Gravity Capillary Free Surface Flows

CUUK1062-Vanden 9780521811903 May 28, 2010 12:43
This page intentionally left blank

CUUK1062-Vanden 9780521811903 May 28, 2010 12:43
GRAVI TYCAPI LLARY FREE- SURFACE FLOWS
Free-surface problems occur in many aspects of science and of everyday life, for
example in the waves on a beach, bubbles rising in a glass of champagne, melting
ice, pouring fl ws from a container and sails billowing in the wind. Consequently,
the theory of gravitycapillary free-surface fl ws continues to be a fertile fiel of
research in applied mathematics and engineering.
Concentrating on applications arising from flui dynamics, Vanden-Broeck
draws upon his years of experience in the fiel to address the many challenges
involved in attempting to describe such fl ws mathematically. Whilst careful
numerical techniques are implemented to solve the basic equations, an empha-
sis is placed upon the reader developing a deep understanding of the structure of
the resulting solutions. The author also reviews relevant concepts in flui mechan-
ics to enable readers from other scientifi field to develop a working knowledge
of free-boundary problems.
i
CUUK1062-Vanden 9780521811903 May 28, 2010 12:43
OT HE R T I T L E S I N T HI S S E RI E S
All the titles below can be obtained from good booksellers or from
Cambridge University Press. For a complete series listing visit
http://www.cambridge.org/uk/series/sSeries.asp?code=CMMA
Waves and Mean Flows
OLI VER B

UHLER
Lagrangian Fluid Dynamics
ANDREW BENNETT
Plasticity
SI A NEMAT- NASSER
Reciprocity in Elastodynamics
JAN D. ACHENBACH
Theory and Computation in Hydrodynamic Stability
W. O. CRI MI NALE, T. L. JACKSON AND R. D. JOSLI N
The Physics and Mathematics of Adiabatic Shear Bands
T. W. WRI GHT
Theory of Solidificatio
STEPHEN H. DAVI S
The Dynamics of Fluidized Particles
ROY JACKSON
Turbulent Combustion
NORBERT PETERS
Acoustics of FluidStructure Interactions
M. S. HOWE
Turbulence, Coherent Structures, Dynamical Systems and Symmetry
PHI LI P HOLMES, JOHN L. LUMLEY AND GAL BERKOOZ
Topographic Effects in Stratifie Flows
PETER G. BAI NES
Ocean Acoustic Tomography
WALTER MUNK, PETER WORCESTER AND CARL WUNSCH
B enard Cells and Taylor Vortices
E. L. KOSCHMI EDER
ii
CUUK1062-Vanden 9780521811903 May 28, 2010 12:43
GRAVITYCAPILLARY
FREE-SURFACE
FLOWS
JEAN-MARC VANDEN-BROECK
University College London
iii
CUUK1062-Vanden 9780521811903 May 28, 2010 12:43
iv
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore,
So Paulo, Delhi, Dubai, Tokyo
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
First published in print format
ISBN-13 978-0-521-81190-3
ISBN-13 978-0-511-72961-4
Cambridge University Press 2010
2010
Information on this title: www.cambridge.org/9780521811903
This publication is in copyright. Subject to statutory exception and to the
provision of relevant collective licensing agreements, no reproduction of any part
may take place without the written permission of Cambridge University Press.
Cambridge University Press has no responsibility for the persistence or accuracy
of urls for external or third-party internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
eBook (NetLibrary)
Hardback
CUUK1062-Vanden 9780521811903 May 28, 2010 12:43
To Mirna and Ada
v
CUUK1062-Vanden 9780521811903 May 28, 2010 12:43
vi
Contents
Preface page xi
1 Introduction 1
2 Basic concepts 7
2.1 The equations of uid mechanics 7
2.2 Free-surface ows 8
2.3 Two-dimensional ows 11
2.4 Linear waves 15
2.4.1 The water-wave equations 15
2.4.2 Linear solutions for water waves 17
2.4.3 Superposition of linear waves 24
3 Free-surface ows that intersect walls 31
3.1 Free streamline solutions 33
3.1.1 Forced separation 33
3.1.2 Free separation 43
3.2 The eects of surface tension 58
3.2.1 Forced separation 58
3.2.2 Free separation 66
3.3 The eects of gravity 73
3.3.1 Solutions with
1
= 0 (funnels) 83
1
= 0 (nozzles and bubbles) 88
1
= /2 (ow under a gate with
gravity) 94
3.4 The combined eects of gravity and surface tension 98
3.4.1 Rising bubbles in a tube 99
3.4.2 Fingering in a Hele Shaw cell 103
3.4.3 Further examples involving rising bubbles 108
3.4.4 Exponential asymptotics 112
vii
viii Contents
4 Linear free-surface ows generated by moving
disturbances 114
4.1 The exact nonlinear equations 115
4.2 Linear theory 116
4.2.1 Solutions in water of nite depth 116
4.2.2 Solutions in water of innite depth 122
4.2.3 Discussion of the solutions 124
5 Nonlinear waves asymptotic solutions 129
5.1 Periodic waves 129
5.1.1 Solutions when condition (5.55) is satised 135
5.1.2 Solutions when condition (5.55) is not satised 138
5.2 The Kortewegde Vries equation 142
6 Numerical computations of nonlinear water waves 148
6.1 Formulation 148
6.2 Series truncation method 151
6.3 Boundary integral equation method 152
6.4 Numerical methods for solitary waves 156
6.4.1 Boundary integral equation methods 157
6.5 Numerical results for periodic waves 160
6.5.1 Pure capillary waves (g = 0, T ,= 0) 160
6.5.2 Pure gravity waves (g ,= 0, T = 0) 164
6.5.3 Gravitycapillary waves (g ,= 0, T ,= 0) 175
6.6 Numerical results for solitary waves 181
6.6.1 Pure gravity solitary waves 181
6.6.2 Gravitycapillary solitary waves 186
7 Nonlinear free-surface ows generated by moving
disturbances 191
7.1 Pure gravity free-surface ows in water of nite depth 192
7.1.1 Supercritical ows 192
7.1.2 Subcritical ows 195
7.2 Gravitycapillary free-surface ows 201
7.2.1 Results in nite depth 201
7.2.2 Results in innite depth (removal of the
nonuniformity) 203
7.3 Gravitycapillary free-surface ows with Wilton ripples 206
8 Free-surface ows with waves and intersections with
rigid walls 210
8.1 Free-surface ow past a at plate 211
8.1.1 Numerical results 212
Contents ix
8.1.2 Analytical results 215
8.2 Free-surface ow past a surface-piercing object 218
8.3 Flow under a sluice gate 226
8.3.1 Formulation 228
8.3.2 Numerical procedure 231
8.3.3 Discussion of the results 233
8.4 Pure capillary free-surface ows 236
9 Waves with constant vorticity 244
9.1 Solitary waves with constant vorticity 245
9.1.1 Mathematical formulation 245
9.1.3 Discussion of the results 249
9.2 Periodic waves with constant vorticity 267
9.2.1 Mathematical formulation 268
9.2.4 Discussion 276
10 Three-dimensional free-surface ows 278
10.1 Greens function formulation for two-dimensional
problems 278
10.1.1 Pressure distribution 278
10.1.2 Two-dimensional surface-piercing object 283
10.2 Extension to three-dimensional free-surface ows 286
10.2.1 Gravity ows generated by moving disturbances
in water of innite depth 286
10.2.2 Three-dimensional gravitycapillary free-surface
ows in water of innite depth 293
10.3 Further extensions 298
11 Time-dependent free-surface ows 301
11.1 Introduction 301
11.2 Nonlinear gravitycapillary standing waves 301
References 308
Index 318
Preface
This book is concerned with the theory of gravitycapillary free-surface
ows. Free-surface ows are ows bounded by surfaces that have to be found
as part of the solution. A canonical example is that of waves propagating
on a water surface, the latter in this case being the free surface.
Many other examples of free-surface ows are considered in the book (cav-
itating ows, free-surface ows generated by moving disturbances, rising
bubbles etc.). I hope to convince the reader of the beauty of such problems
and to elucidate some mathematical challenges faced when solving them.
Both analytical and numerical methods are presented. Owing to space lim-
itations, some topics could not be covered. These include interfacial ows
and the eects of viscosity, compressibility and surfactants. Some further
developments of the theories described in the book can be found in the list
of references.
Many results presented in the book have grown out of my research over the
last 35 years and, of course, out of the research of the whole uid mechanics
community. References to the original papers are given. For this book, I
have repeated the older numerical calculations with larger numbers of grid
points than was possible at the time. I am pleased to report that the new
results are in agreement with the earlier ones!
I am deeply indebted to my mentors, coworkers, students and friends who
participated in the research. I feel very fortunate to have known them and I
look forward to continuing these collaborations in the future. Special thanks
are due to Scott Grandison for help with the gures, to David Tranah for his
patience over the years in waiting for the manuscript, to Susan Parkinson for
very careful copy-editing, to Caroline Brown for her help during production
and to Cambridge University Press for publishing the book.
xi
1
Introduction
Free-surface problems occur in many aspects of science and everyday life.
They can be dened as problems whose mathematical formulation involves
surfaces that have to be found as part of the solution. Such surfaces are
called free surfaces. Examples of free-surface problems are waves on a beach,
bubbles rising in a glass of champagne, melting ice, ows pouring from a
container and sails blowing in the wind. In these examples the free surface
is the surface of the sea, the interface between the gas and the champagne,
the surface of the ice, the boundary of the pouring ow and the surface of
the sail.
In this book we concentrate on applications arising in uid mechanics. We
hope to convince the reader of the beauty of such problems and to present
the challenges faced when one attempts to describe these ows mathemati-
cally. Many of these challenges are resolved in the book but others are still
open questions. We will always attempt to present fully nonlinear solutions
without restricting assumptions on the smallness of some parameters. Our
techniques are often numerical. However, it is the belief of the author that
a deep understanding of the structure of the solutions cannot be gained
by brute-force numerical approaches. It is crucial to combine numerical
methods with analytical techniques, especially when singularities are present.
Therefore analytical treatments will be presented whenever appropriate. We
hope that the techniques discussed will be useful not only to researchers in
the eld but also to those working in areas other than uid mechanics.
For completeness the relevant concepts of uid mechanics are reviewed in
Chapter 2.
Free-surface ows fall into two main classes. The rst is the class of such
ows for which there are intersections between the free surface and a rigid
surface. The classic example in this class is the ow due to a ship moving at
the surface of a lake, which involves an intersection between the free surface
1
2 Introduction
(i.e. the surface of the lake) and a rigid surface (i.e. the hull of the ship).
Other examples are jets leaving a nozzle, cavitating ows past an obstacle,
bubbles attached to a wall and ows under a sluice gate. In each case there is
a rigid surface (the nozzle, the obstacle, the wall or the gate) that intersects
a free surface.
The second class contains free-surface ows for which there are no inter-
sections between the free surface and a rigid wall. Here the classic example
is the ow due to a submerged object moving below the surface of a lake. If
the object is small compared with the size of the lake, it is then reasonable
to regard the lake as being of innite horizontal extent; then there is no
intersection between the free surface and a rigid surface. Other examples
include free bubbles rising in a uid and solitary waves.
Chapter 3 is concerned with the theory of free-surface ows of the rst
class. We use the classical assumptions of potential ow theory (irrota-
tional ows of inviscid and incompressible uids) and proceed in stages of
increasing complexity. In addition we restrict our attention to steady and
two-dimensional ows (time-dependent and three-dimensional ows are con-
sidered in the last two chapters). In the rst stage the eects of gravity
and surface tension are neglected. Such free-surface ows are called free
streamline ows. They are characterized by a constant velocity along the
free surfaces. Conformal mapping techniques can then be used to nd ex-
act nonlinear solutions. This situation is fortunate since there are very few
such solutions for free-surface ows. The main results of the free streamline
theory are summarized in Section 3.1. The most important result for the
remaining part of Chapter 3 is that the velocity and slope of the free surface
must be continuous at a separation point (i.e. the intersection between a
free surface and a rigid surface in two dimensions) but the curvature of the
free surface is in general innite. Since this curvature only enters the equa-
tions when surface tension is included in the dynamic boundary condition,
we expect a gravity ow with small gravity to be a regular perturbation of
a free streamline ows, and a capillary ow with small surface tension to
be a singular perturbation. This is conrmed by the numerical results pre-
sented in Sections 3.2 and 3.3. It is shown in Section 3.2 that the presence
of surface tension does not remove the innite curvature at the separation
points. On the contrary it makes the problem more singular by introducing
a discontinuity in slope at the separation point. Depending on the angle
between the free surface and the rigid boundary, the velocity is innite or
equal to zero at the separation point. The appearance of an innite ve-
locity is a limitation of the model. We show that a basic way to remove
this singularity is to take into account the nite thickness of the rigid walls,
Introduction 3
i.e. to consider the walls as thin objects with a continuous slope. When
surface tension is neglected, the free surfaces leave the wall tangentially but
the position of the separation point along the walls is free. We then have
a one-parameter family of solutions (the parameter denes the position of
the separation point) and the question is to determine which value of the
parameter is physically relevant. This is an example of a selection problem.
Selection problems are usually resolved by imposing an extra constraint on
the problem or by including a previously neglected eect and taking the
limit as this eect approaches zero. Here both approaches work. A unique
position for the separation point is obtained by neglecting surface tension
and imposing a constraint known as the BrillouinVillat condition. Equiva-
lently, the same position for the separation point is obtained by solving the
problem with surface tension and then taking the limit as the surface tension
approaches zero. We will show that the mechanism by which solutions with
a small amount of surface tension are selected is related to the fact that, for
each value of the surface tension, the separation point has only one position
for which the free surface leaves the wall tangentially.
In Section 3.3 we turn our attention to the eects of gravity on free stream-
line solutions. We assume that gravity is acting vertically downwards and
neglect surface tension (the combined eects of gravity and surface tension
are covered in Section 3.4). We show that it is again possible for the free
surface not to leave the walls tangentially, but the angle between the free
surface and the wall must be such that the velocity is nite at the separa-
tion point (innite velocities cannot occur on a free surface in the absence
of surface tension). Local analysis shows that there are only three possible
behaviours at the separation point. In the rst there is a horizontal free
surface at the separation point, in the second there is an angle of 120
be-
tween the free surface and the wall and in the third the free surface leaves
the wall tangentially. We show by examples (e.g. ows emerging or pour-
ing from containers) that these three possibilities occur in free-surface ows
with gravity. The restriction to three local behaviours is to be contrasted
with the cases including surface tension discussed in Section 3.2, where all
angles between the walls and the free surfaces are in principle possible. This
contrast suggests that some interesting behaviours might emerge if we com-
bine the eects of gravity and surface tension; this is conrmed in Section
3.4. We show in this section that some free-surface ow problems possess
a continuum of solutions when surface tension is neglected and an innite
discrete set of solutions when surface tension is taken into account. This
discrete set reduces to a unique solution as the surface tension approaches
zero. Therefore a small amount of surface tension can again be used to select
4 Introduction
solutions. One dierence between this selection mechanism and that de-
scribed in Section 3.2 is that there is a innite discrete set of solutions when
surface tension is included instead of one solution. Another dierence is
that the selection is associated with exponentially small terms in the surface
tension. This implies that exponential asymptotics is required to predict
the selected solutions analytically. As we shall see, exponential asymptotics
plays an important role in many other free-surface ow problems such as
the study of gravitycapillary waves (see Chapter 6) and free-surface ows
generated by moving disturbances for small values of the Froude number or
small values of the surface tension (see Chapter 8).
The results presented in Chapter 3 were obtained by a combination of var-
ious numerical schemes that the author has used successfully over the years
to obtain highly accurate solutions for free-surface ow problems. They in-
clude series truncation techniques and boundary integral equation methods.
The idea of the series truncation methods is to identify a rapidly convergent
series representation for the solution that satises all the appropriate partial
dierential equations (for example the Laplace equation for potential ows)
and all the linear boundary conditions. This often requires a local analysis
to identify and remove the singularities associated with corners, stagnation
points etc. The series is then truncated after a nite number of terms and
the unknown coecients are determined by satisfying the remaining non-
linear boundary condition (the pressure condition for free-surface ows) at
appropriately chosen collocation points. This leads to a system of nonlinear
algebraic equations which can be solved by iteration (for example by using
Newtons method). Boundary integral equation methods are based on a
reformulation of the problem as a system of nonlinear integro-dierential
equations for the unknown quantities on the free surface. These equations
are then discretised and the resultant nonlinear algebraic equations solved
by iteration. Such boundary integral equation methods have been used ex-
tensively by many researchers.
Insight into free-surface ows of the second class can be gained by study-
ing the limitations of the classical linear theories. In particular we study in
Chapter 4 the waves generated by a disturbance moving at a constant veloc-
ity (for example a submerged object or a pressure distribution). The results
are qualitatively independent of the type of disturbance, and so most results
in Chapter 4 are presented just for a pressure distribution with bounded sup-
port. A frame of reference moving with the pressure distribution is chosen
and the ow is assumed to be steady. In the linear theory it is assumed
that the disturbance is small enough for the ow to be a small perturbation
of a uniform stream. The equations are then linearised (around a uniform
Introduction 5
stream) and the resulting linear equations solved by separation of variables
and using Fourier transforms. These linear solutions can be expected to
be a good approximation when the disturbance is small. In other words, if
denotes the size of the disturbance, we expect the nonlinear solutions to
approach the linear solutions as 0. This is usually the case, but the
problem is complicated by the fact that the solutions depend not only on
but also on other parameters such as the Froude number
F =
U
(gH)
1/2
and the capillary number
=
Tg
U
4
.
Here U is the velocity of the disturbance, H the depth of the uid, T the
surface tension, g the acceleration of gravity and the density of the uid.
This leads to nonuniformities when these parameters approach critical
values. These nonuniformities appear in an obvious way in the linear solu-
tions. For example the linear theory for pure gravity ows (i.e. ows with
T = 0) predicts innite displacements of the free surface as F 1. This
is unacceptable since the linear theory assumes small perturbations around
a uniform stream and in particular small displacements of the free surface.
More precisely, for any F ,= 1 the linear theory provides a good approxima-
tion of the nonlinear problem as 0. However, for any , no matter how
small, the linear solutions become invalid as F 1. A similar situation
occurs for gravitycapillary ows. As approaches a critical value
H
, the
linear theory again predicts innite displacements of the free surface. The
critical value
H
depends on the depth H (for example
H
= 0.25 in water
of innite depth).
The resolution of these nonuniformities requires a nonlinear theory. We
develop in Chapter 7 such a theory by solving the fully nonlinear equa-
tions numerically. This approach has the advantage of not pre-assuming a
particular type of expansion. Furthermore it gives solutions without any
assumption on the size of . It also provides a valuable guide in deriving
appropriate perturbation expansions for small or moderate values of . We
show in Chapter 7 that the resolution of the nonuniformities is associated
with solitary waves. Near the critical values of and F, there are not only
solutions that are perturbations of a uniform stream (the nonlinear equiv-
alent of the linear solutions mentioned earlier) but also solutions that are
perturbations of solitary waves. Some of these solitary waves are of the
6 Introduction
well-known Kortewegde Vries type but others are solitary waves with de-
caying oscillatory tails.
As a preparation for the nonlinear results of Chapter 7 we present in
Chapters 5 and 6 analytical and numerical solutions for nonlinear periodic
and solitary waves. Such solutions describe the far-eld behaviour of the
nonlinear free-surface ows past disturbances described in Chapter 7. They
are also interesting canonical free-surface ow problems. In particular we
show that waves of the Kortewegde Vries type have oscillatory tails of con-
stant amplitude when surface tension is included. These waves are referred
to as generalised solitary waves to distinguish them from true solitary waves,
which are at in the far eld.
In Chapter 8 we consider further free-surface ows of the rst class (i.e.
ows for which the free surface intersects rigid walls). The solutions of
Chapter 3 approach either an innitely thin jet in the far eld or are wave-
less. We study in Chapter 8 various extensions for which the free surface is
characterised by a train of nonlinear waves in the far eld. An attractive
feature of some of these ows is that exact formulae can be derived for the
amplitude of the waves in the far eld. These relations provide analytical
insight and can be used to check the accuracy of the numerical codes.
All the ows in Chapters 38 are assumed to be steady, two-dimensional
and irrotational. The nal three chapters of the book describe some ex-
tensions in which these assumptions are removed. In Chapter 9 we study
solitary and periodic waves with constant vorticity. We show that there are
new solution branches that do not have an equivalent for irrotational waves.
In Chapter 10 we study some three-dimensional free-surface ows. In partic-
ular we calculate three-dimensional gravitycapillary solitary waves. These
waves are characterised by decaying oscillations in the direction of propaga-
tion and monotonic decay in the direction perpendicular to the direction of
propagation.
Chapter 11 is concerned with time-dependent free-surface ows. This is a
very large subject involving problems such as breaking waves, stability, the
breaking of jets etc. Here we limit our attention to the subject of gravity
capillary standing waves. This choice is motivated by the fact that these
standing waves have properties similar to those of the travelling gravity
capillary waves presented in Chapter 6.
2
Basic concepts
2.1 The equations of uid mechanics
We start with a brief introduction to the equations of uid mechanics. For
further details see for example Batchelor [8] or Acheson [1].
All the uids considered in this book are assumed to be inviscid and to
have constant density (i.e. to be incompressible).
Conservation of momentum yields the Euler equations
Du
Dt
=
1
p +X, (2.1)
where u is the vector velocity, p is the pressure and X is the body force.
Here
D
Dt
=

t
+u (2.2)
is the material derivative. We assume that the body force X derives from a
potential , i.e. that
X = . (2.3)
In most applications considered in this book, the ow is assumed to be
irrotational. Therefore
u = 0. (2.4)
Relation (2.4) implies that we can introduce a potential function such that
u = . (2.5)
Conservation of mass gives
u = 0. (2.6)
7
8 Basic concepts
Then (2.5) and (2.6) imply that satises Laplaces equation
2
= 0. (2.7)
Flows that satisfy (2.4)(2.7) are referred to as potential ows. Using the
identity
u u =
1
2
(u u) + (u) u, (2.8)
(2.4) and (2.2) yield
Du
Dt
=
u
t
+
1
2
(u u). (2.9)
Substituting (2.9) into (2.1) and using (2.3) and (2.5) we obtain
t
+
u u
2
+
p
+
_
= 0. (2.10)
After integration, (2.10) gives the well-known Bernoulli equation
t
+
u u
2
+
p
+ = F(t). (2.11)
Here F(t) is an arbitrary function of t. It can be absorbed in the denition
of , and then (2.11) can be rewritten as
t
+
u u
2
+
p
+ = B, (2.12)
where B is a constant. For steady ows (2.12) reduces to
u u
2
+
p
+ = B. (2.13)
2.2 Free-surface ows
We introduce the concept of a free surface by contrasting the ow past a
rigid sphere (see Figure 2.1) with that of the ow past a bubble (see Figure
2.2). Both ows are assumed to be steady and to approach a uniform stream
with a constant velocity U as x
2
+ y
2
+ z
2
; the eects of gravity are
neglected. They can interpreted as the ows due to a rigid sphere or a
bubble rising at a constant velocity U, when viewed in a frame of reference
moving with the sphere or the bubble. The pressure p
b
in the bubble is
constant. We denote by S the surface of the sphere or bubble and by n the
outward unit normal.
The ow past a sphere can be formulated as follows:
xx
+
yy
+
zz
= 0 outside S, (2.14)
2.2 Free-surface ows 9
z
y
x
S
U
Fig. 2.1. The ow past a rigid sphere. The surface S of the sphere is described by
x
2
+y
2
+z
2
= R
2
, where R is the radius of the sphere.
z
y
x
S
U
Fig. 2.2. The ow past a bubble. The surface S of the bubble is not known a priori
and has to be found as part of the solution.
n
= 0 on S (2.15)
(
x
,
y
,
z
) (0, 0, U) as x
2
+y
2
+z
2
. (2.16)
Equation (2.14) is Laplaces equation (2.7) expressed in cartesian coordi-
nates. The boundary condition (2.15) is known as the kinematic boundary
condition. It states that the normal component of the velocity vanishes
on S.
10 Basic concepts
Equations (2.14)(2.16) form a linear boundary value problem whose
solution is
= U
_
z +
R
3
z
2(x
2
+y
2
+z
2
)
3/2
_
. (2.17)
Here R is the radius of the sphere.
We note that we have derived the solution (2.17) without using the Bernoulli
equation (2.13), which for the present problem can be written as
1
2
(
2
x
+
2
y
+
2
z
) +
p
=
1
2
U
2
+
p
. (2.18)
Here p
denotes the pressure as x

2
+ y
2
+ z
2
. Equation (2.18) holds
everywhere outside the sphere. In deriving (2.18) we have set = 0 in
(2.13) and evaluated B by taking the limit x
2
+ y
2
+ z
2
in (2.13).
Then, using (2.16) gives B = U
2
/2 +p
/.
Equation (2.18) is nonlinear but it is only used if we want to calculate the
pressure p inside the uid. In other words the main problem is to nd by
solving the linear set of relations (2.14)(2.16). We may then substitute the
values (2.17) of into the nonlinear equation (2.18) if we wish to compute
the pressure.
We now show that we need to use the nonlinear boundary condition (2.18)
to solve for the potential for a ow past the bubble of Figure 2.2. This
implies that, because of its nonlinearity, the ow past a bubble is a much
harder problem to solve than the ow past a sphere. The potential function
still satises (2.14)(2.16). However, the main dierence is that the shape
of the surface S of the bubble is not known and has to be found as part of the
solution. In other words the equation of the surface S is no longer given as it
was for the ow past a sphere. Therefore we need an extra equation to nd
S. This equation uses (2.18) and can be derived as follows. First we relate
the pressure p on the uid side of S to the pressure p
b
inside the bubble
by using the concept of surface tension. If we draw a line on a uid surface
(such as S), the uid on the right of the line is found to exert a tension T,
per unit length of the line, on the uid to the left. We call T the surface
tension coecient. It depends on the uid and also on the temperature. It
can be shown (see for example Batchelor [8]) that
p p
b
= TK = T
_
1
R
1
+
1
R
2
_
. (2.19)
Here R
1
and R
2
are the principal radii of curvature of the uid surface: they
are counted positive when the centres of curvature lie inside the uid. The
quantity
K =
_
1
R
1
+
1
R
2
_
(2.20)
is referred to as the mean curvature of the uid surface. In most applications
presented in this book the surface tension T is assumed to be constant.
We now apply the Bernoulli equation (2.18) to the uid side of the surface
S and use (2.19). This gives
1
2
(
2
x
+
2
y
+
2
z
) +
T
K =
1
2
U
2
+
p
p
b
on S. (2.21)
Equation (2.21) is known as the dynamic boundary condition. This is the
extra equation needed to nd S. To solve the bubble problem we seek the
function and the equation of the surface S such that (2.14)(2.16) and
(2.21) are satised. It is a nonlinear problem that requires the solution of a
partial dierential equation (here the Laplace equation (2.14)) in a domain
whose boundary (here S) has to be found as part of the solution. This
is a typical free-surface ow problem. In this book we will describe various
analytical and numerical methods for investigating such nonlinear problems.
We note that the problem of Figure 2.2 is an idealised one, in which
the viscosity and gravity and the wake behind the bubble are neglected.
Bubbles with wakes and the eect of including gravity will be considered in
Section 3.4.3. Readers interested in the eects of viscosity are referred to,
for example, [117].
The dynamic boundary condition (2.21) is valid for steady ows with
= 0. Combining (2.12) and (2.19) we nd that the general form of the
dynamic boundary condition (for unsteady ows) with ,= 0 is
t
+
u u
2
+ +
T
K = B. (2.22)
Here B is the Bernoulli constant. For steady ows, (2.22) reduces to
u u
2
+ +
T
K = B. (2.23)
2.3 Two-dimensional ows
As we shall see, many interesting free-surface ows can be modelled as two-
dimensional ows. We then introduce cartesian coordinates x and y with
the y-axis directed vertically upwards (at present we reserve the letter z to
denote the complex quantity x + iy). In most applications considered in
12 Basic concepts
this book, the potential (see (2.3)) is due to gravity. Assuming that the
acceleration of gravity g is acting in the negative y-direction, we write as
= gy. (2.24)
An example is the two-dimensional free-surface ow past a semicircular
obstacle at the bottom of a channel (see Figure 2.3). This two-dimensional
conguration provides a good approximation to the three-dimensional free-
surface ow past a long half-cylinder perpendicular to the plane of the gure
(except near the ends of the cylinder). The cross section of the cylinder is
the semicircle shown in Figure 2.3.
x
y
Fig. 2.3. Two-dimensional free-surface ow past a submerged semicircle.
For two-dimensional potential ows, (2.4) and (2.6) become
u
y

v
x
= 0, (2.25)
u
x
+
v
y
= 0. (2.26)
Here u and v are the x- and y- components of the velocity vector u.
We can introduce a streamfunction by noting that (2.26) is satised by
u =

y
, (2.27)
v =
x
. (2.28)
It then follows from (2.25) that
2
=

2
x
2
+

2
y
2
= 0. (2.29)
For two-dimensional ows, equations (2.5) and (2.7) give
u =

x
, (2.30)
v =

y
(2.31)
and
2
=

2
x
2
+

2
y
2
= 0. (2.32)
Combining (2.27), (2.28), (2.30) and (2.31) we obtain
x
=

y
, (2.33)
y
=
x
. (2.34)
Equations (2.33) and (2.34) can be recognised as the classical Cauchy
Riemann equations. They imply that the complex potential
f = +i (2.35)
is an analytic function of z = x + iy in the ow domain. This result is
particularly important since it implies that two-dimensional potential ows
can be investigated by using the theory of analytic functions. This applies in
particular to all two-dimensional potential free-surface ows with or without
gravity and/or surface tension included in the dynamic boundary condition.
It does not apply, however, to axisymmetric and three-dimensional free-
surface ows. Since the derivative of an analytic function is also an analytic
function, it follows that the complex velocity
u iv =

x
i
y
=

y
+i
x
=
df
dz
(2.36)
is also an analytic function of z = x +iy.
The theory of analytic functions will be used intensively in the follow-
ing chapters to study two-dimensional free-surface ows. In particular the
following important tools will be useful.
The rst tool is conformal mappings. These are changes of variable dened
by analytic functions. For example, if h(t) is an analytic function of t, the
change of variables z = h(t) enables us to seek the complex velocity uiv as
an analytic function of t (since an analytic function of an analytic function
is also an analytic function). Such conformal mappings are used to redene
14 Basic concepts
a problem in a new complex t-plane in which the geometry is simpler than
in the original z-plane.
The second tool is Cauchys theorem: If h(z) is analytic throughout a
simply connected domain D then, for every closed contour C within D,
_
C
h(z)dz = 0. (2.37)
The third tool is the Cauchy integral formula: Let h(z) be analytic every-
where within and on a closed contour C, taken in the positive sense (coun-
terclockwise). Then the integral
1
2i
_
C
h(z)
z z
0
dz (2.38)
takes the following values:
0 if z
0
is outside C, (2.39)
h(z
0
) if z
0
is inside C, (2.40)
1
2
h(z
0
) if z
0
is on C. (2.41)
When z
0
is on C the integral (2.38) is a Cauchy principal value.
We now show that for steady ows the streamfunction is constant along
streamlines. A streamline is a line to which the velocity vectors are tangent.
Let us describe a streamline in parametric form by x = X(s), y = Y (s),
where s is the arc length. Then we have
vX
(s) +uY
(s) = 0, (2.42)
where the primes denote derivatives with respect to s. Using (2.27) and
(2.28) we have
x
X
(s) +

y
Y
(s) =
d
ds
= 0, (2.43)
which implies that is constant along a streamline. For steady ows the
kinematic boundary condition implies that a free surface is a streamline.
The streamfunction is therefore constant along a free surface.
For two-dimensional ows the dynamic boundary condition (2.22) be-
comes
t
+
1
2
(
2
x
+
2
y
) +gy +
T
K = B. (2.44)
If we denote by the angle between the tangent to the free surface and the
2.4 Linear waves 15
horizontal then the curvature K can be dened by
K =
d
ds
(2.45)
where again s denotes the arc length. In particular if the (unknown) equation
of the free surface is y = (x, t) then
tan =
x
and
dx
ds
=
1
(1 +
2
x
)
1/2
. (2.46)
Using (2.45), (2.46) and the chain rule gives the formula
K =

xx
(1 +
2
x
)
3/2
. (2.47)
2.4 Linear waves
2.4.1 The water-wave equations
Many fre-surface ows involve waves on their free surfaces. When dissi-
pation is neglected and the ow is assumed to be two-dimensional, these
waves often approach uniform wave trains in the far eld (see for example
Figure 2.3). Therefore a fundamental problem in the theory of free-surface
ows is the study of a uniform train of two-dimensional waves of wavelength
extending from x = to x = and travelling at a constant velocity c.
The ow conguration is illustrated in Figure 2.4.
x
y
y = y
1
y = y
2

Fig. 2.4. A two-dimensional train of waves viewed in a frame of reference moving
with the wave. The free-surface prole has wavelength . The uid is bounded
below by a horizontal bottom with equation y = h. Also shown is the rectangular
contour used in (2.56).
16 Basic concepts
For convenience we have chosen a frame of reference moving with the wave,
so that the ow is steady. Using the notation of Section 2.3, we formulate
the problem as
xx
+
yy
= 0, h < y < (x), (2.48)
y
=
x
x
on y = (x), (2.49)
y
= 0 on y = h, (2.50)
1
2
(
2
x
+
2
y
) +gy
T
xx
(1 +
2
x
)
3/2
= B on y = (x), (2.51)
(x +, y) = (x, y), (2.52)
(x +) = (x), (2.53)
_

0
(x)dx = 0, (2.54)
1
_

0
x
dx = c on y = constant. (2.55)
Here g is the acceleration of gravity (assumed to act in the negative y-
direction), T is the surface tension, is the density, y = h is the equation
of the bottom and y = (x) is the equation of the (unknown) free surface.
Equations (2.49) and (2.50) are the kinematic boundary conditions on the
free surface and on the bottom respectively. Equation (2.51) is the dynamic
boundary condition on the free surface. We have used (2.44) and the formula
(2.47) for the curvature of a curve y = (x). Relations (2.52) and (2.53)
are periodicity conditions, which require the solution to be periodic with
wavelength . Equation (2.54) xes the origin of the y-coordinates as the
mean water level. Finally, (2.55) denes the velocity c as the average value
of u =
x
at a level y = constant in the uid. The value of c is independent
of the constant chosen; this can be seen by applying Stokes theorem to
the vector velocity (u, v) using a contour C consisting of two horizontal
lines y = y
1
, y = y
2
and two vertical lines separated by a wavelength (see
Figure 2.4). Since the ow is irrotational,
_
C
udx +vdy = 0. (2.56)
2.4 Linear waves 17
The contributions from the two vertical lines cancel by periodicity and (2.56)
gives
_

0
[u]
y=y
1
dx =
_

0
[u]
y=y
2
dx. (2.57)
Since y
1
and y
2
are arbitrary, the integral on the left-hand side of (2.55) is
independent of the level y = constant chosen in the uid.
The relations (2.48)(2.55) are referred to as the water-wave equations
because they model waves travelling at the interface between water and air
(although they apply also to other uids).
2.4.2 Linear solutions for water waves
A trivial solution of the system (2.48)(2.55) is
= cx, (x) = 0 and B =
c
2
2
. (2.58)
This solution describes a uniform stream with constant velocity c, bounded
below by a horizontal bottom and above by a at free surface.
Linear waves are obtained by seeking a solution as a small perturbation
of the exact solution (2.58). Therefore we write
(x, y) = cx +(x, y) (2.59)
and assume that both [(x, y)[ and [(x)[ are small. Substituting (2.59) into
(2.48)(2.55) and dropping nonlinear terms in and , we obtain the linear
system
xx
+
yy
= 0, h < y < 0, (2.60)
y
= c
x
, y = 0, (2.61)
y
= 0, y = h, (2.62)
xx
+c
x
+g = 0, y = 0, (2.63)
(x +, y) = (x, y), (2.64)
(x +) = (x), (2.65)
_

0
(x)dx = 0, (2.66)
18 Basic concepts
1
_

0
x
dx = 0 on y = constant. (2.67)
We choose the origin of x at a crest and assume that the wave is symmetric
about x = 0. Thus we impose the conditions
(x, y) = (x, y), (2.68)
(x) = (x). (2.69)
Using the method of separation of variables, we seek a solution of (2.60)
in the form
(x, y) = X(x)Y (y). (2.70)
Substituting (2.70) into (2.60), (2.68) and (2.62) yields
X(x) = X(x), (2.71)
the ordinary dierential equations
X
(x)
X(x)
=
Y
(y)
Y (y)
= constant =
2
, (2.72)
and the boundary condition
Y
(h) = 0. (2.73)
Here we have chosen a negative separation constant in (2.72), so that the
solution is periodic in x. Solutions of the two ordinary dierential equations
(2.72) satisfying (2.71) and (2.73) are written as
X(x) = sin x, (2.74)
Y (y) = cosh (y +h). (2.75)
The periodicity condition (2.64) implies that
= nk, (2.76)
where n is a positive integer and
k =
2
(2.77)
is the wavenumber. Multiplying (2.74) and (2.75) and taking a linear com-
bination of the solutions corresponding to the values (2.76) of , we obtain
(x, y) =
n=1
B
n
cosh nk(y +h) sin nkx. (2.78)
2.4 Linear waves 19
Here the B
n
are constants.
Using the periodicity and the symmetry conditions (2.69) and (2.65), we
express (x) as the Fourier series
(x) = A
0
+
n=1
A
n
cos nkx (2.79)
where the A
n
are constants. The condition (2.66) implies that A
0
= 0.
Substituting (2.78) and (2.79) into (2.61) and equating the coecients of
sin nkx yields
cA
n
= B
n
sinh nkh, n = 1, 2, . . . (2.80)
Similarly substituting (2.78) and (2.79) into (2.63) gives
T
A
n
n
2
k
2
+gA
n
+cB
n
nk cosh nkh = 0, n = 1, 2, . . . (2.81)
Eliminating B
n
between (2.80) and (2.81) yields
_
g +
T
n
2
k
2
c
2
nk
sinh nkh
cosh nkh
_
A
n
= 0, n = 1, 2, . . . (2.82)
Since we seek a nontrivial periodic solution (x) ,= 0, we can assume without
loss of generality that A
1
,= 0; then (2.82) with n = 1 implies that
c
2
=
_
g
k
+
T
k
_
tanh kh. (2.83)
Relation (2.82) for n > 1 gives
A
n
= 0, n = 2, 3, . . . , (2.84)
provided that
g +
T
n
2
k
2
c
2
nk
sinh nkh
cosh nkh ,= 0, n = 2, 3, . . . (2.85)
If (2.85) is satised, the solution of the linear problem is
=
cA
1
sinh kh
cosh k(y +h) sin kx, (2.86)
= A
1
cos kx. (2.87)
If the condition (2.85) is not satised for some integer value m of n, the
solution of the linear problem is
=
cA
1
sinh kh
cosh k(y +h) sin kx
cA
m
sinh mkh,
cosh mk(y +h) sin mkx,
(2.88)
20 Basic concepts
1
= A
1
cos kx +A
m
cos mkx, (2.89)
where A
m
is an arbitrary constant. In the theory of linear waves, it is
usually assumed that A
m
= 0. However, when we are developing nonlinear
theories for water waves in Chapters 5 and 6, i.e. improving the linear
approximations (2.88) and (2.89) by adding nonlinear corrections or solving
the fully nonlinear problem (2.48)(2.55) numerically, we shall see that A
m
,=
0. Two consequences are the existence of many dierent families of nonlinear
periodic gravitycapillary waves and the existence of solitary waves with
oscillatory tails.
The velocity c is called the phase velocity and equation (2.83) is the
(linear) dispersion relation. Relation (2.83) implies that waves of dier-
ent wavenumbers and therefore of dierent wavelengths travel at dierent
phase velocities c.
It is convenient to rewrite (2.83) in the dimensionless form
F
2
=
_
1
kh
+kh
_
tanh kh, (2.90)
where
F =
c
(gh)
1/2
(2.91)
is the Froude number and
=
T
gh
2
(2.92)
is the Bond number. Relation (2.90) is shown graphically in Figure 2.5,
where we present values of F
2
versus 1/(kh) = /(2h) for four values of .
The curves of Figure 2.5 illustrate that F
2
is a monotonically decreasing
function of /h when > 1/3 and that it has a minimum for < 1/3.
As /h , F 1. The dierent behaviours for < 1/3 (minimum)
and > 1/3 (monotone decay) in Figure 2.5 have many implications, in
particular for the study of nonlinear periodic and solitary gravitycapillary
waves (see Chapters 5 and 6).
We now examine two particular cases.
The rst is the case of water of innite depth. This is obtained by taking
the limit kh in (2.83), (2.86) and (2.87) and leads to
= cA
1
e
ky
sin kx, (2.93)
= A
1
cos kx, (2.94)
c
2
=
g
k
+
T
k. (2.95)
2.4 Linear waves 21
0
0.5
1.0
1.5
2.0
0 2 4 6 8 10 12 14
Fig. 2.5. Values of F
2
versus 1/(kh). The curves correspond from top to bottom
to = 1.3, = 1/3, = 0.1 and = 0.05. For < 1/3 the curves have a minimum
whereas for > 1/3 the curves are monotonically decreasing.
Since kh = 2h/, the innite-depth results (2.93)(2.95) provide an ap-
proximation to the nite-depth results (2.83), (2.86) and (2.87) when the
wavelength is small compared with the depth h.
Waves with g ,= 0, T = 0 are referred to as pure gravity waves. They are
characterised in the case of innite depth by the dispersion relation
c
2
=
g
k
. (2.96)
Similarly, waves with g = 0, T ,= 0 are called pure capillary waves and are
characterised in the innite-depth case by the dispersion relation
c
2
=
T
k. (2.97)
A simple calculation based on (2.95) shows that c
2
reaches a minimum value
given by
c
min
=
_
4Tg
_
1/4
(2.98)
when
k = k
min
=
_
g
T
_
1/2
. (2.99)
Graphs of c versus in units of c
min
and
min
= 2/k
min
are shown in Figure
2.6. The solid curve corresponds to (2.95), the dotted curve to (2.97) and
the broken curve to (2.96). These curves show that waves with >
min
are dominated by gravity and can be approximated by pure gravity waves
22 Basic concepts
0
0.5
1.0
1.5
2.0
0 1 2 3 4 5 6
Fig. 2.6. Values of c versus = 2/k in units of c
min
and
min
. The solid curve
corresponds to (2.95), the dotted curve to (2.97) and the broken curve to (2.96).
for large. Waves with <
min
are dominated by surface tension and can
be approximated by pure capillary waves for small.
The second particular case is that of pure gravity waves (i.e. T = 0) in
water of nite depth. Then (2.90) reduces to
khF
2
= tanh kh. (2.100)
Since
d
d (kh)
tanh kh 1, (2.101)
equation (2.100) has a solution kh > 0 when F < 1. For F > 1 the only
real solution of (2.100) is kh = 0. This implies that linear gravity waves
only exist when F < 1; for F > 1, linear gravity waves are not possible.
Flows characterised by F < 1 are called subcritical and those characterised
by F > 1 are called supercritical. The distinction between subcritical and
supercritical ows will appear often in this book.
So far we have discussed linear waves in a frame of reference moving
with the wave. This is a convenient choice because the ow is then steady.
However, it is also useful to look at waves from the point of view of a xed
frame of reference in which the wave moves to the left at a constant velocity
c. The nonlinear governing equations are then
xx
+
yy
= 0, h < y < (x, t), (2.102)
t
=
y

x
x
on y = (x, t), (2.103)
y
= 0 on y = h, (2.104)
2.4 Linear waves 23
t
+
1
2
(
2
x
+
2
y
) +gy
T
xx
(1 +
2
x
)
3/2
= B on y = (x, t), (2.105)
(x +, y, t) = (x, y, t), (2.106)
(x +, t) = (x, t), (2.107)
_

0
(x, t)dx = 0. (2.108)
A trivial solution of the system (2.102)(2.108) is
= 0, = 0 and B = 0. (2.109)
We can construct linear waves by assuming a small perturbation of the exact
solution (2.109) in the form of a wave travelling to the left at a constant
velocity c. Therefore we rewrite and in terms of two new functions

and :
(x, y, t) =

(x +ct, y) and (x, t) = (x +ct) (2.110)
Substituting (2.110) into the system (2.102)(2.108) and dropping nonlinear
terms in

and , we obtain the linear system
xx
+

yy
= 0, h < y < 0, (2.111)
c
x
=

y
on y = 0, (2.112)
y
= 0 on y = h, (2.113)
c
x
+g
T

xx
= 0 on y = 0, (2.114)
(x + +ct, y) =

(x +ct, y), (2.115)
(x + +ct) = (x +ct), (2.116)
_

0
(x +ct)dx = 0. (2.117)
Following the derivation of (2.86)(2.89), we nd that the solution of
(2.111)(2.117) is
=
cA
1
sinh kh
cosh k(y +h) sin k(x +ct), (2.118)
= A
1
cos k(x +ct) (2.119)
24 Basic concepts
if (2.85) is satised and
=
cA
1
sinh kh
cosh k(y +h) sin k(x +ct),
cA
m
sinh mkh
cosh mk(y +h) sin mk(x +ct), (2.120)
= A
1
cos k(x +ct) +A
m
cos mk(x +ct) (2.121)
if for n = m (2.85) is not satised. The dispersion relation is given as before
by (2.83).
2.4.3 Superposition of linear waves
Since the system (2.111)(2.117) is linear, new solutions can be obtained by
superposing solutions corresponding to dierent values of k and/or of A
1
.
We consider two particular superpositions for the solution (2.118), (2.119).
The rst corresponds to the superposition of two waves of the same am-
plitude travelling at the same velocity but in opposite directions. This gives
= A
1
cos k(x +ct) +A
1
cos k(x ct), (2.122)
=
cA
1
sinh kh
cosh k(y +h) sin k(x +ct)
+
cA
1
sinh kh
cosh k(y +h) sin k(x ct). (2.123)
Using the trigonometric identities
cos p + cos q = 2 cos
p +q
2
cos
p q
2
(2.124)
sin p + sin q = 2 sin
p +q
2
cos
p q
2
(2.125)
we can rewrite (2.122), (2.123) as
= 2A
1
cos kxcos kct, (2.126)
= 2
cA
1
sinh kh
cosh k(y +h) cos kxsin kct. (2.127)
The solution dened by (2.126), (2.127) is known as a linear standing wave
because the position of its nodal points and of the maximum displacement
of the free surface are xed as t varies. The wave does not propagate and
2.4 Linear waves 25
its free surface moves periodically up and down as t varies. The period of
this motion is
T
s
=
2
kc
. (2.128)
Since u =
x
= 0 along the lines x = 0 and x = /k = /2, we can replace
these two lines by walls (the kinematic boundary condition on them is then
automatically satised). The resulting ow models the periodic sloshing of
a liquid in a container (see Figure 2.7).
0
0.1
0.2
0 1 2 3
Fig. 2.7. Standing wave for A
1
= 0.1 and k = 1. The broken line is the free-surface
prole at t = 0 and the dotted line is the free-surface prole at t = T
s
/2. This ow
models liquid sloshing in a container bounded by vertical walls at x = 0 and x = .
An interesting question is whether there are similar nonlinear solutions.
This question is addressed in Chapter 11, where we construct analytical
approximations to such solutions.
The second example of superposition that we consider is that of two wave
trains of the same amplitude travelling in the same direction but with slightly
dierent wavenumbers k and

k. We rst introduce the angular frequency
= kc (2.129)
and write = W(k). Using (2.83) we have
W(k) = k
__
g
k
+
T
k
_
tanh kh
_
1/2
. (2.130)
Next we rewrite (2.118) and (2.119) as
= A
1
cos[kx +W(k)t], (2.131)
=
cA
1
sinh kh
cosh k(y +h) sin[kx +W(k)t]. (2.132)
26 Basic concepts
The superposition described above then yields
= A
1
cos[kx +W(k)t] +A
1
cos[
kx +W(
k)t]. (2.133)
Using the identity (2.124), we can rewrite (2.133) as
= 2A
1
cos
__
1
2
[(k +

k)]x +
1
2
[W(k) +W(
k))t
__
cos
_
1
2
[(k
k)]x +
1
2
[(W(k) W(
k)t]
_
. (2.134)
For

k close to k, we may approximate (2.134) by
= (x, t) cos[kx +W(k)t], (2.135)
where
(x, t) = 2A
1
cos
_
1
2
(k
k)x +
1
2
[W(k) W(
k)]t
_
. (2.136)
The expression (2.135) is the same as (2.131) except that the constant am-
plitude A
1
has been replaced by the variable amplitude (x, t).
Dierentiating (2.136) with respect to x and t yields
x
= A
1
(k
k) sin
_
1
2
(k
k)x +
1
2
[W(k) W(
k)]t
_
(2.137)
and
t
= A
1
[W(k) W(
k)] sin
_
1
2
(k
k)x +
1
2
[W(k) W(
k)]t
_
. (2.138)
The derivatives (2.137) and (2.138) are of order k
k and W(k) W(
k)
respectively. They are therefore small for

k close to k. This implies that the
amplitude (x, t) is a slowly varying function of x and t. In other words, the
solution is a wave of wavenumber k, travelling at velocity c, whose amplitude
(x, t) is slowly modulated. The amplitude (x, t) is itself a wave travelling
at velocity
W(k) W(
k)
k
k
. (2.139)
For

k close to k, the velocity (2.139) becomes
c
g
=
dW(k)
dk
. (2.140)
2.4 Linear waves 27
The velocity c
g
is called the group velocity. In general it diers from the
phase velocity
c =
W(k)
k
. (2.141)
For water waves, (2.130) and (2.140) give
c
g
=
1
2
__
gk +
T
k
3
_
tanh kh
_
1/2
__
g +
3Tk
2
_
tanh kh +
_
g +
T
k
2
_
kh
cosh
2
kh
_
. (2.142)
We now examine in more detail the case of innite depth. The phase
velocity c is then given by (2.95). Taking the limit kh in (2.142), we
obtain
c
g
=
1
2
_
gk +
Tk
3
_
1/2
_
g +
3Tk
2
_
. (2.143)
In particular, we have for pure gravity waves (g ,= 0, T = 0)
c
g
=
1
2
_
g
k
_
1/2
=
c
2
(2.144)
and for pure capillary waves (T ,= 0, g = 0)
c
g
=
3
2
_
Tk
_
1/2
=
3c
2
. (2.145)
The phase velocity c is the velocity at which the wave travels. The group
velocity c
g
is the velocity at which the slowly varying amplitude travels.
This phenomenon is illustrated in Figure 2.8, where we present the solution
(2.135) for pure gravity waves of innite depth.
Here we have assumed g = 1, k = 1,

k = 1.1 and A
1
= 0.2 and have
chosen t = 0. The outside curves are the envelope of the wave train. Both
the wave train and the envelope travel to the left. Using (2.96) we nd
that the wave train travels at the speed c = 1 whereas (2.144) shows that
the envelope travels at the speed c
g
= 1/2. Since c
g
< c the waves will
advance in their envelope, and as they approach the nodal points of their
envelope they will progressively die out. However, waves are born just ahead
of the nodal points of the envelope. These graphical results illustrate that
the wave travels at the velocity c whereas the envelope of the wave (i.e. the
amplitude) travels at the velocity c
g
.
28 Basic concepts
0
0 20 40
0.4
60
0.2
Fig. 2.8. The solution (2.135) for A
1
= 0.2, k = 1 and

k = 1.1.
A simple relation between c and c
g
can be derived by combining (2.129)
and (2.140) to give
c
g
= c +k
dc
dk
. (2.146)
Relation (2.146) shows that if c has a minimum for some value of k, then
c = c
g
at this minimum (since dc/dk = 0 at a minimum). For example, in
water of innite depth c has a minimum for k = k
min
, where k
min
is given by
(2.99) (see also Figure 2.6), and c
g
= c when k = k
min
. On the right of the
minimum in Figure 2.6 we have dc/dk < 0, and (2.146) implies that c
g
< c.
Similarly, dc/dk > 0 on the left of the minimum in Figure 2.6 and c
g
> c.
One important property of the group velocity is that it is the speed at
which the energy of a linear wave travels. We will demonstrate this property
in the particular case of pure gravity waves in water of nite depth. The
analysis is similar to that presented in Billingham and King [13]. At a xed
value of x, the rate at which the uid on the left does work on the uid on
the right is given by
_
0
h
p
x
dy. (2.147)
The average of (2.147) over one period is
E
f
=

2
_
t
+2/
t
_
0
h
p
x
dydt, (2.148)
where t
is an arbitrary value of t and is the angular frequency. The value

of p is obtained by linearising (2.12) (with = gy) around u = 0. This
2.4 Linear waves 29
gives
p =
t
gy + constant. (2.149)
Using (2.118) we nd
_
t
+2/
t
x
dt =
cA
1
k
sinh kh
cosh k(y +h)
_
t
+2/
t
cos k(x +ct)dt = 0.

