Reprinted from JOURNAL OF COMPUTATIONAL PHYSICS Vol. 55, No.

2, August 1984
All Rights Reserved by Academic Press. New York and London Printed in Belgium

Analytical and Numerical Aspects of

Certain Nonlinear Evolution Equations.
III. Numerical, Korteweg-de Vries Equation

THIAB R. TAHA* AND MARK J. ABLOWITZ

Departmem of Mathematics and Computer Science,

Clarkson College of Technology,Potsdam,Nel~' York /3676
Received August 10, 1982; revised November 15, 1983

Various numerical methods are used in order to approximate the Korteweg-de Vries
equation, namely: (i) Zabusky-Kruskal scheme, (ii) hopscotch method, (iii) a scheme due to
Goda, (iv) a proposed local scheme, (v) a proposed global scheme, (vi) a scheme suggested by
Kruskal, (vii) split step Fourier method by Tappert, (viii) an improved split step Fourier
method, and (ix) pseudospectral method by Fornberg and Whitham. Comparisons between
our proposed scheme, which is developed using notions of the inverse scattering transform,
and the other utilized schemes are obtained.

1. INTRODUCTION

The Korteweg-de Vries equation (KdV) introduced in [I] was originally derived in
order to describe the behavior of one-dimensional shallow water waves with small but
finite amplitudes. More recently, this equation also has been found to describe wave
phenomena in plasma physics [2, 3], anharmonic crystals [4, 5] bubble-liquid
mixtures 16, 7], etc. There has been great interest in this equation because of its
special properties. A substantial review of this work can be found in [8, 9]. Zabusky
and Kruskal [10] discovered the concept of solitons localized waves with special
interaction properties, while studying the results of a numerical computation
(describing an anharmonic lattice) on the KdV equation. This motivated the work of
Gardner et at. [II] and led to the explosion of both the theoretical and numerical
work which is still growing today. Many analytical results are available for equations
which exhibit exact multisoliton behavior, when an associated scattering problem can
be found. Of course there are many examples of inexact, or quasi-soliton behavior.
For these problems little or no analytical results are known and numerical studies are
essential in order to develop an understanding of the phenomena. This work aims to
compare a proposed scheme which was developed in paper I using notions of the

* Permanentaddress: Department of Computer Science,University of Georgia. Athens, Ga. 30602

231
0021-9991/84 $3.00
Copyright!C\ 1984 by Academic Press,Inc.
All rights of reproduction in any form reserved. 'I
 232 T AHA AND ABLOWITZ

inverse scattering transform (1ST) and certain other known numerical methods for the
KdV equation

u, + 6uux+ uxxx= O.
(1.1)
These results can be extended to cover many other related equations as well.
The following numerical methods are applied to the KdV equation:
1. Finite difference methods.
(a) Explicit methods.
(i) Zabusky and Kruskal scheme 110, 121.
(b) Implicit methods.
(i) Hopscotch method 1131.
(ii) A scheme due to Goda 1141.
(iii) The proposed scheme 1151.
(iv) A scheme suggested by M. Kruskal 1161.
2. Finite Fourier transform or pseudospectral methods.
(i) Split step Fourier method introduced by Tappert [171.
(ii) Pseudospectral method introduced by Fornberg and Whitham [181.
As in paper II in order to compare schemes, our approach for comparison is to (a)
fix the accuracy (L<X) for computations beginning at t = 0 and ending at t = T; (b)
leave other parameters free (e.g., At, or Ax), and compare the computing time
required to attain such accuracy for various choices of the parameters.
Various methods are applied to the KdV equation (1.1) subject to the following
conditions:

(a) The Initial Conditions

(i) I-Soliton Solution
The exact solution of (1.1) on the infinite interval is

U(x. I) = A sech2(kx -wI -'10)' (1.2) j

where

w = 4k2 . A = 2k2, '10= constant.

For initial conditions, Eq. (1.2) is used at t = 0, and different values of A are tested
and. '10 is chosen to be zero.

(ii) Collisions of Two Solitons

The exact solution of (1.1) on the infinite interval is

u(x, t) = 2(logf)xx'
(1.3)
 NONLINEAR EVOLUTION EQUATIONS, IJI

where

f=
" -k X k 3 t + ,,10)
,,/- / -/ ,,/,
and

eA;j= ki-kj ~
For initial conditions,Eq. (1.3) is used at t = 0, and two different setsof valuesof the
parametersare studied, namely,

kl = I, ,,(01-,
-,. - 0 17~O)
= 2 ft.
k2=12.
and
k. = 1, k2= J"S. "(0)- -, 0
.,. ,,~O}
= 10.73,

and the solitons are allowed to interact and return to their original shapes.

(a) The Boundary Conditions

Periodic boundary conditions on the interval [-20,201 are imposed.
The numerical solution is compared with the exact solution. In addition, two of the
conserved quantities are computed, namely; J u2dx, and J [2u] -(ux)21 dx.

2. THE REPRESENTATION OF THE KdV EQUATION USING NUMERICAL METHODS

1. Finite Difference Methods

(i) Zabusky and Kruskal Scheme
In their original work, Zabusky and Kruskal [10], used the explicit leapfrog finite
difference scheme

~t
-2u::'+, + 2u::'-1-U::'-2)' (2.1 )
-W(U::'+2
where u:;' = u(nL1x,mAt); nand m are integers. This scheme is consistent with
Eq. (1,1) and the truncation error is of order (O«At)2) + O«AX)2». The linear
stability requirement for this scheme is

iff I I I 2
AX -2Uo+~ ~~ (2.2)

I
1+e"'+e"2+e"'+"2+A'2
 234 T AHA AND ABLOWITZ

(where Uois the maximum value of u in the range of interest). This means that a very
small time step must be used to preserve stability. For the initial time step one may
use the uncentered scheme
I 0 At 0 0 0
un = Un -~ (Un+ I + Un + Un-I )(Un+
0 I -Un-I 0 )

