Existence of Solitary Internal Waves

in a Two-Layer Fluid of Infinite Height

S. M. Sun

Department of Mathematics

Virginia Polytechnic Institute

and State University

Blacksburg, VA 24061

19960415 030
Abstract

This paper concerns the existence of internal solitary waves moving

with a constant speed at the interface of a two-layer fluid with infinite
height. The fluids are immiscible, inviscid, and incompressible with con-
stant but different densities. Assume that the height of the upper fluid
is infinite and the depth of the lower fluid is finite. It has been formally
derived before that under long-wave assumption the first-order approx-
imation of the interface satisfies the Benjamin-Ono equation, which has
algebraic solitary-wave solutions. This paper gives a rigorous proof of
the existence of solitary-wave solutions of the exact equations govern-
ing the fluid motion, whose first-order approximations are the algebraic
solitary-wave solutions of the Benjamin-Ono equation. The proof relies
on estimates of integral operators using Fourier transforms in L2(R)-
space and is different from the previous existence proof of solitary waves
in a two-layer fluid with finite depth.

DTXC QUALITY INSPECTED 5

 1. Introduction

This paper deals with internal waves of finite amplitude in a stable two-fluid system
with infinite height. The two fluids are immiscible, inviscid, and incompressible with
different but constant densities. The lower boundary is horizontal and no upper boundary
exists. Let the depth of the lower fluid be h in an equilibrium state and the densities of
the lower and upper layers of the flow be p~,p+, respectively, with p~ > p+ > 0. It is of
interest to find an internal solitary wave moving with a constant velocity U at the interface.
This problem was first stiidied by Benjamin using long-wave theory [4]. In the long-wave
analysis, the lengths of waves at the interface are assumed to be considerably larger than
h, but no long-wave approximation is possible for the motion of the upper fluid since the
height of the upper fluid is infinite. In reference to an (x*,t/*)-coordinate system moving
with the same speed U, the fluid motion is steady. It is found in [4] that to have long waves
at the interface, U must be near a critical wave speed UQ = {{p~ — p+)gh/p~)1'2, where
g is the gravitational acceleration constant. Assume that (UQ/U)2 = 1 + 71 e with e > 0
a small long-wave parameter and 71 < 0 a fixed constant and the long-wave expansion of
the solution in the lower fluid is used. A model equation, called Benjamin-Ono equation,
can be derived as follows,

£W,-™.-(£)-., d,
where
1 r+oo y+00

and T]*(x) is the deviation of the interface from the equilibrium state. The equation (1)
has a solitary-wave solution
2 7 3

5(x) (2)
= * + {h,? '
where a = —(471 /3), p. = —(p+/p~'yi) and x is a stretched horizontal variable with x = ex*.
The Benjamin-Ono equation was first derived in [4] for very general stratified fluids
with infinite depth and the two-fluid flow is one of these cases. Also the algebraic solitary-
wave solution (2) was obtained using Fourier transform technique. At the same time, Davis
and Acrivos [11] obtained the equation in an experimental and numerical study of the wave
phenomena. Then Ono [19] derived a corresponding time-dependent equation, which was
later studied using inverse scattering method [9, 12] and has multi-soliton solutions [17].
Maxworthy [18] carried out experiments again to show the existence of internal solitary-
waves and found that the waves are quite stable. Maslowe and Redekopp [16] generalized
the Benjamin-Ono equation to nonlinear waves in stratified shear flows and obtained more
complicated integral model equations. Then Benjamin [6] considered a two-layer fluid
with large surface tension at interface and derived the equation (1) with an extra term
of third-order derivative. Recently Amick and Toland [2, 3] proved that the solitary-wave
solution (2) and its translations in the ^-direction are the only solutions of (1) which decay
to zero as x approaches infinity. Bennett et al. [8] obtained the stability of solitary waves
of the Benjamin-Ono equation with time dependence by a similar technique used in the
proof of stability of solitary waves for the KdV equation [5, 10]. However, we note that
the solitary-wave solution (2) is a solution of the Benjamin-Ono equation, which is an
approximate equation of the exact equations governing the fluid motion. It is importment
to know whether the solitary-wave solution (2) is an approximation of a solution of the
exact equations. The corresponding problem for two-fluid flows with finite depth has been
studied intensively in last fifteen years. The model equation for the interface is so-call
KdV equation, which also has solitary-wave solutions. It has been proved by numerous
literatures that the solitary-wave solution of the KdV equation is an approximation of
a solitary-wave solution of the exact equations. The reader is referred to an article by
Benjamin et al. [7] for a general treatment of such problems and the references therein. As
pointed out in [7], the question whether the solitary-wave solution (2) is an approximation
of a solution of the exact equations for a two-fluid flow with infinite depth needs to be
answered.

This paper gives a rigorous proof of the existence of solitary waves at the interface
for a two-layer fluid with infinite depth, whose first order approximations are the solitary
waves given by (2). The result can be stated as follows. Assume that the depth of lower
fluid is h at infinity and the height of the upper fluid is infinite. The densities are p+ and
p~ for the upper fluid and lower fluid, respectively. If the wave speed U on the interface
satisfies

U = ((/T- - p+)flfÄ/(/»"(l+7iO))1/2

with 7i < 0 a fixed, constant and e > 0 a small parameter, then there is a solitary-wave

solution for the exact equations governing the two-fluid flow. Assume h + i]*(x*) is the
distance between the rigid bottom and the interface. Then the form of the interface for

the solitary-wave solution satisfies

where a =-(471/3),^ =-(p+//9~7i), -:

s
J2 sup |(1 +x2)(<TY*(x)/dxi)\ < Ke:

and K is independent of e but may depend on 5. Therefore the solitary-wave solution

(2) of the Benjamin-Ono equation (1) is the first-order approximation of a solitary-wave
solution of the exact equations.
The idea of the proof can be summarized as follows. Since the interface of the two
fluids is part of the solution, the domain of the fluids is unknown. We use Levi-Civita's
method to transform the unknown domain into a fixed domain using velocity potential
and stream function as independent variables. The main advantage of Levi-Civita's trans-
formation is to put all the nonlinear terms in the boundary conditions rather than in
the differential equations. It has been shown that such transformation is one to one and
onto if the amplitude of the solutions is small [20]. Then we use Fourier transform to in-
vert the differential equations with nonlinear boundary conditions into integral equations.
Thus finding a solution of the differential equations becomes looking for a fixed point of
corresponding integral operators. However, the kernels in the integral operators have a
singularity. In order to remove the singularity, a solvability condition must be satisfied.
Next we introduce an unknown function and obtain an integro-differential equation for this
function from the solvability condition. By using some techniques in operator theory, we
are able to transform the integro-differential equation into an integral equation. By taking
advantage of the fact that the nonlinear terms only appear at the boundaries and obtain-
ing the estimates of the integral operators, we can show that the operators are contraction
in certain Banach spaces. The contraction mapping theorem implies the existence of the
solution. We note here that the special forms of the equations in the two-fluid flow are
important to our approach. The method also works without many changes for a two-fluid
system with finite height in upper fluid and infinite depth in lower fluid, which has been
studied by Amick [1] (it was known by the author after the major work in this paper was
completed). Amick used a different formulation of the problem and a different approach as
well to prove the existence of the solitary wave solutions. Kirchgässner [15] also considered
surface waves on a layer of fluid with infinite depth. However, we note that the argument
given here does not apply to the proof of the existence of solitary waves in continuously
stratified fluids with infinite depths, which were included in [4], and further research is
needed for such cases.
This paper is organized as follows. In Section 2, all the variables are nondimensional-
ized and Levi-Civita's method is used to transform the governing equations in an unknown
domain into equations in a fixed domain. In Section 3, a formal asymptotic expansion of
solutions is used and solitary-wave solutions are formally derived using a different method
from the one used in [4]. In Section 4, the equations in Section 2 are transformed into inte-
gral equations and a long-wave scaling for the horizontal variable is assumed. A solvability
condition is introduced to remove the singularity of the kernels in the integral operators.
In Section 5, the estimates of the integral operators are obtained using properties of the
Fourier transform in L2(R). Finally, the existence of solitary-wave solutions is proved in
Section 6.

