J . Fluid Mech. (1981),vol. 112, p p .

225-251 225
Printed in Great Britain

An investigation of internal solitary waves

in a two-fluid system
By C. G A R Y KOOP AND G E R A L D BUTLER
Fluid Mechanics Department, TRW/DSSG, One Space Park,
Redondo Beach, California 90278

(Received 1 August 1980)

The results of an experimental investigation dealing with finite-amplitude internal

solitary waves in a two-fluid system are presented. Particular attention is paid to
characterizing solitons in terms of their shape and amplitude-wavelength scale
relationship. Two cases are considered, viz., a shallow- and a deep-water configuration,
in order to study the depth effect upon the propagational characteristics of the waves.
Comparisons are made between the experimental results and existing internal-wave
theories. I n addition, discussion is presented describing how these existing theories
may be extended to include higher-order nonlinear and viscous effects.

1. Introduction
The study of finite-amplitude effects in internal-wave systems has received a great
deal of attention in recent years from numerous investigators; notably Benney (1966),
Benjamin (1966, 1967), Davis & Acrivos (1967), Ono (1975), Kubota, KO & Dobbs
(1977)) among others. The recent observations of large-scale solitary wave motions in
both the atmosphere (Christy, Muirhead & Hales 1978) and in the ocean (Osborn &
Burch 1980; Osborne, Burch & Scarlet 1978) have generated additional interest in
this phenomen0n.t Much of the theoretical work which has been done has been
concerned with analysing fluid motion in systems where the internal waves are weakly
nonlinear and are in some sense long relative to the overall depth of the fluid. Following
what is, by now, a familiar procedure, one may describe these wave motions by
balancing nonlinear and dispersive effects in the governing equations. Such a pro-
cedure leads directly to an expression governing the long-time evolution of the stream
function, which is the Korteweg-de Vries equation. This equation has been studied a
great deal within the past decade and methods have been established to construct
exact analytical solutions for arbitrarily prescribed initial conditions (cf. Segur 1973).
I n addition to these shallow-water theories, effort has been spent in analysing non-
linear internal wave motion in fluids of infinite extent (Benjamin 1967; Ono 1975).
Here, long waves are measured relative to an internal length scale characteristic of
the thermocline region, rather than the total fluid depth. For this infinite-depth
problem, dispersion enters a t lower order, and the resulting stream-function evolution
equation has a dispersive term which is a Hilbert transform. This contrasts with the
triple-derivative term found in the Korteweg-de Vries equation. Recently, Kubota
t An interesting account of the practical difficulties associated with oil-drilling activities in
the presence of 100 m internal solitary waves is presented in OiE and Gas JoumaE, 1 7 September,
1979, pp. 66-67.
226 C. G . Koop and G . Butler
et al. (1978) derived an evolution equation which interpolates between these shallow-
and deep-water limits. This equation, which is valid provided the thermocline thick-
ness is sufficiently smaller than the total fluid depth, has a dispersive term that is a
transcendental integral operator, and which reduces to a triple derivative and a Hilbert
transform in the shallow- and deep-water limits, respectively.
In contrast to the great deal of analytical work which has been done, there appears
to be a dearth of experimental study dealing with finite-amplitude effects on long
internal waves. Walker (1973), in an investigation of internal interfacial waves in a
two-fluid system, presents data over a limited range of depth ratios, but these results
are somewhat inconclusive due to the dominance of viscous effects and the small tank
size used. Yates (1978) recently made measurements of nonlinear wave motions in a
continuously stratified system. However, a majority of these experiments were made
a t fairly large wave amplitudes, and no definite conclusion could be drawn regarding
the accuracy of any of the aforementioned weakly nonlinear theories. More recently,
Kao & Pao (1979) presented experimental measurements of solitary-wave propagation
for a thin pycnocline stratification. These results, however, were confined to shallow-
water conditions and no attempt was made to establish whether the spatial scales of
the waves, correlated with some measure of the wave amplitude, were consistent with
the shallow-water theory.
The primary motivation for the present investigation is to try and fill this experi-
mental void and provide a quantitatively reliable data base, upon which one may
assess the validity and determine the regions of applicability of the various theories
dealing with finite-amplitude internal waves. Our particular concern will be to study
how the important characteristics of these waves are altered, owing to the changing
importance of the dispersive terms, as one proceeds from a shallow-water system to
one which is in some sense deep. In pursuing this problem, it is felt that the most
convenient experimental approach is to consider pulse-like permanent waveform
disturbances whose characteristics can be readily compared with the analytic solitary-
wave solutions of the relevant evolution equations. I n these comparisons between the
theories and the experiments, special emphasis will be placed upon some of the geo-
metrical features of the disturbances, such as profile shape, and the functional relation-
ship between the amplitude of a solitary wave and its wavelength. As will be discussed,
we feel that this amplitude-wavelength scaling represents the most fundamental
difference between the various theories, and experimentally is the most definitive way
of establishing the domain of validity of each theory. A less sensitive criterion would
involve the use of phase speed measurements, but as will be shown such measure-
ments would have to be extremely precise in order for the data to be useful in dis-
criminating between theories.
I n addition t o the aforementioned discussion, a secondary intent of this investiga-
tion is to discuss how some of the existing internal-wave theories may be extended to
include higher-order nonlinear effects or viscous effects, and to indicate when these
corrections are important.
 Internal solitary waves in a two-juid system 227

2. Theoretical description
2.1. Review of existing theoretical analyses
All of the existing analytical work relevant to the nonlinear internal-wave studies
conducted in the present investigation may be roughly categorized as follows:
(a) shallow-water theory (Benjamin 1966)
h / H B I, h/H = O(I);
( b ) deep-water theory (Benjamin 1967; Ono 1975)
h/H-+O, h/h B 1 ;
( c ) finite-depth theory (Kubota et al. 1978)
h / h B 1, h / H < 1;
where h is a measure of the horizontal extent of the wave, h is an intrinsic length scale
associated with the density stratification (e.g. the thermocline thickness) and H is
the total fluid depth.
All of these theories may (at least to first order) be cast in the framework of a
generalized evolution equation commonly known as Whitham’s equation which is
written as

Here, y(x,t) measures the internal wave displacement field, c ( k ) is the linear dis-
persion law (with co being the linear long-wave phase speed), and c1 is a functional of
the k = 0 eigenfunction.
For shallow-water internal waves, Benjamin (1966) shows that the linear dispersion
law expanded for small wavenumbers Ic has tlie quadratic dependence

and, for this case, Whitham’s equation reduces to the familiar Korteweg-de Vries
(KdV) equation
all
+ corx + c 1 r r x + c2rzzx = 0.
(The actual definitions o f c 0 ,c1 and c2 may be found in the cited reference and are not
repeated here.) The solitary-wave solution to this equation has been known for some
time and is given by
x - ct
q ( x- ct) = a sech2- (3a)
A ’
where
= Go+>, ac ah2 = 1 2C2 .
3 C1

Here, a is the maximum displacement and c is the phase speed correct to O ( a / H ) .