(2.3)

(b) Implicit Methods

(i) Hopscotch Method
In 1976 Greig and Morris 1131 proposed a hopscotch scheme for the KdV
equation (1.1).
With 1= u2J2, the scheme is

m+ I m 3 At 1m 1m ) Lit ( m 2m 2 m m
Un =Un- A:;-( n+l- n-1 -""2"(A~ Un+2- Un+l+ Un-I-Un-2' ) (2 .a4 )

U m + I = U m -3
n n
~
Lix
(1m + I _ 1 m+ I
n+1 n-1
) -~:.--
2(Lix)J
( u m + I -2u
n+2
m+
n+1
I + 2u m + I -U
n-1
m + I
n-2. )

(2.4b)
To implement the scheme, we employ (2.4a) for those grid points for which
(n + m) is even and (2.4b) for those for which (n + m) is odd. A quasi-tridiagonal
system of equations must be solved at each time level. An optimization of Gaussian
elimination method is used to solve this system (see paper II, Appendix A). The
linear stability requirement for this scheme is that 113]

Lit I 1
-~

(LiX)3 (LiX)2 "0 -2

I
(2.5)

(uo is the maximum value of u in the range of interest). The truncation error of this
scheme is of order (O«Llt)2) + O«LlX)2).

(ii) A SchemeDue to Goda

This implicit scheme for approximating the KdV equation (1.1) is given by

-.
At
( um+
n
I

+ ' { um+1 2 m+I +2 U,,-I

m+1 m+l
2(Ax)3 ,,+2 -U,,+I -U"-2 } =0.
 NONLINEAR EVOLUTION EQUATIONS, III

The truncation error of this scheme is O(At) + O«AX)2). This scheme is uncon-
ditionally stable according to linear analysis. In order to apply this scheme, we have
to solve a quasi-pentagonal system of equations at each time level:

x x x x x
x x x x x
x x x x x
x x x x x
x x x x x

x x x x x
x x x x x
x x x x x

An optimization of Gaussian elimination method is used to solve this system of

equations.

(iii) The Proposed Scheme Which Is Based on the 1ST (see Paper I)
First, consider the local scheme with A (0) = (3j2)(Atj(Ax)3), which can be written
as

-U::'-I(U::'+U::'-I+U::'-2)}. (2.7
The truncation error of this schemeis O«At)2) + O«AX)2).
To implementthis scheme,Eq. (2.7) for the new time level can be written as 116]

um+I-3um+I
n+2 n+1 +( 3 + t: ) um+l_um+I=B
n n-1 n. ( 28 .

)
where

2(Lix)3
6= ~ 1
Lit

(At is supposed to be of the same order as Ax) and

B" = U::'-2-3u::'_1 -U::'+I + (3 + e) u::'

-3(Ax)2( (U::,)2-(u::'+ 1)2]-(AX)2[U::': ..(u::'+. + u::':.. + u::':1)

-u::'-.(u::' +U::'-I + u::'-J].

 236 TAHA AND ABLOWITZ

This can be solved by a version of the Crank-Nicolson back and forth sweep method
for the heat equation [20J. We seek an equation or the form

Um+1
n+1 =aum+l
n +bm+1
n , (2. 10)
which is suitable for computing u~+ I explicitly by sweeping to the right. For stability
lal ~ I. Repeated substitution into Eq. (2.8) to eliminate U;:21, u;:.., and u;+ I in
favor of u;!.. gives
b;:.. + (a -3) b~+ I +(a2 -3a + 3 + e) b;!.. + (a1 -3a2 + 3a+ ea -I) u;!..= Bn.

(2.11 )
Requiring the u:~.. ter.m to drop out determinesa (uniquely since lal ~ 1) as a
solution of
(a -1)3 + ea = 0 (2.12)

and leaves for bn (at the new time level) a second-order difference equation. The
corresponding homogeneous equation of (2.11) has a solution of the form

bn=kn (2.13)
..if the constant k satisfies
k2 + (a -3) k + a2 -3a + 3 + e = O. (2.14)
It can be shown, or verified, that the two roots k of this equation are the two roots
with Ikl > 1 other than that of the cubic equation determining a < 1 above. It follows
that b can be computed explicitly by sweeping to the left
bn-. = (3a -a2) bn -abn+ 1+ aBn (2.15)

(Eq. (2.15) is obtained from (2.11) and (2.12)). To obtain the solution un' first solve
forbn from (2.15) then use (2.10) to calculate un.
In order to implement this sweeping technique, the same iteration method used for
the sweeping technique discussed in paper II is used. The only difference is that we
have to assume initial values not just for bn but also for bn+..
Second, for the global method which can be written as (paper I, Equation (2.30)*)

-sr -I
?"+!{Y/+I N/+ I -N,+M,
/+ I

+ .
T=S":;;- E" + 1 (2.16)

Ii.
 NONLINEAR EVOLUTION EQUATIONS, III 237

where
"
E /I =A(Zlsm
-/I W/I-I -sm+1D(2)
,,- + H """+G _sm+l '5"' ( H k +G k )
"-'
k=-oo
"
+ s::' Wn-ICn-1 -S::'D~), C = A (4)+ ~ pj w-j 1,
" -;;...
j= -00
T" = Y"+IM" + S::':tY"+IZ" -S::'+IN"+I' M " =sm+1WA(4-,-smD(4)
" ,,- ,,-
" "
Z,,=(A~)+ L QjWj-I)W", N,,=. D(2)-~ + ~ Fj'
J=-oo j=-oo
n
Wn= n
1=-00
YI' YI=
( l-sm+1
I-Sj
I
) ,

