ACCURATE NUMERICAL METHOD FOR SINGULAR INITIAL-VALUE PROBLEMS

Advanced Computational Intelligence: An International Journal, Vol.8, No.1/2/3, July 2021
DOI: 10.5121/acii.2021.8301 1
ACCURATE NUMERICAL METHOD
FOR SINGULAR INITIAL-VALUE PROBLEMS
Tesfaye Aga Bullo1*
, Gemechis File Duressa1
and Gashu Gadisa Kiltu2
1
Department of Mathematics, Jimma University, Jimma, Ethiopia
2
Department of Mathematics, Madda Walabu University, Robe, Ethiopia
ABSTRACT
In this paper, an accurate numerical method is presented to find the numerical solution of the singular
initial value problems. The second-order singular initial value problem under consideration is transferred
into a first-order system of initial value problems, and then it can be solved by using the fifth-order Runge
Kutta method. The stability and convergence analysis is studied. The effectiveness of the proposed methods
is confirmed by solving three model examples, and the obtained approximate solutions are compared with
the existing methods in the literature. Thus, the fifth-order Runge-Kutta method is an accurate numerical
method for solving the singular initial value problems.
KEYWORDS
Singular, initial value problem, accurate solution.
1. INTRODUCTION
The studies of ordinary differential equations have attracted the attention of many
mathematicians and physicists. A lot of problems in mathematical physics and astrophysics can
be modelled by several types of differential equations. These equations have different
characteristics depending on order, provided conditions, modelled problems, etc. In particular,
initial value differential equations were used in the theory of stellar structure, thermal behaviour
of a spherical cloud of gas, isothermal gas spheres, and thermionic currents (see [1 - 7].
Numerical solutions for the initial valued problems were presented by different researchers, for
instance, Shiralashetti et. al., [8], was used the Haar wavelet collocation method; Susmita and
Chakraverty [4], were used the Chebyshev Neural Network method [9] also used, an implicit
numerical method for the numerical solution of singular initial value problems. Further, different
applications of differential equations via numerical methods provided in [14 - 16]. Recently,
Shiralashetti et. al., [8], states that the Haar wavelet collocation method to solve the linear and
non-linear differential equations. Yet, there is a lack of accuracy and convergence. So that, it is
important to apply and use efficient numerical methods for solving the second-order singular
initial value problems.
One member of the family of Runge-Kutta methods is a fifth-order Runge Kutta method (RK5)
which means, the error per step is on the order of
6
h , while the total accumulated error has
order
5
h . Thus, in this paper, we use the fifth-order Runge Kutta method to produce a more
accurate solution.

2
2. FORMULATION OF THE METHOD
Consider the singular initial value problem of the form:
( ) ( ) ( , ) ( ), 0
y x y x f x y g x x l
x


 
     , (1)
subject to the initial conditions
(0) , (0)
y y
 

  , (2)
where , , and
    are given constant numbers with 0
  for ( )
y x is an unknown function.
Now, dividing the interval [0, 1] into N equal subinterval of mesh length h and the mesh point
is given by 0 , for 1,2,..., 1.
i
x x ih i N
    For the sake of simplicity,
denote ( )
i i
y x y
 , ( ) , ( )
i i i i
z x z g x g
  , etc.
Further, let us the substitutions ( ) ( ) and ( ) ( )
z x y x z x y x
  
  , then Eq. (1) with Eq. (2) can be re-
written at the nodal point i
x as:
( , , ), (0) ,
( , , ), (0) ,
i i i i
i i i i
y F x y z y
z G x y z z


  


  

(3)
where ( , , )
i i i i
F x y z z
 and ( , , ) ( , )
i i i i i i i
i
G x y z g z f x y
x


   .
To solve Eq. (3), we apply the single-step methods that require information about the solution at
the point i
x to calculate at 1,
i
x  [11, 12]. The general numerical solution of Eq. (3) using the
fifth Runge Kutta method is given as:
1
1
1
5
5
1
,
,
i i i i
i
i i i i
i
y w k
z
z w
y
m











  



(4)
where
4 4
4
1 1
1 1
4
( , , ),
( , , ).
i i i i ij j i ij j
j j
i i i i ij j i ij j
j j
hF x c h y a k z a
h
k m
G x c h y a k
m m
z a
 
 









  
  
 
 

3
Thanks to Christodoulou [10], was present the fifth-order Runge Kutta method to solve a first-
order initial value problem of the form 0 0
( , ), ( )
dy
f t y y t y
dt
  , which is given by the following
equation:
1 1 3 4 5 6
1
(7 32 12 32 7 )
90
n n
y y k k k k k
       , (5)
where
1 ( , ),
n n
k hf x y

