You are on page 1of 23

Second Order Methods for Doubly Singular

Boundary Value Problems


R.K. Pandey

and G. K. Gupta

Department of Mathematics, Indian Institute of Technology, Kharagpur, 721302, India.


Abstract
In this work we develop three numerical methods of second order accuracy for
doubly singular boundary value problems
(p(x)y

(x))

= q(x)f(x, y), 0 < x 1


with boundary conditions y(0) = A(or y

(0) = 0 or lim
x0
p(x)y

(x) = 0) and y(1) +


y

(1) = , where (> 0), ( 0), and A are nite constants. Here p(0) = 0 and
q(x) is unbounded near x = 0. The convergence of the methods is established under
quite general conditions which are veried by two examples.
Keywords: Doubly singular boundary value problems, Finite dierence method, Chawlas
identity, Thomas-Fermi equation.
MSC: 65L10, 65L12, 34B16
1 Introduction
Consider the following class of singular two point boundary value problems
Ly (p(x)y

= q(x)f(x, y), 0 < x 1, (1)


with boundary conditions
y (0) = A, y (1) + y

(1) = (2)
or, y

(0) = 0(or lim


x0
p(x)y

(x) = 0), y (1) + y

(1) = , (3)

Corresponding author. E-mail: rkp@maths.iitkgp.ernet.in

E-mail: gauravkgupta2@gmail.com
1
where > 0, 0 and A, are nite constants. Here p(0) = 0 and q(x) is unbounded
near x = 0. The condition p(0) = 0 says that the problem is singular and as q(x) is
unbounded near x = 0, so the problem is doubly singular ([1]). Let p(x), q(x) and f(x, y)
satisfy the following conditions:
(A1): (i) p(0) = 0, p(x) > 0 in (0, 1],
(ii) p(x) C[0, 1] C
1
(0, 1] ,
(iii) p(x) = x
b
0
g(x), (0 b
0
< 1), g(x) > 0 and
(iv) G C
3
[0, 1] and G
iv
(x) exists on (0, 1),
where G(x) = 1/g(x) on [0, 1].
(A2): (i) q(x) > 0 in (0, 1], q(x) is unbounded near x = 0,
(ii) q(x) L
1
(0, 1),
(iii) q(x) = x
a
0
H(x), H(0) > 0 with (a
0
> 1),
(iv) H(x) C
3
[0, 1] and H
iv
(x) exists on (0, 1) and
(v) 1 + a
0
b
0
0.
(A3): f(x, y) is continuous on {[0, 1] R},
f
y
exists, continuous and
f
y
0 for
all 0 x 1 and for all real y.
Thomas [16] and Fermi [8] independently derived a boundary value problem for deter-
mining the electrical potential in an atom. The analysis leads to the nonlinear singular
second order problem
y

= x
1/2
y
3/2
with a set of boundary conditions. The following are of our interest.
(i) the neutral atom with Bohr radius b given by y(0) = 1, by

(b) y(b) = 0;
(ii) the ionized atom given by y(0) = 1, y(b) = 0.
Furthermore, Chan and Hon [4] have considered the generalized Thomas-Fermi equation
(x
b
y

= cq(x)y
e
, y(0) = 1, y(a) = 0
for parameter values 0 b < 1, c > 0, d > 2, e > 1 and q(x) = x
b+d
.
Such singular problems have been the concern of researchers [2,3,7,9]. The existence-
uniqueness of the solution of the boundary value problem (1) with boundary condition
(2) or (3) is established in [1, 6, 12, 13, 14, 17]. Bobisud [1] has mentioned that in case
lim
x0
+
q(x)
p

(x)
= 0, the condition y

(0) = 0 is quite severe, i.e., it is sucient but not necessary


for forcing the solution to be dierentiable at x = 0. In fact if lim
x0
+
p(x)y

(x) = 0, then
lim
x0
+
y

(x) = lim
x0
+
q(x)
p

(x)
lim
x0
+
f(x, y(x)).
2
Thus lim
x0
+
y

(x) = 0 if either lim


x0
+
q(x)
p

(x)
= 0 or f(0, y(0)) = 0. But lim
x0
+
q(x)
p

(x)
= 0 if q(x)
has discontinuity at x = 0; thus it is natural to consider the weaker boundary condition
lim
x0
+
p(x)y

(x) = 0.
There is a considerable literature on numerical methods for q(x) = 1 but to the best
of our knowledge very few numerical methods are available to tackle doubly singular
boundary value problems. Reddien [15] has considered the linear form of (1) and derived
numerical methods for q(x) L
2
[0, 1] which is stronger assumption than (A2)(ii).
Chawla and Katti [5] have developed three second order accuracy methods M
1
, M
2
and M
3
for dierential equation (1) with q(x) = 1, p(x) = x

, 0 < 1 and boundary


conditions y(0) = A, y(1) = B. The method M
1
is based on nonuniform mesh and
one evaluation of the function f(x, y) while methods M
2
and M
3
are based on uniform
mesh and three evaluations of the function f(x, y). They claimed that the method M
3
based on uniform mesh is superior than other methods. Furthermore, the methods M
1
based on nonuniform mesh and M
2
based on uniform mesh have been extended in [11] for
nonnegative function p(x) satisfying (A1)(i iii) with G(x) analytic at singular point
x
0
and q(x) = p(x).
In this paper we extend all the three methods M
1
, M
2
and M
3
to doubly singular dif-
ferential equation (1) with boundary conditions (2) or (3). All the methods are developed
for both uniform as well as appropriate nonuniform mesh. The second order convergence
of the methods are established under quite general conditions on p(x), q(x) and f(x, y).
For p(x) = 1, q(x) = 1, the methods M
1
and M
2
both reduce to the classical second
order method based on one evaluation of the function f(x, y). For nonuniform mesh
the conditions on the function f(x, y) are weaker than that of uniform mesh. Thus the
methods based on nonuniform mesh can tackle wider class of doubly singular problems.
The method M
3
based on nonuniform mesh is superior to the other methods which is
corroborated through numerical examples.
2 Finite dierence methods
This section is divided in two parts (i) Description of the methods and (ii) Construction
of the methods. All coecients not specied explicitly in this section are specied in the
appendix.
2.1 Description of the Methods
In this section we rst state the methods; their detailed construction process is given in
section 2.2.
For positive integer N 2, we consider the mesh w
h
= {x
k
}
N
k=0
over [0, 1]: 0 = x
0
<
x
1
< x
2
< x
N
= 1, h =
1
N
. Let r(x) := f(x, y(x)) (y(x) is the solution), y
k
= y(x
k
)
3
etc. Now we approximate the dierential operator Ly on the grid w
h
by the following
dierence operators
(L
h
y)
k
= y
k1
/J
k1
(1/J
k
+ 1/J
k1
) y
k
+ y
k+1
/J
k
, k = 1(1)(N 1), (4)
(L
h
y)
N
= y
N1
/J
N1
(1/J
N1
+ /(G
N
)) y
N
+ /(G
N
), (5)
for boundary value problem (1)-(2). In the case of boundary condition (3) the operator
is approximated by
(L
h
y)
1
= y
1
/J
1
+ y
2
/J
1
. (6)
Here y = ( y
k
) denotes the approximate solution, y
k
y
k
, G
k
= G(x
k
), and
J
k
=
x
k+1

