Applied Mathematics and Computation 177 (2006) 737–744

www.elsevier.com/locate/amc

The modified decomposition method and Padé approximants

for a boundary layer equation in unbounded domain
Abdul-Majid Wazwaz
Department of Mathematics and Computer Science, Saint Xavier University, Chicago, IL 60655, United States

Abstract

The modified decomposition method is applied for analytic treatment of nonlinear differential equations that appear on
boundary layers in fluid mechanics. The modified method accelerates the rapid convergence of the series solution, dramat-
ically reduces the size of work. The obtained series solution is combined with the diagonal Padé approximants to handle
the boundary condition at infinity.
Ó 2005 Elsevier Inc. All rights reserved.

Keywords: The modified decomposition method; Nonlinear differential equations; Boundary layers; Blasius equation; Padé approximants

1. Introduction

Nonlinear phenomena, that appear in many areas of scientific fields such as solid state physics, plasma
physics, fluid mechanics, population models and chemical kinetics, can be modelled by nonlinear differential
equations. In particular, the nonlinear differential equations that characterize boundary layers in unbounded
domain, that will be examined here, is of much interest. A broad class of analytical solutions methods and
numerical solutions methods were used to handle these problems. The Adomian decomposition method
has been proved to be effective and reliable for handling differential equations, linear or nonlinear.
In this work we aim to apply the modified decomposition method to handle the nonlinear differential
equation
f 000 þ ðn  1Þff 00  2nðf 0 Þ2 ¼ 0;
ð1Þ
f ð0Þ ¼ 0; f 0 ð0Þ ¼ 1; f 0 ð1Þ ¼ 0; n > 0;
that appear in boundary layers in fluid mechanics [1,2], where f00 (0) < 0. An analytic treatment will be ap-
proached to find the numerical values of f00 (0) for several values of n. The goal will be achieved by combining
the series obtained by the modified decomposition method with the diagonal Padé approximants.

E-mail address: [email protected]

0096-3003/$ - see front matter Ó 2005 Elsevier Inc. All rights reserved.
doi:10.1016/j.amc.2005.09.102
738 A.-M. Wazwaz / Applied Mathematics and Computation 177 (2006) 737–744

It is interesting to point out that Kuiken [1,2] investigated this problem for three cases of n, namely for
0 < n < 1, n = 1 and for n > 1. It was shown in [1] that the algebraic behavior at the outer edge of a boundary
layer in fluid mechanics can sometimes be allowed in regions of non-vanishing extent.
It is well known that the Blasius equation is the mother of all boundary-layer equations in fluid mechanics.
Many different, but related, equations have been derived for a multitude of fluid-mechanical situations, for
instance, the Falkner-Skan equation [3]. Eq. (1) used by Kuiken [1] does not include the Blasius equation
for a particular choice of n. Indeed, this equation is for backward boundary layers, that is, boundary layers
originating at 1.
Although the convergence provided by Adomian decomposition method is rapid [4], the modified decom-
position method developed by Wazwaz in [5] accelerates this rapid convergence of the series solution. The
method provides the series solution by using few iterations; therefore it facilitates the computational work
and minimizes its volume.
Although the modified form introduces a slight change in the formulation of Adomian recursive relation,
but it provides a qualitative improvement over standard Adomian method. While this slight variation is rather
simple, it does demonstrate the reliability, efficiency, and power of the method. It is well known that polyno-
mials are used to approximate truncated power series. Moreover, polynomials can never blow up in a finite
plane and this makes the singularities not apparent. To overcome these difficulties, the obtained series is best
manipulated by Padé approximants for numerical approximations. Using the power series, isolated from other
concepts, is not always useful because the radius of convergence of the series may not contain the two
boundaries.
In this paper, only a brief discussion of the modified decomposition analysis will be emphasized, because
the complete details of the method are found in [5–16] and in many related works.