H
k-
- A (41(s m
-k+IYk-
S m) W
k k-I'
G - (s
k-
m
k-
Sm+ I ) D (41
k+1 -,
F} =A(41
-} (sm+IW } _sm+lw
}-I }-I )+ D(41
-i-i (sm -S,!,+I
J )'
w + D(41 ( sm+l_ s m+l
-J (S,!,+I_sm
P} =A(41 }+I} ) -}+I } y)
},

Q} = (sm+ 1- S'!' ) W A (4) - (S'!' Y -sm} ) D(41

}-I J } -J-I} -,

A (21= -~ A (01+ ~ a D(21 = -~ A (01-~ a

-3- 2' -3- 2'

D(4)=~A(O)+~ .a, a = ~'.1t

-6 -4

A ~) = arbitrary constant t-(4.t)2 U m

and sm-
,,- -e n"

With A~) = ~ u, the same idea is applied and the only difference is in the B n term.
This proposed scheme is unconditionally stable, and ha~ a truncation error of order
(O«Ar)2 + O«AX)2».

(iv) A SchemeSuggestedby M. Kruskal

Kruskal [16] has suggested the numerical scheme

(2.17)

for

u/+uxxx=o. (2.18)
 238 TAHA AND ABLOWITZ

Kruskal did not suggest any particular numerical scheme for the nonlinear part of the
KdV equation (1.1). The following scheme is used to solve the KdV equation (1.1):

+ U':(U':+I -U':-I)] t = o.
(2.19)
To implement this scheme, the above mentioned sweeping/iteration technique is
used. Several values of e are employed alld experimentally we find that e = j gives the
bestresults.
This scheme is unconditionally stable according to linear stability, and has a trun-
cation error of order (O«Lft)2) + O«LfX)2».

2. Finite Fourier Transform or PseudospectralMethods

(i) Split StepFourier Method by F. Tappert
For convenience the spatial period was normalized to [0, 2nI, then Eq. (1.1)
becomes

(2.20)

where p is half the length of the interval of interest, and X = (x+ p) nip.
As discussed in part II of this paper the essence of the solution method is to
alternate between two steps: (1) advance the solution using only the nonlinear term
by means of a (implicit) finite difference approximation. (2) advance the solution
using only the linear term by means of the discrete fast Fourier transform (FFT).
To implement this method for the KdV equation (2.20), as the first step, one first
I
approximates,

.7r
u,+6-uux=O. (2.21)
p
A straightforward discretization is

(2.22)

~
 NONLINEAR EVOLUTION EQUATIONS, III
239

u(Xj' / + £1/)= F-1(elkJ1fJ/pJ.1JF(ii(Xj'/»), (2.23)

+ (ii2)::'!il+ 18(u2)::,+ ( U2 )m
-8
n-J -(U2)'::+2 + (U2)'::-2J}. (2.24)
Also, one may improve the truncation error to be of order (O(Lft)2 + O(Lfx)P) for all
p, see the pseudospectral method (Fornberg and Whitham) below.

(ii) Pseudospectral Method by Fornberg and Whitham [18]

As mentioned in part II of this paper, this is a Fourier method in which u(x, t) is
transformed into Fourier space with respect to x. Again for convenience the spatial
period is normalized to [0,2n]. This interval is discretized by N equidistant points,
with spacing AX= 2n/N. The function u(X, t), numerically defined only on these
points, can be transformed to the discrete Fourier space by

u(k, t) = Fu =

k=-~ 2 ,...,

The inversion formula is

u(jAX, t) = p-IU = u(k,t) e2"jjk/N

(2.26)
These transforms can be performed, efficiently with the fast Fourier transform
algorithm [21 J. With this scheme, Ux can be evaluated as F-1{ikFu}, uxxx as

~~
240 TAHA AND ABLOWITZ

F-11-ik3Fu} and so on. Combined with a leap-frog time step the KdV eql,iation
(2.20) would then be approximated by

At u(X, t) F-1(kF(u))

(2.27)

Fornberg and Whitham make a modification in the last term, however, and take

u(X, I + LlI) -U(X, I -LlI) + 2i~ LIt u(X, t) F-'(kF(u»

p

(2.28)

The difference between Eq. (2.27) and (2.28) is in the approximation of the linear
equation

(2.29)

The linear part of Eq. (2.28) is exactly satisfied for any solution of Eq. (2.29) [18 J.
Also it turns out that the linearized stability condition is less restrictive for (2.28)
than for (2.27): Jtl(JX)3 < 3/2n3 ~ 0.1520 compared to Jtl(JX)3 < I/n3 ~ 0.0323
(for details see [18 J).
Since the Fornberg and Whitham scheme is explicit, it is natural to consider
Crank-Nicolson type implicit version, e.g.,

+F~I(k3F(u(X, t»)} = O. (2.30)

This scheme (2.30) is unconditionally stable according to linear stability. We make
some remarks about (2.30) in the conclusions.

