ULTRASPHERICAL WAVELETS METHOD FOR SOLVING LANE-EMDEN

TYPE EQUATIONS

Y. H. YOUSSRI1 , W. M. ABD-ELHAMEED2,1 , E. H. DOHA1

1
Department of Mathematics, Faculty of Science, Cairo University, Giza 12613, Egypt
E-mail: youssri@sci.cu.edu.eg, eiddoha@frcu.eun.eg
2
Department of Mathematics, Faculty of Science, University of Jeddah, Jeddah, Saudi Arabia
E-mail: walee 9@yahoo.com
Received April 15, 2015

In this paper, a new shifted ultraspherical wavelets operational matrix of deriva-

tives is introduced. The two wavelets operational matrices, namely Legendre and first
kind Chebyshev operational matrices can be deduced as two special cases. Two numer-
ical algorithms based on employing the shifted ultraspherical wavelets operational ma-
trix of derivatives for solving linear and nonlinear differential equations of Lane-Emden
type are developed. The main idea for obtaining the presented algorithm is essentially
based on reducing the linear or nonlinear equations with their initial conditions to sys-
tems of linear or nonlinear algebraic equations, which can be efficiently solved. Some
numerical examples are given to demonstrate the validity and the applicability of the
algorithms.
Key words: Lane-Emden equations, ultraspherical polynomials, wavelets, oper-
ational matrix, spectral methods, collocation methods.
PACS: 02.60.Cb, 02.30.Mv, 02.70.Hm, 02.70.Jn, 02.30.Gp.

1. INTRODUCTION

Wavelets theory is a relatively new and an emerging area in mathematical re-

search. It has been applied to a wide range of engineering disciplines; particularly,
wavelets are very successfully used in signal analysis for wave form representation
and segmentations, time frequency analysis and fast algorithms for easy implemen-
tation. Wavelets permit the accurate representation of a variety of functions and
operators. Moreover, wavelets establish a connection with fast numerical algorithms,
(see [1, 2]).
The application of Legendre wavelets for solving differential and integral equations
is thoroughly considered by many authors (see, [3–8]). Also, Chebyshev wavelets
are used for solving some fractional and integral equations (see, [9–13]).
In recent years, the studies of singular initial value problems (IVPs) for second order
ordinary differential equations (ODEs) have attracted the attention of many math-
ematicians and physicists. One of the equations describing this type is the Lane-
Emden type equation which is formulated as
α
y 00 + y 0 + f (x, y) = g(x), 0 < x 6 1, α > 0, (1)
x
RJP 60(Nos.
Rom. 9-10),
Journ. Phys., Vol. 1298–1314
60, Nos. 9-10,(2015) (c) 2015
P. 1298–1314, - v.1.3a*2015.11.20
Bucharest, 2015
2 Ultraspherical wavelets method for solving Lane-Emden type equations 1299

subject to the initial conditions

y(0) = A, y 0 (0) = B, (2)
where A and B are constants, f (x, y) is a continuous real-valued function, and g(x) ∈
C[0, 1].
Eq. (1) was named after the astrophysicists Jonathan H. Lane and Robert Emden
(1870), as it was first studied by them. Eq. (1) was used to model several phenomena
in mathematical physics and astrophysics such as the theory of stellar structure, the
thermal behavior of a spherical cloud of gas, isothermal gas sphere, and theory of
thermionic currents.
Isothermal gas sphere equation is modeled by
2
y 00 + y 0 + ey = 0, 0 < x 6 1, (3)
x
subject to the initial conditions
y(0) = 0, y 0 (0) = 0. (4)

Singular IVPs of Lane-Emden type have been investigated by a large number of au-
thors. The solution of the Lane-Emden problem, as well as other various linear and
nonlinear singular initial value problems in quantum mechanics and astrophysics,
is numerically challenging because of the singularity behavior at the origin. The
approximate solutions to the Lane-Emden equation were given by many numeri-
cal techniques such as homotopy perturbation method [14–16], variational iteration
method [17], sinc-collocation method [18], an implicit series solution [19] and a
Jacobi-Gauss collocation method [20]. Parand et al. [21] obtained another approxi-
mate solution based on rational Legendre pseudospectral approach. Recently, some
other approximate solutions of Lane-Emden type equations are obtained using per-
turbation techniques [22, 23], optimal homotopy method [24], generalized Jacobi-
Galerkin method [25], and third and fourth kinds Chebyshev operational matrices
[26].
Spectral methods play prominent roles in solving various kinds of differential equa-
tions. It is known that there are three most widely used spectral methods, they are the
tau, collocation, and Galerkin methods. Collocation methods have become increas-
ingly popular for solving differential equations, in particular, they are very useful
in providing highly accurate solutions to nonlinear differential equations (see, for
example [27–34]).
One approach for solving differential equations is based on converting the un-
derlying differential equations into integral equations through integration, approxi-
mating various signals involved in the equation by truncated orthogonal series and
using the operational matrix of integration, to eliminate the integral operations. Spe-
RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20
1300 Y. H. Youssri, W. M. Abd-Elhameed, E. H. Doha 3

