Sie sind auf Seite 1von 63

SIAM REVIEW Vol. 14, No.

4, October 1972

WEAK NONLINEAR DISPERSIVE WAVES: A DISCUSSION CENTERED AROUND THE KORTEWEGmDE VRIES EQUATION*
A.

JEFFREY?

AND

T.

KAKUTANI

Abstract. This article is concerned with wave propagation processes where some balance occurs in the competition between a nonlinear effect and a higher order derivative effect which might be of a dispersive or a dissipative nature. The Burgers equation and the Korteweg-de Vries equation, which are prototype scalar nonlinear dissipative and dispersive equations, are shown to be fundamental to this study, even when quite general systems of equations are involved. The role of the solitary wave solution is shown to be central to the study which is applied to gravity waves, plasma waves and to waves in lattices. Both steady state solutions and initial value problems are reviewed together with questions of stability, existence and uniqueness.

CONTENTS
1. Introduction 2. Physical examples 2.1. Gravity waves 2.2. Plasma waves

(a) Hydromagnetic waves (magneto-acoustic waves and Alfv6n waves) (b) Ion-acoustic waves
2.3. Lattice waves 2.4. A general reductive method 3. Mathematical properties 3.1. Steady solutions (a) The Korteweg-de Vries equation (b) The modified Korteweg-de Vries equation (c) The Korteweg-de Vries-Burgers equation 3.2. Stability of steady solutions (a) The Burgers shock wave (b) The Korteweg-de Vries solitary wave 3.3. Initial value problems (a) The single solitary wave (b) The interaction of two solitary waves (c) The sinusoidal function (d) Gaussian distribution (e) Amplitudes of the emergent solitons 3.4. Existence and uniqueness of the solution

582 585 585 590 590 601 603 608 614 614 614 616 619 622 622 624 625 626 626 627 629 631 636

1. Introduction. What does "dispersion" mean? A primitive concept of dispersion seems to have its origin in the dispersion of light waves, by which familiar phenomena such as the seven rainbow colors are explained. Starting from such familiar ideas, physicists and mathematicians have extracted essential features of the phenomena and established a precise definition of dispersion for general linear waves including acoustic waves, water waves, elastic waves, electromagnetic waves, plasma waves and so on [45].
Received by the editors May 3, 1971, and in revised form January 6, 1972. Department of Engineering Mathematics, The University, Newcastle upon Tyne, England. Department of Engineering Mathematics, The University, Newcastle upon Tyne, England. Now at Department of Mechanical Engineering, Osaka University, Osaka, Japan. The work of this author was supported by the British Science Research Council.

"

582

WEAK NONLINEAR DISPERSIVE WAVES

583

Because of the linearity of the governing equations, the field quantities associated with linear waves may be resolved into Fourier components. Let us consider, for simplicity, one-dimensional plane waves. In this case, the governing equations have elementary solutions of the form

(1.1)

a exp [i(kx

cot)],

where x denotes a one-dimensional space coordinate, the time, a the amplitude, k the wave number, and co the angular frequency. The wavelength 2 is defined to be 2 2rc/k. A consistency condition that the solution should not be trivial leads to a relationship between co and k of the form

(1.2)

o(k)

which is known as the dispersion relation. The general solution is then given in terms of Fourier integrals of the form

(1.3)

fo
Vp

A(k) exp [i(kx

cot)] dk,

where the spectrum function A(k) is determined by appropriate initial or boundary conditions. We define the phase velocity Vp and the group velocity V of the wave by the following relations [44]:

(1.4)

co/k and Vg

c3co/ck.

It will be found from the dispersion relation (1.2) that the phase velocity Vp and the group velocity Vg are generally not equal to each other. In other words, the phase velocity Vp depends upon the wave number. If this is the case, we say that the wave is dispersive, or that the wave has dispersion "in a wider sense." In dispersive systems, the dispersion relation (1.2) generally gives complex co for a real k. Therefore not only the phase velocity depends upon the wavelength (or the wave number), but also the effective amplitude of the wave will be attenuated with time if Im (co) < 0. We call such attenuation dissipation. On the other hand, if Im (co) > 0, the effective amplitude of the wave will grow without bound in the course of time. We call this instability. If, in particular, co co(k) is a real function of a real k, where cVp/ck O, then neither dissipation nor instability occurs. We call this pure dispersion or dispersion "in the narrower sense." Thus the meaning of dispersion is welldefined for linear systems. In nonlinear wave systems, on the other hand, the frequency of the wave is not only a function of the wave number but also of other parameters, such as the amplitude, which are all assumed to be small in the linear approximation. Unlike the case of linear systems, it seems that a precise definition of dispersion has not yet been established for nonlinear systems. In many cases, however, we can obtain corresponding equations of a linear dispersive system by linearizing the governing equations which are originally nonlinear. In these cases, we can
Extensive discussions on the growth and damping of waves are given in [60], [62].

584

a. JEFFREY

AND T. KAKUTANI

consider the dispersion of nonlinear waves. We shall say that the system is dispersive if its linearized form exhibits dispersion in the sense of linear waves. Simple nonlinear waves in a continuous medium are usually governed by hyperbolic differential equations of the form [26]:

(1.5)

8U

t- A(x

U)-x + B(x

8U

U)=O

where U is an unknown column vector with n components U l,U2, .., u,; A(x, t;U) and B(x, t;U) are, respectively, an n x n matrix and a column vector with n components, which are assumed to be differentiable with respect to their arguments x, and U as many times as necessary. All eigenvalues of the matrix A are assumed to be real and distinct so that (1.5) is hyperbolic. Solutions of this equation will generally steepen their shape with time in the (x, t)-plane and will eventually become multivalued at some critical time tc even if we give arbitrarily smooth initial data. It is well known that this will even occur when the initial data is analytic [23]. Actual physical systems, however, are often described by equations containing higher order space derivatives of U with a small (usually the case) coefficient, in addition to the terms in (1.5). In these systems, once steepening of waves occurs, then the higher order derivative terms begin to play a role, even when multiplied by a very small coefficient. This suggests a possibility of developing an asymptotic method somewhat similar to the boundary layer theory for viscous fluids. The higher order derivative terms usually act as a check or balancing effect against the nonlinear steepening of waves, resulting in an eventual steady state. Typical examples of such steady states are the shock waves in dissipative media or solitary waves in purely dispersive media. The simplest equation to describe such a system is probably the Burgers equation for dissipative systems or the Korteweg-de Vries equation for purely dispersive ones. This review is devoted to a study of the Korteweg-de Vries equation and its generalized forms. 2 First of all, we shall see, in 2, how various physical systems can be described by the Korteweg-de Vries equation. It is remarkable that there are so many examples, ranging from classical water waves and lattice waves to plasma waves, which are governed by this simple nonlinear equation in an asymptotic sense [5], [13], [15], [19], [32], [35], [39], [40], [53], [66], [69]. This motivated and encouraged us to write this paper. Among wide classes of general dispersive systems, when the linearized dispersion relation has the form

(1.6)

co

ak

bk 3

for small values of k, where a and b are real constants, then the original system of nonlinear equations can often be reduced asymptotically to the Kortewegde Vries equation if one considers small but finite amplitude waves. Therefore it is instructive to investigate first the dispersion relation in the linearized limit. It should be noted, however, that this is not always the case as will be seen in the
A counterpart for the Burgers equation may be seen in [9], [20].

WEAK NONLINEAR DISPERSIVE WAVES

585

case of the nonlinear Alfv6n wave in a collisionless plasma [31 ], which is described by a modified K orteweg-de Vries equation. There are also a number of examples which are governed by modified or generalized forms of the Korteweg-de Vries

equation [3], [27], [30], [31], [36], [54], [69]. Precise conditions under which general dispersive systems of equations can be reduced to the Korteweg-de Vries equation have not yet been established. An heuristic approach, however, has been made by Taniuti and Wei [63] and by Gardner and Su [17, p. 329], who have developed an asymptotic perturbation method to reduce a certain class of nonlinear equations to the Korteweg-de Vries equation or to one of its generalized forms. Taniuti-Weis method is also applicable to dissipative systems, where the original system of equations is reducible to the Burgers equation or to one of its generalized forms. Section 3 deals with the mathematical properties of the K orteweg-de Vries equation and its solutions. We begin with the steady state solutions, of which the familiar solitary wave solution is a special example. The discussion is extended to steady state solutions of the modified Korteweg-de Vries equation and to those of the Korteweg-de Vries-Burgers equation. We then investigate the stability of such steady solutions [25]. Some typical results for initial value problems obtained by numerical computation are given [6], [69], [71]. It is found that for suitable smooth initial data, the solution steepens its shape and eventually splits into a number of solitary waves. A remarkable fact that will be elaborated on later is that each solitary wave behaves as if it were an independent particle. On account of this, solitary waves are often called "solitons." We should mention here that an analytical approach to initial value problems has also been made in a series of papers entitled The Korteweg-de Vries equation and generalization which are now appearing in the Journal of Mathematical Physics [41], [50], [51], [61]. We give only a brief review of this series of papers, because a number are still in press at the time of writing and they will play a complementary role to this paper. We conclude the section with a proof of existence and uniqueness of the solution under appropriate initial and boundary conditions.
2.1. Gravity waves. Let us first consider the gravity waves which take place on the free surface of a two-dimensional horizontal layer of fluid having a finite depth h o (Fig. 2.1.1). We assume that the fluid is inviscid and incompressible and that the motion is irrotational. This is in fact just the problem from which the K orteweg-de Vries equation was first derived at the end of the last century [40]. The motion of such a fluid may be described by the harmonic equation for

the velocity potential qS(x, y, t):

(2.1.1)

c--x#/cy

0.

The boundary conditions relevant to the present problem may be given in the form

(2.1.2) (2.1.3a)
Oq5 Oy

at y

O,
aty=h,

?h

4
,Oxcx

Ot

586
Y

A. JEFFREY AND T. KAKUTANI

.Y= h(x,t)

"
c3t

II//IIIIIIIIIIII/IL INIIIIIII/ x

FIG. 2.1.1. Geometrical configuration for gravity waves

+-}

+ G(h

ho)=0 aty=h,

where y h(x, t) represents the free surface. All quantities are normalized with respect to a characteristic length Lo and a characteristic speed U o. The nonis known as Froude dimensional parameter G is defined by G gLo/U2o number), g being the acceleration due to gravity. The conditions (2.1.2) and (2.1.3a) are obvious; while condition (2.1.3b), which expresses the continuity of pressure at the free surface, arises from Bernoullis theorem. Since we shall consider waves of long wavelength the effect of surface tension will be ignored. As was mentioned in the Introduction, it is instructive to investigate the appropriate linearized dispersion relation before proceeding to a nonlinear problem. The linearized version of the basic equation (2.1.1) together with the boundary conditions (2.1.2) and (2.1.3a), (2.1.3b) can, respectively, be written as follows:

(1/x//-

2+2
(2.1.2)

0,

=0 y
2@ +G t

aty=0,
=0 aty=ho.

(2.1.3)

Condition (2.1.3) has been obtained by eliminating h between (2.1.3a) and (2.1.3b). It is also to be noted that the conditions at the free surface y h can be replaced by those at the undisturbed surface y ho within the framework of a linear approximation. Assuming a sinusoidal wave of the form

(y)exp [i(kx t)], we obtain for the general solution (y), (y) A cosh ky + B sinh ky,
(2.1.4)

(x, y, t)

where A and B are arbitrary constants. Application of the boundary conditions which is (2.1.2) and (2.1.3) leads to an eigenvalue problem involving k and

WEAK NONLINEAR DISPERSIVE WAVES

587

expressed as the following dispersion relation"


2 (D Gk tanh kho. (2.1.5) It is obvious from this relation that the phase velocity Vp og/k and the group velocity V Ooo/Ok are not equal to each other except in the limit as k--, 0. We are interested here in weakly dispersive waves and in weak nonlinearity.

Therefore we consider waves for which k << 1, thereby implying waves of long wavelength. In this case the phase velocity can be expanded as a power series in /2 to give

(2.1.5) 3

Vp

--6-

which is of the form mentioned in the Introduction (cf. (1.6)). Now let us proceed to the nonlinear problem. According to the linear dispersion relation (2.1.5), the phasor of the wave takes the form

(2.1.6)

kx

ot

k(x

Vot )+

Vohg
6

k3t,

k << 1.

Introducing a small parameter e O(k 2) which measures the weakness of dispersion, we shall stretch our coordinates as follows"

(2.1.7)

l/2(X- Vot),

27

e3/2t, y

y.

This may be interpreted as a "semicharacteristic" coordinate transformation, and it is convenient for the description of progressive waves. In terms of the new coordinates (, y, 27), the basic equation (2.1.1) and the boundary conditions (2.1.2) and (2.1.3a), (2.1.3b) may be written, respectively, in the following forms"

(2.1.8) (2.1.9)
(2.1.10a)

e-5 +

20

020

0,

=0 Oy
Oy

aty=0,

e2

Oh

Vo
+

aty

h,

(2.1 10b) e2

+e.G(h-ho)=0 aty= h,

where Since, on the other hand, we consider weakly nonlinear waves, we expand the dependent variables around the undisturbed uniform state as a power series in terms of the same parameter e"

(2.1.11)

h(, 27) h o + ah(1)(, 27) + 32h(2)(, 27) + , O(, Y, 27) /3/(1)(, Y, 27) + eZO(2)(, Y, 27) + "",
Vp

The other root

Vo{1

(hg/6)k + O(k4)} leads to essentially the same results.

588

A. JEFFREY AND T. KAKUTANI

thereby implying that the small parameter e is also to be regarded as a measure of weakness of the nonlinearity. Thus we arrive at a situation in which the nonlinear effect is balanced by the dispersive effect in the above sense through the scale parameter e. This situation is somewhat similar to that in the case of Prandtls boundary layer theory, in which an appropriate coordinate stretching and expansion of the dependent variables is introduced in order to balance the nonlinear effect by the viscous (dissipative) effect. Substituting the series (2.1.11) into (2.1.8) and equating terms with the same powers of e, we obtain

2j(1)
(2.1.12)

cy z
Oy2

--0,

102

0,

__> 2.

Similarly, the boundary condition (2.1.9) gives

(2.1.13)

=o Integrating (2.1.12) with respect to y, subject to the condition (2.1.13), we find

(2.1 14)

0 (2)
0(3)

where a(, z) and b(, z) are arbitrary functions of (, z) resulting from the integration. The first line of (2.1.14) shows, in the lowest order of perturbation, that the horizontal velocity component does not depend on the vertical coordinate and that the vertical velocity component is zero. This fact means, by virtue of Bernoullis theorem, that the pressure distribution can be approximated by hydrostatic arguments. This is one of the fundamental assumptions employed in the usual long wave approximation [59]. Noting that

-ly
2

cy

n>l.