x
k
(p())
1
d. (7)
Note that the functions p(x) and G(x) are dened in assumption (A1).
We now state three dierent methods to solve the singular boundary value problem (1)
with boundary conditions (2) or (3).
Method M
1
The rst method M
1
for boundary value problem (1)-(2) is given by
(L
h
y)
k
= a
M
1
0,k
r
k
, k = 1(1)(N 1), (8)
(L
h
y)
N
= a
M
1
0,N
r
N
. (9)
In the case of boundary condition (3) the modied equation for k = 1 is
(L
h
y)
1
= a
M
1
0,1
r
1
+ a
M
1
1,1
r
2
. (10)
Method M
2
The second method M
2
for boundary value problem (1)-(2) is given by
(L
h
y)
k
= a
M
2
0,k
r
k
+ a
M
2
1,k
r
k+1
a
M
2
1,k
r
k1
, k = 1(1)(N 1), (11)
(L
h
y)
N
= a
M
2
0,N
r
N
. (12)
In the case of boundary condition (3) the modied equation for k = 1 is
(L
h
y)
1
= a
M
2
0,1
r
1
+ a
M
2
1,1
r
2
. (13)
Method M
3
The third method M
3
for boundary value problem (1)-(2) is given by
(L
h
y)
k
= a
M
3
0,k
r
k
+ a
M
3
1,k
r
k+1
a
M
3
1,k
r
k1
, k = 1(1)(N 1), (14)
(L
h
y)
N
= a
M
3
0,N
r
N
a
M
3
1,N
r
N1
. (15)
4
In the case of boundary condition (3) the modied equation for k = 1 is
(L
h
y)
1
= a
M
2
0,1
r
1
+ a
M
2
1,1
r
2
. (16)
The coecients a
M
i
i,j
are specied in the appendix.
All the three methods are developed for both uniform mesh (with mesh point x
k
=
kh, h =
1
N
) and nonuniform mesh (with mesh point x
k
= (kh)
1
(1b
0
)
).
2.2 Construction of the methods
In this section we describe the derivation of method M
1
and the derivation of other
methods follows in a similar manner. Local truncation errors are mentioned without
proof.
2.2.1 Method M
1
With z (x) = p (x) y

(x) the dierential equation (1) becomes z

= q(x)f(x, y(x)). Then


an approximation for the dierential operator Ly on the mesh w
h
is obtained as follows:
Integrating z

= q(x)f(x, y(x)) twice, rst from x


k
to x and then from x
k
to x
k+1
and
change the order of integration to get the following
y
k+1
y
k
= z
k
J
k
+
x
k+1

x
k

x
k+1

t
(p ())
1
d

q (t) r (t) dt, (17)


where z
k
= z(x
k
), r(x) := f(t, y(x)) and J
k
=

x
k+1
x
k
(p ())
1
d. In an analogous manner,
we get
y
k
y
k1
= z
k
J
k1

x
k

x
k1

x
k1
(p ())
1
d

q (t) r (t) dt. (18)


Eliminating z
k
from (17) and (18) we obtain the Chawlas identity
y
k+1
y
k
J
k

y
k
y
k1
J
k1
=
I
+
k
J
k
+
I

k
J
k1
, k = 1 (1) (N 1), (19)
where
I
+
k
=
x
k+1

x
k

x
k+1

t
(p ())
1
d

q(t)r (t) dt,


I

k
=
x
k

x
k1

x
k1
(p ())
1
d

q(t)r (t) dt.


5
Now putting Taylor series expansion of G(x), r(x) and H(x) about x
k
in I

k
we get the
approximation for the smooth solution y(x) as:
(L
h
y)
k
= a
M
1
0,k
r
k
+ t
(1)
k
, k = 1(1)(N 1). (20)
Here the local truncation error t
(1)
k
is given by
t
(1)
k
= b
10,k
G
k
H
k
r

k
+ (b
10,k
G
k
H

k
+ b
01,k
G

k
H
k
) r
k
+

b
11,k
G

k
H

k
+
1
2
b
20,k
G
k
H

(
k
)

r
k
+
1
2
b
02,k
G

(
k
)H
k
r
k
+ (b
11,k
G

k
H
k
+ b
20,k
G
k
H

k
) r

k
+
1
2
b
20,k
G
k
H
k
r

(
k
),
k = 1(1)(N 1), x
k1
<
k
,
k
< x
k+1
; (21)
coecient is specied in the appendix.
Note that the functions G(x) and H(x) are dened in assumption (A1) and (A2)
respectively.
2.2.2 Discretization of the boundary condition at x = 1
We write (18) for k = N
y
N
J
N1

y
N1
J
N1
= p
N
y

N

I

N
J
N1
. (22)
Now, from boundary condition at x = 1 and Taylor series expansion of r(x) we get the
required discretization as follows
(L
h
y)
N
= a
M
1
0,N
r
N
+ t
(1)
N
. (23)
Here the local truncation error t
(1)
N
is given by
t
(1)
N
=
a

10,N
J
N1
G
N
H
N
r

N
+

10,N
J
N1
G
N
H

N
+
a

01,N
J
N1
G

N
H
N

r
N
+
1
2
a

02,N
J
N1
G

(
N
)H
N
r
N
+

11,N
J
N1
G

N
H

N
+
1
2
a

20,N
J
N1
G
N
H

r
N
+

11,N
J
N1
G

N
H
N
+
a

20,N
J
N1
G
N
H

N
+
1
2
a

20,N
J
N1
G
N
H
N
r

N
), x
N1
<

N
< x
N
; (24)
coecient is specied in the appendix.
6
2.2.3 Discretization of the boundary condition at x = 0
Integrating the dierential equation (1) twice, rst from x
1
to x; then from x
1
to x
2
and
interchanging the order of integration we get
y
2
y
1
= J
1
x
1

0
q(t)r(t)dt +
x
2

x
1

x
2

t
(p())
1
d

q(t)r(t)dt.
Now, using Taylor series expansion of r(x), H(x) and G(x) about x = x
1
we get the
following discretization for k = 1
(L
h
y)
1
= a
M
1
0,1
r
1
+ a
M
1
1,1
r
2
+ t
(1)
1
, (25)
t
(1)
1
=

a
+
11,1
J
1
G

1
H

1
+
1
2
a
+
02,1
J
1
G

1
H
1
+
x
3+a
0
1
(3)
H

1
+
1
2
a
+
20,1
J
1
G
1
H

r
1
+

2x
3+a
0
1
(3)
H

1
+
a
+
20,1
J
1
G
1
H

1
+
a
+
11,1
J
1
G

1
H
1

1
+

x
3+a
0
1
(3)
+
1
2
a
+
20,1
J
1
G
1
H
1

(
1
), 0 <
1
< x
2
. (26)
Here t
(1)
1
is the local truncation error; coecient is specied in the appendix. The functions
G(x) and H(x) are dened in assumption (A1) and (A2).
The methods M
2
and M
3
can be obtained by using central dierence and forward dierence
approximations for r

k
respectively.
2.2.4 Computation of y
0
To compute y
0
we use the following:
y
0
= y
1
+
x
2+a
0
b
0
1
(1 + a
0
)(2 + a
0
b
0
)
H
1
G
1
r
1
+ x
3+a
0
b
0
1

