Mathematical Modelling and Analysis http://mma.vgtu.

lt
Volume 23, Issue 1, 33–43, 2018 ISSN: 1392-6292
https://doi.org/10.3846/mma.2018.003 eISSN: 1648-3510

A New Approach for Solving Nonlinear

Singular Boundary Value Problems∗
Hui Zhua , Jing Niub , Ruimin Zhanga and Yingzhen Lina
a
School of Applied Science and Civil Engineering, Zhuhai Campus,Beijing
Institute of Technology
519088 Zhuhai, China
b
School of Mathematics and Sciences, Harbin Normal University
150025 Harbin, China
E-mail(corresp.): njirwin@163.com
Received March 14, 2017; revised December 6, 2017; accepted December 6, 2017

Abstract. In this paper, an efficient method based on Quasi-Newton’s method and

the simplified reproducing kernel method is proposed for solving nonlinear singular
boundary value problems. For the Quasi-Newton’s method the convergence order is
studied. The uniform convergence of the numerical solution as well as its derivatives
are also proved. Numerical examples are given to demonstrate the efficiency and
stability of the proposed method. The numerical results are compared with exact
solutions and the outcomes of other existing numerical methods.
Keywords: nonlinear singular boundary value problem, numerical solution, Quasi-Newton’s
method, reproducing kernel method.
AMS Subject Classification: 34B16; 34K28; 47B32; 65L10.

1 Introduction
We now consider the following nonlinear singular boundary value problem with
Neumann and Robin boundary conditions:
(
u00 (x) + αx u0 (x) = f (x, u(x)), 0 < x ≤ 1, α ≥ 1,
(1.1)
u0 (0) = 0, au(1) + bu0 (1) = c.

∗ Correspondence to: Jing Niu, School of Mathematics and Sciences, Harbin Normal Uni-
versity, Harbin, 150025, China.This work were supported by Youth Foundation of Hei-
longjiang Province (Grant No.JJ2016QN0211) and College Innovative Talent foundation
of Heilongjiang Province(Grant No.UNPYSCT-2017179).
Copyright
c 2018 The Author(s). Published by VGTU Press
 ∗
This is an Open Access article distributed under the terms of the Creative Commons Attribution
License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribu-
tion, and reproduction in any medium, provided the original author and source are credited.
34 H. Zhu, J. Niu, R. Zhang and Y. Lin
∂f
Here we assume nonlinear function f (x, u) is continuous. ∂u exists and is
∂f
continuous and ∂u ≥ 0 for any 0 ≤ x ≤ 1.
Singular boundary value problems are frequent in a wide variety of applied
mathematics and engineering practices, such as chemical reactions, gas dynam-
ics, control and optimization theory, areas of biology and astrophysics [3, 13].
Due to the difficulty of singularity, it is usually impossible to obtain the ana-
lytical solution of such equations. Hence, it’s quite necessary to seek accurate
numerical solutions of equation (1.1). Taylor series method is applied and the
error analysis is presented in [2]. The authors [6] apply He’s variational itera-
tion method for solving equation (1.1). In [7], cubic spline method to analyze
equation (1.1) is presented. The authors [1] use B-spline functions to develop
a numerical method for computing approximation to the solution of equation
(1.1). A new approach implementing a modified decomposition method in
combination with the cubic B-spline collocation technique is introduced in [8].
Furthermore, the improved Adomian decomposition based on Green’s function
is discussed [18].The finite difference method and non-polynomial cubic spline
method are given in [16] and [17] respectively. Recently, a novel approach based
on the operational matrix of orthonormal Bernoulli polynomial has been pro-
posed [13]. Numerical solution of equation (1.1) based on improved differential
transform method(IDTM) has also been proposed by the authors in [10]. Fur-
thermore, the similar method based on IDTM works well for the other type of
nonlinear boundary value problems [11]. Although, these numerical methods
are efficient and have many advantages, a lot of computational work or a high
degree of smoothness are needed.
Recently, the theory of reproducing kernel has emerged as a powerful frame-
work in numerical analysis, differential and integral equations, and probability
and statistics [5]. Based on the reproducing kernel theory, the reproducing
kernel method has been used and modified by many authors [4, 9, 14, 15, 19]. In
this work, by combining with Newton iteration and modifying the reproduc-
ing kernel method, we will find the numerical solutions of equation (1.1). It’s
a different approach from those previous attempts. Our method can reduce
computation cost and provide highly accurate global approximate solutions.

