Numerical solution of RLW equation using linear nite

elements within Galerkins method

Abdulkadir Dogan
*
Department of Mathematics, Faculty of ArtScience, University of Nigde, 51100 Nigde, Turkey
Received 12 April 2000; received in revised form 25 July 2001; accepted 8 August 2001
Abstract
The regularised long wave equation is solved by Galerkins method using linear space nite elements. In
the simulations of the migration of a single solitary wave, this algorithm is shown to have good accuracy for
small amplitude waves. Moreover, for very small amplitude waves ( 60.09) it has higher accuracy than an
approach using quadratic B-spline nite elements within Galerkins method. The interaction of two solitary
waves is modelled for small amplitude waves. 2002 Elsevier Science Inc. All rights reserved.
Keywords: RLW equation; Finite element methods; Galerkin
1. Introduction
The regularised long wave (RLW) equation is dened as
U
t
U
x
UU
x
U
xxt
0: 1
It is a dierent explanation of non-linear dispersive waves to the more familiar Korteweg-de Vries
(KDV) equation of the form;
U
t
U
x
UU
x
U
xxx
0: 2
The RLW equation, at rst, proposed by Peregrine [1]. The RLW equation plays a major role
in the study of non-linear dispersive waves [1,2] because of its description a larger number of
important physical phenomena, such as shallow water waves and ion acoustic plasma waves.
Applied Mathematical Modelling 26 (2002) 771783
www.elsevier.com/locate/apm
*
Tel.: +90-388-2250292; fax: +90-388-2250180.
E-mail address: abdulkadir@ttnet.net.tr (A. Dogan).
0307-904X/02/$ - see front matter 2002 Elsevier Science Inc. All rights reserved.
PII: S0307- 904X( 01) 00084- 1
Experimental evidence suggests that this description breaks down if the amplitude of any wave
exceeds about 0.28, as wave breaking is observed with water waves [1].
Due to non-linear nature of the RLW equation, not many exact solutions exist in the litera-
ture. Studies mainly consider numerical solution of the problem. Eilbeck and McGuire [3] pro-
posed some methods for the numerical solution of the RLW equation. Bona et al. [2] presented a
number of numerical methods to solve an evolution equation. Jain et al. [4] applied both the
splitting method and the cubic spline method to solve the RLW equation. Gardner and Gardner
[5] solved the RLW equation using the nite element method based on Galerkins method with
cubic splines as element shape function. They showed that the method is accurate and ecient
after applying to migration and interaction of solitary waves problems.
The RLW equation was also solved using the nite element method based on quadratic B-
spline nite elements within Galerkins method [68]. Nguyen and Reynen [7,8] used a least
squares technique and showed that their method eliminates spurious oscillation in advection
diusion problems. In these studies [7,8] Galerkins method with linear nite element was used to
set up a numerical solution. We used a least-squares method using linear spacetime nite ele-
ments to set up a numerical solution for the RLW equation [9]. Gardner et al. [13] presented
a PetrovGalerkin solution for the equal width equation using quadratic B-spline spatial nite
elements. In a subsequent study, Gardner et al. [14] solved the RLW equation with a Petrov
Galerkin B-spline nite element method. More recently, Dogan [15] solved the RLW equation
with a PetrovGalerkin method using quadratic B-spline nite elements.
In the following we set up Galerkins method to the RLW equation using linear space nite
elements. The properties and advantages of this method is discussed and its accuracy is compared
with that of numerical algorithms described in Refs. [4,6,9,15]. Eventually, the interaction of two
solitary waves of small amplitude is studied.
2. The nite element solution
We shall solve a non-linear RLW equation in the normalised form
U
t
U
x
UU
x
lU
xxt
0; 3
where , l are positive parameters and the subscripts t and x denote dierentiation, with the
physical boundary condition U ! 0 as jxj ! 1.
When the RLW equation is used to model waves generated in a shallow water channel, the
variables are normalised in the following way. Distance x and water elevation U are scaled to the
water depth h and time t is scaled to h=g
1=2
, where g is the acceleration due to gravity.
When applying Galerkins method we minimise the functional [10]
_
x
N
x
0
U
t
U
x
UU
x
lU
xxt
W
j
dx 0; 4
where W
j
is a weight function, with respect to the nodal variables.
The region x
0
; x
N
is partitioned into uniformly sized nite elements of length h by the nodes x
m
such that x
0
< x
1
< < x
N
. A typical nite element of size h x
m1
x
m
, mapped by local
coordinates g, where x x
m
gh, 0 6g 61, makes, to integral (4), the contribution
772 A. Dogan / Appl. Math. Modelling 26 (2002) 771783
_
1
0
U
t
_