From equation ( 3 c ) one observes that the solitary-wave amplitude-wavelength
scaling required for a wave to be described correctly by the shallow-water theory is
228 C. G. Koop and G.Butler
Benjamin (1967) and Ono (1975) consider the problem of an infinite-depth fluid,
where the internal wave is long relative to the intrinsic length scale h, associated with
the density stratification. Waves in this system are more dispersive than shallow-
water waves, the expanded dispersion law having linear rather than quadratic
dependence upon k, and the resulting evolution equation (often called the Benjamin-
Ono equation) is written

where H ( v X xdenotes
) the Hilbert transform of qxx.
The solitary-wave solution to this equation has been found by Benjamin (1967) to
be the Lorentzian profile given by
ah2
r(X-Ct) =
(x- C t ) Z + h2’

It is of interest to note that the character of the solitary-wave amplitude-wavelength

scaling required for this deep-water system, viz.

is distinct from the shallow-water scaling given by equation ( 3 d ) . Much of the sub-
sequent discussion regarding the experimental results will focus upon this distinction.
We remark here that, in our opinion, this amplitude-wavelength scaling is the most
fundamental difference between the shallow- and deep-water theories, and represents
the most sensitive way of establishing the domain of validity for each theory. For
parameters typical of this experimental investigation, wavelength predictions of the
two theories may differ by more than 100 % for a given wave amplitude. Thus, high
experimental accuracy is not a requirement for discrimination between the theories.
Furthermore, regression statistics (particularly on log-log plots) utilizing the entire
data base may be used to establish whether the KdV scaling A/h s-* or the N

Benjamin-Ono scaling h/h E--1 (where E = a/h) is most appropriate for a given
N

system. Of lesser concern, we feel, is the degree to which the two theories are able to
predict the experimentally measured phase speeds. This is because both theories
predict a linear dependence of phase speed upon wave amplitude. Thus, to discriminate
between the two predictions one must be able to resolve slight differences in the slopes
of the amplitude-phase-speed relationship. We emphasize that these differences are
small. For parameters typical of our experiment, phase-speed measurements would
have to be made to an accuracy of about 1 % before differences between the two
theories could be resolved. As a consequence, we have chosen to use the amplitude-
wavelength scale relationship as the primary basis for our assessments.
In addition to the KdV and Benjamin-Ono theories, Kubota et al. (1977) derived
an evolution equation governing the propagation of long waves in fluids which are
neither shallow nor deep, and in which the thermocline is much smaller than the total
 Internal solitary waves in a two-fluid system 229

fluid depth. I n this case, the dispersion law has the transcendental wavenumber
dependence
cm CO
N kcothkh,

and the finite-depth evolution equation resulting from this dispersion law is given by
a7 22
a t + ~ o q r , + c l ~ ~ , +-
c
22x2j1,,, q(x’,t) [
77
coth-(x-x‘)-sgn-
2H
(x- x’)
H
]
dx’. (6)

Kubota et al. note that the finite-depth equation reduces to the KdV and Benjamin-
Ono equations in the shallow- and deep-water limits, respectively.
The solitary-wave solution to this equation has been found by Joseph (1977) to be

v(x-ct) = a/(cosh2 h
where
c=co-2c2
2nH
[1--cot- 27
,
H
ahcot- = --.8c2
h C1

As can be seen from equation (7c), the solitary-wave amplitude-wavelength scaling

in a finite-depth system is transcendental, which contrasts with the algebraic rela-
tionships derived in the KdV and Benjamin-Ono analyses.’
It is worth noting at this point that all of the previously described evolution equa-
tions possess soliton as well as solitary-wave solutions, as demonstrated by Xegur
(1973) (KdV), Meiss & Pereira (1978), Chen, Lee & Pereira (1979) (Benjamin-Ono)
and Kubota et al. (1978) (finite depth).
All of the analyses which have been discussed up to this point are first-order theories,
so that terms of order (alh) have been retained in the analysis but the second-order
nonlinear and dispersive terms are neglected. However, the experimental data (to be
discussed) involve waves having amplitudes as large as a/h z 0.6. For these distur-
bances the assumption of weak nonlinearity might no longer be valid. To assess what
effect the inclusion of higher-order terms has upon the solitary-wave amplitude-
wavelength scaling, we have extended the KdV analysis for a two-fluid internal wave
system t o O(a,/h)2(details of the derivation and definition of the various constants
are presented in appendix A). The resulting evolution equation takes the form
27
+ coqz + c1rrz + czrzzr + c3752 + c4(V/xz)z + C5(Y3), + d 7 3 z = 0- (8)

The solitary-wave solution to this extended evolution equation is given by

q(x, t) = a sech2-
h
[
-ct 1 + C tanh2 -
h
The solitary-wave amplitude-wavelength relationship is given by
230 C. G. Iioop and G. Butler
and (for later reference)

where a(1),a(2)and C are defined in appendix A. We remark here that the analysis
presented in appendix A considers both the rigid lid and free surface boundary con-
ditions. For conditions typical of the present experimental configuration, however,
the quantitative differences between the two analyses is not large. For example, in
the shallow-water configuration (discussed in the next section), a t h/h = 10 the
second-order rigid-lid theory yields a / h = 0.0676, whereas in the free-surface analysis
a/h = 0.0579. For comparison, the first-order results are a/h = 0.0577 (rigid lid) and
a/h = 0.0552 (free surface). For consistency with the finite-depth theory (which
incorporates the rigid-lid boundary condition), all of the subsequent KdV calculations
will be based upon the rigid-lid analysis.

2.2. Viscous effects

Preliminary experiments, which are described in 93, revealed that in the present
facility the influence of viscosity upon the internal-wave propagation characteristics
could not be neglected. It was observed, for example, that, for conditions typical of a
majority of the experiments, viscous stresses produce a 50 per cent attenuation in the
internal-wave amplitude as the disturbance propagates one length of the tank. Clearly,
the previously discussed inviscid theories are incapable of describing such phenomena,
and this provided the motivation for a theoretical effort designed to study analytically
the effect viscosity has upon the experimentally generated internal waves.
One simple model which proves to be useful is Keulegan's (1948) analysis for the
gradual viscous damping of surface gravity waves. This model has been modified to
include internal interfacial waves in a two-fluid system. I n essence, the theory assumes
that the energy dissipation in the viscous boundary layers along the solid surfaces
and within the interface region is balanced by a net decrease in the solitary-wave
amplitude. The compensation for this amplitude attenuation is an increase in hori-
zontal extent of the wave such that the fundamental KdV scale relationship
h / H = ae-4 is preserved. Details of the analysis are presented in appendix B. The
resulting expression governing the solitary-wave amplitude decay is given by

where

(One may refer to appendix B for definition of the various parameters appearing in
the above expressions.) These equations reduce to the expressions derived by
Keulegan (1948) (with one small proviso as noted in appendix B) in the limit p2+ 0.
The preceding expressions present a convenient way of calculating wave amplitude
attenuation, providing the initial disturbance is a solitary wave and one is considering
a two-fluid system. It is also of interest, however, to describe the evolution of internal
waves emanating from arbitrary initial conditions and propagating through a viscous
medium having a variable density gradient. A related problem, viz. the evolution
 Internal solitary waves in a two-$uid system 231

equation for a shallow-water surface gravity wa.ve in a viscous fluid, has been derived
by Kakutani & Matsuuchi (1975) and solved numerically by Matsuuchi (1976). This
analysis has been extended here to include internal-wave systems. The details of the
analysis are presented in appendix C. The resulting equation governing the evolution
of the stream function
1c.b2,t ) = fft,7)$1).
is given by

where c1 and c8 are the same coefficients defined for the inviscid KdV equation and