2 The Quasi-Newton’s method

In this section, we describe the classical Newton method which is a very effective
tool for solving nonlinear problems. Consider a nonlinear equation f (x) = 0,
at the point (x0 , f (x0 )),we can get a tangent equation
y = f (x0 ) + f 0 (x0 )(x − x0 ).
Furthermore, the equation f (x) = 0 can be approximated by the following
equation
f (x0 ) + f 0 (x0 )(x − x0 ) = 0. (2.1)
Then, a Newton’s iterative formula for solving f (x) = 0 is defined as following:
f (xk )
xk+1 = xk − , k = 0, 1, 2 . . . .
f 0 (xk )
A New Approach for Solving Nonlinear Singular Boundary Value Problems 35

Now we introduce the Fréchet derivative proposed by paper [12].

Definition 1. Let F : X → Y , where X and Y are Banach spaces. Then a
bounded and linear operator A : X → Y is called a Fréchet derivative of F at
u ∈ X if
kF(u + h) − F(u) − A(h)k
lim = 0,
h→0 khk
for all h ∈ X denoted by F 0 (u).

Lemma 1. If F is a linear operator, there holds F 0 (u)=F, for all u ∈ X.

Proof. Since F is a linear operator, we have

kF(u + h) − F(u) − F(h)k kF(u) + F(h) − F(u) − F(h)k

lim = lim = 0.
h→0 khk h→0 khk

Therefore, F 0 (u)=F. t
u

Especially, we can obtain the following Lemma.

Lemma 2. If F is a constant operator, then F 0 (u) is a zero operator for all

u ∈ X.

Based on Fréchet derivative theory, we can get an approximate formula for

equation (1.1) which is similar to formula (2.1).
Consider an operator F : C 2 [0, 1] → C[0, 1] as
∆
F(u) = xu00 + αu0 +xN (u),

where N (u)= − f (x, u(x)). Equation (1.1) can be rewritten as

(
F(u) = 0, 0 < x ≤ 1, α ≥ 1,
u0 (0) = 0, au(1) + bu0 (1) = c.

According the Definition (1) and Lemma (1), we can get the Fréchet derivative
of F at u0 :
F 0 (u0 ) : u 7→ xu00 + αu0 +xN 0 (u0 )u. (2.2)
Moreover the operator equation F(u) = 0 can be approximated by

F(u0 ) + F 0 (u0 )(u − u0 ) = 0. (2.3)

By substituting (2.2) into (2.3), we get the following linear equation

xu00 + αu0 +xN 0 (u0 )(u − u0 )+xN (u0 ) = 0.

Therefore, we obtain a Quasi-Newton’s iterative process as the following form

(
xu00k+1 + αu0k+1 +xN 0 (uk )uk+1 = xN 0 (uk )uk − xN (uk ),
(2.4)
u0k (0) = 0, auk (1) + bu0k (1) = c, k = 0, 1, 2 . . .

Math. Model. Anal., 23(1):33–43, 2018.

36 H. Zhu, J. Niu, R. Zhang and Y. Lin

and its general form

F(uk ) + F 0 (uk )(uk+1 − uk ) = 0. (2.5)
It can be seen equation (2.4) and equation (2.5) are linear equations. We
could get the numerical solution. Firstly we choose a simple polynomial func-
tion satisfying the boundary conditions as an initial function u0 . Then by
substituting u0 into equation (2.4), we can get a solution u1 . And after several
rounds of iteration, a high accuracy numerical result can be obtained. Next,
we analyze the the convergence of our iterative algorithm.
Theorem 1. Let Fréchet derivative F 0 and N be operators mentioned in equa-
tion (2.4)and equation (2.5). Assume that kF 0−1 kkN 00 k = γ < 1. Further, the
initial function u0 satisfies that ku0 − uk < 1, where u is the exact solution
of equation (1.1). Then, iterative formula equation (2.4) or equation (2.5) is
k
convergence, and the absolute error satisfies |uk − u| ≤ kuk − uk ≤ γ 2 −1 .

