See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/220563393

Fourth-order finite difference method for solving

Burgers' equation

Article  in  Applied Mathematics and Computation · November 2005

DOI: 10.1016/j.amc.2004.12.052 · Source: DBLP

CITATIONS READS
114 1,255

3 authors, including:

Hany A. Hosham
Al-Azhar University
40 PUBLICATIONS   338 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Nonlinear Behavior of a Novel Electronic Switching Circuits Systems View project

Discontinuous Systems and Stochastic Processes View project

All content following this page was uploaded by Hany A. Hosham on 24 November 2018.

The user has requested enhancement of the downloaded file.

 Applied Mathematics and Computation 170 (2005) 781–800
www.elsevier.com/locate/amc

Fourth-order finite difference method

for solving Burgers
equation
I.A. Hassanien a, A.A. Salama a,*
, H.A. Hosham b

a
Department of Mathematics, Faculty of Science, Assiut University, Assiut 71516, Egypt
b
Department of Mathematics, Faculty of Science, Al-Azhar University, Assiut 71524, Egypt

Abstract

In this paper, we present fourth-order finite difference method for solving nonlinear
one-dimensional Burgers
equation. This method is unconditionally stable. The conver-
gence analysis of the present method is studied and an upper bound for the error is
derived. Numerical comparisons are made with most of the existing numerical methods
for solving this equation.

2005 Elsevier Inc. All rights reserved.

Keywords: Burgers
equation; Finite difference method; Stability; Convergence

1. Introduction

In this paper, we consider the one-dimensional quasi-linear parabolic partial

differential equation in the form
ou ou 1 o2 u
þu ¼ ; a 6 x 6 b; t>0 ð1Þ
ot ox Re ox2

*
Corresponding author.
E-mail address: [email protected] (A.A. Salama).

0096-3003/$ - see front matter
2005 Elsevier Inc. All rights reserved.
doi:10.1016/j.amc.2004.12.052
782 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800

with the initial condition

uðx; 0Þ ¼ /ðxÞ; a6x6b ð2Þ
and the boundary conditions
uða; tÞ ¼ q1 ðtÞ and uðb; tÞ ¼ q2 ðtÞ; t > 0; ð3Þ
where Re > 0, called Reynolds number and /, q1 and q2 are the prescribed
functions of the variables. Historically, Eq. (1), which was first introduced
by Bateman [1] who gives its steady solutions. It was later treated by Burgers
[2,3] as a mathematical model for free turbulence and after whom such an
equation is widely referred to as Burgers
equation, is one of a few well-know
nonlinear partial differential equations, which have been solved analytically for
a restricted set of arbitrary initial conditions; see [4,5]. Benton and Platzman [6]
surveyed exact solutions of the one-dimensional Burgers-like equations. Many
problems can be modeled by the Burgers
equation; see [7]. For example, the
Burgers
equation is a one-dimensional analogue of the Navier–Stokes equa-
tions, nonlinear model of the incompressible momentum equations; see [8].
The first term is an unsteady term, the second represents nonlinear convection
and the term on the right-hand side models viscous dissipation.
This problem can be considering a singular perturbation problem of the par-
abolic partial differential equations involving small parameter is considered.
This model has been studied by many researchers for the following reasons:
(i) It contains the simplest form of nonlinear advection term uux and dissipa-
1
tion term Re uxx for simulating the physical phenomena of wave motion. (ii)
Its analytical solution was obtained by [4] so that numerical comparison can
be made; (iii) Its shock wave behavior when Re is large.
Numerical techniques for the solution of Burgers
equation usually fall into
the following classes: finite difference, finite element and spectral methods. For
a survey of these methods, we refer to [9–14].
In our method we develop an extensive analysis of two-level three-point fi-
nite difference method of order 2 in time and 4 in space by computing the local
truncation error. We have applied the present method to Burgers
equation (1)
with a set of boundary and initial conditions given by Eqs. (2) and (3). Numer-
ical computations for a wide range of values of Re, show that the present
method offers better accuracy in comparison with all the previous methods.
Furthermore, this method proposed here is of a general nature and can be used
for solving nonlinear partial differential equations arising in other areas.
The outline of this paper is as follows. In Section 2, we derive the two-level
three-point finite difference method of order 2 in time and 4 in space. Stability
analysis is discussed in Section 3. Section 4, devoted for the convergence anal-
ysis of the method. Numerical comparisons are made with most of the existing
numerical methods for solving Burgers
equation are introduced in the final
section.
 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800 783

2. Methods of solution

In this section, we derive a tridiagonal finite-difference method for solving

Burgers
equation. The solution domain {(x, t):x 2 [a, b], t 2 [0, 1)} is discret-
ized into cells described by the node set (xi, tj) in which xi = a + ih;
i = 0, 1, . . . , N; tj = jDt; j = 0, 1, . . . , J; h = 1/N is a mesh width, Dt = tmax/J
is the time step and tmax is the final time. Here, naturally, N and J are positive
integers. We define the general two-level three-point method by the following

a1 U mþ1 mþ1 mþ1

iþ1;jþ1 þ a0 U i;jþ1 þ a
1 U i
1;jþ1 ¼ b1 U iþ1;j þ b0 U i;j þ b
1 U i
1;j ;

i ¼ 1; 2; . . . ; ðN
1Þ; j ¼ 0; 1; . . . ; ðJ
1Þ: ð4Þ

Here and throughout the work, Ui, j is the approximate solution at the mesh
point (xi, tj) and U mi;jþ1 denotes the mth iterate of the approximate solution at the
mesh point (xi, tj+1) with an initial guess U 0i;jþ1 satisfies the boundary condi-
tions. We note the facts about (4): (i) The matrix represented by AU mþ1 i;jþ1 ,
A = [a
1, a0, a1] is tridiagonal, thus the discretized system is very easily solved.
(ii) No fictitious points or extra boundary conditions are needed.
We derive the present method by computing the local truncation error,
jþ1=2
si , at the point (xi, tj+1/2) as follows:
jþ1=2
si  a1 uðxiþ1 ; tjþ1 Þ þ a0 uðxi ; tjþ1 Þ þ a
1 uðxi
1 ; tjþ1 Þ



b1 uðxiþ1 ; tj Þ þ b0 uðxi ; tj Þ þ b
1 uðxi
1 ; tj Þ

QðLuðxi ; tjþ1=2 ÞÞ; ð5Þ

where Q is tridiagonal operator in the form

Qðuðxi ; tjþ1=2 ÞÞ ¼ Qjþ1 ðuðxi ; tjþ1 ÞÞ þ Qj ðuðxi ; tj ÞÞ;

