Advanced Series on Ocean Engineering — Volume 35

DYNAM ICS OF
WATEM WAVES
Selected Papers of
Michael Longuet-Higgins
Volume 3

jT;--

S G Sajjadi
editor

World Scientific
DYNAMICS OF
WATER WAVES
Selected Papers of
Michael Longuet-Higgins
V olum e 3
ADVANCED SERIES ON OCEAN ENGINEERING

Series Editor-in-Chief
Philip L- F Liu (Cornell University)

Published
Vol. 24 Introduction to Nearshore Hydrodynamics
by lb A. Svendsen (Univ. of Delaware, USA)
Vol. 25 Dynamics of Coastal Systems
by Job Dronkers (Rijkswaterstaat, The Netherlands)
Vol. 26 Hydrodynamics Around Cylindrical Structures (Revised Edition)
by B. Mutlu Sumer and Jorgen Fredsoe (Technical Univ. of Denmark,
Denmark)
Vol. 27 Nonlinear Waves and Offshore Structures
by Cheung Hun Kim (Texas A&M Univ., USA)
Vol. 28 Coastal Processes: Concepts in Coastal Engineering and Their
Applications to Multifarious Environments
by Tomoya Shibayama (Yokohama National Univ., Japan)
Vol. 29 Coastal and Estuarine Processes
by Peter Nielsen (The Univ. of Queensland, Australia)
Vol. 30 Introduction to Coastal Engineering and Management (2nd Edition)
by J. William Kamphuis (Queen’s Univ., Canada)
Vol. 31 Japan’s Beach Erosion: Reality and Future Measures
by Takaaki Uda (Public Works Research Center, Japan)
Vol. 32 Tsunami: To Survive from Tsunami
by Susumu Murata (Coastal Development Inst, of Technology, Japan),
Fumihiko Imamura (Tohoku Univ., Japan),
Kazumasa Katoh (Musashi Inst, of Technology, Japan),
Yoshiaki Kawata (Kyoto Univ., Japan),
Shigeo Takahashi (Port and Airport Research Inst., Japan) and
Tomotsuka Takayama (Kyoto Univ., Japan)
Vol. 33 Random Seas and Design of Maritime Structures, 3rd Edition
by Yoshimi Goda (Yokohama National University, Japan)
Vol. 34 Coastal Dynamics
by Willem T. Bakker (Delft Hydraulics, Netherlands)
Vol. 35 Dynamics of Water Waves: Selected Papers of Michael Longuet-Higgins
Volumes 1-3
edited by S. G. Sajjadi (Embry-Riddle Aeronautical University, USA)
Vol. 36 Ocean Surface Waves: Their Physics and Prediction
(Second Edition)
by Stanisfaw R. Massel (Institute of Oceanology of the Polish Academy of
Sciences, Sopot, Poland)

*F or the co m p lete list o f titles in this se ries, p lea se w rite to the P ub lisher.
 Advanced Series on Ocean Engineering — Volume 35

DYNAMICS OF
WATER WAVES
Selected Papers of
Michael Longuet-Higgins
Volume 3

editor
S G Sajjadi
Embry-Riddle Aeronautical University, USA

№ World Scientific
NE W J E R S E Y • L ONDON • S I N G A P OR E • B E I J I N G • S H A N G H A I • HONG KONG • TAI PEI • C HE NNA I
Published by

World Scientific Publishing Co. Pte. Ltd.

5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.

Advanced Series on Ocean Engineering — Vol. 35

DYNAM ICS OF W ATER W AVES
Selected Papers o f M ichael Longuet-Higgins
Volumes 1 -3
Copyright © 2013 by World Scientific Publishing Co. Pte. Ltd.

All rights reserved. This book, o r parts thereof, may not be reproduced in any form o r by any means, electronic
o r mechanical, including photocopying, recording o r any information storage and retrieval system now known
o r to be invented, without written permission from the Publisher.

For photocopying o f material in this volume, please pay a copying fee through the Copyright Clearance Center,
Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from
the publisher.

ISBN 978-981-4322-51-5 (Set)

ISBN 978-981-4322-52-2 (V ol. 1)
ISBN 978-981-4322-53-9 (V ol. 2)
ISBN 978-981-4322-54-6 (Vol. 3)

Printed in Singapore by В & Jo Enterprise Pte Ltd

 Preface by S G Sajjadi

It is a pleasure and honour to write this brief preface to introduce this three-
volume selection of papers by a close colleague, Professor Michael Longuet-
Higgins, FRS.
As a biographical sketch written by him follows this preface I will limit
myself to some thoughts of mine having to do with his influence on our
community. From the recognition which Professor Longuet-Higgins has received
in appreciation of his work I would like to mention his work on microseisms.
Scientists have long known about microseisms, but, until recently, no one
could determine where they came from. Microseisms were first recorded as a
strange, continuous buzz on the earliest seismometers, devices that measure
earthquakes. Every year, the cumulative energy released by these small vibrations
equals the amount of energy released globally from earthquakes. Records of
microseismic activity give scientists a history of wave interaction in Earth’s
oceans since the early 20th century. They are also used to examine the history
of storms over the ocean. Scientists are interested in learning where these
microseisms originate because the information can help them monitor stress in
Earth’s crust.
The theory of the origin of microseisms was first introduced in 1950 by
Michael Longuet-Higgins whilst at the University of Cambridge in England.
Longuet-Higgins suggested that the vibrations originated in places where ocean
waves were traveling in opposite directions toward each other at the same
frequency and at a certain ocean depth. According to his theory, when these
waves collide, they combine to form stationary waves that remain in a constant
position over large areas of the ocean. These waves create tall, pulsing columns
of pressure that repeatedly beat down on the ocean floor, causing it to vibrate.
The vibrations generate seismic surface waves, which spread out thousands of
miles and are detected by seismometers as noise. This new study on microseisms,
which appeared in the March 2007 issue of the Proceedings of the Royal Society,
Series A was part of interdisciplinary collaboration, which included Longuet-
Higgins and researchers from the California Institute of Technology, Pasadena
and the Hydrologic Research Center in San Diego.
This three volume set presents selected original research papers of Professor
Longuet-Higgins in the various areas of water waves, ocean dynamics, and
fluid mechanics. I believe they will serve as a milestone and beacon for future
generations.
Apart from the subject area covered in these three volumes, Longuet-Higgins
has published papers in other fields such as electromagnetic measurement of
tidal streams and ocean currents, time-varying currents depending on the Earth’s
rotation, projective geometry, etc. These papers have not been included in the
present selection, but it is hoped that they will be published in a separate volume
in the near future.
Finally, my thanks are due to a number of publishers for their permissions for
photographic reproduction of the original papers. In particular I wish to thank the
Royal Society of London, the Oxford University Press, the Cambridge University
Press, MacMillan Publishing Co., the Journal of Physical Oceanography, the
Journal of Marine Research, the American Physical Society, the American
Geophysical Union, Prentice-Hall, Pergamon Press and Elsevier Publishing Co.

S G Sajjadi
ERAU, Florida
December 2008
 Preface by Michael Longuet-Higgins

The practical consequences of wave motion for both the coastal engineer and the
geophysicist are many and varied, and new applications constantly arise to keep
the subject alive. At the same time, surface waves challenge the fluid dynamist to
find an explanation for such spectacular phenomena as wave breaking, when the
upper surface overturns on itself.
The present collection of original papers is not intended to cover the whole
subject. It is simply a selection of basic contributions by a single author and
his collaborators. On reflection, the papers can be seen to have been guided by
certain points of view: (1) a preference for simple or “physical” explanations
where possible, and particularly for geometrical interpretations. (2) a desire
to solve problems by accurate analysis initially, but then to find simpler
approximate models, easier to apply in practice. (3) an acknowledgement of
the need to compare theoretical results with field observations or laboratory
experiments. Like many other subjects, fluid mechanics advances by putting one
foot forward and then the other.
For convenience, this collection is arranged according to subjects. Included
in this first Volume are the author’s principal papers on microseisms; on mass
transport by water waves; on stochastic processes, especially applied to wind-
waves; on various mechanisms of wave generation by wind; and on the theory
and consequences of radiation stresses (momentum flux due to waves).
Conference papers are generally omitted, since usually they consists of reviews
of the original papers, doubtless in a more readable form. Preceding each group
of papers, there is a brief introduction giving the circumstances under which the
papers were written, and providing further background information.
As an appendix to Volume III, a list is given of the author’s papers on other
subjects. Some, as for instance time-varying ocean currents, are closely related to
papers in this Collection, while other subjects are widely different.
My thanks are due to Professor SG Sajjadi for offering to undertake
the editing of this selection of my papers. I would also like to express my
indebtedness to some of my mentors, colleagues and collaborators, aside
from those named explicitly in the subsequent text. In particular I wish to
mention CV Durrell, the senior mathematics don at Winchester College, for his
thorough mathematical grounding, and for his encouragement of my first forays
into mathematical research; to Dr George Deacon, leader of Group W at the
Admiralty Research Laboratory at Teddington, England, from 1944 to 1949 and
subsequently the first Director of the UK National Institute of Oceanography
at Witley, for his consistent support; to Norman Barber, a senior physicist in
group W; he was my first research supervisor who introduced me to “physical”
ways of looking at mathematical problems; to Sir Harold Jeffreys, for his
inspiring lectures on probability and statistics at Cambridge University; and to
Sir Geoffrey Taylor, for his inimitable style of research. Walter Munk has been
a long-time friend and advocate. It is be appropriate to mention those many
anonymous referees who have helped me on my scientific journey.
Perhaps the greatest credit of all should go to the subject of water waves
for affording the student such great pleasure and interest, and to the waves
themselves, for following so obediently the laws of mathematical physics.
 A B rief B iography

In 1943, at the age of seventeen, the author entered Trinity College, Cambridge,
with a scholarship in mathematics from Winchester College, the school where he
had spent the first four years of World War II. Already he had acquired a taste for
research in geometry. By the summer of 1945, he had qualified for a Cambridge
BA in mathematics. He was then required to do three years “work of national
importance.” Fortunately for him, he was assigned to “Group W” (for waves) at
the Admiralty Research Laboratory, Teddington. This Group had been formed in
June 1944 to study the long-distance propagation of ocean waves in preparation
for projected military operations in the Pacific Ocean. The Group had been
spectacularly successful, and its lease of life had been extended, with wider terms
of reference. During three years at Teddington (1945-8), he worked not only
on the theory of wind waves but also on the geomagnetic induction of voltages
by tidal streams, and on the generation of oceanic microseisms. In September
1945, he returned to Cambridge to read for a PhD. There was, however, no
break in his research; he continued to develop the same subjects of interest,
reporting at the end of each term to Sir Harold Jeffreys, and later to Dr Robert
Stoneley. In 1951, he was awarded a 4-year research Fellowship (Title A) at
Trinity College. The first year (1951-2) he spent in the USA as a Commonwealth
Fund Fellow, staying first at the Woods Hole Oceanographic Institute on Cape
Code, Massachusetts, and then at the Scripps Institution of Oceanography at La
Jolla, California, with Walter Munk. At Scripps he became interested in wave
generation by wind, in the statistical properties of sea states, and in several other
topics.
On his return to England in 1952, he spent two years of his Research
Fellowship in Cambridge after which he accepted Dr George Deacon’s invitation
to join the newly formed N10, the UK’s National Institute of Oceanography,
at Witley in Surrey. There he was to spend thirteen happy and fruitful years,
working (mainly) on ocean waves. After 1963, he concentrated more on time-
varying ocean currents, especially those which depend essentially on the rotation
of the Earth. This period also included visiting appointments in the Mathematics
Department at MIT (1957-8), the University of Adelaide, Australia (1964) and
at the University of California, San Diego (1961—2 and 1966-7). In 1963, he
was elected to the Royal Society of London. From 1967 to 1969, he spent two
years assisting in the expansion of the Department of Oceanography at Oregon
State University, in Corvallis, Oregon, but in 1969, he was appointed to a Royal
Society Research Professorship, to be held jointly at Cambridge University, in
the Department of Applied Mathematics and Theoretical Physics, and at The
National Institute of Oceanography at Witley. Once there, he decided on a multi­
faceted attack on the problem of wave breaking, involving both theory and
innovative experiments in the field and laboratory. He was also free to pursue
research in other subjects.
In 1989, on reaching the age of formal retirement, he wished to continue
doing research and for two years joined the La Jolla Institute in San Diego,
California. In 1991, he was appointed to a position as a Senior Research Physicist
at the Institute for Nonlinear Science in the University of California, San Diego,
with an Adjunct Professorship at the Scripps Institution of Oceanography. There
he turned attention to the natural sources of underwater sound, particularly the
sound produced by the creation of bubbles in breaking waves (previously this
sound was call “wind noise”). Financial support came mainly from the US Office
of Naval Research and from the National Science Foundation. Since his second
“retirement” in 2001, he has used his freedom as a Research Physicist Emeritus
to indulge his interest in a variety of problems without the necessity of applying
for outside grants. These interests have included the damping of incoming swells
by sand ripples, and the construction of very simple but accurate approximations
to gravity waves of limiting steepness.
 xi

Contents

Preface by S G Sajjadi v

Preface by Michael Longuet-Higgins vu

A Brief Biography IX

Volume I

Introductory Notes for P art В 1

В. Oceanic Microseisms 3

В 1. Sea Waves and Microseisms, 5

Nature, Lond. 162 (1948) 700 (with F. Ursell).
B2. A Theory of the Origin of Microseisms, 7
Phil. Trans. R. Soc. Lond. A 243 (1950) 1-35.
B3. An Experimental Study of the Pressure Variations in 43
Standing Water Waves,
Proc. R. Soc. Lond. A 206 (1951) 424-435 (with R.I.B. Cooper).
B4. Can Sea Waves Cause Microseisms? 55
Proc. Symposium on Microseisms, U.S. Nat. Acad. Sci. Publ.
306(1952) 74-93.

Introductory Notes for P art С 75

С. Mass Transport in W ater Waves 77

С 1. Mass Transport in Water Waves, 79

Phil. Trans. R. Soc. Lond. A 245 (1953) 535-581.
C2. On the Decrease of Velocity with Depth in an 126
Irrotational Water Wave,
Proc. Cam. Phil. Soc. 49 (1953) 552-560.
C3. The Mechanics of the Boundary-Layer Near the Bottom in a 135
Progressive Wave (Appendix to a paper by R.C.H. Russell
and J.D.C. Osorio),
Proc. 6th Conf. on Coastal Eng. (Miami, 1957), pp. 183-193.

C o n f e r e n c e papers (n ot in clu d ed ).
xii

C4. Mass Transport in the Boundary Layer at a Free Oscillating Surface, 146
J. Fluid Mech. 8 (1960) 293-306.
C5. Steady Currents Induced by Oscillations Round Islands, 160
J. Fluid Mech. 42 (1970) 701-720.
C6. Peristaltic Pumping in Water Waves, 181
J. Fluid Mech. 137 (1983) 393-407.

Introductory Notes for P art D 197

D. Stochastic Processes 199

D 1. On the Statistical Distribution of the Heights of Sea Waves, 201

J. Mar. Res. 11 (1952) 245-266.
D2. The Statistical Distribution of the Maxima of a Random Function, 223
Proc. R. Soc. Lond. A 237 (1956) 212-232 (with D.E. Cartwright).
D3. Statistical Properties of a Moving Wave-Form, 244
Proc. Cam. Phil. Soc. 52 (1956) 234-245.
D4. On the Velocities of the Maxima in a Moving Wave-Form, 256
Proc. Caw. Phil. Soc. 53 (1957) 230-233.
D5. The Statistical Analysis of a Random, Moving Surface, 260
Phil. Trans. R. Soc. Lond. A 249 (1957) 321-387.
D6. Statistical Properties of an Isotropic Random Surface, 327
Phil. Trans. R. Soc. Lond. A 250 (1957) 157-174.
D7. A Statistical Distribution Arising in the Study of the Ionosphere, 345
Proc. Phys. Soc. В 70 (1957) 559-565.
D8. On the Intervals Between Successive Zeros of a 352
Random Function,
Proc. R. Soc. Lond. A 246 (1958) 99-118.
D9. The Statistical Distribution of the Curvature of a 372
Random Gaussian Surface,
Proc. Cam. Phil. Soc. 54 (1958) 439-453.
D 10. The Distribution of the Sizes of Images Reflected in a 387
Random Surface,
Proc. Cam. Phil. Soc. 55 (1959) 91-100.
 xiii

D ll. The Focusing of Radiation by a Random Surface When the 397

Source is at a Finite Distance,
Proc. Cam. Phil. Soc. 56 (1960) 27-40.
D12. Reflection and Refraction at a Random Moving Surface. 411
I. Pattern and Paths of Specular Points,
J. Opt. Soc. Amer. 50 (1960) 838-844.
D13. Reflection and Refraction at a Random Moving Surface, 418
II. Number of Specular Points in a Gaussian Surface,
J. Opt. Soc. Amer. 50 (1960) 845-850.
D14. Reflection and Refraction at a Random Moving Surface, 424
III. Frequency of Twinkling in a Gaussian Surface,
J. Opt. Soc. Amer. 50 (1960) 851-856.
D15.* The Statistical Geometry of Random Surfaces, pp. 105-144
in Hydrodynamic Instability, ed. C.C. Lin, Amer. Math. Soc.,
Providence (1960).
D 16. The Distribution of Intervals Between Zeros of a Stationary 430
Random Function,
Phil. Trans. R. Soc. Lond. A 254 (1962) 557-599.
D 17. Bounding Approximations to the Distribution of Intervals 473
Between Zeros of a Stationary Gaussian Process pp. 63-88
in Time Series Analysis, ed. M. Rosenblatt, John Wiley and Sons,
New York, (1962).
D 18. The Effect of Non-Linearities on Statistical Distributions in the 499
Theory of Sea Waves,
J. Fluid Mech. 17 (1963) 459-480.
D 19. Modified Gaussian Distributions for Slightly Nonlinear Variables, 521
Radio Sci. 68D (1964) 1049-1062.
D20. On the Joint Distribution of the Periods and Amplitudes of 535
Sea Waves,
J. Geophys. Res. 80 (1975) 2688-2694.
D 21. On the Distribution of the Heights of Sea Waves: 542
Some Effects on Nonlinearity and Finite Band Width,
J. Geophys. Res. 85 (1980) 1519-1523.
D22. On the Skewness of Sea-Surface Slopes, 547
J. Phys. Oceanog. 12 (1982) 1283-1291.
xiv

D23. On the Joint Distribution of Wave Periods and Amplitudes in a 556

Random Wave Field,
Proc. R. Soc. Lond. A 389 (1983) 241-258.
D24.1- Can Optical Measurements Help in the Interpretation of -
Radar Backscatter?,
pp. 121-127 in Satellite Microwave Remote Sensing,
ed. T.D. Allan, Ellis Horwood, Chichester (1983).
D25. Statistical Properties of Wave Groups in a Random Sea State, 574
Phil Trans. R. Soc. Lond. A 312 (1984) 219-250.
D26.* Wave Group Statistics, pp. 15-35 in Oceanic Whitecaps, -
eds. E.C. Monahan and G. MacNiocaill, (D. Reidel Publ. Co.,
Dordrecht, 1985).
D27. On the Skewness of Sea-Surface Elevation, 606
J. Fluid Mech. 164 (1986) 487-498 (with M.A. Srokosz).
D28. An Effect of Sidewalls on Waves in a Wind Wave Channel, 618
J. Geophys. Res. 95 (1990) 1765.

Introductory Notes for P art F 619

F. Wave Analysis and Wave Generation 621

FI. Bounds for the Integral of a Non-Negative Function in 623

Terms of its Fourier Coefficients,
Proc. Cam. Phil. Soc. 51 (1955) 590-603.
F2. Observations of the Directional Spectrum of Sea Waves 637
Using the Motions of a Floating Buoy,
pp. 111-136 in Ocean Wave Spectra, Prentice Hall, New York, 1963
(with D.E. Cartwright and N.D. Smith).
F3.t The Directional Spectrum of Ocean Waves, and Processes of -
Wave Generation,
Proc. R. Soc. Lond. A 265 (1962) 286-315.
F4. A Nonlinear Mechanism for the Generation of Sea Waves, 663
Proc. R. Soc. Lond. A 311 (1969) 371-389.
F5. Action of a Variable Stress at the Surface of Water Waves, 682
Phys. Fluids 12 (1969) 737-740.
 XV

F6. Some Effects of Finite Steepness on the Generation of 686

Waves by Wind pp. 393-403,
in A Voyage o f Discovery (George Deacon 70th Anniversary Vol.),
Pergamon Press, Oxford (1977).
F7. Theory of Weakly Damped Stokes Waves: A New Formulation 697
and its Physical Interpretation,
J. Fluid Mech. 235 (1992) 319-324.

Introductory Notes for P art G 703

G. Radiation Stresses 705

G 1. Changes in the Form of Short Gravity Waves on Long Waves 707

and Tidal Currents,
J. Fluid Mech. 8 (1960) 565-583 (with R.W. Stewart).
G2. The Changes in Amplitude of Short Gravity Waves on 726
Steady Non-Uniform Currents,
J. Fluid Mech. 10 (1961) 529-549 (with R.W. Stewart).
G3. Radiation Stress and Mass Transport in Gravity Waves, with 747
Application to ‘Surf Beats’,
J. Fluid Mech. 13 (1962) 481-504 (with R.W. Stewart).
G4. A Note on Wave Set-Up, 771
J. Mar. Res. 21 (1963) 4-10 (with R.W. Stewart).
G5. Radiation Stresses in Water Waves; a Physical Discussion, 778
with Applications,
Deep-Sea Res, 11 (1964) 529-562 (with R.W Stewart).
G6. On the Wave-Induced Difference in Mean Sea Level Between 812
the Two Sides of a Submerged Breakwater,
J. Mar. Res. 25 (1967) 148-153.
G7. Longshore Currents Generated by Obliquely Incident Sea Waves, 1, 818
J. Geophys. Res. 75 (1970) 6778-6789.
G8. Longshore Currents Generated by Obliquely Incident Sea Waves, 2, 830
J. Geophys. Res. 75 (1970) 6790-6801.
G9.f The Average Wave Forces Acting on Wave Power Machines,
J. Soc. Underwater Tech. 2 (1976) 4 -8.
xvi

G 10. The Mean Forces Exerted by Waves on Floating or Submerged 842

Bodies with Applications to Sand Bars and Wave Power Machines,
Proc. R. Soc. Lond A 352 (1977) 463-480.
G il. The Propagation of Short Surface Waves on Longer Gravity Waves, 864
J. Fluid Mech. I l l (1987) 293-306.
G12. The Orbiting Double Pendulum: An Analogue to Interacting 878
Gravity Waves,
Proc. R. Soc. Lond. A 418 (1988) 281-299
(with K. Dysthe, F.S. Henyey and R.L. Schult).
G13. Laboratory Measurements of Modulation of Short-Wave Slopes by 897
Long Surface Waves,
J. Fluid Mech. 233 (1991) 389-404
(with S.J. Miller and O.H. Shemdin).

Volume II

Introductory Notes for P art H 913

H. Waves in Shallow W ater; Beach Processes 915

H 1. The Refraction of Sea Waves in Shallow Water, 917

J. Fluid Mech. 1 (1956) 163-176.

H2. On the Transformation of a Continuous Spectrum by Refraction, 932

Proc. Cam. Phil. Soc. 53 (1957) 226-229.
H3. Sea Waves and Beach Cusps, 936
Geogr. J. 128 (1962) 194-201 (with D.W. Parkin).

H4. On the Trapping of Wave Energy Round Islands, 944

J. Fluid Mech. 29 (1967) 781-821.
Н5.* Recent Progress in the Study of Longshore Currents, pp. 203-248 -
in Waves on Beaches, and Resulting Sediment Transport,
ed. R.E. Meyer, Academic Press, 1971.

Н6.* The Mechanics of the Surf Zone, -

Proc. 13th Int. Cong, on Theoretical and Applied Mechanics,
Moscow, Springer-Verlag (1972) pp. 213-228.

H7. On the Nonlinear Transformation of Wave Trains in Shallow Water, 985

Archiw. Hydrotech. 24 (1977) 445-457.
 xvii

H8. Oscillating Flow Over Steep Sand Ripples, 998

J. Fluid M eek 107(1981) 1-35.

H9. Wave Set-Up, Percolation and Undertow in the Surf Zone, 1033
Proc. R. Soc. Lond. A, 390 (1983) 283-291.
H10. On Wave Set-Up in Shoaling Water with a Rough Sea Bed, 1046
J. Fluid Mech. 527 (2005) 217-234.

Introductory Notes for P art I 1065

I. Nonlinear Interactions in Surface Waves 1067

I I. Resonant Interactions Between Two Trains of Gravity Waves, 1069

J. Fluid Mech. 12 (1962) 321-332.

12. Phase Velocity Effects in Tertiary Wave Interactions, 1081

J. Fluid Mech. 12 (1962) 333-336 (with O.M. Phillips).

13. An Experiment on Third-Order Resonant Wave Interactions, 1085

J. Fluid Mech. 25 (1966) 417-435 (with N.D. Smith).

14. On the Nonlinear Transfer of Energy in the Peak of a 1104

Gravity-Wave Spectrum: A Simplified Model,
Proc. R. Soc. Lond. A 347 (1975) 311-328.

Introductory Notes for P art К 1123

К. Steep, Steady Gravity Waves 1125

K l. On the Form of the Highest Progressive and Standing Waves in 1127

Deep Water,
Proc. R. Soc. Lo nd. A 331 (1973) 445-456.

K2. On the Mass, Momentum, Energy and Circulation of a 1139

Solitary Wave,
Proc. R. Soc. Lond. A 337 (1974), 1-13.

КЗ. On the Mass, Momentum, Energy and Circulation of a 1152

Solitary Wave II,
Proc. R. Soc. Lond. A 340 (1974), 471-493 (with J.D. Fenton).

K4. Integral Properties of Periodic Gravity Waves of Finite Amplitude, 1175

Proc. R. Soc. Lond A 342 (1975) 157-174.
xviii

K5. On the Speed and Profile of Steep Solitary Waves, 1193

Proc. R. Soc. Lond. A 350 (1976) 175-189
(with J.G.B. Byatt-Smith).

Кб. Theory of the Almost-Highest Wave: The Inner Solution, 1208

J. Fluid Mech. 80 (1977) 721-741 (with M.J.H. Fox).

К7. Theory of the Almost-Highest Wave. Part 2. Matching and 1229

Analytic Extension,
J. Fluid Mech. 85 (1978) 769-786 (with M.J.H. Fox).
К8. Some New Relations Between Stokes’s Coefficients in the 1247
Theory of Gravity Waves,
J. Inst. Math. Appl. 22 (1978) 261-283.
К9. The Almost-Highest Wave: A Simple Approximation, 1260
J. Fluid Mech. 94 (1979) 269-273.
КЮ* The Orbital Motion in Steep Water Waves
(Appendix to a paper by J.H. Nath),
Proc. 16th Int. Conf. on Coastal Eng. Hamburg, 1979,
pp. 874-877.

К11. Why is a Water Wave like a Grandfather Clock?, 1265

Phys. Fluids 22 (1979) 1828-1829.

К12. The Trajectories of Particles in Steep, Symmetric Gravity Waves, 1267

J . Fluid Mech. 94 (1979) 497-517.

К13. Spin and Angular Momentum in Gravity Waves, 1288

J. Fluid Mech. 97 (1980) 1-25.

К14. Trajectories of Particles at the Surface of Steep Solitary Waves, 1313

J. Fluid Mech. 110 (1981) 239-247.

К15. On Integrals and Invariants for Inviscid, Irrotational Flow 1322

Under Gravity,
J. Fluid Mech. 134 (1983) 155-159.
К16. New Integral Relations for Gravity Waves of Finite Amplitude, 1327
J. Fluid Mech. 149 (1984) 205-215.

К17. Bifurcation in Gravity Waves, 1338

J. Fluid Mech. 151 (1985) 457-475.
 xix

K18.* A New Way to Calculate Steep Gravity Waves, pp. 1-15 in

The Ocean Surface,
eds. Y. Toba and H. Mitsuyasu, D. Reidel Publishing Co.,
Dordrecht, 1985.

К 19. The Asymptotic Behaviour of the Coefficients in Stokes’s Series 1357

for Surface Gravity Waves,
IM A . J. Appl. Math. 34 (1985) 269-277.

K20. Accelerations in Steep Gravity Waves, 1366

J. Phys. Oceanogr. 15 (1985) 1570-1579.

K 21. Accelerations in Steep Gravity Waves: 1376

II. Subsurface Accelerations,
J. Phys. Oceangr. 16 (1986) 1332-1337.

K22. Eulerian and Lagrangian Aspects of Surface Waves, 1382

J. Fluid Mech. 173 (G.I. Taylor Symposium Vol., 1986)
683-706.

K23.f Eulerian and Lagrangian Wave Measurements,

Proc. Ocean Struc. Dyn. Symp. (Corvallis, Oregon, 1986)
1-32.

K24. Lagrangian Moments and Mass Transport in Stokes Waves, 1407

J. Fluid Mech. 179 (1987) 547-555.

K25. Measurements of the Vertical Acceleration in Wind Waves, 1416

J. Phys. Oceanogr. 17 (1987) 3-11
(with J.A. Ewing and M.A. Srokosz).

K26. Lagrangian Moments and Mass Transport in Stokes Waves. 1425

Part 2. Water of Finite Depth,
J. Fluid Mech. 186 (1988) 321-336.

K27. Asymptotic Theory for the Almost-Highest Solitary Wave, 1441

J. Fluid Mech. 317 (1996) 1-19 (with M.J.H. Fox).

K28. A Close One-Term Approximation to the Highest Stokes 1460

Wave on Deep Water.
Ocean Engrg. 33 (2006) 2012-2024 (with R.C.T. Rainey).
XX

Introductory Notes for P art M 1473

M. Capillary-Gravity Waves 1475

M 1. The Generation of Capillary Waves by Steep Gravity Waves, 1477

J. Fluid Mech. 16 (1963) 138-159.

М2. Limiting Forms for Capillary-Gravity Waves, 1499

J. Fluid M eek 194 (1988) 351-375.

М3. Capillary-Gravity Waves of Solitary Type on Deep Water, 1524

J. Fluid M eek 200 (1989) 451-470.

M4. Capillary Rollers and Bores, 1544

J. Fluid M eek 240 (1992) 659-679.

M5. Capillary-Gravity Waves of Solitary Type and Envelope 1565

Solitons on Deep Water,
J. Fluid Meek 252 (1993) 703-711.

M6. Parasitic Capillary Waves: A Direct Calculation, 1574

J. Fluid Mech. 301 (1995) 79-107.

M7. Surface Manifestations of Turbulent Flow, 1603

J. Fluid Mech. 308 (1996) 15-29.

M8. Capillary Jumps on Deep Water, 1618

J. Phys. Oceanogr. 26 (1996) 1957-1965.

M9. Experiments on Capillary-Gravity Waves of Solitary Type 1627

on Deep Water,
Phys. Fluids 9 (1997) 1963-1968 (with X. Zhang).

M10. Viscous Dissipation in Steep Capillary-Gravity Waves, 1633

J. Fluid Mech. 344 (1997) 271-289.

M il.1- Solitary Waves on Deep Water, pp. 149-167 in

Wind-Over-Wave Couplings. Perspectives and Prospects,
eds. S.G. Sajjadi, N.H. Thomas and J.C.R. Hunt,
Oxford, Clarendon Press, 356 pp. (1999).
 xxi

Volume III

Introductory Notes for P art L 1653

L. Unsteady Free-Surface Flows; Wave Breaking 1657

LI. On Wave Breaking and the Equilibrium Spectrum of 1659

Wind-Generated Waves,
Proc. R. Soc. Lond. A 310 (1969) 151-159.

L2. A Class of Exact, Time-Dependent, Free-Surface Flows, 1668

J. Fluid Mech. 55 (1972) 529-543.

L3. Periodicity in Whitecaps, 1685

Nature, Lond. 239 (1972) 449-451
(with M. Donelan and J.S. Turner).

L4. A Model of Flow Separation at a Free Surface, 1690

J. Fluid Mech. 57 (1973) 129-148.

L5. Review of: ‘Breaking Waves’, (film) by G.B. Olsen and 1710
S.P. Kjeldsen,
J. Fluid Mech 57 (1973) 624.

L6. An ‘Entraining Plume’ Model of a Spilling Breaker, 1711

J. Fluid Mech. 63 (1974) 1-20 (with J.S. Turner).

L7. Breaking Waves — In Deep and Shallow Water, 1731

Proc. I Oth Symp. on Naval Hydrodynamics,
(Cambridge, Mass. 1974), pp. 597-605.

L8.t Recent Developments in the Study of Breaking Waves,

Proc. 15th Conf. on Coastal Eng. (Honolulu, 1976) pp. 441-460.

L9. Self-Similar, Time-Dependent Flows with a Free Surface, 1740

J. Fluid Mech. 73 (1976) 603-620.

L10. The Deformation of Steep Surface Waves on Water. 1758

I. A Numerical Method of Computation,
Proc. R. Soc. Lond. A 350 (1976) 1-26 (with E.D. Cokelet).

LI l.f A Calculation of Unsteady Surface Waves,

Proc. N th JUTAM Cong. (Delft, Holland, 1976), pp. 423-424
(with E.D. Cokelet).
xxii

L12.1" Advances in the Calculation of Steep Surface Waves and

Plunging Breakers,
Proc. 2nd Int. Conf. on Numerical Ship Hydrodynamics,
Berkeley, 1976 pp. 332-346.

L 13? The Calculation of Steep Gravity Waves,

Proc. 2nd Conf. on Behaviour o f Offshore Structures,
Trondhein, Norway (1976) 2, pp. 27-39
(with E.D. Cokelet and M.J. Fox).

LI 4. The Instabilities of Gravity Waves of Finite Amplitude in 1784

Deep Water. I. Superharmonics,
Proc. R. Soc. Lond. A 360 (1978) 471-488.

L 15. The Instabilities of Gravity Waves of Finite Amplitude in 1802

deep water. II. Subharmonics
Proc. R. Soc. Lond A 360 (1978) 489-505.

LI 6. The Deformation of Steep Surface Waves on Water. 1819

II. Growth of Normal-Mode Instabilities,
Proc. R. Soc. Lond A 364 (1978) 1-28 (with E.D. Cokelet).

LI 7^ On the Dynamics of Steep Gravity Waves in Deep Water,

pp. 199-220 in Turbulent Fluxes Through the Sea Surface,
Wave Dynamics, and Prediction,
eds. A. Favre and K. Hasselman, Plenum Publ., 1978.

L 18. A Technique for Time-Dependent Free-Surface Flows, 1847

Proc. R. Soc. Lond. A 371 (1980) 441-451.

LI 9. On the Forming of Sharp Comers at a Free Surface, 1858

Proc. R. Soc. Lond A 371 (1980) 453-478.

L20. Modulation of the Amplitude of Steep Wind Waves, 1884

J. Fluid Mech. 99 (1980) 705-713.

L21 The Unsolved Problem of Breaking Waves, (Keynote Address),

Proc. 17th Int. Conf on Coastal Engineering
(Sydney, Australia, 1980) pp. 1-28.

L22. On the Overturning of Gravity Waves, 1893

Proc. R. Soc. Lond A 376 (1981) 377-400.
 xxiii

L23.* A Parametric Flow for Breaking Waves,

Proc. Int. Symp. on Hydrodynmics in Ocean Engineering
(Trondheim, 1981), pp. 121-135.

L24. Parametric Solutions for Breaking Waves, 1917

J. Fluid Mech. 121 (1981) 403-424.

L25. Bubbles, Breaking Waves and Hyperbolic Jets at a Free Surface, 1939
J. Fluid Mech. 127(1983) 103-121.

L26. Rotating Hyperbolic Flow: Particle Trajectories and 1958

Parametric Representation,
J. Mech. Appl. Math. 36 (1983) 247-270.

L27. Measurements of Breaking Waves by a Surface Jump Meter, 1982

J. Geophys. Res. 88 (1983) 9823-9831 (with N.D. Smith).

L28.1" Towards the Analytic Description of Overturning Waves, pp. 1-24

in Nonlinear Waves, ed. L. Debnath, Cambridge Univ. Press, 1983.

L29.t Surface Wave Interactions, Keynote Address, pp. 29-34

in Proc. 9th Australasian Fluid Mech. Conf.
(Auckland, New Zealand), 1986,29-34.

L30. On the Stability of Steep Gravity Waves, 1991

Proc. R. Soc. Lond. A 396 (1984) 269-280.

L31. Bifurcation and Instability in Gravity Waves, 2003

Proc. R. Soc. Lond. A 403 (1986) 167-187.

L32. A Stochastic Model for Sea Surface Roughness I. Wave Crests, 2024
Proc. R. Soc. Lond A 410 (1987) 19-33.

L33.* Advances in Breaking-Wave Dynamics, pp. 209-230

in Wave Dynamics and Radio Probing o f the Ocean Surface,
eds. O.M. Phillips and K. Hasselmann, Plenum Publ.,
New York, 1987.

L34.* Measurements of Breaking Waves,

in Wave Dynamics and Radio Probing o f the Ocean Surface,
eds. O.M. Phillips and K. Hasselmann, Plenum Publ.,
New York, 1987, 257-264 (with N.D. Smith).
xxiv

L35.* Mechanisms of Wave Breaking in Deep Water, pp. 1-30

in Sea Surface Sound, ed. B. R. Kerman, D. Reidel Publ. Co.,
Dordrecht, 1988.

L36.* The Dynamics of Bragg-Scattering Waves on the Sea Surface,

pp. 57-83 in Mathematics in Remote Sensing,
ed. S.R. Brooks, Oxford, Clarendon Press (1989)

L37. Flow Separation Near the Crests of Short Gravity Waves, 2039
J. Phys. Oceanogr. 20 (1990) 595-599.

L38. A Stochastic Model of Sea-Surface Roughness. II, 2044

Proc. R. Soc. Lond. A 435 (1991) 405-422.

L39. Highly-Accelerated, Free-Surface Flows, 2062

J. Fluid Mech. 248 (1993) 449-475.

L40. Crest Instabilities of Gravity Waves. 2089

Part 1. The Almost-Highest Wave,
J. Fluid Mech. 258 (1994) 115-129 (with R.P. Cleaver).

L41. Crest Instabilities of Gravity Waves. 2104

Part 2. Matching and Asymptotic Analysis,
J. Fluid Mech. 259 (1994) 333-344
(with R.P. Cleaver and M.J.H. Fox).

L42. A Fractal Approach to Breaking Waves, 2116

J. Phys. Oceanogr. 24 (1994) 1834-1838.

L43.1" The Initiation of Spilling Breakers, Keynote Address,

Proc. Int. Symp. on Waves — Physical and Numerical Modelling,
University of British Columbia, Vancouver, B.C.,
21-24 August 1994, pp. 24-48.

L44. On the Disintegration of the Jet in a Plunging Breaker, 2121

J. Phys. Oceanogr. 25 (1995) 2458-2462.

L45. The Crest Instability of Steep Gravity Waves or How Do Short 2126
Waves Break?, pp. 237-246 in The Air Sea Interface,
eds. M.A. Donelan, W.H. Hui and W.J. Plant, Toronto, Canada,
Univ. of Toronto Press (1996) 789 pp.
 XXV

L46. Crest Instabilities of Gravity Waves. Part 3. Nonlinear 2136

Development and Breaking,
J. Fluid МесИ.ЪЪЪ (1997) 33-50 (with D.G. Dommermuth).

L47. On the Crest Instabilities of Steep Surface Waves, 2154

J. Fluid Mech. 336 (1997) 51-68 (with M. Tanaka).

L48.1" Progress Toward Understanding How Waves Break,

Proc. 21st Symp. on Naval Hydrodynamics, Trondheim,
Norway, 24-28 June 1996, Washington, D.C.,
Nat. Acad. Sci. Press (1997), pp. 7-28.

L49. Shear Instability in Spilling Breakers, 2172

Proc. R. Soc. Lond. A 446 (1994) 399-409.

L50. Instabilities of a Horizontal Shear Flow with a Free Surface, 2183

J. Fluid Mech. 364 (1998) 147-162.

Introductory Notes for P art N 2199

N. Standing Waves 2201

N1. Theory of Water Waves Derived from a Lagrangian. 2203

Part 1. Standing Waves,
J. Fluid Mech. 423 (2000) 275-291.

N2. Vertical Jets from Standing Waves, 2220

Proc. R. Soc. Lond A 457 (2001) 495-510.

N3. On the Breaking of Standing Waves by Falling Jets, 2236

Phys. Fluids 13 (2001) 1652-1659 (with D.G. Dommermuth).

N4. Vertical Jets from Standing Waves: The Bazooka Effect, 2244
pp. 195-204 in IUTAM Symp. on Free Surface Flows,
ed. Y.D. Shikhmurzaev, Kluwer Acad. Publ. (2001).

N5. Vertical Jets from Standing Waves. II, 2253

Proc. R. Soc. Lond. A 457 (2001) 2137-2149
(with D.G. Dommermuth).

N6. Asymptotic Forms for Jets from Standing Waves, 2266

J. Fluid Mech. 447 (2001) 287-297.
xxvi

N7. On Steep Gravity Waves Meeting a Vertical Wall: 2277

A Triple Instability,
J. Fluid Mech. 466 (2002) 305-318 (with D.A. Drazen).

N8.* Standing Waves in the Ocean, pp. 201-218 in Wind over Waves II:
Forecasting and Fundamentals o f Applications,
eds. S.G. Sajjadi and J.C.R. Hunt, Horwood Publ.,
Chichester, U.K., (2003) 232 pp.
 1653

In tro d u c to ry N otes fo r P a r t L
L . U nsteady F ree-S u rface Flow s; W ave B reaking

Papers L I to L50

This, the longest group of papers, represents a multi-faceted attack on the

mathematical description of breaking waves, and of their dynamics. The author’s
interest in the subject was rekindled after attending a conference in Vancouver,
BC in 1968.
Paper LI points out that the dimensionless constant a in OM Phillip’s (1966)
expression for the frequency spectrum of breaking waves is not an absolute
constant, but depends upon the dimensionless fetch. The paper provides simple
estimates of the energy lost in wave breaking, and of the energy supplied by
the wind.
Papers L3 to L7 and paper L27 involve field observations of breaking waves
and laboratory wave experiments. Paper L3 explains why whitecaps in deep
water (as seen for example from the air or from the deck of a ship) are observed
to appear and disappear periodically, with a period equal to twice that of
the waves.
Paper L4 proposes a simplified model of the turbulent flow at the crest
of a spilling breaker. It is used in Paper L37 to explain a certain type of
flow instability observed by Koga (1982) at the crest of a short gravity wave.
Paper L6, written with J.S. Turner, proposes a model for the spreading of
turbulence down the forward face of a spilling breaker. Use is made of some of
the laboratory observations in the film by Olsen and Kjeldsen which is reviewed
in L5. Paper L7 describes experiments with breaking waves, and proposes a new
method of creating a breaking wave in the laboratory. Wave energy is focused
at a certain point in a wave channel by decelerating the wavemaker linearly in
time. The method has since been used to great advantage by W.K. Melville and
his students.
Paper L27 describes a “jump-meter” designed to measure wave breaking at
very small length-scales and time-scales, which are thought to have a special
bearing on the mean horizontal wind-stress and on wave generation by wind.
A strikingly realistic flow, which resembles the overturning plume in
a plunging breaker, is described in Papers L23 and L24. Here z is a cubic
polynomial in the parameter со (see Section 12 of L24). As shown in Figures 9,
10, and 11, there is good agreement both with laboratory experiments and with
numerical, time-stepping computations (see below).
1654

Paper L2 is the first in the sequence of papers suggesting mathematical

models of waves as they approach breaking. The series continues with L9,
L I8, L I9, L22, to L26, L28, and N6. Paper L2 itself gives a new derivation of
an exact flow in which the free surface takes the form of a time-varying ellipse.
The analysis in L2 is extended to flows with a hyperbolic free surface. A special,
limiting case is demonstrated experimentally. Paper L9 generalises the solution
given in Paper L2 by constructing flows in two or three dimensions, in which the
length-scale varies as some power of the time t. The analysis involves the use
of a complex variable су as a parameter, so that the velocity-potential X = Ф + \\fj
and the space-coordinate z = x + \y are each analytic functions of со and of the
time /. The two boundary conditions at the free surface are met by requiring that
both the pressure p and its rate of change Dp/Dt vanish on the same surface. The
general method is formulated in Paper L I8, where some examples are given. In
L I9 special attention is given to a solution representing an overturning flow, as
in the jet of a plunging breaker, namely the rotating, time-dependent hyperbola.
Paper L26 studies this flow from a Langangian viewpoint. Lastly in this series,
Paper L39 suggests an approximate parametric solution for a highly accelerated
flow such as sometimes occurs when a wave impacts a vertical wall.
Papers L I4 and L I5 represent an important advance in water-wave theory,
by showing that regular trains of gravity waves, of finite amplitude are unstable
to a number of different harmonic perturbations, besides those of the Benjamin-
Feir type. The most important ranges of the instabilities, and their initial rates of
growth, were calculated, for two-dimensional waves. For a further discussion, see
Papers L30 and L31. The consequences for the “grouping” of ocean waves are
discussed in Paper L20.
Papers L10 and L I7, in collaboration with E.D. Cokelet, introduced a new
numerical time-stepping method for following accurately the development of
waves of finite amplitude. Paper L10 follows a steep, progressive wave after
it had been given enough extra energy to make it turn over and break. This
contribution was the first convincing model of a plunging breaker. Paper L I7
built on the results of Paper L I5 by following some subharmonic instabilities by
numerical time-stepping, to the point in time where overturning of the surface
occurs. This revealed the typical shape of a plunging breaker. Time-stepping
calculations of this kind have been taken up and developed by many others; see
for example the paper by Vinje and Brevig (1981), in Papers L23 and L24.
In most theories of wave generation by wind the sea surface is regarded as a
succession of waves of uniform wave height or amplitude. Papers L32 and L38
develop a model of the sea surface in which short, wind-generated waves ride
on the surface of longer waves which are not of uniform amplitude but instead
have a Rayleigh distribution, as is normally observed in the ocean. Paper L32
 1655

calculates the distribution of the “roughness” (i.e. the short wave steepness) at
the crests of the longer waves. Paper L38 calculates the distribution of roughness
over the whole sea surface.
The profile of a steep Stokes wave close to (but not actually at) its limiting
steepness, was examined earlier in Papers K5 and K6 on the almost-highest
wave. Now, in Papers L40, L41, and L45 to L47 are examined the instabilities
of these very steep waves. In Paper L40 it is shown that the principle (linear)
instability takes the form of a bulge on the forward face of the wave. In L41
it is shown that this instability is analytically continuous with the instability
of the complete wave in deep water and that it becomes unstable at the first
energy maximum of the whole wave, as calculated by Tanaka. A similar result
is predicted for the second energy maximum. In Paper L47 the form of the
instabilities is calculated with great accuracy, and they are shown to obey a
scaling law predicted by the analysis. Similar calculations are performed for
the crest instabilities of solitary waves. In Paper L46 the growth of the lowest
instability is followed numerically to its nonlinear stage, and it is shown to lead
to overturning of the wave crest, as in a plunging breaker. Interestingly it is found
that if the sign of the initial perturbation is reversed, the nonlinear stage is quite
different; the perturbation leads at first to a transition of the wave to an almost
steady wave of lower amplitude, followed by a long-term recurrence to the
original wave, and so on. Paper L48 contains a useful summary and discussion of
the preceding work, and adds new results.
Two papers in this series, both on the instability of a shearing flow, are
closely related to breaking waves. Paper L49 concerns the instability of the
shearing flows seen on the forwards face of the crest of a gravity wave (Type 2
instability). This instability quickly grows and then collapses into turbulence. A
simplified model (Section 4) shows that the instability may be considered as a
train of “vortex waves”, i.e. a horizontal array of vortices. A second Paper A50
considers a more general model, in which a surface layer travelling with uniform
velocity overlies a deeper layer with a smooth velocity profile, tending to zero at
comparative large depths. There is rough agreement before the horizontal scales
of the instabilities and those observed.
O f the remaining papers in this section, Paper L42 describes some paradoxical
results resulting from consideration of the water surface as a fractal, and Paper
L44 discusses how the sheet of water in a plunging jet may become unstable to
three-dimensional perturbations, as is often observed.
L. Unsteady Free-Surface Flows; Wave Breaking
 1659

Proc. Roy. Soc. A. 310, 161-169 (1969)

Printed in Great Britain

On wave breaking and the equilibrium spectrum

of wind-generated waves

By M. S. L o n g u e t -H ig o in s , F.R.S.
Oregon State University, Corvallis, Oregon

(.Received 3 September 1968)

A theoretical calculation ia made of the loss of energy by wave breaking in a random sea state
in terms of the spectral density function. In the special case of the equilibrium spectrum
F(cr) = адха~ь the proportion ro of energy lost per mean wave cycle is found to be given by
m = e-1/Sa
irrespective of the low-frequency cut-off in the spectrum.
Assuming that in the equilibrium state the loss of energy by breaking is comparable to
that supplied by the wind, one can estimate the constant a in terms of the drag coefficient
of the wind on the sea surface. It is found that
a = -Ц Щ Ш О С Ц р ^/р ^)].
Taking a representative value of G one finds a = 1*3 x 10“*, which falls within the range of
observed values of a. The above equation for a is rather insensitive to the various assump­
tions made in the analysis.
There is some evidence, derived from observation, that a may not in faot be quite con­
stant, but may decrease slightly as the wave age (gtjU) or the non-dimensional fetch (gxjU*)
is increased. I t is suggested that the drag coefficient may behave similarly.

I n t r o d u c t io n

The ‘equilibrium law ’ for wind-generated surface waves, as first derived by Phillips
( 1958 ) is that over a certain range of wave frequencies cr the spectral density of the
surface elevation £ is given by

F(cr) = ад2(т~& (crx < (T < <r2), ( 1 -1)

where g denotes the acceleration of gravity and a is an absolute constant. The fre-
quency-dependence of the spectrum ( 1-1) has been approximately verified by ob­
servations over a wide range of frequencies (Phillips 1966). Some experimentally
determined values of a, as quoted by Phillips, are shown in the last column of table 1.
Although there appear to be significant differences among the entries in the last
column, there is at least an order-of-magnitude agreement among them. However,
no independent theoretical estimate of a has yet been given.
Now the basic idea underlying Phillips’s derivation of the equilibrium law is that
over this range of frequencies the spectrum is limited by wave breaking. It seems
natural then to inquire how much energy is indeed lost by wave breaking in a
spectrum of the form ( 1*1), and to relate this to the energy input by the wind, which
can be estimated independently. This is the purpose of the present note.

10 [ 151 ] Vol. 310. A. (6 May 1969)

152 M. S. Longuet-Higgins
The estimate of energy loss must of necessity be somewhat rough, since we are
dealing with a nonlinear phenomenon. Fortunately, however, it turns out that the
appropriate value of a is fairly insensitive to the various assumptions that will be
made.
T a b l e 1. V a l u e s of a a s d e t e r m in e d b y m e a s u r e m e n t s o f

THE SPECTRUM OF SURFACE WAVES

* u
(m) (m/s) gx/U* CL

Burling ( 1959) 70 3-9 180 1*48 x 10-*

Hicks (1960) 55 1-2 380 1-21 x 10 ~*
Kinsman (i 960) 2500 5*6 800 1-04 x 10-*
Pierson (ed.) ( i 960) 3 x10s 9-5 3300 1-33x10-*
Longuet-Higgins, 5 x 10s 100 5 x 10* 0-80 x 10-*
Cartwright & Smith
( 1963)

The calculation is rendered remarkably straightforward by the demonstration,

given in §2 , that the proportional loss of energy per mean wave cycle in the spectrum
( 1*1) is approximately given by
w = e-1/8e (1 -2 )

independently of the frequency range a 1 < <x < or2, provided cr1 < cr2. On the other
hand, if we assume that the energy lost in breaking is comparable with that supplied
by the wind, we are led to the conclusion (see § 2 ) that, to an order of magnitude

® ~ 1 6 0 0 C W /W ). (1-3)
where G is the ‘drag coefficient’ of the wind on the waves. On assuming the value
С = 1*6 x 10-s and equating ( 1*2 ) and (1-3) we find

a = 1*3 x 10~2, (1-4)

a value within the range of observation and moreover extremely insensitive to the
rough assumptions made in the analysis.
The question is raised in §4 whether a is an absolute constant or whether, on the
other hand, it varies throughout the development of a wave spectrum. There is
some evidence (derived from table 1) to suggest that a may in fact be a decreasing
function of the ‘wave age’ as defined by Sverdrup & Munk ( 1947).

2. T h e l o s s o f e n e r g y b y w a v e b r e a k in g

A rough estimate of the energy lost to the breaking waves may be made in the
following way. It is well known that in a standing wave, the maximum downward
acceleration is equal to g (see Taylor 1953 ). It is less well known that in a progressive
wave, whose limiting form at the crest is the Stokes 120 ° angle, the acceleration is
equal to directed away from the crest in all directions (Longuet-Higgins 1963).
 1661

Wave breaking and the equilibrium spectrum 163

This is illustrated in figure 1. In a typical sea state, where the energy is directed
mainly within about 30° of the mean direction, it is reasonable to suppose that white-
caps appear whenever the vertical acceleration at the crest approaches - \g.

® i ь
F ig u b e 1 . The limiting acceleration near the crest in a Stokes 120 ° angle.

I f for a moment we assume that the wave is adequately described by the linear
expression £ = a co b (kx —art) (cr2 = gk), (2 *1)

it follows that the condition for breaking is simply

a<r2 = (2 -2 )

It might be thought unacceptable to use the linearized theory of small-amplitude

waves for an obviously nonlinear effect. But we note that on the same basis the maxi­
mum wave slope would be given by

MOmax. = = 0-5 (2-3)

compared to the actual value
arc tan \ tt = 1/^/3 = 0*547 (2-4)

and that the ratio of the wave height to the wavelength would be given by

2a ak acr2 1
-p = — = ---- = ^ - = 0*159 ..., 2-5
L it ng 2л

compared to the computed value 0*142...(Michell 1893). Hence in adopting the

criterion of equation (2 *2 ) we may be in error by some 10 %.
Now in a sea state a narrow frequency spectrum it has been shown (Longuet-
Higgins 1952 ) that the distribution of wave amplitudes a is theoretically a Rayleigh
distribution: „
P{a) = = i exp { - a2/o2} (2-6)
CL
and, moreover, this distribution has been found to fit the observed distribution of
wave heights remarkably well even when the spectrum is not narrow (Longuet-
Higgins 1952 ; Watters 1953 >Cartwright & Longuet-Higgins 1956 ). This distribution
is shown in figure 2 . It has a maximum density at a/a = 1/^ 2 , and it decreases ex­
ponentially as afa ->• 00. The parameter a represents the root-mean-square wave
amplitude, which for a narrow spectrum is related to the mean energy density E of
1662

154 M. S. Longuet-H iggins

the wave field per unit horizontal area by
E = ipga2. (2-7)

Now let us suppose that all those waves of amplitude greater than the critical
amplitude a0 will break, where
O'0v 2 = i9 (2*8)

and or is a mean frequency which may be determined from the spectral density
* V )b y Г оо Лео
cr2J l^((7)dcr = J (t2F((t) dcr. (2-9)

FiatraE 2. The probability density p(a) of the wave am plitude a, given by

p(a) = 2(a/a*) exp {—a 8/a “}

(the Rayleigh distribution). Here a denotes the r.m.s. wave am plitude and a 0 th e critical
am plitude for breaking waves.

I f determined in this way, a has the property that the curve £ = a cos a t has the
same mean number of zero-crossings per unit time as the record with spectrum
F{cr) (see Rice 1944 ). Let us further assume that if the wave amplitude a exceeds
a0, then wave breaking reduces the amplitude to a 0 exactly, so that the amount o f
energy lost is \рд(аг —a§). With these assumptions it follows that the mean loss of
energy density in one wave cycle T = Ял/сг is given by

f iMia? ~ « 0) Via) da = - f \pg{a- a§) d(exp {- a2/a2})

J at J a,

= f
J а,
exp { —a21а2} А(\рдаг)

( 2 - 10 )

where =\P9< (2-11)

 1663

Wave breaking and the equilibrium spectrum 156

In other words, the proportion m of wave energy lost per mean wave cycle is simply
w = e x ^ {-E J E } . ( 2 *12)

B ut E = pg£2 = pgj^ F{(r)d<r. (2*13)

So from (2*8) and (2 -11) this can also be written

m = exp { i ^ / / ' f'(o')cLo')- (2 -u )

To make further progress we need to make a specific assumption as to the form of
the spectral density F(<r). Let us assume first that F{cr) is the equilibrium spectrum
with sharp low-frequency cut-off:
fO (cr < a \ ) , l
U .; ; : ) <2-i5 >
In this case we find
J > < r ) d<r = g (2-16)

Г a^2o" 3 dcr
and ^ = 7 ^------------- = 2 o f . (2-17)
ocg2a~5dcr
J <Ti
Hence w = ex p {—l / 8a}, (2-18)
a result independent of <гх.
If, on the other hand, F(<r) has a smooth cut-off as suggested by Neumann & Pier-
son ( 1966): F(cr) = ag»<r-‘ exp{-fi(<Tla1)i}, (2-19)
where J3 is an absolute constant (== 0-74) and o’1 = g/U, U being a representative
wind-speed, we find then r<X3 2
j. * № -■ & > (2'20)

f a" 3 exp { —fiio-Jcr)4} dcr

and o=2= - ^ ---------------------------------- (2 -21 )
I cr- 6exp { - /?(cTj/cr)4} do*
Hence w = exp{-l/27ra}, (2 -22 )
independently of both p and crv
The two results (2-18) and (2 -22 ) are quite similar. For the purpose of discussion
we shall adopt equation (2*18).

3. An e s t im a t e o f а

Some of the energy lost to a wave through breaking will doubtless be transferred
to other parts of the gravity wave spectrum. A small but significant part may also
be transferred to capillary waves (Longuet-Higgins 1963 ). However, the visual
1664

156 M. S. L onguet-H iggins

observations of breaking waves in a laboratory flume certainly suggests that the
greater part of the wave energy lost through breaking normally goes directly into
turbulence.
Because of these turbulent energy losses it follows that the energy involved in
wave breaking must be comparable with, though in equilibrium less than, the total
energy supplied to the sea surface by the wind. Thus if W denotes the mean rate of
working by the wind, w E jT ^ w> (3 -i)

where E denotes the total wave energy density per unit area and T the mean period
as defined in § 2 .
On the other hand since the equilibrium spectrum is limited by wave breaking,
if we assume that the wind puts energy mainly into the longer wave components
that are about to be limited by breaking, it is arguable that the loss of energy through
breaking should be comparable with the rate of growth of the total wave energy:
даE /T ~ dEjdt. (3-2)
Of these two inferences, (3*1) appears the better founded. But if both inferences
are correct, then it follows that
W ~ dE /dt (3-3)
as suggested on other grounds by R. W. Stewart ( 1961). Of course it is always true
that the left-hand side of (3-3) is greater than the right.
To estimate the rate of working of the wind we may adopt more than one method.
Perhaps the simplest is to assume that
W ~ ти+у (3*4)
where r denotes the wind-stress and u* the friction-velocity, defined by
T= Pair *4 ■ (3-6)
Since it is known empirically that
7 —Cpa.lT U2, (3*6)
where U denotes the mean wind-speed at a height 10 m above the sea surface, and
С ~ 1*6 x 10-3, it follows that, on this basis, = CXJ2 and so
W~ Pair < = C}Palr UK (3-7)
Alternatively we may adopt the suggestion of Stewart & Grant ( 1962 ) and as­
sume that, since energy and momentum are transferred to a wave in the ratio of the
phase velocity, then W = TCl = Gpalr U \ (3 -8 )

where cxdenotes the phase velocity of the waves that are receiving most of the energy.
The two estimates are the same if it is supposed that
cx ~ C iU ~ 0*04£7. (3*9)
Without making this assumption we shall simply adopt equation (3*7), which is
almost certainly correct to within an order of magnitude.
 1665

Wave breaking and the equilibrium spectrum 167

Now w y by equation (3*1), is given by
w ~ TW jE . (3 *10)
We need then to estimate typical values of T and E. Referring to the observations
of Sverdrup & Munk ( 1947 ) we take the phase speed с of the dominant waves to be
equal to U nearly, so that 0 0 TT
T = (M l)
9 Я '
and we take the ‘significant wave height* H to be equal to 0*25U2jg approximately.
Since for a Rayleigh distribution H = 2-83a (Longuet-Higgins 1952 ) we have
a ~ 0-088172/<7 and so E= _ 3.9 x j 0-»pU4g. (3-12)

On substituting into equation (3-10) from equations (3-11), (3-12) and (3-7) we find
that the terms in U and g cancel each other identically, giving

Ш~ 1600CipaIr/p„ter (3-13)
or, with the values С = 1*5 x lO-3 and p ^ lp^ ter 1*3 x 10-3,
1-0x10-* (314)
Having now estimated the numerical value of w we are in a positionto find a from
equation (2*18): .
“ * - 8 Е ^ ф 1 ’35* 10' г- (316)
We notice that this value of a falls within the range of the experimentally deter­
mined values of a as shown in table 1. However, we may remark that if any of the
relations between W, E, T and U were to be altered by a factor of 2 this would change
the value of In m by only one part in 12 . Hence the corresponding value of a given
by (3*13) would only vary by this amount and so would still be within the range of
observation.
However, the replacement of equation (2*18) by equation (2 -22 ) would put a
just beyond the upper limit of the observed range.

4. Is CL B E A L L Y CONSTANT?

For the appropriate values of E and T we selected, in the above analysis, the
values given by Sverdrup & Munk (1947 ) for large values of the nondimensional
fetch (gxl U2), say {gxj U2) 104. In a sea state which is still growing under the action
of the wind, smaller values of (gxfU2) would be appropriate and hence smaller
values of both E and T. The variation of T under different wind conditions is gener­
ally less marked than the variation of E. Hence w, which is proportional to TIE,
might be expected to decrease with (gx/U)23and so a, by (3*15), would also decrease
with (gx/U2).
This inference gains some support from the data in table 1 . In the second and
third columns are shown the logarithmic-mean values of the fetch x and the wind
1666

158 M. S. L onguet-H iggins

speed V for each of the observations quoted, and in the fourth column the logarith­
mic-mean value of (gx/U2). The values of a in the last column do indeed decrease
as the parameter {gxjU2) increases, with the exception of the value attributed to
Pierson ( i 960 ). This was derived from measurements with a vertical wave pole, the
calibration of which was uncertain (see Chase et al. 1957 ). Apart from this observa­
tion all the values of a decrease monotonically with (gx/U2).
That a should be a decreasing function of the nondimensional fetch or of the ‘ wave
age 5 (gt/U) is in accord with the common observation that when a breeze suddenly
starts to blow white caps appear initially to be more numerous, indicating a tem ­
porarily increased value of a.
The equation whether the wind-stress coefficient С is also a function of the time
would be worth investigation. Some of the scatter in the experimental determina­
tions of С (see Phillips 1966 , figure 4*18) may perhaps be explained in this way.
Some further light is thrown on these conclusions by some recent observations of
fetch-limited waves by Barnett & Wilkerson ( 1967 ). The observations were made
in a practically uniform wind-field U = 16 m/s at distances from the shoreline rang­
ing from 50 to 300 km, so that
2000 < gx/U2 < 12000 .

The most striking observational conclusion was that as the fetch x increased, so
the peak energy density in the spectrum (which occurs at the low-frequency end of
the equilibrium range) first rose to a value 1-5 to 3 times greater than its final,
equilibrium value. The authors suggest, in fact an equilibrium spectrum of the form
F{or) = ад2ог~6/((т1сг0)} (4*1)

where cr0is the radian frequency of the maximum spectral density. This frequency
diminishes as the fetch is increased. Clearly, if an attempt were made to make a best
fit of the equilibrium law ( 1 -1) to a typical spectrum, the constant a would have to
be adjusted differently according to the fetch, the highest values necessarily occur­
ring at the shortest fetches. This is apparently consistent with our conclusions, f
Nevertheless an examination of the original spectra summarized in table 1 shows
that many of them do not exhibit such a spectral peak as found by Barnett &
Wilkerson, even at quite small values of (gx/ll2). We therefore prefer not yet to
adopt any particular form of/(cr/cr0) apart from that implied by equation ( 1 -1).
There is no reason in principal why the calculation of m in terms of F(cr) given in
§ 2 should not be adapted to a function/(cr/<r0) of any reasonable form.

This paper is partly based on a contribution to the Symposium on Turbulence

in the Ocean, held at the University of British Columbia, Vancouver, B.C. on
11-14 June 1968. It was completed at the Woods Hole Oceanographic Institution
during August 1968. The research has been supported under NSF Grant GA 1452
t I f the peak values of the spectral density are neglected, B arnett & Wilkerson find a
rather low value of a, namely 0-6 x 10~a.
 1667

Wave breaking and the equilibrium spectrum 159

and ONR Contract 241-11. The author is indebted to Dr N. P. Pofonoff, Dr C. S.
Cox and to other colleagues at the Woods Hole Summer School for stimulating
discussions.

R eferences

B arn ett, T. P. & Wilkerson, J . C. 1967 On the generation of ocean wind waves as inferred
from airborne measurements of fetch-limited spectra. J . Mar. Res. 25, 292-321.
Burling, R. W. 1959 The spectrum of waves a t short fetches. Dtsch. Hydr. Z. 12, 45-64,
96-117.
Cartwright, D. E. & Longuet-Higgins, M. S. 1956 On the distribution of the m axima of a
random function. Proc. Roy. Soc. A 237, 212-232.
Chase, J . et al. 1957 The directional spectrum of a wind generated sea as determined from
d ata obtained by the Stereo W ave Observation Project. New York Univ. CoU. Eng.
Rep. Ju ly 1967.
Hicks, B. L. i 960 The energy spectrum of small wind waves. Univ. Illinois, C.S.L. Rep.
M-91.
Kinsm an, B. i 960 Surface waves a t short fetches and low wind-speed—a field study.
Chesapeake Bay Inst. Rep. no. 19.
Lam b, H . 1932 Hydrodyanmics (6th ed.). Cambridge University Press.
Longuet-Higgins, M. S. 1952 On the statistical distribution of the heights of sea waves.
J . M ar. Res. 11, 245-266.
Longuet-Higgins, M. S. 1963 The generation of capillary waves by steep gravity waves.
J . FI. Mech. 16, 138-159.
Longuet-Higgins, M. S., Cartwright, D. E. & Smith, N. D. 1963 Observations of the direc­
tional spectrum of sea waves using the motions of a floating buoy. Ocean wave spectra,
pp. 111-136, Englewood Cliffs, N .Y .: Prentice-Hall, Inc.
Michell, J . H . 1893 The highest waves in water. Phil. Mag. 36, 430-437.
Neumann, F. & Pierson, W. J. 1966 Principles of Physical Oceanography. Englewood Cliffs,
N. J ., Prentice-Hall, 545 pp.
Phillips, О. M. 1958 The equilibrium range in the spectrum of wind-generated waves. J .
F I. Mech. 4, 426-434.
Phillips, О. M. 1966 The dynamics of the upper ocean. Cambridge University Press.
Pierson, W. J . (ed.) i 960 The directional spectrum of a wind-generated sea as determined
from d ata obtained by the Stereo Wave Observation Project. NewYork Univ. CoU. Eng.,
Met. Pap. 2, no. 6.
Rice, S. O. 1944 The m athem atical analysis of random noise. Bell Syst. Tech. J . 23, 282-
332.
Stew art, R. W. 1961 . The wave drag of wind over water. J. FI. Mech. 10, 189-194.
Stew art, R. W. & Grant, H. L. 1962 D etermination of the rate of dissipation of tu rb u ­
lent energy near the sea surface in the presence of waves. J . Oeophys. Res. 67, 3177—
3180.
Stokes, G. G. 1847 On the theory of oscillatory waves. Trans. Camb. Phil. Soc. 8, 441-455.
Sverdrup, H . U. & Munk, W. H . 1947 Wind, sea and swell: theory of relations for forecasting.
U.S. N avy Hydrogr. Office Pub. no. 601.
Taylor, G. I. 1953 An experimental study of standing waves. Proc. Roy. Soc. A 218, 44-59.
W atters, J. K. 1953 D istribution of height in ocean waves. N .Z . J . Sci. Tech. В 34. 409-
422.
1668

J. Fluid Mech. (1972), vol. 65, part 3, pp. 629-643 629

Printed in Great Britain

A class of exact, time-dependent, free-surface flows

By M . S. L O N G U E T - H I G G I N S
D epartm ent of Applied M athem atics and Theoretical Physics, University of Cambridge,
and National In stitu te of Oceanography, Wormley, Surrey

(Received 18 December 1969 and in revised form 20 Ju ly 1972)

Attention is drawn to a class of inviscid irrotational flows which satisfy the

conditions at a time-dependent free surface exactly. The flows are related to
the ellipsoids of Dirichlet (1860).
Depending on a parameter P , the cross-section may take the form of a variable
ellipse (P < 0 ), a hyperbola (P > 0 ) or a pair of parallel lines (P = 0 ). The elliptical
case was investigated both theoretically and experimentally by Taylor (1960).
The hyperbolic case (P > 0 ) is remarkable in that the flow develops a singularity
when the angle between the asymptotes approaches a right-angle. It is suggested
that this solution represents a possible instability near the crest of a standing
gravity wave of large amplitude.
In the intermediate case (P = 0 ) the solution describes an open-channel flow
in which the fluid filaments are stretched uniformly in a horizontal direction,
The latter flow is demonstrated experimentally.

1. Introduction
E xact solutions of the time-dependent equations of motion for a non-viscous
fluid with a free surface are quite rare. One class of motions known since the
last century is the ellipsoids of Dirichlet (1860), in which a rotating, gravitating
mass of fluid, of constant density and vorticity, is contained within a closed
ellipsoidal surface, which deforms continuously with time. The solution of the
three-dimensional problem has been derived concisely by Lamb (1932, §382).
A two-dimensional form of these motions was rediscovered by Taylor (1960)
in the course o f an experimental and theoretical study of the form of a jet issuing
from an elliptical orifice.
More recently the present author (1969), who was interested in the form of
a standing gravity wave of maximum amplitude, inadvertently discovered
a hyperbolic form of the same solution. Unlike the elliptical case, however, the
hyperbolic solution shows the interesting feature of a pressure singularity, or
shock, at the instant when the angle between the asymptotes becomes equal
to a right-angle. Though not necessarily describing the limiting form of a standing
gravity wave, the hyperbolic solution may very well describe an instability near
the wave crest, or the time-dependent behaviour of a jet impinging on a plane
wall.
It is the description of the hyperbolic flow which is the main purpose of the
present paper. Section 2 outlines a general approach to the problem of time-
34 FLM 55
 1669

630 M. S. Longuet-Higgins
dependent free-surface flows. The most general two-dimensional solution with
constant rate of strain is derived in § 3. It is seen to depend upon two dimension-
less parameters P and Q, governing respectively the time and length scales of
the motion. The elliptical case (P < 0 ) is described in § 4 , in a form rather
different from that given by Taylor (1960). The description of the hyperbolic
case is given in § 5, followed by a discussion of its applicability to the standing
wave.
The intermediate case (P = 0 ) is also of interest (§ 6 ). This describes a flow
bounded by two parallel plane surfaces, which approach one another. The
intervening fluid is ejected in such a way that any given particle travels outwards
with constant velocity. The flow is related to a well-known open-channel flow
but is exact, not being dependent on the shallow-water approximation or the
hydrostatic assumption. The flow is verified in a simple experiment, described
in § 8 .

2. The general problem

We seek a potential ф(х, у, t, z) satisfying Laplace’s equation
V V = 0, (2 . 1)
and a pressure function p{x,y>z, t) related to ф by Bernoulli’s theorem:

^ = - ^ - i W ) 2- F , (2.2)

where V is the gravitational potential, so that there exists a free surface on

which the pressure vanishes:
p = 0. (2.3)
Since the surface moves with the fluid we must have also

g , = ! +V № = 0 (2.4)

on the same surface. An arbitrary function of the time, which is sometimes added
to (2 .2 ), is considered as being absorbed into ф.

3. A two-dim ensional, gravity-free solution

L etu s write ф = %А(х2—у 2) —j f d t , (3.1)
where A and / are functions of the time, to be determined. The components и
and v of the velocity in the x and у directions are given by
и = A xу v = —A y. (3.2)

This is a simple, two-dimensional shear flow, as illustrated in figure 1. The flow

is symmetrical about both the x axis and the у axis, the instantaneous streamlines
being rectangular hyperbolae. The instantaneous streamlines in this particular
flow coincide with the particle trajectories.
1670

A class of exact, time-dependent, free-surface flows 631

The pattern of streamlines corresponding to the velocity potential

F i g u r e 1.
of equation (3.1), when A > 0. (When A < 0 the flow is reversed.)

Gravity being neglected, we may write V = 0 in (2 .2 ), so that the pressure p

is given by
p lp = - l ( A + A * ) x ‘ + H A -A * )y * + f, (3.3)
where a dot denotes partial differentiation with respect to t. Hence the rate of
change of p following the motion is given by

= ~ l ( A + lA A + 2A3) x * + \ ( A - 4ЛА + 2 +/ (3.4)

p ut

I f p = 0 and D p/D t = 0 are to represent the same surface, the corresponding

terms in (3.3) and (3.4) must be proportional. So we must have
1 + 4АЛ + 2А3 А - 4 А Л + 2Л3 f
(3.5)
A+A2 “ Л -А “ /*
The terms in A alone give
A A —4A2 + 2A* = 0 . (3.6)
Multiplying each side by 2A /А 9 and integrating we obtain

p (3.7)
A8 A *~ '
a constant, so that A 2 = Л*(1 + РЛ*). (3.8)
3+-2
 1671

632 М. 8. Longuet-Higgins

Hence t = + f ----- — ----- -. (з g)

~ J A 2(l + P A 4)i
Now each ratio in (3.6) is equal to the ratio of the differences of any pair of
numerators to the differences of the denominators. Thus we have
/ / / = iA A /A * = 4J./A, (3.10)
which has the solution / = \QA*t (3 -11)
where Q is a constant.
Hence altogether from (3.3)
2p /p = - (A + A 2) x1+ (A - А 2) у2+ QA*, (3 . 12)
the function A{t) being given inversely by (3.9). The free surface p = 0 is an
ellipse or a hyperbola according as the coefficients of x2 and yz are of the same or
opposite sign, that is according as
A* %A 2, P < 0. (3.13)
We shall take these two cases separately.

4. The elliptical case (P < 0 )

Write ( - P ) i = M > 0, (4.1)
so that when 0 < M A < 1 equation (3.9) becomes
, dA
_ J^ А 2(1-М *А*)Ь’ (4-2)
the constant of integration being suitably chosen. The substitution
M A - cos a (0 < cl < in ) (4.3)

leads to t = A f[ — ^ . — т. (4.4)
Jo cos20 ( 2 - sm 2 d)*
This integral may be expressed in terms of the Legendre elliptic integrals
Лас ^
E (a ,k) = I
(4.5)
F(a k) = Г ____
’ J о (1 —A sin26 )i’
—&;22sii
tabulated, for example, by Byrd & Friedman (1954). In fact
t = 2 W [ t a n a ( l - |s i n 2a ) i - ^ ( a ,2 i) + J P (a ,2 i)]. (4.6)
Using a as a parameter, A may now be plotted as a function of t as in figure 2 ,
for fixed values of M . When t < 0 the solution found by reflexion in the line
t = 0 is analytically continuous with the solution for t > 0 . The limiting case
M = 0 , however, requires special treatment (see below). In general A attains
a maximum value M~x at time t — 0 , as can be seen also from (4.2). When on the
other hand t -> oo then A -+ 0 , and from (4.2) A ~ t~x.
1672

A class of exact, time-dependent, free-surface flows 533

A

t
. The function A{t) when P < 0 , giving the velocity a t a fixed point as a function
F ig u r e 2
of the tim e t. Only the curves for A > 0 are shown. Those for A < 0 are obtained by
reflexion in the t axis.

F ig u r e 3. Successive configurations of the free surface when P < 0 (and A > 0 ).

The shading shows the regions where the pressure is positive.

Between t — ± oo the sign of A remains the same (positive in figure 2 ), so that

the flow at any given point remains constant in direction. The curve of A{t)
has points of inflexion (A = 0 ) where P A 4 = —J, that is to say where
M A = 2 -* = 0-8409.... (4.7)
 1673

634 M. S. Longuet-Higgins
The free surface p = 0 may be written in the usual form for an ellipse:
#2 У2
^2 + £2= 1 ’ (4-8)
where a and b are the lengths of the semi-axes, given by
. QA* , QA*
~ A^+A’ Ь-кгА' <49>
Since A 2 > \A | by (3.13) it is clear that for a real free surface to exist Q must be
positive. Also from (4.9) we have

by the differential equation (3.8). Hence the product of the semi-axes ol the
ellipse is equal to ( —Q /P)iy a constant, as we would expect from continuity,
nab being equal to the cross-sectional area.
We also have
Q A 2- A A2 Q A 2+ A
M* A* ’ b ~ M * A* > ( )
and so if A is expressed in terms of A,
Q 1± (1 - M W ) \ Q 1T(1
M2 M 2A 2 ’ M2 M 2A 2 { >

according a s ^ O . When t = 0, then a and 6 are both equal to Qb/M, and the
cross-section is circular. When t > 0 the ellipse is elongated in the x direction
and when t < 0 it is elongated in the у direction, as shown in figure 3.
It will be seen from (3 . 12) that the pressure p is a maximum at the centre of
the ellipse, at which point the velocity is zero, and that in general the pressure
increases towards the centre. This suggests that the flow may be stable provided
that the fluid is contained within the ellipse. The corresponding flow for an
elliptical cavity, when the fluid is outside the ellipse, is probably unstable.
A further property of the flow is that all surfaces of constant pressure {p = p 0)
are similar to the free surface p = 0 , but that these do not move with the fluid
unless we assume that p 0 is not constant but varies in time like A*.
The pressure at the centre of the ellipse (x, y) = (0 , 0 ) is given by

pip = iQA*. (4.13)

This was compared by Taylor (I960) with the pressure measured at the centre
of a jet issuing from an elliptical orifice in a vertical wall, with very good agree­
ment (see Taylor 1960, figure 5).
Taylor’s derivation of (4.13) differs from the one given here, the time dependence
being expressed in the form of an integral, which is then expanded as an infinite
series. I t may be noted that the infinite series
__ 1 _ 1 1
2. 3 + 8. 7 16.11 + ‘ ” *
1674

A class of exact, time-dependent, free-surface flows 535

which, is summed numerically by Taylor is equal to

2 B № ~ K fa ),
where E and К are complete elliptic integrals.

5. The hyperbolic case (P > 0)

Writing now P i = N > 0, (5.1)

Г со ^4.
we have when A > 0 t = ± )A + (6'2)

the constant of integration again being suitably chosen. The substitution

N A = c o t (P ) (5.3)

reduces (5.2) to t = ± J tfj' (S-4)

and hence
t = ± ^ [ t a n ( P ) ( l ~ is m * P )i-E (fiy l/2i) + P4/?, 1/2*)], (5.5)

where 2? and P are given by (4.5).

F igube 4. The function A when P > 0, giving the velocity a t a fixed point as a function
of t. The curves for A < 0 and t < 0 may be obtained by reflexion in the axes.
 1675

536 M. S. Longuet-Higgins
As before, A may now be plotted as a function of t (see figure 4 ). The curves
for A > 0, t > 0 may be extended by reflexion in either the t axis or the A axis.
A new feature, which did not appear in figure 2 , is the presence of a singularity
at t = 0 . In the neighbourhood of this point we have from (6 .2 )
t ~ ± (3N*A*y-\ N A ~ ± (N fit)i. (6 .6)

Thus the velocity (except at x —у = 0 ) becomes infinite like and the pressure
(except at the free surface) like tri. The displacement near t = 0 is like the integral
of the velocity and remains finite.
As we saw earlier, the cross-section of the free surface is a hyperbola, given by
{A + A 2)x2- ( A - A 2)y2 = QA*. (5 .7 )
Since in this case A 2 > A* it follows that the coefficients of x2 and y 2 have the
same signs as A and —A respectively. We may distinguish three cases as follows.

Case 1 : Q > 0
Then if A > 0 the form of the free surface is as shown in figure 5 (a), the shaded
area indicating the region where the pressure is positive. The angle which the
asymptotes make with the x axis, namely

a = arc tan

is greater than 45°. The angle decreases or increases according as A %0 . The

semi-axis a of the hyperbola is given by
a2 = QA*l(A+A2). (5.9)
As t 0 we have asymptotically
a2 ~ QA*/A ~ Q/Pi. (5.10)
If, on the other hand, A < 0 the form of the free surface is as shown in figure 5 (6 ).
The angle a is less than 45°, and the semi-axis b is given by
b2 = Q A * l(-A + A 2). (5.11)
Basically (and in the absence of gravity) the configuration is the same as in
figure 5 (a) but turned through a right-angle.

Case 2 : Q < 0
Then if A > 0 the form of the free surface is as in figure 6 (a), the angle a being
greater than 45 °. The shading shows the region of positive pressure. If A < 0
the configuration is as in figure 6 (b).

Case 3: Q = 0
In this case the free surface reduces to two planes making equal angles +^.m th-.
the x axis. If A > 0 the configuration is as in figure 7 (a), and if
figure 7 (b). j.
1676

A class of exact, time-dependent, free-surface flows 637

* у

(a) №)
F ig u r e 5. The form of th e free surface when P > 0 and Q > 0. (a) A > 0, (6) A < 0.

F i g u r e 6. The form of the free surface when P > 0 and Q < 0. (a) A > 0, (b) A < 0 .

In the hyperbolic solutions just found, both pressure and velocity become
large at infinity. Hence these are essentially local solutions, which may never­
theless describe the local behaviour of some realizable flows. For example the
solution of figure 5(a) may describe the time-dependent behaviour of a jet
impinging on a plane surface (y — 0 ). The flow of figure 6 may partially describe
the collapse of a mound of fluid under gravity. In each case the existence of
a singularity at a = 45° is very interesting. It appears that as this angle is
approached a weak shock will occur and that the pressure gradient will change
sign. Thus the configuration of figure 5 (a) will go over into the configuration of
figure 6 (6 ) (and that of figure 6 (a) into that of figure 5(b)). Where the pressure
 1677

638 M. S. Longuet-Higgins

F ig u r e 7. T h e form o f the free surface when P > 0 and Q = 0. (a) A > 0, (6) Ac 0.

was previously positive it now becomes negative, and a flow that was stable
becomes suddenly unstable.
It is natural to consider how these solutions are related to the problem of the
limiting form of a standing gravity wave of maximum amplitude. Penney &
Price (1952) showed that at the instant of maximum elevation the acceleration
at the crest of a standing wave is equal to g, directed downwards. Thus the fluid
near the crest is in a state of almost free fall. Penney & Price suggested on
theoretical grounds that the limiting slope at the crest should be 45°. These
authors' arguments were not accepted by Taylor (1953), who nevertheless showed
experimentally that the maximum slope was indeed very close to 45°. However,
the occurrence of instabilities near the crest made the precise observation of the
maximum angle rather difficult.
Since the hyperbolic solution described above has a singularity at t = 0, at
which the velocity becomes infinite, it evidently does not describe the motion in
a smooth standing wave. Nevertheless, since the fluid near the crest is in a state
of almost free fall, the flows of figures 5 (b) and 7 (6) (in the region# < 0) may well
represent possible instabilities that could occur near the instant of maximum
elevation.

6. The intermediate case (P = 0)

In this special case (3.9) becomes
t = ± l/A . ( 6 . 1)
Let us suppose, without loss of generality, that A > 0 . Then we have two
possibilities. I f t = 1/A ( 6 .2 )
2 л
we have from (3.12) 2t*'

Then the free surface p = 0 is given by

\y\ = («■*>
1678

A class of exact, time-dependent, free-surface flows 539

(2IQ)*y

I 2 1
/ / / / / yV/ ' /

1 I
/ У / / / Х / / / У / / / sty S / / / / 7
. , L T 2
7 / / / / у / / 7 У г77~

< 0
У / у / у ^ //уу / / / / \ /////'у _
T
S //S //*///, ///A s/ / / / / /
1 —l 1 1

/////A s///, / / / / / / / / / /
1 -2 1 i

F ig u r e 8. Successive configurations of the free surface when P = 0.

The flow is contained between two planes parallel to the x axis, which come
together with velocity (2Q)U~Z (see figure 8 (b)). The pressure on the x axis is
proportional to (r*. Hence as t -> 0 the pressure becomes very large, as a greater
and greater thickness of fluid has to be accelerated outwards.
The second possibility, when t = —1/A > 0 , is essentially the reverse of the
first; the fluid is contained between two lines parallel to the у axis (figure 8 (a)).
A paradoxical feature of the flow described by (6.2)-(6.4) is that although
the horizontal pressure gradient is zero the horizontal velocity и given by ( 3 .2 )
appears at first sight to be decelerating (since А ос 2-1). The paradox is resolved
by recognizing the difference between the Eulerian acceleration du/dt and the
Lagrangian acceleration Du/Dt. In fact any marked particle moves outwards
with constant horizontal velocity и (though with steadily diminishing vertical
 1679

640 M. S. Longuet-Higgine
velocity v), so that the horizontal component of the Lagrangian acceleration
vanishes everywhere.
The flow is such that when t is small all particles start from the neighbourhood
of the line x = 0 . As t increases, an observer at a fixed point sees first the swiftest
particles, since they arrive first, then particles with gradually slower and slower
velocities. Hence to a fixed observer the velocity appears to be decreasing in
time. The flow is analogous to an expanding universe.
An experiment to test the reality of this flow will be described in § 8 .
The flow will be recognized as being related to a well-known channel flow,
based on the nonlinear shallow-water theory in which vertical accelerations are
neglected (see, for example, Forchheimer 1930). The present solution is however
exact, and is independent of the shallow-water approximation.

7. Particle trajectories
The Lagrangian description of the class of flows described in this paper is
almost as simple as the Eulerian. Thus let (X, Y) denote the co-ordinates of
a fixed particle, functions of the time t and of the initial positions (X 0,70) of the
particle at time t0. Then corresponding to the flow described by (3.1) we have
dXjdt = u = дф/дх = Ax = A X , |
(7.1)
dY jdt = v = дф/ду = - A y = - A Y , )

and so (7.2)
т п в — А ®-
Hence Х = Х „Щ , T = (7.3)
'1
~

, ft,
a>
И

(7.4)
II

where
г
f

It can be shown directly that an expression of the general form (7.3) satisfies the
Lagrangian equations of continuity and of motion, and that the pressure is
constant on a free surface of elliptical cross-section provided that
FJP = X F -4 \F -')ld P , (7.6)
Л being a constant related to the axes of the ellipse.
The special flow described in § 6 corresponds to
A = 1Jt, F = t j t . (7.6)
The pressure in this case is given by

where h0 denotes the total depth of water at time t0.

8. Experimental verification
In order to neutralize the effects of gravity on these free-surface flows there
are at least three experimental alternatives.
( 1) The experiments may be conducted under conditions of free fall. This was
in effect the method adopted by Taylor (I960), who replaced the time variation
by a gradual space variation, in a free nearly two-dimensional jet.
1680

A class of exact, time-dependent, free-surface flows 641

F ig u r e 9. Sketch of apparatus to generate the free-surface flow corresponding to Q = 0.

(2) The experiments may be conducted at high speed or on a small scale, so

as to increase the Froude number and reduce the effective value of g. A limit to
the reduction in scale is set by surface tension. Thus high speed appears the more
promising.
(3 ) In some configurations gravity may be counterbalanced by a hydrostatic
pressure gradient. For this to be possible the free surface must be horizontal.
The only instance is when P = 0 (§ 6 ).
This suggested the following simple experiment.
Let a rectangular tank be fitted with a movable block, which initially is near
one end of the tank, as in figure 9. Let the space between the block and that end
of the tank be filled with a volume Q of water initially at rest. At time tQlet the
block be quickly accelerated to a uniform velocity U, say, along the tank. The
water should then follow the plunger in such a way that the free surface remains
horizontal and the depth of water in the tank is given by
h = Cl/Ut. (8.1)
The horizontal component of velocity in the tank will be given by

и = VxfxB, (8 . 2 )
where x is the distance from one end of the tank and xB = Ut is the distance
through which the block has travelled. Thus we have и = A x as in (3.2), since
A = 1It.
Any other motion of the plunger will cause a tilting of the free surface-for
example if the plunger is accelerated or decelerated, or if it is suddenly removed
altogether as in the collapse of a dam. Such flows will invite specifically gravita­
tional effects.
Figure 10 (plates 1 and 2 ) shows the realization of such an experiment in
a rectangular tank of height 18 in., width lOJin. and total length 12 ft 2 in. The
block, which moves from right to left, is in the form of a trolley running on
rubber wheels. Attached to the rear of the trolley is a vertical sheet of plywood,
 1681

642 M. S. Longuet-Higgine
sealed to the floor and side walls of the tank by a thin rubber sheet. Attached to the
front of the trolley is a sponge-rubber buffer and a horizontal steel cable, which
in turn is attached to an electric motor, as in figure 9. When the trolley reaches
the far end of the tank the clutch is automatically released and the trolley comes
to an abrupt halt. The trolley also carried two bricks whose weight helped to
seal the lower edge of the rubber sheet to the floor of the tank.
At the start of the experiment the trolley was held in position by a long pole
(whose shadow on the rear wall is visible to the left of the trolley in figure 10)
while the space to the right of the trolley was filled with dyed water, to a height
of 14in. At the word ‘go’ the pole was released and the clutch of the motor was
engaged. The trolley quickly accelerated to an almost uniform speed of about
1*6 ft/s. The sequence of photographs in figure 10 was taken with a cine camera
running at 64 frames/s. A selection of frames is shown, the serial numbers of
each frame being given on the right.
It can be seen that, at the start of the run, the initial acceleration of the trolley
produces a tilt of the free surface, and hence an incipient wave (figures 10 (a)-(c),
plate 1). But as the motion of the trolley becomes more uniform the wave is
damped out and the surface becomes level to a remarkable degree (figures 10(A)-(j),
plate 2 ).
On reaching the far end of the tank the trolley comes to a halt and water begins
to pile up against the rear face of the trolley (figures 10 (&) and (I), plate 2 ).
Fluid is theD thrown back (figures 10 (m) and (n), plate 2) on top of the horizontal
stream, which continues to flow from right to left and to diminish in depth.
At this stage, and indeed after about figure 10 (e) (plate 1), the horizontal flow is
supercritical, that is to say the velocity и exceeds {gh)b, and hence the oncoming
stream remains unaffected by the return flow. At a later stage (not shown in
figure 10 ) a bore is formed travelling from left to right. Ultimately, viscosity and
turbulence damp the flow, but in the short time corresponding to figures 10 (a)-(n),
the thickness of the boundary layer, given by (vt)i, is of order 1 mm, and so can
be neglected.
In principle it is possible to reverse the experiment and to compress the
fluid by forcing the plunger in the opposite direction (towards the fixed end)
still maintaining a horizontal free surface. But in practice the required initial
flow would be difficult to establish. Any irregularity of the free surface would
cause wave motion, and the horizontal contraction of the fluid by the plunger
would then tend to increase the wave amplitude through work done against the
radiation stresses (see Longuet-Higgins & Stewart 1961; Taylor 1962).
When, on the other hand, the plunger moves away from the fixed end, any
unwanted disturbance of the free surface tends to be reduced.

Dr J. S. Turner kindly assisted me with the experiments described in § 8 . 1 am

indebted to Sir Geoffrey Taylor for interesting discussions.
1682

Journal of Fluid Mechanics, Vol. 55, <parl 3 Plate 1

F io u r e s 10 (a) to (g)

LONGUET-HIGGINS (Facing p. 544)

 1683

Journal of Fluid Mechanics, Vol. 55, part 3 Plate 2

F ig u b e s 10 (Л) to (n)
F igure 10. A model experiment designed to realize the expanding flow described in §6.
Each number on the right indicates the serial number of the corresponding frame. Tho
film was shot at 64 frames per second.

LONGUET-HIGGINS
 A class of exact, time-dependent,free-surface flows 643

REFERENCES
B yrd, P . F . & F r i e d m a n , M. D. 1964 Handbook of Elliptic Integrals for Engineers and
Physicists. Springer.
D ir i c h l e t , G. L. I860 Untersuchungen uber ein Problem der H ydrodynam ik. Abh. Kon.
Oest. Wiss. Gottingen, 8, 3-42.
F o r c h h e e m e r , P. 1930 Hydravlik. Leipzig: Teubuer.
L a m b , H . 1932 Hydrodynamics, 6th edn. Cambridge University Press.
L o n g u e t -H i g g i n s , M. S. 1969 The exact hydrodynamical description of a slosh. Proc.
N A T O Advanced Study Institute on Topics in Geophys. Fluid D yn., p. 44. Univ. Coll.
N. Wales, Bangor.
L o n g u e t -H i g g in s , M. S . & S t e w a r t , R. W. 1961 The changes in am plitude of short
gravity waves on steady non-uniform currents. J . Fluid Mech. 10, 529-549.
P e n n e y , W. G. & P r i c e , A. T. 1952 Finite periodic stationary gravity waves in a perfect
fluid. Phil. Trans. Roy. Soc. A 244, 254-284.
T a y l o r , G. I. 1953 An experimental study of standing waves. Proc. Roy. Soc. A 218,
44-59.
T a y l o r , G. I. 1960 Form ation of thin flat sheets of water. Proc. Roy. Soc. A 259, 1-17.
T a y l o r , G. I. 1962 Standing waves on a contracting or expanding current. J . F luid
Mech. 13, 182-192.
 1685

СReprinted from Nature, Vol. 239, No. 5373. pp. 449-451,

October 20, 1972)

Periodicity in Whitecaps
W h il e flying low over the ocean between Adelaide and Kan­
garoo Island (36° S, 137° E) one of us (J. S. T.) observed the
following interesting phenomenon.
In a moderate wind of 6-9 m/s from the north, whitecaps
were being formed sporadically, each appearing for about a
second or less and then disappearing, leaving a streak of foam
that would persist for a few seconds. Strikingly, the whitecaps
would generally appear in succession, each making its appear­
ance down-wave of the previous one. The interval between the
appearance of successive whitecaps was from about 8 to 10 s.
Another of us (M. D.) has made similar observations from
the deck of a ship steaming westwards in the N. Atlantic
(53° 20' N, 37° 54' W) in a cross-wind of 30 knots and of
nearly constant direction of 350°. Again, whitecaps were
observed to appear in succession, each down-wave of the last,
with a periodicity of about 12.2 s. The period of the larger
waves, on the other hand, was substantially less, being about
8.7 s, as determined visually.
We offer the following explanation. Sea waves are not of
uniform amplitude, but occur in groups of high and low
waves1. The wave envelope progresses with the group velocity
Cf, which for waves in deep water is about half2 the phase
velocity C. Individual wave crests will appear therefore at the
rear of a group, move forward through the group and diminish
again towards the front (Fig. 1). The maximum wave amplitude
will be attained near the centre of the group, at which point
breaking may occur. Now progressive waves tend to break at
a certain limiting value of the wave steepness (=wave ampli­
tude/wavelength) for which the downward acceleration of the
crest reaches 1/2^ (see ref. 3). Provided that the wavelength
of the dominant waves is nearly constant, this limiting con­
dition will occur at a certain critical value of the wave amplitude,
as indicated-in Fig. 1. At this critical amplitude a whitecap
will appear and will persist until the wave amplitude falls
below the critical amplitude.
1686

-------- -С
Q

Fig, 1 Schematic drawing of a wave crest advancing through a

group of high waves. The broken line indicates the critical wave
height for breaking. In a and с the waye crest is breaking. In
b there is no breaking. The time between a and с is two wave
periods.

2
 1687

Because breaking represents the greatest loss of wave energy,

which is replenished much more slowly by the wind3, the
critical amplitude will be exceeded only slightly. Hence not
more than one wave in each group will normally be breaking
at any given time. In some groups there may be no breakers.
Thus, breaking will occur periodically, as each crest in
succession advances through a group and attains the critical
amplitude. But because the wave envelope itself is advancing
with the group velocity, each whitecap will occur further
forward than the previous one. The periodicity o f the whitecaps
is determined by the velocity of the wave crests relative to their
envelope, which equals ( C - Cg). For waves in deep water,
this velocity is C/2. So we would expect the time interval
between successive whitecaps to be just twice the wave period.
This is confirmed by a simple experiment which we have
performed in a wave tank o f length 45 foot, width 2 foot, and
mean depth 2 foot. Waves o f period 0.5 s (effectively deep-
water waves) were generated at one end o f the tank. By
starting the wave-maker from rest, we produced a train of
waves progressing down the tank, the front advancing with the
group velocity. The form of the wave envelope near the front
is described theoretically by a Fresnel integral, which locally
exceeds the steady wave amplitude following the front by
about 9%. If the steady wave amplitude is slightly less than
critical, then on passing through the front o f the group each
wave crest is made to break. The period o f breaking was
observed to be double the wave period.
By starting and then stopping the wavemaker gradually we
found it possible to generate a finite group o f waves which
could be made to begin breaking at a certain point along the
tank. The period of breaking was then also observed to be
twice the wave period.
At the time o f the observations near Kangaroo Island,
mentioned above, the period of the highest wave, corresponding
to a mean wind-speed o f 7.5 m/s and a fetch o f 100 km, would
be expected to be about 4.5 s, or about half the period o f the
whitecaps. This would be in accordance with our hypothesis.
On the other hand, the shipboard observations suggest that
the ratio o f whitecapping period to wave period generally lies
between 1 and 2. The explanation may be that in a locally
generated wave field the frequency spectrum is not always
narrow, and that the shorter, steeper waves are helped to break
by the orbital motion within the larger waves4, which would
not break by themselves. The visual observation o f wave
period is related to surface elevation, whereas for the purpose
o f wave breaking it is the spectrum of wave steepness which is
relevant. The spectrum o f surface elevation has a maximum at
lower frequency than the spectrum o f wave steepness, so that
3
1688

this would tend to reduce the apparent ratio o f whitecap

period to wave period. Against this is the tendency o f an
observer to ignore the breaking o f the shorter waves and to
record only the larger and more persistent whitecaps.
In the observations just quoted the wind had veered through
180° within the previous 12 hours. There may therefore have
been some swell in the opposite direction which would have
tended to increase the rate o f wave breaking.
Under favourable conditions, that is to say with a steady
wind strength and direction, we may expect the period o f the
whitecaps to be more nearly twice the wave period. If so, then
the aerial observation o f whitecaps provides a ready means o f
estimating both the wind-speed and the altitude o f an aircraft.
For, dividing the period o f the whitecaps by two, we obtain the
wave period, and hence the phase velocity C, which is related
to the wind speed5. Knowing the wave period we may also
calculate the actual wavelength, and from the apparent wave­
length, as seen by an observer in the aircraft, we may calculate
the height o f the observer. Knowing the wavelength, and that
the highest waves are in a breaking condition, we may also
estimate the maximum and r.m.s. wave amplitudes1*3.
Two further possible consequences may be noted. (1) Not
only breakers but other features o f the surface elevation will
tend to occur periodically in the ocean. Thus over any
moderate area o f the sea surface, say a few wavelengths in
diameter, the surface elevation may be expected to repeat
itself roughly every two wave periods. This phenomenon
should be observable, for example, in the intensity o f radar
backscatter6 or o f underwater acoustical backscatter from the
free surface, for sufficiently narrow beams. (2) A whitecap
represents a local disturbance to the normal laminar propaga­
tion o f surface waves. The disturbance is intermittent, but its
propagation speed being one-half the phase-speed o f the
dominant waves, we may expect a transfer o f energy to waves
having one-half the dominant wave period, or one-quarter the
wavelength. Spectral maxima at twice the dominant wave
frequency are indeed observed7. There may also be a transfer
o f energy to much longer waves, namely those with wavelengths
comparable to the envelope o f the dominant waves. But the
length o f a wave group not being uniform this reaction will
be less specific.
These suggested effects o f breaking waves are distinct from
the weak non-linear interactions already well known in the
literature8.
We have discussed waves in deep water only. In shallow
water the phase velocity is more nearly equal to the group
velocity, and the occurrence o f breaking is determined largely
by the ratio o f the wave height to the local depth o f water9.
4
 1689

Thus the situation is quite different.

One of us (M. D.) acknowledges a Killam postdoctoral
fellowship from the University of British Columbia. We thank
Professor R. Radok for providing the occasion on which the
aerial observations were serendipitously made.

M. D o n e la n

Canada Centre for Inland Waters,

Burlington, Ontario
M. S. L o n g u e t-H ig g in s
National Institute o f Oceanography,
Wormley, Godalming, Surrey
J. S. T u rn er

Department o f Applied Mathematics

and Theoretical Physics,
Silver Street, Cambridge

Received March 11, 1972.

1 Longuet-Higgins, M S., 7. Mar. Res.,11, 245 (1952).

2 Lamb, H., Hydrodynamics, 6th ed. (Cambridge University Press,
Cambridge, 1932).
3 Longuet-Higgins, M. S., Proc. Roy. Soc., A, 310, 151 П969).
4 Longuet-Higgins, M. S., Proc. Roy. Soc., A, 311, 371 (1969).
5 Sverdrup, H. U., and Munk, W. H., Wind, Sea and Swell, Publ. 601
(US Hydrographic Office, Washington, 1947).
6 Munk, W. H., and Nierenberg, W. A., Nature, 224,1285 (1969).
7 Barnett, T. P., and Wilkerson, J. C., /. Mar. Res., 25, 292 (1967).
8 Phillips, О. М., The Dynamics o f the Upper Ocean (Cambridge
University Press, Cambridge, 1966).
9 Munk, W. H., Ann. NY Acad. Sci., 51, 376 (1949).

Printed in Great Britain by Flarepath Printers Ltd.. St. Alban», Hem.

1690

J. Fluid Mech. (1973), vol. 67, part 1, pp. 129-148 129

Printed in Great Britain

A model of flow separation at a free surface

By M. S. L O N G U E T -H IG G IN S
Department of Applied Mathematics and Theoretical Physics,
University of Cambridge and National Institute of Oceanography,
Wormley, Godalming, Surrey

(Received 2 October 1970 and in revised form 28 July 1972)

Flow separation can be observed ( 1) at the leading edge of a spilling breaker

or ‘ white-cap’, (2) at the lower edge of a tidal bore or hydraulic jump and (3)
upstream of an obstacle abutting a steady free-surface flow. At the point of flow
separation there is a discontinuity in the slope of the free surface. The flow
upstream of this point is relatively smooth; the flow downstream of the dis­
continuity is turbulent.
In this note, a local solution for the flow in the neighbourhood of the discon­
tinuity is derived. The turbulence is represented by a constant eddy viscosity N>
and the tangential stress across the interface between the laminar and turbulent
zones is expressed in terms of a drag coefficient C. It is shown that the inclinations
o f the free surface of the two sides of the discontinuity depend on С only, and are
independent of N and g. As С increases from zero to large values, so the inclina­
tion of the free surface in the turbulent zone increases from 10° 64' to 30°. In the
laminar zone the inclination of the free surface simultaneously decreases from
10° 54' to 0°, the densities in the two zones being assumed equal.
Owing to the possible entrainment of air at the separation point, the effective
density p' in the turbulent zone may be less than the density p in the laminar zone.
When these densities are allowed to be different it is found that the possible
flows are o f two distinct types. Flows of the first type, called ‘ quasi-static’ ,
are contiguous to a state of rest. Flows of the second type, called ‘ dynamic’,
are contiguous with the frictional flows described above, for which p' = p. At
a given positive value of С there exists generally only one quasi-static solution.
There is also just one dynamic solution provided p \p > 0*50012. On the other
hand, if p'lp < 0-5 there may be either two or no dynamic flows, depending on the
value of C; and when 0*5 < p’\p < 0-50012 there may be three such flows.
The inclination of the free surface is studied as a function of С and p'jp.

1. Introduction
The breaking of surface waves has important dynamical consequences in the
ocean, both in deep and in shallow water. In deep water, white-caps may be
responsible for a high proportion of wave energy dissipation (see Stewart &
Grant 1962), and hence for the conversion of wave momentum to larger scales of
motion. Thus if D denotes the mean rate of energy dissipation per unit horizontal
area, and if с is the phase velocity of the breakers, an amount of wave momentum
9 FLU 57
 1691

130 M. S. Longuet-Higgins
equal to Djc is available for conversion to larger scales of motion, including mean
currents.
The turbulence due to breaking waves is probably also the chief agent of
vertical mixing and for vertical transfer of heat and momentum in the upper­
most layers of the ocean, while the entrainment of air by breaking waves must
influence the transfer of dissolved gases between atmosphere and ocean.
Likewise in shallow water it is known that longshore currents, which transport
large quantities of sand and sediment parallel to the coastline, are generated with­
in the surf zone by waves striking the coast at an oblique angle (Galvin 1967).
The mean tangential stress r exerted by the waves is equal to {Djc) sin Q, where
D and с have the same meanings as before and в denotes the angle of incidence
(see Longuet-Higgins 1970a, b). Though most of the dissipation in the surf zone
is due to breaking waves, neither the form of the breaking waves, nor the dis­
tribution of turbulence with depth are well known.
Indeed, while the theory of surface waves of small amplitude has been highly
developed (see, for example, Lamb 1932) and much progress has been made
with the theory of weakly nonlinear interactions (Phillips 1966), our know­
ledge of breaking waves is surprisingly scanty.
It was shown by Stokes (1880) that a steady progressive wave of limiting
amplitude must have a sharp crest with a discontinuity of 60° in surface slope.
Using this criterion, the limiting form of waves in deep water was determined
by Michell (1893) and in shallow water by McCowan (1894), Davies (1952) and
others. Numerical calculations of unsteady waves have also been successfully
carried out by, for example, Street (1972) up to the point of breaking. But
little, if any work has been done on the analytical description of a gravity
wave after it has broken. The reason is no doubt connected with the fact that
such flows are essentially turbulent, and potential theory is no longer applicable
to the whole flow.
Laboratory studies of breaking waves in shallow water have been published
by Mason (1952), Iverson (1952), Ippen & Kulin (1955), Divoky, Le Мё1ти^ &
Lin (1970), and others. Mason (1952) distinguished two types of breaking wave:
on the one hand ‘ plunging breakers’, in which the wave crest topples forwards
and falls violently onto itself; and on the other hand *spilling breakers’, in
which the free surface becomes unstable near the wave crest and forms a quasi­
steady white-cap on the forward slope of the wave. Illustrations of these two types
are given in figures 1 and 2 of Mason (1952).
In contrast, a considerable amount of experimental work has been done on
hydraulic ‘ jumps’ in channel flows; for a comprehensive review see Rajaratnam
(1967). These are essentially shallow-water flows. The flow in the jump has been
explored in its dependence on the Froude number. Evidently some experimental
difficulties remain to be overcome, in particular those arising from the entrain­
ment of air at the free surface. Quantitative measurements by Rajaratnam
(1962) indicate that the mean density in a vertical section can fall as low as
0 -8 g/cm3, and possibly lower.
It must be emphasized, however, that breakers are by no means a shallow-
water phenomenon, as is shown by the presence of white-caps in deep water
1692

Flow separation at a free surface 131

F ig u r e 1. Examples of flow separation, (a ) A spilling breaker, (b) The flow

induced by an obstacle at the surface of a steady stream.

(Monahan 1971). Generally they seem to occur when the vertical acceleration
approaches the limiting value \g (Longuet-Higgins 1969 a). It is also certain that
the extreme shearing motion near the crests of large-scale waves will sometimes
give rise to a shearing instability.
In deep water there are at least two ways in which waves may attain their
limiting amplitude. The first occurs when the frequency spectrum is narrow. Since
the phase velocity of deep-water waves is twice that of their group velocity,
the amplitude of waves will grow to a maximum and then decay as they pass
through their wave envelope. Near the maximum amplitude they may break.
The second mechanism applies when the shorter waves riding on the backs
of longer waves are forced by the latter to steepen and break near the crests of
the longer waves (Longuet-Higgins 1969 b).
In deep water, white-caps appear generally to last for a shorter time than on
gently sloping beaches, but from visual observation white-caps may be better
classed as ‘ spilling’ than as ‘ plunging’. Some observations of white-caps are
given in a recent note by Donelan, Longuet-Higgins & Turner (1972).
The complete analytical description of breaking waves, and of hydraulic
jumps, presents a challenging task. In the present paper we do not attempt a
complete description of either of these phenomena, but we suggest on the other
hand a possible model for one local feature of such flows, namely the flow near the
forward edge of a spilling breaker or hydraulic jump, where the turbulent flow
meets the more tranquil water (see figure 1). In this region the flow is evidently
turbulent, as elsewhere in the breaker. We suggest that it may be treated
as a turbulent wedge with a certain eddy viscosity whose magnitude is deter­
mined by conditions outside the local flow (see figure 2 ). Such a local solution
is developed in the present paper.
In § 2 we show that the only type of smooth irrotational flow which can sustain
a discontinuity of surface gradient is one with an angle of 120° - in fact, a general­
ization of the Stokes 120° angle flow (see figure 2). But to sustain this the fluid on
9-2
 1693

132 M. S. Longuet-Higgins

the left of the discontinuity must be at rest, and the free surface on the left must
be horizontal. There is no way to support a rise in water level on the left, except
by frictional forces, or (in turbulent motion) by Reynolds stresses.
We therefore assume (in § 3) that the fluid on the left of the discontinuity is
turbulent, and is dominated by Reynolds stresses. The latter are assumed to be
represented by a coefficient N of eddy viscosity. The mean velocity in the turbu­
lent zone is assumed small compared to the flow on the right (in a frame of refer­
ence travelling with the phase velocity). Across the boundary of separation, both
normal and tangential stresses are balanced, the tangential stresses on the left
being given by a constant coefficient С times the square of the velocity on the
right. At the free surface, both components of stress vanish. Under these con­
ditions it is shown that a local two-dimensional solution does exist. Moreover
it has the remarkable property that the inclination of the free surface on the two
sides of the discontinuity depends only upon О, and is independent of both N and
g. The angle of elevation on the left is shown to lie between 10° 54' and 30°; the
angle of depression on the right is between 10° 54' and zero. Various special
cases are discussed in § 4.
In §5 we discuss the effect of a possible difference in density on the two sides of
the interface, caused by the entrainment of air at the point of separation. The
flows described above are generalized so as to include an arbitrary ratio of the
densities, and the effect on the inclination of the free surface is determined.
A comparison with some observations of the flow in hydraulic jumps is made
in§ 6 .
In the discussion in §7 it is suggested that a way may be opened up to treat
the hitherto intractable problem of breaking waves, and more generally, of mixed
turbulent and laminar free-surface flows. In such analyses the present local
1694

Flow separation at a free surface 133

solution may be an essential feature, representing the flow near the point of
separation.

2. A frictionless flow
By choosing a reference frame moving with a suitable horizontal velocity we
may reduce the motion to a steady flow. Let polar co-ordinates be taken as in
figure 2 , with the origin at the point of discontinuity and with the line 0 = 0
directed vertically downwards. Let the boundary of the fluid on the right be
given by в = a, that on the left by в = —у, with a possible interface at в = —/?.
Let p and g denote the pressure and gravitational acceleration, and p and p'
the densities on the right and left respectively.
Let us first suppose the flow to be irrotational. Then, in an incompressible fluid,
we seek a stream function \jr such that
V2^ = 0 (2.1)
with boundary conditions
Ф = 0, p = 0, when в = a,
^ = 0, ^continuous, when в = —fi, ■ (2.2)
rjr = 0, p = 0, when в = —у..
We may, or may not, have a discontinuity in дг/г/дв at в ——/3. Following Stokes
(1880) we try f = .4 г” вши(0 - а ) ( - / ? < 0 < a ). (2.3)
This satisfies the first condition at the surface в = ос. To satisfy the second con-
dion we note that in steady frictionless flow
pip —gr cos в + \q2 = constant (2.4)
by Bernoulli's theorem. Since from (2.3)
g2 = n2A*r2n~2 (2.5)
we have p = p(gr cos в — -j- constant. (2.6)
The vanishing of p on в = a then implies that
n= h (2.7)
so that the angle between radial streamlines must be 120°, and further
Л2 ==|^соза, (2.8)
so that if А Ф 0, then cos a > 0 ; in other words, the streamlines must slope
downwards from the origin. This is also clear from the fact that the velocity has
to vanish at the origin, and a particle on a surface of constant pressure can gain
kinetic energy only by going downhill.
If now we try to satisfy the free-surface conditions at 6 = —y by the expres­
sion -ф- = Вгп&шп{6+у)} —у < в < —Д we find, by a similar argument that
either у < \тт, which is obviously impossible when 0 < a < £я, or else В = 0,
that is to say the fluid on the left is stationary. Then the free surface on the left
is horizontal (y = ^n).
 1695

134 M. S. Longuet-Higgins
Can we satisfy the boundary conditions on the interface в = -ft\ On the left
the pressure is simply the hydrostatic pressure p = p'grcos ft. On the right the
pressure is given by equation (2 .6), which, in view of (2.8), may be written as
p = pgr (cos ft - cos a). (2.9)
Equating p and p' on the two sides of в = —ft we have
p' cos ft = p(cos ft—cos a), (2 .10)
or, since a-f ft = ftf,
{p—p') cos ft = p cos (§7Г - ft). (2 . 11)
This leads immediately to
(p-p')lp = y st& n ft-b (2.12)
and hence tan/? = (Ър-2р')ЦЗр. (2.13)
The two extreme cases are as follows. First, when p' = 0, then
a = ft = £tt, A = ±$gi. (2.14)
This is the Stokes 120° angle (Stokes 1880). Secondly, when p' = p then
a= h P = i”, A = 0. (2.15)
The free surface on the right is then horizontal and no flow takes place. Inter­
mediate values of the density ratio p'Jp give flows which have non-zero values of
A and which are in effect generalizations of the Stokes 120° angle flow. All these
have a discontinuity in velocity on the surface 6 = —ft, and all have a horizontal
free surface on the left.
We remark that if a non-zero vorticity (without friction) is allowed in the
flow on either side of the discontinuity, it will not affect the angle of the free sur­
face. For ijr is then the sum of a harmonic function plus a solution ijrQof the equa-
tion (2.16)
VVo = -"о>
where o)0is the limiting vorticity near the comer. But since in radial co-ordinates

(2.17)

it follows that the relevant solution of (2.16) will be of order o)0r2, which tends to
zero with r more rapidly than the harmonic function (2.3). Hence the presence of
vorticity may affect the curvature of the free surface near the crest, but not the
limiting angle. For the Stokes 120° angle this was pointed out by Miche (1944).

3. A turbulent flow
The observation that the free surface generally slopes upwards to the left
of the discontinuity means that the flow there must be effected either by friotional
forces or by Reynolds stresses, for without these it would be impossible for a
particle at the free surface to be brought to rest at the origin (see equation (2.4)).
In this section, therefore, we shall seek a solution representing laminar flow
on the right but highly turbulent flow on the left (see figure 2 ). For simplicity we
1696

Flow separation at a free surface 135

F ig ubk 3. The local flow in a frietionlesa fluid with different densities.

shall first assume that the densities on the left and on the right are equal, i.e.

Most turbulent flows are difficult to represent satisfactorily, the present being
no exception. Moreover there are some outstanding differences between this and
other types of flow, in that the source of the turbulence may be largely external
to the region considered - in a breaking wave, for example, most of the turbulence
may originate nearer to the wave crest. Secondly, the mean flow near the stagna-
tion point must be relatively small to an observer moving with the phase velo­
city of the wave.
These considerations lead us to suggest a very simple model as follows. Let us
suppose that the p« are given in terms of the mean flow by expressions analo­
gous to those for ordinary viscous stresses, that is to say

(3.1)

Here N is an eddy viscosity, which we shall at first assume to be constant. The

quantity P is analogous to (but not equal to) the pressure p. We assume also that
the Reynolds stresses dominate the inertia terms in the equations for the mean
motion. Then in the turbulent zone we have essentially a balance between Rey­
nolds stresses and gravity. The equations for the stream function xjr of the mean
flow become, in rectangular co-ordinates,

(3.2)

(y being directed vertically upwards). Cross-differentiation gives

V^ = 0 (3.3)

and we see also that Pfp + gy is the harmonic conjugate of NVhJr, the expression
(PIp+ gy) -f iNVhjr (3.4)

being an analytic function of x + iy.

 1697

136 M. S. Longuet-Higgins
On в = —у we have the boundary conditions
f = 0, Pm = 0, Pr0 = 0, (3.5)
where now, in radial co-ordinates,

(3.6)
jj,—

(see, for example, Lamb 1932, p. 679).

At the interface в = —/?, which separates the turbulent flow on the left from the
laminar flow on the right, we shall assume that
^ = 0, Pm = - p , Pr„ = Cpq2, (3.7)
where p and q denote the pressure and speed in the laminar flow; that is,
Ф = -9У-Ья\ \ ,38.
%q2 = gr cos a, = gr cos '
The second of the conditions (3.7) may therefore be written as
P J p -'jy = \q2- (3.9)
The third of equations (3.7) is based on the assumptions that ( 1) the mean
flow on the left is small compared with that on the right, and (2) there is a small
entrainment of fluid from the laminar flow into the turbulent flow. If we define
an entrainment constant e by
_ normal velocity of laminar flow across boundary
~ relative tangential velocity of laminar and mean turbulent flow
then conservation of momentum tangential to the boundary suggests that
C = e. (3.11)
In a recent review Turner (1969) has quoted evidence that, at the boundary of
a square jet, e is commonly of order 0*08. Hence we expect О to be of this order
also.
It is remarkable that the above equations (2. 1) and (3.3), with the boundary
conditions (2.2), (3.6), (3.7) and (3.9), admit a simple exact solution. Let us

ijr = Er3sin 3(0+y) + J?Vssin(0+у ), (3.12)

where E and F are constants to be determined. Using (2.17) we have
V2^ = 8JVsin(0+y), (3.13)
which is a harmonic function, so (3.3) is satisfied. Moreover from (3.4) it follows
that
P/p+gy = SNFr cos {в+y). (3.14)
Since трis an odd function oftf + у we see that the two boundary conditions xjr = 0
1698

Flow separation at a free surface 137

and Prj? = 0 on <9+ у = 0 are automatically satisfied. In general we have from (3.6)
and (3.12) Prt/p = _tfr[12tfsin3(0+y) + 4Fsm(0+y)]. (3.16)
This obviously vanishes when 04-y = 0 . We have also
PoelP ~ ЯУ = - 2Nr[6E cos (0+у ) + 2F cos (0+у)] - {Pjp + gy) (3.16)
or using (3.14)
Pee/p-gy = - X2Nr[Ecos3{e+y) + Fcos{6+y)]. (3.17)
The condition that Pee should vanish on 0 = - у now yields
E + F = ~(g/12N) cosy (3.18)
and the three conditions on 0 = - Д yield respectively
2£sin3ft + Psinft = 0,
E cos 3ft + Pcosft = -Q , (3.19)
E sin 3ft + IF sin ft = —2CQ,
where we have written ft = у (3 20)

(which is the vertex angle of the turbulent ‘wedge’) and

(32I)
From the first and third of equations (2.19) it follows that
E = - 30^/sin 38, F = ЗСд/sin 8 (3.22)
and so on substituting in the second of (3.19) we find, if Q Ф 0 ,
cot 3ft—cot 8 = 1/3G. (3.23)
Similarly on substituting in (3.18) and using (3.21) we find

cosec 3ft- cosec 8 = -------- .„C°S^— 7-. (3.24)

3C cos (§tt —у -4*ft) v '
Equations (3.23) and (3.24) constitute the solution to the problem. Given the
drag coefficient G we may first use (3.23) to determine 8 and then use (3.24) to
determine у and hence all the other parameters of the flow. The pattern of the
streamlines is independent of both N and g.

4. Discussion o f the solution

From (3.23) we have
1 _ -sin 2ft - 2 cos 8 .
3G sin 38sin 8 sin 35 *
and so G = - £ sin 3ftsec ft. (4.2)
This function is shown in figure 3. In the range 0 < ft < \тт,it is negative and has a
minimum equal to - 0-196 where
cos 2ft = 1), ft = 34° 16#. (4.3)
 1699

138

F ig u r e 3. Graph of the function f(d) — —£ sin 36 sec 8, giving the drag coefficient С as a
function of the angle S at the apex of the turbulent wedge.

The angles of inclination a ', fi' and y \ and the angle S' of the discontinuity in
F ig u r e 4 .
slope at the separation point, shown as functions of the drag coefficient C.
1700

Flow separation at a free surface 139

Since С is negative, this range 0 < 8 < $7r corresponds to upwards flow along the
interface.
When, on the other hand, < 8 < \тт,G must be positive. Indeed, as 8 \n
so G -> oo. But since in practice С is probably small, only those values of 8 close
to \ttare of interest. In this neighbourhood we have
C ^ 8 - \ tt=8', (4.4)
say. It wiH be noted that 8' is just equal to the discontinuity in the inclination of
the free surface at r = 0.
Beyond the point 8 = the roots of (4.2) do not yield physical solutions, since
they imply (as will be seen) that a > \n.
Prom (3.23) and (3.24) we have, on eliminating G, the relation
cot 3<?-cotfl _ cos (|я —y + <?) .. gv
cosec 38—cosec 8 cos у
relating 8 and у . Manipulation of both sides yields
cos 5/cos 28 = cos(£4-f 7r) + 8in(£+f 7r)tany (4.6)
and hence tan у = tan 28 + 3* sec 28. (4.7)
This simple relation enables us at once to calculate у in terms of 8, and hence also
ft and a. In figure 4 we have plotted, besides 8’, the functions
a! = i ^ - a , ft' = \n-ft, / = y -J tf. (4.8)
The angles a’ and y' represent the inclination of the free surface to the horizontal
on the right and left respectively, and ft' represents the inclination of the interface.
From figure 4 it will be seen that as С increases from 0 to oo, so 8' increases from
0° to 30°. At the same time a! decreases from 10° 54' to 0°; ft' increases from
49° 6' to 60°, and y rincreases from 10° 54' to 30°. We shall now discuss some
special cases.
We take first some non-negative values of С (figures 5a, b, c). A typical
flow is when 8 = 75° (figure 56). The flow is in the direction we might expect
intuitively. The free surfaces on the left and right are inclined at angles
of 21° (upwards) and 6° (downwards) respectively. As the drag coefficient is
increased, so the flow approaches the configuration shown in figure 6 (c). Although
G oo, the flow on the left remains bounded, since at the same time g2 -> 0
in such a way as to make Cpq2finite. In the limit, (3.22) shows that E = F = 3CQ,
so that on the left __
^rocr3sin2(0+y)sin(^-f y). (4.9)
This represents a flow that is concentrated mainly in the neighbourhood of the
interface.
As С moves in the opposite direction we pass through the limiting case G = 0,
8 = 60°, shown in figure 6 (a). Paradoxically, the free surface is not horizontal
in this case. The reason is that, although the tangential stress at the interface
vanishes, the presence of the turbulence results in an increase of the normal
stress Pee (hence a reduction in P) whose effect must be counterbalanced by a
140 M. S. Longuet-Higgins

F ig u r e 6. Flow configurations when С = 0 : (a) 8 ^ 1, (6) 8 — 15°,

(c) 8 = 30°, (d) 8 - 60°.
1702

Flow separation at a free surface 141

reduction in the pressure p on the right of the interface. Hence the flow on the
right is similar to a special case of the frictionless flow described in § 2 . The fluid
on the left, however, is not stationary. Taking the limit of equations (3.22) as
O ^-Owe find E Ф 0 , F -> 0 and hence
xjrac r3sin 3(0 + y), (4.10)
which represents a simple, irrotational flow in a 60° corner. By the symmetry of
the flow it can be seen that this flow ensures the vanishing of Pr0 both on в — —p
and# = —y.
Some solutions representing negative values of С are shown in figure 6 . When
8 1 (figure 6 a) the turbulent zone is very thin and the laminar flow approximates
that in a Stokes 120° angle С is then small.f As 8 increases so does |C| until the
maximum at 8 = 34° 16'. When 8 — 45° the free surface on the left is horizontal,
the mean flow being given by E = —F = —3(2)i CQ, and so
VFocr3sin2(0 + y)cos(0 + y). (4.11)

Finally when 8 = 60° we have the same flow as in figure 5 (a) except that the direc­
tion is reversed.
We note that in deriving (3 .22) and (3.23) it was necessary to divide by Q,
so that the solution Q = 0 , hence E = F — 0 was excluded. This is the trivial solu­
tion in which the fluid is at rest on both sides of the interface. But we see that this
solution is not continuous with either of the ranges examined above. This remark
serves to emphasize the nonlinearity of the present flows. To establish them we
must pass through some configuration other than a state of rest. Once established,
however, observation suggests that they are at least securlarly stable.

5. Turbulent flow with differing densities

Any entrainment of air at the point of separation may be expected to reduce
the effective density of the fluid above the interface; that is, in the turbulent
region. The method of § 5 can also be applied to the problem when the density p'
o f the turbulent fluid on the left differs from the density p in the laminar region
on the right. The drag coefficient С may be expected to depend upon the relative
density difference 7] = (p—p')/p. Ellison & Turner (1969) found that in an almost-
parallel flow the entrainment constant e was a function of the Richardson number
Ri = rjgLj U2, where L and U are typical length and velocity scales. In the present
situation the flows are assumed to be self-similar, and all geometrical length
scales are proportional to the radial distance r. Taking U = q as a typical velo­
city, we see that Rioz ygr/q2, which is independent of r. Hence we may be justified
in assuming О to be a constant independent of r, though depending in some way
on the relative density difference rj, and the geometry of the flow.
Following the method of § 3 we find in place of (3.22)

Е = ~ sm
Ш 3o
й оp- ’ J, = sm
S ^оp (5Л)
t It is possible that this flow approximates the flow near the crest of a spilling breaker.
 1703

142 М. 8. Longuet-Higgins

С
The angle 8 shown as a function of O, for given values of the density ratio
F ig u h e 7.
p'lp. The dashed line corresponds to p'jp — 0-5. The maximum value of С on this curve is
indicated by a solid circle (•).

and in place of (3.23) and (3.24)

_C Q 3^\ \
1 Л
3C\ cosa/’ l
4COSV 1

i ' 1 ^ co sa * J
II
to

where у = (/>—p')lp **1 —

- S2?,
, R = p'lp• (5-3)
By eliminating a and С respectively from equations (5.2)
r = gm3J[sin(a-frr)+78in*] (5
6 (1 —7j) cos 8 sin (£—Jtt)
(sm28 + JS)sm(8—J7r) + 2?7 cos £ sin2 £ .
and tan у — cos 2<5 [sin (£-£ 7r) + i/sm,S] ' '
which are obviously generalizations of (4.2) and (4.7). For any given value of ц
we now have С and у in terms of 8, so that a, /?, у and 8 may all be expressed as
functions of G.
In figure 7 , 8 is shown as a function of m over the whole possible range
0 < 8 < 120°.
1704

Flow separation at a free surface 143

F ig u r e 8. Values of af and y ' for the quasi-static flows (£' < 0 ) , giving the inclination of
the free surface in the laminar and the turbulent zones respectively.

In fact we are interested only in positive values of C, and this limits the relevant
values of 8 to two ranges, (1) when 30° < 8 < 60° and (2) when 60° < 8 < 120°.
In range 1, С can tend to zero for values of 8 given by
sin (£ - J7r)+ 77sin£ = 0. (5.6)
At such values both E and F vanish, by (5.1), so that the fluid on the left is static.
From (5.5) we may verify that in this case у — 90°, hence 8 — 90° —p and (5.6)
reduces to (2 . 11). Thus in the limit as С 0 we recover the flows described in § 2 .
As О is increased, keeping 8 within the range 1, we obtain a set of flows which are
generalizations of the flows in § 2. The tangential stress at the interface no longer
vanishes. It can be seen from figure 7 that these more general flows are continuous
with those corresponding to points on the line 8 — 60° (8' = 0 ) as R ->• 1 (ij -+ 0 )
which correspond to a state of rest in both the laminar and turbulent zones.
For this reason we may call the flows in range 1 ‘ quasi-static’ flows. For
 1705

144 М. 8. Longuet-Higgins
such flows we see that 8' (= £ -6 0 °) is negative, so that the free surface is
convex.
The values of a' and y' corresponding to the quasi-static flows are shown in
figures 8 (a) and (b). It will be seen that as R is diminished from unity, keeping С
constant, so the angle of depression a' on the right generally increases, as we
might expect, though the behaviour of y' is more complicated. In the limit
as jR->0 and С -> 0 , we see that a! 30°, so that the flow on the right is the
Stokes 120° angle flow, as we should expect. However, the flow on the left
depends on the way in which С and R each tend to zero. Such non-uniform
convergence is to be expected, since the ratio of the tangential stress to normal
stress in the turbulent fluid depends on the ratio Сpip1; that is, C/R.
Consider now the flows in range 2 (60° < 8 < 120°). From figure 9 it can be seen
that these are continuous with the flows described in §§ 3 and 4. We shall call the
flows in this range ‘ dynamic’. For dynamic flows С may not tend to 0 unless
either 8 -> 60° or 120°, or unless R -> 0 . There is a critical ratio of the densities:
R = 0-5. When R > 0-5, then С becomes arbitrarily large as 8 approaches a
certain value. On the other hand, when R ^ 0-5, С cannot exceed a certain maxi­
mum value, depending on R.
When R = 0-5 the maximum value of С occurs at 8 = 116° 30', which is inside
the range 2 . Hence there exists a small range of the density ratio, say
0*5 < R < R0,

for which С has two stationary values. Thus we see that when 0 < R < 0-5 there
are generally two dynamic flows or none, for a given value of C; when 0*5 < R < R0
there are either three dynamic flows or one; and when R0 < R < 1 there is always
just one flow for a given value of C.
An approximate method (not described here) shows that the value of R0 is
roughly £(l + £), where £ = (2/135)2 = 0-00022. Hence R0 = 0-50011. By direct
computation using (5.4) it was found that in fact R0 = 0-500117.... The corre­
sponding values of a', /?', y' and 8' are given by
a' = 3°23', ft’ = 56° 37',1
y '= l ° 5 4 ', 8' = 57°42',J
and С has the value 0-26146.
The general values of a! and y' for the dynamic flows are shown in figures
9 (a) and (b). At moderate values of С it appears that, as R is diminished from
unity, so a' and y' both initially increase. For example, if С is of order 0 -1, then
a change of 10 % in the ratio R, from 1-0 to 0-9, will increase a' by about 5° and
will increase y' by about 8°. At very low densities (small values of R) if С also
is sufficiently small, a' must tend to 30°, so that we retrieve the Stokes 120°
angle. At the same time y' lies somewhere between 30° and 60°.

6. Comparison with observations

There appear to be few existing measurements with which the present model
can be directly compared. Hydraulic jumps in open channels have been fairly
1706

Flow separation at a free surface 145

F ig u r e 9. Values of (a) a' and (6) y ' for the dynamic flows (£' > 0), giving the inclination
of the free surface in the laminar and the turbulent zones respectively.

extensively studied (for a review see Rajaratnam 1967). In these, however, the
flow is strongly influenced by the presence of the bottom, especially at high
Froude numbers. Nevertheless, some comparisons may be made. Bakhmeteff &
Matzke (1936) showed that the profile of the hump was dependent on the Froude
number Ft = UJfahJi, where XJ1 and \ denote the mean velocity and water
depth below the jump. From figure 2 of Rajaratnam (1968), based on their data,
zo F L M 57
 1707

146 M . S. Longuet-Higgins

tany' У'
1-98 0-59 29°
2-92 0*44 24°
4-09 0-39 21 °
5-53 0-26 15°
8-63 0-23 13°
T able 1. Inclination of the free surface above the toe of a hydraulic jump
in open channel flow.

the inclination of the free surface just above the toe of the jump is given by
table 1. Comparison with figure 9(b) shows that these inclinations, in the range
13-29°, might be expected on the basis of an entrainment coefficient С of order
0*1 and a density ratio lying in the range 0*7 < p’jp < 1-0 .
Air entrainment in hydraulic jumps has been studied by Rajaratnam (1962).
The volume concentration c, as defined by him, is approximately equal to 1 —p'lp.
Over a comparable range of Froude numbers (2-42 < Fx ^ 8 -12) he found that,
for points near the upper surface, с lay generally between 0-5 and 0-20. No mea­
surements very close to the toe of the jump are reported, but the trend of the
observations suggested that the air concentration there was even higher. Thus,
the density ratio p'/p was certainly sometimes as low as 0*8 and may well have
been less. The range of densities inferred earlier is therefore reasonable.
There appear to be no careful observations of the lower interface of the air-
entrained flow. Rajaratnam (1967, p. 218) states that the angle of depression of
the entrained bubbles increases with the Froude number. This may be due
partly to the increase in the mean flow velocity, relative to the rate at which
bubbles rise to the surface. Generally, interfacial angles indicated by air-bubble
entrainment will tend to underestimate the actual angle of depression. Again,
the presence of the bottom will tend to diminish the apparent angle of depression,
except perhaps very close to the toe, and to reduce the range of validity of the
local solution.
Immediately upstream of the toe it has been generally assumed, on the basis
of nonlinear hydrostatic theory, that the free surface is exactly horizontal.
However, in the neighbourhood of a discontinuity of surface elevation the hydro­
static assumption does not necessarily apply. Figure 9 (a) suggests that close to
the toe the free surface will be inclined to the horizontal at an angle at! lying
between about 10° and 20°. Such small angles may have been overlooked.

7. Discussion and applications

In the foregoing model it has been assumed for simplicity that both N and С
are constant. The fact that the resulting flow field is independent of the value of
N suggests that the velocity of the model does not depend critically on the con­
stancy of N. Nevertheless we may note that comparable analytic solutions could
be obtained if N were assumed proportional to a power of the radial distance
r; for example, if Ncc gbA.
1708

Flow separation at a free surface 147

It must be emphasized that the representation of the turbulence by an eddy

viscosity is necessarily an analytical device, made necessary by the complication
of the actual flow. The chief significance of the present model may be to suggest
a way in which the hitherto intractable problem of white-caps may be made
amenable to analysis, namely by treating certain regions of the flow as laminar
and others as turbulent, the turbulence being represented by an eddy viscosity,
uniform or otherwise.
For example, it may well be possible to construct a solution for a spilling
breaker in which the white-cap is represented by a viscous zone near the wave
crest, the eddy viscosity diminishing towards the rear, as in figure 1. Because of the
dissipation of energy it would be necessary, in an exactly steady state, to assume
that energy was continually being supplied by normal or tangential stresses at
the free surface.
In deep water the waves may be supposed to move forwards through a wave
group, thus changing slowly in amplitude. Such a solution might be considered
as quasi-stationary. If, on the other hand, the waves are in shallow water
approaching a gently sloping beach, the energy is supplied by the incoming
swell.

I am indebted to Dr J. S. Turner for comments on a first draft of this paper, and

to Mr I. D. James for checking the algebra in §5.

REFERENCES
B akhm eteff, B. A. & M a t z k e , A. E. 1936 The hydraulic jump in terms of dynamic
similarity. Trane. A.S.G.E. 101, 630-680.
D a v i e s , Т. V. 1962 Symmetrical, finite amplitude gravity waves. In Gravity Waves,
pp. 65-60. U.S. Nat. Bur. Standards. Circular no. 521.
Divoky, D .,L e M £ h a u t £ , B. & L i n , A. 1970 Breaking waves on gentle slopes. J. Geophys.
Res. 75,1681-1692.
D o n e l a n , М., L o n g u e t - H i g g i n s , M. S. & T u r n e r , J. S. 1972 Periodicity in white-caps.
Nature, 239, 449-451.
E , Т. H. & T u r n e r , J. S.
l l is o n 1959 Turbulent entrainment in stratified flows. J.
Fluid Mech. 6 , 423-448.
G a l v in ,C. J. 1967 Longshore current velocity: A review of theory and data. Rev. Geo-
phys. 5, 287—304.
Ipfen , A. T. & K u l i n , G. 1965 Shoaling and breaking characteristics of the solitary
wave. M .I.T . Hydrodynamics Lab. Rep. no. 15.
I v e r s o n , H. W . 1952 Laboratory study of breakers. In Gravity Waves, pp. 9-32. U.S. Nat.
Bur. Standards. Circular no. 521.
L a m b , H . 1932 Hydrodynamics, 6th edn. Cambridge University Press.
L o n g u e t - H i g g i n s , M. S. 1969a On wave breaking and the equilibrium spectrum of wind-
generated waves. Proc. Roy. Soc. A 310, 151-159.
L o n g u e t -H i g g i n s , M. S. 19696 A non-linear mechanism for the generation of sea waves.
Proc. Roy. Soc. A 311, 371-309.
L o n g u e t - H i g g i n s , M. S. 1970a Longshore currents generated by obliquely incident sea
waves, 1 . J. Geophys. Res. 75, 6778-6789.
L o n g u e t - H i g g i n s , M. S. 19706 Longshore currents generated by obliquely incident
sea waves, 2. J. Geophys. Res. 75, 6790-6801.
M c C o w a n , J. 1894 On the highest wave of permanent type. Phil. Mag. 39, 361-359.
xo-a
148 M . S. Longuet-Higgins
M ason, M. A. 1952 Some observations of breaking waves. In Gravity Waves, pp. 215-220.
U.S. Nat. Bur. Standards. Circular No. 521.
Miche, R. 1944 Mouvements ondulatories de la mer en profondeur constante ou d6crois-
sante. Ann. Ponts et Chaussees, 114, 25-87, 131-164, 270-292, 396-406.
M i c h e l l , J. H. 1893 The highest waves in water. Phil. Mag. 36, 430-437.
M o n a h a n , E. C. 1971 Oceanic whitecaps. J. Phys. Oceanogr. 1, 139-144.
P h i l l i p s , О. M. 1966 The Dynamics of the Upper Ocean. Cambridge University Press.
R a j a r a t n a m , N. 1962 An experimental study o f the air entrainment characteristics o f
hydraulic jump. J. Inst. Engng India, 42, 247-273.
R a j a r a t n a m , N. 1 9 6 7 Hydraulic jumps. Advances in Hydroscience, vol. 4 , pp. 197-280.
Academic.
R a j a r a t n a m , N. 1968 Profile of the hydraulic jump. Proc. A.S.C.E. H Y 3, 663-673.
S t e w a r t , R . W . & G r a n t , H. L. 1962 Determination of the rate of dissipation of turbu­
lent energy near the sea surface in the presence of waves. J. Geophys. Res. 67, 3177-
3180.
S t r e e t , R. L. 1972 Shoaling of finite-amplitude waves on plane beaches. Proc. 12th
Conf. on Coastal Engng, pp. 345-361. New York: A.S.C.E.
S t o k e s , G. G. 1880 On t h e t h e o r y o f o s c i ll a t o r y w a v e s , Appendix B. Math. & Phys.
Papers, 1, pp. 225-228. Cambridge University Press.
T u r n e r , J. S. 1969 Buoyant plumes and thermals. Ann. Rev. Fluid Dyn. 1, 29-44.
1710

J. Fluid Mech. (1973), vol. 67, part 3, p. 624 6 24

Printed in Great Britain

F ILM REVIEW

Breaking Waves. By S. P. K j e l d s e n and G. B. O l s e n . Technical University

of Denmark. 16 mm, black and white, 20 min. 1000 K r (purchase) or
100 K r (rental). In English.
The theory o f surface waves o f small amplitude is by now highly developed. In
contrast, the study of breaking waves is relatively untouched. Breaking waves
play an important role in the maintenance o f our shorelines and in the coastal
transport o f sand and sediment. The time may be overdue for fluid dynamicists
to plunge metaphorically into this mixed laminar and turbulent regime.
The publication of the above film, prepared by two students at the Institute
of Hydrodynamics and Hydraulic Engineering of the Technical University of
Denmark, is thus particularly opportune and welcome. The film first defines and
describes the three main types o f breakers on sloping beaches, namely the
“ spilling breaker” (a relatively gentle affair, in which the wave crest is nearly
symmetrical and foam spreads gradually down the forward face); the “ plunging
breaker” , in which the crest is markedly unsymmetrical and the foam pene­
trates deeply into the forward face of the wave; and the “ surging breaker” ,
which occurs only on steep beaches.
After a careful statement of terminology the film describes how the three
main breaker types are found to depend mainly on two parameters: the gradient
of the beach and the steepness of the waves in deep water. Certain length
scales (such as the local depth o f water) are seen to be irrelevant, at least
when the wavelength is sufficiently large.
The film also shows how the “ breaker height index” is found to depend on
both the beach slope and the deep-water wave steepness. Lastly a fourth type o f
breaker, namely the “ collapsing breaker” is mentioned. This is an intriguing
variant o f the surging breaker, in which the fluid appears suddenly to lose its
forward momentum, and to collapse.
While the whole film is expository in character, being based largely on the
previous work of Galvin and others, it is nevertheless fascinating and may well
stimulate fluid dynamicists to be more active in this neglected branch of their
subject. The photography is excellent, good use being made of an improvised
travelling carriage to maintain the camera opposite the waves, and of flash
photographs to demonstrate the orbital motion of fluid particles. Those viewers
interested in sound effects will note the economical use of percussion both for
background and emphasis.
In one respect only is the film disappointing, namely that in spite of its title it
makes no mention of breaking waves or white-caps in deep water. Acting over a
much greater area, deep-water white-caps are the tell-tale clue for the wide­
spread conversion of wave momentum into surface currents. But to do justice
to this subject perhaps a separate film is needed.
M. S. L o n g u e t -H i g g i n s
 1711

J. Fluid. Mech. (1974), vol. 63, part 1 , pp. 1-20 j

Printed in Great Britain

An ‘ entraining plume’ model of a spilling breaker

By M. S. L O N G U E T -H IG G IN S an d J. S. T U R N E R
Department of Applied Mathematics and Theoretical Physics,
University of Cambridge

(Received 20 August 1973)

It is proposed that a spilling breaker can be regarded as a turbulent gravity

current riding down the forward slope of a wave, and can be treated using methods
which have been successful in other contexts. The whitecap retains its identity
because it is lighter than the water below, owing to the trapping of air bubbles,
and information from laboratory experiments is used to estimate the density
o f the air-water mixture in different circumstances. Entrainment of water from
below, at a rate E(Ri0) which is a function of the overall Richardson number Ri0,
has two opposing effects. It provides increasing mass and buoyancy fluxes which
can produce an accelerating flow and it also gives rise to a drag, because the en­
trained fluid has upslope momentum (in co-ordinates moving with the wave crest).
A similarity solution is obtained under the assumptions that the flow is
steady in time, and that the slope and the density difference remain constant.
In this solution, the thickness of the whitecap is proportional to the distance s
measured from the crest of the wave. The tangential velocity is proportional
to si. Since the velocity in a Stokes 120° angle is also proportional to si, this
implies that such a flow can start from a small disturbance with zero flux, and
propagate with constant acceleration. An important consequence of the analysis
is that solutions of this kind are possible only when the slope and the density
difference between the whitecap and the water below are sufficiently large;
otherwise the upward drag dominates, and a self-sustaining flow cannot form.
For a slope of 30°, near the crest of the breaking wave, the theory predicts that
a density difference greater than 8 % is required to sustain a steady motion,
at which point the downslope velocity is 12 % of the opposing velocity at the
wave surface. A ‘ starting plume’ model of the adv^noing front of the breaker
is also discussed, which suggests that this too will be accelerating uniformly,
but will have a velocity somewhat less than that in the flow behind.
A comparison with the laboratory observations of Kjeldsen & Olsen verifies
several features of the model, including the order of magnitude of the relative
velocities in the whitecaps and the wave beneath. It also reveals the intermittent
nature of the flow, which is here explained as due to the intermittent rounding
o f the wave crest due to damping of the wave by the whitecap on the forward face.

1. Introduction
The physical importance of the phenomenon o f wave breaking, both in shallow
and in deep water, has been discussed in a recent review (Longuet-Higgins 19736).
I PLм 63
1712

2 M. S. Longuet-Higgins and J. S. Turner

For example, it was pointed out that breaking waves are responsible for the
conversion o f wave momentum to surface currents, for the large-scale mixing
o f the surface layers of the ocean, and for enhancing the transfer of heat, salt
and dissolved gases by the production of bubbles and spray.
Very little is known, however, about the mechanism of breaking of surface
waves, especially in deep water. The form of the limiting wave has been calculated,
following the early work of Stokes (1880), who showed that the steady progressive
wave o f maximum amplitude must have a sharp angle of 120° at the crest. It
may be possible to follow the evolution o f a wave up to the point of breaking by
numerical methods similar to those used by Chan & Street (1970) for a wave in
shallow water. However, apart from a study by Longuet-Higgins (1973a) (one
feature o f which will be discussed further below), no theoretical description has
been given o f the motion near the crest and in the resulting whitecap after the
wave has broken. Mathematicians working in this field are accustomed to dealing
with potential flows, and since the same methods cannot be applied to a turbulent
breaker, this part o f the problem tends to be neglected. In the present paper the
turbulent character o f the motion at this stage is taken as the starting-point,
and ideas developed in another field are adapted to describe the progress of
a whitecap down the front of a wave. Our theory is applied more particularly
to breakers on a shelving beach, but there is nothing in the model which prevents
it being applied also to deep-water waves.
The formulation of our model has been strongly influenced by laboratory
studies of breaking waves and some associated phenomena. In particular Mason
(1962) clearly distinguished two types of breaking wave: ‘ plunging breakers’, in
which the wave crest curls forward and plunges deeply into the slope o f the wave
some distance from the crest, and ‘ spilling breakers’ , in which the broken water
seems to develop more gently from an instability at the sharp crest and forms
a quasi-steady whitecap on the forward slope. Only this latter type, the spilling
breaker, will be considered in the present paper. The contrast between these
two types (and a third, the ‘ surging breaker’ which can develop as a wave runs
up on a steep beach) is shown very clearly in a recent cine film made by Kjeldsen &
Olsen (1971; see also Galvin 1972).
A characteristic property of a spilling breaker is that, as it breaks gently at
the crest, it traps enough air bubbles for the resulting air-water mixture to be
significantly lighter than the water below it. This density difference will inhibit
mixing with the face of the wave, so that the whitecap rides on top of the sloping
sea surface. The basic assumption of our theoretical model is that the whitecap
can be regarded as a distinct turbulent flow, which is driven down the slope by
the component of gravity in that direction in just the same way as a turbulent
gravity current on a solid sloping boundary. As the flow continues, the turbulence
will lead to the entrainment of water from the laminar wave surface below, and
it will also result in the further incorporation of air, specially near the front of
the whitecap, so that the density difference can be maintained.
The theory o f Longuet-Higgins (1973a) suggests a model for the local flow
near the forward edge of a spilling breaker, where the turbulent whitecap meets
the undisturbed water surface. Here, on the contrary, we emphasize (in § 2 ) the
 1713

An ‘ entraining plume ’ model of a spilling breaker 3

F i g u r e 1 . Sketch showing the features of a spilling breaker which are incorporated in the
theoretical model. The wave is moving from right to left and has a whitecap on its forward
face. The velocities in both the wave and whitecap are measured relative to the wave
crest, with positive direction downwards.

properties of the flow some distance behind the front, using the balance of forces
at a fixed cross-section in the manner proposed for turbulent gravity currents
by Ellison & Turner (1959). Later (in § 6) it will be shown how the advance of
the front can be described in a way which is consistent with this steady plume
model. Common to both Longuet-Higgins’s and Ellison & Turner’s theories,
however, is the assumption that there is a tangential stress at the boundary
between the turbulent and laminar flows, due to the entrainment across this
boundary
The several elements to be incorporated in the theory of §2 are shown in
figure 1. In co-ordinates moving with the phase velocity the flow at the surface
of the waves is directed upwards towards the crest, with velocity decreasing to
zero at the corner. On top of this is the whitecap, with gravity tending to drive
it down the slope. Entrainment into this turbulent flow will add both water
from below and air from above (or at the front). Mixing into the whitecap will
be driven by turbulence produced by shear across the interface, but it will be
inhibited by the fact that the turbulent air-water mixture in the whitecap is
lighter than the water below which it entrains. The rate at which water is in­
corporated will depend on the velocity difference u0 = и' —и between the white­
cap and the wave, on the density difference p —p' (wherep' is the mean density
o f the air-water mixture) and on the local length scale, say the depth 8. Following
Ellison & Turner (1959) we suppose that the entrainment rate is a function of
the overall Richardson number
Bi0 = g'8 cos вlu%, ( 1)
where g' = g(p—p')lp and в is the slope. Rather less is known about the entrain­
ment of air, but in § 3 we discuss laboratory measurements of air concentrations
in hydraulic jumps (see, for example, Rajaratnam 1962) and in self-aerated flows
on steep slopes (e.g. Straub & Anderson 1958) which allow us to put some limits
on p'.
A similarity solution describing the development of the whitecap with dis­
tance from the crest is obtained in § 4. This must be an oversimplification of the
actual unsteady flow, but it exhibits very clearly the most important physical
features. The problem is unusual in that entrainment is both the driving and
retarding mechanism: driving because it provides an increasing flux of water to
accelerate the whitecap down the slope of the wave, and retarding because the
i«a
1714

4 M. S. Longuet-Higgins and J. S. Turner

entrained water has momentum in the upslope direction. The balance between
these opposing tendencies and gravity determines the behaviour of the spill,
and it is shown in § 5 that only for large enough slopes and density differences can
a downslope flow be formed. This problem has some features in common with
the ‘ layer reversal’ calculation of Elhson & Turner (1959). They showed how
to calculate the strength of an opposing flow necessary to cause an inclined plume
to reverse its natural direction of motion (with gravity) and to be swept along by
the ambient stream.
In §7 we make some comparisons between the results of the theory and
laboratory measurements on spilling breakers, which confirm some o f the
features of our model but also reveal the intermittent character o f the flow.
This is discussed further in § 8, where a physical explanation is suggested.