cial attentions have been given to applications of block pulse functions [35], Leg-
endre polynomials [36], Chebyshev polynomials [37], Haar wavelets [38], Legendre
wavelets [6, 39], and Chebyshev wavelets [9]. Another approach is based on using
operational matrix of derivatives in order to reduce the underlying problem into solv-
ing a system of algebraic equations (see, [3]).
The ultraspherical polynomials have been used in various applications. For instance,
Doha and Abd-Elhameed [40, 41] and Doha et al. [42] have constructed efficient
spectral-Galerkin algorithms using compact combinations of ultraspherical polyno-
mials for solving second-, high even- and high odd-order elliptic equations. More-
over, Doha and Abd-Elhameed [43] have obtained accurate spectral solutions for the
parabolic and elliptic partial differential equations by the ultraspherical tau method
and they have pointed out that the expansion based on Chebyshev polynomials is not
always better than ultraspherical series. The expansion for the solution enables one
to get the sought-for approximation for any possible value of the ultraspherical para-
meter λ > − 21 .
The main aim of this paper is to develop and implement two algorithms for solving
differential equations of Lane-Emden type based on employing shifted ultraspherical
wavelets operational matrix of derivatives.
The paper is organized as follows. In Sec. 2, we give some relevant properties of
ultraspherical polynomials and their shifted forms. Moreover, in this section, ultra-
spherical wavelets are constructed. In Sec. 3, we establish the operational matrix
of differentiation of the ultraspherical wavelets basis. Section 4 is concerned with
presenting and implementing a numerical algorithm for solving linear and nonlinear
Lane-Emden equations. In Sec. 5, we give a comprehensive study for the error ana-
lysis of the suggested wavelets expansion. Section 6 is concerned with considering
some numerical examples aiming to illustrate the efficiency and applicability of the
developed algorithm. Conclusions are given in Sec. 7.

2. SOME PROPERTIES OF ULTRASPHERICAL POLYNOMIALS AND THEIR SHIFTED

FORMS

2.1. ULTRASPHERICAL POLYNOMIALS

The ultraspherical polynomials (a special type of Jacobi polynomials) associ-

ated with the real parameter (λ > − 12 ) are a sequence of orthogonal polynomials on
1
the interval (-1,1), with respect to the weight function w(x) = (1 − x2 )λ− 2 , i.e.
Z1 
2 λ− 12 (λ) 0, m 6= n,
(1 − x ) Cm (x)Cn(λ) (x) dx = (5)
hn , m = n,
−1
RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20
4 Ultraspherical wavelets method for solving Lane-Emden type equations 1301

where √
π n! Γ(λ + 21 ) Γ(n + 2λ)
hn = , (2λ)n = .
(2λ)n (n + λ) Γ(λ) Γ(2λ)
(λ)
It is convenient to weigh the ultraspherical polynomials so that Cn (1) = 1.
(0)
This is not the usual standardization, but has the desirable properties that Cn (x) are
(1)
identical with the Chebyshev polynomials of the first kind Tn (x), Cn2 (x) are the
(1)
Legendre polynomials Ln (x), and Cn (x) is equal to Un+1 n (x)
, where Un (x) are the
Chebyshev polynomials of the second kind.
The derivatives of ultraspherical polynomials are given in the following theorem.
(λ)
Theorem 1. For all q > 1, the qth derivative of the ultraspherical polynomial Ck (x)
is given explicitly by
(λ) 2q k!
Dq Ck (x) = ×
(q − 1)! Γ(k + 2λ)
 k − m + q − 2   k + m + q + 2λ 
k−q (m + λ)Γ(m + 2λ) !Γ
 2 2
X
(λ)
 Cm (x),
k−q−m k + m − q + 2λ + 2 
m=0
(k+m−q) even
m! !Γ
2 2
k ≥ q. (6)
(For a proof of Theorem 1, see, Doha [44]).
(λ)
As a direct consequence of Theorem 1, the first derivative of Cn (x) can be easily
obtained in the following corollary.
Corollary 1. For all n > 1, one has
n−1
2 n! X (λ + k) Γ(k + 2λ) (λ)
DCn(λ) (x) = Ck (x). (7)
Γ(n + 2λ) k=0
k!
(k+n) odd

2.2. SHIFTED ULTRASPHERICAL POLYNOMIALS

(λ) (λ)
The shifted ultraspherical polynomials are defined on [0, 1] by C̃n (x) = Cn (2x−
1). All results for ultraspherical polynomials can be easily transformed to give the
(λ)
corresponding results for their shifted forms. The orthogonality relation for C̃n (x)
1
with respect to the weight function w̃(x) = (x − x2 )λ− 2 is given by

m 6= n,

0,
1
Z 

−λ √
2 λ− 12
(x − x ) C̃n(λ) (x) C̃m
4
(λ)
(x) dx =
π n! Γ(λ + 12 )
0 
 , m = n.
(2λ)n (n + λ) Γ(λ)
RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20
1302 Y. H. Youssri, W. M. Abd-Elhameed, E. H. Doha 5

(λ)
Also, the first derivative of C̃n (x) is given in the following corollary.
Corollary 2. For all n ≥ 1, one has
n−1
4 n! X (λ + k) Γ(k + 2λ) (λ)
DC̃n(λ) (x) = C̃k (x). (8)
Γ(n + 2λ) k=0
k!
(k+n) odd