3. CONCLUSIONS

Various numerical methods are used in order to approximate the KdV equation
(1.1), namely, (i) Zabusky and Kruskal scheme (2.1), (ii) hopscotch method (2.4),
(iii) a scheme due to Goda (2.6), (iv) a proposed local scheme (2.7), (v) a proposed
global scheme (2.16), (vi) a scheme suggested by Kruskal (2.19), (vii) split step
 NONLINEAR EVOLUTION EQUATIONS, III
241
Fourier method by Tappert (2.22>-'(2.24), and (viii) pseudospectral method by
Fornberg and Whitham (2.28).
As in part II, our approach for" comparison is to (a) fix the accuracy (Loo) for
computations beginning at t = 0 and ending at t = T; (b) leave other parameters free
(e.g., At, or Ax), and compare the computing time required to attain such accuracy
for various choices of the parameters.
Two sets of initial conditions were studied: (A) .I-soliton solution with different
values of the amplitude, (B) collisions of two solitons with different values of the
parameters. According to this approach we have made the following conclusions:
(1) The scheme of Goda required a long time compared to the other «i), (ii),
(iv), (vi), (vii), (viii» schemes.
(2) Zabusky and Kruskal's scheme was good for low amplitudes, but it needed
more computing time than the other remaining methods «ii), (iv), (vi), (vii), (viii»
for high amplitudes.
(3) The calculations for the previous two methods; Goda, and
Zabusky-Kruskal, were not carried out for the I-soliton case with amplitude = 4.
They needed very long computing time.
(4) The Tappert and hopscotch schemes took less computing time than the
previous two schemes. For small amplitudes hopscotch was more efficient than
Tappert, and they behave almost the same for fairly small amplitudes. On the other
hand for relatively high amplitudes the Tappert scheme turned out to be better.
(5) The suggested scheme by Kruskal is in general faster than the previous
schemes «i), (ii), (iii), (v), (vii).
(6) The Fornberg and Whitham method is much faster than the suggested
scheme by KruskaI, it is roughly three times faster for small amplitudes and six times
faster for high amplitudes. Also (2..27) was tried, but (2.28) proved to be somewhat
faster. In addition, the implicit .version (2.30) was implemented and did not prove to
be faster than (2.28).
(7) The proposed local scheme is the best amongst all the utilized schemes. It
was roughly eight times faster than the suggested scheme by Kruskal. (See remark
below Eq. (2.18». This certainly shows that the approximation of the nonlinear term
is crucial. Also, it was roughly one and a half times faster than the Fornberg and
Whitham scheme. This despite the fact that the local scheme is only O«At)2, (Ax)2)
whereas Fornberg and Whitham method is of order O«At)2, (L1x)P)for all p. These
results suggest that "1ST generated" schemes will be good approximations for
equations which are in fact exactly solvable by the 1ST. The proposed global scheme
was implemented and proved to be faster than some of the utilized schemes, but much
slower than its local version. Since the global scheme is so complicated and
cumbersome, we are not convinced that our methods of implementation were optimal.
Optimizing the implementation of our global scheme will be under further
investigation in the near future. (The following tables and figures exhibit the results).
All the numerical calculations were inspected at every step by using the conserved
242 TAHA AND ABLOWITZ
"iU
QQ
C
:a
c 0- ,..,
LlJ "3" ~
.., -oi a-
"0
C
~
0
II
"iU
QQ 0-000 -or-N r-000- ~~V)
C r-00r- NOr- NN- 080
'c
c 888888888888
.,
"OJ,
~
000000000000
I I I I I
..
c
0
.;:: a
~ C
~ .-
-0
0.-
S~
0 C" 0-
00
8 OO~
~-o-
-ON
8 00
0--
8 00
ULlJ oOQ 0-"" 00 8 0
...;.. 880888888808
.E"O 000000000000
.,.,~
0 ., I I I I I I
O;.c
0-
V QQ
c
s:E
::=.,~ MMo- ~VlN CONN CO-~
r-CO-o M~o- r-o-r- MOO-
u
~C .- -N~ -N~ -N~ M~~
~ 000000000000
~ "0., 000000000000
000000000000
LlJ u.~
...J U::
~ <-
< c::J
I- ~~
c 0 ..., ..., ..., ...,
"'...,0 "'...,0 "'...,0 "'...,0
~~ -.,
.-.c
00-': od-': dd-': dd-':
<~
0-
-~
"0
., .-...
u ~ ~ M M
N N
.; E
"Z
C"~
u
E
-
--.. f=
~ 0 'oj" 0
...c
.c ...
~ ~ ~ ~
.~ .E
.c
~q a-
MN N
on
"'.,.
§:~ ~8 -0
0
N§ ;::;8
.. 00 00 00 00
(i
S
i=
~II -II ~II -II ~ -
II II
~ -
II II
QQ
~~ ~~ ~~ ~~ ]I
.5
~ '11
0. >.
S
""'- =
0 .,.~
U
., ""'"
= '"
-5
...
0

N 2
':'~
'u -0

.c
0
u
-t
C :: c os "'uC- 0.
.g
~
,. 00=
"0
C
0 O
:c
0.
~
:u
0.
S
0
N ..; ~
U
 NONLINEAR EVOLUTION EQUATIONS, III 243
"':
""' ~
.,., r-:
--N ...,~-C .., ,..., ,,",or-
000 --N -C-C-C -~r-
,...,...,. NNN ...,o~
888888 888 ~oo .E
000000000000 0" r£
I
0 .- LU
J.. ~
~> E
--N
:;'0008-
"'0-- 00-
080
"'N'"
N-OO
u
... .-
C ~":I
.;; ;:- r£
§§8 §8§ 888 888 .c ~.
=ioo 000 000 000 -~.. '<
I c "-""':I
0
-'<~
II ~
.- 1"-"'
~.-
c
N 0
U ~-'"
'" '-.'-.U
NMM IONM OONr- r-NM
:. """.c
00"" ~1Or- M.,.,r- .,.,10-
NM~ N~ --- Q. .c.c-
0 00
0000000
8 8 0000 88 0
0
E
LU
""LU
c
000000000000 --
.c
.-c

.c.cO
3: 3: .-
c .a
3:
c --
0 0
'"
0
0 LULU-
.-=
.~ cr
="
cr LU
'" '" '" '" '" U U ..
"""0 """0 """0 """0 c .» u
00"': 00"': 00"': 00"': 0 '< u
" ~""".c
--~~-
LU N '"
.-~ U U.-
N
roo
..,
«I
~ .~
c
.-"-"'
~--
...c
'- '-
.c
E.
~
C .I 0 0",
LU '< >.>.c
'" ~ '"'::s .~ .~ LU
0 0 0 0 LU N C
C
~_C LULUC 0
~ ~ ~ ~ 0
.~ '-.'-.
'0'- = ='=
cr cr =
15
'" U 0"",,15
U U U '"
..:. .a=»-
",a-\0 . LU~"""LU
-- --,.
> UU"
or) C > '" "'.-
CX),.,.
00
\OM
~8 3:
0 ti
LULU""
ti 0
C C
0 U
...
E
00 00 00 00
~ ~..uu=
II II II II
~- ~II -II
'<- '<- \I \I
~ uu--uU.c.cC
~~ ~~ ~~ ~~ LU .c.c'-'-.c
fo-fo-OO-
~
.-~. II U U
II = =.-'"
.~ -0 0 ~ -: E .
"Q -C C -N ..,~
., e
.. ..
C C
LU
=-_.