2. The motion o f an inclined plume entraining a varying ambient flow

The development of our model is adapted from that given by Ellison &
Turner (1959), to which reference should be made for more details. Take the
origin moving with the wave crest, the s axis along the wave slope, and the
n axis perpendicular to this; the motion is assumed to be two-dimensional,
i.e. independent of the distance along the crest. The mean velocity u' (positive
in the downslope direction), depth 8 and density p' in the whitecap at any s are
defined using integrals of the local properties u'n and p'n across the flow:

u'S - r < d n , (2)

u '4 = j “ u'n4 n , (3)

u'Sp' - r < P n d n . (4)

There are certain differences between the present flow and the turbulent
gravity currents considered previously which should be pointed out immediately.
The density difference which drives the flow is now not the same as that which
inhibits the entrainment. The full density difference between air and the air-
water mixture in the breaker provides the driving buoyancy force, whereas it
is the smaller density difference between the air-water mixture and the water
below it which must be used when calculating entrainment. The entrainment
assumption is summarized in the equation of continuity
d(Su')lds = Е\ий\, (5)
which states that the downslope mass flux is increased by entrainment at a rate
proportional to the velocity difference between the whitecap and the wave
slope below. The factor E is a function E(Ri0) of the overall Richardson number
defined by (1), whose form is known from laboratory experiments. Implied by
the use of Ri0 as the only non-dimensional parameter involving the density
difference is the Boussinesq approximation, which is used in all that follows.
 1715

An ‘ entraining plume * model of a spilling breaker 5

In some cases where the proportion of air in the breaker is large, it may not be
justifiable to neglect the variation of density in the inertia terms; but in view of
the strong simplifications made in other ways it hardly seems appropriate to
make the model too elaborate.
Another difference from the earlier theory becomes apparent in the derivation
of the momentum equation. The velocity of the fluid being entrained into the
breaker is not constant (as it was for the ‘ reversal’ problem) but varies like
и ос sb (see §4), and this introduces an extra term into the equation. It will also
appear from the similarity solution that the plume velocity has this same de­
pendence on s, so that the mass (or buoyancy) flux is increasing with distance. It
is therefore not appropriate to seek solutions in which the buoyancy flux remains
constant (as it does for a simple gravity current entraining ambient fluid which
is at rest or in uniform motion). Instead, we shall suppose that the mean density
p' o f the air-water mixture remains substantially constant at all s, an assumption
which will be shown in § 3 to have some experimental support.
The momentum equation, with an arbitrary exterior flow, can be written in
the form
-^(u‘4 ) = Euu0+ S2g$ sin в — ^ {gS2cos в), (6)

where the Boussinesq approximation is used, and g rather than g' appears in the
driving terms. On the left is the total rate of change of momentum flux. The
first term on the right arises from the entrainment of fluid with velocity и at
a rate Eu0, the second represents the component of gravity accelerating fluid
down the slope, and the last expresses the pressure force on the layer due to
its changing depth, the pressure gradient being assumed quasi-hydrostatic, equal
to p'g cos в normal to the surface. There is no solid boundary, and hence no other
drag term. and S2are profile constants, which are required to allow for arbitrary
density profiles. They are assumed to be independent of s and are defined as in
Ellison & Turner (1959) by
SlP'S* = j^2p^ndn (7)

and S2p'S = (8)

numerical estimates of them for air-water mixtures are given in § 3.

The term on the left of (6) and the first term on the right can be combined to
give, for constant slope,

^-(Su'u0) + Su’ ^ = S ^ s i n t f - ^ c o s f l ^ i p , (9)

which is the form used in the subsequent analysis. This differs from the momentum
equation used by Ellison & Turner through the addition of the term in du/ds,
which was zero in the previous application, where the external flow was uniform.
1716

6 M. S. Longuet-Higgins and J. S. Turner

3. The processes o f aeration

A detailed study of the mechanisms whereby air can be trapped and retained
in a whitecap lies outside the scope of the present work. In order to proceed with
the entraining plume model, however, we need to make some plausible assumption
about the variation of pf with s and в. In this section we must therefore digress
briefly to discuss some relevant laboratory data on air entrainment in turbulent
flows of water.
The first way in which air bubbles can be incorporated in the whitecap is by
the over-running o f a layer of air by the advancing front. This is similar to the
trapping which occurs in the roller zone of hydraulic jumps, and which has been
thoroughly studied experimentally (see the review by Rajaratnam 1967).
According to Rajaratnam (1962), the concentration of air in the jump rises
rapidly initially, then decreases gradually downstream. The air is fairly uniformly
distributed vertically, and the maximum concentration depends systematically
on the Froude number; it reaches values as high as 20 % over the range of con­
ditions he studied. This suggests that the mean density p' in a vertical section
can fall as low as 0-8g/cm3 near a vigorously turbulent front.
Another mechanism for the entrainment of air is the ‘ self-aeration’ which
can occur along the whole upper surface of a thin, vigorously turbulent flow
(see, for example, Straub & Lamb 1966; Straub & Anderson 1958; Lakshmana
Rao, Seetharamiah & Gangadharaiah 1970). High velocity open-channel flows
(on a spillway, for instance) can begin to trap air when the turbulent boundary
layer on the bottom reaches the surface, and ‘ white water’ develops. The
equilibrium distributions of air have been measured, and show a continuous
variation from water containing a few air bubbles at the bottom to all air and
no water at the top. Essentially there are two zones, the lower containing air
bubbles suspended in water, and the upper consisting of water drops in air.
An example of such a profile (converted to densities) and the associated velocity
profile is shown in figure 2. Also of interest to us here is the observation that the
mean concentration is a function mainly of the slope, with a weaker dependence
on the Froude number. In figure 3 we summarize results of Straub & Anderson
(1958) which show the variation with slope of the mean density over the whole
depth, and that over the lower zone alone (the latter, shown in curve (a), being
perhaps the more relevant for the present application). The dashed line represents
a theoretical result which will be discussed in § 5.
Both these sets of observations suggest, therefore, that for whitecaps running
down a given slope with (as will be shown) a constant Froude or Richardson
number, it is appropriate to take the concentration of air, and hence the density,
as constant. The air-water mixture is likely to be lighter on steep slopes (with
p‘ as little as 0-7 near the crest of the wave where the slope is 30°) than it is lower
down, but the effect of this will be investigated in our model by considering
a series of fixed slopes and density differences, rather than a varying profile.
Plots like figure 2 also allow us to estimate the values of Sx and S2 (from (7)
and (8)) to use in the calculations, and we suggest Sx = 0-80 and S2 = 1*00 as
representative values. The deviations of these constants from unity are measures
 1717

An ‘ entraining plume * model of a spilling breaker 7

0 2 4 6
и (m/s)

F ig u r e 2. An example of the density and downslope velocity profiles measured through

a self-aerated flow. (After Straub & Lamb 1966, figure 9.)

Slope (deg)

F i g u r e 3. The mean density of self-aerated flows measured as a function of slope, (a) shows
the result of averaging over a lower region consisting of air bubbles suspended in turbulent
water, and (6) includes a more diffuse region of spray droplets above this. (After Straub &
Anderson 1968, figure 6.) (c) represents our theoretical estimate of the minimum density
differences required to produce a self-sustaining whitecap at various slopes.
1718

8 M. S. Longuet-Higgins and J. S. Turner

o f non-uniformity o f the density distribution, and clearly constant density is
a pretty good assumption; it is even better for the profiles measured in air-
entraining hydraulic jumps.
There is another aspect of the self-aeration process which may have a bearing
on whitecaps and should be mentioned now. As Gangadharaiah, Lakshmana Rao
& Seetharamiah (1970) have pointed out, entrainment can occur only when the
turbulence at the free surface has enough energy to overcome the surface-tension
energy, and so allow eddies to project out of the surface and trap air bubbles.
The criterion can be written, using an argument which will not be reproduced
here, in the form
I = (phU2l<r)(ut hlv)i > Ic, (10)

where h is the depth, U the mean velocity, и* the friction velocity, cr the surface
tension and v the kinematic viscosity. Thus the ‘ inception number’ I must
exceed some critical value Jc (of order 50 when the turbulent boundary layer
has reached the surface) before air can be trapped at the surface in this way.
For a breaker, this means that the flow must be deep enough and fast enough,
or in terms o f the theory in § 4 (which implies that I oc si), must have progressed
far enough from the crest, before this mechanism can operate. In the early stages
of all breakers, over-running at the front will be the only way in which air is
trapped, but it is possible that at a later stage whitecaps on large waves can be
self-sustaining while those on small waves cannot.

4. Similarity solution for a spilling breaker

The boundary conditions and extra relations needed to simplify (5) and (9)
and solve for S and и' can now be written down. It will shortly be seen that the
known velocity variation at the surface of the wave, together with the constant
density difference suggested by the discussion in the previous section, leads to
a simple similarity solution describing the progress of a spilling breaker.
It has already been pointed out that the surface velocity near the sharp crest
varies like и oc s i; and if the slope below the crest is assumed to be constant, the
same functional form will hold everywhere as a consequence of Bernoulli’s equa­
tion. It is implied that the whitecap running down this slope remains thin, and
does not itself substantially change the slope as it mixes with water from the
wave. Since the pressure at the base of the turbulent layer is gp'S cos 6, the
Bernoulli relation applied along this surface gives

i /ж2 = gps sin в —gp'S cos в, (11)

making the same quasi-hydrostatic assumption for the pressure as before. When
the last term, which allows for the effect of the whitecap on the surface pressure,
is neglected, then
и2 = 2gs sin в, (11л)

and we obtain an explicit value for the constant of proportionality in the velocity
relation.
 1719

An *entraining plume ’ model of a spilling breaker 9

It will be assumed that p' is constant, independent of s (though it can still be

a function of в). This suggests that g' = g(Ap/p) should be taken as the second
parameter (in addition to 5) which is required to define the flow. It turns out to
be more convenient to use s and g (not g') as scaling parameters, and to in­
corporate Др/р elsewhere in the analysis, but the assumption underlying this
reduction should be kept in mind. Thus let us write
8 = sk, u = gbsW, и' = дЫ и\ ( 12)

where к, U and U' are dimensionless functions. (It is worth noting that these
imply an ‘ eddy viscosity’ proportional to Succ ghi. Also if all velocities are
proportional to one another, then the ‘ inception number’ I defined by (10)
behaves like si.)
Substituting (12) in (5) gives
d(gh*kU')lds = E g h i(U '-U )
, 2 U '-U
k=- E. (13)
U'
From the definition (1) we obtain
g(Ap/p) ks cos в
BL =
0_ (U'-Ufgs
D. 2/Д,p\ E совв
ОГ 0 3 V/э / £/)(' ( *
Thus for particular values of U' and U (which specify the magnitudes of the
velocities in the whitecap and in the fluid immediately below it), the turbulent
layer thickens linearly with s, and the Richardson number remains constant.
We are therefore dealing with a kind of ‘ normal ’ flow, whose gross properties
can be described in the same terms at all s, in spite of the fact that the flow is
accelerating uniformly down the slope. Note that the linear spread implies (from
(11)) that the same form of velocity variation at the surface of the wave will
hold even when the pressure effect of the whitecap is taken into account.
The momentum equation (9) becomes, using (12) and (13) and simplifying,

^ [U'(U' —V ) gks2] -f- kgslU'-^[siU) = S2gkssin в - Sxgk2s cos в

or U'2{U' - U ) + IU'*U = (J£2sintf) £7' - ^S^U' - U) Я cos в. (15)

Equations (14) and (15), together with the empirical entrainment function
E{Ri0), give three relations from which U\ Ri0 and E can be determined as
functions o f U. It is convenient to have an analytic expression for E , and the
laboratory results of Ellison & Turner (1959) can in fact be represented to good
accuracy by the form

Calculations have been made for three slopes, 10°, 20° and 30°, and for three
density differences at each slope. The values of the profile functions Sx and S2
1720

0 04 0-8 1-2
(о) -U

(Ъ) -U (с) -U

F i g u r e 4 . Non-dimensional whitecap velocity U' plotted against the upslope velocity U

in the laminar flow below, (a) Slope of 30°. (6) Slope of 20°. (c) Slope of 10°. Plots for three
density differences are given in each case; values of Ri0 are marked on the curves. The
vertical dashed line is drawn at the value of U obtained using the Bernoulli relation, and
the inclined dashed lines show the modification due to the pressure effect of the whitecap.
 1721

An *entraining jplume' model of a spilling breaker 11

-U

F ig u r e 5 . T h e r a t io U'I\U\, a s a fu n c t io n o f U, f o r t h r e e s lo p e s a n d s e v e r a l d e n s i t y
d i f f e r e n c e s , c a l c u l a t e d f r o m f ig u r e 4 .

have been fixed at Sx = 0-80 and S2 = 1*00 throughout, as suggested in §3.

The solutions of interest are those for positive U‘ (appropriate for a whitecap
going down the slope) as a function of U (negative, since the basic flow on the
front surface of the wave is always upwards in co-ordinates moving with the
crest). They have been found parametrically by fixing Ri0> calculating E from
(16) and then solving (14) and (15) for U and U'. The rate of spread can be found
from (13), and used to evaluate the correction term in (11).

5. Results and discussion

The results of these calculations for a slope of 30° and Дpjp = (0-1,0*2,0*3)
are shown in figure 4 (a), where U' is plotted logarithmically against U on a linear
scale, and values of Ri0 are marked on the curves. The results for 20° and
Ap/p = (0-1,0-2,0-3) are given in figure 4(6), and for 10° and Ap/p = (0-05, 0-10,
0*15) in figure 4(c) (note the changed scale of U in the latter figure). The corre­
sponding values of U'/U for all nine calculations are plotted as functions of U in
figure 5. The major features of interest occur in the range of parameters plotted.
The most important property of these curves is the maximum in \U\ as U*
varies. This form implies that no solution for a downslope entraining flow, moving
against the upslope velocity at the wave surface, is possible if |U\ is too large.
But we already have an estimate of U (the non-dimensional velocity at the
surface of the wave) in (11) or (11a); in the absence of a whitecap, which is the
appropriate condition for the initiation process, this is just (2sin0)i. The corre­
sponding values, |U\ = 1-00 at 30°, 0*83 at 20° and 0-59 at 10°, are shown as the
1722

12 M. S. Longuet-Higgins and J. S. Turner

vertical lines on figures 4 and 5. The contrasts between these plots demonstrate
clearly the effect o f changes of slope and density difference, and sum up the
main results of this paper. At a slope of 30° (near the crest of the wave) real
solutions to the balance equations exist when Дp/p is in the range suggested by
experiment, and so a breaker can develop from a small disturbance and accelerate
down the slope. The minimum density difference required at a slope of 30° is
8 %, well below the experimental values discussed in § 3.
The minimum density differences for other slopes are plotted as the dashed
line on figure 3, for comparison with the experimental observations. At a slope
o f 34£°, the development of a whitecap becomes possible, according to our theory,
with zero density difference between the whitecap and the water below. As the
slope decreases, larger and larger density differences are required to produce
a self-sustaining flow, whereas the experimental curves on figure 3 show that the
density difference will in fact be smaller at small slopes. Below a certain slope
(between 18° and 24°, depending on the values we adopt for Дp/p) the upslope
drag will dominate, and the whitecap can no longer propagate. This is consistent
with the observation that an unsteady whitecap gets only part of the way down
the face o f the wave.
When the critical condition is passed, there are in general two solutions for
the downslope velocity V for the relevant value of |U\. The lower branch o f the
curves corresponds to larger BiQ, low entrainment rates and much smaller down­
slope velocities. It can also be argued that this represents an unstable state,
since the motion is dominated by gravity; a disturbance which increases the
velocity difference between the laminar flow and the whitecap will lead to an
increase in the mixing rate, larger velocities of the breaker, and a rapid transition
to the upper branch. On the upper branch, however, extra mixing produces
a substantial increase in the drag, which opposes the velocity fluctuation and
restores the equilibrium. (A similar effect was noted by Ellison & Turner in their
solution for flow reversal in a pipe.) At a slope of 30° the intersections o f the
physically relevant upper branches with \U\ = 1 show that the whitecaps will
have a downslope velocity which is much smaller than the upward velocity at
the surface of the wave below. The ratio U'j\U\ ranges between 0*12 at
Дp/p = 0*08, the lower limit for the initiation of the flow, and 0*17 at Ap/p = 0*30.
The variation of the angle of spread к (equation (13)) is also of interest, and
this is plotted in figures 6 (a) and (6) for slopes of 30° and 10° as a function of U
(both on linear scales). The corresponding values of Ri0 are again marked on the
curves. When Ri0 is large к has its largest values in the range considered, which
for в = 30° at least are probably unrealistically high; but we should remember
that these upper (dashed) parts of the curves correspond to the lower, unstable
branches of figures 4(a) and (c). As Ri0is decreased, so is к, and the values along
the more relevant full parts of the curves seem physically much more reasonable.
With these estimates of к available, we can now calculate the effect of the
whitecap on the velocity at the wave surface using (11). In the above discussion
of the marginal state, it was appropriate to use (11a), since a whitecap cannot
affect the wave before it has formed. Once this has happened, however, the
opposing upslope velocity is reduced owing to the pressure effect of the whitecap,
 1723

An *entraining plume * model of a spilling breaker 13

—U — XJ

F ig u r e 6. The angle of spread of whitecaps, calculated as functions of the dimensionless

upslope velocity U = uf\ga)t for two slopes and various density differences. Values of
Ri0 are again marked on the curves, (a) Slope 30°: (i), (ii), (iii) correspond to Дp/p = (0-1 ,
0*2, 0-3). (6) Slope 10°: (i), (ii), (iii) Дp/p = (0-05, 0 1 0 , 0-15).

which will be able to propagate more easily than it could initially. The values of
the non-dimensional surface velocity |J7|, corrected in this way, are plotted
on figures 4(a), (6) and (c) at the values of U' derived in the earlier calculations.
The crossing points (marked by bold dots) of these and the original curves thus
represent possible solutions for a downslope flow, including its effect on the velocity
of the opposing flow which it is entraining. On a slope of 30° (figure 4 a) the effect
is to increase the downslope velocity to 0-21 times the undisturbed surface
velocity (this ratio depends very little on Apjp). For 10° (figure 4c) the effect
is more dramatic, since solutions now become possible where none existed
before (with a downslope velocity 0 *16 times that at the surface of the undisturbed
wave). A possible implication is that a finite disturbance, due perhaps to a breaker
spilling down from the higher slope above, may allow the flow to go further than
would be expected from the argument based on the marginal stability of an
undisturbed flow.

6. The motion o f the front

We must now return to the question raised in the introduction: to what extent
can a theory formulated for steady motion (i.e. assuming и and u’ are independent
of time at a fixed s) be used consistently to describe a flow which is certainly
increasing in length sis time goes on? We believe that this procedure can be
justified, using an extension of the ‘ starting plume’ model proposed by Turner
(19G2) for the axisymmetric plume and applied by Tsang (1970), among others,
to the two-dimensional case.
This kind of model implies that the whole flow, the advancing front as well
as the layer behind, will remain similar as it develops in time. First, it must be
geometrically similar, so that the front also spreads linearly with distance (but
perhaps at a different rate from the plume behind). Second, the power-law
dependence of velocity on distance must be the same, with the front advancing
1724

14 М . S. Longuet-Higgins and J. S. Turner

deeper penetration

F ig u r e 7. Sketch of the ‘ starting plume’ interpretation of an advancing whitecap.

at some fraction o f the layer velocity at the same position. The flow in the layer
behind the front is just the same as that in a completely steady plume extending
to infinity in the s direction (which has been implied in all the previous discussion).
In a ‘ starting plume’ , however, this flow feeds momentum and buoyancy to
the front at just the rate required to maintain similarity; but because of the
accumulation of fluid at the front, the velocity o f advance is reduced.
For both axisymmetric and two-dimensional starting plumes in neutral sur­
roundings (i.e. with a constant buoyancy flux) the above picture has been shown
to be consistent with the detailed equations of motion of the front. That is, if
the region is regarded as a turbulent ‘ thermal’ , which is mixing with its sur­
roundings and whose properties are also being changed by the injection o f
buoyant fluid from behind, then the deduced behaviour of the velocity and
radius does indeed match that in the plume. What we must show now is that the
same matching can be carried out for a two-dimensional plume whose properties
are changing according to the similarity solution of § 4.
The model to be discussed is sketched in figure 7. The steady, established part
o f the whitecap is entraining water from below and air from above, at such a rate
that the density of the air-water mixture remains constant. This also implies,
according to the similarity solution, that the mass and buoyancy fluxes are
increasing down the slope, with 8 oc s and u' oc ght. Thus the whitecap is
accelerating uniformly down the slope, and the position of a fluid particle as
a function of time t is given by s oc t2. At the front the layer will feed fluid into
a circulating region which is like one half of a vortex pair (cf. the two-dimensional
starting plume). Extra entrainment of both air and water will certainly occur
here, and the penetration of the surface of the wave below may well be greater
than it is in the steady part of the flow.
The evidence from experiments on hydraulic jumps (§3) has suggested that
the net effect will be to keep the density of the mixture near the front constant
(at a given Froude number); in fact this result applies more particularly to the
front, rather than to the flow behind. If we now assume that g' (or g) is the major
 1725

An ‘ entraining plume1model of a spilling breaker 15

parameter determining the motion of the front, and seek a similarity solution,
then again we find u' oc gh*. Thus the front and the layer behind it have the
same velocity behaviour, essentially because constant acceleration is the only
behaviour which is consistent with the use of g (or <7') as a scaling parameter.
It may also be useful to sketch how this dimensional argument can be related
to the dynamics of the front, regarded as a line vortex, though a description
which is sufficiently detailed to include the different rates of spread and of
advance must await more experimental information. Following Turner (1960)
the impulse P per unit length of a vortex pair can be written in terms of the
circulation К and the radius r as
P o cK r. (17)

In the case of the similarity solution with constant acceleration this becomes

P ocu 's2octb, (17 a)

so that the rate o f change of momentum is

dPjdtcct4. (18)

Contributing to this change of momentum is the downslope component of the

total buoyancy per unit length of the front, which, assuming that Ap remains
constant, is
jF oc r2 oc t*. (19)

A second contribution comes from the flux o f momentum from the layer behind

i l f o c ^ o c i 4. (20)

Since (18), (19) and (20 ) have the same t dependence, the momentum equation
expressing the balance between these terms will give a consistent description of
the motion of the front.
The starting plume model, therefore, provides some justification for the kind
of theoretical model we have used to describe a whitecap. If the similarity
solution is a good approximation to the real, accelerating flow developing from
a small disturbance at wave crest, then we would expect the front also to advance
initially with constant acceleration. The velocity of the front could, however,
be less than that predicted in figure 4.

7. Com parison with observation

We have compared our theoretical model with measurements taken from the
recent film by Kjeldsen & Olsen (1971) of laboratory experiments on breaking
waves. The measurements were made in a channel about 1 m wide and more than
10 m in length, and with rigid plane beaches of uniform gradient. In the sequences
on spilling breakers the waves had generally the appearance of solitary waves.
We analysed two such sequences in detail, and the conclusions being similar in
the two cases, only one will be presented here.
We measured, frame-by-frame, the following parameters:
1726

16 M .S . Longuet-Higgins and J. S. Turner

F i g u k e 8 . Measurements of the properties I, tj and в of a whitecap (as defined in the text),

taken from a spilling breaker sequence in the film by Kjeldsen & Olsen. Note the inter­
mittent nature of the flow.

X = the horizontal distance o f the wave crest (identified by its maximum

curvature) relative to a fixed mark in the wave tank.
H = the height of the wave crest above the bottom.
I = the distance from the crest to the ‘ toe ’ of the whitecap, measured along the
surface of the wave.
в — the inclination to the horizontal of the line joining the crest to the ‘ toe*
o f the whitecap.
From these we calculated the horizontal and vertical distances (f and rj) o f
 1727

An ‘ entraining plume * modeJ о/ a spilling breaker 17

t 1----------------- 1 i----------- 1 i___________ i

Origin Ф @ G

F ig u r e 9. The measured horizontal position of the crest X , and the horizontal distance £
between the toe of the whitecap and the smoothed position of the crest, plotted for the
same film sequence as figure 8. The time intervals marked along the bottom are those used
to obtain the average values given in the table.

the toe relative to the wave crest. We also measured the maximum thickness $
o f whitecap on each frame, and occasionally the angle of the wedge, approxi­
mately $/l.
In figure 8 are plotted the quantities Z, в and rj = I sin0, as functions of the
frame number. It can be seen at once that, apart from a general increase in I
and q with time, there is also a marked intermittency. The observations showed
that after initiation of the whitecap there was a preliminary period of growth,
after which the wave crest would become more rounded. A part of the whitecap
would then be dragged over the crest, which then became sharp again. The reasons
for this behaviour are discussed in § 8 . Sometimes two wave ‘ crests’ could be
identified on a single frame and in this case the measurements of both are
plotted.
However, most of the intermittency in I shown in figure 8 is due to the oscilla­
tion in the horizontal position of the crest, not the toe. This is shown in figure 9,
where we have plotted the horizontal co-ordinate | of the toe relative to the
smoothed position of the crest. It will be seen that, after an initial jump, | in­
creases much more smoothly than f (note the change in scale). A parabola,
representing a uniform acceleration, can be drawn to fit the points reasonably
2 FLM 63
1728

18 M. S. Ixynguet-Higgins and J. S. Turner

well, apart from a brief excursion in the middle, associated with a transient
increase in the rate o f advance of the crest.
The corresponding velocities can be estimated in the absence of any precise
record o f the film speed used by Kjeldsen & Olsen, and using only the fact that
the mean phase velocity dX/dt is given accurately for a solitary wave by Scott
Russell’s formula (gH)i. According to our model и may be taken as (2grj)i very
nearly, where tj is the measured vertical distance from the crest. The smoothed
downslope velocity dl/dt = sec d(d%/dt) can be found from the parabola fitted
to the points in figure 9, and compared with dX/dt from that same plot. Denoting
the ratio {d%ldt)l{dXjdt) by R say, we obtain finally
(dlldt)lu = Rsec6{Hl2r/)i. (21 )
The results o f this calculation at three times during the run are shown in the
table. H, 7] and в were evaluated locally, using averages of 10-16 values centred
on the time of interest. This procedure is not quite consistent with the fitting of
a line of constant slope over the whole range o f X , but it is sufficiently accurate
for the present purpose. In the early stages, immediately after the initiation of
the whitecap, its behaviour is well described by the theory; it is accelerating
down the slope, and moving with 16% of the upslope velocity in the wave
underneath it. At later times, however, its velocity is an increasing fraction of u.
This could be a consequence of two effects. First, there is the influence of the
whitecap on the velocity of the upslope flow beneath it (which has been described
in the earlier theoretical sections). Second, the change in slope with time leads
to a decrease in the deduced value of q, an effect which is not included in the
simple theory, but which is discussed further below.

t
(frames
from
initiation H V
Interval of spill) R в (cm) (cm) (d lfd tyu
1 13 0078 35° 29*7 5-3 016
2 31 0187 215° 27-8 5-8 0-31
3 64 0-326 17-5° 24-6 8-2 0-42
T able 1. Summarizing the mean properties of a spilling breaker as measured from the
film at three times after the initiation of a whitecap.

8. The energy balance o f the waves

We shall now give an interpretation of the intermittency seen in figures 8 and 9 .
A rough estimate of the energy lost through breaking may be made as follows.
I f we consider the whitecap as a mass of lighter fluid resting inertly on the forward
face of the wave, the weight W of the whitecap must do work against the wave
at a rate Wvaosd, where v is the normal velocity at the interface and в is the
angle between the normal and the vertical. Since v = с sin 0, the rate of working
is с W c o b 0 sin в and the total work done during one wave period is W L cos в sin в,
 1729

An ‘ entraining plume ’ model of a spilling breaker 19

where L is the wavelength. This must be equal to the difference in the integrated
flux of energy across the two ends of a horizontal interval of length L. (We ignore
the loss o f kinetic energy through entrainment into the wedge, which can be
shown to be relatively small.) Now the integrated energy flux is simply the total
energy of the wave, which for a solitary wave of maximum amplitude in water of
uniform depth is equal to ypgh3, where h is the undisturbed depth and у is
a factor equal to about 1-0 (see Longuet-Higgins 1974). However, in shoaling
water у may be of order 5 or more (see Ippen & Kulin 1955). We have then
W Lcosdsine = —Ld{ypgh3)/dX. (22 )
But also W = ip'gl2sin a, (23)
where I is the length of the whitecap and a the angle of the ‘ wedge ’. From these
two equations we obtain
w - M e a . (M)
' sm 26 sin cl v '
Substituting the typical values
pjp' = 1-2, \dhjdX\ = 0-03, a = 20°, 6 = 30°, у = 3*0,
we find Ijh — 1*7. The total depth H (on beaches of this slope) being about 2*6h,
we have
ЦН = 0-65. (25)
We conclude that for a solitary wave to lose energy gradually to a spilling
breaker, the length I of the whitecap should be a certain fraction, about half,
o f the total height of the wave.
However, we have already shown that there is a tendency for the length of
the whitecap to increase and for the whitecap to spread down the forward face
o f the wave, whereas on a shelving beach H tends constantly to diminish. It
follows that a steady state cannot normally be achieved. The whitecap can
therefore exist only intermittently.
From the observations it appears that what actually occurs in a spilling breaker
is that, as the whitecap grows, the rate of damping described above overtakes the
effect o f wave steepening due to shoaling. The wave crest then becomes rounded.
This reduces the slope, and enables part of the whitecap to be dragged over the
crest. The wave can then steepen again, and the whitecap resumes its growth.
The process is repeated, until the wave energy is no longer sufficient to maintain it.

9. Conclusion
We have shown how several features of a spilling breaker can be described
using a model which regards the whitecap as a turbulent plume, running down
the forward slope of the wave and entraining the laminar fluid below it. In
particular, the predicted sensitive dependence on the slope, and the magnitude
of the downslope velocity, are in agreement with laboratory observations of
spilling breakers.
Our explicit similarity solutions, however, are based on the assumption of
2-2
1730

20 M. S. Longuet-Higgins and J. S. Turner

constant density of the air-water mixture in the whitecap. While this has some
support from the limited observations of air-entraining flows o f other kinds,
it is the weakest point of our present model, and further work on this aspect of
the problem would be desirable. It seems probable that the spreading of a white­
cap down the forward face of a wave is sometimes limited not by the surface slope,
but by the decrease of the density difference due to the loss of air bubbles as they
rise through the whitecap. A fully time- and space-dependent description o f the
concentration of air in a whitecap should be our eventual aim. Nevertheless,
the balance of forces we have used in this paper will still be relevant in any
theory which takes the turbulence in whitecaps properly into account.

One o f us (J.S.T.) acknowledges the support of a grant from the British

Admiralty.
REFERENCES
Ch a n , R . К . C. <fc Street , R. L. 1970 Shoaling of finite amplitude waves on plane
beaches. Proc. 12th Conf. on Coastal Engng, pp. 345-362. New York: A.S.C.E.
E llison , Т . H. & T urner , J. S. 1959 Turbulent entrainment in stratified flows. J. Fluid
Mech. 6 , 423-48.
G a l v i n , C. J. 1972 Wave breaking in shallow water. In Waves on Beaches (ed. R. E.
Meyer), pp. 413-466. Academic.
G a n g a d h a r a i a h , Т., L a k s h m a n a R a o , N . S . & S e e t h a r a m i a h , K . 1970 Inception and
entrainment in self-aerated flows. J. Hyd. Div. A.S.C .E. H Y 7, 1549-1665.
I p p e n , A. T. & K u l i n , G. 1955 Shoaling and breaking characteristics of the solitary wave.
M .I.T . Hydrodynamics Lab. Tech. Rep. no. 15.
K j e l d s e n , S. P. & O l s e n , G. B. 1971 Breaking waves (16 mm film). Technical Uni­
versity of Denmark, Copenhagen.
L a k s h m a n a R a o , N. S ., S e e t h a r a m i a h , K . & G a n g a d h a r a i a h , T. 1970 Characteristics
of self-aerated flows. J. Hyd. Div. A.S.C .E. H Y 2, 331-355.
L o n g u e t - H i g g i n s , M . S. 1973a A model of flow separation at a free surface. J. Fluid
Mech. 57, 129-148.
L o n g u e t - H i g g i n s , M . S. 19736 The sea surface. Review lecture to Royal Society of London,
5 April 1973.
L o n g u e t - H i g g i n s , M. S. 1974 On the mass, momentum, energy and circulation of
a solitary wave. Proc. Roy. Soc. A. In the Press.
M a s o n , M . A. 1962 Some observations of breaking waves. In Gravity Waves, pp. 215-220.
U.S. Nat. Bur. Standards. Circular no. 521.
R a j a r a t n a m , N. 1962 An experimental study of the air entrainment characteristics of
hydraulic jump. J. Inst. Engng India, 42, 247-273.
R a j a r a t n a m , N. 1967 Hydraulic jumps. Advances in Hydroscience, vol. 4, pp. 197-280.
Academic.
S t o k e s , G. G. 1880 On the theory of oscillatory waves. Math. <Sc Phys. Papers, 1,
197-229, 314. Cambridge University Press.
S t r a u b , L. G. <feA n d e r s o n , A . G. 1958 Experiments on self-aerated flow in open channels.
J. Hyd. Div. A.S.C .E. H Y 7, 1-35.
S t r a u b , L . G. & L a m b , O. P. 1956 Studies of air entrainment in open-channel flows.
Trans. A.S.C.E. 121, 30-44.
T sang , G. 1970 Laboratory study of two-dimensional starting plumes. Atmos. Envir. 4,
519-544.
T u r n e r , J. S . 1960 A comparison between buoyant vortex rings and vortex pairs.
J. Fluid Mech. 7, 419-432.
T u r n e r , J. S. 1962 The starting plume in neutral surroundings. J. Fluid Mech. 13, 366-
368.
 1731

Reprinted from Tenth Symposium on Naval Hydrodynamics, Hydrodynamics for Safety Fundamental
Hydrodynamics, ACR-204, Office o f Naval Research, Department o f the Navy

BREAKING W A V ES -IN DEEP OR SHALLOW WATER

M. S. Longuet-Higgins
Department o f Applied Mathematics and Theoretical Physics,
Silver Street, Cambridge, England and Institute o f
Oceanographic Sciences, Wormley, Surrey, England.

ABSTRACT It is therefore appropriate to con­

sider possible methods for studying ex­
Free gravity waves in deep water perimentally the breaking of waves in
can be induced to break in a laboratory deep water. In the present paper we
channel either by decelerating the wave- shall describe several such methods.
maker, or by use of the Fresnel envelope
at an advancing wave front. The period In addition we shall discuss a sim­
of breaking is then twice that of the ple model of spilling breakers, which
waves. Quasi-steady breakers can be promises to explain some of the observed
produced by the reflection of a bow-wave features of these mixed laminar and tur­
from the side of a ship towing tank. bulent flows.
Measurements of the mean flow field and
the turbulence show a downwards transfer WAVES IN A LONG CHANNEL
of momentum in the upper part of the
wave. A theoretical model of a spilling Given a rectangular channel, deep
breaker has been proposed which accounts compared to the wavelength to be studied,
quantitatively for some features of the in which there is a wavemaker at one end
whitecap. Some intermittency is observ­ and a wave absorber (perhaps in the form
ed, which is connected with the rounding of a shoaling beach) at the other (see
of the wave crest due to loss of energy. Figure l). How is one to induce wave
breaking at some intermediate point along
INTRODUCTION the tank?
Breaking waves are of importance to
ship hydrodynamics for several reasons.
The breaking of the bow wave near the
hull can contribute appreciably to the
ship's wave resistance Calcula­
tions of wave resistance based on the
amplitude of the Kelvin wave pattern
must also take account of energy lost
in breaking. Not least, the impact of Fig. 1 Schematic drawing of a rectan­
heavy seas against the hull or super­ gular wave channel, with wavemaker at one
structure may cause considerable damage. end.
Oceanographers have additional If one increases the amplitude of
reasons for being interested in breaking the wavemaker, keeping the radian fre­
waves. Breaking is important in the quency <i constant, then two drawbacks
energy budget of wind-generated waves, appear. In the first place one may get
and in the transfer of momentum from cross-waves, of frequency 4r& . Assum­
waves to surface currents. It also ing these to be eliminated oy technical
generates turbulence which mixes the devices such as parallel baffles, one
upper layers of the ocean and influences finds, secondly, that with a steady train
the transfers of heat and momentum be­ of waves the amplitude tends to decay
tween the ocean and atmosphere. with distance from the wavemaker owing to
the progressive loss of energy with hori­
In trying to understand the highly zontal distance X . Hence the waves
nonlinear phenomenon of a breaking, we will break, in deep water, only very
encounter the difficulty that as yet close to the wavemaker, where they are
there exists no satisfactory theory for possibly distorted by local "transients"
any of the various kinds of breaking due to the wavemaker itself.
wave. Moreover, such experimental work
has been done, has usually been con­ To avoid this difficulty we may make
cerned with waves in shallow water, use of the dispersive properties of deep-
where the bottom plays an essential water waves, as follows.
role.

597
1732

Method 1 ; Storm-building. Рог then the wave energy will tend to con­
small-amplitude waves in deep water we verge in the neighbourhood of a single
know that the group-velocity C* (the pro­ point OC , building up a "storm" in
pagation velocity for the wave energy) which breaking waves can occur. A deep-
is given by water breaking wave produced in this
manner is shown in Figure 3.

(1 )
л A»-
(where С is the phase velocity). The
longer waves (with lower O ' ) are there­
fore progagated faster than the shorter
waves (with higher O ' ). Suppose that
at time f ; » the wave maker generates a
train of waves of frequency O ', , and Fig. 3 A breaking wave in relatively
that at time this is followed by a deep water produced in a channel 1 5 ш
train of waves of frequency сГ, > о -» long and 50 cm deep, at a point 7 m from
(see Figure 2). the wavemaker, by decelerating the wave­
maker. (Method 1).

(The author suggested this method to

Dr. N. Hogben in 1959 at the time of the
installation of the wavemaker in the
mile channel at The National Physical
Laboratory. It has been used for build­
ing wave spectra, but apparently not for
studying the process of wave breaking).

Method 2 : Travelling wave fronts.

A second technique depends on the finite
bandwidth of frequencies associated with
the sudden starting or stopping of the
wavemaker.
Suppose that at time 0 tlie wave­
maker is switched on suddenly with con­
stant amplitude fcr and frequency O ' .
A train of waves is propagated down the
tank, the front moving with approximately
the group-velocity 3 /Дв' . According to
linear theory the elevation of the free
Fig. 2 Propagation diagram for two surface at position OC. and time t can
groups of waves of different frequency. be shown to be given by

The second train of waves will overtake

the first at time t and at a distance £(*>tr) ~ I С F (x ) b C(crt- eV j ) (5)
from the wavemaker, given by

(the real part being understood) where

* s (2 ) (3 is a constant depending on the form
tr—tr, t - of the wavomaker. For a wedge of depth
and angle j3 we have
Eliminating f we have
6 - 3<r . (6)
Й - - .Р a = -3 _ - . (3)
Ax
^ is a dimensionless time variable:
Clearly if the frequency O' is varied
continuously so that

dt u>

598
 1733

which eouala zero at the front itself; Figure 6 shows five records taken
and F (X j describes the wave envelope, at the same poljnt , with different wave
in fact amplitudes. As the amplitude is in­
creased, so the arrival of the wave
front is advamced. The bottom record,
with the largest amplitude shows a sharp-
'•> pointed crest typical of a wave of maxi­
mum amplitude.
(see for example ^3 )). This function is
related to the Fresnel integral ( k ) . A A film record of the advancing wave
compressed sketch of the surface eleva­ train shows the occurrence of breaking
tion С is shown in Figure U,. waves at the head of the wave train. 1
Very striking is the fact that the break­
ers occur once every two wave cycles.
This is due to the relative motion of the
waves and their envelope. The waves tend
to break when a crest passes the position
of a maximum in the wave envelope. But
the speed of a crest relative to the
velope is only tJ-с (since the
itself travels with velocity С
JL denotes the wavelength, then
■T time interval between the occurrence of
two successive breakers is

Fig. Ц Sketch of the surface that is to say two wave periods.

elevation at a fixed point X as a
function of the time t . (Method 2). It may be mentioned that the same
difference between phase and group velo­
It can be seen that the envelope function city is responsible for the leap-frogging
F ix ') at first increases exponentially, effect noticed in whitecaps in deep water
reaches the value -Jr at TT* 0 , then (6 ). It was observed from the air that
overshoots its final amplitude by about whitecaps tend to occur in sequences,
1 9 % before finally settling down to its
each whitecap being succeeded by the next
final amplitude / as x -> • This further downwind, at an interval of about
suggests that it may be possible for the two wave periods.
waves to attain their breaking amplitude
close to the wave front, while not break­ WAVES FROM A MOVING SHIP
ing, and so losing energy, in the steady
wave train behind. The time-dependence inherent in a
breaking wave may be partly eliminated
Figures 5 and 6 show an experimental by studying the pattern of waves gener­
verification of this idea, which the ated by a body in steady motion through
author carried out in a tank 330 ft long, the water. If measuring instruments are
and 6 ft wide , at the Hydraulics Research attached to the moving body, the waves
Station, Wallingford. The plotted points are then practically "reduced to rest".
show the amplitude of successive waves in Some fluctuations of course remain owing
a record of surface elevation taken at a to turbulence and possible instabilities
fixed point 1 6 0 ft from the waveraaker. in the flow.
The solid curve is the theoretical ampli­
tude as given by the envelope function With this idea in mind, the present
F (.t) The observed wave amplitude author and Dr. M. Donelan ( 7) experi­
follows the theoretical curve fairly mented with the production of breakers
well, especially ahead of the front, in the wave pattern following a small
where the waves are low. The maximum motor launch. The venture was only part­
wave amplitude is slightly larger than ly successful until we noticed that the
predicted, and occurs a little sooner. wave amplitude was greatly enhanced when
This is probably a nonlinear effect. the launch was set on a course parallel
According to Whitham's theory of non­ to an embankment or jetty (see Figure 7).
linear wave propagation (j>) the group Evidently this was due to the reflection
velocity (defined as energy flux divided of the waves from the embankment or, to
by energy density) is an increasing func­ put it another way, to the presence of am
tion of the wave amplitude. The order of
magnitude of the discrepancy between ob­
served and theoretical group velocities
in Figure 5 is which is of the same
order as would be expected on the basis 1 Shown at the Symposium
of nonlinear theory.

599
1734

Fig.5 Comparison of the observed wave amplitude (plotted points) at a point

160 ft from the wavemaker, as a function of the nondimensional time X . The
solid curve represents the envelope function F(x) of equation (7 ). Wave frequency
O' = Г ' П rad/sec.

Fig. 6 Records of the wave amplitude at a fixed point X = 160 ft, for waves
of the same frequency but different amplitude, showing effects of nonlinearity.

600
 1735

Fig. 7 Bow-wave from a moving launch breaking against an embankment.

image ship moving parallel to the first For practical reasons, the experi­
The wave pattern due to the image ship ment was transferred to the No. 2 towing
then reinforced the first so that the tank at the National Physical Laboratory,
wave amplitude near the embankment was where moreover the experiment could be
easily increased beyond the breaking repeated under controlled conditions.
point. Moreover the wave crests became Figure 8 shows the bow-wave produced by
almost normal to the embankment, making a model ship towed at a speed of 2 m/s
the flow nearly two-dimensional. parallel to the wall, and at a distance

Fig. 8 The bow wave produced by a model ship towed at a speed of 2.0 т/в in
the No. 2 towing tank at the National Physical Laboratory. Teddington.

601
1736

Fig. 9 Velocity measurements in the experiments of Figure 8 . (a) mean velocity

field, after subtraction of the carriage velocity. (b) r.m.s. levels of turbulence;
the horizontal and vertical arms of the crosses represent ( ц ,а) and
respectively. (c) the "shear" component of the turbulence: u*\r. A vector point­
ing to the right indicates a downward transfer of ЭС-momentum,

602
 1737

of 1 m from it. The velocity field was window is essential. Figure 10 shows a
measured with ал electromagnetic flow­
meter capable of recording two compon­
ents of flow, in this case the vertical
component and the horizontal oomponent
in the direction of the motion of the
carriage. The response time of the
flowmeter was of order 0 . 0 5 sec so that
fluctuations with frequency less than
about 20 c/s were measurable. By sub­
tracting out the constant velocity of
the carriage it was possible to con­
struct a picture of the mean velocity
field (l*,\r) relative to a stationary
observer and the fluctuating components
{и/,-**') . Because of cavitation round
the instrument and its support, it was
not possible to measure the velocity
very close to the free surface.
The mean velocity field is shown
in Figure 9a. Evidently this resembles
the particle velocity field in a non­
breaking, irrotational wave, that is to
say upwards on the forward face and
downwards at the rear. These velocities
are of course much less than the steady
carriage velocity, also shown in Figure
9a.

In Figure 9b can be seen the r.m.s.

levels of turbulence in the horizontal
and vertical directions: the crosses
represent ( u 7*) ,h and ( ^ . Clearly
the highest levels of turbulence begin
on the forward face of the wave just
below the mean level, where in fact the
toe of the whitecap is found. The tur­
bulence is then swept back, with the
mean current (in the steady flow refer­
red to axes moving with the wave) to­
wards the rear of the wave. Further to
the rear the turbulence starts to decay
and also spreads downwards.

Figure 9c shows the turbulent

momentum flux (u7\? ) . Conventionally
the arrows are drawn in the positive
oc -direction when the flux of OC -momentum
is downwards. It can be seen that in
the upper part of the wave, beneath the
crest, the flux is strongly downwards,
but below the level of the troughs the
flux is, unexpectedly, in the reverse
direction, though weaker. This suggests
the presence of a Jet-like current
directed towards the rear, at the level
where the momentum transfer changes sign.
Such a current may be associated with
the injection of air near the toe of the
whitecap. On the other hand, in the
wave trough at least, the vertical
gradient of the steady flow и is
negative, even in an irrotational wave,
so that if turbulence exists, a negative
value of u/v* is not unreasonable.
SPILLING BREAKERS
To study the flow close to the free Fig. 10 A sequence showing the
surface, photographic methods are clearly development of a spilling breaker (after
the most useful. For this purpose a side Kjeldsen and Olsen 1970*

603
1738

sequence of pictures of breaking waves The equations for the conservation

in shallow water, taken from a recent of mass and momentum can be shown to be
film by Kjeldsen and Olsen (.8 ). The satisfied by a "similarity solution",
waves are travelling from right to left in which the thickness of the whitecap
in gradually shoaling water, and the is proportional to the distance from
breakers are of the spilling type, in the crest of the wave. The down slope
which the motion is quasi-steady and velocity is proportional to \J . (in
the whitecap spreads rather slowly down Eulerian coordinates the velocity is
the forward face of the wave. steady; in Lagrangian coordinates each
particle undergoes an acceleration
A theoretical model of spilling parallel to the free surface). Since
breakers has recently been suggested by the velocity in a Stokes 120® angle is
Longuet-Higgins and Turner (1973). In also proportional to this implies
this the breaker is regarded as a tur­ that such a flow can start from a small
bulent gravity current riding down the disturbance with zero flux, and propagate
forward slope of the wave, and retaining with constant acceleration.
its identity because it is lighter than
the water below owing to the trapping of Solutions of this kind are possible
air bubbles ( Figure 1l). It is assumed only when the slope and density differ­
ence Eire sufficiently large; otherwise
the upwards drag dominates and a self-
sustaining flow cannot form. For a
slope of 3 0 * near the crest of the break­
ing wave, a density difference greater
than 8 % is required to sustain a steady
motion, at which point the downslope
velocity is 1 2 % of the opposing velocity
at the wave surface. Density differences
at least of this order have been observed
in hydraulic jumps (9,).
Fig. 11 Model of a spilling breaker.
The detailed examination of a film
that across the lower boundary of the sequence of breaking waves has shown a
gravity current entrainment of heavier very interesting characteristic of the
fluid takes place, at a rate governed whitecaps, namely their intermittent
by the overall Richardson number. The behaviour. Figure 1 2 (j_0) shows the
entrainment of mass has the effect of length of whitecap plotted as a function
increasing the volume of the whitecap of time, as given by the frame number in
and helping to produce an accelerating the film record of Kjeldsen and Olsen.
flow, while the entrainment of momentum What evidently happens is that from time
gives rise to a drag which reduces the to time the wave crest becomes rounded
downhill acceleration. and part of the whitecap is swept over

Fig. 12 Measurements of the length of whitecap in a spilling breaker, plotted

against frame number, showing intermittency (from Longuet-Higgins and Turner, 197J«).
 1739

the crest, eventually dissolving on the REFERENCES

rear of the wave. This rounding of the
whitecap can be attributed to a loss of 1. Baba, E. , "Study of separation
wave energy resulting from the work done of ship resistance components,"
by the wave in supporting the whitecap Mitsubishi Technical Bulletin, No. 59»
on its forward face (see Section 8 of 1969, PP. 1-16.
reference (10)).
2. Dagan, G . , and Tulin, М., "Two-
The form of spilling breakers on dimensional free-surface gravity flow
very gently shelving beaches with past blunt bodies," Journal of Fluid
gradients of 0.02 or less is known to Mechanics. Vol. 51, 1972, pp. 529-543.
resemble quite closely that of solitary
waves (see (11) ). It is therefore 3. Miles, J.W., "Transient gravity
relevant to a discussion of this wave response to an oscillating pressure,"
phenomenon that the solitary wave of Journal of Fluid Mechanics, Vol. 13» 1962^
maximum amplitude (with a crest angle PP. U 5 - 150.
of 120°) has recently been found not
to be the most energetic, for a given 4. Abramowitz, М., and Stegun, I.A.,
undisturbed water depth (1 2). Figure Handbook of Mathematiced Functions ,
13 shows the calculated energy and Washington, D.C., U.S. Govt. Printing
Office, 1964, Ю 4 6 pp.
5. Whitham, G.B., "Nonlinear dis­
persive waves," Proceedings of the Royal
Society of London A, Vol. 283, 1965,
pp. 238-261.
6. Donelan, М., Longuet-Higgins,
M.S., and Turner, J.S., "Periodicity in
whitecaps," Nature, London, Vol. 239,
1972, pp. 449-451.
7. Donelan, М., and Longuet-Higgins,
M.S., "Some observations of the field of
motion in breaking waves in deep water,"
(in preparation).
8. Kjeldsen, S.P., and Olsen, G.B.,
"Breaking waves" ( 1 6 mm film), Lyngby,
Technical University of Denmark, 1972.
9 . Rajaratnam, N., "An experi­
mental study of the air entrainment
Fig. 13 The dimensionless energy characteristics of hydraulic jump,"
£ and momentum I of a solitary Journal of the Institution of Engineers
wave, as a function of the wave amplitude of India. Vol. 42, 1962, pp. 2 4 7 I2 7 3 .
G s о. / M , where h is the un­
disturbed depth (from Longuet-Higgins and 10. Longuet-Higgins, M.S., and
Fenton 1974). Turner, J.S., "An 'entraining plume1
model of a spilling breaker," Journal
of Fluid Mechanics, Vol. 63, 1974,
momentum, as a function of wave height. pp. 1 -2 0 .
It appears then that a wave on reaching
the maximum energy will gain further 11. Ippen, A.Т., and Kulin, G.,
height only by losing some energy, per­ "Shoaling and breaking characteristics
haps by breaking or by instabilities on of the solitary wave," Massachusetts
the forward face. Vhen finally the wave Institute of Technology, Hydrodynamics
attains its greatest height, any further Laboratory Technical Report No. 15, 1955.
loss of energy causes it to jump back to
a point lower down on the energy curve, 12. Longuet-Higgins, M.S. and
possibly to a point where the wave momen­ Fenton, J.D., "On the mass, momentum,
tum is the same both before and after the energy and circulation of a solitary wave
jump (as indicated by the arrows to the Ц ," Proceedings of the Royal Society of
left in Figure 13). Thus a non-linear London A, (in press
property of surface waves may help to
produce the observed intermittency.

605
(Page 606 blank)
1740

J. Fluid Mech. (1976), vol. 73, part 4, pp. 603-620 6 03

Printed in Great Britain

Self-similar, time-dependent flows with a free surface

By M I C H A E L S. L O N G U E T -H IG G IN S
Department of Applied Mathematics and Theoretical Physics, Silver Street, Cambridge
and Institute of Oceanographic Sciences, Wormley, Godalming, England

(Received 22 September 1975)

A simple derivation is given o f the parabolic flow first described by John (1963)
in semi-Lagrangian form. It is shown that the scale of the flow decreases like
J-3, and the free surface contracts about a point which lies one-third of the way
from the vertex o f the parabola to the focus.
The flow is an exact limiting form of either a Dirichlet ellipse or hyperbola, as
the time t tends to infinity.
Two other self-similar flows, in three dimensions, are derived. In one, the free
surface is a paraboloid o f revolution, which contracts like t~z about a point lying
one-quarter the distance from the vertex to the focus. In the other, the flow is
non-axisymmetric, and the free surface contracts like t~b.
The parabolic flow is shown to be one of a general class o f self-similar flows in
the plane, described by rational functions of degree n. The parabola corresponds
to n — 2 . When n = 3 there are two new flows. In one of these the scale varies
as fV and the free surface has the appearance of a trough filling up. In the other,
the free surface resembles flow round the end of a rigid wall; the scale varies
as f-4'17.

1. Introduction
The review of free-surface flows by Gilbarg (1960) convincingly emphasizes
the scarcity of known, exact solutions to time-dependent flows with a free
surface, particularly when gravitational terms are included. Among the known
solutions are the accelerated cavities of von K&rman (1949) and Gilbarg (1962),
which have also been extended by Yih (1960); the similarity flows for the impact
of a cone on a free surface (Garabedian 1953), and some interesting examples o f
free-surface flows derived by an inverse method due to John (1953). Not men­
tioned by Gilbarg (1960) are the long-standing exact solutions known as
Dirichlet ellipsoids (Lamb 1932, p. 382), which were first derived by Dirichlet in
1860, and partly rediscovered by John (1952) and Taylor (1960). A hyperbolic
form of this solution was recently discussed by Longuet-Higgins (1972) in relation
to the appearance of instabilities at the crest of a standing wave.
Among the very simplest of non-trivial solutions to the time-dependent
problem is the flow derived in §2 below. In this, the free surface takes the form
of a parabola, whose linear dimensions vary as irz (where t denotes the time) and
which therefore reduces to a thin sheet as t -> oo. This flow was in fact discovered
by John (1953), in whose treatment, however, the nature of the solution is some-
 1741

604 М. 8 . Longuet-Higgins
what hidden by the inclusion of gravity in a non-essential way (see § 7 , below).
When viewed in the natural frame of reference it becomes clear that the flow is
self-similar, and that the parabolic surface contracts (or expands) about a fixed
point one-third of the distance from the vertex to the focus.
The purpose of the present paper is to investigate the possibility of other flows
o f a similar kind. In § 3 we derive two flows in three dimensions, in which the free
surface contracts like t~2and t~5respectively. The possible connexion of all these
flows with the Dirichlet ellipsoids is discussed in §4, and in §§6 and 6 it is shown
that the parabolic flow of § 2 is in fact an exact asymptotic form of the two-
dimensional Dirichlet hyperbola. The analysis is given in some detail because of
a possible future application to the theory of slender breaking waves (Longuet-
Higgins & Cokelet 1976).
In the second part of the paper (§§7-9) the parabolic solution is generalized
in another direction, namely to higher-order rational flows in a plane. For this
purpose, the semi-Lagrangian formulation of John (1953) proves particularly
useful. We show that for each positive integer n there exist two classes of self-
similar, time-dependent flows, with a free parameter Л. By an appropriate choice
o f Л one can generally exclude certain singularities in the flow. The case n = 2
yields the aforementioned parabolic flow and another self-similar flow confined
to the outside of a parabolic surface. The case n = 3 gives rise to a solution
representing a fluid filling up a trough in an otherwise plane surface, and another
solution representing a free-surface flow round the end of a solid wall.
АД the flows discussed in this paper are gravity-free. The paper is intended to
prepare the ground for a future study of time-dependent flows incorporating
gravity in an essential way.

2. A two-dimensional, free-surface flow

We shall first derive by a direct method the parabolic flow mentioned in the
introduction.
Consider the velocity potential

ф - % ^ - У г) + Р ^ , (2-1)

where (x , y) are rectangular co-ordinates, t is the time and P and Л are constants
to be determined. Taking the density as unity, we find that the pressure p , from
Bernoulli's equation, is given by

- ? = |a+ ( l - A ) P ^ i + ^ + / , (2-2)

where / is a function of the time only. Hence the rate of change of p following
a fixed particle is given by

^ = ^ +A (1_ A)p J L + (2Л—D ^ - f . (2.3)

At the free surface both p and DpJDt must vanish. The vanishing of (2 .2 ) and
(2 .3 ) will represent the same surface provided the coefficients of corresponding
1742

Self-similar у time-dependent flows with a free surface 605

terms are in proportion. Hence, either A = 1 or Л = 4. I f A = 1 then from (2 . 1)
we may, by choice of a different frame o f reference, take P = 0 , so
f ___ f Q
d t~ th J ~ t*
say. The free surface is then y2 = —Q/t2, which represents two planes parallel to
the x axis. Leaving aside this flow, which has been described and demonstrated
experimentally by Longuet-Higgins (1972),f consider the case A = 4. The terms
in p and DpjDt dependent on t alone will then be in proportion to the terms in x
provided that

dt t t* ’
whence we have

/ ~ ! ? 4 <»>
where Q is an arbitrary constant. The velocity potential is now

Ф= ^ г- У г)+ Р % ( 2 .6 )
and the free surface is

y2 = 3P b l v - % (2-6)
where P and Q are both arbitrary constants.
To fix the ideas, let us suppose P < 0 and t > 0 . Then setting
a = -fP /^ , с = Q/3P (2.7)
the equation o f the free surface becomes
y2 = —4a{x —\cc —ct). (2 .8 )

When с = 0 this obviously represents a parabola, with vertex at the point

(J<%, 0 ) and distance a between vertex and focus (see figure 1).
As t increases, so a, and all the dimensions of the parabola, vary as t~z. The free
surface contracts towards the origin О as the centre of similitude. This point lies
inside the parabola, at one-third of the distance from the vertex to the focus.
When t is small, the free surface (for bounded values of y) is almost plane.
When on the other hand t -> oo the free surface becomes very elongated in the
x direction, producing a thin jet of fluid, ejected to the left with velocity
<j>x = x/t+Plt*
(this quantity being negative when x < fa). When t is large, фх ~ x/t, so that
a given particle travels leftwards with almost uniform velocity. At a fixed point,
however, the velocity tends to zero.
Remarkably, (2 .8 ) shows that for any value o f Q, not necessarily zero, a free
surface may be found on which the pressure is not only constant but constant

t Note that in equation (6.4) of that paper, the last term should be <-1.
 1743

606 М. 8. Longuet-Higgine

Fi g u r e 1 . A cross-section of the free surface in the two-dimensional flow described in § 2 .

The curve is a parabola which contracts about the point 0 , lying one-third of the way from
V to F . The distance VF is proportional to t~*.

following a particle. The surface always has the form of a parabola, parallel to
the surface corresponding to Q = 0 , but travelling to the right with constant
velocity c. To produce this flow we simply reduce the pressure at infinity by the
amount Q/t4.
The velocity normal to the median plane of the jet is

Фу = ~ у11>
which is always independent of x. It follows that any line of particles parallel to
the median plane always remains so, and that the flow may be realized by the
ejection of fluid from between two approaching parallel plates. A more interesting
realization is likely to be in the jet of water formed by a ‘ plunging breaker*
during the short time that the jet is thin and almost horizontal (see Longuet-
Higgins & Cokelet 1976).
1744

S d f -similar, time-dependent flows with a free surface 607

3. Self-similar flows in three dimensions
Let (ж, у , z) be rectangular co-ordinates in three dimensions, and consider the
more general potential
Ф = ^ ( х 2- т у г- п г г) + Р ~ , (3.1)

where in order to satisfy Laplace’s equation we specify

m + n = 1. (3 .2 )
As in §2 we have
j pT p2
-v = (в-#)
where / is a function o f t only, and

^ = i [ m ( m + l ) V + » ( * + l ) s*s] + A ( l —A ) ^ + ( 2 A - 1 ) ^ j — (3.4)

On comparing coefficients of уг and z2, we see that the surfaces p = 0 and

Dp/Dt — 0 will coincide only if (m, n) = ( 1, 0 ) or (0 ,1), which cases are equivalent
to the two-dimensional flows discussed earlier; or if (m,n) = (2 , —1), (£,£) or
( — 1, 2 ). Taking first the case
(m,n) = (£,£)

Fio x j b e 2 . Sketch of the free surface in the axisymmetric flow given by ( 3 .6 ) . The surface

contracts about the point 0 , lying one-quarter of the distance from V to the focus i^.The
dimensions vary as t~l.
 1745

608 М. 8. Longuet-Higgins
and comparing coefficients of y2 and x we have
A = 2(m+1) = 3,
and from the terms independent of x and у

1 6 1* + t* •
Hence we obtain for the velocity potential

<*■»>
and for the free surface
у2 + 22 = - Щ Х- ± Р - а ) , (3.6)

where yff = - f P /* 2, c = Q/2P. (3.7)

Thus the free surface is a paraboloid o f revolution (see figure 2 ), in which the
dimensions vary as t~2. The surface expands or contracts about a point lying
one-quarter of the distance from the vertex to the focus.
If on the other hand we take (m,n) = ( —1, 2 ) we have
A = 2(w+l) = 6
4P2 0
and / = - 3 ^ +F

F i g u r e 3. Sketch o f the free surface in the flow described by (3.8). Though the flow is
three-dimensional, the surface is a parabolic cylinder. The surface contracts about a line
through 0 , lying two-fifths of the distance from the vertex to the focal line. The dimensions
vary as t~*.
1746

Self-similar, time-dependent flows with a free surface 609

Then the velocity potential is

Ф = ± ( * г+Уг- Ь * ) + Р% (3.8)
and the free surface is
z2 = - 4 y(x - % y - c t ) , (3.9)
where у = — с = Q/5P. (3 . 10 )
Thus the flow is the sum of an axisymmetric flow and o f a uniform translation
in the x direction. The free surface is a parabolic cylinder (see figure 3) whose
dimensions vary as J-5. The surface expands or contracts about a line which lies
| of the distance between the vertex and the focal line.

4. The Dirichlet ellipsoids

It is natural to discuss the relation of the flows described in §§ 2 and 3 to the
ellipsoids o f Dirichlet (1860; see also Lamb 1932, § 382). In these, the free surface
is given by
x2 y2 z2 л 44
a2+ b2 + c2 * ^ )
where a, b and с are functions of the time t only, satisfying
abc — constant = M (4.2)

and, in the absence of gravitational attraction and vorticity,

ad = bb = cc, (4.3)
where a dot denotes differentiation with respect to t. These last equations have
the integral
d2+ b2+ c2 = constant = L2 (4.4)
and the corresponding velocity potential is

* = 1 [1 х*+ ъ у2+ И - (4-6)

In two cases the integrals (4.2) and (4.4) suffice to determine the motion,
namely, first, when с = 0 and the motion is two-dimensional, in which case
b = M/cay Ь = —Md/ca2
and so d = L (1 + M 2jc2a*)-b\ (4.6)
and second, when с = b and the flow is axisymmetric, in which case
b2 = M/at b2 = Ma2!±a*
and so a = L (l + Ml4ta3)~i> (4.7)
One might expect that the solutions found in §§2 and 3 were special cases of
(4.6) or (4.7). This turns out to be untrue, but we can nevertheless show that they
are limiting cases in a certain sense.
40 FLH73
 1747

610 M .S . Longuet-Higgins
In the following sections we shall study a special two-dimensional case which
in fact corresponds to the limit c ^ c o with b2 replaced by —b2. This analysis is
given fully, since it may have later application to the theory of breaking waves
(Longuet-Higgins & Cokelet 1976).

5. The D irichlet hyperbola

We shall now describe how the parabolic flow derived in § 2 is related to a class
of exact, irrotational flows in which the free surface has the form of a variable
hyperbola.
In the hyperbolic flow discussed by Longuet-Higgins (1972) the velocity
potential is given by
ф = \А(хг-у * ) (6 . 1)
and the free surface by
- (A + A 2) x2+ (A - A 2)y2 = R2A\ (6 .2 )
where A is a function of t only, defined by

f 00 dA
А * ( 1 + № А * ) 1 ’ ( 5 ,3 )

JRand N being any positive constants. It is convenient to write

A N — a, t/N = т, (6.4)
so that a and r are related simply by

a ^ l + a1)*' (6 '61

Generally, when a > 0 , the substitution

a = cot (\P)
enables r to be expressed as
r = tan(JA )(l-Jsin*/3)i-E(/3,2-i) + lF(/}, 2~i), (6.6)
where E and F are elliptic integrals, and then a may be plotted against тwith
/3 as a parameter (see figure 4 of Longuet-Higgins 1972).f
Certain aspects o f the motion may be expressed very simply in terms of a.
Thus in Lagrangian co-ordinates we have
х = фх = Ax, У — Фу — —A y

so x = x0exp ^j A dt^ = я0^(т), у = yQexp ^ A dt^ = y0/-^(T)>

where (я0, yQ) denote the co-ordinates at time t = 0 and

F (t) = exp ^J A dt^ = exp ^ J arfrj . (6.7)

f In equation (4.6) of that reference, the modulus of the elliptic integral should be 2“^
as in equation (5.6).
1748

Self-similar, time-dependent flows with a free surface 611

F ig u r e 4 . Graphs of the functions F , dFjdr and d^F/dr*, which give the particle
displacement, velocity and accelerations as functions of the time.

Since (6.8)
J > -
we have also F = {(1 + a4)*+ !}*/«. (6.9)
The particle velocity x and acceleration x are proportional respectively to
dFjdr and d2Fjdr2. The quantities F, dFjdr and d2F/dr2 are plotted in figure 4
above.
From (5.2), the free surface may be expressed as
s 2/a2-$, 2/ 62 = 1, (6.10)

R a R w/4
where
(6 .И )
R a
b=
N {(1 + a4)i+ l}i

So the angle 2y between the asymptotes is given by

tan у = bja = F~2 (6.12)
and we have
tan2y = 2tany/(l —tan2y) = a2. (6.13)
Lastly, the radius of curvature p at the vertex of the hyperbola is given by
p = b*la=(R/N)F- 3. (6.14)
The angle 2y is shown as a function of r in figure 6 .
 1749

612 M .S. Longuet-Higgins

r
F ig u r e 6. Graph of the angle 2 y between the asymptotes of the hyperbola,
as a function of the dimensionless time r.

When r-> + 0 we have from (6.6)

Г00da 1
j a a4 3as
so a ~ (3r)-*-*oo (6.16)
and F (t) ~ exp {4(3r)*}. (6.16)
Hence x ~ а;0{1 + £(3т)?}, Nx ~ x0(3rH, ~ - z 0(3tH (6.17)
and similarly for y. So as r -> 0 the displacements remain finite, though the
velocities and accelerations become infinite. At r = 0 itself (6 .12) and (6.13) show
that the angle between the asymptotes is 90°, and from (6.14) the radius of
curvature at the vertex equals RjN.
Hence we may define r precisely, as the elapsed (dimensionless) time since the
asymptotes were mutually perpendicular.

6. The form of the Dirichlet hyperbola as t-+co

Consider now the asymptotic form, as £-»-oo, of the solution given in §6 . When
r-> oo then a-> 0 , and the appropriate expansion of the integral (6 .6 ) is
A , 1 . l a 8 1.3 a7 , 1.3.6 a11 ,л
T + ct+ 2 3 1 . 2 2 ».7 1 .2 .32s. 11
where A = Щ п , 2~i) - 2E(\ir, 2“ l) = - 0-847213. (6.2)
1750

Self-similarу time-dependent flows with a free surface 613

Writing r —A = cr (6.3)
we have a -1 = cr—^a3 +-^x 7— (6.4)
so by successive approximation
or1 = a--feo 3 + ... (6.5)
and from (5.9)
j?(r)= : 2 *(cr-^(r - 3 + ...). (6 .6 )
It will be seen from figure 4 that the linear term already gives a good approxima­
tion when r > 0-5.
Consider the form of the free surface as r->oo. Then (5.11) and (6 .6 ) show that
a ~ 2\<xRjN = alJ b ~ (2b *)-1R/N = \ (6.7)
say, so that the hyperbola becomes elongated in the x direction and correspond­
ingly thin in the у direction. Moreover the vertex ( —a, 0 ) travels to the left with
almost uniform velocity
- dajdt = - 2bR/N2. ( 6 .8 )
More accurately we have
a = a1+ a2+ ...t b = 61 + 62+ ..., (6.9)

where = Ь* = Ъ Щ 2 ^ (fU 0)

and in general an\an_x and bn/bn_x are of order a-4.

Transforming to new co-ordinates
x' = x + a1} y '= y (6 . 11 )
corresponding to an observer moving to the left with uniform velocity —dajdt,
and setting x = x' —ax in (5.10), we find for the new equation of the free surface
tx, - a 1)2 y2
a2 b2 ~
or y2 = (b2la2) [{x '-a 1)2- ( a 1+ a2+...)*] (6 . 12 )
precisely. If now x' is of order a~*alt we have to a first approximation
y2 = ( - 6f/af) (2axx' + 2a1a2). (6.13)
But writing

we have from (6 . 10)

а2 = -$ссу (6.15)
so that (6.13) reduces to
y2 = - 4 a (z'-J a ), (6.16)
which is of the same form as (2 .8 ).
It follows that the self-similar flow described in § 2 is a limiting form of the flow
near the tip of a Dirichlet hyperbola, as the time t tends to infinity.
 1751

614 М . 8. Longuet-Higgins

It is interesting that the parabolic described in § 2 , though only an asymptotic

form of the hyperbolic flow, is nevertheless an exact solution in itself. Higher
approximations to the hyperbolic flow, for large values of t , can easily be
derived by including higher-order terms in (6 .12). But the corresponding
solutions are not exact, being valid only asymptotically for large r. In the
following section of this paper we shall describe a class of rational solutions which
is indeed exact, and which effectively generalizes the solution described in §2 .

7. A semi-Lagrangian method
John (1963) has given a general method for finding two-dimensional, time-
dependent, irrotational flows with a free surface, as follows. Set x + iy = f, and
let us seek solutions in the form
S = £(M), (7.1)
where a) is a Lagrangian co-ordinate (constant following a given particle) and
£ is an analytic function of a). Thus the particle velocity and acceleration are
respectively and The pressure gradient is in the direction of the vector
(£tt + ig)> where g denotes gravity.
On the free surface oj is assumed to be real, so the tangent is in the direction
of the vector Hence the free-surface condition may be written as
£ t t '+ *9 = ir '(*>» 0 L * w r e a l> ( 7 -2 )

where r is a function which is real on the boundary. The velocity potential

% = ф + ii/r has to satisfy
dxld£ = u - i v = [ £ t(u>*,t)]* (7.3)
on the boundary, which may be done by taking

throughout the fluid. However dx/d£ must be a single-valued analytic function

of £ throughout the fluid. Hence we have the restriction
[£t(w*, t)]* is a single-valued analytic function of £(w, t). (7.4)
John (1953, p. 603) shows that all these conditions are satisfied by the
expressions

(7.6)

X = ~2t + i & + igt)^ '

where С is an arbitrary constant. We remark, first, that if the motion is referred
to a frame of reference accelerating downwards with velocity g the terms in g
disappear. The solution is therefore essentially gravity-free. Second, setting
g = 0 in (7.6) we have
Ф= = (7.6)
1752

Self-similar, time-dependent flows with a free surface 616

which is identical with (2.5) if we rotate the axes through 90° and take
C2 = - 3 P .

8. A general class o f rational flows

Consider the polynomial expression
^ „ • ^ n_1 0)11-2 • wn_3
£ = c0wn + tc1- p r - c i - p i— tc3- зд- + ...,
/« .X
(8.1)

where n is any positive integer, and c0, ..., cn and A are constants to be determined.
On substituting into (7.2), taking g = 0 and r independent of (o, and equating
coefficients of wn_1, o)n~\ ..., we obtain
A(A + 1) 6! = nc0rtx+2t y
2 A(2 A + l)c a = (w—l)c 1riA+2,
3A(3A + 1) c3 = (n - 2 ) c2rf*+2, * ( 8 .2 )

n\(nX + l)cn = 1 . c„_,r(A+2, .

which relations are satisfied by taking r = A/tA+2, c0 = 1 and
n\ 1 _ 4 /Q ox
°m ~ та! (и—m)! (A + 1 ) ( 2 Л + 1)... ( ш Л + 1)’ m .......* ’ ' '

provided that Аф — 1, — —я.-1. So, writing

t*4 = Х + гГ, A-1^ + i[& («*>t)]* « U -iV , txa> = П, (8.4)
we have from (8 .1)
X + iY = Z(Cl), U -iV = W (C l), (8.5)
where
Z = Qn+ ic 1^ n_1 —с20.п~2—гс3£}п-3 + ..., ]
( 8.6)
W= гсх£}n_1 + 2 c 2 Qn“ 2- 3icz Q.n~3 - . . . . J

To avoid a singularity, any zero of dZjdQ. in the domain of the fluid must also be
a zero of dW/dQ. This gives us in general a condition to determine A. Then the
co-ordinates (x, y) and the velocity components (u, v) are given parametrically by
x + iy = tr*xZ> u —iv = AHnA+1)W. (8.7)
In the special case n = 2 we have

Z - n , n f . Q - (A + .H 2A + i r <8-8’

л 4 1 а + (Ш )Щ Т Т )- “ 9)
Now dZjdO. vanishes at Q = —г/(A + l), but since dW/dCl has no zeros, the
branch-point cannot be annulled. Nevertheless the solution given by (8 .8 ) and
(8.9) can, for general A, still represent the flow outside a parabolic free surface,
whose scale, from (8.4) varies like ^_2Л.
 1753

616 M. S. Longuet-Higgins

У/73

Figure 6. The form of the free surface in the self-similar flow described by (8.14). The
points A and В are branch-points; A is annulled by choice of Л. The curve expands about
the origin with dimensions proportional to

In the more typical case n = 3 we have

z = as+A3+г l П2- (A +l) (2A+ 1)

Q-
(Л+ 1) (2 Л+ 1) (ЗА + 1) ,
(8.10)
3i 3i
w= Л+1
Q2+ -
(A+ 1)(2A+1)
Q-
(A + 1) (2Л+ 1),(3A +1) V

The derivatives of Z and W vanish when

1
A+ l (A + l) (2A+ 1)
(8. 11)

A+l (A + l) (2A+ 1)

Hence Q = i/( 2A+ 1) (8.12)

and on substituting in (8 . 11) we find

7A + 4 = 0 ,

so A = —y, Q t=a—li. (8.13)

The Lagrangian co-ordinates are given by

X + i F = £i3+ 7t£l2+ 49Q — (8.14)

15
1754

Self-similart time-dependent flows with a free surface 617

so the free surface, for which Q is a real parameter, is given by
78
Q 4 4 9 fl, Г = 7 а °-_ . (8.16)

This is shown in figure 6 . At large distances the tangent becomes nearly parallel
to the X axis. From (8 .11) the branch-points are given by

П = - 7 i, (8.16)
15
li 16.73 .
and Z= (8.17)
3’ 15.9
The branch-point (8.16), which is on the —Y axis, is already annulled, the other
is not. So the solution is valid for a fluid filling the space on the convex side of the
free surface. The motion is like an expanding trough. From (8.7), the dimensions
of the trough vary like ^-лЛ, that is like By reversing t we have a description
of a trough ‘ filling up*, though it must be remembered that the solution is
essentially gravity-free.

9. A second class of flows

Consider now the alternative expression
(j),n-i 0)ifi—2 ,71-3
£ - d0t(D*+id1^ - d 2- ^ 1- i d ZpKZi + . . . . (9.1)

Substituting into (7.2) and comparing coefficients as before we have

(Л—1) Adj = nd0rtx+z,
(2 Л—1) 2A<Z2 = (n—1) dyrt**2,
(ЗА - 1 ) Ш 3 = (n -2 ) d2rt*+*, (9.2)

(nX —1) nXdn — 1 . d ^ r t ^ 2,

which are satisfied by taking r = Xjtx+2, d0 = 1 and
n\ 1
dm- m\ (n—m)\( A - l) (2Л - 1) ... (m A - 1) ’ (9.3)

provided that А Ф 1, J......» -1. Now writing

t»A-l£= i + »7, <"*[&(&>*,«)]* = V + iV , txo> = П (9.4)

we have X + iF = Z(Q), V + iV = Щ П), (9.6)

where now
z = a n+id1a » - 1 - a»-* - w 3 a » -3+ ...,
(9.6)
W = £2" + (A - l)idxfl ” - 1 + (2 A- 1 ) d2П»-3- ....J
 1755

618 M. S. Longuet-Higgins
We now proceed as before. In the case n = 2,

(9.7)
W = П2+ 2»П+т;-!--.
Л— 1

The only zero of dZ/dQ, is when П = —г/(Л—1), and on inserting this value in
dWjdO. = 0 we get A—1 = 1 so A = 2 . This gives the solution of §7.
In the next case, n = 3, we have

7j — Q3 I 0 2 _______ ______ о __________ _________

Z/- U + A ~ l (A—1) (2A—1) (A —1) (2A—1) (ЗА —1) *
(9.8)

and so any root of

(9.9)

(9.10)

(9.11)

or 7Л8 —20Л2+16Л —4 = 0. (9.12)

The only real root of this equationf is A = 1*7231, corresponding to Cl = —2*644».

So from (9.4) we have £ = t~*'w Z, where
Z = Q* + 4-1488I22—1-6960Й —0-1356». (9.13)
Hence the free surface is given parametrically by
X = £ls—1-6970Q, 7 = 4-1488Q2- 0-1356, П real (9.14)
(see figure 7 ). The branch-points are given by the roots of (9.9), which are
Q = —2*б44г or —0 -2222 i, corresponding to Z ——39*14г or 0*0254г. The branch­
point at —39*14i also corresponds to a root of W, so there is no singularity there.
The branch-point at 0*0254г is an irreducible branch-point of the flow, as we
would expect from the fact that on this side of the origin the free surface inter­
sects itself. The solution will, however, correspond to a physical flow if one sheet
of the Riemann surface is excluded, say by a rigid boundary along the —Y axis,
as in figure 7 . Physically, the solution represents the flow round the end of a solid

f The complex roots of (9.12) do not represent solutions, since A would have to be
replaced by A* in (9.10).
1756

Self-similar, time-dependent flows with a free surface 619

F ig u r e 7. The form of the free surface in the self-similar flow described by (9.13). The
branch-point В is excluded from the flow. The free surface contracts like J-4"17 and on the
right becomes parallel to the X axis.

wall. As t increases to infinity the point of contact of the free surface moves out
to the end of the wall, and the free surface curls tightly round it, becoming
eventually almost plane and perpendicular to it.
Solutions for higher values of n may be investigated similarly.
It will be noticed that because the function r(o), t) is the same in (8 .2 ) and (9.2),
and (7.2) is linear in £, any linear combination of (8.1) and (9.1) will satisfy (7.2)
also, and indeed we may add expressions corresponding to different values of n.
However, (8.4) and (9.4) show that the relation between £ and Z is not the same
in the two cases, or for different values of n, so that the corresponding expressions
will not represent self-similar flows.
An exception might occur for two values of n, say n± and n2, for which
%A = w2A—1, X = ll(n2—n1).
But this possibility is excluded by the restrictions on A which are implicit in
(8.3) and (9.3).
 1757

620 M. S. Longuet-Higgins
Hence the two classes of flows given by (8 .1) and (9.1) are possibly the only
self-similar flows of polynomial type that can be readily realized. (We can of
course replace o) by any rational function of a) alone, without essentially altering
the flow.)

10. Discussion and conclusions

We have shown that the simple, parabolic flow described in § 2 may be general­
ized in two ways: first, to three dimensions, where two similar types of motion
exist, both related to the Dirichlet ellipsoids. Just as the parabolic flow has been
shown to be an exact, limiting form of the Dirichlet hyperbola, so it may be
shown that the other two flows described in §4 are exact limiting forms of axi­
symmetric hyperboloids having two and one sheets respectively.
The second generalization of the parabolic flow was to higher-order rational
flows in the plane, using the semi-Lagrangian formalism of John. For each value
of the positive integer n we found two possible flows. When n = 2 , these are
respectively the parabolic flow mentioned earlier, and another solution repre­
senting flow on the outside of a parabola. When n = 3, we obtained two new
solutions, one representing a ‘ trough ’ expanding like t&; the other,a flow curling
round the end of a solid wall. Higher values of n will yield similar flows, generally
restricted to part of the plane.
All the solutions described in the paper are essentially gravity-free, which may
limit their application to highly accelerated flows, or to special situations where
gravity can be neglected (see for example Longuet-Higgins 1972, § 8). To extend
such solutions to situations where gravity is important, and particularly to the
breaking of surface waves, is a problem to be considered in a subsequent paper.

REFERENCES
D irichlet , G. L. 1860 Untersuchungen iiber ein Problem de Hydrodynamik. Abh. Kon.
Ges. Wiss. Gottingen, 8, 3-42.
G ababedian , P. 1953 Oblique water entry of a wedge. Comm. Pure Appl. Math. 6 ,
167-167.
G ilbabg , D. 1952 Unsteady flow with free boundaries. Z. angew. Math. Phys. 3, 34-42.
G ilbabg , D. 1960 Jets and cavities. Handbuch der Physik, vol. rx (ed. S. Flugge), pp. 311-
445. Springer. (See especially pp. 15-18, 350-358.)
J ohn , F. 1952 An example of a transient three-dimensional flow with a free boundary.
Rev. Gen. Hydratd. 18, 230-232.
J ohn , F. 1953 Two-dimensional potential flows with a free boundary. Comm. Pure Appl.
Math. 6 , 497-503.
KAbm JLn , T. von 1949 Accelerated flow of an incompressible fluid with wake formation.
Ann. Mat. Pure Appl. 29 (4), 247-249.
L amb , H . 1932 Hydrodynamics, 6th edn. Cambridge University Press.
L onguet -H iggins , M. S. 1972 A class of exact, time-dependent, free-surface flows. J . Fluid
Mech. 55, 529-543.
L onguet -H iggins , M. S. & Cokelet , E. D . 1976 The deformation of steep surface
waves: I. A numerical method of computation. Proc. Roy. Soc. A (in press).
T aylo b , G. I. 1960 Formation of thin flat sheets of water. Proc. Roy. Soc. A 259, 1-17.
Y ih , C.-S. 1960 Finite, two-dimensional cavities. Proc. Roy. Soc. A 258, 90-100.
1758

Proc. R. Soc. Lond. A. 350, 1-26 (1976)

Printed in Great Britain

The deformation of steep surface waves on water

I. A numerical method of computation

B y M. S. L onguet -H ig g in s , F.R.S. and E. D. Cok elet

Department of Applied Mathematics and Theoretical Physics,

University of Cambridge, England, and
Institute of Oceanographic Sciences, Wormley, Godalming, Surrey

(Received 28 October 1975)

Plunging breakers are beyond the reach of all known analytical approxima­
tions. Previous numerical computations have succeeded only in integrating
the equations of motion up to the instant when the surface becomes
vertical. In this paper we present a new method for following the time-
history of space-periodic irrotational surface waves. The only independent
variables are the coordinates and velocity potential of marked particles at
the free surface. At each time-step an integral equation is solved for the new
normal component of velocity. The method is faster and more accurate
than previous methods based on a two dimensional grid. It has also the
advantage that the marked particles become concentrated near regions of
sharp curvature. Viscosity and surface tension are both neglected.
The method is tested on a free, steady wave of finite amplitude, and is
found to give excellent agreement with independent calculations based on
Stokes’s series. It is then applied to unsteady waves, produced by initially
applying an asymmetric distribution of pressure to a symmetric, progres­
sive wave. The freely running wave then steepens and overturns. It is
demonstrated that the surface remains rounded till well after the over­
turning takes place.

1. I n tr o d u c tio n

Breaking waves are the agent for many significant processes in the upper ocean
including the transfer of horizontal momentum from wind-waves to surface cur­
rents. Yet remarkably, one of the'most familiar and spectacular properties of the
sea surface-its capacity to turn over on itself-is one of the least well understood.
All the usual theories for surface waves - the small-ampHtude approximations of
Airy and Stokes, the nonlinear shallow-water theory and the Korteweg-De Vries
equations for solitary and cnoidal waves - are essentially approximations, valid only
when the fluid acceleration is sufficiently small compared to gravity. These approxi­
mate theories cease to be valid when the acceleration is comparable to g, or when
the surface elevation is a multivalued function of the horizontal displacement.
The small-amplitude theories can indeed be carried to higher approximations,
showing that the form of steady waves of large amplitude tends, in the limit, to the
simple comer-flow discovered by Stokes (1880 a) in which the free surface has a sharp

z [ 1 ] Vol. 350 . A. (30 July 1976 )

 1759

2 M. S. Longuet-Higgins and E. D. Cokelet

angle of 120°. B u t these theories are for steady, symmetric waves, and do not
describe the development of the flow as the surface overturns. An attempt was made
b y Price (1971) to construct a perturbation of Stokes’s comer-flow, but his expres­
sions contain singularities for both t > 0 and t < 0 (both after and before the break
point) and so do not represent an appropriate solution.
An oft-quoted theory for wave breaking is that of Biesel (1952) who carried to
a second approximation Miche’s first-order Lagrangian solution for waves
approaching a plane beach. In this theory, the free surface appears to develop a cusp,
and then intersects itself, forming a closed loop. I t hardly needs to be pointed out
that the approximation ceases to be valid long before the cusp is formed, and more­
over that the looped surface is topologically impossible, if the fluid is to He always on
one side of the free surface.
The mathematical difficulty of the problem arises essentially from the need to
satisfy the condition of constant pressure (which is generally non-linear in the
velocity) at a free surface which not only is unknown, but whose form is highly time-
dependent. Since no appropriate analytical theory has yet been suggested, it is
natural to seek whatever guidance can be got from numerical computations. A start
in this direction was made by Chan & Street (1970), who employed the so-called
‘ marker-and-celT technique. The flow being assumed two dimensional (that is,
dependent only on the coordinates ж, у and on the time t), the (x , ?/)-plane is covered
b y a rectangular grid, and the velocity components are computed at fixed points
within each cell. The development of the flow is followed in small time-steps, using
difference-equations to represent the conservation of mass and momentum (the
fluid being assumed incompressible and inviscid). For waves in water of finite
uniform depth, this method achieved some success. The authors were able to
demonstrate, for example, the steepening of the forward face of the wave almost up
to the instant when the free surface becomes vertical. A description of the over­
turning was not achieved, though the computational procedure might, with further
trouble, have been modified so as to cope with a multivalued surface elevation. It is
notable that in these computations no use was made of the vorticity equation; it
was assumed simply that the initial flow was irrotational. A t subsequent times the
flow was not quite irrotational.
Now all numerical computations using a rectangular grid have an important
practical limitation. To obtain acceptable accuracy, the grid-spacing must be
reasonably small, at least compared to the local scale of the flow or the radius of
curvature of the free surface. B u t as the spacing is reduced, the number of grid
points has to increase like N 2, where N is the order of magnitude of the number of
points in each direction. In this way, the quick-access storage of even the larger
computers is rapidly used up, and the necessary computation time also exceeds
practical limitations.
To overcome these essential difficulties we have developed and tested a quite
different numerical method, which we describe in the present paper.
For oscillatory waves, the diffusion of vorticity into the interior may be quite slow
1760

The deformation of steep surface waves. I 3

(see Longuet-Higgins 19536 , i 960 ). Hence, up to the point of breaking it may be

a good approximation to neglect viscosity and assume the motion irrotational. This
leads to a velocity potential satisfying Laplace’s equation (V20 = 0 ). Now within
a closed boundary such functions are uniquely determined by their values on the
boundary itself. Moreover the time-evolution of the flow is uniquely determined by
the pressure applied at the moving surface, through Bernoulli’s equation. Hence
the problem is in reality a problem in one space dimension, not two.
Using this insight, and assuming the motion to be periodic in space (though not
necessarily in time) we first transform the coordinates so that the domain of the
fluid lies within a simple closed contour. The boundary С corresponds to the free
surface. We then show that for particles on the surface, the kinematical and
dynamical conditions can be very simply expressed in terms of the rates of change
of ф and of the coordinates following a fixed particle. It follows that by proceeding in
small time-steps we are able to follow the potential ф from one position of the free
surface to the succeeding position. This does not alone solve the problem, however.
From the values of ф on the new position of the surface we can indeed calculate the
tangential component of velocity Ьф/ds. But for the next step we need also to know
the normal component дф/дп. The crucial step is to determine дф/дп at each point
of the boundary by the solution of an integral equation (equation (4.5) below). Once
this is done, with sufficient accuracy, the time-stepping can proceed.
It will be apparent that this mixed Eulerian-Lagrangian method has the out­
standing advantage that the independent variables are all evaluated at the free
surface. The velocity potential and its derivatives in the interior are not used,
though these can easily be calculated from surface values by Cauchy’s theorem.
Hence the number of independent variables in the computation is of order N t
not N2, and for given storage and machine-time much greater accuracy can be
obtained.
The method is also flexible. Not only free waves can be computed, but also waves
to which is applied any smooth distribution of pressure at the free surface.
One welcome but unexpected advantage is also the tendency of the marked
particles to concentrate near regions of sharp curvature, where accurate resolution
is most needed.
The accurate programming of the computation is an essential and by no means
trivial task, described below in §§ 5-8. As will be seen, one of the keys to success was
to devise a simple smoothing technique, which eliminated certain unwanted
instabilities.
The method was first tested on the special case of a free progressive wave of finite
amplitude, for which accurate and independent methods of computation have
recently become available. As will be seen from §9, the present time-stepping
technique was in excellent agreement with the independent computation. Then as
a first application of the method, we describe in § 10 the development of a free wave
which by an initial application of surface pressure is raised to an energy level
exceeding the maximum for a steady, progressive wave. Figures 6-12 show how the
1-2
 1761

4 M. S. Longuet-Higgins and E. D. Cokelet

wave develops in time, ultimately turning over and plunging towards the forward
face of the wave.
Some of the implications of this straightforward calculation are discussed in § 11.
This paper is intended to be the first of a series in which the present method of
computation is employed as one tool in a systematic investigation of free surface
flows and breaking waves.

2. B asic eq u a tio n s

Let (x,y) denote rectangular coordinates with the ж-axis horizontal and the
у-axis vertical, as in figure 1. The motion is assumed to be periodic in the ж-direction,
with period
L = 2п/к. (2 .1)
The fluid is inviscid and incompressible, and the origin is taken to be in the mean
surface level. The motion may be assumed to be started from rest by conservative
forces, so that it is irrotational at all times; any slow diffusion of vorticity inwards

F igube 1. Choice of axes and notation for space-periodic wave motion

in deep water.

from the boundaries is neglected. By a choice of reference frame, the ж-averaged

horizontal velocity й at some given depth у may be taken as zero and then, since the
motion is periodic and irrotational, й must vanish at all other depths у (beneath the
lowest point of the free surface). Thus
и= 0 (2 .2 )
at all depths, and also at all times, since a non-zero value of 0й/Ыwould imply an
infinite rate of input of horizontal momentum.
It is convenient to choose units of mass, length and time so that the density p is
unity and also so that
<7=1, к — i, (2.3)
1762

The deformation of steep surface waves. I 5

where g is the acceleration due to gravity. By (2.1) we have then

L = 2k. (2.4)

We have now the following equations for the velocity potential ф. Since the fluid
is irrotational and incompressible

и = Уф, У-и = У2ф = 0. (2.5)

Since V0 is periodic and Уф = 0 , it may be shown (Longuet-Higgins 1953 a) that at

great depths V0 vanishes exponentially:
\Щ ^ A(x,t)ev. (2.6)

At the free surface у = y8(x, t) we have the kinematic conditions

Da; _ 9ф Dy _ Ъф
(2.7)
Dt дхг Dt dy*

where g = ! + v^.V, (2 .8 )

which denotes differentiation following a given particle. We have also the dynamical
condition derived from Bernoulli’s equation, namely that
Ъф1Ы= - р 3- у - \ ( Ч ф ) \ ( 2 .9 )

where padenotes the pressure applied at the surface. From this we can immediately
derive the rate of change of ф following the motion, namely

D ф1Ы = - р а-у+\(Чф )\ (2 . 10)

The difference between the right-hand sides of (2.9) and (2.10) is simply a change of
sign in the last term.

3. T ra n sfo rm a tio n o f co o rd in a te s

Since the motion is periodic in x with period 2n we may write

reI<? = £ = e“ 4* (z = z + iy), ( 3 .1 )

where f is a new complex variable, analytic and single-valued everywhere inside the
contour С which corresponds to the fluid surface (see figure 2 ). (ir,6) are polar
coordinates in the f-plane, and we have from (3.1)
г = е*, у = In r,
(3 .2 )
в = —x, X —— 6 .

All points at infinite depth in the (x>y) plane are transformed into the origin О in
the £-plane.
 1763

6 M. S. Longuet-Higgins and E. D. Cokelet

Let us write
Х = Ф+ (3.3)
for the complex velocity-potential in the physical (x, y)-plane. Then x is generally
analytic and single-valued inside C. Moreover by equation (2 .6) it follows that as
C->o
X ~ A g. (3.4)
Hence x is analytic and single-valued everywhere inside C.

if- plane

F ig u r e 2. One wavelength in the z-plane transformed to a closed domain in the £-plane by

equation (3.1). The free surface is transformed into the contour C.

At the free surface we have from (2.7) and (3.2)

Dr _.Dy
,fy/ дф__ 2дф
Di D* 0r'
(З.б)
m Bx дф_дф
T>t Ъх дв
Lastly, since
ш у (дфу _ Id* * _ К № | а _ А ( Щ \ П Щ 2]
\0ж/ \ду) Idz |d^?I |d^| LvW \r0^/ J
the dynamical condition (2 . 10) becomes

(3.6)

4. T he D ir ic h l e t problem

Suppose that at some initial instant t = t0 we are given the velocity potential ф
throughout the fluid, and hence the value of ф and its derivatives both inside and
on the contour C(t0). Let (г, в) denote the (Lagrangian) coordinates of a particle on
1764

The deformation of steep surface waves. I 7

C(t0). Then equations (3.5) will determine the position of the same particle a short
time dt later. Similarly (3.6) will determine the value of + df) on the new contour
C(£0 + df). By considering adjacent particles, and differentiating along the surface
we can then obtain the tangential component of velocity дф/дв. But this does not
immediately determine the normal component дф/дп, which is also needed for the
step afterwards.
The problem of determining the normal component of velocity at the boundary
is equivalent to the Dirichlet problem of finding the normal gradient of a function ф

F ig u r e 3. Definition of variables for the Dirichlet problem of § 4 .

whose values are given on a closed contour C, and which is harmonic (V20 = 0)
everywhere inside G. We may formulate the problem as an integral equation as
follows.
Let (s,n) be tangential and normal coordinates at a typical point P on the
boundary (see figure 3), and let (E, oc) be the polar coordinates of P with respect to
an arbitrary point Q{r0, в0) in the interior. Let

S= ^nB (4.1)

so V2S = 0 . Then by Green’s theorem we have

( i '2 )

Denoting ф(г0, 6Q) by ф0) we have then

2пф0 = ^ фdoc—j ^ ln .R d s . (4.3)

 1765

8 M. S. Longuet-Higgins- and E. D. Cokelet

Letting Q(r0,60) approach C, we have in the limit

Jc 0 da = 7u0o+ p J 0 da, (4.4)

where P denotes the principal value (C is assumed to be a smooth at Q). Hence

altogether from (4.3) and (4.4)

J c S b -Bd4 - P j V a - ^ 0. (4.5)

Since the right-hand side involves only the values of ф on G, which are known, and
since JR, a are determined solely by the shape of C, equation (4.5) is an integral
equation for (дф/дп), with a singular kernel In J?.
It remains to express the time-derivatives of г, в and фin terms of the tangential
and normal derivatives of ф. We have in general

0 J . J
0; = c o s ^ +sm % ’
13 • *э в9
7 S0 = sm% - CO8% ’
dr ■ * = т—.
se
where sm/?
с03^ = э ? os

Hence equations (3.5) and (3.6) become

Dr 0дгдф двдф
d t = r a«a7+r aia^’
m ,г е м дгдб
(4.6)
d t. ~ Г dsds ГЪзЪп*

)'}■)
Equations (4 .5 ) and (4 .6 ), together with initial conditions, are the basis for the
following computations.

5. S o lu tio n of the in t e g r a l e q u a t io n

The values of ф, Ъф/ds and дф/дп are to be evaluated at a finite number N of points
on the boundary which will correspond to fixed particles. We label these with the
suffix j , where j = 1, 2 ,..., JV. The corresponding values of г, в are similarly labelled
rp 6j. We will suppose that Q lies in turn at (r{, where г = 1, 2 , ...,N also. Hence
It, a, s take the values JRij} ccip s{j. The integral on the left of (4.5) can be approxi­
mated as the product of the {N x 1) matrix (дф/дп)] multiplied by an (N x N) matrix
1766

The deformation of steep surface waves. I 9

of coefficients, say Ay. The integral on the right can be approximated by a quantity
Hence we have N linear simultaneous equations of the form
Ау(Ъф1Щ = Bi (i = 1, ...,N) (5.1)
to be solved for the values (дф/дп)^
The success of the method will depend critically upon the accuracy with which
the integrals in equation (4.5) are approximated by linear sums. Let г be fixed, and
let us take Q(rit 6^ as origin of s, writing si}- —s(j_i)mo<1N. Since In В is singular both
at s = 0 and 5 = sN, but In (i2/s) or In [BI(sN—5)] is not, we write the integral on the
left of (4.5) as follows

f ^ \ n R d s= PI^ h ( ^ ) d s + f Sl^ ta s d s + f”"^ln.Rds

Jc&n Jo Jo Sn Jsx dn

+0 b ( ^ ) d*+ 0 ,” ‘*»-*>d*- <“ >

where sL is the arclength at a point close to s — 0 and on the positive side, and sMis
a similar arclength just short of the other end of the range of integration. (In our
calculation we take sL to be the larger of s6 and 0.1sN and sM to be the smaller of
&and 0 .95^.)
We denote the integrals in (5.2) by Ilt I2> Is respectively. Iv J3 and / 4 are
no longer singular, and may be calculated by approximating the integrand over
each sub-interval (sj} sj+1) by a 4-point Lagrangian polynomial. More specifically,
a 4-point, closed-range quadrature formula is used at the end-points of each
integral, and a 4-point, open-range quadrature formula is used for the remainder
(Buckingham, 1962). The local errors in this method are of order (As)6, where As is
the maximum arclength between two adjacent points Sj.
The remaining integrals 72 and I6must be handled differently. I2for example, can
be written as the sum of integrals of the form

Each term in the integrand can be evaluated exactly. To find the derivatives of
дф/дп with respect to s, we approximated (S^/dTi),- by a fourth-order Lagrangian
interpolation polynomial over the 5 points centred on j, and differentiated this.
The integral I6is handled in a similar way, except centred on the point (j -f 1). This
technique gives local errors of order (As)6In (As).
To calculate the arclengths sp the quantities and](fy —Znj/N) as functions of,?
(considered as a real variable) were approximated by cubic splines (see Ahlberg,
Nilson & Walsh 1967). From these can be calculated dr/dj, dd/dj and hence

g - [ ( D 4 $ 7 .
at each point j. Equation (5.4) was then integrated by'Simpson’s rule, giving s} at
 1767

Ю M. S. Longuet-Higgins and E. D. Cokelet

each point, with local errors of order (As)4. The tangential derivative dф|dj was
calculated similarly. Hence we found

(5.5)
© ,-© y ® /

with errors of order (As)3.

$ H »J
To evaluate the integral on the right-hand side of (4.5) we need to know ocj at each
point of the contour. In general we have
, lr* s i n ^ - r * s i n & \ .. # ..
cl, = ос, t = a r c t a n ---------------^ | (j ф г) ( 5 .6 )
} \7y COS d j — T i COS 6 J ' ' '

and in the exceptional case we may use

CLi = CL. = arctan \cos (dr/dj )f-r s m 0 {dd/dj )i / • (6-7)

Because the curve is a simple closed contour we have aN = с^+я. The right hand
side of (4.5) can now be written

pL aS ds= pL aHjdi {5-8)

and integrated using Simpson’s rule, since we know the integrand at equally spaced
values of j . This gives local errors of order (As)4.
The above method of solving the integral equation was programmed in F o r t r a n i v
and solved on the I.B.M. 370/165 digital computer at Cambridge University. With
N = 60 points along the free surface, the central processing time was 3.4 s and the
core-store 150 К bytes for a double-precision calculation (17 significant figures).
About 80 % of this time was used to set up the 60 x 60 system of linear equations,
and 20 % was used to invert the matrix. The errors in дф/дп were found to decrease
like (As)3, which is consistent with the errors in the least-accurately determined
integral. The computation time varied roughly as N2.

6. T i m e -s t e p p i n g

For numerical purposes, equations (4.6) may be regarded as first-order ordinary

differential equations in t. For most of the time-stepping we used the Adams-
Bashforth-Moulton (A.B.M.) scheme which, applied to an equation of the form

dy/dt = f(t) (6 .1 )
1768

The deformation of steep surface waves. I 11

is as follows:
У\р = Уо+Y i (55/o - 59/_1+ 37/ _ 2 - 9/_3),l
д , (6 -2 )
2/ic = 2/о + 2^ ( 9 /ip + 1 9 / 0 — 5 / _ x + / _ 2) . J

Here A£ denotes the short time interval, f n denotes f(t+nAt) and ylp, ylc denote the
‘ predicted* and ‘ corrected * values of уг (see Acton 1970). The method is fourth-
order, local errors being 0(At)b, but requires only two evaluations of / at each time
step. Applied to equations (4.6) this means that we have to solve the integral
equation for дф/дп only twice for each time-step.
Since the A.B.M. method needs information from three previous time-steps, a
fourth order Runge-Kutta (R.K.) technique was used to make the first three time-
steps from the initial conditions (see Gerald 1970). This method uses no information
about previous time-steps, but takes four mini-steps forward from the current time.
A weighted average is then used to calculate the function at the new time. The U.K.
method requires four evaluations of the time-derivative at each step and is thus
twice as time-consuming as A.B.M.

7. C h ecks of accuracy

The accuracy of our numerical solution to the integral equation (4.6) was tested
by calculating, at each time-step, the value of

This represents the total ‘ outflow5across С in the £-plane, or across G' in the z-plane,
and it is clear from Green’s theorem, or from considerations of continuity, that
Q should vanish identically. The numerical value of Q was found by writing

Ч Ж *
and integrating by Simpson’s rule.
As further checks we calculated in a similar way the mean level
1 f 2-rr 1 /*2тг
= 2J 0 y^ = - T n ) a Ыгйв (73)
which also should vanish, and the potential and kinetic energies

г - я Г № ‘ - 5 } > ' , ' а'’ < ,'4 )

“ d T ' i r f - . - s j / s 8* <’ •«
 1769

12 M. S. Longuet-Higgins and E. D. Cokelet

(the last step follows from Green’s theorem). When there is no input of energy by
normal pressure at the surface, then E = (T + V) should remain constant.
All these tests verified the accuracy of our solutions, generally to an acceptable
number of decimal places (see below, §§ 9 and 10).
The maximum allowable step At should ideally be determined by considerations
of accuracy and stability. There being no theory directly applicable to the system
(4.6) we have adopted a practical criterion similar to that of Chan & Street (1970)
(though their method of dividing the fluid region into cells is quite different from
ours). Thus At is restricted so that at each step no fluid particle moves more than
a distance (As)min, the minimum arclength from a particle to its nearest neighbour
on C. If during the calculation this limit is exceeded, then At is halved, and the
time-stepping is restarted with three U.K. steps followed by A.B.M. steps as before.

8. I n s t a b il it y a n d sm o o th in g

In nearly all computations, the wave profile, after a sufficiently long time,
developed a saw-toothed appearance, in which the computed positions of the
particles lay alternately above and below a smooth curve (see, for example, figure 4).
The cause of the instability is unknown. Tests showed that, once started, the rate of
growth of the instability, per unit time, was independent of the number of time-
steps. Hence it is not due simply to rounding errors. The growth may be partly
physical, being similar to the growth of short gravity-waves by horizontal compres­
sion of the crests of longer waves (see Longuet-Higgins & Stewart i 960 ; Phillips &
Banner 1974). In reality these instabilities are partly damped by viscosity, which
we have neglected.
The instability was effectively removed by the following procedure. A function
f(x) defined at equally spaced points xi (j = 1, 2 ,3,...), and in which alternate points
lie on a smooth curve, can be locally approximated by two polynomials, say

h(x) = (а0+ а1ж+ а2жа+ ---+ апа?п) + ( “ 1)*(^> + &1ж+**-+&л-1жП“1)- (8-1)

The first bracket represents a smooth mean curve, the remainder a quantity which
oscillates with period 2 in j. The coefficients a0, Oj,... and 60, bv ... may be chosen
uniquely so that h{x) =f{x) exactly at (2n + 1) consecutive points xjy say (j - n) to
(j + n) inclusive. As a smoothed function we can then take the even part:

Цх) = аъ + а1х + а2хг+ ... + anxn. (8 .2)

In the case n - 2 this leads to the 5-point smoothing formula

7j = A ( +Щ + - Л+2) <8-3)
and when n = 3 we find in a similar way

Si — t z ( —f j - 3 + 4/y-i + 16Л + 9/я-1 ~ fj+ з)« (8.4)

1770

The deformation of steep surface waves. I 13

Figube 4. Comparison of the profile of a steady progressive wave in deep water (8 = 0.80)
computed by Pad6 approximants from Stokes’s series (smooth curve) and the corre­
sponding time-stepped profile (unsmoothed) represented by the oircles. N = 30. The
profiles are compared at times (a) t = 0, (b) t = (c) t = $тг, (d) t — n. Note the growth
of the instability at t = n .
 1771

14 M. S. Longuet-Higgins and E. D. Cokelet

Both formulae were tried for smoothing the functions rp 6j and фр at every 5 , 10 or
20 time-steps. Both worked well and gave a smooth profile with no appearance of
the small-scale oscillations. For certain reasons, as stated in the following section,
the formula (8.3) was generally preferred.

9. T est on a pr o g r e ssiv e w ave

An important and essential test of the method was as follows. For initial conditions
we took a symmetric, progressive wave of finite amplitude in deep water (see
figure 4). A very accurate method of computing the wave profile, based on Stokes’s
expansion (1880 b) has recently been developed by Schwartz ( 1972, 1974). This
nees Pad6 approximants to sum otherwise divergent series. In fact we adopted
a modification of Schwartz’s method due to Cokelet ( 1975) in which the expansion
parameter is taken as
usb/c4, (9.1)
where gcre8t and trough denote the particle velocities at the wave crest and the
wave trough, in a frame of reference moving with the phase-speed c.
We selected a wave corresponding to 8 = 0.80, whose crest-to-trough height is
about 90 % of that of the highest wave in deep water. As initial values for the
numerical integration, we inserted the computer coordinates of the wave profile
(see figure 4) and the corresponding values of the velocity potential ф, in the frame
of reference for which the deep water was at rest, and therefore the motion generally is
time-dependent.
In figure 4 the initial wave has its crest at x = n, and the wave progresses from
left to right. At time intervals of two wave profiles are plotted: the continuous
curve represents the Pad6-approximated profile, and the plotted points represent
the positions of surface particles found from numerical time-stepping by the present
method (with N = 30).
The plotted points in figure 4 are without smoothing, and the growth of the
saw-toothed instability from the third to the fourth profile is apparent. Actually,
the unstable wave is found to conserve total energy to one part in 100 up to t = n,
when the computations break down.
Figure 5 now shows the computed points, compared with the steady-wave profile,
at time t = 2k, when smoothing has been applied after every 6 time-steps. As can
be seen, the profiles are indistinguishable except to the sharpest eye. For this
particular plot the 7-point smoothing formula (8.4) was used. The flux Q of equation
(7.1) varied between б x 10" 6 and 4x 10"3, hovering around lO" 4 for most of the
calculation and showing no tendency to increase. The mean surface displacement y9
varied from 3 x 10-5 to 1 x 10-3. The total energy E lay between 0.06938 and
0.07007 (the value from Pad6 -approximants is 0.06996). The wave energy decreased
by 0.14 % at each smoothing with the 7-point formula, compared with 0.06 % using
the б-point formula. Although asymptotically less accurate, the б-point formula
1772

The deformation of steejp surface waves. I 15

seemed to perform better in practice. When the number N of integration points was
raised to 60, only 0.003 % of the energy was then lost at each smoothing.
The numerically calculated profile also remained very closely in-phase with the
Pad£-approximated profile, using the calculated phase speed of the steady wave.
These checks suggest strongly that the numerical method remains accurate over
the time-scales in which we are interested, which are of the order of one wave period.

F igure 5. Comparison of the profile of a steady wave calculated from Stokes’s series (smooth
curve) and the corresponding time-stepped profile, with smoothing, represented by
circles, at time t = 2n. (This is twice the maximum duration in figure 4.) N = 30; 7-point
smoothing every 6 time-steps. The two profiles are indistinguishable.

10 . P l u n g i n g breakers

So far we have described a numerical method of calculating the form of periodic

gravity waves in deep water, when these are subjected to an arbitrary application
of pressure pBat the surface; and we have shown that in the special case of free,
progressive waves (pe = 0 ) the method gives results in very close agreement with
known theory. In this section we shall show that the method can also yield an
accurate description of highly nonlinear and unsteady waves.
There is one feature of the fluid motion which is beneficial to our numerical
solutions. As mentioned in §8 in connection with the instability, the straining
motion in the primary wave results in horizontal convergence near the wave crest.
Two adjacent fluid particles will move closer together as they pass up the forward
face towards the crest of the wave. This causes a bunching of the marked particles
used in the computation, which has the desirable effect of improving the resolution
near a region of sharp curvature. This will be evident in some of the profiles to
be shown.
Unlike the situation for steady, symmetric waves, in which the flows are described
by a restricted range of parameters corresponding to wave height, wavelength and
mean depth, unsteady waves have a much larger diversity, corresponding to various
initial wave profiles and particle velocities, and to various distributions of applied
pressure at the surface.
We present here the results for waves developing from one initial condition and
with one particular form of pressure forcing. As initial conditions we take the
 1773

16 M. S. Longuet-Higgins and E. D. Cokelet

symmetric progressive wave of § 9, for which $ = 0.80. The energy E0 of this wave
is 94 % of the maximum possible energy Emof a progressive wave of the same
length in deep water (which corresponds to £ = 0.92; see Longuet-Higgins 1975 a;
Cokelet 1975). At time t — 0 , a wave crest is at x = n, and the wave progresses from
left to right (see figure 6 ). The pressure applied at the surface is taken to be
[j30sin$sin(a; —ct) (0 < t
(10.1)

Thus p8 consists of a simple sine-wave progressing with the phase-speed c, in

quadrature with the surface elevation. The amplitude of ps increases from zero
when t — 0 to pQat t = \n, and then falls to zero when t = it, remaining zero there­
after. The forcing is thus limited to one-half of a wave period, after which the wave
runs free.
We show the results for four different pressure amplitudes: p0 = 0.0729, 0 .100,
0.126 and 0.146. To resolve the flow near the crest, the number of points N was

У
1

0 x

-1

1
(b)

F io u b e s 6 ar-c. For description see opposite.

1774

The deformation of steep surface waves. I 17

F igu re 6. A time-sequence of profiles for the pressure amplitude p0 = 0.146 at times (a) t = 0,
(b) t = *7t, (o) t = fit, (d) t - л, (e) * = ( /) « = К (?) f = fit. = 60; 6 -point
smoothing every 6 time-steps. The surface pressurep 0 is applied tilH = я; then the wave
runs free.

increased to 60, since 30 points proved too few. The 5-point smoothing formula was
used every 5 time-steps.
Figure 6 shows a sequence of wave profiles at various times for p0 = 0.146. The
dimensionless time interval between plots is except for the last three plots,
which are separated by a time interval of %n. The wave begins at (a) (t — 0 ) as a

2 Vol. 350 . A.
 1775

18 M. S. Longuet-Higgins and E. D. Cokelet

steady wave and slowly increases in amplitude as it moves to the right. The pressure
forcing is removed at (d) and the wave is thenceforth free. The front face quickly
steepens, and at (g) the crest is overhanging.
Figure 7 is a comparison of the profiles of the waves corresponding to p0 = 0.0729,
0 .100, 0.126, 0.146 at the same time t = 5.066 after applying the pressure. It is clear

F i g u r e 7 . A c o m p a r is o n o f w a v e p r o file s a t t h e s a m e t im e t = 6 .0 6 6 a f t e r f i r s t a p p l y i n g t h e
s u r f a c e p r e s s u r e , (a ) pQ = 0 .0 7 2 9 , (b)pQ = 0 .1 0 0 , (o ) p0 = 0 .1 2 6 , ( d)p9 = 0 .1 4 6 .
1776

The deformation of steep surface waves. I 19

that the larger the pressure forcing, the farther towards breaking are the waves.
Notice that the profiles remain smooth and free from instability.
Figure 8 shows a close-up of the overturning crests, in the case p0 = 0.0729. This
is a series of plots of a small region near the crest of the wave, as seen by an observer
moving horizontally with the speed of an infinitesimal wave of the same wavelength
(i.e. (ж—i) is plotted). The positions of the particles have been marked with small
circles, and each profile drawn by connecting adjacent particles with straight lines.

(x - t )
F ig u h e 8. Close-up of the wave-crest at successive times, with a plotting interval of л/160.
p0 = 0.0729; 6.596 < t < 6.890.

On the other hand the imaginary lines joining consecutive positions of each small
particle give an idea of the particle trajectories, in this frame of reference.
Similar profiles for the cases p0 = 0 .100, 0.126 and 0.146 are shown in figures 9-11.
Perhaps the most remarkable feature of figures 8-11 is that the crest does not
tend to develop a sharp angle, as for the highest steady wave, before overturning
2-2
 1777

20 M. S. Longuet-Higgins and E. D. Cokelet

takes place. Instead, a smooth jet of fluid is ejected from the forward face of the
wave.
Our method of computation appears valid till well after the surface is vertical.
However, the curvature at the forward tip of the wave appears to increase in time,
so that the computations cannot be continued beyond a certain point.
Figure 12 shows the complete wave profile corresponding to the last wave crest
plotted in figures 8-11. We notice a tendency for the more highly forced waves
(12 c) and (12 d), which are also more energetic, to start breaking at lower values of
the wave height.
Finally in figures 13-16 we show the time-variation of the kinetic, potential and
total energies, respectively, for each value of the forcing pressure p0. The oscillations
between kinetic and potential energy suggest that there are standing-wave com­
ponents in the motion. The total energy, however, increases smoothly to a constant
value. Also indicated in figure 16 are the values of the final energy E divided by the

(x -t)

F ig u r e 9. As figure 8, but with p 0 = 0.100; 6.046 < t < 6.282.

1778

The deformation of steep surface waves. I 21

energy, Emi of the most energetic progressive wave, that is Em—0.07403, for
$ = 0.92.
We mention the results of the checks listed in § 7. The flux Q remained at about
10~5 to 10-4 for most of the calculation, increasing to 10-3 near the end. The mean
level y8 was of order 10-4 for all the calculations, except near the last profile for
p0 = 0.146, when it reached 1 x 10-3. The total energy E remained very constant
after t = n for all runs, its largest fluctuation being a decrease of about 0.2 %.
The above calculations were programmed in Fortran rv double precision, and
carried out on the I.B.M. 370/165 at Cambridge. The computer time needed for each
pressure forcing to achieve overhanging waves was typically 35min for 270 time-
steps at 169 К bytes of storage.

(я-*)
F iq u b e 10. Ав figure 8, but with p0 = 0.126. 4.830 ^ < 5.046.
 1779

22 M. S. Longuet-Higgins and E. D. Cokelet

11. D iscu ssion and con clu sion s

We have developed a numerical technique to study nonlinear, unsteady, free-

surface waves. Basically the method involves solving an integral equation along the
fluid surface to determine the spatial dependence of the motion at each time step.
This has the advantage that the only variables which need be calculated and stored
are those at the free surface, which is precisely where the flow information is most
needed. The surface is represented b y marked Lagrangian fluid particles, but our
solution technique is neither exclusively Lagrangian nor Eulerian. I t makes use of
the powerful properties of potential theory b y adopting an Eulerian form at each
instant o f time, and it also makes use of the marked-particle quality of the
Lagrangian description b y following the fluid particles and their Eulerian velocity
potential through time.
I t is now possible to follow the development of the surface numerically beyond the

(x -t)
Figubb 11. As figure 8, but withp0 = 0.146. 4.712 < t < 4.928.
1780

The deformation of steep surface waves. I 23

instant when the tangent first becomes vertical. From the results of § 10 we conclude
that waves do not necessarily develop a sharp comer or singularity before the free
surface overturns. Instead they can curl over and plunge towards the forward face
o f the wave, and there is nothing to suggest that the flow does not remain smooth
(neglecting surface tension) up to the instant of impact. The Stokes 120° angle may
у

Figure 12. A comparison, of overturned profiles, for different amplitudes of foroing.

(o) p 0 = 0.0729, t = 6.890; (6) p 0 = 0.100, t = 6.282; (c) p0 = 0.126, t = 6.046;
(d)p0= 0.146, t = 4.928.
 1781

24 M. S. Longuet-Higgins and E. D. Cokelet

not be a typical fluid flow but simply a highly singular form, intermediate between
a steady, symmetric wave and an unsteady, unsymmetric, plunging wave. In the
gravity-free flows investigated by Longuet-Higgins ( 1972, 19756 ) a mass of fluid is
deformed into a continually elongated jet with a thin tip. A plunging breaker which
is nearly free-falling may behave similarly. It is also possible that most spilling

F igure 13. The kinetic energy Г as a function of the time t, for each pre
(a) p 0 = 0.0729, (6) p0 = 0.100, (c) p0 = 0.126, (d) p0 = 0.14

F igure 14. The potential energy V as a function of the time t, for each pressure amplitude.
(a) p0 = 0.0729, (6) p0 = 0.100, (c) p0 = 0.126, (d) p0 = 0.146.
1782

The deformation of steejp surface waves. I 25

breakers begin b y being similar in form to a plunging breaker, but on a smaller scale,
close to the crest o f the wave.
In practice, surface tension, air currents and the growth of instabilities m ay cause
the plunging jet to break up into droplets or spray before the tip hits the front face
o f the wave. In this first study we have neglected surface tension, in order to
concentrate on the simpler, large-scale features o f the flow.

t
F igure 15. The total mean energy [E = T + V) as a function of the time t, for each pressure
amplitude, (a) p0 = 0.0729, EJEm = 1.37; (b) p0 = 0.100, E fE m = 1.56; (c) p0 — 0.126,
E jE m = 1.73; (<Z) p0 = 0.146, E {E m = 1.88

Our computations, which are for deep water, show incidentally that for the
breaking o f irrotational waves a sloping bottom is not necessary. Indeed, b y elimin­
ating the bottom, one of the parameters of the problem, namely the ratio o f the
depth to the wavelength, has been eliminated. On the other hand, similar computa­
tions can easily be done for periodic waves in water of any arbitrary uniform depth.
B y varying the magnitude and duration of the applied surface pressure, it should
in future be possible to gain insight into the energy and momentum lost by sea waves
b y wave breaking, both in deep and shallow water, as a function of the energy input,
and hence to gauge the transfer of momentum from the waves to surface currents.
For simplicity we have assumed that the waves are initially progressive. Here
again this assumption can be generalized and we can apply the method either to
standing waves, or to waves that are a mixture of opposite but unequal progressive
waves. These and other possible applications will be studied in future papers.

This computation was begun in March 1972 by one of us (M.S.L.-H.) on the

I.B.M. 360/50 at the Institute of Theoretical Astronomy, Cambridge. For the first
 1783

26 M. S. Longuet-Higgins and E. D. Cokelet

two years of his research, E. D. C. was supported b y a Graduate Fellowship from the
U.S. National Science Foundation, and for a third year b y a Research Studentship
from the Cambridge Philosophical Society. During a final year, research fees were
remitted b y Cambridge University. To all these bodies past and present we express
our thanks and appreciation.

R eferences

Acton, F. S. 1970 Numerical methods that work. New York: Harper & Row.
Ahlberg, J. H ., Nilson, E . N. & Walsh, J. L. 1967 The theory of splines and their application.
New York: Academic Press.
Biesel, F. 1952 Study of wave propagation in water of gradually varying depth. In Gravity
waves. Giro. U.S. natn. Bur. Stand, no. 621, pp. 243-253.
Buckingham, R . A . 1962 Numerical methods. London: Pitman.
Chan, R. K . & Street, R . L. 1970 SUMMAC - A numerical model for water waves. Tech. Rep.
Dep. Civil Engrg, Stanford University, Calif., no. 135.
Cokelet, E. D . 1975 Steady and unsteady nonlinear water waves. Ph.D. dissertation,
Cambridge University.
Gerald, C. F. 1970 Applied numerical analysis. Reading, Mass.: Addison-Wesley.
Longuet-Higgins, M. S. 1953 a On the decrease of velocity with depth in an irrotational
water wave. Proc. Camb. Phil. Soc. 49, 552-560.
Longuet-Higgins, M. S. 1953 b Mass transport in water waves. Phil. Trans. R. Soc. Lond.
A 245, 636-681.
Longuet-Higgins, M. S. i 960 Mass-transport in the boundary layer at a free oscillating
surface. J. Fluid Mech. 8, 293-306.
Longuet-Higgins, M. S. 1972 A class of exact, time-dependent, free-surface flows. J. Fluid
Mech. 55, 629-543.
Longuet-Higgins, M. S. 19750 Integral properties of periodic gravity waves of finite ampli­
tude. Proc. R. Soc. Lond. A 342, 157-174.
Longuet-Higgins, M. S. 1975 b Self-similar, time-dependent flows with a free surface. J. Fluid
Mech. 73, 603-620.
Longuet-Higgins, M. S. & Stewart, R . W . i 960 Changes in the form of short gravity waves
on long waves and tidal currents. J. Fluid Mech. 8, 565-583.
Phillips, О. M. & Banner, M. L. 1974 Wave breaking in the presence of wind drift and swell.
J. Fluid Mech. 66, 625-640.
Price, R. K . 1971 The breaking of water waves. J. geophys. Res. 76, 1576-1681.
Sohwartz, L. W . 1972 Analytic continuation of Stokes’ expansion for gravity waves. Ph.D.
dissertation, Stanford University.
Schwartz, L. W . 1974 Computer extension and analytic continuation of Stokes’ expansion
for gravity waves. J. Fluid Mech. 62, 653-578.
Stokes, G. G. 1880 a Considerations relative to the greatest height of oscillatory waves which
can be propagated without change of form. Mathematical and physical papers, vol. 1,
pp. 225-228. Cambridge University Press.
Stokes, G. G. 1880b Supplement to a paper on the theory of oscillatory waves. Mathematical
and physical papers, vol. 1, pp. 314-326. Cambridge University Press.
1784

Proc. R. Soc. Lond. A. 360, 471-488 (1978)

Printed in QrecU Britain

The instabilities of gravity waves of finite amplitude

in deep water

I. Superharmonics

B y M. S. L o n g u e t - H i g g i n s , F.R.S.
Department of Applied Mathematics and Theoretical Physics, University of
Cambridge, Silver Street, Cambridge CBS 9EW, U.K. and
Institute of Oceanographic Sciences, Wormley, Surrey, C/.isT.

(Received 13 September 1977)

In this paper we embark on a calculation of all the normal-mode per­

turbations of nonlinear, irrotational gravity waves as a function of the
wave steepness. The method is to use as coordinates the stream-function
and velocity potential in the steady, unperturbed wave (seen in a reference
frame moving with the phase speed) together with the time t. The de­
pendent quantities are the cartesian displacements and the perturbed
stream function at the free surface.
To begin we investigate the ‘ superharmonics i.e. those perturbations
having the same horizontal scale as the fundamental wave, or less. When
the steepness of the fundamental is small, the normal modes take the
form of travelling waves superposed on the basic nonlinear wave. As the
steepness increases the frequency of each perturbation tends generally to
be diminished. At a steepness ak « 0.436 it appears that the two lowest
modes coalesce and an instability arises. There is evidence that this
critical steepness corresponds precisely with the steepness at which the
phase velocity is a maximum, considered as a function of ak.
The calculations are facilitated by the discovery of some new identities
between the coefficients in Stokes’s expansion for waves of finite amplitude.

i. I n t r o d u c t i o n

For a proper understanding of air-sea interaction and of the generation an<

development of a wave field, some knowledge of the form and the dynamics о
steep water waves is clearly essential. To this end several new methods of calcula
tion both for steady, symmetric waves (Longuet-Higgins 1975; Cokelet 1977
Longuet-Higgins & Fox 1977, 1978) and for unsteady, time-dependent flow
(Longuet-Higgins & Cokelet 1976) have recently been put forward. Particularl;
in the last-mentioned paper it was shown how, given a spatially periodic wav
form with arbitrary initial conditions, it is possible to follow numerically th
development of a breaking wave until after the free surface has overturned.
Such numerical methods have much in common with a controlled experimenl
carried out with high accuracy. To understand the results, however, we need t
[471]
 1785

472 M. S. Longuet-Higgins

bring to bear analytical methods and approximations. Leaving aside the later
stages of plunging breakers, our aim in the present paper is to study analytically
the initial development of the motion, when the flow may be treated as a small,
time-dependent perturbation of a steep, steady wave (as viewed b y an observer
travelling along with the phase-speed). The form of the steady wave, though
evidently nonlinear, can be well calculated, and it is permissible to treat the time-
dependent part of the motion as a linear perturbation of this steady flow.
There is a second reason for undertaking such an investigation. Though the
mathematical existence of gravity waves on deep water has not been in doubt
since the work of Levi-Civita (1925), nevertheless the question of their stability
to small perturbations has remained unsettled. Benjamin & Feir (1967), noticing
the tendency towards disintegration of waves in a laboratory channel, successfully
analysed one particular type of instability, due to the growth of sidebands. This
causes a slow modulation of the wave envelope, with a space-scale much larger
than one wavelength. The question whether gravity waves are unstable on short
length-scales, comparable to, or less than, one wavelength, is not answered by their
analysis.
A t the same time, recent computations of steep, perfectly periodic waves
(Longuet-Higgins 1975; Longuet-Higgins & Fox 1978) have shown the existence
of stationary values in the phase-speed, considered as a function of the wave height
(see figure 1 of the present paper). It is not unreasonable to expect that a maximum
in the phase-speed will mark the onset of a new type of normal-mode perturbation,
probably unstable. Such an instability would be of the same scale as a wavelength,
or less. The observed speed with which overturning of the free surface can take
place suggests indeed that this type of breaking may begin as a fairly localized
instability.
In this paper our aim is to investigate all types of instability, both those on a
space-scale that is less than one wavelength (superharmonics) and those with scale
greater than a wavelength (subharmonics). For reasons of exposition we take first
the superharmonics.
In our problem the basic motion is a wave of finite amplitude, which cannot be
treated as a small perturbation. Hence we develop in §2 a type of analysis in
which as independent coordinates we take the stream-function and velocity-
potential of the flow at any instant, together with the time t. This is an extension
of Stokes’s (1880) method for steady flows. The boundary conditions are then
expressed in the new variables (equations (2.8) and (2 .9 )). To analyse the perturba­
tions, we m ay take as basic coordinates the values of ф and ijf in the unperturbed
flow (that is to say, the symmetric wave of finite amplitude) and linearize with
respect to the small perturbations (§ 3 ). The analysis of the perturbations into
normal modes, and the method of calculating the eigenfrequencies, is described
in § 5 and the results are discussed in §§ 6-10.
The calculated eigenfrequencies are shown in figure 1 as functions of the steepness
ak of the unperturbed wave. I t appears that all the superharmonic perturbations
1786

The instabilities o f gravity waves. I 473

remain stable up to ah = 0.42 at least, but there is indeed evidence for the occur­
rence of an instability at about ak = 0.436, the amplitude corresponding to the
maximum phase velocity. For reasons given in §9, other stationary values of the
phase-velocity probably correspond also to the onset of instabilities.
The relation of these unstable modes to the subharmonic instabilities of Ben­
jamin & Feir will be discussed in part II.

2. T r a n s f o r m a t io n of b o u n d a r y c o n d it io n s

Consider any time-dependent, irrotational flow of an inviseid incompressible

fluid, in two dimensions. Let ф and ^ denote the velocity potential and the stream
function respectively. In general these are functions of the horizontal and vertical
coordinates x , у and the time t. The usual conditions to be satisfied at a free surface
/ ( * , * * ) - 0 are that df/dt = 0 when / = 0 , (2 . 1)

where d /d t denotes differentiation following a particle, and also the Bernoulli

condition p/p+\q*+gy+<t>t = p(t), (2 .2 )
where p and p denote the pressure and the density, q2 denotes (V0)2, g is gravity
and фг denotes the partial derivative of ф with respect to t, when x and у are held
constant. fi(t) is an arbitrary function of the time only, which will be incorporated
into gy or фь that is we may take /? = 0 .
In our problem, however, it will be more convenient to take ф, \jr and t as in­
dependent variables and x, у as functions of (ф, tfr, t). This is because we shall be
dealing with small perturbations of a steady flow in which the equation of the
unperturbed free surface is \jr = constant.
Now consider how the boundary conditions must be transformed. Using sub­
scripts for partial derivatives we have the fundamental relations

Ф хх ф+ ФуУф - !. Ф х х ф+ Ф*Уф = °>1

Ф х ^ + ФуУ^г = 0, t x x + + t v y* =U
and hence, using the Cauchy-Riemann relations,

(Фх> Ф у Фх. Фу) = яЧ*ф, Уф, Х+, Уф)-

Also фь = -{ф хх( + фуу1) = q2d{x,y)/b(f,t),
(2.3)
Фь = ~ ( Ф х х 1 + Ф уУ д = 1
- Я гд (х ,у )/ д (ф ,«)./

So when р = 0 equation (2 .2 ) becomes

к2+0У+?29(ж>У)/Э(^М) = о,
and on multiplying by q~2 we obtain
Ъ(х,У) , Ъ(х,у) (0 .
Ш ) +дут ж ) + ь - ° - { ]
 1787

474 M. S. Longuet-Higgins
To transform the kinematic condition (2 . 1) we note that since
d , 9 , 0 0
d< Yx dx bt
we have dф

(2.6)

from equations (2.3). Now let the free surface be given in the form:
+ *шЖ(ф%Ь). (2 .6 )
Then the kinematic condition can be expressed as
d(F —г/г)/dt = 0 ,

that is, F^ +Ft- * ± = 0.

On substituting from (2.5) and multiplying by q~2we obtain

(i I Ь{-Х'У">\ p , д(*’ У) p I n f“ 7)
1 1 + т г ) ) F*+ m r ) F*+ w j ) - °- (2-7)
Finally in equations (2.4) and (2.7) we may use the Cauchy-Riemann relations
for ХфУур Xp, y^. to obtain two boundary conditions for у{ф, t) and F $ , t), namely

-(Уф Vt + У* *t) + 9У(Уф+ yfy + h = °. (2 -8)

and f l-( y * y » + y # * t) ] -^ + (»J+l£)-Ft + ( y * y t - y * * » ) - 0, (2 -9)

where x must be determined as the conjugate function to y. Equations (2 .8 ) and

(2.9) are both to be satisfied on the free surface \jr = P(0, t).

3. P e rtu rb a tio n a n a ly s is

Equations (2 .8 ) and (2 .9 ) hold quite generally, for any time-dependent, irrota­

tional, inviscid incompressible flow. We now wish to consider perturbations of a
steady state, in which the free surface is given by r/r = 0 . In fact we have in mind
that the steady state is a symmetric, deep-water wave of finite amplitude, seen in
a frame of reference moving with precisely the phase-speed —c. Thus we shall
write in general x = Х(ф, # ) + № >
,,*vM
, t);
).1
(3.1)
У = У(ф,</г) + у(ф, <)./
where X, У represent the steady, unperturbed motion with free surface rjf = 0,
and where ?, ij and the perturbed surface, —Р(ф, (), are all quantities of order
e, on the scales of the unperturbed flow. Substituting then in equations (2.8) and
1788

The instabilities o f gravity waves. I 475

(2.9) we obtain two expressions which must vanish on the free surface \Jr = F. But
on expanding each expression in a Taylor series in \jr we obtain expressions which
are to vanish instead on xjr — 0. Thus to order e we have respectively

—(Хф^t + ^fr£t)
+ p (Y +Y^F + ?7)[(7$ + Y\) + (7^ + Y^^F + 2 ( 1^ 77^+ 1^ 77^)] + J = 0, (3.2)

and * , + ( * } + Ц ) *t + (Y+Vt ~ &) = 0. (3.3)

In (3.2) the terms of lowest order yield

2gY {Y l+Y D = - l (3.4)

and those of order e give

- ( Y t Vt+Y+£t) + g ( Y } + Y l ) ( Y t F + V)
+g[Y(Y$+ Yl)+F+2Y(YtVt+ Y f ^ )] = 0 .

Setting then <7=1

we have
(УфЬ + ГфЪ) = (¥1+Ц )'г1+ 2Г(¥фГ1ф+ Г4гг14,) Н У { Ц + Y$j\f F (3.5)

and (Уф - 1V %) = ^ + ( У? + Ц ) Ft (3-6)

to be satisfied on xjr —0. Also £ and 77 are conjugate functions of (ф, ft) in the interior.
Given the basic flow 7(0, iff) satisfying Уфф+ 7 ^ = 0 and the boundary condition
(3.4), we have to find £, 77 and F, satisfying (3.5) and (3.6), so as to represent an
exponentially growing disturbance. Thus we seek solutions in which £, r) and F
are proportional to e~lcrt, with Im (cr) > 0 . Solutions in which Im (cr) = 0 will be
neutrally stable. Greatest importance will be attached to those eigensolutions, if
any, in which Im (<r) is greatest, representing the fastest-growing instabilities.
It is then convenient to write equations (3.4), (3.5) and (3.6) in the form

2P 7 = - 1 (3.7)

and - i<г(Уфri + Y+g) = Pq + Q^ + B y +

-icr(Y^V- Y ^ + P F ) = -F ^ I

where P = + Y\,
Q = 2 YYr
(3 .9 )
R = 2YY+,
S = [ 7 ( r j+ r » ) V , )

all these quantities being evaluated at \jr = 0 .

 1789

476 M. S. Longuet-Higgins

4. P e r t u r b a t io n of a u n if o r m flow

As the simplest example of a flow to which the equations of § 3 can be applied,

consider a uniform horizontal flow given by
Х = ф, Y =s Y0+i/r, (4 .1)
which will satisfy (3.7) provided
(4-2)
Equations (3.8) now become (when \jr —0 )
-fc r g « 9 -fy + j? 7 l
(4.3)
—icr{i}+F) = —Fj. J
These have solutions of the form
£= A
7) = —iA e**
and i 7 = i£ e“^ _w,
provided (<т+ к -1 )А = - В л
crA = (<r+^)5;J
hence (o-+k)2 = k.
This equation has only the real roots
a = ~ k± k i,
which correspond to neutral waves propagating with the phase-speed
a/k = —1 ± k~i
of free gravity waves travelling in the positive or negative sense.

6. C a l c u l a t i o n o f a steady, s y m m e tric w a v e

A steady, symmetric wave in deep water may be represented by the expressions

X = ф/с + H Xe^/c sin (ф/с) +H 2e*^/c sin (2ф/с) + .. .,1
Y = х/г/с+ +H i e ^ cos (ф/с) +H 2e^/cсов(2ф/с) + ...,J

where с represents the stream velocity at infinite depth (equal to minus the phase
velocity in a stationary frame of reference) and # 0, Hlt... are real constants. The
wavelength is taken to be 2n and this together with the choice <7 = 1 fixes the
scales of length and velocity. The crest of the wave is at (ф, iff) = (0 , 0 ). The co­
efficients # 0, Hlt ... and с must be chosen so that the condition (3.4) is satisfied.
This condition leads, as is well known, to a one-parameter system of waves in
1790

The instabilities of gravity waves. I 477

which, however, neither Нг nor с increases monotonically throughout the entire

range (Schwartz 1974; Longuet-Higgins 1975).
In calculating the Fourier coefficients and Cn = nH Jc of the unperturbed
wave, the author noticed some previously undiscovered relations (equations
(A 14) below), of which an analytical proof is given in another paper (Longuet-
H iggins 1978). Use of these relations greatly simplified and expedited the
calculation o f the Hn as functions of the wave steepness ak. Here 2 a denotes the
crest-to-trough height o f the w aves:
a = H 1 + H 3 +H 6+ ... (6*2)

and к denotes the wavenumber

к = 2rc/wavelength = 1 (6 .3)

in our chosen units. The quantity ak is also sometimes called ‘ the dimensionless
w ave am plitude’ .
Th e Gn were calculated in quadruple precision to order a120, b y the simple algo­
rithm described in Longuet-Higgins (1978). The results for the phase velocity с
are shown in fig. 1 of that paper, where c2 is plotted against ak over a range including
the higher waves. The calculations check very precisely the values of c2 from an
earlier calculation (Longuet-Higgins 1975). In particular the maximum value of
the phase velocity as a function of wave steepness are verified. This occurs at
ak = 0 .436 ..., as compared to the maximum wave steepness, which is when
ak = 0 .4434 . A quite independent calculation of the phase-speed b y an asymptotic
analysis valid in the limit as a k ^{a k )mftx has been given by Longuet-Higgins &
F o x (1977, 1978), with practically identical results.

6. P e r t u r b a t io n s of a s t e a d y w a v e

W e shall assume at first that the perturbations £, 17, F are also periodic in ф with
the same basic wavelength 2 n. This temporarily excludes all subharmonics of the
wavelength and also instabilities of the Benjam in-Feir type, though these will be
considered later. In general we have

£oc a0 4- axe^/ccos {ф/с) + a2 e ^ c o s (2 ф/с ) + ...

+ 6Xe^/csin {ф/с) 4- b2 e ^ lcsin {2 ф/с) + ...,
(6 .1)
7} ос Ь0 4- fei e*lc cos {ф/с) + b2 e**lccos (2 ф/с) + ...
— агe*,csin {ф/с) —a2 e ^ lcвт{ 2 ф/с) —... ,

and Fee c0+ cxcos {ф/с) + c2cos (2ф/с) 4-...

(6.2)
4- d\ sin {ф/с) 4- d2 sin (2ф/с)
The factor of proportionality is е~ы . Note that terms Unear in ф and ^ are not
included in f or i] \these would imply a finite change of momentum at infinite depth,
which cannot be brought about by a zero or finite pressure at the free surface.
 1791

478 M. S. Longuet-Higgins

The origins of ф and \Jr, in the perturbed motion, are in general arbitrary. How­
ever, we have fixed the origin of ф by assuming P(t) to vanish in equation (2 .2 ).
The origin of \fr can now be fixed by assuming that in the expansion of the free
surface \jr = F we have
c0 = 0. (6.3)
Now the functions 5^, Y^, P, Q, R, S can all be expanded in Fourier series in ф
(see appendix A). Of these, Y^, P, Q, S are even functions of ф, involving terms in
cos(ti0/c), and Y^ R are odd functions, involving %тпф. Now substituting the
series (6.1) and (6.2) into equations (3.8) and equating coefficients of соапф
(n = 0 , 1, 2 ,...) and siппф (n = 1, 2,...) we obtain an infinite system of equations
for the unknown coefficients an, bn, cn, dn in the form

ХАх-Вх (Л = -i(r),

where x denotes the column-vector whose elements are (an, bn, cn, dn) and A, В
are matrices of coefficients independent of <r. This system of equations may be
solved by truncating the vector x at successively higher values of n, neglecting
terms greater than n = N, say. Thus at the iVth stage there are (4N + 2 ) unknowns
a0, ..., fljyj ..., bjsf; Cj, •••> j с?!)..., dpj and (T. From the coefficients of cos тьф
(n = 0,1, ...,N) and s i n (n = 1, in the two equations (3.8) we get (42У+ 2)
homogeneous equations, whose coefficients form a square array. The vanishing of
the corresponding determinant gives a characteristic equation for Л, namely
|(АЛ„-В„)| = 0 ,

where N denotes the order of the truncation. The roots of this equation, if they
converge as i^-^oo, give the instabilities of the system. Greatest interest attaches
to those roots, if any, for which i?(A) is large and positive.
Details of the matrices AN and BN, which are important for actual computation,
are given in appendix A. The matrices were constructed in F o r t r a n rv double
precision, for the IBM 370/165 at Cambridge University. The eigenvalue equation

\Bn A J - \ I n\= Q (6.4)

was solved by a standard routine ( f o 2 a g f ) for a sequence of values of N, until the
relevant eigenvalues had converged. In some cases convergence was accelerated by
Shanks transforms. For amplitudes less than 0.30, good convergence was obtained
with N = 20 . The higher wave amplitudes required values of N ^ 45. For these the
computation was transferred to the 360/195 at Rutherford Laboratory.

7. T he computed e ig e n v a l u e s

In figure 1 are shown the radian frequencies <r of the computed normal-mode
perturbations, as a function of the steepness ak of the unperturbed wave. These
have been designated n — 1,2, 3,... and n = —1, —2, —3,... on the following basis.
1792

The instabilities of gravity waves. I 479

At small values of ak the perturbations are nearly sinusoidal, and |n| denotes the
corresponding number of spatial cycles contained in one wavelength 2 к of the
unperturbed wave. Thus n — 1 corresponds to a perturbation which simply shifts
the phase of the unperturbed wave by a small amount. The phase-shifted wave
travels with the same phase-speed as the unperturbed wave. Hence the frequency
of the perturbation is precisely zero.
Similarly n — 2 corresponds to a perturbation of twice the wavenumber, and
half the wavelength, of the unperturbed wave. The frequency of the wave, as
afc->0, is given by „ = 2 - 7 2 = 0.5868... (7 . 1)

Fioubb 1. Radian frequencies cr of the normal-mode perturbations, as functions of the

dimensionless amplitude ak of the unperturbed wave.

in the frame of reference moving with the speed of the unperturbed wave. This can
be seen from §4 by writing k = \n\ = 2 . Generally, those perturbations of wave­
number n ^ 1, travelling in the same sense as the unperturbed wave, have a radian
frequency cr = » - rai (7.2)
in the limit as ak 0.
 1793

480 M. S. Longuet-Higgins
Negative values of n correspond to perturbations travelling in the opposite
sense to the unperturbed wave. These have a relative frequency
(T=\n\ + \n\i (7 .3)
in the limit as ak 0 . Thus the perturbation n = —1, having the same wavelength
as the unperturbed wave but travelling in the opposite direction, has a frequency
cr = 2 ; to an observer moving with the speed of the unperturbed wave the frequency
of encounter of crests of the perturbation is just doubled.

(ak)*
0 a05 _ 0.10 _ . 0.15

ak
Figure 2. Aar plotted against (ak)*, showing the decrease in frequency of the normal modes
due to the finite amplitude of the basic wave. Broken lines represent the asymptotes (7.8).

So at small values of ak the perturbation is propagated independently of the

presence of the unperturbed wave. As ak increases, however, we see from figure 1
that the unperturbed wave begins to affect the frequency of the perturbation.
Now it was shown by Longuet-Higgins & Phillips ( 1962) that the speed cx of a
gravity wave of wavenumber kx relative to the water at infinite depth is affected
by the presence of a second wave of amplitude a%and wavenumber k2to the extent

that = ± (?A )* - (“ a**)* c2. (7-4)

where c2 is the speed of the second wave, here assumed travelling in the negative
1794

The instabilities of gravity waves. I 481

direction. The sign of (д/к)Ъ is + or — according to the direction of propagation of

the first wave. Equation (7.4) is valid provided the first wave is of relatively small
steepness, and that it is shorter than the second wave.
However, the speed of the second wave itself is affected by its own amplitude,
so that relative to deep water
c2 = l + J K t f (7-6)
in the present units. Therefore, relative to an observer moving with speed c2, the
speed of the first wave is
o' = Cj + Ca = l± k ~ i- i {a 2k2)2 (7.6)

to order (a2k2)2. Hence the radian frequency, in this frame of reference is

cr = (7-7)
Writing = |n| in this formula we see that the frequency of the first wave is
shifted by an amount
Acr = —\(ак)2\п\, (7.8)

where ak is the steepness of the unperturbed wave and n the number of the harmonic.
Thus regardless of whether n is positive or negative the relative frequency always
tends to be reduced. (The case n = 1 is excluded since then the two wavelengths
are equal.)
Figure 2 shows the computed values of Acr plotted against (ak)2, and compared
to the asymptotic expression (7.8). It will be seen that the shift in frequency agrees
well with the approximate formula up to about ak = 0.30. At higher values of
|n| the relative change in frequency (Acr/cr) is comparatively small.

8. T h e eigenfunctions

Figure 3 shows the set of eigenfunctions corresponding to the modes in figure 1

at the typical wave amplitude ak = 0.20. It may be checked that the number of
maxima and minima in each mode is equal to the corresponding value of |n|. The
modes are progressive, and only the in-phase component is shown; the quadrature
component is shifted along the surface of the unperturbed wave by about a quarter
of the local wavelength.
The pairs of curves in figure 3 were in fact obtained by plotting (х}у) —{Х(ф, 0),
Y(фу0 )} (the unperturbed wave profile) and

х ~ Х( ф, 0) + е£(ф, 0, t) + еР(ф, 1)Х^(ф, 0),1

у = ¥(ф, 0 ) + е1)(ф, 0, {) + еР (ф, t) Гф(ф, 0 ),/

where £, tj, Р are the normalized values of the perturbations, and e is a small
constant. Thus £, i), F are equal to etj, eP. The expression (8 .1) represents the
~~2
1796

The instabilities of gravity waves. I 483

first order expansion of (3 .1) about yjr = 0 . A finite value of e is necessary in order
to distinguish between the members of each pair, but since quantities of order €2
are neglected there has to be some distortion in the representation.
Nevertheless two predictions can be verified: (1) that the local wavelength of
the perturbations is shorter in the crests of the unperturbed wave than in the
trough; (2 ) that the amplitude of the perturbation is greater at the crest of the
unperturbed wave than in the trough. The local wavenumber, according to
Longuet-Higgins & Stewart (i 960 ), is approximately
|?t| ( 1 +afccosfoc) (8 .2 )
and the local amplitude is
a(l + akcoskx) (a = const.), (8.3)
so that the ratio of the maximum to the minimum values of each quantity is
(1 +ak)/(i—ak). (8.4)
Both predictions (8.2) and (8.3) are well verified by the waves in figure 3.

9. B e h a v io u r a t la r g e v a lu e s o f ak
Figure 1 indicates that the frequencies cr of all the normal modes except n — 1
diminish monotonically as ak increases, at least until ak = 0.42. Beyond this value,
convergence becomes difficult to obtain by the present method.
The mode n = 2 is especially interesting, as it appears to be approaching the axis
cr = 0 the most rapidly. If the frequencies сгг and <r2 of modes n = 1 and 2 coalesce
at a critical wave amplitude a = a12, say, and if
о| ~ K(a12- a ) (К > 0), ( 9 .1 )

then we would expect that, when a > a12, <r2 would become imaginary, indicating
an unstable mode. Figure 4 shows a plot of cr\ against ak at high wave amplitudes.
Though the computations cannot yet be carried beyond ak = 0.42, an extra­
polation of the known values certainly suggests that o\ is tending to zero at some
wave amplitude ал2 less than the maximum. It also appears that 2 is close to, if
not coincident with, the amplitude corresponding to the maximum phase speed.
The occurrence of an instability at this wave amplitude might be understood as
follows. Consider a particular perturbation of a symmetric wave which is such as
to change the unperturbed wave to a symmetric wave of slightly greater amplitude.
In general, since the phase speed is a function of the amplitude, the perturbed wave
will travel at a slightly different speed from the unperturbed wave. Hence the
small perturbation (i.e. the difference between the perturbed and unperturbed
waves) will not be a normal mode, since it began by being symmetric about the
crest, and gradually developed asymmetry. However, at the particular wave
amplitude at which the phase speed is a maximum, the perturbation will remain
 1797

484 M. S. Longuet-Higgins

symmetric, and will have zero frequency. Thus, at any stationary value of the
phase-speed a new type of normal mode, with zero frequency, becomes possible.
This leads us to expect that points on the axis a = 0 at which the speed is
stationary will be critical points where the modes will branch or coalesce, and at
which new and unstable modes may first appear. (This does not exclude the possi­
bility that some critical points will occur when |cr| > 0 .)

wave amplitude, ak
F ig u r e The squares of the radian frequencies <rx and <rt for the two lowest
4.
modes, at high values of the dimensionless wave amplitude.

