Professional Documents
Culture Documents
J Cnsns 2018 09 013
J Cnsns 2018 09 013
PII: S1007-5704(18)30293-4
DOI: https://doi.org/10.1016/j.cnsns.2018.09.013
Reference: CNSNS 4640
Please cite this article as: E. Capelas de Oliveira, S. Jarosz, J. VazJr., Fractional Calculus via Laplace
Transform and its Application in Relaxation Processes, Communications in Nonlinear Science and
Numerical Simulation (2018), doi: https://doi.org/10.1016/j.cnsns.2018.09.013
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service
to our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and
all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Highlights
• A unifying approach to fractional derivatives based on the theories of distribu-
tions and of Laplace transform is provided, elucidating the relation between the
Riemann-Liouville and the Caputo definitions, and showing how these and other
definitions can be generalized.
T
IP
CR
US
AN
M
ED
PT
CE
AC
1
ACCEPTED MANUSCRIPT
T
IP
Abstract
The fractional calculus has been receiving considerable in-
CR
terest in recent decades, mainly due to its several interesting
applications. In this paper we provide a very intuitive ap-
proach to the fractional calculus based on the Laplace trans-
form and ideas from the theory of distributions. Our ap-
US
proach reveals the deep relationship between the Riemann-
Liouville and the Caputo definitions of fractional derivative,
and opens the way for the formulation of other definitions,
AN
which we explore accordingly. As an example of its different
many applications, we use it to formulate some generaliza-
tions of a relaxation function model, and discuss some lim-
itations that these models impose on possible definitions of
M
1. Introduction
PT
Science studies nature mostly through the construction of models. The models can be of
phenomenological nature or can be based on theories, and specially in the case of physics,
they are written in mathematical language. Ever since the huge success of the application
CE
of the differential and integral calculus in the modelling of the motion of celestial bodies,
the calculus became the most important tool in physics for the modelling of physical
phenomena. However, the success of differential and integral calculus should not be seen
AC
as a reason to discourage the search for its possible generalizations; as a matter of fact,
the generalization of the concepts of derivative and of integral could be useful in the
modelling of processes that are not suitably described through the ordinary concepts of
1
capelas@ime.unicamp.br
2
stjarosz@gmail.com
3
vaz@ime.unicamp.br
2
ACCEPTED MANUSCRIPT
T
[17, 18, 19, 20, 21, 22]), dissipation [23, 24, 25, 26], relaxation [27, 28, 29], control theory
IP
[30, 31, 32], etc, as well as in formulation of fractional versions of physical theories [33],
like for example fractional quantum mechanics [34, 35, 36, 37].
It is important to notice however that one can find several different definitions in-
CR
volving the fractional versions of the integral and the derivative operators [38]. In this
scenario, it is essential to organize the ideas and clearly settle the rules that a definition
of fractional integrodifferential operator has to satisfy. This issue has been discussed
US
in [39]. The so-called Riemann-Liouville and Caputo derivatives [40, 41, 42] are exam-
ples of fractional derivatives satisfying the criteria of [39], and they are indeed the most
common ones found in the applications of fractional calculus. However, the presenta-
AN
tion of the Riemann-Liouville and the Caputo derivatives as formal definitions, in spite
of being an incontestable formal procedure, lacks motivation and leaves aside the con-
structive aspects of these generalizations. The objective of this paper is to provide an
introduction to the fractional calculus, and particularly to the Riemann-Liouville and
M
the Caputo derivatives, that can be done using the Laplace transform and the theory
of the Dirac delta function, and which can be easily generalized. It is closely related to
the approach of [43]. In spite of its simplicity, our approach reveals the deep relation
ED
between these two generalizations of the derivative concept. We aim to unfold this re-
lationship and to pave the way to expose this formalism in one of its many interesting
applications. We should also note that if we are interested only in the Caputo derivative,
PT
its possible generalizations have been studied in [44] in terms of the theory of complete
Bernstein functions, proving results concerning the existence and uniqueness of solutions
of relaxation and diffusion equations.
CE
This paper has been organized as follows. In section 2 we briefly discuss some impor-
tant issues from the theory of distributions that will be useful in this work, with special
attention to the Laplace transform and the relation between the derivatives in the distri-
AC
bution sense and in the ordinary sense. In section 3 we exploit the former discussion in
order to define a fractional distributional derivative and from it the Riemann-Liouville
and the Caputo versions of the fractional derivative. In section 4 we briefly discuss
how our approach can be generalized in order to encompass other possible definitions
of fractional derivatives. In section 5 we apply the concept of fractional derivative in
order to construct some models for a relaxation process that generalize the usual expo-
nential law and which include as special cases other known models in the literature. In
3
ACCEPTED MANUSCRIPT
T
as a continuous linear functional acting on a space of test functions D, which consists
of smooth functions with compact support on R. The Dirac delta function δ(t − t0 ) is
IP
defined in such a way that [46, 47]
hδ(t − t0 ), φ(t)i = φ(t0 ), (1)
CR
where φ(t) ∈ D. Locally integrable functions f (t) define regular distributions {f (t)}
through Z
h{f (t)}, φ(t)i =
The distribution theory not only places the definition of the delta function on a solid
AN
ground, but also allows us to perform operations on it, like for example calculating its
derivative or its Fourier and Laplace transforms. The derivative T 0 of a distribution T
is defined as [46, 47]
M
where F [φ(ω)](t) is the usual definition of the Fourier transform for ordinary functions.
