BULLETIN OF THE POLISH ACADEMY OF SCIENCES INVITED PAPER

TECHNICAL SCIENCES, Vol. 66, No. 4, 2018

DOI: 10.24425/124254

On fractional vectorial calculus

M.D. ORTIGUEIRA1∗ and J.A.T. MACHADO2
1
CTS-UNINOVA and DEE of Faculdade de Ciências e Tecnologia da UNL, Campus da FCT da UNL, Quinta da Torre,
2829 – 516 Caparica, Portugal
2
Institute of Engineering, Polytechnic of Porto, Dept. of Electrical Engineering, Porto, Portugal

Abstract. This paper reviews the fractional vectorial differential operators proposed previously and introduces the fractional versions of the
classic Green’s, Stokes’, and Ostrogradski-Gauss’s integral theorems. The suitability of fractional derivatives for sciences and the Laplacian
definition are also discussed.
Key words: Grünwald-Letnikov, Liouville, fractional gradient, fractional divergence, fractional curl, fractional Laplacian, Green’s theorem,
Stokes’ theorem, Ostrogradski-Gauss’s theorem.

1. Introduction tools, unlike the classical ones, look more suitable for dealing
with non-homogeneous and anisotropic media.
Fractional calculus (FC) was born almost the same time as the
With those operators the generalized Helmholtz decompo-
integer order calculus [1–6] and resulted from a discussion be-
sition theorem for fractional space and time is reviewed and
tween Leibniz and Bernoulli. Recently FC abandoned a pure
decoupled by means of fractional wave equations for fields or
mathematical perspective and was recognised by the applied
potentials. The generalization is based on differential opera-
sciences community, namely physics and engineering [7–23],
tors and points to the need for an integral formulation.
giving rise to interesting applications in many fields, such
as biology [24, 25], biomedical engineering [4, 26–28], fi- The importance of the classical Green’s, Stokes’, and
nance [29], and signal processing [30] just to mention a few. Ostrogradski-Gauss’s integral theorems is unquestionable.
Hence, attempts to derive a fractional formulation based on
FC is an essential tool for modelling long memory and
two different point of views have been made, namely, frac-
long range processes. This non-locality in time and space can
tal geometry [35, 36, 39, 45] and FC [31, 39, 46]. In this line
be found in many phenomena, such as the diffusion process-
of thought, we propose the fractional version of such theo-
es [31–34], anomalous porous media [35], and fractional
rems for rectangular domains. The starting point is the gen-
spaces [36]. Viscoelasticity is another subject of active stud-
eralization of the fundamental theorem of calculus. Several
ies [37]. The modelling of mechanical systems, from 1-D [38]
researchers tried this topic [16, 46, 47], but without having
to n-D [16, 39, 40], is also an important subject that attract-
the support of the notion of fractional definite integral (FDI)
ed the attention of researchers. This has been accomplished
and, consequently, achieving limited results. However, recent-
through the fractionalization of Hamiltonian and gradient con-
ly this concept was tackled by means of the generalization
cepts [16, 17, 39, 40]. The general case of field theories was
of the Barrow formula [48]. Having defined the FDI on R,
tackled by Herrmann [41, 42].
we can introduce definite integrals in R2 and R3 for rec-
The generalization of the vectorial operators: gradient, di- tangular domains. Here we use those ideas to generalize the
vergence, curl, and Laplacian was discussed in [43], starting classic Green’s, Stokes’, and Ostrogradski-Gauss’s theorems,
by the space version of the fractional forward and backward bearing in mind a full agreement with the fractional vecto-
Grünwald-Letnikov and Liouville derivatives [5, 44]. These rial operators proposed in [43]. In a wider scope, this work
operators are used to define a pair of left and right fractional follows the generalization of the FC the concepts and pro-
gradients that lead to the corresponding divergences and curls. vides a straightforward methodology for the adoption of FC
The fractional Laplacian is obtained from the inner product of in applied sciences.
the left and right gradients. Such operators are backward com-
The manuscript is organized as follows. Section 2 defines
patible in the sense that they recover the classical definitions
the main fractional derivatives suitable for fractional vecto-
when the order is 1.
rial calculus. Time and space derivatives are introduced using
An important aspect of our formulation concerns the use the Grünwald-Letnikov and Liouville formulations. Two-sided
of distinct orders for different directions. These non-local derivatives are also included due to their importance in the
development of differential vectorial tools, namely the Lapla-
∗ e-mail: [email protected] cian. The tools are described in Sec. 3, where we analyze the
Manuscript submitted 2017-12-06, revised 2018-02-14 and 2018-03-06, left and right gradients, divergences, and curls, called “first
initially accepted for publication 2018-03-07, published in August 2018. generation operators”, leading to a second group of “second

Bull. Pol. Ac.: Tech. 66(4) 2018 389

 M.D. Ortigueira and J.A.T. Machado

generation operators” and particularly to the fractional Lapla- • We will work with the multivalued expressions sα and
cian. This is compared with the fractional Laplacian based (−s)α . To obtain functions from them, we fix for branchcut
on the Riesz potential. The generalized Helmholtz decom- lines the negative real half axis for the first and the positive
position is recalled and its solution expressed in terms of real half axis for the second; for both we work on the first
those corresponding to several fractional wave equations. In Riemann surface.
Sec. 4, the fractional definite integrals are described. They are
used in Sec. 5 to extend the classic integral theorems towards 2. Suitable fractional derivatives
the fractional case. The Green’s, Stoke’s, and Ostrogradski-
2.1. Previous comments. In [51], two criteria were proposed
Gauss’s theorems in rectangular domains are formulated. The
for deciding if a given operator can be considered as a frac-
particular cases of uniform fractional orders is also discussed.
tional derivative. According to such criteria, there are several
Finally, Sec. 6 outlines the main conclusions.
acceptable definitions. Later, it was shown that only some of
such derivatives are suitable when thinking on the generaliz-
1.1. Remarks ing classic tools [44]. In particular, they must verify the index
• We will assume that we are working on R3 and that the law in order to ensure that, given a derivative, there exists its
set {e1 , e2 , e3 } constitutes its standard orthonormal base. inverse, that is the anti-derivative, [48]. In the follow-up such
Therefore, each vector has the representation derivatives are recalled.

v = v1 e1 + v2 e2 + v3 e3 . 2.2. About time derivatives. Let f (t), t ∈ R, be a function

of time, having Laplace transform, F (s), with a given re-
In particular, we define the vector
gion of convergence and introduce the Pochhammer symbol
r = x1 e1 + x2 e2 + x3 e3 . (1) (a)n = a(a + 1) · · · (a + n − 1), n ∈ N, with (a)0 = 1. Let
us define two sets of fractional derivatives according to the
• A vectorial function is represented by arrow of time, namely the forward and backward derivatives:
f (x1 , x2 , x3 ) = f1 (x1 , x2 , x3 )e1 + f2 (x1 , x2 , x3 )e2 Definition 2.1 (forward Grünwald-Letnikov).
X∞
+f3 (x1 , x2 , x3 )e3 . (−α)n
Dfα f (t) = lim+ h−α f (t − nh), (6)
• When necessary, we write α = (α1 , α2 , α3 ) and ω = h→0
n=0
n!
(ω1 , ω2 , ω3 ) and similarly for other cases, as s. Definition 2.2 (backward Grünwald-Letnikov).
• We will use the two-sided Laplace transform (LT) of ∞
X (−α)n
f (x1 , x2 , x3 ) defined on R3 and given by: Dbα f (t) = e−iπα lim h−α f (t + nh). (7)
h→0+
n=0
n!
F (s) = L [f (x1 , x2 , x3 )]
Z There are several properties exhibited by (6) and (7) [51]:
(2)
= f (x1 , x2 , x3 )e−(s·r) dx1 dx2 dx3 , • Linearity
R3
• Additivity and commutativity of the orders (index law). If
we apply (6) twice for any two orders α and β, we have
under the usual existence conditions. The inverse LT is [49]
a1Z+i∞ a2Z+i∞ a3Z+i∞
Dfα Dfβ f (t) = Dfβ Dfα f (t) = Dfα+β f (t). (8)
1 • Neutral and inverse elements
f (x1 , x2 , x3 ) = F (s)e(s·r) d3 s,
(i2π)3
a1 −i∞ a2 −i∞ a3 −i∞ Dfα Df−α f (t) = Df0 f (t) = f (t). (9)
(3)
where a = (a1 , a2 , a3 )√is in the region of convergence of From (9) we conclude that there is always an inverse el-
the transform and i = −1. ement, that is, for every order α there is always the −α
• The Fourier transform (FT) is obtained from the LT using order derivative.
the substitution s = iω with ω ∈ R3 is defined by the • Backward compatibility (n ∈ N)
synthesis equation [50] If α = n, then:
Pn

Z (−1)k nk f (t − kh)
1 Dfn f (t) = lim k=0 .
f (x1 , x2 , x3 ) = F (ω)ei(ω.r) d3 ω (4) h→0 hn
(2π)3
R3 We obtain this expression repeating the first order deriva-
and the corresponding analysis equation tive.
If α = −n, then:
F (ω) = F [f (x1 , x2 , x3 )] n
X (n)k
Z Df−n f (t) = lim f (t − kh) · hn ,
(5) h→0 k!
= f (x1 , x2 , x3 )e−i(ω.r) dx1 dx2 dx3 . k=0
R3 that corresponds to a n-th repeated summation [5].