2. Formulation

In this section, we use a formulation devised by Levi-Civita for treating the problem
of existence of waves of finite amplitude [14, 20]. The method is motivated by the desire
to reformulate the problem in terms of the velocity potential <p* and stream function I/J*
as independent variables in order to work in a fixed domain in stead of in the partially
unknown domain in the physical plane. We consider an irrotational flow of two immis-
cible, inviscid, and incompressible fluids of different but constant densities bounded by a
horizontal rigid bottom and without upper boundary. There exists an interface between

5
 these two fluids and there is a wave at the interface moving with a constant speed U. In
the reference to a coordinate system moving with the wave at the same speed, the flow
can be regarded as a steady flow. Let y* — 0 be the interface when x* goes to infinity,
y* = —h be the rigid bottom with h > 0, and y* = rj*(x*) be the interface. Then the
upper fluid with density p+ is in r/*(x*) < y* < +00 and the lower fluid with density p~
is in —h < y* < TJ*(X*), respectively.
Since the flow is irrotational and the fluid is incompressible, we can introduce the ve-
locity potential ip*, the stream function tp*, and the complex velocity potential x*{x* ■, V*) =
X*(z*) with x* = V* + ^*i z* - x* + iy* in -h < y* < rf(x*) and rf(x*) < y-* < +00.
The complex velocity w* = dx*/dz* is given by w* = u* — iv*, where u*, v* are the velocity
components. The corresponding dimensionless quantities will be

z = z*/h, w = w*/U, p = p+/p-<l, X = <Pi+tyi=X*/(hU) (3)

so that dx/dz — w. Furthermore we introduce a nondimensional parameter

7 = (1 - p)gh/U2 , (4)

where (I/7)1'2 is so-called nondimensional wave speed. Therefore ip\ is a constant along
the interface and for the sake of convenience choose tpi = 0 at the interface. The bottom
corresponds to ^ = — 1 since the total flow over a curve extending from the bottom to
the interface is Uh. In the following we use /+, /~ to denote a quantity / in the upper-
and lower-layer fluid, respectively. The boundary conditions are formulated as follows:

v = — Im w = 0 at -0! = —1, (5)

72/ + (l/2)\w~\2 - (p/2)\w+\2 = cojistant, (6)

y+ = y~ at fa = 0, (7)

where the second condition is from the Bernoulli's equation on the interface, w+,y+ and
w~,y~ at ipi =0 are denoted the values of w,y as x\)\ —> 0^. At the infinity, we have

w —> 1 when |x| —*■ +00. (8)

6
Therefore the complex potential function x(z) maps the physical plane (i.e. the z-plane)
onto the strip regions —1 < -01 < 0, 0 < i/>i < +oo in the x-plane. The mapping is one to
as
one and the inverse mapping function z(x) exists as we can easily see later. Taking x
independent variable, we then determine the analytic function w(x) in — 1 < ^l < 0,0 <
ipi < +00 from the boundary conditions.
It is convenient to simplify the form of the boundary conditions by following Levi-
Civita's idea [14, 20] to replace the dependent variable w by 9 + iX through an equation

w exp(-»-(ö + *A)). (9)

Therefore \w\ = eA and 9 = argw, where \w\ is the magnitude of the velocity vector
and 9 is the inclination relative to the x-axis. Now we formulate the conditions for the
determination of 9 and A in the ip\-ip\ plane. Condition (5) is reduced to

0 =0 at 0! = -1. (10)

By differentiating (6) and (7) with respect, to tp\ along ipi = 0, we have

7- \-\w — p\w T— = 0,
dip i dtp i dtpi
+
dy~ dy ,
-— = -— at ipi = 0 .
dtp-i d<pi

From dz/dx = (u + it>)|u;|-2 and the analyticity of 9 '+ iX, we obtain (dy/dipi) = v\w\~2
and (dX/d<pi) = ~{d9/dip\). Thus at rpi = 0 (6) and (7) become

e2A ^—-pe2X - 7e
A
sinö =0, (11)
oy>i otpx
e~X~ sm9~ = e_A+sin0+. (12)

When \tp\\ —> +oo,

0 + z'A-*O. (13)

Since 9 + iA is an analytic function of tp1 + i^, 0 and A satisfy the Cauchy-Riemann

equations

§Vl = A^, AVl = -9^ in - 1 < ^i < 0, 0 < ?/>i < +oo. (14)
In this paper we are interested in finding bounded nontrivial solutions (#, A) of (14) with
the boundary conditions (10) to (13) where A is even and 6 is odd in tpi.

3. Formal derivation of solitary wave solutions

Although a formal solution has been obtained by Benjamin [4] and many others in
different ways, we give another derivation here, which not only gives the first order ap-
proximation but also leads to higher order approximations.
To obtain an approximate solution of (10) to (14), we use a new parameter ip = e<fi

to stretch the horizontal variable ipi. Then we also strech the vertical variable by ■?/> = eißi

in the upper fluid, since the vertical length is infinite in the upper fluid and no long-wave
approximation can be assumed, and let the vertical variable in the lower fluid be the same,
i.e. ip — i\)\. Assume that 9, A and 7 have the asymptotic expansion of the following forms,

0+ ^fl+^/O + e3 0+^/0+ •••,

in 0 < ijf < +00,
Ä+ = e A+(^, r/,) + e A+(<^, </>) + • ■ •,
2 3

0- = e20r(^,^) + e302-(vp,V) + ---,

in - 1 < ?/> < 0,
A-
= e\i(<p, VO + e A^, xp) + ■ ■ ■,
2

1 = lo + Hi ■ (15)

In terms of ip,tp the equation (14) becomes

A{=ö+, 0J = -Aj in 0<V<+oo,

^ = e<^, e- = -e\~ in - 1< </>< 0 .