2. The modified decomposition method

In this section, we will briefly discuss the use of the modified decomposition method for nonlinear differ-
ential equations where details can be found in [5–16].
Without loss of generality, we consider the differential equation

Lu þ Ru þ Nu ¼ gðxÞ ð2Þ

with prescribed conditions, where u(x) is the unknown function, L is the highest order derivative which is as-
sumed to be easily invertible, R is a linear differential operator of less order than L, Nu represents the nonlin-
ear terms, and g is the source term. Applying the inverse operator L1 to both sides of (2), and using the given
conditions we obtain

uðxÞ ¼ f ðxÞ  L1 ðRuÞ  L1 ðNuÞ; ð3Þ

where the function f(x) represents the terms arising from integrating the source term g(x) and from using the
given initial or boundary conditions, all are assumed to be prescribed.
The standard Adomian method defines the solution u(x) by the series
X
1
uðxÞ ¼ un ðxÞ; ð4Þ
n¼0

where the components u0, u1, u2, . . . are usually determined recursively by using the relation

u0 ðxÞ ¼ f ðxÞ;
ð5Þ
ukþ1 ðxÞ ¼ L1 ðRuk Þ  L1 ðNuk Þ; k P 0.

The decomposition method usually identifies the zeroth component u0(x) by the function h(x) defined above.
The components u0, u1, u2, . . . are determined recursively, therefore the solution u(x) in a series form defined
by (4) is readily derived. It is obvious that Adomian method changes the differential equations to obtaining an
 A.-M. Wazwaz / Applied Mathematics and Computation 177 (2006) 737–744 739

easily computable components. The closed form for the solution u(x) if exists can be immediately obtained
because of the rapid convergence presented by the method.
The modified decomposition method [5] suggests that the function f(x) defined above in (3) be decomposed
into two parts, namely f0(x) and f1(x) such that
f ðxÞ ¼ f0 ðxÞ þ f1 ðxÞ. ð6Þ
The proper choice of the parts f0 and f1 is essential and depends mainly on trial basis. In view of this decom-
position of h(x), a slight variation only on the components u0(x) and u1(x) should be introduced. The proposed
variation is that only the part f0(x) be assigned to the zeroth component u0, whereas the remaining part f1(x) be
combined with the other terms given in (5) to define u1(x). In view of this assumption, we formulate the fol-
lowing recursive relation for the modified decomposition method:
8
< u0 ðxÞ ¼ f0 ðxÞ;
>
u1 ðxÞ ¼ f1 ðxÞ  L1 ðRu0 Þ  L1 ðNu0 ðxÞÞ; ð7Þ
>
: 1 1
ukþ2 ðxÞ ¼ L ðRukþ1 ðxÞÞ  L ðNukþ1 ðxÞÞ; k P 0.
Comparing the recursive relation (5) of the standard Adomian method with the recursive relation (7) of the
modified decomposition method leads to concluding that in (5) the zeroth component was defined by the func-
tion f(x) = f1(x) + f2(x), whereas in (7), the zeroth component u0(x) is defined only by f0(x) a part of f(x). The
remaining part f1(x) of f(x) is added to the definition of the component u1(x) in (5).
In our earlier works [5–16], we have shown that only few iterations are sufficient to determine the exact
solution or an approximation of high accuracy for most of the examined cases. The choice of f0(x) for
u0(x) to contain minimal number of terms has a strong influence on facilitating the recurrence relation,
and, as a consequence, accelerates the rapid convergence of the solution. As stated before, the success of
the modified method depends mainly on the proper choice of the parts f0(x) and f1(x). We have been unable
to establish any criterion to judge what forms of f0(x) and f1(x) can be used to yield the acceleration
demanded. It appears that trials are the only criteria that can be applied so far. Numerical applications
used in [5–16] showed that the modified algorithm compares favorably with standard decomposition
method.

