Applied Mathematical Modelling 35 (2011) 5662–5672

Contents lists available at ScienceDirect

Applied Mathematical Modelling

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

Efficient Chebyshev spectral methods for solving multi-term fractional

orders differential equations
E.H. Doha a, A.H. Bhrawy b,⇑, S.S. Ezz-Eldien c
a
Department of Mathematics, Faculty of Science, Cairo University, Giza, Egypt
b
Department of Mathematics, Faculty of Science, King Abdulaziz University, Jeddah, Saudi Arabia
c
Department of Basic Science, Institute of Information Technology, Modern Academy, Cairo, Egypt

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, we state and prove a new formula expressing explicitly the derivatives of
Received 29 December 2010 shifted Chebyshev polynomials of any degree and for any fractional-order in terms of
Received in revised form 29 April 2011 shifted Chebyshev polynomials themselves. We develop also a direct solution technique
Accepted 8 May 2011
for solving the linear multi-order fractional differential equations (FDEs) with constant
Available online 20 May 2011
coefficients using a spectral tau method. The spatial approximation with its fractional-
order derivatives (described in the Caputo sense) are based on shifted Chebyshev polyno-
Keywords:
mials TL,n(x) with x 2 (0, L), L > 0 and n is the polynomial degree. We presented a shifted
Multi-term fractional differential equations
Nonlinear fractional differential equations
Chebyshev collocation method with shifted Chebyshev–Gauss points used as collocation
Tau method nodes for solving nonlinear multi-order fractional initial value problems. Several numerical
Collocation method examples are considered aiming to demonstrate the validity and applicability of the pro-
Shifted Chebyshev polynomials posed techniques and to compare with the existing results.
Gauss quadrature Ó 2011 Elsevier Inc. All rights reserved.

1. Introduction

Fractional differential equations (FDEs) appeared in many fields of science and engineering [1–4]. The analytic results on
the existence and uniqueness of solutions to the fractional differential equations have been investigated by many authors;
among them [4–7]. In general, most fractional differential equations do not have exact analytic solutions, so approximation
and numerical techniques must be used.
Finding accurate and efficient methods for solving FDEs has become an active research undertaking. There are several
analytic methods [8–11], which have been developed to solve the fractional differential equations. In the last decade or
so, extensive research has been carried out on the development of numerical methods which are numerically stable for both
linear and nonlinear fractional differential equations. Diethelm et al. [12] presented Predictor corrector method for numer-
ical solution of FDEs. In [13], the authors have proposed an approximate method for numerical solution of a class of FDEs
which are expressed in terms of Caputo type fractional derivative. In fact, the method presented in [13] takes advantage
of FDEs converting into Volterra-integral equations. Furthermore, the generalization of the Legendre operational matrix to
the fractional calculus has been studied in [14]. In [15,16], the spectral tau method for numerical solution of some FDEs
is introduced. Recently, Esmaeili and Shamsi [17] introduced a direct solution technique for obtaining the spectral solution
of a special family of fractional initial value problems using a pseudo-spectral method, and in [18] Pedas and Tamme devel-
oped the spline collocation methods for solving FDEs.

⇑ Corresponding author. Permanent address: Department of Mathematics, Faculty of Science, Beni-Suef University, Beni-Suef, Egypt.
E-mail addresses: eiddoha@frcu.eun.eg (E.H. Doha), alibhrawy@yahoo.co.uk (A.H. Bhrawy), s_sezeldien@yahoo.com (S.S. Ezz-Eldien).

0307-904X/$ - see front matter Ó 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.apm.2011.05.011
 E.H. Doha et al. / Applied Mathematical Modelling 35 (2011) 5662–5672 5663

In [19–21] the authors have presented spectral methods for solving boundary value problems. The main advantage of
spectral methods lies in their accuracy for a given number of unknowns. For smooth problems in simple geometries, they
offer exponential rates of convergence/spectral accuracy.
In the present paper we extend the application of spectral methods to solve linear and nonlinear fractional initial value
problems. Our main aim is to develop some efficient spectral algorithms based on shifted Chebyshev tau (SCT) method to
approximate the linear multi-order fractional initial value problems such that they can be implemented efficiently. Dealing
with the nonlinear multi-order fractional initial value problems on the interval (0, L), we propose a spectral shifted Cheby-
shev collocation (SCC) method to find the solution uN(x). The nonlinear FDE is collocated only at the (N  m + 1) points. For
suitable collocation points we use the (N  m + 1) nodes of the shifted Chebyshev–Gauss interpolation on (0, L). These equa-
tions together with m initial conditions generate (N + 1) nonlinear algebraic equations which can then be solved using New-
ton’s iterative method. Finally, the accuracy of the proposed algorithms are demonstrated by test problems.
The paper is organized as follows. In Section 2 we introduce some necessary definitions and give some relevant properties
of Chebyshev polynomials. In Section 3 we state and prove the main result of the paper which give explicitly a formula that
expresses the fractional-order derivatives of the Chebyshev polynomials in terms of the shifted Chebyshev polynomials
themselves. In Section 4 we construct and develop algorithms for solving linear and nonlinear multi-order FDEs by using
tau and collocation spectral methods. In Section 5 the proposed methods are applied to several examples. Also a conclusion
is given in Section 6.