Substitution of the asymptotic forms for 9 and A yields different order of the approximations
for the equations and boundary conditions. The first-order approximation is

6
*U = C U = ~Xt m 0 < ^ < + 00,
Krp = °> ^ = ~KV in - 1< V < 0 ,
89 ~
—k--lQ0- = 0, 9- =9+ at V = 0,
of
e~ = 0 at if> = -1.
In order to have bounded nontrivial solutions 9^± ,Xf,
\± we need

70 = 1, 67(<p,i/>) = Sv(<p)ty + 1), A- = -S(<p),

0+ =F e-W F [SM] , +
XJ=F ik)-1\k\e-W*F[Sv(ip)]

where S(<p) is even and to be determined, and

+oo
1
F [/](*) = ^ / eik*f(x)dx,
+ oo
-iktp
F \M<P) = f(k)dk (16)

are the Fourier and inverse-Fourier transforms. The properties of these transforms can be
founded in [21, 22].
To find an equation for S(<p), we need second order approximation of the equations,

X = 6 6
U t^ ti> = ~Xt in
0 < ^ < +oo,

Kib
V
2V> — °» 0-.2ip -X 2ip in - 1 < if? < 0

97 - X~9~ = 9+ at 7p = 0,

(9~ =0 at 0 = -1.

This is a nonhomogeneous boundary value problem. In order to have a solution (A2 ,62),
the nonhomogeneous terms at i\) = 0 must satisfy

' 897 89+

= o.
ip=0

By the first order approximation 9X ,AX and 70 = 1, S(ip) must satisfy

j.SM + ZS^S^^ + pF -1*1 F [SM] (tp) = 0.

Since we are looking for solutions decay to zero at infinity, S(<p) satisfies

PF \k\ F [S(<p)] (^)-7i5(^)-(3/2)52(^) = 0, (17)

 which is so-called Benjamin-Ono equation [4, 19]. This equation has a solitary wave
solution,
s (18)
^ = Ä'
where a = —(47i/3),// = —(p/71) with 71 < 0. The derivation of solution (18) from (17)
can be found in [4]. Therefore the first-order approximation in (15) can be obtained. The
first-order approximation of the wave profile at the interface is

2\ ajj,^e 2,
y = eS(<p) + 0(e2) = -^-~2 + 0(e2). (19)

Since a > 0 with 71 < 0, this is an elevation solitary wave. This formal procedure can be
easily carried over to find the higher order approximations.
In the following sections, we shall show rigorously that there exists a solitary wave
solution of (10) to (14) and (19) is the first-order approximation of this solution at the
interface ?/>i = 0.

4. Transformations and Banach spaces

In this section, we transform (10) to (14) into integral equations. Let 7 = 1 + 7ie with
71 < 0 and e > 0.
First we consider the equations in the domain 0 < i\)\ < +00. The equation (14)
implies

#J, =AJl5 Ki =_Ö

Ji' ö
J^i +0$ivi =0
- (20)
If A+, A- and 9~ are small enough, (12) can be written as

§+ = arcsin(eA+~A~ sin0~) = b{ipi) at fa = 0. (21)

By applying the Fourier transform on (20), we have

(£ I**])** - l*l2 £ I**] = 0,

A A „
where F [f] is defined in (16). If F [8+] is finite as fa —» +00,

F[8+}(k,fa) = e-\k^ F[b(^)},

10
which gives

0+(¥>l,^l)=.F l*#i F arcsin(eA x

sin0 ) (22)
^!=0

By (20) again, we have

\+{Vl,tl>1)=F (i\k\/k)e-W+\F arcsin(e sin0 ) (23)

^1=0

Note that (22) and (23) hold in 0 < fa < +oo.

Next we study the equations (10) to (14) in the domain —1 < fa < 0. The equations
are

K=K^ K,=-K^ ^+^=0 in -1<^<0, (24)

7! e ö-+pe2(Ä+-A-)
dfa
+ (l+7ie)(e"3A~sin(Ö- 6~) at Vi = o, (25)

6~ =0 at fa = —1. (26)

From (22), we can rewrite (25) as

do- e~+pF \k\F 0- = 1\£®~

dfa

+ (1 + 7ie)(e-3X" sin(ö-) - t) + p(l - e^'x+-X^) F \k\F

+
+ pe2(x -x-)F k\ F 9 — arcsin(e X+-X- sm8

=/!(ö-,A-,A+). (27)

By using the Fourier transform, (24), (26) and (27), we obtain

smli(\k\(fa+l))F[f,]
§-(<p1,fa)=F (28)
\k\ cosh |fc| — sinh|&| + p\k\ sinh|A;|

IklcoshdkK^+l^Fjh}
\~(<p1,fa)=F (29)
ik(\k\ cosh|fc| — sinh \k\ + p\k\ sinh|&|)

Thus the equations (10) to (14) are transformed into integral equations (22), (23), (28)
and (29), where f\ is defined in (27).

11
 A crucial step in our procedure is that we make the same change of independent
variables

(p = e<pi, tl> = ip\, (30)

which we made in Section 3. In accordance with the development of Section 3, we put

0* = e~l2a±
6 A± = e~1i\±
A (31)

Then by changing the integration variables in (22), (23), (28) and (29), these equations
become: In 0 < ifi < +oo, ':

-e\k\f p e 2arcsinfee(A+-A") sin(e^-) (32)

V>=o

i\k\/k)e-tm F e arcsin (V<A+-A~>sin(e2r) ) (33)

In -1< V < 0,

sinh(e|£|(</> + 1)) F [f2]

6-(<p,r/>)=F (34)
pe\k\ sinhe|&| + e\k\ coshe|&| — sinhe|&|

lfc|cosh(e|Ä;|(^ + l))F[e/2]
\-{<p^)=F (35)
ik(pe\k\ sinhe|A;| + e\k\ coshe|&| — sinhe|&|)

where

f2(e-,\-,\+) = e-2Me29-,e\-,e\+)
= [7leö- + (1 + 7ie)(e_2e"3eA_ sin(e2 IT) - 6~)]

+ pe2{tX+-tX~) F e\k\ F \$~ - e~2 arcsin(e' A+-eA" sin(e^-))

+ p(l-e2(eX+-eX~))F e\k\F [0"

= 1 + 11+111.

Notice that when e or k is small, the denominators in the inverse-Fourier transforms of (34)
and (35) have a zero point of order e2k2. Therefore we have to single out this singularity

12
in the integrals. Write

sinh(e|fc|0 + 1)) _ qfr + 1

pe\k\ sinh.e|fc| + e\k\ cosli e|A;| — sinhe|&| pe\k\
(*</> + l)(e\k\ cosh(e|ib|) - sinh(e|fc|))
/9e|&|(/>e|&| sinhe|fe| + e\k\ coshe|fc| — sinhe|fc|)
sinh(e|&|(V> + 1)) - (iß + l)sinh(e|fc|)
+ pe\k\ sinhe|Ä| + e\k\ coshe|fc| — sinhe|fc|
def
h(k)(ll> + 1) + k2{k)(ll> + 1) + &3(M), (36)

|fc|cosh(e|fc|ty + l))
ik(pe\k\ sinhe|fc| + e\k\ coshe|A;| — sinhe|&|) ipe2k\k\
(e\k\ cosh(e|&|) — sinh(e|&|))
2
ipe k\k\(pe\k\ smhe\k\ + e\k\ coshe|fc| — sinhe|fc|)
(e\k\ cosh(e|fc|(V> + 1)) - sinh(e|&|))
+ -ike(pe\k\ sinhe|A;| + e\k\ coshe|Ä;| — sinh.e|A;|)
def
(l/iek^k^k) + k2(k) + Ä^(M))- (37)

