Computers and Mathematics with Applications 62 (2011) 1046–1054

Contents lists available at ScienceDirect

Computers and Mathematics with Applications

journal homepage: www.elsevier.com/locate/camwa

Numerical solution of fractional differential equations using the

generalized block pulse operational matrix
Yuanlu Li ∗ , Ning Sun
College of Information and Control, Nanjing University of Information Science and Technology, Nanjing 210044, PR China

article info abstract

Keywords: The Riemann–Liouville fractional integral for repeated fractional integration is expanded in
Operational matrix
block pulse functions to yield the block pulse operational matrices for the fractional order
Fractional calculus
integration. Also, the generalized block pulse operational matrices of differentiation are
Fractional differential equations
derived. Based on the above results we propose a way to solve the fractional differential
equations. The method is computationally attractive and applications are demonstrated
through illustrative examples.
Crown Copyright © 2011 Published by Elsevier Ltd. All rights reserved.

1. Introduction

Fractional calculus has become the focus of interest for many researchers in different disciplines of science and
technology [1–5]. A great deal of research has shown the advantageous use of the fractional calculus in the modeling and
control of many dynamical systems [1–8]. However, the fractional order systems are represented in the frequency domain
by irrational transfer functions, which correspond to linear time varying differential equations in the time domain. One of
the main difficulties is how to solve the fractional differential equations, so some techniques were proposed to solve them.
The most commonly used ones are Adomian decomposition method (ADM) [9], Variational Iteration Method (VIM) [10],
Fractional Differential Transform Method (FDTM) [11], Operational Matrix Method [12], Homotopy Analysis Method [13,14],
Fractional Difference Method (FDM) [15] and Power Series Method [16]. Also there are some classical solution techniques,
e.g. Laplace Transform Method [17].
It is somewhat surprising that among different solution techniques few papers reported application of the orthogonal
function method for the fractional order differential equations [18–21]. However, through the analysis of the orthogonal
function method, we hold that it should be applicable to solve the fractional order systems.
For an ordinary dynamical system, the orthogonal function method is based on converting the underlying differential
equations into integral equations through integration, approximating various signals involved in the equation by truncated
orthogonal series and using the operational matrix of integration to eliminate the integral operations. Typical examples are
the Walsh functions [22,23], block pulse functions [24–27], Legendre polynomials [28–30], Chebyshev polynomials [31],
Laguerre polynomials [32,33], and Fourier series [34,35].
In this paper, our purpose is to extend the orthogonal function method to solve the fractional linear differential equations.
Similar to the process of using the orthogonal function method to solve the ordinary dynamical systems, for the fractional
systems, we need to convert the underlying fractional differential equations into integral equations through the fractional
integration, expand various signals involved in the equation by block pulse functions and using the operational matrix for
the fractional integration to eliminate the fractional integral operations. So, there are some questions to be answered:
(i) How to derive block pulse operational matrices of the fractional integration and differentiation.
(ii) How to solve the fractional order linear systems via block pulse operational matrices of the fractional integration.

∗ Corresponding author.
E-mail address: [email protected] (Y. Li).

0898-1221/$ – see front matter Crown Copyright © 2011 Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.camwa.2011.03.032
 Y. Li, N. Sun / Computers and Mathematics with Applications 62 (2011) 1046–1054 1047

2. Brief review of block pulse functions (BPFs) and the related operational matrices [36]

A set of BPFs Φm (t ) containing m component functions in the semi-open interval [0, T ) is given by

Φm (t ) , [ϕ0 (t ) ϕ1 (t ) · · · ϕi (t ) · · · ϕm−1 (t )]T , (1)

where T denotes transpose.