1
h
U
g


h

UUU
g

l
h
2
U
ggt
_
W
j
dg 0; 5
where to simplify the integral,

UU is taken to be constant over an element. This leads to
_
1
0
U
t
vU
g
bU
ggt
W
j
dg 0; 6
where
b
l
h
2
;
and
v
1
h
1

UU
and b and v are taken as locally constant over each element. The variation of U over the element
x
m
; x
m1
is expressed as
U
e

2
j1
L
j
u
j
; 7
where L
1
, L
2
are linear spatial basis functions and u
1
, u
2
are the nodal parameters. With the local
coordinate system g dened above the basis functions have following expressions [10]
L
1
1 g;
L
2
g:
If the weight functions W
j
are identied with the basis functions L
j
_
1
0
U
t
vU
g
bU
ggt
L
j
dg 0; 8
is obtained. Integrating Eq. (8) by parts leads to
_
1
0
U
t
vU
g
L
j
bU
gt
L
0
j
dg 0: 9
If U which described in Eq. (7) is substituted in Eq. (9), an elements contribution is obtained in
the form of

2
k1
_
1
0
L
k
ou
k
ot
_ _
vL
0
k
u
k
_
L
j
bL
0
k
L
0
j
ou
k
ot
_
dg 0; 10
where the symbol
0
denotes dierentiation with respect to g, which can be written in matrix form
as
D
e
bF
e

ou
e
ot
E
e
u
e
0; 11
A. Dogan / Appl. Math. Modelling 26 (2002) 771783 773
where
u
e
u
1
; u
2

T
:
The element matrices are given by the integrals
D
e
jk

_
1
0
L
j
L
k
dg;
E
e
jk
v
_
1
0
L
j
L
0
k
;
and
F
e
jk

_
1
0
L
0
j
L
0
k
:
For the element x
m
; x
m1
the suces j, k take only the values 1 and 2 so that the matrices D
e
,
E
e
, F
e
are 2 2
D
e

1
6
2 1
1 2
_ _
;
E
e

1
2
v
1 1
1 1
_ _
;
F
e

1 1
1 1
_ _
;
and v given by
v
1
h
1 u
1
;
is constant over the element.
Assembling contributions from all elements leads to the following equation
D bF
ou
ot
Eu 0; 12
and u u
0
; u
1
; . . . ; u
N

T
, contains all the nodal parameters. The three assembled matrices are
tridiagonal. The general row for each matrix has the following form
D :
1
6
1; 4; 1
F : 1; 2; 1
E :
1
2
v
m1
; v
m1
v
m
; v
m
:
774 A. Dogan / Appl. Math. Modelling 26 (2002) 771783
A typical member of Eq. (12) is
o
ot
1
6
_ _
b
_
u
m1

2
3
_
2b
_
u
m

1
6
_
b
_
u
m1
_

1
2
v
m1
u
m1

1
2
v
m1
v
m
u
m

1
2
v
m
u
m1
;
13
where v
m
is given by
v
m

1
h
1 u
n
m
:
We can use a CrankNicolson approach in order to nd a numerical solution for this set of
ordinary dierential equations. Taking a time centre as t n
1
2
Dt. We can write
ou
m
ot