(2.150)
Therefore (2.148) simplies to
E
f
=

2
_
t
+2/
t
_
0
h
x
dydt. (2.151)
Substituting (2.118) into (2.151) and evaluating the integral yields
E
f
=
A
2
1
k
2
c
3
4 sinh
2
kh
_
h +
sinh 2kh
2k
_
. (2.152)
We now dene the kinetic and potential energy per unit horizontal length
by
_
0
h
1
2
_
_
x
_
2
+
_
y
_
2
_
dy (2.153)
and
_

0
gydy =
1
2
g
2
. (2.154)
Averaging the quantities (2.153) and (2.154) over a wavelength gives the
mean kinetic energy
K =

2
_

0
_
0
h
_
_
x
_
2
+
_
y
_
2
_
dydx (2.155)
and the mean potential energy
V =
g
2
_

0
2
dx. (2.156)
Substituting (2.118) and (2.119) into (2.155) and (2.156) gives, after inte-
gration,
K =

V =
1
4
gA
2
1
. (2.157)
Thus the total energy is
E =

K +

V =
1
2
gA
2
1
. (2.158)
30 Basic concepts
Combining (2.152) and (2.158) we obtain
E
f
= E
1
2
c
_
1 +
2kh
sinh 2kh
_
. (2.159)
Using (2.142) with T = 0 gives
c
g
=
c
2
_
1 +
2kh
sinh 2kh
_
. (2.160)
Therefore comparing (2.159) and (2.160) yields
E
f
= Ec
g
. (2.161)
This shows that the energy in the wave travels at the group velocity c
g
.
This property will be used in Chapter 4, where we discuss the radiation
condition.
3
Free-surface ows that intersect walls
We continue our study of free-surface ows by considering the two-
dimensional ow shown in Figure 3.1. The ow domain is bounded be-
low by the horizontal wall AB and above by the inclined walls CD and DE
and the free surface EF. The uid is assumed to be incompressible and
inviscid and the ow is assumed to be irrotational and steady. We introduce
cartesian coodinates with the x-axis along the horizontal wall AB and the
y-axis through the separation point E (here a separation point refers to an
intersection of a free surface and a rigid wall). The angles between the walls
CD and DE and the horizontal are denoted by
1
and
2
respectively.
A B
D
E
F
x
y
2
C
Fig. 3.1. A two-dimensional free-surface ow bounded by the walls CD, DE and
AB and the free surface EF. The separation point E is dened as the point at
which the free surface EF meets the wall DE. Points C, A, F, B are at an innite
distance from E. The ow is from left to right.
The conguration of Figure 3.1 was chosen because it can be used to
describe many properties of free-surface ows that intersect, i.e. adjoin,
31
32 Free-surface ows that intersect walls
rigid walls. These properties when understood for the ow of Figure 3.1 can
then be used to describe locally ows with more complex geometries.
There are various illustrations of the ow of Figure 3.1. The rst is the
ow emerging from a container bounded by the walls CD, DE and AB.
When
1
=
2
= /2, the conguration of Figure 3.1 models the ow under
an innitely high gate (see Figure 3.2). Here the point D is irrelevant and
has been omitted from the gure.
A B
C
E
F
Fig. 3.2. The free-surface ow under a gate. The ow is from left to right.
When
1
= 0 and
2
< 0, Figure 3.1 describes locally the ow near the
bow or the stern of a ship (see Figure 3.3). A clear distinction between
the stern and bow ows will be introduced in Chapter 8, when we discuss
gravity ows with a train of waves in the far eld. Further particular cases
of Figure 3.1, which model bubbles rising in a uid and jets falling from a
nozzle, are described in Section 3.3.2.
A
D C
E
F
B
Fig. 3.3. A model for the free-surface ow near the bow or stern of a ship.
As mentioned in Chapter 1 we will proceed with problems of increasing
complexity. Section 3.1 is devoted to free-surface ows with g = 0 and
T = 0. Such ows are called free streamline ows and the corresponding
free surfaces are called free streamlines. In Section 3.2 we will study the
eect of surface tension (T ,= 0, g = 0). In Section 3.3 we will examine the
eect of gravity (T = 0, g ,= 0). The combined eects of gravity and surface
tension (T ,= 0, g ,= 0) are considered in Section 3.4.
3.1 Free streamline solutions
3.1.1 Forced separation
We consider the ow conguration of Figure 3.1. Here the eects of gravity
and surface tension will be neglected (T = 0, g = 0). We refer to this
problem as one of forced separation because the free surface is forced to
separate at the point E where the wall DE terminates. We denote by u
and v the horizontal and vertical components of the velocity. Using the
incompressibility of the uid and the irrotationality of the ow, we dene
a potential function (x, y) and a streamfunction (x, y). As shown in
Section 2.3, the complex potential f = + i and the complex velocity
w = u iv = df/dz are both analytic functions of z = x +iy.
The wall AB is a streamline along which we choose = 0. The walls CD
and DE and the free surface EF dene another streamline, along which
the constant value of is denoted by Q. We also choose = 0 at the
separation point E. These two choices ( = 0 on AB and = 0 at E)
can be made without loss of generality because and are dened up to
arbitrary additive constants. Bernoullis equation (2.13) with = 0 yields
1
2
(u
2
+v
2
) +
p
= constant (3.1)
everywhere in the uid. The free surface EF separates the uid from the
atmosphere which is assumed to be characterised by a constant pressure p
a
.
In the absence of surface tension, which we are assuming, the pressure is
continuous across the free surface (see (2.19)). Therefore p = p
a
on the free
surface. It follows from (3.1) that
u
2
+v
2
= U
2
on EF, (3.2)
where U is a constant.
A signicant simplication in the formulation of the problem is obtained
by using and as independent variables. This choice was used by Stokes
[144], to study gravity waves, and by Helmholtz [71] and Kirchho [90]
(see also [19] and [69]) to investigate free streamline ows. We shall use
it extensively in our studies of gravitycapillary free-surface ows. The
simplication comes from the fact that the ow domain is mapped into the
strip 0 < < Q shown in Figure 3.4. The free surface EF (whose position
was unknown in the physical plane z = x + iy of Figure 3.1) is now part
of the known boundary = Q in the f-plane. Since u iv is an analytic
function of z and z is an analytic function of f (the inverse of an analytic
function is also an analytic function), u iv is an analytic function of f.
B A
C
D E F
= 0
= Q
Fig. 3.4. The ow conguration of Figure 3.1 in the complex potential plane f =
+i.
A remarkable result is that many free streamline problems can be solved
in closed form (see Birkho and Zarantonello [19] and Gurevich [69]). These
exact solutions are obtained by using conformal mappings, and several meth-
ods have been derived to calculate them. The method we now choose to de-
scribe uses a mapping of the ow domain into the unit circle. It was chosen
because it yields naturally to the series truncation methods used in Sec-
tions 3.23.4 to solve numerically problems with gravity and surface tension
included.
In the absence of gravity and surface tension, the ow approaches a uni-
form stream of constant depth H as x . It follows from the dynamic
boundary condition (3.2) that this uniform stream is characterised by a
constant velocity U. Since = 0 on AB and = Q on EF, H = Q/U.
We dene the logarithmic hodograph variable i by the relation
w = u iv = e
i
. (3.3)
The function i has some interesting properties. First, the quantity
=
1
2
ln(u
2
+v
2
) is constant along free streamlines (see (3.2)). Second, can
be interpreted as the angle between the vector velocity and the horizontal.
Third, (3.3) leads, for steady ows, to a very simple formula for the curvature
of a streamline. This formula can be derived as follows. Since the vector
velocity is tangent to streamlines, is the angle between the tangent to a
streamline and the horizontal. The curvature K of a streamline is given by
(2.45). Using the chain rule, we can rewrite (2.45) as
K =
s
. (3.4)
Along a streamline is constant and therefore
s
= 0 and

s
= e
. (3.5)
Subsituting (3.5) into (3.4) yields the simple formula
K = e
. (3.6)
We now introduce dimensionless variables by using U as the reference
velocity and H as the reference length. Therefore = 1 on the walls CD
and DE and on the free surface EF. The dynamic boundary condition (3.2)
becomes
u
2
+v
2
= 1 on EF. (3.7)
We map the strip ABFC shown in Figure 3.4 into the unit circle in the
t-plane by the conformal mapping
e
f
=
(1 t)
2
4t
. (3.8)
The ow conguration in the t-plane is shown in Figure 3.5. It can easily
be checked that the points A and C are mapped into t = 0 and the points
B and F are mapped into t = 1. The value of t at the point D is denoted
by d. The free surface EF is mapped onto the portion
t = e
i
, 0 < < , (3.9)
of the unit circle. This can easily by shown by noting that the substitution
of (3.9) into (3.8) gives, after some algebra,
=
1
ln sin
2

2
on = 1. (3.10)
As varies from 0 to , varies from to 0, so that (3.9) is the image of
the free surface in the t-plane.
A C D E
F
B
Fig. 3.5. The ow conguration of Figure 3.1 in the complex t-plane.
One might attempt to represent the complex velocity w = u iv by the
series
w =
n=0
a
n
t
n
. (3.11)
However, the series will not converge inside the unit circle [t[ 1, because
singularities can be expected at the corner D and as x (i.e. at t = 0).
Nevertheless we can generalise the representation (3.11) by writing
w = G(t)
n=0
a
n
t
n
, (3.12)
where the function G(t) contains all the singularities of w. As we shall see
in Sections 3.23.4, this type of series representation enables the accurate
calculation of many free-surface ows with gravity and surface tension in-
cluded. For the present problem we require G(t) to behave like w as t 0
and as t d. We can then expect the series in (3.12) to converge for [t[ 1.
To construct G(t), we nd the asymptotic behaviour of w near the singu-
larities by performing local asymptotic analysis near D and as x .
The ow near D is a ow inside a corner. We will nd the nature of the
singularity at D by considering the general problem of a ow inside a corner
of angle (see Figure 3.6).
We introduce cartesian coordinates with the origin at the apex G of the
corner. We choose = 0 on the streamline HGL and = 0 at x = y = 0.
Assuming without loss of generality that the ow is in the direction of the
H
G
x
y
L
Fig. 3.6. Flow in a corner bounded by the walls GH and GL.
arrow, we have < 0 along the wall HG, > 0 along the wall GL and
> 0 in the ow domain. The ow conguration in the complex potential
plane is shown in Figure 3.7.
H
G
Fig. 3.7. The ow conguration of Figure 3.6 in the complex potential plane. The
ow domain is the upper half-plane > 0.
We seek a solution of the form
z = Ae
i
f
, (3.13)
where A > 0, and are real constants. On the wall GL (where > 0),
the kinematic boundary condition can be written as arg z = 0. Therefore
(3.13) implies that
= 0. (3.14)
On the wall GH (where < 0), the kinematic boundary condition can be
written as arg z = . Writing = e
i
[[ and using (3.13), we nd that
+ = . (3.15)
Relations (3.14) and (3.15) imply that
=

; (3.16)
therefore (3.13) gives
z = Af
/
. (3.17)
Since
w =
_
dz
df
_
1
(3.18)
we obtain the formula
w =

A
f
1/
(3.19)
or, eliminating f between (3.17) and (3.19),
w =

A
/
z
/1
. (3.20)
Flows inside corners will occur in many ow congurations described in this
book and we will refer often to the above local analysis. We note that the
formulae (3.17), (3.19) and (3.20) still hold if the boundary GL in Figure
3.6 is an arbitrary straight line through G (i.e. if the angle HGL is rotated).
The only dierence is that is then dierent from zero.
The velocity at the point G is equal to zero when < and is unbounded
when > see (3.20). We will refer to the ow of Figure 3.6 as a ow
inside a corner when < and as a ow around a corner when > .
For the ow of Figure 3.1, =
2
+
1
and (3.19) implies
w = O[(f
D
i)
(
2
1
)/)
] as f
D
+i, (3.21)
where
D
is the value of at the point D. Here we have used the classical
O notation to indicate an estimate of the behaviour of a function. We recall
that writing
f(x) = O[g(x)] as x x
0
(3.22)
means that
f(x)
g(x)
A as x x
0
, (3.23)
where A is a constant. Similarly,
f(x) = o[g(x)] as x x
0
(3.24)
means that
f(x)
g(x)
0 as x x
0
. (3.25)
Using (3.8) yields
f
D
i = O[t d] as f
D
+i. (3.26)
Combining (3.21) and (3.26) gives
w = O[(t d)
(
2
1
)/)
] as t d. (3.27)
This concludes our local analysis near the point D.
As x , the ow behaves like that due to a sink at x = y = 0.
Therefore
f Bln z as x (3.28)
where B is a positive constant. Dierentiating (3.28) with respect to z gives
w =
df
dz
=
B
z
. (3.29)
Since the ux of the uid coming from is 1 and the angle between the
walls CD and AB is
1
, we have
B =
1
1
. (3.30)
Eliminating z between (3.28) and (3.29) gives
w = O[e
1
f
] as f , (3.31)
and relation (3.8) then implies that
e
f
= O(t) as f . (3.32)
Therefore (3.31) and (3.32) give
w = O(t
1
/
) as t 0. (3.33)
Combining (3.27) and (3.33), we can choose
G(t) = (t d)
(
2
1
)/
t
1
/
(3.34)
and write (3.12) as
w = (t d)
(
2
1
)/
t
1
/
n=0
a
n
t
n
. (3.35)
There are, of course, many other possible choices for G(t). For example G(t)
in (3.34) can be multiplied by any function analytic in [t[ 1.
We now need to determine coecients a
n
in (3.35) such that the dynamic
boundary condition (3.7) is satised. This can be done numerically by trun-
cating the innite series in (3.35) after N terms and nding the coecients
a
n
, n = 0, . . . , N 1 by collocation. This is the approach we will use when
solving problems where the eects of gravity or surface tension are included
in the dynamic boundary condition. However, it can checked that
n=0
a
n
t
n
=
_
1
1 td
_
(
2
1
)/
(3.36)
and therefore the present problem has the exact solution
w =
_
t d
1 td
_
(
2
1
)/
t
1
/
. (3.37)
The existence of an exact solution for the ow of Figure 3.1 follows from
the general theory of free streamline ows. This theory was developed by
Kirchho [90] and Helmholtz [71]; see Birkho and Zarantonello [19] or
Gurewich [69] for details.
The free-surface prole is obtained by setting = 1 in (3.8) and (3.37),
calculating the partial derivatives x
and y
from the identity

x
+iy
=
1
w
(3.38)
and integrating with respect to .
As a rst example let us assume that
1
=
2
= /2 (see Figure 3.2).
Then (3.37) reduces to
w = t
1/2
(3.39)
and (3.9), (3.38) and (3.39) yield
x
+iy
= e
i/2
, 0 < < , (3.40)
along the free surface EF. Dierentiating (3.10) with respect to and
applying the chain rule to (3.40) gives
x
+iy
=
1
cotan

2
e
i/2
. (3.41)
Integrating (3.41) with respect to and taking the real and imaginary parts
gives
x =
2
cotan

2
+

1 (3.42)
y =
2
sin

2
+ 1. (3.43)
Relations (3.42) and (3.43) dene the free-surface prole in parametric form.
It is shown in Figure 3.8.
0
0.5
1.0
1.5
2.0
0 2 4 6 8
Fig. 3.8. Free-surface prole for the ow conguration of Figure 3.2. The position
of the separation point E is indicated by a small horizontal line. The vertical scale
has been exaggerated to show clearly the free-surface prole.
A classical parameter associated with this ow is the contraction ratio C
c
,
dened as the ratio y
F
/y
E
of the ordinates of the points F and E. Using
(3.43) with = and = 0, we obtain
C
c
=

+ 2
0.611. (3.44)
As a second example, let us assume
1
= 0 and
2
= /2 (see Figure 3.9).
Then (3.37) becomes
w =
_
t d
1 td
_
1/2
. (3.45)
Proceeding as in the previous example, we obtain
x
+iy
=
1
cotan

2
_
1 e
i
d
e
i
d
_
1/2
(3.46)
on the free surface EF.
A
B
C
D
E
F
Fig. 3.9. A free-surface ow emerging from a container bounded by the horizontal
walls CD and AB and by the verical wall DE.
Integrating (3.46) gives x and y on the free surface as functions of .
There is a solution for each value of 1 < d < 0; the parameter d measures
the length of the vertical wall DE in the complex t-plane. This is an inverse
formulation in the sense that for each value of d the length of the wall DE
in the physical plane is found at the end of the calculations, in the following
way. We rst calculate y
for 1 < t < d by using (3.38) and (3.45). We

then evaluate y
t
for 1 < t < d by using (3.8) and the chain rule. The
length of the wall DE is then obtained by integrating with respect to t from
1 to d. A typical solution for d = 0.5 is shown in Figure 3.10.
0
0.4
0.8
1.2
1.6
0 1 2 3 4
Fig. 3.10. Computed free-surface prole for the ow conguration of Figure 3.9
with d = 0.5. The position of the separation point E is indicated by a small
horizontal line. The vertical scale has been exaggerated to show clearly the free-
surface prole.
As d 0, the length of the vertical wall DE tends to innity and the
ow reduces to that of Figure 3.2. As d 1, the length of the vertical
wall DE tends to zero and the ow reduces to a uniform stream.
As a third example, we assume
2
< 0 and
1
= 0 (see Figure 3.3). As
mentioned at the begining of this chapter, this conguration models the
ow due to a surface-piercing obstacle moving at a constant velocity when
viewed in a frame of reference moving with the obstacle. In particular it is
a simple model for the ow near the stern or the bow of a ship. Again using
(3.37), we obtain
w =
_
t d
1 td
_
2
/
. (3.47)
As in the previous two examples we use (3.47) to calculate x
+iy
on the
free surface. After integration we obtain the shape of the free surface in
parametric form. A typical free-surface prole for d = 0.2 and
2
= /3
is shown in Figure 3.11.
0
0.2
0.4
0.6
0.8
1.0
0 1 2 3 4
Fig. 3.11. Computed free-surface prole for the ow conguration of Figure 3.3
with d = 0.2 and
2
= /3. The position of the separation point E is indicated
by a small horizontal line. The vertical scale has been exaggerated to show clearly
the free-surface prole.
3.1.2 Free separation
In Figures 3.1 and 3.9, on the one hand, the free surface is forced to separate
from the rigid wall DE at E because the wall DE terminates at E. We refer
to this situation as forced separation. On the other hand, if the innitely
thin wall DE is replaced by a wall of nite thickness bounded by a smooth
curve then in principle the point of separation E can be any point on the
smooth curve (see Figure 3.12). We refer to this situation as free separation.
A B
F
C
D
E
Fig. 3.12. The ow conguration of Figure 3.9 but with the vertical wall DE re-
placed by a wall bounded by a smooth curve.
We note that any solution corresponding to free separation represents
also a solution with forced separation if the smooth curve is cut along a line
through the separation point (see Figure 3.13). As we shall see in Section
3.2, the distinction between forced and free separation is important when
studying the eects of surface tension.
A B
F
C
E
D
Fig. 3.13. The ow conguration of Figure 3.12 when the smooth curve is cut by a
vertical line.
3.1.2.1 Open cavities
We now consider some solutions with free separation which will be useful
in Section 3.2 when we consider the eects of surface tension. Figure 3.14
shows a particular case of Figure 3.12 for which the vertical rigid wall DE
of Figure 3.9 has been replaced by a smooth elliptical wall with equation
_
x
a
_
1/2
+
_
y
b
_
1/2
= 1. (3.48)
Here x and y refer to coordinates with the origin at the centre of the ellipse
and a and

b are the semi-axes of the ellipse.
A B
E
F
C
D
Fig. 3.14. The ow of Figure 3.9 when the vertical wall DE is replaced by a smooth
semi-elliptical wall.
If a
b, the semi-ellipse is thin and the conguration of Figure 3.14 can

be viewed as that of Figure 3.9 but with the innitely thin wall DE replaced
by a smooth wall of nite thickness. In other words, Figure 3.14 takes into
account the nite thickness of any real wall but approaches the conguration
of Figure 3.9 as a/
b 0. However, to study ows with free separation we

shall assume that

b = a (i.e. that the semi-ellipse is a semicircle), so that
ows corresponding to dierent positions of the separation point E can be
clearly distinguished on the proles.
The ow of Figure 3.14 can be reected in the wall CD. This yields the
ow of Figure 3.15. It models a ow past a circular cylinder with a cavity
behind it (see for example Batchelor [8] for a discussion of cavitating ows).
Fig. 3.15. Cavitating ow past a circular object in a domain bounded by two hor-
izontal walls.
We shall study the ow of Figure 3.15 when the radius of the circle is very
small compared with the distance between the horizontal walls, so that the
circle can be assumed to be in a uid unbounded in the vertical direction
(see Figure 3.16). The angle between the free surface and the circle at the
separation points is denoted by . For free streamline solutions = 0. This
follows from (3.20) with = , which shows that a value ,= 0 would
generate a zero or an innite velocity at the separation points. This would
contradict (3.2). However, we shall see in Sections 3.2 and 3.4 that values
,= 0 can occur when surface tension is taken into account.
C
B
G
x
A
y
D
E
Fig. 3.16. The cavitating ow past a circle in an unbounded uid domain. When
the surface tension T is zero, the free surfaces leave the circle tangentially and
= 0. When T ,= 0, the angle can be dierent from zero.
We dene dimensionless variables by using the radius R of the circle as the
reference length and the constant velocity U far upstream as the reference
velocity. We introduce the potential function b, the streamfunction b and
the complex potential f = b + ib. Without loss of generality we may
choose = 0 at the point C and = 0 on the streamlines ECAD and
ECBG. Here and in the remaining part of this section, the letters E, C, B,
G, A and D refer to Figure 3.16. The constant b is dened so that = 1
at the separation points A and B. The ow conguration in the complex
potential plane is illustrated in Figure 3.17.
We introduce the complex velocity u iv and dene the function i
by the relation (3.3). Using (3.6), we have
K =
e
. (3.49)
We shall seek i as an analytic function of + i in the half-plane
< 0 (see Figure 3.17). The solution in > 0 can then be obtained by
symmetry. The boundary conditions on = 0 are given by
= 0 on = 0, < < 0, (3.50)
+1
A
B
C D E
G
Fig. 3.17. The ow of Figure 3.16 in the complex potential plane.
e
= 1 on = 0, 0 < < 1, (3.51)

= 0 on = 0, 1 < < . (3.52)
The condition (3.50) follows from symmetry. Equation (3.51) follows from
(3.49) and the fact that the curvature of the rigid boundary ACB is 1.
Relation (3.52) is the dynamic boundary condition rewritten in terms of .
This completes the formulation of the problem. We seek i as an
analytic function of + i in < 0 satisfying (3.50)(3.52). We will solve
the problem by following the series truncation method introduced in Section
3.1.1 (see (3.12)). First we map the ow domain into the unit circle in the
complex t-plane by the transformation
f
1/2
=
_
t
1
t
_
1
2i
. (3.53)
The ow conguration in the t-plane in shown in Figure 3.18. The rigid
surface ACB is mapped onto the circle [t[ = 1 and the free surfaces AD
and BG are mapped onto the imaginary axis. The conditions (3.50)(3.52)
become
= 0 on 0 < t < 1, (3.54)
e
= 1 on t = e
i
, /2 < < 0, (3.55)
= 0 on t = ir, 1 < r < 0. (3.56)
Here we have described the unit circle [t[ = 1 by t = e
i
, where is a real
parameter.
Following Brodetsky [23] and Vanden-Broeck [160] we represent i by
A
B
C
D
G
E
Fig. 3.18. The ow of Figure 3.16 in the complex t-plane.
an expansion, as follows:
i = ln
1 +t
1 t

n=0
B
n
t
n
. (3.57)
The derivation of (3.57) follows that leading to (3.12). There is a singu-
larity at the point C where locally we have a ow inside a right angle corner
(see Figure 3.16). Therefore, (3.19) yields
u iv f
1/2
as f 0. (3.58)
Using (3.3) and (3.53) yields i ln(1 t) as t 1. Thus
i + ln
1 +t
1 t
(3.59)
is not singular and can be represented in the unit circle of the t-plane by a
Taylor expansion. This leads to (3.57). One might argue that other singu-
larities occur at the separation points A and B. However, these singularities
are automatically taken into account by (3.53). We note that (3.57) implies
u iv = G(t) exp
_
n=0
B
n
t
n
_
, (3.60)
where
G(t) =
1 t
1 +t
. (3.61)
Therefore (3.60) is similar to (3.12). The only dierence is that the series
has been rewritten as the exponential of a series.
It can easily be checked that (3.54) and (3.56) are satised by assuming
that the coecients B
n
are real and that B
n
= 0 when n is even. Therefore
we can rewrite (3.57) as
i = ln
1 +t
1 t
+
n=1
A
n
t
2n1
. (3.62)
We now determine coecients A
n
such that (3.55) is satised. This is done
numerically by series truncation and collocation. Thus we truncate the
innite series in (3.62) after N terms, i.e. we write
i ln
1 +t
1 t
+
N
n=1
A
n
t
2n1
. (3.63)
Next, we satisfy (3.55) at the mesh points =
I
, where
I
=

2N
I, I = 1, 2, . . . , N. (3.64)
This is achieved by using (3.62) to evaluate the values of , and / at
the mesh points (3.64) and substituting these values into (3.55). This leads
to a system of N equations for the N + 1 unknowns A
n
, n = 1, 2, . . . , N,
and b. The last equation is obtained by xing the position of the separation
point A. This is done by imposing
(
N
) =

2
, (3.65)
where the angle is dened in Figure 3.16.
The system of N +1 nonlinear algebraic equations with N +1 unknowns
needs to be solved numerically by iteration. In most problems considered in
this book, this is done by Newtons method. This method can be described
as follows. Assume that we want to solve a system of M nonlinear algebraic
equations
f
i
(x
1
, x
2
, . . . , x
M
) = 0, i = 1, 2, . . . , M, (3.66)
with M unknowns x
1
, x
2
, . . . , x
M
. Let (x
(n)
1
, x
(n)
2
, . . . , x
(n)
M
) be the approxi-
mation of the solution at iteration n. Then we linearise the left-hand side
of (3.66) around this iteration as
f
i
(x
(n)
1
, x
(n)
2
, . . . , x
(n)
M
) +
M
j=1
_
x
j
x
(n)
j
_
_
f
i
x
j
_
(n)
. (3.67)
The next approximation, (x
(n+1)
1
, x
(n+1)
2
, . . . , x
(n+1)
M
), is obtained by equat-
ing (3.67) to zero and solving the resulting linear system for x
1
, x
2
, . . . , x
M
.
Each iteration is expensive since it requires solving a linear system of equa-
tions. However, the iterations usually converge quadratically so that only
a few iterations are needed to obtain an accurate solution. The method
also requires an initial guess (x
(0)
1
, x
(0)
2
, . . . , x
(0)
M
) to start the iterative pro-
cess. When facing a problem with several solutions, the solution obtained
after convergence will depend on the initial guess chosen. The matrix with
elements
f
i
x
j
(3.68)
is called the Jacobian matrix; an attractive feature of Newtons method is
that bifurcations from branches of solutions can be found by monitoring
the sign of its determinant. This is a consequence of the fact that the
determinant vanishes at a bifurcation point (Keller [84]).
The free-surface proles are then obtained by integrating numerically the
identity
1
b
_
x
+i
y
_
= e
+i
. (3.69)
The numerical results can be described in terms of the angle (see Figure
3.16). Solutions can be obtained for all values 0 < < . However, only the
solutions for
< <
, where
55
and
124
, have a physical
meaning for cavitating ow past a circle.
For <
the solutions are not acceptable because then the free surfaces
would enter the body (see the solution for = 25
in Figure 3.19). They are

nevertheless useful in describing the cavitating ow past the body obtained
by cutting the circle along the straight line AB in Figure 3.16 and retaining
only the portion on the left of AB. This cutting of the circle is similar to
the cutting seen in Figure 3.13.
For >
the solutions are not acceptable because the free surfaces

cross each other (see the solution for = 150
in Figure 3.19). The last

acceptable solution, at =
, has free surfaces that approach the x-axis

asymptotically as x (see Figure 3.20).
Physically acceptable solutions for >
can be obtained by considering

cusped cavities. Cusped cavities were introduced numerically by Southwell
and Vaisey [142] and analytically by Lighthill [98] and [99]. They will be
calculated numerically by a boundary integral equation method in the next
section.
0
0.5
1.0
1.5
0 1 2 3 4 5
Fig. 3.19. Computed free-surface proles for = 25
, =
and = 150
.
0
1
2
3
0 2 4 6 8 10
Fig. 3.20. The cavitating ow corresponding to =
124
.
3.1.2.2 Cusped cavities
Unwanted intersections of free surfaces, such as those described above for
>
, occur in many applications. A classical example is the exact so-

lution of Crapper [37] for nonlinear capillary waves travelling at a constant
velocity at the surface of a uid of innite depth (see Section 6.5.1 and
Figures 6.86.11). Crappers solutions form a one-parameter family of solu-
tions. The parameter can be chosen as the steepness s of the waves (i.e. the
dierence in height of the crests and the troughs divided by the wavelength).
For small values of s, the waves are close to linear sine waves (see Figure
6.8). As s increases the waves develop rounded crests and sharp troughs (see
Figure 6.9). When s reaches the critical value s
0.73, the free surface

develops a point of contact with itself and a small trapped bubble forms
at the trough of the wave (see Figure 6.10). For s > s
, the free surface

is self-intersecting and the solutions lose their physical meaning (see Figure
6.11). Vanden-Broeck and Keller [185] showed that physically acceptable
solutions for s > s
can be obtained by preventing the free surface from

self-intersecting. The resulting free-surface proles for s > s
have trapped
bubbles at the troughs, as in Crappers solution for s = s
. Since prevent-
ing self-intersection imposes an extra constraint on the solutions, an extra
unknown is needed. This is provided by the pressure in the trapped bubble,
which is found as part of the solution.
The calculations of Vanden-Broeck and Keller [185] will be described in
Section 6.5.1. Here we use a similar approach to nd physically acceptable
cavitating ows for >
, by preventing the crossing of the streamlines

and seeking a family of cusped cavities (see Figure 3.21).
C
E
L
M
B
A
y
x
D
G
Fig. 3.21. Flow past a circle giving rise to a cusped cavity.
As we shall see there is a cusped cavity for each value of >
. These
solutions approach the solution in Figure 3.20 as
. In other words
the x-coordinate of the cusp in Figure 3.21 tends to as
and the
corresponding solution approaches that of Figure 3.20. As 180
, the
x-coordinate of the cusp tends to 2 and the cavity collapses to a point.
Following the work of Vanden-Broeck and Keller [185], as mentioned above
we need to identify a new unknown to prevent the intersection of the free
streamlines. A natural choice is the pressure p
c
in the cavity. This is mo-
tivated by the fact that cusped cavities are closed (they do not extend to
innity as do the open cavities of Figure 3.19) and so we do not have to
require that p
c
= p
b
. Therefore our dynamic boundary condition on the free
surfaces AL and BM of Figure 3.21 is
=
1
2
ln(1 +(), (3.70)
where the cavitation number ( is found as part of the solution. We dene
the potential function b and the streamfunction b and choose b so that
= 1 at the separation points B and A. The ow conguration in the
complex (, )-plane is illustrated in Figure 3.22.
+1
A
B
C D
E
G
Fig. 3.22. The ow of Figure 3.21 in the complex potential plane.
We solve the problem by a boundary integral equation method. This
technique will be used extensively in the remaining part of the book. The
basic idea is to reformulate the problem as a system of integro-dierential
equations that involves only unknowns on the boundary of the ow do-
main. This system is then discretised and the resultant algebraic equations
are solved by iteration (usually Newton iteration). The obvious advan-
tage is that mesh points are only needed on the boundary rather than in
the whole ow domain. In other words the two-dimensional ow problem
of Figure 3.21 is reduced to a one-dimensional problem on the boundary
ECALD. As we shall see in Chapter 10, boundary integral equation meth-
ods can also be used for solving fully three-dimensional problems. There the
three-dimensional problem is reduced to a two-dimensional problem on the
boundary. A convenient way of deriving the system of integro-dierential
equations for two-dimensional ows is to use the Cauchys integral equa-
tion formula (see (2.38)(2.41)). An alternative way which does not rely
on complex variables is to use Greens theorem and Greens functions. For
three-dimensional problems, complex variables are not available and Greens
theorem and Greens functions are the only way to derive the system of
integro-dierential equations.
We can derive such a system for the problem of Figure 3.21 by applying
the Cauchy integral equation formula in the (, )-plane of Figure 3.22 to
the function (, ) i(, ) with a contour consisting of the axis = 0
and a semicircle in < 0 centred on = = 0 and of arbitrary large
radius. Since (, ) i(, ) 0 as , there is no contribution
from the semicircle and we obtain
(, )i(, ) =
1
2i
_

(, 0) i(, 0)
i
d when < 0. (3.71)
(see (2.40)). On the free surface, (2.41) gives
(, 0) i(, 0) =
1
i
_

(, 0) i(, 0)

d. (3.72)
The integral in (3.72) is a Cauchy principal value. Taking the real and
imaginary parts of (3.72) gives
(, 0)) =
1
(, 0)

d, (3.73)
(, 0) =
1
(, 0)

d. (3.74)
Relations (3.73) and (3.74) are known as Hilbert transforms. It can be
shown that one implies the other. Therefore we are free to choose either
(3.73) or (3.74). It turns out that (3.73) is the better choice because it leads
to a relation between and on the portion CAL of the streamline = 0.
This follows from the fact that = 0 on EC and on LD. Therefore (3.73)
simplies to
(, 0) =
1
_
l
0
(, 0)

d. (3.75)
Here l is the value of at the cusp L. If we restrict the values of in (3.75)
to 0 < < l, then (3.75) is a relation between and on CAL.
The kinematic boundary condition on CA and the dynamic boundary
condition (3.70) imply that
e
= 1, 0 < < 1, (3.76)

=
1
2
ln(1 +(), 1 < < l. (3.77)
Finally, we impose y = 0 at the cusp by writing
_
l
0
e
(,0)
sin (, 0)d = 0. (3.78)
This completes the reformulation of the problem as a system of nonlinear
integro-dierential equations. We seek (, 0) and (, 0) such that (3.75)
(3.78) are satised. Once (, 0) and (, 0) are known for 0 < < l, then
the shape of the cusped cavity and the velocity eld in the ow domain can
be calculated by integration in the following way. First the shape of the
cavity is obtained in the parametric form x(, 0), y(, 0) by integrating the
identity
x
+iy
=
1
u iv
= e
(,0)+i(,0)
. (3.79)
Next, (, 0) for < 0 and > l can be calculated from (3.75). The values
of (, 0) and (, 0) are then known for all < < . Substituting
these values in (3.71), we can evaluate by integration (, ) and (, )
everywhere in the ow domain. The velocity eld is then given by (3.3).
We will solve the problem numerically. First we dene the mesh points
I
=
I 1
N 1
, I = 1, . . . , M, (3.80)
and the corresponding unknowns
I
= (
I
, 0), I = 1, . . . , M, (3.81)
where M and N are positive integers and l = (M 1)/(N 1). Since l > 1,
we require M > N. We also use the midpoints
m
I
=

I
+
I+1
2
, I = 1, . . . , M 1. (3.82)
We calculate (
m
I
) in terms of the unknowns (3.81) by applying the trape-
zoidal rule to the integral in (3.75) and summing over the points (3.80). We
justify this discretisation by showing that the symmetry of the quadrature
and of the distribution of mesh points enable us to calculate the Cauchy
principal value as if it were an ordinary integral. First we rewrite the inte-
gral on the right-hand side of (3.75) (evaluated at
m
I
) as
_

I
0
(, 0)

m
I
d +
_

I +1
I
(, 0)

m
I
d +
_
l
I +1
(, 0)

m
I
d. (3.83)
The rst and third integrals in (3.83) are ordinary integrals and can therefore
be evaluated by the trapezoidal rule. The second integral in (3.83) is a
Cauchy principal value, which we rewrite as
_

I +1
I
(, 0) (
m
I
, 0)

m
I
d +(
m
I
, 0)
_

I +1
I
d

m
I
. (3.84)
The second integral in (3.84) is also a Cauchy principal value. Simple inte-
gration shows that its value is zero. The rst integral in (3.84) is an ordinary
integral and can be evaluated by the trapezoidal rule as
I
(
m
I
, 0)
I

m
I
h
2
+

I+1
(
m
I
, 0)
I+1
m
I
h
2
=

I+1
I+1
m
I
h
2
+

I
I

m
I
h
2
. (3.85)
The right-hand side of (3.85) is just the integral
_

I +1
I
(, 0)

m
I
d
evaluated by the trapezoidal rule. Therefore the Cauchy principal value on
the right-hand side of (3.75) can be evaluated by the trapezoidal rule as if
it were an ordinary integral. This approach to evaluating Cauchy principal
values will be used often in this book. We note that the derivation (3.83)
(3.85) can easily be extended to other integration formulae such as Simpsons
rule or for mesh points
M
I
that are not midpoints. The only dierences are
that the second integral in (3.84) might not be zero and that the left-hand
side of (3.85) should be used instead of the right-hand side.
We now return to our problem and satisfy (3.76) at the mesh points
m
I
,
I = 2, . . . , N 1, and (3.77) at the mesh points
m
I
, I = N, . . . , M2. The
last three equations are given by (3.78) and by the geometric conditions
1
=
2
,
M
= 0. (3.86)
This system of algebraic equations is solved by Newtons method. Typical
free-surface proles are shown in Figure 3.23.
It can be seen that as
, ( 0.
We note that the numerical procedure presented here is not restricted to
a circular obstacle and can be generalised to ones of arbitrary shapes in
the following way. First we denote by F(x, y) = 0 the equation of the rigid
boundary CA (see Figure 3.21) and calculate x and y on CA by the formulae
x(, 0) =
_

0
e
(,0)
cos (, 0)d (3.87)
0
1
2
3
0 2 4 6 8 10
Fig. 3.23. Three computed cusped cavities. The cavitation numbers ( from the
smallest cavity to the largest are 0.55, 0.29 and 0.1 respectively.
and
y(, 0) =
_

0
e
(,0)
sin (, 0)d. (3.88)
We then apply the numerical procedure described above, the equations ob-
tained by satisfying (3.76) at the mesh points
m
I
, I = 2, . . . , N 1, being
replaced by the new equations
F[x(
m
I
, 0), y(
m
I
, 0)] = 0, I = 2, . . . , N 1, (3.89)
where x(, 0) and y(, 0) are dened by (3.87) and (3.88).
The solutions derived in this section are examples of cavitating ows with
( < 0. Such cavities were considered analytically by Lighthill [98], [99].
Batchelor [8] notes that such cavities have not been observed, perhaps be-
cause the boundary layer at the rigid surface would separate before reaching
the low-velocity region where the free streamlines begin.
Before concluding this section, let us mention that there are many cavity
models with ( > 0 (the Riabouchinsky model, the re-entrant jet model, the
Roskho model etc). The reader interested in these models is referred to the
books of Birkho and Zarantonello [19] and Gurevich [69].
3.2 The eects of surface tension
In this section we will investigate the eects of the surface tension T on the
free streamline solutions of Section 3.1. We show that the limit T 0 is
singular. When T ,= 0, discontinuities in slope can appear at the separation
points. In particular, values of ,= 0 can occur in Figure 3.16. We shall
also show that the limit T 0 can be used to select solutions.
3.2.1 Forced separation
We start our study by investigating the local behaviour of the ow of Figure
3.1 near the separation point E, in the absence of surface tension. For
simplicity we assume
1
=
2
= /2, i.e. we consider the ow shown in
Figure 3.2. The point E corresponds to t = 1, = 1 and = 0. Using
(3.8) we nd

1
4
(t + 1)
2
as t 1. (3.90)
Relation (3.39) gives
w i
i
2
(t + 1) as t 1. (3.91)
Furthermore (3.3) gives
w i + +

2
as t 1. (3.92)
Combining (3.90)(3.92), we obtain

2
+ ()
1/2
. (3.93)
Since e
= 1 at E, (3.6) implies that

K S
1/2
as 0, (3.94)
where
S =
1
2
1/2
. (3.95)
Therefore the ow leaves the wall DE tangentially (see (3.93)) but the cur-
vature of the free surface at E is unbounded (see (3.94)). It can be shown
that (3.94) holds for
1
,= /2 and
2
,= /2. Of course, the value of S
depends on
1
and
2
.
These free streamline results show that an innite curvature can occur at
the separation points. This singularity does not invalidate the free stream-
line theory because the curvature does not appear explicitly in the equations
and does not have a direct physical meaning. However, when surface tension
is taken into account, the condition p = p
a
on the free surface is replaced by
p = p
a
+TK, (3.96)
where T is the surface tension and K is the curvature of the free surface
(see (2.19)). It follows from (3.1) and (3.96) that the dynamic boundary
condition on the free surface becomes
1
2
(u
2
+v
2
) +
T
K = constant. (3.97)
Equation (3.97) shows that an innite curvature at the separation point
E implies an innite velocity at E. This implies that solutions with T ,= 0
are qualitatively dierent from the solutions with T = 0 of Section 3.1.1. It
also suggests that the limit T 0 is a singular limit. These two properties
are conrmed by the calculations below.
Ackerberg [2], Cumberbatch and Norbury [39], Vanden-Broeck ([159],
[160], [163], [164], [176]) and others studied the ow conguration of Fig-
ure 3.2 (and related free-surface ows) in the limit as T 0. The results
of Vanden-Broeck showed that the inclusion of surface tension in the free
streamline ows of Section 3.1.1 does not remove the innite curvature at
the separation points. On the contrary, it makes the ow more singular
by introducing a discontinuity in slope at the separation points. In other
words there is an angle ,= 0 between the tangent to the free surface at the
separation point and the wall (see Figure 3.24).
A
B
C
E
F
Fig. 3.24. The ow under a gate with surface tension included in the dynamic
boundary condition. The free surface does not leave the gate tangentially: there is
an angle ,= 0 between the free surface and the gate at the separation point E.
This angle is a function of the surface tension. We now demonstrate
these ndings by presenting asymptotic results, for T small. We will present
later fully nonlinear computations for arbitrary values of T.
We assume that the ow in Figure 3.24 is characterised by a uniform
stream with constant velocity U as x . As in Section 3.1.1, we dene
dimensionless variables by taking U as the unit velocity and H = Q/U (for
a denition of Q see Figure 3.4) as the unit length. The dynamic boundary
condition (3.97) in dimensionless form is then
1
2
(u
2
+v
2
) +
1
K = constant, (3.98)
where
=
U
2
H
T
. (3.99)
Since u
2
+ v
2
1 and K 0 as , the constant on the right-hand
side of (3.98) is equal to 1/2. Using (3.3) and (3.6), we rewrite (3.98) as
1
2
e
2
=
1
2
. (3.100)
If we assume = 0 at the separation point then the free streamline
solution (i.e. the solution for = ) for the conguration of Figure 3.1 can
be described near the separation point by

0
C
1/2
as 0, (3.101)
where C is a constant. Here
0
is the value of at the separation point when
= . For example, for the ow of Figure 3.24, (3.93) shows that
C =
1/2
and
0
= /2. (3.102)
Relation (3.101) implies that the curvature of the free surface near E behaves
like
K = e

1
2
C
1/2
as 0. (3.103)
Therefore the curvature of the free surface is unbounded at the separation
point E unless C = 0.
Following Ackerberg [2] we introduce the following scaling of the variables:
f
= f, (3.104)
=
1/2
( i +i
0
). (3.105)
The function
satises Laplaces equation in
< 0. Thus
2
+

2
2
= 0. (3.106)
The kinematic and dynamic boundary conditions linearise in the limit
, so that the boundary conditions on
= 0 are
= 0 on
= 0,
< 0, (3.107)
= 0 on
= 0,
> 0. (3.108)
Relation (3.101) gives the outer behaviour
C(f
)
1/2
as [f
[ , (3.109)
where indicates the imaginary part of a function. Cumberbatch and Nor-
bury [39] showed that the solution of (3.106)(3.108) not containing waves
and having the weakest singularity at the separation point
= 0 is given
on the free surface by
) =
1
2
C
1/2
+
C
2
1/2
ln
as
0, (3.110)
) =
1
2
1/2
C ln
as
0. (3.111)
The solution (3.110), (3.111) is not valid near
= 0 because
is un-
bounded at
= 0 (an unbounded value of
invalidates the linearisation).

The asymptotic scheme can now be described as follows. For large we
have an outer solution whose rst term is the free streamline solution (i.e.
the solution without surface tension). This solution merges with the solution
(3.110), (3.111) obtained for
1, i.e. for
1
. Since the solution
(3.110), (3.111) becomes invalid as
0, we follow Vanden-Broeck [159]

and seek a local solution that corresponds to a ow past a corner of angle
0
. Thus on the one hand, using (3.19) and (3.3), we can write
e
=

A
0
/
0
1
, (3.112)
which implies that, near = 0,

_
1

0
_
ln . (3.113)
On the other hand (3.105) and (3.111) give
=

1/2

1
2
1
()
1/2
C ln . (3.114)
A comparison of (3.113) and (3.114) yields
0
=
C
2
_
_
1/2
. (3.115)
Thus we have matched the solution (3.110), (3.111) with a local solution
corresponding to a ow in a corner of angle
0
(cf. Figure 3.6).
The value of at the separation point is
=
0
C
2
_
_
1/2
. (3.116)
If we denote by the angle between the wall and the free surface at E (see
Figure 3.24) then (3.116) implies that
=
C
2
_
_
1/2
. (3.117)
For the particular ow of Figure 3.24, (3.102) and (3.116) yield

2
1/2
as . (3.118)
We now present fully nonlinear solutions for arbitrary values of . The
presence of surface tension changes drastically the dynamic boundary condi-
tion and invalidates the techniques used for streamline ows. Exact solutions
can no longer be expected and fully nonlinear solutions have to be calculated
numerically. There are, however, a few examples of exact solutions. Those
will be considered in Section 6.5.1.
We can calculate nonlinear solutions for the ow conguration of Figure
3.1 by modifying appropriately the series representation (3.35) to accommo-
date the singularity at t = 1. Using insight given by the asymptotic result
(3.117), we assume that the ow near t = 1 is a ow in an angle + .
Using (3.19) and (3.90) we obtain
w f
/
as 0 (3.119)
and
w (t + 1)
2/
as t 1. (3.120)
Therefore
w = (t d)
(
2
1
)/
t
1
/
(t + 1)
2/
n=0
a
n
t
n
(3.121)
is the appropriate generalisation of (3.35) when surface tension is included.
The asymptotic solution (3.117) for large suggests that should be found
as part of the solution.
We now present explicit calculations in the particular case
1
=
2
= /2.
In other words we consider the ow conguration of Figure 3.24. Then the
expression (3.121) becomes
w = t
1/2
(t + 1)
2/
n=0
a
n
t
n
. (3.122)
The dynamic boundary condition is given in dimensionless variables by
(3.100), where is dened in (3.99).
We truncate the innite series in (3.122) after N terms and calculate the
coecients a
n
, n = 0, . . . , N 1, and by satisfying (3.100) at the N + 1
equally spaced mesh points
I
=

N + 1
_
I
1
2
_
, I = 1, . . . , N + 1. (3.123)
This leads to a system of N +1 equations with N +1 unknowns, which can
be solved by Newtons method. We present numerical results in terms of
the parameter
v
= 2. (3.124)
The factor 2 in (3.124) has been introduced so that
v
coincides with the
parameter used by Vanden-Broeck [163].
Typical free-surface proles are shown in Figure 3.25. For
v
= , the
free-surface prole reduces to the free streamline solution of Figure 3.8. As
v
0, the free-surface prole approaches the horizontal line y = 1 (i.e.
the horizontal line through the stagnation point E). This is consistent with
the fact that the dynamic boundary condition (3.100) predicts that the
curvature of the free surface tends to zero as
v
0 (the straight line y = 1
has zero curvature).
Numerical values of versus
v
are shown in Figure 3.26. As
v
varies
from 0 to , varies continuously from /2 to 0. For
v
= , the dynamic
boundary condition (3.100) reduces to the free streamline condition u
2
+v
2
=
1. The solution is then given by
= 0; a
0
= 1, a
n
= 0, n = 1, 2, . . . (3.125)
Subsituting (3.125) into (3.122) we obtain
w
= t
1/2
. (3.126)
This is the free streamline solution (3.39). Here the subscript refers to
the case
v
= .
1.0
1.1
1.2
1.3
1.4
1.5
1.6
1.7
0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Fig. 3.25. Computed free-surface proles for the ow conguration of Figure 3.24.
The proles from top to bottom correspond to
v
= ,
v
= 50,
v
= 25,
v
= 10
and
v
= 5;
v
is dened by (3.99) and (3.124).
0.4
0.6
0.8
1.0
1.2
1.4
1.6
0 5 10 15 20
Fig. 3.26. Values of the angle between the free surface and the wall at the sepa-
ration point E (see Figure 3.24) versus
v
.
As
v
0, the free surface approaches a horizontal straight line. The
solution is then
=

2
, a
0
= 2; a
n
= 0, n = 1, 2, . . . (3.127)
Substituting (3.127) into (3.122) we obtain the following exact solution:
w
0
=
2t
1/2
t + 1
. (3.128)
The subscript 0 refers to the case
v
= 0.
As in Section 3.1.1, we can dene the contraction ratio C
c
as the ratio of
the ordinate of the free surface as x and the ordinate of the separation
point E.
For
v
= , we have
C
c
=

+ 2
. (3.129)
(see (3.42)(3.44) for an explicit derivation).
For
v
= 0, the free surface is the horizontal line y = 1 and the ordi-
nates of E and of the level of the free surface as x are equal to 1.
Therefore
C
c0
= 1. (3.130)
In Figure 3.27 we present numerical values of C
c
versus
v
. As
v
de-
creases from innity, the contraction ratio increases monotonically from C
c
to C
c0
.
0.8
0.9
1.0
0 5 10 15 20
Fig. 3.27. Values of the contraction ratio C
c
versus
v
for the ow conguration of
Figure 3.24.
As
v
, the values of are described by the asymptotic formula
(3.117). Combining (3.117) and (3.124) we obtain

(2
v
)
1/2
as
v
. (3.131)
The numerical values in Figure 3.26 are in good agreement with (3.131) for
v
large. For
v
= 20, the value of predicted by (3.131) agrees with the
numerical results within two per cent.
We can also construct a perturbation solution for
v
small by writing
=
0
+
v
1
+O(
2
), (3.132)
=
0
+
v
1
+O(
2
). (3.133)
Here
0
and
0
are dened by
0
i
0
= ln w
0
, (3.134)
where w
0
is given by (3.128). Representing the free surface EF by (3.9), we
nd from (3.134) that
0
= ln cos

2
,
0
= 0, 0 < < , (3.135)
on the free surface EF. Substituting (3.132) and (3.133) into (3.100) and
equating coecients of
v
, we obtain
=
1
2
sin
2
(/2)
cos(/2)
. (3.136)
Using (3.10) and the chain rule we can rewrite (3.136) as
d
1
d
=
1
2
sin

2
. (3.137)
Integrating (3.137) and using the condition
1
= 0 at = 0, we obtain
1
=
1
_
cos

2
1
_
. (3.138)
In particular, (3.138) implies that
1
=
1
at = . (3.139)
Combining (3.117), (3.133) and (3.139) we obtain
=

2

v
2
as
v
0. (3.140)
For
v
= 1, the value of predicted by (3.140) agrees with the numerical
results within two per cent.
3.2.2 Free separation
The behaviour (3.103) occurs for all free streamline problems near the inter-
section of a free surface with a rigid wall. The constant C depends on the
geometry of the problem. When C < 0, as in the ow of Figure 3.24, the
asymptotic analysis of Section 3.2.1 for large shows that there is locally
a ow around a corner near the separation point E. The velocity at the
separation point E is then innite. An interesting question is whether there
are free streamline ows for which C 0. Of particular interest are ows for
which C = 0 and therefore for which the singular behaviour (3.103) disap-
pears. For C > 0, the asymptotic analysis of the previous section suggests
that, when surface tension is included, the ow near the separation point
is a ow inside an angle with a stagnation point at the separation. In this
section we will show that there are free streamline ows with C 0 when
the rigid boundaries are curved. The consideration of such ows will enable
us to introduce the concept of selection, which will be very useful when we
are studying gravitycapillary ows.
We consider the open-cavity model of Figure 3.16 but now with the eect
of the surface tension T included in the dynamic boundary condition. Pro-
ceeding as in Section 3.1.2 we seek i as an analytic function of + i
in the lower half-plane, < 0, of the domain shown in Figure 3.17. This
function must satisfy (3.50), (3.51) and
e
=

2
(e
2
1) on = 0, 1 < < . (3.141)
Here is dened by
=
U
2
R
T
. (3.142)
We start our investigation by reconsidering the solutions of Section 3.1.2
and by calculating the curvature K
A
of the free surface at the separation
point A. Using (3.49) and (3.62), we nd that
K
A
=
1
b

1
2
C(b b)
1/2
as 1, (3.143)
where
C = b
1/2
b
1/2
n=1
(1)
n+1
(2n 1)A
n
. (3.144)
Since the A
n
are functions of , (3.144) denes C as a function of . A
graph of C versus the angular position of the separation points is shown
in Figure 3.28. If the angle is counted positive when shown as in Figure
3.16, a comparison of Figures 3.24 and 3.16 shows that (3.117) implies that
=
C
2
_
_
1/2
. (3.145)
In other words has opposite signs in Figures 3.16 and 3.24. This
dierence in sign has been maintained to be consistent with previously pub-
lished results.
0
0.5
1.0
1.5
20 40 60 80 100 120 140
Fig. 3.28. Values of the constant C versus .
The constant C vanishes when =
55
(see Figure 3.28). Thus

(3.143) shows that the curvature of the free surface at the separation points
is innite unless =
.
For >
, Figure 3.28 shows that C > 0 and (3.145) predicts > 0. The
ow near B in Figure 3.16 is a ow inside an angle with zero velocity at B.
For <
, Figure 3.28 shows that C < 0 and the values of predicted by

(3.145) are then negative. The ow near B is then a ow around a corner
with innite velocity at B. These results are only valid for large. As
0, (3.141) shows that the curvature of the free surfaces tends to zero.
Since the ows are characterised by a constant velocity at innity, the free
surfaces must approach two horizontal straight lines. Therefore

2
as 0. (3.146)
Relation (3.146) shows that < 0 in the limit 0 when < /2.
Relation (3.145) shows that > 0 in the limit when >
. If
we assume that, for a given value of , is a continuous function of then
there must exist for each value of a particular value of
< < /2
for which = 0 (i.e. for which the ow leaves the circle tangentially). We
describe these particular values of by the function
= g(). (3.147)
This conjecture is conrmed by the nonlinear computations below. In
particular these results show that
g()
as . (3.148)
This implies that the limit T 0 can be used to select a particular solution
with T = 0. In Section 3.1.2 we calculated solutions for T = 0. Then the
dynamic boundary condition implies = 0. We obtained solutions for all
values of . When T ,= 0, solutions with = 0 exist only for values of
satisfying (3.147). Taking the limit , (3.148) shows that we should
select the solution corresponding to =
, which is known as the one

satisfying the BrillouinVillat condition (see [22], [194], [19] and [69]). This
condition was introduced to select the position of the separation points in
the case of free separation without surface tension. It requires the pressure
to be minimal in the cavity. By Bernoullis equation (3.1), this is equivalent
to the condition that the velocity is a maximum on the free streamlines.
For the conguration of Figure 3.16, the BrillouinVillat condition yields
=
.
The above analysis shows that the selection mechanism based on the
limit T 0 provides a new physical interpretation of the BrillouinVillat
condition.
We will now solve the problem numerically, calculate g() and demon-
strate (3.148). We rst map the ow of Figure 3.16 into the unit circle in
the t-plane by the transformation
t =
1 +if
1/2
1 if
1/2
. (3.149)
The ow conguration in the t-plane is shown in Figure 3.29.
A
C
D
E
Fig. 3.29. The ow conguration of Figure 3.16 in the t-plane dened by (3.149).
Next we note that the ow near A is locally a ow inside a corner with
angle and that the ow near C is a ow inside a right-angle corner
(see Figure 3.16). Therefore, using (3.149), we obtain
u iv (t i)
/
as t i, (3.150)
u iv t 1 as t 1. (3.151)
Following the series truncation method of Section 3.1.2, we represent the
complex velocity by
u iv = e
i
= (1 t)(1 +t
2
)
/
n=0
a
n
t
n
. (3.152)
The multiplicative factors in front of the series in (3.152) remove the singu-
larity (3.150). Therefore we can expect the series in (3.152) to converge in
the unit circle of the t-plane.
If we describe points on the unit circle by t = e
i
, 0 < < , we can
rewrite (3.141) and (3.51) as
b
cos
3
(/2)
sin(/2)
d
d
=

2
(e
2
1),

2
< < , (3.153)
e
b
cos
3
(/2)
sin(/2)
d
d
= 1, 0 < <

2
. (3.154)
Coecients a
n
in (3.152) are found such that (3.153) and (3.154) are sat-
ised. This is achieved by series truncation as in Section 3.1.2. We trun-
cate the innite series in (3.152) after N terms and nd the N coecients
a
0
, a
1
, . . . , a
N1
, the constant b and the angle by collocation. Thus we
introduce the N mesh points
I
=
_
I
1
2
_

N
, I = 1, . . . , N. (3.155)
In order to avoid the value = /2 at which the expression (3.152) is
singular, we choose N to be even. Using (3.152) and (3.155) we obtain
() and

() at the points
I
in terms of the coecients a
n
. Substituting
these expressions into (3.153) and (3.154) we obtain N equations. An extra
equation is obtained by requiring the velocity to be unity at innity. This
leads to
() = 0. (3.156)
The last equation relates and the angle at the separation point:
[
()]
=/2
=

2
+. (3.157)
This system of N + 1 equations with N + 1 unknowns can be solved by
Newtons method. We used this numerical scheme to compute solutions
for various values of and . The coecients a
n
were found to decrease
rapidly as n increases. For example, for = 30
and = 10, a
1
0.6,
a
10
0.310
2
and a
40
0.310
3
. In Figure 3.30 we present numerical
values of / versus for = 1.
0
0.10
0.05
60 70 80 90 100
Fig. 3.30. Values of / versus (in degrees) for = 1.
The curve in Figure 3.30 shows that for = 1 there is exactly one value
of at which = 0. Similar results were found for other values of (see
[176] and [183]). As 0, the free surfaces in gure 3.16 approach two
horizontal lines. Therefore the curve corresponding to = 0 in Figure 3.30
is the straight line (not shown in the gure) of equation
=

2
. (3.158)
For = , the angle is equal to zero for all values of and the curve
corresponding to = in Figure 3.30 is the horizontal line = 0 (not
shown).
These results imply that, for each value of ,= , there is a particular
value for which = 0 (i.e. for which the free surface leaves the obstacle
tangentially). We denote these particular values of by the function (3.147).
In Figure 3.31 we present computed values of = g() versus
1
. As
0, 90
. As ,
. Therefore the particular solution

that satises the BrillouinVillat condition in the absence of surface tension
can be viewed as the limit of the family of solutions in Figure 3.31 as the
surface tension approaches zero.
So far we have mainly considered solutions with = 0. It is of interest
to look at solutions with ,= 0. The angle can then be interpreted as
a contact angle whose value depends on the properties of the uid and of
the rigid boundary. In Figure 3.32 we present values of versus
1
for
= 0.04.
80
70
60
50
0 0.2 0.4 0.6 0.8 1.0
Fig. 3.31. Values of versus
1
. The corresponding free-surface proles leave the
circular object tangentially, i.e. = 0.
80
82
84
86
88
90
92
94
96
98
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Fig. 3.32. Values of versus
1
for = 0.04.
This curve can be viewed as the equivalent of the curve (3.147) but with
= 0.04 instead of = 0. One interesting property to note in Figure 3.32
is that now is not a monotonic function of
1
. However, only one value
of corresponds to each value of the surface tension (i.e.
1
). For
1
small, increases rapidly. This behaviour can be described by substituting
= 0.04 into (3.145) and noting that
C = 2(0.04)
_
_
1/2
. (3.159)
Together with Figure 3.28, this predicts that increases as
1
decreases,
for large.
3.3 The eects of gravity
In this section we study solutions for the ow conguration of Figure 3.1
with surface tension neglected but with gravity included in the dynamic
boundary condition. We assume that gravity acts in the direction dened
by the angle
1
(see Figure 3.33) and write the dynamic boundary condition
as
1
2
(u
2
+v
2
) +gy sin
1
gxcos
1
= B, (3.160)
where B is the Bernoulli constant. In this section we will look for solutions
without waves on EF. Solutions with waves on EF will be studied in
Chapter 8.
A B
C
D
E
F
g
1
y
x
Fig. 3.33. The ow conguration of Figure 3.1 with the eects of gravity included.
Two situations of particular interest are
1
= /2 and
1
= 0. When
1
= /2, gravity is acting in the negative y-direction. One interpretation
of the ow of Figure 3.33 is then the ow emerging from a container with
gravity included. Another is the ow under a sluice gate, a classical topic in
hydraulics (see [11], [54], Larock [93], Chung [30], Vanden-Broeck [181] and
Binder and Vanden-Broeck [15], [16]).
There are now two free surfaces CD and EF (see Figure 3.34). The
model consists of replacing the upper free surface CD by a rigid lid (see
Figure 3.35). The conguration of Figure 3.35 is that of Figure 3.33 with
1
= 0. An accurate numerical study of the complete free-surface ow of
Figure 3.34 is presented in Section 8.3.
A
C
D
E
F
B
2
g
Fig. 3.34. A free-surface ow under a sluice gate.
A B
D
E
F
2
g
C
Fig. 3.35. A model for the ow under a sluice gate, in which the free surface CD
has been replaced by a rigid lid.
When
1
= 0, gravity is acting in the positive x-direction. A realistic
view of the ow is obtained by rotating Figure 3.33 by 90
clockwise. If the
ow is then reected in the wall AB, the result corresponds to a jet falling
from a nozzle (see Figure 3.36).
In Figures 3.33 and 3.36 we have assumed that the free surface EF leaves
the wall DE tangentially. As we shall see, there are in addition solutions
for which EF does not leave the wall DE tangentially.
There is then an angle between the wall DE and the free surface EF at
the separation point E (see Figure 3.37). There are only three possible values
for . One of them is = ; it corresponds to the case already mentioned
where the free surface leaves the wall tangentially. The existence of these
three values of is to be contrasted with the problems including surface
tension discussed in Section 3.2, where all values of were in principle
possible, and with the free streamline problems in Section 3.1, where only
the value = was acceptable.
The existence of the two values of ,= can be established by deriving
a local solution valid in the neighbourhood of the separation point E. The
F
B
E
D
C
A
g
2
Fig. 3.36. The free-surface ow emerging from a nozzle. This is the ow congu-
ration of Figure 3.33 rotated by 90
and reected in the wall AB. Here

1
= 0 and
1
= 0.
A B
C
D
E
F
2
g
y
x
Fig. 3.37. The ow of Figure 3.33 with an angle between the free surface and the
wall at the separation point E.
analysis follows the work of Dagan and Tulin [40]. We dene new local
coordinates x and y with the origin at the separation point and such that
gravity is acting in the negative y-direction. The local ow is illustrated in
Figure 3.38. Here the separation point is denoted by G, the wall by HG and
the free surface by GL. It can easily be seen that the ow of Figure 3.38
describes the ow near E in Figure 3.37 if
2
=

2

1
+
2
. (3.161)
If = , the free surface LG leaves the wall HG tangentially and the
velocity at G is nite and dierent from zero. If < , the ow is locally
a ow inside a corner and the velocity at G is zero (i.e. G is a stagnation
point). If > , the ow is locally a ow around a corner and the velocity
at G is innite (see Figure 3.6 and (3.19)). Therefore values of > are
not possible, since (3.160) requires the velocity at G to be nite.
x
y
G
L
H
2
g
Fig. 3.38. Local-gravity free-surface ow near the intersection of a wall HG with a
free surface GL.
We shall now determine the allowed values of when < . We dene
a potential function and a streamfunction and choose = = 0 at the
point G. The complex potential plane is shown in Figure 3.39.
H G
Fig. 3.39. The ow of Figure 3.38 in the complex potential plane. The ow domain
is < 0.
Using (3.17) we express the local solution in the form
z = Af
/
. (3.162)
We write the complex constant A as
A = ae
i
, (3.163)
where a and are real and is such that the kinematic boundary condition
on the wall HG is satised. On the wall HG, arg z =
2
. However,
the argument of (3.162) evaluated on = 0, < 0, gives arg z = + .
Equating the two expressions for arg z gives
= +
2
. (3.164)
Substituting (3.163) and (3.164) into (3.162) yields
z = af
/
e
i(+
2
)
. (3.165)
Before satisfying the dynamic boundary condition we need to improve the
local solution (3.165) by writing explicitly the next-order correction:
z = af
/
e
i(+
2
)
+bf
e
i
+ (3.166)
Here , b and are real constants. The second term in (3.166) takes into
account the deviation of the free surface from the straight line GL as in-
creases. Since we require the second term in (3.166) to be a small correction
to (3.165) in the limit f 0, we impose
>

. (3.167)
Taking the real and imaginary parts of (3.166) on = 0, > 0, we obtain
x = a
/
cos( +
2
) +b
cos + , (3.168)
y = a
/
sin( +
2
) +b
sin + . (3.169)
Dierentiating (3.168) and (3.169) with respect to yields
x
2
+y
2
=
_
a
_
2
2/2
+ 2
a
b
/+2
cos( +
2
) +b
2
22
+ . (3.170)
Since
2
2 <

+ 2 < 2 2, (3.171)
the last term in (3.170) is of lower order and can be neglected. We can
rewrite (3.170) as
x
2
+y
2
=
_
a
_
2
2/2
_
1 +
2
a
b
/
cos( +
2
)
_
+
(3.172)
and express u
2
+v
2
in the limit f 0 as
u
2
+v
2
=
1
x
2
+y
2
=
_

a
_
2
22/
_
1
2
a
b
/
cos( +
2
+)
_
+ . (3.173)
We now substitute (3.169) and (3.173) into the dynamic boundary condi-
tion (3.160). First we note that
1
= /2 since g is acting in the negative
y-direction in Figure 3.38 and that B = 0 since u = v = 0 at the point G.
This gives
1
2
_

a
_
2
22/
+ga
/
sin( +
2
) +gb
sin + = 0. (3.174)
We then equate the coecients of the leading-order terms in (3.174). If
+
2
,= then the rst and second terms in (3.174) give
= 2
2
or =
2
3
(3.175)
and
a =
_
1
2
2
1
g sin(
2
)
_
1/3
. (3.176)
Since = 2/3, we require /3 <
2
< 2/3, for otherwise the free
surface in Figure 3.38 would descend towards the stagnation point G and
this would be in contradiction with the dynamic boundary condition (3.160)
with
1
= /2, which implies that a stagnation point on a free surface is the
highest point on it. In the remaining part of this chapter we will assume
2
> 0; however, solutions with
2
< 0 will be considered in Chapter 8 (see
Figure 8.5, where
3
=
2
). Furthermore, the solution with
2
= /6
will be used in Section 6.5.2 to describe the singularity near the crests of
the highest gravity waves. If
+
2
= or =
2
, (3.177)
the second term in (3.174) vanishes and the balance of the remaining terms
gives
= 2
2
. (3.178)
The conditions (3.167), (3.177) and (3.178) yield
=
2
<
2
3
. (3.179)
The condition (3.177) implies that the free surface LG is horizontal.
In summary we have the following possibilities. If
2
2/3, there are
three possible values for : , 2/3 and
2
. However, if
2
2/3 then the
only possible value for is .
We note that on the one hand the solution (3.162) with = 2/3 and
A dened by (3.163), (3.164) and (3.176) is an exact solution for the ow
conguration of Figure 3.38 for all values of z inside the angle HGL. On
the other hand the solution (3.162) with =
2
is only a local solution in
the limit z 0.
We shall solve the ow problem of Figure 3.37 numerically by using a series
truncation method similar to that used in Section 3.2. All the solutions
constructed in this section are waveless as x . Solutions with waves
as x will be computed in Chapter 8. The ow conguration in the
complex potential plane is shown in Figure 3.4. As in Section 3.1.1, we map
the complex potential plane onto the inside of the unit circle in the t-plane
by using the transformation (3.8). The ow conguration in the t-plane is
illustrated in Figure 3.5. Next we represent the complex velocity w = uiv
by the expansion (3.12) where G(t) contains all the singularities of w in
[t[ 1. In this case there are two singularities: one is at the separation
point E and the other at t = 1 (i.e. as x ).
The former corresponds to a ow inside an angle (see Figure 3.37) and
is described by
w f
1/
as f 0 (3.180)
(see (3.19)). Using (3.8) we have
w (1 +t)
22/
as t 1. (3.181)
The singularity at t = 1 depends on the value of
1
. It can be seen from
(3.160) that
u
2
+v
2
2gxcos
1
as x when
1
,= /2 (3.182)
and that
u
2
+v
2
constant as x when
1
= /2. (3.183)
In (3.183) we have used the fact that there are no waves as x .
We rst examine the case ,= /2 (see Figure 3.37). Relation (3.182)
shows that u
2
+v
2
as x . Since the ux between the free surface
EF and the wall AB is nite (and equal to Q), it follows by conservation
of mass that the free surface approaches the wall asymptotically as x .
In other words the ow reduces to an arbitrarily thin jet in the x-direction
as x . Therefore u v as x , and (3.182) implies that
u (2gxcos
1
)
1/2
. (3.184)
If we denote by y = (x) the equation of the free surface then from the
conservation of mass we have as x
u(x) = Q. (3.185)
Combining (3.184) and (3.185) we have
(x) Q(2gxcos )
1/2
. (3.186)
Writing successively
x
and
y
for u in (3.184) gives
x
= (2gxcos )
1/2
, (3.187)
y
= (2gxcos )
1/2
. (3.188)
Integrating (3.187) with respect to x and (3.188) with respect to y gives
expressions for and . Combining them gives
f = +i = (2g cos
1
)
1/2
_
2
3
x
3/2
+ix
1/2
y
_
. (3.189)
We note that, for x large,
z
3/2
= x
3/2
_
1 +
iy
x
_
3/2
x
3/2
_
1 +i
3y
2x
_
= x
3/2
+i
3x
1/2
y
2
. (3.190)
Combining (3.189) and (3.190) gives
f z
3/2
or z f
2/3
. (3.191)
Dierentiating (3.191) with respect to z and eliminating z by using the
second of the relations (3.191) yields
df
dz
= u iv z
1/2
f
1/3
. (3.192)
Next we examine the case
1
= /2 (see Figure 3.37). Then the ow
approaches a uniform stream with constant velocity U and constant depth
H as x (see (3.183)). The exact equations that describe the ow as
x are, in dimensional variables,
xx
+
yy
= 0, 0 < y < H +(x), (3.193)
y
=
x
x
on y = H +(x), (3.194)
1
2
(
2
x
+
2
y
) +gy =
1
2
U
2
+gH on y = H +(x), (3.195)
y
= 0 on y = 0. (3.196)
Here y = H + (x) is the equation of the free surface. Equations (3.194)
and (3.196) are the kinematic boundary conditions on the free surface and
on the bottom, and equation (3.195) is the dynamic boundary condition on
the free surface.
We write
w = U + w + , (3.197)
= Ux +

+ , (3.198)
= + , (3.199)
where w,

and are assumed to be small perturbations. We have assumed
as before that the ow approaches a uniform stream as x . Substituting
(3.197)(3.199) into (3.193)(3.196) and linearising yields
xx
+

yy
= 0, (3.200)
U
x
=

y
on y = H, (3.201)
U

x
+g = 0 on y = H, (3.202)
y
= 0 on y = 0. (3.203)
Eliminating between (3.201) and (3.202) gives
U

xx
+
g
U
y
= 0 on y = H. (3.204)
We use separation of variables, to nd a solution of (3.200) in the form
(x, y) = X(x)Y (y). (3.205)

Substituting (3.205) in (3.200) and in (3.203) yields the ordinary dierential
equations
X
(x)
X(x)
=
Y
(y)
Y (y)
=
2
(3.206)
and the boundary condition
Y
(0) = 0. (3.207)
Here
2
is the separation constant. The solutions of the two dierential
equations (3.206) satisfying (3.207) are
X(x) = Be
x
+Ce
x
, (3.208)
Y (y) = Dcos y, (3.209)
where B, C and D are constants. We set C = 0, so that

remains bounded
as x . Multiplying (3.208) and (3.209) yields the solution
= Ae
x
cos y, (3.210)
where A = DB is a constant. Substituting (3.210) into (3.204) yields
H =
1
F
2
tan H, (3.211)
where
F =
U
(gH)
1/2
(3.212)
is the Froude number. We note that the derivation leading to (3.210) is
similar to that leading to (2.74) and (2.75) in the theory of linear waves.
The main dierence is that we chose a negative separation constant in (2.72)
whereas we have chosen a positive one in (3.206).
Substituting (3.210) into (3.198) and dierentiating with respect to x gives
w =

x
i
y
= U A e
z
. (3.213)
Next we rewrite (3.213) in terms of t. Using (3.8) we obtain, as x
or equivalently as t 1,
e
f
(1 t)
2/
as t 1. (3.214)
Combining (3.213) and (3.214) and using dimensionless variables, i.e. setting
U = 1 and H = 1, we have
w 1 A (1 t)
2 /
as t 1. (3.215)
Similarly (3.211) gives in dimensionless variables
=
1
F
2
tan . (3.216)
Relations (3.215) and (3.216) dene the singular behaviour of w as t 1
for
1
= /2.
Finally we use (3.214) to rewrite (3.192) as
w [ln(1 t)]
1/3
as t 1. (3.217)
This demonstrates the singular behaviour of w as t 1 when ,= /2.
1
= 0 (funnels)
We rst consider the conguration of Figure 3.37 with
1
= 0. Following
the approach of Sections 3.1.1 and 3.2, we nd that the complex velocity w
can be represented by the expression
w =
_
t d
1 td
_
(
2
1
)/
t
1
/
[ln C
p
(1 t)]
1/3
(ln C
p
)
1/3
(1 +t)
22/

G(t), (3.218)
where t is dened by (3.8). Here C
p
is an arbitrary constant. The vari-
ous factors appearing in the numerator of the expression multiplying

G(t)
remove the singularities in w at t = 0 (see (3.33)), t = 1 (see (3.217)),
t = d (see (3.27)) and t = 1 (see (3.181)). The factors appearing in
the denominator of the expression multiplying

G(t) are not essential and
the computations could have been performed without them. The function
G(t) in (3.218) is then free of singularities and can be written as the Taylor
expansion
G(t) =
n=0
a
n
t
n
. (3.219)
There are of course alternative representations for

G(t). For example we
shall see in Section 3.3.2 that another convenient representation for

G(t) is
G(t) = exp
_

n=0
a
n
t
n
_
(3.220)
(see also (3.60)). We choose 0 < C
p
< 0.5 in (3.218), so that
[ln C
p
(1 t)]
1/3
is real for 1 < t < 1. Then it can be easily checked that the kinematic
boundary conditions on the walls CD, DE and AB are automatically satis-
ed by assuming that the coecients a
n
in (3.219) are real. For the compu-
tation presented we chose C
p
= 0.2. We note that dierent values of C
p
and
dierent choices for

G(t) (see (3.219) or (3.220)) will yield dierent values
for the coecients a
n
. However these various series representations yield
the same values of w provided that, all the singularities in the unit circle of
the complex t-plane have been properly removed.
We present explicit solutions for
1
=
2
,= 0 (the analysis follows Lee
and Vanden-Broeck [94]). After reection in the wall AB, this models a jet
of uid emerging from a funnel (see Figure 3.40).
A
B
F
E
C
2
y
x
Fig. 3.40. A free-surface ow emerging from a funnel. Gravity is acting vertically
downwards. This ow can be obtained by rotating the ow of Figure 3.33 by 90
and reecting it in the wall AB.

