Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.

org/page/policies/terms
DOI:10.1190/1.9781560803737
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms
DOI:10.1190/1.9781560803737

SM
NUMBER 24
GEOPHYSICAL MONOGRAPH SERIES

volume editors
Tijmen Jan Moser
Enders A. Robinson

Wei Liu, managing editor

BASIC WAVE ANALYSIS

Huafeng Liu, Alexey Stovas, and Yunyue Li,

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Front cover: Portrait of Christiaan Huygens painted by Caspar Netscher

DOI:10.1190/1.9781560803737

(1639 –1684) on display in the Museum Boerhaave in Leiden on loan

from the Collectie Haags Historisch Museum. Used with permission.

Back cover: Portrait of Constantijn Huygens (1596– 1687) and his five chil-
dren painted by Adriaen Hanneman in 1640 on display in the Mauritshuis,
The Hague. Used with permission.

ISBN 978-0-931830-56-3 (Series)

ISBN 978-1-56080-372-0 (Volume)

Library of Congress Control Number: 2020934330

Society of Exploration Geophysicists

8801 S. Yale Ave., Ste. 500
Tulsa, OK 74137-3575

# 2020 by Society of Exploration Geophysicists

All rights reserved. This book or parts hereof may not be reproduced in any
form without written permission from the publisher.

Published 2020
Printed in the United States of America
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

This book is dedicated to

Guy A. Navarra, M.D.

Seacoast Medical Associates, Newburyport, MA
and grandchildren
DOI:10.1190/1.9781560803737

Chloe Alexandra Robinson

Bjorn Knowlton Robinson
Ashtyn Adelle Robinson

—Enders A. Robinson

My little nephew
Felix Iskander Henrick Kalkhoven
—Tijmen Jan Moser
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms
DOI:10.1190/1.9781560803737
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Contents
About the Authors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv

Part 1: Velocity Analysis

Chapter 1: Survey of Seismic Imaging. . . . . . . . . . . . . . . . . . . . . . . . . 3

Exploration geophysics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Earthquake seismology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Traveling waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Physical seismology and geometrical seismology . . . . . . . . . 14
DOI:10.1190/1.9781560803737

Tomography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Depth points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Prestack reverse-time depth migration . . . . . . . . . . . . . . . . . 24
Iterative improvement method . . . . . . . . . . . . . . . . . . . . . . . 26
Saddle point or pass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Full waveform inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

Chapter 2: Time, Distance, and Velocity . . . . . . . . . . . . . . . . . . . . . . 39

Time-distance curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Alternative expression for the time-distance curve . . . . . . . . 45
Single horizontal interface . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Critical angle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Interval velocity and average velocity. . . . . . . . . . . . . . . . . . 56
Oblique and conventional root-mean-square velocity . . . . . . 64
Snell’s parameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Linear approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

Chapter 3: Velocity Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

Four important corrections . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Multiple coverage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
Common midpoint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Coherency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

v
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

vi Basic Wave Analysis

Stacking (NMO) velocity . . . . . . . . . . . . . . . . . . . . . . . . . . .100

Dix formula for interval velocity . . . . . . . . . . . . . . . . . . . . .106
Approximation of stacking velocity . . . . . . . . . . . . . . . . . . .113

Part 2: Raypath Analysis

Chapter 4: Traveling Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .121
Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .126
Wave motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .128
Sinusoidal waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .132

Chapter 5: Basic Properties of Waves . . . . . . . . . . . . . . . . . . . . . . . 139

DOI:10.1190/1.9781560803737

Rays and wavefronts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .139

Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .140
Wave equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .144
Telegrapher’s equations . . . . . . . . . . . . . . . . . . . . . . . . . . . .147
Downgoing waves and upgoing waves . . . . . . . . . . . . . . . . .152

Chapter 6: Eikonal Equation and Ray Equation. . . . . . . . . . . . . . . 159

Gradient and directional derivative . . . . . . . . . . . . . . . . . . . .159
Principle of least time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .162
Eikonal equation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .163
Ray equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .168
Diving waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .171
Fresnel zone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .174
Fermat’s derivation of Snell’s law . . . . . . . . . . . . . . . . . . . .177
Huygens’ derivation of Snell’s law . . . . . . . . . . . . . . . . . . . .183
Comparison of Fermat and Huygens . . . . . . . . . . . . . . . . . . .185

Chapter 7: Ray Tracing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

Classical ray tracing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .189
Determination of wavefront curvature. . . . . . . . . . . . . . . . . .200
Propagation and depropagation . . . . . . . . . . . . . . . . . . . . . . .210
Normal moveout velocity . . . . . . . . . . . . . . . . . . . . . . . . . . .221
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Contents vii

Interval velocities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .229

Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .234

Part 3: Waveform Analysis

Chapter 8: Three Prototype Waves . . . . . . . . . . . . . . . . . . . . . . . . . 239

Pioneers of wave study and the invention of radio . . . . . . . .239
Early days of seismic exploration . . . . . . . . . . . . . . . . . . . . .243
Seismic waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .245
Plane waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .251
Cylindrical waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .260
Spherical waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .261
DOI:10.1190/1.9781560803737

Chapter 9: Singularity Functions. . . . . . . . . . . . . . . . . . . . . . . . . . . 265

Dirac delta function and Heaviside step function . . . . . . . . .265
Fourier transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .271
Fourier transforms of delta and step functions. . . . . . . . . . . .275
Fourier transform of a derivative . . . . . . . . . . . . . . . . . . . . .277

Chapter 10: Waves Traveling in Opposite Directions . . . . . . . . . . . 281

To see a star . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .281
Interference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .283
Matter waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .291
Diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .296
Seismic convolutional model . . . . . . . . . . . . . . . . . . . . . . . .303
Huygens’ principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .306
One-dimensional wave equation . . . . . . . . . . . . . . . . . . . . . .311
Initial value problem for the wave equation . . . . . . . . . . . . .317
Two-dimensional wave equation. . . . . . . . . . . . . . . . . . . . . .323
Three-dimensional wave equation. . . . . . . . . . . . . . . . . . . . .325

Chapter 11: Green’s Function for the Wave Equation . . . . . . . . . . 331

Wavefront cones. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .331
Three spatial dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . .333
One spatial dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . .338
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

viii Basic Wave Analysis

Two spatial dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . .341

Green’s function for the wave equation. . . . . . . . . . . . . . . . .346

Chapter 12: Initial and Boundary Conditions . . . . . . . . . . . . . . . . . 351

The convolution integral. . . . . . . . . . . . . . . . . . . . . . . . . . . .351
Source signature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .355
Surface convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .358
Wave equation with initial conditions . . . . . . . . . . . . . . . . . .363
Wave propagation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .364
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .372

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
DOI:10.1190/1.9781560803737

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

About the Authors

Enders A. Robinson is professor emeri-

tus of geophysics at Columbia University in
the Maurice Ewing and J. Lamar Worzel
Chair. He received a B.S. in mathematics
in 1950, an M.S. in economics in 1952, and
a Ph.D. in geophysics in 1954, all from
Massachusetts Institute of Technology. As a
research assistant in the mathematics depart-
ment at MIT in 1950, Robinson was assigned
to seismic research. Paper-and-pencil math-
ematics on the analytic solution of differential
DOI:10.1190/1.9781560803737

equations was expected. Instead, Robinson

digitized the seismic records and processed them on the MIT Whirlwind
digital computer. The success of digital signal processing led to the for-
mation of the MIT Geophysical Analysis Group in 1952 with Robinson as
director. Almost the entire geophysical exploration industry participated
in this digital enterprise. In fact, this effort later would be recognized in
the Boston Globe newspaper, which published a special magazine (May
15, 2011) recognizing the 150 most valuable contributions MIT has made
in science and technology– #32 was the Geophysical Analysis Group for
spurring the “digital revolution” in oil prospecting. In 1965, Robinson and
six colleagues formed Digicon, one of the first companies to do commercial
digital seismic processing. In 1996, Digicon and Veritas combined to form
VeritasDGC, which combined with CGG in 2007.
With Sven Treitel, Robinson received the SEG award for best paper in
GEOPHYSICS in 1964, the SEG Reginald Fessenden Award in 1969, and
the Conrad Schlumberger Award from the European Association of
Exploration Geophysicists (EAGE), also in 1969. In 1983, Robinson was
made an honorary member of SEG. In 1984, he received the Donald
G. Fink Prize Award from the Institute of Electrical and Electronic Engin-
eers (IEEE). In 1988, he was elected to membership in the National
Academy of Engineering. He received the SEG Maurice Ewing Medal
and the SEG award for best paper in GEOPHYSICS in 2001, the Blaise
Pascal Medal for Science and Technology from the European Academy
of Sciences in 2003, and the Desiderius Erasmus Award from EAGE in
2010. Robinson is the author of 20 books and the coauthor of 13.

ix
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

x Basic Wave Analysis

Tijmen Jan Moser received his Ph.D.

from the University of Utrecht, concentrating
on the shortest-path method for seismic ray
tracing. He has worked as a geophysical con-
sultant for a number of companies and insti-
tutes, including Amoco, Institut Français du
Pétrole, Karlsruhe University, Bergen Uni-
versity, Statoil/Hydro, Geophysical Institute
of Israel, Fugro-Jason, and Horizon Energy
Partners. Moser was an Alexander-von-Hum-
boldt stipendiat 1996– 1997. Since 2005, he
has been working independently with SGS-
Horizon, Seismik, and many other associations. Based in The Hague, The
Netherlands, he is close to the former estate of Christiaan Huygens,
whom he considers the greatest geophysicist of all time. Moser’s main inter-
DOI:10.1190/1.9781560803737

ests include seismic imaging, asymptotic methods, seismic reservoir charac-

terization, diffraction, and geothermal exploration. He has authored many
influential papers on ray theory and ray methods, Born inversion and mod-
eling, macro-model independent imaging, and diffraction imaging, several
of which have received best paper awards, including an Honorary
mention in 2005 from SEG and the Eötvös Award in 2007 and 2009 from
EAGE. Moser has co-edited two reprint volumes for SEG, Classical and
Modern Diffraction Theory and Seismic Diffraction, and teaches courses
on diffraction for EAGE and SEG. He co-organizes the APSLIM (Active
and Passive Seismics in Laterally Inhomogeneous Media) workshops in
the Czech Republic (2015, 2021). He is a member of SEG and the Math-
ematical Association of America (MAA), an honorary member of EAGE,
and Editor-in-Chief of Geophysical Prospecting.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Preface
This monograph is devoted to the study of waves. Many natural
phenomena involve waves, and there are three main types: mechanical
waves, electromagnetic waves, and de Broglie waves. Mechanical waves
require a medium through which to travel and examples include seismic
waves, sound waves, water waves, and a vibrating string. Electromagnetic
waves, such as radio waves, microwaves, light, X-rays, and gamma rays,
travel in a vacuum. Recognizing that electromagnetic radiation has charac-
teristics of both particles and waves, Louis-Victor de Broglie of France is
credited as the first to suggest a unifying hypothesis in 1924, proposing
that matter has wave properties as well as particle properties. Even so, the
wavelength of everyday objects is much smaller than that of electrons, so
wave properties have not been detected for them; rather these objects
DOI:10.1190/1.9781560803737

around us show only particle behavior, and de Broglie waves, also known
as matter waves or probability waves, are studied only with regard to suba-
tomic particles.
Perusal of the internet reveals that the study of a single aspect of wave
phenomena can occupy the attention of an entire field of science. Thanks to
the shared concepts that are implicit in the word wave, however, a specialist
in wave propagation in one field can become a specialist in a very different
field surprisingly quickly. For example, a geophysicist can communicate
easily with scientific counterparts in other fields, such as quantum physics,
computer science, digital signal processing, non-destructive testing, radar,
sonar, radio astronomy, and meteorology. The feature in common is basic
wave analysis.
The purpose of this book, Basic Wave Analysis, is to provide infor-
mation required for understanding the fundamental aspects of the elaborate
computer processing schemes prevalent in exploration geophysics. Basic
Wave Analysis does not treat any specialized field, all of which are
covered by the excellent books available in the bookstores of SEG and
EAGE. Basic Wave Analysis has three parts. Part 1 addresses velocity analy-
sis. The correct determination of velocity is the most important problem in
seismic exploration, and an understanding of velocity analysis is a valuable
asset for a geophysicist. Part 2 discusses raypath analysis. Raypaths provide
a geometrical picture of how waves travel, so that a person can visualize
raypaths in their imagination. Geometrical pictures are as important in seis-
mology as they are in optics. Part 3 addresses wavefront analysis. A person
cannot easily visualize traveling wavefronts in his imagination; however,
a computer can follow their motion, and give the geophysicist the final

xi
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

xii Basic Wave Analysis

outcome. Knowledge of wavefront analysis helps a geophysicist understand

many modern computer methods.
We live in a computer age. Computers are very good at doing tedious
repetitive tasks at a rate and scale far beyond what people can do. For
example, robots, automation, and software can replace people in many
aspects of manufacturing. But, although computers may be taught to
“learn,” through such activities as analysis, self-training, observation, and
experience, machine learning and artificial intelligence (AI) have yet to
replace people in all activities. Indeed, the seismic method is faced with
almost insurmountable obstacles in the remote sensing of objects in the
unreachable depths of the earth’s interior. For true understanding, an
image of the subsurface is required, but an accurate image can be very dif-
ficult to produce. Reflected waveforms recorded in a seismic section are
smeared and distorted. The excessive smearing noise lowers the image
quality considerably, often requiring uniform and dense distribution of
DOI:10.1190/1.9781560803737

sources and receivers for the imaging scheme to work at all.

Now suppose that a research geophysicist is assigned the task of writing
a full waveform inversion (FWI) code. Forget about AI and machine learn-
ing. The computer would be an invaluable tool, but in no way can the com-
puter do the thinking required for such a task. The person must depend upon
their own intellect, not upon AI. As a prerequisite, the person would have to
be well versed in geophysical theory conveyed by books and journals.
In the clash between human learning and machine learning, we do
not know what the final result will be. If we take any standard imaging
method, there is no method yet for making use of any sort of machine-
learning process that would automatically transform the existing code into
some more advanced code. For example, the solution to a one-dimensional
geophysical problem for simple structures is generally mathematically tract-
able. The geophysicist would write a simple computer code that would solve
the problem. Ideally, AI would use the simple code and turn it into a three-
dimensional code that would handle arbitrary complex structures, but this
ideal has not been attained yet.
Since the beginning of digital seismic processing in 1950, there have
been tremendous advances in seismic acquisition and computing machines.
The science of geophysics also has undergone major advances, many of
which are associated with the computer as a tool. None of these major
advances has been the result of machine learning. No computer has auto-
matically reprogrammed itself by means of machine learning. All of the
existing codes have been written by geophysicists. As always, fundamental
changes in code must be carried out by skilled geophysicists. Now imagine
that there were no more geophysicists. In their place, however, human
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Preface xiii

operators could continue running the computers to produce subsurface

maps. The operators most likely would become more adept at using the
existing codes, thereby producing better maps. This a process of human
learning, not machine learning.
At some point, people might become satisfied with what they already
have and not strive for new inventions. As a result, inventive science
would disappear. No longer would there be a need for scientific books
and journals. At the moment of disappearance, there well might be a
feeling that this is wonderful. Think of it. Vast computational algorithms
can do whatever you like. No more thinking is needed. But, inevitably,
something would break down. No one would be able to fix it. Things
would not remain the same. They would get worse. The answer for the
future is not that computers will be able to take the place of human thinking.
Rather, as computers become more powerful, geophysicists must become
more knowledgeable. Geophysicists always will be needed to assimilate
DOI:10.1190/1.9781560803737

the existing situation in order to produce better imaging programs.

So, what can be done in geophysics? One option is that geophysicists
use the programs on their computers. The geophysicist (1) enters the input
data, (2) enters parameters by the use of a menu, and (3) interprets the
output. The computer is a black box, with internal structure regarded as for-
bidden territory.
A second option is that geophysicists, in addition, learn about the
detailed workings of the specific programs on their computers. This
approach, however, would be a monumental undertaking because of the
utter complexity of such programs.
A third option is that geophysicists, instead, learn about the basic pur-
poses of the computer programs. Fortunately, this approach is a not a monu-
mental undertaking because it is constructed in terms of basic wave analysis.
As the geophysicist becomes more proficient, the territory of the second
option opens up. We can keep our identity as geophysicists as long as we
can retain our knowledge of basic science. Refinements of these geophysical
tools, refinements of our talents, and refinements of the geological models to
be manipulated by data processing are extending the reach of our unknown
in tremendous strides.
Of course, we honor Galileo and Newton. However, two additional 17th
century scientists must be remembered too. It is Fermat and Huygens who
give us the underpinnings of basic wave analysis. Christiaan Huygens has
provided us with the velocity of light and the optical Doppler effect,
which establishes that light is a moving wave. This is the greatest contri-
bution ever in the field of velocity analysis. Pierre de Fermat has given us
the early version of the principle of least action, which is the greatest
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

xiv Basic Wave Analysis

contribution ever in the field of raypath analysis. In addition, Huygens has

given us Green’s function for the wave equation as well as the convolution
integral in diagrammatic form, and these are the greatest contributions ever
in the field of wavefront analysis.
This book has not been written to address advanced subjects. For
example, items not covered include velocity-model independent techniques,
such as the common-reflection surface stack (CRS) and multifocusing (MF).
Likewise, not covered in depth are the advanced topics of anisotropy, shear
waves, migration, and full waveform inversion (FWI), which require entire
books in their own right. Rather, this book concentrates on the basic con-
cepts of Fermat and Huygens to explore and understand basic wave analysis.
This book is based upon inventive science. It deals with ideas, and not with
numerical algorithms. It does not explain the details the many migration and
inversion methods use, but it does provide the reader with the tools needed to
make those topics more understandable. The three parts of this book are in
DOI:10.1190/1.9781560803737

the order of increasing difficulty, and the most important part is Part 1,
because velocity analysis is central to every seismic investigation.

It was a privilege to have Susan Stamm, books manager for the Society
of Exploration Geophysicists, as the editor of this book. At every step of
the way, Susan would provide links that would tie the book together into
one coherent whole. This extraordinary feat makes much of the material
in the book valuable to those who do not want to struggle with mathematical
intricacies. To Susan, in the words of Shakespeare, we “can no other answer
make but thanks, and thanks; and ever thanks.”
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Introduction
Julius Caesar wrote, “Gallia est omnis divisa in partes tres” (“Gaul is a
whole divided into three parts”). In keeping with the division by three, this
book contains the same number of parts: Part 1 on velocity analysis, Part 2
on raypath analysis, and Part 3 on waveform analysis.
Velocity analysis (more precisely, conventional velocity analysis) is
based upon stacking. Velocity refers to the velocity of seismic waves in
their passage through an elastic medium. In this book, we only address com-
pressional waves in an isotropic medium. In other words, we do not explore
shear waves and surface waves, and we assume that the velocity at any given
point has equal magnitude in all directions. The velocities in various rock
types vary widely. Generally, velocity increases with increasing depth. Ve-
locity also varies laterally. Thus, velocity can be represented as a function
DOI:10.1190/1.9781560803737

v(x, y, z), where x, y, z are the three-dimensional (3D) spatial coordinates.

The velocity function can serve as a mathematical model of the subsurface
structure. A major task in seismic exploration is the estimation of the veloc-
ity function from knowledge of observed waveforms. Stacking is the basis
for the conventional estimation of velocities from seismic data. The efficacy
of stacking depends upon good alignment of the primary reflection events in
a common midpoint (CMP) gather. Various test velocities are used to deter-
mine the velocity which yields the best alignment.
Raypath analysis is based upon the eikonal equation. The underlying
concept is Fermat’s principle of least time, which states that the raypath
between two points is the path that can be traversed in the least amount of
time. A more modern statement is that rays traverse the path of stationary
time with respect to variations of the path. In other words, a ray prefers
the path such that there are other paths, arbitrarily close on either side,
along which the ray would take almost exactly the same time to traverse.
Waveform analysis is based upon the wave equation. The underlying
concept is Huygens’ principle, which states that each point on the wavefront
acts as a point source which emits spherical wavelets. These wavelets travel
in the medium, and, at a later instant of time, the total wavefront is the envel-
ope which encloses all of these wavelets. Fermat’s principle follows math-
ematically from Huygens’ principle in the limit of small wavelength. At
source points, seismic signals are sent into the earth (the medium) as
input. At receiver points, the returned seismic signals are recorded as
output. In an exploration program, we observe the source waveforms (i.e.,
swept-frequency signal, air-gun signature, etc.) and the received waveforms
(i.e., the seismograms). These observed data serve as the known quantities.

xv
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

xvi Basic Wave Analysis

The subsurface geologic structure (e.g., in the form of a velocity function) is

the unknown quantity. The inverse problem consists of using the known data
to estimate the unknown structure.
The conventional approach to seismic inversion conceptually separates
the observed wavefields into their component parts. The components are
sorted into two groups: one group is designated as signal; the other group
is designated as noise. In turn, noise is sorted into two groups: one group
is designated as outside noise; the other group is designated as source-
generated noise. Outside noise consists of environmental signals that exist
completely independent of the seismic experiment, such as microseisms,
wind noise, and traffic noise. In some prospects, outside noise is so minor
that it can be neglected; however, some outside noise, such as swell noise
and tug noise on marine seismic data, must be taken into consideration.
Although most microseismic noise can be neglected, the fact that it may
be the dominant feature at low frequencies is making it an issue of increasing
DOI:10.1190/1.9781560803737

interest because of the rise of full waveform inversion (FWI).

Consider these two examples affected by source-generated noise.
Source-generated noise includes source signatures, ghosts, reverberations,
and any other unwanted multiple events. The first example is traditional
seismic refraction prospecting. The signal consists of refracted waves
(i.e., head waves) and diving waves; all of the remaining information on
the seismograms is source-generated noise. The second example is tra-
ditional seismic reflection prospecting. The signal consists of the primary
reflection events. In certain cases, some multiple reflection events also
are considered as signal. Everything else on the seismograms is source-
generated noise. Of particular concern is a reverberation, which is a multiple
reflection in a layer. Although it usually occurs in the water layer in marine
work, reverberation sometimes occurs on land records too. Importantly,
reverberations are unwanted because they hide the desired primary
reflected waves. Therefore, the elimination of the source-generated noise
(as much as can be done) greatly simplifies the inverse problem. Then,
the signals can be imaged by use of either raypath or wavefront methods.
Alongside contemporary technical information, however, this book also
serves to remind the readers of our pioneering ancestors of scientific re-
search. Too often, the study of science pays minimal attention to the histori-
cal forebearers to whom we owe much. On the contrary, the stories of these
important figures provide fascinating insight into the development of ideas
which underpin our work today.
An important figure to mention was Carl Friedrich Gauss (1777–1855).
Not one to let difficulties hinder him, Gauss pioneered an approach for deter-
mining the path of heavenly objects. The problem began with Giuseppe
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Introduction xvii

Piazzi’s discovery of the asteroid Ceres on 1 January 1801. Piazzi was able
to make 22 observations over a period of 41 days before he lost the asteroid
in the face of the Sun. From these data points, the (ill-posed) problem was to
find the orbit of Ceres. Although corrections were needed for possible errors
and effects, e.g., Earth’s rotation axis, its solar orbit, and artifacts from
Piazzi’s observations, Gauss undertook the challenge using what would
become Gauss’s method. First, he determined a rough approximation of
the unknown orbit. Second, he used this approximation to revise his
initial calculation and derive a more precise orbit. Then, he repeated this
process iteratively until all of the values in the calculation became coherent
with each other and with the observations. In September 1801, several fore-
casts of the prospective orbit were published, including Gauss’s. In fact,
using his predicted locations, Ceres was observed again between 25 Novem-
ber and 31 December. Gauss wrote, “It is now clearly shown that the orbit of
a heavenly body may be determined quite nearly from good observations
DOI:10.1190/1.9781560803737

embracing only a few days; and this without any hypothetical assumption.”
From this success, Gauss continued to improve and refine his methods while
calculating orbits for additional asteroids as they were discovered. Gauss
published his methods in 1809 and the astronomical methods described
there are still in use today.
If we look back even further, the 1670s were remarkable because both
Gottfried Wilhelm von Leibniz and Sir Isaac Newton independently derived
their versions of calculus at that time. In Philosophiæ Naturalis Principia
Mathematica (1687), Newton established the law of gravity. However,
Newton did not use his calculus, but instead used Greek geometry affixed
with special manipulations to deal with infinitesimals. Few could fully
appreciate Newton’s work until the Bernoulli family, d’Alembert, Euler,
and others transformed Newton’s arcane mathematics into the calculus of
Leibniz. Indeed, this situation was not uncommon. Excellent theoretical
work often remained unused until it was (and is) transformed into a more
understandable account.
The same situation prevails in geophysics. Over the years, a tremendous
amount of mathematical material has been published. However, in our
(recent) past, most of that material could not be computed because we
lacked the necessary computer power and we lacked extensive acquisition
data. But now, the great computer power and acquisition methods available
today make it possible to use much of this mathematical material in order to
determine the paths of seismic waves through the intricacies of the subsur-
face. From the recorded data, the computer can reveal a comprehensive
description of the structure of the earth. Here, much depends upon geophy-
sicists who can choose relevant mathematics and have it converted into
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

xviii Basic Wave Analysis

computer code. The value of a new geophysical process is assessed in terms

of obtaining as output an accurate description of the subsurface, and not in
terms of the elegance of the mathematics involved. Good computer results
depend upon both the quality of the mathematical model and the efficacy of
the numerical methods used. Good results also depend upon an understand-
ing of the geophysical processes in play. To understand these processes, this
book will examine the basic tenets of our three pillars—velocity, raypaths,
and waveforms.
Geophysicists constantly use visual thinking, from initial layout to final
interpretation. The productive use of the principles given here will come
from geophysicists who make good use of visual imagery. Such visualiza-
tion is especially useful in solving problems where shapes, forms, or patterns
are concerned. Scientific results must be put into language and images that
can be easily understood and interpreted. When writing Philosophiæ Natur-
alis Principia Mathematica in 1687, Newton is said to have often wandered
DOI:10.1190/1.9781560803737

alone in the gardens of Cambridge University, drawing geometric diagrams

with a stick in the sand of the walkways (with others stepping around the
diagrams so as not to disturb them). Writing in the sand is also the inspiration
for Greek geometry; in turn, Greek geometry is the inspiration for math-
ematical analysis (calculus). Now, of course, mathematical analysis has
inspired computer methods. So, it is our current approach and methods
which are essential in forging ahead in the new world of massive data sets
and powerful computers. Despite shortcomings, current geophysical
methods work very well; even so, we know that we can do better. Today
the bounty of data resources and computing facilities is beyond anything
that could have been imagined a few years ago. As Gauss observed, it is
not knowledge but the act of learning, and not possession but the act of
getting there, which grants the greatest enjoyment.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms
DOI:10.1190/1.9781560803737

Part 1: Velocity Analysis

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms
DOI:10.1190/1.9781560803737
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1

Survey of Seismic Imaging

“Boldness be my friend!
Arm me, audacity, from head to foot!”
DOI:10.1190/1.9781560803737

—William Shakespeare

Exploration geophysics
Exploration geophysics has a dual scientific role: one is the continuous
search for new petroleum and mineral reserves, the other is the continuity of
research which provides better exploration tools and techniques and a more
complete knowledge of the physics of the earth. Recognizing, and develop-
ing, the interdependent nature of the search and research roles is essential for
progress in exploration geophysics. In the past, geophysical field work and
geophysical research often have been quite removed from each other, even
though the interplay between them has been central to a comprehensive view
of exploration geophysics. Indeed, the interchange of ideas is essential for
processing seismic data with computers. Any idea that can be embodied
in a computer program can be utilized immediately in the processing of
field data. In order to further develop this potential of the interchange of
ideas, every geophysicist must have a broad objective within which to
operate, but they also must be given full freedom to use their initiative,
imagination, and creativity. New ideas must come from both the search
and research aspects of geophysics.
As a result, it is important to maintain lines of communication among
geophysicists as well as with geologists and management. One way to
improve the communication of information is through the increased use
of models whose basic principles are understood by all geophysicists.
Such models can be mathematical models, physical models, or hybrid
models, and they should incorporate the field data as well as prior
3
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

4 Basic Wave Analysis, Part 1: Velocity Analysis

geophysical experience. Of course, in the usual case, the model cannot be

deduced by exact consideration but only represents the simplest explanation
which fits the data most plausibly. The selection of a model can be regarded
as taking place in three stages, i.e., the limitation of the choice to a finite
number of plausible model types, the selection of one of these types, and
the selection of the final model from all possible models within this type.
The development of good geophysical models increases our ability to
search for oil and conduct the necessary research; using better instruments
and better techniques helps not only geophysics but industry in general.
Because of the large mass of data involved in geophysical processing,
it is natural to build a statistical basis for geophysical models. To keep the
data tractable, they need to be sifted and sorted. In order to establish a cor-
respondence with the field situation, the research models also must allow for
noisy and uncertain situations. In this regard, models must incorporate stat-
istical parameters. Statistical data and statistical methods are in everyday
DOI:10.1190/1.9781560803737

use in most, if not all, sciences that unravel the world around us, and statisti-
cal techniques have proven indispensable in a great many areas of human
activity. In this respect, the field geophysicists historically have been
ahead of the research geophysicists, because every day field geophysicists
must come to terms with nature and make decisions about uncertain
events based on their evaluation and assessment of statistical data;
whereas, all too often, research geophysicists have confined themselves to
the narrow bounds of a strictly deterministic mathematical model which
can never adequately describe the complex factors present in the earth
(Robinson, 1982a, 1982b). However, in geophysics as in any other branch
of science, the scientific method is synonymous with the methods of
model building, or, to put it otherwise, the output of research takes the
form of scientific models. With the introduction of large-scale digital pro-
cessing of seismic data, the operative uses of scientific models have
become essential, and, as a result, research geophysicists have made signifi-
cant advances in model building. These advances make current geophysical
models useful in both the search for oil and the research for better methods,
and they place geophysics at the forefront of contemporary scientific knowl-
edge (Robinson, 1981). Today, geophysics strives to coordinate the practical
and theoretical aspects of model building with specific regard to their simi-
larities and differences. A first incentive for an integrated treatment is, of
course, the controlled experiment aspect of geophysics. Controlled exper-
iments can be conducted, but the expense typically is very high, and the
model builder, using field data, would utilize the experimental results as
much as possible. The second direct need for an integrated treatment
results from the fact that the available geophysical data is usually of a
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 5

composite nature, partially experimental and partially observational.

Indeed, great rewards will come to those who can piece together all of
this information. Geophysicists have the tools, ability, and know-how to
strive for and finally reach this end.
Now, let us consider some of the fundamental notions of model build-
ing. First, what are models? A model is simply a representation of something
else. Typically, only those details which seem essential for the intended use
of the model are included in the representation. A geophysical model is sup-
posed to represent the real earth in certain significant respects. However, the
advantage of a model over the real earth is the relative ease with which the
model may be manipulated and experimented with, as well as the conceptual
understanding of the earth made possible by the model. As such, the objec-
tive of model building is not to create an exact duplicate of the real earth;
rather, models simply should be useful substitutes for what they are repre-
senting. In fact, if there exists a useful connection between the behavior
DOI:10.1190/1.9781560803737

of the earth and the corresponding behavior of the model, then the model
can be useful in analyzing data from the earth and making geophysical, geo-
logical, and management decisions (Robinson, 1981). When we describe
models, we may do so verbally, graphically, or by mathematical functions
and equations. Models also may take the form of computer programs.
Some modes of expression might be more convenient than others for
describing certain types of models. Moreover, most models can be
expressed in several different modes.
In geophysics, it is essential to have several expressions of the same
model, so that we may employ various levels of sophistication across
these representations. The approach of model building is broad and flexible,
and several types of a model can be constructed for a variety of purposes. At
the highest level, any given model should be connected with other models,
both for geologic and managerial purposes. At the intermediate level, the
model should take into consideration all of the important geophysical pa-
rameters, so the geophysicist has the flexibility and scope to experiment
with and adjust the model to the field situation. At the lowest level, the
model should be built on a firm mathematical and physical foundation in
order to support the ultimate consequences resulting from the model.
In essence, model building is a systematic coordination of theoretical
and empirical elements of knowledge into a joint construct. It is important
to discern between these two pillars of the joint construct, i.e., the hypothe-
tical assumptions that constitute the theoretical part of the model and the
empirical observations that the model serves to interpret (Robinson,
1982b). Empirical observations enter model construction in different
ways: at an early stage when observations and experience are accumulated,
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

6 Basic Wave Analysis, Part 1: Velocity Analysis

whether or not a tentative model has been formed, and at more advanced
stages in assessing the parameters of the model on the basis of observations
and testing the theoretical model against the empirical evidence.
Returning to our juxtaposition of search and research, it is important
to note that the “search” aspect of geophysics represents empirical explora-
tion. More specifically, a seismic crew exploring for oil may be regarded
as a scientific organization conducting an empirical investigation on the
structure of sedimentary layers. On the other hand, the “research” aspect
of geophysics represents an investigation of a more theoretical nature.
Empirical investigation and theoretical investigation (“search” and “re-
search”) are somewhat independent vehicles for geophysical progress. In
some respects, theoretical investigation is ahead of what has been verified
empirically, and, in other respects, the converse is true. To put it otherwise,
and contrary to what is stated at times, it is not a fact that the theoretical
approaches have a kind of general priority or advantage over empirical
DOI:10.1190/1.9781560803737

approaches. However, it is usually quite another matter when it comes to

the reporting of fresh results, or to expository treatments in articles and
textbooks; in those cases, often it is appropriate and convenient to present
first the theoretical results and then the empirical aspects. This sequence
does not necessarily represent the order of discovery, especially in geophys-
ics where the empirical development of the exploration seismic method in
the 1920s preceded the corresponding theoretical development.
In addition, it is important to note the interaction between observational
and experimental in scientific approaches. Because it is not possible to
manipulate distant stars, astronomy is an observational science. High-
energy physics, geophysics, geology, paleontology, epidemiology, and
social sciences also are largely observational sciences. Even so, most
fields of science have both observational and experimental characteristics,
and this applies to our endeavors, too.
Remote sensing is a basic element in the observational sciences; in
particular, remote sensing is fundamental to reflection seismology. The
explorers of old would traverse the seas to find new lands and then sail
home with their treasures. Geophysicists, however, cannot travel through
thousands of meters of solid rock to find new horizons. Instead, they must
seek treasure indirectly by sending signals into the ground and interpreting
their echoes. Exploration geophysicists must make use of sophisticated
methods of remote detection in order to obtain their goal. The use of
seismic waves represents a way to obtain knowledge of a remote physical
body which cannot be reached by direct means. The investigation is depen-
dent on the organization of what is originally haphazard into a coherent
pattern. The result is a very detailed picture of the subsurface structure of
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 7

the earth. The raw seismic records would appear to be misleading and incon-
clusive to an uninitiated observer, but, with a very large degree of skill and
know-how, the seismic data are processed. A picture of the underlying
subsurface begins to appear as an undeniable fact. The accomplishment
of having this picture gradually emerge from a mass of seemingly non-
interpretable data is made possible by the expenditure of massive computer
power. The problem is solved when all of the evidence points in one clear
direction.
Furthermore, we can employ two kinds of remote sensing: passive and
active. In passive remote sensing, the observer waits for signals from
unknown and unreachable regions. In active remote sensing, the observer
emits signals into such regions and then records the resulting reflected,
refracted, or scattered signals. Earthquakes generate seismic waves which
then serve as the signals required for remote sensing. Because we must
wait for earthquakes to occur, earthquake seismology represents passive
DOI:10.1190/1.9781560803737

remote sensing. To locate petroleum and other mineral resources in the

upper layers of the earth’s crust, however, we may prefer to take an active
approach. In seismic exploration, source signals are transmitted into the
ground, and the reflected seismic waves are recorded and processed.
Thus, exploration seismology represents active remote sensing.

Earthquake seismology
Among the many great earthquakes that have wreaked havoc, none had
such significance and publicity as the catastrophe of Lisbon. On 1 November
1755, much of the City of Lisbon, Portugal was destroyed. Several tens of
minutes after the earthquake, an enormous tidal wave (tsunami) engulfed
the harbor and downtown, rushing up the Tagus River on which Lisbon is
built. Two additional tidal waves followed and, in the areas unaffected by
the tsunamis, a devastating fire raged for five days. As many as 100,000
people lost their lives. The property damage was enormous. A few buildings
survived, which serve as tourist attractions today. The vast royal archives
were lost, including precious works of art and detailed historical records
of explorations by Vasco da Gama and other early navigators.
The Lisbon earthquake heightened the debate between intellectuals of
the age on their views of reason and religion. Gottfried Wilhelm Leibniz
(1646– 1716) coined the term theodicy in an attempt to justify God’s exist-
ence in light of the apparent imperfections of the world. To Leibniz, the
concept that our world is the best possible world was not a sentimental
idea but one that could be demonstrated by reason and faith. In contrast, Vol-
taire (1694–1778) reacted sharply and impatiently to such a concept, and he
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

8 Basic Wave Analysis, Part 1: Velocity Analysis

gave his ultimate answer to the optimists in his satirical novel, Candide
(1759). Indeed, the great minds of Europe pondered the philosophic impli-
cations of the Lisbon earthquake during this period later known as the Age of
Enlightenment.
During this period of reflection, important advances in scientific study
were made as well. In fact, it was John Michell who conducted the first
scientific study of the earthquake. In doing so, he established that earth-
quakes travel in waves, and his research signaled the birth of modern
seismology.
John Michell (1724–1793) was educated at Queen’s College, Cam-
bridge, earning an M.A. in 1752 and B.D. in 1761. Among his accolades,
he was a fellow of his college, was elected a fellow of the Royal Society
in 1760 (the same year as Henry Cavendish), and became Woodwardian pro-
fessor of geology in 1762. In 1767, he became rector of St. Michael’s
Church of Thornhill, near Leeds, Yorkshire, England, at which he accom-
DOI:10.1190/1.9781560803737

plished much of his most important scientific work. Sir Edmund Whittaker
reportedly observed in 1910 that, during the century after Newton’s death,
“the only natural philosopher of distinction who lived and taught at Cam-
bridge was Michell,” although his “researches seem to have attracted little
or no attention among his collegiate contemporaries and successors, who
silently acquiesced when his discoveries were attributed to others and
allowed his name to perish entirely from Cambridge tradition.” Although
Michell remains virtually unknown today, there is a blue plaque on the
church wall in Thornhill that commemorates him (Figure 1.1).
In 1759, Michell deliberated on the great Lisbon earthquake. Michell
observed that the source of the Lisbon earthquake seemed to have been
under the sea, resulting in the tsunami. At that time, there was little scientific
understanding of earthquakes. Thus, Michell’s 1759 paper, “Conjectures
concerning the cause, and observations upon the phenomena of earth-
quakes,” marked the beginning of the scientific discipline of seismology.
Michell was the first to propose that earthquakes are caused by “shifting
masses of rock miles below the surface.” Faults are displaced fractures in the
earth’s crust. They are the outcomes of different forces pushing or pulling on
the crust, causing rocks to slide up, down, or past each other. Based on direc-
tion of slip, faults can be categorized as: (1) strike-slip, in which the offset is
predominantly horizontal, parallel to the fault trace; (2) dip-slip, in which
the offset is predominantly vertical and/or perpendicular to the fault trace;
and (3) oblique-slip, which combines elements of strike-slip and dip-slip.
The first depiction of a fault was drawn by Michell (Figure 1.2). It
showed a dip-slip fault in which the movement is absolutely vertical.
Michell’s concept of vertical earth motion persisted until about 1960,
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 9

Figure 1.1.
Thornhill plaque.
(Photo courtesy of
Bill Henderson.)
DOI:10.1190/1.9781560803737

Figure 1.2. A rendering of Michell’s original illustration of a dip-slip fault,

showing vertical movement of rock layers, that would cause an earthquake.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

10 Basic Wave Analysis, Part 1: Velocity Analysis

during which time it was believed the mountain belts were formed by iso-
stasy. Isostasy held that all elements in a system are either in or attempting
to reach a hydrodynamic equilibrium. For example, a piece of balsa wood
would float in water higher than an ice cube, because balsa wood is less
dense than ice. The balsa wood may be pushed down, but it would return
to its floating position after the downward pressure was removed, i.e., it
would “pop back up” to return to its status of equilibrium. If molasses
were used instead of water, things would happen more slowly because mol-
asses is more viscous than water. The pre-continental drift thinking in
geology may be described in this way. The pressure of contraction along
with isostatic adjustments were thought to have caused heavy portions of
the crust to sink and lighter portions to rise, operating in a hypothetical
interior of the earth conceived as a viscous fluid able to accommodate this
sinking and rising. Likewise, paleontologists used this theory to propose
that sinking land bridges could explain how identical fossil species were
DOI:10.1190/1.9781560803737

found on continents later separated by oceans.

Regarding location, Michell recognized that earthquakes were observed
to occur multiple times in certain areas, such as Lisbon, across different time
periods, and that they often were found near volcanic activity. Michell also
observed that earthquakes were the likely source of faults in rock strata, and
his illustration (Figure 1.2) depicted how rock layers are displaced during an
earthquake. Furthermore, Michell was the first to realize that earthquakes
can travel long distances as waves, and he observed that major shocks
always were followed by aftershocks. Indeed, Michell suggested that obser-
vations on the time of shock at several places would permit the determi-
nation of the place of origin of the earthquake.
In other scientific endeavors, Michell proposed several inventive con-
cepts which, though not necessarily recognized at the time, demonstrate
the breadth of his intellect. Michell proposed a heavenly object massive
enough to prevent light from escaping, correctly predicting that this dark
star (i.e., black hole) would not be visible directly but could be identified
by the motions of a companion star in a binary system. He also suggested
using a prism to measure what now is known as gravitational redshift, the
gravitational weakening of starlight due to the surface gravity of the
source. Although Michell himself acknowledged that some of these ideas
were not technically practical at the time, his 1784 letter to his good
friend Cavendish on the effect of gravity on light noted that he hoped
they would be useful to future generations.
Another topic addressed by Michell was magnetism. It was known that
naturally magnetized pieces of magnetite, called lodestone, would attract
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 11

small pieces of iron. Indeed, the first magnetic compasses were pieces of
lodestone in the form of needles suspended so they could turn. Such mag-
netic compasses were used in navigation at sea and surveying on land, but
they were expensive. In 1750, Michell published, at Cambridge, A Treatise
of Artificial Magnets, in which he disclosed an easy and expeditious method
of making non-natural magnets superior to the best natural ones. Michell
had discovered that magnets could be created by stroking strips of steel
with existing magnets; moreover, these permanent magnets were stronger
than the original and their production was less expensive than naturally
occurring lodestone. Besides the description of the method of magnetization
which still bears his name, this work contained a variety of accurate mag-
netic observations and a lucid exposition of the nature of magnetic
induction.
Furthermore, it was Michell who determined how the mass of the earth
could be measured. To this end, Michell built a torsion balance that
DOI:10.1190/1.9781560803737

measured the gravitational competition between different masses. (It

should be noted that the torsion balance is a scientific apparatus for measur-
ing very weak forces, usually credited to Charles-Augustin de Coulomb,
who invented it in 1777, although Michell independently invented one
sometime before 1783.) The competing masses were the earth versus two
large lead spheres. Gravity caused the apparatus, suspended on ropes, to
twist through a tiny angle. The angle was related to the force of gravity
exerted by the earth versus the two large lead spheres. Unfortunately,
Michell died in 1793, at age 68, before he could complete the measurement.
Instead, Cavendish obtained Michell’s torsion balance, rebuilt it, and, in
1797 –1798, conducted the experiment that weighed the world for the first
time. The result of Cavendish’s experiment was an estimated density of
the earth as 5.448 + 0.038 times that of water, which is about 1%
greater than the value recognized today. Because the volume of the earth
was known, the measurement of its density allowed its mass to be deter-
mined accurately.
Lastly, recognition of Michell’s achievements would not be complete
without including the large reflector telescope (3 m focal length, 75 cm
mirror) he built for use in his garden. Using this impressive instrument,
he was able to expand the understanding of star groupings, asserting how
double stars did not appear close together by chance but rather were held
together by gravity. Indeed, this paper, published in 1767, using probability
theory, was said to have inspired William Herschel, discoverer of Uranus,
and his sister, Caroline, to construct their famous catalog of binary stars,
after which Herschel proved that binary stars orbit a common center.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

12 Basic Wave Analysis, Part 1: Velocity Analysis

Traveling waves
There are two kinds of waves—standing waves and traveling waves. A
standing wave, which also is known as a stationary wave, is a wave that
remains in a constant position. A standing wave can arise in a stationary
medium as a result of interference of two waves traveling in opposite direc-
tions. The waves on a string of a musical instrument are examples of stand-
ing waves. Traveling waves propagate through a medium. In other words, a
traveling wave is not confined to a given space in the medium. The most
commonly observed traveling wave is an ocean wave. The main property
of a traveling wave is that it transfers energy from one point to another.
There are two main types of waves—mechanical waves and electro-
magnetic waves. A mechanical wave requires a medium in which to
travel. Examples are water waves, sound waves, and seismic waves.
Mechanical waves result from an oscillation of an elastic material. Each
DOI:10.1190/1.9781560803737

particle of the material oscillates in a small region around its equilibrium

position. The wave travels by a transfer of energy from one particle of the
medium to the next particle. Only the energy propagates; the material does
not. In other words, a mechanical wave conveys not material but energy.
Seismic waves within the earth are caused by earthquakes or man-made
sources. The waves travel through the solid earth and are recorded on
seismographs.
In each period of a mechanical wave, the substance of this medium is
deformed. The deformation reverses itself owing to restoring forces result-
ing from its deformation. For cases in which the medium is an elastic sub-
stance, there is no permanent displacement of the particles of the medium.
The particles of the medium are analogous to a boat anchored in the
ocean waves. The boat moves up and down in the same place while the
passing waves move onward. For example, sound waves propagate when
air molecules collide with their neighbors, transferring energy from neigh-
bor to neighbor down the line, causing a cascade of collisions in a given
direction. Because the collisions are elastic, the air molecules oscillate
about their equilibrium positions. This keeps the molecules from continuing
to travel in the direction of the wave. As a result, each molecule oscillates
about its same place, whereas energy travels in the direction of the wave.
While mechanical waves can only travel through a physical substance
(i.e., the lack of a vacuum), electromagnetic waves can travel through a
vacuum, because they do not need the substance to propagate. They
include radio waves, microwaves, infrared radiation, visible light, ultra-
violet radiation, X-rays, and gamma rays. These waves consist of periodic
oscillations in electrical and magnetic fields, and they travel at the speed
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 13

Figure 1.3. Sinusoidal wave plotted as a function of time.

DOI:10.1190/1.9781560803737

of light. It is because of electromagnetic waves that we can have the science

of astronomy. It is because of electromagnetic waves in the visible spectrum
that animals can see. A wave can be expressed in terms of energy, or wave-
length, or frequency. Frequency is measured in hertz, wavelength is
measured in meters, and energy is measured in joules.
The main property of a traveling wave is that it transfers energy from
one point to another. Traveling waves are found everywhere in nature.
They even occur in chemical reactions and epidemic outbreaks. The basic
parameters used to describe a traveling wave are amplitude, period, fre-
quency, wavelength, and velocity. As shown in Figure 1.3, the amplitude
of a wave is the magnitude of the maximum disturbance (from the central
value) during one wave cycle. The wavelength is how far the wave moves
during one cycle. The wavelength is the distance from the top of the wave
to next top of the wave. In other words, it is the distance from the crest to
crest. The wave velocity is how fast the wave is moving. Because velocity
is the ratio of distance over time, the velocity of the wave is equal to the
quotient of wavelength over period. Alternatively, the velocity is equal to
wavelength multiplied by frequency.
The science of seismology is concerned with the transmission of infor-
mation by the means of seismic waves. A disturbance in the earth is propa-
gated as waves moving away from their source. Some of these waves
eventually reach the receiving instrument, the seismograph. Classified
according to the range of transmission, the two primary branches of seis-
mology are: teleseismology, for which the range might be several thousand
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

14 Basic Wave Analysis, Part 1: Velocity Analysis

kilometers, and exploration seismology, for which the range might be

several thousand meters. For near-surface applications, the range might
be even tens of meters. The most important types of sources in teleseis-
mology are earthquakes and nuclear explosions; in exploration seismology,
the most important sources are small chemical explosions (shots), weight-
droppings, air guns, and vibration-inducing machines (known as vibroseis).
One of the most important characteristics of the earth, as it concerns
the transmission of seismic waves, is that it is composed of layered strata.
The layers considered in teleseismology would be the crustal layers, the
mantle, and the core; the layers considered in exploration seismology
would be post-Cambrian sedimentary layers in which deposits of petroleum
are found.
Exploration seismology is divided into the branches of reflection seis-
mology and refraction seismology. Most of petroleum exploration is done
by reflection seismic methods. Reflection seismology is a method of
DOI:10.1190/1.9781560803737

mapping the subsurface structure of the earth from the knowledge of the
arrival times of events reflected from the subsurface layers. The earth’s sedi-
mentary layers are predominantly sub-horizontal, but some of those layers
do have features, such as anticlines, unconformities, and faults, that can
serve as traps for petroleum.

Physical seismology and geometrical seismology

Let us first discuss light waves. The wave nature of light accounts for
effects such as interference and diffraction. Physical optics involves
systems in which the wave nature of light must be taken into consideration;
geometrical optics involves situations in which the size of the physical
objects of a system is much larger than the wavelength of the light. In geo-
metrical optics, rays are used to model propagation of light from a certain
source through a particular medium. This usage makes the description of
complex optical systems easier. A ray is a construction line used to deter-
mine what the wave does when affected by mirrors, lenses, and changes
of medium. You can think of a ray being a very narrow beam of light.
Numerical ray tracing is a method for calculating the path of the rays
through a system. Of course, every real system does experience diffraction
effects, so geometric optics is necessarily an approximation. The simplicity
arising from treating only rays, however, enables many uses.
All electromagnetic (EM) radiation has a wavelength and a correspond-
ing frequency. Also, a wave has a certain effective speed depending upon the
medium. A wave is a disturbance (oscillation) which transfers energy. Light
is an example of an electromagnetic wave where electric and magnetic fields
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 15

are oscillating. When we consider EM radiation, it is much more likely that

the term “ray” will be used in reference to frequencies equal to or greater
than visible light. At frequencies equal to or lower than visible light, you
also see the term “wave” being used. The higher the frequency, the less
likely you will see anything other than the term “ray” being used; the
lower the frequency, the more likely you will see the term “wave” being
used. For example, there are x-rays and gamma rays (which are high fre-
quency) versus radio waves and microwaves (which are low frequency).
In the same way, seismology can be treated by either its physical nature
or its geometrical nature. In physical seismology, we make use of numerical
wave propagation, which takes into consideration interference and diffrac-
tion. In geometrical seismology, we make use of numerical ray tracing,
which does not take into consideration interference and diffraction. Raytra-
cing, however, still allows multiple arrivals at the same receiver point
adding and interfering with each other. In essence, numerical wave propa-
DOI:10.1190/1.9781560803737

gation utilizes the wave equation whereas ray tracing utilizes the eikonal
equation. The wave equation is a second-order linear hyperbolic partial
differential equation. It is used to describe waves such as those occurring
in physics, e.g., sound waves, light waves, and water waves. It arises in
several fields, including acoustics, electromagnetics, and fluid dynamics.
The eikonal equation is a non-linear partial differential equation encoun-
tered in problems of wave propagation. On a digital computer, it is easier
to implement the eikonal equation than its parent wave equation.
At the turn of the 21st century, seismic instrumentation and computer
technology advanced to a high level of sophistication and power. Pre-
viously, numerical ray tracing and simplified numerical wave propagation
were the prevalent methods used in seismic imaging. Now it was possible
to use highly complex methods of numerical wave propagation. Tomo-
graphy and reverse time migration benefited from these improvements.
The ultimate formulation of the improved numerical wave propagation
became known as full waveform inversion (FWI).
Yet again, our contemporary achievements are made possible by the
visionaries of the past. Based on the mathematical classics of antiquity,
in 1662, the great mathematician Pierre de Fermat (1601– 1665) showed
that a ray of light passing from a high-speed medium (air) to a low-speed
medium (water) follows the path which takes the least time (see
Figure 1.4). This result is called Fermat’s principle or the principle of
least time. A modern version restates the principle in this way: rays of
light traverse the path of stationary optical length with respect to variations
of the path; i.e., a ray of light will prefer the path such that there are other
paths, arbitrarily nearby on either side, along which the ray would take
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

16 Basic Wave Analysis, Part 1: Velocity Analysis

almost exactly the same time to traverse.

Using this principle, it is possible to
describe the properties of light rays which
have been reflected from mirrors, refracted
through various media, or experienced com-
plete internal reflection. In fact, Fermat’s
principle follows mathematically from
Huygens’ principle (at the limit of small
wavelength), and it has the same form as
Hamilton’s principle of least action, which
is the basis of Hamiltonian optics. Further-
Figure 1.4. The bent line more, in Analyse des réfractions, Fermat
ABC takes the least time, employs the technique of adequality (i.e.,
because its segment AB approximate equality) to derive Snell’s
spends more time in high-
law of refraction and the law of reflection,
velocity air than straight-line
DOI:10.1190/1.9781560803737

and then, in Methodus ad disquirendam

segment AD and its segment
BC spends less time in low- maximam et minimam, he again uses adequ-
velocity water than straight- ality to analyze several issues of calculus,
line segment DC. including calculating maxima and minima
of functions, tangents to curves, area,
center of mass, and least action. Indeed,
Fermat’s work is an antecedent to the infinitesimal calculus of Newton
and Leibniz.
Moreover, Fermat’s first statement of his idea that the law of refraction
might be deduced from a minimum principle appears in a letter to Cureau de
la Chambre in 1657. Then, in a letter of 1 January 1662, Fermat announced
that he had accomplished his proof, demonstrating that the path of a
refracted ray of light was that which takes the least time! Importantly,
Fermat achieved this feat before the invention of calculus. Needless to
say, the objections came. In May 1662, Clerselier, an expert in optics and
leading spokesman for the Cartesians on this matter, wrote:

As it is not time which moves, it cannot be time which determines

the motion, and when a body is once moved and set on a certain
path, there is no reason to believe that time, greater or lesser,
could cause this body to change its path, since time does not act
on it and has no power over it. But, as the entire speed and direction
of this body depend on its force and the disposition of its force, it is
much more natural, and, in my view, much more scientific, to say as
Mr. Descartes does, that the speed and direction of this body are
altered by the alteration which takes place in the force and the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 17

disposition of this force, which are the true causes of its movement,
and not to say as you do, that they change by an intention which
nature possesses of always taking the path it can pursue most
quickly, an intention which it cannot have, since it acts without
knowledge, and thus has no effect on this body.

Although stated eloquently, this rebuke of Fermat by Clerselier proved not

to withstand the test of time.
Continuing with the great mathematicians of the age who influenced
our field, next was Johann (also known as Jean or John) Bernoulli
(1667– 1748), one of several prominent mathematicians in the Swiss Ber-
noulli family. Known for his contributions to calculus, and for educating
a young Leonard Euler, Johann Bernoulli was the first to solve the bra-
chistochrone problem. In June 1696, he posed the problem to others and
wrote:
DOI:10.1190/1.9781560803737

I, Johann Bernoulli, address the most brilliant mathematicians in

the world. If someone communicates to me the solution of the pro-
posed problem, I shall publicly declare him worthy of praise. Given
two points A and B in a vertical plane, what is the curve traced out by
a point acted on only by gravity, which starts at A and reaches B in
the shortest time.

In response, he received four solutions: Newton, Jacob Bernoulli,

Leibniz, and Guillaume de L’Hôpital. According to Newton’s biographer,
John Conduitt (1688– 1737), Newton “did not come home till four (in the
afternoon) from the Tower very much tired, but did not sleep till he had
solved it, which was by four in the morning.”
For our purposes, however, it is the solution obtained by Johann
Bernoulli that is the most instructive. A heavy point object (a bead) starts
from rest at the point A and slides without friction along a curve to a
lower point B. Although there are an infinite number of curves, the
problem is determining the curve down which a bead sliding from rest
and accelerated by gravity will slip (without friction) from A to B in the
least time. This curve is called the brachistochrone curve. The term is
derived from the Greek brachistos (shortest or minimum) and chronos
(time or delay); the word, brachistochrone, means shortest time or, alterna-
tively, minimum delay.
Johann Bernoulli considered an arbitrary curve from A down to B in the
vertical x, z plane. The x-axis is horizontal and the z-axis is vertically down-
ward. The bead is subject to the acceleration g of gravity. The numerical
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

18 Basic Wave Analysis, Part 1: Velocity Analysis

Figure 1.5. Raypath in layer i.

DOI:10.1190/1.9781560803737

value for g is about 9.8 m/s2. There are slight variations in this numerical
value (to the second decimal place) that are dependent primarily upon
altitude and latitude. In sliding down the curve, the bead accelerates, so
its velocity continues to increase.
Johann Bernoulli’s solution divides the vertical plane into layers
(see Figure 1.5). He assumes that the bead follows a straight line in each
layer, so that the path is piecewise linear. Let v1 , v2 , v3 , . . . be the velocity
downward in the successive layers, and let the ray crossing the interfaces
successively have incident angles u1 , u2 , u3 , . . . Snell’s law gives

sin u1 sin u2 sin u3

= = = ··· (1.1)
v1 v2 v3

In the limit, as the layers become infinitely thin, the line segments tend
to a curve. If v is the velocity at (x, z) and u is the angle of incidence, then the
curve satisfies

sin u
=p. (1.2)
v
Now, we come to an important point. We need the velocity function,
which gives velocity v versus depth z. Fortunately, Galileo already has
derived the velocity function, which is

v= 2gz . (1.3)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 19

Substituting equation (1.3) into equation (1.2), we obtain the equation of the
curve as

z = k2 sin2 u , (1.4)
1
where k2 = . So, we have
2gp2
dz sin2 u 1 1
z′ = = cot u and sin2 u = = = .
dx sin u + cos u
2 2 1 + cot2u 1 + z′ 2
(1.5)
Thus, equation (1.4) for the curve becomes

k2
z = k2 sin2 u = , (1.6)
1 + z′ 2
DOI:10.1190/1.9781560803737

which yields

z(1 + z′ 2 ) = r , (1.7)

where r ; k2 . As shown by substitution, equation (1.7) is satisfied by the

cycloid
x(t) = r(t − sin t), z(t) = r(1 − cos t) . (1.8)

Although Bernoulli makes the analogy to a light ray, we will make the
analogy to a seismic ray. The seismic ray traversing a medium is a brachis-
tochrone, i.e., a minimum-delay curve. In the case that the velocity function
is given by equation (1.3), the brachistochrone is a cycloid (see Figure 1.6).
A cycloid is described by a point in the circumference of a wheel that rolls

Figure 1.6. An (upside down) cycloid.

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

20 Basic Wave Analysis, Part 1: Velocity Analysis

Figure 1.7. A diving wave.

upon a straight line. The variable t measures the angle through which the
wheel of radius r has rotated. In our case, the straight line is the x-axis
DOI:10.1190/1.9781560803737

and the rolling proceeds upside down (i.e., the circle rolls under the x-
axis). Johann Bernoulli concludes his discussion of the brachistochrone
problem with these words:

Before I end, I must voice once more the admiration I feel for the
unexpected identity of Huygens’ tautochrone and my brachisto-
chrone. Nature always tends to act in the simplest way, and so it
here lets one curve serve two different functions.

Turning next to a diving wave (see Figure 1.7), the velocity function
often used to model this wave is given by
v(x, z) = v0 + az , (1.9)

with v0 and a each constant. The x-axis depicts the surface of the ground.
We see that the velocity increases linearly with depth but does not vary
with horizontal coordinate. At the shot point (the origin of coordinates),
the ray makes the angle u0 with the vertical. The particle travels along the
raypath (x, z). The method of Johann Bernoulli shows that the raypath is
an arc of a circle of radius
v0
r= = constant . (1.10)
a sin u0
The center of this circle is point C with coordinates
 
v0 v0
(xC , zC ) = , − . (1.11)
a tan u0 a
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 21

The tangent to the circle at the shot point makes an angle u0 with the
z-axis. Note that the vertical component zC does not depend upon the
angle. As a result, the family of rays with the given velocity function is com-
posed of circular paths that pass through the origin (which is the shot point).
All of the circular paths have their centers on the line z = −v0 /a, where
formally v(x, z) = 0, and which is referred to as the velocity depth.