1
(1 + a
0
)(2 b
0
)
+
1
(2 + a
0
b
0
)(3 + a
0
b
0
)
+
1
(2 b
0
)(3 + a
0
b
0
)

1
H
1
r
1
+

1
(1 + a
0
)(2 + a
0
)
+
1
(3 + a
0
b
0
)

G
1
H

1
r
1

+x
2+a
0
b
0
1

1
(1 + a
0
)(2 + a
0
)
+
1
(3 + a
0
b
0
)

(r
2
r
1
)
+O(h
4+a
0
b
0
). (27)
which can be obtained by integrating (py

= q(x)r(x) rst from 0 to x, then from 0 to


x
1
.
7
3 Convergence of the methods
In this section we show that under suitable conditions methods M
1
and M
2
are of second
order accuracy on uniform mesh as well as nonuniform mesh. The convergence analysis for
the method M
3
follows in a similar manner. For convergence analysis we rst introduce
the matrix notations.
Let

Y = ( y
1
, ..., y
N
)
T
, R(

Y ) = ( r
1
, ..., r
N
)
T
, Q = (q
1
, ..., q
N
)
T
and T = (t
1
, ..., t
N
)
T
,
then the nite dierence methods (M
1
, M
2
and M
3
) can be expressed in matrix form as
B

Y + PR(

Y ) = Q, (28)
and for smooth Y = (y
1
, ..., y
N
)
T
, the methods can be written as
BY + PR(Y ) + T = Q, (29)
where B = (b
ij
) and P = (p
ij
) are (N N) tridiagonal matrices.
Let E =

Y Y = (e
1
, ...e
N
)
T
, then from Mean Value Theorem R(

Y ) R(Y ) = ME,
M = diag{U
1
, ..., U
N
} (U
k
=
f
k
y
k
0), then from equation (28) and (29) we get the error
equation as
(B + PM)E = T. (30)
We recall that by the notation Z 0 we mean that all components z
i
of the vector Z
satisfy z
i
0. Similarly by B 0 we mean all the elements b
ij
of the matrix B satisfy
b
ij
0.
From Corollary of Theorem 7.2 and Theorem 7.4 ([10]) a tridiagonal matrix B is
monotone if
b
k,k+1
0, b
k,k1
0 and
N

j=1
b
k,j

0, k = 1, 2, ..., N
> 0, for at least one i.
Furthermore, from Theorem 7.3 ([10]), B
1
exists and the elements of B
1
are nonneg-
ative. Now, if B + PM is monotone and B + PM B, then from from Theorem 7.5
([10])
(B + PM)
1
B
1
. (31)
Let vector norm E and matrix norm B be dened by
E = max
1kN
|e
k
| and B = max
1kN
N

j=1
|b
kj
|,
then from (30) we get
E = (B + PM)
1
|T|. (32)
Furthermore, if B + PM B and B is also monotone matrix, then from (31) we get
E B
1
|T|. (33)
8
3.1 Uniform Mesh Case
In this section we establish the convergence of the methods M
1
and M
2
. The convergence
of the method M
3
follows in the similar manner.
Let V = (v
1
, . . . , v
N
)
T
where v
j
= (2/) +2
1
2
(x
j
+ 1)
2
. Let W = (w
1
, . . . , w
N
)
T
=
BV , then w
1
> 0, w
N
> 0 and
w
k
=
1
J
k
(v
k
v
k+1
)
1
J
k1
(v
k1
v
k
) ,
=
h
J
k

1 + x
k
+
h
2

h
J
k1

1 + x
k

h
2

1
J
k

1
J
k1

. (34)
Now from J
k
=

x
k+1
x
k
x
b
0
G(x)dx, for xed x
k
and h 0, we get
J
k
=
x
1b
0
k+1
x
1b
0
k
1 b
0
G
k
+ O(h
2
)
hx
b
0
k
G
k
.
Thus, for xed x
k
and suciently small h we get
w
k
>

b
0
h
2G
k

x
b
0
1
k
, k = 2 (1) (N 1). (35)
Method M
1
with boundary conditions y(0) = A, y(1) + y

(1) =
The elements of the matrices B, P and the vector Q for the method M
1
are as follows:
b
k,k1
=
1
J
k1
, k = 2(1)N; b
k,k
=

1
J
k
+
1
J
k1

, k = 1(1)(N 1);
b
k,k+1
=
1
J
k
, k = 1(1)(N 1); b
N,N
=

1
J
N
+

G
N

;
q
1
=
A
J
0
, q
N
=

G
N
, q
k
= 0, k = 2(1)(N 1);
p
k,k
= a
M
1
0,k
, k = 1(1)(N 1); p
k1
= 0, p
N,N
=
a

00,N
J
N1
G
N
H
N
. (36)
As J
k
> 0 and for suciently small h, the matrices B and B+PM are irreducible and
monotone. Thus B
1
and (B + PM)
1
exist and nonnegative. Now, since PM 0 for
suciently small h, it follows that (B + PM)
1
B
1
.
Let B
1
= (

b
ij
), from B
1
W = V , w
1
> 0 and w
N
> 0 we get
N1

j=2

b
k,j
w
k
v
j
< (2/) +
3
2
, k = 1 (1) N.
9
Now, in view of w
1
> 0, w
N
> 0 and (35) we get

b
0
h
2

N1

j=2

b
k,j
G
k
x
b
0
1
k
< (2/) +
3
2
, k = 1 (1) N. (37)
Let Z = (1, , 1)
T
, then S = (s
1
, , s
N
)
T
= BZ denotes the vector of row sums of
B. Now, from B
1
S = Z we obtain

b
k,1
= 1/s
1
= J
0

b
k,1

h
1b
0
(1 b
0
)
G
0
, k = 1 (1) N (38)
and

b
k,N
=
1
s
N

G
N
, k = 1 (1) N. (39)
Next we obtain bounds for truncation error t
(1)
k
. For this we assume that for x (0, 1]
x
1+a
0
b
0
|r
(i)
(x)| N
i
, i = 0(1)2,
for suitable constants N
i
, i = 0(1)2. Now, from the approximation of coecients given in
appendix it follows that for xed x
k
and suciently small h
|t
(1)
k
| C
1
h
3
x
b
0
1
k
, |t
(1)
N
| C
1
h
2
, (40)
for suitable constant C
1
. From (33) and (37)-(39) we get
|e
i
|

b
k,1
|t
(1)
1
| +
N1

j=2

b
k,j
|t
(1)
j
| +

b
k,N
|t
(1)
N
|
h
2

C
1
h

b
k,1
+
b
0
h
2
N1

j=1

b
k,j
G
k
x
b
0
1
k
+

b
k,N

.
Thus E = O(h
2
) .
Method M
2
with boundary conditions y(0) = A, y(1) + y

(1) =
For method M
2
the elements of matrix B and the vector Q can be given by (36), while
the elements of matrix P are as follows:
p
k,k
= (b
00,k
G
k
H
k
+ b
01,k
H
k
G

k
+ b
10,k
G
k
H

k
) , k = 1(1)N;
p
k1
=
1
2h
b
10,k
G
k
H
k
, p
N,N
=
a

00,N
J
N1
G
N
H
N
. (41)
Here one may note that we can not bound B + PM by B as PM is not necessarily be
nonnegative and hence we work with (B + PM)
1
.
10
Let Z = (1, , 1)
T
then S

