Nuno R. O. Bastos, Rui A. C. Ferreira and Delm F. M.

Torres

THE FRACTIONAL DIFFERENCE CALCULUS OF VARIATIONS

Nuno R. O. Bastos1 , Rui A. C. Ferreira2 , Delm F. M. Torres3
Department of Mathematics Polytechnic Institute of Viseu 3504-510 Viseu, Portugal e-mail: nbastos@estv.ipv.pt Faculty of Engineering and Natural Sciences Lusophone University of Humanities and Technologies 1749-024 Lisbon, Portugal e-mail: ruiacferreira@ua.pt
3 2 1

Department of Mathematics University of Aveiro 3810-193 Aveiro, Portugal e-mail: delm@ua.pt

Abstract. We introduce a discrete-time fractional calculus of variations. First order fractional necessary optimality conditions of Euler-Lagrange type for the time scale hZ, h > 0, are established. Examples illustrating the use of the new condition are given. They show that the solutions to the considered fractional problems: (i) converge to the classical discrete-time solutions when the fractional order of the discrete-derivatives tend to integer values, (ii) converge to the fractional continuous-time solutions when h tends to zero. 2010 Mathematics Subject Classication: 26A33, 39A12, 49K05. Keywords: fractional difference calculus, fractional summation by parts, calculus of variations, Euler-Lagrange equations, time scale hZ.

INTRODUCTION

The Fractional Calculus is currently an important research eld in several different areas: physics (including classical and quantum mechanics as well as thermodynamics), chemistry, biology, economics, and control theory [13, 15, 17]. It has origin more than 300 years ago when LHopital asked Leibniz what should be the meaning of a derivative of non-integer order. After that episode several more famous mathematicians contributed to the development of Fractional Calculus: Abel, Fourier, Liouville, Riemann, Riesz, just to mention a few names. 1

The Fractional Difference Calculus of Variations

In [12] Miller and Ross dene a fractional sum of order > 0 via the solution of a linear difference equation. Namely, they present it as (see Section 2 for the notations used here) 1 f (t) = ()
t

(t (s))(1) f (s).
s=a

(1)

This was done in analogy with the RiemannLiouville fractional integral of order > 0,
a Dx f (x) =

1 ()

(x s)1 f (s)ds,
a

which can be obtained via the solution of a linear differential equation [12, 13]. Some basic properties of the sum in (1) were obtained in [12]. More recently, F. Atici and P. Eloe [4, 5] dened the fractional difference of order > 0, i.e., f (t) = ((1) f (t)), and developed some of its properties that allow to obtain solutions of certain fractional difference equations. Fractional differential calculus has been widely developed in the past few decades due mainly to its demonstrated applications in various elds of science and engineering. The study of fractional problems of the Calculus of Variations and respective Euler-Lagrange equations is a fairly recent issue see [1, 2, 7, 9, 14] and references therein and include only the continuous case. It is well known that discrete analogues of differential equations can be very useful in applications [10, 11]. Therefore, we consider pertinent to start a fractional discrete-time theory of the calculus of variations. Our objective is two-fold. On one hand we proceed to develop the theory of fractional difference calculus, namely, we introduce the concept of left and right fractional sum/difference (cf. Denition 2.1). On the other hand, we believe that the present work will potentiate research not only in the fractional calculus of variations but also in solving fractional difference equations, specically, fractional equations in which left and right fractional differences appear. We formulate our main result in the time-scale hZ, h > 0. Because the theory of fractional difference calculus is in its infancy [4, 5, 12], the paper is self contained. We begin, in Section 2, to give the denitions and results needed throughout. In Section 3 we present the new results of the paper. Finally, in Section 4 we show some examples.

PRELIMINARIES

We begin by introducing some notation used throughout. The notation is borrowed from the calculus on time scales [6]. Let a, b Z with a < b. We put T = [a, b] Z, T = [a, b 1] Z 2 and T = [a, b 2] Z. Denote by F the set of all real valued functions dened on T. Also, we will frequently write (t) = t + 1, (t) = t 1 and f (t) = f ((t)). The usual conventions c1 1 t=c f (t) = 0, c T, and i=0 f (i) = 1 remain valid here. As usual, the forward difference is dened by f (t) = f (t) f (t). If we have a function f of two variables, f (t, s), its partial (difference) derivatives are denoted by t and s , respectively. For arbitrary x, y R dene (when it makes sense) x(y) = (x + 1) , (x + 1 y) 2