Proof. Based on equation (2.5), we have

uk+1 − uk = −[F 0 (uk )]−1 F(uk ).
Supposing u(x) is a solution of F(u) = 0, and Ek (x) = uk (x) − u(x). Then,
Ek+1 = Ek − [F 0 (uk )]−1 (F(uk ) − F(u)).
Thus,
F 0 (uk )Ek+1 = F 0 (uk )Ek − (F(uk ) − F(u)). (2.6)
Note that
F 0 (uk )Ek = xEk00 + αEk0 + xN 0 (uk )Ek , (2.7)
F(uk ) − F(u) = xu00k + αu0k + xN (uk ) − (xu00 + αu0 + xN (u))
= xEk00 + αEk0 + x(N (uk ) − N (u)) = xEk00 + αEk0 + xN 0 (ξk )Ek . (2.8)
Substituting (2.7), (2.8) into (2.6), we obtain
F 0 (uk )Ek+1 = xEk00 + αEk0 + xN 0 (uk )Ek − (xEk00 + αEk0 + xN 0 (ξk )Ek )
= x(N 0 (uk ) − N 0 (ξk ))Ek = xN 00 (ηk )(uk − ξk )Ek .
As |uk − ξk | ≤ |uk − u| = |Ek |, we have
kEk+1 k ≤ kF 0 (uk )−1 kkN 00 (ηk )kkEk k2 < γkEk k2 .
Furthermore, we discuss the convergence order based on the mathematical
induction. Firstly,
0
kE0 k = ku0 − uk < 1 = γ 2 −1 .
Let k
−1
kEk k = kuk − uk = γ 2 ,
Hence,
k k+1
−1 2 −1
kEk+1 k < γkEk k2 = γ(γ 2 ) = γ2 .
Thus, the proof is completed. t
u
A New Approach for Solving Nonlinear Singular Boundary Value Problems 37

3 The simplified reproducing kernel method

In this section, we focus on solving the linear equation by simplified reproducing
kernel method. For convenience, we discuss the simplified equation of equation
(2.4) as follows
(
xu00 (x) + αu0 (x)+xa2 (x)u = h(x),
(3.1)
u0 (0) = 0, au(1) + bu0 (1) = c,

where a2 (x) = N 0 (uk ), h(x) = x(N 0 (uk )uk − N (uk )).

Now, we introduce the reproducing kernel method for equation (3.1). From
[4], it follows that W23 [0, 1] , W21 [0, 1] are reproducing kernel spaces with repro-
ducing kernel functions Rt (x) and rt (x) respectively. Considering an operator
L : W23 [0, 1] → W21 [0, 1],

Lu = xu00 (x) + αu0 (x)+xa2 (x)u(x),

equation (3.1) can be written as

(
Lu = h(x),
(3.2)
u0 (0) = 0, au(1) + bu0 (1) = c.

Lemma 3. Let L∗ be a conjugate operator of L, then (L∗ rt )(x) = (LRx )(t).

∞ ∞
Taking a dense set {xi }i=1 on [0, 1], we get a function system {ψi (x)}i=1 in
W23 ,where ψi (x) = L∗ rx (xi ).
n
Theorem 2. {ψi (x)}i=1 is linear independent.

For the proof, one may refer to [4].

Let
∂Rt (x)  ∂Rt (x) 
ϕ1 (x) = t=0 , ϕ2 (x) = aR1 (x) + b t=1 ,
∂t ∂t
∞
Sn+2 = span {ψi }n1 ∪ ϕ1 ∪ ϕ2 , S = ∪ Si+2 .

i=1

Then Sn+2 ⊂ S ⊂ W23 .

Theorem 3. Let u be the exact solution of the operator equation (3.2). Then
Pu is also the solution of (3.2), where P is the orthogonal project operator
from W23 to S .

Let Pn+2 is the orthogonal project operator from W23 to Sn+2 .

Theorem 4. Pn+2 u is the solution of the following linear equation system
(
hv, ψi i = h(xi ), i = 1, 2, . . . n,
(3.3)
hv, ϕ1 i = 0, hv, ϕ2 i = c.

Math. Model. Anal., 23(1):33–43, 2018.