~ ~
."",,-

> >
.
"0
.,"'~.,E "'" .. :

-00
:.u.Q .,
crCC
UU~~'.u'.u~
'" E E
~~UUE.
E U U U ~~ = = I
0
Co"'
u
'"
0
0,-
u
'"
g.E"
",o.c """, ~~"--.
>
U
" E
-c; 0-
...OS
0 =
...~ o LI...~ '" C C ~I ~I "LU ,,-
LU'~
~
'" Cou 0,0 -g >,.c o
c U U U U ..
., 0 .c.c -N.c.c LU
~
~
~-
I-
.."00
..:
f-o ~
"'.,.Q~ "fo-fo--=--=-fo-fo-E
~ II II II II II II II
fo-,<- -N-N8
.,.; -c ,..: <Xi a ":I":I""~~~
 244 T AHA AND ABLOWITZ
;;
co
c
:0
c
UJ
"0
c
~
0
II
;;
co
c
'c
c
'0iJ
u
~ 0
N
"'
.-.
c
0
-c

0'
N
I
oS 0
~ ':;'
O-~ -.;
>
E ~
0 C" ~
uUJ .5
..> .,
.2"0
~ -5
0 u c
0- .oC 0
N
V ~
c II
°-
--:"> .,
..,
.oJ 0
~CI1
.E
()>'C .- "Co
E
-~
f"O
u =
-() () .-N -5
UJ ==
-I
~ C::> .~
-I: ~"' c
L- c"O
r- .-0 .g
~OC '6
~ ~ c
<~ 0
u
0-
-~
"0 .~ ]
..u
u ..
:5
.:; E u c
C"~ .u
~Uz
e
i= ;
--
"'oC
oC ..
() 0

U
.g

c
.-'"'"
oC
~o. ~
c
~-
Gj' II
U
.:'"
~
0
E
1= .,
...
co =
.5 '"
'5
0- :;
E c=
u
0 "co
U
~>
'--oS ...
.,
0 '" c
C
0
"'0 u
'c 0
~
0- ~
E I-
0
0
U
J
 NONLINEAR EVOLUTION EQ\)A TIONS,III 245
~
00
c
:0
~
.~
c
tlJ -a~
"! "': or)
~ 0: ~
"0 ..
e
M
M
N
N
'"
C
z
0
0
II
~
00 oor-oo-~O O~- ~No- NV)o- "-'0-
c V)V)V)OOO V)~,,-, V)V)~ ~V)N v)N8
'co
c
'OJ,
::."' 888§§§ §§§ 888 §§§ 8So
u 000000000000000000
~ I I I I I 0
N
"'
c c5
O;:
0
'"
. c
N
I
~ 0
e00. .;:'"
:I
"tU
..
>
008000 "-0000 N",..- -0--0 r--00 u
urZ 88~888 §8~ ~§§ 8§8 ~g~ .5
...> 880888 080 ~oo 808 ~o~ u
..9;"0 000000000000000000 -5
...~ I I I I I I I I I
...u
C!.c S
0- v
v~ c II
u
s:E
~ 0 NO\O
"0
~rn NOO\"2""2"«) 00\0 NNt- ON.., 8 ,,",N .~
>,=.-
u 8
N"2"",N-",
t-NN\Ot--
«)\0-
"'0-
0\t-"2"
O\Nt-
..,t-\O
,,",0- \OOt-
0\",
c.
~ O-N--N O-N 8 _- 8 -N 0-- E
cu
== ~"O
:I U 000000000 00 00000
ciciocicio cicio cicici cicici cicici -5
tlJ tl~
..J ..(.;: .~
~ c~ c
..( ","'
.g
k
...c"O
'" -
.-~
~ U
-II
or)
Nor)ONor)O
00"';00"';
or) or)
Nor)O
00"';
or)
Nor)O
00"';
or)
Nor)O
00"';
or)
Nor)O
00"';
:a
c
..(~ 8
0-
-'"
"0 .-u u ~ ""' ~ ~
U
...U
...
~
0-
M
0
M
0
N N :5
.;
,,":I
e ., c
cu
~"'uZ U
E
..., ... '"
j:: c
r-
r- '"\0 N
N
N
'" cu
c
5 ~
.c ...
°e
~ ~ ~ ~ ~ .g
.~ ..9; ~ "0
.c I '?
~o. 0
~- ~
§:II
U
N
.~ ",X ~8
OX)
""- inN