The expression (3.218) reduces to
w = t
2
/
[ln C
p
(1 t)]
1/3
(ln C
p
)
1/3
(1 +t)
22/

G(t), (3.221)
where

G(t) is dened by (3.219). The dynamic boundary condition (3.160)
with
1
= 0 gives
1
2
(u
2
+v
2
) gx = B. (3.222)
We dene dimensionless variables by using (Q
2
/g)
1/3
as the unit length and
(Qg)
1/3
as the unit velocity; here Q is the value of on the streamline CEF.
This scaling is dierent from that used in Sections 3.1.1 and 3.2 since the
velocity tends to innity as x instead of approaching a constant U. In
dimensionless variables (3.222) becomes
1
2
(u
2
+v
2
) x = B. (3.223)
The ow domain in the f = (+i)-plane is, as before, the strip 0 < < 1
(see Figure 3.4). Here the point D is irrelevant since
1
=
2
.
We will nd the coecients a
n
by truncating the series in (3.221) and
satisfying (3.223) at suitably chosen collocation points. Therefore we need to
express x in (3.223) in terms of w = uiv. This is achieved by dierentiating
(3.223) with respect to and using (3.10) and (3.38); is dened in (3.9).
Then the chain rule gives
uu
+vv
+
1
cotan

2
u
u
2
+v
2
= 0. (3.224)
The local analysis near E described at the beginning of this chapter shows
that there are three possible values of : , 2/3 and /2 +
2
. Numerical
experimentation shows that, for a given value of
1
=
2
, the solutions
corresponding to = and = /2 +
2
form a one-parameter family of
solutions whereas there is only one solution corresponding to = 2/3.
A convenient choice for the parameter is
H =
1
W
, (3.225)
where W is the dimensionless distance between the separation point E and
the wall AB. Therefore
H =
1
y()
, (3.226)
where y() denotes the value of y at = .
For = and = /2 +
2
, we truncate the innite series within (3.221)
after N terms and satisfy (3.224) at the N 1 collocation points
I
=

2(N 1)
_
I
1
2
_
, I = 1, 2, . . . , N 1. (3.227)
An extra equation is obtained by satisfying (3.226), where

H is given. The
value of y() in (3.226) is obtained by integrating numerically the identity
(3.38), where w is dened by (3.221). This gives a system of N algebraic
equations for the N unknowns a
0
, a
1
, . . . , a
N1
. This system is solved by
Newtons method.
For = 2/3, there is one fewer equation since now we do not have to
satisfy (3.226). Therefore again we truncate the innite series within (3.221)
after N terms but satisfy (3.224) at the N collocation points
I
=
1
2N
_
I
1
2
_
, I = 1, 2, . . . , N. (3.228)
As before, this gives a system of N nonlinear algebraic equations that can be
solved by Newtons method. We nd that there are solutions for all values
of

H > 0. Dierent behaviours are found for
2
> /6 and
2
< /6.
For
2
< /6, there is a solution with = 2/3 for a particular value, say
H
c
, of

H. Solutions with = and = /2 +
2
occur for

H >

H
c
and
H <

H
c
respectively. Therefore
= when

H >

H
c
, (3.229)
=
2
3
when

H =

H
c
, (3.230)
=

2
+
2
when

H <

H
c
. (3.231)
The solutions corresponding to (3.229) and (3.231) approach the solution
corresponding to (3.230) as

H approaches

H
c
from above and from below
respectively. The value of

H
c
depends on the value of
2
.
For
2
> /6, all solutions with

H > 0 are characterised by = . This
is consistent with the fact that the local analysis (see Figure 3.38) shows
that there are no solutions with = 2/3 or = /2+
2
when
2
> /6.
These results are illustrated in Figure 3.41, where we plot values of
the velocity q
E
at the separation point E versus

H for various values
of
2
.
The solutions with = 2/3 and = /2 +
2
have stagnation
point at E and are therefore characterised by q
E
= 0. For
2
/6,
q
E
,= 0 for all

H > 0 and q
E
0 as

H 0. For < /6, q
E
0 as
H

H
c
and q
E
= 0 for

H <

H
c
. Therefore the curves corresponding to
2
= /4 and
2
= /6 approach 0 as

H 0, whereas the curve corre-
sponding to
2
= /12 intersects the horizontal axis at the value

H =

H
c

0.528.
Typical free-surface proles are shown in Figures 3.423.45.
The three proles of Figures 3.423.44 are for
2
= /12. Figure 3.42
corresponds to

H = 0.81, i.e. the case (3.229). The free surface leaves the
wall tangentially. Both the free surface and the wall position can be seen
in the gure (the separation point E corresponds to the point on the curve
with ordinate y = 0).
Figure 3.43 shows the prole for

H =

H
c
0.528 (case (3.230)). There is
a 2/3 angle between the free surface and the wall at the separation point.
0
0.2
0.4
0.6
0.8
1.0
0 0.2 0.4 0.6 0.8 1.0
Fig. 3.41. Values of the dimensionless velocity q
E
at the separation point E versus
H (see (3.225)). The values of

2
corresponding to the curves from left to right are
/4, /6 and /12.
Figure 3.44 is the solution for

H = 0.25. Since

H <

H
c
, it corresponds to
the case (3.231). The free surface is horizontal at the separation point.
The prole in Figure 3.45 corresponds to
2
= /6 and q
E
= 0.5. The
free surface leaves the wall tangentially. This prole is typical in the sense
that all the solutions with
2
= /6 leave the wall tangentially (see Figure
3.41).
0
1
1 2 3 4 5 6
Fig. 3.42. A free-surface prole EF of Figure 3.40 for
2
= /12. The ordinate of
the separation point is zero. The dimensionless velocity q
E
at the separation point
is 0.5. The free surface leaves the wall tangentially and = .
0
1
1 2 3 4 5 6
2
the separation point is zero. This prole corresponds to = 2/3 and

H =

H
c

0.528.
0
1
0 1 2 3 4 5 6
2
= /12 and

H = 0.25.
The ordinate of the separation point is zero. The free surface is horizontal at the
separation point.
1
= 0 (nozzles and bubbles)
Next we present computations for
1
=
2
= 0 (see Figure 3.46). This
conguration diers from that of Figure 3.40 because the ow approaches a
uniform stream with constant velocity U as x instead of approaching
a zero velocity. This problem was considered by many investigators (see [18],
[62], [63], [165], [166], [169] and others). The study presented below follows
Vanden-Broeck [165], [166] and [169].
We will describe the problem by reverting to dimensionless variables in
which the unit length is H = Q/U and the unit velocity is U. Then the
0
1
1 2 3 4 5
2
the separation point is zero. The dimensionless velocity q
E
at the separation point
is 0.5. The free surface leaves the wall tangentially.
B
F
E
C
A
x
y
Fig. 3.46. The free-surface ow from a nozzle.
dynamic boundary condition (3.160) becomes, in dimensionless variables,
u
2
+v
2
1
F
2
x = B, (3.232)
where F is the Froude number dened by
F =
U
(2gH)
1/2
. (3.233)
We note that F is related to

H in (3.225) by
F =
_

H
3
2
_
1/2
. (3.234)
The factor 2
1/2
in (3.233) was introduced for consistency with previous cal-
culations. Similarly (3.224) becomes
2uu
+ 2vv
+
1
F
2
cotan

2
u
u
2
+v
2
= 0. (3.235)
The ow of Figure 3.46 has interesting applications. First, it clearly mod-
els a jet emerging from a nozzle. Second, it models a bubble rising in a tube
when viewed in a frame of reference moving with the bubble (see Figure
3.47). This follows from the symmetry of the ow: the portion EF of
the bubble surface in Figure 3.47 is identical, for the same value of the
Froude number F, to the portion EF of the jet surface in Figure 3.46.
A
C
B
F
x
y
E
Fig. 3.47. A bubble rising in a tube, viewed in a frame of reference moving with
the bubble. Physical bubbles are characterised by a continuous slope at the apex.
Returning to our calculation, we represent the complex velocity w by the
expansion
w =
[ln C
p
(1 t)]
1/3
(ln C
p
)
1/3
(1 +t)
22/

G(t). (3.236)
In this section we represent

G(t) by (3.220) instead of (3.219). Therefore we
write
G(t) = exp
_

n=1
a
n
t
n
_
, (3.237)
where we have set a
0
= 0 so that w = 1 at t = 0.
As for the solutions with
1
=
2
,= 0 of Section 3.3.1, there are solutions
with = , = 2/3 and = /2 +
2
= /2. The solution for =
2/3 corresponds to a critical value F
c
0.3578 of the Froude number F.
Solutions with = occur for F > F
c
, and those with = /2 +
2
= /2
occur for F < F
c
. Therefore we have
= when F > F
c
, (3.238)
=
2
3
when F = F
c
, (3.239)
=

2
when F < F
c
. (3.240)
As expected these solutions are the limit of those of Figure 3.40 as
2
0.
The three relations (3.229)(3.231) reduce to (3.238)(3.240) as
2
0, with
H
c
= (2F
2
c
)
1/3
. (3.241)
The computations were performed as follows. For the solutions with =
and = /2, we truncated the series (3.237) after N terms and satised
(3.235) at the N collocation points (3.228). For a given value of F, this
gives a system of N algebraic equations with N unknowns. This system is
solved by Newtons method. For = 2/3, F is one unknown. Therefore
we truncate the series (3.237) after N1 terms and satisfy (3.235) at the N
collocation points (3.228). This leads again to a system of N algebraic
equations with N unknowns. Typical free-surface proles are shown in
Figures 3.483.52.
0
0 0.2 0.4
Fig. 3.48. Rising bubble in a tube for F = 0.1.
The solutions that we have computed model the ow emerging from a
nozzle (see Figure 3.46) or a bubble rising in a tube (see Figure 3.47). On
0
0 0.2 0.4
Fig. 3.49. Rising bubble in a tube for F = 0.3.
0
0 0.2 0.4
Fig. 3.50. Rising bubble in a tube for F = F
c
. There is a 120
angle at the apex

of the bubble.
physical grounds we expect a bubble to be characterised by a continuous
slope at its apex. This implies that = /2. Therefore all the solutions for
0 < F < F
c
should model a rising bubble. However, experiments (see [32]
and [107]) showed that bubbles are only observed for a unique value,
F
e
0.25, (3.242)
of the Froude number. Clearly the value F
e
is in the interval 0 < F < F
c
for which we have computed bubbles but the question is to nd what is
special about the value F
e
. This is an example of a selection problem. We
have already encountered such a problem in Section 3.2.2. There we found
that cavitating ow past a circular cylinder could be calculated for all values
< <
when surface tension was neglected. We then showed that a

unique solution, for 55
, could be selected by solving the problem with

0
0 0.2 0.4
Fig. 3.51. Rising bubble in a tube for F = 0.4. A cusp is appearing at the apex.
0
0 0.2 0.4
Fig. 3.52. Rising bubble in a tube for F = 1. There is a pronounced cusp at the
apex of the free-surface prole.
T ,= 0 and then taking the limit as T 0. We shall show in Section 3.4
that a unique bubble can again be selected by introducing surface tension
and then taking the limit as T 0.
We conclude this section by mentioning that our ndings with T = 0 are
consistent with analytical results derived by Garabedian [62]. Garabedian
[62] proved that there are mathematical solutions describing bubbles with
a continous slope at the apex for all values of F smaller than a critical value
F
c
. He then used an energy argument to suggest that the only physically
signicant solution is the one for which F = F
c
. In addition he showed that
F
c
> 0.2363. However, our computations using series truncation showed that
F
c
0.36. This value is about 40 percent higher than the experimental value
(3.242). Furthermore the solution corresponding to F = F
c
does not have
a continuous slope at the apex since = 2/3 (see Figure 3.50). Therefore
Garabedians energy argument does not select the relevant solution.
1
= /2 (ow under a gate with gravity)
In this section we consider solutions for the ow conguration of Figure
3.37 with
1
= /2. The analysis follows [171] and [94]. We introduce
dimensionless variables by using U as the unit velocity and H as the unit
length. The Bernoulli equation (3.160) can be written as
1
2
(u
2
+v
2
) +
1
F
2
y = B, (3.243)
where
F =
U
(gH)
1/2
. (3.244)
A
B
C
E
F
x
y
Fig. 3.53. The ow under a gate showing, the angle between the free surface EF
and the wall CE.
We represent the complex velocity w by
w =
_
t d
1 td
_
(
2
1
)/
t
1
/
(t + 1)
2
22/
e
A(1t)
2 /
G(t), (3.245)
where A is a real constant to be found as part of the solution. The function
G(t) is free of singularities. The condition w = 1 at t = 1 implies that
G(1) = 0. Therefore we can write
G(t) = e
n=1
a
n
(t
n
1)
. (3.246)
Following the derivation leading to (3.224) we rewrite (3.243) as
uu
+vv

1
F
2
cotan

2
v
u
2
+v
2
= 0. (3.247)
We now present explicit computations for
1
=
2
= /2 (i.e. for the ow
conguration of Figure 3.53). As we shall see, these numerical computations
show that there are solutions corresponding to the three possible values ,
/2 and 2/3 of .
We rst calculate the solutions corresponding to = (i.e. solutions for
which the free surface leaves the wall tangentially). We truncate the innite
series in (3.245) after N2 terms and introduce the N1 collocation points
I
=

N 1
_
I
1
2
_
, I = 1, 2, . . . , N 1. (3.248)
We then use (3.245) to evaluate uiv at the mesh points (3.248) and obtain
N 1 algebraic equations by satisfying (3.247) at these points. One more
equation is given by (3.216). The nal equation is obtained by xing a
parameter characterising the ow. An obvious choice for this parameter is
the Froude number F. However, the computations reveal that there can be
several solutions corresponding to the same value of the Froude number; a
better choice for the parameter is the dimensionless velocity at the separation
point E. Therefore we set
q
E
= [(w)
t=1
[. (3.249)
This leads to a system of N+1 equations for the N+1 unknowns a
1
, a
2
, . . . ,
a
N2
, A, F and .
Typical proles are presented in Figures 3.54 and 3.55.
0
1
2
0 1 2 3
Fig. 3.54. Computed free-surface prole for the ow under a gate with F = 2. The
angle is equal to .
The values of the dimensionless velocity q
E
at the separation point E
versus the Froude number F are shown in Figure 3.56.
0
1
2
0 1 2 3
Fig. 3.55. Computed free-surface prole for the ow under a gate with F = 1.8.
The angle is equal to .
0
0.2
0.4
0.6
0.8
1.0
1.6 2.0 2.4 2.8 3.2
Fig. 3.56. Values of the dimensionless velocity q
E
at the separation E versus the
Froude number F.
As F , so q
E
1 and the solution reduces to the free streamline
solution (3.39). As q
E
decreases from 1, the Froude number F rst decreases
to a minimum value F
2
1.8 and then increases up to the value F
1
1.87.
The value F = F
1
corresponds to q
E
= 0. These results show that a unique
solution with = exists for all values of F > F
1
. For F
2
< F < F
1
,
two dierent solutions with = are possible. For F < F
2
, there are no
solutions with = . The coecients a
n
were found to decrease rapidly as
n increases. However, as q
E
approches zero, the rate of convergence of the
series deteriorates and larger and larger values of N are needed to obtain
accurate solutions. This is due to the fact that solutions with q
E
= 0 must
correspond to = /2 or = 2/3. Therefore they cannot be computed
by the expansion (3.245), (3.246) with = .
We now consider solutions with = 2/3. Numerical experimentation
shows that there is only one solution. This is the limit of the family of
solutions with = as q
E
0. To calculate it, we set = 2/3 in (3.245),
truncate the innite series in (3.246) after N3 terms and satisfy (3.247) at
the mesh points (3.248). This leads to N 1 equations for the N unknowns
a
1
, a
2
, . . . , a
N3
, A, F and . The last equation is (3.216). The resulting
numerical solution is shown in Figure 3.57.
0
1
2
3
0 1 2 3
Fig. 3.57. Computed free-surface prole for the ow under a gate with = 2/3.
The solution is unique and the corresponding value of the Froude number is F =
F
1
1.86.
Finally, we look for solutions with = /2. Numerical experimentation
shows that there is a one-parameter family of solutions (the parameter can
be chosen as the Froude number F). This family exists for F > F
1
. As
F F
1
, the solutions approach the solution corresponding to F = F
2
and q
E
= 0. A typical free-surface prole is shown in Figure 3.58. To
compute these solutions we set = /2 in (3.245), truncate the innite
series in (3.246) after N 2 terms and again satisfy (3.247) at the mesh
points (3.248). For a given value of F, this leads to N 1 equations for the
N unknowns a
1
, a
2
, . . . , a
N2
, A and . The last equation is given as before
by (3.216).
Although these solutions are mathematically interesting they are unstable
since in them the heavy uid is lying on top of the light one.
Values of the contraction ratio C
c
versus F for the solution branch with
= are shown in Figure 3.59.
Budden and Norbury [24] derived the following asymptotic solution for
C
c
:
C
c
=

+ 2

(4j + 2)
2
+
3
( + 2)
5
0.0007
2
+ (3.250)
0
1
2
3
0 1.0 2.0
Fig. 3.58. Computed free-surface prole for the ow under a gate with = /2.
The value of the Froude number is F 1.9.
0.4
0.5
0.6
1.6 2.0 2.4 2.8 3.2
c
versus the Froude number F.
Here j = 0.915 0965 and is dened by
=
F
4
4
_
1
2
+
2
F
2
_
3
(3.251)
where = q
E
. For F = 2, the value of the contraction ratio predicted by
(3.250) agrees with the numerical results within one per cent.
3.4 The combined eects of gravity and surface tension
When gravity is included in the dynamic boundary condition and surface
tension is neglected, only three values of the angle between the free surface
EF and the rigid wall DE are allowed for the ow conguration of Figure
3.37. When surface tension is included and gravity neglected, all values of
are in principle possible. Therefore we can expect interesting behaviours
to occur when both gravity and surface tension are taken into account,
especially in the limit T 0.
3.4.1 Rising bubbles in a tube
We present our main ndings for the ow of Figure 3.46. The analysis follows
Vanden-Broeck [166]. We recall that this conguration also models the ow
past a bubble in a tube (see Figure 3.47). As in Section 3.3.2 we assume
without loss of generality that there is an angle between the wall CE and
the free surface EF at the point E (see Figure 3.46). If the free surface
leaves the wall tangentially then = . If < then the ow near E is
locally a ow inside an angle and the speed at E is zero. If > then the
ow near E is locally a ow around a corner and the speed at E is innite.
Following the analysis in Section 3.3.2, we introduce dimensionless variables
by taking the constant velocity U at x = as the reference velocity and
the distance H between AB and CE as the reference length. The dynamic
boundary condition on the free surface EF can be written in dimensionless
variables as
1
2
q
2
1
2F
2
x +
2
K = B. (3.252)
Here q is the magnitude of the velocity variable, K the curvature of the
free surface EF, B the Bernoulli constant, F the Froude number dened in
(3.233) and the Weber number, dened by
=
2U
2
H
T
. (3.253)
As noted in Section 3.3.2 only values of are allowed when T = 0
because the dynamic boundary condition (3.160) requires the velocity q
E
at the point E in Figure 3.46 to be nite. However, when T ,= 0, values
of > are in principle possible because an innite value of q in (3.252)
can be balanced by an innite value of the curvature. Examples of such
ows with surface tension included and gravity neglected were covered in
Section 3.2. In this section we show examples of gravitycapillary ows
for which q
E
is innite. The physical relevance of such ows can of course
be questioned. However, we shall see that their consideration is crucial in
constructing systematically other physical solutions.
The ow conguration in the complex potential plane is shown in Figure
3.4. We map it into the complex t-plane by using (3.8). The complex
t-plane is illustrated in Figure 3.5. Proceeding as in Section 3.3.1, we write
the complex velocity w as
w =
[ln C
p
(1 t)]
1/3
(ln C
p
)
1/3
(1 +t)
22/
G(t), (3.254)
where
G(t) = e
n=1
a
n
t
n
. (3.255)
0
0.2
0.4
0.6
0.8
1.0
0 0.2 0.4 0.6 0.8 1.0 1.2
Fig. 3.60. Values of versus F for the ow conguration of Figure 3.47 when
= , i.e. T = 0.
0.2
0.4
0.6
0.8
1.0
0 0.2 0.4 0.6 0.8 1.0 1.2
Fig. 3.61. Values of versus F for the ow conguration of Figure 3.47 when
= 10.
Next we dierentiate (3.252) with respect to and use (3.10) and the
chain rule to obtain
2uu
+ 2vv
+
1
F
2
cotan

2
u
u
2
+v
2
+
4
_
tan

2
uv
vu
(u
2
+v
2
)
1/2
_
= 0.
(3.256)
0.8
0.9
1.0
1.1
0.10 0.15 0.20 0.25 0.30
Fig. 3.62. Enlargement of part of Figure 3.61 showing clearly oscillations around
1.0.
0.0
0 0.2 0.4
Fig. 3.63. The selected bubble for T = 0. The value of the Froude number is
F = F
0.23.
We truncate the innite series in (3.255) after N1 terms and satisfy (3.256)
at the mesh points
I
=
1
2N
_
I
1
2
_
, I = 1, . . . , N. (3.257)
This is achieved by substituting (3.254) and its derivative with respect to
into (3.256). It leads to N nonlinear equations for the N unknowns
a
1
, . . . , a
N1
and . This system is solved by Newtons method for given
values of F and .
We start the presentation of the numerical results by recalling the ndings
of Section 3.3.2 when T = 0 (i.e. = ). They are shown graphically in
Figure 3.60 where we plot the parameter
= 2

(3.258)
versus F. There is a unique solution with = 2/3 (i.e. = 2/3) for
F = F
c
0.36. Solutions with = /2 (i.e. = 1) and = (i.e. = 0)
occur for F < F
c
and F > F
c
respectively. The curve of Figure 3.60 is
discontinuous with a jump at F = F
c
. When surface tension is included
in the dynamic boundary condition, the discontinuity disappears and the
curve of Figure 3.60 is replaced by a continuous one. This is illustrated
in Figure 3.61, where we present values of versus F for = 10. An
interesting feature is that the curve of Figure 3.61 oscillates innitely often
around = 1. An enlargement of Figure 3.61 is shown in Figure 3.62. As F
decreases, the amplitude and wavelength of the oscillations decrease. These
results suggest that there is a countably innite set of values of F for which
= 1. We denote this set by
F
i
, i = 1, 2, 3, . . . , (3.259)
where
F
1
> F
2
> F
3
> .
We recall that physically relevant bubbles are identied as those for which
= 1. Therefore all bubbles with F < F
c
are physically relevant when
T = 0 whereas only those corresponding to the set (3.259) are physically
relevant when T ,= 0.
The numerical computations show that, for each given value of i,
F
i
F
as , (3.260)
where
F
0.23. (3.261)
This is illustrated in Figure 3.64, where we plot values of F
1
versus
1
. As
T 0 (i.e.
1
0), F
1
F
in agreement with (3.260).

Our ndings can be summarised as follows. When T = 0, there is a bubble
(with = /2) for each value of 0 < F < F
c
. When T ,= 0, there is a bubble
(with = /2) for a discrete set of values of F (see (3.259)). As T 0,
the discrete set reduces to a unique value F = F
of F. Therefore we have
succeeded in selecting a unique solution by including surface tension and
taking the limit as T 0. Moreover the selected value F
0.23 is close
to the experimental value F
e
0.25 (see [32] and [107]).
0
0.5
1.0
1.5
2.0
0 0.1 0.2 0.3
Fig. 3.64. Values of
1
versus F
1
. As
1
0, F
1
F
0.23.
3.4.2 Fingering in a Hele Shaw cell
Another classical example of the selection of solutions via surface tension
occurs in the study of ngering in a Hele Shaw cell, This problem can be mo-
tivated as follows. It is well known that an instability may occur in a porous
medium when a less viscous uid drives a more viscous uid (Saman and
Taylor [132]). To study this instability, experiments have been performed
in a Hele Shaw cell, a channel formed by two closely spaced parallel glass
plates; this provides a model of a two-dimensional ow through a porous
medium. It was found that the unstable interface develops a number of
ngers. After some time, one nger dominates and suppresses the growth
of the others and the ow reaches a steady state in which a single nger
propagates without change of shape. McLean and Saman [109] modelled
this nger by a two-dimensional potential ow with surface tension included
at the interface (see Figure 3.65). They denoted by U the velocity of the
nger, 2a the lateral width of the channel, b the transverse thickness, T
the surface tension and the viscosity (here we use a tilde to avoid con-
fusion with the angle used earlier in this section). In addition they de-
noted the ratio of the width of the nger and the width of the channel
by .
Taking a as the unit length and (1)U as the unit velocity, McLean and
Saman derived a nonlinear integro-dierential equation for the unknown
shape S of the free surface:
ln q(S) =
S
_
1
0
(S
)
S
(S
S)
dS
, (3.262)
A
B
y
x
q
Fig. 3.65. Model for a nger in a Hele Shaw cell. Only half the nger is shown.
qS
d
dS
_
qS
d
dS
_
q = cos , (3.263)
(0) = 0, q(0) = 1, (3.264)
(1) =
2
, q(1) = 0. (3.265)
Here
=

, (3.266)
q = (1 ) q, (3.267)
=
Tb
2
2
12Ua
2
(1 )
2
. (3.268)
The integral in (3.262) is of Cauchy principal-value form. The variables
and q in (3.266) and (3.267) are dened in terms of the dimensionless

complex velocity u i v by the relation
u i v = qe
i
. (3.269)
The ow conguration is illustrated in Figure 3.65. The x-axis is parallel
to the walls of the channel and is the axis of symmetry of the nger. The
points A and B correspond to S = 0 and S = 1 respectively.
After a solution for and q is obtained, the shape of the nger is given
by
x(S) +i y(S) =
1
_
1
S
e
i
Sq
dS. (3.270)
For = 0 (i.e. in the absence of surface tension), Saman and Taylor
[132] obtained the following exact solution:
q =
_
(1 S)/(1 )
2
(1 )
2
+S(2 1)
_
1/2
, (3.271)
= cos
1
q. (3.272)
The solution (3.271), (3.272) leaves the parameter undetermined. In other
words a solution can be found for each value of 0 < < 1. This nding
is not consistent with experiments, which show that for small values of the
surface tension there is only one nger, corresponding to 0.5. This is
again a selection problem that can be resolved by solving the problem with
surface tension and then taking the limit as the surface tension approaches
zero.
Early numerical calculations with ,= 0 were performed by McLean and
Saman [109], who identied one family of solutions. Romero [129] then
found two other families. As we shall see there is in fact a countably innite
set of families of solutions.
There is a strong analogy between the ngering problem and the bubble
problem of Figure 3.47. The procedure to nd the discrete set (3.259) al-
lowed the angle to be found as part of the solution. Therefore we shall use
a similar approach for the ngering problem. The analysis follows Vanden-
Broeck [161].
We dene a modied problem that has solutions for all values of and .
This modied problem is obtained by replacing (3.265) simply by
q(1) = 0. (3.273)
Therefore (1) becomes an unknown to be found as part of the solution.
We solve the modied problem dened by (3.262)(3.264) and (3.273) and
obtain solutions for all values of and . We will then obtain solutions of the
original problem by selecting among the solutions of the modied problem
those for which (1) = /2.
Following McLean and Saman [109] we introduce the change of variables
S
= 1
. (3.274)
Here is the smallest root of
1
2
cotan = . (3.275)
With (3.274), is twice dierentiable with respect to at both end points.
McLean and Saman [109] chose = 2 in (3.274). In order to solve the
modied problem we will choose = 4.
We introduce the N mesh points
I
=
I 1
N
, I = 1, . . . , N. (3.276)
We also dene the unknowns
I
= (1
I
), I = 1, . . . , N. (3.277)
We discretise the system (3.262)(3.264) and (3.273) by following the
procedure outlined by McLean and Saman [109]. Thus we obtain N 1
nonlinear algebraic equations for the N 1 unknowns
I
, I = 2, . . . , N. For
given values of and this system is solved by Newtons method.
In Figure 3.66 we present numerical values of (1) versus for = 0.273.
As approaches zero, (1) and the nger collapses on the negative
x-axis. As approaches unity, (1) oscillates innitely often around /2.
An enlargement of part of Figure 3.66 is shown in Figure 3.67 to illustrate
the oscillations clearly.
Figures 3.66 and 3.67 show that there is a countably innite set of values
of for which (1) = /2. We denote this set by
i
, i = 1, 2, 3, . . . (3.278)
where
1
>
2
>
3
.
1.56
1.60
1.64
1.68
1.72
0.5 0.6 0.7 0.8 0.9 1.0
Fig. 3.66. Values of (1) versus for = 0.273.
The solutions corresponding to the values (3.278) of are the solutions
of the original problem.
1.569
1.570
1.571
1.572
0.70 0.75 0.80 0.85 0.90
Fig. 3.67. Enlargement of part of Figure 3.66 showing clearly the oscillations around
/2.
Vanden-Broeck [161] showed numerically that, for a given value of i,
i

1
2
as 0. (3.279)
It can be seen from relation (3.279) that a unique solution corresponding
to = 1/2 is selected in the limit as the surface tension tends to zero. We
note that (3.279) is comparable with (3.260). This nding is illustrated in
Figure 3.68, where values of
1
,
2
and
3
versus are plotted. As 0,
the three curves approach = 1/2.
0.82
0.74
0.66
0.58
0.50
0 0.2 0.4 0.6 0.8 1.0 1.2
Fig. 3.68. Values of versus .
3.4.3 Further examples involving rising bubbles
We now return to potential ows and present three other examples of free-
surface ows for which a unique solution can be selected by taking the
limit T 0. These ows are similar to the ow past a bubble shown in
Figure 3.47. The rst two are given in Figures 3.69 and 3.70. They model
a two-dimensional bubble rising at a constant velocity U in an unbounded
uid when viewed in a frame of reference moving with the bubble. Both
congurations include a model for the wake behind the bubble. In that
sense they are improvements on the ow of Figure 2.2, in which the wake
was neglected. The free-surface ows of Figures 3.69 and 3.70 can be solved
by series truncation methods similar to those used earlier in this section.
Details can be found in [170] and [174] respectively. We summarise here the
main results.
D
x
A
y
U
E
S S'
J' J
E'
Fig. 3.69. A free streamline model for a rising bubble.
In Figure 3.69, a free streamline model (similar to the cavitating models
of Section 3.1.2) is used. This implies that the velocity is equal to U on
the boundaries SJ and S
of the wake. When surface tension is neglected

there is one solution for each value of the Froude number
F =
U
(gD)
1/2
, (3.280)
where D is dened in Figure 3.69. If we denote by the angle between the
x
A
y
U
E
L
S' S
J' J
Fig. 3.70. Joukovskiis model for a rising bubble.
symmetry line EA and the free surface AS then it is found that
=

2
when F < F
c
, (3.281)
=
2
3
when F = F
c
, (3.282)
= when F > F
c
, (3.283)
where
F
c
0.9. (3.284)
These ndings are very similar to those obtained in (3.238)(3.240) for
the ow of Figure 3.47. Introducing the surface tension T on the surface
SAS
of the bubble yields a discrete set of values of F for which = /2.

A unique solution for which F 0.51 is then obtained by taking the limit
T 0 (see [170] for details).
The ow of Figure 3.70 is similar to that of Figure 3.69 except that the
boundary of the wake is now approximated by two vertical lines, SJ and
S
. This ow can be characterised by the Froude number

F =
U
(gL)
1/2
, (3.285)
where L is dened in Figure 3.70. This is a crude model for the wake. It is,
however, of mathematical interest because Joukovskii (see [69]) found an ex-
act solution in the absence of surface tension. This solution is characterised
by F = (2)
1/2
. The results obtained for the ow of Figure 3.69 suggest
by analogy that there is a solution for each value of F and that Joukosvkiis
solution is just a member of this family of solutions. This was conrmed
by the numerical computations in [174], where it was shown that there is a
solution for the ow conguration of Figure 3.70 for 0 < F < satisfying
(3.281)(3.283), where is the angle between EA and AS in Figure 3.70
and where F
c
0.66. As before a unique solution with = /2 can be
selected by introducing surface tension and taking the limit T 0. Inter-
estingly, the numerical computations suggest that the selected solution is
Joukovskiis exact solution (see [174]).
The third example is a generalisation of the problem of a bubble rising
in a tube considered in Sections 3.3.2 and 3.4.1. The ow conguration is
shown in Figure 3.71(a). Gravity is acting vertically downwards. The angle
between the left-hand wall and the horizontal is denoted by , and the angle
between the negative x-axis and the tangent line to the free surface JS at
J is denoted by . When < /2, Figure 3.71(a) models a physical bubble
rising in an inclined tube.
1.5
1.0
0.5
0
(a)
0 0.5
S
I
U
J
y
x
H
J'
I'
1.5
1.0
0.5
0
(b)
0 0.5
S
I
U
J
y
x
J'
I'
Fig. 3.71. The ow domain and the coordinates. This is a computed prole for
= 7/12, F = 0.11 and = 10.
This problem was studied experimentally by Maneri [107] and has been stud-
ied theoretically in [36] and [95]. The ow conguration of Figure 3.71(a)
also describes a jet emerging from a nozzle and falling down along a wall.
In this case the ow is viewed as bounded on the left by an innite wall and
on the right by a semi-innite wall and a free surface (Figure 3.71(b)).
The ow can be characterised by the Froude number
F =
U
(gH)
1/2
(3.286)
and the Weber number
=
U
2
H
T
. (3.287)
Here is the density of the uid, U the velocity as x and H the width
of the tube.
When T = 0 (i.e. = ), the admissible values of depend on and
can be predicted by the local analysis of Section 3.3 (see Figure 3.38). When
0 < < 2/3, there is a critical value F
c
of the Froude number F such that
solutions with = 0, = /3 and = occur for F > F
c
, F = F
c
and F < F
c
respectively. Values of F
c
versus are shown in Figure 3.72.
However, for 2/3 there is no such critical value of F, and = 0
for all 0 < F < (see [95]).
0
0 30 60 90 120
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Fig. 3.72. Values of F
c
versus .
We note that these results include those of Section 3.3.2 as a particular
case, since the ow of Figure 3.71(a) is simply half that of Figure 3.47.
Figure 3.72 predicts F
c
0.506. Dividing this value by

2 to take into
account the dierent denitions of H in Figures 3.3.2 and 3.71(a) gives 0.36,
in agreement with the value obtained in Section 3.3.2.
We now consider in more detail the problem of a bubble rising in an
inclined tube (i.e. the ow of Figure 3.71(a)). The experimental data of
Manieri [107] showed that for each value of 0 < < /2, there is only one
value of F for which a bubble exists. This does not agree with the numerical
results, which predict a solution for each value of F and 0 < < /2.
This discrepancy can be removed by generalising the procedure described in
Section 3.4.1. Thus again we introduce surface tension and take the limit as
the surface tension tends to zero. Couet and Strumolo [36] chose for each
value of and the particular solution corresponding to the largest value
of F for which = /2. The selected results were found to be in good
agreement with experiment.
The results in [95] summarised above show that the only possible values
of when T = 0 are , /3 and 0. Therefore there are no solutions with
= /2 when T = 0 unless = /2. This implies that the criterion, in [36]
has to be used with T small but dierent from zero.
Here we follow [95] and use a dierent selection criterion, in which we take
the limit T 0 instead of keeping T ,= 0 as in [36]. The numerical results
in [95] show that for each value of 0 < /2 and there is a discrete set
of values
F
i
, i = 1, 2, 3, . . . , (3.288)
where
F
1
> F
2
> F
3
> , (3.289)
for which = . This nding reduces to (3.259) when = /2. The
numerical computations show that for a given value of i
F
i
F
as . (3.290)
The selected values of F
and the corresponding proles are found to be

in close agreement with the experimental data of Manieri [107] (see [95] for
details). A typical selected prole for = /3 is shown in Figure 3.73.
3.4.4 Exponential asymptotics
Since the ngering problem of Section 3.4.2 has an exact solution when
T ,= 0 (see (3.271), (3.272)), it is tempting to try to construct asymptotic
solutions for T small in the form of a power series in T. This was achieved
by McLean and Saman [109], who showed that an arbitrary number of
terms can be calculated. However, they found that this expansion leads
to solutions for each value of 0 < < 1. This is to be contrasted with
the discrete set of solutions (3.278) found by direct numerical computation.
Fig. 3.73. Computed solution for = /3 and F = F
0.527. The dots are the

experimental values of Manieri [107].
This paradox was resolved by noting that the selection mechanism leading
to (3.278) is associated with exponentially small terms in T. Such terms are
smaller than any positive integer power of T as T 0 and therefore cannot
be calculated by a power series in T. Exponential asymptotics has been
used by many investigators to study this problem analytically (see Saman
[131] for a review and references).
Vanden-Broeck showed that the selection mechanism for a rising bubble
that leads to (3.259) (see Section 3.4.1) is also associated with exponentially
small terms.
Over the last 30 years, it has been found that exponentially small terms
play a surprisingly signicant part in free-surface ow problems. We will
encounter further examples later in this book. One example concerns the
eect of surface tension on solitary waves (see Chapter 6). Other examples
are the free-surface ows generated by moving disturbances for small values
of the Froude number or small values of the surface tension (see Chapter 8).
4
Linear free-surface ows generated
by moving disturbances
In this chapter we consider free-surface ows generated by a disturbance
moving at a constant velocity U. We choose a frame of reference moving
with the disturbance and assume that the ow is steady in that frame.
We will restrict our attention to two-dimensional linear ows. Nonlin-
ear, three-dimensional and unsteady free-surface ows will be discussed in
Chapters 7, 10 and 11 respectively. The disturbance could be a submerged
object, say a submarine (see Figure 4.1), an obstacle at the bottom of a
channel (see Figure 4.2), a surface-piercing object, say a ship (see Figure
4.3), or a distribution of pressure (see Figure 4.4).
Fig. 4.1. The two-dimensional free-surface ow generated by a submerged object
moving at a constant velocity U, when viewed in a frame of reference moving with
the object.
Fig. 4.2. The two-dimensional free-surface ow generated by an object moving at
a constant velocity U along the bottom of a channel, when viewed in a frame of
reference moving with the object.
114
4.1 The exact nonlinear equations 115
Fig. 4.3. The two-dimensional free-surface ow generated by a surface-piercing ob-
ject moving at a constant velocity U, when viewed in a frame of reference moving
with the object.
y
x
Fig. 4.4. The two-dimensional free-surface ow generated by a pressure distribution
moving at constant velocity U, when viewed in a frame of reference moving with
the pressure distribution.
4.1 The exact nonlinear equations
We introduce cartesian coordinates x, y with the x-axis parallel to the bot-
tom (see Figure 4.4). The fully nonlinear problem can be formulated in
terms of a potential function (x, y) as
xx
+
yy
= 0, h < y < (x), (4.1)
y
=
x
x
on y = (x), (4.2)
1
2
(
2
x
+
2
y
) +gy
T
xx
(1 +
2
x
)
3/2
+
P(x)
= B on y = (x), (4.3)
y
= 0 on y = h, (4.4)
n
= 0 on the wetted surfaces of the objects (if any). (4.5)
Here y = (x) is the equation of the free surface and
n
denotes the normal
derivative of . The bottom of the channel is at y = h. Equations (4.2),
(4.4) and (4.5) are the kinematic boundary conditions on the free surface, on
the bottom and on the surface of any object present, respectively. Equation
(4.3) is the dynamic boundary condition on the free surface. The quantity
P(x) is the prescribed pressure distribution; the parameter measures its
size. We x the Bernoulli constant B in (4.3) by choosing y = 0 as the
116 Linear free-surface ows generated by moving disturbances
constant level of the free surface corresponding to P = 0 and to a uniform
stream with constant velocity U. This leads to
B =
1
2
U
2
. (4.6)
If the object is fully submerged (i.e. if its surface does not intersect the
free surface y = (x)) and does not touch the bottom y = h, we need
to specify, in addition to equations (4.1)(4.5), the circulation of the vector
velocity around the object.
4.2 Linear theory
4.2.1 Solutions in water of nite depth
A rst insight into the problem will be gained by developing a linear theory.
The analysis presented in this chapter follows Rayleigh [128]. Nonlinear
investigations are presented in Chapter 7. The results are qualitatively
independent of the nature of the disturbance; therefore we will restrict our
attention to the case where the disturbance is a pressure distribution (see
Figure 4.4). The nonlinear problem is described by (4.1)(4.4). To derive
a linear theory we assume that the parameter in (4.3) is small and seek a
solution in the form of a small perturbation around a uniform stream with
constant velocity U and constant depth h. This is motivated by the fact
that (x, y) = Ux, (x) = 0 is an exact solution of (4.1)(4.4) when = 0.
The linear equations are obtained by expanding the solutions in powers of
and retaining only the terms of order . Therefore we write
(x, y) = Ux +
1
(x, y) +O(
2
), (4.7)
(x) =
1
(x) +O(
2
). (4.8)
Substituting (4.7) and (4.8) into (4.1)(4.4), using (4.6) and equating terms
of order yields
1xx
+
1yy
= 0, h < y < 0, (4.9)
1y
= U
1x
on y = 0, (4.10)
U
1x
+g
1
1xx
+
P
= 0 on y = 0, (4.11)
1y
= 0 on y = h. (4.12)
When P = 0 the system (4.9)(4.12) reduces to the system (2.60)(2.63)
with c = U. Therefore if
1
=
1p
,
1
=
1p
is a particular solution of
(4.9)(4.12) with P ,= 0, the general solution of (4.9)(4.12) with P ,= 0 is
obtained by adding to the particular solution
1p
,
1p
an arbitrary multiple
of a solution of (2.60)(2.63). When the condition (2.85) is satised, (2.86)
and (2.87) give
1
=
1p
UA
1
sinh kh
cosh k(y +h) sin k(x +), (4.13)
1
=
1p
+A
1
cos k(x +) (4.14)
where A
1
is an arbitrary amplitude and an arbitrary phase. The constant k
in (4.13) and (4.14) satises the equation (2.83) with c = U. Relations (4.13)
and (4.14) mean that we can add a train of periodic waves of wavenumber k
and of arbitrary amplitude A
1
to any particular solution of (4.9)(4.12) and
obtain a new solution. Cases for which the condition (2.85) is not satised
will be considered in Section 7.3.
Rayleigh [128] showed that a unique solution of the linear problem (4.9)
(4.12) can be obtained by using the radiation condition. This condition
requires that there is no energy coming from innity. In other words, if
there is a train of waves at x = in Figure 4.4 then the energy in this wave
train must be travelling to the right. Similarly, if there is a wave train at
x = then the energy must be travelling to the left. The discussion at
the end of Section 2.4.3 shows that the energy of a wave train travels at the
group velocity c
g
. In a frame of reference in which the disturbance moves
to the left at a constant velocity U, the waves and the energy in the waves
move also to the left at velocities c = U and c
g
respectively. In the frame of
reference moving with the disturbance, the waves are not travelling and the
energy travels to the left at velocity c
g
U. Therefore we require c
g
> U
for a wave train at x = and c
g
< U for a wave train at x = . We
note that for < 1/3 and U > c
min
there are exactly two possible wave
trains (see Figure 2.5). The wave train with k > k
min
is characterised by
c < c
g
and can therefore appear at x = . The wave train with k < k
min
is characterised by c > c
g
and can appear at x = .
Following Rayleigh [128] we shall see that the radiation condition is auto-
matically satised by introducing some dissipation in the equations (4.9)
(4.12), solving the problem and then taking the limit as the coecient of
dissipation tends to zero. One way to do this is to introduce the so-called
Rayleigh viscosity > 0 and to rewrite (4.11) as
U
1x
+g
1
1xx
+
P
+
1
= 0 on y = 0. (4.15)
We now solve the system (4.9), (4.10), (4.15) and (4.12) and then take
the limit 0. Eliminating
1
between (4.10) and (4.15) gives
U
1xx
+
g
U
1y

T
U
1yxx
+
P
x
+
1x
= 0 on y = 0. (4.16)
We represent
1
(x, y) by the Fourier integral
1
(x, y) =
_

F(a, y)e
iax
da. (4.17)
Substituting (4.17) into (4.9) and (4.12) gives
2
F
y
2
a
2
F = 0, h < y < 0, (4.18)
F
y
= 0 on y = h. (4.19)
The solution of (4.18) satisfying (4.19) is
F(a, y) = A(a) cosh a(y +h), (4.20)
and (4.17) can be rewritten as
1
(x, y) =
_

A(a) cosh a(y +h)e

iax
da. (4.21)
Here A(a) is an arbitrary function of a. Next we write the pressure distri-
bution as the Fourier integral
P(x) =
_

B(a)e
iax
da. (4.22)
The inverse Fourier transform gives
B(a) =
1
2
_

P(x)e
iax
dx. (4.23)
Substituting (4.21) and (4.22) into (4.16) gives
A(a) =
iB(a)
UD(a) cosh ah
, (4.24)
where
D(a) = a
g
U
2
tanh ah
Ta
2
U
2
tanh ah i
1
. (4.25)
Here
1
= /U.
Combining (4.10), (4.21) and (4.25) yields after integration
1
(x) =
1
U
2
_

e
iax
B(a) tanh ah
D(a)
da. (4.26)
To simplify the presentation we assume that P(x) is an even function (i.e.
P(x) = P(x)). It follows from (4.23) that
B(a) =
1
_

0
P(x) cos axdx. (4.27)
Therefore B(a) is a real even function of a, and the integrand in (4.26) can
be written as the ratio of a complex numerator and a complex denominator:
R
N
+iI
N
R
D
+iI
D
=
R
N
R
D
+I
N
I
D
R
2
D
+I
2
D
+i
R
D
I
N
R
N
I
D
R
2
D
+I
2
D
; (4.28)
here R
N
and R
D
are real odd functions of a and I
N
and I
D
are real even
functions of a. The real part of (4.28) is an even function of a and the
imaginary part of (4.28) is an odd function of a.
Therefore (4.26) can be written as
1
(x) =
2
U
2
'
_

0
e
iax
B(a) tanh ah
D(a)
da, (4.29)
where ' denotes the real part.
We now take the limit of (4.29) as
1
tends to zero.
If 'D(a) ,= 0 for 0 < a < , we obtain
1
(x) =
2
U
2
'
_

0
e
iax
B(a) tanh ah
a (g/U
2
) tanh ah (Ta
2
/U
2
) tanh ah
da. (4.30)
If 'D(a) vanishes for some values of a, some care has to be taken in taking
the limit. This occurs when there are values a
such that
a
g
U
2
tanh a
h
Ta
2
U
2
tanh a
h = 0. (4.31)
Comparing (4.31) with (2.83) we see that a
= k, where k is a wavenumber
satisfying the dispersion relation (2.83) with c = U. If 1/3, there is a
minimum value c
min
of c (see Figure 2.5). On the one hand there are then
two values a
1
and a
2
of a
for U > c
min
; without loss of generality we can
assume that 0 < a
1
< a
2
. On the other hand there are no values a
when
U < c
min
, and the solution is given by (4.30).
Let us consider the case < 1/3. We rewrite D(a) in the form
D(a) =
a
U
2
[U
2
C
2
(a)] i
1
, (4.32)
where
C
2
(a) =
_
g
a
+
T
a
_
tanh ah. (4.33)
If a
is a root of 'D(a), we seek a root of D(a) for

1
0 by writing
a = a
+ (a
), (4.34)
where [ (a
)[ is small. Subsituting (4.34) into (4.32) and expanding in a

Taylor expansion for (a
) small gives
D(a
+ (a
)) (a
)
2a
U
2
C(a
)C
(a
) i
1
, (4.35)
where the prime denotes the derivative with respect to a. Therefore
(a
) =
i
1
U
2
2a
C(a
)C
(a
)
(4.36)
to leading order. Since C(a
) = U, we can rewrite (4.36) as

(a
) = i
0
(a
), (4.37)
where
0
(a
) =

1
U
2a
(a
)
. (4.38)
We now evaluate the integral in (4.29) by contour integration in the complex
plane = a +ib. The quantity C(a) is the phase velocity of linear gravity
capillary waves of wavenumber a (see (4.33)). Relation (2.146) then shows
that the group velocity C
g
(a) is greater than C(a) when C
(a) > 0 and

smaller than C(a) when C
(a) < 0. The rst possibility occurs for a = a
2
and the second for a = a
1
(since we have assumed that 0 < a
1
< a
2
).
Relation (4.38) shows that
0
(a
) > 0 when C
(a
) < 0 and
0
(a
) < 0 when
C
(a
) > 0. Therefore the pole a
1
+i
0
(a
1
) is in the upper half of the complex
-plane whereas the pole a
2
+ i
0
(a
2
) is in the lower half. Similarly, it can
be shown that there is also a pole at a
1
+ i
0
(a
1
) in the upper half-plane
and a pole at a
2
+i
0
(a
2
) in the lower half-plane.
We now consider the integral
2
U
2
_
C
c
e
ix
B() tanh h
D()
d (4.39)
where = a +ib. The denition of the contour ( depends on the sign of x.
For x > 0, the contour ( is anticlockwise and consists of the a-axis and a
semicircle of radius R in the upper half-plane.
For x < 0, the contour ( is clockwise and consists of the a-axis and a
semicircle of radius R in the lower half-plane. We will refer to the contours
( for x > 0 and x < 0 as (
+
and (
respectively. These contours were

chosen so that the contributions from the semicircles will tend to zero as
R , which follows from the fact that the exponential
e
ix
= e
iax
e
bx
(4.40)
appearing in the integrand of (4.39) tends to zero as b (b )
when x > 0 (x < 0).
Therefore (4.39) reduces to
2
U
2
_
C
e
ix
B() tanh h
D()
d =
2
U
2
_

e
iax
B(a) tanh ah
D(a)
da. (4.41)
We now evaluate the integral in (4.39) by using the residue theorem. This
gives
2
U
2
_
C
+
e
ix
B() tanh h
D()
d = 2iR
+
I
+ 2i[residue at a
1
+i
0
(a
1
)]
+ 2i[residue at a
1
+i
0
(a
1
)] (4.42)
for x > 0 and
2
U
2
_
C
e
ix
B() tanh h
D()
d = 2iR
I
2i[residue at a
2
+i
0
(a
2
)]
2i[residue at a
2
+i
0
(a
2
)] (4.43)
for x < 0. Here R
+
I
and R
I
denote the sums of the residues on the positive
and negative parts of the imaginary -axis.
Noting that the real part of the rst integral on the right-hand side of
(4.41) is 2
1
(x) (see (4.29)), combining (4.41) with (4.42) and (4.43) and
taking the limit
1
0, we obtain
1
(x) = '(iR
+
I
) +
2
U
B(a
1
) tanh a
1
h
a
1
C
(a
1
)
sin a
1
x for x > 0 (4.44)
and
1
(x) = '(iR
I
)
2
U
B(a
2
) tanh a
2
h
a
2
C
(a
2
)
sin a
2
x for x < 0. (4.45)
Taking the limit [x[ we nd
1
(x)
2
U
B(a
1
) tanh a
1
h
a
1
C
(a
1
)
sin a
1
x as x (4.46)
and
1
(x)
2
U
B(a
2
) tanh a
2
h
a
2
C
(a
2
)
sin a
2
x as x . (4.47)
Relations (4.46) and (4.47) show that the free-surface prole is characterised
by a periodic wave train in the far eld. These waves have wavelengths 2/a
1
and 2/a
2
respectively. Since a
1
< a
2
, the train of longer wavelength occurs
as x and the train of shorter wavelength as x .
In conclusion, the linear free-surface proles are given by (4.30) for U <
c
min
and by (4.44) and (4.45) for U > c
min
.
4.2.2 Solutions in water of innite depth
We now examine the particular case of water of innite depth. The exact
relations are given by (4.1)(4.3), (4.6) and
x
U,
y
0 as y . (4.48)
Proceeding as in Section 4.2.1, we assume (4.7) and (4.8) with small and
derive the linearised system
1xx
+
1yy
= 0, y < 0, (4.49)
1y
= U
1x
on y = 0, (4.50)
U
1x
+g
1
1xx
+
P
+
1
= 0 on y = 0, (4.51)
1x
0
1y
0 as y . (4.52)
Eliminating
1
(x) between (4.50) and (4.51) yields (4.16). The remaining
analysis follows closely that of Section 4.2.1. We nd
1
(x, y) =
_

A(a)e
|a|y
e
iax
da (4.53)
with
A(a) =
iB(a)
U

D(a)
, (4.54)
where B(a) is given by (4.27) and
D(a) = a
g
U
2
[a[
a

Ta
2
U
2
[a[
a
i
1
. (4.55)
The expression for the prole
1
(x) is
1
(x) =
2
U
2
_

0
e
iax
B(a)
D(a)
da. (4.56)
Since only positive values of a are involved in (4.56) we can now rewrite
D(a) as
D(a) = a
g
U
2

Ta
2
U
2
i
1
. (4.57)
We now take the limit
1
0. If '
D(a) ,= 0 for 0 < a < , we obtain
1
(x) =
2
U
2
'
_

0
e
iax
B(a)
a (g/U
2
) (Ta
2
/U
2
)
da. (4.58)
The expression (4.58) is the equivalent of (4.30) in water of innite depth.
As expected, it can be formally derived from (4.30) by replacing the tanh
factors by 1.
If '
D(a) vanishes for some values of a then some care is needed in taking
the limit
1
0, and we follow the approach of Section 4.2.1. This situation
occurs when there are values a
such that
a
g
U
2

Ta
2
U
2
= 0. (4.59)
Comparing (4.59) and (2.95), we see that a
= k, where k is a wavenumber
satisfying the dispersion relation (2.95) with c = U. Referring to Figure 2.6,
we see that there are two values a
1
< a
2
of a
for U > c
min
, where c
min
is
given by (2.98). These values are given explicitly by solving the quadratic
equation (4.59); they are
a
1
=
U
2
2T
_
1
_
1
4Tg
U
4
_
1/2
_
, (4.60)
a
2
=
U
2
2T
_
1 +
_
1
4Tg
U
4
_
1/2
_
. (4.61)
However, there are no values a
when U < c
min
; then the solution is given
by (4.58).
When U > c
min
, we rewrite

D(a) in the form
D(a) =
a
U
2
[U
2
C
2
(a)] i
1
, (4.62)
where
C
2
(a) =
g
a
+
T
a. (4.63)
The remaining part of the calculations follows step by step that of Section
4.2.1, with the tanh factors replaced by 1. This leads to
1
(x) = '(iR
+
I
) +
2
U
B(a
1
)
a
1
C
(a
1
)
sin a
1
x, x > 0, (4.64)
and
1
(x) = '(iR
I
)
2
U
B(a
2
)
a
2
C
(a
2
)
sin a
2
x, x < 0. (4.65)
Taking the limit [x[ we nd
1
(x)
2
U
B(a
1
)
a
1
C
(a
1
)
sin a
1
x as x (4.66)
and
1
(x)
2
U
B(a
2
)
a
2
C
(a
2
)
sin a
2
x as x . (4.67)
As in water of nite depth, (4.66) and (4.67) and the inequality a
1
< a
2
imply
that the solutions are characterised by a wave train of long wavelength as
x and a wave train of short wavelength as x . For U values
not close to c
min
, the long waves are dominated by gravity and the short
waves are dominated by surface tension.
The solution in the absence of surface tension is obtained by taking the
limit T 0. Relations (4.60) and (4.61) show that
a
1
g/U
2
and a
2
as T 0. (4.68)
It follows that the right-hand side of (4.67) tends to zero. Therefore the free
surface is at as x and is characterised by wave trains of wavenumber
g/U
2
as x .
4.2.3 Discussion of the solutions
We illustrate these results rst in the particular case of water of innite
depth; then c
min
is given by (2.98). Following Vanden-Broeck [184] and
others, we dene the parameter
=
Tg
U
4
. (4.69)
Then U < c
min
and U > c
min
correspond respectively to > 0.25 and
< 0.25.
We present in Figures 4.5 and 4.6 typical free-surface proles predicted
by (4.58) for (see (4.7), (4.8)) equal to 0.001 and equal to 0.27. In
these calculations we chose
P(x) =
U
2
2
e
5g
2
x
2
/U
4
. (4.70)
Since > 0.25, the free surface is at in the far eld (i.e. there are no
wave trains in the far eld). Since the pressure distribution is symmetric
about x = 0, the free-surface proles are also symmetric about x = 0. They
are characterised by oscillations of decaying amplitude as [x[ . This is
consistent with the fact that (4.59) does not have real roots a
for > 1/4.