Since F [D] 6⊂ D (that is, the Fourier transform of a function with compact support does
not have compact support), we have to enlarge the set of test functions in order to this
AC
definition makes sense. We do this by defining the set of test functions of slow growth S
as the set of smooth functions that decreases together with their derivatives, as |t| → ∞,
faster than any power of |t|−1 . Clearly D ⊂ S , and Schwartz proved that F [S ] ⊂ S .
Continuous linear functionals acting on S are called tempered distributions and its
space is denoted by S 0 . Note that S 0 ⊂ D. It follows from the above definitions that
F [T 0 (t)](ω) = −iωF [T (t)](ω). (7)
4
ACCEPTED MANUSCRIPT
T
0
0 0 0
h{f (t)} , φ(t)i = −hf (t), φ (t)i = − f (t)φ (t) dt − f (t)φ0 (t) dt, (9)
t+
IP
−∞ 0
CR
{f (t)}0 = {f 0 (t)} + ∆f (t0 )δ(t − t0 ), (10)
where {f 0 (t)} is the regular distribution associated with the usual derivative of the
ordinary function f (t) and
with
∆f (t0 ) = f (t+ −
0 ) − f (t0 ) US (11)
AN
f (t±
0 ) = lim+ f (t0 ± ). (12)
→0
n−1
X
(n) (n)
{f (t)} = {f (t)} + ∆f (k) (t0 )δ (n−1−k) (t − t0 ). (13)
k=0
ED
For a causal function f (t) we have f (0− ) = 0 and f (0+ ) = f (0) [48, 49], and its derivative
in the distributional sense is
CE
n−1
X
(n) (n)
{f (t)} = {f (t)} + f (k) (0)δ (n−1−k) (t). (16)
k=0
Eq.(15) and eq.(16) are very important results and are key formulas in our approach to
fractional derivatives (section 3).
5
ACCEPTED MANUSCRIPT
T
L [δ(t − t0 )](s) = e−st0 , (18)
IP
and that
L [T 0 (t)](s) = sL [T (t)](s). (19)
CR
The latter property apparently contrasts with the well-known property involving the
Laplace transform of the derivative of an ordinary function, that is,
US
L [f 0 (t)](s) = sL [f (t)](s) − f (0). (20)
This situation can be cleared after considering the expression for the derivative of a
discontinuous function. From eq.(14) we have
AN
L [{f (t)}0 ] = L [{f 0 (t)}] + [f (0+ ) − f (0− )]. (21)
Let us remember that the theory of Laplace transform involves causal functions [48, 49].
M
Then, if we use eq.(15) in eq.(19), and since L [δ(t)] = 1, we obtain eq.(20) with the
identification {f 0 (t)} = f 0 (t)H(t), where H(t) is the Heaviside step function.
ED
Remark: We can interpret eq.(21) in terms of the usual theory of the Laplace transform
once we define L± as Z ∞
L± [f (t)] = lim+ e−st f (t) dt. (22)
→0 ±
PT
If we compare the above equations with eq.(21) and take into account that L− [f (t)] =
L+ [f (t)], we see that eq.(21) can be interpreted in terms of the usual theory of Laplace
transform once we make the identifications
AC
6
ACCEPTED MANUSCRIPT
with the derivative calculated in the distribution sense. Consequently, since G1 (t) =
H(t), we have G0 (t) = δ(t). It follows that
T
L [Gν (t)] = s−ν ,
IP
ν > 0, (27)
CR
L [Gν (t)] = s−ν , ν ≤ 0. (28)
Z t
1
L f (τ ) dτ = L [f (t)]. (31)
0 s
If we use the notations
PT
Z t
D−1
t [f (t)] = f (τ ) dτ,
0 (32)
CE
0
D1t [f (t)] = Dt [f (t)] = {f (t)} ,
7
ACCEPTED MANUSCRIPT
T
(36)
−1
= L s L [G1−α (t)] · L [f (t)] ,
IP
L −1
L [G1−α (t)] · s L [f (t)] ,
CR
which can be written, after using eq.(19) and the convolution theorem, as
L −1
US
s L [G1−α (t)] ? f (t)] = {G1−α (t) ? f (t)}0 ,
AN
−1
L L [{G1−α (t)} ] · L [f (t)] = {G1−α (t)}0 ? f (t),
0
(37)
−1
L L [G1−α (t)] · L [{f (t)} ] = G1−α (t) ? {f (t)}0 .
0
M
The equivalence of these three different expressions is expected due to the well-known
result that (f ? g)0 = f 0 ? g = f ? g 0 .