2. Preliminaries and notation

2.1. The fractional derivative in the Caputo sense

The Riemann–Liouville fractional integration of order m is defined as

Rx
J m f ðxÞ ¼ C1ðmÞ 0
ðx  tÞm1 f ðtÞdt; m > 0; x > 0;
ð2:1Þ
0
J f ðxÞ ¼ f ðxÞ:
One of the basic property of the operator Jm is
Cðb þ 1Þ
J m xb ¼ xbþm : ð2:2Þ
Cðb þ 1 þ mÞ
The next equation defines Caputo fractional derivative of order m
 m 
d
Dm f ðxÞ ¼ J mm m f ðxÞ ; ð2:3Þ
dx
where m  1 < m 6 m, m 2 N and m is the smallest integer greater than m (see [22,23]).
The Caputo’s fractional differentiation is a linear operation
Dm ðkf ðxÞ þ lgðxÞÞ ¼ kDm f ðxÞ þ lDm gðxÞ; ð2:4Þ
where k and l are constants. For the Caputo derivative we have
Dm C ¼ 0; ðC is a constantÞ; ð2:5Þ
(
0; for b 2 N0 and b < dme;
D m xb ¼ Cðbþ1Þ bm ð2:6Þ
Cðbþ1mÞ x ; forb 2 N0 and b P dme or b R N and b > bmc:

We use the ceiling function dme to denote the smallest integer greater than or equal to m, and the floor function bmc to denote
the largest integer less than or equal to m. Also N = {1, 2, . . . } and N0 = {0, 1, 2, . . . }.

2.2. Properties of shifted Chebyshev polynomials

The well-known Chebyshev polynomials are defined on the interval (1, 1) and can be determined with the aid of the
following recurrence formula:
T iþ1 ðtÞ ¼ 2tT i ðtÞ  T i1 ðtÞ; i ¼ 1; 2; . . . ;
where T0(t) = 1 and T1(t) = t. In order to use these polynomials on the interval x 2 (0, L) we defined the so-called shifted
 
Chebyshev polynomials by introducing the change of variable t ¼ 2x L
 1. Let the shifted Chebyshev polynomials T i 2x
L
1
be denoted by TL,i(x). Then TL,i(x) can be generated by using the following recurrence relation:
 
2x
T L;iþ1 ðxÞ ¼ 2  1 T L;i ðxÞ  T L;i1 ðxÞ; i ¼ 1; 2; . . . ; ð2:7Þ
L
5664 E.H. Doha et al. / Applied Mathematical Modelling 35 (2011) 5662–5672

where TL,0(x) = 1 and T L;1 ðxÞ ¼ 2x

L
 1. The analytic form of the shifted Chebyshev polynomials TL,i(x) of degree i is given by

Xi
ði þ k  1Þ!22k k
T L;i ðxÞ ¼ i ð1Þik x; ð2:8Þ
k¼0 ði  kÞ!ð2kÞ!Lk

where TL,i(0) = (1)i and TL,i(L) = 1.

The orthogonality condition is
Z L
T L;j ðxÞT L;k ðxÞwL ðxÞdx ¼ djk hk ; ð2:9Þ
0

where wL ðxÞ ¼ pffiffiffiffiffiffiffiffiffi

1
Lxx2
and hk ¼ c2k p; with c0 ¼ 2; ci ¼ 1; i P 1:
Any function u(x), square integrable in (0, L), may be expressed in terms of shifted Chebyshev polynomials as
X
1
uðxÞ ¼ aj T L;j ðxÞ;
j¼0

where the coefficients aj are given by

Z L
1
aj ¼ uðxÞT L;j ðxÞwL ðxÞdx; j ¼ 0; 1; 2; . . . ð2:10Þ
hj 0

In practice, only the first (N + 1)-terms shifted Chebyshev polynomials are considered. Hence we can write
X
N
uN ðxÞ ’ aj T L;j ðxÞ: ð2:11Þ
j¼0

The special values

ðqÞ iði þ q  1Þ! pffiffiffiffi

T L;i ð0Þ ¼ ð1ÞðiqÞ p; q 6 i; ð2:12Þ
Cðq þ 12Þði  qÞ!Lq
will be of important use later.
Now, we turn to the Chebyshev–Gauss interpolation. We denote by xN,j, 0 6 j 6 N, the nodes of the standard Chebyshev–
Gauss interpolation on the interval (1, 1). The corresponding Christoffel numbers are -N,j, 0 6 j 6 N. The nodes of the
shifted Chebyshev–Gauss interpolation on the interval (0, L) are the zeros of TL,N+1(x), which are denoted by xL,N,j, 0 6 j 6 N.
Clearly xL;N;j ¼ 2L ðxN;j þ 1Þ. The corresponding Christoffel numbers are -L,N,j = -N,j, 0 6 j 6 N. Let SN(0, L) be the set of all poly-
nomials of degree at most N. Thanks to the property of the standard Chebyshev–Gauss quadrature, it follows that for any /
2 S2N+1(0, L),
Z L Z 1   XN   X N
1 L L
wL ðxÞ/ðxÞdx ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi / ðx þ 1Þ dx ¼ -N;j / ðxN;j þ 1Þ ¼ -L;N;j /ðxL;N;j Þ: ð2:13Þ
0 1 1  x2 2 j¼0
2 j¼0

3. The fractional derivatives of TL,i(x)

The main objective of this section is to prove the following theorem for the fractional derivatives of the shifted Chebyshev
polynomials.