The asymptotic analysis of steep gravity waves by Longuet-Higgins & Fox

( 1977 , 1978) has recently shown the presence of stationary points in the phase velocity
not only at ak = 0.436 but also at a sequence of wave amplitudes approaching
very closely to the maximum amplitude. It seems plausible that these may represent
points of coalescence between the mode n = 1 and some of the higher modes
n = 3 , 4 ,... which are approaching the real axis at greater values of ak. The verifica­
tion of this conjecture must however await an investigation of the stability based
on the asymptotic analysis itself, which is more suited to the problem of very
steep waves.

A p p e n d ix . C o n s t r u c t i o n o f th e m atrices A and В

From equations (6 . 1) we have

IV = a0+C'1cos^+O2cos2(?+...,| ^ ^

Yj = Gxsin 6+C 2sin 20 +

1798

The instabilities of gravity waves. I 485

where 0 ——ф/с and

C0 = 1/c, Gn = n H jc (71=1,2,3...). (A 2 )

Hence P = ( Yl + Y \ ) ^ = P0+ 2PXcos 0 + 2 P2cos 2 0 + ...Л

Q = (2 YY^ =о = Oo + 2^i cos 0 + 2 £ 2 cos 20 + ..., | (A 3)
R = {2 Y Y ^ 0 = 2 jR1sin0 + 2i 2aein20 + ..J
where
Д - Ona0 + On+1 C1+ (7n+2O2+ ..., "j
2g n = 2Я* C0+ (Яп_х+Яп+1) <7, + (Яп_2 +Ял+2) C2+ j (A 4)
2Pn = (Яп_1- Я л+1)0 1+ (Яп_2 - Я л_2)С2+ ...,]

and we define Я_, = Я,. Lastly

5 = [{7(7J+ r^JVV-o = t^+O ^W + ^ V - o - (A 5)

Writing Dn = (n/c)Cn =(n/c)*Hn (A 6 )

we have PY^= U0+ 2£71cos0 + 2£/2 cos 20 + ...Л

+
= To 2Ti cos 0 + 21^ cos 20 + ..., J (A 7)
ВУфр = W0+ 2WXcos 0 + 2WZcos 20 + ... J

where
2Un = 2PnC0 + (Pn+l+-Pn-i) ^l+(Pn+2+Pn-2) C2+ ...Л
2T^ = 2Qn D 0 + (Qn+1 + Q n_ i) Dx + ( § n+2 + Qn-2) ^ 2 + •••> / (A 8 )
2Wn — (Rn+l —i?n_x) Dx+ (Rn+2—Rn_2) + •••J
and we have defined
P-i = Pi> Q-j = Qi> R-i = — (A 9)

From (A 5) and (A 7) we have

S = £ o= 2 £ 1c o s0 + 2 $ 2 c o s 2 0 + (A 10)

where Sn = Un+ Vn+ Wn. ( A H)

Now from the definitions of Qnand Bn in (A 4) we have

Qn + -Kn = <?n> (A 12)

where Gn = Hn C0 +Hn_1C, +Hn_%Ct + .... (A 13)
It has been shown (Longuet-Higgins 1978) that

(?0 = —1/C0, Qn = 0 (n = 1,2,3,...). (A 14)

Hence we have Q0 = - i/O l), Rn = - Q n ( n = 1,2 ,3,...), (A 15)

 1799

486 M. S. Longuet-Higgins
and from (A 8 ) and (A ll)
= %Pn C0+ (i ^+1 + Px_j) Сг + [Pn+2+ Рп-ъ) C2+ ...
+ 2^nZ)0 + Z)x+ 2^n_zZ)2+ ... + Q0Dn. (A 16)
On multiplying the above series for Y^} 7^, P, Q, R and S with the Fourier series
(6 .1) and (6 .2 ) for £, tj and F we can now construct the matrices AN and BN of
§ 6 . The general case will be made clear from figures 5 and 6 . These show the layout of
Л3 and B3. The terms whose coefficients are elements of the matrix are indicated
by the row vector (%, ...,d3) above the matrix and by the column vector on the
left. All elements of A3 are zero except those of the submatrices / (1), I{2\ / (3), J(4)
which are as follows:
•P
ьэ

to

CiС CqG +CiС

20,
C +С
С 203
p

/<u =
G 3 GC
2
3
+
C
i+G 3
ъ+х
0
4 2

LCf3 0.\„ C»+ (70

/(2) =
C
cl c2— o C<-O
< xO t-G0 03- 31

[c ct O
2 03
3 s Co ok-<
c ,-c j

PP
zt
Г2
+ P P,*+PiР*+Рг
P
*P
+ Р„iP
2 2Д 1

»P
+*+P
0
P3+PiР
LР.+Л b+P P j 1
»

=
[P
P»-P
~PiP
2 P
*
-P-P
o0 ш -щ 3 a
- л .
1

U -a P
^— Pl ^-• p j
a

Similarly in figure 5 all elements of BN are zero except those of Ja), ..., J<e), which
are as follows:
"2P0 2PX 2P2 2P3
Jd ) =
2PX P0+ P2+ 2Q'0/с P±+ P3 P2 + P4
2Pt P1+ P3+ 2Q1/c P0+ Pi + 4Q'0/c P1+ P6
.2P3 Рг + Pl + 2QJc Р ^ Р ь + iQJc P0+ P3+ 6Q'0/cJ

'P i-P i P0- P , + 2Q'0/c A - P 3 Рг-Р*

J(« = P ,-P 2 P1- P 3+ 2Q1/c P0—Pl +iQ'0/c P i - P b
P3- P 3 P3- P t + 2Q3/c Px- P s+iQi/c P0- P 3+ 6Q'0/c_
For clarity we have written Q'0 = IQ0. Next J<3>and are the same functions of
the Sn as and are of the Pn. And lastly
- 1 /e 0 0
J<« = 0 —2 /c 0
0 0 —3 /c
1800

The instabilities of gravity waves. I 487

a0 Oi аг a3 b0 bt b2 b3 c2 c3 dx d2 d3

a0 ax a2 a3 b0 bx b2 b3 c, c2 c3 dy d2 d3

F ig u r e 6. Layout of the matrix BN when N = 3.

x7 Vol. 360. A.
 1801

488 M. S. Longuet-Higgins

The computations described in this paper were begun while the author was a
guest lecturer in the Institute of Port and Ocean Engineering at the Norwegian
Institute of Technology, University of Trondheim, in March 1977. The author is
indebted to Professor P. Bruun for the kind hospitality of his Department and to
others of his staff, particularly Dr 0. G. Houmb, for many practical arrangements.

R e fer en ces

Benjamin, Т. B. & Feir 1967 The disintegration of wave trains on deep water. 1. Theory.
J. Fluid Mech. 27, 417-430.
Cokelet, E. D. 1977 Steep gravity waves in water of arbitrary uniform depth. Phil. Trane.
R. Soc. Lond. A 286, 183-230.
Levi-Civita, T. 1925 Determination rigoureuse des ondes permanentes d’ampleur finie.
Math. Ann. 93, 264^314.
Longuet-Higgins, M. S. 1975 Integral properties of periodic gravity waves of finite amplitude.
Proc. R. Soc. Lond. A 342, 167-174.
Longuet-Higgins, M. S. 1978 Some new relations between Stokes’s coefficients in the theory
of gravity waves. J. Inst. Math. Applic. (in the press.)
Longuet-Higgins, M. S. <fe Cokelet, E. D. 1976 The deformation of steep surface waves on
water. I. A numerical method of computation. Proc. R. Soc. Lond. A 350, 1-26.
Longuet-Higgins, M. S. & Fox, M. J. H. 1977 Theory of the aJmost-highest wave. The
inner solution. J. Fluid Mech. 80, 721-741.
Longuet-Higgins, M. S. & Fox, M. J. H. 1978 Theory of the almost-highest wave. П .
Matching and analytic extension. J. Fluid Mech. 85, 769-782.
Longuet-Higgins, M. S. & Phillips, О. M. 1962 Phase velocity effects in tertiary wave
interactions. J. Fluid Mech. 12, 333-336.
Longuet-Higgins, M. S. & Stewart, R. W . i 960 Changes in the form of short gravity waves
on long waves and tidal currents. J. Fluid Mech. 8 , 666-583.
Schwartz, L. W . 1974 Computer extension and analytic continuation of Stokes’s expansion
for gravity waves. J . Fluid Mech. 62, 553-578.
Stokes, Sir G. G. 1880 Supplement to a paper on the theory of oscillatory waves. Mathe­
matical and Physical Papers, vol. 1, pp. 314—326. Cambridge University Press.
1802

Proc. JK. Soc. Lond. A. 360, 489-606 (1978)

Printed in Great Britain

The instabilities of gravity waves

of finite amplitude in deep water
II. Subharmonics

B y M . S. L o n g u e t - H i g g i n s , F.R .S.
Department of Applied Mathematics and Theoretical Physics, University of
Cambridge, Silver Street, Cambridge CBS 9EW, U.K., and Institute of
Oceanographic Sciences, Wormley, Surrey,

(Received 13 September 1977)

Calculation of the normal-mode perturbation of steep irrotational

gravity waves, begun in part I, is here extended to a study of the sub­
harmonic perturbations, namely those having horizontal scales greater
than the basic wavelength (2it/к).
At small wave amplitudes a, it is found that all perturbations tend to be­
come neutrally stable; but as ak increases the perturbations coalesce in
pairs to produce unstable modes. These may be identified with the
instabilities analysed by Benjamin & Feir ( 1967) when ak is small. How­
ever, as ak increases beyond about 0.346, these modes become stable
again. The maximum growth scale of this type of mode in the unstable
range is only about 14 % per wave period, which value occurs at ak « 0.32.
At values of ak near 0.41 a new type of instability appears which has
initially zero frequency but a much higher growth rate. It is pointed out
that this type might be expected to arise at wave amplitudes for which
the first Fourier coefficient in the basic wave is at its maximum value, as a
function of the wave height. The corresponding wave steepness was found
by Schwartz ( 1974) to be ak = 0.412.
A comparison of the calculated rates of growth are in rather good
agreement with those observed by Benjamin ( 1967) in the range
0.07 < a k < 0.17.

1. I n t r o d u c t i o n

In this paper we continue and extend the stability analysis of irrotational water
waves of finite amplitude which was begun in part I (Longuet-Higgins 1978 ). In
that paper we determined the normal-mode perturbations of gravity waves on deep
water, for values of the wave steepness up to about ak = 0.42. Consideration was,
however, restricted to perturbations having the same space periodicity as the basic
wave, or less. The existence of instabilities having a much longer wavelength has
been known for some time (Benjamin & Feir 1967; Benjamin 1967). Our present
aim is to obtain an overall view of the instabilities, including scales both shorter
and longer than one basic wavelength.
[ 489 ] 17-2
 1803

490 M. S. Longuet-Higgins
We show first in § 2 how the calculations of the subharmonics can be carried out
by a relatively simple modification of the analysis of part I. The simplest case,
when the perturbations have a horizontal scale of twice the wavelength of the
unperturbed wave, is studied in §3 (see figures 1 and 2 ). This reveals at once the
presence of modes which are unstable over an intermediate range of wave steepness.
By examining a sequence of modes of increasing length scales (§§4 and 6 ) we are
able to delineate the boundary of the unstable zone (figure 7). It is shown that the
lower left hand part of the diagram corresponds to the Benjamin-Feir instability
(§ 6 ). However the diagram shows also that above a certain amplitude each of these
instabilities is temporarily suppressed. At a still higher value of ak, namely
ak « 0.41
a new type of instability occurs, which we show might be expected at around the
value of ak for which the first Fourier coefficient in the analysis is a maximum.
A comparison of the calculated growth rates with the available observations
is made in § 7.

2. G e n e r a l m eth od

As in part I suppose that we are given the basic unperturbed wave motion of
wavelength 2 tc as described
Y = \jr/c+ +Hxe^/ccos (ф/с) + # 2e2^/ccos (2ф/с) + ...,|
Х = ф/с +HXe^/c sin {ф/с) +H2e^ lc sin (2ф/с) +
where the constants Hn have been calculated so as to satisfy the nonlinear con­
ditions at the free surface \{r = 0 . Since the only physical constant is g the accelera­
tion of the gravity (which we take as equal to 1) there exists a precisely similar
solution in which the wave height and length are each reduced by a factor m,
namely
r = f/ c' + \Щ + я ; cos (ф/с') + Щ e ^ cos (2ф/с') + ...
and similarly for X', where
я , = ( Я1"ЩПIm. n/m integral| (2 2)
ft \o otherwise J
and the new phase velocity c' is reduced by mb\
c' —m~bc. (2.3)
The coefficients C'n in the expansion of Y'^ are
C'0 = 1/c = mtC0 (2.4)
Tjt t t ( miCnlm, n/m integrall
and in general C„ = nE Jc = otherwise. j <2 6)

We now have a basic wave with precisely m wavelengths in the interval 0 < X' < 2 tc,
and we may proceed, exactly as in part I, to find all the perturbations of this wave
having length scales 2n or less. These perturbations are described by equations
1804

The instabilities of gravity waves. I I 491

(6 . 1) and (6 .2 ) of part I, with primed quantities replacing the unprimed. Having

determined all the possible normal modes on this scale, we can then convert our
solutions back to a scale in which the wavelength of the original wave is 2 n (not
2 я/га) by multiplying each eigenvector by ra and each eigenfrequency by m~i.
The phase speed с is multiplied by rai. Such a procedure requires only a minor
modification to the original F o r t r a n program.
However, since many of the coefficients C'n and H'n will be zero, the modes will
fall into mutually independent classes, each of which may be treated separately.
To fix the ideas, consider first the simplest case ra = 2 . The perturbations now have
a space periodicity of precisely two wavelengths. From an examination of the
coefficients in the matrices A'2Nand B'2Ncorresponding to part I it will be seen that
the solutions fall into two categories. On the one hand we have modes for which

a2n+l = &2n+l = C2n+1 = ^2n+l = n —0 , 1, 2 ,...

and the coefficients of cos{(2w+ 1)ф/с'}, sin{(2?i + 1)ф/с'] all vanish identically;
these may be called the even modes. Clearly since they contain only even harmonics
they are periodic with wavelength 2ny not only 4n. They are in fact physically
identical with the modes already found and discussed in part I. In the second
category of solutions we have

a2 n ~ bin —C2n = ^2n ~ n —0 , 1, 2,...

and the coefficients of cos (2пф/с')} sin [2пф'/с) all vanish identically. These are the
‘ odd ’ modes - the class which is not yet computed. In these, the perturbations of
adjacent waves are of opposite sign.
To compute the odd modes, we can simply throw out every other row and column
of the matrices A'2N and B2N, reducing the order of the matrices from (82V+ 2) to
4N, where N is the order of the highest harmonic of the basic waves. Hence the
calculation of the odd perturbations can be achieved to the same order of accuracy
as that of the even perturbations.
In fact it was found useful for purposes of checking, first to calculate all the
modes together at some low value of ak and to verify that half of the modes (the
even category) were indeed identical with those found previously. Then the program
was modified so as to calculate only the odd modes, which were again checked at
low ak before proceeding to higher values of ak where core storage becomes a
limiting factor.

3. R e s u l t s : ra = 2

For the odd modes, the eigenfrequencies were found to converge at all values of
the unperturbed wave amplitude up to about ak = 0.42. The results for both odd
and even modes are shown together in figure 1.
The branches have been labelled according to the wavenumber of the dominant
harmonic at low values of ak. Thus n = 6 /2 corresponds to a mode which, at low
ak, has a wavelength 2/5 of the basic wavelength 2n. The even modes correspond
 1805

492 M. S. Longuet-Higgins

to even values of the numerator. Thus the modes n = 2/2 and n = 4 /2 have already
appeared as n = 1 and n = 2 in fig. 1 of part I. (For clarity, the vertical scale is
here doubled.) The new modes are n = 1/ 2 , 3/2 and 5/2.
When ak-+ 0 the radian frequencies of the odd modes, like those of the even
modes, tend to the limiting values
ar=\n±ni\. (3.1)
Thus for the three lowest odd modes (n = 1/ 2 , 3/2, 5/2) the frequencies are res­
pectively 0.2071, 0.2753 and 0.9189.

wav© steepness, ak
Figutue 1. Radian frequencies of the normal-mode perturbations having twice the wavelength
of the fundamental wave, shown as functions of the unperturbed amplitude ak.

A8 ak increases from zero, the steepness of the basic wave begins to affect
the frequency of the perturbations. It was shown by Longuet-Higgins & Phillips
( 1962 ) that the speed Cj of a gravity wave of wavenumber kx (relative to the
water at infinite depth) is affected by the presence of a second wave of amplitude
a2 and wavenumber k2. To order (a2k2)2 they found
_ Г± (y/kift—(a2^2)2ca ^1 > ^2*1 /g 2 )
^ i {я/к^ —{аг ^ 2с2{к1/к^) fcj < k2,J
1806

The instabilities of gravity waves. I I 493

assuming the second wave is travelling in the negative direction. By the same
argument as in § 7 of part I, this leads to a change in the frequency of the first wave,
as seen in a frame of reference travelling with the second. Identifying kx with
|n|and a2k2with ak we obtain

* _ / - 1 М И 02. M > 11 l33)

\(|«|-J)|n|(ofc)a, |n| < l.J

It will be noticed that when |n| = \ or 0 the coefficient of (ak)2 vanishes. This
accounts for the zero curvature in the branches n — 1/2 and n = 0 at their junctions
with the vertical axis.

wave steepness, ak
F igure 2. Rates of growth of the normal-mode perturbations having
twice the wavelength of the fundamental wave.
 1807

494 M. S. Longuet-Higgins
As ak increases from zero, it appears from figure 1 that the two branches n —3/2
and 7i = 1/2 coalesce, and the joint eigenfrequency is then of the form

cr = ± a ± i P

where a and p are real and we take p ^ 0. The mode corresponding to the positive
sign is then unstable. This instability, as we shall see later, is of the Benjamin-Feir
type.
As ak increases still further, however, stability is unexpectedly re-established
at about ak = 0.366. Thereafter, the frequency of the upper branch remains almost
constant at about or = 0.20, but the frequency of the lower branch continues the
downward trend begun by the branch n = 3/2 before it coalesced with n = 1/2.
Ultimately, at about ak = 0.406, the lower branch meets the branch n = 2/2,
which as we saw in part I represents a perturbation which simply shifts the phase
of the basic wave by a constant. At the junction a new instability springs up, with
a much higher rate of growth, as seen in figure 2. Though the corresponding eigen­
function is less well determined than the eigenfrequency, it does appear that this
second instability is much more localized than the first, and may represent the
initial stage of a plunging breaker.
In part I we showed that the junction of the modes n —4/2 and 2/2 probably
coincided with the calculated maximum in the phase speed с as a function of ak.
The junction of n = 3/2 and n = 2/2 is clearly at a different point. How are we to
understand the occurrence of an instability at this lower wave amplitude?
At the critical point the frequency of the perturbation is zero, so that the per­
turbed wave travels along with the same phase velocity as the unperturbed wave.
But the new perturbation is odd, not even. The perturbed wave may therefore
represent a train of waves of alternately higher and lower amplitude. Since we are
not at a velocity maximum, however, the natural increase in velocity of the higher
wave due to its greater steepness must be exactly counterbalanced by a decrease in
velocity owing to a shortening of its wavelength. What type of perturbation is
required in order to produce simultaneously an increase in height and decrease in
length of every alternate wave, and a decrease in height and increase in length of
the intervening waves ? On reflection it becomes clear that the perturbation must
have (at least) three maxima and three minima in every two wavelengths. Hence
we would expect the wavenumber n = 3/2 to be dominant.
Confirmation of this interpretation comes from the remark that the perturbation,
being odd, contains no even harmonics, in particular no harmonic n = 2/2. Now the
harmonic n = 2/2 = 1is simply the first harmonic of the fundamental, unperturbed
wave train. If the height of every alternate wave is to be increased without adding
to it any first harmonic the magnitude of the first harmonic must clearly be stationary
with respect to any increase in the wave height. But Schwartz ( 1974) foimcl by
computation that the amplitude a± of the first Fourier coefficient of a regular wave
train does indeed attain a maximum value, and diminishes again before the height
1808

The instabilities of gravity waves. I I 495

is a maximum. From figs. 3 and 8 of Schwartz’s paper it appears that the wave
steepness for which the first harmonic is a maximum is given by

2,a/L = 0.131.

This is equivalent to ak = 0.412, which is not far from the amplitude ak = 0.406,
where the onset of the new instability is found.
The above argument is not exact, because any alteration in the perturbed wave
train cannot be precisely the same as in a uniform train of waves; there must be a
slight distortion of the symmetry in order that neighbouring waves shall join
smoothly. Moreover, our argument would apply equally to all the higher har­
monics, which attain their maxima at nearly but not quite the same wave amplitude
(see Schwartz 1974, fig- 8).

4. T he general case m > 2

When m ^ 3 the normal modes may be separated into mutually exclusive cate­
gories in a somewhat analogous way. Let I be any integer such that

0 <l<\m. ( 4 .1 )

From equations (2 .2 ) and (2 .6 ), all the coefficients Я* and Cn will vanish except
those for which n is a multiple of m (see figure 3). From the addition rules for sines
and cosines it follows that all the coefficients Pn, Qn, Rn, Sn defined in (I) will also
vanish except when n = jm ,j = 0 , 1 , 2 , . . . . Hence the only non-vanishing coefficients
in the matrices A and В will be those corresponding to the elements

a jm ± l> b jm ±i> c j m ±l> d jm ± i

and to the coefficients of cos пв, sin пв when n = jm ± I. The situation is illustrated
in figure 3 for the case m = 7, I = 2 . Each value of n = jm for which a non-zero
coefficients exists is flanked by two values n = j m ± l for which there may be a
non-zero element of the eigenvector (Oj,...,dN) with the exception that when
m —0 , there is only one value n = I.
So, if we include, at the iVth stage, only the coefficients of cos пв and sin пв having
n ^ Nm, we obtain matrices Л * , B* of order 8N precisely, from which to determine
the normal modes. (When m = 2 , I = 1, the rows and columns coincide in pairs,
and the matrix is of order 4iV only.) This doubling of the order, from 4>N to 8N

I I I I 1 1

m m m
F ig u r e Harmonic components of the class of perturbations having wavenumbers
3.
n = \pm + ql\, where m and I are fixed and p and q are running integers.
 1809

496 M. S. Longuet-Higgins
means that to obtain a comparable accuracy, at any given value of N, almost
four times the core-storage is required compared to the requirement for calculating
the superharmonics. Nevertheless convergence of the eigenvalues to at least three
significant figures (and generally many more) was obtained for wave amplitudes
up to and including ak = 0.40.

6. R e s u l t s w h e n m > 2
In order to check the programming of the calculation in Fo r t r a n iv it was found
useful first to try the case m = 4, knowing that the modes corresponding to m = 4,
1 = 1 should be among the complete set of modes for m = 4. (For moderate ak, the
complete set can be found by a much simpler modification of the existing program
for the superharmonics.) This being verified precisely for moderate ak it was then
possible to carry out confidently the calculations for larger values of the parameter.

wave steepness, ak
Radian frequencies o f subharmonics corresponding to the case
F ig u r e 4.
m = 4 . Circular plots correspond to calculated values.
1810

The instabilities of gravity waves. I I 497

The results when m = 4, I = 1 are shown in figure 4, together with previous
results from figure 1. For clarity the vertical scale is again doubled. The branches
from figure 1 are those labelled n = 2/4, 4/4, 6/4 and 8/4; the new branches are
n = 1/4, 3/4, 6/4 and 7/4.
It will be seen that the branches n = 3/4 and 5/4 coalesce to form an unstable
mode similar to n = 2/4 and 6/4; and so also do n = 1/4 and 7/4. With increasing
amplitude ak, the unstable modes eventually become stable again, and one branch
continues down to form a new instability with the axis n = 4/4, at about the same
value of ak as before.
Consistently with our previous interpretation, we expect that the point where the
branch (1/4, 7/4) meets the axis will correspond to a mode in which one wave in
every four is increased in height and reduced in wavelength; and the other three are
decreased in height and increased in wavelength. So it may be designated by the
symbol ( + , —, —, —). The point where (3/4, 6/4) meets the axis will correspond
to a mode in which two successive waves in every four are increased in height (and
reduced in wavelength) and the next two are decreased in height. So it may be
designated by the symbol ( + , + , —, —). Moreover the corresponding perturbation
described in § 3 may be designated ( + , —, + , —). These symbols clearly exhaust
the possibilities, if we include any sign reversals or shifts in phase through an
integral number of wavelengths.
Figure 6 shows the complete set of eigenfrequencies when m = 8 . As Sn decreases
from 16 to 8 , so the limiting frequency
cr_ = \n—rd\ (6.1)

decreases from 0.586 to 0 . Then cr_ increases again to a maximum value 0.25 at
8n = 2 , before decreasing again towards zero in the limit n -+ 0 .
For every I less than m it appears that the modes n = l ± l /m coalesce to form an
unstable mode. As ak increases, this mode becomes stable again. Then one of the
stable branches descends to the axis (n — 8 / 8 ) where a new instability arises.
The corresponding rates of growth are shown in figure 6 . It will be seen that the
maximum growth rate in the Benjamin-Feir type of instability is limited to
P = Im (<7) < 0.0234. (5.2)
This value occurs with the mode n = ( 1/ 2 , 3/ 2 ) when the basic wave amplitude
ak « 0.32.
The second type of instability grows much more rapidly. For practical reasons
(see § 4) the rates of growth could be determined accurately only in the degenerate
case m — 2 and when n = ( 1/ 2 , 3/2), that is to say 8n = (4, 12 ). At an amplitude
ak = 0.42, for example we find
p = 0.126 (5.3)

which in one wave cycle corresponds to a magnification

e2*A = 2.21 (6.4)
 1811

498 M. S. Longuet-Higgins
or an increase of 121 %. (The increase corresponding to (6 .2 ) is only 14%.) The
trend of the curve on the right of figure 6 suggests that at larger values of ak the
rates of growth are still higher.

wave steepness, ak
F i g u r e 6. Radian frequencies of subharmonics when m = 8.

The points in figure 5 where pairs of modes coalesce and become unstable all
lie on a continuous locus as shown in figure 7. This forms the boundary between
stable and unstable parts of the diagram in the sense that in the ‘ unstable’ part
there is always some unstable mode at the given frequency and amplitude. The
boundary on the left appears to extend to the origin, and the tangent there, as we
shall see, corresponds to the Benjamin-Feir theory.
The boundary on the right of the unstable region is the locus of points in figure 5
where the tangent to the right hand branch of each curve is vertical. For the pair
1812

The instabilities of gravity waves. I I 499

8n = ( 1, 16) this point is difficult to determine accurately, but already the left
hand part of the curve has a very sharp curvature and we expect that the right
hand branch will have also, the two branches coming very close together. When
l/m < 1 /8 further computation shows that the lower branch n = l/m descends
towards the axis cr = 0 and becomes almost straight. Where it coalesces with the

wave steepness, ak
F i g u r e 6. Growth rates of subharmonics when m = 8 .

upper branch n = 2 —l/m the curvature becomes very sharp, indicating only a
slight degree of interaction. At a value of l/m marginally less than 1 /8 the two
branches appear to intersect, so that the left and right hand boundaries of the
region of instability are joined.
At its lower end, the right hand boundary of the unstable region appears to meet
the axis cr = 0 at about the point where ak = 0.347. It is to be noted that by an
application of Whitham’s average-Lagrangian technique, Lighthill ( 1967 ) found
 1813

500 M. S. Longuet-Higgins
that for very slowly varying wave trains at finite amplitude, the Benjamin-Feir
instability should appear when 2a /L — 0.054 approximately. This gives ak ж 0 .34 ,
in good agreement with the present value.

wave steepness, ak
F igu re 7. Stability diagram for gravity waves in deep water. The tangent at the origin
corresponds to the asymptotic theory of Benjamin & Feir ( 1967 ). (The resemblance to a
breaking wave is coincidental.)

6. T h e B e n j a m i n - F e i r i n s t a b i l i t y

The analysis proposed by Benjamin & Feir ( 1967 ) to explain the observed dis­
integration of wave trains in laboratory channels envisaged a primary, deep-water
wave of finite amplitude a and radian frequency 0) on which was superposed two
‘ side-bands’ of infinitesimal amplitude ea and of frequencies cu(l ± 8) respectively.
Their analysis indicated that the side-bands would tend to grow in amplitude at
the expense of the primary waves provided that
2{akf > (6 . 1)
The ultimate rate of growth of the side-bands was given by
d(lne)/d* = Zo)8\a?/al- 1)*, (6 .2 )
where a0 denotes the critical amplitude 8a>/k<j2.
The effect of the side-band amplification is to produce a modulation of the wave
envelope with a spatial wavenumber Aк related to $ by the group velocity:
8a)/Ak = dcr/dk = (o/2 k. (6.3)
1814

The instabilities of gravity waves. I I 501

Hence in units such that о) = к = 1,
Ak = 28. (6.4)
Observations suggest that the increasing depth of modulation may even cause the
total disintegration of the train of waves.
A different but equally illuminating description of the phenomenon was given by
Hasimoto & Ono ( 1966 ), who began by deriving the partial differential equation

Л — + <6-6)
for the complex wave envelope eA (^r) of a slowly varying wave train on deep
water.
Here £ and r are the dimensionless variables
£ = ek(x—tdcr/dk), т= e2t, (6 .6 )
and quantities of order e3 are neglected. On assuming a slightly modulated wave
train
A = a { l + a ') e t f , (6.7)

where a' and x' are small perturbations proportional to exp (iAк£ + /3т), one finds
after substituting in (6.5) and considering real and imaginary parts, that
^ = i(A ^ {W -i(A ^ } (6 .8 )
or from (6.4)
p = al(a2/a l ~ l)i, where a\ = J(Ak)2 = \82. (6.9)
So, as before, the motion becomes unstable when (ak)2 > \82. The calculated rate
of growth is similar.
The above theory is vaUd only asymptotically for sufficiently small values of the
wave steepness ak and of the modulation frequency 8 and wavenumber Ak. Note
that in a frame of reference travelling with the phase speed the apparent frequency
of the perturbations will be given by
A<r « \Ak±a\(a2/a l — 1)*. (6.10)
Now the two wavenumbers of each pair of modes which coalesce in the bottom
left hand comer of figure 5 are
n = t ± l / m = l ± A k = 1±28 (6.11)
according to equation (6.4). So to compare with the asymptotic theory we should
write
or = 28 = Ak = |w—1|. (6 . 12)
and from (6 . 1) the critical wave amplitude a0 is given by
<r = V2a 0 fc. (6.13)
Equation (6.13) corresponds to the straight asymptote through the origin in
figure 7. The calculated curve appears to touch this line at the origin. The deviations
from the line are of order {ak)i, however, not ak)2, so that in improving the asymp­
totic theory some care should be taken with the expansions in powers of ak or 8.
 1815

502 M. S. Longuet-Higgins
To account for this, we remark that the instability can be regarded as a resonance
brought about by the effect of the finite-amplitude wave on the frequencies of the
two side-bands (see Benjamin & Feir 1967, § 1). Explicitly, we have only to equate
the two expressions (3.3) for the velocity, the first with kx = 1 + Дк and the second
with kx = 1 —Ak, to obtain a result equivalent to (6.13). This is because for small ak,
the second-order corrections to the phase velocities are valid over the greater part
of the interval 0 < a < a0. However, in the neighbourhood of the critical point the
curves are approximately parabolic, introducing terms like (a0—a)t. It is these
singularities that are responsible for the deviations of the boundary from its
asymptote being of a lower order than suggested by Benjamin ( 1967 ).

7. C o m p a r i s o n w it h o b s e r v a t io n

Benjamin ( 1967 ) has reported experiments on the spatial rate of growth of side­
band instabilities in a train of waves generated in a laboratory channel. The
quantity plotted in his fig. 4 is d(lne)/d#, where z denotes kz/2n. To compare
with our results this must be converted to a time rate of change. We must assume
that d/dt ~ cgb/bz, where cg denotes the linear group velocity, although for finite-
amplitude waves the assumption is only approximately correct. Since in our units
cg = \ this implies that
d(lne)/d5 = 4Ttd(lne)/d< = 4я1т(о-). (7.1)
The experiments were carried out at a constant side-band frequency 8 = 0.10
and variable wave steepness ak. In figures 5 and 7 they therefore correspond to
points along the line гг = 0.10 parallel to the axis of ak. Our calculations of growth
rates in figure 6 were, however, carried out at fixed values of the nominal wave­
number n. To obtain the comparable rate of growth we must first find the appro­
priate (fractional) value of n and then interpolate in an appropriate manner.
First from figure 7 we see that the critical wave steepness a0k corresponding
to a frequency <r = 0.10 is a0k = 0.083 (as compared to 0.071 on the asymptotic
theory). For any given value of the wave steepness ak > a0kvre can determine the
appropriate value of n by interpolating along a line parallel to the stability boundary
in figure 7 . Thus when 0.083 ^ ak ^ 0.20, we find values of n ranging monotonically
from 9.63 to 9.75.
Now in figure 8 we have plotted the calculated values of
7} = Im(cr)/al against £ = а/я 0 (7.2)
at different values of n. The uppermost curve corresponds to the limiting value
8n = 8 (or to = 1). From equation (6.9) this curve is simply 7) = (f 2 —1)*. The remain­
ing curves show the extent to which the calculated growth rates depart from the
asymptotic theory. For any given value of ak we know the corresponding values
of £ and n so we may easily interpolate to find the appropriate value of rj. Then
from (7.1) and (7.2) we estimate
d(lne)/d# = 4na%7j. (7.3)
1816

The instabilities of gravity waves. I I 503

normalized steepness, (a/a0)

F ig u r e Calculated growth rates, plotted at constant values of the wavenumber n. The
8.
coordinates are £ = a/aQ, ij = Im (сг)/о®, where a0 denotes the critical wave amplitude.

wave steepness, ak
F ig u r e 9. Comparison of the calculated growth rate (solid curve) with the measured rates of
growth (plotted points) as reported by Benjamin (1967 ). The broken line represents the
asymptotic theory of Benjamin & Feir (1967 ).
 1817

504 M. S. Longuet-Higgins
The resulting curve for d(lne)/dx is shown in figure 9 (solid curve), compared
with the measured rates of growth according to Benjamin ( 1967 , fig. 4 ), and the
asymptotic theory of § 6 .
It will be seen at once that the present calculations agree rather well with the
observed growth rates, indeed more closely than the asymptotic theory with which
they were originally compared. The part of the axis to the left of the critical point
is strictly part of the theory, representing zero rate of growth. This passes through
the first plotted observation. The second, representing a positive growth rate, also
lies on the theoretical curve. The remaining points all lie somewhat below the curve,
perhaps on account of dissipation, which we have neglected.
The foregoing measurements all correspond to low values of the amplitude:
ak < 0.17. In preliminary experiments at higher wave amplitudes, using a wave-
maker of the conventional wedge or plunger type, the author has found that
steeper waves tend to become irregular, hindering accurate observations. No doubt
this is directly due to the fact that unless the wavemaker exactly matches the
particle-motion in a progressive wave, transient motions are generated which
quickly turn into instabilities as the waves progress.
One method of overcoming these difficulties and of reaching higher amplitudes
which may be more stable would be to design a wavemaker so as to match more
closely the nonlinear motions in steeper waves.
An attractive alternative is to conduct numerical experiments, possibly with the
forward integration technique devised by Longuet-Higgins & Cokelet ( 1976 ),
inserting the appropriate initial conditions. Investigations on these lines will be
reported in a subsequent paper.

8. D is c u s s io n

We saw in part I that a localized, superharmonic instability occurs near the

point when the phase speed is a maximum (i.e. ak та 0.436). This instability has
zero frequency Re (cr) but a high rate of growth Im (cr). The perturbed wave is of
greater height than the original. Energetically this is possible because at this
amplitude the energy of a steady wave is a decreasing function of the wave height.
The wave of maximum energy has an amplitude ak only 0.428 (Longuet-Higgins
1975)-
In this paper we have seen that the type of subharmonic instability discovered
by Benjamin & Feir depends upon a coalescence of the harmonics n = 1 + l/m.
But it occurs only over a certain range of wave amplitudes. Beyond this there is
first a more stable range of amplitudes. At still higher amplitudes there occurs a
second type of instability with a much higher growth rate, formed by the coalescence
o f the harmonics n - 1 + l/m and n = 1. This type is similar to the superharmonic
instability of part I, in that it has zero real frequency. It involves an increase in the
height of some individual waves, but also a decrease in height of the other waves.
Now, however, the increasing waves draw upon the energy of the decreasing waves
1818

The instabilities of gravity waves. I I 505

and so can grow even though the initial amplitude ak is less than the amplitude of
maximum energy for the average wave train.
In this investigation we have so far treated only inviscid, irrotational gravity
waves on deep water because this is the simplest case with the fewest possible
parameters. Physically the model may apply best to surface waves long enough for
surface tension and viscosity to be negligible. Nevertheless the analysis clearly
needs to be extended so as to include capillarity and the important effects of surface
shear.
In water of finite depth h the work of Whitham ( 1967 ) and Benjamin ( 1967 )
suggests that the subharmonic instabilities may be suppressed at sufficiently small
values of kh. On the other hand from observations in shallow water one would
expect the local, superharmonic instabilities to be enhanced. Further investigations
of these effects are clearly desirable.

R eferences

Benjamin, Т. B. 1967 Instability of periodic wavetrains in nonlinear dispersive systems.

Proc. R. Soc. Lond. A 299, 59-67.
Benjamin, Т. B. & Feir 1967 The disintegration of wave trains on deep water. 1. Theory.
J. Fluid Mech. 27, 417-430.
Haeimoto, H. <fe Ono, H . 1972 Nonlinear modulation of gravity waves. J. Phys. Soc. Japan
33, 805-811.
Lighthill, M. J. 1967 Some special cases treated by the Whitham theory. Proc. R. Soc. Lond.
A 299, 28-53.
Longuet-Higgins, M. S. 1975 Integral properties of periodic gravity waves of finite ampli­
tude. Proc. R. Soc. Lond. A 342, 157-174.
Longuet-Higgins, M. S. 1978 The instabilities of gravity waves of finite amplitude in deep
water. I. Superharmonics. Proc. R. Soc. Lond. A 360, 471-488.
Longuet-Higgins, M. S. & Cokelet, E. D. 1976 The deformation of steep surface waves on
water. I. A numerical method of computation. Proc. R. Soc. Lond. A 350, 1-26.
Longuet-Higgins, M. S. & Phillips, О. M. 1962 Phase velocity effects in tertiary wave
interactions. J. Fluid Mech. 12, 333-336.
Schwartz, L. W . 1974 Computer extension and analytic continuation of Stokes’s expansion
for gravity waves. J. Fluid Mech. 62, 553-578.
Whitham, G. B. 1967 Variational methods and applications to water waves. Proc. R. Soc.
Lond. A 299, 6-25.
 1819

Proc. R. Soc. Lond. A. 364, 1-28 (1978)

Printed in Great Britain

The deformation of steep surface waves on water

II. Growth of normal-mode instabilitiesf

B y M. S. L onguet -H ig g in s :): and E. D. Co k elet §

X Department of Applied Mathematics and Theoretical Physics,

Silver Street, Cambridge, U.K.
t§ Institute of Oceanographic Sciences, Wormley, Surrey, U.K.

(Received 3 April 1978)

Studies of the normal-mode perturbations of steep gravity waves (Longuet-

Higgins 19786 , c) have suggested two distinct types of instability: at low
wave steepnesses we find subharmonic instabilities with fairly low rates
of growth, and at higher wave steepnesses there are apparently local
(‘ superharmonic ’ ) instabilities leading directly to wave breaking. Between
these two types of instability is an intermediate range of wave steepnesses
where the unperturbed wave train is neutrally stable.
In the present paper we employ the time-stepping method of an earlier
paper (Longuet-Higgins & Cokelet 1976 ) to test the rate of growth of each
type of instability. For the initial linear stages of each instability, the
computed rates of growth are accurately confirmed, and it is verified that
the local instability does indeed lead to breaking.
The later nonlinear stages of the subharmonic instabilities are further
investigated. In the two examples so far computed it is found that the
gradual rates of growth of the subharmonic instabilities are maintained,
and that ultimately every alternate crest develops a fast-growing local
instability which quickly leads to breaking.

1. I n t r o d u c t io n

The task of understanding and predicting the formation of breaking waves on deep
water has recently been approached in two different ways. On the one hand a
general method has been developed for calculating the deformation of the free
surface in any irrotational motion which is periodic in the horizontal coordinate
(Longuet-Higgins & Cokelet 1976 ; referred to as paper I). This has been applied
successfully to a variety of initial conditions (see I and also Cokelet 1977 ). On the
other hand a more analytical approach (Longuet-Higgins 19786 , c) has yielded some
understanding of the initial rates of growth of any small perturbations to a uniform
train of waves o f finite amplitude.
The conclusions reached in the more recent papers (Longuet-Higgins 19786 , c)
were somewhat surprising. For instance it was found that there were at least two
distinct types of instability: subharmonic instabilities of the Benjamin-Feir type,
t Part I appeared in Proc. R. Soc. Lond. A 350, 1-26.

1 Vol. 364. A. (12 December 1978)

1820

2 M. S. Longuet-Higgins and E. D. Cokelet

which were however confined to waves whose ‘ steepness ’ ak lay within a certain
finite range (the upper limit being at ak « 0.37 and the maximum growth-rate at
ak « 0.32); and secondly, local or superharmonic instabilities which first appeared
when ak «0 .4 1 . These had much higher rates of growth, and it was suggested that
they led directly to overturning of the free surface, that is to say breaking waves.
With regard to the subharmonic instabilities, rather good confirmation of the
calculated growth rates was provided by the observations reported by Benjamin
( 1967 ). But these were only for small values of the wave steepness (ak < 0.17). At
higher values of ak no direct measurements are yet available.
The calculations reported by Longuet-Higgins ( 19786 , c) were lengthy. Because
of the need to verify the somewhat unexpected conclusions relating to instabilities
at higher values of ak, the present authors have, in this paper, set out to determine
the initial growth-rates of each type of instability by the quite independent time-
stepping method of paper I. This method is indeed well suited to the present prob­
lem whenever the initial perturbation has a length-scale equal to an integral number
m of wavelengths of the unperturbed wave. Here we concentrate particularly on
the case m = 2 when the perturbations are two wavelengths long. This case has the
advantage that the numerical calculations are the most accurate for any value of
m at the higher values of ak, where confirmation is most needed. Also, they display
all the unexpected features mentioned earlier.
These features are set out in more detail in § 2 . In § 3 we briefly describe the time-
stepping method, to be applied in the present instance. Sections 4 and 5 describe
the main results as they relate to the initial rates of growth. From table 1 and figure 5
it will be seen that the agreement between the two methods is remarkably good.
But the time-stepping method used in the present paper has the further advantage
that (unlike perturbation analysis) its validity is not limited to small perturbations
of the finite-amplitude wave. Being without analytic approximation, the numerical
calculation can be carried to the point when the perturbations themselves are of
finite amplitude. This is done in §§ 6 and 7. As a result it is found, first, that the
‘ local’ instabilities do indeed lead directly to wave breaking, in which the final
overturning takes place very rapidly (figures 15 and 18). Secondly, the subharmonic
instabilities, when followed to their later stages of growth, develop local instabilities
at every alternate wave crest. These local instabilities are precisely similar to the
local instabilities found previously. In other words, the disintegration of the wave
train, which is brought about by the subharmonic instabilities, extends as far as
to make the waves break, in all those cases that we have tried. Moreover, the
computed behaviour of the wave crests is so similar in all cases that there is hope
for a local, asymptotic theory of the dynamics.
Finally we should like to emphasize that, unlike those in paper I, the waves
treated here are theoretically free waves, with constant and uniform pressure at
the free surface. Thus there is no input of energy; we are studying the dynamics o f
breaking waves under the simplest possible conditions.
 1821

The deformation of steep surface waves. I I 3

2. The n o rm a l-m o d e a n a ly sis

In the perturbation analysis of Longuet-Higgins (1978 6 ,c) the normal modes were
calculated in the form ж _ Х{ф> f ) + ^ f ) e_lrt

У = Y (ф, ф) + е Щ , ф) e~i<rtJ 1*

where ф and \jr denote the velocity potential and stream function in a frame of
reference moving with the speed с of the unperturbed wave, and x ,y are rect­
angular coordinates (x horizontal, у vertically upwards) expressed in terms of ф, rjr
and the time t as independent variables. X and Y represent the coordinates of the
unperturbed wave, and £, fj the normalized perturbation, e being an arbitrary small
parameter. The free surface, which in the unperturbed wave is given by \jr = 0 , is
given by f = е /(ф )е ~ м (2.2)

in the perturbed state. In fact £, ij and/have both real and imaginary parts, corre­
sponding to the ‘ in-phase’ and ‘ quadrature’ components of the normal mode, and
each of these components is given, in turn, by Fourier series in cos {1ф/с) and in
sin (1ф/с), where I is a non-negative integer and фruns from 0 to 2mn/c.
Throughout this paper the units of length and time are normalized by taking
g = 1 and the wavelength of the unperturbed wave to be 2 k .
It was shown by Longuet-Higgins ( 19786 , c) that each normal mode may be
designated by one or more rational numbers (mode numbers) which indicate its
principal wavenumber components when ak is small. For example n = (£, J) de­
notes a subharmonic normal mode with principal components having wavenumbers
\ and §. An index + or — may be added to designate that the perturbation is
either growing or decaying, respectively.
In general cr is a complex quantity, and the frequencies occur in conjugate pairs.
The unstable modes are those for which <r has a positive imaginary part. The cal­
culations revealed two distinct types of instability. The first are disharmonies, with
wavelengths greater than that of the unperturbed wave. At small wave steepness
ak these are clearly of the Benjamin-Feir type (Benjamin 1967). If we imagine the
side-band frequency Д<r as fixed and the steepness ak of the unperturbed wave to
increase gradually, then these subharmonic modes first become unstable when ak
slightly exceeds ДсгД/2 . The maximum rate of growth is found for a normal per­
turbation two wavelengths long, at an amplitude ak ж 0.32. But when ak slightly
exceeds 0.37 the perturbation apparently becomes neutrally stable again: the
instability disappears (a result first predicted by Lighthill ( 1967)). Beyond ak = 0.37
the normal-mode analysis of Longuet-Higgins (1978 c) indicates a narrow range of
neutral stability, culminating in the onset of a second type of instability which (it
was suggested) is more local, being concentrated near the crests of individual waves.
For this type, cr is pure imaginary, and the initial rate of growth is much higher.
It was foreseen that these instabilities might develop directly into breaking waves.

1-2
1822

4 M. S. Longuet-Higgins and E. T>. Cokelet

3. T he t i m e -s t e p p i n g method

In the present paper we aim to test and extend these results by the entirely
independent time-stepping method developed in I. This method is based on the
fact that in an irrotational, incompressible fluid the kinematic and dynamic
boundary conditions describe the evolution of the flow in terms of quantities speci­
fied only on the boundaries. For deep-water waves, the only boundary is the free
surface, the region of greatest interest. Its location is given by the coordinates,
= 1, 2 of specified fluid particles. In order to follow the motions
of the surface we need know only the tangential and normal velocities of the par­
ticles on it. Given the value of the velocity potential ф* along the surface we find the
tangential velocity 00*/0s by numerical differentiation. To find the normal velocity
00*/0n we use Green’s third identity ((4.2) of paper I) to derive an integral equation
for дф*/дп in terms of other surface-evaluated quantities. Having solved this
equation numerically we march forward in time with the evolution equations
((2.7) and (2.10) of paper I).
The motion is assumed to be periodic in the horizontal coordinate, though not
necessarily in the time t. This assumption enables us to transform the semi-infinite
region of fluid ( ж * , у * - р 1 а п е ) into a finite enclosed region (£-plane) such that the
free surface maps to the exterior boundary (see § 3 of I). The basic numerical
computations are performed in this transformed plane.
Evidently the time-stepping method can be adapted without essential change to
the present problem, provided the spatial period L contains m wavelengths of the
unperturbed motion (ra an integer). In the present paper we confine ourselves to
the case ra = 2 , first because the initial perturbations, as calculated by the normal-
mode method (Longuet-Higgins 1978 b, c), are given most accurately when ra = 1
or 2 , at the larger values of ak. Secondly, the case ra = 1 excludes subharmonic
instabilities and yields only the localized instabilities; and these occur at slightly
higher values of ak than when ra = 2 so the accuracy is correspondingly less. By
considering ra = 2 we shall include both types of instability, and with greater
initial accuracy.
A third advantage of the case ra = 2 is that, of all integer values of ra, this yields
the subharmonic instability having the highest growth-rate, namely when

n = (h$)> ak = 0.32.
In the time-stepping method the reference frame is stationary, relative to the
water at infinite depth. So if we denote the velocity-potential and stream-function
by ф*, ifr*, and the rectangular coordinates by x*, y*, we have the relationsf
x* = x + ct, ф* = ф+ сх*Л
У* = У, т
/r* = i/r + cy*.)

t In (Longuet-Higgins 19 78 b, c) the undisturbed waves were taken to be propagating in

the negative ж-direction. Here we assume they propagate in the positive sense.
 1823

The deformation of steep surface waves. I I 5

To begin the time-stepping we need only to know the values of x*, y* and <j>* when
\jr = e/, as a function of some parameter which increases monotonically along the
free surface. (Here one can take this to be ф/с in the first place.) Now on the free
surface i/r = e f we have from (2 . 1) to first order,
* = Х(ф, 0 ) +е/(ф ) Х+(ф, 0 )+е£(ф, 0 )Л
у = Г {ф ,0)+ е}(ф )Г ^ ф ,0 )+ еЩ ,0 ), } ' ’
when t = 0 ; and it remains now only to write
x * = x, у* - у, ф* = ф + сх, (3.3)
where x and у are given by (3.2).
The computation points were at first taken at N equally spaced values of ф
along the free surface. Later it was found that improved accuracy could be ob­
tained by distributing the points at equally spaced intervals of arc length in the
transformed £-plane (see I, § 3). Normally N was taken as 90 but occasionally 60
or 120 according to the accuracy required. After the initial distribution, the same
points were followed as Lagrangian marker points throughout the rest of the com­
putation. This proved beneficial because they tended to concentrate at the crests,
the regions of largest curvature.
The computations were programmed in Fo r t r a n i v and performed on Ruther­
ford High Energy Laboratory’s IBM 360/195 digital computer. The array storage
requirements were proportional to (JV2-h 77iNT) and in double precision (8 bytes/
word) the core storage (arrays + buffers + system overheads) was 310 К bytes
(1 К = 1024) for N = 90. The computations were first limited by the availability
and expense of computer time. With N = 90 the time required (ca. N 2) for one
solution of the integral equation (4.5) of I was 2.2 s. This usually had to be solved
twice per time-step when Adams-Bashforth-Moulton time-stepping was used.
The time-step At was adjusted so that no particle moved further in a time-step
than the distance between itself and an adjacent particle. When At was altered,
three Runge-Kutta steps requiring four solutions of the integral equation were
taken. The amount of computer time per run varied a great deal, but generally
higher waves took longer because of large particle velocities and hence smaller
time-steps. In the examples to be described, the runs typically required 5-10 min of
machine time.
Throughout the computation, the surface pressure was assumed to be zero.

4. E xam ple

Now as an example of the time-stepping computation described in § 3 we show

in figure 1 the results of a computation for the perturbation n = (J, f )+ when
ak = 0.32. According to the normal-mode calculations (Longuet-Higgins 19786 , c)
this should in fact be the fastest-growing mode of the Benjamin—Feir type. In the
left hand column of figure 1 are shown two successive crests of the perturbed wave,
1824

6 M. S. Longuet-Higgins and E. D. Cokelet

with time increasing down the page. At the top right is shown the initial pertur­
bation, according to equation (2.3) butwith the vertical scale multiplied by a factor 5,
for clarity. Below this, corresponding to each profile on the left, is the perturbation
calculated by taking the difference between the profile on the left and the unper­
turbed wave advanced with the theoretical phase-speed c.
It will be noticed, first, that every alternate crest of the original profile is higher,
in general, and the intermediate crests are lower. This is because the perturbation
is an odd one, being equal, but of opposite sign, on adjacent crests.
Next it will be seen that after half a cycle of the perturbation (namely at the foot
of figure 1) the waves that were the higher have become lower, and vice versa.

t= 0

Ь -

t=\T

F igube 1 . Development of the subharmonic instability $)+ over one half-period of the
perturbation, when ak = 0.32. The left hand column shows the time-stepped surface
profiles. The right hand column shows the resulting perturbation, enlarged vertically
by a factor 5. Time increases downwards. In this case N = 90, e = 0.025.
 1825

О х 4тг О х 4тг

F ig u b e 2. As in figure 1, but with e = 0 (zero perturbation). This tests the accuracy of the
time-stepping technique. On the right, the vertical scale is enlarged times 20 .

This is because after a half-cycle the perturbation is exactly proportional to its

initial value, but of opposite sign.
Third it will be seen, by comparing the perturbations at the top and bottom of
figure 1, that the magnitude of the perturbation has indeed increased.
Figure 2 shows a precisely similar computation but with e = 0 , that is, the initial
perturbation is zero. This is simply a test of the accuracy of the time-stepping. After
the same interval of time the perturbation shown on the right (magn. x 20 ) has
in fact grown very little; it is a measure of the error inherent in the calculation.
The r.m.s. amplitude of the perturbation was less than 4 x 10-4.
Now figure 3 shows the companion mode n = (J, f)~ which is expected to decay.
Again, comparison of the perturbations at top and bottom of the figure shows that
the perturbation is indeed diminished.
1826

F ig u r e 3. Development of the decaying mode (£, $)- when ak = 0.32.

Parameters as in figure 1 .

5. Calculation of the grow th rate

To determine the rate of growth of the perturbation h(x*, t) over a given interval
of time, say (0 , t)ywe aim to determine the ratio J?(i) of the root-mean-square values
of h(x*t t) and h(x*, 0). By definitionf

{R(t)Y = {h(x,0)Ydx. (5.1)

It is also useful to define the correlation coefficient between h(x, t) and h(x + * 0. °) by
n, « \h(x,t)h(x + x 0>0 )da:
C(xQ,t) = ----------------------- ;-------------------------------- (5. 2)
[{W ^, 0 }2c b j + 0 )}2da:]*
t In this section x and у are written for x* and y*. This should cause no confusion with the
notation of 5 2 .
 1827

The deformation of steep surface waves. I I 9

We might expect the maximum correlation between h(x,t) and k(x + x0, 0 ) when
x0 = ct, where с is the phase velocity. However, since ^e-I<rt has both a real and
an imaginary part, corresponding to components in-phase and in quadrature with
the initial perturbation, the maximum (absolute) correlation will in general fluc­
tuate in time. If T denotes the period of the perturbation (T = 2л/R e (<r)) then at
times t = sT , where s is an integer, we should expect C{ct, t) to be near unity; while
at times t = (s + ^)T we should expect C(ct, t) to be nearly —1. Provided the maxi­
mum value of |C(a:0,<)| is near unity, then a useful measure of the growth-rate
should be derived from the corresponding value of jR(f). In fact we would expect
that while the perturbation is still small
Я (0 * « Л p = Im(o'). (5.3)
The actual determination of the integrals in (5.1) and (5.2) must be done with
care. The computer program begins at t = 0 , and has its own method of determining
the time-step length, At. In general the flow will not be computed at exactly
t = \sT , but rather at times before and after this. Hence it is necessary to inter­
polate the particle positions to this time. This can be done in a variety of ways.
Let us assume we have a simple equation of the form
dy/dt = /( 0 , (5.4)
and we store у and / at every time-step. Our time-stepping technique is fourth-
order (local errors ~ (A<)5) which means it would be consistent to use five pieces
of information about у and / for interpolation. However it seems a good idea to
use information only at times tx and t2 which bracket the time of interest, t. This
implies a cubic interpolation formula (errors ~ (ДО4) based on y v y2, f x and f 2
(here subscripts denote time levels). We tried interpolating the particle positions
using such a formula, and it was clear that the results were ‘ noisy*. A linear inter­
polation (errors ~ (At)2) based on just the particle positions themselves was much
smoother. This is to be expected since the displacements are calculated from the
time derivatives by integration, a smoothing process. In order to have a smooth
interpolant y(t), and to make use of both the displacement and derivative infor­
mation at tx and t2, we took the average of two quadratics each fitted through tx and
t2 but with one using f x and the other using f 2. Thus we have

y(t) = At2+ Bt + D , (5.5)

where A f 2- f x
Л -~ Ш '
D / 1 + / 2 , Уг~~У\ , (5.6)
~ 2 At Д< ’

d = - m \ + Ч) - № + « ■ ,

This interpolation formula is exact for quadratics and has local errors which vary
as (At)3.
1828

10 M. S. Longuet-Higgins and E. D. Cokelet

Once the positions of the fluid particles are interpolated to the required time we
wish to determine the shape of the perturbation. The free surface is specified by

у = H(x,t) + h(x,t), (5.7)

where H represents the contribution from an unperturbed steady wave and h

represents the perturbation. The steady wave profile is calculated by the technique
of Longuet-Higgins ( 1978 a) at t = 0, and for some later time it is just horizontally
displaced by an amount ct where с is its phase speed. The perturbed wave profile is
represented by the collection of points {x^y^, and to calculate
hj = h(Xj, t) we simply subtract H} = H(xj}t) from y f. However in practice we
represent H(x , t) by another collection of N points, and the horizontal coordinates
of these will not necessarily coincide with the x^ Therefore we use a periodic cubic
spline to interpolate H(xt t) in space.

F ig u b e 4. The correlation coefficient C{ct, i) between k(x, () and h(x + ct, 0) as a function of
the time t for the perturbation (-Jr, f)+ when ak = 0.32 (see figure 1).

We calculated the integrals in (5.1) and (5.2) by interpolating h(x,t) to equally

spaced values of x with periodic cubic splines, and then using the Simpson rule.
The value of C(ct, t) is plotted in figure 4 as a function of the time t, for the particular
mode n = (J, J)+ discussed in § 4. It can be seen that C(ct,t) does indeed fluctuate
in time, with extrema at about t = \sT (в = 0 ,1 , 2 ,...) as expected.
Of particular interest is the value of R(t) at t = \sT (« = 1,2,3,...) and the corre­
sponding rate of growth
= m in ify ).
For ak = 0.32, n = (J, §)+ we find that, when t = \T then ft' = 0.0254 and when
t = T, then ft' = 0.0254. This is to be compared with the value p — 0.0234 obtained
by the normal-mode analysis (see table 1).
 1829

The deformation of steep surface waves. I I 11

T a b l e 1. C o m p a r i s o n o f c a l c u l a t e d g r o w t h -r a t e s o f n o r m a l - m o d e

pertur batio ns

time-stepping with
normal-mode calculation integral equation
, ------------------------------- * -------------------------------4
i-------------- A--------------*
* \T fi ft'
ak c(k/g)* n Re (<г) (я/а) Im (cr) C(cl, t) m t~l \nR
.2072 15.17 .0000 -.9 9 9 0.999 - .0001
0.10 1.005013
ii .2676 11.74 .0000 - .9 9 7 0.995 - .0004
.2098 14.97 .0000 -.9 9 6 1.002 .0001
0.20 1.020203
(I .2406 13.06 .0000 -.9 9 0 1.000 .0000
fib i)+ .2150 14.61 .0132 -.9 9 2 1.219 .0135
0.25 1.031746
НЫ Г .2150 14.61 -.0 1 3 2 -.9 8 0 0.833 -.0 1 2 5
/(*. i)+ .1931 16.27 .0234 -.9 6 9 1.512 .0254
0.32 1.052512
Hi. i>- .1931 16.27 -.0 2 3 4 -.7 8 6 0.752 -.0 1 7 5
.1804 17.41 .0000 -.9 9 4 0.985 - .0009
0.38 1.074399
(f .1309 24.00 .0000 - .8 5 3 0.909 - .0040
0.40 1.082225 * .1840 17.07 .0000 -.9 0 8 1.118 .0065
0.41 1.086045 (*)+ .0000 oo .065

6. R esults

Similar calculations were carried out for the modes n = (£, j ) at the unperturbed
wave amplitudes ak = 0.10, 0.20, 0.25, 0.32, 0.38, 0.40 and 0.41. For the normal
modes under consideration the perturbation theory (Longuet-Higgins 1978 c) pre­
dicts that at the two lowest values of ak the waves will be stable, at the next two
they will be unstable, at the next two they will be stable again, and at the highest
value of ak they will be violently unstable. This general behaviour is confirmed by
our computations. Figure 5 shows a graph of the growth-rate /3 plotted against the
unperturbed wave amplitude ak. The solid curve represents the growth-rates
according to Longuet-Higgins (1978 b), and the points represent the present results.
The surprising prediction of a return to stability with increasing wave amplitude
is confirmed reasonably well. Departures of the observed growth-rates from the
predicted ones are probably due to computational inaccuracies, although the finite
amplitude of the perturbation may play a role. The results for the highly unstable
mode at ak = 0.41 are not shown on figure 5 but will be discussed later. In all
cases except ak = 0.32 we took e — 0.0125.
Table 1 lists the quantitative results of the calculations. The first six columns
give the unperturbed wave amplitude and phase speed, the mode of the pertur­
bation, its frequency in a reference frame moving with the unperturbed wave, the
semi-period of the perturbation and the growth-rate. These quantities are all
derived from the normal-mode theory. The last three columns give the maximum
correlation coefficient, the r.m.s. amplitude ratio and the observed growth-rate,
all evaluated at t = \T. We shall now discuss the results in detail beginning with
the lowest waves.
1830

12 M. S. Longuet-Higgins and E. D. Cokelet

At ak = 0.10, the modes n = \ and n = $ display a negligible growth-rate, as
shown in the last column of table 1. The correlation coefficients are —0.999 and
- 0.997 respectively, after one semi-period. These imply very little change of
shape, and this is verified by an inspection of the profiles in figures 6 and 7. On the
left of each figure are the complete wave profiles, plotted every one tenth of a semi­
period. The reference frame is at rest, i.e. the average horizontal fluid velocity is
zero at any point which lies always below the free surface. The perturbation
profiles h(x,t)t plotted with a vertical exaggeration of 20:1, are on the right. Note
how in figure 6 the smooth profile of h is very nearly a sine wave of length 4л. This

F ig u r e 5. The linear growth-rate /? of the normal perturbations n = J and (£, f)* as a

function of the steepness ak of the unperturbed wave train. Curves represent the normal-
mode calculations of Longuet-Higgins (19786 ). Plotted points are from the time-
stepping method of the present paper. O, n = \ and ($, §)+; x , n = } and (|, f)“ .
 1831

The deformation o f steep surface waves. I I 13

]
E- - - - - - - - - - - - 3- - - - E— - - - - - - - - H
E- - - - - - - - - - - - 3- - - - E- - - - - - - - - - - - - 3
E- - - - - - - - - - - - 3- - - - E— - - - - - - - - — j
E- - - - - - - - - - - - 3- - - - E- - - - - - - - - - - 4
E- - - - - - - - - - - - 3- - - - E-- - - - - - - - - - - - - 3
------------------------------------------- -------] t= \ T J----------------------------------- ------------- =
O x 4тг 0 x 4i r

F ig u r e 6. The mode n = $ when ak = 0.10 followed over one half-cycle of the perturbation.
The left hand column shows the time-stepped profiles. The right hand column shows the
resulting perturbation enlarged vertically by a factor 20. Time increases downwards.
e = 0.0126.

is to be expected since this mode is exactly a sine wave in the limit as ak -> 0 (which
is the basis for the notation n = J, signifying one sine wave in two fundamental
wavelengths). Likewise the profile of the mode n = J in figure 7 is predominant^
three sine waves.
The profiles for ak = 0 .20 , n = J and f are illustrated in figures 8 and 9. Com­
paring these to figures 6 and 7 we see that the increased nonlinearity of the un­
perturbed waves enhances the higher harmonic content of the perturbation profiles,
local bumps and dips begin to appear. The correlation coefficients are —0.996 and
— 0.990, and the growth-rates are 0.0001 and 0.0000 for n = \ and f respectively.
This extremely good agreement with theory is heartening when it is realized that
1832

14 M. S. Longuet-Higgins and E. D. Cokelet

0.05
t—o
-0.05

Figure 7. The mode n = | when ak = 0.10 followed over one half-cycle

of the perturbation.

we first compute the complete wave profile (unperturbed wave + perturbation),

interpolate in time, interpolate in space, subtract the unperturbed wave, and then
make the comparison to find С andR.
For ak = 0.25 theory predicts one growing mode n = (J,$)+ and one decaying
mode n = (£, \)~, which indeed are found. The profiles for the growing mode are
displayed in figure 10 in which the localized nature of the extrema are more appar­
ent than before. The correlations at t = \T still exceed 0.98 in magnitude. The
growth-rates differ from those predicted by at most 5 %, with the decaying mode
diminishing more slowly than predicted. This is not surprising since any numerical
inaccuracies can feed energy to a potentially growing mode and so obscure the
decaying one.
 1833

The deformation of steep surface waves. I I 15

10 .05.
t= 0
J- Q 0 5

4i r 0 4 it

F ig u r e 8. The mode n = £ when ak = 0 .2 0 followed over one half-cycle

of the perturbation. Compare figure 6 .

The modes (|, f)* at ak = 0.32 have already been discussed in § 3. (£, f)+ is of
particular interest since it represents the fastest-growing Benjamin-Feir mode.
One of the unexpected results of the normal-mode theory is that a wave whose
steepness is near ak = 0.38 will be stable to the modes n = £ and n — |. The results
given in table 1 and plotted in figure 11 certainly suggest this. The correlation
coefficient equals —0.994 for n = and there is a very slight decay in amplitude.
We observe a lower correlation coefficient and an increased rate of decay from
w = f . This is probably due to a variety of factors such as: ( 1) insufficiently con­
verged eigenfunctions used as initial values, (2 ) the increased size of the pertur­
bation amplitude for the mode n = § due to the manner in which the eigenfunctions
were normalized, and (3) inaccuracies in the time-stepping. Whatever the cause,
1834

16 M. S. Longuet-Higgins and E. D. Cokelet

t=\T

F ig u r e 9. The mode n = $ when ak = 0.20 followed over one half-cycle

of the perturbation. Compare figure 7.

the graph of С against t reveals a slightly noisy signal. The profiles for the two
modes are shown in figures 11 and 12.
The normal-mode theory predicts that ak = 0.40 is just on the neutrally stable
side of a very rapidly growing instability. We have tested the n = £ mode and have
observed growth but at a rate significantly slower than that of the unstable waves
on either side. This discrepancy is probably due to the factors mentioned previously
and to the close proximity of the rapid-growth region. The profiles are shown in
figure 13. One interesting feature of these is the near step-function shape of the
perturbation at t = 0 and t = \T = 17.07. This is also apparent in the n = J profiles
for ak = 0.38 at t = \T = 8.71 as shown in figure 11. The perturbation elevates
one wave trough and lowers the next. This also happens but to a lesser extent for
 1835

The deformation o f steep surface waves. I I 17

0.05
f=0
0.05

t=\T

F ig u r e 10 . Development of the unstable mode ($, §)+ when ak = 0.25, over one half-cycle of
the perturbation. The vertical scale on the right is enlarged times 20.

n = f, ak = 0.38 at / = \T = 12.00 . We have not calculated the mode n = | for

ak = 0.40 because the relatively low frequency of the perturbation would have
required an excessive amount of computer time, and the results would probably
be too inaccurate to justify this.
The instability at ak = 0.41 differs from the others considered in two ways: ( 1)
it has zero frequency with respect to the unperturbed waves (i.e. it remains ‘ locked
on ’ to the crest) and (2 ) it grows very much more rapidly. The first fact means that
we can measure its growth-rate at every time-step, and the second means we need
only follow the v/ave a short time before it breaks. The profiles (with a 5 :1 vertical
exaggeration of the perturbation) are plotted in figure 14 over a non-dimensional
time of slightly more than 2 units. The nature of the perturbation is such as to
1836

18 M. S. Longuet-Higgins and E. D. Cokelet

-,0.05
t= 0

F ig u r e 1 1 . The neutrally stable mode n = £ when ak = 0.38 followed over one half-cycle of
the perturbation. The vertical scale on the right is enlarged times 20.

reduce one wave crest and raise the other with not much alteration of the profile
elsewhere. The crest-plots of figure 15 show the overturning. These computations
were done with 120 points along the profile for increased resolution. The comparison
of theoretical and observed growth-rates is given in figure 16, a plot of In R against
time. The spread of observed values at each data point represents the variability
of R for slightly different methods of calculating the integrals. This gives a rough
indication of the minimum size of the errors involved. The agreement with the
linear theory is quite good especially during the initial stages of growth. At later
times the deviations are not too great even when the perturbation has grown large
enough to cause marked asymmetries in the wave profiles.
 1837

The deformation o f steep surface waves. I I 19

F igure 12. The neutrally stable mode n = f when ak = 0.38; as in figure 1 1 .

7. T he n o n l in e a r stages of grow th

Since the time-stepping method is not subject to any assumptions of linearity,

we may use it to follow the later, nonlinear stages of the instabilities studied earlier.
In figure 17 we show the mode (£, f ) + at ak = 0.32, carried through a second half­
cycle 0.65 T < t < 1.05 T. It will be seen that in the final stages every alternate
crest ultimately steepens and breaks. Figure 18 shows a close-up of the crest in a
frame of reference which is at rest relative to deep water. It will be seen how like
the profiles are to those in figure 15. Indeed, the two figures are almost superposable.
It follows that when ak = 0.32 this instability, which began as a subharmonic of
Benjamin-Feir type, later develops a different, local type of instability which is in
1838

F ig u r e 13. Development of the mode n = £ when ak — 0.40.

all respects similar to that arising at ak = 0.41. Since wave breaking involves
energy loss, this leads to the disintegration of the original wave train.
To test whether the same phenomenon occurs with lower waves, a similar run
was carried out on the mode (J, §)+ at the lower wave amplitude ak = 0.25. Figure
19 shows the surface profile, after nine successive half-periods. In the final profile
the waves do indeed break (see also figure 20 ). A close-up of the breaking crest is
shown in figure 21 . Again, the close similarity to the local instability in figure 15
will be noted.
The calculated values of G(ct, t) and ft are shown in table 2 . The constancy of ft
throughout the whole stage of development is indeed remarkable, as also is the
closeness of ft to the value 0.0132 derived from the independent normal-mode
analysis.
 1839

The deformation of steep surface waves. 11 21

0.2
*=o
4 0 h

- 0.2

4
v-"
h -

h -

h - 4
h -

H -

H -

H -

2.05
I— ^
0 , 4*
F io u h e 14. Development of the highly unstable mode n = (|)+ when ak = 0.41.
The vertical scale on the right is magnified times 6 .

Do all subharmonic instabilities lead to breaking? Because of the necessary

machine time, the question cannot readily be answered by computation. One may
reason as follows. It appears (from Longuet-Higgins 19786 ) that an individual
gravity wave becomes unstable when ak « 0.436. Roughly, then, one would expect
an instability to lead to wave breaking if the local wave steepness exceeds this
amount. If now we assume, first, that the subharmonic instability tends to increase
the amplitude of every alternate wave but leave the wavelengths roughly unchanged,
and secondly that the energy in each wave varies roughly as the square of the local
amplitude, then we should expect subharmonics to lead to breaking only when

ak > 0 .4 3 6 Д /2 = 0.3 0.
1840

22 M. S. Longuet-Higgins and E. D. Cokelet

------------------------------------1____________________ l ____________________ I____________________ I____________________ i

16 2.0 x 2.4 2.8

Figure 16. The mode n = (|)+ when ak = 0.41; enlargement of the wave crest, seen in a frame
of reference at rest. Time interval between profiles is 2 я /б 0 ; the range, 1.37 ^ t < 1.87.

Figube 16. Growth of the mode n = (Jj+whenoi = 0.41, shown by In R(t) ae a function of (.
1 he broken lrne represents the growth fit where fi is the rate corresponding to the normal-
mode analysis.
 1841

The deformation o f steep surface waves. I I 23

t=UT
1 20

t= T

F ig u r e 17. Later stages in the development of the mode (J, f)+

when ak = 0.32 (continuation of figure 1 ).

The example ak = 0.25 shows that this condition is contradicted. The reason
appears to be that every alternate wave is not only increased in amplitude, but also
shortened relative to its neighbours, so that locally a higher value of the wave steep­
ness is attained.
This is confirmed by figure 22 . In figure 22 a is shown the variation with time of
the individual wave amplitude a', defined as half the height of each crest above the
trough in front of it. This is normalized by multiplying by the constant wavenumber
k. The amplitude of each wave fluctuates, and the fluctuations grow in time. Figure
226 shows the individual wavenumber k' corresponding to the horizontal distance
between adjacent troughs. This also fluctuates, by as much as 12 % in either direc­
tion, though these fluctuations are almost in quadrature with the fluctuations in a'.
1842

24 M. S. Longuet-Higgins and E. D. Cokelet

Finally, figure 22 c shows the fluctuations in the individual wave steepness a'k'.
These are slightly greater than the fluctuations in a'k above. The wave finally
breaks when the individual wave steepness is about 0.39.
Hence it may be that waves of even lower amplitude than ak = 0.25 will develop
instabilities (m = 2 ) to the point of breaking, assisted by a shortening of the indi­
vidual wavelength.

F ig u r e1 8 . Enlargement of the wave crest in figure 2 0 , near the instant of overturning, seen
in a reference frame at rest. Time interval between profiles is 2 я / б 0 ; the range,
3 4 .6 6 t 3 5 .1 6 .

The profiles in figures 15, 18 and 21 are so similar as to suggest that the dynamics
of the final stages of overturning are determined mainly by local conditions near
the wave crest. If this is so then we would expect that the local length-scale I and
time-scale r would be related by
l/дт‘ = 0 ( 1),

where g denotes gravity. In fact if we take I to be the *width *of the crest, measured
by the horizontal distance over which the profile is convex, then I « 0.4 while the
time r for the evolution of the crest in figure 15 is above five plot intervals, i.e.
about 0 .6 . Since g = 1 in these units the above relation is indeed satisfied.
However it will be noted that the time-scale for the initial growth-rate of the
instability when ak = 0.41 is much longer than т since /?, though larger than for
the Benjamin-Feir instabilities, is still only equal to 0.065. Thus is about 15.5.
 1843

The deformation o f steep surface waves. I I 25

*=0
,
0 .2

- 0.2

2T
f-----------------------------1

3T

AT

F i g u r e 19. Later development of the unstable mode ($, f ) + when ak = 0.25. The profiles
are seen at times t = \вТ, а = 0, 1, 2, 9 where T is the period of the mode.

F ig u r e 20. Profiles of the mode (ф, f)+ near breaking, when ak = 0.25.
The time interval between profiles is 2 л /4 ; t = 140.86 and 142.42.
1844

26 M. S. Longuet-Higgins and E. D. Cokelet

F ig u r e2 1 . Enlargement of the overturning crests in the mode (4 , $)+ when ak = 0 .2 5 , seen

in a reference-frame at rest. Time interval between profiles is 2 n / 5 0 ; the range, 1 4 1 .9 2
< t< 1 4 2 .4 2 .

T a b l e 2 . O v e r a l l g r o w t h -r a t e o f t h e
i n s t a b i l i t y (£, f ) + w h e n ak = 0.25

t C(ct, t) P' = In R
= 1 4 .6 1 -0 .9 9 2 0 .0 1 3 5
T 0 .9 7 9 0 .0 1 3 4
Iт -0 .9 8 2 0 .0 1 3 1
2T 0 .9 6 6 0 .0 1 3 4
fT -0 .9 6 3 0 .0 1 3 3
ВТ 0 .9 4 6 0 .0 1 3 3
IT -0 .9 1 6 0 .0 1 3 6
4T 0 .8 6 3 0 .0 1 3 6
!т -0 .7 6 6 0 .0 1 4 1

8. C o n c l u s i o n s
By applying the independent time-stepping method of paper I we have confirmed
and extended the conclusions of recent studies (Longuet-Higgins 19786 , c) relating
to the stability of gravity waves. The initial rate of growth of each type of instability
has been verified, and it is confirmed that subharmonic instabilities of Benjamin-
Feir type are indeed confined to a certain range of wave steepnesses. Beyond this
range, the perturbations are neutrally stable, until at about ak = 0.41 a second
type of instability is found which leads directly to wave breaking.
 1845

The deformation of steep surface waves. I I 27

Figure 22. Time-dependence of local properties of the surface: (a) the local wave amplitude о ;
(6) the local wavenumber k'\ (c) the local wave steepness a'k'.
1846

28 M. S. Longuet-Higgins and E. D. Cokelet

In the cases we have tested, the subharmonic instabilities apparently grow to the
point where the wave crests develop local instabilities which again lead to breaking.
The final stages of these breaking waves appear to be very similar in all cases, and
their dynamics are related to local scales of length and time. This gives hope for a
nonlinear theory of the latter stages of wave breaking.
As emphasized in earlier papers, the present calculations assume potential flow
and neglect the possible influences of viscosity and capillarity, which must introduce
qualitatively new effects. Moreover the motion is assumed to be two-dimensional,
and the depth of water to be infinite. For sufficiently large-scale waves,
viscosity and capillarity may indeed be negligible. The possible effects of three-
dimensionality and of finite depth are, however, very interesting in the appropriate
circumstances, and require further study.

Throughout this investigation we received support from the Departments o f

Energy and Industry, through a contract with the Institute of Oceanographic
Sciences.

R eferences

Benjamin, Т. B. 1967 Instability of periodic wavetrains in nonlinear dispersive systems. Proc.

R. Soc. Lond. A 299, 69-67.
Cokelet, E. D. 1977 Breaking waves. Nature, Lond. 267, 769-774.
Lighthill, M. J. 1967 Some special cases treated by the Whitham theory. Proc. R. Soc. Lond.
A 299, 28-63.
Longuet-Higgins, M. S. 1978 a Some new relations between Stokes’s coefficients in the theory
of gravity waves. J. Inst. Math. Appl. (in the press).
Longuet-Higgins, M. S. 19 78 b The instabilities of gravity waves on deep water. I. Super­
harmonics. Proc. R. Soc. Lond. A 360, 471-488.
Longuet-Higgins, M. S. 1978 c The instabilities of gravity waves on deep water. II. Subhar­
monics. Proc. R. Soc. Lond. A 360, 489-506.
Longuet-Higgins, M. S. & Cokelet, E. D. 1976 The deformation of steep surface waves on
water. I. A numerical method of computation. Proc. R. Soc. Lond. A 350, 1-26.
 1847

Proc. R. Soc. Lond. A 371, 441-451 (1980)

Printed in Great Britain

A technique for time-dependent free-surface flows

B y M . S. L o n g u e t - H i g g i n s , F.R.S.

Department of Applied Mathematics and Theoretical Physics,

Silver Street, Cambridge, and Institute of Oceanographic Sciences,
Wormley, Surrey

(Received 4 October 1979)

In this paper we present a reformulation of the problem of time-dependent

irrotational free-surface flows under gravity. Compact expressions are
derived for the pressure p and its rate of change Dp/DJ following a particle.
Some exact but special solutions are discussed, and a method of approach
for the general case is proposed.

1. In tro d u ctio n

Over 130 years after the formulation by Stokes (1845) of the general equations for
inviscid irrotational flow, it is remarkable that few if any exact solutions to the
problem of time-dependent free-surface flows under gravity are known. Most known
solutions refer to flows that are steady, or, like progressive waves, are easily re­
ducible to steady flows by an appropriate choice of the frame of reference. Among
the time-dependent gravitational flows that have been considered are those for
shallow-water waves and standing waves, but these all appear to involve expansions
in powers of some small parameter, such as surface slope, and hence must be con­
sidered as approximations at best.
Other time-dependent solutions, such as the Dirichlet parabola (John 1953),
ellipse (Taylor i960) and hyperbola (Longuet-Higgins 1972) are exact but do not
contain gravity in an essential way.
An interesting method for obtaining solutions to time-dependent free-surface
flows was indeed suggested by John (1953), who in this way derived some special
flows (see also Longuet-Higgins 1976). However, the method cannot be general,
for it assumes a Lagrangian coordinate a) which is an analytic function of the velocity
potential. This would not include, for instance, a progressive irrotational wave
with plane bottom, in which the mass-transport velocity induces a strong vertical
gradient of the mean displacement.
The present situation is highlighted by the absence of any satisfactory analytic
solution to the problem of an overturning wave, although the possibility of accurate
numerical calculation has been demonstrated by Longuet-Higgins & Cokelet (1976,
1978 ).
In our quest for a general method, we may be guided by the following considera-

16 [ 441 1 Vol. 371 . A U August 1980)

442 M. S. Longuet-Higgins
tions. For free-surface flows there are two boundary conditions. One of these, the
kinematic condition, states that a particle in the free surface must move with the
surface. This is most easily expressed in terms of Lagrangian coordinates. However,
in the Lagrangian analysis the equations of continuity and of irrotationality for the
interior of the fluid are both highly nonlinear and difficult to handle, whereas in
the Eulerian description both conditions are automatically ensured by the assump­
tion of a velocity potential ф satisfying V 20 = 0, or equivalently the assumption
that
X = ф+ ift ( 1. 1)

is an analytic function of the space coordinate z. This overwhelming advantage of

the Eulerian description suggests strongly that it be incorporated in our attack on
the problem.
The second essential boundary condition for free-surface flows is the condition
that the pressure at the free surface be constant (say zero). Now the pressure, for
a time-dependent irrotational motion, is expressed quite simply in terms of the
complex potential x(M ) b y the Bernoulli equation, so that the equation of the
free surface can be expressed very simply in Eulerian coordinates as p — 0.
Returning to the first condition, it is important to realize that the free surface,
in unsteady flow, is not a streamline. (Indeed the trajectories of particles throughout
the fluid are not streamlines, but only tangent to the streamlines instantaneously.)
We may however express the kinematic boundary condition by saying that for a
particle in the free surface ТУр/Dt = 0, that is to say the rate of change of the
pressure following a particle must vanish. This also can be expressed in Eulerian
coordinates. However until the present investigation (described in §2) it was
possibly not realized how compact this boundary condition becomes when ex­
pressed in terms of the complex coordinate z.
Thus the problem, as formulated in the present paper, reduces to ensuring that
the two surfaces p = 0 and Dp/Dt = 0 are coincident.
As we shall see later, this formulation, together with the fact that calculations
in complex coordinates are very easily performed on a modem computer, places
in our hands a powerful tool for investigating time-dependent free-surface flows.

2. Z AND t AS IN DE P E N D E N T V ARIABLES

Throughout this paper we assume the flow to be two-dimensional and work in

terms of the complex variables

z — x + iy, z* = x - i y , ( 2 . 1)

where x and у are rectangular coordinates, x being directed vertically downwards.

The flow being irrotational and incompressible there exists a complex velocity
potential
Х = ф+ ’и/г (2.2)
 1849

A technique for time-dependent free-surface flows 443

which is an analytic function of z and a smooth function of the time t, such that the
particle velocity и + iv = W* is given by

w = X , W * = (*)• = (**),.. (2.3)

Here a suffix denotes partial differentiation, and an asterisk denotes the complex
conjugate. The pressure p is then given by the Bernoulli equation in the form

p = Re (gz-Xt - hXtX* + /), (2.4)

where g denotes gravity and/is a function of t only.

To find an expression for Dp/Dt, where D /D t denotes differentiation following a
particle, we have first

|2S>
Now if F(z,z*) is any differentiable function of both z and z* we have

Fx = [Fz + Ft.)t Fv = i № (2-6)

Also
« = i t e + X**). « = (2-7)
Therefore in general
B F /D t = Ft + (x *F g+x,F,*)- (2-8)
Writing
F = -2p = (x t + Xt*) + X.Xt* - g(z + z*) - 2/. (2-9)
we find
- 2Dp/Di = (Xtt + Xtt*) + (XztX.*+X't*Xz) ~ 2ft

+x,*(xzt+xa x * - g )
+ Х'<.Хг* + Х * * Х .-9 ) (2-10)
and so
D p/D t = R e [jff» -C ¥ K+ 2 ^ *xrt+ W * x J + f t), (2.11)
whprp
W* = Xz* = T>zfDt. ( 2. 12)

The remarkably compact expression (2 . 11) seems not to have been recognized
previously.
For steady flows or for motions such as progressive waves when referred to the
appropriate coordinates, equation (2 .11) reduces to

Dp/D t = R e \ W * ( g - W * x a)l (2.13)

The first term on the right represents the pressure change due to vertical motion
in the hydrostatic pressure field. The second term, involving Xa* represents the
effect of the curvature of the streamlines.

16-2
1850

444 M. S. Longuet-Higgins

3. X AND * AS I N D E P E N D E N T v a r ia b l e s

For some problems (for example in waves of finite amplitude) it may be more
convenient to take x and t as independent variables, and to express the space co­
ordinate z in terms of them. For steady waves (no time dependence) this method
was suggested by Stokes ( 1880 ).
To find the corresponding expressions in the new variables, we have for example,
by the method of differentials,
dz = zx d^ + ztt df,
(3.1)
= X.dz + Xtcfc.J
On eliminating from these equations and then equating coefficients of the inde-
pendent increments dz and dt we obtain

■ W b -M (3.2)
z x Xt + 2t = 0.J
Hence
Xz= l / z x = W (3.3)
and
* = - Z t/ Z x = - Wzt. (3.4)

The expression (2.4) for the pressure then becomes

p = Re(gz+ W*zt - i W W * + f ) . (3.5)

To obtain the corresponding expression for Dp/TH, we note that if F(z, t) is any
differentiable function which on substitution z — z(x) becomes equal to G(x, 0
then, by the same argument,
F. = Ox/z x (3.6)
and
I ',= Ot+ X tGx, (3.7)

so on substituting for Xt from (3.4) we have

Ft = Gt - G xzt/z x. (3.8)

By repeated application of the results (3.6) and (3.8) we obtain

xa = - W \ x,

x w --T P V * V «r (39)
* „ = - W z tt- 2 W % z xt~W>z\zxx..

Finally on substitution in (2.11) we find, after some rearrangement,

Ър/TH = Re + H 4 XX) + f t], (3 . 10)

H = W ( W * - z t). (3. 11)

 1851

A technique for time-dependent free-surface flaws 445

Since from (2 .8 ) the rate of change of x following a particle is

I>x/T>t = Xt + Xz*Xz = - Wzt + W W * (3.12)
(from (3.3) and (3.4)) it follows also that

H = J>x/Vt, (3.13)
showing the analogy between (2 . 11) and (3.10). However, since W ( = Xz — I /2*) i®
linear in H is clearly not linear in z, showing that (2 . 11) is generally the simpler
expression of the two.
In steady flows equation (3.10) reduces to

T>p/T>t = R e [ W ( g + W * W * \ x)\. (3.14)

The term due to curvature of the streamlines can also be written as

R e(W<fzxx), (3.15)
where q denotes the particle speed. In progressive gravity waves this term must
exactly balance the hydrostatic term at the free surface itself.

4. X A N D 2 B O T H F U N C T I O N S OF 0)
The analysis of the two previous sections may be included in the general case
when x and z are each functions of a third complex variable w, and of the time
that is
X = X(a),t), z = Z(w,*). (4.1)
Thus §§2 and 3 correspond to the cases Z = ш and X = ш respectively. The form
(4.1) also includes as a special case the formalism of John (1953) but is more general,
since John assumed that to was a constant following a particle, and moreover that
(o was real at the free surface. This restricted the solution of flows of a special class.
When we adopt the assumption (4.1), the equation of the free surface is given by a
more general (but real) function of <0and (o*.
Expressions forp and D p/D t can be derived by an argument similar to that of
§ 3. Thus we easily find
X. = X J Z „= W (< d ,t) (4.2)
say, and
Xt = Xt - W Z t (4.3)
leading to
- 2 p = iW W * + (Xt - W Z t) - g Z - f + c . c „ (4.4)

where c.c. stands for complex conjugate. The right hand side of equation (4.4) is
a real symmetric function of w, ы' and t.
Now if jP(z,z*, t) is any real symmetric function of z and z* which by the sub­
stitution z = Z((o, t), z* = Z*(o)*, t) becomes 0(o), w*, t) it can easily be shown that

F. = GJZ„, Ft. = (F,)’ = O J 4 Z * (4.6)

1852

446 M. S. Longuet-Higgins
and
Ft = Qt - Q „ Z t/Z ',—Qu*Z t* /Z j* . (4.6)
Therefore from (2.8) we have
DG/Di = D F /O t = F ,+ W *F ,+ W F *
= a t + ( W * - Z t) Q J Z „ + ( W - Z t* ) G * I Z * . (4.7)
Applying this result to the function —2p of (4.4) we obtain finally

- 2Dp/Dt = (Xtt- WzH) + 2K{XU - WZM)

+ K t(Xai- W Z J ) - g W - f t, (4.8)

W = X J Z M= Dz/D< (4.9)
and
К = ( W * - Z ,) /Z U= Dw/Di. (4.10)
It can be seen that when Z з to we have
Sw= l , Z t = 0 ; Zww= Zwt = ^tt = 0 , (4.11)
so
= (4-12)

and equation (4.8) reduces to (2 . 11). On the other hand when I = w w e find
W = 1/Zmi К = W ( W * - Z t) = Я , (4.13)

and so (4.8) reduces to (3.10).

In his formulation of the free-surface problem John ( 1953) defines a Lagrangian
variable w. This is equivalent to assuming initially that
Dw/D t = 0, (4.14)
and so
K = 0, W* = Zt. (4.16)

But we shall see that the description of many flows is in the end simpler if we do
not make this assumption.
In order that the solution shall represent a permissible flow having a velocity
field that is both finite and single-valued everywhere within the fluid, we must have
ZJX„ = 1/Ж Ф 0 (4.16)
everywhere within the fluid. This implies, for example, that any zero of Zu within
the relevant domain must coincide with a corresponding zero of Хы.

6. S o m e exact s o l u t io n s

We mention now some known exact (but non-trivial) solutions to the equations
of §4.
First is the Dirichlet parabola (John 1953 ; Longuet-Higgins 1976) in which
Z^w-iR/fi, X=№/t, (6.1)
 1853

A technique for time-dependent free-surface flows 447

R being an arbitrary real constant. For, from equations (4.9) and (4.10) we have

W = ш/t, K = (o*/t + R/t* (6 .2 )

and so from (4.4), if g Ф 0,

p = J(« -< tf* )* /| * + iR{a) + (o*)/t*+f, (6.3)

while from (4.8) we find

Dp/Dt = - 4 p / t (5.4)

provided / = \R2/t*. From (6.3) and (5.4) it is clear that both p and Dp/T)t vanish
on the free surface
(yt*)2 = R(xP + -±*R) (5.5)

which represents a parabola, expanding or contracting like t~z.

Other self-similar flows, in which Z and X are both polynomials in a), have been
derived by Longuet-Higgins ( 1976).
Closely related to the flow given by equations (5.1) is the Dirichlet ellipse (Lamb
1932 ; Taylor i 960 ) corresponding to

Z = io, X = i Aar*. (5.6)

Here the function A (t) is given inversely in terms of t by

(= (6.7)

and we have also/ oc A*, g = 0 .

Similarly we have the Dirichlet hyperbola in which

t= А-*(1 + А*)-*АА. (5.8)

This was discussed by Longuet-Higgins ( 1972, 1976 ). Both (5.5) and (5.6) have
surfaces which are conics, with axes in fixed directions in space. However, some
asymmetric and more general (irrotational) flows in which the free surface rotates
about 0 with a non-zero angular velocity will be described in another paper (Longuet-
Higgins 1980 a).
The above-mentioned solutions are all gravity-free, and so their application is
restricted to situations where the fluid is in free fall, or where the motion develops
so rapidly that gravity can be neglected. Of great interest therefore is the Stokes
comer flow, which in our formulation corresponds to

Z = oj\ X=lig*aj*. (5.9)

The free surface is a pair of lines (or planes) inclined at angles + to the vertical.
To see that this is a solution, we have from equations (4.9) and (4.10)

W = ig*o), К » - 1\дЦа)*/(о), (5.10)

1854

448 M. S. Longuet-Higgins
and so from (4.4)
p = g((i)2 —ww*+ a)*2). (6.11)

On the other hand from equation (4.8)

Dp _ 1 1 6>3 —2a)2(i)* + 2оно*2 —w** j 2\

Dt ~ 4i^ aw*

Comparing equations (6 .6 ) and (5.7) we see that

I* (5.13)
Dt 4i o)(o*

from which it follows that Dp/DJ vanishes everywhere on the surface p = 0 , as

required. Equation (5.11) shows that the free surface is in fact a degenerate conic
consisting of the two straight lines
argw= ± J tt (5.14)

through the origin. Since' z = ш2, this implies argz = ± Jtt.

The above method of solution is at first sight more complicated than Stokes’s
original method. However, as we shall see later (Longuet-Higgins 1979 b), by
merely adding to X a quadratic function of the variable w, a simple but accurate
solution may be obtained to the hitherto intractable problem of the flow in an
overturning wave.
Meanwhile, in the next section of this paper, it will be convenient to derive some
compact formulae relating to certain singularities in the pressure field. Not only
will these be useful later, but they also illustrate the unexpected simplicity of the
present analysis.

6. S i n g u l a r it ie s in the pressure f ie l d

The free surface, being a surface of constant pressure, can have a sharp comer
only when the pressure gradient vanishes. Consider then the condition that the
pressure gradient vanish while the pressure p remains a differentiable function of
x and y, or equivalently of z and z*.
From equation (2.9) and the general formulae (2 .6 ) for differentiation of a smooth
function we have
-2z>x = (xa X * + Хгг*Х,) + (Xzt + Xrt*)-2 ?. ! (6

At a stationary point, where px = p y = 0 , we then have

Re(X=X.* + * « t-0 ) = O, I
(6.2)
 1855

A technique for time-dependent free-surface flows 449

In other words the condition for a stationary value of the pressure p is simply

XzzXz* + Xzt = V' (6-3)

For a frame of reference in free fall, that is to say if the origin is accelerated down­
wards with the acceleration of gravity, we may set g = 0 in equation (6.3) so that
the condition for a stationary value of the pressure reduces to

XzzXz* + Xzt = 0- (6-4)

Consider now the form of the contours p = constant in the neighbourhood of a
stationary point. Assuming the second derivatives of p to exist we have, to second
order,
d 2? = hiPxx^+ZPxvdxdy+Pvvdy2)- (6 -5 )

We shall be concerned mainly with saddle points that is to say points at which two
different contours p = constant intersect one another. In that case the angle у
between the two contours is given by

cos* у = -— -. (6 .6 )
(P z x -P ,v ) + *Р ху2
Now from (2 .6 ) we have in general

P xx Ргг "1" 2P a *

P xy = i ( P * z - P z ' A (6.7)
P vv = ~ P z z + 2Pzz* - Pz*z»>.
so that
P xx+P yv = *Pzz*>
( 6. 8 )
P x x ~ P w ~ ^(Pzz~b~Pz*t*) . }

and hence (6 .6 ) becomes simply

cos2у = - , (6.9)
PzzPt
or since рг*г. = pu* this is equivalent to

cosy = A (6. 10)

Pxz

provided we choose у so that —Jtc ^ у < Jtc.

Now from equation (2 .9 ) we see that in terms of the velocity potential %■

~ 2Pzz -= д и + д и я Л 1 ( 6 1 1 )

- 2p = XzzXzz*' '

Therefore on substitution in (6 . 10) we obtain

“ ' - B S f e r (6,12)
1856

450 M. S. Longuet-Higgins
The expressions for the mean curvature and the Gaussian curvature of the
surface p = p{xyy) are also very simple. In fact from (6 .8 ) we have that

м = p „ + p yv = (6.13)

showing that the mean curvature of p is in general negative. Exceptionally where the
velocity gradient vanishes, that is when

X«* = i(“ * + i»*) = 0 . <614)

then the mean curvature will also vanish
The Gaussian curvature, from (6 .8 ) is given by

я = P XIP VV~ P „ ? = ^ P ^ - P t z P j ) (6 1 6 >

and from (6 . 11) this is
a = <6 16 >
which may be either positive or negative. Thus the pressure field can have both
saddle points {Q < 0) or maxima (Q > 0, M < 0 ). On the other hand minima
(£> > 0 ,M > 0) are ruled out by (6.13).
Lastly, the characteristic condition for a cusp in the free surface is that Q shall
vanish, that is to say
= |XmX.* + XEtl- <617>

7. D i s c u s s i o n
We have seen that the equations for time-dependent irrotational flow of a perfect
fluid with a free surface become remarkably compact when formulated in terms
of the complex variables z - x + iy and z* — x —iy, instead of x and у respectively.
This is because the velocity potential ^ is a function of z and t only, independent
of 2*, and because the pressure p is a real function of z, z* and t which is symmetric
in z and 2*. The kinematic boundary condition T>p/Dt = 0 is also much simpler
than expected.
This simplicity is maintained when z and x are ©ach expressed in terms of a third
complex variable w, which need not be a Lagrangian coordinates, as in the formu­
lation of the problem by F. John.
In all the exact solutions known at present it appears that p and T>p/T>t are
related by an equation of the form

D p/D t = 4>(cj,(o*,t)p, (7.1)

where Ф is a real function, symmetric in (o and o>*. In the Dirichlet conics, which
are gravity-free, Ф is independent of ш and w*, and is a function of t only. In the
Stokes corner flow, on the other hand, Ф is a function of (o and ш* only, and inde­
pendent of t, while in the self-similar solutions described by Longuet-Higgins ( 1976 )
X, 2 and Ф all involve functions of a similarity variable a>tx, where Лis a constant.
 1857

A technique for time-dependent free-surface flows 451

However, equation (7.1) is not the most general relation which ensures the vanish­
ing of Djj/D* when p = 0 . For example, we may also have
Цр/D* = Ф1р-\-Ф2рг + . . . . (7.2)
Relations of this kind will be explored in the papers to follow.

R eferences

John, F. 1953 Two-dimensional potential flows with a free boundary. Commune pure appl.
Math. 6 , 497-503.
Lamb, H. 1932 Hydrodynamics, 6th edn. Cambridge University Press.
Longuet-Higgins, M. S. 1972 A class of exact, time-dependent, free-surface flows. J. Fluid
Mech. 55, 529-543.
Longuet-Higgins, M. S. 1976 Self-similar, time-dependent flows with a free surface. J. Fluid
Mech. 73, 603-620.
Longuet-Higgins, M. S. 1980 a On the forming of sharp comers at a free surface. Proc. R. Soc.
Lond. A 371, 453-478.
Longuet-Higgins, M. S. 1980 b A branch-point model of a breaking wave (in preparation).
Longuet-Higgins, M. S. & Cokelet, E. D. 1976 The deformation of steep surface waves on
water. I. A numerical method of computation. Proc. R. Soc. Lond. A 350, 1-26.
Longuet-Higgins, M. S. & Cokelet, E. D. 1978 The deformation of steep surface waves on
water. II. Growth of normal-mode instabilities. Proc. R. Soc. Lond. A 364, 1-28.
Stokes, G. G. 1845 On the theories of the internal friction of fluids in motion, and of the equili­
brium and motion of elastic solids. Trans. Camb. phil. Soc. 8 , 287-341 (reprinted in
Mathematical and physical papers, vol. 1, pp. 75-129. Cambridge University Press, 1880).
Stokes, Sir G . G. 1880 Supplement to a paper on the theory of oscillatory waves. Mathematical
and physical papers, vol. 1, pp. 314-326. Cambridge University Press.
Taylor, G. T. i 960 Formation of thin flat sheets of water. Proc. R. Soc. Lond. A 259, 1-17.
1858

Proc. R. Soc. Lond. A 371, 453-478 (1980)

Printed in Great Britain

On the forming of sharp corners at a free surface

By M. S. L onguet -H ig g in s , F.R.S.
Department of Applied Mathematics and Theoretical Physics,
Silver Street, Cambridge and Institute of Oceanographic Sciences,
Wormley, Surrey

(.Received 4 October 1979)

By applying the technique for time-dependent irrotational flows proposed

in the preceding paper, a new class of exact free-surface flows is derived.
In these, the free surface has the form of a variable hyperbola, whose axes
rotate in space. The angle у between the asymptotes, and the angle 8 of
orientation of the axes, are found explicitly as functions of the time.
The solutions fall into three groups. First there are those in which у
diminishes smoothly from 90° (a rectangular hyperbola) to zero (a slender
hyperbola) while the angle of orientation 8 increases towards a finite limit.
Secondly there are solutions in which у diminishes to a positive minimum,
and then returns again to 90°. Thirdly у may begin from small values,
increase to less than 45° and return again towards zero. In each case the
total angle 8m&x remains finite.
An exceptional but very interesting solution, in which the vertex angle
у remains constant at 45° and the free surface rotates with uniform
angular velocity 8, is described in terms of elementary functions of the
time t.
It is suggested that these flows, which are generalizations of the sym­
metrical Dirichlet hyperbolae, are relevant to the flow near the tip of a
breaking gravity wave. Since for large values of t the angle 8 tends to its
limit like t~l, the flows may be matched asymptotically to the parabolic
arch of a plunging breaker. In other cases, the tip of the wave can curl over
and appear to form a vortex. By the inclusion of terms cubic in the space
coordinates it is also possible to represent a sharp crest pointing upwards
and tending towards a cusp.

1. I n t r o d u c t io n

This paper is a sequel to the preceding paper paper I (Longuet-Higgins 1980 a;

to be referred to as I) in which a fresh approach was suggested to the problem of
finding an analytical (as opposed to a numerical) description of time-dependent
free-surface flows.
The hitherto unsolved problem which prompted this investigation is how to
understand and analyse the flow in a breaking water wave. At the outset it seems
essential to distinguish two different aspects of the problem. One is to find the local
form of the sharp-pointed crests observed just before the free surface erupts into
spray. This is the special subject of the present paper. The other aspect is the
[ 463 ]
 1859

454 M. S. Longuet-Higgins
overturning of the whole mass of fluid such as occurs in a plunging breaker. This
will be the subject of a subsequent paper (Longuet-Higgins 19806).
That these two aspects are in fact distinct is shown, first, by the observation that
upwards-pointing cusps sometimes occur, and secondly by the demonstration both
numerically (Longuet-Higgins & Cokelet 1976) and by laboratory experiment
(Longuet-Higgins 1977) that in plunging breakers it is possible for the upper part
of a breaking wave to overturn entirely before a sharp corner is actually formed.
Among previous authors who have attempted an analysis of the local flow, Biesel
(1952) gave an approximation which is valid only for small surface slopes, and
which cannot in fact be carried to the limit when a cusp is formed. Beyond this
limit, his solution is topographically impossible. The ‘ spiral vortices *of Tulin (1964)
again are multi-valued flows which cannot be incorporated into a rational
solution without the invocation of ‘ turbulence’. Such solutions are perhaps
more appropriate to the state of the fluid after a wave has broken, rather than
before.
N ot long ago some attention was attracted by the suggestion of Zeeman (1969)
that the formation of a sharp cusp corresponds topologically to one of the seven
‘ elementary catastrophies’ enumerated by Thom (1969). However, Thom pre­
supposes that the free surface is defined by the condition that a certain ‘ potential
function’ shall be stationary on the surface. Despite unpublished attempts by
Professor Zeeman and by the present author, no appropriate ‘ potential’ for the
gravity-wave problem has yet been discovered. A t the present time, therefore,
Zeeman’s suggestion lacks any hydrodynamical content. (In a later discussion
Zeeman (1975) has expressed reservations about the validity of this model.)
The method of the present paper is relatively simple. As in paper I we seek a
surface on which the pressure p and its rate of change T>p/T>t following a particle
will both vanish together. In this way both the dynamical and the kinematic
conditions on the free surface will be satisfied. In the neighbourhood of some point
where the free surface has a sharp (though finite) curvature, we can expect there
to be a saddle point in the pressure field, not generally on but outside the fluid.
Such a point has downwards acceleration g, so that to an observer attached to this
point the fluid appears to be in free fall, and the equations are greatly simplified.
In fact this procedure effects for free-surface flows a simplification quite similar
to that achieved for progressive waves by taking axes moving with the phase speed
(see § 2).
In § 3 we therefore seek solutions appropriate to the neighbourhood of a saddle
point of p. We obtain a family of solutions which includes the Dirichlet hyperbolae
(Longuet-Higgins 1972) but is more general. For, whereas the solutions found
earlier had surfaces in the form of conics with principal axes fixed in direction, the
new solutions allow the principal axes to rotate in space; this despite the fact that
the flow itself is irrotational (see § 4 ).
The time history of these flows is described in § 5 and it is shown that the vertex
angle (i.e. the angle between the asymptotes) of the hyperbola can tend to zero
1860

On the form ing of sharp comers at a free surface 455

as the time t increases. Moreover, although zero angle is not reached in finite time,
nevertheless arbitrarily small angles may be attained.
Some very interesting special cases are described in §§6 and 7. In one of these
the fluid has the form of a simple rotating slab, whose thickness tends to zero as
t -+ oo. In the other, the vertex angle of the hyperbola tends exponentially to the
value 45°, while the axes of the hyperbola rotate at a constant (but non-zero)
angular velocity.
The formation of a sharp cusp requires not only the vanishing of the vertex angle
in the surface as described by the quadratic terms, but also the presence of some
terms cubic in the space coordinates. In § 10 we show that the exact solutions of
the previous sections may be modified by the addition of cubic and higher-order
terms, which are determined ultimately by the conditions for matching to the outer
flow (this involves the gravitational acceleration <7). Figures 11 and 12 show that
some of the resulting forms may be quite realistic. The argument also suggests
that upwards-pointing cusps are more easily formed than those that point down­
wards.

2. F rame of references

We shall assume that the velocity field is two-dimensional, irrotational and

divergence-free, and hence may be represented by a complex potential x which is
an analytic function of the coordinate z — x + '\y. We take the real axis of z (i.e.
the x axis) to be directly vertically downwards.
Throughout the interior x(z) is regular, and we shall assume that generally x
may be extended analytically a short distance beyond the free surface. Cases such
as the Stokes corner flow, where there is a branch point actually on the free surface,
will be considered in a later paper (Longuet-Higgins 19806 ). In other words in the
present paper we consider only regular singularities either on or just beyond the
free surface.
We are particularly interested, however, in situations where the free surface
develops a sharp curvature, as in figure 1. We may then expect the existence either
on or just outside the fluid, of a sharp angle in the contours of constant pressure.
The simplest such case is when the pressure 2>(z,z*) has a saddle point 0 , say, at
which two contours p = p0 intersect. The free surface itself (shown hatched in
figure 1) will be given by different contour p = 0 . Since it has been shown in paper I
that pressure minima cannot occur in the interior, but only maxima or saddle
points, it follows also that p0 < 0 in general.
At the point О itself the pressure gradient vanishes. Therefore the particle
acceleration at this point must be equal to g, directed downwards, in a stationary
frame of reference. We shall now make the special assumption that the saddle point
О moves with thefluid and shall take a new frame of reference with the origin attached
to 0. In this new frame, since О accelerates downwards with acceleration g , the
fluid motion will appear to be gravity-free; all terms involving g must vanish. This
greatly simplifies the analysis locally. A compensating disadvantage is that if in
 1861

456 M. S. Longuet-Higgins

F io u r e i . Sketch of pressure contours p = constant in the neighbourhood of a saddle point

O. The free surface, shown hatched, corresponds to p = 0. The frame of reference is
attached to the moving point O.

the original frame the flow at large distances is steady or horizontal, then in the
new reference frame the velocity potential must contain a term - gtz which must
be matched to the local flow in some intermediate zone.
In this paper we shall be concerned almost exclusively with the local flow. Since
the flow is regular at the origin О it follows that x т а У be expanded in a power
series:
X = Lz + iAz* + iBz* + ... (2 . 1)
in which the term independent of z, which has no effect on the velocity, is taken to
be zero. The remaining coefficients are complex functions of the time t.
Since the origin moves with the fluid, Xt must vanish at 0. Hence
L = 0 (2 .2 )
and we may verify that the condition for a stationary value of p in a free-fall
reference frame, namely
+ = 0 <2-3)
(see paper I, equation (6.4)) is indeed satisfied.
General expressions for the pressure p and its rate of change q following a particle
have been shown in Paper I to be

- 2p = X.X.* + (Xt + Xt*) - 1 (2.4)

- 2 ? = {Хс,Х*г + 2XztX* + * „) + c-c- - 2*'>J
 1862

On the forming of sharp comers at a free surface 457

where F is some real function of the time t and a dot denotes its rate of change
(and c.c. is the complex conjugate). The formal problem is to find solutions x(z>0
such that p and q vanish together on the same (moving) surface.
In the first part of this paper we shall study solutions in which A is the only
non-vanishing coefficient in equation (2 . 1). Later we shall investigate the possibility
that В also does not vanish.

3. Q u a d ra tic s o lu tio n s
Suppose first, then, that
(3-1)
the remaining terms in equation ( 2 . 1) being zero. On substituting for x *n equations
(2.4) we obtain
- 2p = \{Az2 + 2 AA*zz* + A*z* 2) - 2 F, (3.2)
and — 2q = £[(Л + 2A2A *)z2 + ±A A * zz* + c.c.] —2fi. (3.3)
A sufficient condition for equations (3.2) and (3.3) both to vanish on the same
surface is that the coefficients of corresponding terms are proportional, that is
A + 2 A 2A * 2 {AA* + AA* ) A* + 2A A *2 /
(3.4)
AA* A* F’
We shall now obtain solutions to these equations which are a generalization of
those obtained in Longuet-Higgins ( 1972 ), where it was assumed that A was real.
Now let us write .
A = a e l'r, (3.5)
where a and cr are both real. Then
A = (a + iatr) e|/r (3.6)
and A = [(a —i<r2a)-H ( 2dd--|-a0-)]eUr. (3.7)
So on substituting in equations (3.4) we obtain
(a —ol&2+ 2as) + i(2a<7 + acr) 4a a F .
. ' ■ . — a — C.C. — t j , (3. о J
a + ia<r a2 F
Because each ratio equals its conjugate complex expression we find, by adding or
subtracting the corresponding numerators and denominators,
a -a & 2+ 2a3 2(а& + а&) 4a
(3.9)
a eta cl F’
From the second and third expressions we get

1 + 1 - 4 (3-10)
oc, cr a

! - « * (3. 11)

\
 1863

458 M. S. Longuet-Higgins
and on integration with respect to t,
& as Лее2, (3.12)
where Л is an arbitrary real constant.
In the first expression in (3.9) we may now substitute A2a 4 for <72, and then from
the first and third terms we get
ad —4a2 + 2a4 —A2ae = 0. (3.13)
This is a generalization of equation (3.6) of Longuet-Higgins ( 1972), to which (3.13)
reduces when A = 0 . Now multiplying both sides of (3.13) by the integrating factor
2d/а 9 as before we obtain, on integration,

- 8, Aa 4A a 2- r .
a <з “ >
where P is a constant. Hence
d2 = a 4 - A 2ae+ Pa8, (3.16)
a generalization of equation (3.8) of Longuet-Higgins ( 1972).
We shall see that the essential feature of this generalization is that it allows the
existence of a class of flows that is asymmetric, whereas all the flows treated pre­
viously were symmetric about the axes of x and y.
The general solution of equation (3.15) above is simply

t= f------ ------------- J, (3.16)

Ja2(l —AJa*+Pa4)i '
which can be expressed, as in the special case A = 0 , by means of elliptic integrals.
For both calculation and discussion, however, it is just as convenient to have
equation (3.16) in integral form.
Having obtained a implicitly as a function of the time t, we have from (3.12) that

Aa»d<, (3.17)

or what is the same thing

r ------- Ado------ (3.18)
J(1-Лга*+ Ра*)*
Lastly we have from the third and fourth terms of (3.9)
F = -iS < x \ (3.19)
where S is an arbitrary constant (cf. Longuet-Higgins (1972), equation (3.11) in
which Q is written for —£). Since we wish the pressure at z = 0 to be negative or
at least not positive, we take
Я > 0. (3.20)
1864

On the forming of sharp comers at a free surface 459

4. Form o f th e f r e e su rfa ce
The equation of the free surface is p = 0 , or from equation (3.2)
A z* + 2 A A * zz * + A * z *2 = 4F. (4.1)
Introducing polar coordinates
z = re* (4.2)
and substituting the expressions (3.5), (3.6), (3.12) and (3.19) we have
(d + iAa8) r2 e1*2* ^ + c.c. + 2aV 2 = - 2Sa*. (4.3)
Now writing a + iAa3 = —/?eIe, (4.4)
where cose = —a/fi, sine = —Ла3/Д, (4 б )
and ft1 = а 2 + Л2ав = а4(1 + Р а 4), (4.6)
by (3.15), we see that (4.3) becomes
pr2el(2fl‘Hr+e) —aV 2+ c.c. = 25a4. (4.7)
Referred to cartesian coordinates
x' = rcos [в+\{<т + e)],|
(4.8)
y' = rsin [0 + i(cr + e)],J

inclined at an angle S = i(o4*e) (4 9)

to the axis of x> equation (4.7) becomes
(у? - а2) ж'2- (p + a2) y ’2 = Sa\ (4.10)
which represents a hyperbola or an ellipse according as
(/? - a 2)0? + <x2) £ 0, (4.11)
that is p2 ^ & A (4 . 12)
or 0 (4.13)
by equation (4.6).
In this paper we shall deal mainly with the hyperbolic case, that is to say when
P > 0 , and (4.10) may be written in the standard form

x'*/a2- y ’2/b* = 1, (4.14)

the semi-axes a and b being given by
Sot4 S ol4 iA
а4 = д т т - 6 (416)
(we recall that 8 > 0 in general). In the special case P > 0 , S = 0 the surface de­
generates to a pair of intersecting straight lines. When P = 0 but S > 0 the surface
consists of two parallel lines (see § 6 ).
 1865

460 M. S. Longuet-Higgins
We notice also that in the general case

S2a* S2
a‘b* = j ^ = p (4Лб>
a constant. The angle у between the asymptotes is also given by

b2- a 2 a2 1
cosr= - ^ = 7 = (7Т Щ * (4Л7)
It will be seen from (4.16) that in general a2 > b2, so the angle which each asymptote
makes with the axis of x' is less in magnitude than 45°, and у is less than 90°.

F i g u r e 2. Sketch of the hyperbolic form of the free surface as given by equation (4.14).
у denotes the angle between the asymptotes. The axes of x' and y' rotate about the
origin О with (clockwise) rate of rotation 6.

The rate of rotation of the axes can be found from the identity

e = cose^- (sine) —sine— (cose), (4.18)

dr at

On substituting from equations (4.6) and using (4.6) with the differential equation
for a (equation. (3.16)) we find
e = Aaa ? ~ ^ * 1 . (4.19)
1 + Pa*
Together with (3.12) we have then
t i / • -4 (4.20)
* = *(tr+e) = I T P ^
 1866

On the forming o f sharp corners at a free surface 461

5. D isc u s sio n : the gen eral case

From equation (3.15) we see that for a solution to exist we must have
1 —A2a 2 + Pa 4 ^ 0 . (5.1)
For sufficiently small values of a (excluding а = 0 ) it is clear that a solution will
always exist. In fact from equation (3.16) we have, for small values of |a|,

(5.2)
J a2 a

(we shall agree for the present to take the negative square root in (3.16) so that ±
signs can be omitted). So for small a we have
a ~ l/t (a 0) (5.3)
and we see also from equations (4.18) and (4.21) that in this limit the angle between
the asymptotes tends to zero and the rate of rotation of the axes tends to zero also.
At the opposite extreme when |a| -> oo solutions also exist, provided P Ф 0 . In
the limit а -* oo we see from (3.16) that

t-t ~ ---------L - . (5.4)

" J P*a* 3Pia3
Hence aoc\t —f0|*, (5.5)
and as t-> t0 the angle у tends to 90°. At this angle there is a weak ‘ shock’ , as
described in Longuet-Higgins ( 1972).
At intermediate values of a the character of the solution depends upon the nature
of the roots of the quartic equation
1 —A2a 2 4- Pa* = 0 . (5.6)
To fix the ideas, take Л > 0 . (The symmetric flows corresponding to A = 0 have
already been discussed by Longuet-Higgins ( 1972).) If P > JA4 then equation (5.6)
has no real roots, so the sign of ri remains unchanged in the range t0 < t < 00, say
(see figure 3). The solution is essentially similar to the symmetric case described by
Longuet-Higgins ( 1972, §5). For calculations it is convenient to write
P/A 4 = *7, Aa = 1/ / , (5.7)
and then (3.16) becomes
l/A a «J/
M (5.8)
A Jo (w - / 2+ / 4)*'

The parameter A is evidently a time scale for the motion. In (5.7) we have fixed the
lower limit of integration at / = 0 , but any other choice simply changes the origin
of the time t, that is to say it shifts each curve in figure 3 horizontally by a fixed
constant.

\
 1867

462 M. S. Longuet-Higgins

F ig u r e 3. Graph of A a as a function of t/ A when w = P/M > J,

so that equation (6 .6) has no real roots.

If on the other hand 0 < P < £A4, equation (5.6) has four real roots, a = alt ± a2,
g i v e n b y 1 /A * a f = / ! = i + ( i - w ) * . l
(5.9)
1/А*<Х| = / I = * - ( i - T77)*J
Separate solutions exist in the ranges 0 < a < ax and a 2 < a < oo, and not in the
intermediate range ax < a < a 2 (see figure 4). In the first interval, the typical
solution may be taken as ^ - 1/Ae ^
(5.10)
A = Л ( ^ - / ! + / 4)*'
The function a = 1/А / is small at large negative values of t, when the angle у
between the asymptotes is also small. Then a increases to its maximum value ax
at / = 0 and diminishes again towards zero as t -> oo.
1868

On the forming of sharp comers at a free surface 463

F igure 4. Graph of Aa as a function of t /A when ru = P/A* <

so that equation (5.6) has two real roots.

In the second interval (a2 < a < oo) the typical solution is

H fd f (6 . 11)

The solutions start at t = —t' with a and/ -1 large and у approximately 90°. Then
a and у diminish to a minimum value when a = a2, afterwards increasing again to
infinity 8L8t-+t'.
The value of t' is finite, and some values are given in table 1 for certain values of
ш, along with the corresponding minimum angles ymin.
The rate of rotation of the axes, which is given by equation (4.20), is always of
 1869

464 M. S. Longuet-Higgins
constant sign, positive if A > 0 . However, it will be seen that |<j| takes its maximum
value when Pa 4 passes through the value 1, making
д = A / 2P* = 1 /2 mX. (6.12)
This can take place in one or other of the two ranges of a.

T able 1

rotation of the axes, for modes in class 1

® (<*»-<Wdeg TU ( ^ -S ^ /d e g 777 ( * * - * 0)/deg

0.25 oo 0.30 114.5 1.0 62.3
0.26 147.7 0.35 100.2 2.0 50.2
0.27 133.5 0.40 92.0 3.0 44.8
0.28 125.2 0.45 85.9 4.0 41.6
0.29 110.2 0.50 81.7 5.0 39.3

(6) Parameters for modes in classes 2 and 3

class 2 class 3

w 7mi»/deg ffX ( ^ -A J /d e g 7m«/deg (<*00-<Wdeg

0.25 45.00 OO oo 45.00 oo
0.24 50.77 0.595 30.04 39.23 119.9
0.23 53.22 0.470 23.23 36.78 113.1
0.22 55.13 0.397 19.31 34.87 109.2
0.21 56.79 0.345 16.56 33.21 106.5
0.20 58.28 0.304 14.45 31.72 104.4
0.15 64.62 0.176 8.03 25.38 98.0
0.10 70.38 0.099 4.41 19.62 94.4
0.05 76.72 0.044 1.90 13.28 91.9
0.00 90.00 0.000 0.00 0.00 90.0

6. T he s p e c ia l case P = 0
In this and the following section we shall discuss two particularly interesting
special cases which have been excluded from the previous discussion.
The first of these is when P = 0 . Then a2 goes to infinity, = 1/A, and the
solution must lie in the range 0 < a < exclusively. In fact from (3.16) we have

da (1 —A’ a*)*
■ u ■V -A W j*-------- a----- ‘ ( >
Hence a = l / ( < s+A*)t (6 .2 )

and from (3 . 12) <r = J"Aa2d< = arctan (tfA). (6.3)

Combining these last two equations gives altogether

A s « e " = l/(A-it). (6.4)

1870

On the forming of sharp corners at a free surface 465

Since p = (a4+ P a 8)* = a*, the free surface, given by (4.10), consists simply of
the two paraUel planes x- = ± S a (6.5)

(see figure 5) which rotate at a rate

S = Aa2 = &. (6.6)
The instantaneous inclination of the two planes is simply
£ = cr = arctan (t/ A). (6.7)
Thus the fluid consists of a slab, contained between the two rotating parallel
planes given by equation (6.6). As the inclination increases, so the thickness of the
slab diminishes like (f2+ A2)“*. The distance intercepted by the fluid on the ж-axis
remains constant. This solution is a generalization of the symmetric flow studied
experimentally by Longuet-Higgins (1972, §8), to which it reduces when A = 0.

Figure 6. Form of the free surface in the special flow

described by equation (6-4), when m = 0.
 1871

466 M. S. Longuet-Higgins

7. T h e case P - JAa
The second case of special interest is when the two roots ax and a* of § 5 are
coincident. Then we must have ax = a2 = J 2 / A, and from (3.16)
f da
(7.1)
*“ ”Ja2(l-iA2a2
*)
which can be integrated immediately to give
Aa + 7 2
1 - i - J_i (7.2)
А Ла + 2^2 Aa-^/2
Solutions may exist in either of the ranges 0 < a < 2^/A and 2/A < a < oo. The
limiting value a = ^/2/A may be approached either from above or from below.
Interestingly, since
t ----- (ЛД/2)1п|А(ас —a,)|, (7.3)
weseethat |a “ a i| ~ e~Vitlx. (7.4)

F igure 6. A a shown as a function of t f A in the special case when xn = J;

see equation (7.2).

In other words, the free surface approaches its limiting form exponentially in t,
that is much more rapidly than when а -► 0. The limiting angle between the
asymptotes is given (see equation (4.17)) by

c08r = lim n r W * = j 2 (75)

1872

On the forming of sharp comers at a free surface 467

and therefore у -*46°. (7.6)

The rate of rotation of the axes, from equation (4.20), is

8 — JAa8 = 1/Л. (7.7)

In other words the limiting configuration is a hyperbola, with a constant vertex

angle of 45°, rotating with arbitrary, constant but non-zero angular velocity (see
figure 9c).
That this limiting form is itself a possible solution of the original equations may
be verified by substituting the expressions

a = constant, a = a = 0л
I (7.8)
& = constant, cr = 0, J

with equations (3.9). These are clearly satisfied provided that

—a&2-f- 2a3 = 0 (7.9)

or & - ±<j2a. (7.Ю)

Substituting now into the equation of the free surface (4.3), we find this becomes

J 2 ( x ''- y '2) + {x '2+ y^2) = -S o * , (7.11)

which represents a hyperbola with semi-angle \y given by

to n ^ = ( v § ? i ’:)*• (7 12 )

Therefore \y = Jn, у = Jir (7.13)

as expected.
From (4.5) it is clear that in this case
cose = 0, e = —Jrc, (7.14)

and so 8 - i(<r + e) = \& = аД/2 (7.16)

o r, since Aa = ^2, 8 = 1/Л. (7.16)

Altogether we have then
a = ^2/X, ar = j& dt = 2t/X, (7.17)

and so x = \Az% = («V-ty2) (7-18>

8. The a n g le s у and 8

For computing the vertex angle 7 it is convenient to write

А<x = £, (8-1)

and then from (4.17) we have in general

у = arccos (1 +ет£4)-*. (8.2)
 1873

468 M. S. Longuet-Higgins
y/deg

m
y/deg

m
FiaimE 7. The angle у between the asymptotes as a function of the time t
(a) when то > (Л) when m <
1874

On the form ing of sharp comers at a free surface 469

The angle у is shown as a function of t/A in figure 7a, b corresponding to m > 0.25
and ш < 0.25 respectively.
In figure 7a the behaviour of у is very similar to that of a in figure 3, except that
the range of у is finite. The broken line corresponds to one of the limiting curves
ш = 0.25. It can be seen that when w is near to but not equal to 0.25 the angle у
tends to ‘ linger’ in the neighbourhood of 45°.
In figure 7 b (ш < J) the behaviour of у is similar to that of a in figure 4, except
that when m = 0 (the rotating ‘ slab’) the graph of у is the straight line у = 0.
Again the solutions fall into two classes. The first consists of those for which у
starts from 90° (at t = —*'), decreases to a minimum (greater than 45°) at t = 0,
and then increases again to at finite time t = t\ In the second class, the angle у
starts from near zero at t = —oo, increases to a maximum at t = 0 and then falls
towards zero again as t -+ -f oo.
In figure 7a we show one of the limiting curves for w = 0.25. However, the other
limiting curve, in which у starts from near 45° and decreases monotonically to
zero, cannot be shown in figure 7b, since it is not symmetrical about the instant
t = 0.
To evaluate the angle of orientation $ we may begin from equation (4.20), which
can be written in the form

(8.3)

by equation (8.2). Hence

(8.4)

which can be calculated from the expressions for у as a function of t/A which have
just been found. In certain cases, however, it is more convenient to evaluate S
independently in terms of £. For, from (3.15) and (8.1) we have

А£ = £2(1 - £ 2+ ш£*)* (8.5)

and hence from (8.3)

(8 .6 )

In particular when w — J this becomes

(8.7)

which can be expressed in terms of elementary functions:

( 8 . 8)

Graphs of 8 are shown in figure 8a, b for the cases ro ^ J respectively. In figure 8a
it will be seen that when m > 0.25 the angle of orientation increases monotonically
 1875

470 M. S. Longuet-Higgins

F i g u r e 8 . For legend see facing page.

to a finite, non-zero value. The limiting values are indicated on the right of the
diagram, and also in table 1. In the special case ш = J, the angle 8 increases without
limit, as we would expect from equations (7.2) and (8.8).
In figure 86 we see the behaviour of 8 when w < J. In this case also 8 tends mono-
tonically to some positive value (the limits are given also in table 1). When m = 0,
as shown by the lower broken curve, the limiting value of 8 is 90°, that is the axes
have turned through a right angle from their position at t = 0. The upper broken
curve in figure 8b corresponds not to equation (8.8) but to the limiting solution of
§7, in which the rate of rotation is uniform, by equation (7.15). In fact we have
simply
8 — t/\ . (8.9)
1876

On the form ing of sharp comers at a free surface 471

m\
F i g u r e 8. T h e angle 8 o f rotation o f the principal axes, as a function o f
the tim e t (a ) when m > J, (fr) when w < J.

9. I l l u s t r a t io n s

As an example of the use of figures 7 and 8 we show in figure 9a the evolution of

the free surface in the typical case w = 0.30. Taking as the starting point the
instant when the asymptotes of the hyperbola are at right angles to one another,
and supposing, for illustration, that the axes are at 45° to the vertical, we can
follow the form as the axes rotate clockwise and the vertex angle у diminishes
monotonically to zero. The total angle 8 through which the axes have turned as
t oo is in this case 108° (see table 1). The distance a of the vertex from the origin
satisfies, by equation (4.16),
ab = constant, (9.1)
where 6 = a tan £y. Hence
a oc (tan £y)“*. (9.2)

In figure 96 we show the case w = 5.0 in which the total angle of rotation 8 is
only 39°. W e have also started at time t — 0 with a different initial orientation.
Figure 9c illustrates the critical case w = J when the axes rotate indefinitely with
a uniform angular velocity.
Figure 10 shows two cases when w < J. In figure 10a the axes rotate through 14.5°,
at which instant the vertex angle is 52°, before returning again to a vertex angle of
90°. In figure 106 the angle between the asymptotes starts from 0, increases to
nearly 32° when t = 0, and then falls again to 0.
 1877

472 M. S. Longuet-Higgins

tjA = 0 0.5 1.0 1.5 2.0

F i g u r e 9. Examples of solutions in class 1 , when w ^ J; (a) w = 0.30, (6) w = 6.0, (c) the
critical case xn — 0.25. The origin О is here stationary, or moving in a horizontal straight
line.

In all the examples of figures 9 and 10 it was assumed that the origin 0 was
stationary, or at least progressed uniformly in a horizontal straight line. In general,
however, О is in a free-fall trajectory. Such a trajectory is described by the para-
metric equations x - x, = - <,)»,■>

У -V i = U V -h )’ ) 3>
where U denotes the horizontal velocity and {xv y j denote the coordinates of the
highest point of the trajectory, attained when t = tv By an appropriate choice of
xv y v tx and U we can construct some suggestive time histories (figure 11).
It will be noted that in the trajectory (9.3) we have
dу dy/dt _ U
(9.4)
da; dx/dt g(t - у ’
so that the inclination 6 of the trajectory to the vertical is given by
U
в = arctan (9.5)
g(t —f|)
1878

On the form ing of sharp comers at a free surface 473

-0.152 0.152 0.304

f/A=—2.0 -1.0 0 1.0 2.0

F i g u r e 10. Examples of solutions in classes 2 and 3.
(a ) w — 0.20 (class 2), ( b) ш = 0-20 (class 3). О moves linearly.

Now for the hyperbolae described in § 5 we have, as t -* oo,

a ~ l /*, AS ~ A2a2 ~ A2/t2, (9.6)

from equations (4.20) and (5.3). Therefore

$ - 8 00 ~ - A / t . (9.7)

Comparing equations (9.5) and (9.7) (and bearing in mind that the hyperbola is
rotating clockwise) we see that the axes of the hyperbola will follow the parabolic
path of О rather well if we have $тлх = 180° and A = U /g (see figure 11 a). This
leads us to define the matching parameter

fi = A g/U . (9.8)

For an optimum match we must have ц —

If > 180°, the tip of the breaking wave will (if allowed sufficient time) curl
right over, as in figure lib , while if < 180° it will be pointing forwards relative
to the trajectory of O. In the critical case ц = 1 the flow will look particularly
simple.
It will be seen that in the special case w = 0 the angle 8, given by equation (6.7),
can be precisely equal to в over the whole range of t. Thus for a very slender hyper­
bola the direction of the tip of the jet can be perfectly matched to the parabolic
trajectory.

17 Vol. 37i. A
 1879

474 M. S. Longuet-Higgins

s -v .
(a)

F ig u r e 11. Local solutions with the origin О in a free-fall trajectory, (а) ш = 0.3, /* = 1 and
= 180°. (b) w = 0.26, ц = 1 and = 204°; (с) та = 0.261, = 2.0 and = 263°.
ft denotes the ‘ matching parameter’ Ag / U .

It should be borne in mind that the tip of the wave is not generally at О but at
a distance a from О given by equation (9.2). But since from (4.17)

cosy ~ 1—£ш(Аа)4 (9.9)

as / - » oo it follows that
у ~ ш*(Аа)2 ~ w*(A/<)2, (9.10)
and so from (9.2)
a/ ao ~ (2/y)* ~ (4/rn)U/X. (9.11)
This is small compared with the vertical displacements in the parabola, which are
of order gt2.
1880

On the forming of sharp comers at a free surface 475

10. H ig h e r - o rder t e r m s : m a tc h in g

In this section we investigate the possibility of solutions which are of higher order,
leading to the formation of a cusp at the free surface.
From figure 1 it will be seen that the simplest way in which this can occur is
for the saddle point at О to coalesce with another stationary point of the pressure
field. This second stationary point O' must clearly be a minimum of p. In general
the pair of stationary points may be represented locally by a cubic expression in z.
Accordingly we now write
X = \Az2+ \ B z * (10.1)
and calculate p and Dp/DJ from equations (2.4) as before. To order r3 we now
obtain
— 2p = — 2F + %(Az2+ ^^4*zz* + c.c.) + J(^z3+ 3^*Bz2z* + c.c.)
and —2q = — 2& + Щ А + 2A2A * )z 2+ 4AA*zz* + c.c.]
+ $[(£ + 6A A * B ) z3+ 6[A 2B * + А * В + А * Ё ) z *z * + c.c.]. (10.3)
I f we assume that the four cubic terms in equation (10.2) are in the same proportion
as before to the corresponding terms in (10.3) then we obtain four further relations
for the two unknown functions \B\ and arg.fi. But our assumption is unnecessarily
strong. It is quite sufficient, in order that p and q shall both vanish on the same
surface, to suppose that , / # _ ( , / * ) ( ! + * ), (10.4)

where R denotes a real power series. Provided that we retain the original solution
for the coefficient A of the quadratic terms in x> we easily see that R has to be at
least of third order :
R = i(i21z» + 3i?2z*z* + c.c.) + ... • (Ю.5)
On substituting the above expressions for p , q and R into equations (10.4) and
equating coefficients of the cubic terms we have

B + Q A A *B + R X 2(A 2B * + A * B + A * & ) + R 2 , 1ЛЛЧ

6 2 *в " °°- “ T 1 '

The functions A and F being already known from § 3, there are four equations for
the six functions consisting of the real and imaginary parts of R v R 2 and B. The
coefficient В is therefore undetermined by local conditions; if В is given, equation
(10.6) serves simply to determine Ri and Д2.
The addition of higher-order terms to R in equation (10.5) does not alter the
situation, since at each stage new coefficients i?3, J?4, ... are introduced which are
sufficient to accommodate the higher coefficients C, D , ... in x-
Thus the higher coefficients in x arc available for satisfying other conditions, for
example of convergence, and of matching the local flow to a particular flow at large
z. Details of this procedure will be described in a future study.
Meanwhile a particularly simple case may be noted here, namely when
Б = е »^ + т )-2, (10.7)
17-3
 1881

476 M. S. Longuet-Higgins

where v and r are constants (r > 0). On substituting for В in equation (10.2) and
defining x', y\ e and £ as in § 4 we obtain for the free surface, as far as the terms in r3,

(ft + a2) у/2= - Sa* + (ft - a2) s'2

+ + г)"3[cx(*/3- ЗяУ2) - 5l(3xy 2- у

'*)]
- 2a(* + г )"2(ж'2+ у'2) (с2ж' - s2y'), ( 10. 8)

where Cj = cos (v - 3$), s1 = sin (v - 3S), 1

(10.9)
c2 = cos(y-3 £+e), 8г — s i n (v -3 $ + € )J

Now as t со we have a ~ t~x -> 0 and from the definition of ft in (4.6)

(ft + a 2) - 2a2, (ft - a2) - \Pa\ ( 10.10)

Hence when t p Л equation (10.8) becomes

y '2 = - S a * + lPa*x'*

+ а ( т 1ж/3+ т2х,гу>+ т 3ж У 2+ т 4у/3), ( 10. 11)

where mlt m2) m2 and m4 are all of order 1 in a, and in particular

ml — §ci —C2 —£с1э ( 10.12)

since e -> 0 as t -> oo. It will be seen that when xf is 0(1) then y ' is 0(a\) and the
cubic terms in equation (10.11) dominate the right hand side. Hence the form of
the free surface tends towards a cusp (figure 12).
Moreover, we have seen that in the outer flow field the fluid, in the free-fall
reference frame, will have an upwards (i.e. negative) acceleration. This makes it
probable that in flows closely matched to the outer field the real part of В will be
negative, producing a cusp pointing upwards. On the other hand in an overturning
flow, for example the tip of a plunging breaker, the region in the tip is connected
only very remotely to the flow in the rest of the wave. In this case it does not appear
necessary for cubic terms to enter the expansion. The appropriate curvature of the
jet can be produced by quartic terms or terms of even higher order.

11. D iscussio n

W e have obtained certain exact solutions to the equations of time-dependent

free-surface flow, which appear to represent possible limiting forms for breaking
gravity waves. Generally the solutions involve elliptic integrals, which are readily
evaluated numerically, but in one interesting case, when vj — J, the solution

becomes <f>+ i f = ( 1/AJ 2) e™ ‘ (x + iy)\ (11.1)

which is so extremely simple that it seems surprising that it was not discovered
1882

On the forming of sharp corners at a free surface 477

.0

l
F i g u r e 12. The class 1 solution rn = 6.0, with a cubic term included, showing the near
formation of a cusp. In equation (10.7), v = 220° and r = A = 1 . Also S = 1/64.

previously, f Equation (11.1) represents an irrotational flow with a free surface in

the form of a hyperbola, with angle 45° between the asymptotes, rotating at a
constant angular velocity A.
Apart from the case w — J, the flows that we have obtained may be classed as
follows.
Class 1 (тп > J). The vertex angle starts at 90° and tends monotonically to zero
as t -> oo.
Class 2 (rn < J). The vertex angle starts at 90° and decreases to a minimum greater
than 45°, then returns again to 90° at a finite value of the time t.
Class 3 (m < J). The vertex angle starts from small angles when t is large and
negative, increases to a maximum less than 46°, and then diminishes to zero again
as t -> +oo.
We have suggested that solutions in class 1 can represent limiting forms near the
tip of a breaking wave. Although the vertex angle у does not reach zero in finite
t There is considerable literature on rotating ellipsoids and cylinders of elliptical cross
section, under the influence of a mutual attraction of the particles. See, for example,
Gieenhill ( 1879, 1880). The hyperbolic case seems to be generally excluded owing presumably
to the infinite gravitational potential. There is no elliptic case corresponding to equation (11.1).
 1883

478 M. S. Longuet-Higgins

time, nevertheless the protruding jet becomes indefinitely thin and would in practice
be easily broken up into water droplets and spray by the action of surface tension
and air currents. In upwards-pointing waves the appearance of a cusp will be
enhanced by the presence of cubic and higher-order terms which will tend to domi­
nate over the quadratic terms as t tends to infinity.
Solutions in class 2, on the other hand, represent waves which at first seem to
advance towards breaking, but then subside into their former state. They may
represent limiting forms of the surface in partly reflected waves.
Class 3 solutions may represent the tip of a hydraulic arch.
In classes 1 and 2 our choice of initial instant as being the time when the angle у
is equal to 90° is to some extent arbitrary, and is based on the fact that for angles
у greater than 90° the pressure at О becomes greater, not less than, the pressure
at the free surface. Before this instant, the solutions would hardly be applicable.
Finally, in this paper we have studied only the local form of the free surface
near a region of sharp curvature. It remains to study the flow in the rest of the fluid
(the outer region). In a sense our solutions play a role in time-dependent flows
similar to that played in steady progressive waves by the Stokes 120° comer flow,
or by the almost-highest waves of Longuet-Higgins & Fox (1977). To obtain a
complete picture of a breaking wave, for example, we must match our local solutions
to some appropriate flow in the outer zone, such as will be described in Longuet-
Higgins (19806).
R eferences

Biesel, F. 1952 Study of wave propagation in water of gradually varying depth. Gravity
Waves, U.S. National Bureau of Standards, Washington, Circular no. 621, pp. 243-263.
Greenhill, A. G. 1879 On the rotation of a liquid ellipsoid. Proc. Cam. phil. Soc. 3, 233-246.
Greenhill, A. G. 1880 On the general motion of a liquid ellipsoid under the gravitation of itR
own parts. Proc. Cam. phil. Soc. 4, 4-14.
Longuet-Higgins, M. S. 1972 A class of exact time-dependent, free-surface flows. J. Fluid
Mech. 55, 629-543.
Longuet-Higgins, M. S. 1977 On breaking waves. Proceedings of the symposium on waves on
water of variable depth, Canberra, July 1976. Lecture Notes in Physics no. 64, pp. 129-
150. Berlin: Springer.
Longuet-Higgins, M. S. 1980a A technique for time-dependent free-surface flows. Proc . R.
Soc. Lond. A 371, 441-451. (Paper I.)
Longuet-Higgins, M. S. 19806 A branch-point model of a breaking waves (in preparation).
(Paper III.)
Longuet-Higgins, M. S. & Cokelet, E. D. 1976 The deformation of steep surface waves on
water. I. A numerical method of computation. Proc. R. Soc. Lond. A 350, 1-26.
Longuet-Higgins, M. S. & Cokelet, E. D. 1978 The deformation of steep surface waves on
water. II. Growth of normal-mode instabilities. Proc. R. Soc. Lond. A 364, 1-28.
Longuet-Higgins, M. S. & Fox, M. J. H. 1977 Theory of the almost-highest wave. J . Fluid
Mech. 80, 721-741.
Thom, R. 1969 Stability structurelle et morphogenbse. New York: Benjamin.
Tulin, M. P. 1964 Supercavitating flows - small perturbation theory. J . Ship Res. January,
pp. 16-37.
Zeeman, E. C. 1969 Breaking of waves. Proceedings of the Symposium on Differential Equations
and Dynamical Systems, University of Warwick, 1968, 3. Berlin: Springer.
Zeeman, E. C. 1975 Catastrophe theory: its present state and future perspectives. Warwick
dynamical systems (Springer Lecture Notes), pp. 1-34. Berlin: Springer.
1884

J. Fluid Mech. (1980), vol. 99, part 4, pp. 705-713 706

Printed in Cheat Britain

Modulation of the amplitude of steep wind waves

B y M . S. L O N G U E T - H I G G I N S
Department of Applied Mathematics and Theoretical Physics, Silver Street, Cambridge,
England, and Institute of Oceanographic Sciences, Wormley, Godalming, Surrey

(Received 1 November 1979)

Lake & Yuen (1978) have suggested that in very steep wind waves the modulation -
frequency of the wave amplitude may correspond to the frequency of the fastest-
growing subharmonic instability of a uniform train of waves whose amplitude equals
the mean wave amplitude d. The approximate theory of Benjamin & Feir (1967)
gives this frequency as (ak)fd, where к is the wavenumber and f d the frequency of the
unperturbed waves. This expression applies strictly only to very small values of the
wave steepness ak.
More recently (Longuet-Higgins 1978) the present author calculated accurately
all the normal-mode instabilities of steep gravity waves on deep water. In this note
these calculations are used to determine the frequency of the fastest-growing sub-
harmonic instabilities precisely. When compared with the experimental data of Lake
& Huen, these frequencies show even closer agreement.

1. Introduction
In the study of random seas, questions concerning the fluctuation in the height of
the waves - for example, how many waves, on the average, are there in a wave group ?-
have important applications to naval architecture and ocean engineering, especially
when we are concerned with nonlinear phenomena (such as ship slamming) stimulated
by a resonant response to the waves.
In the past such questions have been treated mainly by linear theories (Longuet-
Higgins 1956; Goda 1970; Ewing 1973). Thus, there is a demonstratable relation
between the group-length of a gaussian stochastic process and the ‘width’ of the
corresponding frequency spectrum, suitably defined. Such linear theories do not
predict any particular value for the width of the spectrum, but leave this as a matter
either for measurement or to be estimated by wave-forecasting techniques.
In very steep waves, however, the waves become non-gaussian, and the frequency
spectrum, as ordinarily defined, becomes contaminated by the presence of second
and higher harmonics which are phase-bound to the corresponding fundamental
Fourier components. So the appropriate width of the spectrum tends to be over­
estimated, when spectral moments are used. Against this, it has appeared that the
width of the main peak in the spectrum, under conditions of active wave generation,
can be extremely narrow (Hasselmann el al. 1973).
Now it is known (Benjamin & Feir 1967) that a uniform train of gravity waves of
finite amplitude a in deep water is inherently unstable to certain subharmonic per­
turbations. Benjamin & Feir represented these as side-bands, with radian frequencies
o' ± A<r, where cr is the frequency of the fundamental and До* a positive perturbation;
0022-1120/80/4513-1320 $02.00 <g> 1980 Cambridge University Preaa
23 flm 99
 1885

706 M . S. Longuet-Higgins

F ig u r e 1. Sketch o f the surface elevation rj at a fixed point,

as a function of the time t.

and they showed that the side-bands would grow at the expense of the fundamental
wave provided
Acr/cr < л/2 ak, (1.1)
к being the fundamental wavenumber. Moreover, the most rapidly growing perturba­
tion (of this type) has a frequency such that
До-/сг = ak. (1.2)
This interesting result was taken up by Lake & Yuen (19 7 8 ) who suggested that the
fluctuations of the wave envelope in a steep, irregular wave train might correspond to
the most rapidly growing perturbations of a uniform wave train. In support of this
not unreasonable suggestion they measured the ratio of the modulation frequency
f e to the dominant wave frequencyf dof wind-waves in a short channel, with fetches up
to 30 ft and wind speeds up to 30 ft s_1. Despite some scatter, f j f dappeared to increase
about linearly with mean wave steepness (see Lake & Yuen (1978) figure 13; also see
figure 6 below).
Now the theoretical results (1.1) and (1.2) are valid strictly only for small but finite
values of A cr and ak. Recently the present author has calculated accurately the in­
stabilities of deep-water gravity waves over the greater part of the range of wave
steepness ak (Longuet-Higgins 1978). These calculations have been confirmed by an
entirely different method (Longuet-Higgins & Cokelet 1978). The results show that the
instabilities are of two distinct types: local instabilities, which lead directly to
whitecapping, and subharmonic instabilities of the Benjamin-Feir type, which
however are confined to the range 0 < ak < 0-36 approximately. It was shown that
the relations (1.1) and (1.2) are valid only at the lower end of the range of ak.
In this note it will be shown that the results of Longuet-Higgins (1978) allow the
frequency of the most rapidly growing subharmonic perturbation to be determined
accurately. When the resulting curve of f e/fd is substituted for the straight line (1.2)
used by Lake & Yuen, rather better agreement with their data is obtained.

2. The modulation frequency

Consider a record rj(t) of the sea surface elevation at some fixed horizontal position,
say x = 0, as shown schematically in figure 1. The broken line represents the modulus
1886

Modulation of the amplitude of steep mind waves 707

rje = |^4(f) | of the envelope function, drawn so as to touch the dominant waves near
to their crests. The envelope is in fact the smoothest such function (in a certain sense)
which can be so drawn. We wish to find the mean frequency (in some sense) of the
fluctuations of 7fe(t).
Now in the proposed model we regard 7}{t) as the result of modulating a uniform
train of waves of finite amplitude, that is an unperturbed wave

rj = f(k x — o i ) — 'Z af cob j{kx —oi) (2 .1 )

with af real. Here x and t represent the horizontal co-ordinate and the time. To retrieve
rj(t) we set x = 0.
It is useful to consider the perturbations of 7} in a frame of reference moving with the
phase-speed с = <r/к. Then 7} itself becomes independent of the time:

rj —f(k x ') = 2 %cos jkx', (2.2)

where x' = x —ct. It is further convenient to express 7} as a function of the velocity

potential ф in this relative motion, so that

} =* F ty ) = S Л,сов( jty/c).
7 (2.3)

At a wave crest we take x' = 0, ф — 0 so the wave amplitude a is given by

= ( 2 -4 )

W e now may express the elevation 7} in the form

ч ~ ч (Ф )+ Ш * )> (2-6)
where т)' is a small perturbation of the basic wave rj. Squares and higher powers of
tj' (but not rj) are neglected. It is relatively straightforward to calculate the TLormal
mode perturbations, which take the form

= (2.6)
where Pn and Qn are real and imaginary parts of the complex amplitude function: n is
a rational number characteristic of the particular normal mode; and

<r» = * n+ iPn (2-7)

say. The real part an gives the radian frequency of the normal mode, in the travelling
reference frame, and the imaginary part fin gives its e-folding rate of growth.
The normal modes were calculated explicitly in Longuet-Higgins (1978). It was
shown that in general we can write
n = l/m, (2.8)
where I and m are integers, m denoting the number of wavelengths (in a horizontal
direction) over which the particular mode is periodic. Thus m = 1,1 ^ 1 corresponds
to all the superharmonics having length scales equal to or less than the wavelength
2тг/к, while m = 2, 1 = 2, for example, corresponds to all the subharmonics which are
two wavelengths long; and so on.
The values of I are assigned to each harmonic in such a way that the ratio n = l/m
gives the number of maxima and minima of each harmonic at low values of ak) positive
33-a
 1887

708 M . S. Longuet-Higgins

ak
F i g u r e 2. Real part of the radian frequency <r of normal-mode perturbations of deep-water
gravity waves, as a function of the wave amplitude ak (after Longuet-Higgins 1978).
Im ( a )

0 01 0-2 0-3 0-4

ak
F i g u r e 3. Imaginary part of the radian frequency <r (i.e. growth rate) of normal-mode instab­
ilities of deep-water gravity waves, as a function of the wave amplitude ak (after Longuet-
Higgins 1978).
1888

Modulation of the amplitude of steep wind waves 709

F ig u r e 4. Sketch of the surface elevation tj of a perturbed train of waves as a function

of the horizontal distance x.

values of n correspond to modes propagated in the same sense as the unperturbed

wave, negative values of n to modes propagated in the opposite sense.
Some of the modes were connected analytically to more than one mode at low ak.
For these it is necessary to define a pair (or more) of wavenumbers, as in figure 3 below.
Figures 2 and 3 respectively show the real and imaginary parts of a when m = 8, so
I = 8n.
Now we need to obtain the envelope function 7je{t) in the stationary frame of reference.
To do this we sample Tj(t) at points near the crests ф = N cL and at times t = N r,
where N is an integer and т is the basic wave period,
L = 2n/к, т= 2n/crd. (2.9)
Thus we consider the set of points
V(N t ) = a + [P (N c L ) + iQ (N c L )] e- ***»*, (2.10)
where t = N t , as defining the wave envelope r)t approximately (see figure 4).
For example, when m = 2 and we consider odd harmonics (I odd), P and Q will
reverse sign when ф is increased by cL (t being fixed). In other words, on spatially ad­
jacent waves the perturbations are of opposite sign. Hence, as each crest passes the
fixed point of observation (x = 0) the vector [P (N c L ) + iQ (N c L )] will be constant in
magnitude but will alternate in sign; hence it will oscillate with radian frequency
\(fd. Meanwhile the vector e- **»»4will rotate with radian frequency an. So the apparent
frequency, in the stationary frame of reference, is (£ard- a n). The term £crd clearly
represents a Doppler shift, compounded of the phase-velocity and the wavenumber
\ of the basic disturbance.
More generally, when the normal mode has two dominant components with wave­
numbers nx and л 2, say, the dominant spatial wavenumber will be Дn = \{пх—пг).
For the unstable subharmonic modes shown in figure 3 we have
8An= 1,2, ...,6. (2.11)
Clearly the ‘ Doppler shift’ is just equal to сДп. From this it follows that in general the
frequency of modulation in the stationary frame of reference is related to the basic
frequency f d = 2nard by
 1889

M . S. Longuet Higgins

8Дn An ak <*n/(r9 <rd/ar9 *n/<T4 Л/Л

10 0126 0-072 0-062 1 003 0-062 0-063
20 0-260 0-155 0-118 1-012 0-117 0-133
3-0 0-376 0-246 0-163 1-031 0-168 0-217
4-0 0600 0-309 0-198 1-049 0-189 0-311
60 0-626 0-361 0-221 1-063 0-209 0-416
6-0 0-700 0-382 0-231 1076 0-216 0-635
Т а в ь е 1. Calculation of the modulation frequency f t induced by the most unstable mode

3. The most unstable mode

To determine the frequency of the most unstable mode at any given wave steepness
ak we may proceed as follows.
In figure 3, we have drawn with a broken line the envelope of the solid curves.
No solid curve lies above this envelope. Therefore at a point of contact of the envelope
with any particular curve the corresponding mode has the fastest rate of growth of any
mode at that particular value of the steepness ak. To determine the corresponding
modulation frequency a, we now go to figure 2 and read off the value of Re (<r) corres­
ponding to that particular mode at the same value of ak.
The resulting values of oLt/a 0for each of the unstable modes in figure 3 are shown in
table 1. Here <r0 represents the radian frequency of infinitesimal waves (ak -+■ 0). To
obtain the ratio al/crd we must divide by the relative speed of waves of finite amplitude,
which may be calculated, for example, from figure 1 of Longuet-Higgins (1975).
Finally f j f d is found from equation (2.12).
The figures in the final column have been used to draw the curve in figure 5, showing
crt/crd against ak. Also shown are the tangent at the origin, representing equation (1.2),
and a vertical line marking the amplitude of the most unstable wave which, from
figure 3, is found to occur at a steepness ak = 0*32 approximately. It can be seen that
for the lower part of the range of ak the corrected curve lies somewhat below the
sloping line representing the Benjamin-Feir theory, but that it crosses this line at
about ak = 0-30 and thereafter lies above the line.
In fact since the steepest waves in an irregular sea will exceed the average steepness
by a factor of 1*5 or more (Longuet-Higgins 1952) we hardly expect to find mean
values of ak much above 0*3 and for those the corrected curve always lies below the
Benjamin-Feir tangent.