In fact solving (4.59) for a
yields
a
=
U
2
2T
[1 (1 4)
1/2
], (4.71)
and a
is complex when > 0.25.

0
0.001
0.002
0 10 20
Fig. 4.5. Linear free-surface prole for the pressure distribution (4.70) with =
0.001. The value of is 0.27.
In Figure 4.7 we present a typical prole calculated from (4.64) and (4.65)
with = 0.001 and = 0.222. The pressure distribution is dened by (4.70).
Since < 1/4, there are wave trains in the far eld, in accordance with (4.66)
and (4.67).
We now examine some limitations and nonuniformities of the linear theory.
Those limitations will be resolved in Chapter 7 where we develop a nonlinear
theory.
First, the linear result (4.58) predicts unbounded displacements of the
free surface as 1/4
+
. This is shown analytically by noting that when
0
0.001
0 10 20
0
0.001
0.002
0 5 10 15
= 0.25, the denominator of the integrand in (4.58) can be written as
T
U
2
_
a
U
2
2T
_
2
(4.72)
and so the integral in (4.58) diverges as 0.25
+
, because of the double
zero in (4.72). Similarly
a
1

U
2
2T
and a
2

U
2
2T
as 0.25
. (4.73)
Furthermore,
C
_
U
2
2T
_
= 0. (4.74)
Therefore (4.66) and (4.67) predict waves of unbounded amplitude as
0.25
.
This behaviour is shown graphically in Figure 4.8, where we present values
of the elevation of the free surface at x = 0, i.e.
1
(0) =
2
U
2
'
_

0
B(a)
a (g/U
2
) (Ta
2
/U
2
)
da, (4.75)
versus for > 1/4. Here P(x) is dened by (4.70) and = 0.001. As
1/4
+
,
1
(0) tends to innity in accordance with (4.72).
0
0.004
0.008
0.25 0.26 0.27 0.28 0.29 0.30
Fig. 4.8. Values of the amplitude of the free-surface prole versus for the pressure
distribution (4.70) with = 0.001 (upper curve) and = 0.001 (lower curve). The
unit length in the vertical scale is U
2
/g.
These results show that, for a given value of (however small),
1
(x)
and
1
can be made arbitrarily large by taking suciently close to 1/4.
This is in contradiction with the assumptions leading to (4.7) and (4.8). In
other words, the expansions (4.7) and (4.8) are nonuniform as 1/4. As
we shall see in Chapter 7, appropriate solutions for close to 1/4 can be
obtained by using a nonlinear theory. We note that the nonuniformity of
the linear theory in water of innite depth is associated with the existence
of a minimum phase velocity (see (2.98)). Similar nonuniformities occur in
water of nite depth and are associated with the minimum occurring in the
phase velocity when < 1/3 (see Figure 2.5).
Second, a similar nonuniformity occurs for pure gravity ows (g ,= 0,
T = 0) in nite depth as F 1. This can be shown by setting T = 0 in
(4.30) and noting that for a small the denominator of the integrand in (4.30)
then behaves like
_
1
1
F
2
_
a +
1
F
2
a
3
h
2
3
+ (4.76)
Therefore the integral in (4.30) diverges as F 1. This phenomenon is
illustrated graphically in Figure 4.9, where we plot values of
1
(0) versus
F for F > 1. The curve has a vertical asymptote at F = 1, and the
linear theory is not valid for F close to 1. Again, this nonuniformity will be
removed in Chapter 7 where we develop nonlinear theories.
0
0.01
0.02
0.03
0.04
0.4 0.6 0.8 1.0 1.2 1.4
Fig. 4.9. Values of the amplitude of the free-surface prole versus F for the pressure
distribution (4.70) with = 0.001. The unit of length in the vertical scale is U
2
/g.
As we shall see in Chapter 7 the nonuniformities in Figures 4.8 and 4.9 are
associated with the existence of solution branches corresponding to solitary
waves.
5
Nonlinear waves asymptotic solutions
The linear results of Chapter 4 show examples of free-surface ows that ap-
proach either periodic wave trains or at free surfaces in the far eld. As we
shall see this is also the case for nonlinear free-surface ows. However, peri-
odic waves if present are then nonlinear. This motivates a study of nonlinear
travelling gravitycapillary waves since they describe the far-eld behaviour
of nonlinear ows past disturbances. These solutions are useful in nonlinear
computations because they can be employed to derive ecient truncation
procedures (see for example Section 7.2.1). They are also interesting canon-
ical free-surface ow problems. This chapter is concerned with asymptotic
solutions for nonlinear waves. Fully nonlinear numerical solutions will be
presented in Chapter 6.
5.1 Periodic waves
As shown in Section 2.4.1, nonlinear travelling gravitycapillary periodic
waves are described by the equations
xx
+
yy
= 0, h < y < (x), (5.1)
y
=
x
x
on y = (x), (5.2)
1
2
(
2
x
+
2
y
) +gy
T
xx
(1 +
2
x
)
3/2
= B on y = (x), (5.3)
(x +, y) = (x, y), (5.4)
(x +) = (x), (5.5)
1
_

0
x
dx = c, (5.6)
129
130 Nonlinear waves asymptotic solutions
_

0
(x)dx = 0. (5.7)
The ow conguration is illustrated in Figure 2.4. We will derive analytical
approximations by improving on the linear approximations of Section 2.4.
We rst introduce an amplitude parameter to measure the size of the waves
(a precise denition will be given later) and seek solutions in the form of
expansions in powers of . Thus we write
(x, y) = cx +
1
(x, y) +
2
2
(x, y) +
3
3
(x, y) +O(
4
), (5.8)
(x) =
1
(x) +
2
2
(x) +
3
3
(x) +O(
4
), (5.9)
c = c
0
+c
1
+
2
c
2
+
3
c
3
+O(
4
), (5.10)
B = B
0
+B
1
+
2
B
2
+
3
B
3
+O(
4
). (5.11)
Such expansions were rst used by Stokes (see [143]) and (5.8)(5.11) are
often referred to as the Stokes expansion. As we shall see,
1
(x, y),
1
(x)
are the linear approximations (2.88), (2.89) and the terms of order
2
,
3
, . . . are nonlinear corrections.
One diculty is that the boundary conditions (5.2) and (5.3) are expressed
on y = (x), where (x) is itself expressed as an expansion in powers of
(see (5.9)). Therefore it is not possible to equate powers of directly. We
resolve this diculty by expanding expanding
y
(x, (x)) and
x
(x, (x)) in
Taylor expansions about y = 0. This leads to
y
(x, (x)) =
y
(x, 0) +
yy
(x, 0)(x) +
yyy
(x, 0)
2
(x)
2
+
yyyy
(x, 0)
3
(x)
6
+ , (5.12)
x
(x, (x)) =
x
(x, 0) +
xy
(x, 0)(x) +
xyy
(x, 0)
2
(x)
2
+
xyyy
(x, 0)
3
(x)
6
+ . (5.13)
Subsituting the expansions (5.8)(5.10) into (5.12) and (5.13) gives
y
(x, (x)) =
1y
+
2
(
2y
+
1yy
1
)
+
3
_
3y
+
2yy
1
+
2
1yy
+
2
1
1yyy
2
_
+O(
4
). (5.14)
Similarly, we obtain
x
(x, (x)) = c
0
+(c
1
+
1x
) +
2
(c
2
+
2x
+
1xy
1
)
+
3
_
c
3
+
3x
+
2xy
1
+
2
1xy
+
2
1
1xyy
2
_
+O(
4
).
(5.15)
Substituting (5.8)(5.11) into (5.1)(5.7), using (5.14) and (5.15) and equat-
ing equal powers of yields
B
0
=
c
2
0
2
(5.16)
and the three linear problems
1xx
+
1yy
= 0, h < y < 0, (5.17)
1y
c
0
1x
= 0, y = 0, (5.18)
1y
= 0, y = h, (5.19)
1xx
+c
0
1x
+g
1
= B
1
c
0
c
1
, y = 0; (5.20)
2xx
+
2yy
= 0, h < y < 0, (5.21)
2y
c
0
2x
= c
1
1x
1yy
+
1x
1x
, y = 0, (5.22)
2y
= 0, y = h, (5.23)
2xx
+c
0
2x
+g
2
= c
0
1xy
c
1
1x
1
2
(
2
1x
+
2
1y
)
+B
2
1
2
c
2
1
c
0
c
2
, y = 0; (5.24)
3xx
+
3yy
= 0, h < y < 0, (5.25)
3y
c
0
3x
=
1
2yy

2
1yy

1
2
2
1
1yyy
+c
1
2x
+c
2
1x
+
1x
2x
+
1x
1xy
+
1x
2x
, y = 0, (5.26)
3y
= 0, y = h, (5.27)
3xx
+c
0
3x
+g
3
=
3T
2
1xx
2
1x
c
1
c
2
c
1
2x
c
1
1xy
1x
c
2
1x
2x
1x
1xy
1
c
0
c
3
c
0
2xy
1
c
0
1xy
1
2
c
0
2
1
1xyy

1y
2y

1y
1yy
1
+B
3
, y = 0.
(5.28)
We shall restrict our attention to solutions that are symmetric. Assuming
that the origin of x is chosen so that the wave is symmetric about the axis
y = 0, we impose
(x, y) = (x, y), (5.29)
(x) = (x). (5.30)
Using the expansions (5.8) and (5.9) we have
i
(x, y) =
i
(x, y), i = 1, 2, . . . , (5.31)
i
(x) =
i
(x), i = 1, 2, . . . , (5.32)
Applying the method of separation of variables, we seek a solution of (5.17)
in the form
1
(x, y) = X(x)Y (y). (5.33)
Substituting (5.33) in (5.17) yields
X
(x)
X(x)
=
Y
(y)
Y (y)
= constant =
2
. (5.34)
Here we choose the separation constant to be negative, so that the solution
is periodic in x. Relations (5.33) and (5.19) yield
Y
(h) = 0. (5.35)
Equation (5.34) gives two ordinary dierential equations. Using the condi-
tions (5.31) and (5.35), we write their solutions as
X(x) = sin x, (5.36)
Y (y) = cosh (y +h). (5.37)
The periodicity condition (5.4) implies that
= nk, (5.38)
where n is a positive integer and
k =
2
(5.39)
is the wavenumber. Multiplying (5.36) by (5.37) and taking a linear combi-
nation of the solutions corresponding to the values (5.38) of , we obtain
1
(x, y) =
n=1
B
n
Similarly, we can write the solutions of (5.21) and (5.25) as
2
(x, y) =
n=1
F
n
cosh nk(y +h) sin nkx, (5.41)
3
(x, y) =
n=1
G
n
Here B
n
, F
n
and G
n
are constants.
Using (5.7), the periodicity condition (5.5) and the symmetry conditions
(5.32), we may express
1
(x),
2
(x) and
3
(x) as Fourier series:
1
(x) =
n=1
A
n
cos nkx, (5.43)
2
(x) =
n=1
E
n
cos nkx, (5.44)
3
(x) =
n=1
D
n
cos nkx, (5.45)
where A
n
, E
n
and D
n
are constants.
We now dene the parameter as
=
a
, (5.46)
where a is the rst Fourier coecient of (x), i.e.
a =
2
_

0
(x) cos kxdx. (5.47)
It follows from (5.43)(5.45) that
A
1
= (5.48)
and
E
1
= D
1
= 0. (5.49)
Substituting (5.40) and (5.43) into (5.18) and equating coecients of
sin nkx yields
c
0
A
n
= B
n
sinh nkh. (5.50)
Similarly, substituting (5.40) and (5.43) into (5.20) gives B
1
= c
0
c
1
and
T
A
n
n
2
k
2
+gA
n
+c
0
B
n
nk cosh nkh = 0. (5.51)
Eliminating B
n
between (5.50) and (5.51) yields
_
g +
T
n
2
k
2
c
2
0
nk
tanh nkh
_
A
n
= 0, n = 1, 2, . . . (5.52)
Since A
1
,= 0 (see (5.48)), (5.52) with n = 1 implies that
c
2
0
=
_
g
k
+
T
k
_
tanh kh. (5.53)
Relation (5.52) with n > 1 gives
A
n
= 0, n = 2, 3, . . . , (5.54)
provided that
g +
T
n
2
k
2
c
2
0
nk
tanh nkh
,= 0. (5.55)
The solution of the rst order problem is then
1
=
c
0
A
1
sinh kh
cosh k(y +h) sin kx, (5.56)
1
= A
1
cos kx. (5.57)
However, if for some value m of the integer n, the condition (5.55) is not
satised i.e. if
g +
T
m
2
k
2
c
2
0
mk
tanh mkh
= 0, (5.58)
then the solution is
1
=
c
0
A
1
sinh kh
cosh k(y +h) sin kx
c
0
A
m
sinh mkh
cosh mk(y +h) sin mkx,
(5.59)
1
= A
1
cos kx +A
m
cos mkx, (5.60)
where A
m
is arbitrary at this stage of the calculations. As we shall see the
values of A
m
are determined by solvability conditions at higher order in .
We conclude this subsection by noting that in the particular case of water
of innite depth, so that h , the condition (5.55) and the solutions
(5.56), (5.57) and (5.59), (5.60) reduce to
g +
T
n
2
k
2
c
2
0
nk ,= 0, (5.61)
1
= c
0
A
1
e
ky
sin kx, (5.62)
1
= A
1
cos kx, (5.63)
1
= c
0
A
1
e
ky
sin kx c
0
A
m
e
mky
sin mkx, (5.64)
1
= A
1
cos kx +A
m
cos mkx. (5.65)
Using (5.53), equation (5.61) reduces further to
,=
1
n
, (5.66)
where is dened by
=
4
2
T
g
2
. (5.67)
5.1.1 Solutions when condition (5.55) is satised
We consider rst the case when the condition (5.55) is satised. Substituting
(5.56) and (5.57) into the right-hand sides of (5.22) and (5.24), we obtain,
after some algebra,
2y
c
0
2x
= c
1
A
1
k sin kx +c
0
A
2
1
k
2
coth khsin 2kx
2xx
+c
0
2x
+g
2
=
c
2
0
A
2
1
k
2
4

1
4
c
2
0
k
2
A
2
1
cosh
2
kh
sinh
2
kh
(5.68)
+c
0
c
1
A
1
k
cosh kh
sinh kh
cos kx
+
_
3
4
c
2
0
A
2
1
k
2
1
4
c
2
0
k
2
cosh
2
kh
sinh
2
kh
_
cos 2kx +B
2
.
Using the expansions (5.41) and (5.44) in the left-hand sides of the previous
two relations and equating coecients of sin nkx and cos nkx yields
B
2
=
c
2
1
2
+
1
4
c
2
0
k
2
A
2
1
sinh
2
kh
, (5.69)
F
1
k sinh kh +c
0
E
1
k = c
1
A
1
k, (5.70)
F
1
k cosh khtanh kh = c
1
A
1
k, (5.71)
2kF
2
sinh 2kh + 2kc
0
E
2
=
A
2
1
k
2
c
0
tanh kh
, (5.72)
2kF
2
cosh 2kh +
_
4k
2
T
+g
_
E
2
c
0
=
3
4
c
0
A
2
1
k
2
1
4
c
0
A
2
1
k
2
tanh
2
kh
, (5.73)
nkF
n
sinh nkh +nkc
0
E
n
= 0, n = 3, 4, . . . (5.74)
T
E
n
n
2
k
2
+c
0
nkF
n
cosh nkh +gE
n
= 0, n = 3, 4, . . . (5.75)
Equations (5.49), (5.70) and (5.71) give
F
1
= 0, c
1
= 0. (5.76)
Solving (5.72) and (5.73) yields
E
2
=
A
2
1
k
2
c
2
0
4
4 tanh kh 3 tanh
2
khtanh 2kh + tanh 2kh
tanh
2
kh[2kc
2
0
(4k
2
T/ +g) tanh 2kh]
, (5.77)
F
2
=
1
2k sinh 2kh
_
A
2
1
k
2
c
0
tanh kh
2kc
0
E
2
_
. (5.78)
Solving (5.74) and (5.75) gives
F
n
= 0, E
n
= 0, n = 3, 4, . . . (5.79)
Therefore
2
= E
2
cos 2kx, (5.80)
2
= F
2
cosh 2nk(y +h) sin 2nkx, (5.81)
where E
2
and F
2
are dened by (5.77) and (5.78). Combining (5.9), (5.47)
and (5.80) gives
= A
1
cos kx +
2
E
2
cos 2kx. (5.82)
To conclude the calculation up to order
2
we need to calculate c
2
. This is
done by substituting the expressions (5.43), (5.46), (5.47), (5.76), (5.80) and
(5.81) for
1
,
2
,
1
,
2
, c
0
and c
1
into the right-hand sides of (5.26) and
(5.28) and the series (5.42) and (5.45) into the left-hand sides. Equating the
coecients of cos kx and those of sin kx gives two equations. Eliminating
G
1
between these two equations yields, after some algebra,
c
2
=
c
0
A
2
1
k
2
4
_
(coth kh + 4 coth 2kh 3 tanh kh)S
2
+3 2 coth khcoth 2kh
3Tk
2
4gS
1
tanh kh
_
, (5.83)
where
S
1
=
_
1 +
Tk
2
g
_
tanh kh (5.84)
and
S
2
= 2
(Tk
2
+g) tanh kh[tanh 2kh 2 coth kh tanh 2kh(2 sinh
2
kh)
1
]
(16Tk
2
+ 4g) tanh 2kh 8(Tk
2
+g) tanh kh
.
(5.85)
Therefore the phase velocity c is given up to order
2
by (5.10), where c
0
, c
1
and c
2
are dened by (5.53), (5.76) and (5.83) respectively.
We conclude this section by examining the behaviour of the solution (5.10)
in the shallow-water limit (i.e. as kh 0). Using the properties
sinh kh kh as kh 0
and
cosh kh 1 as kh 0,
we nd that (5.83) yields
c
2
c
0
1
(kh)
2
as kh 0. (5.86)
Since c
1
= 0, (5.10) then implies that, on the one hand, the ratio of the rst
two terms of the expansion (5.10) for c satises
2
c
2
c
0

2
(kh)
2
as kh 0; (5.87)
on the other hand, the expansion (5.10) is only valid asymptotically if the
ratio
2
c
2
/c
0
is small in the limit 0. Relation (5.87) shows that for any
value of > 0 the ratio
2
c
2
/c
0
can be made arbitrarily large by taking kh
suciently small. In other words the asymptotic expansions derived in this
section become invalid as kh 0. Appropriate asymptotic expansions that
are valid when both and kh are small will be derived in Section 5.2.
5.1.2 Solutions when condition (5.55) is not satised
We now assume that the condition (5.55) is not satised for some value m
of n. We will present explicit calculations for m = 2. Therefore we have
g + 4
T
k
2
2kc
2
0
sinh 2kh
cos 2kh = 0. (5.88)
The solutions for
1
and
1
are obtained by setting m = 2 in (5.59) and
(5.60). This gives
1
=
c
0
A
1
sinh kh
cosh k(y +h) sin kx
c
0
A
2
sinh 2kh
cosh 2k(y +h) sin 2kx, (5.89)
1
= A
1
cos kx +A
2
cos 2kx. (5.90)
We determine the constants A
2
and c
1
by substituting (5.89) and (5.90)
into the right-hand sides of (5.22) and (5.24). To simplify the algebra and to
illustrate the main ideas we shall present these calculations in the particular
case of innite depth. Taking the limit h in (5.53), (5.88) and (5.89)
gives
c
2
0
=
g
k
+
T
k, (5.91)
g + 4
T
k
2
2kc
2
0
= 0, (5.92)
1
= c
0
A
1
e
ky
sin kx c
0
A
2
e
2ky
sin 2kx. (5.93)
As noted earlier, the condition (5.92) can be rewritten in the simple form
=
1
2
, (5.94)
where is dened by (5.67).
In water of innite depth, cx as y . Therefore (5.8) implies
that
i
0, i = 1, 2, . . . as y . (5.95)
Using (5.95) with i = 2 and following the derivation of (5.40) we obtain
2
(x, y) in the form
2
(x, y) =
n=1
F
n
e
nky
sin nkx. (5.96)
Substituting (5.44), (5.96), (5.90) and (5.93) into (5.22) and (5.24) and
equating coecients of sin kx and cos kx yields
k

F
1
= c
1
A
1
k +
3
2
c
0
A
1
A
2
k
2
, (5.97)
c
0

F
1
k =
c
2
0
A
1
A
2
k
2
2
+c
0
c
1
A
1
k. (5.98)
Similarly, equating the coecients of sin 2kx and cos 2kx yields
2k

F
2
+ 2kc
0
E
2
= 2c
1
A
2
k +c
0
A
2
1
k
2
, (5.99)
4
T
k
2
E
2
+gE
2
+ 2

F
2
kc
0
=
c
2
0
A
2
1
k
2
2
+ 2c
0
c
1
A
2
k. (5.100)
Solving (5.97) and (5.98) gives
c
1
=
c
0
A
2
k
2
. (5.101)
Similarly (5.99) and (5.100) give
4
T
k
2
E
2
gE
2
+ 2kc
2
0
E
2
=
c
2
0
k
2
2
A
2
1
4c
0
c
1
kA
2
(5.102)
and thus
E
2
_
2kc
2
0
4Tk
2
g
_
= 4c
0
c
1
A
2
k +
c
2
0
A
2
1
k
2
2
. (5.103)
Using (5.91) and (5.92), it can easily be checked that the contents of the
parentheses on the left-hand side of (5.103) equal zero. Therefore
c
1
A
2
=
A
2
1
kc
0
8
. (5.104)
Combining (5.101) and (5.104) yields
A
2
=
A
1
2
(5.105)
1
= A
1
cos kx
1
2
A
1
cos 2kx, (5.106)
1
= c
0
A
1
e
ky
sin kx
c
0
2
A
1
e
2ky
sin 2kx, (5.107)
c
1
=
1
4
c
0
kA
1
. (5.108)
We dene the steepness s of the wave by the equation
s =
(0) (/2)
(5.109)
Noting that = s/2 to leading order and using (5.9), (5.10), (5.48), (5.106)
and (5.108) we obtain
c = c
0
s
4
c
0
+ , (5.110)
=
s
2
cos kx
s
4
cos 2kx + . (5.111)
Relations (5.106) and (5.111) show that there are two solutions when
(5.94) is satised (one corresponding to the + sign and one to the sign).
The corresponding free-surface proles in water of innite depth are shown
in Figures 5.1 and 5.2.
0
0.5
1.0
1.5
0 5 10 15 20
Fig. 5.1. The Wilton ripple (5.106) with A
1
= 1 and k = 1. This solution corre-
sponds to the + sign in (5.106).
They are usually referred to as the Wilton ripples (because Wilton [198]
was one of the rst to calculate them). One solution is characterised by a
crest dimple and the other by a trough dimple. We note that (5.82) and
(5.111) both involve cos kx and cos 2kx. However, a crucial dierence is that
the cos 2kx term is of order
2
in (5.82) whereas it is of order s, and therefore
of order , in (5.111). This implies that (5.82) approaches A
1
cos kx as
0 and that (5.111) approaches (s/2)[cos kx (1/2) cos 2kx] as 0.
Therefore the solution (5.111) exhibits the trough and crest dimples shown
in Figures 5.1 and 5.2 however small and s are.
The above calculations illustrate the nonuniqueness of gravitycapillary
periodic waves: we obtain two dierent solutions if the condition (5.58) is
0
0.5
1.0
0 5 10 15 20
Fig. 5.2. The Wilton ripple (5.106) with A
1
= 1 and k = 1. This solution corre-
sponds to the sign in (5.106).
satised for m = 2. Similar nonuniqueness occurs when (5.58) is satised for
values of m > 2. Calculations were performed by Nayfeh [116] for m = 3.
The analysis can also be extended to nite depth, and solutions valid when
(5.58) is satised can be derived. However, the calculations for larger values
of m become quickly intractable because it is necessary to calculate the
solution to high order in in order to nd the values of A
m
in (5.60). Such
solutions will be calculated numerically in Sections 6.5.3.1 and 6.5.3.2.
We also note that the denominator in (5.77) vanishes when (5.58) is satis-
ed for m = 2. Similarly, the denominators in
n
and
n
for n 3 are also
found to vanish when (5.58) is satised for some integer value m. This is
consistent with the solutions of Section 5.1.1, which were derived under the
assumption (5.55). However, the expansions of Section 5.1.1 are nonuniform
in the sense that the ratios of successive terms (for example the ratio of
2
2
and
1
) can be made arbitrarily large by making the left-hand side of (5.55)
suciently small by appropriate choices of the parameters.
For future comparison of the present results with the numerical compu-
tations of Chapter 6 we now rewrite (5.91) and (5.110) in terms of the
parameter (see (5.67)) and a parameter dened by
=
2c
2
g
. (5.112)
This gives
= 1 + (5.113)
and
= 1 +
s
2
(1 +) =
3
2

3
4
s + (5.114)
Finally let us mention that Wilton [198] calculated further terms in the
expansion and found
=
3
2

3
4
s
2
s
2
4
(5.115)
5.2 The Kortewegde Vries equation
As we saw in Section 5.1.1, the Stokes expansion becomes nonuniform as h
0. This suggests that a dierent expansion should be used when both the
depth h and the amplitude of the waves are small. We study in this section
asymptotic expansions valid when h is small with various assumptions on
the size of the amplitude of the waves. For the sake of generality we will
derive these equations in the time-dependent case. Travelling waves will
then be found as particular cases.
The exact time-dependent-potential free-surface ow equations are given
by (2.102)(2.105). We rewrite these equations by choosing the origin of y
on the bottom. This gives
xx
+
yy
= 0, 0 < y < h +, (5.116)
y
= 0 on y = 0, (5.117)
t
+
x
y
= 0 on y = h +, (5.118)
g +
t
+
1
2
(
2
x
+
2
y
)
T
xx
(1 +
2
x
)
3/2
= C on y = h +, (5.119)
where h is the depth of the uid. Here we have written the elevation of the
free surface as y = h +, where h is the undisturbed depth.
We introduce the velocity scale c
0
= (gh)
1/2
and denote the typical am-
plitude of by a. We also dene the amplitude parameter
=
a
h
(5.120)
and the depth parameter
=
h
2
l
2
. (5.121)
Next we rescale the variables as
x =
x
l
, y =
y
h
,

t =
c
0
l
t, (5.122)
=

a
,

=
c
0
gla
. (5.123)
That the scalings in the x- and y- directions in (5.122) are dierent is crucial
in deriving shallow-water approximations.
We now rewrite the nonlinear equations (5.116)(5.119) in terms of the
new variables (5.122) and (5.123). After dropping the tilde this leads to
xx
+
yy
= 0, 0 < y < 1 +, (5.124)
y
= 0 on y = 0, (5.125)
t
+
x
y
= 0 on y = 1 +, (5.126)
+
t
+
1
2
2
x
+
1
2
2
y

xx
(1 +
2
2
x
)
3/2
= 0 on y = 1 +, (5.127)
where
=
T
gh
2
.
We seek in the form of an expansion in powers of y. Thus we write
(x, y, t) =
n=0
y
n
f
n
(x, t). (5.128)
Substituting (5.128) into (5.124) and (5.125) yields
f
n
= 0 for n = 1, 2, 3, . . . , (5.129)
f
n
=

n(n 1)
2
f
n2
x
2
, n = 2, 4, 6, . . . (5.130)
Solving (5.130) recursively gives
=
n=0
(1)
n
y
2n
(2n)!
2n
f
x
2n
n
, (5.131)
where
f = f
0
. (5.132)
Substituting (5.131) into the boundary conditions (5.126) and (5.127) gives
t
+ [(1 +)f
x
]
x
[
1
6
(1 +)
3
f
xxxx
+
1
2
(1 +)
2
x
f
xxx
] +O(
2
) = 0,
(5.133)
+f
t
+
1
2
f
2
x

1
2
(1 +)
2
(f
xxt
+f
x
f
xxx
f
2
xx
)

xx
(1 +
2
2
x
)
3/2
+O(
2
) = 0. (5.134)
If all terms of order are dropped and (5.134) is dierentiated with respect
to x, we obtain
t
+ [(1 +)w]
x
= 0, (5.135)
w
t
+ww
x
+
x
xxx
= 0, (5.136)
where
w = f
x
. (5.137)
Equations (5.135) and (5.136) are the shallow-water equations. They
dene a hyperbolic system of partial dierential equations that does not
admit travelling waves (see Whitham [195]).
The Kortewegde Vries equation with the eects of surface tension in-
cluded can be derived from (5.133) and (5.134) by specialising to waves
moving to the right. To lowest order in and (i.e. to order
0
and
0
),
(5.135) and (5.136) reduce to
t
+w
x
= 0, (5.138)
w
t
+
x
= 0 (5.139)
and a solution moving to the right is
w = ,
t
+
x
= 0. (5.140)
Following Whitham [195], we seek solutions correct to order and order
in the form
w = +A+B +O(
2
+
2
). (5.141)
Here A and B are functions of and its derivatives. Equations (5.133) and
(5.134) become
t
+
x
+(A
x
+ 2
x
) +
_
B
x
1
6
xxx
_
+O(
2
+
2
) = 0, (5.142)
t
+
x
+(A
t
+
x
) +
_
B
t
1
2
xxt
_
xxx
+O(
2
+
2
) = 0. (5.143)
Since
t
=
x
+ O(, ), we can replace the t-derivatives by minus the
x-derivatives in the rst-order terms and rewrite (5.143) as
t
+
x
+(A
x
+
x
)+
_
B
x
+
1
2
xxx
_
xxx
+O(
2
+
2
) = 0. (5.144)
Equations (5.142) and (5.144) are consistent if the coecients of and in
both equations are the same. This yields
A
x
+ 2
x
= A
x
+
x
, (5.145)
B
x
1
6
xxx
= B
x
+
1
2
xxx
xxx
. (5.146)
Integrating (5.145) and (5.146) gives
A =
1
4
2
, (5.147)
B =
_
1
3

1
2
xx
. (5.148)
Substituting (5.147) and (5.148) into (5.144) gives the Kortewegde Vries
equation,
t
+
x
+
3
2
x
+
1
6
(1 3)
xxx
+O(
2
+
2
) = 0. (5.149)
We can now rewrite (5.149) in terms of the unscaled dimensional variables
(see (5.120)(5.123)) as
t
+c
0
x
+
3
2
c
0
h

x
+
1
6
c
0
h
2
(1 3)
xxx
= 0. (5.150)
We construct travelling-wave solutions of (5.150) by assuming that
= h(X), where X = x Ut. (5.151)
The function (X) represents a wave travelling to the right with constant
velocity U. Substituting (5.151) into (5.150) gives the ordinary dierential
equation
hU
+c
0
h
+
3
2
c
0
h
+
1
6
c
0
h
2
(1 3)h
= 0, (5.152)
where the primes denote derivatives with respect to X. Integrating (5.152)
and dividing by c
0
h yields
1
6
h
2
(1 3)
+
3
4
_
U
c
0
1
_
+G = 0, (5.153)
where G is a constant of integration. Multiplying (5.153) by
and inte-
grating again gives
1
3
(1 3)h
2
2
+
3
2
_
U
c
0
1
_
2
+ 4G +J = 0, (5.154)
where J is another constant of integration.
Equation (5.154) admits periodic travelling solutions. These were calcu-
lated analytically by Korteweg and de Vries [91] (see also [195] and [191]).
These solutions are called cnoidal waves because they are expressed in terms
of the Jacobian elliptic function cn. As the ratio of their wavelength and the
depth h tends to innity, the cnoidal waves approach solitary waves. These
solitary waves consist of a single hump or a single depression, depending on
the value of the parameter . As x , and its derivatives tend to
zero. It then follows that G = J = 0, and (5.154) reduces to
1
3
(1 3)h
2
2
+
3
2
_
U
c
0
1
_
2
= 0. (5.155)
It can easily be checked that the solution of (5.155) is then
= a sech
2
_
3 a
4h
2
(1 3)
_
1/2
X, (5.156)
where
a = 2
_
U
c
0
1
_
. (5.157)
When < 1/3, a > 0 and (5.156) is an elevation wave with U > c
0
. The
value of increases from = 0 at X = , rises to a maximum = a and
then returns symmetrically to = 0 at X = .
When > 1/3, a < 0 and (5.156) is a depression wave with U < c
0
. The
value of decreases from = 0 at X = , reaches a minimum = a and
then returns symmetrically to = 0 at X = .
The solution (5.156) predicts that d/dX becomes unbounded as 1/3.
This is due to the fact that the dispersive term
xxx
in (5.150) disappears
when = 1/3. We note that in the expansion procedure leading to (5.150)
it was assumed that and are both small and of the same order of mag-
nitude, i.e. = , = where 0 < 1. To obtain a model equation
valid near = 1/3, we need to include higher-order dispersive terms. This
can be achieved by assuming that
=
2
, = (5.158)
and expanding near 1/3 by writing
=
1
3
+
1
+ (5.159)
The resulting equation rewritten in terms of the dimensional variables is
t
+c
0
x
+
3
2
c
0
h

x
+
1
6
c
0
h
2
(1 3)
xxx
+
c
0
h
4
90

xxxxx
= 0. (5.160)
This is known as the fth-order Kortewegde Vries equation. It was derived
by Hunter and Vanden-Broeck [76] in the steady case and by Hunter and
Scherule [75] in the unsteady case (see also [83]).
An important question is whether there are fully nonlinear solutions con-
sistent with (5.156) when their amplitude is small. We shall see in Chapter
6 that the answer is positive when = 0 or > 1/3. However, there are
no fully nonlinear solutions of small amplitude consistent with (5.156) when
0 < < 1/3. The fully nonlinear solutions are then characterised by a train
of ripples in the far eld. Such waves are often referred to as generalised
solitary waves, to distinguish them from true solitary waves, which are at
in the far eld.
6
Numerical computations of nonlinear water waves
6.1 Formulation
We consider again a train of periodic waves like that shown in Figure 2.4.
Suppose that the waves have wavelength and propagate to the left in
a channel bounded below by a horizontal bottom. We take a frame of
reference moving with the wave and seek steady solutions. We will present
the numerical methods of interest in terms of cartesian coordinates with
x = y = 0 at a crest of the wave (see Figure 6.1). Some alternative choices
for the origin of coordinates will be used in Sections 6.4 and 6.5. Gravity is
taken as acting in the negative y-direction.
A
B
C x
y
A'
B'
C'
= 0
= Q
= 2Q
Fig. 6.1. A wave of wavelength travelling in a channel bounded below by a hor-
izontal bottom. A frame of reference moving with the wave is chosen. The free
surface = 0, the bottom = Q and the image of the free surface = 2Q are
shown.
The potential function and the streamfunction were introduced in
Section 2.3. We choose = 0 at x = y = 0 and = 0 on the free surface
148
6.1 Formulation 149
and denote by Q the constant value of on the horizontal bottom and by
u and v the horizontal and vertical components of the velocity.
As in Section 2.4 and Chapter 5, we dene the phase velocity as the
average horizontal velocity at a constant level of y in the uid. Using (5.6),
we obtain
c =
1
_

o
u(x, y)dx =
1
_

0
x
dx =
1
[(, y) (0, y)] =

1
(, y). (6.1)
Therefore (, y) = c and it follows from the assumed periodicity and
symmetry of the wave that
_
n
2
, y
_
= c
n
2
, n = 0, 1, 2, . . . (6.2)
Following Stokes [144] and the analysis in Chapter 3, we will use the
potential function and the streamfunction as independent variables. We
dene the complex potential
f = +i.
The kinematic boundary condition on the horizontal bottom can then be
satised by using the method of images. Thus we reect the ow in the
bottom (see Figure 6.1). The ow conguration in the complex potential
plane is shown in Figure 6.2: it is the strip < < , 2Q < < 0.
B C A
A'
B' C'
= 0
= 2Q
Fig. 6.2. The ow conguration of Figure 6.1 in the complex potential plane.
As noted in Chapter 3, the obvious advantage of working in the complex
f-plane is that the unknown free surface and its image in Figure 6.1 have
been mapped onto the known boundaries = 0 and = 2Q.
150 Numerical computations of nonlinear water waves
We dene the function +i by the relations
+i =
1
u iv
= x
+iy
. (6.3)
Since u iv is an analytic function of f, +i is also an analytic function
of f.
Next we map the f-plane into the unit disk in the complex t-plane by the
transformation
t = e
2if/c
= e
2i/c
e
2/c
. (6.4)
The ow conguration in the t-plane is shown in Figure 6.3.
A
C B B'
A'
C '
Fig. 6.3. The ow conguration of Figures 6.1 and 6.2 in the complex t-plane. The
solid curve [t[ = 1 corresponds to the free surface and the broken curve [t[ = r
2
0
to
the image of the free surface.
The free surface is mapped into the circle [t[ = 1, the bottom into the circle
[t[ = e
2Q/c
and the image of the free surface into the circle [t[ = e
4Q/c
.
To compute solutions, we need to express the fact that +i is an analytic
function of t in the annulus r
2
0
< [t[ < 1, where
r
0
= e
2Q/c
. (6.5)
There are two methods for doing this. The rst is to represent + i as a
Laurent expansion in powers of t. This leads to the numerical procedure we
call series truncation. The second is to apply the Cauchy integral formula to
the function +i in the annulus of Figure 6.3. This leads to the boundary
integral equation method. Both methods were introduced in Chapter 3.
They are described within the framework of water waves in Sections 6.2 and
6.3 respectively.
6.2 Series truncation method 151
6.2 Series truncation method
The function +i is an analytic function of t in the annulus r
2
0
< [t[ < 1.
Therefore it can be represented by the Laurent expansion
+i = a
0
+
n=1
a
n
t
n
+
n=1
b
n
t
n
= a
0
+
n=1
a
n
e
2inf/c
+
n=1
b
n
e
2inf/c
. (6.6)
Since = 2Q is the image of the free surface = 0 in the bottom we have
(, 2Q) +i(, 2Q) = (, 0) i(, 0), (6.7)
and it follows from (6.6) that
b
n
= a
n
r
2n
0
, n = 1, 2, . . . (6.8)
Furthermore (6.1) implies that
a
0
=
1
c
. (6.9)
Substituting (6.8) and (6.9) into (6.6) gives
+i =
1
c
+
n=1
a
n
e
2inf/c
+
n=1
a
n
r
2n
0
e
2inf/c
. (6.10)
We shall determine coecients a
n
in (6.10) such that the dynamic bound-
ary condition (5.3) is satised. To do so, we need to rewrite (5.3) in terms
of and . Using (3.3), (3.6) and (6.3) we obtain
1
2
1
2
+

2
+g
_

0
()d
T
2
+

2
)
3/2
= B, (6.11)
where B is the Bernoulli constant and

() = (, 0) and

() = (, 0)
denote the values of and on the free surface = 0.
In the remaining part of this section, we introduce dimensionless variables
by using as the unit length and c as the unit velocity. In terms of the
dimensionless variables, (6.11) becomes
1
2
1
2
+

2
+
2
_

0
()d

2
2
+

2
)
3/2
= B, (6.12)
where and are dened by (5.67) and (5.112) respectively, and (6.10)
becomes
+i = 1 +
n=1
a
n
e
2inf
+
n=1
a
n
r
2n
0
e
2inf
. (6.13)
We truncate the innite series in (6.13) after N 1 terms and determine
the N + 1 unknowns B, and a
n
, n = 1, . . . , N 1, by collocation. Thus
we introduce the N collocation points
I
=
1
2
I 1
N 1
, I = 1, . . . , N, (6.14)
and satisfy the dynamic boundary condition (6.12) at these points. This
gives N nonlinear algebraic equations. The nal equation is obtained by
xing the amplitude of the wave, for example by imposing
_
1/2
0
()d = s, (6.15)
where s is given and is called the steepness of the wave. It is dened as the
dierence in height between a crest and a trough of the wave prole divided
by the wavelength.
In the particular case of water of innite depth, r
0
= 0 and (6.8) implies
b
n
= 0, n = 1, 2, . . . The Laurent series (6.13) reduces then to the Taylor
series
+i =
n=0
a
n
t
n
=
n=0
a
n
e
2inf
, (6.16)
where a
0
= 1. The representation (6.16) can be derived directly by noting
that when r
0
= 0, (6.4) maps the ow domain into the unit circle 0 < [t[ < 1
of the t-plane. Therefore +i can be represented by the Taylor expansion
(6.16).
Various variations of the series truncation method have been proposed.
For example, Vanden-Broeck [172] calculated waves in water of innite depth
by expanding the complex velocity w = u iv instead of + i in powers
of t, i.e. by writing
w = 1 +
n=1
c
n
e
2inf
. (6.17)
6.3 Boundary integral equation method
Instead of using the series representation (6.10), we now derive a relation
between and by using Cauchy integral formula (see (2.38)). Such an
approach was used in Chapter 3 to compute cusped cavities (see Section
3.1.2.2). The presentation in this section follows [135], [162] and [76].
We apply the Cauchy integral formula to the function + i in the
complex t-plane with the contour C chosen as the boundaries of the an-
nulus in Figure 6.3. This yields
(t
0
) +i(t
0
) =
1
i
_
C
(t) +i(t)
t t
0
dt. (6.18)
The contour C consists of the circle [t[ = 1 oriented clockwise and of the
circle [t[ = r
2
0
oriented anticlockwise. Here t
0
is assumed to be on the unit
circle [t[ = 1 of Figure 6.3. Therefore the integral in (6.18) is a Cauchy
principal value.
Applying (6.4) with = 0 as a change of variables and using (6.7), we
can rewrite (6.18) as
() +i
() =
1
_
_

() +i
()]
e
i/
e
i/
e
i/
d
r
2
0
_

() i
()]
r
2
0
e
i/
r
2
0
e
i/
e
i/
d
_
, (6.19)
where = c/2. Taking the real part of (6.19), we obtain the following
relation between

() and

():
() =
1
_
_

'
_
[
() +i
()]
e
i/
e
i/
e
i/
_
d
r
2
0
_

'
_
[
() i
()]
r
2
0
e
i/
r
2
0
e
i/
e
i/
_
d
_
, (6.20)
where ' denotes the real part.
Relations (6.11) and (6.20) dene a system of nonlinear integro-dierential
equations for the unknown functions

() and

(). We will solve this
system numerically. First we introduce the mesh points
I
=
c
2
+
c
N 1
(I 1), I = 1, . . . , N, (6.21)
the midpoints
m
I
=

I
+
I1
2
, I = 1, . . . , N 1, (6.22)
and the unknowns
I
=

(
I
), I = 1, . . . , N, (6.23)
I
=

(
I
), I = 1, . . . , N. (6.24)
Since

1
= 0 =

N
= 0 by symmetry, there are only 2N 2 unknowns
I
and

I
. The integrals in (6.20) can be approximated by the trapezoidal
rule (which is spectrally accurate for periodic functions) with =
J
, J =
1, . . . , N, and =
m
I
, I = 1, . . . , N 1. The symmetry of the quadrature
and of the discretisation enable us to evaluate the Cauchy principal value
as if it were an ordinary integral (see Section 3.1.2). This gives N 1
equations.
Next we evaluate the derivatives

and

at the midpoints by centred

dierence formulae. For example, using two points we have
(
m
I
) =

I+1
I
h
, I = 1, . . . , N 1, (6.25)
and using four points we have
(
m
I
) =

I1
27
I
+ 27
I+1
I+2
24h
, I = 2, . . . , N 2. (6.26)
Here
h =
2
N 1
=
c
N 1
is the mesh size. We also evaluate

and

at the midpoints by centred
interpolation formulae. For example, using two points we have
(
m
I
) =

I+1
+

I
2
, I = 1, . . . , N 1, (6.27)
and using four points we have
(
m
I
) =

I1
+ 9
I
+ 9
I+1
I+2
16
, I = 2, . . . , N 2. (6.28)
We note that formulae like (6.26) and (6.28) can be extended for I = 1 and
I = N1 since

I
can be dened for I < 1 and I > N by using the assumed
periodicity and symmetry of the wave.
Now we need to approximate the integral in (6.11). A simple way to do
this is rst to dene
Y () =
_

0
()d. (6.29)
If N is odd we can calculate Y (
I
) by using
Y (
I+1
) = Y (
I
) +

I+1
+

I
2
h, I =
N + 1
2
, . . . , N 1, (6.30)
and
Y (
I1
) = Y (
I
)
I1
+

I
2
h, I =
N + 1
2
,
N 1
2
, . . . , 2, (6.31)
where
Y (
(N+1)/2
) = 0. (6.32)
A similar formula can be derived when N is even. The values of Y (
m
I
) can
then be evaluated by interpolation, for example by using the interpolation
formula
Y (
m
I
) =
Y (
I
) +Y (
I+1
)
2
, I = 1, . . . , N 1. (6.33)
To describe the numerical procedure further we need to introduce dimen-
sionless variables. There are several possible choices. For example, we could
choose Q/c as the unit length and c as the unit velocity. Here, however, we
will follow the choice of Section 6.2 and choose as the unit length and c
as the unit velocity. We then set c = 1 and = 1 in (6.20) and use (6.12)
instead of (6.11). By satisfying (6.12) and (6.20) at the midpoints (6.22) we
obtain 2N2 algebraic equations for the 2N unknowns

I
, I = 1, . . . , N,

I
,
I = 2, . . . , N1, B and . A further equation expresses that the wavelength
is (here = 1):
_
1
0
()d = 1. (6.34)
The nal equation xes the amplitude of the wave, for example by writing
_
1/2
0
()d = s, (6.35)
where s is the steepness. The integrals in (6.34) and (6.35) are evaluated
numerically (for example by the trapezoidal rule). The resulting system of
2N nonlinear algebraic equations with 2N unknowns is solved by Newtons
method for given values of and s.
When r
0
= 0, the water is of innite depth and, using = c/2, (6.20)
reduces to the simple integral relation
() =
1
c

1
c
_
c/2
c/2
() cot

c
( ) d. (6.36)
We conclude this section by mentioning that the eciency of the numerical
method can be improved by using the assumed periodicity and symmetry of
the waves. The symmetry implies that
() =
() and

() =

(). (6.37)
Therefore the integrals in (6.20) and (6.36) can be reduced to integrals from
0 to c/2. For example, we can rewrite (6.36) as
() =
1
c

1
c
_
c/2
0
()
_
cot

c
( ) + cot

c
( +)
_
d. (6.38)
We can then solve problems by following the numerical scheme described
above but with the mesh points restricted to the interval 0 < < c/2, i.e.
by replacing (6.21) by
I
=
c
2(N 1)
(I 1), I = 1, . . . , N. (6.39)
6.4 Numerical methods for solitary waves
Some solitary waves have already been introduced as solutions of the
Kortewegde Vries equation (see Section 5.2). As we shall see, there are
three dierent types of solitary wave. The rst type is a solitary wave with
a free-surface prole that approaches monotonically a constant level in the
far eld (see Figure 6.4).
Fig. 6.4. A solitary wave with monotonic decay to a constant level.
The solution (5.156) of the Kortwegde Vries equation is an example of
a solitary wave of this type. Solitary waves of the second type also have a
free-surface prole that approaches a constant level in the far eld but with
decaying oscillations. Examples of such waves occur when both gravity and
surface tension are included. They are described in Section 7.2.2 and an
extension to three-dimensional waves is presented in Chapter 10. There are
both elevation and depression waves (see Figures 6.5 and 6.6).
The third type of solitary wave does not approach a uniform stream in
the far eld but is characterised by oscillations of constant amplitude (see
Figure 6.7). We will refer to these waves as generalised solitary waves to
contrast them with true solitary waves, whose proles approach constant
levels in the far eld. As we shall see these generalised solitary waves occur
when both gravity and surface tension are taken into account.
Fig. 6.5. An elevation solitary wave with a decaying oscillatory tail.
Fig. 6.6. A depression solitary wave with a decaying oscillatory tail.
There are two approaches to the numerical investigation of solitary waves.
The rst is to derive schemes to compute them directly. The second is to
compute periodic waves by using the schemes outlined in Sections 6.2 and
6.3 and then to take larger and larger values of the wavelength. The rst
approach works particularly well for solitary waves of the rst two types (see
Figures 6.46.6). The innite domain < x < is then truncated to
A < x < A, where A is a large real number, and the contributions from
A < x < and from < x < A are either neglected or approximated
by asymptotic solutions. The second approach is more appropriate for com-
puting generalised solitary waves, i.e. solitary waves of the third type (see
Figure 6.7). This reason is that no approximations are made about the
behaviour of the ow in the far eld and we can then be condent that
the oscillations in the far eld are not generated by the truncation of the
domain.
6.4.1 Boundary integral equation methods
In this section we follow the rst approach and derive a boundary integral
equation method to compute directly the solitary waves of Figures 6.4
6.6. Therefore we assume that the ow approaches a uniform stream with
constant velocity U and constant depth H as [x[ . The value Q of
Fig. 6.7. A generalised solitary wave.
the streamfunction on the bottom is then given by
Q = UH. (6.40)
To obtain an integral relation between

() and

() on the free surface,
we take the limit in (6.20). This leads to
()
1
U
=
1
()

d +
1
()( ) + 2Q[
() 1/U]
( )
2
+ 4Q
2
d. (6.41)
To proceed further we need to dene dimensionless variables. A natural
choice is to take H as the unit length and U as the unit velocity. In terms
of these dimensionless variables (6.41) becomes
() 1 =
1
()

d +
1
()( ) + 2[
() 1]
( )
2
+ 4
d.
(6.42)
We also write the dynamic boundary condition (6.11) in dimensionless form
as
1
2
1
2
+

2
+
1
F
2
_

0
()d
2
+

2
)
3/2
= B, (6.43)
where
F =
U
(gH)
1/2
(6.44)
is the Froude number and
=
T
HU
2
(6.45)
is the Bond number, already introduced in Section 2.4.2 (see (2.92)).
Relations (6.42) and (6.43) dene an integro-dierential system for

()
and

(). It was solved numerically by Hunter and Vanden-Broeck [76] for
pure gravity solitary waves. Another, similar, scheme was derived by Byatt-
Smith and Longuet-Higgins [25]. Here we will follow the approach in [76]
and present a numerical procedure to solve the general problem with both
gravity and surface tension included in the dynamic boundary condition.
I
= (I 1)h, I = 1, . . . , N, (6.46)
where h is the interval of discretisation. We also dene the midpoints
m
I
=

I
+
I+1
2
, I = 1, . . . , N 1. (6.47)
We satisfy (6.42) and (6.43) at the midpoints (6.47). This leads to 2N 2
nonlinear algebraic equations for the 2N+2 unknowns F, B,

(
I
) and

(
I
),
I = 1, . . . , N. The details of the discretisation follow those of Section
6.3. The integrals in (6.42) are approximated by the trapezoidal rule with
=
J
, J = 1, . . . , N, and =
m
I
, I = 1, . . . , N 1. The values of

(
m
I
),
(
m
I
),

(
m
I
) and

(
m
I
) needed to satisfy (6.43) at the mesh points (6.47)
are approximated by formulae similar to (6.25)(6.28). The integral in (6.43)
is calculated using (6.29)(6.33). Four more equations are needed. One is
obtained by imposing the symmetry condition
(
1
) = 0. (6.48)
Another equation xes the amplitude of the wave. Several choices are possi-
ble for this; for example the distance between the crest (or trough) and the
level of the free surface at innity can be xed by writing
_

0
()d = A, (6.49)
where A is given. Here A > 0 for an elevation wave (see Figures 6.4 and
6.5) and A < 0 for a depression wave (see Figure 6.6). An extra equation is
obtained by satisfying the dynamic boundary condition (6.43) as [x[ .
This leads to
B =
1
2

A
F
2
. (6.50)
There are also various possible choices for the nal equation. One possibil-
ity is to relate an unknown on the free surface to unknowns at neighbouring
points by an extrapolation formula. For example, we could express

(
M
)
in terms of

(
M1
) and

(
M2
) by a linear extrapolation formula. A
variation is to use an extrapolation formula that reects the fact that

decays exponentially as [[ (see [76]). The particular choice is not

crucial provided that the formulae used become exact in the limit h 0.
The resulting system of 2N + 2 equations with 2N + 2 unknowns is solved
by Newtons method.
Series truncation methods have also been used to compute solitary waves.
We will describe one of them in Section 6.6.1.
6.5 Numerical results for periodic waves
We describe in this section the numerical computation of nonlinear periodic
waves. We examine successively pure capillary waves, pure gravity waves
and gravitycapillary waves.
6.5.1 Pure capillary waves (g = 0, T ,= 0)
Numerical solutions for pure capillary waves can be computed using the
numerical procedures of Sections 6.2 and 6.3. However, Crapper [37] derived
an exact solution for pure capillary waves in water of innite depth. His
work was generalised to water of nite depth by Kinnersley [89]. A simpler
derivation of Kinnersleys solutions was provided by Crowdy [38].
Crappers exact solution can be written as
z =
f
c
+
4i
k
1
1 +Ae
2if/c

4i
k
, (6.51)
c
2
=
2T
1 A
2
1 +A
2
. (6.52)
The steepness s of the wave is related to the constant A by
s =
4A
(1 A
2
)
. (6.53)
Relation (6.52) is a dispersion relation. It reduces to the linear dispersion
(2.97) when A 0. The nonlinear dispersion (6.52) is also consistent with
the asymptotic solution (5.82). To check this, we expand (6.52) and (6.53)
in powers of A. This yields on the one hand
c
2
=
2T
_
1
s
2
2
8
_
+o(A
2
). (6.54)
On the other hand, taking the limits h , g 0 in (5.83) gives
c
2
=
c
0
A
2
1
k
2
16
. (6.55)
Using (5.43) with g = 0 and h , (5.48), (5.10) and (6.55) yields
c
2
=
2T
_
1

2
2
2
_
+o(
2
). (6.56)
The denition (5.46) of implies that = s/2 to leading order. Therefore
relation (6.56) agrees with (6.54).
Fully nonlinear free-surface proles are obtained by taking the real and
imaginary parts of (6.51). This gives a prole in the parametric form x =
x() and y = y(). Typical free-surface proles are shown in Figures 6.8
6.11.
0
0.2
0.4
0.6
0 0.2 0.4 0.6
Fig. 6.8. Crappers free-surface prole for k = 2. The steepness s = 0.1.
0
0.2
0.4
0.6
0 0.2 0.4 0.6
Fig. 6.9. Crappers free-surface prole for k = 2. The steepness s = 0.55.
For small values of the steepness s, the waves are close to linear sine waves
(see Figure 6.8). As s increases the waves develop more rounded crests and
0
0.2
0.4
0.6
0 0.2 0.4 0.6
Fig. 6.10. Crappers free-surface prole for k = 2. The steepness s = 0.73. There
is a small trapped bubble at the trough of the wave.
0
0.2
0.4
0.6
0.8
0 0.2 0.4 0.6 0.8
Fig. 6.11. Crappers free-surface prole for k = 2. The steepness s = 0.9. This
prole is unphysical because it is self-intersecting.
sharp troughs and ultimately become overhanging near the troughs (see
Figure 6.9). A limiting conguration is reached for s = s
0.73, when
the free-surface prole develops a point of contact with itself and a small
trapped bubble at the trough (see Figure 6.10). For s > s
, (6.51) predicts
unphysical self-intersecting free-surface proles (see Figure 6.11).
Vanden-Broeck and Keller [186] computed new physical solutions for s >
s
. They all have proles with a point of contact with themselves and a
small trapped bubble at their troughs, just like Crappers highest wave with
s = s
. As described at the beginning of Section 3.1.2.2, preventing the

self-intersection of the prole (see Figure 6.11) imposes an extra constraint
on the solutions. Therefore an extra unknown is needed. This is achieved by
nding the pressure P(s) in the trapped bubble as part of the solution. Such
solutions can be computed by an appropriate modication of the boundary
integral equation technique of Section 6.3. Details can be found in [186].
The pressure P(s) is found to be an increasing function of s, with P = 0 at
s = s
(see Figure 6.12).

1.0
0.5
P
s
0.5 1.0 1.5
Fig. 6.12. The bubble pressure P (in units of c
2
) versus the steepness s for the
family of waves computed by Vanden-Broeck and Keller [186]. Taken from J. Fluid
Mech. 98, 161169 (1980). The axis P = 0 corresponds to Crappers solution,
which is physically meaningful for 0 s 0.73 and unphysical for s > 0.73.
The waves exist for s
< s < s
. Here s
0.663. Some typical proles

of the waves and bubbles are shown in Figure 6.13
s = 0.6635
The bubble for s = 0.6635, expanded by a factor of 10
s = 0.7
s = 0.75
s = 1.05
s = 1.4
Fig. 6.13. Free-surface proles for the family of waves computed by Vanden-Broeck
and Keller [186]. Taken from J. Fluid Mech. 98, 161169 (1980).
The exact solutions of Kinnersley [89] are more complicated than Crap-
pers and involve elliptic integrals. Kinnersley solved the more general
problem of waves propagating on a sheet of nite thickness. The ow con-
guration is shown in Figure 6.14.
= Q
= 0
n
Fluid sheet
x
y
Fig. 6.14. A travelling wave on a uid sheet.
The uid sheet is surrounded by regions of constant pressure (e.g. air)
and gravity is neglected. Dimensionless variables are introduced by choosing
the wavelength as the unit length and c as the unit velocity. The solutions
are then characterised by two parameters. The rst is the value Q of the
streamfunction on the lower boundary of the sheet (see Figure 6.14). The
second is the parameter
W =
T
c
2
. (6.57)
Kinnersley obtained exact solutions for both symmetric waves (see Figure
6.15) and antisymmetric waves (see Figure 6.16).
The symmetric waves satisfy by symmetry the condition
y
= 0 on y =
0. Therefore they describe also capillary waves in water of nite depth
when the ows of Figure 6.15 are restricted to the domain y > 0. Figures
6.15 and 6.16 show that both symmetric and antisymmetric waves have
limiting congurations with trapped bubbles. Solutions for higher values of
the amplitude can be obtained by following the approach of Vanden-Broeck
and Keller [186] described earlier.
An interesting question is whether the solutions of Crapper [37] and Kin-
nersley [89] are unique. In other words, are there additional branches of
solutions? Blyth and Vanden-Broeck [20] showed that the answer is in fact
positive. They computed new branches of solutions that bifurcate from the
symmetric solution branches. Their numerical procedure follows that of Sec-
tion 6.2. Typical free-surface proles for these new solutions are shown in
Figures 6.176.19.
6.5.2 Pure gravity waves (g ,= 0, T = 0)
Using the asymptotic results of Section 5.1, Stokes [143], [144] noticed that
pure gravity waves tend to develop sharp crests and broad troughs as their
0.8
(a)
x x
0.4
0
0 0.5 1.0 1.5 2.0
0.8
0.4
(b)
0
0 0.5 1.0 1.5 2.0
1.2
0.8
0.4
(c)
0
0 0.5 1.0 1.5 2.0
1.2
0.8
0.4
(d)
0
0 0.5 1.0 1.5 2.0
Fig. 6.15. Symmetric waves for Q = 0.5: (a) W = 0.1013, (b) W = 0.1154,
(c) W = 0.129 49, (d) W = 0.141 03. Part (d) shows the limiting conguration
with trapped bubbles. Taken from J. Fluid Mech. 507, 255264 (2004).
amplitude is increased. Furthermore the heights of the crests increase with
the amplitude. It follows from (5.3) (with T = 0) that the velocity at the
crests decreases as the amplitude increases. This prompted Stokes to conjec-
ture that, as the amplitude increases, the waves ultimately reach a limiting
conguration with a stagnation point at their crests. Stokes then showed
that the ow near the crest of this limiting conguration is locally the same
as the ow inside a corner with an enclosed angle of 120
(see Figure 6.20).