Tomography
Tomography is a method for finding the velocity and reflectivity distri-
bution from a multitude of observations using combinations of source and
receiver locations. Generally, space is divided into cells and data are
expressed as line integrals along raypaths through those cells. In seismic
tomography, slowness (or velocity), and sometimes an attenuation factor,
is assigned to each cell and traveltimes (and amplitudes) are calculated
DOI:10.1190/1.9781560803737

by tracing rays through the model. Depending on the focus, there are differ-
ent specializations within tomography. Cross-hole tomography uses sources
and receivers in different boreholes in an effort to reconstruct an object from
wave projections. Traveltime tomography focuses on information gained
from arrival times; similarly, attenuation tomography focuses on amplitude
and diffraction tomography analyzes the scattered wavefield. Reflection
tomography is a method for determining the velocity and/or reflectivity dis-
tribution from observations of reflection events at various locations and
source-receiver offsets. In tomography, the source points and the receiver
points are known. Received waves are calculated by tracing rays through
a model consisting of a multitude of cells, each having constant velocity
(and constant attenuation). The results are compared with the observed
waves. Then, the model is perturbed and the process repeated iteratively
to minimize the errors, measured by the difference between the calculated
and observed waves. After each significant change of assumed velocity,
raypaths must be recalculated; for small changes, raypaths can be reused.
When, after a number of iterations, the calculated wave matches the
observed wave, we have reason to assume that we have arrived at a plaus-
ible velocity (attenuation) model that represents the medium in the field.
In the case of marine seismic surveys, where the boats drag receivers at
the sea surface, a common datum can be used as a reference of the vertical
locations of sources and receivers. On the other hand, when the seismic
survey is not on a leveled surface, such as land or ocean bottom, the topo-
graphy of these surfaces causes varying positions of sources and receivers.
Such variations create computational difficulty in certain processing steps
which assume a flat acquisition level. In these cases, a seismic reference
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

22 Basic Wave Analysis, Part 1: Velocity Analysis

datum (SRD) often is used. A seismic datum correction is a value added to

reflection times of seismic data to compensate for the location of the geo-
phone and source relative to the seismic datum. Mathematically, the
sources and receivers are transferred to a level reference datum by means
of static corrections. The static corrections include elevation corrections
and weathering corrections for removing the effects of near-surface layers.
First arrivals often are associated with diving waves due to the velocity
gradients within the near-surface layers. The downgoing incident wave
rapidly turns around before being reflected and is recorded by the receiver
as the first arrival. Just as reflection traveltime tomography can be used to
update an initial estimate of a subsurface velocity-depth model, turning-
ray tomography may be used to update an initial estimate of a near-
surface velocity-depth model. As delineated in Yilmaz (2008), the process
is as follows. Start with an initial velocity-depth model composed of near-
surface layers with constant velocities and thicknesses, and then determine
DOI:10.1190/1.9781560803737

the first-arrival times by ray tracing. Begin with the picking of the first
arrivals that represent the diving waves through the near surface. Define
an initial velocity-depth model by a set of near-surface layers with constant
velocities and thicknesses. Model the first-arrival times by ray tracing.
Compute the difference between the modeled and observed first-arrival
times. Estimate the change in parameters vector by perturbing the velocities
of the near-surface layers only. Iterate as necessary to minimize the dis-
crepancy between the modeled and actual first-arrival times. Then, the
final velocity-depth model resulting from the iterative application of
turning-ray tomography is used to compute the one-way traveltimes
through the near-surface model along vertical raypaths. These are used to
apply the necessary source and receiver statics corrections to the prestack
data. Update the parameter vector to obtain a new near-surface velocity-
depth model. Iterate as necessary to minimize the discrepancy between
the modeled and actual first-arrival times (Yilmaz, 2008).

Depth points
Subsurface rocks exist in depth. Seismic reflection data portray this sub-
surface in recorded two-way time. At the source point, we (as an industry)
always record the downgoing vibroseis swept frequency signal. However, at
sea, we do not always record the airgun source-signature signal, although in
an ideal world we should. For the purposes of this discussion, we assume the
downgoing signal (whatever it is) is recorded. At the receiver point, we
record the upgoing signal. A portion of the downgoing energy is converted
at the geologic depth points and becomes upgoing energy. In situations in
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 23

Figure 1.8. Level

interface with midpoint
M directly above depth
point D (left) and dipping
interface with midpoint
M not above depth point
D (right).

which layers do not dip, the depth point is the halfway point when a wave
travels from a source to a reflector to a receiver. In the case of flat layers,
the depth point is vertically below the midpoint. In the case of dipping
beds, the depth point is not vertically below the midpoint. Given the
source signal, the receiver signal, and the two-way time, the problem is
the determination of the corresponding depth point. The principle used to
solve the problem is the same in the case of either ray-based methods or
DOI:10.1190/1.9781560803737

wave-based methods.
Figure 1.8 depicts two versions of reflected ray SDR. The assumption that
the reflection event comes from a horizontal interface is shown on the left. A
horizontal interface is a restrictive assumption. What if the interface is not
horizontal? This is shown on the right of Figure 1.8. Reflector positions
are valid when the total distance from the source to reflection point to receiver
is constant. The locus of all possible refection points is an ellipse with the
source at one focus and the receiver at the other focus. The problem is to
find the depth point. It is easy in the case of a level interface (shown at
left), but difficult in the case of a dipping interface (shown at right).
We want to trace the path of a seismic ray, such as shown in Figure 1.9.
We will make the analogy to a miner who travels in tunnels dug into the
earth. A miner goes down the shaft at point S. The miner arrives at depth
point D. The time from from S to D is t1 . At D, the miner abruptly turns
and goes up a second shaft from D to R. The time from D to R is t2 . The
total time elapsed is t.

Problem: Given the total time t, solve the equation t = t1 + t2 for each
of t1 and t2 . This problem is the case of solving one equation to determine
two required unknowns.

Solution: Send a second miner down the shaft starting at R for the
amount of time t. He goes from R to D. At D, the second miner does not
turn but continues downward. In this way, he goes down a third shaft
from D to Q. With the first miner at S and the second miner at Q, set the
clock at time zero. The first miner starts at S (at time 0) and heads down.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

24 Basic Wave Analysis, Part 1: Velocity Analysis

DOI:10.1190/1.9781560803737

Figure 1.9. To facilitate ray tracing from paths, here t1 ¼ traveltime from S to D,
t2 ¼ traveltime from D to R, and t ¼ two-way traveltime ¼ t1 + t2 .

The second miner starts at Q (at time 0) and heads up. They will both meet at
D at the same time, i.e., the required unknown time t1 . It then follows that the
required unknown time t2 is t2 = t − t1 .

Application: Both reverse time migration (RTM) and full waveform

inversion (FWI) use the above procedure to find a depth point D. However,
D is a mathematical entity. Is it a geologic entity? At each grid point, corre-
late the downgoing wave of the first miner with the upgoing wave of the
second miner. A strong correlation indicates that D is a geologic turning
point; a weak correlation indicates that the point in question is not a geologic
turning point.

Prestack reverse-time depth migration

Migration, also called imaging, is the transformation of seismic data
recorded as a function of arrival time into a scaled version of the true geome-
try of subsurface geologic features that produced the recorded seismic
energy. Migration may be described as an inversion method that transforms
the seismic data observed at the earth’s surface into a subsurface geologic
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 25

map. Time migration assumes that velocity varies only in the vertical direc-
tion, whereas depth migration also allows for horizontal variation of velocity.
Both time and depth migration results can be displayed in either time or
depth. Imaging involves focusing and positioning, and it depends on a
specific earth model. Focusing involves such things as collapse of diffractors,
maximizing amplitude, and reproducing wavelet character. Positioning
involves the correct location events and the sharpening of event terminations
relative to faults, salt flanks, and unconformities. Migration is a computer
operation that makes use of raypath analysis or some approximation to the
wave equation. Migration by application of the wave equation is accom-
plished in one of several ways: integration along diffraction surfaces
(Kirchhoff migration); numerical finite-difference, downward-continuation
of the wavefield or phase-shift; and equivalent operations in frequency-
wavenumber or other domains (frequency-domain migration). Poststack
migration involves mush less computing than prestack migration. However,
DOI:10.1190/1.9781560803737

common-midpoint stack does not correctly stack dipping events. Dip-

moveout can partially correct the deficiency. The less expensive poststack
migration (even with dip moveout) is inferior to more-expensive prestack
migration.
Prestack reverse-time depth migration (RTM) is based on full-wave
seismic modelling on a numerical grid. It is composed of both forward and
backward processes, i.e., the seismic forward modelling of the wave field
originating from the source and the reverse-time modelling of the received
records. The following of events both downward and upward through the
earth model provides RTM the capability to image all possible arrivals
without dip limitation. RTM accounts for extreme lateral velocity variations.
It explicitly handles turning waves and all other complex propagation
paths. The ability to image these complex wave modes reveals aspects of
the subsurface that otherwise would have had poor direct illumination.
Specifically, the prestack migration of RTM means that shot-records are
individually migrated before any stacking. The basic assumption is that only
first-order (i.e., primary) reflection events are present in the seismograms.
The source signature (i.e., the vibroseis swept frequency signal or the air-
gun signal) should be removed from the raw shot records by deconvolution.
In this way, all of the shot records from the various shots in the prospect will
have the same unit impulse as source.
A numerical grid is used for the computations. RTM consists of two
processes:

Process (1). Forward-time modeling: Source waves are computed

downward (forward in time) from surface to depth.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

26 Basic Wave Analysis, Part 1: Velocity Analysis

Process (2). Reverse-time modeling: Received waves are computed

downward (backward in time) from surface to depth.

In both processes (1) and (2), seismograms (seismic responses recorded

as a time series) are generated for each grid node. At the grid node, the
seismograms of the two processes are correlated. The resulting correlation
is stored. It is assumed that strong correlation between the amplitudes of
(1) and (2) only exists at a subsurface grid node where a reflection
occurs. In other words, a reflection occurs at a certain location and time if
there is a down-going wave from the source and a corresponding up-
going wave to the receivers. The total time for the two waves must be the
two-way traveltime. Reverse-time depth migration assumes that only
primary reflection events are present in the seismograms. Accordingly,
the direct wave (first arrival), head waves, and multiples must be muted in
the records prior to RTM. The RTM system uses the two-way wave
DOI:10.1190/1.9781560803737

equation, thus improving imaging in areas where complex geology violates

the assumptions made by less computationally intensive methods of
migration. Until recently, the enormous computing resources needed for
RTM has impeded its use.

Iterative improvement method

A direct method is able to solve a problem by a finite sequence of
operations. In the absence of rounding errors, a direct method would
deliver an exact solution (as in the solution of a linear system of equations
by Gaussian elimination). However, many problems, especially non-linear
problems, do not admit a direct solution. Moreover, in cases where direct
methods theoretically still exist, the cost of computing the often large
number of variables may be prohibitive. In such cases, iterative methods
are computationally feasible alternatives. In computational mathematics,
an iterative improvement method is a mathematical procedure that uses
an initial guess to generate a sequence of improving approximate solutions,
in which the nth approximation is derived from the previous approxi-
mations. An iterative method is called convergent if the corresponding
sequence converges for given initial approximations. Heuristic-based itera-
tive improvement methods are common.
To discuss the application of an iterative method, consider this ima-
ginary case study. Suppose that the detailed historical records of explora-
tions by Vasco da Gama and other early navigators were not destroyed by
the Lisbon earthquake. Also suppose that all of the original historical
records of explorations by Columbus and his contemporaries remain
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 27

intact. These voyage records would include ship’s logs and other sailing
and port-of-call information; even pertinent weather conditions often
were inscribed in the log. In the study of climate change, there is interest
in the climatologic conditions in past ages, such as the age of exploration.
In particular there is always interest in tides, currents, and prevailing
winds.
Now, we can add the forward problem and the inverse problem to our
example. Tides, currents, and prevailing winds represent the cause of a
ship’s motion. The resulting voyage of the ship represents the effect. The
forward problem is: Given the cause, find the effect. The inverse problem
is: Given the effect, find the cause. In our case, the forward problem is:
Given the tides, currents, and prevailing winds, determine the course of
the ship. The inverse problem is: Given the course of the ship, determine
the tides, currents, and prevailing winds. The direct problem involves deduc-
tive reasoning, whereas the inverse problem involves inductive reasoning.
DOI:10.1190/1.9781560803737

In inductive reasoning, the premises are viewed as supplying strong

evidence for the truth of the conclusion. In contrast to a deductive argument
in which the conclusion is certain, the truth of an inductive argument’s con-
clusion is probable, based upon the evidence given.
We have the voyage records and we want to find the tides, currents,
and prevailing winds. It is a difficult inverse problem. How do we go
about it? We make an educated guess (i.e., a guess based on knowledge
and experience, therefore likely to be nearly correct) of the tides, currents,
and prevailing winds. Then, we solve the forward problem in order to find
the synthetic ships’ voyages. The objective function is a measure of the
goodness of fit between the synthetic data and the actual data. The computed
synthetic data yields only one point on the objective function. The objective
function looks like a mountain range, and the goal is to climb to the top of the
highest peak. The peak represents the best possible fit.
We are at one point on the mountain range (i.e., on the objective func-
tion) in a thick fog so we cannot see where we are going. However, we can
measure things in a very small neighborhood of the point. We choose to
measure the gradient of the slope. The gradient tells us which direction is
up and the steepness in that direction. Our direction of movement is the
same direction as given by the gradient. By climbing a certain distance in
the direction of the gradient, we correct our initial guess about the tides, cur-
rents, and prevailing winds. The result is an improved guess. Next, we take
the improved guess as our starting point. We follow the same procedure
again in order to further improve the guess. We continue this iterative
process until we are satisfied with the answer we derive for the tides, cur-
rents, and prevailing winds.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

28 Basic Wave Analysis, Part 1: Velocity Analysis

Saddle point or pass

Historically, maximum and minimum problems have influenced the
development of calculus. In its geometrical meaning, the derivative
df (x)/dx is the slope of the tangent to the curve y = f (x) at the point
(x, y). At a maximum or minimum of f (x), the tangent to the curve is hori-
zontal, that is, its slope is equal to zero.
An extremum (plural extrema) is any point at which the value of a
function is largest (a maximum) or smallest (a minimum). There are both
absolute and relative (or local) maxima and minima. For example,
Figure 1.10 shows the tangent is horizontal at the five designated points,
so that f (x1 ) and f (x3 ) are minima. We see that f (x1 ) is a relative
minimum, while f (x3 ) is the absolute minimum. Similarly, f (x2 ) represents
a relative maximum and f (x4 ) the absolute maximum. Finally, f (x5 ) is
neither a maximum nor a minimum, even though the tangent is horizontal.
DOI:10.1190/1.9781560803737

This shows that the vanishing of the derivative is a necessary, but not a suf-
ficient, condition for the occurrence of an extremum of a smooth function
f (x). A point where the derivative vanishes, whether it is an extremum or
not, is called a stationary point.
Let us find the extreme values of a function z = f (x, y) of two variables.
The function represents the height z of a surface above the x, y-plane.
In effect, the function represents a mountain landscape. As shown in

Figure 1.10. Stationary points, i.e., points where the tangent is horizontal.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 29

Figure 1.11. Surface plot of peak (left) and contour plot of peak (right).
DOI:10.1190/1.9781560803737

Figure 1.12. Surface plot of pass (left) and contour plot of pass (right).

Figure 1.11, the maximum value of the function corresponds to the top (the
peak) of a mountain. Pikes Peak is the highest summit of the southern Front
Range of the Rocky Mountains in North America. A minimum corresponds
to the bottom (the deep) of a depression. Death Valley is the deepest point on
the North American continent. A depression that contains water is called a
pond or lake. At both maxima and minima, the tangent plane to the surface is
horizontal.
In addition to peaks and deeps, however, there are other points for which
the tangent plane is horizontal, such as mountain passes (see Figure 1.12).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

30 Basic Wave Analysis, Part 1: Velocity Analysis

The Khyber Pass is a mountain pass in the north of Pakistan, close to the
border with Afghanistan. Generally speaking, a pass is a route through a
mountain range. The pass is the lowest possible route between two
mountains. Also, a pass is the highest point on that very route.
Because many of the world’s mountain ranges have presented formidable
barriers to travel, passes have been important since before recorded
history. There are other words for pass such as gap, notch, col, gorge,
and saddle. A deep is an inverted peak. More specifically, on the
contour map, regard any level line of elevation h as if it had the elevation
−h. Then the map is inverted. It becomes the map of the bottom of the
oceans. The peaks become deeps, the deeps become peaks, but the passes
remain passes.
Now, consider a different visual representation of this scenario, as
shown in Figure 1.13. There are two mountains A and B on a range and
two points C and D on different sides of the range. Suppose that we want
DOI:10.1190/1.9781560803737

to cross the range using an optimal path. Each possible path will have a
highest point; however, there will be one path CD for which the altitude
of its highest point is lowest. The point E of maximum altitude on this
path is the mountain pass, called in mathematical language the saddle
point. It is clear that E is neither a maximum nor a minimum, because we
can find points near E which are higher and lower than E.
Similarly, if we choose to travel from peak A to peak B, each possible
path will have a lowest point; however, there will be one path AB for

Figure 1.13. A mountain pass E.

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 31

which the altitude of its lowest point is highest, and the minimum for this
path is also point E. Thus, this saddle point E has the property of being a
highest minimum (a maxi-minimum) or a lowest maximum (mini-
maximum). Because E is the minimum point of AB, the line tangent to
AB at E must be horizontal. Because E is the maximum point of CD, the
line tangent to CD at E must be horizontal. Therefore, the tangent plane,
which is the plane determined by these lines, also is horizontal. Thus, we
can recognize three different types of points which have horizontal
tangent planes: maxima, minima, and saddle points.

Full waveform inversion

With the great advances in seismic acquisition, large volumes of seismic
data are now available. The enormous power of modern computers makes
DOI:10.1190/1.9781560803737

possible the consideration of complex seismic problems. Generally, direct

computational methods do not exist for the solution of such complex prob-
lems, so iterative methods must be used. Iterative improvement methods are
the mainstay for many depth migration techniques and for full waveform
inversion (FWI).
Full waveform inversion is a method of inverting the recorded seismic
traces so as to obtain a satisfactory model of seismic velocity. Generally
in FWI, the recorded seismic traces do not have their source signatures
removed by deconvolution. The source signatures appear in the forward
modelled and adjoint source parts of the algorithm, and they are never
strictly deconvolved away. In fact, many FWI algorithms can update the
source signature as part of the inversion process. Initially, a model of
smoothly varying velocity is used as the educated guess. If the resulting
synthetic records are not satisfactory (i.e., if the resulting synthetic
traces are a poor fit to the recorded traces), then the velocity model is
adjusted and new synthetic records are formed. This process is repeated
until satisfactory synthetic records are obtained. In summary, iterative
improvement is a means of adjusting the velocity function until hopefully
an excellent subsurface image of the geology is obtained. Iterative
improvement corresponds to finding one’s way out of a maze. The path
taken must be adjusted successively in order to reach the endpoint. Some-
times simple adjustments can be made, but usually the entire imaging
process has to be redone several more times and often many more
times. Even then, a satisfactory solution may not be obtained. It is impor-
tant to separate the final objective (hypothesis) from what is really feasible
(factual).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

32 Basic Wave Analysis, Part 1: Velocity Analysis

Full waveform inversion methods are among the most recent tech-
niques for seismic data processing and they are under continuous devel-
opment. The resolution limit of FWI depends on the wavelength of the
seismic data, so it cannot be said that FWI is able to image arbitrarily
heterogeneous subsurfaces. The objective of FWI is to develop a high-
resolution, high-fidelity velocity model of the subsurface that will
facilitate the matching of individual synthetic seismic waveforms with
an original field dataset. A good match is a necessary condition, but
generally is not a sufficient condition, for a model to be close to the
true subsurface earth. Refraction-based FWI utilizes diving waves (i.e.,
wavefronts continuously refracted upward through the earth due to the
presence of a vertical velocity gradient). Ideally but not always realisti-
cally, FWI can resolve small-scale velocity features, in particular, in
shallow-water environments where reflection-based methods are limited.
Utilizing refracted and diving waves, FWI overcomes the restric-
DOI:10.1190/1.9781560803737

tions posed by conventional reflection tomography and highly accurate,

high-resolution shallow velocity models can be obtained. In addition,
minimum pre-processing is required, and free-surface effects in the data
do not need to be removed. Reflection FWI now is commonly used in
the industry.
In seismic prospecting, a seismic source is placed at or near the
surface of the earth (whether land or sea). The seismic waves propagate
downward through the earth, and, due to the geological structure of the
site, multiple reflection and refraction occur. Sensors (geophones), also
placed at or near the surface of the earth, receive the returning waves.
Both the source waves and the returning waves are recorded. The input
to the FWI algorithm consists of the observed data (i.e., source waves
and received waves) and an initial guess of the velocity model. Full
waveform inversion uses an iterative improvement scheme to minimize
the difference between the observed data and synthetic data which is
generated from a wave simulator with an estimated (velocity) model
of the subsurface. As such, each FWI framework essentially consists
of a wave simulator for modeling the synthetic data and an adjoint
simulator for calculating a velocity update from the data misfit. The
use of the phrase “correlation coefficient” to generally describe the
objective function of FWI is misleading. Some FWI algorithms may
use a correlation to remove the effects of the waveform in early iter-
ations (to help with the cycle-skipping issue), but most FWI objective
functions are defined in terms of full waveform misfit and are not
simply correlations.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 33

DOI:10.1190/1.9781560803737

Figure 1.14. Iterative improvement method.

As shown in Figure 1.14, iterative improvement can be described in

terms of mathematical symbols. We use the following notation:

Y = observed shot records (known)

P = wave equation (known)
x0 = initial guess of velocity function (known) (1.12)
xn = velocity function at step n
yn = synthetic shot records at step n

Let xn be the velocity function, P be the wave equation, and yn be the syn-
thetic shot records. In the linear case, we can write
yn = P xn . (1.13)
Generally, y does not linearly depend on x (although the wave equation is
linear between the source and the wave function). In the nonlinear case,
the relation between x and y should be written as y = F(x). Here, we need
to make the transition from the general problem to its iterative solution.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

34 Basic Wave Analysis, Part 1: Velocity Analysis

The general statement of the problem is that we do not know x, and the itera-
tive solution is given by a sequence of xn for n = 1, 2, 3, . . . .
In the linear case, the slightly wrong value xn + dxn yields

yn + dyn = P (xn + dxn ) . (1.14)

Thus

dyn = P (xn + dxn ) − yn . (1.15)

Expanding equation (1.15), we have

dyn = Pxn + P dxn − yn . (1.16)

We know all of the entries on the right side of equation (1.15). Thus, we
know the value of dyn . Using equation (1.13) in equation (1.16), we obtain
DOI:10.1190/1.9781560803737

dyn = P dxn . (1.17)

We have reduced the problem of solving equation (1.17) for dxn .

Now, at this point, iterative improvement is used to find the value of dxn .
Equation (1.17) is the local linearization of the optimization problem
at xn . Generally, there are two nested iterations:

1. To solve the global non-linear optimization problem by sequential

linearizations.
2. To solve the local linearized problem by linear algebra.

There will be disagreement between synthetic records yn and the actual

observed records Y. These records involve a large number of values. We
want to condense all of the numbers into only one number by means
of a function, called the objective function c(x). The objective function
measures goodness of fit between synthetic and observed Y. The optimum
xn would be that value of x for which the objective function c(x) is the
maximum. We can think of the objective function as a mountain range in
which we are looking for the top of the highest mountain. As Voltaire
writes, “Perfection is attained by slow degrees; it requires the hand of time.”
Iterative improvement is a technique to solve optimization problems.
In this technique, we start with a sub-optimal velocity function x0 (i.e., an
initial guess), and the velocity function is improved repeatedly step by
step. Essentially, the initial velocity function x0 is an educated guess as to
what the velocity function might be. At worst, the initial velocity function
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 35

could be chosen at random. However, it is prudent to make a judicious

choice. The idea of starting with a sub-optimal solution is compared to start-
ing from somewhere near the base of a mountain. The improvement of
the sub-optimal solution is compared to walking in steps up the mountain.
We finally stop when we are close to the top of the mountain. To start the
process, we begin with the initial guess. We can plot the objective function
as a contour map (i.e., level-line map). There are three kinds of stationary
points on a contour map: peaks (or highs) H, passes (or saddle points with
a horizontal tangent plane) S, and deeps (or lows) L. We not only want to
find a peak, but also we want to find the highest peak. The problem of
finding the highest peak is already difficult in two dimensions. With more
dimensions/parameters, the topology of the stationary points can be much
more complicated. Another frequently occurring problem is that peaks
and deeps actually may be long and narrow stretched valleys/ridges.
The direction of the gradient vector indicates the direction in which the
DOI:10.1190/1.9781560803737

function increases most rapidly, thereby indicating the orientation of the

tangent plane. Therefore, the tangent plane to the graph of the function at
a point can be determined by the gradient vector and the value of the func-
tion at that point.
How do we find the top? We do it in steps. We are at xn which is at
elevation c(xn ). We want to take the next step to the top. There
are various algorithms, such as the conjugate gradient algorithm and the
steepest-ascent algorithm. The conjugate gradient algorithm has faster con-
vergence compared to the steepest-ascent algorithm, but we choose to use
the latter to illustrate the concept due to its simplicity. For the steepest-
ascent algorithm, we compute the gradient gn of the objective function at
xn . Let an be the chosen step-length. Then, the new value of the velocity
function is given by the correction equation
xn+1 = xn + an gn . (1.18)
The result xn+1 is the corrected value of the velocity function. The corrected
value xn+1 replaces the value xn just used, and the process starts over. The
procedure is iterated until some velocity model appears with a satisfactory
level of goodness of fit.
In summary, we want to determine the velocity function of the subsur-
face from observed seismic data. First, an initial velocity function (an edu-
cated guess) is used to implement an iterative improvement method. The
global maximum of the objective function gives the particular velocity func-
tion that most resembles the physical velocity structure of the subsurface.
Many mean-square-error functions are multi-peaked, resembling a mountai-
nous landscape with many peaks and many deeps. A number of local maxima
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

36 Basic Wave Analysis, Part 1: Velocity Analysis

exist. Theoretically, only one global maximum exists. However, in practice,

there may be many maxima that approximately qualify as the unique global
maximum. Indeed, in the presence of noise, the global maximum loses it
absolute meaning. Unfortunately, there is no simple way to find the global
maximum. Generally, the iterative improvement technique focuses on a
specific region and determines its maximal value. This peak may or may
not be the global maximum, and this peak may represent a sufficient but
not ideal solution. Thus, the iterative improvement scheme can locate a
local maximum, but it may not produce an optimal solution. If the objective
function is extremely rugged with several peaks and valleys, the iterative
improvement technique will need to begin with a very good initial guess
in order to be effective. The following words of Alexander Pope in An
Essay on Criticism (1711) effectively describe the objective function:
Here hills and vales, the woodland and the plain,
Here earth and water seem to strive again,
DOI:10.1190/1.9781560803737

Not chaos-like together crushed and bruised,

But, as the world, harmoniously confused:
Where order in variety we see,
And where, though all things differ, all agree.
Typically, we use an iterative algorithm to solve the fully non-linear
problem in FWI. It begins with an initial model, which then is improved
iteratively by using a sequence of linearized local inversions. The choice
of the initial model is critical to avoid the local minima in FWI. Often,
reflection tomography is a good choice for its efficiency and global consist-
ency. At this stage, primary reflection events are used to resolve the long
wavelength velocity trend and shallow gradients, providing a suitable
initial model for any subsequent FWI iteration.
The goal of FWI is to estimate a discrete parametrization of the subsur-
face by minimizing the misfit between the observed shot records of a seismic
survey and numerically modeled shot records. The predicted shot records
are obtained by solving an individual wave equation per shot location.
Both refraction- and reflection-based FWI can be used. Full waveform
inversion is ideally suited to broadband seismic data particularly rich in
low frequency information. Technically, the data should be rich in low-
frequency information provided by long offset acquisition. This is a
crucial point. The long offsets available today are the result of the great tech-
nological advances made in acquisition equipment in recent years. Reflec-
tion-based FWI is designed for use with backscattered arrival, such as
reflection and diffraction. A benefit is that much deeper velocity model
updates are possible for standard streamer lengths. The input data should
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 1: Survey of Seismic Imaging 37

have sea-surface effects removed. Recording the full wavefield enables the
model builder to progressively add the detail required to resolve complex
velocity problems in the overburden.
Full waveform inversion can be formulated as an attempt to solve for
different aspects of an earth model. Elastic FWI solves for elastic parame-
ters, including P-wave velocity, S-wave velocity, and density. Acoustic
FWI usually solves for P-wave velocity; a density model typically is not
extracted. Related to implementation, FWI also can be formulated in
either the time or frequency domain.
To conclude this section, in spite of its common use, a complete and
objective account of FWI cannot fail to mention that it is subject to
certain fundamental limitations. These include the following.
1. Geophysical inversion is an inherently ill-posed problem. There is no
guarantee that a solution exists, nor that it is unique, nor that it
depends continuously on the input data. In many cases, FWI suffers
DOI:10.1190/1.9781560803737

from non-uniqueness (the well-known cycle-skip problem). Due to the

stochastic nature of the subsurface, a unique correct model generally
does not exist, but rather, if we take non-uniqueness and uncertainty
into account, many models can be equally useful for the purposes of
the geophysical interpreter (focusing and matching possible well data).
2. Generally, there is no description of forward wave propagation in the
earth which is sufficiently accurate, complete, or adequate to model
observed data. Common choices are modeling by finite-difference sol-
ution of the wave equation or by asymptotic ray methods. Both modeling
techniques, along with other techniques, have their particular limitations
(assumptions, artifacts). Therefore, the earth model resulting from FWI
generally will carry the imprint of the selected forward modeling
technique.
3. With regard to the problem of imperfect forward modeling, another
problem is that there is generally no adequate parametrization of the
earth. For computational purposes, the earth needs to be represented
by a mathematical description which can be programmed in a computer.
Some techniques for forward modeling, such as finite differences,
require the earth to be represented by a rectangular grid; other tech-
niques, such as ray methods, require a representation by a layered or
blocky model with sufficiently smooth layer and interface functions.
In this context, Glogovsky et al. (2009) ask the rhetorical question
whether the earth can be represented by splines.
For a further detailed and critical review of FWI, see Tarantola (2006),
Landa et al. (2006), Weglein (2013), and Landa (2013).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms
DOI:10.1190/1.9781560803737
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2

Time, Distance, and Velocity

“All truths are easy to understand once they are discovered; the
point is to discover them.”
DOI:10.1190/1.9781560803737

—Galileo Galilei

Time-distance curve
To create a map of the subsurface, a geophysicist converts received
seismic traces, which record events as a function of time, into a format
that records them as a function of depth. In other words, a time function
recorded at the surface must be transformed into a depth function. Unlike
radio waves, the velocity of seismic waves is very dependent on
the medium through which they travel. Thus, the velocity changes as the
waves travel into the earth. Generally, velocity increases with depth,
although occasionally there may be layers in which a decrease in velocity
occurs. For a given surface point, the velocity plotted as a function of
depth is called the velocity function. Thus, in reflection seismology, there
are two equally important variables: reflection time and velocity.
Knowing these variables allows the depth to the reflecting horizons to be
determined. Because there are important lateral changes in velocity, that
is, because the velocity function varies from one location to another, a
given velocity function cannot be assumed to be valid for an entire prospect.
As a result, the velocity depth function must be corrected from place to place
over the area of exploration (Robinson, 1982a). The problem of velocity
estimation is not easy in the case of isotropic rocks. In the case of anisotropic
rocks, it is, in fact, even more difficult.

39
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

40 Basic Wave Analysis, Part 1: Velocity Analysis

One method of measuring the velocity function is to drill a deep hole,

i.e., an exploration well, and determine the velocity by placing seismic
detectors in the hole at various depths. However, it also is possible to
estimate the velocity function by measurements confined to the surface.
In such a case, the velocity function is estimated by analyzing the time
differentials of a particular event which is received by a lateral array of
detectors. These estimates always will depend upon a ceteris paribus
(other things being equal) assumption. In the following sections, we will
consider the layered-earth seismic model which provides a method for deter-
mining the velocity function.
In most cases of seismic data acquisition, the detectors are located
upon, or close to, the surface. In ocean bottom acquisition, however, the
receivers can be either ocean bottom cables or ocean bottom nodes. Typi-
cally, ocean bottom cables are positioned by a receiver boat, and ocean
bottom nodes are positioned freely in the ocean depth. Ocean bottom
DOI:10.1190/1.9781560803737

cables are appropriate for shallow water areas only, usually less than 1 km
in depth; ocean bottom nodes, however, may be used at depths greater than
1 km or more. Compared to conventional streamer acquisition on the water
surface, ocean bottom acquisition has several benefits, such as the use of
dual sensors with both hydrophone receivers that record pressure and geo-
phone receivers that record particle velocity (as well as other elastic proper-
ties of waves). Ocean bottom acquisition can provide a much wider azimuth
range than streamer acquisition, which is limited by the number of towed
streamers. While surface acquisition can be affected by the noise of water
waves, streamer boats, other vessels, and oil platforms, typically seismic
data recorded by ocean bottom acquisition will be cleaner because it is not
affected by these noise sources. Towed streamers, however, can be affected
significantly by oil tankers and platforms. An additional benefit of ocean
bottom acquisition is that it can record continuously for a long period of time.
Borehole geophysics addresses the vertical analysis of the conditions of
the subsurface and its fluid content. Borehole seismic data collected during
and after the drilling of exploration and appraisal wells delivers valuable
information that is critical for reservoir characterization. Combined with
surface seismic data, borehole seismic surveys image specific reservoir fea-
tures around the borehole. One of the primary results of a borehole geophy-
sical experiment is the time to depth relation curve. This curve can be used to
generate an accurate calculation of interval, root-mean-square (RMS), and
average velocities. The defining characteristic of a vertical seismic profile
(VSP) is that either the energy source or the detectors (or sometimes
both) are in a borehole. In the most common type of VSP, receivers in the
borehole record reflected seismic energy originating from a seismic
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 41

source at the surface. There are many other surveys that can be done in a
borehole other than the simple VSP; each type of survey will produce differ-
ent results and help exploration in a different way.
The principle of the refraction seismograph is based on the idea that the
first arrivals of the sound reach the detectors by the least traveltime path
along the interface of some buried contrastingly high-speed layer. The
principle of the reflection seismograph is based on the idea that the sound
of the explosion penetrates down to the buried contrastingly high-speed
layer and is reflected back to the detector (without traveling along the
interface of the reflecting bed). The first arrivals on all seismic records
usually are either direct or refracted events. Later events on seismic
records may be either reflected or refracted events. In all cases, time is the
basic quantity measured.
In seismic exploration, nothing is obvious. Everything requires some
assumption or some estimate. It has been clear since the outset of seismic
DOI:10.1190/1.9781560803737

exploration about a century ago that the surface layers present a formidable
problem. Along a seismic land line from shot point to shot point, the surface
elevation changes, the depth of the shot changes, and the thickness of the
slow-speed surface material changes. For that reason, the seismic records
have been reduced to an arbitrary reference surface known as a datum.
The reduction to a seismic datum minimizes local topographic and near-
surface effects. Seismic times and velocity determinations are referred to
the datum plane (usually, but not necessarily, horizontal and planar) as if
sources and geophones have been located on the datum plane and as if no
low-velocity layer exists. In other words, usually a datum level is chosen
near the surface of the ground, yet everywhere deeper than the estimated
surface layer. All of the recorded time above the datum level is removed
(by weathering calculations) at each individual shot point and by spread
setup. This removal of time above the datum, in effect, places the shot
point and detectors below the surface and on the datum.
Well surveys are reliable methods for determining seismic velocity. In
most cases, however, it is necessary to estimate the velocity function by
means of seismic surveys on the surface of the earth. These surveys excite
sources of energy at various surface points and record the resulting
seismic traces by means of a lateral array of detectors, also on the surface
of the earth. Each trace is a function of time and is associated with the hori-
zontal coordinates of source and receiver. The distance between the source
and the receiver is called the (full) offset x. The (two-way) traveltime is the
time it takes a wave to go from the source to a deep horizon and then, by
reflection, to the receiver. Such a deep horizon appears on the seismic
data as a reflection event. The locus of an event is a curve which we can
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

42 Basic Wave Analysis, Part 1: Velocity Analysis

consider as a function of time t versus offset x. Such a curve is called a time-

distance curve. The time-distance curve is fundamental in the consider-
ations which follow.
Consider Figure 2.1, in which the horizontal x-axis represents the
surface of the earth. A moving wavefront, at any particular instant, is the
totality of points which are in the same phase of the cycle (e.g., crest,
trough, etc.) in the propagation. In this book, we consider only isotropic
media. As a result, each raypath always intersects the wavefronts at right
angles. The angle at which the upcoming seismic ray strikes surface of
the earth is called the angle of emergence. More precisely, the emergence
angle is the angle between a straight-line raypath (or the tangent to the
raypath if it is curved) and the normal to the earth’s surface.
Referring to Figure 2.2 (bottom), because the medium is isotropic, the
wavefront is perpendicular to the raypath. From the geometry, it is shown
that the ray emerges at an angle with the vertical, and the wavefront
DOI:10.1190/1.9781560803737

makes the same angle with the horizontal. In other words, the wavefront

Figure 2.1.
The emergence
angle u0
(Robinson and
Treitel, 2008).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 43

Figure 2.2.
Fundamental
seismic time-
distance
relationship:
time-distance
curve (top) and
wavefront curve
(bottom).
DOI:10.1190/1.9781560803737

angle u is the same as the emergence angle u. In a small increment of time dt,
the wave travels a distance v dt along the raypath. In the same amount of
time dt, the wavefront sweeps out a distance dx along the x axis. Slowness
is defined as the reciprocal of velocity. The derivative dt/dx is called the
horizontal slowness. Thus,
 
dt
v dt v velocity dx
sin u = = = =  
dx dx horizontal velocity 1
dt v
horizontal slowness
= . (2.1)
slowness
In other words, equation (2.1) says that the sine of the emergence angle
is equal to the ratio of the wave velocity v to the horizontal velocity dx/dt.
Because |sin u| ≤ 1, it follows that the horizontal velocity is always greater
than or equal to the wave velocity. Equivalently, we may say that the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

44 Basic Wave Analysis, Part 1: Velocity Analysis

slope dt/dx of the time-distance curve must be less than or equal to 1/v.
Thus, horizontal slowness is always less than or equal to the slowness.
Now, consider Figure 2.2 (top). The traveltime t of raypath A is plotted
above its emergent point x. This gives one point on the time-distance curve.
The traveltime t + dt of raypath B is plotted above its emergent point x + dx.
This gives another point on the time-distance curve. The locus of all such
points for a given event composes the time-distance curve. The time-
distance curve of a seismic event is defined as the plot of the traveltime t
of the seismic event versus the horizontal distance x.
What is the value of the time-distance curve in exploration? Its value is
that it can be kinematically extracted from surface observations. In the past,
a geophysicist would identify reflection events on paper seismic records. For
each event, the various values of (x, t) could be recorded and the time-
distance curve could be plotted. Then, the slope dt/dx of the time-distance
curve could be measured. Also suppose that velocity v could be measured.
DOI:10.1190/1.9781560803737

Then, the value of sin u could be computed by means of equation (2.1). In

this way, the emergence angle u is obtained.
When the receiver point is directly above the depth point, the emergence
angle u = 0, from which it follows that p = 0. The reciprocal of Snell’s
parameter p is

1 v dx
= = , (2.2)
p sin u dt

which is the horizontal velocity of the seismic event, i.e., the apparent veloc-
ity of the event along the x-axis. We see that

v dt v dt
sin u = and cos u = . (2.3)
dx dz

We have the Pythagorean relationship

sin2 u + cos2 u = 1 ,

which yields
 2  2  2
dt dt 1
+ = . (2.4)
dx dz v

Equation (2.4) is known as the eikonal equation. It is a very fundamental

equation for describing ray and wavefront propagation in general
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 45

inhomogeneous media. The eikonal equation can be written as

(horizontal slowness)2 + (vertical slowness)2 = (slowness)2 . (2.5)

Alternative expression for the time-distance curve

Let l be the wavelength. In Figure 2.3, l represents the distance from O
to P. The period T is the time for the wavefront to travel the distance l from
O to P. The velocity with which the wave travels this distance is v = l/T. It
is called the wave velocity (or more specifically, the phase velocity).
However, if we merely observe motion along the horizontal axis, we see
it takes the same time T to travel the distance lx from O to P1 . The velocity
with which the disturbance travels this distance is called the horizontal
phase velocity. We have
DOI:10.1190/1.9781560803737

lx l v
horizontal phase velocity = = =
T T sin u sin u
phase velocity
= . (2.6)
sin u

Because sin u is less than or equal to one in magnitude, we see that

the horizontal phase velocity is greater than or equal to phase velocity v.

Figure 2.3. Plane wave traveling with respect to distance.

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

46 Basic Wave Analysis, Part 1: Velocity Analysis

Likewise, the vertical phase velocity is

lz l v phase velocity
vertical phase velocity = = = = . (2.7)
T T cos u cos u cos u

Likewise, the vertical phase velocity is greater than or equal to phase

velocity v.
The slowness is the reciprocal of velocity; that is, the slowness is
n = 1/v. We have

1 sin u
n = slowness = , horizontal slowness = ,
v v
(2.8)
cos u
vertical slowness = .
v
DOI:10.1190/1.9781560803737

It should be noted that both the horizontal slowness and vertical slowness
is less than or equal to the slowness. Figure 2.4 shows a plane wave
which passes through the origin at time 0 and passes through point P at
time t. The coordinates of P are (x, z). If r is the length of the ray OP and
u is the angle between the ray and the z-axis, then x = r sin r u and
z = r cos u. The traveltime t from O to P is equal to t = r n = , which
v

Figure 2.4. Plane wave traveling with respect to time.

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 47

we can write as

r sin u cos u
t = (sin2 u + cos2 u) = (r sin u) + (r cos u)
v v v

or

sin u cos u
t=x +z
v v
= x (horizontal slowness) + z (vertical slowness) . (2.9)

Single horizontal interface

A conic section is a curve obtained as the intersection of a cone in 3D
with a dipping plane. The three types of conic section are hyperbola, para-
DOI:10.1190/1.9781560803737

bola, and ellipse. The circle is a special case of the ellipse, but sometimes it
is called a fourth type of conic section. Often, the ellipse and hyperbola are
defined using two points, each of which is called a focus. The combined
distances from these foci are used to create an equation of the ellipse and
hyperbola. A parabola has one focus point. The graph of a parabola wraps
around its focus. A cone is the surface traced by a straight line (the genera-
trix) moving so that the line always passes through a fixed point (the vertex).
The path is directed by a closed plane curve (the directrix), along which the
line always slides. In the case of a right circular cone, the directrix is a circle.
The axis of this cone is a line through the vertex and the center of the circle,
the line being perpendicular to the plane of the circle.
The greatest progress in the study of conics by the ancient Greeks is due
to Apollonius of Perga (ca. 265– 190 BC), whose eight volume Conic Sec-
tions summarizes and greatly extends the existing knowledge of the time.
Pappus of Alexandria (ca. AD 290–350) is the most important mathematical
author writing in Greek during the late Roman Empire, producing a volumi-
nous account of the most important work accomplished in ancient Greek
mathematics. Concerned with the existence of a sequence of basic construc-
tions leading from the known to the unknown, somewhat as in algebra,
Pappus’s work includes commentary on the geometry of Euclid, Apollonius,
and Eratosthenes. Moving forward, Descartes publishes La Géométrie
in 1637, in which he unifies algebra and geometry into the single subject
of analytic or coordinate geometry, and geometric lines and curves are
transformed into algebraic equations. As an example of the power of
the method, Descartes offers his solution to the “Pappus problem.”
Indeed, the methods of analytic geometry have been indispensable in the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

48 Basic Wave Analysis, Part 1: Velocity Analysis

development of calculus and what followed. And, as René Descartes com-

ments, “You just keep pushing. You just keep pushing. I made every mistake
that could be made. But I just kept pushing.”
Depending on the nature of the interface, reflection from the interface is
either specular (mirror-like) or diffuse (retaining the energy, but losing the
image). The laws of specular reflection are as follows.

1. The incident ray, the reflected ray, and the normal to the reflection
surface at the point of the incidence lie in the same plane.
2. The angle of incidence (i.e., the angle which the incident ray makes with
the normal) is equal to the angle of reflection (i.e., the angle which the
reflected ray makes to the same normal).
3. The reflected ray and the incident ray are on opposite sides of the normal.

Now, let us find the time-distance curve for the case of a single horizon-
DOI:10.1190/1.9781560803737

tal reflecting interface. As shown in Figure 2.5, let z be the depth of the
reflecting plane below the surface. Let S designate the position of the
source and let S′ designate the position of the source image with respect
to the reflecting interface. The source image S′ is a mathematical construc-
tion only. The mathematical points S and S′ are both on the same vertical to
the reflecting interface but in opposite directions, each at a distance z from

Figure 2.5. Time-distance curve for a simple flat interface.

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 49

the interface. Let R be the position of the receiver, at a horizontal distance

x from S. The length of the reflection path SDR is the same as the length
S′ DR . This length is traversal with a velocity v in a time t. Note that time
t is the so-called “two-way” traveltime, that is, the traveltime down to the
reflector plus the traveltime back up to the surface. Thus, the length of the
travel path is vt. The vertical distance is 2z. The horizontal distance x is
the offset of R from S. Because triangle RSS′ is a right triangle, the Pythagor-
ean theorem yields

x2 4z2
x2 + (2z)2 = (vt)2 or t2 − = 2 . (2.10)
v2 v
The two-way vertical time is t0 = 2z/v, which is where the time-
distance curve intersects the vertical t-axis. Thus, equation (2.10) becomes

x2
t2 − = t02 .
DOI:10.1190/1.9781560803737

(2.11)
v2
In equation (2.11), t0 and v are constant, and t and x vary. The resulting
curve, which is the time-distance curve, is shown to be a hyperbola with
asymptotes t = +x/v. The t-intercept t0 is equal to the two-way traveltime
of the vertical (or normally incident) reflection. As x 2 increases from x = 0,
the traveltime increases. If we solve equation (2.11) for t, we obtain
√
x2 + 4z2
t= . (2.12)
v

Differentiating with respect to x, we obtain the slope p of the time-

distance curve as
dt x x
p= = √ = . (2.13)
dx v x2 + 4z2 v2 t

Furthermore, we see that the sine of the emergence angle u is

SR x x
sin u = ′
= √ = . (2.14)
SR x2 + 4z2 vt

If we use the expression for p given by equation (2.13), we obtain

Snell’s law
sin u
=p. (2.15)
v
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

50 Basic Wave Analysis, Part 1: Velocity Analysis

Let us write the slope, equation (2.13) in the form

dt 1
= +  . (2.16)
dx 4z2
v 1+ 2
x
Thus, as x  +1, the slope approaches +1/v; that is, the slope of
the hyperbola approaches the slope of the asymptote. That is, the horizontal
velocity (which is greater than the wave velocity) approaches the wave
velocity as the receiver moves farther and farther from the source. Consider
a wave which radiates from source S along the surface with velocity v. The
time-distance curve (to the right of the origin) for this wave is a straight line
through the origin with horizontal velocity v. That is, the slope of this line
is 1/v, so its equation is t = x/v. We recognize this line as the asymptote,
so we have obtained a physical interpretation of the asymptotes as the
hyperbola.
DOI:10.1190/1.9781560803737

Critical angle
Figure 2.6 shows two layers separated by a horizontal boundary at depth
z1 . Let v1 be the velocity in the upper layer and v2 be the velocity in the lower
layer. Assume that v2 is greater than v1 . A ray starting at source S and reach-
ing the interface at point D will produce a reflected ray DR and a refracted
(i.e, transmitted through the interface) ray DT. The ray DR, which is
reflected back into the first layer, makes an angle u1 with the normal. The
ray DT, which is refracted into the second layer, makes an angle u2 with
the normal. Snell’s law is

sin u1 sin u2
= . (2.17)
v1 v2

Figure 2.6.
Illustration of
Snell’s law.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 51

There is a unique case in which the refraction angle u2 is 908. In this

case, sin u2 is unity. Thus, equation (2.17) becomes

sin u1 1
= . (2.18)
v1 v2

By definition, the incident angle u1 that results in the refracted angle u2 equal
to 908 is called the critical angle of incidence. The critical angle is denoted
by f. Therefore, equation (2.18) can be rewritten as

sin f 1
= . (2.19)
v1 v2

Thus, the expression for the critical angle is

DOI:10.1190/1.9781560803737

v1
f = sin−1 . (2.20)
v2

Equation (2.20) is called the critical condition. When u1 equals the criti-
cal angle f, the refracted ray DT travels in the lower layer approximately
along the interface. If u1 exceeds the critical angle f, then no refracted
ray such as DT is produced. This result is that the energy in the incident
ray SD is totally reflected.
Figure 2.7 represents two homogeneous, isotropic horizontal layers with
velocities v1 and v2 . The shot point is S. The receiver is R located at offset x.
Following the explosion at S, three types of longitudinal waves are received
at R, provided that x is sufficiently great. Each of these arrivals, as well as
other such arrivals, is called a seismic event. Between S and R, there is

Figure 2.7. Head wave SABR and reflected wave SDR.

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

52 Basic Wave Analysis, Part 1: Velocity Analysis

one unique direct ray, one unique reflected ray, and one unique refracted ray.
Their raypaths are:

1. The direct wave traverses raypath SR.

2. The reflected wave traverses raypath SDR.
3. The refracted wave (or head wave) traverses raypath SABR; this assumes
that v2 . v1 , but, if v2 , v1 , then there are no head waves.

The traveltime t required for the direct wave to traverse its raypath SR is
equal to the quotient of the horizontal distance x divided by velocity v1 .
Thus, the equation of the time-distance curve is
|x|
t(x) = . (2.21)
v1
It is composed of two straight lines passing through the origin with slopes
DOI:10.1190/1.9781560803737

1/v1 and −1/v1 .

The traveltime t required by the reflected wave to traverse its raypath
SDR is given by the hyperbola which cuts the t axis at 2z1 /v1 and whose
asymptotes correspond to the traveltime, equation (2.21), of the direct
arrival. The equation of the time-distance curve is

x2 + 4z21
t(x) = . (2.22)
v1
The traveltime t required by the head wave to traverse its raypath SABR
is determined as follows. The head wave leaves S and is refracted at A at the
critical angle. Then, it travels in the lower layer along the path AB parallel to
the interface. At B, it travels upward to R. The traveltime t is the sum of the
traveltimes in the upper and lower layers. The traveltimes in the two layers
are
SA + BR AB
and . (2.23)
v1 v2
We see that
z1
SA = BR = and AB = x − 2z1 tan f . (2.24)
cos f

Thus, the traveltime is the sum

2z1 x − 2z1 tan f
t(x) = + . (2.25)
v1 cos f v2
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 53

Let us replace the trigonometric functions by their equivalents in terms

of the velocities. It follows from the critical condition (2.20) that

 2
v1 v1
cos f = 1 − and tan f = . (2.26)
v2 v2 cos f

Thus, the traveltime curve for the refracted wave is


 2  2
1 1 x
t = 2z1 − + . (2.27)
v1 v2 v2

It is a straight line having a slope of 1/v2 .

We have obtained these results in the case of one horizontal interface.
However, the same reasoning holds for any interface in a multiple horizon-
DOI:10.1190/1.9781560803737

tal-interface model. The head wave leaves S and refracts at the interface in
question, and travels along it within the underlying layer. This motion
causes a disturbance in the overlying layer that produces the upgoing leg
that ends at R. In refraction seismic surveying, we measure the earliest
times of arrival of the seismic waves at various distances from the source.
Often, it is convenient to make S the origin of the coordinate system, but
it is not required.
Let us now return to the one-interface model as shown in Figure 2.8. We
let S be the position of the source. The source initiates a wavefront which
expands spherically in three dimensions. Let us consider the raypaths associ-
ated with this wavefront. The direct wavefront moves along the surface with
velocity v1 . The time-distance line SE for this direct wave is given by
equation (2.21).
Some of the energy from the source travels in a wavefront to the inter-
face and is reflected back to the surface. The raypath is SD0 R0 . In Figure 2.8
(top), the time-distance curve (solid line) for this reflected wave is the
reflection hyperbola CDE given by equation (2.22). The asymptotes of
this hyperbola are the two straight lines (dashed) composing the time-
distance curve of the direct wave.
In addition to the direct path and the reflected path, there is another path
of basic importance. This third path, called the refraction path (also known
as the head wave path), occurs whenever velocity v2 in the lower bed is
greater than velocity v1 in the upper bed. Some of the energy from the
source travels downward and strikes the interface at the critical angle f.
The path SD0 R0 represents the critical reflection with angle f. Any
reflection such as SDR with an angle less than f will produce a refracted
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

54 Basic Wave Analysis, Part 1: Velocity Analysis

DOI:10.1190/1.9781560803737

Figure 2.8. Model with one horizontal interface: raypath SD0 (which is at critical
angle f) produces refracted energy that travels along the underside of the interface,
continually bleeding energy upward to reach the surface; this bleeding energy
constitutes the head waves with raypaths such as SD0 D1 R1 , SD0 D2 R2 , and
SD0 D3 R3 .

path (i.e., transmission path) heading downward in layer 2. However, any

reflection with an angle greater than f cannot produce a downward refrac-
tion in layer 2. The net result is that we obtain raypaths such as SD0 D1 R1 ,
SD0 D2 R2 , and SD0 D3 R3 , where each of the segments D1 R1 , D2 R2 , and
D3 R3 have angle f. These raypaths represent the refracted wave (or head
wave).
Upon inspection of Figure 2.8, we see that raypath SD0 R1 is the raypath
of both the reflected wave and the head wave. Thus, at the emergence
point Ro , both the reflected wave and head wave have the same traveltime.
Their time-distance curves have the common tangent point A. We also know
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 55

that, for horizontal interfaces, any time-distance curve satisfies

dt sin u
= . (2.28)
dx v1
In particular, the slope of the reflected time-distance curve at point A (which
has emergence angle f) is
dt sin f
= . (2.29)
dx v1
The emergence angles of the head waves are all the same, i.e., the criti-
cal angle f. The critical angle is given by equation (2.20), which we write as
v1
sin f = . (2.30)
v2
In particular, the slope of the head-wave time-distance curve is always
DOI:10.1190/1.9781560803737

dt sin f
= . (2.31)
dx v1
Equations (2.29) and (2.31) are identical. In other words, at point A, the
slope of the reflected time-distance curve is the same as the slope of the
head-wave time-distance curve. This common value is found by substituting
the critical angle, equation (2.20), into either equation (2.29) or (2.31). Then,
we obtain
   
dt sin f 1 −1 v1 1 v1 1
= = sin sin = = = constant . (2.32)
dx v1 v1 v2 v1 v2 v2

Let us now find the time-distance curve of the head wave, and the
relationship of this time-distance curve to that of the reflected wave.
Because the slope is constant, the head-wave time-distance curve is a
straight line. The horizontal velocity of the refracted wave is dx/dt = v2 .
At point A, the time-distance straight line of the head wave is tangent to
the time-distance hyperbola of the reflected wave. In summary, we have

1. The direct arrival. Its traveltime lines are two straight lines passing
through the origin with slopes 1/v1 and −1/v1 .
2. The reflected arrival. Its traveltime curve is a hyperbola which cuts the t
axis at 2z1 /v1 and with asymptotes as the traveltime lines of the direct
refracted arrival.
3. The refracted arrival from the first interface. Its traveltime lines are two
straight lines with slopes 1/v2 and −1/v2 . These lines are tangent to the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

56 Basic Wave Analysis, Part 1: Velocity Analysis

reflection time-distance curve at the points corresponding to critical

reflection and extend outward from these tangent points. We have
assumed that v2 . v1 . If v2 , v1 , then there are no refraction arrivals
from the interface.

Interval velocity and average velocity

A level is a straight-edged device with a sealed glass tube partially
filled with alcohol or other liquid, containing an air bubble whose
position reveals whether a surface is perfectly level. A bricklayer must con-
stantly use a level to make sure that the bricks are level at each stage of con-
struction. Over the course of ages, many sedimentary basins have areas in
which the sedimentary layers are approximately level. Various terms
can be used. We can say that their interfaces are level, or alternatively hori-
DOI:10.1190/1.9781560803737

zontal, or alternatively flat. An ideal condition in petroleum exploration is

the case of more-or-less flat-lying layers. The changes in wave velocity in
the horizontal direction are usually the result of slow changes in the
density and elastic properties of the strata. Thus, the changes in wave veloc-
ity in this direction typically are small. However, as one moves in the ver-
tical direction, the changes in wave velocity result from lithological
changes of the layering as well as increasing pressure due to increasing
depth. As a result, the vertical changes in wave velocity generally are
much more rapid than the horizontal changes. Because of the gradual
nature of the horizontal changes, the usual practice is to divide the prospect
into smaller sub-areas, within each of which the horizontal wave-velocity
distribution is assumed to be constant, so that the same vertical wave-
velocity distribution can be used over the entire sub-area. Then, the
interpreter must link these separate sub-areas together in constructing the
final map of the prospect.
For the case in which the interfaces are level, the mathematics of wave
propagation is simplified significantly. This model is known as the flat
layered earth model. A horizontal line represents the surface, and below
the surface there are media whose interfaces are parallel to the surface.
The media have various thicknesses and compressional velocities. The
path of a wave originating at the shot point and arriving at an offset point
after being reflected at depth satisfies certain parametric equations. The
first equation provides an expression for the offset distance, the second
equation yields the travel time, and the third equation represents Snell’s
law of refraction of the wave at each interface. From these equations, we
obtain the equation for the time-distance curve of the waves originating at
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 57

DOI:10.1190/1.9781560803737

Figure 2.9. Flat layered earth model.

the shot point and reflected back to the surface from the reflecting interface.
This earth model composed of horizontal layers, although a simplification
of the true situation in the earth, provides us with the equation of the
time-distance curve that is used in the interpretation of seismic records.
Next, consider the model illustrated in Figure 2.9. The top horizontal
line represents the surface under which there are N layers whose inter-
faces are parallel to the surface. The depth axis z points downward, with
z = 0 indicating the surface. We designate the interfaces of the N layers
by z0 = 0, z1 , z2 , . . . , zN .
Layer i is the layer which lies between interface zi−1 (on top) and inter-
face zi (on bottom). The thicknesses of the layers are
Dz1 = z1 − z0 , Dz2 = z2 − z1 , . . . , DzN = zN − zN−1 (2.33)

The compressional velocities within the layers are, respectively,

v1 , v2 , . . . , vN . According to standard geophysical terminology, velocity vi
is called an interval velocity. In other words, interval velocity vi is the
velocity in the interval between interface zi−1 (on top) and interface zi
(on bottom). An important consideration is the estimation of the interval
velocities from the seismic reflection data.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

58 Basic Wave Analysis, Part 1: Velocity Analysis

The vertical time T0 is

2Dz1 2Dz2 2DzN

T0 = t(0) = + + ··· + . (2.34)
v1 v2 vN
The average velocity is

2(Dz1 + Dz2 + · · · + DzN )

vA = . (2.35)
T0
A source is initiated at shot point S. Waves travel down into the earth
where they suffer reflection and refraction at the interfaces. We will consider
only one raypath, that is, the primary reflection event composed of the path
from shot point S down to depth point D, and the corresponding path from D
up to receiver R. This raypath is shown in Figure 2.10 (left). The receiver R
is offset a certain horizontal distance from the shot point. This distance
DOI:10.1190/1.9781560803737

represents the full offset. We can choose either leg and treat it as a separate
raypath. Figure 2.10 (right) shows the upgoing leg from depth point D to
receiver R. The receiver R is at a certain horizontal distance from the mid-
point M. In this case, this distance represents the half offset.
Now, look at the oblique downgoing raypath in layer i, shown in
Figure 2.11. The x-axis (pointing right) is the horizontal axis. The z-axis
(pointing down) is the vertical axis. The angle that the ray makes with the
vertical is ui . The fundamental right triangle has vertical leg Dzi , horizontal
leg Dx, and hypotenuse Dsi . The vertical leg is the layer thickness Dzi .
The horizontal leg is

Dxi = Dzi tan ui . (2.36)

Figure 2.10.
Two-way raypath
(left) and one-way
raypath (right).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 59

Figure 2.11. Downgoing raypath in layer 1.