1
Dt
u
n1
m
u
n
m
; 14
u
m

1
2
u
n1
m
u
n
m
: 15
From here we nd the recurrence relationship
1
6
_
b
Dt
4
v
m1
_
u
n1
m1

2
3
_
2b
Dt
4
v
m1
v
m

_
u
n1
m

1
6
_
b
Dt
4
v
m
_
u
n1
m1

1
6
_
b
Dt
4
v
m1
_
u
n
m1

2
3
_
2b
Dt
4
v
m1
v
m

_
u
n
m

1
6
_
b
Dt
4
v
m
_
u
n
m1
: 16
The boundary conditions Ux
0
; t 0 and Ux
N
; t 0 require u
0
0 and u
N
0. The above set
of quasi-linear equations has a matrix which is tridiagonal in form so that a solution using the
Thomas algorithm is direct and no iterations are necessary. A linear stability analysis shows the
algorithm to be unconditionally stable since the CrankNicolson method is used implicit in spatial
derivations.
3. Test problems
The RLW equation has the solitary wave solution [1]
Ux; t 3csech
2
kx x
0
vt; 17
where
k
1
2

c
l1 c
_
;
and
v 1 c;
A. Dogan / Appl. Math. Modelling 26 (2002) 771783 775
is the wave velocity. The RLW equation has only three constants of motion given by [11]
I
1

_
1
1
U dx;
I
2

_
1
1
U
2
lU
x

2
dx;
I
3

_
1
1
U
3
3U
2
dx;
and corresponding to conservation of mass, momentum and energy. In the simulations the in-
variants I
1
, I
2
and I
3
are monitored to check the conservation of the numerical algorithm. The root
mean square and maximum error norms L
rms
and L
1
dened as
L
rms
jjU
exact
U
n
jj
2
h

N
1
jU
exact
j
_
U
n
j
j
2
_
1=2
;
and
L
1
jjU
exact
U
n
jj
1
max
j
jU
exact
j
U
n
j
j;
and are used to measure the dierence between the numerical and analytic solutions and to show
how well the numerical scheme reproduces the exact solution. The L
rms
error norm gives a measure
of the average dierence between the solutions while the L
1
error norm measures the maximum
dierence between the exact and numerical solutions.
Firstly, we consider solitary wave motion, secondly we study the interaction of two solitary
waves.
(a) In the simulations of the motion of a single solitary wave reported here we let l 1 and,
to allow comparison with earlier work [4,6,9]. Eq. (16) is taken as initial condition with range
40 6x 660, h 0:125, Dt 0:1, x
0
0 and c 0:1 so that the solitary wave has amplitude 0.3.
Analytic values for the invariants are determined from Eq. (17) as
I
1