390 Bull. Pol. Ac.: Tech. 66(4) 2018

 On fractional vectorial calculus

• We can apply the LT to (6) and (7) to obtain LT of the left derivative. In fact, if we apply the LT to (13)
  and (14), then we obtain
L Dfα f (t) = sα L [f (t)] , (10)
α
with Re(s) > 0 in the first and Re(s) < 0 in the second. L [Dlrx f (x)] = (±s)α L [f (x)] , (15)

Let N = ⌊α⌋ + 1. For functions with LT or FT, there are with Re(s) > 0, in the left derivative (+ sign), and Re(s) < 0,
integral formulations for the fractional derivatives, enjoying in the right (− sign) case.
the same set of properties.
The most general integral formulations of derivatives, in 2.4. Two-sided derivatives. The composition of derivatives
the sense of being valid for any functions defined in R, are of the same type (e.g. left) is a derivative of the same type.
introduced in the folowing definitions [44, 52]. If the composition is mixed, say a left with a right one, then
we obtain a two-sided derivative, that can be called centred
Definition 2.3 (forward regularised Liouville derivative). derivative.
Z∞ Two centred derivatives were introduced in [55–57]. Let
1
Dfα f (t) = us consider the composition of a left and a right derivatives.
Γ(−α) α β
0 As shown in [55, 56], the composition Dlx Drx f (x) leads to
(11) the GL centred (two-sided) fractional derivative:
" N −1
#
X (−1) fm (m)
(t)
· f (t − τ ) − u(α) τ m τ −α−1 dτ, +∞
X
0
m! α+β
Dcx f (x) := lim+ h−α−β (−1)n
h→0
n=−∞
where u(·) is the unit step function. (16)
Γ(α + β + 1)
Definition 2.4 (backward regularised Liouville derivative). · f (x − nh).
Γ(α − n + 1)Γ(β + n + 1)
Z∞
α e−iαπ In a general setup we would proceed as in [57] by in-
Db f (t) =
Γ(−α) troducing two parameters: γ = α + β defining the derivative
0
(12) order and θ = α − β sometimes called skewness. In this paper
" N
#
X −1 (m)
f (t) m −α−1 it will be called dissymmetry, since it determines the symme-
· f (t + τ ) − u(α) τ τ dτ. try of the binomial parameters in (16) [57]. Nonetheless, here
0
m!
we retained the order as α + β to enhance the two sources.
Other Liouville derivatives can be defined on R. One, sim- The above formula has the usual integer order derivatives
ply called “Liouville derivative”, is similar to the Riemann- as particular cases, but it introduces also others not existing
Liouville [1] and the other is the “Liouville-Caputo deriva- before as shown in the following examples.
tive” [44, 53, 54], similar to the Caputo derivative.
Examples. Some examples show the relation between the
2.3. On space derivatives. For derivatives in the time do- centred and the classic derivatives
main we consider two cases with the same LT and we inter-
preted them as causal and anti-causal according to the region • α = 1 and β = 0
of convergence.
When dealing with derivatives in the space domain there 1+0 f (x) − f (x − h)
Dcx f (x) = lim+ .
is no need to impose causality, since we can move in all direc- h→0 h
tions. Therefore, two fractional space derivatives called “left” This is one of the classical derivative definitions.
and “right”, are adopted. The left (in space) is defined by the • α = 0 and β = 1
same expression as for the forward derivative (in time) and
α f (x) − f (x + h)
denoted Dlx f (x). 0+1
Dcx f (x) = lim+ .
h→0 h
Definition 2.5 (left and right Grünwald-Letnikov derivatives).
X∞ Aside a factor (−1) this is another classical derivative.
α −α (−α)n Substituting −h for h we obtain the previous expression.
Dlx f (x) = lim h f (x − nh). (13)
h→0+
n=0
n! • α = 1/2 and β = 1/2
α
The right space derivative, Drx f (x), is defined as the +∞
X
backward, but removing the exponential factor – see (7) 1/2+1/2
Dcx = lim h−1 (−1)n
∞ h→0+
X (−α)n n=−∞
α
Drx f (x) = lim h−α f (x + nh), (14)
h→0+
n=0
n! 1
· f (x − nh).
Γ(3/2 − n)Γ(3/2 + n)
The lack of the exponential factor in the right deriva-
tive, when compared with the backward definition (7), has, as This represents a new derivative without having an inter-
consequence, that the corresponding LT is different from the pretation in classical terms.

Bull. Pol. Ac.: Tech. 66(4) 2018 391

 M.D. Ortigueira and J.A.T. Machado

• α = 1 and β = 1 Remark 2.1. All the results derived using the GL derivative
can be obtained with the integral formulations, due to their
1+1 2f (x) − f (x + h) − f (x − h) equivalence, at least for functions with LT [52].
Dcx f (x) = lim .
h→0+ h2
Remark 2.2. Partial derivatives are readily obtained from the
This derivative is the classical centred derivative of order 2.
Aside a factor (−1) coincides with the classic order two above definitions. Usually the symbol ∂x is adopted for denot-
ing the partial derivatives. Here, we will continue using Dx .
derivative
For example
2 f (x) − 2f (x − h) + f (x − 2h)
Dcx f (x) = lim+ . +∞
h→0 h2 X (−α)n
α
We can obtain other order 2 derivatives with other com- Dlx = lim+ h−α (−1)n f (x1 , x2 − nh, x3 ).
2
h→0
n=0
n!
binations. For example with α = 3/2 and β = 1/2 we
obtain For the other derivatives, the notations are similar.
+∞
X
3/2+1/2
Dcx = lim+ h−1 (−1)n
h→0
n=−∞
3. Fractional vectorial differential operators
1 3.1. The gradients, divergences, and curls. Let us consider
· f (x − nh). a scalar field f (x1 , x2 , x3 ). The usual integer order gradient is
Γ(5/2 − n)Γ(3/2 + n)
a vector with components corresponding to the partial deriva-
This represents also a new derivative. tives of the scalar field. The gradient points in the direction of
We must note that the LT of the centred derivatives does the largest rate of function increase and its magnitude is the
not exist. In fact, it would be given by (s)α (−s)β , but the slope of the graph in that direction. This important and useful
region of convergence is the empty set. However, the corre- tool has some limitations since it requires a smooth field and
sponding Fourier transform is given by [5, 55, 56] uses derivatives of order one.
π Here we extend its applications by considering not on-
lim (s)α (−s)β = |k|α+β ei 2 (α−β)·sgn(k) . ly fractional derivatives, but also (and more important) dif-
s→ik
We can write ferent orders. With this in mind, we are able to model non-
 α+β  α+β i π
homogenous and non-isotropic spaces. Essentially we can face
F Dcx f (x) = |k| e 2 (α−β)·sgn(k) F (ik), (17) complexity in different directions.
where sgn(·) is the signum function and F (ik) is the FT of For any α ∈ R, we define left gradient operator by
f (x).
Definition 3.1.
The special case where α = β
 2α  α1 α2 α3
F Dcx
2α
f (x) = |k| F (ik) (18) gradα α
l (·) = ∇l (·) := Dlx1 (·)e1 + Dlx2 (·)e2 + Dlx3 (·)e3 .
(21)
will be very important due to its relation with the FT of the 3
Similarly, for any α ∈ R , we define right gradient oper-
Laplacian. The properties of the two-sided derivatives are dis-
ator through
cussed in detail in [55, 56]. In the following we describe the
most important: Definition 3.2.