(20(1) a(, ), + O2 (;34 (1) 1 y232a y4 + b(, z), 2

0(, h, z)= 0(, ho, z) +

y ,=ho(h

ho) +

c32q
y=ho

(h- ho) 2 q-

and using the relations (2.1.14), we obtain from the boundary conditions

(2.1.10a), (2.1.10b),

cy
(2.1.15a)

y=ho

(0

(3)

h(X)

20(2)1

Oh (1)

63h (2)

100 (1) Oh (1)

WEAK NONLINEAR DISPERSIVE WAVES

589

(2.1.15b)

21 a

Gh 2)

O,

Using the relations (2.1.14) and eliminating the higher order terms in (2.1.15a), (2.1.15b), we finally obtain a single equation for

(2.1.16)

Ohl)c3z

0 )3V q- h(

63h(1)

2h 0

c3

Vohg 633h(1) 6 63

This is the equation known as the Korteweg-de Vries equation. It should be noted that the coefficient of the right-hand side member (which we call the dispersive term) of this equation is just the same as the coefficient of k 2 in the linear dispersion relation (2.1.5). A similar equation is valid for ll//(1)/ which corresponds to the x-component of the perturbation velocity. Once the lowest order solution h 1) and () have been found, it is a straightforward matter to proceed to a higher order approximation. It should be remarked here that (2.1.16) thus derived does not necessarily represent all aspects of the original system of equations (2.1.1), (2.1.2) and (2.1.3a), (2.1.3b). The most obvious restriction is that (2.1.16) can describe progressive waves propagating in one direction only, whereas this was not true of the original problem. That this is so is mainly due to the semicharacteristic coordinate stretching that has been employed and to the use of asymptotic expansions for the dependent variables. In general an asymptotic solution cannot characterize the totality of properties belonging to the original problem, and the consequence of such an approach in this case is that it leads to the Korteweg-de Vries equation with its unidirectional propagation properties. It is our belief that the disadvantage of this incomplete description is more than compensated by the relationship that we shall establish between the steady problem and the asymptotic solution (t >> 1) to the unsteady problem, and by the highly stable nature of solitary waves which will be proved later. The above method of derivation of the Korteweg-de Vries equation is essentially that due to Gardner and Morikawa [15], except for the formulation in terms of the velocity potential. The original equation derived by K orteweg and de Vries themselves [40] has a slightly different form from (2.1.16) though a simple transformation reduces it to this form. The Gardner-Morikawa procedure has been adopted here because it seems more elegant and systematic than the original method adopted by Korteweg and de Vries. This classical problem of gravity waves has been extended to various problems concerning waterwave phenomena. From amongst these we mention only a few typical examples. A system of equations governing gravity waves in the presence of a shear flow U(y) may be reduced to the Korteweg-de Vries equation, the coefficients of which are then given by functionals of U(y) [13]. The behavior of so-called internal waves, where a density stratification given by p(y) is present, is also found to be described by the Korteweg-de Vries equation, the coefficients

590

A. JEFFREY AND T. KAKUTANI

of which are then functionals of p(y) [19]. Another example is to be found in the gravity waves which take place on the free surface of a viscous fluid layer flowing down an inclined plane [3]. In this case the resulting equation is not so simple. However, in some special cases, it reduces to the Korteweg-de Vries equation, the Burgers equation or a combined form of them. In another paper [4] Benney has shown how when dealing with long nonlinear waves the system of equations may be reduced to the Korteweg-de Vries equation even in the presence of vorticity, density stratification and rotation. It is also found that the propagation of waves on an elastic tube filled with a viscous fluid, which may be regarded as a simple model of an artery, is described by the Korteweg-de Vries-Burgers equation in a particular limit [27]. A new type of solitary wave has been identified by Benjamin I-2] in connection with steady internal waves of permanent form occurring in fluids of great depth. It can be shown that in the corresponding unsteady case of internal waves in a fluid of great depth the governing equation is a modified Korteweg-de Vries equation. Finally, the damping of solitary wave solutions to a modified Kortewegde Vries equation with various dissipative terms has been examined by Ott and Sudan [55]. One special case of this corresponds to the equation for dissipative lattice waves to be discussed in 2.3 and is similar to the equation for gravity waves in a fluid covering an uneven sea bed derived by Kakutani [29].

2.2. Plasma waves.

(a) Hydromagnetic waves. In recent years, much attention has been paid to finite amplitude hydromagnetic waves propagating in a collision-free plasma
under a uniform external magnetic field. These investigations are related more or less to the problem of whether or not the so-called collision-free shock wave is possible. Up to the present, no satisfactory theory has yet been established to explain the existence of the collision-free shock wave. Nevertheless, many interesting results are known concerning nonlinear hydromagnetic waves. Amongst them, a notable result is that the behavior of weak nonlinear hydromagnetic waves can be described either by the Korteweg-de Vries equation or by one of its generalized forms. This is the main subject of this section. Let us consider, for simplicity, one-dimensional plane waves and take Cartesian coordinates (x, y, z) with the x-axis parallel to the wave normal. The one-dimensional motion of a cold quasi-neutral collision-free plasma may be described by the following nondimensional system of equations [32]:

(2.2.1) (2.2.2) (2.2.3) (2.2.4)

n
c3t

O(nu) =13, Ox

A 2du A2 A2
dv dt
dw dt

+ -nxx (B2
B 8B r n Ox
1 d

c3l
x

=0,

R dt n Ox]

B OB
n

WEAK NONLINEAR DISPERSIVE WAVES

(2.2.5)

B,

-x + B, cqx
cw
U

R, d-x
1

(2.2.6) (2.2.7)

dB
dt d dt

Bx Bzox
+
Ot
Ox,

cqu_

R Ox

591

dr)

where the density n, the magnetic field B(B, By, B), the ion-fluid velocity v(u, v, w) and the coordinates (x, y, z) are normalized, respectively, in terms of the density of the undisturbed plasma no, the intensity of the undisturbed magnetic field Bo, a characteristic speed U o and a characteristic length Lo. The Alfv6n Mach number A is defined by

(2.2.8)
parameters

Uo/VA,

Va

Bo/nho(m + me)

m and me being, respectively, the mass of an ion and an electron. The dispersive

R and R are defined by


Ri

(2.2.9)

mi/mo and Re me/mO, where i eBo/(mic)) and me eBo/(meC)) are, respectively, the ion and electron cyclotron (Larmor) frequencies; and o is the characteristic frequency of the wave defined by Uo/L o Here the time in the above equations has been normalized

with respect to Since we are concerned with plane waves propagating along the x-direction under the presence of a uniform magnetic field Bo(COS sin 0), all physical quantities are assumed to be functions of one space coordinate x and the time t, except for B (= cos ) which, by virtue of the Maxwells equations, can be shown to be identically constant. As for the case of gravity waves, it is instructive to investigate first the linearized dispersion relation in order to determine an appropriate scale change. Linearizing the system of equations (2.2.1) to (2.2.6) and assuming a sinusoidal wave proportional to exp [i(kx t)], we obtain the following dispersion relation"

, ,

Vp

2A(1 +

R R 1A-2k2)
+

(1 + cos ) +
A- R[ R2

coso + sinp + 2 cos


cos

(..0

k
+

(1-coso)a +

+ sin

where are the phase velocities, k the wave number and m the angular frequency. Inspection of (2.2.10) shows that there are three kinds of nondispersive waves

2cos

592

A. JEFFREY AND T. KAKUTANI

(i.e., the phase velocity coincides with the group velocity). Two of them occur when k 0 and the other at some value k ko, say. In this paper we consider only the former situation. 4 Now, for waves with small values of k (i.e., for waves with long wavelength) we can expand (2.2.10) as a power series in k 2 as follows:

Vv+
(2.2.1 la)

2RiReA2

V V(2.2.11 b)
V-d

2RiReA 2

cot2q9 k 2

-ll-

O(k 4)
cos q9

A
+ corresponds to the magneto-acoustic wave, and where Vp Vp- to the Alfv6n wave, because V and Vff represent, respectively, the phase velocities of the magneto-acoustic and the Alfv6n waves for ideal (nondispersive) magnetohydrodynamics [26]. It is obvious that the coefficient of k 2 for the magneto-acoustic wave vanishes at the critical angle given by

(2.2.12)

qgc

tan

R,e RVfe )
{

tan

(mm_ m/)

while the coefficient of k 2 for the Alfv6n wave is negative definite. Therefore the Alfv6n wave has a uniform negative dispersion irrespective of the propagation direction relative to the external magnetic field, whilst the magneto-acoustic wave has positive or negative dispersion depending upon the propagation angle. Here we have adopted the convention that a dispersion effect is said to be positive when the phase velocity increases with the wave number. When q9 qgc, the phase + velocity for the magneto-acoustic wave Vp may be expressed as follows"

(2.2.11c)
condition

V+

2A4RR

sin q

k4+O(k6)}.
2 cos q

It should be noted here that the expansions (2.2.1 la), (2.2.1 lb) are valid under the

(1

cos (49) 2

>)

//

which is roughly equivalent to

(2.2.13)

COS 2 q9

sin 2 q9

A-2R 1R Ik 2,

q9

>>

4mi A -2R/-

me

R lk2

The steady state wave with wave number k

k0 was discussed by Mizutani and Taniuti [52].

WEAK NONLINEAR DISPERSIVE WAVES

593

provided that (p << and me/m << 1. Therefore the limit q --. 0 must be considered in an asymptotic sense because the waves become arbitrarily smooth as q approaches zero. (i) Magneto-acoustic waves (q)=/= q)c). Let us first consider the magnetoacoustic waves with 0 4= qc [32]. According to the dispersion relation (2.2.11a), the phasor of the magneto-acoustic waves (q) (Pc) becomes

kx

cot

(2.2.14)

k(x
2

Vt)
1-

V# +k3t,
cot2q

-2RiRA
(2.2.15)

which suggests the introduction of a coordinate-stretching of the form

G1/2(X

Vgt),

"c

G3/2t,

where G is a scale parameter. In terms of the new coordinates (, r), the system of equations (2.2.1) to (2.2.6) becomes

(2.2.16) (2.2.17) (2.2.18)


(2.2.19)

G- + -{n(u

c3n

c3

+ Vo )}

0,

c3 A2dU +-nb { (B2 + B2) d B By AzdV d n d


G 1/2

A2

dw dr

B, B n 8
V

G 1/2

d
dr,

R
U
8u

(2.2.20)
(2.2.21) (2.2.22)

dBy
d

B.-+ By
8w

1/2

Ri

B(dw)
8

dB=
d d

G 1/2

= G--r + (u

V-)

Now we expand the dependent variables around the undisturbed uniform state as the power series in G"
n
U
U

-t-

e/q (1)

(2.2.23)
W

G2 g(2) GU (1) @ G2U (2) + GU(1) --t- G2U (2) +

__

By

G1/2(GW (1) GZw (2) sin qo + GB(1) + G2B (2) + y gl/2(GB(zl) + e2B(z2)+

--

), ),

594

A. JEFFREY AND T. KAKUTANI

where the half powers in e precede w and B in order to eliminate the half powers in e in the system of equations (2.2.16) to (2.2.21). As a result of this the perturbations of w and B, may be assumed to be of a smaller order of magnitude than those of the other quantities. We assume that the plasma is in a uniform state upstream at infinity, and therefore we have the following boundary conditions:
n (i)
u
(i)

O, O,

u( i)

O,
0,

(2.2.24)

B(ri)

Bz

) --,

0,

i= 1,2,...

Substituting the expansions (2.2.23) into the system of equations (2.2.16) to (2.2.21) and equating terms with the same powers of e, we arrive at the following
sets of equations for each power in e. First we have for e 1,

Vgn
(2.2.25a)

ulna= O,
(1)
-(1) sin (_p By

VAZu

VAZv(t) + B(yt) cos O, V- By(1) + V(1) cos q) U(1) sin q


VgA2w

O, O,

cos q

VgR c3B)
gV(1)

(2.2.25b)

V-B(1) + w(1)COS q) + V-R/-1- -0


where the boundary conditions (2.2.24) have been used. Remembering that l/A, it follows that the determinant of the matrix formed by the coefficients of the homogeneous set of equations (2.2.25a) is zero, so that then

(2.2.26a)

n (1)

Au (1)

V (1)

-u (1) cot qg,

B(yl)

On the other hand, the set of equations (2.2.25b) together with (2.2.26a) gives
W (1)

(R21 A sin3
(4

R-

COS

q?)

(2.2.26b)

cos q0 (R/- sin 3 q9

R- 1) (u(I) 0

U (1)

sin q)

Thus we can express all the first order quantities in terms of u (1) and its derivative with respect to Next we have for e2,

WEAK NONLINEAR DISPERSIVE WAVES

595

u2)=

8(n()u(1))

--(VAu
(2.2.27a)

a-

B sin O)=

A[.,Ou + u).
u n)

sin

(VAZv(2)+
+

()--/ B vgg + n(i)-cos U a{(V oB2)+ V(2) cos- U(2) sin)= B az +u(1)a{
+u
1)
1)

B V) cos)= A 2

1)(1)

02B )

1)

+ By

(u(

+ g- R/-1

(2W(

;( Vg(2.2.27b)

A2w (2) + B(2) cos q)

Vg- R-

----/
8B() cos
(p

(S2)/

+ u 1)

(W(1))
)

(2

V-B(2) + w(2) cos qo + VR;


)

OB + U(1)--8B( + B(z1)
8u (1)

Using the relations (2.2.26a), (2.2.26b) and eliminating the second order quantities in (2.2.27a), we finally obtain the following Korteweg-de Vries equation for u():

(2"2"28)

8u(1)363ru(1)(u8
t-

(1)

2A3RiRe

{ R,.. R/_ee)
1

cot 2 q9

}(3U(1)

It should be noted again that the coefficient of the dispersive term is exactly the same as the coefficient of k 2 in the linear dispersion relation (2.2.11a). Similar () equations are also valid for n (), v (), B(), d and B( ) d. Once the lowest
order quantities have been found it is a straightforward matter to obtain the higher order quantities if necessary.

