Computers Math. Applic. Vol. 29, No. 6, pp.

75-80, 1995
Pergamon Copyright1995 Elsevier Science Ltd
Printed in Great Britain. All rights reserved
0898-1221/95 $9.50 + 0.00
0 8 9 8 - 1221 ( 95 ) 0 0 0 0 9 - 7

Modification of
Adomian's D e c o m p o s i t i o n M e t h o d
to Solve Equations Containing Radicals
S. N. VENKATARANGAN AND K. RAJALAKSHMI
Department of Mathematics
Indian Institute of Technology, Madras 600 036, India

(Received January 1994; accepted February 1994)

Abstract--Adomian's decomposition method has been used in obtaining appproximate solutions

to a wide class of nonlinear deterministic/stochastic operator equations. Adomian, Rach and Sarayan
have investigated the use of the decomposition method for equations involving radicals. But in such
equations, when the linear operator is of higher degree, the direct use of Adomian polynomials for the
radicals produces computational difficultieswhile inverting the operator, in particular, if the initial
approximation is not a constant. In this paper, we propose a method in which the nonlinear term in
the equation involving radicals is firstexpanded formally in a power series. The power series is then
approximated by Padd approximant, thereby replacing the radical term by a ratio of polynomials.
The decomposition method of Adomian is now applied to the modified equation without any difficulty
as the radical type nonlinearity is now converted into a polynomial type nonlinearity.

Keywords--Decomposition method, Nonlinear equations, Radicals, Pad6 approximant.

1. I N T R O D U C T I O N

Mathematical modelling of most of the physical systems leads to nonlinear differential equations.
Adomian's decomposition method [1] has been used in obtaining approximate solutions to a wide
class of nonlinear deterministic/stochastic operator equations, involving systems of algebraic [2]
differential [3], integro-differential [4], and partial differential equations [5]. Adomian's method
consists of:
(i)splitting the given equation into linear and nonlinear parts;
(ii)inverting the linear operator on both sides;
(iii)decomposing the unknown function into a series whose components are to be determined;
(iv) identifying the initial and/or boundary conditions and the terms involving the independent
variable alone as initial approximant;
(v) decomposing the nonlinear term in terms of Adomian polynomials; and
(vi) finding the successive terms of the series solution by successive iteration using Adomian
polynomials.
A modification of the method named as shooting type Adomian's method was given in [6],
to solve boundary value problems with homogeneous boundary conditions and no source terms
where the initial approximant turned out to be zero. When the ordinary differential equation is
singular, inverting the highest order derivative leads to an indeterminate form. A proper choice
of the linear operator, with the inverse defined, is metioned in [7]. Also [7] gives a new technique
of handling the boundary condition at infinity.

Typeset by .AA,~-TEX

75
 76 S. N. V E N K A T A R A N G A N AND K. RAJALAKSHMI

Adomian, Rach, and Sarafyan [8] have investigated the use of decomposition method for equa-
tions involving radicals. But in such equations, when the linear operator is of higher degree,
the direct use of Adomian polynomials for the radicals produces computational difficulties while
inverting the operator, in particular, if the initial approximation is not a constant.
In this paper, we propose a method in which the nonlinear term involving radicals is approx-
imated using Padd approximants, and then Adomian's decomposition method is used to solve
the equation. We consider nonlinear ordinary differential equations containing radicals in the
unknown and not in its derivatives. The method is explained in Section 3, and is applied in
Section 4 to the Thomas-Fermi equation, which has no analytic solution. The result is compared
with those obtained by Anderson and Arthurs [9] using a dual variational principle, and with the
numerical solution provided by Kobayashi [10]. The accuracy of the result proves the success of
this method which overcomes the above said computational difficulty.

2. D E F I N I T I O N OF A PAD]~ A P P R O X I M A N T
A Padd approximant [11] is the ratio of two polynomials constructed from the coefficients of
the Taylor series expansion of a function. The L / M Padd approximant to a formal power series
A ( x ) = ~j=o aj x j is given by [L/M] = PL(Z)/QM(X), where PL(X) is a polynomial of degree
at most L and QM(X) is a polynomial of degree at most M. Without loss of generality, we can
assume QM(O) to be 1. Further, PL and QM have no common factors. This means that the
formal power series A ( x ) equals the [L/M] approximant through L + M + 1 terms.