Lemma 3.1. Let TL,i(x) be a shifted Chebyshev polynomial then

Dm T L;i ðxÞ ¼ 0; i ¼ 0; 1; . . . ; dme  1; m > 0: ð3:1Þ

Proof. This lemma can be easily proved by making use of relations (2.4)–(2.6) with relation (2.8).

Theorem 3.2. The fractional derivative of order m in the Caputo sense for the Chebyshev polynomials is given by
X
1
Dm T L;i ðxÞ ¼ Sm ði; jÞT L;j ðxÞ; i ¼ dme; dme þ 1; . . . ; ð3:2Þ
j¼0

where
X
i
ð1Þik 2iði þ k  1Þ!Cðk  m þ 12Þ
Sm ði; jÞ ¼ m  :
k¼dme
cj L C k þ 12 ði  kÞ!Cðk  m  j þ 1ÞCðk þ j  m þ 1Þ
 E.H. Doha et al. / Applied Mathematical Modelling 35 (2011) 5662–5672 5665

Proof. The analytic form of the shifted Chebyshev polynomials TL,i(x) of degree i is given by (2.8). Using Eqs. (2.5), (2.6) and
(2.8) we have

Xi
ði þ k  1Þ!22k m k X i
ði þ k  1Þ!22k k!
Dm T L;i ðxÞ ¼ i ð1Þik k
D x ¼i ð1Þik xkm ; i ¼ dme; dme þ 1; . . . ð3:3Þ
k¼0 ði  kÞ!ð2kÞ!L k¼dme ði  kÞ!ð2kÞ!Lk Cðk  m þ 1Þ

Now, xkm may be expressed in terms of shifted Chebyshev series, so we have

X
1
xkm ¼ bkj T L;j ðxÞ; ð3:4Þ
j¼0

where bkj is given from (2.10) with u(x) = xkm, and

8
> Lkm Cðkmþ12Þ
< pffiffipffi
> 1
Cðkmþ1Þ ; j ¼ 0;
bkj ¼ ð3:5Þ
jLkm P
j
>
> ðjþr1Þ!22rþ1 Cðkþrmþ1Þ
ð1Þjr ðjrÞ!ð2rÞ!Cðkþrmþ1Þ2 ;
: pffiffipffi j ¼ 1; 2; . . .
r¼0

Employing Eqs. (3.3)–(3.5) we get

X
1
Dm T L;i ðxÞ ¼ Sm ði; jÞT L;j ðxÞ; i ¼ dme; dme þ 1; . . . ; ð3:6Þ
j¼0
Pi
where Sm ði; jÞ ¼ k¼dme hijk ; and
8
> ið1Þik ðiþk1Þ!22k k!Cðkmþ12Þ
>
> Lm ðikÞ!ð2kÞ!pffiffipffiðCðkmþ1ÞÞ2 ;
>
>
j ¼ 0;
>
< ð1Þik ijðiþk1Þ!22kþ1 k!
hijk ¼ Lm ðikÞ!ð2kÞ!Cðkmþ1Þpffiffipffi
>
>
>
> Pj
ð1Þjr ðjþr1Þ!22r Cðkþrmþ12Þ
>
> ; j ¼ 1; 2; . . .
: ðjrÞ!ð2rÞ!Cðkþrmþ1Þ
r¼0

After some lengthly manipulation, hi,j,k may be put in the form

ð1Þik 2i ði þ k  1Þ! Cðk  m þ 12Þ

hijk ¼ ; j ¼ 0; 1; . . . ð3:7Þ
cj Lm Cðk þ 12Þ ði  kÞ! Cðk  m  j þ 1Þ Cðk þ j  m þ 1Þ
Eq. (3.7) leads to the desired result. h

4. Spectral methods for FDEs

In order to show the fundamental importance of the formula we proved in the last section, we apply it for solving multi-
order linear FDEs with SCT method. Also we proposed the SCC method for solving multi-order nonlinear FDEs. For the exis-
tence and uniqueness and continuous dependence of the solution of the problem, see [24].