.w

596

A. JEFFREY AND T. KAKUTANI

(ii) Magneto-acoustic waves at the critical angle (q qc). As was already noted, when the propagation angle q of the wave relative to the external magnetic field becomes the critical angle qc given by (2.2.12), the coefficient of the dispersive term in (2.2.28) vanishes. Therefore we must treat this special case separately [31]. Since the phasor of the wave then becomes (cf. (2.2.11c)) kx oot k(x V t) V #+ kSt, (2.2.29)

2R2R2A4i--e-- sin2 qOc


we should stretch our coordinates as follows"

(2.2.30)

el/Z(x

V t),

es/zt,

where e is a scale parameter. Then the system of equations (2.2.1) to (2.2.6) takes on an analogous but slightly different form from (2.2.16) to (2.2.21). A similar procedure to that employed in (i) leads to the following results. Firstly,

(2.2.31)

n(1)= u(1)

/)(1)

14)(1)=

B(yl)

Bz

1)

0,

and secondly, the second order quantities are governed by a generalized form of the Korteweg-de Vries equation. For example, the equation for u (z) becomes

(2.2.32)

OU (2)

& +

"2)6U(2)

5b/(2)

It should be remarked that the coefficient of the dispersive term is equal to the coefficient of k 4 in the dispersion relation (2.2.11c) except for the sign. (Note that (ik) 3 -ik 3, but (ik) ikS.) Equations similar in form to (2.2.32) are also (2) valid for n (2), v (2,, B(y2), d and B(z2) d. (iii) AlfvOn waves. In the case of the Alfv6n waves [31] the linear dispersion relation (2.2.11b) gives

IW

kx

oot

(2.2.33)
#-

k(x

V t)
+

Vg#- k3t,

2RReA
{

We introduce the following coordinate-stretching"

(2.2.34)

gl/2(X

V-t),

e,3/zt,

where, as before, e is a scale parameter. If we attempt a similar procedure to that used in (i), where the expansion of the dependent variables was as shown in (2.2.23), then with (2.2.24) we have n()= u (1)- v (1)- B(y1)-- 0, (2.2.35)
and we obtain a linear equation for w (1) or

B(z1) of the type

(2.2.36)

c3fz

V-/-

633f c33"

WEAK NONLINEAR DISPERSIVE WAVES

597

This is one of the examples where the original system of equations is not reducible to the K orteweg-de Vries equation, even when the linear dispersion relation is of the form mentioned in the Introduction (cf. (1.6)). Physically speaking, this failure to reduce to the Korteweg-de Vries equation might be taken to reflect the fact that the Alfv6n wave is not accompanied by a "density change." On the other hand, however, earlier investigations [37], [38] of nonlinear steady state Alfv6n waves have shown that even the Alfv6n mode has a solitary wave solution associated with it, though this is clearly not possible for a linear equation such as

(2.2.36).
The difficulty which arises here stems from the fact that to balance dispersion and nonlinearity effects in this case, the stretching of x and must be accompanied by a scaling of the dependent variables which this time differs from the one used in (2.2.23). The situation can be resolved by employing a simple dimensional argument to show that here the appropriate expansion of the dependent variables is as follows:

(2.2.37)

together with the boundary conditions


n (i) u (i)
V
(i)

(2.2.38)

W (i)

i)

_ -

O,
0,

n ()
U ()

1,

0,

O, O,
0, except that

B(zi)

sin

,
as

O,

i=0,1,2,

The essential point of (2.2.37) which is different from (2.2.23) is that we take into account "zeroth" order deviations of the "transverse" components of v, w, By and B from the uniform state. Even if the zeroth order deviations of u and n are taken into account, it turns out that n () and u () 0. It should also be noted that the series expansions (2.2.37) are not in integral powers of e, but in half integral powers of Thus the perturbations of all quantities are of the same order of magnitude with respect to one another. Now substituting (2.2.37) into the system of equations analogous to (2.2.16) to (2.2.21) with V0+ replaced by Vff, and remembering the boundary conditions (2.2.38), we obtain the following sets of equations for each half integral power of e.

598

A. JEFFREY AND T. KAKUTANI

(2.2.39)

+ B(zO)2 Av()+ B(y)=


B(O)2

sin 2 sin

Aw ()
for 1/2
n (1) cos (p

B(2

0;

Au (1)

0,

Au (1) cos q)

/n(0)j(1) y ..y

+ B (o) B (1))
OB(
)

Av (1)

+ B(y1)-+ B(zl)

(2.2.40)
Aw(1)

ReA

(w ()

ReA c3
B(yO)n (1)
B(O)n (1)
R

Av (1)
Aw (1)
which gives

+ B(y1)
+ B( 1)

53
10v ()

Ri

(2.2.41)

for 81,
n (2) cos p

Au (2)

An(1)u (1),
B(_O) B (2))

Au

(2,

cos q)

--(B)B(y2, +

1/2(B1,2 + B(z 1,2 ),


n(1)

B(z)) a
a /

(2.2.42)

CV(2)

OB
(C

- -(r IBI ))
y/,l,B?

(2W(1)

(UT,r/,1,2)

WEAK NONLINEAR DISPERSIVE WAVES

599

c3w (2)

c3

+ 3Bz
c3

c3

(BOn(e)
2V(1)

c3B)
nl)

RiA

(BO)nl)2)

where (2.2.39) to (2.2.41) have been used. Eliminating the higher order terms in (2.2.42) by means of the relations (2.2.39) to (2.2.41), we have
COS 2

(2.2.43)

B
)

OB

632B(y
2RiRe A2

where
~o) By,z =--y,z

B) d

Now it is convenient to introduce new variables (B, 0) defined as

(2.2.44)

B cos 0,

B)
sin 2

B sin 0.

The first line of (2.2.39) gives at once

(2.2.45)

B2

const.

-. Using the new The boundary condition for 0 is given by 0 0 as dependent variable 0 we can rewrite the right-hand side of (2.2.43) as

B(rO)C3/(z c
"r,

B(zo) c3/(r)/ c3 "r,

sin2q9
2R R A 2

)1 4-

cot2 (40

where it should be noted that #- has already been defined by (2.2.33). Differentiating both sides with respect to and eliminating c3/(y)/c3z and c3/)/Oz, we obtain

F(O) + #where

FO

OF(O)

2
30 #- --5 c

c30 V g c3.c

600

A. JEFFREY AND T. KAKUTANI

This equation can be integrated once with respect to equation for 0"

Thus we have the following

where
2_ lim v(o)

Since, however, 0 is very close to the undisturbed uniform state near -, the linearized dispersion relation (2.2.11b) must hold near -o. Substituting 0 c exp [i(kx cot)] into the definition of e2, we then have
(2__

(1 + #-k2- V_ _)
2V

V-

co-

which, by virtue of (2.2.1 lb) and (2.2.33), implies 0 2 0. Therefore we have finally the following third order nonlinear dispersive equation for 0"

(2.2.46)

c30

z
On ()
3

601

V 3"
R )0/ (see (2.2.41) and (2.2.44)),

30

Remembering that nt)= A-(R we can rewrite (2.2.46) as an equation for

(2.2.47)

2#-VA2(R

R)-2n ()2

or, putting n 1)

(2.2.47)

{-- Vff A2(R? R; 1)- 2}- /2f, we have O3f Of F Of f2 Vff

Since u)= Vn ), u 1) is also governed by an equation similar to (2.2.47) or (2.2.47). It is worth remarking that (2.2.47) is a modified form of the Kortewegde Vries equation in the sense that the nonlinear term f df/O in the Kortewegde Vries equation is replaced by f2 Of/O. An equation of this type also appears in the study of anharmonic lattices (see ff 2.3). It is found from (2.2.47) that not only the dispersive term but also the nonlinear term is affected by the dispersive effect. The problem of hydromagnetic wave propagation considered here has been extended to two different cases. One involves the inclusion of the plasma temperature. Starting from the kinetic equation (Vlasov-equation), Kever-Morikawa showed that the magneto-acoustic wave is governed by the K ortewege Vries equation even in a warm plasma [39] (also [16]). By taking account of the electron temperature, Kawahara also showed that the magneto-acoustic wave is governed by the K orteweg-de Vries equation, whereas the Alfv6n wave is governed by a modified Korteweg-de Vries equation similar to (2.2.47) [35]. In those cases,

On ()

O3n()

WEAK NONLINEAR DISPERSIVE WAVES

601

the coefficients of the resulting equations are, of course, functions of plasma temperature. The other extension involves the examination of the effect of electron-ion collisions. In this case, the resulting equation describing the magnetoacoustic wave can be reduced to the K orteweg-de Vries-Burgers equation [36]. The steady propagation of such waves was investigated by Grad and Hu [18].

(b) Ion-acoustic waves. Another weak nonlinear wave propagating in a plasma whose behavior can be described by the K orteweg-de Vries equation is the ion-acoustic wave 66]. We assume that the plasma under consideration is collision free and consists of cold ions and isothermal warm electrons. There is assumed to be no external magnetic field. We shall consider, as before, one-dimensional plane waves propagating along the x-direction. Then the basic system of equations relevant to the problem takes the following form [66]:

(2.2.48)

E
c3n

(2.2.49)
(2.2.50) (2.2.51)

t + t +

c3u

-Ux (nu) cu

O,

e + l On
n

O,

where n and n denote, respectively, the densities of ions and of electrons, u is the ion-fluid velocity, E the electric field, x the space coordinate and the time. All these quantities are normalized in terms of the following quantities" the undisturbed density no, a characteristic speed Uo, the characteristic electric field E o (=_ 4nenoLo, e being the electronic charge), a characteristic length L o and the characteristic frequency COo (-= Uo/Lo). The nondimensional parameters Rpi and lD are defined, respectively, as (2.2.52) Rpi Opi/O) 0 and D LD/Lo,
where OOpi (--

LD (=

x//4ne2no/mi) is the frequency of the ion plasma oscillation and w/kTe/(4ne2no)) is the Debye length; mi, k and Te being the ion mass, the

Boltzmann constant and the electron temperature (which is assumed to be constant), respectively. In the above equations, the electron inertia has been neglected, since we assume that Rpi O(1) and 1D O(1), and for most plasmas the mass ratio of an electron to an ion, me/mi, is much smaller than unity. Linearization of the system of equations (2.2.48) to (2.2.51) gives the following dispersion relation"

(2.2.53)

Vv

where M is the effective Mach number defined as

(2.2.54)

Uo/a

co

MN//1 + 12Dk2,
a

w/kTe/mi.

602

A. JEFFREY AND T. KAKUTANI

2 2 Here a represents the effective sound speed (note that Rpil M -2) It is easily o found from (2.2.53) that the phase velocity Vp is identical to the group velocity if and only if k 0. For small values of k, the dispersion relation can be expanded

as the power series in

k2:
Vo(1
1,zt.2 o,.

(2.2.55)

Vp

o/k

+ O(k4)),

Vo

1/M.

Therefore the phasor of the wave becomes

(2.2.56)

kx

ot

k(x

Vot + 1/212k3t.
z

Introducing, as before, the coordinate-stretching

(2.2.57)

el/2(x- Vot),

e3/2t,

we can rewrite the system of equations

(2.2.48) to (2.2.51) in terms of (, z):

(2.2.58) (2.2.59) (2.2.60)


(2.2.61)

e 1/2

cE

o=n-ne,
+
Vo)}
e

0,
--pi-,

ezz + (u- Vo)g-I/2E

12u +n

Cne= O. c3

On the other hand, let us expand the dependent variables around the undisturbed uniform state as the power series in e:

(2.2.62)

+ en() + 2rt(2) d- , n 1 + enter) + eZn2) + ..., U 2U (2) + U (1) E e/z(e.E () + eZE (2) + ...).
n

We assume that upstream at infinity the plasma is in a uniform state, so that we impose the following boundary conditions:
n (i)

0,

(2.2.63)

n
U i)

E (i)

--

0,

O, O,

as

--,

,
i= 1,2,

WEAK NONLINEAR DISPERSIVE WAVES

603

Substituting (2.2.62) into the system of equations (2.2.58) to (2.2.61) and equating terms with the same powers of e, we obtain the following sets of equations for each power in e:

For e 1,

n (1)-

(2.2.64)

Vo nil) Vo
3u(

E + lo
and for e 2,

-nel)
u
(1)

0,
--0,

-+- R 2 E
O lrl

O,

o,

(2.2.65)

where the boundary conditions (2.2.63) have been used. Using the relations (2.2.64), we can eliminate higher order terms in (2.2.65), obtaining a final result:

(2.2.66)

c3u ()
Oq7

u(x)

U (1)

2 VI2D

3 U(1)

JE

Again this is the K orteweg-de Vries equation. In view of (2.2.64), n {), ) and ) d are also governed by an equation similar to (2.2.66). To conclude this section, it should be remarked that the problem of ionacoustic waves considered here has been extended, recently, to two different cases. One is to include the effect of Landau damping [54] and the other is to examine the effect of electron-ion collisions [30]. The former introduces an additional dissipative term, which is proportional to the principal value of j (au)/a{)d{/( ), whereas the latter introduces a nonlinear dissipative term proportional to c3/c3{(c3u)/c3z + u ) 3u)/a{). These additional terms to the Korteweg-de Vries equation (2.2.66) should be compared with 2U(1)/02 in the Korteweg-de VriesBurgers equation.
2.3. Lattice waves. In this section we show how a study of nonlinear lattice waves leads to a continuum approximation to the lattice problem in the form of a generalized K orteweg-de Vries equation. Here the term lattice waves is used to describe atomic vibrations in solids and for a detailed account of the linear theory

ne

604

A. JEFFREY AND T. KAKUTANI

FIG. 2.3.1. One-dimensional lattice comprising N particles

we refer the interested reader to Dean [10, p. 561], [11]. We take as our starting point a simple discrete model comprising a set of identical masses m equi-spaced on a straight line between given endpoints and connected one to the other by identical springs. This was the model first introduced by Born and von Karman [7] in linear form and in a modified nonlinear form by Debye [12] in an attempt to account for the anomalous thermal conductivity properties of certain crystals. Indeed, it was this latter problem that led first, Fermi, Pasta and Ulam [65, Chaps. 7, 8], and then Zabusky [69], to examine matters further thereby stimulating much of the current interest in the K orteweg-de Vries equation. As a preliminary to the nonlinear case let us consider briefly the linear model illustrated by Fig. 2.3.1. We shall suppose that there are N identical particles of mass m and that they are joined as indicated in Fig. 2.3.1 by identical springs of length h, each with spring constant Then the equation of motion of the ith particle is

(2.3.1)

mYi

(Yi+l

Y,) + z(Y,-I

Y,),

where y is the displacement of the ith particle from its equilibrium position and dots signify differentiation with respect to time. If we assume that the group of N particles is repeated periodically along the axis with each group attached to the next, then only periodicity conditions need be given at points A and B distance L apart. A solution to the system of equations (2.3.1) subject to some suitable initial conditions of the form

Yil,
with
i=

, /i 1,2, ..., N,

oi

and ,f ilt o

fli

arbitrary constants, may be obtained by writing Yi C e with when we find from (2.3.1) that o must satisfy the algebraic

equation

(2.3.2)
with

-co2C,

(C,+, m

2C, + C,_ t),

N and the cyclic condition __+ N i. 1, 2, From these equations we deduce that for a nontrivial solution involving constants C that are not all zero it is necessary that (_D 2 satisfy the matrix eigenvalue problem

...,

(2.3.3)

(rc/m)A k- 6021

O,

WEAK NONLINEAR DISPERSIVE WAVES

605

where
2 -2
0 0 0

0 0

(2.3.3a)

-2

-2

and I denotes the unit matrix of order N. 2 In terms of mo /m, the eigenvalues o92 are readily seen to be 2 N 2 09 =4o9sin 2(rej/N) for j= (2.3.4)

Here o9o is the natural frequency of the spring mass system. When N is large it is usual 11] to consider j to be a continuous variable and to write rcj/N, when 0 </ < t. Equation (2.3.4) then assumes the form

(2.3.5)

o9

209 o sin/ for 0 </

=< t,

which is simply the dispersion relation in this linear model. Whereas in linear lattice theory the solution to an initial value problem for a lattice usually involves the use of higher order modes corresponding, to large values of j in (2.3.4), in nonlinear lattices Zabusky [69] has shown that only the lowest order modes seem to be involved. In view of this, a continuum approximation may be used in which the system of coupled ordinary differential equations (2.3.1) is replaced by a partial differential equation in one space variable and time. This partial differential equation is derived from the discrete model in such a way that its linearized dispersion relation only involves the lowest modes of the discrete model; in other words its linearized dispersion relation approximates (2.3.5) for small k. Later we see that the simplest continuum approximation in the linear case is the ordinary one-dimensional linear wave equation. Let us now follow Zabusky [69 and introduce nonlinearity in the form of a nonlinear spring force F between particles through the relation

(2.3.6)

where A is the displacement from equilibrium and and p are constants. In addition, and differently from Zabusky, we assume that the ith point mass with displacement Yi is subjected to a resisting force 237, where 2, q are positive
constants.