The ith component ϕi (t ) of the BPF vector Φm (t ) is defined as

iT /m ≤ t < (i + 1)T /m

1
ϕi (t ) =
0 otherwise

where i = 0, 1, 2, . . . , (m − 1).
A square integrable time function f (t ) of Lebesgue measure may be expanded into an m-term BPF series in t ∈ [0, T ) as

f (t ) = [c0 c1 · · · ci · · · c(m−1) ]Φm (t ) , C T Φm (t ). (2)

The constant coefficients ci ’s in Eq. (2) are given by

∫ (i+1)h
ci = (1/h) f (t )dt , (3)
ih

where h = T /m is the duration of each component BPF along time scale.

In the m-term BPF domain, an operational matrix for integration Pm has been given by Deb et al. [37] as the following
upper triangular matrix:
1/2 1 1 ··· 1 1

 0 1/2 1 ··· 1 1 
 0 0 1/2 ··· 1 1 
Pm , h    . (4)
··· ··· ··· ··· ··· ···
0 0 0 ··· 1/2 1

0 0 0 ··· 0 1/2

The matrix Pm performs as an integrator in the BPF domain and it is pivotal in any BPF domain analysis. Thus, approximate
integration of a function f (t ), using Eqs. (2) and (4), is
∫ T
f (t )dt ≈ C T Pm Φm (t ). (5)
0

3. Block pulse operational matrices for fractional calculus

We now derive the operational matrix for the fractional calculus. Several definitions of a fractional calculus have been
proposed [2,3,5]. we formulate the problem in terms of the Riemann–Liouville fractional integration, which is defined as
∫ t
1 1
(I α f )(t ) = (t − τ )α−1 f (τ )dτ = t α−1 ∗ f (t ), 0 ≤ t < T, (6)
Γ (α) 0 Γ (α)
where α is the order of the integration, Γ (α) is the Gamma function and t α−1 ∗ f (t ) denotes the convolution product of
t α−1 and f (t ). Now if f (t ) is expanded in block pulse functions, as shown in Eq. (2), the Riemann–Liouville fractional integral
becomes
1 1
(I α f )(t ) = t α−1 ∗ f (t ) ≈ C T {t α−1 ∗ Φm (t )}. (7)
Γ (α) Γ (α)
Thus if t α−1 ∗ f (t ) can be integrated, then expanded in block pulse functions, the Riemann–Liouville fractional integral is
solved via the block pulse functions.
1
According to the linear nature of Laplace transform, in order to compute the convolution product C T Γ (α) {t α−1 ∗ Φm (t )},
we only need to compute 1
Γ (α)
t α−1 ∗ ϕi (t ).
For ϕi (t ), applying definition of the convolution, we have
∫ t
α−1
t ∗ ϕi (t ) = τ α−1 · ϕi (t − τ )dτ . (8)
0
1048 Y. Li, N. Sun / Computers and Mathematics with Applications 62 (2011) 1046–1054