ED
Γ(1 − α) α
0 (t − τ )
We can easily generalize this procedure for n = dαe = min{n ∈ Z | n ≥ α}, by writing
sα = sn sα−n in place of sα = ssα−1 in eq.(36). We define the fractional distributional
CE
derivative of order α as
Z t
α (n) 1 n f (τ )
Dt [f (t)] = {Gn−α (t) ? f (t)} = (Dt ) dτ, (39)
AC
Γ(n − α) α−n+1
0 (t − τ )
8
ACCEPTED MANUSCRIPT
f (0)t−α
+
Dαt [f (t)] = C Dαt [f (t)] + , (41)
Γ(1 − α)
Since t−α
+ /Γ(1 − α) → δ(t) for α → 1, we can interpret eq.(41) as the generalization of
eq.(15) for the case of a fractional derivative. In other words, the fractional distributional
T
derivative is the fractional generalization of the derivative in the distribution sense, while
IP
the Caputo derivative is the fractional generalization of the derivative in the ordinary
sense [50]. The generalization of eq.(42) is
CR
Z t
α (n) 1 f (n) (τ )
C Dt [f (t)] = Gn−α (t) ? {f (t)} = dτ, (43)
Γ(n − α) 0 (t − τ )α−n+1
Dαt [f (t)] = α
C Dt [f (t)]
US+
n−1
X f (k) (0)
tk−α , (44)
AN
k=0
Γ(k − α + 1) +
(n−1−k)
Gn−α ? δ (n−1−k) (t) = Gn−α (t) ? δ(t)
tk−α
+
(45)
= Gk+1−α (t) = .
ED
Γ(k − α + 1)
Moreover, since
(n)
tk+
PT
(n−1−k)
δ (t) = , (46)
k!
we can also write
CE
(n)
tk
Gn−α (t) ? δ (t) = Gn−α (t) ? +
(n−1−k)
k!
(n) (47)
AC
tk tk
= Gn−α (t) ? + = Dαt + ,
k! k!
9
ACCEPTED MANUSCRIPT
The difference between the fractional distributional derivative and the Riemann-Liouville
derivative follows from eq.(16), and it gives
n−1
X
Dαt [f (t)] = α
RL Dt [f (t)] + (Gn−α (t) ? f (t))(k) (0)δ (n−1−k) (t). (50)
k=0
T
Dαt [f (t)] = RL Dαt [f (t)], t > 0. (51)
IP
Remark: It is also important to consider the Laplace transform of the different defini-
tions of fractional derivatives. By construction, the Laplace transform of the fractional
CR
distributional derivative is
US
The Laplace transform of the Caputo derivative follows using eq.(44), that is,
n−1
X
L [ RL Dαt [f (t)]] α
= s L [f (t)] − sn−1−k (Gn−α (t) ? f (t))(k) (0). (54)
k=0
ED
This expression for the Laplace transform of RL Dαt [f (t)] clearly shows why it has not so
much interest in physics; indeed, while in eq.(53) we have the presence of the derivatives
f (k) (0), which have an obvious interpretation in terms of an initial value problem, in
PT
eq.(54) we have the terms (Gn−α (t) ? f (t))(k) (0), which lack any apparent interpretation.
4. Possible generalizations
CE
Our approach to the fractional integrodifferential operator was based on choosing eq.(34)
as a generalization of eq.(33). However, if we look to that situation from a more general
AC
point of view, we can think of other possible generalizations; that is, let us look for a
fractional integrodifferential operator defined in terms of a function Φ(s, α) as
and such that Φ(s, 1) = s, Φ(s, 0) = 1 and Φ(s, −1) = s−1 . Then eq.(34) was based on
the very natural choice Φ(s, α) = sα . However, other choices are also possible.
10
ACCEPTED MANUSCRIPT
In order to exemplify our idea, let us consider the slightly more modest task of defining
a fractional differential operator for 0 ≤ α ≤ 1. If we write Φ(s, α) in the form
Φ(s, α) = sψ(s, α), (56)
where ψ(s, 1) = 1 and ψ(s, 0) = s−1 , then, following the same steps of section 3, we
obtain that Z t
∗ α
Dt [f (t)] = Dt Ψ(t − τ, α)f (τ ) dτ, (57)
0
where
L [Ψ(t, α)] = ψ(s, α) (58)
T
and we have used the asterisk to distinguish this new case. A Caputo-like fractional
derivative, in analogy to eq.(42), is
IP
Z t
∗ α
C Dt [f (t)] = Ψ(t − τ, α)f 0 (τ ) dτ. (59)
CR
0
US
Recently, in their study of constitutive equations in materials, Fabrizio and Caputo
(FC) [51] have proposed a new definition of fractional derivative which can be modelled
in our approach by the function
AN
1
ψ1 (s, α) = , (61)
(1 − α)s + αt−1
0
Z t
∗ 1
α
FC Dt [f (t)] = e−(t−τ )α/t0 (1−α) f 0 (τ ) dτ. (62)
1−α 0
ED
In [51] the authors used t0 = 1. We notice, however, that recently Ortigueira and
Tenreiro Machado [52] and Tarasov [53] have criticized this definition as a candidate for
fractional derivative. As we will see in the next section, we also have some concerns
PT
s−1
ψ2 (s, α) = . (63)
(1 − α) + α(st0 )−α
The inverse Laplace transform of ψ2 (s, α) is given in terms of the Mittag-Leffler function
AC
[55], so that Z t
∗ 1 −α(t − τ )α
α
AB Dt [f (t)] = Eα f 0 (τ ) dτ, (64)
1−α 0 tα0 (1 − α)
where Eα (·) is the Mittag-Leffler function with parameter α. In [54] the authors used
t0 = 1. In the next section, we will see this definition has the same problem as the FC
one when we consider its use in fractional relaxation models.