2.1. Padé approximants

It is well known that polynomials are used to approximate truncated power series. It was examined by Boyd
[17] and Wazwaz [6–8] that polynomials tend to exhibit oscillations that may give an approximation error
bounds. Moreover, polynomials can never blow up in a finite plane and this makes the singularities not appar-
ent. To overcome these difficulties, the obtained series is best manipulated by Padé approximants for numer-
ical approximations. Using the power series, isolated from other concepts, is not always useful because the
radius of convergence of the series may not contain the two boundaries.
It is now well known that Padé approximants [18] have the advantage of manipulating the polynomial
approximation into a rational functions of polynomials. By this manipulation we gain more information
about the mathematical behavior of the solution. In addition, power series are not useful for large values
of x, say x = 1. Boyd [17] and others have formally shown that power series in isolation are not useful to
handle boundary value problems. This can be attributed to the possibility that the radius of convergence
may not be sufficiently large to contain the boundaries of the domain. It is therefore essential to combine
the series solution, obtained by the decomposition method or any series solution method, with the Padé
approximants to provide an effective tool to handle boundary value problems on an infinite or semi-infinite
domains. Recall that the Padé approximants can be easily evaluated by using built-in function in manipulation
languages such as Maple or Mathematica.
For better numerical approximation, our approach stems mainly from the combination of the modified
decomposition method and the diagonal approximants [n/n].
The essential behavior of the solution f(x) of Eq. (1) will be addressed by using several diagonal Padé
approximants of different degrees.
740 A.-M. Wazwaz / Applied Mathematics and Computation 177 (2006) 737–744

3. The boundary layer problem

The nonlinear third order boundary layer problem

f 000 þ ðn  1Þff 00  2nðf 0 Þ2 ¼ 0; n > 0;

0 0
ð8Þ
f ð0Þ ¼ 0; f ð0Þ ¼ 1; f ð1Þ ¼ 0; n > 0;

that appear in boundary layers in fluid mechanics [1,2], where f00 (0) = a < 0, will be examined in this work.
To apply the modified decomposition method, we first rewrite (8) in an operator form

Lf ¼ ðn  1Þff 00 þ 2nðf 0 Þ2 ; ð9Þ

where L is a third order differential operator, and hence L1 is a three-fold integration operator defined by
Z t Z t Z t
L1 ðÞ ¼ ðÞdt dt dt. ð10Þ
0 0 0

Operating with L1 on both sides of (9) and using the initial conditions f(0) = 0, f 0 (0) = 1, f00 (0) = a, we obtain

1 2
f ðxÞ ¼ x þ ax2  L1 ððn  1Þff 00  2nðf 0 Þ Þ. ð11Þ
2

The standard Adomian decomposition method defines the solution f(x) by the series

X
1
f ðxÞ ¼ un ðxÞ; ð12Þ
n¼0

where the components un(x), n P 0, will be determined recursively. Substituting (12) into both sides of (11)
gives
!
X
1
1 2 X1 X1
1
un ðxÞ ¼ x þ ax  L ðn  1Þ An  2n Bn ; ð13Þ
n¼0
2 n¼0 n¼0

where An(x) and Bn are the Adomian polynomials that represent the nonlinear terms ff00 and (f 0 )2 respectively.
Adomian polynomials can be generated for various classes of nonlinearity according to specific algorithms [4–
16].
The modified decomposition method [5] suggests that the function h(x) be divided into two parts
1
hðxÞ ¼ x þ ax2 ¼ h0 ðxÞ þ h1 ðxÞ; ð14Þ
2
where h0(x) = x is assigned to the component f0(x), and the other part h1 ðxÞ ¼ 12 ax2 is added to the definition of
the component f1(x). Under this assumption, we set the modified recursive relation
f0 ðxÞ ¼ x;
1
f1 ðxÞ ¼ ax2  L1 ððn  1ÞA0  2nB0 Þ; ð15Þ
2
fkþ2 ðxÞ ¼ L1 ððn  1ÞAkþ1  2nBkþ1 Þ; k P 0.
It is obvious that the slight variation is made on the components f0 and f1 only if compared with the standard
Adomian method.
For convenience, we list below, by using the algorithms introduced in [4–16], few terms of the Adomian
polynomials An(x):
 A.-M. Wazwaz / Applied Mathematics and Computation 177 (2006) 737–744 741