3. D E S C R I P T I O N OF T H E M E T H O D
Consider the equation of the form

Lu Nu + Ru= f, in [a,b],
uk(a)=dk, k=O, 1,...,r-1, u l(b) = d l , l=r,...,n,

where L is the linear n t h order ordinary differential operator d n / d x n, which is easily invertible,
N is the nonlinear operator involving radicals, that is

N u = (ao + al u + . . . + am um) p/q ,

R denotes the remaining linear and nonlinear operators, and f is the term containing the inde-
pendent variable. By Adomian's decomposition method,
oo

u = ~'~ u,~ and uo = h + L - l f,

where h is the solution of L u = 0 satisfying the r boundary conditions at the initial point a,
with n - r arbitrary constants. These constants are determined by applying the remaining n - r
boundary conditions to the series solution. The nonlinearity N u is written as a series of Adomian
polynomials. An easy computation of the Adomian polynomials for any type of nonlinearity is
given in [3]:
(2O

Nu = ZAn,
n=O

where An's are the Adomian polynomials. We have

ul = - L - 1 A o - L - 1 R u o ,
u2 = - L -1 A1 - L - 1 R u l ,

un+l = - L -1 An - L - 1 R u n .
 Equations Containing Radicals 77

Here
Ul : - L -1 (ao -[- al Uo - [ - ' " -)- am u ~ ) p/q - L -1 ( R uo),

where L -1 is an n-fold integration. When u0 is a function of the independent variable, say

u0 -- p + q t, then
~1. *.
ul . . . . [a0 + a l ( p + q x n ) + + a m ( p + qx,~)m] p/q

+ R (p + qx,~)} dx,~ d x n - 1 ... d x l ,

whose computation itself becomes difficult as is seen from above.

In the present method, the radical involved in the equation is expanded formally in a power
series. But its domain of validity is restricted due to the use of the binomial expansion. Then, we
approximate the power series by a rational approximation, namely a Pad~ approximation. The
Pad~ approximant of the power series is valid in the entire domain, except perhaps at the poles,
if they exist. Hence, the approximant is valid in the domain of the equation. Now the radical
terms in the equation have been substituted by a ratio of polynomials in the unknown function
of desired degree. The choice of the approximant depends on the initial condition of the problem.
T h a t is, at the initial value of the function, the approximant should match the radical almost
exactly or to any desired degree of accuracy. Now, on cross multiplying, the equation has only
di u
nonlinearities of the form uk -~r, i -- 0, 1 , . . . ,n, and k = 0, 1 , . . . ,m, and there is no difficulty in
obtaining the successive terms of the series solution using Adomian polynomials. The method is
illustrated with an example in the next section.

4. EXAMPLE
Consider the Thomas-Fermi equation occuring in physics which has no analytic solution:

d2y y3/2
dx 2 - z l / 2 , y(O) = 1, y(co) = O. (1)

One measure of convergence of the technique is given by calculating the initial value y~(0) of the
Thomas-Fermi potential. This quantity, which is difficult to compute, plays an important role
in determining many of the physical properties of the Thomas-Fermi atom. For example, the
energy for a neutral atom of atomic number Z is

6(4~ 2/3
E= Z 7/3y'(o).

To obtain a power series representation for the radical y3/2, we use the simple transformation
y -- u 1. The equation then becomes

d2u _ (u + 1) 3/2 u(0) 0, u(co) -1. (2)

dx 2 xl/2 '

Now we expand (u + 1) 3/2 in power series:

( u + l ) 3 / 2 = 1 + ~ u3+ ~ 3 u2 + . . . . (3)

The [1/1] Pad~ approximant for the series is (4 + 5u)/(4 - u). Hence, equation (2) becomes

d2u 4 + 5u
dx 2 -- x 1/2 (4 - u)' u(0) = 0, u(co) = - 1 . (4)
 78 S.N. VENKATARANGAN AND K. RAJALAKSHMI

Equation (4) can be rewritten in the form

d2u 1 5u u d2u
u(0) = 0, u(oo) = - 1 . (5)
dx 2 = ~ + ~ + -~ dx---'-
if,
Now, by applying Adomian's method, we have
d2u - u d2u -5u 1
L u = ~x2, N u = --4- dx---~, R u -- 4xl/2, f = xl/2, L - 1u = ~0~~0xlu dx2 dx 1.
Now,
u=L -1{ 5u ud2u~ 4x3/2 (6)
4--~-~ + ~ dx2 j + A x + ~
Let oo

U ---- EUn, with uo = A x + 4] X3/2.