4. Comparison with observation

In figure 6 the theoretical curve representing f j f d has been inserted in figure 13 of
Lake & Yuen (1978), which includes the data derived from a wind-wave channel at
rather short fetches. It will be seen that the new curve agrees with the data better than
the original theory, in so far as it goes more nearly through the centre of the cloud of
observations.
However, it must be said, first, that the measurement of the envelope frequency f t
from a typical record may be a somewhat subjective procedure and, second, that there
is no obvious reason to prefer the mean wave amplitude ak as a parameter of the wave
 Modulation of the amplitude of steep wind waves

ak
F ig u r e 5. The frequency of modulation of the wave envelope in a wave of initial
amplitude a perturbed by the fastest-growing instability.
 1891

M . S. Longuet-Higgins

F i g u r e 6. Ratio of the modulation frequency to the dominant wave frequency, accordng to the
laboratory wind-wave data of Lake and Yuen (1978), compared to the most unstable
modulation frequency. Broken line represents Benjamin & Feir’s (1967) theory. Solid curve
represents present calculations. О» Uw = 15fts-1; □ , Uw = 2 0 ft а-1; Д, Uw = 26fts-1;
0 , Uw = ЗОЛв-1; V , Uw = 36 ft s-1.

field rather than say the root-mean-square wave amplitude, which would be somewhat
greater.
Nevertheless one can state, on the evidence of figure 6, that in the steepest waves
encountered in these experiments the frequency of modulation was about 0*2 times
the wave frequency, implying that in a very steep sea every fifth wave, on average, is
the highest. And, further, that this is supported by our calculations.
It must be emphasized that this would be the wave grouping as seen by an observer
or wave recorder at the fixed point x = 0, measuring the elevation as a function of
the time. When seen in space, the wavenumber of the envelope would appear to be
simply
Д п = | ( п 1- п 2). (4 .1 )

For waves of low amplitude, and fairly long groups we have in deep water
ал = Дo 'Ф \c Дn, (4.2)
1892

Modulation of the amplitude of steep wind waves 713

since the group velocity equals half the phase velocity. Hence the wave groups appear
twice as long in time as they do in space.
The lower the waves, according to these ideas, the longer the groups of waves become,
until the width of the spectrum ceases to be governed by the local nonlinear dynamics
and is determined instead by other factors depending on the synoptic wind pattern
and the path of wave propagation.
For field measurements of ocean waves, the wave age, as given by the ratio c /U
where U is wind-speed, is generally much greater than in the laboratory experiments
reported by Lake & Yuen (1978). It appears that typically c /U lies between 0-6
and 1*5, and hence that ak, according to Sverdrup & Munk (1947) lies between 0-3
and 0*05, but with the bulk of the observations corresponding to c /U ~ 1*0, ak ~ 0-10.
It will be interesting to see to what extent the observed values of f j f d lie near the
theoretical curve of figure 6.

This paper was prepared during February 1979 while the author was a visitor in the
Department of Engineering Sciences at the University of Florida, Gainesville, U.S.A.
The author is indebted to Dr К . Millsaps and other members of the Department for
their kind hospitality and assistance.

REFERENCES
B e n j a k i n , Т . B. <fe F eer, J. E. 1967 The disintegration of wave trains in deep water. Part I.
Theory. J . Fluid Mech. 27, 417-430.
E w c s g , J. A. 1973 Mean length of runs of high waves. J. Oeophys. Res. 78, 1933-1936.
G o d a , Y . 1970 Numerical experiments on wave statistics with spectral simulation. Rep., Port
Harbour Res. Inst., M in. Transport, Yokosuka 9, 3.
H a s s e lm a n n , K . 1973 Measurements of wind-wave growth and swell decay during the joint
North Sea Wave Project (JONSWAP). Deut. Hydrogr. Z. Suppl. A 8 (12).
L a k e , В. M. & Y u e n , H. C. 1978 A new model for nonlinear wind waves. Part 1. Physical
model and experimental evidence. J. Fluid Mech. 88, 33-62.
L o n g u e t - H ig g in s , M. S. 1962 On the statistical distribution of the heights of sea waves.
J. Mar. Res. 11, 246-266.
L o n g u e t - H ig g in s , M. 8 . 1966 The statistical analysis of a random, moving surface. Phil. Trans.
Roy. Soc. A 249, 321-000.
L o n g u e t - H ig g in s , M. S. 1976 Integral properties of periodic gravity waves of finite amplitude.
Proc. Roy. Soc. A 342, 157-174.
L o n g u e t - H ig g in s , M. S. 1978 The instabilities of gravity waves of finite amplitude in deep
water. Proc. Roy. Soc. A 360, 471-605.
L o n g u e t - H ig g in s , M. S. & С о к е ь е т , E. D. 1978 The deformation of steep surface waves on
water. II. Growth of normal-mode instabilities. Proc. Roy. Soc. A 364, 1-28.
S v e b d ru p , H. U. <b M u n k , W. H. 1947 Wind, sea and swell: Theory of relations for forecasting.
UJS. Hydrographic Office, Publ. no. 601.
 1893

Proc. R. Soc. Lond. A 376, 377-400 (1981)

Printed in Great Britain

On the overturning of gravity waves

B y M. S. L o n g u e t - H i g g i n s , F.R.S.
Department of Applied Mathematics and Theoretical Physics ,
Silver Street, Cambridge CB3 9E W , and Institute of Oceanographic Sciences,
Wormley, Surrey G U 8 5UB, U.K.

(Received 30 October 1980 )

It is shown that an approximation to the initial stages of overturning in an

irrotational gravity wave is given by the potential function
X = §igr*z* + 2^4zi,
where g denotes acceleration due to gravity, z is a complex coordinate in the
plane of motion, and A is a linear function of the time t.
B y adding to the above expression a third term, linear in z, an expression
is obtained which can describe the development of sharp corners or cusps
in the free surface.

1. I n t r o d u c t i o n
The sight of the sea surface overturning on itself, as when waves break in deep water
or on a sloping beach, is familiar to most mathematicians. Y e t surprisingly little
progress has been made in finding an analytical description of the free surface
profile and of the corresponding field of flow. Recently Longuet-Higgins & Cokelet
(1976,1978) demonstrated a purely numerical method of computation in short time-
steps, at each step solving an integral equation. But to comprehend the results of
such a computation or to handle them conveniently, the numerical calculation
needs to be complemented by some exact or approximate analysis.
A purely local solution describing how the free surface can develop a sharp
curvature near the overturning tip of a wave crest was described in a companion
paper (Longuet-Higgins 19806). Here we propose to find simple expressions to
describe the process whereby the whole body of fluid overturns. W e are concerned,
however, only with the upper part of the wave, not too far from the wave crest.
It has sometimes been suggested (for example by Price (1971)) that a wave breaks
only after attaining the limiting form described by Stokes (1880) in which the crest
has a sharp comer of 120 ° (see figure 1). Thus Price proposed an approximate
solution in which the 120° corner-flow was taken as an initial configuration. H ow ­
ever, observation suggests that, on the contrary, waves generally break without
passing through the Stokes configuration, and that the 120° angle is a very special
case, not generally attained. This conclusion is strengthened by the recent demon­
stration that a limiting wave, with a 120° angle at the crest, actually has less energy
than is possessed by lower, symmetric waves with rounded crests (see Longuet-
Higgins & Fox 1978). Hence the limiting wave must be very difficult to attain
[ 377 ] x4-a
1894

378 M. S. Longuet-Higgins

experimentally, for it cannot be built up by gradual shoaling or by normal pressure

applied to the surface without the aid of some dissipative process.
The Stokes 120° comer-flow is nevertheless of considerable interest for our
investigation. Thus we shall obtain a tentative representation of the initial stage of
overturning simply by adding to the Stokes expression some time-varying term
representing a finite perturbation of the 120° comer-flow. In our model the flow does
not actually pass through the Stokes configuration. Indeed, the extent to which the
comer-flow is by-passed is a measure of the total dimensions of the overturning.
In any overturning flow, not only is the surface elevation multivalued but more
fundamentally the velocity field (extended analytically beyond the free surface) is
a multivalued function of the position coordinate. This is dramatically realized
when the tip of a ‘ plunging ’ breaker first comes into violent contact with the smooth
forward face of the wave. The analytic description of such a multivalued velocity
field must involve (at the very least) a branch point in the complex potential. The
simplest such branch point is of order £, that is to say it involves a square root.
Now the Stokes comer-flow already involves a branch point of order J. I f we
assume that the branch-point of order J coincides with that of order f we are led
immediately to the simple expression for the velocity potential x which is given in
equation (2.5) below. A discussion of the corresponding field of flow is described in § 2.
In §§ 3 and 4 we show that the boundary conditions may be satisfied asymptotically
at infinity by taking the quantity A in equation (2.5) to be a linear function of the
time t. Then the solution contains two arbitrary constants which may be chosen
(§ 5) so as to obtain the best fit of the boundary conditions over the finite part of the
wave. The resulting surface profiles are shown in figures 5a and b.
In the second half of the paper we show that an improved family of solutions can
be derived by adding to the potential function x a term that is linear in z, representing
a uniform upwards velocity. Moreover it is then possible to obtain solutions that
develop a sharp-pointed cusp at the tip of the wave, as seen in figure 12.

2. D e s c r i p t i o n o f th e flo w

For simplicity we shall assume the flow to be two-dimensional, incompressible,

inviscid and irrotational. In general the velocity potential x is then an analytical
function of the complex coordinate z = x + iy (where x and у are rectangular
coordinates in the plane of motion), and also of the time t. It is convenient to take
the real axis of z (i.e. the ж-axis) pointing vertically downwards.
Consider a wave crest that is being propagated horizontally to the right. Let us
take a reference frame moving to the right with a uniform velocity equal to the
phase-speed (suitably defined). Then the flow at some distance below the wavecrest
will appear to be backwards, i.e. to the left.
Figure 1 is a representation of Stokes’s comer-flow, in which

X = f i ? 12*, (2.1)
 1895

On the overturning o f gravity waves 379

F i g u r e 1 . Streamlines in Stokes’ 120° corner-flow. The fluid

occupies the region — 60° < в < 60°.

where g denotes acceleration due to gravity. The free surface consists of the two
planes
argz = ±$rc. (2.2)

The lower streamlines are in the water. The upper streamlines represent an analytic
extension of the flow into the ‘ air’ above. Because of the branch-point at z = 0 we
have to assume a cut in the z-plane. In this instance we have taken the cut to lie
along the line argz = Jrc.
Now in an overturning wave the streamlines are not altogether as in figure 1, but
curl back over to the right, as in figure 2. This diagram represents the streamlines of
the potential
X = 2Azi, (2.3)
where A is a complex constant:
A = -a e -* . (2.4)
The streamlines are parabolae with foci at z = 0, and with axes along the line
arg z = 2e. In figure 2 we have taken e = 30°, so the axes happen to lie along the line
arg z = 60° which is part of the free surface in figure 1.
1896

380 M. S. Longuet-Higgins

N ow let us combine the two flows (2.1) and (2.3) by writing

X = %izt + 2Az*. (2.5)
(For convenience we choose units in which g — 1.) The streamlines are shown in
figure 3b. It will be seen that for large values of \z\ when (2.5) is dominated by the
term in zi, the flow is asymptotically like the Stokes comer-flow. On the other hand
for small values of \z\, the term in z\ dominates, and the flow resembles the parabolic
flow of figure 2.
In general the stagnation point z — S is given by
d x/dz = iz b + A z -t — 0. (2.6)
Hence
z = iA — —ia e~Ie (2.7)
so the vector OS is inclined at an angle e to the horizontal (negative у -axis).
Another example of the streamlines corresponding to (2.5) is illustrated in
figure 3a. This corresponds to e = 0. W e shall show later that it is necessary to add
to the expression (2.6) for x a third term representing a uniform upwards flow.
Nevertheless it will be useful to discuss first the simpler form (2.6).
 1897

On the overturning of gravity waves 381

Re г

F i g u r e 3 (o). Streamlines of the flow given by equation (2-6) when e = 0°.

3. B oundary conditions

The flow being time-dependent, the free surface is not in general a streamline, but
instead may be specified by the two conditions
p = 0, D p /D t = 0, (3.1)

where p denotes the pressure and J )/D t denotes the rate of change following a
particle.
382 M. S. Longuet-Higgins

F i g u r e 3 (6). Streamlines of the flow given b y equation (2.5) when e = 30°.

It is convenient to write - 2p - P and - 2Dp/Dt = Q. General expressions for

P and Q were derived in Longuet-Higgins (1980 a), namely

P = XzXt + iXt + X t)-(z + * * ) - 2F (3.2)

and
Q = (x !x b + 2x*xit+ x*t) - X t - f + «•«•. <3-3)
where F is some function of the time t only, and c.c. denotes the conjugate complex
of all preceding terms.
 1899

On the overturning of gravity waves 383

In these expressions let us now substitute

X = %ш3+ 2А(о, (3.4)

where
со = zi (3.5)

and A is a function of the time, to be determined. W e have then

P = —((o2 —(0(0* + (o*2) -f 2 (Aoj + A *(o*)

+ i( A *(o2- А (о*2) /шо* —2F + A A */(0(0*, (3.6)

the terms being ordered in descending powers of |ш|. Also from (3.2)

(.A - A * ) ( ( o - ( o *) A*(os + A(o* 6 2(A A * + A *A )

10(0* 2(o*(o*3 ^ (0(0*

\AA*((o* — (o**) i(A * 2a) — A 2(o*) A 2A*cj + A A * 2(o*

+ 2(o2(o* 2 2(o3(o* 3 * '

The general problem, as formulated in Longuet-Higgins ( 1980a), is to determine the

functions A and F so that both P and Q shall vanish on the same surface (the free
surface). Although this may not be possible over the entire ranges of a) and t, we
shall show that over significant ranges of (O and t this can be achieved quite closely.

4. C o n d i t i o n s a s oj - > 00

I f A vanished for all t, then from equations (3.6) and (3.7) we should have

precisely, so that the vanishing of P implies also the vanishing of Q. From (3.5) the
free surface P = 0 is then given by

(o2— (0(0* + (o*2 = 0. (4.2)

This represents the pair of straight lines &Tg((o/(o*) = ± Jtu, that is argz = ± Jtc,
as in figure 1.
When A is not zero, but (o is large, we can still ensure that Q vanishes on the same
surface as P asymptotically, by making

(4.3)
1900

384 M. S. Longuet-Higgins

to order 1 as a) -> oo, where A is a real constant to be determined. Substituting for P

and Q from (3.6) and (3.7), and considering only the terms of order u>and 1, we see

that A = 0, (4.4)
and
2i(A *u )2— A w *2) — 2 = —i (A o ) + А * й )* ) (o) — (o*) —A(w2—ш * + o>*2), (4.5)

whence
2iA * = — iA — AA
\ (4.6)
—2Li = У * - Л J
and
- 2 ^ = i ( i - i * ) + A. (4.7)
From (4.6) we have
A + A * = 0, (4.8)
so that A is pure imaginary, and also
A = ii. ( 4 .9 )
Then from (4.7)
A = $i*\ (4.10)
Equation (4.4) also implies that

A = constant, A = A 0+ A t , (4.11)
and from (4.10)

where A 0and denote the values of A and F when t — 0. The constants A 0, F 0 and f 1
are at our disposal, A being related to F by equation (4.10).
I f А Ф 0 we may choose the origin of time t so that t = 0 corresponds to the
instant when A is real. In other words we may choose A 0 to be real. Assuming A 0 to
be negative, we may choose units so that

Л 0 = - 1, ( 9 = 1). (4.13)

F i g u r e 4. Locue o f the point A ( t ): (a ) w hen U — V = 0 ;

(6) w hen U < 0, V > 0.
 1901

On the overturning of gravity waves 385

as in figure 4. W e see then that the expression (2.5), if it is to satisfy the boundary
condition as o) -+ oo, has essentially two degrees of freedom, controlled by the two
real parameters F0 and t .
It would be possible to obtain further relations for determining FQ and by
considering lower-order terms in equation (4.3) but such relations would not
necessarily improve the fit of the boundary conditions except at large cj. Instead we
may choose and P so as to optimize the fit of the boundary conditions over the
finite part of the boundary, in the following way.

5. B oun dary c o n d itio n s f o r fin it e (o

To determine the two independent parameters F0 and we may agree to satisfy
the boundary condition D p /D t = 0 at a finite number of points (generally not more
than two) on the free surface p — 0. Alternatively, for any given values of F0 and P
we may calculate

а д . ^ ) = / ( Й ) г^ (6.1)

along the free surface p = 0, and then adjust F0 and / so as to make R a minimum.
The latter course has the advantage that the boundary condition is, so to speak,
distributed over the range —oo < у < oo, rather than being concentrated at one or
two discrete points in the free surface.
Adopting the second method, we find for the values of F0 and / the following:

F0 = - 0 .1 7 6 , f 1= 0.124. (5.2)

The corresponding free surface, at t — 0, is shown in figure 5a. A comparison with

the streamlines for e = 0, which are shown in figure 3 a, shows that the free surface
does indeed intersect the streamlines at a non-zero angle.
To find the form of the free surface at earlier times t < 0 we may proceed back­
wards, by writing
F = F0+ J^t, }
(5.3)
A = Ло + fiityJ

and taking t = - 1, - 2, say. (Note that f 1, not F , remains constant.) The

resulting profiles are shown in figure 5a for t = 0, — 2, — 4 , — 10. As time
regresses, the profiles become increasingly rounded. Indeed at t = — 10 there is
little sign of the overturning to follow, apart from a slight asymmetry in the surface
profile.
On the other hand as t increases from 0 the profiles rapidly become steeper, as
shown in figure 5b. The profiles are quite similar to those calculated numerically by
Longuet-Higgins & Cokelet ( 1978, figure 21).
1902

386 M. S. Longuet-Higgins

(a)

(M

F i g u r e 6. The form of the free surface that minimizes J(DP/D*)*d« at time t = 0, for
U = V = 0, A 0 = — 1 ; jF0 = — 0.176, F = 0.124. (a ) Successive profiles of the free
surface at previous times: t = — 10, — 8, ..., 0. (6) Successive profiles of the free surface
at subsequent times: t = 0, 1 ,..., 6.
 1903

On the overturning of gravity waves 387

The solution is nevertheless valid only for a limited range of the time t. In the
second half of this paper we shall show that, by adding to (2.6) a further term, linear
in 2, it is possible to obtain solutions in which the surface inclination exceeds 90°,
and the free surface develops a sharp comer or cusp.

6. A MORE G E N E R A L M ODEL
N o w let us adopt the more general expression

X — §iw3+ Ua) 2+ 2A(ot (6.1)

where U is a constant and A = A(t) as before. On substituting this expression in

equation ( 3 .2 ) we find

P = - (a)2— Q)0J* + w*2)

+ {2A + iU * )(o + (2A *-iU )< o *

+ i( A *a>2— A at*2)/оно* + ( U U* — 2F)

+ (UA*(o+ U*A<o*+AA*)/<o<o*. ( 6. 2)

Likewise if we substitute in equation ( 3 .3) and retain only the terms of highest
order in |w| we obtain

Q = i(w3- 2(o2(o* + 2(oo* 2- (0*3)/2(0(0* + 2(Ato + A*(o*)

+ [(U + 2iA *)(0 2- { U + U* + 2$ ) оно* + { U * - 2iA)(o* 2]/axo*

+ .... ( 6 .3 )

I f we now assume the relation ( 4 .3 ), then on equating coefficients of the terms of

highest order on each side we find

A = 0 = A* ( 6 .4 )
as before, but now
( U + 2i A * ) + A - i ( U * - 2iA) = 0 , \

(t/* —2 L i)+ A —i(l/ + 2 L 4 *) = 0, | (6.6)

_ ( t f + U * + 2f ) - \ + l ( U * - 2iA) + l ( U + 2iA *) = 0.J

On eliminating Л from these expressions we have

U + 2\A* = U* — 2iA , (6.6)

and
U + U* + 2# = $ ( V * - 2iA). ( 6 .7 )
Hence
A = $i(/ + i t / - i I 7 * ) , (6.8)

which generalizes equation (4.10).

Let us write
U = U + iV , (6.9)
1904

388 M. S. Longuet-Higgins

where V and V are real. Then equation (6.8) becomes

A = $ i( f+ iU ) -iV . (6.10)
Conversely
J = - l U - } i ( V + 2A). (6.11)

The form of the free surface is obtained by writing P = 0 in (6.2). Since from
(4.11) and (6.11) we have

4 « ( 4 o- * W ) + f i ( * + K 0 < (612)
we obtain
tо*—ом* -f cj** = |i(i^ + 17) (a>—(o*)
+ i ( ^ o- J F 0 (<o*-G>**)/oxi)*

+ $i(J?+ i U ) t (d)2 + w*2)/(iXD*

+ ( U 2+ V 2—2F0) — 2flt

+ 0 ( M " 1). (6.13)

Now, introducing polar coordinates by writing

z = re w, w= (6.14)
we find, to order r*1,

cos в - £ = - KJ?1+ C7) r“i sin 0

+ [|( J?■+ J U ) cos в + \ V sin 0 - 1 ] tr- 1

- [ Л sin в + F0- J (U 2+ F 2)] r - 1. (6.16)

On the forward face of the wave set

d = in + 8, (6.16)

where 8 is of order r-i. Thus by expanding each side of (6.16) in powers of r_* and by
successive approximation we easily obtain

rS = - , $ , ( * +

+ A 0+ M * 'o ~ i ( 17* + F*)]. ( 6 - 17 )

Similarly on the rear face of the wave if we set в = - (J* + 8) we find

rS = + и )г * + ^ ( # - 1 и - ^ У ) 1

-4 ,+ £ № > -* (t f* + F * )]. (6.18)

Consider now the implication of these expressions. In (6.17), the quantity r8
represents approximately the normal displacement of the free surface above the
straight line в = $n. For large values of r, the leading term in (6.17) represents a
parabola which is concave or convex upwards according to the sign of + U. For
a concave forward face (and convex rear face) we need to have

P + T J < 0. (6.19)
 1905

On the overturning o f gravity waves 389

The second term on the right of (6.17) or (6.18) represents an upward displacement
which is independent of r but increases linearly with the time. W e require that as
t -> — oo, so the normal displacement shall be negative, at least when r oo. This
implies that the coefficient of / in (6.17) or (6.18) shall be positive, that is

> 0. (6.20)

In the simplest case U = V = 0 discussed in §§ 4 and 6 the two conditions (6.19) and
(6.20) are clearly mutually contradictory. In the next-simplest case when V = 0,
U ^ 0, the conditions (6.19) and (6.20) imply that

U < 82?1< — SU. (6.21)

Hence U < 0, in other words the imposed uniform velocity must be upwards.
A physical interpretation is as follows (see figure 6). On the forward face of the
wave the upward uniform velocity combines with the upward component of flow
in the Stokes comer-flow (seen in the present frame of reference) to produce a larger
value of the absolute velocity, compared with the velocity at an equal distance on
the left, where the velocity in the Stokes comer-flow has a downward component.
N ow from Bernoulli’s equation

gx = + + (6.22)

it follows that, at two comparable positions 1 and 2 on the right and left respectively,

g{x х- х г) = l(? i-? | ), (6.23)

since at large distances <f>t is of order only and so relatively small. The right-hand
side of (6.23) is seen to be positive, showing that the displacement of the free surface
face is downwards on the forward face but upwards on the rear face. The whole wave
therefore tends to tilt clockwise.

F i g u r e 6. A p h y s ic a l in te rp re ta tio n o f th e ro le o f th e u p w a rd s
v e r t ic a l v e lo c it y (U < 0 ) in e q u a tio n (6 .1 ).
1906

390 M. S. Longuet-Higgins

(a)

F i g u r e 7 (a). Successive positions of the free surface corresponding to equation (6.1)

with U — — 1, V = 0.6. A 0, F 0 and F are chosen so ад to minimize J (D P / D 0 2de at
time t = 0: A 0 = -1 .7 0 2 , F 0 - -2 .4 92 , F = 0.403.

For given values of U and V we may choose the three variables A 0, F0 and J?* so as
to minimize JQ2ds taken along some central part of the free surface (say— 10
< Y < 10). In the typical case U = — 1, V = 0.5 we find the optimum values

A 0 = - 1.702, F0 = - 2.49, ^ = 0.402.

The corresponding r.m.s. value of T>p/T>t is 0.281.

Successive profiles of the free surface are shown in figure 7. For a quantitative
comparison with the profiles calculated numerically by Longuet-Higgins & Cokelet
( 1978), we may consider the time-history of the maximum surface slope a max,
shown by the broken curve in figure 8. This suggests that damax/d£ reaches a
maximum when a max is about 60°, or t = 6. The solid curve in figure 8 corresponds
to the profiles from figure 21 of Longuet-Higgins & Cokelet ( 1978). In these calcu­
lations, damax/d£ clearly continues to increase steeply beyond a max = 60°. Moreover
the corresponding time-scales in figure 8 are in the ratio т л 8.8. The length-scales,
on the other hand, are in the ratio I « 17.6, so that т/И is about 2.1 instead of unity
as we might expect (g is taken as unity in both cases).
 1907

On the overturning of gravity waves 391

F ig u r e 7 ( b ). A s figure 7a, but viewed in a frame of reference travelling

with velocity ( — 0.6, — 1.7).

The cause of the discrepancy can be traced to the fact that as t increases Dp/Df,
instead of remaining small, becomes of order unity near the wave crest. In general
the crest appears less sharp-pointed than in observations.

7. S h ar p cor ner s at the fr ee su r fa c e

In this section we shall enquire whether the form of the velocity potential pro­
posed in equation (6.1) can lead to the formation of a sharp corner at the interface,
and possibly even to a cusp.
As shown in Longuet-Higgins ( 1980a, paper I), the simplest type of sharp corner
corresponds to a saddle-point in the pressure field. The general condition for a
saddle-point, in terms of the velocity potential #(z, t), was shown in § 6 of paper I
to be
= i (7Л)
1908

392 M. S. Longuet-Higgins

F ig u r e 8. The maximum inclination a mex of the free surface as a function of the time t.
------ , From numerical computation (Longuet-Higgins & Cokelet 1978); upper time-scale.
------ , From analytic model (figure 7); lower time-scale.

(<7 being taken as unity). On substitution for x from (6.1) this gives

l(i/a)-A /< D *)(-i< o*+ V * + A*/w*) + A / (o = 1, ( 7 .2 )

where o) = z* as before. Hence

A = w -J w * (l + iB )(l+ ii? * ) + J (B -i )l/ * , ( 7 .3 )

where
В = A/o ) 2 = A jz . ( 7 .4 )

N o w , if we satisfy the free surface conditions at oo, then b y § 6

A+A*=-V. ( 7 .5 )

On substituting for A from ( 7 .3 ) and simplifying we obtain

(й > + о > *)(1 + В Б *) + Р (й )- а > * )+ Г 7 ](5 + ^ * ) - 1 Г ( Б - 5 * ) = 0. (7.6)

Then writing z = r ew as before, so that

В = — (a/r)e~I(0+e>, ( 7 .7 )

we find that equation ( 7 .6 ) becomes

(a/r)2— 2/?(a/r) + 1 = 0, ( 7 .8 )
where
p = (- t a n £ 0 + \XJr~\ sec \6) cos (0 + e ) - $ V r~ i sec £0sin(0 + e ). (7.9)
 1909

On the overturning of gravity waves 393

Equation (7.8) has two positive real roots

a /r = f i ± (f i 2- l ) i (7.10)
provided that
fi> l. (7.11)

When ft = 1, the two roots are coincident. Given r and в (but not a or e) a necessary
and sufficient condition for (7.11) to be satisfied is that

(tan £ 0 - {7r-*seci0)a+ (i 7 r “*sec£0)* ^ 1. (7.12)

On writing
ri cos \6 =
(7.13)
risin \Q
this becomes
(7.14)

a region whose boundary is a hyperbola in the (£,т;)-р1апе. W e recover the per­

missible region in the z-plane by noting that

z = x + \y = (f+iT/)2+ 2 i^ . (7.16)

The region is sketched in figure 9 in the typical case V = 0 .6 . When V = 0 the

boundary consists of two intersecting parabolas.
To find the angle у contained between the tangents to the free surface at the sharp
corner itself we may make use of the formula

cosy =*ХаХ~/\Ха,Х'!.+Хы\< (7-16)

which was proved in Longuet-Higgins (1980 a). On substituting from equation (6.1)
of the present paper and making use of (7.3) we find

cosy = r\ |1 +LB|7[21«>- B U * + i B (l +LB*)w*|]. (7.17)

The condition for a cusp is that у — 0 and so

c o s y = l. (7.18)

For given values of U and V, one further condition may be imposed on the
solution. It is natural to specify that D p /D t = 0 at the sharp corner itself. Now in
general we have
Dp/Vt = p t+ xivа+ XtP* ’ (7■1

At the sharp comer, p e also vanishes; hence

p t = 0. (7.20)

From the general expression (3.2) for the pressure it follows that we must have

X«Xt + Xtt-P + c-c-= 0' (7-21)

1910

394 M. S. Longuet-Higgins

■*— *y /U 2

x lU 2

F i g u r e 9. Sketch of region in the z-plane allowing real solutions of

equation (7.8). Shaded areas indicate no solution.

Further, since in our model A = 0, we have \tt — 0. Thus (7.21) becomes

(A/a>)(-i<o*+U*+A*/(o*)-#+c.Q. = 0, (7.22)
where f is given in terms of A by equation (6.11).
Altogether, equations (7.10), (7.18) and (7.22) provide three conditions for
determining a/r, в and e, given U and V.
To determine the nature of the solution it is convenient to start with the case when
U = V - 0 (see the appendix) or equivalently when r -* oo but rja remains finite.
Then the boundary (7.14) of the permissible zone of solutions to the first condition
collapses onto the (horizontal) у-axis. Solutions may occupy the upper half-plane
x < 0, but differ radically in the left-hand and right-hand quadrants (see figure 10).
Solutions with values of у less than 60°, including the cusps, for which у = 0, are
confined to the left-hand quadrant (-1 8 0 ° > в > —90°). On the right the only
comer-angles are much blunter. The loci corresponding to equation (7.22) reduce to
two arcs, shown by dotted lines in figure 10. The cusp-locus у = 0 intersects these in
two points, each a possible position of a sharp crest.
To obtain solutions for finite values of r = rc it is convenient to take units so that
*7 = - l , 0=1 (7.23)
1912

396 M. S. Longuet-Higgins

11. Loci of possible sharp comers at the free surface when U = —1. Solid
F ig u r e
curves show admissible solutions. Broken curve shows inadmissible solutions.

and to seek first solutions at large but finite values of r in the neighbourhoods of the
two limiting solutions in figure 10. This was done, by starting with r = 10* and
solving (7.18) and (7.22) by iteration. The value of r was then gradually reduced,
each solution being used as a starting point for the next.
O f the solutions obtained in this way, those starting from the inner of the two
solution points in figure 10 corresponded to high values of $ and were therefore
unsuitable, by the criterion (6.19). Solutions starting from the outer point in
figure 10 corresponded to smaller values of t . Some loci of possible cusps are shown
in figure 11. The corresponding values of

are generally not positive, as is required by equation (6.20). Nevertheless they can
be made small enough to be acceptable.
An example is shown in figure 12, where we have/ = —0.039, which is sufficiently
small that the slight rise in level as t -► —oo for large r may be accepted. The shape of
the forward face, while concave as r oo, is slightly convex at smaller values of r.
 1913

On the overturning o f gravity waves 397

F i g u r e 12. A time-sequence of surface profiles culminating in a cusp.

U = - 1 , V = 0, re = 6.0.

8. D i s c u s s i o n and conclusions

W e have argued that, the flow in a breaking wave being multivalued, we must
expect at least a branch-point in the velocity-potential x • The simplest possible
branch-point is of order J, and we have shown that the three-term expression (6.1)
is capable of describing, at least qualitatively, the most obvious features of plunging
breakers, namely the overturning of the forward face and the apparent forming of
a cusp at the tip.
It should be emphasized that the solutions are valid only approximately, and over
a limited range of the time t. A t the later stages of the flow it will be necessary to
match the solution near the tip of the wave to a locally valid solution such as was
described in Longuet-Higgins ( 19806). The reason for our interest in the cusp-like
solutions of § 7 is that these particular forms, with a small crest-angle, are most likely
to be suitable for matching to the local flow.
Likewise, the earlier stages of the flow, when t is large and negative, must be
matched to the flow in the rest of the wave, in a manner similar to that used for the
almost-highest wave by Longuet-Higgins & Fox (1978), but the time-dependence
also being included.
1914

398 M. S. Longuet-Higgins

Thus we see the present solutions as essentially intermediate, in both space and
time, between expressions describing the wave as a whole, which must take account
of boundary conditions such as finite or decreasing depth of water, and the ultimate
stage of breaking described locally as a jet or sharp comer.
The expression (6.1) may possibly be the first three terms of an asymptotically
convergent series of decreasing powers of 2*. Further terms in the series could
perhaps be determined in the same way in which we have determined A (t ). However,
different forms would be required for non-progressive waves. For example, in a
standing wave, with a vertical plane of symmetry, we should require a solution with
at least two singularities, situated symmetrically on either side of the plane, or
reflecting wall.
Naturally, any of the foregoing profiles may be viewed in a frame of reference
moving with an arbitrary steady velocity U ', so that the branch-point appears to
move in the opposite direction with speed — V . Moreover we can obtain new,
solutions relative to axes with any acceleration a, on replacing the gravitational
acceleration g by g ’ = g — a and referring the new solutions to axes pointing in the
direction of g\

A ppendix. S h arp c o rn e rs: th e case U = V = 0

The following analysis provides starting values for obtaining solutions in the
more general case.
When U = 0, equation (7.9) reduces to

P = —tan \6 cos(0 + e). (A l)

Since p has to satisfy (7.11), there is a critical point where

др/дв = 0, p= 1 (A 2)

(hence a /r = 1). Equations (A 1) and (A 2) lead immediately to

cos0 = i ( l - V 3 ) ; 0 = -111.47°, (A 3 )
and
e — 64.41°. (A 4)

The loci of г/а = (г/а) ew are shown in figure 10. The critical point is designated by
P. W e show the quadrants — 90° > в > — 180° and — 180° > в > — 270°, the others
being obtained by reflection in the origin.
In figure 10 the contours of у are obtained from (7.17) by writing U = 0. However,
the loci for T>p(T>t = 0, obtained from (7.22), simplify very considerably in this
instance. For, since from § 6 A * = —A and equation (7.22) implies that
either
A = 0 (A 5)
or
<u-1( —iw* + A */w*) — —c.c. = 0. (A 6)
 1915

On the overturning of gravity waves 399

Consider the first alternative (A 6). In that case Xt — 0 and the motion must be
steady (clearly an approximation). However, assuming this is so, we have from

(7.1) that

*><
(A 7)

II
hence
\ [}/о > -А /ь Р )(-\ а )* + А */ь)*) = 1» (A 8)
that is
( i - B ) ( - i + 5 * ) = 2ew. (A 9)
The real part gives
1 -B B * = 2 cos в. (A 10)

But from (7.4) we have B B* = a2/r*. Combining this with (A 10) gives

r/a = (1 —2cos0)“*. (A ll)

This is the locus shown by the outer dotted lines in figure 9.

The second alternative (A 6) gives

i ((o * -Зано*+ a>**) + ( A - A * ) = 0, (A 12)

that is
r(2 cos 0 - 3 ) + 2a sine = 0. (A 13)
Therefore
r/a = 2 sin e/(3 —2 cos в). (A 14)

This locus is represented in figure 9 by the two dotted arcs closest to the origin in
each quadrant.
When the free surface corresponding to the above values is plotted as a function of
time, however, it is found that the profiles have unphysical features; either they are
strongly convex on their forward face, or else the limiting profiles with sharp comers
are approached from the outside , as t increases, and not from the inside as desired.
W e have shown in § 7 that this difficulty is overcome by taking non-zero values
of U and V.

This paper was begun in Cambridge and completed during a visit by the author
to the Jet Propulsion Laboratory of the California Institute of Technology,
Pasadena, between November 1979 and February 1980. For the hospitality and
assistance given him there, he is much indebted to Dr О. H. Shemdin and Professor
P. A. Saffman, and members of the Division of Earth and Space Sciences headed
by D r М. T. Chahine.

R e f e r e n c e s

Longuet-Higgins, M. S. 1980a A technique for time-dependent, free-surface flows. Proc. R .

Soc. Lond. A 371, 441-461.
Longuet-Higgins, M. S. 19806 On the forming of sharp comers at a free surface. Proc. R.
Soc. Lond. A 371, 463-478.
1916

400 M. S. Longuet-Higgins
Longuet-Higgins, M. S. & Cokelet, E. D. 1976 The deformation of steep surface waves on
water. I. A numerical method of computation. Proc. R . Soc. Lond. A 350, 1-26.
Longuet-Higgins, M. S. & Cokelet, E. D. 1978 The deformation of steep surface waves on
water. II. Growth of normal-mode instabilities. Proc. R . Soc. Lond. A 364, 1-28.
Longuet-Higgins, M. S. & Fox, M. J. H . 1978 Theory of the almost-highest wave. Part 2.
Matching and asymptotic expansion. J . Fluid Mech. 85, 769-786.
Price, R. K . 1971 The breaking of water waves. J . geophys. Res. 76, 1576-1581.
Stokes, Sir G. G. 1880 Supplement to a paper on the theory of oscillatory waves. Mathematical
and Physical Papers , vol. 1, pp. 314-326. Cambridge University Press.
 1917

J. Fluid Mech. (1982), vol. 121, pp. 403-424

Printed in Great Britain

Parametric solutions for breaking waves

By M . S. L O N G U E T - H I G G I N S
Department of Applied Mathematics and Theoretical Physics,
University of Cambridge, Silver Street, Cambridge, England,
and Institute of Oceanographic Sciences, Wormley, 8urrey

(Received 8 September 1081 and in revised form 16 February 1982)

Time-dependent flows such as occur in breaking surface waves are often most con­
veniently described in parametric form, with the coordinate z and velocity potential
X each expressed in terms of a third complex variable oj and the time t.
In this paper we discuss some interesting flows given in terms of elementary func­
tions of a) and t. Included are the Stokes 120° corner flow, the 46° rotor or rotating
wedge, and a decelerated upwelling flow, with an exactly plane surface.
Lastly it is shown that a class of cubic flows, which are related to the plane upwelling
flow just mentioned, has a free surface that corresponds with remarkable accuracy
to the forward face of an overturning, or plunging, breaker.

1. Introduction
The phenomenon of overturning exhibited by surface gravity waves in both deep
and shallow water still lacks a satisfactory mathematical description. Figures 1 and 2
illustrate two salient features of the flow. First, it must be highly time-dependent,
with particle accelerations comparable to, or even much larger than, g. Mathematical
techniques for dealing with free-surface flows of this kind have been suggested in some
recent papers (Longuet-Higgins 1980a, b, 1981),f particularly in paper I.
The second feature, illustrated more especially by figure 2, is that when the jet
impinges on the forward face of the wave the relatively smooth flow becomes sharply
discontinuous. Mathematically, the flow can be regarded as a multivalued function.
It is as though the colliding particles are attempting to pass onto another sheet of the
Riemann surface. This in turn implies at least a simple branch point in the velocity
potential, located somewhere in the ‘tube’ of the breaking wave.
(We are here interested only in describing the flow up to the moment of impact.
Afterwards, a different type of description will become necessary, involving perhaps
turbulence and entrained flow.)
In conventional descriptions of irrotational flow, the velocity potential x i® usually
expressed directly in terms of the complex coordinate z in the plane of motion, with
the time t as an additional variable. In papers I and II a more flexible scheme was
suggested, in which x an<^ 2 were each expressed as analytic functions of a third
complex variable o>, and also t. As pointed out in § 2 below, this has the advantage
that the simplest kind of zero, where dz/dto vanishes at some point a) — a>0, say,
corresponds to the simplest kind of branch point in the velocity potential x and its

f To be referred to as I, I I and I I I respectively.

1918

404 M . S. Longuet-Higgins

F i g u r e 1 . Profile of a plunging breaker (from Miller 1967).

F i g u r e 2. Tip of a plunging breaker striking the forward face of the wave

(from Miller 1957).

derivatives. Such a parametric description therefore seems possibly very appropriate

to our problem.
Some years ago John (1963) suggested a particular type of parametric representa­
tion in which the parameter и was assumed to be Lagrangian, in the sense of being
constant following a given particle. This leads to certain simplifications (see I). How­
ever, in deriving physical solutions, John’s chief preoccupation was to avoid the
occurrence of zeros in dz/do.;, or at least to ensure that the corresponding singularities
were annulled by zeros in the velocity gradient. In our problem, on the other hand,
the zeros of dz/dto are very much to be desired, provided that they occur fairly close
to the domain of the flow, and outside it.
 1919

Parametric solutions for breaking waves 406

A second feature of John’s analysis was to assume the parameter ы to take only
real values at the free surface. This assumption enables a velocity potential x(w, t)
to be found in the interior, corresponding to a given motion of the free surface. For
the general problem, the assumption is probably too restrictive. Nevertheless, our
purpose here will be to examine whether certain simple flows that do accord with
John’s second assumption (namely that ш is real at the free surface) are of any uee in
describing the overturning of surface waves.
Already some results in this direction have been given in a previous paper (Longuet-
Higgins 1976). But all the flows described there were gravity-free, in the sense that
the acceleration g did not appear in the solution. The flows were appropriate to a
reference frame in free fall, or to the cases where the particle accelerations are large
compared with g. As pointed out in §6 below, gravity-free flows correspond to a
certain linear and homogeneous boundary condition for z(w, t). To include gravity,
we must solve a non-homogeneous equation. But, once a particular integral is
found, further solutions of the homogeneous equation may be added as a kind of
complementary function.
As a first example we present a simple parametric description of the well-known
Stokes corner flow, in which gravity is of course included. An analogous flow is the
‘ 45° rotor ’ discovered in paper II, for which a parametric representation is given in § 7.
We then pass on to discuss an interesting type of gravity flow (§8). This can be
described as a ‘ decelerated upwelling ’, in which the pressure gradient from the surface
velocities is exactly balanced by that due to the deceleration. Associated with this
is a family of flows given as simple polynomials in w. Remarkably, it appears that
one of the cubic polynomials agrees very closely with the observed profiles of over­
turning waves.
Further discussion is given in § 14.

2. Parametric representation; branch points

We consider incompressible, irrotational flow in two dimensions, where the velocity
potential x = Ф+ ifr is general an analytic function of the position coordinate
z = x + iy and of the time t. Instead of expressing x directly as a function of z and t
we assume that x and 2 are each analytic functions of a third complex variable ш,
and of t. Thus by eliminating w between x a^d 2 we could if we wished express x directly
as an analytic function of 2, except at a singularity.
Consider the flow in the neighbourhood of a branch point. If suffixes are used to
denote partial differentiation, with respect to w or t, then the particle velocity is given
by W*. where W = XJ z „ ( 2 . 1)

and an asterisk denotes the conjugate complex quantity. Suppose that гыhas a simple
zero at a) = (i)Q, where 2 = z0. Then in the neighbourhood of this point we have

z ~ zo ~ i ( w“ wo)22W (2-2)
Hence (o -(J 0cc ( z - z 0)i, (2.3)

and if x is a regular and analytic function of с0 with хЫ9 + ® then

X - X v ~ (w - w 0)X«,oc ( z - * „ ) i (2.4)
1920

406 M . S. Longuet-Higgins

also. This represents the simplest type of branch-point singularity. The velocity x*
is 0(z —z0)* and so less singular than in a source of vortex, where it is 0{z —z0)”1.
If, as in § 4, the velocity W * is finite at o> = w0, as well as the potential x> then by
a similar argument W — W0 is generally 0(z —z0)* and x ~ Xo *s O(z —z0)^.

3. Boundary conditions
General expressions for the pressure p and its rate of change D p /D t following a
particle were given in paper I. Thus if g denotes gravity, and if the ж-axis is taken
pointing downwards, from Bernoulli’s equation we have

- 2 p = (xt- Wzt) + c.c.+ W W * -g (z + z * ) - 2 f , (3.1)

in which / is a function of t only, and ‘ c.c.’ denotes the complex conjugate of the
preceding terms. Also

- 2g = [ ( x « - Wztt) + 2K (3 U - Wzu ) + К Ч Х ш ~ * 0 1 + -g W + W )-2 f „

(3.2)

where к=— = (3.3)

Dt z„

At a (moving) free surface, the boundary conditions are that

p = 0, D p /D t = 0. (3.4)

In the special case when w is a Lagrangian coordinate, constant following a particle,

then clearly 2( = W *. К = 0, (3.6)
and so (3.2) reduces to

- 2^ = ( * „ - Wzu) + C.C.- g (W + W * )-2 f . (3.6)

Some solutions making use of the equations (3.1) and (3.2) in this general form were
given in papers I I and III.

4. M ethod o f John
The above equations may be considered as a generalization of the approach of
John (1953), who assumed, in addition to the flow being Lagrangian, that the para­
meter to was real at the free surface. So at the free surface we have both

W* = ztH (4.1)

and x j z . = W = zf(w *) = z ? H (4.2)

since со = a)*. Therefore (at the free surface)

# « - * » , (« ) * ? ( « ). (4-3)
and a flow satisfying at least the kinematic boundary condition may be found by
integrating with respect to (o:

{a))zf{(o)d<D. (4.4)
 1921

Parametric solutions for breaking waves 407

This is always provided that \ has no singularities, including zeros of гы unaccom­

panied by zeros of Wu, within the domain of the fluid. On the contrary, branch points
outside the fluid may be welcome, for reasons given in § 1.
In John’s formulation the dynamical boundary condition becomes especially
simple for, since cd is Lagrangian at the surface, the particle acceleration there is
simply ztt. Hence the pressure gradient is zu —g, which must be normal to the free
surface. But, since ы is real, the tangent to the surface is in the direction of ztt. So we
must have
zit- g = irz„ (4.5)

where r is some function of o) and t which is real on the boundary.

In the interior of the fluid, however, w is not necessarily Lagrangian. There, by
virtue of (4.4), we have
jt . -*«(< »)*? (») (*-e>
(see (4.3)), and so the particle velocity is given by

W = ( X j z j * = zt(u>*)t (4.7)

not zt(cj), as it would be if и were Lagrangian everywhere. Only at the free surface,
where (o = oj*, is the velocity given by zt[aj).
For this reason we shall need the more general expressions forp and D p /D t given
by (3.1) and (3.2). Making use of (4.7) these may be rewritten more explicitly as follows.
First

~2 p = (^t + ^ f)+ 2 t(w *)2 ? H -^ (2 -| -2 *)-2/~ [2 tH 2 f(w )+ 2 t(w#)2f(w*)]. (4.8)

It will be noted that this is a function of со, w* and t. I f F[u), (o*} t) is any such function,
it is clear that
^ = Ft + K F a+ K *F u., (4.9)

where К = D cj/D t is given by (3.3). In the present case, since W* = 2t(w*), we have

К = Ы а > * )-г ь{а>))/гы. (4.10)

With F =* - 2 p and К given by (4.10), wc find in fact that (4.9) reduces to

" 2 75jf =■ KXtt- Wztt) + 2Kz„ztt(a>) + K h „ z t M ] + c . c . -g (W + W * ) - 2 /,. (4.11)

But from (4.6) Xut ~ Wzut = 2«(ш) z„, (4.12)

= <4' 13>
Hence (4.11) is equivalent to (3.2).
Lastly we note that, though the particle acceleration a at the surface, in John s
formulation, is simply equal to z^t in the interior it must be calculated from the more
general formula
a = ~ z t(w*) = ztt(w*) + K * z tu.{(o *), (4.14)

К * being the conjugate o f (4.10).

1922

408 M . S. Longuet-Higgins

5. Frames o f reference
Given a suitable function r(w,*), real when w is real, the boundary condition

ztt~9 — irzw (5.1)

may be regarded as a non-homogeneous linear differential equation for z(k),t). Let
z0(w, t) be any solution of (5.1). Then it is clear that Zq+ z^ w, t) is also a solution,
provided only that z± satisfies the homogeneous equation

ztt = irzw (5.2)

with the same integrating function r. We may call zQa particular integral of the equa­
tion (5.1). Then zlt in general form, is the complementary function, and z0+ zx the
complete integral.
Since (5.2) can be derived from (5.1) simply by setting g = 0, (5.2) can be regarded
also as a gravity-free form of (5.1), and we may think of solutions to (5.2) as free-fall
solutions, that is solutions to (5.1) but seen in a reference frame moving with downward
acceleration g.
Thus, if z0(w,J) is any solution of (5.1), then

— Zq — \gt? (5.3)

is a solution to (5.2), and conversely. It is easy to verify from (4.8) that the pressure
p(u)} (o*,t) remains unaltered, apart from an additive function of the time t. The
corresponding velocities, however, will differ by the amount^, which may be important
in matching inner flows to an outer flow as t ^ —oo.
In other respects the distinction between a solution to (5.1) and (5.2) need not be
important, particularly in situations where the particle accelerations are large com­
pared with g} or are highly variable. In § 6 we shall examine a flow which in one
reference frame, including gravity, is steady, but in another, free-falling frame is an
expanding, self-similar flow.

6. The Stokes corner flow

As a first application consider the familiar flow proposed by Stokes (1880) to
describe the limiting form of a progressive wave crest, seen by an observer moving
horizontally with the phase speed. Take the origin О at the wave crest and the axis
of &t[z) pointing vertically downwards as in figure 3(a). (The wave is moving to the
right.) A particle on the forward face of the wave travels up the slope with deceleration
\gt the inclination to the vertical being Jn or arccosj. If each particle is labelled by
the time w at which it reaches the wave crest, then the position z(w, t) of the particle
(i) at time t < w is given by
z = \g e'”l* (o) —t)2. (6.1)

(A characteristic of progressive motion is that z is a function of w —t only.)

To verify the free-surface condition, note that

- igeil,l3{< o -t) (6.2)

and ztt~~9 — Ы ^ я/3“ 2) = У М д ^ , (6.3)
 1923

Parametric solutions for breaking waves 409

z-planc

<*)-planc
(b)

О
j ------------■*----------------- V 60° ~7777777t 7777777777*
\

\
F i g u r e 3. Parametric representation of the Stokes comer flow
(a) in the z-plane, (6) in the o>-plane.

Hence (6.1) is satisfied if J3 ....

r= (6-4>
which is real when ш is real. Moreover
*?(«) = - \g<rivl\ a > -t ), (6.6)

so from (4.3) x« = (6 6)
Integration with respect to a) gives

X = ~ iV?2(w “ = I (6-7)
the well-known expression for the velocity potential in a Stokes corner flow.
Paradoxically, when t > ы the particles in the free surface, according to (6.1),
reverse their direction and slide again down the forward face of the wave, instead of
down the rear face as might be expected. The paradox is resolved by considering the
situation in the w-plane (figure 36). There it is seen that the forward face of the wave
(O A in figure 3a) corresponds to the positive real axis of со—t, but the rear face OB
corresponds to the line arg(o> —t) = The point to —t = 0 is a branch point of z,
and when и —t < 0 the particles go onto another sheet of the Riemann surface.
1924

410 M . S. Longuet-Higgins

All the particles in the interior of the fluid correspond to S{(o) < 0), and since o) is
not real the representation there is not Lagrangian. Thus the particles do not have to
follow a path on which o) is constant.
It remains to verify that the two boundary conditions (3.4) are in fact satisfied on
OB. This would follow from the symmetry of the expression (6.9) for but from (6.1)
it is less obvious. However, direct substitution into (4.8) gives us

2p = i^ (w - ш*) [(w - 1) eiwli - (со* - 1) e - W ] , (6.8)

which vanishes both when о) — о)* (that is on ОA , ОA'), and when

(w -O e '*'3 = (o > *-r)e -^ 8 (6.9)

(that is on O B ).
Clearly if z(a>, <) is a polynomial in o), then the expression forp(o>, oj*, t) will, by (4.8),
be a polynomial in oj and oj*, with a factor oj —oj*, so we can write in general

^ = (oj —oj*)G(oj,oj*,t), (6.Ю)

9
and the free surface, apart from oj = oj*, will be given by О = 0. To show that the
second condition D p /D t = 0 is satisfied on this surface, it is sufficient to show that
D O /D t = 0 when G = 0.
In the present case
О = (oj — t) einl3—(oj* —t) е~*”1*, (6.11)

and a direct calculation using (4.9) with F = 0 gives

= - M ( w - i ) ae8i" 3- ( " - < ) 8« - 2i" s]. (6-12)

Since О is a factor of the right-hand side of (6.12), it follows that D O /D t vanishes on

O B as required.
W e remark that, if № denotes the steady flow (6.1), then
= (6.13)

represents a gravity-free flow satisfying the homogeneous boundary condition

= irzw (6.14)
with r given by (6.4). Moreover
z<2>= ±geinizS, (6.16)

where S = (o) — t)‘l — 2 е~и1йt2

=* oj2 - 2a)t + 3it2

= t2[{o>/t)2-2 (a )/t ) + p i ] . (6.16)

Thus z® represents a self-similar flow, expanding with time proportionally to t2.

Higher-order polynomial expressions also exist satisfying (6.16) with the same
function as in (6.4). For example, the cubic

г = w3- | ( l + i^j3i)to)2+ % j3it2o) + ltz, (6.17)

for which the contours p = 0 are shown in figure 4. One branch, other than the real
axis of oj , passes through the point oj/t = 1. The branch points, given by the vanishing
 1925

Parametric solutions for breaking waves 411

F i g u r e 4. Loci of p = 0 corresponding to the flow given by equation (7.8)

in the (j-plane.

of zu, are at w / t = 1 and | 3i. Hence a possible domain of flow would appear to be the
shaded area shown in figure 4. However, on further investigation it appears that the
second boundary condition D p / D t == 0 is not satisfied on any of the contours p = 0
other than the real axis. Hence (6.17), and similar expressions of higher order, do not
represent complete solutions.

7. The 45° rotor

An interesting exact solution of the homogeneous equation (6.2) is the expression

z = we^1+^ +<‘, (7*1)

where e is a constant phase, to be determined. We have clearly

ztt =* 2iwzut (7.2)

so that z satisfies (5.2) with r * 2to. Moreover from (4.3)

Xы= - ( l - i ) ^ e - * , (7.3)

hence ^
* - V2 '• ( ’
Choosing e «= f 7Г, we may write
ut
X- ф * <7-6>

which identifies the flow with the 46° ‘rotating hyperbola’ discovered in paper II.
The flow is sketched in figure 5 (a).
I4 ILM 121
1926

412 M . S . Longuet-Higgins

(a) (ь)

F i g u r e 6. The 45° rotating corner flow: (a) equation (7.1); (6) equation (7.10).
In each case the flow is extended by reflection in the origin O.

Direct calculation using (4.8) gives for the pressure field

— 2p = (w —w*) [(1 *f i)w —(1 —i)(j>*]e~u . (7.6)

So, apart from the real axis w = to* there is a second free surface p = 0 given by
arg (i) — — \ t i . Setting
О s [ ( l + i ) w - ( l - t ) w * ] e - w, (7.7)
we find also from (4.9) that
(7.8)

Hence the second boundary condition is also satisfied on О = 0.

W e see from (7.1) that

W * — zt(w *) = —(l-ft )w * e -(1+<)t+ic. (7.9)

For points on the trailing face of the wedge ( oj = <o*) we have zt — —(1 + i)z . Hence
the particles spiral inwards towards the centre along 46° spirals.
For points on the other face, arg w = - { n , we have o)*/o) = t, hence ze - (1 —i )z
and the particles spiral outwards.
The single-term expression (7.1) represents only the special case of the ‘ 45° rotor’
when the boundary has a sharp corner. By adding to (7.1) a second term, so that
z = s ( i ) e -{l+ i)t+ M n l9 + w - i eu-ot+<*/e (7 .1 0 )

we obtain a more general flow in which the free surface is a 46° hyperbola, with rounded
vertex, rotating about its centre 2 = 0 (see figure 6b). The potential x (v, 0 satisfies
,at
(7.11)

and the equation of the free surface is

1гегигг +с.с. +Лгг* = - 1, (7.12)
 1927

Parametric solvXions for breaking waves 413

a special case of paper II, equation (4.1). In the present formulation z((o,t) has a
branch point (гы- 0) when
o) = ee-iwe, 2 = 2e-«+<*/4f (7j3)
which lies at a focus of the hyperbola, that is in the interior of the flow. But this is
also a branch point of the velocity zt{a)*), so that altogether the flow is non-singular.
The flow just described is itself a highly special case of the class of hyperbolic flows,
with variable angle between the asymptotes, suggested in II as representing the tip
of a plunging breaker. However, their parametric representation would in general
seem to involve transcendental functions of t.

8. U pw elling flows
We now discuss a different, and in some ways simpler, class of flows than the two
types just described. These may be called ‘upwelling flows’ for reasons that will
become apparent.
We first choose a very simple form of (5.1). In fact let us take r = —2/1, independent
of a), and to save writing set . /Q 44
ь ш = w (8.1)

so that the non-homogeneous boundary condition (6.1) becomes

\t(zH- g ) = z„. (8.2)

This must be satisfied when w is pure imaginary, that is when ш + w* = 0.

Clearly (8.2) is satisfied by the expression
z --\ g t w . (8.3)
Moreover zm = - ±gt, (8.4)
W = z t { - w ) = \gw, (8.5)

so from (4.6) Xw ~ (8,e)

giving x = ~ = - 2(•

This flow is shown in figure 6. The free surface @ (w ) = 0 is the jr-axis, x = 0. The
velocity potential is
( 8. 8)
r 21
and the streamlines are yjr = —^ = constant, (8.9)

which are rectangular hyperbolae. For t > 0 the flow represents a decelerated up­
welling, in which the vertical and horizontal components of flow are given by

f. f <8-10>
respectively, the axis of x being vertically downwards.
To verify that horizontal surface x = 0 is indeed a surface of constant pressure,
note that the horizontal pressure gradient is in general given by
—Py = vt + {uvx + w y), (8.11)
x+-a
1928

X
Fioube в. Streamlinee of the upwelling flow (8.7).

where (u,v) = {фх,фу), From (8.10) the terms on the right of (8.11) are equal to
— y/t*, 0 and y jt 2 respectively. Hence p v = 0 everywhere. Alternatively we may
remark that in the Bernoulli equation

- V = Фь + МФ1 + Ф 1 )-9 х (8Л2)

the acceleration term фь is exactly balanced by the usual pressure defect - J(V0)2.
For t < 0, фх becomes positive (when x > 0), and instead of a decelerated upwelling
we have an aocelerated downwelling.

9. Complementary upwelling flows

We now seek flows complementary to the upwelling flow described in § 8. That is
to say, we seek solutions to the homogeneous boundary condition

itztt (9.1)
to be satisfied when w + w* = 0.
Let us try the polynomial expression

* = Ьпшл 4-Ьп_1шп- 1+ ... + Ь0» (e*2)

the coefficients bn being functions of the time t only. Substituting into (9.1), we obtain

V K = 0,

i tbn- i - nbn,

( 9 '3 )

)
Clearly bn must be of the form At + B, with A and В constant. Since the equations
(9.3) are linear, the simplest solutions of (9.2) can be separated into two fundamental
 1929

Parametric solutions for breaking waves 416

classes, those in which A = 1, В = 0, and those in which A = 0, В = 1. Denoting these

by Pn and Qn respectively, wo find by straightforward integration, as far as n = 3,

P0 = t,

Pi - tw + t2,
(9.4)
P2 — trn2+ 2t*w 4- §J3,

P 3= tw* -f 3thn24- 2t3m +

and Q0 ш 1,

Qx - w + 2t\n |*|,
(9.6)
Q2 = w2+ 4twIn |t[ -f 4J2(ln |J|- f),

Q3 = ш8+ 6^ш21п |<|+ 12<*ш(1п |/|- f ) + 4/3(ln |<|-y ).

The complete polynomial solution will be of the form

z = Z ( A nPn + B nQn). (9.6)

n

Each of the flows Pn is self-similar, as can be seen by writing it in the form

Рл(ш,0 = /п+1Д (ш Д 1). (9.7)

In fact these flows are limiting cases of the self-similar flows described in §9 of Longuet-
Higgins (1976), in the limit as Л -> 1.
On the other hand, the flows represented by Qn are not self-similar, nor is any com­
bination of the Pn and Qn involving two or more non-zero coefficients A n or B n.

10. Discussion: linear flows

To interpret the solutions physically, we note first that all expressions linear in w
must correspond to a free-surface profile that is a straight line. Similarly, any expres­
sion that is quadratic in w has a surface profile that ie a parabola, whose orientation
and dimensions may vary in time. Every solution of the third degree in w corresponds
to a parametric cubic, and so on.
As an example consider the linear flow Pv We have then
z ~ t w + t2y (Ю.1)
so = t, zt = m + 2t. (10.2)

To find the velocity potential % we note that on the free surface w* = —w, so the
general boundary condition
Хш = * ? (- « 0 z e(w) (10.3)
gives хда = - -f 2t2} (Ю.4)

hence X = ~ itw2+ 2tw (10.6)

to within a function of the time t. The pressure p is now found from the equivalent of
(4.8), namely
- 2 p = (Xt + X?) + 2 ? (-w )z t(-T i7 *)--2 / -[zt(lZ7)z*(-tI7) + Z?(TIT*)ze(--n7*)]. (10.6)
1930

416 M . S . Longuet-Higgins

With suitable choice of/ this gives

- 2 p = (TU+ TZ7*) [ £ ( ш + п7 *) + 2<]. (10.7)

The locus of p = 0 is therefore

td + v j* — 0, 2 + 2* = 2f2, (1 0 .8 )

together with the parallel line

G7+ ra* = - 4£, z + z* = -2 t* . (10.9)

To calculate D p /D t we use the general formula analogous to (4.9), namely

2 £ = Ft + K 'F w+ K '* F ,. , (10.10)

where F = -2 p a n d K ' = [zt( - w * ) - z t(w )]/z w. (10.11)

Hence ^ = (ш + ш*) [(ш + шф)//+ 1]. (10.12)

This vanishes on the free surface (10.8) but not on the parallel surface (10.9). In other
words, (10.9) represents a surface that is no* moving with the fluid particles.
Clearly the non-homogeneous flow (8.3) is a special case of the flow Px when seen
in an accelerated frame of reference.

11. Quadratic flows

The second-order flows (n » 2) are the simplest to have a branch point (zw = 0)
in the finite part of the plane. This occurs always at the focus of the parabola. Thus for
P 2 we have
zw = 2tm + 2<2, (И -1)

which vanishes when m = - t , z - - \t3. The particle velocity W * is given by

W = zf{ —m) = w2—4tru + 2£2, (11.2)

and since Ww = 2тя- 41, which does not vanish at the branch point, this is a genuine
singularity. The flow must therefore be taken in some domain excluding the focus,
i.e. outside the parabola.
The corresponding velocity potential is found to be

X — i tvjA— 2J%* —2t*w2+ 4(4ш3, (11.3)

and the pressure field is given by

2p — (m + tet*) [ \( td —w *) (то2—тп**) + 4/(ro2+ w *2) —8/2(ш + ш*) — 8*3]. (11.4)

The loci of p = 0 in the planes of w and z are shown in figure 7. On the branches of
the locus other than w + w* = 0 it is found that D p /D t does not vanish, nor ie it a
function of t only, so these do not correspond to free surfaces.
Similar remarks apply to the flow z =• Qz.
 1931

417

F i g u r e 7. The quadratic flow г = P t. (a) Free surface in the z-plane;

(6) contoure of p = 0 in the 6>-plane.

12. Cubic flows

The case n = 3 is of special interest as being the simplest case for which the free
surface can intersect itself.
Consider the flow P3. We have
z = tw * + 3 * % * + 2 * 3ш + $ f 4. (1 2 .1 )
1932

418 M . S . Longuet-Higgins

i=P>

(b)
F i g u r e 8. The cubic upwelling z = P ,. (a) Free surface in the г-plane;
(6) contours o f p = 0 and branch points in the to-plane.

A t the free surface, w being imaginary, we may write w/t = ifi} ц real, so on separating
real and imaginary parts in (12.1),

*/«< = _ 3/t*+ J, | (12.2)

y/t* = -ft ? + 2fi.)

The profile is shown in figure 8(a). I t intersects the ж-axis when ft = 0 and ±<J%,
that is in the points z/t* = $ and —-у- respectively, the latter being a double point on
the curve. Stationary values o f у occur when ft1 = § and z/i4 = - f i V t ? * *
 1933

Parametric solutions for breaking waves 419

aspect latio o f the loop, that is to say the width of the loop at its widest section
divided by the distance between the vertex and the node, is

= 0-3629. (12.3)
The maximum width is one-third of the way from the vertex to the node.
The branch points o f z are given by the vanishing of

za = 3tm2-+- 6/2o7-h 2t3, (12.4)

that is ®/< = - 1 ± V i (12.6)

80 *A* = J(1±2VJ). (12.6)
These are marked В and B ' in figure 8 (a). In figure 8 (i) В lies very close to the origin
and B ' is not shown because it lies on another sheet of the Riemann surface.
The velocity potential corresponding to (12.1) is found to be

X~ ~ + 4*3Ш4- - 2t*W2+ | / в0 7 , (12.7)

and the pressure is given by

2p = [ — £(тпв-f П73ш *3) — %t(4zn5 — 5 m h n * 2) + Qt2(2m* —tu3vj* — 3 w 2w * 2)

+ 4*3(7ш* + S w h D *)- Ш 4(ш* + ш ш *)-8*5ш] + с.с. (12.8)

13. Comparison with observation

In figure 9 we compare the observed surface profile o f figure 1 with the curve of
equation (12.2). The agreement with the forward face o f the wave is remarkable.
A second comparison can be made with the recent numerical calculations of breaking
waves by Vinje & Brevig (1981), shown in figure 10. Inserted in the figure are straight
lines, each representing the axis o f symmetry o f a cubic curve fitted to the lower part
o f the profile. Table 1 shows the lengths o f the corresponding axes; 2a is the distance
between the two points where the curve intersects the axis, and 26 is twice the maxi­
mum distance of the lower part of the profile from the axis. The ‘ aspect ratios’ 6/a
are given in the third column of table 1, and will be seen to be close to the theoretical
ratio (12.3).
In figure 10 the orientation o f the axes is nearly constant; the angle o f inclination
o f the horizontal varies only between about 37° and 47°. The length 26 o f the minor
axis is shown as a function o f the time in figure 11. The plotted points have been fitted
with a curve of the form C(t —10)4, where t0 > 0. I f this were followed precisely it would
indicate that the dimensions o f the ‘ tube’ tended to zero as t -> t0.
The curve o f figure 8 may indeed be typical o f plunging breakers. I t was remarked
by New (1981) that profiles were very often fitted quite accurately by an ellipse with
axes in the ratio ^/3:1. In figure 12 we show a comparison between the cubic curve of
figure 8(a) and a ‘ ^3-ellipse’ having the same minimum radius o f curvature. Again
the agreement is close.
The only part of the free surface desoribed so far is the forward face o f the wave.
However, the complete locus p = 0 (see (12.8)) contains another branch within the
fluid, labelled I I in figure 13(a). On this branch the second boundary condition
D p /D t = 0 is not generally satisfied (as it is on I). Moreover the curve is symmetric
1934

420 M . S. Longuet-Higgins

F i g u r e 9. Comparison of г =* А г Рл with the observed profile in figure 1.

0-6

£
02

05 0-9 t"«f 1-7 21

F ig u r e 10. Successive profiles of a wave breaking in deep water

(after Vinje & Brevig 1981, figure 4.12).

about the real axis o f z. Nevertheless, since the pressure p has a stationary point
(saddle point) on the real axis between I and n, it appears that only a slight asymmetric
perturbation o f the flow would be sufficient to alter the free surface to the form hi,
corresponding very nearly to the surface o f a plunging breaker. Near the tip o f the jet,
the perturbed flow may locally take the fqrm described in paper I I.
Figure 13 (b) shows the same flow z = P 3but at a different time t *= 0*5. Apart from
the perturbation, which is seen to be a small part of the total flow, the surfaces in
figures 13 (a) and 13 (6) are precisely similar.
 1935

Parametric solutions for breaking waves 421

t 2a (mm) 26 (mm) b/a в

6 840 280 0-41 37°
7 656 210 0-38 38°
8 620 19*0 0-37 39°
9 416 150 0-30 40°
10 40-5 140 0-36 41°
11 36-5 12-5 0*34 43°
12 326 11-2 0-36 46°
13 295 10-5 0-38 47°

T a b l e 1. Measured parameters of deep-water wave orests,

from Vinje <k Brevig (1981, figure 4.12)

о Vinje &. Brevig ( 1981)

30

----- 26=C(r-/o)4

20

«5ч_

10

10 IS 20

Time t (arbitrary units)

F ig u re 1 1 . Time variation of the length of the minor axis in figure 10.

In the expression for the pressure p the terms of highest order in w are given by

2 p = J(wf — w *3), + 6)(zi7s) (13.1)

so that one of the asymptotes is argw = corresponding to a surface inclined

asymptotically at 120° to the forward face of the wave.
The second solutions Qn will not be discussed in detail, except to remark that @3
has one branch point on either side of the free surface ш - w* whenever In |f| < }.
Thus z = Qz, unlike z = P 3, is not a physically possible flow. However, it may be worth
noting that over limited intervals of time the expression

z — P 3+ iijQs (13.2)
1936

422 M . S. Longuet-Higgins

v/3-ellipse

F i g u r e 12. Comparison of the cubic curve z = P , given by (12.2)

and an ellipse having axes in the ratio 3:1.

F i o u r e 13. Contours of the pressure for г « P , in the physical plane:

(a) when t = 1 ; (b) when I = 0*5.
 Parametric solutions for breaking waves 423

Figure 14. Successive profiles of the free surface for the flow z = P, + 0‘ltQ,.

where tj is a real, positive constant, < 1, represents a flow in which the asymptotes
rotate clock wise, as does the nodal loop (see figure 14). This agrees qualitatively with
the numerical profiles shown in figure 10.

14. Conclusion
W e have shown how a parametric description o f the fluid flow, involving only
elementary functions o f the parameter o>, can describe some outstanding features of an
overturning gravity wave, particularly the asymptotic form of the front face of the
wave. W e have confirmed the suggestion in I I I that the analytic description probably
involves a branch point. We have also described the flow near the tip o f the ‘ je t ’, in
a special case.
A complete solution should ideally give an account o f the whole development of
the flow, including the rear face o f the wave. There is reason to believe that this may
be possible by combining or perturbing the solutions given in the present paper.

REFERENCES
J o h n , F. 1963 Two-dimensional potential flows with a free boundary. Commune Pure Appl.
Math. 6, 497-503.
L o n o u e t -H ig o in 8, M. S. 1976 Self-similar, time-dependent flows with a free surface. J. Fluid
Mech. 73, 603-620.
L o n g u e t - H i g g i n s , M. S. 1980a (Paper I) A technique for time-dependent, free-surface flows.
Proo. R. Soc. Lond. A 371, 441-461.
L o n g u e t - H i o o i n s , M. S. 19806 (Paper II) On the forming of sharp comers at a free surface.
Proc. R. Soc. Lond. A 371, 463-478.
L o n g u e t - H i g g i n s , M. S. 1981 (Paper II I ) On the overturning of gravity waves. Proc. R. Soc.
Lond. A 376, 377-400.
M i l l e r , R. L. 1967 Role of vortices in surf zone prediction, sedimentation and wave forces.
In Beach and Nearshore Sedimentation (ed. R. A. Davis & R. L. Ethington), pp. 92-114.
Tulsa, Oklahoma: Soc. of Economic Palaeontologists and Mineralogists.
424 M . S. Longuet-Higgins

N e w , A. L. 1981 Breaking waves in water of finite depth, British Thtor. Mech. Colloq., Bradford,
6-9 April 1981 (abstract only).
Stokes, G. G. 1880 On the theory of oscillatory waves. Appendix B : Considerations relative to
the greatest height of oscillatory waves which can be propagated without change of form.
Mathematical Physical Papers, vol. I , p p . 225-228. Cambridge University Press.
V in je, T. Sc B re v io , P. 1981 Breaking waves on finite water depths: a numerical study. Ship
Res. Inst, of Norway, Rep. R - 111.81.
 1939

J. Fluid Mech. (1983). vol. 127, pp. 103-121 103

Printed in Great Britain

Bubbles, breaking waves and hyperbolic jets

at a free surface
By M . S. L O N G U E T - H I G G I N S
D ep a rtm en t o f A p p lied M athem atics and Theoretical Physics, S ilver Street, Cambridge,
and In stitu te o f O ceanographic Sciences, W orm ley, Surrey

(R eceived 7 May 1982)

Experiments have shown that bubbles approaching an air-water interface give rise
to axisymmetric jets projected upwards into the air. Similar jets occur during the
collapse o f cavitation bubbles near a solid surface. In this paper we show that such
jets are well modelled by a Dirichlet hyperboloid, a hyperbolic form o f the
better-known ellipsoid. The vertex angle o f the hyperboloid is calculated as a function
o f time and found to agree with the observations o f Blake & Gibson (1981) and others.
The jet is initiated, according to this model, when the vertex angle passes through
2 arctan\/2, or 109-47°, at which instant the fluid accelerations become large. This
compares with a vertical angle of 90° in the corresponding two-dimensional flow.
Further experiments demonstrate that an axisymmetric standing wave, when
driven beyond its maximum amplitude, can break by throwing up a jet of the same
hyperbolic form. Hyperbolic jets may occur commonly in free-surface flows.

1. Introduction
In a recent paper Blake & Gibson (1981) studied the growth and collapse of vapour
bubbles close to a free surface, both numerically and experimentally. One o f their
experiments is reproduced, in modified form, in figure 1. As the bubble rises and
expands it can be seen to raise and accelerate a thin layer o f fluid in the form of a
spherical dome. Then (starting at about frame no. 30) there emerges a remarkable
jet rising from the dome along the axis of symmetry. The jet ultimately approximates
a circular cone with diminishing vertex angle. Such a phenomenon is clearly of
environmental interest for transfer o f water droplets, salt nuclei and organic particles
from the-ocean to the atmosphere by bursting bubbles (see Blanchard & Woodcock
1980). As indicated by Blake & Gibson, similar jets issuing from the interior surface
o f a cavitation bubble near a solid surface, and directed towards the surface, are
probably responsible for much of the damage to propellers and other fast-driven
hydraulic machinery.
The numerical calculations described by Blake & Gibson agree with their observa­
tions as far as the initial formation o f the jet. However, the mechanism still appears
somewhat mysterious. In the present paper we propose an analytic model for the
development of the jet. The model is in fact a hyperbolic form o f the flow known in
astrophysics as a Dirichlet ellipsoid, and described for example in Lamb (1932). The
hyperbolic form, however, does not appear to have been seriously considered, perhaps
on account o f the infinite masses involved, and certain singularities in the time
dependence, described below. A two-dimensional hyperbolic form has been studied by
Longuet-Higgins (1972, 1976) in connection with the breaking o f surface waves, and
generalized in more recent papers (1980, 1982).
1940

104 M . S. Longuet-Higgins

Frame no. 10

F i g u r e 1. J e t produced b y a gas bubble approaching a free surface (a fte r B la k e & G ibson 1981).
T h e dashed line represents the tra je cto ry o f the assumed centre o f the h yp erb oloid.

The theory for the three-dimensional hyperboloid is given below in §2, and the case
o f axial symmetry is discussed in detail in §3. One special feature o f the flow is the
occurrence o f a ‘ jo lt’ , or local infinity in the acceleration when the vertex angle o f
the hyperboloid equals 2 arctan\/2, or 109*5°. This in fact corresponds to the initiation
o f the jet. A comparison with Blake & Gibson’s observations follows in §4, with
reasonable agreement (see figure 4).
The degree of correspondence between the observations and this simple theoretical
model prompted the author to enquire whether a similar flow might be observed in
other situations. One example that came to mind was the jet thrown up by a standing
water wave when excited by a vertical or horizontal oscillation of the container
(Faraday 1831). In §6 we describe a simple experiment which in effect verifies that
these jets are, in suitable circumstances, well described by the same theory.
In the appendix we enquire further into the initial ‘ jo lt’ and give a physical
explanation for it. Finally, in §7 we summarize the conclusions and note possible
implications for the modelling of a breaking wave.
 1941

Bubbles, breaking waves and hyperbolic jets at a free surface 105

2. The Dirichlet hyperboloids

The theory for Dirichlets’ ellipsoids (see Dirichlet 1860) is given in Lamb (1932,
§382), where it is generalized to cover both self-gravitation and rotation of the fluid
as a whole. For convenience we give here an alternative derivation for the hyperboloids
in the simplest situation, that is without rotation or self-gravitation.
The fluid is assumed inviscid and incompressible and the flow irrotational. The
velocity potential, relative to rectangular coordinates (x,y,z) is taken to be of the
form i /• * • \
Ф = 2 & + Ь У ' + 1 Л ( 2 1 )

where a, 6, с are functions of the time t only, and a dot (") denotes time-differentiation.
Equation (2.1) represents, in fact, the most general expression for a pure straining
flow that is symmetric in regard to the three coordinate planes. The velocity vector

!* !* )• (2-2)
and by continuity we have a b с __
J J - + т + - = 0, (2.3)
a b с
so that abc = 2L 3, (2.4)
is a constant o f the motion.
The fluid being assumed homogeneous and o f unit density and gravity negligible,!
the pressure p at any point is given by

— 2p = 2<j>t + (V<t>)* + F( t) = (^ x г+ j!y a+ ^ г 2) + ^,. (2.5)

H РПРР О Io n

Surface tension being also neglected, we may take as boundary conditions that both
p and D p /D t vanish on the same surface (cf. Longuet-Higgins 1972, 1976). Thus the
coefficients o f corresponding terms in (2.5) and (2.6) must be in proportion. Hence

* + 1 * + * £ + « ' (2 .7 )
a a b b с с г

and on integration ad = K XF, bb = К г F, cc = K 3F, (2.8)

where the are constants. A change in scale o f a, 6 or с does not alter the velocity
potential ф, so without loss of generality we may take K x = — 1, K 2 = К л = 1. Then
from (2.8)
aa%= —bb = —cc = — F t (2.9)

and by (2.5) the free surface p = 0 becomes simply

a2 b2 c2 (2Л° )

a hyperboloid with semi-axes a, b and c. Moreover, the continuity condition (2.3) can
be written aa — bb—cc = 0, (2.11)

t A lte rn a tiv e ly , the flow m ay be view ed in a free-fall reference frame.

1942

106 M . S. Longuet-Higgins

so that an integration d2—62— c2 = U2, (2.12)

a second constant. When a , b and с have been determined as functions o f tt then F
is given by (2.9).
In two special cases the equations can be integrated completely.
Two-dimensional flow. This corresponds to the limit

c-o, i- o , — = г2 < oo. (2.13)

С С

Then ab — lz, d*-6*= U \ (2.14)

so t 12 , 14
b — —, 0 = ----- r-, (2.15)
a a2

(2.16)
' 0 - Э - *
hence p®/
(2.17)
± Г ( 1“ 5 = )а *-
On the other hand we may have
Axisymmetric flow. This corresponds to

с = 6, (2.18)

so ab2 = 2L 3, аг - 2 Ъ г = и г \ (2.19)

hence b = (2 L 3)i a-i, 6 = - (if,3)! a a ^ (2.20)

and
(2.21)

hence
(2.22)
- ± f !'1
In both (2.17) and (2.22) the right-hand side is expressible as an elliptic integral.
I t is noteworthy that, in the general case, a second surface p = ps(t) may be found
from (2.5) and (2.6) provided that p = ps and D p /D t = ps represent the same surface.
The equation o f the surface must clearly be
x2 у2 z2 F+2ps
(2.23)
a2 fe2 c2 F

from (2.5), and from (2.6) a similar equation with (/ + 2ps) /Z1on the right. I f we then
take ps = AF (2.24)
with Л constant, the new surface is a hyperboloid similar to (2.10), and with semi-axes
a', b ', c' given by ,
- = i = - = ( l + 2A)l. (2.25)
a b с

3. The axisymmetric hyperboloid

The two-dimensional case (2.17) has already been discussed (with a different
approach) by Longuet-Higgins (1972, 1976), showing, in particular, that the fluid
accelerations become infinite when the angle between the asymptotes passes through
90°. We give now an analogous discussion for the axisymmetric case.
 1943

Bubbles, breaking waves and hyperbolic jets at a free surface 107

The free surface {k = 0) is given by
x2 y2+ z2
= 1. (3.1)
b2

This represents a hyperboloid o f two sheets, having the x-axis as axis of symmetry
(see figure 2). A section by the plane z — 0 gives the hyperbola

^ ! _ 2 ' ! == i (3-2)
a2 b2
To fix the ideas we shall discuss the lower branch, for which x > 0.

F i g u r e 2. A x isym m etric h yperboloid corresponding to (3.1).

Introduce dimensionless variables

a 0 b Ut
(3.3)
a ~L ' L ’ T L *

so that cep2 = 2, (3-4)

and (2.22) becomes (3.5)

т = J (1—a ~3)bda,

the lower limit being suitably chosen. The angle between the asymptotes is 2y, where

tan у = —
6
= —
f i
= —
21
j w-o)
a a oc*

by (3.4), and the radius of curvature R at the vertex (x ,y ) = (a,0) is given by

R b2 L p 2_ 2 L
(3.7)
1944

108 M . S. Longuet-Higgins

The velocity o f a particle at the vertex is (d,0), where

V (3 -8 )

Lastly „ .. §t/!a - 3
F = - a a = - lP a a „ = (1 _ a _ 3)2- (3.9)
(1—a -3)2'
Consider the limit o f (3.5) as a - » oo. When a > 1 we have

)da, (3.10)
r = 1,

and so writing ^ _ j » [(1 _ a _ 3), _ l]d a = _ ,.2936

(3.11)

we have т —t 0 = а + }а ~ г+ ^ а - 5+ ... (3.12)

or a = (t - t 0) -J ( t - t 0) - 2+ .... (3.13)
Н рпрр ol
tan y ~ [1 + | (т—T,,)-3] (3.14)

and л 2L
Й ~ { Г ^ 5 * [1 'Н (Т _Т 0 ) (3 ,5)
I t can be shown that the limiting form of the surface is a paraboloid whose dimensions
contract like t~2 about a point of similarity 0 lying on the axis o f symmetry at
one-quarter the distance from the vertex of the parabola to the focus. This limiting
form is one of the self-similar flows discussed in Longuet-Higgins (1976, §3).
Consider on the other hand the limit as а - » 1. Writing a = 1 + y , rj > 0, and
substituting in (3.5), we have

= £ [ i - ( i + 7 ) - 3)]irf7 ~ - ^ i , i . (3.16)

Hence
а ~ 1 + ( ^ г т ) 1, ( 3 1 7 )

and (3.6) and (3.7) give

(3.18)

Also, 2* ,
(3.19)

Thus as t -+0 the velocities (represented by a) tend to infinity like l~b and the
accelerations (also the normal pressure gradient) tend to infinity like Hence the
motion starts with a weak ‘ jo lt’ or shock, just as in the well-known two-dimensional
case (Longuet-Higgins 1972). Further discussion o f this phenomenon is given in the
appendix.
Equation (3.19) shows that the limiting angle between the asymptotes is
lim 2y = 2 arctan 2* = 109-47° (3.20)
compared with 90°, or 2 arctan 1*, in the two-dimensional case.
 1945

Bubbles, breaking waves and hyperbolic jets at a free surface 109

1946

110 M . S. Longuet-Higgins

The function r was calculated numerically by quadrature over the complete range
1 ^ a < 10. Representative values are given in table 1.
Figure 3 shows the development o f the hyperbola as a function o f the dimensionless
time т. For illustration we have shown not only the free surface p — 0, but also a
second surface p = ps = AF(t), with Л = 1*5.
W e note that values o f a lying in the range (0 < a < 1) correspond to angles 2y
greater than (3.20). However, we then have d2 < 2b2 and U2 would be negative. The
normal pressure gradient at the free surface also changes sign, and there is reason
to doubt that the flow is stable (cf. Longuet-Higgins 1972).

a r 2y R /l F /U 2
100 0 0000 109-47° 2-0000 1-5000
1-04 00090 106-26° 1-8491 1-3335
116 00676 97-08° 1-4863 0-9610
1*36 0-2078 83-45° 10813 0-5963
1-64 0-4417 67-91° 0-7436 0-3401
2-00 0-7798 53-13° 0-5000 01875
2*44 1-1887 40-71° 0-3359 01033
2-96 1-6951 3104° 0-2283 00578
3*56 2-2862 23-78° 0-1578 00332
4-24 2-9604 18-40° 0-1112 0-0197
500 3-7165 14-42° 0-0800 0-0120
5*84 4-5538 11-44° 00586 0-0075
6-76 5-4719 9-20° 00438 00049
7-76 6-4706 7-49° 00332 0-0032
8-84 7*5496 6-16° 0-0256 0-0022
1000 8-7089 512° 00200 0-0015

T a b l e 1. Values of r, 2y, R /L and F / U 2 as functions of a

4. Comparison with observation

To establish a coordinate system for figure 1 we chose two particular coordinate
frames, numbers 50 and 80, in which the hyperbolic form o f the jet was sufficiently
clear, and inserted, in each case, the centre of the hyperbola at the intersection of
the asymptotes. The frames being equally spaced in time, a straight line was drawn
through these two points; the other centres, some of which were less clear, were
assumed to lie on this line. From each centre, asymptotes were then drawn to the
hyperbolic part o f each profile. The angles 2у between the asymptotes are shown in
table 2.

Frame number 2y r
30 104° 0
40 47° 1-125
50 24° 2-25
60 15° 3*375
70 10° 4*5
80 8*5° 5-625
90 7° 6-75
100 6-5° 12-6
T a b l e 2. Measured values of the angle 2y between the asymptotes in figure I
 1947

Bubbles, breaking waves and hyperbolic jets at a free surface 111

F i g u r e 4. Time history of the angle 2y between the asymptotes according to (3.6). The plotted
points correspond to the observations in figure 1 .

In theory, since the reference frame in §§2 and 3 above is inertial, i.e. in free fall,
the trajectory o f the centre С should actually be in a free-fall parabola. However,
the elapsed time between frames 30 and 100 - 5*5 ms - is so short that the curvature
o f the trajectory is quite negligible: the maximum deflection would be about one part
in 103.
To obtain a dimensionless time r, it was noted that the angles 2у corresponding
to the two frame numbers 50 and 80 were 24° and 7°, corresponding to r = 2*25 and
6-75 respectively. The values o f r for the other frames were then inferred by linear
interpolation (and extrapolation) giving the numbers shown in column 3 of table 2.
These have been plotted in figure 4. The agreement with the theoretical curve will
be seen to be quite good. Especially notable is the observed angle 104° in frame no.
30, which is only a little less than the critical angle o f 109-5°. I t appears a coincidence
that this was one o f the frames selected for display.
In figure 10 of Blake & Gibson (1981) there was one further figure (10.11)
corresponding to a frame number 142. Extrapolation gives for this frame r = 12-6
and hence, from the model o f §3, 2у = 3-6°. The angle measured from the photograph
is about 3°, so that the agreement is maintained even as far as this point.
Near the base o f the jet in figure 1, where fluid is apparently being drawn into the
hyperboloid, close agreement is not to be expected. In fact the photographs from
frame 50 onwards show the formation o f a smaller, reverse jet directed into the cavity,
which by frame 100 has actually penetrated the lower cavity wall. Nevertheless, given
the unplanned character of the initial conditions, it is remarkable that the form of
the upwards jet should be so stable.
1948

112 M . S. Longuet-Higgins

The downwards jet appears similar to that produced by a shaped charge (see
Birkhoff et al. 1948). These authors model the phenomenon by a steady flow. Their
observations, however, indicate a uniform rate o f strain in the jets, more in accord
with the present theory.
The asymmetrical collapse o f cavitation bubbles in the interior o f a fluid or near
a solid surface has been discussed by Benjamin & Ellis (1966), Plesset & Chapman
(1971) and many others. From the theory o f §3 one would expect the maximum
acceleration o f the inwards jet to occur when the ‘ angle o f indentation ’ of the bubble
reached a critical value o f around 109-5°. Such a result appears not inconsistent with
the numerical calculations shown in figures 1 and 2 o f Plesset & Chapman (1971),
for example. However the development o f an inwards jet is necessarily limited in time
and space by the presence o f the cavity walls, so that comparison with the ideal
hyperbolic form is less feasible than for a bubble breaking a free surface.
Jets formed by a hollow liquid surface (without lining) have been produced
artificially by Bowden (1966), and are probably quite a common phenomenon. They
appear to be similar to the jets arising from the bowl o f any small bubble (diameter
1-2 mm) bursting at a free surface (see Blanchard & Woodcock 1980; MacIntyre 1968,
figure 1). Similar jets may model the ‘ splash’ produced after a solid body has fallen
rapidly through a free surface; see Worthington (1908).

5. Standing gravity waves

The observed tendency towards the formation o f jets in cavitation or vapour
bubbles prompts one to enquire whether a similar phenomenon may occur in other
situations involving free-surface flows. One example o f such a flow is a standing
gravity wave, driven slightly beyond its maximum amplitude.
Already in the case o f standing waves in two dimensions, it was suggested
(Longuet-Higgins 1972) that the Dirichlet hyperbola, with a critical angle o f 90°,
might model some o f the instabilities found experimentally by Taylor (1953) in
standing waves o f finite amplitude. I t was verified recently by M clver & Peregrine
(1981) that numerically calculated profiles of overdriven standing waves do indeed
conform to the Dirichlet hyperbola.
Turning now to axieymmetric gravity waves of finite amplitude, we note that in
a theoretical study Mack (1962) has suggested that a standing wave of maximum
amplitude should have a critical interior angle of 2 arctan y/2 at the crest - the same
critical angle that was found above in §3. Mack’s argument, which depends on the
pressure p being expansible everywhere as a Taylor series, even at the crest o f the
wave, resembles the argument put forward by Penney & Price (1952) for a crest angle
o f 90° in two-dimensional standing gravity waves of limiting amplitude, and so is open
to the same criticism as was directed against it by Taylor (1953), namely, there
appears no conclusive reason for assuming that at the instant o f maximum elevation
there is not a singularity in the pressure p at the crest. Taylor nevertheless showed
experimentally that the angle at the crest of a two-dimensional wave was close to
90°. W e may wonder whether, just as the two-dimensional Dirichlet hyperbola
describes an overdriven standing wave in two dimensions, so the axisymmetric
Dirichlet hyperboloid may describe an overdriven axisymmetric standing wave.
Some controlled experiments on axisymmetric gravity waves of finite amplitude
were undertaken by Fultz & Murty (1963; see also Edge & Walters 1964). These
experiments tended to verify some of Mack’s other theoretical results, but they were
apparently not carried as far as waves o f maximum amplitude. The reason no doubt
 1949

Bubbles, breaking waves and hyperbolic jets at a free surface 113

lay in the method of wave generation, which was by a plunger close to the free surface,
on the axis o f symmetry. This would tend to generate turbulence, and to interfere
with the surface at the crest.
However, Faraday (1831) showed that standing waves could be generated in a layer
o f liquid on the surface of a vibrating plate, or by the vertical oscillation o f a relatively
deep vessel o f liquid, and that the waves so produced were often of half the frequency
o f excitation. Physically, as we now know, this subharmonic resonance is related to
the existence, in ordinary standing water waves, o f pressure fluctuations at great
depths having a frequency double the fundamental frequency of the wave (see Miehe
1944; Longuet-Higgins & Ursell 1948). These second-order pressure fluctuations
remain o f finite amplitude, i.e. not tending to zero as one descends in the fluid, and
they have been shown to be responsible for oceanic microseisms of double the
frequency o f the sea waves (Longuet-Higgins 1950; Haubrich, Munk & Snodgrass
1963). Conversely, one finds that, by creating widespread pressure fluctuations at
great depths, standing surface waves of half the existing frequency can be pro­
duced. f
In §6 we shall describe an experiment in which this type of excitation is used to
generate an upwards pointing jet.

6 . Experiments on breaking waves

A cylindrical glass beaker of height 27 cm and internal diameter 16-4-16-5 cm was
filled with water to a depth o f 18*5 cm and placed on a horizontal wooden platform,
attached to a Derritron vibrator (type VP4B) as in figure 5. The vibrator was driven
by an electronic oscillator (Type 300WT, Mk 2) capable of producing sinusoidal
vertical movements o f amplitude 0-5 mm at frequencies from 0*5 to 1000 Hz.
In the centre o f the beaker was placed a thin baffle, consisting of two mutually
perpendicular sheets o f aluminium, each o f width 5*08 cm, mounted vertically on a
brass base, the top of the baffle being about 6-0 cm below the water surface. The
purpose o f this was to damp out the lower asymmetric modes of oscillation, while
leaving the axisymmetric modes relatively unhindered.
The oscillator was driven at a constant frequency /e = 6*64 Hz - slightly less than
twice the calculated resonant frequency of 3-43 Hz for the lowest axisymmetric mode
o f free oscillation (including surface tension but disregarding viscous effects at the
walls o f the cylinder). As the amplitude of the oscillator was increased from zero to
about 0-3 mm, there first appeared at the free surface synchronous surface oscillations
with a fundamental frequency equal to the forcing frequency. These were axisym­
metric, with radial modenumber 6. They then gave way to synchronous oscillations
with azimuthal modenumber 4 or 6. Finally there was a dramatic transition to a
subharmonic axisymmetric oscillation of half the driving frequency. With the
amplitude o f the vertical excitation set at about 0-3 mm the subharmonic axisym­
metric mode could be maintained in a quasi-steady state indefinitely, with the
silhouette o f the free surface approaching a sharp corner of about 110°, as, for
example, seen in figure 6(a). The slight rounding o f the crest is due presumably to
surface tension.
The attainment of this form was however very sensitive to the chosen frequency
o f excitation. A t slightly lower frequencies/e the subharmonic axisymmetric mode
o f frequency £/e itself became subject to subharmonic oscillations, axisymmetric or
otherwise, notably with frequencies \fe or £/e.
t F o r a m athem atical analysis based on M atthieu functions see Benjam in & Ursell (1954).
1950

114 M . S. Longuet-Higgins

F i g u r e 5. Glass beaker placed on a vertical Derritron vibrator, in a state of rest. The vertical flanges
on the axis are to suppress asymmetric modes of oscillation.

When the axisymmetric oscillation was overdriven by increasing the amplitude of

excitation to about 0*5 mm, the amplitude o f oscillation quickly increased (see figures
66 c) and the profile became unstable (figure 6/), ultimately producing an upwards
jet along the vertical axis (figures 6g, 7). The most spectacular jets rose to a height
o f more than 170 cm above the free surface (i.e. ten times the diameter o f the flask),
as proved by splashes found at this height on the white cardboard background. These
however were initiated on the axis of symmetry, near the instant when the free surface
was at its lowest point; the standing wave had already degenerated so far that it
somewhat resembled a collapsing bubble (see MacIntyre (1968, figure 1). The fact that
the tip o f the jet was well-formed (i.e. sharply pointed) by the time that it rose above
the mean level meant that the time history of the angle could not be measured very
accurately, at least in its early stages.
Nevertheless, the initial acceleration of some jets, which produced velocities o f
order 8 m/s in less than 0 05 s must have been in excess of 150.
The conclusions from this experiment are, first, that steady axisymmetric standing
waves can apparently be generated with a vertex angle less than the supposed critical
 Bubbles, breaking waves and hyperbolic jets at a free surface

Figure 6. Crest profiles of successive standing waves (frequency approximately 3*43 Hz) excited subharmonically
by a vertical oscillation (frequency 6*64 Hz). The interval between frames is about 0-29 s.
1951

115
М . S. Longuet-Higgins
 1953

Bubbles, breaking waves and hyperbolic jets ai a free surface 117

angle o f 109-5°. Second, that by overdriving the waves an unsteady, axisymmetric

jet, similar to that produced by an expanding gas bubble, can indeed be produced;
and, third, that the jet originates not near the crest of the wave (where it might have
been expected), but in the trough of the wave, in a similar way to the main splash
produced by a small bursting bubble (MacIntyre 1968). The fact that a small
oscillation o f amplitude as little as 0-5 mm can ultimately produce a jet rising to a
height o f over 1 7 m must also be considered remarkable.

7. Discussion
W e have shown that the axisymmetric jet produced by a gas bubble near a free
surface is well modelled by the inertial, inviscid, hyperbolic flow described in §3.
Moreover, a very similar jet has been shown to occur in axisymmetric standing waves
driven beyond their limiting amplitude. By inference we may expect that such jets
are a possible characteristic o f other types of unsteady free-surface flow. In particular,
we may take note o f a previous suggestion (Longuet-Higgins 1980) that the tip of
an overturning gravity wave may be suitably modelled by a hyperbolic jet of
generalized form, in which the principal axes are in rotation.
I t should be emphasized that the jets discussed in this paper are to a high degree
inertial, that is independent o f both gravity and surface tension. In the initial stages
o f formation, the time scale is too short for these other forces to exert an effect. A t
later stages both gravity and surface tension may become effective. Gravity causes
the jet to fall back, and the origin to describe a free-fall trajectory. Surface tension
can cause the jet to break up into droplets.
However, the evident stability o f the jet under surface tension is remarkable. The
reason may be that the flow is not simply a uniform inextensible stream, as is assumed
in the well-known perturbation analysis o f a jet of circular cross-section (see Lamb
1932, §274). In the hyperbolic jet, on the contrary, the free surface is constantly being
extended, with the consequence that any wavelike perturbation of the flow is being
drained o f energy by the radiation stresses. Conversely, if time were reversed so that
the free surface were being contracted, the flow would be highly unstable, quite
regardless o f surface tension.
In two-dimensional breaking waves o f plunging type, the jet is continually being
stretched in the direction o f motion, but not in the transverse direction. One therefore
expects instabilities to appear at first across the flow, with crests aligned parallel to
the jet. This is what is commonly observed.

The author is indebted to John Blake for arousing his interest in cavitation bubbles
and for subsequent stimulating discussions, and to Norman Smith for assistance with
the experiments described in §6.

Appendix. The initial ‘ shock’ : a physical discussion

The outstanding feature o f the flows described above is the occurrence o f a
singularity in both velocity and acceleration at the critical instant t = 0. What is the
reason ?
Consider first the two-dimensional flow (2.13) and choose the simplest case when
the hyperbolic cross-section reduces to a pair o f straight lines enclosing the variable
angle 2y, as in figure 8. It is useful to express the flow in Lagrangian coordinates in
1954

118 M . S. Longuet-Higgins

F i g u r e 8. Illu stration o f the stream lines and the free surface in a tw o-dim en sion al wedge. (T h e
v e rte x angle 2у is shown increasing.)

which the position (ж, у) o f a particle is given as a function o f t and the initial position

(* ° '2'о )Ь у * = ax0, у = by„. (A l)

Then the velocity (±, y) is (ax0, by0), satisfying (2.2). The equation o f continuity yields

1 = J M L = a6, (A 2)
d{x0,y0)

as in (2.14), with I = 1. The particle trajectories are therefore rectangular hyperbolae

given by /a ox
xy = x0y0 = constant (A 3)
(see figure 8).
On the other hand the pressure field, from the Lagrangian equations

Px0 = - x Xo* - y Xoy = |

(A 4)
Ръ = - х*.х-Уу'У = -Ы>. J
is given by p = -\ {a d x l + bbyl) + F {t ). (A 5)
By taking F = 0 we ensure that the free surface (p = 0) consists o f two straight lines
through the origin, or alternatively that we are observing the flow out at infinity in
the (x, y)-plane.
Consider then a particle P [x } y) in the free surface, and to fix the ideas suppose the
flow is reversed in time, so that the vertex angle 2у is less than 90° and is increasing.
The tangent P T to the trajectory makes an acute angle O P T with the vector PO.
Moreover since the trajectory curves always to the right (i.e. clockwise) the angle O P T
is increasing towards 90°. On the other hand, the only force accelerating the particle
is the pressure gradient, which is in the direction o f the normal P N to the free surface.
I f there were no force at all, the direction of the tangent to the trajectory would not
vary. Clearly the only way О Т can be made to curve towards the right by the purely
normal pressure gradient is by making the normal component of velocity increase
to infinity.
 1955

Bubbles, breaking waves and hyperbolic jets at a free surface 119

F ig u r e 9. As in figure 8, but for an axisymmetric, or conical, flow.

A direct but very brief analytical proof is as follows. From (7.3) we have

- + y
x - = o, (A 6)
x у

hence
(A 7)
x у
say. Thus
^ -(х г - у г) = 2 (x x -y y )
at
= 2 G (xx+yy)

= 2Gr(x cos 0 + y sin в), (A 8)

where ( r,6 ) are polar coordinates in the plane. The expression in brackets on the right
represents the radial acceleration, which vanishes identically. Hence x2—y2 is a
positive constant. But, because of the symmetry of the trajectory about в = 45°, we
see that when x - * y then x2tends to y2. These simultaneous requirements are possible
only if x and у each tend to infinity. In other words, the velocity becomes infinite.
Turning again to the axisymmetric case we can give a very similar physical
explanation. The essential difference is that in the axisymmetric case the trajectory
o f a given particle is no longer symmetric about a line through 0 inclined at 45° to
the x-axis. Instead, the tangent to the trajectory becomes normal to the radius vector
at a different angle у = arctan y/2 , as shown in figure 9.
A formal proof is as follows. I f p denotes (y2+ z2)b, the equation o f continuity now
requires that each trajectory satisfy

xp2 = x0pl = constant, (A 9)

hence
* = - ^ = G, (A 10)
x P
1956

120 M . S. Longuet-Higgins
say; therefore
j i (x2- 2 p i ) = 2 (x x -2 p p )

= 2 G (xx+p p )

= 2Gr(x cos в + р sin 0), (A ll)

where (r, 6) are now polar coordinates in the (a:,p)-plane. The right-hand side vanishes
as before. Hence x2— 2p2 is a constant, and each o f x2 and p2 must become infinite
when x2-+ 2p2, that is when p2—*•2x2, by (A 10). The critical angle is therefore
2 arc tan д/2.

REFERENCES
B e n ja m in , Т. B . & E l l i s , A. T. 1966 The collapse of cavitation bubbles and the pressures thereby
produced against solid boundaries. Phil. Trans. R. Soc. Lond. A 260, 221-240.
B e n ja m in , Т. B. & U r s e l l , F.
1954 The stability of the plane free surface of a liquid in vertical
periodic motion. Proc. R. Soc. Lond. A 225, 505-515.
B i r k h o f f , G., M a c D o u g a l l , D . P ., P u g h , E. M. & T a y l o r , G. I. 1948 Explosives with lined
cavities. J. Appl. Phys. 19, 563-582.
B l a k e , J. R. & G ib s o n , D. C.
1981 Growth and collapse of a vapour cavity near a free surface.
J. Fluid Mech. I l l , 123-140.
D. C. & W o o d c o c k , A. H. 1980 The production, concentration and vertical
B la n c h a r d ,
distribution of the sea-salt aerosol. Ann. New York Acad. Sci. 338, 330-347.
B o w d e n , F. P. 1966 The formation of microjets in liquids under the influence of impact or shock.
Phil. Trans. R. Soc. Lond. A 260, 94-95.
D i r i c h le t , P. L. 1860 Untersuchungen uberein Problem der Hydrodynamik. Abh. Kon. Ges. №4'ss.
Gottingen 8 , 3-42.
E d g e , R. D. & W a l t e r s , G. 1964 Period of standing gravity waves of largest amplitude on water.
J. Geophys. Res. 69, 1674-1675.
F a r a d a y , M. 1831 On a peculiar class of acoustical figures, and on certain forms assumed by groups
of particles on vibrating elastic surfaces. Phil. Trans. R. Soc. Lond., pp. 299-340.
F u l t z , D. & M u r t y , T. S.
1963 Experiments on the frequency of finite-amplitude axisymmetric
gravity waves in a circular cylinder. J. Geophys. Res. 68, 1457-1462.
H a u b r i c h , R. A., M u n k , W . H . & S n o d g r a s s , F. E. 1963 Comparative spectra of microseisms
and swell. Bull. Seism. Soc. Am. 53, 27-38.
Lam b, H . 1932 Introduction to Hydrodynamics. 6th edn. Cambridge University Press.
L o n g u e t - H i g g i n s , M. S. 1950 A theory of the origin of microseisms. Phil. Trans. R. Soc. Lond.
A 243, 1-35.
L o n g u e t - H i g g i n s , M. S. 1972 A class of exact, time-dependent, free-surface flows. J. Fluid Mech.
55, 529-543.
L o n g u e t - H i g g i n s , M. S. 1976 Self-similar, time-dependent flows with a free surface. J. Fluid
Mech. 73, 603-620.
L o n g u e t - H i g g i n s , M. S. 1980 On the forming of sharp corners at a free surface. Proc. R. Soc. Lond.
A 371, 453-478.
L o n g u e t - H i g g i n s , M. S.
1983 Rotating hyperbolic flow: particle trajectories and parametric
representation. Q. J. Mech. Appl. Math. 36 (May 1983).
L o n g u e t - H i g g i n s , M . S. & U r s e l l , F. 1948 Sea waves and microseisms. Nature 162, 700.
M a c I n t y r e , F. 1968 Bubbles: a boundary-layer ‘ microtome ’ for micron-thick samples of a liquid
surface. J. Phys. Chem. 72, 589-592.
M c I v e r , P . & P e r e g r i n e , D. H. 1981 Comparison of numerical and analytical results for waves
that are starting to break. In Proc. Symp. on Hydrodynamics in Ocean Engineering, Trondheim,
Norway, August 1981, pp. 203-215. University of Trondheim.
M a c k , L. R. 1962 Periodic, finite-amplitude, axisymmetric gravity waves. J. Geophys. Res. 67,
829-843.
 1957

Bubbles, breaking waves and hyperbolic jets at a free surface 121

M ic h e , R . 1944 M ou vem en ts ondulatoires de la mer en profondeur constante ou decroissante. Ann.

Ponts et Chaussees 114, 27-78, 131-164, 270-292, 369-406.
P e n n e y , W . G. & P r i c e , A . T . 1952 F in ite periodic stationary g ra v ity waves in a perfect fluid.
Phil. Trans. R. Soc. Lond. A 244, 254-284.
P l e s s e t , M. S. & C h ap m an , R . B.
1971 Collapse o f an in itia lly spherical vapou r c a v ity in the
neighbourhood o f a solid boundary. J. Fluid Mech. 47, 283-290.
T a y l o r , G. I. 1953 A n experim ental study o f standing waves. Proc. R. Soc. Lond. A 21 8 , 44-59.
W o r t h i n g t o n , A . M. 1908 A Study o f Splashes. Longm ans, Green & Co.
1958

R O T A T I N G H Y P E R B O L I C FLOW:
PARTICLE TRAJECTORIES
AND PARAMETRIC REPRESENTATION

By M. S. L O N G U E T -H IG G IN S

(Department of Applied Mathematics and Theoretical Physics, Silver Street,

Cambridge, and Institute of Oceanographic Sciences, Wormley, Surrey)

[Received 7 October 1981. Revise 2 June 1982]

SUM M ARY
A class of time-dependent, irrotational flows in which the velocity potential has the
form \ A z 2 (where z is a complex coordinate and A is a complex function of the
time) are of special interest in relation to the flow in the jet of a plunging breaker. In
this paper we calculate the trajectories of marked particles, and make use of the
result to construct a semi-Lagrangian representation of the flow. Special attention is
paid to the limit when the free-surface profile tends to an infinitely thin hyperbola,
and to another special case (the “ 45°-rotor”) in which the axes rotate with uniform
angular velocity, and the angle between them is constant at 45°.

1. Introduction
The problem of describing mathematically the breaking of gravity waves in
deep or shallow water involves, at the very least, solving the equations of
motion for time-dependent inviscid flow with a free surface. Such solutions
are rare, but an interesting class of flows, which may describe the flow in the
tip of a plunging breaker, was recently pointed out by the present author ( 1 ).
In these the surface takes the form of a hyperbola, in which the origin falls
freely in a parabolic path while the principal axes rotate non-uniformly in
time. The angle between the asymptotes also varies, and in an important
class of cases it tends to zero, giving rise to a slender or sharp-pointed jet.
The solutions in (1) were given essentially in Eulerian coordinates, using a
velocity potential \ as a function of the complex coordinate z and of the
time t. If the flows were steady, the particle trajectories would be given
simply by the streamlines Im \ = constant. But in a time-dependent flow the
streamlines continually vary, and only touch the trajectories instantaneously.
The first aim of the present paper is to determine the particle trajectories
analytically.
A second aim is more far-reaching. As shown in a companion paper (2)
the complete description of an overturning wave probably will involve a
solution given in parametric form, with \ and z each expressed in terms of a
third complex variable o>, and t. Now in the simplest type of parametric
representation, due originally to John (3), it is assumed that at the free
surface w takes only real values, and moreover that at the surface (but not

[Q . Л Mech, appl. Math., Vol. 36, PL 2, 1983]

248 M. S. L O N G U E T - H I G G I N S

elsewhere) <d is constant following a particle. This representation of the flow

is, so to speak, semi-Lagrangian. It was shown in (2) that some highly
suggestive flows, including one that resembles the forward face of a plunging
breaker, can be described in this semi-Lagrangian form. As a next step,
therefore, we propose to describe the hyperbolic flow of ( 1 ) in semi-
Lagrangian form, so that possibly the two types of flow in (1) and (2) may be
combined.
One would naturally expect that the semi-Lagrangian description of a flow
would be closely related analytically to the general analysis of the particle
trajectories. This relation is explored in the present paper.
In sections 2 to 5 we give the general solution for particle trajectories in a
hyperbolic flow. The solution is illustrated in sections 6 and 7 by two simple
cases: the “ Dirichlet hyperbola” , where the orientation of the axes is fixed
in space, and the “ 45°-rotor” , where the whole flow rotates with constant
angular velocity. The physically most important case of the slender jet is
worked out in section 8. The remainder of the paper concerns the relation
between the general solution of sections 3 to 5 and the desired semi-
Lagrangian description. Section 9 states the basic result, which is then
proved in section 10. The conclusions are discussed in section 11, and
numerical computations are given in section 12 .

2. Statement of the problem

W e are given a flow in the Eulerian representation by
Х Ч А Л (2.1)

where x is the complex velocity potential, z is the complex coordinate x + iy

and A = a e fa, (2.2)

with a and cr real functions of the time t. It was shown in (1) that (2.1)
represents a flow with a time-dependent free surface, attached to the
particles, provided that
cr, = A a 2 (2.3)

and
aa„ - 4a f + 2a4- \2a 6= 0, (2.4)

where Л is a constant, and we use suffices to denote differentiation. Equa­

tions ( 2 .3 ) and ( 2 .4) can be integrated to give
 R O T A T IN G H YPE R B O LIC FLOW 249

respectively, where P is a second constant of integration. The equation of

the free surface is

A ,z 2 4 -2 A A * z z * + A * z * 2= ~2S<*4, (2.7)

in which a star denotes the complex conjugate, and S is a third arbitrary

constant. When S > 0 equation (2.7) represents a surface whose profile is a
hyperbola, rotating about its centre with angular velocity —S„ where

8f =A<x2/(1 + Pa4) (2 .8)

(see section 4 of (1)). In the special case A = 0 the axes are stationary and
the profile reduces to a Dirichlet hyperbola, studied earlier (see (4,5)).
Unless otherwise stated we shall assume A.^0.
The problem which we consider first is how to determine the trajectories
of fixed particles. The flow being time-dependent, the trajectories are
different from the instantaneous streamlines, which provide only the
tangents to the trajectories.

3. Equations lor trajectories

Suppose the coordinate z is expressed in Lagrangian form, so that the
particle velocity is simply zt. Then from (2.1) we have
(3.1)

Introduce the new time variable

(3.2)

Then our basic equation becomes

(3.3)

Hence

(3.4)

say, where ф is real. Therefore

(3.5)

W e obtain an equation for ф as follows. Let z = reie; then from (3.5)

 1961

250 M . S. L O N G U E T - H IG G IN S

But from (3.4)

ф+ сг = -26, (3.7)
so that
фт+ сгг = —2вт= —2 sin ф ( 3 .8)
by (3.6). Also from (2.5) and (3.2)

<rT = crj г, = Aa. (3.9)

W e have then
фт+ 2 sin ф = —Aa, (3.10)
a being a given function of f, hence of r.
It is somewhat more convenient to replace r by a as independent
variable. Then (3.10) takes the form

+ (2/Aa) sin ф = - 1 . (3.11)

The function a can be expressed directly in terms of cr by eliminating t from

equations (2.5) and (2.6), giving

(312)
where
ш = Р/А4.

Then in place of (3.5) we use

J
2 = R exp ^A-1 а " 1е,фdcrj (3.13)

in which the constant R depends on the lower limit in the integral. For
points on the free surface, R may be related to S by substitution in equation
(2.7).

4. The Riccati equation

By the substitution
v = —cot (4.1)

equation (3.11) is transformed into the Riccati equationt

vtT-(2l\oL)v+\{\ + v2) = 0> (4.2)

which can be reduced to a linear (second-order) equation by setting

17 = exp Qji/dcr^. (4.3)

t F o r an account o f this equation see W atson (6).

 R O T A T I N G H Y P E R B O L IC FLOW 251

Hence \ve obtain

tw ~^ Ur + Jt) = 0. (4.4)

This generally has two independent solutions t](1) and V 2) say, so that

rj = ari(l) + bri(2), (4.5)

where a and b are constants. Then from (4.3) we have

2 71•o’ n П (1) +’ СП
'Iff (2>
c Mcr i л s\
v = ---- = 2 - ц Г Г — (2), (4.6)
TJ T| "b CT]

where с = bla. So the solution to (4.2) involves only one arbitrary constant c,
as we would expect. T o find z we have now from (4.1) that

e * = (v -i)/ (i/ + i), (4.7)

which expression can be substituted into (3.13).

5. A linear equation for z

A method which is less direct but ultimately simpler is as follows. In the
first-order relations

zf = A * z * , z* = A z (5.1)

between z and z * let us differentiate with respect to t:

zn = A * z * + A * z * t (5.2)
and substitute for z * and z* from (5.1). This gives

z,t = (A */ A *)zt + A A * z (5.3)

or
A * z tt- A t z t- A A * 2z = 0, (5.4)
a second-order equation for z alone.
However, not every solution of (5.4) is a solution of (5.1). Whereas the
general solution of (5.4) involves two arbitrary complex constants, hence
four degrees of freedom, the solution of ( 5 .1 ) involves only two degrees of
freedom.
Let z (1) and z <2) denote any two linearly independent solutions of (5.1),
which we know to exist from section 4. These must also be solutions of (5.4).
Moreover the general solution of (5.4) is of the form

z = L z (1) + M z (2), (5.5)

where L and M are arbitrary (possibly complex) constants. Under what

conditions does this satisfy (5.1)? Substitution in (5.1) gives

L z ;i) + M zi2) = A * a * 2 (1)* - f M * z (2)*). (5.6)

 1963

252 M. S. L O N G U E T - H IG G IN S

But z (1) and z <2) each individually satisfies (5.1). Hence

L A * z il)* + M A * z (2)* = A * ( L * z (l)* + M * z (2)*). (5.7)

Dividing by A * and noting that z (1>* and z (2)* are independent we have

L = L *, M = M* (5.8)
i.e. L and M are both rea/. Conversely, if L and M are real, then (5.5) is a
solution of (5.1).
Now let z = z (1) denote any solution of (5.1) and hence (5.4). A second
solution Z to (5.4) may be obtained by writing
Z = uz(1), (5.9)

and substituting in (5.4). This gives

А *(ы„г(1) 4- 2 + uzj,n) - A?(w,z(1) + uz\l)) - A A * 2uz(1> = 0.

(5.10)

Since z = z (1) satisfies (5.4), this reduces to

( A * z ) m„ + (2 A * z t - А ? г )ц =0 . (5.11)

Hence

(5.12)
ц z A*
which on integration gives
u, = iK A *fz 2, (5.13)

where К is a constant (possibly complex). Then from (5.9) our second

solution is

,(2) = i K z J ( A * / z 2) dt, (5.14)

the lower limit being undetermined.

Under what conditions is z = z (2) a solution of (5.1)? If

z[2) - A * z (2)* , (5.15)

then we must have

ije jz .J ( A * lz 2) dt + A * lz J = - ilC * A * z * | (A / z * 2) dt, (5.16)

that is

* { a * z * J ( A * / z 2) d t+ A * / z J = - e - ‘M * z * J ( A / z * 2) dr, (5.17)
1964

R O T A T IN G H Y P E R B O L IC F L O W 253

where у = arg K. Writing

e i7| (A * / z 2) d t = Ф, (5.18)

we obtain
г г * (Ф + Ф * ) + е * = 0 . (5.19)

Since the first terms are real, it follows that e'y = ± l . Without loss of
generality, take eiy = l. Then (5.19) becomes

z z * j(A * / z 2+ A/z *2) dt + 1 = 0 . (5.20)

But the integral in (5.20) can be written

Therefore (5.20) becomes simply

z z * / (z z \ = 0; (5.22)
in other words the lower limit of integration in (5.14) must correspond to
|z|=o°.
T o summarise, we have shown that if z ~ z (l) is any solution of (5.1) and
(5.4), then a second solution z = z (2) is given by

z ^ = i z ^ \ ^ - 2dt, (5.23)

provided the low er limit o f integration corresponds to |z|=°°. Furthermore,

the general solution to equations (5.1) is o f the form

z = L z (1) + M z (2), (5.24)

where L and M are both real constants.

W e note that from (5.1) and (5.2) the particle acceleration z„ can be
expressed linearly in terms of z and z * by

z„ = A A * z + A * z * , (5.25)

so that the equation (2.7) of the free surface can also be written

zz* + z * z tl= -2 S a 4. (5.26)

Substitution of (5.24) into the above identity leads to a normalising relation

between L, M and S.
Lastly we note that (5.3) can be expressed solely in terms of the real
function a. For since A, = (a, + iaat)eitr and crt = Aa2, (5.3) becomes

= (a t/a - iAa2)z f 4- a 2z. (5.27)

W e shall now give some examples.
 1965

254 M. S. L O N G U E T - H IG G IN S

6. The case Л = 0
In this very special case, equations (2.3) and (2.8) show that crt and 8, both
vanish, and the principal axes of the hyperbola are stationary. Accordingly
we may take
<r = 0, A = a, real, (6.1)

and equation (5.27) can be written

a (a - 'z t)t = a 2z. ( 6.2)

Using the time variable r of (3.2) we have simply

z„ = z, (6.3)

so that

z = L exp ^J a dtj + iM exp J a dtj, (6.4)

and substitution in (3.3) or (5.1) shows that L and M are both real.
Moreover the functions

z (1) = exp (J,«*),

(6.5)
z (2)
4 Ч -Ы
are related as in (5.23).
Taking real and imaginary parts of (6.4) we have

x = L exp Q a у = M exp J a d?j, ( 6.6)

so that the trajectories are simply

xy = LM, a constant, (6.7)

that is to say they are rectangular hyperbolae, the origin assumed stationary.
By (2.4) the function a satisfies
aa„ —4a? + 2 a 4 = 0 ( 6.8)

and hence

Н д т т к ч !- ( 6 '9 )

This is the non-rotating case discussed by Longuet-Higgins (4 ,5).

7. The 45°-rotor
An interesting case mentioned in (1) is when m = J and
Aa = constant = V2, (7.1)
1966

R O T A T IN G H Y P E R B O L IC F L O W 255

Fig . 1. Particle trajectories in the 45°-rotor (w = *). The free surface is

shown at times r/Л = constant + n, n = 1,2, 3,4.

so that by (2.3)

Aat = 2 , (7.2)

and

<7 = 2t/A, A = (v/2 M )e2',/\ (7.3)

The angle between the asymptotes is constant at 45°, and the whole flow
rotates with a constant angular velocity - 8 t = - A -1 (see Fig. 1).
On choosing the time scale so that A = 1, equation (5.4) becomes simply

zn + 2izt - 2 z = 0. (7.4)
Thus

z = le~il+i)t + me(1-l)t, (7.5)

where I and m are constants. Substituting in (5.1) gives

- 1 ( 1 4 - 0 < Г (1+о‘ + m ( l - i) e (1“ 01 = V 2 e " 2lt{ l* e ~ (l-I)t + m * e (1+0‘}. (7 .6 )

 1967

25 6 M . S. L O N G U E T - H IG G IN S

Therefore

(7.7)

(7.8)

(7.9)

It will be seen that the functions

z ( l ) _ e -(l-M)r-*-frrri

Z(2 )_ _ L e (l—0l+i7ri (7.10)

are related as in (5.23). Also, substitution of (7.9) in the free-surface

equation (5.26) gives

L M —2S. (7.11)
The same result can be obtained, though at greater length, by solving
equation (4.4) for 17 and making the substitutions indicated in sections 3 and
4.
Consider (7.9); when t is large and negative the first term on the right of
(7.9) is dominant and so

zt e —( l + i)z. (7.12)

This is the equation of a 45° equiangular spiral. The particles are far out
along the trailing face of the rotating hyperbola and are spiralling inwards
towards the vertex (see Fig. 2). On the other hand when t is large and
positive the second term is dominant and so
(7.13)

The particles are far out on the leading face of the hyperbola and are
spiralling outwards.
A t intermediate values of t the two terms in (7.9) may be comparable. In
fact the vertex of the hyperbola corresponds to

e2t = V2L/M. (7.14)

8. Slender hyperbolae
In one particularly interesting class of flows described in (1), the angle 7
between the asymptotes tends to zero as * —►«>, while the orientation 6 of
the axes tends towards a constant limiting value. Such flows may be of
particular interest for comparison with the tip of a plunging breaker.
1968

R O TATING H YPER BO LIC FLOW 257

F ig . 2. Particle trajectories relative to a fixed origin O, when m = 0-5. The

surface profiles are shown at times corresponding to / = 0-0, 1*0, 2-0, 3-0
and 4-0. The real axis of z is directed upwards.

This limiting case is represented by a - > 0 in (2.6), hence t ~ l f a -+<*>.

Also, from (3.12), < r ~ -k a .
It is clear from section 3 that A occurs only in the combinations f/A and
Aa. So if А ф 0 we may, by a suitable choice of time scale, set A = 1. Then, by
a straightforward expansion in powers of a or cr, we find from (3.12)

cr = - a - i a 3+ ( j b w - & ) a 5 + . . . , (8.1)
 1969

258 M. S. L O N G U E T - H IG G IN S

and hence
= ~cr + б€Г3 —(io^r + Т2б)сг 5 + . . . , (8.2)
OJ- 1 = _ 0г-1_1 0.+ (Х ш __7_)0.з_|„ ? ( 8.3)

and so on. By substitution for a -1 in equation (4.4) we may find solutions

for rj in the form

t) = a0o-p + a2crp+2 + a4<rp+4 + . . . , (8.4)

where a0^ 0 and p is a constant to be determined. The indicial equation, got
by equating coefficients of crp-2 in (4.4), is

p ( p - l ) + 2p = 0. (8.5)
Hence p = - 1 or 0, and we find for the two independent solutions

1^1) = о-- , + £ < 7 -(б Ь е т -5 ’щ )<г3+ . • • > 1 ,

V 2,= i - ^ 2+ ^ 4+ . . . . J
On substitution in (4.6) and (4.7) we find that
е*Ф = l + K r+ (ic —%)<r2—(c+|i)<r 3 + . . . , (8.7)

and hence

ф = cr + ccr2+ 0 (<74), ( 8 .8)

leading to

z « o - - 1 - i- (J + | ic )o - - £ c o -2 + ....... (8.9)
To express this result in terms of r, we note that when m > J the integrand
in equation (2.6) is always negative. Hence t is monotonic in a, and we may
choose the lower limit of integration as a = oo. Then

(8 .10)
Ja a (1 - a +WGL p

making t —>0 as a —><» and vice versa. When |a2 —w a 4|<l,. the integrand
may be expanded in powers of a. Hence

, = tl+Ji-ia+(iOT-J)a3-...}, (8.11)
where is a constant, given by

ti = lim ( f - —Y ( 8. 12 )
a-+0 \ aJ

For numerical values of ti see Table 1. The series on the right of (8.11)
involves only odd powers of a. Thus writing
(8.13)
1970

R O T A T IN G H Y P E R B O L IC F LO W 259

T able 1. Values of tl and <r0.

20 -0-5568 1*7175
1*0 -0*2656 2-1565
0-5 0-0997 2-8542
0-4 0*2698 3*2146
0-3 0*6413 40196
0-25 oo 00

and inverting the series we find

1 1 /w 3\ 1 /ОЛА.
а ~ Т ~ 2 Р + \ 6 + 8 ) ^ “ - - - ( 8 Л 4 )

Now from (8.1) we can express cr in terms of t':

1 1 / т а г 1 \ 1 / o i c 4

° = — +^ ~ \ T s + (815)
From (8.9) and (8.15) we find that

z = J ? ( r '4 - i - ^ + g ^ + . ..), (8.16)

where R is a constant. By substituting in equation (5.1) and equating the

coefficients of leading order we obtain

K = R *. (8.17)

Hence R is real.
The solution corresponding to JR = 1 and с = 0 is clearly
z (1) = ( ' + * + . . . , (8.18)

which becomes infinite as * —*<». So from (5.23) we have a second solution

(8.19)

which is easily identified as the coefficient of с in equation (8.16). As proved

in section 5, the general solution of (5.1) must be of the form (5.24), where
from (8.16)
L = R } M = Rc. (8.20)

This shows further that in (8.16) the parameter с occurs at most linearly in
the coefficients of t,n.
T o carry the expansions further it is now easiest to use equation (5.4) in
the form
(«t /а “ ia 2)z f ~ = 0, (8.21)
 1971

260 M. S. L O N G U E T - H I G G I N S

where a is expanded as a series in lit', as in (8.14). In this way we find

. ш InУ
zw= f'+, ....
24t 120t'
(8.22)
,( 2) _ i 1 i
+ ---------,
2*' 6 f'2 6 f'3 10f'4

as far as the terms in 1 /f'4.

Lastly, on substituting from (5.24) into (5.26) and comparing coefficients
of l/f '4 we obtain

R 2(m —c2) —2S, (8.23)

that is

v j L 2- M 2 = 2S. (8.24)

W e leave the discussion of this result until section 11.

9. A semi-Lagrangian representation
For the reasons given in section 1, we are interested in finding for the flow
(2.1) a Lagrangian or semi-Lagrangian representation of the type proposed
by John (3). In this, the position coordinate z is expressed as a function of a
complex variable <o and of the time, in such a way that on the free surface <o
is real and also Lagrangian, that is to say the velocity and acceleration are
given by zt(<o, t) and z„(o>, f). In general, however, the velocity equals
z,(oj*, f), not z,(<t>, f). Only on the free surface are these expressions
equivalent.
The representation has the advantage, first, that the free-surface con­
dition can be expressed in the simple form
(9.1)

where r(co, r) is to be real when <o is real. Second, the velocity potential, if
required, can be found from
X «,=X 2z„ = z?(<»)z„(<o) (9.2)

by direct integration with respect to cj.

A restriction is that any zeros of within the fluid become singularities
(branch-points) unless also z*„(a)) vanish there at the same time. In other
words
гш(со) = 0 implies z*(<*>) = 0 (9.3)

for all points in the fluid.

T o find such a representation for the flow (2.1), consider the expression
z = F cosh <0 + G sinh (9.4)
1972

R O TA TIN G H YPERBO LIC FLOW 261

where F and G are functions of t only, possibly complex. When со is real,

equation (9.4) represents one branch o f a hyperbola. Also

z ^ - F sinh o) + G cosh со (9.5)

which vanishes only at a focus o f the hyperbola, that is in the interior of the
fluid. But

z £,(<*>) = F * sinh co + G ? cosh со, (9.6)

so the simultaneous vanishing of z*, and г * ш implies that

F*/F = G*/G. (9.7)

On the other hand the boundary condition (9.1) becomes

(F„ cosh co + G„ sinh g>) = ir(F sinh со + G cosh со), (9.8)

and hence
F„ = irG ,}
(9.9)
G tt = irF,l
where r must be real for real со.
The non-rotating case of section 6 corresponds to F real and G pure
imaginary, say G = »G'. Then (9.7) becomes

FJF = - G 'tIG ' = l3, (9.10)

where 0 is real. Let us take 0 = a. Then

F = exp ^ J a d t 'j, G ' = exp (9.11)

Elimination of r from equation (9.9) gives

FJG * = —r = —G\JF, (9.12)

and then substituting from (9.11) we have

(a, + a 2) exp ^2 J a dt j = (оц - a 2) exp ^-2 J a dt^. (9.13)

Differentiating each side logarithmically we find

tn + 2aat ] ^ q n- 2 o
^ +2а^ + 2 а = ^ ф - 2 а , (9.14)
a ,+<*2 a, —a

aa„ —4a? + 2 a 4 = 0, (9.15)

the equivalent of equation (2.4) when Л = 0.

Finally, from (9.13) we have

r2 = - F ttG\JFG' = a? - a 4. (9.16)
 1973

262 M. S. L O N G U E T - H I G G I N S

But from (2.6), when A = 0 ,

a? = a 4( l + P a 4), (9.17)
so that

r2= Pa8. (9.18)

10. General solution

W e return now to the general case when А ф0, and the hyperbola rotates.
Choose units so that A = 1. W e shall prove the following theorem:

T h e o r e m . Let c, t) be the general solution of equation (3.11) based on

the function v of equation (4.6), and set

H(a, c, 0 = exp ^J a “ V * d tj . (10.1)

Then
z = F (r)cosh a> + G(0sinh<«) ( 10 .2)
is a semi-Lagrangian representation of the flow (2.1) if we take

(a ,00,, t),'
F = (2 S / w № Г(а, f),l
(10.3)
G = (2 S )iH c(<
(a,0, f). J

Proof. W e have to show that these expressions satisfy equation (9.7) and
also equations (9.9) with a suitable choice of r.
First, since z = H (a , c, r) is a solution of equation (3.1) we must have

H* = AH (10.4)

for all values of t and c. But A is independent of c. So differentiating each

side of ( 10 .5 ) partially with respect to с we have
H * = H * —A H C. (10.5)

(W e note z = H c is also a solution of (3.1).) From (10.4) and (10.5) it follows

that
Н ?/Я = Н £/Д , (10.6)

In particular setting с = 0 we obtain

F f/ F = G*/G, (Ю.7)

as required.
T o prove (9.9) we establish the following lemma.

L emma. Let Z be any solution of (3.1) and let

Z„ = ire, (10.8)
1974

R O T A T IN G H Y P E R B O L IC F L O W 263

where r = -m ^ a 4 and a satisfies (2.4). Then reciprocally

£„ = irZ (10.9)
with the same function r.
Proof. Writing r = N a 4 we have from (10.8) and (5.27) that
INC = (a ~5a, —ia ~2) Z t + a ~2Z. (10.10)

On differentiating each side with respect to t and substituting for Z „ from

(5.27), we get

iN£t = а ” 6( а а „ - 4 a 2+ a 4- a 6) Z t - ( a ~3at + i)Z , (10.11)

which by (2.4) reduces to

iN£t = —a _ 2Z, —( а _ 3а, + i)Z . (10.12)

Again differentiating each side and substituting for z„ from (5.27) we find

iNS„ = - a ~ 4(aatl- 3 a 2+ a 4)Z . (10.13)

So using (2.4) we obtain

zN£„ = - a ~ 4{ a t - a 4+ a 6)Z . (10.14)

But by (2.6)
a2 = a 4( l —a 2 + tzra4). (10.15)
So (10.14) gives
£tl = iN~1/aja4Z = irZy (10.16)

provided that
N 2 = ш. (10.17)

This proves the lemma.

Now suppose that Z = /z(l) where z (1) is a solution of (5.1), and that
£ = mz(2) where z (2> is related to z (1) as in (5.23), I and m being real. W e
shall show that with a suitable choice of I and m the relation ( 10 .8) is indeed
satisfied.
For then (10.8) can be written

a - 4z j z = —(Nm/I)J (A */z2) dt, (10.18)

where z = z (1). On substituting for z„ as before, then differentiating and

substituting for zn and att> we obtain
( а _ 3а, + 0 + 2 a “ 2(zl/z) + (o T 5af - ia~2)(z t/z)2 = NmA*/lz2. (10.19)

Now since z, = A * z * this becomes

A tz2+ 2 A A * zz* + A * z*2= Nma4Jl. (10.20)
 1975

264 M. S. L O N G U E T - H IG G IN S

But for points on the free surface, equation (2.7) is satisfied by z = fz(1).
Hence equation (10.20) is indeed satisfied provided that

Nm/l = —2S//2, (10.21)

that is to say
Im = 2S!vjK (10.22)

W e have then only to verify equation (10.18) for one particular value of f, or
in the limit as t —> °°.
Consider now the solution of section 8. From the expansion (8.22) we
easily find that the two sides of (10.18) are equal asymptotically provided

Nm/l = - w , (10.23)
that is
m/l = m*, (10.24)
and comparing (10.22) and (10.24) we have

/= (2S/m)±, m = (2S)*. (10.25)

Finally, since it was shown that

z (1) = H (a, 0, 0 , z (2) = Hc(a, 0 , t), (10.26)

the theorem (10.3) is proved.

11. Discussion
For points on the free surface, the trajectories given by the two represen­
tations of sections 8 and 10 should be equivalent.
T o see this, we may set in (8.24)
L - (2S/or)* cosh o>, M = (2S)* sinh со, (11.1)

which by ( 8.20) implies

с = M IL = tanh w. (11.2)

So for points on the surface of hyperbolae we have

z = (2S/m)*(z(l) cosh w + w*z(2) sinh <o), (11.3)

where z (1) and z (2) are the two functions given in (8.22). It will be seen that
this agrees precisely with the representation of section 10 .
Consider the interpretation of (11.3). Equations (8.22) show that, in the
limit as t —> °o,
z (1)~ r ' + i , z (2)----- i/2t\ (11.4)

so
z ~ {2 S !m )K t'+ i) cosh a>. (11.5)
1976

R O T A T IN G H Y P E R B O L IC F L O W 265

The point a) = 0 is at z ~(2S/w )H f' + i), which travels with constant limiting
velocity
C = (2S/m)K ( И . 6)
and if we take the origin at this point, the limiting position of each particle
as t —> oo is given by

z '- W Q S M K t ' + i), (П .7 )

when u) is small. In other words the limiting form of the profile is a straight
line along the real axis. The particles are being stretched out along this line,
linearly in time.
The line is in reality an infinitely thin hyperbola, given by

x' = |(2SM*ta>2,
( 11 .8)
y' = -(2 S )W 2 r.

The tip of the hyperbola continues to rotate with diminishing angular

velocity -5 , given by ( 2 .8), i.e.

8, = a 2/(l + та4) ~ a 2. (11-9)

Since the transverse angular velocity of a particle is, by (3.7) and (6.9),
0t = + о-), ~ a 2( l + со-), (11.10)

the particle gains on the tip of the hyperbola by a relative angular velocity

a 2ccr---- ct~3' — m W -3. ( 1 1 .1 1 )

W e note that for the 45°-rotor, in which the angle between the asymptotes
does not tend to zeo at any time r, a semi-Lagrangian representation
corresponding to ( 10 .2 ) is

2 = F (0 e a’ + G (r)e -tt>, (11.12)

where
F = (2S)*H (a,0, t),
(П .1 3 )
G = (2S)*HC(<
(a ,0 , (),J

and

H (a , c, t) = е-<1* ° '+* " + - j^ gO-O'+ и . (11.14)

The expression (11.12) is in agreement with (7.9), provided

L = (2S)*ew, M = (2 S)ie-w (11.15)

and o) is real. These values satisfy (7.11). Moreover, the necessary condi­
tions (9.1) and (9.3) are also satisfied provided
r = 2. (11.16)
 1977

266 M. S. L O N G U E T - H IG G IN S

It can also be shown that z = H (a, c, t)y с real, represents the general
particle trajectory as derived by the method of section 4.

12. Numerical calculations

Any member of the family of trajectories given by (11.3) is easily found,
once the two functions z (1) and z (2) have been evaluated. To do this we may
first evaluate the functions a and t by quadrature from ( 2 .6) and ( 3 . 12 ),
setting a = /3 = Ilf. Thus

( 12. 1)

and

(12.2)

Generally it suffices to use Simpson’s rule, with equal increments of /. When

f is small we have

(12.3)

1 1
t~f+ h, o '------ (12.4)
I

provided that

(12.5)

and

( 12.6)

Some values of crQ and tj calculated in this way are given in Table 1.
T o calculate z (1) and z (2) we may compute two trial functions za(t) and
*ь (0 beginning with any convenient initial conditions, say
za( 0) = zaO> zb( 0) = zb0, (12.7)

and integrate forward using equation (3.1) in the form

( 12.8)

hence

(12.9)
af
1978

R O T A T IN G H Y P E R B O L IC FLOW 267

For graphical accuracy it is sufficient to use the “ leap-frogging” formulae

( 12 . 10)
z- = z "- 2 + 2 ( | L a/’
the two initial values z 0 and Zj being calculated from the expansion

2 = г ° + й Н й ' 2 + й ' 3 - - ( 1 2 1 1 )

in which / = 0 and Д/ respectively.t Now by section 8 we must have

za = aa)zm + a^’z
( 12.12)
zb = bm z w + b m zm :)

i ivj. j . jtoxiiuic u a jc c iu iic a w iuw ii w u cu iiic o u g u i is ш a u c c -i a u pai

w = 0 '3 , (b) да = 0-5, (с) да = l'O, (d) да=2-0.

t Неге Д/ is double the increment used in the calculation of (12.1) and (12.2).
 1979

Fio. 3. (Contd.)
1980

R O T A T IN G H Y P E R B O L IC F L O W 269

Fig. 3. ( Contd.)

where a (1), a (2), b(1) and bi2) are constants. So as t-+ ° ° and /-►

z ^ a V + a V , (12.13)
where s(1\ s(2) are the truncated series for z (1), z(2) on the right of (8.22).
Hence

a (1) = lim za/s(1) (12.14)

and

aC2) = lim(^a - a (1)s(1))/s(2). (12.15)

In practice a(1) and a(2) can each be determined from a sequence of values at
uniformly spaced values of f, using Shanks transforms; and similarly for b(1)
and b(2). A check on the results is that

arg a (1) = arg a(2) = arg b(l) = arg b(2\ (12.16)

 1981

270 M . S. L O N G U E T - H I G G I N S

Finally we solve equations (12.12) for z (1) and z (2) in terms of za and zb.
The trajectories calculated in this way are shown in Fig. 2, in the typical
case m = 0*5. The instantaneous surface profiles at successive times are also
shown, corresponding to equal increments of / (J~l is shown as a function of
t in Fig. 3 of (1)). As can be seen, the profile at time t = 0 is a rectangular
hyperbola, and as time increases the angle between the asymptotes di­
minishes monotonically towards zero. A t the same time the principal axes
rotate clockwise through a finite angle.
The trajectories start with a marked component of velocity inwards into
the fluid, as the angle between the asymptotes diminishes. Later, as the
vertex angle tends to zero, the trajectory is determined more and more by
the ejection of the fluid to infinity, in the x -direction.
In Fig. 2 it has been assumed that the origin О is in a free-fall reference
frame. In a stationary reference frame, however, О moves in a parabolic
trajectory, with downwards acceleration g. T o transform the trajectories to
such a reference frame it is necessary to add to z the terms Uf +£gr2, where
U and g are (complex) constants representing a uniform velocity and the
acceleration of gravity respectively. Figure 3 shows examples, with ш = 0*3
to 2-0, in which U is taken as -1 *5 i and g as -1 , with A = 1 as before.
It will be seen that as w diminishes from 2-0 to 0-3 the total angle 8
through which the hyperbola turns gradually increases. As ш approaches
0-25, the angle у between the asymptotes “ hesitates” near 45°, and the total
angle 8 increases without limit, as in Fig. 1.

REFERENCES
1. M. S. L o n g u e t - H i g g i n s , Proc. R . Soc. A371 (1980) 453.
2. , J. Fluid Mech. 121 (1982) 403.
3. F. John, Comm uns pure appl Math. 6 (1953) 497.
4. M. S. L o n g u e t - H i g g i n s , J. Fluid Mech. 55 (1972) 529.
5. , ibid. 73 (1976) 603.
6. G. N. W a t s o n , Theory of Bessel Functionsy 2nd Ed. (Cambridge University Press,
1966).
1982

JOURNAL O F GEOPHYSICAL RESEARCH. VOL. 88. NO. С14. PAGES 9823-9831, NOVEMBER 20, 1983

Measurement o f Breaking Waves by a Surface Jump Meter

M . S. L o n g u e t - H ig g in s and N . D . S m it h

Institute o f Oceanographic Sciences

Quantitative information on the strength and size distribution of whitecaps in a given wave field is
very scarce. During the MARSEN field experiments, observations of surface elevation were made with a
capacitance wire wave recorder attached to a free floating spar buoy. Automatic analysis of the records
with a differentiating circuit and counter allowed a histogram of jump-heights to be constructed corre­
sponding to any preset critical rise rate R of the surface elevation. Over a certain range of R the
histogram was nearly independent of the precise value of R. This occurred usually when 0.6 < R/c0 < 1.0
where c0 was the speed of the dominant waves. Records were obtained in wind speeds ranging from I to
14 m/s. At 14 m/s the number of “jumps” indicating either steep or breaking waves was of the order of 1
every 100 wave periods. It is shown that this number is consistent with previous theoretical estimates,
and with visual observations of whitecap coverage. Because of the dispersive properties of waves, white-
capping in deep water is intermittent. Both theoretical calculation and laboratory experiments lead us to
expect steep or breaking waves to induce separated flow, with a high local input of momentum to the
wave field. It is concluded that such local events could contribute significantly to the total horizontal
stress exerted by the wind.

1. In tr o d u c tio n disc w ould leave un tu rn ed a sim ilar disc of sm aller diam eter.
B reaking waves in deep w ater are intim ately involved in a T hus the m ethod ap p aren tly was unreliable.
n um ber o f processes at the air-sea boundary, including the W e therefore resorted to the m ore direct m ethod described
horizo n tal stress exerted by wind [Banner and Melville, 1976], in section 2 below, in which the aim was to detect and m ea­
th e generation and dissipation of surface waves, the vertical sure the small ju m p s o r discontinuities in surface elevation
mixing in the upper ocean, the exchange of heat and gases associated with spilling o r plunging breakers. T h e instrum ent
(enhanced by vertical mixing), and the generation o f aerosols designed for this purpose is described in section 3. After suc­
by b u rstin g bubbles [ Blanchard and Woodcock, 1957]. cessful testing of a m odel in the 1980-ft (604 m) wave channel
O n the oth er hand, the frequency of occurrence of breaking at W orm ley, the p ro to ty p e was first deployed from the N ord-
waves and also their intensity have been little studied. C ertain wijk observation tow er off the D utch coast, as a p a rt o f the
descriptive properties are indeed im plicit in the Beaufort wind M A R SEN field pro g ram (see section 4). An analysis of the
scale. But the best know n quantitative inform ation seems to results will be given in sections 5 an d 6. W e also discuss the
be in term s o f w hitecap coverage [see, e.g., Monahan, 1971]. im plication of the results for w ind stress and wave generation.
D isregarding any possible subjectivity in the visual observa­
tions. this still gives only an indirect m easure of the dynam ical 2. P r in c ip l e of M easurem ent
characteristics of the breakers. M oreover, the scatter in the
d a ta when plo tted against local wind speed suggests strongly Suppose th at the elevation r\ o f the sea surface is m easured
th a t the local wind speed is not the only relevant physical as a function of tim e i a t an alm ost fixed h o rizontal position
param eter. (x, y), as in Figure 1. A progressive wave passing the recorder
T h ere is clearly a need for direct, instrum ental m easure­ will generally show a sm o o th rate o f rise dtj/dt. But if a break­
ments of b reaking waves to give quantitative inform ation of ing wave passes the recorder, we expect a sudden ju m p in
their physical properties such as the height of the “ju m p ” at surface elevation. F o r a plunging breaker (Figure la) this is
the surface, the distribution of w hitecap dim ensions, and the relatively large; for a spilling b reaker (Figure lb) it is smaller.
intensity of the accom panying turbulence. M oreover, these We rem ark th a t spilling breakers m ay start either as instabil­
q u antities need to be related to the local wave field and to the ities on the forw ard face of the w ave o r as small-scale plunging
tim e history of any relevant winds. breakers near the w ave crest, which then spread dow n the
As a first step in this direction, we have (since 1974) tried forw ard face [Longuet-Higgins and Turner, 1974].
various m ethods of m easuring breakers in deep w ater. O ne Suppose then th at the o u tp u t voltage V{t) from a capaci­
m ethod was to record their acoustical signature. However, the tance wire recorder is passed through a differentiating circuit,
m ain difficulty with (passive) acoustical detection is th at the as in Figure 2. W hen dV/dt cxceeds a certain threshold value
signal depends rather strongly on the distance from the source ц, say, a gate G triggers an integrating circuit I. This integrates
to receiver; thus under breaking wave conditions only the the signal dV/dt between the time t, when dV/dt rises through
m ean acoustical intensity can be easily interpreted. /j until the next instant i 2 when dV/dt falls below ц again. T he
W e also attem pted to calibrate breaking waves by their resulting “ju m p ”
ability to o verturn floating discs of different diam eters. O n the f'l av
whole, larger waves overturned larger discs. However, it was
found th a t occasionally the sam e wave that overturned a large
J = \ T idt~ K('2) “ K(t,) (1)
is then autom atically sorted into a histogram by the co u n ter
Copyright 1983 by the American Geophysical Union. C, which divides the range o f ju m p s into intervals with preset
divisions:
Paper number ЗС0481.
0 148-0227/83/003C-0481 S0S.00 0= -,J m (2)
9823
 1983

9824 L o n g u e t -H ig g in s a n d Sm ith : M easu rem en t o f B re a k in g W a ves

experiment in O ctober-N ovem ber 1979. F or the times of ob­

servation see Table 1.
The buoy was launched from a platform on the SE corner
of the tower (see Figures 4 a and 4b), and times of observation
were restricted to those periods when the tidal currents carried
the buoy aw ay from the structure tow ard the south.
D uring the first half o f the period of observation (September
27 to O ctober 15), preliminary results were obtained, with the
critical rate of rise R set arbitrarily. D uring the remaining
(a) PLUNGING
period the electrical o u tput K(r) from the capacitance wire was
recorded on m agnetic tape, so that the o u tp u t could be rerun
later a t various settings of ц or, equivalently, R.