8
> A0 ðxÞ ¼ u0 ðxÞu00 ðxÞ;
>
>
> A ðxÞ ¼ u ðxÞu00 ðxÞ þ u ðxÞu00 ðxÞ;
>
>
> 1 0 1 1 0
>
< A2 ðxÞ ¼ u0 ðxÞu002 ðxÞ þ u1 ðxÞu001 ðxÞ þ u2 ðxÞu000 ðxÞ;
ð16Þ
> A3 ðxÞ ¼ u0 ðxÞu003 ðxÞ þ u1 ðxÞu002 ðxÞ þ u2 ðxÞu001 ðxÞ þ u3 ðxÞu000 ðxÞ;
>
>
>
>
> A4 ðxÞ ¼ u0 ðxÞu004 ðxÞ þ u1 ðxÞu003 ðxÞ þ u2 ðxÞu002 ðxÞ þ u3 ðxÞu001 ðxÞ þ u4 ðxÞu000 ðxÞ;
>
>
:
A5 ðxÞ ¼ u0 ðxÞu005 ðxÞ þ u1 ðxÞu004 ðxÞ þ u2 ðxÞu003 ðxÞ þ u3 ðxÞu003 ðxÞ þ u4 ðxÞu001 ðxÞ þ u5 ðxÞu000 ðxÞ;
and for Bn(x), we find
8 2
>
> B0 ðxÞ ¼ ðu00 Þ ðxÞ;
>
>
>
> B1 ðxÞ ¼ 2u00 ðxÞu01 ðxÞ;
>
>
>
< B ðxÞ ¼ ðu0 Þ2 ðxÞ þ 2u0 ðxÞu0 ðxÞ;
2 1 0 2
ð17Þ
>
> B ðxÞ ¼ 2u 0
ðxÞu 0
ðxÞ þ 2u 0
ðxÞu 0
3 ðxÞ;
>
>
3 1 2 0
>
> B ðxÞ ¼ ðu0 Þ ðxÞ þ 2u0 ðxÞu0 ðxÞ þ 2u0 ðxÞu0 ðxÞ;
2
>
> 4
>
:
2 1 3 0 4
B5 ðxÞ ¼ 2u00 u05 þ 2u01 ðxÞu04 ðxÞ þ 2u02 ðxÞu03 ðxÞ.
With f0 defined in (15), this can be valuable in determining the other components. It then follows:
f0 ðxÞ ¼ x;
1
f1 ðxÞ ¼ ax2  L1 ððn  1ÞA0  2nB0 Þ;
2
f2 ðxÞ ¼ L1 ððn  1ÞA1  2nB1 Þ; ð18Þ

f3 ðxÞ ¼ L1 ððn  1ÞA2  2nB2 Þ;

f4 ðxÞ ¼ L1 ððn  1ÞA3  2nB3 Þ;
and so on. This in turn gives
f0 ðxÞ ¼ x;
1 1
f1 ðxÞ ¼ ax2 þ nx3 ;
2 3
1 1
f2 ðxÞ ¼ að3n þ 1Þx4 þ nðn þ 1Þx5 ;
24 30
1 2 1 1 ð19Þ
f3 ðxÞ ¼ a ð3n þ 1Þx5 þ að19n2 þ 18n þ 3Þx6 þ nð2n2 þ 2n þ 1Þx7 ;
120 720 315
1 2 1
f4 ðxÞ ¼ a ð27n2 þ 42n þ 11Þx7 þ að167n3 þ 297n2 þ 161n þ 15Þx8
5040 40320
1
þ nð13n3 þ 38n2 þ 23n þ 6Þx9 ;
22680
where other components are obtained as well but not listed. This in turn gives
     
ax2 nx3 1 1 4 1 2 1 2 1 2 1 5 19 2 1 1
f ðxÞ ¼ x þ þ þ na þ a x þ n þ na þ a þ n x þ n aþ a þ na x6
2 3 8 24 30 40 120 30 720 240 40
 