,,

The governing equation replacing (2.3.1) now becomes

(2.3.7)
where
o9

m
1, 2,

(y+

+ [(Yi+

...,

sEn ()11 N or, in terms of the linear spring frequency o90,

Yi) p+I

(Yi

Yi- 1)

p+

13

(2.3.8)

o- 2yi

(Yi+
sgn

2yi

+ y_ 1) + [(Y+

Yi) p+

(Y

Y- 1)P+1]

for/= 1,2,

..., N.

606

A. JEFFREY AND T. KAKUTANI

To convert this to a continuum approximation we now assume that Taylors theorem may be used to interpret Yi_+l in terms of partial derivatives of Yi with respect to x. In a first approach to the nondissipative version of this problem (2 0), Zabusky only retained terms in the Taylor series to order h 3. As a result, by a process we describe later, he arrived at the lowest order nonlinear continuum representation
6t 2
c2

+ cz(1 + p)h p

-x

where c 2 oh is the propagation speed of waves in the linear continuum case 0 and h 0. In the linear case when z 0 this reduces to the wave in which equation already mentioned. By an ingenious use of the hodograph transformation Zabusky 68 solved this equation subject to the initial conditions

y(x,O)=asinx,

t(x,0)=0

for0=<x__< 1,

and showed that a differentiable solution only exists for a finite elapsed time, so that obviously (2.3.9) cannot describe the nondissipative lattice problem satisfactorily. The breakdown of this solution has been established using different techniques both by Lax [42] and Jeffrey [23. If a continuum approximation does exist, then it must be derived by retaining higher order terms in h, so let us now assume that terms of order h 4 must be retained in the Taylor series, and generalize Zabuskys results to the dissipative system (2.3.8). We start by writing
Yi+

[x
y +_ h
Yi
Yi-

from which we find at once that

(2.3.10)
and

Yi+

h-- -t

c3ZY + h333yhg3Y i 6x 2 3! OX 3 4! c3x 4


h2
h 2 O2y 21 (X 2 h 2 2y 21 IX 2

t- O(h 5)

h 3 O3y 31 OX, 3
h 3 3y 3 OX, 3

h t?,y t-O(h 5) 41 t?x

(2.3.11)

Yi

hc3x

t?y

h4

c3Y

41 dx

t- O(h 5)

Again setting 2 o)oZh 2, and then using (2.3.10) and (2.3.11) in (2.3.8) and deleting the suffix i, gives rise to the partial differential equation

(2.3.12)

2y__ C22y +(1 t2 [_x2


-sgn

which is a fourth order continuum approximation to the discrete system (2.3.8). Notice that in the nondissipative case (2 -= 0), when working only to the lowest significant order in h, (2.3.12) reduces to the simplest nonlinear continuum approximation already displayed in (2.3.9).

--

+p)h p

y)pZy + Lh24y 12 c3x -x

+O(h 4)+O(oh p+2)

WEAK NONLINEAR DISPERSIVE WAVES

607

Neglecting terms of order higher than O(h 4) or O(oh p+ 2) in (2.3.12) we thus arrive at the following continuum approximation to the nonlinear dissipative system (2.3.8):

In the nondissipative case (2 0) this reduces to the equation studied by Zabusky [69]. To transform (2.3.13) into a generalized Korteweg-de Vries equation we now employ the semicharacteristic variable-stretching transformation

(2.3.14)
whereby

[_x2 + e(1 + p)h

xz]-+-

h2

sgn

--

(2.3.15)

c3x

c3

ct

e c-,

where e is a small parameter. In terms of the new variables (, z) we then find that (2.3.13) assumes the form

(2.3.16)

sgn

where O(e 2) and O(2e/h 2) have been neglected. Giving to e the value e c(1 + p)h/2, and setting y/ (2.3.16) to the equation

(2.3.17)
where

uw

3u

"

cq+l

h2

=0,
u, then reduces

-sgn (u)vlul

tt

h2-p/{120(1 -+- p)},

Rcq/{(1 + p)h p+ 2}.

This is a generalized dissipative K orteweg-de Vries equation and, as would be expected from the construction of the model from which it was derived, dissipation only enters as a power of the undifferentiated dependent variable. This suggests that the effect of dissipation will be relatively small, and if the scaling is such that c 1, then the dispersive and dissipative effects in (2.3.17) will only be comparable when O(h z-p) and 2 O(h4). To conclude, let us check whether the dispersion relation for the linearized form of (2.3.13) does, in fact, involve only the lowest order modes of the discrete model. First, notice that only when q 1 will the last term on the right-hand side of (2.3.13) be linear, and that the second term will always be nonlinear if e 0. and neglecting the nonlinear term involving we set So, taking q

(2.3.18)

const,

exp [i(kx

oat)]

608

A. JEFFREY AND T. KAKUTANI

in the resulting equation to obtain


co

ckll
toohk
hzk2
24

1/2

or

(2.3.19)

2c

2h2k_]

-]

The imaginary term accounts for the dissipation produced by the linear damping when 2 4: 0, so let us confine attention to the real part of to. Identifying x with h we see from (2.3.18) that y is periodic in kh with period 2zc, so that to compare dispersion relation (2.3.19) with dispersion relation (2.3.5) we must make the identification k kh/2. We find that

Re [to]

2COo(/- 3/3 !),

where the bracketed term is simply the first two terms of the series expansion of sin k. Thus the dispersion relation of the linearized form of the fourth order nonlinear continuum approximation does depend mainly on the lowest order modes of the discrete model.

2.4. A general reductive method. As we have seen in the foregoing sections, there are various physical systems which can be described by the K ortewegde Vries equation. Up to the present, however, there is no general selection rule to determine whether or not a given system of nonlinear equations can be reduced to the K orteweg-de Vries equation. In other words, the necessary and sufficient conditions under which general nonlinear systems of equations can be reduced to the Korteweg-de Vries equation have not yet been established. The form of the linear dispersion relation given in (1.6) is merely a necessary condition, but not a sufficient one. An heuristic approach, however, has been made by Taniuti-Wei [633 and Gardner-Su 173 to reduce a certain class of nonlinear equations to the K ortewegde Vries equation or one of its generalized forms. Taniuti-Weis method is also applicable to dissipative systems, where the original system of equations is reduced to the Burgers equation or to one of its generalized forms. Here we shall follow Taniuti-Weis method. We consider the following equation for a column vector U with n components
Ul,U2,

Un,

1"1

2,

+A

+
-1

x H+ K

U=O,

p>2,

and where A, are n n matrices with elements depending upon U alone and are assumed to be differentiable with respect to U as many times as necessary. The method is based on the following assumptions: (A1) There exists a constant solution U0 representing a uniform state, and U can be expanded around the uniform state as a power series in a scale

WEAK NONLINEAR DISPERSIVE WAVES

609

parameter

(2.4.2)

U
j=0

8JUj.

(A2) The eigenvalues of Ao (= A(Uo)) are real, at least one of them is nondegenerate, finite and nonzero and the eigenspace does not contain any invariant subspace. (A3) U tends to the uniform state U o as x ---, -, that is, (2.4.3)
which means that

--,

Uo

as x -,

-,
j= 1,2,....

U j-,0 asx-, -, (2.4.3) By virtue of assumption (A1) and the differentiability of A, may expand A, H and K as power series in

H and K, we

"

A
j=o

;JAj,

(2.4.4)

H
j=0 j=0

:JH

j,

Ho Ko

U. (V,A)o and similar expressions may be defined for Here g, and K,. V, denotes the gradient operator with respect to U. Before proceeding to the nonlinear problem, let us first derive the linear dispersion relation. Setting U U o + U, IU[ << [Uo[, and linearizing (2.4.1) with and assuming a sinusoidal wave exp [i(kx mr)I, we obtain respect to
where

Ao

A(Uo),

the linear dispersion relation

(2.4.5)

det

--VpI + A o + (ik) p-1


=1=1

(-VpHo +

Ko

O,

where I denotes the unit matrix and Vv(= m/k) is the phase velocity. For small values of k, we may express the phase velocity Vp as a power series in k p- 1:

(2.4.5)

Vv= Vo{1 +a lk v-1

+ O(k 2(p-

where Vo is the nondegenerate, finite and nonzero eigenvalue of A o and al is a constant given by

(2.4.5")
This last assumption can be extended to more general boundary value problems. In fact, we may replace (2.4.3) by U U(x o, t) at x x o, where U(x o, t) is a given boundary value. We may also con-oe or, more generally, U sider the initial value problem, for example, U U(x, to) Uo as at to, where U(x, to) is a given initial value. For details see [63].

610

A. JEFFREY AND T. KAKUTANI

r o and lo being respectively the normalized right and left eigenvectors of A o associated with the eigenvalue Vo. Since the phasor of the wave becomes

(2.4.6)

kx- cot

k(x- Vot)- VoakVt,

we shall introduce the following coordinate-stretching:

(2.4.7) ea(x- Vot), r, ea+lt, a=_(p- 1)-1. In terms of the new coordinates (, z), (2.4.1) can be expressed as follows:

(2.4.8)

+ (- VoI + A)-5- + e

eH-z + (- VH + K)

O.

Substituting (2.4.2) and (2.4.4) into (2.4.8) and equating the terms with the same powers in e, we obtain for e 1,

(2.4.9)
and for g2,

U (-VoI + Ao)- O,

(- VoI +

Ao)CU2

c3U1
p

{U (VuA)o}

(2.4.113)
Integrating (2.4.9) with respect to we obtain

{(--goHo-Jl- Ko)}
/=1 =1

(PU1

subject to the boundary condition (2.4.3),

(2.4.11)

ro u(:),

where u (1) is one of the components of U and the corresponding component of r o is normalized to unity. The nonappearance of an arbitrary function of : in (2.4.11) is because U tends to the uniform state U o as x --, -o, and by (2.4.7), oe implies --, oe forj or Uj 0 as x 1, 2, (See footnote 5 in connection with boundary condition (2.4.3).) Then, multiplying (2.4.10) by the normalized left eigenvector lo associated with the eigenvalue Vo, we can eliminate the second order quantities in (2.4.10), thereby obtaining a single equation for ul):

(2.4.12)
where

U(1)

27

t-

ClU(I)

(U(1)

(Pu(1)
c2

i0

lOOP

(2.4.12a)
(2.4.12b)

1
c2

/o{ro (VuA)o}ro/(loro),

-lo
/

lo=1

I-I (-VoHo + Ko

ro/(loro)

By virtue of (2.4.11) other components of U are also governed by equations similar to (2.4.12). It is obvious that if p is odd, then (2.4.12) is a generalized Korteweg-de Vries equation, while if p is even, it is a generalized Burgers equation.

WEAK NONLINEAR DISPERSIVE WAVES

611

Apart from the sign factor

-i p- 1, the coefficient of the highest derivative term c 2

is exactly the same as the coefficient of kp- in the linear dispersion relation (2.4.5). The ion-acoustic wave discussed in 2.2(b) belongs to this class. If the eigenspace A o contains invariant subspaces, the above method is not necessarily valid. For instance, let us consider the simplest case in which it is composed of two invariant subspaces of rn and n rn dimensions, so that Ao has an irreducible representation

(2.4.13)
while

Ao

A0

A;-

taking the form

Z=I-IP= (H O/& + K O/Ox) is an interchange between these spaces,


0

(2.4.14)

D0

D+

say. Then c2 given in (2.4.12b) becomes zero 6 and consequently it turns out that (2.4.12) does not necessarily admit a smooth solution, which is inconsistent with the assumption (A1). Since it seems very difficult to establish a general theory, we content ourselves with the following special case. We consider the equation

c--- + Axx +

H+K
A
A+ C

p> 2

where U, A, H and K have analogous meanings to those in (2.4.1). We modify the fundamental assumptions (A1)-(A3) as follows: (B1) The matrix A has the representation

(2.4.16)

B A-

in which A + and A- are respectively rn m and (n- m) (n m) matrices, which are functions of U + only, while the elements of B and C are linear combinations of the components of U- multiplied by functions of U + only, where U + are defined by
U

(2.4.17)

U+
U-

UUm

/Urn+ urn+

(B2) There exists a uniform state given by

(2.4.18)

U=

Uff

This corresponds to the vanishing of the coefficient a of k in the expansion of the linear dispersion relation. Thus the lowest significant order may become O(k 2(p- )).

612

A. JEFFREY AND T. KAKUTANI

and U -+ can be expanded around the uniform state as power series in the scale parameter

(2.4.19a)
(2.4.19b)
(B3) A o
=_

U+
j=O

eJU+,
e;U]-.
j=l

U-= el/2

A(Uo) has an irreducible representation


Ao

(2.4.20)

A0

which means B(Uo) C(Uo) 0. The eigenvalues of A0 are real; at least one of them is nondegenerate, finite, and nonzero and is not equal to any eigenvalue of A-. (B4) H and K are functions of U + only, and the operator in the last term of (2.4.15) takes the following form:
P

(2.4.21)
say.

=1

1--I

H + Kxx)
U

c3

3\

0 O+

D0

(B5) U tends to the uniform state U o as x (2.4.22)


which means that

- Uo
as x
:1

-, that is, -,
j= 1,2,....

(2.4.22)

Uf 0
b(x
Vt),

asx-,

We now introduce a coordinate-stretching defined by

(2.4.23)

b+It, b

=- 2(p- 1)
A-,
which is not

where V; is the nondegenerate, finite, nonzero eigenvalue of equal to any eigenvalue of A-. Then (2.4.15) takes the form

(- VgI + A)---+

(- VH + Ks)

+ eH

U =0.

The last term of this equation may be expressed as

(2.4.25)
in which

e/2

(-VgH + K)

+ eH ) U

e /2

Z eDU,

Dq takes the form

(2.4.26)

Dq

WEAK NONLINEAR DISPERSIVE WAVES

613

by virtue of the assumption (B4). Hence (2.4.24) splits into two parts" p (U + GU 8U + B (2.4.27a) a + (-VI+ A + + 1/2 eD U =0,

(2.4.27b)
and

)---+ a 8U8U + I + A- )SU- + C Vg (+ e--r (


cz
A+

D as the power series in

In view of the assumptions (B1), (B2) and (B4), we may expand A -+- ,B,C

(2.4.28a)

(2.4.28b)
(2.4.28c)

B--81/2
C---1/2

q=O
p

81/2

q=O

E qD; U +

O.

j=0

E ,JAf-,
j=l

E 8JBj
E 8"Jcj

j=l

(2.4.28d)
in which

D y J d,
j=0

A U .(V.+A+-)o,
C1
and

B1

UU

U-

(Vu_C)O

dl

"

(V,_ B)o, (V + D)o,

^+ where d6o are constant matrices. Substituting (2.4.19a), (2.4.19b) and (2.4.28a)-(2.4.28d) into (2.4.27a), (2.4.27b) and equating terms with the same powers in e, we have the following sets of equations for each power in First we have for e 1,

,.