4. A n a ly s is o f R e s u lts
Figure 5 shows a typical length o f record taken from run 2
on September 27, 1979. O n the trace below can be seen the
jum ps J as detected by the circuits o f Figure 2. T he scaJc of
the jum p trace is the same as th at of the record above.
T he way in which the histogram of ju m p heights J depends
Fig. 1. Illustration of (a) a plunging breaker and (b) a spilling
breaker passing a vertical wire recorder. on the critical rate of rise R can be seen, in a typical case, in
Figure 6. The vertical coordinate N, represents the total
num ber of jum ps, in a 10-min record, which exceed (i — 1)5.
T hus the upperm ost curve f = 1 in Figure 6a represents the
It is convenient to m ake the J , correspond to integer multiples total num ber of jum ps in the record; the next curve i = 2
o f a certain ju m p height Дft = 6, say. represents the total num ber of jum ps exceeding <5, and so on.
T h e resulting histogram depends clearly on the selected Clearly, as R increases, so N , decreases m onotonically. How-
values of \i and S, to an extent which we shall aim to m ini­
mize.

3. T h e M e a s u r in g A p p a ra tu s

A sketch o f the ap p a ra tu s is shown in Figure 3. The sensing

wire is carried on a vertical spar buoy of total length 5.04 m.
Buoyancy is provided by tw o dahn buoys near the center, and
angular stability is provided by a weight at the bottom . The
total weight is 120 kg. A central hollow tube piercing the
w ater surface gives vertical stability, and a directional fin a t­
tached to the central tube keeps the buoy in line either with
the current o r with the wind, whichever is the stronger. The
signal from the sensor is carried by cable to the recording
equipm ent on a ship o r stationary platform.
A scale m odel, of overall length 2.0 m, was first built and
tested in the 30-m wave channel at W orm ley. in mechanically
generated waves of period about 1 s. It was verified that the
critical rate of rise dn/dl = R could be set in such a way that
the jum p-detecting circuit responded only to spilling o r plung­
ing breakers and not to steep, nonbreaking waves.
T he full-scale apparatus, show n in Figure 4, was first de­
ployed from the N ordw ijk observation tow er 10 km from the
D utch coast, at 52°I6.4'N , 4°17.8'E during the M A RSEN field

0 > DIFFERENTIATING CIRCUIT G • GATE

1 • INTEGRATING CIRCUIT С • COUNTER

Fig. 2. Block diagram of the circuit for automatic measurement of

the histogram of jump heights.
1984

L o n g u e t - H ig g in s a n d S m ith : M e a s u re m e n t o f B r e a k in g W a v e s 9825

Fig. 4a. The buoy in horizontal position before launching

ever, for i Ss 2 there is a certain range of R over which each of analyzed in this way. T he given value o f R denotes the m id­
the curves for N , is nearly horizontal, in this case 6.5 point, approxim ately, of the corresponding range.
m /s < R < 9.0 ш/s. O ver this range the distribution is nearly Also show n in T able 2 is the m ean wind speed U , as record­
independent of R. ed by an anem om eter on the lower, a t a height of 20 m above
An exception is the curve for i = 1, which generally de­ m ean sea level; and the phase speed c0 of the do m in an t waves
scends m ore steeply than the others. This curve, which in­ in each record, th at is, the speed of a regular train of waves, of
cludes ju m ps of very small am plitude, will clearly be affected low am plitude, having the sam e num ber of zero crossings as
by any high-frequency noise, and we shall therefore disregard the actual record.
it.
T h e corresponding histogram o f ju m p heights is found by 5. I n t e r p r e t a t io n
tak in g the differences
H ow should we interpret the results of Table 2? C onsider
AN , = N l - N , . l (3) first the relation of R to c0. Figure 10 shows schem atically a
a t some particular value of R w ithin the chosen range. Figure regular wave progressing to the left. T he ap p aren t rate o f rise
6 b shows the histogram when R = 8.2 m/s, for example. of the surface at a given point is clearly
T he anaJysis of other typical records is show n in Figure 7-9.
In Figure 9 there is clearly no range of R over which the
curves N , can be considered as horizontal. In such a case we
conclude that there are no significant jum ps in the record. where с is the phase speed and Ax = cAf. But Aq/Ax = tan a,
Table 2 shows the results from all the records that were where a is the angle of inclination o f the surface to th e hori-
 1985

Fig. 4c. The buoy in position 60 m from the tower.

1986

L o n g u e t - H ig g in s a n d S m ith : M e a s u re m e n t o f B r e a k in g W a v e s 9827

TABLE 1. Times of Observation With the IOS Surface Jump Meter 18. T his show s a single occurrence o f w hat m ay have been a
From the Nordwijk Platform plunging breaker actually striking the sp a r buoy. In the re­
Wind Wind m aining cases we m ay sup p o se th a t the waves were on the
Run Date Time Speed, m/s Direction point of breaking.
T h e question o f the phase o f th e ju m p s relative to th at of
2 Sept. 27 1554-1614 8.0 240° the corresponding w ave m ust also be exam ined. It seems that
3 Sept. 27 1654-1714 7.5-7.0 240°
4 Sept. 28 in every case the ju m p s occurred on the forw ard face of the
1430-1450 4.0 25°
5 Sept. 28 1525-1543 1-2 25° wave, that is ahead o f the crest, as seen on the record. Some
6 Oct. 4 0830-0850 8.4 165° allow ance m ust be m ade for th e phase shift in the response of
7 Oct. 4 0915-0935 8.0-6.4 160° the buoy. T he appendix gives a simplified theoretical analysis.
8 Oct. 18 0920-0940 13.7 300° F ro m F igure 12b it appears th a t generally the total phase-lag
9 Oct. 18 1005-1035 13.7 300°
10 Oct. 18 1038-1115 14.2 300° {Фл + Фв) *s negative. T his im plies th a t th e phase of the re­
11 Oct. 19 1015-1100 11.4 220° sponse leads the phase of the surface elevation. At high fre­
14 Nov. 20 1100-1140 6.6 036° quencies the total phase lag tends to zero, so high-frequency
15 Nov. 20 1140-1200 5.2-4.3 039° “jum ps” are registered w ithout appreciable time lag, b u t their
16 Nov. 23 1230-1355 10.8 252° occurrence is delayed relative to a longer wave. In oth er words
17 Nov. 24 1330-1440 6.7-7.2 220°
a “ju m p ” is further dow n the forw ard face o f the w ave than
appears on the record. H owever, the to tal phase shift \фА
+ фв\ is not large, being less th an 30° for waves o f period 7 s
zontal. H ence
o r less (corresponding to phase speeds c0 of less th an 11 m/s).
R = с ta n a (5)
6. D is c u s s io n
or
T he last colum n o f T able 2 indicates th a t un d er the given
a = arctan (R/c) (6)
range of conditions the n um ber N o f ap p aren tly breaking
N ow in a regular progressive gravity wave, the m axim um waves in a 10-min record was typically between 0 an d 8.
inclination of the surface is 30.37° [see Longuet-Higgins and Is this result reasonable in term s of present know ledge?
Fox, 1977]. H ence the m axim um value o f R/c is tan Let ro denote the p ro p o rtio n of d o m in an t waves th a t have
30.37° = 0.586. Any progressive wave in which R/c exceeds whitecaps. Table 2 indicates th a t o n O cto b er 18, for example,
this critical value m ust be breaking. T hus it is interesting th at the wave period was aro u n d 6.0 s, so th at the n um ber of waves
in T able 2 all th e ratio s R/c0 lie som ew hat above this value. in 10 min was a b o u t 100. W ith N = 3 this gives w = 3
O u r conclusion m ust be that the jum ps registered occur x 1 0 " 2, which is within the range calculated theoretically by
mainly on the forw ard face of steep, unstable waves. They are Longuet-Higgins [1969].
n o t necessarily either w hite-caps or plunging breakers. N ever­ Again, if the foam from a typical “ju m p ” has a persistence
theless, in the absence of a com plete know ledge o f the local time T„, then the p ro p o rtio n of the sea surface covered by
dynam ics o f steep, progressive but uniform waves, it seems foam (the visual w hitecap coverage (VW C) will be
reasonable to suppose th at these are m ainly waves th a t are
VW C = ш Tp/T (?)
ju st a b o u t to break, it n o t actually breaking.
E xam ination of all the plots of N t versus R showed that in where T is the wave period. F o r salt water, Monahan and
only one case did N 2, N 3, ••*, not become zero when R > c0 Zietlow [1969] found Tp = 3.85 s; w hence V W C = 2 x 10- 2 .
(that is, when the angle o f inclination exceeds 45°). T he excep­ This value lies well w ithin the range of o bservations reported
tion is in F ig u re 6, for the run from 1005 to 1015 on O ctober by Monahan [1971, Figure 2] for a w ind o f 14 m/s.

FROM RUN No. 2.27 SEP 7fl

Fig. 5. (e) A typical length of record taken from nin 2 on September 27,1979. (b) The corresponding trace of jumps.
 1987

9828 L o n g u e t -H io g in s a n d Sm ith : M easu rem en t o f B re a k in g W a ves

18/Ю/79. 1005-1015

Fig. 6. (e) The number Ы, of jumps of magnitude h belween (i - l)£ and IS, as a function of the critical rate of rise Л,
during the 10-min interval 1005—1015 (local lime) on October 18. (fc) The corresponding histogram of jump heights;

AN,

хтггх

Fig. 7. As in Figure 6, for the interval 1035-1045 on October 18.

20/11/79 . П 0 0 -Ш 0

Дм,

ja

Fig. 8. As in Figure 6, for the interval 1100-1110 on November 20.

1988

L o n g u e t - H ig g in s a n d S m ith : M e a s u re m e n t o f B r e a k in g W a v e s

23/11/79 1 2 5 5 -1 3 0 5

An,

5 -

Fig. 9. As in Figure 6, for the interval 1320-1330 on November 23.

In spite o f the prelim inary n ature o f the observations it m ay when the w ave is near its p o int o f m axim um steepness. Since
be interesting to discuss the possible im plications o f this result the speed o f the w ave relative to its envelope is (c — ct) or
for tw o aspects o f air-sea interaction: the horizontal stress ab o u t i c , w hitecaps are seen interm ittently, with a frequency
exerted by the w ind and the input of energy to the waves. ab o u t half th a t of the waves themselves [see Donelan el. a i,
H orizontal stress. It was show n very clearly by Banner and 1972].
M elville [1976] th a t the occurrence of w hitecaps and short T he separation of the airflow m ust therefore be in term ittent
breaking waves n ear the crest o f a steep gravity wave tended also. N o precisely ap p ro p riate calculations of the airflow are
inevitably to induce a separation of the airflow over the wave. available, and it is reasonable to com pare the onset of sep ara­
M easurem ents indicated th a t the flux of horizontal m om en­ tion w ith som e recent calculations o f the flow started from rest
tum from the air to the w ater was thereby increased locally by over a steep but regular train o f w aves [ Longuet-Higgins ,
a factor of o rd er 50. T hough occasional separation of the 1980]. T he variation o f the d rag coefficient c D with tim e is
airflow m ay also occur when the waves are not breaking, show n in Figure 11; cD begins at t = 0 w ith a relatively high
nevertheless, it seems probable th at breaking waves induce value of 0.15 due to the form ation o f a circulating eddy behind
sep aratio n far m ore strongly and consistently. each crest. O nce the eddy is form ed, however, th e m om entum
T w o questions then arise: H ow frequently are breaking of the airflow changes less rapidly, an d the d rag coefficient
waves observed in the ocean, and w hat p ro p o rtio n of the total falls drastically to a level w hose m ean is far less, by perhaps
w ind stress can the local patches of breaking be expected to tw o orders o f m agnitude. T he high d rag is thus a m om entary
acco u n t for? phenom enon, lasting for a tim e TD o f o rd er L/(U — c), where L
W ith regard to the first question we note first th a t wave is the w avelength and U the w ind speed. Since the wave
b reaking is an essentially interm ittent phenom enon, especially period T is equal to L/c, it follows th at
in deep w ater. T his is due to the fact th a t surface waves are
T J T Ф c/(U - с ) (8)
b o th dispersive and irregular; high waves occur in groups,
w ith individual waves traveling through the group at a speed с which, in the case of the d ata o f Table 2, is of o rd er 1 a t least.
w hich is greater than the group velocity cr T hus a wave will T he contribution of steep waves to the to tal m ean d rag coef­
start a t the rear of a group, grow to its m axim um steepness as ficient is then of order
it passes thro u g h the group, and die aw ay as it reaches the
front of the group. A w hitecap will be seen only m om entarily сD' = wCmiU ~ c)2/c2 (9)
where c D0 denotes the initial d rag coefficient for th e high
waves. Conservatively, we m ay take, from F igure 11, 4=
TABLE 1 Analysis of Data 0.075. In table 2, for the d ata of O cto b er 18, (U - c)/c is of
u. order 0.3, giving cD' = 0.6 x 1 0 " 3. T his is sm aller th an the
Co R.
Date Time m/s m/s m/s N custom ary coefficient cD 4» 1.5 x 10“ \ b u t nevertheless o f the
sam e order of m agnitude.
Oct. 18. 1979 0922-0932 13.7 7.6 6.5 3
Oct. 18, 1979 0935-0945 13.7 9.2 8.0 1
Oct. 18. 1979 1005-1015 13.7 9.9 8.2 6
Oct. 18, 1979 1035-1045 14.2 10.4 8.5 7
Oct. 18. 1979 1055-1105 14.2 10.6 8.5 5
Oct. 19. 1979 1020-1030 14.4 9.5 _ 0
Oct. 19, 1979 1051-1101 14.4 7.9 6.5 5
Nov. 20, 1979 1100-1110 6.6 6.7 5.0 2
Nov. 20, 1979 1130-1140 5.2 5.2 _ 0
Nov. 23. 1979 1230-1240 10.8* 8.4 5.5 1
Nov. 23, 1979 1255-1305 10.8* 8.4 5.5 1
Nov. 23. 1979 1320-1330 10.8* 8.4 _ 0
Nov. 24. 1979 1335-1345 6.7 6.6 5.0 2
Nov. 24, 1979 1412-1422 6.9 6.7 5.0 2
TTTmT77rfrf1'
•Heavy rain. Fig, 10. Interpretation of the rate of rise R for a progressive wave.
 1989

L o n g u e t -H ig g in s a n d S m i t h : M easu rem en t o f B r e a k in g W aves

Hence
a = 2 n/TD = 0.598 rad/s

K/a = - In ~ - 0.63
n D

0)0/a = (I + \ K 2/a2) 112 - 1.048 (A9)

w0 = a(o»0/a) = 0.627 rad/s

T0 = 2 n/w0 — 10.0 s
-0.2 L
In forced oscillations, if Y denotes the elevation of the free
Fig. 11. The drag coefficient c0 as a function o f the time I for flow surface above its mean value, the restoring force is p ro p o r­
over a sharp-crested wave form o f steepness tf/L = 0.169, started
from rest at time ( = 0.
tional to (y — Y) and the frictional force is proportional to
(j) — f i t ), where p is a factor depending on the depth d of the
m ain source of friction (in ou r case the buoyancy chambers)
below the surface. In deep w ater we expect that
E quation (9) suggests also th at for the sam e values of ro and
Coo, thc con trib ution cD' will be a strongly increasing function (3 = e~u = е ~ " и * (A10)
of U/c. T his is in agreem ent with the ideas and observations
expressed by Longuet-Higgins [1969] and Kitaigorodskii where к is the wave num ber and w the radian frequency of the
[1973, p. 88]. (There is also experim ental evidence from the waves. W hen d is small com pared to the wavelength L = 2тс/к,
analysis of storm surges from hurricanes that the effective drag then p is nearly equal to 1. But if d > 0.25 L, then p is small.
coefficient should be increased at high values of U/c [see Whi­ From (A3) it can be seen that the (linearized) equation gov­
taker et a i, 1975].) erning the m otion m ust in general be
W e conclude that flow separation induced by steep waves
у + K (y — p V ) + <o0 J(y — У) = 0 (A ll)
could well m ake a significant contribution to the mean hori­
zontal wind stress. But the signal V (t) recorded by the spar buoy is proportional
to {Y — y) = if, say. F rom (A ll), writing у = Y — ij, we have
A p p e n d ix : Response o f a S p a r Buov
for t] the equation
A S im p lifie d T h e o r y

Theoretically, the vertical displacem ent у o f the buoy from r j + K t j + w024 = (A12)
its equilibrium position m ay be assum ed to be governed by
the equation

y+C\y\y + wo 2y = 0 (A l)

in free m otion, where С is a drag coefficient per unit mass, and

to0 is the natu ral (radian) frequency of oscillation. It is con­
venient to replace the dam ping term by a linear expression
Ky, w ith К chosen so th at total dissipation over one cycle is
the sam e, th a t is,

J K y 2 dt m I C|yiJ dt (A2)

T h u s K , though a constant, depends in fact on the am plitude

o f the oscillation. T he resulting equation of free oscillation is

у+ K y + o»0гу - 0 (A3)
W hen К < 2w0, this has solutions of the form

у = A e~ K'12 cos (ai + e) (A4)

w here

a = (a>0J-i/Cl),/J (A5)
an d A an d i are constant. T he “period" TD of the dam ped
oscillation is

TD = 2n/a (A6)

an d the atten u atio n factor over one cycle is

e- it w e *-«*/« = d (A7)

say. F o r the present buoy it was determ ined by experiment

th a t for a dam ped oscillation of initial am plitude A = 1.0 m,
Fig. 12. Calculated frequency response o f the spar buoy and jump
T„ = 10.5 s D - 0.14 (A8) meter (a) amplitude and (b) phase lag ф =» + фг
1990

L o n g u e t - H ig g in s a n d S m i t h : M e a s u r e m e n t o f B r e a k in g W aves 9831

H ence th e response to the sinusoidal oscillation shifts larger th an 30° occur only for w ave periods T greater
th an ab o u t 7 s.
Y = ае‘ш (A 13)

w here a an d со are real constants, is given by Acknowledgments. The authors are much indebted «о B. McCart­
ney for assistance in the design of the elecironic circuits used in the
П— (A 14) apparatus, to R. Spanhoff and his colleagues at the Rijkswaterstaat
for collaboration and logistic support on the Nordwijk tower, and to
provided О. H. Shemdin and his group at the Jet Propulsion Laboratory,
California Institute of Technology, for allowing us to use channels on
( - c o l + iKa> + (o02)be~‘* = [ - w 2 + iK (l - p)w ]a (A15) their magnetic tape recorder. Their work was carried out under con­
tract with the National Aeronautics and Space Administration. This
manuscript was begun in England and completed during a visit by
{oj02 — со2) + iK io
one of the authors (M.S.L-H.) to the Department of Engineering
{а/Ь)е‘* (A 16) Sciences at the University of Florida, Gainesville, Florida, during
- t o 2 + iK (t February 1981. Comments and discussions were provided by mem­
bers of his class.
and th e p hase lag ф will be given by

Kto K(1 - p) R eferences

(A 17) Banner, M. L, and W. K. Melville, On the separation of air flow over
water waves, J. Fluid Mech., 77, 825-842, 1976.
T h e beh avior of ф can be seen qualitatively from Figure Blanchard, D. C., and A. H. Woodcock, Bubble formation and modi­
12b. F o r very high frequencies, as w/u>0 —►oo, the argum ent of fication in the sea and its meteorological significance, Tellus, 9,
145-158, 1957.
b oth the n u m e rato r and d en o m in ato r in (A 16) tends to 180°, Donelan, М., M. S. Longuet-Higgins. and J. S. Tunner, Periodicity in
so ф —►0°. T h a t is, for sudden discontinuities o r “jum ps", r\ is whitecaps. Nature, 239, 449-451, 1972.
in-phase w ith Y (as we would expect). W hen ш > w0, ф lies Kitaigorodskii, S. A., The Physics o f Air-Sea Interaction, translated
between 0° and - 9 0 ° , so r\ leads the surface elevation У. from the Russian by A. Baruch, 237 pp., Israel Program for Scien­
W hen cu < co0, ф lies between —90° and —180°, as show n in tific Translations. Jerusalem. 1973.
Longuet-Higgins. M. S., On wave breaking and the equilibrium spec­
Figure 12b. Finally, as w -* 0, (A 10) shows that (1 — {}) is of trum of wind-generated waves. Proc. R. Soc. London Ser. A, 310,
ord er to2; hence ф -* — 180°. So for very long waves, is 180° 151-159, 1969.
o u t of phase w ith У. Longuet-Higgins, M. S., Polygon transformations in fluid mechanics,
F o r calculation it is convenient to write (A 17) as in Proceedings o f the 7th International Conference on Numerical
Methods in Fluid Dynamics, Stanford, California, June 23-24, 1980,
ф = фл + ф, (A 18) Springer, Berlin, 1980.
Longuet-Higgins. M. S., and М. H. Fox. Theory of the almost-highest
where wave: The inner solution, J. Fluid Mech., 80, 721-741, 1977.
Longuet-Higgins, M. S., and J. S. Turner, An “entraining plume"
_____ (К/а>0Хсо/ш0) model of a splitting breaker, J. Fluid Mech., 63, 1-20, 1974.
arctan --------- r — r- Monahan, E. C.. Oceanic whitecaps, J. Phys. Oceanogr„ /, 139-144,
1 - W 2 /(O0 2
1971
(A 19) Monahan. E. C., and С R. Zietlow, Laboratory comparisons of fresh­
_ (K/co0Xl - e - i “ 1'“0,)
an ; water and salt-water whitecaps, J. Geophys. Res.. 74, 6961-6966,
co/a>0 1969.
Nath, J. J.. and F. L. Ramsey, Probability distributions of breaking
and wave heights emphasizing the utilization of the JONSWAP spec­
trum, J. Phys. Oceanogr., 6, 316-323, 1976.
Я = dw02/g (A20) Whitaker, R. E., R. O. Reid, and A. C. Vastano, An analysis of drag
F ro m Figure 3 we m ay tak e d 3 m, and then from (A9) we coefficient at hurricane windspeeds. Tech. Memo 56, Coastal Eng.
Res. Cent., Fort Bel voir, Va.. 1975.
find Я = 0.120°. Also
M. S. Longuet-Higgins and N. D. Smith, institute of Oceanograp­
K/w0 = (X/eV(c«»0/e ) = 0.60 (A21) hic Saences, Wormley, Surrey, England.
T he angles фл and фв are show n in Figure 12b. Clearly, фв is
quite sm all (less than 7°) and the main contribution comes (Received August 17, 1982;
from фА. T he “resonant” period T0 in this case is 10.0 s. Phase accepted March 21. 1983.)
 1991

Proc. R. Soc. Lond. A 396. 269-280 (1984)

Printed in Great Britain

On the stability o f steep gravity waves

B y M. S. L o n g u e t -H iggins , F.R.S.
Department of Applied Mathematics and Theoretical Physics, University of
Cambridge, Silver Street, Cambridge CBS 9EW , U.K.
and Institute of Oceanographic Sciences, Wormley, Godalming,
Surrey GUS 5UB, U.K.

(Received 9 July 1984 - Revised 12 November 1984)

Previous calculations o f the normal mode perturbations of steep gravity

waves have suggested that the lowest superharmonic mode n = 2 becomes
unstable at around ak = 0.436, where 2a is the crest-to-trough height o f
the unperturbed wave and к is the wavenumber. This would correspond
to the wave steepness at which the phase speed с is a maximum
(considered as a function o f ak). However, numerical calculations at such
high wave steepnesses can become inaccurate. The present paper studies
analytically the conditions for the existence o f a normal mode at zero
limiting frequency. I t is proved that for superharmonic perturbations
such conditions will occur only for a pure phase-shift (corresponding to
n = 1) or when the speed с is stationary with respect to the wave
steepness, that is when dc = 0. Hence the limiting form o f the instability
found by Tanaka (J. phys. Soc. Japan 52, 3047-3055 ( 1983)) near the
value ak = 0.429 must be a pure phase-shift.

1. I n tr od uc tio n

Since the pioneering work o f Whitham ( 1967) and Benjamin & Feir ( 1967) it has
been known that gravity waves on water o f infinite depth are unstable, at least
to certain subharmonic perturbations. Precise calculations o f the normal modes
by the present author ( 1978 a) and by McLean ( 1982) have shown also the existence
o f some unstable perturbations with length scales shorter than the basic
wavelength.
While many o f the instabilities are three-dimensional, some important and
interesting examples occur even in two dimensions (one horizontal and one
vertical). Thus it was shown in Longuet-Higgins ( 1978a; hereinafter referred to
as I ) that some superharmonic disturbances, with a dominant wavelength one half
that o f the original wave (i.e. having wavenumber n = 2 ) tend to become unstable
when the steepness ak o f the unperturbed wave exceeds about 0.43. Although the
calculations could not be carried accurately beyond about ak = 0.42, nevertheless
they were consistent with the suggestion (made on physical grounds) that a
transition to instability occurs at the wave steepness ak for which the phase speed
с is a maximum. From independent calculations (Longuet-Higgins & Fox 1978)
this was known to be at about ak = 0.436 (less than the maximum steepness
ak = 0.4434).

[ 269 ]
1992

270 M. S. L o n gu et-H iggin s

Recently, however, Tanaka ( 1983 ) has proposed a different method o f calculation,

claiming to have overcome the difficulties inherent in using Fourier series
expansions at high values o f ak. His numerical results agreed closely with those
o f I up to ak = 0.41, but then tended to diverge, so that instability o f the mode
n = 2 appeared to occur at around ak = 0.429. This corresponds closely to the
steepness for which the total energy density E is a maximum. H owever, no physical
reason for this coincidence could be found.
Since numerical calculations, however accurate, often lack the certainty o f
mathematical analysis, the present author sought an analytical approach to the
problem. An opportunity came with the discovery that by making use o f certain
identities between the coefficients an in the Stokes series for deep-water waves it
was possible to simplify considerably both the expression o f some integral
quantities (Longuet-Higgins 1984a) and the numerical calculations o f finite-
amplitude gravity waves and their points o f bifurcation (Longuet-Higgins 19846).
In this paper we show that the analysis o f the normal-mode perturbations can be
simplified also.
In brief, we shall show that if there exists a normal instability with radian
frequency cr tending to zero at a certain wave steepness ak, then at the critical
point either (i) the phase speed is stationary: dc = 0 , or (ii) the limiting form o f
the perturbation is a pure phase-shift. The points at which dc = 0 and dE = 0 are
quite distinct (see Longuet-Higgins 19846). Therefore if a zero-frequency normal
mode occurs at dE = 0, its limiting form must be a pure phase-shift.
The present paper further illustrates the utility o f a matrix analysis for gravity
waves on deep water, which in turn is based on the quadratic relations ( 3 . 12) for
the Fourier coefficients an. Although the analysis applies strictly only to waves
in two dimensions, it would seem worthwhile to investigate the possibility o f a
three-dimensional analogue.

2. E q uatio ns for no r m a l modes

In any two-dimensional, irrotational flow o f an inviscid, incompressible fluid,

the rectangular coordinates (x,y) may be expressed as functions o f the velocity
potential ф, the streamfunction \\r and the time t. In the interior o f the fluid ( x + '\y)
is everywhere an analytic function o f (0 + i^r), and at a free surface, where

f = f (? U ). <2 J >

the two boundary conditions expressing that the pressure p is constant and that
a particle at the surface remains at the surface become respectively

-(УфУ1+У*Х1)+ЯУ(Уф + У \ )+ \ = ® <2'2*

"a (УфУ1-УфХ,) + [1 -(У ф У ,+ У ф Ъ )']*'ф + (У ф +У ф )Р1 = 0 <2-3)

(see I, §2). Here suffixes are used to denote partial derivation. In ( 2 .2 ) g denotes
gravity, the у -axis being taken vertically upwards. W e shall choose units so that
0=1-
 1993

On the stability of steep gravity waves 271

We seek solutions to these equations in the form
x = Х{ф, xjr) + е£{ф, xjr) e-i<TM
y = У{ф,\1г) + ег}{ф,\1г)е-ш \ (2.4)
F = е/(ф) е~ш , )
in which (X, Y ) represents a steady, progressive wave travelling with speed с in
the positive ж-direction, as seen in a frame of reference travelling with the waves.
The flow then appears steady. The terms e(£,^,/) represent a small perturbation
of the steady wave, varying harmonically with the time t.
We now substitute (2.4) into (2.2) and (2.3) and expand each of the left-hand
sides in a Taylor series in \j/ about xjr — 0 (the unperturbed surface). From the terms
independent of e we obtain
2У(У*+У£)+1 = 0 (2.5)
and Fq = 0 (2 .6 )
to be satisfied when i/r = 0. For waves in deep water we must also have
Y ~ -cf (2.7)
as \jr-+ oo.
Assuming these satisfied, the terms in e then give us the following equations for
the perturbations:
- i< r ( 7 ^ + УфТ}) = Ру + <2ч+ + ЯУф+в/, (2 .8 )
- i < r ( Y ^ - Y + V) = f t , (2.9)

where we have written

P = Y \+ Y ly
Q = (Y*)+ = 2YY<„
(2 . 10)
Д = ( П , = 2УТ*,
S = [ Y ( Y l + Y%)\ = Y f P + ¥ ( Ц + Ц)+,)
all to be evaluated on ijr = 0. In addition
£ + i ^-»*0 as xjr-xx) (2 . 1 1 )
(see I, §3).

3. T he basic wave

In the unperturbed flow, F = 0 by ( 2. 4 ) so th at ( 2 . 6 ) is already satisfied. T o

satisfy ( 2. 5) and ( 2. 7 ) we assume a steady, symmetric wave given by the Stokes
expansion oo
(y -iJT ) + ( ^ - i ^ ) / e = K + 2 вя е*»<*+1*>/« (3.1)
П -1

in which the coefficients an are all real, and units have been chosen so that the
1994

272 M. S. L onguet-H iggins

wavelength Л equals 2n. As so the right side of (3.1) tends to a constant
Ja0. On the unperturbed surface [i/r = 0) we have, from the real part of (3.1),

Y = £ * am cos mQ (3.2)
TO-0
where в = ф/с (3.3)
and an asterisk means th at whenever a 0 occurs in the sum it is to be replaced by
Ja0. F urther oo
—cYj, = £ bn созпв,
n-o
00 (3.4)
- с У ф = £ bn sin пв,

where К 1» bn — 7ia,nt ti — 1, 2 , 3 ,.... (3.5)

Hence, when i/r = 0,
c2{ Y \ + Y \ ) = £ £ bnbm c o s (m - 7i) 0 , (3.6)
n-o m- 0
oo
th a t is c2P = E * 2/} cos Ю, (3.7)
i-o

where Pi = 2 6m+I6m, / = 0,1,2..... (3.8)

TO—
о
The boundary condition (2.5) can be written
2 c 2 YP —— c 2. (3.9)
On substituting for Y and c2P from ( 3 . 2) and ( 3. 7 ) we have
00 00
2 £ * E * ат Pz[cos (ra + l) в + cos (тп—1)в] ——с2 (3.10)
то-о г-о
and on equating the coefficients of coswtf, * * 0 , 1 , 2 ...... on each side we obtain
the infinite set of equations
а 0 ^ о + (а 1 + а 1 ) Л - М ° 2 + а 2 ) Р 2 + (а з + а з ) Р з + - - - = - с 2 ,'

a \ ^ 0 + ( а 0 + а 2 ) ^ 1 + (а 1 + а з ) ^ 2 + (а 2 + а 4 ) ^ 3 + - - - =
(3.11)
а 2 ^ 0 + ( а 1+ а з ) ^ 1 + ( а 0 + а 4 )/ 2 + ( а 1+ а 5)^ 3 + -* - =

between c, and the original coefficients an.

Formally, equations (3.11) are cubic in the an. I t is a remarkable fact, first noted
explicitly in Longuet-Higgins ( 1 9 7 8 6 ), th at these equations are equivalent to a
simpler, quadratic set of equations, namely
ao b0 + a1b1+ a2b3+ az b3+ ... = - c 2,>
® i ^0 a o “ b a i b 2 ■+■ ^ 3 “ 1" • • • =
(3.12)
a2b0 + a16j + a0 b2+ al 63 + ... = 0,
 1995

On the stability of steep gravity waves 273

For, if we denote the left-hand sides of equations (3.12) by Fn, n = 0 , 1 ,2, and
the corresponding expressions in (3 . 1 1 ) by On then we have identically
b0 F0+ 6 j Fx+ b2 F2+ 6 3 Fz + ... = G0,
60 Fx+ bxF2+ 6 2 Fz + . . . = Glt
(3.13)
b0 F2+ bxF3+ ... = G2,

th at is to say B x F = G, (3.14)
bo bx b2 6 3
0 b0 bx b2
where B = (3.15)
0 0 60 b,

and F — (F0,FlyF2, ...)T,

(3.16)
G = (G0,GlyG2, ...)T.
Since 60 = 1 we have |B | = 1 (3.17)
and the matrix В is non-singular. Hence any non-vanishing solution to (3.11) Hs
equivalent to a non-vanishing solution to (3.12) and vice versa.
The above proof, which was given in Longuet-Higgins ( 1 9 7 8 6 ), relies on the
convergence of all the series involved, which is, however, assured at all interior
points of the fluid domain. An alternative proof due essentially to J. G. B. Byatt-
Smith, is given in the same paper (see also Longuet-Higgins 1 9 8 4 0 , 6 ).

4. T h e n o r m a l m o d e s ; l i m i t a s <r->0
To solve the perturbation equations (2.8) and (2.9) it was noted in I that (f, rj)
being conjugate functions, tending to 0 as i/r/c-^ао, may be expanded in the form

0?-i£> = E + (4.1)
n- 0
and th a t in general ^
/ = S ( r n + ii„ )e ‘" ^ , (4.2)
71—0
where a n, fin>y n and $n are real constants. Substitution of these expressions into
(2 .8 ) and (2 .9 ) and equating the coefficients of cosn# (n = 0 , 1 , 2 ,...) and sinn#
(n = 1 , 2 , 3 ,...) yields a system of equations for the constants л п,/Зп, y n, Sn and the
radian frequency or, which can be solved numerically by successive truncations.
However, we are here interested chiefly in the limiting case when <r-*0. From
(2 .9 ) we see immediately th at in the limit
Л - 0, / —Уо- (4.3)
Equation (2 .8 ) then reduces to
Pv + QVf + ^ + s V 0 = 0, (4.4)
to be satisfied when \jr = 0.
1996

274 M. S. Longuet-Higgins
We rem ark th a t this equation can be derived directly from the steady boundary
condition 2y{yl + y \ ) = - 1 on xjr = F (4.5)
(compare (2.5)) by writing
y = Y + e T j } F — ey0 (4.6)
and then considering the coefficient of e.

5. N ormal modes (c o n t in u e d )
To proceed further with (2.8) we must evaluate Q, R and S. From (2.10) and
(3.4) we have oo oo
—cQ = £ * E a n 6n[cos(m*f n )e + cos(m —n)6], (5.1)
n —0 TO-0

—c R = X* £ a n 6m[sin(m + 7i)0 + sin(m —n)6], (5.2)

n-o m-0

th a t is - c Q = £ Qt cos Id, (5.3)

о

—cR — £ Ri sin 16, (5.4)

where Q0 = £ am6m (5.5)

m-0
and when I > 0,
г-i да
Ql = 2 ai-m bm+ E {ambi+m+ a l + m ^m)>
m -0 m -o
( 5.6)
t-1 oo
R t = S a i - m &m + ^ (a m ^ Z - m — a l + m ^ m ) '
m-0 m-0
Therefore when I > 0,

Ql + R l = 2 E ttf-m^m+ 2 E a m fy+ж —0
m-0 m-o
by (3.12). Hence for I = 0 ,1 ,2 ,3 ,... we have

Qi= E az+m6TO (5-8)

m-0
and in particular
Qo= 2 am bm = - c \ (5.9)
m-0
Thirdly we have from (2.10)
S = Y ^ P + Y M Q + Y H R. (5.10)
Hence
00 00
—c3£ = E bm cos тд E * 2Pn cos
m-o

+ £ E rabm($n cosm0 cosn6 + R n smm6 sinn0). (5.11)

m-o n -0
 1997

On the stability of steep gravity waves 275

On making use of the relations (5.7) we obtain

—c3S = 2 2* bm Pn[cos{m + n )e + co8{m—n)6]

m -o n -o
OO 00
+ 2 2 ™bmQn c o s(m + n )6 (5.12)
m -o n -0

= 2*25, cos W, (5.13)

1-0
say
Now adopting the expansions (4.1) and (4.2) we have when \jr = 0
0° \
V = 2 ( a n cos п в —/Зп sinn0),
о
00
—cij# = 2 (wan cosntf — sinn0) (5.14)
о
oo
—сг}ф = 2 {nan sin п в + п р п cosn#)

On substituting these expressions into the boundary condition (4.4) and equating
to zero the coefficients of J, cos#, cos 26, ... and sin#, sin 26, ... we obtain a linear
system of equations for
a = (aQ, a l ta 2, ...),]
(5.15)
= [fiv fit ’ ■••) J
and y0, which may be written in the form

' M 0 S
i
x ( oe ; p ; To )
т _ (5.16)
0 N i 01

where
(Po + ^o) (Л + ^i) (^2 + П) (P3 + P3 )
(Pi + Pi) (P0+ P2+ Q0) (Pi + P>) (P* + P<1)
M = (P2+ P2) ( P ^ P t + Q j (Pq+ Pa+ 2Qq) (/> + />) (5.17)
(P3 + P3) (P2 + ^ + 0 2) (P1 + P5 + 2«1) (P0 + Pe + 3Q0)

'(Po-Pt + Qo) (P1 - P 2) (Рг ^4 )

and ЛГ = [ P t - P t + Qi) (P0- P 4+ 2^o) (Р -Р ь)
(р0- р 6+ з а д
(5.18)
( A - J i + e .) (P i-P b + W i)

and О denotes the zero square matrix, 0 a zero column vector. Finally
S = ( S 0, S l ,S 2,...)T (5.19)

where S n is given by (5.13).

1998

276 M. S. L on guet-H iggins

6 . C o n d i t i o n s w h e n <t = 0
We shall make use of the following lem m as:
M = BxC, (6.1)
N = В x £>, (6.2)
in which В denotes the m atrix (3.15) and C ,D are the symmetric matrices:
2 ((1,4*0.,)
(aj + aJ (2a24-2a2) (Зл34-Зй3)
(1 4- a04- 2a2) (2^14" 3a3) (3a24-4a4)
C = (2a24- 2a2) (2a14-3a3) (1 4- 2a04-4a4) (З ^ -Ь б ^ ) (6.3)
(3a34-3a3) (3a24-4a4) (3®! 4* 5as) (14- 3a04-6ae)

and
(l4 -a 0- 2 a 2) (2«! —3a3) (3 a 2 —4 a 4)
(2aj —3a3) (1 + 2 a 0—4a4) ( 3 a j - 5 a 5)
£> = (6.4)
(3a2—4a4) (3a2—5a6) (1 4 - 3 a 0 — 6 a e )

Moreover, M x b T = S, (6.5)
N x b'T = 0, (6.6)
where S is given by (5.19) and
* “ (боА Л . « 0 , ( 6 -7 >

b' = (61,62,63,...). (6.8)

These results are all proved in the Appendix.
Consider the consequences. Equation (6.5) implies th a t the last column of the
m atrix in (5.16) depends linearly on the columns of M, and hence is redundant.
The value of y0 is thus arbitrary. This is not obvious a priori, since although a
non-zero value of y0 implies the addition of a mere constant to the streamfunction
\jr, it also changes the position of the free surface, to order e, hence the values of
the coefficients an in the Fourier expansion (3.1).
On omitting the last column of the matrix, (5.16) reduces to the ‘square’ form
м iо
— -!........I * I — - ) = I ......... I (69)
o ; n / \ r ) \ 0

This again reduces to the two independent systems

M x a T= 0 (6 . 10)

and N x p = 0 ( 6 . 11)

for the in-phase and quadrature components of the perturbation (4.1).

Consider first the symmetric perturbations. The system (6.10) can have a
non-vanishing solution only if | M | = 0. But by (6.1)
\ M \ = | B | x | C | = |C | (8.12)
 1999

On the stability of steep gravity waves 277

since | В | = 1 by (3.15). Therefore for a non-vanishing solution we require
| C | = 0. (6.13)
Now it was shown in Longuet-Higgins ( 1 9 8 4 6 ) and can indeed be seen from
equations (3.12), th a t the vanishing of | С | is the necessary condition that the phase
speed с be stationary with respect to changes in the wave amplitude, that is
dc =s 0. (6.14)
This then shows th at there can exist symmetric normal perturbations at zero
frequency only if dc = 0 .
Consider on the other hand the asymmetric perturbations. From (6 .6 ) it follows
that there always exists an antisymmetric perturbation, given by
fi= b '. (6.15)

This represents a simple phase-shift of the original wave through the horizontal
distance ec.
Now (6.11) implies also th at the determinant of N vanishes. Hence the first row
of N is linearly dependent on the others. The corresponding equation is thus
redundant and may be replaced by a condition on the phase, for example that
A = 0. (6.16)
The modified system then becomes
N' x ft'1 = 0, (6.17)

where fi' = [/32,J33, ...) and N' is the matrix derived from N by omitting the first
row and column, th at is
(Po-P* + 2Qo) « -/» -
( /> - /> + 2 0 0 (/>0- ^ e + 3<20) (6.18)

Thus an antisymmetric perturbation exists only if

|ЛГ| = 0 . (6.19)

Now, as in the proof of (6.1) given in the Appendix, it may be shown that
N' = B x D \ (6.20)
where D ' is the matrix derived from D by omitting the first row and column, that
is
(1 + 2a0—4<z4) (3oj 5a5)
D' = (3ах- Ь а ъ) (1 + 3a0 —6 ae) ( 6 . 21 )

B ut it has already been shown numerically th at | D' \ does not vanish anywhere
in the range of uniform waves (see Longuet-Higgins 1 9 8 4 6 ), and particularly not
2000

278 M. S. Longuet-Higgins
when E = E mgLX. Hence }ЛГ | does not vanish anywhere in the range, and there are
therefore no antisymm etric normal modes a t zero frequency apart from a pure
phase-shift.
Since any perturbation to the original wave form may be expressed as the sum
of a symmetric part and antisymmetric part, this shows th a t there are no other
antisymmetric normal modes of any form a t zero frequency, except a t points where
the phase speed has a stationary value, i.e. when dc = 0 .
The form of the normal mode, a t points close to a point of stationary phase speed,
is presumably given by
a = [a0, a lfa2, ...] = a(c) + e(Aa + /a/f) + 0(e2), (6 .2 2 )
where a <c) denotes the values of the Fourier coefficients a t the critical wave
amplitude, a and p correspond to solutions of (6.10) and (6.11), and A, fi are
constants. To determine the ratio A\ц and the corresponding (non-zero) value of
the radian frequency cr the full equations (2.8) and (2.9) may be carried to higher
order in e.

7. D iscu ssio n

We have shown, essentially, th a t for irrotational waves on deep water the

conditions for a normal-mode perturbation a t zero limiting frequency are the same
as those for a bifurcation of the steady motion. B ut it was already shown previously
(Longuet-Higgins 1 9 8 4 6 ) th a t the only possible 4 -1 bifurcation consists of a pure
phase-shift, except possibly a t a stationary value of the phase speed, th a t is when
dc = 0. I t follows th a t if there exists a superharmonic normal mode with zero
limiting frequency a t any value of a к other than dc = 0 , then its limiting form must
be a pure phase-shift.
This conclusion applies in particular to the numerical result of Tanaka ( 1 9 8 3 )»
who found a zero-frequency mode a t around ak — 0.429.
In a recent correspondence Dr Tanaka has kindly informed the author th a t the
limiting form of the instability found by him is indeed a pure phase-shift. However,
it is still not clear why this situation should occur so close to the point a t which
сIE = 0. A further investigation of the question is in progress.

(6.1), (6.2), (6.5) a n d (6.6)

A p p e n d i x A. P r o o f o f e q u a t i o n s
To prove (6.1) let the matrix C, defined by (6.3), be written in the apparently
asymmetric form
(60 + 60) (04-61+ a 1) ( 0 + 6 24 - 2 a 2) (04-634-3a3)
(b i+ b j (60+ 62+ a 0) (0 4- 63 + 2ax) (0 4- 64 4- 3 a 2)
C= (62 4 - 6 2) {bl + b3+ a l) (604-644-2a0) (0 4 - 654 - 3 ^ ) (Al)
(6 3 + 6 3 ) (62+ 64+ a 2) (614-654-2a1) (b0 + 4- 3 a 0)

On multiplying the ith row of В (see equation (3.15)) by the jth column of (A 1)
and using the definition of Pt in (3.8), we see th at the terms in 6n alone yield the
 2001

On the stability of steep gravity waves 279

inner product (/^_д + Рщ). On the other hand, from equations (3.12) and (5.3)
the terms in b{ and yield 0 if i < j , and (j —i)Q i 4 if i > j . This proves (6.1).
To prove (6.2) let us define
-(бо —60) (0—b1+ a1) ( 0 - 6 2+ 2a2) (0 —63+ 3a3)
(b r-b j (b0~ b 2+ a0) (0 —63+ 2a,) (0 —64+ 3a2)
(b2- b 2) ( ^ - Ь з + aJ (60- 6 4 + 2a0) (0 —65 + 3 0 ^
—
(6 3 )
6 3 (62—64+ a2) (6i —66+ 2ai) (b0- b 6+ 3a0)

(A 2)
Then it is clear th at
D[ = D (A 3)
(where a prime denotes the matrix derived from a given matrix by omitting the
first row and first column). As before, we now have
B x D l —N t, (A 4)
where
(Po-Po) (Рг-Рг + 0) (P2- P 2 + 0) (Рз-^з + 0)
(Pi~ Pi) (Po-P* + Qo) ( P i-P * + 0) (Рг-Р* + Ъ)
(Рг ~Рг) (Pi - Р з + Qr) (Po-P* + 2Qo) (Pi-Pb + 0)
(P* ~~P3 ) (Pi~~Pi + Ф2) (P i-P b + W i) ( / > - / > + з д 0)

(A 5)
B ut in D 1 the elements of the first row all vanish. So from (A 4) we have
B' x D\ = N[ (A 6)
and since В' = В (see equation (3.15)) this is (6.2).
To prove (6.6) we note that since the elements in the first row of N t all vanish,
it follows from (A 4) that
b' x D = 0T (A 7)
and on taking the transpose matrix of each side of this equation we obtain
D x b ' T = 0, (A 8)
D being symmetric. We now multiply on the left by B, using the commutative
law for matrices, to obtain
( B x D ) x b'T = 0 (A 9)
and (6.6) follows, by equation (6.2).
Lastly, to prove (6.5) note first that from the definition of St in (5.13) we have
^ = (^ + ^o)^o + № + ^ + 0)fti + № + P2 + 0)62+ . . . ,
Si = (JP1+ / >i)60+ (/i + ^ + ^o)6i + (A + ^3 + 0)62+ . . . 1 (A 10)
^3= ( P 2+ P2) 604-(P1-fP3+ <?1)61+ (P0+ P4 + 2 g 0)62 + . . . >
280 M. S. Longuet-Higgins
From (5.17) and (6.6) it is clear th a t St is the inner product of b with the (I —l)th
row of M. This proves the lemma.

This paper was begun while the author was visiting Cal. Tech. J e t Propulsion
Laboratory, Pasadena. These and other results were presented at the Workshop
on Surface Gravity Waves a t U.C. Santa Barbara on 31 May-1 June 1984; also
a t a meeting a t the University of Bristol on 3 July 1984. To Dr M. Chahine, Head
of the Earth Sciences Division of J.P .L ., and to Professor M. Tulin, U.C.S.B., the
author expresses thanks for their hospitality and assistance.
I am indebted to Dr P. A. Saffman and Dr M. Tanaka for interesting corre­
spondence and discussion.

R eferences
Benjamin, Т. B. & Feir 1967 The disintegration of wave trains on deep water. J. Fluid Mech.
27, 417-430.
Longuet-Higgins, M. S. 1975 Integral properties o f periodic gravity waves o f finite amplitude.
Proc. R. Soc. Lond. A 342, 157-174.
Longuet-Higgins, M. S. 19 7 8 a The instabilities of gravity waves of finite amplitude in deep
water. I. Superharmonics. Proc. R. Soc. Lond. A 360, 471-488.
Longuet-Higgins, M. S. 19786 Some new relations between Stokes’s coefficients in the theory
of gravity waves. J. Inst. Maths. Applies. 22, 261-273.
Longuet-Higgins, M. S. 198 4 a New integral relations for gravity waves of finite amplitude.
J. Fluid Mech. (In the press.)
Longuet-Higgins, M. S. 19846 Bifurcation in gravity waves. J. Fluid Mech. (In the press.)
Longuet-Higgins, M.S. & Fox, M. J. H. 1978 Theory of the alm ost-highest wave. Part 2.
Matching and analytic extension. J. Fluid Mech. 85, 769-786.
McLean, J. W. 1982 Instabilities of finite-amplitude water waves. J. Fluid Mech. 114,315-330.
Tanaka, M. 1983 The stability of steep gravity waves. J. phys. Soc. Japan 52, 3047-3055.
W hitham, G. B. 1967 Variational methods and applications to water waves. Proc. R. Soc. Lond.
A 299, 6-25.
 2003

Proc. R. Soc. Lond. A 403, 167-187 (1986)

Printed in Great Britain

B ifu rcatio n and in stab ility in g r a v ity w aves

B y M. S. L onguet-H iggins, F.R.S.

Department of Applied Mathematics and Theoretical Physics,
University of Cambridge, Silver Street, Cambridge CBS 9EW , U.K.
and Institute of Oceanographic Sciences, Wormley, Godalminq,
Surrey GUS 5 UB, U.K.

(Received 1 August 1985)

Calculations of the two-dimensional normal-mode perturbationsofgravity

waves on deep water (Longuet-Higgins, Proc. R. Soc. Lond. A 360 ,
471-488 (1978a) ; Longuet-Higgins, Proc. R. Soc. Lond. A 360 , 489-505
(19786)) are here extended to values of the wave steepness ak as high as
0 .43 . This is achieved (a) by using a new method to evaluate the
coefficients in Stokes’s series to high order, and (6) by rearrangement of
the matrix equations so as to reduce the order by half.
The behaviour of the normal-mode frequencies c n in the range
0.35 < ak < 0.43 is clarified. Subharmonics of the form n = (l/m, 2 — l/m)
where I and m are integers and I < 2m are shown to combine in pairs to
form type II instabilities with relatively high rates of growth. For these
modes, the critical values of ak at which a vanishes correspond precisely
to bifurcation points. In the special case l/m = J the two modal curves
coincide.
The family of frequency curves is bounded by the lowest superharmonic
(71 = 2 ). It is verified that this mode becomes unstable when ak = 0 .4292,
corresponding to the lowest maximum of the energy density E. The
boundaries of the parameter regions for instabilities of both type I and
type II are determined.

1. I n tro du ctio n

Steep, irrotational gravity waves on deep water display in simple form some
striking nonlinear wave phenomena. These include the presence of local maxima
in the phase speed с or the energy density E (Longuet-Higgins 197s); the
bifurcation of regular, uniform waves into steady wavetrains of non-uniform
amplitude (Chen & Saffman 1980; Saffman 1980; Longuet-Higgins 1985 a ) ; and the
existence of subharmonic instabilities (Benjamin 1967; Longuet-Higgins 1978 a, 6;
McLean et al. 1981; McLean 1982). All these phenomena are already present in the
analytically simpler two-dimensional waves, as well as in more general three-
dimensional wave motions.
The first accurate calculations of the normal modes of perturbation for gravity
waves of finite amplitude were made by Longuet-Higgins (19780,6) and were
confirmed by Longuet-Higgins & Cokelet (1978). These revealed the existence of
two types of subharmonic instability. The type I, or Benjamin-Feir, instabilities
were present, generally in the range 0 < ak < 0 .41 , though some modes returned

7 [ 167 ] Vol. 403. A (8 February ig86)

2004

168 M. S. L onguet-H iggins

to stability at around ak = 0.35 as predicted by Lighthill (1967). (Here a denotes
the wave amplitude and к denotes the wavenumber of the unperturbed wave.) A
new, type II, instability with a much faster rate of growth arose at around
ak = 0 .40 . This was shown by McLean et al. (1981) and McLean (1982) to be a
special case of a more general three-dimensional instability, which dominates over
type I when ak > 0 .32 .
For practical reasons, the earlier method of calculation could not be performed
accurately beyond ak = 0 .41 . This left unresolved certain questions, particularly
as to the behaviour of the normal modes near and beyond the points where the
frequency <тп of the normal modes passed through zero. Such points will generally
correspond to the critical values of ak at which a uniform train of wave bifurcates
into a steady but non-uniform train of waves (Saffman 1982; Longuet-Higgins
1985 a). However, an interesting exception occurs in the neighbourhood of the
energy maximum (di£ = 0 ) at ak = 0 .429 , at which point the lowest superharmonic
(n = 2 ) becomes unstable (Tanaka 1983) but there is no bifurcation, apart from a
pure phase shift (Longuet-Higgins 1984, 1985a; Тапака 1985; Saffman 1985).
The purpose of the present paper is to carry the calculation of the subharmonics
and of the bifurcation points up to wave steepnesses as high as ak = 0.43 and to
relate these to the lowest superharmonic mode, n = 2.
Several critical values of the wave steepness ak for bifurcation into wave trains of
class m /l (in which the horizontal period ism waves long) were given approximately
by Saffman (1980, see table 1). However, some of these are very close together.
In §3 we use the slightly different method of Longuet-Higgins (1985 a) to achieve
higher accuracy. The results (see table 1 and figure 1), show that the critical
steepnesses lie rather precisely on a parabolic curve, symmetric about m /l = 2.
In §§4-6 we turn to the general calculation of normal-mode frequencies. We
succeed in extending the range of ak as far as 0 .43 , (a) by calculating the Fourier
coefficients to much higher order, with the use of the method described in
Longuet-Higgins (1985 a, 6) ; and (6) by using a suggestion of K. Tanaka (personal
communication, 1984) which has the effect of halving the order of the matrices
encountered in the calculation. After testing the method against Tanaka’s own
calculation of the superharmonic mode n = 2 (see §5), the method is applied, in
§6, to the subharmonic modes. It proves possible to delineate accurately the
frequency curves, and to show how in most cases, after passing through a = 0, they
combine in pairs to form type II instabilities (see figures 3- 6 ). Some questions
regarding possible degeneracy of the modes are also resolved. Further, by relating
the subharmonic modes to the superharmonic mode n = 2, it is possible to deter­
mine the boundaries of the instability regions for both types I and II (see figure 8).
We also evaluate the growth rates (figure 9 ).

2. D efin itio n s and eq u atio n s

As in previous papers (Longuet-Higgins 1985 a, 1984, to be referred to as I and II)
we consider a progressive irrotational wave of finite amplitude, travelling to the
right, with speed с relative to the fluid at great depth. The motion is horizontally
 2005

Bifurcation and instability in gravity waves 169

periodic, with period 2n. If we take axes OX and 0 Y moving to the right with speed
c, then in the unperturbed state the coordinates X and Y can be expressed in terms
of the velocity potential ф and streamfunction \jr by Stokes’s series

Y —iX = (\/r—ф) + £а0+ 2 anein(*+i]/r)/c. (2.1)

n-1
If we take units so that g — 1 , and chose the origin so that the free surface ^ = 0
has mean level —Jc2, then the coefficients an can be shown to satisfy the quadratic
relations
aQbQ+ a*bx+ a * b 2+ a * b z + ... = - c 2/
ахЪ0+ а 0Ъх+ а* 62 + а* 6 3+ ... = 0,

a2 bQ+ a1b1+ a 0 62 + a f 6 3+ ... = 0, ’ (2 -2 )

а3Ь3+ а261+ а1Ь2+ а0Ь3+ ... = °»

in which we have written

60 = 1 , bn = nan, n = 1 ,2 ,3 ,..., (2.3)
(see paper I). With a particular choice of origin the coefficients an become real,
so the asterisks (denoting complex conjugates) may be dropped.
Equations (2.3) may in fact be expressed in the compact form
0F/aan = O, 71 = 0 ,1 ,2 ,..., (2.4)
where F is a cubic polynomial in the at, related to the Lagrangian density (see I).

3. B ifu rc a tio n : th e c ritic a l w ave steep n esses

We state here the analytical conditions to be satisfied at a critical point (Saffman
1 9 8 0 ). I t will be convenient to use the same notation as papers I and II.
By (2.2), any small increments dan in the Fourier coefficients must satisfy
С da = —dc2, (3-1)
where С is the Jacobian
- (1 + 1) (^ + <4 ) (2 a 2 + 2 a2)
(ai + a j ( l + a 0+ 2a2) [2ax+ 3a3) ( 3-2)
C=
(2a2+ 2a2) (2^ + Заз) ( l+ 2 a 0+ 4a4)

and where
dfl = {$da0,d a v da2,...)
(3.3)
dc 2 = (dc2, 0 , 0 ,...)T.
7*»
2006

170 M. S. Longuet-Higgins
In class m /l waves, the regular series, in which the wavelength is 2n /m , is
characterized by
= 1 (34)
“ mn+i = 0. J =
where an is the corresponding Fourier coefficient for class 1 waves. In this series
we have damn = ( \ / m ) d a n and damn47 = 0. On the bifurcated branch, however,
we seek solutions such th at
< 4 ,,+ ,# 0, 1 = 1 ,2 ,... (m—1). (3.5)
Now when m is odd (and greater than 1) it will be seen th a t the corresponding set
of equations (3.1) breaks up into \(m —1) subsets, of the form
C(m, I) da(m, I) = 0 (3.6)
where C(m, I) is a submatrix of C, containing only the rows and columns numbered
I, (m ± l ), (2m ± l ) y (3m ± l), etc., and similarly for da. For solutions of (3.6) to exist,
the determinant of C(m, I) must vanish. And on substituting for the am n from (3.4)
and suppressing the primes we obtain

d- - ° - ,s ', )

1
2«, За,
[ 1 + s a '] ( , + s ) “ > i K b

° i 2°. H b i 3° > н ь

( l + s b
4а4
[ 1 + ( 1 + s ) “ -] i K b

2«, За, .] ! 40.

( 2 - i b [i + H ) ° ' H h

3«, 4a4
K b ; [ , + K b ]

За, 4a< ! 5а»

H - H b [ I + (3 - s ) ° « ]
i
i -------------------- -----------------
1
!

(3.8)
Equation (7.4) of I is the special case when m = 3 and I = 1.
When m is even we obtain (Jra—1) equations similar to (3.7) and one equation,
corresponding to I = £m, which is the same as th at for m = 2 and I = 1, and is in
fact equation (6.5) of 1.
In the general case, Dm t is clearly a function of l/m only. It will also be seen
th at Dm t remains invariant if I is replaced by [m —I) so l/m is replaced by (1 —l/m).
So the curve of ak against l/m is symmetric about l/m = J.
To find the first roots of (3.7) we calculated the ratios D m JD"m t (where D'
denotes the principal minor of D) as a function of the wave amplitude
a = а х+ a3-f a5+ — (3-9)
The first zeros of D m t/D"m j are shown in table 1 for some representative low values
of m/l. These are plotted against l/m in figure 1. I t can be seen th at these lie on
 2007

Bifurcation and instability in gravity waves 171

F i g u r e 1. T h e critical w a v e steep n ess ak for b ifu rcation s o f class m /l.

a smooth curve, almost a parabola with vertex at l/m = The very small difference
between m /l = 2 and m /l = 3, which was found by Chen & Saffman ( 1 9 8 0 ), is also
accounted for by the symmetry about l/m = £.
In the limit when m-> 00 and I remains fixed, Dm t takes the limiting form

1 a1 ! ax 2a2 2at 3«з

°1 ( l + a 0) j 2at 2ax 3a3 3a2
1

ьэ

3az 4a4
—

2a, j (l+«o)
w
а

«1
2аг 2a x i 3a3 (1+2 a0) 4a4 3a,
2a, 3«, j 2a j 4a, (1 + 2 a 0) 5a6
3as 3at , 4a 4 3a! 5a5 (1 + 3a0)
1
1

The first zero of D ^ /D ^ was also calculated and is shown in table 1 and
figure 1. This clearly lies on the same smooth curve.
We note that the limiting value itself cannot be expected to correspond to a
bifurcation, because any train of waves having infinitely slow modulation would
be indistinguishable from a uniform wavetrain.
2008

172 M. S. Longuet-Higgins

4. N orm al m o d e s : g en er a l v alues of th e fr e q u e n c y

In the method of calculation proposed by Longuet-Higgins ( 1 9 7 8 a, 6 ), the

coordinates (ж, у) of the perturbed flow were expressed in terms of the complex
potential (ф + ixjr) in the form
x = Х (ф , 1/г)+е£{ф, 1/г)е-ш Л
(4.1)
y = У{фу1/г) + ет}{ф,}1г)е- ш J
where X, Y are the solutions for the steady wave, as in §2 above, and the remaining
terms in (4.1) represent small perturbations, of order e and of radian frequency <r.
The free surface was assumed to be given by
xjr = е/(ф) e~i<rt. ( 4. 2 )

The functions £, rj and / were then expanded in the form

y - % = E (an + i/?n)e*"<*+W c,
n-0
(4.3)
/= S (y„ + ian)eIn(*+ W c ,

where a n,/3n,yn>8n were real constants, to be determined. Application of the

boundary conditions then led to an eigenvalue problem
(/U -B )t> = 0, (4.4)
where Л = —ia*; A and В are matrices involving the coefficients an in (2.1), and
v is a column vector of the coefficients a.n,fin,y n,8n. The forms of A and В are
specified in the Appendix to Longuet-Higgins (1 9 7 8 a). Truncation of the series
after terms in е1Лг^/с leads to finite matrices A N and B N, each of order
(42V+ 2) x (4 iV+ 2 ). The eigenvalue problem (4.4) can be solved by successively in­
creasing N until the eigenvalues Л and eigenvectors v have converged numerically,
with or without the help of accelerating convergence techniques.
This method was used by Longuet-Higgins (1 8 7 8 a, b) to calculate the frequencies
crn of the superharmonic modes for values of ak up to about 0.42 for the super­
harmonic modes and for the simplest subharmonic mode n — (£,f). For the more
general subharmonic modes, which require matrices of order (8 iV+ 4), the calcu­
lations could be carried accurately only to ak < 0.40. The main limitation, a t th at
time, was on the size of the matrices th at could be handled conveniently by the
available computers.
However, the above limitation can be overcome, in part, by the observation,
suggested by M. Tanaka (personal communication, 1 9 8 4 ) that by rearranging rows
and columns in the matrices A and В the system (4.4) could be reduced to an
equivalent system having half the number of rows and columns, for the same value
of N. In this way the order of the highest harmonics in the calculation can be a t
least doubled, in principle.! Moreover, the new method of calculation of the Fourier
coefficients which was introduced in II makes it possible to calculate the values
of an accurately to much higher order than before (n at least 600) and hence also
the elements of the matrices A N and BN.
 2009

Bifurcation and instability in gravity waves 173

The rearrangement of the matrices is as follows. In figures 5 and 6 of
Longuet-Higgins (1 9 7 8 a) let the last N columns be placed immediately after the
first (iV+ 1 ) columns; also let the last N rows be interchanged with rows (N + 2)
to (2 iV +l). Then each of the matrices A N and BN is divided into four square
submatrices of order (2 iV+l), the top left-hand and bottom right-hand quarters
o i A N vanish, and so do the top right-hand and bottom left-hand quarters of BN.
In the present notation, if we take
v x = (a 0,a „ a 2,...; - y j c , - y 2/c, ...)T,
(4.5)
v2= ( ^ 0 >/?i> ^ 2 > — S2/c, , )
then equations (4.4) can be rewritten as

Thus
(l, Ш ( во I ) ( 3
XAn V2 = Bu Vt A
(4.7)
XA v x = B22v 2,]
and on eliminating v 2 and v x in turn, we have
(AM21—B22 A 12l B n ) v x = 0 ,|
(4.8)
(A2A 12—B l l A 2l1 B 22) v 2 = 0 /

The forms of the submatrices A 12, A 2l, Bu and B22 are given in Appendix В

5. S u p e r h a r m o n i c s : t h e m o d e n — 2
To check the method, and to verify the calculations of Tanaka ( 1 9 8 3 ) on the
stability of the lowest superharmonic mode, the above method was first applied
to superharmonicsj in the range of wave steepness 0 < ak < 0.43.
The Fourier coefficients an were computed directly by the method of paper I,
up to order n = 600. These agreed well with an asymptotic expression for a n, valid
at large wave steepnesses, which has been derived in Longuet-Higgins (1 9 8 5 b). The
eigenfrequencies, or rather their squared values, were then computed by the
method of §4 , a t selected values of the wave steepness ak , and with values of N
as great as 72 when ak ^ 0.425, and 144 when ak = 0.43. The results for the two
lowest eigenfrequencies and <r2 are shown in table 2a. In the same table are
shown the first differences AN between successive convergents, and the ratios r of
successive values of A^. As this ratio is nearly constant in each case, we may
estimate the final value by Richardson’s extrapolation formula
/00 = / t f + A W ( 1 ” r )> (6 -1)

corresponding to a geometric series for AN.

The resulting values of a\ and <r\ are shown in table 2b. The lowest eigen­
frequency cr1 corresponds always to a pure phase shift (see Longuet-Higgins
f T h e p o ssib ility is in fa ct a con seq u en ce o f th e fore-an d -aft sy m m e tr y o f an irro ta tio n a l w a v e .
J This work was done in July 1984: see Tanaka ( 1985 ).
2010

174 M. S. Longuet-Higgins
1 9 7 8 a) and should be precisely zero. The estimated value of erf therefore gives
some indication of the accuracy of the computation. This suggests th a t the last
decimal place of the calculated values of v \ is uncertain. Nevertheless when
plotted in figure 2 it can be seen th at there is close agreement with T anaka’s ( 1 9 8 3 )
curve. In particular there is no doubt th at <r\ changes sign between ak = 0.425
and 0.430, very close to the energy maximum a t ak —0.4292.

F ig u re 2. Square of the radian frequency, <r£, for the lowest non-trivial superharmonic mode;
+ , Longuet-Higgins ( 1978 ) ; x , Tanaka ( 19 73 ); ©, present paper.

6. S ubh a rm o n ic modes

The notation for the subharmonics is explained in Longuet-Higgins ( 1 9 7 8 6 ). At

low or moderate values of ak> two modes having dominant harmonics n = v and
(2 —v), say, combine to form an unstable (Benjamin-Feir) mode denoted by
(v,2 —v). As ak is increased, such a mode eventually returns to neutral
stability (cr real). After splitting into two, the frequency of the upper branch a t
first increases, then at higher values of ak it remains nearly constant. The
frequency of the lower branch decreases towards zero, at a wave steepness around
ak = 0.40; see, for example, figures 4 and 5 of Longuet-Higgins ( 1 9 7 8 6 ). The first
question to be resolved is the precise behaviour of the frequency curve as <r->0 .
For the general subharmonic we have m > 2 (see §5 ) and the order of the
matrices A t] and Bi} was 4N. Successive calculation of the eigenfrequencies at, say,
N = 6 8 , 7 2 , 7 6 and 8 0 gave a sequence of convergents for each eigenfrequency. On
deriving the corresponding differences An and ratios rn as in § 6 it was found th at
in each case the rn were nearly constant. Hence a convergence formula similar to
(6 . 1 ) was applied, but with a higher order correction. In this way an accuracy of
a t least five decimal places was usually obtained. (For the special case m = 2 , the
corresponding order of the matrices was only 2iV.) Numerical values of the
frequencies for the modes 8 n = ( / , 1 6 —/) are given in table 3, and are plotted in
figures 3-5.
 2011

Bifurcation and instability in gravity waves 175

F ig u r e 3 . The radian frequency Re (cr) aa a function of the wave steepness ak , for the simplest
subharmonic mode, n = (£,|).

The simplest mode, n = (£, §), is shown in figure 3. I t will be seen th at the curve
has (by symmetry) a vertical tangent at D where cr = 0,ak = 0.40496 (see table 1).
Figure 4 shows the two next lowest modes, n = (},J) and (J, J), which are in fact
all roots of one analytic equation (cf. §§4,5). These pass through zero at a steepness
ak = 0.40433 (see table 1). At the point of zero frequency neither curve has a
vertical tangent, but each passes continuously into the reflection of the other in
the axis cr = 0. At higher values of ak the combined curve does have a vertical
tangent, at which a new instability appears. This can be appropriately labelled by
the combined symbol 4л. = (1,3,5,7) or 8n = (2,4,10,14).
Figure 5 shows the intermediate modes 8n = (/,16 —/), where / is odd. In each
2012

176 M. S. L onguet-H iggins

Re (cr)

F ig u r e 4 . The radian frequency Re(<r) as a function of the wave steepness ak , for the
subharmonic modes 4n = (1,7) and (3,5).
 2013

Bifurcation and instability in gravity waves 177

Re (cr)

F ig u r e 5. The radian frequency Re(or) as a function of the wave steepness ak, for the modes
8n = (1,15) (3,13), (5,1») and (7,9).
 2014

(2.14) 178

<=>
Re(<r)
M. S. Longuet-Higgins

Fiqueb 6. The family of frequency curvee 8n ~ (1,16—1), including the euperharmonic 8n = (0,16).
 (2,14)

Re (a)
Bifurcation and instability in gravity waves

F ig u r e 7. Enlargement of figure 6 in the neighbourhood of the bifurcation points.

179
2015
2016

180 M. S. L onguet-H iggins

case it will be seen th a t a mode (1, 16 —J) combines smoothly with the reflection
of one of the other modes, say (l\ 16 —Г), where \ l —1'\ — 8. The intersections of
these curves with the axis <r = 0 correspond precisely to the bifurcation points
ra = I/ 8 or V/ 8 listed in table 1. For an analytic proof, see Appendix A. Thus all
the intersections lie between the point В corresponding to l/m = 0, ak = 0.40223
and the point D corresponding to l/m = \ yak = 0.40496.
As ak is increased, each combined mode becomes unstable. The two corresponding
symbols for these modes are 8n = (1,7,9,15) and (3,5,11,13) respectively.
A closer look at the crossover zone (figure 6) shows how the branches are
regularly spaced. The pattern is completed by the curve 8n = (0,16), th a t is
n = (0,2), the same superharmonic mode as th at studied in §5. Here it is a 2 th at
is plotted, not <r\ , as in figure 1. This curve passes through the point E at
ak = 0.4292.
All critical points P, where each curve has a vertical tangent, clearly lie on a
single smooth arc. If further curves, given by n = [l/m, 2 —l/m), were added and
if m /l were increased without limit, it appears th a t the curves would approach both
the curve n = 2 and the axis cr — 0. The points P would then tend towards the
corner at E. Moreover, the tangent a t E to the locus of P must be vertical.
In figure 6 it seems a t first sight th at four of the curves pass through a single
point just above the horizontal axis dt around ak — 0.405, and th at four also pass
through a second, lower, point. An enlargement of the area is shown in figure 7,
from which it can be seen th at the coincidences are not exact. This may be
understood by considering th at as m tends to infinity, the corresponding curve
(l/m , 2 —1/га) will in the limit cross the axis <r = 0 at the point B, corresponding
to the same value of ak as the limiting point l/m = 0 in figure 1, th a t is, a t
ak — 0.40223. At the same time as the gradient decreases, the curve will intersect
the curve n = (J, |) in a point which comes increasingly close to the horizontal axis.

ak
F ig u r e 8. Regions o f stab ility for modes of types I and II.
 201 7

B ifurcation an d instability in gravity waves 181

F ig u r e 9. Rates of growth Im(o*) for modes of types I and II, over the range
0.34 < ak < 0.43.

To mark the boundary between unstable and neutrally stable modes of the
present type, we may draw a smooth curve through all the critical points where
the frequency curves have a vertical tangent. The result is figure 8 .
On the left of the diagram the stability boundary meets the axis cr = 0 in the
point A where ak = 0.347. This has been shown to be the limiting wave steepness
for slow modulations, as given by Whitham’s averaged Lagrangian theory
(Lighthill 1 9 6 7 ; Longuet-Higgins 1 9 6 8 b). Here we note th at the point A is quite
distinct from the limiting point В discussed earlier in §3. This may be seen simply
2018

182 M. S. L onguet-H iggins

from an examination of figures 5 and 6 . A formal proof was given in § 6 of Saffman
( 1 9 8 0 ).
I t will be noted from figure 4 th a t the lower left-hand branch of 8 n = (1,15),
corresponding to n — J, is actually lower than th a t corresponding to 8 n = (2,14),
i.e. n = J. The limiting height, as ak-+ 0 is given by <rc = | n —rd | (see Longuet-
Higgins 1 9 7 8 6 ). As n-*0 we have trc->0. Hence we may expect the loci of the
critical points to approach the same point E on the axis a = 0, and with vertical
tangents a t th at point. This is shown in figure 8 .
The rates of growth of the unstable modes are shown in figure 9. Clearly the
type II instabilities generally grow much faster than type I.

7. D isc u ssio n

The subharmonic instabilities th a t we have studied are of two kinds. Those with
dominant wavenumbers n = (v,2 —v) may be called type 1 , and have relatively
low growth rates. They have been identified as Benjamin-Feir instabilities (see
Longuet-Higgins 1 9 7 8 6 ). They extend throughout the range of ak betweenO and E.
The point E is not associated with a bifurcation.
Secondly, we have type 2 instabilities, with dominant wavenumbers
n = (v, \ —v, 1 + y, 2 —v), and relatively high growth rates, see figure 9. In two
dimensions, these instabilities extend throughout the range from В to E. The case
v = \ is associated with a bifurcation at B.
There are presumably higher types of instability in ranges of ak beyond the
point E.
Tanaka ( 1 9 8 5 ) has shown numerically that the next superharmonic n = 3
becomes unstable at ak = 0.442, close to the next energy extremum d E = 0.
Presumably the frequencies of the subharmonic modes could be calculated
similarly. However, for an understanding of these and other results at still higher
values of ak it may be better to use an asymptotic analysis based on the theory
of the almost-highest wave (Longuet-Higgins & Fox 1 9 7 8 ).
Some authors have suggested th at the type II instabilities, and other instabilities
of higher order, are essentially three-dimensional, being finite-amplitude examples
of the higher order resonances noted by Zacharov ( 1 9 6 8 ). There is, however,
nothing three-dimensional about the phenomena discussed here. This contrasts
with the situation in shear flows, for example, where stretching of vortex lines
produces some essentially three-dimensional effects which would be absent in two
dimensions. Yet irrotational waves in only two dimensions still exhibit a wealth
of interesting nonlinear properties. In this paper we have taken the alternative view
th at such properties will be most easily understood, both numerically and
analytically, by first concentrating attention on the simplest possible system,
namely the two-dimensional case. From recent papers (Saffman 1 9 8 0 , 1985*
Tanaka 1 9 8 3 , 1 9 8 5 ; Longuet-Higgins 1 9 8 5 a) there is evidence that this approach
has already proved fruitful.
Since this investigation was begun, the author has learned of a report by Branger
et al. ( 1 9 8 4 ) in which some of the above results were obtained, but apparently with
less accuracy. To accelerate convergence, these authors used the ‘e-algorithm’ of
 2019

Bifurcation and instability in gravity waves 183

Wynn ( 1 9 5 6 ), without greatly increasing the order of their matrices beyond those
used in Longuet-Higgins ( 1 9 7 8 6 ). Similar calculations by Kharif & Roux ( 1 9 8 4 )
for the euperharmonic n = 2 are certainly less accurate than those of Tanaka ( 1 9 8 3 ,
1 9 8 5 ). In addition, Branger et al. have not resolved the questions of degeneracy
or determined the stability boundaries to the extent done in this paper. Their report
should nevertheless be consulted for additional information, particularly regarding
the eigenfunctions.

A ppe n d ix A. N ormal modes in the lim it as <r->0

In paper II it was shown that the algebraic condition for the frequency o' of a
superharmonic mode to pass through zero is, in general, the same as the condition
for a bifurcation point. (An exception occurs if the mode represents, in the limit,
a pure phase shift.) Here the same analysis will be extended to the subharmonic
modes.
For the frequency <r to pass through zero, it was shown that the matrix M should
be singular, where
(Po + Po) (Л + Л) (^2+^)
м = (Л + Л ) № + Л + О о ) ( Л + J i) (А 1)
(Л + Л ) ( Л +A +G o) Л + ^ + Э Д о ) ••

and where
Рп — 2 bn+jb} ,
J- о
oo (A 2 )
Qn = 2 Яп+jbj.
i- 0
An exception to the rule could occur if the limiting form of the instability is a pure
phase-shift.
We shall now show that for a subharmonic normal mode of class m /l , the
condition that | M | = 0 is equivalent to the condition that D m t = 0.
I t was proved in II that
M = J3 x C , (A3)
where
Л Ьг Ь,
0 К 6. Ь2
B= 0 0 К Ьг (A 4)
0 0 0 ь,

and С is the matrix (3.2). In a subharmonic normal mode of class m the only non-zero
components of the unperturbed wave are those of order n = j m , where
j = 0 ,1 ,2 ...... Hence also the only non-zero Pn and Qn are those of order jm.
Now in a subharmonic of class m /l (that is, one having a wavelength m times
2020

184 M. S. L onguet-H iggins

the unperturbed wavelength), the only harmonic components arc those of order
i j m ± l ) (see Longuet-Higgins 1 9 7 8 6 , §4). Hence the condition for the frequency
of such a perturbation to pass through zero is th at the matrix M {m , /) be singular,
where M(m, I) denotes the matrix constructed out of the rows and columns of M
numbered (j m ± l ). But all the elements of В vanish except those numbered sm,
where 5 is an integer. Hence we find, corresponding to (4.3), th at
M(m, I) = B{m) C(m, I), (A 5)
where
0 bm 0 ^2m
0 b0 0 bm 0
0 0 fro 0 bm (A 6 )
1=
0 0 0 bo 0
0 0 0 0 bo

except when m = 2 : then

-bo 62 b<
0 bQ
0 b22
B{ 2) = (A 7)
0 0 bo ...
: ; :

Becausc |£ (m )| = 1 for all m, B(m) is non-singular. Therefore M (m J) is singular

if and only if C(ra. /) is singular. This proves the result.

A p p e n d i x B. F o r m s o f t h e m a t r i c e s A i} a n d B fj

We give here, for reference, the forms of the submatrices A l2, A 2l, B n and B22
introduced in §4. These will be clear from the case N = 2. Thus
(60+ 60) ( K +6,) (62+ 62) "
bi (60+ 62) №,+6a) 0

-<4,2 — b2 bz (bQ+ b<) (В 1)

-b y (b0 b2) (bx-b3) (1*0-1\) U \-^ )

-Tv

<5
<2

L -ь2
1
1

1—

-b 9 (b0- b i )

( + 0)
6 0 6 ( . + ,)
6 6 ( + 2)
6 2 6 (Л+Р.) [Pt + P t )~

Ьг ( + 2)
6 0 6 (6 1 + ) 6 3 (/u n ) (Л + /У
ОI —
b2 b> ( + 4)
6 0 6
(Л + /у (П+Л) (B 2)

*. i - b 0+ b 2) (- 6 1 + ) 6 3
\ ()
. 6 2 b3 (- 6 0 + 4)6 -
 2021

Bifurcation and instability in gravity waves 185

'(/'» + / Jo) (/,i + / ,«) (»S'2+ .S'.,)-

(Л + Л ) {P0 + 1*2 + Qo) (Л + Л) ! (Я0+ Я2) (*Sf, +iV3)

Bn =
(/>,+/>) ( П + ^ + ЗУо) I № + «,) (tf0+ « 4>

I Co 0
0
. i о *0. -

! о 0 ■
0 1 -в о 0 1
B z2 = ! о -2 ^ o

(Л -Л ) [P '-P t+ Q o ) (Л — j (*0- Я 2) (Д,-Л ’,)

Л Р г-Р г) (/> -/> + 2 < W i № --» .) (*S’o - ^ ) -

w h e re Pn a n d Qn a r e g i v e n b y e q u a t i o n s (4 .2 ), a n d

= £ • ( Pn-.+sQn-.)bs+ S Pn+A- <B 5)

T able 1. P o in t s o f b if u r c a t io n

m /l l/m ak
2 .5 .404961
в .375 .404813
3
3 .3333 .404687
4 .25 .404335
6 .1666 .403823
8 .125 .403501
oo .402233

T a b l e 2. (a ) C o n v e r g e n t s t o <
t\ and (t\
N г? Л г <т\ A

ak = 0.41
40 .000657 .078297
339 897
48 .000318 .49 .077400 .36
165 325
56 .000153 .48 .077075 .34
80 112
64 .000073 .48 .076963 .32
38 36
72 .000035 .076927
ale = 0.42
40 .002949 .049594
1043 2412 .45
48 .001906 .67 .047 182
694 1096
56 .001212 .65 .046086 .43
448 475
64 .000764 .63 .045611 .39
284 186
72 .000480 .045425
2022

186 M. S. L onguet-H iggins

T a b l e 2 . ( a ) ( cont.)
N Л r
ak = 0.425
40 .006803 .032706
48 .005327 1476 4190
.88 .028516 .55
56 .004032 1295 2324
1054 .81 .026192 .55
64 .002978 .77 1279
816 .024913 .53
72 .002162 681
.024232
ak = 0.43
80 .004862 -.0 1 6 2 7 4
96 1349 2074
.003513 .68 -.0 1 4 2 0 0 .94
112 911 1957
.002601 .74 -.0 1 2 2 4 3 .84
128 675 1639
.001926 .77 -.0 1 0 6 0 4 .79
144 528 1288
.001408 -.0 0 9 3 1 6
(6 ) E s t i m a t e s or o-f, cr\ a n d a 2
ak Nr
.41 300 .00000 .07691 .2773
.42 300 .0000 .04531 .2129
.425 300 .0006 .0235 .153
.43 600 .0003 -.0 0 4 5 .067 i
T able 3. F r e q u e n c i e s o f s u b h a r m o n i c m o d e s o f t y p e 8 n — (1, 1 6 — /) i n t h e
RANGE OF WAVE STEEPNESSES ak FROM 0 .4 TO 0 .4 2 5
ak °Y <r< 07 *7 <*i <*r
I= 1
.400 .05028 .0 .00508 .0 .22420 .0 .24092 .0
.4025 .05040 .0 .00156 .0 .22772 .00506 .227 72 .00506
.405 .05051 .0 .00253 .0 .22236 .00890
.00890 .22236
.40625 .05057 .0 .00485 .0 .21944 .00923 .21944 .00923
.4075 .05063 .0 .00741 .0 .31632 .21632 .00856
.00856
.410 .05077 .0 .01344 .0 .2139 .0 .20479 .0
.415 .05107 .0 .03236 .0 .2179 .0
.0 .16350
.420 .05147 .0 .086 .037 .2185 .0 .037
.086
.425 .0518 .0 .069 .095 .2187 .095
.0 .069
1= 2
.400 .09892 .0 .01459 .0 .23044 .0 .17967 .0
.4025 .09911 .0 .00681 .0 .23126 .0
.0 .16718
.405 .09931 .0 .00278 .0 .231 85 .0
.0 .15281
.40625 .09941 .0 .00860 .0 .23209 .0
.0 .14454
.4075 .09951 .0 .01545 .0 .23230 .0 .13516 .0
.410 .09973 .0 .03524 .0 .23265 .0 .11008 .0
.415 .10022 .0 .0667 .0595 .23315 .0 .0667 .0595
.420 .1008 .0 .0592 .0983 .2335 .0983
.0 .0592
.425 .1017 .0 .0493 .134 .2339 .134
.0 .0492
Z= 3
.400 .14426 .0 .03130 .0 .21523 .0
.0 .11388
.4025 .14450 .0 .01787 .0 .21560 .0
.0 .09831
.405 .14475 .0 .00197 .0 .21592 .0
.0 .07622
.40625 .14487 .0 .021 10 .0 .21607 .0 .0
.05683
.4075 .14499 .0 .03791 .02723 .21621 .0 .02723
.03791
.410 .14525 .0 .03664 .05488 .21647 .0 .03664 .05488
.415 .14580 .0 .03374 .09033 .21695 .0 .03374 .09033
.420 .14648 .0 .03016 .1206 .21743 .0 .03016 .1206
.425 .14744 .0 .0254 .152 .2180 .0 .0254 .152
 2023

Bifurcation and instability in gravity waves 187

R e fe r e n c e s

Benjamin, Т. B. 1967 Instability of periodic wavetrains in nonlinear dispersive systems. Proc.

R. Soc. Lond. A 299, 59-67.
Branger, H., Ramamonjiarisoa, A. & Kharif, C. 1984 On the subharmonic instability of very
steep deep water gravity waves. Report, Inst. Mech. Stal. Turb., Marseille, October 1984.
(11 pages.)
Chen, B. & Saffman, P. G. 1980 Numerical evidence for the existence of new types of gravity
waves of permanent form on deep water. Stud. appl. Math. 62, 1-21.
Kharif, C. & Roux, B. 1984 Stabilite des ondes de gravite de fortes cambrures. J. Mec. Thior.
Appl. 3, 717-723.
Lighthill, M. J. 1967 Some special cases treated by the Whitham theory. Proc. R. Soc. Lond.
A 299, 28-53.
Longuet-Higgins, M. S. 1975 Integral properties of periodic gravity waves of finite amplitude.
Proc. R. Soc. Lond. A 342, 157-174.
Longuet-Higgins, M. S. & Cokelet, E. D. 1978 The deformation of steep surface waves on water.
II. Growth of normal-mode instabilities. Proc. R. Soc. Lond. A 364, 1-28.
Longuet-Higgins, M. S. 1978 a The instabilities of gravity waves of finite amplitude in deep
water. I. Superharmonics. Proc. R. Soc. Lond. A 360, 471-488.
Longuet-Higgins, M. S. 19786 The instabilities of gravity waves of finite amplitude in deep
water. II. Subharmonics. Proc. R. Soc. Lond. A 360, 489-505.
Longuet-Higgins, M. S. 1984 On the stability of steep gravity waves. Proc. R. Soc. Lond. A
396, 269-280. (Referred to as II in text.)
Longuet-Higgins, M.S. 1985 a Bifurcation in gravity waves. J. Fluid Mech. 151, 457-475.
(Referred to as I in text.)
Longuet-Higgins, M. S. 19856 The asymptotic behaviour of the coefficients in Stokes’s series
for surface gravity waves. J. Inst. Math. Applies 34, 269-277.
McLean, J. W., Ma, Y. C., Martin, D. U., Saffman, P. G. & Yuen, H. C. 1981 Three dimensional
instability of finite amplitude water waves. Phys. Rev. Lett. 46, 817-820.
McLean, J. W. 1982 Instabilities of finite-amplitude water waves. J. Fluid Mech. 114, 315-330.
Saffman, P. G. 1980 Long wavelength bifurcation of gravity waves on deep water. J. Fluid
Mech. 101, 567-581.
Saffman, P. G. 1985 The euperharmonic instability of finite amplitude water waves. J. Fluid
Mech. 159, 169-174.
Tanaka, M. 1983 The stability of steep gravity waves. J. phys. Soc. Japan 52, 3047-3055.
Tanaka, M. 1985 The stability of steep gravity, waves. II. J. Fluid Mech. 156, 281-289.
Wynn, P. 1956 On a device for computing the em(Sn) transformation. M.T.A.C. 10, 91-96.
Zakharov, V. E. 1968 Stability of periodic waves of finite amplitude on the surface o f a deep
fluid. J. appl. Mech. tech. Phys. 2, 190-194.
2024

Proc. R. Soc. Lond. A 410, 19-33 (1987)

Printed in Great Britain

A stochastic model of sea-surface roughness

I. Wave crests
By M. S. L o n g u e t -H ig g in s , F.R.S.
Department of Applied Mathematics and Theoretical Physics, Silver Street ,
Cambridge CBS 9 EW, U.K., and Institute of Oceanographic Sciences,
Wormley , Surrey GUH bUB, U.K.

(Received 26 June 1986)

In this paper we develop a two-scale model of sea-surface roughness, in

which for the first time the randomness of both long and short waves is
taken fully into account. The model includes long-wave-short-wave
interactions, dissipation of the short-wave energy by breaking, and
regeneration by the wind. This leads to an integral equation for the
short-wave steepness, which is solved by iteration.
The effects of wind speed and of long-wave steepness upon the
distribution of roughness a t the long-wave crests are calculated and
discussed. Also the effect of a band-width param eter for the long-wave
spectrum. A random noise source can be included.

1. I n t r o d u c t i o n
In almost all sea states under the action of wind there is a rather broad spectrum
of wavelengths, ranging from long gravity waves of period several seconds down
to short gravity waves and capillary waves, which are highly responsive to the
local wind speed. The latter are known to be modulated by the longer gravity
waves on which they ride (Evans & Shemdin 1980) and it has been shown by
numerous laboratory and field studies (Keller & Wright 1975; Wright et al. 1980;
Plant & Keller 1983; Hoogeboom 1985) th at these modulations are responsible for
backscattering of centime trie radar waves from the ocean surface, and hence are
the principal factor in imaging the surface by X-band and L-band radars.
Nevertheless, the problem of how to account for the observed distribution of
short-wave energy with regard to the phase of the longer waves has remained
unsolved.
I t has long been known theoretically th at straining of the short waves by the
orbital motion of the long waves, together with work done against the short-wave
radiation stresses, tends to produce shortening and steepening of the shorter waves
near the long-wave crests (Longuet-Higgins & Stewart i 9 6 0 ). Recently, some much
more accurate calculations of this effect according to the principle of action
conservation (Longuet-Higgins 1987) have shown the importance of including full
nonlinearity of the longer waves, in some circumstances. A linear model for
ehort-8 urface waves, which includes both growth and dissipation, has been
proposed by Smith (1986). Nevertheless, all models so far proposed are unrealistic,
in th at it has been assumed, first, th at the short-wave steepness is somehow rigidly
[ 19 1
 2025

20 M. S. Longuet-Higgins
determined, rather than having a distributed probability, and second, that the
height of the longer waves is uniform and given. The purpose of the present paper
is to replace these by more realistic assumptions, and indeed to show that the
randomness of the longer waves plays an essential role in distributing the short
wave steepness.
The basic idea, introduced in §3, is to suppose th at the short-wave steepness 8
a t the crest of a long wave of amplitude A has some probability density depending
on A , and then to relate the density on one wave crest to the density on the wave
crest immediately before. This procedure is made possible by the fact that we
already know, at least approximately, the joint probability density of the
amplitudes A 1 and A2 of two successive waves. (The approximate joint distribu­
tion, sometimes called the bivariate Rayleigh distribution, has been used with
success in predicting the properties of wave groups.) With suitable assumptions
regarding the history of the short waves over the intervening time, including
growth by wind action and possible dissipation by breaking, we are able to show
th a t the density p(a\A) of the short-wave slopes on a long wave of given amplitude
A satisfies a certain integral equation, which can be solved in a straightforward
manner by iteration (§4). In fact, the presence of both growth and dissipation are
essential to the convergence of the solution.
In the subsequent sections, we present some results for typical values of the
parameters. These show how the solutions depend upon the rm s steepness of
the longer waves, and on the wind-induced growth rate. I t is notable th at the
short-wave density p(e\A) can be biomodal, with a peak both a t the limiting
steepness s0 and at some lower value of в. In § 6 we investigate the effect of group
length of the longer waves. Somewhat different distributions are obtained for ocean
swell (long-wave groups) and wind waves (short groups), respectively. In §7, we
investigate the result of adding a certain amount of noise to the process of wave
growth.
Sections 8 and 9 contain a physical discussion and a statement of the conclusions.

2. D escriptio n of the m odel

Suppose th at the longer waves consist of a fairly narrow-band, gaussian

disturbance, in which the time-interval т between successive crests is nearly
constant, as shown in figure 1. The amplitudes of two typical successive waves are
denoted by A x and A 2. Superposed on the longer waves is a group of shorter waves
whose r m s surface slopes, *at the crests of longer waves, are denoted by and s2
respectively. I t is convenient a t first to consider the time t in figure 1 as the time
measured a t the short-wave group. Thus, the elapsed time between the appearance
of the short-wave group at crests 1 and 2 is given by
t — t2 —L/(C cg), (2 . 1)

where L and С are the wavelength and phase speed of the long waves, and cg is
the group velocity of the short waves. Assuming cg -4 С .we have
r « L / C = T, (2.2)
approximately.
2026

Sea-surface roughness 21

?«)

F i g u r e 1. T im e h isto r y o f a sh o r t-w a v e g ro u p , w ith s te e p n e s s s ,, a2 a t tim e s t — tlt tv

During this time the wind will generally act on the short waves so as to increase
their slope, and if there were no dissipation by wave breaking the short waves
would grow by a factor e^T, where /? is a time rate of growth, related to the wind
speed (see, for example, Plant 1 9 8 2 ). T hat is to say
8 j 8 x = ев , В - fir. (2.3)
On the other hand, if there were no generation or dissipation of short-wave
action, the straining of the long waves would induce a change in the short-wave
steepness depending only on the vertical elevation rj of the waves, for a given long
wavelength (Longuet-Higgins 1 9 8 7 , figure 9). Thus we should have
(2.4)
where К = 2 n /L is the wavenumber of the long waves. In fact, it is a good
approximation to take
f{Ky) = (2.5)
I t will be sufficient initially to retain only the first term in this expression. Then
from (2.4) and (2.5) we have
«*/«1 = e2-08*M«-^>. (2.6)
Supposing now th at in the absence of wave breaking the rate of growth due to
the wind is independent of the long waves, we shall have altogether
в ^ ^ е Я + ^ - 'Ч у = 2.08#, (2.7)
provided th at neither s, or 8г is limited by breaking. In such a case it is convenient
to write (2.7) reciprocally in the form
8l = 82F(A1, A i )t (2 .8 )
where F{Av A 2) = e~B^ {A'-A*). (2.9)
However, the e m s steepness 8 of the short waves is necessarily limited by
breaking. Here we shall assume a sharp limit
*1 ^ e0’ ^ so> (2.Ю)
 2027

22 M. S. Longuet-H iggins
an appropriate value of s0 being given by (5.3). Hence 8l is given by the smaller of
(2.8) and (2.10), and e2 by the smaller of (2.7) and (2.10). The situation is shown
schematically in figure 2 .

Physically, the introduction of both wave generation by wind and energy

dissipation by breaking provides the means for attaining a statistically steady
state.

3. P ro b a b ility d en sities
We propose now to treat both the slope s and the amplitude A as random
variables. The distribution of the long-wave heights will be regarded as given, and
independent of the short waves over a sufficiently long period of time. The problem
then is to determine the probability density of s at given wave amplitude A.
Now the density p(A v A 2) of successive wave amplitudes A x and A 2 for narrow
spectra has been derived and used in connection with the theory of group lengths
(Rice 19 4 4 - 1 9 4 5 , 1 9 5 8 ; Kimura 1 9 8 0 ; Longuet-Higgins 1 9 8 4 )- In fact it is the
‘two-dimensional Rayleigh’ density

f( 11 * ' M l * 4 ( 1 - * 2 M 7 K >

in which A and к are constants and / 0(z) is the modified Bessel function of order
zero. A is equal to the r m s value of the surface elevation rj and к is a ‘groupiness
param eter’ explicitly related to the spectral density of rj. I t can be shown that for
narrow spectra к* « 1 — 4 n2v*, where v is a dimensionless bandwidth (Longuet-
Higgins 1 9 8 4 ). The density of A x or A 2 alone is the Rayleigh density
p(A) = (A / A 2)e~At/i7[i. (3.2)
2028

Sea-surface roughness 23
Hence the probability density of A x given A 2, which is obtained by dividing
p {A x, A 2) by p {A 2), is
Р (А М г) = (1_ ‘t ) i l gi e“» [e-J-Ч /„(*£, S ,)]|

where £x = A x/{1 - к 2)* A, £2 = A J ( i —к2)1*A.

By using the model of § 2 we shall now derive an equation for the probability
density p(s\A) of the short-wave slope on long waves of a given amplitude A.

p(«U)

F i g u r e 3. Sketch of the density p(a\A ), showing the continuous part ф(з, A ) and the singular
part P(A)S(8—80).

First, the fact th a t the slope 8 is limited by breaking means that, for a typical
value of A , there must be an exceptionally large probability density in the
neighbourhood of 8 = s0, as shown in figure 3. In fact, the density p(s\A) has two
components, a continuous component, which we denote by ф{з, A) in 0 < 8 < s0,
and a singular component P(A)S(s —80) in the neighbourhood of s = s0, S(x) being
the Dirac delta function. Normalization of clearly requires th at

|* V (« ,^ )d e + P ( ^ ) = 1. (3.4)

We now need to relate p(s^A2) to p(sl\Al ). We shall regard p(82\At ) as being

determined by the values of p ^ l ^ ) on the previous wave, over the whole possible
range of A v Thus,
Дв2 = j p K I A ) As, p (A x\At ) <L4„ (3.5)

but this takes somewhat different forms according as the continuous or the singular
parts of p(s^A^) отр(8^Аг) are considered. First, the contribution to ф(82, А г) from
Ф(81»A ) presents no difficulties. 8Xis related to s2 by (2 .8 ), so th at we have

b8j di 82 * dsx/d82 = F{AX, A 2). (3.6)

 2029

24 M. S. Longuet-H iggins
Moreover, the integral in (3.4) must be taken over the range of A x for which 8X< 80,
hence
0 < A l < Лг+у-Ч .В + 1п( 40/«г)] = Н{аг, А г), (3.7)
say. The function H is shown schematically in figure 4.

F i g u r e 4. Sketch of the function H(st , A t ) of equation (3.7).

The contribution to ф(зг, At ) from P(AX) can be expressed as a double integral

[P(4.) S ( s , - s 0) Цв, F - s J de, F(AU A,) p(A JA t ) dA , (3.8)

and from the formula - .

J % (z )]d * = ^ ^ — (3.9)
we have - .
j i ( s s F - s }) dA , = [d K f)/c M i]ji_ji/? = ^ F (310)

from (2.10). Therefore, on reversing the order of integration in (3.8) the double
integral becomes
(l/y « ,)/4 A )p (A M i). А = Ж «..Л )- (311)
Altogether, then, when 0 < 8X< s0 we have
ГН(в0.Аш) . i
Ф(8г> A%) — I ф{8ХУА х) F(AX, A 2)p (A 1\A 2)d A x+ ——P(Ax)p(A^A 2)>
Jo 2

A x = H(s2, A 2). i

On the other hand, when s2 = s0 we may integrate each side of (3.5) to obtain

P(A,) = + (3.13)

Physically, this expresses the probability P(A2) as the sum of two parts. The first
comes from the lower long waves, as a result of steepening by straining and by
 Sea-surface roughness 25
the effect of the w ind; the second part comes from the slightly higher long waves,
as a, result of regeneration by the wind only, and in spite of the straining.
Equations (3.12) and (3.13) are clearly a coupled pair of integral equations to
determine ф(з, A) and P{A) simultaneously.

4. S o l u t i o n o f e q u a t i o n s (3.12) a n d (3.13)
To find ф{8,А) and P{A) we may proceed by successive approximation.
Arbitrary starting values 0 (1>(s, A) and Z*1^ ) satisfying the condition (3.4) may
be substituted on the right of (3.12) and (3.13) and used to calculate 0 (2)(s, A) and
F*Z)(A). Substituting these again on the right-hand sides of the equations, we may
calculate 0 (3)(s, A) and /* 3)(Л), and so on, the process imitating the development
of the short wave probabilities from one wave to the next. If the functions ф{п){8, A)
and P^n)(A) converge, we have a solution.
This was tried with two different sets of initial conditions, namely
фЫ(8,А) = в ; \ P<1>M) = 0 (4.1)
and Ф{1)(з,А) = 0, = 1. (4.2)
W ith either set, the procedure was found to converge to the same solution,
although with (4.2) the convergence was somewhat faster; typically four decimal
places were obtained after only seven or eight iterations.
At each step, the accuracy of the integrations was checked by evaluating the
left-hand side of (3.4) fdr ф — ф{п+1) and P = P<n+1>; for it may be shown th a t if
any pair ф{п), P<n) satisfy (3.4) then 0 (n+1), P<n+1>, if calculated by the iteration
procedure, should also satisfy (3.4). For graphical accuracy (0 .1 %) it was found
sufficient to take 101 integration points in the range 0 < s /s 0 < 1 and 51
integration points in the range 0 < A / A ^ 5.

5. T h e r a n g e s o f AK, 8 a n d В
The steepness A K of the longer waves will generally be limited by the fact th a t
the ‘ significant waves’, i.e. those with amplitude 2 A, cannot have a steepness much
exceeding the limiting steepness A K = 0.443 for steady progressive waves. Thus
we would expect л __ „ __ 4.
^ 2 A K ^ 0 .4 4 , A K < 0.22 (5.1)
approximately. The above estimate is consistent with the laboratory wind-wave
data of Lake & Y.uen ( 1 9 7 8 ), who found a mean value of A K for the dom inant
waves not exceeding 0.28. For a Rayleigh distribution this would correspond to
{\n)*AK; hence A K ^ 0.224.
We note th at in experiments with unsteady, plunger-generated waves, Ochi &
Tsai ( 1 9 8 3 ) found signs of breaking in individual waves for which A K = 0.35, less
than the limiting value 0.443, but this is not inconsistent with (5.1).
In a similar way, if we consider the short waves as a narrow-band process in
which the local wave steepness a = ak has a Rayleigh distribution,
р(л) = (ot/82) e-at' 2at, (5.2)
 2031

26 M. S. Longuet-H iggins
it is reasonable to suppose th at local breaking of the short waves imposes a limit
5 < e0 = 0.22. (5.3)
We note th a t according to (5.2) and (5.3) the local mean-square slope surface slope
a2 satisfies ^ Q Q5 ^ 4j

which is not far from the limiting value 0.04 + 0.02 found by Plant ( 1 9 8 2 ).
For the rate of growth of surface slopes under the action of the wind, Plant
( 1 9 8 2 ) has shown th at the formula
P = 0.02(w*/c)2 o’ cos в (5.5)
fits many different observations. Here u+ denotes the friction velocity, taken to
equal 0.04 times the wind speed U at a standard elevation, с and a are the phase
speed and the radian frequency of the short waves, and в is the angle between
the directions of the short waves and the wind. Because our analysis refers to the
short-wave slopes rather than to their spectral density, we have here divided the
right-hand side of Plant’s equation (1) by a factor 2 . In the special case в = 0> when
wind and waves are in the same direction, (5.5) reduces to
/?= 3.2x10 ~*(U/c)2<r. (5.6)
For example, if we consider short-gravity waves of length I = 20 cm, we have
or = (2ng/l)t = 1 7 .6 rad s -1 and с = д/o' = 0.56 m s_l. Hence with U = 6 .0 m s-1
we find p = 0.065 s_1. With the period of the long waves about 7.5 s (Evans &
Shemdin 1 9 8 0 , table 2), we have В = 0.49 and eB = 1.63. With the long waves
having period 2 s, we have B = 0.129 and eB — 1.13 only. Thus swell and
wind-waves may correspond to rather different values of the wind amplification,
even for the same wind-speed.
With shorter (X-band) short waves, В is generally much greater, but the
amplification of the short waves is then more limited by breaking.

6. R e s u l t s
To explore the behaviour of the solutions, calculations were at first carried out
with representative values of A K and В , and with к = 0. The latter condition
implies zero correlation between successive waves, so th at р(А х\Аг) is independent
of A 2 and is, in fact, given by the Rayleigh density p(A) of equation (3.2), with
A = AV
Figure 5 shows the solution when A K = 0.1 and В = 0.1. I t will be seen th at
as the height A of the waves increases, so the curves of ф(з, A) tend to move to
the right, th at is to say the short waves on the whole become steeper. The
probability P{A) of the short waves attaining their limiting steepness also increases
monotonically with A. For the lowest values of A / A the probability^of finding
breaking waves a t the crests of the longer waves is quite low; when A / A = 0.5 the
probability is less than 20% . When А / A = 2, P(A) is more than 90%.
The probability density ф(з, A) does not always increase monotonically with 8.
In fact, when A / A = 0 and 0.5, the density has maxima near s/80 = 0.83 and 0.92
respectively, as well as the delta-function peak a t s /s Q= 1. The distribution of
2032

Sea-surface roughness 27

F i o u r e 5. (a) The calculated probability density ф{в. A). (6) The probability o f breaking P{A)
when A K = 0.1, В = 0.1 and к = 0.

slopes is thus bimodal in these cases. When s / s Q< 0 .5 we see th a t ф(з, A) is

generally quite negligible.
In figure 6 we see the result of increasing the wave-generation param eter В to
0.2, while keeping A K = 0.1. This clearly shifts the curves ф(з, A) to the right,
increasing the short-wave steepnesses in general. The probability P of breaking
is now never less than 0.3.
In figures 7 and 8 we see similar results, but for steeper long waves: A K = 0.2.
When В = 0.1 most of the densities р( 5 |Л) are now bimodal. Physically, this is
because the distributions of short-wave steepness on the crests of the lower long

s/80 A lA

F ig u r e 6 . As fig u r e 5 , w h e n A K = 0 .1 , В = 0 .2 , к = 0 .
 2033

28 M. S. Longuet-Higgins
L0
(«)

/
AlA-0.0/ /Г \
/ / \ p 05
//
/ / A X
/
S i i i
0 0£ LO О 2 4
s f Sq A./A.
F ig u re 7. As figure 5, when A K = 0.2, В = 0.1, к = 0.

lJO
4

P 0L6

0 0t5 1J0 0" 2 4

«/«• A /A
F i g u r e 8. As figure 6, w hen A K — 0.2, В = 0.2, к — 0.

waves, say А / A < 1 , are determined by the fact th at the short-wave steepness is
limiting on the higher long waves and is then reduced by the subsequent fall in
long-wave height. The regeneration of short-wave steepness by the wind is
insufficient to overcome this effect.
In figure 8 , however, when В is increased to 0.2, the curves for ф(8, A) are shifted
back again toward the right.

Variation of к
In the results so far, the correlation parameter к has been set equal to zero.
Figures 9, 10 and 11 show the effect of increasing к by stages, while keeping A K
and В constant a t 0.2 as in figure 8 .
2034

Sea-surface roughness 29

L0
(a) (b) ' -----’

/
,^05
>4/4=00// ' V t
/ 0& / /

/ / /
1 1 _l
о 05 1.0 • 0 2 4
e/s0 ■ A /A
F ig u r e 9. As figure 8, when к = 0.5.

F i g u r e 10. As figure 8, when к — 0.75.

When к =■ 0.5 (figure 9) there is little effect on either A) or P (A ), though the

curves for ф(в,А) will be seen to have shifted slightly to the right. This value of
к, corresponding to кг = 0.25 or v « 0.2 is typical of local wind-waves (Longuet-
Higgins 1 9 8 4 , figure 14).
However, when к = 0.75 (figure 1 0 ) the shift to the right is very distinct, and
even more so when к = 0.85 (figure 11). The corresponding values of к 2 and v are
typical of ocean swell (Longuet-Higgins 1 9 8 4 , figure 13).
Physically, the effect of increasing к toward unity is to increase indefinitely the
mean length of the wave groups, and so to make the long waves appear locally
of uniform height. In the limit к - + 1 it is clear th a t the regeneration of short-wave
steepness by the wind will prevail over the redistribution of short waves by
 2035

30 M. S. Longuet-Higgins

8/Sq А /A
F i g u r e 11. A s f ig u r e 8 , w h e n к = 0 .8 5 .

variability of the long waves, and so will force the short waves up to the breaking
steepness a t every long-wave crest. Thus, we shall have
ф(в,А)-> 0, P(A)-+1. (6.1)
This is the situation th at has been assumed, in effect, by previous authors (Phillips
& Banner 1 9 7 4 ).
Generally, however, we see that this condition is by no means attained, and that
the variation of long-wave heights is an important factor in reducing the probable
short-wave steepness, even a t the long-wave crests.

7. E ffect of b a c k g r o u n d ‘n o ise ’
So far we have assumed th at the short waves are present initially and, in spite
of some dissipation through breaking, are always regenerated by the wind. In other
words, we have a ‘boot-strap’ situation, the regeneration of the waves depending
on their initial presence. However, apart from the exponential growth rate of the
waves, as expressed through the factor in (2 .8 ), there may in fact be other sources
of short-wave energy, arising from splashing, nonlinear wave interactions, etc.
W ithout specifying the exact mechanism, we may represent such a noisy source
by a constant input of probability density to the small slopes, a sort of ‘probability
rain ’. For such an input, Аф, a plausible form is
Аф = N(s/s*) e-**'2**, (7.1)

in other words a Rayleigh distribution of the r m s slope s and total amplitude N

per wave period T.
Accordingly, we tried the effect of adding on a small extra probability, Аф,
having the above form, at each iteration and at the same time normalizing by
dividing the probabilities by (1 + J A^ds). For numerical values we chose s = 0.1s0
(so th at J A0ds « N), and N = 0 .0 5 or 0 .1 . Also A K = 0 .2 = В, as in figure 8.
2036

Sea-surface roughness 31
The iteration was found to converge just as before (although a t a slower rate,
comparable to the initial conditions ( 4 . 1 ) ) . The resulting steady solution in the case
N = 0.05 is shown in figure 12. This may be compared with figure 8. Instead of
the almost vanishing values of ф found previously for low slopes (a/e0 < 0.25, say)
we now see a substantial ‘background probability ’ extending over the whole range
of low slopes, well beyond the original r m s value 5 of the noise input ( 7 . 1 ) . This,
of course, is caused by amplification of the noise by the wind stress.
Figure 13 shows the result when N is increased from 0.05 to 0.1. Now the two
maxima in ф are of comparable magnitude, when A / A ^ 2 . A t the larger values
of A / A the short waves are almost uniformly distributed over 0 < 8 < 80. The
probability P of breaking is correspondingly reduced.

e/e* A /A
F ig u r e 12. As figure 8, but showing the effect of added ‘noise’; N = 0.05.

F ig u r e 13. As figure 12; N = 0.10.

 2037

32 M. S. Longuet-Higgins

8. D iscu ssio n

One of the simplifying assumptions of the model is the existence of a sharp

critical steepness for wave breaking. In reality, there may be a range of possible
short-wave steepnesses, rather than a single value, leading to a smoother distri­
bution for p{8 , A) a t high values of 8. A modified assumption might be made that
a t high steepnesses a high degree of damping occurs, to represent the wave
breaking. In th at case, a similar type of integral equation for p(s, A) could still be
formulated.
A second limitation of the model is the simplified relation (2.7) between 8Xand
s2. This includes only the linear part of the expression (2.5), and the nonlinear term
becomes significant when A K exceeds 2, say. Inclusion of this term leads to a
modification of (3.6), but no essential difficulty.
A more subtle deficiency of (2.7) is that it implies a possibly steady rate of growth
due to the wind, even though the slope 8X may already be limiting. Thus (2.7)
strictly would imply th at near the long-wave crest, where the rate of straining is
small, 8 would have to grow beyond 80, a t least initially. This difficulty may be
overcome by determining 8 through a step-by-step integration throughout the
interval tx < t < t2. If this is done, then (2.7) is replaced by a more consistent, but
a t the same time less simple, relation.
A fourth limitation of our model is that we have used the joint density p{A x, A 2)
for consecutive long waves as given by (3.1). Strictly, this is justified only for
narrow wave spectra, associated with fairly long groups of (long) waves, but, like
the Rayleigh distribution for single waves, it may well have an unexpectedly wide
range of validity. Again, it might theoretically be replaced by a more general
expression, but only with a consequent loss of simplicity in the calculations.

9. C o n clu sio n s

We have developed a simplified model of sea-surface roughness th at takes into

account the randomness of both the longer waves and of the short-scale waves
superposed on them. Essentially the model depends .upon three parameters: (1)
the r m s steepness A K of the longer waves; (2) the natural rate of growth В due
to the wind; and (3) a correlation parameter к for the long wave amplitudes. The
model demonstrates that:
(1) the density p{e\A) of the short-wave slopes s is not necessarily unimodal,
especially a t low wave amplitudes A ;
(2) as A / A increases, so the density p(s\A ) tends to become monotonic, and the
probability P(A) of breaking a t the crest increases;
(3) the effect of increasing В is to move the curves for ф(з, A) (the continuous
p art of p(a\A)) to the right and to increase the probability of breaking at the crest;
( 4 ) the effect of decreasing the long-wave band width v , or letting a: approach

closer to 1, is again to increase the probability of breaking a t the crests of both

higher and lower long waves;
(5) the effect of adding a small amount of noise to the short waves is to increase
2038

Sea-surface roughness 33
the probabilities of low short-wave slopes. Paradoxically, the probability P{A) of
breaking is reduced.
In a further paper (in preparation) we shall show how the analysis can be
extended so as to calculate the probable distribution of surface slope, s, over the
complete profile of the long waves. This leads to expressions for the mean-square
short-wave slope, the average phase lag between the surface slope and the long
waves, and other observed quantities.

The calculations presented in this paper were begun in Cambridge, England

during November 1985, and were continued during a visit by the author to the
University of Florida, Gainesville, in February and March 1986. To Dr K. Millsaps,
Head of the Departm ent of Aerospace Sciences a t the University of Florida, and
to members of his staff, the author expresses his appreciation of their hospitality
and assistance.

R eferences
Evans, D. A. &, Shemdin, О. H. 1980 An investigation of the modulation of capillary and short
gravity waves in the open ocean. J. geophys. Res. 85, 5019-5024.
Hoogeboom, P. 1985 The determination o f modulation transfer functions from the ‘ Noordwijk ’
radar measurements in 1978 and 1979. TNO Physics and Electronic Lab. Report no. FEL-TNO
1985-70.
Keller, W. C. & Wright, J. W. 1975 Microwave scattering and the straining o f wind-generated
waves. Radio Sci. 10, 139-147.
Kimura, A. 1980 Statistical properties of random wave groups. In Proc. 17th Int. Conf. on
Coastal Engng, Sydney, pp. 2955-2973. New York: American Society of Civil Engineering.
Lake, В. M. & Yuen, H. C. 1978 A new model for nonlinear waves. Part 1. Physical model and
experimental evidence. J. Fluid Mech. 8 8 , 33-62.
Longuet-Higgins, M. S. & Stewart, R. W. i 960 Changes in the form o f short gravity waves on
long waves and tidal currents. J. Fluid Mech. 8 , 565-583.
Longuet-Higgins, M. S. 1984 Statistical properties of wave groups in random sea state. Phil.
Trans. R. Soc. Lond. A 312, 219-250.
Longuet-Higgins, M. S. 1987 The propagation of short surface waves on longer gravity waves.
J. Fluid Mech. (In the press.)
Ochi, М. K. & Tsai, C.-H. 1983 Prediction of breaking waves in deep water. J. phys. Oceanogr.
13, 2008-2019.
Phillips, О. M. &. Banner, M. L. 1974 Wave breaking in the presence of wind drift and swell.
J. Fluid Mech. 6 6 , 625-640.
Plant, W. J. 1982 A relationship between wind stress and wave slope. J. geophys. Res. 87,
1961-1967.
Plant, W. J. & Keller, W. C. 1983 Parametric dependence o f ocean wave-radar modulation
transfer functions. J. geophys. Res. 88 , 9747-9756.
Rice, S. O. 1 944~S The mathematical analysis of random noise. I. Bell. Syst. tech. J. 23, 282-332
and 24, 46-156.
Rice, S. O. 1956 Distribution of fades in radio transmission - gaussian noise model. Bell. Syst.
tech. J. 37, 581-635.
Smith, J. A. 1986 Short surface waves with growth and dissipation. J. geophys. Res. 91,
2616-2632.
Wright, J. W., Plant, W. J., Keller, W. C. & Jones, W. 1980 Ocean wave-radar modulation
transfer functions from the West Coast Experiment. J. geophys. Res. 85, 4957-4966.

2 Vol. 410. A
 2039

R eprinted from J o u r n a l o f P h y sic a l O c e a n o g r a p h y , Vol. 20, No. 4, April 1990

American Mtteorolopcal Society

Flow Separation near the Crests of Short Gravity Waves

M ic h a e l S. L o n g u e t - H ig g in s

Center for Studies o f Nonlinear Dynamics, La Jolla Institute. La Jolla. California and Department of
Applied Mathematics and Theoretical Physics. Cambridge, England

27 July 1989 and 18 October 1989

ABSTRACT
Laboratory observations of short gravity waves of length about 1 m by Koga show the occurrence of a
downward flow separation near the wave crests, making angles of 10® to 50° with the horizontal. Koga's
observations are compared with a theoretical model of flow separation suggested by Longuet-Higgins. Some
features of agreement are found.

1. Introduction directions, are shown in Fig. 2 (adapted from Koga

1982).
Steep surface waves in deep water are known to be­
The phenomenon is comparable to, but different
com e unstable in a variety o f ways, som e o f which
from, the flow studied by Banner and Phillips (1974).
have been described in a recent review ( Longuet-Hig-
They observed the separation at the edge o f a turbulent
gins 1987). Am ong the better known instabilities is the
roller artificially induced by placing a submerged ob­
whitecap, or bubbly roller, which develops on the for­
stacle in a steady current. In that case the angle o f
ward face o f a progressive gravity wave having a wave­
depression o f the separated flow appears to be some­
length greater than a few meters. A theoretical model
what smaller, and the flow resembles more the flow
for this flow was proposed by Longuet-Higgins and
beneath a whitecap.
Turner (19 7 4 ). At much shorter wavelengths, less than
Returning to Koga’s observations, we emphasise that
about 10 cm , capillary-gravity waves can trap pockets
such a separated flow, involving as it sometimes does
o f air in the wave troughs, as first suggested by Crapper
the entrainment o f air bubbles, is o f considerable in­
(195 7 ) and observed experimentally by Toba (1961). terest in connection with the production of underwater
However, at intermediate length-scales, o f the order 1 sound at low wind speeds. Although the turbulent zone
m, a different type o f instability has been observed by upwind o f the line of separation may not be as easily
Koga (1 9 8 2 ). It is this that we wish to discuss in the visible from above as in a whitecap, nevertheless the
following note. process o f trapping of air pockets, and their subsequent
Koga (1 9 8 2 ) found that near the crest o f a steep distortion into spherical bubbles, can be expected to
gravity wave having a period o f about 0.73 s (wave­ contribute significantly to the level o f underwater sound
length 75 cm ), a structured flow could be observed. A (Longuet-Higgins 1989a,b).
cluster o f bubbles was carried downwards into the in­ In the present note we draw attention to the fact that
terior o f the fluid at an angle o f about 40° to the hor­ several years before Koga’s (1982) paper, a theoretical
izontal (see Fig. la ). Sometimes several such bubble model for the separation o f flow at a free surface was
clusters were visible, as in Figs. lb and lc. The flow suggested by Longuet-Higgins (19 7 3 ). The model, in
appeared to separate from the free surface along a line its simplest form, has several features corresponding
which marked the boundary between relatively smooth closely with Koga’s observations. In section 2 we out­
flow ahead, ( i.e. dow nw ind) o f the separation line, and line the main features o f the model, and in section 3
a rough, apparently turbulent region behind (i.e., up­ we shall compare it with Koga’s observations. A dis­
w ind) o f the line. The separation was confirmed by cussion follows in section 4 and conclusions in sec­
photographs o f the surface taken from below and tion 5.
pointing vertically upwards (Koga 1982, Fig. 5 ). The
measured velocities in the bubble sheets, including their
2. A theoretical model
The model is sketched in Fig. 3. In the unshaded
region to the right, bounded by the free surface в = a
Corresponding author address: Dr. Michael S. Longuet-Higgins,
La Jolla Institute, 7855 Fay Avenue, Suite 300, P.O. Box 1434, La and the sloping plane в = —/3, the flow is assumed to
Jolla, CA 92038. be inviscid and irrotational. The density p in this region

© 1990 American Meteorological Society

596 J O U R N A L OF P H Y S I C A L O C E A N O G R A P H Y Vo l u m e 20

FlO. 1. Entrainment o f bubble clusters in breaking wind waves. Wind-speed 16 m s -1.

Fetch 16.4 m. (From Koga 1982).

is assumed constant To satisfy both Laplace’s equation and dissipation o f the short waves. Here we shall as­
and the condition o f constant pressure when fi = a w e sume that the flow is so strongly turbulent that it is
take as stream function dominated by Reynolds stresses which can be repre­
sented by a constant eddy viscosity N , say. (This is a
A ri a sin | (в - a ) (2. 1)

where

- g cosa. ( 2. 2 )

(T h e case a = ‘/ j t corresponds to Stokes’ well-known

com er flow ). The pressure p and the velocity q are
then given by

- = gr cosd - - q1 (2.3)
D L

(2.4)
It will be seen that both p and q are proportional to
the radial distance r. There is a stagnation point at r
= 0.
In the shaded region to the left, where - у < в < - 0 ,
the flow is assumed to be turbulent with a constant R e . 2. Estimated velocities V and directions в o f the (low ncai
density p'. The main source o f the turbulence is as­ the crest of wind waves, as calculated from observations of entrained
sumed to be the free surface, where there is breaking bubbles. (From Koga 1982).
 2041

A pril 1990 NOTES AND CORRESPOND ENC E 597

The above equations admit an exact solution in the

form
ф = Егг sin3(0 + 7 ) + Fr* sin(0 + 7 ) (2.9)
where E and F are constants. (For details see Longuet-
Higgins 1973, sections 3 to 5). Some representative
flows when p = p' and С > 0 are shown in Fig. 4. The
corner angle in the laminar flow to the right is always
120°. The angle of depression of the free surface lies
between 0° and 11 ° (actually arctan3“3/2, = I0°54').
The upwards inclination of the free surface in the flow
to the left is supported by the turbulent Reynolds
stresses.
The limit С -*• 0 is of special interest. We find then
that F -*■ 0 and so from (2.9)
R g . 3. Notation and coordinates for the theoretical ф = Егг sin3(0 + 7 ) ( 2. 10)
model of section 2.
representing a simple, irrotational mean flow in a 60°
corner. By symmetry, the tangential stress P„ vanishes
“turbulent Stokes flow” model). The streamfunction both on the interface в = - 0 and on the free surface
^ for the mean flow then satisfies в = - у . The constant E is given by
V4* = 0. (2.5) E = 0.0156*/# ( 2. 11)
On the interface в = —0 we assume that (see the Appendix to this paper).
ф= 0 ( 2.6 ) The existence of the flow depends upon the eddy
viscosity N being greater than zero. If N were replaced
so that there is negligible flow across the boundary, by the ordinary kinematic viscosity, then E would still
and the turbulent stresses are given by be finite, but improbably large.
Pm = - P , P„ = Cpg2 (2.7) The above solution may be generalized to the case
when p'/p < I; see Longuet-Higgins (1973, section 5).
wherep and дате given by Eq. (2.3) and (2.4), and С In Figs. 9a and 9b of that paper the angles of inclination
is a positive constant. We shall study in particular the
case when С is small. (C is likely to be of the same ol = 90° —a, 7 ' = 7 “ 90c ( 2.12)
order as the entrainment constant at the boundary of of the free surface on either side of the origin are shown
a turbulent jet, that is, of order 0.08; see Turner 1969). as functions of the coefficient C. In the limiting case
On the free surface 6 = —у we assume when С -*> 0 both of the angles d and y ‘ lie generally
£ = 0, P„ = 0, P„ = 0 (2.8) between 10°54' and 30°.

(•)

RG 4. (from Longuet-Higgins 1973). Examples of particular flows when p - p' and C '> 0.
(a) a = 60°; ( b ) « = 75e; (c) 6 = 90».
2042

598 JOURNAL OF PHYSICAL OCEANOGRAPHY V o l u m e 20

W e m ay m ention that when p '/p < 1 there exists a applicable to the im mediate neighbourhood o f the
second class o f solutions in which the flow velocity in point (or lin e) o f separation. Like the Stokes 120° cor­
the left-hand sector - у < в < ~(3 is relatively small. ner flow, it does not aim to describe the total flow in
Indeed when we let С 0 the fluid on the left becomes the wave, or even in the crest. Nevertheless a complete
static. The inclination o f the interface в = —0 is then description o f the flow must take into account the spe­
given by cial conditions existing near the separation point, which
may largely determine the global characteristics o f the
tan/3 = (3 p /p ' - l ) / 3 1' 2. (2 .1 3 ) flow.
If p '/p = 0, we see that the configuration is that o f the Perhaps the most controversial aspect o f the model
sym m etric, 120° com er-flow. If on the other hand p '/ is its m odeling o f the turbulence by a uniform eddy
p = 1, it is found that the flow velocity vanishes ev­ viscosity. Such an assumption can be expected to yield,
erywhere. at best, only approximate, or qualitatively correct, pre­
dictions. Underlying the assumption is that the tur­
3. Comparison with observation bulence arises only partly from the shear layer along
the boundary (в = - /3 ) o f the laminar flow. A much
In Fig. 2b, which shows the direction o f the flow as stronger source o f turbulence is the dissipation of en­
indicated by the bubble clusters (but allowing for the ergy by short capillary-scale features near the wave
rate o f rise o f the bubbles relative to the surrounding crest. Such dissipation occurs naturally as a result of
fluid; see Koga 1982, section 2 ) the direction varies the breaking o f capillary waves riding on the longer
from 190° to 2 30°, i.e. 10° to 50° downwards from gravity waves. Such capillary waves are blocked by the
the horizontal, with a mean value 39°. This compares gradients o f velocity on the forward face o f the gravity
with the theoretical value 49° shown in Fig. 4 in the waves; see Phillips (1981). The turbulcnce so created
case p' = p, С -► 0. If we assume p' < p the theoretical will tend to be circulated by the mean flow in the tur­
angle lies between 49° and 30°. bulent sector o f the fluid.
The magnitude o f the velocities shown in Fig. 2a We have mentioned that the turbulence will help to
above range from 0.4 to 1.2 m s 4 , with a mean value support the inclination o f the free surface to the left
0.8 m s" 1. In the theoretical model the flow velocity q (i.e., up-wave) o f the point o f separation. In addition
in the laminar region is given by equations (2 .4 ) and we may expect a contribution from the radiation
(2 .2 ), hence stresses in the short capillary waves. The effect would
be similar to the wave “set-up” maintained by waves
q 2 = 2 grcosa. (3 .1 )
in a coastal surf zone (Longuet-Higgins and Stewart
This depends both on r and a, but if as representative 1963).
values we take r = 0.05 m , a = 39°, we obtain q = 0.87 We note that Koga (1 9 8 2 ) has proposed an expla­
m s -1, within the range o f observation. nation for the entrainment o f bubbles at the point o f
Consider now the inclination o f the free surface. The separation. He suggested that it was analogous to the
photographs in Fig. 1 were taken from an angle slightly entrainment o f air by a jet entering the surface from
oblique to the line o f the wave crests on the left. It can above. The presence o f such jets is not evident in the
nevertheless be seen that the oblique line o f bubbles photographs o f Fig. 2. Rather, we believe the entrain­
separates from the surface at a point slightly forward ment o f air is to be associated with the stability o f the
o f the wave crest. In Figs. la and lc the crest profile is flow near the point o f separation. However the inves­
fairly sharply curved, so that the inclination o f the tan­ tigation o f this problem is beyond the scope o f the
gent is not well defined. Nevertheless it can be seen present paper.
that the oblique angle between the line o f bubbles and
the sm ooth part o f the free surface ahead o f the point
o f separation is not far from 120° as predicted by the 5. Conclusions
theoretical model.
In Fig. lb there appears to have been more than one Several features o f the flow separation observed by
separation event. The crest may also have become less Koga (1982) near the crests o f steep gravity waves arc
steep since the earlier events. Nevertheless the angle in agreement with the simple theoretical model pro­
between the lines o f bubbles and the tangent to the posed earlier by Longuet-Higgins (1 9 7 3 ). The latter
mean surface is about 135°, only slightly greater than assumes that on one side o f the separating streamline
the predicted value. the flow is irrotational, and on the other side it is tur­
From these comparisons we may conclude that there bulent. The turbulence, however, arises mainly from
is a rough, but quantitative agreement between the ob­ dissipation o f energy at the free surface, and not from
servations and the predictions o f the model. shearing at the separating streamline. The flow is es­
sentially nonlinear, and is not contiguous to a slate of
4. Discussion rest. Aeration of the fluid by the entrainment o f bubbles
We emphasize first that the model o f flow separation has only a small effect on the bulk density o f the fluid.
which has been suggested is only a local model, possibly A possible difference in density between the laminar
 2043

A p r il 1990 NOTES AND CO RRESPONDENCE 599

and turbulent regions has been included in the model Assuming i Ф 0, we have, to lowest order
in order to allow for the bulk effect o f bubbles pene­
3 V3 c o s t + sin ? = 0 (Al l )
trating the free surface. However on the small scales
considered here such density differences will probably whence
be slight. Only in the case o f large-scale spilling breakers
where foaming is evident do we expect significant den­ 7 = - r + arctan3 "3/2 100°54'. (A12)
sity differences (Longuet-Higgins and Turner 1974).
We shall denote this value o f у by 70 .
APPENDIX Now from equations ( A l ) we have
The Evaluation of E and F 3C<2
E- - X<2 ( A13)
In section 3 o f Longuet-Higgins (1973) it is shown sin 35 ' 37 ~ 0
that the boundary conditions (2 .6 ) and (2.7) lead to 3 CQ
the equations ( AI 4)
sin 6 sin 7r/ 3
£ sin 36 + F sin 6 = 0
( A l) as t -*• 0. Moreover from (A 4)
E sin 36 + ^ F sin 5 ■2Cej
(!« •+ « -* )
and
E cos36 + F c o sS (A2) ~ -jj ( g/ W) c o s ( v - 70)
where
6 = y - 0 (A3)
(A15)
Hence the limiting value o f E is
(A 4)
<3 = J2 co s
M - Eo = Q - 0.0156(£/Л 0 ( A16)
Hence
as stated in section 2.

cot 35 (A 5) REFERENCES
Banner, M. L., and О. M. Phillips, 1974: On the incipient breaking
of small scale waves. J. Fluid \fcch.. 65, 647-656.
csc 35 - csc 5 (A 6) Crapper, G. D. 1957; An exact solution for progressive capillary waves
3С
cos of arbitrary amplitude. J. Fluid Mech., 96, 417-445.
Koga, M. 1982: Bubble entrainment in breaking wind waves. Tellus.
34, 481-489.
We exam ine the solution as С -+■ 0 and 5 т /3 . Longuet-Higgins, M. S. 1973: A model of Oow separation at a free
Writing surfacc. J. Fluid Mech.. 57, 129-148.
------ , 1987: Mechanics o f wave breaking in deep water. Proc. NATO
3 С = A, 6 = ir/3 + 5', (A7) Adv. Workshop on Natural Mechanisms o f Surface Generated
Noise in the Ocean. Lerici, Italy, 15-19 June 1987.
we find from (A 5 ), to order <2, that ------ ■, 1989a: Monopole emission of sound by asymmetric bubble
oscillations. I. Normal modes. J. Fluid Mech., 201, 525-541.
J_ 1 1 1 ------ , 1989b: Monopole emission of sound by asymmetric bubble
(A 8)
3« V3 + 3 e ' x - oscillations. II. An initial-vaiue problem. J. Fluid Mech.. 201,
543-565.
Now substituting into (A 6 ) and expanding similarly ------ , and R. W. Stewart, 1963: A note on wave set-up. J. Mar. Res..
in powers o f « we obtain 21,4-10.
------ , and J. S. Turner, 1974: An “ entraining plume" model of a
spilling breaker. J. Fluid Mech., 63, 1-20.
(l + 2Уз -^ < 2| co s ( 7 - с) Phillips, О. М., 1981: The dispersion of short wavelets in the presence
of a dominant wave. J. Fluid Mech., 107, 465-485.
= (1 - V3< + <2 ) c o s 7 (A 9) Toba, Y., 1961: Drop production by bursting of air bubbles on the
sea surface. III. Study by use o f a wind flume. Mem. Coll. Sci.
whence to order e2 Univ. Kyoto. Ser. A. 29, 313-344.
Turner, J. S.. 1969: Buoyant plumes and thermals. Ann. Rev. Fluid
(3V3< - 2e2) COS7 + (« + 2l/3<2) sin7 = 0. (A10) Dyn., 1, 29-44.
2044

A stochastic model o f sea-surface roughness. II

B y M ich ae l S. L o n g u e t -H ig g in s

Institute for Nonlinear Science, University of California San Diego, La Jolla,

California 92093-0402, U.S.A.

A two-scale model of a wind-ruffled surface is developed which includes (1)

modulation of the short waves by orbital straining in the long waves, (2) dissipation
of short-wave energy by breaking, and (3) regeneration of the short-wave energy by
the wind. For simplicity the long waves are at first assumed to be uniform. I t is
shown th a t the character of the surface is governed by the parameter Q = /?/(cryKA ),
where is the proportional rate of short-wave growth due to the wind, а, К and A
are the long-wave frequency wavenumber and amplitude, and у = 2.08. When
Q < 1 the short waves break over only part of the long-wave surface. When 1
they break everywhere.
The mean-square steepness s2 of the short waves is an increasing function of
but a decreasing function of the long-wave steepness AK. The phase angle between
s2 and the long-wave elevation rj is an increasing function of Q. The correlation
between s2 and rj is largest when Q < 1, but tends to 0 as I.
The simple model is extended to the case when the long-wave amplitude A has a
Rayleigh probability density. To take account of the ‘sheltering ’ effect of high waves
we compute the case when any two successive waves have a bivariate Rayleigh
density.
The application of the model to laboratory and field data is discussed.

1. Introduction
The hydrodynamical interaction of short waves with longer waves on the sea surface
has important consequences both for air-sea transfer processes and for the imaging
of the ocean surface by certain types of remote sensors which depend upon Bragg
scattering from the shorter waves. References are given in Longuet-Higgins (1987;
referred to as part I). In that paper, a model of the short-waves dynamics was
presented which included the following features: (a) modulation of the short-wave
steepness by orbital straining in the longer waves; (b} dissipation of short-wave
energy by breaking, particularly near crests of the longer waves; (c) regeneration of
the short-wave energy by the wind; (d ) a random distribution of long-wave
amplitude.
Particular attention was paid to the behaviour of the short waves at the long-wave
crests, and it was shown how an integral equation could be formulated for the
probability density p(s | A ) of the short-wave steepness s at the crest of a longer wave
of given amplitude A. In part I the author looked forward to a study of the
characteristics of the short waves over the whole profile of the longer waves. From
these one could obtain several parameters of interest such as the mean-square short­
wave steepness s2 over the whole wave surface, and the phase and amplitude
modulation of s2 relative to the long waves.
Proc. R. Soc. Lond. A (1991), 435, 405-422
Printed in Great Britain 405
 2045

406 M. S. Longuet-Higgins
The present paper gives the results of this study. To present and interpret them
we first describe a simplified model which includes the elements (a), (b) and (c) above
but omits, for the time being, the random element (rf). This simplified model is
described in §§2 and 3 below, and its consequences are discussed in §§4 and 5.
Randomness is then introduced in two stages. A partially random model, in which
the long-wave amplitude has a Rayleigh density but the wave height is locally
uniform (a narrow-band spectrum) is studied in §6. Then in §7 we re-introduce the
full bivariate Rayleigh distribution for the long wave amplitude, and present the
numerical results. In §§8 and 9 we make comparisons with laboratory and field data.
A discussion follows in § 10.

2. Physical assumptions
We assume that the surface-slope spectrum may be effectively divided into two
parts as follows. First, a narrow, low-frequency part, to be called the ‘long waves’.
A representative amplitude and wavenumber for these being denoted by A and K,
the corresponding surface elevation is given by
ij = Acosd, 6 = a t+ K x + e, (2.1)
where a; is a horizontal coordinate, t is the time. A and e are functions of x and t which
vary slowly compared with the phase в. If cr and К are both positive, the waves move
to the left, as in figure 1.
Superposed on the long waves we assume a much shorter, wind-generated, ‘surface
roughness’ with local mean-square slope s 2. Following Plant (1982) we assume that,
below saturation, s grows exponentially with time:
5 oc e* (2.2)
say, where the growth-rate ft depends upon both the wind-speed W and on the
frequency of the short waves. Here we shall take /? to correspond to the mean short­
wave frequency cr', say. Note that the empirical law (2.2) already takes some account
of the nonlinear interactions among the short waves themselves. However, the short­
wave steepness is assumed to be limited by breaking at a critical steepness s0, say.
For the reasons given in part I we choose
s0 = 0.22, s* = 0.0484, (2.3)
which is consistent with the limits on s found by Plant (1982).
In the absence of wind, the relative steepening of the short waves due to the orbital
motion in the long waves has been shown to be given by
5 oc eyK* (2.4)
approximately, where у = 2.08 (see Longuet-Higgins 1986). Hence (when the short
waves are not limited by breaking) we combine (2.2) and (2.4) to give
s „ efit+ук^ (2.5)
rj being given by (2.1). Thus
3« = \(р+уК.Ъг)/Ы)в, a < s0l 26
d< to, s = «oJ
The case s = s0 corresponds to the parts of the surface where the short-wave
steepness is limited by breaking.
Proc. R. Soc. Lond. A (1991)
2046

A stochastic model of sea-surface roughness. II 407

(a) p
er ^ -
о 0 ^

________ ►в
о ^ ____^
Figure 1. Time-history of short waves on a uniform train of long waves. The broken line
indicates intervals where the short waves are breaking, (a) Q < 1 , (6 ) Q ^ 1 .

Lastly, we assume th at the short waves have a negligible short-term influence on

the long waves, and that the system is in a statistically steady state. Thus there is
an overall balance between the input of short-wave energy by the wind and the loss
of short-wave energy by breaking at the limiting root-mean square (r.m.s.) steepness
V
We shall now explore the properties of this model.

3. A uniform train of long waves

Suppose first th at the wave amplitude A is uniform and the phase e is zero (figure
1). Since the waves are statistically steady, and short-wave energy is being supplied
by the wind, the short waves must be breaking somewhere. Therefore they are
certainly breaking at the long-wave crests, where the short-wave steepening due to
the long-wave orbital motion is at its greatest. Let us follow a group of short waves
as it is carried by the orbital motion of the long waves. We assume th at its group-
velocity is small compared to Act. A s the long waves move forward beneath the short
waves there may come a point P, on the windward face of the long waves, where the
straining of the short waves overcomes the wind input, i.e. ds/dt falls below zero.
From equation (2.6) this point is given by 6 = 6 \ where
f i - y K A a s in # '= 0. (3.1)
We have then sin 6' —Q, (3.2)
where we have written Q = (/3/(г)/уКА . (3-3)
For P to be real we must have Q ^ 1. When Q < 1, P lies above the mean level M t
at which the orbital straining is a maximum. Thus 0 < O' < \n.
Between P and Q the waves are not breaking, so by equation (2.5) we must have
5 = s0e x p [fi{t-t/) + yK(7i-7i'))) (3-4)
where t' and rj' denote the values of t and у when 6 = 6'. In figure 1, 6" denotes the
point where 5 again equals s0. This will be when the exponent in (3.4) vanishes. For
fixed x we have a(t —t') = 6 —6'. So
(0/<t ) (6"-в') + уКА(совв"-со&6') = 0. (3.5)
From (3.5) it follows that
(cos6"-cos6/)/(6 " -6 /) = - Q , (3.6)
where Q is the same combination of constants as in (3.3).
Proc. R. Soc. Lond. A (1991)
 2047

408 M. S. Longuet-Higgins

0 0.4 0.8
Q
Figure 2. The critical phase-angles в ' and в” aa a function of Q = fi/yaA K .

Suppose Q to be given. If в > 1 then equation (3.2) has no solution. Physically this
means that the wind is blowing so strongly that the short waves are everywhere
breaking. If on the other hand Q ^ 1 then equation (3.3) may be solved for 6 \ and
from (3.7) we can then find в".
The results are shown in figure 2, where 6' and OFare plotted against Q throughout
the range 0 < Q ^ I. For example when Q 4 1 (a relatively light wind) then в' is
small and 0" approaches 2n. Thus P and Q approach the crest from either side. This
means that short-wave breaking is confined to the neighbourhood of the long-wave
crests.
On the other hand when Q approaches 1, then both 6' and 6" approach 90°, so P
and Q approach the p o in tif (в = £л). Physically, the short waves are breaking almost
everywhere, and only near the mean water level do we have 5 < s 0. By expansion of
the left-hand sides of (3.2) and (3.7) in powers of the small angles
а.' = \ п - в \ <х" = в " - \к (3.8)
we find a ' = ( l — Я 2)*, a " = 2(1 - Q 2) l (3.9)
In other words, Q is twice as far from Ж as is P.

4. Statistical properties
For comparison with observation, three numerical quantities are specially useful,
namely the mean-square slope

(4.1)

the correlation function

1 f2*
С(ф) = s2(0 )t/(0 -0 ) = 2^J 82{в)т}(в-ф)&ф (4.2)

Proc. R. Soc. Lond. A (1991)

2048

A stochastic model of sea-surface roughness. II 409

and the correlation coefficient
С*(ф) = С{ф)/(аУ)К (4.3)
where a bar denotes an average with respect to x or t. Of particular interest are the
maximum value Cmax of С{ф) and the phase-angle фт at which the maximum is
attained.
Now in the simple model of §3 we have
« * » e S [l+ * m (4.4)
where
—1 + exp [{2ft/а ) [в—в') + 2yAK(ao% в —cos в')] when в' < в < в ",
F(d)
-I:1 otherwise.
(4.5)

Hence s2 = 5о|^ + <^ F(6)d6 (4.6)

(4.7)
and ? = ^ [ 1 + ^ -Г (2 F + -f’2)d(?

Further, in (4.2) we can substitute for rj from equation (2.1) to obtain

C(0) = s;j4(.X:eos0 + 7sin0)/2jt, (4.8)

where X =
2
ГW ) соавйв,
7 2
Y = J - Г F(d) sin вАв.
71Jp
(4.9)

Equation (4.8) shows that C(6) takes its maximum value when ф = ф0, where
cos фи = X/{X* + Г)1, sin ф„ = У/(Хг + Г)* (4.10)
and th at Стм = а\А (Х г + Т ) У (4.11)
Hence also c*M = ( ^ + n V 2 te (4.12)
and C*(0) = X /2 iz, (4.13)

where Z2 = 1 + ^ J (2F+F*)d6. (4.14)

We consider now some limiting values. When Q approaches 1 from below, then
[$* —6')-+0 and from equation (4 .6 ) we have

52/ s J - > 1 . ( 4 .1 5 )

Also from (4 .9 ) and (4 .1 0 ), we see that Z - > 0 and 0 (since F < 0) and Y/X -+ 0 ,
hence
0 O-> — 9 0 °, (4 1 6 )

while from (4 .1 4 ), i2-*(2Tt)s, hence

С*и ^ 0 , C*(0)^0. (4.17)
Note that in physical space (0 decreasing) s2 leads rj by 9 0 °. This is because s2 has a
small dip in the non-breaking region on the rear slope of the wave, which leads the
minimum of tj (the wave trough) by 9 0 °.
Proc. R. Soc. Lond. A (1991)
 2049

410 M. S. Longuet-Higgins

AK
Figure 3. The relative mean-square slope s*/5o and correlation coefficient С
as functions of AK in the limiting case 12->-0.

On the other hand when Q 1 and so the wind-generation factor is relatively

small, breaking will occur only close to the wave crests. Hence в' ->0 and в "^ 2п and
from equation (3.5) we have in the limit
s2/Sq —exp \2уАК(соъ&—1)] (4.18)
throughout 0 < в < 2л. From (4.1) we find then
7 / s \ = e - 4 0{z)y (4.19)
where z = 2yA K (4.20)
and I0 denotes the modified Bessel function of order zero. Similarly from equations
(4.9) we have
X = e~zI 1(z), 7 = 0 (4.21)
so that the phase ф0 vanishes. Also from (4.14)
Z 2 = e - ^ /0(22). (4.22)
Hence by (4.12) and (4.13)
= <7*(0) = 2 '/ , ( 2 ) / [ / 0(2 z )]I. (4 .2 3 )

The above expressions for s2/s l and are plotted as functions of AK = z /2 y in

figure 3.
5. Numerical results
For general values of A K and /?/cr the integrals in equation (4.6)jJ4.7) and j4.9) are
easy to compute, and in figure 4 a-d we have plotted contours of s2/ 5o> Фо> ^max an<^
(5(0) respectively, as functions of AK and p /a . From (3.3) it is clear that the loci
Q = const, are straight lines at a 45° angle. In particular the locus & = 1 corresponds
to the straight line on the lower right-hand corner_of each figure.
I t will be seen that each of the quantities s2/ s l , С*, ф0 and C*(0) varies
monotonically with AK or with p /o ’. In particular, as the long-wave steepness is held
constant but the wind generation factor P/o’ is increased, so s2/s$ increases, as one
Proc. R. Soc. bond. A (1991)
 2050

Proc. R. Soc. Lond. A (1991)

A stochastic model of sea-surface roughness. II

Figure 4. Statistical parameters for short waves on a uniform train of long waves, (a) s2/ 5o» (6) Стлх, (с) ф0, (d) C(0).
 2051

412 M. S. Longuet-Higgins
might expect. More interestingly, for constant wind, s2/s\ is a decreasing function of
the long-wave steepness AK. This is due to the increased amount of short-wave
stretching, away from the long-wave wrests.
As /?/(г-*0, the limiting values of s2/ s 02 and Cmax are given by (4.19) and (4.23)
respectively. Since for small values of z
I0(z) = 1 +i*2, 1 ,(2 ) = \z( 1 + \ z 2) (5 . 1)

we have $7$ 2 = 1 - z + f z 2 (5.2)

and C*mix = (z/ 2 i ) ( l - |22), (5.3)
where z = 2yAK. (5.4)
In general, the phase ф0 appears from figure 4c to be a function of Q alone. At least
for small values of A K and /?/сг this may be understood as follows. From (3.2) and
(3.7),
d' = Q, and 0" = 2я-(4лЯ)* (5.5)
to lowest order in Q , and so from (4.5)
F(d) = 2 (P/a) (в -в ')-¥2 уА К (со8 в-созв') (5.6)
On substituting for F(6) in (4.9) we find, to lowest order,
X = 2nyAK, Y = —4л/?/<г, (5.7)
hence tan ф0 = (2/3/а)/уАК = 2Q. (5.8)
In this limiting case, the mean-square steepness s2 still leads the surface elevation rj
in physical space, but this is due mainly to the exponential growth factor ft from the
wind input. Also since (2к —6") is greater than в' (see figure 2) the region of breaking
extends further down the forward face of the wave than down the next rear face.
In each of figure 4 a-d, the theoretical values to the right of the line Q = 1 are
constant.
6. R andom long w aves: the Rayleigh distribution
So far we have assumed the height of the long waves to be uniform. In a real sea
state having a fairly narrow-band spectrum, the amplitude A of the dominant waves
has a probability density that is closely Rayleigh (see Longuet-Higgins 1980). Thus
V(A) = (A/Al)e~Ali2AK (6.1)
where A 0 is a constant related to the mean-square wave amplitude <A 2> by
C42> = 2A 2. (6.2)
With the aid of (6.1) we may generate the results of §2 to the case when the amplitude
A is not quite uniform but instead varies slowly, the groups of waves being relatively
long.
Since cr is assumed constant, the parameter p / a will be fixed, for a given wind-
speed. The effect of distributing the amplitude A will be to average the values of $2
in figure 3, for example, by means of the probability density j)(A). So we may extend
the definitions of §4 by calculating

<s2> = ( ^ | s 2(0)d<?) (6.3)

Proc. R. Soc. Lond. A (1991)

2052

A stochastic model of sea-surface roughness. II 413

II

long-wave amplitude.
EW с.

Figure 5. As figure 4, but for a Rayleigh-distributed

C>I

Proc. R. Soc. Lond. A (1991)

 2053

414 M. S. Longuet-Higgins
in which < ) denotes _the statistical average. Similarly we may define the average
quantities <C'*(^)), <s4> and <^2)0 , and finally, as in equation (4.3),
C(0) = «T*(0)>/[<?><7‘>]». (6.4)
The angle ф, is defined as previously, as the angle for which С{ф) takes its maximum
value Cmax.
The results are shown in figure 5 a-d as functions of A*К and where
A* = <Л> = (1я)!Л0 (6.5)
denotes the average long wave amplitude. Figure 5a showing <s2>/s;j, may be
compared with figure 4 a. The form of the contours is very similar, except that the
diagonal contour in figure 4a corresponding to Q = 1 is now smoothed over.
Similarly figure 56 showing ф corresponds closely to figure 46, and figure 5c
showing Cmax corresponds closely to figure 4c. In both figures 4a and 5a, for example,
the contours of constant ф0 have a slope of nearly unity, indicating that 0O is a
function of A*K/{cr/B) very nearly.