6c
k
; I
2

12c
2
k

48kc
2
l
5
; I
3

36c
2
k
1
_

4c
5
_
:
The simulations are run to time t 20, and the L
rms
and L
1
error norms and the invariants I
1
,
I
2
, I
3
, recorded in Table 1. The prole of the solitary wave at times t 0 and 20 are compared in
Fig. 1. It is obvious, by time t 20, there to is little degradation of the wave amplitude and that
any non-physical oscillations that may have developed on the wave are too small to be observed.
The distribution of error plotted in Fig. 2 is concentrated near the wave maximum and oscillates
smoothly between 2 10
4
and 3 10
5
. Results are compared in the Table 2 with quadratic B-
spline nite elements, of length h 0:1, within a standard Galerkin approach [6], with a nite
dierence scheme based on cubic spline interpolation functions [4,6] with space step h 0:1 and
with a least squares method with linear elements [9].
For the solitary wave of amplitude 0.3 at t 20, L
1
error norm has a value of 0:198 10
3
and
the constants of motion I
1
, I
2
, I
3
change by less than 0.1 percent. The least squares algorithm leads,
776 A. Dogan / Appl. Math. Modelling 26 (2002) 771783
at t 20, to an L
1
1:755 10
3
, while the constants of motion I
1
, I
2
, I
3
change by up to 0.25
percent. In a corresponding simulation using a B-spline method with quadratic spline elements the
error norm at time t 20 is only 0:086 10
3
and the constants of motion I
1
, I
2
, I
3
vary by less
than 8 10
4
percent. The nite dierence scheme studied by Jain et al. [4] is based on cubic spline
interpolation functions. We have implemented this algorithm [6] and obtained that for a solitary
wave of amplitude 0.3 at t 20 the L
1
68 10
3
, it is also obtained that the constants of
motion I
1
, I
2
, I
3
increase from the analytic value by about 10 percent. These present results are also
compared with data obtained with PetrovGalerkin quadratic scheme [15] in Table 2. At t 20
the PetrovGalerkin quadratic scheme [15] leads to L
1
0:081 10
3
, while the constants of
motion I
2
, I
3
change by about 8 10
3
percent, I
1
changes by less than 2 10
3
percent.
These errors are considerably larger than those obtained with the present numerical method
and conservation is correspondingly poor. We see that for solitary waves of amplitude 0.3
Table 1
Single solitary wave amplitude 0.3, h 0:125, Dt 0:1, 40 6x 660 (I
1
3:9799497, I
2
0:81046249, I
3
2:579007)
Time L
rms
L
1
I
1
I
2
I
3
0 0:002 10
3
0:007 10
3
3.97993 0.810461 2.57901
4 0:116 10
3
0:054 10
3
3.98039 0.810610 2.57950
8 0:224 10
3
0:100 10
3
3.98083 0.810752 2.57996
12 0:325 10
3
0:139 10
3
3.98125 0.810884 2.58041
16 0:417 10
3
0:171 10
3
3.98165 0.811014 2.58083
20 0:511 10
3
0:198 10
3
3.98206 0.811164 2.58133
Fig. 1. Solitary wave proles at t 0 and 20.
A. Dogan / Appl. Math. Modelling 26 (2002) 771783 777
Galerkins method with linear elements is more accurate than the least squares approach with
linear elements but is less accurate algorithm than Galerkin with quadratic splines and Petrov
Galerkin quadratic, while the nite dierence scheme is the least accurate of all.
In a second numerical experiment c 0:03 so that the amplitude of the solitary wave is 0.09.
Analytic values for the constants of motion are determined as I
1
2:109407, I
2
0:127302,
I
3
0:388806. The results of this simulation are recorded in Table 3. This simulation of a solitary
wave of amplitude 0.09 leads to an L
1
0:20 10
3
, at time t 20, while the constants of
motion I
2
, I
3
change by less than 0.03 percent, I
1
changes by less than 0.1 percent.
In Table 4 it is found that simulation using a B-spline method with quadratic spline elements [6]
has an L
1
error norm of 0:432 10
3
, while the constants of motion I
2
, I
3
change by less than
8 10
4
percent, I
1
changes by about 0.12 percent. In a corresponding simulation with the least