• Linearity gradβr (·) = ∇βr (·) := Drx

β1
1
β2
(·)e1 + Drx 2
β3
(·)e2 + Drx 3
(·)e3 .
• Additivity and commutativity (22)
If 2α + 2β > −1, [55, 56] 3
The gradients act on a scalar function defined on R and
2α
 2β  2β
 2α  2α+2β generate vectors having as components the partial derivatives
Dcx Dcx f (x) = Dcx Dcx f (x) = Dcx f (x) (19)
of the function: the nabla is a vectorial differential opera-
• Neutral and inverse elements tor. Their action over other vectors, namely vectorial func-
In particular, with 2α + 2β = 0, relations (19) show that, tions with components defined on R3 , originates two pairs of
for any centred derivative of order α, with |α| < 1/2, there differential vectors obtained using the inner or scalar prod-
is an anti-derivative, with order −α, [55, 56] and it can be uct and the cross or vectorial product. Let f (x1 , x2 , x3 ) be
obtained by using formula (18) a vectorial function. Calculating the inner product of the left
2α
 −2α  0 gradient and f we get the left divergence of f
Dcx Dcx f (x) = Dcx f (x) = f (x). (20)
α1 α2 α3
This implies that the derivative of order 2α = 1 does not divlα (f ) = ∇α
l · f := Dlx1 f1 + Dlx2 f2 + Dlx3 f3 . (23)
have inverse given by (16).
• Relation with the Riesz-Feller derivative With the right nabla we obtain the right divergence
In the 1-D case, the Riesz and the Riesz-Feller operators are (divrα (f )), similarly defined.
closely related with the centred derivative defined in (16). If, instead of the inner product, we use the cross product,
In particular such operators can be obtained for α = β, and then we derive the left and right curls. For the left curl we
for α − β = ±1, respectively ( [55, 56] and [1], page 214). have

392 Bull. Pol. Ac.: Tech. 66(4) 2018

 On fractional vectorial calculus
α1 α2 α3
Dlx 1
Dlx 2
Dlx 3
2. (a) (∇α α
l · ∇l ),
curllα (f ) = (∇α With the results presented in 3.1 we have
l × f ) := f1 f2 f3 . (24)
2α1 2α2 2α3
e1 e2 e3 (∇α α
l · ∇l ) = Dlx1 + Dlx2 + Dlx3

We can write (b) (∇α α

r · ∇r )

α2 α3
 This is similar to the previous case
(∇α
l × f ) = Dlx2 f3 − Dlx3 f2 e1
 α3 α1
 (∇α α 2α1 2α2 2α3
r · ∇r ) = Drx1 + Drx2 + Drx3 .
+ Dlx3 f1 − Dlx 1
f3 e2 (25)
 α1 α2
 Remark 3.3. The two last operators are scalar one-
+ Dlx1 f2 − Dlx2 f1 e3 . sided. They are similar to the fractional Laplacian
The corresponding right (curlrα (f )) is readily obtained. that we will define next.
The three pairs of operators defined in (21)–(24) will be (c) (∇α α
l · ∇r ),
called first generation operators [43], because they use only Attending to the results presented in Subsec. 3.1, we
the left or right derivatives, that is, not both. obtain easily:
Remark 3.1. With the LT such operators can be formulated α1 α2 α3
(∇α α α1 α2 α3
l · ∇r ) = Dlx1 Drx1 + Dlx2 Drx2 + Dlx3 Drx3
easily from a transform perspective. To exemplify, the LT of
and from (16) we derive that
the divergence is given in the LT domain by the inner product
of the vector sα and the transform of the vector field, α1 α1
Dlx Drx f (x1 , x2 , x3 ) = Dc2α1 f (x1 , x2 , x3 )
 α    1 1

L divlr (f ) = (±s)α · L [f ] . ∞
X n
(−1) Γ(2α1 +1)
= lim h−α (27)
Remark 3.2. The choice of the region of convergence h→0
n=−∞
Γ(α1 +n+1)Γ(α1 −n+1)
(Re(s) > 0 or Re(s) < 0) and the sign determines the type f (x − nh, y, z).
of divergence: left or right.
This expression is the centred fractional derivative [5]
Similarly, we obtain for the LT of the curl of order 2α1 relatively to the variable x1 . For x2 and
 α
  α

x3 the expressions are similar.
L curllr (f ) = (±s) × L [f ] ,
Definition 3.3. With (27) we can define the fractional Lapla-
under the conditions stated in Remark 3.2. cian with order α by
∆α := (∇α α 2α1 2α2 2α3
l · ∇r ) = Dcx1 + Dcx2 + Dcx3 . (28)
3.2. Mixed operators. The above operators can be combined
to get second generation operators in several distinct ways, but Remark 3.4. When α1 = α2 = α3 = 1 the centred derivative
in agreement with the previous ideas. The most important are equals the classic centred derivative of order 2, but we do not
those resulting from combinations of left-right operators that recover the classic Laplacian, because it uses a non centred
produce two-sided derivatives. derivative and as consequence, it yields a (−1) factor.
Attending to these considerations about the use of LT in
3.2.1. Internabla operations. Divergence and curl operators Remarks 3.1 and 3.2 and relations (17) and (18) we conclude
were defined by inner and outer products of the nabla (left or that, for this operator, the region of convergence degenerates
right) and a vectorial function. If we substitute this function into the imaginary axis. This means that the suitable trans-
by the left/right nabla, then we get the following operators form is the FT.
(a) (∇α α For the FT of the Laplacian we use (18) and the properties
1. l × ∇r )
Using (25) we can write of the two-sided derivatives [55, 56] to obtain
  2α
 α2 α3 α3
 F T ∆α f (x1 , x2 , x3 ) = kkk F (ik), (29)
(∇α α
l × ∇r ) = Dlx2 Drx3 − Dlx3 Drx2 e1
α2

 α3 α1 α1
 where
α3 2α
+ Dlx 3
Drx1 − Dlx 1
Drx 3
e2 kkk = |k1 |2α1 + |k2 |2α2 + |k3 |2α3 . (30)
 α1 α2 α2 α1

+ Dlx1 Drx2 − Dlx Drx 1
e3 When αi = 1/2, i = 1, . . . , 3, we obtain a new Laplacian of
2
order α = 1/2 that results from the application of derivatives
It is well known that in integer order calculus we of order 1/2.
have the important property: ∇1 × ∇1 ≡ 0. This is The above Laplacian is valid for orders verifying αi >
not valid with the operator (∇α α
l × ∇r ) that in general −1/2, due to the conditions of existence of the two-sided
is not null. derivatives, [55,56]. However, due to the relation between the
(b) (∇α α α α
l × ∇l ) and (∇r × ∇r ) centred derivative and the Riesz potential referred above, we
We have: can compute the partial derivatives in (28) for αi ≤ −1/2
with the 1-D Riesz potential, which enlarges the validity of
(∇α α α α
l × ∇l ) = (∇r × ∇r ) = 0 (26)
the definition to any orders. This means that the Laplacian
that will be useful in 3.3. introduced in (28) can be computed for any real order.

Bull. Pol. Ac.: Tech. 66(4) 2018 393

 M.D. Ortigueira and J.A.T. Machado

It is relevant to refer here that we can define a more gen- In the literature several authors have proposed expressions
eral Laplacian by using operators with different orders: for the Riesz Laplacian by looking for an operator satisfying
the generalization of (32), that is, by verifying:
∆α,β = (∇α β α1 +β1
l · ∇r ) = Dcx1
α2 +β2
+ Dcx 2
α3 +β3
+ Dcx 3
F ((−∆)α f )(k) = |k|2α F (f )(k), (36)
with FT given by
h i with 0 < α ≤ 1 and
F T ∆α,β f (x1 , x2 , x3 ) = (k)α,β F (ik),  α
X3

where |k|2α = kj2  . (37)