7. The bivariate Rayleigh distribution

Even in the randomized model of §6, the short waves always break at long-wave
crests. In actual sea states, however, a high wave may often be followed by a
significantly lower wave. Because of orbital stretching, the short waves may not
break at the crests of the second waves, even if they break on the first. Similarly a
non-breaking wave may be followed either by a breaking or by a non-breaking wave.
A general stochastic model which takes into account all such possibilities was
developed in part I. There, the joint probability density of two successive wave
amplitudes А г and A 2 was assumed to be given by the bivariate Rayleigh density:

р{А" Аг) = ( 1 ^ ^ ехр[_(Л?+^ ?)/2(1_лг)^ ]/0( ( Г ^ м |} (7-1)

Here A 2 denotes half the mean-square amplitude as before, 70 is the modified Bessel
function of order zero and к is a ‘groupiness ’ parameter, related explicitly to the
spectral density of 7). For narrow spectra it can be shown that к2 = 1—(2nv)2, where
v is the spectral band-width. For very narrow spectra we have к » 1. This is the
limiting case corresponding to the model of §6. On the other hand for very broad
spectra /c->0 and then A y and A 2 are statistically independent:
p(A v A 2) = р(Аг)р(А2), (7.2)
where p(A J and p(A2) are each Rayleigh densities, as in (6.1).
In part I, it was shown how the assumptions of §2 above enable one to determine
the probability density p(S/A) for the steepness s of the short waves at the crest of
a long wave of given amplitude A v Given s on A v and given the amplitude A 2 of the
next wave, we can now integrate forwards using equation (2.6) to find the value of
s(0) at all points between the two wave crests. In this way we can determine the
statistics of s(6) over the complete surface profile.
As a check, this numerical procedures was first carried out in the limiting case
к = 1, corresponding to A 2 = A lf and it was verified that the calculated values of s2,
ф0, Cmax and 0(0) agreed with those found by the uniform (i.e. slowly varying) model
in §6.
Proc. R. Soc. Lond. A (1991)
 2054

ASt°Cha°4crnod e l o f
f ^ r M e г0ид/1Пш ц

SK W - A 0991)
 2055

416 M. S. Longuet-Higgins
For general values of к, when the amplitudes Ax and A2 are unequal, the
calculation of C(82,7f) requires that we define the elevation 7) over the interval
0 < в < 4л. We assumed that the amplitude varied ‘piecewise linearly’, that is
([А1+ (Аг —Ах)6] costf, О < 0< 2я,
* “ 1 [Аг + (А3—Аг) (в—2ji)]cos0, 2ti < в < 4л,
where А3 denotes the wave amplitude following A2. To take full account of the
variability of A3 would be impractical computationally, but in the case к = 0 we can
reasonably replace Аъ by its constant mean value <^4> = A*.
The results when к = 0 are shown in figure 6 a~d. Again it will be seen that the
contours of s2/sj, ф, Cmax and C0 are qualitatively similar to those in figure 4.

8. Com parison with observation

The strong effect of long-period waves in reducing the amplitude of short surface
waves seems to have been first measured in the laboratory by Mitsuyasu (1966); see
Phillips & Banner (1974). These authors generated long waves mechanically with a
paddle wave maker, in a wind-wave flume. The quantity measured and tabulated by
Phillips & Banner was the proportional reduction (r2) in the mean-square elevation
of the short wind-waves, as a function of the long-wave steepness AK. On the other
hand the ratio s2/sl in the present theory refers to the relative mean-square steepness
of the short waves. The slope spectrum of wind-waves being rather broad, these two
quantities are not identical. Nevertheless one would expect them to behave in a
qualitatively similar way. In fact, figures 7 and 8 of Phillips & Banner (1974) shows
r 2 decreasing monotonically with AK from 1 at AK = 0 to about 0.35 at AK = 0.1.
This is similar to the proportional reduction in s2/sl in figure 4 above, but about twice
as great. Phillips & Banner (1974) attribute their observations and those of
Mitsuyasu (1966) to the presence of a surface ‘wind-drift’ current, which we have
neglected. The effectiveness of the wind-drift current in this respect has been
questioned by Plant <fc Wright (1977).
Similar reductions in the spectral density of the surface elevation have also been
measured in the laboratory by Donelan (1987). He finds that the reduction is greatest
near the peak of the wind-wave spectrum, and becomes relatively small at higher
frequencies, for a given value of AK. This again is in qualitative agreement with
figure 4 a. In his figure 6 , for example, where AK = 0.105, the peak frequency at a
distance 12 = 55 m from the wave maker is 1.4 Hz. The corresponding growth rate,
according to his observations, was /? = ££ = 0.0012 s”1. Hence /?/<r, in figure 3a
above, would be 0.00037. According to figure 3 above the value of s2/sl at
AK = 0.105 is 0.72. In figure 6 of Donelan (1987) the reduction lies between 0.20 at the
peak frequency/ = /p, and 1 at / = /p±0.5 Hz.
Direct measurements of the slope spectrum have been made recently by Miller &
Shemdin (1991; see also Miller et a l 1991). In these data the basic long-wave
steepness was fixed at AK = 0.041 and the wind-speed varied from 4.0 to 10.0 m s” 1
(one measurement at 1.5 m s-1 was rejected). The relevant parameters are given in
table 1 . Here s\ is the measured mean-square_surface slope in the absence of a
mechanically generated long wave (AK = 0) and s2 is the same quantity with the long
wave added. The next column shows the ratio of these two; it is always less than 1.
Proc. R. Soc. Lond. A (1991)
2056

A stochastic model o f sea-surface roughness. II 417

Table 1. Parameters for the experiments of Miller et al. (1991)

* 7 sl
W/{m s *) Л
sl2 s2 s*/s\ / p/ H z A/s' 1 fi/cr (theory)
4.0 0.038 0.022 0.68 3.0 0.039 0.012 0.90
6.5 0.056 0.045 0.80 2.5 0.075 0.024 0.92
9.0 0.063 0.055 0.87 2.0 0.079 0.025 0.93
10.0 0.063 0.059 0.94 2.0 0.079 0.025 0.93

Table 2. Amplitude and slope of the Fourier harmonics of the mechanically generated wave
n л пк п
1 4.11 0.043
2 0.40 0.015
3 0.46 0.038
4 0.04 0.013
5 0.01 0.008
6 0.02 0.003

The next columns of table 1 show /p, the peak frequency of the slope spectrum and
ft, the growth-rate of the wind-generated waves, as calculated from Donelan’s (1987)
formula
0 = 2.05 x 10-b{ULI2/c-X )2A3(o. (8 . 1 )
Here ULI2 denotes the wind-speed at a height of half a wavelength above the waves.
We equate this to W. We also set w = 2 nfp and с = g/o). If we equate sx to s0, then
we may compare the fourth and the last columns of table 1 . The agreement is closer
at the higher wind-speeds, presumably because the short waves are closer to
equilibrium with the wind and the underlying long waves, as assumed in the theory.
It has, however, been noted by Miller et al. (1991) that the mechanically
generalized waves contained higher harmonics which, while contributing negligibly
to the surface elevation, nevertheless produced significant components of the surface
slope. This is shown in table 2, where the amplitudes An of the various harmonics are
given. The harmonic n = 3, for example has an amplitude A3 which is only one-tenth
of the fundamental Ax. But because the frequency crn is proportional to n, the
wavenumber K n is proportional to n2. Hence А3К г is comparable in magnitude to Ax
K x. The ratio Р/<тп, however, is proportional only to n , and so varies less
dramatically.
In this paper the wavenumber К of the long waves has been assumed to be
constant, or nearly so. To take full account of the higher harmonics would require
development of the theory beyond our present limits. However figure 3 suggests that
if the values of AK and f}/cr for the third harmonic were used instead of those for the
first harmonic, then s2/s\ would be somewhat less than the values given in table 1 .
Presumably the effect of both the first and third harmonics acting_together is to
reduce s2/s\ still further, bringing it closer to the observed value of s2/s\.
It should be emphasized that in these experiments the higher harmonics in the
waveform were practically invisible to the naked eye, yet their effects on the mean-
equare slope s2(6) were quite marked (see Miller et al. 1991). Similar effects may have
been present in the previous experiments mentioned above.
Proc. R. Soc. Lond. A (1991)
 2057

418 M. S. Longuet-Higgins
Table 3. Comparison of theoretical and observed values of s2/slfor the Pierson-Moskomtz spectrum
W/(m s-1) fil*9 s2/*l s2= s2„+sl s2 (obs)
6 0.100 0.906 0.051 0.030
8 0.154 0.949 0.053 0.041
10 0.215 0.976 0.054 0.051
12 0.282 0.993 0.055 0.061
14 0.356 0.998 0.056 0.071

9. Field observations
We have seen that the orbital contraction of the long waves is measured not by the
long-wave amplitude A alone, but by the long-wave steepness AK. Thus the
requisite information for the long waves is not the surface elevation spectrum E(a)
which is usually measured, but the slope spectrum
S(cr) = кгЕ(<т)> кг = (т2/д. (9.1)
Often a low-frequency, or swell, component which is visible in the elevation spectrum
E(cr) may not be significantly present in the slope spectrum S{a).
Commonly used expressions for E(<r) such as proposed by Pierson & Moskowitz
(1964) are valid only at frequencies below about 0.5 Hz, and tend to underestimate
the spectral density at higher frequencies. Direct measaurements of the slope
spectrum are rare. Those by Shemdin & Hwang (1988) show a single broad peak, with
a high-frequency behaviour like f~l &approximately.
Field observations of sea surface slopes are known to be strongly influenced by
other parameters such as atmospheric instability, as measured by the air-sea
temperature difference (Hwang & Shemdin 1988). In fact, after removal of the
temperature instability effect from the available data the residuals show a large
scatter and no obvious dependence on the steepness of the dominant waves. (The
value of A used seems to have been calculated from the significant wave height, not
the mean wave height.) It is also unclear whether the ‘significant waves’ tabulated
for example by Cox & Munk (1954) and used in this dataset were sufficiently
separated from the short, steep waves to be counted as swell in the sense of the
present paper.
10. Discussion
The present simplified model of long-wave/short-wave interaction, while including
the features listed in § 1 , has neglected others that may be significant in some
circumstances; particularly the wind-drift surface current discussed by Phillips &
Banner (1974) and the weak nonlinear interactions between the short waves
(Valenzuela & Laing 1972). The importance of the latter relative to the wind­
generated growth rate has been discussed by Miller et al. (1991), in relation to the
laboratory experiments described there. The two growth rates were shown to be of
comparable magnitude. It is, however, arguable that the empirical formulae for the
growth-rate derived by Plant (1982) or Donelan (1987), which we have used, do in
fact include the short-wave interactions, so that these are already accounted for,
though roughly.
A further deficiency of the model is the use of a single frequency to characterize the
short-wave spectrum; especially since the growth-rate due to the wind depends
Proc. R. Soc. Lond. A (1991)
 A stochastic m odel o f sea-surface roughness. II 419
that frequency. However, it does appear from figures 3-5
strongly on the (nrT S П 10 2’ b°th S^ S2° and °™x and °* dePend much more
short г ng-wave steepness AK than on f$/a. Hence the dependence on
Р е г Ы ^ ^ иеПСУ iS n0t aS critical “ first ^PPears.
cases when th m?St ser*ous limitation of the model is that it applies strictly only in
wind-wave ' 6ft ШГ °*ear seParation between the long waves (swell) and the shorter
laboratorvS |П t и S °^в sPectrum- Such a separation can certainly be achieved in the
tpmnprof П f ocean’ weH-documented cases of such spectra in neutrally stable
temperature conditions are still rare.
Wa^ to аРР^ ^ e theory, which suppresses the assumption of
wavp о П11S ^lven *n fcbe Appendix. In this we consider simply an equilibrium wind-
FUm a s*ngle peak frequency or The low-frequency part of the
hicrh f m 1S assumec^ to be given by the Pierson & Moskowitz (1964) formula. The
loft " ГеяиепсУ Part follows the laser observations of Shemdin & Hwang (1988). The
hicrhi* 1^ a^Prc^^matec^ by a single band of frequencies at the mean frequency of the
, Г, an 2 6 *ow'frequency part of the spectrum makes a small but constant
oht£»n л ^°П S л t 0 mean"s4uare slope. The high-frequency contribution s H 2 is
f ITt к " " ^ure above, with the assumption that s%= 0.0484. Whereas s 2L is
•° 6 a c<^ns^?,nt independent of the wind-speed, s2Hincreases with W on account
? M 8 1j1^rease in s2/sl with ft/cr in figure 6 a. The sum (s\ + s2H) is compared with Cox
un s empirical formula for the total mean squared slope in table 3. It will be
Ti6n ^ 616 a 8 reement between theory and observation at about W — 1 0 m s-1.
so e rate of increase in (s^ + s2,) with W is of the correct sign but too low in
a so u va ue. ГЧо doubt the reason is that much of the observed increase in mean-
square slopes with increasing wind-speed is due to the broadening of the slope
spectrum due to the progressive lowering of the low-frequency cut-off. The highest
requencies in the spectrum are affected less by the ‘swell’ than by the short waves
of intermediate frequency on which they ride; it is only the intermediate waves
which are reduced by the ‘swell’.
Indeed, in a broad spectrum of wind-waves we may expect that reduction of the
energy in the peak frequencies by interaction which swell waves will lead to an
increase of energy in the highest frequencies. Evidence of this effect can be clearly
seen m the spectra recorded by Donelan (1974). To represent the phenomenon
satisfactorily an 7i-scale model is required, with n > 2.

T he calculations in §5 were begun while the author was visiting the California In stitu te o f
T echnology J e t Propulsion Laboratory m 1987. The author thanks particularly Victor Z lotnieki for
technical assistance and Omar Shemdin for general discussions. The completion of the work was
made possible by the Office of Naval Research under Contract N00014-86-C-0303 to Ocean
Research and Engineering, Pasadena, California, U.S.A.

A ppendix. Application to an equilibrium wind-wave spectrum

Consider the surface-elevation spectrum E(cr) proposed by Pierson & Moskowitz
(1964) for winds between 0 and 10 m s-1, namely
E{a) = ctg2cr" 5 e ' ^ 4, ^ ^j
where a = 0.0081, /3 = 0.74 and
a ° = g ! W' (A 2)
Proc. R. Soc. Land. A (1991)
 2059

420 M. S. Longuet-Higgins
W being the wind-speed. This has a maximum when
or/cr0 = [4/0/5]* = 0.877. (A 3)
The slope spectrum corresponding to (A 1 ) is
SL(a) = c u r l erW*>\ (A 4)
which has a maximum when
<r/<r0 = (4# = 1.312, (A 5)
differing from (A 3) by a factor 1.5, approximately. We shall denote the frequency
corresponding to (A 5) by crp, so
o-p = 1.312g/W. (A 6 )
Suppose that equation (A 4) adequately represent the low-frequency part of the
spectrum, that is in the range
0 < cr < cr1} a l = n ap, (A 7)
where я. is a factor of order 2. Then in the two-scale model it is reasonable to assume
that the long wave has a mean steepness A*К given by
(A*K)* = \K((AK)*> = iKs[, (A 8 )
where -
4 = J sL(a)A(r. (A 9)

On substituting from (A 4) and changing the variable of integration to x =; АК/Ч)4

we obtain
(A 10 )

When n = 2, for example, this gives

i l = 0.00728, (A ll)
hence
A*K = 0.107 (A 12 )
by (A 8 ). We assign to the long waves the peak spectral frequency <rp given by
equation (A 6 ).
Consider now the high-frequency part of the spectrum. The laser observations by
Shemdin & Hwang (1988) at wind-speeds between 2.8 and 6 .2 m s-1 show that over
the range 1 Hz </< 16 Hz the slope spectrum falls off roughly like /~16, that is
slightly more steeply than in (A 4). Above 16 kHz the fall-off is again steeper. Notice
that any spectrum of the form S oc cr~l‘s over a range cr, < cr < <r2 has a mean
frequency f(,t ,-
J (T/SdoyJ S do- = (ct-j o-j)*. (A 13)

Thus we shall assign to the short waves in the model a mean frequency
o) = (n(7po-q)i (A 14)
where <rp is the peak frequency (A 6 ) and «rq is a cut-off frequency: 2n x 16 rad s
Proc. R. Soc. Land. A (1991)
2060

A stochastic model o f sea-surface roughness. II 421

For the rate of growth ft of the short waves we adopt Plant’s (1982) formula
ft —0.04 (и щ/с )2 (i) cos в , (A 15)
where u+ is the wind friction velocity, с the phase speed and в the angle between the
wind and the wave velocity. Taking u+ = 0.04W, с = g/cj and 0 —0 this yields
ft = 0.64 x 10~4 (W/g)2aA (A 16)
From this and (A 12) we have
/?/(Tp = 0 . 7 3 x l 0 - 4 («crq lf/ g )i
= 0.0024(теИО| (A 17)
if W is measured in m s-1
Suppose first that n = 2. Then knowing the two parameters A*K and /?/wP given
by (A 12) and (A 16) we may enter figure 6 a, for example, to find s2H, the mean-square
slope of the short waves. To this we must add s2Lgiven by (A 1 1 ) to obtain the total
mean-square slope _ _
s* = sl+s*H. (A 18)
In table 3 the results are evaluated at various wind-speeds W and are compared with
the empirical formula
s2 = 5.12x 10-5 W (A 19)
given by Cox & Munk (1954), from sun glitter measurements.

References
Cox, C. S. & Munk, W. H. 1954 Statistics of the sea surface derived from sun glitter. J. Mar. Res.
13, 198-227.
Donelan, M. A. 1987 The effect of swell on the growth of wind waves. Johns Hopkins Univ. APL
Tech. Dig. 8, 18-23.
Hwang, P. A. & Shemdin, О. H. 1988 The dependence of sea surface slope on atmospheric
stability and swell conditions. J. geophys. Res. 9 3 , 13903-13912.
Jahne, B. 1989 Energy balance in small-scale w av es-an experimental approach using optical
slope measuring technique and image processing. In Radar scattering from modulated wind-waves
(ed. G. J . Komen & W. A. Oost), pp. 105-120. Dordrecht: Kluwer.
Longuet-Higgins, M. S. 1980 On the distribution of the heights of sea waves: some effects of
nonlinearity and finite band width. J. geophys. Res. 85, 1519-1523.
Longuet-Higgins, M. S. 1987 A stochastic model of sea surface roughness. I. Wave crests. Proc. R.
Soc. Lond. A 410, 19-34.
Miller, S. J . & Shemdin, О. H. 1991 Measurement of the hydrodynamic modulation of centimeter
waves. J. geophys. Res. 9 6 , 2749-2759.
Miller, S. J ., Shemdin, 0 . H. & Longuet-Higgins, M. S. 1991 Laboratory measurements of
modulation of short wave slopes by long surface waves. J. Fluid Mech. (In the press.)
Mitsuyasu, H. 1966 Interactions between water waves and wind. I. Rep. Res. Inst. Appl. Mech.
Kyushu Univ. 14, 67-88.
Phillips, О. M. & Banner, M. L. 1974 Wave breaking in the presence of wind drift and swell.
J. Fluid Mech. 66, 625-640.
Pierson, W. J . & Moskowitz, L. 1964 A proposed spectral form for fully developed wind seas based
on the similarity theory of S. A. Kitaigorodskii. J. geophys. Res. 6 9 , 5181-5190.
Plant, W. J. & Wright, J . W. 1977 Growth and equilibrium of short gravity waves in a wind-wave
tank. J. Fluid Mech. 82, 767-793.
Proc. R. Soc. Lond. A (1991)
 2061

422 M. S. Longuet-Higgins
Plant, W. J . 1982 A relationship between wind stress and wave slope. J. geophys. Res. 87,
1961-1967.
Shemdin, О. H. & Hwang, P. A. 1988 Comparison of measured and predicted sea surface spectra
of short waves. J. geophys. Res. 93, 13883-13890.
Valenzuela, G. R. & Laing, М. B. 1972 Nonlinear energy transfer in a gravity-capillary wave
spectrum, with applications. J. Fluid Mech. 54, 507.520.
Wenz, F. J . 1976 Cox and Munk’s sea surface slope variance. J. geophys. Res. 81, 1607-1608.
Received 20 February 1991; accepted 25 March 1991

Proc. R. Soc. Lond. A (1991)

2062

J. Fluid Mech. (1993), vol. 248, pp. 449-475 449

Copyright <D 1993 Cambridge University Press

Highly accelerated, free-surface flows

B y M I C H A E L S. L O N G U E T -H IG G IN S
Institute for Nonlinear Science, University of California, San Diego, La Jolla,
CA 92093-0402, USA

(Received 23 January 1992 and in revised form 29 September 1992)

Accelerations exceeding 20 g in surface waves have been observed both in

experiments and in numerically computed flows with a free surface. The present
paper describes a family of analytic solutions which display such behaviour. They are
expressible in parametric form as
z = F sinh a) + iG cosh (D+ yw + iH,
where F , G and H are functions of the time t only, and у is linear in I. w is a complex
parameter which is real at the free surface. The functions F(t) and G(t) satisfy two
nonlinear, coupled ODEs, which can be solved numerically. Typically the solutions
pass through an ‘inertial shock’, or singularity in the time, where the displacements
vary as J5, the velocities as О and the accelerations as H. In this class of solution the
free surface develops a cusp as t-+ oo. In a special case, F and G vary as and the
cusp is reached in finite time. Gravity is neglected, but plays a part in setting up the
initial conditions for the highly accelerated flow.
In future papers it will be shown that more general solutions exist in which the
acceleration is momentarily large but bounded.

1. Introduction
Some interesting examples of very high accelerations occurring in surface gravity
waves were recently reported by Cooker & Peregrine (1990); see also Peregrine &
Cooker (1991). Their numerical calculations of steep waves meeting a vertical wall (or
of two such opposite waves meeting) show that the formation of a wave trough near
the wall is sometimes followed surprisingly by a thin vertical jet which shoots
upwards with accelerations typically of 20-50 g, and occasionally as great as 1 0 0 0 g
(Peregrine & Cooker 1991). The jet is accompanied by correspondingly large
pressures and pressure gradients at the wall itself.
Upward-pointing jets induced by solitary waves meeting a wall have also been
recorded experimentally by Nishimura & Takewaka (1990). Such jets are often seen
against the bow of a ship. It is likely that the axial jets seen in standing waves and
collapsing bubbles (Longuet-Higgins 1983 c) are closely related.
In spite of these calculations and observations, there is little understanding of the
phenomenon, or of the range of conditions under which it occurs. For this purpose,
purely numerical calculations are inadequate, and cannot always describe accurately
the asymptotic form of the fluid flow when the free surface develops a sharp
curvature. What is required is an analytic solution which can unify the different
numerical computations and define the asymptotic behaviour of the flow.
Exact solutions for time-dependent flows with a free surface are rare. A flow which
has some of the appropriate features is the ‘Dirichlet hyperbola’ ; see Longuet-
Higgins (1972). In this flow the free surface has the form of hyperbola with varying
 2063

450 M. S. Longuet-Higgins
angle в between the asymptotes. When в approaches 90°, the velocity and
acceleration become infinite, like Г 5 and Г* respectively, where t is the time measured
from the critical instant. A physical explanation has been given by Longuet-Higgins
(1983c). There are two weaknesses in this solution, however. One is the very
restricted form of the free surface, which does not allow the formation of a cusp -
only a thin, ultimately parabolic, jet as t-+ oo. Secondly, it is not immediately clear
how the solution is to pass through the infinity at time t = 0 .
These considerations prompted the author to search for a more general analytic
solution. In this and two companion papers use is made of a parametric
representation of irrotational flow first suggested by John (1953). The representation
is sometimes called ‘semi-Lagrangian \ The present author has already expressed the
Dirichlet hyperbola and its extension to a rotating free surface, in a semi-Lagrangian
form; see Longuet-Higgins (1983a, 6 ). With this representation, as is shown in the
present paper, it is a simple matter to generalize the Dirichlet hyperbola by adding
an extra term linear in w and t. This has a non-trivial effect; for example it makes
possible the formation of a cusp at the free surface. It also gives rise to a beautiful
family of solutions which have the principally desired feature, the occurrence of an
inertial shock at a certain instant. It can be shown, moreover, that by the addition
of two further terms the infinite pressure and acceleration are replaced by
momentarily large but finite values (see §13 below).
The plan of the present paper is as follows. Since it may be convenient for many
readers, a brief introduction to the semi-Lagrangian representation is given in §2. In
the following section we give the expression for the Dirichlet hyperbola in semi-
Lagrangian form, and in §4 state the first generalization. It is shown that this leads
to two coupled, nonlinear, ordinary differential equations for the coefficients^/) and
G(t) as functions of the time. One of these equations (equation (4.10)) follows from
the free-surface condition, the other, (4.11), from the condition that the flow be free
of singularities within the fluid. In §5 we see that in the simplest case, (4.10) leads to
a simple mapping of the trajectories in the (F, G)-plane. The time-dependence of the
solutions is described in §6 , including the mapping of the ‘shock line’. In §7 we
describe the behaviour of the solutions in the neighbourhood of the shock line, and
show that F and G behave like J* there, just as for the Dirichlet hyperbola. There is
one exceptional trajectory (§ 10) where the time-dependence is like F. Knowing this
behaviour one can integrate up to and across the shock line without difficulty (§8 ).
The mapping of the solutions in the general case is described in §§ 11 and 1 2 . Section
13 contains some discussion and an indication of further developments.

2. Param etric representation

The basis of the following analysis is the semi-Lagrangian method of representing
any time-dependent, irrotational, free-surface flow in two dimensions [x, y). In this
representation, both the complex coordinate z = x+iy and the complex velocity
potential x = Ф+ lxk are given in terms of a complex parameter со, and the time t.
Thus we let
z = z(<v,t), Х= Х(<°’ 1)> (21)
where z and x are analytic functions of cj. On the free surface, where the pressure is
constant, it is supposed (i) that w is real and (ii) that w is a Lagrangian coordinate,
that is id is constant following a particle. Hence the velocity vector (u ' , v ) is given by
u + iv = zt(o),t), o) real, (2.2)
2064

Highly accelerated, free-surface flows 451

a subscript denoting partial differentiation. Moreover, the particle acceleration at the
surface is ztt{cot t), and since the pressure gradient has to be normal to the free surface
it follows from the equations of motion that the acceleration vector ztt is normal to
the free surface, gravity being neglected, or in the frame of reference in free-fall. But
the tangent to the free surface is in the direction of z^w j). Hence the free-surface
condition may be written
Ztt = LRzw, (2.3)
where R(o)yt) is some function of w and t which is real at the free surface.
In the interior of the fluid where a) is not real, (2 .2 ) does not apply, but rather
u + \v = zt[u)*yt) (2.4)
where an asterisk denotes the complex conjugate. The reason for this requirement is
immediately clear, for then we have
dx/dz —u —\v = zf { ( i ) y t ) y (2.5)
where z* denotes the complex conjugate function to z. Hence

й = 2 £ = г > '( ) г > ’ <)- (2-6)

Both sides being now functions of w and not involving oj*, we may integrate with
respect to ш to obtain in general

X= J 2*(w>0 0 (2-7)

An important subsidiary condition is the following. Generally, the coordinate

transformation represented by z((o, t) is regular, z being an analytic function of cj and
t. However, there will usually be singularities, the simplest of which occurs when
zw= 0 , say when z = zQ. In the neighbourhood of such a point
( z - z 0) = (<t>-w0) 2,J+|(w-w 0)22ww+ .... (2 .8 )
Since 2Wvanishes, we see that
((x) (Jq) oc (z 2q)* (2.9)
so that the velocity field (u —iv), represented by z*(a>,t) has a branch point unless
z*((o,t) also has a similar singularity, that is to say
— 0 when w — oj0 . (2 . 1 0 )
The condition (2.10), which is applied at one or more points inside the fluid, will
be seen to play a vital role in the following solutions.

3. The Dirichlet hyperbola

One interesting solution to the equations in §2 is given by
2 = F sinh ы + \Gcosh ш (3-1)
(see Longuet-Higgins 1983a, b), F and G being real functions of the time only. For
then
zM= F" sinh w+iCT cosh w, (3-2)
 2065

452 M. S. Longuet-Higgins
where a prime denotes d/d/, while
zu = F cosh (o + iGsinh iо. (3-3)
Hence the boundary condition (2.3) is satisfied provided that
Fn = -RG, G" = RF. (3.4)
On eliminating R from these equations we have
FF" + GG" = 0. (3.5)
As for the interior condition, it is clear from (3.3) that zu will vanish on the axis
of symmetry (oj pure imaginary) when

(3.6)

and so (3.7)
However, we have also from (3.3) that
z j = F' cosh и) —iG' sinh to. (3.8)
The zeros of z*t therefore coincide with those of zu provided that
F'/F = - G'/G, (3.9)
that is F'G+FG' = 0 (3.10)
or FG = constant. (3.11)
From the two equations (3.4) and (3.11) a differential equation for R(t) and hence
F and G can be obtained (see Longuet-Higgins 19836). An alternative method is
given in Appendix A.
The free surface corresponding to (3.1) is clearly a hyperbola:

(3.12)

with semi-axes equal to F and G respectively. The branch points (3.7) are at the two
foci, and both are annulled by the condition (3.11).
We remark that since the fluid is on one side only of the free surface, it is really
unnecessarily to annul both of the branch points; only the branch point within the
fluid itself need be annulled. Hence we suspect the existence of a more general
analytic expression of this type in which only one branch point is annulled.

4. A first generalization
The key to a more general type of solution than that of §3 is to add to the right-
hand side of (3.1) a term linear in both ш and t. Thus let
z = Fsinh aj + iGcosh w + yw + i//, (4.1)
where у = a + /ft, (4.2)
2066

Highly accelerated, free-surface flows 453

a and /ff being real constants, and H a function of the time only, to be determined.
Then we have
ztt = F" sinh u) + \G" cosh o) + iH" (4.3)
and = F cosh w+ i<?sinh <v+ y. (4.4)
Equation (2.3) is therefore satisfied provided that
F" = —RGy G" = RF (4.5)
as before, and further H" = R(a + /%)• (4.6)
Clearly H" represents an additional upward acceleration of the fluid as a whole.
To free the flow of branch points, any zeros of zw(equation (4.4)) that lie in the fluid
must be annulled by a zero of
z*t = jP'coshw—i6?'sinh<i>-l-/?, (4.7)
since /? = y' by (4.2). Thus we have two equations in ew:
2zw = [F + iG) ew+2 y + {F- iG) е ' ы= 0,
(4.8)
2z* = (F'-iG ')e» + 2/l+(F' + iG')e °=‘ 0 1
which must share at least one common root. The condition for this is the vanishing
of the discriminant of the two equations, that is
(F + iG) 2у (F-iG ) 0
(.F '-iG ') 2/3 {F' + iG') 0 = 0. (4.9)
0 (F + iG) 2y (F-iG )
0 {F'-iG') 2/3 (F' + iG')
(Equation (4.9) may be derived by multiplying each of (4.8) by ew and then
eliminating e2w, еы, 1 and e-w from all four equations.)
Though apparently possessing both real and imaginary parts, the determinant can
be seen to be purely real. For, if we interchange the third and fourth rows, the
resulting determinant has rotational symmetry about its central point except that i
is replaced by —i. The determinant thus equals its complex conjugate, and so is real.
On expansion we find
Ф = ( F 'G + F G y - t fF - y F 'f- W G + yG')* = 0. (4.10)
This is a nonlinear, ordinary differential equation relating F and G. In addition, we
have the second-order equation
FF" + GG" = 0 (4-11)
as before, obtained by eliminating R from (4 .5 ). If F and G can be determined from
(4.10) and (4.11) we can then find R from either of equations (4.5). Then H is found
by integrating (4.6) twice with respect to t.
The branch points of the representation z(a), t) correspond to the two zeros of zw, see
(4.8). These are
ew= [—y± i(F2+ G2—y 2)*]/(F + i£). (4.12)
Since |e"| = 1 , cj is pure imaginary. Hence z, given by (4.1) reduces to
2 = ±i(F2+ G2—y 2)* + yo) + iH. (4.13)
 2067

454 M. S. Longuet-Higgins

X
F i g u r e 1 . (a )
A typical profile of the free surface corresponding to (4.1), showing the two branch
points corresponding to zeros of гы. [F ——1.5, G = —0.8.) (6) Streamlines corresponding to the
solution of (a) when (2.10) is satisfied at the lower branch point. (F' = 1.0, G' = —2.0598.)

Provided (F2+ G2—y 2) > 0, z also is pure imaginary, and the branch points lie on the
imaginary axis (see figure la). This axis intersects the surface at oj —0, z = i((7+//).
Thus provided (F2—y 2) > 0 one of the branch points always lies ‘above’ the free
surface, and one *below ’ the surface, so it is the negative branch point that is to be
excluded.
2068

Highly accelerated, free-surface flows 455

From (4.1) it will be seen that when [F + y) = 0, then for small values of the
parameter o) we have
x = \Fw\ у -(У (1 + ^ )+ Я , (4.14)
so the free surface (to real) develops a cusp. Generally, for small oj, we have хы= F + y
whereas for large &>, хы will have the same sign as F. Hence the range —y < F <0
must be excluded since the free surface would then have a loop.

5. M ap of the solutions: /? = 0
To fix the ideas, let us take first the simplest case: /? = 0 . Equation (4.10) then
reduces to
Ф = {F'G+FG')2- a 2{F'2+ G'2) = 0, (5.1)
that is ( ^ - а ^ 'Ч д а 'С Ч ^ - а 2) ^ 2 = 0. (5.2)
For any given values of F and G this defines in general two directions in the (F , G)-
plane given by
F' -F G ± ol[F2+ G2- o.2)'
G' G2- a 2 ( *)
provided that F2+ G2—a 2 ^ 0. (5.4)
Assuming a to be positive, the set of directions corresponding to the positive sign in
(5.3) is shown in figure 2. The circular region in the centre is excluded by (5.4). The
pattern clearly has a 4-fold rotational symmetry about the origin.
The set of directions corresponding to the negative sign in (5.3) is exactly similar
to figure 2 , but with the opposite direction of rotation; the two patterns are mirror
images of each other.
Equation (5.2) or its solution (5.3) is in fact the condition that at least one of the
zeros of z w be annulled by one of the zeros of z * t . Now the zeros of z * t are given, as
in (4.12), by
ew= [ - p±i(F'2+ G'2- - \ G ' ) (5.5)
and when /? = 0 it is evident that the two roots cannot both annul roots of (4.12) (in
which у = a) except in the special case а = 0 . Thus in general, only one root of the
roots of zw= 0 is annulled. It turns out that choosing the plus sign in (5.3) always
annuls the singularity below the free surface (see figure 1 ) while choosing the minus
sign annuls the singularity above the free surface. We shall be concerned only with
the former case, for which the pattern of directions is as in figure 2 .
It should be noted that while the subsidiary condition (2.10) is satisfied exactly at
all times, this removes only the principal part of the singularity at zu = 0. In figure
1 (6 ) we have plotted the instantaneous streamlines for a solution in which the free
surface is as in figure 1 (a). This shows that on the part of the axis of symmetry below
the branch point the streamlines are very nearly, but not quite, parallel to the axis.
To accommodate the flow, we must assume a weak line sink along the y-axis between
the branch point and у = —oo. Thus, while the solution is not exact in this respect,
it appears as a very good approximation.
We are interested especially in flows that develop upward-pointing cusps, without
loops. This implies that, over some part of the trajectory, F < —a and G < 0. Clearly
the trajectories of interest are confined to the region — oo <F < — a , — oo <G < cl
shown in figure 3 .
 2069

456 M. S. Longuet-Higgins

F/a
F ig u r e 2. Directions of the trajectories of the coefficients F and О in the (F, (?)-plane.

F/a.
F ig u r e 3. Trajectories of [F, 0) in the relevant part of the plane: F < —a ,0 < 1.

The trajectories are of three types. Type A starts from F = —oo, G —a and
becomes asymptotic to the cusp line F ——a, G < 0. Each trajectory crosses the
6?-axis at some point (JP0>0 ) between —oo and —a.
Type В also starts from F ——oo, G ~ a, but crosses the cusp line F = —cl at some
point where G > 0.
2070

Highly accelerated, free-surface flows 457

Type C, the ‘critical trajectory’, is intermediate between Types A and B. It
touches the cusp line F = —a at G = 0. The section of the line F = 0, G < 0 should
also be considered as part of this trajectory.

6. Time-dependence of the solutions

To solve the two coupled equations (4.11) and (5.1) for/*(0 and G(t) we may first
differentiate (5.1) with respect to the time to obtain
AF"+BG* + C = 0, (6 . 1 )
where
A = (.F'G+FG') G - a 2F\ В = (F'G+FG')F- ol2G\ С = (F'G+FG') 2 F'G'.
(6 .2)
Substituting for F" and G" from (4.5) we find
R = C/A (6.3)
where A = (AG-BF) = (F'G+FG')(G2- F 2) - o c 2(F'G-FG'). (6.4)
Then F", G" and H" are found from (4.5) and (4.6). Choosing initial values of F, G,
F' and G' to satisfy (5.1) at a given time t = t0 we can then integrate forwards (or
backwards) to determine these quantities as functions of t.
As one would anticipate from the case a = 0, the solution develops a singularity at
a finite time t = tc at which time A vanishes and R becomes infinite. An explanation
can be given as follows. If we think of (^,0) as coordinates of a particle in the
(F,G)~plane, then (F",G") is the vector acceleration of the particle. Equation (4.11)
then states that the radial component of the acceleration vanishes. Thus the particle
moves as a small heavy steel ball in a thin, frictionless tube which is constrained to
rotate about the origin (0 , 0 ).
Suppose the particle moves on a curved trajectory. If the curvature and the
velocity are both non-vanishing, the particle has a non-zero acceleration normal to
the trajectory. A crisis must therefore occur when the normal to the trajectory passes
through the origin, that is when
FF' + GG' = 0. (6.5)
As confirmation, note that by (5.1)
a 2 = (FG'+F'G)2/(F'2+G'2). (6 .6 )
Hence on substitution in (6.4) we find, after some simplification,
A = (FG'+F'G)(F2F'2- G 2G'2)/(F'2+ G'2). (6.7)
So by (6 .2 ) and (6.3)
_ 2r e r + o _ 8)
(FF' + GG')(FF'-GG'Y
Clearly R becomes infinite when (FF' -I- GG') = 0 and if F'G' is not zero, in accordance
with (6.5).
To interpret the second factor in the denominator, note that on the critical circle
F2+G2 = oc2 (6.9)
we have, again by (5.1) or (6.6),
(FF'-G G ')2 = 0. (6.Ю)
 2071

458 M. S. Longuet-Higgins
Thus the critical circle is also part of the locus A = 0 or of R = oo. However, it lies
outside the region of interest for us.
Lastly we note that a general expression for the curvature к of the trajectory is
F'G"-F"G'
K~ + ' (6,11)
So on substitution for F" and G" from (4.5),

Hence R = ~FF^~+GGi ' (6.13)

where q is the ‘ velocity ’ (F'2+ G'2)VHence if neither к nor q vanishes, R must become
infinite when (6.5) is satisfied.
A physical interpretation of the singularity in terms of the fluid flow was
previously given by Longuet-Higgins (1983c) in the special case a = 0. Essentially
what occurs is an inertial shock necessitated by the kinematics of the flow field,
combined with the local form of the free surface. For this reason it is appropriate to
call the locus (FF' + GG') = 0 the ‘shock line’. It is of course a line drawn in the (Ft
6?)-plane, not in physical space.
In the next section we investigate the behaviour of F(t) and G{t) in the
neighbourhood of the shock line.

7. Neighbourhood of the shock line

In the neighbourhood of the shock line let us write
F = Fc + £, G = GC+ V (7.1)
where the suffix с denotes values on the shock-line itself, and (£, rj) are of order ea,
say, where e 1. Substituting in (5.1) we have then
Ф = [(<?c + , ) 2 - а 2] г +m + Z)(GC+V) SV + № + S)2 - a 2] v n = 0. (7.2)
Solving for f 'In' as before, and retaining terms of order e° and e1 only we find
i ' / i = а + Ы+ст,, (7.3)
aD -FG , aF -O D aO—FD 2G(FG—ccJD)
where а ~ 0г _ аг ' 6 (G*-a*)D' ° ( в г- а 2) 0 (G2 - a 2)2 ' '
and we have written D = (F2+ G2—a 2)*, suppressing the suffices. After some
reduction we find that
c —ab. (7.5)
Suppose now that on the trajectory £ is expanded in a Taylor series in rj:
£= а 7 +|Л?/2 + ..., (7.6)
where Л is a constant to be determined. This implies that
S h ' = dg/d* = a + A*. (7.7)
2072

Highly accelerated , free-surface flows 459

But from (7.3) we have, to order e ,
£'/у' = a-\-{ab + c)7j. (7.8)
Therefore A = ab + с = 2ab (7.9)
by (7.5). Altogether then we have
£ = ar/+ abrj2 (7.10)
on the trajectory.
On the other hand from (4.11) we have
(Fe+ ar+ (< yc+ * )“ o. (7.11)
But from (7.6) £" = (a + \r))i}" + \ri'2. (7.12)
Substituting for E," in (7.7) and noting that the lowest-order terms must vanish, i.e.
aFc -\-Gc —0, (7.13)
we obtain correct to order e2 the equation
(a2 + XFC4-1) rjrj" + AFCrj'2 = 0 (7.14)
for ij(t). The only solution that vanishes as t->tc is
(7.15)
T
1!
-a

, a2-f- AFc -f-1 a2+ 2abFc + 1 (7.16)

wbere 71 ~ a 2 + 2 AFC+ 1 ~ a? + 4abF0+ 1
Now from the expressions for a and b given in (7.4) it is straightforward to establish
the identity
a 2 + l= 2 abF. (7.17)
From (7.16) it then follows that
n =l (7.18)
Thus we have shown that, in the neighbourhood of the shock line, F and G tend
to finite values Fc and Gc, but that velocities, like F' and G\ become infinite (t —tcy 3;
the accelerations and the pressure behave like (/—tc)~*. This generalizes the results
previously obtained in the case cl — 0 (Longuet-Higgins 1972).

8. Integrating through the shock line

As a consequence of the analysis in §7, we now have a method of continuing the
integration of F and G through the singularity. For, when sufficiently close to the
shock line we have from (6 .2 ) and (6.3), since F\G' ос H,
С ~(*-*сГ\ Д~(<-*с)“*. <8Л>
Hence A = С/R ~ { t - t j . (8.2)
Thus Az behaves linearly with t, so by linear extrapolation of J 3 in the neighbourhood
of A —0 we can determine accurately the critical instant tc.
Then, if F(t) and G(t) are integrated up to a time tv say, close to tc, we can find the
value Fc of F at the critical time by linear extrapolation of F with respect to (t —tc)*.
Having found Fc and likewise Gc we can cross over to the time
t2 = 2tc - t x (8.3)
 2073

460 M. S. Longuet-Higgins

t
F ig u r e 4. The behaviour of A and A3 in the neighbourhood оf t = te for the typical trajectory
(F0, g0) = ( - 1.3,0), f ; = 1 .0.

on the far side of the singularity by taking

F(t2) = 2F(tc) - F ( t ll F'2(t2) = F'M , (8.4)
and similarly for G(t2) and G'(t2). Step-by-step integration can then start again from
I = t2.
Figure 4 shows the result of this procedure for a typical trajectory of Type A, when
F0 = —1.30 and F'0 = 1.0. Both A and A3 are plotted as functions of t. It will be seen
that the graph of Avs. t is that of a typical cubic curve, and that A3behaves smoothly
and linearly versus the time.
The corresponding velocity y t = G' + H' (8.5)
and acceleration y tt = G"+H" (8 .6 )
of the surface particle w = 0 are shown in figure 5. In the limit as t-+tc, y tt is
symmetric about t = tc, and y u is antisymmetric. The pressure gradient py is similar
to y tt.
As a practical point we mention that sufficiently close to I —tc it is convenient to
take time increments d/ corresponding to equal increments of the function/ = In A.
Since (t —tc) oc A3 this implies that
dt oc 3|£—£c|d f oc |f—Jj. (8.7)
The time steps in the integration were reduced until the results converged
sufficiently accurately. Typically, as in figures 4-6, d/ was taken as 0.5 x 10-4. In the
neighbourhood of the shock line, defined by \A\ < 0.005, d* was chosen to correspond
to increments 0.005 in In A. The limiting txwas chosen to be the first instant at which
\A\ < 10"e. On the far side of the shock line (/ > t2) the procedure was reversed. Note
that since tc and t2 were not round numbers, neither were the chosen times t, for
t > tx.
2074

Highly accelerated, free-surface flows 461

10 t — ,— ,— ,—

-5

-1 0 J____I___ I___ I 1 I I__|___ |___ I I t___ I___ L

-0 0.1 0.2 0.3

F i g u r e 5.The vertical velocity yt and acceleration yu of the surface particle to = 0 in the

neighbourhood of t = lc, for the trajectory {F0,G0) = ( —1-3,0), F'0 = 1.0.

9. Typical solutions, Types A and В

Figure 6 shows a sequence of surface profiles for the Type A trajectory F0 = —1.3.
These were obtained by integrating backwards from t = 0, when (F, G) =
F'0 = 1 , to t = —5, and then forwards from t = 0 (with the same initial conditions)
to about t = 0.5.
For large negative values of t we may show that

(9.1)

where an and bn are constants. Thus the gradient of the surface at w = ± oo, namely
G/\F\, tends to zero like When t = —oo the surface becomes flat.
As t increases from —oo to about —1 the value of G/\F\ increases to a maximum,
and the surface develops a hollow, as in figure 6 (a), with maximum gradient of about
13° when t = —1 . Then the gradient decreases to zero at t = 0 , when (F , G) crosses the
axis G = 0; the hollow ‘fills up’ and the surface is plane; see figure 6 (6 ). The particle
velocities, however, do not vanish. The velocity vectors are shown at various points
along the surface.
For positive values of t (figure 6 c) the surface becomes convex, and since G < 0 the
gradient at infinity is negative. The shock line is crossed at time tc = 0.14866. As
shown by the velocity vectors, the velocity undergoes a sharp kick, resulting in an
S-shaped trajectory for each particle (see figures 6 d and 126).
As t increases beyond tc the surface develops a sharp curvature at the crest; see
figure 6 (c). It may be shown that for large positive times t
F(t) ~ —a + axГ 1 + a 2 1~2+ ...,
(9.2)
G(t) ~ C+1+ 60 + 6 j t~l + 62 1~2+ ... j
 462 M. S. Longuet-H iggins

F ig u r e 6 (a , 6 ). F o r c a p tio n see fa c in g p a g e .

The condition for a cusp, F = —a, is not attained in finite time, but only in the limit
as t ^ c o . Nevertheless the form of the surface is quite close to a cusp even at
moderately large times t.
Figure 7 shows a similar sequence of profiles for a typical trajectory of Type B,
which does not cross the axis G —0. Thus G is always positive and the surface is
always concave. The profiles were obtained by integrating backwards and forwards
 H ighly a ccelerated, free-su rfa ce flow s 463

т---- 1---- j---- 1---- 1---- 1---- 1---- 1---- 1---- 1---- 1---- 1---- 1---- 1---- г

J___ I___ 1 , I___ I___ I___ I___ I___ I___ I___ I___ I___ I------1------L
-2 0 2

F ig u r e 6. A
sequence of surface profiles for the Type A solution (F0,G0) = ( —1.3,0), F'0 —1,
Я 0 = 0. H'0 =
10.0. (a) - 5 < * < - 1 ; (6) —1 ^ < 0, and (c) 0 ^ i < 0.4532. Velocity vectors are
marked for the particles at w = —5.0 (0.25) 5.0. (d) Typical paths of particles in the free surface.

from time /= 0 (say) with initial conditions (FyG) = ( —2.0,0 .8 ), F'0 = 1. The
behaviour as t ^ —oo is similar to that in Type A trajectory; see (9.1). As t increases
from —oo the surface gradient s^ at (o = oo again increases, so that a hollow
develops. The cusp line is always reached in finite time in this type of trajectory but
 2077

464 M. S. Longuet-Higgins

F ig u r e 7. (a) A sequence of surface profiles for the T ype В solution (^0,(?о) = ( —2.0,0.8), F'0 = 1 ,
H0 = 0 ,H'0 = 3.0, when —5 < I < 1.0600. (6 ) The downward-pointing cusp in (a) a t tim e t = 1.0600.

the cusp points downwards. Beyond F ——a, the surface profile develops a loop, and
the solution becomes unphysical. What happens in practice may be that the sharp
cusp entrains some air below the water surface at this point.
The third type of trajectory, which is intermediate between Types A and В will be
discussed in the following section.
2078

Highly accelerated, free-surface flows 465

10. Type C: the critical trajectory

By definition, the critical trajectory is the unique trajectory passing through the
point (F,G) = ( —a,0); see figure 3.
To integrate the solution we must first determine the behaviour of F{t) and G{t)
near ( —a , 0 ). This point lies on the shock line (F'G-\-FG') = 0, the cusp line F = —a
and the critical circle F2+ G2 = a 2. Following the method of §7 by writing
F = - a ( 1 + 0 , G = ocV, (1 0 .1)
where now £, tj are of order e 2 and e respectively, and substituting in (5.1) and setting
a = 1 we obtain, to lowest order in e,
Фa - £ ,2 + 2 ^ y +
2& ?'2 = 0 . ( 1 0 .2 )
Now assuming £ = \\r)2, £' = Atjtj', (10.3)
where A is a constant to be determined, we have from (10.2) that
( - ЛV + 2AV2+ Xtj2) t)'2 = 0. (Ю.4)
Hence Л2 —ЗА = 0 (10.5)
so A = 0 or 3. ( 1 0 .6 )
The solution A = 0 applies to the part of the trajectory consisting of the line F = —a,
—oo < G < 0, and will be considered later. The solution A = 3 is the first term in the
expansion
£ = §^2 + !b 4 + l i V + . . . , (10-7>
which can be used for starting the integration of (F, G) along the part of the
trajectory corresponding to t < 0 .
In addition we have from (4.11) and (7.1)
(1 + 0 Г + ^" = О. (Ю.8)
To lowest order, (10.7) and (10.8) reduce to
£= № (Ю.9)
and Z"+ 44" = 0 (1 0 .1 0 )
respectively. Differentiating (10.9) twice with respect to t and substituting in (10.10)
gives
4^ " +3 ^ 2 = 0 . (1 0 .1 1 )
The general solution of (10.11) tending to 0 as <->0 is
rj —ctn ( 10. 12)

where ±n(n—l) + 3 n2 = 0. (10.13)

Hence n =f (10.14)
Thus we have rj —ct\ £ = §c2t* (10.15)
to lowest order. In addition
C - 2 £V2 =§Sfc’H,
R = Z 'h — « в г » , (10.16)
Л = -C/R = ЦсН\
 2079

466 M. S. Longuet-Higgins

F i g u r e 8. The surface profiles for the solution corresponding to the critical trajectory (Type C).
(a) / ^ 0, (6) 0.

Thus the approach to the shock line is still described by a power law, but the powers
are different from the general case.
The sequence of surface profiles is shown in figure 8 . Again, as /< —oo the surface
is plane. As t increases the surface becomes concave upwards, with a maximum
” 8™ ЬУ U0..7,
when t = —0.56. As £->0 the surface again reverts to a plane (figure 8b), but with a
non-zero velocity field. In this case, therefore, it appears that the cusp line is reached
in finite time.
2080

Highly accelerated, free-surface flows 467

-(^ o + l ) - 1
F i g u r e 9. The ratio CJC_ shown as a function of — (F0+ 1)"1, when G0 = 0.

C. c+ cjc. -(^o + i r 1
—00 1 .0 0 0 0 - 1 .0 0 0 0 - 1 .0 0 0 0 .0 0
-2 .0 1.0 4 8 3 - 1 .1 8 8 3 -1 .1 3 4 1 .0 0
—1.5 1.0 9 9 6 - 1 .7 2 1 8 -1 .5 6 6 2 .0 0
-1 .3 1 .1 5 3 0 -2 .1 9 6 0 - 1 .9 0 5 3 .3 3
-1 .2 1 .2 0 5 6 -2 .6 2 7 7 - 2 .1 8 0 5 .0 0
-1 .1 1 .3 1 0 2 -3 .6 4 5 1 -2 .7 8 2 10 .0 0
-1 .0 5 1 .4 3 0 0 -5 .0 0 1 3 -3 .4 9 7 2 0 .0 0
T a b l e 1. Computed values of C+./C_ for various values of F0

For small negative values of ty the solution near the origin is self-similar. In fact
by setting ш = FQ, Z = ftz, с = —J and choosing a suitable expression for H{t) the
surface profile (4.1) reduces to the self-similar form

ZocG 3 + 7iG2 + 4 9 f l - ^ i (10.18)

15
studied by Longuet-Higgins (1976).
When t > 0, the trajectory of (F, G) continues with uniform velocity down the line
F = —a, G < 0. It is clear that such a solution, with
z = a(w-sinhw) + iWcosh(t>-ftf0 (10.19)
and R =o ( 1 0 .2 0 )
is possible with any arbitrary values of the constant velocity Valthough since R = 0
the flow is essentially unaccelerated. The question arises: after passing through the
shock line, what should be the appropriate value of K?
This question cannot be answered by direct numerical integration as was done in
the general case, because as F0-+—a the Taylor expansions about (Fc, Gc) which were
previously used have a radius of convergence which tends to zero in the limit. Hence
the convergence is non-uniform, and in the limiting case a Taylor expansion is
invalid.
However, we may approach the question another way by enquiring what is the
ratio of the two constant velocities C_ and C+in (9.1) and (9 .2 ). Does the ratio <7+/C_
tend to a finite limit as F0-+—a?
 2081

468 M. S. Longuet-Higgins
Table 1 shows the values of C_ and C+calculated by numerical time-stepping and
extrapolation of F'{t) and G'(t) to t ——oo and + oo respectively, for a sequence of
values of F0 approaching —a. When plotted logarithmically in figure 9 we see that
they follow rather closely a power law; in fact
C+/C_~(Fq+1)-* (1 0 .2 1 )
approximately. We conclude that the upwards velocity of the crest of the surface
profile should be considered as infinite, in the limiting case. In figure 8 (6 ) we show the
surface profiles developing in time, but scaled by this infinite velocity. Apart from
the scaling factor, the solution is quite similar to the finite solution shown in figure
6 (c), beyond about t = 0 .2 .

11. The general case ft Ф 0

From §5 onwards we have assumed that ft = 0. We come now to the general case
i? + 0 .
In the first place, since у —a+ fit, we can change the origin of the time t so as to
make a = 0. Thus
у = fit where t = t + a/ft. ( 1 1 -1 )
We shall assume this done, and then suppress the tilde.
From (4.10) the expression for Ф becomes simply
Ф = (F'G+FG')2- f t 2[(F -tF /)2+ (G+ tG')2]. (11.2)
To simplify the factor multiplying fi2 we set
F = tP(t), G = t~lQ(t) (11.3)
giving (P/Q+PQ')2- F ( t iP'2+ Q'2) = 0, (И.4)
which can be written
(Q2- f i 2t*)P'2+ 2PQP'Q' + {P2- f t 2) <T = 0. (11.5)
Provided that D = P2+ Q2ft 4 - fi2 > 0 ( 1 1 .6 )
( 1 1 .5 ) defines two characteristic directions in the (P,Q)-plane given by
P' 1 (-PQ/t*±№ri) {U1)
Q '~ t2 Q2J t ' - F K ’ }
As before, if fi > 0, a choice of the positive sign in (11.7) annuls the branch point
which lies below the free surface. The condition ( 1 1 .6 ) implies now that (P,Q) must
lie outside the ellipse _
F P 2+ Q2/t2 = ft2, (11.8)
having semi-axes ft and fit respectively. This ellipses touches the cusp line, which is
given by P = —ft.
The time-dependence can be found by differentiating ( 1 1 .2 ) with respect to t as
before. Substituting for the second derivatives F" and G" from (4.5) we then get
R = C/A (И.9)
where now
С = [F'G+FG,)ZF'G'-ft2(G+tG')G\ 1
( 11. 10)
Л = [F'G+FG'){F2- G 2) - f t 2t[{F, -tF )G + tFG,l
ni
In terms of P and Q, A can be expressed as
Aft2 = (P'Q+PQ')(P2- Q 2/t*)+ft2(P'Q-PQ'y (11-11)
2082

Highly accelerated, free-surface flows

(a)

-5 -4 -3 -2 -1 0
Р
F i g u r e 10. (a ) Trajectories through the point ( —2,0) in the (P, (J)-plane, when P\ is positive.
(6) Trajectories through the point ( - 2, 0) in the (P,£)-plane, when P\ is negative.
 2083

470 M. S. Longuet-Higgins
Now substituting for fP from ( 1 1 .4) we obtain
A = (PQ'+P'Q) (.PP' + QQ'/П [РГ - QQ'/t*)
t2 P'2+ Q'2/tA ' (И.
The factor (PP' —QQ'/t*) can be shown to vanish only on the ellipse (11.3), and the
shock line is given by
[PP' + QQ'/t*) = 0. (11.13)
Now (11.13) can be written
t2P' Q/t2
-Qr = - ^ j r - (1114)
Since t2P'/Q' is also given by (11.7), we see that the equation of the cusp locus is a
function only of P and Q/t2, independently of other variables. And it is the same
equation as in the case /? = 0 treated earlier, except that now P, Q/t2 and /? take the
place of F, G and a.
Suppose then that we set
Q = Q/t2. (11.15)
Then in the (P, §)-plane, but not in the [P, Q)-plane or the (F, (?)-plane, the equation
of the shock line is invariant with respect to the time. So also is the boundary ( 1 1 .8 ),
which becomes the ‘circle’
P 2+ Q2 = ft2, (11.16)
and also the cusp line, P = —/?. It is clearly advantageous, therefore, to map the
trajectories of (F, G) in the (P, $)-plane, where these crucial features stay put (see
figure 1 0 ), rather than in either of the other two planes.

12. Typical solutions: /? Ф 0

Consider the initial conditions. Unlike the case ft = 0 , the origin of time has now
been fixed. Moreover in ( 1 1 .2 ), /3 defines the scale of the velocities. The solutions to
(11.2) and (4.13) have arbitrary lengthscale, but this is related to their timescale by
the chosen velocity f$.
We may assume /3 > 0. For the solution to be valid below the free surface, fit must
be positive, hence t > 0. As a ‘typical’ instant let us take t = 1 . The value of {P, Q)
at time t = 1 then determines the lengthscale of motion. The velocity scale may be
determined by choice of P' (or F') at time 1=1. The value of Q' (or G') is then
determined by (11.2). Thus through any given point (Р^Ф,), at a given time t = 1 ,
there is a singly infinite family of solutions, with parameter P[, say.
To fix the ideas let us take (P1? QJ on the negative $-axis, so = 0 . This has the
advantage that the gradient (P'^Q'x) at the initial point is fixed; for then
Q' = (t2Q)' = t2Q' + 2tQ = t2Q' (12.1)
and so by (11.7) P ‘/Q' = ±(РгIP1- ^ ( 1 2 .2 )
independently of the initial value of P'. Therefore the trajectories passing through
such a point touch either of two fixed lines. These lines are the two tangents from
(Pj.O) to the critical circle 27; see figure 1 0 .
Figure 10 shows the family of trajectories of (P, Q) passing through the typical
point (P1,0 1) = (-2 ,0 ) on the <?-axis. When P\ > 0 (figure 10a) each of the
trajectories tends to an asymptote going to infinity in the upper left-hand quadrant,
as t ->0. The coefficients F and G then tend to constant, non-zero values. On the other
hand i f + oo, then so long as P[ > P ;crU « 1.087, each trajectory tends to a limit
2084

Highly accelerated, free-surface flows 471

F ig u r e 11. Trajectories through the point ( —2 , 1 ) in the (Р,ф)-р1апе.

point ( —ft, Q) on the cusp line (P = —Д). Each trajectory also crosses the shock line
at some instant t = tc , where P and Q behave like (t —£c)*, as described in §7. Since the
limit points of P and Q are not at the origin, F and G must behave asymptotically
like t. The free surface is then of the form
z ~ t(Lsinhoj + iMcosh (jj—fioj), (12.3)
where L and M are constants. This is a cusped curve of fixed shape, with dimensions
increasing linearly with time.
However, when о < p ; < P'.crit the trajectories do not cross the shock line, but
instead tend to limit points on the Q-axis. In the critical case P '^ P lc rit
trajectory touches the shock line.
The asymptotic behaviour of F and G at the limit points can be investigated by
Wntmg F = t(A+0, G = t(M+ v ), (12.4)
where A and fi are constants and £, tj tend to 0 in the limit. On substituting in ( 1 1 .2)
we find to lowest order ^ „
Ф=е4(А2 - Л / ^ 2> (12-5>
which must vanish for all values of t. Hence either A = —ft or fi = 0. The first case
describes the limit points on the cusp line. The second case describes the limit points
on the $-axis.
For sufficiently small values of PJ, the asymptote to the trajectory as /-►0 makes
an angle greater than 45° with the negative $-axis. Hence the trajectory must
intersect the second shock line, touching the circle £ at the point (0 , /?); see figure
1 0 (a).
On the other hand, when P\ < 0 (figure 106), each trajectory has an asymptote
P = P0 as J->0 . It then crosses the shock line and as t-+ oo tends to a limit point on
the negative Q-axis. The free surface is asymptotically flat.
Figure 11 shows the family of trajectories through a point [P^Qx) lying off the
Ф-axis. The family contains some curves which cross the cusp line above the point
 2085

472 M. S. Longuet-Higgins
{—ft, 0). In these, the free surface develops downward-pointing cusps at a finite time
t,which subsequently develop into loops. There exists a family of such trajectories
passing through the point (—/?,0). The limiting form of the free surface is derived
below in Appendix B.

13. Discussion and conclusions

We have found and described a class of simple, time-dependent solutions of the
non-linear equations for potential flow with a free surface which are of special
interest for two reasons: (a) the free surface can form a cusp, in finite or infinite time,
and (6 ) the flow develops an inertial shock (not related to the formation of the cusp)
through which it can pass and continue on the other side. At the shock itself the fluid
velocity, pressure and acceleration become momentarily infinite, but the dis­
placements are bounded.
The solutions described in this paper satisfy the boundary conditions for time-
dependent, irrotational flow exactly. Both gravity and surface tension have of course
been neglected. This means that the solutions apply strictly only to those parts of the
flow where the acceleration greatly exceeds g. The role of gravity is simply to help
set up the initial conditions under which an inertial shock can develop. Since gravity
and surface tension are always present, thesolutionsdescribed here must be regarded as
asymptotic forms to which the actual flow will tend, as is suggested by the
experiments noted in § 1 . However, the range of initial conditions which ultimately
give rise to a flow having this asymptotic form, though a very interesting problem,
is beyond the scope of this paper.
It is natural to ask whether the occurrence of an ‘inertial shock’ is in any way
related to the formation of a cusp at the free surface. However, by considering the
special case of the Dirichlet hyperbola which displays shock behaviour (see Appendix
A) but no cusp, it will be apparent that the two phenomena are not closely related.
The Dirichlet hyperbola is, moreover, an exact solution, with no weak discontinuities
in the interior of the fluid such as occur in the general case (see §5).
Do these weak discontinuities completely negate the interest of these flows in the
general case? In our opinion they do not; because the discontinuities are only weak,
in a certain sense, it may be possible to treat each solution as the first in a series of
approximations to a more exact solution.
The present model is the beginning to a more extensive study of this hitherto
unrecognized phenomenon. In future work, the theory will be developed in several
different directions:
(i) Analogous solutions in which circular functions of ш take the place of
hyperbolic functions in the expression for z ( o j , t ) . These describe motions that are
periodic (wave-like) in the я-direction, though not necessarily in time (see Longuet-
Higgins 1993a).
(ii) By adding to the right-hand side of (4.1) two further terms in sinh2w and
cosh2 w respectively one obtains a class of time-dependent solutions in which the
pressure and acceleration attain very large values near a critical instant, but
nevertheless remain finite (Longuet-Higgins 19936). This dismisses any doubts as to
the physical significance of the model.
(iii) By allowing the coefficients/^/) and G{t) to become complex instead of real, one
can obtain solutions with a sense of rotation, as was previously done for the Dirichlet
hyperbola (Longuet-Higgins 1983a, 6 ). These solutions may be applicable to breaking
waves, particularly plunging breakers, which sometimes develop high accelerations.
2086

Highly accelerated уfree-surface flows 473

(iv) There is reason to suspect the existence of analogous axisymmetric solutions,
in parametric form, which will describe the well-known occurrence of inertial jets in
collapsing bubbles.

This work has been supported by the Office of Naval Research under Grant No.
N00014-91-J-1582.

Appendix A. Particle paths in a Dirichlet hyperbola

Equations (3.5) and (3.11 ) admit an exact analytic solution as follows. Let us write
В = F/F' = —G/G' (A l)

so that F = exp ^ d/j, G =—exp ( J t 4 ,a21

where we have chosen the unit of length so that the constant on the right of (3.11)
equals —1 . Then on differentiating (A 1 ) we have
( l - B ' ) F ' 2+ (l+B')G'2 =FF" + GG" = 0 (A3)
by (3.5). Hence ( B ' - l ) F 2 = [B' + 1 ) G2, (A 4)
and on differentiating each side logarithmically we get
B" 2 B" 2 /A
+ -К--КГТТ—
B ' - l В B' + 1 B'K. (A 5 >

hence BB" + W"‘ = 2. (A 6 )

A first integration yields
B*B'2—B* = constant =1, (A 7)
say, by choice of the unit of time. Therefore
B'2 = ( 1 +£ 4)/В4 (A 8 )

and so £= -f f—^ l dB. (A 9)

We may choose the origin of time at В = 0. It follows from (A 2) that

f ’ =exp[±I(i^ )H (A10)
and that G = —1/F. The path of a particle at the free surface (w fixed and real)
relative to a frame moving downwards with speed Vis now given parametrically by
x = F s inhw, у = Vt—F~xcosh a). (A ll)
When В 4: 1 we have from (A 9) that t ~ IB3, so В ~ (30* and so from (A 10)
F ~ 1±|(30*- (A 12)
Thus the particle velocity becomes infinite like H and there is a typical weak ‘shock
Ifl FLM 248
 2087

474 M. S. Longuet-Higgins
(*)

(b)

F ig u r e 12. Particle paths in a Dirichlet hyperbola, (a) V = 0; (6) V = 2.

In the special case V= 0, the particle paths satisfy xy —constant and all are
hyperbolae; see figure 12(a). The large velocities near /= 0 are not immediately
obvious. However, if we view the trajectories in a moving reference frame by taking
V= 2, for example, we get figure 1 2 (6 ), in which the typical 5-shaped kinks appear.
This shows that the kinks are in a sense artifacts of the reference frame. On the
other hand they could be useful experimentally as a diagnostic for weak shocks.
2088

Highly accelerated, free-surface flows 475

Appendix B. The lim iting form of trajectories through the point ( —/?, 0) as
t—>00
For large values of the time t, the coefficients F and 0 have the asymptotic
expansions
F = -/ ft(l+ a1*2*'+ a2*4*'+...)} G = pt(bxГ+ 6 2 Р + ...). (В 1 )
where v is a negative index. Substituting into ( 1 1 .2 ) and (4.1 1 ) we find to lowest order
Ф a [2(y + 2) (Зу + 2) а г 6 J —(2v)2a\] t*v+2 = 0 (B 2 )
and FF" + GG" a [2v{2v + 1 ) ax+ v(v + 1 ) b\] t2v = 0 (B 3)
respectively. On eliminating the ratio ax/b\ from these equations we obtain a cubic
equation:
7i'3 + 20y2+ 16^ + 4 sb 0 (B 4)
for the exponent v, which has only one real solution, namely
v ——1.7231. (B 5)
Correspondingly we find a 1/6 f = —0.1478. (B 6 )
When ш/Р <€ 1 it is easily shown that the free surface has the self-similar form

2 = й ’ ~ М :ТС 2 _ (»'+1)(2«+1)й + ((-+1)(2|/+1)(3('+1) <B7)

in which Й= and г = (B8)

(see Longuet-Higgins 1976, §9).

R E FE R E N C E S
C o o k e r, M. J . & P e r e g r i n e , D. H. 1990 Violent water motion at breaking-wave impact. In
Coastal Engineering (Proc. ICCE Conf., Delft, Holland), vol. 1, pp. 164-176. ASCE.
J o h n , F. 1953 Two-dimensional potential flows with a free boundary. Commun. PureAppl. Maths
6 , 497-503.
L o n g u e t - H i g g i n s , M. S. 1972 A class of exact, time-dependent, free-surface flows. J. Fluid Mech.
55, 529-543.
L o n g u e t - H i g g i n s , M. S. 1976 Self-similar, time-dependent flows with a free surface. J. Fluid
Meek. 73, 603-620.
L o n g u e t - H i g g i n s , M. S. 1983a On the forming of sharp corners at a free surface. Proc. R. Soc.
Lond. A 371, 453-478.
L o n g u e t - H i g g i n s , M. S. 19836 Towards the analytic description of overturning waves. In
Nonlinear Waves (ed. L . Debnath), pp. 1-24. Cambridge University Press, 360 pp.
L o n g u e t - H i g g in s , M. S. 1983c Bubbles, breaking waves and hyperbolic jets at a free surface.
J. Fluid Mech. 127, 103-121.
L o n g u e t - H i g g in s , M.S. 1993a Highly-accelerated, free-surface flows II. Space-periodic
solutions. (To be submitted.)
L o n g u e t - H i g g i n s , M. S. 19936 Highly-accelerated, free-surface flows. III. Inertial shocks of
bounded intensity. (To be submitted.)
N is h im u r a , Н. & T a k e w a k a , S. 1990 Experimental and numerical study on solitary wave
breaking. In Coastal Engineering (Proc. ICCE Conf., Delft, Holland), vol. 1 , pp. 1033-1045.
ASCE.
P e r e g r i n e , D. Н. & C o o k e r , M. J . 1991 Violent motion as near-breaking waves meet a wall.
Proc. IDTAM Symp. on Breaking Waves, Sydney, Australia, July 1991.
 2089

J.Fluid Mech. (1 994), vol. 258, pp. И5-129 115

Copyright © 1994 Cambridge University Press

Crest instabilities of gravity waves.

Part 1. The almost-highest wave
By M IC H A E L S. LONGUET-HIGGINS' a n d R. P. CLEA V ER 2
1 Institute for Nonlinear Science, University of California, San Diego, La Jolla,
CA 92093-0402, USA
aGas Research Centre, British Gas, Ashby Road, Loughborough, LE11 3QU, UK

(Received 16 April 1993 and in revised form 17 June 1993)

It is shown theoretically that the crest o f a steep, irrotational gravity wave, considered
in isolation, is unstable. There exists just one basic mode of instability, whose
exponential rate of growth fi equals 0.1 23(g/R)\ where g denotes gravity and R is the
radius o f curvature at the undisturbed crest. A volume o f water near the crest is shifted
towards the forward face o f the wave; the ‘ toe’ of the instability is at a horizontal
distance 0.45Л ahead o f the crest. The instability may represent the initial stage of a
spilling breaker. On small scales, the ‘ toe’ may be a source o f parasitic capillary waves.

1. Introduction
Despite much interest during the last twenty years, the problem of how and why
gravity waves break in deep water has remained incompletely solved; for recent reviews
see Longuet-Higgins (1988) and Banner & Peregrine (1993).
As pointed out previously (Longuet-Higgins 1981), a breaking wave does not
necessarily pass through the limiting configuration for steady, steep irrotational waves
first calculated by Michell (1893), with a sharp angle of 120° at the crest (the ‘ Stokes
comer-flow’). The actual flow being unsteady, this limiting configuration is by-passed,
by a less or greater margin. Thus from observation it is found that the mean ratio H /L
o f height H to wavelength L in a breaking wave is usually less than the value 0.141 for
limiting waves. The latter would correspond to a ‘steepness’ parameter ak — nH /L =
0.443. For breakers resulting from the fastest-growing Benjamin-Feir instability it is
found that ak « 0.38 (see Longuet-Higgins & Cokelet 1978, figure 22(c)). In a typical
sea-state, ak may lie between 0.05 and 0.20 (see Holthuijsen & Herbers 1986).
The Benjamin-Feir instability also does not account for the marked horizontal
asymmetry o f steep waves prior to breaking (Kjeldsen & Myrhaug 1980; Bonmarin
1990), although Tulin & Li (1991) have shown that this type of instability does produce
an increased supply of energy to the forward face of the wave.
The general approach o f regarding the initial stages of wave breaking as being due
to an instability o f a steady flow is attractive, however. In the present paper we consider
first the instability o f a single, steep wave crest, as described by the theory of the
‘ almost-highest’ wave (Longuet-Higgins & Fox 1977, 1978). This theory describes the
flow near the crest o f any steady, irrotational wave having slightly less than the
maximum height. The curvature o f the surface profile is everywhere finite. In the ‘ inner
flow’ the surface slope tends to 120° in each direction. We here consider time-
dependent perturbations of this flow. If the crest is perturbed by the addition o f extra
energy, how does the deformation develop? What are the normal modes of
2090

116 M .S . Longuet-Higgins and R. P. Cleaver

perturbation? If any o f the modes are unstable, which mode has the fastest rate o f
growth? How does this compare with observation?
The method o f analysis is similar to that developed earlier for periodic progressive
waves in deep water (Longuet-Higgins 1978 a, b). Here, however, it is applied to an
isolated wave crest; the lengthscales in the basic flow and its perturbations are assumed
small compared to a wavelength L. As will be shown, there do indeed exist both time-
periodic normal-mode perturbations and an exponentially growing normal mode, with
a characteristic rate o f growth.
A further reason for interest in this problem may be mentioned. It is known that a
steep, fairly short gravity wave, with wavelength o f order 10 cm, can develop parasitic
capillary waves on its forward face, leading sometimes to a ‘ capillary bore’ (see
Longuet-Higgins 1992). The question arises: how is such a bore initiated? Is it due to
capillarity combined with the sharp curvature at the wave crest, as is currently believed,
or does it arise from an instability o f the gravity wave crest?
The work will be presented in two papers. In the present paper, §§2-4, we first
describe an improved calculation o f the inner flow, making use o f some new relations
corresponding to the quadratic relations between Stokes’s coefficients in periodic
waves. In §§5 and 6 we proceed to give a method for calculating the perturbations o f
the inner flow, and the corresponding results. These are discussed in Section 7.
In a second paper (Longuet-Higgins, Cleaver & Fox 1994) we will describe the
corresponding results when the lowest-order correction to the outer flow is included in
the calculations.

2. The alm ost-highest w ave: definitions

Consider a steady irrotational wave propagating to the left with speed с as in figure
1. In a frame of reference moving with the phase speed the flow appears steady, and
we may take rectangular coordinates (x,>>) with the x-axis horizontal, the y-axis
upwards and the origin 0 vertically above a wave crest, at such a distance that
Bernoulli’s equation can be written,
P+y+kq2 = 0, (2.1)
where p is the pressure, and q the particle speed. We have chosen the units of mass and
time so that the density p and the acceleration due to gravity g are both unity. On the
free surface p = 0 we have then 9
2y + q2 = 0. к*-*)
If ф and \}f denote the velocity potential and the stream function respectively, then on
the free surface we may take \jf = 0. Further writing z = i(x + i>>) and x = we
haVe 2 y = - (z + z*), q2 = x,xt = (г* # -1 (2-3)
so that (2 .2 ) becomes simply (z + z*)z z* - 1 (2 4 )

For very steep waves the dimensionless parameter

e = q jy/ lc (2-5)
becomes small, q0 being the particle speed at the crest (x = 0). It was shown by
Longuet-Higgins & Fox (1977, 1978) that as e -> 0 the flow near the wave crest tends
asymptotically to a certain ‘ inner solution’, described in Longuet-Higgins & Fox
(1977, referred to herein as LHF1). As unit of length we may choose the vertical
distance of the wave crest from the origin x = 0; see figure 1. Then at the crest
(Л: = 0 ) we have (7 6)
у — 1» q = qo = л/2 , (2 .0 )
 2091

117

F igure 1. Profile of the almost-highest wave, showing coordinates and the 30° asymptotes
(from LHF1).

by (2.2). For large values o f z the inner flow tends to Stokes’s 120° comer flow,
z ~ ( fi* ) ’ as z-> oo. (2.7)
To calculate the complete inner flow LHF1 make the transformation
p-'\X (2 .8)
l* = /?T W
where fi is a positive constant, and then suppose that inside the circle \<o\ = 1 (which
corresponds to the fluid domain xjr < 0) the solution can be expressed in the form
z = (8+iX)l B(a>), (2.9)
where 8 is a positive constant and
B((j) = BQ4- (о + В 2uP + •. • (2.10)
is a power series in u with real coefficients Bn. It is convenient to take 8 = /?. To ensure
that the free surface passes through the wave crest, where x = 0, <o = 1, we must specify
that я ( 1) = г 0+ г , + г г+ ... = r * . (2. 11 )
Since the coefficients Bn are all real, the flow will be symmetric about the line x = 0.
Now to satisfy the free-surface condition (2.4), we have from (2.9)
= K8+ix)-bH(wl (2 . 12 )
where H(oj) = fB(oj) —( 1 + o>) B'{oj) (2.13)
and a prime denotes d/dw. The boundary condition (2.4) then becomes
+ = 1, (2.14)
to be satisfied on the circle M = 1 when r = argw lies in the open interval ( —я, я).
Since w- * may be expanded in a Fourier series:

(2.15)
—oo

sin(rt + i)7C
where = n = 0 , ± 1, ± 2 , (2.16)
(л+|)я
we can substitute into the boundary condition (2.14) and by equating coefficients o f
cos пт obtain an infinite set o f cubic equations for the coefficients Bn.
In LHF1 this system o f equations was solved numerically, together with (2.10),
taking 8 — 10*. On successive truncation o f the system at increasing values o f л a
solution was found which converged satisfactorily. The profile o f the free surface is
plotted in figure 9 o f LHF1.
2094

120 M. S. Longuet-Higgins and R. P. Cleaver

In n
F igure 2. Plot of ln|2?J vs. In я when n = 1, ...,9 0 , showing the asym ptotic behaviour o f the
coefficients.

5. Sm all perturbations: method of approach

To derive the equations governing small perturbations of the inner flow we use the
general method introduced by Longuet-Higgins (1978 b) which was applied there to the
calculation of the normal-mode perturbations of a periodic Stokes wave. Thus if we
take as independent variables the velocity potential ф and the stream function ф in the
perturbed motion, together with the time t, the free surface may be specified by
f = F&t) (5-1)
and the two boundary conditions (kinematic and dynamic) when transformed to the
new variables become , „ „
( * * + * * > -O J+ tf) * - 1 (5-2)
and ( y ^ y - y ^ t) = {ft +f^)Ft + [ \ - ( y t y t+ yt xt)]Ft (5.3)
respectively, to be satisfied when i/r = F. We now write
x = Х{ф, ф) + &ф, ф, t\ у = Y(4>t f ) + ф, /), (5-4)
where X and Y represent the unperturbed flow and £ and 17 are small perturbations of
the same order as F. On substituting these expressions into (5.2) and (5.3), expanding
in a Taylor series about т /r = 0 the lowest-order terms give
2Y{Y\+Y\) = - \ (5.5)
as for the steady flow, and the terms of order F give
1*6 + УФп, = (У1+ Ц ) ч + г п г фЬ+ У^)+[У(У1+ Ц )\*г (5.6)
and Уф^-Уф^ = РфНУгф+Уф)Р„ (5 7)
both equations to be satisfied on xjr = 0 .
It is convenient to write
£= + ig, Z = -Y + iX . (5.8)
 2095

Crest instabilities o f gravity waves. Part 1 121

Then on subtracting i times (5.7) from (5.6) we have
z j ( t + i z , i 9 - - * z , z * e + p ) - K z + z ^ ( z ; f 1 + zjr5 )
+*i[(Z+Z*) (zxz*x- z*xzxx)+(ZJ - zj Z^ZJ] F
(5.9)
to be satisfied on Im (#) = 0 .
We now assume that the perturbation £ may be expanded in a series of positive
powers of (j) and that Fmay be expanded in a Fourier series on the unit circle: that is
£ = Z (bn + ian)a)ny F= 2 (cn cos пт + dn sin nr), (5.10)
n -0 n -0
where an, bn, c n and dn are time-dependent coefficients. As in Longuet-Higgins
(19786), we can always choose the origin of \jr so that c0 = 0.
To expand the boundary condition (5.9) in power series we write, following equation
(2,10)’ z - A2 ( 1 +«)-*£(«),'
zx = iA~\\+a))'H{<o), ' (5.11)
zxx = A~*(l +w)*G(w), ,
where A = (2£)*, Н(ш) is given by (2.13) and
GM = |5(a>)-Kl + w ) 5 » - ( l +*>)2Я » . (5.12)
For the perturbation £ we write
£ = C(w), £, = - i J " 3(l +«)£(«), (5.13)
where Дю) = (l+ w )C #(w). (5.14)
Then (5.9) becomes
- i i [/>(«.) C(w) - Л(ш) Яг)] = Re (6 H Ды) - R M <Ы]

+ 2 i ^ l + c o s r ^ ) + S(w )^ r)J, (5.15)

Where P(a>) = А г(\ + м - ^ Я И ,

g(a>) = J (1 + w)* [йГ*£(о>) + W’5(a»_1)] tf(оГ1),
R(oj) = A( 1+ «A l +^- 1)*Я(б>)Я(о;-1), (5.16)
5 (w) = Re[(l + w) (jj~*H(u)) Н(аГ1) tf(w)
—{w~*2?(w) + аЛ^оТ1)} (1 + (o) tf(w-i) G(o>)].
We may express each of P, Q, R and S as power series in w:

P M = 2 * > ", <2 Н = 2 2 „*Л

—00 —
oo (5.17)
jR(^) = 2 ^nw"> *^(w) = 2 $n 0)11»
where clearly Fn = 0 when л > 0. (5.18)
From (3.1) it also follows that
Q(u) = A(l+a))i/H((o) (5.19)
and so Qn = 0 when n < 0 . (5.20)
From their original forms in (5.15), R and S are symmetric functions of r, so
R.„ = Rn, S.„ = S„ (5.21)
2096

122 M. S. Longuet-Higgins and R. P. Cleaver

and we can write

R(t) = R0+ 2 2 Яя c o s m , S(t) = S0+ 2 2 Sn cos nT■ (5.22)

i i
The two original free-surface conditions are retrieved from the real and imaginary
parts of (5.15):

£ Im [Р И C(«)] = Re [Q(<o) E(u) - R(w) CM] + S(t) F(t),

ot (5.23)
i{R e[/> W C M ]-«(T )P(r)) = - 2 ( l+ c o s r )^ .

Equating coefficients of cosw and sin пт in each of these equations gives four
simultaneous equations for the vector
x —(Aq, aXi..., bQ,bXi..., Cj, c2, . •., dxi d2i ...), (5.24)
see the Appendix. These four equations all have real coefficients. We now seek normal-
mode solutions in which Ъ/bt ~ ft. Thus equations (5.23) yield a matrix equation of the
form , л
PAx = Bx, (5.25)
where >5 and В are matrices of infinite order. Truncating the vector x by setting
Bn = 0 when n > N, say, and letting ant bn, cn, dn also vanish when n > N, we can
choose (4N+2) of the equations (5.25) to determine a sequence of eigenfrequencies or}
and the corresponding eigenvectors xy Further details are given in the Appendix.

6. Results
The computations were programmed in FORTRAN (double or quadruple
precision), with the aid of a CLAMS library routine SGEEV for extracting the
eigenvalues and eigenvectors of a real, non-symmetric square matrix, of order (2 N+ 1)
(see the Appendix). The lower eigenvalues, for N = 30(2)50 are shown in table 1.
They appear to have converged reasonably well. No use has been made of Pade
approximants.
As expected, there is one (double) root p —0, for, from equations (A 7) and ( A ll)
of the Appendix, we see that the elements of the first column of A22 are all zeros. Hence
by (A 1) and (A 2) there is a solution in which p —0, a 0 Ф 0 and all the other
coefficients vanish. This corresponds to a purely horizontal shift of the surface profile
through a distance a0, without change of form.
Table 1 shows further that p has just one positive value, corresponding to an
exponential rate of growth. All the remaining eigenvalues p are pure imaginary,
corresponding to oscillatory normal modes.
The most interesting mode is clearly the exponentially growing mode. Table 1 shows
that by N = 50 the corresponding value of p has converged effectively to
P = 0.0544.... (6.1)
The form of this mode is shown in figures 3 and 4, at successive times /, illustrating
the exponential growth. The stability consists of a forward displacement of the fluid,
confined mainly to the wave crest. The displacement begins rather abruptly at the ‘ toe’
of the instability, which is located on the forward face of the unperturbed wave. The
location of the toe can be defined as the point of maximum upwards curvature of the
perturbation profile. This occurs at a horizontal distance у = —2.3 from the centreline.
 2097

Crest instabilities o f gravity waves. Part 1 123

У
F ig u re 3. Profile of the unstable mode /? = 0.05442 at times t = 0, 10, and 20. The amplitude a
at t = 0 is 0.01.

N ± fi ±0-1 ±<гг ±<r* ±<?A ± ^s

30 0.05452 0.00000 0.08000 0.26393 0.61308 0.80706
32 0.05447 0.00000 0.08026 0.26380 0.61313 0.80701
34 0.05444 0.00000 0.08045 0.26372 0.61316 0.80698
36 0.05443 0.00000 0.08058 0.26367 0.61318 0.80697
38 0.05442 0.00000 0.08068 0.26364 0.61320 0.80696
40 0.05442 0.00000 0.08075 0.26362 0.61321 0.80695
42 0.05442 0.00000 0.08081 0.26361 0.61322 0.80695
44 0.05442 0.00000 0.08086 0.26360 0.61322 0.80694
46 0.05443 0.00000 0.08090 0.26359 0.61322 0.80694
48 0.05443 0.00000 0.08093 0.26359 0.61323 0.80694
50 0.05444 0.00000 0.08097 0.26359 0.61323 0.80694
T able 1. Calculated growth rates fi and eigenfrequencies crm = i/?O
Tfor the lowest normal modes,
when N = 30(2)50.
2098

124 M. S. Longuet-Higgins and R. P. Cleaver

У
F ig u r e 4. Close-up of the profile in figure 3.

У
F ig u r e 5. As in figure 3, but at times t = 20, 25 and 30, indicating qualitatively the later
development of the instability.

The tangent to the unperturbed surface at this point makes an angle of 21° with the
horizontal.
It will be realized that the present analysis is valid for a small perturbation only. If
the amplitude of the perturbation were increased, the theoretical profile would develop
as in figure 5, so that the crest would overturn. But one criterion for the validity of the
perturbation expansion is that the corresponding perturbation in the surface slope
 2099

Crest instabilities o f gravity waves. Part I 125

should not be too large. Thus figure 5 is at best a qualitative indication of the later
development of the flow. A more realistic form is likely to be as in the rough analytic
model given by Longuet-Higgins (1981, figures 7 and 12 ), in which the forward face of
the wave has a finite concave curvature. A strong forwards jet can then erupt from the
tip of the wave (see for example Longuet-Higgins & Cokelet 1976, 1978). However,
these later, nonlinear shapes of the flow must be treated by different analytic models,
or else by numerical time stepping.
By reversing the sign of the eigenvector x in (5.24) we obtain a perturbation which
has a toe on the rear face of the wave. Such instabilities are less often observed. This
may be due to the fact that any small disturbances on the forward face of the wave are
accentuated by orbital straining, whereas on the rear face they are diminished. The sign
of the instability is that corresponding to an excitation of the forward face, rather than
the rear face.
The remaining eigenvalues in table 1 correspond to purely harmonic oscillations of
the flow, without exponential growth. An example is shown in figure 6 . This
corresponds to a bound progressive wave superposed on the unperturbed flow, which
is being swept backwards over the wave crest. Its relative phase speed is much smaller
than the (backwards) velocity of the unperturbed flow, in the chosen frame of
reference.
Returning to the unstable of figure 3, we may convert the rate of instability to more
practical units as follows.
The radius of curvature at the crest of the unperturbed profile is
R = 5.15 (6.2)
in the present units (see LHF1, §9). From dimensional considerations, since the only
physical constant involved is the acceleration due to gravity g , the rate of growth fi
must be of the form
P —C{g/R)\ (6.3)
where С is an absolute constant. In the present units g —1, hence
С = 0.0544 x (5.15)i = 0.123. (6.4)
It can also be shown that the e-folding time of the instability is about two and a half
times the time taken for a particle near the crest to traverse a diameter of the circle of
curvature.
In a progressive gravity wave of length 2я, the ratio of the lengthscales is equal to
e2 (see LHF1, equation (2.4)) where e is the parameter defined by equation (2.5) above.
The ratio of the timescales is therefore e1, by Froude scaling. For such a gravity wave
the ratio of fi to the radian frequency cr of a wave of small amplitude is
fi/a = 0.0544e-1. (6.5)
Lastly, to relate e to the steepness parameter ak of the gravity wave, we have, from
equation (5.4) of LHF1,
ak = 0.443—0.50e2+ 0(e3), (6 .6 )
so that e2 = 2Aak (6.7)
where Aak denotes the difference between ak and its limiting value 0.4432. From (6 .6 )
we then have
PI a = 0.0385(Aa/t)-i. (6 .8 )
This shows that for very steep waves (if they are attainable) the rate of growth, on a
fixed timescale, goes to infinity like (Aak)~*.
2100

У
F ig u r e 6. Profiles of the neutrally stable mode a = 0.263 59 at four different phases, (a) a t = 0,
(fc) a t = (с) a t = n, ( d ) a t = 2тс.

7. Discussion
The existence of an inherent instability in the crest of a steep gravity wave has
important implications for the way in which two-dimensional gravity waves actuall>
break. From our results it seems extremely unlikely that a steep wave will attain the
Stokes limiting form; in practice it must topple over before that steepness is attained
In previous theoretical work on periodic, irrotational waves (Tanaka 1983; Longuet-
 2 101

Crest instabilities o f gravity waves. Part 1 127

Higgins 1986) it was found that the lowest superharmonic perturbation became
unstable at around ak = 0.4292, corresponding to the lowest energy maximum. The
instability analysed in the present paper is very probably a limiting form of this same
instability. To trace the connection requires that we consider the stability of a wave
crest modified by an ‘ outer flow’ corresponding to the rest of the wave, as was
described in Longuet-Higgins & Fox (1978). We shall carry out this analysis in future
work. The present paper has established that there is an instability which is essentially
a local property of an isolated wave crest, and hence may occur in other types of wave,
for example waves in finite depth, or in irregular seas.
On a small scale, where surface tension forces become significant, there arises the
interesting possibility that the true source for the parasitic capillary waves found on the
forward face of a short gravity wave is not the sharp negative curvature at the gravity
wave crest itself, as has been previously suggested, but rather the sharp (positive)
curvature occurring at the ‘toe’ of the crest instability. This conclusion appears to be
consistent with many careful observations of parasitic capillaries. A full analysis will
involve the incorporation of surface tension into the boundary conditions.

The present paper is based in part on Chapters 5 to 7 of R.P.C.’s PhD dissertation

(Cleaver 1981); the numerical calculations have been revised and extended by
M.S.L.-H. who is indebted to Jon Wright for advice on eigenvector subroutines.
The work has been supported by the Office of Naval Research under Contract
N00014-91-J-1582.

Appendix. C alculation of the eigenvalues

It will be seen that the order of equation (5.25) can be reduced by splitting the vector
x into two halves. Thus if we write
и = (д0, flj,...,aN',dx,dv . . . , v = (6 0, by,..-tb^, c2, c#) (A 1)
equation (5.25) becomes
и = A12 fiA2l 0 = A22и, (A 2)
where An denotes the matrix derived from the terms proportional to cos nr in the first
of the two equations (5.22), together with the terms proportional to sin пт in the second
equation, and so on.
Thus we can write
Jim [P(w)C(w)l
iC0S"T ), (A3)
A"'\R c[P(w)C( w) ] -R ( t)F( t), sin ПТ J

and similarly
J Re [Q(w) E(w) - R(w) C(w)) + S(r) F(r),
iC0S"T
smm- J
L (A 4)
21 ‘ —2 ( 1 + cos t) d/ydr,
fRe[/>MC(w)]-i?(T)F(T), *:cosnT), (A5)
21 *|lm [/’(w) C(>v)] sin m J
f —2(1 + cost) d/ydr
^ C0S"r] , (A 6 )
: {Re [Q(w) E{w) - R{w) C(w)] 4 - S(t) F(t), Sin ЛТ J
2102

128 M. S. Longuet-Higgins and R. P. Cleaver

where n runs from 1 to N. In the case n = 0 it is convenient to take the constant in the
Fourier series to be | rather than 1. This multiplies the corresponding coefficients
by 2 .
In this way we find
( /*+>: 0\ /<3-Я(+) i S<+)\
A4 1 ..................)» Л ги .......................I , (A 7)

p</*">
+) i R f
-R<+)\|
\ 0 \ J' J
jl 0 : - j \
A2l — 1 » ^22= 1 ... , (A 8 )
_/*-> : I1 \

T
cb
b
fc
0

1
1
where in the case N = 3, for example,
( (.P0+ P0.) (Р-г + Р- J (P- 2 + P- 2) (Р-3 + Р - Л
(/>.,+Л) (P- 2 +P0 ) (P-3+ P -J (P-, + P~d \ (A9)

(P-1 + P2) (P-z + P,) (P -' + P») (P-b + P - , ) } '

(P- 3+ P3) (P-I + P J (P-b + P J (P-e + Po)/

/(.P-i-Pi) (P- 2 - P 0) (P-z-P-i) (P-l - P ^ ) \

P*-' = I (Р -*-Р г) (P-3- P ,) (P-,-P«) (P -s -P -J) (A 10)
\(J“-3-^3) ( P - - P J (P-b-P,) (P-e-P«) /
and
0 -(fi. + fi,) HQ., + <?„) 4(Q-, + Q-,) 6(Q_, + Q_2)\
(0 - (в , + вг) ЬШо + е .) 2 (Q-. + Q,) 3(fi-, + fi-t)\
0 -(fi, + C J b (fi. + f i j 2 (fi, + fi,) 3(fi-, + e j / '
(A n )

о ч б з + е ,) i - ( f i . + e j 2 ( e , + e 2) 3 ( f i .+ e ,) /
Q' is the same as Q but without the first row. Also Я(+) is similar to Я<+) but without
the first column while
/ (Л ..- Л ,) (Л-2 -Л „) (Л -з-Л -,) (Л-4 -Л _ 2)\
« (->= I (Л -.-Л .) (Л -з-Л .) ( * - ,- * „ ) (Л -5-Л .,) • (А 12)
\(Л-з~Лз) (Л .4 - Л 2) (Л_5 - Л ,) (Л_в- Л 0) /
Я denotes Я(_) without the((S_t-st ) ( S Then
first column. .,- S .,)
S (+) (isS similar
. , - S . 2)\
to Я(+>, while
s<-> = (S- 3 - S .) (S . 4 - S 0) (S_s-S _ ,) . (A 13)
\ (5- 4 - 5 2) (S_6 —5,) (S _ ,-S 0) I

Lastly, •/ = I t t 3 1* (A 14)

while J ' is the same as J but without the first row. Note that the first rows of J and
Q are double the values corresponding to the pattern in the matrices as a whole.
From (A 2) one obtains immediately
f i u - B 1v, ftv = B 2u, (A 15)
 2103

Crest instabilities o f gravity waves. Part 1 129

where B x= A~'AU and B 2 = A~}A22. Whence
fPu = B xB 2u. (A 16)
Therefore Л = fP is a root of the eigenvalue equation
det(B,B 2 —Л/) = 0. (A 17)
For a given order N of truncation, the matrix B xB 2 has (2N+ 1) rows and columns,
half the number corresponding to the matrices in (5.23). We can therefore expect
improved accuracy and speed of solution, at a given value of N.

REFERENCES
B an n er, M. L. & P e regrin e, D. H. 1993 W ave breaking in deep water. Ann. Rev. Fluid Mech. 25,
373-397.
Bonm arin, P. 1989 Geometric properties of deep-water breaking waves. J. Fluid Mech. 209,
405-433.
C le a v e r , R. P. 1981 Instabilities o f surface gravity waves. PhD thesis, University o f Cambridge,
UK, 224 pp.
H o lt h u u s e n , L. H. & H e r b e r s , Т. H. C. 1986 Statistics of breaking waves observed as white-caps
in the open sea. J. Phys. Oceanogr. 16, 290-297.
K je ld s e n , S. P. & M y r h a u g , D. 1980 Wave-wave interactions current-wave interactions and
resulting extreme waves and breaking waves. Proc. 17th Inti Conf. Coastal Engng, pp. 2277-2303.
ASCE.
L o n g u e t - H i g g i n s , M. S. 1978a Some new relations between Stokes’s coefficients in the theory of
gravity waves. J. Inst. Maths Applies. 22, 261-273.
L o n g u e t - H i g g i n s , M. S. 19786 The instabilities of gravity waves of finite amplitude in deep water.
I. Superharmonics. Proc. R. Soc. Lond. A 360, 471-488.
L o n g u e t - H i g g i n s , M. S. 1981 On the overturning of gravity waves. Proc. R. Soc. Lond. A 376,
377-400.
L o n g u e t - H i g g i n s , M. S. 1986 Bifurcation and instability in gravity waves. Proc. R. Soc. Lond. A
403, 167-187.
L o n g u e t - H i g g i n s , M. S. 1988 Mechanisms of wave breaking in deep water. In Sea Surface Sound
(ed. B. R. Kerman), pp. 1-30. Kluwer. 639 pp.
L onguet-H iggins, M . S., C leaver, R. P. & Fox, M . J. H. 1994 Crest instabilities of gravity waves.
Part 2. Matching and asymptotic expansion. J. Fluid Mech. 259, 333-344.
M. S. & C o k e l e t , E. D. 1976 The deformation of steep gravity waves. I. A
L o n g u e t - H ig g in s ,
numerical method of computation. Proc. R. Soc. Lond. A 350, 1-26.
L onguet-H iggins, M . S. & C okelet, E. D. 1978 The deformation of steep gravity waves on water.
II. Growth of normal-mode instabilities. Proc. R. Soc. Lond. A 364, 1-28.
M .S. & Fox, M.J . H. 1977 Theory of the almost-highest wave: the inner
L o n g u e t - H ig g in s ,
solution. J. Fluid Mech. 80, 721-741 (referred to herein as LHF1).
L o n g u et-H ig g in s, M . S. & Fox, M . J. H. 1978 Theory of the almost-highest wave. Part 2. M atching
and analytic extension. J. Fluid Mech. 85, 769-786.
M i c h e l l , J. H. 1893 The highest waves in water. Phil. Mag. 36, 430-437.
T a n a k a M. 1983 The stability of steep gravity waves. J. Phys. Soc. Japan 52, 3047-3055.
T u l i n , M. P . & Li, J. J. 1991 A mechanism for wave deformation and breaking intermediated by
resonant side-bands. Proc. IUTAM Symp. on Breaking Waves, Sydney, Australia July 1991 (ed.
M. L. Banner & R. H. J. Grimshaw), pp. 21-37. Springer.
2104

J. Fluid M ech. (1994), vol. 259, pp. 333-344 333

Copyright © 1994 Cambridge University Press

Crest instabilities of gravity waves.

Part 2. Matching and asymptotic analysis
By M IC H A E L S. LO N GU E T-H IGG IN S1, R. P. C L E A V E R 2
a n d M. J. H. FO X3
1 Institute
for Nonlinear Science, University of California, San Diego, La Jolla,
CA 92093-0402, USA
2Gas Research Centre, British Gas, Ashby Road, Loughborough LEI 1 3QU, UK
3 Berkeley Nuclear Laboratories, Nuclear Electricity pic, Berkeley,
Gloucestershire GL13 9PB, UK

(Received 11 June 1993 and in revised form 16 September 1993)

In a previous study (Longuet-Higgins & Cleaver 1994) we calculated the stability of the
flow near the crest of a steep, irrotational wave, the ‘almost-highest’ wave, considered
as an isolated wave crest. In the present paper we consider the modification of this
inner flow when it is matched to the flow in the rest of the wave, and obtain the normal­
mode perturbations of the modified inner flow. It is found that there is just one
exponentially growing mode. Its rate of growth is a decreasing function of the
matching parameter e and hence a decreasing function of the wave steepness ak. When
compared numerically to the rates of growth of the lowest superharmonic instability
in a deep-water wave as calculated by Tanaka (1983) it is found that the present theory
provides a satisfactory asymptote to the previously calculated values of the growth
rate. This suggests that the instability of the lowest superharmonic is essentially due to
the flow near the crest of the wave.

1. Introduction
The present paper is a sequel to a recent investigation (Longuet-Higgins & Cleaver
1994, referred to herein as LHC) in which the problem of wave breaking was tackled
by considering the stability of the flow in an ‘almost-highest wave’, that is to say the
asymptotic form assumed by the crest of a steep, irrotational gravity wave as the
limiting Stokes corner flow is approached, but while the crest is still rounded (see
Longuet-Higgins & Fox 1977, referred to herein as LHF1). It was found that there
exists just one unstable normal-mode perturbation of this flow, with a form resembling
the initial stages of a spilling breaker.
The question then arises: how is this local instability of a gravity wave crest related
to the known unstable modes of perturbation of a progressive gravity wave? For
example, it was shown numerically by Tanaka (1983) and Longuet-Higgins (1986) and
analytically by Saffman (1985), that the lowest superharmonic perturbation becomes
unstable at a value of the wave steepness parameter ak = 0.4292, corresponding to the
first maximum in the wave energy density as a function of ak. Is this superharmonic
instability essentially the result of the local flow near the wave crest?
Now the ‘almost-highest wave’ considered in LHF1 and LHC is the inner flow near
the wave crest, in terms of a small parameter e which measures departure of the flow
from an ideal Stokes corner flow. Thus we may define
e = q J2 c, 0-1)
 2105

334 Л/. S. Longuet-Higgins, R. P. Cleaver and M. J. H. Fox

where q0is the particle speed at the crest, in a frame moving horizontally with the wave-
speed c. For small values of e the flow near the crest (within a distance of order e2) is
described by the ‘almost-highest’ wave. However, if we wish to calculate the properties
of progressive waves with slightly less than the limiting steepness, this inner flow must
be matched asymptotically to the ‘outer flow’ in the rest of the wave, as described by
Longuet-Higgins & Fox (1978, referred herein to as LHF2). It was there shown that
this approach of working downwards from the wave of limiting steepness predicts
successfully the surface profile, the wave speed, and other integral properties as a
function of the steepness ak.
The process of matching the inner and outer flows introduces a small correction to
the inner flow, of order e30*"1*, where A is the root of a transcendental equation; see
LHF2 . Our hypothesis is that the perturbations of the corrected inner flow will yield
the asymptotic behaviour of the superharmonic instability, and hence show that it is
essentially an instability of the flow in the wave crest.
An outline of the paper is as follows. In §2 we describe the matching technique and
in §§3 and 4 determine the corrected inner flow, as in Fox (1977). In §5 we derive
equations for the perturbation of this inner flow. In fact these are given by a
modification to the correcting equations for the original inner flow, as derived in LHC.
The results of the calculations are described and discussed in § 6 .

2. The matching technique

We shall use the same notation and coordinates as in LHC, choosing units of mass,
length and time so that the fluid density and the gravitational acceleration g are both
unity. The wavelength is taken to be 2n.
As was seen in §2 of LHF2, the fluid domain in any steep, progressive wave may be
divided into three distinct zones: an inner zone I of radius e2centred on the wave crest,
in which typical lengths are of order e2 and typical velocities are 0(e); an intermediate
zone II, where typical lengths and velocities are of order e and e* respectively; and an
outer zone III where lengths and velocities are of order unity; see figure 1 . Thus in Zone
I we have
z = *i> X = «3Xi. (2 . 1)
where zx and X\ are of order unity. From LHF1 we know that
(fi*i)S(l + Л), (2 .2 )
where R = C (i^)-<,+1'* + C*( - i*,)-"-1'", (2.3)
С is a constant and fi is the positive root of

T tanhf = 2 7 2 - <2'4>
viz. fi = 0.7143— As я ,-* 0 0 , so clearly /?->0 and
*1 ~ (§i*A (2.5)
representing the Stokes 120° comer-flow.
On the other hand in Zone III we know from LHF2 (see also Grant 1973) that in
the limit e -> 0 (a sharp-crested, limiting wave), and as ^ - > 0 (the neighbourhood of the
sharp crest), so
2~ (ii*)J[l+ 7(i*)A- 1], (2 .6 )
2106

Crest instabilities o f gravity waves. Part 2 335

---------------------L -------------------- ►

F ig u re 1. (after LH F2). Sketch showing regions of validity for the inner solution (Zone I), outer
solution (Zone III) and matching of the two solutions (Zone II).

where Л is the lowest positive root of

nA TiA я /м #ч\
T tanT = - V 3 ’ (2л)
namely A = 1.8027... (2.8)
and у is a constant. It was found numerically in LHF2, §3, that for a progressive wave
on deep water
(|)5r = 0.131 so у = 0.100. (2.9)
Now by matching the solutions in Zone II, as in §2 of LHF2, we see that the inner
flow (2 .2 ) has to be modified as in equation (2 .6 ), that is by multiplying by a factor
F such that when y, -» oo
F -l+ yd x Y -1
= 1 +y(ie3Xi)A_1
= 1 +эт 0 х,)А“ . (210>
where r) = e3(A_1) = e2,481. (2 . 11 )
This defines the behaviour of the inner flow for large values of Xv

3. Determination of the inner flow

To derive complete equations for the inner flow we introduce as in LHF1 the
transformation
м <1 (зл >
and write
*, = (*+ % )’ B(m) + 7,(8 + B(o>), (3.2)
Where B(u>) = B„+B1a>+Вгшг+... (3-3)
as before, and _ _ ....
% ) = e, +e l w + V + - - (3-4)
By equation (2.10) we have that as x ,-» oo so
2.~ ( г + ;Х1)! ад[1+ у1/ (% )А' ‘]
~ (i+ iA!l)i5((1>)[l+r , ( i + ^ 1)A- 1)
= (i+ %)* B(a>) + yri(S+ В(ы). (3-5)
Comparing (3.2) and (3.5) we see that
B((o) ~ yB(b)) as y, -*■oo, (3-6)
that is, as ы -+- 1 . Hence *
Д ( - 1) = у Д ( - 1). (3-7)
 2107

336 M. S. Longuet-Higgins, R. P. Cleaver and M. J. H. Fox

But from (2.2) and (3.2) we have
B ( - 1)-(f )* -1.3104. (3.8)
Thus we must have
Д ,-Д 1+ Я2 + ... = Я ( - 1) = 0 . 100 . (3 .9 )
We specify also that B(oj) vanishes at the wave crest (w = 1), and so
BQ+ Bv+ B2+ ... = 1) = 0. (3.10)

4. The free-surface condition for B ( gj)

The condition of constant pressure at the free surface may be written

(z+z’ > W = l (41)

(see LHC §2), to be satisfied on \(u\ = 1. The same equation is valid when z is replaced
by zx and x by X\- But fr°m equation (3.1) we have
2H
(*+*>) = T ^ (4-2)

(43)
and from (3.2)
dz = ( J + i ^ « N + ? ( i + i * 1)*-S» M , (4.4)
d(iXi)
where Н(ш) = Щш)-(1+ш)В'(ш) (4.5)
as in LHC §2, while _ _
H(w) = (Л- i) B(to) - H(w) B'(w). (4.6)
Thus from (3.2) and (4.2) we have
z, = Дг( 1+ <o)~>B(ca) + r/A2i-\ I + w)-J-i B(<o), )
dz , 1 (4.7)
- i p - = д - 4 1 + «)! а д + , “ - , (1 +«)-•'-> HM,
dXi J
and A = (2S)\
It is convenient to write the boundary condition for z1 as

<4-*>
where c.c. denotes the complex-conjugate expression. Then on substituting for z, and
dzj/d^, and considering the terms in i/° and t}' respectively we obtain
ш-'>В(ш)Н(<о)Н(а>-1)+ c.c. = 1, (4.9)
and (1 + a>y~* ot~* B(tit) H(o>) H(a>~l)+c.c.
+ (1 + ш)‘-дйГ! B(w) Н(ш) H(aj~')+c.c.
+ (1 + <o)'-A« Н B(u) Н{ш) Й(оГ') + С .С. = 0. (4.10)
On extracting the real factor
[ (l+ w )( l+ & rl)]i(,_A> (4.11)
2108

C rest instabilities o f gra vity waves. Part 2 337

z
S
n N = 50

II
о

и
0 0.047765 0.047761 0.047758
1 -0 .0 7 0 7 1 1 -0 .0 7 0 7 0 6 -0 .0 7 0 7 0 2
2 0.014920 0.014921 0.014921
3 0.003529 0.003531 0.003532
4 0.001474 0.001475 0.001476
5 0.000858 0.000858 0.000859
6 0.000570 0.000571 0.000571
7 0.000371 0.000371 0.000372
8 0.000292 0.000292 0.000293
9 0.000192 0.000192 0.000193
10 0.000167 0.000167 0.000167
U 0.000108 0.000108 0.000108
12 0.000101 0.000101 0.000101
13 0.000063 0.000062 0.000062
14 0.000064 0.000064 0.000064
15 0.000037 0.000037 0.000037
16 0.000041 0.000041 0.000041
17 0.000022 0.000022 0.000022
18 0.000028 0.000027 0.000027
19 0.000013 0.000013 0.000013
T able 1. Coefficients B n in the expansion o f the modified inner solution, equations (3.2) and (3.4)

from each of the terms in equation (4.10) and writing

?= ? = |-|A (4.12)
we find that (4.10) reduces to
в д я и # 'а д + м .
+ B(w) Я(«Г‘) ш” H(o>)+c.c.
+ В(и>-')Н(ш-1)ш‘1Щ(о)+с.с. = 0. (4.13)
The functions В(и>) and H(w) are already given in terms of the known coefficients Bn,
calculated in LHC, §4. Moreover when \<o\= 1, < jjp may be expanded in the Fourier
series
(ov = 2 (o = elT (4.14)
-00
valid in —n < т < л, where
_ Sin («-/>) я (415)
(« -Р )* •
Similarly for ojq. Thus by equating coefficients of cosтт, m = 0 , 1 , 2 , in equation
(4.13) we obtain a sequence of linear equations for the unknown coefficients Bn. By
taking the first (N—1) of these equations and truncating the sequence Bn after n = N
we may solve the equations together with (3.9) and (3.10) to find 2?0, Then N
is allowed to increase until the lower coefficients have numerically converged.
The result of this procedure is displayed in table 1, where B0, Bl t ...,B 19 are given
when N = 30, 40 and 50. It will be seen that the coefficients have converged effectively
to 4 or 5 decimal places.
 2109

338 M. S. Longuet-Higgins, R. P. Cleaver and M. J. H. Fox

У\

У
F ig u r e 2. Free-surface profiles for the cases e = 0, 0.1, 0.2 and 0.3: (a) on the scale of
zx = e*z and (b) on the scale of z.

The corresponding surface profiles are shown in figure 2 for e = 0, 0.1, 0.2 and 0.3.
In figure 2(a) the profiles are plotted with the scaled coordinate zx = e- 2z. In figure 2(b)
they are plotted against z. It should be borne in mind that these represent only the inner
solution near the crest of the wave, so that in figure 2 (6 ) the region of validity is only
within a distance 0(e2) from the wave crest.
The relation between the parameter e, defined by equation (1.1), and the steepness
2110

Crest instabilities o f gravity waves. Part 2 339

ak

F igure 3. R elation between the param eter e and the wave steepness ak. The broken
line shows the asym ptote (4.16).

parameter ak for progressive irrotational waves in deep water may be calculated

directly using Stokes’s expansion. The result is shown in figure 3. The dashed line
corresponds to the asymptote
ak = 0.4432 - 0.50e24 *0.503e3cos (2.143 In e -1.54) (4.16)
derived in LHF2, equation (5.4).

5. The normal-mode perturbations

To find the normal-mode perturbations of the modified flow we proceed as in §5 of
LHC1, that is we write
^ -Z + C , (5Л)
where Z represents the unperturbed, steady flow of equation (3.2) and £is a small, time-
dependent perturbation whose square we shall neglect. The free surface is specified by
fi = 0. ( 5 -2 )
where \frx = Im(Xi) is of the same order as £. Then as in LHC1 we are led to the free-
surface condition
(z+z*)zxz; =i (5.3)
at lowest order, and at order £
ZJ ( 6 4- iZ, Ft) = - \ZX2*(C + {*) ~|(Z4- Z*) (Z* ^4- Z, ф
+Ji [(Z+Z*) (Zx7 * - Z* Z J +(Z* - Zx) ZXZ*]F
- i F„ (5-4)
both to be satisfied on \jrx= 0 .
Now from (3.2) and (4.2) we have

(5.5)

that is Z = Д2(1 +ы)-> J(w), (5.6)

where A = (2Я)> and
S(a>) = В(ы) + 1/Д3<А- 1>(1 + о)'-лД И , (5.7)
 2111

340 М. S. Longuet-Higgins, Л. P. Cleaver and M. J. H. Fox

a known function, expansible in positive powers of oj. From (5.6) we may also write

(5.8)
where
(5.9)
6( oj) = f i ( a O - f ( l + « ) # ( « ) - ( 1 + w ) 2J > ) ,
and a prime denotes d/dw. Similarly if
(5.10)
then C, = -iA -3(I + «)£(«), (5.11)
where £(w) = (l+ w )C /(w). (5.12)
then on substituting in equation (5.4) we obtain

to be satisfied on

In equation (5.14), P, Q, R, S are precisely the same functionals of B^H and G as they
are of By H and G in equation (5.16) of LHC1. Moreover, since B(o)), like B(oj), is
expansible in positive powers of w, all the remainder of §5 of LHC1 applies also. Thus
we may seek normal-mode solutions in which £ and F have a time dependence like e^,
and proceed to determine fi in a similar manner.

6. Results
Calculations of the eigenvalues were carried out with 8= 10* as in LHC1 and e in
the range 0 to 0.105. It was found that all the eigenvalues fi2were real. Moreover, only
the lowest eigenvalue was positive, corresponding to an exponential growth or decay
rate /?0. In figure 4 is plotted as a function of e, and will be seen to be a steadily
decreasing function of e, diminishing from = 0.00292 at e = 0 to zero at about
e = 0.102, apparently. The computation is valid, however, only for sufficiently small
values of e.
It is more revealing to consider the unsealed growth rate fi/ae, relative to the radian
frequency o' of a progressive deep-water wave of low amplitude (o'2 = gk, where к is the
wavenumber). Figure 5 shows ( f i j e o ) 2 plotted against the wave steepness ak. The
curve on the upper right of the diagram corresponds to that in figure 4, the same
portions being shown by solid or broken curves. The remaining plots in figure 5 are of
the frequencies of the lowest (non-trivial) superharmonic mode of a progressive wave
of finite amplitude, as calculated by Tanaka (1983) and Longuet-Higgins (1978, 1986).
The quantity —a\/<r2 is plotted from figure 7 of Tanaka (1983) and table 2(b) of
Longuet-Higgins (1986). These calculations demonstrated that the lowest mode
becomes unstable at around ak —0.4292, where the energy density has a local
maximum (see also Saffman 1985). This point is marked by a cross (x ) in figures 4
and 5.
2112

Crest instabilities o f gravity waves. Part 2 341

e
F ig u r e 4. The rate of growth /?0 of the unstable mode, is shown as a function of e.

ak
F ig u r e 5. Growth rates of the crest instability compared with those of the lowest superharmonic
instability in a progressive, irrotational wave in deep water: О ------О O V 0”)2» from ^ 8ure 4 >
• ----- • —(<r,/<r)2 from Longuet-Higgins (1978, 1986); +------ b —{vjo)* from Tanaka (1983).

From figure 5 it appears that the calculations of the present paper provide an
asymptote for the previous calculations of the lowest superharmonic instability, in the
limit as e->-0 and ak-^(ak)m&x. In other words, the superharmonic instability is
associated primarily with the instability of the flow near the wave crest; it is a ‘crest
instability’.
 2113

342 M. S. Longuet-Higgins, R. P. Cleaver and M. J. H. Fox

У\
F ig u r e 6. Form of the free-surface profile in the growing instability, when e = 0.08.

Secondly, the asymptotic analysis, so far as it has been carried in the present paper,
is valid only for sufficiently small values of e (as one would expect). Figure 5 suggests
that the limit of validity is at about e = 0.08 (ak = 0.440). More accurate calculations
of this instability from the present point of view would have to take account of the flow
not only in the inner zone I (see §2 above) but also in the matching zone II and hence
also the flow in zone III. This would imply corrections depending upon higher powers
of e.
The displacement of the free surface corresponding to the growing unstable mode is
shown in figure 6 , for the case e = 0.08. It will be seen to be practically identical in form
to the case e = 0, shown in figures 3 and 4 of LHC. That is, the fluid near the crest is
displaced forwards, and the ‘ toe’ of the breaker occurs at about y x = —2.3. The chief
discernible difference is in the rate of growth /?0, which is 0.0544 when e = 0 and 0.0421
when e = 0.08. According to figure 4, fil decreases almost parabolically with e, so that
$ = 0.00296-О.ОЗОе2. (6.1)
Hence in the unsealed time frame the rate of growth /?' = fio/ea relative to the radian
frequency cr of a progressive wave in deep water (and of small slope) may be written
P = 0.0544e_1(l -0.05e2), (6.2)
generalizing equation (6.5) of LHC. This is valid only when 0 ^ e ^ 0.08. Since over
this range e2 a 2Aak, where Aak is the difference between the steepness parameter ak
and its limiting value 0.4432, we have also
P % 0.0385(Aafc)“* (l-OAOAak) (6.3)
over the same range.
Tanaka (1983) does not describe the surface profile of the instability in his
calculations, so that changes in the form of the instability in the range
0.4292 < ak< 0.44 cannot be ruled out.
2114

Crest instabilities o f gravity waves. Part 2 343

7. Conclusions and discussion

We have extended the stability calculations for the almost-highest wave to waves of
slightly lower steepness ak, taking into account the lowest-order perturbation to the
inner flow, which is of order е3(Л-1). As before, there is just one unstable mode for the
inner flow, with a form resembling the initial stages of a spilling breaker. Its rate of
growth is a decreasing function of the perturbation parameter e. The curve of p j e
as a function of the wave steepness parameter ak is found to be a satisfactory
asymptote to Tanaka’s previously calculated values of the lowest superharmonic
instability (n = 2). The form of the instability is almost independent of e and к when
0.08, but the rate of growth depends strongly on e.
The lowest superharmonic instability becomes unstable at the lowest value of ak for
which the wave energy density is a maximum {ak = 0.4292). As mentioned by Tanaka
(1985, §4) the next highest superharmonic mode (n = 3) becomes unstable also, but at
the next highest stationary value of the energy, as a function of ak. We may conjecture
that this mode also tends asymptotically to a form described by the theory of the
almost-highest wave.
From these results we have drawn the conclusion that the reason for the existence
of the superharmonic instability is the form of the flow in the crest of the wave; it is
essentially a ‘crest instability’. The fact that the superharmonic mode first becomes
unstable at a stationary value of the energy in the wave as a whole need not deter us
from this conclusion. For the maximum of the energy can itself be regarded as a result
of the peculiar form of the flow in the almost-highest wave, as described by the inner
flow near the wave crest.
The form of the inner flow described in LHC1 is not of course limit to steep, periodic
waves in deep water; it will apply also to periodic waves in water of finite depth, or to
solitary waves in shallow water, for example. However the modification to the inner
flow, which depends on the nature of the outer flow, will be different in the various
cases. From our results for deep water, it seems likely that this modification to the inner
flow will affect the rate of growth, but not the form, of the crest instability.
Finally, it is possible that similar conclusions will apply to the final breaking of
waves that are not strictly periodic, such as those associated with wave groups which
arise from swbharmonic instabilities of the Benjamin-Feir type; see Longuet-Higgins
& Cokelet (1978), Dold & Peregrine (1987). Such numerical calculations certainly
suggest that the ultimate overturning of the crest of a gravity wave has a characteristic
form, largely independent of its previous history of growth.
In mixed seas, having a broad frequency spectrum, the situation is different. There,
the field observations by Holthuijsen & Herbers (1986) show that breaking can occur
at much lower wave steepnesses; ak = 0.3 is a typical value. Such breaking is probably
initiated by a more global type of instability, leading to strong fore-and-aft asymmetry
of the waves prior to overturning. The process of formation of the final jet, however,
may have some features in common with the much smaller jet arising from a crest
instability.
The implications for very short gravity waves should also be recalled from LHC.
There it was pointed out that as soon as the crest instability grows sufficiently to induce
a sharp curvature on the forward shape of the wave, surface tension may intervene to
produce parasitic capillary waves ahead of the sharp curvature, as is often observed.
There may also be a ‘capillary roller’ following the sharp curvature, which will lead to
the roughness and ultimate crumpling of the wave crest - a quite distinct style of wave
breaking; see Longuet-Higgins (1993).
 2115

344 M. S. Longuet-Higgins, R. P. Cleaver and M. J. H. Fox

The present paper is based in part on Chapter 7 of Fox (1977) and chapter 8 of
Cleaver (1981). The calculations have been completely reworked and extended by
M.S.L.-H. This study has been supported by the Office of Naval Research under
Contract N00014-91 -J-1582.

R E FE R E N C E S
C le a ve r ,R. P. 1981 Instabilities of surface gravity waves. PhD thesis, University of Cambridge.
D o l d , J. W. & P eregrine , D . H. 1987 Water-wave modulation. In Proc. 20th Inti Conf. on Coastal
Engng, Taipeh, Nov. 1986, pp. 163-175. ASCE.
Fox, M .J .H . 1977 Nonlinear effects in surface gravity waves. PhD. thesis, University of
Cambridge.
G r a n t , M. A. 1973 The singularity at the crest of a finite amplitude Stokes wave. J. Fluid Mech. 59,
257-262.
H o l t h u ijse n , L . H . & H e rbe rs , Т . H . C . 1986 S ta tis tic s o f b r e a k in g w a v es o b serv ed a s w h ite-ca p s
in th e o p en se a . J. Phys. Oceanogr. 16, 290-297.
L o n g u et -H ig g in s , M. S. 1986 Bifurcation and instability in gravity waves. Proc. R. Soc. Lond. A
403, 167-187.
L o n g u et -H ig g in s , M. S. 1993 The crest instability of steep gravity waves or How do short waves
break? Proc. Air-Sea Interface Symp. June 1993, Marseille, France (ed. M. A. Donelan) (to
appear).
L o n g u et -H ig g in s , M. S. & C le ave r , R. P. 1994 Crest instabilities of gravity waves. Part I. The
almost-highest wave. J. Fluid Mech. 258, 115-129 (referred to herein as LHC).
L o n g u et - H ig g in s , M. S. & C o kelet , E. D. 1978 The deformation of steep surface waves on water.
II. Growth of normal-mode instabilities. Proc. R. Soc. Lond. A 364, 1-38.
L o n g u et-H ig g in s, M .S . & Fox, M .J.H . 1977 Theory of the almost-highest wave: the inner
solution. J. Fluid Mech. 80, 721-741 (referred to herein as LHF1).
L o n g u et - H ig g in s , M .S . & Fox, M . J . H . 1978 Theory of the almost-highest wave. Part 2.
Matching and analytic extension. J. Fluid Mech. 85, 769-786 (referred to herein as LHF2).
S a f f m a n , P. G. 1985 The superharmonic instability of finite amplitude water waves. J. Fluid Mech.
159, 169-174.
T a n a k a , M. 1983 The stability of steep gravity waves. J. Phys. Soc. Japan 52, 3047-3055.
T a n a k a , M. 1985 The stability of steep gravity waves. Part 2. J. Fluid Mech. 156, 281-289.
2116

1834 J O U R N A L OF P H Y S I C A L O C E A N O G R A P H Y V o l u m e 24

Reprinted from J o u r n a l of P h y sica l O c ea n o g ra ph y , Vol. 2 4, No. 8, August 1994

A n o i c u MororolopcaJ Society

A Fractal Approach to Breaking Waves

M i c h a e l S . L o n g u e t -H ig g in s
Institute for Nonlinear Science. University of California, San Diego. La Jolla, California
24 June 1993 and 29 November 1993

ABSTRACT
Fractal models of breaking waves in a random surface should preferably describe dynamical as well as geo­
metrical properties. This becomes feasible if there is a wide separation between the length scales of component
waves. Using this idea, a simple model of breaking waves is constructed, which shows that whereas the downward
acceleration of particles at a wave crest is limited to g, the upward accelerations in a wave trough are unbounded.
Owing to tangential stretching or contraction, certain phases of a progressive or standing wave can be identified
as being stable or unstable. The most striking instabilities are expected on the forward slopes of progressive
waves, and in the troughs of steep waves meeting a vertical wail.

1. Introduction or their immediate consequences, did not enter directly.

Such a task is indeed very difficult.
Because of the involvement of breaking waves in The dynamical discussion of superimposed trains of
many important transfer processes at the sea surface, finite-amplitude surface waves is much simplified if we
much attention has been given in recent years to un­ assume that 1) the waves are all pure gravity waves on
derstanding the various types and mechanisms of wave deep water and 2) there is a large proportional differ­
breaking in deep water. For reviews, see Longuet-Hig­ ence in scale between each component wave and the
gins (1988) or Banner and Peregrine (1993). Here we next largest wave. In the present paper we use these
shall consider some aspects of a breaking sea state from assumptions to prove certain properties of breaking
a fractal point of view. waves. Thus, whereas the maximum downward accel­
The idea of self-similarity in a wind-driven sea state eration in an upward-point crest is equal to g (section
is not new. Beginning with Phillips (1958), various 3), the maximum upward acceleration is shown to be
power-law spectra, which imply a dynamical self-sim­
unlimited; see section 4. The very high upward accel­
ilarity, have been suggested. To lowest order in the
eration may be realized in the smooth interaction of
surface slopes, such sea states are generally considered
as Gaussian, so that the phases of the various harmonic random waves with a vertical wall, as has been found
components are uncorrelated. Correlation between the in the numerical calculations by Cooker and Peregrine
phases occurs only at higher orders. (1991). In sections 5 and 6 we discuss the reasons why
A fractal representation, however, involves generally certain phases of a progressive or standing wave may
not only self-similarity with respect to different length be more or less stable to short-wavelength disturbances.
scales, but also a nonrandom relation between the A discussion follows in section 7.
phases. Recently, Huang et al. (1992) showed that the The dynamical mechanisms that will be described
“unwrapped phase” of an ocean wave record has a are fractal in character, insofar as they involve ampli­
fractal property, but no clear dynamical explanation fication of the tendency toward breaking when wave
is forthcoming. A rather thorough analysis of photo­ components of different scales are added in a self-sim­
graphic data of the sea surface in terms of a multifractal ilar manner.
representation has been performed by Kerman (1990)
(see also Kerman 1993). While successfully separating
out a “breaking wave” or fractal component of the 2. Superposing self-similar waves
data from a “nonbreaking” or Gaussian component,
this analysis was essentially that of a static surface, fro­ In the simplest case, imagine a short surface wave
zen in time. The dynamical equations of the surface, of shape given in rectangular coordinates (x , y) by у
= f(x), as in Fig. 1. On this is superposed a similar
wave smaller by a factor p, and on top of this a still
Corresponding author address: Dr. Michael S. Longuet-Higgins,
smaller wave reduced by a factor p2, and so on to (#
Institute for Nonlinear Science, University of California, San Diego, + 1) waves where N is finite. Analytically, the surface
La Jolla, СА 92093-0402. elevation rj is given by

& 1994 American Meteorological Society

 2117

AUGUST 1994 N O TES AND C O R R E SPO N D E N C E 1835

Fto. 2. Superposition of profiles n “ / ( j) with nonzero

FIg. 1. Superposition of wave profiles у - f( x ) having differing phase shift t.
length scales, with zero phase shift at x « 0.
curvature in the first wave will behave locally as if it
were propagated on a steady stream parallel to the sur­
v =/( JC) + />/(X/р) + p1f ( x ! p 1) face of the first wave, but in a field of gravity
+ ■■• + p"A x/f " ). ( 2. 1 )
g ' = g - a = {1 - A)*. (3.2)
At the origin x = 0 the elevation is given by
Accordingly a particle at the crest of the second wave
1- p"*' will have a downward acceleration
1?=/ ( 0) (I +P+ • • • + р * ) = Л 0 )
1-P (3.3)
(2.2)
relative to the first wave. A third wave riding on the
which is finite as N -*> oo. The surface gradient 0) second will therefore experience a gravitational field
is given by
g * = gf — a ' = ( I —X')( 1 —X)g, (3.4)
VX=M0)(N+ 1 ), (2.3)
and so on. After the (N + 1)th wave is added, the
which vanishes in the case of a profile symmetric about effective gravity will be
x = 0. The curvature at the origin is given by
* < « - ( l - X ) ( l - X ') - - - ( l - X < " « > ) f . (3.5)
Tlx* =/xx(0)( 1 + p -' + p - 2 + . • • + p - N), (2.4)
If the waves are all self-similar, then clearly g {ff) -*■ 0
which tends to infinity with N. Thus, in the limit the as N-*- oo. Hence the total downward acceleration of
surface must develop a sharp point. a particle, which is given by
A more interesting case is when the normal dis­
placement n at the rth stage of the superposition is + a' + a' + (* -* ') + (* '-* ') +
given as a scaled function p rf ( e + s/pr) where s is the
arc length measured along the surface after the rth wave = g - g lU), (3.6)
is added,« being some phase constant. Then by a sim­ will tend to g (cf. Longuet-Higgins 1985).
ilar argument, the coordinates ( x , y ) of the crest will This result is in accordance with our expectation
remain finite as TV-*- oo, but the surface curvature will that at any sharp point in the free surface of a fluid the
become infinite. If <is independent of r, the angle be­ pressure gradient, which always has a zero component
tween the tangent and a fixed line will also increase tangential to the free surface, should be identically zero.
indefinitely so that locally the surface will take the form Hence, if the ordinary momentum equation
of a tight spiral (see Fig. 2).
a = -V p + g (3.7)
3. Acceleration of a particle at a wave crest applies there, we might expect a = g.
Returning to the original case <= 0, as in Fig. 1, let However, this expectation fails in the next example.
us suppose that the original profile у =/( x) represents
the surface of a Stokes gravity wave in deep water. The 4. Acceleration of a particle in a wave trough
downward acceleration a of a particle of water as it
passes through the wave crest will be some fraction of Consider on the other hand the acceleration of a
g %the acceleration of gravity, say particle in a wave trough (Fig. 3). In a single Stokes
wave, the upward acceleration has a magnitude a given
a = \g, 0 < X < 0.39 (3.1) by
(see Longuet-Higgins and Fox 1977). A second gravity
wave whose length is short compared to the radius of a = \g t 0 < X < 0.30; (4.1)
2118

1836 JOURNAL OF PHYSICAL OCEANOGRAPHY V o lum e 24

see Longuet-Higgins (1985). A similar, much shorter

wave near the trough of a first wave then experiences
an effective gravity horizontal current U interact with it so as to produce
an additional momentum flux associated with the short
g 1 = a + g = (1 + \ )g (4 .2 ) waves. The additional horizontal flux of horizontal
in contrast to Eq. (3 .2 ). After adding N such waves in momentum is given by the leading component Sxx of
succession, we see by a similar argument that the ef­ the radiation stress tensor Stj . In deep water
fective gravity g lS) is given by
(5.1)
*(*> = (1 + X)(l + X') - • -(1 + Х(ЛГ-'>)£. (4.3)
If the waves are all self-similar, or even if only where E is the mean energy density of the short waves.
Z X(r>-*> oo, it follows that If the current U is varying in space, work is done by
the current on the short waves, so as to produce a
(AO _► (4.4) change in the short-wave energy. Thus, if U varies only
Hence the particle acceleration in the trough of the in the x direction,
composite wave is in defin itely large. dE dU dU
We remark that the demonstrations given in this - + (C t+ U )- = - S „ - , (5.2 )
and the previous section apply equally well both to
progressive and to standing waves (or even to combi­ where cg is the group velocity of the short waves. For
nations of progressive and standing waves). In each short waves propagating on long waves, we may take
case, however, it is necessary that the phases of the С/to be the orbital motion in the long waves. The term
component waves on different scales be suitably related. {cg + U)dU/dx will then be relatively small and we
In the next section we shall see under what circum­ shall have roughly
stances this is most likely to occur. Here, however, we
may note that in the special situation where random к— (5 .3 )
waves are impinging on a vertical wall (see Fig. 4 ), all dt 2 ax '
the wave components are in effect standing waves with
an antinode in surface elevation at the wall itself. Thus, Thus the parts of the long wave that are most unstable
the phases at the wall are either 0° or 180° and hence to short-wave disturbances are those for which d(J/dx
have a greater probability of being precisely correlated. < 0; that is to say, where the surface is contracting
The above result helps to show that the very high horizontally.
accelerations (20 g and higher) found by Cooker and
Peregrine (1991) in numerical studies of waves im­
pinging on a vertical wall are not in themselves unrea­
sonable.

S. Stability of the free surface

Those phases of a surface wave that are most liable
to instability may be investigated very simply by con­
sidering the radiation stress associated with short waves
riding on longer waves. It was shown by Longuet-Hig­
gins and Stewart (1960) that short waves riding on a RG. 5. Stable and unstable phases in a progressive gravity wave.
 2119

AUGUST 1994 NOTES AND CORRESPONDENCE 1837

-0.08 -0.04 0 0.04 0.08 0.12

(У-П/L
Fig. 6. Relative steepening factor r for short waves riding on a long wave of finite steepness
AK, shown as a function of the surface elevation у in the long wave (from Longuet-Higgins 1987).

In a p rogressive wave (see Fig. 5) the surface is con­ done by radiation stress, and reduction in the effective
tracting on the forw ard face of the wave, and stretching gravity. All these effects will be cumulatively greater
horizontally on the rear face. Thus, the forward face on the forward face of the longest waves.
of the wave is unstable to short-wave disturbances, and
the rear face is stable. This is reflected in the common 6. Surface stability in standing waves
observation that the forward face of a steep wind wave
is generally rough, while the rear face is smooth. Consider on the other hand a standing surface wave
The wavelength of the short waves is also affected (Fig. 7). In such a wave the amplitude of the crests
by the contraction of the free surface, and the amplitude and troughs are either increasing simultaneously as in
of the short waves is further increased by the relatively
smaller value of the apparent gravity g' on the upper
part of the long waves (Longuet-Higgins and Stewart
1960).
The extent of the short-wave steepening on long
waves of finite amplitude is, in fact, greater than would .stable
be expected on a linear theory for the long waves. Exact
calculations (see Fig. 6 ) show that even on a long wave
with steepness parameter AK = 0.3, the short waves
steepen between the long-wave trough and long-wave
crest by a factor as great as 4. When AK = 0.4 the
steepening factor exceeds 10. b
Thus, it is most probable that the steepest short waves
will be found near the crests of the long waves, and unstable unstable
preferably on their forward face. For, if the short waves
break ahead of the long-wave crests, they will not regain
enough energy from the wind to break on the rear face
(cf. Longuet-Higgins 1991). unstable
Still shorter waves riding on the short waves will be Flo. 7. Stable and unstable phases in a standing gravity wave: (a)
subject to the similar processes of steepening by con­ crests and troughs increasing in amplitude; (b) crests and troughs
traction of the free surface due to the short waves, work diminishing.
2120

1838 J O U R N A L OF P H Y S I C A L O C E A N O G R A P H Y V o lu m e 24

Fig. 7a or decreasing simultaneously (Fig. 7b). In the enon. But even there the local contraction of the free
first case the free surface is everywhere being stretched. surface in the unperturbed flow is a significant factor
Hence, it is everywhere stable to short-wavelength per­ in the instability.
turbations. In the second case the free surface is every­ A satisfying fractal model of a breaking sea that fully
where contracting and so it is unstable to short-wave- incorporates the dynamics of a free surface under the
length perturbations. The latter case, when the wave action of gravity still awaits development. The present
trough is rising, corresponds to the situations found by paper draws attention to this challenge and makes a
Cooker and Peregrine (1991) to give rise to very high modest start.
upward accelerations.
REFERENCES
Banner, M. L., and D. H. Peregrine, 1993: Wave breaking in deep
7. Discussion water. Annu. Rev. Fluid Mech.. 25, 313-397.
Cooker, M. J., and D. H. Peregrine, 1991: Violent motion as near-
We have shown how, by incorporating dynamical breaking waves meet a wall. Proc. IUTAM Symp. on Breaking
considerations into a simple fractal model of surface Waves. Sydney, Australia, Springer, 291-297.
waves it is possible to understand some observed fea­ Huang, N. E., S. R. Long, С.-С. Tung. M. A. Donelan, Y. Yuan,
and R. J. Lai, 1992: The local properties of ocean surface waves
tures of the sea surface. The model is highly simplistic by the phase-time method. Geophys. Res. Lett.. 19, 685-688.
and is made possible only by the assumption of a wide Kerman, B. R., 1990: Fractal geometry and sound generation of a
separation of scales between the component waves. wind-swept sea. Proc Conf. on Natural Physical Sources of Un­
This enables us to assume that the short waves are in derwater Sound. Cambridge, England, KJuwer Academic, 467-
a sense controlled by the longer waves on which they 482.
------ , 1993: A multifractal equivalent of the Beaufort scale for sea-
ride and to ignore the reaction of the short waves on state. Geophys. Res. Lett.. 20, 297-300.
the much more energetic long waves. Longuet-Higgins, M. S., 1985: Accelerations in steep gravity waves.
Nevertheless it is worth noting that the short-wave J. Phys. Oceanogr , 15, 1570-1579.
steepening shown in Fig. 6, for example, is nearly in­ ------ , 1987: The propagation of short surface waves on longer gravity
waves. J. Fluid Mech.. 177, 293-306.
dependent of the wavelength of the short waves when ------ , 1988: Mechanisms of wave breaking in deep water. Sea Surface
the ratio of the wavelengths exceeds about 2 ( see Lon­ Sound. B. R. Kerman, Ed., Kluwer Academic, 1-30.
guet-Higgins 1987, Figs. 4 and 5). Thus the aspects of ------ , 1991: A stochastic model of sea surface roughness. II. Proc.
wave steepening discussed here may not be very de­ Roy. Soc. London A, 435, 405-422.
------ , and R. W. Stewart, 1960: Changes in the fonn of short gravity
pendent on the assumption concerning the component waves on long waves and tidal currents. J. Fluid Mech.. 10,
length scales. 529-549.
It should also be noted that the horizontal contrac­ ------ , and M. J. H. Fox, 1977: Theory of the al most-highest wave:
tion, which destabilizes certain portions of the free sur­ The inner solution. J. Fluid Mech.. 80, 721-741.
face, is merely the mechanism that initiates the break­ ------ , and R. P. Cleaver, 1994: Crest instabilities of steep gravity
waves. Part 1. The inner solution. J. Fluid Mech., 258, 115-
ing. Another such mechanism is the instability of the 129.
crest of a steep wave ( see Longuet-Higgins and Geaver Phillips, О. М., 1958: The equilibrium range in the spectrum of wind­
1994). It appears the latter is mainly a local phenom­ generated waves. J. Fluid Mech.. 4, 426-434.
 2121

2458 J O U R N A L OF P H Y S I C A L O C E A N O G R A P H Y Volume 25

Reprinted from J ournal of Physical Oceanography. Vol. 25, No. 10, October 1995
A m m a n MnextJopoJ Society

On the Disintegration of the Jet in a Plunging Breaker

M ich ae l L onguet -H iggins
Institute for Nonlinear Science. University of California, San Diego. La Jolla, California
13 March 1995 and 14 April 1995

ABSTRACT
An inviscid mechanism is proposed foe the breakup of the jet in a plunging surface wave. Streamwise per­
turbations of the original surface are shown to grow rapidly owing to stretching of the thin jet and to drastic
reduction in the normal pressure gradient. This converts transverse gravity waves into almost pure capillary
waves. Conservation of wave action for the perturbations then implies a strong increase in the perturbation
amplitude.

1. Introduction wave as in Fig. 2, with crests aligned in the plane of

The spontaneous growth of longitudinal striations in undisturbed flow. If s is measured along the surface,
the jet of an overturning wave is a well-observed phe­ the normal displacement £ will be given by
nomenon; usually it is an intermediate stage in the С = a (s ) cos(*y - a t), ( 2. 1 )
breakup of the jet into droplets and spray. Some inves­
tigators have sought an explanation in the frictional ef­ where к and a denote the wavenumber and the radian
fect of the air or in the formation of streamwise rolls frequency. These will be related by
in the jet, analogous to Gortler vortices. Here we shall
show on the contrary that the phenomenon can be a 2 = g*k + Tk\ (2.2)
largely accounted for by the behavior of irrotational
waves on a thin sheet of fluid. Two essential ingredients where T denotes surface tension and g * is the effective
are the longitudinal stretching of the jet and the highly value of g for the surface; thus,
reduced pressure gradient normal to its surface.
g* = g cos a + n - Kq3, (2.3)
2. Initial stages where a is the local angle between the surface and the
Figure 1 illustrates the development of the jet in a horizon, n is the local normal acceleration of the un­
typical plunging breaker, calculated numerically with disturbed surface, к is the curvature of the surface, and
the assumption that the flow is irrotational and two- q is the particle speed in a reference frame moving with
dimensional. The pressure at the free surface is con­ the wave. In fact, we have pg* = dp/dn, the normal
stant. Near time t = t {. when the tangent to the surface gradient of the pressure.
near the crest makes a sharp right-angled turn, there is The wave amplitude a (s ) is assumed to vary rela­
a large pressure gradient in the fluid that accelerates the tively slowly compared to the phase (ky - a t), so that
fluid near the crest horizontally and propels a jet for­ we may apply the principle of conservation o f wave
ward from the crest. action where the action density A is defined by
This part of the flow has been modeled analytically
as a rotating Dirichlet hyperbola (see Longuet-Higgins A = l- a 2a/k. (2.4)
1980, 1983). From about t2 < t < h the jet narrows
rapidly, and the pressure gradient within the jet dimin­ Now at time t = t0, when a = o<>, say, g * will be of
ishes drastically; the fluid is then almost in a state of the same order as g . But at time t = t3, say, the pressure
free-fall in a parabolic trajectory. gradient will be small and hence g * will be small com­
Now imagine a perturbation in the form of a short pared to g , and perhaps also compared to Tk2. In that
surface wave of amplitude a propagated across the case we shall have simply
tr2 = 7 1 '; (2.5)
Corresponding author address: Dr. Michael S. Longuet-Higgins.
Institute for Nonlinear Science, University of California, San Diego. that is, the waves are controlled dominantly by capil­
9500 Gilman Drive. La Jolla, СА 92093-0402. larity. Since k remains constant, it follows that

Л I0Q S A m r r i m n M rt^ r tm ln e irn l S n r i r f v

2122

O c to b e r 1995 NOT E S AND C O R R E S P O N D E N C E 2459

Fig. I. Successive profiles of a plunging breaker in water of finite depth, calculated by numerical time stepping.
(After Vinje and Brevig 1981.)

A_ If the initial perturbations are dominantly gravity

( 2.6 ) waves, they will, on becoming converted to capillary
Ao
waves, have their amplitude increased. For example, if
and hence the wavelength of the perturbations lies between 2 and
20 cm, their amplitude will be increased by a factor
a2 / 8 + Tk2\',2 A
(2.7) between 1.23 and 3.40.
al \ Tk2 ) A0 Supposing that the perturbations survive this initial
increase in amplitude without breaking, let us consider
Now we may apply the law of action conservation in
their subsequent history.
the form
A As = const = A 0( A s ) o. (2.8) 3. Perturbations of a thinning jet
where As denotes the distance along the surface be­ It was shown by Taylor in a celebrated paper (1959)
tween two neighboring particles moving with the fluid. that the sinusoidal small perturbations of a thin sheet
At about the time t = t2, As/(As)0, is still of order 1, of fluid of uniform thickness 2h are of two kinds: (a )
so that from (1.7) and (1.8) symmetric and (b ) antisymmetric (see Fig. 3 ). In case
g + Tk2\ (a ) the displacements on the two sides of the plane of
(2.9) symmetry are opposite in sign, the velocity potential ф
al \ Tk2 being given by

(b)

F ig . 2. Sketch o f short transverse waves on a plunging breaker, Fig. 3. Sketch of waves on a thin sheet of fluid: (a) symmetric and
at an early stage o f development. (b) antisymmetric (after Taylor 1959).
 2123

2460 J O U R N A L OF P H Y S I C A L O C E A N O G R A P H Y V o lu m e 25

ф= aa and
cosh kz sin (ky - at) (3.1)
к sin bkh E = 2V = T(ak)2. (3.10)
in our notation, z = 0 being the central plane. The dis­ The action density A is then given by Eta. For the
persion relation is symmetric waves this leads to
Tky tanh kh (3.2) A = (T lh ),/2a 2. (3.11)
as in pure capillary waves on water of finite depth A.
In case (b) the displacements on each side are of the We then have two relations, derived from the conser­
sam e sign, corresponding to a velocity potential vation of mass and of wave action, respectively. From
mass conservation
фш sinh kz sin(ky - at) (3.3) hAs = const, (3.12)
к cosh kh
and from action conservation
and dispersion relation
( a 2/hU2)As = consL (3.13)
a 2 = Tk* cosh kh. (3.4)
On dividing (3.13) by (3.12) we obtain
It is obvious that an initial perturbation confined to
one side of the jet will be resolved into the sum of a a 2/h: const, (3.14)
symmetric and an antisymmetric perturbation, which
then will be propagated independently, each with its
own phase and group velocity. a a /r (3.15)
When the sheet is thin compared to the length of the
perturbation, that is, kh < 1, the above expressions are
greatly simplified. For the symmetric perturbations we
have
aa .
= j^ s m i k y - a t)
(3.5)
= Thk*

so that the particle velocity is entirely tangential and

the phase speed с is given by
с = a Ik = ( T h)U2k. (3.6)
These waves are dispersive.
On the other hand, for the antisymmetric perturba­
tions
ф - aaz sin(Jbr (T/)J
(3.7)
72 = ( T/h)k2

giving
с = (T lh )m . (3.8)
These waves are therefore nondispersive and, in fact, (b)

are closely analogous to the waves on a stretched

membrane of tension T and thickness 2h .
Consider now the behavior of the waves when the
jet (which we assume to be already thin compared to
the wavelength) is stretched in the x direction. For both
types of wave the potential energy V is derived from
the extension of both the upper and lower surfaces
against the surface tension T. Hence we have
Fia. 4. Ultimate mode of disintegration of waves on a thin !
V = ± T (a k )2 ( 3 .9 )
of fluid: (a) symmetric waves and (b) antisymmetric.
2124

O c t o b e r 1995 NOTES AND CORRESPONDENCE 2461

(b)
Fig. 5. Sketch showing the change in orientation o f obliquely propagating waves on a sheet o f water,
when the sheet is stretched in the x direction.

and same way. The nonlinear theory for antisymmetric

waves on thin sheets of fluid (Kinnersley 1976, section
a/h « h' (3.16) 5) indicates that the surface profile is described para­
From (3.15) it follows that stretching of the sheet re­ metrically in terms of the elliptic integrals Е (в,т ) and
duces the amplitude of the symmetric perturbations, but F(0, m) by
not as fast as the thickness 2Л. From (3.16) we see that
relative to the thickness h the amplitude grows, and (2E - F)
though our analysis does not go beyond linear pertur­ * 2т иг (3.21)
bations, it appears that the perturbations will finally z = a cos
pinch off the sheet into cylindrical drops.
From the nonlinear theory of capillary waves on thin where m denotes the modulus. The velocity potential
sheets of water (Kinnersley 1976) we find that in the is proportional to F. In the critical case m = 0.7312
limit the surface profile is one of a family of circular when a/L = 1.330, the surfaces of adjacent waves
arcs with centers on the line of symmetry [ see Kinner­ touch, as shown in Fig. 4b.
sley 1976; Eq. (1 0 )]. This is illustrated in Fig. 4a. The
limiting flow in the “ neck” of the wave is simple and 4. Oblique perturbations
has been discussed by Longuet-Higgins (1988, sec­
tion 4 ). We have so far considered only sinusoidal pertur­
Consider, on the other hand, the antisymmetric bations that are propagated exactly transversely to the
waves. We have for the action density direction of the plunging wave, that is, to the direction
of extension of the fluid sheet. Now consider a pertur­
A = ( Th)U2a2k . (3.17) bation such that the wave crests (lines of constant
Therefore we have (3.12) together with phase) make a general angle /? with the direction of the
breaker, as in Fig. 5a. A consequence of extending the
( а 2Л|/2)Д * = const, (3.18) sheet in the x direction in the ratio с > 1 will be to
leading to distort the wave pattern so that two points P, Q, say,
on the x axis are now separated by a distance P 'Q '
a * h u* (3.19) = ePQ, whereas as two points P, R in the transverse
direction remain at the same separation as before.
and Hence, the wavenumber
a/h а /Г3'4. (3.20) ( kx, ky) = (k sin/0, к cos/?) (4.1)
Hence the amplitude of the disturbance increases with­ becomes
out limit, but because the displacements are of the same
sign they do not pinch off the sheet into strips in the ( * ; , * ; ) = <«-'* sin/?,* cos/?). (4.2)
 2125

2462 J O U R N A L OF P H Y S I C A L O C E A N O G R A P H Y V o lu m e 25

Hence, the vertical distance separating two marked par­

ticles following each other at a time-interval of Д/ will
increase like gt Д/. By continuity we see that again the
thickness 2h of the sheet forming the unperturbed wa­
terfall must decrease like 1/r. In other models of break­
ing waves (Longuet-Higgins 1980, section 10; Lon­
guet-Higgins 1981, Fig. 12), a sharp cusp is attained
in a finite time t.
The most important and drastic stage of the instabil­
ity appears to be the initial transition of the perturbation
from a gravity to a capillary wave. This may be re­
sponsible for the breakup of the jet at an earlier stage
of its development. Once the jet has been broken up
into independent strips in the manner described, the
ensuing breakup of the strips into droplets can be ex­
pected to follow swiftly.
We have ignored the frictional and inertial influence
of the air surrounding the jet, which may be apprecia­
ble. However, we note that if the inviscid mechanism
just described is indeed significant, then an increase in
the amplitude of the initial cross-wave perturbation
Fig. 6. The tip of a plunging breaker, modeled as a rotating would be expected to advance the process of disinte­
Dirichlet hyperbola in free-fall (Longuet-Higgins 1980.) gration. This is a prediction amenable to experimental
verification.
The effect is to turn the direction of the perturbation Acknowledgments. The stimulus for the work came
more nearly transverse to the flow. In the limit, as the from a discussion at the ONR Workshop on Free-Sur-
sheet becomes infinitely stretched, the striations be­ face Turbulence in Pasadena, 28 February 1995. The
come aligned with the flow, as is observed. The dy­ author is supported by ONR Grants N00014-91-J-1582
namics of the perturbation in the general case are sim­ and N00014-94-1-0008.
ilar to those in the case of directly transverse waves,
except that some allowance must be made for propa­ REFERENCES
gation of the wave action, with the group velocity, nor­ Dias. F.. and E. O. Tuck, 1993: A steady breaking wave. Phys. Fluids
mal to the wave crests. A. 5, 271-279.
Jenkins, A. D.. 1994: A stationary potential-flow approximation for
a breaking wave crest. J. Fluid Mech., 280, 335-347.
5. Discussion Kinnersley. W.. 1976: Exact large amplitude capillary waves on
sheets of fluid. J. Fluid Mech.. 77, 229-241.
In theoretical models of plunging breakers, the Longuet-Higgins. M. S.. 1980: On the forming of sharp comers at a
plunging jet has sometimes been treated as a steady free surface. Proc. Ray. Soc. London A. 371,453-478.
------, 1981: On the overturning of gravity waves. Proc. Roy. Soc.
flow (see, for example, Dias and Tuck 1993; Jenkins London A. 376, 377-400.
1994), but in reality the flow is time dependent. A pos­ ------ , 1983: Rotating hyperbolic flow: Panicle trajectories and para­
sible analytic model is the rotating hyperboloid (Lon­ metric representation. Quart. J. Mech. AppL Math. 36,247-270.
guet-Higgins 1980, 1983), in which the thickness 2h ------, 1988: Limiting forms for capillary-gravity waves. J. Fluid
of the sheet (following a fixed particle) diminishes like Mech.. 194, 351-375.
Taylor. G. I.. 1959: The dynamics of thin sheets of fluid II. Waves
1//; see Fig. 6. on fluid sheets. Proc. Roy. Soc. London A. 253, 296-312.
In a steady waterfall, where the particles are almost Vinje, Т.. and P. Brevig, 1981: Breaking waves on finite water
in free-fall, the vertical distance of a marked particle depths: A numerical study. Ship Res. Inst, of Norway, Rep. R-
below a given level will increase ultimately like 1/2gt2. 111.81.
2126

237

TH E C R E S T IN ST A B IL IT Y OF ST E E P G R A V IT Y W A V E S,
O R H O W DO SHO RT W AVES B R E A K ?
M ic h ae l S . L o n g u e t-H ig g in s
I n stitu te f o r N on lin ear S cie n ce , U n iversity o f C aliforn ia, San D iego, La J o lla , C a liforn ia
92093-0402, U.S.A.