Nuno R. O. Bastos, Rui A. C. Ferreira and Delm F. M. Torres

where is the gamma function. The following property of the gamma function, (x + 1) = x(x), will be frequently used. It was mentioned in Section 1 that equality (1) was introduced in [12] as the fractional sum of order > 0. In fact one nds the denition of left and right fractional sum (Denition 2.1) when proving the fractional formula of summation by parts (Lemma 3.1), which is a crucial result in order to fulll our goal (see proof of the Euler-Lagrange equation (10) given by Theorem 3.4). Denition 2.1. Let f F . The left fractional sum and the right fractional sum of order > 0 are dened, respectively, as
a t (f )(t)

1 = ()

(t (s))(1) f (s),
s=a

(2)

and
t b (f )(t)

1 = (s (t))(1) f (s). () s=t+

(3)

Remark 2.1. The above sums (2) and (3) are dened for t = a + mod (1) and t = a mod (1), respectively, while f (t) is dened for t = a mod (1) (see [12]). Throughout we will write (2) and (3), respectively, in the following way:
a t f (t)

1 = () 1 ()

(t + (s))(1) f (s),
s=a

t T,

t b f (t) =

(s + (t))(1) f (s),
s=t

t T.

Remark 2.2. The left fractional sum dened in (2) coincides with the fractional sum dened in [12]. The analogy of (2) and (3) with the RiemannLiouville left and right fractional integrals of order > 0 is clear:
a Dx f (x)

1 = () 1 ()

(x s)1 f (s)ds,
a b

x Db f (x) =

(s x)1 f (s)ds.
x

The Fractional Difference Calculus of Variations

It was proved in [12] that lim0 a f (t) = f (t). We do the same for the right fractional t sum using a different method. Let > 0 be arbitrary. Then,
t b f (t) =

1 ()

(s + (t))(1) f (s)
s=t

= f (t) +

1 ()
b

(s + (t))(1) f (s)
s=(t)

= f (t) +
s=(t) b

(s + t) f (s) ()(s t + 1) + i) f (s). (s t + 1)

st1 i=0 (

= f (t) +
s=(t)

Therefore, lim0 t f (t) = f (t). It is now natural to dene b

0 a t f (t)

= t 0 f (t) = f (t), b

which we do here, and to write

a t f (t) = f (t) +

( + 1)

t1

(t + (s))(1) f (s),
s=a b

t T,

0,

(4)

t b f (t)

= f (t) + ( + 1)

(s + (t))(1) f (s),
s=(t)

t T,

0.

The next theorem was proved in [5]. Theorem 2.2. Let f F and > 0. Then, the equality
a t f (t)

= (a f (t)) t

(t + a)(1) f (a), ()

t T ,

holds. Remark 2.3. It is easy to include the case = 0 in Theorem 2.2. Indeed, in view of previous denitions, we get
a t f (t)

= (a f (t)) t

(t + a)(1) f (a), ( + 1)

t T ,

(5)

for all 0. We now present the counterpart of Theorem 2.2 for the right fractional sum: 4

Nuno R. O. Bastos, Rui A. C. Ferreira and Delm F. M. Torres

Theorem 2.3. Let f F and 0. Then, the equality

t (b) f (t)

(b + (t))(1) f (b) + (t f (t)), b ( + 1)

t T ,

(6)

holds. The proof of Theorem 2.3 is very similar to the one of Theorem 2.2. Denition 2.4. Let 0 < 1 and set = 1 . Then, the left fractional difference and the right fractional difference of order of a function f F are dened, respectively, by
a t f (t)

= (a f (t)), t

t T , t T .

and
t b f (t)

= (t f (t))), b

MAIN RESULTS

Our aim is to introduce the discrete-time fractional problem of the calculus of variations and to prove corresponding necessary optimality conditions. For that we need a fractional version of the summation by parts formula.

3.1 Fractional summation by parts

Lemma 3.1 (Fractional summation by parts). Let f and g be real valued functions dened on Tk and T, respectively. Fix 0 < 1 and put = 1 . Then,
b1 b2

f (t)a g(t) = f (b 1)g(b) f (a)g(a) + t

t=a t=a

t (b) f (t)g (t) b1

Proof. From (5) we can write

b1

g(a) (t + a)(1) f (t) ( + 1) t=a

b1

b1

t=(a)

(t + (a))(1) f (t) . (7)

f (t)a g(t) t
t=a

=
t=a b1

f (t)(a g(t)) t f (t)

t=a b1 a t g(t)