DOI:10.1190/1.9781560803737

The hypotenuse is the oblique distance

Dzi
Dsi = . (2.37)
cos ui

Let Dti be the traveltime in layer i. We divide the oblique distance by the
velocity to obtain the traveltime in layer i:

Dzi
Dti = . (2.38)
vi cosui

Referring also to Figure 2.12, our immediate goal is to find the relation-
ship between horizontal distance

x = Dx1 + Dx2 + · · · + DxN (2.39)

and traveltime

t = Dt1 + Dt2 + · · · + DtN . (2.40)

The relationship between x and t will be expressed by a pair of parametric

equations. The horizontal distance x is the sum of the individual offset dis-
tances, equation (2.36) for all of the layers. The first parametric equation is


N 
N
x= Dxi = tan ui Dzi . (2.41)
i=1 i=1
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

60 Basic Wave Analysis, Part 1: Velocity Analysis

Figure 2.12. One-

way raypath.
DOI:10.1190/1.9781560803737

The total traveltime is the sum of the individual traveltimes, equation (2.38),
for all of the layers. The second parametric equation is

N N
Dzi
t= Dti = . (2.42)
i=1
v
i=1 i
cosui

At this point, we will define the ray parameter. For the laterally homo-
geneous media discussed here, the ray parameter represents a geometric
property of a seismic ray that remains constant throughout its path. In
such a case, the ray parameter is invariant in transmission, reflection, and
refraction. Snell’s law implies that the ratio of the sine of angle ui to velocity
vi is the same for each layer. Thus, the ray parameter (or Snell’s parameter)
p is defined as
sin ui
p= = constant for i = 1, 2, . . . , N . (2.43)
vi
The ray parameter is a number associated with the raypath in question. A
different raypath would have a different value for p.
We can use Snell’s law, equation (2.43), to replace the trigonometric
quantities in parametric equations (2.41) and (2.42). This yields

p
cos ui = 1 − p2 v2i and tan ui =  . (2.44)
1 − p2 v2i
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 61

Thus, the parametric equations (2.41) and (2.42) become

N
p vi Dzi 
N
Dzi
x=  and t=  . (2.45)
i=1 1 − p2 v2i i=1 vi 1 − p2 v2i

For N . 1, these equations cannot be solved explicitly for t in terms of

x; that is, the Snell’s parameter p cannot be eliminated by algebraic manip-
ulations. The curve t(x) is the time-distance curve. Sometimes we will write
the time-distance curve simply as t. However, in the case of a vertical ray
(i.e., when x = 0), we will write either t(0) or t0 .
We want to find an alternative expression for the time-distance curve for
a one-way path, as given by equations (2.45). The expression for x will be
the same as that given in equations (2.45). The expression for t is obtained
using equations (2.9). We have
N
sin ui cos ui
DOI:10.1190/1.9781560803737

t= Dxi + Dzi . (2.46)

i=1
vi vi

Snell’s law states that sin ui /vi = p = constant. We see that the hori-
zontal slowness, i.e., sin ui /vi , is the same for each layer, and, in fact, this
common horizontal slowness is Snell’s parameter p. Thus, we can take
the horizontal slowness p outside of the summation sign. However, the ver-
tical slowness,

cos ui 1 − p2 v2i
= , (2.47)
vi vi
is different for each layer, and thus we cannot take this factor outside of the
summation sign. We have

N N
1 − sin2 ui
t=p Dxi + Dzi
i=1 i=1
vi

or


N
1 − p2 v2i
t = px + Dzi . (2.48)
i=1
vi

This equation gives the alternative expression for the time-distance

curve. It represents a new way of representing the time-distance curve.
From expression (2.48), immediately we can see that
dt
=p. (2.49)
dx
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

62 Basic Wave Analysis, Part 1: Velocity Analysis

Equation (2.49) is an important equation, to which we will refer

often. It says that the slope of the time-distance curve is equal to Snell’s
parameter p.
The quantities in equation (2.48) represent a one-way path. For a two-
way path, we merely double the one-way quantities, which produces

N
p vi Dzi 
N
Dzi
x=2  and t=2  . (2.50)
i=1 1 − p2 v2i i=1 vi 1 − p2 v2i

The significance of Snell’s parameter p is realized by taking i = 1 in

equation (2.43), in which case we have

sin u1
p= . (2.51)
v1
DOI:10.1190/1.9781560803737

The first raypath segment is special, because it intersects the surface of the
earth, and so its properties can be obtained by surface measurements.
Figure 2.13 and Figure 2.14 again depict the fundamental seismic time-
distance relationship. Figure 2.13 indicates that the slope of the time-dis-
tance curve at any point is equal to the value of the parameter p correspond-
ing to that point. We want to establish this result. In Figure 2.14, let the
upcoming ray have an angle of emergence u1 . We see that the wavefront
makes an angle u1 with the surface of the earth. A wavefront coming up
the ray meets the surface at time t. Because of the slant of the raypath, we
see that a wavefront coming up the adjacent ray will meet the surface at

Figure 2.13. The

slope dt/dx of the
time-distance
curve at a point is
equal to the value
of Snell’s
parameter p at that
point.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 63

Figure 2.14. The

wavefront angle u at
the earth’s surface.
DOI:10.1190/1.9781560803737

Figure 2.15.
Example of a one-
way raypath.

time t + dt. Thus, the extra path length is v1 dt. The extra offset is dx. We
thus have
v1 dt
sin u1 = . (2.52)
dx
Snell’s parameter p is sin u1 /v1 . Thus, the foregoing equation produces the
desired result:
dx sin u1
= =p. (2.53)
dt v1
Figure 2.15 illustrates an example of the downgoing leg in the case of
two horizontal layers. The first layer is 1.0 km thick and has an interval
velocity of 2 km/s. The second layer is 1.5 km thick and has an interval ve-
locity of 3 km/s. We consider three rays, with Snell’s parameter p equal to
0, 1/12, and 1/6, respectively. The angles of these one-way rays, shown in
Figure 2.15 are obtained from the equations in Table 2.1.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

64 Basic Wave Analysis, Part 1: Velocity Analysis

Table 2.1. Determination of angles of one-way rays, using Figure 2.15.

sin u1 sinu2
Ray A: p=0= = Thus u1 = 0, u2 = 0
2 3
1 sinu1 sinu2
Ray B: p= = = Thus u1 = 9.59o , u2 = 14.48◦
12 2 3
1 sinu1 sinu2
Ray C: p= = = Thus u1 = 19.47o , u2 = 30.00◦
6 2 3

Table 2.2. Determination of horizontal

distance of one-way rays, using Figure 2.15.

Ray A: Dx1 = (Dz1 ) tan 0◦ = (1.0) tan 0◦ = 0

Dx2 = (Dz2 ) tan 0◦ = (1.5) tan 0◦ = 0
x = Dx1 + Dx1 = half-offset = 0
DOI:10.1190/1.9781560803737

Ray B: Dx1 = (1.0) tan 9.59◦ = 0.169

Dx2 = (1.5) tan 14.48◦ = 0.387
x = half-offset = 0.556
Ray C: Dx1 = (1.0) tan 19.47◦ = 0.353
Dx2 = (1.5) tan 30.00◦ = 0.886
x = half-offset = 1.219

The horizontal distance x for each of these one-way rays are shown in
Table 2.2.
The one-way traveltimes along the raypaths are shown in Table 2.3.

Oblique and conventional root-mean-square velocity

Next, let us specify that we have an oblique travel path, and not merely
the special case of a vertical travel path. Recall the first parametric equation
given as equation (2.41). It may be rewritten as


N    N
Dzi
x= (sin ui ) = ( pvi )(vi Dti ) , (2.54)
i=1
cos ui i=1

which is

N
x=p v2i Dti . (2.55)
i=1
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 65

Table 2.3. Determination of one-way

traveltimes, using Figure 2.15.

Ray A: Dz1 1.0

Dt1 (0) = = = 0.500
v1 2
Dz2 1.5
Dt2 (0) = = = 0.500
v2 3
t = vertical one-way traveltime = 1.000
Ray B: Dt1 (0)
Dt2 (x) = = 0.507
cos 9.59◦
Dt2 (0)
Dt2 (x) = = 0.516
cos 14.48◦
t = oblique one-way traveltime = 1.023
Ray C: Dt1 (0)
Dt2 (x) = = 0.530
cos 19.47◦
Dt2 (0)
Dt2 (x) = = 0.577
DOI:10.1190/1.9781560803737

cos 30.00◦
t = oblique one-way traveltime = 1.107

This equation states that horizontal distance x is equal to parameter p times

the summation of the squared velocities weighted by the traveltime incre-
ments. If we normalize this summation by the total traveltime t(x), we
obtain a quantity that is called the oblique mean-square velocity. Its square
root, called the oblique root-mean-square (ORMS) velocity, is given by

1  N N
vORMS =  v2i Dti (x) where t(x) = Dti (x) . (2.56)
t(x) i=1 i=1

This velocity is different for each raypath; that is, the ORMS velocity
depends upon Snell’s parameter p. In terms of ORMS velocity, the time-
distance curve is specified by the parametric equations

N 
N
Dti (0)
x = ptv2ORMS , t = Dti = . (2.57)
i=1 i=1
cos ui

In the case of normal incidence (i.e., a vertical raypath), Snell’s param-

eter p is zero, and the oblique RMS velocity reduces to the conventional
RMS velocity, defined as

1  n N
vRMS = v2i Dti (0) where t0 ; t(0) = Dti (0) . (2.58)
t(0) i=1 i=1
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

66 Basic Wave Analysis, Part 1: Velocity Analysis

Continuing our example in the preceding section, we have the

following:
Ray A:

22 (0.5) + 32 (0.5)
vRMS = = 2.550
0.5 + 0.5
(conventional RMS velocity for p = 0) , (2.59)
Ray B:
 
x 0.556
vORMS = =   = 2.554
pt  1
(1.023)
12
 
1
oblique RMS velocity for p = , (2.60)
12
DOI:10.1190/1.9781560803737

Ray C:
 
x 1.2196
vORMS = =   = 2.570
pt  1
(1.107)
6
 
1
oblique RMS velocity for p = . (2.61)
6
We see that, as the angle of issue increases, the ORMS velocity increases,
because the ray spends an increased proportion of its journey in the lower
(high-velocity) layer.

Snell’s parameter
Figures 2.16 and 2.17 again depict the fundamental seismic time-dis-
tance relationship. Before we proceed with the hyperbolic approximation
problem, it would be helpful to understand more about the time-distance
curve for the horizontally layered model (see Figure 2.16). One important
property is that the slope of the time-distance curve at any point is
equal to the value of the parameter p corresponding to that point (see
Figure 2.17). To establish this result, let the upcoming ray have an angle
of emergence u1 , with the normal to the surface of the earth. Hence, the
wavefront makes an angle u1 with the surface of the earth. A pulse
coming up the ray meets the surface at time t. Because of the slant of the
raypath, we see that a pulse coming up the adjacent ray will meet the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 67

Figure 2.16. The

slope of the time-
distance curve at a
point is equal to the
value of Snell’s
parameter p at that
point.
DOI:10.1190/1.9781560803737

Figure 2.17.
Emergent wave at
a level surface.

surface at time t + dt. Thus, the extra path length is vi dt. The extra offset is
dx. Therefore, we have

v1 dt
sin u1 = . (2.62)
dx
Snell’s parameter p is defined as (sin u1 )/v1 . Thus, the foregoing
equation produces the required result:

dx sin u1
= =p. (2.63)
dt v1
Snell’s parameter p represents the slowness of the wavefront along the
horizontal interface (i.e., the horizontal axis or x-axis). Snell’s parameter
p has units of inverse of velocity and is measured in units of
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

68 Basic Wave Analysis, Part 1: Velocity Analysis

milliseconds per meter (i.e., seconds per kilometer). The quantity dx/dt is
called the phase velocity of the wavefront along the horizontal interface.
In other words, the quantity dx/dt is the horizontal phase velocity.
The parameter p is equal to the reciprocal of the horizontal phase velocity;
that is,

1 Dt
p= = = horizontal slowness . (2.64)
Dx Dx
Dt

A geophysicist (who is constrained to make measurements at the surface

of the ground) cannot directly observe either the wave velocity v(z) or the
propagation angle u (z) for any depth z. However, Snell’s parameter p can
be measured. The reason is the horizontal slowness (i.e., inverse horizontal
velocity) at which the intercept of a wavefront with the surface of the ground
DOI:10.1190/1.9781560803737

moves along the surface; that is,

sin u dt 1 1
p= = = = . (2.65)
v dx dx horizontal velocity at z = 0
dt

If we trace a curving ray traveling by a refraction path from a source to a

receiver, Snell’s parameter p is constant for the entire ray path; that is

sin u1 sin u2 sin uN

p= = = ··· = . (2.66)
v1 v2 vN

In other words, Snell’s parameter p for the entire seismic ray path is the reci-
procal of the horizontal velocity dx/dt of the wavefront at the surface of the
ground. Snell’s law says that the wave disturbance as apparent at any depth z
also moves horizontally at the same velocity as at the surface. Thus, Snell’s
law can be obtained as a consequence of the fact that the horizontal velocity
dx/dt at one depth must equal the horizontal phase velocity at any other
depth. This result holds in any medium in which velocity v(z) is a function
of depth z only.
In summary, in the model with horizontal interfaces, Snell’s parameter
p is the ratio of the sine of the angle between a ray and the normal to an
interface divided by the velocity in the layer. Snell’s law states that this
parameter is constant on the transmission of the ray from layer to layer.
As we go vertically down in depth, the velocity function changes at each
interface. Because the interfaces are level, the velocity function is the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 69

same no matter at what horizontal point x we take. Thus, the velocity func-
tion does not depend upon x, but is a step function v(z) of depth z only. In
the limiting case where the flat layers become infinitesimally thin, we
obtain a more general heterogeneous medium in which the velocity v(z) is
a function of depth z only. The angle u (z) is also a function of z. It
follows that Snell’s parameter p is given by a constant p. This constant,
which is independent of z, is
sin u (z)
p= . (2.67)
v(z)
The time-distance curve is symmetrical; that is,
t(x) = t(−x) . (2.68)
Thus, we may confine ourselves to non-negative values of Snell’s parameter
p. When the receiver point R is directly above depth point D (i.e., when
DOI:10.1190/1.9781560803737

x = 0), the raypath is not oblique. It is vertical. We see that u1 = 0, so

p = 0. In such a case, we have the vertical travel path given by

n
Dzi
x = 0, t0 ; t(0) = . (2.69)
i=1
vi

The one-way traveltime t(0) is the one-way traveltime in the vertical direc-
tion. We also use the symbol t0 for t(0). It is shown to be the sum of the one-
way vertical traveltimes in each of the layers.
Let vm be the greatest of the interval velocities v1 , v2 , . . . , vn . As an exer-
cise, show that the time-distance curve has as its asymptote as the line whose
equation is

 2
 n
1 vi
vm t − x = vm 1− . (2.70)
v
i=1 i
vm

Linear approximation
The early days of seismic analysis has curves fitted to the data as an
aid for visualization. The data is plotted on graph paper, and the curves
are fitted by hand. Much emphasis is given to methods of constructing a
curve, or mathematical function, that has the best fit to a series of data
points. Curve fitting can involve either interpolation, where an exact fit to
a specific set of data is required, or statistical regression, in which a
smooth function is constructed that approximately fits over a range of
data points. Today, in the age of digital seismic processing, curve fitting
is still essential. More complicated methods are used. In order to understand
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

70 Basic Wave Analysis, Part 1: Velocity Analysis

what is happening inside the computer, knowledge of the simple methods is

necessary. Simple methods help us grasp the essence of intricate problems.
Geophysicists can attack intricate seismic problems, largely because they
know that simple seismic problems have solutions that work. As Leonardo
da Vinci has said, “Simplicity is the ultimate sophistication.”
The most readily available mathematical techniques are linear. The only
recourse in the early days is to convert nonlinear things into linear things in
order to handle them mathematically. Generally, such a transformation
cannot hold in the large scale but is valid only in a small region surrounding
the point of greatest interest. The Taylor series is very important in geophys-
ics because it offers a way to find a linear approximation to a function.
Linear problems are relatively easy to solve. Non-linear problems require
repeated linearized solutions.
Indeed, the development of our field was indebted to Brook Taylor
(1685 –1731) for the creation of the Taylor series, among other things.
DOI:10.1190/1.9781560803737

Taylor was born in Edmonton, near London, and he entered St. John’s
College, Cambridge, in 1701. In 1708, he obtained a solution to the
problem of the “center of oscillation.” In 1715, Taylor’s Methodus Incre-
mentorum Directa et Inversa added a new branch to higher mathematics:
the calculus of finite differences. Much of the numerical analysis completed
on a digital computer makes use of finite differences. Taylor was the one
who first linked wave motion to physical properties in the form of the
wave equation, and he developed the Taylor series, which is a representation
of a function as an infinite sum of terms that are calculated from the values of
the function’s derivatives at a single point. (Notably, James Gregory also
worked in this area and published several such series.) In 1715, Brook
Taylor gave a general method for constructing Taylor series for all functions
for which they exist. His method was based on the idea of linear perspective,
which is a type of perspective used by artists in which the relative size,
shape, and position of objects are determined by drawn or imagined lines
converging at a point on the horizon. In his 1715 essay, Linear Perspective,
Taylor set forth its true principles in an original and more general form than
any of his predecessors.
Once again, the great minds of the past were developing the ideas that
would be used by many generations to come. The Swiss family Bernoulli
burst upon the mathematical scene with an overwhelming force, and few
could challenge them. In England, only Sir Isaac Newton, Brook Taylor,
and possibly Roger Cotes could qualify. Roger Cotes (1682– 1716)
worked with Newton by proofreading the second edition of the Principia.
Cotes also invented the quadrature formulas (known as Newton– Cotes
formulas), and he first introduced what is known today as Euler’s
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 71

formula. Unfortunately, the effect of Taylor’s demonstrations often was lost

through his failure to express his ideas fully and clearly; fortunately, the
application of the Taylor series remained.
A Taylor’s series is an important tool for analysis, because its finite
number of terms may be used to approximate a function. Errors introduced
by the use of the approximation are addressed by the quantitative estimates
given by Taylor’s theorem. Taking some initial terms of the Taylor series, a
polynomial is determined (i.e., a Taylor polynomial). Then, the Taylor
series of a function is the limit of that function’s Taylor polynomials as
the degree increases, for a case in which a limit exists. It should be noted
that, even if its Taylor series converges at every point, a function may not
be equal to its Taylor series. Instead, a function that is equal to its Taylor
series in an open interval (or a disc in the complex plane) is known as an
analytic function in that interval. Indeed, analytic functions gain predomi-
nance following the work Augustin-Louis Cauchy (1789– 1857), who is
DOI:10.1190/1.9781560803737

one of the first to state and prove theorems of calculus rigorously, rejecting
the heuristic methods of earlier authors.
Now, applying these ideas to our discussion of linear approximation, we
can begin by stating that the single-horizontal-interface time-distance curve
is the hyperbola
x2
t2 − = t02 . (2.71)
v2
We can convert this curve into a straight line by the transformation
u = x2 , w = t2 . (2.72)
The hyperbola (2.71) becomes the straight line
u
w = 2 + t02 . (2.73)
v
This line has slope 1/v2 and w-intercept t02 . Graphically, straight lines are
much easier to work with than hyperbolas. In fact, it is not necessary here
to consider only one interface, because the conversion works in the large
scale.
When we consider two or more horizontal interfaces, the time-distance
curve is no longer a hyperbola but a higher-order curve. However, it is
almost hyperbolic in shape. Thus, the transformation equation (2.72) con-
verts it into an almost straight line, curved slightly downward as u increases.
To make things simple in our geophysical analysis, we would like to
approximate this slightly curved line by a straight line. Such an approxi-
mation cannot hold in the large scale, but we can pick some point at
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

72 Basic Wave Analysis, Part 1: Velocity Analysis

which we want good approximation. The accuracy will diminish as we move

away from that point. Let us call the required point (u1 , w1 ). We will pick the
straight line as the one that goes through this point and has the same slope as
the slightly curved line at this point. These two conditions are enough to
determine the straight line (the local Taylor approximation at non-zero x).
The parametric equations for the horizontal-interface time-distance
curve t(x) are


N 
N
Dti (0)
x = ptv2ORMS , t = Dti = . (2.74)
i=1 i=1
cos ui

Under the transformation equation (2.72), the curve

u = x2 , w = t2 (x) (2.75)
DOI:10.1190/1.9781560803737

is a slightly curved line. The slope of this line is

dw dt2 dx dt(x) 1
= = 2t(x) . (2.76)
du dx du dx 2x
dt(x)
We know that the slope of the time-distance curve is p = .
Thus, equation (2.76) becomes dx

dw 1 t(x)p
= 2t(x)p = . (2.77)
du 2x x

The left side of equations (2.74) is

x = ptv2ORMS .

Thus, equation (2.77) becomes

dw 1
= . (2.78)
du v2ORMS

Therefore, we see that the slope of the slightly curved line is equal to the
reciprocal of the square of the oblique root-mean-square (ORMS) velocity.
In Figure 2.18, let us pick the required point as the w-intercept of
the slightly curved line; that is, u = 0, w = t02 , where t(0) is the vertical
one-way time. Snell’s parameter p for this point is zero, and vORMS
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 73

Figure 2.18. The

approximating
straight line
(dashed line) is
tangent to the
slightly curved
line (solid line)
at the point
(u, w) = (0, t02 ).

reduces to the usual RMS velocity given by


DOI:10.1190/1.9781560803737

1  n N
vRMS = v2i Dti (0) where t(0) ; t0 = Dti (0) . (2.79)
t(0) i=1 i=1

The desired straight line, therefore, is the one that goes through the point
u = 0, w = t02 and has slope 1/v2RMS . Its equation is

u
w= + t02 . (2.80)
v2RMS

Converting back to x and t, we see that this straight line becomes the
hyperbola

x2
t2 − = t02 . (2.81)
v2RMS

This hyperbola, depicted in Figure 2.19, is the time-distance curve for a

single-layered medium, with vertical two-way time t0 and constant velocity
vRMS . The given approximation is one in which the N-layered medium is
replaced by a single-layered medium whose velocity is given by the root-
mean-square velocity of the N-layered medium.
As a second example, let us pick the required point as (u1 , w1 ), where u1
is not zero (see Figure 2.20). Snell’s parameter p will not be zero, and there
is an oblique root-mean-square velocity corresponding to this value of p.
Under the usual condition of velocity increasing with layer depth, vORMS
will be greater than vRMS . The desired straight line is the one that goes
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

74 Basic Wave Analysis, Part 1: Velocity Analysis

Figure 2.19. The

approximating
hyperbola (dashed
line) is tangent to
the time-distance
curve t(x) (solid
line) at the point
(x, t) = (0, t0 ).

Figure 2.20. The

approximating straight
line (dashed line) is
DOI:10.1190/1.9781560803737

tangent to the slightly

curved line (solid line)
at (u1 , w1 ).

Figure 2.21. The

approximating
hyperbola (dashed
line) is tangent to
the time-distance
curve t(x) (solid
line) at x = +x1 .

through the point (u1 , w1 ), and has slope 1/v2ORMS ; that is,
u − u1
w − w1 = . (2.82)
v2RMS

Converting back to x and t, we obtain the hyperbola (depicted in

Figure 2.21)

x2 x21
t2 − = t12 − , (2.83)
v2ORMS 2
vORMS
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 2: Time, Distance, and Velocity 75

where x1 and t1 are defined by u1 = x21 and w1 = t12 . This hyperbola is

the time-distance curve for a single-layered medium, with vertical two-
way time

x2
T0 = t12 − 2 1 (2.84)
vORMS

and constant velocity vORMS . In this single-layered approximation, the

two-way vertical time is different from the two-way vertical time t0 in the
given N-layer medium.
In seismic data processing, neither of the two foregoing approximations
are used. Instead, a straight line
u
w= + t02 (2.85)
v2S
DOI:10.1190/1.9781560803737

would be fitted according to a coherency criterion. Actually, the fitting is

accomplished in the x, t domain, so that, in fact, the corresponding hyper-
bola is fitted by a coherency criterion to each possible hyperbola-like
reflected event appearing on the CMP gather of traces. The empirically
determined quantities vs and ts are called the stacking velocity (also called
NMO velocity) and stacking time, respectively. However, vs and ts are
usually interpreted as the RMS velocity vRMS and two-way vertical time
t0 , respectively. Such an interpretation can only be an approximation, as
has been shown. In the next chapter, we will discuss how the stacking veloc-
ities are computed.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms
DOI:10.1190/1.9781560803737
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3

Velocity Estimation

“Measure what can be measured, and make measureable

what cannot be measured.”
DOI:10.1190/1.9781560803737

—Galileo Galilei

Four important corrections

An important step in the processing sequence consists of making
traveltime adjustments, or corrections. Time adjustments are characterized
either as static corrections or as dynamic corrections. For a static correction,
a time shift (or translation) is applied to an entire trace; that is, regardless of
record time or reflector depth, a constant time correction term is added to or
subtracted from all reflection times. In contrast, dynamic corrections vary
with record time and therefore depend on reflector depth.
First, let us consider static corrections. A region of very low velocity
occurs in a near-surface (or weathered) layer that extends from the earth’s
surface to a depth of tens to hundreds of meters. In the near-surface layer,
velocities change either gradually or abruptly from values of about 600 to
1500 m/s or greater. Time delays associated with this weathered layer
disrupt reflection continuity (i.e., trace-to-trace alignment) and represent a
significant challenge when processing data acquired on land. In addition,
the lateral variations, including elevation changes and near-surface inhomo-
geneities, severely degrade trace-to-trace continuity. To improve trace-to-
trace continuity, each trace can be corrected by a time shift, which has the
effect of placing the source and receiver on a fictitious horizontal datum
plane. This time shift is additively composed of a source correction and a
receiver correction, which together are referred to as static corrections.

77
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

78 Basic Wave Analysis, Part 1: Velocity Analysis

As we will demonstrate, the purpose of static computations is to remove

time variations caused by anomalous conditions at the earth’s surface. In
comparison, typically seismic lines recorded at sea will not require static
corrections because the water layer is largely uniform at constant elevation.
Even so, it is becoming more common to use “water column statics” to
correct for variations in the velocity of the water layer.
When performing a correction, the underlying assumption is that the
resulting model trace provides the information that would have been
recorded if the sources and receivers had been displaced vertically down-
ward to a reference plane with no weathering material present. This time
delay is assumed to be surface-consistent, i.e., it represents the sum of a
source-related component and a contribution characteristic of a given
receiver position. The significant success of modern static correction
programs appears to validate these assumptions in the general sense. A
more complete analysis, however, shows that the near surface layers
DOI:10.1190/1.9781560803737

behave as complicated filters, the impulse responses of which can distort

amplitude and phase characteristics of seismic wavefronts.
The dynamic corrections convert each trace to the equivalent trace that
would have been received if the source and receiver were at the same lateral
point—the point midway between the actual source and receiver locations.
In this conversion, we are referring only to traces composed of so-called
primary reflected waves. According to ray theory, a primary reflection
event results from a raypath down from the source to the reflecting
horizon and then directly upward to the receiver. Given the velocity func-
tion, this raypath can be computed by means of Snell’s law. If the layers
are horizontal, then all of the reflection points (or depth points) are
always directly beneath the midpoint between source and receiver. If the
layers are dipping, then the depth points are offset from the midpoint
(Robinson, 1977). Thus, the dynamic corrections depend on both the veloc-
ity function and the dip of the reflecting beds. The component of the correc-
tion caused by the separation of source and receiver is called the normal
moveout (NMO) correction and the component caused by dip is called the
dip correction. In conclusion, four important corrections can be performed
on recorded seismic traces: source correction, receiver correction, NMO
correction, and dip correction.

Multiple coverage
During the Crimean War (1854– 1856), the British wanted to capture the
Russian naval base at Sebastopol in the Crimea, and they required more
effective artillery. They turned to Robert Mallet (1810–1881), the son of
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 79

the owner of a Dublin iron foundry. Notable achievements supplied by the

foundry included ironwork for the railway network, a lighthouse, and a
swing bridge over the River Shannon, all located in Ireland. For this
project, Mallet designed a 43-ton mortar of 0.91 m caliber capable of throw-
ing a 1100 kg shell a distance of 2.4 km. This huge mortar actually arrived
too late to be used in action, even though it was built in sections to allow
transport. Fortunately, Mallet did not confine his work to foundry and
later became known for his work in seismology. In fact, he originated
such terms as seismology, isoseismal map, and epicenter. His seminal
paper, On the Dynamics of Earthquakes (1846), laid the foundation of
modern seismology. Working with his son, John, from 1852 to 1858,
Robert Mallet prepared The Earthquake Catalogue of the British Associ-
ation (1858) and conducted blasting experiments to determine the speed
of seismic propagation in sand and solid rock.
During the time that Robert Mallet was conducting his work in seismol-
DOI:10.1190/1.9781560803737

ogy, a great calamity occurred. On 16 December 1857, Padula, Italy was

devastated by an earthquake which caused 11,000 deaths. This disaster
was almost a repetition of the Lisbon earthquake one century earlier,
which motivated John Michell. Once again, history repeated itself and a dis-
aster helped move science forward. Located 100 km south-east of Salerno,
Padula was Mallet’s destination, from which he wrote, Report on the Great
Neapolitan Earthquake of 1857. For this major scientific work, Mallet made
use of the new tool of photography to record the devastation caused by the
earthquake. In 1862, he published two volumes; in these, he estimated that
the earthquake’s hypocenter was about 8 to 9 geographical miles deep.
Indeed, the science of what lies beneath continued to progress following
the work of Mallet.
When one looks out over an expanse of land or ocean and considers
the problem of discovering what lies below, the scope of seismic explo-
ration is realized. Of course, we cannot make a continuous survey of each
and every point but must pick relatively few points that are representative
of the entire area. Such a procedure represents sampling in the space
domain.
So, how do we determine our survey? Consider the multiple coverage
method. Let us envisage a single horizontal reflecting plane at depth. We
suppose that our exploration efforts are restricted to a straight line along
the surface. We focus our attention on a single point D on the horizontal
reflecting plane under the surface line. In order to record a reflection from
a point, the shot s and the detector r must be placed in symmetric positions
(relative to a specified coordinate system) with respect to the surface point m
directly above the subsurface point D.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

80 Basic Wave Analysis, Part 1: Velocity Analysis

Figure 3.1 shows three possible shots and the corresponding detectors
required to record a reflection from point D. The surface point m is the
midpoint between s1 and r1, and also the midpoint between s2 and r2, and
also between s3 and r3. That is,
s1 + r1 s2 + r2 s3 + r3
m= = = . (3.1)
2 2 2
The corresponding offsets are
x1 = r1 − s1 , x2 = r2 − s2 , x3 = r3 − s3 . (3.2)
If this setup is used, we record three separate reflected waves from the given
depth point D. We have, in effect, sampled depth point D three times; that is,
the diagram represents a three-fold coverage of depth point D.
Why do we want to do
this? Why is one sample of
DOI:10.1190/1.9781560803737

depth point D not enough?

There are several important
reasons for the use of
multiple-fold coverage. One
reason is that it provides
ways of suppressing signal-
generated noise (e.g., mul-
tiple reflection, refraction,
and diffraction). Another
reason is that it provides
ways of estimating the velo-
Figure 3.1. Multiple coverage and common
city of the seismic waves.
midpoint m. We will have much to say
about these matters.
Figure 3.2 illustrates a
typical geometry by which
seismic reflection data is
recorded. A single impulsive
source and several detectors
are spaced evenly along a
line on the surface of the
ground. A single shot will
instigate the recording of
a seismic trace at each
detector position. We let
Figure 3.2. One-sided spread of receivers. s represent the horizontal
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 81

coordinate of the source (shot) and r the horizontal coordinate of the receiver
(detector). Two additional descriptive coordinates are defined, i.e., the
midpoint m ¼ (s + r)/2 between the source and the receiver and the offset
x ¼ r 2 s from the source to the receiver.
The arrival time of a reflection from a given horizontal plane interface is
given by the time-distance curve t(x). In the case of a horizontal reflector,
the same time-distance curve describes the traveltime curve for both the
geometry with source s a constant, as shown in Figure 3.3, and the geometry
with midpoint m a constant, as shown in Figures 3.4 and 3.5. The traces

a)

Figure 3.3.
Common shot
point (CSP)
gather:
DOI:10.1190/1.9781560803737

(a) geometry and

(b) data display.
b) The dashed
curve shows the
locus of the
reflected waves
from the given
horizontal
reflecting plane.

Figure 3.4. Geometry of common midpoint (CMP) gather.

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

82 Basic Wave Analysis, Part 1: Velocity Analysis

DOI:10.1190/1.9781560803737

Figure 3.5. Data display of a CMP gather. Trace r1 results from shot s1, trace r2
from shot s2, etc. The curve shows the locus of the reflected waves from the given
horizontal reflecting plane.

with a common source s comprise a common shot point (CSP) gather. The
traces with a common midpoint m comprise a common midpoint (CMP)
gather. The CSP gather is realized from a single shot. However, a CMP
gather requires several shots. With reference to Figure 3.4, we see the r1
trace results from shot s1, the r2 trace from shot s2, and so on. The field pro-
cedure is shown in Figure 3.6. The traces from shot s1 are recorded at
receiver positions r1, r3, r5, r7. The distance between r1, r3 or between r3,
r5 or between r5, r7 is called the group interval. Then, the entire experiment
is moved to the right by one-half group interval. (The reason for the word
“group” is that the seismic receiver is composed of a group of geophones
spread out in an array or pattern.) Thus, the traces from shot s2 are
recorded at receiver positions r2, r4, r6, r8. The field procedure is
repeated in this manner, resulting in seismic traces for many offsets for
any midpoint. A collection of such traces for the given midpoint comprise
the CMP gather.
A convenient way to display this field procedure is shown in
Figure 3.7. Each point represents a seismic trace for a typical recording geo-
metry. The seismic source moves upward. As the source moves ahead, each
one-half group interval (i.e., where the group interval is defined as the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 83

Figure 3.6. Geometry for two successive shots s1 and s2 showing data redundancy
that allows us to compile CMP gathers. For example, the CMP gather for a midpoint
at location r2 would include trace s1r5 and trace s2r4.
DOI:10.1190/1.9781560803737

Figure 3.7.
Relationship among
the horizontal
(surface of earth)
coordinates r, s, m, x.
Each point
represents a seismic
trace.

distance between two adjacent receivers), the source sets off a source signal
and the receivers record the resulting traces. From the geometry, there are
four types of common point gathers of traces: the common shot point
(CSP), the common receiver point (CRP), the common midpoint (CMP),
and the common offset point (COP). Displays of the traces of such
gathers are called profiles or sections. Note these additional specifications:
COP means x = (s − r)/2 is constant; CMP means m = (s + r)/2 is
constant; CSP means s is constant; and CRP means r is constant.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

84 Basic Wave Analysis, Part 1: Velocity Analysis

Common midpoint
Time-distance and equi-time
curves can be interpreted as conic
sections, as follows. Consider
Figure 3.8, which shows a cone
cut by a plane perpendicular to the
axis. The resulting circle is the
equi-time curve (isochrone) of a
zero-offset event. Next, the cone
is cut by a plane parallel to the
axis; the resulting hyperbola is the
time-distance curve of the zero-
offset event. Finally, the cone is
Figure 3.8. Conic section. cut by a slanting plane, and the
DOI:10.1190/1.9781560803737

resulting ellipse is the equi-time

curve for a non-zero-offset event.
Next, as we continue this discussion, we will arrive at the defini-
tion for the term normal moveout (NMO). First, consider a horizontal
(i.e., non-dipping) multi-layer model as shown in Figure 3.9a. Each
interface is depicted by a horizontal line. A depth point D on interface
3 is shown, together with the point m on the earth’s surface directly
above D. Using M as a CMP, various source si and receiver ri pairs
(i = −3, −2, −1, 0, 1, 2, 3) are shown. On each trace, we only consider
the primary reflection event from depth point D. Of course, in actuality,
there also would be many other events on each trace due to primaries
from other interfaces as well as multiples from all of the interfaces. The
raypaths of some of the primaries, such as from source s2 to receiver r2,
via reflection point D, are drawn. In Figure 3.9b, the time-distance curve
is shown. It plots the (two-way) arrival time of the primary reflection
versus receiver position ri.
The traveltimes of the reflected pulses from D vary from trace to
trace depending upon the raypath taken. Instead of using receiver positon
r as the horizontal axis of this traveltime curve, we can use offset distance
x = r − s as the horizontal axis. Offset distance is the distance from the
source to the receiver, and thus the total horizontal distance that the
seismic pulse travels. We have drawn the offset axis below the receiver-pos-
ition axis in Figure 3.9. Thus, the time-distance curve can be written as t(x).
It now plots the (two-way) arrival time of the primary reflection event versus
offset distance x.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 85

a)

Figure 3.9.
(a) Primary
reflection events
from interface 3,
and (b) the time-
distance curve t(x)
DOI:10.1190/1.9781560803737

or these primary
b) reflection events.

We know that the slope of the time-distance curve t(x) is equal to Snell’s
parameter p; that is,
dt(x)
=p. (3.3)
dx
As a result, the traveltimes t(x) vary in a smooth manner with offset x. Thus,
we can connect the set of empirical points (xi, ti) of primary events due to
reflection point D in the form of a smooth curve. The curve of traveltime
t versus offset x (i.e., the curve drawn through the seismic reflection
events from interface 3) is the time-distance curve t(x) for the geometry
shown in Figure 3.9. This geometry provides all of the traces that are
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

86 Basic Wave Analysis, Part 1: Velocity Analysis

centered around a fixed common midpoint M. We recognize this layout at

the CMP gather. Thus, the t(x) curve is the CMP time-distance curve for
interface 3.
Another important concept is that of zero-offset traveltime. The zero-
offset traveltime is the two-way traveltime associated with midpoint M.
Because offset x is zero at M, the zero-offset time is t(0), i.e., the traveltime
from M to D and back to M. Now, we finally can define normal moveout
(NMO) in the general case. The normal moveout Dt is defined as the differ-
ence between t(x) and t(0). That is, NMO is
Dt = t(x) − t(0) . (3.4)
Equation (3.4) represents the NMO correction (also known as dynamic
correction). It is a time correction applied to reflection times t(x) to make
them all compatible with t(0).
Seismic time as shown on seismic records is always two-way time; that
DOI:10.1190/1.9781560803737

is, seismic time is the time from the source to the depth point plus the time
from the depth point to the receiver. Often, we want to make use of one-way
time; for example, the traveltime from depth point to receiver. In the case
of flat layers, which we are considering here, the one-way time is simply
one-half of the two-way time.
The simplest case in seismic prospecting is that of a single flat layer,
as shown in Figure 3.10. The source is S, the depth point is D, and the
receiver is R. The image S′ of S with respect to the interface is shown
(i.e., the interface is the perpendicular bisector of the line SS′ ). Because
SD equals S′ D, the travel distance SDR is equal to the line S′ R. This
line is equal to vt(x), where v is velocity and t(x) is the value from the
time-distance curve. The zero-offset traveltime is S′ S, which is equal to
vt(0). The horizontal distance SR is the offset x. By the Pythagorean
theorem, we have
v2 t2 (x) = v2 t2 (0) + x2 , (3.5)
which is the equation for the hyperbola
x2
t2 (x) − = t2 (0) . (3.6)
v2
Thus, in the one-interface case, the time-distance curve for the CMP
gather is exactly a hyperbola. The hyperbola is symmetric about the
common shot point x = 0; that is, t(x) = t(−x). The NMO is

x2
Dt = t(x) − t(0) = t2 (0) + 2 − t(0) . (3.7)
v
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 87

Figure 3.10. The case

of a single flat
reflector.
DOI:10.1190/1.9781560803737

Various approximations to this expression often are used. The square

root expansion

√ 1 1
1 + A = 1 + A − A2 + · · · for |A| ≤ 1 (3.8)
2 8
yields

x2
Dt = t(0) 1 + 2 2 − t(0)
v t (0)
 
x2 x4 x2 x2
= 2 − + ··· = 2 1 − 2 2 + ··· . (3.9)
2v t(0) 8v4 t3 (0) 2v t(0) 4v t (0)

If we take only the first term in this expansion, we produce the

approximation
x2
Dt = t(x) − t(0) ≈ . (3.10)
2v2 t(0)
From approximation (3.10), we see that NMO increases as the square of
offset x, inversely as the square of velocity v, and inversely as the vertical
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

88 Basic Wave Analysis, Part 1: Velocity Analysis

DOI:10.1190/1.9781560803737

Figure 3.11. The case of a single straight-line dipping interface. Note that point S′
is not a physical point that lies below the interface but merely a mirror image of
physical point S which lies above the interface.

traveltime t(0). The second derivative of the time-distance curve t(x) is

always positive. It increases rapidly as more distant geophones are used,
and it becomes less for deeper horizons.
The concept of normal moveout provides the main criterion by which
we can decide whether an event observed on a seismic record is a primary
reflection and not a multiple reflection or a diffraction event. If the NMO
for an observed event differs from the value required for primary reflection
by more than an allowable experimental error, then we cannot treat the event
as primary reflection.
To this point, we have addressed only the case of flat-lying beds.
Next, we consider the case of a single straight-line dipping interface;
Figure 3.11 depicts the basic geometry involved. Figure 3.12 depicts
the same geometry as Figure 3.11, but it includes additional information.
We let angle u be the dip angle, and we let SG measure the perpendicular
distance to the dipping interface from source point S. We now want to
draw the raypath for a primary reflection event from the dipping interface.
We draw a line between S′ (which is the mirror image of S with respect
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 89

DOI:10.1190/1.9781560803737

Figure 3.12. The normal ray is perpendicular to the interface, whereas the image
ray is perpendicular to the earth’s surface. The angles ER ′ S and SGD are right
angles.

to the interface; i.e., SG = GS′ ) and the receiver R. The line S′ R

intercepts the interface at D. The raypath is SDR, which is equal in length
to the straight line S′ R. The triangle SES′ is isosceles. Also, the triangle
SDS′ is isosceles. The line EK is drawn parallel to line SS′ . It is apparent
that line EK bisects angle SER′ . The image ray is defined as the line from
image point S′ that meets the surface at right angles. In other words, the
image ray is S′ R′ .
As shown in Figure 3.12, the length of the offset raypath SDR is vt(x).
The offset distance is x. The zero-offset path is SS′ with length vt(0). Note
that angle S′ SR = p/2 + u. We now apply the law of cosines to the triangle
S′ SR to obtain
p 
v2 t2 (x) = v2 t2 (0) + x2 − 2vt (0) x cos + u , (3.11)
2
which is
x2 2t(0) x sin u
t2 (x) = t2 (0) + + . (3.12)
v2 v
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

90 Basic Wave Analysis, Part 1: Velocity Analysis

Figure 3.13. Time-

distance curve for a
dipping interface in
the case of a CSP
located at x ¼ 0.

If we complete squares, we obtain

v2 t2 (x) (x + v t(0) sin u)2
− =1. (3.13)
v2 t2 (0) cos2 u v2 t2 (0) cos2 u
DOI:10.1190/1.9781560803737

This time-distance curve t(x) is a hyperbola, as before, but the axis

of symmetry is with respect to the line x = −v t(0) sin u instead of the line
x = 0 (de Bazelaire, 1988). This means that t(x) is not equal to t(2x),
unlike the case for zero dip where t(x) = t(−x). The hyperbola is shown
in Figure 3.13. This hyperbola represents the time-distance curve for reflec-
tion events as would appear on a CSP gather.
From Figure 3.12, an important duality may be grasped. The normal ray
is the ray between the source and the interface, striking the interface at right
angles. The image ray is the ray between the source image and the earth’s
surface, striking the earth’s surface at right angles (Hubral, 1977). The
point where the image ray meets the earth’s surface is
x = −v t(0) sin u . (3.14)
In other words, the image ray meets the earth’s surface at the point where
the traveltime t(x) is a minimum. The image raypath is the raypath with
the minimum traveltime from the source image to the surface; that is, the
reflection SER′ has minimum traveltime among all receivers.
The time-distance hyperbola can be written as

x2 + 2v t(0) x sin u
t(x) = t2 (0) +
v2
 
x2 + 2v t(0) x sin u
≈ t(0) 1 + , (3.15)
2v2 t2 (0)

where we have used only the first term in the square root expansion.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 91

The most common method of finding the dip angle u is to consider two
receivers equally distant from the source, but on opposite sides. Then, we
subtract the traveltimes to obtain

4v t(0) x sin u
t(x) − t(−x) ≈ t(0)
2v2 t2 (0)

or

2x sin u
t(x) − t(−x) ≈ . (3.16)
v
The distance between the two receivers is denoted by
Dx = x − (−x) = 2x . (3.17)
The quantity
DOI:10.1190/1.9781560803737

Dtd = t(x) − t(−x) (3.18)

is called the dip moveout. Thus, we have the equation

Dtd
sin u = v , (3.19)
Dx
which gives an expression which can be used to find the dip angle. For small
dips, u is approximately equal to sin u; in such cases, the dip angle is directly
proportional to the dip moveout. Because the dip moveout is proportional to
Dx, we would want to use as large a value of Dx as the data warrants in order
to obtain an accurate estimation of dip. For example, for a symmetric spread
of geophones about a source point, we would measure dip moveout between
receivers at the opposite ends of the spread.
In principle, the method of 3D prestack depth migration avoids the
assumptions needed in stacking-to-zero-offset and post-stack migration.
However, before the turn of the 21st century, computers were not powerful
enough to conduct such burdensome tasks as 3D prestack migration on a
production basis. As a result, partial methods were used. Migration was
somewhat insensitive to velocity when only small angles were involved.
When wide angles were involved, migration became quite sensitive to ve-
locity, so it was necessary to estimate the velocity to be used for migration.
Thus, migration became interwoven with velocity estimation. Sherwood
et al. (1976) showed that the moveout correction can be considered in two
parts: one part depends upon offset, and the other part depends upon dip.
Sherwood called the process Devilish. Yilmaz (1979) developed the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

92 Basic Wave Analysis, Part 1: Velocity Analysis

Figure 3.14. A
CMP geometry with
dipping interface
showing that each
raypath has a
different depth
point.
DOI:10.1190/1.9781560803737

process more explicitly and called it prestack partial migration. In time, the
process was called dip moveout (DMO).
So, next we want to consider the time-distance curve t(x) on a
CMP gather in the case of a dipping interface. As shown in Figure 3.14,
the CMP is denoted by M, and two source-receiver pairs are displayed.
The zero-offset ray is the normal ray; this raypath is perpendicular from
the common midpoint M to the interface. We designate the reflection
point by D0. The reflection points for the offset paths are as follows.
The reflection point D1 is for the path S1D1R1 and reflection point D2
for the path S2D2R2. None of these reflection points are the same, and, in
fact, there is only one source-receiver pair which would produce a reflec-
tion at point D, which is defined as the depth point directly under midpoint
M. This situation contrasts sharply with the situation in the case of flat
beds, where all of the reflection points are at D (i.e., directly under mid-
point M).
Recall that we have defined NMO correction, which modifies the trav-
eltime along all of the traces with the same CMP as if they travel along the
normal ray. In other words, the NMO correction removes the additional
traveltime caused by the offset between source and receiver.
Let us review. For a flat interface, the CMP gather shows a primary
reflection event that lies on a hyperbola t(x), symmetric with respect to
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 93

a)

b)
DOI:10.1190/1.9781560803737

c)

Figure 3.15. (a) A CMP geometry with flat interface, (b) seismic reflection event
falling on a hyperbola (symmetric about CMP), and (c) seismic reflection event after
NMO correction.

the zero-offset point x = 0 (see Figure 3.15). Equation (3.10) can be

rewritten as
x2
t(x) ≈ t(0) + or t(x) ≈ t(0) + NMO . (3.20)
2v2 t(0)
If we subtract the NMO correction, we obtain
t(x) − NMO ≈ t(0) , (3.21)

which is approximately a straight line that equals t(0) for all values of x.
Thus, we can add all of these corrected traces, and in so doing the event
on each trace adds in phase. This is the process of stacking, and the result
is that the primary reflection event is reinforced at the expense of any
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

94 Basic Wave Analysis, Part 1: Velocity Analysis

multiple reflection events which arrive at approximately the same depth

point but with different moveout.
Now, consider traveltime correction in a dipping interface case. The
CSP gather (with the source point at x = 0) also shows a hyperbola, but
this hyperbola is not symmetric about the source point x = 0. Instead, this
hyperbola is symmetric about a point that is shifted in the up-dip direction.
Naively, one might think that the hyperbola would be tilted along with
the reflector, but it is not. Of course, its main axis (and axis of symmetry)
always remains vertical (see Figure 3.16). As a result, we turn instead to
the CMP gather.
As shown in Figure 3.17, here the primary reflection event lies on
a hyperbola t(x), and (as will be demonstrated) the hyperbola is indeed sym-
metric about the common midpoint x = 0. Thus, we can subtract an NMO
correction Dt from t(x) to obtain t(0) = t(x) − Dt. The result is that the
primary reflection event lines up on the corrected traces, and hence we
DOI:10.1190/1.9781560803737

may stack (i.e., add) them. The resulting stacked trace is one in which the
primary event stands out more sharply against multiple events and noise.

a)

b)

Figure 3.16. (a) A CSP geometry with dipping interface and (b) seismic reflection
event falling on hyperbola (not symmetric about CSP).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 95

a)

Figure 3.17. (a) CMP

geometry with dipping
interface, (b) seismic
reflection event falling
DOI:10.1190/1.9781560803737

on hyperbola
(symmetric about
b) CMP), and (c) seismic
reflection event after
NMO correction.

c)

Now, we can derive the hyperbolic equation and the NMO expression
for the CMP gather in the case of a dipping interface. Consider the geometry
shown in Figure 3.18. The interface has angle u. Let M be the midpoint
between S and R, so SM = MR = x/2. The point S′ is the image of S with
respect to the interface. Also M′ is the image of M with respect to the
interface, and hence MM′ is the normal raypath of length vt(0). Let SS′
cut the interface at S0 and let MM′ cut the interface at D0. Let a
horizontal line from S0 cut MD0 at P. Triangle S0D0P is a right triangle
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

96 Basic Wave Analysis, Part 1: Velocity Analysis

Figure 3.18. The

geometry for the
CMP hyperbola in the
case of a dipping
interface.
DOI:10.1190/1.9781560803737

by construction (with right angle at D0). Thus,

 x
PD0 = S0 P sin u = sin u . (3.22)
2
We then have

vt(0) x sin u
SS0 = MD0 − PD0 = − , (3.23)
2 2
which produces

SS′ = vt(0) − x sin u . (3.24)

Now, we apply the law of cosines to triangle S′ SR. The angle S′ SR is

(p/2) + u. Side S′ R is vt(x). We have
p 
v t (x) = [vt(0) − x sin u] + x − 2x[vt(0) − x sin u ] cos + u ,
2 2 2 2
2
(3.25)
which yields

v2 t2 (x) = v2 t2 (0) + x2 (1 − sin2 u)

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 97

or
v2t2(x) = v2t2(0) + x2 cos2 u . (3.26)
This is the equation of a hyperbola with t(x) symmetric about the point
x = 0. Of course, we know on a priori grounds that t(x) = t(−x) because
of the principle of reciprocity; that is, we know that the traveltime from a
point s to a point r with offset x = r − s must equal the traveltime from
point r to point s with offset
s − r = −(r − s) = −x . (3.27)
Thus, we have found the required time-distance curve for the CMP gather in
the case of a dipping interface. We can write this equation as
x2
t2 (x) −  = t2 (0) .
v 2
(3.28)
DOI:10.1190/1.9781560803737

cos u
If we write

x2 cos2 u
t(x) = t(0) 1 + 2 2 (3.29)
v t (0)

and use the square root expansion

√ 1
1 + A = 1 + A + ··· , (3.30)
2
we obtain the approximation

x2 cos2 u
t(x) ≈ t(0) + . (3.31)
2v2 t(0)

Thus, the NMO correction in the case of an interface dipping at angle

u is

x2 cos2 u
Dt = t(x) − t(0) ≈ . (3.32)
2v2 t(0)

When u = 0, this equation becomes

x2
Dt = t(x) − t(0) ≈ , (3.33)
2v2 t(0)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

98 Basic Wave Analysis, Part 1: Velocity Analysis

which is the same as the NMO equation in the case of a flat interface
(equation 3.20):

x2
Dt ≈ . (3.34)
2v2 t(0)

Coherency
Applying the “new” mathematics of statistics to the study of stars, John
Michell’s 1767 paper demonstrated that many more stars occur in pairs or
groups than could be accounted for in a perfectly random distribution.
Studying the Pleiades cluster, Michell calculated the likelihood of discover-
ing such a close grouping of stars at about one in one-half million. To
explain these double and multiple star systems, he posited that the stars
might be drawn together by gravitational pull. Indeed, by employing prob-
DOI:10.1190/1.9781560803737

ability theory, his work provided the first evidence of the existence of binary
stars and star clusters and, as noted previously, inspired the work of William
and Caroline Herschel as well.
With the advent of the digital computer, the mathematics of statistics
has come into use in seismology. One such application is that of coherency.
Trace-to-trace coherence (or similarity) can be given a quantitative measure
in several ways. For two traces, the cross correlation can be used as a
measure of coherence. For a large number of traces, we could make use
of the fact that similar traces add (i.e., stack) in phase and hence reinforce
each other, whereas dissimilar traces stack out-of-phase and hence cancel
each other. Thus, when we stack several traces together, the resulting ampli-
tude is generally large where the individual channels are similar (coherent)
and small where they are not alike (incoherent). The ratio of the energy of
the stack to the sum of the energies of the individual traces, therefore,
would be a measure of the degree of coherence.
Let Ait be the amplitude of the individual trace i at time t. We suppose
that there are M traces in the record section, so that the trace index i ranges
from 1 to M. Furthermore, we suppose that the arrival time t of a specific
primary reflection event follows some trajectory across the traces compos-
ing the record section. In velocity analysis, this trajectory is the hyperbola
specified by some particular value of velocity v. Thus, the actual value of
the trajectory time t for any given trace depends upon the trace index i.
The amplitude of the stack for trajectory t is given by Ait where the
summation is over i = 1, 2, . . . , M. The square of this stack amplitude is
the instantaneous energy of the stack for that particular trajectory. The
instantaneous energy of each trace along the trajectory is A2it , so the sum
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 99

of the energies of the input traces is A2it where the summation is over
i = 1, 2, . . . , M. Next, the ratio Et associated with trajectory t can be
defined by means of the formula

M 2
(instantaneous stack energy) i=1 Ait
Et = = M 2
. (3.35)
M (instantaneous input energy) M i=1 Ait

The reason for introducing the number M in the numerator is that it acts as a
normalization factor. Thus, the instantaneous energy ratio Et is one when all
of the trace amplitudes are the same along the trajectory.
A coherent event extends over a time interval on each trace. In order to
account for this effect, we usually use in place of Et the semblance St, which
we will now define. Let us pick a time gate on each trace. This gate is cen-
tered on the trajectory t and extends the same width t on each side. Thus, the
total width of the time gate is 2t, which is the same for each trace. In
DOI:10.1190/1.9781560803737

summary, the time gate is an interval of length 2t on each trace and is

symmetrically disposed about the trajectory t. This choice makes the gate
boundaries parallel to the trajectory t. The semblance is a measure of the
coherence of a signal which crosses the gate and is defined as

(stack energy within gate)

St = . (3.36)
M (input energy within gate)

Not only does the semblance tend to be large when a coherent event is
present, but the magnitude of the semblance also is sensitive to the ampli-
tude of the event. Strong events exhibit large semblances and weak events
exhibit moderate values of semblance, whereas incoherent data have very
low semblances.
It should be noted, however, that a seldom mentioned problem with
semblance is that it has a fundamental singularity near 0. For example, let
us define S as

2
ai
S= . (3.37)
N a2i

For a = (0, 0, 0, 1), we have S = 1/4. However, for a = (1, 1, 1, 0), we

have S = 3/4. For larger N, the first goes to 0, the second to 1; yet, we
would like to consider these two sequences as having the same
semblance/coherence. Despite this issue, semblance is used with great
benefit in seismic processing.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

100 Basic Wave Analysis, Part 1: Velocity Analysis

Semblance and other coherence measures are used to determine the

values of parameters that will optimize a stack. Semblance is calculated
for various time shifts between the component channels; the optimum
time shift is one that maximizes the semblance. Semblance, therefore, can
be used to determine static corrections. The most important application of
semblance is in the computation of a velocity spectrum from which we
determine stacking velocity as a function of stacking time.

Stacking (NMO) velocity

Because of the redundancy inherent in the multiple coverage method
of seismic prospecting, unwanted interference and noise can be attenuated
by stacking. Stacking also is a means of estimating the velocity function.
The dynamic corrections put all primary reflection events in phase on the
traces in each CMP gather. Because the raypaths of multiple reflection
DOI:10.1190/1.9781560803737

events are different, the dynamic corrections do not put the multiples
in phase. Hence, if we add together (i.e., stack) all of the corrected traces
in a CMP gather, we severely attenuate the multiples as well as other
incoherent noise. Thus, we obtain one output trace for each midpoint; this
composite trace is called the stacked trace for that midpoint. Sometimes,
this type of stack is called the common depth point (CDP) stack, although
the more accurate term common midpoint (CMP) stack. The reason for
using the term midpoint instead of depth point is that, for slanting sub-
surface horizons, the depth points (i.e., points of reflection) do not lie
directly under the source-receiver midpoints. In other words, they are
dispersed. The CMP stack is based on the fact that the reflection times of
multiple reflection events generally increase faster with increasing shot-
receiver distances than do those of primary reflection with the same vertical
traveltime.
Stacking improves specific data quality. However, the indiscriminate
use of stacking has fallen out of favor because stacking also destroys infor-
mation necessary for good migration results. Even so, stacking has very
important uses in velocity estimation, such as can be demonstrated in the fol-
lowing discussion of CMP stacking. Velocity analysis techniques utilize the
variation of NMO with record time to find velocity. The primary objective of
velocity analysis is to find the amount of NMO that should be removed to
optimize the stacking of primary events. Multiples as well as primaries
will give rise to peaks, and the results must be interpreted to determine
the best values of velocity with which to stack the data. In many areas,
where the velocity increases more-or-less monotonically with depth, the
peaks associated with the highest reasonable stacking velocities represent
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 101

primary reflection events, and the peaks associated with lower velocities
represent multiple reflection events of various sorts.
Stacking can be described in various ways; here, we will describe it
in terms of NMO. Each trace on a CMP gather is corrected for NMO.
That is, an event on a trace that appears at time t is time-corrected by
subtracting the NMO from t so that the corrected event will appear at
time t(0). As we have discussed, the NMO for a given trace decreases as
t(0) increases. Thus, the NMO correction changes with record time, so the
correction is said to be dynamic (as opposed to static). After correcting
the set of traces for NMO, the events due to a horizontal reflector
will line up on all of the traces of the CMP gather at the same time t(0).
Thus, when the corrected traces are added together, the primary reflection
events are reinforced. This is why the process is called stacking. The
resulting output trace is called the stacked trace. Each CMP gather yields
one stacked trace. The collection of all stacked traces constitutes the
DOI:10.1190/1.9781560803737

stacked section.
As we have shown, the NMO correction puts all primary reflection
events in phase on the traces in each CMP gather. Because the raypaths of
multiple reflection events are different, the dynamic corrections do not put
the multiple reflection events in phase. Hence, if we add together (i.e.,
stack) all corrected traces in a CMP gather, we attenuate the multiples as
well as incoherent noise.
In summary, NMO represents the variation of reflection arrival time
due to the variation in the source-to-receiver distance (offset). Normal
moveout depends upon velocity and to a lesser extent upon the dip of the
interfaces, as well as on offset. Normal moveout decreases with reflection
time t0; that is, for a given offset, the NMO for a deep reflection is less
than the NMO for a shallow reflection. Many velocity analysis techniques
are based on the NMO measurements.
Velocity is used in the calculation of the positions of subsurface
reflectors from the observed traveltimes. It can be said that velocity is the
most important variable used in seismic prospecting. A seismic wave pro-
pagates with a velocity that depends on many factors, such as chemical
composition and local geology. The velocities of the rocks generally
increase with depth and vary from values in a range of about 335 m/s in
the air to values approaching 6400 m/s in deep sedimentary basins.
The objective of stacking is to increase signal-to-noise ratios to a level
sufficient to ensure reliable identification of primary reflected events on
the seismograms. However, velocity as a function of seismic depth must
be known quite accurately in order to apply proper NMO corrections
prior to the processing method known as stacking. Only in this way can
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

102 Basic Wave Analysis, Part 1: Velocity Analysis

Figure 3.19. Single-

horizontal-interface
model.
DOI:10.1190/1.9781560803737

we obtain the correct hyperbola-shaped curves used in the stacking

process. It is this hyperbolic characteristic of reflection time-distance
curves that provides a means of establishing the necessary velocity-time
relationship.
The offset is the horizontal distance x from the source to the receiver.
Velocity can be determined from seismic measurements of the variation
of arrival time t as x changes. Often, the measurement of velocity is based
on the single-horizontal-interface model illustrated in Figure 3.19. It has a
single horizontal reflecting interface at depth z. The layer between the inter-
face and the surface has constant velocity v. For a receiver located at the
source point, the reflected energy will travel vertically down to the reflecting
interface and vertically back to the geophone. Its arrival time t0 is the two-
way vertical distance divided by the velocity; that is, t0 = 2z/v. A receiver at
horizontal distance x away from the source will detect reflected energy at a
greater arrival time t, because the energy has traveled farther. The distance
traveled is vt. Think of the reflecting interface as a mirror. An observer at R
sees the light source S at its mirror image S′ . In other words, the process of
reflection in effect replaces the source point S with an image point S′ . Plato
said, “There is no harm in repeating a good thing,” and in this spirit we
repeat: The Pythagorean theorem for the hypotenuse of a right triangle
gives the equation

v2 t2 = v2 t02 + x2 . (3.38)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 103

Now, let us describe the velocity spectrum, starting with equation

(3.38). This is the equation for the hyperbola that relates t and x, and it
involves the two parameters t0 and v. The essential idea behind velocity
analysis is that the data in the CMP gather can be used to estimate velocity.
The data-processing method used is called a velocity search procedure. In
the velocity search procedure, the vertical time t0 is fixed at successive
values ts (called the stacking times), and the velocity v that is optimum for
this stacking time is called the slacking velocity vs. To examine this in
greater detail, consider Figure 3.20a. At a fixed zero-offset time t0 = ts ,
the common midpoint set of traces, with full offsets x1 , x2 , . . . , xN , is
overlaid with a set of hyperbolas. These hyperbolas are indexed by a trial
velocity v. In other words, there is a different velocity v for each hyperbola

a)
DOI:10.1190/1.9781560803737

b)

Figure 3.20. (a) Hyperbolic sweep of the CMP gather to determine which value
should be chosen for velocity v; this process is repeated at constant time intervals
down the record. (b) Semblance of the traces within a small time gate centered on a
particular hyperbola; the peak value occurs at velocity 2, which indicates that this
velocity would be selected as vs for the given value of t0.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

104 Basic Wave Analysis, Part 1: Velocity Analysis

in the set. For each hyperbola, the semblance of the traces (within a small
time gate centered on the hyperbola) is computed. Figure 3.20b shows the
semblance plotted for a set of values v. The velocity v that produces the
maximum semblance represents the best-fitting hyperbola and thus produces
the best stack. As a result, this optimum velocity is called the stacking
velocity vs at time ts. This procedure is repeated for a collection of zero-
offset (vertical) times ts, generally at equally spaced time increments. The
result is called the velocity spectrum at the common midpoint. The velocity
spectrum consists of a 3D display of semblance as a contour in v, ts space.
A curve showing the stacked semblance maxima as a function of zero-
offset (stacking) time ts identifies the coherent events and the stacking
velocities vs (see Figure 3.21).
Stacking combines data (after correction for NMO) which have a CMP
between the source and the receiver. The correction for NMO is referred to
as a dynamic correction. In the usual case, velocity increases with depth. As
DOI:10.1190/1.9781560803737

a result, long-path multiples travel at a lower average velocity than do the

primary reflection events with the same arrival time. Thus, the long-path
multiples show a greater NMO than that of these primaries. Because stack-
ing is performed according to the velocity of the primaries, the multiples will
not stack in phase and thus will be attenuated. Stacking, like all summation
methods, also attenuates random noise.
For a stacked trace, the CMP represents both source and receiver. Thus,
a stacked record section represents a coincident source/receiver geometry,
or, in other words, a zero-offset geometry. In a true zero-offset experiment,
the raypaths from the source to the reflector and back to the same source
point (i.e., receiver point) may be complex, but we do know that the down-
ward and upward legs must be identical, and that the raypath strikes the
interface at right angles. Such paths are called normal incidence paths, or
normal paths. As shown in Figure 3.22, the point at which the normal
path strikes the target interface at right angles is called the normal incidence
point (NIP). It is assumed that the stacked record section represents data pro-
duced by coincident source/receiver experiments. In other words, it is
assumed that each stacked trace has the receiver location coincident with
the source location.
In summary, a velocity spectrum is obtained by scanning the CMP
gather of traces along hyperbolic trajectories for signal coherence. These
scans establish a velocity function (i.e., velocity versus vertical seismic
time) that is used in calculating the NMO corrections prior to the stacking
of the CMP traces. One further point that deserves mention is that the
time-distance curves for hyperbolic interfaces are not hyperbolic, except
in the case of the first interface. Nevertheless, the curves are almost
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms
DOI:10.1190/1.9781560803737

Chapter 3: Velocity Estimation

Figure 3.21. (a) A CMP gather, and two ways of displaying the velocity spectrum computed from this gather: (b) gated raw plot and
105

(c) contour plot. (From Yilmaz, 2008, Chapter 3, page 303.)