1
2 ( , ),
2 2
n n
k
h
k f t y
  
1 2
3
3
( , ),
4 16
n n
k k
h
k f t y

  
3
4 ( , ),
2 2
n n
k
h
k f t y
  
5 2 3 4
3 1
( , ( 3 6 9 ))
4 16
n n
h
k f t y k k k
      ,
6 1 2 3 4 5
1
( , ( 4 6 12 8 ))
7
n n
k f t h y k k k k k
       .
Thus, to solve Eq. (3), the fifth-order Runge Kutta method can be re-written as:
1 1 3 4 5 6
1 1 3 4 5 6
1
(7 32 12 32 7 ),
90
1
(7 32 12 32 7 ).
90
i i
i i
y k k k k k
z m m
z m m m
y 

     
 




   



(6)
where: 1 ( , , ),
i i i
k F x y z

1 ( , , ),
i i i
m G x y z

1 1
2 ( , , ),
2 2 2
i i i
k m
h
k hF x y z
   
1 1
2 ( , , )
2 2 2
i i i
k m
h
m hG x y z
    ,
3 1 2 1 2
1 1
( , (3 ), (3 )),
4 16 16
i i i
h
k hF x y k k z m m
     
3 1 2 1 2
1 1
( , (3 ), (3 ))
4 16 16
i i i
h
m hG x y k k z m m
      ,
3 3
4 ( , , ),
2 2 2
i i i
k m
h
k hF x y z
    3 3
4 ( , , )
2 2 2
i i i
k m
h
m hG x y z
    ,
5 2 3 4 2 3 4
3 1 1
( , ( 3 6 9 ), ( 3 6 9 ))
4 16 16
i i i
h
k hF x y k k k z m m m
          ,
5 2 3 4 2 3 4
3 1 1
( , ( 3 6 9 ), ( 3 6 9 ))
4 16 16
i i i
h
m hG x y k k k z m m m
          ,

4
6 1 2 3 4 5 1 2 3 4 5
1 1
( , ( 4 6 12 8 ), ( 4 6 12 8 ))
7 7
i i i
k hF x h y k k k k k z m m m m m
            ,
6 1 2 3 4 5 1 2 3 4 5
1 1
( , ( 4 6 12 8 ), ( 4 6 12 8 )).
7 7
i i i
m hG x h y k k k k k z m m m m m
           
In the determination of the parameters, since the terms are up to 5
( )
O h be compared, the
truncation error is 6
( )
O h and the order of method is 5
( )
O h [11 - 13].
The method is given in Eq. (6) is applied on Eq. (3) for 0
i  , 0
( . 0)
i
i e x x
  , So that, it is
necessary to modify the given second-order singular initial value problems at 0
i  , as follow:
Modification at Singular
The fifth-order Runge Kutta method given in Eq. (6) cannot be used for the stated problem
at 0
i  , since it is not defined at 0
x x
 . Hence we have a modified form on Eq. (1) at the
singular point 0 0
x x
  as:
0 0 0 0
0
( ) lim ( ) ( , ) ( )
x
y x y x f x y g x
x



 
   .
Using L. Hospital’s Rule of limit evaluation, we obtain:
0 0 0 0
(1 ) ( ) ( , ) ( )
y x f x y g x
 

   at 0 0
x x
  ,
0 0 0 0
1
( ) ( ( ) ( , )),
1
y x g x f x y


  

for 1
   .
Thus, from the relations i i
y z
  , the above equation re-written as:
0 0 0 0
1
( ( , )), (0) .
1
z g f x y z
 

   
 (7)
Now, by applying Eq. (6) on Eq. (3) for 0
i  and on Eq. (7) for 0
i  , we can find stable and
more accurate approximate solutions for the proposed problem.
3. STABILITY ANALYSIS
The problem in Eq. (1) is reduced into Eq. (3) and let take the second equation from Eq. (3), then
we have:
( , , ), (0)
i i i i
z G x y z z 
   , (8)
where ( , , ) ( , )
i i i i i i i
i
G x y z g z f x y
x


   .