Therefore (34) and (35) become

v V
(
+ !)(> (l/pe\k\)F[f2] + F k2(k) F [f2
V
+ F h(k^)F[h] > (38)

V
\-(<p,il>)=F (l/ipek\k\)F[f2] + F (l/ik)k2(k) F [f2]
V
+ F (l/ik)k^(k,iP)F[f2] ■
(39)

Note that k^(k,iß) is finite as ek —> 0. Therefore we need to remove first two terms in the
right hand sides of (38) and (39). Let

A~ = -T)(ip) + C and then 6 = {ip + l)rjv((p) + £ . (40)

If we assume
v
-T] =F (l/ipek\k\)F[f2] +F (l/zk)k2(k) F [f2] (41)

13
 which is called solvability condition, then (38) and (39) are changed into

S=F h(k,1>)F[f2] (42)

C=F [l/ik)hi,(k,iJ>)F[f2] (43)

By the definition, f2(6 , A ,A+) can be written as

f2(6-,\-,\+)=f2(r]ip + t,-rj + (,\+)

^=o
= 7ie6~ - Ze\~8- + e ft(0_, A~, A+)
2

= 7i erlv + 3er/?/v + 71 e£ - 3e((r]^ - £7/ + (0

+ e2ft(0-,A-,A+)

= 7ie?/v + 3eVVv + eMTh t, C, A+) - (44)

Thus by taking the Fourier transform on both sides of (41), we can obtain

-pe\k\kiF[V] = (l+pe\kMk))F[f2],

which implies
v
F p\k\(l + pe\k\k2(k))-'F [rj] (l/e)/2(r,A-,A+)
j (p

= 7i^ + (3/2)(r,2)¥,+/2(?7,£,C,A+). (45)

If we let e = 0 and f2 = 0, (45) becomes (17) and it has a solution

ap
%0 = <p + p2
2 (46)

where a — —(471/3),/* = —(p/71). Assume

v = S(<p) + w(y). (47)

Then substitution of (47) into (45) yields

p[F \k\{l + pe\k\k2{k))-1 F [u y-[Uv — 3(Su)v

= (3/2)(u2^ + p F \k\2pek2(k)(l + pe\k\k2(k))~' F [S(<p)] + /2(r?,£,C,A+).

14
By applying the integration once, we have

pF \k\(l + pe\k\k2{k))~l F[u] 7iu> — 2>Suj

r+oo
= (3/2V2 +PF \k\2 pek2(k)(l + pe\k\k2{k))~l F [S(ip)}
J if
+
= <7a(u;,e,C,A ). (48)

In the following, we shall show that (32), (33), (42), (43) and (48) have a solution
(LO, £, C, 9+, A+) in some Banach space for small e > 0, where u>, (, A+ are even and £, 9+
are odd in <p. -
Before proving the existence, we define some Banach spaces for later use. Let iP(R)
be the usual Sobolev spaces on R = (—oo, +oo) with s > 0 [22]. Define

W = {/(V) G F'(R) | ^/(p) € #5(R) and ||/||w. = ||(1 + V2)1/2/||^(R)},

53 = {/fe^)6l2(Rx [-1,0]) I for any Ve [-1, 0],/(</?,</>) G W5, and

ll/lls» = sup 11/(^,^)11^ <+00}.

-l<t/><0

We use Wes, B*,H* or W0S, Bs0,Hl to denote the corresponding spaces with even functions
A
or odd functions in ip. Note that f(ip) G Ws if and only if F [/](&) is differentiable in k
and
(1 + P)*/2 F [/](*) 6 i2(R), (1 + £2)s/2^ F [/](*) G L\R).

Also by using the Sobolev embedding theorem, we have

Lemma 1. If u,v G iP(R) for s > 1, then uv G iP(R) and

IHI#'(R) < Ä"H|#'(R)N|ff.(H.)

where K is a fixed constant.

5. Estimates of the operators

First let us consider (32) and (33). Rewrite these equations into

ff+far/,) =F e-'W* F [Sv] +F e-^Fluy + ZfaO)]

15
 V
+F -e|A# p (e~2 arcsin(ee(A+-A_) sine2r) - Ö")
^=o
def
^ L,(S) + L2(U,O + L>(Lü,Z,(A ),
+ (49)

A+(V,^)=F
v
HAI/i^e-'1*1* F (e~2 arcsin(ee(A+-A") sine20_) - 0
#=o
d
= M1(5)+M2(u;)0 + M3(a;,e,C,A+), (50)

where L%,M% are of order e for small e > 0. By using the properties of Fourier transforms,
Lemma 1, and the Plancherel equality, we can have that if ||£(v>0)||w» + IKCVJ'O)!! W« +
IK>lk0* + ||A+(^,0)||W; < K for s > 2, then for 0 < r/> < +oo,

and
Mw,e,C,A+)|U. + ||M3(a,,e,C,A+)||ff; <iT6.

Since |AT| is piecewise differentiate and w^,^ are odd,

r+°° A
<pM2(u>, 0 = 2(P cos(ktp)(ie)e-€\k^ F [uv + £(<p, 0)] d&
Jo

e-^(sgn(k))e [^^^ F K + ftp, 0)] J <tt ,

where sgn(fc) = 1 for k > 0 and sgn(Ar) = —1 for A: < 0. Thus for xj) bounded,

||^2(W,0lkj+lbM2(a;,0||^

< ür(||(l + Ar2)3/2 (d (e-W* F [OJV + £(p, 0)]] M

L*(R)

+ (1 + Jb2)*/2sgn(A) ( d ((e"«'*'* F [uv + £(<*>, 0)] I /<&

L*(R)

<K{\\Uv\\w.+M(<p,0)\\w:).

By using the same argument, we have

\\ipL3(u,UA+)\\H: + \\vM3(u,UA+)\\Hi<K€.

16
Therefore we obtain
Theorem 1. If ||wv||w. + ||£O,0)||^ + ||CO,0)||^ + ||A+(^,0)||w- < K with s > 2,
then for 0 < ifi < K\ < +oo

\\L2(^0\\W:+\\M2(U,O\\W: <Ä"(Klk; +11^,0)11^.),

||Z3(u,,£,C,A+)|k, + ||M3(u,,£,C,A+)||^<A'e.