j=1
3
X π
(k)α,β = |km |αm +βm ei 2 (αm −βm )·sgn(km ) . This operator can be implemented by [1] means of
1 Z
1 (∆ly f )(x)
(−∆)α f (x) := dy, (38)
3.2.2. On the Laplacian. The Laplacian is an important op- γ3 (α) |x − y|2α+3
R3
erator in physics and engineering. However, the standard de-
finition does not cope with domains that are neither homo- where (∆ly f )(x) is the l-th difference of f (x). For l > 2α the
geneous, nor isotropic. Researchers in applied sciences have above integral is absolutely convergent. Often we have α < 1
been interested in a definition of fractional Laplacian suitable and l is chosen as l = 1 yielding [59, 60]
for describing such type of media. In recent studies the frac- Z
1 f (y) − f (x)
tional Laplacian was implemented by means of the inverse of (−∆)α f (x) := dy. (39)
γ3 (α) |x − y|2α+3
the Riesz potential [58–60]. However, it is not straightforward R3
that this is the suitable option when thinking in applications
Operator (39) has been used to implement the fractional
where the properties change with direction. In fact, the inverse
Laplacian [59, 60]. However, the Riesz based Laplacian does
of the Riesz potential cannot be expressed as a sum of partial
not fit into our proposal pointing to the generalization of clas-
derivatives and is only defined using the inverse of Fourier
sic vectorial operators.
transform of the operator [61].
The classical Laplacian is defined in 3-D by
3.3. Vectorial Laplacian and mixed operators. Starting
∂ 2 f (x) ∂ 2 f (x) ∂ 2 f (x) from the Laplacian formulation, (28), we can define the vec-
∆f (x) = + + , (31)
∂x21 ∂x22 ∂x23 torial Laplacian
where x = (x1 , x2 , x3 ). This operator is fundamental in lapα f = ∆α f1 e1 + ∆α f2 e2 + ∆α f3 e3 (40)
physics, namely in electromagnetism. Applying the FT to
that enjoys a very important relation shared with two pairs of
equation (31), we obtain
mixed operators
F (−∆f )(k) = |k|2 fb(k) (32) 1. Gradient of a divergence
with We define gradient of a divergence as
3
X
|k|2 = kj2 . (33) gradα α α α
r (divl f ) := ∇r (divl f )
j=1 2α1 2α2 2α3
= Dcx f e + Dcx
1 1 1
f e + Dcx
2 2 2
f e
3 3 3
α1 α2 α2 α3 α3 α1 (41)
Remark 3.5. If instead of the usual right derivative used to +Drx1 Dlx2 f2 e1 + Drx2 Dlx3 f3 e2 + Drx3 Dlx 1
f1 e3
define the Laplacian, the centred one is adopted, then the sign α1 α3 α2 α1 α3 α2
+Drx Dlx f3 e1 + Drx Dlx f1 e2 + Drx Dlx f2 e3 .
− can be removed. 1 3 2 1 3 2

Remark 3.6. We obtain gradα α

l (divr f ) with the substitu-
In his work [58], Riesz implicitly suggested several pos-
tions l for r and r for l.
sible definitions of fractional Laplacian [3,58], but he did not
present the explicit realization of such operators. Nonethe- Remark 3.7. We can also define gradα α
l (divl f ) and
less, Riesz deduced expressions suggesting that the potential gradαr (divr
α
f ), but they are one-sided operators of limited
defined by interest.
Z 2. Divergence of a curl
1
(Iα
R f )(x) := f (y)|x − y|α−3 dy, (34) We can define several divergence of a curl operators as
γ3 (α)
R3 before. The most interesting are the following
for any positive α, could implement the inverse of the integer divlα (curllα (f )) = divlα (∇α α α
l × f ) = ∇l · ∇l × f = 0, (42)
order Laplacian. The normalizing constant γ3 (α) is given by divrα (curlrα (f )) = divrα (∇α α α
r × f ) = ∇r · ∇r × f = 0, (43)
3
π 2 22α Γ (α) π 2 Γ (2α) because they are identically null as in the usual integer
γ3 (α) =
=2

. (35)
Γ 3−2α2 Γ 3−2α
2 Γ 21 + α order case.

394 Bull. Pol. Ac.: Tech. 66(4) 2018

 On fractional vectorial calculus

3. Curl of a curl 2. For the above pair of fields there are four potentials, two
We start defining curl of a curl by curlr (curll (f )). To ob- scalar, φ1 and φ2 , and two vectorial, A1 and A2 , that are
tain a closed form for this operator we use the vectorial able to represent those fields according to
Laplacian to get:
∂ γ A1 1
curlrα (curllα (f )) = gradα α α F1 = −gradα
r φ1 − + curllα A2 , (50)
l (divr (f )) − lap f (44) ∂tγ a1
in agreement with the classical result. Obviously we can ∂ γ A2 1
have alternative order for the symbols “l” and “r” obtaining F2 = −gradα
l φ2 − + curlrα A1 . (51)
∂tγ a2
a second valid relation:
curllα (curlrα (f )) = gradα α α Remark 3.10. In this generalization of the Helmholtz decom-
r (divl (f )) − lap f . (45)
position, the gradient and the rotational have different char-
Remark 3.8. It is possible to generalize and to combine op- acter: one is left and the other right-sided.
erators with any triplets of orders: α = (α1 , α2 , α3 ) 6= β =
(β1 , β2 , β3 ) as we did before for the Laplacian. Remark 3.11. The scalars r1 and r2 and the vectors j1 and
j1 have the role of inputs to the system defined by (46) to (49).
Remark 3.9. In the above expressions it was assumed that
the orders of the operators are fixed. Nevertheless, we can Remark 3.12. The two fields are interlaced. The solution of
consider changing the orders of derivation. As concerns the (46) to (49) for obtaining the fields consists of decoupling that
derivation in time, we only have to perform another deriva- relationship.
tive using a suitable order, since the proposed time derivative
Remark 3.13. Another manifestation of that relationship can
verifies the additivity of the orders that can assume any real
be obtained with the divergence operators in (48) and (49)
value. Concerning the space derivatives, the problem may not
using (42) and (43) to get
have a such simple solution. It was conjecturated [43] that
we can substitute a given field, F, by gradβr divlα F, but that ∂ γ r2
conjecture remains to be proved. divrα j1 = a2 , (52)
∂tγ
∂ γ r1
3.4. Fractional generalized Helmholtz decomposition the- divlα j2 = a1 . (53)
orem. The Helmholtz decomposition theorem is a classical ∂tγ
result that allows the decomposition a given time independent Remark 3.14. The scalar and vectorial potentials serve as in-
field into curl-free and divergence-free components [62, 63]. termmediate tools to obtain the fields (outputs of the system).
This theorem was extended by Kapuścik for the time vari-
ant case and simultaneously deduced some gauge invariants
of classical field theories [64]. This development was used by 3.4.1. Solution in terms of potentials. To justify this theo-
Nevels [65] to present a derivation of Maxwell equations. Fol- rem it was shown that the pair of fields can be determined
lowing Kapuścik’s procedure, similar conclusions for a frac- from the four potentials [43]. As referred above we consid-
tional setup were obtained in [43]. Essentially the procedure er r1 , r2 , j1 , and j2 as inputs to the system and F1 and F2
consists in the decomposition of two fields in terms of time- the outputs. The scalar and vectorial potentials are auxilliary
dependent scalar and vectorial potentials. The decomposition functions. Therefore, we need relations involving such poten-
involves the vectorial tools presented in the previous sections tials serving as intermmediate tools to compute the fields from
and the forward fractional time derivative (to impose causal- the input functions.
ity). Such relations are obtained by applying the divergence
and rotational operators for both fields and manipulating the
Theorem 3.1. Helmholtz decomposition theorem
equations with the help of the vectorial operators introduced
1. Consider a pair of time-dependent vector fields F1 and in Sec. 3. As shown in [43] the four potentials are completely
F2 that are null at an infinitely long distance. Such pair is determined from four equations and once they are known we
uniquely determined by obtain the fields F1 and F2 . Such equations assume the form
r1 = divlα F1 , (46)  
α α ∂ γ φ2
a1 j1 = gradl divr A2 − a1 a2 γ
r2 = divrα F2 , (47) ∂t
  (54)
and α ∂ 2γ A2
− lap A2 + a1 a2 ,
∂ γ F2 ∂t2γ
j1 = curlrα F1 + a2 , (48)
∂tγ  
γ ∂ γ φ1
∂ F1 a2 j2 = gradα α
divl A1 − a1 a2 γ
j2 = curllα F2 + a1 , (49) r
∂t
∂tγ (55)
 
where a1 and a2 are two constants with physical meaning ∂ 2γ A1
used to match the physical dimensions. − lapα A1 + a1 a2 ,
∂t2γ

Bull. Pol. Ac.: Tech. 66(4) 2018 395

 M.D. Ortigueira and J.A.T. Machado
 
∂γ α ∂ γ φ1 3.4.2. Decoupled solution involving the fields. We can
r1 = − divl A1 − a 1 a 2
∂tγ ∂tγ avoid passing by intermmediate potentials if we use wave
  (56) equations for the fields (see [43]). Such equations are
α ∂ 2γ φ1
− ∆ φ1 + a1 a2 2γ , ∂ 2γ F1
∂t lapα F1 + a1 a2 = −curllα j1
γ
  ∂t2γ
∂ ∂ γ φ2 (66)
α
r2 = − γ divr A2 − a1 a2 γ ∂ γ j2
∂t ∂t + gradα
r r1 + a2 ,
(57) ∂tγ
 