3. SHIFTED ULTRASPHERICAL WAVELETS OPERATIONAL MATRIX OF DERIVATIVES

Wavelets constitute a family of functions constructed from dilation and trans-

lation of single function called the mother wavelet. When the dilation parameter a
and the translation parameter b vary continuously, we have the following family of
continuous wavelets:
 
−1/2 t−b
ψa,b (t) = |a| ψ , a, b ∈ R , a 6= 0. (9)
a
(λ)
Ultraspherical wavelets ψnm (t) = ψ(k, n, m, λ, t) have five arguments: k, n can as-
sume any positive integer, m is the order for the ultraspherical polynomial, λ is the
known ultraspherical parameter and t is the normalized time. They are defined on
the interval [0, 1] by

k+ 1
 2 2 (λ) k+1
  
(λ) √ C m 2 t − (2n + 1) , t ∈ [ 2nk , n+1
2k
],
ψnm (t) = A m (10)

 0, otherwise,
where m = 0, 1, . . . , M, n = 0, 1, . . . , 2k − 1, and
(2λ)m (m + λ) Γ(λ)
Am = √ . (11)
π m! Γ(λ + 12 )
A function f (t) defined on [0, 1] may be expanded in terms of ultraspherical wavelets
as
X∞ X ∞
(λ)
f (t) = cnm ψnm (t), (12)
n=0 m=0
where Z 1
  1
(λ)
cnm = f (t), ψnm (t) = (t − t2 )λ− 2 f (t) ψnm
(λ)
(t) dt. (13)
w 0
Assume that f (t) can be approximated in terms of ultraspherical wavelets as:
k −1 M −1
2X X
(λ)
f (t) ' cnm ψnm (t) = CT Ψ(λ) (t),
n=0 m=0
RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20
6 Ultraspherical wavelets method for solving Lane-Emden type equations 1303

where C and Ψ(λ) (t) are 2k (M + 1) × 1 matrices given by

 T
C = c0,0 , c0,1 , . . . , c0,M , c1,0 . . . , c1,M , . . . , c2k −1,0 , . . . , c2k −1,M , (14)

h iT
(λ) (λ) (λ) (λ) (λ) (λ) (λ)
Ψ(λ) (t) = ψ0,0 , ψ0,1 , . . . , ψ0,M , ψ1,0 , . . . , ψ1,M , . . . ψ2k −1,0 , . . . , ψ2k −1,M . (15)
The ultraspherical wavelets operational matrix of the first derivative is stated and
proved in the following theorem.
Theorem 2. Let Ψ(λ) (t) be the ultraspherical wavelets vector defined in (15). The
derivative of the vector Ψ(λ) (t) can be expressed by
d (λ)
Ψ (t) = DΨ(λ) (t), (16)
dt
where D is 2k (M + 1) × 2k (M + 1) operational matrix of derivative defined as fol-
lows  
F 0 . . . 0
 0 F . . . 0 
 . . . . 
D=  .
, (17)
. . . 
. . . .
 
0 0 . . . F
where F is the (M + 1) square matrix whose (r, s)th element is defined as follows:
2k+1

√ , r > 2, r > s and (r + s) odd,

Fr,s = Ar−1 As−1
0, otherwise,


where Ar is as given in (11).

Proof. If we make use of the shifted ultraspherical polynomials, the rth element of
vector Ψ(λ) (t) in (15) can be written as
1
2k+ 2 (λ)  k 
Ψ(λ)
r (t) = (λ)
ψnm (t) =√ C̃m 2 t − n χ[ n , n+1 ] ,
Am 2k 2k

i = 1, 2, . . . , 2k (M + 1), (18)
where r = n(M + 1) + (m + 1); m = 0, 1, . . . , M ; n = 0, 1, . . . , (2k − 1), and
t ∈ [ 2nk , n+1

1, 2k
];
χ[ n , n+1 ] =
2 k 2 k 0, otherwise.
If we differentiate (18) with respect to t, then we get
(λ) 1
dΨr (t) 0(λ) 22k+ 2 d h (λ)  k i
= ψnm (t) = √ C̃m 2 t − n χ[ n , n+1 ] . (19)
dt Am dt 2k 2k

RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20

1304 Y. H. Youssri, W. M. Abd-Elhameed, E. H. Doha 7

It is to be noted here that the r.h.s. of (19) is zero outside the interval [ 2nk , n+1
2k
], and
hence its ultraspherical wavelets expansion only have those elements in Ψ (t) that (λ)
(λ)
are nonzero in the interval [ 2nk , n+1
2k
], i.e. Ψi (t), i = n(M + 1) + 1, n(M + 1) +
(λ)
dΨr (t)
2, . . . , n(M + 1) + M + 1, this enables one to expand in terms of shifted
dt
(λ)
ultraspherical wavelets Ψi (t); i = n(M + 1) + 1, n(M + 1) + 2, . . . , n(M + 1) +
M + 1, in the form
(λ) (n+1)(M +1)
dΨr (t) X (λ)
= ai Ψi (t),
dt
i=n(M +1)+1

and accordingly (19) implies that the operational matrix D is a block matrix defined
as in (17). Moreover, we have
(λ)
dC̃0 (t)
= 0,
dt
d (λ)
and this implies that Ψ (t) = 0 for r = 1, (M + 1) + 1, 2(M + 1) + 1, . . . , (2k −
dt r
1)(M + 1) + 1.
Consequently, the first row of matrix F defined in (17) is zero. Now and with the aid
(λ)
of Corollary 2, the first derivative of Ψr (t) may be expressed in the form
(λ) 3 m−1
dΨr (t) 22k+ 2 X (λ)
= √ (λ + j) C̃j (2k t − n) χ[ n , n+1 ] . (20)
dt Am j=0
2k 2k
j+m odd