1 1 2 11 3 2
þ na2 þ nþ n3 þ a2 þ n2 a2 þ n2 x 7
120 315 315 5040 560 315
 
11 3 33 2 3 3 2 23 1 167 3 1 3
þ a þ n aþ an þ na þ aþ n aþ a n x8
40320 4480 4480 5760 2688 40320 960
 
1 527 3 2 19 3 709 2 23 2 2 23 2 13 4 43
þ nþ na þ n þ na þ na þ n þ n þ a x9
2
3780 362880 11340 362880 8064 22680 22680 120960
þ 
where other terms up to O(x13) were derived.
742 A.-M. Wazwaz / Applied Mathematics and Computation 177 (2006) 737–744

3.1. The Padé approximants

Kuiken [1,2] studied the behavior of f(x) for three specific cases, namely, for 0 < n < 1, n = 1, and for n > 1.
To study the mathematical behavior of the f(x), it is normal to derive approximations for f00 (0) = a < 0 for the
three prescribed cases of n. It was formally shown by [17,5–9] that this goal can be achieved by forming Padé
approximants [18] which have the advantage of manipulating the polynomial approximation into a rational
function to gain more information about f(x). It is well known that Padé approximants will converge on
the entire real axis if f(x) is free of singularities on the real axis. Moreover, it is to be noted that Padé-finding
algorithms are built-in utilities in most manipulation languages such as Maple and Mathematica.
More importantly, the diagonal approximant is the most accurate approximant, therefore we will construct
only the diagonal approximants in the following discussions. Using the boundary condition f 0 (1) = 0, the
diagonal approximant [M/M] vanish if the coefficient of x with the highest power in the numerator vanishes.
Using the Maple built-in utilities to solve the resulting polynomials gives the values of the initial slope f00 (0) = a
listed in the following table:
For the case where 0 < n < 1, Graph 1 shows the behavior of the Padé approximants for n = 0.2, 0.4, 0.6,
and 0.8. Graph 1 shows the algebraic decay as pointed out by Kuiken [1,2].
It is interesting to point out that Kuiken [1,2] examined the specific case where n ¼ 13. For this value of n,
Eq. (8) can be reduced to
1
f 0 ¼ f 2  ax þ 1; ð20Þ
3
upon substituting n ¼ 13 and integrating the resulting equation twice. In [1], the general solution for this case
was derived as
 2 16 0
a Ai
f ðxÞ ¼ 3 . ð21Þ
9 Ai

where Ai is Airy function. Using this exact solution, it was found that
a ¼ 0:56144919346; ð22Þ

where our result obtained above in Table 1 is consistent with this result.
For n = 1, and using the average of all five diagonal Padé approximants, we find that
a ¼ 1:154948004. ð23Þ

5
n=0.2
4

n=0.4
3

n=0.6
2
n=0.8

0
2 4 6 8 10
x

Fig. 1. Graph of the diagonal Padé approximants of f(x) where 0 < n < 1.
 A.-M. Wazwaz / Applied Mathematics and Computation 177 (2006) 737–744 743

Table 1
Numerical values for a = f00 (0) for 0 < n < 1 by using Padé approximants
n [2/2] [3/3] [4/4] [5/5] [6/6]
0.2 0.3872983347 0.3821533832 0.3819153845 0.3819148088 0.3819121854
1
3 0.5773502692 0.5615999244 0.5614066588 0.5614481405 0.5614491934
0.4 0.6451506398 0.6391000575 0.6389732578 0.6389892681 0.6389734794
0.6 0.8407961591 0.8393603021 0.8396060478 0.8395875381 0.8396056769
0.8 1.007983207 1.007796981 1.007646828 1.007646828 1.007792100

Table 2
Numerical values for a = f00 (0) for n > 1 by using Padé approximants
n a
4 2.483954032
10 4.026385103
100 12.84334315
1000 40.65538218
5000 104.8420672

For the third case where n > 1, and using the diagonal Padé approximants [2/2]–[6/6] we obtained the results
shown in Table 2 above.
The results shown above confirms the exponential decay for f(x), for n > 1 as formally derived by Kuiken
[1].