∂ 2γ φ2 ∂ 2γ F2
− ∆α φ2 + a1 a2 2γ . lapα F2 + a1 a2 = −curlrα j2
∂t ∂t2γ (67)
This formulation suggests that for each pair (φi , Ai ), ∂ γ j1
+ gradα
l r2 + a1 .
i = 1, 2, we expect to have: ∂tγ
1. Remark 3.15. The above theory only makes sense if it is
∂ γ φ1 a true generalization of the classic integer order counterpart.
divlα A1 − a1 a2 γ = 0, (58) Therefore, when the orders in the gradients, divergences, and
∂t
curls become 1 we have to obtain the classic Helmholtz de-
∂ γ φ2 composition theorem. From the derivative definitions given in
divrα A2 − a1 a2 = 0, (59)
∂tγ Sec. 2 we conclude that the fractional left and right gradi-
allowing us to obtain 4 decoupled equations. ents, divergences, and curls recover the classic corresponding
2. Let u1 and u2 be two scalar potentials, and B1 and B2 operators, since there is no difference between integer order
be two vectorial potentials verifying (58), (59), and (61). left and right derivatives. On the other hand, these operators
Then, the fields (50) and (51) remain invariant under the are continuous functions of the orders. The unique difference
transformations: lies in the Laplacian definition. The one proposed here leads
∂ γ u1 1 α to an expression that has a “−” sign when compared to the
φ1 −→ φ1 − ∂tγ − a1 a2 divl B1
classic one, due to the use of the two-sided derivative.
∂ γ u2 1 α
φ2 −→ φ2 − ∂tγ − a1 a2 divr B2
Remark 3.16. There are several approaches for obtaining
(60) the fractional Maxwell equations [22, 39, 40, 66], but without
∂ γ B1 1 the coherence given by the previous formalism. In [43] a ver-
A1 −→ A1 + gradα
r u1 − ∂tγ − α
a1 curll B2 sion of fractional Maxwell equations based on the generalized
Helmholtz decomposition was proposed.
∂ γ B2 1
A2 −→ A2 + gradα
l u2 − ∂tγ − α
a2 curlr B1
3. Any other scalar or vector potential used to obtain the in- 4. On the fractional definite integrals
variants must give a null contribution to the fields, meaning
4.1. Previous considerations. The concept of fractional def-
that they must verify the condition:
inite integrals (FDI) was introduced in [48]. Starting with
∂ 2γ ψ a generalization of Barrow formula an FDI definition was
∆α ψ + a1 a2 = 0. (61)
∂t2γ proposed, followed by the formulation of the “fractional fun-
According to the above relations (54) to (61) we conclude damental theorem of calculus”. These developments allowed
that we can always choose the scalar and vectorial potentials the definition of double and triple integrals on rectangular
verifying (61). This together with (60) implies that the con- spaces. Here we will recall those results in the scope of the
ditions (58) and (59) are also satisfied leading to simplified present work.
decoupled equations for the potentials
 
α ∂ 2γ A2 4.2. A generalized Barrow formula. Let f (x), x ∈ R, such
a1 j1 = − lap A2 + a1 a2 , (62) (α)
∂t2γ that flr (x), α > 0, exist – it is enough to assume that f (x)
  has LT with a non void region of convergence. We denote the
∂ 2γ A1 (−α)
two cases of left and right anti-derivatives by flr (x).
a2 j2 = − lapα A1 + a1 a2 , (63)
∂t2γ
  Definition 4.1. The α-order fractional integral (FI) of f (x)
α ∂ 2γ φ1 over the interval (a, b) is defined through the fractional Bar-
r1 = − ∆ φ1 + a1 a2 2γ , (64)
∂t row formula
  Zb
∂ 2γ φ2
r2 = − ∆α φ2 + a1 a2 2γ , (65) α
Ilr f (a, b)
(−α) (−α)
= − f (x)dxα = flr (b) − flr (a), (68)
∂t
a
establishing a set of 8 independent equations involving the
Laplacian. Once the potentials are computed, the fields F1 where the dashed integral symbol represents the FI. Using the
and F2 are readily obtained using (50) and (51). Liouville anti-derivative expression we can write

396 Bull. Pol. Ac.: Tech. 66(4) 2018

 On fractional vectorial calculus
Z∞ where each integration is obtained by the fractional Barrow
α 1
Ilr f (a, b) = [f (b ± x) − f (a ± x)] dxα , (69) formula.
Γ(α + 1)
0
If the orders of the FD are equal (i.e., α1 = α2 = α),
Dxα α−1
since α = x
, as x ≥ 0. From the standard (integer then we can consider the standard expression for surface,
Zb x1 x2 = S, to get the fractional surface integral
Z
order) Barrow formula f ′ (x)dx = f (b) − f (a) we obtain
Ilr g(a1 , b1 , a2 , b2 ) = − g(x1 , x2 )dS α .
α
(77)
a
the expression S

Z∞ Zb The integral in R3 is obtained in a similar way.

α 1
Ilr f (a, b) = f ′ (y ± x)dydxα . (70) Definition 4.3. The FI in R3 is defined by
Γ(α + 1)
0 a α
Ilr g(a1 , b1 , a2 , b2 , a3 , b3 )
Since the function f (x) has LT, the integral in (69) exists.
The same happens with inner integral in (70) that is uniformly Zb1Zb2Zb3 (78)
convergent. Consequently, we can commute the integrations = − − − g(x1 , x2 , x3 )dxα α2 α3
1 dx2 dx3 ,
1

to obtain a1 a2 a3
Zb
α (−α+1) where each integration is performed by means of the fraction-
Ilr f (a, b) = flr (x)dx. (71) al Barrow formula.
a
If the derivative orders are equal (i.e., α1 = α2 =
Considering as variable the upper limit, b = z ∈ R, (71) α3 = α), then we get
leads to a fractional formulation of the fundamental theorem Z
α
of integral calculus. In fact, setting f (x) = Dlr g(x) we can α
Ilr g(a1 , b1 , a2 , b2 , a3 , b3 ) = − g(x1 , x2 , x3 )dV α , (79)
write [48]
V
Zz
α α (1) where we considered the standard expression for volume,
Ilr Dlr g(a, z) = Dlr g(x)dx = g(z) − g(a), (72) V = x1 x2 x3 .
a

and 5. Fractional classic integral theorems

h i
α
Dlr α
[Ilr f (a, z)] = α
Dlr
(−α) (−α)
flr (z) − flr (a) = f (z), (73) in rectangular domains
In this section we generalize the classic integral theorems of
since, for the adopted formulations of derivatives, the deriva- Green, Stokes, and Ostrogradski-Gauss by following a pro-
tive of a constant is zero. cedure similar to the one adopted in [46]. Furthermore, we
will use the left derivative unless another option is explicitly
4.3. Integrals in R2 and R3 . Let us assume that the func- mentioned.
tion g is dependent of a variable, x1 , and a parameter, x2 , that
is kept fixed, f (x1 , x2 ). We define the parametric integral 5.1. The Green’s theorem. Let us consider a closed
Zb1 path γ defining a rectangle in the horizontal plane
Ilr g(a1 , b1 , x2 ) = − g(x1 , x2 )dxα
α 1 Σ = {(x1 , x2 ) : a1 ≤ b1 , a2 ≤ b2 } and a vectorial function
1
(74) f (x1 , x2 ) = f1 (x1 , x2 )~e1 + f2 (x1 , x2 )~e2 , where ~ei , i = 1, 2,
a1
define the usual orthogonal base in R2 . Let αi , i = 1, 2, be
(−α1 ) (−α1 )
= glr (b1 , x2 ) − glr (a1 , x2 ). the order of the derivative with respect to the variable xi . The
Similarly, fixing x1 and having x2 for variable, we define the line FI along γ is defined by:
parametric integral Zb1 Zb2
Zb2
α
Ilr f (γ) = − f1 (x1 , a2 )dx1 + − f2 (b1 , x2 )dxα
α1
2
2

α
Ilr g(x1 , a2 , b2 ) = − g(x1 , x2 )dxα
2
2 a1 a2
(75) Za1 Za2
a2
(−α ) (−α2 ) + − f1 (x1 , b2 )dxα
1
1
+ − f2 (a1 , x2 )dxα
2
2

= glr 2 (x1 , b2 ) − glr (x1 , a2 ).

b1 b2
This motivates to the following definition. Zb1
Definition 4.2. The FI on a rectangular region (a1 , b1 ) × = − [f1 (x1 , a2 ) − f1 (x1 , b2 )] dxα
1
1