Next we consider (42) and (43). From the forms of k^(k,^) and kz^(k,ij)), it is easy
to see that for any small e > 0, k £ (—oo, +oo) and — 1 < ?/» < 0,

\h(k,iß)\ + |fc3v,(fc,VOI + |^&fc3(fc, V)| + |(^)_1*3^(*,^)| < A2 < +oo,

where K-i is a fixed constant. From the definition of fi in (34) and (35), we also have
v V
F h(k,i>)F[f2] ■-F h(k,iP)F[I]
v
+ F h(k,p)F[II] + F h{k^)F[III]

Assume that at ip = 0, +
||A ||HA< + ||A \\w> + ||# \\w> + \\ö+\\w° < K for s > 2. By Lemma
1, the forms of /, II and III, and the Plancherel equality, we have that for — 1 < tf> < 0
v
F h(k,^)F[I] <K\\I\\H,<Ke.
Hi

Also
V
F h{k,xl>)F[II] <\\h(k^)F{II}\\lHK)<K\\II\\l2{K)
A r
<K F k\eF r-e-2arcsin(ee(A+-A_) sine2^-)
i2(R)
2 n2
<Ke (\\e- liA-fe + l|A+fe),
and for 5 > 1
ds v
&3(fc,V>)F[II] <i|FÄ;3(Ä;^)F[J/]||2L2(R)
dc/? L*(R)

< Ä-p-^"1 F [II] |ß,

2 1 2
(a) < KiKi/eX^- ///^)!! ^^
d»-i v A r 2
<A \k\F 8 —e arcsin(e e(A+-A~)sine2ö-
ds3'1 L*(R)
2 2 2
<AV(|r|| +||A-|| +||A+|| ).
17
 V A
Tims || F [k3(k,tß) F [II]]\\H> < Ke. By using similar estimates for III, we can have
|| F [h(k,4>) F [III\]\\H* < Ke. Therefore for -1< ij> < 0

v
F h{k^)F[f2] < Ke.
Hi

Also note that \k\ is piecewise differentiable and the derivatives of k%, k3^, ekk% and (l/ek)k3^
with respect to k are bounded. Therefore using similar arguments, we can have

v
<?F h(k,^)F[f2] <K F {dh(k^)/dk)F[f2
H> H*

+ h(k,^)F[cpf2]
H*
< Ke.

Thus F [h(k,i>) F [f2]] e W° and

v
F h(k,4>)F[f2] < Ke.
w*

By a similar reason,

v
F (l/ik)k^(k,iP)F[f2\ <\\(l/zke)k3lP(k,^)eF[f2}\\LHIl)
L*(R)
<K\\f2\\LHK)<Ke,

and for s > 1,

ds v
F (l/ik)kSlP(k,^) F [f2 < ks-lk^(k,^)F[f2}\\LHJL)
dip
<K\\{ds-lf2/d<ps-l)\\rM^
L*(R) <Ke.

V A
Thus || F [{l/ik)kzy(k,%j;) F {f2]}\\H* < Ke. By applying the similar argument, we can
v A
have \\(p F [{l/ik)kzxi){k,'ip) F [f2\\\\Hs < Ke since \k\ is piecewise differentiable. Therefore
F {(l/ik)k^(k,^) F [f2]] e W: and

v
F (l/ik)k3i,(k,il>) F [f2] < Ke.
W!

We summarize the above results as follows.

18
Theorem 2. If 6~ = (5 + w)^ + 1) + £ € B*, A" = -(S + w) + ( € Bae, \+(<p, 0) e We<
and Hwll^+i + ||£||B. + IICIIs. + ||A+(^,0)||^ < K with 5 > 2, then

v
h{k^)F[f2] < Ke,
B~:
V
F (l/ik)h^(k,^)F[f2] < Ke.
Bf

where K is a constant independent of e and ■</>.

Finally we study the equation (48). Let

N(w) = pF \k\{l + pe\k\k2{k))~l F[UJ — jioj — 3Su. (51)

Then (48) becomes

(52)

We first study the general nonhomogeneous equation

N(u) = g(if), uv(0) = 0, (53)

where u>(<p) and g(<p) are even functions in ip. By applying the Fourier transform on both
A
sides of (53) and using F [S(ip)] — (aju/2)exp(—p\k\), we have

p\k\(l + pe\k\k2(k))-' F H (k) -7lF H (k)

/ + °° A A
o^e""1*-*'! F [üj](k')dk' =F [g](k), (54)
-oo

A
where a and p are defined in (46). Thus F [<*>](&) must satisfy

Zap '+0°
£ M(*)- 2(/>|*|(l + ,e|*l*2(*))-1-7i)./-
e-H*-*'l F [w](k')dk'

F [g]{k) def
= 9o(k) (55)
p\k\(l + pe\k\k2(k))-^-7l)

Note that by the definition of k2(k),

e\k\ cosh e|fe| — sinhe|&|

{l + pe\k\k2{k))-1 = 1
/9e|fc|sinhe|fc| + e\k\ coshe|&| — sinhe|A;|
e|&| coshel&l — sinheifcl
= 1+ ' ' ,,,'.' ,,, ' ' >1 (56)
p\k\esmhe\k\

19
Denote the integral operator in (55) by

T
<(«(*» = of inn-u izlmw i /+0° «-"l*-*'l«(fc')^' (57)

and T(u) =f T0(u). Then (55) becomes

F[u]-Te(F[u]) = g0(k). (58)

Now consider an integral equation

u-T(u) = /, ^ (59)

where / G L2(R) and u,/ are even in ip. Let

/+oo
e-"lfc-*'lu(ife')^'-
-oo

By (59), W(k) satisfies

Wkk-fi2W+ ,y W = /(fc). (60)

Since u(fc) and /(&) are even, we consider equation (60) with Wjt(0) = 0. It is easy to find
a solution uj(k) of the corresponding homogeneous equations (60) for k > 0,

w1(Jfe) = Jb(Jb + /z-1)c-"*.

Then using reduction of order, we can obtain another linearly independent solution U2(k)
with 14(0) = 0 and u\u'2 — u[u2 = 1. Also as fc —> +00, u2(k) and its derivatives have
order of 0(ife-2e'i*). Thus the solution of (60) is
pk r + 00
W(k) = Ml(fc) / u2(s)/(a)<fc + u2(fc) / u1(s)f(s)ds
Jo Jk
d
£ Af(f(k)) d= ker(k,s)f(s)ds for k > 0, (61)
Jo
and evenly extended to k < 0.
Lemma 2. If f(k) G i2(R), then jV(/) e #e2(R) and

• II^(/)I|^(R)<A1|/||^(R)
20
Proof: Since Af(f) is even,
r+°° 2
ll^(/)llij(R) = 4|pV(/)fL2(R+) < 4 jf ker(*, *)/(*)<** LHB-+)
r + OO / i» + CO \ / f + OO \
<4/ (/ |ker(fc,s)|ds) ( / \kev(k,s)\f2(s)ds)dk

<4( sup / |ker(fc,a)|ck) ( sup / | ker(k,s)\dk) ||/||L2(R+)

\0<k< + ooJo / \0<s<+oo7o /
But

sup / \ker(k, s)\ds < sup f / \ui(k)u2(s)\ds

o< k<+ooJo 0<fc<+co Wo
f + OO
+ [ °° MkfaWlds) < K sup ( / (fc2 + l)e-"*(52 + l)"1e'iSds
Jj
7fc ' 0<fc<+co Wo
H-oo
f+OO N
+ / (A:2 + l)"1e'4fc(52 + l)e-^sdsj < Kx < +00 .

The same estimate holds for sup0<s<+oo J0 °° | ker(fe, s)\dk. Thus ||A/"/||L2(R) < ä"||/||L2(R).