Expanding this equation in terms of ultraspherical wavelets basis, we get

(λ) m−1
dΨr (t) X 1 (λ)
= 2k+1 p Ψn(M +1)+j (t). (21)
dt j=0
Ar−1 Aj−1
j+m odd

So if we choose Fr,s as
2k+1

√ , r=2,. . . ,M+1, s=1,. . . , r-1 and (r+s) odd;

Fr,s = Ar−1 As−1
0, otherwise,


then Eq. (16) is obtained, and the proof of the theorem is completed.

Remark 1. It is worthy to note here that the ultraspherical wavelets operational

matrix given in (17) is considered as a generalization of the Legendre wavelets oper-
ational matrix given in [3].
RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20
8 Ultraspherical wavelets method for solving Lane-Emden type equations 1305

Corollary 3. The operational matrix for the nth derivative can be obtained from
dn Ψ(t)
= Dn Ψ(t), n = 1, 2, . . . , (22)
dtn
where Dn is the nth power of matrix D.

4. LANE-EMDEN DIFFERENTIAL EQUATIONS

In this section, we give two numerical algorithms for solving linear and non-
linear differential equations of Lane-Emden type based on employing the shifted
ultraspherical wavelets operational matrix of derivatives that was introduced in Sec.
3.

4.1. LINEAR DIFFERENTIAL EQUATIONS

Consider the linear Lane-Emden differential equation

α 0
y 00 (x) + y (x) + f (x) y(x) = g(x), y(0) = A, y 0 (0) = B,
x (23)
0 < x 6 1, α > 0.
If we set
f1 (x) = x f (x), g1 (x) = x g(x),
then Eq. (23) is turned into
xy 00 (x) + α y 0 (x) + f1 (x) y(x) = g1 (x), y(0) = A, y 0 (0) = B,
(24)
0 < x 6 1, α > 0.
If we approximate y(x), x, f1 (x) and g1 (x) by the ultraspherical wavelets, then one
can write
k −1 M
2X X
y(x) ' cnm ψnm (x) = CT Ψ(x),
n=0 m=0
k −1 M
2X X
x' qnm ψnm (x) = QT Ψ(x),
n=0 m=0
k −1 M
(25)
2X X
f1 (x) ' f˜nm ψnm (x) = FT1 Ψ(t),
n=0 m=0
k −1 M
2X X
g1 (x) ' gnm ψnm (x) = GT Ψ(x),
n=0 m=0
RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20
1306 Y. H. Youssri, W. M. Abd-Elhameed, E. H. Doha 9

where CT , QT , FT1 and GT are defined similarly as in (14). Relations (17) and (22)
enable one to approximate y 0 (x) and y 00 (x) as
y 0 (x) ' CT DΨ(λ) (x), y 00 (x) ≈ CT D2 Ψ(λ) (x). (26)
With the aid of Eqs. (25) and (26), the residual of Eq. (24) can be written explicitly
in the form
R(x) =QT Ψ(λ) (x)(Ψ(λ) (x))T (D2 )T C + α CT DΨ(λ) (x)
(27)
+ FT1 (Ψ(λ) (x))T Ψ(λ) (x) C − GT Ψ(λ) (x).
Applying tau method (see [45]), the following 2k (M + 1) − 2 linear equations are

generated
Z 1
(λ)
Ψj (x)R(x) dx = 0, j = 1, 2, . . . , 2k (M + 1) − 2. (28)
0
Moreover, the initial conditions of Eq. (23) yield
CT Ψ(λ) (0) = A, CT DΨ(λ) (0) = B. (29)
Thus Eqs. (28)-(29) generate 2k (M
+ 1) set of linear equations that can be solved
for the unknown components of the vector C, and hence the approximate spectral
wavelets solution of y(x) can be obtained.

4.2. NONLINEAR LANE-EMDEN DIFFERENTIAL EQUATIONS

Consider the nonlinear equation

α
y 00 (x) = ϕ x, y(x), y 0 (x) = − y 0 (x) − f (x, y) + g(x),


x (30)
0
y(0) = A, y (0) = B, 0 < x 6 1, α > 0,
where f (x, y) is a nonlinear function in the two variables x, y. If we set
Θ x, y(x), y 0 (x) = x ϕ x, y(x), y 0 (x) = −α y 0 (x) − x f (x, y) + x g(x),

then Eq. (30) is turned into

x y 00 (x) = Θ x, y(x), y 0 (x) , y 0 (0) = B,


y(0) = A, 0 < x 6 1, α > 0. (31)
If we approximate y(x) as in (25) and make use of (17) and (22), then after following
the same procedure of Sec. 4.1, we get
 
x CT D2 Ψ(λ) (x) ≈ Θ x, CT Ψ(λ) (x), CT DΨ(λ) (x) . (32)
Also, the initial conditions yield
CT Ψ(λ) (0) = A, CT DΨ(λ) (0) = B. (33)
RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20
10 Ultraspherical wavelets method for solving Lane-Emden type equations 1307

To find an approximate solution of y(x), we compute (32) at the first (2k (M +1)−2)
(λ)
roots of C̃2k (M +1) (x). These equations with the two Eqs. (33) generate 2k (M + 1)
nonlinear equations in the expansion coefficients, cnm , that can be solved with the
aid of the well-known Newton’s iterative method.