4.1. Linear multi-order FDE

We are interested in using the SCT method to solve the linear multi-order FDE
X
r1
Dm uðxÞ þ ci Dbi uðxÞ þ cr uðxÞ ¼ f ðxÞ; in I ¼ ð0; LÞ;
i¼1 ð4:1Þ
ðiÞ
u ð0Þ ¼ di ; i ¼ 0; . . . ; m  1:
where ci (i = 1, . . . , r) and 0 < b1 < b2 <    < br1 < m, m  1 < m 6 m are constants. Moreover, Dmu(x)  u(m)(x) denotes the Caputo
fractional derivative of order m for u(x) and the values of di (i = 0, . . . , m  1) describe the initial state of u(x) and f(x) is a given
source function.
Let wL ðxÞ ¼ pffiffiffiffiffiffiffiffiffi
1
Lxx2
, then we denote by L2wL ðIÞðI :¼ ð0; LÞÞ the weighted L2 space with inner product:
Z
ðu; v ÞwL ¼ wL ðxÞuðxÞv ðxÞdx;
I

and the associated norm kukwL ¼ ðu; uÞ1=2 2

wL . It is known that {TL,n(x) : n P 0} forms a complete orthogonal system in LwL ðIÞ.
Hence, if we define
 
SN ðIÞ ¼ span T L;0 ðxÞ; T L;1 ðxÞ; . . . ; T L;N ðxÞ ; ð4:2Þ
5666 E.H. Doha et al. / Applied Mathematical Modelling 35 (2011) 5662–5672

then the standard shifted Chebyshev tau approximation to (4.1) is to find uN 2 SN(I) such that

X
r1
ðDm uN ; T L;k ðxÞÞwL þ ci ðDbi uN ; T L;k ðxÞÞwL þ cr ðuN ; T L;k ðxÞÞwL
i¼1
ð4:3Þ
¼ ðf ; T L;k ðxÞÞwL ; k ¼ 0; 1; . . . ; N  m;
ðiÞ
uN ð0Þ ¼ di ; i ¼ 0; 1; . . . ; m  1:
Let us denote

X
N
uN ðxÞ ¼ aj T L;j ðxÞ; a ¼ ða0 ; a1 ; . . . ; aN ÞT ;
j¼0
ð4:4Þ
fk ¼ ðf ; T L;k ðxÞÞwL ; k ¼ 0; 1; . . . ; N  m;
f ¼ ðf0 ; f1 ; . . . ; fNm ; d0 ; . . . ; dm1 ÞT ;
then Eq. (4.3) can be written as
"
X
N X
r1
aj ðDm T L;j ðxÞ; T L;k ðxÞÞwL þ ci ðDbi T L;j ðxÞ; T L;k ðxÞÞwL
j¼0 i¼1
i
þcr ðT L;j ðxÞ; T L;k ðxÞÞwL ¼ ðf ; T L;k ðxÞÞwL ; k ¼ 0; 1; . . . ; N  m; ð4:5Þ
X
N
ðkNþm1Þ
aj T L;j ð0Þ ¼ dkNþm1 ; k ¼ N  m þ 1; N  m þ 2; . . . ; N:
j¼0

Let us denote
s
A ¼ ðakj Þ0<k;j<N ; Bs ¼ ðbkj Þ0<k;j<N;s¼1;2;...;r1 ; C ¼ ðckj Þ0<k;j<N ;

where
(
ðDm T L;j ðxÞ; T L;k ðxÞÞwL ; k ¼ 0; 1; . . . ; N  m; j ¼ 0; 1; . . . ; N;
akj ¼
DkNþm1 T L;j ð0Þ; k ¼ N  m þ 1; . . . ; N; j ¼ 0; 1; . . . ; N;
(
s ðDbs T L;j ðxÞ; T L;k ðxÞÞwL ; k ¼ 0; 1; . . . ; N  m; j ¼ 0; 1; . . . ; N; s ¼ 1; 2; . . . ; r  1;
bkj ¼
0; otherwise;
(
ðT L;j ðxÞ; T L;k ðxÞÞwL ; k ¼ 0; 1; . . . ; N  m; j ¼ 0; 1; . . . ; N;
ckj ¼
0; otherwise;

then by virtue of (2.12), (3.2) and making use of the orthogonality relation of shifted Chebyshev polynomials (2.9), and after
s
some lengthly manipulation, one can show that the nonzero elements of akj ; bkj ; s ¼ 1; 2; . . . ; r  1; and ckj are given explic-
itly in the following form
8
< hk Sm ðj; kÞ; k ¼ 0; 1; . . . ; N  m; j ¼ m; . . . ; N;
pffiffiffi
akj ¼ jðjkþNmþ2Þ2ðkNþmÞ3 p
: ð1ÞjkþNmþ1 CðkNþm12ÞLjkþNmþ1
; k ¼ N  m þ 1; . . . ; N; j ¼ 0; 1; . . . ; N;

s hk Sbs ðj; kÞ; k ¼ 0; 1; 2; . . . ; N  m; j ¼ dbs e; dbs e þ 1; . . . ; N; s ¼ 1; 2; . . . ; r  1;

bkj ¼
0; otherwise;

hk ; k ¼ j ¼ 0; 1; . . . ; N  m;
ckj ¼
0; otherwise:
Thereby, we can write Eq. (4.5) in the following matrix system form

X
r1
ðA þ ci Bi þ cr CÞa ¼ f: ð4:6Þ
i¼1

In the case of ci – 0, i = 1, . . . , r, the linear system (4.6), can be solved by forming explicitly the LU factorization; i.e.,
Pr1 i
A þ i¼1 ci B þ cr C ¼ LU: The expense of calculating LU factorization is O(N3) operations and the expense of solving the linear
system (4.6), provided that the factorization is known, is O(N2).
 E.H. Doha et al. / Applied Mathematical Modelling 35 (2011) 5662–5672 5667

