Ex.26 The Riemann-Liouville derivative of order α > 0 for a sufficiently
well-behaved causal function f (t) is defined as the the left inverse of the
corresponding fractional integral of order α, that is

Dα J α f (t) = f (t) , α > 0.

In order to have, for α not integer, representations with integrable kernels

the only choice is to consider m − 1 < α ≤ m with m ∈ IN and define

Dα f (t) := Dm J m−α f (t) , with m − 1 < α ≤ m ,

namely

dm t f (τ )
 Z
1
 Γ(m − α) dtm 0(t − τ )α+1−m dτ , m − 1 < α < m,



Dα f (t) :=
m
 d f (t),



α = m.
dtm
For complementation we define D0 = I.
We easily verify in view of the semigroup property of the fractional integral

Dα J α = Dm J m−α J α = Dm J m = I .

We note that for any non integer α > 0 the kernel in the R-L fractional
derivative is always weakly singular. In particular for 0 < α < 1 we have

d t f (τ )
Z
α 1
D f (t) := dτ , 0 < α < 1 ,
Γ(1 − α) dt 0 (t − τ )α

Ex.27 Prove
Γ(γ + 1)
Dα tγ = tγ−α , α ≥ 0, γ > −1 , t > 0,
Γ(γ + 1 − α)
which is the straightforward generalization of the corresponding formula for
the derivative of order n.
This identity can be proved by using the definition of the fractional derivative
in terms of a fractional integral, Ex.26, by recalling the results in Ex.22 and
Ex.24.
Note the remarkable fact that when α is not integer (α 6∈ IN) the fractional
derivative Dα f (t) is not zero for the constant function f (t) ≡ 1. In fact we
have
t−α
Dα (1) = , α ≥ 0, t > 0,
Γ(1 − α)

11
which identically vanishes for α ∈ IN, due to the poles of the Gamma function
in the points 0, −1, −2, . . ..
Ex.28 Prove that the operator of the Riemann-Liouville derivative is con-
tinuous with respect to the integer order, namely, if m − 1 < α ≤ m with
m∈N
α → (m − 1)+ Dα f (t) → Dm−1 f (t) = f (m−1) (t) .
In fact, in view of the definition in Ex.26,
Dα f (t) → Dm J 1 f (t) = Dm−1 f (t) = f (m−1) (t) .

Ex. 29 Prove the following statement about functions which for t > 0 admit
the same Riemann-Liouville fractional derivative of order α
Xm
α α
D f (t) = D g(t) ⇐⇒ f (t) = g(t) + cj tα−j ,
j=1

where m − 1 < α ≤ m , m ∈ IN , and the coefficients cj are arbitrary

constants.
Ex.30 In addition to the Riemann-Liouville derivative we introduce the so-
called the Caputo derivative of order α > 0 with m − 1 < α ≤ m, m ∈ IN for
a sufficiently well-behaved causal function f (t), denoted by ∗ Dα , and defined
by
α m−α
∗ D f (t) := J Dm f (t) , with m − 1 < α ≤ m .
Thus it differs from the R-L derivative in that the processes of differentiation
and fractional integration are interchanged, so that it explicitly reads
 Z t (m)
1 f (τ )
 Γ(m − α) 0(t − τ )α+1−m dτ , m − 1 < α < m,



α
∗ D f (t) :=
m
 d f (t),



α = m.
dtm
In this way the Caputo derivative of a constant is zero. To distinguish the
Caputo derivative from the Riemann-Liouville derivative we have decorated
it with the additional apex ∗. For non-integer α the definition of the Caputo
derivative requires the absolute integrability of the derivative of order m.
Whenever we use the operator ∗ Dα we (tacitly) assume that this condition
is met.
Now it is worthwhile to exhibit the relation between the two fractional deriv-
ative. It turns out
m−1
α α
X tk−α
∗D f (t) = D f (t) − f (k) (0+ ) ,
k=0
Γ(k − α + 1)

12
and therefore, recalling the R-L fractional derivative of the power functions
in Ex.27, " #
m−1 k
α α
X t
∗ D f (t) = D f (t) − f (k) (0+ ) .
k=0
k!
We recognize that for non integer order the Caputo fractional derivative
represents a sort of regularization in the time origin for the Riemann-Liouville
fractional derivative. We also note that for its existence all the limiting
values,
f (k) (0+ ) := lim+ Dk f (t), k = 0, 1, . . . , m − 1 ,
t→0

are required to be finite. In the special case f (k) (0+ ) ≡ 0, we recover the
identity between the two fractional derivatives.
Let us now restrict our attention for 0 < α < 1 (i.e. m = 1) where we have
t
f (1) (τ )
Z
α 1
∗ D f (t) := α
dτ
Γ(1 − α) 0 (t − τ )
t−α
= Dα f (t) − f (0+ ) = Dα f (t) − f (0+ ) .
 