5. ERROR ANALYSIS

In this section, we give a comprehensive study for the error analysis of the
suggested ultraspherical wavelets expansion. In this respect, we will state and prove
two important theorems, in the first one we show that the ultraspherical wavelets
expansion of a function f (x) with a bounded second derivative, converges uniformly
to f (x), and in the second one we give an upper bound for the error (in L2ω̃ -norm,
1
ω̃(t) = (t − t2 )λ− 2 ) of the truncated ultraspherical wavelets expansion.
The following lemma is needed.
Lemma 1. (see, [46], p. 742) Let f (x) be a continuous, positive, decreasing function
X∞
P
for x > n. If f (k) = ak , provided that an is convergent, and Rn = ak , then
k=n+1
Z ∞
Rn 6 f (x) dx.
n
1
Theorem 3. A function f (x) ∈ L2ω̃ [0, 1], 0 < λ < 1, w̃(x) = (x − x2 )λ− 2 can be
expanded as an infinite series of ultraspherical wavelets, which converges uniformly
to f (x), given that |f 00 (x)| 6 L. Explicitly, the expansion coefficients in (13) satisfy
the inequality
4L (1 + λ)2 (m + 1 + λ)2
|cnm | < 5 , ∀ n > 0, m > 2. (34)
(m − 2)4 (n + 1) 2
Proof. For the proof of Theorem 3, see, [47].

 large values of n and m the expansions

Note 1. It is worthy to note here that for
5
coefficients |cnm | is of O n− 2 m−2 .
Note 2. It is worthy to note here that the result obtained in Theorem 3 may be
improved but with some extra conditions on the function f (x). To be concise, if
(p)
f is differentiable
 p-times
1
 some p > 1 and |f (x)| 6 L, we can show that
for
|cnm | is of O n−p− 2 m−2 .
Theorem 4. If f, λ satisfy the hypothesis of Theorem 3, and if we consider the ul-
k −1 M −1
2X X
(λ)
traspherical wavelets expansion fk,M (t) = cnm ψnm (t), then the following
n=0 m=0
RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20
1308 Y. H. Youssri, W. M. Abd-Elhameed, E. H. Doha 11

error estimate (in L2ω̃ -norm) is obtained

(1 + λ)2 L (M + λ)
kf − fk,M kω̃ < 7 , ∀M > 3. (35)
4k (M − 3) 2
(λ)
Proof. From Eq. (12), and making use of the orthonormality property of {ψnm (t)},
we get
∞ X
X ∞
kf − fk,M k2ω̃ = c2nm .
n=2k m=M
In virtue of Theorem 3, one can write
∞ ∞
X X (m + 1 + λ)4
kf − fk,M k2ω̃ 2
< 16L (1 + λ) 4
,
(m − 2)8 (n + 1)5
n=2k m=M
and the application of Lemma 1 leads to
Z ∞ Z ∞
2 2 4 (x + 1 + λ)4
kf − fk,M kω̃ < 16L (1 + λ) 8 5
dx dy
2k −1 M −1 (x − 2) (y + 1)
4 2 3−2k 15λ2 − 15λ + 35λ M


(λ + 1) L 4 + 21(M − 1)M + 9
=
105(M − 3)7
(1 + λ)4 L2 (M + λ)2
< ,
42k (M − 3)7
and hence
(1 + λ)2 L (M + λ)
kf − fk,M kω̃ < 7 ,
4k (M − 3) 2
which completes the proof of the theorem.

 of k and M the expansions

Note 3. It is worthy to note here that forlarge values
− 25 −k
coefficients ek,M = |f − fk,M | is of O M 4 .

6. ILLUSTRATIVE EXAMPLES

In this section, we illustrate the two proposed algorithms for solving initial and
boundary value linear and nonlinear differential equations of Lane-Emden type, by
presenting four examples.