A b s t r a c t . T h e local, normal-mode perturbations of the crest of a steep g rav ity wave (an
“alm o st-h igh est” wave) have been calculated . T here is ap paren tly only one un stab le m ode,
which is sta tio n ary and has an exponential grow th-rate 0 = 0.0544, on the local tim e-scale. In
th is in stab ility the fluid is displaced forwards, creatin g a region of sh arp (concave) cu rv atu re,
or “toe” , on the forward face of the wave.
T h e in sta b ility is shown num erically to be an asym ptotic form of the low est un stab le
superh arm on ic in stab ility of a finite-am plitude Stokes wave.
It is suggested th at in short gravity waves (w avelength 5 to 50 cm ) th is in sta b ility is the
source of p arasitic c ap illary waves, which originate a t the “toe” of the in stab ility rath er than at
the wave crest. T he p arasitic capillaries produce strong vorticity, leading to flow sep aratio n and
the generation of a roller beneath the gravity-w ave crest. T he roller then becomes turb ulent
and the crest crum ples. T h is prediction is supported by recent labo rato ry observations d ue to
Professor J.H . Duncan.

1 Introd uction
A steep g ra v ity w ave, w ith wavelength in the range 5.0 cm to 50 cm ap p ro x i­
m ately, is com m only observed to support a train o f “p arasitic” c a p illa ry w aves
(C o x , 19 5 8 ). T h e capillaries not only contribute stron gly to the d am p in g o f
th e g ra v ity w ave (see Longuet-Higgins, 19 6 3 ) b u t in so doing th e y g en erate
enough v o rtic ity to contribute su b stan tially to the vo rtex o r “ro lle r” which is
g en erally found beneath the crest o f such waves; see Ebuchi, K a w am u ra and
T ob a (1 9 8 7 ), Longuet-Higgins (19 9 2 ). In early theories o f p arasitic cap illaries
(Longuet-H iggins, 1963; C rap p er, 1970) it was supposed th a t the source o f the
c ap illa ry waves was the sharp curvature at the crest o f the g ra v ity w ave. B u t
close observation suggests th at the capillaries originate from some p oint on the
forw ard face o f the wave, rath er than at the crest itself. W h y is th is so?
To answ er this question, we have investigated carefully the form o f th e in sta ­
b ility o f th e crest o f a very steep g ravity wave. In a recent pap er by L onguet-
Higgins and C leaver (19 9 3 ), here referred to as LHC, we took as o u r m odel
th e crest o f an “alm ost-highest wave” (see Fig. 1 ), whose asym p to tic form was
first found by Longuet-Higgins and Fox (19 7 7 ), (or LH F 1 ), and we found the
n orm al-m ode pertu rb ation s of this flow, th a t is to say those p ro p o rtio n a l to
etct, where a is the radian frequency and t is the tim e. A m ong the sp ectru m
o f such oscillations we found ju s t one mode w ith an im aginary frequ en cy cry
th a t is w ith a positive exponential growth having /3 = - io > 0. T h e form o f
the in stab ility is shown in Fig. 2, and it will be seen th at it does indeed have
a sh arply concave “toe” on the forw ard face o f the crest, which could becom e
the source o f cap illary waves. (These m ust ride ahead o f the source, since th e ir
group velocity exceeds th eir phase velocity; see Lamb, 19 3 2). An o u tlin e o f the
calculations leading to this result is given in Sections 2 to 4 below.
238 The Crest Instability of Steep Gravity Waves, or How Do Short Waves Break?

Fig. 1 . (from L H F l). Profile of the alm ost-highest wave, showing coordinates and 30°
asym ptotes.

The occurrence of this instability suggests th at short gravity waves su p port­

ing parasitic capillary waves break in a way different from longer gravity waves
or from very short capillary waves. F irst, the crest becomes unstable, leading
to parasitic capillaries originating at the wto e” o f the instability. Then the vo r­
ticity generated by the capillaries causes a shear flow beneath the gravity-w ave
crest, as shown by Longuet-Higgins (199 2 ). Finally the crest becomes turbulent
and the wave collapses, or ‘'crumples”. Some prelim inary experim ents carried
out by Professor Duncan at the University o f M aryland support this conclusion,
and are described in Section 7 below.

2 T he inner solution: basic flow

W e consider a steady, irrotation al wave propagating to the left as in Fig. 1 ,
seen in a reference fram e moving to the left with the phase-speed c. We take
rectangular coordinates (x,y) with the у-axis vertically upwards and the origin
0 vertically above a crest, at such a height th at on the free surface

» +5?2 =o (1 )
precisely. (The acceleration o f gravity g is taken to be u nity). If ф and ф
denote the velocity potential and the stream -function in the steady flow and if
X = ф + хфу z - i(x + iy), then equation ( 1 ) becomes
(z + z-)zx (zxy = l (2 )

to be satisfied on the free surface ф = 0, say. (Here a subscript denotes differ­

entiation and an asterisk denotes the com plex conjugate).
If q0 denotes the p article velocity at the crest o f the wave, where x = 0 and
ф = 0 , then for steep waves the param eter e defined by
<2 = ?3/2c2 (3)
2128

Michael S. Longuet-Higgins 239

becomes small, and as с —►0 the flow near the crest tends asymptotically to a
certain inner flow (see Fig. 1 ) tending to the Stokes corner-flow z = (Зг'х/2)2^3
as |z| or |x| tend to infinity. We choose a length-scale so that the crest is at unit
distance below the origin (which is the point of intersection of the asymptotes).
Thus at the crest we have

у = - 1 , q = q0 = V2 (4)
To calculate the form of the inner flow, LHF1 made the transformation
1 -w 6 - i\ /,ч
= <5—— , и = - (5)
1 +w <5+ *X
where 6 is a suitable constant. This maps the flow onto the interior of the circle
|u;| = 1. Then the solution was expressed as

« = (< + ! # % ) (6)
where
B(u) = Bo + B\ lj + J?2 ^ Hh ■ (7)
The coefficients Bn (which are real, by symmetry) were determined by substi­
tuting ( 6 ) into the boundary condition (2) and equating coefficients ofu;n. This,
together with the necessary condition that

B{ 1 ) = B0 4- Bx + B2 + ... = 6 ' 2/3 (8 )

yields an infinite system of nonlinear equations for the coefficients Bn> which
can be solved numerically by truncation, i.e. setting Bn = 0 when n > N, and
then allowing N to tend to infinity.
The system of equations derived from equation (*2) is in general cubic in the
coefficients Bn. However, it was shown in LHC that this system can be replaced
by an equivalent system of quadratic equations, which renders the calculation
both faster and more accurate, for given N. As a result, the flow in Fig. 1
was found to satisfy the condition of constant pressure at the surface to one
part in 5 x 10s, at least, for all values of \z\ < 68.5 (the second crossing of the
asymptote; see LHFl).

3 N orm al-m ode p ertu rb atio n s

To calculate the normal-mode instabilities we need boundary conditions for the
time-dependent flow at the free surface $ = -F(<M). $аУ- Writing
i = Х{ф,-ф) + ^{ф1ф,г) Л
i (9)
У = У ( ф , ф ) + 7](ф1ф ^ ) J

where Z = i(X + iY) denotes the unperturbed flow found in Section 2 and
С = *(£ + irj) is a small perturbation, LHC show that the kinematical and the
 2129

2 40 The Crest Instability of Steep Gravity Waves, or How Do Short Waves Break?
dynam ical boundary conditions can be combined as the real and im aginary parts
o f the single equation

z-x« , + iZxFt) = - \ z x z-x( С + О - \{Z + z - ) ( z ; c x + ZXQ )

+ \ i [(Z + Z -)(Z X z ; x - z ■ z xx) + ( z ; - z x ) z x z ; ] f - iF* ( 10 )

to be satisfied on Im (x) = 0 .
Now z = Z is alread y expanded in a known power series in u ; see equation ( 6 ).
T his can be expanded in a Fourier series on the unit circle и = e t r , -тг < r < гг.
W e m ay assume sim ilarly the expansions

( = E ~ o(4-. + '« n K
(11)
/ = EST=o (cn cos nT + <*» Sin nr)
w here the coefficients a n , 6n , c„ and dn are functions o f the tim e t only. Then
su b stitu tin g in equation ( 10 ) and equating coefficients o f cos n r and sin n r in
both real and im aginary parts we obtain a system o f equations (with real coef­
ficients) for the vector

x = (а0, « ь — ; &о,Ьь— ; сь с2, . . . ; ) ( 12)

To seek norm al-m ode instabilities we w rite d/dt = (3 and hence obtain a
m atrix equation o f the form
M x = Bx (13 )
w here A and В are infinite m atrices. These can be solved num erically by setting
B n = 0 for n > N , say, and also an = bn = c„ = dn = 0 when n > jV, and
then choosing the lowest (AN + 2 ) equations to determ ine the eigenfrequencies
<jj = i/3j and the corresponding eigenvectors X j .
A n im portant im provem ent in accuracy is attained by noting th at the form
o f A is such th at the vector x can be split into two parts:

u = (ао,вь
(1 4 )
v = ( 6o , 6b .. . 6# ; cb c2l ...c;v)

so th a t equation (1 3 ) can be w ritten

P A U u = A 12v }
(15)
(3A2i v = A 22u J
w here Д ц , etc. are subm atrices of A. Then we have

pu = B jv , B i = А 7/ A j2
(16)
p\ - B 2u, B 2 = A ji1 A 22
Michael S. Longuet-Higgins 241

Fig. 2 . Close-up of the in stab ility a t tim es t = 15,20 and 25.

and hence
^ 2u = B i B 2u (17)
T his reduces th e ord er o f the m atrices by one-half, for a given value o f N , and
leads to a considerable im provem ent in accuracy for a given available m achine­
tim e. F u rth er details o f the calculation are given in LHC.

4 R esu lts of th e calcu latio n

It was found th at by N = 50 all the lower eigenvalues effectively converge to at
least 3 significant figures. Am ong these, there is one eigenvalue /?2 = 0 which
corresponds to a pure horizontal shift o f the wave profile. A ll the rem aining
eigenvalues (32 are negative, corresponding to o scillatory modes, except for one
root (32 = 0 .0 02 9 6, or P = ± 0 .0 5 44 , which corresponds to p ertu rb a tio n s which
grow or decay exponentially in time.
O f most interest is the exponentially growing m ode, (3 = 0.0544. F igure 2
shows the form of the free surface at a succession o f times t. T he in sta b ility
consists of a forward displacem ent of the free surface confined alm ost e n tire ly to
the wave crest. The displacem ent begins ab ru p tly at the “to e” of the in stab ility,
which is located on the forward face of the wave at about a distance 2 .3 units
from the center-line у = 0. The tangent to the unperturbed surface at this p oint
makes an angle o f about 21 ° with the horizontal. Because o f the sharp c u rv atu re
 2131

242 The Crest Instability of Steep Gravity Waves, or How Do Short Waves Break?
the flow becomes highly nonlinear, so th at subsequent profiles in Figure 2 are
at best q u alitative. In fact from previous studies of breaking gravity waves
(Longuet-Higgins and Cokelet, 1978) we expect to see the eruption o f a strong
je t from th e tip of the overturning wave - unless the scale is so sm all th at the je t
is contained by surface tension. It is clear th a t the sharp concave curvature at
the toe o f the instability, which precedes the overhanging je t, is a likely source
for the generation o f cap illary waves when surface tension is taken into account.
T his is m ore in accord w ith observation (see below) than previous theories in
which it was assumed th a t the source o f parasitic capillary waves was the sharp
cu rvatu re a t the wave crest itself.
W e note th at by reversing the sign o f each component in the eigenvector x
in equation ( 12 ), and w ith the same value o f /?, we obtain a pertu rb ation which
p rojects m ainly from the rear face of the wave. This is less frequently observed.
A possible explanation is th at the original stim ulus for the p ertu rb ation is likely
to come from shorter waves which are swept backwards up the forw ard face
o f the advancing wave, and are increased in steepness by the o rb ital straining
combined with the radiation stress. These disturbances will excite the instability
on the forw ard face. By contrast, on the rear face straining tends to reduce the
steepness o f any shorter, superposed waves. Thus the sign o f the eigenvector x
is m ore likely to be determ ined by disturbances on the forward face than on the
rear face. Once the sign o f x is determ ined, the instability will grow accordingly.
On the present scaling, the ratio of the length-scale to th at of a gravity wave
o f wavenum ber к = 1 and wavelength 2тг is equal to ( 2, w here e is the param eter
defined by equation (3). The tim e-scales are in the ratio c. Thus the rate of
grow th /3 to the radian frequency a of a sm all-am plitude wave in deep w ater
(<72 = gk) is given by
P!<j = 0.0544 c"1 (IS)

T he p aram eter € is related to the steepness ak o f a gravity wave by equation

(5.4) o f LH F 1 , which can be w ritten

ak = 0.4432 - 0.50 (2 + 0(<3) (19)

This shows th at in very steep waves (if th ey are attain ab le) the grow th-rate
o f the instab ility becomes infinite like (0.4432 - ak)~1/2. Thus it appears most
unlikely th a t a steep gravity wave will ever attain the lim iting Stokes form w ith
a sharp 120° corner. In practice it must topple over, probably forw ards, before
the lim iting form is attain ed .

5 R elatio n to superharm onic in stab ilities

The above result is true for waves o f alm ost lim iting steepness. Is it also true
for waves o f som ewhat lower am plitude? In a second paper (Longuet-H iggins,
C leaver and Fox, 1993, o r LHCF) we have investigated this question by consid­
ering the flow in the wave crest when it is m atched asym p to tically to the flow in
the rest o f a periodic wave in deep w ater, as was first done by Longuet-Higgins
and Fox (197 8 , or LHF2). This introduces a correction to the inner flow at the
Michael S. Longuet-Higgins 243

ak

Fig. 3. T he dim ensionless grow th-rate 0 ' = 0/c<r of the unstable mode on the outer scale. 0 *
is plotted as a function of the wave steepness ak. Also shown are the values of w here
<72 is the radian frequency of the lowest normal-mode perturbation of a deep-w ater Stokes wave
of finite am p litude: ( • ) from Longuet-H iggins (1978, 1986), (+ ) from Tanaka (1983).

crest. In LH CF we calculated the norm al-m ode p e rtu rb ation s o f this corrected
flow. It turns out th at the slightly less steep waves are still unstable, b u t w ith a
reduced rate o f grow th /3. M oreover the rate o f grow th can be com pared to the
rate o f grow th o f the lowest unstable (superharm onic) mode o f a fin ite-am p litu d e
Stokes wave, which was calculated by Longuet-Higgins (1 9 7 8 , 1986) and T anaka
(19 8 3 ). This shows th a t the crest in stab ility is sim ply an asym p totic form o f
the la tte r in stab ility; see Figure 3. T he norm al m ode retu rns to s ta b ility when
the wave steepness ak is less than 0.4292.

6 D iscussion
W e have shown th a t for very steep Stokes waves, w ith ak > 0.40, th ere is ju s t
one local in stab ility o f the wave crest, and th at it is a lim iting form o f the
superharm onic in stab ility discovered earlier by Tanaka (198 3 ). In its lim itin g
form , the in stab ility produces a ra th e r sharp concavity in the free surface on the
forw ard face o f the wave. In large-scale gravity waves the in stab ility p ro b ab ly
develops a forwards je t, which falls forw ard and traps air, as in the in itia l stage
of a spilling or plunging breaker.
 2133

244 The Crest Instability of Steep Gravity Waves, or How Do Short Waves Break?
W e have of course assumed ra th er special initial conditions, nam ely the weak
p ertu rb ation o f a perfectly periodic train of progressive waves. We may expect
a sim ilar crest in stab ility to be found in a wider class o f non-periodic waves.
W e emphasize, however, th at this is alm ost certainly not the only type o f in sta­
bility which will lead to wave breaking, since in random seas gravity waves are
com m only observed to break at much lower wave steepnesses; see for exam ple
Holthuijsen and Herbers (1986).
From an oceanographic point o f view the most interesting consequences o f
th e crest in stab ility axe likely to be for very short g ravity waves, with wave-
lengths between 5 and 50 cm, say. In these, the instability may be prevented
by surface tension from developing a sharp-pointed je t. Instead, the sharply
concave curvature at the toe o f the instab ility will become a point of origin
for capillary waves propagating down the forward face o f the wave: the well-
known “parasitic capillaries 4 (Cox, 1958). Such capillaries are known to damp
th e gravity wave rapidly, through th eir increased dissipation of energy due to
th eir short wavelength (Longuet-Higgins, 1963, Longuet-Higgins and S tew art,
19 6 4). A more interesting consequence has been dem onstrated by Longuet-
Higgins (199 2 ), nam ely th at the parasitic capillaries will shed enough vorticity
(associated with the energy dissipation) to produce a vo rtex, or “capillary ro ller”
in the crest of the gravity wave. This roller not only helps to sustain the capillary
waves, but it gives rise to turbulent motion in the wave crest, with consequent
increase in the tran sfer o f heat and gases between ocean and atm osphere. It
m ay also happen th at the roller is so strong as to advance upon the capillaries
and overtake them , producing a quasi-steady flow as at the toe o f a spilling
breaker, where air is entrained in term itten tly in the form of bubbles (see Koga,
1982; Longuet-Higgins, 1990).
The sequence o f events — from the developm ent o f the instability, to the
form ation of the parasitic capillaries, to the shedding o f vo rticity and form ation
o f the roller, to the generat ion o f turbulence in the wave crest — m ust take place
very rapidly, and be fully visible only with the aid o f high-speed photography.

7 Com parison w ith observation

P relim inary confirm ation is provided by some recent observations made by P ro ­
fessor Jam es Duncan at the U niversity o f M aryland, and reproduced here by
permission. A train o f waves with period 1.0 to 1.4 Hz was produced by a
plunger at one end of a long (40 m ) wave tank. The free surface was illum i­
nated in a vertical plane by a parallel beam o f light incident from above, and
viewed from the side by a cam era m ounted on a carriage moving horizontally
w ith approxim ately the phase speed o f the waves; see Figure 4. The light rays
being refracted a t the free surface, ray caustics appeared w herever there was
focusing o f the beam due to the cu rvature o f the w ater surface. In this way,
very small p erturbations could be d etected, especially when the wavelength o f
the perturbation was short.
Figure 4 , with waves o f frequency 1.4 Hz, or wavelength 1.0 m, shows the
developm ent o f an in stab ility very sim ilar to th at described in Figure 2. It is
2134

Michael S. Longuet-Higgins 245

5 cm 5 cm

5 cm

Fig. 4. Successive profiles of a m echanically-generated wave of frequency 1.4 Hz (by courtesy

of Professor J . H. D uncan).

accom panied by the appearance of a parasitic capillary wave. T he in tern al flow

is not visible, b u t according to the argum ents given in Longuet-H iggins (1 9 9 2 )
we expect vo rticity to be shed from the parasitic capillaries. T he longer ripples
seen above the toe o f the in stab ility are probably in stabilities o f the resu ltin g
shear flow. S h o rtly afte r the last fram e shown, the surface became tu rb u le n t.
It should be noted th a t in these experim ents there was no im posed w ind field.

8 Conclusion
W e have described a mechanism, inferred from both an alysis and la b o ra to ry
observations, whereby short g ravity waves can essentially d isintegrate. T h is
form o f breaking is to be distinguished both from the spilling or plunging o f
large-scale gravity waves and from the pinching o f air bubbles by the trou g h s
of very short capillary waves. A descriptive name for this in term ed iate form o f
breaking would be “crum pling” . B y generating turbulence in the crests o f sh o rt
gravity waves it may contribute significantly to the transfers o f heat and gases
 2135

246 The Crest Instability of Steep Gravity Waves, or How Do Short Waves Break?
between ocean and atmosphere at low and moderate wind-speeds.
This work has been supported by the Office of Naval Research under Contract
N0014-91-J 1582.

References
C rapper, G. D.: 1970, ‘Nonlinear capillary waves generated by steep gravity waves’, J. Fluid
M ech. 40, 149-159.
Ebuchi, N., Kawamura, H. and Toba, Y.: 1987, ‘Fine structure of laboratory wind-wave
surfaces studied using an optical method’, B oun dary-L ayer M et. 39, 133-151.
Holthuisen, L. H. and Herbers, Т. H. C.: 1986, ‘S tatistics of breaking waves observed as
w hite-caps in the open sea’, J. P hys. O ceanogr. 16, 290-297.
Koga, М.: 1982, ‘Bubble entrainm ent in breaking wind waves’, Tellus 34, 481-489.
Lamb, H.: 1932, H ydrodynam ics, 6th ed., Cambridge Univ. Press, 738 pp.
Longuet-Higgins, M. S.: 1963, ‘The generation of capillary waves by steep gravity waves’, J.
F luid M ech. 16, 138-159.
Longuet-Higgins, M. S.: 1978, ‘The instabilities of gravity waves of finite am plitude in deep
w ater, I. Superharm onics’, P roc. R. Soc. Lond. A 36 0 , 489-505.
Longuet-Higgins, M. S.: 1986, ‘Bifurcation and instab ility in gravity waves’, P roc. R. S oc.
Lond. A. 403, 167-187.
Longuet-Higgins, M. S.: 1990, ‘ Flow separation near the crests of short gravity waves’, J. P hys.
O ceanogr. 20 , 595-599.
Longuet-Higgins, M. S.:1992, ‘ C apillary rollers and bores’, J. Fluid M ech. 2 4 0 , 659-679.
Longuet-Higgins, M. S. and Cleaver, R.P.: 1994, ‘Crest instabilities of gravity waves, I. The
inner solution’, J. Fluid M ech. 258, 115-129.
Longuet-Higgins, M. S., Cleaver, R. P. and Fox, M. J. H.: 1994, ‘Crest in stab ilities of gravity
waves. II. M atching and asym ptotic expansion’, J. Fluid M ech. 2 5 9 , 333-344.
Longuet-Higgins, M. S. and Cokelet, E. D.: 1978, ‘The deformation of steep surface waves on
w ater. II. Growth of normal-mode in stab ilities’, P roc. R. S oc. Lond. A 3 64 , 1-38.
Longuet-Higgins, M. S. and Fox, M. J. H.: 1977, ‘Theory of the alm ost-highest wave: the inner
solution’, J. Fluid M ech. 80, 721-741.
Longuet-Higgins, M. S. and Fox, M. J. H.: 1978, ‘Theory of the alm ost-highest wave. II.
M atching and an alytic extension’, J. Fluid M ech. 85, 769-786.
Longuet-Higgins, M. S. and Stew art, R. W .: 1964, ‘Radiation stresses in w ater waves. A
physical discussion with applications’, D eep-Sea Res. 1 1 , 529-562.
Tanaka, M. 1983: ‘The stab ility of steep gravity waves’, J. P hys. Soc. Japan 5 2 , 3047-3055.
2136

J. Fluid M ech . (1997), vol. 336, pp. 33-50 33

Copyright © 1997 Cambridge University Press

Crest instabilities of gravity waves. Part 3.

Nonlinear development and breaking
By M IC H A E L S. L O N G U E T -H IG G IN S1
a n d D O U G L A S G. D O M M E R M U T H 1
1 Institute for Nonlinear Science, University of California, San Diego, La Jolla,
CA 92093-0402, USA
2Naval Hydrodynamics Division, Science Applications International Corporation,
10260 Campus Point Drive, San Diego, CA 92121, USA

(Received 15 December 1995 and in revised form 14 August 1996)

The ‘ almost-highest wave’, which is the asymptotic form of the flow in a steep
irrotational water wave of less than the limiting height, was recently shown to be
unstable to infinitesimal disturbances (see Longuet-Higgins & Cleaver 1994). It was
also shown numerically that the lowest eigenfrequency is asymptotic to that of the
lower superharmonic instability of a progressive wave in deep water (Longuet-Higgins,
Cleaver & Fox 1994). In the present paper these calculations are revised, indicating the
presence of more than one such instability, in -agreement with recent calculations on
steep periodic and solitary waves (Longuet-Higgins & Tanaka 1997).
The nonlinear development of the fastest-growing instability is also traced by a
boundary-integral time-stepping method and the initial, linear growth rate is
confirmed. The subsequent, nonlinear stages of growth depend as expected on the sign
of the initial perturbation. Perturbations of one sign lead to the familiar overturning
of the wave crest. Perturbations of the opposite sign lead to a smooth transition of the
wave to a lower progressive wave having nearly the same total energy, followed by a
return to a wave of almost the initial wave height. This appears to be the beginning of
a nonlinear recurrence phenomenon.

1. Introduction
In a recent paper (Longuet-Higgins & Cleaver 1994) it was shown that the local flow
near the crest of a steep, progressive gravity wave on water is itself subject to an
exponentially growing instability. To distinguish it from instabilities involving the
whole wave, this was called a ‘ crest instability’. In a second paper (Longuet-Higgins,
Cleaver & Fox 1994) it was further shown numerically that the crest instability is the
asymptotic form of the lowest superharmonic instability of a complete, periodic gravity
wave in deep water when the wave steepness ak tends to its limiting value
Сak)maz - 0.4423. As shown by Tanaka (1983) this instability first occurs when
ak = 0.4292, the value at which the total energy E attains its first maximum (see also
Saffman 1985). For solitary waves, a parallel result was demonstrated numerically by
Tanaka (1986).
In the above work all of the calculations, whether analytic or purely numerical, were
carried out on the basis of linear perturbation theory, and the departures from a
steady, nonlinear wave were assumed infinitesimally small. An interesting paper by
Tanaka et al. (1987) traced the subsequent nonlinear development of the fastest-
growing instability for the solitary wave by numerical time stepping. Depending on the
 2137

34 M. S. Longuet-Higgins and D. G. Dommermuth

sign o f the initial perturbation, it either developed into a plunging breaker or else
underwent a smooth transition to a solitary wave o f lower amplitude and nearly the
same total energy. A similar investigation for periodic waves in deep water was
attempted by Jillians (1988) with partial success (see below). Meanwhile the general
interest in the subject has intensified owing to its relevance to wave breaking and
turbulent mixing in the uppermost layer o f the ocean, and to the microstructure o f the
sea surface.
The purpose o f the present paper is twofold. First we revise and extend the linear
stability calculations on deep-water waves reported in Longuet-Higgins & Cleaver
(1994) and Longuet-Higgins et al. (1994). The need for such revision was prompted by
a recent independent investigation by Tanaka (1995), undertaken at the request o f one
o f us (M .S .L .-H .).f Our results are described in §3 below.
Our second and main purpose in this paper is to follow by numerical time stepping
the nonlinear development o f the fastest-growing crest instability o f a periodic wave.
The approach adopted is slightly different from that o f Tanaka et al. (1987) or Jillians
(1988). Those authors used as initial values for the time stepping a numerically
determined normal-mode instability for the complete wave. Here we use instead the
lowest crest instability, as calculated in the present paper.
The nonlinear calculation is described in general terms in §4, and the results are
given in §5. As expected, the nonlinear development o f the instability depends crucially
on the sign o f the initial (linear) perturbation. A perturbation o f positive sign leads on
to overturning o f the wave crest in the familiar manner (see figure 9). One might have
expected a perturbation o f negative sign to lead initially to a wave crest with a slight
concavity on the forward face but this is not the case. Instead the wave undergoes a
smooth transition to a periodic wave o f lower amplitude and then returns to a wave
o f almost the same amplitude, in a kind o f recurrence phenomenon. Thus the
behaviour is similar to that o f the crest instability o f a solitary wave, except that for
a solitary wave the period o f recurrence is infinite.

2. Initial conditions: linear perturbation theory

In the theory o f the almost-highest wave (Longuet-Higgins & Fox 1977, 1978) the
flow near the crest o f a steep, irrotational gravity wave is characterized by the
horizontal particle velocity q at the crest o f the wave, as seen in a reference frame
moving horizontally with the phase speed. Together with gravity g this defines a
characteristic length scale
l = q 2/2g (2 . 1)
and a small parameter
e2 = q*/2cl ( 2 . 2)
where c0 is the linear phase speed given by
cl = g/k = gL/2n , (2.3)
к and L being the wavenumber and wavelength.
The limiting form o f the wave crest as e-> 0 was determined by Longuet-Higgins &
Fox (1977) and is shown in figure 7 o f that paper. If axes o f x and у are chosen as
horizontal (opposite to the direction o f propagation) and vertically upwards, then we
can define
z = y —ix, х = Ф+ i^> (2-4)
t The details of T anaka’s calculations and an analysis of the results are given in the paper by
Longuet-Higgins & Tanaka (1997).
2138

Crest instabilities of gravity waves. Part 3 35

where ф and ijf are the velocity potential and streamfunction. For large values of z we
have
as Ixl *< 0, (2.5)
corresponding to Stokes’s 120° corner flow. Generally, the unperturbed flow is
expressed as a series
z = (8+iX)m Щш)9 B = ^ B n*Л (2 -6 )
n -0

where w= ^ (2. 7)
8 + ix
and 8 is a constant at our disposal (<$was taken to be 103/2). The series (2.6) is assumed
to converge inside and on the circle M = 1 which corresponds to the unperturbed free
surface. By applying suitable conditions on the boundary, the coefficients B0, B19..., B90
were determined to a high degree of accuracy.
The normal-mode perturbations of this flow were calculated in Longuet-Higgins &
Cleaver (1994). We write
z = Z + £, (2.8)
where Z is the previously determined flow, given by equation (2.6) and £ is a small
perturbation. The free surface is defined by \fr = F(0, t) and only terms linear in £ and
Fare retained. The kinematic and dynamical boundary conditions combined then lead
to a single (complex) condition to be satisfied on \jr = 0. The perturbations £ and Fare
expanded in series:

n -0
(2.9)
F — 2 (cncosm- + rfnsinrtr),
n -l

where e‘T= w and the coefficients an,bnicn,dn depend on the time t only. Assuming a
time dependence e* and equating coefficients of cosm- and sin/rr in the boundary
conditions as far as n = TV, say, one is led to a set of (4N4-2) simultaneous linear
equations for the vector
л: = (aQ,ai i ...iaN't b0,bl9... ybN' c19...,c N; dx, . . . 9dN) (2 .Ю)
in the form
(ЗАх = Bx (2.П)
where A and В are matrices of order (4ЛГ+2). The numerical solution of these
equations, for a sufficiently large value of the truncation parameter N, then yields the
required set of eigenvectors (3 and the corresponding eigenvector x , which in turn yields
a normal-mode perturbation. There is always one mode in which /? = 0, a0 Ф 0 and all
the other coefficients an vanish. This solution corresponds to a perturbation consisting
of a pure phase shift. In Longuet-Higgins et al. (1994), the results of Longuet-Higgins
& Cleaver (1994) for vanishingly small e were extended to positive values of e by
matching the ‘ inner’ solution for the flow near the crest (that is to say when z = 0(e2))y
to an ‘ outer’ flow representing the rest of the wave. The matching process requires a
modification to the inner flow by an amount of order e3(A-1), where Л is the lowest
positive root of
— tan— ____ (2.12)
2 2 2V 3
 2139

36 M. S. Longuet-Higgins and D. G. Dommermuth

hence Л = 1.8027. Once the modification is made, the calculation of the normal-mode
perturbations to this flow can be carried out as before.

3. Results of the linear perturbation analysis

In Longuet-Higgins & Cleaver (1994), equations (2.5) were solved numerically for
N ^ 50, using a standard F o r t r a n subroutine s g e e v which unfortunately developed
significant errors, when N exceeded about 20. These were only later detected. We have
now used a more robust routine which simply determines the sign of the determinant
D(P)=\\/3A-B\\. (3.1)
Hence the zeros of D(fi) could be found by a straightforward bracketing technique.
The largest positive eigenvalue A determined by this method is shown in table 1 for
N = 60(2) 80 in the column under e = 0. It appears to be converging to a value near
A = 0.0434. (3.2)
This differs significantly from the value 0.0544 in table 1 of Longuet-Higgins &
Cleaver, but is close to the value (0.0435) found by Cleaver in his PhD dissertation
(Cleaver 1981), obtained by the use of a Shanks transform.
The remaining columns of table 1 show the corresponding values of A when
e = 0.02(0.02)0.14. When e > 0.15 convergence is difficult to obtain. All the values of
Д at iV= 80 are plotted against e in figure 1. It will be seen that A decreases
monotonically with e.
From equation (2.9) we should expect small departures of A from its value at e = 0
to vary as e3(A“1}. In figure 1 we show the curve
A = 0.0434-Cie**-», (3.3)
where С = 0.046 is an empirically determined constant.
The calculations also suggested the presence of a second positive zero of /)(/?) at
around 0.01; see table 2. This implies a second exponentially growing mode, as was
found also by Tanaka (1995).
For periodic waves on deep water, the parameter e is related to the wave steepness
ak by the formula
ak = 0.4432 —O.Se2—0.160e3cos (2.143 In e—1.54) (3.4)
(see Longuet-Higgins & Fox 1978, equation (5.4)), so that we may plot the scaled
growth rate (/?„/e) against ak. Thus figure 2 shows f i je plotted against ak and
compared with the growth rates An as calculated by Tanaka (1995) for steep periodic
waves of less than the limiting steepness. Although Tanaka’s calculations were not
carried beyond about ak —0.4424 it is clear that the values of A or A derived from the
theory of the almost-highest wave do indeed form a satisfactory asymptote to Tanaka’s
values. Figures 1 and 2 should be regarded as superseding figures 4 and 5 of Longuet-
Higgins et al. (1994).
A more stringent comparison is made in figure 3, which shows the squared
eigenvalues (Pn compared to Tanaka’s scaled eigenvalues, as a function of e = q/y/2c0.
In particular the limiting value = 0.00188 is a good match to Tanaka’s calculations.
After determining the eigenfrequencies A»» the calculation of the eigenfunctions was
carried out by setting a0 or axequal to 1 and then solving all but one of equations (2.9)
2140

Crest instabilities of gravity waves. Part 3 37

F igure 1. The positive growth rate /?, of the lowest crest instability, plotted against e. The broken
curve represents the asymptotic formula (3 .3).

N\e 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14

60 0.04325 0.04320 0.04299 0.04251 0.04167 0.04020 0.03749 0.03224
62 0.04327 0.04323 0.04301 0.04254 0.04168 0.04016 0.03737 0.03199
64 0.04330 0.04325 0.04304 0.04256 0.04169 0.04012 0.03725 0.03175
66 0.04332 0.04328 0.04306 0.04258 0.04169 0.04008 0.037 14 0.03153
68 0.04334 0.04330 0.04308 0.04260 0.04169 0.04004 0.03702 0.03134
70 0.04336 0.04331 0.043 10 0.04261 0.04169 0.04000 0.03691 0.03116
72 0.04337 0.04333 0.04311 0.04262 0.04168 0.03996 0.03681 0.03101
74 0.04339 0.04334 0.04313 0.04263 0.04168 0.03991 0.03671 0.03089
76 0.04340 0.04335 0.043 14 0.04264 0.04167 0.03987 0.03661 0.03081
78 0.04340 0.04336 0.04314 0.04264 0.04165 0.03982 0.03652 0.03078
80 0.04340 0.04336 0.04315 0.04264 0.04164 0.03978 0.03645 0.03079
Table 1. Calculated values of the growth rate /?t

for the remaining coefficients. This yields xfr) and 1\ф) by equations (2.7). However,
since the free surface is given by \Jr = Ft not \jr = 0, we have for the surface profile to
calculate
у=У(ф,0) + [г{ф)^(ф,0) + ч(ф, 0 ) p ‘. (3.5)
We write this as
y —Y+aAyc^\ (3-6)
 2141

38 M. S. Longuet-Higgins and D. G. Dommermuth

ak
Figure 2. Growth rates Ля of the two lowest superharmonic instabilities of steep periodic waves, as
calculated by Tanaka (1995) (crosses). The circles are values of (ije from figure 1 of this paper.

N\e 0.00 0.02 0.04 0.06

60 0.00731 0.00758 0.00838 0.00873
62 0.00763 0.00787 0.00858 0.00878
64 0.00790 0.00812 0.00875 0.00881
66 0.008 14 0.00834 0.00889 0.00883
68 0.00835 0.00853 0.00902 0.00884
70 0.00853 0.00869 0.00912 0.00883
72 0.00869 0.00884 0.00921 0.00882
74 0.00883 0.00897 0.00929 0.00880
76 0.00895 0.00908 0.00935 0.00878
78 0.00906 0.00917 0.00941 0.00874
80 0.00915 0.00926 0.00945 0.00870
Table 2. Calculated values of /?4

where Ay denotes the expression in square brackets on the right o f (3.5), normalized
so that
max|A>»|=l. (3.7)

Essentially Ay gives the form o f the perturbation. In figure 4, Ay is plotted against the
unperturbed coordinate x/l with ф as parameter. The curves are shown for € = 0 and
0.08. Each curve has a relatively sharp maximum when x < 0 (on the forward face o f
2142

Crest instabilities o f gravity waves. Part 3 39

0.003 —[— 1 1 i 1 1 1 1 ■1 1
-
- -
- -
- -
0.002
о Ъ X xo xn=\
- -
xox
*°x
- X -
Cx
“ X “
(fin/e)2 0.001 — <? —
- X -
- X -
- -
- X -
0 X
X
<N
I X
С

X
- -
X

-
-0.001 1 __ 1__ —1------ 1------ 1____I. 1 .1 1 1
0.1 0.2

F igure 3. Graph of f t (circles, this paper) and (еЛ„)2 (crosses, T anaka 1995) against e.

the wave) and a less pronounced minimum on the rear face. There is little difference
between the curves, however.
Figure 5 shows a comparison of the eigenfunction corresponding to e = 0.08 with
T an aka’s calculated eigenfunction when o j = 0.92 (e = <?/\/2c0 = 0.07533). The curves
are similar but by no means identical. It should be remembered, however, that the
eigenfunction corresponding to e = 0.08 is that of the inner solution, which is valid
over a distance x = 0(e2) only. For any non-zero value of e there has to be a matching
zone of width O(e) where the solution differs from the solution in the inner zone by a
proportional difference which is 0(1). In the region where x /e 2 is about 10, say, x /e is
about 0.8 so that some significant differences are to be expected.

4. Nonlinear stage of growth; method of calculation

To investigate the nonlinear growth of the instability we use a boundary-integral
technique for numerical time stepping, essentially similar to that pioneered by
Longuet-Higgins & Cokelet (1976) but in the modified form due to Vinje & Brevig
(1981). The problem is formulated as finding the irrotational flow of a homogeneous,
incompressible and inviscid fluid with a free surface and in infinite depth. If ф (х,у,0
and ifr{x,y, t) denote the velocity potential and streamfunction, and t the time, then ф
and \{/ are both solutions of Laplace’s equation and we may write

X(z,t) = ф + ii/r, z = x +i y (4.1)

 21 4 3

40 M. S. Longuet-Higgins and D. G. Dommermuth

x/l
F igure 4. Vertical displacement Ay of the free surface plotted against the scaled horizontal
coordinate x in the cases e = 0 and 0.08. The wave travels to the left.

so that x is an analytic function of z. At the free surface then the kinematic condition
can be written
D z _ fd * y
(4.2)
Dt \dz) ’

where D/D/ denotes (0/6/ + V0-V) and a star denotes the complex conjugate. The
condition of constant pressure can be written

Dф _ 1
(4.3)
Df gy 2

where g denotes gravity. We shall take g —1 and also in this application the
wavelength 1 = 1 . The mapping w = exp(—2niz) maps one wavelength onto the
interior of a closed contour C(t) in the w-plane, and we may apply Cauchy’s theorem

(4.4)

to any point w inside C. On letting ш approach С we obtain an integral equation

for x •
2144

Crest instabilities of gravity waves. Part 3 41

&У

F igure 5. Comparison o f the lowest normal-mode instability o f the almost-highest wave when
e = 0.08 (solid curve) with the lowest instability o f a periodic wave when e = 0.0753 (dashed curve).

At any instant /, the boundary conditions (4.2) and (4.3) provide new values of the
coordinates z and velocity potential фon С at the time (/ + dr). Then a discretized form
of (4.4) is used to find the unknown streamfunction xfr on C. On C, piecewise-linear
functions are used to represent фand \{r; the resulting influence coefficients are as given
in Vinje & Brevig (1981). To advance in time we use a fourth-order Runge-Kutta
scheme. In order to prevent the growth of saw-tooth instabilities, five-point smoothing
and re-gridding was applied after every ten time steps.
As a check on the accuracy of the method we first applied it to a steep Stokes wave
with steepness parameter
ak —0.43982. (4.5)
Here 2a denotes the crest-to-trough wave height and к = 2n/L is the wavenumber.
(This corresponds to H/L = 0.14.) The wave profile was found by expanding the
surface elevation and velocity potential in Fourier series with 4000 terms, and adjusting
the coefficients so as to satisfy boundary conditions (for steady flow) at 4000 equally
spaced grid points on the surface. To proceed with the time stepping, we used 2000
unequally spaced points; near the wave crest the spacing was reduced to provide better
resolution. The smallest spacing Ax was 0.61 x 10"4 at the crest and 0 .4 3 x 10"3 in the
trough.
A measure of the numerical error in the Bernoulli equation is provided by the
expression
\В+ г1+ №фУ + сфх\ (46)
\В\+ Н+\ФФУ + с\фх\
which was found to be less than 3 x 10-6. (Here В is the Bernoulli constant, the mean
level being taken as zero.)
 2145

42 M. S. Longuet-Higgins and D. G. Dommermuth

A further check was on the phase speed. According to the asymptotic theory of the
almost-highest wave (Longuet-Higgins & Fox 1978), the speed с and steepness ak are
related by equation (3.4) and
сг/с\= 1.1931 —1.18 e3 cos (2.143 In e+2.22), (4.7)
where e is a small parameter defined by equation (2 .2 ), q being the particle speed at the
wave crest and c0 the speed of linear gravity waves (cj = 2ng/L). From the initial
profile we determine that q/c0 = 0.12002 and so e = 0.084866. Equations (3.4) and
(4.7) then yield ak —0.43966 and с = 0.43589 respectively, in good agreement with our
numerically determined values; the numerically calculated speed was 0.435 89c0.

5. Growth of normal instabilities: results

As initial condition for the normal instability at time t = 0 we add to the
unperturbed wave profile a perturbation aAy where a is a small parameter and
Ay = e\V+ F Y ^ 0, (5.1)
see (3.6) and (3.7). The velocity potential <f>in §4 is related to the potential ф' in §§2 and
3 by ф = ф' —сх. Hence the perturbation in ф is given by
Аф = - e 3c£. (5.2)
The two perturbations Ay and Афcan be given as functions of the unsealed coordinate
x = e2X (5.3)
in the linearized approximation.
In addition, since the wavelength in §4 is taken as unity (not 2л), Ay, Афand x must
be rescaled by multiplying by factors (2 л)-1, (2 л)3/2 and (2 л)"1 respectively.
Figures 6(a) and 6{b) show the graphs of the initial perturbations aAy and осАф to
be added to the profile and velocity potential of the steady wave, when a = 1. These
correspond to e = 0.08194, almost the same as in §4. The results, as we shall see,
depend crucially on the sign of a. Figure 7 shows the initial development of the
perturbation when a = 0.2 (a ‘positive’ perturbation). Each curve represents the
difference between the perturbed and the unperturbed surface elevation, seen in a
stationary reference frame. After an initial adjustment in which the amplitude
diminishes, the perturbation settles down to a steady rate of growth. This is illustrated
also in figure 8 , where the total range of the perturbation (i.e. the difference between
the maximum and minimum elevation) is plotted versus the time t (crosses) up to
t = 3.5. The dashed line shows the theoretical linear growth rate. It appears that
nonlinearity affects the growth rate significantly after about / = 1.5. The initial period
of adjustment may be due to the fact that the calculated eigenfunctions refer to the
‘ inner’ solution of the almost-highest wave, which is valid only within a distance 0(e2)
from the crest, and hence do not perfectly match the eigenfunctions for the whole wave.
(The non-matching part of the initial perturbation is presumably resolved into
eigenfunctions which either decay or grow less rapidly.)
The development of the complete surface profile is shown in figure 9, in the travelling
reference frame. The wave clearly curls over and breaks in a characteristic way, similar
to that in a solitary wave; see Tanaka et al. (1987, figure 5). Breaking may be said to
occur when the surface profile develops a sharp comer at around t = 2 .0 .
2146

Crest instabilities of gravity waves. Part 3 43

X
F igure 6 . Form o f the initial perturbations when a = 1, as functions o f the unsealed horizontal
coordinate x: ( a) Ay, (b ) Дф.
 2147

44 M. S. Longuet-Higgins and D. G. Dommermuth

x/L
F igure 7. Initial development of the positive perturbation, when a = 0.2, in a moving
reference frame.

The development o f the negative perturbation corresponding to a = —0.2 is more

interesting and is shown first in figure 10. A fter a similar initial period o f adjustment
the perturbation grows exponentially, as expected; see the plotted points (squares) in
figure 8. The departures from the linear growth rate are o f similar magnitude, but
opposite sign, to the positive perturbation.
W ith the negative perturbation, however, the surface profile remains convex and
rounded, as seen in figure 11. By analogy with the solitary wave (Tanaka et al. 1987)
we would expect the initial perturbed curve to undergo a transition to a lower wave o f
the same total energy E as the original. From the asymptotic formulae for the potential
and kinetic energies given in Longuet-Higgins & Fox (1978) we find for the total energy
density (Г + V)
E = 0 .0 7 2 8 6 - 0 .3 8 3 e3c o s (2 .1 4 3 1 n e + 1.59). (5.4)

Figure 12 shows £ as a function o f the wave steepness ak. Thus the initial wave, for
which ak = 0.4398 and e = 0.08487, has an energy E = 0.07306, which is the same as
that o f a lower wave with e = 0.2239 and hence ak = 0.4182 (H/L = 0.1331).
The actual long-term development o f the negative instability is shown in figures 13
and 14. Figure 13 shows the crest-to-trough wave height 2 a as a function o f the time,
and figure 14 shows the phase difference, as measured by the position o f the maximum
elevation o f the crest. From figure 13 it appears there is indeed a transition to a lower
wave o f height 0.1303, approximately as expected, but that the wave then returns
almost to the original height, in a kind o f recurrence, and then oscillates. The difference
in phase speed relative to the original wave may be measured by the gradient o f the
2148

Crest instabilities of gravity waves. Part 3 45

F igure 8 . Range o f the perturbations as a function o f the time /. Positive and negative
perturbations (a = ± 0.2) are marked by crosses and square plots, respectively.

x/L
F igure 9. Development o f the surface profile after the positive perturbation (a = 0.2) at time intervals
A/ = 0.05, as seen in a reference frame moving with the phase speed o f the unperturbed wave.
 2149

46 M. S. Longuet-Higgins and D. G. Dommermuth

x/L
Figure 10. Development of the negative perturbation, when a = -0 .2 , in the moving reference
frame of figure 9.

(x+ct)/L
F igure 11. Development of the surface profile after the negative perturbation (a = —0.2) at time
intervals Л/ = 0.2, as seen in the moving reference frame.

curve in figure 14. The maximum gradient corresponds to a difference Ac = 0.0016

hence (2я)/1/2Дс = -0.0040 and this does agree roughly with the difference in speeds
Ac = c1—c2 = —0.0037, (5.5)
see table 3. However, in figure 14 the maximum phase-speed difference is at around
/ = 5, which is not when the wave has its minimum amplitude. Also there is an interval
of time near /= 8 where Ac is apparently negative. Thus the phenomenon is less simple
than it appears at first.
2150

Crest instabilities of gravity waves. Part 3 47

ak
Figure 12. The total energy E and the phase speed с as functions of the wave steepness ak.

t
F igure 13. Crest-to-trough height of the perturbed wave, as a function of the time /.

Figure 15 shows the complete surface profiles at times t = 0,8.4 and 16.0, the vertical
scale being exaggerated. The recurrent character of the transition is clearly different
from the corresponding behaviour in the case of the solitary wave (Tanaka et al. 1987)
where the wave height tends to its lower value exponentially in time. As a check on the
accuracy of the numerical time stepping in figures 13 and 14, we note that the total
energy at time /=19.6 was 1.85022 x 10~3 compared with the initial energy
1.850 31 x 10~3 at time /= 0, a proportional difference of less than 5 x 10"5.
 2151

48 M. S. Longuet-Higgins and D. G. Dommermuth

t
Figure 14. Phase of the perturbed wave, as a function of the time t.

x/L
Figure 15. Surface profiles at times t = 0, 8.4 and 16.0. The vertical scale is enlarged.

e ak H/L E c/ c0
0.08487 0.43982 0.1400 0.07306 1.09262
0.22400 0.41818 0.1331 0.07306 1.08894
Table 3. Comparison of wave parameters

6. Discussion and conclusions

Previous calculations of the normal-mode instabilities of the ‘almost-highest wave’
have been revised. The rate of growth of the fastest-growing mode agrees well with
recent calculations by Tanaka (1995) (see Longuet-Higgins & Tanaka 1997) and
2152

Crest instabilities of gravity waves. Part 3 49

provides an asymptote for them as the crest parameter e = <?д/2c0 tends to zero. The
limiting form of the lowest normal-mode instability is as close as can be expected to
that computed by Tanaka for e = 0.0753 (the smallest value of e for which he made
calculations).
Time stepping of the initially perturbated wave by a boundary-integral method has
shown that the subsequent development of the instability at first follows the linear
theory faithfully. The subsequent nonlinear stages of growth depend crucially on the
sign of the initial perturbation. ‘ Positive’ perturbations lead to overturning of the wave
crest; ‘ negative’ perturbations lead on to a transition of the wave to an almost steady
wave of lower amplitude, followed by a kind of recurrence, the wave returning very
nearly to its initial height and speed.
This behaviour differs from that found for the solitary wave by Tanaka et al. (1987),
where a negative perturbation led to a monatonic transition to a lower solitary wave.
It might be considered that in that case the period of recurrence was infinitely long.
Our results also differ from those of Jillians (1989) who did not report recurrence,
possibly owing to his use of a larger initial perturbation.
Finally it must be emphasized that the type of motion considered in the present
paper is purely irrotational flow without surface tension or viscosity. Especially in the
case of the positive perturbation, surface tension can be expected to have important
effects on the ‘je t’ of the overturning breaker, particularly when the wavelength L is
less than about 2 m, and may prevent it from plunging altogether; see Duncan et al.
(1994). At a later stage in the process, vorticity can be introduced by viscous
dissipation, leading to a capillary roller (Longuet-Higgins 1992). These effects are at
present under investigation.

By arrangement with Professor Peregrine, M. Jervis has applied the perturbations

shown in our figure 6 to steep solitary waves, obtaining results similar to those
described by Tanaka et al. (1987). M.S.L.-H. is supported by the Office of Naval
Research under Contract N00014-91-J-1582 and D.G.D. under Contract N00014-93-
C-0046.

REFEREN CES
C leaver, R. P. 1981 Instabilities o f gravity waves. PhD thesis, University o f Cambridge.
D uncan , J. H., Philomin, V., Q iao, H. & K immel, J. 1994 The formation o f a spilling breaker. P hys.
Fluids 6 , S2.
J illians, W. J. 1989 The superharmonic instability of Stokes waves in deep water. J. Fluid M ech.
204, 563-579.
Longuet-H iggins, M . S. 1992 C apillary rollers and bores. J. Fluid M ech. 240, 659-679.
Longuet-H iggins, M. S. & C leaver, R. P. 1994 Crest instabilities of gravity waves. Part 1. The
almost-highest wave. J. Fluid M ech. 158, 115-129.
L onguet-H iggins, M. S., C leaver, R. P. & Fox, M . J. H. 1994 Crest instabilities of gravity waves.
Part 2. M atching and asymptotic analysis. J. Fluid M ech. 259, 333-344.
Longuet-H iggins, M. S. & C okelet, E. D. 1976 The deformation of steep surface waves on water.
I. A numerical method of computation. P roc. R. S oc. Lond. A 350, 1—26.
Longuet-H iggins, M .S . & Fox, M . J . H . 1977 Theory of the almost-highest w ave: the inner
solution. J. Fluid M ech. 80, 721-741.
Longuet-H iggins, M. S. & Fox, M. J. H. 1978 Theory of the almost-highest wave. Part 2.
Matching and analytic extension. J. Fluid M ech. 259, 333-344.
Longuet-H iggins, M. S. & T anaka , M . 1997 On the crest instabilities of steep surface waves.
J. Fluid M ech. 336, 51-68.
 2153

50 M. S. Longuet-Higgins and D. G. Dommermuth

S affman, P. G. 1985 The superharmonic instability of finite-amplitude water waves. J. Fluid M ech.
159, 169-174.
T an aka , M. 1983 The stability of steep gravity waves. J. Phys. S oc. Japan 52, 3047-3055.
T anaka , M. 1986 The stability of solitary waves. Phys. Fluids 29, 650-655.
T anaka , M. 1995 On the “ crest instabilities” of steep gravity waves. Proc. Workshop on
M athem atical P roblem s in the T heory o f N onlinear Water Waves, CIRC, Luminy, France, 15-19
M ay 1995.
T anaka , М ., Dold, J. W ., L ewy , M. & Peregrine, D. H. 1987 Instability and breaking of a solitary
wave. J. Fluid M ech. 185, 235-248.
V inje, T. & Brevig, P. 1981 Breaking waves on finite water depths: a numerical study. Ship Res.
Inst, o f N orway, Rep. R -111.81.
2154

J. Fluid Mech. (1997), vol. 336, pp. 51-68 51

Copyright © 1997 Cambridge University Press

On the crest instabilities of steep surface waves

By M IC H A E L L O N G U E T -H IG G IN S 1
AND M IT S U H IR O T A N A K A 2
1 Institute of Nonlinear Science, University of California, San Diego, La Jolla,
CA 92093-0402, USA
2Department of Applied Mathematics, Faculty of Engineering,
Gifu University, 1-1 Yanagido, Gifu 501-11, Japan

(Received 19 April 1996 and in revised form 28 August 1996)

The forms of the superharmonic instabilities of irrotational surface waves on deep

water are calculated for wave steepnesses up to 99.9 % of the limiting value. It is found
that as the limiting wave steepness is approached the rates of growth of the lowest two
unstable modes (n = 1 and 2 ) increase according to the asymptotic law suggested by the
theory of the almost-highest wave (Longuet-Higgins & Cleaver 1994; Longuet-
Higgins, Cleaver & Fox 1994; Longuet-Higgins & Dommermuth 1997). Moreover,
each eigenfunction becomes concentrated near the wave crest, with a horizontal scale
proportional to the local radius of curvature at the crest. These are therefore 4crest
instabilities’ in the original sense.
Similar calculations are carried out for the normal-mode instabilities of solitary
waves in shallow water, at steepnesses up to 99.99% of the limiting steepness. Similar
conclusions are found to apply, though with greater accuracy.

1. Introduction
The stability of a train of steep, progressive, irrotational waves in water of finite or
infinite depth is of some interest in connection with the theory of wave breaking.
Subharmonic instabilities, having wavelengths larger than the basic wavelength 2n/k,
can occur at relatively low wave steepness. But if we confine attention to the
superharmonic instabilities, that is to say two-dimensional instabilities having the same
space-periodicity as the unperturbed wave, then it is found that only very steep gravity
waves are appreciably unstable.t In deep water it was shown numerically by Tanaka
(1983) and analytically by Saffman (1985) that gravity waves first become unstable in
this way when the steepness parameter ak attains the value 0.4292, corresponding to
the first maximum in the total energy density. Similarly, for solitary waves Tanaka
(1986) showed that instability first occurred at a height-to-depth ratio a/h —0.7824,
which again corresponded to the first maximum in the total wave energy E.
The wave steepnesses just mentioned are already high, being close to the limiting
steepnesses: ak = 0.4432 for a wave in deep water and a/h = 0.8332 for the solitary
wave. Calculation of the properties of such steep waves by ordinary methods presents
increasing difficulties as the limiting wave is approached. It was therefore of some
interest when Longuet-Higgins & Fox (1977) showed that the crest of any steep Stokes
wave (not the steepest) approached a certain asymptotic form characterized by a length
scale
/ = <?72g, (11)
t We are excluding the ‘ bubbles’ of non-stationary instability pointed out by MacKay & Saffman
(1986) and Kharif & Ramamonjiarisoa (1990).
 2155

52 M. Longuet-Higgins and M. Tanaka

where q is the particle velocity at the wave crest in a reference frame moving with the
phase speed с, and g denotes gravity. The flow has a rounded crest at the origin, and
at infinity tends to the well-known 120° corner flow. In a second paper Longuet-
Higgins & Fox (1978) showed how to fit this inner flow to the outer flow in a wave in
deep water by an asymptotic matching technique assuming the parameter

= gl/4 (1.2)
to be small, c0 = (g/k:)1/2 being the linear phase speed. A corresponding theory for
solitary waves, in which c0 = (gd)1/2, has been given by Longuet-Higgins & Fox (1996).
Now in two recent papers (Longuet-Higgins & Cleaver 1994; Longuet-Higgins,
Cleaver & Fox 1994) it was suggested that for very steep waves the lowest
superharmonic instability of a progressive wave should correspond, in the limit as
e-*0, with an instability of the inner flow of the almost-highest wave. The lowest
instability of the inner flow was therefore calculated, also the first-order correction to
this instability resulting from the matching process. The growth rates of the instability
apparently agreed with those found numerically by Tanaka (1983) for progressive
waves of finite amplitude. Tanaka’s calculations, however, at that time extended only
as far as ak = 0.439, and no eigenfunctions were available. One of us (M.S.L.-H.)
therefore proposed to Tanaka to extend his calculations (with a more modem
computer) to higher values of ak and to calculate if possible the corresponding
eigenfunctions. It is these calculations, first presented last year (Tanaka 1995), that are
reported fully in the present paper. We report also similar calculations for solitary
waves which are more accurate than those for periodic waves.
The calculations by Tanaka (1995) showed the need for a re-examination of the
numerical calculation reported in Longuet-Higgins & Cleaver (1994) and Longuet-
Higgins et al. (1994). This has led to a significant improvement in their accuracy (see
Longuet-Higgins & Dommermuth 1997, referred to herein as LHD). As will be shown
here, the revised values of the rate of growth as € - * 0 agree satisfactorily with those
reported by Tanaka (1995), especially for the lowest mode {n = 1). The revised
calculations also show the presence of a second mode (n = 2 ) as was originally
suggested by Cleaver (1981); this also is consistent with Tanaka’s (1995) calculations.
The plan of this paper is as follows. In §§2 and 3 we describe the method of
computation, and the numerical results, for steep periodic waves in deep water, and in
§§4 and 5 we present corresponding results for the solitary wave in shallow water. In
§ 6 we compare the two sets of results with one another and with the eigenfunctions
obtained by LHD for the limiting case e-*0. A discussion follows in §7.

2. Numerical results for Stokes waves in deep water

To calculate the flow in a steady, irrotational gravity wave of finite amplitude in deep
water we have used the method described in Tanaka (1983). Waves of a given
amplitude are distinguished by a parameter o j = \—qc/qt>where qc and qt are the
particle speed at the crest and the trough, respectively, observed in a reference
frame moving with the Stokes wave. The parameter o j is more suitable as the amplitude
parameter than ak in the sense that its range is known a priori to be 0 ^ oj < 1. Thus
(o = 0 corresponds to the wave with infinitesimal amplitude, while o j = 1 corresponds
to the limiting wave with a stagnation point at its crest (i.e. qc = 0). The range of o j that
we treat here is 0.80 ^ o j ^ 0.97. (Note that o j behaves similarly to the parameter o j '
used by Longuet-Higgins & Fox 1978.)
2156

On the crest instabilities o f steep surface waves 53

N ak с error (%)
1[ 64 0.4267367789 1.0915699348 1.805 x 10-“
w = 0.80 , 128 0.4267368619 1.0915699350 3.820 x 10“10
1[ 256 0.4267368619 1.0915699350 1.602 x Ю" 10
512 0.4426166715 1.0922690822 5.260 x 10°
w = 0.97 - 1024 0.4427541436 1.0922769604 5.666 x 10-3
2048 0.4427542224 1.0922769604 1.258 x 10-®
T able 1. Convergence of the Stokes wave in deep water

(O ak £ c/c0 Rk/e2
0.80 0.42674 0.18858 1.09157 5.3274
0.81 0.42855 0.17917 1.09201 5.3063
0.82 0.43022 0.16974 1.09235 5.2869
0.83 0.43173 0.16030 1.09261 5.2692
0.84 0.43312 0.15086 1.09279 5.2529
0.85 0.43437 0.14141 1.09290 5.2382
0.86 0.43551 0.13196 1.09295 5.2248
0.87 0.43655 0.12251 1.09294 5.2128
0.88 0.43749 0.11307 1.09289 5.2021
0.89 0.43835 0.10362 1.09282 5.1926
0.90 0.43912 0.09419 1.09272 5.1843
0.91 0.43983 0.08475 1.09261 5.1772
0.92 0.44047 0.07533 1.09251 5.1711
0.93 0.44106 0.06591 1.09242 5.1659
0.94 0.441 58 0.05649 1.09235 5.1618
0.95 0.44204 0.04707 1.09231 5.1585
0.96 0.44244 0.03766 1.09228 5.1560
T able 2. Calculated values of the parameters for periodic waves in deep water

Table 1 shows the convergence of the steepness parameter ak and the phase speed
с as the number N of Fourier modes used to express the flow is increased, for the cases
0) = 0.80 and и) = 0.97. The maximum deviation of the Bernoulli constant \q2 + gy
along the free surface is also shown. The limiting value of ak is known to be 0.4432,
hence the ak of the two cases shown in the table are about 96.3% and 99.9% of the
limiting value. The table shows that N = 128 is sufficient to get the convergence of с
to the ten decimal places for w = 0.80, while more than N = 1024 is necessary to get
the same convergence when w = 0.97. The speed of convergence of the series
representing the Stokes wave also determines the speed of convergence of the solution
of the eigenvalue problem in the stability analysis, and this rapid slowdown of
convergence for larger values of шcauses serious difficulty in the stability calculation
later.
The range 0.80 ^ cj ^ 0.97 corresponds to 0.1886 ^ e ^ 0.0282, see table 2. Within
this range of e the asymptotic relations for ak and с derived by Longuet-Higgins & Fox
(1978) coincide with our numerical results up to four significant figures. The last
column of table 2 shows the radius of curvature R at the crest converging to the value
5.15c2 predicted in Longuet-Higgins & Fox (1977). According to the asymptotic
theory for the deep-water wave given in Longuet-Higgins & Fox (1978), the lowest-
 2157

M. Longuet-Higgins and M. Tanaka

5.35 i I r 1 '■-J—1 — 1 1 1 1— 1Г i
1 1“ 1 1 1 1 “I

- -

X
“
5.30
- X
-
- -
- X -

RJl 5.25 — X
—
- X -
- -
X
- -
X
- _
5.20 ___ X
_
_ * _
X

X -
_ X _
X
X

5.15 x* , , , , 1 1__ 1_ _J_ 1 1 1 _]__ L . 1_ 1 J1__ L_J __

0 0.01 0.02
e3(«-l)
Figure 1. The scaled crest curvature R/l for deep-water waves, as a function o f e3**"1’.

order correction to the inner solution is proportional to e3(e-1) where a is the lowest
root of the transcendental equation

|tia tan \ncc - , (2 . 1 )

that is to say a = 1.8027. In figure 1 we have plotted R/l against e3<e-1>, that is e2,4081
and it can be seen that the plots lie almost on a straight line. (A small oscillation about
this line is also to be expected from the analysis.) From figure 1 we may deduce the
more accurate expression
R/l ^ 5.1525 + 9.5 e3^ 1* (2.2)
for the crest curvature R.

3. N orm al modes of instability

In this Section we consider the normal modes of perturbation of the steady waves
calculated in § 2 , that is to say we seek those small perturbations whose time-
dependence is like eA‘, where Л is a constant. The numerical method whereby the
problem is reduced to finding the eigenvalues and eigenfunctions of a pure imaginary
square matrix is explained in detail in Tanaka (1983). As mentioned in §1, for low
values of the wave steepness ak, or equivalently for low values of w all the eigenvalues
Ля are pure imaginary and the normal modes behave sinusoidally in time. At
comparatively large values of ak or w, there appear one or more real eigenvalues, each
corresponding to a perturbation which either grows or decays exponentially with t.
2158

On the crest instabilities of steep surface waves 55

К A? At A?
10 -0.0 2 2 7 6 w = 0.97 70 2.94771
15 -0 .0 1 8 2 0 120 2.63295
20 -0 .0 1 5 0 6 200 2.45113
30 -0 .0 1 4 8 8 300 2.40413
50 -0 .0 1 4 8 8 400 2.39750
70 -0 .0 1 4 8 8 500 2.39659
T able 3. Convergence of the eigenvalue of the lowest mode

tO 90
OJ E A? (eAj)2 (eA2)2
0.80 0.073991 -0 .0 14 8 8 -0.000529 -0 .53 7 23 -0 .0 1 9 1 0 5
0.81 0.074032 -0 .00 3 99 -0.00 0 12 8 -0 .5 0 8 17 -0 .0 1 6 3 12
0.82 0.074026 0.00766 0.000221 -0.4 7 8 6 4 -0 .0 13 7 9 0
0.83 0.073979 0.02026 0.000521 -0 .44 8 60 -0 .0 1 15 2 8
0.84 0.073898 0.03411 0.000776 -0 .4 1 8 0 0 -0 .0 0 9 5 13
0.85 0.073791 0.04957 0.000991 -0.3 8 6 7 5 -0.00 7 73 4
0.86 0.073668 0.06718 0.001170 -0.3 5 4 8 0 -0.00 6 17 8
0.87 0.073534 0.08774 0.001317 -0.3 2 2 0 6 -0.004834
0.88 0.073399 0.11238 0.001437 -0.28 8 49 -0.003688
0.89 0.073268 0.14288 0.001534 -0.25 4 06 -0.00 2 72 8
0.90 0.073149 0.18197 0.001614 -0 .2 18 8 3 -0 .0 0 19 4 1
0.91 0.073046 0.23405 0.001681 -0 .18 2 9 4 -0 .0 0 13 14
0.92 0.072964 0.30643 0.001739 -0 .14 6 5 7 -0.00 0 83 2
0.93 0.072904 0.41179 0.001789 -0 .10 9 8 1 -0.000477
0.94 0.072867 0.57390 0.001831 -0 .0 7 2 19 -0.000230
0.95 0.072849 0.84216 0.001866 -0 .0 3 19 7 -0.000071
0.96 0.072847 1.33468 0.001893 0.01593 0.000023
0.97 0.072852 2.39659 0.001912 0.08742 0.000070
T able 4. Calculated values of the total energy density E and of the squared eigenfrequencies of the
normal-mode perturbations of a deep-water wave

Table 3 shows the convergence of the square of the eigenvalue of the mode
n = 1 for the deep-water waves with o>= 0.80 and o) = 0.97. In these tables, Ne is
the number of Fourier modes which are used to express the eigenfunctions and
does not have any direct relation with N of table 1 besides the trivial relation Ne < N.
In terms of Ne the size of the matrix whose eigenvalue problem we need to solve is
(2Ne+ 1) x (2Ne+ 1 ). When o>= 0.80 we can have convergence of A* to five decimal
places with Ne = 30, while for o j = 0.97 we get the convergence to only two decimal
places even when Ne is increased to 400. When o j = 0.97 and Ne —400, the total
CPU time used in the construction of the matrix and the calculation of all the
eigenvalues and the eigenvectors was about 12 s on VP2600 of the Computation
Center of Nagoya University.
In the analysis of the local flow around the crest by LHD time and space are
normalized in such a way that g = 1 and qc = \/2, while in our calculations they are
normalized such that g = к = 1 for the Stokes waves in deep water. The propagation
speed of the basic steady wave c0 in the limit of infinitesimal amplitude equals 1. In
order to compare our results with those of LHD, we need to multiply our quantities
by e~2 for each dimension of space and e"1 for each dimension of time.
 2159

56 M. Longuet-Higgins and M. Tanaka

ез<«-1)

F igure 2. Squares of the growth rates A„ of the two lowest modes of instability of the deep-water
wave, scaled by the parameter e, shown as a function of e3'*-1’. Crosses denote present calculation;
circles, the asymptotic theory of the almost-highest wave, from Longuet-Higgins & Dommermuth
(1997).

The square of the scaled eigenvalues еЛ1 еЛ2 of the two lowest eigenmodes n = 1 and
n = 2 are shown in table 4 and plotted in figure 2. The time-dependence of the
perturbation is assumed here to be like eA‘, hence Л2 > 0 implies instability. On the
other hand the asymptotic theory given in Longuet-Higgins et al. (1994) and LHD
indicates that the lowest-order corrections to the eigenfrequencies of the inner flow
vary as e3(a_1>. In figure 2 we therefore plot the results as functions of e3(e_1). As
expected we see that the plots indeed fall approximately on a straight line, for small e.
A slight oscillation about the straight line is also seen.
Figure 2 also shows the corresponding results from LHD, and it will be seen there
is close agreement over the expected range of e.
Figure 3 shows the profile of the lowest unstable mode (n = 1) as a function of the
horizontal coordinate x for oj = 0.84, 0.88 and 0.92. It can be seen that as ш increases
the perturbation progressively shrinks and is confined to a narrower region around the
crest (which is at x = 0). The arrow in the figure indicates the direction of propagation
of the Stokes wave. The amplitude of the normal mode, which is arbitrary within the
framework of the linear stability analysis, is normalized in such a way that the
maximum deviation of the profile is equal to unity, and this convention will be used
throughout this paper.
Figure 4 shows the profiles of the same normal mode as those shown in figure 3, but
as functions of the scaled horizontal coordinate x/e2. The widths of the curves are now
2160

On the crest instabilities o f steep surface waves 57

F ig u r e 3. Vertical displacement rjl in the lowest unstable mode of a deep-water wave as a function
of the horizontal coordinate x , when <
j = 0.84, 0.88 and 0.92.

F ig u re 4. Vertical displacement tj1 in the lowest unstable mode of a deep-water wave as a function
of the scaled coordinate x/e2, when w = 0.84, 0.88 and 0.92.
 2161

58 M. Longuet-Higgins and M. Tanaka

comparable, but the values of w do not seem to be large enough for a discussion of the
asymptotic behaviour as e->0. Nevertheless we can clearly observe a tendency to
converge towards some specific profile as 1 (e->0 ).
The second mode n = 2 has just become unstable for the steepest case o>= 0.97, and
it seems obvious that we need to carry on to larger values of w in order to discuss the
asymptotic behaviour of the normal modes. For the case of oj = 0.97, we need to solve
the eigenvalue problem of a matrix larger than 800 x 800 in order to obtain the
convergence of Л2 to three significant figures as we mentioned before. Although the
calculation for steeper cases is simple and straightforward, we did not carry it out for
reasons of expense. As we shall see, greater accuracy is attainable in the case of solitary
wave perturbations.

4. Numerical results for solitary waves in shallow water

The method which we employ here for calculating steady solitary waves is the same
as that explained in detail in Tanaka (1986). In this method, steady solitary waves with
different wave heights are distinguished by a parameter q/c where q = qet the particle
velocity at the wave crest. In common with oj used for the deep-water waves in the
previous section, the range of q/c is known a priori to be 0 ^ q/c < 1. Unlike w,
however, q/c decreases as the wave gets steeper: q/c = 1 corresponds to the solitary
wave with infinitesimal wave height, while q/c = 0 corresponds to the limiting solitary
wave characterized by a 12 0 ° angular corner with a stagnation point (q = 0 ) at the
crest. The range of q/c we treat here is 0.01 < q/c ^ 0.30. Values of the wave height
H and phase speed с at given values of q/c are shown in table 5; also shown is the
radius of curvature R at the wave crest. As in §§2 and 3 we may define an inner length
scale /= q*/2g and a small parameter
e'* = gl/cl cl = gd, (4.1)
which tends to zero as the wave approaches its limiting steepness, but since the value
of cl here differs from the deep-water case the corresponding value of e' is not precisely
equivalent to the e of §§2 and 3 and so will be distinguished with a prime.
Unlike the deep-water wave in which we employed spectral representation for both
the deep-water wave and the perturbations to it, we use for the solitary wave a more
direct method in which the basic solitary wave and the perturbations are both
represented by the values at a discrete set of M points properly distributed along the
free surface. In our calculations a solitary wave and the perturbations to it are always
calculated at the same set of points.
Table 6 shows the convergence of the phase speed c, the wave height H and the
square of the eigenvalues A, and A2 of the two lowest unstable modes, in the case
q/c = 0 .0 1 , as the number of nodes is increased.
Under the normalization g = 1 and d= 1, the limiting height for the steady solitary
wave Hmax is known to be 0.83322, hence the wave height of the highest solitary wave
shown in table 5 is about 99.99% of Hmax. For the steepest deep-water wave that was
treated in the previous section q/c was about 0.05, so it can be said that the solitary
wave shown in table 6 is much closer to the limiting wave than the steepest deep-water
wave that we treated. Table 6 shows that even for such a steep solitary wave, we can
obtain convergence up to nine significant figures for с and up to four significant figures
for Aj with M = 60 only. The size of the matrix whose eigenvalue problem needs to be
solved in the stability analysis of the solitary wave is (2M —1) x (2Af —1). This much
faster convergence suggests that the solitary wave case should be much more promising
2162

On the crest instabilities o f steep surface waves 59

q/c H/d e' c/c0 R/e'2d

0.01 0.83312 0.00913 1.29089 5.15264
0.02 0.83286 0.01826 1.29089 5.15334
0.03 0.83243 0.02738 1.29088 5.15476
0.04 0.83183 0.03651 1.29086 5.15703
0.05 0.83106 0.04564 1.29085 5.16028
0.06 0.83106 0.05477 1.29086 5.16459
0.08 0.82798 0.07303 1.29098 5.17674
0.10 0.82536 0.09131 1.29127 5.19417
0.12 0.82228 0.10961 1.29174 5.21750
0.14 0.81869 0.12793 1.29233 5.24740
0.16 0.81448 0.14628 1.29296 5.28457
0.18 0.80954 0.16464 1.29356 5.32979
0.20 0.80374 0.18300 1.29401 5.38390
0.22 0.79695 0.20133 1.29421 5.44785
0.24 0.78908 0.21961 1.29407 5.52266
0.25 0.78472 0.22872 1.29385 5.56450
0.26 0.78005 0.23781 1.29352 5.60950
0.28 0.76798 0.25590 1.29249 5.70966
0.30 0.75824 0.27385 1.29092 5.82458
T able 5. Parameters for the solitary wave

M c/(gd)m H/d {.Kd/cf (A2d/с)г

30 1.2908876618 0.8331121581 13.0763 0.682318
40 1.2908906712 0.8331160426 13.9659 0.713888
50 1.2908908571 0.833 1162825 13.8925 0.714442
60 1.2908908823 0.8331163150 13.8940 0.714884
70 1.2908908866 0.8331163205 13.8931 0.714946
80 1.2908908874 0.8331163216 13.8930 0.714966
90 1.2908908876 0.8331163218 13.8929 0.714973
T able 6 . Convergence of the solitary wave and the eigenvalues of the two lowest eigenmodes when
q/c = 0.01

than the deep-water wave case for our present purpose which is to clarify the asymptotic
behaviour of unstable normal modes as e-^O.
Table 5 indicates that the scaled radius of curvature R/l approaches the theoretical
value 5.15 as before. According to the asymptotic theory for steep solitary waves
(Longuet-Higgins & Fox 1996) we expect the lowest-order corrections to the inner
solution to be of order e'[3<e-1)], as for the deep-water wave. Accordingly in figure 5 we
have plotted R/l against e't3<«-D3 an(j we obtain as previously an almost straight-line
plot.

5. Eigenmodes for the solitary wave

As shown in figure 6 , the total energy of the solitary wave exhibits three stationary
values within the range 0.01 < q/c < 0.30. This behaviour of E is in agreement with the
known asymptotic theory for steep solitary waves (Longuet-Higgins & Fox 1996).
Hence when q/c = 0.01 we expect the existence of three unstable normal modes.
 2163

60 M. Longuet-Higgins and M. Tanaka

Figure 5. The scaled crest curvature R/l for solitary waves, as a function of e',3<e_l>1.
2164

On the crest instabilities o f steep surface waves 61

q/c / E A? A’
0.01 2.54347 0.97268 13.89290 0.71497
0.02 2.54347 0.97268 3.45947 0.15537
0.03 2.54337 0.97260 1.52518 0.05006
0.04 2.54315 0.97246 0.84689 0.01488
0.05 2.54286 0.97233 0.53209 0.00242
0.06 2.54261 0.97228 0.36050
0.08 2.54275 0.97275 0.188 86
0.10 2.54449 0.97439 0.10866
0.12 2.54839 0.97738 0.06489
0.14 2.55467 0.98164 0.03875
0.16 2.56312 0.98684 0.02242
0.18 2.57325 0.99248 0.01211
0.20 2.58438 0.99795 0.00577
0.22 2.59567 1.00260 0.00215
0.24 2.60620 1.00578 0.00043
0.25 2.61089 1.00661 0.00008
0.26 2.61506 1.00685
0.28 2.62138 1.00527
0.30 2.62436 1.00057
T able 7. C alculated values o f the im pulse I and total energy E o f the so litary w ave, and o f the
squared grow th rates of the two lowest instabilities

e'[3(a-D)
Figure 7. Scaled values of the squared growth rates of the two lowest modes of instability o f the
solitary wave, as a function of
 2165

62 M. Longuet-Higgins and M. Tanaka

'" I гтгтугт Г Т -.-.Н Г П -Т -.- Г Т Т Г Т Г Т Т Т Г Т 1 r p - i . . . . . . .

[(d)

x x
Figure 8. Profile of the first unstable mode o f the solitary wave (solid line). The dashed curve shows
the unperturbed wave profile: (a) q/c = 0.25, (b) q/c = 0.20, (c) q/c = 0.10, ( d ) q/c = 0.02.

F igure 9. Profiles of the first unstable mode of the solitary wave plotted against the scaled
coordinate x/e'2.
2166

On the crest instabilities of steep surface waves 63

x
F igure 10. Comparison of the first unstable mode of the solitary wave when q/c = 0.25 (solid curve)
with a pure phase shift (dashed curve).

Numerical values of the total impulse and energy are given in table 7, along with the
squared growth rates of the two lowest instabilities.
In figure 7 we show the square of the scaled eigenvalues e'An of the two lowest
unstable modes plotted as functions of e,[3(a-1>1. As yet there is no corresponding
asymptotic theory for the eigenvalues at general values of e, but the points approach
the vertical axis along a straight line, as expected. The limiting values of e\n as e-*0,
according to the asymptotic analysis of LHD, were shown in figure 2 above. It will be
seen that these are in reasonable agreement with the present calculations.
Figure 8 (a-d) shows the profiles of the lowest unstable mode (n = 1) for q/c = 0.25,
0.20, 0.10 and 0.02. The dashed line in each figure shows the profile of the unperturbed
solitary wave for the corresponding value of q/c. The horizontal axis is the original
horizontal coordinate x for which g = 1 and d = 1 .It can be seen clearly that the length
scale of the normal mode gradually diminishes as q/c^0 and the perturbation is
confined to a narrower region around the crest in a similar manner as that we observed
for the first unstable mode of the deep-water wave (see figure 3). When we replot in
figure 9 those profiles as functions of the scaled coordinate x/l (that is x/e'2 in our
scaling), there appears to be a convergence of the profile for sufficiently small values
of q/c (but see § 6 below). Figure 10 compares the profile of the first unstable mode
when q/c = 0.25 (solid line) with the profile of the trivial neutral mode corresponding
to a horizontal shift (dashed line); this is simply the derivative of the profile of the
steady solitary wave, for this value of q/c. According to Saffman (1985) the trivial
neutral mode and the non-trivial mode should become identical at the point where the
non-trivial mode exchanges its stability. As the critical point for the lowest non-trivial
mode, which is equal to the first maximum of the total energy, is q/c « 0.259 as figure
6 indicates, the case shown in figure 10 (q/c = 0.25) is supercritical (in this sense) only
marginally. The closeness of the two profiles shown in figure 10 can be regarded as an
indication that our numerical results are in accord with Saffman’s theory.
Figure 11 (a-d) shows the profile ij2 of the second unstable mode as a function of *
for q/c = 0.05, 0.04, 0.03 and 0.02, respectively. The first three curves are also plotted
 2167

64 M. Longuet-Higgins and M. Tanaka

_, l F TT Г ’-1'" ’- ' 1 ................ 1~1 ' 1 1 ■ 1 ' >

: Ф)

: ^

x x
F ig u re 11. Profile of the second unstable mode of the solitary wave (solid line). The dashed curve
shows the unperturbed wave profile:(a) q/c = 0.25, ( b ) q/c = 0.20, (c) q/c = 0.10, (d ) q/c = 0.02.

F ig u re 12. Profiles of the second unstable mode of the solitary wave plotted against the scaled
coordinate х/е'г.
2168

On the crest instabilities of steep surface waves 65

as functions of x/l in figure 12. The profile for q/c = 0.02 is practically indistinguishable
from the profile for q/c = 0.03.

6. Comparison o f the deep-water wave and solitary wave

For general values of the deep-water wave steepness ak or the solitary wave steepness
H/d the rates of growth of the instabilities are not directly comparable as functions of
e, since the parameters e and e' depend on external parameters and hence are defined
differently in the two cases. However the inner length and time scales

depend only on the velocity q, for fixed g, and not on e, so that the scaled frequencies

( 6 .2)
g
can be compared in the limit as e-+0.
For the limiting deep-water wave we have from figure 13, since q = \/2e,
Axq
= 0.0621, ^ = 0.016. (6.3)
g g
This can be compared with the limiting solitary wave for which, from figure 7,
A, q Aj? (6.4)
= 0.0621, ^ = 0.014.
g S
On the other hand the limiting values according to the asymptotic theory of the almost-
highest wave (LHD) are

^ = 0.061(4), ^ = 0.01(8). (6.5)

g g
The above values are reasonably consistent, but the extrapolated eigenvalues in (6.3)
and (6.4) are closer to each other than to those in (6.5), suggesting that the present
calculations are more accurate.
For a similar reason we can legitimately compare the eigenfunctions only in the limit
as 6 -* 0 , by taking as a horizontal scale the length /= q2/2g which is independent of
e. In figure 14 is a comparison of the first unstable mode of the deep-water wave when
w = 0.84 (e = 0.1509) with the corresponding mode of the solitary wave when q/c =
0.16 (e' = 0.1279), but neither of these is necessarily close to the limiting form. For, the
agreement may be due partly to a similarity between the next-lowest-order corrections
to the inner solutions in these two cases.
In figure 15 we compare the first mode of the solitary wave when q/c = 0.02 (e' =
0.0183) with the first unstable mode of the almost-highest wave (from LHD). The
general similarity is clear, but there are significant differences. We remark that even at
this value of e some differences are still to be expected, because the eigenfunction for
the almost-highest wave is that for the inner solution only, valid only when x is of order
e2, i.e. x/l = 0 (1 ). Even in this zone there are necessarily corrections of order e3(A_1) (see
LHD, §2). Further, in the matching zone where x = O(e), i.e. x/l = 0 (e _1), there are
0 ( 1 ) corrections to the inner solution. Thus even for the solitary-wave instability
shown in figure 15, when e' = 0.0183, there are 0 ( 1 ) corrections to the inner solution
 M. Longuet-Higgins and М. Tanaka

e 3 ( « - i)

F ig u re 13. Extrapolation of the eigenfrequencies eA, and еЛ2 for the deep-water wave, from the three
lowest calculated values of e. The fitted lines are given by у = 0.0019305-0.10138 * when n = 1 and
by у = 0.0013251 -0.3147 7 л: when n = 2.

F igure 14. Comparison of the first unstable mode of a deep-water wave (w = 0.84, solid curve) with
that of a solitary wave (q/c = 0.16, dashed curve).
2170

On the crest instabilities of steep surface waves

(q/c = 0.02, dashed curve) with

F igure 15. Comparison of the first unstable mode of a solitary wave
the lowest mode for the inner solution when e = 0 (LHD, solid curve).

when x fl is of order 50. However, when x/l is of order 5, say, we would expect closer
agreement than is evident in the figure.
The eigenfunction for the solitary wave at q/c —0.02 shown in figure 15 is very close
to the solitary-wave eigenfunction at q/c = 0.16 shown in figure 14, which suggests that
it has effectively converged, especially in the inner zone where x/l is of order ■
Probably, therefore, the present determination of the lowest mode is more accurate
than that presented in LHD.

7. Conclusions and discussion

Our extended calculations of the lowest superharmonic instabilities of Stokes waves
in deep water strongly support the conclusion that these are indeed ‘ crest instabilities ,
as first suggested by Longuet-Higgins & Cleaver (1994) and Longuet-Higgins et al.
(1994); the numerical details being as revised by LHD.
Each crest instability first arises at a wave steepness ak corresponding to a stationary
value of the energy density E. At that value of ak it has the form of a pure phase shift
and so is not localized at the wave crest. However, our calculations have followed the
development of the two lowest instabilities (n = 1 and 2 ) as the wave steepness a
approaches its limiting value. For the lowest mode (n = 1) the eigenfrequencies agree
very closely with the asymptotic values given in LHD. The eigenfunctions also
converge well and are in reasonable agreement with the lowest instability of the almost-
highest wave as determinated by LHD.
Contrary to a conclusion stated in Longuet-Higgins & Cleaver (1994) and Longuet-
Higgins et al. (1994), a second superharmonic instability (n = 2 ) has been found and
its eigenfrequency is consistent with the results of LHD. There is no doubt that higher
instabilities (n > 2 ) exist also.
Similar results have been obtained for solitary waves in shallow water, with
increased accuracy. There again the scaling introduced by Longuet-Higgins & Cleaver
(1994) and Longuet-Higgins et al. (1994) is found to govern the form and growth rates
 2171

68 M. Longuet-Higgins and M. Tanaka

of the instabilities, as the solitary waves approach their maximum steepness. It should
be emphasized that the present computations apply only to the linear stages of the crest
instability. The nonlinear stages have been investigated by LHD and have been shown
to lead in some cases to overturning and breaking of the wave crests; see also Tanaka
et al. (1987) and Jillians (1989). In this study we have neglected the effects of surface
tension, which may have a strong effect on the crest instability of any surface wave whose
length is less than 2 m - perhaps more - for example through the production of para­
sitic capillary waves. Once such short-scale features occur, the influence of viscosity will
increase drastically. These effects are presently under investigation by the authors.
We admit that our results can apply to wind waves in only a limited sense. For,
whereas we have treated the instability of a uniform wave train, a typical ocean wave
spectrum displays strong wave grouping and a wide range of length scales. Both of
these features will be associated with other types of wave instability, and hence wave
breaking. Nevertheless, the crest instabilities that we have considered here are
essentially localized phenomena. Hence they may typify the final stages of breaking in
other circumstances, provided the local scale is sufficiently great to avoid the effects of
surface tension.
M.S. L.-H. gratefully acknowledges the support of the US Office of Naval Research
under Contract N00014-91-J-1582.

REFERENCES
C leaver, R. P. 1981 Instabilities of surface gravity waves. PhD thesis, University of Cambridge.
J illians, W. J. 1989 The superharmonic instability of Stokes waves in deep water. J. Fluid Mech.
204, 563-579.
K h a r if, C. & R am am onjiarisoa, A. 1990 On the stability of gravity waves on deep water. J. Fluid
Mech. 218, 163-170.
Longuet-H iggins, M. S. & C leaver, R. P. 1994 Crest instabilities of gravity waves. Part 1. The
almost-highest wave. J. Fluid Mech. 258, 115-129.
Longuet-H iggins, M. S., C leaver, R. P. & Fox, M. J. H. 1994 Crest instabilities of gravity waves.
Part 2. Matching and asymptotic analysis. J. Fluid Mech. 259, 333-344.
Longuet-H iggins , M. S. & D ommermuth, D. G. 1997 Crest instabilities of gravity waves. Part 3.
Nonlinear development and breaking. J. Fluid Mech. 336, 33-50 (referred to herein as LHD).
L onguet-H iggins, M. S. & Fox, M. J. Н. 1977 Theory of the almost-highest wave: the inner
solution. J. Fluid Mech. 80, 721-741.
Longuet-Higgins , M .S. & Fox, M .J.H . 1978 Theory of the almost-highest wave. Part 2.
Matching and analytic extension. J. Fluid Mech. 85, 769-786.
Longuet-Higgins, M. S. & Fox, M. J. H. 1996 Asymptotic theory for the almost-highest solitary
wave. J. Fluid Mech. 317, 1-19.
M a c K a y , R. S. & S affman, P. G. 1986 Stability of water waves. Proc. R. Soc. Lond. A 406,
115-125.
S affman , P. G. 1985 The superharmonic instability of finite-amplitude water waves. J. Fluid Mech.
159, 169-174.
T a n aka , M. 1983 The stability of steep gravity waves. J. Phys. Soc. Japan 52, 3047-3055.
T a n aka , M. 1986 The stability of solitary waves. Phys. Fluids 29, 650-655.
Ta n a k a , M. 1995 On the “crest instabilities” of steep gravity waves. Workshop on Mathematical
Problems in the Theory of Nonlinear Water Waves, CIRM, Luminy, France, May 1995 (Abstract
only).
T a n a ka , М., D old , J. W., L ew y , M. & Peregrine, D. H. 1987 Instability and breaking of a solitary
wave. J. Fluid Mech. 185, 235-248.
Z ufiria, J. A. & S affman , P. G. 1986 The superharmonic instability of finite-amplitude surface
waves on water of finite depth. Stud. Appl. Maths 74, 259-266.
2172

Shear in sta b ility in spilling breakersf

B y M i c h a e l S. L o n g u e t - H i g g i n s
Institute fo r Nonlinear Science, University of California, San Diego, La Jolla,
California 92093-0402, U.S.A.

High-speed photographs of a gently breaking water wave have shown the partic­
ular instability of the wave crest predicted by Longuet-Higgins et al. (J. Fluid
Mech. 259, 333-344 (1994)) and Longuet-Higgins & Cleaver ( J. Fluid Mech. 258,
115-129 (1994)), with a ‘bulge’ on the forward face of the wave and the genera­
tion of parasitic capillaries ahead of the instability, emanating from the ‘toe’ of
the bulge. The photographs also show the unexpected occurence of longer (type
2) capillary waves above the toe of the bulge. In this paper it is shown that the
type 2 capillaries are probably shear-flow instabilities arising from the vorticity
shed by type 1 (parasitic) capillaries.
At large amplitudes the initially linear shear instabilities must break up into
discrete vortices. By formulating a simple model of the shear instabilities as
vortex waves it is shown that the wavelength of the type 2 capillaries should
be about 4 times the wavelength of the type 1 capillaries, a result in agreement
with observation.
When the flow separates at the toe of the bulge, the use of Prandtl’s rule to
estim ate the strength of the vortices confirms the above relation. At larger length-
scales when surface tension is negligible, the argument leads to a simple rule for
predicting the wavelength of the initial surface roughness of a spilling breaker.

1 . In tro d u ctio n
Gently breaking water waves (‘spilling breakers’) have long been only partly un­
derstood. Although a physical model for the quasi-steady flow, in which a tur­
bulent whitecap propagates steadily down the forward face of wave, was given
some years ago by Longuet-Higgins & Turner (1974) and a simple model of the
forward tip of the wave was proposed by Longuet-Higgins (1973, 1990), the ques­
tion of how a spilling breaker is initiated remained obscure until recently when
Duncan et al. (1994 a, b) published high-speed observations that showed that the
process begins with the formation of a ‘bulge’ or instability on the forward face
of the wave, in accordance with theoretical calculations by Longuet-Higgins &
Cleaver (1994) and Longuet-Higgins et al. (1994). In fairly short gravity waves,
with length 1 m or less, the sharp curvature at the ‘toe’ of the bulge produces a
train of ‘parasitic’ capillary waves (see Cox 1958) propagated ahead of the bulge,
owing to anomalous dispersion. Such capillary waves have been shown to be a
surprisingly strong source of vorticity, capable of producing a ‘roller’, or closed
circulation in the crest of the gravity wave (see Longuet-Higgins 1992). However,

t This paper was produced from the author’s disk by using the Tfe* typesetting system.
Proc. R. Soc. Lond. A (1994) 446, 399-409 © 1994 The Royal Society
Printed in Great Britain 399
 2173

400 M. S. Longuet-Higgins
-4 --------------

type 2

Figure 1 . Sketch of the crest instability of a steep gravity wave, slowing the initial bulge, with
type 1 (parasitic) capillaries ahead of the toe and type 2 capillaries above the toe. The gravity
wave travels to the left.

the photographs by Duncan et al. also showed the unexpected growth of a train
of longer capillary waves above the toe of the bulge, as in figure 1 . The latter cap­
illary waves (which we call type 2 to distinguish them from the type 1 parasitic
capillaries) grew in amplitude until they broke and collapsed into turbulence.
The purpose of this paper is to suggest an explanation for the type 2 capillaries.
In short, it is proposed that they are shear instability waves arising from the
shear layer due to vorticity generated by the type 1 capillaries. The theory of
the instability of horizontal shear flows with a free upper surface was initiated by
Stern & Adam (1973) and has been developed by Kawai (1977), Voronovich et
al. (1980), Caponi et al. (1991), Morland et al. (1991), Shrira (1993) and others.
Stern & Adam (1973) consider a simple layer of constant vorticity and uniform
depth overlying a stationary, infinitely deep fluid. Their model leads to a cubic
equation for the dispersion relation, which was analysed in detail by Caponi et
al. (1991). The theory has been developed in connection with the generation of
surface waves by wind. Kawai (1977) showed that the neglect of viscosity had little
effect on the growth rates. Morland et al. (1991) have compared the ‘piece-wise
uniform’ shear model with three smooth distributions of vorticity and have shown
that the wavelength of maximum growth is not sensitive to the model, although
the rates of growth are. Voronovich et al. (1980) added a thin, discontinuous shear
layer at the base of the uniform layer, and showed that this had a de-stabilizing
effect on the rates of growth. They also discussed the possibility of an ‘explosive
instability’ arising from the nonlinear interaction of the resonant triads of linearly
unstable modes.
In §§2 and 3 below we calculate the wavelengths predicted by the model of
Stern & Adam (1973) and show they are in good agreement with the type 2
instabilities observed by Duncan, on the present hypothesis. On the other hand a
much simpler model is proposed in § 4 for the later (nonlinear) stages of growth.
It is argued that the shear layer will tend to break up quickly into an array
of discrete vortices, whose strength is determined by the strength of the initial
shear-layer and the wavelength of the instability. The model leads to a very simple
Proc. R. Soc. Lond. A (1994)
2174

Shear instability in spilling breakers 401

Figure 2. A surface current with uniform shear, in the model proposed by Stern & Adam
(1973). In this diagram, the x-axis is directed to the left.

prediction, namely that the wavelength of the type 2 capillaries will be about four
times that of the type 1 capillaries, if there is no flow separation at the Чое’ of
the crest instability. If there is flow separation, as predicted in Longuet-Higgins
(1992) then the application of Prandtl’s rule allows a sim ilar prediction to be
made (§5).
Such predictions apply in the first place only to the breaking of rather short
gravity waves, with lengths less than about 1 m. The most prominent whitecaps
in the ocean occur on larger scales. It is therefore of some interest that the use of
Prandtl’s rule enables one to generalize the earlier results to much larger spilling
breakers and to estimate the horizontal scale of their surface roughness (see § 6 ).

2. Instability of a surface layer

To consider the initial instability, we adopt the model introduced by Stern &
Adam (1973) (see figure 2 ). A uniform shear layer of thickness H and vorticity Q
lies above a fluid of infinite depth, otherwise at rest. The surface velocity relative
to deep water is thus given by .
ua = flH. (2-1)
Small perturbations of the flow are sought in the form of sin e - waves moving
to the left with speed c. (Note that in figure 2 the x-axis is directed to the
left.) Thus relative to an observer travelling with this speed the velocity in deep
water appears as a steady stream to the right. If x and у are horizontal and
vertical coordinates in this frame, the vertical displacement of the upper and
lower boundaries of the shear layer are given by
rf\ = a i e ,fcx, rj2 = a2elkx ( 2 -2 )

and the stream-functions in and beneath the shear layer have the forms
= \Пу2
2 + (ПН - c)y + (АекУ + Ве~кУ)ёкх
(2.3)
Ф2 = -\П Н 2 - су + Dekyeikx, '}
where а ь а2, А ,£ and D are constants, possibly complex. On applying the con con­
ditions of constant pressure at у = 771, continuity of pressure across у = r)2y and
Proc. R. Soc. Lond. A (1994)
 2175

402 M. S. Longuet-Higgins
the kinematic boundary conditions, one obtains a dispersion relation in the form
of a cubic equation for
в = (с - tis)/c0, where c0 = (g/k + Tk)1/2 (2.4)
and g and T denote gravity and surface tension. (The density is taken as 1 .) The
equation for 9 is
в3 + (F + 2q)62 - ( 1 + 0F )6 + F = 0, (2.5)
where
q = u./c„ 0 = q/kH, F = 10(1 - e~2kH) - q. (2.6)
After solving (2.5), the radian frequency a of the waves relative to the stationary
axes is found from
а = ck = (в + q)c0k. ( 2 .7 )
If any of the roots в of equation (2.5) are complex, then <
7 * = Im(cr) is interpreted
as a time-rate of the growth of the perturbation.
It is convenient to take units such that
<7 = 1 , T = 1, (2 .8 )
so that the wavenumber km and speed Cm of ordinary surface waves of minimum
phase speed are 1 and y/2 respectively. The corresponding wavelength Am is 27r.
In c.g.s. units, Am and cm would be 1.74 cm and 23 cm s-1, for clean water at
20 °C.
In analysing equation (2.5) Caponi et al (1991) showed that instability occurs
only if us > cm and if q and /? lie within a certain wedge-shaped zone in the ( q, /?)
plane. Morland et al (1991) showed that the rate of growth satisfies the Howard
semi-circle theorem, namely that с lies inside a circle with the line joining (0 , 0 )
and (us,0) as diameter. For a typical dispersion relation derived from equation
(2.5) see figure 3 below.

3. Application o f the model

Figure 2 a-d of Duncan (19946) shows gentle breaking of a 1.4 Hz wave, with
the initial bulge and formation of a train of parasitic capillaries ahead of the toe.
From the wavelength Ai = 0.11 cm of the parasitic capillaries we deduce their
phase-speed
Cl = (Tfc0 1/2 = 65cm s ' 1. (3.1)
Because the capillaries are steady, C\ is also the velocity in the stream beneath.
Now it was shown in Longuet-Higgins (1992) that the total velocity difference
across the shear layer shed by the parasitic (type 1 ) capillaries is
uB= 0.S7{ak)2od (3.2)
approximately, where (afc)0 represents the maximum steepness of the capillaries,
that is the steepness near the toe. Because (ak)Qis seen from the photograph to
be of order 1 or greater we may take us = Ci = 4.0(pT)1/4.
As regards Я , the thickness of the viscous boundary-layer beneath the type 2
capillaries will be of order 4 (vt)1! 2, where t denotes the time taken for the stream
to traverse the type 1 and half of the type 2 capillaries, a distance of about 3 cm,
from the photograph. Hence t « (3/65) = 0.05 s and H w 0.1 cm = 0.4(T/p)1/2.
Figure 3a shows dispersion relation, i.e. a plot of aT against к in the case
Proc. R. Soc. Lond. A (1994)
2176

Shear instability in spilling breakers 403

Figure 3. Dispersion relation according to the model of figure 2, when uB= 4.0(gT)l/4 and
H = 0.4(p/T)1/2. (a) crr as a function of к and (6) (7* as a function of к in dimensionless units.

us = 4.0, H = 0.4 in dimensionless units. In the range 0 < A < 1.31 there are
three stable solutions, the uppermost curve corresponding to the surface gravity-
capillary mode, and the lower two curves mainly waves on the interface у — —H.
Sim ilarly when к > 7.35. In the intermediate range, however, cr has a non-zero
im aginary part, and one mode is unstable. Prom figure 3 6 , which shows o* plotted
against fc, it can be seen that the maximum rate of growth occurs when fc = 3.41,
corresponding to a (dimensional) wavelength of 1.74/3.41 = 0.51 cm. This is not
fax from the observed value of 0 .5 5 cm.
The corresponding growth-rate in figure 3 6 is <
7 * = 0.965, or dimensional units

<Ti = 0.965(g3/T)1/4 = 57.5s"1.

The interval between frames in figure 2 of Duncan (1994 6 ) being 0.009 s, we see
that this implies a growth by a factor e0 52 = 1 .6 8 between each frame; somewhat
less than the observed value (about 1 .8 ).
The streamlines corresponding to the instability with the maximum rate of
growth are shown in figure 4. The frame of reference moves with the phase-
velocity c. The surface stream is to the left relative to deep water, as also is the
phase-velocity. Some circulation in closed circuits can be seen, also an asymmetry
in the streamlines and in the lower boundary of the shear layer (shown by a broken
curve).
Because of the thinness of the shear layer (in this example kH = 1.1) the
perturbations will rapidly become nonlinear, and the layer will tend to break
up into discrete vortices, spaced at intervals of one wavelength of the initial
instability. It will therefore be helpful to model the latter stages of the flow as
the response of the free surface to an array of discrete vortices, as follows.

4. Vortex waves
Consider a horizontal array of vortices, having circulation 27Г7, each at a depth
h below the near surface level and separated by a distance 2тг/fc (see figure 5).
We shall later take h = \H and fc = fc2. If the free surface were a perfect plane,
there would be an image array of vortices with circulation —27Г7 at a height h
above the surface. Under the influence of the image array the whole system would
move to the left with speed
Proc. R. Soc. bond. A (1994)
 2177

404 M. S. Longuet-Higgins

- X

Figure 4. Steamlines in the most unstable mode corresponding to figure 3 a, b. The broken
curves indicate the boundaries of the vortical layer. (The я-axis is to the left).

c = —\^k coth kh (4.1)

(see Lamb 1932, §64). Relative to axes (Ox, Oy) moving with the array, the particle
velocity (u,v) induced by the vortex array plus its image at a general point
z = x + iy is given by
u —iv = |i7 &[cot \k(z + ih) —cot \k(z —i/i)], (4.2)
which can be simplified to
jk s W k h
cosk z-cosh kh K J
Consider the horizontal velocity of a particle in the plane у = 0. Assuming for
simplicity that
4e =6^1 (4.4)
we find that the relative to stationary axes
и = —yk[l + cos k(x —ct)]. (4.5)
Thus at points immediately above a vortex, say (x —ct) = 0, the particle speed
|u| will be slightly greater than at intermediate points, say (x —ct) = 7r/к. If the
surface were held flat, then by Bernoulli’s theorem the pressure p on the plane
у = 0 would fluctuate by an amount
p' — —ecyk cos k(x —ct)> (4.6)
being less immediately above a vortex.
Now if the surface is relaxed, this travelling disturbance will induce a surface
W2LV6
TJ — a cos k(x —ci), (4.7)
say, whose amplitude a can be determined from the boundary conditions. In fact
Proc. R. Soc. Lond. A (1994)
2178

S hea r in sta b ility in spilling breakers 405

*---------- 2n/k ----------- ►

Q -y 0 " y Q

G О G
5 ук coth kh

Figure 5. Sketch of vortex waves, due to an array of vortices travelling to the left.

using th e general expression

sinh kh
u — \v = c + £c6e_lfc(z“ ct) + 7 к (4.8)
cos k ( z - ct) - cosh kh

for th e particle velocity and applying the conditions th a t

g “ (2/ - v ) — 0 and P = const. - T d 2r ) / d x 2 (4.9)

on у = 77 and th a t th e irrotational or non-singular p a rt of the velocity vanishes

a t z = - i / i we find, to first order in c, th a t

ak = € (4.10)
с2 - с Г

where c 0 is the speed of the free capillary-gravity waves, th a t is

d = ^ + Tk. (4.11)

From the equation (4.10) it follows th a t when cj > c2, th a t is when the surface is
relatively ‘stiff’, the surface elevation 77 im mediately above the vortices is negative,
i.e. the surface is drawn down by the Bernoulli effect. On the other hand when
cj < c2, th a t is for a relatively relaxed surface, the sign of a k is positive and the
surface above each vortex is raised.
When с = c0 the model does not apply, but equation (4.10) suggests th a t as
с approaches Co there is a kind of ‘resonance’, when the response of the fluid
becomes large, possibly causing the waves to break.
Proc. R. Soc. Lond. A (1994)
 2179

406 M. S. Longuet-Higgins

5. Vortex ripples
From the above simple model we may predict a relation between the wave­
lengths of type 1 and the type 2 capillaries. (We denote corresponding quantities
by suffixes 1 and 2.) For, the total vorticity in one wavelength of the shear layer
is given by
2tt7 = (2тг/ к 2) П Н = ( 2 п / к 2)ив. (5.1)
So
7^2 = иь « Ci (5.2)
by §3. B ut we saw th at c2 = §7 &2. Hence

сг « Jci. (5.3)
B ut for capillary waves of wavenumber к we have in general th at c2 = T k . Thus
if the type 2 waves are in resonance we have
k i / k 2 = c j/c j ~ 4. (5.4)
In other words, the wavelength of the type 2 capillaries should be about four
times th a t of the type 1 capillaries. The observed wavelengths being 0.55 cm and
0 .1 1 cm we see th at this relationship was approximately satisfied.
Equation (5.4) is valid only approximately, depending as it does on the relation
u s « Ci, which in turn depends on the observation ( a k ) 0 « 1. If for example, in
equation (3.2), (a k 0) > 1.07 we shall have u8 > Ci, so th at the surface velocity
above the toe of the instability will be negative relative to the stream below, and
mass will be transported down the forward face of the wave towards the toe. At
still larger values of ия this may cause the bulge itself to advance furthur down
the forward face, as is observed. Another reason may be th at, as the wave crest
becomes flatter owing to dissipation of the wave energy or to the advance of the
wave through the wave group, the horizontal scale of the ‘crest instability’, which
is roughly proporational to the radius of the curvature at the crest of the gravity
wave, m ust grow in time, and so advance down the forward face of the wave.
An approxim ate rule for the length of the most unstable waves in a shear flow
can also be deduced. Thus in §3, if we equate 7 к with f l H then the speed of
the instability waves is | u s. If the waves are in resonance with the free capillary
wave, then ,
\ u > * c 0 = (g /k + T k ) l' \ (5.5)
For a given value of u B, this gives us a quadratic equation to estim ate k. For
example, when u s — 4.0 then к = 2 + \/3 = 3.73, compared with the value
к = 3.41 in figure 36. However, the estim ated speed of the instability is \ u t = 2.0,
compared with the calculated value a T/ k — 0.82.

6. Flow separation
W hen the steepness (a k ) 0 of the parasitic (type 1) capillaries approaches its
limiting value 2.29 the flow will tend to separate from the sharp troughs of the
capillaries (see Longuet-Higgins 1992). Flow separation can also be expected on
the forward face of the larger gravity waves, when the tendency of the ‘crest
instability’ to overturn (see Longuet-Higgins & Cleaver 1994; Longuet-Higgins et
al. 1994) can no longer be prevented by the action of the capillarity near the toe

Proc. R. Soc. Lond. A (1994)

2180

S hea r in sta b ility in spilling breakers 407

of th e instability. One m ay expect overturning to occur first a t some critical value
of th e W eber num ber
W e = \ U 2I T k , (6 . 1 )
where U denotes th e tangential velocity of the fluid in a reference frame moving
w ith th e waves, and к is th e m axim um curvature of the free surface.
On large scales, flow separation, accom panied by the entrainm ent of air a t the
toe of th e crest instability, occurs normally. However, if th e shear layer remains
w ithin a wavelength of th e free surface we can still apply the results of § 4 through
the use of th e P ra n d tl’s rule (see P ran d tl & T ietjens 1934). T his states th a t
the ra te of shedding of circulation into th e free shear layer is equal to |Q 2,
where Q denotes th e velocity difference across th e boundary-layer ju st before
flow separation (see figure 1 ). In our case, if the downward flow Q' from above
the flow separation point is sm all com pared w ith {/, we m ay take Q = {/. T hus
d C /d t = \U \ ( 6 .2 )

(It is convenient to th in k of the right-hand side of (6.2) as U tim es \ U , i.e. the

vorticity in teg rated across th e boundary-layer tim es th e m ean stream velocity.)
On integrating over one wave period r 2 of the shear instability, we see th a t the
to ta l circulation in one wavelength A2 of the shear instability is given by

2 тг7 = Г \ U 2d t = \ U 2t 2. (6.3)
Jo
Hence
7 fc2 = \ U 2t 2/ A2 = \ U 2/ c2, (6.4)
where c2 is th e speed of the shear waves. B ut we saw th a t if the shear waves are
in resonance w ith th e free surface waves then | 7 k 2 — c2. So from equation (6.4)

c* = itfVc*. (6.5)
Therefore
Cj = i U, (6 .6 )
as in § 3.
Now equation (6 .6 ) is applicable not only to ripples bu t to all types of gravity-
capillary waves, including large-scale gravity waves. Since for such waves c | =
p'/fc2, where g' is the effective value of gravity allowing for the tilt of the unper­
tu rb ed surface and the real acceleration of the unperturbed flow (c.f. Longuet-
Higgins 1987), we have the result th a t
A2 = n U 2/ 2 g r (6.7)
relating the wavelength A2 of the shear instabilities to the speed U of the flow a t
th e point of the flow separation. More generally, if capillarity is included, equation
(6.7) becomes
A2 = ^ [ U 2 ± (U 4 - 64g ' T ) 1' 2] (6.8)
provided th a t U 2 > 8 (s 'T )1/2.
To apply equation (6.7), for example, we have only to determ ine the velocity U
at the toe of the spilling breaker. This may be done through Bernoulli’s equation,
if we know the speed с of the gravity wave, together with the elevation Y of th e
Proc. R. Soc. bond. A (1994)
 2181

408 M. S. Longuet-Higgins

toe of the spilling breaker above the mean surface level. As the particle speed at
the mean level is equal to с (see Lamb 1932, §250), U 2 may be found from
U 2 = c2 - 2g Y } (6.9)
g being the true value of gravity.

7. Discussion
We have shown th at the type 2 capillary waves observed by Duncan et al.
(1994 a, 6 ) in their experiments are very probably instabilities in the shear flow
induced by the type 1 capillaries, i.e. those propagated ahead of the toe of the crest
instability. The type 2 capillaries quickly grow and collapse into turbulence. We
may infer the existence of such a phenomenon for most gently-breaking surface
gravity waves with wavelengths between about 10 cm and 1 m. For the shear
instability to occur, the speed of the type 2 capillaries, which is about half the
speed of the type 1 capillaries, must be greater than the minimum velocity of
capillary-gravity waves (23 cm s-1). Hence the speed of the type 1 capillaries,
which equals the particle speed U in the gravity wave before instability, must
exceed about 46 c m s '1.
In §3 we considered instabilities th at were uniform in space and growing in
time. In some circumstances it may be more appropriate to consider instabilities
th a t гиге stationary in tim e but growing exponentially in a horizontal direction.
Their rate of growth will be described by a complex wave number к = kr -1- ikit
the imaginary p art k{ being related to the growth-rate cr* by k{ = сг{/ с д where cg
denotes the group-velocity dcrr /dA:. This is readily determined from graphs such
as figure 3 a.
Instabilities above the toe of the bulge are not limited to capillary waves. On
a larger scale, particularly when there is flow separation at the toe, there may
be type 2 instabilities which are essentially gravity waves. B ut their phase-speed
will be about | С/, and will be close to the speed of free surface waves of the same
wavelength.
In §4 we have of course assumed th a t the discrete vortices, though possibly
unstable themselves, remain effectively stable and two-dimensional for a certain
period of time.
It will be understood th at all of the analysis in this paper applies solely to
the initial stages of a spilling breaker. Later stages have been partly modelled
by Longuet-Higgins (1973, 1990) and by Longuet-Higgins к Turner (1974). From
qualitative observations Peregrine & Svendsen (1978) have suggested th at the
flow separation seen to occur at the toe of a well-established spilling breaker or
bore can be considered as a free shear layer or mixing layer. The break-up of such
a layer into vortices, and the subsequently thickening of the layer by amalgama­
tion of pairs or triplets of vortices, was beautifully dem onstrated by W inant &
Browand (1974); see also Brown & Roshko (1974). Such mixing layers may not
be too far from the free surface to be associated directly with surface waves. Al­
though some preliminary observations of the turbulence in spilling breakers have
been made by Longuet-Higgins and Donelan (see Longuet-Higgins 1974) and by
B attjes & Sakai (1981), further experim ental studies of mixing layers with a free
surface are desirable.
Proc. R. Soc. Lond. A (1994)
2182

S hea r in sta b ility in spilling breakers 409

This work was supported by the Office of Naval Research under Contract N00014-94-1-0008.

R e fe re n c e s
Battjes, J. A. & Sakai, Т. 1981 Velocity field in a steady breaker. J. Fluid Mech. I l l , 421-437.
Brown, G. L. & Roshko, A. 1974 On density effects and large structure in turbulent mixing
layers. J. Fluid Mech. 64, 775-816.
Cox, C. S. 1958 Measurements of slopes of high-frequency wind waves. J. Mar. Res. 16, 199-225.
Caponi, E. A., Yuen, H. C., Milinazzo, F. A. & Saffman, P. G. 1991 Water-wave instability
induced by a drift layer. J. Fluid Mech. 222, 207-213.
Duncan, J. H., Philomin, V., Behres, M. & Kimmel, J. 1994 a The formation of spilling water
waves. Phys. Fluids. (In the press.)
Duncan, J. H. et al. 1994 6 Gallery of fluid motions. Phys. Fluids 6(9). (In the press.)
Karman, T. Von 1911 Fliissigkeits-u. Luftwiderstand. Phys. Zeitschr. 13, 49.
Kawai, S. 1977 On the generation of wind waves relating to the shear flow in water. A preliminary
study. Sci. Rep. Tohoku Univ. ser. 5, Geophys. 24, 1-17.
Lamb, H. 1932 Introduction to hydrodynamics, 6th edn. Cambridge University Press.
Longuet-Higgins, M. S. 1973 A model of flow separation at a free surface. J. Fluid Mech. 57,
129-148.
Longuet-Higgins, M. S. 1974 Breaking waves - in deep or shallow water. In Proc. 10th Syrup,
on Naval Hydrodynamics, Cambridge, Mass., pp. 597-605. Arlington, VA: Office of Naval
Research.
Longuet-Higgins, M. S. 1987 The propagation of short surface waves on longer gravity waves,
J. Fluid Mech. 117, 293-306.
Longuet-Higgins, M. S. 1990 Flow separation near the crests of short gravity waves. J. Phys.
Oceanogr. 20, 595-599.
Longuet-Higgins, M. S. 1992 Capillary rollers and bores. J. Fluid Mech. 240, 659-679.
Longuet-Higgins, M. S. 1993 The crest instability of steep gravity waves, or How do short waves
break? In Proc. Symp. on the Air-Sea Interface, Marseille, France, 24-30 June 1993.
Longuet-Higgins, M. S. & Cleaver, R. P. 1994 Crest instabilities of gravity waves. I. The inner
solution. J. Fluid Mech. 258, 115-129.
Longuet-Higgins, M. S. & Turner, S. J. 1974 An entraining plume model of a spilling breaker.
J. Fluid Mech. 63, 1-20.
Longuet-Higgins, M. S., Cleaver, R. P. & Fox, M. J. H. 1994 Crest instabilities of gravity waves.
II. Matching and asymptotic expansion. J. Fluid Mech. 259, 333-344.
Morland, L. C., Saffman, P. G. & Yuen, H. C. 1991 Waves generated by shear layer instabilities.
Proc. R. Soc. Lond. A 433, 441-450.
Peregrine, D. H. & Svendsen, I. A. 1978 Spilling breakers, bores and hydraulic jumps. In Proc.
16th Conf. on Coastal Energy, pp. 540-550. New York: ASCE.
Prandtl, L. & Tietjens, O. G. 1934 Fundamentals of hydro- and aeromechanics (translated by
L. Rosenhead). New York: Dover.
Stern, М. E. & Adam, Y. A. 1973 Capillary waves generated by a shear current in water. Mem.
Soc. R. Liege 6, 179-185.
Shrira, V. I. 1993 Surface waves on shear currents: solution of the boundary-layer problem.
J. Fluid Mech. 252, 565-584.
Voronovich, A. G., Labanov, E. D. & Rybak, S. A. 1980 On the stability of gravitational-
capillary waves in the presence of a vertically non-uniform current. Izv. Atm. Ocean. Phys.
16, 220-222.
Winant, C. D. & Browand, F. K. 1974 Vortex pairing: the mechanism of turbulent mixing-layer
growth at moderate Reynolds number. J. Fluid Mech. 63, 237-255.

Received 7 A pril 1994; accepted 1 June 1994

Proc. R. Soc. Lond. A (1994)

 2183

J. Fluid Mech. (1998), vol. 364, pp. 147-162. Printed in the United Kingdom 147
© 1998 Cambridge University Press

Instabilities o f a horizontal shear flow with a

free surface
By M I C H A E L S. L O N G U E T - H I G G I N S
Institute for Nonlinear Science, University of California, San Diego, La Jolla, CA 92093-0402, USA

(Received 4 August 1997 and in revised form 15 December 1997)

The simple shear-flow model of Stern & Adam (1973), in which a layer of uniform
vorticity and depth overlies an infinitely deep fluid, is here extended by the addition
of an upper fluid layer of uniform thickness and constant velocity. In this way many
experimentally observed velocity profiles can be approximated. The normal mode
instabilities of such a model can be found analytically, and their properties calculated
through the solution of a quartic polynomial equation. The dispersion relation is here
determined and illustrated in its dependence on the Froude number and on the ratio
Я 1/ Я 2, where Hi and H 2 denote the mean depths of the surface layer and the base
of the shear layer, respectively. It is found that two branches of instability which are
distinct when Я 1/ Я 2 is moderate or small can become merged when H i / H 2 ^ 0.4924.
Also calculated are the fastest-growing modes, and their wavelengths. The results are
applied to some examples of surface flows generated by towed bodies, and to steady
spilling breakers.

1. Introduction
The theory of the stability of surface shear flows has important applications to the
study of surface wakes behind ships or towed bodies (Dimas & Triantyfallou 1994);
to the stability of the crest of a spilling breaker (Coakley & Duncan 1997); and to
wind-drift currents at the sea surface (Stern & Adam 1973). Such currents are usually
unstable, and over large horizontal distances may develop into a turbulent mixing
layer. But over short distances they may usefully be treated as steady horizontal
shear layers; see for example Peregrine (1974). Theoretical interest then focuses
on the normal instabilities of such layers, especially those with the largest growth
rates, which can give an indication of the horizontal length scales of the observed
perturbations.
The presence of a free surface is an additionally interesting feature. For, whereas
mixing layers in an unbounded fluid have been well studied (see especially Winant &
Browand 1974), those near a free surface are less well understood.
The simplest theoretical model of a surface shearing current was that due to Stern
& Adam (1973), who assumed a surface layer of uniform thickness and vorticity
overlying a stationary layer of infinite depth. Other models or developments of
the same model are due to Kawai (1977), Voronovich, Lobanov & Rybak (1980),
Milinazzo & Saffman (1990), Caponi et a l (1991), Morland, Saffman & Yuen (1991)
and Shrira (1993).
Recently Dimas & Triantyfallou (1994) have computed the linear stability of a
shear flow with a continuous velocity profile of the form
u = U s e c h 2 by (1.1)
2184

148 M. S. Longuet-Higgins

F ig u r e 1. S k etch o f the velocity profile for the m od el flow.

(see figure 9 below) and have followed the nonlinear development of a typical
sinusoidal instability. It is the linear aspect of the problem which will be considered
here. To be precise, Dimas & Triantyfallou find, as in an earlier paper (Triantyfallou
& Dimas 1989) that the normal instabilities of the profile (1.1) appear to be of
two general kinds: Branch I, which are im portant at low wavenumbers, and Branch
II which are im portant at high wavenumbers for low Froude numbers, or at low
wavenumbers for high Froude numbers. However, their method o f calculation is
elaborate and their numerical results are consequently incomplete. There seems to be
good reason for adopting a much simpler model, representing the observed current
profile almost equally well, but amenable to an analytical treatment.
In fact such a model is readily available. It involves simply an extension of the
shear-flow model of Stern & Adam (1973) by the addition of a layer of uniform flow
immediately below the surface. The model profile depends on three parameters: the
surface current U and the depths Hi and H2 of the two upper layers. One particular
com bination of these parameters fits equation (1.1) reasonably well. Moreover, there
is no difficulty in exploring other values of the parameters since the solution to
the stability problem is now reduced to determining the roots o f a simple quartic
polynomial. Not only is this procedure much faster than a purely numerical method,
it is also more complete and accurate, and reveals the relation between different
branches of the solution in a more transparent way.
In one special case (H\ = 0) the flow reduces to that of Stem & Adam (1973), and
in another case (Hi = H2) it reduces to a thin shear layer at a depth H 2 below the
free surface.
The plan of the paper is as follows. Section 2 introduces the model and the
basic equations. Section 3 treats especially the boundary conditions, and derives the
dispersion relation. This is then discussed, with examples, in §4; see figures 2 to 5.
The boundaries of unstable regions are shown in figure 6 . In §5 we discuss some
special cases. Some comparisons with continuous velocity profiles are given in § 6 and
in §§ 7 and 8 we apply the theory to the flow observed in a steady spilling breaker. A
general discussion follows in § 9 .

2. Basic equations
Consider a volume of inviscid, incompressible fluid whose velocity и is directed
horizontally in the x-direction, and depends only on the vertical coordinate y \ see
 2185

Instabilities o f a horizontal shear flow with a free surface 149

figure 1. In particular we take
—Hi < у < 0
u(y) i= u'
^ Q {y 4- Я 2), —H2 < у < —Hi (2.1)
1 °. У< -н 2
so that by continuity at у = —Hi
U = Q (H2 - H i ) . (2.2)
It will be convenient to denote the mean depth of the shear layer by
H = i ( H , + H 2) (2.3 )

and to define the Froude number F by

F2 = U 2/g H . (2.4)

Then, if the surface tension T is neglected, the basic flow depends only on two
dimensionless numbers: H \ / H and F.
We shall consider sinusoidal perturbations of this flow travelling with phase speed
с to the right. In a frame of reference travelling with the phase speed the two
components (u, u) of the fluid velocity will be given by expressions of the following
form. In the upper layer,
и = ( и —с) + ЩВе** - В е - к”)е*х, \
v= к(Аеку + Bc~kf)e'kx. ] ‘
It is understood that on the right the real part is to be taken. In the intermediate
layer,
u = Q ( y + H 2) - c + ik(C<*> - De~k> )e^ , \
u= *(Ce*' + D e-^e**. J l '
In the lowest, semi-infinite layer
u = —c + i*Ee*'+ftjc,
(2.7)
v= kEeky+'kx.
;}
These expressions satisfy the equation of continuity d u /d x + d v / d y = 0. The vorticity
is given by

c = S - | = ° - - fl ’ ° (z8)
in the three regions respectively.

3. B o u n d a ry co n d itio n s
It is assumed that at the free surface the pressure is a constant (zero), and that at
the two interfaces both the pressure and the normal velocity are continuous.
The horizontal derivative of the pressure p is found from the momentum equation

[p + g y + j(“2 + o2)] = (3.1)

The density is taken to be unity. On linearization, equation (3.1) becomes

^ -(p + uu') = tv , (3.2)

2186

150 M. S. Longuet-Higgins
where a prime denotes the perturbation velocity. Using the equation of continuity we
then have

(3.3)
ox dy
At the free surface у = r\, say, the vertical velocity v is equal to u d tj/d x f . In
the linearized theory the boundary condition at the surface can be replaced by the
condition that
p = ( g q - T d ^ / d x 2) (3.4)
at the mean surface level у = 0. Differentiating with respect to x then gives us

и — gv — T d 2v / d x 2 on у — 0. (3.5)

From now on until § 7 we neglect surface tension by taking T = 0. Since £ vanishes

in the upper layer we obtain as the upper boundary condition

( U — c)2-^—— gv on у = 0. (3.6)
dy

Hence from equation (2.5)

(17 - c f k ( A — B) = g(A + B). (3.7)

At the interface between the two upper layers, continuity of the normal velocity
implies, to first order, the continuity of v . This and the continuity of d p / d x gives us

Ac~kHl + BekHl = Ce~kHl + B t kHl, }

( U - c ) 2(Ae~kH' - B e kH') = ( V - c ) 2(Ce~kH' - D e k'H') ? (3-8)
- ( U - c)(Q/k)(Ce~kH' + DekH'). J

At the lower interface we find similarly

Ce~kHl + DekHli =
= E e - klH\ )
(3.9)
c2(Ce~kHl - DekHl) + c(Q /k)(C e~kH' + De*H ) = c2£ e - ‘H!.J

If we write for short

Я] = e~2kH‘, A2 = e- 2*"1 (З.Ю)
and introduce the notation

«. = (gA)1/2, ч = a/co, z = ^CO (эн)

and
(3.12)
kc o’

t This is an approximation, based on the assumption that the rate o f growth o f the instability is
sufficiently small.
 2187

Instabilities o f a horizontal shear flow with a free surface 151

then equations (3.7), (3.8) and (3.9) become
Z 2(A - В ) A + B,

X\A + В X\C + D,

Z(X\A - B) Z (X l C - D ) + p(Xl C + D), (3.13)

x2c + d X2E t

( Z + q ) ( X 2C - D ) p(X2C + D) = ( Z + q ) X 2E.
Eliminating E between the last two of these equations gives us
( Z + q ) D - t f ( X 2C + D ) = 0 (3.14)
and we are left with four linear equations for А, В, С and D with matrix
/ ( Z 2 - 1) ~ ( Z 2 + 1) 0 0 \
Xi 1 -Xy -1
(3.15)
X{Z -Z -x ^ z+ fi) (Z -P )

\ о 0 - \ X 2fi (Z + q -tf)J
The vanishing of the determinant of this matrix gives an equation for Z and hence
the dispersion relation.

4. T h e disp ersio n relation

By manipulating rows and columns in (3.15) the vanishing of the determinant
becomes equivalent to
( Z 2 — 1) 2 0 0
Xi -{X x+1) X\ -1
= 0, (4.1)
0 Z XyOL (Z —a)
0 0 X2a (Z + q - a)

where
_ l (4.2)
p-
On expansion of the determinant, equation (4.1) reduces to
A,Z(Z2- l ) ( Z + 4 )

+ Я,(Я, - A 2) a Z ( Z 2 + l)

+ a IMq - (A, - Хг)а ] [X,(Z2 + 1) + (Z 2 - 1)] = 0, (4.3)

a quartic equation for Z , which may be solved precisely by radicals; see for example
Turnbull (1957).
Having obtained Z , the relation between the wave frequency со (relative to deep
water) and the wavenumber /с, can be found from
(o =* kc = fcco(Z + q \ (4.4)
where c0 = (g / k ) l/2. For real values of the wavenumber k, the phase speed is given
2188

152 M. S. Longuet-Higgins

Figure 2. The radian frequency <rr and rate of growth <7,- as functions of к in dimensionless units,
when h\ = 0.5, F — 1.5.

F ig u r e 3. As in figure 2, when hi = 0.7, F = 1.5.

by (oT/ к and the rate of growth by co(, where cor and со,- are the real and imaginary
parts of со. It is appropriate to plot cor and ct>,- as functions of к for given values of
the Froude number F = U / ( g H ) i/2 and the depth ratio H \ / H .
For convenience in presentation we introduce the dimensionless wavenumbers and
frequency
к = kH, (a r + <7j) = (ci>r + cOi)H/U (4-5)
and also the depth ratios
h , = Я ,/Н , h2 = H2/ H (4-6)
so that h i + h 2 = 2 .
A typical example, when hi = 0.5 and F = 1.5, is shown in figure 2. For large values
of к there are four real roots Z yielding four separate branches of <rr (figure 2d).
 2189

Instabilities o f a horizontal shear flow with a free surface 153

F ig u r e 4. A s in figure 2, when hi = 0.3, F = 0.5.

R g u r e 5. Enlargement of figure 4, showing the ‘bubble’ of instability near к = 4.73.

However between Q and R two of the roots become a conjugate complex pair, and
the segment QR is doubled; this is indicated by a ‘parallel’ mark. Similarly between 0
and P another branch is doubled. In each of the intervals OP and Q R , is non-zero
(figure 2b).
Figure 3 shows a second example, hy — 0.7, F = 1.5. In this case the two points P
and Q have coalesced into PQ. The double branch is now continuous through P Q \
the single branch crosses it with no exchange of stability.
A third example, F — 0.5, hi = 0.3 is shown in figure 4. This case is similar to figure
2, except that Q and R have moved very close together. A blow-up of the region QR
(figure 5) shows that there is a small ‘bubble’ of instability between Q and K.
To obtain a synoptic view, we have plotted in figure 6 (a) the wavenumbers к
corresponding to the points P , Q, R, that is to say the points of marginal stability, for
2190

M. S. Longuet-Higgins

FIgure 6. Synoptic stability diagrams: full curves, loci of marginal stability; crosses, points of
maximum growth rate at constant Froude number F. (a) fi, = 0.3, (b) hi = 0.5, (c) /ii = 0.7, (d)
h\ = 0.9.

values of the Froude number F (on the vertical axis). It will be seen that there are
two distinct areas of stability and two areas of instability. Similarly in Figure 6 (b),
when h\ = 0.5. However in figure 6 (c), when h\ = 0 .7 , the two areas of instability
have joined in the middle, creating altogether three separate areas of stability. This
situation is accentuated in figure 6 (d), when h\ = 0.9. Now two of the stable areas have
moved far to the right, and only a very thin wedge of stability at small wavenumbers
remains.
The pinch-off point between figures 6{b) and 6 (c), when the two unstable areas just
meet, can be determined numerically (see the Appendix) as hx = 0.6599 (i.e. H J H 2 =
0.4924), and F = 1.5361. The dispersion relation for this particular configuration is
shown in figure 7. Three modes coincide at the wavenumber к = 1.1091.
In each of figures 6 (a) to 6(d) the wavenumbers corresponding to maximum rates
 2191

Instabilities o f a horizontal shear flow with a free surface 155

R g u r e 7. crr and <7, as functions of к for the triple-point mode at h\ = 0.6599, F = 1.5361.

1
tT

Branch I

F i g u r e 8. (a) The maximum growth rate a/ as a function of F, at various values of hi.

(b) The corresponding wavenumber /c(a,).

of growth (<T/)raax have been marked by cross plots. In figures 6 (c) and 6(d) the circular
plots along the arc Л В correspond to minimum growth rates \<rt\.
Of special interest are the wavelengths (2n /k) of the fastest-growing modes. The
wavenumbers к are shown in figure 8 (a) as functions of F, for depth ratios h{ ranging
from 0.1 to 0.7. Generally, the wavenumbers increase with increasing hu but decrease
with increasing F, both for Branch I and for Branch II.
The corresponding rates of growth are shown in figure 8 (b). Again, the rates of
growth generally increase with increasing depth ratio, that is, with decreasing thickness
of the vortical layer. The growth rates on Branches I and II are of the same order of
magnitude.
2192

156 M. S. Longuet-Higgins
5. Limiting cases
As hi -*> 0 so that the upper layer becomes infinitesimally thin, it is found that the
branch point P in figure 2 moves leftwards towards the origin O. In the limit, the
left-hand curve in figure 6 (a) coincides with the vertical axis к = 0. Setting X\ — 1 we
find that equation (4.3) becomes
Z (Z 2 - 1)(Z + q) + (1 - Л2)а Z (Z 2 + 1) + 2a [q + (X2 - l)a ]Z 2 = 0 (5.1)
which has a zero root Z . Removing this root we are left with a cubic equation which
reduces to
Z 3 + (m + 2<j)Z2 - ( l + 2 a m )Z + m = 0, (5.2)
where
m = a (1 —A2) — q. (5*3)
This is the same equation as found by Stern & Adam (1973).
Suppose on the other hand that Iii tends to 1, so that (H 2 — H Y) / H -*• 0 and the
shear layer becomes very thin, while remaining at a finite distance H below the free
surface. Then we have
Ai —A2 ~ 2 k (H2f c—
-H iU
(5.4)
Q = U / ( H 2 -Я 0 J
and hence а = Q /2kco becomes large while

(Ai —A2)a -> — = q. (5.5)

Co
Then equation ( 4 .3 ) reduces to
A,(Z 2 + 1) + (Z 2 - 1) = 0 (5.6)
so
Z 2 = 7 —ф- = tanhfcHi (5-7)
1 + Ai
and
a = k U ± (gk tanh kH i)1/2. (5*8)

Thus о has two finite values, which correspond to waves travelling at a speed
±[(g/& )tanhfcHi] 1/2 relative to the surface current U. They are similar to gravity
waves in the upper layer, propagated as though this shear layer were a rigid boundary.
The remaining roots of equation (4.3 ) tend to infinity as (Я 2 — H \ ) / H —►0 for
finite k. They represent instabilities of the shear layer itself. As shown in Batchelor
(1967, Section 7.1), for example, a shear layer of thickness Ah in an unbounded fluid
is unstable to all modes with wavenumber к of order less than (Ah)~l . In the presence
of a free surface, however, the modes can be significantly modified.
Thirdly let us consider the case of low Froude number F. This is the ‘rigid-lid’
approximation, when the free surface is assumed to be planar. Then q and a are both
of order F -1/2. Equation (4.3) has two roots
Z ~±1, c~ U ± c0 (5-9)
which correspond to surface gravity waves travelling in either direction with high
speed c0 relative to the surface current U. The remaining roots are approximately
 2193

Instabilities o f a horizontal shear flow with a free surface 157

F i g u r e 9. The velocity profile given by equation (1.1) (continuous curve), and its approximation by
the model profile (2.1).

solutions of the quadratic equation

Z (Z + q) + {Xi - X2)aZ + a [q - (1 —X2/X l)oi](Xl — 1) = 0 (5.10)
which is equivalent to
(c - U)c + (Я, - X2)(Q /2k)(c - U ) + (Q /2 k )[U - (1 - X2/X {)[Q /2 Щ А , - 1). (5.11)
In the further limit /c —►0 (waves long compared to the depths of the two layers) we
have
(A, - X2)(Q /2k) = (H2 - H {)C2 = U (5.12)
and so
(c-U )(c+ U ) = 0 (5.13)
or с = ± U , to lowest order.

6. Comparisons with continuous profiles.

The profile given by equation (1.1), which was adopted by Dimas & Triantyfallou
(1994) to fit experimental data of the shear flow in the wake of a hydrofoil is shown
in figure 9 (continuous curve). The constant b in equation (1.1) is chosen so that at
the mid-point where и = \ V the depth —у is equal to unity. This implies that

e* = V 2 - l (6 . 1)
when у — —1, and hence
b = 0.8814. (6.2)
To approximate this profile we may draw a tangent to the curve at the mid-point
(u,y) = { \ U , —1) and use this to represent a layer of constant shear fl. The velocity
2194

158 M. S. Longuet-Higgins

R g u r e 10. The growth rate or, as a function o f wavenumber for the model profile in figure 9.

gradient at the mid-point is given by

~ — —2b U tanh by sech2by = ( b / ^ 2 )1 ) (6.3)

and hence
- h x = 1 - ( b j . 2) ' 1 = 0.1977. (6.4)
In figure 10 we show the growth rates cr, calculated as in §5, when h{ takes the value
(6.4), and for Froude numbers F = 0.5, 1.5, 2.5 and 5.5. The two families of curves
are qualitatively similar to those calculated numerically for the velocity profile ( 1. 1)
by Dimas & Triantyfallou (1994), and shown in their figure 9. Apparently they were
unable to present accurate results when crf < 0.01. As noted by Morland et al. (1991)
for the simpler case of the wind-induced drift current, the piecewise linear solution
has narrower bands of unstable modes and somewhat higher growth rates.
As a second application consider the velocity profile
u = U [ I - t & n h (by2/ b 2)] (6-5)
representing the shearing current in the second wave trough behind a steady breaking
wave (Duncan & Dimas 1996). Following the same procedure as before we find that
the mid-point of the current occurs when
t byX/Jl = 7 3 , y = d (6 .6 )
and hence
b = 0.5493. (6.7)
Fitting the shear layer by a tangent at the mid-point of the velocity profile we find in
a similar way that
hi = 1 —(3b)"1 =0.3132. (6.8)
F i g u r e 11. (From Coakley & Duncan 1996). Observed fluid velocities relative to a 15 cm hydrofoil
towed at a speed of 69 cm s-1 to the left; average of ten runs.

The Froude number F — U / { g H ) l/2 may be determined from their figure 16, in which
U = 14.8 cm s-1 and H = 3.9 cm, giving F = 0.239. The dispersion relation, which is
similar to figure 4 above, shows there is one maximum growth rate at a dimensionless
wavenumber K\ — 0.338 and a very small bubble of instability at k 2 = 18.3. These
correspond to dimensional wavelengths 2n /k\ = 72.5 cm and 2п /к2 = 1.34 cm
compared with the two wavelengths of 36 cm and 1.4 cm calculated by Duncan and
Dimas (1996)|. The present calculations indicate that the rate of growth of the second
instability was exceedingly small.

7. A p p licatio n to a steady, spilling breaker

Thirdly we may apply our model to a set of laboratory measurements of the velocity
field in a wave induced by a towed hydrofoil (Coakley & Duncan 1996). Figure 11,
reproduced from their paper, shows an averaged velocity field from a 15 cm foil towed
at a speed of 69 cm s_I, in a reference frame travelling with the foil. This indicates
that the fluid at the surface has a low velocity compared to the flow beneath. By
fitting selected profiles at five equally spaced profiles, as in table 1, we can deduce the
parameters h\ and F for each profile, as shown in the table. Then from figure 8 above
we find the wavenumbers k { and k2, and the corresponding wavelengths Li and L2.
A complete comparison with the observed surface profile in figure 11 would have
to take account of energy fluxes, including the nonlinear interaction of the waves
with the mean flow. At present we simply note that the longer wavelengths L\ range
from 11 to 22 cm, which is comparable to that of the more prominent fluctuations
in the observed surface elevation. The shorter wavelengths L 2 lie between 1.6 and
4.0 cm. These are subject to capillarity, which can be taken into account as in §7
below. Additional data, including an instantaneous time sequence of the free surface,
would enable the comparison to be pursued further.

f It may be noted that the authors adopt a different definition of F from that given in §4 above
or in Triantyfallou & Dimas (1989).
160 M. S. Longuet-Higgins

X H, H2 U H L\ l2
(mm) (cm) (cm) (cm s-1) (cm) hi F Ki (cm) *2 (cm)
20 0.20 0.42 66 0.31 0.64 3.8 0.13 15 1.24 1.6 (1.6)
38 0.25 0.79 58 0.52 0.48 2.6 0.29 11 1.07 3.1 (3.0)
56 0.27 0.85 49 0.56 0.48 2.1 0.27 13 1.16 3.0 (2.9)
74 0.27 1.17 45 0.72 0.37 1.7 0.24 19 1.21 3.8 (3.5)
92 0.27 1.32 45 0.79 0.34 1.6 0.23 22 1.25 4.0 (3.8)
T a b le 1. Fitted parameters for the current profiles in figure 11, and the predicted dominant
wavelengths

8. The inclusion of capillarity

The foregoing analysis can easily be modified to include the effect of capillarity.
For, it will be seen that the solution of the dispersion relation (4.3) is formally
unaltered, for any given wavenumber к , if only the gravity-wave velocity Co = (g / k ) l/1
is replaced by the full capillary-gravity velocity Co = ( g /k 4- T7c)1/2. In other words g
has only to be replaced by
g' = g + T k 2. (8.1)
Equivalently the Froude number F = C//(gH ) 1/2 is replaced by t//( g 'H )1/2. The
corresponding dispersion relations, and the synoptic diagrams, can be constructed
as in figures 2-7, but these will not be shown here. Each depends on the new
dimensionless param eter T /g H 2, which in the present paper has been set to zero.
One example of a dispersion relation in which this parameter is not zero, but on the
other hand hi is zero, is shown in figure 3 of Longuet-Higgins (1994).
Proceeding in this way we have recalculated the wavenumbers and wavelengths
shown in table 1. The changes to k\ and Li are negligble. The modified values of L i
are shown in parentheses.

9. Discussion and conclusions

Extending the single-shear-layer model of Stern & Adam (1973) by the addition
of an extra surface layer, we have been able to reproduce the stability properties of
typical current profiles in the wake of hydrofoils or behind breaking waves, and to
simplify their discussion. The previous classification of unstable normal modes into
‘Branch Г and ‘Branch II’ has been confirmed, but with the restriction that over some
ranges o f Froude number and depth ratio the two branches are found to merge.
A reasonable agreement has been found with the calculations in Duncan & Dimas
(1996), and also with the observations by Coakley & Duncan (1997).
It is clear that a simplified model such as that used here can have several advantages
over an elaborate numerical scheme. The calculations can be carried out much faster
and with greater completeness and accuracy, so that a synoptic view of the solutions
can quickly be achieved. At the same time, the slight loss in precision resulting from
the initial fitting of the model to the observed current profile is probably unim portant
for most purposes.
We have discussed here only the initial development of the instabilities, using linear
theory, it being generally assumed that the subsequent flows will be dominated by the
most unstable modes. A discussion of the nonlinear stages of development is outside
 2197

Instabilities o f a horizontal shear flow with a free surface 161

1.56 I I i i | i i i i | < i i i j i i l l

x
X

1.54
Ft")

1.52
X
X

j---1---1---1---1---1---1---1---1---1---1---1---1---1---1---1---1---1---L
0.6599 0.66
A.
F i g u r e 12. Plot of F(l) and F(2) as functions o f hi, close to the critical point.

the scope of this paper, though a nonlinear theory might well be based on the present
simplified model.
In this paper we have considered only two-dimensional instabilities. The three-
dimensional instabilities of a surface shear layer are likely to be of considerable
interest under typical oceanic conditions. Such instabilities could well be investigated
by means of a similar simplified approach.

I am indebted to Professor J. H. Duncan for correspondence and for supplying a

preprint of Coakley & Duncan (1997). This work has been supported by the Office
of Naval Research under Contract N00014-94-1-0008.

Appendix. Determination of the critical values of h\ and F

We take advantage of two properties of the limiting case: (a) Z is real, and (b) Z is
a triple root of the quartic equation (4.3). Thus if the polynomial expansion of (4.3)
is
/( Z ) s Z 4 + a ,Z 3 + a2Z 2 + a3Z + a 4 = 0 (A 1)
then we have also
f \ Z ) = 4Z 3 + 3a,Z 2 + 2a2Z + a3 = 0 (A 2)
and
f " ( Z ) = 12Z 2 + 6 a,Z + 2a 2 = 0. (A 3)
From (A 3) we have at once
Z = —[3a, + (9a? - 24a2)1/2]/12 (A 4)
and on substituting into (A 1) and (A 2) we find two simultaneous relations:
<t>\(F,hUK) = 0, <t>2(F ,h U K) = 0. (A 5)
2198

162 M. S. Longuet-Higgins
For any given value of hi slightly exceeding the critical value, one can solve equations
(A 5) for F and к by Newton’s method of successive approximation. There are two
solutions (F (1),k (I)) and (F(2),k (2)), in general. Close to the critical point, (F (n),/ii) lies
closely on a parabola; see figure 12. The critical values of hi and F correspond to the
vertex of the parabola, and similarly for к(п) and h i.