This conclusion can easily be derived by noticing that the ow of Figure 6.20
is a particular case of the ow of Figure 3.38 for
2
= /6. Satisfying the
dynamic boundary condition on GL yields the exact solution (3.162) with
= 2/3 and A dened by (3.163), (3.164) and (3.176). In deriving this
solution in Chapter 3, we satised the dynamic boundary condition on GL
but not on HG. However, the symmetry of the ow of Figure 6.20 implies
that the dynamic boundary condition is then automatically satised on HG
by symmetry.
The existence of a limiting conguration with a stagnation point at
the crest suggests that we should characterise gravity waves by the
0.6
(a)
0.4
0.2
0
0 0.5 1.0 1.5 2.0
0.8 (b)
0.4
0
0 0.5 1.0 1.5 2.0
(c)
x
0.8
0.4
0
0 0.5 1.0 1.5 2.0
x
1.2
(d)
0.4
0.8
0
0 0.5 1.0 1.5 2.0
Fig. 6.16. Antisymmetric waves for Q = 0.5: (a) W = 0.1853, (b) W = 2.038,
(c) W = 0.2223, (d) W = 0.2426. Part (d) shows the limiting conguration with
trapped bubbles. Taken from J. Fluid Mech. 507, 255264 (2004).
parameter
0
= 1
1
c
2
(0)
(1/2)
(6.58)
instead of the steepness s (see [28] and [172]). Since 1/
(0) is the speed at

the crest of the wave,
0
= 1 for the limiting conguration. Furthermore,
0
0 as the steepness s tends to zero since 1/
c as s 0. Therefore
the advantage of the parameter
0
is that the range 0 <
0
< 1 of possible
values of
0
is known a priori. Similar parameters were used by Chen and
Saman [28] and Vanden-Broeck [162]. For example, Vanden-Broeck [162]
used the parameter
1
= 1
1
c
2
2
(0)
(6.59)
instead of the parameter
0
. It is easily checked that 0 <
1
< 1.
Numerical computations of the limiting congurations with a stagnation
point at the crest were performed as early as 1883 (see [112]). However, it
was not until 1982 that a proof of the existence of waves with a 120
angle
at their crest was provided (see [5]).
0.2
(a)
0.1
0
0 0.5 1.0 1.5
(c)
0.1
0
0 0.5 1.0 1.5
(d)
0
0 0.5 1.0 1.5
(b)
0.1
0
0 0.5 1.0 1.5
x x
Fig. 6.17. New free-surface proles for Q = 0.1: (a) W = 0.2783, (b) W = 0.2794,
trapped bubbles. The vertical scale has been exaggerated to show the trapped
bubbles clearly. Taken from J. Fluid Mech. 507, 255264 (2004).
We rst present numerical results for pure gravity waves in water of in-
nite depth. Results for pure gravity waves in water of nite depth will be
described after that.
6.5.2.1 Waves in water of innite depth
The asymptotic results (5.10), (5.76) and (5.83) predict that the speed of a
nonlinear gravity wave (i.e. ) is an increasing function of . A tempting
conjecture is then that is a monotonically increasing function of that
reaches its maximum value for the limiting conguration with a 120
angle.
However, this is not the case and it was shown in the 1970s (see [133],
[101] and [106]) that the values of oscillates innitely often as the limiting
conguration is approached. This is illustrated in Figure 6.21, where values
of versus
0
are shown.
The gure shows clearly the rst maximum. This maximum is followed
by an innite succession of minima and maxima as
0
1. This is a very
ne structure and only the rst maximum can be seen on the scale of Figure
6.21.
0.2
0.1
0
0 1.0 0.5 1.5
(a)
0.2
0.1
0
0 1.0 0.5 1.5
(b)
0.2
0.1
0
0 1.0 0.5 1.5
(d)
0.2
0.1
0
0 1.0 0.5 1.5
(c)
x x
Similarly it can be shown that the parameter used in the asymptotic
analysis of Chapter 5 oscillates innitely often as the wave of maximum
steepness is approached.
In Figure 6.22 we present values of the steepness s versus
0
. These
results show that the steepness s is a monotonically increasing function of
0
. Therefore s is a suitable parameter to describe steep waves since there is
only one solution for each value of s. However, it is not as convenient as
0
because the maximum steepness (and therefore the range of s) is not known
a priori.
Typical free-surface proles for
0
= 0.4, 0.6 and 0.99 are shown in Fig-
ure 6.23. The prole corresponding to
0
= 0.99 is close to the limiting
conguration with a 120
angle at the crest.

The results presented in Figures 6.216.23 can be computed by using
either of the two numerical procedures outlined in Sections 6.2 and 6.3.
The boundary integral equation method of Section 6.3 works well for
waves of small or moderate amplitude. However, the accuracy deteriorates
as the limiting conguration is approached and larger and larger numbers
0.2
(a)
0.1
0
0 0.25 0.50 0.75 1.00
(b)
x
0.2
0.1
0
0 0.25 0.50 0.75 1.00
0.2
(c)
0.1
0
0 0.25 0.50 0.75 1.00
x
(d)
0.2
0.1
0
0 0.25 0.50 0.75 1.00
(c) W = 3083, (d) W = 0.3176. Part (d) shows the limiting conguration with
6
2
3
G
x
y
L H
Fig. 6.20. The local ow near the crest of the highest pure gravity wave.
N of mesh points are required. This is due to the fact that the velocity at
the crest decreases and tends to zero as
0
1. Since the magnitude of the
velocity is [[, the equally distributed mesh points (6.21) or (6.39) lead to
fewer and fewer points near the crest of the wave in the physical (x, y)-plane
as
0
1. This is particularly bad for accuracy because the curvature at
the crest gets larger and larger as
0
1. These problems can be overcome
by concentrating mesh points near the crest. There are several ways to do
1.00
1.04
1.08
1.12
1.16
1.20
0 0.2 0.4 0.6 0.8 1.0
Fig. 6.21. Values of the parameter versus
0
.
0
0.04
0.08
0.12
0.16
0 0.2 0.4 0.6 0.8 1.0
Fig. 6.22. Values of the steepness s versus
0
.
this. Schwartz and Vanden-Broeck [135] used the change of variable
=
n
(6.60)
where n is an integer greater than 1. In their computations, they chose n = 3
and reformulated the problem in terms of . The problem was then solved
numerically by using points equally spaced in . The change of variables
(6.60) concentrates points near the crest in the physical plane and enables
computations for values of
0
very close to 1. Chen and Saman [28] used
another change of variable,
= c
_

2
sin 2
_
, (6.61)
0
0.2
0.4
0.6
0.8
0 0.2 0.4 0.6
Fig. 6.23. Typical free-surface proles for
0
= 0.4, 0.6 and 0.99 (respectively the
bottom, middle and top lines).
where the variable and the parameter

are between 0 and 1. They then
used equally spaced points in and concentrated mesh points near the crest
by choosing

very close to 1 (for example

= 0.99 or even

= 0.999). The
advantage of (6.61) over (6.60) is that variables that are periodic in , with
period c, are still periodic in , with period 1.
The series truncation method of Section 6.2 also gives accurate results for
waves of small and moderate amplitude. For the limiting conguration, the
ow near the crest at f = 0 is a ow inside a 120
angle, and (3.19) implies

that
w f
1/3
as f 0. (6.62)
Since the crests are mapped to t = 1 under the transformation (6.4), (6.62)
implies that
w (1 t)
1/3
as t 1. (6.63)
Because of the singularity of (6.63) at t = 1, we cannot expect the expansions
(6.16) or (6.17) to converge when
0
= 1 (i.e. for the limiting conguration).
Also, we can expect these expansions to converge more and more slowly as
0
1.
Michell [112], Olfe and Rottman [118], Vanden-Broeck [172] and Vanden-
Broeck and Miloh [188] calculated the wave of maximum steepness by writing
w = (1 t)
1/3
_
1 +
n=1
d
n
e
2inf
_
, (6.64)
where t is dened by
t = e
2if
.
The expansion (6.64) satises (6.63) and is therefore convergent. The phi-
losophy here is similar to that used in Chapter 3: the singularities of w in
the unit circle [t[ 1 are removed by writing w in the form (3.12), where
G(t) contains all the singularities of w. In this case
G(t) = (1 t)
1/3
,
since w is not analytic at t = 1. The expansion (6.64) leads to very accurate
results. In particular it predicts the value = 1.1931 for the dimensionless
wave speed.
Havelock [70] and Vanden-Broeck [172] calculated waves of arbitrary am-
plitude up to the limiting conguration by expressing w as
w = (1 t)
1/3
_
1 +
n=1
g
n
e
2inf
_
(6.65)
and nding the coecients g
n
and by series truncation and collocation.
By taking the cube of (6.17), (6.64) and (6.65), we see that these three
expressions are particular cases of
w
3
= 1 +
n=1
f
n
e
2inf
(6.66)
for appropriate choices of the coecients f
n
. Therefore gravity waves of
arbitrary amplitude (including the highest) can also be calculated by using
(6.66) and series truncation. This approach was used by Vanden-Broeck and
Miloh [188].
We conclude this section by asking whether there are other solution
branches. As in Section 6.5.1, the answer is positive. Chen and Saman
[28] discovered new branches that bifurcate from the solutions described
above. We will refer to the solutions described above as the basic branches.
The new solutions can be calculated as follows. We rst recompute the
basic branch by dening the wavelength as a multiple of the fundamental
wavelength (here the fundamental wavelength refers to the smallest value
of for which (5.4) and (5.5) are satised). We then choose as the unit
length. As an example, Figure 6.24 shows a solution with the wavelength
chosen as twice the fundamental wavelength. Since the wavelength is taken
as the unit length, the distance between every second crest is 1.
We then follow the solution branch associated, for example, with the solu-
tion of Figure 6.24 by the boundary integral equation method of
0
0.02
0.04
0 0.2 0.4 0.6 0.8 1.0
Fig. 6.24. Free-surface prole on the basic solution branch. The wavelength was
chosen as twice the fundamental wavelength. This solution corresponds to the
bifurcation point.
Section 6.3, using the solution in Figure 6.24 as the initial guess in the New-
ton iterations, together with continuation. This means that a previously
computed solution is used as the initial guess to compute a new solution
for a slightly smaller or larger value of the steepness. Following the general
discussion of Newtons method (see (3.66)(3.68)), we look for possible bi-
furcation points by monitoring the sign of the determinant of the Jacobian
matrix. Chen and Saman [28] identied such a bifurcation point, and the
solution shown in Figure 6.24 in fact corresponds to this bifurcation point.
The next step is to compute the bifurcating branches. Chen and Saman
[28] used a general numerical method developed by Keller [84] to move from
the solution of Figure 6.24 to the new branches. Here we use a simpler
approach, which consists of using the solution of Figure 6.24 as the initial
guess and replacing the amplitude equation (6.35) or (6.58) by
Y (0) Y (0.5) = , (6.67)
where Y () is dened by (6.29) and is given. Figures 6.25 and 6.26 show
two computed solutions, with > 0 and < 0 respectively. As [[ increases
the waves ultimately approach the limiting conguration with a 120
angle,
at x = 0 and x = 1 for > 0 and at x = 0.5 for < 0.
Chen and Saman [28] and Saman [130] computed further solutions by
starting with basic solutions in which the wavelength is n times the funda-
mental wavelength, for n > 2.
The existence of these new branches of waves is related to the presence of
the rst maximum, shown in Figure 6.21. This maximum implies that two
0
0.02
0.04
0 0.2 0.4 0.6 0.8 1.0
Fig. 6.25. Free-surface prole for a solution on the bifurating branch with > 0.
0
0.02
0.04
0 0.2 0.4 0.6 0.8 1.0
Fig. 6.26. Free-surface prole for a solution on the bifurating branch with < 0.
waves of dierent amplitudes can travel at the same speed. As mentioned in
Section 5.1.2, this phenomenon occurs for linear gravitycapillary waves. In
that case linear superposition leads to new solutions (see (5.89) and (5.90)).
Although linear superposition is not applicable for nonlinear solutions, the
new solutions of Figures 6.25 and 6.26 can be roughly interpreted in that
way. This suggests that further bifurcation points might be associated with
the innite sequence of maxima and minima of as
0
1. This was
conrmed by the calculations of Vanden-Broeck [162], who identied a bi-
furcation point, between the rst maximum and the rst minimum, leading
to solutions similar to those of Figures 6.25 and 6.26.
6.5.2.2 Waves in water of nite depth
All the computations and results presented in the previous subsection extend
to the case of nite depth. In particular, waves in water of nite depth also
have a limiting conguration with a 120
angle at their crests and values of

oscillating innitely often as
0
1.
For a given value of r
0
(i.e. of the depth), the waves approach the linear
waves of Section 2.4 as
0
0. However, a new feature of nite depth
is that for any given value of
0
(no matter how small) the waves deviate
from the linear waves of Section 2.4 as r
0
approaches 1 (i.e. as the depth
tends to zero). This behaviour is consistent with the nonuniformity of the
expansions of Section 5.1 as h 0 (see the end of Section 5.1). It is
also consistent with the asymptotic analysis of Section 5.2: the numerical
calculations show that the waves (with
0
small) approach the cnoidal waves
mentioned in Section 5.2 as r
0
1. For r
0
= 1, these waves become solitary
waves.
Vanden-Broeck [162] showed that the results of Chen and Saman [28]
can be extended to nite depth and that there are again solution branches
that bifurcate from the basic branches. These results can be described as
follows. Using the notation in [162] we characterise the amplitude of the
waves by the parameter
1
(see (6.59)) instead of
0
. In Figures 6.27 and
6.28 we present values of
=
c
2
c
2
0
(6.68)
versus
1
for r
2
0
= 0.4 and r
2
0
= 0.8. Here c is the nonlinear wave speed and
c
0
=
_
g
k
tanh kh
_
1/2
is the phase speed corresponding to linear waves (see (2.83)).
In these gures, the solid curves correspond to the basic solution branch
(see Figure 6.24 and the associated discussion for a denition of the basic
branch). The broken curves are the bifurcating branches. A typical free-
surface prole on the bifurcating branch for r
2
0
= 0.4 is shown in Figure
6.29.
6.5.3 Gravitycapillary waves (g ,= 0, T ,= 0)
The analysis of Section 5.1 illustrates the nonuniqueness of gravitycapillary
waves when the condition (5.58) is satised for some integer value of m.
0.24
0.23
0.22
0.85 0.90 0.95 1.0
1
Fig. 6.27. Values of the parameter
versus
1
for r
2
0
= 0.4. Taken from Phys.
Fluids 26, 23852387 (1983).
0.48
0.47
0.46
0.45
0.85 0.90 0.95 1.0
1
Fig. 6.28. Values of the parameter
versus
1
for r
2
0
= 0.8. Taken from Phys.
Fluids 26, 23852387 (1983).
For example, (5.106) and (5.107) predict two dierent solutions in water of
innite depth when m = 2. The numerical computations of Schwartz and
Vanden-Broeck [135], Chen and Saman [29] and Hogan [74] show that there
are indeed many dierent families of solutions. We illustrate these results by
considering rst gravitycapillary waves in water of innite depth. Results
for water of nite depth will be presented later.
6.5.3.1 Waves in water of innite depth
We use as the unit length and c as the unit velocity. The gravitycapillary
waves are characterised by the three parameters , (see (5.67) and (5.112))
0
0.2 0.4 0.6 0.8 1.0
0.01
0.02
0.03
0.04
y
x
Fig. 6.29. Computed free-surface prole for a wave on the bifurcating branch for
r
2
0
= 0.4. Taken from Phys. Fluids 26, 23852387 (1983).
and the steepness s. Using these parameters and assuming innite depth we
can rewrite the linear solutions (5.57) and (5.53) as
(x) =
s
2
cos 2x, (6.69)
= 1 +. (6.70)
These solutions hold for s suciently small provided that the condition
(5.55) is satised. We recall that in water of innite depth (5.55) takes the
simple form
,=
1
n
, (6.71)
where n > 1 is an integer.
When (6.71) is not satised for some value n = m, i.e. when
=
1
m
, (6.72)
the results of Section 5.1 shows that several solutions are possible. For
example when m = 2 (i.e. = 0.5), there are two solutions, described by
(5.106) and (5.108). In terms of the dimensionless variables, these solutions
can be written as
(x) =
s
2
cos 2x
s
4
cos 4x + , (6.73)
=
3
2

3
4
s (s)
2
+ (6.74)
The above results are valid only for s suciently small. We now extend
them by fully nonlinear calculations obtained by using the boundary integral
equation method of Section 6.3.
Figure 6.30 shows numerical values of versus for s = 0.03.
0.5
0.5
1.0
1.5
2.0
1.0
1
3
4
2
Fig. 6.30. Values of versus for s = 0.03. Taken from J. Fluid Mech. 95, 119139
(1979). The long-broken line corresponds to = 1 +.
These results show that there are many dierent families of solutions.
Four families, numbered 1 to 4, are shown in Figure 6.30. The long-broken
straight line corresponds to the linear solution (6.70). Relations (6.74) pre-
dict = 1.42 and = 1.56 for s = 0.03. The point with coordinates (0.5,
1.42) in Figure 6.30 is close to family 1 whereas the point with coordinates
(0.5, 1.56) is close to family 2. This illustrates that the Wilton ripples are
members of two dierent families of solutions (families 1 and 2). The corre-
sponding free-surface proles are shown in Figure 6.31. Using the notation
of Schwartz and Vanden-Broeck [135] we have plotted 2y versus 2x. Only
half a wavelength is shown in Figure 6.31. Solutions on the other branches
of solutions are characterised by a larger number of dimples on their proles.
This is clearly shown in Figure 6.32, where we present free-surface proles
for = 0.33 and s = 0.03. Further proles for other values of and s can
be found in [135].
0.10
2
1
0.05
0
Fig. 6.31. Free-surface proles for = 0.5 and s = 0.03. Taken from J. Fluid Mech.
95, 119139 (1979).
0.2
0.1
0
2
1
3
4
Fig. 6.32. Four free-surface proles for = 0.33 and s = 0.03. Taken from J. Fluid
Mech. 95, 119139 (1979).
In Figure 6.33 we plot values of versus for = 0.5. For small the
values of agree with (6.74) as expected.
6.5.3.2 Waves in water of nite depth
It is convenient to describe the properties of gravitycapillary waves in water
of nite depth in terms of the parameters , r
0
and F dened in (2.92),
(6.5) and (2.91) respectively. We dene dimensionless variables by taking
the undisturbed depth Q/c as the reference length and c as the reference
1.5
1.0
0.5
0.05 0.10 0.15
2
1
m
s
Fig. 6.33. Values of versus s for = 0.5. The solid curves correspond to the
numerical computation and the broken curves to the asymptotic solution (6.74).
Taken from J. Fluid Mech. 95, 119139 (1979).
velocity. In this section the amplitude of the waves is measured by the
dimensionless velocity u
0
at x = y = 0 (see Figure 6.1). Values of u
0
close
to 1 correspond to waves of small amplitude.
The condition (5.55) is satised when > 1/3. Periodic waves are then
unique and described for small values of s by the expansions of Section 5.1.1.
When < 1/3, the condition (5.55) is not satised for some values of the
parameters, and numerical calculations show the existence of many dierent
families of solutions. The situation is then similar to that described in the
previous section for waves in water of innite depth.
We examine in more detail the properties of waves with < 1/3 for r
0
close to 1. Figure 6.34 shows values of F versus
l =
2
ln r
0
(6.75)
for = 0.24 and u
0
= 0.97. Here l is the dimensionless wavelength. There
are many dierent families of waves and three of them are presented, as
(a),(b) and (c). Typical free-surface proles for each family are shown in
Figure 6.35.
As expected there are dimples on the free-surface proles, as was found
for water of innite depth. However, for values of r
0
close to 1 these dimples
tend to concentrate in the troughs of the waves. The results of Figure 6.35
show that as one moves to the right in Figure 6.34, from one family to the
next, two crests or two troughs are added to the proles. For example there
are 11 crests and 12 troughs per wavelength on the prole on Figure 6.35(a),
13 crests and 12 troughs per wavelength in the prole of Figure 6.35(b) and
1.010
(a) (b) (c)
F
1.008
1.006
30 32 34 36
l
Fig. 6.34. Computed values of the Froude number F versus the wavelength l for
periodic waves with = 0.24 and u
0
= 0.97. Taken from J. Fluid Mech. 134,
205219 (1983).
13 crests and 14 troughs per wavelength on the prole of Figure 6.35(c). A
most interesting question is what happens as one moves further to the right
of Figure 6.34 and considers the limit l (i.e. r
0
1). We shall see in
Section 6.6.2 that generalised solitary waves are obtained in this limit.
6.6 Numerical results for solitary waves
Many of the periodic solutions described in Section 6.5 have nontrivial limits
as . These limiting solutions are solitary waves, which belong to
one of the three types illlustrated in Figures 6.46.7 (i.e. solitary waves
with monotonic decay, solitary waves with decaying oscillatory tails and
generalised solitary waves.). No solitary waves were found for pure capillary
waves (i.e. T ,= 0, g = 0). In the next two sections, we explore the properties
of pure gravity solitary waves and of gravitycapillary solitary waves.
6.6.1 Pure gravity solitary waves
In Section 6.5.2.2, we mentioned that periodic gravity waves develop at
troughs for /h large. As /h these waves approach solitary waves.
For small values of the amplitude, they agree with the solution (5.156) of the
Kortewegde Vries equation. As the amplitude increases, the waves deviate
from (5.156) and ultimately approach the limiting conguration with a 120
angle at its crest. An existence proof for solitary waves of nite amplitude
was given by Friedrichs [61]. Numerical computations of solitary waves of
nite amplitude were performed by many investigators (see Schwartz and
y
(a)
0 5 10 15
0
y
(b)
0 5 10 15
0
y
0
x
(c)
0 5 10 15
Fig. 6.35. (a) Computed free-surface prole of a periodic wave with = 0.24 and
u
0
= 0.97. The wavelength is l = 32. (b) Same as (a) but with l = 34. (c) Same as
(a) but with l = 36.6. Taken from J. Fluid Mech. 134, 205219 (1983).
Fenton [134] for a review and Tanaka et al. [149] for more recent references).
Here we follow mainly the numerical approach of Hunter and Vanden-Broeck
[77] (see also Section 6.4.1).
It is found that there is a one-parameter family of solutions. This param-
eter can be chosen as the quantity A dened in (6.49). However, a better
choice is
= 1
F
2
2
(0)
, (6.76)
where F is dened by (6.44). This choice is similar to (6.58) and (6.59).
Since 1/
(0) is the dimensionless speed at the crest, = 1 for the limiting

conguration. Also,

(0) 1 and F 1 as A 0. Therefore (6.76) implies
that 0 as A 0. The advantage of the parameter over A is that the
range of possible values 0 < < 1 is known a priori.
We present in Figure 6.36 numerical values of F versus for waves close
to the limiting conguration (i.e. for close to 1).
a b
c
x
F
1.295
1.290
1.285
0.80 0.85 0.90 0.95 1.00
Fig. 6.36. Values of F versus . Taken from J. Fluid Mech. 136, 6371 (1983).
The curve a corresponds to numerical values obtained by the method
described in Section 6.4.1. It shows that F is not a monotonic function of
for close to 1. In fact there is an innite number of maxima and minima
as 1 (only the rst maximum is shown in Figure 6.36). This behaviour
is similar to that already encoutered for periodic waves (see Figure 6.21).
The curves c and b in Figure 6.36 correspond to earlier results obtained
by Longuet-Higgins and Fenton [105] and by Byatt-Smith and Longuet-
Higgins [25]. These authors also found the rst maximum. A comparison of
the curves b and c with the curve a show that the solutions of Byatt-Smith
and Longuet-Higgins are more accurate than those of Longuet-Higgins and
Fenton. The cross in Figure 6.36 corresponds to the limiting conguration,
which will be calculated by series truncation in what follows.
6.6.1.1 A series truncation method for gravity solitary waves
We now describe how the series truncation method is used to compute soli-
tary waves up to the limiting conguration. The ow is shown in Figure
6.37. We introduce cartesian coordinates such that the level of the free
y
x
0 1.0 2.0
Fig. 6.37. The ow and of the coordinates. The prole is the computed free surface
of the highest solitary wave. Taken from J. Fluid Mech. 136, 6371 (1983).
surface in the far eld corresponds to y = 0.
As previously, we introduce the complex potential f = + i and the
complex velocity w = u iv. We choose = 0 at the crest and = 0 on
the free surface. As [x[ , the ow approaches a uniform stream char-
acterised by a constant velocity U and a constant depth H. We introduce
dimensionless variables by taking U as the unit velocity and H as the unit
length. On the free surface the Bernoulli equation yields
u
2
+v
2
+
2
F
2
y = 1, (6.77)
where F is dened by (6.44).
We map the ow domain in the region [t[ < 1 of the complex t-plane by
the transformation
f =
2
ln
1 +t
1 t
i. (6.78)
The transformation (6.78) maps the bottom of the channel onto the real
diameter 1 < t < 1 and the free surface onto the half-circumference [t[ = 1
in the upper half of the t-plane. We will use the notation t = re
i
, so that
the free surface is described by r = 1 and 0 < < .
Hunter and Vanden-Broeck [77] calculated the highest solitary wave by
representing the complex velocity w by the expansion
w =
_
1 +t
2
2
_
1/3
exp
_
A(1 t
2
)
2
+
n=1
a
n
(t
2n
1)
_
, (6.79)
where is the smallest positive root of

tan
F
2
= 0. (6.80)
The term A(1t
2
)
2
in (6.79) comes from the fact that the complex velocity
on the free surface satises
u iv 1 +Ae
||
as [[ . (6.81)
The constants a
n
and A must satisfy the free-surface condition (6.77).
The analysis leading to (6.65) and (6.66) shows that solitary waves of ar-
bitrary amplitude (including the highest) can be calculated by representing
w
3
as follows:
w
3
= 1 + (1 t
2
)
2
n=1
a
n
t
2n2
. (6.82)
Alternatively, we can set
w =
_
1 +t
2
1 +
_
1/3
exp
_
A(1 t
2
)
2
+
n=1
a
n
(t
2n
1)
_
. (6.83)
Here is another constant to be found as part of the solution. The expansion
(6.79) is similar to the Michell expansion for the periodic wave of largest
steepness (6.66) and was used before by Lenau [96]. The factor (1 + t
2
)
1/3
removes the singularity associated with the 120
angle at the crest of the

highest wave. The expansions (6.82) and (6.83) are equivalent to the Davies
approximation [41] and the Havelock expansion [70] (see Vanden-Broeck and
Miloh [188]).
As shown by Vanden-Broeck and Miloh [188], equivalent numerical results
are obtained by using (6.82) or (6.83). Here we describe results based on
(6.83).
First we dierentiate (6.77) with respect to . Then using (6.78) we obtain
F
2
[u()u
() +v()v
()]
2
v()
u
2
() +v
2
()
1
sin
= 0. (6.84)
We will characterise the amplitude of the solitary wave by the parameter
dened in (6.76).
We now truncate the innite series in (6.82) after N terms and determine
the N +4 unknowns a
1
, a
2
, . . . , a
N
, A, , and F
2
by collocation. Thus we
introduce the N + 2 collocation points
I
=
E
2
+ (I 1)E, I = 1, . . . , N + 2, (6.85)
where E = /(2N + 4), and satisfy (6.84) at the collocation points (6.85).
This yields N +2 nonlinear algebraic equations. The nal two equations are
given by (6.80) and (6.76). Thus for a given value of we have a system of
N + 4 equations with N + 4 unknowns. This system is solved by Newton
iteration. In some calculations, it is more convenient to use a variant of the
scheme in which is xed and is found as part of the solution.
As an example we present numerical results for the highest wave. Then
we x = 1 (or = 1 if we use the above-mentioned variant). The scheme
is then equivalent to that used by Hunter and Vanden-Broeck [77]. We note
that for the highest wave (for which u = v = 0 at the crest), the elevation
of the crest is related to F by the simple relation
F
2
=
2
(6.86)
(see 6.77). Hunter and Vanden-Broeck [77] computed solutions for values of
N 100 and concluded that = 0.833 22. This estimate can be improved by
increasing N. They found the values 0.833 217 02, 0.833 208 77, 0.833 205 44,
0.833 203 15 and 0.833 202 00 for N = 200, 400, 600, 900 and 1200 respec-
tively. An extrapolation for N gives the value 0.833 198 6. This value
is in close agreement with the value 0.833 197 obtained by Williams [197]
and the value 0.833 199 79 obtained by Evans and Ford [53]. The computed
prole of the highest solitary wave is shown in Figure 6.37.
6.6.2 Gravitycapillary solitary waves
The numerical results of Section 6.5.3.2 suggest that periodic gravity
capillary waves with < 1/3 approach generalised solitary waves as l
(see Figure 6.35). To conrm this idea we repeated the calculations of Fig-
ure 6.35 for larger values of l (i.e. for r
0
closer to 1) and for various values
of u
0
and . The calculations follow closely Vanden-Broeck [177].
In Figure 6.38, we present values of F versus l for = 0.24 and u
0
= 0.99.
Curves (a) and (b) correspond to two computed families of solutions; cor-
responding proles of the waves for these two families are shown in Figures
6.39 and 6.40 respectively.
Comparing the proles in Figures 6.39 and 6.40, we see that the number
of inexion points (and therefore the number of ripples) increases as one
moves in Figure 6.38 from one family to another family further to the right.
Furthermore it follows from the symmetry of the ow that the last point
on the graphs in Figures 6.39 and 6.40 is either a crest or a trough of the
ripples. These facts indicate that, for l suciently large, there is an innite
96
1.002
1.004
1.006
1.008
1.010
104 112 120
(a)
(b)
(c)
128
l
F
Fig. 6.38. Values of the Froude number F versus the dimensionless wavelength l
for periodic waves with = 0.24 and u
0
= 0.99. The solid curves (a) and (b) are
the numerical solutions. The broken curves show families obtained by translation.
Taken from Phys. Fluids A 3, 263266 (1991).
number of families of solutions. The corresponding curves in Figure 6.38
can be obtained from any particular curve by translating it horizontally
by a multiple of the wavelength of the ripples. Generalised solitary waves
are then obtained by jumping from one curve to the next as we take the
limit l . After each jump, two more crests or troughs appear on each
wavelength of the proles, one on the right and one on the left. In the limit
l , we obtain a generalised solitary wave with innite trains of ripples
in the far eld. For each value of 0 < < 1/3, these generalised solitary
waves form a two-parameter family of solutions. The numerical results in
Figures 6.386.40 are for periodic waves with l > 90. We will use these long
waves to approximate generalised solitary waves.
The proles in Figures 6.39 and 6.40 show that the steepness of the ripples
(i.e. the dierence in height between a crest and a trough of the ripples
divided by their wavelength) is small for all the proles presented. Therefore
the dimensionless wavelength L of the ripples satises the linear dispersion
relation
F
2
=
L
2
_
1 +
4
2
L
2
_
tanh
2
L
(6.87)
(see (2.90)).
Starting with curve (a) in Figure 6.38, we used (6.87) to calculate the
wavelength L corresponding to each value of F. Then we performed the
translations mentioned earlier. The resulting curves are the broken lines in
(c)
0 10 20 30 40 50
0.004
0.002
0.000
0.006
0.008
0.010
(d)
0 10 20 30 40 50
0.004
0.002
0.000
0.006
0.008
0.010
(a)
0 10 20 30 40 50 60
0.000
0.002
0.004
0.006
(b)
0 10 20 30 40 50
0.004
0.002
0.000
0.006
0.008
0.010
(e)
0 10 20 30 40 50
0.010
0.005
0.000
0.015
0.020
0.025
Fig. 6.39. (a) Computed free-surface proles for a periodic wave with = 0.24,
u
0
= 0.99 and F = 1.0005. Only half a wavelength is shown. (b) Same as (a) but
with F = 1.0035. (c) Same as (a) but with F = 1.003 58. (d) Same as (a) but with
F = 1.0036. (e) Same as (a) but with F = 1.008. Taken from Phys. Fluids A 3,
263266 (1991).
Figure 6.38. We note that the 12th broken line coincides within graphical
accuracy with curve (b). This constitutes a check on our calculations.
The solution corresponding to the prole in Figure 6.39(a) is very close
to a train of periodic waves extending from x = to x = . This shows
that the generalised solitary waves bifurcate from a train of periodic waves.
As one moves away from the bifurcation point, one crest of the train of
periodic waves is progressively lifted.
(c)
0 10 20 30 40 50 60 70
0.004
0.002
0.000
0.006
0.008
0.010
(b)
0 10 20 30 40 50 60 70
0.004
0.002
0.000
0.006
0.008
0.010
(a)
0 10 20 30 40 50 60 70
0.004
0.002
0.000
0.006
0.008
0.010
Fig. 6.40. (a) Computed free-surface proles for a periodic wave with = 0.24,
u
0
= 0.99 and F = 1.0038. Only half a wavelength is shown. (b) Same as (a) but
with F = 1.003 87. This prole is close to that corresponding to the right-hand
cross in Figure 6.38. (c) Same as (a) but with F = 1.003 95. Taken from Phys.
Fluids A 3, 263266 (1991).
Figure 6.39 shows that the amplitude of the ripples rst decreases and
then increases as one moves along the of solution branch. There is one
point on each curve in Figure 6.38 for which the amplitude of the ripples
is a minimum. These points are indicated by the crosses in Figure 6.38,
and the prole corresponding to the cross on the left is shown in Figure
6.39(c). Within graphical accuracy, the amplitude of the ripples in Figure
6.39(c) is zero and the corresponding solution looks like a true solitary wave
(i.e. a wave without ripples in the far eld). However, the calculations of
Champneys et al. [26] showed that the amplitude of the ripples is always
dierent from zero (although it can be very small) for 9/50 < < 1/3.
This was also shown theoretically by Sun [146] for suciently close to 1/3.
One feature of the ripples is that their amplitude is an exponentially small
function of F 1; this was shown by exponential asymptotics [147] and
by application of centre-manifold and normal-form theory [97]. For further
references see the review [43].
We conclude this chapter by comparing our numerical results with the pre-
dictions of the Kortewegde Vries equation (5.154). The curve (c) in Figure
6.38 corresponds to the periodic solutions of (5.154) (i.e. the cnoidal waves).
As l , curve (c) approches the solitary-wave solution. The solutions of
the Kortewegde Vries equation (see curve (c)) do not have ripples in the far
eld. However, the above discussion shows that the numerical solutions of
the complete nonlinear problem (curves (a) and (b) and the broken curves)
always have ripples of nonzero amplitude in the far eld.
Figure 6.38 shows that the numerical solutions of the complete nonlinear
problem (curves (a) and (b) and the broken curves) intersect curve (c) at
a discrete set of points. The intersections with curves (a) and (b) are very
close to the crosses. Therefore, for u
0
close to 1 the solutions for which
the amplitude of the ripples is a minimum are closely approximated by the
cnoidal wave solutions, although ripples are not predicted for the cnoidal
waves.
7
Nonlinear free-surface ows generated
by moving disturbances
We now return to the free-surface ow generated by a moving disturbance
and extend the results of Chapter 4 to the nonlinear regime. We shall
see that the wave trains in the far eld (if they exist) are then described
by the nonlinear theories of Chapters 5 and 6. Furthermore we will show
that the nonuniformities of Figures 4.9 and 4.8 are removed when a nonlinear
theory is used. Some nonlinear solutions described in this chapter approach
the linear solutions of Chapter 4 as the size of the disturbance approaches
zero, while others approach solitary waves.
We have organised the results in the following way. In Section 7.1 we
present pure gravity free-surface ows (i.e. g ,= 0, T = 0) in water of
nite depth and show in Section 7.1.1 that the nonuniformity of the lin-
ear supercritical solutions near F = 1 (see Figure 4.9) is removed when a
nonlinear theory is used. Subcritical ows are considered in Section 7.1.2.
In Section 7.2 we consider gravitycapillary free-surface ows. Solutions in
water of nite depth are described in Section 7.2.1. We show in Section
7.2.2 that the nonuniformity of the linear theory near = 0.25 (see Figure
4.8) is removed when a nonlinear theory is used. We examine in Section 7.3
the implications of the existence of multiple branches of periodic gravity
capillary waves (see Sections 5.1.2 and 6.5.3) for free-surface ows generated
by moving disturbances in water of innite depth.
Qualitatively similar numerical results were obtained for the three types
of disturbance shown in Figures 4.14.4. Therefore, in each section or sub-
section we will restrict our attention to one type of disturbance (due to a
pressure distribution in Sections 7.1.1, 7.2.2 and 7.3 and due to a submerged
object in Sections 7.1.2 and 7.2).
191
192 Nonlinear free-surface ows generated by moving disturbances
7.1 Pure gravity free-surface ows in water of nite depth
7.1.1 Supercritical ows
We will assume that the disturbance is a pressure distribution, and we will
neglect the eect of surface tension. The ow conguration is shown in
Figure 4.4. The fully nonlinear problem is described by equations (4.1)
(4.4) with T = 0. In this section we restrict our attention to supercritical
free-surface ows. Therefore the ow approaches a uniform stream with
constant velocity U and constant depth H in the far eld. We dene the
Froude number
F =
U
(gH)
1/2
. (7.1)
Since the ow is assumed to be supercritical, F > 1. We introduce di-
mensionless variables by taking U as the reference velocity and U
2
/g as
the reference length. In terms of the dimensionless variables, the dynamic
boundary condition (4.3) (with T = 0) becomes
1
2
(
2
x
+
2
y
) +y +
P(x)
U
2
=

B on y = (x). (7.2)
Here

B = B/U
2
. It follows from the choice of the origin of y that

B = 1.
In order to compare the nonlinear computations with the linear results
of Chapter 4, we assume that the pressure distribution is given by (4.70).
Therefore, using the dimensionless variables we write
P(x)
U
2
=
e
5x
2
2
. (7.3)
Since P(x) = P(x), the solutions are symmetric with respect to x = 0.
We solve the problem numerically by using a boundary integral equation
method similar to those used in Sections 3.1.2.2 and 6.4.1. We rst dene
the complex potential f = +i. We choose = 0 on the free surface and
= 0 at the point x = 0 on the free surface. The value Q of on the
bottom satises
Q =
1
F
2
, (7.4)
where F is dened by (7.1). Next we dene the complex function + i
by (6.3). We seek + i as an analytic function of the complex potential
f = +i. We satisfy the kinematic boundary condition = 0 on = Q
by reecting the ow in the bottom surface. Therefore we seek +i as an
analytic function in the strip 2Q < < 0. The values of (, ) and of
(, ) on the free surface = 0 and on its image = 2Q are related by
(, 2Q) = (, 0) and (, 2Q) = (, 0). (7.5)
We derive an integral equation on the free surface by applying the Cauchy
integral formula to the function 1 + i in the complex f-plane with a
contour consisting of the free surface, its image and two vertical lines at
= . Since 1 +i 0 as , there are no contributions from
the two vertical lines at = . The contributions from the free surface
and its image give, after taking the real part and using (7.5),
() 1 =
1
()

d +
1
()( ) + 2Q[
() 1]
( )
2
+ 4Q
2
d.
(7.6)
Here

() = (, 0) and

() = (, 0).
We can rewrite (7.2) in terms of

and

as
1
2
+

2
+ 2y + 2p() = 1, (7.7)
where
p() =
e
5
2
2
(7.8)
and
y =
_

()d. (7.9)
Comparing (7.3) and (7.8), we see that P(x) has been replaced by p(). This
change in the nonlinear dynamic boundary condition is compatible with the
linear theory of Chapter 4, since = x + O() (see (4.7) and (4.8)) implies
that p() = p(x) + O(
2
) as 0. Therefore the nonlinear problem
dened by (7.6) and (7.7) reduces formally to that of Chapter 4 in the limit
0.
I
= A+ 2A
I 1
N 1
, I = 1, . . . , N, (7.10)
the midpoints
m
I
=

I
+
I+1
2
, I = 1, . . . , N 1, (7.11)
and the unknowns
I
=

(
I
),
I
=

(
I
), I = 1, . . . , N. (7.12)
We now obtain 2N 2 equations by satisfying (7.6) and (7.7) at the mid-
points (7.11). The details of the discretisation follow that used in Sections
6.3 and 6.4.1. We force the free surface to be at in the far eld by imposing
the conditions
1
=
N
= 0. (7.13)
This gives two more equations. The nal equation xes the elevation of the
free surface at = 0:
_

0
()d = a, (7.14)
where a is given. This system of 2N +1 nonlinear equations for the 2N +1
unknowns
I
,
I
and Q is solved by Newtons method. We note that after
the value of Q has been calculated, the corresponding value of F can be
deduced from (7.4).
Numerical values of a versus F are shown in Figure 7.1 for = 0.001. The
corresponding linear results are shown in Figure 4.9. Figure 7.1 shows that
the nonuniformity of the linear theory (i.e. the vertical asymptote in Figure
4.9) has been removed in the nonlinear calculations: all the values of a are
now nite.
0
0.01
0.02
0.03
0.04
1 1.01 1.02 1.03
Fig. 7.1. Values of a versus F for = 0.001.
The curve of Figure 7.1 has a turning point at F = F
t
1.004. For
F > F
t
, the lower portion of the curve is close to that predicted by the
linear theory of Chapter 4. The corresponding solutions are perturbations
of a uniform stream. This can be checked in the following way. Select a
solution corresponding to a point on the lower part of the curve of Figure
7.1 (let us say for F = 1.02). Then use this solution as an initial guess to
compute a new solution for F = 1.02 and a value
1
of slightly smaller than
0.001. Then use this solution as an initial guess to compute a new solution
for a value of slightly smaller than
1
, and so on. Ultimately we obtain a
solution corresponding to = 0. This solution is a uniform stream.
We shall use this procedure, of computing solutions for a decreasing se-
quence of values of , several times in this chapter and in those that follow.
We shall refer to it as a continuation in (or in whatever parameter we are
using).
An interesting question is what happens if, using continuation in , we
start with a solution on the upper portion of the curve of Figure 7.1. The
answer is that the resulting solution for = 0 is a solitary wave. Soli-
tary waves form a solution branch bifurcating from F = 1 in Figure 7.1
(see Chapter 5 for analytical solutions for solitary waves of small amplitude
and Chapter 6 for numerical computations of solitary waves of arbitrary
amplitude). These solitary waves are solutions of the present problem cor-
responding to = 0. The upper portion of the curve of Figure 7.1 is a
perturbation of that branch of solitary waves.
Our ndings can be summarised as follows. For = 0, the solution in
Figure 7.1 consists of the F-axis (a uniform stream) and of the branch
of solitary waves bifurcating from the F-axis at F = 1 and extending for
F > 1. For ,= 0, the curve in Figure 7.1 has a turning point at F =
F
t
> 1 (the value of F
t
depends on ). The solutions corresponding to the
lower portion of the curve are perturbations of a uniform stream, while the
solutions corresponding to the upper portion of the curve are perturbations
of solitary waves.
In Figure 7.2 we present two free-surface proles corresponding to F =
1.02 and = 0.001. The prole of lower amplitude is a perturbation of a
uniform stream. The prole of larger amplitude is a perturbation of a soli-
tary wave. Results similar to those presented in this section were obtained
in [173], [45] and [48] for other disturbances.
7.1.2 Subcritical ows
We now consider subcritical ows (i.e. we assume that F < 1). Although
the results could be described by using the pressure distribution of Sec-
tion 7.1.1, we will choose to present them for another type of disturbance,
namely a submerged semicircular obstacle on the bottom of a channel (see
Figure 7.3). We introduce cartesian coordinates with the x-axis along the
bottom and the origin at the centre of the semicircular obstacle. The accel-
eration of gravity g is acting in the negative y-direction. The linear theory of
Chapter 4 indicates that waves should be expected on the free surface as
0
0.01
0.02
0.03
0.04
0 10 20
Fig. 7.2. Two free-surface proles generated by a moving pressure distribution for
= 0.001 and F = 1.02. The lower prole is a perturbation of a uniform stream
while the upper prole is a perturbation of a solitary wave.
x . As x , the ow approaches a uniform stream with a constant
velocity U and constant depth H. We introduce dimensionless variables by
taking H as the unit length and U as the unit velocity.
H
x
y
U
Fig. 7.3. The ow conguration. When surface tension is neglected, as in the gure,
the free surface is at as x and wavy as x . If surface tension were
taken into account, there would be in addition a train of waves as x .
We then dene the following complex variables, z = x + iy, the complex
potential f = + i and the complex velocity u iv. Without loss of
generality we choose = 0 on the bottom, so that = 1 on the free surface.
The condition of constant pressure on the free surface can be written as
1
2
F
2
(u
2
+v
2
) +y =
1
2
F
2
+ 1. (7.15)
Here F is the Froude number dened in (7.1). Following Forbes and Schwartz
[59] we dene the new variable
= +i (7.16)
by the relation
+i =
1
2
_
z +

2
z
_
, (7.17)
where = R/H is the dimensionless radius of the semicircle.
Relation (7.17) is the classical Joukowskii transformation; it maps the
z-plane into the -plane, in which the bottom streamline is a straight line.
We shall seek + i as an analytic function of f = + i in the strip
0 < < 1. In terms of the variable (, ), (7.15) becomes
F
2
(z
2
2
)(z
2
2
)
8(zz)
2
+z
1
2
F
2
1 = 0, (7.18)
where an overbar signies the complex conjugate and denotes the imagi-
nary part.
By using the Cauchy integral equation formula, Forbes and Schwartz [59]
derived the following integral relation between (, 1) and (, 1):
_
(, 1)
1
2
_
_
+
(, 1)
1
2
_
d
( )
2
+ 4
=
1
__
+
(, 1)( )
( )
2
+ 4
d +
_
+
(, 1)
( )
d
_
. (7.19)
Their derivation is similar to that leading to (7.6), therefore the details will
not be repeated here. The second integral on the right-hand side of (7.19)
is a Cauchy principal value.
This concludes the formulation of the problem. For given values of and
F, we seek two functions and satisfying (7.18) and (7.19).
The linear theory of Chapter 4 indicates that waves should be expected
on the free surface. Furthermore the radiation condition requires the waves
to be behind the obstacle and the free surface to be at as x . This
suggests that an easy way to impose the radiation condition in the nonlinear
computations is to truncate the integrals from to in (7.19) at some
nite values t
1
and t
2
and force the free surface to be at at = t
1
.
This approach was used sucessfully by Forbes and Schwartz [59], Vanden-
Broeck [158], Mekias and Vanden-Broeck [110] and others. A more accurate
approach is to approximate the integrals from to t
1
and from t
2
to
by asymptotic solutions for [[ large. These two approaches are described
for the ow conguration of Figure 7.3 in Sections 7.1.2.1 and 7.1.2.2.
7.1.2.1 Results obtained using truncation of the integrals
The numerical method follows closely the work of Forbes and Schwartz [59],
Forbes [55] and Grandison and Vanden-Broeck [65]. The reader is referred
to those papers for details.
We rst truncate the three integrals from to in (7.19) to integrals
from t
1
to t
2
, where t
1
and t
2
are large positive numbers. Then we dene
equally spaced points
I
= t
1
+
t
2
+t
1
N
I, I = 0, . . . , N,
and corresponding unknowns
I
=
(
I
, 1), I = 0, . . . , N,
I
=
(
I
, 1), I = 0, . . . , N.
The problem is then discretised by following the approach used in Section
7.1.1. Relations (7.18) and (7.19) yield a system of nonlinear algebraic equa-
tions for the unknowns
I
and
I
. The radiation condition in the absence of
surface tension is equivalent to the requirement that the free-surface prole
is at as x . This is imposed by forcing = 1/2 and
= 0 at the
rst mesh point
0
. The system of nonlinear algebraic equations is solved
by Newton iteration.
The above numerical scheme gives very good results (see Forbes and
Schwartz [59] for examples). However, as noted by Forbes and Schwartz,
spurious small waves appear in front of the obstacle (i.e. in the region
x < 0). These waves do not satisfy the radiation condition and are an arti-
fact of the numerical procedure. There is also a distortion in approximately
the last quarter wavelength of the downstream waves, which, although in-
signicant, can prevent the numerical method from converging for values of
the Froude number close to 1.
We present below a numerical method which improves the accuracy, re-
moves the spurious waves and reduces the distortion.
7.1.2.2 Results obtained without using truncation of the integrals
The inaccuracies in the scheme of Section 7.1.2.1 are caused by the replace-
ment of the integrals from to in (7.19) by integrals from = t
1
to
= t
2
.
We shall show that these inaccuracies are greatly reduced by includ-
ing suitable approximations of the integrals from to t
1
and from t
2
to .
We consider rst the integrals from to t
1
. As , the
ow approaches a uniform stream with velocity 1. An asymptotic solu-
tion describing a waveless perturbation of a uniform stream was derived in
Section 3.3 (see formulae (3.210) and (3.211)). Using (3.210), (3.211) and
the properties x and y 2 as we obtain
Ae
as , (7.20)
where A is a constant and is the smallest positive root of

tan
F
2
= 0. (7.21)
We also have
C as , (7.22)
where C is a constant.
We now use (7.22) and (7.20) to approximate the three integrals from
to t
1
as follows:
_
t
1
(, 1)
1
2
_
d
( )
2
+ 4

_
C
1
2
__
1
2
arctan
(t
1
+)
2
+

4
_
,
(7.23)
_
t
1
(, 1)( )
( )
2
+ 4
d A
_
t
1
( )
( )
2
+ 4
d, (7.24)
_
t
1
(, 1)

d A
_
t
1

d. (7.25)
The constants A and C are found by requiring that (7.20) and (7.22) are
satised at the last mesh point
N
.
We now consider the integrals fromt
2
to . This is more complicated than
the previous case, because there are nonlinear waves on the free surface for
> 0. However, the waves ultimately approach a train of periodic waves as
. Therefore we can assume that the waves repeat themselves without
change of shape or amplitude for > t
2
. This is a good approximation
when t
2
is large. We dene at each iteration two values p
1
< p
2
of that
correspond to the two crests of the wave trains closest to = t
2
(i.e. the
two crests further to the right). We dene
for > t
2
by
(t
2
+s) =
(t
2
+p
1
p
2
+s) for s < p
2
p
1
(7.26)
and, more generally,
(t
2
+s) =
(t
2
+p
1
p
2
+s n(p
2
p
1
))
for n(p
2
p
1
) < s < (n + 1)(p
2
p
1
). (7.27)
0 4 8
0.97
0.98
0.99
1.00
1.01
12
Fig. 7.4. Free-surface ow for = 0.2, F = 0.4.
Here s > 0 and n is a positive integer.
Similarly, we write
(t
2
+s) =
(t
2
+p
1
p
2
+s n(p
2
p
1
))
for n(p
2
p
1
) < s < (n + 1)(p
2
p
1
). (7.28)
Relations (7.26)(7.28) dene
and
for > t
2
in terms of their values
for < t
2
. This enables us to extend the trapezoidal-rule (or Simpson-rule)
approximation of the integrals to values of much larger than t
2
without
introducing extra unknowns
I
and
I
or increasing signicantly the com-
puting time. Figure 7.4 shows a typical free-surface prole. We note the
elimination of waves upstream of the disturbance. All the results presented
were obtained with N = 300 mesh points. We checked that these results are
correct within graphical accuracy by varying N.
Making the above downstream approximation removes the distortion in
the last quarter wavelength experienced by Forbes and Schwartz [59] and
allows us to compute solutions for ow speeds greater than those used by
these authors. Convergence was obtained for values of the Froude number
up to F = 0.555, for = 0.2 (see Figure 7.5).
The prole of Figure 7.5 contains nonlinear waves with sharp crests and
broad troughs. This prole demonstrates that the radiation condition is
eectively satised and that there is no downstream distortion.
0 2 4 6
0.95
1.00
1.05
1.10
Fig. 7.5. Free-surface ow for = 0.2, F = 0.555.
7.2 Gravitycapillary free-surface ows
In this section, we include the eects of both the gravity g and the sur-
face tension T. We will consider solutions for water of nite depth in
Section 7.2.1. Results for innite depth are presented in Section 7.2.2.
7.2.1 Results in nite depth
We generalised the calculations of Section 7.1.2 by including the eect of sur-
face tension. The ow conguration is sketched in Figure 7.3. The dynamic
boundary condition (7.15) now becomes
1
2
F
2
(u
2
+v
2
) +y +WK =
1
2
F
2
+ 1, (7.29)
where K is the curvature at the free surface and W is the Weber number,
dened by
W =
T
gH
2
.
Following Forbes [56] we nd that (7.29) can be replaced by
1
8
F
2
(z
2
2
)(z
2
2
)
(zz)
2
+[z]
1
2
F
2
1
1
4
iW
(z
2
2
)
1/2
(z
2
2
)
1/2
(zz)(
)
3/2
_
(
) + 4
2
_
z
(z
2
2
)
2

z
(z
2
2
)
2
__
= 0. (7.30)
This expression reduces to (7.18) when W = 0.
The linear theory of Chapter 4 shows that in general there are wave trains
both upstream and downstream of the disturbance. We use the method
described in Section 7.1.2 to approximate the integrals from t
2
to +; the
method used in Section 7.1.2 to approximate the integral from to t
1
is no longer applicable because of the upstream train of waves. Here we
use the approach developed by Vanden-Broeck [182] and Grandison and
Vanden-Broeck [65]. It consists in approximating
for < < t

1
by
a linear wave. We will assume in this section that the condition (5.55) is
satised (solutions when (5.55) is not satised will be considered in Section
7.3). Using (5.57) and noting that to leading order x we can write, for
< < t
1
,
= Dsin(k +), (7.31)

where D, k and are to be found as part of the solution.
Relation (7.31) approximates the waves in < < t
1
by linear waves.
It is therefore a good approximation if these waves are of small amplitude.
An extension of (7.31) consists of writing
n=1
D
n
sin(nk +). (7.32)
Relation (7.32) is now an approximation for nonlinear waves and is exact
in the limit

N . The constants D
1
, D
2
, . . . , D
N
, k and are found by
tting values of
at points on the discretised mesh close to t

1
.
We repeated the calculations for dierent values of

N and checked that
the results presented are independent of

N.
This truncation procedure enabled us to conrm and improve the results
of Forbes [56]. The results presented by Forbes do not have a constant mean
upstream height. Furthermore the distortion that occurs in the T = 0 case is
even more pronounced in the T ,= 0 case and extends over two wavelengths.
In Figure 7.6 we present results in the absence of surface tension to con-
trast them with the gravitycapillary solution of Figure 7.7. As can be seen
in Figure 7.7, the modied model not only demonstrates the existence of a
solution with wave trains both upstream and downstream but also that the
previous problems associated with this computation, namely the decreasing
mean height of the upstream waves and the violation of the radiation condi-
tion, has been eliminated. The linear dispersion relation (2.83) predicts an
upstream wavelength 0.8796 and a downstream wavelength 3.4247.
These predictions agree to within 2% with the values predicted by the non-
linear computations. We note that for 3.4247 and W = 0.07 the con-
dition (5.55) is satised. Further solutions, which involve Wilton ripples
15 10 5 0 5 10 15
0.98
0.99
1.00
1.01
1.02
Fig. 7.6. Free-surface ow for = 0.05, F = 0.8, W = 0.
0.98
0.99
1.00
1.01
1.02
15 10 5 0 5 10 15
Fig. 7.7. Free-surface ow for = 0.05, F = 0.8, W = 0.07.
and which are valid when condition (5.55) is not satised, will be presented in
Section 7.3.
The approximations of the far eld described in this section are general
and can be used for any steady two-dimensional gravitycapillary free-surface
ow past a disturbance.
7.2.2 Results in innite depth (removal of the nonuniformity)
In this section, we show that the nonuniformity of the linear gravitycapillary
free-surface ows shown in Figure 4.8 as 1/4 is removed when nonlinear
solutions are calculated. Here the parameter is dened by (4.69).
The linear theory of Chapter 4 predicts solutions with waves when < 1/4
and solutions with a at free surface in the far eld when > 1/4. In
this section we restrict our attention to values > 1/4 and show that the
nonuniformity of Figure 4.8 as 1/4
+
is removed when a nonlinear theory
is developed. In the far eld we have
x
U as [x[ . (7.33)
We introduce dimensionless variables by using U
2
/g as the reference length
and U as the reference velocity and assume that the disturbance is the pres-
sure distribution (7.8). The formulation and numerical procedure follow that
of Section 7.1.1. The only dierences are that surface tension needs to be
included in the dynamic boundary condition and that the uid is of innite
depth. Therefore we rewrite the dynamic boundary condition (7.7) as
1
2
+

2
+ 2y + 2p()
2
2
+

2
)
3/2
= 1. (7.34)
When = (i.e. T = 0), (7.34) reduces to (7.7).
The appropriate integral relation between

and

can be derived by taking
the limit Q in (7.6). This yields
() 1 =
1
()

d. (7.35)
Equations (7.34) and (7.35) dene a system of equations for the unknown
functions

and

. To solve it numerically we introduce the mesh points
(7.10), the midpoints (7.11) and the unknowns (7.12). We obtain 2N 2
equations by satisfying (7.34) and (7.35) at the midpoints (7.11); three more
equations are provided by (7.13) and (7.14). The resulting system of 2N +1
equations for the 2N + 1 unknowns
I
,
I
and is solved by Newtons
method for given values of and a. Details about the numerical procedure
can be found in Vanden-Broeck and Dias [184] and Dias et al. [44].
In Figure 7.8 we present numerical values of a (see (7.14) for a denition)
versus . The upper curve corresponds to = 0.001 and the lower curve to
= 0.001. The corresponding linear results are shown in Figure 4.8. Figure
7.8 shows that the nonuniformity of the linear theory (i.e. the vertical
asymptote at = 0.25) has been removed in the nonlinear calculations: all
the values of a are now nite.
The curves in Figure 7.8 have turning points at =
. Here
+
and
correspond to the upper and lower curves respectively. The values
are
greater than 0.25 and depend on the value of .
The solutions corresponding to the lower portion of the upper curve for
>
+
and to the upper portion of the lower curve for >
are pertur-
bations of a uniform stream, in the sense that continuation in (see Section
0
0.04
0.25 0.26 0.27 0.28 0.29 0.30
Fig. 7.8. Values of a versus for = 0.001 (upper curve) and = 0.001 (lower
curve).
0
0.1
0 20 40
Fig. 7.9. Depression solitary wave for = 0.264.
7.1.1 for a denition) ultimately leads to a uniform stream. These solutions
are described by the linear theory of Chapter 4 when [[ is small.
If we use continuation in by starting with a solution on the upper portion
of the upper curve in Figure 7.8 for >
+
, we obtain elevation solitary
waves with a decaying tail; similarly, if we use continuation in by starting
with a solution on the lower portion of the lower curve in Figure 7.8 for >
, we obtain depression solitary waves with a decaying tail. The shapes of

such solitary waves are indicated in Figures 6.5 and 6.6. The solitary waves
with decaying tails form solution branches that bifurcate from = 0.25 in
Figure 7.8. Typical free-surface proles are shown in Figures 7.9 and 7.10.
The fact that the solitary waves bifurcate from = 0.25 can be explained
intuitively as follows. The proles of Figures 7.9 and 7.10 look like a wave
train whose amplitude is slowly varying. According to linear theory we
should expect these waves of varying amplitude to travel at the phase ve-
locity while their envelopes travel at the group velocity (see Section 2.4.3).
Since in general phase and group velocities have dierent values, the waves
of Figures 7.9 and 7.10 cannot be expected to be steady unless the phase
and group velocity are equal. This is exactly what happens when = 0.25.
0
0.1
0 20 40
Fig. 7.10. Elevation solitary wave for = 0.261.
We recall that nonuniformities similar to that of Figure 4.8 occur in water
of nite depth. Here we have restricted our attention to innite depth in
order to keep the presentation simple. However, similar calculations can be
performed for water of nite depth (see [44]). In particular it can be shown
that branches of solitary waves bifurcate from the minima of the curves in
Figure 2.5 when < 1/3.
Existence proofs for solitary waves with decaying tails were given by Iooss
and Kirchgassner [79] and others (see Dias and Khari [43] for a review).
Furthermore, analytical approximations valid near the bifurcation points
were obtained by Dias and Iooss [42], Akylas [3] and Longuet-Higgins [103].
7.3 Gravitycapillary free-surface ows with Wilton ripples
When < 1/4, linear solutions in water of innite depth for the ow con-
guration of Figure 4.4 are characterised in the far eld by trains of linear
periodic waves. Here is dened by (4.69). There are two wave trains.
The one of shorter wavelength occurs as x and the one of longer
wavelength as x . These linear waves are described by the formulae
(4.66) and (4.67) of Chapter 4. They are consistent with the formulae (5.62)
and (5.63) derived in Chapter 5. However, we showed in Section 5.1 that
the formulae (5.62) and (5.63) are only valid when the condition (5.66) is
satised. Therefore the linear solutions of Section 4.2.2 are not valid when
=
4
2
T
g
2
=
1
m
, m = 2, 3, . . . (7.36)
because there is no nonlinear train of periodic waves that can approach a
linear wave train at x = as the size of the disturbance approaches zero.
A similar diculty arises in problems in water of nite depth when the
condition (5.55) is not satised (nonlinear solutions when (5.55) is satised
were described in Section 7.2.1).
We describe in this section the appropriate solutions in water of innite
depth when = 1/m in the particular case m = 2. Then the linear waves
as x should be described by (5.107) and (5.106) instead of (5.62)
and (5.63). In this section we use the numerical procedure of Section 7.2.2
together with the truncation techniques of Section 7.2.1 to compute solutions
when = 1/2 (see also Vanden-Broeck [182]).
We note that (5.112), (5.113) and (5.67) imply that
=