(2.4.29a)

(-Vg-I + A-) aui


(- V- I + A-)

(2.4.29b)

Equation (2.4.29a) can be solved immediately, giving

(2.4.30)

where the boundary condition (2.4.22) has been used. In this expression, U (1)+ is one of the components of Ui and is the normalized right eigenvector of Ag associated with the eigenvalue V0+. Substitution of (2.4.30) into (2.4.29b) gives an expression for Ui- in terms of u (1)+, that is,

(2.4.31)

0,

c-

+ d-o U
()

r- u

+,

r-

614

A. JEFFREY AND T. KAKUTANI

Next, we have for 2,

doU
(- Vg + A)--;_ + dgo U]

u;

u;

u
l-, say, and using

(2.4.32b)
Multiplying (2.4.32a) by the normalized left eigenvector of A-, the relation (2.4.31), we obtain finally

(2.4.33)
where

- SU()+ST,
c-

(2.4.33a) (2.4.33b)

C-U (1)+ --tOU(1) C

82P- lU(1)+

+ ^+ + + r;-). l-o( VoI + A) dooo/(o

Equation (2.4.33) is simply a generalized form of the Korteweg-de Vries equation, and reduces to it when p 2. The magneto-acoustic wave (q 4= q)) discussed in 2.2(a) belongs to this class, v
3.1. Steady solutions.

(a) The Korteweg-de Vries equation. Let us consider the Korteweg-de Vries
equation in the following form"

(3.1.1)

since any Korteweg-de Vries equation with arbitrary coefficients can always be transformed into this form by an appropriate rescaling of the variables. In this equation, the dispersive parameter # is not necessarily small and may be either positive or negative. Without loss of generality, however, we may assume/ > 0, since we can obtain an equation of the form

u--x + #-x

0,

(3 1.1)

8u

03U Ou -/+ uux- --x o,

by making the variable changes u --, -u, x -x and --, t. Steady solutions of (3.1.1) have already been obtained by Korteweg and de Vries themselves [40], who showed that these solutions are either solitary or
0 including the magneto-acoustic wave at q A generalized system in which c obtained in a similar way if we utilize the coordinate-stretching defined by

qc can be

gb(X
(see [31]).

V-t),

g/+2t, b

2(p

1)

WEAK NONLINEAR DISPERSIVE WAVES

615

periodic waves propagating with constant velocity relative to the medium. Following Korteweg and de Vries, we seek a solution of the following form:

(3.1.2)

u(x,t)=u(), =x-2t, 2=const. Substituting (3.1.2) into (3.1.1) and integrating twice with respect to

,
>=

we obtain

(3.1.3)

-u a

32u

6Au

6B

f(u),

say, where A and B are integration constants. It is interesting to note that this equation may be regarded as an equation of motion of a particle with unit mass under the potential -f(u)/(6), or as an equation of an anharmonic oscillator provided that we interpret and u as time and space coordinates, respectively. It is obvious that a real solution of (3.1.3) is possible only for f 0 (note that g > 0). Since there is no bounded real solution if f(u) 0 has only one real root, we shall assume here that the function f(u) has three real zeros; that is, we set

(3.1.4)
where c < c 2 <
C3

f(u)
2

(u

Cl)(U

c2)(c 3

u),

which implies that

(3.1.5)

A
B

1/2(c + c2 + c3), ---(C1C 2 + C2C 3

-c lC2C3.

In order that c l, c 2 and c3 should be real, the following conditions must hold: 22 +2A>0, (3.1.6) f(2 + N/// 2 2A) >= 0 and /(2 //22 -+- 2A) N 0. The general behavior of the function f(u) with u is shown schematically in Fig. 3.1.1 (curve A). Thus the real solution of (3.1.3) represents a nonlinear oscillation between c2 and c3. There are two special cases. One is that c2 Cl (curve B), which corresponds to the solitary wave, and the other is that c2 c3 (curve C)
giving rise to a constant solution given by u
f(u)

C3C1)

c 3 which is related to linear sinu-

FIG. 3.1.1. The behavior off(u). A." cnoidal wave, B: solitary wave, and C." constant solution (u

c3)

616

A. JEFFREY AND T. KAKUTANI

soidal waves. In fact we can reproduce sinusoidal waves by allowing c2 c3 (i.e., k 2 << 1) in (3.1.7) below. If three zeroes c c 2 and c3 coalesce, we have again a constant solution corresponding to this triple zero. The explicit solution of (3.1.3) can be obtained in terms of the Jacobian elliptic function cn:

(317)..

u--c 2

--(C 3 --C2)cn2

I312#--c1-{x---}(c1
/-_ 3 4K.c
[/3
V
3
Cl

nt-Cznt-c3)t};k

where k 2 (C 3 C2)/(C 3 C) (see [8]). On account of this, the periodic wave train (3.1.7) is often called a "cnoidal wave." Since the real period of the function cn 2 is 2K, K being the complete elliptic integral of the first kind, the period of the cnoidal wave u is given by

(3.1.8)

Tp

If, in particular, c2 c, then k 1, and the elliptic function cn degenerates into the hyperbolic function sech, and the period becomes infinite; that is,

Te

(3.1.9)
If we set

c + (c 3

c)sech 2

Cl {X 12#

(2C + C3)

}].

u and c 3
u u

a, then (3.1.9) takes the following form:

(3.1.9)

+asech2

[&{

)}

Thus u and a may be interpreted as the uniform state at infinity and the amplitude of the solitary wave, respectively. It shoMd be noted that the speed of the wave relative to the uniform state is proportional to the amplitude, the width of the wave is inversely proportional to the square root of the amplitude, and the amplitude is independent of the uniform state u. We can obtain the solitary wave solution (3.1.9) directly from (3.1.3) if we impose the conditions

fl.=. af /aul.=. o.
The same result may also be obtained if we impose the boundary conditions
asy,ffp

as

or

+@

where u u means that not only u -m or { respect to { tends to zero as {

u but also every derivative of u with + m.

(b) The modified Krteweg-de Vries equation. The steady solutions of a Korteweg-de Vries equation
0t

(..0)

+eu +g=0
.

1, m=positiveinteger,

> 0, a

u -x and -u, x we change the sign of

We may assume g > 0 in the sense mentioned in 3.1(a), i.e., if m is odd, the transformations give an equation for < 0, and if m is even, the same is true provided that

WEAK NONLINEAR DISPERSIVE WAVES

617
x

may also be obtained in a similar manner [69]. In fact, if we put u u(), 2t, 2 const., then (3.1.10) can be integrated to take the following form"

fro(U), u(au/ct) f,,(u) -2[(m + 1)(m + 2)]-lu+ 2 + 2u 2 + 2Au + 2B, where A and B are constants of integration. Thus the solution u of (3.1.11) generally represents nonlinear wave trains between real zeros of the function f,,(u), where f,,(u) > 0 between the zeros.

(3.1.11)

There are many interesting special cases which include solitary waves and even shock waves depending upon the values of ct, 2, A and B. To illustrate this let us consider some examples below. 1. In this case, if u 0 and fml,=,=O dfm/dul,,=,,=o 0, then (i)

(3.1.11) becomes

-uZ{2u"[(m + 1)(m + 2)] -1


Thus a possible solitary wave has the speed

2}.

(3.1.13)

2a[(m + 1)(m + 2)]- 1, where a is the amplitude of the solitary wave. In fact, integration of (3.1.12) gives
2
u
a sech 2/m (/D),

(3.1.14)
D=

I-21a(m +

l)(m +

If we put m 1, we recover the result (3.1.9) with uoo 0. It is interesting to note that the larger the index m becomes, the more slowly the amplitude tends towards the uniform state. For m 2, (3.1.14) gives two possible solitary waves, the compressive one with the amplitude a and the rarefactive one with the amplitude ar (see Fig. 3.1.2). If we restore the uniform state u we obtain

(3.1.15)
showing that

ac- ar
ac ar if u
)= u 2

-4u,

0. The corresponding wave speeds obtained are

(3.1.16)

+-aca=

+ -ac(ac + 4u uo + -a(ar 4u
u2
2

There is another interesting special case for m 2. This occurs when fzlu=u dfz/dulu=u dZfz/duZ[u=u 0, where u 0. This corresponds to the coalescence of three out of four real zeros of the function f2 in Fig. 3.1.3. In this case, f2 takes the following form"

(3.1.17)

f2
c

--(U

Uoo)3(/,/
U2

C),

-3uoo, 2

618
U

A. JEFFREY AND T. KAKUTANI

Uoo+

FIG. 3.1.2. Compressive and rarefactive solitary waves. A" Compressive solitary wave (2 u B" Rarefactive solitary wave (2 u

+ ac(a + 4uoo)/6) + at(at 4uoo)/6).

(See [69].)

whence the amplitude a of the wave is given by a integration of (3.1.11)now gives


-1

[c

o[

14uoo[. In fact,

(3.1.18)

uoo

which represents a solitary wave that tends algebraically (i.e., much more slowly when compared with the hyperbolic function) to the uniform state as ][--, oo. If u < 0, we have a compressive solitary wave, whereas if u > 0, we have a rarefactive one. (ii) 0 -1. In addition to the wave trains and solitary waves, we have an interesting special case for -1, which corresponds to a "shock" wave. For

c=

u=""
/
Fie;.

3.1.3. Algebraic solitary wave (2

uoo

aZ/16). (See [69].)

WEAK NONLINEAR DISPERSIVE WAVES

619

FIG. 3.1.4. Transition through shock wave (2

-u2/3)

example, f2 can take the following form for m

2"

(3.1.19)

f2
j

(u
2

u)2(u + uoo) 2,

Integration of (3.1.11)now gives

(3.1.20)

u=

_+utanh

(up),
>= 3,

which represents a "shock" wave in the sense that it connects two different uniform states as shown in Fig. 3.1.4. It is obvious that we may have more interesting possibilities for m 3. Some discussion of this matter is given in 69]. We may also generalize the K orteweg-de Vries equation to the following form"

(3.1.21)

though so far an explicit solution to this equation has not even been obtained for the simplest case m 1 and p 3. An example of this type of equation is obtained for a magneto-acoustic wave propagating along the critical direction qc (cf. 2.2(a)). (c) The Korteweg-de Vries-Burgers equation. In this subsection we consider an equation which represents a combination of the K orteweg-de Vries and Burgers equation, namely,

(2P- lU +__ um(U + #Ox 2p-1

-x

0,

(3.1.22)

c3 + Uxx

6U

v-x2 +/-x
3

2U

3U

0,

v > 0,

> 0.

620

A. JEFFREY AND T. KAKUTANI

2.1 and 2.2(a), this type of equation occurs in some classes of nonlinear dispersive systems with dissipation. Physical considerations require that the dissipative parameter v must always be positive, while the dispersive parameter / may be either positive or negative. We may, however, assume here/ > 0 in the sense discussed at the start of 3.1(a). As is well known, the Burgers equation, (3.1.22) with # 0, has a steady solution of the form

As we have mentioned in

(3.1.23)

}(Uo + U+)- (UT

+ uoo)tanh

4V

where x 2t, 2 const. 2*(uo + u) and uoo > 0 with uT and + denoting, respectively, the uniform states at infinity upstream ("- -o) and downstream ( + oo). This solution may be regarded as a "shock" wave which + On the other hand, connects two different uniform states designated by and uoo. we have seen that the Korteweg-de Vries equation ((3.1.22) with v 0) has a steady solution (3.1.7) exhibiting a cnoidal wave which reduces, in a special case, to a solitary wave (cf. (3.1.9) or (3.1.9)). Therefore, it is interesting to ask what sort of steady solution belongs to the (3.1.22) and how it depends on the relative magnitude of the parameters v and #. Now setting u u(), x- 2t, 2 const, in (3.1.22) and integrating it with respect to we obtain

"

(3 24)

dZu ..... d
#
"2

1 2 u 2

+ 2u+ A +

du

Vd--(,

where A is a constant of integration. This equation may be interpreted as an equation of motion for a particle or of an anharmonic oscillator under the action of a nonlinear force together with a friction proportional to the velocity, provided that we regard and u as the time and space coordinates, respectively. Pursuing this analogy, we introduce the "velocity" v by

(3.1.25)
dill

du/d,

and then (3.1.24) reduces to the following pair of first order simultaneous equations"

(3.1.26)
where

dv

(u- u o)(u- u +)

u is defined by
2

(3.1.27)
and we assume 2

u =2-T-v/22 + 2A,
u u

+ 2A > 0 (cf. (3.1.6)). It is obvious that the system of equations (3.1.26) has two singular points, namely,
(3.1.28a) (3.1.28b)
uoo,
v
v

0,
0.

u,

WEAK NONLINEAR DISPERSIVE WAVES

621

The elementary theory of the ordinary differential equations then establishes that
a nodal point

if V 2

_ - 4#x//22 + 2A,
(4/tx//, + 2A,
2

(3.1.29a)
and that

(u L,0) is

a spiral point a center

if 0 < 2

ifv =0,

(3.1.29b)

(u+, 0) is always a saddle point.

Since the solution of (3.1.26) may include the Burgers shock wave (3.1.23) as a special case, it seems probable that the singular point (uS,, 0) corresponds to the + upstream infinity while (u 0) corresponds to the downstream infinity (vice versa if we assume # < 0). A sketch of the phase diagram in the (u, v)-plane and of the possible profiles of the solutions are shown in Fig. 3.1.5 and Fig. 3.1.6, respectively. When the dissipation is dominant (1 2 4/x//22 4- 2A), the solution represents a monotone shock wave qualitatively similar to the Burgers shock wave (/ 0). On the other hand, when the dispersion is dominant (0 < v 2 < 4#x//22 2.,), the solution exhibits an oscillatory shock wave which may be regarded as a combination of a quasi-solitary wave and the damped cnoidal wave. In the limiting case of v 0, we recover the solitary wave or the cnoidal wave.

FIG. 3.1.5. Phase diagram in the (u, v)-plane. A: Solitary wave (v 0), B cnoidal wave (v 0), C: oscillatory shock wave (0 < D monotone shock wave (v __>

< 4/t,/- -+- 2A), 4#w/-, + 2A).

622

A. JEFFREY AND T. KAKUTANI

FIG. 3.1.6. Sketch

of typical wave profiles.


A Solitary wave (v 0), C: oscillatory shock wave (0 < D: monotone shock wave (v >=
< 4#x//2 + 4/w/2 + 2A).

2A),

Numerical calculations for typical values of # and v were carried out by Grad-Hu [18], and an asymptotic solution for 0 < v2<< + 2A was obtained by Johnson [28] by matching a perturbed solitary wave solution with a lightly damped cnoidal wave.

4/tx//2

3.2. Stability of the steady solution. We consider, in this section, the stability of the Burgers shock wave and of the Korteweg-de Vries solitary wave [25]. These solutions are not only typical steady solutions of the equations considered in the preceding section, but are also obtainable in closed form in terms of elementary functions. This latter fact makes the mathematical analysis much simpler.