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

106 Basic Wave Analysis, Part 1: Velocity Analysis

Figure 3.22.
Normal incidence
paths for the same
source-receiver
(zero-offset) point
M. Each path strikes
its target interface at
right angles at the
normal incident
point (NIP).
Generally, in the
case of dipping
interfaces, there is a
different normal
path for each
DOI:10.1190/1.9781560803737

interface, as shown
here.

hyperbola-shaped to the extent that the second-order hyperbolic approxi-

mation can be used in the velocity search even in areas of relatively
complex structures.

Dix formula for interval velocity

In conventional seismic processing, normal moveout is computed by
assuming straight-line propagation in a constant velocity model. The ques-
tion addressed in this section is which mathematical expression for velocity
we should use for the constant velocity appearing in this simple model. This
is not a simple question to answer, but generally the answer given by Dix
(1955) is used in petroleum exploration.
One of the most important problems in reflection seismology is how to
address vertical variations of velocity. The simplest methods utilize various
modifications of the constant velocity model, a model, in which the actual
earth layers existing between the earth’s surface and an interface at some
given depth, are replaced by a single layer of some constant velocity V
(see Figure 3.23). By varying the depth of the interface, this assigned veloc-
ity, which is constant between the surface and interface, becomes a function
V(z) of the depth z of the interface. If T0 is the vertical traveltime to the
interface, then, alternatively, V may be considered to be a function of T0.
Note that we are assuming the layers are horizontal (i.e., that they are not
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 107

Figure 3.23. Model of earth

layers with interval velocities
(left), and model of earth
sections with constant average
velocities (right).
DOI:10.1190/1.9781560803737

Figure 3.24. Sides of a right

triangle.

dipping). Thus, the laterally constant velocity model is one in which the
earth is assigned a different constant velocity V(z) for each interface
(where z is the depth of the interface).
Certainly, V(zN) must be some sort of average of the interval velocities
v1 , v2 , . . . , vN down to interface zN. Thus, in Figure 3.23, V(z1) would be
some function of v1, V(z2) some function of v1, v2, and V(z3) some function
of v1, v2, v3. The functional form we choose for V(z) would depend on the
purpose for which we want to use it. One purpose of a velocity function
V(z) is to compute normal moveout. Thus, we would want to pick a
convenient functional form that will serve this purpose.
The time-distance curve for the single-horizontal-layer model is
obtained by applying the Pythagorean theorem to the sides of a right tri-
angle. As shown in Figure 3.24, the hypotenuse is the slanted straight-line
ray of length VT, one leg is the full offset x, and the other leg is the vertical
distance VT0. Here, T is the two-way oblique time and T0 is the two-way
vertical time. Thus, the Pythagorean equation is

V 2 T 2 = V 2 T02 + x2 . (3.39)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

108 Basic Wave Analysis, Part 1: Velocity Analysis

If we consider this equation as the equation of a curve of T versus x, we

see that the curve is a hyperbola. The T-intercept is equal to the vertical time
T0. If we consider this equation as the equation of a curve of T 2 versus x 2,
we see that the curve is a straight line of slope 1/V 2. The NMO equation
for the single-layer model is

x2
DT = T02 + 2 − T0 , (3.40)
V
where T0 is vertical time, V is velocity, x is offset distance, and T is oblique
time. To gain insight, let us experiment with this equation by means
of examples. The oblique time T approximates the oblique time of the
horizontally layered model. In the following two examples, we let T0 be
equal to the vertical time t(0) of the horizontally layered model and exper-
iment with different definitions of velocity V. In these examples, we let hori-
zontal distance x represent the full offset, and all times are two-way times.
DOI:10.1190/1.9781560803737

In Examples 1 and 2, the unit of distance is kilometer and the unit of time
is second.
Example 1. Figure 3.25 has one layer with an interval velocity v1 = 2.4,
a layer thickness Dz1 = 1.5, and an offset distance x ¼ 1.8. The two-way
vertical time is
2(1.5)
T0 = t(0) = = 1.25 . (3.41)
2.4
If we set V = v1 = 8, then
 
x2 (1.8)2
T = T02 + 2 = (1.25)2 + = 1.458 . (3.42)
V (2.4)2

Figure 3.25.
Illustration of
Example 1. The
interval velocity
is 2.4.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 109

Normal moveout DT is
DT = T − T0 = 1.458 − 1.250 = 0.208 . (3.43)

Example 2. Figure 3.26 has two layers with interval velocity v1 = 1.8
for the first layer and v2 = 3 for the second layer. The thicknesses of the
layers are Dz1 ¼ 0.5625 and Dz2 ¼ 0.9375, which make a total thickness
of 1.5. As a result, the vertical time T0 is the same as for Example 1; that is,
2(0.5625) 2(0.9375)
T0 = t(0) = + = 1.25 . (3.44)
1.8 3
The average velocity is
2(Dz1 + Dz2 ) 2(0.5625 + 0.9375)
vA = = = 2.4 . (3.45)
T0 1.8
DOI:10.1190/1.9781560803737

We will use vA as the velocity in a single-layer approximation. The

Pythagorean equation gives the oblique time as
 
x2 (1.8)2
T = T02 + 2 = (1.25)2 + = 1.458 . (3.46)
vA (2.4)2

Figure 3.26. Illustration of Example 2. The average velocity in the vertical

direction is 2.4, the same as the interval velocity in Example 1. The raypath on the
downgoing leg is not the straight path SBD, but is the bent raypath SAD. The reason
is that, for minimum traveltime (Fermat’s principle), the raypath must consist of the
shorter distance SA (instead of SB) in the low-velocity medium and the longer
distance AD (instead of BD) in the high-velocity medium.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

110 Basic Wave Analysis, Part 1: Velocity Analysis

Normal moveout is

DT = T − T0 = 1.458 − 1.250 = 0.208 . (3.47)

Instead of using the single-layer approximation, let us compute DT for this

second case by using the time-distance curve for the given two-layer model.
The resulting angles are u1 = 20◦ 46′ , u2 = 36◦ 13′ . The two-way oblique
time is

2(0.5625) 2(0.9375)
t(x) = + = 1.443 . (3.48)
1.8 cos 20 46 3 cos 36◦ 13′
◦ ′

The resulting Dt is

Dt = 1.443 − 1.250 = 0.193 . (3.49)

DOI:10.1190/1.9781560803737

This Dt is true NMO, as we have calculated it from the actual time-distance

curve and not from approximation.
Examples 1 and 2 have the same offset distance, average velocity, and
vertical traveltime. As a result, DT as computed from the single-layer
model is the same for both examples; i.e., DT = 0.208. In other words,
the single-layer model does not consider any effect caused by the layering
in Example 2. If we do consider the layering in Example 2, the oblique
time t(x) is less than T. Thus, Dt = 0.193 is less than DT = 0.208. The
value 0.193 is the correct value, whereas the value 0.208 given by the
single-layer approximation is wrong. To make the single-layer approxi-
mation give the correct Dt, or nearly the correct Dt, we cannot use the
average velocity vA in the equation but instead must use the root-
mean-square (RMS) velocity vRMS. In the case of Example 1, the RMS
velocity is
 

 1 1 v21 t(0)
vRMS = v2i Dti (0) = = v1 = 2.4 . (3.50)
t(0) i=1 t(0)

As we would expect, in a single-layered medium, the RMS velocity is the

same as the interval velocity as well as the average velocity. In the case of
Example 2, the RMS velocity is

(1.8)2 (0.625) + (3)2 (0.625)
vRMS = = 2.4738 , (3.51)
1.250
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 111

which is greater than the average velocity vA = 2.4. Let us now use
the RMS velocity in the Pythagorean equation. The equation gives the
oblique time as
 
x2 (1.8)2
T = T02 + 2 = (1.25)2 + = 1.446 . (3.52)
vRMS (2.4738)2

This oblique time 1.446 is close to the true value of 1.444 computed pre-
viously. Therefore, the resulting DT,
DT = T − T0 = 1.446 − 1.250 = 0.196 , (3.53)
is close to the true value Dt = 0.193, given previously.
In summary, if we use a multi-layer raypath calculation, we obtain the
correct Dt of 0.193. If we use the average velocity in the single-layer
approximation, we obtain the computed DT of 0.208. If we use the RMS
DOI:10.1190/1.9781560803737

velocity, we obtain the computed DT of 0.196 s. Thus, the RMS velocity pro-
duces a better result than the average velocity when using the single-
layer Pythagorean equation in any case in which there is layering. In fact,
the RMS velocity will differ from the average velocity more and more as
the layering becomes more complex. Figure 3.27 shows the true time-dis-
tance curve and the approximating hyperbola with the same t-intercept
and with the RMS velocity.
As demonstrated in the preceding section, the velocity spectrum as
computed from the CMP data yields an empirical velocity function that
gives velocity vs versus time ts. The quantities vs and ts are called stacking
velocity and stacking time, respectively. The stacking velocity vs provides
the greatest semblance for the given stacking time ts as compared to any
other velocity v. Because vs is empirically determined, we do not have a

Figure 3.27. The RMS hyperbola (dashed) has the same vertical time as the true
time-distance curve (solid), but otherwise always has a greater traveltime.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

112 Basic Wave Analysis, Part 1: Velocity Analysis

mathematical equation for it. However, we want to set vs equal to a theoreti-

cal velocity V that does have a mathematical expression. The reason we need
a mathematical formula is so that we can compute the interval velocities
v1 , v2 , . . . , vN from the observed stacking velocities at their respective stack-
ing times.
In work conducted before 1955, the empirically determined velocity vs
is set equal to the weighted average velocity vA, defined as

total distance 2(Dz1 + Dz2 + · · · + DzN )

vA = = . (3.54)
total time t(0)

This is a poor choice, as demonstrated by the foregoing examples. In 1955,

C. H. Dix sets the empirical velocity vs equal to RMS velocity (see
Figure 3.27). As our examples show, this produces good results. After the
stacking velocities have been determined (e.g., from the velocity spectrum)
DOI:10.1190/1.9781560803737

at two successive stacking times ts, the interval velocity can be found from
the Dix formula [which will be shown below in equation (3.58)].
The Dix formula shows how the RMS velocities for two successive
interfaces can be used to find the intervening interval velocity. The
formula for RMS velocity is

1 n 2 N
vRMS = v Dti (0) where t(0) = Dti (0) . (3.55)
t(0) i=1 i i=1

Writing vRMS(N ) for the RMS velocity for the Nth interface and
vRMS(N 2 1) for the RMS velocity for the interface above it, we have

N N−1 N
v2i Dti (0) = v2i Dti (0) + v2N DtN (0) = v2RMS (N) Dti (0)
i=1 i=1 i=1
(3.56)
N−1 N−1
v2i Dti (0) = v2RMS (N − 1) Dti(0) .
i=1 i=1

Note that Dti (0) denotes the two-way vertical time in layer i. Subtracting
and then dividing both sides by Dti (0), we obtain the Dix formula for the
interval velocity vi. The Dix formula is
 
N N−1
1
v2N = v2 (N) Dti (0) − v2RMS (N − 1) Dti (0) . (3.57)
DtN (0) RMS i=1 i=1
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 113

Table 3.1. Velocity analysis.

CALCULATED:
GIVEN: observed GIVEN: observed interval velocities
stacking velocities stacking times computed by Dix
N vsN (km/s) tsN (s) formula vN

1 2.5 0.6
2 2.6 0.8 2.9
3 3.1 0.9 5.7
4 3.3 1.0 4.7
5 3.4 1.2 3.9
6 3.8 1.8 4.5
7 4.0 2.0 5.5
DOI:10.1190/1.9781560803737

Now, the basic assumption of velocity analysis is invoked. Given two

successive stacking times ts,N−1 and ts,N , and the corresponding empirically
determined stacking velocities vs,N−1 and vs,N , the basic assumption is to set
the stacking velocities equal to the RMS velocities and the stacking times
equal to the vertical traveltimes appearing in the Dix formula. Thus, the
Dix formula becomes

v2s,N ts,N − v2s,N−1 ts,N−1

v2N = . (3.58)
ts,N − ts,N−1

“Things won are done; joy’s soul lies in the doing” (Shakespeare). No one is
known to have published a continuous version of the Dix equation.
From a given velocity spectrum, we can measure observed stacking ve-
locities vsN (as shown in Table 3.1, column 2) and observed stacking times
tsN (as shown in Table 3.1, column 3). In other words, columns 2 and 3 depict
the given values. Then, the interval velocities may be calculated. They are
computed by means of the Dix formula (3.58). Thus, we obtain the sequence
of interval velocities (as shown in Table 3.1, column 4). This is the classic
way of determining interval velocities from seismic data.

Approximation of stacking velocity

Now, let us approach the velocity problem from the opposite direction.
In Examples 1 and 2 of the preceding section, the unit of distance is
kilometer and the unit of time is second. From Example 2, let us take the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

114 Basic Wave Analysis, Part 1: Velocity Analysis

actual Dt of 0.193 and then calculate what velocity in the equation will
produce this Dt. The result of this calculation is the velocity vx, given by
 
x2 (1.8)2
vx = 2 = = 2.496 . (3.59)
t (x) − t2 (0) (1.443)2 − (1.250)2

A velocity calculated by this method is called the offset velocity vx. By defi-
nition, offset velocity (for a given x) provides the exact Dt when used in the
single-layer model for that particular offset x. Thus, offset velocity yields the
correct result when making a Dt correction on a seismic record for that
specific offset. The average velocity yields poor results unless the interval
velocities are nearly the same. The RMS velocity yields nearly the correct
result. Offset velocity is a velocity chosen to fit the data at one fixed value
of x, so that, as we move away from that particular x value, the results
DOI:10.1190/1.9781560803737

become poorer. If we use a shorter offset in Example 2, for example, such

as 4 instead of 6, the offset velocity is different. For x ¼ 4, the true
oblique time is

2(0.5625) 2(0.9375)
t(x) = + = 1.340 , (3.60)
6 cos 15 01 10 cos 25◦ 36′
◦ ′

so that the offset velocity is

 
x2 (1.2)2
vx = 2 = = 8.280 . (3.61)
t (x) − t2 (0) (1.340)2 − (1.250)2

Thus, offset velocity is a function of offset distance as well as layering. We

can summarize the velocities in Example 2 as:

Average velocity ¼ vA ¼ 2.4

RMS velocity ¼ vRMS ¼ 2.4738
Offset velocity ¼ vx ¼ 2.484 for x ¼ 4
Offset velocity ¼ vx ¼ 2.49 for x ¼ 6.

Average velocity should be used for depth conversion, because it is the

true vertical velocity in the ground. Average velocity should never be used
for NMO calculations, because it produces a value for DT that is too large. If
the interval velocities are available from a well survey, then the RMS veloc-
ity can be calculated and can be used for a DT correction. The RMS velocity
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 115

Figure 3.28. The slightly curved (solid) line is the x 2, t 2 transformation of the true
time-distance curve. The two straight lines are the approximations for offset
velocity (dotted) and RMS velocity (dashed).
DOI:10.1190/1.9781560803737

Figure 3.29. The time-distance curve (solid) and the approximating hyperbolas for
offset velocity (dotted) and RMS velocity (dashed).

will always give a better result for DT than the average velocity calculated
from the same survey.
Considering Figures 3.28 and 3.29, let (x1, t1) be a point on the time-
distance curve. Snell’s parameter p is equal to the slope dt/dx of the time-
distance curve. The minimum point (0, t0) of the time-distance curve is
the point where the slope is zero. Thus, the minimum point is the point
that corresponds to p = 0. The approximating hyperbola is


x2
T = t02 + 2 where t0 = t(0) . (3.62)
V
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

116 Basic Wave Analysis, Part 1: Velocity Analysis

It also goes through the point (0, t0) and has a zero slope at that point.
The offset velocity vx is the value of V that makes this hyperbola also
go through the point (x1, t1) on the time-distance curve. Thus, vx can be
found by solving the equation

x2
t1 = t02 + 12 . (3.63)
vx

The resulting value of vx is called the offset velocity for offset x1. Offset
velocity increases with increased spread length or with increased dip of
the reflecting horizon. The average velocity is less than the RMS velocity,
and the RMS velocity is less the offset velocity; that is,
vA ≤ vRMS ≤ vx . (3.64)
Offset velocity vx might be characterized as a simple stacking velocity
DOI:10.1190/1.9781560803737

as compared to the sophisticated statistical determination of empirical

stacking velocities by means of the velocity spectrum. The empirical stack-
ing velocity as derived from the seismic data always should be used for DT
corrections, but the stacking velocity never should be used for depth
conversion.
There is another approach for determining a mathematical expression
for the empirical stacking velocity. This approach does not require the
approximating hyperbola to go through the minimum point (0, t0) of the
time-distance curve. The hyperbola still will have a minimum and a zero
slope at x ¼ 0, but its intercept T0 will not necessarily be equal to t0. The
approximating hyperbola is

x2
T = T02 + 2 . (3.65)
V

The two arbitrary constants T0 and V are determined so as to yield a good fit
of the hyperbola to the time-distance curve.
Next, we will describe the use of the oblique RMS velocity as an
approximation to the stacking velocity. At some point (x1, t1) on the time-
distance curve, as evidenced by the seismic data, we measure the slope
(dt/dx)1. This slope is equal to Snell’s parameter, so we can write
(dt/dx)1 = p1 . From our study of the horizontally layered model, we
know that the oblique RMS velocity for parameter p1 is

x1
vORMS = . (3.66)
p1 v1
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 3: Velocity Estimation 117

Now, let V = vORMS . We solve for T0 by requiring that the hyperbola go

through the point (x1, t1). Thus,
x21
T02 = t12 − . (3.67)
v2ORMS

This method, in which we choose the oblique RMS velocity to approximate

the empirical stacking velocity, provides a good approximation of the hyper-
bola to the time-distance curve near (x1, t1).
The mathematical expression used to approximate the empirical
stacking velocity provides a hyperbola that, in some sense, best fits the
time-distance curve. The time-distance curve is the curve of a reflected
event on the seismic data, so that the final result is a velocity determined
directly from the data. The best-fit criterion varies according to the
method used in the mathematical approximation to the stacking velocity.
Al-Chalabi (1973) gives the following least-squares criterion. Let the
DOI:10.1190/1.9781560803737

approximating hyperbola be

T 2 = a + bx2 , (3.68)

where a = T02 and b ¼ 1/V 2 are the two arbitrary constants to be deter-
mined. From the seismic data, we obtain M points (xi, ti) on the time-
distance curve. According to the principle of least-squares, we want to
minimize

M
f (a, b) = (a + bx2i − ti2 )2 . (3.69)
i=1

If we set the partial derivatives with respect to a and b to equal zero, we

obtain a set of two simultaneous equations for a and b, which can √be
solved easily. Then, √  best-fitting hyperbola has vertical time T0 = a
the
and velocity V = 1/ b. As an exercise, further details would be useful.
This is a very general approach to hyperbolic NMO fitting, but how does
this estimate of velocity relate to the conventional estimates?
In summary, velocity analyses are made by determining how much
stacked energy (or the quantity called semblance) results from a sequence
of trial stacking velocities. This determination is made for a number of
stacking velocities over a series of narrow windows of data. The variation
of arrival time as we change the offset distance from the source to the
receiver carries the information that allows us to determine the stacking
velocity. This variation is embodied in the form of the time-distance
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

118 Basic Wave Analysis, Part 1: Velocity Analysis

hyperbola for the single-flat-interface model. An estimated stacking velocity

vs is a velocity determined from the seismic data by requiring a good fit of
the hyperbola to the seismic event (the manifestation of the time-distance
curve). The stacking velocities so computed are empirical quantities and
thus do not have a mathematical expression in terms of the interval veloci-
ties. In most cases, the stacking velocity is set equal to the RMS velocity,
which does have a mathematical expression. Thus, if we know the RMS ve-
locity vRMS, the interval-velocity can be calculated by the Dix formula:

v2RMS (i) ti (0) − v2RMS (i − 1) ti−1 (0)

v2i = , (3.70)
ti (0) − ti−1 (0)
where vRMS (i) and vRMS (i − 1) indicate RMS velocities for interfaces i and
i − 1, respectively, and ti (0) and ti−1 (0) are the corresponding vertical tra-
veltimes. The notation ti (0) should be used for the vertical traveltime,
DOI:10.1190/1.9781560803737

because the oblique time, using the true raypath determined by Snell’s
law, is denoted by ti (x) when x is offset. Hence, ti (0) would be the zero-
offset or vertical time. Equation (3.70) resembles a finite difference approxi-
mation of a derivative. As an exercise, it would be interesting to determine
the continuous variant.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms
DOI:10.1190/1.9781560803737

Part 2: Raypath Analysis

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms
DOI:10.1190/1.9781560803737
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 4

Traveling Waves

“The simplicity of nature is not to be measured by that of our con-

ceptions. Infinitely varied in its effects, nature is simple only in its
causes, and its economy consists in producing a great number of
DOI:10.1190/1.9781560803737

phenomena, often very complicated, by means of a small number

of general laws.”

—Pierre-Simon Laplace

Coordinates
In 1672, Gottfried Wilhelm (von) Leibniz went to Paris where he was
taught mathematics by Christiaan Huygens. In a letter to James Bernoulli
in 1703, Leibniz recalled, “I quickly saw how great was Huygens.”
Huygens advised Leibniz to study the mathematics of Blaise Pascal
(1623– 1662). In 1659, Pascal had
published Traité des sinus du
quart de cercle, in which Pascal
made use of the characteristic tri-
angle (also known as the infinitesi-
mal triangle or the differential
triangle).
In a stroke of genius, Leibniz
labeled the sides of the triangle
with the notation dy and dx, so
that the slope of the tangent line is
given by the derivative dy/dx (see Figure 4.1. Differential triangle with
Figure 4.1). The result was the infinitesimals dy and dx.

121
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

122 Basic Wave Analysis, Part 2: Raypath Analysis

universally used calculus of Leibniz. Leibniz’s form of calculus allowed

someone who does not understand infinitesimals to teach calculus to
people who will never understand infinitesimals. In the words of Leibniz,
“Why is there anything at all rather than nothing whatsoever?”
The two axes divide the plane into four sections called quadrants. The
quadrants are labelled with quadrant 1 starting at the positive horizontal axis
and going around counter-clockwise. Angles are measured positively in the
counter-clockwise direction. This method of numbering the quadrants and
measuring angles is universally accepted. For points in the first quadrant,
both abscissa and ordinate are positive. For this reason, the first quadrant
generally is used in the teaching of mathematics. Another cardinal rule of
teaching is to teach plane geometry before solid geometry. In other
words, one learns about straight lines and circles before planes and
spheres. In particular, straight-line waves and circle waves come before
plane waves and spherical waves.
DOI:10.1190/1.9781560803737

The world is three dimensional. The horizontal coordinates are x and y.

The depth coordinate is z. In the spirit of first things first, we usually
limit ourselves to plane geometry by using only the x and z coordinates.
Now comes an important point. We will make the z-axis the horizontal
axis and the x-axis the vertical axis, and do the mathematics in the first
quadrant.
In Figure 4.2, the Cartesian coordinate system in two dimensions is
given by an ordered pair of perpendicular axes with an orientation
for each axis. Here, the axes are labeled as the z- and x-axes. The horizontal
z-axis (the abscissa) represents distance into the ground. The vertical x-axis
(the ordinate) represents distance along the ground. The point where the
axes meet is called the origin. Angles are positive in the counter-clockwise
direction. Projection onto the z-axis involves the cosine of the angle, and
projection onto the x-axis involves the sine of the angle.

Figure 4.2. Cartesian

first quadrant with
horizontal axis
representing depth
z into the ground,
and vertical axis
representing distance
x along the ground.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 4: Traveling Waves 123

Figure 4.3.
Cartesian fourth
quadrant as used in
geophysics.

In geophysics, we measure things positively going down into the

ground. In order to obtain this situation, rotate Figure 4.2 by 908 clockwise
in order to obtain Figure 4.3. The positive z-axis (the ordinate) is now ver-
DOI:10.1190/1.9781560803737

tical down into the ground. Angles are positive in the counterclockwise
direction. In this way, projection onto the x-axis (the abscissa) involves
the sine of the angle, and projection onto the z-axis (the ordinate) involves
the cosine of the angle. Thus, in Figure 4.3, the sine occurs where you would
otherwise expect the cosine, and vice versa.
A vector is a directed line segment. A vector has direction as well as
magnitude. Vectors are denoted by boldface letters. There is a special
kind of vector called a unit vector. A unit vector has magnitude 1, with no
units. The three unit vectors for Cartesian coordinates are:

1. unit vector ex pointing in the +x direction,

2. unit vector ey pointing in the +y direction, and
3. unit vector ez pointing in the +z direction.

Any vector can be expressed in terms of unit vectors. For example, the
vector (x, y, z) can be expressed as x ex + y ey + z ez . For example, the graph
of the vector valued function,
r(t) = (sin t)ex + (cos t)ez , (4.1)
represents a circle of radius one centered at the origin. A raypath (in
two dimensions) can be represented by r(t) = (x(t), z(t)) where t is travel-
time. The foot of this vector is at the origin (0, 0) and the arrowhead is at
the point (x(t), z(t)) on the raypath. A 2D straight ray starting from the
source point (x0 , z0 ) and going at constant speed v in the direction of the con-
stant angle u (as measured from the horizontal) has the vector representation
r(t) = (x(t), z(t)) = (x0 + vt sin u, z0 + vt cos u) . (4.2)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

124 Basic Wave Analysis, Part 2: Raypath Analysis

Figure 4.4. Radius

vector and tangent
vector.

The velocity vector v(t) is the derivative of r(t) is, that is,
v(t) = r ′(t) = (x ′(t) , z ′(t) ) = v sin u ex + v cos u ez . (4.3)
The derivative has an important geometric interpretation—the vector r′ (t) is
DOI:10.1190/1.9781560803737

tangent to the curve when this vector is positioned so its initial point is at the
terminal point of the radius vector. The length of the vector r′ (t) is equal to the
speed v.
Now, we can introduce the arc length s. The arc length s is defined as the
length along the curve measured in distance. As shown in Figure 4.4, as the
value of the arc length increases, the corresponding point P moves along
the curve. The coordinates of P become functions of s; that is, the curve
can be written as (x(s), z(s)). If the coordinates of P are expressed in some
other parameter t, then the curve can be written as (x(t), z(t)). The
formula for computing arc length s is

t  2  2
dx dz
s= + dt . (4.4)
t0 dt dt

Here, t0 is the initial value of traveltime t. Thus, we have


 2  2
ds dx dz
= + . (4.5)
dt dt dt
In vector notation, we have
  t t  
ds  dr  ds dr
 dt .
= and s= dt = (4.6)
dt  dt  t0 dt  
t0 dt

Note that we distinguish between vector norms denoted by  , and the

absolute value of a scalar denoted by | |.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 4: Traveling Waves 125

Figure 4.5. Radius vector

and unit tangent.

The unit tangent vector u = u(t) is defined by

 
dr
r′ (t) dt dr
u(t) = ′ = = . (4.7)
DOI:10.1190/1.9781560803737

r (t) ds ds
dt

As shown in Figure 4.5, the angle u is the counter-clockwise angle from the
direction of the positive z-axis to the unit tangent. The unit tangent vector is
u = (sin u)ex + (cos u)ez . (4.8)
Let us summarize the pertinent facts about the unit tangent. Let the vector
r = (x, z) represent a point on a given ray. Let s denote arc length along
the ray. Let an adjacent point on the same ray be given by

r + Dr = (x + Dx, z + Dz) . (4.9)

The vector Dr = (Dx, Dz) is (approximately) a tangent vector to the ray. The
vector Dr connects the two points in question. The length of this vector is

Dr = Dx2 + Dz2 . (4.10)

Let the path length between the two points be Ds, which is approximately
equal to the increment of the arc length on the ray. The unit tangent to the
raypath is the limit of Dr/Ds as the points approach each other; that is

dr Dr
= lim . (4.11)
ds Ds 0 Ds
The length of the vector Dr is approximately equal to the path length
difference Ds. As a result, the vector dr/ds is a unit vector. As a result,
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

126 Basic Wave Analysis, Part 2: Raypath Analysis

the unit vector tangent to the ray is

 
dr dx dz dx dz
u= = ex + ez = , . (4.12)
ds ds ds ds ds
The vector u is directed along the tangent to the curve in the direction of
increasing values of arc length s. Let u be the angle that the ray makes
with the vertical axis. Then, the unit tangent vector is

u = (sin u, cosu) . (4.13)

Curvature
Next, let us discuss some properties associated with the unit tangent
vector u. The curvature is denoted by the lowercase Greek letter kappa k.
The curvature is defined as
 
du
DOI:10.1190/1.9781560803737

k= 
 ds  . (4.14)

We have
du
= (cos u)ex + (−sin u)ez . (4.15)
du
Because
du du d u
= , (4.16)
ds d u ds
we have
    
du  
  = du d u = d u (cos u)2 + (−sin u)2 =
du
. (4.17)
 ds  d u ds ds ds
Thus, the curvature is
du
k= . (4.18)
ds
This expression for curvature is not useful unless the formula for the curve is
a function of arc length s. Now, let us give an expression for curvature when
the formula for the curve is a function of an arbitrary parameter t. We have
 
 du 
   
du du ds du  dt 
= and k =   =    
  

.
dt ds dt ds  ds 
 
dt
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 4: Traveling Waves 127

 
ds  dr
Because of equations (4.6), i.e., =

, the curvature is
dt dt 
 
 du 
 
 dt  u′ (t)

k =  = ′ . (4.19)
 dr  r (t)
 
dt

For a plane curve given as (x(t), y(t)), the signed curvature is

x′ y′′ − y′ x′′
k= 3/2
, (4.20)
(x′ 2 + y′ 2 )
where primes refer to derivatives d/dt with respect to parameter t. In the case
of a plane curve given explicitly as y ¼ f (x), the signed curvature reduces to
DOI:10.1190/1.9781560803737

d2 y
k= dx2 . (4.21)
 2 3/2
dy
1+
dx

In the case of the vibrations of a taut stretched string, it is usual to assume

that the slope dy/dx is small compared with unity (see Figure 4.6). This con-
dition yields the approximation, which is called the taut string curvature,
given by
d2y dy
k= 2
under the condition that ≈0. (4.22)
dx dx

Figure 4.6. The taut

string condition
(small amplitude
means small slope).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

128 Basic Wave Analysis, Part 2: Raypath Analysis

Wave motion
Two major categories of linear traveling waves are mechanical
waves and electromagnetic waves. Mechanical waves require physical
matter in which to travel. They cannot travel in a vacuum. On the other
hand, electromagnetic waves can travel in a vacuum at the speed of light,
with that speed commonly denoted c. Originally, the standard meter was
in the form of a platinum bar which was kept in Paris. This in turn was
replaced in 1889 by 30 platinum-iridium bars kept at locations around the
world. However, using such physical objects for the definition of the
meter was imprecise. In 1960, the meter was defined as 1,650,763.73 wave-
lengths of light emitted by the krypton-86 isotope. In 1983, the meter
was defined as the length of the path travelled by light in a vacuum in
1/299,792,458 of a second. In other words, the meter is defined in such a
manner that the speed of light is exactly c ¼ 299,792,458 m/s.
DOI:10.1190/1.9781560803737

Of course, we know that electromagnetic waves can travel through

matter, such as air, water, and glass. The reason is that the interior of any
matter is largely a vacuum, because the molecules are spaced widely
apart. The mechanism of such transport involves the absorption and reemis-
sion of the wave energy by the molecules of the matter. When an electro-
magnetic wave impinges upon the molecules, the energy of that wave is
absorbed. This energy absorption is what causes the molecule’s electrons
to vibrate for a short period of time, and the vibrating electrons create a
new electromagnetic wave with the same frequency as the original electro-
magnetic wave. Although the electron vibrations occur for only a very short
period of time, they delay the motion of the wave through the medium. Once
the energy of the electromagnetic wave is reemitted by a molecule, it travels
through a small region of vacuum between molecules. When it reaches the
next molecule, the electromagnetic wave is absorbed, transformed into elec-
tron vibrations, and then reemitted as an electromagnetic wave. The electro-
magnetic wave travels at a speed of light through the vacuum of interatomic
space. The process of absorption and reemission causes the net speed of the
electromagnetic wave through matter to be less than the speed of light in a
vacuum. For that reason, the speed of light through air or water or glass is
less than the speed of light in a vacuum. In water, you would have to take
into account the absorption and transmission of the light particle between
every water droplet. Thus, the speed of light in water depends on the
quality and density of the water. The value 225,000 km/s is roughly the
measurement for the speed of light in water found experimentally.
For the case of a mechanical wave, the substance of the medium is
deformed. For the case in which the medium is a perfectly elastic substance,
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 4: Traveling Waves 129

there is no permanent displacement of the particles of the medium. As an

example, sound waves propagate when air molecules collide with their
neighbors, transferring energy from neighbor to neighbor down the line,
causing a cascade of collisions in a given direction. Because the collisions
are elastic, the air molecules oscillate about their equilibrium positions.
The molecules do not travel in the direction of the wave. Each molecule
stays in its same place whereas energy travels in the direction of the
wave. An ocean wave is a typical traveling wave. The particles of the
water move up and down in the same place while the passing wave
moves onward. The main property of a traveling wave is that it transfers
energy from one point to another.
Waves can be classified as either standing waves or traveling waves.
A standing wave, also known as a stationary wave, is a wave that remains
in a constant position. A traveling wave is not confined to a given
space in the medium, but propagates through the medium. The waves on
DOI:10.1190/1.9781560803737

a string of a musical instrument are examples of standing waves. Such

waves are the result of interference of traveling waves propagating in oppo-
site directions.
When water ripples reach an obstacle, these ripples (waves) are reflected
and a new set of ripples is produced. If the surface of the obstacle occurs at a
908 angle in relation to the initial direction of wave travel, the reflected
ripples will travel back in the opposite direction. If the waves confront an
obstacle at a different angle, the resulting ripples will be reflected in new
directions. Sound is a disturbance that is produced by the vibration of
some material body which can be detected by the ear. Sound is transmitted
through a transmission medium, such as air, water, or other materials, in the
form of longitudinal (compressional) waves. Sound waves are reflected
from surfaces, such as walls, mountains, clouds, or the ground. A sound is
seldom heard without accompanying reflection, especially inside a building
where the walls and furniture supply the reflecting surfaces. Even rolling
thunder is due primarily to successive reflection from clouds and land sur-
faces. A popular practice at the U.S. Capitol building in Washington,
D.C. is to send a friend to the far side of the capitol dome. Ordinary conver-
sation can be heard if the speaker and listener are both close to the wall of the
dome. Many buildings with dome-like rotundas exhibit this guided reflec-
tion phenomena. Locations where the rotunda effect is experienced are
sometimes called “whispering galleries.” The dome of St. Paul’s Cathedral
in London is a famous example.
Suppose that the earth’s atmosphere extends uniformly as far as the
moon. Take the distance to be 384,000 km, and use 340 m/s as the speed
of sound. It would take sound more than 13.3 days to reach the moon.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

130 Basic Wave Analysis, Part 2: Raypath Analysis

As a comparison, consider that the Apollo mission, which began on 16 July,

1969, took 51 hours and 49 minutes (i.e., 2.16 days) to reach the moon.
The ear is able to distinguish two sounds as separate only if they reach it
at least 0.1 s apart. Otherwise, they blend together and the listener hears what
appears to be a single sound. Suppose a sound of short duration is reflected
back to a listener after 0.1 s or longer. The listener would hear a repetition of
the original sound, i.e., an echo. In air, sound travels at a speed of approxi-
mately 340 m/s. In order for an echo to occur, the reflecting surface must
be at least 17 m away, so that the sound can travel the two-way distance
of 34 m from the listener to the reflector and back within the allotted 0.1 s.
In summary, the direction of advance of sound waves may be changed by
reflection from suitable surfaces. An echo occurs when a reflected sound
wave (echo) returns to a listener 0.1 s or more after the original wave
reaches the listener. The echo is a distinct repetition of the original sound.
In fact, a listener may calculate the approximate distance between their
DOI:10.1190/1.9781560803737

location and a thunder storm—the rough calculation in kilometers is the

number of seconds elapsed between the flash of lightning and the arrival
of the sound of the thunder divided by three.
A fathometer is an instrument for determining ocean depths. It makes
use of the reflection of sound waves in the water. A ship sends a sound
pulse down into the water. After being reflected from the sea bottom, the
returned sound is detected by an underwater receiver also mounted on the
ship. The time interval is recorded by a special device. The depth of the
sea at that point can be computed from knowledge of the elapsed time
and the speed of sound in water.
Indeed, sound waves (energy) may be reflected from curved surfaces so
that they travel in a specific direction, thereby making the sound more
readily audible at a distance. For example, in an auditorium a curved sound-
ing board may be placed behind a speaker so that sound waves that might
otherwise be lost to listeners are instead reflected back to them. Similarly,
an ear horn or ear trumpet can collect sound waves and convey their
energy to a listener or a different detector. In fact, a sound wave may be
refracted if the speed is not the same in all parts of the medium or if parts
of the medium are moving. It also may be refracted as it passes from one
medium to another.
A sound wave moving through air is composed of alternating regions of
higher and lower density. Regions of higher density are called compressions.
Regions of lower density are called rarefactions. When the prong of a tuning
fork vibrates in the air, the layer of air adjacent to the prong undergoes com-
pression; in effect, the prong movement squeezes the air molecules together
producing the compression segment. When the prong springs back in the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 4: Traveling Waves 131

opposite direction, an area of reduced air pressure is produced, resulting in

the segment of rarefaction. A succession of rarefactions and compressions
composes the longitudinal wave motion that emanates from an acoustic
source.
When two waves pass through the same medium at the same time, the
resulting particle motion represents the combined disturbances of the two
sets of waves. Generally known as interference, these combined effects
are relevant for all types of wave motion. If a listener is present in a room
with a loud whistle and walls which reflect the sound, then that person
can observe how the sound will be exceptionally loud at certain points
and quieter in other locations within the room. For locations at which a com-
pression of the reflected wave and a compression of the direct wave arrive at
the same time, the effects of both waves are combined and the sound is loud.
In contrast, when a rarefaction of one wave arrives with a compression of the
other, the effects will be partially, or even completely, cancelled and the
DOI:10.1190/1.9781560803737

sound will be quieter.

Interference also can result when two sets of sound waves with slightly
different frequencies travel through the air at the same time. The listener will
observe a “beat,” the swelling and fading of the sound. Because the com-
pressions and rarefactions of the set of waves do not possess the same
spacing, the resulting sound patterns will vary. When two compressions
arrive at the same time, the listener will hear sounds that are louder than
when a compression of one set of waves arrives with the rarefaction of
the other. The number of beats occurring each second is equal to the differ-
ence of the two frequencies. For example, two sets of waves of frequency 10
cycles/s and 12 cycles/s, respectively, combine and produce a resultant
sound wave which beats at 12 minus 10, or 2 cycles/s. In summary, beats
occur when two sources of different frequency are produced at the same
time and the listener experiences how the resulting sound will rise and
fall periodically in intensity as the waves alternately reinforce and cancel
each other.
Furthermore, wave motions may produce heat. This occurs because
some of the regular motion of particles in a wave motion can be converted
into irregular motion (heat) as the wave motion travels through a single
medium or between two mediums. This constitutes absorption of energy
from the wave, and the amount of sound that is absorbed during a wave’s
passage will depend on the material involved. Porous materials are good
absorbers of sound, because much of the wave energy is changed to heat
energy in the pores. For situations in which the objective is to reduce the
amount of sound that is transmitted through walls or floors or reflected
from walls, a material which generally absorbs sound should be used.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

132 Basic Wave Analysis, Part 2: Raypath Analysis

Importantly, waves may be characterized by their wave length, which is

the distance between either two successive compressions or rarefactions.
Relatedly, the frequency of a vibrating body is the number of complete
vibrations per second, and the frequency of a wave motion sent from a
source is the number of waves per second passing a specific point. These
two frequencies, in fact, will have the same value. For all wave motion,
the velocity v, frequency f, and wave length l are related by the equation
v = f l.

Sinusoidal waves
Let us now examine the sinusoidal wave u(x, t) that is given by the
equation
u(x, t) = A cos k(x − vt) . (4.23)
DOI:10.1190/1.9781560803737

The so-called phasor is defined as

phasor = A eik(x−vt) . (4.24)
Thus, the wave can be written as

u(x, t) = Re[A eik(x−vt) ] . (4.25)

Here, x is distance, t is time, v is velocity (speed), and the positive constant A

is amplitude. The positive constant k is known as the angular wave number,
or propagation number. The product kx is in radians. The cosine varies in
magnitude from +1 to – 1, so that the maximum value of u(x, t) is A.
Equation (4.25) represents a progressive wave traveling at speed v. This
sinusoidal disturbance is periodic in both space and time.
Leonhard Euler writes, “For the sake of brevity, we will always rep-
resent this number 2.718281828459. . . by the letter e. After exponential
quantities, the circular functions sine and cosine should be considered
because they arise when imaginary quantities are involved in the exponen-
tial.” An increase or decrease in the argument k(x − vt) by the amount 2p
leaves the cosine representation unaltered; that is,

A eik(x−vt)+2pi = A eik(x−vt) e+2pi = A eik(x−vt) . (4.26)

An increase/decrease by +p leads to a polarity change +1. We define the

spatial period l by the equation

A eik(x−vt)+kli = A eik(x−vt) e+kli = A eik(x−vt) . (4.27)

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 4: Traveling Waves 133

Thus, we have |kl| = 2p. If we take both k and l as positive numbers, then
we obtain the fundamental equation

2p
k= . (4.28)
l
A cycle is a complete alteration in which a phenomenon attains a maxi-
mum and minimum value, returning to a final value equal to the original
one. In our case, one cycle is the portion of the sine wave between two suc-
cessive crests (or two successive troughs or any other two corresponding
points).
The spatial period l is known as the wavelength. The wavelength l
is the distance between two successive crests (peaks) of the spatial
sinusoidal wave and commonly is expressed in units of distance. An
increase or decrease in x by the amount l leaves the sinusoidal wave u
unaltered; i.e.,
DOI:10.1190/1.9781560803737

u(x, t) = u(x + l, t) . (4.29)

In a completely analogous fashion, we can examine the temporal period T,

usually called simply the period (Robinson and Treitel, 2008). It is the
amount of time it takes for one complete wave to pass a stationary observer.
In this case, it is the repetitive behavior of the wave in time which is of inter-
est, so that

u(x, t) = u(x, t + T) . (4.30)

In terms of phasors, the equation (4.30) is

A ei[k(x−v(t+T))] = A eik(x−vt) e+kvTi = A eik(x−vt) . (4.31)

Therefore, we have |kvT| = 2p. Because these are all positive quan-
tities, we can write

kvT = 2p . (4.32)

As we have shown, k = 2p/l, so 2pvT/l = 2p, from which it follows that

l
T= . (4.33)
v
To describe the period, we first choose a fixed point in space. Then, we
measure the time interval between successive peaks (or troughs) to obtain
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

134 Basic Wave Analysis, Part 2: Raypath Analysis

the period. In Figure 4.7, the period T is the number of units of time per
cycle. The period commonly is expressed in seconds. More precisely, the
period is expressed in seconds per cycle.
To describe the wavelength, we take a picture of the wave at a fixed
instant of time. Then, we measure. Figure 4.8 shows a sine wave plotted
as a function of distance at a given time. The wavelength is the interval of
DOI:10.1190/1.9781560803737

Figure 4.7. A sine wave at a fixed point in space.

Figure 4.8. The sine wave at a fixed instant of time.

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 4: Traveling Waves 135

distance between two adjacent crests. Thus, the velocity is

l
v= . (4.34)
T
In words, this equation says that the velocity (which is distance divided
by time) is equal to the spatial wavelength l divided by the time period T.
When the wave is plotted as a function of distance at a given distance, the
wavelength is the interval of distance between two adjacent crests. When
the wave is plotted as a function of time at a given distance, the period is
the interval of time between two adjacent crests. It is helpful to have a
good grasp of the relationship between a period of time and the correspond-
ing frequency. Mathematically, the period is the reciprocal of the frequency
and vice versa. Because the symbol f is used for frequency and the symbol T
is used for period, this relationship is
DOI:10.1190/1.9781560803737

1 1
period = or T= . (4.35)
frequency f

The reciprocal of the period is the frequency f, or the number of cycles per
unit of time. The frequency f commonly is expressed in Hertz (Hz), or
cycles/s. The equation T = l/v becomes v = lf . The velocity v commonly
is expressed in m/s. The two quantities which often are used in the literature
of wave motion are the angular frequency (expressed in radians per second
or rad/s)
2p
v= (4.36)
T
and the angular wave number (expressed in radians per meter or rad/m)
2p
k= . (4.37)
l
We see that the angular wave number also is called the propagation number.
The wavelength, angular wave number, period, and frequency all describe
aspects of the repetitive nature of a wave in space and time. These concepts
can be applied to waves which are not sinusoidal as long as each wave is
composed of a regularly repeating pattern (i.e., periodic waves) (Robinson
and Treitel, 2008).
Next, consider the equation v = lf . If we divide wave velocity v
(expressed in meters per second) by wavelength l (expressed in meters
per cycle), the length term (meters) cancels out, and the result is frequency
f (expressed in cycles per second). Frequency f indicates the number of
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

136 Basic Wave Analysis, Part 2: Raypath Analysis

cycles which pass a given point in 1 second. Wavelength l is expressed in

meters per cycle. Then, we have the equation

v meters per second cycles

= = =f . (4.38)
l meters per cycle second

Thus, we can write

f l = frequency × wavelength = velocity = v . (4.39)

In a medium of constant velocity, we see that a wave with a short wavelength

has a high frequency, and one with a long wavelength has a low frequency.
For example, if the motion repeats every 0.01 second, then the frequency is
100 Hz (i.e., 100 cycles per second). Furthermore, as the waves move, the
spatial distance between any two adjacent pulses is always the same, and
is the wavelength l. Because the pulses are separated by a distance l and
DOI:10.1190/1.9781560803737

because each pulse moves over this distance in a time T, it follows that
the velocity of propagation is v = l/T (Robinson and Treitel, 2008).
Using the equation f = 1/T, we again find that v = f l, or that the velocity
of propagation of a periodic wave is the product of the frequency and the
wavelength. Because
 
2p
l
kl l
kv = =   =v, (4.40)
T 2p
v

we have

u(x, t) = A cos k(x − vt) = A cos(kx − v t) . (4.41)

In summary, the basic parameters used to describe a traveling wave are

amplitude, period, frequency, wavelength, and velocity. We have defined
these terms in the case of a sinusoidal wave. The amplitude is the magnitude
of the maximum disturbance from the central value. The period, denoted by
T, is the length of time taken by one cycle, and the frequency f is the recipro-
cal of period. The wavelength l indicates how far the wave moves during
one cycle. In other words, the wavelength is the distance from one crest
to the next crest. The wave velocity v is how fast the wave is moving.
Because velocity is the ratio of distance over time, the velocity of the
wave is equal to the quotient of wavelength over period. Alternatively,
the velocity is equal to wavelength multiplied by frequency.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 4: Traveling Waves 137

A cycle is one of a succession of periodically recurring events. A cycle is

defined as the part of the wave between one crest and the next, or else
between one trough and the next. In general, cycle is the part of the
wave between any point and the next corresponding point in the
wave motion.
The frequency f is the number of cycles per second, otherwise known
as a hertz.
The period T = 1/f is the reciprocal of frequency. The period T is
measured in units of seconds per cycle.
The wavenumber k is the number of cycles per meter.
The wavelength l = 1/k is the reciprocal of wavenumber. The wave
length l is measured in units of distance per cycle.
There are 2p radians per cycle.
The angular frequency v = 2pf is the number of radians per second.
The angular wavenumber k = 2pk is the number of radians per meter.
DOI:10.1190/1.9781560803737

The wavelength l is the distance between one crest and the next or
between one trough and the next. In general, the wave length is the distance
between any point and the next corresponding point in the wave motion.
Wavenumber (sometimes called the propagation number) is a measure of
spatial scale. It can be thought of as a spatial analog of temporal frequency,
and can be called spatial frequency. Often, it is defined as the number of
wavelengths per unit distance, or in terms of wavelength k = 1/l.
A simple relation exists between frequency f, wave length l, and wave
speed v. For example, a source will vibrate for a time t, and the number of
waves produced will be ft. At the end of this time, the first wave will have
reached a distance equal to vt. This distance is equal to the number of
waves times the length of each wave. Therefore,

vt = ftl or v = fl . (4.42)

The speed of sound in air is 340 meters per second. The wave length of a
sound with a frequency of 256 cycles per second is

v 340 meters per second

l= = = 1.33 meters per cycle . (4.43)
f 256 cycles per second
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms
DOI:10.1190/1.9781560803737
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 5

Basic Properties of Waves

“Those who are accustomed to judge by feeling do not understand

the process of reasoning, because they want to comprehend at a
glance and are not used to seeking for first principles. Those, on
DOI:10.1190/1.9781560803737

the other hand, who are accustomed to reason from first principles
do not understand matters of feeling at all, because they look for
first principles and are unable to comprehend at a glance.”

—Blaise Pascal

Rays and wavefronts

Let us suppose that a point source is emitting light waves of wavelength
l in all directions. These waves may be represented by a set of spheres (with
radial spacing l) spreading from their center source. Because every point on
each sphere is equidistant from the source, it can be thought of as represent-
ing the crest of a wave. If straight lines are drawn outward from the source,
each line (ray) represents the direction along which the wave is advancing.
Figure 5.1 shows spherical waves spreading from a small source and
also rays drawn to show the direction in which the waves are moving. In
an isotropic medium, the rays always cross the waves perpendicularly.
The rays are merely convenient construction lines that often enable us to
discuss the travel of light more simply than by drawing the waves. A ray
is a line showing the direction of flow of radiant energy. It is a mathematical
device rather than a physical entity. In practice, one can sometimes produce
very narrow beams or pencils, and we might imagine a ray to be the unattain-
able limit on the narrowness of such a beam (Robinson and Treitel, 2008).

139
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

140 Basic Wave Analysis, Part 2: Raypath Analysis

The dual of the ray is the wavefront,

and vice versa. When we look at the
waves emanating from the point where
we dropped a stone into a still pond, we
see the wave motion traveling outward
as wavefronts, and the raypaths are
only mental constructions. Likewise, in
seismic interpretation, the rays are only
visualized as mathematical abstractions,
whereas in various cases wavefronts can
be observed on seismic sections. In an
isotropic medium (i.e., one in which
Figure 5.1. Light waves and rays. properties are the same in all directions),
The concentric arcs represent rays are orthogonal trajectories of the
sections of wavefronts; the straight wavefronts. That is to say, they are
DOI:10.1190/1.9781560803737

lines represent rays. lines normal (i.e., perpendicular) to the

wavefronts at every point of intersec-
tion. In such a medium, a ray is parallel to the propagation vector.
However, this is generally not true in anisotropic substances.

Velocity
One of the most important equations in physics involves distance,
rate, and time. Before we write down the equation, let us explain each
factor. Rate is the speed at which something travels. Time refers to how
long it takes to travel a certain distance at a certain speed. Distance is
how far something travels in a certain amount of time at a certain speed.
We will now give the equation from which we can easily calculate distance,
velocity, and time. The velocity equation is:

distance = velocity × time . (5.1)

In ordinary life, we encounter friction. Friction makes it more difficult

to push things and it wears out moving engine parts. How nice it would be
without friction! However, life without friction would be impossible. You
would slip on anything and you could not stop your moving vehicle. In
seismic analysis, we are confronted with the many variations of velocity.
Velocity in all its forms and values presents the most difficult problem in
all aspects of seismic analysis. Astronomers must deal with the velocity of
light, but fortunately its velocity in a vacuum is always the same constant.
It would be convenient if seismic velocity were always the same constant.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 5: Basic Properties of Waves 141

However, such a situation would not be good. Reflection from an interface

is possible because of a change in velocity (more precisely, impedance) at
the interface. The accurate estimation of seismic velocities is critical. It
opens up the entire vista of remote sensing. Without reflection, there
would be no way to explore for hydrocarbons or minerals at depth. Fur-
thermore, we could argue that an earth without reflecting interfaces is
not likely to bear any hydrocarbon reservoirs. Hence, there are both geo-
logical and wave-theoretical reasons why we study reflection seismology.
In the 17th century, Robert Hooke writes, “By the help of microscopes,
there is nothing so small, as to escape our inquiry; hence there is a new
visible world discovered to the understanding.” The focusing property
of velocity analysis gives us a sharp reflectivity image with the main
reflected events positioned at their correct depths, much like focused
binoculars.
Velocity is a vector quantity that denotes the rate of change of position
DOI:10.1190/1.9781560803737

with respect to time, or a speed with the directional component. In geophys-

ics, we often deal with isotropic rock which has the property that, at a point,
velocity has the same magnitude in all directions. As a result, often the term
velocity is used in situations in which more correctly the term speed should
be used. Velocity is the rate of displacement of an object. It is measured in
m/s. Acceleration is the rate of change of velocity of an object. It is
measured in m/s2. They are both vector quantities; i.e., both magnitude
and direction are required to fully specify them.
Let us look at the motion of a particle moving along a curve on a plane.
First, consider the simplest case, which is the case in which the curve is a
straight line. Let the position of the particle on the line be given by the
arc length s(t). The instantaneous speed and acceleration of the particle
are defined by
ds
speed = v(t) = (5.2)
dt

dv d 2 s
acceleration = a(t) = = . (5.3)
dt dt2
On the straight line, there are only two possible directions for the motion.
The sign of v(t) distinguishes the direction of motion. A positive speed
means that the particle is moving in the positive direction of the straight
line, and a negative speed means that it is moving in negative direction.
A vector can change in both length and direction.
In Figure 5.2, a particle moves along a curve in two coordinates. At each
instant, the direction of motion is considered to be the direction of the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

142 Basic Wave Analysis, Part 2: Raypath Analysis

Figure 5.2. Velocity vectors on a curve.

tangent to the curve. Vectors are used to deal with this problem. At each
DOI:10.1190/1.9781560803737

instant, the speed of the particle is the instantaneous rate at which the arc
length traveled by the particle is changing with time. The speed and direc-
tion of motion together form a vector quantity called the velocity of the par-
ticle. This velocity can be represented geometrically by an arrow that points
in the direction of motion and its magnitude is equal to the speed of the
particle.
If a particle moves along a curve so its position vector at time t is r(t),
then the instantaneous velocity, instantaneous speed, and instantaneous
acceleration of the particle are, respectively:
dr
velocity = v(t) = (5.4)
dt
 
dr ds
speed = v(t) = v(t) =  
 dt  = dt (5.5)

dv d 2 r
acceleration = a(t) = = 2 . (5.6)
dt dt

In summary, velocity is the vector v(t) that is tangent to the curve and points
in the direction of motion. Moreover, speed is the scalar v(t) that is the rate at
which arc length traveled changes with time.
Let us examine the sinusoidal wave given by

u(x, t) = A cos k(x − vt + 1) = A cos (kx − vt + 1) . (5.7)

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 5: Basic Properties of Waves 143

The argument of the sinusoidal function is known as the phase f of the

wave; that is,

f = kx − vt + 1 . (5.8)

At t = x = 0, we have f = 1. Thus, 1 is called the initial phase.

Let us briefly explain the partial derivative. Let f (x, y) = x3 y2 . To cal-
∂f (x, y)
culate the partial derivative , we view y as being a fixed number
∂x
and calculate the ordinary derivative with respect to x. To give a simple
demonstration, set y = b, where b is a constant. Then, the partial derivative
of x3 y2 with respect to x is the same as the ordinary derivative of x3 b2 . The
ordinary derivative is 3x2 b2 . Thus, the required partial derivative is 3x2 y2 .
The partial derivative of f with respect to t, if we hold x constant, is the
rate of change of phase with respect to time, and is equal to the negative
DOI:10.1190/1.9781560803737

angular frequency

∂f
= −v . (5.9)
∂t
Similarly, the rate of change of phase with distance x, if we hold t constant, is
the wavenumber

∂f
=k. (5.10)
∂x
The condition of constant phase is expressed as

f(x, t) = kx − vt + 1 = constant . (5.11)

Taking differentials, we have

d f = k dx − v dt = 0 or k dx = v dt . (5.12)

Solving this equation, we obtain


∂x v
= . (5.13)
∂t f=constant k

The term on the left represents the velocity of propagation, subject to the
condition of constant phase. In the case of a plane wave without geometrical
spreading, the displacement u of the crest remains constant as the wave
moves through the material. Choose any point on the wave profile, e.g.,
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

144 Basic Wave Analysis, Part 2: Raypath Analysis

the crest of the wave. Because the only variable in the sinusoidal wave func-
tion is the phase, it too must be constant. That is, the phase is fixed at a value
producing the constant displacement u at the chosen point. The point moves
along with the profile at the velocity v, and so does the constant phase
condition as well (Robinson and Treitel, 2008).
Because velocity is equal to frequency times wavelength, i.e., v = f l,
and because v = 2pf and k = 2p/l, it follows that v = v/k. Therefore,
the above equation also can be written as

∂f v
 = =v. (5.14)
∂t f=constant k

Thus, the speed at which the profile moves is the wave velocity v or, more
specifically, the phase velocity. The phase velocity carries a positive sign
when the wave moves in the direction of increasing x, and carries a negative
DOI:10.1190/1.9781560803737

sign when the wave moves in the direction of decreasing x.