= (s

1
, , s

N
)
T
= (B +PM)Z denotes the vector of row
sums of B + PM. From S

= (B + PM)Z, it is easy to see that


s

1
>
1
2J
0
, s

N
>

G
N
, and s

k
> hH
k
x
a
0
k
, k = 1(1)(N 1).
Now, for xed x
k
and suciently small h, from (B + PM)
1
S

= Z we get
(B + PM)
1
k,1

1
s

2h
1b
0
G
1
(1 b
0
)
, (B + PM)
1
k,N

1
J
N1
s

h
1b
0

, and
(B + PM)
1
k,j

1
min
2jN1
s

j
, j = 1(1)(N 1)
1
hH
k
x
a
0
k
. (42)
The local truncations errors t
(2)
k
and t
(2)
N
are given as
t
(2)
k
=

b
11,k
G

k
H

k
+
1
2
b
20,k
G
k
H

(
k
) +
1
2
b
02,k
G

(
k
)H
k

r
k
+ (b
11,k
G

k
H
k
+ b
20,k
G
k
H

k
) r

k
+
1
2
b
20,k
G
k
H
k
r

(
k
)
h
2
6
b
10,k
G
k
H
k
r

(
k
), k = 1(1)(N 1), x
k1
<
k
< x
k
,
t
(2)
N
=
a

10,N
J
N1
G
N
H
N
r

N
+

10,N
J
N1
G
N
H

N
+
a

01,N
J
N1
G

N
H
N

r
N
+
1
2
a

02,N
J
N1
G

(
N
)H
N
r
N
+

11,N
J
N1
G

N
H

N
+
1
2
a

20,N
J
N1
G
N
H

r
N
+

11,N
J
N1
G

N
H
N
+
a

20,N
J
N1
G
N
H

N
+
1
2
a

20,N
J
N1
G
N
H
N
r

(
N
), x
N1
<
N
< x
N
. (43)
Next we obtain bounds for truncation error t
(2)
k
. For this we assume that for x (0, 1]
|r
(i)
(x)| N
i
, i = 0(1)2,
for suitable constants N
i
, i = 0(1)2. Now, from the approximation of coecients given in
appendix it follows that, for xed x
k
and suciently small h
|t
(2)
k
| C
1
h
3
x
a
0
k
, k = 1(1)(N 1); |t
(2)
N
| C
3
h
2
, (44)
where C
1
and C
3
are suitable constants. Now, since ||E|| ||(B + PM)
1
T||, with the
help of (42) and (44) we obtain ||E|| = O(h
2
).
3.2 Nonuniform Mesh Case
In this section also we establish the second order convergence for the methods M
1
, M
2
and the analysis for the method M
3
follows in similar manner.
11
Let V = (v
1
, . . . , v
N
)
T
where v
j
= (2/) +2
1
2
(x
j
+ 1)
2
. Let W = (w
1
, . . . , w
N
)
T
=
BV , then
w
k
=
1
J
k
(v
k
v
k+1
)
1
J
k1
(v
k1
v
k
) ,
=
h
J
k

1 + x
k
+
h
2

h
J
k1

1 + x
k

h
2

1
J
k

1
J
k1

. (45)
Now, from J
k
=

x
k+1
x
k
x
b
0
G(x)dx, for xed x
k
(= (kh)
1
1b
0
) and h 0, we get
J
k
=
x
1b
0
k+1
x
1b
0
k
1 b
0
G
k
+ O(h
2
)

h
1 b
0
G
k
.
Thus for xed x
k
and suciently small h we get
w
k
>

b
0
h
(1 b
0
) G
k

x
2b
0
1
k
, k = 2 (1) (N 1). (46)
Method M
1
with boundary conditions y(0) = A, y(1) + y

(1) =
For this method the elements of the matrices B, P and the vector Q are given by (36). As
J
k
> 0 then for suciently small h we get that the matrices B and B+PM are irreducible
and monotone. Furthermore, in view of PM 0 we conclude that B
1
, (B + PM)
1
exist, nonnegative and (B + PM)
1
B
1
.
Since
Let B
1
= (

b
ij
), from from B
1
W = V , w
1
> 0 and w
N
> 0 we get
N1

j=2

b
k,j
w
k
v
j
< (2/) +
3
2
, k = 1 (1) N.
Now in view of w
1
> 0, w
N
> 0 and (46) we get

b
0
h
(1 b
0
)

N1

j=2

b
k,j
G
k
x
2b
0
1
k
< (2/) +
3
2
, k = 1 (1) N. (47)
To obtain the bounds for the truncation error t
(1)
k
and t
(1)
N
, we assume that for x (0, 1]
|x
1+a
0
+b
0
r
(i)
(x)| N
i
, i = 0(1)2,
for suitable constants N
i
, i = 0(1)2, then from suciently small h
|t
(1)
k
| C
1
h
3
12(1 b
0
)
3
G
k
x
2b
0
1
k
, |t
(1)
N
| C
4
h
2
12(1 b
0
)
3
G
N
12
and
|t
(1)
1
| C
5
h
2
x
b
0
1
(1 b
0
)
3
,
where C
1
and C
4
and C
5
are suitable constants. Now following the analysis of uniform
mesh case we get that ||E|| = O(h
2
).
Method M
2
with boundary conditions y(0) = A, y(1) + y