In
r- .,.,.,.,
NO
N-
-0
C
~
U ...
.:'" OM 00 00 00 00 ..,0 0
e O-C 00 dd 00 00 00 ~
~ ~ ~II -II II II II II
'0:- II II
10:- II II
'< - II II
'< - ..
u
.:
00 ~~ ~~ ~~ ""1""1 ~~ ~~
cu
'"
~0.
.,
"1:)
c
:~
c
e .u
cu
"U .,E
OJ
0 .u ..
-co ~
u "tj E b ., c,-
U
"8
.c .c
OJ
"'.c
OJ
.,.,.c
0 .,u
u.c
., c "0
0 t.)
Co- u
'--50 ~
;)
u
u
'0 t:
., "tU
.,.
Clo'"
0-
..to
O
0
CO
Co
~
..E
0 .u ..
>
"
'"
'"'" .,- "1:)~0%:
c C-
0.
0.
Clot.)
OJ 0
.c
Coo
CO
=' >..~ ~
o .c- .,.co%: 0
.~
0
:I: ~00
~ t- ~ ~ ~ (.)
0
;0.
u
e0 z
() N M ,.f' vi -0 .~
f-o
 246 T AHA AND ABLOWITZ
-=
OJ) ""
c "
'6 .~
"I: V? ~ ~
c
~ ~ or) v 'D
~toJ ..,.
-0
c 5
= Z
0
II
-=
c
OJ) N~O~M
~ ~ -M M
-N-~- 8
--0 ~ N
-N-~- O'
--0 ~ N
N~O~~
0 0 -0 -
'c
c
.aiI
88888 8§888 888§8 §§8§§
u 00000000000000000000
~ I I I I I I I I
~
0
'Z 0
= c
"0.-
-0 .-
E =
0 "
00000 ~N-~N ~N-~N 00000
ul'li 0000
0000
0 0
8 ONV~OO
000
00 8
ONv~OO
000000000000000000
0000
8
..;.. 0880000880000000080
,B-o 00000000000000000000
"'~ I I I I I I I I I I
8 OJ
0- ..c
v OJ)
c
si
,..J 0
-.-CI) ~~~-~ N~-O~ N~~-N -~O~~
u
>.c.- 88888888880888888888
0
V-~~~
O--N-
MO~O~
8MOv~v ~~~N~
;.. E-o -.J
-"U
,., u.-
N 00000000000000000000
...u-
..J -('Z
~
-( c~ '"
=
E- c"8
=-
.-.c
~ u
(...
II
-~ooo
oo"':r-i..;
-~ooo
oo"':r-i..;
-~ooo
oo"':r-i..;
-~ooo
oo...:r-i..;
-(~
0-
-=
-0 .-U
U ..
u 0- ~ ,., N
..u '"
OJ
V ~ N
.; E u
0-"
uZ
E
or) 10
~ i=
--
",.c
.c ..
u c
Os
t;,j
..,.
~ ~ ~
.~ ,B
.c
~o. ."
M ..
N

'"
'"
0
on
0
N
\0."
~~ "in
.c
~8 -8
cici
~8 ~8
U
E
.,
.. 00 00 00
i= ~ II II
'< - ~II -II II II
'< - II II
~-
OJ) ~~ ~~ ~~ "J"J
.S
='
0. >-
E .¥~
WI-
::I .,
0
u "'6
.r..¥
., WI
-5
u ~
u ~..
N
-~
::I ~
u
...
0 ~ .u "0 .u
'0
u
'" t
:= c Co
~
c
~
..
Q.., "0
0
0
0.
0
:I:
Co
tV
t-
'c
=
0.
E
0 z
0 N ~ ~
u
l
~
 EQUATIONS, III 247
U
ou
~
~
on
r-: '"! .5
g
~
..~
.S?
-;u
..
:u
~ON-M ~~N~M OO~~- ~-~MN
~N~~~ -O~-N ~~~N~ -~~M~ ..>.
~~~8~
~~~O~ --o~~
888~~ 8°~o~
o8~8~ ~~~~N
o88~8 -5
~
00000000000000000000 C
I I I I ou
>.
U
.~
..
U
0-
'"
0--°"""" 00 -""-°0- 00-'-V)~ ~
888°88°0""" 8888'" 8~"'8-
88888 8888§ 88888 88§88 ~
C
0':;000 00000 00000 00000 OU
I I I I I I I I I -IN
'"
..
~
.~
-c.
-oo~~~ OM~OOOO ~-O~~ ~-~-~ E
OU
NOOM~~ OO-MM~ N~OO-O NO~OO
80888888888008888888
08--- 0 °88-N -5
.~
00000000000000000000 C
.g
'i5
C
0
-~ooo -~ooo -~ooo -~ooo u
OO~NM OO~NM OO~NM OO~NM
~
:5
'0
'oj" ~ ;::;
~
N
C
ou
'"ou
0 ~ 0 ~
.?;
~ '"i ~ !OJ
"0
'"
0
~
l-
~
~-...,
00
..,. 0--0
00
N-
""'i-
\00
t:
~
0
00 00 00 00
II II
~
II II
~ - II II
'<- '< -
II II
'<- ~
ou
"'q"'q ~~ ~~ ~~ '"
""C ..
.. .. :g -'-
OJ
Co
E ~
u
E
E
"0 .. -co
..'" ~ .
tUN
.c
OJ
u u
"'~
...c
'" u
~
u .0
"
E ~N
C"O
"'
u 0 '" 0
0.-
'" "c.. ..I
>-
0-
0.'"-
0.'"
-;; 0 LV .nO-=
0 u. .- "'-
0 ""~ ~ '"..OU >
..-
..tU
.¥
"'=' 0.", 0.0 >-::> C ...
~ ~-
U
f-
0
.caD
".0;0-
.n
Cl.
0 ..
U -
0.- C
~ ..
1--5C
...; -0 ,..: oci Q
0
248 T AHA AND ABLOW.ITZ
co
co "0
c OJ
:0c .~
-= ~ -0 .,.,
~ ci r.:
"0
C
..
0
E -0 N
0\
= Z
0
II
co
co 00""--0"" --V)-q", OV)r-MM
c
~=8S?;~ OOOOOr-o- -N-qO-
"c
c
"OJ,
88888
00000
§ 8 8 8 8 8,8 § 8 §
u 0000000000
~ I I I I I
c
"'
0
";:; 'c
~ 0
:0 ";:;
0.=
E :0
0 0- 009;00 --"'...,N O-"-""'D
U~ 00~08
..>
0"0
88080
00000
88"-'88888°8
8080880880
=~
N
00000 0000000000
I I
0 u
0-..:
V CO
c
B'>
~o
~C/) r-V)N-.rO \Ca-a-\CM NOa-N\C
>'c r-V)-.r..,N a-V-CX)- MNV)OCX)
() .- IJ -""~r-N -\Ca-V)O NMCX)CX)O
>
e"O
:oU
~ OOO-N
00000
OOO-N
0000000000
--O-N
() .~ 00000 0000000000
~ :;::
oJ c:>
~
-< ="'
f-- .:"8
--
=.:
-U
Eo.
-II
-~NO","
OO";r-ir-i
-~"'O""
00"':"';"';
-~"'O.."
00"':"';"';
-<~
0-
-= ()
"0
U .-
.. ..
u or)
" .
E u 0- ~
..U "' O- ;>.
= N '01"
';; E
0-:0
., ~OJ)~
...0 ~
C U"O '"
"
E
~Uz
"' U 1= c
v
r-
OU"C
_=,,0 N
~
--5 Os
~ ..."
;>."0
"...
C"..
0 -"" =
,.,
.: .. > .-C
'" ~ '" U
~ ~ ~
.:.~ .E. " CT.- -;;
~~ N o¥""...
=
~N .,
N
00 ~~~~
"O~~~
OC"~
V)-
NO
0- or)
N
~~ °;n -08 .L:"- ...~ r- 0
\ON
.c 00 ..,S.E° 0;0; ~8
.,'"
U
E 00 E'iU~C:- 00 00
t:: ~ ~II -II "~ 0 ~;
.L:~~U
II II
~-
II II
~-
CO
.:
"1"1 ~ ~~ ~~
:;
0. >.
E
0
..-~
"'-
~ ~
u -g .0..-
~ '"
U .c
~..
N ~ ~
'--5 U u
0 ~ -~
.0 ~
.. '8 t
..
c
0
::
..Q.~
c
""
0
~
0
0.
0.
.u
.~ UJ 0 X I-
~
0.
E ci
0 Z
U
 NONLINEAR EVOL UTI ON EQUATIONS, III 249
.9;
]
.., ~
0 0 .9.
.., ~
OJ
~
>,
OJ
-5
""c
000--0 ~~~M- OO-MOO~ ~~OM- .u
~~g~~ -N-~~ ~~N~O -~~~-
>.
§§g§§ 8g§§g §§§§§ §§§g~ -0;
00000000000000000000 .~
I I I I uOJ
0.'
'"
~
N
---
or)
8-~0~ -~~~~ -NOOMM ~OOMOON ""
c
.u
~~o8~ 8~~~§ 888§§ 8§§8~ N
00800880000000000088 ::;.
00000000000000000000
I I I I I I I I I I '"
""OJ
.~
"Q.
E
.u
-'CNvO\ O\-MMN NCX)NON V-M-on
-5
NO\VM-
o 8
N'CCX)onO
0-N