(a) The Burgers shock wave. With an appropriate rescaling and choice of the reference frame of coordinates, we may express the Burgers shock wave solution (3.1.23) in the following form:

(3.2.1)

Uo(X)

u tanh

2v

]
<<

> 0.

We superimpose a small disturbance v(x, t) upon this steady state solution:

(3.2.2)

Uo(X) + v(x, t),

Ivl

luol.
0 and linearizing it

Substituting (3.2.2) into the Burgers equation (3.1.22) with # with respect to v, we obtain the following equation for v:

(3.2.3)
Assuming that v(x, t)

c3v

c3--/+ Ux + -d-x v
ff

c3v

duo

1021)
v

Ox 2

f(x)g(t), it follows that

(3.2.4)

g(t) c exp(rt),

const.

WEAK NONLINEAR DISPERSIVEWAVES

623

and that f(x) is governed by the equation

(3.2.5)

v)-x 2-uo x-

/:f

/Uo f=0. +

As the disturbance should vanish at infinity, we must assume the following


boundary condition for f:

(3.2.6)

It is obvious from (3.2.4) that if Re () > 0 the disturbance will grow with time without bound. This means that the original steady state solution is unstable. On the other hand, if Re () < 0 the disturbance will eventually die away, which means that the steady state solution is asymptotically stable in the Lyapunov sense. If, in particular, Re () 0, then the original steady state solution is merely stable in the Lyapunov sense. By setting ux/(2v) y and 4w/u equation (3.2.5) and the boundary condition (3.2.6) reduce, respectively, to the following one-parameter system"

0 as

Ixl --,

(3.2.7)
(3.2.8)

dZf + 2 tanh ydf


dy

+(2sech

y-)f L(f)

f0 asyl.

Since the differential operator L in (3.2.7) is even in y, both odd and even functions of y can separately be solutions of (3.2.7). In fact, the required solutions have the form

(3.2.9)

foaa

sech y(k sinh ky sech y(k cosh ky

tanh y cosh ky),

where k + If, in particular, k 0, then foaa =-sechytanhy and fv, 0. Therefore, we must replace the even solution by fv,=o) sech y(1 -ytanhy). Similarly, if k 1, foaa 0 and fvn sechZy- In this case, foaa(= ) tanh y + y sech 2 y should be taken as the odd solution. Now necessary and sufficient conditions under which the solutions (3.2.9) satisfy the boundary condition (3.2.8) are given by Re (k)< 1, from which we obtain the condition

f,

tanh y sinh ky),

(3.2.10) (3.2.11)

Re () <

Remembering the definition of

{Im ()}2/4 N 0.
Re () < 0,

we have

which shows that the Burgers shock wave solution (3.2.1) is asymptotically stable. It should be noted that when k 1 (i.e., 0 or 0), the even solution sech2 the corresponding a stable whereas solution, fv,,(= x) Y represents merely odd solution foaa(=,) tanh y + y sech2 y is not an adequate solution because this cannot satisfy the boundary condition (3.2.8). Thus we may conclude that the Burgers shock wave is either asymptotically stable or stable with respect to infinitesimal disturbances.

624

A. JEFFREY AND T. KAKUTANI

Interesting results have been deduced about the asymptotic stability of traveling wave solutions to Burgers equation by IIin and Oleinik [21] and also by Peletier [56]. In a sense their work is the inverse of the results presented here since IIin and Oleinik show that for a certain class of initial data the solution to Burgers equation converges to a traveling wave. Peletier develops the argument further by giving an inequality determining the rate of convergence of the solution to the asymptotic traveling waveform related to (3.2.1).

(b) The Korteweg-de Vries solitary wave. By choosing an appropriate reference frame of coordinates together with an appropriate rescaling, the solitary wave solution (3.1.9) takes the form

(3.2.12)

Uo(X)

-u(1_=

3 seek2

,k

4--x),
0

uoo

> 0.

A similar procedure to that employed in the preceding subsection leads to a governing equation for the disturbance function f of the form
(3.2.13)

d3f
dy3

4(1

3 sech 2

df_(24sech y)yy

y tanh y +e)f

where y =_ again be given by

w/(uoo/41)x and

(-8a/uo)w@/oo. The boundary condition may


as lyl -- oz.

(3.2.14)

f --, 0
d2

Three independent solutions of this equation can be obtained as follows:

2(2 (3.2.15)

2) 2 e x +

4--2 {e(X-

eX[2(2

2) + 4 e

where the ). are the distinct roots of the cubic equation

sech y}

sech y{2(2

2)

2(2

1) tanh y + 2 tanh 2 y}], k 1,2,3,

(3.2.16)
If, in particular, 2

/3

4,

O.

0, 2 or 2, that is, if e 0, thereby representing a stable all then the are case, fk proportional to sech 2 y tanh y for k 1, 2, 3. Therefore, in this case, we must take our three independent solutions in the following form"

(3.2.17)

fx f2 f3

sechZy tanh y f0, say, 3yfo + tanh2 Y- 2 sech 2 y, 15yfo + 2 sinh2 y _31_ 7 tanh2 y

8 sech2 y.

It should also be noted that the cubic equation (3.2.16) has double roots, 2 2/x// when e -T-16/(3x/-). In this case two of the solutions given by (3.2.15) become identically equal to each other, say fl =- f2. Therefore, we must then choose three

WEAK NONLINEAR DISPERSIVE WAVES

625

new independent solutions as follows:

f
(3.2.18)

20(20

2)z eXO, + 4 __{eXO-1), sech y},


ay-

d2

f2
f3

2 Yfl + (32o

e{x- 1), sech y), 820 + 4) ex + 8-_.{ ay

23(23

2) 2 13 +

4--y2 {e(X3-1), sech y},


-4/w/

d2

or 4/x//, where 20 denotes either or -2/x//; and 23 is either or It is obvious that the cubic equation (3.2.16) according as 2o is has no triple roots. Now it can be easily seen that any linear combination of the solutions given by (3.2.15), (3.2.17) or (3.2.18) cannot satisfy the required boundary condition (3.2.14) except for thefo given in (3.2.17) which represents a stable solution. Thus we may conclude that the solitary wave solution of the Korteweg-de Vries equation is stable in the Lyapunov sense with respect to infinitesimal disturbances. The fact that the solitary wave solution is stable is also confirmed by numerical calculation. (See 3.3.) In fact, it has been shown numerically that the solitary wave is so stable that it behaves as though it were a single particle. It is on account of this that it is sometimes called a "soliton." A generalization of the stability analysis to a more general system of equations, which includes the Burgers equation and the Korteweg-de Vries equation as special cases, may be made using Lyapunovs direct method [24]. The stability of a class of periodic wave trains of water waves was investigated by Benjamin [1], although his analysis was not concerned directly with the cnoidal wave solution of the K orteweg-de Vries equation. He showed that most periodic waves are unstable with respect to a certain kind of infinitesimal disturbance and he found that the instability is due to a resonance mechanism between the original wave and the superimposed perturbations.

2/xf 2/x//J -2/x//.

3.3. Initial value problems. Ever since the derivation of the equation to which Korteweg and de Vries gave their name there has been relatively little attention paid to this equation. However, as we have already seen at the start of this paper,
in the last decade or so, it has turned out that a very wide class of nonlinear dispersive systems can be described by this equation or by one of its modified forms. Thus various attempts are now being made to obtain an analytical solution of the unsteady (time-dependent) Korteweg-de Vries equation. A remarkable transformation discovered by Miura et al. [14], [50] is indeed probably one of the major contributions in this direction. 9 The task is, however, not as easy as in the corresponding case of the Burgers equation, which can be transformed into a heatconduction or diffusion equation by a simple transformation known as HopfCole transformation [93, [20]. Fortunately, to assist with this task, we have the aid of high speed computers which help to bridge the gaps left when an analytical approach fails. Generally
An analytical approach to the problem has been made in a series of papers: The Kortewegde Vries equation and generalization [41], [50], [51], [61].

626

A. JEFFREY AND T. KAKUTANI

speaking, even a carefully designed numerical computation gives only a "discrete" knowledge of problems, but the results obtained are usually straightforward and self-explanatory. Moreover, a numerical result often suggests a clue to an analytical approach. In fact, a combination of numerical and analytical approaches, first called "synergetics" by S. Ulam [69], has played an important part in the last decade and will almost certainly continue to do so in the future. We now discuss some typical initial value problems for the Korteweg-de Vries equation which have been solved by numerical methods.

(a) The single solitary wave. One of the simplest and most obvious ways of checking whether the steady solitary wave solution is an asymptotic solution of the unsteady (time-dependent) K orteweg-de Vries equation is to examine an initial value problem in which the solitary wave solution itself provides the initial condition. We shall reproduce here briefly results obtained by Zabusky [69] (see also [70]). The problem is to solve the Korteweg-de Vries equation

(3.3.1)

cu

3 ct + Uxx + #x

Ou

c3u

0,

subject to the initial condition

(3.3.2)

u(x, O)

a sech 2

(x/D),

where D x//i21a/a, and the uniform state at infinity, u, has for convenience been chosen to be zero (cf. (3.1.9)). We choose the amplitude (or height) of the as 100 and 0.02, respectively; whence pulse a and the dispersive coefficient D 0.0069. In the numerical calculation which was carried out by Zabusky, the solitary wave initial condition was replaced by a periodic function comprising a series of well-separated identical solitary waves. In fact, the separation distance was taken to be 28.6D. It was found that during the time interval of 0.0024 time units, the initial pulse proceeded through a distance l l.4D with the average speed 33.18 units, which should be compared with the value 100/3 units in the steady state case (cf. (3.1.9)). The amplitude did not deviate from its original value by more than 0.19 This result suggests strongly that the steady solitary wave solution is indeed an asymptotic solution of the time-dependent Korteweg-de Vries equation.

xf

(b) The interaction of two solitary waves. It is already known that the propagation speed of a solitary wave is proportional to its amplitude (cf. (3.1.9)). Therefore, if initially we consider two solitary waves on an infinite line, with the larger to the left of and well-separated from the smaller, then the larger one will approach the smaller. To clarify this interaction, Zabusky [69] solved (3.3.1), again with 0.02 but subject to the following initial condition" Two pulses with amplitudes al 180 and a 2 80 are considered which are separated by the distance 12.0D1, where D refers to the first pulse. These are considered relative to the uniform state at infinity u -26, so that the second pulse remains fixed. The trajectories and amplitude histories of the two pulses are given in Fig. 3.3.1, from which we can see that the larger overtakes the smaller, accelerates during the overlapping process, and then emerges again relatively unaffected. The dotted

WEAK NONLINEAR DISPERSIVE WAVES

627

12

."
3OD,
DISTANCE
-2

180
50 100

AMPLITUDE

FIG. 3.3.1. Trajectory and amplitude history of two interacting pulses [69]

lines signify that amplitudes are ambiguous during interaction. When the two pulses are again separated from one another after the overlapping process and the separation distance is equal to the original separation distance 12.0D1, then the amplitudes of the first and second pulses differ from their original values only by 0.065 and 0.487 respectively. A remarkable fact is that the pulses preserve their initial identities (both shape and strength) to high numerical accuracy despite the nonlinear interaction, in other words, the solitary wave behaves essentially as though it were a single particle. As remarked, it was this property which gave rise to the name "soliton." Recently, Lax [43] confirmed by a rigorous analysis that two solitons preserve their identities in the above sense. This tendency will be observed more definitely in the following examples, where many solitons interact with one another.

(c) The sinusoidal function. Let us now proceed to a more general case where the initial condition is given by the sinusoidal function

(3.3.3)

u(x, 0)

cos nx.

Historically speaking, this was the first example of an initial value problem for the Korteweg-de Vries equation which was solved numerically by Zabusky and Kruskal [71]. They chose the dispersive coefficient of (3.3.1) to be 0.022. Thus, initially, the ratio of the dispersive term to the nonlinear term is extremely small, because

(3.3.4)

max

ax

Ux

0.004.
=o

628

A. JEFFREY AND T. KAKUTANI

If we neglect the dispersive term in (3.3.1), we obtain a simple hyperbolic equation

(3.3.5)

ct

uux

0,

whose implicit solution, subject to the initial condition (3.3.3), is given by


u cos r(x ut). (3.3.6) As is well known, the approximate solution u given by (3.3.6) tends to steepen its shape and eventually becomes multivalued at x 1/2 when t 1/rt (breakdown time). In Fig. 3.3.2, the numerical solutions of (3.3.1) with x// 0.022 are shown at 0. three different time stages. Curve A is the initial cosine function (3.3.3) at

Curve B is the solution at the breakdown time t. This shows the start of an oscillatory structure for x 1/2, which is obviously due to the effect of the dispersive term in (3.3.1). Curve C at 3.6t shows a fully developed oscillatory structure, or a train of solitons. In fact, each pulse of this train can be regarded as a soliton, because when it is "well"-separated from the neighboring pulses it moves with the "same" speed and has the "same" width as that of the steady solitary wave solution. Table 3.3.1 shows a comparison of the widths and speeds of solitons observed in the numerical solution at 3.6t to those calculated from the steady solitary wave solution (cf. (3.1.9)). For the first seven solitons, the agreements of both results are fairly good. It should also be noted that the locus of the maxima of these solitons is a straight line (note that the periodic initial condition has been employed).
u

(R)..--

U
-1.O

O.

1.O

1.S

Fo. 3.3.2. Numerical solution of the Korteweg-de Vries equation with condition u(x, 0) cos (=x). (See [69].) 0 (cos (=x)), A B: t,
C:t=

0.022 subject to initial

3.6tn.

WEAK NONLINEAR DISPERSIVE WAVES

629

TABLE 3.3.1 Comparison of the widths and speeds of solitons observed in the numerical solution at 3.6tB with those calculated from the steady solitary wave solution (see [71])
Soliton

Width (D)

Speed
observed calculated

number

observed

calculated

0.0455 0.0475 0.0492 0.0522 0.0567 0.0636 0.0769 0.099

0.0456 0.0476 0.0493 0.0516 0.0568 0.0639 0.0767 0.109

227 110 0 -99 169 -273 -361 -443

254 131 4 105 -202 -289 -354 -353

Figure 3.3.3 shows the trajectory of the maxima of nine solitons together with the amplitude history of the first soliton. It can be seen that each soliton moves along a straight trajectory when it is "well"-separated from the others. When solitons approach one another, their trajectories deviate from straight lines, they accelerate through one another, and then emerge again relatively unaffected. It is also observed that during the overlapping period, the joint amplitude of the interacting solitons decreases. This is a remarkable contrast to the case of linear interaction. At 1/2tR (t R 30.4t), all the odd numbered solitons coalesce at x 0.385 and the even numbered solitons coalesce at x 1.385. The resulting waveform for u is found to be very similar to that of the second harmonic of the initial cosine function. At tR (not shown in Fig. 3.3.3), all nine solitons coalesce and the resulting waveform for u is almost the same as that of the original cosine function. At 2tR, one can observe again a "near recurrence" though it is not as good as the first recurrence. For > tR, the successive coalescences become poorer due to solitons arriving more and more out of phase with each other, and then eventually the situation improves again when their phase relationships change favorably. The result thus obtained shows that the solitary wave is extremely stable (cf. 3.2) and preserves its own identity despite numerous nonlinear interactions.