(a2 , b2 ) is given by a1
Zb1Zb2 Zb2
Ilr g(a1 , b1 , a2 , b2 ) = − − g(x1 , x2 )dxα
α α2
1 dx2 ,
1
(76) + − [f2 (b1 , x2 ) − f2 (a1 , x2 )] dxα
2 .
2

a1 a2 a2

Bull. Pol. Ac.: Tech. 66(4) 2018 397

 M.D. Ortigueira and J.A.T. Machado

Using the fractional formulation of the fundamental theorem

of integral calculus (72) and the Fubini theorem [67], we can
write
Zb1Zb2
α
Ilr f (γ) =−−
a1 a2
 
· −Dxα22 f1 (x1 , x2 ) + Dxα11 f2 (x1 , x2 ) dxα α2
1 dx2 .
1

Z
In this line of thought, introducing the notation − f (γ)dγ α =
γ
α
Ilr f (γ), we can formulate the following theorem.
Theorem 5.1. The fractional Green’s theorem for a rectangu-
lar domain Σ is given by
Z Zb1Zb2 Fig. 1. Example of a rectangular domain in R2 . The boundary is
α a rectangular line γ
− f (γ)dγ = − −
γ (80)
a1 a2
 
· Dxα11 f2 (x1 , x2 ) − Dxα22 f1 (x1 , x2 ) dxα α2
1 dx2 .
1
5.2. The Stokes’ theorem. In this case we adopt a strategy
similar to the one followed for the Green’s theorem, with the
This theorem is valid for a region formed by juxtaposing
difference that now the integration path is a line in R3 with
a finite number of small rectangles. In this case the integra-
6 segments parallel to the coordinate axis.
tion paths is a succession of horizontal and vertical segments.
Let us consider a surface constituted by 3 contigu-
We can write dγ α = cdxα α2
1 + (1 − c)dx2 , with c = 1 and
1

ous faces of a parallelepiped Ψ = {(x1 , x2 , x3 ) : a1 ≤

c = 0 for horizontal and vertical segments, respectively.
b1 , a2 ≤ b2 , a3 ≤ b3 } and a vectorial function f (x1 , x2 , x3 ) =
Consider two rectangles placed side by side, Σ1 = f1 (x1 , x2 , x3 )~e1 +f2 (x1 , x2 , x3 )~e2 +f3 (x1 , x2 , x3 )~e3 , and αi ,
{(x1 , x2 ) : a1 ≤ b1 , a2 ≤ b2 }, and Σ2 = {(x1 , x2 ) : b1 ≤ i = 1, 2, 3, the derivative orders relatively to variables xi .
2b1 − a1 , a2 ≤ b2 }. The corresponding surrounding lines are The 3 faces define a path, γ, with 6 segments. To compute
γ1 and γ2 . Let Σ be the union of the two rectangles and γ the fractional line integral along γ we apply to each face the
the surrouding line. We have: Green’s theorem above introduced. The direction of the walk
Z Z must be the same in each face. This leads to a cancellation of
− f (γ)dγ α + − f (γ)dγ α the contributions in edges belonging to two contiguous faces.
γ1 We can write
γ2

Zb1 Zb2 • Upper horizontal face

=− f1 (x1 , a2 )dxα
1
1
+ − f2 (b1 , x2 )dxα
2
2

a1 a2
Z Zb1Zb2

− f (γ)dγ = − − Dxα11 f2 (x1 , x2 , b3 )
α
Za1 Za2 Zbb1 (81)
γ3 a1 a2
α1 α2
+− f1 (x1 , b2 )dx1 + − f2 (a1 , x2 )dx2 + − f1 (x1 , a2 )dxα
1
1


b1 b2 b1 − Dxα22 f1 (x1 , x2 , b3 ) dxα1 α2
1 dx2 .
Zb2 Zb1 Za2
• Close frontal face
+− f2 (b1 , x2 )dx2 + − f1 (x1 , b2 )dx1 + − f2 (b1 , x2 )dxα
α2 α1
2
2

a2 bb1 Z Zb2Zb3
b2 
− f (γ)dγ = − − Dxα22 f3 (b1 , x2 , x3 )
α
Zbb1 Za2 γ1 (82)
a2 a3
= − f1 (x1 , a2 )dx1 + − f2 (2b1 − a1 , x2 )dxα
α1
2
2

a1 b2 − Dxα33 f2 (b1 , x2 , x3 ) dxα2 α3
2 dx3 .
Zb1 Zb2 Z • Right hand face
+− f1 (x1 , b2 )dxα
1
1
+ − f 2 (a 1 , x2 )dxα2
2 = − f (γ)dγ α .
bb1
Z Zb1Zb3
a2 γ 
− f (γ)dγ = − − Dxα11 f3 (x1 , b2 , x3 )
α
γ2 (83)
We set 2b1 − a1 = bb1 . This procedure can be used with re- a1 a3
gions bounded by integration paths similar to that illustrated 
in Fig. 1. −Dxα33 f1 (x1 , b2 , x3 ) dxα α3
1 dx3 .
1

398 Bull. Pol. Ac.: Tech. 66(4) 2018

 On fractional vectorial calculus

Attending to the definition of the fractional curl operator in- According to the fractional fundamental theorem of inte-
troduced in Sec. 3, we have gral calculus, summing the flux contributions of two parallel
 α2 α3
 faces, we get
(∇α l × f ) = Dlx2 f3 − Dlx3 f2 e1
 α3 α1
  α1 α2
 (84) Zb1Zb2
+ Dlx f1 − Dlx f3 e2 + Dlx f2 − Dlx f1 e3
3 1 1 2
− − [f3 (x1 , x2 , b3 ) − f3 (x1 , x2 , a3 )] dxα α2
1 dx2
1

we can rewrite the relations (81)–(83) in the form a1 a2

• Upper horizontal face (x3 = b3 ) Zb1Zb2Zb3

= − − − Dxα33 f3 (x1 , x2 , x3 )dxα α2 α3
1 dx2 dx3 .
1
Z Zb1Zb2
a1 a2 a3
− f (γ)dγ = − − ∇α
α
e3 dxα
l ×f ·~
α2
1 dx2 .
1
(85)
γ3 a1 a2 We can sum of the six contributions and represent sym-
bolically the result by
• Close frontal face (x1 = b1 ) I
Z Zb2Zb3 (f (x1 , x2 , x3 ) · ~n)dS α
− f (γ)dγ = − − ∇α
α
e1 dxα
l ×f ·~
α3
2 dx3 .
2
(86) S

γ1 a2 a3 Zb2Zb3
= − − Dxα11 f1 (b1 , x2 , x3 )dxα α3
2 dx3
2
• Right hand face (x2 = b2 )
a2 a3
Z Zb1Zb3 Zb1Zb3
− f (γ)dγ = − − ∇α
α
e2 dxα
l ×f ·~
α3
1 dx3 .
1
(87) +− − Dxα22 f2 (x1 , b2 , x3 )dxα α3
1 dx3
1

γ2 a1 a3
a1 a3
Adding the 3 terms we obtain the following theorem. Zb1Zb2
Theorem 5.2. The fractional Stokes’s theorem for a paral- +− − Dxα33 f3 (x1 , x2 , b3 )dxα α2
1 dx2 ,
1

lelepipedic domain, Ψ, becomes a1 a2

Z b
3 Z j Zbk
where j, k = 1, 2, 3, j 6= k 6= i and xi = bi . The normal ~n is
X α equal to one of the base vectors, ~ei , i = 1, 2, 3, according to
α
− f (γ)dγ = − − ∇α ei dxj j dxα
l ×f ·~ k ,
k
(88)
the integration surface.
γ i=1a a
j k
With this result we can state the following theorem.
where j, k = 1, 2, 3, j < k, j, k 6= i, and, for each i = 1, 2, 3, Theorem 5.3. Let the left divergence operator be defined in
the variable xi is kept constant equal to bi . (23). Then the fractional Ostrogradski-Gauss or Divergence
As in the classical case, we define the flux, Φ3 , through a theorem can be written as
I
horizontal face by
(f (x1 , x2 , x3 ) · ~n) dS α
Zb1Zb2 S
Φ3 = − − f (x1 , x2 , x3 ) · ~e3 dxα1 α2
1 dx2 Zb1Zb2Zb3 (89)
a1 a2 = − − − divlα (f ) dxα α2 α3
1 dx2 dx3 .
1

Zb1Zb2 a1 a2 a3

= ±− − f3 (x1 , x2 , x3 )dxα α2
1 dx2 ,
1
In particular, f (x1 , x2 , x3 ) can be set equal to the gradient
a1 a2 of a scalar function: f (x1 , x2 , x3 ) = gradα r (g) leading to

where the − or + signs apply when x3 = a3 or x3 = b3 . The Corollary 5.1.

fluxes Φ1 and Φ2 , through frontal and lateral faces are defined I


in a similar way. Therefore, the factional Stokes’ states that gradα n dS α
r (g) · ~
the fractional integral of a vectorial field along a rectangular S
line is equal to the total flux of the curl of such field through Zb1Zb2Zb3 (90)
a surface supported on such line.
= − − − ∆α (g) dxα α2 α3
1 dx2 dx3 .
1

a1 a2 a3
5.3. The Ostrogradski-Gauss’s theorem. Let us consider
a parallelepipedic region Ψ previously defined and a vectorial Remark 5.1. As referred previously, the results deduced for
function f (x1 , x2 , x3 ) = f1 (x1 , x2 , x3 )~e1 +f2 (x1 , x2 , x3 )~e2 + a parallelepiped can be generalized to the region that results
f3 (x1 , x2 , x3 )~e3 . Since each face is parallel to a coordi- from joining as many parallelepipeds as needed in such a way
nate plane, its normal is parallel to one of the unity vectors that all the small surfaces are parallel to one coordinated
~ei , i = 1, 2, 3. plane.