(The various quantities are defined in appendix C.) Although no exact solution to the
above equation has been identified analytically, solutions have been obtained numeric-
ally using the pseudo-spectral technique of Fornberg (1977). Presentation of the
results of these calculations, however, is deferred until after discussion of some of the
experimental results.

3. Experimental results and comparison with theory

3.1. Experimental apparatus and procedure
The experiments were performed in the TRW internal-wave facility which is shown
schematically in figure 1 . The wave tank, constructed from 4in Plexiglas, is approxi-
mately 6 metres long and has a 45 ern x 60 cm cross-section. The lower boundary of
the tank consists of a specially constructed false bottom, fabricated with reinforced
blancher ground tooling plate, which was introduced in order to minimize variations
in bottom topography. Over the 6 metre length of the channel, the average variation
in bottom elevation is less than about 0.02 cm. Note that there is no wave-absorp-
tion device a t the downstream end of the tank, so that waves are free t o reflect off the
end wall and propagate back through the test section.
The fluid system chosen for use in experimentation consists of two mutually
immiscible liquids having different specific gravities. Several considerations went into
the choice of this system. First of all, the two-layer stratification is an extremely
simple system, and it is believed that there is no advantage in using a system which
is, in some sense, more complicated (e.g. continuous salt stratification). All of the
physics relevant to the present study, vix. nonlinearity, dispersion, ability to vary
overall fluid depth, existence of an intrinsic internal length scale, etc., are contained
in the simple two-fluid system. Secondly, the two-fluid system is particularly well
suited for providing quantitatively reliable experimental data, because the interface
is always well defined and experiments are quite repeatable. As a by-product of
having an int,erface which is, for all practical purposes, infinitesimally thin, wave-
amplitude measurements may be made in a straightforward and unambiguous manner
232 C . G . Koop and G.Butler
Probe: 5 b 5 a 4 b 4a 3b 3a 2b 2a 1

, FIGURE
6m _I
1. Schematic diagram of experimental apparatus.

using capacitance gauges, and since each displacement measurement requires the use
of only a single probe (in contrast to the probe arrays required in continuously strati-
fied fluids) one is free to use several such probes longitudinally displaced in order to
study the spatial evolution of a disturbance.
The immiscible fluids chosen for use in this investigation were Freon T F (specific
gravity 1.58, v = 0.0044 cm2 s-l) and deionized water. To keep the fluid system as
clean as possible, both the water and the Freon were filtered using 5 pm filters prior
to being introduced into the tank. I n addition, a surface skimmer was utilized in
order to ensure that the interface between these two fluids remained clean during the
experiments.
Interfacial disturbances were generated using a displacement-type wavemaker,
similar to that described by Walker (1973). I n the quiescent state, the wavemaker
paddle straddles the interface, as depicted in figure 1. Pulse-like disturbances are
generated by imposing a single downward displacement of the paddle. This is accom-
plished by coupling the paddle to a geared drive motor through a scotch yoke. The
drive motor, gear reduction unit, and the various moving parts are all mounted on
structures attached directly t o the floor of the laboratory. This assures that vibrations
associated with operation of the wave-generating mechanism are decoupled from the
tank itself.
The wavemaker has two degrees of freedom, viz. the length and duration of the
paddle stroke, and these may be varied between 0-5 cm, and 0.5-5 seconds, respec-
tively. It is important to recognize that the wavemaker does not have sufficient
degrees of freedom to produce arbitrarily prescribed waveforms (e.g. solitary waves)
as initial conditions to the experiment. It can only produce pulse-like disturbances
which, in general, are not solitary waves. But, as previously noted, all of the evolution
equations being studied in this investigation admit soliton solutions, so that the
permanent waveform disturbances desired for the experiment were generated by
adjusting the amplitude and duration of the paddle displacement (in an empirical
fashion) such that a single soliton (plus a dispersive tail) emerged downstream of the
wavemaker. Because the phase velocity of the soliton is greater than the group
velocity of the tail, it separates from the oscillatory portion of the signal. Once the
soliton has evolved and separated from the dispersive wavetrain, one is able to
examine the important characteristics of this permanent waveform disturbance as it
propagates through the test section. Following the aforementioned procedure, soliton
disturbances in the amplitude range 0.025 < E < 0.5 are readily generated in this
facility.
 Internal solitary waves in a two-$uid system 233
A 2 L

Incident wave t- Reflected wave

FIGURE
2. Definition of wavelength.

Measurement of the interfacial displacement were made using nine longitudinally

displaced single-element capacitance gauges of the type described by Lake et al. (1977).
These were mounted on five separate traversing mechanisms positioned above the
water surface. The first probe was positioned 20 cm downstream of the wavemaker in
order to record the initial disturbance created by the paddle displacement (see figure 1
for location and nomenclature of the remaining probes). The paddle displacement was
monitored using a linear variable-displacement transducer (LVDT) mounted on the
wavemaker. During the course of a run, the outputs of the nine capacitance gauges
plus that of the LVDT were simultaneously recorded on a multichannel strip chart
and on a 14-channelFM tape recorder.
Typically, experiments were initiated by starting the strip chart and tape recorder
with the fluid in a quiescent state. The wavemaker was then actuated to yield a
single downward thrust of the paddle. The resulting initial interfacial disturbance
was recorded by the first probe, probe 1. Subsequently, this initial pulse-like wave
evolved into a soliton followed by a dispersive wavetrain. These propagated down the
tank past the eight data probes, probes 2a-5b, reflected off the endwall, and propag-
ated back through the test section in the opposite direction. When the reflected wave
passed probe 2a, the experiment was considered to be completed, and the monitoring
equipment was shut off. Once the fluid had returned to a quiescent state (roughly
1-2 minutes later), the nine capacitance probes were calibrated by traversing them
through the interface and correlating the probe output with its vertical position. The
total elapsed time between the start of the experiment and the completion of the
calibration was normally less than 5 minutes.
The above procedure was repeated several times using different values of the
paddle displacement to generate various-amplitude waves. Following such a series of
experiments, the FM analog tape was digitized at a rate of 100 samples/s per channel
(real time), and the resulting digital data were stored for numerical processing.
Typical quantities which were computed in the data-reduction program included the
interfacial displacement (in engineering units) versus time for all nine capacitance
gauges, and the maximum amplitude and horizontal extent of the incident and
reflected waves recorded by each probe. Figure 2 illustrates the manner in which the
amplitude and wavelength were computed. Note that only the forward portion of the
soliton profile was used, since conditions in the ‘wake ’ of the disturbance were never
truly quiescent. Two definitions of the disturbance wavelength were used. The first,
denoted hW5,simply represents the half-amplitude point and is defined by
234 C . G. Roop and G . Butler
A second definition, denoted A,, was also used and is given by