38 H. Zhu, J. Niu, R. Zhang and Y. Lin

Proof.

hPn+2 u(·), ψi (·)i = hu(·), Pn+2 ψi (·)i = hu(·), ψi (·)i = hu(·), L∗ rxi (·)i
= hLu(·), rxi (·)i = hh(·), rxi (·)i = h(xi ),
D ∂Rx (·) E
hPn+2 u(·), ϕ1 (·)i = hu(·), Pn+2 ϕ1 (·)i = hu(·), ϕ1 (·)i = u(·), |x=0
∂x
∂ ∂ 0
= hu(·), Rx (·)i |x=0 = u(x) |x=0 = u (0) = 0,
∂x ∂x
hPn+2 u(·), ϕ2 (·)i = hu(·), Pn+2 ϕ2 (·)i = hu(·), ϕ2 (·)i
D ∂Rx (·)  E ∂ 
= u(·), aR1 (·) + b x=1 = au(1) + b hu(·), Rx (·)i x=1

∂x ∂x
∂ 
= au(1) + b u(x)x=1 = au(1) + bu0 (1) = c.

∂x
t
u

Denote un = Pn+2 u. It can be shown that un is an approximate solution

of u. Furthermore, we can prove the uniform convergence.
(i)
Theorem 5. un (x) uniformly converge to u(i) (x) on [0, 1], i = 0, 1, 2.

Proof.
(i)
|u(i) (i)
n (x) − u (x)| = |hun (·) − u(·), Rx (·)i |
D ∂i E ∂iR
x
=  un (·) − u(·), i Rx (·)  ≤ kun − uk
 
∂x ∂xi
.
i
Note that k ∂∂xRix k is bounded on [0, 1], thus

|u(i) (i)
n (x) − u (x)| ≤ M kun − uk → 0.

The proof is completed. t

u

Hence, un is a good approximate solution of (3.2). Since un ∈ Sn+2 , we get

n
X
un = αi ψi +β1 ϕ1 +β2 ϕ2 . (3.4)
i=1

As un is the solution of equation (3.3), we have

 n
P


 αj hψj , ψi i + β1 hϕ1 , ψi i + β2 hϕ2 , ψi i = h(xi ), i = 1, 2, . . . n,

 j=1
n

 P
αj hψj , ϕ1 i + β1 hϕ1 , ϕ1 i + β2 hϕ2 , ϕ1 i = 0, (3.5)
 j=1
n


 P


 αj hψj , ϕ2 i + β1 hϕ1 , ϕ2 i + β2 hϕ2 , ϕ2 i = c.
j=1

The coefficients of formula (3.4) are the solution of linear equation system (3.5),
hence we get the approximate solution un of equation (3.1).
A New Approach for Solving Nonlinear Singular Boundary Value Problems 39

4 Numerical examples
In this section, numerical examples are tested to demonstrate the efficiency and
stability of our method. All examples are computed by Mathematica 7.0.

Example 1. Consider the nonlinear differential equation [1, 18]:

(
u00 (x) + x1 u0 (x) = −eu(x) , 0 < x < 1,
u0 (0) = 0, u(1) = 0.
√
The exact solution is u(x) = 2 ln (3−24−2√ 2


2)x2 +1
. In our method, we need
to take a group of parameters, such as initial function u0 (x),the number of
iterations k, and the number of dense points n. Let u0 (x) = x4 . The absolute
error function is E = |uk,n (x) − u(x)|, 0 < x ≤ 1, where uk,n (x) is the approx-
imate solution. Table 1 compares the results for this example by the proposed
method and the methods in [1, 18].

Table 1. Maximum absolute error for Example 1

k n E n E in [18] n E in [1]
3 n=4 5.67E-05 12 4.01E-04 20 3.16E-05
3 n=6 3.14E-06 16 2.54E-05 40 7.87E-06
3 n=8 6.35E-07 20 2.10E-06 60 3.50E-06

Figure 1 shows comparison of exact and approximate solution for k = 3,

n = 8 and also plots the absolute errors for k = 3, n = 4, 6, 8. We can see that
the presented method converges rapidly to the exact solution by iterating only
three times.