Qj ðuðxi ; tj ÞÞ ¼ qþ 0

j uðxiþ1 ; t j Þ þ qj uðxi ; t j Þ þ qj uðxi
1 ; t j Þ;

Qjþ1 ðuðxi ; tjþ1 ÞÞ ¼ qþ 0

jþ1 uðxiþ1 ; t jþ1 Þ þ qjþ1 uðxi ; t jþ1 Þ þ qjþ1 uðxi
1 ; t jþ1 Þ:

The operators Lu(xi, tj) and Lu(xi, tj+1) in (5) are defined in the form
Luðxi ; tj Þ  ut ðxi ; tj Þ þ U i;j ux ðxi ; tj Þ
euxx ðxi ; tj Þ ð6aÞ

and
Luðxi ; tjþ1 Þ  ut ðxi ; tjþ1 Þ þ U mi;jþ1 ux ðxi ; tjþ1 Þ
euxx ðxi ; tjþ1 Þ; ð6bÞ
1
respectively, where e ¼ Re.
For a sufficiently smooth function u(x, t), the standard Taylor development
jþ1=2
of si is given by
784 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800

jþ1=2
si ¼ T 0;0 uðxi ; tjþ1=2 Þ þ T 1;0 uð1Þ ðxi ; tjþ1=2 Þ þ    þ T 6;0 uð6Þ ðxi ; tjþ1=2 Þ
þ T 0;1 ut ðxi ; tjþ1=2 Þ þ T 1;1 utð1Þ ðxi ; tjþ1=2 Þ þ    þ T 6;1 uð6Þ
t ðxi ; t jþ1=2 Þ

þ T 0;2 utt ðxi ; tjþ1=2 Þ þ T 1;2 uð1Þ

tt ðxi ; t jþ1=2 Þ þ    þ T
6;2 ð6Þ
utt ðxi ; tjþ1=2 Þ
 
þ O ðDtÞ3 þ h5 : ð7Þ

We can compute the coefficients T0,0, T1,0, . . . , T6,0 in the following form

T 0;0 ¼ a1 þ a0 þ a
1
ðb1 þ b0 þ b
1 Þ;


h 1
1;0
T ¼ a1
a
1
ðb1
b
1 Þ
qþ Um þ q0jþ1 U mi;jþ1 þ q
m
jþ1 U i
1;jþ1
1! h jþ1 iþ1;jþ1

þ 0

þqj U iþ1;j þ qj U i;j þ qj U i
1;j ;

h2 2 þ 
T 2;0
¼ a1 þ a
1
ðb1 þ b
1 Þ þ qjþ1 þ q0jþ1 þ q
jþ1 þ q þ
j þ q 0
j þ q

j
2! hz

2 

qþ U m

q

U m
þ q þ
U iþ1;j
q

U i
1;j ;
h jþ1 iþ1;jþ1 jþ1 i
1;jþ1 j j


h3 6 þ 
T 3;0 ¼ a1
a
1
ðb1
b
1 Þ þ qjþ1
q
jþ1 þ q þ
j
q

j
3! hz
3 

qþ U m
þ q

U m
þ q þ
U iþ1;j þ q

U i
1;j ; ð8aÞ
h jþ1 iþ1;jþ1 jþ1 i
1;jþ1 j j

h‘ n
T ‘;0 ¼ a1 þ ð
1Þ‘ a
1
ðb1 þ ð
1Þ‘ b
1 Þ
‘!
‘ð‘
1Þ h þ i
þ qjþ1 þ ð
1Þ‘ q
þ
jþ1 þ qj þ ð
1Þ qj
‘

hz
‘ h
‘
1

qþ Um þ ð
1Þ q
m
jþ1 U i
1;jþ1
h jþ1 iþ1;jþ1
io
‘
1

þqþ j U iþ1;j þ ð
1Þ qj U i
1;j ; ‘ ¼ 4; 5; 6; ð8bÞ

where z ¼ he . Also, we compute the remaining coefficients in the following form


ðDtÞk k
1
T 0;k ¼ k a1 þ a0 þ a
1 þ ð
1Þ ðb1 þ b0 þ b
1 Þ
2 k!
2k h þ i
k
1 þ

qjþ1 þ q0jþ1 þ q

jþ1 þ ð
1Þ ðq j þ q 0
j þ q

j Þ ;
Dt
 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800 785

hðDtÞ n
k
k
1
T 1;k ¼ a1
a
1 þ ð
1Þ ðb1
b
1 Þ
2k k!
2k h þ k
1 þ
i

qjþ1
q
jþ1 þ ð
1Þ ðqj
q
j Þ
Dt
1h

qþ Um þ q0jþ1 U mi;jþ1 þ q
m
jþ1 U i
1;jþ1
h jþ1 iþ1;jþ1
io
k
þ ð
1Þ ðqþ 0

j U iþ1;j þ qj U i;j þ qj U i
1;j Þ ;
k 
h2 ðDtÞ k
1
T 2;k ¼ a1 þ a
1 þ ð
1Þ ðb1 þ b
1 Þ
2! 2k k!
2k h þ k
1 þ
i