5
The nonlinear function Eq. (8) can be linearized by expanding the function G in Taylor series
about the point 0 0 0
( , , )
x y z and truncating it after the first term as:
0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
( , , ) ( ) ( , , )
( ) ( , , ) ( ) ( , , ).
G
z G x y z x x x y z
x
G G
y y x y z z z x y z
y z

    

 
  
 
(9)
By the differentiation rules of the function of several variables Eq. (9) can be written as:
0 0 0 0 0 0 0 0 0
2
0 0
0
0 0 0 0 0
0
( , ) ( )( ( , ))
( ) ( , ) ( )( ).
f
z g z f x y x x g z z x y
x x x
x
f
y y x y z z z
y x
  
 



  
       



    

Using Eq. (12) and substitution, we get:
0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0
1 1
( ( ( , ))) ( , ) ( )( ( ( ( , )))
1 2 1
1
( ( ( , ))) ( , )) ( ) ( , )
1
1
( )( ( ( , ))).
1
f
z g g f x y f x y x x g g x y
x
f f f
g x y x y y y x y
x x y
f
z z g x y
x

   
 
   

 


  
       
  
  

    
   


   
 
This can be re-written in the form of:
z z c

   , (10)
where 0 0 0
1
( ( ( , ))
1
f
g x y
x
  



  
 
,
0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0
1
( , ) ( )( ( (
1
1
( , )) ( , )) ( ) ( , ) ( ( ( , )) ,
1
c g z f x y x x g g
f f f f
x y x y y y x y g x y z
x x y x
  

    

  
      

   

    
    
Which is linear in the variable z, since dividing both sides of Eq. (10) by  , we obtain
z c
z
 

  and let ,
c
u z

  then we have:
c
z u

  . (11)
Substituting Eq. (11) into Eq. (10), we get the following:
c c
u u c

 

   
   
   
   
u u

  , (12)

6
which is called the linear test equation for the non-linear Eq. (8). The solution of the test
equation, Eq. (8) is:
0 ( )
h n
u u e
 , (13)
where 0
u is constant.
Now, by applying Eq. (5) on Eq. (12), we have:
1 ,
i
k hu


2
1
(1 ),
2
i
k hu h
 
 
2 2 3 3
3 ,
4 32
i i i
h h
k h u u u
 

  
2 2 3 3
4 (1 ),
2 8 64
i
h h h
k hu
  

   
2 2 3 3 4 4 5 5
5
3 9 21 9
,
4 32 256 1024
i i i i i
k hu h u h u h u h u
    
    
2 2 3 3 4 4 5 5 6 6
6
1 15 15 9
2 112 224 896
i i i i i i
k hu h u h u h u h u h u
     
     
By substituting the values of 1 3 6
and
k k k
 into
1 1 3 4 5 6
1
(7 32 12 32 7 )
90
i i
u k k k k k
u        ,
We obtain:
1 ( )
i i
u E h u

  , (14)
where: 2 2 3 3 4 4 5 5 6 6
1 1 1 1
.
6 24 120
1
( )
0
1
28
2 1
h h h h
h h
E h  
    

     
The errors in numerical computation don’t grow, if the propagation error tends to zero or if at
least bounded, [13]. Now, from Eq. (13), it is easily observed, the exact value of ( )
i
u x increases
for the constant 0
  and decreases for 0
  with the factor of
h
e
. While from Eq. (14) the
approximate value of i
u increases or decreases with the factor of E ( h). If 0
h
  ,
then 1
h
e
 ; So the fifth-order Runge Kutta method is relatively stable. If 0
h
  , (i.e., 0
  )
then the fifth-order Runge Kutta method is stable in the interval of 5.604 0.
h

  
4. NUMERICAL EXAMPLES AND RESULTS
To demonstrate the applicability of the methods, three model singular initial value problems have
been considered. These examples have been chosen because they have been widely discussed in
the literature and their solutions are available for comparison.

7
Example 1: Consider the singular initial value problem:
2
2
2(2 3) 0, 0 1
y y x y x
x
 
      , (15)
subject to the initial conditions: (0) 1, (0) 0
y y
  and its exact solution is given as
2
( ) x
y x e

Table 1: The comparison of pointwise relative errors at 0.1
h  for Example 1.
Exact Solution Pointwise relative errors
x Method [4]
Our method
(RK5)
0.1 1.010050167084168 0.0007 2.2109e-05
0.2 1.040810774192388 0.0013 6.6993e-05
0.3 1.094174283705210 0.0003 7.9959e-05
0.4 1.173510870991810 0.0134 8.5615e-05
0.5 1.284025416687741 0.0026 8.8478e-05
0.6 1.433329414560340 0.0021 9.0031e-05
0.7 1.632316219955379 0.0085 9.0904e-05
0.8 1.896480879304951 0.0041 9.1408e-05
0.9 2.247907986676471 0.0087 9.1714e-05
1.0 2.718281828459046 0 9.1923e-05
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8
x
Exact
and
Numerical
Solution
Numerical Solution
Exact Solution
Figure 1: The Relationship between the Exact and Numerical solutions for Example 1.
Example 2: Consider the non-linear singular initial value problem:
2
2
4(2 ) 0, 0 1
y
y
y y e e x
x
 
      , (16)
with the initial conditions (0) 0, (0) 0
y y
  . Its exact solution is 2
( ) 2log(1 )
y x x
   .