Being an integral quantity, it is felt that A, provides a better (i.e. less noisy) measure
of the soliton wavelength than does the half-amplitude point. This turns out to be an
important consideration in analysing the results, since most of the data scatter is
introduced through the experimental determination of the soliton wavelength. A, is
related to the A's usedin the KdV, Benjamin-Ono, and finite-depth analyses (equations
(3a ) , (5a ) and (7 a ) respectively) as follows :
KdV, A, = AKdV;
Benjamin-Ono, A, = &A,,o;
Finite depth, A, = H cot H/A,,.

Two test conditions were chosen for consideration so that the depth effect upon
the soliton amplitude-wavelength relationship could be investigated. These are
identified as follows :
Configuration
Shallow water
' = hFreon
1.366 cm
= 'Freon
8.314 cm
+ 'water
6.086
Deep water 1.366 cm 49.236 cm 36.046

3.2. Qualitative results and discussion of viscous effects

The purpose of the present investigation is to study how the propagation charac-
teristics of internal solitons are altered as one proceeds from a shallow-water system,
presumably governed by the KdV equation, to a deep-water regime where the
Benjamin-Ono analysis should be more appropriate. It has been previously empha-
sized that our primary concern is the study of permanent waveform disturbances. Of
less importance for our purposes is what might be termed the period of evolution or
generation; i.e. that period of time when an initial pulse-like disturbance (which is not
a soliton) evolves into one or more solitons followed by a dispersive wavetrain. From
an experimental point of view, it is important to ensure that this transient period of
evolution is not long, relative to the time required for the disturbance to propagate
to the end of the tank. Figure 3 presents the results of a preliminary test where the
above criterion was not satisfied. The measurement was made using a Freon depth of
about 7 cm, an overall fluid depth of about 40 cm. A single capacitance probe was
positioned roughly 4 m from the wavenumber. Time increases to the right in this
figure. The several pulses depicted in these records are a result of multiple reflections
of the initial disturbance off the endwalls of the tank. The interesting feature to be
observed in this figure is the extremely long evolution time required before a soliton
is produced. Note, for example, that the initial pulse (whichis not a soliton) propagates
at least 36 m or roughly 6 tank lengths before one could say that the leading soliton
is reasonably well separated from the remaining portion of the wavetrain. One also
observes from this figure that the wavelength of the disturbance is quite long; being
almost 50 yoof the length of the tank.
Obviously, such features are undesirable from the standpoint of studying soliton
characteristics, owing to the extremely long length of the disturbance (relative to the
 Internal solitary waves in a two-$uid system 235

x = ct
= 2 tank lengths

=1 0 2

=1 2 2

t increasing
___)
=1 4 3
FIQURE
3. Results of a preliminary test; h N 7 cm, H N 40 cm.
236

Probe

46

4a

36

i--+tl-*: 'I - I I I I : I : I 1 : ! I 1 I !

3a

2b

2a

FIGURE
4. Example of multiple-soliton generation; shallow-water configuration.

length of the tank) and the predominance of endwall effects. These problems may be
alleviated by sufficiently decreasing the fluid layer depths. Hammack & Segur (1974)
present a simple argument for estimating the sorting time required for a soliton to
separate from the dispersive portion of the wavetrain. Using their model (modified
 Internal solitary waves in a two-Jluid system 237

3a

2b

2a

FIGURE
5. Example of single-soliton generation ; shallow-water configuration.
238 G. G . Koop and G . Butler

26

FIGURE
6. Example of the generation of a dispersive wavetrain; shallow-water configuration.

for an internal wave system), one estimates that, for a Freon layer depth of 1.5 em,
the sorting time for a typical wave in the present facility should be less than the time
required for the pulse to propagate 50 per cent of the length of the tank. This is con-
sidered to be acceptable, and all of the remaining experiments were conducted using
this nominal value for the depth of the Freon layer.
Having, in some sense, identified the best configuration for testing, several experi-
ments were performed using the probe arrangement depicted in figure 1 to study
qualitatively the evolution of the various types of disturbances which can be produced
in this facility. Some typical results of these tests are presented in figures 4-6. The
test conditions for these runs are those of the shallow-water configuration identified
in 5 3.1. Time increases to the left in these figures, and the second wave in each record
is the reflection of the incident wave of€the endwall of the tank. It should be pointed
out that the sensitivities and gain settings of the various probes are not the same, so
that relative-amplitude measurements between probes are not meaningful. Figure 4
presents a case where multiple solitons are formed from the initial condition. One
observes here that a t least two (and possibly three) solitons emerge. Note that, as the
disturbance propagates down the tank, the relative spacings between adjacent peaks
irlcreases; a result of the difference in nonlinearity between the various solitons.
Figure 5 shows the formation of a single soliton, followed by what appears to be a
dispersive wavetrain. The data shown in this figure typify those which were used to
obtain the quantitative results described in 333.2 and 3.3.
 Internal solitary waves in a two-Juid system 239

1.2 Inviscid KdV

Viscous KdV
(equation (1 1))

Xlh
FIGURE7 . Viscous decay of interfacial wave amplitude. Initial conditions:
measured profile for case C. 0, case A ; 0,case B ; A,case C .

Finally, figure 6 shows a case where a dispersive wavetrain, but no soliton, is formed.
The initial condition for this disturbance was a wave of depression, which was pro-
duced by reversing the direction of the wavemaker. That only an oscillatory waveform
is generated from this initial condition is qualitatively consistent with the results of
Hammack & Segur (1974) (their figure 7 ) and is predicted by the inverse-scattering
solution of the KdV equation.
If we re-examine the single-soliton example shown in figure 5 , it is found that, in
actuality, a wave of permanent form is never realized experimentally. Note, for
example, in the record for probe 2a, that as a result of viscous dissipation the ampli-
tude of the reflected wave is only about 30 per cent of the incident-wave amplitude.
Clearly, viscosity has a non-negligible influence upon the propagation characteristics
of the waves being studied in this investigation. To quantify the amount of such
viscous attenuation, experiments were performed where multiple reflections off the
endwalls were monitored in order to increase effectively the distance travelled by a
given disturbance. The amplitude history of three such experiments (normalized by
the incident-wave amplitude measured at probe 2a) is shown in figure 7 . Cases A
and B represent data obtained in the deep-water configuration with initial amplitudes
eOa= 0.305 and 0.109 respectively. Case C represents shallow-water data with
eZa= 0.269. From these data, one sees that viscous dissipation acts to attenuate the
wave amplitude by roughly 50 per cent per pass through the test section. These data
are also useful as a basis for testing the viscous theories described in 8 2.3. For example,
figure 7 presents the numerically calculated solution to the viscous KdV
equation (1l),using as an initial condition the measured wave form recorded by probe 1
(20 em from the wavemaker). Two calculations were performed, one with and one
240 C. G . Koop and G . Butler