Γ(1 − α)

We limit to sketch the proof of the relationship between the two fractional
derivatives for this case. It is clear that for this purpose we must starting
from the R-L derivative and use the general formula
Z b(t) Z b(t)
∂ db da ∂f (t, τ )
f (t, τ ) dτ = f [t, b(t)] − f [t, a(t)] + dτ .
∂t a(t) dt dt a(t) ∂t

However we offer an alternative (formal) prove playing with operator symbols

and using the commutative property of the fractional integrals. Let us recall
the following identities

Dα := D1 J 1−α , ∗D
α
= J 1−α D1 ,

D1 J 1 f (t) = f (t) , J 1 D1 f (t) = f (t) − f (0+ ) .

Then
Dα [f (t) − f (0+ )] = Dα J 1 D1 f (t) = D1 J 1−α J 1 D1 f (t) = D1 J 1 J 1−α D1 f (t)
= J 1−α D1 f (t) := ∗ Dα f (t) .

Ex.31 Analyze the different behaviour of the R-L and Caputo fractional
derivatives at the end points of the interval (m − 1, m) namely when the
order is any positive integer.

13
With this exercise we complement the result in Ex.28 for the R-L derivative.
We note that whereas for α → m− both derivatives reduce to Dm , due to
the fact that the operator J 0 = I commutes with Dm , for α → (m − 1)+ we
have
 α
 D f (t) → Dm J 1 f (t) = D(m−1) f (t) = f (m−1) (t),
α → (m − 1)+ :
α
∗D f (t) → J 1 Dm f (t) = f (m−1) (t) − f (m−1) (0+ ).


As a consequence, formally speaking, we can say that Dα is, with respect to

its order α , an operator continuous at any positive integer, whereas ∗ Dα is
an operator only left-continuous.
The above behaviours have induced us to keep for the Riemann-Liouville
derivative the same symbolic notation as for the standard derivative of integer
order, while for the Caputo derivative to decorate the corresponding symbol
with subscript ∗.
Ex.32 Complement the Ex.29 analyzing the different behaviours of two func-
tions that have the same fractional derivative in the R-L and Caputo sense.
We can prove that, with m − 1 < α ≤ m , and cj arbitrary constants,
m
X
α α
D f (t) = D g(t) ⇐⇒ f (t) = g(t) + cj tα−j ,
j=1
Xm
α
∗D f (t) = ∗ Dα g(t) ⇐⇒ f (t) = g(t) + cj tm−j .
j=1

On this respect, we observe that the Caputo derivative for non integer order
m − 1 < α < m satisfies the same rule as the ordinary derivative of orderm.
Ex.33 Provide examples for which the fractional derivative of non integer
order
Dα Dβ 6= Dβ Dα , Dα Dβ 6= Dα+β .
Some examples are provided in [Gorenflo and Mainardi (1997)], see below.
However in the specialized literature there are conditions ensuring that

Dα Dβ = Dβ Dα = Dα+β .

In conclusion, in case of a non-integer order the semi-group property (valid

for the standard derivative of integer order as outlined in Ex.21) does not
hold in general.

14
To show how the Law of Exponents does not necessarily hold for the R-L
fractional derivative, we provide two simple examples (with power functions)
for which

(a) Dα Dβ f (t) = Dβ Dα f (t) 6= Dα+β f (t) ,
(b) Dα Dβ g(t) 6= Dβ Dα g(t) = Dα+β g(t) .

In the example (a) let us take f (t) = t−1/2 and α = β = 1/2 . Then, us-
ing Ex.24, we get D1/2 f (t) ≡ 0 , D1/2 D1/2 f (t) ≡ 0 , but D1/2+1/2 f (t) =
D f (t) = −t−3/2 /2 .
In the example (b) let us take g(t)
√ = t1/2 and α = 1/2 , β = 3/2 . Then, again
using Ex.24, we get D1/2 g(t) = π/2 , D3/2 g(t) ≡ 0 , but D1/2 D3/2 g(t) ≡ 0 ,
D3/2 D1/2 g(t) = −t3/2 /4 and D1/2+3/2 g(t) = D2 g(t) = −t3/2 /4 .
Ex.34
We recall that under suitable conditions the Laplace transform of the m-
derivative of f (t) is given by
m−1
X
m m
L {D f (t); s} = s fe(s) − sm−1−k f (k) (0+ )
k=0
m−1
X
= sm fe(s) − sk f (m−1−k) (0+ ) ,
k=0

where
f (k) (0+ ) := lim+ Dk f (t) .
t→0

For the Caputo derivative of order α with m − 1 < α ≤ m we have

m−1
X
L { ∗ Dα f (t); s} = sα fe(s) − sα−1−k f (k) (0+ ) ,
k=0

f (k) (0+ ) := lim+ Dk f (t) .

t→0

The corresponding rule for the Riemann-Liouville derivative of order α is

m−1
X
L {Dα f (t); s} = sα fe(s) − sm−1−k g (k) (0+ ) ,
k=0

g (k) (0+ ) := lim+ Dk g(t) , where g(t) := J (m−α) f (t) .

t→0

Hint: Let us prove the above rules for 0 < α < 1.