By differentiating (61) once to have

—^(/(fc)) = u[{k) J u2(s)f(s)ds + u'2(k) J u1{s)f{s)ds,

and using the same argument for J\f(f(k)), we obtain |K-V(/))/dfc||z,2(R) < -K"||/||z,2(R)-
Since Af(f) satisfies (60), \\d2(Af(f))/dk2\\Ll{K) < K\\f\\Llm- Therefore ||^/|U»(R) <

Ä"ll/IUj(R).
Thus the solution of (60) is W(k) = Af(f) for / G L2(R). By (57) and the definitions
of T and W(k), we can write (59) into

/»Fl-71
or
u(k) = f(k)--^-Af(f). ' (62)
PFI-71
From (62), Lemma 2, and the fact that the first order derivative of \k\ is piecewise con-
tinuous, we have that (59) has a unique solution u G £2(R) with ||u||i2(R) = \\(I —
Tr'fhHK) = \\Wkk - fi2W\\LH1L) < Ä-||/|U*(R) and

||(i- rrVll^w = INI/zum < KW/WHHB.),

\\k(I-T)-'f\\mm = \\ku\\mm < K{\\k{df/dk)\\L,m + Www). (63)

21
 Next let us consider
u - Tt(u) = f (64)

for e > 0 small, where / and u are even in (p with / € Z/2(R). We shall show that I — Te
is invertible in L2e(R) for / 6 L2e(R). Write (64) as

u-T(u) + (T-T€)(u)=f. (65)

Now we estimate the operator norm of T — Tt. If u E Ll(R), then by (56)

def 3a///)|Ä;|(e|A;|coshe|^| — sinh e|Ar|)

Te(u) = (Te - T)(«) = -
2pe\k\smhe\k\(p\k\-ll)(p\k\(l+pe\k\k2(k))^^ll]
/+oo
x
-oo

Therefore
+ 00
|£|2(e|Ä;|coshe|Ä:|-smlie|&|^
I^(«)IILJ(R) <K oo {e\k\smhe\k\y{p\k\-liy{p\k\{l + pe\k\k2{k))-^ -^f

x f f °° e-^k-k'\u(k')dk'} dk
+°° k1,22(e\k\ cosh e\k\- sinh e\k\f
< KhWhm dk
(|fc| + l)4(e|jfe|sinhe|ife|)2
K\\u\\2L,wI2(e).

However
e\k\ coshe|&| — sinhe|&|
< Kz < +oo for all e > 0, k G (-oo, -foo).
e\k\ sinhe|&|
By the dominated convergence theorem, /(e) —► 0 and ||Te[| < KI{e) —► 0 when e —> 0. For
small e > 0 the equation of u — Tt(u) = / can be transformed into

u - (/ - T)-\Te - T)u = (I - T)~lf.

Since (I - T)-1 is bounded from L2e{R) to L2e(R), the operator norm of (I - T)~l (Tf - T)
goes to zero as e —>• 0, which implies I - (I - T)~l(Te - T) is invertible and the solution
of u — Te(u) = / satisfies

u Uj(R) = ||(/-Te)-7|Ui(R)

< I-il-Ty^Tt-T)) \l-T)~lf ^ <K\\f \i>im (66)

Ll(K)

22
if e > 0 is small. By (64) and the definition of Te in (57), we also have that

|HU*(R) < H*Te(u)|U;(R) + P/||LJ(R) < ä"(II*/HLJ(R) + ll/IU;(R)) ■ (67)

For f(k) G if^R), by taking the derivative of (64), we obtain

d/(*)
+ /(i) +
s-w L'(R) ^ U(fc) 2
£ (R) ^(ll*s i*(R) dk i2(R)

+ ||(*2 + l),/2/(fc)IU»(R) (68)

If (1 + k2)'f2f(k) and (1 + k2)sl2{df/dk) are in L2(R), then by using (64) repeatedly and
noting that the integral operator
/+oo
e-H*-*'i(i + k2yi2{\ + \k'\2)-si2f(k')dk'
:
co

maps i2(R) to L2(R) continuously for any s > 0, we obtain

;i + *a)'/2«(*)iu,(R) + |(i + k2y/2^u(k)

i2(R)

< K (j|(i + k2yi2f{k)\\L,m + ||(i + k2yi2j-kf{k) (69)

£2(R),

In order to have (69) we have used the fact that u = Te(u) + / implies

IMMR) < ||*Te(u)||La(R) + ||fc/||z,*(R) < JC(||«||La(R) + ||fc/|U*(R)).

Now we can consider (58). From above derivation, for e > 0 small enough, (58) has a
solution F [u] such that

d A
l(i + k2y'2 F H||L2(R) + (i + k2yi2-
dk
FM
i2(R)
d
< K ( ||(1 + k2y'2gQ{k)\\Llm + (1 + **)'/2_0O(fc)
i2(R)

if the right hand side is finite. Then we use the Plancherel equality, the properties of the
Fourier transform, and the definition of go(k) in (55) to obtain that for g(<p) 6 W/-1 with
def , _ix
s > 1, the equation (53) has a solution u = Nr (g) G Wes satisfying

IIJV-^^ii^^üriMi^-i.
23
We summarize it as follows.
Theorem 3. If N(u) is defined in (51) and g(<f) G Wes for s > 0, then for small e > 0 the
equation N(u>) — g with uv(0) = 0 has a solution u = N'1^) in W*+1 satisfying

6. Existence Proof

We consider (42), (43), (49), (50) and (52). Let

XvO=^K1/2, pi(y>)= (ö+(v,0)-L1(5))e-1/2,

p2(^)=(A+(¥3,0)-M1(5'))6-1/2,

3i(v,^) = £(^>~1/2, ?2(v,^) = C(</v/>K~1/2- (70)

Then we rewrite (42), (43), (49), (50) and (52) by

P(¥') = ^1/2JV-I(yi(«1/2P^1/2«1^1/292^1/2P2(V) + ^1(5')))

= P(p,Pl,P2,?l,?2), (71)

p1(v3) = e-1/2(i:2(^1/2,3l£1/2) + i:3(pe1/^e1/V,e1/2g2,e1/2p2(^) + M1(5)))

= -Pl(p,Pl,P2,?l,?2), (72)

= p2(p,Pi,P2,qi,q2), (73)

qi(<p,1>) = e-1,2F \h(k,rj>)F \f2 ((S + e^2p)v(i> + 1) + q,el'\

_(5+el/2p)+?2el/2?Ml(5) + el/2p2)

Ql(P,Pl,P2»9l»92), (74)
A r
<Z2 (V,,^) = e-1/2F (l/ifc)M*. VO i71 /2((S + e^p)^ + 1) + gi61/2,

_(5+el/2p) + (?2el/2)Ml(5) + el/2p2)

Q2(P,Pl,P2,ffl,g2)- (75)

24
We are looking for fixed points of mapping (P,Pi,P2, Q\, Q2) in some Banach spaces.
Define a closed convex set in a Banach space W/+1 x W03 x W* x Bs0 x Bse as follows:

s ={x = (p,p,,P2,quq2) € w:+i x w: x w: x Bs0 X S*

= Ibllv^+i + Ibilk; + INk* + Ikibi + ||?2||s; < 61 < +00},

where b\ > 0 is a small fixed number with s > 2. Assume that X = (p, Pi, P2, <7i, <72) 6 <?.
By Theorem 2, we obtain

|Qi(i>>Pi.P2>?i,?2)||ßj + ||<32(P,PI,P2,3I,<?2)||B| < #e1/2 (76)

In order to use Theorems 1 and 3, we need to change (71) to (73) into the following forms,

p((p) =P(P,PI,P2,QUQ2) = P(p,pi,P2,qx,q2), (77)

Pi fa) =Pl(P,Pl,P2,Ql,Q2) =Pl(P,Pl,P2,<Zl,<?2), (78)

p2(<?) =P2(P,P\,P2,Ql,Q2) =-P2(p,Pl,P2,?l,92). (79)

From Theorem 3, we need to show that

g,(e^P,e^Que^Q2,e^p2+M1(S)) < Ke. (80)

W!