squares algorithm [9] L
1
0:24 10
3
and while constants of motion I
1
, I
2
, I
3
change by similar
amounts to those above.
Fig. 2. The error is exact-numerical solution at t 20 for the solitary wave.
Table 2
Single solitary wave amplitude 0.3, at t 20, h 0:125, Dt 0:1, 40 6x 660
Method L
rms
L
1
I
1
I
2
I
3
Galerkin linear 0:511 10
3
0:198 10
3
3.98206 0.811164 2.58133
Galerkin quadratic [6] 0:220 10
3
0:086 10
3
3.97989 0.810467 2.57902
l.s linear [9] 4:688 10
3
1:755 10
3
3.98203 0.808650 2.57302
f.d. [4,6] cubic 196:1 10
3
67:35 10
3
4.41219 0.897342 2.85361
PetrovGalerkin
quadratic [15]
0:227 10
3
0:081 10
3
3.97986 0.810399 2.57880
778 A. Dogan / Appl. Math. Modelling 26 (2002) 771783
It is found that L
1
4 10
3
with the cubic nite dierence scheme [4] and while the constants
of motion I
1
, I
2
, I
3
increase from the analytic value by about 10 percent during the course of the
experiment. It is found that simulation using PetrovGalerkin method with quadratic B-spline
elements [15] has L
1
0:316 10
3
and while the constants of motion I
2
, I
3
change by less than
0.013 percent and I
1
changes by less than 0.1 percent. These errors are considerably larger than
those found with the present numerical method and conservation is poor. We nd that the least
squares algorithm [9] has the highest mean accuracy and also, for this smaller solitary wave, better
conservation than shown in Table 1.
Various space/time-step combinations are used in Table 5 and we nd that the most accurate
simulation is that with the choice h 0:25 combined with Dt 0:2.
As the amplitude of a solitary wave is reduced the pulse broadens and it may be necessary to
increase the solution range in order to maintain accuracy. The eect of doubling the range from
Table 5
Error norms for single solitary wave amplitude 0.09, at t 20, 40 6x 660
h Dt L
rms
L
1
0.025 0.025 35:9 10
3
10:3 10
3
0.05 0.05 3:21 10
3
1:023 10
3
0.125 0.1 0:535 10
3
0:198 10
3
0.25 0.2 0:177 10
3
0:067 10
3
0.5 0.4 0:31 10
3
0:094 10
3
1.0 0.8 1:11 10
3
0:345 10
3
4.0 0.8 6:00 10
3
1:89 10
3
Table 3
Single solitary wave amplitude 0.09, h 0:125, Dt 0:1, 40 6x 660
Time L
rms
L
1
I
1
I
2
I
3
0 0:138 10
3
0:390 10
3
2.10702 0.127302 0.388804
4 0:150 10
3
0:193 10
3
2.10840 0.127303 0.388809
8 0:283 10
3
0:142 10
3
2.10931 0.127303 0.388809
12 0:401 10
3
0:151 10
3
2.10985 0.127304 0.388812
16 0:480 10
3
0:155 10
3
2.10986 0.127305 0.388814
20 0:535 10
3
0:198 10
3
2.10906 0.127305 0.388815
Table 4
Single solitary wave amplitude 0.09, at t 20, h 0:125, Dt 0:1, 40 6x 660
Method L
rms
L
1
I
1
I
2
I
3
Galerkin linear 0:535 10
3
0:198 10
3
2.10906 0.127305 0.388815
Galerkin quadratic [6] 0:563 10
3
0:432 10
3
2.10460 0.127302 0.388803
l.s linear [9] 0:347 10
3
0:239 10
3
2.10769 0.127260 0.388677
f.d. [4,6] cubic 14:45 10
3
3:996 10
3
2.333 0.140815 0.430052
PetrovGalerkin quadratic
[15]
0:537 10
3
0:316 10
3
2.10908 0.127318 0.388854
A. Dogan / Appl. Math. Modelling 26 (2002) 771783 779
40 6x 660 to 80 6x 6120 is shown in Table 6. When L
rms
and L
1
error norms are reduced by
a factor of about 2.3, the maximum improvement in accuracy is found for space step 0.25, time
step 0.2.
In Table 7 we give the error norms and invariants for an even smaller solitary wave, 3c 0:03.
We nd excellent results with the range 80 6x 6120, h 0:25 and Dt 0:2.
Throughout the simulation both error norms remain less than 5 10
5
, while the constants of
motion I
2
and I
3
vary by less than 5 10
3
percent and I
1
changes by about 0.023 percent. The
eect of changes in the space and time steps is examined in Table 8. With the choice h 0:25 and
Dt 0:2 we nd the smallest error norms.
(b) The second problem we shall study will be the interaction of two solitary waves. As initial
condition we use [2]
Ux; t 3c
1
sech
2
k
1
x v
1
t x
1
3c
2
sech
2
k
2
x v
2
t x
2
; 18
where
k
j