2
=
(1 +)
2
=
9
8
. (7.37)
The problem is formulated in equations (7.34) and (7.35). To solve them
numerically we introduce the mesh points
I
=
(N 1)E
2
+ (I 1)E, I = 1, . . . , N, (7.38)
and the unknowns
I
=

(
I
). (7.39)
Here E is the interval of discretisation.
The integro-dierential equations are discretised and satised at the in-
termediate mesh points
I+1/2
=

I
+
I+1
2
, I = 1, . . . , N 1. (7.40)
This leads to N 1 nonlinear algebraic equations. The Cauchy principal
value in (7.35) is approximated by the trapezoidal rule with a summation
over the mesh points (7.38) (see Section 3.1.2.2). This yields
(
I+1/2
) = 1
1
J=1
w
J
J

I+1/2
, (7.41)
where w
1
= w
N
= E/2 and w
J
= E otherwise. The symmetry of the
quadrature has enabled us to evaluate the Cauchy principal value integral
in (7.35) as if it were an ordinary integral.
The approximation (7.41) replaces the integral from to in (7.35)
by an integral from A to A, where A = (N 1)E/2. To obtain the results
to be presented here, we improve this truncation by modifying appropriately
the approach of Section 7.2.1. We rst rewrite (7.35) as
() = 1
1
_
A
A
()

d
1
_
A
B
()

d
1
_
B
A
r
()

d, (7.42)
before applying the trapezoidal rule. Here B A. For A suciently large,
and

r
are the periodic wave trains of Chapter 5. We present results for
the value = 9/8 derived in (7.37). For this value of , we showed that
the linear solution of Chapter 5 fails in the nonlinear regime because there
are no nonlinear periodic waves approaching the linear wave on the far right
as 0. The correct nonlinear solutions for = 9/8 should have a train
of Wilton ripples on the far right. As we shall see this is conrmed by our
numerical calculations. We will restrict our calculations to 1, so that a
formula for

r
can be derived from (5.106) and (5.107). This yields
r
= Ak
sin k
( +) Ak
sin 2k
( +). (7.43)
Similarly, (5.62) and (5.63) give
(
1
) = A
k
+
sin k
+
( +
). (7.44)
The constants A, A
, ,
are found by imposing the continuity conditions
r
(
N
) =
N
,

r
(
N1
) =
N1
,

r
(
N2
) =
N2
, (7.45)
(
1
) =
1
,

(
2
) =
2
. (7.46)
Relation (7.44) imposes a train of linear sine waves of short wavelength at
the far left. Relation (7.43) imposes a train of Wilton ripples at the far
right.
Relations (7.45) and (7.46) together with the N 1 equations obtained
by discretising the integro-dierential equations dene a system of N + 4
nonlinear algebraic equations for the N + 4 unknowns A, A
, ,
and
I
,
I = 1, . . . , N. This system is solved by Newtons method.
There are two nonlinear solutions, corresponding to the plus and minus
signs in (7.43). These are shown in Figures 7.11 and 7.12. As x , these
solutions approach Wilton ripples. A further discussion of this problem can
be found in [182].
10
0
5
0 5 10
Fig. 7.11. Free-surface prole of a nonlinear solution for = 9/8 and = 10
5
.
The vertical scale is in units of 10
5
.
10
0
5
0 5 10
Fig. 7.12. Free-surface prole of a nonlinear solution for = 9/8 and = 10
5
.
The vertical scale is in units of 10
5
.
8
Free-surface ows with waves and intersections
with rigid walls
In Chapter 3 we calculated pure gravity ows (g ,= 0, T = 0) which approach
either an innitely thin jet in the far eld (see Figure 3.36) or a uniform
stream characterised by a Froude number F greater than unity (see Figure
3.53). In the latter case the ow is waveless in the far eld, in accordance
with the linear theory of Section 2.4.
We now consider pure gravity ows for the conguration of Figure 3.37,
for which the ow is subcritical in the far eld (i.e. for which F < 1). The
linear theory of Section 2.4 and the nonlinear computations of Chapter 7
suggest that we should expect a wave train in the far eld. We shall present
computations for three particular cases of Figure 3.37. In the rst two cases
(see Figures 8.1 and 8.2) it will be assumed that the ow is of innite depth
(i.e. that the distance between the bottom AB and the streamline CEF in
Figure 3.37 is so large that the layer of uid can be assumed to be of innite
depth). The value of the Froude number F is then 0. The analysis follows
Vanden-Broeck [158], Vanden-Broeck et al. [190], Vanden-Broeck and Tuck
[192] and Vanden-Broeck [168].
A
B
y
x
C
U
S
H
S
P
S
L
S
R
S
F
Fig. 8.1. The free surface past a at plate. The contour used in Sections 8.1.2 and
8.4 is also shown. The ow leaves the plate tangentially (i.e. = 0) when surface
tension is neglected.
210
A
D
C
B
3
U
Fig. 8.2. The free-surface ow past a surface-piercing obstacle. The free surface
separates tangentially from the inclined wall at point B.
The third case is the ow under a sluice gate, shown in Figure 3.34 (see
also Vanden-Broeck [181] and Binder and Vanden-Broeck [15], [16]). This
is a ow in water of nite depth which is subcritical as x and
supercritical as x .
The ow of Figure 8.1 is probably the simplest ow for which a free surface
intersects a rigid surface. As we shall see, an attractive feature of this ow
is that some exact formulae for the amplitude of the waves in the far eld
can be derived. These formulae provide analytical insights and can be used
to check the accuracy of numerical codes.
It is tempting to try to compute solutions for the ow congurations of
Figures 8.1, 8.2 and 3.34 by using a series representation similar to (3.245),
(3.246). However, it is not easy to remove the singularity of the ow as
x in Figures 8.1 and 8.2 because the complex velocity w = u iv
oscillates innitely often as x . The singularity at t = 1 corresponds to
a nonlinear wave train. This is to be contrasted with the corresponding su-
percritical ows of Section 3.3, where the ow approaches a uniform stream
as x and can therefore be described by linear theory (which assumes
a small perturbation around a uniform stream). In view of these remarks it
is preferable to use a boundary integral equation method to compute free-
surface ows having a wave train in the far eld. Solutions for the ow
congurations of Figures 8.1 and 8.2 are presented in Sections 8.1 and 8.2.
The ow under a sluice gate is treated in Section 8.3. Finally, an extension
of the ndings of Section 8.1 to pure capillary ows is presented in Section
8.4.
8.1 Free-surface ow past a at plate
We start with the ow conguration of Figure 8.1. Numerical results are
described in Section 8.1.1. The exact analytical results are derived in Section
8.1.2. The presentation follows [158].
212 Free-surface ows with waves and intersections with rigid walls
8.1.1 Numerical results
Let us consider the ow conguration of Figure 8.1. As y , the ow
is uniform and characterised by a constant velocity U. We introduce di-
mensionless variables by using U
2
/g as the reference length and U as the
reference velocity. We dene cartesian coordinates with the y-axis directed
vertically upwards. Gravity is acting in the negative y-direction. We in-
troduce a potential function and a streamfunction . We choose = 0
at the separation point B and = 0 on the streamline ABC. The ow
is then mapped into the lower half-plane < 0 of the complex potential
plane f = +i. We introduce the complex velocity u iv and dene the
function i by the formula (3.3). Since u iv 1 as ,
i 0 as . (8.1)
The dynamic boundary condition in dimensionless variables yields
e
2
+ 2y =

B. (8.2)
We choose the origin of y such that

B = 1.
The solutions are characterised by the Froude number
F =
U
(gH)
1/2
, (8.3)
where y = H is the dimensional ordinate of the separation point B in
Figure 8.1.
Next we apply Cauchys integral equation formula (2.41) to i in the
complex f-plane with a contour consisting of the -axis and a circle C
R
of
radius R in the lower half-plane, < 0. This yields
() i
() =
1
i
_
R
R
() i
()

d
1
i
_
C
R
(f) i(f)
f
df, (8.4)
where the rst integral is a Cauchy principal value. Here () and

()
denote the values of and on the streamline = 0.
We now take the limit R in (8.4). It can be shown by using (8.1)
that the integral over C
R
in (8.4) tends to 0 as R . Therefore (8.4)
yields
() i
() =
1
i
_

() i
()

d. (8.5)
Taking the real and imaginary parts of (8.5) gives
() =
1
()

d (8.6)
and
() =
1
()

d. (8.7)
Relations (8.6) and (8.7) are similar to the Hilbert transforms (3.73) and
(3.74). As mentioned in Section 3.1.2.2, each implies the other. Therefore
we are free to choose either (8.6) or (8.7) for our boundary integral equation
formulation. However, (8.6) is more convenient than (8.7) because it only
involves unknowns on the free surface = 0, > 0. This follows from the
fact that the kinematic boundary condition on the wall AB can be written
as
= 0 on = 0, < 0. (8.8)
Substituting (8.8) into (8.6) gives
() =
1
_

0
()

d. (8.9)
If we assume that > 0 then (8.9) involves only values of and on the free
surface. Next we eliminate y from (8.2) by dierentiating the latter with
respect to . Using the identity
x
+y
=
1
u iv
= e
+i
(8.10)
gives
e
2

+e

sin

= 0. (8.11)
Relations (8.9) and (8.11) dene a system of integro-dierential equations
for the values of () and

() on the free surface = 0, > 0. An equation
involving only

() can be obtained by substituting (8.9) into (8.11). We
solve this equation numerically by nite dierences. We rst dene the mesh
points
I
= (I 1)E, I = 1, . . . , N, (8.12)
and the corresponding unknowns
I
= (
I
). (8.13)
As before, E is the interval of discretisation. We shall require (8.11) to be
satised at the midpoints
m
I
=

I
+
I+1
2
, I = 1, . . . , N 1. (8.14)
Following the analysis in Section 3.1.2.2, we rst evaluate (
m
I
) by applying
the trapezoidal rule to the integral (8.9) with a summation over the points
(8.12). Next we evaluate the values of

and / at the midpoints (8.14)
by linear interpolation and centred nite dierence formulae.
There are N unknowns
I
, I = 1, . . . , N. One equation forces the free
surface to leave the plate tangentially:
1
= 0. (8.15)
Another denes the value of F by the requirement
e
2 (0)
= 1 +
2
F
2
, (8.16)
where F is prescribed. Equation (8.16) follows because (8.2) holds at =
= 0 and the dimensionless ordinate of the separation point B is 1/F
2
.
In (8.16) the value of (0) is approximated by the two-point extrapolation
formula
(0) 3 (
m
1
) 2 (
m
2
). (8.17)
The nal N 2 equations are obtained by satisfying (8.11) at
m
I
, I =
1, . . . , N 2. The resulting system of N equations with N unknowns is
solved by Newton iteration. Typical free-surface proles for F = 3.34 and
F = 6.3 are shown in Figure 8.3. For convenience the origin of cartesian
coordinates has been chosen as the separation point B. The gure shows
that there is a wave train on the free surface. As x , the amplitude of
the waves tends to a constant. When F , the ow reduces to a uniform
stream. For F large, the waves are of small amplitude and are close to linear
sine waves. As F decreases, the amplitude of the waves increases and the
wave proles develop sharp crests and broad troughs in accordance with the
nonlinear properties of steep gravity waves (see Chapter 6).
There are two sources of error in the numerical scheme. The rst is that
the innite domain in the integral (8.9) has been truncated at = NE.
The second is the size of E. Accurate solutions are obtained in the limit
E 0 and NE . One way to take this double limit is to x a value
of NE and compute solutions for smaller and smaller values of E until they
become independent of E (for example, within graphical accuracy) and then
to repeat the procedure for larger and larger values of NE until the solutions
become independent of NE as well.
The above numerical method truncates the integral from 0 to to an
integral from 0 to NE. In other words

is assumed to be zero from NE to
. As was noted in Section 7.1.2, this truncation has a pronounced eect
only on the last wavelength of the wave train, and so very accurate solutions
0
0.1
0.2
0.3
0.4
0.5
0 5 10 15 20 25 30 35 40
Fig. 8.3. Free-surface proles for the ow shown in Figure 8.1. The solid curve
corresponds to F = 3.34 and the broken curve to F = 6.3. The plate is along the
negative x-axis.
for gravity free-surface ows with waves can be obtained. Furthermore, the
distortion in the last wavelength can be removed by using the approach of
Section 7.1.2 to approximate for NE < < .
8.1.2 Analytical results
In this section we supplement the numerical computations of Section 8.1.1
with exact analytical results. We rst note that for two-dimensional steady
ows the Euler equations (2.1) and the conservation of mass equation (2.6)
imply
_
S
_
V(V n) +gyn +
p
n
_
d = 0. (8.18)
Here S is a closed contour inside the uid region, V the vector velocity, p
the pressure, the density, the arc length along S and n the unit normal
exterior to the contour S. Relation (8.18) expresses the conservation of
momentum; it follows from (2.1) and (2.6) by application of the divergence
theorem. We now choose S to consist of the plate S
p
, the free surface S
F
,
a vertical line S
R
at x = , a horizontal line S
H
at y = and a vertical
line S
L
at x = (see Figure 8.1). We take the component of (8.18) along
the x-axis. This gives
_
S
_
u(V n) +gyn
x
+
p
n
x
_
d = 0. (8.19)
Here u and n
x
are the components of V and n along the x-axis. It is
convenient to replace the line S
H
by a horizontal line at y = d, where
d is arbitrarily large. Without loss of generality we may assume that S
R
intersects the free surface at the level y = 0.
The integrals over S
p
and S
H
in (8.19) do not contribute since V n = 0
and n
x
= 0 along them. The integration over S
F
, S
R
and S
L
in (8.19) gives
gH
2
2

_
H
d
_
u
2
+
p
+gy
_
x=
dy
gd
2
2
+
S
W
= 0. (8.20)
Here u
is the uniform velocity at x = . The quantity S

W
is dened by
S
W
=
_
0
d
(p +u
2
)
x=
dy (8.21)
and represents the momentum ux per unit span of the waves far from the
plate. Using Bernoullis equation we can rewrite the integral in (8.20) as
_
H
d
_
u
2
+
p
+gy
_
dy =
_
1
2
u
2
+
1
2
u
2
_
(H +d). (8.22)
We note that S
W
involves only quantities dened as x (see (8.21)).
Since the ow as x is characterised by a train of periodic waves, S
W
can be calculated as a function of the amplitude of the waves by using the
computations of periodic waves described in Chapter 6. Longuet-Higgins
[101] showed that S
W
can be written as
S
W
= 3V +dU
2
+
1
2
d
2
g, (8.23)
where
V =
g
2
_
s+
s
2
(x)dx (8.24)
is the mean potential energy per unit horizontal area of the waves. Here s
is a large number and y = (x) is the equation of the free surface in the far
eld. We note that
V =

V , (8.25)
where

V is given by (2.156). Substituting (8.22) and (8.23) into (8.20) we
have
gH
2
2
=
3V

1
2
U
2
H
1
2
u
2
H +
1
2
(u
2
d U
2
d). (8.26)
Longuet-Higgins [101] also showed that the conservation of mass can be
written as
u
(d H) = Ud
2K
U
, (8.27)
where
K =

2
_
s+
s
_

d
_
_
x
_
2
+
_
y
_
2
_
dydx (8.28)
is the mean kinetic energy per unit horizontal area of the waves. We note
that when the waves are of small amplitude the upper limit of integration
can be replaced by zero, and so
K =

K, (8.29)
where

K is dened by (2.155). Multiplying (8.27) rst by u
and then by U
and adding the results yields
u
2
d U
2
d = u
2
H +u
UH
2K

2K
U
. (8.30)
Substituting (8.30) into (8.26), we obtain
gH
2
2
=
3V
U
+
u
UH
2

U
2
2
. (8.31)
As d we have u
= U. Thus
gH
2
2
=
3V

2K
. (8.32)
Using the denition (8.3) of the Froude number we can rewrite (8.32) as
F = U
_
6g
V
4g
K
_
1/4
. (8.33)
As shown in Section 6.5.2, periodic gravity waves can be characterised by
their steepness s, dened as the dierence in ordinates between a crest and a
trough divided by the wavelength. The quantities U, V and K are functions
of the steepness s of the waves. Values of U versus s were calculated in
Section 6.5.2 (see Figure 6.21). Similarly values of V and K versus s were
calculated by Longuet-Higgins [101], Cokelet [31] and Schwartz and Vanden-
Broeck [135]. Thus (8.33) amounts to a relation between the Froude number
F and the steepness s of the waves far from the plate. This relation is
shown graphically in Figure 8.4 (see the solid curve). It was obtained by
substituting Cokelets results into (8.33). The numerical results of Section
8.1.1 show that the steepness of the waves decreases away from the plate
and reaches a constant value after a few periods. The crosses in Figure 8.4
show these constant values for a few values of the Froude number. These
numerical results are in close agreement with those predicted by the exact
relation (8.33).
0.05
0.10
0.14
0 2 4 6
s
F
Fig. 8.4. Values of the steepness s versus F. The solid curve shows the exact relation
(8.33) in which Cokelets [31] results are used. The crosses give the numerical results
obtained in Section 8.1.1. The broken line corresponds to the numerical results
derived in Section 8.2.1.
We showed in Section 6.5.2 that the values of U oscillate innitely often
as the wave of maximum steepness is approached. Longuet-Higgins [101]
and Cokelet [31] showed that the values of V and K have a similar property.
Therefore we can expect the values of F predicted by (8.33) also to oscillate
innitely often as the waves in the far eld approach their maximum steep-
ness. This was conrmed by the calculations of Vanden-Broeck [158]. We
note that these oscillations are too small to be seen on the scale of Figure
8.4.
8.2 Free-surface ow past a surface-piercing object
We now turn our attention to the ow shown in Figure 8.2. For the ow
conguration of Figure 8.1, the free surface leaves the plate AB tangentially
at the point B. The numerical and analytical calculations of Sections 8.1.1
and 8.1.2 show that there is a one-parameter family of solutions, for the
ow of Figure 8.1. By analogy we can expect that there is a two-parameter
family of solutions for the ow of Figure 8.2. A possible choice for the rst
parameter is the Froude number F dened by (8.3), where again H is the
dimensional ordinate of corner D in Figure 8.2. The second parameter is
then the length l
w
of the wall DB. As l
w
0, the ow of Figure 8.2 reduces
to the ow of Figure 8.1. As l
w
increases, the velocity at the separation B
decreases, and a limiting conguration is reached when a stagnation point
(i.e. a point with u = v = 0) occurs at the separation point B.
C
B
3
U
D A
Fig. 8.5. Free-surface ow past a surface-piercing object with a stagnation point at
B. When
3
> /3, the free surface is horizontal at the separation point B. When
3
< /3, there is an angle of 2/3 between the wall DB and the free surface.
The local analysis of Section 3.3 can be used to determine the angle be-
tween the free surface and the wall DB when B is a stagnation point. Com-
paring Figures 3.38 and 8.5, we nd that
3
=
2
. Therefore the discussion
following (3.175) and (3.176) implies that an angle of 2/3 between the free
surface and the wall at the stagnation point B will occur if
3
< /3. If
3
> /3, the free surface has to be horizontal at the stagnation point.
We now present numerical results for the limiting conguration of Figure
8.5 with
3
= /2. Since /2 > /3, as discussed above the free surface is
then horizontal at B.
We choose = 0 at the stagnation point B and denote by b the value of
at the corner D. The formulation and the numerical method follow that
of Section 8.1.1. Our analysis also follows that in [190], [192] and [168]. For
consistency we shall use the same notation as that used in these publications.
In particular we introduce dimensionless variables by taking b/U as the unit
length and U as the unit velocity. The dynamic boundary condition on the
free surface then becomes
e
2
+y = , > 0, (8.34)
where
=
U
3
2gb
. (8.35)
Dierentiating (8.34) with respect to and using the identity (8.10) yields
2e
2

+e

sin

= 0, > 0. (8.36)
The ow is mapped into the lower half-plane < 0 of the complex po-
tential f-plane. We then introduce the complex velocity u iv and dene
the function i by (3.3). The values of () and

() on the streamline
= 0 are related by (8.6). The kinematic boundary conditions on the walls
AD and DB yield
= 0, < < 1, (8.37)
= /2, 1 < < 0, (8.38)

since the dimensionless value of at corner D is 1. Substituting (8.37)
and (8.38) into (8.6) yields
() =
1
2
_
0
1
1

d +
1
_

0
()

d. (8.39)
Evaluating the rst integral on the right-hand side of (8.39) gives
() =
1
2
ln
+ 1
+
1
_

0
()

d. (8.40)
Relations (8.36) and (8.40) dene an integro-dierential equation for

().
We discretise the problem by following the procedure used in Section
8.1.1 for the ow past a at plate. We introduce the mesh points (8.12), the
unknowns (8.13) and the midpoints (8.14) and evaluate (
m
I
) by applying
the trapezoidal rule to the integral on the right-hand side of (8.40) with a
summation over the mesh points (8.12). We then evaluate

(
m
i
) by linear
interpolation and / at the midpoints (8.14) by centred dierences. We
then satisfy (8.36) at the midpoints (8.14). This leads to N 1 equations.
The nal equation is obtained by expressing

3
in terms of

2
:
3
=

2
_
2
_
1/2
. (8.41)
Equation (8.41) becomes exact in the limit E 0, since

1/2
as 0.
The motivation for the choice (8.41) is given below in Section 8.2.2 (see also
[190], [192] and [168]).
A typical free-surface prole is shown in Figure 8.6. Further free surface
proles can be found in [190], [192] and [168].
0
0 5 10 15 20 25 30
Fig. 8.6. Values of y versus x for = 0.45.
There is again a wave train in the far eld. As 0 (i.e. as F 0),
the amplitude of the waves approaches zero and the free surface approaches
a horizontal straight line. This solution is often referred to as the rigid-lid
solution. An exact solution is obtained on setting

= 0 in (8.40). This gives

0
() =
1
2
ln
+ 1
0
= 0 for > 0. (8.42)
Here the subscript 0 signies that = 0.
Relation (8.42) and the identity (8.10) imply
dz
0
df
=
_
+ 1
_
1/2
for > 0. (8.43)
It is tempting to try to construct analytically a solution for F small (or
equivalently for small) by perturbing (8.42) and then seeking dz/df as an
expansion in powers of . Thus we write
dz
df
= e
+i
=
n=0
n
z
n
(f), (8.44)
where z
0
is given on the free surface by the right-hand side of (8.43).
Substituting (8.44) into the system (8.36), (8.40) and equating coecients
of
n
leads to a recurrence relation for the coecients z
n
(f), and an arbitrary
large number of coecients can in principle be calculated. However, the
series is divergent and furthermore the partial sums predict a at free surface
in the far eld, without wave trains. This discrepancy between the expansion
(8.44) and the numerical results is due to the fact that the amplitude of the
waves is an exponentially small function of . This means that as 0 the
amplitude a
w
of the waves tends to zero faster than any powers of , i.e.
a
w
= o(
m
) (8.45)
for any positive integer m, where the o symbol was dened in (3.24) and
(3.25).
The work of [190], [192] and [168] showed that the expansion (8.44) di-
verges as n!. More precisely it was found that
z
n
C()[
1
()]
n
(n +a)! for n large. (8.46)
Here a 0.6. The functions
1
() and C() are complex valued and were
calculated in [190], [192] and [168].
Combining (8.46) and (8.44) we obtain
dz
df

N
n=0
n
z
n
+C()
n=N+1
n
[
1
()]
n
(n +a)!. (8.47)
Using the denition of the gamma function,
(n +a)! =
_

0
e
t
t
n+a
dt,
we can rewrite the innite sum in (8.47):
C
n=N+1
n
1
(n +a)! = C
_

0
e
t
t
a
(
1
t)
N+1
m=0
(
1
t)
m
dt. (8.48)
The innite series
m=0
(
1
t)
m
(8.49)
appearing in the integrand of (8.48) is immediately recognizable as the
Taylor expansion of the function
1
1
1
t
. (8.50)
Therefore we have
C
n=N+1
(
1
)
n
(n +a)! C
_

0
e
t
t
a
(
1
t)
N+1
1
1
t
dt. (8.51)
We note that the Taylor expansion (8.49) converges only if [t[ < ([
1
[)
1
.
The above procedure, known as Borel summation, is based on the assump-
tion that the series (8.49) can be identied with the function (8.50) for all
t. Using the identity
(
1
t)
N+1
1
1
t
=
(
1
t)
N
1
1
t
(
1
t)
N
,
we rewrite the right-hand side of (8.51) in the form
C
_

0
e
t
t
a
(
1
t)
N
1
1
1
t
dt C(
1
)
N
(N +a)! C(
1
)
N
1
(N
1
+a)!, (8.52)
where N
1
is the smallest integer such that N
1
+a > 1. Thus, substituting
(8.52) into (8.51) we obtain
z
(, ) C
_

0
e
t
t
a
(
1
t)
N
1
1
1
t
dt+
N
n=N
2
[
n
z
n
C(
1
)
n
(n+a)!]+F
N
1
, (8.53)
where N
2
= N
1
if N
1
> 0 and N
2
= 0 otherwise, and
F
N
1
= C
1
n=N
1
(
1
)
n
(n +a)! if N
1
< 0,
F
N
1
= 0 if N
1
= 0,
F
N
1
= C
N
1
1
n=0
n
z
n
if N
1
> 0.
Using terminology introduced by Dingle [51] we describe the function
T
N
1
_
1
_
= C
_

0
e
t
t
a
(
1
t)
N
1
1
1
t
dt (8.54)
as the terminant of the divergent series expansion.
The main diculty about the terminant (8.54) is its lack of uniqueness.
We can describe this nonuniqueness in terms of a branch-cut location. For
this purpose we rewrite (8.54) in the form
T
N
1
_
1
_
= C(
1
)
N
1
1
e
1/
1
W
N
1
_
1
_
, (8.55)
where the function W
N
1
(t) can be represented by the following convergent
expansions (see [51]):
W
N
1
(t) =
1
t
(N
1
+a)!e
t
_
t
N
1
+a
_
1
t
N
1
+a 1
+
t
2
(N
1
+a 1)(N
1
+a 2)

__
t
N
1
+a
sin (N
1
+a)
(8.56)
when a is not an integer and
W
N
1
(t) = e
t
i=0
t
i
[(i) ln t]
i!
, (8.57)
when a is an integer. Here
(i) =
d
di
(ln i!).
The branch cut of W
N
1
(t) is dened as the branch cut of ln t appearing
explicitly in the expansion for integer a and implicitly in the expansion for
noninteger a, since
t
N
1
+a
= e
(N
1
+a) ln t
.
We note that when a is an integer the function W
N
1
(t) reduces to the expo-
nential integral function E
1
(t).
Dierent functions W
N
1
(t) can be obtained by specifying dierent cuts
in the complex t-plane. In the present problem, we are interested in the
following three choices for W
N
1
(t), with 't < 0:
W
N
1
(t) = W
0
N
1
(t) =
_

t
( t)
N
1
+a
e
d; (8.58)
W
N
1
(t) = W
+
N
1
(t) = W
0
N
1
(t), t > 0,
W
N
1
(t) = W
+
N
1
(t) = W
0
N
1
(t) 2i(t)
N
1
+a
, t < 0, (8.59)
W
N
1
(t) = W
N
1
(t) = W
0
N
1
(t) + 2i(t)
N
1
+a
, t > 0, (8.60)
W
N
1
(t) = W
N
1
(t) = W
0
N
1
(t), t < 0.
Here (t)
N
1
+a
is dened as the principal value of the power. The function
W
N
1
(t) is dened in the complex t-plane cut along the negative real t-axis,
whereas the functions W
+
N
1
(t) and W
N
1
(t) are dened in the complex t-plane
cut along the positive real t-axis.
The usefulness of the original asymptotic expansion can now be increased
by substituting (8.54) and (8.55) into (8.53). This yields
z
(, ) F
N
1
+
N
n=N
2
[
n
z
n
C(
1
)
n
(n +a)!]
C(
1
)
N
1
1
e
1/
1
W
N
1
(1/
1
) . (8.61)
We note that the three solutions (8.61) corresponding to the three choices
(8.58)(8.61) dier by a term that is exponentially small in the limit 0.
Let us rst consider the choice (8.58). Since W
0
N
1
(t) is dened in the
complex t-plane cut along the negative real t-axis, the corresponding func-
tion in (8.61) is dened in the complex
1
-plane cut along the positive real
1
-axis. The numerical results show that this branch cut is crossed at some
point on the free surface. Hence the corresponding free surface (8.61) has a
discontinuity at that point and is therefore not physical.
Let us now consider the choices (8.59) and (8.61). Since the functions
W
+
N
1
and W
N
1
are dened in a t-plane cut along the positive real t-axis, the
corresponding function in (8.61) is dened in the complex
1
-plane cut along
the negative real
1
-axis. This branch cut is not crossed at any point along
the free surface. Hence the corresponding free surfaces are continuous and
can be expected to have a physical meaning.
Vanden-Broeck [168] showed that the solution (8.61) corresponding to
W
+
N
1
has a wave train in the far eld and agrees with the numerical solutions
of Section 8.2.1 when (or equivalently F) is small. By using (8.59) and
(8.61), Vanden-Broeck [168] estimated the steepness of the waves in the far
eld as
(0.5/
0.4
)e
1/D
, (8.62)
where D 0.4. The quantity (8.62) is exponentially small in the limit
0, in accordance with the discussion at the beginning of this section.
An improved estimate (which also shows that the steepness of the waves
is exponentially small) has recently been derived by Trinh, Chapman and
Vanden-Broeck.
Before discussing the solution corresponding to the choice W
N
1
, let us
mention that the ows of Figures 8.1 and 8.2 can be interpreted as free-
surface ows caused by the plate AB or the object ADB moving at a con-
stant velocity U to the left. This models the ow near the stern of a ship.
For these reasons the ows of Figures 8.1 and 8.2 are referred to as stern
ows. The presence of waves on the free surface is consistent with the radi-
ation condition (see Chapter 4), which requires gravity waves (if present) to
be at the back of a moving disturbance. Since potential ows are reversible,
we can reverse the direction of U in Figures 8.1 and 8.2. Therefore the ows
of Figures 8.1 and 8.2 can also be interpreted as free-surface ows caused
by the plate AB or the object ADB moving at a constant velocity U to the
right; the ows are then referred to as bow ows. However, the presence
of waves on the free surface is now incompatible with the radiation condi-
tion: appropriate bow ows should be characterised by a uniform stream as
x . In other words, none of the free-surface ows computed so far in
this chapter can model bow ows.
The solution (8.61) corresponding to W
N
1
is in fact characterised by a
uniform stream as x . However, the summation procedure based on
Borel summation fails to yield reliable solutions near the stagnation point.
Vanden-Broeck and Tuck [192] conjectured that bow ows are characterised
by a jet or spray near the rigid wall DB. This conjecture was conrmed
by Dias and Vanden-Broeck [47], who computed such solutions by series
truncation methods similar to those introduced in Chapter 3. A typical
free-surface prole is shown in Figure 8.7.
0
4
8
0 4 8 12
x
y
Fig. 8.7. Computed bow ow with a spray.
We conclude this section by mentioning that bow ows without spray
exist in water of nite depth (see Vanden-Broeck [175]).
8.3 Flow under a sluice gate
The free-surface ow under a sluice gate is a classical problem of uid me-
chanics. The ow conguration is shown in Figure 3.34. We shall rst
discuss results for
2
= /2 (see Figure 8.8). Extensions for
2
,= /2 will
be considered at the end of this section. The ow of Figure 8.8 is bounded
C
H U
D
y
x
B A
Fig. 8.8. The free-surface ow past a vertical sluice gate.
below by a horizontal bottom and above by two free surfaces AB and CD
and a vertical wall BC. Far downstream there is uniform ow with a con-
stant velocity U and a constant depth H. As we shall see, the problem can
be characterized by the downstream Froude number
F =
U
(gH)
1/2
, (8.63)
where g is the acceleration of gravity.
Over the years, many analytical and numerical approximations have been
obtained (see [11], [54], [93], [30], [181], [15] and [16]). It is usually assumed
that B is a stagnation point and that there is a uniform stream far upstream.
This last assumption is the radiation condition, which requires that there
are no waves far upstream. It is then possible to dene an upstream Froude
number,
F
U
=
V
(gD)
1/2
(8.64)
where V and D are the constant velocity and depth far upstream. The
Froude numbers F and F
U
are related by the identity
F
2
U
=
F
2
8
_
_
8
F
2
+ 1
_
1/2
1
_
3
(8.65)
(see [17] for a derivation). Many previous results suggest that there is a
one-parameter family of solutions and that the ow is subcritical upstream
(i.e. F
U
< 1) and supercritical downstream (i.e. F > 1).
In this section we will compute accurate numerical solutions for the fully
nonlinear problem. The analysis follows Vanden-Broeck [181] closely. The
problem is rst formulated as an integral equation for the unknown shapes
of the free surfaces. This equation is then discretised and the resulting
algebraic equations are solved by Newtons method, following the approaches
of Sections 3.1.2.2, 6.3 and Chapter 7. We again nd a one-parameter family
of solutions. However, our results show that the solutions do not satisfy the
radiation condition: there is a wave train on the upstream free surface. The
amplitude of the waves comes as part of the solution. Our numerical ndings
indicate the nonexistence of solutions that satisfy the radiation condition.
For F > 2.4, the waves are of very small amplitude and the upstream
free surface is essentially at. However, for F < 2.4 the waves are quite
noticeable and ultimately become large nonlinear waves as F is decreased.
As mentioned in Sections 3.1.1 and 3.3.3, an interesting quantity is the
contraction ratio
C
c
=
H
y
C
. (8.66)
Here y
C
is the distance from the separation point C to the bottom. We will
show that our computed values of C
c
are in good agreement with those of
Fangmeier and Strelko [54] for F large. This is consistent since the scheme
of Fangmeier and Strelko neglects the waves upstream and our calculations
show that the waves are indeed very small for F large.
The problem is formulated in Section 8.3.1. The numerical procedure is
described in Section 8.3.2 and the results are discussed in Section 8.3.3.
8.3.1 Formulation
The uid is assumed to be inviscid and incompressible and the ow to be
irrotational. The ow domain is bounded below by a horizontal bottom and
above by free surfaces AB and CD and a vertical wall BC (see Figure 8.8).
We introduce cartesian coordinates with the origin on the bottom and the
y-axis along the vertical wall. Gravity g is acting in the negative y-direction.
Far downstream the ow approaches a uniform stream with constant velocity
U and constant depth H. We dene dimensionless variables by choosing H
as the unit length and U as the unit velocity.
We introduce the complex potential function
f = +i (8.67)
and the complex velocity
w = u iv. (8.68)
Here as before u and v are the horizontal and vertical components of the
velocity. Following Benjamin [11] and others, we assume that B is a stag-
nation point (i.e. u = v = 0 at B). Without loss of generality, we choose
= 0 at B and = 0 on the streamline ABCD. It follows from our choice
of dimensionless variables that = 1 on the bottom. We denote by
C
the
value of at the separation point C. The ow conguration in the f-plane
is shown in Figure 8.9.
B C D A
D A
Fig. 8.9. The ow conguration of Figure 8.8 in the complex potential plane.
In terms of the dimensionless variables, the dynamic boundary condition
on the free surfaces AB and CD can be written as
u
2
+v
2
+
2
F
2
y = 1 +
2
F
2
. (8.69)
Here F is the Froude number dened by (8.63).
The kinematic conditions on the bottom and on the gate BC imply that
v = 0 on = 1 (8.70)
and
u = 0 on = 0, 0 < <
C
. (8.71)
This concludes the formulation of the problem. We seek w as an analytic
function of f in the strip 1 < < 0. This function must approach 1
as and must satisfy (8.69)(8.71). As we shall see, there is a one-
parameter family of solutions. It is convenient to choose this parameter as
C
.
We now reformulate the problem as an integral equation. First we dene
the function i by
w = e
i
(8.72)
and we map the ow domain onto the lower half of the -plane by the
transformation
= +i = e
f
. (8.73)
The ow in the -plane is shown in Figure 8.10.
A D B C D
Fig. 8.10. The ow conguration of Figure 8.8 in the -plane.
Next we apply the Cauchy integral formula (2.41) to the function i
in the complex -plane. We choose a contour consisting of the real axis and
a semicircle of arbitrary large radius in the lower half-plane. After taking
the real part we obtain
() =
1
. (8.74)
Here () and () denote the values of and on the axis = 0. The
integral in (8.74) is a Cauchy principal value. The kinematic conditions
(8.70) and (8.71) imply
() = 0 for < 0 (8.75)
and
() = /2 for 1 < <
C
, (8.76)
where
C
= e
C
.
Substituting (8.75) and (8.76) into (8.74), we obtain
() =
1
2
ln
[
C
[
[1 [
+
1
_
1
0
(
+
1
C
(
. (8.77)
Equation (8.77) provides a relation between and on the free surfaces.
We can obtain another relation between and on the free surfaces in the
following way. First we substitute (8.72) into (8.69). This yields
e
2
+
2
F
2
y = 1 +
2
F
2
. (8.78)
Next we evaluate the values of y on the free surfaces by using (8.73) and
integrating the identity
d(x +iy)
df
= w
1
. (8.79)
This gives
y() = 1 +
F
2
2
+
1
_

1
e
(
0
)
sin (
0
)
0
d
0
for 0 < < 1 (8.80)
and
y() = 1 +
1
e
(
0
)
sin (
0
)
0
d
0
for >
C
. (8.81)
Equations (8.77), (8.78), (8.80) and (8.81) dene a nonlinear integral equa-
tion for the unknown function () on the free surfaces 0 < < 1 and
>
C
.
8.3.2 Numerical procedure
We will solve the integral equation dened by (8.77), (8.78), (8.80) and
(8.81) numerically, using equally spaced points in the potential function .
Thus we introduce the change of variables
= e
(8.82)
and rewrite (8.77) as
() =
1
2
ln
[
C
e
[
[1 e
[
+
_
0
(
0
)e
0
e
0
e
d
0
+
_

(
0
)e
0
e
0
e
d
0
.
(8.83)
Similarly we rewrite (8.80) and (8.81) as
y
() = 1 +
F
2
2
+
1
_

0
e
(
0
)
sin
(
0
) d
0
, (8.84)
y
() = 1 +
_

(
0
)
sin
(
0
) d
0
. (8.85)
Here
() = (e
),
() = (e
) etc.
Next we introduce the equally spaced mesh points
U
I
= (I 1)
1
, I = 1, . . . , N
1
, (8.86)
and
D
I
=
C
+ (I 1)
2
, I = 1, . . . , N
2
(8.87)
on the upstream and downstream free surfaces. Here
1
> 0 and
2
> 0
are the mesh sizes. The corresponding unknowns are
U
I
=
(
U
I
), I = 1, . . . , N
1
, (8.88)
and
D
I
=
(
D
I
), I = 1, . . . , N
2
. (8.89)
Since
U
1
= 0 and
D
1
= /2, there are only N
1
+N
2
2 unknowns
U
I
and
D
I
.
We evaluate the values
U
I+1/2
and
D
I+1/2
of
() at the midpoints; thus
U
I+1/2
=

U
I
+
U
I+1
2
, I = 1, . . . , N
1
1, (8.90)
and
D
I+1/2
=

D
I
+
D
I+1
2
, I = 1, . . . , N
2
1. (8.91)
Now the trapezoidal rule is applied to the integrals in (8.83) with summa-
tions over the points
U
I
and
D
I
. The symmetry of the quadrature and of
the distribution of mesh points enables us to evaluate the Cauchy principal
values as if they were ordinary integrals (see Section 3.1.2.2). In the calcu-
lations presented here, we follow Hocking and Vanden-Broeck [72] and rst
rewrite the last integral in (8.83) as
_

D
N
2
C
[
(
0
)
I+1/2
]e
0
e
0
e
d
0
+

I+1/2
ln
[e
D
N
2
e
[
[
C
e
[
(8.92)
before applying the trapezoidal rule. The values
I+1/2
of at the mesh
points (8.90), (8.91) are evaluated in terms of the unknowns (8.88), (8.89)
by four-point interpolation formulae.
Next we evaluate y
U
I
= y
(
U
I
) and y
D
I
= y
(
D
I
) by applying the trape-
zoidal rule to (8.84) and (8.85). This yields
y
U
1
= 1 +
F
2
2
, (8.93)
y
U
I
= y
U
I1
e
U
I +1/2
(sin
U
I+1/2
)
1
, I = 2, 3, . . . , N
1
1, (8.94)
y
D
N
2
= 1, (8.95)
y
D
I
= y
D
I+1
e
D
I +1/2
(sin
D
I+1/2
)
2
, I = N
2
1, N
2
2, . . . , 1. (8.96)
We use these values to evaluate y
() at the midpoints (8.90) and (8.91)

by interpolation formulae.
We now satisfy (8.78) at the midpoints (8.90) and (8.91). This yields
N
1
+N
2
2 nonlinear algebraic equations for the N
1
+N
2
1 unknowns F,
U
I
, I = 2, . . . , N
1
, and
D
I
, I = 2, . . . , N
2
.
The nal equation is obtained by xing the length of the plate BC. Using
(8.72), (8.76) and (8.79), we obtain
y
= e
on 0 < <
C
. (8.97)
We use (8.83) to evaluate
() for 0 < <

C
and integrate (8.97) numer-
ically. This yields the length L of the plate BC in terms of the unknowns.
The last equation is then
y
U
1
y
D
1
L = 0. (8.98)
For a given value of
C
this system of N
1
+ N
2
1 equations with N
1
+
N
2
1 unknowns is solved by Newtons method.
8.3.3 Discussion of the results
We used the numerical scheme described in Section 8.3.2 to compute solu-
tions for various values of
C
.
Most calculations were performed with N
1
= 320, N
2
= 360,
1
= 0.02
and
2
= 0.01. We also calculated solutions with smaller values of
1
and
2
and larger values of N
1
and N
2
and checked that the results to
be presented here were independent of these parameters within graphical
accuracy. An example of such a check is presented at the end of this section.
Typical free-surface proles are shown in Figure 8.11.
There is a wave train on the upstream free surface. For large values of
F (let us say F > 2.4), the waves are so small that they cannot be seen
on the gures and the proles are essentially at far upstream (see Figures
8.11(a), (b)). However, for F < 2.4, the waves are clearly noticeable on
the proles (see Figures 8.11(c), (e)). As F decreases, the waves become
large-amplitude nonlinear waves with broad troughs and sharp crests (this
tendency is beginning to shown in Figure 8.11(e)). We expect that they will
ultimately reach the limiting conguration with a 120
angle at their crests

(see Figure 6.20) as F is further decreased.
The proles of Figures 8.11(a)(e) show that the mean elevation of the up-
stream free surface increases as F increases. Since the ux UH
is normalized to 1, it follows that the mean velocity far upstream decreases
as F increases (this is also consistent with (8.65), which is a valid ap-
proximation when the waves are of small amplitude). Therefore the phase
velocity of the waves decreases as F increases. This explains why the wave-
length of the waves increases as we move from Figure 8.11(a) to Figure
8.11(e).
2
3
4
5
6
7
1
2
3
1
2.2
2.6
1.4
1.8
1.4
1.6
1.8
2.0
2.2
1.2
1.0
2
3
4
1
0 5
0 5 0
0 5
0 5
(a) (b)
(c)
(e)
(d)
Fig. 8.11. Computed proles of the free surfaces in the vicinity of the gate. The
arrows indicate the position of the point at which the downstream free surfaces
separate from the gate. (a)
C
= 0.71 and F = 3.25; (b)
C
= 0.41 and F = 2.41;
(c)
C
= 0.26 and F = 2.03; (d)
C
= 0.19 and F = 1.83; (e)
C
= 0.075 and
F = 1.51.
In Figure 8.12, we show values of the contraction ratio C
c
(dened in
(8.66) versus y
C
/y
B
. Here y
C
and y
B
are the ordinates of the points C and
B (see Figure 8.8). As y
C
/y
B
approaches zero, y
B
and F . The
problem reduces then to a classical free streamline ow (3.39) and
C
c
=

+ 2
0.615
0.610
0.605
0.600
0.595
0.590
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
y
C
/y
B
C
c
c
versus y
C
/y
B
. The symbols correspond
to the calculations of Fangmeier and Strelko [54].
(see (3.44)). The dots in Figure 8.12 are numerical values taken from Figure
13 in Fangmeier and Strelko [54]. These numerical values are in good
agreement with ours for small values of y
C
/y
B
(i.e. for large values of F).
This is consistent, since Fangmeier and Strelko assumed that there are
no waves upstream and our results show that the waves are of very small
amplitude for F large.
In Figure 8.13, we present an example of the checks that we used to test
the accuracy of the numerical results.
The two curves, which lie close together, are computed upstream free
surfaces for
C
= 0.19. Figure 8.13 shows that the results are independent
of
U
N
1
.
The results presented in this section show that there is a wave train on the
upstream free surface. The amplitude of these waves is dierent from zero,
except in the limit F . This indicates the nonexistence of solutions
satisfying the radiation condition, which requires there to be no waves far
upstream. In particular the numerical procedure of Section 8.3.2 diverges
if we try to impose the radiation condition by, for example, forcing the free
surface to be at far upstream.
In the calculations presented here we have assumed that B is a stagnation
point. We could have assumed instead that the upstream free surface is
tangential to the gate at the separation point (see Figure 8.2). Such solutions
were calculated by Binder and Vanden-Broeck [15]. Their results showed
2.7
2.6
2.5
2.4
2.3
0
Fig. 8.13. Computed proles of the upstream free surface with N
1
= 320, N
2
= 360,
1
= 0.04 and
2
= 0.01 (solid curve) and N
1
= 320, N
2
= 360,
1
= 0.02 and
2
= 0.01 (broken curve).
that there is always a train of waves upstream. Recently Binder and Vanden-
Broeck [16] showed analytically and numerically that these upstream waves
can be eliminated if a disturbance is introduced in front of the gate.
8.4 Pure capillary free-surface ows
In this section we supplement the ndings of Section 8.1 by considering the
ow of Figure 8.1 with gravity neglected but surface tension included in
the dynamic boundary condition (i.e. g = 0 and T ,= 0). As we shall see,
there is then a train of pure capillary waves as x . Pure capillary
free-surface ows were considered in Chapter 3 (e.g. the ow emerging from
the container of Figure 3.1 and the cavitating ow of Figure 3.16). The
absence of waves in these ows is justied because a wave train as x
would violate the radiation condition. However, waves are compatible with
the radiation condition if the direction of the ow (i.e. the direction of U)
is reversed in Figure 8.1. Following the discussion at the end of Section
8.2.2, we refer to such ow as a bow ow. It models the ow in front of the
semi-innite plate AB when it moves to the right at a constant velocity U,
as viewed in a frame of reference moving with the obstacle.
We compute solutions for the ow of Figure 8.1 by a boundary integral equa-
tion method similar to that used in Section 8.1.1. The numerical procedure
follows that described in Anderson and Vanden-Broeck [6]. As y ,
the ow approaches a uniform stream with constant velocity U. We in-
troduce dimensionless variables by taking T/(U
2
) as the reference length
and the velocity U as reference velocity. We introduce the complex velocity
u iv and dene and by (3.3). The dynamic boundary condition gives
in dimensional variables
1
2
(u
2
+v
2
) +K = B, (8.99)
where K is the curvature of the free surface and B is the Bernoulli constant.
Using (3.3) and (3.6) we rewrite (8.99) in terms of and as
1
2
e
2
e

= B, (8.100)
where () and

() are the values of and on the free surface = 0.
For the present problem B = 1/2 since, for periodic waves of wavelength
in innite depth,
_

0
(u
2
+v
2
)dx = and
_

0
Kdx = 0. (8.101)
The latter is a general property of periodic waves; the former is only valid
for innite depth and may be found in Wehausen and Laitone [196].
Equation (8.100) provides a relation between and on the free surface
= 0, > 0. A second relation is provided by (8.9).
Our experience with capillary ows in Section 3.2 suggests that the free
surface does not leave the plate tangentially: there is an angle ,= 0 be-
tween the tangent to the free surface and the plate at the separation point
B (see Figure 8.1). This expection is conrmed by the analytical results
presented below in Section 8.4.2. There we show analytically that there are
no solutions with = 0 except for the trivial solution corresponding to a
uniform stream with constant velocity 1 and a at free surface. However,
we will not use this analytical insight when solving the problem numerically;
rather, we will assume that there is an angle (measured counterclockwise
from the x-axis) at the separation point and nd its value numerically. This
allows solutions that leave the plate tangentially ( = 0) and also solu-
tions with a discontinuity in slope ( ,= 0). The computed values of are
then compared with the exact analytical values to check the accuracy of the
numerical procedure.
We introduce the mesh points
I
= (I 1)E, I = 1, . . . , N + 1, (8.102)
the midpoints
M
I
=
1
2
(
I
+
I+1
), I = 1, . . . , N, (8.103)
and the unknowns
I
= (
I
), I = 1, . . . , N + 1. (8.104)
We dene the angle by imposing
1
= . (8.105)
Now we evaluate (
M
I
) by applying the trapezoidal rule to (8.9) with sum-
mation over the points (8.102) (see Section 3.1.2 for a justication of the
discretisation); a truncation procedure similar to that presented in Section
8.1.1 is used. We then require (8.100) to hold at the midpoints (8.103). This
leads to N equations for the N +1 unknowns
I
, I = 1, . . . , N +1. The last
equation xes an extra parameter (for example the value of at some point
on the free surface). Once again the system is solved by Newtons method.
Typical free-surface proles are shown in Figures 8.14 and 8.15.
(a)
(b)
0 5 10 15 20 25
0
0 5 10 15 20 25
x
x
0.10
0.05
y
0.15
0.10
0.05
0
y
Fig. 8.14. Computed free-surface proles for (a) = /48 and (b) = /48.
Taken from Proc. Roy. Soc. A 452, 19851997 (1996).
2
1
0
5 0 10 15 20 25
5 0 10 15 20 25
2
1
0
(a)
(b)
x
x
y
y
Fig. 8.15. Computed free-surface proles for (a) = /4 and (b) = /4. Taken
from Proc. Roy. Soc. A 452, 19851997 (1996).
For [[ small, the waves are close to linear sine waves (see Figure 8.14).
As [[ increases, the waves become nonlinear with broad crests and narrow
troughs (see Figure 8.15), in accordance with Crappers exact solution (see
Section 6.5.1).
We use the conservation of momentum equation (8.19) with g = 0, i.e.
_
S
_
u(V n) +
p
n
x
_
d = 0, (8.106)
to derive an exact relationship between the angle , the steepness s and the
wavelength of the waves in the far eld. The derivation below is similar
to that used in Section 8.1.2 (see also [6]). However, it is entirely analytical
because the train of waves in the far eld is described by Crappers exact
solution (see Section 6.5.1). We will derive the exact relation between s,
and for innite depth by rst considering the ow past a plate in nite
depth d and then taking the limit as d . We let y = 0 represent the
mean free-surface level of the waves in the far eld and y = H the level of
the plate. We dene a reference velocity U such that u = U + u, where
_
x
0
+
x
0
udx = 0 (8.107)
as x
0
, the integral being taken at an arbitrary level y inside the ow
region. Bernoullis equation gives in the ow domain
1
2
(u
2
+v
2
) +
p
=
1
2
U
2
, (8.108)
where the right-hand side is Bernoullis constant. As d , U
U.
As in Section 8.1.2, we choose S in (8.106) to consist of the plate S
p
, a
vertical line S
R
at x = M, a horizontal line S
H
at y = d and a vertical line
S
L
at x = M (see Figure 8.1). Here M is a large positive real number.
There is no contribution to (8.106) from S
p
and S
H
since on these curves
n
x
= 0 and u n = 0. On S
L
, n
x
= 1 and the contribution to (8.106) is
_
H
d
_
U
2
+
p
_
dy, (8.109)
where U
is the constant velocity of the uid as x .