6.1. LINEAR LANE-EMDEN EQUATIONS

Example 1. Consider the Lane-Emden equation given in [48] by

2
y 00 + y 0 + y = 6 + 12x + x2 + x3 ; 0 < x 6 1, y(0) = 0, y 0 (0) = 0, (36)
x
RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20
12 Ultraspherical wavelets method for solving Lane-Emden type equations 1309

with the exact solution y = x2 + x3 . We solve Eq. (36) using the algorithm described
in Sec. 4.1 for the case corresponding to k = 0, M = 3. After performing some
manipulations, the components of the vector C are given by
r r
A0 11 + 6λ A1 31 + 14λ
c0,0 = , c0,1 = ,
2 16(1 + λ) 2 16(2 + λ)

r r
A2 5 + 10λ A3 1 + 2λ
c0,2 = , c0,3 = ,
2 16(1 + λ) 2 16(2 + λ)
and consequently
y(x) = CT Ψ(λ) (x) = x2 + x3 ,
which is the exact solution.
Example 2. Consider the Lane-Emden equation given in [49] by
2 0
y 00 + y + y m = 0; 0 < x 6 1, y(0) = 1, y 0 (0) = 0. (37)
x
We solve Eq. (37) using the algorithm described in Sec. 4.1 for the two cases corre-
spond to m = 0, 1.
2
Case (i) (m = 0): In this case the exact solution of (37) is y = 1 − x6 . If we make use
of (17) and (22), then the two operational matrices D and D2 are given respectively
by
 
0 0 0
D =  F2,1 0 0 ,
0 F3,2 0
and
 
0 0 0
2
D =  0 0 0 ,
F2,1 F3,2 0 0
where
p p
Γ(1 + λ) 2(1 + λ) Γ(1 + λ) 2(1 + λ)(2 + λ)(1 + 2λ)
F2,1 = √ , F3,2 = √ ,
π Γ( 12 + λ) π Γ( 12 + λ)
moreover the vector Ψ(λ) (x) can be evaluated to give
 
√ 1
2 A2 (2x − 1)
Ψ(λ) (x) = 
 
,
 √ 2
2(1 + λ)(2x − 1) − 1 
2 A3 ,
1 + 2λ
RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20
1310 Y. H. Youssri, W. M. Abd-Elhameed, E. H. Doha 13

where Am is as given in (11).

Now, the residual of Eq. (37) is given by
R(x) = x CT D2 Ψ(λ) (x) + 2 CT DΨ(λ) (x) + x. (38)
If we apply the tau method, then we get
p 2 + 2λ p 1
2A2 c0,1 + 2A3 c0,2 + = 0, (39)
1 + 2λ 8
moreover, the use of the two initial conditions yields
p p
c0,0 − 2A2 c0,1 + 2A3 c0,2 = 1, (40)
and
p 4 + 4λ p
2A2 c0,1 − 2A3 c0,2 = 0. (41)
1 + 2λ
The solution of the linear system of Eqs. (39)-(41) gives
45 + 46λ −1 −(1 + 2λ)
c0,0 = , c0,1 = √ , c0,2 = √ ,
48(1 + λ) 12 2 A2 48(1 + λ) 2 A3
and consequently
x2
y(x) = 1 − , (42)
6
which is the exact solution.
Case (ii) (m = 1): In this case the exact solution of (37) is y = sinx x . We solve Eq.
(37) using the algorithm described in Sec. 4.1. The maximum absolute error E is
given in Table 1 for various values of k, M , and λ. The numerical and exact solutions
are depicted in Fig. 1, in case of k = 0, M = 3 for some values of λ. Also, aiming
to illustrate the super convergence of our algorithm, Eq. (37) is solved by recasting
it as a system of first-order differential equations and then applying the following
backward differentiation formula (BDF) (see, [50])
4 1 2
yn+2 = yn+1 − yn + h f (tn+2 , yn+2 ), y(t0 ) = y0 .
3 3 3
0
We argue as follows, let y = u, then Eq. (37) is equivalent to the following system of
first-order differential equations
y 0 = u, y(0) = 1,
0
xu = −2u − x y, u(0) = 0.
Finally, Table 2 displays a comparison between our proposed wavelets solution for
the case k = 0, M = 10 and λ = 1, with the solution obtained by BDF method. The
results in this Table show that our method is more accurate if compared with the BDF
method.
RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20
14 Ultraspherical wavelets method for solving Lane-Emden type equations 1311

Table 1

Maximum absolute error E for Example 2 case (ii)

k M λ E λ E λ E λ E
0 4 1.22 . 10−6 1.95 . 10−7 3.09 . 10−7 4.22 . 10−7
6 0 1.40 . 10−9 1
2 1.93 . 10−10 1 3.42 . 10−10 3
2 5.13 . 10−10
1 3 5.90 . 10−8 3.94 . 10−8 7.14 . 10−8 1.54 . 10−8
4 0 5.19 . 10−10 1
2 3.51 . 10−9 1 3.27 . 10−6 3
2 2.03 . 10−4

Table 2

A comparison of the wavelets solution with the BDF solution for Example 2 case (ii)
x Wavelets solution BDF solution Wavelets error BDF error
0.2 0.993347 0.992666 4.2 × 10−15 6.8 × 10−4
0.4 0.973546 0.972218 2.5 × 10−13 1.3 × 10−3
0.6 0.941071 0.939127 4.8 × 10−12 1.9 × 10−3
0.8 0.896695 0.894182 4.9 × 10−11 2.5 × 10−3
1.0 0.841471 0.838448 3.4 × 10−10 3.0 × 10−3

1.02

1 Exact
λ=0
λ=0.5
0.98 λ=1

0.96

0.94
y(x)

0.92

0.9

0.88

0.86

0.84
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

Fig. 1 – Numerical and exact solutions of Example 2 case (ii) for k = 0, M = 3.