1
2

c
j
l1 c
j

_
;
Table 6
Error norms for single solitary wave amplitude 0.09, at t 20, 80 6x 6120
h Dt L
rms
L
1
0.05 0.05 3:072 10
3
1:021 10
3
0.125 0.1 0:106 10
3
0:041 10
3
0.25 0.2 0:078 10
3
0:029 10
3
0.5 0.4 0:260 10
3
0:094 10
3
Table 7
Single solitary wave amplitude 0.03, h 0:25, Dt 0:2, 80 6x 6120
Time L
rms
L
1
I
1
I
2
I
3
0 0:021 10
3
0:042 10
3
1.205551 0.024167 0.072938
4 0:014 10
3
0:028 10
3
1.205685 0.024168 0.072938
8 0:012 10
3
0:019 10
3
1.205766 0.024168 0.072939
12 0:012 10
3
0:013 10
3
1.205811 0.024168 0.072939
16 0:014 10
3
0:008 10
3
1.205832 0.024168 0.072939
20 0:015 10
3
0:006 10
3
1.205834 0.024168 0.072940
Table 8
Error norms for single solitary wave amplitude 0.03, at t 20, 80 6x 6120
h Dt L
rms
L
1
0.125 0.1 0:136 10
3
0:035 10
3
0.25 0.2 0:015 10
3
0:006 10
3
0.5 0.4 0:050 10
3
0:015 10
3
780 A. Dogan / Appl. Math. Modelling 26 (2002) 771783
and
v 1 c
j
;
evaluated at t 0 produce two solitary waves. Again in these simulations we take l 1. The
one of amplitude 3c
1
sited at x x
1
and that of amplitude 3c
2
at x x
2
. An interaction occurs
when the larger is placed to the left of the smaller. We study such an interaction with c
1
0:2,
x
1
177, c
2
0:1 and x
2
147 running the simulation for a time 400 and using the range
200 6x 6400, h 0:12 and Dt 0:1. As there is no exact analytic two wave solution, the
Table 9
Invariants for interaction of two solitary waves amplitudes 0.6 and 0.3, h 0:12, Dt 0:1
Time I
1
I
2
I
3
0 9.8586 3.2449 10.7788
40 9.8642 3.2456 10.7809
80 9.8683 3.2475 10.7872
120 9.8719 3.2491 10.7928
160 9.8751 3.2506 10.7979
200 9.8786 3.2523 10.8036
240 9.8825 3.2544 10.8109
280 9.8854 3.2557 10.8156
320 9.8883 3.2569 10.8197
360 9.8907 3.2576 10.8220
400 9.8930 3.2585 10.8251
Fig. 3. Interaction proles of the solitary waves at times from t 0 until t 400.
A. Dogan / Appl. Math. Modelling 26 (2002) 771783 781
accuracy of the simulation is gauged by degree of conservation produced by the algorithm. We
nd that with the space/time step combination 0.12/0.1 the constants of motion I
1
, I
2
, I
3
show a
higher degree of conservation than with the choice 0.05/0.05.
The variation of the invariants during the simulation with h 0:12, Dt 0:1 are given in Table
9; each changes by less than 0.45 percent while Fig. 3 shows the interaction prole at times from 0
until 400 in steps of 100.
By time t 400, the larger wave has passed through the smaller to reach the point x 311:56
while the smaller has reached x 281:68. A very small wave of amplitude 0:63 10
4
has been
left behind at x 233:8. Undisturbed by an interaction, the larger wave would reach 303 and the
smaller 293 by time t 400. The interaction has caused a phase advance of dx 8:56 in the larger
wave and a phase retardation of dx 11:32 in the smaller. This observation is in qualitative
agreement with earlier numerical experiments on very much larger amplitude waves [12]. The
accuracy of these results is expected to be aected by the relatively large space and time steps used.
4. Conclusion