= =
t=a

(t + a)(1) g(a) ( + 1) (t + a)(1) f (t)g(a). ( + 1) (8)

b1

f (t)a g(t) + t
t=a

The Fractional Difference Calculus of Variations

Using the relation (4) we get

b1

f (t)a g(t) t
t=a b1

=
t=a b1

f (t) g(t) + ( + 1) f (t)g(t) + ( + 1) f (t)g(t) + ( + 1)

t1

(t + (s))(1) g(s)
t1

s=a b1

=
t=a b1

f (t)
t=a b2 s=a

(t + (s))(1) g(s)
b1

=
t=a

g(t)
t=a b2 s=(t)

(s + (t))(1) f (s)

= f (b 1)[g(b) g(b 1)] +

t=a

g(t)t f (t), (b)

where the third equality follows from the fact that if and are two functions dened on T and is a function dened on T T , then the equality
b1 t1 b2 b1

(t)
t=a s=a

(t, s)(s) =
t=a

(t)
s=(t)

(s, t)(s)

holds. We proceed to develop the right hand side of the equality

b1 b2

f (t)a g(t) = f (b 1)[g(b) g(b 1)] + t

t=a t=a

g(t)t f (t) (b)

as follows:
b2

f (b 1)[g(b) g(b 1)] +

t=a

g(t)t f (t) (b) g(t)t f (t) (b) g(a) ( + 1)

t=b1 t=a b2

= f (b 1)[g(b) g(b 1)] + = f (b 1)g(b) f (a)g(a)

b2 t (b) f (t) t=a

t=a

g (t)(t f (t)) (b)

b1

(s + (a))(1) f (s)
s=(a)

g (t) ,

where the rst equality follows from the usual summation by parts formula. Putting this into (8), 6

Nuno R. O. Bastos, Rui A. C. Ferreira and Delm F. M. Torres

we get:
b1 b2

f (t)a g(t) t
t=a

= f (b 1)g(b) f (a)g(a) +
t=a b1

t (b) f (t)

g (t) (s + (a))(1) f (s).

g(a) ( + 1)

t=a

(t + a) ()

(1)

f (t)

g(a) ( + 1)

b1

s=(a)

The theorem is proved. Remark 3.1. When = 1 and f and g are dened on {a, a + 1, . . . , b}, then we obtain from our formula (7) the usual integration by parts formula in difference calculus:
b1 b2

f (t)g(t) = f (b 1)g(b) f (a)g(a)

t=a b1 t=a

f (t)g (t)

= f (t)g(t)|b a
t=a

f (t)g (t) .

3.2 Necessary optimality condition

We begin to x two arbitrary real numbers and such that , (0, 1]. Further, we put = 1 and = 1 . Let a function L(t, u, v, w) : T RRR R be given. We assume that the second-order partial derivatives Luu , Luv , Luw , Lvw , Lvv , and Lww exist and are continuous. Consider the functional L : F R dened by
b1

L(y()) =
t=a

L(t, y (t), a y(t), t y(t)) t b

(9)

and the problem, that we denote by (P1 ), of minimizing (9) subject to the boundary conditions y(a) = A and y(b) = B (A, B R). Our aim is to derive necessary conditions of rst and second order for problem (P1 ). Denition 3.2. For f F we dene the norm f = max |f (t)| + max |a f (t)| + max |t f (t)|. t b
tT tT tT

A function y F with y (a) = A and y (b) = B is called a local minimizer for problem (P1 ) provided there exists > 0 such that L() L(y) for all y F with y(a) = A and y(b) = B y and y y < . Denition 3.3. A function F is called an admissible variation for problem (P1 ) provided = 0 and (a) = (b) = 0. 7

The Fractional Difference Calculus of Variations

The next theorem presents a rst order necessary condition for problem (P1 ). Theorem 3.4 (The fractional discrete-time EulerLagrange equation). If y F is a local mini mizer for problem (P1 ), then Lu [](t) + t Lv [](t) + a Lw [](t) = 0 y y y t (b) holds for all t T , where the operator [] is dened by [y](s) = (s, y (s), a y(s), s y(s)). s b Proof. Suppose that y () is a local minimizer of L(). Let () be an arbitrary xed admissible variation and dene the function : () , () R by () = L(() + ()). y This function has a minimum at = 0, so we must have (0) = 0, i.e.,
b1
2

(10)

Lu [](t) (t) + Lv [](t)a (t) + Lw [](t)t (t) = 0, y y y t b

t=a

which we may write, equivalently, as

b2

Lu [](t) (t)|t=(b) + y
t=a

Lu [](t) (t) y
b1 b1

+
t=a

Lv [](t)a (t) y t

+
t=a

Lw [](t)t (t) = 0. (11) y b

Using Lemma 3.1, and the fact that (a) = (b) = 0, we get for the third term in (11) that
b1 b2

Lv [](t)a (t) y t
t=a

=
t=a

y t (b) Lv [](t)

(t).