Since ϕi (t ) = 1 in [iT /m, (i + 1)T /m), then we have

tα
∫ t
 τ α−1 dτ = , 0 ≤ t < T /m


α
t α−1 ∗ ϕ0 (t ) = ∫0 t (9)
α− t α − (t − T /m)α
τ dτ =
1
, T /m ≤ t < T


α

t −T / m

0, 0 ≤ t < T /m

 (t − T /m)α


, T /m ≤ t < 2T /m

t α−1 ∗ ϕ1 (t ) = α (10)
 (t − T /m)α − (t − 2T /m)α

, 2T /m ≤ t < T


α
0, 0 ≤ t < 2T /m

 (t − T /m)α


, 2T /m ≤ t < 3T /m

t α−1 ∗ ϕ2 (t ) = α (11)
α α
 (t − 2T /m) − (t − 3T /m) , 3T /m ≤ t < T



α
..
.
0, 0 ≤ t < (m − 1)T /m

t α−1
∗ ϕm−1 (t ) = [t − (m − 1)T /m]α (12)
, (m − 1)T /m ≤ t < T .
α
Set
(I α Φm )(t ) ≈ Fα Φm (t ). (13)
Next, we derive Block pulse operational matrix for fractional integration. For (I α ϕ0 )(t ) we have
(I α ϕ0 )(t ) = f11 ϕ0 (t ) + f12 ϕ1 (t ) + · · · + f1m ϕm−1 (t ) (14)
where
b  α
tα
∫
m m b 1
f11 = dt = · (15)
b 0 Γ (α + 1) Γ (α + 2)
m
iT /m
 α
t α − (t − T /m)α
∫
m b 1
f1i = dt = · [iα+1 − 2 · (i − 1)α+1 + (i − 2)α+1 ],
T (i−1)T /m Γ (α + 1) m Γ (α + 2)
i = 2, . . . , m. (16)
For another (I α ϕi )(t ), i = 1, . . . , m − 1, applying shifting property, we can directly obtain the coefficients expanded by
BPFs.
Finally, we can get
1 ξ1 ξ2 ··· ξm−1
 
0 1 ξ1 ··· ξm−2 
Fα = hα
1 0 0 1 ··· ξm−3 
, (17)
.. 

Γ (α + 2)  ..
0 0 0 . . 
0 0 0 0 1
where
ξk = (k + 1)α+1 − 2kα+1 + (k − 1)α+1 , (k = 1, 2, . . . , m − 1). (18)
Fα is called the block pulse operational matrix of fractional integration. When α = 1, matrix Fα is equal to matrix Pm , so
matrix Fα is a generalization of the block pulse operational matrix for integration Pm .
Let Dα is the block pulse operational matrix for the fractional differentiation, According to the property of fractional
calculus, Dα Fα = I, we can easily get matrix Dα by inverting the Fα matrix.
According to matrix theory, we know that the inverse matrix of an upper triangular matrix is also upper triangular matrix.
So, we can write

1 ξ1 ξ2 ··· ξm−1 −1 d0 d1 d2 ··· dm−1

   
0 1 ξ1 ··· ξm−2  0 d0 d1 ··· dm−2 
0 0 1 ··· ξm−3 
 = 0
 0 d0 ··· dm−3 
. (19)
..  .. 

 ..   ..
0 0 0 . . 0 0 0 . . 
0 0 0 0 1 0 0 0 0 d0
 Y. Li, N. Sun / Computers and Mathematics with Applications 62 (2011) 1046–1054 1049

Fig. 1. 0.5-order integration of the function f (t ) = t.

That is
1 ξ1 ξ2 ··· ξm−1 d0 d1 d2 ··· dm−1
   
0 1 ξ1 ··· ξm−2   0 d0 d1 ··· dm−2 
0 0 1 ··· ξm−3 
·0
 0 d0 ··· dm−3 
 = Im , (20)
..   .. 

 ..  .. 
0 0 0 . . 0 0 0 . .
0 0 0 0 1 0 0 0 0 d0

where Im is the identity matrix of order m × m. After solving the matrix equation of Eq. (20), we get
m−1
−
d0 = 1, d1 = −ξ1 d0 , . . . , dm−1 = − ξk dm−k−1 . (21)
k=1

For example, let α = 0.5, m = 8, T = 1, the operational matrices F0.5 and D0.5 are computed below:
0.2660 0.2203 0.1434 0.1160 0.1001 0.0894 0.0816 0.0755
 
 0 0.2660 0.2203 0.1434 0.1160 0.1001 0.0894 0.0816
 0 0 0.2660 0.2203 0.1434 0.1160 0.1001 0.0894
0.2660 0.2203 0.1434 0.1160 0.1001
 
 0 0 0
F0.5 = 
0.2660 0.2203 0.1434 0.1160

 0 0 0 0
0.2660 0.2203 0.1434
 
 0 0 0 0 0
0 0 0 0 0 0 0.2660 0.2203
 
0 0 0 0 0 0 0 0.2660
3.7599 −3.1148 0.5527 −0.4178 −0.0091 −0.0998 −0.0442 −0.0460
 