1
0.04 0.06 0.08 0.1 0.2 0.4 0.6 1 2
E = afh

FIGURE
8. Effects of viscosity upon amplitude- wavelength scale relationship.
__ , inviscid theory; - - -, viscous theory.

without the viscous term, and the output included the maximum disturbance ampli-
tude as a function of propagation distance. From the results of the inviscid calculation,
one finds that during the initial evolution period (when the leading soliton is being
formed) the maximum amplitude grows. However, at about x/h equal t o 80, the
soliton has separated from the rest of the waveform, and beyond this point the
maximum amplitude is constant. Examining the second calculation, one can clearly
see the effect viscosity has upon the evolution of the initial disturbance. One finds,
for example, that, for x/h less than about 10, the peak amplitude grows, but beyond
this point viscosity becomes important, and the amplitude is attenuated by roughly
40 per cent before the solution merges into the data. Numerical difficulties prevented
continuation of this calculation beyond about x/h = 200, but application of Keulegan’s
modified analysis, given by equation (lo), yields reasonably good agreement with the
remaining portion of the data.?
From this demonstration of the importmaneeof viscosity on the disturbance ampli-
tude, it is natural to ask how viscosity affects the soliton wavelength. This is an
important question. Since the propagation characteristics of the interfacial waves
are to be characterized in terms of their amplitud+wavelength relationship, it is
necessary t o establish whether this relationship has any functional dependence upon
viscosity. I n a qualitative sense, one may reason that, for a sufficiently large-amplitude
soliton, the important dynamical balance lies between the nonlinear and dispersive
terms in the governing equations. Viscosity should be important only in that it acts
t o attenuate the wave slowly, but should not directly influence the soliton scaling
given by ah2/h3= constant (this may be considered to be a quasi-steady approxima-
tion). It is clear, however, that, as viscosity acts to decrease the disturbance ampli-
tude monotonically, eventually, the strengths of the viscous and nonlinear terms
will become of comparable order, At this point, the quasi-steady approximation
becomes invalid, and viscosity must enter directly into the scaling. The degree to
which this scale relation is affected by viscosity is not easily predicted analytically,
t Note added in proof: use of the more-accurate expression of Hammack, Leone & Segur
(1981) yields slightly better agreement with these data.
 Internal solitary waves in a two-Juid system 241

1 2 3 4
Xlh0.5

FIGURE shallow-water configuration. - -0- -, ‘seeha’;

9. Profile shape r](z)/aversus z/hO.&;
-_ x-- Lorentzian. Data compiled from runs: 4125-1, E = 0.04; 4125-2, E = 0.05;
4125-1, E = 0.11; 4124-3, E = 0.18; 4125-6, E = 0.26; 4129-1, E = 0.57; 4129-1, E = 0.68.

but we may again appeal to the numerical solutions of the viscous KdV equation and
theoretically quantify what effects should be found in the present facility. The results
of such a calculation are shown in figure 8 where h,(t)/h is plotted versus e(t) (with
time appearing parametrically). For reference, the inviscid scaling ah:/h3 = constant
is also presented. The initial condition for this calculation was a sechz profile (with
e(t = 0) = 0.4)having the proper solitary-wave amplitude-wavelength relationship
as predicted by the inviscid theory. From these calculations, one finds that, for e(t)
greater than about 0.2, the amplitude-wavelength history follows the inviscid line,
verifying the quasi-steady nature of the problem. As the amplitude decreases, how-
ever, viscosity begins to alter the scale relation in such a manner that h,/h grows
more slowly than €4.The effect is not unduly large, though (except for very small
waves, viz. e < 0.02)) and we conclude that viscous effects on the amplitude-wave-
length scaling in the present facility are sufficiently small that meaningful measure-
ments of this relationship may be obtained and compared with inviscid theoretical
results.
3.3. Quantitative results
I n the shallow-water configuration ( H l h = 6-086, Ap/p = 0.33) tweIve runs were
made to examine the profile shape and amplitude-wavelength scale relationship for
nonlinear internal solitons. Each run consisted of 16 realizations of a waveform
(8 probes measuring incident and reflected waves), so that the shallow-water experi-
ments yielded almost 200 profile measurements. I n a strict sense, however, not all of
these measurements may be considered to be totally independent, since any one run
yields 16 measurements of the same disturbance, although one must remember that
this wavo is slowly varying in time due to viscosity, so that it is not exactly the same
wave a t each measurement station. The amplitudes of these waves varied over a
242 C . G . Koop and G . Butler

IUU t

.
-E
2
10

1 .o
0.0 1 0.1 1
E = a/h

FIGURE10. Amplitude-wavelength scale relationship; shallow-water configuration. (Solid

symbols represent reflected waves.)

decade between 0.025 < E < 0.75 with corresponding wavelength variations
4 < h,/h < 15. Figure 9 presents the measured wave profile shape, r(x)/aplotted
versus xlh,,.,. I n order to provide some estimate of the variance between profiles,
seven individual wave-form realizations in the amplitude range 0.04 < E < 0.7 are
presented. For reference, both the ‘sechz’ and Lorentzian profiles are also shown in
this figure. Examination of the results reveals that the agreement between the
shallow-water data and the ‘sech2’profile predicted by the KdV theory is quite good.
Such agreement between the theoretically predicted and experimentally measured
wave forms is encouraging; but we have not subjected the theory to a very stringent
test, because such profile comparisons introduce the artificial constraint of requiring
the theory and experiment to agres at xlh,., = 0 and I . For wave profiles that are
reasonably similar, one would not expect large differences to occur between the two
curves. A stronger test of the theory would be to establish whether the theoretically
predicted one-parameter family of such ‘sech2’profiles or, in essence, the amplitude-
wavelength relationship, is in agreement with the data. This is most conveniently
displayed on a log-log scale, as shown in figure 10,where h,/h is plotted versus E .
These data represent amplitude-wavelength measurements of both the incident and
reflected waves. As there may be some question rogarding the validity of the reflected-
wave data, since these waves are in effect seeing their own ‘wake’, these data are
identified using solid symbols. I n general one finds, though, that the incident and
reflected wave data behave in a reasonably similar manner. Also shown in figure 10
are the theoretical scale relations predicted by the KdV, Benjamin-Ono and
finite-depth analyses. Several comments may be made regarding the data shown
in this figure. First of all, it is not unreasonable to expect that the Benjamin-Ono
infinite-depth theory should poorly describe the results of a shallow-water experi-
 Internal solitary waves in a two-&id system 243