qjþ1 þ q
jþ1 þ ð
1Þ ðqj þ q
j Þ
Dt
2h þ k þ
i
þ qjþ1 þ q0jþ1 þ q
jþ1 þ ð
1Þ ðq j þ q 0
j þ q

j Þ
hz
h
2 þ m i

m k þ


qjþ1 U iþ1;jþ1
qjþ1 U i
1;jþ1 þ ð
1Þ ðqj U iþ1;j
qj U i
1;j Þ ;
h
k
h3 ðDtÞ
T 3;k ¼ a1
a
1 þ ð
1Þk
1 ðb1
b
1 Þ
3! 2k k!
2k h þ k
1 þ
i

qjþ1
q
jþ1 þ ð
1Þ ðqj
q
j Þ
Dt
6h þ k
1 þ
i
þ qjþ1
q
jþ1 þ ð
1Þ ðqj
q
j Þ
hz
3h i
k þ

qþ U m
þ q

U m
þ ð
1Þ ðq U iþ1;j þ q

U i
1;j Þ ;
h jþ1 iþ1;jþ1 jþ1 i
1;jþ1 j j

ð9aÞ

h‘ ðDtÞk n ‘ k
1 ‘
T ‘;k ¼ a1 þ ð
1Þ a
1 þ ð
1Þ ðb1 þ ð
1Þ b
1 Þ
‘! 2k k!
2k h þ ‘ k
1 þ ‘
i

qjþ1 þ ð
1Þ q
jþ1 þ ð
1Þ ðqj þ ð
1Þ q
j Þ
Dt
‘ð‘
1Þ h þ ‘ k þ ‘

i
þ qjþ1 þ ð
1Þ q
jþ1 þ ð
1Þ ðqj þ ð
1Þ qj Þ
hz
‘h þ m ‘
1 k þ

qjþ1 U iþ1;jþ1 þ ð
1Þ q
m
jþ1 U i
1;jþ1 þ ð
1Þ ðqj U iþ1;j
h io
‘
1

þð
1Þ qj U i
1;j Þ ; ‘ ¼ 4; 5; 6 and k ¼ 1; 2: ð9bÞ
jþ1=2
The truncation error is said to be formally of order p if si ¼
Oðhp Þ as h ! 0 (e fixed) for i = 1, 2, . . . , (N
1), j = 1, 2, . . . , (J
1). In the
present work, we derive our method for solving (1) by the following conditions.
786 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800

T ‘;k ¼ 0; ‘ ¼ 0; 1; 2; k ¼ 0; 1 ð10Þ
and

T ‘;k ¼ Oðh4 Þ; ‘ ¼ 3; 4; k ¼ 0; 1: ð11Þ

From the conditions (10), the coefficients a1, a0, a
1, b1, b0 and b
1 are deter-
mined in terms of qþ;0;

j and qþ;0;

jþ1 as follows:
~sþ 1 h ~ i
a1 ¼
2S jþ1
zð3qþ U m
jþ1 iþ1;jþ1 þ q 0
U m
jþ1 i;jþ1
q

U m
Þ
jþ1 i
1;jþ1 ;
Dt 2hz
~s0 2 h~ i
a0 ¼ þ S jþ1
zðqþ U m
jþ1 iþ1;jþ1
q

U m
jþ1 i
1;jþ1 ; Þ
Dt hz
~s
1 h ~ i
a
1 ¼
2S jþ1
zðqþ U m
jþ1 iþ1;jþ1
q 0
U m
jþ1 i;jþ1
3q

U m
Þ
jþ1 i
1;jþ1 ;
Dt 2hz
~sþ 1 h ~ i
b1 ¼ þ 2S j
zð3qþ 0

j U iþ1;j þ qj U i;j
qj U i
1;j Þ ;
Dt 2hz
~s0 2 h~ i
b0 ¼
S j
zðqþ j U iþ1;j
q

j U i
1;j Þ ;
Dt hz
~s
1 h ~ i
b
1 ¼ þ 2S j
zðqþ j U iþ1;j
q 0
U
j i;j
3q

j U Þ
i
1;j ;
Dt 2hz
ð12Þ

where
~sþ ¼ qþ þ
s0 ¼ q0j þ q0jþ1 ; ~s
¼ q

j þ qjþ1 ; ~


j þ qjþ1 ;

S~jþ1 ¼ qþ 0
~ þ 0

jþ1 þ qjþ1 þ qjþ1 ; S j ¼ qj þ qj þ qj :

Substituting from (12) into (8) and separate T3,k and T4,k, k = 0, 1 at the level
j and j + 1 as follows:
eh h þ i
T 3j ¼ 6qj
6q
þ 0

j þ zð
2qj U iþ1;j þ qj U i;j
2qj U i
1;j Þ ;
3!
ð13Þ
4 eh2 h þ 0
þ

i
Tj ¼ 10qj
2qj þ 10qj
2zðqj U iþ1;j
qj U i
1;j Þ
4!
and
eh h þ i
T 3jþ1 ¼ 6qjþ1
6q
þ m 0 m
m
jþ1 þ zð
2qjþ1 U iþ1;jþ1 þ qjþ1 U i;jþ1
2qjþ1 U i
1;jþ1 Þ ;
3!
eh2 h i
T 4j ¼ 10qþjþ1
2q 0
jþ1 þ 10q

jþ1
2zðq þ
U m
jþ1 iþ1;jþ1
q

U m
Þ
jþ1 i
1;jþ1 :
4!
ð14Þ
 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800 787