4.2. Nonlinear multi-order FDE

In this section, we use the SCC method to numerically solve the nonlinear multi-order FDE, namely

Dm uðxÞ ¼ F x; uðxÞ; Db1 uðxÞ; . . . ; Dbk uðxÞ ; x 2 I; ð4:7Þ

with initial conditions

uðiÞ ð0Þ ¼ di ; i ¼ 0; 1; . . . ; m  1; ð4:8Þ

where m  1 < m 6 m, 0 < b1 < b2 <    < bk < m. It is to be noted here that F can be nonlinear in general.
In fact, Diethelm and Ford in [24] have proved that the nonlinear equation (4.7) subject to the initial conditions (4.8) has
(under natural Lipschitz conditions imposed on F) a unique continuous solution. For more detail about the theorems of the
existence and uniqueness of (4.7) subject to (4.8), see [24, pp. 627–633].
In order to use the shifted Chebyshev collocation method to solve Eq. (4.7) subject to (4.8). Let
X
N
uN ðxÞ ¼ aj T L;j ðxÞ; ð4:9Þ
j¼0

then, making use of formula (3.2) enables one to express explicitly the derivatives Dm uðxÞ; Db1 uðxÞ; . . . ; Dbk uðxÞ in terms of the
expansion coefficients aj. The criterion of spectral shifted Chebyshev collocation method for solving approximately (4.7),
(4.8) is to find uN(x) 2 SN(0, L) such that

Dm uN ðxÞ ¼ F x; uN ðxÞ; Db1 uN ðxÞ; . . . ; Dbk uN ðxÞ ð4:10Þ

is satisfied exactly at the collocation points xL,Nm+1,k, k = 0, 1, . . . , N  m. In other words, we have to collocate Eq. (4.10) at the
(N  m + 1) shifted Chebyshev roots xL,Nm+1,k, which immediately yields
!
X
N
m
X
N X
N
b1
X
N
bk
aj D T L;j ðxL;Nmþ1;k Þ ¼ F xL;Nmþ1;k ; aj T L;j ðxL;Nmþ1;k Þ; aj D T L;j ðxL;Nmþ1;k Þ;    ; aj D T L;j ðxL;Nmþ1;k Þ ;
j¼0 j¼0 j¼0 j¼0

k ¼ 0; 1; . . . ; N  m; ð4:11Þ
with (4.8) written in the form
X
N
ðiÞ
aj T L;j ð0Þ ¼ di ; i ¼ 0; 1; 2; . . . ; m  1: ð4:12Þ
j¼0

This constitute a system of (N + 1) nonlinear algebraic equations in the unknown expansion coefficients aj (j = 0, 1, . . . , N),
which can be solved by using any standard iteration technique, like Newton’s iteration method.

5. Numerical results

To illustrate the effectiveness of the proposed methods in the present paper, several test examples are carried out. Com-
parisons of the results obtained with those obtained by other methods reveal that our methods is very effective and
convenient.

Example 1. Consider the equation

m
aDm uðxÞ þ bD 2 uðxÞ þ cDm1 þ euðxÞ ¼ f ðxÞ; ð5:1Þ

0 < m1 6 1; 1 < m2 6 2; 3 < m 6 4;

and
2b 2c c
f ðxÞ ¼ x2m2 þ x2m1  x1m1 þ eðx2  xÞ;
Cð3  m2 Þ Cð3  m1 Þ Cð2  m1 Þ
subject to
uð0Þ ¼ 0; u0 ð0Þ ¼ 1; u00 ð0Þ ¼ 2; u000 ð0Þ ¼ 0: ð5:2Þ

The analytic solution of this problem is u(x) = x2  x. El-Mesiry et al. [25] applied the Euler’s method in Eqs. (5.1), (5.2),
and then replaced the integrals by the product trapezoidal quadrature formula (ET). Also, they applied Euler’s method in Eqs.
(5.1), (5.2), then replaced the integrals by using the product rectangle rule (ER). Regarding problem (5.1), (5.2), we study two
different choices of a, b, c, e, k, m1, m2 and m.
5668 E.H. Doha et al. / Applied Mathematical Modelling 35 (2011) 5662–5672

Table 5.1
Maximum absolute error using SCT for Case I.

N 4 8 16 32 48 64
SCT 1.0  103 5.6  104 6.9  105 1.1  105 3.9  106 1.9  106

Table 5.2
Maximum absolute error using SCT for Case II.

N 4 8 16 32 48 64
4 4 5 6 7
SCT 6  1.10 3  0.10 2.0  10 1.8  10 4.8  10 1.8  107

Case I: In the case of a = 1, b = 1, c = 1, e = 1, m1 = 0.77, m2 = 1.44 and m = 3.91, the maximum absolute errors achieved in
[25], with 1000 steps, are 9.98  104 and 9.53  104 by using ET and ER methods respectively. While the maximum abso-
lute error using SCT method, with the same choices pofffiffi a, b, c, p e,ffiffiffim, m1, m2 and various choices of N, are presented in Table 5.1.
pffiffiffiffiffiffi
Case II: For a ¼ 1; b ¼ 1; c ¼ 0:5; e ¼ 0:5; m1 ¼ 202 ; m2 ¼ 2 and m ¼ 11; the maximum absolute errors in [25], with
1000 steps, are 9.95  104 and 9.80  104 using ET and ER methods, respectively. While the maximum absolute error,
using SCT method, with the same choices of a, b, c, e, k, m1, m2 and various choices of N, are presented in Table 5.2. Numerical
results of this problem show that the SCT method is more accurate than the ET and ER methods [25]. see Tables 3 and 4 in
[25].