(d) Gaussian distribution. Another interesting initial value problem for the K orteweg-de Vries equation was solved by Berezin and Karpman [6]. They considered an initial condition of the following form"

(3.3.7)

Uoqg(x/l) at

0,

where Uo is the amplitude and is a characteristic length scale for the initial disturbance. Introducing the new normalized variables

(3.3.8)

v=u/u o,

x/1,

Uot/1,

630

A. JEFFREY AND T. KAKUTANI

o.ste

.o

t/3

t/4

tR/5
tR/6

FIG. 3.3.3. Soliton trajectories on a space-time diagram beginning at hand diagram shows the amplitude history of the first soliton 0). (See [69].)

O.ltR

3.04tn.

The right-

we can express the Korteweg-de Vries equation (3.3.7) as follows"

(3.3.1) and the initial condition


0,

(3.3.9) (3.3.10)

8v
v

+ v- +

Ov

1 O3v
O-2

C3 3
0,

q9() at

where O- l(uo/) 1/2. Thus we arrive at a one-parameter family of similar solutions in the sense that the solutions of (3.3.1) subject to the initial condition (3.3.7) are similar to each other provided that the similarity parameter O- is fixed. It should be noted here that the soliton initial condition (3.3.2) gives

(3.3.11)

O-

O-,

x2,

D (cf. 3.3.3(a)). provided that we set Uo a and Actual numerical calculations were carried out by employing the Gaussian distribution function as the initial condition

(3.3.12)

qg()

exp (- 2),

since this is a typical symmetric function which tends to zero as Ix[ or [l tends to infinity. It was found that the behavior of the numerical solutions was completely different, according as O- >> O-s or O- << O-s. As shown in Fig. 3.3.4(a) and Fig. 3.3.4(b), for sufficiently large values of O- compared with O-z, the initial disturbance (3.3.12)

WEAK NONLINEAR DISPERSIVE WAVES

631

1.0

2.0

u(x, O)

FIG. 3.3.4(a). Numerical solutions of the Korteweg-de Vries equation subject to initial condition Uo exp {-(x/l)2}. a 5.9 > ac. (See [6].)

splits into a number of solitons in the course of time or, more precisely, the initial disturbance (3.3.12) breaks up into two solitons when 4 < a < 7, into three when 7 < a < 11, into four when a 11, into six when a 16, and thereafter the number of emergent solitons increases indefinitely with increasing numerical values of a (cf. 3.3(e) below). On the other hand, if a << as, no solution breaks up into solitons at all, but the solutions for tr << as exhibit rapidly oscillating wave packets. For certain intermediate values of a a mixed type of solution was found which consists of a leading soliton and an oscillating "tail," very similar to the rapidly oscillating wave packet (Fig. 3.3.4(c)).

(e) Amplitudes of the emergent solitons. Starting from the conservation laws associated with the Korteweg-de Vries equation, Berezin and Karpman 6] succeeded in estimating the amplitudes of solitons which emerge from the initial disturbance given in (3.3.10). It is known that the Korteweg-de Vries equation has an infinite number of conservation laws [51], the first three of which are expressed as follows:

(3.3.13)

tQm( ,r)

OPm( "r)

=0,

m= 1,2,3,

632

A. JEFFREY AND T. KAKUTANI

/
10

1.0

2.0

3.0

u(x, O)

FIG. 3.3.4(b). Numerical solutions of the Korteweg--de Vries equation subject to initial condition Uo exp {-(x//)2}. tr 16.5 > ac. (See [6].)

where

v,

= e-

+ lv l(cv lv4 /--1 "--i -


e=5

102/9

0.-

The first one is the Korteweg-de Vries equation (3.3.9) itself written in conservation form. The second and third can be obtained from (3.3.9) by multiplying each term by v and v respectively. These three conservation laws were first found by Whitham [67].

cv

?vcv)

llcv

WEAK NONLINEAR DISPERSIVE WAVES

633

--_..o

v V

V jo
Qm(, .c) d

o.,

.o-.

u(x, O)

FIG. 3.3.4(c). Numerical solutions of the Korteweg-de Vries equation subject to initial condition Uo exp {-(x//)2}. a 1.9 < ac. (See [6].)

If the initial disturbance is such that it tends to zero as ]l --* oo (the Gaussian distribution (3.3.12) is an example), it follows from (3.3.13) that

(3.3.15)

Sm

Qm(, 0) d,

1,2,...

are time-invariant or conserved quantities. Now let us assume that the initial disturbance (3.3.10) breaks up into N solitons. For sufficiently large values of r, each soliton hardly overlaps its neighbors. Therefore, for r >> 1, we may assume that
N

v(, )

(3.3.16)

--I
N

Z v)(, ),

Sm"
where

E S,

(3.3.17)
Qm

Qmlv

sech2{,

634

A. JEFFREY AND T. KAKUTAN!

from which we obtain the following equations for the amplitudes of the solitons
qm d, m 1,2,.... Qm(, O) d Os The right-hand sides of these equations can be expressed in terms of the given initial condition (3.3.10) by using the conservation laws (3.3.14). For example, if N 2, we find the amplitudes of two solitons as follows:

(3.3.18)

{a(r)}m-1/2

(3.3.9)

{al,} 1/

1/2[1
o() a,

-+ {(4/31) /3}

where a (1), a (2) correspond, respectively, to the

(3.3.20)

and

signs, and

Since the right-hand side of (3.3.19) must be real and positive (note that all the square roots in (3.3.19) have been taken in the arithmetic sense), we have e2 < 31 < 4e2, from which we obtain a necessary condition for the initial disturbance (3.3.10) to break up into two solitons
o-

(3.3.21)

ffc

< cr < 2ac,

{o()}

e() a

2.0--

X
1.5

1.0

X X

0.5--

<

FIG. 3.3.5. Dependence of the amplitude a of the solitons on the similarity parameter tr [6]. Crosses "experimental" values, Curve I: decay into two solitons, Curve II: decay into three solitons.

WEAK NONLINEAR DISPERSIVE WAVES

635

In particular, for the Gaussian distribution (3.3.12), (3.3.21) gives ac 4, which is of the same order of magnitude as that of as (cf. (3.3.11)). Thus the similarity parameter as associated with the soliton initial condition (3.3.2) indeed provides a rough criterion whether or not the initial disturbance (3.3.12) breaks up into solitons. When a > ac, the initial disturbance (3.3.12) always splits into solitons. The crosses in Fig. 3.3.5 show the "experimental" values of amplitudes which were obtained from numerical calculations for various values of the parameter a. On the other hand, the "theoretical" values calculated from (3.3.18) for N 2 and N 3 are also shown in Fig. 3.3.5, where Curve I is the two branches of (3.3.19) and Curve II is obtained numerically by solving (3.3.18) for N 3. Both results agree with each other very well except for the critical values of a, i.e., in the vicinity of a 2ac, where the "experimental" points already go over to Curve II. By extending the above analysis, Karpman [33] estimated a "distribution function" for the amplitude of the emergent solitons as r oe (g 0 with Uo and fixed; note that a l(uo/p)l/2). Let us denote by f(a)da the number of solitons whose normalized amplitudes with respect to Uo fall into the interval (a, a + da), and assume that the initial disturbance (3.3.10) has the form of a symmetric
positive pulse"

qg() qg(-) > 0, qg(oe) 0. (3.3.22) By solving an integral equation for f(a), which is simply the asymptotic form of (3.3.18) as a oe, Karpman obtained the result (3.3.23) f (a) l[uo/(48#)]l/2a 1/2.
Thus the total number of emergent solitons is given by

(3.3.24)

f(a) da

l[uo/(61a)] 1/2

(It was found in the course of analysis that the upper limit of soliton amplitude is 2uo.) The above result (3.3.23) was obtained for an initial disturbance of the form q)() sech 2 (Note that this initial condition does not necessarily imply the soliton initial condition discussed in 3.3(a). In fact, qg() sech 2 is equivalent to the soliton initial condition if and only if we choose a as.) But Karpman also
asserts that qualitatively similar results are also valid for other initial disturbances which have the shape of a symmetrical pulse. In concluding this section, we must make an important remark concerning the relationship between the generation of solitons and the initial conditions. We saw in 3.3(d) that when r < r the initial disturbance (3.3.12) does not generate any soliton at all except for the leading soliton accompanied by a rapidly oscillating tail when r a (cf. Fig. 3.3.4(c)) and only rapidly oscillating wave packets when a << ac. However, it should also be noted that there are solutions subject to a certain class of initial condition which do not split into solitons at all. For example, if the "area" of the profile of the initial disturbance is negative, that is, if

<

v(, o) d

e() d __< 0,

636

A. JEFFREY AND T. KAKUTANI

then it is impossible for solutions subject to these initial conditions to break up into solitons. This is because the area is nothing but S which is one of the conserved quantities Sin, whereas the profile of solitons given by (3.3.17) has positive area. Note that we have assumed here that # > 0. If # < 0, the area of the profile of solitons is negative. Consequently, any initial disturbance whose profile has positive area cannot produce solitons. The same general argument applies not only to $1 but also to the other conserved quantities Sin, with m > 2.

3.4. Existence and uniqueness. Because the mathematical issues raised when existence and uniqueness of the solution of the K orteweg-de Vries equation are studied are very different from those involved when examining the physical properties of the solution, we have chosen to discuss them last. The results we now present are in no sense the best possible, since the form of analysis used has been chosen to illustrate some of the techniques which may be employed rather than to obtain sharp theorems. As always, existence is much harder to establish than uniqueness so we prove uniqueness first in the special case of strongly periodic initial data; to be specific, for initial data which is periodic together with its first two spatial derivatives. We start from the K orteweg-de Vries equation

(3.4.1)

U 3U U + + tSx3 ct

Ux

and consider a solution u defined on [0, 1] conditions

[0, T] satisfying the periodic initial

(3.4.2) (3.4.3) (3.4.4) (3.4.5)


where 0

u(x, O) u(O, t)
c3x

Uo(X),
u(1, t),

u (o, t)

@ (1, t),

c3x2

c2u c2U(o t) x2(1, t),

__< x __< 1,0=<t__< T. Assume that there is another solution v satisfying (3.4.1) and subject to the same initial conditions (3.4.2) to (3.4.5) so that
(3.4.6)
Then, setting w equations in w,
u

c + vxx +/-5x3
c3w

0.

v and subtracting the two equations leads to the linear

(3.4.7)

u wux
+

/3

O3W

o.

WEAK NONLINEAR DISPERSIVE WAVES

637
then

Multiplication of this equation by w and integration with respect to x over [0, yields

tLJo (3.4.8)

/[1 (l w a

xl ffaw fol
+
wu

dx

w2

Ov

dx

O3w j w-x
o

dx

O.

Integration by parts, use of the result w initial conditions establishes that

v and appeal to the periodic

w2

dx

and

so that (3.4.8) now becomes

(3.4.9)
Writing

at

W2

dx

W2

dx

W2

dx

(3.4.10)

Ilwl

w 2 dx

and taking the modulus of (3.4.9) we arrive at the differential inequality

(3.4.11)

-dllwll <= Ilwl max xx(X,)

max

xx(X,t)

To proceed further we need a priori estimates for max Ic?u(x, t)/cxl and max Icv(x, t)/cxl. These may be obtained by use of Sobolevs lemma [34], [58] in the following form. SOBOLEVS LEMMA. Let p, q be integers with 0 <= q < p. Then for every constant > 0 there exists a constant M(e), such that for all functions u sufficiently differentiable on the x-interval [0, 1],
max cqu
O<_x=l

<e

dx

+ M(e)

bl

dx.

In addition to this lemma we also require the following identity [51] satisfied by a solution of the Korteweg-de Vries equation"

(3.4.12)

u4

12#u

+ -#

dx

C,

const.

This result is essentially due to Whitham [67], though it is one of the family of conservation laws found by Miura et al. [51].

638

Using the notation 11. as defined in (3.4.10), and taking the modulus of identity (3.4.12), we arrive at the inequality

36#2

Now Sobolevs lemma gives estimates of max [ul 2 and thus max lul in terms of u 2 dx and j (cu/c3x) 2 dx, from which we may immediately deduce that there exists a constant K such that

A. JEFFREY AND T. KAKUTANI

-Sx2] ff ic32ul2

dx

<__ CI

ff

u 4 dx

+ 1211
u dx

f/ 6U)2
u

dx

N Cl + maxlu

+ 12 maxul

dx.

(3.4.13)
such that

-XXl
max

dx

<__ K.

A further application of Sobolevs lemma then shows that there exists a constant K
(3.4.14)

A similar argument gives rise to a corresponding inequality for the derivative v(x, t)/x,
(3.4.15)
max

Ux(X, t)

<K

Incorporating these estimates into differential inequality (3.4.11) finally gives

(3.4.16)

d- Ilwl
0

3K w

After integration differential inequality (3.4.16) leads to the comparison solution

(3.4.17)
for

w(x, t)

IIw(x, 0) e 3Kt

IIw where 0 =< < T. However w(x, 0) 0 by virtue of initial condition (3.4.2), which must be satisfied both by u and v, so that IIw(x, 011 0 for all 0 =< T. This can only be true if w(x, t) O, showing that u(x, t) v(x, t) in [0, 1] =< x [0, T]. We have established uniqueness on the assumption that a solution
exists.

To prove existence of the solution u subject to the periodic initial conditions (3.4.2) to (3.4.5) we employ the results and the method of regularization of parabolic equations as exploited by Lions [46], [48] and Temam [64] by first augmenting (3.4.1) by the addition of a term e(c34u/c3x4). Following their work we denote by f the interval (0, 1), and by H(f), with integer s >= 0, the Sobolev space
v L2(), 0 =< j =< s VIV e L2(n), xje
dj

WEAK NONLINEAR DISPERSIVE WAVES

639

equipped with the norm


dJr

where
that

I.

signifies the LZ(f)-norm.

Let Uo be a family of infinitely differentiable functions defined on [-0, 1] such

dJuo.(O)
dx

dJuo(1)
dx
--*

for all j >_ 0, and let Uo converge weakly to u o in the space H 1(") as e can be shown [64] that for every fixed e > 0 there exists a solution

0. Then it satisfying

(3.4.18)
(3.4 19)

u
c3u,:
8t

L(0, T; L(f)) fl L2(0, T; H2(f)),

u-x +
Uo.(x),
t)

-bx o,
j=0,..., 3.

(3.4.20) (3.4.21)

u(x, O)

c3Ju (0 cx

c3iu -8-x (1, t),


L2(0, T; L(fl)),

By virtue of property (3.4.18) we have

and also

--

L2(0, T; HI())

(3.4.22)

U--X

L2(0, T; L2()).

Here Lz(S;E) is the space of functions defined on S, a locally compact measure space with measure # _>_ 0, with values in a Hilbert space such that

lf(t) I dl(t)

and L(S;E) is the space of functions f defined on S with values in E and essentially bounded, so

From (3.4.19) we see that u(Su/Sx) may be written as a linear combination of derivatives linear in u, so that taken together with property (3.4.22) and the regularity properties of linear parabolic equations this shows that

(3.4.23)

ct

e L2(0, T; L2(fl)),

U,

L2(0, T; H4()).