On S
R
, n
x
= 1 and we have likewise the contribution
_

d
_
u
2
+
p
_
dy. (8.110)
Using the fact that p = TK along the free surface, where K denotes its
curvature, we nd that
_
S
F
pn
x
d = T(cos cos ). (8.111)
Here denotes the angle between the free surface and the x-axis at the point
C where S
R
intersects the free surface S
F
(see Figure 8.1). Since u n = 0
on S
F
, we have
_
S
F
_
u(u n) +
p
n
x
_
d =
T
(cos cos ). (8.112)

Altogether, then, from (8.106) we obtain
0 =
_
S
L
+S
R
+S
F
_
u(u n) +
p
n
x
_
d
=
_
H
d
_
U
2
+
p
_
dy +
_

d
_
u
2
+
p
_
dy
T
(cos cos ) . (8.113)

Using (8.108) we have
_
H
d
_
U
2
+
p
_
dy =
_
H
d
1
2
U
2
+
_
1
2
U
2
+
p
_
dy
=
_
H
d
1
2
U
2
+
1
2
U
2
dy =
1
2
(U
2
+U
2
)(d H). (8.114)
Following Hogan [74], we dene the mean wave momentum
I =
_

d
udy (8.115)
and the excess momentum ux due to the waves
S
xx
=
_

d
u
2
+
p
dy +
T
(1 cos ), (8.116)
where the overbar denotes an average over one wavelength. These are related
to the average kinetic energy K and the average potential energy V by the
relations
I =
2K
U
, (8.117)
S
xx
= 4K V + (U
2
U
2
)d (8.118)
(see Hogan [74]). Hence
_

d
u
2
+
p
dy =
_

d
(U + u)
2
+
p
dy
= U
2
_

d
dy + 2U
_

d
udy +
_

d
_
u
2
+
p
_
dy
= U
2
d + 2UI +S
xx
(1 cos )
= V +U
2
d
T
(1 cos ). (8.119)
Taking the average in equation (8.113) over one wavelength, we have
0 =
_
1
2
(d H)(U
2
+U
2
)
_
+
_
V +dU
2
(1 cos )
_
+
_
T
(1 cos )
T
(1 cos )
_
= V
1
2
(U
2
U
2
)d +
1
2
(U
2
U
2
)d
(1 cos ) +
1
2
(U
2
+U
2
)H. (8.120)
Equating the mass ux on the left- and right-hand sides of the surface S,
we have
U
(d h) = Ud +
_

d
udy = Ud
2K
U
. (8.121)
Multiplying by U and U
and adding the two resulting equations gives
1
2
(U
2
U
2
)d = K
_
1 +
U
U
_
1
2
(U
2
+UU
)H (8.122)
Substituting this into equation (8.120) gives
0 = V +K
_
1 +
U
U
_
(1cos ) +
1
2
(U
2
UU
)H+
1
2
(U
2
U
2
)d.
(8.123)
This is an exact relation for ow in nite depth. As d , we have
U
U, U
U and (U
2
U
2
)d 0. Hence for innite depth
0 = V + 2K
T
(1 cos ). (8.124)
The energies V and K were calculated by Hogan [74] using Crappers exact
solution. They are given by
V =
2T
_
_
1 +
1
4
2
s
2
_
1/2
1
_
(8.125)
and
K =

2
s
2
T
4
_
1 +
1
4
2
s
2
_
1/2
. (8.126)
Substituting these into equation (8.124) gives
cos = 2
_
1 +
1
4
2
s
2
_
1/2
1. (8.127)
Equation (8.127) shows that if there are waves on the free surface (i.e.
s ,= 0) then the ow will not leave the plate tangentially. The numerical
results of Section 8.4.1 conrm that are indeed waves on the free surface.
Equation (8.127) provides a check on the numerical results. For example, on
the one hand the numerical proles of Figures 8.14 show waves of steepness
0.0292 in the far eld. On the other hand, (8.127) with = /48 predicts
s = 0.0295. The theoretical and numerical values of the steepness agree
within one per cent.
9
Waves with constant vorticity
Two fundamental approaches have been used in the previous chapters to
calculate free-surface ows. The rst involves perturbing known exact so-
lutions. Often these exact solutions are trivial, e.g. a uniform stream. To
leading order this approach gives a linear theory (see for example the calcu-
lations of Chapter 4) and at higher order a weakly nonlinear theory (see for
example the small-amplitude expansions and the Kortewegde Vries equa-
tion of Chapter 5).
In the second approach fully nonlinear solutions are computed. This ap-
proach involves a discretisation leading to a system of nonlinear algebraic
equations, which is then solved by iteration (e.g. using Newtons method).
Iteration requires the choice of an initial guess. These initial guesses are
often trivial solutions or asymptotic solutions derived in the rst approach.
After convergence of the iterations, the solution obtained is then used as
an initial guess to compute a new solution for slightly dierent values of
the parameters. For example, the linear solutions of Section 2.4 were used
as an initial guess in Chapter 6 to compute a nonlinear solution of small
amplitude. This solution was then used as an initial guess to compute a so-
lution of larger amplitude and so on. This method of continuation leads to
families of solutions; an application is the continuation in used in Section
7.1.1. We can then investigate whether other solution branches bifurcate
from these branches (see Section 6.5.2.1 for an example). A limitation of
this second approach is that we can only compute branches of solutions that
extend the simple initial guesses we have tried. It is always possible that
we are missing solution branches that could be calculated by starting with
other, unknown, initial guesses.
In this chapter we will solve a problem for which additional solution
branches (not connected by continuation to simple initial guesses) are found:
the eect of vorticity on water waves. The assumption of irrotationality used
244
in Chapters 5 and 6 is often a very good approximation. However, vorticity
can be generated by wind stress or by friction at the bottom of a channel.
Here we assume for simplicity that the vorticity is constant in the uid. This
enables us to reduce the problem to one described by Laplaces equation.
Solitary waves and periodic waves are considered in Sections 9.1 and 9.2
respectively.
9.1 Solitary waves with constant vorticity
9.1.1 Mathematical formulation
We consider a two-dimensional solitary wave in an inviscid incompressible
uid (see Figure 9.1).
5
4
3
2
1
0
0 5 10 15
x
y
Fig. 9.1. The ow and the coordinates. The free surface shown is a computed
solution for = 0.474, A = 3.8 and G = 0. Taken from J. Fluid Mech. 274, 340
(1994).
The uid is bounded below by a horizontal bottom. The ow is assumed
to be rotational and characterized by a constant vorticity . We take a
frame of reference with the x-axis along the horizontal bottom and in which
the ow is steady. We assume that the ow is symmetric with respect to
the y-axis.
We denote by H the depth at innity and by c the velocity on the free
surface at innity. The variables are made dimensionless by choosing H as
the unit length and c as the unit velocity.
246 Waves with constant vorticity
The ow is characterized by the following parameters:
=
H
c
, (9.1)
G =
gH
c
2
, (9.2)
=
A
H
. (9.3)
Here g is the acceleration of gravity, counted as positive when acting verti-
cally downwards, and A is the elevation of the wave crests above the level of
the free surface at innity. The parameter is the dimensionless vorticity, G
is the dimensionless gravity and is an amplitude parameter. When = 0,
the problem reduces to that considered in Sections 5.2 and 6.6.1.
Benjamin [12] derived asymptotic solutions for solitary waves of small
amplitude. His results showed that for each value of there is a one-
parameter family of solutions that bifurcates from the uniform shear ow at
the critical value
G = 1 +. (9.4)
Teles da Silva and Peregrine [150], Pullin and Grimshaw [127] and Vanden-
Broeck [178], [179] calculated numerically solutions for solitary waves with
constant vorticity. Their results extended Benjamins results for waves of
nite amplitude. We shall show that there are in addition solution branches
that are not connected by continuation to Benjamins solutions. The analysis
below follows [178] and [179].
It is convenient to describe the problem in terms of a streamfunction
(x, y) satisfying
2
= (9.5)
in the ow domain. Following [141], [150], [127], [178] and [179], we can
reduce the problem to one described by Laplaces equation by subtraction
of a particular solution of (9.5). Thus if we write
=

2
y
2
+ (1 +)y (9.6)
then
2
(x, y) = 0. Therefore the quantity w(z) = u iv =
y
+ i
x
is an analytic function of z = x + iy, where the uid velocity vector is
(u (y 1) +1, v). We satisfy the kinematic condition that v must equal
zero on the bottom by reecting the ow in the bottom. The function w(z)
vanishes at innity. Hence the Cauchy integral formula (2.41), when z is on
the free surface, gives
w(z) =
1
i
_
C
w()d
z
(9.7)
with a Cauchy principal-value interpretation. Here C denotes the free sur-
face and its image in the bottom.
We parametrize the free surface by x = X(t), y = Y (t), where t is the arc
length with t = 0 at the crest of the wave. Then
X
(t)
2
+Y
(t)
2
= 1, (9.8)
X(0) = 0, Y (0) = 1 +, (9.9)
where is the amplitude parameter dened in (9.3). We now consider u and
v to be functions of t. Taking the real part of (9.7) and using the symmetry
of the wave with respect to the y-axis, we obtain, after some algebra,
u(t) =
_
0
X
st
[u(s)Y
(s) v(s)X
(s)] Y
st
[u(s)X
(s) +v(s)Y
(s)]
(X
st
)
2
+ (Y
st
)
2
ds
_
0
X
s+t
[u(s)Y
(s) v(s)X
(s)] Y
st
[u(s)X
(s) +v(s)Y
(s)]
(X
s+t
)
2
+ (Y
st
)
2
ds
+
_
0
X
st
[v(s)X
(s) u(s)Y
(s)] +Y
s+t
[u(s)X
(s) +v(s)Y
(s)]
(X
st
)
2
+ (Y
s+t
)
2
ds
+
_
0
X
s+t
[v(s)X
(s) u(s)Y
(s)] +Y
s+t
[u(s)X
(s) +v(s)Y
(s)]
(X
s+t
)
2
+ (Y
s+t
)
2
ds,
(9.10)
where X
st
= X(s) X(t) and Y
st
= Y (s) Y (t); equation (9.10) holds
for all 0 < t < .
On the free surface, the kinematic condition and Bernoulli equation yield
(u + + 1 Y )Y
(s) = vX
(s), (9.11)
[u + 1 (Y 1)]
2
+v
2
+ 2GY = 1 + 2G. (9.12)
Here G is the gravity parameter dened in (9.2). Equation (9.12) expresses
the fact that the pressure is constant on the free surface.
We x two of the three parameters , G and and seek four functions
u, v, X
and Y
and the value of the third parameter such that the system
(9.8), (9.10)(9.12) is satised.
We are seeking a numerical solution of the nonlinear integro-dierential sys-
tem (9.8), (9.10)(9.12). First we dene N distinct mesh points on the free
surface by specifying values of the arc length parameter t = S
I
, where
S
I
= E(I 1), I = 1, . . . , N; (9.13)
here as before E is the interval of discretization. We shall also make use of
the intermediate mesh points S
I1/2
= (S
I+1
+S
I
)/2, I = 1, . . . , N 1.
We now dene the 4N corresponding fundamental unknown quantities
u
I
= u(S
I
), I = 1, . . . , N, (9.14)
v
I
= v(S
I
), I = 1, . . . , N, (9.15)
X
I
= X
(S
I
), I = 1, . . . , N (9.16)
and
Y
I
= Y
(S
I
), I = 1, . . . , N. (9.17)
We will estimate the values of the x- and y- coordinates X
I
= X(S
I
),
Y
I
= Y (S
I
) in terms of the fundamental unknowns by the trapezoidal rule,
i.e. X
1
= 0, Y
1
= 1 + and
X
I
= X
I1
+X
(S
I3/2
)E, I = 2, . . . , N, (9.18)
Y
I
= Y
I1
+Y
(S
I3/2
)E, I = 2, . . . , N, (9.19)
where X
(S
I3/2
) and Y
(S
I3/2
) are evaluated from X
I
and Y
I
by a four-
point interpolation formula. We satisfy (9.11) and (9.12) at the mesh points
S
I
, I = 1, . . . , N 1 and (9.8) at S
I
, I = 1, . . . , N. This yields 3N 2
nonlinear equations. Next we evaluate X(S
I1/2
) and Y (S
I1/2
) by four-
point interpolation.
We then satisfy (9.10) at the point t = S
I1/2
, I = 1, . . . , N 1, by
applying the trapezoidal rule to (9.10), with a sum over the points s =
S
J
, J = 1, . . . , N. The symmetry of the discretisation and of the trapezoidal
rule with respect to the singularity of the integrand at s = t enables us
to evaluate this Cauchy principal-value integral by ignoring the singularity,
with an accuracy no less than for a nonsingular integral (see Section 3.1.2.2
for a justication). This yields N 1 extra nonlinear equations.
Four more equations are obtained by imposing the symmetry condition
Y
1
= 1 (9.20)
and the far-eld conditions
Y
N
= 0, (9.21)
u
N
= 0, (9.22)
v
N
= 0. (9.23)
We now have 4N + 1 equations for the 4N + 1 unknowns (9.14)(9.17)
and G. This system is solved by Newtons method for given values of
and . To obtain the results presented in the next section, we also used a
variation of the scheme in which we xed G and and found as part of
the solution.
9.1.3 Discussion of the results
We used the numerical scheme of Section 9.1.2 to compute solutions for
various values of the parameters. In all cases presented we checked that the
numerical results were independent of E and N within graphical accuracy.
9.1.3.1 Solitary wave branches
Numerical values of versus G are shown in Figure 9.2 for various values
of .
For all values of and G, the uniform shear ow = 0, u = 0, v = 0,
X
= 1 and Y
= 0 is a trivial solution. This branch of trivial solutions

corresponds to the G-axis in Figure 9.2. The solitary waves are solution
branches that bifurcate from this trivial solution at the critical values (9.4).
Benjamin [12] derived an asymptotic solution for small values of . In
particular he obtained the following relation between G, and (see his
equation (46)):
G = 1 +
_
1 + +

2
3
_
. (9.24)
Our numerical values agree with (9.24) as approaches zero.
As we progress along the solution branches, we nd that there is a critical
value
c
0.32 of such that dierent limiting congurations occur
for <
c
and for >
c
. For >
c
, the limiting conguration is
0.4 0.8
0
4
8
G
Fig. 9.2. Values of versus G for various values of . The curves from right to
left correspond to = 0.11, 0.11, 0.32, 0.35 and 0.8. The broken curve
corresponds to (9.25). Taken from J. Fluid Mech. 274, 344 (1994).
characterized by a stagnation point at the crest with a 120
angle. Using
(9.12), we nd that these limiting congurations satisfy
=
1
2G
. (9.25)
Relation (9.25) corresponds to the broken line in Figure 9.2. These solu-
tion branches were studied by Pullin and Grimshaw [127] and Teles Da Silva
and Peregrine [150]. In the particular case = 0, the solutions agree with
previous numerical calculations for solitary waves in the absence of vorticity
(see Section 6.6.1). Typical proles are shown in Figures 9.3(a), (b).
For <
c
a new phenomena occurs: the solution branches in Figure
9.2 extend for arbitrarily large values of without intersecting the broken
curve. As , the wave approaches a closed region of constant vor-
ticity in contact with the bottom of the channel. Below we consider these
limiting congurations further. A typical prole for <
c
is shown in
Figure 9.3(c).
Figure 9.2 shows that there are solitary waves with constant vorticity in
the absence of gravity (i.e. for G = 0). These solutions form a one-parameter
family of solutions. Numerical values of versus for G = 0 are shown
in Figure 9.4. There is again a trivial solution corresponding to a uniform
1.0
2.0
3.0
1.00
1.10
1.20
(a) (b)
2.0
1.6
1.2
0.8
0 5 10 15
0 5 10 15
0 5 10 15
(c)
Fig. 9.3. Computed free-surface prole for: (a) = 0.11, A = 0.2 and G = 0.754;
(b) = 0.11, A = 1.0 and G = 0.49 (this solution is close to the limiting
angle at the crest); (c) = 0.8, A = 2.5 and G = 0.19.

12
8
4
0
0.0
Fig. 9.4. Values of versus for solitary waves without gravity, i.e. for G = 0.
shear ow. This trivial solution corresponds to the -axis in Figure 9.4. The
solitary waves bifurcate from this trivial solution at = 1.
A typical free-surface prole for G = 0 is shown in Figure 9.1. As ,
the wave approaches a circular region made up of uid in rigid body rotation
and in contact with the bottom of the channel at a single point. Such
a conguration was suggested by Teles da Silva and Peregrine [150] as a
possible limiting conguration for solitary waves without gravity.
The solution branches for 1 < <
c
in Figure 9.2 intersect the -axis.
As discussed above, the corresponding solutions are the solutions without
gravity of Figure 9.4. For
c
< < 0, the solution branches of Figure 9.2
do not intersect the -axis and the solutions without gravity are members of
a new solution branch. These branches were computed by continuation: we
used the solution without gravity as the initial guess to compute a solution
for a small value of G. This solution was then used as the initial guess to
compute a solution for a larger value of G and so on. Values of versus G
for the new branch corresponding to = 0.11 are shown in Figure 9.5.
30
40
50
60
0
G
Fig. 9.5. Values of versus G for the new branch with = 0.11. The cross
corresponds to the solution (9.26).
Typical free-surface proles are shown in Figures 9.6(a)(d).
The new branches do not bifurcate from a uniform shear ow and evolve
between two dierent limiting congurations.
The rst limiting conguration is a wave whose prole has a point of
contact with itself (see Figure 9.6(a)). It is a wave without gravity (i.e.
G = 0) and the closed region is a circular domain made up of uid in rigid
body rotation. The radius of the circular domain is A/2. Since the velocity
on the free surface is 1 when G = 0, we have
A =
4
. (9.26)
0
40
80
120
0
20
40
60
0 40
0 20 40
0
40
80
120
0 40
0
20
40
60
0 20 40
(a) (b)
(c) (d)
Fig. 9.6. Computed free-surface proles for (a) = 0.11, A = 38 and G = 0.0016;
(b) = 0.11, A = 35.5 and G = 0.0064; (c) = 0.11, A = 41.5 and G = 0.02;
(d) = 0.11, A = 50.5 and G = 0.034.
For = 0.11, (9.26) gives A = 36.364. This value corresponds to the
cross in Figure 9.5. It is consistent with our numerical calculations.
The second limiting conguration is the same as for the solution branches
with <
c
in Figure 9.2. It is a closed region of constant vorticity in
contact with the bottom of the channel along a segment. The approach to
this limiting conguration is apparent in Figures 9.6(c), (d).
As increases, the curve in Figure 9.5 moves upwards. We expect that
this curve will coalesce into the point = , G = 0 as 0 and
that the new solutions do not exist for > 0. This is suggested by the
fact that the solution corresponding to G = 0, = and = 0 has
the properties of both the rst and the second limiting conguration (i.e.
the prole of the wave has a point of contact with itself and is also in con-
tact with the bottom). As approaches
c
from above, we expect the new
family of solutions to merge with the solution branches that bifurcate from
a uniform shear ow.
The results rst discussed show that there are solitary waves of arbi-
trary large amplitude. Figures 9.6(c), (d) suggest that these waves approach
closed regions of constant vorticity touching the bottom along a segment as
. These closed regions are dicult to compute accurately using the
previous numerical scheme because more and more mesh points are needed
on the free surface as increases. Therefore we shall calculate these limiting
congurations directly (see Figure 9.7(a)).
1.2
0.8
0.4
0
0 1.0
1.2
0.8
0.4
0
0 0.4 0.8
1.2
0.8
0.4
0
0 0.5 1.0
(a) (b)
(c)
Fig. 9.7. Computed solutions for a closed region of constant vorticity in contact
with a wall with (a) = 0.045 and = 3.04; (b) = 0.6 and = 3.33;
(c) = 0.21 and = 4.34. Taken from J. Fluid Mech. 258, 105113 (1994).
We rst rescale the variables by choosing the velocity Q at the separation
points (i.e. the extremities of the segment of contact) as the unit velocity
and the unit length such that the total length of the free surface is equal
to 4. We choose cartesian coordinates with the origin in the middle of the
segment of contact. Then the problem is characterized by the dimensionless
gravity
=
2gL
Q
2
(9.27)
and the dimensionless vorticity
=
L
Q
. (9.28)
Following the formulation of Section 9.1.1 and Vanden-Broeck and Tuck
[193], we introduce the streamfunction and write
=

2
y
2
; (9.29)
then
2
(x, y) = 0. As in Section 9.1.1, w(z) = uiv =
y
+i
x
is an anal-
ytic function of z = x + iy, where the uid velocity is now (u y, v). We
parametrise the free surface by x = X(t), y = Y (t), where t is the arc length,
with t = 0 at the left-hand separation point. Then X
(t) and Y
(t) satisfy
(9.8) and X(2) = 0. Proceeding as in Section 9.1.1, we nd that the integro-
dierential equation (9.10) still holds if the upper limit is replaced by 2.
(u Y )Y
(s) = vX
(s), (9.30)
(u Y )
2
+v
2
+Y = 1. (9.31)
This completes the formulation of the problem. For a given value of
we seek four functions, u, v, X
and Y
, satisfying (9.8)(9.10), (9.30) and

(9.31). The value of comes as part of the solution.
We solved the problem using a numerical procedure similar to that de-
scribed in Section 9.1.2. Most results were obtained with 60 equally spaced
mesh points between s = 0 and s = 2.
Typical free-surface proles are shown in Figures 9.7(a)(c). For = 0,
the solution is a circular region of uid in rigid body rotation with one point
of contact with the bottom. The corresponding value of is . A graph of
versus is shown in Figure 9.8.
3
1 3 5 7
5
7
Fig. 9.8. Values of versus .

We found that there is a solution for each value of < 0. As becomes
more negative, increases and the length of the segment of contact increases.
Vanden-Broeck and Tuck [193] considered closed regions of constant vor-
ticity in contact with curved boundaries. The results that we have discussed
generalize some of their ndings by including the eect of gravity.
9.1.3.2 More branches of solitary waves
In Section 9.1.3.1 it was shown that there are branches of solitary waves
that do not bifurcate from a uniform shear ow. These branches start from
a limiting conguration consisting of a circular region of uid in rigid body
rotation with no gravity, i.e. with G = 0 (see Figure 9.10). As one progresses
along the solution branch in Figure 9.5 G increases and then decreases back
to 0. The corresponding solution is a solitary wave with the familiar shape of
a single hump (see Figure 9.9). As one continues along the branch G becomes
negative and the wave ultimately approaches the form of a closed region of
constant vorticity in contact with the bottom of the channel. Solutions with
G negative were calculated in the Section 9.1.3.1 and will not be considered
further.
0
5
10
15
20
25
30
0 5
x
y
10 15 20
Fig. 9.9. Computed solution for = 0.245 and G = 0. Taken from J. Mech. B
(Fluids) 14, 761771 (1995).
The two solutions without gravity shown in Figures 9.9 and 9.10 can be
combined to generate further solutions without gravity. For example, we
can have two circular regions of uid in solid body rotation on top of each
other (see Figure 9.11(b)) or the solitary wave of Figure 9.9 with one or two
circular regions of uid in rigid body rotation on top of its crest (see Figures
9.11(a), (c).
0
5
10
15
20
25
30
0 5 10 15 20
Fig. 9.10. Solution for G = 0 consisting of a circular region of uid in rigid body
rotation. The value of is = 0.245.
In fact, an arbitrary number of circular regions of uid in rigid body
rotation can be added in a similar way. Since the velocity is constant on the
free surface when G = 0, the radius R of the circular regions in rigid body
rotation are related to by the simple formula
R
H
=
2
[[
. (9.32)
The solutions of Figures 9.11(a)(c) are trivial extensions of the solutions of
Figures 9.9 and 9.10. However, it is not obvious that these solutions can be
perturbed to generate solutions with G ,= 0.
Below, however, we will show that there are branches of solitary waves
with gravity that approach the solutions in Figures 9.11(a)(c) as G 0.
These new branches do not bifurcate from a uniform shear ow. We found
them by rst calculating a solution close to the limiting conguration with
a 120
angle at the crest of the wave. We then computed the corresponding

family of solutions by a continuation method. All the solutions presented
were obtained by using the numerical scheme described in Section 9.1.2.
To construct the new families of solutions, we look rst at solutions which
are close to the limiting conguration with a 120
degree angle at the crest

of the wave. These limiting congurations are characterized by a stagnation
point at their crest. Using this property and (9.12), we nd that they
satisfy
=
1
2G
. (9.33)
60
40
20
0
60
(a)
(b)
(c)
40
20
0
60
40
40 20 20 0 40
20
0
Fig. 9.11. (a) Solution for G = 0 consisting of the solitary wave of Figure 9.9 with
one circular region of uid in rigid body rotation on top of its crest. The value of
is 0.245. (b) Solution for G = 0 consisting of two circular regions of uid in
rigid body rotation on top of each other. The value of is 0.245. (c) Solution
for G = 0 consisting of the solitary wave of Figure 9.9 with two circular regions of
uid in rigid body rotation on top of its crest. The value of is 0.245.
We dene waves close to the limiting conguration with a 120
angle as
solutions for which 2G is close to 1. In Figure 9.12, we show values of
versus for such waves with 2G = 0.96.
d
a
b
c
0
5
10
15
20
25
30
35
40
Fig. 9.12. Values of versus for solitary waves with 2G = 0.96. These waves
are close to the limiting conguration with a 120
angle at the crest. The branches

associated with points a, b, c and d are shown in Figure 9.13.
As increases, the values of oscillate between a succession of minima
and maxima. The lowest minimum in coincides with the critical value
c
introduced in the Section 9.1.3.1. For >
c
, the solution branches
that bifurcate from a uniform shear ow at the critical values (9.4) have a
limiting conguration with a 120
angle at their crests; see Teles da Silva

and Peregrine [150], Pullin and Grimshaw [127] and the discussion in Section
9.1.3.1. Each limiting conguration corresponds to a point on the lower part
of the curve in Figure 9.12 with >
c
.
For <
c
, the solution branches that bifurcate from a uniform shear
ow at the critical values (9.4) do not have limiting congurations with
a 120
angle at their crests; see the references mentioned above. As we

showed, these branches approach a conguration similar to that of Figure
9.9 as G 0 and ultimately tend to closed regions of constant vortic-
ity in contact with the bottom of the channel if G is allowed to become
negative.
This discussion shows that the lower portion of the curve in Figure 9.12
for >
c
consists of limiting congurations of branches that bifurcate
from a uniform shear ow at the critical values (9.4). Each point on the
remaining part of the curve of Figure 9.12 corresponds to a solution close to
the limiting conguration for a new family of waves, which does not bifurcate
from a uniform shear ow.
We calculate these new families by continuation, i.e. we x a value of
associated with a point on the curve of Figure 9.12 and use the corresponding
solution as the initial guess to compute a solution for a slightly perturbed
value of or G; this solution is then used as the initial guess for a further
calculation. Explicit calculations were performed for the points a, b, c and d
of Figure 9.12. The solution branches are shown in Figure 9.13, where we
present values of versus G.
6
12
18
24
30
36
42
0.03 0.06 0.09
G
d
e
a
b
c
Fig. 9.13. Values of versus G for = 0.245 (curves a, b, c), = 0.3 (curve d)
and = 0.267 (curve e). The curves a, b, c and d intersect the curve of Figure
9.12 at points a, b, c and d respectively. The broken curve corresponds to the
relation 2G = 0.96. The three crosses correspond, from bottom to top, to the
solutions without gravity of Figures 9.11(a)(c).
The broken curve corresponds to the relation 2G = 0.96. For conve-
nience we denote the branches associated with points a, b, c and d in Figure
9.12 by the same letters in Figure 9.13. Figure 9.13 shows that the new
solution branches do not bifurcate from a uniform shear ow (i.e. from the
G-axis in Figure 9.13). Furthermore the new branches can be extended up
to G = 0. However, our numerical scheme becomes more sensitive as G
approaches zero; the numerical calculations for curves a, b and c had to be
stopped at some small positive value of G (see Figure 9.13).
Typical free-surface proles are shown in Figures 9.149.16.
The proles of Figure 9.14 are for = 0.245 and correspond to the
family of solutions starting at the point a in Figure 9.12 (see curve a in
Figure 9.13). As G 0, the proles of Figure 9.14 suggest that the wave
approaches a conguration consisting of a single hump wave with a circular
region of uid in rigid body rotation at its crest. This conguration is shown
in Figure 9.11(a) and the corresponding value of is indicated by the lowest
cross on the -axis in Figure 9.13. The position of this cross is consistent
with our numerical calculations.
The proles of Figure 9.15 are for = 0.245 and correspond to the
family of solutions starting at the point b in Figure 9.12 (see curve b in
Figure 9.13). As G 0, the wave approaches two circular regions of uid
in rigid body on top of each other. This conguration is shown in Figure
9.11(b) and the corresponding value of is indicated by the middle cross
on the -axis in Figure 9.13.
The proles of Figure 9.16 are also for = 0.245 but correspond to
the family of solutions starting at the point c in Figure 9.12 (see curve c
in Figure 9.13). As G 0, the wave approaches a single hump wave
with two circular regions of uid in rigid body rotation at its crest. This
limiting prole is shown in Figure 9.11(c) and the corresponding value of
is indicated by the top cross on the -axis in Figure 9.13.
The proles of Figure 9.17 correspond to = 0.3. They are on the
solution branch starting at point d in Figure 9.12 (see curve d in Figure
9.13). As G 0, the wave proles approach a solitary wave with the shape
of a single hump. The corresponding prole is shown in Figure 9.17(c).
Next we show the connection between these new solutions and the solu-
tions in Section 9.1.3.1. There it was shown that there are solution branches
for
c
< < 0 that do not bifurcate from a uniform shear ow. These
branches evolve between two congurations with G = 0. The rst is a circu-
lar region of uid in rigid body rotation (see Figure 9.10) and the second is a
single hump conguration (see Figure 9.9). These branches were obtained
by rst calculating single hump solutions for G = 0 and
c
< < 0 and
(a)
(b)
(c)
60
50
40
30
20
10
0
60
50
40
30
20
10
0
60
50
40
30
20
10
0
0 20 40
Fig. 9.14. Computed free-surface proles for solutions on curve a of Figure 9.13.
The value of is 0.245. The values of G are 0.024 (Figure 9.14(a)); 0.018 (Figure
9.14(b)) and 0.005 (Figure 9.14(c)).
(c)
20 0 40
60
50
40
30
20
10
0
(b)
60
50
40
30
20
10
0
(a)
60
50
40
30
20
10
0
Fig. 9.15. Computed free-surface proles for solutions on curve b of Figure 9.13.
The value of is 0.245. The values of G are 0.02 (Figure 9.15(a)); 0.011 (Figure
9.15(b)) and 0.0045 (Figure 9.15(c)).
(c)
20 0 40
60
40
20
0
(b)
60
40
20
0
(a)
60
40
20
0
Fig. 9.16. Computed free-surface proles for solutions on curve c of Figure 9.13.
The value of is 0.245. The values of G are (a) 0.015; (b) 0.01 and (c) 0.0085.
0
4
8
12
16
(c)
0 5 10
(b)
0
4
8
12
16
(a)
0
4
8
12
16
Fig. 9.17. Computed free-surface proles for solutions on curve d of Figure 9.13.
The value of is 0.3. Part (a) corresponds to the intersection of curve d in
Figure 9.13 with the broken line in Figure 9.13. Part (b) is for G = 0.05 and
Part (c) is for G = 0.
then constructing the corresponding family of solutions by continuation;
see Section 9.1.3.1 for details. Such families are shown in Figure 9.5 for
= 0.11 and in Figure 9.13 for = 0.267 (see curve e). Corresponding
proles for = 0.267 are presented in Figure 9.18.
These results indicate that, as decreases, the branches in Section 9.1.3.1
approach the broken curve in Figure 9.13 and ultimately intersect it at
=
d
. Here
d
is between 0.3 and 0.267. For
c
< <
d
, the branch
0
10
20
30
(c)
0 10 20
0
10
20
30
(b)
0
10
20
30
(a)
Fig. 9.18. Computed free-surface proles for solutions on curve e of Figure 9.13.
The value of is 0.267. The proles correspond to (a) = 8.5, G = 0.054, (b)
= 15.9, G = 0.015 and (c) = 16.14, G = 0.0065.
in Section 9.1.3.1 becomes a branch that evolves between a single hump
solution and a point on the broken curve of Figure 9.13 (i.e. a solution close
to the limiting conguration with a 120
angle at its crest); an example of

such a family is curve d in Figure 9.13. These families exist up to =
c
.
For <
c
there are still families of solutions that start from the single
hump conguration with G = 0. However, they do not intersect the broken
curve of Figure 9.13 and can be extended up to the G-axis. These are the
solutions for <
c
that bifurcate from a uniform shear ow. Such families
of solutions are shown in Figure 9.2 for = 0.35 and = 0.8.
We have shown that there are families of solutions with gravity that ap-
proach the congurations of Figure 9.11 as G 0. Our solutions are not
exhaustive. Indeed, we expect that there are an innite number of families of
solutions that, as G 0, approach congurations similar to those of Figures
9.11(a)(c) but with an arbitrary number of regions of uid in rigid body
rotation. These could be obtained by extending the calculations in Figure
9.12 for higher values of . Similarly we expect that there are an innite
number of branches similar to that computed in Section 9.1.3.1 (i.e. similar
to curve e in Figure 9.13). These families should connect the congurations
of Figure 9.11, and their generalization with an arbitrary number of circular
regions, without intersecting the broken curve of Figure 9.13.
9.2 Periodic waves with constant vorticity
Simmen and Saman [141] obtained numerical solutions for periodic waves
with constant vorticity in water of innite depth. Their results showed that
the waves have either a limiting conguration with a 120
angle at the crest

or a trapped bubble at their troughs.
The results of Section 9.1.3 showed that there are solitary waves with con-
stant vorticity that approach limiting congurations with trapped bubbles at
their crests as the acceleration of gravity approaches zero (see Figure 9.11).
In this section we extend the calculations of Simmen and Saman [141]
by showing that there are additional families of periodic solutions that, like
the solitary waves of Section 9.1.3.2, approach limiting congurations with
trapped bubbles at their crests as the acceleration of gravity tends to zero.
Each bubble is circular and contains uid in rigid body rotation. The main
dierence between solitary and periodic waves is that there are no limiting
congurations with trapped bubbles at the troughs for solitary waves.
The problem is solved numerically by a boundary integral equation method.
The numerical approach is similar to that used in Section 9.1.2 for solitary
waves. It is also similar to the approach used by Simmen and Saman
[141]. The essential dierence between our scheme and that of Simmen and
Saman is that we do not scale the variables so that the acceleration of
gravity is 1. This enables us to consider waves in the limit as the accelera-
tion of gravity approaches zero. The new families of solutions exist only for
suciently large values of the amplitude. As we shall see, they are con-
structed by a continuation method similar to that used in Section 9.1.3.
The problem is formulated in Section 9.2.1. The numerical procedure is
discussed in Section 9.2.2 and the results are presented in Section 9.2.3. The
presentation follows Vanden-Broeck [180] closely.
9.2.1 Mathematical formulation
We consider a two-dimensional periodic wave in an inviscid incompressible
uid. The uid is of innite depth. The ow is assumed to be rotational
and characterized by a constant vorticity . We take a frame of reference
with the x-axis along the mean water level and in which the ow is steady.
Gravity is acting in the negative y-direction. We assume that the ow is
symmetric with respect to the y-axis.
The ow is described in terms of a streamfunction (x, y) satisfying
2
= (9.34)
in the ow domain.
As before, we reduce the problem to one described by Laplaces equation
by subtracting a particular solution of (9.34). Thus if we write
= +

2
y
2
cy (9.35)
then
2
(x, y) = 0. We require that
0 as y . (9.36)
This denes uniquely the quantity c in (9.35). Following Simmen and
Saman [141], we refer to c as the wave speed.
The variables are made dimensionless by choosing the wavelength as
the unit length and c as the unit velocity. In terms of the dimensionless
variables, (9.35) becomes
= +

2
y
2
y, (9.37)
where
=

c
. (9.38)
The quantity w(z) = uiv =
y
+i
x
is an analytic function of z = x+iy,
and the uid velocity vector is (u + y 1, v). The function w(z) vanishes
at innity.
We map the ow domain within the wavelength 1/2 < x < 1/2 from the
z-plane into the interior of a domain of the -plane by the transformation
= e
2iz
. (9.39)
Since w(z) is periodic in x with period 1, it gives rise to a single-valued
analytic function of . Applying the Cauchy integral formula (2.38) to the
function u iv in the -plane we obtain
w() =
1
i
_
C
w()d

, (9.40)
where C denotes the closed curve bounding the ow in the -plane. In
(9.40), is on C and the integral is a Cauchy principal value.
Suppose that the free surface is parametrised by x = X(t), y = Y (t),
where t is the arc length with t = 0 at the crest of the wave. Then
X
(t)
2
+Y
(t)
2
= 1, (9.41)
X(0) = 0, Y (0) = , (9.42)
where is the dimensionless elevation of the crests. We now consider u and
v to be functions of t. Taking the real part of (9.40) and using the symmetry
of the wave with respect to the y-axis, we obtain, after some algebra,
u(t) =
_
1/2
0
[u(p) iv(p)][2ix
(p) + 2y
(p)]
1 exp2i[x(t) x(p)] + 2[y(t) y(p)]
dp
_
1/2
0
[u(p) +iv(p)][2ix
(p) 2y
(p)]
1 exp2i[x(t) +x(p)] + 2[y(t) y(p)]
dp, (9.43)
where as before denotes the imaginary part.
(u 1 +Y )Y
(s) = vX
(s), (9.44)
[u +Y 1]
2
+v
2
+ 2GY = B. (9.45)
Here B the Bernoulli constant and G is the gravity parameter dened by
G =
g
c
2
. (9.46)
Equation (9.45) states that the pressure is constant on the free surface.
For given values of and we seek four functions, u, v, X
and Y
,
satisfying (9.41)(9.45). The parameters G and B are found as part of the
solution.
We seek a numerical solution of the nonlinear integro-dierential system
(9.41)(9.45). First we dene N distinct mesh points on the free surface by
specifying values of the arc length parameter t = S
I
, where
S
I
= b
I 1
2N 2
, I = 1, . . . , N. (9.47)
Here b is the length of the free surface between two successive crests. We shall
also make use of the intermediate mesh points S
I1/2
= (S
I+1
+S
I
)/2, I =
1, . . . , N 1.
We now dene the 4N corresponding fundamental unknown quantities
u
I
= u(S
I
), I = 1, . . . , N, (9.48)
v
I
= v(S
I
), I = 1, . . . , N, (9.49)
X
I
= X
(S
I
), I = 1, . . . , N, (9.50)
and
Y
I
= Y
(S
I
), I = 1, . . . , N. (9.51)
We estimate the values of the x- and y- coordinates X
I
= X(S
I
) and
Y
I
= Y (S
I
) in terms of the fundamental unknowns by the trapezoidal rule,
i.e. X
1
= 0, Y
1
= and
X
I
= X
I1
+X
(S
I3/2
)
b
2N 2
, I = 2, . . . , N, (9.52)
Y
I
= Y
I1
+Y
(S
I3/2
)
b
2N 2
, I = 2, . . . , N. (9.53)
Here X
I3/2
and Y
I3/2
are evaluated from X
I
and Y
I
by a four-point
interpolation formula. We satisfy (9.44) and (9.45) at the mesh points S
I
,
I = 1, . . . , N, and (9.41) at the mesh points S
I
, I = 1, 2, . . . , N. This yields
3N nonlinear equations.
Next we satisfy (9.43) at the point t = S
I
, I = 2, . . . , N 1, by applying
the trapezoidal rule to (9.43), with a sum over the points s = S
J1/2
, J =
1, . . . , N 1. The symmetry of the discretisation and of the trapezoidal
rule with respect to the singularity of the integrand at s = t enables us
to evaluate this Cauchy principal-value integral by ignoring the singularity,
with an accuracy no less than that of a nonsingular integral (see Section
3.1.2.2). This yields N 2 extra nonlinear equations.
Two equations are obtained by imposing the symmetry conditions
v
1
= 0 (9.54)
and
v
N
= 0. (9.55)
We obtain two more equations by xing the origin of Y at the mean water
level and by requiring that w vanishes at innite depth. This leads to
_
b/2
0
Y X
ds = 0 (9.56)
and
_
b/2
0
(uX
+vY
) ds = 0. (9.57)
The nal equation xes the wavelength by imposing
X
N
= 1/2. (9.58)
We now have 4N + 3 equations for the 4N + 3 unknowns (9.48)(9.51),
B, b and G. This system is solved by Newtons method for given values of
and . To obtain the results presented in the next section, we also used
a variation of the scheme in which we xed G and and found as part of
the solution.
We used the numerical scheme of Section 9.2.2 to compute solutions for
various values of and . Most calculations were performed with N = 121.
In all cases we checked that the numerical results were independent of N
within graphical accuracy.
We rst conrmed the ndings of Simmen and Saman [141]. There are
solution branches that bifurcate from the uniform shear ow u = v = 0.
As one progresses along the solution branches, the waves ultimately reach
limiting congurations with a trapped bubble at the trough or a 120
angle
at the crest.
We now construct the new families of solutions. Following the approach
of Section 9.1.3, we look rst at solutions that are close to the limiting
angle at the crest. Such limiting congurations

are characterized by a stagnation point at the crest. We dene waves close
to the limiting conguration with a 120
angle at the crest as solutions for

which the velocity at the crest is equal to a small quantity . Therefore they
satisfy
u
1
+ 1 = . (9.59)
Here we choose = 0.1.
In Figure 9.19, we show values of the steepness s (i.e. the dierence in
heights between a crest and a trough divided by the wavelength) versus
for waves close to the limiting conguration. We see that there is a mini-
mum value
c
such that there are no solutions for <
c
and denote the
corresponding value of s by s
c
. We now calculate families of solutions by
continuation. To do this we x a value of associated with a point on the
curve of Figure 9.19 and use the corresponding solution as the initial guess
to compute a solution for a slightly perturbed value of or G. This solution
is then used as an initial guess for a further calculation.
0
3.0
2.0
1.0
0.0
a
s
b
c
Fig. 9.19. Values of the steepness s versus for waves close to the limiting congu-
ration with a 120
angle at the crest. The waves are characterised by = 0.1. The

arrow indicates the point (
c
, s
c
). Taken from IMA J. Appl. Math. 56, 207217
(1996).
We nd that the points on the lower part of the curve in Figure 9.19, for
s < s
c
, are limiting congurations, or more precisely solutions close to the
limiting conguration, of branches that bifurcate from the uniform shear
ow u = v = 0. These branches were described by Simmen and Saman
[141] and will not be considered further here.
The points on the upper part of the curve in Figure 9.19, with s > s
c
, are
limiting congurations of new families of waves. Explicit calculations were
performed for points a and b in Figure 9.19. The corresponding solution
branches are shown in Figures 9.20 and 9.21, where we present values of s
versus G for the values = 2.202 and = 1.045 corresponding to the
points a and b of Figure 9.19.
0.74
s
0.72
0.70
0.68
0.66
0 0.10 0.20 0.30 0.40
G
Fig. 9.20. Values of s versus G for = 2.202. The highest point on the curve
corresponds to point a in Figure 9.19. Taken from IMA J. Appl. Math. 56, 207217
(1996).
1.70
1.60
1.50
0 0.010 0.020 0.030 0.040
G
s
Fig. 9.21. Values of s versus G for = 1.045. The highest point on the curve
corresponds to point b in Figure 9.19. Taken from IMA J. Appl. Math. 56, 207217,
(1996).
Figures 9.20 and 9.21 show that the new branches do not bifurcate from
the uniform shear ow (i.e. the G-axis in Figures 9.20 and 9.21). Fur-
thermore the new branches can be extended up to G = 0. However, the
numerical calculations in Figure 9.21 become more sensitive for G small and
had to be stopped at a small positive value of G.
Typical free-surface proles corresponding to points on the curves of Fig-
ures 9.20 and 9.21 are shown in Figures 9.22 and 9.23 respectively.
0.8
0.4
0.0
0 0.5 1.0 1.5
y
0.8
0.4
0.0
0 0.5 1.0 1.5
y
x
x
(a)
(b)
Fig. 9.22. Typical free-surface proles for points on the curve of Figure 9.20. The
value of is 2.202. The proles correspond to (a) G = 0.38 and (b) G = 0. Two
wavelengths are shown in both parts (a) and (b). Taken from IMA J. Appl. Math.
56, 207217 (1996).
The proles of Figure 9.22 show that as G 0 the waves approach, a
wave similar to those computed by Simmen and Saman [141]. The prole
of Figure 9.22(b) is close to a conguration with a trapped bubble at the
trough. Our calculations show that there is a wave with a trapped bubble
when G = 0 and that the corresponding value s
1
of the steepness is approxi-
mately 0.7. This value is consistent with the results in Figure 12 of Simmen
1.0
0.5
0.0
0 0.5 1.0 1.5 2.0
y
x
(a)
1.0
0.5
0.0
0 0.5 1.0 1.5 2.0
y
x
(b)
1.0
0.5
0.0
0 0.5 1.0 1.5 2.0
y
x
(c)
Fig. 9.23. Typical free-surface proles for points on the curve of Figure 9.21. The
value of is 1.045. The proles correspond to (a) G = 0.03, (b) G = 0.011 and
(c) G = 0.007. Two wavelengths are shown in each of parts (a) and (b). Taken
from IMA J. Appl. Math. 56, 207217 (1996).
and Saman [141]. Their broken curve has a vertical asymptote at about
s = s
1
(the value of the parameter
used by Simmen and Saman is equal

to when G = 0 because their solutions are scaled such that g = 1).
The proles of Figure 9.23 are more exotic. A comparison of the proles
in Figures 9.23(a)(c) show that as G 0 the waves ultimately approach,
a conguration consisting of a wave with a circular region at the crest. The
uid in the circular region is in rigid body rotation.
Our new families of waves can also be limited by trapped bubbles at their
troughs. This is illustrated in Figure 9.24, where we present the free-surface
prole corresponding to the point c in Figure 9.19. The wave has almost
developed a closed region at the crests and a trapped bubble at the troughs.
9.2.4 Discussion
We have used a boundary integral equation method to compute periodic
waves with constant vorticity are have shown that there are solution branches
that approach congurations with a closed region of uid in rigid body rota-
tion as G 0. Our results supplement those of Simmen and Saman [141],
which show that there are limiting congurations with a 120
angle at the
crest or a trapped bubble at the trough.
In Section 9.1, we showed that there are solitary waves with circular closed
regions at their crests. Since solitary waves can be viewed as the limit of
periodic waves as the ratio of the wavelength and the depth goes to innity,
our results suggest that congurations with circular regions at the crest are
a general property of waves in water of arbitrary depth.
1.0
0.5
0.0
0 0.4 0.8 1.2 1.6
y
x
Fig. 9.24. Free surface corresponding to the point c in Figure 9.19. The value of
is 0.52. Two wavelengths are shown. Taken from IMA J. Appl. Math. 56,
207217 (1996).
The results presented are only representative. We expect that there are
a large number of families similar to the family shown in Figure 9.23. In
particular we should expect waves with several closed regions of uid in rigid
body rotation at their crests (such congurations were computed for solitary
waves in Section 9.1.3.2). However, for periodic waves the admissible solu-
tions are limited by the appearance of trapped bubbles at the troughs. This
is to be contrasted with solitary waves, for which there are no congurations
with trapped bubbles at the troughs.
Finally we should point out that some of our exotic waves, such as those
in Figure 9.23(c), are likely to be unstable. However, others such as those
in Figure 9.23(a) have more classical shapes.
10
Three-dimensional free-surface ows
As shown in the previous chapters, ecient methods for two-dimensional
free-surface ows can be derived by using the theory of analytic functions. In
particular, free streamline problems, series truncation methods and bound-
ary integral equation methods based on the Cauchy integral formula can be
used to obtain highly accurate solutions. Unfortunately such techniques are
not available for three-dimensional free-surface ows. However, as we shall
see in this chapter, boundary integral equation methods can still be derived
using Greens theorem (see also [122][125]) .
Boundary integral equation methods based on Greens theorem can also
be used for two-dimensional free-surface ows as an alternative to meth-
ods based on the Cauchy integral formula. We rst show this for two-
dimensional free-surface ows generated by moving disturbances in water of
innite depth. Gravity is included in the dynamic boundary condition but
surface tension is neglected.
10.1 Greens function formulation for two-dimensional problems
We describe the numerical method based on Greens functions by consider-
ing the free-surface ows generated by a moving pressure distribution (see
Figure 4.4) or by a moving surface-piercing object (see Figure 4.3). We will
assume that the water is of innite depth. The corresponding method based
on the Cauchy theorem was described in Chapter 7 for a moving pressure
distribution.
10.1.1 Pressure distribution
We consider the two-dimensional free-surface ow generated by a pressure
distribution moving at a constant velocity U at the surface of a uid of
278
10.1 Greens function formulation for two-dimensional problems 279
innite depth. The uid is assumed to be inviscid and incompressible and
the ow is assumed to be irrotational. We choose a cartesian frame of
reference moving with the pressure distribution and assume that the ow is
steady. Gravity is acting in the negative y-direction and the eect of surface
tension is neglected. Following the formulation of Section 4.1, we introduce
the potential function (x, y), so that the velocity is given by (
x
,
y
). In
the ow domain, satises
2
= 0, < x < , < y < (x), (10.1)
with the condition
(
x
,
y
) (U, 0) as y . (10.2)
Here we denote by y = (x) the equation of the free surface. The kinematic
and dynamic boundary conditions give
x
=
y
on y = (x) (10.3)
and
1
2
(
2
x
+
2
y
) +g +
p
=
U
2
2
on y = (x), (10.4)
where g is the acceleration of gravity, is the uid density and p is the
prescribed pressure distribution. The choice of the Bernoulli constant on
the right-hand side of (10.4) xes the origin of y. The upstream radiation
condition gives
(
x
,
y
) (U, 0) and 0 as x . (10.5)
The physical quantities are made dimensionless by taking U as the unit
velocity and the length L of the support of the pressure distribution as the
unit length. The Froude number is dened by
F =
U
gL
. (10.6)
The formulation involves Greens second identity,
_
V
(
2

2
)dV =
_
C
_
n
_
ds. (10.7)
Here C is a closed curve bounding a region V of the plane. The curve C is
characterised by its arc length s and its outward normal n. Assuming that
satises Laplaces equation and that is the two-dimensional free space
280 Three-dimensional free-surface ows
Greens function g = 1/4 ln[(x x
)
2
+ (y y
)
2
], (10.7) gives
(x
, y
) = r
_
C
_
g
n
g
n
_
ds. (10.8)
Here r = 1 when (x
, y
) is inside C and r = 1/2 when (x
, y
) is on C.
We now choose = x and assume that C consists of the free surface
and a semicircle of arbitrarily large radius in the region y < (x). Using the
arc length s and describing the free surface parametrically by x = X(s) and
y = Y (s), we obtain
1
2
T(s
) =
_
T(s)
G
n
(s, s
) G(s, s
)
T(s)
n
_
ds. (10.9)
Here (s) = (X(s), Y (s)), T(s) = (s)X(s), G(s, s
) = (1/4) ln[X(s)
X(s
)]
2
+ [Y (s) Y (s
)]
2
and n = (Y
(s), X
(s)). The denition of the

arc length requires that
X
2
+Y
2
= 1. (10.10)
The kinematic and dynamic boundary conditions on the free surface are
rewritten in terms of the dimensionless variables as
n
= 0 (10.11)
and
1
2
2
s
+
Y
F
2
+P =
1
2
, (10.12)
where P is the dimensionless pressure. We choose
P(s) =
_
e
1/(s
2
1)
for [s[ < 1,
0 otherwise.
(10.13)
The unknown functions (s), X(s) and Y (s) are obtained by solving nu-
merically the nonlinear equations (10.9)(10.12) subject to the radiation
condition, which requires that there is no energy coming from innity (see
Chapter 4 for a discussion of the radiation condition).
We dene N equally spaced points s
1
= e(N 1)/2, s
i
= s
1
+e(i 1),
i = 2, . . . , N, where e is the interval of discretisation. We choose N to be
odd. Here s
1
approximates and s
N
= s
1
approximates +. We use
the notation x
i
= X(s
i
), y
i
= Y (s
i
) etc. The domain of integration for
(10.9) is (s
1
, s
N
).
In order to satisfy the Bernoulli equation at the rst point, we impose
y
1
= 0, x
1
=
1
= 1, x
1
=
1
= s
1
. (10.14)
Equations (10.10)(10.12) and the trapezoidal rule yield
x
k
=
_
1 y
2
k
,
x
k
= x
k1
+
1
2
e
_
x
k
+x
k1
_
,
y
k
= y
k1
+
1
2
e
_
y
k
+y
k1
_
,
k
=
_
1 2
y
k
F
2
2p
k
,
k
=
k1
+
1
2
e(
k
+
k1
),
(10.15)
for k = 2, . . . , N.
The values of the functions at the midpoints are calculated by interpola-
tion with two or four points (x
k1/2
=
1
2
(x
k1
+x
k
) etc.). Equation (10.9) is
evaluated at the midpoints s
i1/2
, i = 2, . . . , N 1. The integral is approx-
imated by the trapezoidal rule with a summation over the mesh points s
i
,
i = 2, . . . , N. Substituting (10.15) yields N2 nonlinear algebraic equations.
The nal two equations are obtained by imposing the radiation condition,
using the relations
y
1
= 0 and 3y
1
+ 4y
2
y
3
= 0. (10.16)
The second of these relations imposes y
1
= 0, approximately. This system of
N nonlinear equations for the N unknowns y
1
, . . . , y
N
is solved by Newtons
method.
The initial guess for the unknowns y
i
is zero when 1 or previous
computed solutions obtained for slightly dierent values of F and when
is large.
The numerical accuracy of the scheme was checked by varying N and e
(see Figure 10.1).
We found that the solutions presented here are independent of N and
e within graphical accuracy for N 200 and e 0.1. In the numerical
calculations, the integral fromto in (10.9) was replaced by an integral
from s
1
to s
N
. We found that these upstream and downstream truncations
only aect the rst and last half-wavelength of the free-surface proles. A
similar numerical behavior was found in Asavanant and Vanden-Broeck [7].
Improved truncation methods similar to those used in Section 7.1.2 can be
developed.
We compared our numerical solutions with those obtained using the
method of Chapter 7. A typical pair of proles is shown in Figure 10.2.
Similar results were found for other values of and F. The conclusion of
8 4 0 4 8
8
4
0
4
8
x 10
Fig. 10.1. Computed free-surface proles for F = 0.7, = 0.001 with grids N = 721,
e = 0.025 (broken line) and N = 361, e = 0.05 (solid line).
the comparison is that numerical results as accurate as those of Chapter 7
can be obtained without using complex variables. This suggests that ac-
curate results for three-dimensional free-surface ows can be obtained by
generalising the Greens function formulation of Section 10.1.1 to three di-
mensions. This is done in Section 10.2.
8 4 0 4 8
8
4
0
4
8
x 10
Fig. 10.2. Computed free-surface proles obtained with an algorithm based on
Greens functions (solid line) and with the algorithm of Chapter 7 (dotted line).
The parameters are F = 0.7, = 0.001. The grid used is N = 721 and e = 0.025.
10.1.2 Two-dimensional surface-piercing object
As noted in Chapter 7, once a solution of (10.9)(10.12) has been computed
for a given pressure distribution (10.13), we can replace the free surface
under the support 1 < s < 1 of the pressure distribution by a rigid sur-
face. Here, the support refers to the interval over which the pressure is
nonzero.
Therefore the schemes described in Section 10.1.1 and in Chapter 7 pro-
vide an inverse method for calculating free-surface ows past surface-piercing
objects, e.g. two-dimensional ships (see Figure 4.3). The shape of the ship
is given at the end of the calculations by the shape of the streamline under
the support of the pressure distribution. One drawback of this approach is
that the shape of the ship depends on the Froude number F.
It is therefore desirable to have an approach that enables a direct calcu-
lation of the free-surface ow past a given surface-piercing object. This was
achieved by Asavanant and Vanden-Broeck [7] using complex variables.
In the present section we will explore a corresponding approach using the
Greens function formulation. As before we will assume that the water is of
innite depth.
We shall present results for a parabolic object dened by the equation
y =

2
(x
2
1). (10.17)
In general we might expect a spray or splash at the front of the object (see
for example Dias and Vanden-Broeck [47]). Here, we restrict our attention
to ows that separate smoothly from the object.
Let us denote by s
a
and s
b
the values of s at the left- and right-hand
separation points. Since we need to nd s
a
and s
b
as part of the solution,
we introduce a new variable t by
s = s
a
+ (s
b
s
a
)t. (10.18)
The relation (10.18) maps the unknown interval (s
a
, s
b
) into the xed
interval (0, 1). The new unknown functions are

(t) = (s),

X(t) = X(s),
Y (t) = Y (s), where s is dened by (10.18).

The system of nonlinear equations to be solved is now obtained by
substituting (10.18) into (10.9), (10.10) and (10.12). This yields the
integro-dierential equation
2[
(t
)

X(t
)]
=
_
2[
(t)

X(t)]
[

X(t)

X(t
)][
(t)] + [
Y (t)

Y (t
)]

X
(t)
[

X(t)

X(t
)]
2
+ [
Y (t)

Y (t
)]
2
ln
_
[

X(t)

X(t
)]
2
+ [
Y (t)

Y (t
)]
2
_

Y
(t)
_
dt, < t
< ,
(10.19)
the Bernoulli equation
1
2
_

t
s
b
s
a
_
2
+

Y
F
2
=
1
2
, for t < 0 or t > 1 (10.20)
and the arc length equation
X
2
+

Y
2
= (s
b
s
a
)
2
. (10.21)
In addition the kinematic condition boundary condition on the object gives
Y =

2
(

X
2
1) for 0 < t < 1. (10.22)
At the separation points t = 0 and t = 1 we must satisfy both (10.20) and
(10.22), so we have
1
2
_

t
s
b
s
a
_
2
+
(

X
2
1)
2F
2
=
1
2
at t = 0 or t = 1. (10.23)
For the numerical computation, again we introduce N equally spaced
points t
1
= e(N1)/2, t
i
= t
1
+e(i 1), i = 2, . . . N, and use the notation
x
i
=

X(t
i
), y
i
=

Y (t
i
),
i
=

(t
i
), x
i
=

X
(t
i
), y
i
=

Y
(t
i
) and
i
=

t
(t
i
).
The values of

t
at the surface of the object cannot be determined, as
in (10.15), by using the Bernoulli equation. At the surface of the object,
between t = 0 and t = 1 there are M = 1/e 1 mesh points (we will choose
e such that 1/e is integer, but this is not a necessary condition). At each
mesh point there are two unknowns,
i
and y
i
. So we have N + M + 2
unknowns: y
1
, . . . , y
N
,
(N+1)/2+1
, . . . ,
(N+1)/2+M
and s
a
, s
b
.
The integral equation is evaluated at the midpoints t
i1/2
, i = 2, . . . , N1,
as before, so we obtain N 2 equations. Another M equations are given by
y
(N+1)/2+j
=

2
_
x
2
(N+1)/2+j
1
_
, j = 1, . . . , M. (10.24)
The equations at the separation points (10.23) give another two equa-
tions and the radiation condition (10.16) provides the nal two equations.
The values of

at the separation points,
(N+1)/2
and
(N+1)/2+M+1
, are
obtained using an extrapolation formula with four points (taken from the
object).
The usual initial guess is y
i
= 0, i = 1, . . . , N, s
a
= 1, s
b
= 1,
(N+1)/2+j
= s
b
s
a
, j = 1, . . . , M.
At the rst point we impose
y
1
= 0, x
1
=
1
= s
b
s
a
, x
1
=
1
= s
a
+ (s
b
s
a
)t
1
, (10.25)
and the remaining functions are calculated as before, using equations (10.20),
(10.21) and the trapezoidal rule. Again, the values of the functions at the
midpoints are calculated by interpolation with two points.
The numerical scheme described above was used to calculate solutions
for various values of F and . The accuracy of the results was checked by
varying N and e. We present typical free surfaces for > 0 and < 0 in
Figures 10.3 and 10.4.
40 20 0 20 40
0.008
0.004
0
0.004
0.008
Fig. 10.3. Computed free-surface proles with N = 421, e = 0.1 for F = 1.5,
= 0.004. The parabolic object and the separation points (the two crosses) are
also shown.
We observe that if F is held constant and the value of is varied then
the wavelength of the waves downstream does not change much, but their
amplitude is aected.
Our calculations cannot be directly compared with those of Asavanant
and Vanden-Broeck [7] because their study was for nite depth. Also, they
chose the position of the separation points and obtained the position of the
vertex of the obstacle as part of the solution. In our case the position of
40 20 0 20 40
0.015
0.010
0.005
0
0.005
0.010
0.015
Fig. 10.4. Computed free-surface proles for F = 1.5, = 0.006 (dotted line),
= 0.004 (broken line) and = 0.001 (solid line).
the vertex is known and we calculate the position of the separation points
as part of the solution.
10.2 Extension to three-dimensional free-surface ows
The results of Sections 10.1.1 and 10.1.2 show that two-dimensional free-
surface ows can be computed accurately by using the Greens function
formulation. In this section we extend this approach to three-dimensional
ows. Our method follows Forbes [58] and Parau and Vanden-Broeck [122].
We rst present explicit results for pure gravity ows generated by moving
pressure distributions or submerged disturbances in water of innite depth.
Extensions to uids with surface tension are discussed in Section 10.2.2.
10.2.1 Gravity ows generated by moving disturbances
in water of innite depth
10.2.1.1 Formulation
We consider a three-dimensional pressure distribution moving at constant
velocity U at the surface of a uid of innite depth. The ow is shown in
Figure. 10.5.
As in Section 10.1.1, we choose a frame of reference moving with the
pressure distribution and assume that the ow is steady. We introduce
cartesian coordinates x, y, z with the z-axis directed vertically upwards
and the x-axis in the opposite direction to the velocity U. We denote by
x
z
y
U
P
Fig. 10.5. The ow in the three-dimensional case. The pressure distribution is
moving to the left at a constant velocity U.
z = (x, y) the equation of the free surface. The potential function (x, y, z)
satises the Laplace equation,
2
= 0, < x < , < y < , < z < (x, y),
(10.26)
in the ow domain. The kinematic boundary condition (10.3), the dynamic
boundary condition (10.4) and the radiation condition (10.5) are now
x
+
y
y
=
z
on z = (x, y), (10.27)
1
2
(
2
x
+
2
y
+
2
z
) +g +
p
=
U
2
2
on z = (x, y), (10.28)
no waves as x . (10.29)
Equation (10.7) holds in three dimensions, where V now represents a
volume bounded by the surface C. Proceeding as in Section 10.1.1 and
using the three-dimensional free surface Greens function
G =
1
4
1
[(x x
)
2
+ (y y
)
2
+ (z z
)
2
]
1/2
, (10.30)
we obtain
1
2
[(x
, y
) Ux
]
=
__
R
2
[(x, y) Ux]
1
4
(x, y) (x
, y
) (x x
)
x
(y y
)
y
(x x
)
2
+ (y y
)
2
+ [(x, y) (x
, y
)]
2
3/2
dxdy
+
__
R
2
1
4
U
x
(x x
)
2
+ (y y
)
2
+ [(x, y) (x
, y
)]
2
1/2
dxdy,
(10.31)
where (x, y) = (x, y, (x, y)).
The pressure is chosen as
p(x, y) =
_
_
_
P
0
exp
_
L
2
(x
2
L
2
)
+
L
2
(y
2
L
2
)
_
, [x[ < L and [y[ < L,
0 otherwise.
We introduce dimensionless variables by using U as the unit velocity and
L as the unit length. Combining equations (10.27) and (10.28) and using
the chain rule we obtain
1
2
(1 +
2
x
)
2
y
+ (1 +
2
y
)
2
x
2
x
y
1 +
2
x
+
2
y
+