 0 3.7599 −3.1148 0.5527 −0.4178 −0.0091 −0.0998 −0.0442
 0 0 3.7599 −3.1148 0.5527 −0.4178 −0.0091 −0.0998
3.7599 −3.1148 0.5527 −0.4178 −0.0091
 
 0 0 0
D0.5 =  .
 0 0 0 0 3.7599 −3.1148 0.5527 −0.4178
3.7599 −3.1148 0.5527 
 
 0 0 0 0 0
0 0 0 0 0 0 3.7599 −3.1148
 
0 0 0 0 0 0 0 3.7599

The fractional integration and differentiation of the function t was selected to verify the correctness of matrices Fα and
Dα . That is because the fractional integration and differentiation of the function f (t ) = t is easily obtained as following
(2) α+1 Γ (2)
(I α f )(t ) = Γ Γ(α+ 2)
t and (Dα f )(t ) = Γ (2−α) t 1−α , respectively, which is easily used to compare the results obtained by the
proposed method. When α = 0.5, T = 1, m = 32, the comparison results for the fractional integration and differentiation
are shown in Figs. 1 and 2, respectively.

4. Solution of the fractional differential equations by the generalized block pulse operational matrix

4.1. Linear multi-order fractional differential equation

In this section, we are concerned with providing a numerical solution to multi-order fractional linear system of the form
n
Dα y(t ) = aj (t )Dβj y(t ) + a0 (t )y(t ) + f (t )
−
(22)
j =1
1050 Y. Li, N. Sun / Computers and Mathematics with Applications 62 (2011) 1046–1054

Fig. 2. 0.5-order differentiation of the function f (t ) = t.

subject to the initial conditions

y(k) (0) = bk , k = 0, 1, . . . , ⌈α⌉ − 1, (23)

where α > β1 > β2 > · · · > βn , Dα denotes the Caputo fractional derivative of order α , aj (t ) is known function for
j = 0, 1, . . . , n, and f (t ) is input signal, y(t ) is output response.
The general procedure of numerical solution of fractional differential equations consists of two steps.
First, initial conditions are used to reduce a given initial-value problem to a problem with zero initial conditions. At this
stage, instead of a given equation a modified equation, incorporating the initial values, is obtained.
Then, the generalized block pulse operational matrix of fractional integration is used to transform the fractional
differential equation into an algebraic equation.
The solution of the initial-value problem Eq. (22) can be written in the form
y(t ) = y∗ (t ) + z (t ) (24)
(k)
where y∗ (t ) is some known function, satisfying the conditions y (0) = bk , k = 0, 1, . . . , ⌈α⌉− 1, and z (t ) is a new unknown
function.
Substituting Eq. (24) into the Eq. (22) and the initial conditions Eq. (23), we obtain for the function z (t ) an initial-value
problem with zero initial conditions.
n
Dα z (t ) = aj (t )Dβj z (t ) + a0 (t )z (t ) + g (t )
−
(25)
j =1

subject to the initial conditions

z (k) (0) = 0, k = 0, 1, . . . , ⌈α⌉ − 1. (26)

The input signal g (t ) and output response Dα z (t ) may be expanded by the BPFs as

g (t ) ≈ GT Φm (t ), (27)
α
D z (t ) ≈ C Φm (t )
T
(28)
where G = [g0 , g1 , . . . , gm ]T is a known but C = [c0 , c1 , . . . , cm ]T is an unknown m × 1 column vector.
Similarly, aj (t ) for j = 0, 1, . . . , n, may also be expanded by the BPFs as

aj (t ) ≈ ATj Φm (t ) (29)
where Aj is a known m × 1 column vector.
Using Eq. (28) together with property of fractional calculus, we have

Dβj z (t ) = I α−βj [Dα z (t )] = I α−βj [C T Φm (t )] = C T Fα−βj Φm (t ). (30)

Substituting Eqs. (27)–(30) into (25), we have
n
−
C T Φm (t ) = ATj Φm (t )[Φm (t )]T [Fα−βj ]T C + AT0 Φm (t )[Φm (t )]T [Fα ]T C + GT Φm (t ). (31)
j =1

According to the properties of BPFs, we can get

ϕ1 (t ) O
 
ϕ2 (t )
Φm (t )[Φm (t )]T =  .
 