(12)

Using (6) we can write

b1

Lw [](t)t (t) y b
t=a b1

=
t=a b1

Lw [](t)(t (t)) y b (13) Lw [](t) t (t) y (b)

t=a b1

= =

(b + (t))(1) (b) ( + 1)
b1

Lw [](t)t (t) y (b)

t=a

(b) ( + 1) 8

(b + (t))(1) Lw [](t) . y
t=a

Nuno R. O. Bastos, Rui A. C. Ferreira and Delm F. M. Torres

It follows that
b1

Lw [](t)t (t) y (b)

t=a b1

=
t=a b1

Lw [](t) (t) + y

( + 1)

b1

s=(t)

=
t=a b1

Lw [](t)(t) + y ( + 1) Lw [](t)(t) + y ( + 1) (t)a Lw [](t). y t

b2

(s + (t))(1) (s)
b1

(14)

Lw [](t) y
t=a b1 s=(t) t1

(s + (t))(1) (s) (t + (s))(1) Lw [](s) y

=
t=a b1

(t)
t=a s=a

=
t=a

We apply again the usual summation by parts formula, this time to (14), to obtain:
b1

(t)a Lw [](t) y t
t=a b2

=
t=a

(t)a Lw [](t) + ((b) ((b)))a Lw [](t)|t=(b) y y t t

b2 t=b1 (t)a Lw [](t) t=a y t

(t)(a Lw [](t)) y t 1)a Lw [](t)|t=(b) y t

b2

(15)

+ =

t=a (b)a t Lw [](t)|t=(b) (b y

(b)a Lw [](t)|t=(b) y t

(a)a Lw [](t)|t=a y t

t=a

(t)a Lw [](t). y t

Since (a) = (b) = 0 it follows, from (14) and (15), that

b1 b2

Lw [](t)t (t) y (b)

t=a

=
t=a

(t)a Lw [](t) y t

and, after inserting in (13), that

b1 b2

Lw [](t)t (t) y b
t=a

=
t=a

(t)a Lw [](t). y t

(16)

By (12) and (16) we may write (11) as

b2

Lu [](t) + t Lv [](t) + a Lw [](t) (t) = 0. y y y t (b)

t=a

The Fractional Difference Calculus of Variations

Since the values of (t) are arbitrary for t T , the EulerLagrange equation (10) holds along y. Remark 3.2. If the initial condition y(a) = A is not present (i.e., y(a) is free), we can use standard techniques to show that the following supplementary condition must be fullled: Lv (a) + ( + 1)
b1

(t + a)(1) Lv [](t) y
t=a b1

t=(a)

Similarly, if y(b) = B is not present (i.e., y(b) is free), the equality Lu ((b)) + Lv ((b)) Lw ((b)) + ( + 1)
b1

(t + (a))(1) Lv [](t) + Lw (a) = 0. (17) y

(b + (t))(1) Lw [](t) y
t=a b2

t=a

((b) + (t))(1) Lw [](t) y

= 0 (18)

holds. We just note that the rst term in (18) arises from the rst term on the left hand side of (11). Equalities (17) and (18) are the fractional discrete-time natural boundary conditions.

3.3 Generalization
Theorem 3.4 can be generalized from the time scale Z to hZ, h > 0. The proof is technical, and we present the result here without proof. Illustrative examples of the application of Theorem 3.6 are given in Section 4. We begin with some standard notation of the calculus of variations on time scales [8]. Let hZ = {hk : h > 0, k Z} and a, b hZ with a < b. The previous considered notation, valid for h = 1, is now generalized for t hZ by denoting (t) = t + h, (t) = t h, and (t) = h. 2 We put T = [a, b] hZ, T = [a, (b)] hZ and T = [a, 2 (b)] hZ. The forward difference is dened by f (t) = The delta h-integral or h-sum of f is
b
b 1 h

f (t) f (t) . (t)

f (t)t =
a
a k= h

f (kh)h .