We define qj
;0;þ as polynomials in z at each mesh point xi in the form

X
2
qj
;0;þ ¼ n
;0;þ
‘ z‘ ; ð15Þ
‘¼0

where the coefficients n‘
;0;þ , ‘ = 0, 1, 2 are independent of e. We determine these

coefficients by substituting from (15) into (13) and imposing (11). This result is
written in the form
eh h i
T 3j ¼ c0 þ c1 z þ c2 z2 þ Oðz3 Þ ¼ Oðh4 Þ;
3!
eh2 h i
T 4j ¼ c3 þ c4 z þ Oðz2 Þ ¼ Oðh4 Þ;
4!
where
c0 ¼ 6nþ
3 þ 0
2
0
6n0 ¼ Oðh Þ; c3 ¼ 10n0
2n0 þ 10n0 ¼ Oðh Þ;

c1 ¼ 6nþ
þ 0
2
1
6n1
2n0 U iþ1;j þ n0 U i;j
2n0 U i
1;j ¼ Oðh Þ;
ð16Þ
c4 ¼ 10nþ 0
þ

1
2n1 þ 10n1
2ðn0 U iþ1;j
n0 U i
1;j Þ ¼ OðhÞ;

c2 ¼ 6nþ
þ 0

2
6n2
2n1 U iþ1;j þ n1 U i;j
2n1 U i
1;j ¼ OðhÞ:

Now, we use the approach in Berger et al. [15], to determine the coefficients
n‘
;0;þ , ‘ = 0, 1, 2 in the form
nþ

0 ¼ n0 ¼ 2; n00 ¼ 20; nþ 0
1 ¼
U i;j ; n1 ¼ 0 and n

1 ¼ U i;j : ð17Þ
It can be verified by direct substitution that, (17) satisfies (16). Then, the
coefficients qj
;0;þ are given in the form
q

j ¼ 2 þ zU i;j ; q0j ¼ 20; qþ
j ¼ 2
zU i;j : ð18aÞ
In the same manner, from (11) and (14), the coefficients q
;0;þ
jþ1 are written in
the form
q
m
jþ1 ¼ 2 þ zU i;jþ1 ; q0jþ1 ¼ 20; qþ m
jþ1 ¼ 2
zU i;jþ1 : ð18bÞ
Finally, the present method can be written in the following matrix form
AU mþ1
i;jþ1 ¼ BU i;j ð19Þ
with the discretization matrices A and B
A ¼ ½a
1 ; a0 ; a1  and ½b
1 ; b0 ; b1 :
The matrix A is tridiagonal, therefore the resulting linear systems can be
solved very efficiently linearly in time using a special form of Gaussian elimina-
tion known as the Thomas algorithm. Also, substitution from (13), (18) into
(7), we can show that the local truncation error of order 2 in time and 4 in
space for e fixed.
788 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800

3. Stability considerations

In this section, we prove that the present method is unconditionally stable.

We use the von Neumann linear stability analysis. The present method (4) ad-
mits a solution of the following form

U i;j ¼ Zj expðIaihÞ; ð20Þ

pffiffiffiffiffiffiffi
where I ¼
1 and a is a positive constant. Substituting from (20) into (4)
gives

Z jþ1 ½a1 expðIahÞ þ a0 þ a
1 expð
IahÞ

¼ Z j ½b1 expðIahÞ þ b0 þ b
1 expð
IahÞ:

The growth of the Fourier component by amplification factor

Zjþ1 b0 þ ðb1 þ b
1 Þ cos ah þ Iðb1
b
1 Þ sin ah

G¼ ¼ : ð21Þ
Zj a0 þ ða1 þ a
1 Þ cos ah þ Iða1
a
1 Þ sin ah

~ ¼ zU i;j ¼ zU mi;jþ1 in Eq. (12) and substitution into

For simplicity, we let q
Eq. (21) gives,

4 þ ~cH ðahÞ
G¼ ;
4
~cH ðahÞ

where ~c ¼ Dt~
k > 0 and

ðcos ah
1Þð48 þ 4~q2 Þ
Ið24~q sin ahÞ

H¼ :
10 þ 2 cos ah
Ið~q sin ahÞ

The strict von Neumann condition for stability is that jGj 6 1; see Smith
e 6 0. Direct
[16]. This condition is satisfied if and only if real part of H equal H
computation of H e yields

q2 Þð10 þ 2 cos ahÞ þ 24~

e ¼ ðcos ah
1Þð48 þ 4~ q2 D 1
H ; ð22Þ
D2
h
where D1 i¼ sin2 ah describes the interval [0, 1] and D2 ¼ ð10 þ 2 cos ahÞ2 þ
q sin ahÞ2 > 0, then we can write (22) in the form
ð~


q2 ð1
cos ahÞ :
ðcos ah
1Þ 480 þ 96 cos ah þ 16~
e
H ¼ ð23Þ
D2
e 6 0, then the present method (4) is uncondition-
From (23), it is clear that H
ally stable.
 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800 789

4. Convergence analysis

In this section, we study the convergence of the present method. For simplic-
ity, let ui, j = u(xi, tj), be exact solution of (1) and Fe i;j ¼ BU i;j . Then, we can
rewrite (19) in the form
ðA0 þ hA1 þ h2 A2 þ h3 A3 ÞU i;jþ1 ¼ Fe i;j ; ð24Þ
where A0 is (N
1) · (N
1) matrix in the form
2 3

2 1
6 7
6 1
2 1 7
6 7
6 7
A0 ¼ 6 0 1
2 1 7;
6 7
6     7
4 5
0 0 0 1
2

and the explicit form of the matrices A1, A2 and A3 can be obtained in a similar
way. The exact form of the system (24) is written as follows

h2 jþ1=2
ðA0 þ hA1 þ h2 A2 þ h3 A3 Þui;jþ1 ¼ Fe i;j
s : ð25Þ
24e i
Our convergence analysis is given by the following theorem:

Theorem 1. Let Ui, j+1 be obtained by the method (19). Then, at each mesh point
(xi, tj+1), we have the following
~ 2
kEjþ1 k 6 KððDtÞ þ h4 Þ;
where kÆk represent the 1-norm in matrix vector, kEj k ¼ max kei;j k, ei,j =
i
~ be a constant.
ui,j
Ui,j and K

To prove this theorem, we will need the following lemma:

Lemma 1 (See Fröberg [17]). If U = (Ul, m) is a square matrix of order (N
1)

with kUk < 1, then (I0 + U) is a nonsingular matrix and

1 1
kðI 0 þ UÞ k 6 ;
1
kUk
where
X
N
1
kUk ¼ max jUl;m j
l
m¼1

and I0 is the identity matrix.