11
ACCEPTED MANUSCRIPT
5. Fractional relaxation
A relaxation process is a response of a system to an external perturbation. Among the
characteristics properties of the response function, we will assume it to be a completely
monotonic function [56], that is, a function f (t) such that (−1)n f (n) (t) ≥ 0 for n =
0, 1, 2, . . ., since it ensures, for example, that in isolated systems the energy decays
monotonically. Completely monotonic functions have also a very useful characterization
as a result of Bernstein’s theorem [57]. Relaxation process is a very important process
in several areas of physics [58, 59, 60, 61, 62, 63]. Let us mention some cases: (i)
In the study of nuclear magnetic ressonance and magnetic ressonance imaging where
the term, associated with the relaxation, describes how a signal can change with time
T
[64, 65, 66]; (ii) The vibrational energy relaxation [67, 68, 69] is a process in which
population distribution amended by an external perturbation returns to the classical
IP
Maxwell-Boltzmann distribution [70]; (iii) Dieletric relaxation, which was the object of
a recent study [71], where the Cole-Cole [72, 73], the Davidson-Cole [74], the Havriliak-
CR
Negami [75, 76] and the Kolhrausch [77] were discussed, as models based on Mittag-
Leffler functions [55]; (iv) The structural relaxation [78, 79, 80] which is responsible for
the glass transitions was also a material of recently study [81, 82]. In all those cases, the
equilibrium position. US
term associated with the relaxation means the return of a perturbated system into the
The simplest model of relaxation function f (t) is the well-known exponential law
AN
f (t) = f (0) exp (−t/τ ), where τ is the relaxation time. If we want to study a model that
deviates from this exponential behaviour but which reproduces it in an appropriate limit,
fractional calculus can help us. In order to do this, we note that f (t) = f (0) exp (−t/τ )
satisfy the first order differential equation f 0 (t) = −τ −1 f (t); then, as a generalization of
M
it, we can consider a function satisfying fractional differential equations of order α that
has this ordinary differential equation as limit for α → 1.
The most natural fractional generalization of the differential equation f 0 (t) = −τ −1 f (t)
ED
is
α
? Dt [f (t)] = −Λ(τ, α)f (t), (65)
with 0 < α ≤ 1 and Λ(τ, α) is a function such that Λ(τ, 1) = τ −1 , like for example τ −α
PT
where τ is the relaxation time. The question mark in the above equation is a reference
to a question we must answer: which derivative should we consider? Riemann-Liouville
or Caputo? In order to answer this, we first note that, from eq.(102) and eq.(105) (in
CE
t−α
Dαt [1] = 0, Dαt [1] = , (t > 0). (66)
AC
C RL
Γ(1 − α)
But, from the physical point of view, we expect that, the larger the relaxation time is,
the closer the function is to its initial value. Consequently, for τ → ∞, we expect that
α
? Dt [const] = 0, and as we see from the above equation, this property is satisfied by the
Caputo derivative or by the Caputo-like generalizations as in eq.(59).
12
ACCEPTED MANUSCRIPT
Our model for a fractional relaxation function is given therefore by the solution of the
fractional differential equation
∗
C Dαt [f (t)] = −Λ(τ, α)f (t), (67)
with 0 < α ≤ 1. Just like in the α = 1 case, we can solve this equation by using the
Laplace transform. Using eq.(60) we have
ψ(s, α)sF (s) − ψ(s, α)f (0) = −Λ(τ, α)F (s), (68)
which gives
F (s) = H(s, α)f (0), (69)
T
where
IP
ψ(s, α)
H(s, α) = . (70)
sψ(s, α) + Λ(τ, α)
CR
The solution of eq.(67) is therefore
where
US
h(t, α) = L −1 [H(s, α)]. (72)
AN
Different relaxation models are therefore characterized by different choices of the function
ψ(s, α) which models the fractional derivative. However, the choice of the function
ψ(s, α) has to satisfy one extra condition, that is,
M
Using the initial value theorem of the Laplace transform [57], this condition can be
ED
replaced by
lim sH(s, α) = 1, (74)
s→∞
1
lim = 1, (75)
s→∞ 1 + Λ(τ, α)[sψ(s, α)]−1
CE
which gives
1
lim = 0. (76)
s→∞ sψ(s, α)
AC
The condition eq.(76) is clearly satisfied the Caputo derivative with ψ(s, α) = sα−1
for α > 0. However, the FC derivative with ψ(s, α) given by eq.(61) does not satisfy it,
as well as the AB derivative with ψ(s, α) given by eq.(63). This means that solutions of
eq.(67) with these derivatives will not satisfy the initial condition (except for the integer
case α = 1). This fact completely excludes the usefulness of the FC and AB fractional
derivatives in fractional relaxation models of the form of eq.(67). In appendix B we
explicitly show this fact by presenting the solutions of the models with these derivatives.