6.2. NONLINEAR LANE-EMDEN EQUATIONS

Example 3. Consider the Lane-Emden equation given in [49] by

2 0
y 00 +
y + y 5 = 0; 0 < x 6 1, y(0) = 1, y 0 (0) = 0. (43)
x
 − 1
x2 2
with the exact solution y = 1 + 3 . We solve Eq. (43) using the algorithm
described in Sec. 4.2 for the case corresponds to k = 0, M = 3. In Table 3, the
RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20
1312 Y. H. Youssri, W. M. Abd-Elhameed, E. H. Doha 15

components of the vector C and the maximum absolute error E are illustrated for
various values of λ. The numerical and exact solutions are depicted in Fig. 2, in case
of k = 0, M = 3 for some values of λ.

Table 3

Components of C and the maximum absolute error E for Example 3

λ c00 c01 c02 c03 E
1
2 0.951426 −0.0696688 −0.0185703 0.00274369 2.19 . 10−4
3
2 0.95514 −0.0708446 −0.0223878 0.00397709 3.80 . 10−4
5
2 0.956739 −0.0715075 −0.0240569 0.00468697 4.97 . 10−4
0 0.946943 −0.0686796 −0.0140174 0.00172343 1.02 . 10−4

Exact
0.98 λ=0.5
λ=1.5
λ=2.5
0.96

0.94
y(x)

0.92

0.9

0.88

0.86
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

Fig. 2 – Numerical and exact solutions of Example 3 for k = 0, M = 3.

Example 4. Consider the Lane-Emden equation given in [48, 49] by

2 0
y 00 + y + ey = 0; 0 < x 6 1, y(0) = 0, y 0 (0) = 0. (44)
x
We solve (44) using the algorithm described in Sec. 4.2 for the case corresponding to
k = 0, M = 3. In Table 4, we give a comparison between the present solution for the
three cases corresponding to λ = 12 , 23 , 1, with the two solutions obtained by Pandey
et al. [49] and Wazwaz [48].
Remark 2. The results in Table 4 show that the two numerical solutions for various
values of λ are closer to the Runge-Kutta solution obtained from Mathematica than
the solutions obtained by Pandey [49] and Wazwaz [48].
RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20
16 Ultraspherical wavelets method for solving Lane-Emden type equations 1313

Table 4

A comparison of the present solution with [49] and [48] for Example 4
1 2
x Pandey Wazwaz λ= 2 λ= 3 λ=1 Mathematica
0 0.0000000 0.00000000 0.0000000000 0.0000000000 0.0000000000 0.000000000000
0.2 -0.0066533 -0.00665337 -0.0066533671 -0.0066533670 -0.0066533669 -0.006653367099
0.5 -0.0411541 -0.04115400 -0.0411539572 -0.0411539573 -0.0411539571 -0.041153954402
1 -0.1588624 -0.15882700 -0.1588276774 -0.1588276776 -0.1588276775 -0.158825034266

7. CONCLUDING REMARKS

In this paper, numerical solutions of Lane-Emden linear and nonlinear equa-

tions are given with the aid of ultraspherical wavelets algorithm. The derivation of
this algorithm is essentially based on constructing the shifted ultraspherical wavelets
operational matrix of differentiation. One advantage of the developed algorithms is
that high accurate approximate solutions are achieved using a few number of terms
in ultraspherical expansion. Another advantage is that for every value of the ultras-
pherical parameter, a numerical solution is obtained. The obtained numerical results
are favorably compared with the analytical ones.

REFERENCES

1. A. Constantinides, W.P. Schowalter, J.J. Carberry, J.R. Fair, Applied numerical methods with per-
sonal computers (McGraw-Hill, Inc., 1987).
2. D.E. Newland, An introduction to random vibrations, spectral and wavelet analysis (Courier Cor-
poration, 2012).
3. F. Mohammadi, M.M. Hosseini, J. Frankl. Inst.-Eng. Appl. Math. 348 (8), 1787–1796 (2011).
4. F. Mohammadi, M. Hosseini, J. Adv. Res. 2 (1) (2010).
5. F. Mohammadi, M. Hosseini, S.T. Mohyud-Din, Int. J. Syst. Sci. 42 (4), 579–585 (2011).
6. M. Razzaghi, S. Yousefi, Int. J. Syst. Sci. 32 (4), 495–502 (2001).
7. M. Razzaghi, S. Yousefi, Math. Comput. Model. 34 (1-2), 45–54 (2001).
8. M. Razzaghi, S. Yousefi, Math. Meth. Appl. Sci. 25 (7), 529–539 (2002).
9. E. Babolian, F. Fattahzadeh, Appl. Math. Comput. 188 (1), 417–426 (2007).
10. M. Dehghan, A. Saadatmandi, Int. J. Comput. Math. 85 (1), 123–130 (2008).
11. A. Saadatmandi, M. Dehghan, Comput. Math. Appl. 59 (3), 1326–1336 (2010).
12. L. Zhu, Y. Wang, Q. Fan, in The 2011 International Conference on Scientific Computing, Las
Vegas, USA, pp. 126130, 2011.
13. L. Zhu, Q. Fan, Commun. Nonlinear Sci. Numer. Simul. 17 (6), 2333–2341 (2012).
14. M. Chowdhury, I. Hashim, Phys. Lett. A 365 (5-6), 439–447 (2007).
15. M. Chowdhury, I. Hashim, Nonlinear Anal-Real 10, 104–115 (2009).
16. A. Yildirim, T. Özis, Phys. Lett. A 369 (1-2), 70–76 (2007).
17. A. Yildirim, T. Özis, Nonlinear Anal.-Theory Methods Appl. 70 (6), 2480–2484 (2009).
18. K. Parand, A. Pirkhedri, New Astron. 15 (6), 533–537 (2010).
19. E. Momoniat, C. Harley, Math. Comput. Model. 53 (1-2), 249–260 (2011).
RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20
1314 Y. H. Youssri, W. M. Abd-Elhameed, E. H. Doha 17