(1) =
For the method M
2
the elements of the matrix B and the vector Q are given by (36) and
elements of matrix P are given by (41).
To obtain the bounds for the truncation error t
(2)
k
and t
(2)
N
, we assume that for x (0, 1]
|x
2b
0
r
(i)
(x)| N
i
, i = 0(1)2,
for suitable constants N
i
, i = 0(1)2. Then for suciently small h
|t
(2)
k
| C
1
h
3
(1 b
0
)
3
x
a
0
+b
0
k
, |t
(2)
N
| C
6
h
2
,
where C
1
and C
6
are suitable constants. Now following the analysis of uniform mesh case
we get that ||E|| = O(h
2
).
Thus we have established the following result.
Theorem 3.1 Assume that p(x), q(x) and f(x, y) satisfy assumptions given in (A1),
(A2) and (A3) respectively. Let r(x) := f(x, y(x)), then the method M
1
is of second
order accuracy for suciently small mesh size h provided
(i) x
1+a
0
b
0
r
(i)
(x), i = 0(1)2 are bounded on (0, 1] in case of uniform mesh x
k
= kh;
and
(ii) x
1+a
0
+b
0
r
(i)
(x), i = 0(1)2 are bounded on (0, 1] in case of nonuniform mesh x
k
=
(kh)
1
1b
0
.
Theorem 3.2 Assume that p(x), q(x) and f(x, y) satisfy assumptions given in (A1),
(A2) and (A3) respectively. Let r(x) := f(x, y(x)), then the method M
2
is of second
order accuracy for suciently small mesh size h provided
(i) r
(i)
(x), i = 0(1)2 are bounded on (0, 1] in case of uniform mesh x
k
= kh; and
(ii) x
2b
0
r
(i)
(x), i = 0(1)2 are bounded on (0, 1] in case of nonuniform mesh x
k
=
(kh)
1
1b
0
.
Theorem 3.3 Assume that p(x), q(x) and f(x, y) satisfy assumptions given in (A1),
(A2) and (A3) respectively. Let r(x) := f(x, y(x)), then the method M
3
is of second
order accuracy for suciently small mesh size h provided
13
(i) x
1+a
0
b
0
r
(i)
(x), i = 0(1)2 are bounded on (0, 1] in case of uniform mesh x
k
= kh;
and
(ii) x
1+a
0
+b
0
r
(i)
(x), i = 0(1)2 are bounded on (0, 1] in case of nonuniform mesh x
k
=
(kh)
1
1b
0
.
4 Remarks
1. The method M
1
based on one evaluation of data function f(x, y), while methods
M
2
and M
3
are based on three evaluations of f(x, y).
2. Chawla and Katti [5] have claimed that the method M
3
based on uniform mesh is
superior than other methods but from the tables it can bee seen that method M
3
based on nonuniform mesh provides better results than that of M
3
based on uniform
mesh. So the method based on nonuniform mesh is superior than other methods.
3. This work also extends the work of Pandey and Singh [11] to wider class of functions
p(x), q(x) and f(x, y) as apart from other weaker conditions we do not assume that
G(x) is analytic at the singular point.
4. The nonuniform mesh methods can tackle wider class of problems that corresponding
uniform mesh method.
5. The method M
3
based on nonuniform mesh is better that other methods as evident
from the maximum absolute errors displayed in Table 1 and Table 2.
5 Numerical Illustrations
To illustrate the convergence of the methods and to corroborate their order of the accuracy,
we apply the methods to following examples. The maximum absolute errors are displayed
in Table 1 and Table 2.
Example 5.1 In this example lim
x0
p(x)y

(x) = 0 but y

(0) = 0.
(x
b
0
y

=
x
b
0
1
1 + e
x

b
0
e
x
+ xe
2x
e
y
xe
x

lim
x0
p(x)y

(x) = 0(or y(0) = ln(1/2))


and y(1) + 5y

(1) = ln

1
1 + e

5e
(1 + e)
with exact solution y(x) = ln(1/(1 + e
x
)). Here a
0
= b
0
1.
14
Example 5.2 In this example G(x) = (1+x
3.5
). Thus G(x) C
3
[0, 1] and G
iv
(x) exists
on (0, 1), but G(x) is not an analytic function.

x
b
0
1
1 + x
3.5
y

=
(3.5 b
0
)
(1 + x
3.5
)(4 + x
3.5b
0
)

(3.5 b
0
)x
5.5b
0
e
y

2.5x
2
+
3.5x
5.5
(1 + x
3.5
)

(0) = 0(or y(0) = ln(1/4))


and y(1) + 5y

(1) = ln(1/5) (3.5 b


0
)
with exact solution y(x) = ln(1/(4 + x
3.5b
0
)). Here a
0
= 0.5.
All the three methods M
1
, M
2
and M
3
are applied on the both the examples for b
0
=
0.2, 0.5. Maximum absolute errors and order of accuracy for Example 5.1 and Example
5.2 are displayed in Table 1 and Table 2 respectively.
5.1 Numerical Results for Uniform/Nonuniform Mesh Case
Table 1: Maximum absolute errors and order of the methods for Example 5.1
N b
0
= 0.2 Order b
0
= 0.5 Order b
0
= 0.2 Order b
0
= 0.5 Order
Uniform Mesh:y(0) = ln(1/4) Nonuniform Mesh:y(0) = ln(1/4)
Method M
1
512 1.37(6)
a
3.71(-6) 5.40(-7) 7.84(-7)
1024 3.72(-7) 1.88 9.99(-7) 1.90 1.35(-7) 2.00 1.96(-7) 2.00
2048 1.01(-7) 2.68(-7) 3.39(-8) 4.90(-8)
4096 2.70(-8) 1.90 7.14(-8) 1.91 8.55(-9) 2.00 1.22(-8) 2.00
Method M
2
512 1.40(-7) 3.30(-7) 2.05(-7) 1.17(-6)
1024 3.50(-8) 2.00 8.25(-8) 2.00 5.11(-8) 2.00 2.91(-7) 2.00
2048 8.77(-9) 2.06(-8) 1.28(-8) 7.29(-8)
4096 2.20(-9) 2.00 5.14(-9) 2.00 3.20(-9) 2.00 1.82(-8) 2.00
Method M
3
512 4.96(-8) 1.04(-7) 6.28(-9) 2.57(-8)
1024 1.24(-8) 2.00 2.59(-8) 2.00 1.57(-9) 2.00 6.41(-9) 2.00
2048 3.09(-9) 6.47(-9) 3.91(-10) 1.60(-9)
4096 7.62(-10) 2.00 1.63(-9) 2.00 9.23(-11) 2.08 4.00(-10) 2.00
15
N b
0
= 0.2 Order b
0
= 0.5 Order b
0
= 0.2 Order b
0
= 0.5 Order
Uniform Mesh:lim
x0
p(x)y