13
ACCEPTED MANUSCRIPT
We also note that the FC and AB derivatives does not satisfy the conditions supposed
by Kochubei for the proposed generalizations of the Caputo derivative in [44].
Let us consider therefore fractional relaxation model based on the Caputo derivative
in the form
α −α
C Dt [f (t)] = −τ f (t). (77)
In this case we have
sα−1 s−1
H(s, α) = = , (78)
sα + τ −α 1 + (τ s)−α
which can be inverted using
s−1
T
−1
f (t) = f (0) L
1 + (τ s)−α
IP
X∞ (79)
= f (0)τ (−1)n L −1 (τ s)−αn−1 ,
CR
n=0
f (t)
1.0
ED
0.8
0.6
PT
0.4
0.2
CE
t
1 2 3 4 5 6
Figure 1: Plots of exp(−t) = E1 (−t) (continuous curve), E0.8 (−t0.8 ) (dashed curve) and
E0.6 (−t0.6 ) (dotted curve).
AC
The model eq.(77) can also be solved using the usual power series method. In order
to include the fractional case we consider a series of the form
∞
X
f (t) = an tαn . (81)
n=0
14
ACCEPTED MANUSCRIPT
which gives
∞
X
α Γ(α(n + 1) + 1) αn
C Dt [f (t)] = an+1 t . (83)
n=0
Γ(αn + 1)
Then eq.(67) gives the recurrence relation
τ −α Γ(αn + 1)
T
an+1 = − an , (84)
Γ(α(n + 1) + 1)
IP
which gives
(−τ −α )n
CR
an = a0 , (85)
Γ(αn + 1)
and the solution of eq.(67) is
f (t) = f (0)
∞
X [−(t/τ )α ]n
n=0
US
Γ(αn + 1)
≡ f (0) Eα [−(t/τ )α ], (86)
AN
as expected.
Another possible generalization of f 0 (t) = −τ −1 f (t) is
with 0 < α ≤ 1 and where τ is the relaxation constant with usual dimension [T ]−1 . This
is a special case of the class of models for anomalous relaxation studied in [29, 83]. The
ED
n=0
Γ(n − α + 2)
an+1 = −τ −1 an , n = 0, 1, 2, . . . (89)
Γ(n + 2)
and the general term can be written in the form
AC
(−τ −1 )n G(2 − α + n)
an = a0 , (90)
n! G(2 − α) G(1 + n)
15
ACCEPTED MANUSCRIPT
T
for different values of α for f (0) = τ = 1.
IP
f (t)
1.0
CR
0.8
0.6
0.4
0.2
US
AN
t
1 2 3 4 5 6
M
Figure 2: Plots of exp(−t) = F1 (−t) (continuous curve), F0.8 (−t) (dashed curve) and
F0.6 (−t) (dotted curve).
ED
6. Concluding remarks
PT
The fractional derivative definitions of Riemann-Liouville and of Caputo are the most
common ones found in the applications of fractional calculus. We have provided an
introduction to the Riemann-Liouville and the Caputo derivatives based on the Laplace
CE
transform and on the theory of distributions, from which we had naturally led to a
definition of a fractional distributional derivative. We have clarified the deep relation
between these generalizations of the derivative concept, and obtained some important
AC
properties of them. We have also seen how to deal with some other possible definitions
of fractional derivative in our approach, in particular the definitions of Fabrizio-Caputo
and Atangana and Baleanu. As an example of possible applications of the formalism,
we have discussed some generalizations of the exponential law relaxation model. We
have found that only the Caputo derivative provides a consistent model satisfying the
initial conditions for an arbitrary order α (0 < α ≤ 1) of the fractional derivative. We
provided the solutions for two specific models.
16
ACCEPTED MANUSCRIPT
T
and on the right of a point, and in the sequel, we will use these integrals to define the
IP
fractional distributional derivative, and from it the Riemann-Liouville and the Caputo
derivatives.