20. A. Bhrawy, A. Alofi, Commun. Nonlinear Sci. Numer. Simul. 17 (1), 62–70 (2012).
21. K. Parand, M. Shahini, M. Dehghan, J. Comput. Phys. 228 (23), 8830–8840 (2009).
22. R. Van Gorder, New Astron. 16 (2), 65–67 (2011).
23. R. Van Gorder, Celest. Mech. Dyn. Astron. 109 (2), 137–145 (2011).
24. S. Iqbal, A. Javed, Appl. Math. Comput. 217, 7753–7761 (2011).
25. W. Abd-Elhameed, Eur. Phys. J. Plus 130 (3), 52 (2015).
26. E.H. Doha, W.M. Abd-Elhameed, M.A. Bassuony, Rom. J. Phys. 60 (3-4), 281–292 (2015).
27. A.H. Bhrawy, A.A. Al-Zahrani, Y.A. Alhamed, D. Baleanu, Rom. J. Phys. 59, 646–657 (2014).
28. A.H. Bhrawy, M.A. Zaky, D. Baleanu, Rom. Rep. Phys. 67(2), 340–349 (2015).
29. A.H. Bhrawy, E.A. Ahmed, D. Baleanu, Proc. Romanian Acad. A 15 (4), 322–330 (2014).
30. E.H. Doha, A.H. Bhrawy, D. Baleanu, M.A. Abdelkawy, Rom. J. Phys. 59 (3-4), 247–264 (2014).
31. A.H. Bhrawy, M.A. Zaky, D. Baleanu, M.A. Abdelkawy, Rom. J. Phys. 60 (3-4), 344–359 (2015).
32. A.H. Bhrawy, E.H. Doha, D. Baleanu, S.S. Ezz-Eldien, M.A. Abdelkawy, Differential Equations
15, 23 (2015).
33. M.A. Abdelkawy, M.A. Zaky, A.H. Bhrawy, D. Baleanu, Rom. Rep. Phys. 67(3), 773–791 (2015).
34. M. Mirzazadeh, M. Eslami, A.H. Bhrawy, A. Biswas, Rom. J. Phys. 60 (3-4), 293–310 (2015).
35. C. Cheng, Y. Tsay, T. Wu, J. Frankl. Inst.-Eng. Appl. Math. 303 (3), 267–284 (1977).
36. R. Chang, M. Wang, J. Optim. Theory Appl. 39 (2), 299–307 (1983).
37. I.-R. Horng, J.-H. Chou, Int. J. Syst. Sci. 16 (7), 855–861 (1985).
38. J.-S. Guf, W.-S. Jiang, Int. J. Syst. Sci. 27 (7), 623–628 (1996).
39. F. Khellat, S. Yousefi, J. Frankl. Inst.-Eng. Appl. Math. 343 (2), 181–190 (2006).
40. E.H. Doha, W.M. Abd-Elhameed, SIAM J. Sci. Comput. 24, 548–571 (2002).
41. E.H. Doha, W.M. Abd-Elhameed, Math. Comput. Simul. 79 (11), 3221–3242 (2009).
42. E.H. Doha, W.M. Abd-Elhameed, A.H. Bhrawy, Appl. Math. Model. 33 (4), 1982–1996 (2009).
43. E.H. Doha, W.M. Abd-Elhameed, J. Comput. Appl. Math. 181 (1), 24–45 (2005).
44. E.H. Doha, Comput. Math. Appl. 21 (2-3), 115–122 (1991).
45. C. Canuto, M. Hussaini, A. Quarteroni, T. Zang, Spectral methods in fluid dynamics (Springer
Verlag, 1988).
46. J. Stewart, Single Variable Essential Calculus: Early Transcendentals (Cengage Learning, 2012).
47. W.M. Abd-Elhameed, Y. Youssri, Abstract and Applied Analysis, Article ID 626275, 8 pp. (2014).
48. A.-M. Wazwaz, Appl. Math. Comput. 118, 287–310 (2001).
49. A. B. R. K. Pandey, N. Kumar, G. Dutta, Appl. Math. Comp. 218 (14), 7629–7637 (2012).
50. E. Süli, D. Mayers, An introduction to numerical analysis (Cambridge University Press, 2003).

RJP 60(Nos. 9-10), 1298–1314 (2015) (c) 2015 - v.1.3a*2015.11.20