(x) = 0 Nonuniform Mesh:lim


x0
p(x)y

(x) = 0
Method M
1
512 1.38(-6) 2.28(-6) 1.63(-6) 8.43(-7)
1024 3.70(-7) 1.88 7.80(-7) 1.89 4.08(-7) 2.00 2.10(-7) 2.00
2048 9.73(-8) 2.10(-7) 1.02(-7) 5.25(-8)
4096 2.56(-8) 1.90 5.55(-8) 1.91 2.55(-8) 2.00 1.31(-8) 2.00
Method M
2
512 6.81(-7) 7.37(-7) 8.87(-7) 2.44(-6)
1024 1.68(-7) 2.02 1.84(-7) 2.00 2.22(-7) 2.00 6.10(-7) 2.00
2048 4.19(-8) 4.61(-8) 5.56(-8) 1.52(-7)
4096 1.05(-8) 2.00 1.15(-8) 2.00 1.39(-8) 2.00 3.82(-8) 2.00
Method M
3
512 4.97(-8) 1.04(-7) 6.28(-9) 2.57(-8)
1024 1.24(-8) 2.00 2.59(-8) 2.00 1.57(-9) 2.00 6.41(-9) 2.00
2048 3.09(-9) 6.47(-9) 3.91(-10) 1.60(-9)
4096 7.62(-10) 2.02 1.61(-9) 2.00 9.23(-11) 2.08 4.01(-10) 2.00
a
1.37(6) = 1.37 10
6
Table 2: Maximum absolute errors and order of the methods for Example 5.2
N b
0
= 0.2 Order b
0
= 0.5 Order b
0
= 0.2 Order b
0
= 0.5 Order
Uniform Mesh:y(0) = ln(1/4) Nonuniform Mesh:y(0) = ln(1/4)
Method M
1
512 8.01(-7) 1.53(-6) 6.53(-7) 9.76(-7)
1024 2.00(-7) 2.00 3.82(-7) 2.00 1.64(-7) 2.00 2.45(-7) 2.00
2048 5.01(-8) 9.55(-8) 4.09(-8) 6.13(-8)
4096 1.25(-8) 2.00 2.38(-8) 2.00 1.02(-8) 2.00 1.53(-8) 2.00
Method M
2
512 4.13(-7) 5.02(-7) 4.45(-7) 7.94(-7)
1024 1.03(-7) 1.99 1.26(-7) 2.00 1.11(-7) 2.00 1.97(-7) 2.00
2048 2.58(-8) 3.16(-8) 2.78(-8) 4.92(-8)
4096 6.45(-9) 2.00 7.78(-9) 2.00 6.91(-9) 2.00 1.23(-8) 2.00
Method M
3
512 3.11(-7) 4.43(-7) 3.42(-8) 4.11(-7)
1024 7.76(-8) 2.00 1.11(-7) 2.00 8.51(-9) 2.00 1.04(-7) 2.00
2048 1.94(-8) 2.76(-8) 2.12(-9) 2.62(-8)
4096 4.83(-9) 2.00 6.77(-9) 2.00 5.24(-10) 2.01 6.59(-9) 2.00
16
N b
0
= 0.2 Order b
0
= 0.5 Order b
0
= 0.2 Order b
0
= 0.5 Order
Uniform Mesh:y

(0) = 0 Nonuniform Mesh:y

(0) = 0
Method M
1
512 2.47(-6) 6.13(-6) 3.42(-6) 1.80(-6)
1024 9.69(-7) 1.64 1.92(-6) 1.68 8.67(-7) 1.98 4.52(-7) 1.99
2048 3.05(-7) 5.50(-7) 2.18(-7) 1.13(-7)
4096 8.76(-8) 1.81 1.49(-7) 1.89 5.41(-8) 2.01 2.84(-8) 2.00
Method M
2
512 9.14(-6) 9.07(-6) 3.11(-7) 2.58(-6)
1024 1.99(-6) 2.20 1.95(-6) 2.21 6.70(-8) 2.21 6.40(-7) 1.99
2048 4.43(-7) 4.34(-7) 1.54(-8) 1.59(-7)
4096 1.01(-7) 2.12 9.96(-8) 2.12 3.85(-9) 2.00 3.97(-8) 2.00
Method M
3
512 3.06(-7) 3.97(-7) 3.43(-8) 1.04(-7)
1024 7.71(-8) 1.99 1.05(-7) 1.92 8.51(-9) 2.01 2.62(-8) 1.99
2048 1.93(-8) 2.69(-8) 2.12(-9) 6.59(-9)
4096 4.83(-9) 2.00 6.69(-9) 2.01 5.24(-10) 2.02 1.65(-9) 2.00
A Appendix
Here we mention the truncation errors which are not mentioned in the previous sections.
The coecients involved in the methods and their approximations are also mentioned.
Truncation errors for methods M
2
and M
3
t
(2)
1
=

a
+
11,1
J
1
G

1
H

1
+
1
2
a
+
02,1
J
1
G

(
1
)H
1

f
1
+

x
3+a
0
1
(1 + a
0
)(2 + a
0
)(3 + a
0
)
H

(
1
) +
1
2
a
+
20,1
J
1
G
1
H

(
1
)

f
1
+

2x
3+a
0
1
(1 + a
0
)(2 + a
0
)(3 + a
0
)
H

1
+
a
+
20,1
J
1
G
1
H

1
+
a
+
11,1
J
1
G

1
H
1

1
+

x
3+a
0
1
(1 + a
0
)(2 + a
0
)(3 + a
0
)
+
1
2
a
+
20,1
J
1
G
1
H
1

(
1
), t
(3)
1
= t
(2)
1
,
t
(3)
k
=

b
11,k
G

k
H

k
+
1
2
b
20,k
G
k
H

(
k
) +
1
2
b
02,k
G

(zeta
k
)H
k

f
k
+(b
11,k
G

k
H
k
+ b
20,k
G
k
H

k
) f

k
+

1
2
b
20,k

h
2
a
+
10,k
J
k
+
h
2
a

10,k
J
k1

G
k
H
k
f

(
k
), k = 1(1)(N 1),
t
(3)
N
=
1
2
a

02,N
J
N1
G

(
N
)H
N
f
N
+

11,N
J
N1
G

N
H

N
+
1
2
a

20,N
J
N1
G
N
H

(
N
)

f
N
17
+

11,N
J
N1
G

N
H
N
+
a

20,N
J
N1
G
N
H

N
+
1
2
a

20,N
J
N1
G
N
H
N
f

(
N
).
Coecients for method M
1
a
M
1
0,k
=

b
00,k
G
k
H
k
+ b
10,k
G
k
H

k
+ b
01,k
G

k
H
k
+ b
11,k
G

k
H

k
+
1
2
b
20,k
G
k
H

k
+
1
2
b
02,k
G

k
H
k

,
a
M
1
0,N
=
1
J
N1

00,N
G
N
H
N
+ a

10,N
G
N
H

N
+ a

01,N
G

N
H
N
+ a

11,N
G

N
H

N
+
1
2
a

02,N
G

N
H
N
+
1
2
a

20,N
G
N
H

,
a
M
1
0,1
=

x
1+a
0
1
1 + a
0
H
1
+
a
+
00,1
J
1
H
1
G
1

x
2+a
0
1
(2)
H

1
+
a
+
10,1
J
1
G
1
H

1
+
a
+
01,1
J
1
H
1
G

1
(x
2
x
1
)

x
2+a
0
1
(3)
H
1
+
a
+
10,1
J
1
G
1
H
1

a
+
11,1
J
1
G

1
H

1
+
1
2
a
+
02,1
J
1
G

1
H
1

x
3+a
0
1
(3)
H

1
+
1
2
a
+
20,1
J
1
G
1
H

, a
M
1
1,1
=
1
(x
2
x
1
)

x
2+a
0
1
(3)
H
1
+
a
+
10,1
J
1
G
1
H
1

.
Coecients for method M
2
a
M
2
0,k
=

b
00,k
G
k
H
k
+ b
01,k
H
k
G

k
+ b
10,k
G
k
H

k
+ b
11,k
G

k
H

k
+
1
2
b
20,k
G
k
H

k
+
1
2
b
02,k
G

k
H
k

,
a
M
2
1,k
=

1
(x
k+1
x
k1
)
b
10,k
H
k
G
k
+
1
(x
k+1
x
k1
)
(b
11,k
G

k
H
k
+ b
20,k
G
k
H

k
)