Example 2. Consider the equation, see [26]

3 8:533333333 2:25
D2 uðxÞ þ D4 uðxÞ þ uðxÞ ¼ x3 þ 6x þ x ; uð0Þ ¼ 0; u0 ð0Þ ¼ 0; ð5:3Þ
Cð0:25Þ
whose exact solution is given by u(x) = x3.
Ford and Connolly [26] introduced three alternative decomposition methods for the approximate solution of Eq. (5.3)
using the Caputo form of the fractional derivative. Methods 1 and 2 produce different systems of equations. In the case of
method 1, the system may be of quite high dimension. Method 2 keeps the dimension of the system reasonably small
and independent of the precise orders in the equation. However, method 3 decomposes the multi-term equation into a sys-
tem of fractional equations of varying orders.
Regarding problem (5.3), and in [26] the best result is achieved with 512 steps and the maximum absolute errors are
6.93  105, 1.18  104 and 3.10  106 by using method 1, method 2 and method 3, respectively.
In Tables 5.3 we introduce maximum absolute error, using SCT method with various choices of N. Moreover, ln of the
L2wL , L1 1
wL and HwL - errors using the SCT method with various choices of N have been plotted in Fig. 1. Numerical results of this
fractional-order differential equation show that the SCT method is more accurate than the three decomposition methods, see
Table 3 in [26].

Example 3. We consider the following initial value problem in the case of the inhomogeneous Bagley–Torvik equation of the
form

uð2Þ ðxÞ þ uð2Þ ðxÞ þ uðxÞ ¼ f ðxÞ;

3
in I ¼ ð0; 1Þ; uð0Þ ¼ 0; u0 ð0Þ ¼ a; ð5:4Þ
with an exact solution u(x) = sin(ax).
Table 5.4 lists the maximum absolute error, using the SCT method with various choices of a and N.

Example 4. Consider the initial value problem

uð2Þ ðxÞ  3uð3Þ ðxÞ ¼ f ðxÞ;

5 2
in I ¼ ð0; 1Þ; uð0Þ ¼ 1; u0 ð0Þ ¼ c; u00 ð0Þ ¼ c2 ; ð5:5Þ
with an exact solution u(x) = ecx.

Table 5.3
Maximum absolute error with various choices of N, for Example 2.

N SCT N SCT
4 8.82  106 28 7.39  1011
8 1.91  107 32 3.16  1011
12 1.52  108 40 7.58  1012
16 2.52  109 48 2.32  1012
20 6.22  1010 56 8.35  1013
24 1.96  1010 64 3.39  1013
 E.H. Doha et al. / Applied Mathematical Modelling 35 (2011) 5662–5672 5669

22

20

wL
1
error
18

error,L wL errorandH
16

14
L2wL error

L
12
Lw
2
Ln LwL error

10 H1wL error

4 8 12 16 20 24
N Nfrom4to24

Fig. 1. L2wL , L1 1
wL and HwL of SCT method for Example 2.

Table 5.4
Maximum absolute error using SCT method, for Example 3.

N a SCT a SCT
4 1 3.4  104 4p 3.9  100
8 4.3  107 4.7  101
16 1.8  108 3.5  105
32 7.1  1010 1.4  106
48 9.9  1011 1.9  107
64 2.4  1011 4.8  108

Table 5.5
Maximum absolute error using SCT method, for Example 4.

N a SCT a SCT
8 1 6.2  103 6 1.8  102
16 7.5  104 8.4  103
32 1.0  104 7.5  104
48 3.5  105 1.8  104
64 1.6  105 6.5  105

Table 5.5 lists the maximum absolute error, using the SCT method with various choices of c and N.

Example 5. In this example we consider the following nonlinear initial value problem [27,28]

Da uðxÞ ¼ 1 þ uðxÞ2 ; p  1 < a 6 p; ð5:6Þ

ðiÞ
u ð0Þ ¼ 0; i ¼ 0; 1; . . . ; p  1: ð5:7Þ
The exact solution of the initial value problem (5.6), (5.7) for a = 1, i.e. the ODE case, is u(x) = tan (x).