Xlh0.5

FIGURE 1 1 . Profile shape ~ / ( z ) /versus

a x / A 0 5 ; deep water configuration. - -0- -, ‘seeha’;
_ _ x - -, Lorentzian and finite depth ( E = 0.6); - -a- -, finite depth (8 = 0.06). Data
compiled from runs: 4/27-3, E = 0.06; 6/1-7, E = 0.08; 6/1-4, E = 0.11; 4/27-2, E = 0.11;
6/1-1, 6 = 0.17; 6/1-2, E = 0.26; 6/1-2, E = 0.56.

ment. Indeed, this is observed in figure 10 where one finds that the slope of the
Benjamin-Ono line is in total disagreement with the data. The finite-depth theory
does better, but the shallow-water KdV theory yields by far the best agreement with
the data. The agreement, though, is by no means exact. One notes, for example, that
for e less than about 0.05 the theory slightly overpredicts the experimentally measured
wavelengths by 10-15 per cent,. This is not inconsistent with our previous discussion
(3 3.2) regarding the effects of viscosity on the amplitude-wavelength scale relation-
ship. There it is shown that numerical solutions of the viscous KdV equation ( 1 1 )
yield wavelengths which are somewhat smaller than those the inviscid theory would
predict for values of e less than about 0.1. One suspect,s, then, that, for the very-
small-amplitude portion of the data, the nonlinear effects have been weakened to such
an extent that the viscous terms are of comparable order. This argument seems
plausible, but could be significantly strengthened by additional experimentation.
Over the remaining portion of the data the slope of the KdV theory and the data are
reasonably close, but for e greater than about 0.2 the theory slightly underpredicts the
measurements. However, the second-order KdV theory, derived in 3 2.2, shows
noticeably better agreement with the data, indicating that the small discrepancy
between the data and the first-order theory for e > 0-2 is likely due to higher-order
nonlinear effects.
Upon completion of the shallow-water experiments, the overall fluid depth ratio
was increased to H l h = 36.044 and eight additional experiments were conducted to
ascertain what effect the increased depth has upon the amplitude-wavelength scale
relationship. The range of amplitudes studied varied between 0-03 < E < 0.6 with a
corresponding span of wavelengths 4 < h,/h < 20. Note that‘ in terms of the total
244 C . B. Koop and G.Butler

Benjamin-Ono (infinite depth)

.
<
x‘ 10

1st order KdV

1 .o
0.01 0.1 1
E = a/h

FIGURE
12. Amplitude-wavelength scale relationship ; deep-water configuration. (Solid symbols
represent reflected waves.)

fluid depth H , 0.11 < h,/H < 0.56. Hence, for these experiments the depth is always
greater (and in some cases much greater) than the wavelength of the disturbance.
Figure 11 presents the measured deep-water profile shape, qlh,.,. Again, to provide
some degree of statistical significance seven individual wave records in the amplitude
range 0.06 < E < 0.6 are presented. For reference both the ‘sech2’and Lorentzian
profile shapes are shown. In addition, wave profiles corresponding to Joseph’s (1977)
solitary-wave solution to the finite-depth equation are also presented. The two finite-
depth profiles depicted are for E = 0.06 and E = 0.6; i.e. the range of E over which the
data were compiled (note that the B = 0.6 finite-depth profile is indistinguishable from
the infinite-depth Lorentzian profile). We note from this figure that most of the data
lie between the ‘sech2’and Lorentzian profiles particularly in the tails of the waves.
The E = 0.06 finite-depth theory ostensibly yields good agreement with the data, but
actually all the data should lie between the two curves E = 0.06 and E = 0.6, rather
than being centred on the E = 0.06 profile. Thus, one must conclude that even this
relatively weak comparison between the theory and the data, involving the normalized
profile shape, does not yield a conclusive result.
Figure 12 presents the amplitude-wavelength scaling measured for the deep-water
runs, together with the theoretical scaling predicted by the KdV, Benjamin-
Ono, and finite-depth analyses. Again, one sees that none of the three theories
accurately (or even approximately) describes the experimental results. The extent
t o which all of these theories are in disagreement with the data is, perhaps,
somewhat surprising, particularly in light of the reasonably good results obtained in
the shallow-water experiments. For example, one notes that the KdV theory has a
slope which agrees fairly well with the data (on a log-log plot), but overpredicts
h,/h by a factor of almost 2. The second-order KdV theory exhibits much the same
 Internal solitary waves in a two-$uid system 245

behaviour that was observed in the shallow-water configuration, but does not lead
to any improvement in agreement with the data. For the Benjamin-Ono and finite-
depth analysis, one finds that both theories intersect the data, but clearly have the
wrong slope.