,
a
M
2
1,k
=

1
(x
k+1
x
k1
)
b
10,k
H
k
G
k
+
1
(x
k+1
x
k1
)
(b
11,k
G

k
H
k
+ b
20,k
G
k
H

k
)

,
a
M
2
0,N
=
1
J
N1

00,N
G
N
H
N
+ a

10,N
G
N
H

N
+ a

01,N
G

N
H
N
+ a

11,N
G

N
H

N
+
1
2
a

02,N
G

N
H
N
+
1
2
a

20,N
G
N
H

,
a
M
2
0,1
=

x
1+a
0
1
1 + a
0
H
1
+
a
+
00,1
J
1
H
1
G
1

x
2+a
0
1
(1 + a
0
)(2 + a
0
)
H

1
+
a
+
10,1
J
1
G
1
H

1
+
a
+
01,1
J
1
H
1
G

1
(x
2
x
1
)

x
2+a
0
1
(1 + a
0
)(2 + a
0
)
H
1
+
a
+
10,1
J
1
G
1
H
1

+
a
+
11,1
J
1
G

1
H

1
+
1
2J
1
a
+
20,1
G
1
H

1
(x
2
x
1
)

2x
3+a
0
1
(1 + a
0
)(2 + a
0
)(3 + a
0
)
H

1
+
a
+
20,1
J
1
G
1
H

1
+
a
+
11,1
J
1
G

1
H
1

+
x
3+a
0
1
(1 + a
0
)(2 + a
0
)(3 + a
0
)
H

1
+
1
2J
1
a
+
20,1
G
1
H

,
18
a
M
2
1,1
=

1
(x
2
x
1
)

x
2+a
0
1
(1 + a
0
)(2 + a
0
)
H
1
+
a
+
10,1
J
1
G
1
H
1

+
1
(x
2
x
1
)

+
2x
3+a
0
1
(1 + a
0
)(2 + a
0
)(3 + a
0
)
H

1
+
a
+
20,1
J
1
G
1
H

1
+
a
+
11,1
J
1
G

1
H
1

.
Coecients for method M
3
a
M
3
0,k
=

b
00,k
G
k
H
k
+ b
01,k
H
k
G

k
+ b
10,k
G
k
H

k

1
(x
k+1
x
k
)
a
+
10,k
J
k
G
k
H
k
+ b
11,k
G

k
H

k
+
1
2
b
20,k
G
k
H

k
+
1
2
b
02,k
G

k
H
k

1
(x
k+1
x
k
)

a
+
11,k
J
k
G

k
H
k
+
a
+
20,k
J
k
G
k
H

1
(x
k+1
x
k
)

a
+
11,k
J
k

1
(x
k+1
x
k
)
a

11,k
J
k1

k
H
k

1
(x
k+1
x
k
)

a
+
20,k
J
k

G
k
H

k
+
1
(x
k
x
k1
)

11,k
J
k1
G

k
H
k
+
a

20,k
J
k1
G
k
H

+
1
(x
k
x
k1
)
a

10,k
J
k1
G
k
H
k
+
1
(x
k
x
k1
)

20,k
J
k1

G
k
H

,
a
M
3
1,k
=
1
(x
k+1
x
k
)

a
+
10,k
J
k
G
k
H
k
+
a
+
11,k
J
k
G

k
H
k
+
a
+
20,k
J
k
G
k
H

,
a
M
3
1,k
=
1
(x
k
x
k1
)

10,k
J
k1
G
k
H
k
+
a

11,k
J
k1
G

k
H
k
+
a

20,k
J
k1
G
k
H

,
a
M
3
1,N
=
1
(x
N
x
N1
)

10,N
G
N
H
N
+ a

11,N
G

N
H
N
+ a

20,N
G
N
H

,
a
M
3
0,N
=
1
J
N1

00,N
G
N
H
N
+ a

10,N
G
N
H

N
+ a

01,N
G

N
H
N
+
a

10,,N
(x
N
x
N1
)
G
N
H
N
+a

11,N
G

N
H

N
+
a

11,N
(x
N
x
N1
)
G

N
H
N
+
a

20,N
(x
N
x
N1
)
G
N
H

N
+
1
2
a

02,N
G

N
H
N

+
1
2
a

20,N
G
N
H

,
where
a

00,k
=
1
(1 b
0
)

(x
2+a
0
b
0
k1
x
2+a
0
b
0
k
)
(2)
+
x
k1
(1)

x
1+a
0
k1
x
1+a
0
k

,
19
a

01,k
=
1
(1 b
0
)

(x
k
x
k1
)x
2+a
0
b
0
k1
(2)
+
1
(3)

x
3+a
0
b
0
k1
x
3+a
0
b
0
k

+
(x
k1
x
k
)x
1b
0
k1
(1)

x
1+a
0
k1
x
1+a
0
k

x
2b
0
k1
(1)(2)

x
1+a
0
k1
x
1+a
0
k

+
1
(2)(3 + a
0
b
0
)

x
3+a
0
b
0
k1
x
3+a
0
b
0
k

,
a

02,k
=
1
(1 b
0
)

(x
k1
x
k
)
2
x
2+a
0
b
0
k1
(2)
+
1
(3)

2(x
k1
x
k
)x
3+a
0
b
0
k1

2
(4)

x
4+a
0
b
0
k1
x
4+a
0
b
0
k

+
(x
k1
x
k
)
2
x
1b
0
k1
(1 + a
0
)

x
1+a
0
k1
x
1+a
0
k

+
2(x
k1
x
k
)
(2)(3 + a
0
b
0
)
x
3+a
0
b
0
k1

2(2)
(2)(4)

x
4+a
0
b
0
k1
x
4+a
0
b
0
k

2(x
k1
x
k
)
(2)(1)
x
2b
0
k1

x
1+a
0
k1
x
1+a
0
k

+
2
(3)(1)

x
1+a
0
k1
x
1+a
0
k

2
(3)(4 + a
0
b
0
)

x
4+a
0
b
0
k1
x
4+a
0
b
0
k

,
a

10,k
=
1
(1 b
0
)

(x
k1
x
k
)
(1)
x
2+a
0
b
0
k1
+
(x
k1
x
k
)
(1)
x
2+a
0
b
0
k1

1
(2)
x
1b
0
k1

x
2+a
0
k1
x
2+a
0
k

+
1
(3)

x
3+a
0
b
0
k1
x
3+a
0
b
0
k

,
a

20,k
=
1
(1 b
0
)