0.30 0.00006

0.25 0.00005

0.20 0.00004

0.15 0.00003

0.10 0.00002 n=4

u@xD
0.05 0.00001 n=6
u3,8 @xD n=8
0.00 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0

Figure 1. Comparison of exact solution u(x) and approximate solution u3,8 (x) (left side)
and E (right side) for Example 1(k = 3).

Example 2. Consider the nonlinear singular boundary value problem describing

the equilibrium of the isothermal gas sphere [3, 10, 18]
(
u00 (x) + x2 u0 (x) = −u5 (x), 0 < x < 1,
p
u0 (0) = 0, u(1) = 3/4.

Math. Model. Anal., 23(1):33–43, 2018.

40 H. Zhu, J. Niu, R. Zhang and Y. Lin
p
The exact solution is u(x) = 3/(3 + x2 ). Let u0 (x) = x2 . Table 2 shows the
comparisons of the presented method and the method used in [3, 10, 18]. After
three iterations, the maximal absolute error can reach 10−7 at n=10. We have
higher accuracy compared to the other mentioned methods.
Figure 2 gives comparison of u(x) and u4,10 (x) and also shows the absolute
errors for k = 3, n = 4, 6, 10. It indicates that the curves of the exact solution
and the approximate solution are overlap. The approximate solution is get-
ting more and more accurate as the number of dense points n increases. Our
approach has good convergence and precision for numerical calculation.

Table 2. Maximum absolute error for Example 2

k n E n E in [18] n E in [3] n E in [10]
4 n=4 3.25E-05 12 1.40E-03 16 3.64E-04 12 1.68E-04
4 n=6 6.74E-06 16 2.47E-04 32 2.49E-05 16 1.65E-05
4 n = 10 7.30E-07 20 4.86E-05 64 1.60E-06 20 1.66E-06

Figure 2 gives comparison of u(x) and u4,10 (x) and also shows the absolute
errors for k = 3, n = 4, 6, 10. It indicates that the curves of the exact solution
and the approximate solution are overlap. The approximate solution is get-
ting more and more accurate as the number of dense points n increases. Our
approach has good convergence and precision for numerical calculation.

1.00
0.00004
0.98

0.96 0.00003

0.94
0.00002
0.92

0.90 u@xD 0.00001 n=4

n=6
0.88 u4,10 @xD
n=8
0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0

Figure 2. Comparison of exact solution u(x) and approximate solution u4,10 (x) (left
side) and E (right side) for Example 2(k = 4).

Example 3. Consider the nonlinear singular boundary value problem studying

the distribution of heat sources in the human head [7, 10, 13, 16].
(
u00 (x) + x2 u0 (x) = −e−u(x) , 0 < x ≤ 1,
(4.1)
u0 (0) = 0, u(1) + u0 (1) = 0.

Let u0 (x) = x3 . Since the exact solution of equation (4.1) is not known,we
calculate function values of approximate solution in ten points which are pre-
sented in [7,10,13,16]. The numerical results for k = 3, n = 10 are listed in Ta-
ble 3. The proposed method is in good agreement to other methods [7,10,13,16].
A New Approach for Solving Nonlinear Singular Boundary Value Problems 41

Table 3. Numerical results for Example 3

x Method in [16] Method in [7] Method in [13] Method in [10] Present Method
0 0.3675169710 0.3675179806 0.3675152742 0.3675167997 0.3675166189
0.1 0.3663623697 0.3663634922 0.3663623292 0.3663623137 0.3663622527
0.2 0.3628941066 0.3628952219 0.3628940661 0.3628940507 0.3628940189
0.3 0.3570975862 0.3570986892 0.3570975457 0.3570975301 0.3570975087
0.4 0.3489484612 0.3489495463 0.3489484206 0.3489484049 0.3489483815
0.5 0.3384121893 0.3384132503 0.3384121487 0.3384121330 0.3384121172
0.6 0.3254435631 0.3254445926 0.3254435224 0.3254435063 0.3254434945
0.7 0.3099860810 0.3099870706 0.3099860402 0.3099860240 0.3099860196
0.8 0.2919711440 0.2919720836 0.2919711030 0.2919710864 0.2919711009
0.9 0.2713170512 0.2713179290 0.2713170101 0.2713169936 0.2713170010
1.0 0.2479277646 0.2479285660 0.2479277233 0.2479277073 0.2479277424