790 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800

Proof of Theorem 1. Subtracting (24) from (25) we have

2
e
ðA0 þ hA1 þ h2 A2 þ h3 A3 ÞEjþ1 ¼ R; e ¼
h sijþ1=2 ;
R
24e
using Lemma 1 we have
 
kA
1 e X 3 
0 k:k Rk  m
1 
kEjþ1 k 6
1
; kMk ¼  Am h ; ð26Þ
1
hkA0 k  kMk  m¼1 

provided that

1
h< :
kA
1
0 k  kMk

In order to find kA
1 0 k ; see Usmani and Marsden [18], the matrix A is a
0

1
monotone matrix and if A0 ¼ ðal;m Þ, we have
8
< mðl
N Þ ; l P m;
N
al;m ¼
: lðm
N Þ ; l 6 m:
N

We notice that al, m < 0 for all l, m = 1, 2, . . . , N
1. Then

X
N X
l X
N
1
lðN þ 2ÞðN
lÞ
Sl ¼ jal;m j ¼
al;m
al;m ¼ ;
m¼1 m¼1 m¼l
2N

we can easily prove that the sequence {Sl}, is maximum at l ¼ N
1

2
, hence

N 3 þ 2N 2
N
2
kA
1
0 k ¼ S ðN
1Þ=2 ¼ ¼ h
2 Z;
~ ð27Þ
8N
where
" #
3 2 2 3
ðb
aÞ þ 2hðb
aÞ
h ðb
aÞ
2h
Z~ ¼ :
8ðb
aÞ

Now, from our method based on (7), we have

e ¼ Oðh2 ðDtÞ þ h6 Þ:
k Rk
2
ð28Þ

Thus using (26)–(28) we get

2
kEjþ1 k ¼ OððDtÞ þ h4 Þ:

From this result we deduce unconditional convergence as Dt ! 0 and

h ! 0. h
 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800 791

5. Numerical experiments and discussions

In this section, we consider the numerical solution of Burgers
equation (1)

for three test examples.

Example 1. We consider Burgers
equation (1) with boundary and initial

conditions in the following form
uð0; tÞ ¼ uð1; tÞ ¼ 0; t > 0: ð29Þ

uðx; 0Þ ¼ sin px; 0 6 x 6 1: ð30Þ

The exact solution for this example given by Cole [4] is
P
2pe 1 2 2
n¼1 nd n sinðnpxÞ expð
n ep tÞ
uðx; tÞ ¼ P 1 ð31Þ
d 0 þ n¼1 d n cosðnpxÞ expð
n2 ep2 tÞ
with the Fourier coefficients
Z 1
d0 ¼ expf
ð2peÞ
1 ½1
cosðpxÞg dx;
0

Z 1
dn ¼ 2 expf
ð2peÞ
1 ½1
cosðpxÞg cosðnpxÞ dx; n ¼ 1; 2; 3; . . . :
0

The results for this example are presented in Tables 1–6 for Re = 0.1, 1.0,
10.0, 100.0 and 10,000.0. We solve this example for different values of N and
t. These tables show a comparison between the exact solution with our present
method and other methods which are introduced by [19–24]. The numerical
comparisons show the present method offers better results than the other
methods.

Table 1
Comparison between exact and numerical solutions of Example 1 for Re = 0.1
x t = 0.01, Dt = 0.0001 t = 0.02, Dt = 0.0001
Exact Hon and Present Exact Hon and Present
solution Mao [21] method solution Mao [21] method
0.1 0.1146 0.1152 0.1146 0.0428 0.0433 0.0428
0.2 0.2182 0.2192 0.2182 0.0815 0.0823 0.0815
0.3 0.3006 0.3021 0.3006 0.1122 0.1133 0.1122
0.4 0.3539 0.3556 0.3539 0.1320 0.1333 0.1320
0.5 0.3727 0.3745 0.3727 0.1389 0.1403 0.1389
0.6 0.3550 0.3567 0.3550 0.1322 0.1335 0.1322
0.7 0.3024 0.3039 0.3024 0.1125 0.1136 0.1125
0.8 0.2200 0.2211 0.2200 0.0818 0.0826 0.0818
0.9 0.1157 0.1163 0.1157 0.0430 0.0434 0.0430
792 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800

Table 2
Comparison between exact and numerical solutions of Example 1 for Re = 1.0 and t = 0.1
x Exact Kutluay et al. [23] Aksan and Present method
solution Özdes [19]
Explicit Exact-explicit
Dt = 0.00001 Dt = 0.0001
N = 80 N = 10 N = 80 N = 10 N = 10 N = 80
0.1 0.10954 0.10952 0.11048 0.10955 0.10958 0.10954 0.10954
0.2 0.20979 0.20975 0.21159 0.20981 0.20989 0.20980 0.20979
0.3 0.29190 0.29184 0.29435 0.29192 0.29199 0.29190 0.29190
0.4 0.34792 0.34786 0.35080 0.34795 0.34709 0.34794 0.34792
0.5 0.37158 0.37151 0.37458 0.37161 0.37173 0.37159 0.37158
0.6 0.35905 0.35898 0.36189 0.35907 0.35920 0.35906 0.35905
0.7 0.30991 0.30985 0.31231 0.30993 0.31003 0.30992 0.30991
0.8 0.22782 0.22778 0.22955 0.22783 0.22792 0.22783 0.22782
0.9 0.12069 0.12067 0.12160 0.12070 0.12071 0.12069 0.12069