Wave equation
For the simple case in which the medium is uniform everywhere, the ray
paths are straight lines. This leads to the principle of rectilinear propagation.
This type of wave propagation may be illustrated by a disturbance originat-
ing at a point in a medium which is perfectly homogeneous. The disturbance
propagates as a spherical wavefront and radial ray paths from a point source
in a uniform medium. The source itself could, for instance, be a shot of dyna-
mite. Because the medium is uniform, the wave proceeds outward in all
directions from the source, and its successive positions will assume the
form of expanding concentric spheres with the source as center. As a conse-
quence of uniformity, the velocity of the wave will be constant everywhere
in the medium. The successive positions of the wave after equal intervals of
elapsed time will, accordingly, be concentrically spaced spheres whose radii
differ by equal increments. Each wavefront, which represents the locus of
the disturbance after certain time interval, is a spherical arc. If the total
elapsed time is t and the velocity is v, then the radius of the sphere is
equal to vt.
Each radius represents a raypath associated with a spherical wavefront.
A raypath may be given physical significance if we think of it as the path
over which the energy of the disturbance travels from the source S to the
spherical wavefront. It is evident in this simple case that the ray paths are
everywhere at right angles to the wavefronts. For isotropic media, the ray
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 5: Basic Properties of Waves 145

paths are always orthogonal to the wavefronts regardless of the wavefront

shape (Robinson and Treitel, 2008).
For our present purpose, we define the leading-edge wavefront as the
most forward position of the advancing region of disturbance at any partic-
ular instant of time. Behind the leading-edge wavefront, the medium has
been disturbed. Ahead of the leading-edge wavefront, the medium is undis-
turbed. It is merely the physical effect caused by the original disturbance
being propagated away from its source as a consequence of the natural
elastic behavior of the medium.
The three familiar dimensions of space are depth, length, and width. In
Einstein’s general relativity, space and time are not modeled as separate
entities but instead are unified to a four-dimensional spacetime. Physicists
sometimes use more dimensions in order to make a model more mathemat-
ically tractable and to gain general insights more easily. However, there
could be more than four dimensions of spacetime, in which case the
DOI:10.1190/1.9781560803737

additional dimensions have managed somehow to escape detection. String

theories require more dimensions of spacetime for their mathematical
consistency. In bosonic string theory, spacetime is 26-dimensional, while
in superstring theory it is 10-dimensional, and in M-theory it is 11-
dimensional.
In the 19th century, seismology was one-dimensional (depth) in the
sense that a single seismograph was used at an earthquake recording
station. In the 20th century, seismology was two-dimensional (depth,
length) in the sense that a line of geophones on the surface was used in
exploration. In the 21st century, seismology has been three-dimensional
(depth, length, width) in the sense that an array of geophones on the
surface is used in exploration.
In comparison to modern physics, geophysics appears quite modest.
Generally, geophysics does not extend beyond classical physics.
However, geophysics has enough complications to satisfy the enquiring
mind. For example, wave propagation in 2D (even-numbered dimensions)
is fundamentally different from 3D (odd-numbered dimensions). [Note
that “4D” is semantics to denote the situation in which the evolution of a
hydrocarbon reservoir is monitored in a time-lapse fashion; this, however,
is not related to wave propagation in any 4D space.] With this caveat, let
us illustrate wave propagation in two dimensions. The example is that of
tossing a stone into a pond of water. The initial water disturbance is set
up at the point where the stone strikes the surface, and the appearance of
the resultant circular expanding waves is a matter of common experience.
This illustration of wave motion is so captivating that there are stories
that Huygens himself pondered it.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

146 Basic Wave Analysis, Part 2: Raypath Analysis

Galileo’s pupil, Vincenzo Viviani, wrote that, in 1589, Galileo dropped

two balls of different masses from the Leaning Tower of Pisa. According to
legend, Galileo thereby discovered that their time of descent was indepen-
dent of their mass. Galileo also discovered that the objects undergoing
the force of gravity fell with the same acceleration. Actually, the objects
fell so fast that he could not discover such things with the instruments
then available. The genius of Galileo was that he devised an experiment
whereby he could discover such things with his crude instruments. His
experiment consisted of letting gravity roll a ball down an inclined plane,
in which case he could make the necessary measurements. However,
today with modern instruments and clocks, we could easily make the necess-
ary measurements for Galileo’s Pisa experiment. Like him, we would find
that force was equal to mass times acceleration. This relationship became
Newton’s second law of motion. In 1660, Hooke discovered the law of elas-
ticity. Known as Hooke’s law, it described the linear variation of tension
DOI:10.1190/1.9781560803737

with extension in an elastic spring. Hooke first described this discovery in

the anagram “ceiiinosssttuv,” whose solution he published in 1678 as “Ut
tensio, sic vis,” meaning “As the extension, so the force.” The wave equation
could be derived using Newton’s second law of motion together with
Hooke’s law.
Physicists are well aware of the concept of force. A mathematician
uses mathematical expressions for the various forces. Seismic waves
are, first of all, elastic waves. Such waves are determined by elastic
forces present in rocks under deformation within the elastic limits. Start-
ing from these first principles, one can derive the wave equation. We will
provide an elementary description here and then a derivation from first
principles in the next section. Let v represent the velocity of the traveling
wave. For our purposes, we assume that v is a constant. The dependent
variable u can be particle displacement, or pressure, or some other
related quantity. If the independent variable is time t, then the first deriva-
tive is a velocity and the second derivative is an acceleration. On the other
hand, if the independent variable is travel distance x, then the first deri-
vative is a slope and the second derivative is the so-called taut-string
curvature.
In essence, the wave equation says that taut-string curvature is pro-
portional to acceleration. The proportionality constant is 1/v2 . Thus, the
wave equation is

1
taut-string curvature = acceleration . (5.15)
v2
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 5: Basic Properties of Waves 147

In one dimension (the x-axis), the wave equation is

∂2 u 1 ∂2 u
= . (5.16)
∂x2 v2 ∂t2
The time coordinate is t, the space coordinate is x, the velocity is v, and the
wave is u.

Telegrapher’s equations
Oliver Heaviside (1850–1925) was born in London. He suffered from
scarlet fever when young, which left him with a hearing impairment. He
was a good student, but his parents could not afford to keep him at school
after he was 16. Heaviside’s uncle by marriage was Sir Charles Wheatstone
(1802– 1875). Although the telegraph had many inventors, it was Wheat-
DOI:10.1190/1.9781560803737

stone with William Fothergill Cooke (1806– 1879) who patented a telegraph
system in 1837 that used a number of needles on a board that could be moved
to point to letters of the alphabet. (Samuel Morse [1791–1872] indepen-
dently developed and patented a recording electric telegraph in 1837.) In
1867, Wheatstone gave Heaviside a job at one of his telegraph companies;
from this experience, Heaviside adapted complex numbers to the study of
electrical circuits, invented mathematical techniques for the solution of
differential equations, and independently co-formulated vector analysis. In
the 1880s, Oliver Heaviside developed the transmission line model.
The transmission line model makes use of the so-called telegrapher’s
equations, which are a pair of coupled, linear differential equations that
describe the voltage and current on an electrical transmission line with
distance and time. The model demonstrates that electromagnetic waves
can be reflected on a wire, and the resulting wave patterns can appear
along the line. The theory applies to high-frequency transmission lines
(e.g., telegraph wires and radio frequency conductors), audio frequency
(e.g., telephone lines), low-frequency transmission lines (e.g., power
lines), and direct current. It also applies to seismic analysis, which we
do in this section.
The term stress expresses the loading in terms of force applied to a
certain cross-sectional area of an object. The internal force acting on a
small area of a plane can be represented by three components: two com-
ponents which are parallel to the plane and one which is normal to the
plane. The normal stress is calculated as the normal force component
divided by the area; the shear stress is derived as the parallel force
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

148 Basic Wave Analysis, Part 2: Raypath Analysis

components divided by the area. In a finite area, these stresses are average
stresses. When the area is allowed to approach zero, however, the stresses
become stresses at a point. It is notable that the stresses on any plane can
be computed from the stresses on three orthogonal planes passing through
the point. Furthermore, because each plane has three stresses, the stress
tensor has nine stress components, which completely describe the state of
stress at a point.
The response of a system to an applied stress is strain. Stress is
produced when a material is loaded with a force, causing the material to
deform. Engineering strain is defined as the amount of deformation in the
direction of the applied force divided by the initial length of the material.
In the example of a bar stretched in tension, the strain is calculated as the
amount of change in length (elongation) divided by the material’s initial
length. Strain distribution may or may not be uniform in a complex struc-
tural element; similar to uniformity of stress, the uniformity of strain will
DOI:10.1190/1.9781560803737

depend upon the nature of the loading condition.

Furthermore, the degree of stress will affect the temporary or permanent
response of the material. For a case in which a stress is minor, the strain on
the material may be minor, so the material will be able to return to its orig-
inal size after the stress is removed. Called elastic deformation, this form of
material deformation will occur only when stresses are lower than a critical
stress (i.e., the yield strength). In contrast, plastic deformation occurs when a
material is loaded beyond its elastic limit; the material remains in a
deformed condition after the load is removed.
Now, we will derive the wave equation for plane compressional waves
from first principles. Let the direction of propagation be the direction of the
vertical axis with depth z measured positively down. Let p denote the
pressure defined as negative stress; that is,

p = −szz . (5.17)

Let w denote the z component of particle displacement, so the normal

strain is

∂w
e= . (5.18)
∂z

In the case of particle displacement w, we will make use of the overhead dot.
The dot indicates ∂/∂t. Thus, particle velocity ∂w/∂t is denoted ẇ. The
reason for this simplified notation is that particle velocity is an important
quantity in its own right.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 5: Basic Properties of Waves 149

Hooke’s law states that, for a small deformation, the stress in a body is
proportional to the corresponding strain; i.e.,

stress = E × strain . (5.19)

Here, E is a constant called modulus of elasticity. Depending upon the nature

of force applied on the body, the modulus of the elasticity is classified into
three types: Young’s modulus of elasticity, bulk modulus EB of elasticity
(which we will use), and modulus of rigidity.
When a uniform pressure (normal force) is applied all over the surface
of a body, the volume of the body changes. For small strains, Hooke’s law
states that

stress szz = bulk modulus EB × strain ∂w/∂z . (5.20)

Because p = −szz , Hooke’s law becomes

DOI:10.1190/1.9781560803737

∂w
−p = EB . (5.21)
∂z
If we differentiate equation (5.21) with respect to time, we obtain

∂p ∂ẇ
= −EB . (5.22)
∂t ∂z
This is the first of two equations which we need. The other equation is
Newton’s law, which says force equals mass × acceleration. Force per
unit volume is

∂szz ∂p
=− . (5.23)
∂z ∂z
∂2 w
Mass per unit volume is r. Acceleration is 2 .
Thus, the Newton’s law equation is ∂t

∂p ∂ẇ
= −r . (5.24)
∂z ∂t
This is the second of two equations which we need. These two equations,
called the telegrapher’s equations, are

∂p ∂ẇ ∂p ∂ẇ
= −EB and = −r . (5.25)
∂t ∂z ∂z ∂t
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

150 Basic Wave Analysis, Part 2: Raypath Analysis

On the left sides are, respectively, the time and space derivatives of pressure.
On the right sides are, respectively, the space and time derivatives of particle
velocity. Such a pair of equations is known in electrical engineering as the
telegrapher’s or transmission line equations. Lord Kelvin devised the pair
to describe transmission on the first submarine Atlantic Ocean cable. In
analogy with Kelvin’s results, we call p the voltage, w the current, 1/EB
the capacitance, and r the inductance. The two telegrapher’s equations
can be combined by differentiating one with respect to t and the other
with respect to z and then eliminating either p or ẇ. Then, we have either

∂2 p r ∂2 p ∂2 ẇ r ∂2 ẇ
= or = . (5.26)
∂z2 EB ∂t2 ∂z2 EB ∂t2

Equations (5.26) are the one-dimensional scalar wave equation for p and for
ẇ, respectively. Thus, pressure and particle velocity each propagate as tra-
DOI:10.1190/1.9781560803737

veling waves. The general solutions of the two wave equations (5.26) are
 z  z
p = p1 t − + p2 t +
v v
 z   z
(5.27)
ẇ = ẇ1 t − + ẇ2 t + .
v v
Here, p1 and ẇ1 are downgoing waveforms and p2 and ẇ2 are upgoing wave-

EB
forms. The wave velocity is the constant v = r .
Let us now find the relationships between the waveforms using the
telegrapher’s equations. Substituting expressions (5.27) for p and ẇ into
Hooke’s law and Newton’s law equations, we obtain, respectively,
EB ′
p′1 + p′2 = (ẇ1 − ẇ′2 )
v
(5.28)
1 ′
(p − p′3 ) = r (ẇ′1 + ẇ′2 ) .
v 1
The prime indicates the derivative of a function with respect to its argument,
z in this case. Solving these equations, we obtain the pair

EB
2p′1 = ẇ′1 + vr
v
 (5.29)
EB
2p′2 = ẇ′2 + vr .
v
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 5: Basic Properties of Waves 151

Define the impedance Z by

EB 
Z= = vr = rEB . (5.30)
v
Equations (5.29) and (5.30) yield

p′1 = ẇ′1 Z
(5.31)
p′2 = ẇ′2 Z .
Upon integration, we obtain

p1 = Z ẇ1 (downgoing waves)

(5.32)
p2 = −Z ẇ2 (upgoing waves) .
Therefore, we have shown that the pressure for each wave is proportional to
DOI:10.1190/1.9781560803737

the particle velocity of that wave, and that the constant of proportionality is
simply the impedance Z. The pressure and particle velocity for a downgoing
wave (i.e., traveling in the plus z direction) are in phase (i.e., p1 = +Z ẇ1 ),
whereas, for an upgoing wave (i.e., traveling in the minus z direction), they
are 1808 out-of-phase (i.e., p2 = −Z ẇ2 ).
Finally, we know that instantaneous power is the product of force ×
velocity. Hence, the energy flux for downgoing waves is

p21
p1 ẇ1 = Z ẇ21 = (downgoing energy flux) (5.33)
Z
and for upgoing waves is
p22
p2 ẇ2 = −Z ẇ22 = − (upgoing energy flux) . (5.34)
Z
The negative signs in equation (5.34) merely indicate that the flow of energy
is in the negative z direction. Thus, the net downgoing energy flux is the sum
of the energy fluxes, that is,

p21 p22
p1 ẇ1 + p2 ẇ2 = Z ẇ21 − Z ẇ22 = − . (5.35)
Z Z
For land exploration, geophones are used as receivers. Most of the geo-
phones typically used in exploration sense vertical particle velocity, that
is, the quantity we have designated as ẇ. For sea exploration, hydrophones
are used as receivers. Hydrophones record pressure; that is, the quantity that
we have designated as p. Because p and ẇ are proportional, the seismic
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

152 Basic Wave Analysis, Part 2: Raypath Analysis

record sections from a land survey can be merged with the seismic record
sections from a sea survey. However, because the two types of records
can be 1808 out-of-phase, the polarities of the recording instruments must
be known. Even so, this is not the only step in reconciling land and sea
records. In the words of Shakespeare, “But no perfection is so absolute,
That some impurity doth not pollute.” Generally, the sensitivity and fre-
quency response of the instruments will be different as well.

Downgoing waves and upgoing waves

The seismic method relies on the fact that seismic traveling waves
propagate through the earth. In reflection seismic surveying, seismic
waves are generated at shot points and the waves travel downward
through the earth. As the downward-traveling waves encounter various
reflecting surfaces between successive subsurface layers, they are partially
DOI:10.1190/1.9781560803737

reflected upward. The resulting upward-traveling waves also are partially

reflected downward when they encounter reflecting surfaces between suc-
cessive layers. The waves received at buried or submerged detection
points are composed of both upward-traveling waves and downward-travel-
ing waves.
A dual sensor consists of both a pressure-sensitive seismic detector
and a particle-velocity-sensitive seismic detector. The outputs of the two
detectors are combined in order to attain separately the upcoming waves
and the downward-traveling waves. In other words, the dual sensor
renders the receiving system suitable for differentiating between various
waves, such as longitudinal waves traveling in one direction and longitudi-
nal waves traveling in the opposite direction.
In order to effectively use all of the seismic processing methods avail-
able today, it is essential to be able to measure the traveling waves upon
which ray theory is based. A hydrophone cannot measure a traveling
wave as such. Neither can a geophone. However, in a dual sensor, both
types of detectors can be used in conjunction with one another to measure
the traveling waves required for effective geophysical signal processing.
The dual sensor involves two detectors. One seismic wave detector at
each receiving station is pressure-sensitive, while the other seismic wave
detector located at the same receiving station is velocity-sensitive.
A geophone measures particle velocity (denoted by ẇ). A hydrophone
measures pressure (denoted by p). However, we do not want to use those
measurements as such. Instead, we want to make use of the downgoing com-
ponent of the particle velocity disturbance (denoted by D) and the upgoing
component of the particle velocity disturbance (denoted by U ). Similarly,
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 5: Basic Properties of Waves 153

we let d denote the downgoing component of the pressure disturbance and

let u denote the upgoing component of the pressure disturbance.
Augustin Jean Fresnel (1788–1827) is credited with showing that the
definition of the reflection coefficient of an interface requires the consider-
ation of both the particle velocity attribute ẇ and the pressure attribute p
of the wave motion on each side of the interface. Both of these attributes
must be continuous across the interface. A disturbance satisfying the
wave equation is equal to the sum of two traveling waves, one of which
travels downward and the other upward. The downgoing and upgoing
waves transport the energy to and from a reflecting horizon, and the particle
velocity and pressure attributes of Fresnel determine the partition of energy
at that horizon.
The net result is that seismic processing simply cannot be done with
conventionally recorded data unless one is willing to make traveling-wave
assumptions. However, no traveling-wave assumption must be made if a
DOI:10.1190/1.9781560803737

dual-sensor (that is, both a geophone and a hydrophone) is used as the re-
ceiver. A dual sensor provides both the hydrophone signal and the geophone
signal. Consider this example. The signals from a geophone and a hydro-
phone are out-of-phase; therefore, it can be concluded that the event is a tra-
veling upgoing wave. In contrast, if the geophone and hydrophone signals
are in-phase, then the event would be a downward-traveling wave (Robinson
and Treitel, 2008).
Now, consider the case of a buried receiver. The source also can be
buried, but it should be located at a level above the level of the receiver.
Then, the layered earth system (lying above the basement rock) is divided
into two subsystems by the horizontal plane which passes through the re-
ceiver. The upper subsystem of layers is called the shallow system, and
the lower subsystem is called the deep system (see Figure 5.3). Because
the shallow system contains the source of energy, it is active. Because the
deep system does not contain a source of energy, it is passive.
In the deep system, the downgoing wave at the receiver is one input; the
upgoing wave at the lowermost interface is the other input (which is null).
The upgoing wave at the receiver is one output; the downgoing wave at
the lowermost interface is the other output. The deep system is a passive
system with only one input (the downgoing wave at the receiver). If we con-
sider the output given by the upgoing wave at the receiver, then we have a
conventional convolutional model with one input and one output (see
Figure 5.4).
The reflection response involves only the reflection coefficients of the
deep system, and in no way depends upon the reflection coefficients of
the shallow system. Two wavelets have to be estimated, i.e., the input
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

154 Basic Wave Analysis, Part 2: Raypath Analysis

Figure 5.3. A deep system with two inputs and two outputs.
DOI:10.1190/1.9781560803737

Figure 5.4. The convolutional model for the deep-layered system.

wavelet and the output wavelet, which both occur at the receiver location.
A geophone measures the particle-velocity attribute of the seismic disturb-
ance, and a hydrophone measures the pressure attribute of the seismic
disturbance. Seismic processing is based upon the availability of downward-
and upward-traveling waves. However, a traveling wave is never recorded
as such in seismic acquisition. The signal recorded by a geophone is the
sum of the particle velocity attributes ẇ of the downgoing and upgoing
waves. The signal recorded by a hydrophone is the sum of the pressure attri-
butes p of the downgoing and upgoing waves (Robinson, 2000).
When the receiver is placed at the surface of the earth, we naturally
assume that the received wave is upgoing. This assumption is called a
traveling-wave assumption. However, no such assumption is possible
when the receiver is placed below the surface. What can we do? We must
use a dual sensor and record both the particle velocity attribute ẇ and the
pressure attribute p. Then, we must go back in time and appeal to Augustin
Jean Fresnel and Jean Baptiste Le Rond d’Alembert (1717–1783) for
their help.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 5: Basic Properties of Waves 155

As discussed in this section, Fresnel shows that the determination of the

reflection coefficient of an interface requires the consideration of both the
particle velocity attribute and the pressure attribute of the wave motion on
each side of the interface; both of these attributes must be continuous
across the interface. D’Alembert shows that a disturbance satisfying the
wave equation is equal to the sum of two traveling waves, one of which
travels downward and the other upward. The downgoing and upgoing
waves of d’Alembert transport the energy to and from a reflecting
horizon, and the particle velocity and the pressure attributes of Fresnel deter-
mine the partition of energy at that horizon.
Let us return to the use of a dual-sensor (that is, both a geophone and a
hydrophone) acting as the receiver. A traveling-wave assumption is
unnecessary, but the directions of the waves still must be determined.
Figure 5.5 shows an illustration of geophone and hydrophone signals
being out-of-phase. Although neither by itself would be enough to determine
DOI:10.1190/1.9781560803737

whether the event is upgoing or downgoing, the two signals are out-of-
phase, so (by the convention used here) the event is upgoing. On the other
hand, if the geophone and hydrophone signals were in-phase, then the
event would be a downward-traveling wave.
Furthermore, in the case of dual sensors, the traveling waves can be
computed directly from the data by means of the d’Alembert equations.
To do so, consider Figure 5.6. The inputs are particle velocity, pressure,
and acoustic impedance, all measured at the receiver location. The
outputs are the downgoing particle-velocity wave and the upgoing par-
ticle-velocity wave, both occurring at the receiver location. The receiver
is the dual sensor buried below the source of seismic energy to measure
the particle velocity and pressure signals at the receiver location. By use
of the d’Alembert equations, the particle velocity signal and the pressure
signal are converted into the downgoing wave and the upgoing wave at
the receiver location (Robinson and Treitel, 2008). The downgoing wave

Figure 5.5. The

event is an upward-
traveling wave.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

156 Basic Wave Analysis, Part 2: Raypath Analysis

Figure 5.6. The d’Alembert

equations.
DOI:10.1190/1.9781560803737

is the input signal and the upgoing wave is the output signal that occurs in the
convolutional model (as shown in Figure 5.4) of the deep system.
We can derive the d’Alembert equations as follows. For a given rock
layer, let r denote the density and let v denote the wave propagation velocity.
The product Z = rv is the acoustic impedance. A dual sensor is composed of
a geophone and a hydrophone. The geophone records the particle-velocity
trace ẇ and the hydrophone records the pressure trace p. Each trace is
equal to the sum of the downgoing wave motion plus the upcoming wave
motion at the sensor. Let D denote the downgoing wave motion of the par-
ticle-velocity trace and let U denote the upgoing wave motion of the particle
velocity trace. Similarly, let d denote the downgoing wave motion of the
pressure trace and let u denote the upgoing wave motion of the pressure
trace. Thus, we have the two equations (the first equation for the particle ve-
locity trace and the second for the pressure trace):

V =D+U and p=d+u . (5.36)

There are various conventions used. We will use the convention given by
Berkhout (1987), which is as follows.

1. The downgoing wave motion d has the same polarity as the downgoing
wave motion D, and the two are related by a scale factor given by the
acoustic impedance.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 5: Basic Properties of Waves 157

2. The upgoing wave motion u has the opposite polarity as the upgoing
wave motion U, and the same scale factor relates the two.

Thus, we have
d = ZD and u = −ZU . (5.37)
The solution of these equations yields the d’Alembert equations for the
downgoing and upgoing particle-velocity wave motion,

ẇ + p/Z ẇ − p/Z
D= and U= . (5.38)
2 2
Alternatively, we could solve for the downgoing and upgoing pressure wave
motion to obtain the corresponding d’Alembert equations,

Z ẇ + p −Z ẇ + p
d= u=
DOI:10.1190/1.9781560803737

and . (5.39)
2 2
A dual geophone/hydrophone sensor is required to obtain the downgoing
input signal and upgoing output signal. The d’Alembert equations are
used, so no traveling-wave assumption is required. Einstein deconvolution
removes all of the effects that are introduced by everything above the
receiver location. Einstein deconvolution not only eliminates the source
signature, but also eliminates the ghosts and reverberations due to the
layers above the receiver. Similar to conventional signature deconvolu-
tion, Einstein deconvolution requires knowledge of the input signal as
well as the output signal. The input wavelet is the downgoing wave at
the receiver and the output signal is the upgoing wave at the receiver.
The deconvolved signal is an estimate of the reflection response of the
deep system (that is, the earth layering below the receiver) (Robinson
and Treitel, 2008).
In addition, there is up-over-down (Einstein) deconvolution. Ideally,
the source and the receiver each can consist of multiple arrays of com-
ponents. The receiver must be buried at a depth below the depth of the
source. As shown in Figure 5.7, this deconvolution method will consist of
two steps. The first step is the conversion of the particle-velocity signal
and pressure signal into the downgoing wave and the upgoing wave at the
location of the receiver. In order to carry out this step, the d’Alembert
equations are used. The second step is the deconvolution of the upgoing
wave by the downgoing wave. This up-over-down (Einstein) deconvolution
removes the (unknown and not necessarily minimum-delay) source signa-
ture as well as the reverberations and ghosts due to the layers above the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

158 Basic Wave Analysis, Part 2: Raypath Analysis

DOI:10.1190/1.9781560803737

Figure 5.7. Up-over-down deconvolution.

receiver. The result of Einstein deconvolution is the unit-impulse reflection

response of the deep earth system (i.e., the layers below the receiver).
In other words, the resulting deconvolved seismogram is the unit-impulse
reflection response that would be produced as if there were no layers at
all above the buried receiver.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 6

Eikonal Equation and Ray Equation

“And perhaps, posterity will thank me for having shown

that the ancients did not know everything.”
DOI:10.1190/1.9781560803737

—Pierre de Fermat

Gradient and directional derivative

Let us turn to traveltime t. Traveltime represents the time required for
seismic energy emanating from the source point to reach a given depth
point (x, z). Traveltime has magnitude but no direction; thus, traveltime
may be represented by the scalar function t(x, z). The traveltime surface is
a plot of traveltime t against (x, z). The traveltime function can be depicted
by a surface plotted against horizontal coordinate x and vertical (depth)
coordinate z. An imaginary terrain, depicted by a topographic map, can be
used to visualize the traveltime configuration. Topographic maps provide
information about elevation of the surface above sea level, representing
elevation with contour lines (level lines).
The contour map of the terrain represents a scalar function. Each point
on a contour line has the same elevation. In other words, a contour line rep-
resents a horizontal slice through the land surface. A set of contour lines tells
you the shape of the land. For example, hills are represented by concentric
loops, whereas stream valleys are represented by V-shapes. Steep slopes
have closely spaced contour lines, while gentle slopes have very widely
spaced contour lines. The contour interval is the elevation difference
between adjacent contour lines. Now, consider this in terms of waves. A
wavefront is the locus of all points with a given traveltime. The contour
line t(x, z) = T represents the wavefront for traveltime T. We can imagine

159
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

160 Basic Wave Analysis, Part 2: Raypath Analysis

how the traveltime surface looks from a study of the contour lines (i.e., a
study of the wavefronts). The traveltime surface rises relatively steeply
where wavefronts are close to one another, and rises relatively gently
when they are far apart (Robinson and Treitel, 2008).
In two spatial dimensions, the traveltime function is similar to a hill,
whose height at the point r = (x, z) is t(x, z). The gradient of t(x, z) at a
point is a vector pointing in the direction of the steepest slope at that
point. The magnitude of the gradient vector provides the steepness of the
slope. The gradient depends only upon the partial derivatives of t(x, z) eval-
uated at the point in question. The gradient is the vector defined by the
equation
 
∂t ∂t ∂t ∂t
grad t = , = ex + ez . (6.1)
∂x ∂z ∂x ∂z
DOI:10.1190/1.9781560803737

Here, ex , ez are the unit vectors in the x, z directions, respectively. The

gradient operator,
 
∂ ∂ ∂ ∂
grad = , = ex + ez , (6.2)
∂x ∂z ∂x ∂z

is a generalization of the familiar differentiation operator. When the gradient

operator acts on a function t(x, z), it produces a vector, namely the gradient.
Given the function t(x) of a single variable, the chain rule is

dt dt dx
= . (6.3)
d s dx d s
The chain rule can be extended to functions of several variables. For
example, the chain rule for t(x, z) is

dt ∂t dx ∂t dz
= + . (6.4)
d s ∂x d s ∂z d s
The directional derivative is a generalization of partial derivative. The
partial derivatives indicate the rate of change of the traveltime in the direc-
tions of the axes. The directional derivative indicates the rate of change in
any specified direction. The traveltime t(x, z) depends upon both coordinate
axes x, z. We may hold z constant, and consider the curve that gives the vari-
ation of t with x alone. The slope ∂t/∂x of this curve is called the partial
derivative of t with respect to x. Similarly, the partial derivative ∂t/∂z is
defined. In general, we want to know the slope in some arbitrary direction
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 6: Eikonal Equation and Ray Equation 161

(Robinson and Treitel, 2008). Recall that a unit vector is a vector with
length, or magnitude, of 1. We can convert any vector into a unit vector
in the same direction by dividing the vector by its magnitude. Hence, the
unit vector for direction angle g (as measured from the horizontal) can be
represented by w = (sin g, cos g). If we let s represent distance in the direc-
tion of the vector w, then
dx dz
= sin g, = cos g . (6.5)
ds ds
The slope of t(x, z) in the direction of this unit vector is called the direc-
tional derivative. The directional derivative is the weighted average of the
two partial derivatives, the weights being the component of the unit direc-
tional vector. Thus, the directional derivative is given by the chain rule
dt ∂t dx ∂t dz ∂t ∂t
= + = sin g + cos g . (6.6)
d s ∂x d s ∂z d s ∂x ∂z
DOI:10.1190/1.9781560803737

This shows that the directional derivative is the dot product

 
dt ∂t ∂t
= , · ( sin g, cos g) = grad t · w . (6.7)
ds ∂x ∂z
The first vector in the dot product is the gradient of traveltime. The second
vector is the unit vector in the desired direction. The partial derivatives are
special cases of the directional derivatives. For instance, ∂t/∂x is the direc-
tional derivative in the x direction.
The directional derivative may be written as
dt
= grad t · w = grad t · w cos a , (6.8)
ds
where a is the angle between the gradient and the directional vector. The
maximum value of the directional derivative is obtained when the direction
vector points in the same direction as the gradient, that is, when a = 0 and
cos a = 1. This maximum value is equal to the magnitude of the gradient.
In other words, the gradient gives the direction of maximum slope.
Suppose the directional vector points along a contour line. Because the
contour line is level, the directional derivative in the direction of a
contour line must be zero. That is, cos a must be zero, so a is 908; thus,
the gradient vector is orthogonal (or perpendicular) to the contour line
(Robinson and Treitel, 2008).
Now, let us introduce the important concept of a flow line. A vector field
is a rule that assigns a vector to each point (x, z). An important case is the
vector field defined by the gradient. In visualizing this vector field, we
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

162 Basic Wave Analysis, Part 2: Raypath Analysis

imagine that the vector grad t is attached to each point. Thus, the vector field
associates a direction and a magnitude to each point. If a hypothetical par-
ticle moves in such a manner that its direction at any point coincides with the
direction of the gradient at that point, then the curve traces out a so-called
flow line. Because the direction of the flow line is uniquely determined by
the vector field, it is impossible to have two different directions at the
same point. Therefore, it is impossible to have two flow lines cross each
other. The contour lines and the associated flow lines are important tools
in the understanding of seismic wave motion.

Principle of least time

As an ant goes up a mound, it experiences the rate of change of elevation.
The gradient is the vector (because it has both magnitude and direction) that
points in the steepest direction (i.e., the direction in which the rate of change is
DOI:10.1190/1.9781560803737

greatest). The magnitude of the gradient vector is the rate of change of the
elevation along this path. Thus, if the ant continually follows the gradient,
it will take the steepest path to the top of the mound. Assume that the
speed of the ant at any point on the mound is the same in all directions (an
isotropic medium). Thus, by taking the steepest path, the ant will reach the
top in the shortest time. In other words, the path that follows successive gra-
dient vectors represents the least-time path for such a climber.
Next, this reasoning can be stated in terms of mathematics. Let an arbi-
trary path between two wavefronts be given. This arbitrary path is called the
test path. First, we would like to determine what happens between two
closely spaced wavefronts. Because the wavefronts are so close together,
we may consider the parts of them within a small region to be two parallel
straight lines. The test path would be a straight line between the two wave-
fronts, and the flow line would be a straight line orthogonal to both wave-
fronts. If u is the angle between the test path and the flow line, if ds is
equal to the length of the flow line, and if d s is the length of the test path,
then s = ds/ cos u. As usual, we assume an isotropic medium, so that the
velocity at any point is the same in all directions. Thus, the time it takes
to transverse the flow line is dt = ds/v. Likewise, the time it required to
transverse the test path is
ds 1
dt = = ds . (6.9)
v v cos u
The traveltime along the flow line between the two wavefronts is

1
t= ds . (6.10)
v
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 6: Eikonal Equation and Ray Equation 163

It is understood, of course, that the path of integration is along the flow line.
The traveltime along the test path is

1
t= ds . (6.11)
v cos u
It is understood that the path of integration is along the test path. Because
cos u at any point is always less than or equal to one, it follows that t ≤ t.
Thus, the traveltime along the flow line is less than the traveltime along
any test path, except the flow line itself. The flow line, that is, the line
whose direction at any point coincides with the direction of the gradient,
is the least-time path (Robinson and Treitel, 2008).

Eikonal equation
According to the dictionary, the word slowness means the quality of
DOI:10.1190/1.9781560803737

moving at a low speed. The word slowness also is used to denote a quantity
introduced in seismology which is the reciprocal of velocity. The traveltime
of a wave is the distance that the wave travels times the slowness of the
medium. Thus,
1
slowness = . (6.12)
velocity
Slowness indicates the amount of time required for an object to travel a
given distance. If we would like to rewrite equation (6.12) so that is has a
more pleasing quality, we can use the synonym swiftness for the word speed:

1
slowness = . (6.13)
swiftness
Let v be velocity, n be slowness, Dx be the increment of distance, and Dt
be the increment of time. Then, velocity equation is

Dx = v Dt . (6.14)

The slowness equation is

Dt = n Dx . (6.15)

In the limit Dt  0, we have

dx dt
v= and n= . (6.16)
dt dx
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

164 Basic Wave Analysis, Part 2: Raypath Analysis

Figure 6.1. Right

triangle with sides a, b, c
and altitude s.

Next, consider the triangle shown in Figure 6.1, which is a right triangle
with legs a and b. The hypotenuse is c, and the altitude to the hypotenuse is s.
The angle between hypotenuse c and leg a is u. The angle between leg b and
altitude s is also u. The altitude s can be expressed in terms of the two legs as
 
b
DOI:10.1190/1.9781560803737

s = a cos u = a . (6.17)
c
Employing the Pythagorean theorem,
a2 + b2 = c2 . (6.18)
we obtain
ab
s = √ . (6.19)
a2 + b2
We now wish to establish the so-called reciprocal Pythagorean theorem
a−2 + b−2 = s−2 . (6.20)
In order to do so, we take the square of the reciprocal of altitude s and obtain
the desired result:
 2  2  2
1 a2 + b2 a2 b2 1 1
= 2
= 2
+ 2
= + . (6.21)
s (a b) (a b) (a b) a b
Figure 6.2 depicts the wavefront of a plane wave at a given instant of
time and the wavefront at a subsequent time. The time increment is Dt. In
time increment Dt, the wavefront moves a horizontal distance Dx. In time
increment Dt, the wavefront moves a vertical distance Dz. However, the
bulk of the energy moves along the ray. Let s measure the path length
along a given ray. The raypath is orthogonal to the wavefront. In time incre-
ment Dt, the bulk of the energy moves distance Ds traveled along the
raypath.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 6: Eikonal Equation and Ray Equation 165

Figure 6.2.
Wavefronts and
raypath.
DOI:10.1190/1.9781560803737

The reciprocal Pythagorean theorem (for legs Dx and Dz and

altitude Ds) is
 2  2  2
1 1 1
+ = . (6.22)
Dx Dz Ds

If we multiply this equation by (Dt)2 , we obtain

 2  2  2
Dt Dt Dt
+ = . (6.23)
Dx Dz Ds

In the limit as Dt  0, we obtain the eikonal equation

 2  2  2
∂t ∂t ∂t
+ = . (6.24)
∂x ∂z ∂s

Now, consider these statements:

† horizontal direction represents apparent slowness ∂t/∂x,

† vertical direction represents apparent slowness ∂t/∂z, and
† raypath direction represents slowness dt/ds.

Therefore, the eikonal equation can be interpreted as: the sum of squares
of the apparent slownesses in the coordinate directions equals the square
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

166 Basic Wave Analysis, Part 2: Raypath Analysis

of the actual slowness in the raypath direction. Because the actual slowness
in the raypath direction is the reciprocal of seismic velocity v, the eikonal
equation may be written as
 2  2  2
∂t ∂t 1
+ = . (6.25)
∂x ∂z v

The given wavefront is at time t, and the new wavefront is at time t + dt. The
traveltime along the ray is dt. If s measures path length along the given ray,
then the travel distance in time dt is ds. The increments dt and ds are related
by the slowness, that is dt = n ds. Thus, the directional derivative in
the direction of the raypath is equal to the slowness, that is, n = dt/ds.
The directional derivative may be written in terms of its components as

dt ∂t dx ∂t dy dr
= + = grad t · . (6.26)
DOI:10.1190/1.9781560803737

ds ∂x ds ∂y ds ds

Because dr/ds = u, it follows that

dt
= grad t · u or n = grad t · u . (6.27)
ds
The Fermat requirement that the unit tangent u have the same direction
as the vector grad t(x, z) means that these two vectors are related by the
scalar n(x, z). Thus, the relationship can be written as

grad t(x, z) = n(x, y) u(x, z) or ∇t = n u . (6.28)

This equation is the eikonal equation in vector form (Robinson and Treitel,
2008). Because u is a unit vector in the same direction as the gradient, it
follows that n = |grad t|. In other words, the slowness is equal to the
magnitude of the gradient of traveltime. If we take the square of each
side, we obtain the eikonal equation in scalar form
 2  2
∂t ∂t
n =
2
+ . (6.29)
∂x ∂z

The eikonal equation says that the magnitude of the gradient of the trav-
eltime is equal to the slowness. Note that the flow lines of the vector field
n(x, z) u(x, z) are the same as those of grad t(x, z), because only the direction
of the vector grad t(x, z) at any point is relevant in determining the direction.
The eikonal equation says that, at any point, the gradient of the wavefront is
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 6: Eikonal Equation and Ray Equation 167

equal to the slowness n times the unit tangent to the ray. Therefore, the
gradient and the tangent both go in the same direction. Because the gradient
is orthogonal to the wavefront, and the tangent is along the ray, it follows
that the ray is orthogonal to the wavefront, and
 
∂t ∂t
, = n (cos u, sin u) . (6.30)
∂x ∂z

The left side of equation (6.30) involves the wavefront and the right side
involves the ray. As we mentioned in reference to expression (6.13), velocity
is called swiftness. The reciprocal of velocity is slowness. In equation (6.28),
the function t(x, z) is the traveltime from the source to the point with the
coordinates(x, z). The slowness (or reciprocal velocity) at that point is
n(x, y) = 1/v(x, y). The apparent swiftnesses, each along its coordinate
direction, are, respectively, (∂x/∂t, ∂z/∂t). Thus, the apparent slownesses,
DOI:10.1190/1.9781560803737

each along its coordinate directions, are, respectively (∂t/∂x, ∂t/∂z). The
actual swiftness along the raypath direction is v = ds/dt. The actual slow-
ness along the raypath direction is n = dt/ds.
The eikonal equation describes the traveltime propagation in an isotro-
pic medium. The eikonal equation does not fix the direction of the ray; that
follows from initial conditions. To obtain a well-posed initial value problem,
it is necessary to know the velocity function v(x, y) at all points. Moreover,
as an initial condition, the source or some particular wavefront must be
specified. Furthermore, a choice must be made between one of the two
branches of the solutions (i.e., either the wave going from the source or
the wave going to the source). Then, the eikonal equation yields the travel-
time field t(x, y) in the heterogeneous medium as required for migration and
other seismic processing needs. For more information on the computation of
traveltimes for migration, see Moser (1991, 1994).
The eikonal equation is a restatement of Fermat’s principle of least time.
In other words, a raypath must be a flow line. A flow line is orthogonal to all
of the wavefronts. The eikonal equation is the fundamental equation that
connects the ray (which represents the path) to the wavefront (which
spreads out on both sides of the path). In effect, the wavefront feels the
effects of points away from the raypath. The eikonal equation makes a
traveling wave (as envisaged by Huygens) fundamentally different from a
traveling particle (as envisaged by Newton). Furthermore, William
Rowan Hamilton (1805–1865) bases his work on the principle of least
action, which is a more general formulation of the principle of least time.
Hamilton is credited with proposing the wave-particle duality, which pro-
vides the mathematical foundation of quantum mechanics.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

168 Basic Wave Analysis, Part 2: Raypath Analysis

To summarize, we have defined the gradient of the traveltime surface

and defined the raypath curve. We have established that the ray direction
is always perpendicular to the traveltime surface. A wave as it travels
must follow the path of least time. As an example, consider a point
source. The source would be at the lowest point of a basin. The wavefronts
are similar to contour lines for which there are no maxima and minima away
from the source. The traveltime is monotonically increasing along the ray.
The eikonal equation is a non-linear partial differential equation for which
only very few analytic solutions are known. Therefore, there is a need for
eikonal equation solvers, as testified by the extensive publications describ-
ing numerical algorithms.
Rays and wavefronts are interesting to study. In certain cases, rays can
touch each other (caustics). In other cases, rays can cross each other (mul-
tiplication of traveltime). Now, suppose that the height of the hill is
measured in time, and select a point on a contour line. In what direction
DOI:10.1190/1.9781560803737

will the ray point? Suppose the ray points in the direction of the contour
line. In other words, suppose that the raypath lies directly upon the wave-
front. As the wave travels a certain distance along this ray, it takes time.
But all time is the same along the wavefront. Thus, a wave cannot travel
along a wavefront. It follows that a ray must point away from a wavefront.
Suppose now that a ray points away from the wavefront. The wave wants to
take the least time to travel to the new wavefront. By isotropy, the wave ve-
locity is the same in all directions. Because the traveltime is velocity multi-
plied by distance, the wave wants to take the raypath that goes the shortest
distance. The shortest distance is along the path with no component along
the wavefront; that is, the shortest distance is along the normal to the wave-
front. In other words, the raypath must be orthogonal (i.e., at right angles) to
the wavefront. The ray’s unit tangent vector u must be orthogonal to the
wavefront. By definition, the gradient is a vector that points in the direction
orthogonal to the wavefront. Thus, the ray’s unit tangent vector u and the
gradient grad t of the wavefront must point in the same direction (Robinson
and Treitel, 2008).

Ray equation
The various forms of the wave equation permit analytic solutions only
for a very limited class of velocity models. The power of modern computers
makes possible numerical solutions that could not be done previously. Ray
tracing is a method for calculating the path of waves or particles through a
system with regions of varying propagation velocity, absorption, and reflect-
ing surfaces. Under these circumstances, the ray path may bend, change
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 6: Eikonal Equation and Ray Equation 169

direction, or reflect off of surfaces. Ray theory allows interference of

separate ray arrivals. Interference is not incompatible with ray theory.
Ray tracing solves the problem by iteratively advancing rays through the
medium by discrete amounts. Ray tracing often relies on approximate
solutions of the wave equation. The approximations are valid as long as
the waves propagate through and around objects whose dimensions are
much greater than the wavelength. Ray theory does not describe diffraction
at surfaces with structural details smaller than the wavelength and
around objects which are smaller than the wavelength. Ray theory in its
simplest form (standard ray theory) also breaks down in the neighborhood
of surfaces where the ray field is singular, such as caustics (see Červený,
2001).
A moving single ray must take an extremal path (such as the least-time
path). How does the ray know how to do this? For example, consider a
seismic wave traveling through a medium whose slowness n increases in
DOI:10.1190/1.9781560803737

the direction of travel. Because the raypath is parallel to the gradient of

the slowness, it undergoes no bending. However, if the contour lines of
slowness are at an angle to the ray, then the ray will bend, even though
the slowness at each point along the path is identical to what it was
before, when there was no bending. This shows that the path of a ray
cannot be explained solely in terms of the value of the slowness on the
path. We also must consider the slowness along neighboring paths, i.e.,
along paths not taken.
Huygens provides the classical explanation—light does not propagate in
the form of a single ray. According to Huygens, light propagates as a wave-
front possessing transverse width. If a small section of a propagating wave-
front encounters the same value of slowness along its width, then the ray will
not bend. If a small section of a propagating wavefront encounters the differ-
ent values of slowness along its width, then the ray will bend. The amount of
bending depends upon the gradient of the slowness. The wavefront propa-
gates more rapidly on the side where the slowness is low (i.e., where the ve-
locity is high) than on the side where the slowness is high. As a result, the
wavefront naturally turns in the direction of increasing slowness (Robinson
and Treitel, 2008).
In this book, the position vector

r = (x, z) (6.31)

always represents a point on a specific raypath, and not any arbitrary point
in space. As time increases, r traces out the particular raypath in question.
The ray, in general, will follow a curved path. The equation for the unit
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

170 Basic Wave Analysis, Part 2: Raypath Analysis

tangent vector is
 
dr dx dz
u= = , . (6.32)
ds ds ds
The vector n u (called slowness vector) is tangent to this curved raypath.
We recall that the eikonal equation is
n u = ∇t . (6.33)
The eikonal equation says that the raypath is a flow line of traveltime. In
other words, the seismic ray at any given point follows the direction of
the gradient of the traveltime field t(r).
The slowness n(x, z) is a scalar function that depends upon both coordi-
nates x and z. However, we may hold z constant, and consider the curve that
gives the variation of slowness with x alone. The slope of this curve ∂n/∂x is
called the partial derivative of slowness with respect to x. Similarly, the
DOI:10.1190/1.9781560803737

partial derivative ∂n/∂z is defined. The gradient of the slowness surface is

the vector with these partial derivatives as components; that is,
 
∂n ∂n
grad n = ∇n = , . (6.34)
∂x ∂z
Now, we will derive the so-called ray equation. We know that the gra-
dient of the traveltime is
 
∂t ∂t
grad t = ∇t = , . (6.35)
∂x ∂z
Because slowness n = dt/ds, it follows that the derivative of ∇t along
the ray is
 
d dt
(∇t) = ∇ = ∇n . (6.36)
ds ds
We take the derivative of the eikonal equation with respect to path length
s and obtain
d d
(n u) = (∇t) . (6.37)
ds ds
The right side of the equation (6.37) is the gradient of the slowness, so we
obtain the ray equation
d
(n u) = ∇n . (6.38)
ds
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 6: Eikonal Equation and Ray Equation 171

The ray equation indicates how the slowness vector n u changes along the
curved raypath. Almost tautologically, the ray equation says that the rate
of change of the slowness vector is equal to the gradient of the slowness.
The ray equation does not say the raypath is a flow line of slowness. The
ray equation says the rate of change of the raypath is a flow line of slowness.
The flow line, that is, the line whose direction at any point coincides with the
direction of the gradient, is the least-time path.

Diving waves
Eventually, raypaths will return to the surface whenever the velocity
increases with depth. Such waves are called diving waves or turning
waves. Complex overburden always presents a problem in seismic explora-
tion. Full waveform inversion (FWI) can be used as a method for determin-
ing an accurate description of the near surface and thereby preventing
DOI:10.1190/1.9781560803737

distortions in the imaging of the deep subsurface. The diving waves

can be modeled by ray tracing with the use of initial velocity models at
various locations throughout the survey. These velocity models are
updated iteratively in order to reduce the misfit between the observed and
modeled seismic data. The maximum depth of penetration, and hence the
FWI update, varies significantly depending on the local velocity and geolo-
gic environment. Full waveform inversion is not always used for the near
surface, yet it is being used increasingly at depth.
Horizontal velocity layering presents symmetry. The angle of emer-
gence is the same as the angle u0 of incidence. Suppose we have a two-
dimensional stratified medium, with horizontal coordinate x and depth
coordinate z. First, let us consider a velocity function v(z) that depends
upon depth only. We further assume that v(z) increases with depth
(i.e., ∂v/∂z . 0. Thus, the slowness n(z) = 1/v(z) decreases with depth;
i.e., ∂n/∂z , 0. It follows that the gradient of the slowness is
   
∂n ∂n ∂n
∇n = , = 0, . (6.39)
∂x ∂z ∂z
The gradient points directly uphill; that is, it points in the direction of stee-
pest ascent. In this case, we see that the gradient of slowness points verti-
cally toward the surface of the earth.
Now, consider the special case in which the velocity function
v(z) = v0 + az (6.40)
increases linearly with depth z. Specifically, the velocity function is speci-
fied completely by the two constants v0 (the velocity at the surface) and a
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

172 Basic Wave Analysis, Part 2: Raypath Analysis

(the slope of the linear velocity function). It follows that the slowness
1 1
n(z) = = (6.41)
v(z) v0 + az

decreases with depth. Because

∂n ∂v−1 ∂v a
= = −v−2 = − 2 , (6.42)
∂z ∂z ∂z v

it follows that the gradient of the slowness is

  
∂n ∂n a
∇n = , = 0, − 2 , (6.43)
∂x ∂z v

which points vertically toward the surface of the earth. The ray will follow a
DOI:10.1190/1.9781560803737

curved path. The equation for the unit tangent vector is

 
dr dx dz
u= = , = (sin u, cos u) . (6.44)
ds ds ds

The ray equation becomes

d  a
(n(sin u, cos u)) = 0, − 2 . (6.45)
ds v
The first component of the ray equation is
d
(n sin u) = 0 . (6.46)
ds
Thus, n sin u is a constant. This constant is Snell’s parameter p. Thus, the
first component of the ray equation is the none other than the familiar
Snell equation
sin u
p= = constant . (6.47)
v
At the shot point, the ray makes the angle u0 with the vertical. Choose a
specific point (x, z) on the raypath. Designate the angle at this point by u.
Snell’s equation says that
sin u0 sin u
p= = . (6.48)
v0 v
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 6: Eikonal Equation and Ray Equation 173

The second component of the ray equation is

d ∂n
(n cos u) = . (6.49)
ds ∂z
If we take the derivative, the left side becomes

du ∂n dz du ∂n
−n sin u + cos u = −n sin u + cos u cos u . (6.50)
ds ∂z ds ds ∂z
The second component of the ray equation becomes
du ∂n ∂n
−n sin u + cos2 u =
ds ∂z ∂z
or
d u ∂n
DOI:10.1190/1.9781560803737

−n sin u = (1 − cos2 u)
ds ∂z
so
du ∂n
= −v sin u . (6.51)
ds ∂z
From equations (6.51), we have

du  a  a
= −v sin u − 2 = sin u = ap = constant . (6.52)
ds v v
The reciprocal of this equation (6.52) produces

ds 1
= ; r. (6.53)
d u ap

Equation (6.53) is the differential equation of a circle with radius r.

In conclusion, the solution of the ray equation tells us that the raypath is
an arc of a circle of radius r. Let a particle of energy start at the origin
(0, 0) where the path makes an initial angle u0 with the vertical. The particle
travels along a circular arc to point (x, z) where the path makes an angle u
with the vertical. It can be shown that the center of this circle is the point
C with coordinates
 2
(1 − p2 v20 ) v0
C= ,− . (6.54)
ap a
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

174 Basic Wave Analysis, Part 2: Raypath Analysis

Figure 6.3. Diving

waves. The arcs of
each circle represent
raypaths with
different initial
angles.

Note that the z-coordinate of the center is independent of the parameter

p. More particularly, it does not depend upon u0 . As a result, the family of
rays with the given velocity function is composed of circular paths that pass
through the origin and that have their centers on the line z = −v0 /a (see
Figure 6.3). The circular raypath starts at the origin (which is the shot
point). The tangent to the circle at the shot point makes an angle u0 with
the positive z-axis.
DOI:10.1190/1.9781560803737

Fresnel zone
A traveling sine wave has the form
u(x, t) = A cos (kx − vt) = A cos k(x − vt) , (6.55)
where v is angular frequency, t is time, k is angular wavenumber, and x is
distance. The temporal frequency is f = v/2p and the period is T = 1/f .
The wavelength is l = 2p/k. During one period T of time, the wave
travels one wavelength l of distance. Thus, the velocity v is equal to wave-
length over period; that is, v = l/T. Lateral resolution refers to how close
two reflecting points can be situated horizontally so that the two points do
not blend together, but instead appear as two separate points.
As shown in Figure 6.4a, the sum of two waves in step has complete
reinforcement. The sum of two waves out of step by l/4 has partial
reinforcement (Figure 6.4b), and the sum of two waves out of step by
l/2 has complete cancellation (Figure 6.4c).
Figure 6.5 depicts a spherical wavefront incident on a horizontal reflec-
tor. Each point on this reflector represents a point diffractor. We assume that
the source and receiver are coincident at S on the earth’s surface. The
primary reflection event from subsurface point P has two-way traveltime
t = 2z/v, so z = vt/2. Now, let the incident wavefront advance in depth
by the amount l/4. Energy from subsurface location B, or A, will reach
the receiver at time t1 = 2y/v, so y = vt1 /2. The energy from all of the
points within the reflecting disk with radius r will arrive sometime
between t and t1 . The total energy arriving within the time interval t1 − t
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 6: Eikonal Equation and Ray Equation 175

a)

Figure 6.4. (a)

The sum of two
b) waves in phase; (b)
the sum of two waves
90o out-of-phase
DOI:10.1190/1.9781560803737

(i.e., displacement of
l/4); and (c) the
sum of two waves
180o out-of-phase
(i.e., displacement of
l/2 = l/4 + l/4).
c)

interferes constructively. The reflecting disk AB is called the first Fresnel

zone. Two reflecting points that fall within this zone generally are
considered indistinguishable as observed from the earth’s surface.
The distance r is the radius of the first Fresnel zone. Recall that v = l/T,
so l = vT = v/f . We have

vt l v
z= and y=z+ =z+ . (6.56)
2 4 4f
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

176 Basic Wave Analysis, Part 2: Raypath Analysis

Figure 6.5.
Two wavefronts
out of step by
l/4.

Thus,
DOI:10.1190/1.9781560803737

 2
v 2zv v2
y = z+
2
= z2 + + . (6.57)
4f 4f 16f 2

For large f, the last term on the right becomes negligible, so we may
write

2zv zv
y2 ≈ z2 + = z2 + . (6.58)
4f 2f

We note that SPB is a right triangle. Thus,

2zv zv
r 2 = y2 − z2 ≈ z2 + − z2 = . (6.59)
4f 2f

Recall that z = vt/2. Thus,

 v 2 t
r2 ≈ . (6.60)
2 f

Thus,

v t
r≈ . (6.61)
2 f

The Fresnel zone offers a generalization for finite frequencies of the

geometric ray. For a given frequency f and a ray between source S and
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 6: Eikonal Equation and Ray Equation 177

the coincident receiver point, the Fresnel zone contains all points around
the geometric ray that contribute to constructive interference at the receiv-
er end of the ray. For decreasing frequency it expands, for increasing fre-
quency it shrinks. In the limiting case of frequency going to infinity, the
Fresnel zone shrinks to the classical geometric ray. Because all points
within the Fresnel zone contribute to the wave observed at the receiver
point, it provides a measure of the resolution along the ray.
See Hubral et al. (1993) and Moser and Červený (2007) for detailed dis-
cussions of the construction of the Fresnel zone in isotropic and generally
anisotropic models.

Fermat’s derivation of Snell’s law

An initial approach to the study of seismic wave propagation is the
method of approximation commonly called geometric seismology. It is
DOI:10.1190/1.9781560803737

the counterpart of geometric optics. The basic element in geometrical

optics is the light ray. It is a hypothetical concept that indicates the direction
of the propagation of light at any point in space. This concept originates with
the Pythagorean notion of visual rays. Geometrical optics consists of a set of
rules that determines the paths followed by light rays. In any uniform
medium, the rays travel in straight lines. An ideal light ray has zero
width. The light emitted from a small localized source can be represented
as a collection of rays pointing radially outward from an idealized point
source. Similarly, light flowing with uniform intensity through space can
be represented by a collection of parallel rays (e.g., light from a star or a
laser). The formation of a sharp shadow when an object is illuminated by
a parallel beam of light can be explained easily by tracing the paths of the
rays that are not blocked by the object. Importantly, in this conception,
aspects of the ray of the light, such as its phase, are not significant; the
central issue is the ray path. The method used to create ray diagrams of
systems is called ray tracing, and it can be used to estimate the path of
light in a medium. The phenomenon of light is not considered as a wave
in the ray theory of light; therefore, wave-related phenomena, such as scat-
tering, diffraction, and interference, cannot be described using the ray
model. Some specific types of light rays are named after the places they
occur: a light ray which falls on an object is known as the incident ray; a
light ray reflected by an object is a reflected ray; a light ray refracted by
an object is a refracted ray; and a light ray which bends around an obstacle
is a diffracted ray.
At first glance, the words ray and beam may seem to mean the same
thing, but these terms have two different meanings. It is easy to think of a
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

178 Basic Wave Analysis, Part 2: Raypath Analysis

narrow beam of light as being composed of a collection of parallel arrows—

a bundle of rays. If the beam of light (1) moves from one medium to another,
(2) reflects off surfaces, (3) disperses, or (4) comes to a focus, then the
bundle of rays traces the beam’s progress in a simple geometrical manner.
A beam is a thin projection of particles or waves. Because a beam has a
finite width, it can be observed physically. Because a ray has zero width,
it is a concept that cannot be observed physically. Wave properties, such
as wavelength, amplitude, and phase, are omitted when a geometrical ray
is discussed. Any property of the waves or particles can be described in
terms of a beam (or using dynamic ray theory, see Červený, 2001).
A beam is a narrow projection of a set of waves or particles. Beams are
used in various fields and applications. The two main types of beams are
light (or electromagnetic wave) beams and particle beams. A laser produces
a light beam. More precisely, a laser is any of a class of devices that produces
an intense beam of light of a very pure single color. (“Laser” is actually an
DOI:10.1190/1.9781560803737

acronym for “light amplification by stimulated emission of radiation.”) In

fact, a light beam from a laser may be intense enough to vaporize extremely
hard and heat-resistant materials. Also, whereas natural light moves in
beams of different wavelengths, a laser in its ideal form will have only
one wavelength.
A particle beam is a concentrated stream of charged or neutral particles,
such as electrons, neutrons, or protons, generated for studying such things as
particle interactions, nuclear structure, or crystal structure. In most cases, the
stream of particles is moving at or near the speed of light. Charged particles,
or ions, are easy to accelerate using basic magnetic repulsion, as in the Large
Hadron Collider. Because these ions repel one another, the unrestrained
beam would spread enormously. For that reason, the particle beam needs
a neutralizing chamber in order to maintain a focused beam of neutral
particles.
Geometric seismology is usable when the wavelengths involved are
small compared to the scale of the dimensions of the geologic structure.
A ray is a portion of a line which starts at a certain point and goes off in a
particular direction. Rays of light are narrow beams of light. Seismic rays
are the counterparts of the rays of light, but physically seismic rays are
never narrow beams. Beams can be defined as ray bundles that generally
depend on two parameters, i.e., two take-off angles or two independent coor-
dinates at an initial surface. Gaussian beams are characterized by Gaussian
dependence of the amplitude as a function of the beam parameters, and they
are being used frequently in seismology and seismic exploration because of
their regular amplitude over the ray field (thereby resolving the singularities
of standard ray theory).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 6: Eikonal Equation and Ray Equation 179

An isotropic material is one that has identical values of a property in

all directions. Glass and metals are examples of isotropic materials. In
contrast, wood is an anisotropic material because its properties are different
parallel and perpendicular to the grain. A layered rock such as slate is ani-
sotropic. In an isotopic rock, the wave velocity is the same in all directions.
It is Christiaan Huygens who first differentiates between the properties of
wave propagation in isotopic rock as opposed to anisotropic rock, based
on experiments with double refraction in Icelandic spar. In particular,
he shows that, in isotropic material, the rays are at right angles to the
wavefronts. In this book, we assume that all materials are isotropic, so the
wavefronts and rays are always perpendicular to each other.
In a homogeneous medium, the seismic waves travel in straight lines;
that is, the seismic rays are straight. When a seismic ray hits an interface,
it is reflected in such a way that the angle of the incident ray with the inter-
face normal is equal to the angle of the reflected ray with the interface
DOI:10.1190/1.9781560803737

normal. Figure 6.6 illustrates the law of reflection

u1 = u1′ . (6.62)

Here, u1 is the angle of incidence and u1′ is the angle of reflection.