.. (32)
 . 
O ϕm (t )
 Y. Li, N. Sun / Computers and Mathematics with Applications 62 (2011) 1046–1054 1051

Set [Fα−βj ]T C = Xj = [xj1 , xj2 , . . . , xjm ]T , then we have

xj1 O
 
xj2
Φm (t )[Φm (t )]T [Fα−βj ]T C =   Φm (t ) = diag (Xj )Φm (t ).
 
.. (33)
 . 
O xjm
Using Eq. (33) we can rewrite Eq. (31) as
n
−
C T Φm (t ) = ATj diag (Xj )Φm (t ) + AT0 (diag [Fα ]T C )Φm (t ) + GT Φm (t ) (34)
j=1

or
n
−
C = diag (Xj )Aj + (diag [Fα ]T C )A0 + G. (35)
j =1

Solving the system of algebraic equations, we can obtain the coefficients C T . Then, we can get the output response

z (t ) = C T Fα Φm (t ). (36)

4.2. Nonlinear multi-order fractional differential equation

Consider the nonlinear multi-order fractional differential equation

n
Dα y(t ) = aj Dβj y(t ) + a0 [y(t )]m + g (t )
−
(37)
j =1

subject to the initial conditions

y(k) (0) = ck , k = 0, 1, . . . , ⌈α⌉ − 1 (38)

α
where α > β1 > β2 > · · · > βn , D denotes the Caputo fractional derivative, aj is constant for j = 0, 1, . . . , n.
Compared with the linear case, a major distinction lies in the computation of [y(t )]i , when we use the generalized Block
pulse operational matrix to solve the nonlinear multi-order fractional differential equation.
Set
Dα y(t ) ≈ C T Φm (t ) (39)
then we have
y(t ) ≈ C T Fα Φm (t ). (40)
Set C Fα = [x1 , x2 , . . . , xm ], then we have
T

[y(t )]i = [xi1 , xi2 , . . . , xim ]Φm (t ). (41)

The derivation is similar to the linear case, we have
n
−
C T Φm (t ) = aj C T Fα−βj Φm (t ) + a0 [xi1 , xi2 , . . . , xim ]Φm (t ) + GT Φm (t ). (42)
j=1

This is a nonlinear system of algebraic equation, we use the ‘‘fsolve’’ in Matlab to solve it.

5. Numerical examples

Example 1. Consider the fractional equation [11]

D2 y(t ) + Dα y(t ) + y(t ) = 8, t > 0, 0 < α < 2. (43)

Subject to
y(0) = 1, y′ (0) = 0.
This problem was solved in [11], for α = 0.5 and α = 1.5. Our results compared to Ref. [11] are given in Table 1, where
h = 0.001. According to the exact solution given in Ref. [11], ours results are in high agreement with results obtained using
the FDTM and better than those obtained using the ADM. We show the approximate solution in Fig. 3 for h = 0.1.
1052 Y. Li, N. Sun / Computers and Mathematics with Applications 62 (2011) 1046–1054

Table 1
Our results compared to Ref. [11].
t α = 0.5 α = 1.5
Ours FDTM [11] ADM [11] Ours FDTM [11] ADM [11]