Developing this argument further then shows that f x [0, T].

u is infinitely differentiable in

640

A. JEFFREY AND T. KAKUTANI

By ingenious reasoning involving Sobolev inequalities and interpolation spaces 47], [49], Temam [64] has established the following a priori estimates involving u
(3.4.24) lUILo(O,T;L2(n)) <= const., (3.4.25) IC3U/c3XlLO,T;L=(a)) <= const., (3.4.26) -IOZue/XZIL2(O,T;LZ(f)) < const., (3.4.27) /-I3U2X31LO,T;L2, < const., (3.4.28) IC3Uf3tIL2O,T;V2,) const., where V2 is the subspace of HS(f) comprising functions v for which
djv

(0)=
and

-d-x (1),

0, 1,

..., s

1,

Vz is the dual space.

The existence of the solution u of (3.4.1) then follows from these results as a consequence of passing to the limit as e --, 0. From results (3.4.24), (3.4.25) and (3.4.28) we see that

weakly in L(O, T; L2(f)),


c3x

cu
t

cx cu

weakly in L(0, T; L2()),

weakly in L2(0, T; Vz).

The first two results imply that u u weakly in L(0, T; HI()), and as the injection of Hl(f) in L2() is compact, it follows from the third result that

strongly in L2(0, T; L2()). The passage to the limit in (3.4.18) to (3.4.21) now offers no difficulty because

weakly in L(O, T; L l(r)),


and so (3.4.19) tends in the limit to (3.4.1). Other existence and uniqueness proofs are possible which are either more or less restrictive than the one outlined here and may involve different techniques. By way of example, in the case of periodic initial conditions we refer the interested reader to Sj6berg [57] for an approach to this problem using finite difference methods. This paper does not, however, present a definitive answer to the problem since it still contains a number of unresolved difficulties. Perhaps the most significant of these arises in connection with the attempt to obtain a global existence theorem by a stepwise process in time, for it is there that an appeal is made to the various conservation laws reported first by Whitham [67] and later in greater generality by Miura et al. [51]. It will suffice here to say that while Sj6berg demonstrates the fact that a certain type of solution u(x, t) obtained by the

WEAK NONLINEAR DISPERSIVE WAVES

641

Arzela-Ascoli lemma from the limit of a sequence of interpolation functions based on a discrete approximation is differentiable once with respect to and three times With respect to x, with Ux(X, 0), uxx(x, 0), U,x(X, 0) L2(), the derivation of the conservation laws used requires the use of higher order derivatives than he has shown his solution to possess. It seems probable that to establish the necessary differentiability properties for his solution will require differentiability properties for Uo(X) that are unnecessarily restrictive. Although it would seem that a similar criticism could be offered in connection with the use of conservation law (3.4.12) to prove uniqueness in the present paper, the situation here is actually rather different. In point of fact this form of uniqueness proof was only adopted to illustrate an application of the Sobolev lemma in conjunction with one of the conservation laws associated with the K orteweg-de Vries equation. Had the method of parabolic regularization been employed further, then the use of (3.4.12) could have been avoided in the uniqueness proof so that the objection would no longer arise. For a proof of this type involving weaker periodicity conditions on the initial data in which (3.4.5) is no longer required see Temam [64].

Acknowledgments. We would like to take this opportunity to express our gratitude to the editor and referees for a number of valuable criticisms which have led to the clarification of various points and to the correction of an error.
REFERENCES

1] T. B. BENJAMIN, Instability ofperiodic wave trains in nonlinear dispersive systems, Proc. Roy. Soc. Ser. A, 299 (1967), pp. 59-75. Internal waves of permanent form in fluids of great depth, J. Fluid Mech., 29 (1967), [2]
pp. 559-592.

[3] D. J. BENNEY, Long waves on liquid films, J. Math. and Phys., 45 (1966), pp. 150-155. [4] --, Long nonlinear waves in fluid flows, Ibid., 45 (1966), pp. 52-63. [5] Y. A. BEREZIN AND V. I. KARPMAN, Theory of non-stationary finite-amplitude waves in a lowdensity plasma, Zh. Eksper. Teoret. Fiz., 46 (1964), pp. 1880-1890 Soviet Physics JETP, 19 (1964), pp. 1265-1271. [6] ----., Nonlinear evolution of disturbances in plasmas and other dispersive media, Zh. Eksper. Teoret. Fiz., 51 (1966), pp. 1557-1568 Soviet Physics JETP, 24 (1967), pp. 1049-1056. [7] M. BORN AND T. VON KARMAN, Ober Schwingungen in Raumgittern, Phys. Z., 13 (1912), pp.
297-309.

[8] P. F. BYRD [9]

[10] I11] [12]


[13] [14] [15] I16]

AND M. D. FRIEDMAN, Handbook of Elliptic Integrals Jbr Engineers and Physicists, Springer, Berlin, 1954. J. D. COLE, On a quasi-linear parabolic equation occurring in aerodynamics, Quart. Appl. Math., 9 (1951), pp. 225-236. P. DEAN, Lattice dynamics, R. F. Wallis, ed., Pergamon Press, New York, 1965. Atomic vibrations in solids, J. Inst. Math. Appl., 3 (1967), pp. 98-165. P. DEBYE, VortGge fiber die kinetische Theorie des Materie und der Elektrizitt, Teubner, Leipzig, Germany, 1914. N. C. FREEMAN AND R. S. JohNsoN, Shallow water waves on shear flows, J. Fluid Mech., 42 (1970), pp. 401-409. C. S. GARDNER, J. M. GREEN, M. D. KRUSKAL AND R. M. MIURA, A method for solving the Korteweg-de Vries equation, Phys. Rev. Lett., 19 (1967), pp. 1095-1097. C. S. GARDNER AND G. K. MORIKAWA, Similarity in the asymptotic behaviour of collision-free hydromagnetic waves and water waves, Rep. NYO 9082, Courant Institute, New York, 1960. ----, The effect of temperature on the width of a small-amplitude solitary wave in a collision-free plasma, Comm. Pure Appl. Math., 18 (1965), pp. 35-49.

642

A. JEFFREY AND T. KAKUTANI

[17] C. S. GARDNER AND C. H. Su, Derivation of the Korteweg-de Vries equation, 1966 Annual Rep., Princeton Univ. Plasma Phys. Lab., Matt-Q-24, Princeton, 1967. [18] H. GRAD AND P. N. Hu, Unified shock profile in a plasma, Phys. Fluids, l0 (1967), pp. 2596-2602. [19] K. HAMER, The three variable expansion procedure, in preparation. [20] E. HOPE, The partial differential equation u + uu, #Uxx, Comm. Pure Appl. Math., 3 (1950),
pp. 201-230.

[21] A. M. ILIN

O. A. OLEINIK, Asymptotic behaviour of solutions of the Cauchy problem for of time, Mat. Sb., 51 (1960), pp. 191-216. [22] A. JEFFREY, The development of singularities of solutions of nonlinear hyperbolic equations of order greater than unity, J. Math. Mech., 15 (1966), pp. 585-598. The evolution of discontinuities in solutions of homogeneous nonlinear hyperbolic equations [23] having smooth initial data, Ibid., 17 (1967), pp. 331-352. [24] ---, Stability of parabolic systems, Indiana Univ. Math. J., to appear. [25] A. JEFFREY AND T. KAKUTANI, Stability of the Burgers shock wave and the Korteweg-de Vries soliton, Ibid., 20 (1970), pp. 463--468. [26] A. JEFFREY AND T. TANIUTI, Nonlinear Wave Propagation, Academic Press, New York, 1964. [27] R. S. JOHNSON, Nonlinear waves in fluid-filled elastic tubes and related problems, Doctoral thesis, Univ. of London, London, 1969. A nonlinear equation incorporating damping and dispersion, J. Fluid Mech., 42 (1970), 28] pp. 49-60. [29] T. KAKUTANI, Effect of an uneven bottom on gravity waves, J. Phys. Soc. Japan, 30 (1971), pp. 272-276. (See errata, Ibid., 30 (1971), p. 593.) [30] T. KAKUTANI AND T. KAWAHARA, Weak ion-acoustic shock waves, Ibid., 29 (1970), pp. I0681073. [31] T. KAKUTANI AND H. ONO, Weak nonlinear hydromagnetic waves in a cold collision-free plasma, Ibid., 26 (1969), pp. 1305-1318. [32] T. KAKUTANI, H. ONO, T. TANIUTI AND C. C. WEI, Reductive perturbation method in nonlinear wave propagation. II." Application to hydromagnetic waves in cold plasma, Ibid., 24 (1968), pp. 1159-1166. [33] V. I. KARPMAN, An asymptotic solution of the Korteweg-de Vries equation, Phys. Lett., 25A (1967), pp. 708-709. [34] T. KATO, Perturbation Theory jbr Linear Operators, Springer, Berlin, 1966. [35] T. KAWAHARA, Oblique nonlinear hydromagnetic waves in a collision-free plasma with isothermal electron pressure, J. Phys. Soc. Japan, 27 (1969), pp. 1331-1340. Weak nonlinear magneto-acoustic waves in a cold plasma in the presence of effective 36] electron-ion collisions, Ibid., 28 (1970), pp. 1321-1329. [37] A. P. KAZANTSEV, Stationary plasmaflow in a magnetic field, Zh. Eksper. Teoret. Fiz., 44 (1963), Soviet Physics JETP, 17 (1963), pp. 865-868. pp. 1283-1288 E38] P. J. KELLOGG, Solitary waves in cold collisionless plasma, Phys. Fluids, 7 (1964), pp. 1555-1571. [39] H. KEVER AND G. K. MORIKAWA, Korteweg-de Vries equation for nonlinear hydromagnetic waves in a warm collision-free plasma, Ibid., 12 (1969), pp. 2090-2093. [40] D. J. KORTEWEG AND G. DE VRIES, On the change of./brm of long waves advancing in a rectangular channel, and on a new type of long stationary waves, Philos. Mag., 39 (1895), pp. 422-443. [41] M. D. KRUSKAL, R. M. MIURA, C. S. GARDNER AND N. J. ZABUSKY, Korteweg-de Vries equation and generalization. V." Uniqueness and nonexistence of polynomial conservation laws, J. Math. Phys., (1970), pp. 952-960. [42] P. D. LAX, Development of singularities of solutions of nonlinear hyperbolic partial differential equations, Ibid., 5 (1964), pp. 611-613. Integrals of nonlinear equations of evolution and solitary waves, Comm. Pure Appl. Math., [43] 21 (1968), pp. 467-490. [44] M. J. LIGHTHILL, Group velocity, J. Inst. Math. Appl., (1965), pp. 1-28. On waves generated in dispersive systems to travelling effects, with applications to the [45] dynamics of rotatingfluids, J. Fluid Mech., 27 (1967), pp. 725-752. [46] J. L. LIONS, Equations diffOrentielles opOrationnelles et problkms aux limites, Springer, Berlin, 1961. Problkmes aux limites dans les Oquations aux drives partielles, Montreal Univ. Press, [47] Montreal, 1962.
AND

some quasi-linear equations for large values

WEAK NONLINEAR DISPERSIVE WAVES

643

[48] [49] [50]


[51]

, Sur certaines Oquations paraboliques non linOaires, Bull.

[52] [53]
[54]

[55] [56]

[57]
[58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68]
[69]

[70] [71]

Soc. Math. France, 93 (1965), pp. 155-175. J. L. LONS AYD J. PEEaRE, Sur une classe despaees dinterpolation, Inst. Hautes ltudes Sci. Publ. Math., 19 (1964), pp. 5-68. R. M. MURA, The Korteweg-de Vries equation and generalization. I." A remarkable explicit nonlinear transformation, J. Mathematical Phys., 9 (1968), pp. 1202-1204. R. M. MIURA, C. S. GARDNER AND M. D. KRUSKAL, The Korteweg-de Vries equation and generalization. H." Existence of conservation laws and constants of motion, Ibid., 9 (1968), pp. 1204-1209. A. MIZUTANI AND T. TANIUTI, Oblique hydromagnetic waves in a cold plasma, Phys. Fluids, 12 (1969), pp. 1167-1172. K. W. MORXON, Finite amplitude compression waves in a collision-free plasma, Ibid., 7 (1964), pp. 1800-1815. E. Oa AYD R. L. SUDAN, Nonlinear theory of ion-acoustic waves with Landau damping, Ibid., 12 (1969), pp. 2388-2394. , Damping of solitary waves, Ibid., 13 (1970), pp. 1432-1434. L. A. PELETIER, Asymptotic stability of travelling waves, IUTAM Syrup. on Instability of Continuous Systems, Springer-Verlag, Berlin, 1971, pp. 418-422. A. SJOBERG, On the Korteweg-de Vries equation, existence and uniqueness, J. Math. Anal. Appl., 29 (1970), pp. 569-579. S. L. SOBOLEV, Applications of Functional Analysis in Mathematical Physics, vol. 7, Translations of Mathematical Monographs, Amer. Math. Soc., Providence, 1963, p. 56 et. seq. J. J. SaOER, Water Waves, Interscience, New York, 1957. P. A. STURROCK, Kinematics of growing waves, Phys. Rev., 112 (1958), pp. 1488-1503. C. H. Su AYD C. S. GARDYER, Korteweg-de Vries equation and generalization. III." Derivation of the Korteweg-de Vries equation and Burgers equation, J. Mathematical Phys., 10 (1969), pp. 536-539. R. N. SUDAY, Classification of instabilities from their dispersion relations, Phys. Fluids, 8 (1965), pp. 1899-1904. T. TANIUTI AND C. C. WEI, Reductive perturbation method in nonlinear wave propagation./, J. Phys. Soc. Japan, 24 (1968), pp. 941-946. R. TEMAM, Sur un problkme non linOaire, J. Math. Pures Appl., 48 (1969), pp. 159-172. S. M. ULArVl, A Collection of Mathematical Problems, John Wiley, New York, 1960. H. WASHIMI AND T. TANIUTI, Propagation of ion-acoustic solitary waves of small amplitude, Phys. Rev. Lett., 17 (1966), pp. 996-998. G. B. WHIHAM, Non-linear dispersive waves, Proc. Roy. Soc. Ser. A, 283 (1965), pp. 238-261. N. J. ZAatSY, Exact solution for the vibrations of a nonlinear continuous model string, J. Mathematical Phys., 3 (1962), pp. 1028-1039. A synergetic approach to problems of nonlinear dispersive wave propagation and interaction, Proc. Symp. Nonlinear Partial Differential Equations, W. Ames, ed., Academic Press, New York, 1967, pp. 223-258. Solitons and bound states of the time independent Sehr6dinger equation, Phys. Rev., 168 (1968), pp. 124-128. N. J. ZABUSY AND M. D. KRtSAL, Interaction of solitons in a collisionless plasma and the recurrence of initial states, Phys. Rev. Lett., 15 (1965), pp. 240-243.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Das könnte Ihnen auch gefallen