NNN-O
88
NMO'Con
---o
8 v'C'C8
0\~'C~
-N
8 onO\CX)-O
V~~M~
888 -- .~
0000 0000000 00
c
00000000000000000000
.g
:0
c
0
u
-\DMO~
OO"':NN
-\DMO~
OO"':NN
-\DMO","
OO"':NN
-\DMO","
OO"':NN ~
:5
\0
V)
- 0
or) ~
c
.u
'"
.u
'"
- .,.,
M .g
-0
c
'"
~ ~ ~ ~ 0
~
I-
vO
or,
v ...,...,
,
..,-""'
""'
or)
N"i"
C
~
00 00 00 ;;;8 .c
0
00 00 00 00 '"
II II OJ-=-
~ ""
- ~-
II II II II
"I - ~ -
11 II ..0
.uN
~~ ~~ ~~ "'0N
.!!
""c --
.~ I
OJ ~ c
.u-
u u ~ tV
~ ..
.c
E
OJ "tJ E
u
"'.c
u
"0
u
"'0.:
u
E
-00
b Q)
u.ro E
a>
""-
..
OJ
~ 0 u
0.'"
0 u
o."Q) c ~
> .-C
OJ
-;;
-'"
0-
..m 0-
0.'"
..'"
'" 0 .c
""0 II..~.c
..OJ
~.c
o.u 0..0 " >.> c -
~
" .c-
u 0 u 0
'"Q).ro~ 0 c
~ 1--
0.:-
I- CO
~
u
0
0
.
~ U
...:
I- ~
...; \0 cO ~ OJ
.5
-
I
~
 "
~
~

1. labusky and K..uskal.

1. Goda
3. Hopscotch. .
4 Tappe..t.
5. K..uskal.
6. The p..opoSed local scheme.
7. The p"oposed global scheme
8. Fo..nbe..g and WOitOam

-' Pi' , P6 ~, Method.

FIG. 1. Displays the computing time (E) which is required by each utilized method given in Table I
I-soliton, amplitude = I.

18
17
-: 16
~ 15 ~~
~
;= 14

13

11

I. Zabuskyand Kruskal.
l. Goda.
3. Hopscotch.
4. Tappert.
5. Kruskal.
6. The proposed local scherne.
7. The proposed 910bal scherne
B. Fornberg and Whitham.

.,..678 -
Method

FIG. 2. Displays the computing time (E) which is required by each utilized method given in
Table II.
I-soliton. amplitude = 2.
250
 ,;
..~
coo CM
.~

!.
~
~

1. Hopscotch.
Z. Tappert.
3. "ruskal.
4. The proposed local schene.
S. The proposed 91obal schene
6. Fornber9 and Whitham.

1 2 3 4 5 6 Metood

'I unit in time" 2 min.