Bull. Pol. Ac.: Tech. 66(4) 2018 399

 M.D. Ortigueira and J.A.T. Machado

5.4. Uniform derivative order cases. In the previous sec- Considering rectangular domains the Green’s, Stoke’s, and
tions we assumed different orders for the three space direc- Ostrogradski-Gauss’s theorems were presented. A discussion
tions. Herein, we derive formulae for the particular case of of the particular case having uniform fractional orders was
uniform order, that is, for α1 = α2 = α3 = α. also included.
In this case the FI on a rectangular region constituted by
contiguous rectangles is given by (77) and is reproduced here Acknowledgement. This work was partially funded by
Z National Funds through the Foundation for Science and
Ilr f (a1 , b1 , a2 , b2 ) = − f (x1 , x2 )dS α .
α
(91) Technology of Portugal, under the project PEst-OE/EEI/
S UI0066/2013.
3
The integral in R was obtained in a similar way and the FI REFERENCES
in R3 was defined in (79) and reads
Z [1] S.G. Samko, A.A. Kilbas, and O.I. Marichev, Fractional Inte-
α
Ilr f (a1 , b1 , a2 , b2 , a3 , b3 ) = − f (x1 , x2 , x3 )dV α . (92) grals and Derivatives. Theory and Applications, Gordon and
Breach Sc. Publ., London, 1993.
V [2] S. Dugowson, Les Différentielles Métaphysiques, Ph.D. Thesis,
A particular case of the Green’s theorem occours when Université Paris Nord, Villetaneuse, France, 1994, [In French].
α1 = α2 = α and can be written as [3] A.A. Kilbas, H.M. Srivastava, and J.J. Trujillo, Theory and Ap-
Z Z plications of Fractional Differential Equations, Elsevier, Am-
 
− f (γ)dγ α = − Dxα1 f2 (x1 , x2 )−Dxα2 f1 (x1 , x2 ) dS α . (93) sterdam, 2006.
[4] R.L. Magin, Fractional Calculus in Bioengineering, Begell
γ S
House, Connecticut, 2006.
Similarly, for Stokes’ theorem 5.2, if αi = α, i = 1, 2, 3, [5] M.D. Ortigueira, Fractional calculus for scientists and engi-
then we can write from (88) neers, Lecture Notes in Electrical Engineering, 84, Springer,
Z 3 Z
Dordrecht (2011).
X [6] D. Valério, J.J. Trujillo, M. Rivero, J.A.T. Machado, and D.
− f (γ)dγ α = − ∇α × f · ~ei dSiα , (94)
Baleanu, “Fractional calculus: A survey of useful formulas”,
γ i=1S
i Eur. Phys. J. Special Topics 222, 1827–1846, EDP Sciences,
where Si , i = 1, 2, 3, is a surface with normal given by ~ei , Springer-Verlag (2013).
[7] K.S. Cole and R.H. Cole, “Dispersion and absorption in di-
i = 1, 2, 3.
electrics – I alternating current characteristics”, J. Chem. Phys.
Concerning the Ostrogradski-Gauss’s theorem 5.3, in the 9, 341–352 (1941).
case of uniform orders we can write, from (89) [8] K.S. Cole and R.H. Cole, “Dispersion and absorption in di-
I Z
electrics – II direct current characteristics”, J. Chem. Phys. 10,
f (x1 , x2 , x3 ) · ~n dS = − div α (f ) dV α .
α
(95) 98–105 (1942).
S V [9] S. Westerlund and L. Ekstam, “Capacitor theory”, IEEE Trans.
Dielectr. Electr. Insul. 1(5), 826–839 (1994).
With α = 1, we recover the classical results.
[10] B.E. Conway, Electrochemical Supercapacitors: Scientific
Principles and Technological Application, Plenum, New York,
6. Conclusions (1999).
In this paper, a generalization of classical fractional vectori- [11] I. Schäfer and K. Krüger, “Modelling of coils using fraction-
al calculus was proposed, by considering the differential and al derivatives”, Journal of Magnetism and Magnetic Materials
307, 91–98 (2006).
integral formulations. Concerning the first we presented frac-
[12] K. Biswas, S. Sen, and P. Dutta, “Realization of a constant
tional versions of the vectorial tools, namely, the gradient,
phase element and its performance study in a differentiator
divergence, curl, and Laplacian. For this purpose, suitable circuit”, IEEE Trans. Circuits Syst. II, Exp. Briefs 53(9), 802–
formulations of fractional derivatives were described. Time 806 (2006).
and space derivatives were introduced using the Grünwald- [13] D. Baleanu, A.K. Golmankhaneh, and M.C. Baleanu, “Frac-
Letnikov and Liouville formulations. Two-sided derivatives tional electromagnetic equations using fractional forms”, Int
were also presented due to their importance in the develop- J. Theor. Phys. 48, 3114–3123 (2009).
ment of differential vectorial tools, namely the Laplacian. We [14] I.S. Jesus and J.T, Machado, “Development of fractional order
described left and right gradient, divergence, and curl, calling capacitors based on electrolyte processes”, Nonlinear Dynam-
them “first generation operators”, since they led to the “sec- ics 56, 45–55 (2009).
ond generation operators” and, in particular, to the fractional [15] I.S. Jesus and J.T. Machado, “Comparing integer and fraction-
al models in some electrical systems”, Proceedings of FDA’10.
Laplacian. This formulation was compared with the fraction-
The 4th IFAC Workshop Fractional Differentiation and its Ap-
al Laplacian based on the Riesz potential. The generalized
plications, Badajoz, Spain, October 18–20 (2010).
Helmholtz decomposition was recalled and its solution ex- [16] V.E. Tarasov, Fractional Dynamics: Applications of Fractional
pressed in terms of the solutions of several fractional wave Calculus to Dynamics of Particles, Fields and Media, Higher
equations. Before going into the integral tools the fractional Education Press and Springer, Beijing (2010).
definite integrals were defined. They were adopted to gen- [17] C.S. Drapaca and S. Sivaloganathan, “A fractional model of
eralize the classic integral theorems to the fractional case. continuum mechanics”, J. Elast. 107, 105–123 (2012).