Example 4. Consider the nonlinear singular boundary value problem arising in

the study of steady-state oxygen diffusion in a spherical cell [1, 10, 13, 17].
(
δu(x)
u00 (x) + αx u0 (x) = u(x)+µ , δ > 0, µ > 0,
(4.2)
u0 (0) = 0, 5u(1) + 5u0 (1) = 5,
where δ and µ are often taken as 0.76129 and 0.03119, respectively. And we
take the value of α as 2. This equation does not have exact solution. Instead,
we discuss the absolute residual error function which is a measure of how well
the approximation satisfies the equation (4.2) as
 
α δuk,n 
R = u00k,n + u0k,n −

, 0 < x ≤ 1.
x uk,n + µ 
Taking u0 (x) = x3 , a good approximation is obtained for k = 4, n = 10. The
numerical results of different methods are listed in Table 4.

Table 4. Numerical results for Example 4

x Method in [17] Method in [1] Method in [13] Method in [10] Present Method
0 0.8284833089 0.8284832730 0.8284832819 0.8284832870 0.8284833754
0.1 0.8297060779 0.8297060752 0.8297060924 0.8297060890 0.8297061155
0.2 0.8333747471 0.8333747169 0.8333747336 0.8333747303 0.8333747353
0.3 0.8394899001 0.8394898981 0.8394899140 0.8394899106 0.8394899068
0.4 0.8480527729 0.8480527704 0.8480527850 0.8480527816 0.8480527759
0.5 0.8590649189 0.8590649140 0.8590649272 0.8590649239 0.8590649167
0.6 0.8725283096 0.8725283084 0.8725283200 0.8725283166 0.8725283100
0.7 0.8884452969 0.8884452959 0.8884453056 0.8884453023 0.8884452975
0.8 0.9068185417 0.9068185403 0.9068185481 0.9068185448 0.9068185422
0.9 0.9276509838 0.9276509825 0.9276509884 0.9276509853 0.9276509849
1.0 0.9509457969 0.9509457946 0.9509457985 0.9509457960 0.9509457969

Table 5 describes the comparison of absolute residual error between pre-

sented method and IDTM [10]. The implementation of our approach in good
line with existing methods [1, 10, 13, 17] has a fast convergence speed and high
accuracy.

Math. Model. Anal., 23(1):33–43, 2018.

42 H. Zhu, J. Niu, R. Zhang and Y. Lin

Table 5. Maximum absolute residual error for Example 4

k n R n R in [10]
4 n=4 4.25E-05 4 4.84E-04
4 n=6 6.14E-06 6 6.78E-05
4 n=8 1.38E-06 8 9.45E-06
4 n = 10 8.39E-07 10 1.31E-06

5 Conclusions
To summarize, in this paper, we propose a new approach for solving nonlin-
ear singular boundary value problem. First, we reduce the nonlinear problem
to a linear problem based on the Quasi-Newton’s method. And second, the
simplified reproducing kernel method which has less computation than before
is presented to solve the linear problem. The approximate solution can be
obtained after iterative computation. Our method is tested by four examples
arising in applied science. The numerical results show that our method has
higher accuracy, less calculation and high efficiency. It is worthy to note that
the approach detailed here can be readily adapted to the case of nonlinear inte-
gral and integro-differential equation with various boundary value conditions.