By (48) and noting that

def
l'i F \k\pe\k\k2(k)(l + pe\k\k2{k))-' F [S(<p)]

<K\ F [\k\pe\k\k2{k){l + pe\k\k2(k))-1 exp(-/*|fc|

Since \k\2e ßk
\ \ decays exponentially as k —> +00 and k2(k)(l + pe\k\k2(k)) 1
is uniformly
bounded, ||Ii||w« < Ke for any s > 0. Therefore the W/-norm of first two terms in g\
of (48) are less than Ke. Next we need to show ^°° f2(S + w, f, C, \+)d<p G Wes. By the
definition of f2 in (44),

/2(5 + cu,e,C,A+) = 7ie + [-3(C(5 + a;)¥,-e(5 + u;) + a)

+ eH((S + u)v + £, -(5 + cu) + C, A+)]
def
7if + I/i(u;,e,C)- (81)

25
Now we need

Lemma 3. If m{ip) G Ws0, u2(<p) G W/ for s > 1, then fj°° ui(p)u2(¥>)<fy> € W°+1 and

+00
u!(^)u2(v?)^ w.
< iriMvOlkJMvOlk;

r+00 ailc
Proof: Let v(<p) = / ui((p)u2(ip)d<f. Then (dv/dip) = u\u2 G W^ * °y Lemma 1

||(du/<V)lk* < Ä"||«i(^)||w/||u2(v)lk/ •

Thus we only need to show the estimates for ||v||wo. Since v is even,

+00 +00
v L*(R)
r.2 u\{i)u2{t)dt dip
<p

2
< A / {^ + l)-2 Ul(t)u2(t)(l+t )dt dip

<Al(H-^)1/2u1i|2L2(R)||(l + ^)1/2u2||2L2(R),

and

•+00 f + OQ

MIL^R) <K '

I -2< -2 + 1)-
^(^

2
< K\\(l + ip f' u, || 2 2
2(R)
J (p
Ul(t)u2(t)(l

||(1 + ip2f'2u2\\l 2(R).

+ t2)dt dip

Thus |H|w° < A'||UI||VK! 11^21| w» and the proof is completed.

From (81), it is easy to check that every term in II\ is either a product of u\ and
u2 with ui G W*-\u2 G Ws0~l or a term like F [\k\ F [f(ip)}} with ||/||^ < Ke. The
integral of the latter is F [(\k\/ik) F [f(<p)}} G W/. Thus

+00
III dip
w.
. <^+IICIk- + ||flk;).

But by the definition of P in (77),

h(S + co,e^2Que'/2Q2A+) = liQi\^o+n1(uj,el/2Q1,e^2Q2,X+)

= II1(S + u,e1'2Q1,e1'2Q2,\+).

26
since Qi|^=o = 0 by our definition of Qi in (74) and the construction of k3(&,-0) in (36).
Thus
r+oo
/ /(5 + o;,e1/2g1,e1/2g2,A+)^
■J<fi ^e

< K(e + \\^ Qi\\w: + ||e1/2Qi|k*) < Ke,

2

and (80) is proved. From Theorem 3,

\\P(p,Pi,P2,qi,q2)\\w:+i <AV/2. (82)

Finally by (78), (79) and Theorem 1,

\\Pi(p,Pi,P2,quq2)\\w:+\\P2(p,Pi,P2,qi,q2)\\wS <Ke1'2. (83)

Define
7"(p,Pi.P2,gi,?2) = [-P,-Pi,-P2,Qi,Q2](p, Pi,J»2,31,92)-

By (76), (82) and (83), T maps S into itself if e > 0 is small. Also by using the same proof
as (76), (82) and (83) from Theorems 1, 2 and 3, we have
Theorem 4. If X™ = (p™,P?\£\q?\q™) and X^ = {p^ J? J?,q?^) are
in S, then
l\T(X^)-T(X^)\l <Kel/2\\X^ -X(2)||.

We choose e so small such that the constant Ke1!2 in Theorem 4 is also less than 1/2.
Then T is a contraction in S. By the contraction mapping theorem, we can have a unique
fixed point of T in S. Therefore we obtain a solution (p, Pi, P2, 9i, 92) in S for (71) to (75).
However from (73), (79) and (50), it is easy to see that if ||üVIIW* + ||^(^,0)||^ < Ke
and ||A+||^ < K, then ||iMp,Pi,P2,9i,92)||iyf < Ke3/2. Therefore we have the following
Theorem.
Theorem 5. For e > 0 small enough, the integral equations (32) to (35) have a solution

A+(^)=£ (i\k\e/k)e-W* F [Sv] +A+(^),

B-(iP,rl,) = ^ + l)Sl/> + ^(ip,i,)1

\-(ip, ?/>) = -S + Af (<p, i>),

27
where
a/.r
S(v0 =
(f + ji2
2

wiith a = -(471/3),)u = -(phi), Of are odd and Af are even in y>, (l + (p2)1/29f(-,i/;),(l +
ip2)1/2\t(;T/>)€ HS(R) for 5 > 0 with

sup (ii(i+^2)1/2ör(-,v)ii//-(R) + ii(i+^)1/2Ar(vA)ii^(R)

+ SUp (||ö+(-,V)||^(R) + ||A+(-,^)||^(R))

+ sup (yeti-MlH'w + Wf^ti-MlH'm

v
0<i/j<Kl

< Ke,

and ii is a constant independent of e but may depend on 5 and K\.