400 Bull. Pol. Ac.: Tech. 66(4) 2018

 On fractional vectorial calculus

[18] M. D’Ovidio and R. Garra, Fractional Gradient and its [35] M. Ostoja-Starzewski, “Continuum mechanics models of frac-
application to the Fractional Advection Equation, arX- tal porous media: Integral relations and extremum principles”,
iv:1305.4400v1 math-ph, May 19, (2013). Journal of Mechanics of Materials and Structures 4(5), 901–
[19] A.S. Ali, A.G. Radwan, and A.M. Soliman, “Fractional order 912 (2009).
Butterworth filter: Active and passive realizations”, IEEE Jour- [36] G. Calcagni, “Geometry of fractional spaces”, Adv. Theor.
nal on Emerging and Selected Topics in Circuits and Systems Math. Phys. 16, 549–644 (2012).
3(3), (2013). [37] F. Mainardi, Fractional Calculus and Waves in Linear Vis-
[20] S.I.R. Arias, D.R. Muñoz, J.S. Moreno, S. Cardoso, R. Fer- coelasticity: An Introduction to Mathematical Models, Imperial
reira, and P.J. Freitas, “Fractional modeling of the AC large- College Press, (2010).
signal frequency response in magnetoresistive current sensors”, [38] M.P. Velasco and L. Vázquez, “On the fractional Newton and
Sensors 13, 17516–17533 (2013). wave equation in one space dimension”, Appl. Math. Modelling
[21] A.E. Çalik, H. Ertik, B. Öder, and H. Şirin, “A fractional cal- 38(13), 3314–3324 (2014).
culus approach to investigate the alpha decay processes”, Inter- [39] V.E. Tarasov, “Fractional generalization of gradient and Hamil-
national Journal of Modern Physics E 22(7), 1350049 (2013). tonian systems”, J. Phys. A: Math. Gen. 38, 5929–5943
[22] F.A.A. El-Salam, “Fractional approach of Maxwell equations (2005).
in the curved spacetime”, Journal of Taibah University for Sci- [40] V.E. Tarasov, “Review of some promising fractional physical
ence 7, 173–179 (2013). models”, International Journal of Modern Physics B 27(9),
[23] V. Martyniuk and M.D. Ortigueira, “Fractional model of an 1330005 (2013).
electrochemical capacitor”, Signal Processing 107, 355–360 [41] R. Herrmann, “Gauge invariance in fractional field theories”,
(2015). Vol. 2, Physics Letters A 372, 5515–5522 (2008).
[24] I.S. Jesus, J.T. Machado, and J.B. Cunha, “Fractional electrical [42] R. Herrmann, Fractional Calculus: An Introduction for Physi-
impedances in botanical elements”, Journal of Vibration and cists, World Scientific Publishing Co. Pte. Ltd. (2011).
Control 14, 1389 (2008). [43] M.D. Ortigueira, M. Rivero, and J. Trujillo, “From a gen-
[25] U. Skwara, J. Martins, P. Ghaffari, M. Aguiar, J. Boto, and eralized Helmholtz decomposition theorem to the fractional
N. Stollenwerk, “Applications of fractional calculus to epi- Maxwell equations”, Commun Nonlinear Sci Numer Simulat
demiological models”, Numerical Analysis and Applied Math- 22, 1036–1049 (2015).
ematics, ICNAAM 2012, AIP Conf. Proc. 1479, 1339–1342 [44] M.D. Ortigueira and J. Machado, “Which derivative? Fractal
(2012). and fractional”, Fractal and Fractional 1(1), 3 (2017).
[26] M.J. Lewis and M.A. McNarry, “Twenty-four hour variation [45] M. Ostoja-Starzewski, J. Li, H. Joumaa, and P.N. Demmie,
in heart rate variability indices derived from fractional dif- “From fractal media to continuum mechanics”, Z. Angew.
ferintegration”, Clin. Physiol. Funct. Imaging 35(1), 57–63 Math. Mech. 94, 373–401 (2014).
(2015). [46] V.E. Tarasov, “Fractional vector calculus and fractional
[27] C. Ionescu, J.T. Machado, and R. De Keyser, “Modeling of the Maxwell’s equations”, Annals of Physics 323(11), 2756–2778
lung impedance using a fractional order ladder network with (2008).
constant phase elements”, IEEE Transactions on Biomedical [47] E.C. Grigoletto and E.C. de Oliveira, “Fractional versions of
Circuits and Systems 5(1), 83–89 (2011). the fundamental theorem of calculus”, Applied Mathematics 4,
[28] D. Copot, R. De Keyser, E. Derom, M.D. Ortigueira, and C.M. 23–33 (2013).
Ionescu, “Reducing bias in fractional order impedance estima- [48] M.D. Ortigueira and J. Machado, “Fractional definite integral”,
tion for lung function evaluation”, Biomedical Signal Process- Fractal and Fractional 1(1), 2 (2017).
ing and Control 39, 74–80 (2018). [49] A. Aghili and P.B. Moghaddam, “Certain theorems on two di-
[29] J.A.T. Machado, “And I say to myself: What a fractional mensional Laplace Transform and non-homogeneous parabolic
world!”, Fractional Calculus and Applied Analysis 14, 635 partial differential equations”, Surveys in Mathematics and its
(2011). Applications 6, 165–174 (2011).
[30] R. Magin, M.D. Ortigueira, I. Podlubny, and J. Trujillo, “On [50] M.J. Roberts, Signals and Systems: Analysis Using Transform
the Fractional Signals and Systems”, Signal Processing 91, Methods and Matlab, McGraw-Hill, 2003.
350–371 (2011). [51] M.D. Ortigueira and J.T. Machado, “What is a fractional deriv-
[31] M.M. Meerschaert, J. Mortensen, and S.W. Wheatcraft, “Frac- ative?”, J. Comput. Phys. 293, 4–13 (2015).
tional vector calculus for fractional advection–dispersion”, [52] M.D. Ortigueira, R.L. Magin, J.J. Trujillo, M.P. Velasco, “A re-
Physica A 367, 181–190 (2006). al regularized fractional derivative”, Signal Image Video P, 6,
[32] M.M. Meerschaert, Fractional Calculus, Anomalous Diffusion, 351–358 (2012).
and Probability, in Fractional Dynamics: Recent Advances, [53] J. Liouville, “Memóire sur le calcul des différentielles à indices
edited by: J. Klafter, S. C. Lim, and R. Metzler, World Scien- quelconques”, Journal de l’École Polytechnique 13, 71–162
tific Publishing Co. Pte. Ltd. (2011). (1832). (In French)
[33] R.L. Magin, C. Ingo, L. Colon-Perez, W. Triplett, and T.H. [54] J. Liouville, “Memóire sur l’usage que l’on peut faire de
Mareci, “Characterization of anomalous diffusion in porous bi- la formule de Fourier, dans le calcul des différentielles à in-
ological tissues using fractional order derivatives and entropy”, dices quelconques”, Journal für die reine und angewandte
Microporous and Mesoporous Materials 178, September 15, Mathematik (Journal de Crelle) 13, 219–232 (1835). (In
pp. 39–43 (2013). French)
[34] L. Boyadjiev and Y. Luchko, “Multi-dimensional α-fractional [55] M.D. Ortigueira, “Riesz potential operators and inverses via
diffusion–wave equation and some properties of its funda- fractional centred derivatives”, International Journal of Math-
mental solution”, Computers & Mathematics with Applications ematics and Mathematical Sciences 2006, ID 48391, pp. 1–12
73(12), 2561–2572 (2017). (2006).

Bull. Pol. Ac.: Tech. 66(4) 2018 401

 M.D. Ortigueira and J.A.T. Machado

[56] M.D. Ortigueira, “Fractional central differences and deriva- [62] D.A.W. Pecknold, “On the role of the Stokes-Helmholtz de-
tives”, Journal of Vibration and Control 14(9–10), 1255–1266 composition in the derivation of displacement potentials in
(2008). classical elasticity”, Journal of Elasticity 1(2), 171–174 (1971).
[57] M.D. Ortigueira and J.J. Trujillo, “A unified approach to frac- [63] D.D. Joseph, “Helmholtz decomposition coupling rotational to
tional derivatives”, Commun Nonlinear Sci. Numer. Simulat. irrotational flow of a viscous fluid”, Proceedings of the Nation-
17, 5151–5157 (2012). al Academy of Sciences of USA 103(39), 14272–14277 (2006).
[58] M. Riesz, “L’integral de Riemann-Liouville et le probleme de [64] E. Kapuścik, “Generalized Helmholtz theorem and gauge in-
Cauchy”, Acta Mathematica 81(1), 1–222 (1949). variance of classical field theories”, Lett. Nuovo Cimento 42,
[59] W. Chen and S. Holm, “Fractional Laplacian time-space mod- 263–266 (1985).
els for linear and nonlinear lossy media exhibiting arbitrary [65] R.D. Nevels, A Simple ‘Derivation’ of Maxwell’s Equations
frequency dependency”, The Journal of the Acoustical Society Relying on the new Extended Helmholtz Theorem, IEEE Ant.
of America 115(4), 1424–1430 (2004). Prop. Soc. Int. Symp. Newport Beach, California, pp. 822–825
[60] L. Caffarelli and L. Silvestre, “An extension problem related (1995).
to the fractional Laplacian”, Communications in Partial Dif- [66] E.K. Jaradat, R.S. Hijjawi, and J.M. Khalifeh, “Maxwell’s
ferential Equations 32(8), 1245–1260 (2007). equations and electromagnetic Lagrangian density in fractional
[61] M.D. Ortigueira, T.-L. Laleg-Kirati, and J.T. Machado, “Riesz form”, Journal of Mathematical Physics 53, 033505 (2012).
potential versus fractional Laplacian”, Journal of Statistical [67] T. Apostol, Mathematical Analysis, Addison Wesley, Fifth Edi-
Mechanics: Theory and Experiment 2014(9), P09032 (2014). tion, (1981).

402 Bull. Pol. Ac.: Tech. 66(4) 2018