A lawyer by day and a mathematician by night, Fermat was a man of
obvious genius who never published his findings during his life. Fortunately,
his discoveries were saved by the mathematicians he corresponded with in
Paris. Although the inventors of calculus are said to be Newton and Leibniz,
Fermat certainly played a part. Fermat found a method of finding maxima

Figure 6.6. Reflection at an interface.

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

180 Basic Wave Analysis, Part 2: Raypath Analysis

Figure 6.7. Geometric construction to show that u1 = u1′ by Fermat’s principle of

DOI:10.1190/1.9781560803737

least time.

and minima which students today would recognize as setting the derivative
equal to zero. Here we give Fermat’s method.
Fermat’s principle states that, out of all possible paths that the wave
might take to go from one point to another, the wave takes the path that
requires the shortest time. Let us now show that Fermat’s principle contains
both the law of straight-line propagation and the law of reflection. Figure 6.7
shows two points, A and B, and a straight interface. On the other side of the
interface, we construct an artificial point B′ , which is the same distance
below the interface as point B is above the interface. Consider now some
path ADB. The sum of the distances AD + DB is proportional to the time
of propagation, because the wave travels with constant velocity in the
medium. This sum also is equal to AD + DB′ . Thus, the problem
becomes: When is the sum of two such lengths the least? By inspection,
we see that the answer is obtained when the line goes through the point C
as a straight line from A to B′ . In such a case, it follows from the geometry
that u1 = u1′ . The statement that the angle of incidence is equal to the angle
of reflection is equivalent to the statement that the wave travels from A to the
interface and back to B in such a way that it takes the least possible time.
Now, let us show that the principle of least time gives Snell’s law of
refraction. In Figure 6.8, a straight interface separates two media, the
upper one with wave velocity v1 and the lower one with wave velocity v2 .
For definiteness, let us assume v2 is greater than v1 . The problem is to go
from A to B in the shortest time. If we proceed by means of a straight
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 6: Eikonal Equation and Ray Equation 181

Normal to interface

= angle of incidence

= velocity in upper medium

Interface

= slightly longer-me path

= least-me path
= velocity in lower medium = angle of refracon =

Figure 6.8. Geometric construction to show that Snell’s law results from Fermat’s
DOI:10.1190/1.9781560803737

principle of least time.

line, then we would travel a certain distance in each medium. The wave
travels more slowly in the first medium than in the second medium. Thus,
it is advantageous to travel a lesser distance in the first medium and a
greater distance in the second medium. Let us assume that the path ACB
produces the shortest time of all possible paths. This means that any other
path will take a longer time. However, for paths close to the optimum
path, there is essentially no change in time. (Of course, there is a positive
infinitesimal change of the second order for displacements in either
direction from C.) In regard to changes of the first order, we consider a
nearby point D, and compare the new path from A to B through D with
the old path through C.
Provided that the distance DC is small, the difference in time between
the two paths should be nearly zero. For a small distance DC, lines AC
and AD will be nearly parallel (as shown in Figure 6.8). Enlarging
Figure 6.8 in the region of line DC, we can specify additional details
using our new Figure 6.9. If we draw the perpendicular DE, we see that
the least-time path is shortened by the amount CE. Thus, we gain a
certain amount of time by not having to go that extra distance. On the
other hand, by drawing the corresponding perpendicular CF, we see that
the least-time path is lengthened by the amount DF. We lose a certain
amount of time by having to go this extra distance. The time we gain and
the time we lose must be equal because, in this first-order approximation,
there is to be no change in time.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

182 Basic Wave Analysis, Part 2: Raypath Analysis

= angle of incidence =
= velocity in upper medium

Normal to interface
Interface

= angle of refracon =
= least-me path
= velocity in lower medium
DOI:10.1190/1.9781560803737

Figure 6.9. Enlargement of Figure 6.8 in the region of line DC.

The angle of incidence is

u1 = /ACN = /FCD . (6.63)

Thus, segment FD has length

FD = CD sin u1 . (6.64)

The time required to traverse segment FD is

FD CD sin u1
t1 = (slowness in medium 1) FD = = . (6.65)
v1 v1
The angle of refraction is

u2 = /N ′ CE = /EDC . (6.66)

Thus, segment CE has length

CE = CD sin u2 . (6.67)

The time required to traverse segment CE is

CE CD sin u2
t2 = (slowness in medium 2) CE = = . (6.68)
v2 v2
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 6: Eikonal Equation and Ray Equation 183

Following this argument, if we set t1 = t2 , we obtain

CD sin u1 CD sin u2
= . (6.69)
v1 v2
As a result, we obtain

sin u1 sin u2
= . (6.70)
v1 v2
This result, known as Snell’s law of refraction, states that, for the least-time
path, the angles should be such that the ratio of their sines is equal to the ratio
of the velocities.
An interesting consequence of the principle of least time is the principle
of reciprocity. If we have determined the least-time path in one direction,
then (in the isotropic case in which the wave velocity is the same in any
DOI:10.1190/1.9781560803737

direction) the same path will have the least time in the opposite direction.
Thus, if we determine the wave propagation path in one direction, then
the opposite path is also a propagation path.

Huygens’ derivation of Snell’s law

In 1672, Newton states that his experiments prove that light is made of
particles and not waves. He argues that the geometric nature of the laws of
reflection and refraction could only be explained if light is made of particles,
because waves do not tend to travel in straight lines. Robert Hooke’s Cri-
tique of Newton’s Theory, delivered in 1672, reads: “The motion of light
in a uniform medium, in which it is generated, is propagated by simple
and uniform pulses or waves, which are at right angles with the line of direc-
tion. But falling obliquely on the refracting medium, it receives another
impression or motion, which disturbs the former motion, somewhat like
the vibration of a string: and that, which was before a line, now becomes
a triangular superficies, in which the pulse is not propagated at right
angles with its (original) line of direction, but askew, as I have more at
large explained in my Micrographia.”
In 1678, Christiaan Huygens claims to have disproven Newton’s theory
by showing that the laws of reflection and refraction can be derived from his
wave theory of light. Instead of considering rays as Fermat did, Huygens
uses wavefronts. As such, we can explore Huygens’ derivation of Snell’s
law. At a very large distance from a point source, the curvature of the wave-
front is negligible in a limited viewing range. Thus, such waves appear as
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

184 Basic Wave Analysis, Part 2: Raypath Analysis

DOI:10.1190/1.9781560803737

Figure 6.10. A plane wave incident on an interface in the case of v2 . v1 .

plane waves over the region of observation, and, in this sense, we may speak
of plane waves. In Figure 6.10, a plane wave is incident on a plane horizontal
interface between two layers with wave velocities v1 and v2 . In the seismic
case, velocity usually increases with depth. For that reason, we have drawn
the figure for the case of v2 . v1 . The figure depicts successive positions of
the wavefronts of both the incident and transmitted waves. The reflected
wavefronts are not shown. In the given time increment Dt, the wavefront
has moved a distance Dr1 in layer 1 and Dr2 in layer 2. In the same time
increment, the contact of the wavefront at the interface has moved from O
to P, which represents a distance of OP along the interface. Angle u1 is
the angle of incidence (the angle between the ray and normal in the upper
medium), and similarly u2 is the angle of refraction (the angle between
the ray and normal in the lower medium). These angles also can be described
as the angles between the wavefronts and the interface.
Now, we can discuss the geometry highlighted in Figure 6.11. The
various distances in the two layers can be expressed as

Dr1 Dr2
OP = = . (6.71)
sin ui sin ut
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 6: Eikonal Equation and Ray Equation 185

DOI:10.1190/1.9781560803737

Figure 6.11. Reproduction of Figure 6.10 with lines removed to show only items
pertinent to the discussion.

so that

1 sin u1 sin u2
= = . (6.72)
OP Dr1 Dr2
The velocities yield

Dr1 = v1 Dt, Dr2 = v2 Dt . (6.73)

Therefore, it follows that

Dt sin u1 sin u2
= = . (6.74)
OP v1 v2
This result is recognized as Snell’s law.

Comparison of Fermat and Huygens

Fermat’s principle, or the principle of least time, is the assertion that
the path taken between two points by a ray is the path that can be traversed
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

186 Basic Wave Analysis, Part 2: Raypath Analysis

in the least time. Now, we can state the principle of least time in a more
accurate way. A particular path has the property of least time if there is
no first-order change in the time for any small change in the ray in any
manner whatsoever. Therefore, if we plot the time required to travel a
certain path against its crossing point on the interface, we would produce
a U-shaped curve. The minimum value on the time-curve corresponds to
the shortest of all possible times. This means that, if we move to points
near to the minimum value, there is essentially no change in time (in the
first approximation), because the slope is zero at the bottom of the curve.
There will be only a second-order change in the time. In other words, the
principle of least time says that the waves take a path such that there are
many other paths nearby that take almost the same time. More precisely,
we should say stationary time instead of least time, because the optimum
time could be either a maximum or an inflection point with zero slope in
addition to a minimum.
DOI:10.1190/1.9781560803737

Figure 6.12 shows point P as the common reflection point. Curve E is

the elliptical reflecting surface with the two given points A and B as foci.
Curve M is a reflecting surface tangent to the elliptical surface E at P, but
with more curvature. If Q is any point on curve M close to point P, then
the hypothetical travel path AQB is shorter than the ray path APB,
because the path is constant for all points on the elliptical curve E.

Figure 6.12. APB is a maximum-time path with respect to surface M, and APB is a
minimum-time path with respect to surface L.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 6: Eikonal Equation and Ray Equation 187

Because Q is any neighboring point on curve M, it follows that raypath APB

is a maximum with respect to reflector M. Likewise, if L is a reflecting
surface tangent to the elliptical surface at P, but with less curvature, then
raypath APB is a minimum with respect to reflector L. Therefore, if any
arbitrary reflecting surface has more curvature than the elliptical surface
at P, then the ray path has maximum time. If the arbitrary reflecting
surface has less curvature at P, then the ray path has minimum time. The
tangent plane T always has less curvature than the elliptical surface so
that a reflection event from a plane always has minimum time (Robinson
and Treitel, 2008).
Fermat described his principle of least time in a letter dated 1 January
1662. It was met with objections that stated the principle was a moral,
and not a physical, statement, so it could not be the cause of any physical
effect. However, this negative opinion proved to be wrong. Fermat’s prin-
ciple was the forerunner of the principle of least action, or, more accurately,
DOI:10.1190/1.9781560803737

the principle of stationary action. The principle could be used to derive

Newtonian, Lagrangian, and Hamiltonian equations of motion and even
general relativity. The principle of least action was fundamental to the
development of quantum mechanics.
In 1678, Huygens puts forth his rule for the propagation of light as
waves (Huygens’ principle) and his wave proof of Snell’s law. From a
physical point of view, Fermat’s principle can be justified by Huygens’ prin-
ciple. Let us explain. Suppose that waves propagate from A to B by all poss-
ible paths. Most of the waves arriving at B will interfere destructively.
However, a considerable number of neighboring routes will be close to
the least-time path and therefore will interfere constructively. Thus, the
least-time path is the route along which the wave appears to travel. In
current terminology, this is called a stationary phase argument.
Consider these practical examples. A lifeguard on the beach sees a
person needing assistance in the water a long distance down the beach. A
lifeguard can run faster that he can swim. Therefore, the lifeguard will
run most of the distance down the beach before he swims to the person.
Optimally, the lifeguard would take the least-time path. In a recent study,
ants are allowed to travel on two different surfaces, with one surface
where they can walk more quickly (corresponding to the beach) than
another surface (corresponding to the water) (Figure 6.13). This experiment
concludes that the ants behave the same way as the lifeguard. Similar to the
lifeguard, the ants choose the fastest route rather than the direct route. The
ants obey Fermat’s principle of least time (Oettler et al., 2013). Further dis-
cussion of this remarkable behavior is found at https://phys.org/news/
2013-04-ants-fermat-principle.html.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

188 Basic Wave Analysis, Part 2: Raypath Analysis

Figure 6.13.
Example of
Fermat’s principle
of least time.
DOI:10.1190/1.9781560803737

Proving theories, however, requires mathematics in addition to

practical examples. Major areas of mathematics are geometry, arithmetic,
algebra, and calculus. Calculus uses infinitesimals in order to address
rates of change, accumulated change, and many other rates. In fact, Archi-
medes uses infinitesimals in a manner that foreshadows integral calculus.
Through the “Method of Exhaustion,” he approximates the value of p. By
this method, we can determine areas and volumes of figures with curved
lines and surfaces, such as pyramids, cones, circles, and spheres. In a
similar way, Fermat’s proof of the principle of least time uses infinitesimals
in a manner resembling differential calculus. Indeed, the invention of calcu-
lus provides the needed symbols (e.g., the differential sign and the integral
sign). This symbolic approach makes it possible to systematically acquire
the final results without the burden of breaking the problem down into its
infinitesimal components as done by Archimedes and Fermat. In effect,
Fermat uses a ray (see ACB shown in Figure 6.8) to describe the motion
of a particle. By means of infinitesimal analyses, he transfers the problem
into a picture (see Figure 6.9) that is remarkably similar to the wave
figure of Huygens (see Figure 6.11). In other words, Fermat encompasses
both the particle theory and the wave theory. In so doing, Fermat anticipates
the wave-particle duality which is the core of quantum mechanics.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7

Ray Tracing

“Mathematical analysis is as extensive as nature itself. It defines all

perceptible relations, measures times, spaces, forces, temperatures.
This difficult science is formed slowly, but it preserves every prin-
DOI:10.1190/1.9781560803737

ciple which it has once acquired. It grows and strengthens itself

incessantly in the midst of the many variations and errors of the
human mind. Its chief attribute is clearness. It has no marks to
express confused notations. It brings together the most diverse
phenomena, and discovers the hidden analogies which unite them”

—Joseph Fourier

Classical ray tracing

A common assumption of ray theory is that energy transport occurs
along the ray paths, not across them. This assumption is violated when
the ray passes a caustic or when diffraction at a model discontinuity
occurs. The three-dimensional theory of classical ray tracing is given in
the works of Červený et al. (1977), Hubral (1980), Hubral and Krey
(1980), Červený (2001), and many others. In this chapter we will present
a simplified two-dimensional version. In isotropic media, a raypath is a
line everywhere perpendicular to the wavefronts. In the situation of horizon-
tal layering, a raypath can be characterized by its direction at the surface.
The ray parameter p = sin ui /vi is constant along any ray, where ui is the
angle with the vertical and vi represents velocity at the given point.
Seismic energy does not travel only along raypaths, as seismic energy
can reach a point by diffraction even if the raypath is blocked. However,
much of the energy does travel along the raypaths. Raypaths constitute a

189
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

190 Basic Wave Analysis, Part 2: Raypath Analysis

useful method for determining arrival times through geologic models by

means of ray tracing. Ray tracing provides a way of determining the
arrival time at receivers by following raypaths which obey Snell’s law
through a geologic model for which the velocity distribution is known.
If we look at a continuous velocity log recorded in an oil well, we
immediately see that the sedimentary rock layers do not compose a
simple geologic structure consisting of only a few homogeneous layers
bounded by sharp interfaces. Instead, the rocks consist of highly stratified
layers which can have complex folding and faulting. Because the true
geology in all of its details is so complicated, it can never be determined
completely by a seismic reflection survey. Instead, the purpose of a
seismic survey is to obtain a reasonable geologic model which can lead to
the discovery of petroleum. As a rule, the subsurface cannot be adequately
parametrized to account for its full complexity. Moreover, we do not know
how to model wave propagation for the full complexity of the subsurface.
DOI:10.1190/1.9781560803737

The typical models used in the study of raypaths and traveltimes cannot
account for the large number of interfaces found in an exploration
prospect. For practical reasons, such models must be limited to only a rela-
tively small number of layers. However, such models can be useful in
gaining insight as well as providing practical approaches for the interpret-
ation of seismic data. If the agreement between the model and the actual sub-
surface structure is poor, then the model must be altered. The complexity of
the model depends upon the type and amount of information we want to
extract, as well as upon the quantity and quality of the input data. The accu-
racy of the data, the relevance of the model, and the applicability of the
theory are factors which always must be balanced in order to obtain the
best estimates of the desired subsurface parameters.
A homogeneous material is a material which has uniform composition
and uniform properties throughout. The opposite term of homogeneous is
inhomogeneous or heterogeneous. An isotropic material is a material for
which the physical and mechanical properties are equal in all orientations
or directions. If the grains of the material are not oriented uniformly in all
directions, it is an anisotropic material. Properties such as Young’s
modulus, thermal expansion coefficient, and magnetic behavior can vary
with directions in anisotropic materials. A discussion of isotropy is based
primarily on the properties of the context. In physics, if the thermal expan-
sion coefficient of a solid is the same in all directions, then it is said to be
isotropic in that physical classification. Also, concepts such as optical iso-
tropy and electromagnetic isotropy are discussed in physics.
Now, consider a piecewise-constant model, or simply a blocky model,
as depicted in Figure 7.1. For ease of presentation, we will consider only
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 191

Figure 7.1.
A piecewise-
constant model
(also known as a
blocky model).
DOI:10.1190/1.9781560803737

two-dimensional models, with x representing the horizontal (lateral) coordi-

nate and z the vertical (depth) coordinate. The top surface of the model is
a straight line which represents the earth’s surface or, more realistically,
the common datum plane to which all surface measurements have been
reduced. Below this surface line are a series of smoothly curved lines,
falling in approximately horizontal directions, which represent the geologic
interfaces. The material between any two consecutive interfaces is taken
to be a homogeneous isotropic rock layer. As we know, a rock layer can
transmit both compressional waves and shear waves. If the interfaces do
not intersect each other, then the blocky model also is referred to as a
layer-cake model.
A compressional wave (also called longitudinal wave or P-wave) is
a wave in which the disturbance is an elastic deformation in the direction
of motion of the wave. A shear wave (also called a transverse wave or
S-wave) is a wave in which the disturbance is an elastic deformation
perpendicular to the direction of the motion of the wave. An acoustic
wave (also known as a sound wave) is a type of longitudinal wave that pro-
pagate by means of adiabatic compression and decompression. Important
quantities for describing acoustic waves are sound pressure, particle veloc-
ity, particle displacement, and intensity. In the most basic exploration
models, it is assumed that compressional waves but no shear waves
are transmitted. In such cases, the scalar wave equation with the compres-
sional wave velocity can be used. In summary, the blocky 2D model
consists of a horizontal line for the earth’s surface, and a sequence of
smoothly curved lines for the subsurface interfaces. Between any two con-
secutive interfaces is a homogeneous isotropic rock layer which transmits
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

192 Basic Wave Analysis, Part 2: Raypath Analysis

only compressional waves. Each layer has a given constant compressional

wave velocity.
Next, we would like to derive a recursive algorithm for tracing a
raypath through the model. The raypaths between any two consecutive inter-
faces are straight lines. Ray theory requires that Snell’s law is satisfied at
each interface. Thus, we see that the solution of the ray-tracing problem is
of a purely geometrical nature. We want to consider the forward ray-
tracing problem; that is, we want to trace a ray that starts at a source point
in a given direction. This problem also is referred to as the initial-value
ray-tracing or Cauchy problem. We assume that all of the layer velocities
and all of the interface shapes are known. According to Snell’s law, the
initial ray direction determines the entire raypath through the given
model. However, there is no way of choosing such an initial direction a
priori so that the raypath passes through some specified end point. This dif-
ficulty is the problem of two-point ray tracing. The two main techniques for
DOI:10.1190/1.9781560803737

solving it are shooting and bending. The term shooting method is used for
any iterative method which by successive approximations on the initial
direction of the ray (through trial and error or search methods such as
Newton’s root finding algorithm) determine the ray trajectory to a specific
end point. Bending involves starting with an arbitrary initial curve connect-
ing the specified end points and iteratively updating (bending) until it
satisfies the ray equation, or, equivalently, Fermat’s minimum traveltime
principle. In our discussions, we are not trying to solve the two-point ray-
tracing problem, so with the location where an initial ray hits the surface
is not an issue.
We will use various coordinate systems. One is a fixed-frame coordinate
system (x, z) in which x is the horizontal axis and z is the vertical axis. The
model is described in terms of this system; that is, each interface is provided
as a locus of points with respect to the (x, z) system. See Figure 7.2 for an
illustration of a fixed-frame coordinate system.
In addition, Figure 7.2 shows a right-handed moving-frame coordinate
system (X, Z ). This coordinate system accompanies a given wavefront as it
travels along a selected raypath. We will define the Z-axis as the axis that
points in the direction of the advancing wavefront. That is, the Z-axis lies
in the direction of the raypath. The X-axis is the axis that is perpendicular
to the Z-axis, and thus the X-axis is tangent to the wavefront.
The moving-frame coordinate system (X, Z ) at any point can be
fully described by three vectors, i.e., the origin vector O, the unit vector
i in the X direction, and the unit vector k in the Z direction (see
Figure 7.3). The vectors O, i, k are specified in terms of the fixed-frame
system (x, z).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 193

DOI:10.1190/1.9781560803737

Figure 7.2. Fixed-frame coordinate system (x, z) and moving-frame coordinate

systems (X, Z ). Note that the z-axis is turned in the direction opposite of typical
figures (i.e., upside down), so that the z-axis (depth axis) points up.

Figure 7.3. The three

vectors which specify the
moving-frame system
(X, Z ).

Using the following numbering system, let the source point be O0, the
initial direction of the ray be k0, and i0 be the corresponding unit tangent
to the wavefront. The vectors i0, k0 are chosen so that they form a right-
handed system.
As shown in Figure 7.4, let the first interface encountered by the raypath
be called interface 1. The layer between the source and interface 1 is called
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

194 Basic Wave Analysis, Part 2: Raypath Analysis

Figure 7.4.
Moving-frame
coordinate system
with numbering
system employed
in the model.
DOI:10.1190/1.9781560803737

Table 7.1. Moving-frame coordinate systems (each on the refracted side of

the interface), as shown in Figure 7.4.

Source O0 i0 k0
Interface 1 O1 i1 k1
Interface 2 O2 i2 k2
Interface 3 O3 i3 k3

layer 1 with velocity v1, and the raypath segment between the source and
interface 1 is called s1.
The raypath segment s1 is a straight line. (We also will let s1 denote the
length of this segment.) According to Snell’s law, the incident raypath bends
at interface 1 into a refracted raypath on the other side of interface 1. Let O1
be the point of intersection of the raypath with interface 1. There are two
moving-frame systems at point O1, one corresponding to the incident side
of the interface and the other to the refracted side. Let us denote the
moving-frame system on the refracted side by O1, i1, k1.
From O1, the raypath is a straight line in direction k1 to the next interface,
which we call interface 2. The point of intersection of the raypath with
interface 2 is O2. The layer between interface 1 and interface 2 is called
layer 2 with velocity v2, and the raypath segment between O1 and O2 is s2.
According to Snell’s law, the incident raypath bends at interface 2 into a
refracted raypath on the other side of interface 2. Let us denote the moving-
frame system on the refracted side by O2, i2, k2. Thus, we have the sequence
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 195

Table 7.2. Sequence of layers in moving-frame coordinate

system, as shown in Figure 7.4.

Layer 1 with velocity v1 Segment s1 with direction k0

Layer 2 with velocity v2 Segment s2 with direction k1
Layer 3 with velocity v3 Segment s3 with direction k2

of moving-frame coordinate systems (each on the refracted side of the inter-

face) shown in Table 7.1. We also have the sequence of layers shown in
Table 7.2.
Before engineering designs could be created on computers, designers
used drafting tools. In particular, draftsmen drawing curves in shipbuilding
would use splines—long, thin, flexible strips of wood, plastic, or metal.
These splines were held in place at control points with lead weights. The
DOI:10.1190/1.9781560803737

elasticity of the spline material combined with the constraint of the control
points would cause the strip to take the shape that minimized the energy
required for the bending between the fixed points. As a result, the spline
took the smoothest possible shape. Then, the spline curve could be trans-
ferred to the material used to build the ship. With the advent of computers,
computational counterparts of wood, plastic, or metal splines have gained
importance.
The simplest computational method is linear interpolation (i.e., the
connection of consecutive data points with lines). The next simplest
method is quadratic. The use of the quadratic function on each interval
makes the graph smoother. We can make the graph even smoother by
matching the derivatives at the data points. If this is done, the result is
called a quadratic spline. Using cubic functions or fourth-degree functions
should make the graph smoother still. Where should we stop? An overfitted
model contains more parameters than can be justified by the data. The
essence of overfitting is to have unknowingly used some aspect of noise
as if it were underlying the model structure. An underfitted model is a
model in which some parameters or terms that would appear in a correctly
specified model are missing. Underfitting occurs when the model cannot
adequately capture the underlying structure of the data. The usual decision
is that cubic is the optimal degree for splines. The spline also ensures that the
interfaces do not generate spurious diffracted rays at the interpolation points,
in other words, the spline representation makes the model raytracing-
friendly. The basic idea of the cubic spline is that we represent the function
by a different cubic function on each interval between data points. In order to
keep the mathematics simple in the following presentation, we will not use
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

196 Basic Wave Analysis, Part 2: Raypath Analysis

splines as such. Instead, we will use a quadratic function (more specifically,

a parabola) fitted by least squares. For our purpose of kinematic raytracing
(computation of ray paths and traveltimes along them), this is sufficient.
For the purpose of computation of wavefront curvature, the piecewise
quadratic function may generate instabilities, which we will ignore for sim-
plicity in this book.
Now, we want to provide a recursive scheme to trace the raypath.
We will give the algorithm to move from one interface to the next. Then,
the algorithm can be used in a recursive manner to trace the ray from the
source to any interface. For simplicity, let us describe the algorithm in
terms of specific interfaces, e.g., from interface 2 to interface 3.
As initial data, we assume that we know the moving-frame coordinate
system O2, i2, k2 on the refracted side of interface 2. What we mean is that
the vectors O2, i2, k2 are known in terms of the fixed-frame coordinates
(x, z). The algorithm is as follows (see Figure 7.5).
DOI:10.1190/1.9781560803737

Step 1. Recall that the locus of interface 3 is known in terms of the fixed-
frame system. We select a set of points on this locus, in the vicinity of
the raypath, and transform these points into the moving-frame system as
follows (see Figure 7.4). Let the selected points on interface 3 be denoted
by P3j for j = 1, 2, . . . , N, where the vectors P3j are in terms of the fixed-
frame system. Let (X3j , Z3j ) be the desired moving-frame coordinates of
P3j. These coordinates can be computed by means of the equation

X3j i2 + Z3j k2 = P3j − O2 . (7.1)

Thus, we obtain a set of points on interface 3 in terms of the known moving-

frame coordinate system O2, i2, k2.

Figure 7.5. The

moving-frame
coordinates
(X3j, Z3j) of the
interface point
P3j. Here, k2
denotes the
tangent to the ray,
i2 the normal.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 197

Step 2. In this step of the algorithm, we want to find an expression

for interface 3 in the vicinity of the raypath. By least-squares, we fit a
parabola to the chosen points P3j (where j = 1, 2, . . . , N). This parabola
represents the desired approximation to interface 3. This parabola is
determined in terms of the moving-frame coordinates (X, Z) of the
system O2, i2, k2. An explicit expression for the parabolic approximation
of interface 3 is

Z = bX 2 + cX + d , (7.2)

where the coefficients b, c, d are determined by least-squares. The value of Z

for X ¼ 0 provides the length of the raypath segment from O2 to O3; that is,
we can find s3 as s3 ¼ d.
Thus, as shown in Figure 7.6, we can determine the point O3 as
DOI:10.1190/1.9781560803737

O3 = O2 + s3 k2 . (7.3)

An implicit expression for the parabolic approximation of interface 3 is

F(X, Z) = Z − bX 2 − cX − d = 0 . (7.4)

The traveltime between O2 and O3 is

s3
Dt3 = . (7.5)
v3

Figure 7.6. The

parabolic
approximation to
the interface.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

198 Basic Wave Analysis, Part 2: Raypath Analysis

Step 3. In this step, we want to cross interface 3. First, we compute the

unit normal vector n3 to the interface. The vector n3, by definition, points in
the direction of the gradient

∂F ∂F
∇F = i2 + k2 = (−2bX − c)i2 + k2 . (7.6)
∂X ∂Z

In terms of the moving-frame system O2 , i2 , k2 , the point O3 has

coordinates X ¼ 0, Z ¼ s3. Because we want n3 to be the normal at point
O3, we must therefore let X ¼ 0 in ∇F. Thus, the gradient at O3 is

∇F = −ci2 + k2 . (7.7)
√
The length of this vector is 1 + c2 . The required unit normal at O3 is
DOI:10.1190/1.9781560803737

−c 1
n3 = √ i2 + √ k2 . (7.8)
1+c 2 1 + c2

Next, we must find the angle of incidence (see Figure 7.7). Recall that
the incident raypath to interface 3 has direction given by k2. The angle of

Figure 7.7. The raypath is bent at the interface by the angle f ¼ uI 2 uR.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 199

incidence uI is the angle between the incident raypath and the normal; that is,

cos uI = n3 · k2 , (7.9)

where the dot indicates the dot product. Because i2 · k2 = 0 and k2 · k2 = 1,

we have

1
cos uI = √ . (7.10)
1 + c2
Thus, we have found the angle of incidence uI.
Next, we must find the angle of refraction uT. This angle can be deter-
mined immediately from Snell’s law,

sin uI sin uT
= , (7.11)
v3 v4
DOI:10.1190/1.9781560803737

because the velocities v3 and v4 are given constants in the model.

Step 4. In this step of the algorithm, we find the unit vectors i3 and k3 of
the moving-frame system on the refraction side of interface 3. As shown in
Figure 7.8, the vectors i3, k3 are obtained by rotating the vectors i2, k2 by the
net amount that the raypath is bent at interface 3; that is, the angle of rotation
f is the difference of the incident angle and the refraction angle:

f = uI − uT . (7.12)

Thus, we compute i3, k3 as

i3 = cos f i2 + sin f k2
k3 = −sin f i2 + cos f k2

Figure 7.8. The

vectors (i2, k2) are
rotated by the angle
f to obtain the
vectors (i3, k3).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

200 Basic Wave Analysis, Part 2: Raypath Analysis

Figure 7.9.
Rotation angle f in
the case of
reflection.
DOI:10.1190/1.9781560803737

or
    
i3 cos f sin f i2
= . (7.13)
k3 −sin f cos f k2

This step ends the algorithm.

In the four steps of the algorithm, we have computed the moving-frame
system O3, i3, k3 on the refraction side of interface 3 from the given moving-
frame system O2, i2, k2 on the refraction side of interface 2.
An easy modification allows us to find reflection paths as well as
refraction paths. Because the angle of incidence is equal to the angle of
reflection, the raypath is bent by the angle

f = p − 2uI . (7.14)

Thus, we use this value of f as the rotation angle in Step 4 (see Figure 7.9).

Determination of wavefront curvature

Next, we consider a wave that emerges at the surface of the earth.
Note that additional information on curved wavefronts and curved interfaces
can be found in Hubral (1979, 1980), Krey (1980), and other papers. We rep-
resent the earth’s surface by a horizontal straight line with coordinate x.
Suppose the wave emerges at point x. At point x, we can describe the
wave by both its raypath and its wavefront. At x and in the vicinity of x,
we place a string of detectors. Let us assume that the wave is due to an
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 201

impulsive point source initiated at time t ¼ 0, and that the wave travels in
a raypath beam. This raypath suffers refraction and reflection at various
interfaces until it finally emerges at surface point x. The emergence of
this wavefront is recorded as a seismic event by the string of detectors.
This event can be represented by a time-distance curve t(x) which gives
the arrival time of the wavefront as a function of horizontal distance. In
other words, what we can measure (in an ideal case) at the earth’s surface
by a seismic experiment is a time-distance curve for each subsurface path,
whether this raypath represents a primary reflection or a multiple reflection
event. What can we infer about the actual seismic wave which produced
this time-distance curve? We will address this question now.
At a given emergence point x of the ray, there are three important
quantities which we will use. These quantities are:

1. the ordinate t(x) of the time-distance curve,

DOI:10.1190/1.9781560803737

2. the slope dt/dx of the time-distance curve, and

3. the second derivative dt2 /dx2 of the time-distance curve. (Note that this
quantity is related to the geometrical spreading along the ray and, there-
fore, the amplitude at x.)

The value t(x) provides the traveltime along the raypath from the source
to the receiver at x. As we know from previous sections, the slope gives us
the angle u of emergence. Let us review this result (see Figure 7.10). The
slope of the time-distance curve at x is

dt
tan a = , (7.15)
dx
which is the horizontal slowness of the wavefront. Furthermore, from the ray
diagram we have
v1 dt
sin u = , (7.16)
dx
which gives
dt sin u
= . (7.17)
dx v1
Because we can physically measure the angle a on the time-distance curve,
we can find the emergence angle u by the equation
sin u
tan a = . (7.18)
v1
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

202 Basic Wave Analysis, Part 2: Raypath Analysis

Figure 7.10.
Relationship of the
time-distance
curve to the
emerging wave.
DOI:10.1190/1.9781560803737

(We assume the velocity v1 is known.) We immediately recognize the

equation (7.18) (i.e., the tangent ¼ sine equation) as Snell’s law.
Snell’s law often is called the refraction law. More precisely, it is the
refraction law of bearing, because it allows us to find the bearing (i.e.,
angle u) of the emergent ray. There is another law that supplements
Snell’s law, which we call the refraction law of range or Huygens’ law. It
allows us to find the range (i.e., radius) of the hypothetical spherical wave
that has the same curvature as the emergent wavefront. Huygens is the
first to devise the concepts of curvature and radius of curvature in a math-
ematically precise way.
In geometry, the radius of curvature (designated by r) of a curve at a
point is a measure of the radius of the circular arc which best approximates
the curve at that point. It is the inverse of the curvature. In the case of a space
curve, the radius of curvature is the length of the curvature vector.
Huygens’ law (or the refraction law of range) can be obtained in the
following way. Again, we reference Figure 7.10. The emergent wavefront
has a certain curvature. Let us draw a circle, with a center on the straight-
line extension of the raypath, and with radius r, such that the arc of the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 203

circle best fits the emergent wavefront. Thus, the center is the center of the
hypothetical spherical wave, and the radius r of the circle is the radius
of curvature of the wavefront. Let du be the increment of the bearing
angle u. Then, the corresponding increment of the wavefront arc is r du.
The little right triangle with acute angle u shown in Figure 7.10 has adjacent
side r du and hypotenuse dx. Thus, we have the formula
r du
cos u = . (7.19)
dx
Now, we take the refraction law of bearing (i.e., Snell’s law)
dt sin u
= (7.20)
dx v
and differentiate to obtain
d 2 t cos u d u
DOI:10.1190/1.9781560803737

= . (7.21)
dx2 v dx
Because, as has been derived,
d u cos u
= , (7.22)
dx r
we have
d 2 t cos2 u
= . (7.23)
dx2 rv
This equation (7.23) relates the range r of the spherical wave to the second
derivative of the time-distance curve, and thus is the required refraction law
of range (i.e., Huygens’ law).
In answer to the question posed at the beginning of this section, the
following information can be found from the time-distance curve about
the emergent waveform:
traveltime = t(x)
 
−1dt dt sin u
bearing = u = sin v1 so =
dx dx v1

cos2 u d 2 t cos2 u
range = r = so = . (7.24)
d2 t dx2 r v1
v1 2
dx
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

204 Basic Wave Analysis, Part 2: Raypath Analysis

Next, let us find the hyperbolic approximation to the time-distance

curve. The Taylor expansion is
 
dt  1 d 2 t 
t(x) = t(x0 ) +  (x − x0 ) + (x − x0 )2 + · · · (7.25)
dx x0 2 dx2 x0

The bearing u0 and range r0 refer to the point x0. Using expressions (7.20)
and (7.23) for the derivatives, we have
sin u0 cos2 u0
t(x) = t(x0 ) + (x − x0 ) + (x − x0 )2 . (7.26)
v1 2r0 v1

If we square equation (7.26) and drop higher-order terms, we obtain the

required hyperbolic expression:
 2
sin u0 t(x0 ) cos2 u0
t (x) = t(x0 ) +
2
(x − x0 ) + (x − x0 )2 .
DOI:10.1190/1.9781560803737

(7.27)
v1 r0 v1

This is equivalent to the equation for move-out in the context of

common-reflection surface (CRS) stacking. For more information, see
the publications by Peter Hubral and his group at the Geophysical Institute
at the Karlsruhe Institute of Technology (Mann et al., 1999; Bergler
et al., 2002).

Figure 7.11.
Change of the
radius of
curvature due to
refraction (case
of horizontal
interface).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 205

DOI:10.1190/1.9781560803737

Figure 7.12. Change of the radius of curvature due to refraction (case of dipping
interface).

Figure 7.11 shows the case of a horizontal straight-line interface,

and Figure 7.12 shows the case of a dipping straight-line interface. The
following actions will be taken with respect to this interface. As a result,
the resulting arguments will hold for either case. Now, the results of
this section can be extended from the case of the horizontal surface of the
earth to the case of any dipping straight-line interface.
Let an incident wave be refracted at the interface using the notation in
Table 7.3. Let t(x) be the time-distance curve as measured at point x on
the interface. Although previously the variable x referred to the coordinate
as measured along the earth’s surface, now the variable x refers to the coor-
dinate as measured along the interface.

Table 7.3. Notation for incident wave refracted at interface.

Incident side Refracted side

Velocity vI vT
Angle uI uT
Radius rI rT
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

206 Basic Wave Analysis, Part 2: Raypath Analysis

To measure t(x), we would have to put a string of detectors along the

interface. Mathematics is precise; geology is not. In mathematics, the inter-
face is an infinitesimally thin line. If we put the detectors an infinitesimal
distance from the interface on the incident side, then Snell’s law and
Huygens’ law are, respectively,
dt sin uI
= (7.28)
dx vI
and
d 2 t cos2 uI
= . (7.29)
dx2 rI vI
If we put the detectors an infinitesimal distance from the interface on the
refracted side, then Snell’s law and Huygens’ law are, respectively,
DOI:10.1190/1.9781560803737

dt sin uT
= (7.30)
dx vT

and
d 2 t cos2 uT
= . (7.31)
dx2 rT vT
According to geology, the traveltime t(x) is the same on both sides of
the interface. We thus have:
Snell’s refraction law of bearing:

sin uI sin uT
= . (7.32)
vI vT

Huygens’ refraction law of range:

cos2 uI cos2 uT
= . (7.33)
rI vI rT vT

Case 1: A curved wavefront striking a straight interface

First, we will address the case of a curved wavefront striking a straight
interface (Figure 7.13). Here, we will consider only the case for convex
(diverging) wavefronts. However, a wavefront striking the interface can
be concave (converging) as well. In three dimensions, it can even have
two curvatures with different signs. In geophysics, as in all sciences, there
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 207

Figure 7.13. Curved

wavefront striking a
straight interface.

are two mathematical formulas that are used frequently and which we will
DOI:10.1190/1.9781560803737

employ: the Pythagorean theorem and Newton’s binomial expansion.

Let us consider a moving frame (X, Z) on the incident ray, with origin at
the point of intersection of the ray with the interface. As shown in
Figure 7.12, the Z axis points in the direction of the ray, and X is perpendi-
cular to Z. The circular arc that approximates the wavefront is the arc of the
circle given by the equation

X 2 + (Z + rI )2 = r2I . (7.34)

Observe that Z is negative, as shown in Figure 7.12. This equation

(7.34), of course, is an expression of the Pythagorean theorem. Let us
write the circle equation as


X2
Z + rI = r2I − X 2 = rI 1 − 2 . (7.35)
rI

Now, we use the square-root approximation (which is an expression of

Newton’s binomial theorem) and obtain
X2
Z + rI = rI − , (7.36)
2rI

which is the parabola

X2
Z=− . (7.37)
2rI
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

208 Basic Wave Analysis, Part 2: Raypath Analysis

(Note that the minus sign confirms that Z is negative.) This equation
approximates the wavefront arc by the arc of a parabola. The constant in
this parabola is given a name, the curvature aI, where

1
aI = . (7.38)
rI

The curvature is the reciprocal of the radius of curvature. Note that the
minus sign in the equation for the parabola means that, for positive curva-
ture, the convex side of the wavefront points in the direction of travel
(i.e., in the plus Z direction). Here, we restrict the argument to the case of
convex wavefronts. A similar treatment can be made for other cases. To
this point, we have addressed the case of a curved wavefront striking a
straight interface.
DOI:10.1190/1.9781560803737

Case 2: A straight wavefront striking a curved interface

Next, we address the case of a straight wavefront striking a curved
interface as shown in Figure 7.14. Let (x, z) be the coordinate system
for the interface, with the origin at the point of incidence, with z normal
to the interface and pointing into the refracted side, and x tangent to the inter-
face and thus normal to z. We approximate the curved interface by an arc of a
circle with its center on the normal and radius ri . We take ri positive if the
center is on the incident side. Figure 7.14 illustrates the case in which uI ¼ 0.

Figure 7.14. Straight wavefront

striking a curved interface with angel
of incidence uI ¼ 0.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 209

Figure 7.15. Straight wavefront

striking a curved interface with
angle of incidence uI = 0.

Figure 7.15 illustrates the case in which uI = 0. In either case, the projection
DOI:10.1190/1.9781560803737

of ri on the incident ray is ri cos uI. Heuristically, the value ri cos uI may be
regarded as the radius of curvature of an “approximate” curved interface.
The parabolic approximation can be found by replacing rI in equation
(7.37) by ri cos uI. Thus, the straight wavefront shows the interface as the
curve with parabolic approximation
X2
Z=− . (7.39)
2 ri cos uI

The case of a straight wavefront and curved interface has the same form
as the case of a curved wavefront and straight interface, except the sign of
dt is the opposite. Huygens’ law is
d2 t cos2 uI
= . (7.40)
dx2 ri cos uI vI

In the case of both a curved wavefront and a curved interface, the second
derivative is the sum of both separate curvature effects; that is,
d 2 t cos2 uI cos2 uI
= − . (7.41)
dx2 rI vI ri cos uI vI

Thus, Huygens’ law for the case of a curved interface (with radius ri of cur-
vature) and curved wavefronts is
   
cos2 uI 1 1 cos2 uT 1 1
− = − . (7.42)
vI rI ri cos uI vTI rT ri cos uT
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

210 Basic Wave Analysis, Part 2: Raypath Analysis

For example, suppose that a curved wavefront strikes a curved interface at

normal incidence. Then, uI ¼ uT and Huygens’ law becomes
   
1 1 1 1 1 1
− = − . (7.43)
vI rI ri vTI rT ri

Suppose that the curved interface has the same shape as the curved
incident wavefront, so there is an exact match. Then, rI ¼ ri, so the left
side of the equation (7.43) is zero. Then, it follows that the right side is
zero, so rT ¼ ri. Thus, the transmitted (i.e., refracted) wavefront also has
the same shape.

Propagation and depropagation

The customary laws of physics, except the second law of thermodyn-
DOI:10.1190/1.9781560803737

amics, are symmetric in time. Let us assume that the material is perfectly
elastic, so that wave motion in the material produces no loss to heat. In
such a situation, the reversal of time in the equations of motion simply
describes everything going backwards. If the detailed path information of
all of the motions is available, then reversal of all of the motions should
be exactly like a movie played backwards. The wave equation can be
solved in the case of time going forward to yield the propagation of a
seismic wave. Conversely, the wave equation can be solved in the case of
time going backward to yield the so-called depropagation of a seismic
wave. Depropagation is the inverse of propagation. A wave propagates
forward in time from point A to point B. The same wave depropagates back-
ward in time from point B to point A. The second law of thermodynamics is
different from the other physical laws. Entropy must never decrease in time,
except statistically and briefly. Many natural processes are apparently irre-
versible. Irreversibility is intimately connected to the direction of time.
Identifying the physical reasons for observed irreversibility would contrib-
ute to the understanding of physical processes. However, such consider-
ations are beyond the scope of data processing.
A primary reflected ray has a raypath composed of one leg consisting
of downgoing segments from the source to an interface and another leg con-
sisting of upgoing segments from the reflection point to the receiver. An
important special case occurs when the source and receiver positions
coincide. For a coincident source and receiver, the two legs coincide and
the two-way path is called a normal ray (see Figure 7.16). Because a
normal ray represents the minimum traveltime from the coincident
source-receiver pair to the reflecting interface, it follows that the normal
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 211

Figure 7.16. Primary reflection SDR and normal ray MNM, with N the normal
incidence point (NIP).
DOI:10.1190/1.9781560803737

Figure 7.17. Symmetric multiple reflection SDR and normal ray MPN, with N the
normal incidence point (NIP).

ray is perpendicular to that interface. The point N of reflection is called the

normal incidence point (NIP). The two-way time along a normal ray is
called by various names: the zero-offset reflection time, the t(0) reflection
time, or the two-way normal reflection time. Normal rays play an important
role in the theory of stacking.
A multiple reflected ray has a raypath which consists of more than one
reflection point (see Figure 7.17). We also can consider a normal multiple
ray. A multiple is called symmetric if the downgoing and upgoing segments
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

212 Basic Wave Analysis, Part 2: Raypath Analysis

Figure 7.18. A scattering point D. Of all scattered rays, the image ray is the one that
DOI:10.1190/1.9781560803737

is perpendicular to the earth’s surface.

reflect and refract at the same interfaces; otherwise, a multiple is called

asymmetric.
The law of reflection requires that the angle of incidence equals the
angle of reflection. However, seismic energy is returned to the surface not
only by reflection but also by scattering, a process in which the law of reflec-
tion does not hold. Using Figure 7.18, suppose that the depth point D is a
subsurface scatterer. For a source at S1 and a receiver at R1, there is a
minimum travel time path S1DR1 for the downgoing and backscattered
wave. If the source and receiver coincide, the downgoing and upgoing
legs also will coincide. For example, S2DR2 is the two-way path scatter
ray for the coincident source-receiver pair S2R2. Because the positive of
S2R2 is arbitrary, it follows that, for every coincident SR position, there is
a two-way path scatter ray. However, one particular scatter ray has great
importance in migration. It is the two-way path image ray (Hubral, 1977).
The image ray SIDRI represents the minimum traveltime from the scatterer
D to the earth’s surface. Because we assume that the earth’s surface is hori-
zontal, it follows that the image ray strikes the earth’s surface at right angles.
That is, the image ray is vertical at the earth’s surface. The two-way time
along the image ray is called the two-way image time. For the case in
which the receiver of a coincident source-receiver pair is made to respond
to only vertically emerging rays, then the image rays point out the subsur-
face locations from which mostly scattered energy is received. Image rays
play an important role in the theory of migration.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 213

Seismic analysis is almost always conducted through a process of

abstraction in which the data are reduced to those essentials which allow
mathematical treatment. The accuracy required in the results influences
the extent to which such reductions are made, and this accuracy is limited
by the reliability of the data. Today, the analytic methods are implemented
by means of the digital computer and their relative values are judged by their
speed, reliability, and clarity. Two of the important processes in seismic data
processing are stacking and migration, and much can be learned about them
by almost purely geometrical lines of thought. In addressing stacking and
migration from a geometrical point of view, one can take either the
concept of raypath or the concept of wavefront as the point of departure.
Geometric seismology, like geometric optics, may be described as an
approximation that holds for the case in which the wavelengths are of the
order of a fraction of the dimension of any object encountered. The wave-
lengths of light are extremely small in comparison to common objects
DOI:10.1190/1.9781560803737

(such as a spoon) that people usually encounter. As a result, geometric

optics is useful in daily living. However, seismic wavelengths are not
small in comparison to many geological objects that they encounter. As a
result, geometric seismology must be used with care and understanding.
When the dimension of some obstacle is of the same order of magnitude
as the wavelength, then the phenomenon of diffraction takes place. In the
case of an impulsive source (e.g., a dynamite explosion), the seismic
energy is transported in the form of a short compressional wavelet whose
breadth is the same order of magnitude as the geological bodies that we
wish to measure. Thus, within the usual ranges of distances, velocities,
and frequencies encountered in seismic work, we do not have pure raypaths
in the sense of infinitely narrow beams, so we must take into consideration
diffraction effects (Robinson and Treitel, 2008).
The dual of the ray is the wavefront. A wavefront can be defined as the
surface (in three dimensions) or curve (in two dimensions) over which
the phase of a traveling-wave disturbance is the same. In an isotropic
medium, the wavefront moves along rays perpendicular to the wavefront.
A wavefront also can refer to the leading edge of a waveform. The wave-
fronts show the position of a traveling seismic disturbance at successive
times. The shapes of the wavefronts depend upon the velocity distribution
of the rocks. Consider the wavefront representing the front surface of the
wave traveling away from an explosive source. We can physically visualize
such a wavefront and, in certain situations, we can measure it within narrow
limits. For example, when we look at the waves emanating from the point
where we drop a stone into a still pond, we see the wave motion traveling
outward as wavefronts and the raypaths are only mental constructions.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

214 Basic Wave Analysis, Part 2: Raypath Analysis

Likewise, in seismic interpretation the rays are interpreted as mathematical

abstractions, whereas wavefronts actually can be seen on the seismic sec-
tions as reflection events. It is for this reason that we now turn our attention
to the geometrical study of wavefronts.
Now, let us find a way to address wavefronts in our geologic model.
Recall that our model is a two-dimensional model in which the earth’s
surface is flat, and the subsurface interfaces are curved lines; the seismic ve-
locity in the layer between any two interfaces is constant. For pedagogical
purposes, we will simplify the model; we will assume that the subsurface
interfaces are straight lines. However, these straight lines do not have to
be horizontal; each may be tilted at a different angle from the others
(see Figure 7.19).
We want to show how the wavefront at any point on the raypath can be
expressed in terms of seismic parameters along the raypath. We make use of
the moving-frame (X, Z ) where the Z-axis points in the direction of pro-
DOI:10.1190/1.9781560803737

pagation along the ray (see Figure 7.20). At any point P on the raypath,
the wavefront at that point is approximated by the parabola in the
moving-frame coordinates given by

1
Z=− aX 2 . (7.44)
2
The coefficient a is called the wavefront curvature. The radius r of wave-
front curvature is defined as

1
r= . (7.45)
a

Figure 7.19. Type of

model used in which
constant velocity
layers are bounded by
straight-line
interfaces.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 215

Figure 7.20. Moving-frame (X, Z ) with parabola Z = −(1/2)a X 2 which

approximates wavefront.
DOI:10.1190/1.9781560803737

Let O be a point where the raypath crosses a straight-line interface.

Then, let aI, aT, aR denote, respectively, the wavefront curvatures of
the incident I, refracted (i.e., transmitted) T, and reflected R wavefronts.
Let uI be the angle of incidence and let uT be the angle of refraction. The
angle of reflection equals uI. Let vI be the velocity on the incident side
and let vT be the velocity on the refracted side of the interface. The law con-
necting the radii rI, rT of curvature is Huygens’ refraction law of range,
which we derived previously; it is

cos2 uI cos2 uT
= . (7.46)
rI vI rT vT
(This is a special case of the more general law for curved interfaces, also
given in the section, Determination of wavefront curvature; a curved-
interface model can be addressed using that law.) The reflected radius rR
of curvature is related to the incident radius rI of curvature by the reflection
law, which for a straight-line interface is

aR = aI . (7.47)

This law says that the process of reflection does not change the curvature. (A
reflection law for curved interfaces can be derived by the same methods as
we used to derive the curved-interface refraction law.)
These two laws can be transformed into corresponding laws for the
curvatures aI and aT (see Figure 7.21). The curvature aT of the refracted
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

216 Basic Wave Analysis, Part 2: Raypath Analysis

DOI:10.1190/1.9781560803737

Figure 7.21. Upon refraction at an interface, the wavefront curvature is changed

from aI to aT.

wavefront is given in terms of the curvature aI of the incident wavefront by

aT cos2 uT aI cos2 uI
= . (7.48)
vT vI

This equation follows from Huygens’ law by letting rT ¼ 1/aT and rI ¼ 1/

aI. The curvature aR of the reflected wavefront is equal to the curvature a of
the incident wavefront; that is,

aR = aI . (7.49)

This equation follows from the reflection law by noting that

1 1
rT = = rI = . (7.50)
aT aI
These laws provide the rules as to how a wavefront changes shape at an
interface. Next, we need a law as to how a wavefront changes as it propa-
gates within a layer. Here, we have only spherical spreading, so the radius
of curvature merely increases by an amount given by the distance which
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 217

the wavefront travels. If the velocity is v and the traveltime is Dt, then the
distance traveled along the straight-line raypath segment is simply v Dt.
Let r1 be the radius of curvature of the wavefront at the start of the
raypath segment, and let r2 be the radius of curvature of the wavefront at
the end of the raypath segment. Then, the transmission law of curvature is

r2 = r1 + v Dt . (7.51)

We can use these laws to compute the wavefront curvature at any

point along a raypath in our layered earth model. For a point source,
the radius of curvature at the source point is simply r ¼ 0, so a ¼ 1. For
a source along a straight line, the source radius of curvature is r ¼ 1, so
a ¼ 0. Usually, we assume that we have a point source, such as a dynamite
explosion.
Figure 7.22 depicts a raypath from a point source S to a receiver R. We
DOI:10.1190/1.9781560803737

assume that the raypath has been traced by the method given in the section,
Classical ray tracing. Now, we want to find the curvature of the wavefront at
the receiver point R. We make use of the following conventions. At interface
i, the incident angle is denoted by uIi, the refracted angle by uTi, and the
reflected angle by uRi. Ray segments si and interfaces i are counted succes-
sively from S to R, starting with i ¼ 1. The segment si is the length of the
raypath from interface i 2 1 to interface i. (The source point S may be

Figure 7.22. Typical model for propagation and depropagation.

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

218 Basic Wave Analysis, Part 2: Raypath Analysis

considered as interface 0.) The point of intersection of the raypath with inter-
face i is denoted by Oi. Then, we can compute the wavefront curvatures with
respect to the moving-frame coordinates by the recursion displayed
in Table 7.4.
The recursion in Table 7.4 can be applied to more complex models
(blocky models in general) than the one we have illustrated. This recursion
is the forward recursion; that is, one which goes in the direction of the
forward propagating, or expanding, wavefront. However, the recursion
can be reversed to produce a backward recursion; that is, one which goes
in the direction of the depropagating, or shrinking, wavefront. The depropa-
gation recursion is shown in Table 7.5. This depropagation recursion can be
used to solve a range of geophysical inverse problems.
There is symmetry in the case of two flat layers of thickness h1 and h2
with velocities v1 and v2, respectively (see Figure 7.23). Pascal observes
that, “Our notion of symmetry is derived from the human face. Hence, we
DOI:10.1190/1.9781560803737

demand symmetry horizontally and in breadth only, not vertically nor in

depth.” Let us find the radius of curvature of a vertically traveling wave.
Thus, the angle of emergence is u ¼ 0, and all other angles are zero.
Huygens’ law is simply
 
vI
rT = r , (7.52)
vT I

Table 7.4. Recursion to compute wavefront curvatures with respect to

moving-frame coordinates, as shown in Figure 7.22.

Incident side of O1: a−1

I1 = s1
v2 cos2 uI1
Refracted side of O1: aT1 = aI1
v1 cos2 uT1
−1 −1
Incident side of O2: aI2 = aT1 + s2
v3 cos2 uI2
Refracted side of O2: aT2 = aI2
v2 cos2 uT2
Incident side of O3: a−1 −1
I3 = aT2 + s3
Reflected side of O3: aR3 = aI3
Incident side of O4: a−1 −1
I4 = aR3 + s4
v2 cos2 uI4
Refracted side of O4: aT4 = aI4
v3 cos2 uT4
(v4 = v3 , v5 = v2 )
Incident side of O5: a−1 −1
I5 = aT4 + s5
(the receiver R)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 219

Table 7.5. Backward recursion to compute wavefront curvatures with

respect to moving-frame coordinates, as shown in Figure 7.22.

Refracted side of O4: a−1 −1

T4 = aI5 − s5
v3 cos2 uT4
Incident side of O4: aI4 = aT4
v2 cos2 uI4
−1 −1
Reflected side of O3: aR3 = aI4 − s4
Incident side of O3: aI3 = aR3
Refracted side of O2: a−1 −1
T2 = aI3 − s3
v2 cos2 uT2
Incident side of O2: aI2 = aT2
v3 cos2 uI2
−1 −1
Refracted side of O1: aT1 = aI2 − s2
v1 cos2 uT1
Incident side of O1: aI1 = aT1
v23 cos2 uI1
Refracted side of O1: a−1 −1
I0 = aI1 − s1
DOI:10.1190/1.9781560803737

(the source S)

Figure 7.23. A
constant-velocity
horizontal layer
model. The normal
ray is the vertical
raypath for
coincident source-
receiver at M,
corresponding to
offset x ¼ 0.

and the transmission law involves the thickness of the layer. The raypath for
coincident source and receiver (offset x ¼ 0) on the earth’s surface to and
from the second interface yields the propagation equations:

rI1 = h1
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

220 Basic Wave Analysis, Part 2: Raypath Analysis

 
v1
rT1 = r
v2 I1

rI2 = h2 + rT1

rR2 = rI2

rI3 = h2 + rR2
 
v2
rT3 = r
v1 I3

r0 = h1 + rT3 . (7.53)
DOI:10.1190/1.9781560803737

These equations provide

      
v2 v1 2
r0 = h1 + h2 + h2 + h1 = (v2 h2 + v1 h1 ) . (7.54)
v1 v2 v1

This formula readily generalizes for the case of N layers, so we see that the
radius of curvature at the surface is

N
2
r0 = vi hi . (7.55)
v1 i=1

Now, we can now use equation (7.26) from our previous section, Deter-
mination of wavefront curvature, to find the approximate time-distance
curve. For convenience, we repeat the equation here:

sin u0 cos2 u0
t(x) = t(x0 ) + (x − x0 ) + (x − x0 )2 . (7.56)
v1 2r0 v1

We put x0 ¼ 0 and sin u0 ¼ 0 into this equation, together with the above
value of r0, and obtain

t(0)
t2 (x) = t2 (0) + N x2 . (7.57)
2 i=1 vi hi
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 221

Equation (7.57) can be written as

x2
t2 (x) = t2 (0) + , (7.58)
v2RMS

because
N N
2 1
v2RMS = vi hi = v2i Dti (0) , (7.59)
t(0) i=1
t(0) i=1

where
N
2hi
Dti (0) = , t(0) = Dti (0) . (7.60)
vi i=1

Let us note that v2RMS is the average of the square of the velocity. The
DOI:10.1190/1.9781560803737

RMS velocity vRMS, or root-mean-square velocity, is the square root of

v2RMS . As such, vRMS has units of velocity. Actually, the RMS calculation
yields root-mean-square speed, not velocity. This is because velocity is a
vector quantity.
Equation (7.58) is the equation for a hyperbola. Equation (7.58) is the
same as equation (2.81) in Part 1. Note that t(0) ; t0 . This hyperbola,
depicted in Figure 2.19 in Part 1, is the time-distance curve for a single-
layered medium, with vertical two-way time t0 and constant velocity
vRMS. The given approximation is one in which the N-layered medium is
replaced by a single-layered medium whose velocity is given by the root-
mean-square velocity of the N-layered medium.