8
Table 2: The comparison of Maximum Absolute errors for Example 2.
N Method [8] Our method (RK5)
8 4.5110e-04 3.5374e-04
16 1.0863e-04 2.5646e-05
32 3.0001e-05 1.7753e-06
64 7.1538e-06 1.1832e-07
128 1.2981e-06 7.7075e-09
256 4.3423e-07 4.9457e-10
Table 3: Pointwise Absolute errors for Example 2 at different values of mesh length h.
x 0.1
h  0.02
h  0.01
h  0.001
h 
0.1 4.4016e-05 2.9774e-07 2.0257e-08 2.1701e-12
0.2 1.3151e-04 3.1078e-07 2.0126e-08 2.0735e-12
0.3 1.5161e-04 2.9602e-07 1.8848e-08 1.9144e-12
0.4 1.5196e-04 2.6984e-07 1.7012e-08 1.7132e-12
0.5 1.4290e-04 2.3772e-07 1.4876e-08 1.4879e-12
0.6 1.2890e-04 2.0307e-07 1.2625e-08 1.2544e-12
0.7 1.1251e-04 1.6827e-07 1.0391e-08 1.0258e-12
0.8 9.5351e-05 1.3488e-07 8.2663e-09 8.0946e-13
0.9 7.8442e-05 1.0388e-07 6.3049e-09 6.1151e-13
1.0 6.2402e-05 7.5743e-08 4.5336e-09 4.3521e-13
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.5
1
1.5
2
2.5
x 10
-3
x
Pointwise
Relative
errors
h = 0.1
h = 0.05
h = 0.001
Figure 2: Relative pointwise errors decrease as the mesh size decreases for Example 2.
Example 3: Consider the following Lane–Emden equation:
5
2
0, 0 1
y y y x
x
 
     , (17)
With the initial conditions (0) 1, (0) 0
y y
  ; whose exact solution is:
2
1
( )
(1 )
3
y x
x


.

9
Table 4: The comparison of pointwise relative errors at 0.1
h  for Example 3.
x Exact solution
Pointwise Relative errors
Method [4] Our method (RK5)
0.1 9.9834e-01 0.0002 1.8472e-06
0.2 9.9340e-01 0.0001 5.6146e-06
0.3 9.8533e-01 0.0046 6.6915e-06
0.4 9.7435e-01 0.0032 7.0461e-06
0.5 9.6077e-01 0.0076 7.0582e-06
0.6 9.4491e-01 0.0038 6.8632e-06
0.7 9.2715e-01 0.0032 6.5281e-06
0.8 9.0784e-01 0.0002 6.0945e-06
0.9 8.8736e-01 0.0044 5.5917e-06
1.0 8.6603e-01 0.0009 5.0421e-06
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
1
2
3
4
5
6
7
x 10
-6
x
Pointwise
Absolute
errors
h = 0.1
h = 0.05
h = 0.001
Figure 3: Absolute errors decrease as the number of mesh size h decreases for Example 3.
5. DISCUSSION AND CONCLUSION
In this paper, the fifth-order Runge Kutta method has been presented for solving the second-order
singular initial value problems. First, after the given interval or domain is discretized, the
problem is reduced into a system of two first-order initial value problems, and then the obtained
system is solved by using the fifth-order Runge Kutta method. Its stability analysis has been
established and the method is stable in the interval of 5.604 0
h

   . To validate the
applicability of the proposed method, model examples have been considered and solved for
different values of mesh sizes. As it can be observed from the numerical results presented in
Tables 1 - 4 and graphs (Figure 1), the present method approximates the exact solution very well
and compared with the numerical results presented by Susmita and Chakraverty [4] and
Shiralashetti et al., [8]. Furthermore, it observed from Tables 1, 2, and 4, the approximate
solution presented in this paper is more accurate than the approximate solutions presented by the
mentioned kinds of literature. Generally, all the pointwise absolute errors, pointwise relative
errors, and maximum absolute errors decrease rapidly as the number of mesh size h decreases,
and the graphs (Figs. 2 and 3) reflect that the pointwise relative and absolute errors decrease