(x
k1
x
k
)
2
(2)
x
2+a
0
b
0
k1
+
2(x
k1
x
k
)
(3)
x
3+a
0
b
0
k1
+
(x
k1
x
k
)
2
(1)
x
2+a
0
b
0
k1

2
(4)

x
4+a
0
b
0
k1
x
4+a
0
b
0
k

2(x
k1
x
k
)
(2)
x
3+a
0
b
0
k1
+
2
(3)
x
1b
0
k1

x
3+a
0
k1
x
3+a
0
k

,
a

11,k
=
1
(1 b
0
)

(x
k1
x
k
)
2
x
2+a
0
b
0
k1
(2)
+
(x
k1
x
k
)
2
(1)
x
2+a
0
b
0
k1

(x
k1
x
k
)
(2)
x
1b
0
k1

x
1b
0
k1
x
1b
0
k

+
2(x
k1
x
k
)
(3)
x
3+a
0
b
0
k1

2
(4)

x
4+a
0
b
0
k1
x
4+a
0
b
0
k

+
1
(2 b
0
)

(x
k1
x
k
)
(3 + a
0
b
0
)
x
3+a
0
b
0
k1

(x
k1
x
k
)
(1)
x
3+a
0
b
0
k1
+
1
(2)
x
2b
0
k1

x
2+a
0
k1
x
2+a
0
k

(2)
(4)

x
4+a
0
b
0
k1
x
4+a
0
b
0
k

,
b
ij,k
=

a
+
ij,k
J
k
+
a

ij,k
J
k1

, and
20
(i) =
i

j=2
(j + a
0
b
0
), (i) =
i

j=1
(j + a
0
), (i) =
i

j=2
(j b
0
).
Approximations of the coecients for uniform mesh case
For uniform mesh x
k
= kh, k = 0(1)N and h =
1
N
then for xed x
k
as h 0, we get
the following approximations for the coecients
b
00,k

hx
a
0
k
G
k
, b
01,k
(a
0
2b
0
)
h
3
x
a
0
1
k
4G
k
, b
02,k

h
3
x
a
0
k
2G
k
b
11,k

h
3
x
a
0
k
4G
k
, b
10,k
(2a
0
3b
0
)
h
3
x
a
0
1
k
12G
k
, b
20,k

h
3
x
a
0
k
6G
k
.
Furthermore, for suciently small h we get
|a

00,N
| <
h
2
2
, |a

11,N
| <
h
4
4
, |a

20,N
| <
h
4
6
, |a

01,N
| <
2h
3
3
, |a

02,N
| <
h
4
6
, |a

10,N
| <
h
3
3
.
Approximations of the coecients for nonuniform mesh case
For nonuniform mesh x
k
= (kh)
1
1b
0
, k = 0(1)N, k = 0(1)N, and h =
1
N
then for
xed x
k
as h 0, we get the following approximations for the coecients
b
00,k

hx
a
0
+b
0
k
(1 b
0
)G
k
, b
01,k
(a
0
+ 2b
0
)
h
3
x
3b
0
+a
0
1
k
4(1 b
0
)
3
G
k
, b
10,k
(2a
0
+ 3b
0
)
h
3
x
3b
0
+a
0
1
k
12(1 b
0
)
3
G
k
b
11,k

h
3
x
3b
0
+a
0
k
4(1 b
0
)
3
G
k
, b
20,k

h
3
x
3b
0
+a
0
k
6(1 b
0
)
3
G
k
, b
02,k

h
3
x
3b
0
+a
0
k
(1 b
0
)
3
2G
k
.
Furthermore, for suciently small h we get
|a

00,N
| <
h
2
2(1 b
0
)
, |a

01,N
| <
2h
3
3(1 b
0
)
3
, |a

10,N
| <
h
3
3(1 b
0
)
3
,
|a

11,N
| <
h
4
4(1 b
0
)
3
, |a

20,N
| <
h
4
6(1 b
0
)
3
, |a

02,N
| <
h
4
2(1 b
0
)
4
.
Acknowledgements
We are thankful to the referee for valuable suggestions. The work is supported by
Department of Science and Technology, New Delhi, India.
References
[1] Bobisud L. E., Existence of solutions for nonlinear singular boundary value problems,
Applicable Analysis, 35 (1990) 4357.
21
[2] Bobisud L. E., ORegan D., Royalty W. D., Singular boundary value problems, Ap-
plicable Analysis 23 (1986) 233243.
[3] Bobisud L. E., ORegan D. and Royalty W. D., Solvability of some nonlinear bound-
ary value problems, Nonlinear Analysis: Theory Methods & Applications, 12 (1988)
855869.
[4] Chan C. Y., Hon Y. C., A constructive solution for a generalized Thomas-Fermi
theory of ionized atoms, Qutr. Appl. Math., 45 (1987) 591599.
[5] Chawla M. M. and Katti C. P., Finite dierence methods and their convergence for
a class of singular two point boundary value problems, Numer. Math., 39 (1982)
341350.
[6] Dunninger D. R. and Kurtz J. C., Existence of solutions for some nonlinear singular
boundary value problems, J. Math Anal. Appl., 115 (1986) 396405.
[7] Einar Hille, Some aspects of the Thomas-Fermi equation, Journal dAnalyse Math-
matique , 23 (1) (1970) 147170.
[8] Fermi E., Un methodo statistico per la determinazione di alcune proprieta dellatomo,
Rend. Accad. Naz. del Lincei. Cl. sci. s., mat. enat., 6 (1927) 602607.
[9] Granas A., Guenther R. B. and Lee J. W., Nonlinear boundary value problems for
ordinary dierential equations, Dissertationes Mathematcae, Warsaw, 1985.
[10] Henrici P., Discrete variable methods in ordinary dierential equations, John Wiley
& Sons, N. Y. (1962).
[11] Pandey R. K. and Singh A. K., On the convergence of a nite dierence method
for weakly regular singular boundary value problems , J. Comp. Appl. Math., 205
(2007) 496478.
[12] Pandey R. K. and Verma A. K., Existence-uniqueness results for a class of singular
boundary value problems arising in physiology , Nonlinear Analysis: Real World
Appl., 9 (2008) 4052.
[13] Pandey R. K. and Verma A. K., A note in existence-uniqueness results for a class
of doubly singular boundary value problems, Nonlinear Analysis: Theory Methods &
Applications, 71 (2009) 34773487.
[14] Pandey R. K. and Verma A. K., On solvability of derivative dependent doubly sin-
gular boundary value problems, J. Appl. Math. Comp., to appear.
[15] Reddien G. W., Projection Methods and Singular Two Point Boundary Value Prob-
lems Numer Math., 21 (1973) 193205.
22
[16] Thomas L. H., The calculation of atomic elds, Proc. Camb. Phil. Soc., 23 (1927)
542548.
[17] Zhang Y. A note on solvability of singular boundary value problems, Nonnlinear
Analysis: Theory Methods & Applications, 26 (10) (1996) 16051609.
23

You might also like