Normal moveout velocity

Before we proceed with our analysis, consider this analogy. Suppose we
are on a train riding on a railroad track that has experienced an earthquake.
At a juncture, the train is switched to a different, unaffected track. The train
conductor notices that the gauges of the two tracks are not quite the same.
One passenger is a thoughtful mathematician. He says that this is a
crucial approximation, so the train should not proceed until valid tests as
to safety have been completed. Another passenger is a sensible physicist.
He says there is only a small offset in the sizes of the gauges, so the train
should proceed. The mathematician asks: What do you mean by small?
The physicist might answer: I know by experience. According to Napoleon
Bonaparte, “Victory belongs to the most persevering.”
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

222 Basic Wave Analysis, Part 2: Raypath Analysis

The question of relative size may be kept in mind as we next consider a

common midpoint (CMP) gather. The raypaths for different offsets x gener-
ally do not reflect at the normal incidence point (NIP). Unfortunately, the
CMP rays do pass through the NIP. A geophysicist might say that this is
an approximation good for small offset distances, so we can assume that
all CMP rays do pass through the NIP. Associated with the reflecting inter-
face, there is a time-distance curve which appears as a hyperbolic-shaped
primary reflection event on the CMP gather. By means of the velocity spec-
trum, or by some corresponding curve-fitting method, we find the hyperbola
which best fits this empirical seismic event. We denote this hyperbola
(called the stacking hyperbola) by
x2
ts2 (x) = ts2 (0) + . (7.61)
v2s
In this equation, ts(x) denotes the best-fitting hyperbola, the variable x
DOI:10.1190/1.9781560803737

is the offset between source and receiver, and the numbers ts(0) and vs
are the two empirically determined constants which characterize the
hyperbola. The hyperbola ts(x) is the empirical time-distance curve, the
constant vs is called the (optimum) stacking velocity, and the constant
ts(0) is called the (optimum) stack time. In actual practice, the best-
fitting hyperbola will be fitted over some specified range of the actual
reflection event on the CMP gather instead of over its entire range.
Stacking velocity typically is associated with the given offset interval
from xmin to xmax.
Suppose that we, in fact, know the geologic structure of the earth. Then,
in principle, we can find an analytic expression for the true time-distance
curve in terms of the earth parameters. In other words, as long as the
layer interfaces and the layer velocities are known, the traveltime of a ray
and the intersection point of the ray with the earth’s surface are mathemat-
ical functions of the starting point and starting direction of the ray. By the
principle of reciprocity, a primary reflection produces a symmetric time-
distance curve on the CMP gather; that is, t(x) is always symmetric about
the zero-offset point x ¼ 0. We can express this symmetry mathematically
by the equation
t(x) = t(−x) . (7.62)
Here, x is the offset (i.e., shot to receiver distance in a CMP gather).
Thus, the time-distance curve can be expanded into a Taylor series in
even powers of x; that is,
t(x) = t(0) + a2 x2 + a4 x4 + · · · , (7.63)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 223

where a2 , a4 , . . . are the Taylor coefficients. The delta-t value, or normal

moveout (NMO) correction, is defined as
Dt = t(x) − t(0) . (7.64)

Suppose that a primary event and a multiple event occur on the CMP
gather at the same zero offset time t(0). Then, the difference in NMO of
these two events indicates how well the multiple can be suppressed by
stacking.
A more useful Taylor series expansion is the expansion of t 2(x); i.e.,
t2 (x) = t2 (0) + c2 x2 + c4 x4 + · · · , (7.65)

where c2 , c4 , . . . are the Taylor coefficients. In the case of a single constant

velocity layer with a flat interface, this expansion terminates after x 2, and we
have the familiar equation
DOI:10.1190/1.9781560803737

x2
t2 (x) = t2 (0) + , (7.66)
v21

where v1 is the velocity of the single layer. Equation (7.66) is the equation of
a hyperbola, and it represents the single most useful equation in understand-
ing the subject matter of reflection seismology. For any other subsurface
structure, the time-distance curve still is described as hyperbolic-like, but
in actuality the curve is higher-order and much more complicated. That
is why we must empirically fit the best-fitting hyperbola ts(0) in order
to grasp the time-distance curve t(x). In the case of flat constant-velocity
layers (with thicknesses h1 , h2 , . . . , hn and velocities v1 , v2 , . . . , vn , respect-
ively), the expansion (7.65) becomes
x2
t2 (x) = t2 (0) + + c4 x4 + · · · , (7.67)
v2RMS

where vRMS is the RMS velocity (for vertical travel path)

 
 
 2 N  1 N
vRMS =  vi hi =  v2 Dti (0) . (7.68)
t(0) i=1 t(0) i=1 i

In this equation, the two-way vertical traveltime is

N
t(0) = Dti (0) . (7.69)
i=1
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

224 Basic Wave Analysis, Part 2: Raypath Analysis

and the two-way vertical traveltime in layer i is

2hi
Dti (0) = . (7.70)
vi
If we truncate the series, we obtain the approximating hyperbola t(x) given
by the equation
x2
t2 (x) = t2 (0) + . (7.71)
v2RMS
Equation (7.71) is the same as equation (7.58) in the previous section.
Generally, we will denote the constant c2 in the expansion by 1/v2NMO , so
the truncated series is
x2
t2 (x) = t2 (0) + . (7.72)
v2NMO
DOI:10.1190/1.9781560803737

The constant vNMO is called the NMO velocity. As just demonstrated in the
case of flat constant-velocity layers, the NMO velocity and the RMS veloc-
ity are equal; i.e., vNMO ¼ vRMS. For general models, however, vNMO is
different from vRMS. The NMO correction is discussed in Part 1, Chapter 3.
What is the consequence of this assumption that the raypath strikes the
target at or near the NIP? Typically, questions of this sort are resolved by
trial and error as implemented by an iterative improvement algorithm.
Today, we have the mathematics and machines to handle immense problems
of this nature. And how did we reach this stage?
Gottfried Leibniz remarked, “It is unworthy of excellent men to lose
hours like slaves in the labor of calculation which could be relegated to
anyone else if machines were used.” Blaise Pascal (1623– 1662) created a
differential triangle, and Leibniz enlarged upon this concept to create calcu-
lus. Pascal invented a successful digital mechanical calculator (known as the
Pascaline) at age 19. Leibniz learned about the Pascaline and expanded upon
Pascal’s mechanism so it could multiply and divide. Leibniz’ machine
(called the step reckoner) was invented around 1672 and completed in
1694. Leibniz said that its purpose was to make calculations easy, fast,
and exact. Leibniz added that theoretically the numbers calculated might
be as large as desired, if the size of the machine was appropriately adjusted.
Today, the work of Pascal and Leibniz has come to fruition. Calculus
can create the mathematical edifice. The digital computer can complete
the calculations needed in the investigation of the mathematical fine
points. However, this process never ends. More calculus is needed to
create a mathematical edifice of the fine points, and more calculations
will allow us to investigate the finer points of the fine points. Often, it is
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 225

DOI:10.1190/1.9781560803737

Figure 7.24. Constant-velocity layers with dipping interfaces. The normal raypath
is shown, i.e., the raypath that strikes the target interface at right angles at point NIP.
The offset raypath SDR does not strike the target interface at NIP; however, the
approximation used in the text assumes that D and NIP are close together.

better to go as far as possible in creating the mathematical edifice, so that

less reliance has to be placed on the computer.
In the case of certain selected subsurface models, we can obtain math-
ematical expressions for the normal moveout velocity vNMO. Next, we will
find such an expression in the case of straight-line dipping interfaces.
Figure 7.24 shows the normal ray for interface 3 for the common midpoint
M. The downgoing leg of the normal ray coincides with its upgoing leg. The
upgoing leg of the normal raypath departs from interface 3 at right angles at
the NIP. The emergence angle u0 is shown, as well as the angles uIi and uTi at
each interface (i ¼ 1, 2), and the raypath segments si and velocities vi in each
layer (i ¼ 1, 2). Also shown is a typical offset ray for offset x. The offset ray
does not strike the reflecting interface at the NIP. However, for small values
of offset x, we will assume that the offset ray (for offset x) does approxi-
mately strike the interface at the NIP.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

226 Basic Wave Analysis, Part 2: Raypath Analysis

Now, we can add traveltime to this discussion. Let t(0) be the two-way
traveltime along the normal ray, so t(0)/2 is the one-way time. Let td (−x/2)
be the one-way time for the downgoing path of the offset ray with offset x,
and let tu (x/2) be the one-way time for the upgoing path of the same offset
ray. By the principle of reciprocity, we can reverse the direction of travel on
the downgoing offset path, and hence consider td (−x/2) as the time-distance
curve of an upgoing path from the NIP to S.
At this point, recall equation (7.25) from our previous section, Determi-
nation of wavefront curvature. For convenience, we repeat the equation
here:
 
dt  1 d 2 t 
t(x) = t(x0 ) +  (x − x0 ) + (x − x0 )2 + · · · . (7.73)
dx x0 2 dx2 x0

We must find expressions for tu (x/2) and td (−x/2). First, we will regard
DOI:10.1190/1.9781560803737

tu (x/2) as the time-distance curve of an emerging wave at the receiver point

R. At least approximately, this upgoing wave travels from a hypothetical
source at NIP to receiver R. The horizontal distance x/2 is the (positive) dis-
tance from midpoint M to receiver R. Previously, we have used equation
(7.73) to determine an expression for the time-distance curve of an emerging
ray. In the present context, the reference point x0 is zero, and the horizontal
distance is x/2, so the equation becomes
x sin u0 x cos2 u0 x2
tu = tu (0) + + +··· . (7.74)
2 v1 2 2r0 v1 2

Second, consider td (−x/2). We regard td (−x/2) as the time-distance

curve of an emerging wave at source point S. At least approximately, this
upgoing wave travels from a hypothetical source at NIP to source point S.
The horizontal distance −x/2 is the (negative) distance from midpoint M
to source S. Equation (7.73) becomes
x sin u0  x cos2 u0  x2
td = td (0) + − + − +··· . (7.75)
2 v1 2 2r0 v1 2

Note that both tu (x/2) and td (−x/2) are expressed as Taylor series cen-
tered at point M, and thus the emergent angle u0 and radius r0 of curvature
refer to the upgoing wave that emerges at point M. This reference upgoing
wave starts at NIP and travels to M along the normal raypath. We will call
this wave the NIP wave, and it is the one that leaves NIP at right angles to the
interface. The traveltime of the NIP wave is one-half of the two-way
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 227

traveltime t(0) of the normal raypath; that is,

1
td (0) = tu (0) = t(0) , (7.76)
2
so

td (0) + tu (0) = t(0) . (7.77)

As usual, we let t(x) denote the time-distance curve for the CMP gather
for the given reflecting interface. Let us assume that the reflecting point for
the source-receiver pair with offset x lies close to NIP. Then, we can write
(approximately)
 x x
t(x) = tu − + td . (7.78)
2 2
DOI:10.1190/1.9781560803737

If we carry out this summation, we obtain

cos2 u0  x2
t(x) = t(0) + +··· . (7.79)
r0 v1 2

If we square this equation, and drop higher-order terms, we obtain the hyper-
bola given by

cos2 u0 2
t2 (x) = t2 (0) + t(0) x . (7.80)
2r0 v1

Comparing equation (7.81) with equation (7.72), we see that the normal
moveout velocity for the straight-line dipping interface model satisfies

2r0 v1
v2NMO = . (7.81)
t(0)cos2 u0

Thus, vNMO depends upon the velocity v1 in the first layer, the angle u0 of
emergence of the NIP wave, and the radius r0 of curvature at the surface of
the NIP wave, i.e., the wave which originates at the NIP and travels up the
raypath of the normal ray. If we now express r0 in terms of the seismic
parameters vi, si, ai, bi along the normal ray, then we will have our required
expression for vNMO in terms of the seismic parameters. A worthwhile (but
not easy) endeavor would be to find a relationship between vNMO and vRMS
for dipping layers.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

228 Basic Wave Analysis, Part 2: Raypath Analysis

Figure 7.25.
Normal ray in a
model with
straight-line
interfaces.
DOI:10.1190/1.9781560803737

Even so, we can to find a relationship between vNMO and vRMS for
dipping layers using the following example and Figure 7.25. Let the
hypothetical source be at the NIP. The transmission laws in layers 3,2,1
are, respectively,
rI2 = s3
rI1 = s2 + rT2
r0 = s1 + rT1 (7.82)

and the refraction laws at interfaces 2 and 1 are, respectively,

v3 cos2 uT2
rT2 = r
v2 cos2 uI2 I2

v2 cos2 uT1
rT1 = r . (7.83)
v1 cos2 uI1 I1
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 229

Combining equations (7.82) and (7.83) in the sequential manner

rI2  rT2  rI1  rT1  r0, we obtain
 
v2 cos2 uT1 v3 cos2 uT2
r0 = s3 + s2 + s1
v1 cos2 uI1 v2 cos2 uI2
 
1 cos2 uT1 cos2 uT2 cos2 uT1
= v3 s3 + v2 s2 + v1 s1 . (7.84)
v1 cos2 uI1 cos2 uI2 cos2 uI1

This is the required expression for r0, and hence we have achieved
our objective for finding a mathematical expression for the normal
moveout velocity vNMO in terms of the seismic parameters along the
normal ray.
Let us check this formula in the case of flat layers. In this case, angles are
zero, and the travel path is vertical (so si ¼ hi). Then, we have
DOI:10.1190/1.9781560803737

1 1
r0 = [v3 s3 + v2 s2 + v1 s1 ] = [v3h3 + v2h2 + v1h1] , (7.85)
v1 v1
so

2r0 v1 2
v2NMO = 2
= [v1 h1 + v2 h2 + v3 h3 ]
t(0)cos u0 t(0)
1 2
= [v Dt1 (0) + v22 Dt2 (0) + v23 Dt3 (0)] , (7.86)
t(0) 1
which agrees with the conventional definition of RMS velocity (i.e., RMS
velocity along the vertical travel path). Note that Dt1(0) are two-way time
increments, so 2hi = v2i Dti (0).

Interval velocities
The layer velocities vi are called the interval velocities, and the main
objective of velocity analysis is to determine the interval velocities from
the seismic data. We will assume that the geology is described by a 2D
model with straight-line dipping interfaces, as shown in Figure 7.24. For
ease of exposition, we let the number of interfaces be N ¼ 3.
We are going to describe an algorithm for finding the interval velocities.
Before we present the algorithm, however, we can review some of the salient
properties of the model and show how the normal ray is propagated. The
normal ray represents a hypothetical wave with its source at the NIP. As
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

230 Basic Wave Analysis, Part 2: Raypath Analysis

shown in the figure, the normal ray leaves interface 3 at right angles. We can
trace the ray in terms of its radii of curvature, and the algorithm for propa-
gation from the NIP to the CMP is:

rI2 = s3

v3 cos2 uT2
rT2 = r
v2 cos2 uI2 I2

rI1 = rT2 + s2

v2 cos2 uT1
rT1 = r
v1 cos2 uI1 I1
r0 = rT1 + s1 . (7.87)
DOI:10.1190/1.9781560803737

The value r0 is the radius of curvature of the emergent wavefront at

the CMP.
We can simply invert algorithm (7.87) to obtain the following algorithm
for depropagation from the CMP to the NIP:

rT1 = r0 − s1

v1 cos2 uI1
rI1 = r
v2 cos2 uT1 T1

rT2 = rI1 − s2

v2 cos2 uI2
rI2 = r
v3 cos2 uT2 T2
s3 = rI2 . (7.88)

The normal ray is a ray associated with a particular interface.

The normal ray which we are considering is the one for interface 3; this
normal ray will be used to find the quantities v3, s3, and b2 ¼ uI2 as
follows. These three quantities are associated with layer 3; that is, s3 is
the velocity of layer 3, s3 is the length of the raypath segment in layer 3,
and b2 ¼ uI2 is the angle of incidence of the raypath segment in layer 3 strik-
ing the next interface (interface 2).
We assume that we already have determined the structure down to and
including interface 2. Then, the normal ray shown will be used to extend our
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 231

knowledge down to and including interface 3. Thus, the method is a recur-

sive one, proceeding one interface at a time, with the corresponding normal
ray used for each interface.
At the start of the algorithm, we know from the previous steps the values
of v1 and v2 and the positions of the interfaces 1 and 2. From the recorded
seismic data, we must determine certain basic parameters that we will
need to find out about layer 3. These parameters are the two-way traveltime
t(0) along the normal ray, the NMO velocity vNMO, and the angle of emer-
gence u0. Now, we will show how these parameters are measured from the
seismic data.
First, let us find t(0) and vNMO for interface 3. The seismic data used to
determine these parameters is the CMP gather for the CMP point in ques-
tion. We must identify the seismic reflection event corresponding to depth
point NIP, and then fit the best-fitting hyperbola to this event. We can
write this hyperbola as
DOI:10.1190/1.9781560803737

x2
t2 (x) = t2 (0) + , (7.89)
v2NMO

and from it we obtain the numerical value of t(0) and the numerical value of
vNMO. We keep in mind that this hyperbola refers to interface 3.
Next, let us find the angle u0 of emergence of the normal ray in question.
This ray arrives at the surface point which we have labeled CMP. Let the
horizontal coordinate of this point (with respect to some arbitrary origin)
be denoted by m0. We also must consider other CMP points (each with its
own CMP gather) in the neighborhood of m0. For each of these gathers,
we find t(0), as we did for the previous CMP gather in question. We will
label these values more explicitly as t(0, m), which represents a time-dis-
tance curve t as a function of horizontal coordinate m. This curve provides
the two-way traveltimes of the normal waves reflected from interface 3
resulting from many coincident source-receiver pairs. Instead of considering
two-way time, let us divide by 2 to obtain the curve (1/2)t(0, m), which pro-
vides one-way time. (We can obtain one-way time in this simple way
because the normal-reflected wave travels down and up on the same
raypath.) Thus, we may consider the event given by (1/2)t(0, m) as the
event due to the hypothetical NIP waves that originate at interface 3 and
move up the normal raypaths to the surface. Now, we have the data required
to determine the emergence angle u0.
Recall that Figure 2.16 shows the slope of the time-distance curve at a
point is equal to the value of Snell’s parameter (or ray parameter) p at that
point. The required slope of the (one-way) time-distance curve is
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

232 Basic Wave Analysis, Part 2: Raypath Analysis

(1/2)dt/dm evaluated at m0, that is, at the CMP point in question. The ray
parameter of the emergent normal ray is sin u0 /v1 , where v1 is the velocity
of the first layer. We will use the equation

1 dt(0, m) sin u0
= . (7.90)
2 dm m0 v1

Because we know v1, we can determine u0 immediately from this equation.

Thus, we have found t(0), vNMO, and u0 from the observed seismic data.
We have shown in the preceding section that
2r0 v1
v2NMO = , (7.91)
t(0)cos2 u0
which gives
t(0)cos2 u0 v2NMO
r0 =
DOI:10.1190/1.9781560803737

. (7.92)
2v1
We know all of the quantities on the right, so we immediately can calculate
the radius r0 of curvature of the NIP wave.
Now, we are ready to depropagate the NIP wave. Using the depropaga-
tion algorithm, we find
v1 cos2 uI1
rT2 = (r − s1 ) − s2 . (7.93)
v2 cos2 uT1 0
We know the emergence angle u0 and we know the position of interface
1. Thus, we can compute the raypath segment s1, as well as the angles uT1
and uI1 by Snell’s law. (Recall that we know both v1 and v2.) We know
the position of interface 2, so we can compute raypath segment s2 as well
as the angle uT2. Thus, we know all of the quantities on the right side of
the equation (7.93), so we can compute r T2
The question is: do we have enough information to depropagate further
to interface 3? The required equation is
v2 cos2 uI2
s3 = r . (7.94)
v3 cos2 uT2 T2
In this equation, we do not know uI2, s3, and v3. However, they are related.
Let us find the relationships. We know that
 
1
s3 = v3 Dt3 (0) . (7.95)
2
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 233

The factor 1/2 is due to the fact that we require one-way time. Here, the
difference of two-way traveltime for interfaces 3 and 2 is

Dt3 (0) = t(0)t2 (0) . (7.96)

Recall that we measured t(0) from the best-fitting hyperbola, and t2(0) can be
computed as

2s1 2s2
t2 (0) = + . (7.97)
v1 v2

Thus, we know the two factors on the right side of equation (7.96), from
which we obtain Dt3(0). Equation (7.95) says that s3 is equal to the now
known (1/2)Dt3 (0) multiplied by the unknown velocity v3. Next, let us
also relate uI2 to the unknown velocity v3. Snell’s law says that
DOI:10.1190/1.9781560803737

sin uT2 sin uI2

= , (7.98)
v2 v3

so

v23
cos2 uI2 = 1 − sin2 uI2 = 1 − sin2 uT2 . (7.99)
v22

Thus, we can substitute for s3 and cos2 uI2 in equation (7.94) and obtain
   
2 1 v2 rT2 v23
v3 Dt3 (0) = 1 − 2 sin uT2 ,
2
(7.100)
2 v3 cos2 uT2 v2

which is
 
Dt3 (0) rT2 tan2 uT2 v2 rT2
v23 + = . (7.101)
2 v2 cos2 uT2

We know all of the quantities in this equation. Thus, we can solve for v23 , and
let v3 be the positive square root. After we find v3, we immediately can find
 
1
s 3 = v3 Dt3 (0) . (7.102)
2

This allows us to proceed down the raypath leg a distance s3 in layer 3 to give
us the NIP point. Then, interface 3 is specified, as this interface is perpen-
dicular to the ray segment at the NIP. Therefore, we have accomplished
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

234 Basic Wave Analysis, Part 2: Raypath Analysis

Figure 7.26. Condensed depiction of an iterative improvement method

(see Figure 1.14 for a more detailed version).

our objective and extended our knowledge to interface 3. Furthermore, we

can repeat the algorithm as far as we want to any interface.
DOI:10.1190/1.9781560803737

To summarize, we have completed several tasks in this chapter. From

observed seismic data, we are able to use the methods of ray tracing to
depropagate the wave motion and so discover the boundaries and velocities
of the subsurface layers of sedimentary rocks. We have presented the geo-
metrical principles of seismic exploration based upon ray theory. A ray is
a line drawn in space corresponding to the direction of flow of seismic
energy. As such, it is a mathematical device rather than a physical entity.
However, many of the methods used to interpret and understand seismic
events are based on the geometry of rays. Ray theory has a long history
in seismology, going back nearly 100 years to the early days of earthquake
seismology. In exploration seismology, many important contributions have
been made in both the theory and applications of rays and wavefronts.

Verification
Francis Bacon maintained that, “Things alter for the worse spon-
taneously, if they be not altered for the better designedly.” The prefix allo
is a combining form indicating difference, variation, or opposition. The
word allothetic refers to information concerning an animal’s orientation in
an environment that is obtained by the animal from external spatial clues.
The prefix idio is a combining form meaning one’s own, private, personal.
The word idiothetic refers to information concerning an animal’s orientation
in an environment that is obtained by the animal by reference to its previous
orientations and movements, and without external spatial clues.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 7: Ray Tracing 235

A ship of old would try to stay close to land. The steersman would look
for allothetic clues, such as points of land and other ships. The steersman
would use these allothetic clues as feedback to continuously correct his
course. Suppose that the ship is alone at sea out of sight of land. The steers-
man is devoid of allothetic clues. Now the steersman can use no feedback, so
would resort to dead reckoning. What is dead reckoning? It is the use of idio-
thetic clues, such as starting point and measurements of direction, speed, and
time. The steersman must use these idiothetic clues as feedforward to con-
tinuously correct his course. We, too, can use these devices in our course of
study. Allothetic clues are external clues that can be used as feedback to
determine the correct action. Idiothetic clues are internal clues that can be
used as feedforward to determine the correct course. Dead reckoning is
subject to cumulative errors. Dead reckoning is the process of calculating
one’s current position using a previously determined position, and advan-
cing that position based upon known or estimated speeds over elapsed
DOI:10.1190/1.9781560803737

time and course. The corresponding term in biology, used to describe the
processes by which animals update their estimates of position or heading,
is path integration.
The chief characteristic of a global positioning system (GPS) is that
external signals are used. The GPS satellites circumnavigate the earth
twice each day transmitting signals to earth. The GPS device uses this infor-
mation as feedback to calculate the user’s location. In contrast, the chief
characteristic of the inertial navigation system (INS) is that no external
signals are used. It is a device that does dead reckoning for a moving
vehicle. Inertial navigation systems are used on aircraft, submarines,
guided missiles, and spacecraft. It only uses measurements provided by
its self-contained accelerometers and gyroscopes. By means of feedforward,
it tracks the position and orientation of the vehicle relative to a known start-
ing point, orientation, and velocity. Because the device makes use of no
external signal, it cannot be blocked by a hostile signal.
Often a system will use both GPS and INS. In GPS, satellite signals are
used to correct or calibrate the position given by an inertial navigation
system. Inertial navigation systems usually are subject to drift, so they
can provide accurate positions only for a short period of time. The GPS
gives an absolute drift-free position value that can be used to reset the
INS solution.
The velocity function is like a road map or chart. If the wave is at any
given point, then the velocity function tells the wave how to go to reach
the next point. We have the observed records and we wish to obtain the
correct velocity function. We start with an initial guess of velocity function
and compute the synthetic records. This part of the iterative improvement
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

236 Basic Wave Analysis, Part 2: Raypath Analysis

process involves no feedback (Figure 7.26). It only involves dead reckoning;

that is, the internal feedforward action of following the road map (i.e., the
velocity function). Then, we introduce the external observed records. This
outside information is used as feedback to correct the velocity function.
This process can be iterated to obtain a good estimate of the velocity
function.
Everything in this chapter on ray tracing comes under the heading of
dead reckoning. In navigation, dead reckoning is the determination
(without the aid of celestial observations) of the position of a ship from
the record of the courses sailed, the distance made (which can be estimated
from velocity), the known starting point, and the known or estimated drift.
Because the uncertainties of dead reckoning cumulatively increase over
time, celestial observations are taken intermittently to determine a more
reliable position (called a fix), from which a new dead reckoning is begun.
Well-log information can be used as fixes in seismic work. Dead reckoning
DOI:10.1190/1.9781560803737

is implanted in Kalman filtering methods. These methods combine a

sequence of navigation solutions to obtain the best estimate of the naviga-
tor’s current position and velocity. According to René Descartes, “We do
not describe the world we see, we see the world we can describe.”
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms
DOI:10.1190/1.9781560803737

Part 3: Waveform Analysis

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms
DOI:10.1190/1.9781560803737
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 8

Three Prototype Waves

“Reason is nothing without imagination.”

—René Descartes
DOI:10.1190/1.9781560803737

Pioneers of wave study and the invention of radio

The study of exploration seismology is founded upon many historical
legacies. Our science would not have progressed to the advanced level we
experience today without the contributions of countless visionaries and
scientists. As we begin the third part of this book on seismic waveforms,
we must return to history for insights on the cumulative knowledge base
to which we are indebted.
Samuel F. B. Morse (1791–1872), an accomplished portrait painter and
the inventor of the telegraph, was born in Massachusetts, the first child of a
Calvinist pastor and strong supporter of Puritan traditions. In 1844, Morse
sent the first telegraph message over a wire: “What hath God wrought?”
The invention was based on the telegraph key, a switching device to turn
the electric current off and on. In this way, a series of dots and dashes
could be transmitted over the wire. To send a message composed of
written words, the text was encrypted into Morse code (e.g., the letter “a”
is dot dash and the letter “n” is dash dot). One person sent the Morse code
over the electric wire; then, another person at the receiver converted the
series of dots and dashes back into text.
For centuries, the concept of the telephone evolved. Based on the idea of
two diaphragms connected by a taut string or wire, sound waves were
carried as mechanical vibrations along the string or wire from one dia-
phragm to the other. In the 19th century, several individuals experimented

239
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

240 Basic Wave Analysis, Part 3: Waveform Analysis

with ways to replace the string with a wire carrying electric current. The
transmission—a voice—was spoken into a movable disc that would make
and break the current; at a distance, another disc received vibrations, con-
verting them back into speech. The invention of the telephone was the
culmination of work done by many, and involved an array of lawsuits
founded upon the patent claims of several individuals and numerous com-
panies. Although the first telephone was invented by Antonio Meucci
(1808–1889), Alexander Graham Bell (1847–1922) was credited with the
development of the first practical telephone. In 1876, Bell made his first tel-
ephone call: “Mr. Watson, come here. I want to see you.” It was sent over a
wire to his assistant, Thomas Watson, who was in another room in the build-
ing. Similar to the telegraph, the telephone represented a new leap in elec-
trical telecommunication, accomplished by sending electricity over a wire.
But what of transmission without the wire? Although Maxwell’s theory
of electromagnetism, establishing the existence of electromagnetic waves
DOI:10.1190/1.9781560803737

moving at the speed of light, was not published until 1865, earlier inventors
experimented with electricity. One important invention was the capacitor—
a device for storing electrical energy, consisting of two conductors in close
proximity and insulated from each other. The first capacitor was the Leyden
jar, which stores static electricity between two electrodes on the inside and
outside of a glass jar. The Leyden jar was invented in 1745 independently
by Ewald Georg von Kleist (1700–1748) of Prussia and by Pieter van
Musschenbroek (1692–1761) of Leiden (Leyden), Netherlands, the city
for which it was named. Lord Kelvin (1824–1907) found that, under the
right conditions, a Leyden jar could be discharged so as to give alternating
current of very high frequency. In 1886, Heinrich Hertz (1857–1894), using
simple coils, transmitted electromagnetic waves and received them 1.5 m
away from the transmitter. Ironically, Hertz concluded, “I do not think
that the wireless waves I have discovered will have any practical appli-
cation.” In 1897, Guglielmo Marconi (1874–1937) attached a telegraph
key to the transmitter and thereby sent electromagnetic signals in Morse
code (dots and dashes) by wireless (that is, without a wire) over a distance
of 2.5 km.
Reginald Fessenden (1866–1932), a Canadian, was always fascinated
with the work of Bell. Similar to Christiaan Huygens’ inspiration for the
principle that bears his name, conceived after throwing a stone into a
canal and observing the resulting circular water waves, it was said that Fes-
senden threw a stone into a lake, examined the resulting circular water
waves, and realized that electromagnetic waves must radiate from the
antenna at the transmitting end. More specifically, Fessenden asserted that
the waves must keep going in a steady stream until they encircle the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 8: Three Prototype Waves 241

antenna at the receiving end in order to carry the entire frequency range of
voice sounds. In Marconi’s scheme, the waves would stop and go, stop and
go. Fessenden reasoned that continuous waves should be used for voice
transmission.
The idea of transmitting the human voice by wireless dominated the
early wireless experiments of Fessenden. He wanted to transmit speech
and music, and not plain text in the form of dots and dashes. To this
end, Reginald Fessenden invented heterodyning—a signal processing tech-
nique that shifts one frequency range into another frequency range. In other
words, Fessenden discovered how electromagnetic waves could be modu-
lated (known today as AM or amplitude modulation) so as to transmit
information such as speech and music. By means of heterodyning, electro-
magnetic continuous waves could be created, loaded with a voice message,
transmitted, and then eliminated, leaving only the voice message for the
listeners. Although other researchers still pursued the “stop-and-go” idea
DOI:10.1190/1.9781560803737

because they could not accept his heterodyning concept, Fessenden’s math-
ematical background made the abstract idea of heterodyning easy for him
to grasp.
In 1900, Reginald Fessenden set out to demonstrate that his direct-con-
version heterodyne receiver could make continuous wave radiotelegraphy
signals audible. Fessenden set up his transmitter at Cobb Island in Maryland.
He also set up a receiving station in Virginia, 50 miles away, operated by his
assistant, Alfred Thiessen. On December 23, 1900 Fessenden spoke into his
microphone, “One, two, three, four. Is it snowing where you are, Mr. Thies-
sen? If so, telegraph back and let me know.” Thiessen replied by telegraph in
Morse code that indeed it was snowing. Fessenden sat at his desk and wrote,
“This afternoon here at Cobb Island, intelligible speech by electromagnetic
waves has for the first time in World’s History been transmitted.” Thus, Fes-
senden invented “the radio,” which we know today, and he has been credited
as the inventor of AM technology. The original spark-gap radio transmitters
were impractical for transmitting audio, because they produced discontinu-
ous pulses known as “damped waves.” Fessenden realized the need for a
new type of radio transmitter that produced steady “undamped” (better
known as “continuous wave”) signals, which then could be “modulated”
to reflect the sounds being transmitted.
Nikola Tesla (1856–1943) was an engineer known for designing the
alternating-current (AC) electric system, which remains the predominant
electrical system used throughout the world today. He also created the
“Tesla coil,” which continues to be used in radio technology. Although
Tesla failed to accomplish his dream of wireless radio, he saw it accom-
plished through the utilization of the system he envisioned. Between 1890
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

242 Basic Wave Analysis, Part 3: Waveform Analysis

and 1895, Tesla built the high-frequency alternators which produced up to

20 kHz; then, Fessenden was the first to demonstrate that such equipment
could produce the required quiet carrier wave for voice modulation. Thus,
Fessenden avoided the piercing background roar of the damped wave spark
and arc transmitters that other inventors were using. Fessenden agreed with
Tesla that the discontinuous spark-gap wave transmitters were unacceptable
and that radio could only succeed with the use of continuous wave generators.
Fessenden knew how to transmit voice and music by means of continuous
waves, but no one else understood AM radio transmission.
Indeed, amplitude modulation continues to be used worldwide, primar-
ily for medium wave (also known as “AM band”) transmissions, but also on
the longwave and shortwave radio bands. Recent technological advances
have led to digital audio broadcasting (DAB). Digital is nothing more
than an off–on technique. It is ironic that DAB is more efficient in its use
of the spectrum than analog (i.e., Fessenden’s continuous wave) radio,
DOI:10.1190/1.9781560803737

and thus offers more radio services for the same given bandwidth.
Contemporaneous with Fessenden, Marconi accomplished his greatest
achievement in 1901, when he sent a message in Morse code from
England to Newfoundland, Canada. This one-way transatlantic transmission
won him worldwide fame. Ironically, detractors of the project were correct
when they declared that radio waves would not follow the curvature of the
earth, as Marconi believed. In fact, the raypath of Marconi’s transatlantic
radio signal left England headed into space. It was reflected off the iono-
sphere and bounced back down to Newfoundland. Instead, Fessenden per-
fected equipment similar to Marconi’s but more reliable.
To begin conducting transatlantic transmissions, Fessenden built a wire-
less station at Brant Rock, Massachusetts, with an antenna tower 128 m tall
(the height of a 43-story building), and built another station in Scotland.
Then, in January 1906, he achieved the first two-way transatlantic trans-
mission, exchanging Morse code messages between the two stations. Fur-
thermore, the implementation of Fessenden’s system could be put to
many uses, including the promotion of safety at sea, as was demonstrated
by his establishment of wireless stations for the United Fruit Company in
New Orleans, on their ships, and at their plantations in Guatemala.
In addition to Morse code transmissions, Fessenden pioneered voice
transmission (the radio). In June 1906, Fessenden built a small testing
station at Plymouth, Massachusetts, 18 km from Brant Rock. Fessenden
and his assistants conversed regularly by voice transmission (radio)
between these two Massachusetts stations, and, in November 1906, they
confirmed that their conversations also were heard in Scotland; thus, the
first human voice to be transmitted across the Atlantic was that of
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 8: Three Prototype Waves 243

Fessenden’s operator at Brant Rock, Mr. Stein. On Christmas Eve, 24

December 1906, Fessenden transmitted from Brant Rock station the first
audio radio broadcast of music and entertainment in history. Radio operators
on ships in the Atlantic heard the broadcast that included Fessenden playing
the song “O Holy Night” on the violin and reading a passage from the Bible.
This broadcast also was received by listeners on the East Coast of the United
States, who sent him congratulatory messages. For this accomplishment,
Fessenden was using a high-frequency alternator which he built based
upon Tesla’s design.
Thus, Fessenden should be remembered as the first to discover how elec-
tromagnetic waves could be modulated in order to transmit information such
as speech and music. Fessenden validated his discovery by building a direct-
conversion heterodyne receiver that made the modulated electromagnetic
signals audible to the human ear. Unfortunately, application of his receiver
was scant because of its local oscillator’s stability problem; a stable yet inex-
DOI:10.1190/1.9781560803737

pensive local oscillator was not available until Lee de Forest (1873–1961)
invented the triode vacuum tube oscillator. So, while Samuel Morse, Alex-
ander Graham Bell, and Guglielmo Marconi received full credit for their
inventions even though great improvements were later made by others,
Lee de Forest received complete credit for the invention of the radio and
Fessenden received none. Through years of successes as well as failures,
Fessenden was relatively unknown by the time of his death. At his gravesite,
a stone was engraved, “By his genius distant lands converse and men sail
unafraid upon the deep.”

Early days of seismic exploration

Although the general public may be unaware of the great minds who
have been pioneers in the development of the field of seismic exploration,
experts in our profession are not. Indeed, the Society of Exploration Geo-
physicists (SEG) has established The Reginald Fessenden Award, which
recognizes an individual for a specific technical achievement, such as an
invention or a theoretical or conceptual advancement, which has made a sig-
nificant contribution in exploration geophysics.
In addition to his work with radio transmission, Fessenden conducted
experiments which relate directly to techniques we use today. In 1913,
while studying iceberg detection and measurement of ocean depths, Fessenden
invented a seismic instrument that was used to record both refracted and
reflected waves. Using a mechanical oscillator and microphones, his
device measured deviations in reflection and refraction waves to identify
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

244 Basic Wave Analysis, Part 3: Waveform Analysis

anomalies in the ground. He patented both refraction and reflection tech-

niques for use in locating geologic formations.
Ludger Mintrop (1880–1956), a German mine surveyor and geophysi-
cist developed his own seismic refraction field instrument. In his honor,
the European Association of Geophysicists and Engineers (EAGE) estab-
lished The Ludger Mintrop Award. Mintrop’s system consisted of a vertical
pendulum (the geophone) and a photographic unit (the recording instru-
ment). The mass of the pendulum was a 4 kg lead ball suspended elastically
from a laminated spring. As a result of its inertia, the mass of the ball
remained stationary when seismic waves affected the housing, causing the
housing to vibrate. The relative movement between the static mass and
the moving housing was amplified and recorded. This ingenious system
was composed of laminated springs, mirrors, a directing magnet that
returned the mirror to its resting position after tilting, a convex lens, and a
light recorder.
DOI:10.1190/1.9781560803737

On 4 April 1921, Mintrop founded the exploration company SEISMOS,

appealing at first to mining companies interested in the instrument as a means
of detecting ore and coal deposits. In 1923, SEISMOS was hired by the
Mexican Eagle Oil Company (El Aquila), which had been created by the
legendary Lord Cowdray and became a subsidiary of Shell in 1918; a
crew, led by Otto Geussenhainer, was sent to the jungle between Tampico
and Veracruz in Mexico. To begin, the technical equipment consisted of
one Mintrop seismograph, one photographic recorder, and one observation
tent, but soon two more seismographs were added to the crew (one of
these systems remains on display at the Deutsches Museum in Munich,
Germany). The company’s second crew also began working on the American
continent in 1923, led by Mintrop. Though at first hired by Marland Oil in
Oklahoma, soon the crew moved to the Gulf coast of Texas and Louisiana
seeking to use the seismic reflection method to expose salt domes for the
Gulf Production Company. In 1924, Geussenhainer’s crew left Mexico for
Texas and, in a personal letter, Geussenhainer wrote, “My English was not
so good at the time but I could tell that there were a number of doubters in
the Gulf organization.” In fact, skepticism of the new technique was con-
siderable and the Mintrop seismograph was seen as little more than a new
type of divining rod. Yet, on 20 September 1924, Mintrop and Geussenhainer
fixed a point “K” on the map of the area in Fort Bend County under investi-
gation and proposed to Gulf that they drill a test well exactly at that spot.
On 19 November 1924, the well hit caprock at a depth of 110 m, almost
the exact depth Mintrop had predicted; in December, the well came in as
an oil producer. After this discovery, attitudes changed dramatically and
Mintrop established the relevance of seismic exploration.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 8: Three Prototype Waves 245

Consequently, many more individuals played a role in the exploitation

of seismic exploration as a scientific and practical endeavor. As stated in
Lawyer et al. (2001), “Through his Cambridge University contacts, Everette
DeGolyer had heard of Reginald Fessenden and subsequently had many dis-
cussions with him in Boston during the early 1920s about using seismic
waves for locating ore bodies and oil fields. As a consequence, DeGolyer
organized the Geophysical Research Corporation (GRC) in 1925 as an
Amerada subsidiary and hired J. C. Karcher, as GRC’s new vice president,
to acquire a license from Fessenden for Amerada and GRC to use the funda-
mental Fessenden patent for ore body location issued eight years before.”
Now recognized by an SEG award in his name, J. Clarence Karcher
began the construction of a set of seismic (reflection) instruments that
deployed advances in vacuum tubes and radio. Indeed, Karcher and
others, notably E. A. Eckhardt and Burton McCollum, later improved the
seismic reflection method and made its application more practical. It was
DOI:10.1190/1.9781560803737

from these beginnings that the study of seismic waves could proceed.

Seismic waves
Seismic waves are mechanical waves, meaning that these waves involve
the actual motion of the rock particles. Seismic waves that travel through the
entire body of the rock are called body waves. In contrast, seismic surface
waves travel close to the surface rather than through the rock. Seismic
body waves are of two fundamental types: longitudinal and transverse. In
a longitudinal wave, the oscillating particles of the medium are displaced
parallel to the direction of propagation (i.e., the direction of energy trans-
mission) of the wave. In a transverse wave, the particles are displaced in a
direction perpendicular to the propagation direction. When a steel rod is
struck on one end by a hammer, a wave pulse in the form of a longitudinal
compression of the rod travels down its length. If the rod is struck period-
ically, a succession of such pulses, known as a wave train, travels down
the rod. Sound propagates through the air as longitudinal waves. In contrast,
the familiar water waves that come from the point where a stone is dropped
into a quiet pond are transverse waves. Furthermore, because these waves are
confined to the water layer close to the surface, they are surface waves rather
than body waves. Importantly, various types of seismic surface waves also
can be identified on seismograms. Surface waves usually travel more
slowly and have larger amplitudes and longer wavelengths than body waves.
The phenomena perceived by our eyes as light and with our ears as
sound are propagated as wave motion. Their motion occurs not in the
two-dimensional (2D) surface of a pond but in three-dimensional (3D)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

246 Basic Wave Analysis, Part 3: Waveform Analysis

space. However, many of the properties of all wave motion can be observed
by studying the familiar water waves. Of course, it should be noted that there
are fundamental differences between 2D water waves and 3D light and
sound waves. The water waves that are produced by the stone move out
in circular rings at a constant speed. This wave speed is called the velocity
of propagation, which is denoted by v. The waves themselves have crests
and troughs, that is, points where the water level is elevated and points
where the level is depressed. The water surface undulates rhythmically
between crest and trough, crest and trough. The distance between successive
crests, or alternatively the distance between successive troughs, is called the
wavelength; typically, it is denoted by the Greek letter l (lambda). As the
waves travel past a fixed point on the surface of water, they cause a vertical
up-and-down motion of the water at this given point. This up-and-down
motion repeats itself in time in a periodic manner. The number of times
per second that this up-and-down motion repeats itself is called the fre-
DOI:10.1190/1.9781560803737

quency of the wave, which is denoted by f or by the Greek letter n (nu).

(Note that we prefer to use f rather than n for frequency, because the
Greek letter n is too similar to the Roman letter v, normally used for veloc-
ity.) Three fundamental aspects of wave motion are:
1. the velocity v with which the wave travels onward or propagates;
2. the wavelength l, which is the distance between crests (or between
troughs); and
3. the frequency f with which the wave pulsates.

We assume an isotropic medium, in which case the rays are orthogonal to

the wavefronts (see Figure 8.1). Basic raypath analysis is based upon
Fermat’s principle of least time. The raypath is the path that takes the least
time for the particle to go from one point to next point. Basic waveform
analysis is based upon Huygens’ principle. A traveling wavefront is the com-
position of spherical wavelets emanating from each point of the prior wave-
front. A wave equation governs the corresponding wave motion. There are
many types and forms of wave equations; this book makes use of the
classic scalar wave equation with the parameter v, the velocity of the travel-
ing wave. For ray theory, we use the simplification of the wave equation
known as the eikonal equation. For wavefront theory, we use the wave
equation itself. The implementation of ray analysis takes much less computer
power than the implementation of wavefront analysis. For this reason, much
of seismic processing is historically based upon ray analysis and interpret-
ation. With the advent of the mammoth acquisition systems and computers
available today, attention has turned to full waveform inversion (FWI).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 8: Three Prototype Waves 247

DOI:10.1190/1.9781560803737

Figure 8.1. An orthogonal network illustrating one set of curves which are
raypaths (as discussed in Part 2) and another set which are wavefronts (as will be
discussed in this part).

Full waveform inversion (FWI) methods are among the most recent
techniques for geotechnical site characterization and are still under continu-
ous development. The method is fairly general and is capable of imaging the
arbitrarily heterogeneous compressional and shear wave velocity profiles
of the earth. Two key components are required for the profiling based on
FWI. These components are: 1) a computer model for the simulation of
elastic waves in semi-infinite domains; and 2) an optimization framework,
through which the computed response is matched to the measured response,
via iteratively updating an initially assumed material distribution for the
earth. The computer model makes use of wave equations that are able to
incorporate complications such as anisotropy and various types of waves.
Although ray theory can address anisotropy and shear waves (to a certain
degree), wavefront analysis permits more complicated scenarios.
The true origin of wave analysis began with Pythagoras (6th century
BC) and the school he founded in ancient Greece. The fundamental differ-
ence between the school of Pythagoras and other scholarly institutions of the
time was that Pythagoras sought to educate not only his followers but also
the population in general. To Pythagoras, the universe was not chaotic,
but contained a deep underlying order which pervaded the visible world
and the minds of men. He examined the order of the universe as seen in
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

248 Basic Wave Analysis, Part 3: Waveform Analysis

the motions of the heavens, mathematics, music, and science. Pythagoras

observed that, when the blacksmith struck his anvil, different notes were
produced according to the weight of the hammer. The number (in this
case, amount of weight) seemed to govern musical tone. He examined the
wave motion on stretched strings, as found on a lyre. Again, number (in
this case, length of string) seemed to govern musical tone.
In 1747, Jean-le-Rond d’Alembert applies differential calculus to the
problem of a vibrating string, deriving the one-dimensional wave equation:

∂2 u(x, t) 1 ∂2 u(x, t)
= , (8.1)
∂x2 v2 ∂t2
in which the constant v is the wave’s phase velocity. Then, d’Alembert
solves equation (8.1) in the case of v = 1. His solution employs an especially
simple form in the case of zero initial velocity. His proof begins by
saying that, if ∂u/∂x is denoted by p and ∂u/∂t by q, then du is the exact
DOI:10.1190/1.9781560803737

differential
du = p dx + q dt . (8.2)

In these quantities, the wave equation (8.1) becomes

∂q ∂p
= . (8.3)
∂t ∂x
Therefore,
dw = p dt + q dx (8.4)

is also an exact differential. Therefore,

du − dw = p dx + q dt − p dt − q dx = (p − q)(dx − dt) ,
(8.5)
du + dw = p dx + q dt + p dt + q dx = (p + q)(dx + dt) .

Thus,

d(u − w) = (p + q) d(x − t)
(8.6)
d(u + w) = (p + q) d(x + t) .

Therefore, u − w must be a function of x − t and u + w must be a function of

x + t, and we also may establish that
u − w = 2f (x − t), u + w = 2g(x + t) . (8.7)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 8: Three Prototype Waves 249

Through addition, we can eliminate the w terms, leaving

u(x, t) = f (x − t) + g(x + t) , (8.8)

where f and g are arbitrary functions. Equation (8.8) is the d’Alembert sol-
ution of the wave equation (8.1) with v = 1. If v is an arbitrary constant, then
the d’Alembert solution becomes

u(x, t) = f (x − vt) + g(x + vt) . (8.9)

The two terms on the right side of equation (8.9) have the following
interpretation. The term f (x − vt) represents a waveform f moving with ve-
locity of magnitude v in one direction with no change in size or shape. The
term g(x + vt) represents a similar undistorted waveform g moving with ve-
locity of magnitude v in the opposite direction.
A stretched string in a vacuum represents an authentic one-dimensional
DOI:10.1190/1.9781560803737

problem. If we assume that the string is perfectly elastic so that no energy is

lost to heat, then all of the energy remains on the string. Let us examine more
closely the first term in equation (8.9), i.e.,

u(x, t) = f (x − vt) . (8.10)

For convenience of presentation, we shall assume that v is a positive con-

stant. We want to reveal the significance of velocity v.
At time t = t0 = 0, function (8.10) defines the curve u(x, t0 ) = f (x), as
shown in Figure 8.2.
At time t = t1 . 0, function (8.10) defines the curve u(x, t1 ) =
f (x − vt1 ), as shown in Figure 8.3. The curves f (x) and f (x − vt1 ) are iden-
tical in shape. The latter is shifted to the right by a distance equal to vt1 .
At time t = t2 , 0, function (8.10) defines the curve (x, t2 ) = f (x − vt2 ),
as shown in Figure 8.4. The curves f (x) and f (x − vt2 ) are identical in shape.
The latter is shifted to the left by a distance equal to vt2 .

Figure 8.2. The

wave function
u(x, t0 ) = f (x − vt0 )
at time t = t0 = 0.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

250 Basic Wave Analysis, Part 3: Waveform Analysis

Figure 8.3. The

wave function
u(x, t1 ) = f (x − vt1 )
at time t = t1 . 0.

Figure 8.4. The

wave function
u(x, t2 ) = f (x − vt2 )
at time t2 , 0.
DOI:10.1190/1.9781560803737

In any case, the waveform moves along the x-axis (without distortion)
by a distance vt in time t. The velocity with which the wave moves is,
therefore,
vt
propagation velocity = = v . (8.11)
t
In summary, the one-dimensional wave equation produces a waveform
function (8.10) whose propagation is explained by simply shifting the func-
tion f (x). That is, the wave does not change its shape as it travels. In terms of
the physics of wave propagation, this means that the wave travels in a
medium whose physical properties are homogeneous, isotropic, and nondis-
persive. In terms of the wave equation, this means that the velocity v is inde-
pendent of the spatial position x (i.e., the medium is homogeneous), the
physical properties governing v are the same in all directions (i.e., the
medium is isotropic), and that v is independent of the wave’s frequency
(i.e., the medium is nondispersive).
The remainder of this book will be devoted to a study of the wave
equation. The three prototype waves are plane waves, cylindrical waves,
and spherical waves. The fundamental difference is that the spatial function
of the amplitude is planar in the first case, cylindrical in the second case, and
spherical in the third case. The intensity is independent of distance traveled
in a plane wave. The intensity varies approximately as the inverse distance
in the case of the cylindrical wave. The intensity varies as the inverse square
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 8: Three Prototype Waves 251

of distance in the case of the spherical wave. These statements about inten-
sity hold only in a perfectly lossless medium. A point source produces a
spherical wave in an isotropic homogeneous medium. If that source is at
an infinite distance away, then it appears as a plane wave.

Plane waves
A 1D wave is a wave that propagates along a straight line, which we may
take to be the s-axis. A 3D form of a sinusoidal plane traveling wave can be
obtained by a rotation of the coordinate system used to describe a 1D sinu-
soidal plane traveling wave. However, more elements are implied by the
change of variables than only the extra dimensions. Quantitative new fea-
tures appear because of the new degrees of freedom resulting from the
increased dimensionality. For example, in three dimensions we can have a
wave that is a pure traveling wave in one dimension, a pure standing
DOI:10.1190/1.9781560803737

wave in another, and an exponential (i.e., evanescent) wave in the third

dimension.
As depicted in Figure 8.5, let a sinusoidal plane wave propagate in the
direction of the unit vector es along the s-axis. Suppose that, on the plane
s = 0, the wave function u(s, t) is

u(0, t) = A cos vt , (8.12)

Figure 8.5. Sinusoidal plane wave propagating in the direction given by the s-axis
(dashed lines indicate wavefronts at the crests).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

252 Basic Wave Analysis, Part 3: Waveform Analysis

where A is a constant. At the plane defined by s = constant, the wave

function is

u(s, t) = A cos(vt − ks) . (8.13)

Such a plane (s = constant) represents a wavefront. Now, we want to

express the wave function as u(x, y, z, t), where (x, y, z) represents a
general Cartesian coordinate system, instead of u(s, t), where s is the coor-
dinate along the propagation direction.
We let the origin (0, 0, 0) of the (x, y, z) system be in the plane s = 0.
Let ex , ey , ez be the three unit vectors in the (x, y, z) coordinate system. Let
the vector from the origin (0, 0, 0) to the spatial point (x, y, z) be denoted as
r = x ex + y ey + z ez . (8.14)
The plane (or wavefront) s = constant is described in the (x, y, z) system as
DOI:10.1190/1.9781560803737

the plane
s = r · es = constant , (8.15)
where es is the unit vector along the s-axis. Thus, the quantity ks can be
written as
ks = k (r · es ) = (k es ) · r . (8.16)
The vector k es is defined as the propagation vector k; i.e.,
k = k es . (8.17)
Thus, the magnitude of k is k and the direction of k is the direction of the unit
vector es along the wave propagation axis s. If we let the components of k
be kx , ky , kz so

k = kx ex + ky ey + kz ez , (8.18)

then we see that the quantity ks is

k s = k · r = kx x + ky y + kz z . (8.19)

The wave function is, therefore,

u(s, t) = u(x, y, z, t) = A cos(vt − kx x − ky y − kz z) . (8.20)

To simplify our diagrams, we can work with two dimensions (x, z)

instead of three. We let x be the horizontal coordinate and z be the vertical
coordinate (with the positive direction down). The physical meaning of the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 8: Three Prototype Waves 253

DOI:10.1190/1.9781560803737

Figure 8.6. Plane wave propagating in direction of the s-axis.

wavenumber k is the number of radians of phase-lag per unit displacement

along the propagation direction es (see Figure 8.6). It follows that ks is the
phase-lag accumulated in a distance s. The physical meaning of the com-
ponent kx (called the horizontal wavenumber) is the number of radians of
phase-lag per unit displacement along the x-axis, i.e., along the unit
vector ex . A similar meaning can be ascribed to kz .
Suppose the propagation vector makes an angle u with the z-axis. Let l
denote the wavelength in the direction of propagation. If we advance along
the s-axis by a distance of l (at a fixed time t), the phase-lag increases by
one cycle, or 2p radians. On the other hand, if we advance along the x
axis, we travel a distance of l/sin u before s increases by one wavelength
l. That is, the wavelength in the x direction is
l
lx = . (8.21)
sin u
The phase-lag has increased by 2p in a distance of lx along the x-axis.
Because lx is larger than l by a factor 1/sin u, it follows that the increase in
phase-lag per unit distance along x is smaller than k by the factor sin u. That is,
kx = k sin u . (8.22)
◦
In the example shown in Figure 8.7, we have u = 60 and k = 2p(3), so
◦
kx = 2p(3) sin 60 = 2p(3)(0.866) = 2p(2.598) . (8.23)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

254 Basic Wave Analysis, Part 3: Waveform Analysis

DOI:10.1190/1.9781560803737

Figure 8.7. Illustration of the horizontal wavenumber kx = 2p (2.598). There are

3(0.866) = 2.598 cycles or 2p (2.598) radians along the unit vector ex . (Note that,
in the direction of propagation, there are three cycles in one unit of distance, or
equivalently 2.598 cycles in 0.866 units of distance.)

Similarly, we have
kz = k cos u . (8.24)
Therefore, we have
◦
kz = 2p(3) cos 60 = 2p(3)(0.5) = 2p(1.5) . (8.25)

Figure 8.8 depicts the propagation vector. Because kx = k sin u and

kz = k cos u, we see that kx and kz are the projections of the vector k on the
x and z axes, respectively. That is, if we take the projection of the vector k
along the vector ex , we obtain

k · ex = kx , (8.26)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 8: Three Prototype Waves 255

Figure 8.8.
Components kx
and kz of
propagation
vector k.

a number less than the magnitude of the vector by a factor equal to sin u. A
similar condition holds for kz . These conditions ensure that the sum of
DOI:10.1190/1.9781560803737

squares of the components equals the square of the magnitude; i.e.,

kx2 + kz2 = k2 . (8.27)

As we have shown, kx and kz are the legs of a right triangle with hypotenuse k.
It is important to remember that such a Pythagorean relationship does not
hold for horizontal wavelength lx , vertical wavelength lz , and wavelength
l. That is, lx and lz cannot be the components of a “wavelength vector” as
would be defined as l = les . Such a wavelength vector is never used. Let
us give the reason. We recall that the wavelength l is defined as the distance
between the wave crests for displacement along the s-axis.
Also, we know that lx is the distance between wave crests for displace-
ment along the x-axis. In Figure 8.9, we see that
l l
lx = and lz = . (8.28)
sin u cos u

Suppose the direction of propagation is not along a coordinate axis (so u is

not equal to 0 or 2p). In such a case, each of lx and lx is greater than or equal
to l, and hence they cannot be components of l. For this reason, a wave
length vector is never used. The propagation vector k is the correct vector
to use in wave propagation problems.
For reference, we will express the relationships between wavenumbers
and wavelengths; these fundamental relationships are

kx = k sin u, kz = k cos u , (8.29)

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

256 Basic Wave Analysis, Part 3: Waveform Analysis

DOI:10.1190/1.9781560803737

Figure 8.9. Values lx and lz .

l l
lx = , lz = , (8.30)
sin u cos u

2p 2p 2p
k= , kx = , kz = . (8.31)
l lx lz

The Pythagorean equation for the wavenumber is

k2 = kx2 + kz2 . (8.32)

An interesting relationship exists between lx , lz , and l. A well-known

Euclidean theorem (i.e., the reciprocal Pythagorean equation, see Figure
8.10) states that, if h is the altitude on the hypotenuse of a right triangle
whose legs are a and b, then

1 1 1
= + . (8.33)
h2 a2 b2
It follows that

1 1 1
2
= 2+ 2 . (8.34)
l lx lz
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 8: Three Prototype Waves 257

The sinusoidal plane wave given by

u(x, z, t) = A cos(vt − ks) (8.35)

can be written as
u(x, z, t) = A cos(vt − k · r) . (8.36)

The argument of sinusoidal wave function is called its phase. It is

f(x, z, t) = vt − k · r = vt − kx x − kz z . (8.37)

At fixed time t, the locus of points of equal f defines a line called a wave-
front. Thus, a wavefront is given by the condition
d f = d(vt − k · r) = v dt − k · dr = 0 (8.38)

for fixed t. Thus, dt = 0, so d f = −k · dr. We recall that

DOI:10.1190/1.9781560803737

r = x ex + z ez ,

so
dr = dx ex + dz ez . (8.39)

Thus, the wavefront is given by

k · dr = 0 (8.40)

or
(kx ex + kz ez ) · d(x ex + z ez ) = kx dx + kz dz = 0 . (8.41)

In three dimensions, of course, this equation would represent a plane. This is

why such a wave is called a plane wave.
The general expression for a plane wave in 2D is
 
x sin u z cos u
u(x, z, t) = f t − − , (8.42)
v v

where f represents an arbitrary waveshape and u is the angle between the

s-axis (the axis of propagation) and the z-axis. The rays are parallel to the
s-axis and the wavefronts are perpendicular to the s-axis. A wavefront is
a plane of constant phase; i.e.,

x sin u z cos u
t− − = constant , (8.43)
v v
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

258 Basic Wave Analysis, Part 3: Waveform Analysis

Figure 8.10. The

structure of the reciprocal
Pythagorean equation.

for a fixed value of t. Differentiating this equation, we find

sin u dx + cos u dz = 0 , (8.44)

DOI:10.1190/1.9781560803737

so the slope of the wavefront is

dz
= − tan u . (8.45)
dx
Thus, the wavefront makes an angle of 2u with the x-axis. Any point on
a given wavefront corresponds to the same value of time t. The time incre-
ment between two wavefronts is

sin u cos u
Dt = Dx + Dz , (8.46)
v v
which can be written as

Dx Dz
Dt = + , (8.47)
vx vz
where
v v
vx = and vz = . (8.48)
sin u sin u
are called the horizontal phase velocity and the vertical phase velocity,
respectively.
Now, Figure 8.11 depicts the time separation of the two wavefronts as
Dt. The energy propagates along the rays at velocity v, so the distance
along a ray between the two wave crests is v Dt. Consider any two points
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 8: Three Prototype Waves 259

DOI:10.1190/1.9781560803737

Figure 8.11. Wave crests and rays.