The RLW equation is numerically solved using the Galerkin approach with nite elements.
The numerical solution leads to an unconditionally stable algorithm which models the ampli-
tude, position and velocity of a single solitary wave of small amplitude over a extended time
scale.
It is found that solitary waves of amplitude 0.3, Galerkins method with linear elements is more
accurate than the least squares approach with linear elements but is less accurate algorithm than
Galerkin with quadratic splines.
The interaction of two solitary waves, both of small amplitude, is similarly simulated. By time
t 400 the interaction is virtually complete and the waves have emerged with, practically, their
former amplitude and velocity. Phase shifts in line with those observed by earlier workers [12] are
obtained.
Acknowledgement
The author gratefully acknowledges a grant in scope of the NATO Science Fellowship Pro-
gramme by the Scientic and Technical Research Council of Turkey (TUBITAK).
References
[1] D.H. Peregrine, Calculations of the development of an undular bore, J. Fluid Mech. 25 (1966) 321330.
[2] J.L. Bona, W.G. Pritchard, L.R. Scott, Numerical schemes for a model of nonlinear dispersive waves, J. Comp.
Phys. 60 (1985) 167196.
[3] J.C. Eilbeck, G.R. McGuire, Numerical study of the regularized long-wave equation. I. numerical methods,
J. Comp. Phys. 19 (1975) 4362.
[4] P.C. Jain, R. Shankar, T.V. Singh, Numerical solutions of regularized long wave equation, Comm. Num. Meth.
Eng. 9 (1993) 579586.
782 A. Dogan / Appl. Math. Modelling 26 (2002) 771783
[5] L.R.T. Gardner, G.A. Gardner, Solitary waves of the regularised long wave equation, J. Comp. Phys. 91 (1990)
441459.
[6] L.R.T. Gardner, G.A. Gardner, I. Dag, A B-spline nite element method for the RLW equation, Comm. Num.
Meth. Eng. 11 (1995) 5968.
[7] N. Nguyen, J. Reynen, A space-time least-square nite element scheme for advectiondiusion equations, Comp.
Meth. Appl. Mech. Eng. 42 (1984) 331342.
[8] N. Nguyen, J. Reynen, A space-time nite element approach to Burgers equation, in: C. Taylor, E. Hinton, D.R.J.
Owen, E. Onate (Eds.), Numerical Methods for Non-linear Problems, vol. 2, Pineridge, 1982, pp. 718728.
[9] L.R.T. Gardner, G.A. Gardner, A. Dogan, A least-squares nite element scheme for the RLW equation, Comm.
Num. Meth. Eng. 12 (1996) 795804.
[10] O.C. Zienkiewicz, The Finite Element Method, third ed., McGraw-Hill, London, 1977.
[11] P.J. Olver, Euler operators and conservation laws of the BBM equation, Math. Proc Camb. Phil. Soc. 85 (1979)
143159.
[12] J.L. Bona, W.G. Pritchard, L.R. Scott, Solitary wave interaction, Phys. Fluids 23 (1980) 438441.
[13] L.R.T. Gardner, G.A. Gardner, F.A. Ayoub, N.K. Amein, Simulations of the EW undular bore, Comm. Num.
Meth. Eng. 13 (1997) 583592.
[14] L.R.T. Gardner, G.A. Gardner, F.A. Ayoub, N.K. Amein, Modelling an undular bore with B-splines, Comp.
Meth. Appl. Mech. Eng. 147 (1997) 147152.
[15] A. Dogan, Numerical solution of regularized long wave equation using PetrovGalerkin method, Comm. Num.
Meth. Eng. 17 (2001) 485494.
A. Dogan / Appl. Math. Modelling 26 (2002) 771783 783