10
rapidly as the number of mesh size decreases. Thus, the fifth-order Runge Kutta method is a
stable, more accurate, and convergent method to find the approximate solution of the second-
order singular initial value problems.
REFERENCES
[1] Batiha, B. (2009). Numerical solution of a class of singular second-order IVPs by variational iteration
method. International Journal of Mathematical Analysis, 3(40), 1953-1968.
[2] Hasan, M. K., Ahamed, M. S., Huq, M. A., Alam, M. S., & Hossain, M. B. (2014). An implicit
method for the numerical solution of second-order singular initial value problems. The Open
Mathematics Journal, 7(1).
[3] Turkyilmazoglu, M. (2013). Effective computation of exact and analytic approximate solutions to
singular nonlinear equations of Lane–Emden–Fowler type. Applied Mathematical Modelling, 37(14-
15), 7539-7548.
[4] Mall, S., & Chakraverty, S. (2015). Numerical solution of nonlinear singular initial value problems of
Emden–Fowler type using Chebyshev Neural Network method. Neurocomputing, 149, 975-982.
[5] Kiltu, G. G., & Duressa, G. F. (2019). Accurate numerical method for Liénard nonlinear differential
equations. Journal of Taibah University for Science, 13(1), 740-745.
[6] Kiltu, G. G., Duressa, G. F., & Bullo, T. A. (2020). Numerical treatment of singularly perturbed delay
reaction-diffusion equations. International Journal of Engineering, Science and Technology, 12(1),
15-24.
[7] Siraj, M. K., Duressa, G. F., & Bullo, T. A. (2019). Fourth-order stable central difference with
Richardson extrapolation method for second-order self-adjoint singularly perturbed boundary value
problems. Journal of the Egyptian Mathematical Society, 27(1), 1-14.
[8] Shiralashetti, S. C., Deshi, A. B., & Desai, P. M. (2016). Haar wavelet collocation method for the
numerical solution of singular initial value problems. Ain Shams Engineering Journal, 7(2), 663-670.
[9] Mall, S., & Chakraverty, S. (2014). Chebyshev neural network-based model for solving Lane–Emden
type equations. Applied Mathematics and Computation, 247, 100-114.
[10] Christodoulou, N. S. (2009). An algorithm using Runge-Kutta methods of orders 4 and 5 for systems
of ODEs. International journal of numerical methods and applications, 2(1), 47-57.
[11] File, G., & Aga, T. (2016). Numerical solution of quadratic Riccati differential equations. Egyptian
Journal of basic and applied sciences, 3(4), 392-397.
[12] Grewal B’S, (202) Numerical method in engineering and science with programs in FORTRAN 77, C
and c++, Khanna publisher sixth edition.
[13] Jain, M. K., Iyengar, S. R. K., & Jain, R. K. (2007). Numerical methods for scientific and
Engineering Computations; equations.
[14] Bouhous, A. (2015). Artificial Neural Network In The Design Of Rectangular Microstrip
Antenna. Advanced Computational Intelligence: An International Journal (ACII), 2(2).
[15] Jensi, R., & Jiji, G. W. (2015). Hybrid data clustering approach using k-means and flower pollination
algorithm. arXiv preprint arXiv:1505.03236.
[16] Kusi, G. R., Bullo, T. A., & Duressa, G. F. (2021). Quartic Non-polynomial Spline Method for
Singularly Perturbed Differential-difference Equation with Two Parameters. Journal of Engineering
Advancements, 9-15.

11
AUTHORS
TESFAYA AGA BULLO received B.Ec. from Addis Ababa University, and M.Sc. from
Bahr Dar University. Currently, he is a Ph.D. scholar in Jimma University. His research
interests are computational mathematics, numerical solution of singularly perturbed
boundary value problems. He has published more than 20 research articles in reputable
journals.
GEMECHIS FILE DURESSA received his M.Sc. degree from Addis Ababa University,
Ethiopia and Ph.D. degree from National Institute of Technology, Warangal, India. He is
currently working as an Associate professor of Mathematics at Jimma University. His
research interests include Numerical Methods for Singularly Perturbed Differential
Equations (both ODE and PDE). As far as this, he has published more than 70 research
articles in reputable journals.
GASHU GADISA KILTU received B.Sc and M.Sc. from Jimma University, and
currently he is an Assistant Professor in Mathematics at Madda Walabu University. His
research interests are computational mathematics, numerical solution of singularly
perturbed problems. He has published more than 15 research articles in reputable
journals.

ACCURATE NUMERICAL METHOD FOR SINGULAR INITIAL-VALUE PROBLEMS

More Related Content

What's hot (18)

Similar to ACCURATE NUMERICAL METHOD FOR SINGULAR INITIAL-VALUE PROBLEMS (20)

Recently uploaded (20)

ACCURATE NUMERICAL METHOD FOR SINGULAR INITIAL-VALUE PROBLEMS