Let Ω = [a, b] with −∞ < a < b < ∞ be a finite interval on the real axis R. The
CR
α
integrals of arbitrary orders in the Riemann-Liouville sense, denoted by, (Ia+ f )(t) ≡
a It f (t) and (Ib− f )(t) ≡ t Ib f (t) both of order α ∈ C, with Re(α) > 0 are defined by
α α α
Z t
α
Liouville sense on the left and on the right, respectively. Note that (Ia+ f )(t) and
α
(Ib− f )(t) are adjoint, in the sense that
Z Z
ED
b b
α α
f (t)(Ia+ g)(t) dt = (Ib− f )(t)g(t) dt. (96)
a a
The left and right fractional distributional derivatives in a finite interval on the real
PT
axis, denoted by Dαa+ y and Dαb− y respectively, of order α ∈ C with Re(α) ≥ 0, are
Dαa+ [y(t)] = (Dt )n Ia+
n−α
y (t)
CE
Z t
1 n y(τ ) (97)
= (Dt ) α−n+1
dτ,
Γ(n − α) a (t − τ )
AC
17
ACCEPTED MANUSCRIPT
The left and right Caputo derivatives in a finite interval on the real axis, denoted by
α α
C Da+ y and C Db− y respectively, are defined like in eq.(48), that is,
" n−1 (k)
#
X y (a)
α α
C Da+ [y(t)] = Da+ y(t) − (t − a)k+ (99)
k=0
k!
and " #
n−1 (k)
X y (b)
C Dαa+ [y(t)] = Dαa+ y(t) − (b − t)k+ . (100)
k=0
k!
Using the beta function [85], we can find that
k!(t − a)k+α
T
0
α0 +
Ia+ (t − a)k+ = , (101)
Γ(α0 + k + 1)
IP
and consequently that
k!(t − a)k−α
CR
+
Dαa+ [(t − a)k+ ] = . (102)
Γ(k − α + 1)
An analogous expression holds for Dαb− [(b − t)k ], and then we have that
−
n−1
X US
Dαa+ [y(t)] = Dαa+ [y(t)]
y (k) (a)
Γ(k − α + 1)
(t − a)k−α
+ ,
(103)
AN
k=0
and
C Dαb− [y(t)] = Dαb− [y(t)]
n−1
X
M
a+
C
Γ(n − α) a (t − τ )α−n+1
and Z b
α (−1)n y (n) (t)
C Db− [y(t)] = dτ, (106)
CE
Γ(n − α) t (τ − t)α−n+1
where n = dRe(α)e.
The left and right Riemann-Liouville derivatives in a finite interval on the real axis,
AC
18
ACCEPTED MANUSCRIPT
d
with t < b, and where dt
denotes the usual derivative and n = dRe(α)e.
T
derivatives
IP
In this appendix we illustrate with examples the problem discussed in the main text
about the failure of solutions of fractional differential equations of the form of eq.(65)
CR
with FC and AB derivatives to satisfy the initial condition. Let us consider models of the
form of eq.(77), which are particular cases of eq.(65), but using FC and AB derivatives.
Let us start with FC derivative,
∗
FC
US
Dαt [f (t)] = −τ −α f (t),
with 0 < α ≤ 1 and τ is a constant with physical dimension [T ]. We can solve this
(109)
AN
equation by the Laplace transform method. The Laplace transform of the FC derivative
is
sL[f ] − f (0)
L [ ∗FC Dαt [f (t)]] = . (110)
(1 − α)s + αt−10
M
(τ α + 1 − α)s + α
from which we obtain the solution
τα α
PT
AB (113)
with 0 < α ≤ 1 and τ is a constant with physical dimension [T ]. Once again, we can
solve this equation by the Laplace transform method. Since the Laplace transform of
the one-parameter Mittag-Leffler function is given by [55]
sα−1
L [Eα (−λ tα )] = , (114)
sα − λ
19
ACCEPTED MANUSCRIPT
1.0
0.8
0.6
0.4
0.2
1 2 3 4 5 6
T
Figure 3: Plots of solution (112) α = 1 (continuous curve), α = 0.8 (dashed curve) and
IP
α = 0.6 (dotted curve).
1.0
CR
0.8
US
0.6
0.4
AN
0.2
1 2 3 4 5 6
M
τ α sα−1
F (s) = f (0) , (116)
(τ α + 1 − α)sα − α
AC
20
ACCEPTED MANUSCRIPT
References
[1] K. S. Miller and B. Ross, An Introduction to the Fractional Calculus and Fractional
Differential Equations, Wiley, New York, (1993).
[2] B. Ross, (ed.), Fractional Calculus and its Applications: Proceedings of the Inter-
national Conference, New Haven, June 1974, Springer Verlag, New York, (1974).
[3] R. Hilfer (ed.), Applications of Fractional Calculus in Physics, World Scientific Publ.
Co., Singapore (2000).
[4] K. Diethelm, The Analysis of Fractional Differential Equation, Lecture Notes in
T
Mathematics 2004, Springer-Verlag, Berlin-Heidelberg (2010).
IP
[5] M. D. Ortigueira, Fractional Calculus for Scientists and Engineers, Lectures Notes
in Electrical Engineering 84, Springer, Dordrecht (2011).