0.1 0.039754 0.039750 0.039874 0.033510 0.033507 0.036478

0.2 0.157043 0.157036 0.158512 0.125226 0.125221 0.140640
0.3 0.347373 0.347370 0.353625 0.267611 0.267609 0.307485
0.4 0.604699 0.604695 0.622083 0.455439 0.455435 0.533284
0.5 0.921768 0.921768 0.960047 0.684336 0.684335 0.814757
0.6 1.290458 1.290457 1.363093 0.950395 0.950393 1.148840
0.7 1.702007 1.702008 1.826257 1.249959 1.249959 1.532571
0.8 2.147286 2.147287 2.344224 1.579558 1.579557 1.963033
0.9 2.616998 2.617001 2.911278 1.935832 1.935832 2.437331
1.0 3.101902 3.101906 3.521462 2.315526 2.315526 2.952567

Fig. 3. Numerical and exact solution of Example 2 for h = 0.1.

Table 2
Absolute error in y(t ) for different values of h.
t h = 0.1 h = 0.05 h = 0.025 h = 0.0125 h = 0.00625

0.5 0.0024 0.6084E−3 0.1524E−3 0.3815E−4 0.9549E−5

1.5 0.0020 0.5058E−3 0.1267E−3 0.3173E−4 0.7945E−5
2.5 0.0018 0.4403E−3 0.1102E−3 0.2759E−4 0.6902E−5
3.5 0.0016 0.4124E−3 0.1032E−3 0.2581E−4 0.6457E−5
4.5 0.0016 0.3995E−3 0.0999E−3 0.2499E−4 0.6249E−5

Example 2. Consider the fractional variable coefficient linear differential equation [38]

aD2 y(t ) + b(t )Dα2 y(t ) + c (t )Dy(t ) + e(t )Dα1 y(t ) + k(t )y(t ) = f (t ), t ∈ [0, T ] (44)

where 0 < α1 ≤ 1, 1 < α2 ≤ 2, and

b(t ) e(t )
 
1
f (t ) = a − t 2−α2
− c (t )t − t 2−α1
+ k(t ) 2 − t 2
,
Γ (3 − α2 ) Γ (3 − α1 ) 2
subject to

y(0) = 2, y′ (0) = 0.

It is easily verified that the exact solution of this problem is

1
y(t ) = 2 − t 2.
2
For a = 1, b(t ) = t 1/2 , c (t ) = t 1/3 , e(t ) = t 1/4 , k(t ) = t 1/5 , α1 = 0.5, α2 = 1.5 we present numerical values of the
solution to Eq. (44) by our method with h = 0.01 and exact solution in Fig. 3, which is in perfect agreement with the exact
solutions, the absolute errors of y(t ) at given points for different values of h are shown in Table 2.

Example 3. Consider the nonlinear fractional differential equation, which has been studied in [9,11].

Dα y(t ) = y2 (t ) + 1, t ∈ (0, 1), m − 1 < α ≤ m, (45)

 Y. Li, N. Sun / Computers and Mathematics with Applications 62 (2011) 1046–1054 1053

Table 3
Our results compared to Refs. [9,11].
t α = 0.5 α = 1.5
Ours FDTM [11] ADM [11] Ours FDTM [11] ADM [11]

0.1 0.023800 0.023790 0.023790 0.000954 0.000952 0.000951

0.2 0.067335 0.067330 0.067330 0.005385 0.005383 0.005382
0.3 0.123900 0.123896 0.123896 0.014836 0.014833 0.014833
0.4 0.191368 0.191362 0.191362 0.030455 0.030450 0.030449
0.5 0.268862 0.268856 0.268856 0.053203 0.053197 0.053196
0.6 0.356244 0.356238 0.356238 0.083931 0.083925 0.083924
0.7 0.453956 0.453950 0.453950 0.123418 0.123412 0.123412
0.8 0.563014 0.563007 0.563007 0.172397 0.172391 0.172391
0.9 0.685067 0.685056 0.685056 0.231582 0.231574 0.231574
1.0 0.822525 0.822509 0.822511 0.301686 0.301676 0.301676

subject to
yk (0) = 0, k = 0, 1, . . . , m − 1.
It has been studied by using FDTM [11] and ADM [9]. Our results compared to Refs. [9,11] are given in Table 3, where
h = 0.01. Our results are in good agreement with the results in Refs. [9,11].