Since if1(R) is embedded in bounded continuous function space C°(R) and 61 , A-,
satisfy the Cauchy-Riemann equations (20) or (24), (l + <p2)l/20f, (1 + ^2)1/2A± and their
derivatives are bounded and continuous in either Rx ( — 1,0) or Rx (0, +00). Then from
(34), (35) and (40), we write

where Hi consists of all nonlinear terms in /2 and (p2Hi € HS(R) for s > 0. Therefore
(42) can be written as
v v
h(k,i>)F [7i<^ + 0] ^(Ä^/O^M (84)

But from (45), we have

1
^ Nv] = (-71 + P\W + pe\k\k2{k))--n-
1
) [F [7le£] + F [11,

Substituting this expression into (84) and taking the Fourier and inverse-Fourier trans-
forms, we can obtain

Z{<prf)=F
l-71eh(k,il>)(l+<n(-<y1+p\k\(l + Pe\k\k2(k))-1)-1)
def£
h(k,if>) F [Hi

28
However, notice that 71 < 0 and fc4(&,?/>) is bounded and even in k. Since 7Z\ is odd and
tp2%i is in fP(R) for 3 > 0, integration by parts twice yields

f+°° / d2 A
\

But 9?27?-i is in 1P(R) and the derivatives of k±{k,il)) are bounded, which implies <^2£ G
iP(R) for s > 0 and any fixed ?/>. Then by (45), ip2r}v G HS(R). Therefore from (40)
ip26~ G HS(TV) for 5 > 0. Now we use integration by parts in (43) and the oddness of
(l/ik)k^(k,^) to have (1 + ip2)( G CS(R) for integer s > 0. From (45) and the fact that
^=0 = QiU=o — 0 in (74),

v _i A r r+c° 1
rj=F (-ll+p\k\{l + pe\k\k2{k))-1) lF / TZ2dip

where K2 = (3/2)(r/2)¥, + /2, and ^Tl2^2 J+°° Tl2d<p G fP(R) using Lemma 3. Thus
integration by parts twice yields (1 + <p2)rj G CS(R) for s > 0. Finally we use the Fourier
transforms of exp( — |&|q/>) and ei(\k\/k)exp(—e\k\ip) in (49) and (50) and the classical
estimates of elliptic operators in [13] to obtain

sup ((|0+| + |A+|)(1+V + I<M)) <K

<p€'R.,4>eTl+

and the derivatives of #+, A+ are bounded by K(l + ip2 + l^el2)-1. Therefore Theorem 5
can be improved as follows.
Theorem 6. (32) to (35) have a solution stated in Theorem 5 with 6 and A satisfying

sup ((10*(^)| + |Af (<p, tf)|)(l + ^ + \e

s

+ \5]sup ((|D'0±(^)| + |0'Af (V,tf)|)(l +<p2 + M2)) < Ke

for any integer s > 0, where DJ is denoted as the i-th derivative with respect to either ip
or xp or both and K is independent of e but may depend on s.
After transforming back to Ö^A*, and ip\,ij)\ in (30) and (31), and calculating the
Fourier transforms explicitly, we have

29
 Corollary 1. For e > 0 small, the equations (20), (21), (24) to (26) have a solution

x\ + f 1 N e2aM(^i + /Q2-(eyi)2) , 2 +, .
(<*.*) = ((^+/i)2 + (^l)2)2 ^M^,*),

Ö (*>l,lM = (^2+^)2)2 +^l(^l^l),

2

//- + (e<^)2

where a = —(471/3),// = — (p/71), #1 and Ax satisfy the properties stated in Theorem 6.

From the Bernoulli's equation (6) at the interface ?/>i = 0, the interface of two fluids

is determined by

V = (1 + 7i0_1(l/2)((l - e2A_) - p(l - e2A+))

^1=0

= -A" + eY^e^) = eS(epi) + e^e^j),

SU
where Y^i=o P^GR- 1(1 ~^~ ^^{^Y (ip) / d(ps)\ < Ä"e and ÜT is independent of e. Finally we
use the dimensional variables to state the existence of the solitary waves.
Theorem 7. There is a solitary wave moving with constant speed

1/2
U=((p--P+)gh/(p~(l + 7le)))

in a two-fluid flow with 71 < 0 fixed and e > 0 a small parameter, where the height of
the upper fluid is infinite while the depth of the lower fluid is h, and the densities of the
upper and lower fluids are p+ and p-, respectively. The profile of the interface between
two fluids is given by

y =7l{x) =
(eX*y + (h^+eY{ex/h)>

where a = -(471/3),ju = -{p+/p~l\),

s
SU
J2 P 1(1 + x2)(dsY*(x)/dxs)\ < Ke,

and K is independent of e.

30
 Obviously the first-order approximation of the solution given in Theorem 7 was ob-
tained by Benjamin [4] using asymptotic method.

References

[1] C. J. Amick, On the theory of internal waves of permanent form in fluids of great
depth, Trans. AMS, in press.

[2] C. J. Amick and J. F. Toland, Uniqueness and related analytic properties for the
Benjamin-Ono equation — a nonlinear Neumann problem in the plane, Acta Math.
167 (1991), 107-126.

[3] C. J. Amick and J. F. Toland, Uniqueness of Benjamin's solitary wave solution of the
Benjamin-Ono equation, IMA J. Appl. Math. 46 (1991), 21-28.

[4] T. B. Benjamin, Internal waves of permanent form in fluids of great depth, J. Fluid
Mech. 29 (1967), 559-592.

[5] T. B. Benjamin, The stability of solitary waves, Proc. R. Soc. Lond. A 328 (1972),
153-183.

[6] T. B. Benjamin, A new kind of solitary wave, J. Fluid Mech. 245 (1992), 401-411.

[7] T. B. Benjamin, J. L. Bona and D. K. Bose, Solitary-wave solutions of nonlinear

problems, Phil. Trans. R. Soc. Lond. A 331 (1990), 195-244.

[8] D. Bennett, J. L. Bona, R. Brown, S. Stansfield and J. Stroughair, The stability of

internal waves, Proc. Camb. Phil. Soc. 94 (1983), 351-379.

[9] T. L. Bock and M. D. Kruskal, A two-parameter Miura transformation for the Benjamin-
Ono equation, Phys. Lett. 74A (1980), 173-176.

[10] J. L. Bona, On the stability of solitary waves, Proc. R. Soc. Lond. A 344 (1975),
363-374.

[11] R. E. Davis and A. Acrivos, Solitary internal waves in deep water, J. Fluid Mech. 29
(1967), 593-607.

31
[12] A. S. Fokas and M. J. Ablowitz, The inverse scattering transform for the Benjamin-
Ono equation — a pivot to multidimensional equations, Stud. Appl. Math. 68 (1983),

1-10.

[13] G. Gilbarg and N. S. Trudinger, Elliptic partial differential equations of second order,

2nd ed., Springer-Verlag, 1983.

[14] T. Levi-Civita, Determination rigoureuse des ondes permanentes d'ampleur finie,

Mathematische Annalen 93 (1925), 264-314.

[15] K. Kirchgässner, private communication.

[16] S. A. Maslowe and L. G. Redekopp, Long nonlinear waves in stratified shear flows, J.

Fluid Mech. 101 (1980), 321-348.

[17] Y. Matsuno, Exact multi-soliton solution of the Benjamin-Ono equation, J. Phys. A:

Math. Gen. 12 (1979), 619-621.

[18] T. Maxworthy, On the formation of nonlinear internal waves from the gravitational
collapse of mixed regions in two and three dimensions, J. Fluid Mech. 96 (1980),

47-64.

[19] H. Ono, Algebraic solitary waves in stratified fluids, J. Phys. Soc. Japan 39 (1975),

1082-1091.

[20] J. J. Stoker, Water Waves, Interscience Publishers, Ine., New York, 1957.

[21] E. C. Titchmarsh, Introduction to the Theory of Fourier Integrals, 2nd ed., Oxford

Univ. Press, London, 1975.

[22] K. Yosida, Functional Analysis, Springer-Verlag, New York, 1980.

32