4. Discussion of results and conclusions

Based upon the experimental results discussed in the previous section, the following
general observations may be made. First, in the shallow-water configuration the
first-order inviscid Korteweg-de Vries theory agrees in a reasonably quantitative
manner with the experimental data with regards to the profile shape and the soliton
amplitude-wavelength relationship. I n particular, one finds that the profile shape
matches the predicted ‘sech2 ’ profile almost exactly. Furthermore, the amplitude-
wavelength relationship predicted by the first-order theory is valid for wave ampli-
tudes as large as e x 0.2, and the inclusion of second-order nonlinear terms extends
the useful range of the KdV theory to wave amplitudes of e x 0.8. The viscous effects
discussed in 3 3.2 appear to slightly alter the amplitude-wavelength relationship for
small amplitudes, roughly e < 0-1, but these deviations from the inviscid theory are
not large, and presumably could be accounted for using the viscous KdV equation
given by equation (11).I n short, one concludes that the present investigation has
quantitatively validated the adequacy of the Korteweg-de Vries analysis as applied
to internal waves in fluids of limited vertical extent.
Having noted the reasonably good results of the first portion of the investigation,
the obvious question which arises is why the experimental results of the deep-water
experiments are described so poorly by any of the available theories. Since the identical
procedures and data-processing techniques were used on both the deep- and the
shallow-water data, we feel that the deep-water experimental results are not in error.
With regard to the theoretical analyses, the following comments may be made. All of
the theories discussed in this paper consider fluid systems having wave motions which
are both weakly nonlinear and in some sense long. In the Korteweg-de Vries a.nalysis,
‘long’ is measured relative to the total depth of the fluid. In the deep-water configura-
tion, using the theoretical KdV scale relationship, the wavelength h is predicted to be
‘long’ relative to the total depth H (say, h / H 2 10) when e is smaller than about
0.001. Hence, in the deep-water configuration the radius of convergence of the KdV
theory is extremely limited. Since most of the data lie in the amplitude range
0.025 < e < 0.6, the disagreement between the theory and experiment is not
unexpected.
A similar argument may be made regarding the Benjamin-Ono analysis. Here, the
assumption is made that the wavelength is much shorter than the overall fluid depth.
I n the deep-water configuration, this theory would predict that h is much smaller
than H (say AIH < 0.1) when 8 is greater than about 0.5. However, in this region the
weak nonlinearity assumption is violated, and one again concludes that the dis-
agreement with the data probably results from the theory being applied outside of its
domain of validity.
The failure of the finite-depth theory to describe the data adequately is a more
difficult matter to explain. Ostensibly, the important assumptions which are inherent
in this theory are that the waves are weakly nonlinear, and that the thermocline is
246 C . G. Koop and G . Butler
thin relative to the total fluid depth. No constraint is placed upon the wavelength of
the disturbance, other than it must be long relative to the thermocline thickness.
Consider first the weak nonlinearity assumption. If one may extrapolate from the
results of the shallow-water experiments, one would conclude that this assumption
is justified for waves having amplitudes smaller than about E = 0.2. At least a portion
of the deep-water data lie in this parameter range, but even here there is total dis-
agreement between the theory and the experiment.
Regarding the second assumption dealing with the thinness of the thermocline, one
might question whether a depth ratio H / h = 36.044 is sufficiently large to be
classified as a ‘thin’ thermocline. This question may be resolved in the following
manner. I n the finite-depth theory, the thin-thermocline assumption is made in order
to simplify the dispersion relation (for arbitrary stratification) to an analytic form.
I n the two-layer system, however, the dispersion relation is already known in exact
analytic form (cf. Lamb 1932, p. 371). Hence, if one is willing to accept numerical
solutions, one could substitute this exact analytic representation into the dispersive
term of Whitham’s equation, and check to see if the computed results are significantly
different from those predicted by the finite-depth theory. These calculations were
performed, using the pseudo-spectral code previously mentioned, with depth and
density ratios corresponding to those of the deep-water configuration, and wave
amplitudes ranging between 0.01 < E < 1 . For a given value of E , the initial condition
to the program w a s the wave profile predicted by Joseph’s (1977) solitary-wave
solution to the finite-depth equation, and the calculations were run for times between
t = 0 and t = 20 seconds, which roughly corresponds to the time required for a pulse
to make two transits through the test section of the channel. If the thin-thermocline
assumption is a valid approximation in the present deep-water configuration, then
Joseph’s profile should also be the solitary wave solution of the more general equation.
This, in fact, turns out to be the case. The calculations revealed that use of the exact
dispersion relation had almost negligible influence on Joseph’s profile, causing a
modest change in the wave amplitude of typically about 1yoduring the computation
time. On the basis of these calculations, then, one must conclude that the disagreement
between the finite-depth theory and the experimental data is not a result of applying
the theory outside of its domain of validity.
At this time, the present authors do not see a way of resolving this apparent dis-
crepancy between our data and the existing theoretical analyses. We note, however,
that similar results have recently been reported by Hammack, Leone & SeguP (1981).
In this work, the experimentally measured soliton half-amplitude wavelength
( E = 0-18, H l h = 9) is underpredicted by the finite-depth theory by about 2074,
quite consistent with our results. Clearly, further work in this area is required before
a conclusive statement may be made regarding the proper theory to be used in physical
systems where the disturbance wavelength is comparable to the total fluid depth.

The authors gratefully acknowledge several valuable discussions with Dr B. M. Lake

and Dr C. Hindman. We are also indebted to Dr H . Xegur for comments made regard-
ing our viscous theory. Finally, they would like to thank Mr D. J . Rowland for his
help in constructing the experimental facility. This study was sponsored by The Johns
Hopkins University/Applied Physics Laboratory, under Contract no. 60061 3.
 Internal solitary waves in a two-jluid system 247

Appendix A. Extension of the KdV equation to second order

Consider a two-fluid system boundedaboveand below by rigid walls. The lower- and
upper-layer densities are p1 and p,, and the corresponding fluid depths are h, and h,.
The relevant equations and boundary conditions are given by

v2+1,2 = 0, &,(z = h,) = #2,@ = h,) = 0,

Introducing the slow space and time scales

5 = E J ( X - cot), 7 = E%,

where E = a/(h, +h J , and expanding the velocity potential and interfacial dis-
placement as
# = 4#(1)(& 792) + 4 2 ) ( 1 77 2 2 ) + .I,* *

T = %l) (E, 7 )+ E 2 T ( 2 ) ( E , 7)+ * . 3 9

and solving (A 1) recursively, one finds cg a t O ( s ) ,

at 0 ( e 2 ) ,and

The amplitude-wavelength scale relationship is given to second order by

where
248 C . G . Koop and G . Butler
The constant C in (9a) is defined by
C = 3E4/4E2- 15E3d 1 ) / E 2 - Z6/8E2 + E6/2E2, (A 6)
The preceding analysis assumes the upper boundary condition t o be a rigid lid. An
analysis assuming a free-surface condition proceeds along similar lines, and the
resulting amplitude-wavelength scale relationship is given by
a = +$ l + r s [ l - ( l - ~ ) ~ ] / uh,
- 1 + rs[l - (1- u 3 ) ] / u
h, r2- u3s (x)2/(i-+r2 r2+ mu2
where
r
u = - ( 1 - s ) -[I;-,
S
( (-
l;r), -r(l-s) )+I-' --r
S'

Appendix B. Viscous damping of internal solitary waves

Consider a two-fluid system confined between two rigid walls where the inter-
facial displacement 9 is a solitary wave given by

~ ( xt ),= a sech2 (yt)

-,
where
ah2
- = 4 P l , P 2 , h,, h,)
h!
andp,, p,, h,, h, are the lower- and upper-layer densities and fluid depths, respectively.
Following Keulegan (1948), the wave-induced fluid velocity in the wall boundary
layer for the lower layer is given by

where v, is the lower-layer kinematic viscosity. A similar expression may be written

for the upper-boundary-layer flow. The rate of energy dissipation, dEJdt, in the lower
boundary layer is given by (Keulegan 1948)

where B is the breadth of the cha,nnel. The total energy dissipated in the lower layer
includes contributions from the channel side walls and the sheared interfacial region.
The side-wall contribution may be calculated through geometrical parameters. The
interfacial contribution could be calculated by examining the boundary-layer flow
in this region, but, for simplicity, we assume that the contribution from the interfacial
shear region (in the lower layer) is simplyequal to that due to the lower-wall boundary
layer. Although approximate, we feel that this is not too unreasonable, and the
analysis is somewhat simp1ified.t Dissipation in the upper layer is computed in a
similar manner, except that the contribution from the upper-wall boundary layer is
neglected. This is because the upper boundary in the experiment is a frec surface and
should behave as a stress-free boundary.
t Note added in. proof: Hammack et al. (1981) perform the analysis within the interfacial
region.
 Internal solitary waves in a two-$uid system 249
The total energy contained in the wave field, E,, is given by
E, = $BgAph”,.ad,

where = a/h, and Ap = p 2 - p l . Assuming 01 $: a(t)(i.e. a quasi-steady assumption),

we may differentiate this expression with respect to time, and equate this with the
total rate of energy dissipation. After some manipulation, one gets a n expression for
the amplitude decay ofthe solitary wave, given by

where

I n the limit p2+0, the above expression reduces to that derived by Keulegan (one
needs to exclude the contribution from the interfacial region) for the slow viscous
damping of a surface solitary wave.t