6. Conclusion

In this paper, we derive a numerical method for the fractional differential equations based on the operational matrix
Fα for the fractional integration and differentiation. A general procedure of forming this matrix Fα is summarized. Several
examples are given to demonstrate the powerfulness of the proposed method. The matrix Fα can also be used to solve
problems such as fractional system identification, fractional order optimal control.

Acknowledgments

The work was supported by the Jiangsu Ordinary University Science Research Project under Grant (08KJD410002,
09KJB510007) and in part by the Foundation of Nanjing University of Information Science & Technology under Grant
(20080305 and 20080256).

References

[1] E. Laroche, D. Knittel, An improved linear fractional model for robustness analysis of a winding system, Control Eng. Pract. 13 (2005) 659–666.
[2] A. Calderon, B. Vinagre, V. Feliu, Fractional order control strategies for power electronic buck converters, Signal Process. 86 (2006) 2803–2819.
[3] J. Sabatier, M. Aoun, A. Oustaloup, G. Grgoire, F. Ragot, P. Roy, Fractional system identification for lead acid battery state of charge estimation, Signal
Process. 86 (2006) 2645–2657.
[4] V. Feliu-Batlle, R. Perez, L. Rodriguez, Fractional robust control of main irrigation canals with variable dynamic parameters, Control Eng. Pract. 15
(2007) 673–686.
[5] B. Vinagre, C. Monje, A. Calderon, J. Suarez, Fractional PID controllers for industry application. A brief introduction, J. Vib. Control 13 (2007) 1419–1430.
[6] C. Monje, B. Vinagre, V. Feliu, Y. Chen, Tuning and auto-tuning of fractional order controllers for industry applications, Control Eng. Pract. 16 (2008)
798–812.
[7] M. Tavazoei, M. Haeri, Chaos control via a simple fractional-order controller, Phys. Lett. A 372 (2008) 798–807.
[8] M. Tavazoei, M. Haeri, S. Jafari, S. Bolouki, M. Siami, Some applications of fractional calculus in suppression of chaotic oscillations, IEEE Trans. Ind.
Electron. 55 (2008) 4094–4101.
[9] N.T. Shawagfeh, Analytical approximate solutions for nonlinear fractional differential equations, Appl. Math. Comput. 131 (2002) 517–529.
[10] S. Das, Analytical solution of a fractional diffusion equation by variational iteration method, Comput. Math. Appl. 57 (2009) 483–487.
[11] A. Arikoglu, I. Ozkol, Solution of fractional differential equations by using differential transform method, Chaos Solitons Fractals 34 (2007) 1473–1481.
[12] A. Saadatmandi, M. Dehghan, A new operational matrix for solving fractional order differential equations, Comput. Math. Appl. 59 (2010) 1326–1336.
[13] M. Dehghan, J. Manafian, A. Saadatmandi, Solving nonlinear fractional partial differential equations using the homotopy analysis method, Numer.
Methods Partial Differential Equations 26 (2010) 448–479.
[14] M. Dehghan, J. Manafian Herris, A. Saadatmandi, The Solution of the linear fractional partial differential equations using the homotopy analysis method,
Z. Naturforsch. 65a (2010) 935–949.
[15] S. Momani, Z. Odibat, Numerical comparison of methods for solving linear differential equations of fractional order, Chaos Solitons Fractals 31 (2007)
1248–1255.
[16] Z. Odibat, N. Shawagfeh, Generalized Taylor’s formula, Appl. Math. Comput. 186 (2007) 286–293.
[17] I. Podlubny, The Laplace transform method for linear differential equations of the fractional order, 1997, eprint arXiv:funct-an/9710005.
[18] C. Chen, Y. Tsay, T. Wu, Walsh operational matrices for fractional calculus and their application to distributed systems, J. Franklin Inst. 303 (1977)
267–284.
[19] C. Hwang, Y. Shih, Laguerre operational matrices for fractional calculus and applications, Internat. J. Control 34 (1981) 577–584.
[20] G. Maione, Inverting fractional order transfer functions through Laguerre approximation, Systems Control Lett. 52 (2004) 387–393.
[21] G. Maione, A digital, noninteger order, differentiator using laguerre orthogonal sequences, Int. J. Intell. Syst. 11 (2006) 77–81.
[22] C. Chen, C. Hsiao, A state-space approach to Walsh series solution of linear systems, Int. J. Syst. Sci. 6 (1975) 833–858.
[23] W. Glabisz, Direct Walsh-wavelet packet method for variational problems, Appl. Math. Comput. 159 (2004) 769–781.
[24] H. Chyi, S. Yen-Ping, Solution of population balance equations via block pulse functions, Chem. Eng. J. 25 (1982) 39–45.
1054 Y. Li, N. Sun / Computers and Mathematics with Applications 62 (2011) 1046–1054