REFERENCES
B a t c h e l o r , G. K. 1967 An Introduction to Fluid Dynamics. Cambridge University Press, 615 pp.
C a p o n i, E. A., Y u e n , H. С., M i l i n a z z o , F. A. & S ajffm an, P. G. 1991 Water-wave instability induced
by a drift layer. J. Fluid Mech. 222, 207-213.
C o a k le y , D . B. & D u n c a n , J. H. 1997 The flow field in steady breaking waves. Proc. 21st Symp.
on Naval Hydrodynamics, Trondheim, Norway, 24-28 June 1996, pp. 534-549. Washington, DC,
Natl Acad. Press.
D im as, A. A. & T r i a n t y f a l l o u , G. S. 1994 Nonlinear interaction of shear flow with a free surface.
J. Fluid Mech. 2 6 0 , 211-246.
D u n c a n , J. H. & D im a s, A. A. 1996 Ripples generated by steady breaking waves. J. Fluid Mech.
329, 309-339.
K a w a i, S. 1977 On the generation o f wind waves relating to the shear flow in water. A preliminary
study. Sci. Rep. Tohoku Univ. 5, Geophys. 24, 1-17.
L o n g u e t - H i g g i n s , M. S. 1994 Shear instability in spilling breakers. Proc. R. Soc. Lond. A 4 4 6 ,
399-409.
M i l i n a z z o , F. A. & S a f fm a n , P. G. 1990 Effect of a surface shear layer on gravity and gravity-
capillary waves o f permanent form. J. Fluid Mech 2 1 6 , 93-101.
M o r l a n d , L. С., S a f f m a n , P. G. & Y u e n , H. C . 1991 Waves generated by shear layer instabilities.
Proc. R. Soc. Lond. A 433, 441-450.
P e r e g r i n e , D. H. 1974 Surface shear waves. J. Hydraul. Div. ASCE 1 0 0 , 1215-1227.
S h r i r a , V. I. 1993 Surface waves on shear currents: solution o f the boundary-layer problem. J.
Fluid Mech. 252, 565-584.
S t e r n , М. E. & A d a m , Y. A . 1973 Capillary waves generated by a shear current in water. Mem.
Soc. R. Liege 6, 179-185.
T r i a n t y f a l l o u , G. S. & D im as, A. A. 1989 Interaction o f two-dimensional separated flows with a
free surface at low Froude numbers. Phys. Fluids A 1, 1813-1821.
T \ j r n b u l l , H. W. 1 9 5 7 Theory o f Equations, 5 th ed n . E d in b u r g h : O liv e r an d B o y d , 166 pp .
V o r o n o v ic h , A. G., L o b a n o v , E. D. & R y b a k , S. A. 1 9 8 0 O n th e s ta b ility o f g r a v ita tio n a l-c a p illa r y
w a v e s in th e p r e se n c e o f a v e rtic a lly n o n u n ifo r m cu rren t. Izv. Atmos. Ocean Phys. 16, 2 2 0 - 2 2 2 .
W i n a n t , C. D. & B r o w a n d , F. K. 1 97 4 V o rtex p a ir in g : th e m e c h a n is m o f tu r b u le n t m ix in g -la y e r
g r o w th a t m o d e r a te R e y n o ld s n u m b er. J. Fluid Mech. 6 3 , 2 3 7 - 2 5 5 .
 2199

Introductory Notes for Part N

N. Standing Waves

P a p e rs N1 to N 8

AM Balk introduced a simple system o f equations for the motion o f a fluid

with a free surface under the action o f gravity. In this system, both the kinetic
energy T and the potential energy V are simple polynomial expressions o f
the coordinates, o f m axim um degree 4; hence the equations o f motion can be
derived easily from the Lagrangian (T -V ). In Paper N1 the case o f standing
w aves is examined, when all the coefficients are real functions o f the time.
Successive approxim ations to a standing wave are examined, using 1, 2, or 3
Fourier coefficients. When either 1 or 2 coefficients are non-zero there do exist
periodic solutions, but when three coefficients are involved the motion becomes
chaotic.
Papers N2 to N 6 were m otivated by the vertical jets that are conspicuously
thrown upwards when a steep, progressive wave meets a vertical wall. These
jets, analogous to the vertical jets thrown up by standing waves in a circular
vessel (see Paper L25) can achieve very high accelerations and correspondingly
great pressure pulses. In this series o f papers the developm ent o f the jet is first
studied by exploiting the Lagrangian system o f equations developed in Paper N1,
with various sim ple initial conditions. In Paper N2 the initial surface profile is
taken as a pair o f circular arcs (representing the wave crest and the wave trough
respectively). M axim um vertical accelerations exceeding 10 g are found. In Paper
N5 the calculations are carried to greater accuracy by applying the numerical
tim e-stepping technique proposed by Longuet-Higgins and Cokelet (1976; see
Paper L10) in the modified form due to Vinje and Brevig (1981). Starting with
the sam e initial condition as before, the surface profile develops, in general, into
a back-to-back pair o f plunging breakers. By starting with a slightly modified
wave trough, vertical acceleration exceeding 40 g are obtained. The form o f the
upwards je t is also studied and in Paper N 6 is com pared with a sim ple asymptotic
(tim e-dependent) profile.
Paper N3, with D.G. Dom m ermuth, points out that if the upwards je t from a
standing wave falls back into the wave trough, the surface can close around the
jet, causing the wave to “break” in a novel way. Paper N4 sum m arises, with a
discussion, previous papers in this series.
Paper N7, with D.A. Drazen, is an experim ental paper which studies what
happens when a steep, progressive wave meets a vertical barrier. O ver a certain
range o f wave steepness, the occurrence o f a “triple instability” is observed;
2200

every third w ave crest is flattened, the next is sharp-pointed and the third is
interm ediate. T he rate o f grow th o f the instability is m easured, and is found to be
practically independent o f the initial w ave steepness.
N. Standing Waves
 2203

J. Fluid Mech. (2000), vol. 423, pp. 275-291. Printed in the United Kingdom 275
© 2000 Cambridge University Press

Theory o f water waves derived from a

Lagrangian. Part 1. Standing waves
By M I C H A E L S. L O N G U E T - H I G G I N S
Institute for Nonlinear Science, University of California, San Diego,
La Jolla, CA 92093-0402, USA

(Received 18 April 2000)

A new system of equations for calculating time-dependent motions of deep-water

gravity waves (Balk 1996) is here developed analytically and set in a form suitable
for practical applications. The method is fully nonlinear, and has the advantage of
essential simplicity. Both the potential and the kinetic energy involve polynomial
expressions of low degree in the Fourier coefficients Yn{t). This leads to equations of
motion of correspondingly low degree. Moreover the constants in the equations are
very simple. In this paper the equations of motion are specialized to standing waves,
where the coefficients Y„ are all real. Truncation of the series at low values of |n|, say
n < N, leads to ‘partial waves’ with solutions apparently periodic in the time t. For
physical applications N must however be large. The method will be applied to the
breaking of standing waves by the forming of sharp corners at the crests, and the
generation of vertical jets rising from the wave troughs.

1. In tro d u c tio n
The equations of motion for the irrotational motion of an ideal, incompressible fluid
with a free surface are commonly derived from a Hamiltonian expression, in which the
surface displacement and the velocity potential at the surface are taken as canonical
variables (Zakharov 1968). The evaluation of the potential function generally requires
a Hilbert transformation, and especially in high-order expansions in powers of a
parameter the coefficients in these expansions become extremely complicated; see for
example Glozman, Agnon & Stiassnie (1993). A very much simpler and more natural
system of equations has been introduced by Balk (1996), based on a Lagrangian. In
this system not only are the kinetic and potential energies polynomial expressions of
finite degree « 4 ) in the independent coordinates (as opposed to infinite order in
the case of a Hamiltonian system) but the coefficients in these expressions are mainly
low integers. This makes the equations of motion relatively simple to program for
numerical computation, and facilitates the discussion of various approximations.
The Lagrangian scheme introduced by Balk for any general, time-dependent defor­
mation of the free surface (including overturning waves) is in fact a generalization
of the cubic system of equations for steady, progressive Stokes waves discovered by
the present author (Longuet-Higgins 1978). These equations also were found to be
derivable from a low-order polynomial Lagrangian (see Longuet-Higgins 1985).
However, the comparative simplicity of the general Lagrangian system has so far
remained unexploited. The purpose of the present sequence of papers is to show how
Balk’s (1996) analysis can be developed, and to examine some of its applications. We
begin in the present paper with the case of standing waves, periodic in space but not
2204

276 M. S. Longuet-Higgins
necessarily in time. General equations are derived in §§ 2 to 4. An im portant part in the
analysis is played by the determ inant Л of the equations of motion, which must not
vanish in order for time-stepping to proceed. It is shown in § 5 that this determinant
factorizes. Some insights can be gained by considering low-order approximations (i.e.
truncations o f the Fourier series) as is shown in §§ 6 to 8 . High-order approximation
will be used in later papers for a discussion of two phenomena displayed by standing
waves, namely breaking at the wave crests and the forming of strong vertical jets
(‘flip-through’) rising out o f the wave trough.

2. The Lagrangian
It is shown by Balk (1996, § 2) that in any two-dimensional irrotational wave motion,
periodic in the (horizontal) x-direction with period 2л, the coordinates (X , Y ) of a
point on the free surface may be expressed in the form
oo
X + i Y = « + ^ ( X n + i r „ ) e - in{, (2.1)
—00
where X_n = X* and Y_„ = Y*, a star denoting the complex conjugate; also

X n = i c nYn, (2-2)
where a n equals + 1 , - 1 or 0 as n is positive, negative or zero respectively. From (2.1)
it follows that

00
—CO (2.3)
У = £ у „ е - “ <,

so that the Yn(t ) may be taken as the time-dependent coordinates of the fluid. These
are subject to the constraint that the mean level Y, given by

Y = ^ f Y d X = У° + Y , no* YmY„, (2.4)

® т+л=0

shall vanish. Thus we have always

Уо = - £ |п|У„У_. (2-5)
-0 0

The Lagrangian i f is defined as

<£ = Г - V, (2.6)
where V and T are the mean potential and kinetic energy densities per unit horizontal
distance. As in (2.4), V is found directly from (2.3):

2V= У2 d X = Y Y iY m+ Y , n a „ Y , Y m Y„, (2-7)

^ l+m=0 l+m+n=0
all the summations running from —oo to oo.
In a similar way, the velocity potential 4>{£,t) and the streamfunction at the
 2205

Theory o f water waves derived from a Lagrangian. Part 1 277

free surface can be expressed in Fourier series analogous to (2.3):

i< r „ B n e
—
OO ( 2.8)
00

—00
Then the kinetic energy density T is given by
l г2к 00
2T = 2 iij0 = H N B»B— (2.9)

The crucial step taken by Balk (1996) was to show that the kinematic boundary
condition at the free surface can be expressed as
y>( = X t Y( - Y tX ( , (2.10)
which allows the coefficients Bn to be related directly to the Y„ by

inB„ = Y„ + ^ 2 ( a , - < 7 m) l Y , t m ( п ф 0); (2.11)

l+m=n
a dot denotes d /d t. We can always choose
Bo « 0 . (2 . 12 )
It is convenient to write
2\n\Y„
/ 2! пф 0
(2.13)
a- = { i! /1 = 0 ,

so that when, for example, n < 0 (2.11) gives

i nB„ = (...# 2 ^n—2 + a\Yn-\ + aoY„) + \ a nYo + {on- i Y \ + a„_2 Y2 H-----)• (2.14)
From (2.11) we have also
Yo = —(•.. a$Y —4 + a}Y-3 + a{Y -2 + a \ Y - \ ) — (Д- i Yi + 0 - 2*2 H-----)• (2.15)

3. S ta n d in g waves
So far the equations are quite general. However, in standing waves the coordinates
Y„(t) may be chosen to be all real. The above formulae are then simplified and, from
equations (2.8) to (2.13), we find
4 Г = {[{(1 4* Я2 —flifli)<*i + \i.a \ + —0102)02 + |(o 2 + 04 — 0103)03 H-----]2

+^[{( —o2Oi)ai + 5(1 + o4 —a2a2)a2 + 5(01 + а 5 — о2оз)6 з H---- ]2

+ |[{ ( - 0301)01 + 5 ( o5 —0302)62 + |(1 + Об —0303)63 H---- ]2

+ (ЗЛ)
or more compactly
00/00 \ 2
4 T = ^ 2 ( X / ^ mn йи J *
m = 1 \n = l /
2206

278 M. S. Longuet-Higgins
where
(3-3)
provided that a„_m is replaced by 0 when n < m. Writing equation (3.2) in the form

(3.4)

and reversing the order of the summations we obtain

00 00
(3.5)

where
1 00 (3.6)
Qki = 2 ^ ,nk
m=l
In other words, the m atrix of the quadratic form (3.5) is the column-by-column
product of the matrices (Р„*) and (Pm/). Note that P„* is of maximum degree 2 in
the coefficients an. Hence T is of degree 2 in and of maximum degree 4
in fli, Й2, . . . .
Consider now the degree o f the terms in the potential energy V , given by equation
(2.7). The right-hand side o f (2.7) still contains the dependent variable Yo- The first
sum m ation contains term Y02, while the second contains some terms with no more
than one o f J, m, n equal to 0 (for, if any two of /, m, n are zero, then so are all three,
making |/| vanish). Since by (2.5) Yo is quadratic in the remaining Yn, altogether V
has maximum degree 4 in Yi, Y2, —

4. The equations of motion

As independent coordinates we may take the coefficients а ь ^ г ,..- defined by
equation (2.13). Lagrange’s equations of motion are then

(4.1)

which can be written

A (4.2)
dt \ d a n) dan dan'
Consider first the terms involving T. From equation (3.5) we have

(4.3)

and therefore
(4.4)

where
(4.5)
 2207

Theory o f water waves derived from a Lagrangian. Part 1 279

Also on the left of (4.2) we have from equation (3.5)
дТ 1 г—\ v—\ dQid
я ;-5 ? ? -£ * •* ' ««>
After combining the two quadratic forms (4.5) and (4.6) and writing the latter more
symmetrically, the left-hand side of equation (4.2) reduces to

^ 2 Qm - X) X T ^ (4-7)
/ / к
where

<«>
Note that when к = n or / = n, two of the terms in this expression cancel.
In order to calculate d Q u /d a n we have from (3.6)

<«>
while from (3.3)

^ = 0
dan
+1 if n = I- m
+1 if n = I+ m
—am if n= I
—ai if n = m. (4.10)
Lastly for the terms involving V in equations (4.2) and (4.7), note that
d v _ * d v d Y k
8a„ £ д у к да„ ’ (4Л1)
к ——oo

where
-i- if k = ± n
д П = < 2|n| (4.12)
да„ - Y k if * = 0
0 otherwise;
also that
1!

HI YmY„
5 (|/| Y,YmY„) = < |/|
I/I Y,Y„ if
i ( kк = m (4.13)
dYk
*-4
II

I/I Y,Ym
a

To summarize, Lagrange’s equations of motion (4.1) can be expressed in the form

£ <2-< */ = E E S^ k’ № й' + U” (4.14)

/ I к

where Q„, is of maximum degree 4 in the coordinates ab a2,---> while S„ and U„ =

—d V / d a n are each of maximum degree 3.
2208

280 M. S. Longuet-Higgins
5. Solution of the equations
In order to solve equations (4.14) we may suppose an(t) = 0 for all n greater than
N, say, and proceed by successive approximations as N is increased indefinitely. Given
some starting values for an and a„ at an initial time t = 0, equations (4.13) comprise
a set of simultaneous equations for determining йь й2,...й ^ . These can be solved
provided that the determ inant
Д * -К М (5-«
does not vanish. Some interest therefore attaches to the conditions under which
AN = 0 . (5.2)

Now from equation (3.6) we have

Д* = 2 n (5.3)

so that the vanishing of A N depends solely on the vanishing of ||Py| Moreover from
equation (3.3) we have
(5.4)
= (N \)V 2^ n’
where DN is the (N x N ) determ inant
(1 + a 2 — a ia i) (ax + a3 — a {a2) (a2 + (З4 — a\a$)

( a3 - a 2a {) (1 + 0 4 - 02 ^ 2) (a\ + as — a2a$)
Dn = (5.5)
( ал - а га х) ( а $ - а ъа2) (1 + a 6 - a 3a3)

Consider now the (N + 1) x (N + 1) determinant

1 a\ a2 a3
a\ ( 1 + Я 2) ( a\ + аз) (a2 + a4)

En = a2 a3 (l+ « 4 ) ( a\ + as) (5.6)

аз a4 as (1 + Дб)

On subtracting from the zth column of the determinant a{ times the first column we
reduce each element of the first row to 0, except for the first element which is 1. It
follows that
Dn = En . (5-7)
Now in E n the sum of the elements of each column is
v-^\(+) (5.8)
1 + a x + a2 H-------1- a^ = ^ ,

say. Each element in the first row may therefore be replaced by Hence 5D a
factor of the determinant E n . N o w multiplying the second row by 2, adding to this
2 x each of the other even rows and subtracting from this the first row, we see that
 2209

Theory o f water waves derived from a Lagrangian. Part 1 281

all the elements of the second row become equal to + ) where

E (~) = 1 - a x + a2 - a3 + a4 ------. (5 .9 )

Hence £ (-) is also a factor of EN. After extraction of these two factors the remaining
determinant is
1 1 1 1

-1 1 -1 1

a2 a3 0+Я 4) (<*1 + as)

ЯЗ a4 as ( 1 +*б)

Altogether then, from equations (5.3), (5.4) and (5.10) we have

‘■ ■ а д Е ’ Г Ч ' p")
As will be seen below, the vanishing of £ (+) or is associated with the formation
of a cusp at the free surface.
If TN and VN denote the truncated forms of the kinetic and potential energies, and
if = TN — VN we note that an exact integral of the equations

m v s t’
is
HN = TN + VN = constant. (5.13)
Equation (5.13) may be derived on multiplying each side of (5.12) by an and then
summing with respect to n from 1 to N. For, if / is any function of a i , . . . a N and
formally independent of the time t, we have identically

S - t( ‘S +8-S)
-I ( £ * ■ ! : ) ,5Л4>
When / = <?n the last group of terms vanishes by (5.12), and since TN is quadratic
in A t,... 6 jv we obtain

Т Г - 1 27'" ' (! , i >

whence on integration (5.13) follows.
In the following two sections we shall explore the two lowest-order approximations
N = 1 and N = 2. Although their validity is strictly limited to small values of a x and
a2, the behaviour of the approximations at larger values of a\ and a2y and the manner
in which the approximations fail, will be seen to be of interest in later work.
2210

282 M. S. Longuet-Higgins

(x-n )

R g u r e 1. Family o f surface profiles given by equations (6.1) (the case N = 1).

6 . T h e case N = 1
In the simplest case we set a2 = a 3 = • • • = 0, so that the free surface profile is
given by
X = t + o tsin £ , | (61)

у = Y0 4 - a cos J
where а = ai and Yo = —5 a2. This is the parametric equation of a cycloid; see figure
1. Physically admissible solutions are limited to |a| < 1. When a = ±1 the profile has
a downward-pointing cusp.
The dynamical equations take into account only the self-interaction of the funda­
mental harmonic. Thus from § 2 we have
4T, = (1 —а 2)2а2, 1
( 6 .2)
Щ = a2- \<x\ J

The single Lagrange equation (5.12) becomes

— [ |( 1 —а 2)2а] = —а (1 —а 2)а 2 — ^а (1 —а2) (6-3)

and after dividing through by (1 —a2) we obtain

(1 —а2)а —2 аа 2 -f а = 0 . (6-4)
The energy integral (5.13) is
(1 - a 2)2* 2 + (a2 - ±a4) = 4 H, (6-5)
whence
4 Я - « 2( 1 - 1 « 2) (66)
a2 =
(1 —a 2)2
 2211

Theory o f water waves derived from a Lagrangian. Part 1 283

l.O

0.8

0.6
t

0.4

0.2

0
-2 - 1 0 1 2
a
F i g u r e 2. Graphs o f a(f) covering half a wave period, for different values of ao = a(0),
assuming that dc(0) = 0.

and
( i - « 2)
da. (6.7)
- f [4Я —a2(l — | a 2) ] 1/2
Equation (6.7) can also be written as

t = ± ^ L [(1 — a 2) 2 — C 2] ‘/2 d a ’ ( 6 8 )

where
C 2 = (1 —(Xq)2 = 1 —8 Я. (6.9)
Figure 2 shows a as a function of the time t for various values of ao = a(0), assuming
that at t — 0 the fluid is at rest. When a 0 is small, a(t) is simply a sine-wave with
period 2k. Only the section of each curve for which a > 0 is shown. In the hypothetical
limiting case ao = 1 we have С — 0 and so equation (6.9) gives
*= ±V 2( l - a ) , (6 . 10 )
that is to say a pair of straight-line segments. The crest of the wave then rises or falls
uniformly when t Ф 0, with an instantaneous, large negative acceleration at t — 0.

7. The case N = 2
On writing ai = a and a2 = P, the free surface is now given by
x = £ + asin<^ + ! 0 s in 2 £ ,
(7.1)
у = Уо + acos^ + {P cos 2 £,

where
Y0 = - W - \ f . (7.2)
2212

284 М . S. Longuet-Higgins

F i g u r e 3. O u te r lim its fo r th e p o in t (a ,/J ) = {au a2) in th e (a ,/? )-p la n e .

Outer limits for a and p are given by the vanishing of A# in equation (5.11). Since in
(5.10)
1 1 1
-1 1 -1 = (!-/* ), (7.3)
0 1
the outer limits are given by
( l + a + /? )(l-« + /? )(l-0 ) = O ( 7- 4 )

which represents the three lines

a + 0 = -l, o c-0 = l and P = 1. (7.5)
These are shown in figure 3, indicated by the symbols aa, bb and cc respectively.
It appears that whenever the point (a,/J) lies on any of the lines (7.5) the corre­
sponding profile (7.1) has a downward-pointing cusp. Some examples, indicated by
the points А, В, C, D and E in figure 3, are shown in figure 4. In addition, there
are limits arising from the need to avoid loops or double points in the profile. These
limits correspond to the curved boundaries in figure 3. Thus (a,/?) is limited to the
interior of the triangle defined by the lines ( 1), (2 ) and (3 ), and excluding the shaded
areas in the two upper corners.
The evolution equations are derived from the Lagrangian S?2 = T 2 — V2 where

4Г2= [(Ц-)5-а2)а+1(а-ар)Й2+ |[^а-1(1-/?2)№. 1

> (7.6)
4К2= (а2+ 1/52) + а2/5_1(а2+ 1/?2)2. J
The equations of motion for a and ft are then
La + M p = - 1 [L„a2 + 2Lf afi + (2Mp - N a)P2} + C u

Mix + N'p = —i [(2Ma - L f )a2 + 2Na* p + Npj}2] + C2,

 2213

Theory o f water waves derived from a Lagrangian. Part 1

R g u r e 4. (a -/) Surface profiles corresponding to the points A to F in figure 3.

where
L = 2 ( 1 + 0 —a2)2 + a 20 2,

M = a(l —m + P- a2) - № ( l - P2), (7.8)

N = № ( 1 - P ) 2 + \ { 1 - P 2)2,
2214

286 M. S. Longuet-Higgins

Trajectory o f (a ,/3 ) when N = 2 and with starting values a = 0 .4 , /? = 0 , a = ft = 0.

F i g u r e 5. (a )
Time interval 0 ^ t < 2 2 .3 . Interval Ar between plots = 0 .1 . (b) As in (a ) but with 0 < t < 4 4 .6 .
(c) Graph o f a(t) for (a), showing ten oscillations. (d) Graph o f /?(r) for (b), showing seven oscillations.

and
Ci = 2a(l + /? —a2) —a/J2, 1 (?9)
C2 = a2(l —Д) — |/?(1 —/52). J
Some examples of the trajectories of (a,/?) are shown in figures 5 to 7. In each case
the motion starts from rest: a(0) = /?(0) = 0, so that the total energy H is just equal
to the initial potential energy K(0 ).
For disturbances of very small initial amplitude one would expect the coefficients a
and /? to oscillate independently, each with its own frequency. Thus a would oscillate
harmonically with frequency 1 and /? with frequency J 2 = 1.414... . The ratio of
the frequencies being irrational, the combined motion would be non-periodic, in the
small-amplitude limit. However, very slight nonlinearity may cause the frequencies to
become compatible, resulting in a periodic orbit. One such example is shown in figure
5(a) where the initial amplitudes are
a = -0.4, p = 0. (7.10)
 2215

Theory o f water waves derived from a Lagrangian. Part I 287

Figure 6. (a) Trajectory of (a,fi) when N = 2 and with starting values a = 0.4, fi = —0.4,
a = P = 0. t = 0.0(0.1)12.7. (b) Graph of a(t) for (a). (с ) Graph of 0(t) for (a).

The crosses mark points on the trajectory separated by time-intervals 0.1 (one hundred
time steps Ar). After time t = 22.3 the point (a,/?) is close to the opposite point (0.4,
0.0) where, because of conservation of energy, the velocities a are very small. At time
t = 44.6, (a, p) has returned practically to its initial position. In the intervening time,
a has executed seven complete oscillations and P has executed ten (see figures 5b and
5c). So the ratio o f ‘frequencies’ is 10 ч -7, i.e. 1.429 approximately.
A simpler example is shown in figure 6 (a), where the starting point is
a = 0.4, jJ = -0.4. (7.11)
At t = 12.8, (a ,P) returns to its starting point, a having executed two oscillations
and p three, the ratio of frequencies being 1.5. Compared to a(f), the form of p{t) is
remarkably sinusoidal.
A third example is shown in figure 7(a). Here the initial amplitudes are
a = 0.5, P =0. (7.12)
The initial energy is sufficiently large that after just over half an oscillation of the
fundamental a(r) the trajectory of (a,/?) runs off the scale. The point (a,/?) accelerates
toward the boundary, as is shown by the curve of a(t) in figure 7(b). The critical
2216

288 M . S. Longuet-Higgins

Trajectory of (a ,/? ) when N = 2 and with starting values a = 0 .5 , = 0 , a = ft = 0.

R g u r e 7. (a )
(b) Graph of a(f) for (a), (c) Logarithmic plot o f a(0 near the critical instant tc = 4.211.

time t at which (a,/?) crosses the boundary is tc = 4.211 approximately. The final
acceleration |a| as t approaches tc is shown in figure 7(c) on a logarithmic scale. The
acceleration varies as \t —tc\x, where Я is a constant. This behaviour is typical o f any
function f ( t ) satisfying an equation of the form
/ / = C f2 (7.13)
when f passes through a zero.
In all of the above numerical integrations the total energy H remained a constant
to within one part in 10 5.

8 . T h e case N = 3
This last special case to be discussed is the simplest for which not all the factors
of A n are linear. Thus from equation (5.11) on writing N = 3 and a i,a 2>fl3 —
respectively we have

Дз = j ^ j ( l + а + P + У)2(1 - л + Р - y)2Fi, (8. 1)

 2217

Theory o f water waves derived from a Lagrangian . Part 1 289

F i g u r e 8. Outer limits for (a,/?) = {ai,ai) in the case N = 3, when аз = 0.5.

where by (5.10)
0 1 0 1

-1 1 -1 1
F< = (8 .2 )
P У 1 a
У 0 0 1
which on expansion gives
( 1 - /? ) + ( « - -y)y (8.3)
Whereas in (8.1)
l+ a + (8.4)
+

0
II

and
l- a + 0 -y = O (8.5)
represent bounding planes in the space of (a, f$, y), the last factor
(1 - Д ) + ( а - у ) у = 0 (8 .6 )
represents a quadric surface (hyperboloid). Nevertheless the intersection of this surface
with any plane у = constant is a straight line in that plane. When у — 0, of course,
equations (8.4), (8.5) and (8 .6 ) reduce to the lines aa, bb and cc of figure 3. In the
more typical case when у = 0.5, these lines appear shifted as in figure 8 . Thus aa
is lowered by an amount y, bb is raised by у and cc is tilted with inclination у and
passes through the point { a j ) = (y,l). We may call the triangle formed by aa, bb
and cc the basic triangle.
The origin (0,0), which corresponds to infinitesimal orbits, always lies in the
interior of the basic triangle so long as |y| < 1, and it can be verified that points (a,/?)
lying on the sides of the basic triangle always correspond to surface profiles having
downwards-pointing cusps.
We show one example of a time-history a (t)yfi{t)>y{t) when the initial conditions
2218

290 M . S. Longuet-Higgins
T ’ V -------------------- i ' i — 1-------- ----------1-------- ---------Г------- ---------1 . J. ’ T — •— 1—
♦♦ ♦
0 .6 ♦ ♦ -
♦ ♦

4 ± ♦ ♦

0 .4 ♦
tl ♦ ♦ ♦
4 ♦ ♦
♦ * j\
V + + + + + + H rt^ 0 .2
♦ - + ii ♦ /= 1 0 ♦ % j
f = 10 + + +
I * + + + + , = 0 Л
♦ J;.........................
\ . r - o
.
'
- ...... * ................................................. 1 r 0
+ ♦ Г **
w !! ♦ / - 0 .2
\ ♦♦ *
♦ I**♦
- 0 .4
*

■ (a) - 0 .6 . 0ь ) V
i . t i i . i 1 1 . I i - » - i - i

N>
04

о1
- 0 .4 - 0 .2 i0 0 .2 0 .4 0 0 .2 0 .4 0 .6

0I

0
a a

t
R g u r e 9 . (a ) Trajectory o f (a , P) when N = 3 and with starting values a = 0 .4 , p = у = 0,
a = p = у = 0 (compare figure 5a). (b ) Trajectory of (a .y ) corresponding to (a), (c) Root-mean-square
values o f a, P and у as functions of the time t.

are th at at time t = 0
a = 0.4, P=y = 0 (8-7)
and а, 'р,у all vanish as before. A plot of p vs. a alone is shown in figure 9(a), and
the corresponding plot of у vs. a in figure 9(b). Comparing figure 9 (a) with figure
6 (a), where the initial conditions were similar except that у was restricted to be 0 , we
see that trajectories of (a,/?) appear very different and the motion is less periodic. An
approximate complete cycle 0 < t < 44.0, is shown in figure 9(c).
It will be noted that the r.m.s. values of a, p and у are all of comparable magnitude.

9. Conclusions and discussion

We have transformed Balk’s (1996) system of equations into a more usable form
and specialized them to the case of standing waves. The general system of equations
(4.14) for the independent coefficients a„(t),n = 1,2,... may be truncated at n = N
and used for time-stepping the motion, provided that the determinent A n of the
system does not vanish, as is true for sufficiently small an\ see equation (5.5).
Low-order representations of standing waves have well-defined zones of existence
in a„-space. The boundaries are mostly straight lines or planes. As these boundaries
 2219

Theory o f water waves derived from a Lagrangian. Part 1 291

are approached, the corresponding surface profiles are found to develop downward-
pointing cusps. In the case N = 2 a choice of simple initial conditions strongly suggests
the existence of periodic ‘waves’ analagous to those corresponding to a truncated
Hamiltonian which were found by Glozman et al. (1993). However, the addition
of a non-zero third harmonic (N = 3) radically affects the periodicity. The motion
then appears to be chaotic, in a way resembling that found in truncated models
of progressive waves. However, as noted by Glozman et al. (1993) the stochastic
properties of truncated models depend strongly on the order of truncation. In order
to study chaos or periodicity in ‘real’ waves it is necessary to take the approximation
to high orders, as was done by Agnon & Glozman (1996) for a Hamiltonian system.
We do not pursue the subject here.
An application of the equations to some interesting aspects of standing-wave
behaviour which require approximations of large order N leading to high numerical
accuracy will be given in a future paper (Longuet-Higgins 2000).

This work has been supported by the US Office of Naval Research under Contract
N00014-94-1-0008.

R EFEREN CES
A g n o n , Y. & G lo z m a n , M. 1996 Periodic solutions for a complex Hamiltonian system: New
standing water waves. Wave motion 24, 139-150.
B a lk , A. M. 1996 A Lagrangian for water waves. Phys. Fluids 8, 416-420.
G lo z m a n , М ., A g n o n , Y. & S tia s s n ie , M . 1993 High-order formulation of the water-wave problem.
Physica D 66, 347-367.
L o n g u e t - H i g g i n s , M . S. 1978 Some new relations between Stokes’s coefficients in the theory of
gravity waves. J. Inst. Maths Applies. 22, 261-273.
L o n g u e t - H i g g i n s , M. S. 1985 Bifurcation in gravity waves. J. Fluid Mech. 151, 457-475.
L o n g u e t - H i g g i n s , M. S. 2000 Vertical jets from standing waves, (submitted).
Z a k h a r o v , V. E. 1968 Stability o f periodic waves on the surface of a deep fluid. J. Appl. Mech.
Tech. Phys. 2, 190-194.
2220

ТНЕ ROYAL
Ш S O C IE T Y

Vertical jets from standing waves

B y M ichael S. L o n g u e t -H iggins

In stitu te f o r N onlinear Science, University of California,

San Diego, La Jolla, С A 92093-0402, USA

Received 6 A pril 2000; revised 16 August 2000; accepted 8 November 2000

W hen a shallow-water wave meets a steep cliff or harbour wall, a jet is sometimes
throw n upwards w ith very high initial acceleration. In the present work it is shown
th a t a similar phenomenon can occur in a standing wave in deep water, provided th at
the initial shape of the trough is narrow and an almost circular arc. During collapse of
the trough, vertical accelerations exceeding 10 g are found, with subsequent vertical
velocities exceeding 1.7c, where с is the phase-speed of a small-amplitude progressive
wave. The vertical acceleration, plotted as a function of the time, is shown to be very
sensitive to the initial shape of the wave trough. The presence of a much shorter wave
superposed on the initial wave profile can further increase the maximum acceleration
and velocity.
K eyw o rd s: w a ter w aves; su rface w aves; je t form ation ; sta n d in g w aves o n w ater;
w ave breaking; a cceleration s

1. Introduction
Among the more spectacular phenomena displayed by ocean waves is the vertical
jet sometimes thrown upwards when a steep wave encounters a steep cliff face or
harbour wall. For laboratory experiments (see Chan & Melville 1988). Although
commonly observed around coastlines, the first theoretical representation of such
jets was not attem pted until Cooker & Peregrine (1990, 1991) used a boundary-
integral technique to follow the development of a shallow-water wave impacting on
a vertical wall. Under certain conditions, they found vertical acceleration of the
free surface at the wall exceeding 250# (where g is the acceleration due to gravity),
and correspondingly high pressures just below the water surface. The phenomenon,
term ed a ‘flip-through 5 by Cooker & Peregrine, was analysed by Longuet-Higgins &
Oguz (1997), who showed th at it was typical of a class of jets occurring in near-
critical situations where two surfaces come together and nearly, but not quite, trap a
pocket of ‘air’. Other examples of such behaviour are known to occur in a x i s y m m e t r i c
standing waves when a jet arises out of a wave trough (Longuet-Higgins 1983) and in
collapsing axisymmetric bubbles and cavities (Longuet-Higgins &; Oguz 1995, 1997).
It is natural then to ask what initial conditions, in the case of two-dimensional sur­
face waves, will lead to a high vertical acceleration or ‘flip-through’. In the absence
of any complete analytical solution we must resort to numerical time-stepping or for­
ward integration. The first problem is how best to characterize the initial conditions.
For the sake of simplicity, it is obviously preferable to find a description involving as
few free parameters as possible, and this will be our aim.

Proc. R. Soc. bond. A (2001) 4 5 7 , 495-510 © 2001 T h e Royal S ociety

495
 2221

496 м . S. Longuet-Higgins

W hat kind of initial profile might be expected to lead to high accelerations? As

a preliminary hypothesis we might suppose that a wave trough in the form of a
circular arc combined with low initial velocities would be promising, on the following
grounds. Because the pressure gradient at the surface is always normal to the surface,
each particle of water at the surface will then be accelerated initially towards the
centre of the circular arc. The convergence of particles towards the centre of the arc,
together with the conservation of mass, must be expected to lead to high pressures
near the centre, as is the case with a collapsing underwater cavity. (In axisymmetric
flow the convergence is more extreme.) Since the pressure at the surface itself remains
constant, this will produce a high-pressure gradient near the free surface, hence a
large vertical acceleration. Note that, in the example shown by Cooker Sz Peregrine
(1990, 1991), the shape of the wave trough is indeed almost circular.
Consider a surface wave that is periodic in the horizontal direction. Perhaps the
simplest profile having a constant trough curvature is one consisting of successive
circular arcs: upwardly concave arcs in the wave troughs and downwardly concave
arcs at the wave crests. Such a ‘piecewise-circular’ profile would not be analytic.
Nevertheless, we shall see that it can, in some cases, be approximated very accurately
by a simple analytic expression.
For the purpose of time-stepping the initial profile, we shall adopt an alternative
to the usual boundary-integral technique. Our method has its origin in a new for­
mulation of the equations of motion for space-periodic surface waves discovered by
Balk (1996) and developed (for standing waves) by Longuet-Higgins (2000). As out­
lined in § 3, the motion at each instant t is given in terms of a series of coefficients
an (£), n = 1 ,2 ,..., and their time derivatives an(t). The equations of motion are
exact, nonlinear and of finite degree (less than or equal to 4) in the an . It turns
out, as shown in § 5 , that the piecewise-circular profile in figure 1 can, in a family
of cases, be represented with remarkable accuracy by using only two non-zero coeffi­
cients, namely ai and a2. Moreover, by time-stepping these initial profiles, it is found
(see § 6 ) th at high vertical accelerations sometimes exceeding 10 # axe subsequently
generated, even in deep water.
Further, it is shown in §7 that even higher accelerations may be generated by
superposing on the initial profile a second, smaller-scale profile, f

2. The piecewise-circular profile

We seek first a simple initial form of wave profile which might be expected a priori
to yield a high vertical acceleration at subsequent times. For reasons given in § 1, a
promising form for such a profile would be one having a constant upwards curvature
in the wave trough.
As a template for such a profile, it is natural to choose a curve consisting of
the upward-facing circular arc, in the wave trough, joined smoothly on each side to
another downward facing arc, as in figure 1 . Here, p\ denotes the radius of curvature
of the wave trough and p 2 is the radius of the two outer arcs. They are assumed to
t A referee has draw n th e a u th o r’s a tte n tio n to a recent p ap er by T uck (1998), in which th e problem
of a tw o-dim ensional flow, initially a t rest b u t w ith a sem icircular cavity a t th e free surface, is solved
num erically by a rough technique. T h e sem icircular arc occupies one-half of th e horizontal distance
betw een two vertical walls. FYom th e intersections of successive surface profiles w ith th e vertical centreline
in his figure 1 (which do n o t behave very sm oothly), we estim a te th e m axim um vertical acceleration in
th is exam ple to be a b o u t 2.7g.

Proc. R. Soc. Lond. A (2001)

2222

Vertical je t s fro m standing waves 497

join smoothly at the points A and A ' at a distance x \ from the line of symmetry
(x = 0). At the joins, the surface slope is continuous and the slope angle is ± a . Thus
the equation of the inner curve can be w ritten parametrically,
x — pi sin в ,
when —a < в < a , (2.1)
У = 2/1 - p ic o s f l

relative to the level of the points A and A! . Here, y \ is the у-coordinate of the
centre C \ of the inner arc. The equations of the two outer arcs are similar. W ith the
wavelength being taken as 27t, it is easily shown that, in this coordinate system, the
mean surface level is

y d x = p\ (2 a —sin 2 a). (2.2)

4тГ

Since later we shall assume the mean water level to be zero, the constant у has to
be subtracted from the expression for у in (2 . 1 ). Thus, altogether, we can write for
the surface profile ys {x),

Vs(x)
_ f(v i-y )~ (Pi - X2 ) 1/ 2 , |i| <X i, (2.3)
\ ( У 2 - y ) + [p\ - (5Г - Z l ) 2] 1 / 2 , X l < | x | < 5Г,

where у i is the vertical coordinate of the centre of curvature C 2 relative to the level
AA '.

3. Equations of motion
The equations th at we shall use are essentially the ‘natural’ system derived by Balk
(1996) and developed by Longuet-Higgins (2000). It will be convenient to restate
them in a shortened form.

Proc. R. Soc. Lond. A (2001)

 2223

498 M. S. Longuet-Higgins

Consider an irrotational standing wave in deep water, periodic in the horizontal

direction with period 2i r. Let axes of x and у be taken in the horizontal and vertical
directions, у upwards. Then the coordinates of a point on the free surface may be
expressed as

x = f + ^ 2 r n sm n 4 ,
n=l
oo (3.1)
V = Y0 + £ > n cosn*.
71=1

where У0, *i, • • • are functions of the time t (see Balk 1996). The coefficients
an = 2nYn(t), n = l,2 ,..., (3.2)
and their rates of change

dn = ^ 5 ti = 1 , 2 , . . . , (3.3)

are then a natural system for defining the motion at any given instant. If the mean
level is assumed to be constant at у = 0, the conservation of mass requires th at Yq
be a function of the time t given by

y 0 = - f > K n2 = - f ^ < 4 (3.4)

It is shown in Longuet-Higgins (2000) that Balk’s equations of motion for the Yn are
equivalent to

^ ^ Qn}iai = ^ ^ ] &п{к} l)akdi + £/n, n = 1, 2, . . . , (3*5)

l l к
where Q nj is a polynomial expression of maximum degree 4 in the coordinates a*,
while S n and Un are each of maximum degree 3.
Given the values of an and an at some initial instant t = 0 , it is clear th at (3.5)
can be used for forward time-stepping of a n.
In practice, the sequence of coefficients an and the system of equations (3.5) гиге
each truncated at some finite value N of n. The finite system is then solved, and
N is increased until the solutions converge numerically to one part in 106. It is the
converged solutions that will be presented.
These equations have the distinct advantage that the coefficients involve only
simple integers. Moreover, a surprisingly low number of terms in (3.1) is needed in
order to define a sequence of initial profiles closely approximating those in § 2 , as we
shall now show.

4. Initial conditions
We first seek a simple approximation to the initial profile (2.3), th at is, one given
by a series of the form (3.1) with as few non-zero coefficients Yn as possible. It turns
out th at this may be done remarkably well by the use of only two such coefficients,
Yi and Y2.

Proc. R. Soc. Lond. A (2001)

2224

Vertical j e t s fro m standing waves 499

Figure 2. (a) A piecewise-circular profile (dotted curve) corresponding to X i/n = 0.20, a = 52.9°
in (2.3), compared with a truncated series expression (3.1) with a i = 2Y\ — 0.7814 and
a2 = 4У2 = 0.2764, shown by the solid curve. (b) Central section enlarged.

Let us initially fix X\, the half-width of the wave trough. Then by choosing any
particular value of a , the form of the profile is determined; in particular, the coordi­
nates ут and y c of the wave trough and the wave crest. Then in (3.1) we can chose
the coefficients Vi and Y2 so as to make у — ут when £ = 0 and у — y c when £ = тт.
Then, to determine a closest-fitting profile, we may adjust the angle a , keeping x i
constant, so as to minimize the rectified difference

#= [ \ y - y B\d x . C4-1)
J-x i
For example, when x i / n = 0.25 (so the wave trough is one-fifth of the total wave­
length), we obtain the profile shown by the full line in figure 2a. This is compared
with the piecewise-circular profile shown by the dotted line.
Figure 26 is an enlargement of the central segment of figure 2a, showing th a t the
agreement between the original piecewise-circular profile and the truncated-series
expression is indeed close. This may be attributed in part to the fact th a t the first
term in the series (3 . 1 ) by itself represents a profile that, like the solid curve, has a
sharper curvature in the trough than at the crest of the wave.

Proc. R. Soc. Lond. A (2001)

 2225

500 м . S. Longuet-Higgins

Table 1. Parameters of the piecewise-circular profiles in figure 3 and their approximate

representations by the truncated series (3.1)

Xi / tt a a i = 2 Yi a 2 = 4V2

0.25 47.3° 0.6879 0.2379

0.20 52.9° 0.7815 0.2764
0.15 58.4° 0.8779 0.3112
0.10 63.4° 0.9701 0.3270

Figure 3 shows a sequence of such profiles, ranging from x i / к = 0.25 to x i / n — 1.0.

The corresponding values of a\ and are given in table 1 .
In this way, we have constructed a one-parameter sequence of initial wave profiles
with which we can test our hypothesis regarding the consequent high accelerations.

5. Tim e-developm ent of the initial profiles

Now, taking the initial conditions corresponding to the profile of figure 2 , that is,
ai = 0.6879, a2 = 0.2379,
(5.1)
di = 0 , <22 = 0 ,

we can use the time-stepping equations (3.5) to follow the ensuing motion, and so
obtain the sequence of profiles shown in parts (a) and (6) of figure 4. The vertical
acceleration of the surface particle on the central plane x = 0 is plotted as a function
of the time in figure 4c. As will be seen, this rises from a value just less than g
at t = 0 to a maximum of nearly 2.8g near t = 8 .8 . It then falls away steeply
to negative values. The computation is carried only as far as t = 0.96, when the
free surface begins to form a sharp corner and the present method loses accuracy.
However, in the region of maximum acceleration, the surface is locally almost flat.
No loss of accuracy is evident until after the acceleration becomes negative.
In this integration, the time-step A t was taken as 0.001 and the number N of
harmonics was 100. The total energy E remained a constant to one part in 106 until
t = 0.95 and to one part in 10s thereafter.
Similar calculations were carried out for other profiles in figure 3, and the results
are shown in figures 5-7. The accelerations are compared in figure 8 . As the half-width
X i / n of the trough is decreased and the maximum surface slope a simultaneously
increases, so the maximum acceleration j/(0 ) increases and the time to maximum
acceleration falls. In figure 7, when х\/тг = 0.10 and a = 63.4°, the maximum
vertical acceleration is seen to exceed Юр.
This is to be compared with the acceleration of a particle in the trough of an
ordinary time-periodic standing wave, which does not exceed about ^ g .
The vertical velocities of particles at the central point x = 0, which may be found
by integrating the acceleration in figure 8 , are shown in figure 9. The maximum
velocities attained are of order unity, in non-dimensional units, and are somewhat
less variable than the maximum accelerations. It will be recalled th at (g / k ) 1^2 is the
phase speed of a progressive wave of the same wavelength 2тг/к as the standing wave.
(Here, к = 1.) The maximum vertical velocities in figure 11 are between 1.4 and 1. 8
times this phase speed.

Proc. R. Soc. Lond. A (2001)

2226

Vertical j e t s fr o m standing waves 501

Figure 3. A sequence of piecewise-circular profiles (dotted curves) similar to figure 2, with the
corresponding truncated-series curves (solid): (a) x \ / i r = 0.25; (b) x i / i t = 0.20; (c) xx/'K = 0.15;
( d ) x \ / t z = 0 .1 0 .

The initial accelerations in figure 8 are rather low, showing th at the pressure field
initially changes only slowly. However, if the normal velocities in the wave trough
were to be assumed non-zero and directed inwards, we should expect the resulting
accelerations to be consistently higher.
Over the ranges of time indicated in figure 8 , no instabilities in the surface profile,
numerical or otherwise, were observed. This may be due to the fact th at the vertical
acceleration always exceeded —g, not only at the central point x = 0 , but also at
other points along the free surface.

Proc. R. Soc. Lond. A (2001)

 2227

502 M. S. Longuet-Higgins

W mt
Figure 4. (a) Development of the initial wave profile corresponding to a\ = 0.6879, <22 = 0.2379
and di = a.2 = 0; A t = 0.001, N = 100. Time-interval between successive profiles is 0.02.
(6) Central section of (a) enlarged, (c) Vertical acceleration of the surface at x = 0 as a function
of time.

In figure 8 , the increase in acceleration from case (a) to case (d) might be thought
to be due to the increase in depth of the wave trough, which tends to increase
the pressure locally, hence the inward acceleration of the particles. However, this

Proc. R. Soc. Lond. A (2001)

2228

Vertical je ts from standing waves 503

kx

kx

Figure 5. (a) Development of the initial wave profile corresponding to a\ = 0.7815, a-i — 0.2764
and ai = a.2 —0; Д£ = 0.001, N — 100. Time-interval between successive profiles is 0.02.
(6) Central section of (a) enlarged, (c) Vertical acceleration of the surface at x = 0 as a function
of time.

would not be strictly true, for, if the scale of any wave is enlarged while leaving
the shape of the surface profile unaltered, the resulting particle acceleration would
remain unchanged, despite the changes in the corresponding particle velocities. The

Proc. R. Soc. Lond. A (2001)

 2229

M. S. Longuet-Higgins

kx

W mt
Figure 6. (a) Development of the initial wave profile corresponding to oi = 0.8779, a2 = 0.3112
and di = <Z2 = 0; A t — 0.001, N = 150. Time-interval between successive profiles is 0.02.
(6) Central section of (a) enlarged, (c) Vertical acceleration of the surface at x = 0 as a function
of time.

real cause of the increased acceleration must be sought in the narrowing of the circular
trough relative to the overall wavelength of the wave, together with an increase in the
maximum slope angle a. The latter is a purely geometrical effect enabling relatively
more fluid to be directed inwards towards the centre of the circular arc.

Proc. R. Soc. bond. A (2001)

2230

Vertical je ts from standing waves 505

kx

Figure 7. (a) Development of the initial wave profile corresponding to ai = 0.9701, аг = 0.3270
and di = fa = 0; A t = 0.001, N = 200. Time-interval between successive profiles is 0.02.
(b) Central section of (a) enlarged, (c) Vertical acceleration of the surface at x = 0 as a function
of time.

6. Further exam ples

The sensitivity of the developing surface to short-scale perturbations of the initial
wave profile is shown by the following two examples.

Proc. R. Soc. Lond. A (2001)

 2231

506 M. S. Longuet-Higgins

W mt
Figure 8. Comparison of the accelerations shown in figures 4-7: (a) x i/ж = 0.25;
(6) x i /я- = 0.20; (с) xi/7r = 0.15; (d) x \/ n = 0.10.

In figure 10a axe shown successive profiles corresponding to the initial conditions
ai = 0.50, di = 0,1
02 = —0.25, a.2 = 0, > (6 . 1 )
03 = —0.18, аз = 0 , J
and N = 80. A close-up of the central part of the profile is seen in figure 106. The
parameters in (6 . 1) were chosen so that the initial profile had a rounded trough,
somewhat similar to that in Cooker & Peregrine (1990) (see also fig. 11 of Longuet-
Higgins & Oguz (1997)).
The acceleration у of a particle at the free surface on the axis of symmetry (я = n)
is shown as a function of time in figure 106. Fairly high accelerations у ~ 3 are
attained at around t = 0.47, well before the occurrence of the steepest surface slopes.
The next example shows the sensitivity of the maximum acceleration to small
changes in the initial conditions. If, instead of (6.1), we initially take
a i = 0.50, <21 = 0,
a 2 = —0.25, <22 = 0, ► (6-2)
аз = —0.25, аз = 0,
with N = 80 as before, we obtain the sequence of profiles shown in figure 11a.
The corresponding curve of acceleration versus time is shown in figure 116. This is

Proc. R. Soc. Lond. A (2001)

2232

Vertical j e t s f ro m standing waves 507

Figure 9. The vertical velocities y /(g k ) 1/,a corresponding to the accelerations shown in figure 8.

generally higher than in figure 106 and, instead of falling off after t = 0 .4 7 5 , it has a
second spurt, leading to a maximum of over 5g. This may be attributed to the fact
th a t the wave trough at around t = 0 .6 , instead of developing monotonically, itself
develops a smaller secondary trough superposed on the main trough. This acts as
a secondary standing wave, in the gravitational field defined by the first, and hence
produces a secondary vertical acceleration.
Indeed, it is possible to imagine a succession of smaller and smaller standing waves
superposed on each other, as in a fractal (see Longuet-Higgins 1994), leading to
indefinitely large vertical accelerations.

7. D iscussion and conclusions

We have shown th at high vertical accelerations and velocities can arise in the troughs
of standing waves in deep water, particularly if the wave trough has a circular form.
By conservation of mass, the particles directed towards the centre cannot reach
their destination simultaneously, and so require a vertical outlet in the form of a jet.
Dynamically, the destruction of the horizontal momentum requires large horizontal
pressure gradients, hence a high-pressure maximum at a point on the vertical plane of
symmetry below the free surface. But the pressure at the surface itself remains a con­
stant, by hypothesis. Hence there is a strong vertical pressure gradient, accelerating
the particles near the surface vertically upwards.

Proc. R. Soc. LoruL. A (2001)

 2233

M. S. Longuet-Higgins

(a)

(gk)m t
Figure 10. Development of a wave with initial conditions a\ = 0.5, 0.2 = —0.25, аз = —0.18 and
dx = a,2 = аз = 0. (a) Central section enlarged. (6) Acceleration of the surface at x = 0, as a
function of the time. Time-interval between successive profiles is 0.02.

The process has much in common with that of the shaped charge, where it has
been called the ‘Munro effect’ or the ‘Neumann effect’ (see Birkhoff et al. 1948). Note
that the mathematical model offered by Birkhoff et al. is a steady two-dimensional
flow resulting from the convergence of two sheets of fluid onto a central plane. The
situation considered in this paper is different, in that the whole body of fluid below

Proc. R. Soc. Lond. A (2001)

2234

Vertical j e t s f ro m standing waves 509

0.0 0.2 0.4 0.6

(gk)ll2t
Figure 11. Development of a wave with initial conditions as in figure 9 but with аз — —0.25.
(a) Central section enlarged. (6) Acceleration of the surface at x = 0, as a function of the time.
Time-interval between successive profiles is 0.02.

the surface participates in the motion. Moreover, here we give an account of the
complete history of the time-dependent flow, starting from a state of rest.
The calculations have been carried as far as the times when the wave profile
approaches a sharp angle; then the present method, using Fourier analysis, loses
accuracy. The further development of the jet, which displays several interesting types

Proc. R. Soc. bond. A (2001)

 510 M. S. Longuet-Higgins

of behaviour, requires a different computational approach and will be described in

another paper, f
In shallow water, with the aid of a rigid bottom, much higher accelerations can
undoubtedly be attained, as is shown by the computations of Cooker & Peregrine
(1990, 1991). However, there is no apparent reason why the same methods of calcu­
lation used for waves in deep water should not be extended to shallow water, with a
moderate increase in complexity.
This work has been supported by the US Office of Naval Research under Contract N00014-94-
1-0008.

R e fe ren c es
Balk, A. M. 1996 A Lagrangian for water waves. Phys. Fluids 8, 416-420.
Birkhoff, G., MacDougall, D. P., Pugh, E. M. & Taylor, G. 1948 Explosives with lined cavities.
J. Appl. Phys. 19, 563-582.
Chan, E. S. & Melville, W. K. 1988 Deep-water plunging wave pressures on a vertical wall. Proc.
R. Soc. Lond. A 417, 95-131.
Cooker, M. J. & Peregrine, D. H. 1990 Water wave impact: computations, theory and a compar­
ison with measurements. In Proc. 6th Int. Workshop on Water Waves and Floating Bodies,
Woods Hole, MA, USA, pp. 49-53.
Cooker, M. J. & Peregrine, D. H. 1991 Violent surface motion as near-breaking waves meet a
wall. In Breaking waves (ed. M. L. Banner & R. H. J. Grimshaw), Proc. IUTAM. Symp.,
Sydney, Australia, pp. 291-287. Springer.
Longuet-Higgins, M. S. 1983 Bubbles, breaking waves and hyperbolic jets at a free surface. J.
Fluid Mech. 127, 103-121.
Longuet-Higgins, M. S. 1994 A fractal approach to breaking waves. J. Phys. Oceanogr. 24,
1834-1838.
Longuet-Higgins, M. S. 2000 Theory of water waves derived from a Lagrangian. I. Standing
waves. J. Fluid Mech. 423, 275-291.
Longuet-Higgins, M. S. & Oguz, H. N. 1995 Critical microjets in collapsing cavities. J. Fluid
Mech. 290, 183-201.
Longuet-Higgins, M. S. & Oguz, H. N. 1997 Critical jets in surface waves and collapsing cavities.
Phil. Trans. R Soc. Lond. A 355, 625-639.
Tuck, E. O. 1998 Solution of nonlinear free-surface problems by boundary and desingularised
integral equation techniques. In Computational Techniques and Applications Conf. (ed.
J. Noye, M. D. Teubner & A. W. Gill), pp. 11-26. World Scientific.

t In recent work, vertical accelerations as high as 100$ in deep w ater have been obtained.
2236

PHYSICS OF FLUIDS VOLUME 13. NUMBER 6 JUNE 2001

On the breaking of standing waves by falling jets

Michael S. Longuet-Higgins
Institute fo r Nonlinear Science, University o f California, San Diego, La Jolla, California 92093-0402
Douglas G. Dommermuth
Naval Hydrodynamics Division, Science Applications International Corporation,
10260 Campus Point Drive, La Jolla, California 92121

(Received 4 August 2000; accepted 22 February 2001)

By direct numerical time stepping, using two different methods of calculation, we follow the
development of a standing gravity wave on the surface of deep water when it is given an initial
energy exceeding that of the most energetic periodic wave of the same wavelength. Two aspects of
the motion are studied in detail: the sequence of wave profiles close to the instant when the
maximum surface slope angle is close to 45° and, second, the conditions under which a sharp cusp
is formed at the free surface. When a cusp is formed, it can fall vertically into the wave trough,
enclosing “ bubbles of air.” © 2001 American Institute o f Physics. [DOI: 10.1063/1.1369141]

I. INTRODUCTION arc not exclusively a shallow-water phenomenon but can oc­

Standing water waves are often seen to result from the cur also in deep water, and lend to arise whenever the surface
reflection of waves at a steep shoreline or harbor wall. They profile of the wave trough has the form of a circular arc.
can also be observed near the eye of a circular storm, or in Particles accelerated toward the center of the arc are forced
the wake of a fast-moving circular depression.1 The breaking by the conservation of mass to develop a high-pressure gra­
of a standing wave often throws up jets or spray into the air. dient and vertical acceleration. Subsequently the acceleration
Yet compared to the many studies of progressive wave falls sharply and the surface profile develops a sharp comer
breaking, the present literature on the breaking of standing with a maximum slope of about 45°. The surface accelera­
water waves is remarkably sparse. Penny and Price2 devel­ tion then falls below zero. The numerical calculations were
oped a series expansion of deep-water standing wave to fifth not carried beyond that point. The emerging jet at that stage
order in the wave steepness, and put forward an argument resembled a small-scale standing wave forced beyond a 45°
that the highest wave should have a sharp angle of 90° at the slope. A separate study of this situation but on a larger hori­
crest Taylor3 doubted their argument but nevertheless car­ zontal scale seemed to be required.
ried out laboratory experiments suggesting that the limiting In the present paper we aim to supply such a study. We
slope angle was indeed about 45°. Saffman and Yuen4 used a focus on the following problem (Fig. 1). At dme f = 0 the
boundary-integral technique of Longuet-Higgins and free surface of an infinitely deep, incompressible, and non-
Cokelet5 to calculate periodic waves numerically, and found viscous fluid is assumed to be flat, and the velocity of par­
that their steepest wave was in fair agreement with Taylor’s ticles at the surface у = 0 is assumed to be sinusoidal:
measurements. Aoki6 corrected and extended the small-
J>= C cosx, (1-1)
amplitude series due to Penny and Price.2 Schwartz and
Whitney7 introduced a new type of series expansion and car­ in normalized coordinates, gravity g being taken as unity and
ried it to a much higher order. Mercer and Roberts8 used an the wavelength L as 2ir. What is the subsequent motion?
improved boundary-integral technique to study very steep, At small values of the initial velocity С we expect the
periodic waves, and also studied their stability. Agnon and subsequent motion to be a nearly periodic standing wave of
Glozman9 discovered some new types of periodic standing low amplitude. At larger values of C, say 0.1< C < 0.5, we
wave, which they calculated to great accuracy; they did not expect a wave of finite steepness, but it will not be perfectly
study limiting waves. periodic since, as noted by Penny and Price,2 in a perfectly
On the other hand there seem to have been no serious periodic standing wave the free surface is never quite flat;
studies of actually breaking, nonperiodic waves of standing every half-period the surface differs from the mean level by
type until the numerical calculations by Cooker and a small amount proportional to C4cos2x, where x is the
Peregrine1011 of progressive, shallow-water waves impacting horizontal coordinate. But if С is gradually increased without
a vertical wall. This confirmed that a vertical jet could some­ limit, what then happens? How does the surface behave
times be thrown upward from the trough of the wave and that when the total energy exceeds that of a periodic standing
the acceleration of a surface particle at the wall could exceed wave of the same wavelength? Does the surface develop a
250 g. Their data were further analyzed and interpreted by sharp cusp, and if so, under what conditions?
Longuet-Higgins and Oguz.12 As to the numerical time-stepping procedure, we adopt
A recent study by the present author13 has shown that two different but complementary methods. The first (Method
such vertical jets from the trough of a steep standing wave A) makes use of Balk’s14 parametric representation, further

1070-8631/2001/13(eyi 652/8/S18.00 1652 О 2001 American Institute ol Physics

 2237

Phys. Fluids. Vol. 13, No. 6, June 2001 On tfie breaking of standing waves 1653

FIG. 1. (a) Profiles of the free surface when C=0_5 and /=0(0.05) 1.65. (b) Enlargement of the central segment of (a), (c) Profiles of the free surface when
С = 0.5 and i - 1.65(0.05)3.25. (d) Enlargement of the ccntral segment of (c).

developed by Longuet-Higgins.15 This is useful and accurate where Y\ ,Y 2,...,Y n are functions of the time t only. Y0,
as long as the initial velocity С does not exceed about 0.5. which determines the mean level, is given in terms of
The results obtained by this method are described in Sec. Ш. .
у ...... У* by
The second method of calculation is the boundary- N
integral technique described in Longuet-Higgins and К0= - Е 2лК*. (22)
Dommcrmuth,16 which is a refinement of the technique in­
troduced by Longuet-Higgins and CokeleL5 The results are
described in Sec. V. In Sec. VI we make comparisons with It is convenient to write
periodic waves. Conclusions and a discussion follow in an = 2nYn, (2.3)
Sec. VO.
and then the equations for time-stepping the coefficients
a„(r) have the form
II. METHOD A

In Balk,14 the horizontal and vertical coordinates (xy) of

the free surface are given parameterically by a series of the (2.4)
form
N where $„(*,/), and Un are all polynomials of low
(*S 4) degree in the coefficients aa . An overdot ( ') denotes
* = £ + 2 2Y„ sin n (,
differentiation with respect to I. As long as ЦбЛ does not
vanish, which can be verified, Eqs. (2.4) may be solved for
'N <2 1 >
the Д„, and time stepping can procccd. A numerical
2 r„ c o s n f, check of the accuracy is to calculate the total energy
2238

1654 Phys. Fluids, Vol. 13. No. 6, June 2001 M. S. Longuet-Higgins and D. G. Dommermuth

FIG. 2. Plot of the maximum surface slope angle о against the curvature к FIG. 3. Plot of the vertical acceleration у of a panicle at the crest, as a
at the wave crest jc = 0, when С =0.5. function of i, in the case С =0.5.

EN(Y0,Y),...,YN\Y0,fi,...,YN). In all the calculations re­

ported here, En deviated from its initial value by less than J - V 21 ОД
one pan in 105. It was found practical to do calculations with
values of N up to 200, usually less. But from (2.2) we find, when Yn = 0, that
y0= - 4 ^ = _ £j2 (3.3)
III. RESULTS
Also, from (2.4) we find
Figure 1(a) shows a succession of surface profiles in the
case С = 0.5, that is to say, with the initial conditions a\ d l = 0, <S2 = 4d?, dn = 0, when л > 2 . (3.4)
= 0, d] = 0.5, and a n= d „ = 0 when 2 ^n * * N . The surface
Hence
rises from being horizontal at time f = 0 to its maximum
displacement at around r = 1.65. Figure 1(b) is an enlarge­ y = - d 21+ 2d2l=d2l. (3.5)
ment of the central part of the wave. From /=1.65 until /
When C = d j = 0.5, this yields у =0.25.
= 3.25, the surface falls; see Figs. 1(c) and 1(d). However, at
/= 3.25, although the mean level is necessarily zero, the free
surface is not quite flat. The vertical displacement is given by IV. THE TOTAL ENERGY
y ,= ccos2x, (3.1) In the absence of dissipation the total energy E (per unit
approximately, where e= 0.015. This is a measure of the of horizontal distance) must remain a constant. E is given
departure of the motion from periodicity over half a quite simply in terms of C. For, at the initial instant the
“ cycle.” surface is flat and the potential energy vanishes. Therefore E
It is illuminating to plot the maximum angle of inclina­ can be found from the initial kinetic energy alone. But the
tion or of the free surface against к, the curvature of the vertical and horizontal components of motion at the initial
profile on the central line x = 0; see Fig. 2. When С = 0.5, for instant are given by
example, the angle a rises to a maximum of about 32° at / фх=С е~у cosx,
= 1.65, where к —2.2. Then a falls to near zero at r=3.25.
However, the path of (к, a), is not quite closed. This implies ^фу=Се
r -yy smx,
• (4Л)
that for any given value of the maximum slope a there are where ф is the velocity potential. Hence, we have
two possible values of the crest curvature к. On the upward
path the curvature is less and the wave crest is more rounded.
On the downward path, the curvature is greater, implying E= h (4-2)
that the wave crest is narrower. When the starting velocity С
Now the maximum energy E* for a periodic standing
is small, the point (к, a) returns nearly along the same curve
so the wave is more nearly periodic. wave has been found by Mercer and Roberts8 to be
The acceleration у of the particle at the wave crest is ££„= 0.077 74. (4.3)
shown as a function of the time in Fig. 3. This falls to below
-0 .7 g at the time of maximum elevation /= 1.65. The value If we compare Eqs. (4.2) and (4.3), we find that the value of
of the acceleration al time r = 0 is not zero, as might have С yielding the same total energy as a periodic wave of maxi­
been expected, but positive. This can be verified from (2.1) mum energy (but not maximum amplitude) is
and (2.4). For, when x = it and / = 0 we have from (2.1), C = 0.5576. (4-4)
 2239

Phys. Fluids. Vol. 13, No. 6, June 2001 On the breaking of standing waves 1655

Oc
FIG. 4. Graphs of the maximum surfacc slope angle a vs the curvaiure к at FIG. 6. Profiles of the free surfacc when С =0.7 and f “ 0(0.5)4.0.
the wave crest, when С =0.3, 0.4, and 0.5.

conditions are that at time f = 0 the free surface is given by

Thus, wc expect to observe interesting behavior in the non­
y = 0 , and that on the free surface,
periodic wave at around this value of C.
ф = С ь in*, (5.1)
V. HIGHER VALUES OF С
from Eqs. (4.1).
For values of С exceeding about 0.5, the method of cal­ When 0 = 0 .3 , 0.4, and 0.5, both the surfacc profiles and
culation described in Sec. П loses accuracy on account of the the curves o f a vs к are recovered very precisely; see Fig. 4.
high number of terms N required in the Fourier series (2.1). When С = 0.6, 0.7, and 0.8 we obtain the sequence of
Wc therefore resort to the boundary-integral method de­ profiles shown in Figs. 5,6, and 7, respectively. In Fig. 6, for
scribed by Longuet-Higgins and Dommermuth.16 The initial example, when С = 0.7, it can be seen how the crest curva­

FIG. 5. Profiles of the free surface when C = 0.6 and r=0(0.5)4.0. FIG. 7. Profiles of the free surface when 0 0 . 8 and l = 0(0.5)4.0.
1656 Phys. Fluids. Vol. 13, No. 6. June 2001 M. S. Longuet-Higgins and D. G. Dommermuth

FIG. 8. Graphs of the maximum slope angle or vs the curvature к at the FIG. 10. Graph of the maximum surface slope or vs the crest curvature *
wave crest * = 0 when С =0.55. 0.6, 0.7, and 0.8.
when C - 1.0 and 2.0.

ture continues to increase from ( = 0 onward, well beyond smoothly and without any accompanying singularity in the
the times of maximum crest elevation or maximum surface time when a = 4 5 °.
slope. The situation is clearer still from Fig. 8, where the Profiles for some higher values of С are shown in Figs. 9
maximum slope a has been plotted against the crest curva­ and 11. From Fig. 10, it will be seen that when С = 1.0 the
ture к for similar values of C. The times of maximum cur­ maximum slope a reaches 83° and then the maximum cur­
vature are, respectively, t = 1.85, 1.95, 2.4, and 3.2, well be­ vature к is nearly 400, so that the crest is essentially a cusp.
yond the times of maximum elevation, which never exceed The subsequent fate is also interesting, for from Fig. 9(b) it
2 .0. can be seen that the falling jet produces a cavity in the wave
From Figs. 7 and 8 it is also clear that whenever С trough. The effect is a time reversal o f the “ cavity effect”
> 0 .7 the maximum surface slope increases beyond 45° described by Longuet-Higgins.13

(a) oc (b) X

FIG. 9. (a) Profiles of the free surface when C = 1.0 and i = I ). (b) Profiles of the free surface when C * 1.0 and f = 2.0(0.1)5.0.
 2241

Phys. Fluids, Vol. 13, No. 6. June 2001 On the breaking of standing waves 1657

2 0 2 4
(a)

FIG. 11. (a) Profiles of ibe free surfacc when C = 2.0 and f = 0(1.0)5.0. (b)-(e) Profile of the free *uifa< : when C=2 end 1= 5.5, 6.0,6.5. and 6.65. Number
of grid points: 751.

Figure 12, in which the crest elevation y, velocity y, and The case С - 2 . 0 is somewhat similar. From Fig. 10 the
acceleration у are plotted against time, shows that when maximum surface slope a is above 88°, and the maximum
1 .6 < f< 2 .6 the ratio jl/g was nearly - 1 , so that the particles curvature к is over 900. From Fig. 13, the fluid near the crest
near the crest were effectively in freefall. was effectively in free fall when 1 < /< 6 .0 . This time, how­
2242

1658 Phys. Fluids, Vol. 13, No. 6, June 2001 M. S. Longuet-Higgins and D. G- Dommenmuth

VI. COMPARISON WITH PERIODIC WAVES

From Figs. 2, 5, and 8 we have seen that waves o f a
given maximum slope a can exist in two different forms,
with either a smaller or a larger curvature к at the crest.
There is a parallel here with steep, lime-periodic standing
waves, as computed by Mercer and Roberts.8 In their Fig. 3
they show the profiles o f two steep, periodic standing waves
having the same overall steepness but different crest curva­
tures; one of the wave crests is sharp-crcstcd, with a down­
ward acceleration of 0.9800 g. The other crest is relatively
rounded, and the downward acceleration only 0.8855 g.
Note that each of the periodic waves is time symmetric,
that is to say, it passes through the same sequence o f profiles
on its upward as on its downward motion. The wave with the
rounded crest, therefore, does not have a sharp-crested pro­
file at any subsequent time. Nevertheless the comparison
suggests that the behavior of the nonperiodic waves studied
in this paper is in a sense generic, and that the periodic
waves can be regarded as a special case.
We draw attention to the fact that just as in progressive
t waves of constant wavelength, so also in standing waves, the
total energy E of the wave does not increase monotonically
FIG. 12. Plois of the crest elevation y, vertical velocity and acceleration
у as functions of the lime t, when C = 1.0.
with the overall wave steepness s or with the maximum sur­
face inclination a (see Mercer and Roberts,8 Figs. 4 and 7).
Instead, E reaches a maximum at a ccrtain value o f s and
ever, the cavity produced by the falling jet is much deeper, then diminishes. At higher values of s the crest is more
and it appears from Figs. 11(b) and 11(e) that at rime t sharply curved and the surface not far from the crest be­
= 6.65 the cavity walls are about to touch the sides of the jet, comes concave, resulting in a decrease in the potential en­
enclosing two “ bubbles o f air.” The effect is similar to that ergy. Thus, for any given value of E, close to E ш , there
found by Prosperetti and Oguz17 for jets with a large aspect exist two distinct standing waves: one with a more rounded,
ratio, falling onto a free surface. and the other with a sharper, crest. So it is also for nonperi­
Figure 11(e) represents a previously unsuspected way odic waves, except that the same wave passes through both
for a standing wave to break. types of profile.
The theoretical form of the highest periodic standing
wave has not been determined, nor do we even know that, if
such a wave exists, its form is unique. If it does exist, then its
profile is probably sharp-crested, because by the above argu­
ment a wave with a rounded crest could presumably be sup­
planted by a wave that was shaip-crested, with a profile that
was concave on each side of the crest. Considering the time
sequence of the wave profiles, they would then pass from
round-crested to sharp-crested and back to round-crested, in
a time-symmetric manner. In the nonperiodic waves o f the
present paper, particularly those with energy close to E ^ , a
similar transition occurs but with the time symmetry broken.

VII. CONCLUSIONS
We have investigated the response of a body o f deep
water bounded by an initially flat horizonal surface when that
surface is given an initial normal velocity v of the form
v = Ccoskx (g = k = \ ) . (7.1)
When C <0.55 the surface oscillates approximately as a
periodic standing wave, but as С is increased there is an
increasingly pronounced asymmetry between the upward and
t subsequent downward motions. In the upward motion the
FIG. 13. Graphs of the crest elevation y, vertical velocity >, and accelera­ profile o f the crest is more rounded, and on the downward
tion у as functions of the time t, when С =2.0. motion the crest is narrower and more sharply curved. When
 2243

Phys. Fluids, Vol. 13. No. 6. June 2001 On the breaking of standing waves 1659

C > 0 .8 the falling crest develops into a sharp, upward point­ ety in November 1999, and at the IUTAM Symposium on
ing spike that falls back into the wave trough. When С Free-Surface Flows, Birmingham, UK, in July 2000.
> 2 .0 , the falling crest may trap a bubble of air in the trough. M.S.L.H. is supported by the Office of Naval Research under
Since internal friction is ignored, we can, of course, Contract No. N00014-94-1-0008. D.G.D. is supported partly
imagine time to be reversed. Starting with a circular cavity in under Contract No. N00014-97-C-0345.
the trough of a standing wave, a jet will rise up vertically. Its
*M. S. Longuet-Higgins. "A theory of the origin of microseisms.” Philos.
tip will become rounded before falling back again.
Trans. R. Soc. London, Ser. Л 243, I (1950).
A comparison with the forms of time-periodic standing 3W G. Penny and A. T. Pricc. “ Finite periodic stationary gravity waves in
waves of energy slightly below the maximum has high­ a perfect fluid. Pan 2," Philos. Trans. R. Soc. London, Ser. A 244, 254
lighted the existence, at the same total energy £, of two (1952).
JG. L Taylor, “ An experimental study of standing wives," Proc. R. Soc.
distinct kinds of standing-wave profiles for periodic waves London. Ser. A 218. 44 (1953).
also: profiles with rounded crests and those with relatively 4P. G. Saffman and H. C. Yuen, "A note on numerical computations of
sharp crests. Urge amplitude standing waves." ). Fluid Mech. 95. 707 (1979).
In the above investigation, both the viscosity and the 5M. S. Longuet-Higgins and E. D. Cokelet, “ The deformation of steep
surfacc waves on water. L A numerical method of computation." Proc R.
surface tension T o f the water have been neglected, so that Soc. London. Ser. A 350. I (1976).
the conclusions are really limited to standing waves on a eH. Aold, "Higher order calculation of finite periodic standing gravity
scale much larger than ( T /p g ) in. For shorter waves, the waves by means of the computer,” J. Phys. Soc. Jpn. 49, 1598 (1980).
occurrence of local regions o f sharp curvature would un­ 7L. W. Schwartz and A. K. Witney. “ A semi-analytic solution for nonlinear
standing waves in deep water." J. Fluid Mech. 107, 147 (1981).
doubtedly produce a wealth o f parasitic capillaries. At the *G. N. Mercer and A. J. Roberts, "Standing waves in deep water Their
tips of the jets, we expect to find the formation of droplets. stability and extreme form," Phys. Fluids A 4. 259 (1992).
Note added in proof. Since this paper was submitted, the *Y. Agnon and M. Glozman, “ Periodic solutions for a complex Hamil­
tonian system: New standing water waves." Wave Motion 24. 139 (1996).
authors’ attention has been drawn to two different laboratory
10M. J. Cooker and D. H. Peregrine, "Water wave impact: Compulations,
experiments in which the trapping o f air by a falling jet has theory and a comparison with measurements." in Proceedings o f the 6th
apparently been seen. Chan and Melville18 generated a fo­ International Workshop on Water Waves and Floating Bodies. Woods
cused train of progressive waves which then impinged on a Hole, MA. 1990. pp. 49-53.
n M. J. Cooker and D. H. Peregrine, "Violent surface motion as near-
vertical wall. They observed both the formation of a jet brcalring waves meet a wall." in Breaking Waves. Proceedings of the
against the wall and the entrainment of air when the jet fell IUTAM Symposium, Sydney. Australia, edited by M. L Banner and R. H.
back into the wave trough. The jet may have been influenced J. Grimshaw (Springer-Vcrlag, Berlin, 1991), pp. 287-291.
by frictional forces at the wall. Second, Jiang et a i 19 gener­ HM. S. Longuet-Higgins and H. N. Oguz, "Critical jets in surfacc waves
and collapsing cavities," Philos. Trans. R. Soc. London, Ser. A 355, 625
ated standing waves subharmonically in a tank 60 cm long (1997).
(one wavelength). The waves sometimes produced a vertical 1JM. S. Longuet-Higgins, "Vertical jets from standing waves in deep wa­
jet that then fell back into the wave trough. By this procedure ter," Proc. R. Soc. London, Ser. A 457, 495 (2001).
MA. M. Balk. “ A Lagrangian for water waves," Phys. Fluids 8,416 (1996).
they avoided frictional effects at the endwalls o f the tank. Air
15M. S. Longuet-Higgins, “ Theory of water wnves derived from a Lagrang­
was certainly entrained, as was evidenced by the emission of ian. L Standing waves," J. Fluid Mech. 423. 275 (2000).
sound from the entrained bubbles. But a lack of stabib'ty in I6M. S. Longuet-Higgins and D. G. Dommermuth, "Crest instabilities of
the falling jet, leading (on such a small scale) to its breakup gravity waves. Pan 3. Nonlinear development and breaking," J. Fluid
Mech. 336. 33 (1997).
into droplets, will have affected the phenomenon. 17A. Prospcreui and H. N. Oguz, "Air entrainment upon liquid impact,”
Philos. Trans. R. Soc. London. Ser. A 355, 491 (1997).
ACKNOWLEDGMENTS UE. S. rhan and W. K. Melville, “ Deep-water plunging wave pressures oo
a vertical plane wall," Proc. R Soc. London, Ser. A 417, 95 (1988).
A part of the calculations described here was presented »L. Jiang, M. Pcrlin, and W. W. Schultz, “ Period tripling and energy dis­
at the New Orleans meeting of the American Physical Soci- sipation of breaking standing waves,” J. Fluid Mech. 369, 273 (1998).
2244

VERTICAL JETS FROM ST A N D IN G

WAVES; THE BAZOOKA EFFEC T

Michael S. Longuet-Higgins
Institute for Nonlinear Science, University of California, San Diego,
9500 Gilman Drive, La Jolla, California 92093-0402
[email protected]

A b stra c t It is shown that a sufficient condition for the occurrence of a high vertical
acceleration in a two-dimensional standing wave is for the initial profile
of to have a narrow trough that is approximately circular. In such cases
■we find vertical accelerations of the free surface exceeding lOg. Much
higher accelerations are obtained when the initial wave trough is some­
what flattened so that jets emerging from the two corners of the trough
meet on the axis of symmetry. In such a case we find accelerations
exceeding lOOg.
Lastly we consider a different, but complementary, class of initial
conditions when the free surface is initially flat but is given an upwards
velocity dependent sinusoidally on the horizontal coordinate. When the
kinetic energy injected into the fluid exceeds the maximum possible for
a time-periodic standing wave it is found that a new type of breaking
can occur in which air is trapped by the collapse of the wave trough.
Throughout this paper surface tension is ignored and the fluid flow
is treated as inviscid and irrotational.

K eyw ords: Jets, standing waves, breaking, air entrapment.

1. INTRODUCTION
Vertical jets thrown upwards when a steep progressive wave meets a
cliff or harbor wall are a common phenomenon around coastlines and in
shallow water. A numerical model for shallow water, using a boundary-
integral technique, was constructed by Cooker and Peregrine (1991), who
found that in some cases the vertical acceleration of a particle at the wall
could exceed 250g. Longuet-Higgins and Oguz (1997) showed that such
high accelerations were typical of a class of jets occurring in near-critical
situations where two surfaces came together and nearly, but not quite,
trapped a pocket of “air.” Other examples of high vertical accelerations
195
A.C. King and YD. Shikhmurzaev (eds.), UJTAM Symposium on Free Surface Flows, 195-203.
© 2001 Kluwer Academic Publishers. Printed in the Netherlands.
 2245

196

Figure 1. Definition diagram for a piecewise circular wave profile.

are known to occur in axisymmetric standing waves, as shown by the

laboratory experiments of Longuet-Higgins (1983) and ZefF et al. (2000);
also in collapsing bubbles and cavities.
In deep water, jets have been observed in two-dimensional waves im­
pacting a vertical wall (Chan and Melville 1988) and in subharmoni-
cally forced two-dimensional standing waves (Jiang et al. 1998). In the
present paper we shall show theoretically how high vertical accelerations
can occur in such deep-water wave motions also.

2. A CLASS OF INITIAL CONDITIONS

Because of the large number of degrees of freedom of a space-periodic
wave on deep water, it is essential to narrow the choice of initial con­
ditions to a class defined by at most a few parameters. On physical
grounds we may begin with wave profiles having the form of an almost
circular arc. For, initially the acceleration of the particles in the trough
will then be directed inwards towards the centre of the circle. By con­
tinuity the fluid cannot all simultaneously reach the centre; hence a jet
will emerge along the vertical axis.
A two-parameter class of profiles with circular troughs and a contin­
uous tangent can be defined as in Figure 1. Here the parameters are (1)
the ratio p = 2 x \ /L, where 2x\ is the width of the trough and L is the
total wavelength. Surprisingly, we can fit such a profile remarkably well
by just two terms of the general series
N N
s = £ + ^ 2 Y „ sm n f, у = Yo + ^ 2 Y „ c o s n £ (1)
П=1 n=1
2246

197

Figure 2. A sequence of analytic profiles of the form (2.1) (solid curves) approx­
imating piecewise circular profiles of the type shown in Fig. 1 (dotted curves), (a)
p - 0.25, (b) p = 0.20, (c) p = 0.15, (d) p = 0.10.

where x and у are the horizontal and vertical coordinates of a point on

the water surface, £ runs from —n to тг and the Yn 's are constants. In
fact we can choose Yo,Yi and У2 so that the mean surface level is the
same as in Figure 1, and the profiles agree both at the crest ( = 0 and
the trough £ = ±7r. By fixing p we can select the value of a for which
the fit is optimal. Then by varying p we have a one-parameter sequence
of analytic profiles, of the form (2.1) with N = 2, which have troughs
that are very closely arcs of circles; see Figure 2. The dotted curves in
Figure 2 which represent the piecewise circular profiles, are practically
indistinguishable from the solid curves representing the analytic profiles.
 2247

198

'h

-2 J----- ------ 1------— Ll___ ____ I___ ____ i. . i . -1.4 jJ

-3 • - 2 -1 0 1 2' 3 -.0.3- - 0.2 -0.1 0.0 0.1 0.2, 0.3
ikx
I \kx

Figure 3. (a) Successive wave profiles corresponding to the initial profile (d) of Fig.
2. Number of harmonics N = 100. Time-stepping interval dt — 0.001. Interval At
between successive profiles: Дt = 0.02 (b) central section of (a) enlarged.

use either of two methods to calculate the subsequent motion. The

first method is to regard Yn(t),n = 1 , 2 , . . . , as generalised Lagrangian
coordinates, Yq being dependent on Yi,Y2 , . . . through the condition
that expresses the constancy of the mean level:

a)
l
We can then use the system of O.D.E.’s for the Yn first given by Balk
(1996) and developed by Longuet-Higgins (2000). These are much sim­
pler than the corresponding Hamiltonian equations.
As a result we obtain, for example in the case 2x \/L = 0.25, the
sequence of profiles shown in Figures 3a and 3b. The vertical acceleration
у of a point particle on the mid-line cr = 0 is plotted as a function of the
time in Figure 4, curve d. Evidently у rises to a maximum of about lOg
at about t = 0.41, before falling abruptly to below zero as the curvature
of the wave crest becomes increasingly sharp.
Figure 4 shows the sequence of such acceleration curves, corresponding
to the four initial profiles in Figure 2. It is notable that as the width of
the wave trough 2xi diminishes in relation to the wavelength L, so the
maximum acceleration increases; see Longuet-Higgins (2001a).
When, however, the crest curvature becomes large, the number of sig­
nificant harmonics Yn in the series (2.1) increases rapidly, and the above
method is no longer useful. From this time on, or alternatively from the
initial instant t = 0, we can use a standard time-stepping program using
2248

199

Figure 4• Vertical acceleration of surface particles at x = 0 with the four different

initial conditions shown in Fig. 2: p = 0.25, 0.20, 0.15 and 0.10 respectively.

Figure 5. Subsequent development of the surface profiles in Fig. 3.

 2249

200

1
(fr>
|

/
I
!
1
$

1. . Л-1

Figure 6. (a) Successive profiles of a standing wave with an initially flattened

trough (ai = 5/8, аг = —1/4, аз = —3/8). Total wavelength L = 2nfk. (b) vertical
acceleration у of a surface particle on the central line x = 0 . The acceleration exceeds
lOOg. The calculations could not be carried beyond t = 5.4.

a boundary-integral technique; see Longuet-Higgins and Dommermuth

(1996). In the four cases (a) to (d) above the surface develops similarly to
Figure 5. It can be seen that the vertical jet bifurcates, and falls back on
itself as in a plunging breaker or sometimes a less violent spilling breaker.
For further details see Longuet-Higgins and Dommermuth (2001a).

4. THE BAZOOKA EFFECT

Much higher accelerations can be obtained if the initial wave trough,
instead of being exactly circular, is flattened as in Figure 6. Two areas of
shaxp curvature then arise at the sides of the wave trough. Each of these
gives rise to a jet directed inwards towards the centre line, and when they
are about to meet a very high vertical acceleration is produced. Figure
6b shows that in this case the acceleration exceeded lOOg, comparable
to that found in shallow water.
This effect can be compared with that in the two-dimensional model
of a bazooka proposed by Birkhoff et al. (1948). In their Figure 13,
two steady jets of fluid with velocity V\ making an angle with their
axis of symmetry converge to produce two jets emerging along the axis,
having velocities V2 in opposite directions. In the region of convergence
of the jets there is a sudden increase in the component of the velocity
of a fixed paxticle parallel to the axis of symmetry, and hence a high
acceleration along the axis. But the comparison between that case and
Figure 6 is only rough, since in the model of Birkhoff et al. the flow is
2250

201

Кос

Figure 7. Successive profiles of a standing gravity wave (wavelength L = 2'n/k)

which at time i = 0 is flat and is given a vertical velocity C (gk)1^2 cosfcx, where С =
1.0 (a) t = 0.0 to 2.0, (b) t = 2.5 to 5.0.

assumed to be steady from an Eulerian viewpoint, whereas in Figure 6

it is time-dependent.

5. IMPULSIVELY GENERATED WAVES

Lastly we describe some examples of standing waves initiated at a
different phase of the motion, that is to say with initially zero potential
energy but positive kinetic energy.
Suppose that at time t = 0 the free surface is initially flat so у = 0,
but the surface is given an upwards vertical velocity of the form

у = С{дк)1/2 cos kx (t = 0) (1)

where С is a dimensionless constant. When С « 1, the subsequent

motion is approximately a periodic standing wave. For small but finite
values of С the motion must be nonperiodic, for, as shown by Penny
and Price (1952) the free surface of a standing wave is never quite flat
but has at least a vertical displacement of order C4.
Now the total initial energy of the wave (5.1), is \ C 2, whereas the
maximum energy density E * of a periodic standing wave has been shown
by Mercer and Roberts (1992) to be E* = 0.0774 in the present units.
To impart to the fluid an energy E* we would require С = 0.5576 in
equation (5.1). What then happens when С substantially exceeds this
value?
 2251

202

Figure 8. As in Fig. 7, but with С = 2.0. (a) t = 0 to 5, (b) t = 5.5, (c) t = 6.0,
(d) t = 6.65.
2252

203

An example is shown in Figure 7, where С is set equal to 1.0. When

the free surface is falling (Figure 7b) the crest becomes a narrow jet in
reverse. By the time t = 5.0 this has created a rounded cavity, rather
similar to the cavities investigated in Section 2 above.
When we take С — 2, as in Figure 8, the same phenomenon is accen­
tuated, but now the reverse jet, on falling into the wave trough, actually
traps “air” in a cavity, rather like the falling cylindrical columns of fluid
investigated by Prosperetti and Oguz (1997). Note that the trapping of
air in the trough of a deep-water standing wave has been observed by
Jiang et al. (1998, Figure 8c) in experiments on forced standing waves.

Acknowledgement
The above work has been supported by the U.S. Office of Naval Re­
search under Contract N00014-94-1-0008.

References
Balk, A.M. 1996 A Lagrangian for water waves. Phys. Fluids 8 , 416-420.
Birkhoff, G., MacDougall, D.P., Pugh, E.M. and Taylor, Sir G. 1948 Explosives with
lined cavities. J. Appl. Phys. 19, 563-582.
Chan, E.S. and Melville, W.K. 1988 Deep-water plunging wave pressures on a vertical
plane wall. Proc. R. Soc. Lond. A 417, 95-131.
Cooker, M.J., and Peregrine, D.H. 1991 Violent surface motion as near-breaking
waves meet a wall. pp. 291-297 in Breaking Waves eds., M.L. Banner and R.H.J.
Grimshaw, Berlin, Springer 387 pp.
Jiang, L., Perlin, М., and Schultz, W.W. 1998 Period tripling and energy dissipation
of breaking standing waves. J. Fluid Mech. 369, 273-299.
Longuet-Higgins, M.S., and Oguz, H.N. 1997 Critical jets in surface waves and col­
lapsing cavities. Phil. Trans. R. Soc. Lond. A 355, 625-639.
Longuet-Higgins, M.S. and Dommermuth, D.G. 1996 Crest instabilities of gravity
waves. Part 3. Nonlinear development and breaking. J. Fluid Mech. 336, 33-50.
Longuet-Higgins, M.S. 2000 Theory of water waves derived from a Lagrangian. I.
Standing waves. J. Fluid Mech. 423, 275-291.
Longuet-Higgins, M.S. 2001 Vertical jets from standing waves in deep water. Proc. R-
Soc. A 457, 495-510.
Longuet-Higgins, M.S. and Dommermuth, D.G. 2001 a Vertical jets from standing
waves. II (submitted).
Longuet-Higgins, M.S. and Dommermuth, D.G. 2001b On the breaking of standing
waves (submitted).
Mercer, G.N. and Roberts, A.J. 1992 Standing waves in deep water: Their stability
and extreme form. Phys. Fluids A 4 , 259-269.
Prosperetti, A., and Oguz, H.N. 1997 Air entrainment upon liquid impact, Phil. TYans.
R. Soc. Lond. A 355, 491-506.
Zeff, B.W., Kleber, B., Fineberg, J., and Lathrop, D.P. 2000 Singularity dynamics in
curvature collapse and jet eruption on a fluid surface. Nature 403, 401-404.
 2253

ШЩ THE ROYAL
10.1098/rspa.2001.0832 Ш 1 SOCIETY

Vertical jets from standing waves. II

B y M i c h a e l S. L o n g u e t - H i g g i n s 1 a n d D o u g l a s G. D o m m e r m u t h 2
1Institute for Nonlinear Science, University of California, San Diego,
La Jolla, CA 92093-0402, USA
2 Naval Hydrodynamics Division, Science Applications International Corporation,
10260 Campus Point Drive, San Diego, С A 92121, USA

Received 6 March 2001; accepted 3 April 2001

In a previous paper it was shown that in steep standing waves on water the collapse
of a rounded cavity in the wave trough can produce quite high local vertical acceler­
ations, initiating the growth of a strong vertical jet. In one example given here, the
acceleration exceeds lOOp. The main purpose of the present paper is to follow the
subsequent development of the jet by means of a boundary-integral time-stepping
technique. It is found that the jets tend to bifurcate into two halves, each forming
a plunging or spilling breaker. The transition of the rising wave trough into a thin
jet is here compared with an asymptotic flow in which the free surface is given by a
quartic equation.
K e y w o rd s: w ater w aves; su rfa ce w aves; je t form ation ; wave break in g;
sta n d in g w aves; v e rtical acceleration

1. Introduction
In a recent paper (Longuet-Higgins 2001a, hereafter referred to as paper I) it was
shown th at steep standing waves on the surface of deep water can, in some circum­
stances, produce remarkably large vertical accelerations, giving rise to vertical jets
of water arising out of the wave troughs. A similar phenomenon could occur through
the impact of a water wave against a vertical wall or cliff face or against the side of
a ship in deep water. A sufficient condition for moderately large accelerations is that
the wave trough have the form of an almost circular arc, as in a shaped charge.
The previous calculations were carried out using a new method due essentially to
Balk (1996), which is particularly convenient for parametrizing the initial conditions
for the flow. The examples given in paper I showed th at the resulting vertical accel­
erations of a particle in the wave trough could exceed 10 <7, before falling off rapidly
to smaller values.
After the acceleration had fallen to near zero, the tip of the jet began to develop
a sharp curvature. For the mathematical description of such a sharp corner, Balk’s
analysis, which is in terms of a Fourier series, is not so well adapted. In fact the
calculations tended to lose accuracy slightly before the sharp corner was formed.
The purpose of the present paper is to continue and extend the previous results by
making use of another but complementary method of numerical time-stepping, which
was devised for the study of breaking waves by Longuet-Higgins &; Cokelet (1976).
This boundary-integral technique has been further developed by many authors: see,

Proc. R. Soc. Lond. A (2001) 457, 2137-2149 © 2001 The Royal Society
2137
2254

2138 M. S. Longuet-Higgins and D. G. Dommermuth

for example, Vinje & Brevig (1981), Dold & Peregrine (1986), Roberts (1983) and
Mercer & Roberts (1992). In the present paper we shall make use of an accurate
version of Vinje & Brevig’s formulation, which has been used previously to follow
the nonlinear development of steep gravity waves when subjected to a normal-mode
perturbation (Longuet-Higgins & Dommermuth 1997). It can equally well be used
to study the time-history of any space-periodic surface waves, and in particular to
follow the standing-wave jets th a t concern us here.

2. Initial conditions
In paper I the horizontal and vertical coordinates (ж, у) of a point on the free surface
were given in general by equations of the form
N
x = £+ n - 1a„ sin n f ,
n=1
(2 .1)
N
У = Уо + n~l an cos n f ,
n= 1

where £ is a param eter running from 0 to 27r over one wavelength L = 27Г, and the
coefficients an(t) are functions of the time t only. Moreover, to conserve the total
mass of fluid,

» - - £ s “2 <22)
n=l

In the present system, the initial conditions are specified by the values of the
coordinates (x ,y ) of each point on the surface at time t = 0 , together with the
velocity potential at the free surface. In all of the examples considered here, the
initial velocity is zero. Hence we may simply take
d „(0 )= 0 , n = l, 2 (2.3)

The particular examples chosen in paper I had very simple initial conditions. In the
first four examples А, В, С and D we took N = 2 . The initial values of the coefficients
a\ and a 2 were given in table 1 of paper I, and will be restated in the figure captions
below. They correspond to cases where the total width 2x\ of the wave trough is
given by
x i /тг = 0.25, 0.20, 0.15 and 0.10, (2.4)
respectively. The time t is measured in units in which the wavenumber к and the
acceleration due to gravity g are both equal to 1 .

3. Results
Figure 1 shows the development of the surface profile in case A , x i/n = 0.25, from
t = 0-2.4 at intervals of 0.2. A little more than half a wavelength is shown. W hen
t = 0, the wave trough between x /L = 0.375 and 0.625 is very nearly circular. The
‘crest’ arising out of the wave trough first develops a sharp corner at around t = 1.0
(the limit of the previous calculations). At around t — 1.2 a small protrusion arises

Proc. R. Soc. Lond. A (2001)

 2255

Vertical jets from, standing waves. II 21

0.3 r

—0 з Г-i. i , l i I i i i i I i i i i I i i i i I i i » i I i i i i
-0.5 -0 .4 -0.3 -0.2 -0.1 0 0.1
xlL
Figure 1. Profiles of the free surface at times t — 0 (0.2) 2.4 in case A,
x \ / L = 0.25. Initial values: ai = 0.6879, аг = 0.2379 and di = аг = 0.

Figure 2. Continuation of figure 1, (case A) f = 2.4 (0.2) 4.0.

out of the crest and then becomes sharper (t = 1.4 and 1.6), while the main body
of the crest thickens and continues to rise. By the time t = 2.4 it appears that the
sharp tip of the jet is being reabsorbed by the main body.
In figure 2 it can be seen that after t = 2.4 the tip is completely reabsorbed, while
on each side the surface overturns as in a plunging breaker. The calculation could

Proc. R. Soc. Lond. A (2001)

2256

2140 M. S. Longuet-Higgins and D. G. Dommermuth

0.3

0.2

0.1

- 0.

-о.:

-о.:
-0 .5 -0 .4 -0 .3 -0 .2 -0.1 0 0.1
xJL
Figure 3. Profiles of the free surface at times t = 0 (0 .2) 2.8 in case B,
X i/L = 0.20. Initial values a i = 0.7815, a2 = 0.2764 and di = d2 = 0.

not be continued much beyond t = 4.0, owing to the sharp curvature at the tips of
the two breakers.
The corresponding acceleration у of the particle at the central point (y /L = 0.5)
is shown in figure 12, curve A. The acceleration у reaches a peak of about 2 .8*7
at around t = 0.75, when the free surface is still nearly flat locally, and then falls
abruptly to —g, when the sharp-pointed tip is formed. In other words, th a t part
of the fluid is in free-fall. The acceleration remains at —g until t = 2.5, when the
sharp tip has been reabsorbed into the main jet. The acceleration remains negative,
however, until after t = 4 .0 .
Next consider case B, when x \fL — 2.0; the initial width of the trough is only
one-fifth of the wavelength. Figure 3 shows the jet developing more rapidly than in
case A, but otherwise in a similar way. The sharp tip that is formed between t = 0.8
and 1.0 rises higher, and when it falls back into the main body it produces a small
circular cavity, shown enlarged in figure 4. This reabsorption of the mini-jet can be
viewed as the same process as the ‘shaped charge’ effect, but in reverse.
The vertical acceleration for case В is shown in figure 12, curve B. This tim e у
rises to almost bg at around t = 0.71, before falling to —g and remaining there till
after t = 2 .8 ; throughout this time the tip of the jet is in free-fall.
The very small scale of the features near the cavity in figure 4 prevents the com­
putation in this case being carried beyond about t = 3.1. However, on the reasonable
assumption th at this small part of the flow does not significantly affect the devel­
opment of the rest of the profile we may cut out a small central section of total
width 0 .01 , say, and splice both у and ф across the gap, using a simple interpola­
tion formula, before continuing the time-stepping. This procedure yields figure 5.
An enlarged view (figure 6 ) shows the detail of one of the overturning jets for two
different grid solutions.

Proc. R. Soc. bond. A (2001)

 2257

Vertical jets from standing waves. II 2141

xIL
Figure 4. Enlargement of the top of the jet in figure 3 at time t = 2.8 (0.1) 3.1.

Figure 5. Continuation of figure 3 after cutting and splicing a

narrow central section (|x| ^ 0.1).

In case C, when x i/L = 0.15, the corresponding profiles are shown in figures 7-10.
From figure 7 we see th at the tip of the jet is reabsorbed at around t = 2.0. The
maximum vertical acceleration, from figure 12 , curve C, is now greater than 8 g.
Again, to carry the computation beyond t = 2 it was necessary to cut and splice a
narrow section of the profile including the small cavity, as in figure 8 . The resulting

Proc. R. Soc. Lond. A (2001)

2258

2142 M. S. Longuet-Higgins and D. G. Dommermuth

xJL
Figure 6. Enlargement of the overturning jet in figure 5 (case C) at t — 3.8
(-------, 256 p oints;------ , 512 points).

Figure 7. Profiles of the free surface at times t = 0 (0 .2) 2.0 in case C,

x \ / L — 0.15. Initial values: ai = 0.8779, a-i = 0.3112 and di = аг = 0.

profiles are shown in figures 9 (centre rising) and figure 10 (centre falling). The side
jets clearly become very sharp pointed and thin.
Case D, when X \ / L = 0.10, shows a somewhat different behaviour. Figure 11
suggests th at the sharp tip does not have time to develop before it is overtaken by
the main body of the jet. The absence of this feature enables us to continue the

Proc. R. Soc. Lond. A (2001)

 2259

Vertical jets from standing waves. II 214'

xIL
Figure 8. Splicing the central section of figure 7 at t = 2.0 (------, original;-------, splice).

Figure 9. Continuation of figure 7 (case C) after splicing.

calculation until the plunging breakers on each side develop very sharp corners. This
time the vertical acceleration у rises to over Юр.
Figure 12 combines and compares the acceleration curves for all the cases A-D
considered so far. It is essentially an extension of fig. 8 in paper I to negative values
of у (and larger values of t) and confirms very satisfactorily the agreement between
the two quite different methods of numerical calculation.

Proc. R. Soc. bond. A (2001)

2260

2144 M. S. Longuet-Higgins and D. G. Dommermuth

Figure 10. Continuation of figure 9 (case C).

xIL
Figure 11. Profiles of the free surface at times t = 0 (0.5) 3.0 in case D, x i / L = 0.10.
Initial values: ai = 0.9701, a -2 = 0.3270 and di = d2 = 0.

4. Transition to a jet
One of the most interesting features of the profiles in figures 1 , 3, 7 and 11 is the
smooth transition of the rising trough from a sharp corner, with slope angle less than
45°, to a more narrow, sharp-tipped jet. In figure 7, for example, this occurs between

Proc. R. Soc. Lond. A (2001)

 2261

Vertical jets from standing waves. II 2145

Figure 12. The vertical acceleration as a function of the time in cases A-D.

x
Figure 13. Intermediate profiles in case С at intervals of A t = 0.02,
showing the transition from a sharp corner to a jet.

t = 0.6 and 0.8. An enlargement, with intermediate profiles at intervals A t = 0.05, is

shown in figure 13. It is notable that there is no discontinuity at the instant when the
angle of maximum slope is 45°, as there is in a Dirichlet hyperbola (Longuet-Higgins
1972). Instead, the sequence of profiles more closely resembles one of the canonical
forms for the development of a shaxp-pointed jet which have been derived in a recent

Proc. R. Soc. Lond. A (2001)

2262

2146 M. S. Longuet-Higgins and D. G. Dommermuth

x
Figure 14. Profiles given by the asymptotic expression (4.1) when 7 = —6 ,
at intervals of A t = 1.0.

paper (Longuet-Higgins 2001b). These are given by the quartic equation

x 2 = ± 5 t~ 2 - (y3 + 9x2y ) /t + [(77 - § )s4 - 9(27 + 1)x2y 2 - (7 + | ) 2/4] / t 2, (4.1)
where x and у are horizontal and vertical coordinates in a free-fall frame of reference,
and 6 = ±1 and 7 is an arbitrary dimensionless constant. Expression (4.1) is valid
asymptotically for small values of x /t and у ft. The scales of length and time, and
the value of 7 , are chosen so as to match the outer flow. An example is shown in
figure 14, where 7 = —6.0. A uniform vertical velocity V = 0.05 has been imposed.
To make figures 13 and 14 directly comparable, the profiles in figure 13 have been
replotted in an inertial frame of reference, by giving each curve a uniform vertical
displacement \g { t —t i ) 2, where g — 1 and t\ = 0 .6 .
Similar comparisons can be made between equation (4.1) and the profiles in fig­
ures 1, 3 and 11 at the appropriate times.

5. Example of an extremely high acceleration

So far we have assumed the maximum number N of harmonics in our representation
(2 .1) of the initial conditions to be just 2 . Consider, however, a case when N = 3,
namely let
ei = | , «a = - § , 03 = - § , (5 1 )
with a n (0) = 0, n = 1,2,3, as before. Figures 15 and 16 show th at the initial wave
trough is very flat, with two rather sharp corners at each end. Each corner is roughly
the arc of a circle. When t > 0, each corner collapses at first independently of the
other, but then the two resulting jets converge to form a smaller semicircular trough
near the centre. This gives rise to extremely high accelerations, у exceeding 400#
(see figure 17). The calculations cease at around t = 0.54 = tc, say. A logarithmic

Proc. R. Soc. bond. A (2001)

 2263

Vertical jets from standing waves. II 2147

x/L
Figure 15. Profiles of the free surface at times t = 0 (0.1) 0.5 with initial
conditions ai = | , 02 = — аз = —| and di = <22 = аз = 0.

Figure 16. Enlargement of the central portion of figure 15.

plot is interesting. In figure 18 we have plotted \n(y/g) against In |£c —1\. It can be
seen that the plot is linear over l | decades, showing that
y/ g oc |<c - t f (5.2)
over this range of t. This type of power-law behaviour has been shown to be typical
of free-surface flows that are close to being critical, in the sense that they almost
pinch off a ‘bubble of air’ (see Longuet-Higgins & Oguz 1997).

Proc. R. Soc. Lond. A (2001)

2264

2148 M. S. Longuet-Higgins and D. G. Dommermuth

t
Figure 17. The vertical acceleration of a particle on the central line in
figures 15 and 16, as a function of t.

t
Figure 18. Logarithmic plot of y / g against 11 — t c \, where t c = 0 .54.

6. Conclusion
When a standing gravity wave is allowed to develop from a rounded cavity, the
vertical acceleration may attain quite high values compared with g. However, in the
su sequent development of the jet the accelerations are relatively small and the tip
о the jet is practically in free-fall. In the examples given here the jet bifurcates into

Proc. R. Soc. Lond. A (2001)

 2265

Vertical jets from standing waves. II 2149

two plunging or spilling breakers falling away to either side. As shown elsewhere
(Longuet-Higgins & Dommermuth 2001), other initial conditions can give rise to
single, sharp-pointed jets, which relapse into the surface and form a temporary cavity:
a time-reversal of the cases discussed here. Both types of behaviour seem to have
been observed by Jiang et al. (1998) in their experiments on subharmonically forced
standing waves. However, a full dynamical explanation of these authors’ observations
has yet to be given.
The calculations in §§3-5 were presented at the IUTAM Symposium on Free-Surface Flows,
Birmingham, UK, in July 2000. M.S.L-H. is supported by the Office of Naval Research under
contract N00014-00-1-0248. D.G.D. is supported under contract N00014-97-C-0345.

References
Balk, A. M. 1996 A Lagrangian for water waves. Phys. Fluids 8 , 416-420.
Dold, J. W. & Peregrine, D. H. 1986 An efficient boundary-integral method for steep, unsteady
water waves. In Numerical methods for fluid dynamics (ed. K. W. Morton & M. J. Baines),
vol. II, pp. 671-679. Oxford University Press.
Jiang, L., Perlin, M. & Schultz, W. W. 1998 Period tripling and energy dissipation of breaking
standing waves. J. Fluid Mech. 369, 273-299.
Longuet-Higgins, M. S. 1972 A class of exact, time-dependent free-surface flows. J. Fluid Mech.
55, 529-543.
Longuet-Higgins, M. S. 2001a Vertical jets from standing waves. Proc. R. Soc. Lond. A 457,
495-510.
Longuet-Higgins, M. S. 20016 Asymptotic forms for jets from standing waves. J. Fluid Mech.
(In the press.)
Longuet-Higgins, M. S. & Cokelet, E. D. 1976 The deformation of steep surface waves on water.
Proc. R. Soc. Lond. A 350, 1-26.
Longuet-Higgins, M. S. & Dommermuth, D. G. 1997 Crest instabilities of gravity waves. Part 3.
Nonlinear development and breaking. J. Fluid Mech. 336, 33-50.
Longuet-Higgins, M. S. & Dommermuth, D. G. 2001 On the breaking of standing waves by
falling jets. Phys. Fluids 13, 1652-1659.
Longuet-Higgins, M. S. & Oguz, H. N. 1997 Critical jets in surface waves and collapsing cavities.
Phil. Trans. R. Soc. Lond. A 355, 625-639.
Mercer, G. N. & Roberts, A. J. 1992 Standing waves in deep water: their stability and extreme
form, Phys. Fluids A 4, 259-269.
Roberts, A. J. 1983 A stable and accurate numerical method to calculate the motion of a sharp
interface between fluids. IMA J. Appl. Math. 31, 13-35.
Vinje, T. & Brevig, P. 1981 Breaking waves on finite water depths: a numerical study. Report
R-111.81, Ship Research Institute of Norway.

Proc. R. Soc. Lond. A (2001)

2266

J. Fluid Mech. (2001), vol. 447, pp. 287-297. © 2001 Cambridge University Press 287
DOI: 10.1017/S0022112001006061 Printed in the United Kingdom

A sym ptotic forms for jets from standing waves

By M I C H A E L S. L O N G U E T - H I G G I N S
Institute for Nonlinear Science, University o f California, San Diego, La Jolla, CA 92093-0402, USA

(Received 15 February 2001 and in revised form 1 June 2001)

Standing gravity waves forced beyond the maximum height for perfect periodicity
can produce vertical jets with sharp-pointed tips. In this paper, canonical forms for
the wave crests are derived which display sharp cusps in the limit as the time t tends
to infinity. The theoretical profile is in general quartic in the space coordinates, and
can describe the smooth transition of a fairly low wave crest to a cusped form. There
is no singularity as the surface slope passes through 45°.

1. Introduction
The forming of vertical jets in standing waves and in steep progressive waves
meeting a vertical wall has been studied in several recent laboratory experiments (Chan
& Melville 1988; Jiang, Perlin & Schultz 1998; Bredmose et al. 2000) and numerical
com putations (Cooker & Peregrine 1991; Longuet-Higgins 2001a,b; Longuet-Higgins
& D om m erm uth 2001). While the early, highly accelerated phase of the m otion has
received some attention and a certain degree of explanation (Longuet-Higgins &
Oguz 1997; Cooker 2000; Longuet-Higgins 2001a,b), later stages of the flow, when
the vertical acceleration at the crest has fallen almost to —g and the tip of the jet is in
free-fall, have yet to be represented analytically, even when the influences o f surface
tension and viscosity are theoretically absent.
Both the experiments and the more recent numerical calculations, especially those
by Longuet-Higgins & Dommermuth (2001), strongly suggest that the tip o f the
jet can tend towards a sharp cusp with increasing time t. This implies the presence
of cubic or higher-order terms in the expression for the velocity potential. Such an
analytic form has been suggested previously for a plunging breaker in a progressive
gravity wave (Longuet-Higgins 1980, §10). Here we shall apply the same type of
analysis to the simpler case of a symmetrical jet produced by a standing wave.
In all of the present calculation it is assumed that the flow is inviscid and irrotational
and that surface tension can be ignored. Moreover the motion will be viewed in a
free-fall frame of reference, so that there is no local effect of gravity. The solutions
are to be valid asymptotically near the tip of the jet and as the time t tends to infinity.
As far as fourth order in the radius r, the procedure yields a family o f canoni­
cal solutions with one free parameter; see §4. These are illustrated by numerical
computations in § 5, and a discussion follows in § 6 .

2. General solution
Take rectangular coordinates as in figure 1, with the x-axis horizontal and the y-axis
vertically upwards; r and 9 are polar coordinates such that (x,y) = r(sin 0, —c o s 6).
 2267

288 M. S. Longuet-Higgins

F ig u r e 1. Rectangular and polar coordinates in a free-fall reference frame.

Consider a velocity potential

ф = \A r 2 cos 20 4- \В гг cos 30 + \ Cr4 cos 40, (2.1)
where А, В, С are functions of the time t. This expression is to represent a free-surface
flow valid in the limit as r /t —> 0. In fact we shall take A = 1/f, with В and
С proportional to higher inverse powers of the time t. The reference frame being
inertial, the pressure p is given by

- P= 0 , + ^ 2- m (2.2)
where q is the particle velocity and F an arbitrary function of t. On substitution from
(2.1)

—p = r2(\A 2 + \A cos 29) + r3(AB cos 0 + \B cos 30)

+ r \^ B 2 + AC cos 29 + i С cos40) - F, (2.3)

terms of higher degree in r being neglected. Similarly for the rate of change of p
following a particle we find

- 5 ? = r2[2AA + (\A + A3) cos 20]

+ r 3 [2(A2B -f-AB + A B )cos0 + (2A2B + ±B )cos30]

+r*[2BB + {4AB2 + 2AC + ЛС) cos 20

+(5Л2С + ЛЯ 2 + |С ) cos 40] - F . (2.4)

On the free surface, both p and Dp/T)t are to vanish simultaneously. This can be
ensured by specifying that

g = p ^ (1 + r3R + r4S + 0 (r5)), (2.5)

2268

Asymptotic forms for jets from standing waves 289

where R and S are functions of 0 and t only. By symmetry we may assume that
R = Ri cos 0 + Дз cos 30, 1
( 2 .6)
S = So + S2 cos 26 + S4 cos 49, J
where the Rn and Sn are functions of t only.
On equating coefficients of r2 and r 2 cos 26 in (2.5) we obtain
2AA _ jA + A3 _ F
\A 2 " \A ~ F’ ( ’
of which the general solution was derived by Longuet-Higgins (1972). The particular
solution in which we are interested here is

A = ±, F= J , (2.8)

where F0 is an arbitrary constant. Note that the first term in equation (2.1) represents
a stagnation point flow:

Ф= \ ( у * ~ Л (2.9)
which from (2.3) has a real free surface
x2 = tF = Fo/t3 (2.10)
consisting of two parallel vertical planes, but only if F0/ ( 3 > 0.
Next, on equating coefficients of r 3 cos 9 and r 3 cos 39 in equation (2.5) we obtain
2(А2В + A B + AB) + FRi 2A2В + \B + FR3 _ F „
AB \B F'
These constitute two equations for Ry and R3 in terms of A, В and F, namely
FRi = AB(F/F) - 2(A2B + A B + AB), 1
(2. 12)
FRi = \B (F /F )- (2 A 2B + \B ). J
For example, if we choose В = B0/ t 2 where BQ is a constant we find

(2.13)

and hence
«1=0, R3 = ^ . (2-14)
3Fo
Smiilarly by equating coefficients of r 4,r 4 cos20 and r 4 cos 40 in equation (2.5) we
can obtain three equations for S0, S2 and S4 which, if С = Q / t 3, yield
с _ s _ B l-C % Bq + 5C0 p t <5)
S° - 2 To’ S2~ — F T ' S* - — * F T - {
Note that as t increases the term in r 3 in equation (2.5) will come to dom inate the
series. The term in r 4 is relatively small, since by hypothesis r /t <C 1, and similarly
for terms of higher order in r.
 2269

290 M. S. Longuet-Higgins
3. The free surface
The free surface p = 0 can be found by equating to zero the right-hand side of
(2.3). If we now substitute

A -- B -M
A~ t< B~ fi’ cr ~-fi
- 4y НП
(3-1)
in this expression we obtain

^ ( 5 - 5 co s 20 )
j.3
+ (3/? cos 9 — cos 30)

F0
t*K2r ' ' 7 t4
On replacing 1 by (cos2 0 + sin 2 0)” where appropriate we get
r2
sin 2 0
t2
r2
4- -у (P cos 3 0 + 9p cos 0 sin2 0)

Fo
+ 7 ЩР2 + У)cos4 0 + 9(p2 + 2y) cos2 9 sin2 9 + ( \p 2 - ly) sin4 9] = (3.3)

Now writing
(Z, П) = W t , y /t) = (r/0(sin 9, - cos 9) (3.4)
we have

e - pin' + П гп) + [(§/?2 + y)n4 + 9(P2 + 2 y K V + (5/32 - Ш 4) = <f, (3.5)

where
e = F0/ f 4. (3.6)
The left-hand side of equation (3.5) being quadratic in £2, with coefficients which
are polynomials in rj, for any given rj we may solve for £2. If either of the roots is
non-negative we can then plot £ versus rj.
Equation (3.5) is asymptotically valid for sufficiently small values of £ and t]t that is
to say for sufficiently large values of t when x and у are bounded, and for extremely
small values of e. Near the origin of (£, tj), where the quartic terms are relatively
small, we have approximately
{ 2 = e + « i 73 + 9{2rt. (3.7)

In the limit as e —> 0 this curve (or surface) has an upward- or downward-pointing
cusp according as p is negative or positive 0. If e < 0 the curves p = 0 lie inside the
cusped curve and if € > 0 they lie outside. Since e — Fo/14 —►0 as t —►oo we see that
the limiting curves are in general cusped.
Note that if all the fourth-order terms in equation (3.5) are neglected we obtain the
lower-order approximation
Z2 - p ( r i 3 + 9 e r i ) = e (3.8)
2270

Asymptotic forms for jets from standing waves 291

F igure 2. Surface profiles corresponding to equation (4.1) when S = —1 and V = 1/9.

with solution
к2 _ С + Й 3 (3.9)
i - w
The corresponding profile will have a horizontal asymptote given by rj = (9f i r 1-

4. Canonical forms
W hen fi < 0, giving an upward-pointing cusp when e —►0, we may by a suitable
choice of velocity scale choose = —1. (Setting fi = +1 would simply reflect the
profile in the horizontal axis, i.e. turn it upside down.) Then by a further choice of
time scale we may set e = +t~4. Thus equation (3.8) is reduced to the canonical form
x2 = St~2 —(y3 -f 9x2y )/t, (4-1)
where <5 = +1. Similarly equation (3.5) is reduced to
** = s r 2 - (y3 + 9x2y ) / t + [(7y - \ )x4 - 9(2y + l ) x V - (7 + | ) / ] / Л (4.2)
Exceptionally we may have /? = 0, and then by choice of scales
x 2 = д Г 2 + <5;(7x4 - 1 8 х У - у4)/? . (4.3)
As always, these expressions are valid only when x /t and y / t are sufficiently small.

5. Numerical calculations
Consider first the approximation (4.1). The case S = —1 is shown in figure 2. For
clarity, the profiles have been separated by adding successive vertical displacements
At/9, where At is the time interval between profiles. Physically this is equivalent to
 2271

292 M. S. Longuet-Higgins

Figure 3. Surface profiles corresponding to equation (4.1) when <5 = +1 and V = 1/9.

Figure 4. The inclination a to the horizontal of the asymptotes in the profile (4.2),
as a function o f у (or y/fi2).

imposing a uniform vertical velocity V = 1/9. The effect is to maintain the horizontal
asymptote at the same level у — 0 .
As t increases, the central part of the profile approaches the form of a semi-cubical
parabola, with increasing accuracy.
The case S = +1 is shown in figure 3. The most noticeable difference between
these profiles and those of figure 2 is that they can become very steep, indeed almost
vertical, at points other than the central cusp. If a vertical velocity were not imposed,
these curves would all be nested inside each other.
In order to classify the profiles corresponding to equation (4.2) it is convenient to
2272

Asymptotic forms for jets from standing waves 293

F igure 5. (a) Surface profiles corresponding to the canonical form (4.2) when 8 = —1 and у = —5.
The time interval At between successive profiles is 2. To separate the curves a vertical velocity
V = 0.06 has been imposed, (b) Extension o f (a) but with At = 10 and V = —0.015.

consider the asymptotes for large values of x and у (but small values of x / t and
y /t). The directions of the asymptotes are found by equating to zero the fourth-order
terms, hence

S / у2 = [9(2? + 1) + 7{81(2y + l )2 + (14? - 9)(2y + 9)}]/(14y - 9). (5.1)

The angle o f slope
a = | arctan(y/x)| (5-2)
is plotted against у (or more generally y/fi2) in figure 4. In a central range, given by
 2273

294 M. S. Longuet-Higgins

F ig u r e 6. Surface profiles corresponding to the canonical form (4.2) when <5 = -f-1 and у = —5.
Time interval At = 10. Imposed velocity V = 1/9.

R gure 7. Surface profiles corresponding to the exceptional case 0 = 0, equation (4.3), when
<5 = —1,5' = —1. Time interval between profiles: At = 10. Imposed velocity V = 0.

—1.227 < у < 0.643, there are no real asymptotes. On the other hand as у —►±oo, the
angle у tends to a finite value: a = 31.68° (see below).
A typical example, when у = —5, is shown in figures 5(a) and 5(b). In figure 5(a),
in order to separate the curves, a vertical velocity V = 0.05 has been imposed. The
2274

Asymptotic forms for jets from standing waves 295

F igure 8. Numerically calculated profiles of wave crests generated at time t = 0 by imposing the
vertical velocity (6.1) on a flat water surface. Here С — 1. Enlargement o f figure 9 o f Longuet-Higgins
& Dommermuth (2001), seen in a free-fall reference frame.

successive profiles seem to represent the transition of a rounded wave crest into an
upwards vertical jet. It is noticeable that the maximum slope of each curve passes
through 45° without any singular behaviour with respect to the time t.
The sequence is extended to larger times t in figure 5(b), and here, for the sake of
clarity we have imposed a uniform downwards velocity V = —0.01. As t —►oo we see
the development of a sharp cusp.
Similarly when 5 = 1 and у = —5 we obtain the sequence of surface profiles shown
in figure 6 . The curves are steeper, as in figure 4. The profile corresponding to t = 10
is not shown, since that joins with another branch coming from above the crest to
yield an unphysical solution.
When <5 = 1 and у = 5 the curves are again uninteresting.
Lastly we consider the exceptional case /? = 0, given by (4.3). The only physically
interesting case is when 5 = —1. The profiles are then as in figure 7. No vertical
velocity V has been imposed. For large values of x and у the curves approach the
asymptotes
7x4 - 1 8 x V - / = 0, (5.3)
that is to say
Х2/У г = (9 + V§8)/7 = 2.6258 (5.4)
or
y /x = ±0.6171 = ± arctan 31.68°. (5.5)
In each of figures 2, 3, 5, 6 and 7 the necessary condition that x /t and y / t shall be
small can be seen to be satisfied.
 2275

296 M. S. Longuet-Higgins

Figure 9. Enlargement o f the crests in figures 5(a) and 5(b), but with superposed velocity
У = —0.020. These correspond to equation (4.2) with <5 = —l,y = —5. Time interval between
profiles, At = 4.

6. Discussion
Like the limiting Stokes comer flow in progressive surface waves, and its extension
to progressive waves of near-limiting height (Longuet-Higgins & Fox 1977, 1978) the
expressions given above are valid only asymptotically, in this case in the limit as
t —►oo and r /t —►0. The functions B(t) and C(r) in §§2 and 3 are to some extent
arbitrary, and must be determined, in any particular case, by fitting the inner flow
which they represent to the outer flow in the rest of the fluid. The fitting must of
course be carried out in both the space and time domains.
At first sight it appears that by adding suitable terms of order r 5 or higher powers
on the right of equation (2.5) one could proceed indefinitely to higher approximations.
However, there is no guarantee that the process will converge. It may well be more
convenient to adopt a low-order approximation which is certainly valid asymptotically
in the prescribed limits.
There is strong evidence that the asymptotic expressions for the surface profile
given in §§ 3 and 4 above do indeed describe the development of jets in some recent
numerical calculations. For example Longuet-Higgins & Dommermuth (2001) have
shown by precise calculation that if an upwards vertical velocity v of the form
v = С(gk)~l/2 cos kx (6 . 1)
is imparted at time £ = 0 to an infinitely deep body of fluid, where к is a horizontal
wavenumber, g denotes gravity and С is a constant of order 1, then the crest of the
wave so generated tends to develop a sharp upwards jet. Figure 8 is an enlargement
of the wave crest, seen in a free-fall frame of reference; a displacement \g{t —t0)2 has
been added to the computed wave profile, where t0 is taken equal to 2.5. This may be
compared with figure 9, which is an enlargement of the crests in figures 5(a) and 5(b),
but with a superposed uniform velocity V = —0.015. The profiles in figure 9 represent
2276

Asymptotic forms for jets from standing waves 297

equation (4.2) when 3 = —1 and у = —5. The similarity of figures 8 and 9 is striking.
N o special attem pt has been made to select the optimum value o f у for figure 8 .
A further interesting feature of the canonical form (4.2) is that it provides a sm ooth
transition for a wave crest with maximum slope less than 45° to pass to a profile
having maximum slope much greater than 45° (as is shown in figures 5a and 6 , for
example) without having to pass through a time-singularity such as occurs in the
Dirichlet hyperbola (Longuet-Higgins 1972) at time t = 0. For further discussion and
references see Longuet-Higgins & Dommermuth (2001).
The limiting forms described above may apply not only to steep, progressive waves
meeting a vertical wall but also to the bow waves of certain ships.

This research has been supported by the Office of Naval Research under C ontract
N00014-00-1-0248.

REFERENCES
М., P e r e g r i n e , D. H . & T k a is , L. Experimental investigation and
B r e d m o s e , H ., B r o c c h i n i ,
numerical modelling of steep forced water waves. Euromech Colloquium 416, Genova, Italy,
17-20 Sept. 2000.
C h a n , E . S. & M e l v i l l e , W . K . 1988 Deep-water plunging wave pressures on a vertical plane wall.
Proc. R. Soc. Lond. A 417, 95-131.
C ooker, M. J. 2001 Violently erupting free-surface jets. Proc. IUTAM Symp. on Free Surface Flows,
Birmingham, UK, 10-14 July 2000. Kluwer.
C ooker, M. J. & P e r e g r i n e , D. H. 1991 Violent surface motion as near-breaking waves meet a
wall. In Breaking Waves (ed. M. L. Banner & R. H. J. Grimshaw), pp. 291-297. Springer.
J ia n g , L., P e r l i n , M. & S c h u l t z , W. W. 1998 Period tripling and energy dissipation o f breaking
standing waves. J. Fluid Mech. 369, 273-299.
L o n g u e t - H i g g i n s , M. S. 1972 A class of exact, time-dependent, free-surface flows. J. Fluid Mech.
55, 529-543.
L onguet-H iggins, M. S. 1980 On the forming of sharp corners at a free surface. Proc. R. Soc.
Lond. A 371, 453-478.
L o n g u e t - H i g g i n s , M. S. 2001a Vertical jets from standing waves: the bazooka effect. In IUTAM
Symp. on Free Surface Flows, Birmingham, UK, 10-14 July 2000, pp. 195-204. Kluwer.
L o n g u e t - H i g g i n s , M. S. 2001b Vertical jets from standing waves in deep water. Proc. R. Soc. A
457, 495-510.
L o n g u e t - H i g g i n s , M. S. & D o m m e r m u th , D . G. 2001 On the breaking o f standing waves by
falling jets. Phys. Fluids 13, 1652-1659.
L o n g u e t - H i g g i n s , M. S. & Fox, M. J. Н. 1977 Theory o f the almost-highest wave: The inner
solution. J. Fluid Mech. 80, 721-741.
L o n g u e t - H i g g i n s , M. S. & Fox, M. J. Н. 1978 Theory of the almost-highest wave. II. Matching
and analytic extension. J. Fluid Mech. 85, 769-786.
L o n g u e t - H i g g i n s , M. S. & O g u z , Н. N. 1997 Critical jets in surface waves and collapsing cavities.
Phil. Trans. R. Soc. Lond. A 355, 625-639.
 2277

J. Fluid Mech. (2002), vol. 466, pp. 305-318. © 2002 Cambridge University Press 305
DOI: 10.1017/S0022112002001246 Printed in the United Kingdom

On steep gravity waves meeting a vertical wall:

a triple instability
By M I C H A E L S. L O N G U E T - H I G G I N S 1
a n d D A V I D A. D R A Z E N 2
‘Institute for Nonlinear Science, University o f California, San Diego,
La Jolla, CA 92093-0402, USA
2 Scripps Institution o f Oceanography, University o f California, San Diego,
La Jolla, CA 92093-0213, USA

(Received 21 October 2001 and in revised form 20 March 2002)

Theoretical arguments suggest that progressive gravity waves incident on a vertical

wall can produce periodic standing waves only if the incident wave steepness ak
is quite small, certainly less than 0.284. Laboratory experiments are carried out in
which an incident wave train of almost uniform amplitude meets a vertical barrier.
At wave steepnesses greater than 0.236 the resulting motion near the barrier is non­
periodic. A growing instability is observed in which every third wave crest is steeper
than its neighbours. The steep waves develop sharp crests, or vertical jets. The two
neighbouring crests are rounded, flat-topped, or of intermediate form. The instability
grows by a factor of about 2.2 for every three wave periods, almost independently of
the incident wave steepness.

1. Introduction
Surface gravity waves of low amplitude, when reflected from a vertical wall, will
produce standing waves of double the amplitude of the incident waves. If on the other
hand the incident waves are steep enough, they are found to throw up vertical jets
of water against the reflecting barrier; see for example Chan & Melville (1988). The
same phenomenon is often observed when incoming waves meet a cliff or harbour
wall. Moreover, this is not merely a shallow-water phenomenon but occurs also in
deep water. As will be seen below, it follows from a simple energy argument that
progressive waves of more than a certain steepness cannot produce periodic standing
waves; they must be aperiodic.
In his well-known experiments on steep standing waves Sir Geoffrey Tayor (1954)
considered only periodic waves, whose maximum slope he found to be about 45°. In
accurate computations Mercer & Roberts (1992) showed that periodic standing waves
of given wavelength cannot have more than a certain energy, although at energies
slightly below that maximum two different periodic waves having the same energy
can exist. Jiang, Perlin & Schultz (1998) have carried out experiments on deep-water
standing waves forced subharmonically by a vertical oscillation of the wave tank. It
was shown that the waves could be made to break periodically once every three wave
‘periods’. Numerical studies of ‘super-energetic’ standing waves, with various initial
conditions, have been carried out by Longuet-Higgins & Dommermuth (2001a, b)
and it was found that either single jets could be produced, which fell back vertically
into the trough of the wave, creating a semi-circular cavity, or in other cases, starting
2278

306 M. S. Longuet-Higgins and D. A. Drazen

F ig u r e 1. Reflection o f a progressive wave from a vertical wall, when the maximum

surface slope is small.

with a circular cavity, the wave crests could become flat-topped and then break on
either side of the wave crest, like a pair of spilling or plunging breakers.
The purpose of the present paper is to describe experiments in which free progressive
waves are allowed to impinge on a vertical wall, and the complete history of the motion
is then followed. There is no forcing of the wave motion by a vertical or other kind
of oscillation of the boundary, apart from the remote wavemaker. It is found that
when the steepness ak of the incident wave train exceeds about 0.236 the waves in
the neighbourhood of the boundary develop a growing instability in which every
third wave (in time) is the steepest, the two intermediate waves having flat-topped or
rounded crests, or of a mixed type. The steepest waves become sharp-crested; see § 5.
A discussion and conclusions follow in § 6 .

2. E n erg y a n d p erio d icity

Some general conclusions may be drawn immediately from a consideration o f the
total energy density of the waves; see figure 1.
Let Ep denote the mean energy density of the incident progressive wave, averaged
over time and horizontal distance. According to the linearized theory o f surface waves,
in which the surface slopes are small, we have
EP = l>Pga\ ( 2. 1)

where p is the density, g the acceleration due to gravity and a the wave amplitude.
The reflected wave is similar, and the two waves combine to form a standing wave of
maximum amplitude 2 a (crest-to-trough height 4 a) and of time-averaged energy
Es = 2EP. (2.2)
Hence
EP = \E S. (2.3)
According to the linearized theory there is no limit to either Ep or Esy assuming the
wavelength L to be fixed. However from the fully nonlinear theory of gravity waves
the total energy Es of a standing wave has a maximum given by
( £ s )ma* = 0.07774 (2.4)
m units where p,g and the wavenumber к are all unity; see Mercer & Roberts (1992).
Hence, if we assume that all the energy of the incident and the reflected wave goes
 2279

On steep gravity waves meeting a vertical wall 307

ak

Figure 2. Graph of the energy density Ep of a progressive gravity wave of finite steepness ak.

to form a periodic standing wave, and if we ignore the contribution of the higher
harmonics to the total energy (but not to the surface profile), then we must have
Ep ^ 5 С^5)тах — Есгц (2.5)
say, where from (2.4)
Ecru = 0.03887. (2.6)
If on the other hand Ep > Ecrit we see that the resulting motion cannot be periodic.
Now a plot o f the energy density Ep of a progressive wave against its steepness ak
(figure 2 ) shows that a progressive wave may have an energy as great as
(EP) max 0.0745 (2.7)
which is achieved when
ak = 0.429. (2.8)
The maximum wave steepness ak for a progressive wave is 0.4432. So from figure 2
and equation (2 .6 ) there is a certain range of progressive wave steepnesses, namely
0.285 < ak < 0.443 (2.9)
for an incident wave, such that the incident-plus-reflected wave system cannot be a
perfectly periodic standing wave. The resulting motion must be irregular or chaotic
in some way, leading possibly to breaking.
Incident progressive waves whose steepness lies in the range 0 < ak < 0.285 we
shall call subcritical, while those that lie in the range (2 .8 ) we shall call supercritical.
It is not of course implied that all subcritical incident waves will necessarily produce
motions that are perfectly periodic. This is a matter for experiment.

3. Experimental apparatus
The experiments were carried out in a glass-sided wave tank in the Hydraulics
Laboratory at the Scripps Institution of Oceanography. Its dimensions are shown in
figure 3. A computer-controlled, horizontal-movement wavemaker was at A, and an
impervious, sloping beach with gradient 1:10 was at C. A removeable plane wall,
or barrier, was inserted at B. The distances AB and AC were 15.65 m and 26.96 m
2280

308 M. S. Longuet-Higgins and D. A. Drazen

А в с

R gure 3. Sketch o f the wave tank.

F ig u r e 4. Theoretical envelope of the surface elevation in the neighbourhood o f a wave front,

from equation (4.1).

respectively. The width W of the tank and the Stillwater depth h were both equal to
0.50 m.
The vertical displacement of the water surface was measured with resistence-wire
gauges, each consisting of two parallel vertical wires separated by 0.3 cm and oriented
across the tank at B, near the mid-plane of the wave tank. With the barrier in place,
the wires were 2 mm from the surface of the wall. Calibration before and after each
run indicated that the gauges were linear to within about 1 % over the range of
measurement. The output from the gauges was digitized and recorded at a rate of
100 Hz.
A video recording of the surface at a rate of 30 frames/s was made from a position
slightly above the mean water level and to the left (wave side) of the barrier. Blue
vegetable dye was added to the water to increase the contrast between air and water,
the background on the far wall being white.

4. P ro c e d u re
The experiments were conducted at a wave frequency / = 1 . 0 or 1.1s-1, which
is somewhat less than the cut-off frequency (1.25 s"1) for the lowest-order three-
dimensional waves in a channel of width 0.5 m. Nevertheless, after a certain duration
(about 45 s) some higher-order three-dimensional instabilities invariably made their
appearance. The most prominent of these was a cross-wave with wavelength 25 cm
(half the width of the channel) which appeared when / = 1.0 s” 1. Therefore most
experiments were done at / = 1.1 s-1.
 2281

On steep gravity waves meeting a vertical wall 309

a0 a <*s
G (cm) (cm) aa2/g ak (cm)
1.6 4.53 4.22 0.206 0.200 7.50
1.8 5.19 4.63 0.226 0.216 6.95
2.0 5.52 5.66 0.251 0.236 7.87
2.2 5.82 5.51 0.268 0.252 8.49
2.4 6.61 5.85 0.285 0.266 9.28
2.6 7.52 6.36 0.310 0.285 10.89
Table 1. Range o f experiments with X = 0.1, N = 40; ak is determined by the procedure described
in the Appendix.

С
(cm)

-2

~0 10 20 30 40 50
/(s)

R gure 5. Record o f the surface elevation at x = 2.30 m when X = 0.1 and N = 40 (progressive
wave: G = 1.6).

As is well known, if a wavemaker is started from rest, the wave front advances down
the channel with the group velocity cg. If the wavemaker is switched on suddenly at
time t = 0 , then according to linear theory the surface elevation £(x,t) at a horizontal
distance x from the wavemaker is given by
С = iBF{x)eio(t- ° x/g) (4.1)
approximately, where В is a constant depending on the type of wavemaker, a is the
radian frequency of the waves, т is a dimensionless time:

-(± Г ( - т )
which vanishes at the wave front t = 2gx/ g, and F(t) describes the complex wave

ю-i+ArГ*****
envelope:

I 1 -г l Jo (43)
see for example Miles (1962). The function F ( t ) is related to the Fresnel integral
(Abramowitz & Stegun 1964). A sketch of the envelope | F ( t )| as a function of time
is shown in figure 4. The wave amplitude at first increases exponentially, reaches the
value | В at time т = 0 and then oscillates about its final value В as т —►oo. The
2282

310 M. S. Longuet-Higgins and D. A. Drazen

s)

F ig u r e 6. Surface elevation at x = 15.65 m when G = 1.6 (a) with no barrier and (b) with the
barrier in place.

first maximum in the amplitude is about 19% greater than the final amplitude. It has
been shown experimentally (see Longuet-Higgins 1976) that the effect of finite wave
steepness is to increase the effective group velocity so that the wave front arrives
slightly sooner than predicted by the linear theory, and to increase the maximum
wave amplitude considerably.
In the present experiments, in order to suppress the oscillations of the envelope, the
wavemaker was started gradually from rest with a horizontal displacement given by

C= { °’ t< 0 (4.4)
[ С tanh At sin 2n f t, t > 0,
where С and Я are constants. It was found convenient to take Я = 0.1. When Я
was much smaller, the wave train often did not approach its final amplitude (e-2^1
negligible) before the onset of the three-dimensional instabilities mentioned above.
On the other hand, it was found useful to terminate the wave train after a certain
number N of wave cycles, by switching off the wavemaker suddenly. This produced
a corresponding Fresnel pattern of the wave envelope at the rear of the wave train,
including some waves which were steeper than the steady waves.
The input voltage to the wavemaker was governed by a certain gain factor, which
we have denoted by G. Experiments were carried out over the range 1.6 < G ^ 2.6;
see table 1 for the corresponding wave parameters.
 2283

On steep gravity waves meeting a vertical wall 311

10 20 30 40 50 60 70
/(s)

R gure 7. Surface elevation at x = 15.65 m when G = 2.0 (a) with no barrier

(b) with the barrier in place.

At each value o f the gain G, four types of measurements were made. First, the
surface elevation £ was recorded at a point close to the wavemaker (x = 2.30 m) but
still far enough away that local effects were negligible, in general. The barrier at В was
not in place. Second, similar measurements were made at the point B, still without
the barrier, so that the waves passed by as progressive waves. Thirdly the barrier was
inserted, and the surface elevation was recorded at the same point B. In all three
cases the wave gauge was situated on the centreline of the channel. Simultaneously
with the third recording, a video sequence of the waves near the barrier was taken as
described above.
In order to prevent any horizontal displacement of the barrier it was constructed of
1.0 in. laminated plywood, strengthened by angle brackets, and was secured in place
by a steel bar at the top, clamps above the water level and a firm tubular rubber seal
at each sidewall. During the experiment no surface waves on the down-wave side of
the barrier were detected.
A step-calibration of the wire wave gauges was earned out at the beginning and
end of each series of experiments.

5. Results
Figures 5 and 6 show the case G = 1.6. In figure 5 we see the wave amplitude at
x = 2.30 m starting almost immediately to increase monotonically towards the value
312 M. S. Longuet-Higgins and D. A. Drazen

/(s)
R g u r e 8. Surface elevation at x = 15.65 m when G = 2.2 (a) with no barrier
(b) with the barrier in place.

4.5 cm, which is attained after about 25 s. At t = 38 s one can see the maximum T of
the Fresnel envelope created by the abrupt shut-off of the wavemaker after 40 cycles
(36 s).
Figure 6 (a) shows the same progressive wave train on arriving at В (x — 15.65 m).
The final amplitude a is now only 4.2 cm, attained at about t = 35 s. It remains
constant until about t = 42 s, after which it is affected by the Fresnel pattern from
the wave cut-off. From equation (4.2) the width of the Fresnel pattern is proportional
to (2 x /g )1/2 and so is increased over the width at x = 2.30 m by a factor 2.61. When
t > 30 s there are slight indications of a Benjamin-Feir instability, but these are small
compared to the oscillations of the Fresnel envelope. The period of the envelope
oscillations diminishes with distance from T. When t = 43 s their period is less than
two wave periods.
Figure 6(b) shows the same situation as 6 (a) but with the barrier in place. Note the
difference in vertical scales. The wave amplitude a5 is roughly equal to 2a (see table
1) as one would expect from linear theory.
The corresponding three records when G = 1.8 were quite similar to those in
figures 5 and 6 , but with larger amplitudes; see table 1. However when G = 2.0
some qualitatively new features appeared. Figures 7(a) and 7(b) show the records
taken at the point В (x = 15.65 m) without the barrier and with the barrier in place,
respectively. Figure 7(a) shows the usual Fresnel envelope for a progressive wave, with
 2285

On steep gravity waves meeting a vertical wall 313

*(«)
R g u r e 9. Surface elevation at x = 15.65 m when G = 2.4 (a ) with no barrier
(b) with the barrier in place.

the maximum at T. Before t = 40 s there is a slight modulation of the envelope due

either to a Benjamin-Feir instability or to some three-dimensionality in the motion.
Figure 7(b), however, taken with the barrier in place, shows that between t = 40s and
t = 50 s there is apparently a new instability in which every third wave, marked with
the symbol 5, (i = 1 to 4) is higher than its two neighbours.
This is confirmed by figures 8 (a) and 8 (b), taken when G = 2.2. In figure 8 (b),
which shows the surface elevation in the reflected wave, the three-fold pattern now
extends as far as from t = 35 s to t = 50 s. It appears to have overwhelmed the Fresnel
pattern even as far as the maximum T.
Figure 9(b), corresponding to G = 2.4, shows the same pattern extending as far back
as t = 30 s, but by t = 45 s the waves have become chaotic and the Fresnel pattern
is quite ragged. A similar phenomenon is apparent in figure 10(b), corresponding to
G = 2.6. Here the pattern begins and breaks down even earlier.
An examination of the photographic record, see figure 11 for the case G = 2.4,
reveals that the highest peaks in each triplet are always sharp-pointed. The lower
peaks are either round-crested or flat-topped or sometimes have profiles that are
o f intermediate form; see figure 11. After the crest S5, at t = 40s, the motion
becomes markedly three-dimensional, which contributes to the chaotic appearance of
the record of surface elevation.
314 M. S. Longuet-Higgins and D. A. Drazen

FIg u r e 10. Surface elevation at x = 15.65 m when G = 2.6 (a ) with no barrier

(b) with the barrier in place.

6. Discussion and conclusions

A rough measure of the amplitude of the instability noted in figures 7(6) to 10(b) is
the difference AC in crest height between the highest and lowest waves of each triplet.
In figure 12, AC has been plotted against the suffix i in Sf on a log-linear scale, for
each value of G, except that when G = 2.6, i has been increased by 2 to bring the
plots closer together. (This does not of course affect the proportional rate of increase
of A(.) It will be seen that in every case except one, namely i = 1 and G = 2.0, the
plots lie close to the same straight line. This indicates an increase in AC by a factor of
about 2.2 for every three wave cycles, that is an increase of 1.3 per wave cycle. The
exceptional plotted point (shown in parentheses) corresponds to a very small value
of AC, lying within the noise-level of the experiment.
Thus we have detected a subharmonic instability which tends to occur at values of
G greater than about 2.0, that is to say incident wave steepnesses ak ^ 0.236 (see table
1). The observed rate of growth is about 1.3 per wave cycle, practically independent
of the incident wave amplitude.
The above instability is probably related dynamically to the ‘period tripling’
phenomenon observed by Jiang et al. (1998) in forced standing waves. There are
some differences, however.
(1) In Jiang et al. (1998) standing waves were forced subharmonically by oscillating
the wave tank vertically at a frequency twice that of the resulting surface waves. Such
 2287

On steep gravity waves meeting a vertical wall 315

FIg ur e 11. A sequence of frames from the video corresponding to figure 9(b) (G — 2.4), showing
consecutive wave crests between t = 28 s and 42 s. The video was taken at 30 frames/s, and the
timing of each frame is as close as possible to a maximum of the surface elevation shown in
figure 9(b), that is within 1 /6 0 s. Each frame in the left-hand column above corresponds to one of
the maxima marked S, in figure 9(6).
2288

316 M. S. Longuet-Higgins and D. A. Drazen

i
F i g u r e 12. Growth o f the triple-period instability, as measured by the difference AC in
crest-elevation between the highest and lowest waves o f a triplet.

a method of excitation is of course unlikely to be found in nature except, for example,

in an earthquake at sea. In our experiments the ‘quasi-standing’ waves were the result
of the reflection o f free progressive waves from a vertical cliff or wall.
(2) In the experiments of Jiang et al. (1998) a steady state was achieved by balancing
the input of wave energy from the vertical forcing against the loss o f energy due to
wave breaking. In our experiments there was no energy input due to vertical m otion
of the bottom, the loss of energy due to wave breaking was negligible, and the
instability grew in time.
(3) In their experiments the observed sequence of wave crests was: sharp-crested
—►flat-topped —> rounded -* sharp-crested, and so on. In our experiments a sharp-
crested wave was often preceded by a flat-topped wave, though not invariably. Other
types of crest form were also observed, as illustrated in figure 11 .
N ote that some instabilities of periodic standing waves that are subharmonic in
space were found analytically by Mercer & Roberts (1992). Those described here,
however, were subharmonic in time. Recording of the spacial behaviour o f the
instability as a function of the horizontal coordinate x would have involved much
more elaborate instrumentation, particularly since the position of the wave crests,
other than at the wall, appeared to be slightly variable.
In the profiles shown in figure 11 the effects of both viscosity and surface tension
are apparently negligible. Viscous effects at the vertical barrier would tend to be
reduced by the oscillatory nature of the boundary layer. Although surface tension
may affect the wave profiles at smaller scales, at larger scales one can expect Froude
scaling to apply.

M.S.L.H. is supported by the Office of Naval Research under Contract N00014-00-

1-0248, and D.A.D. by the National Science Foundation under G rant О С Е - 9 8 - 1 2 1 8 2
to W. K. Melville. The authors thank C. Coughran, J. Lyons and D. Aglietti for
technical assistance. An account of the experiments was presented at the IN I Program
on Surface Water Waves, Cambridge, England, 13-31 August 2001, and at the
IM A Conference on Wind-over-Waves, Churchill College, Cambridge, 3-5 September
2001.
 2289

On steep gravity waves meeting a vertical wall 317

ak
R gure 13. Graph o f ao2/g against ak for nonlinear progressive gravity waves in deep water.

CD ak (c2 — 1) aa2/g
0.00 0.00000 0.00000 0.00000
0.10 0.14222 0.02042 0.14512
0.20 0.20216 0.04173 0.21060
0.30 0.24877 0.06385 0.26465
0.40 0.28843 0.08674 0.31349
0.50 0.32346 0.11020 0.35911
0.55 0.33958 0.12203 0.38102
0.60 0.35488 0.13384 0.40238
0.65 0.36936 0.14552 0.42311
0.70 0.38303 0.15687 0.44312
0.75 0.39582 0.16767 0.46219
0.80 0.40765 0.17757 0.48000
0.85 0.41839 0.18601 0.49621
0.90 0.42782 0.19211 0.51001
0.95 0.43578 0.19454 0.52056
10.00 0.4432 0.1931 0.5288
T a b le 2. Corresponding values o f ak and aa2/g for deep-water gravity waves of finite amplitude.

Appendix. Determination of the steepness parameter ak

Given the crest-to-trough wave height 2a and the radian frequency a — 2n/T,
where T is the wave period, our problem is to find the wave steepness ak, where к is
the wavenumber.
It is assumed that the waves are effectively in deep water, that is to say if A is the
still-water depth, then is negligible. Now in table 2 of Longuet-Higgins (1975),
the phase-speed с and the quantity
a/n = 2a/L = 2ak (A 1)
are both given as functions of a monotonic parameter со which runs from 0 to 1 as
the wave passes from zero steepness to its limiting configuration with a sharp-angled
crest. See also figure 1 of that paper, where ak and (с2 — 1) are both plotted against
a . From the tabulated entries we may thus obtain the first three columns of table 2
2290

318 M. S. Longuet-Higgins and D. A. Drazen

below. Hence for each value of ak we find the corresponding value of
ae2/g = (c2/g)ak. (A 2)
The values are plotted in figure 13. In the experiments, a2/g is a known constant.
Hence for every value of a we can calculate aa2/g and by interpolation in figure 13
find the corresponding value of ak.

REFERENCES
A b r a m o w itz , M. & S t e g u n , I. A . 1964 Handbook o f Mathematical Functions. Washington, D.C.,
U S Govt. Printing Office, 1046 pp.
C h a n , E. S. & M e l v i l l e , W. K. 1988 Deep-water plunging wave pressures on a vertical wall. Proc.
R. Soc. Lond. A 417, 95-131.
J ia n g , L., P e r l i n , M. & S c h u l t z , W. W. 1998 Period tripling and energy dissipation o f breaking
standing waves. J . Fluid Mech. 369, 273-299.
L o n g u e t - H i g g i n s , M. S. 1975 Integral properties o f periodic waves o f finite amplitude. Proc. R.
Soc. Lond. A 342, 157-174.
L o n g u e t - H i g g i n s , M. S. 1976 Breaking w aves-in deep or shallow water. In Proc. 10th Symp. on
Naval Hydrodynamics, Cambridge, Mass., June 1974 (ed. R. D. Cooper & S. W. Doroff), pp.
597-605. Arlington, VA, Office o f Naval Research, 792 pp.
L o n g u e t - H i g g i n s , M. S. & D o m m e r m u t h , D . G. 2001a On the breaking o f standing waves by
falling jets. Phys. Fluids 13, 1652-1659.
L o n g u e t - H i g g i n s , M. S. & D o m m e r m u t h , D . G. 20016 Vertical jets from standing waves. II. Proc.
R. Soc. Lond. A 457, 2137-2149.
M e r c e r , G. N. & R o b e r t s , A. J. 1992 Standing waves in deep water: their stability and extreme
form. Phys. Fluids A 4, 259-269.
M i l e s , J. W. 1962 Transient gravity wave response to an oscillating pressure. J. Fluid Mech. 13,
145-150.
 Advanced Series on Ocean Engineering — Volume 35

DYNAMICS O F WATER WAVES

Selected Papers of Michael Longuet-Higgins

This is a three-volume selection of classical papers by Michael Longuet-Higgins,

who for many years has been a leading researcher in the fast-developing field
of physical oceanography. Some of these papers were first published in scientific
journals or in conference proceedings that are now difficult to access. All the
papers are characterized by the novelty of their content, and the clarity of their
style and exposition.

The papers are quite varied in their approach. They range from basic theory and
new computational methods to laboratory experiments and field observations.
An overall feature is the frequent comparison between theory and experiment
and the constant attention to practical applications.

Among the many advances and achievements to be found in these three volumes
are: the now generally accepted solution to the longstanding problem of how
oceanic microseisms can be generated in deep water or near steep coastlines;
a theoretical explanation of the strong drifting near the bottom in shallow water;
the first introduction of a boundary-integral technique for calculating free surface
flows; simple analytic expressions for the form and time-development of plunging
breakers; and so on.

The book will be of particular interest to advanced students in ocean engineering;

also more generally to fluid dynamicists and physical oceanographers concerned
with the interaction of the ocean with the atmosphere and with sandy shorelines.