Table 5.6
Approximate solution of (5.6), (5.7) for some values of a using N = 24 for Example 5.

x a = 0.5 a = 1.5 a = 3.5 a = 5.5

15 17 18
0.0 2.01488131  10 2.48756635  10 2. 33577937  10 1.20385247  1020
0.1 3.63409577  101 2.38287193  102 2.76526325  105 1.13518430  108
0.2 5.18146037  101 6.74166264  102 3.10122047  104 5.05150837  107
0.3 6.36998093  101 1.24064046  101 1.27830333  103 4.67241895  106
0.4 7.41575044  101 1.91715625  101 3.49420950  103 2.26750005  105
0.5 8.51143880  101 2.69244272  101 7.62388850  103 7.72431731  105
0.6 9.74812314  101 3.56528295  101 1.44234433  102 2.10327575  104
0.7 1.11896129  100 4.54300832  101 2.47295293  102 4.90654344  104
0.8 1.32738002  100 5.63467003  101 3.94514169  102 1.02208179  103
0.9 1.75904713  100 6.85559288  101 5.95664317  102 1.95269176  103
1.0 4.65181514  100 8.23079611  101 8.61156243  102 4.84555392  103
5670 E.H. Doha et al. / Applied Mathematical Modelling 35 (2011) 5662–5672

13
7.5 10

13
5 10

13
2.5 10

13
2.5 10

13
5 10

13
7.5 10

12
1 10
0 0.2 0.4 0.6 0.8 1

Fig. 2. Error for a = 1 at N = 24, for Example 5.

2
0.9

1.6 0.95

1.2 ux
uN x

1.5
0.8
2.5

3.5
0.4

0
0 0.2 0.4 0.6 0.8 1
x

Fig. 3. Comparison of uN(x) for N = 24 and a = 0.9, 0.95, 1, 1.5, 2.5, 3.5, and u(x) at a = 1 for Example 5.

Table 5.7
Absolute error using SCC method for N = 32 for Example 6.

x a = 2.1, c = 1.1, b = 0.1 a = 2.5, c = 1.99, b = 0.99 a = 3, c = 2, b = 1

0 4.44  1016 4.44  1016 0
0.2 3.79  108 1.01  106 6.93  1018
0.4 3.07  108 2.57  106 5.55  1017
0.6 4.77  108 4.45  106 2.77  1017
0.8 6.73  108 6.28  106 2.22  1016
1.0 7.78  108 7.60  106 0
1.2 7.23  108 8.11  106 0
1.4 4.70  108 7.80  106 4.44  1016
1.6 1.68  108 6.82  106 0
1.8 1.77  108 5.43  106 0
2.0 3.44  109 3.73  106 0

In Table 5.6 we introduce the approximate solutions for (5.6), (5.7) using SCC method at N = 24 and various choices of a
(a = 0.5, 1.5, 3.5, 5.5). Fig. 2 shows the error of analytical and numerical solution for N = 24 and a = 1. Furthermore the approx-
imate solutions for uN(x) for N = 24 and a = 0.9, 0.95, 1, 1.5, 2.5, 3.5, and u(x) at a = 1 are plotted in Fig. 3. Again we see that as a
approaches 1, the solution of the fractional differential equation approaches to that of the integer-order differential
equations.

Example 6. In this example we consider the following nonlinear initial value problem

6 36
Da uðxÞ þ Dc uðxÞDb uðxÞ þ u2 ðxÞ ¼ x3a þ x6bc þ x6 ; a 2 ð2; 3Þ; c 2 ð1; 2Þ;
Cð4  aÞ Cð4  cÞCð4  bÞ
b 2 ð0; 1Þ; x 2 ð0; 2Þ; ð5:8Þ
 E.H. Doha et al. / Applied Mathematical Modelling 35 (2011) 5662–5672 5671

Exact

u10 x
6

ux
4

0
0 0.5 1 1.5 2
x

Fig. 4. Comparison of the approximate solution uN(x) at N = 10 and a = 2.7, c = 1.7, b = 0.7, and u(x) for Example 6.

subject to the initial conditions

uð0Þ ¼ 0; u0 ð0Þ ¼ 0; u00 ð0Þ ¼ 0;

which has the following analytical solution:

uðxÞ ¼ x3 :
In Table 5.7, we list the absolute errors obtained by the shifted Chebyshev collocation method, with different values of a,
c, b and at N = 32.
In the case of a = 2.7, c = 1.7, b = 0.7, the approximate solution obtained by the present method at N = 10 is shown in Fig. 4
to make it easier to compare with the analytic solution.

6. Conclusion

An explicit expression for the fractional-order derivatives of the shifted Chebyshev polynomial of any degree in terms of
the shifted Chebyshev polynomials themselves has been derived. Also, we have presented some efficient direct solvers for
the general fractional-order differential equations by using the Chebyshev tau approximation. The fractional derivatives
are described in the Caputo sense because the Caputo fractional derivative allows traditional initial and boundary conditions
to be included in the formulation of the problem.
In this paper, we proposed a numerical algorithm to solve the general nonlinear fractional-order differential equations
using Gauss-collocation points and approximating directly the solution using the shifted Chebyshev polynomials. The
numerical results given in the previous section demonstrate the good accuracy of these algorithms. Moreover, The algo-
rithms introduced in this paper can be well suited for handling general linear and nonlinear fractional-order differential
equations with boundary conditions.
We note that similar techniques can be applied to tau and collocation methods using Legendre polynomials or other Ja-
cobi polynomials.

Acknowledgment

The authors are very grateful to the referees for carefully reading the paper and for their comments and suggestions
which have improved the paper.