Table 3
Comparison between exact and numerical solutions of Example 1 for Re = 10.0
x Exact t = 0.5, Dt = 0.01 Exact t = 1.0, Dt = 0.01
solution Hon and Present solution Hon and Present
Mao [21] method Mao [21] method
N = 10 N = 10 N = 100 N = 10 N = 10 N = 100
0.1 0.1099 0.1104 0.1099 0.1099 0.0663 0.0664 0.0663 0.0663
0.2 0.2180 0.2186 0.2181 0.2180 0.1312 0.1314 0.1312 0.1312
0.3 0.3222 0.3227 0.3222 0.3222 0.1928 0.1930 0.1928 0.1928
0.4 0.4190 0.4194 0.4191 0.4190 0.2480 0.2483 0.2481 0.2480
0.5 0.5028 0.5028 0.5029 0.5028 0.2919 0.2923 0.2921 0.2919
0.6 0.5623 0.5618 0.5625 0.5623 0.3161 0.3167 0.3163 0.3161
0.7 0.5759 0.5744 0.5763 0.5759 0.3081 0.3090 0.3084 0.3081
0.8 0.5055 0.5030 0.5063 0.5056 0.2537 0.2548 0.2541 0.2537
0.9 0.3093 0.3059 0.3101 0.3094 0.1461 0.1468 0.1464 0.1461

Figs. 1–5, show the present numerical solutions for N = 300 at different
values of t and Re, which exhibit the physical behaviour of this example. Figs. 4
and 5 agree well with Figs. 1d and 2 of Hon and Mao [21] and Daǧ et al. [9],
respectively.

Example 2. We consider Burgers
equation (1) with the initial condition

uðx; 0Þ ¼ 4xð1
xÞ; 0 6 x 6 1 ð32Þ
and the boundary conditions (29). The exact solution to this example is given
by (31), but in this case the Fourier coefficients are
 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800 793

Table 4
Comparison between exact and numerical solutions of Example 1 at different times for Re = 10.0
x t Kutluay et al. [23] Özis et al. [24] Present method Exact solution
Explicit Exact-explicit
N = 40, Dt = 0.001
0.25 0.4 0.30834 0.30891 0.31429 0.30889 0.30889
0.6 0.24039 0.24075 0.24373 0.24074 0.24074
0.8 0.19543 0.19568 0.19758 0.19568 0.19568
1.0 0.16238 0.16257 0.16391 0.16256 0.16256
3.0 0.02718 0.02720 0.02743 0.02720 0.02720
0.50 0.4 0.56911 0.56964 0.57636 0.56963 0.56963
0.6 0.44676 0.44721 0.45169 0.44721 0.44721
0.8 0.35888 0.35924 0.36245 0.35924 0.35924
1.0 0.29162 0.29192 0.29437 0.29192 0.29192
3.0 0.04017 0.04021 0.04057 0.04021 0.04021
0.75 0.4 0.62555 0.62542 0.62592 0.62544 0.62544
0.6 0.48701 0.48721 0.49034 0.48722 0.48721
0.8 0.37366 0.37392 0.37713 0.37392 0.37392
1.0 0.28723 0.28748 0.29016 0.28748 0.28747
3.0 0.02974 0.02977 0.01334 0.02977 0.02977

Table 5
Comparison between exact and numerical solutions of Example 1 at different times for Re = 100.0
x t Kutluay et al. [23] Present method Exact solution
Explicit Exact-explicit
N = 80, Dt = 0.001
0.25 0.4 0.34244 0.34164 0.34191 0.34191
0.6 0.26905 0.26890 0.26896 0.26896
0.8 0.22145 0.22150 0.22148 0.22148
1.0 0.18813 0.18825 0.18819 0.18819
3.0 0.07509 0.07515 0.07511 0.07511
0.50 0.4 0.67152 0.65606 0.66071 0.66071
0.6 0.53406 0.52658 0.52942 0.52942
0.8 0.44143 0.43743 0.43914 0.43914
1.0 0.37568 0.37336 0.37442 0.37442
3.0 0.15020 0.15015 0.15218 0.15018
0.75 0.4 0.94675 0.90111 0.91027 0.91026
0.6 0.78474 0.75862 0.76724 0.76724
0.8 0.65659 0.64129 0.64740 0.64740
1.0 0.56135 0.75862 0.55605 0.55605
3.0 0.22502 0.22502 0.22481 0.22481
794 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800

Table 6
Comparison between numerical solutions of Example 1 for Re = 10,000.0 and t = 1.0
x Caldwell Iskander and Hon and Present method
et al. [20] Mohsen [22] Mao [21] N = 180
Dt = 0.001 Dt = 0.1 Dt = 0.001 Dt = 0.01
0.05 0.0422 0.0419 0.0424 0.0379
0.11 0.0844 0.0839 0.0843 0.0834
0.16 0.1266 0.1253 0.1263 0.1213
0.22 0.1687 0.1692 0.1684 0.1667
0.27 0.2108 0.2034 0.2103 0.2044
0.33 0.2527 0.2666 0.2522 0.2469
0.38 0.2946 0.2527 0.2939 0.2872
0.44 0.3362 0.3966 0.3355 0.3322
0.50 0.3778 0.2350 0.3769 0.3769
0.55 0.4191 0.5480 0.4182 0.4140
0.61 0.4601 0.2578 0.4592 0.4584
0.66 0.5009 0.6049 0.4999 0.4951
0.72 0.5414 0.6014 0.5404 0.5388
0.77 0.5816 0.4630 0.5805 0.5749
0.83 0.6213 0.7011 0.6201 0.6179
0.88 0.6605 0.6717 0.6600 0.6533
0.94 0.6992 0.7261 0.6957 0.6952

1
t = 0.004, 0.008, 0.012,
0.8
0.014 and 0.02
0.6

U 0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
x

Fig. 1. Numerical solution of Example 1 at different times for N = 300 and Re = 0.1.