References
[1] H. Çağlar, N. Çağlar and M. Özer. B-spline solution of non-linear singular
boundary value problems arising in physiology. Chaos Solitons and Fractals,
39(3):1232–1237, 2009. https://doi.org/10.1016/j.chaos.2007.06.007.
[2] S.H. Chang. Taylor series method for solving a class of nonlinear singular bound-
ary value problems arising in applied science. Applied Mathematics and Compu-
tation, 235:110–117, 2014. https://doi.org/10.1016/j.amc.2014.02.094.
[3] M. Chawla, R. Subramanian and H. Sathi. A fourth order method for
a singular two-point boundary value problem. BIT, 28(1):88–97, 1988.
https://doi.org/10.1007/BF01934697.
[4] M.G. Cui. Nonlinear Numerical Analysis in the Reproducing Kernel Space. Nova
Science, NewYork, 2009.
[5] A. Daniel. Reproducing Kernel Spaces and Applications. Springer, NewYork,
2013.
[6] A.S.V. Ravi Kanth and K. Aruna. He’s variational iteration method
for treating nonlinear singular boundary value problems. Com-
puters and Mathematics with Applications, 60(3):821–829, 2010.
https://doi.org/10.1016/j.camwa.2010.05.029.
[7] A.S.V. Ravi Kanth and V. Bhattacharya. Cubic spline for a class
of non-linear singular boundary value problems arising in physiol-
ogy. Applied Mathematics and Computation, 174(1):768–774, 2006.
https://doi.org/10.1016/j.amc.2005.05.022.
[8] S.A. Khuri and A. Sayfy. A novel approach for the solution of a class of singular
boundary value problems arising in physiology. Mathematical and Computer
Modelling, 52(3–4):626–636, 2010. https://doi.org/10.1016/j.mcm.2010.04.009.
A New Approach for Solving Nonlinear Singular Boundary Value Problems 43

[9] Y.Z. Lin, J. Niu and M.G. Cui. A numerical solution to nonlinear sec-
ond order three-point boundary value problems in the reproducing kernel
space. Applied Mathematics and Computation, 218(14):7362–7368, 2012.
https://doi.org/10.1016/j.amc.2011.11.009.
[10] L.J.Xie, C.L.Zhou and S.Xu. An effective numerical method to solve a class
of nonlinear singular boundary value problems using improved differential trans-
form method. SpringerPlus, 5:1066–1084, 2016. https://doi.org/10.1186/s40064-
016-2753-9.
[11] L.J.Xie, C.L.Zhou and S.Xu. A new algorithm based on differ-
ential transform method for solving multi-point boundary value prob-
lems. International Journal of Computer Mathematics, 93(6):981–994, 2016.
https://doi.org/10.1080/00207160.2015.1012070.
[12] P.R. Mcgilliuray and D.W. Olenburg. Methods for calculating Frechet derivatives
and senstivities for the non-linear inversion problem. Geophysical Prospecting,
38(5):499–524, 1990. https://doi.org/10.1111/j.1365-2478.1990.tb01859.x.
[13] M. Mohsenyzadeh, K. Maleknejad and R. Ezzati. A numerical approach
for the solution of a class of singular boundary value problems aris-
ing in physiology. Advances in Difference Equations, 231(1):231, 2015.
https://doi.org/10.1186/s13662-015-0572-x.
[14] J. Niu, Y.Z. Lin and C.P. Zhang. Approximate solution of nonlinear multi-point
boundary value problem on the half-line. Mathematical Modelling and Analysis,
17(2):190–202, 2012. https://doi.org/10.3846/13926292.2012.660889.
[15] J. Niu, Y.Z. Lin and C.P. Zhang. Numerical solution of nonlinear three-point
boundary value problem on the positive half-line. Mathematical Methods in the
Applied Sciences, 35:1601–1610, 2012. https://doi.org/10.1002/mma.2549.
[16] R.K. Pandey and Arvind K.Singh. On the convergence of a finite difference
method for a class of singular boundary value problems arising in physiol-
ogy. Journal of Computational and Applied Mathematics, 166(2):553–564, 2004.
https://doi.org/10.1016/j.cam.2003.09.053.
[17] J. Rashidinia, R. Mohammadi and R. Jalilian. The numerical so-
lution of non-linear singular boundary value problems arising in phys-
iology. Applied Mathematics and Computation, 185(1):360–367, 2007.
https://doi.org/10.1016/j.amc.2006.06.104.
[18] R. Singh and J. Kumar. An efficient numerical technique for the solution of non-
linear singular boundary value problems. Computer Physics Communications,
185(4):1282–1289, 2014. https://doi.org/10.1016/j.cpc.2014.01.002.
[19] Z.H. Zhao, Y.Z. Lin and J. Niu. Convergence order of the reproducing kernel
method for solving boundary value problems. Mathematical Modelling and Anal-
ysis, 21(4):466–477, 2016. https://doi.org/10.3846/13926292.2016.1183240.

Math. Model. Anal., 23(1):33–43, 2018.