4. Discussion

The modified decomposition method was applied to compute an approximation for the function f(x) for
several values of n, n > 0. This advantage demonstrates the reliability and the efficiency of the method. Fur-
ther, the accuracy level can be dramatically enhanced by computing further components of f(x). Moreover,
combining the series obtained with the Padé approximants provides a promising tool to handle problems
on an unbounded domain.
An important conclusion, that was thoroughly investigated by Kuiken [1,2], can be made here. It is clear by
examining Tables 1 and 2 that the function f(x) decays algebraically for the case where 0 < n < 1, and decays
exponentially for the case where n > 1 when x tends to infinity. Kuiken [1,2] derived his results by integrating
backwards, using asymptotic series for large values of x. In this work we derived the same results by combin-
ing the series, obtained by the modified decomposition method, with the diagonal Padé approximants.

References

[1] H.K. Kuiken, On boundary layers in field mechanics that decay algebraically along stretches of wall that are not vanishing small,
IMA J. Appl. Math. 27 (1981) 387–405.
[2] H.K. Kuiken, A backward free-convective boundary layer, Quart. J. Mech. Appl. Math. 34 (1981) 397–413.
[3] A. Asaithambi, A finite-difference method for the Falkner-Skan equation, Appl. Math. Comput. 92 (1998) 135–141.
[4] G. Adomian, Solving Frontier Problems of Physics: The Decomposition Method, Kluwer, 1994.
[5] A.M. Wazwaz, A reliable modification of Adomian’s decomposition method, Appl. Math. Comput. 102 (1999) 77–86.
[6] A.M. Wazwaz, Analytical approximations and Padé approximants for Volterra’s population model, Appl. Math. Comput. 100 (1999)
13–25.
[7] A.M. Wazwaz, A study on a boundary-layer equation arising in an incompressible fluid, Appl. Math. Comput. 87 (1997) 199–204.
[8] A.M. Wazwaz, The modified decomposition method and the Padé approximants for solving Thomas–Fermi equation, Appl. Math.
Comput. 105 (1999) 11–19.
[9] A.M. Wazwaz, A new method for solving differential equations of the Lane-Emden type, Appl. Math. Comput. 118 (2/3) (2001) 287–
310.
[10] A.M. Wazwaz, A new method for solving singular initial value problems in the second order ordinary differential equations, Appl.
Math. Comput. 128 (2002) 47–57.
744 A.-M. Wazwaz / Applied Mathematics and Computation 177 (2006) 737–744

[11] A.M. Wazwaz, Adomian decomposition method for a reliable treatment of the Emden-Fowler equation, Appl. Math. Comput. 161
(2005) 543–560.
[12] A.M. Wazwaz, Partial Differential Equations: Methods and Applications, Balkema, The Netherlands, 2002.
[13] A.M. Wazwaz, A First Course in Integral Equations, WSPC, New Jersey, 1997.
[14] A.M. Wazwaz, A reliable treatment for mixed Volterra–Fredholm integral equations, Appl. Math. Comput. 127 (2002) 405–414.
[15] A.M. Wazwaz, The existence of noise terms for systems of inhomogeneous differential and integral equations, Appl. Math. Comput.
146 (1) (2003) 81–92.
[16] A.M. Wazwaz, The numerical solution of special fourth-order boundary value problems by the modified decomposition method,
Internat. J. Comput. Math. 79 (3) (2002) 345–356.
[17] J. Boyd, Padé approximant algorithm for solving nonlinear ordinary differential equation boundary value problems on an unbounded
domain, Comput. Phys. 11 (3) (1997) 299–303.
[18] G.A. Baker, Essentials of Padé Approximants, Academic Press, London, 1975.