References

[1] R.L. Magin, Fractional Calculus in Bioengineering, Begell House Publishers, 2006.
[2] M. Dalir, M. Bashour, Applications of fractional calculus, Appl. Math. Sci. 4 (2010) 1021–1032.
[3] S. Das, Functional Fractional Calculus for System Identification and Controls, Springer, New York, 2008.
[4] I. Podlubny, Fractional Differential Equations, Academic Press Inc, San Diego, CA, 1999.
[5] M. Amairi, M. Aoun, S. Najar, M.N. Abdelkrim, A constant enclosure method for validating existence and uniqueness of the solution of an initial value
problem for a fractional differential equation, Appl. Math. Comput. 217 (2010) 2162–2168.
[6] J. Deng, L. Ma, Existence and uniqueness of solutions of initial value problems for nonlinear fractional differential equations, Appl. Math. Lett. 23 (2010)
676–680.
[7] A.A. Kilbas, H.M. Srivastava, J.J. Trujillo, Theory and Applications of Fractional Differential Equations, Elsevier, San Diego, 2006.
[8] A.M.A. EL-Sayed, I.L. El-Kalla, E.A.A. Ziad, Analytical and numerical solutions of multi-term nonlinear fractional orders differential equations, Appl.
Numer. Math. 60 (2010) 788–797.
[9] S. Momani, Z. Odibat, Homotopy perturbation method for nonlinear partial differential equations of fractional order, Phys. Lett. A 365 (2007) 345–350.
5672 E.H. Doha et al. / Applied Mathematical Modelling 35 (2011) 5662–5672

[10] F. Dal, Application of variational iteration method to fractional Hyperbolic partial differential equations, Math. Problems Eng., doi:10.1155/2009/
824385.
[11] Z. Odibat, S. Momani, H. Xu, A reliable algorithm of homotopy analysis method for solving nonlinear fractional differential equations, Appl. Math.
Modell. 34 (2010) 593–600.
[12] K. Diethelm, N.J. Ford, A.D. Freed, A predictor–corrector approach for the numerical solution of fractional differential equations, Nonlinear Dynam. 29
(2002) 3–22.
[13] P. Kumer, O.P. Agrawal, An approximate method for numerical solution of fractional differential equations, Signal Process. 84 (2006) 2602–2610.
[14] A. Saadatmandi, M. Dehghan, A new operational matrix for solving fractional-order differential equations, Comput. Math. Appl. 59 (2010) 1326–1336.
[15] F. Ghoreishi, S. Yazdani, An extension of the spectral Tau method for numerical solution of multi-order fractional differential equations with
convergence analysis, Comput. Math. Appl. 61 (2011) 30–43.
[16] S.K. Vanani, A. Aminataei, Tau approximate solution of fractional partial differential equations, Comput. Math. Appl., doi:10.1016/
j.camwa.2011.03.013.
[17] S. Esmaeili, M. Shamsi, A pseudo-spectral scheme for the approximate solution of a family of fractional differential equations, Commun. Nonlinear Sci.
Numer. Simul. 16 (2011) 3646–3654.
[18] A. Pedas, E. Tamme, On the convergence of spline collocation methods for solving fractional differential equations, J. Comput. Appl. Math. 235 (2011)
3502–3514.
[19] E.H. Doha, W.M. Abd-Elhameed, A.H. Bhrawy, Efficient spectral ultraspherical-Galerkin algorithms for the direct solution of 2nth-order linear
differential equations, Appl. Math. Modell. 33 (2009) 1982–1996.
[20] E.H. Doha, A.H. Bhrawy, Efficient spectral-Galerkin algorithms for direct solution for second-order differential equations using Jacobi polynomials,
Numer. Algor. 42 (2006) 137–164.
[21] E.H. Doha, A.H. Bhrawy, Efficient spectral-Galerkin Algorithms for direct solution of fourth-order differential equations using Jacobi polynomials, Appl.
Numer. Math. 58 (2008) 1224–1244.
[22] K. Miller, B. Ross, An Introduction to the Fractional Calaulus and Fractional Differential Equations, John Wiley & Sons Inc., New York, 1993.
[23] M. Caputo, Linear models of dissipation whose Q is almost frequency independent, part II, J. Roy Austral. Soc. 13 (1967) 529–539.
[24] K. Diethelm, N.J. Ford, Multi-order fractional differential equations and their numerical solutions, Appl. Math. Comput. 154 (2004) 621–640.
[25] A.E.M. EL-Mesiry, A.M.A. EL-Sayed, H.A.A. EL-Saka, Numerical methods for multi-term fractional (arbitrary) orders differential equations, Appl. Math.
Comput. 160 (2005) 683–699.
[26] N.J. Ford, J.A. Connolly, Systems-based decomposition schemes for the approximate solution of multi-term fractional differential equations, J. Comput.
Appl. Math. 229 (2009) 382–391.
[27] O. Abdulaziz, I. Hashim, S. Momani, Application of homotopy-perturbation method to fractional IVPs, J. Comput. Appl. Math. 216 (2008) 574–584.
[28] I. Hashim, O. Abdulaziz, S. Momani, Homotopy analysis method for fractional IVPs, Commun. Nonlinear Sci. Numer. Simul. 14 (2009) 674–684.