[25] A. Deb, G. Sarkar, M. Bhattacharjee, S.K. Sen, All-integrator approach to linear SISO control system analysis using block pulse functions (BPF), J. Franklin
Inst. 334 (1997) 319–335.
[26] K. Maleknejad, M. Shahrezaee, H. Khatami, Numerical solution of integral equations system of the second kind by block-pulse functions, Appl. Math.
Comput. 166 (2005) 15–24.
[27] E. Babolian, Z. Masouri, Direct method to solve Volterra integral equation of the first kind using operational matrix with block-pulse functions, Comput.
Math. Appl. 220 (2008) 51–57.
[28] C. Hwang, M. Chen, Analysis and optimal control of time-varying linear systems via shifted Legendre polynomials, Internat. J. Control 41 (1985)
1317–1330.
[29] P. Paraskevopoulos, Legendre series approach to identification and analysis of linear systems, IEEE Trans. Automat. Control 30 (1985) 585–589.
[30] X.T. Wang, Numerical solutions of optimal control for time delay systems by hybrid of block-pulse functions and Legendre polynomials, Appl. Math.
Comput. 184 (2007) 849–856.
[31] K. Maleknejad, S. Sohrabi, Y. Rostami, Numerical solution of nonlinear Volterra integral equations of the second kind by using Chebyshev polynomials,
Appl. Math. Comput. 188 (2007) 123–128.
[32] C. Hwang, Y. Shih, Parameter identification via Laguerre polynomials, Int. J. Syst. Sci. 13 (1982) 209–217.
[33] S. Zaman, A. Jha, Parameter identification of non-linear systems using Laguerre operational matrices, Int. J. Syst. Sci. 16 (1985) 625–631.
[34] M. Razzaghi, S.D. Lin, Identification of time-varying linear and bilinear systems via Fourier series, Comput. Electron. Eng. 17 (1991) 237–244.
[35] M. Razzaghi, A. Arabshahi, S.D. Lin, Identification of nonlinear differential equations via Fourier series operational matrix for repeated integration,
Appl. Math. Comput. 68 (1995) 189–198.
[36] A. Deb, A. Dasgupta, G. Sarkar, A new set of orthogonal functions and its application to the analysis of dynamic systems, J. Franklin Inst. 343 (2006)
1–26.
[37] A. Deb, G. Sarkar, S. Sen, Block pulse functions, the most fundamental of all piecewise constant basis functions, Int. J. Syst. Sci. 25 (1994) 351–363.
[38] A. El-Mesiry, A. El-Sayed, H. El-Saka, Numerical methods for multiterm fractional (arbitrary) orders differential equations, Appl. Math. Comput. 160
(2005) 683–699.