10

Nuno R. O. Bastos, Rui A. C. Ferreira and Delm F. M. Torres

If we have a function f of two variables, f (t, s), its partial (difference) derivatives are denoted by t,h and s,h , respectively. For arbitrary x, y R dene (when it makes sense) x
(y) x ( h + 1) =h , ( x + 1 y) h y

where is the gamma function. Often, the left fractional delta integration of order > 0 is denoted as a f (t) and the t right fractional delta integration of order > 0 by t b f (t). Here, like Ross et. al. in [16], we omit the subscript t on the operator and we will write a f (t) and h f (t). h b Denition 3.5. Let f F . The left fractional h-sum is dened as
a h f (t)

= h f (t) +

( + 1)

(t + h (s))(1) f (s)s,
a

t T,

0,

while the right fractional h-sum of order 0 is given by

h b f (t)

= h f (t) +

( + 1)

(b)

(s + h (t))(1) f (s)s,
(t)

t T.

As before, we begin to x two arbitrary real numbers and such that , (0, 1], and we put := 1 and := 1 . Given a function L(t, u, v, w) : T R R R R we consider the problem
b

L(y()) =
a

L(t, y (t), a y(t), h y(t))t min h b y(a) = A , y(b) = B ,

(Ph )

where A, B R are given. Theorem 3.6 (The h-fractional EulerLagrange equation for problem (Ph )). If y F is a local minimizer for problem (Ph ), then the following equality must be satised: Lu [](t) + h Lv [](t) + a Lw [](t) = 0, y y y (b) h where the operator [] is dened by [y](s) = (s, y (s), a y(s), h y(s)) . h b Remark 3.3. The Euler-Lagrange equation (19) coincides with (10) in the particular case h = 1. If the initial condition y(a) = A is not present (y(a) is free), one can prove that the following supplementary condition must be fullled: h Lv (a) + ( + 1)
b

t T ,

(19)

(t + h a)(1) Lv [](t)t y
a b

(a)

(t + h (a))(1) Lv [](t)t + Lw (a) = 0. (20) y

11

The Fractional Difference Calculus of Variations

Similarly, if y(b) = B is not present (y(b) is free), the following equality hLu ((b)) + h Lv ((b)) h Lw ((b)) + ( + 1)
b

(b + h (t))(1) Lw [](t)t y
a (b)

((b) + h (t))(1) Lw [](t)t y

= 0, (21)

holds. The natural boundary conditions (20) and (21) reduce to (17) and (18) when h = 1.

EXAMPLES

In this section we present two illustrative examples. The results were obtained using the open source Computer Algebra System Maxima (cf. http://maxima.sourceforge.net). All computations were done running Maxima on an Intel CoreT M 2 Duo, CPU of 2.27GHz with 3Gb of RAM. Example 4.1. Let us consider the following problem:
1

J(y) =
0

1 (0 y(t))2 y (t) t min , h 2

y(0) = 0 ,

y(1) = 0 .

(22)

We begin by considering problem (22) with = 1 and different values of h. Our numerical results show that when h tends to zero the extremal of the problem tends to the extremal of the corresponding continuous classical problem of the calculus of variations. More precisely, when h 0 problem (22) tends to the fractional continuous problem [3, Example 2] with solution given by 1 y(t) = t(1 t) 2 when = 1. This is illustrated in Figure 1. The computation times ranged from 0.02 seconds for h = 0.5 to 0.44 seconds for h = 0.0625. Consider now problem (22) with h = 0.05 and different values of . Our numerical results show that when tends to one our extremal tends to the classical (integer order) discrete-time case. This is illustrated in Figure 2. The computation times ranged from 2.5 minutes for h = 0.7 to 6.5 minutes for h = 0.99. Example 4.2. Let us consider the following problem:
1

J(y) =
0

1 (0 y(t))2 t min , h 2

y(0) = 0 ,

y(1) = 1 .

(23)

We consider problem (23) with = 0.75 and different values of h. Our numerical results show that when h tends to zero our extremal tends to the fractional continuous extremal. Indeed, 12

Nuno R. O. Bastos, Rui A. C. Ferreira and Delm F. M. Torres

y (t)

0.15

+ *
0.10

* * + *

+ *
*

+ *
*

+ *
*

0.05

+ *
*

+ *
*

0+ * 0

+ *
1 t

Figure 1: Extremal y(t) for problem of Example 4.1 with = 1 and different hs (: h = 0.50; :
h = 0.25; +: h = 0.125; : h = 0.0625).

y (t) 0.35 0.30 0.25 0.20 0.15

0.10

0.05

+ + *
*

+
*

+
*

+++++ ++ * * * * * ++ * *
* *

+
*

+
*

+
*

0+ * 0

+ *

+ *

+ *

+ *
1 t

Figure 2: Extremal y(t) for problem of Example 4.1 with h = 0.05 and different s (: = 0.70; : = 0.75; +: = 0.95; : = 0.99).