15
For the Caputo derivative let us put h(t) := D1 f (t) so that
h(s)
e sfe(s) − f (0+ )
L[ ∗ Dα f (t); s] = L[J 1−α D1 f (t); s] == = sα fe(s)−f (0+ ) .
s1−α s1−α
For the R-L derivative let us put g(t) := J 1−α f (t) so that
L[Dα f (t); s] = L[D1 J 1−α f (t); s] = se
g (s) − g(0+ )
fe(s)
= s 1−α − g(0+ ) = sα fe(s) − J 1−α f (0+ ) .
s
Remark: We point out the major utility of the Caputo fractional derivative
with respect to the R-L fractionl derivative in treating initial-value problems
for physical and engineering applications where initial conditions are usually
expressed in terms of integer-order derivatives. This can be easily seen using
the corresponding rules for Laplace transformation. In fact the rule for R-L
derivative requires initial values concerning an extra function g(t) related to
the given f (t) through a fractional integral. However, we must note that,
when all the limiting values f (k) (0+ ) for k = 0, 1, 2, . . . are finite and the
order is not integer, we can prove that the corresponding g (k) (0+ ) vanish so
that the formula for R-L drivative simplifies into
L {Dα f (t); s} = sα fe(s) , m − 1 < α < m .
We suggest to verify this result for 0 < α < 1 by considering that g(t) =
J 1−α f (t) as t → 0+ is vanishing necessarily when f (t) is sufficiently well-
behaved around t = 0 with f (0+ ) finite.
Ex.35 On fractional differential equation for anomalous relaxation. The dif-
ferent roles played by the R-L and Caputo derivatives and the Mittag-Leffler
functions are clear when one wants to consider the corresponding fractional
generalization of the first-order differential equation governing the phenom-
enon of (exponential) relaxation. Recalling (in non-dimensional units) the
initial value problem
du
= −u(t) , t ≥ 0 , with u(0+ ) = 1 ,
dt
whose solution is
u(t) = exp(−t) .
We consider the following generalizations involving the Caputo and R-L frac-
tional derivatives with α ∈ (0, 1):
α
(a) ∗D u(t) = −u(t) , t ≥ 0, withu(0+ ) = 1 ,
and
(b) Dα u(t) = −u(t) , t ≥ 0, with lim J 1−α u(t) = 1 .
t→0+

16
In analogy with the standard problem we solve the problems (a) and (b) with
the Laplace transform technique, using respectively the rules given for the
Caputo and R-L derivative. In the case (a) we get
sα−1
u
e(s) = α =⇒ u(t) = Eα (−tα ) ,
s +1
whereas in the case (b) we get
1 sα−1 d
u
e(s) = α
= 1 − s α
=⇒ u(t) = tα−1 Eα,α (−tα ) = − Eα (−tα ) .
s +1 s +1 dt
Ex.36 Taking u(0+ ) finite, prove the following chain of equivalent fractional
equations for t ≥ 0
α + α du
∗ D u(t) = −u(t) =⇒ u(t) − u(0 ) = −J u(t) =⇒ = −D1−α u(t) ,
dt
where ∗ Dα , Dα , J α denote the Caputo fractional derivative, the Riemann-
Liouville fractional derivative and the Riemann-Liouville fractional integral,
respectively, with 0 < α ≤ 1.
Solution: We denote
α
(1) ∗ D u(t) = −u(t) ,
(2) u(t) − u(0+ ) = −J α u(t) ,
du
(3) = −D1−α u(t) .
dt
To prove (1) =⇒ (2) we recall ∗ Dα = J 1−α D1 so that, applying to both sides
of (1) the operator J α , we get
J α J 1−α D1 u(t) = J 1 D1 u(t) = u(t) − u(0+ ) .
To prove (2) =⇒ (3) we recall D1−α = D1 J α so that, applying to both sides
of (2) the operator D1 , we get
du
D1 [u(t) − u(0+ )] = = −D1 J α u(t) = −D1−α u(t) .
dt
The equivalence of the three fractional equations can also be proved in the
Laplace domain
sα−1
(1) sα u
e(s) − sα−1 u(0+ ) = −eu(s) =⇒ u e(s) = u(0+ ) α ,
s +1
sα−1
(2) e(s) − u(0+ )/s = −e
u u(s)/sα =⇒ u e(s) = u(0+ ) α ,
s +1
α−1
s
(3) sue(s) − u(0+ ) = −s1−α ue(s) =⇒ u e(s) = u(0+ ) α .
s +1
As a matter of fact the use of the Laplace transform yields the same solution
u(t) = u(0+ ) Eα (−tα ) .

17
Main references
- Gorenflo, R. and Mainardi, F. (1997). Fractional calculus: integral and
differential equations of fractional order, in: Carpinteri, A. and Mainardi, F.
(Editors) Fractals and Fractional Calculus in Continuum Mechanics, Springer
Verlag, Wien, pp. 223–276. [E-print: http://arxiv.org/abs/0805.3823]
- Mainardi, F. and Gorenflo, R. (2007). Time-fractional derivatives in relax-
ation processes: a tutorial survey, Fractional Calculus and Applied Analysis
10, 269–308. [E-print: http://arxiv.org/abs/0801.4914]
- Mainardi, F. (2010). Fractional Calculus and Waves in Linear Viscoelas-
ticity, Imperial College Press, London and World Scientific, Singapore.

18