Z 1

1
d0 ¼ expf
x2 ð3eÞ ½3
2xg dx;
0

Z 1
dn ¼ 2 expf
x2 ð3eÞ
1 ½3
2xg cosðnpxÞ dx; n ¼ 1; 2; 3; . . . :
0

The numerical results for this example are presented in Tables 7 and 8 for
Re = 1.0 and 10.0, respectively. We solve this example for N = 80 and different
values of t. These tables show a comparison between the exact solution with
 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800 795

0.8

t = 0.25, 0.2, 0.15, 0.1 and 0.05

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
x

Fig. 2. Numerical solution of Example 1 at different times for N = 300 and Re = 1.0.

0.8 t = 0.2, 0.4, 0.6, 0.8 and 1.0

0.6

U
0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
x

Fig. 3. Numerical solution of Example 1 at different times for N = 300 and Re = 10.

our present method and the numerical results introduced by Özis et al. [24]. It is
clearly seen that the present results are in good agreement with the exact solu-
tion. Fig. 6 shows the present numerical solution of this example for N = 40
and Re = 100.0 at different times which exhibit the correct physical behavior
of this example. All solutions are drawn on the same figure since they are very
close to each other. Fig. 6 agree well with Fig. 2 of Kutluay et al. [23].
796 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800

1
t = 0.2, 0.4, 0.6, 0.8, 1.0
0.8
0.6
U 0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
x

Fig. 4. Numerical solution of Example 1 at different times for N = 300 and Re = 100.

1
t=0.2, 0.4, 0.6, 0.8 and 1.0
0.8

0.6

0.4
U
0.2

0
0 0.2 0.4 0.6 0.8 1
x

Fig. 5. Numerical solution of Example 1 at different times for N = 300 and Re = 10,000.

Example 3. We consider the particular solution of Burgers
equation in the

form
½a0 þ l0 þ ðl0
a0 Þ expð^
gÞ 
uðx; tÞ ¼ ; t P 0; ð33Þ
1 þ expð^
gÞ

where ^g ¼ a0 Reðx
l0 t
b0 Þ, the constants a0, l0 and b0 are arbitrary. The ini-
tial condition is found from (33) when t = 0. For this function u ! 1 as x ! 0
and u ! 0.2 as x ! 1. The numerical results for this example are presented in
Table 9 for a0 = 0.4, b0 = 0.125, l0 = 0.6, t = 0.5, N = 36 and Re = 100.0. This
table shows a comparison between the exact solution with our method and
other methods which are introduced by [9,25,26]. The numerical comparison
shows that our method is good agreement with the exact solution and other
 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800 797

Table 7
Comparison between exact and numerical solutions of Example 2 at different times for Re = 1.0
x t N = 80 Exact solution
Özis et al. [24] Present method
Dt = 0.00001 Dt = 0.001
0.25 0.10 0.26245 0.26148 0.26148
0.15 0.16157 0.16148 0.16148
0.20 0.09948 0.09947 0.09947
0.25 0.06111 0.06109 0.06108
0.50 0.10 0.38314 0.38342 0.38342
0.15 0.23394 0.23405 0.23406
0.20 0.14287 0.14289 0.14289
0.25 0.08729 0.08723 0.08723
0.75 0.10 0.28004 0.28157 0.28157
0.15 0.16948 0.16974 0.16974
0.20 0.10261 0.10265 0.10266
0.25 0.06230 0.06229 0.06229

Table 8
Comparison between exact and numerical solutions of Example 2 at different times for Re = 10.0
x t N = 80 Exact solution
Özis et al. [24] Present method
Dt = 0.00001 Dt = 0.001
0.25 0.4 0.32679 0.31752 0.31752
0.6 0.25117 0.24614 0.24614
0.8 0.20270 0.19955 0.19956
1.0 0.16780 0.16559 0.16560
3.0 0.02804 0.02776 0.02775
0.50 0.4 0.59661 0.58454 0.58454
0.6 0.46581 0.45798 0.45798
0.8 0.37293 0.36740 0.36740
1.0 0.30253 0.29834 0.29834
3.0 0.04155 0.04106 0.04106
0.75 0.4 0.64680 0.64562 0.64562
0.6 0.50852 0.50268 0.50268
0.8 0.39117 0.38534 0.38534
1.0 0.30066 0.29586 0.29586
3.0 0.03081 0.03044 0.03044

methods. Fig. 7 illustrates the numerical solution of this example for Re = 100,
N = 36 at different time. The exact solution given by Eq. (33) also is drawn on
798 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800

0.8

0.6 t = 0.4, 0.6, 0.8 and 1.0

U 0.4

0.2
t = 3.0

0
0 0.2 0.4 0.6 0.8 1
x

Fig. 6. Numerical solution of Example 2 at different times for N = 40 and Re = 100.

Table 9
Comparison between exact and numerical solutions of Example 3 for Re = 100.0
x t = 0.5, N = 36 Exact solution
Ali et al. [25] Dogan [26] Daǧ et al. [9] Present method
Dt = 0.025 Dt = 0.05 Dt = 0.025 Dt = 0.01
0.000 1.0 1.0 1.0 1.0 1.0
0.056 1.0 1.0 1.0 1.0 1.0
0.111 1.0 1.0 1.0 1.0 1.0
0.167 1.0 1.0 1.0 1.0 1.0
0.222 1.0 1.0 1.0 1.0 1.0
0.278 0.999 0.999 0.999 0.998 0.998
0.333 0.985 0.994 0.986 0.982 0.980
0.389 0.847 0.848 0.850 0.849 0.847
0.444 0.452 0.407 0.448 0.455 0.452
0.500 0.238 0.232 0.236 0.239 0.238
0.556 0.204 0.204 0.204 0.204 0.204
0.611 0.2 0.2 0.2 0.2 0.2
0.667 0.2 0.2 0.2 0.2 0.2
0.722 0.2 0.2 0.2 0.2 0.2
0.778 0.2 0.2 0.2 0.2 0.2
0.833 0.2 0.2 0.2 0.2 0.2
0.889 0.2 0.2 0.2 0.2 0.2
0.944 0.2 0.2 0.2 0.2 0.2
1.000 0.2 0.2 0.2 0.2 0.2

the same figure, but the graphs can not be distinguished due to the closeness of
the numerical solutions to the exact one.
 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800 799

1.2
1
0.8 t=0.05 t=0.5 t=1.0
U
0.6
0.4
0.2
0 0.2 0.4 0.6 0.8 1
x

Fig. 7. Numerical solution of Example 3 at different times for N = 36 and Re = 100.