F
2
+P =
1
2
, (10.32)
where F = U/(gL)
1/2
and = P
0
/(U
2
). Now P(x, y) is e
1
x
2
1
+
1
y
2
1
for
[x[ < 1 and [y[ < 1, and 0 otherwise.
Equation (10.31) can be rewritten as
2[(x
, y
) x
] = I
1
+I
2
, (10.33)
where
I
1
=
_
0
[(x, y) (x
, y
) x +x
]K
1
dxdy, (10.34)
I
2
=
_
0
x
(x, y)K
2
dxdy, (10.35)
K
1
=
_
(x, y) (x
, y
) (x x
)
x
(y y
)
y
(x x
)
2
+ (y y
)
2
+ [(x, y) (x
, y
)]
2
3/2
+
(x, y) (x
, y
) (x x
)
x
(y +y
)
y
(x x
)
2
+ (y +y
)
2
+ [(x, y) (x
, y
)]
2
3/2
_
(10.36)
K
2
=
_
1
_
(x x
)
2
+ (y y
)
2
+ [(x, y) (x
, y
)]
2
+
1
_
(x x
)
2
+ (y +y
)
2
+ [(x, y) (x
, y
)]
2
_
. (10.37)
In deriving (10.33) we used the fact that the solutions are symmetric in
the y-direction. We note that the integral I
2
is singular, whereas I
1
is not.
10.2.1.2 The numerical scheme
We truncate the intervals < x < and 0 < y < to x
1
< x < x
N
and
y
1
< y < y
M
and introduce the mesh points x
i
= x
1
+(i1)x, i = 1, . . . , N
and y
j
= (j 1)y, j = 1, . . . , M. Following Forbes [58] and Parau and
Vanden-Broeck [122], the integral I
2
is written in the form I
2
= I
2
+ I
2
,
where
I
2
=
y
M
_
y
1
x
N
_
x
1
[
x
(x, y)K
2
x
(x
, y
)S
2
]dxdy,
I
2
=
x
(x
, y
)
y
M
_
y
1
x
N
_
x
1
S
2
dxdy,
with
S
2
=
1
_
A(x x
)
2
+B(x x
)(y y
) +C(y y
)
2
+
1
_
A(x x
)
2
B(x x
)(y +y
) +C(y +y
)
2
and
A = 1 +
2
x
(x
, y
), B = 2
x
(x
, y
)
y
(x
, y
), C = 1 +
2
y
(x
, y
).
The integral I
2
, which contains the singularity, can be calculated using
_ _
dsdt
As
2
+Bst +Ct
2
=
t
A
ln[2As +Bt + 2
_
A(As
2
+Bst +Ct
2
)]
+
s
C
ln[2Ct +Bs + 2
_
C(As
2
+Bst +Ct
2
)].
The 2NM unknowns are
u = (
x
11

x
12

x
N, M 1

x
N M

x
11

x
N M
)
T
.
The integrals and the Bernoulli equation are evaluated at the points
(x
i+1/2
, y
j
), i = 1, . . . , N 1, j = 1, . . . , M, so we have 2(N 1)M equa-
tions. Another 2M equations are obtained from the radiation condition
x
1j
= 0,
x
1j
= 1, j = 1, . . . , M. The values of and are obtained
by integrating
x
and
x
with respect to x by the trapezoidal rule. The in-
tegration is started by using the values derived from the radiation condition
(10.29) and the free-surface condition (10.32) satised at the rst row:
1j
= 0,
y
1j
= 0,
1j
= x
1
,
y
1j
= 0, j = 1, . . . , M.
The values of
y
and
y
are then calculated by central dierences. The
values of the variables and at (x
i+1/2
, y
j
) are obtained by interpolation.
The 2NM nonlinear equations are solved by Newtons method. In most
calculations we chose
x
ij
= 0,
x
ij
= 1 for i = 1, . . . , N, j = 1, . . . , M, as
the initial guess.
10.2.1.3 Numerical results
We used the scheme described in Section 10.2.1.2 to calculate solutions for
dierent values of the Froude number F and of the parameter ; the results
are qualitatively similar. We present a typical free-surface prole for F = 0.7
and = 1 (see Figure 10.6).
0
2
4
6
8
10
12
5
4
3
2
1
0
1
2
3
4
5
0
0.2
0.2
x
y
z
Fig. 10.6. The solution for the wave eld due to a moving pressure distribution for
F = 0.7 and = 1. The grid parameters are N = 75, M = 25, x = 0.2, y = 0.2.
The transverse waves are perpendicular to the direction of the velocity U, which is
parallel to the x-axis. In this graph and the following three-dimensional gures the
darker parts correspond to the troughs and the brighter parts to the peaks of the
waves.
The wake and the two dierent families of waves (transverse waves and
short-length divergent waves) can easily be observed. When F increases, the
amplitude of the divergent waves becomes greater than that of the transverse
waves (see Figure 10.7).
The wavelength of the transverse waves increases with the Froude number
(see Figure 10.8).
The inuence of the truncation upstream and downstream is seen to be
negligible (see Figure 10.9). Here we show the centreline (i.e. the intersection
0
5
10
15
20
25
30
35
15
10
5
0
5
10
15
0
0.5
0.5
x
y
z
Fig. 10.7. These waves were generated for a higher Froude number (F = 1.2) than
in Figure 10.6. The grid used was N = 61, M = 19, x = y = 0.6. The divergent
waves can be observed more easily than in Figure 10.6; their amplitudes are larger
than those of the transverse waves.
2 0 2 4 6 8 10 12
5
4
3
2
1
0
1
2
3
4
5
x
y
Fig. 10.8. The wake in the cases F = 0.7 (lower half) and F = 0.5 (upper half). In
both cases = 1.
of the free surface with the plane y = 0) and two curves corresponding to
dierent truncations, x = (3, 12) and x = (6, 6).
The accuracy of the solutions was tested by varying the number of grid
points and the intervals x and y between grid points (see an example in
0 2 4 6 8 10 12
0
0.05
0.10
0.15
0.20
x
z
Fig. 10.9. The free-surface elevation at the plane y = 0 for two dierent truncations,
x = (3, 12) (the broken curve) and x = (6, 6) (the solid curve), is shown. In
both cases F = 1 and = 1.
2
0
2
4
6
8
10
12
14
16
5
4
3
2
1
0
1
2
3
4
5
2
0
2
x
y
z
Fig. 10.10. The accuracy check: F = 0.7, = 0.0001; N = 89, M = 13; x =
y = 0.2 (lower half); N = 61, M = 17, x = y = 0.3 (upper half).
Figure 10.10). The upper part of Figure 10.10, y > 0, was calculated for
N = 61, M = 17, x = y = 0.3 and the lower part, y < 0, for M = 89,
M = 13, x = y = 0.2. The values of the parameters are the same in
both cases: F = 0.7, = 1.
The algorithm can be easily modied to include two or more pressure
distributions and to study the interaction of the wakes produced by each of
them. We present an example in Figure 10.11 for two pressure disturbances
moving in parallel.
2
1
0
1
2
3
4
5
6
0
1
2
3
4
5
0.05
0
0.05
x
y
z
Fig. 10.11. The case of two pressure distributions moving in parallel (F = 0.4).
The V-shapes of the waves become a W-shape downstream. This case
can be viewed as representing the wave interactions between ships moving
in parallel in deep water.
There are various possible generalisations of our code. For example, it
can be used to consider moving submerged objects. An inverse method to
compute the resulting waves is by superposing singularities. An example of
the waves generated by a source and a sink is given in Figure 10.12.
10.2.2 Three-dimensional gravitycapillary free-surface
ows in water of innite depth
In this section we consider steady three-dimensional fully nonlinear gravity
capillary free-surface ows. We show how to extend the results of Section
7.2.2 to three dimensions and start by considering free-surface ows gener-
ated by a moving pressure distribution. We nd that some solutions are
perturbations of a uniform stream while others are perturbations of solitary
waves. The solitary waves have decaying oscillations in the direction of prop-
agation and monotone decay in the direction perpendicular to the direction
of propagation. They travel at a velocity U smaller than the minimum ve-
locity c
min
of linear gravitycapillary waves. It is shown that the structure
4 2 0 2 4 6 8 10 12 14 16
4
2
0
2
4
0.2
0
0.2
0.4
0.6
x
z
y
Fig. 10.12. The waves generated by a sourcesink pair (F = 0.7). The source is at
(0, 0, 1) and the sink at (1, 0, 1).
of the solutions in three dimensions is similar to that found in Section 7.2.2
for two-dimensional waves (see Figure 7.8).
10.2.2.1 Formulation
We consider a three-dimensional solitary wave travelling at a constant ve-
locity U at the upper surface of a uid of innite depth. The uid is in-
compressible and the ow is irrotational. We choose a frame of reference
moving with the wave and assume that the ow is steady. We introduce
cartesian coordinates x, y, z with the z-axis directed vertically upwards
(opposite to the direction of gravity) and the x-axis in the direction of wave
propagation. As in Section 10.2.1.1 we denote by z = (x, y) the equation
of the free surface and dene the potential function (x, y, z). We introduce
dimensionless variables by using U as the unit of velocity and T/(U
2
) as
the unit of length. Here T is the surface tension. The velocity potential
function (x, y, z) satises Laplaces equation
2
= 0, < x < , < y < , < z < (x, y),
(10.38)
in the ow domain.
The kinematic and dynamic boundary conditions can be written as
x
+
y
y
=
z
on z = (x, y), (10.39)
1
2
(
2
x
+
2
y
+
2
z
) +
_
_

x
_
1 +
2
x
+
2
y
_
_
x
_
_

y
_
1 +
2
x
+
2
y
_
_
y
=
1
2
on z = (x, y); (10.40)
we have used the conditions
(
x
,
y
,
z
) (1, 0, 0) and 0 as (x
2
+y
2
)
1/2
(10.41)
to x the value of Bernoullis constant in (10.40). The parameter in (10.40)
is dened by
=
gT
U
4
.
We solve the problem numerically by a boundary integral equation method
similar to that used in Section 10.2.1.2. Details can be found in [123] and
[124]. One dierent feature for the numerical scheme is that the surface ten-
sion term introduces higher derivatives in (10.40). They are approximated
by centred-dierence formulae. Another new feature is that no radiation
condition is needed. Instead the solutions are assumed to be symmetric
about the x- and y-axes. The discretisation involves a regular grid with
N points in the x-direction and M points in the y-direction. The uniform
mesh sizes on the x- and y-axes are denoted by x and y. The algebraic
equations obtained after discretization are solved by Newtons method. A
suitable initial guess to compute solitary waves is obtained by adapting the
method of Section 7.2.2 from two dimensions to three dimensions. This
consists of rst obtaining solutions for the problem with an extra pressure
term in equation (10.40) and then taking the limit as the magnitude of the
pressure tends to zero.
10.2.2.2 Results
We computed solutions for various values of by using the numerical scheme
described in the previous section. In all cases is assumed to be greater than
1/4, which corresponds to waves moving steadily with a constant velocity
U smaller than the minimum phase speed c
min
dened by (2.98). In this
case only a highly localised disturbance of the water surface is predicted. In
two dimensions, capillarygravity waves were calculated in the same regime
of parameters ( > 1/4). Most computations were performed with x =
y = 0.8 and N = 40, M = 50. The accuracy of the solutions was tested by
varying the number of grid points and the intervals x and y between grid
points. To indicate the numerical consistency, we found that, for = 0.35,
if we changed x and y from 0.8 to 0.6 the corresponding change in the
30
20
10
0
10
20
30
40
20
0
20
40
1.5
1.0
0.5
0
0.5
z
y
x
Fig. 10.13. Solitary gravitycapillary wave with central depression for = 0.35.
The surface elevation is vertically exaggerated by a factor 20.
wave surface elevation was less than 5% everywhere. The smallness of x
and y is limited by memory capacity in order to obtain a truncated domain
large enough to encompass nearly all the wave disturbance.
We present typical free-surface proles in Figures 10.1310.15 and 10.17.
As expected, there is a localised disturbance of the water surface. For each
there is a central-depression wave (for such a wave (0, 0) < 0; see Fig-
ure 10.13 for the full solution and Figure 10.14 for half the solution) and
a central-elevation wave (for this wave (0, 0) > 0; see Figure 10.15). The
waves have decaying oscillations in the direction of propagation and mono-
tonic decay in the direction perpendicular to the direction of propagation.
Figure 10.16 shows the zx and zy cross-sections and a close-up with the
same horizontal and vertical scale for = 0.35. As decreases and ap-
proaches 1/4, more and more oscillations appear in front and behind the
main disturbance.
The amplitude of the free capillarygravity waves decreases to zero as
decreases to 1/4. Their form (see Figure 10.17) suggests that they approach
a train of two-dimensional (constant in the y-direction) periodic waves in
the limit as decreases to 1/4. When is increased, the solitary capillary
gravity wave elevation decreases quickly in every direction and the surface
has the form of a central-depression or central-elevation three-dimensional
fully localised solitary wave.
The solitary-wave branches are shown in Figure 10.18. These branches
bifurcate from the uniform stream at = 1/4.
30
20
10
0
10
20
30
0
10
20
30
40
1.5
1.0
0.5
0
0.5
x
y
z
Fig. 10.14. Central-depression solitary gravitycapillary wave for = 0.35, showing
half the solution. The surface elevation is vertically exaggerated by a factor 20.
30
20
10
0
10
20
30
0
10
20
30
40
1.5
1.0
0.5
0
0.5
x
y
z
Fig. 10.15. Central-elevation solitary gravitycapillary wave for = 0.35. Only
half the solution is shown. The vertical exaggeration is by a factor 20.
For the two-dimensional problem, the solutions are described in Section
7.2.2.
The numerical results presented in this section are consistent with the
asymptotic ndings of Kim and Akylas [87], [88] and of Milewski [115] and
with the rigorous results of Groves and Sun [68]. The graphs presented in
this section are based on [123]. We have recently repeated the calculations
with a ner grid and found results which are qualitatively similar: there are
still two branches bifurcating from = 0.25 in Figure 10.18. However, the
shape of the curves is slightly dierent. These rened computations will be
reported elsewhere.
30 20 10 0 10 20 30
1.5
1.0
0.5
0
0.5
(a)
z
8 4 0 4 8
8
4
0
4
8
z
(b)
Fig. 10.16. (a) The centreline in the Ox direction (solid line) and in the Oy direction
(broken line) for a central depression wave and = 0.35. The vertical exaggeration
is by a factor 20. (b) A close-up with the same scaling in the horizontal and vertical
directions.
10.3 Further extensions
An extension of the numerical method of Section 10.2.1.2 to nite depth
was developed in [124]. The basic idea is to use the formulation of Section
10.2.1.2 but with the free space Greens function replaced by a Greens
function that satises the kinematic boundary condition on the bottom.
Three-dimensional gravitycapillary solitary waves similar to those in water
of innite depth were obtained.
10.3 Further extensions 299
30
20
10
0
10
20
30
0
20
40
1.5
1.0
0.5
0
0.5
Fig. 10.17. Solitary capillarygravity waves for = 0.266. Only half the solution
(y 0) is shown. The vertical exaggeration is by a factor 20.
0.24 0.26 0.28 0.30 0.32 0.34
1.2
1.0
0.8
0.6
0.4
0.2
0
0.2
0.4
a
z(0,0)
Fig. 10.18. Values of the amplitude (0, 0) versus .
The three-dimensional solitary waves of Section 10.2.2 were calculated
by considering free-surface ows generated by a moving pressure distribu-
tion when > 0.25. In [125], the properties of three-dimensional gravity
capillary free-surface ows generated by a moving pressure distribution when
< 0.25 were studied. There are then two wave trains on the free surface.
The radiation condition forces waves of longer wavelength to occur at the
back of the disturbance and waves of shorter wavelength to occur at the
front of the disturbance. This radiation condition cannot be as easily im-
posed as in the pure gravity case, where we simply required the free surface
to be at at some distance in front of the disturbance (see Section 10.2.1).
It was shown in [125] that an ecient way to impose the radiation condition
when < 0.25 is to introduce some dissipation in the dynamic boundary
condition in the form of a Rayleigh viscosity. The Rayleigh viscosity was
introduced in Chapter 4 as a way of satisfying the radiation condition for
linearised problems. The results in [125] showed that it can also be used to
compute nonlinear solutions.
Another extension is to consider three-dimensional waves propagating at
the interface of two uids of constant density (see [126]).
Finally, let us mention that the Greens function formulation employed in
this chapter can be used to compute axisymmetric free-surface ows. The
idea is to use cylindrical coordinates r, , z with the z-axis along the axis
of symmetry and to integrate the Greens function over (see [113], [114],
[60], [187] and [73] for examples of such computations).
11
Time-dependent free-surface ows
11.1 Introduction
The rst ten chapters of this book were devoted to steady free-surface ows.
An equally important topic is that of time-dependent free-surface ows.
Boundary integral equation methods can still be used to investigate these
problems. The idea is to march in time and to solve at each time step a
linear integral equation similar to those derived in the previous chapters, by
using Cauchy integral equation formula or Greens theorem. Such methods
have been developed and used by many authors (see for example [104], [9],
[35], [136] and the references cited in these papers). In particular, results
have been obtained for breaking waves. An obvious use of time-dependent
codes is to study the stability of steady solutions.
In this chapter we will conne our attention to one type of time-dependent
free-surface ow, namely gravitycapillary standing waves. We will solve the
problem by a series expansion similar to that used in Section 5.1 to study
periodic travelling waves. The analysis follows Vanden-Broeck [167] closely.
The choice of this problem is motivated by the fact that gravitycapillary
standing waves have properties similar to those of Wilton ripples (see Section
6.5.3.1).
We note that a proof of the existence of nonlinear gravity standing waves
was provided only recently [81].
11.2 Nonlinear gravitycapillary standing waves
The concept of linear standing waves was introduced in Section 2.4.3. Here
we extend the theory of standing waves to the nonlinear regime. More pre-
cisely we consider the time-periodic two-dimensional ow of a uid bounded
below by a horizontal bottom and above by a free surface. We assume the
301
302 Time-dependent free-surface ows
motion to be periodic in the horizontal direction with wavelength and
measure lengths in units of k
1
= /(2).
Following Concus [33], we dene the parameters and by the relations
=
Tk
2
g
, (11.1)
=

1 +
. (11.2)
Here T is the surface tension, is the density and g is the acceleration of
gravity. For 1, the capillary eects are small, whereas for [1 [ 1
they predominate.
We dene cartesian coordinates such that the motion is symmetric about
the vertical axis, x = 0, and such that y = 0 corresponds to the mean level.
Let k
1
h denote the mean depth, [kg(1 + )]
1/2
the angular frequency,
[kg(1 +)]
1/2
1
t the time and a the amplitude of the linearised surface-
wave motion. Then we dene = ak and let k
1
(x, t) be the elevation of
the free surface above the mean level and [g(1 + )]
1/2
k
3/2
the velocity
potential.
In dimensionless variables, the nonlinear problem is described by the equa-
tions
xx
+
yy
= 0 in 0 < x < , h < y < (x, t), (11.3)
(1 )

xx
(1 +
2
2
x
)
3/2
+
t
+

2
(
2
x
+
2
y
) = 0 on y = (x, t), (11.4)
y
=
t
+
x
x
on y = (x, t), (11.5)
n
= 0 on x = 0 and x = , y = h, (11.6)
x
= 0 on x = 0 and x = , (11.7)
_

0
(x, t)dx = 0, (11.8)
(x, y, t + 2) = (x, y, t), (11.9)
_
0
h
_

0
_
2
0
(x, y, t) sin t cos xdtdxdy = 0, (11.10)
_
0
h
_

0
_
2
0
(x, y, t) cos t cos xdtdxdy =
1
2
2
(tanh h)
1/2
. (11.11)
As noted by Tadjbakhsh and Keller [148] and Concus [33], a unique solution
does not exist for values of h for which the frequency of the nth spatial
harmonic, n[1 + (n
2
1)] tanh nh
1/2
, is an integral multiple of the fun-
damental frequency (tanh h)
1/2
. This yields the uniqueness condition
n[1 +(n
2
1)] tanh nh
tanh h
,= j
2
, n = 2, 3, . . . , j = 1, 2, . . . (11.12)
Concus [34] showed that the values of h for which the uniqueness condi-
tion (11.12) is not satised form a denumerably innite set that is densely
distributed over the entire positive real line.
Following Tadjbakhsh and Keller [148], Concus [33] sought a solution as
an expansion in powers of . Thus
=
0
(x, t) +
2
1
(x, t) +
1
2
2
(x, t) +O(
4
), (11.13)
=
0
(x, y, t) +
2
1
(x, y, t) +
1
2
2
(x, y, t) +O(
4
), (11.14)
=
0
+
1
+
1
2
2
+O(
3
). (11.15)
Substituting (11.13)(11.15) into (11.3)(11.11) and equating powers of
gives a succession of linear systems of equations, whose solutions are:
0
= sin t cos x, (11.16)
0
=

0
sinh h
cos t cos xcosh(y +h), (11.17)
2
0
= tanh h; (11.18)
1
=
1
8
_
2
0
+
2
0
1 + 3
+

2
0
3
6
0
1 3
4
0
cos 2t
_
cos 2x, (11.19)
1
=
0
+
1
8
(
0
3
0
)t
1
16
(3
0
+
3
0
) sin 2t
3[
0
2
3
0
(1 + 2)
7
0
]
16(1 3
4
0
) cosh 2h
sin 2t cos 2xcosh(2y +h),
(11.20)
1
= 0. (11.21)
Higher-order terms can be found in Concus [33] for ,= 0 and in [148] for
= 0.
We note that
1
is unbounded when
1 3
4
0
= 0. (11.22)
This critical value corresponds to n = j = 2 in (11.12).
We now derive a perturbation solution valid when (11.22) holds. We
substitute the expansions (11.13), (11.14) into the system (11.3)(11.11)
and collect all terms of like power in . The terms of order in (11.4) and
(11.5) give
(1 )
0
0
xx
+
0
0
t
= 0, on y = 0, (11.23)
0
y

0
0
t
= 0 on y = 0. (11.24)
Equations (11.3) and (11.6)(11.11) remain unchanged in form as equations
for
0
and
0
.
The terms of order
2
in (11.4), (11.5) and (11.11) give
(1 )
1
1
xx
+
0
1
t
= F
0
on y = 0, (11.25)
1
y

0
1
t
= G
0
on y = 0, (11.26)
_
0
h
_

0
_
2
0
1
cos t cos xdtdxdy = 0. (11.27)
Here F
0
and G
0
are dened by
F
0
=
1
2
[(
0
x
)
2
+ (
0
y
)
2
]
0
0
1y

1
0
t
, (11.28)
G
0
=
0
x
0
x
0
yy
+
1
0
t
. (11.29)
Equations (11.3) and (11.6)(11.10) remain of the same form as the equa-
tions in
1
,
1
and
1
.
The solution of the zeroth-order problem dened by (11.3), (11.23), (11.24),
(11.6)(11.11) and (11.22) is
0
= sin t cos x +Acos 2t cos 2x, (11.30)
0
=

0
sinh h
cos t cos xcosh(y +h)
A
0
sinh 2h
sin 2t cos 2xcosh 2(y +h),
(11.31)
2
0
= tanh h. (11.32)
Here A is an arbitrary constant. Thus the solution of the zeroth-order
problem is not unique when (11.22) is satised.
Dierentiating (11.25) with respect to t, substituting
1
t
from (11.26) and
1
xxt
from (11.26) after dierentiating twice with respect to x, we obtain
1
yxx
+ (1 )
1
y
+
2
0
1
t
= H
0
on y = 0. (11.33)
Here H
0
is dened by
H
0
=
0
F
0
t
+ (1 )G
0
G
0
xx
. (11.34)
Separation of variables yields, for the solution of (11.3) subject to (11.6),
1
(x, y, t) =
m=0
A
m
(t) cos mxcosh m(y +h). (11.35)
Substituting (11.35) into (11.33) we obtain
2
0
cosh mhA
m
(t) + [(1 )m+m
3
] sinh mhA
m
(t)
=
1
_
2
0
H
0
cos mxdx. (11.36)
Here = 1 for m > 0 and = 2 for m = 0. Using (11.28)(11.32), we can
rewrite (11.36) in the form
2
0
A
m
(t) =
1
4
(3
2
0
+
1
0
) sin 2t 2A
2
(
2
0
coth
2
2h + 3
2
0
) sin 4t, (11.37)
2
0
cosh hA
1
(t) + sinh hA
1
=[2
1
+
1
4
A(
1
0
3
3
0
+ 4
0
coth 2h)] cos t
+
1
4
A(4
0
coth 2h +
1
0
+ 21
3
0
) cos 3t,
(11.38)
2
0
cosh 2hA
2
(t) + 2(1 + 3) sinh 2hA
2
(t)
=
3
4
[
2
0
(1 + 2)
1
0
] 2A
1
6A
1
4A
1
2
0
coth 2h sin 2t,
(11.39)
2
0
cosh 3hA
3
(t) + (3 + 5) sinh 3hA
3
(t)
=
1
4
A
0
cos t[(4 + 48) coth 2h (3 + 24)
2
0
3
2
0
],
1
4
A
0
cos 3t[(12 + 48) coth 2h + (3 + 24)
2
0
21
2
0
], (11.40)
2
0
cosh 4hA
4
(t) + (4 + 60) sinh 4hA
4
(t)
= A
2
0
sin 4t[(2 + 30) coth 2h + 2
0
coth
2
2h 6
2
0
], (11.41)
2
0
cosh mhA
m
(t) + [(1 )m+m
3
] sinh mhA
m
(t) = 0, m = 5, 6, . . .
(11.42)
From (11.9) and (11.35) it follows that A
m
(t) must be periodic in t with
period 2 for m 1 and from (11.22) and (11.42) that A
m
= 0 for m 5.
The periodicity of A
1
requires the coecient of cos t in (11.38) to be equal
to zero. Thus
1
=
1
6
A(3
3
0

1
0
4
0
coth 2h). (11.43)
If we set A = 0 in (11.37)(11.43), the solution of (11.39) is
A
2
=
3[
0
2
3
0
(1 + 2)
7
0
]
16(1 3
4
0
) cosh 2h
sin 2t, (11.44)
in agreement with (11.20). As mentioned earlier this solution is unbounded
when (11.22) is satised.
We shall determine the constant A in such a way that the solution of
(11.39) is bounded. The appropriate compatibility condition is obtained by
multiplying (11.39) by sin 2t, integrating with respect to t from 0 to 2,
applying integration by parts twice to the term containing A
2
(t) and using
(11.22). Thus we nd that the coecient of sin 2t on the right-hand side of
(11.39) must be equal to zero. This yields the relation
A
1
=
3[
3
0
(1 + 2)
1
0
]
8 + 24 + 16
2
0
coth 2h
. (11.45)
A =
_
3[
3
0
(1 + 2)
1
0
]
(1 3 + 2
2
0
coth 2h)(3
3
0

1
0
4
0
coth 2h)
_
1/2
. (11.46)
The remaining part of the calculation follows closely the work of Tad-
jbaksh and Keller [148] and Concus [33]. Integrating (11.37)(11.41), we
obtain
A
0
=
1
16
(3
0
+
3
0
) sin 2t +
1
8
A
2
(
0
coth
2
2h + 3
0
) sin 4t +
0
t +
0
,
(11.47)
A
1
=
A(4
0
coth 2h +
1
0
+ 21
3
0
)
32 sinh h
cos 3t, (11.48)
A
2
=
2
sin 2t, (11.49)
A
3
=
A
0
cos t[(4 + 48) coth 2h (3 + 24)
2
0
3
2
0
]
(12 + 20) sinh 3h
2
0
cosh 3h
A
0
cos 3t[(12 + 48) coth 2h + (3 + 24)
2
0
21
2
0
]
(12 + 20) sinh 3h 36
2
0
cosh 3h
, (11.50)
A
4
=
A
2
0
sin 4t[(2 + 30) coth 2h + 2
0
coth
2
2h 6
2
0
]
(4 + 60) sinh 4h 16
2
0
cosh 4h
. (11.51)
Here
0
,
0
and
2
are constants to be determined.
(1 )
1
1
xx
= F
0
0
4
m=0
A
m
(t) cos mxcosh mh, (11.52)
where F
0
and A
m
(t) are dened by (11.28) and (11.47)(11.51). The function
1
is therefore dened as the solution of (11.52) subject to (11.7).
The constant
0
in (11.47) is evaluated by integrating (11.52) with respect
to x between 0 and and using (11.7) and (11.8). Thus we nd
0
=
1
8
1
8
3
0
+
1
2
A
2
0
(1 coth
2
2h). (11.53)
This completes the determination of the rst-order solution. It contains
an arbitrary constant
2
. This constant may be determined at second order
in the way in which A was determined at rst order. The way in which the
calculations proceed should be clear to the reader.
Equation (11.46) shows that there that there are two solutions when
(11.22) is satised. These solutions are very similar to the Wilton rip-
ples of Section 6.5.3.1. Vanden-Broeck [167] showed that the two solutions
corresponding to (11.46) are members of two dierent families.
In conclusion, we have illustrated in this chapter that gravitycapillary
standing waves have properties qualitatively similar to those of the gravity
capillary travelling waves considered in Section 6.5.3.1. In particular, there
are many dierent families of solutions with dimples on their proles.
References
[1] Acheson, D. J. 1990, Elementary Fluid Dynamics Oxford University Press.
[2] Ackerberg, R. C. 1975, The eects of capillarity on free-streamline separa-
tion. J. Fluid Mech. 70, 333352.
[3] Akylas, T. R. 1993, Envelope solitons with stationary crests. Phys. Fluids
A 5, 789791.
[4] Akylas, T. R. & Grimshaw, R. 1992, Solitary internal waves with oscilla-
tory tails. J. Fluid Mech. 242, 279298.
[5] Amick, C. J., Fraenkel, L. E. & Toland, J. F. 1982, On the Stokes
conjecture and the wave of extreme form. Acta Math. 148, 193214.
[6] Anderson, C. D. & Vanden-Broeck, J.-M. 1996, Bow ows with surface
tension. Proc. Roy. Soc. Lond. A 452, 19851997.
[7] Asavanant, J. & Vanden-Broeck, J.-M. 1994, Free-surface ows past a
surface-piercing object of nite length. J. Fluid. Mech. 273, 109124.
[8] Batchelor, G. K. Fluid Dynamics. Cambridge University Press, 615 pp.
[9] Baker, G., Meiron D. & Orszag, S. 1982, Generalised vortex methods
for free surface ow problems J. Fluid Mech. 123, 477501.
[10] Beale, T. J. 1991, Solitary water waves with capillary ripples at innity.
Comm. Pure Appl. Maths 64, 211257.
[11] Benjamin, B. 1956, On the ow in channels when rigid obstacles are placed
in the stream. J. Fluid Mech. 1, 227248.
[12] Benjamin, B. 1962, The solitary wave on a stream with arbitrary distribu-
tion of vorticity. J. Fluid Mech. 12, 97116.
[13] Billingham, J. & King, A. C. 2000, Wave Motion. Cambridge University
Press.
[14] Binder, B. J., Dias, F. & Vanden-Broeck, J.-M. 2005, Forced solitary
waves and fronts past submerged obstacles. Chaos 15, 037106.
[15] Binder, B. J. & Vanden-Broeck, J.-M. 2005, Free surface ows past
surfboards and sluice gates. Euro. J. Appl. Math. 16, 601619.
[16] Binder, B. J. & Vanden-Broeck, J.-M. 2007, The eect of disturbances
on the ows under a sluice gate and past an inclined plate. J. Fluid Mech. 576,
475490.
308
References 309
[17] Binnie, A. M. 1952, The ow of water under a sluice gate. Q. J. Mech. Appl.
Math. 5, 395407.
[18] Birkhoff, G. & Carter, D. 1957, Rising plane bubbles. J. Math. Phys. 6
769779.
[19] Birkhoff, G. & Zarantonello, E. 1957, Jets, Wakes and Cavities Aca-
demic Press, 353 pp.
[20] Blyth, M. G. & Vanden-Broeck, J.-M. 2004, New solutions for capillary
waves on uid sheets. J. Fluid Mech. 507, 255264.
[21] Blyth, M. G. & Vanden-Broeck, J.-M. 2005, New solutions for capillary
waves on curved sheets of uids. IMA J. Appl. Math. 70, 588601.
[22] Brillouin M. 1911, Les surfaces de glissement de Helmoltz at la resistance
des uides. Ann. de Chim. Phys. 23, 145230.
[23] Brodetsky, S. 1923, Discontinuous uid motion past circular and elliptic
cylinders. Proc. Roy. Soc. London A 102, 114.
[24] Budden, P. & Norbury, J. 1982, Uniqueness of free boundary ows under
gravity. Arch. Rat. Mech. Anal. 78, 361380.
[25] Byatt-Smith, J. G. B. & Longuet-Higgins, M. S. 1976, On the speed
and prole of steep solitary waves. Proc. Roy. Soc. London. A 350, 175189.
[26] Champneys, A. R., Vanden-Broeck, J.-M. & Lord, G. J. 2002, Do
true elevation gravitycapillary solitary waves exist? A numerical investigation.
J. Fluid Mech. 454, 403417.
[27] Chen, B. & Saffman, P.G. 1979, Steady gravitycapillary waves on deep
water, Part I: Weakly nonlinear waves. Stud. Appl. Math. 60, 183210.
[28] Chen, B. & Saffman, P.G. 1980a, Numerical evidence for the existence of
new types of gravity waves on deep water. Stud. Appl. Math. 62, 121.
[29] Chen, B. & Saffman, P.G. 1980b, Steady gravitycapillary waves on deep
water, Part II: Numerical results for nite amplitude. Stud. Appl. Math. 62,
95111.
[30] Chung, Y. K. 1972, Solution of ow under a sluice gates. ASCE J. Eng.
Mech. Div. 98, 121140.
[31] Cokelet, E. D. 1977, Steep gravity waves in water of arbitrary uniform
depth, Phil. Trans. Roy. Soc. London A 286, 183230.
[32] Collins, R. 1965, A simple model of a plane gas bubble in a nite liquid. J.
Fluid Mech. 22, 763771.
[33] Concus, P. 1962, Standing capillarygravity waves of nite amplitude. J.
Fluid Mech. 14, 568576.
[34] Concus, P. 1964, Standing capillarygravity waves of nite amplitude: Cor-
rigendum. J. Fluid Mech. 19, 264266.
[35] Cooker, M. J., Weidman, P. D. & Bale, D. S. 1997, Reection of a
high-amplitude solitary wave at a vertical wall. J. Fluid Mech. 342, 141158.
[36] Cou
et, B. & Strumolo, G. S. 1987, The eects of surface tension and

tube inclination on a two-dimensional rising bubble. J. Fluid Mech. 213, 114.
[37] Crapper, G. D. 1957, An exact solution for progressive capillary waves of
arbitrary amplitude. J. Fluid Mech. 2, 572540.
[38] Crowdy, D. G. 1999, Exact solutions for steady capillary waves on a uid
annulus. J. Nonlinear Sci. 9, 615640.
310 References
[39] Cumberbatch, E. & Norbury, J. 1979, Capillarity modication of the
singularity at a free-streamline separation point. Q. J. Mech. Appl. Math. 32,
303312.
[40] Dagan, G. & Tulin, M. P. 1972, Two-dimensional free surface gravity
ows past blunt bodies. J. Fluid Mech. 51, 529543.
[41] Davies, T. V. 1951, Theory of symmetrical gravity waves of nite amplitude.
Proc. Roy. Soc. London A 208, 475486.
[42] Dias, F. & Iooss, G. 1993, Capillarygravity solitary waves with damped
oscillations. Physica D 65, 399423.
[43] Dias, F. & Kharif, C. 1999, Nonlinear gravity and capillarygravity waves.
Ann. Rev. Fluid Mech. 31, 301346.
[44] Dias, F., Menasce, D. & Vanden-Broeck, J.-M. 1996, Numerical study
of capillarygravity solitary waves. Eur. J. Mech. B Fluids 15, 1736.
[45] Dias, F. & Vanden-Broeck, J.-M. 1989, Open channel ows with sub-
merged obstructions. J. Fluid Mech. 206, 155170.
[46] Dias, F. & Vanden-Broeck, J.-M. 1992, Solitary waves in water of innite
depth and related free surface ows. J. Fluid Mech. 240, 549557.
[47] Dias, F. & Vanden-Broeck, J.-M. 1993, Nonlinear bow ows with
splashes. J. Fluid Mech. 255, 91102.
[48] Dias, F. & Vanden-Broeck, J.-M. 2002, Generalized critical free-surface
ows. J. Eng. Math. 42, 291301.
[49] Dias, F. & Vanden-Broeck, J.-M. 2004a, Trapped waves between sub-
merged obstacles. J. Fluid Mech. 509, 93102.
[50] Dias, F. & Vanden-Broeck, J.-M. 2004b, Two-layer hydraulic falls over
an obstacle. Eur. J. Mech. B Fluids 23, 879898.
[51] Dingle, R. B. 1973, Asymptotic Expansions: Their Derivation and Inter-
pretation. Academic Press.
[52] Eggers, J. 1995, Theory of drop formation. Phys. Fluids 7, 941953.
[53] Evans, W. A. B. & Ford, M. J. 1996, An exact integral equation for
solitary waves (with new numerical results for some internal properties). Proc.
Roy. Soc. London A 452, 373390.
[54] Fangmeier, D. D. & Strelkoff, T. S. 1968, Solution for gravity ow
under a sluice gate. ASCE J. Eng. Mech. Div. 94, 153176.
[55] Forbes, L.-K. 1981, On the resistance of a submerged semi-elliptical body.
J. Eng. Math. 15, 287298.
[56] Forbes, L.-K. 1983, Free surface ow over a semicircular obstruction includ-
ing the inuence of gravity and surface tension. J. Fluid Mech. 127, 283297.
[57] Forbes, L.-K. 1988, Critical free-surface ow over a semi-circular obstruc-
tion. J. Eng. Math. 22, 313.
[58] Forbes, L. K 1989, An algorithm for 3-dimensional free-surface problems in
hydrodynamics. J. Comput. Phys. 82, 330347.
[59] Forbes, L. K. & Schwartz, L. W. 1982, Free-surface ow over a semicir-
cular obstruction. J. Fluid Mech. 114, 299314.
[60] Forbes, L. K. & Hocking, G. C. 1990, Flow caused by a point sink in a
uid having a free surface. J. Austral. Math. Soc. Ser. B 32, 231249.
References 311
[61] Friedrics, K. O. & Hyers, D. H. 1954, The existence of solitary waves.
Comm. Pure Appl. Math. 7, 517550.
[62] Garabedian, P. R. 1957, On steady state bubbles generated by Taylor
instability. Proc. Roy. Soc. London A 241, 423431.
[63] Garabedian, P. R. 1985, A remark about pointed bubbles. Comm. Pure
Appl. Math. 38, 609612.
[64] Gleeson, H., Papageorgiou, D. T. & Vanden-Broeck, J.-M. 2007, A
new application of the Kortewegde-Vries BenjaminOno equation in interfacial
electrohydrodynamics. Phys. Fluids 19, 031703.
[65] Grandison, S. & Vanden-Broeck, J.-M. 2006, Truncation methods for
gravity capillary free surface ows. J. Eng. Math. 54, 8997.
[66] Grilli, S. T., Guyenne, P. & Dias, F. 2001, A fully non-linear model for
three-dimensional overturning waves over an arbitrary bottom. Int. J. Numer.
Meth. Fluids 35, 829867.
[67] Grimshaw, R. H. J. & Smyth, N. 1986, Resonant ow of a stratied uid
over topography. J. Fluid Mech. 169, 429464.
[68] Groves, M. D. & Sun, M. S. 2008, Fully localised solitary-wave solutions
of the three-dimensional gravitycapillary water-wave problem. Arch. Rat. Mech.
Anal. 188. 191.
[69] Gurevich, M. 1965, Theory of Jets and Ideal Fluids. Academic Press, 585
pp.
[70] Havelock, T. H. 1919, Periodic irrotational waves of nite amplitude. Proc.
Roy. Soc. London Ser. A 95, 3851.
[71] Helmholtz, H. 1868,

Uber discontinuierliche Fl ussigkeitsbewegungen.
Monatsber, Berlin Akad., 215228, reprinted in Phil. Mag. 36, 337346.
[72] Hocking, G. C. & Vanden-Broeck, J.-M. 1997, Draining of a uid of
nite depth into a vertical slot. Applied Math. Modelling 21, 643649.
[73] Hocking, G. C., Vanden-Broeck, J.-M. & Forbes, L. K. 2002, A note
on withdrawal from a uid of nite depth through a point sink. ANZIAM J. 44,
181191.
[74] Hogan, S. J. 1980, Some eects of surface tension on steep water waves.
Part 2. J. Fluid Mech. 96, 417445.
[75] Hunter, J. K. & Scherule, J. 1988, Existence of perturbed solitary wave
solutions to a model equation for water waves. Physica D 32, 253268.
[76] Hunter, J. K. & Vanden-Broeck, J.-M. 1983a, Solitary and periodic
gravitycapillary waves of nite amplitude. J. Fluid Mech. 134, 205219.
[77] Hunter, J. K. & Vanden-Broeck, J.-M. 1983b, Accurate computations
for steep solitary waves. J. Fluid Mech. 136, 6371.
[78] Iooss, G. & Kirrmann, P. 1996, Capillary gravity waves on the free
surface of an inviscid uid of innite depth existence of solitary waves. Arch.
Rat. Mech. Anal. 136, 119.
[79] Iooss, G. & Kirchgassner, K. 1990, Bifurcation dondes solitaires en
presences dune faible tension supercielle. C.R. Acad. Sci. Paris 311 I, 265
268.
[80] Iooss, G. & Kirchgassner, K. 1992, Water waves for small surface tension:
an approach via normal form. Proc. Roy. Soc. Edinburgh 122A, 267299.
312 References
[81] Iooss, G., Plotnikov, P. & Toland, J. F. 2005, Standing waves on an
innitely deep perfect uid under gravity. Arch. Rat. Mech. Anal. 177, 367478.
[82] Kang, Y. & Vanden-Broeck, J.-M. 2002, Stern waves with vorticity
ANZIAM J. 43, 321332.
[83] Kawahara, T. 1972, Oscillatory solitary waves in dispersive media. J. Phys.
Soc. Japan 33, 260264.
[84] Keller, H. B. 1977, Applications of Bifurcation Theory. Academic Press.
[85] Keller, J. B. & Miksis, M. J. 1983, Surface tension driven ows. SIAM
J. Appl. Math. 43, 268277.
[86] Keller, J. B., Milewski, P. & Vanden-Broeck, J.-M. 2000, Wetting
and merging driven by surface tension. Euro. J. Mech. B Fluids 19, 491502.
[87] Kim, B. & Akylas, T. R. 2005, On gravitycapillary lumps. J. Fluid Mech.
540, 337351.
[88] Kim, B. & Akylas, T. R. 2006, On gravitycapillary lumps, Part 2. Two
dimensional Benjamin equation. J. Fluid Mech. 557, 237256.
[89] Kinnersley, W. 1976, Exact large amplitude capillary waves on sheets of
uid. J. Fluid Mech. 77, 229241.
[90] Kirchhoff, G. 1869, Zur Theorie freier Fl ussigkeitsstrahlen. J. Reine
Angew. Math. 70, 289298.
[91] Korteweg, D. J. & G. de Vries 1895, On the change of form of long
waves advancing in a rectangular channel and on a new type of long stationary
waves. Phil. Mag. 39, 422443.
[92] Lamb H. 1945, Hydrodynamics, 6th edn, Cambridge University Press.
[93] Larock, B. E. 1969, Gravity-aected ow from planar sluice gate. ASCE J.
Engng Mech. Div. 96, 12111226.
[94] Lee, J. W. & Vanden-Broeck, J.-M. 1993, Two-dimensional jets falling
from funnels and nozzles. Phys. Fluids A5, 24542460.
[95] Lee, J. W. & Vanden-Broeck, J.-M. 1998, Bubbles rising in an inclined
two-dimensional tube and jets falling from along a wall. J. Austral. Math. Soc.
B 39, 332349.
[96] Lenau, C. W. 1966, The solitary wave of maximum amplitude. J. Fluid
Mech. 26, 309320.
[97] Lombardi, E. 2000, Oscillatory Integrals on Phenomena Beyond All Orders:
with Applications to Homoclinic Orbits in reversible systems. Lecture Notes in
Mathematics 1741, Springer.
[98] Lighthill, M. J. 1946, A note on cusped cavities. Aero. Res. Councial Rep.
and Mem. 2328.
[99] Lighthill, M. J. 1953, On boundary layers and upstream inuence, I. A
comparison between subsonic and supersonic ows. Proc. Roy. Soc. London A
217, 344357.
[100] Lighthill, M. J. 1978, Waves in Fluids, Cambridge University Press, 504
pp.
[101] LonguetHiggins, M. S. 1975, Integral properties of periodic gravity
waves of nite amplitude. Proc. Roy. London A 342, 157174.
[102] Longuet-Higgins, M. S. 1989, Capillary-gravity waves of solitary type on
deep water. J. Fluid Mech. 200, 451478.
References 313
[103] Longuet-Higgins, M. S. 1993, Capillarygravity waves of solitary type
and envelope solitons on deep water. J. Fluid Mech. 252, 703711.
[104] Longuet-Higgins, M. S. & Cokelet, E. 1976, The deformation of steep
surface waves on water, I. A numerical method of computation. Proc. Roy. Soc.
London A 350, 126.
[105] Longuet-Higgins, M. S. & Fenton, J. D. 1974, On the mass, momen-
tum, energy and circulation of a solitary wave, II. Proc. R. Soc. Lond. A 340,
471493.
[106] Longuet-Higgins, M. S. & Fox, M. J. H. 1978, Theory of the almost
highest wave, Part 2. Matching and analytical extension. J. Fluid Mech. 85,
769786.
[107] Maneri, C. C. 1970, The motion of plane bubbles in inclined ducts. Ph.D.
thesis, Polytechnic Institute of Brooklyn, New York.
[108] McCue, S. W. & Forbes, L. K. 2002, Free surface ows emerging from
beneath a semi-innite plate with constant vorticity. J. Fluid Mech. 461, 387
407.
[109] McLean, J. W. & Saffman, P. G. 1981, The eect of surface tension on
the shape of ngers in a Hele Shaw cell. J. Fluid Mech. 102, 455469.
[110] Mekias, H. & Vanden-Broeck, J.-M. 1991, Subcritical ow with a stag-
nation point due to a source beneath a free surface. Phys. Fluids A 3, 26522658.
[111] Michallet, H. & Dias, F. 1999, Numerical study of generalized interfacial
solitary waves. Phys. Fluids 11, 15021511.
[112] Michell, J. H. 1883, The highest wave in water. Phil. Mag. 36, 430437.
[113] Miksis, M., Vanden-Broeck, J.-M. & Keller, J. B. 1981, Axisymmet-
ric bubble or drop in a uniform ow. J. Fluid Mech. 108, 89101.
[114] Miksis, M., Vanden-Broeck, J.-M. & Keller, J. B. 1982, Rising bub-
bles. J. Fluid Mech. 123, 3141.
[115] Milewski, P. A. 2005, Three-dimensional localized solitary gravity
capillary waves. Comm. Math. Sc. 3, 8999.
[116] Nayfeh, A. H. 1970, Triple and quintuple-dimpled wave proles in deep
water. J. Fluid Mech. 13, 545550.
[117] Ockendon, H. & Ockendon, J. R. 2004, Viscous Flow. Cambridge Texts
in Applied Mathematics.
[118] Olfe, D. B. & Rottman, J. W. 1980, Some new highest-wave solutions
for deep-water waves of permanent form. J. Fluid Mech. 100, 801810.
[119] Osher, S. & Fedkiw, R. 2003, Level Set Methods and Dynamic Implicit
Surfaces. Applied Mathematical Sciences 153, Springer.
[120] Papageorgiou, D. T. & Vanden-Broeck, J.-M. 2003, Large amplitude
capillary waves in electried uid sheets. J. Fluid Mech. 508, 7188.
[121] Papageorgiou, D. T. & Vanden-Broeck, J.-M. 2004, Antisymmetric
capillary waves in electried uid sheets. Eur. J. Appl. Math. 15, 609623.
[122] Parau, E. & Vanden-Broeck, J.-M. 2002, Nonlinear two- and three-
dimensional free surface ows due to moving disturbances. Eur. J. Mech. B
Fluids 21, 643656.
[123] Parau, E., Vanden-Broeck, J.-M. & Cooker, M. 2005a, Nonlinear
three dimensional gravity capillary solitary waves. J. Fluid Mech. 536, 99105.
314 References
[124] Parau, E., Vanden-Broeck, J.-M. & Cooker, M. 2005b, Three-
dimensional gravitycapillary solitary waves in water of nite depth and related
problems. Phys. Fluids 17, 122 101.
[125] Parau, E., Vanden-Broeck, J.-M. & Cooker, M. 2007a, Three-
dimensional capillarygravity waves generated by a moving disturbance. Phys.
Fluids 19, 082 102.
[126] Parau, E., Vanden-Broeck, J.-M. & Cooker, M. 2007b, Nonlinear
three dimensional interfacial ows with a free surface. J. Fluid Mech. 591, 481
494.
[127] Pullin, D. I. & Grimshaw, R. H. J. 1988, Finite amplitude solitary
waves at the interface between two homogeneous uids. Phys. Fluids 31, 3550
3559.
[128] Rayleigh, Lord 1883, The form of standing waves on the surface of run-
ning water. Proc. Lond. Math. Soc. 15, 6978.
[129] Romero, L. 1982, Ph.D. thesis, California Institute of Technology.
[130] Saffman, P. G. 1980, Long wavelength bifurcation of gravity waves on
deep water. J. Fluid Mech. 101, 567581.
[131] Saffman, P. G. 1986, Viscous ngering in Hele Shaw cells. J. Fluid Mech.
173, 7394.
[132] Saffman, P. G. & Taylor, G. I. 1958, The penetration of a uid into
a porous medium or Hele Shaw cell containing a more viscous uid. Proc. Roy.
Soc. London A 245, 312329.
[133] Schwartz, L. W. 1974, Computer extension and analytic continuation of
Stokes expansion for gravity waves. J. Fluid Mech. 62, 553578.
[134] Schwartz, L. W. & Fenton, J. 1982, Strongly nonlinear waves. Ann.
Rev. Fluid Mech. 14, 3960.
[135] Schwartz, L. W. & Vanden-Broeck, J.-M. 1979, Numerical solution of
the exact equations for capillarygravity waves. J. Fluid Mech. 95, 119139.
[136] Schultz, W. W., Vanden-Broeck, J.-M., Jiang, L. & Perlin, M.
1998, Highly nonlinear water waves with small capillary eect. J. Fluid Mech.
369, 253272.
[137] Sethian, J. A. Level Set Methods. Cambridge Monographs on Applied and
Computational Mathematics, Cambridge University Press.
[138] Sha, H. & Vanden-Broeck, J.-M. 1993, Two-layer ows past a semicir-
cular obstacle Phys. Fluids A 5, 26612668.
[139] Sha, H. & Vanden-Broeck, J.-M. 1997, Internal solitary waves with
stratication in density. J. Austral. Math. Soc. B 38, 563580.
[140] Shen, S.-P. 1995, On the accuracy of the stationary forced Korteweg-de-
Vries equation as a model equation for ows over a bump. Quart. J. Appl. Math.
53, 701719.
[141] Simmen, J. A. & Saffman, P. G. 1985, Steady deep water waves on a
linear shear current. Stud. Appl. Maths 75, 3557.
[142] Southwell, R. V. & Vaisey G. 1946, Fluid motions characterised by
free streamlines. Phil. Trans. Roy. Soc. A 240, 117161.
[143] Stokes, G. G. 1847, On the theory of oscillatory waves. Camb. Trans. Phil.
Soc. 8, 441473.
References 315
[144] Stokes, G. G. 1880, in Mathematical and Physical Papers, Vol. 1, p. 314,
Cambridge University Press.
[145] Sun, S. M. 1991, Existence of generalized solitary wave solution for water
with positive Bond number less than 1/3. J. Math. Anal. Appl. 156, 471504.
[146] Sun, S. M. 1999, Nonexistence of truly solitary waves in water with small
surface tension Proc. Roy. Soc. London A 455, 21912228.
[147] Sun, S. M. & Shen, M. C. 1993, Exponentially small estimate for the
amplitude of capillary ripples of generalised solitary waves. J. Math. Anal. Appl.
172, 533566.
[148] Tadjbakhsh, I. & Keller, J. B. 1960, Standing surface waves of nite
amplitude. J. Fluid Mech. 8, 442451.
[149] Tanaka, M., Dold, J. W., Lewy, M. & Peregrine, D. H. 1987, In-
stability and breaking of a solitary wave. J. Fluid Mech. 185, 235248.
[150] Teles da Silva, A. F. & Peregrine, D. H. 1988, Steep solitary waves
in water of nite depth with constant vorticity. J. Fluid Mech. 195, 281305.
[151] Tooley, S. & Vanden-Broeck, J.-M. 2002, Waves and singularities in
nonlinear capillary free-surface ows. J. Eng. Math. 43, 8999.
[152] Tsai, W. T. & Yue D. K. 1996, Computation of nonlinear free surface
ows. Ann. Rev. Fluid Mech. 28, 249278.
[153] Tseluiko, D., Blyth, M. & Papageorgiou, D. T. 2008a, Electried
viscous thin lm over topography. J. Fluid Mech. 597, 449475.
[154] Tseluiko, D., Blyth, M. & Papageorgiou, D. T. 2008b, Eect of an
electric eld on lm ow down a corrugated wall at zero Reynolds number. Phys.
Fluids 20, 042 103
[155] Turner, R. E. L. & Vanden-Broeck, J.-M. 1986, The limiting cong-
uration of interfacial gravity waves. Phys. Fluids 29, 372375.
[156] Turner, R. E. L. & Vanden-Broeck, J.-M. 1988, Broadening on inter-
facial solitary waves. Phys. Fluids 31, 24862490.
[157] Turner, R. E. L. & Vanden-Broeck, J.-M. 1992, Long internal waves.
Phys. Fluids A 4, 19291935.
[158] Vanden-Broeck, J.-M. 1980, Nonlinear stern waves. J. Fluid Mech. 96,
601610.
[159] Vanden-Broeck, J.-M. 1981, The inuence of capillarity on cavitating
ow past a at plate. Quart. J. Mech. Appl. Math. 34, 465473.
[160] Vanden-Broeck, J.-M. 1983a, The inuence of surface tension on cavi-
tating ow past a curved obstacle. J. Fluid Mech. 133, 255264.
[161] Vanden-Broeck, J.-M. 1983b, Fingers in a Hele-Shaw cell with surface
tension. Phys. Fluids 26, 20332034.
[162] Vanden-Broeck, J.-M. 1983c, Some new gravity waves in water of nite
depth. Phys. Fluids 26, 23852387.
[163] Vanden-Broeck, J.-M. 1984a, The eect of surface tension on the shape
of the Kirchho jet. Phys. Fluids 27, 19331936.
[164] Vanden-Broeck, J.-M. 1984b, Numerical solutions for cavitating ow of
a uid with surface tension past a curved obstacle. Phys. Fluids 27, 26012603.
[165] Vanden-Broeck, J.-M. 1984c, Bubbles rising in a tube and jets falling
from a nozzle. Phys. Fluids 27, 10901093.
316 References
[166] Vanden-Broeck, J.-M. 1984d, Rising bubbles in a two-dimensional tube
with surface tension. Phys. Fluids 27, 26042607 and 1992, Rising bubbles in
a two-dimensional tube: asymptotic behavior for small values of the surface
tension, Phys. Fluids A 4, 23322334.
[167] Vanden-Broeck, J.-M. 1984e, Nonlinear gravitycapillary standing waves
in water of arbitrary uniform depth. J. Fluid Mech. 139, 97104.
[168] Vanden-Broeck, J.-M. 1985, Nonlinear free-surface ows past two-
dimensional bodies. In Advances in Nonlinear Waves, Vol. II, L. Debnath, ed.,
Boston, Pitman.
[169] Vanden-Broeck, J.-M. 1986a, Pointed bubbles rising in a two dimensional
tube. Phys. Fluids 29, 13431344.
[170] Vanden-Broeck, J.-M. 1986b, A free streamline model for a rising bubble.
Phys. Fluids 29, 27982801.
[171] Vanden-Broeck, J.-M. 1986c, Flow under a gate. Phys. Fluids 29, 3148
3151.
[172] Vanden-Broeck, J.-M. 1986d, Steep gravity waves: Havelocks method
revisited. Phys. Fluids 29, 30843085.
[173] Vanden-Broeck, J.-M. 1987, Free-surface ow over an obstruction in a
channel. Phys. Fluids 30, 23152317.
[174] Vanden-Broeck, J.-M. 1988, Joukovskiis model for a rising bubble. Phys.
Fluids 31, 974977.
[175] Vanden-Broeck, J.-M. 1989, Bow ows in water of nite depth. Phys.
Fluids A1, 13281330.
[176] Vanden-Broeck, J.-M. 1991a, Cavitating ow of a uid with surface ten-
sion past a circular cylinder. Phys. Fluids A 3, 263266.
[177] Vanden-Broeck, J.-M. 1991b, Elevation solitary waves with surface ten-
sion Phys. Fluids A 3, 26592663.
[178] Vanden-Broeck, J.-M. 1994, Steep solitary waves in water of nite depth
with constant vorticity. J. Fluid Mech. 274, 339348.
[179] Vanden-Broeck, J.-M. 1995, New families of steep solitary waves in water
of nite depth with constant vorticity. Eur. J. Mech. B uids 14, 761774.
[180] Vanden-Broeck, J.-M. 1996a, Periodic waves with constant vorticity in
water of innite depth. IMA J. Appl. Math. 56, 207217.
[181] Vanden-Broeck, J.-M. 1996b, Numerical calculations of the free-surface
ow under a sluice gate. J. Fluid Mech. 330, 339347.
[182] Vanden-Broeck, J.-M. 2002, Wilton ripples generated by a moving pres-
sure distribution. J. Fluid Mech. 451, 193201.
[183] Vanden-Broeck, J.-M. 2004, Nonlinear capillary free-surface ows. J.
Eng. Math. 50, 415426.
[184] Vanden-Broeck, J.-M. & Dias, F. 1992, Gravitycapillary solitary waves
in water of innite depth and related free-surface ows. J. Fluid Mech. 240, 549
557.
[185] Vanden-Broeck, J.-M. & Keller, J. B. 1980, A new family of capillary
waves. J. Fluid Mech. 98, 161169.
[186] Vanden-Broeck, J.-M. & Keller, J. B. 1989, Surng on solitary waves.
J. Fluid Mech. 198, 115125.
References 317
[187] Vanden-Broeck, J.-M. & Keller, J. B. 1997, An axisymmetric free
surface with a 120 degree angle along a circle. J. Fluid Mech. 342, 403409.
[188] Vanden-Broeck, J.-M. & Miloh, T. 1995, Computations of steep gravity
waves by a renement of the DaviesTulin approximation. Siam J. Appl. Math.
55, 892903.
[189] Vanden-Broeck, J.-M. & Schwartz, L. W. 1979, Numerical computa-
tion of steep gravity waves in shallow water. Phys. Fluids 22, 18681871.
[190] Vanden-Broeck, J.-M., Schwartz, L. W. & Tuck, E. O. 1978, Diver-
gent low-Froude-number series expansion in nonlinear free-surface ow problems.
Proc. Roy. Soc. London A 361, 207224.
[191] Vanden-Broeck, J.-M. & Shen, M. C. 1983, A note on solitary and
periodic waves with surface tension. Z. Angew. Math. Phys. 34, 112117.
[192] Vanden-Broeck, J.-M. & Tuck, E. O. 1977. Computation of near-bow
or stern ows, using series expansion in the Froude number. In Proc. 2nd Int.
Conf. on Num. Ship Hydrodynamics, Berkeley, California, 371381.
[193] Vanden-Broeck, J.-M. & Tuck, E. O. 1994, Steady inviscid rotational
ows with free surfaces J. Fluid Mech. 258, 105113.
[194] Villat, H. 1914, Sur la validite des solutions de certains probl`emes
dhydrodynamique. J. de Math. 10, 231290.
[195] Whitham, G. B. 1974, Linear and nonlinear waves. Wiley Interscience,
John Wiley & Sons.
[196] Wehausen, J. V. & Laitone, E. V. 1960, Surface waves. In Handbuch
der Physik, C. Truesdell, ed., Vol. IX, pp. 446778, Springer.
[197] Williams, J. M. 1981, Limiting gravity waves in water of nite depth. Phil.
Trans. R. Soc. Lond. A 302, 139188.
[198] Wilton, J. R., 1915, On ripples. Phil. Mag. 29, 688700.
[199] Zufuria, J. A. 1987, Symmetry breaking in periodic and solitary gravity
capillary waves on water of nite depth. J. Fluid Mech. 184, 183206.
Index
Bernoulli constant, 11, 73
Bernoulii equation, 8, 94
Bond number, 20, 158
boundary integral, 53
boundary integral equation, 150, 157
bow ow, 226, 236
BrillouinVillat condition, 3, 69, 71
capillary wave, pure, 21
Cauchys integral formula, 14
Cauchys theorem, 14
cavitation number, 53
cnoidal wave, 146, 190
conformal mapping, 13
contraction ratio, 41, 91, 227
Crappers solution, 52, 160, 242
dispersion relation, linear, 20
dynamic boundary condition, 11
Eulers equations, 7
ow
bow, 32
stern, 32
subcritical, 22
supercritical, 22
free surface, 1, 8, 14
Froude number, 5, 20, 82, 91, 92, 95, 158, 229,
279
Garabedian energy argument, 93
gravity wave, pure, 21
group velocity, 27
Hele Shaw cell, 103
Hilbert transform, 54
intersection of ows and walls, 31
Jacobian matrix, 50
Joukowskii models, 110
Joukowskii transformation, 197
kinematic boundary condition, 9
Kortewegde Vries equation, 156
fth-order, 147
linear standing wave, 24
logarithmic hodograph, 34
mean curvature of uid surface,
11
Newtons method, 49
phase velocity, 19, 26, 27
potential ows, 8
pure capillary wave, 21
pure gravity wave, 21
radiation condition, 30, 117,
225
Rayleigh viscosity, 300
series truncation, 40, 47, 70,
150
shallow water equations, 144
solitary wave, generalised,
147
stern ow, 225
subcritical ow, 22
supercritical ow, 22
terminant, 223
wave speed, 268
Weber number, 99, 201
Wilton ripple, 140, 206
318

Gravity Capillary Free Surface Flows

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Gravity Capillary Free Surface Flows

Hochgeladen von

Copyright:

Verfügbare Formate

CUUK1062-Vanden 9780521811903 May 28, 2010 12:43

This page intentionally left blank

denotes the pressure as x

is an arbitrary value of t and is the angular frequency. The value

cos k(x +ct)dt = 0.

from the identity

for 1 < t < d by using (3.38) and (3.45). We

b, the semi-ellipse is thin and the conguration of Figure 3.14 can

b 0. However, to study ows with free separation we

= 1 on = 0, 0 < < 1, (3.51)

in Figure 3.19). They are

the solutions are not acceptable because the free surfaces

in Figure 3.19). The last

, has free surfaces that approach the x-axis

can be obtained by considering

, occur in many applications. A classical example is the exact so-

0.73, the free surface

, the free surface

can be obtained by preventing the free surface from

, by preventing the crossing of the streamlines

= 1, 0 < < 1, (3.76)

= 1 at E, (3.6) implies that

satises Laplaces equation in

= 0 (an unbounded value of

invalidates the linearisation).

0, we follow Vanden-Broeck [159]

(see Figure 3.28). Thus

, Figure 3.28 shows that C < 0 and the values of predicted by

, which is known as the one

. Therefore the particular solution

and reected in the wall AB. Here

(x, y) = X(x)Y (y). (3.205)

and reecting it in the wall AB.

H (see (3.225)). The values of

angle at the apex

when surface tension was neglected. We then showed that a

, could be selected by solving the problem with

in agreement with (3.260).

and q in (3.266) and (3.267) are dened in terms of the dimensionless

of the wake. When surface tension is neglected

of the bubble yields a discrete set of values of F for which = /2.

. This ow can be characterised by the Froude number

and the corresponding proles are found to be

0.527. The dots are the

A(a) cosh a(y +h)e

is a root of 'D(a), we seek a root of D(a) for

)[ is small. Subsituting (4.34) into (4.32) and expanding in a

) = U, we can rewrite (4.36) as

(a) > 0 and

(a) < 0. The rst possibility occurs for a = a

) > 0. Therefore the pole a

respectively. These contours were

D(a) ,= 0 for 0 < a < , we obtain

for > 1/4.

is complex when > 0.25.

[(, y) (0, y)] =

at the midpoints by centred

decays exponentially as [[ (see [76]). The particular choice is not

. As described at the beginning of Section 3.1.2.2, preventing the

(see Figure 6.12).

0.663. Some typical proles

(see Figure 6.20).

(0) is the speed at