Appendix C. Evolution equation for long internal waves in a viscous

medium
The following analysis closely follows that of Kakutani & Matsuuchi (1975). Con-
sider a fluid system with a mean density profile p ( z ) which is confined between an
upper and lower boundary, separated by a distance H ; po is a reference density, g is
the acceleration due to gravity, and the vertical co-ordinate z is zero a t the lower
boundary. The flow field is divided into two inner regions, lying in the neighbourhood
of the viscous boundary layers on the rigid surfaces, and an outer region sufficiently
far removed from the boundaries that viscous effects are negligible.
Introducing the KdV scaling
< = €.a(X-cot), 7 = BQt,

where B measures the wave displacement and co is the linear long-wave phase speed,
expanding the outer dependent variables (denoted do),w(O), ...) as
+ €2up+ ...,
u(0)= € U p

+ €2wp+ ...1,
W(0)= 8 W(0)= €[€wp

and requiring the O($) system of equations to have non-singular solutions yields

= p[$wp)’ - $ h ’ w p ] I ~ o . (C 1 )
Here, primes denote differentiation with respect to z, the stream function
II. 7)$@I, and J ( p , $) is given by
= f(<,

t Note that Keulegan approximates 124/4n$ as &.

250 C . G . Koop and G . Butler
The velocity potential Q is determined through the eigenvalue problem (with as yet
unspecified boundary conditions)
(PQ’)’-L
H - ’Q E 0.
c;
In order to proceed, one needs to consider the inner viscous problem so that boundary
conditions on d, wF), and wko)’may be prescribed a t z = 0,H.
Within the viscous inner regions, the characteristic vertical length scale is the
boundary-layer thickness, which for the present experiment can be shown to be
O ( E ) Thus,
. the appropriate inner variables are defined as
71 = Z / E , 72 = (H--X)/E,
where yl and v2 are the stretched co-ordinates in the lower and upper boundary
layers, respectively. E and r remain as previously defined. Substituting this scaling
into the governing equations, expanding the inner dependent variables in powers
of E , and matching with the outer flow solution yields
$(0) = 0 ,

Wp”(0) = 0,
where RT = ( g H ~ H / v , ) & and
, v1 is the kinematic viscosity at the lower boundary. A
similar expression may be derived for the upper boundary layer.
Substituting (C 2 ) into (C 1) yields the long-wave evolution equation

where

and co and $ are determined through the eigenvalue problem

(pQ’)‘-@-’$ = 0, 4(0) = # ( H ) = 0.
c;
We note here that this model was formulated under the assumption that p ( z ) is a
continuous function. Hence the boundary layer within the interfacial shear region
of the present experiment has not been explicitly included. For the purposes of making
comparisons with the experimental data, we included the interfacial shear effects (as
well as side-wall effects) in an ad hoc fashion by making appropriate modifications to
the coefficient c3 in (C 3), consistent with what was done in appendix B.
 Internal solitary waves in a two-&id system 25 1

REFERENCES
T. B. 1966 Internal waves of finite amplitude and permanent form. J . Fluid Mech.
BENJAMIN,
25, 241.
BENJAMIN, T. B. 1967 Internal waves of permanent form in fluids of great depth. J . Fluid
Mech. 25, 559.
BENNEY,C. J. 1966 Long nonlinear waves in fluid flows. J . Math. Phys. 45, 52.
CHEN, H. H., LEE, Y. C. & PEREIRA, N. R. 1979 Algebraic internal wave solitons and the
integrable Calogero-Moser-Sutherland N-body problem. Phys. Fluids 22, 187.
CHRISTIE,D. R., MUIRHEAD,K. & HALES,A. 1978 On solitary waves in the atmosphere. J .
Atm. Sci. 35, 805.
DAVIS,R . E. & ACRIVOS, A. 1967 Solitary internal waves in deep water. J . Fluid Mech. 29, 593.
FORNBERG, B . E. 1977 On a Fourier method for the integration of hyperbolic equations. S I A M
J . Numerical Analysis 12, 509.
HAMMACK, J., LEONE,C. & SEGUR,H. 1981 Long internal waves. (To be published.)
HAMMACK, J. L. & SEGUR,H. 1974 The Korteweg-de Vries equation and water waves. Part 2.
Comparison and experiments. J . Fluid Mech. 65, 289.
JOSEPH,R. I. 1977 Solitary waves in a finite depth fluid. J . Phys. A, Math. General 10, L225.
KAO,T. & PAO,H. P. 1979 Wake collapse in the thermocline and internal solitary waves.
J . Fluid Mech. 97, 115.
KAKUTANI, T. & MATSUUCHI, K. 1975 Effect of viscosity on long gravity waves. J . Phys. SOC.
Japan 39, 237.
KEULEGAN, G. H. 1948 Gradual damping of solitary waves. N.B.S. J . 40, 480.
KUBOTA, T., KO, D. R. S. & DOBBS,L. D. 1978 Propagation of weakly nonlinear internal
waves in a stratified fluid of finite depth. A.I.A.A. J . Hydronautics 12, 157.
LAKE,B. M., YUEN,H . C., RUNGALDIER, H. & FERGUSON, W. 1977 Nonlinear deep-water
waves: theory and experiment. Part 2. Evolution of a continuous wave train. J . Fluid
Mech. 83, 49.
LAMB,H. 1932 Hydrodynamics, 6th edn. Cambridge University Press.
MATSUUCHI,K. 1976 Numerical investigations on long gravity waves under the influence of
viscosity. J . Phys. SOC.Japan 41, 681.
MEISS, J. D. & PEREIRA, N. R. 1978 Internal wave solitons. Phys. Fluids 21, 700.
ONO, H. 1975 Algebraic solitary waves in stratified fluids. J . Phys. SOC.Japan 39, 1082.
OSBORNE,A., BURCH, T. & SCARLET, R. 1978 The influence of internal waves on deep water
drilling. J . Petroleum Tech. 1497.
OSBORNE,A. & BURCH,T. 1980 Internal solitons in the Andaman Sea. Science (submitted).
SEGUR, H. 1973 The Korteweg-de Vries equation and water waves. Solutions of the equations.
Part 1. J . Fluid Mech. 59, 721.
WALKER, L. R , 1973 Interfacial solitary waves in a two fluid medium. Phys. Fluids 16, 1796.
YATES,C. 1978 An experimental study of internal solitary waves. A I A A 16th Aerospace
Sciences Meeting, Huntsville, Alabama, Paper no. 78-262.

9 F L M I12