CR
[6] R. Herrmann, Fractional Calculus: an Introduction for Physicists, 2nd ed., World
Scientific Publ. Co., New Jersey/Singapore (2014).
US
[7] T. M. Atanacković, S. Pilipović, B. Stanković and D. Zorica, Fractional Calculus
with Applications in Mechanics: Vibrations and Diffusion Processes, ISTE Ltd,
London (2014).
AN
[8] T. M. Atanacković, S. Pilipović, B. Stanković and D. Zorica, Fractional Calculus
with Applications in Mechanics: Wave Propagation, Impact and Variational Prin-
ciples, ISTE Ltd, London (2014).
M
[9] H. Fallahgoul, S. Focardi and F. Fabozzi, Fractional Calculus and Fractional Pro-
cesses with Applications to Financial Economics: Theory and Application, Aca-
demic Press, Elsevier Limited, London (2017).
ED
[10] A. I. Saichev and G. Zaslavsky, “Fractional kinetic equations: solutions and appli-
cations”, Chaos 7, 753-764 (1997).
PT
[11] R. Metzler and J. Klafter, “The random walk’s guide to anomalous diffusion: a
fractional dynamics approach”, Phys. Rept. 339, 1-77 (2000).
CE
[12] F. Mainardi, Y. Luchko and G. Pagnini, “The fundamental solution of the space-
time fractional diffusion equation”, Fract. Calc. Appl. Anal. 4, 153-192 (2001);
arXiv:cond-mat/0702419v1.
AC
21
ACCEPTED MANUSCRIPT
[15] R. Figueiredo Camargo, E. Capelas de Oliveira and J. Vaz Jr., “On anomalous dif-
fusion and the fractional generalized Langevin equation for a harmonic oscillator”,
J. Math. Phys. 50, 123518 (2009).
[16] J. Klafter, S. C. Lim and R. Metzler (editors), Fractional Dynamics: Recent Ad-
vances, World Scientific Publ. Co., Singapore (2012).
[17] S. Fomin, V. Chugunov and T. Hashida, “The effect of non-Fickian diffusion into
surrounding rocks on contaminant transport in a fractured porous aquifer”, Prof.
R. Soc. A 461, 2923-2939 (2005).
T
impedance spectroscopy of electrolytic cells”, J. Phys. Chem. B 113, 11371-11374
IP
(2009).
CR
models for anomalous electrodiffusion in nerve cells: infinite domain solutions”, J.
Math. Biol. 59, 761-808 (2009).
[20] N. Korabel and E. Barkai, “Anomalous infiltration”, J. Stat. Mech. Theor. Exp.
P05022 (2011); arXiv:1102.2592v1.
US
[21] P. A. Santoro, J. L. de Paula, E. K. Lenzi and L. R. Evangelista, “Anomalous
AN
diffusion governed by a fractional diffusion equation and the electrical response of
an electrolytic cell”, J. Chem. Phys. 135, 114704 (2011).
[22] O. C. Ibe, Elements of Random Walk and Diffusion Processes, Wiley, New Jersey
M
(2013).
[23] L. Gaul, P. Klein and S. Kemple, “Damping description involving fractional oper-
ED
[24] Ya. E. Ryabov and A. Puzenko, “Damped oscillations in view of the fractional
oscillator equation”, Phys. Rev. B 66, 184201 (2002).
PT
[25] A. Tofighi, “The intrinsic damping of the fractional oscillator”, Physics A 329,
29-34 (2003).
CE
[26] M. Naber, “Linear fractionally damped oscillator”, Int. J. Diff. Equ., 2010, Article
ID 197020, (2010).
AC
22
ACCEPTED MANUSCRIPT
[29] E. Capelas de Oliveira, F. Mainardi and J. Vaz Jr., “Fractional models of anomalous
relaxation based on the Kilbas and Saigo functions”, Meccanica 49, 2049-2060
(2014).
T
[32] P. Shah and S. Agashe, “Review of fractional PID controller”, Mechatronics 38,
IP
29-41 (2016).
CR
ics of Particles, Fields and Media, Higher Education Press, Beijing and Springer-
Verlag Berlin Heidelberg (2010).
[34] E. Capelas de Oliveira, F. S. Costa and J. Vaz Jr., “The fractional Schrödinger
US
equation for delta potentials”, J. Math. Phys. 51, 123517 (2010).
[35] E. Capelas de Oliveira and J. Vaz Jr., “Tunneling in fractional quantum mechanics”,
AN
J. Phys. A: Math. Theor. 44, 185303 (2011).
[36] S. Kondej and J. Vaz Jr., “Fractional Schrödinger equation with delta potential
localized on circle”, J. Math. Phys. 53, 033503 (2012).
M
[37] S. Jarosz and J. Vaz Jr., “Fractional Schrödinger equation with Riesz-Feller deriva-
tive for delta potentials”, J. Math. Phys. 57, 123506 (2016).
ED
tives: Theory and Applications, Gordon and Breach Science Publishers, Switzer-
land, (1993).