13

The Fractional Difference Calculus of Variations

when h 0 problem (23) tends to the fractional continuous problem of the calculus of variations [1, Example 1], with solution given by
x

y(x) = 2
0

dt . [(1 t)(x t)]0.25

(24)

This is illustrated in Figure 3. The computation times ranged from 10.46 seconds for h = 0.5 to 25 minutes for h = 1/30.
y (t) 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

+ *
*

0+ * 0

* +

+ *

+ *

+ *

+ *

+ *

+ *

Figure 3: Extremal y(t) for problem of Example 4.2 with = 0.75 and different hs (: h = 0.50; +:
h = 0.125; : h = 0.0625; : h = 1/30). The solid line represent (24).

ACKNOWLEDGEMENTS This work is part of the rst authors PhD project which is carried out at the University of Aveiro under the Doctoral Programme Mathematics and Applications of Universities of Aveiro and Minho. The second author was supported by The Portuguese Foundation for Science and Technology (FCT) through the PhD fellowship SFRH/BD/39816/2007; the third author was supported by FCT through the R&D unit Centre for Research on Optimization and Control (CEOC), conanced by FEDER/POCI 2010.

REFERENCES [1] O. P. Agrawal, Formulation of Euler-Lagrange equations for fractional variational problems, J. Math. Anal. Appl. 272 (1) (2002) 368379. 14

Nuno R. O. Bastos, Rui A. C. Ferreira and Delm F. M. Torres

[2] O. P. Agrawal, A general formulation and solution scheme for fractional optimal control problems, Nonlinear Dynam. 38 (1-4) (2004) 323337. [3] O. P. Agrawal, A general nite element formulation for fractional variational problems, J. Math. Anal. Appl. 337 (2008), no. 1, 112. [4] F. M. Atici and P. W. Eloe, A transform method in discrete fractional calculus, Int. J. Difference Equ. 2 (2) (2007) 165176. [5] F. M. Atici and P. W. Eloe, Initial value problems in discrete fractional calculus, Proc. Amer. Math. Soc. 137 (3) (2009) 981989. [6] M. Bohner, Calculus of variations on time scales, Dynam. Systems Appl. 13 (3-4) (2004) 339349. [7] R. A. El-Nabulsi and D. F. M. Torres, Necessary optimality conditions for fractional action-like integrals of variational calculus with Riemann-Liouville derivatives of order (, ), Math. Meth. Appl. Sci. 30 (15) (2007), 19311939. [8] R. A. C. Ferreira and D. F. M. Torres, Higher-order calculus of variations on time scales, in Mathematical control theory and nance, Springer, Berlin, 2008, 149159. [9] G. S. F. Frederico and D. F. M. Torres, Fractional conservation laws in optimal control theory, Nonlinear Dynam. 53 (3) (2008) 215222. [10] R. Hilscher and V. Zeidan, Nonnegativity and positivity of quadratic functionals in discrete calculus of variations: survey, J. Difference Equ. Appl. 11 (2005), no. 9, 857875. [11] W. G. Kelley and A. C. Peterson, Difference equations Academic Press, Boston, MA, 1991. [12] K. S. Miller and B. Ross, Fractional difference calculus, in Univalent functions, fractional calculus, and their applications (K riyama, 1988), 139152, Horwood, Chichester, 1989. o [13] K. S. Miller and B. Ross, An Introduction to the Fractional Calculus and Fractional Differential Equations John Wiley and Sons, Inc., New York, 1993. [14] S. I. Muslih and D. Baleanu, Fractional Euler-Lagrange equations of motion in fractional space, J. Vib. Control 13 (2007), no. 9-10, 12091216. [15] M. D. Ortigueira, Fractional central differences and derivatives, J. Vib. Control 14 (2008), no. 9-10, 12551266. [16] B. Ross, S. G. Samko and E. R. Love, Functions that have no rst order derivative might have fractional derivatives of all orders less than one, Real Anal. Exchange 20 (1994/95), no. 1, 140157. [17] M. F. Silva, J. A. Tenreiro Machado and R. S. Barbosa, Using fractional derivatives in joint control of hexapod robots, J. Vib. Control 14 (2008), no. 9-10, 14731485. 15