n=O

The first few Adomian polynomials for u ~d2~ are:

d 2uo
Ao = uo dx 2 ,
d2uo d2ul
At=Ul~+Uo dx 2,
d2uo d2ul d2u2
A2 = u2 ~ + Ul ~ + Uo dx 2.
Hence,

Ul = A x s/2 + ~1 x 3 ,

3 A2 x7/2 + 2 A x 4 + ~--~x 9/2.

u2 = -~

By taking three terms of the series, we have

2 2
EYi = I + Eu~,
i={~ i=O
2
Ey~=l+Ax+ 43 3x/ 2 + g2 A x 5/2 + ~lX3 + ~60A2 xT/2 + 215A T 4 + ~ - - ~ x 9 / 2 . (7)
i=O
The constant A is to be determined by applying the condition at infinity. By putting x 1/2 = t,
equation (7) can be written as a power series in t. Then,

EY'=l+At2+ 3 t 3 + - 5 A t S + ' a t 6 + ,o + ta A t 8 + ~ - ~ t . (8)

i=0
The condition y(c~) = 0 cannot be applied to (8) directly. So by the method given in [7], we
apply the infinity condition (i) to the diagonal Pad6 approximant of the series (8), and (ii) to the
diagonal Padd approximant of the Laplace transform of the series (8), and the resulting algebraic
equation is solved to obtain the constant A. In the case (ii), the final value theorem for the
Laplace transform is used. The approximants to which the condition y(oo) = 0 is applied and
the values of A obtained in both the cases are listed below.

Case (i). Case (ii).

[L/M] A [L/M] A
3/3 --1.64 3/3 -1.48
4/4 --1.5505259 4/4 -1.6
5/5 -1.5645838
 Equations Containing Radicals 79

From (7), we have y'(0) = A. From the numerical solution provided by Kobayashi [10], the value
of y'(0) is given to be -1.5880710. Anderson, Arthurs and Robinson [9] have obtained upper and
lower bounds for y~(0) using complementary variational principles, which are -1.589 < y'(0) <
-1.5631.
The values obtained above indicate the convergence of A. Accuracy improves by taking higher
order Padd approximants. In Figures 1 and 2, we have compared the [1/4] approximant of the
series (7), for A -- -1.6 and A = -1.5645838, with the solution obtained by Anderson, Arthurs
and Robinson [9].

I
O8 - -

....
b o u n d s of the vorlotionol

I/4 opproximont
solution

0.6

0.4

0.2

5 10 15
X A =-1.6
F i g u r e 1.

1,0!

- - bounds of the variational solution

0.8 .... 1/4 approximant

0.6

0.4

0.2

10 15
A =-1.5645838

Figure 2.
80 S.N. VENKATARANGAN AND K. RAJALAKSHMI

5. C O N C L U S I O N S
The method described above provides a computationally easier approach to solve equations
containing radicals. The difficulty in using Adomian polynomials directly to the radical terms is
overcome here.

REFERENCES
1. G. Adomian, Stochastic Systems, Academic Press, London, (1983).
2. G. Adomian, On the solution of algebraic equation by the decomposition method, Jour. Math. Anal. Appl.
105 (1), 141-166 (1985).
3. G. Adomian, Nonlinear Stochastic Operator Equations, Academic Press, London, (1986).
4. G. Adomian and It. Rach, On composite nonlinearities and the decomposition method, Jour. Math. Anal.
Appl. 113 (2), 504-509 (1986).
5. G. Adomian, A new approach to nonlinear partial differential equations, Jour. Math. Anal. Appl. 102 (2),
420-434 (1984).
6. S.N. Venkatarangan and T.It. Sivakumar A modification of Adomian's decomposition method for boundary
value problems with homogeneous boundary conditions, Jour. Math. Phys. Sci. 24, 19-25 (1990).
7. T.It. Sivakumar, Some variants of Adomian's decomposition method and their applications, Ph.D. Thesis,
Indian Institute of Technology, Madras, ((1991)).
8. G. Adomian, R. Rach and D. Sarafyan, On the solution of equations containing radicals by the decomposition
method, Jour. Math. Anal. Appl. 111 (2), 423-426 (1985).
9. N. Anderson, A.M. Arthurs and P.D. Robinson, Complementary variational principles for a generalised
diffusion equation, Proc. Royal Soc. A303, 497-502 (1968).
10. S. Kobayashi, T. Matsukuma, S. Nagai and K. Umeda, Some coefficients of the TFD function, J. Phys. Soc.
Japan 10, 759 (1955).
11. G.A. Baker, Essentials of Padd Approximants, Academic Press, London, (1975).