23
ACCEPTED MANUSCRIPT
[43] R. Gorenflo and F. Mainardi, “Fractional Calculus: Integral and Differential Equa-
tions of Fractional Order”, in A. Carpinteri and F. Mainardi (Editors) Fractals and
Fractional Calculus in Continuum Mechanics, pp. 223-276, Springer Verlag, Wien
and New York (1997).
[45] L. Schwartz, Théorie des Distributions, I-II, Paris: Hermann & Cie, (1950-1951).
[46] A. H. Zemanian, Distribution Theory and Transform Analysis, Dover Publ. Inc.,
New York (1987).
T
[47] R. P. Kanwal, Generalized Functions: Theory and Technique, Academic Press, New
IP
York (1983).
CR
[48] K. H. Lundberg, H. R. Miller and D. L. Trumper, “Initial conditions, generalized
functions and the Laplace transform”, Control Systems IEEE 27, 22-35 (2007).
[49] R. J. Beerends, H. G. ter Morsche, J. C. van der Berg and E. M. van de Vrie,
US
Fourier and Laplace Transforms, Cambridge University Press, Cambridge (2003).
[50] I. Podlubny, “Fractional Differential Equations”, Academic Press, San Diego (1999).
AN
[51] M. Caputo and M. Fabrizio, “Applications of new time and spatial fractional deriva-
tives with exponential kernels”, Progr. Fract. Differ. Appl. 2, 1-11 (2016).
Fabrizio operator”, Commun. Nonlinear Sci. Numer. Simul. 59, 608-611 (2018).
[54] A. Atangana and D. Baleanu, “New fractional derivatives with nonlocal and non-
singular kernel: theory and application to heat transfer model”, Thermal Science
PT
[57] D. V. Widder, The Laplace Transform, Dover Books on Mathematics, Dover Pub-
lications, Inc. (2011).
24
ACCEPTED MANUSCRIPT
[58] P. Debye, “Zur Theorie der anomalen Dispersion im Gebiete der langwelligen elek-
trischen Strahlung”, Verh. Dtsch. Phys. Ges. 15, 777, (1913). Reprinted in The
Collected Papers of Peter J. W. Debye, Interscience Publishers Inc., New York
(1954).
[59] S. Brawer, Relaxation in Viscous Liquids and Glasses, American Ceramic Society,
Columbus, Ohio (1985).
[61] R. Hilfer, “On fractional relaxation”, Fractals 11, Supplementary Issue, 251–257
T
(2003).
IP
[62] V. V. Novikov, K. W. Wojciechowski, O. A. Komkova and T. Thiel, “Anomalous
relaxation in dielectrics. Equations with fractional derivatives”, Material Science -
CR
Poland 23, 977–984 (2005).
US
[64] J. L. Dormann, D. Fiorani and E. Tronc, “Magnetic relaxation in fine-particle
systems”, in Advances in Chemical Physics, 98, (eds I. Prigogine and S. A. Rice),
AN
John Wiley & Sons, Inc., Hoboken (1997).
York (1999).
[67] R. L. Taylor and S. Bitterman, “Survey of vibrational relaxation data for processes
important in the CO2-N2 laser system”, Rev. Mod. Phys. 41, 26–47 (1969).
PT
(2002).
[69] H. Fujisaki and J. E. Straub, Vibrational energy relaxation in proteins, PNAS 102,
6726-6731 (2005).
AC
[71] E. Capelas de Oliveira, F. Mainardi and J. Vaz Jr., Models based on Mittag-Leffler
functions for anomalous relaxation in dielectrics, Eur. Phys. J. Special Topics 193,
161–171 (2011). arXiv:1106.1761.
25
ACCEPTED MANUSCRIPT
[73] K. S. Cole and R. H. Cole, “Dispersion and absorption in dielectrics. II. Direct
current characteristics”, J. Chem. Phys. 10, 98–105 (1942).
[75] S. Havriliak Jr. and S. Negami, “A complex plane analysis of α-dispersions in some
polymer systems”, J. Polymer Sci. C 14, 99–117 (1966).
T
[76] S. Havriliak and S. Negami, “A complex plane representation of dielectric and
mechanical relaxation processes in some polymers”, Polymer 8, 161–210 (1967).
IP
[77] R. Kohlrausch, “Theorie des Elektrischen Rückstandes in der Leidener Flasche”,
CR
Pogg. Ann. Phys. Chem. 91, 179–214 (1854).
[81] Chao Liu, Gao Li, D. R. Kauffman, G. Pang and R. Jin, “Synthesis of ultrasmall
PT
[82] Q. Guo, B. Zheng, S. Zhou, B. Xu, Y. Qiu, Y. Yu, and J. Qiu, “Origin of structural
relaxation dependent spectroscopic features of bismuth-activated glasses”, Opt. Ex-
press 22, 15924–15931 (2014).
AC
26
ACCEPTED MANUSCRIPT
T
IP
CR
US
AN
M
ED
PT
CE
AC
27