FIG. 3. Displays the computing time (E) which is required by each utilized method given in
TableIll.
-soliton, amplitude = 4.

c c~
~ .~

I. Zabus'y and 'rus'al.

Z. Goda.
3. Hopscotch.
4. Tappert.
5. Krus'al.
6. The proposed local scheme
7. The proposed global scheme
8. Fornberg and Whitha..

I Z 3 4 5 6 7 8 Met"'d

FIG. 4, Displays the computing time (E) which is required by each utilized method given in Table
IV ..
Two solitons with amplitudes 0.5 and 1
251
 252 TAHA AND ABLOWITZ

.~ j
.;
e m
c
~ 0

c ,
~ ~I
!:

1 Z.busky .nd Krusk.l.

1. God.
J Hopscotch.
4. T.pperL
5 Krusk.l.
6. The proposed 10c.1 schBne
7. The proposed glob.1 schBne
8. Fornberg and Whith.m.

I Z 3 4 5 6 7 8 Method

*1 unit in time E Z min.

FIG. 5. Displays the computing time (E) which is required by each utilized method given in
Table V.
Two solitons with amplitudes 0.5 and 2.5.

quantities f U2 dx, and f (2uJ -(Ux)2) dx. (Table I-V). The two conserved quantities
were calculated by means of Simpson's rule. In the finite difference schemes we have
discritized Ux using a central difference approximation. In the Fourier methods the
derivatives are calculated using Fourier method. The proposed global scheme is the
only utilized scheme which has an infinite number of conserved quantities, and true
soliton solutions. It is worth mentioning that we calculate the L2 error norm and find
it reflects the same conclusions as the Loo norm.

ACKNOWLEDGMENTS

This researchwas partially supported by the Air Force Office of Scientific Research,Grant 78-3674-
C, the Office of Naval ResearchGrant NOOOI4-76-C~867 and the National ScienceFoundation Grant
MCS79-03498-01. We are grateful for useful conversationswith ProfessorM. D. Kruskal and Dr. H.
Segur.

I
 NONLINEAR EVOLUTION EQUATIONS, III 253

REFERENCES

I. D. J. KORTEWEGAND G. DEVRIES,On the Change of Form of Long Waves Advancing in a

Rectangular Canal, and on a New Type of Long Stationary Wave, Phi/os. Mag. 39 (1895),
422-443.
2. C. S. GARDNERAND G. K. MORIKAWA,"Similarity in the Asymptotic Behavior of Collision-Free
Hydromagnetic Waves and Water Waves;' New York Univ., Courant Inst. Math. Sci" Res. Report
NYO-9082, 1960.
3. H. WASHIMIANDT. T ANIUTI,Propagation of ion-acoustic solitary waves of small amplitudes,Phys.
Rev. Lett. 17 (1966),996-998.
4. (a) N. ZABUSKY,"A Synergetic Approach to Problems of Nonlinear Wave. Propagation and
Interaction in Nonlinear Partial Differential Equations" (W. F, Ames, Ed.), pp. 223-258, New
York, 1967. (b) N. ZABUSKY,Computational synergetics and mathematical innovation, J. Compo
Phys. 43 (1981),195-249.
5. M. D. KRUSKAL, Asymptotology in numerical computation: progress and plans on the
Fermi-Pasta-Ulam problem, in, "Proceedings, IBM Scientific Computing Symposium on Large-
Scale Problems in Physics;' 1965.
6. L. VAN WUNGAARDEN, "Linear and Non-Linear Dispersion of Pressure Pulses in Liquid-Bubble
Mixtures," 6th Symposium on Naval Hydrodynamics, Washington, D. C., 1966.
7. L. VANWUNGAARDEN, On the equations of motion for mixtures of liquid and gas bubbles,J. F/uid
Mech. 33 (1968),465-474.
8. V. E. ZAKHAROV,S. V. MANAKOV,S. P. NoVIKov, ANDS. P. PITAYEVSKY,
"Theory of Solitons,"
Moscow, Nauk, 1980.
9. M. ABLOWITZAND H. SEGUR,Solitons and the inverse scattering transform, SIAM, Philadelphia,
1981.
10. N. J. ZABUSKYANDM. D. KRUSKAL,Interaction of "solitons" in a collisionless plasma and the
recurrenceof initial states,Phys. Rev. Lett. IS (1965), 240-243.
II. C. GARDNER, J. GREENE,M. KRUSKAL,ANDR. MIURA,Phys. Rev. Lett. 19 (1967), 1095.
12. A. C. VLIEGENTHART, J. Engrg. Math. S (1971), 137.
13. I. S. GREIGANDJ. LL. MORRIS,J. CompoPhys. 20 (1976), 60-84.
14. K. GODA, On instability of some finite difference schemesfor the Korteweg-deVries equation, J.
Phys. Soc. Japan I (1975).
15. T. TAHA, Ph.D. thesis,-Clarkson College, Potsdam, New York, 1982.
16. M. D. KRUSKAL,private communication, 1981.
17. F. TAPPERT,Lect. App/. Math. Am. Math. Soc. IS (1974),215-216.
18. B. FORNBERG ANDG. B. WHITHAM,Phi/. Trans. Roy. Soc. 289 (1978), 373.
19. B. FORNBERG, Math. Compo27 (1973),45-57.
10. R. D. RICHTMYERANDK. W. MORTON,"Difference Methods for Initial Value Problems;' 2nd ed.
Wiley-lnterscience, New York, 1967.
21. J. W. COOLEY,P. A. W. LEWIS,ANDP. D. WELCH,IEEE Trans. Educ. E-/2 I (1969),27-34.
22. G. D. SMITH,"Numerical Solution of Partial Differential Equations," Oxford Univ. Press, New
York,1965.

Printed by the St. Catherine Press Ltd., Tempelhof 41, Bruges,Belgium