6. Conclusion

In this paper, a numerical algorithm for the solution of the Burgers
equa-
tion based on two-level three-point finite difference is developed. The present
method is a second-order accurate in Dt and a fourth-order accurate in h. It
was shown that this method is good for any value of Re. The present method
is a simple and accurate method for solving Burgers
equation for wide range of
Reynolds numbers. The present solution is an alternative solution to the exact
(Fourier) one. But if the initial condition of a problem in known only at the
finite number of the mesh points, for such a problem the present method is
much more practical than the Fourier method. It is believed that the present
approach will also prove useful for solving more general problems in fluid
dynamics.

References

[1] H. Bateman, Some recent researches on the motion of fluids, Mon. Weather Rev. 43 (1915)
163–170.
[2] J.M. Burgers, Mathematical examples illustrating relations occurring in the theory of turbulent
fluid motion, Trans. Roy. Neth. Acad. Sci. Amsterdam 17 (1939) 1–53.
[3] J.M. Burgers, A mathematical model illustrating the theory of turbulence, Adv. in Appl.
Mech., vol. I, Academic Press, New York, 1948, pp. 171–199.
[4] J.D. Cole, On a quasi-linear parabolic equations occuring in aerodynamics, Quart. Appl.
Math. 9 (1951) 225–236.
[5] E. Hopf, The partial differential equation ut + uux = luxx, Commun. Pure Appl. Math. 3
(1950) 201–230.
[6] E. Benton, G.W. Platzman, A table of solutions of the one-dimensional Burgers equations,
Quart. Appl. Math. 30 (1972) 195–212.
[7] C.A. Fletcher, Burgers
equation: a model for all reasons, in: J. Noye (Ed.), Numerical
Solutions of Partial Differential Equations, North-Holland, Amsterdam, 1982, pp. 139–225.
[8] W.F. Ames, Nonlinear Partial Differential Equations in Engineering, Academic Press, New
York, 1965.
800 I.A. Hassanien et al. / Appl. Math. Comput. 170 (2005) 781–800

_ Daǧ, D. Irk, B. Saka, A numerical solution of the Burgers
equation using cubic B-splines,
[9] I.
Appl. Math. Comput. 163 (2005) 199–211.
[10] S. Kutluay, A. Esen, A linearized numerical scheme for Burgers-like equations, Appl. Math.
Comput. 156 (2004) 295–305.
[11] S. Kutluay, A. Esen, I._ Daǧ, Numerical solutions of the Burgers
equation by the least-squares
quadratic B-spline finite element method, J. Comput. Appl. Math. 167 (2004) 21–33.
[12] R.C. Mittal, P. Singhal, Numerical solution of Burgers
equation, Commun. Numer. Meth.
Eng. 9 (1993) 397–406.
[13] R.C. Mittal, P. Singhal, Numerical solution of periodic Burgers
equation, Indian J. Pure
Appl. Math. 27 (1996) 689–700.
[14] T. Özisß, A. Özdesß, A direct variational methods applied to Burgers
equation, J. Comput.
Appl. Math. 71 (1996) 163–175.
[15] A.E. Berger, J.M. Solomon, M. Ciment, S.H. Leventhal, B.C. Weinberg, Generalized OCI
schemes for boundary layer problems, Math. Comput. 35 (1980) 695–731.
[16] G.D. Smith, Numerical Solution of Partial Differential Equations, second ed., Oxford
University Press, Oxford, 1978.
[17] C. Fröberg, Introduction to Numerical Analysis, second ed., Addison-Wesley, New York,
1969.
[18] R.A. Usmani, M.J. Marsden, Numerical solution of some ordinary differential equations
occurring in plate deflection theory, J. Eng. Math. 9 (1975) 1–10.
[19] E.N. Aksan, A. Özdes, A numerical solution of Burgers
equation, Appl. Math. Comput. 156
(2004) 395–402.
[20] J. Caldwell, P. Wanless, A.E. Cook, Solution of Burgers
equation for large Reynolds number
using finite elements with moving nodes, Appl. Math. Model. 11 (1987) 211–214.
[21] Y.C. Hon, X.Z. Mao, An efficient numerical scheme for Burgers
equation, Appl. Math.
Comput. 95 (1998) 37–50.
[22] L. Iskander, A. Mohsen, Some numerical experiments on the splitting of Burgers
equation,
Numer. Meth. Partial Differen. Equat. 8 (1992) 267–276.
[23] S. Kutluay, A.R. Bahadir, A. Özdesß, Numerical solution of one-dimensional Burgers
equation: explicit and exact-explicit finite difference methods, J. Comput. Appl. Math. 103
(1999) 251–261.
[24] T. Özis, E.N. Aksan, A. Özdes, A finite element approach for solution of Burgers
equation,
Appl. Math. Comput. 139 (2003) 417–428.
[25] A.H.A. Ali, G.A. Gardner, L.R.T. Gardner, A collocation solution for Burgers
equation
using cubic B-spline finite elements, Comput. Meth. Appl. Mech. Eng. 100 (1992) 325–337.
[26] A. Dogan, A Galerkin finite element approach to Burgers
equation, Appl. Math. Comput. 157
(2004) 331–346.

View publication stats