(the two black circles) on the respective wavefronts, e.g., (x, z) and
(x + Dx, z + Dz). From the geometry, we have

v Dt = Dx sin u + Dz cos u . (8.49)

The time increment Dt between the two points is equal to the sum of two com-
ponents, i.e., the horizontal distance Dx divided by the horizontal phase ve-
locity v/sin u plus the vertical distance Dz divided by the vertical phase
velocity v/cos u. However, energy does not travel along such a path, but
instead travels on the ray path from (x, z) to (x + v Dt sin u, z + v Dt cos u).
The time increment Dt can be obtained by the same formula (8.49),
because Dt is equal to the sum of the horizontal distance v Dt sin u divided
by v/sin u plus the vertical distance v Dt cos u divided by v/cos u.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

260 Basic Wave Analysis, Part 3: Waveform Analysis

Cylindrical waves
Figure 8.12 depicts a cylindrical wave. The coordinate z is equal to
the height of P above the xy-plane. Let the point Q be the projection of P
to the xy-plane. Then, the coordinate f is the angle between the positive
x-axis and the line from the origin to Q. The coordinate r is the distance
from Q to the origin. With this established, now we can briefly examine
another idealized waveform: the infinite circular cylinder. The cylindrical
coordinates are given by

x = r cos f, y = r sin f, z = z . (8.50)

The simple case of cylindrical symmetry requires that

u(r, t) = u(r, f, t) = u(r, t) . (8.51)
DOI:10.1190/1.9781560803737

The f-independence means that a plane perpendicular to the z-axis will

intersect the wavefront in a circle of radius in r. In addition, the z-indepen-
dence further restricts the wavefront to a right circular cylinder centered on
the z-axis and having infinite length. The solution of the wave equation is in
the form of a Bessel function, which for large values of r asymptotically
approach simple trigonometric forms. Finally, when r is sufficiently large,
we can write
A A
u(r, t) ≈ √ eik(r+vt) or u(r, t) ≈ √ cos k(r + vt) . (8.52)
r r
Equations (8.52) represent a set of coaxial circular cylinders filling all
space and traveling toward or away from an infinite line source. In contrast

Figure 8.12. Illustration of

the definition of cylindrical
coordinates r, f, z of a
point P.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 8: Three Prototype Waves 261

to spherical and plane waves, here no solutions can be found in terms of arbi-
trary functions f and g. The geometrical √ spreading
 factor for cylindrical
waves is shown to be approximately 1/ r . For further discussion, see
Baker and Copson (1987) for a detailed analysis of Hadamard’s solution
for 2D wave propagation. In the days of 2D seismic processing, the
migration algorithms necessarily had to use cylindrical waves despite
their difficulties. With advent of 3D seismic processing, spherical waves
now are the dominant choice because of their advantages.

Spherical waves
In mathematics, a spherical coordinate system is a coordinate system
for 3D space where the position of a point is specified by three numbers:
the radial distance r of that point from a fixed origin, its polar angle u
measured from the z-axis, and the azimuth angle f of its orthogonal projec-
DOI:10.1190/1.9781560803737

tion on the xy-plane. It can be represented as the 3D version of the polar

coordinate system. The transformation equations are

x = r sin u cos f, y = r sin u sin f, z = r cos u . (8.53)

Figure 8.13 depicts a spherical shell with point P on the surface. The
coordinate r is the length of the line segment from the origin to P. The coordi-
nate u is the angle between the positive z-axis and the line segment from the
origin to P. The point Q is the projection of P onto the xy-plane. The coordinate
f is the angle between the positive x-axis and the line from the origin to Q.
Although the concept of a wave being the propagation of a disturbance
whose profile is unaltered is generally true in the 1D case, it is not generally
true in the 3D case. The plane wave (harmonic or not) is able to move
through space with an unchanging profile, but there is special type of 3D
wave which retains its shape but not its amplitude as it propagates: the
spherical wave. In order to obtain a wave, we assume that the function
u(r, t) has spherical symmetry about the origin. We assume that
u(r, t) = u(r, t), where r is the length of vector r, that is,

r = r = x2 + y2 + z2 . (8.54)
In such a case, the solution of the equation is simply
f (r − vt)
u(r, t) = . (8.55)
r
We can verify that this is a solution by inserting it into the wave equation.
This represents a spherical wave progressing radially outward from the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

262 Basic Wave Analysis, Part 3: Waveform Analysis

DOI:10.1190/1.9781560803737

Figure 8.13. Illustration of the definition of spherical coordinates r, u, f of a

point P.

origin, at a constant speed v, and having an arbitrary functional form f but

attenuated as 1/r. Another solution is given by

g(r + vt)
u(r, t) = , (8.56)
r
and in this case, the wave is converging toward the origin. A special case of
the general solution

f (r − vt) g(r − vt)

u(r, t) = c1 + c2 , (8.57)
r r
where c1 and c2 are arbitrary constants, is the harmonic spherical wave

A
u(r, t) = cos k(r + vt) , (8.58)
r
where the constant A is called the source strength. At any fixed value of time,
all solutions represent a cluster of concentric spheres filling all space. Each
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 8: Three Prototype Waves 263

wavefront, or surface of constant phase, is given by kr = constant. Notice

that the amplitude of this spherical wave is the function A/r, where the
term r−1 serves as an attenuation factor. The attenuation factor is a direct
consequence of energy conservation. Often, the attenuation factor is referred
to as geometric spreading. That is, unlike the plane wave, a spherical wave
decreases in amplitude, thereby changing its profile, as it expands and moves
out from the origin.
DOI:10.1190/1.9781560803737
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms
DOI:10.1190/1.9781560803737
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 9

Singularity Functions

“Nothing will come of nothing: speak again”

DOI:10.1190/1.9781560803737

—Shakespeare

Dirac delta function and Heaviside step function

One of the greatest physicists of the 20th century was Paul Adrien
Maurice Dirac (1902–1984). Known for his fundamental contributions to
the early development of both quantum mechanics and quantum electrody-
namics, he shared the 1933 Nobel Prize in Physics with Erwin Schrödinger
“for the discovery of new productive forms of atomic theory.” The Lucasian
Professor of Mathematics at the University of Cambridge and later a
visiting-turned-permanent Professor at Florida State University, he was
considered to be precise and taciturn; it was said that his colleagues in Cam-
bridge jokingly defined a unit called a “dirac,” which represented one word
per hour. He was known for formulating the Dirac equation, which describes
the behavior of fermions and predicted the existence of antimatter, and
his name appears in the classification of Fermi–Dirac statistics for half-
integer-spin particles and Bose–Einstein statistics for integer-spin particles.
For wave analysis, the Dirac function would prove to be pivotal. So,
although the Dirac delta function has solid roots in the work in Fourier
analysis by Cauchy and others in the 19th century, anticipating Dirac’s
discovery by over a century, it is Dirac’s understanding which is central
to contemporary analysis.

265
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

266 Basic Wave Analysis, Part 3: Waveform Analysis

It should be noted that, in theory, digital signal processing is much easier

to perform than analog signal processing. The delta function in signal
processing is analogous to the scalpel in surgery; it is the basic cutting
tool. In analog processing, the delta function is the Kronecker delta. The
Kronecker delta is a function of two non-negative integers. The function
is 1 if the integers are equal, and 0 otherwise. In other words, the Kronecker
delta is a discrete unit spike. In practice, it seems that nothing could be easier
to handle than the Kronecker delta. Unfortunately, it seems that nothing used
in practice could be more difficult to handle than the Dirac delta.
The Dirac delta function d(t) and the Heaviside unit step function u(t)
are known as singularity functions. By definition, a stable function has
finite energy. A stable time function h(t) and a stable space function g(x)
have well-defined Fourier transforms, designated by H(v) and G(k), respect-
ively. However, the study of wave propagation involves functions that are
not stable. For example, the function sin k0 x does not have finite energy,
DOI:10.1190/1.9781560803737

so it is not stable. The Fourier transforms of such unstable functions are

not well defined. Thus, in wave propagation, we must consider using quan-
tities that involve a certain kind of infinity.
In order to deal with these infinities, we introduce the Dirac delta func-
tion d(t) defined by

1
d(t) = 0 for t = 0 and d(t)dt = 1 . (9.1)
−1

Note that d(t) is undefined at t = 0. The quantity d(t) was used extensively
by Dirac, which explains why it is called the Dirac delta function. It also is
called the impulse function. To understand the function, first let us give a
description of d(t). Consider a function of t that satisfies the two conditions:

1) it vanishes everywhere except inside a small interval of length Dt

centered at t ¼ 0, and
2) it is positive and large enough inside this interval so that its integral over
this interval is unity.

The exact shape of the function inside this infinitesimal interval

is not important, provided there are no unnecessarily wild variations
(e.g., provided the function is always of order 1/Dt). In the limit as
Dt  0, this function will go over into d(t). In other words, d(t) is an
infinitesimally narrow positive function that is centered at t = 0 and that
has area equal unity.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 9: Singularity Functions 267

Figure 9.1. Rectangular

approximation of the Dirac
delta function.

Figure 9.1 illustrates such an approximation. The entity d(t) is not a

DOI:10.1190/1.9781560803737

function of t according to the usual mathematical definition of a function.

The Dirac delta can be loosely thought of as a function (of unit area) that
is zero everywhere except at the origin, where it is infinite. This is a heuristic
characterization. The Dirac delta is not a function in the traditional sense as
no function defined on real numbers has these properties. The Dirac delta
function is called a generalized function. It can be rigorously defined
either as a distribution or as a measure. Thus, d(t) is not a quantity which
generally can be used in mathematical analysis as an ordinary function
could be. Our use of the integral notation in expression (9.1) is merely a con-
venient way of describing the properties of d(t).
Now, we want to know the value of the product of t d(t). The value of
d(0) standing by itself is not defined. Thus, we have

0 × d(0) = indeterminate, t = 0
t d(t) = . (9.2)
t × 0 = 0, t = 0

Various arguments are used to settle this matter. The result is that the
required product is

t d(t) = 0 . (9.3)

The most important property of d(t) is exemplified by the equation

1
h(t) d(t)dt = h(0) , (9.4)
−1
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

268 Basic Wave Analysis, Part 3: Waveform Analysis

where h(t) is any continuous function of t. We can see easily the validity of
equation (9.4) by examining Figure 9.1. It is apparent that the left side of
equation (9.4) can depend only on the values of h(t) very close to the
origin, so that (without essential error) we may replace h(t) by its value at
the origin, i.e., h(0). Then, equation (9.4) follows from the second equation
in expression (9.1). This result can be written mathematically as
1 1
h(t)d(t)dt = h(0) d(t)dt = h(0) . (9.5)
−1 −1

Next, consider a shifted version of the Dirac delta function, d(t − t0 ).

We can deduce the so-called sifting property, given by
1
h(t) d(t − t0 )dt = h(t0 ) , (9.6)
−1
DOI:10.1190/1.9781560803737

where t0 is any real number. Equation (9.6) says that:

When integrated, the product of any (well-behaved) function and

the Dirac delta yields the function evaluated where the Dirac
delta is singular.

In other words, the process of multiplying a function of t, e.g., h(t), by

d(t − t0 ) and integrating over all t is equivalent to the process of assigning
to the function h(t) the number h(t0 ). In this sense, the delta function is inter-
preted as a functional and not as an ordinary function. The integral notation
is used merely as a convenient way of describing the properties of this
functional. A delta function itself does not have a well-defined value at its
center point. Equations (9.4) and (9.6) show that, when a delta function
occurs as a factor in an integrand, then the integral does have a well-
defined value. In wave propagation, whenever a delta function appears,
it will be something which is to be used ultimately in an integrand. There-
fore, we can treat d(t) as an ordinary function obeying all of the formal
rules of integration, provided that all of our conclusions are based on
either equation (9.4) or equation (9.6) and not on any point properties of d(t).
The Heaviside step function, or the unit step function, is denoted by
the lower-case Greek letter u. We will use the upper-case Greek letter Q
to denote the Fourier transform of u. Named for Oliver Heaviside (1850–
1925), this unit step function u is a discontinuous function represented as

0, t,0
u(t) = . (9.7)
1, t>0
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 9: Singularity Functions 269

This function has the value zero for negative times and the value one for
positive times. The value of u(t) for t = 0 need not be defined.
The unit step function represents a signal that “switches on” at time 0
and remains switched on indefinitely. The general class of step functions
can be represented as linear combinations of translations of the Heaviside
step function. An alternate way to define the delta function or impulse func-
tion is as the derivative of the Heaviside step function. Similarly, we may
say that the Heaviside step function can be defined as the integral of the
delta function. Thus, we may write
t
d u(t)
d(t) = and u(t) = d(t) dt . (9.8)
dt −1

We see that the delta function appears when a discontinuous function is

differentiated.
DOI:10.1190/1.9781560803737

The function h(t) is a function of time, so it is called a time function.

The function h(x, z) is a function of space, so it is called a spatial function.
For the 1D spatial delta function in the variable x, we simply use d(x) instead
of d(t). As a natural generalization of the 1D delta function, we define the 2D
delta function d(x, z) by the relation

d(x, z) = 0 for x = 0 and y = 0 (9.9)

 1
d(x, z) dx dz = 1 .
−1

The 2D delta function satisfies

d u(x) d u(z)
d(x, z) = d(x) d(z) = . (9.10)
dx dz
We may conclude that
1
h(x, z)d(x) d(z) dx dz = h(0, 0) (9.11)
−1

and
1
h(x, z)d(x − x0 ) d(z − z0 ) dx dz = h(x0 , z0 ) . (9.12)
−1

Equation (9.11) describes the so-called shifting property of the delta

function. The function h(x, z) has been shifted to h(x0 , z0 ). As an example,
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

270 Basic Wave Analysis, Part 3: Waveform Analysis

a mass m0 concentrated at the point x0 , z0 may be represented by

1
m0 = m0 d(x − x0 ) d(z − z0 ) dx dz . (9.13)
−1

These results given for two dimensions can be extended to three dimensions
with the 3D delta function d(x) d(y) d(z).
In addition, we can provide some useful equations that hold for the
delta function. The delta function is even; that is

d(x) = d(−x) . (9.14)

The scaling equation for a non-zero scalar c is

1 1 1
d(cx) d(x) 1
d(cx) dx = d(cx) = dx = . (9.15)
−1 |c| −1 |c| |c|
DOI:10.1190/1.9781560803737

−1

Thus,

d(x)
d(cx) = . (9.16)
|c|

At this point, let us define convolution. The convolution of two func-

tions f (x) and g(x) is written as f ∗ g, where the asterisk denotes convolution.
The convolution is defined as the integral of the product of the two functions
after one is reversed and shifted. Thus, the convolution of f and g is
1 1
f ∗g= f (x) g(x − x0 ) dx = f (x − x0 ) g(x) dx . (9.17)
−1 −1

The integral of a function with a shifted delta function is given by:

1 1
f (x) d(x − x0 ) dx = f (x) d(x0 − x) dx = f (x0 ) . (9.18)
−1 −1

This is referred to as the sifting property. The delta function is said to “sift
out” the value of the function at x0 . It follows that the effect of convolving a
function f (t) with the delayed delta function d(t − t0 ) is to delay f (t) by the
same amount; that is,
1
f (t) ∗ d(t − t0 ) = f (s) d(t − t0 − s) ds = f (t − t0 ) . (9.19)
−1
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 9: Singularity Functions 271

As a special case, we have

1
d(t2 − t) d(t − t1 )dt = d(t2 − t1 ) . (9.20)
−1

The Dirac delta function is the name given to a mathematical structure

that is intended to represent an idealized point object, such as a point mass
or point charge. In mathematics, generalized functions, or distributions, are
objects extending the notion of functions [following the distribution theory
of the French mathematician Laurent Schwartz (1915–2002)]. The Dirac
delta can be thought of as a generalized function that (1) is zero everywhere
except at the origin, where it is infinite, and (2) has unit area. Generalized
functions are especially useful in making discontinuous functions more
similar to smooth functions, and describing discrete physical phenomena
such as point charges. In a very specific sense, the Dirac delta function is
DOI:10.1190/1.9781560803737

the most basic generalized function and plays the role of the unit with
respect to the convolution operation. Treating the Dirac delta as an ordinary
function is similar to the way that physicists often treat the derivative dy/dx
as an ordinary fraction. Even so, there is a schism between the physicist’s
use of pragmatic approaches (e.g., using infinitesimal arguments) on
the one hand, and the mathematician’s desire for formal precision on
the other hand. In this book, we employ the pragmatic approach. For
example, we do not consider what class of ordinary functions h(t) are per-
mitted in the Fourier transform. We even use generalized functions. An
excuse for this playact (if one could exist) is that the theory will be
carried out on a digital computer; therefore, any error in method eventually
will be exposed.

Fourier transform
The conventional engineering usage of the Fourier transform depends
upon whether the function to be transformed is a time function or a
spatial function. According to this convention, the Fourier transform H(v)
of a time function h(t) is defined as
1
H(v) = h(t) e−ivt dt . (9.21)
−1

Note that equation (9.21) contains a negative exponent (that is, −ivt). There
is no constant scale factor outside of the integral sign. The continuous
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

272 Basic Wave Analysis, Part 3: Waveform Analysis

variable v is the angular (or circular) frequency. If time t has the dimension
of seconds, then the angular frequency v has dimension of rad/s. The
inverse Fourier transform h(t) of a frequency function H(v) is defined as
1
1
h(t) = H(v) eivt d v . (9.22)
2p −1

Note that equation (9.22) contains a positive exponent (i.e., iv t).

In addition, equation (9.22) contains a constant scale factor of 1/2p
outside of the integral sign. The reason for placing this constant is
the following. If we substitute v = 2pf (where the continuous variable
f is the cyclical frequency) into equations (9.21) and (9.22), then we
obtain
1
h(t) e−i2pft dt
DOI:10.1190/1.9781560803737

H(2pf ) =
−1
1 (9.23)
h(t) = H(v) e i2pft
df .
−1

Except for the sign in the exponent, these two integrals are perfectly
symmetric. It is important to remember that the negative sign (in the
exponent) goes with the time function and positive sign (in the expo-
nent) goes with the frequency function.
According to conventional engineering usage, the signs in the exponent
for spatial functions are the opposite of those for time functions. Thus,
Fourier transform G(k) of the spatial function g(x) satisfies
1
G(k) = g(x) eikx dx
−1
1 (9.24)
1
g(x) = G(k) e−ikx dk .
2p −1

The continuous variable k is the angular (or circular) wavenumber,

given by

2p
k= , (9.25)
l
where l is the wavelength. If the spatial variable x and wavelength l
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 9: Singularity Functions 273

have dimension of meters, then k has dimension of rad/m. It is important to

remember that:

† angular wavenumber k corresponds to angular frequency v, and

† cyclic wavenumber k/2p corresponds to cyclic frequency f = v/2p.

For one spatial dimension x, we have used the symbol k to denote the
angular wavenumber. In the case of two spatial dimensions x and z, we
will use the subscripts x and z to indicate that kx is the angular wavenumber
with respect to the x-axis and kz is the angular wavenumber with respect to
the z-axis. With this notation, we define the 2D Fourier transform G(kx , kz )
of a complex-valued spatial function g(x, z) of two independent variables x
and z as
 1
G(kx , kz ) = g(x, z) ei(kx x+kz z) dx dz . (9.26)
DOI:10.1190/1.9781560803737

−1

The transform G(kx , kz ) is itself a complex-valued function of two

independent variables kx , kz , which we refer to as angular wavenumbers.
Similarly, the 2D inverse Fourier transform g(x, z) of a function G(kx , kz )
is defined as
 1
1
g(x, z) = 2 G(kx , kz ) e−i(kx x+kz z) dkx dkz . (9.27)
4p −1

As mathematical operations, the transform (9.26) and inverse transform

(9.27) are quite similar, except that the sign in the exponent of transform
(9.26) is positive whereas the sign in the exponent in transform (9.27) is
negative. Also, the inverse transform (9.27) has a constant scale factor of
1/4p2 . Except for this scale factor, we could compute these 2D Fourier
transforms by the same algorithm simply by using a plus sign in the expo-
nent for the forward transform (9.26) and a negative sign in the exponent
for the inverse transform (9.27).
The position vector is

r = (x, z) = r = x ex + z ez with length r = r = x2 + z2 . (9.28)

The wavenumber vector is


k = (kx , kz ) = kx ex + kz ez with length k = k = kx2 + kx2 . (9.29)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

274 Basic Wave Analysis, Part 3: Waveform Analysis

The magnitude of k is the wavenumber k = 2p/l. Notice that the

exponentials in transforms (9.26) and (9.27) contain the expression
kx x + kz z, which is the dot product of the wave vector k and the position
vector r; that is,
kx x + kz z = k · r . (9.30)

For notational convenience, we sometimes write the 2D transforms as

 1
G(k) = g(r) eik·r dV ,
−1
 1 (9.31)
1
g(r) = 2 G(k) e−ik·r dK ,
4p −1

which are shorthand notations for equations (9.26) and (9.27), respect-
ively. (Note: When interpreting the notation used in equations (9.31) the
DOI:10.1190/1.9781560803737

quantities dV and dK are understood to be the scalar infinitesimal areas

dx dz in the x, z plane and dkx dkz in the kx , kz , respectively). These short-
hand equations for the Fourier transforms also hold in the 1D and 3D
cases.
We let the symbol F designate the mathematical operator for
Fourier transform. Thus, the Fourier transform pair may be written as

G(kx , kz ) = F [g(x, z)]

(9.32)
g(x, z) = F −1 [G(kx , kz )] .

In terms of black boxes, equations (9.32) are depicted in Figure 9.2.

Let us consider the 2D function
u = u(x, t) = f (x − vt) , (9.33)

where f is an arbitrary twice differentiable function, x and t are the continu-

ous space and time variables, respectively, and v is a positive constant. For a
fixed value of time t, let us determine the Fourier transform of the spatial

Figure 9.2.
Depiction of
the Fourier
transform pair in
equation (9.32).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 9: Singularity Functions 275

function u(x, t). From definition (9.24), we obtain

1
U(k, t) = u(x, t) eikx dx , (9.34)
−1

where the spatial transform U is a function not only of k but also of the
untransformed variable t. Let
1
F(k) = f (y) eiky dy . (9.35)
−1

With the change of variable y = x − vt, the spatial transform U(k, t) for time
t becomes
1 1
U(k, t) = f (y) e ik(x+vt)
dx = e ikvt
f (y) eiky dy = eikvt F(k) . (9.36)
−1 −1
DOI:10.1190/1.9781560803737

Fourier transforms of delta and step functions

If h(t) is continuous at t, then
1 1
h(t) = h(s)d(t − s)ds = d(s)h(t − s)ds (9.37)
−1 −1
or
h(t) = h(t) ∗ d(t) = d(t) ∗ h(t) ,
where the asterisk (∗ ) denotes convolution. These equations show that the
delta function plays the role of unity in convolution. In other words, any
function convolved with the delta function is equal to itself. Because
F [h ∗ g] = F [h]F [g] , (9.38)
we have
F [h(t)] = F [h(t) ∗ d(t)] = F [h(t)] F [d(t)] . (9.39)
Thus, the Fourier transform of d(t) is
1
F [d(t)] = 1 or d(t) e−ivt dt = 1 . (9.40)
−1

The inverse Fourier transform of F [d(t)] is

1
e−ivt d v = d(t) . (9.41)
−1
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

276 Basic Wave Analysis, Part 3: Waveform Analysis

Now, let us look the Heaviside unit step function u(t). Its Fourier
transform is

Q(v) = F [u(t)] . (9.42)

To determine an expression for Q(v), we start with the relation

d u(t)
d(t) = . (9.43)
dt

The Fourier transform of each side of this equation produces

 
d u(t)
F [d(t)] = F or 1 = iv Q(v) . (9.44)
dt
DOI:10.1190/1.9781560803737

Let c be a constant. Because v d(v) = 0, equation (9.44) may be written as

iv Q(v) = 1 + cv d(v) , (9.45)

which produces

1 c
Q(v) = + d(v) . (9.46)
iv i

It can be shown that c = pi. Thus, the Fourier transform of the unit step
function is

1
F [u(t)] = Q(v) = + p d(v) . (9.47)
iv

Now, we can determine the Fourier transform of the 2D delta function

d(x)d(z). As we have observed, this function is zero everywhere except at
the origin, where it is not defined. But because the Fourier transform is an
integral operation, we need not concern ourselves with the point properties
of d(x)d(z). Because the two delta functions are well behaved when they
appear in an integral, we can define their Fourier transform as
1
F [d(x)d(z)] = d(x)d(z) ei(kx x+kz z) dx dz . (9.48)
−1
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 9: Singularity Functions 277

Then, equation (9.48) reduces to

F [d(x)d(z)] ei(kx x+kz z) |x=z=0 = e0 = 1 for all kx and kz . (9.49)

Similarly, the inverse Fourier transform of the delta function is

1
F −1 [d(kx )d(kz )] = for all x and z . (9.50)
4p2

Fourier transform of a derivative

The Fourier transform of a derivative of a 1D function f (x) is
F [ f ′ (x)] = −iv F [ f (x)] . (9.51)

In the study of waves and spatial phenomena, often it is of interest to deter-

DOI:10.1190/1.9781560803737

mine the Fourier transform of the gradient and the Laplacian, which are,
respectively,

∂g ∂g ∂2 g ∂2 g
∇g = ex + ez and ∇2 g = + . (9.52)
∂x ∂z ∂x2 ∂z2
The inverse Fourier transform is
 1
1
g = g(x, z) = 2 G(kx , kz ) e−i(kx x+kz z) dkx dkz . (9.53)
4p −1

When we differentiate g once with respect to x and once with respect to z, we

obtain, respectively,
 1
∂g 1
= [(−ikx ) G(kx , kz )] e−i(kx x+kz z) dkx dkz
∂x 4p2 −1
 1 (9.54)
∂g 1
= [(−ikz ) G(kx , kz )] e−i(kx x+kz z) dkx dkz .
∂z 4p2 −1

From equations (9.54), we deduce that the Fourier transform of the left
side is equal to the expression in square brackets on the right side. Thus,

∂g
F = (−ikx ) G(kx , kz )
∂x
 (9.55)
∂g
F = (−ikz ) G(kx , kz ) .
∂z
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

278 Basic Wave Analysis, Part 3: Waveform Analysis

It follows that the Fourier transform of the gradient is

     
∂g ∂g ∂g ∂g
F [∇g] = F ex + ez = F ex + F ez
∂x ∂z ∂x ∂z
= −G(kx , kz )[kx ex + kz ez ] = −G(kx , kz )k . (9.56)

Similarly, a second differentiation of the two equations (9.54), res-

pectively, with respect to x and with respect to z produces
 1
∂2 g 1
= [(−kx2 ) G(kx , kz )] e−i(kx x+kz z) dkx dkz
∂x2 4p2 −1
2  1 (9.57)
∂g 1 −i(kx x+kz z)
2
= 2 [(−kz2 ) G(kx , kz )] e dkx dkz .
∂z 4p −1
DOI:10.1190/1.9781560803737

From equations (9.57), we deduce that

 
∂2 g
F = −kx2 G(kx , kz )
∂x2
 2  (9.58)
∂g
F = −kz2 G(kx , kz ) ,
∂z2

so the Fourier transform of the Laplacian is

   2   2 
∂2 g ∂2 g ∂g ∂g
F [∇ g] = F
2
2
+ 2 =F 2
+F
∂x ∂z ∂x ∂z2
(9.59)
= −kx2 G(kx , kz ) − kz2 G(kx , kz ) = −(kx2 + kz2 )G(kx , kz )
= −k2 G(kx , kz ) .

These spatial differentiation properties will be useful when we discuss

Green’s theorem and the wave equation in the frequency domain.
In summary, a 2D spatial function g(x, z) has the Fourier transform
G(kx , kz ), which also is called the spectrum of g(x, z). The inverse Fourier
transform,
 1
1
g(x, z) = 2 G(kx , kz ) e−i(kx x+kz z) dkx dkz , (9.60)
4p −1
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 9: Singularity Functions 279

is actually a decomposition of the function g(x, z) into two-dimensional

plane waves, given by

1 −i(kx x+kz z)
e . (9.61)
4p2
The complex weighting factor for this decomposition is the spectrum
G(kx , kz ), which gives the magnitude and phase factors that must be
applied to each 2D plane wave in order to synthesize the desired
function g(x, z).
DOI:10.1190/1.9781560803737
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms
DOI:10.1190/1.9781560803737
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10

Waves Traveling in Opposite

Directions

“This is an art
Which does mend nature, change it rather, but
DOI:10.1190/1.9781560803737

The art itself is nature.”

—Shakespeare
To see a star
From planets, stars, and galaxies to the smallest subatomic particles, all
matter, and energy, the universe includes all of time and space. Although the
observable universe is estimated to be 93 billion light years in diameter, the
size of the entire universe is not known and may be either finite or infinite.
For millennia, it was believed that the Sun, moon, and planets orbited
the Earth and the stars were holes in a large dark outer sphere. In the 16th
century, Nicolaus Copernicus finally presented a Sun-centered solar
system. This theory, called heliocentrism, featured the planets of the solar
system all orbiting the Sun in perfect circles. In the early 17th century,
Johannes Kepler established that the planets have elliptical orbits.
The astronomer and scientist most recognized for his explanation of
heliocentrism, of course, was Galileo. Indeed, it was his explanation of
heliocentrism in Dialogue Concerning the Two Chief World Systems
(1632) which sparked the charges of heresy against him. Ironically, the
book was published in Italian in Florence under a formal license from the
Inquisition, but after his conviction it was placed on the Index of Forbidden
Books. Fortunately, in 1635, Bonaventura Elsevier and Abraham Elsevier at
Leiden published a Latin edition, Systema Cosmicum. Likewise, in 1638,

281
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

282 Basic Wave Analysis, Part 3: Waveform Analysis

Lodewijk Elsevier in Leiden was willing to publish Galileo’s final book

(passed over by other companies in France, Germany, and Poland), The Dis-
courses and Mathematical Demonstrations Relating to Two New Sciences,
which covered much of his work in physics over the preceding 30 years.
Galileo was adept at studying the universe, from the rotation of planets
to the unseen forces at play on Earth—velocity, gravity, inertia, and much
more, which contributed to his applied work on pendulums, thermoscopes,
compasses, and telescopes. In astronomy alone, he was able to confirm the
phases of Venus, discover the four largest satellites of Jupiter, observe the
rings of Saturn, and analyze sunspots.
Also combining science and astronomy, Christiaan Huygens (1629–
1695) studied at the University of Leiden and later at the College of
Orange in Breda, graduating in 1649. Together with his brother Constantijn,
Christiaan began grinding and polishing his own lenses, and eventually
designed the Huygenian eyepiece—a telescope ocular consisting of two
DOI:10.1190/1.9781560803737

lenses. Using these improvements, he built the first astronomical telescope.

In 1655, he was the first to observe Saturn’s largest moon, Titan, and he
noted that Saturn was “surrounded by a thin flat ring, nowhere touching,
and inclined to the ecliptic.” He also observed the Orion Nebula and suc-
cessfully subdivided it into different stars.
Huygens first communicated his wave theory of light to the Paris
Academy of Sciences in 1678. It was published in 1690 as Traité de la
lumière. In it, he gave a numerical value of the speed of light and explained
how light is propagated by means of spherical waves emitted along the
wavefront. Huygens’ principle satisfies the requirement that stars appear
as points of light. Light in a vacuum does not disperse or dissipate. If star-
light dispersed, it would spread out like water waves and we would not see
any individual stars. Like fish in the ocean, we would have no concept of the
celestial world above.
It could be argued that the question, “why can we see stars,” rarely was
asked, so it is little wonder that Huygens’ solution made no impact. Over
time, however, the opposite question arises, “why do we not see more
stars?” Is the sky populated by an infinite number of stars? Olbers’
paradox, named for German astronomer Heinrich Wilhelm Olbers (1758–
1840) and also called the dark night sky paradox, asserts that the darkness
of the night sky conflicts with the assumption of an infinite and eternal
static universe. In an infinite universe with an infinite number of stars,
would not the night sky be incredibly bright? The challenge that Olbers’
paradox presents to the infinity assumptions has been tackled by scientists
as well as writers, such as Edgar Allan Poe (1809–1849). In effect, Poe
resolves the paradox by arguing that, because the universe is of finite age
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 283

and the speed of light is finite, we can observe only a finite number of stars.
The density of stars within this finite volume is low, and the night sky under
which we live appears dark.
The intersection of science and poetry and attempts to explain the uni-
verse also were apparent in Huygens’ final work, Cosmotheoros, which was
published posthumously in 1698. In this book, Huygens speculated about the
existence of extraterrestrial life on other planets, which he imagined would
be similar to that of Earth. Huygens went into great detail, for example,
stating that the availability of water in liquid form was essential for
life and that the properties of water must vary from planet to planet to suit
the temperature range. Here, Huygens uses the word “water” to mean a
liquid necessary for life. Today, scientists believe that life might be
present on Huygens’ moon, Titan, with the “water” there being CH4
rather than H2O. Huygens took the observations of bright spots on the
surface of Mars to be evidence of water and ice. There is water on Mars,
DOI:10.1190/1.9781560803737

as we now know, but it is frozen under the north and south polar ice caps.
The polar ice caps, which were once thought to be made up only dry ice
(frozen carbon dioxide), are now known to have water ice beneath a top
layer of dry ice.

What a wonderful and amazing scheme have we here of the magni-

ficent vastness of the universe! So many suns, so many earths, and
every one of them stocked with so many herbs, trees and animals,
and adorned with so many seas and mountains! And how must
our wonder and admiration be increased when we consider the
prodigious distance and multitude of the stars? (Huygens, 1695).

Huygens took the analogy further and speculated that some of these planets
could have rational creatures that perform astronomy, geometry, music,
have hands and feet, walk upright, and live in a society.

Interference
Drop a stone (along a vertical path) into a pond to create a water wave.
The stone begins at zero speed and it accelerates to its final downward speed
as it hits the water. The water wave begins at its nominal speed and main-
tains that speed as it travels. In other words, water waves do not begin as
motionless and then slowly gain speed as they travel. Water waves are
already traveling at their nominal speed at the moment they are created.
Likewise, all other waves (e.g., sound waves, seismic waves, light waves)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

284 Basic Wave Analysis, Part 3: Waveform Analysis

are already at a certain non-zero speed at the moment they are created,
without needing to be accelerated.
Light and all other electromagnetic waves consist of particles called
photons. Photons travel like waves, not like particles. A photon of light
does not accelerate to light speed. A photon is already traveling at light
speed at the instant that it is created. In other words, a photon is not like a
Heaviside function, which jumps from a speed of zero to light speed instan-
taneously at time zero. Instead, a photon is always traveling at the speed of
light, even at time zero. We can summarize this situation by saying that a
photon is a particle that travels like a wave. The fundamental idea is that
a photon is not a traditional particle. A photon is a quantum object, which
from one point of view acts like a wave. From another point of view, a
photon acts like a particle.
A traveling wave is created because a deformation (in the material-
medium or in the field-medium) causes the medium to snap back toward
DOI:10.1190/1.9781560803737

the equilibrium state. However, the medium overshoots the equilibrium

state, and ends up oscillating back and forth, all the while bringing neighbor-
ing regions into the same motion. The wave speed, therefore, is determined
by the medium, not by an external agent (such as the stone thrown into the
water) pushing on the medium. Pushing harder on the medium makes the
amplitudes of the wave larger. It does not make the wave travel faster.
This picture of a traveling wave is the same for all waves.
We know that photons travel at speed c in a vacuum. A photon is mass-
less, and it always travels at speed c in a vacuum in all reference frames.
Photons cannot penetrate through metal because the free electrons that
make metals conductors easily absorb the photons immediately. However,
photons do pass through a transparent or translucent material medium,
such as glass, water, or air. Light in its passage through such media will
travel at a pace slower than its speed in a vacuum. The ratio of light
speed c to the observed phase velocity is called the refractive index of the
medium. When light passes through glass, the photons come into contact
with the electrons in the glass. However, photons in visible light do not
have enough energy in them to change the state of electrons in glass. In
other words, the energy is not enough to move the electrons in the glass
to a higher level. As a result, the photons do not interact with the electrons
in the glass at all and simply pass through. A wide range of gasses and solids
transmit light in varying degrees. Light slows down slightly in clear air. The
speed of light through air is about 99.97% of the speed of light in vacuum;
this is enough to cause stars to twinkle.
In 1665, Sir Isaac Newton was interested in learning about light and
colors. Newton darkened his room and made a hole in his window shutter,
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 285

allowing only one beam of sunlight to enter the room. Then he took a glass
prism and placed it in the sunbeam. The result was a spectacular multico-
lored band of light similar to a rainbow. The multicolored band of light is
called a color spectrum. Dispersion is a phenomenon in which the phase ve-
locity of a wave depends on its frequency (or wavelength). Each and every
wavelength of light (in vacuum) travels at the same velocity. This behavior
is almost true in air. However, in most transparent materials, such as glass
and water, the different wavelengths are slowed down by different
amounts. Different velocities mean different amounts of bending in refrac-
tion. This explains why white light is split into its component colors by a
prism. The most familiar example of dispersion is the rainbow. Dispersion
in the raindrops separates white light into its component colors.
Jean-Félix Picard (1620–1682) was the first person to measure the size
of the Earth to a reasonable degree of accuracy. In 1670, he determined that
the terrestrial diameter is 12,658 km. This is almost exactly equal to the
DOI:10.1190/1.9781560803737

modern value. Picard collaborated and corresponded with many scientists,

including Sir Isaac Newton, Christiaan Huygens, Ole Rømer, Rasmus
Bartholin, Johann Hudde, and even his main competitor, Giovanni
Cassini, although Cassini was often less than willing to return the gesture.
The astronomical unit (AU) is a unit of length that represents the dis-
tance from the Earth to the Sun. This actual distance varies as Earth orbits
the Sun, going from a maximum (aphelion) to a minimum (perihelion)
and back again, so an average distance is taken. The result is that now the
AU is defined as exactly 149,597,870,700 m, or roughly 150 million km.
In 1653, Christiaan Huygens was the first to calculate an accurate value
for the AU. He knew that Venus showed phases when viewed through a tele-
scope, similar to the moon. He used the phases of Venus to find the angles in
a Venus–Earth–Sun triangle. For example, when Venus appears half-illumi-
nated (corresponding to a half moon) by the Sun, the three bodies form a
right triangle. Estimating correctly the size of Venus, Huygens was able
to determine the Venus–Earth distance. He also measured the angle
between Venus and the Sun. Huygens used these measurements and com-
puted the AU as 12,000 earth diameters. Picard’s value for the diameter
of the earth was 12,658 km. Thus, the value of the AU, according to
Huygens, was 153,696,000 km, or roughly 154 million km. We recall that
the true value is roughly 150 million km, so Huygens’ value was quite
accurate. In 1672, Giovanni Domenico Cassini (1625–1712) calculated
the distance to the Sun. He had access to more advanced instruments,
and the French government paid for observations in Paris and French
Guiana. Cassini obtained an AU value of 140,316,000 km, or roughly
140 million km.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

286 Basic Wave Analysis, Part 3: Waveform Analysis

On 22 August 1676, Ole Rømer (1644–1701) made an announcement to

the French Royal Academy of Sciences in Paris that he would be changing
the basis of calculation for his tables of eclipses of Jupiter’s moon, Io. The
original record of the meeting of the French Royal Academy of Sciences has
been lost; however, Rømer’s presentation was recorded as a news report in
the Journal des Sçavans on 7 December 1676. It stated that Rømer reported
that it would take 11 minutes for light to travel a distance equal to the radius
of Earth’s orbit. Cassini, who was Rømer’s superior at the Royal Observa-
tory, was an early and tenacious opponent of Rømer’s ideas, and it seems
that Picard also shared many of Cassini’s doubts. Rømer never published
the formal description of his method, possibly because of the opposition
of Cassini and Picard to his ideas. Huygens took Rømer’s side, and wrote
to the French Controller-General of Finances Jean-Baptiste Colbert in
Rømer’s defense.
For years, Huygens had claimed that the velocity of light is not infinite,
DOI:10.1190/1.9781560803737

but finite. Although Huygens’ book, Treatise on Light, was not published
until 1690, it had been presented to French Royal Academy of Sciences in
1678. In his book, Huygens used Cassini’s AU value and Rømer’s traveltime
to compute the velocity of light by

Cassini’s AU value 140,316,000

c= = = 212,600 km/s . (10.1)
Rømer’s traveltime 11 × 60

Actually, Rømer’s value of 11 minutes was in error. The true value is

8.317 minutes. If Huygens had used the true traveltime value and his own
AU value, Huygens would have obtained

Huygens’ AU value 153,696,000

c= = = 307,996 km/s . (10.2)
true traveltime 8.317 × 60

Huygens was working in the era before Leibniz’s concept of a math-

ematical function had come into importance. We represent a wave as a func-
tion u(x, y, z, t) of space coordinates x, y, z and time t. The function provides
the oscillation about some equilibrium position at any given point in space at
any given time. Huygens did not have the luxury of using this approach.
Instead, he merely worked with straight lines, curved lines, planes, and
curved surfaces.
In classical mechanics, particles have well-defined positions and trajec-
tories. However, mechanical waves are not localized in space. Rather, waves
fill regions of space. Some waves are more localized than others, and so it is
useful to distinguish two broad classes.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 287

1. A wave pulse (or wavelet) is a relatively localized disturbance. For

example, when a single raindrop falls into a pond, the resulting ripples,
which constitute a surface wave, extend over only a small portion of the
surface at any instant of time.
2. At the opposite extreme, a wave is a general disturbance that extends
over distance and time. If the water surface is repeatedly disturbed by
rainfall for a period of time, the surface wavelets eventually blanket
a large area.

Rainfall over water has been observed from time immemorial, but in
1688 it was Huygens who devised Huygens’ principle, which connects (1)
and (2). Huygens, of course, was working with light waves and not water
waves. The word wavelet (French, ondelette), first used in 1813, was the
diminutive of the word wave; that is, a wavelet was a little wave or a
ripple. Over time, however, the word wavelet was used as a synonym for
DOI:10.1190/1.9781560803737

the secondary wave in the statement of Huygens’ principle. Later, Fresnel

seems to have been the first to use the word wavelength and its symbol,
the Greek letter l.
Energy is work. Power is energy per given time. In signal processing,
the energy E of a continuous-time signal w(t) is defined as
1
E= |w(t)|2 dt . (10.3)
−1

Energy in this context is not, strictly speaking, the same as the conven-
tional notion of energy in physics and other sciences. The two concepts,
however, are closely related, and it is possible to convert from one to the
other. A signal with finite energy is called an energy signal, or (in the
context of wave propagation) a wavelet. Energy signals have significant
non-zero values only for a finite duration of time. The power of an energy
signal is zero, because it is the result of dividing finite energy by infinite
time. Power signals have significant non-zero values for an infinite duration
of time. Because a power signal is not limited in time, the energy of a power
signal is infinite. An energy signal has zero power (i.e., a finite energy
divided by infinity). A non-rechargeable battery is an energy signal. It
holds a finite amount of energy. It will supply significant power only for a
limited time. A power signal is always adding energy, so integration over
all time (to obtain energy) produces an infinite result. The sun is a power
signal (assuming that the sun will never burn out). It radiates power. The
sun would supply infinite energy when its energy is accumulated over all
time.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

288 Basic Wave Analysis, Part 3: Waveform Analysis

Figure 10.1. Central

portion of an infinitely long
Ricker wavelet.

Norman Ricker (1940) introduces the idea of a seismic wavelet. In fact,

he is the first to use the term “wavelet” in such a context. Ricker makes use
of a modified wave equation first given by George Gabriel Stokes in 1845.
The Ricker wavelet is a non-causal energy signal (of infinite duration
−1 , t , 1), and it is symmetric about its origin time (see Figure 10.1).
DOI:10.1190/1.9781560803737

Furthermore, the Ricker wavelet is a zero-phase signal. The amplitude

w(t) of the Ricker wavelet with peak frequency fM at time t is given by

w(t) = (1 − 2p2 fM2 t2 ) exp (−p2 fM2 t2 ) . (10.4)

Special seismic experiments have been conducted to discover an actual

physical Ricker wavelet, but they have all failed. A physical wavelet must be
causal; i.e., w(t) = 0 for t , 0. The Ricker wavelet is not causal. Because
seismic experiments are costly, instead one may compute the wavelet
from the recorded seismic records taken in exploration. No special
experiment is needed. The result is a causal wavelet such as shown in
Figure 10.2.
We can regard an earthquake seismic wave or an exploration seismic
wave as a section of a power signal over the time period of interest. A
seismic wavelet would be a constituent of such a wave. The wavelet is an
effect. We must look for the cause of that effect. Is that wavelet due to
single or multiple reflection? Where are the reflecting interfaces? How do
we undertake this endeavor? One way is to think of the wavelet as the
impulse response of a point source. One drop of rain falls onto a body of
still water (a calm pond). The resulting surface disturbance spreads out as
a circular time-varying wavelet. Over time, this wavelet is attenuated
because of circular spreading. It also changes its shape because of dis-
persion. A rainfall produces many wavelets, all of which superimpose to
form waves on the pond.
When you look at a pond in a rainstorm, you do not see a collection of
raindrops, but you do see many waves. To quote Shakespeare, “When
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 289

Figure 10.2. Spiking deconvolution operator (left) and causal seismic wavelet
DOI:10.1190/1.9781560803737

(right) (Robinson, 1957).

sorrows come, they come not single spies, but in battalions.” When you
look at the sky on a dark and cloudless night, you do not see a blurry
white haze, but you do see many individual stars twinkling in the night.
The twinkle comes from the earth’s atmosphere encountered in the last
few miles that the starlight has traveled. However, there is no such deleter-
ious effect due to the millions of light years that the starlight has traveled to
reach the earth’s atmosphere. Dispersion causes pulses to spread out as
they travel and thus degrade the signals over long distances. An infinitesi-
mally thin spherical shell produces no dispersion. Huygens correctly
concludes that the wavelet (i.e., his secondary wave) must be confined to
such a spherical shell.
Huygens maintains that each wavefront of light consists of many point
sources. Each point source yields a spatial disturbance that spreads out as a
spherical time-varying secondary wave. The secondary wave meets our
definition of wavelet (i.e., an impulse response). Over time, this wavelet
is attenuated because of spherical spreading. However, the wavelet
retains its impulsive shape because there is no frequency dispersion. In
summary, the wavelet is in the form of an infinitesimally thin spherical
shell. The new wavefront of light is formed as the superposition of all of
the wavelets.
Huygens’ view on wave propagation is limited, because he had no
concept of interference. For this reason, the term Huygens’ wavelet may
seem to be an historical anachronism. However, the Encyclopedia Brit-
annica certainly credits Huygens for the wavelet. It gives this definition:
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

290 Basic Wave Analysis, Part 3: Waveform Analysis

Huygens’ principle, in optics, a statement that all points of wave-

front of light in a vacuum or transparent medium may be regarded
as new sources of wavelets that expand in every direction at a rate
depending on their velocities. Proposed by the Dutch mathemati-
cian, physicist, and astronomer, Christiaan Huygens, in 1690, it is
a powerful method for studying various optical phenomena.
A surface tangent to the wavelets constitutes the new wave front
and is called the envelope of the wavelets. If a medium is homo-
geneous and has the same properties throughout (i.e, is isotropic)
the three-dimensional envelope of a point source will be spherical;
otherwise, as is the case with many crystals, the envelope will be
ellipsoidal in shape.

Certainly, two centuries later, Fresnel, Fraunhofer, and Kirchhoff

have refined Huygens’ principle, but this is no reason to diminish the priority
DOI:10.1190/1.9781560803737

of Huygens. Exactly who invented the automobile is a matter of opinion.

If credit is given to one inventor, it would be Karl Benz of Germany in
1885. The Mercedes-Benz automobile of today is certainly superior
in every way, but its origin goes to the early mechanism of Benz two
centuries ago.
Thomas Young was the first person to recognize the importance of inter-
ference in the description of wave propagation. In “A Reply to the Animad-
versions of Edinburgh Reviewers” Young (1804) wrote:
Suppose a number of equal waves of water to move upon the surface
of a stagnant lake, with a certain constant velocity, and to enter a
narrow channel leading out of the lake. Suppose then another
similar cause to have excited another equal series of waves, which
arrive at the same time, with the first. Neither series of waves will
destroy the other, but their effects will be combined: if they enter
the channel in such a manner that the elevations of one series
coincide with those of the other, they must together produce a
series of greater joint elevations; but if the elevations of one series
are so situated as to correspond to the depressions of the other,
they must exactly fill up those depressions. And the surface of the
water must remain smooth; at least I can discover no alternative,
either from theory or from experiment.
A defining characteristic of all waves is superposition, which describes
the behavior of overlapping waves. The superposition principle states that,
when two or more waves overlap in space, the resultant disturbance is equal
to the algebraic sum of the individual disturbances. This principle gives a
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 291

number of effects that are collectively called interference phenomena. In

constructive interference, the crests of two waves coincide, and the waves
are said to be in phase with each other. Their superposition results in a
reinforcement of the disturbance. Conversely, in destructive interference,
the crest of one wave coincides with the valley of a second wave, and they
are said to be out-of-phase. The amplitude of the combined wave equals the
difference between the amplitudes of the individual waves. For the special
case in which those individual amplitudes are equal, the destructive interfer-
ence is complete, and the net disturbance to the medium is zero.
A Green’s function is the impulse response of an inhomogeneous
linear differential equation with specified initial conditions or boundary con-
ditions. Through the superposition principle, the convolution of a Green’s
function with an arbitrary function f (x) is the solution to the inhomogeneous
differential equation for f (x). In other words, given a linear differential
equation, L(solution) = source, one can first solve L(Green’s function) =
DOI:10.1190/1.9781560803737

d, and realizing that, because the source is a sum of delta functions,

the solution is a sum of Green’s functions, by the linearity of L. We only
deal with the wave equation, which explains our use of the term wavelet.

Matter waves
Unifiers are people whose driving passion is to find general principles
which will explain everything. They leave the universe looking a little
simpler than before. Diversifiers are people whose passion is to explore
details. They leave the universe a little more complicated than before.
Both are essential to the progression of science.
In 1674 in Delft, Antoni van Leeuwenhoek (1632–1723) looked through
the microscope that he had carefully constructed. The specimen was a small
glass tube containing water from the Berkelse Meer, a lake a few kilometers
away. Leeuwenhoek was shocked because he did not see pure clear water.
Instead he saw another world, a world unknown to human beings. It was
an aquarium filled with minuscule little animals swimming in all directions,
each about one thousand times as small as a tiny cheese mite. Some were
shaped like spiral serpents, some were globular, and some were elongated
ovals. Leeuwenhoek had discovered the microscopic world inhabited by
microorganisms (also known as microbes), for which he has been credited
with revolutionizing biological science. Leeuwenhoek showed that the bio-
logical world was much more diversified than previously believed.
Emil Johann Wiechert (1861–1928) was a German physicist and geo-
physicist who made many contributions to both fields, including being
among the first to discover the electron and presenting the first verifiable
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

292 Basic Wave Analysis, Part 3: Waveform Analysis

model of a layered structure of the earth. From 1890 to 1897 at Königsberg

University, Wiechert conducted important research on the atomic structure
of electricity and matter. In 1897 at the University of Göttingen, he became
the world’s first Professor of Geophysics. Wiechert’s students included
Beno Gutenberg, Karl Zöppritz, Ludwig Geiger, and Gustav Angenheister.
From 1897 to 1914, the Göttingen school of geophysics produced much
important work under Wiechert’s leadership. By 1900, Wiechert had pro-
duced his famous inverted-pendulum seismograph and had worked out its
theory. Early tables prepared by Zöppritz gave the traveltimes of earthquake
waves through the interior of the earth. Wiechert was one of the first to
suggest the presence of a dense core in the earth. Gutenberg’s calculation
gave the value of 2900 km for the depth of the earth’s outer core. Composed
primarily of liquid iron and nickel, the core controls earth’s magnetic field
which protects us from the sun’s solar winds. With Gustav Herglotz (1881–
1953), Wiechert developed the basic mathematical process whereby the ve-
DOI:10.1190/1.9781560803737

locities of seismic waves deep in the interior of the earth can be derived from
the traveltime tables.
In addition to his work on seismology, Wiechert contributed to other
branches of geophysics, particularly atmospheric electricity. Under his direc-
tion, various methods were developed for measuring potential gradients and
conductivity in the atmosphere. In 1914, Wiechert turned to the transmission
of sound waves through the atmosphere with a view to determining features of
the stratification of the atmosphere. He also investigated the fine structure of
the crust of the earth, using specially designed portable seismographs to
record waves from artificial explosions. In this research, he was one of the pio-
neers of geophysical prospecting by seismic methods. As a geophysicist,
Wiechert was a diversifier. He exposed a diversity of details in the earth
and created tools for the discovery of those details. In 1896, Wiechert wrote:

So far as modern science is concerned, we have to abandon comple-

tely the idea that by going into the realm of the small we shall reach
the ultimate foundations of the universe. I believe we can abandon
this idea without any regret. The universe is infinite in all directions,
not only above us in the large but also below us in the small. If we
start from our human scale of existence and explore the content of
the universe further and further, we finally arrive, both in the large
and in the small, at misty distances where first our senses and then
even our concepts fail us.

In 1924, Louis de Broglie (1892–1987) maintained that, just as light has

wave and particle properties, all microscopic material particles (e.g.,
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 293

electrons, protons, atoms, molecules) also have dual character. They behave
as a particle as well as wave. Such a wave is called a matter wave, a de
Broglie wave, or a probability wave. (A probability wave is a wave whose
amplitude at a given region and over a given time interval corresponds to
the probability of observing a given particle within that region in that
time.) In fact, the concept that matter behaves like a wave is referred to as
the de Broglie hypothesis. This means that an electron which has been
regarded as a particle also behaves like a wave. Thus, according to de
Broglie, all of the material particles in motion possess wave characteristics.
However, the dual nature is significant for only microscopic bodies. For
large bodies, the wavelengths are too small to be measured. Probability
waves play a central role in the theory of quantum mechanics.
The wave-particle duality refers to the exhibition of both wave-like and
particle-like properties by a single entity. For example, electrons undergo
diffraction and can interfere with each other as waves, but they also act as
DOI:10.1190/1.9781560803737

point-like masses and electric charges. The measurement of the wave func-
tion will randomly “collapse,” or rather “de-cohere,” the probability wave
into a sharply peaked function at a well-defined position, (subject to uncer-
tainty), a property traditionally associated with particles. More precisely, a
wave function collapse is said to occur when a wave function (initially in a
superposition of several eigenstates) appears to reduce to a single eigenstate
(by observation). Such a collapse is the essence of measurement in quantum
mechanics and connects the probability-wave function with classical obser-
vables such as position and momentum. In simple words, the measurement
of a probability wave yields a particle. We can observe the interference of
mechanical waves whenever we throw stones into a pond. We can never
observe the interference of probability waves because the very act of obser-
vation collapses the waves into particles, which in turn are subject to the
Heisenberg Uncertainty Principle.
Quantum mechanics solves the problems of the stability of atoms and the
nature of spectral lines. It also brings chemistry into the framework
of physics. Quantum electrodynamics (i.e., the quantum field theory of
photons, electrons, and positrons) has led to the Standard Model. This
model unifies physics, except for gravitation, dark matter, and many unex-
plained numerical quantities. The Standard Model is a set of equations that
can fit onto a single sheet of paper. This sheet of paper arguably represents
the greatest intellectual achievement of human beings up to the present. It
is a major step toward the Holy Grail, i.e., a unified field theory (UFT). It is
a theory that allows all that is usually thought of as fundamental forces and
elementary particles to be written in terms of a single field. There is currently
no accepted unified field theory, and thus it remains an open line of research.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

294 Basic Wave Analysis, Part 3: Waveform Analysis

So, when it comes to “seeing” our research, an additional aspect we

should discuss involves sight, resolution, and probability. The resolution
of a microscope, for example, refers to the shortest distance between two
points that a user can still see as separate images. The resolution of Leeu-
wenhoek’s microscope, which uses light as its illumination source, is suffi-
cient to see microbes. In comparison, an electron microscope uses a beam of
accelerated electrons as a source of illumination. Because the wavelength of
an electron can be up to 100,000 times shorter than that of visible light
photons, the electron microscope has a higher resolving power than a
light microscope and can reveal the structure of smaller objects. Can
atoms be seen under an electron microscope? The answer is partially yes.
In certain situations, the images can be interpreted as representations of
atoms or even atomic nuclei. Has anyone ever seen electrons and protons
under an electron microscope? Electrons and protons are subatomic par-
ticles that travel as waves. A subatomic particle is smaller than an atom.
DOI:10.1190/1.9781560803737

When we talk about waves at the atomic level, we are referring to matter
waves, i.e., probability waves which are 3D graphs that describe the prob-
ability of finding an electron at a certain point around a nucleus. Electrons
are much smaller than most atomic nuclei, so we are able to use them as a
medium to view the features of atoms. Electrons are incredibly tiny and
have extremely low mass. They move extremely fast and, due to the Heisen-
berg Uncertainty Principle, their exact position is practically unknowable.
We can only approximate their position to within a certain uncertainty.
Without being able to acquire an exact position makes it impossible to
view an electron. That aside, their mass is so low that even the smallest inter-
action with them (involving another electron or photon) will send them
flying off so that we cannot view them. Because there is no way to see elec-
trons, we use the electron cloud model of the atom, which indicates where
electrons are likely to be, but never where they actually are. The electron
cloud is, in fact, a probability field.
Gravity and magnetic methods are an essential part of oil exploration
which, despite being comparatively low resolution, have some very big
advantages. These geophysical methods passively measure natural variations
in the earth’s gravity and magnetic fields over a map area and then attempt to
relate these variations to geologic features in the subsurface. Lacking a con-
trolled source, such surveys are usually environmentally unobjectionable. At
a comparatively low cost, airborne potential field surveys can provide cover-
age of large areas. Magnetic surveys are used to determine magnetic
anomalies associated with iron formations which contain magnetite.
Magnetic surveys have been conducted by geophysicists for the past 100
years, and magnetic compasses have been used for a much longer period
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 295

than that. So, although scientific research has made greater progress since,
for example, the magnetic compass of Christopher Columbus, a few
words should be said about how much more could be learned from the
natural world about the use of magnetic information.
Indeed, geophysical methods that use magnetic reference points are
crude in comparison to the high-resolution procedures used by a bird or
turtle, a butterfly or a fish. Animals have the ability to navigate and find
their way accurately without maps or instruments. In particular, an ant con-
tinuously determines its position relative to a known starting point (e.g., its
nest) while traveling on a crooked path. At every point along its journey, the
ant can always return directly in a straight line to its nest. A specific example
is the Sahara Desert ant, which forages for insects that have died of heat
stress. These ants can sustain surface temperatures of up to 708C (1788F).
The ant will venture out a distance as much as a half of a kilometer on a tor-
tuous path that continually meanders and twists in all directions. The ant
DOI:10.1190/1.9781560803737

completes its journey when it finds a dead insect. The ant then goes directly
back to its nest by the shortest route. This skill is necessary to its survival
under the harsh desert conditions.
With respect to size, this ant travels farther from its nest than any other
animal that lives on the Sahara. There are no landmarks on a desert. It is all
sand. An ant does not know where or when it will find a dead insect. As a
result, at each point along its path, the ant computes the distance and direc-
tion to its nest. How does the ant do this? Ants can do this in the dark, so they
do not need the sunlight. At the same time the ant must correct for terrain. It
takes more steps to traverse a fixed horizontal distance on a bumpy terrain
than on a flat terrain. However, the ant gets the correct value in either
case. It seems that ants have a “mental integrator” which allows them to
accumulate changes in direction, and from there to keep track of their
location. The same seems to apply to moles when they dig their underground
tunnels. The “mental integrator” is a form of dead reckoning (as used by
Columbus) which requires a magnetic compass to determine direction. It
is generally accepted that many animals obtain their orientation or naviga-
tional behavior by magnetic field variations. They have internal compasses
just as real as the compass that Columbus used. What kind of compass
within an ant or a mole can sense and amplify something as weak as
earth’s magnetic field?
Magnetoception is the ability of an animal to perceive earth’s mag-
netic field. The earth’s magnetic field can have a decisive effect on
living things. By magnetoception, animals can find food, return their
nests, and meet each other. Bats and geese use magnetoception for naviga-
tional, altitude, and location purposes. Magnetoception allows honeybees
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

296 Basic Wave Analysis, Part 3: Waveform Analysis

to possess the extraordinary ability to create a map of their surroundings.

Invertebrates with magnetoception include fruit flies, lobsters, and certain
bacteria. Vertebrates with magnetoception include many species of birds,
turtles, sharks, and stingrays. The type of magnetoception present in sea
creatures such as sharks, stingrays, and chimaeras is called inductive
sensing. The mollusks (Tochuina tetraquetra) orient themselves between
magnetic north and east before a full moon. The evidence is that magnetite
in the beak of pigeons accounts for their ability to navigate using
magnetoception.
The use of magnetic fields to navigate is found in magnetotactic bac-
teria. They are crucial for understanding other forms of magnetoception
in animals. Magnetotactic bacteria move according to the direction of the
North Pole, and hence they are magnetosensitive bacteria. The sensitivity
of magnetotactic bacterium to the earth’s magnetic field arises from the
fact the bacterium precipitates chains of crystals of magnetic minerals
DOI:10.1190/1.9781560803737

within its interior. These crystals, and sometimes the chains of crystals,
can be preserved in the geological record as magnetofossils. Reports of mag-
netofossils extend to 1.9 billion years ago.
Magnetism is an effect of quantum mechanics. At the subatomic level,
an electron acts as a tiny magnet. It is a rotating electrically charged particle.
Thus, to obtain a fine representation of the magnetic field, one must go
beyond geophysical prospecting, and descend into the subatomic level,
which is the realm of quantum mechanics. In that domain, traveling elec-
trons and other subatomic entities exist as matter waves (probability
waves), not as individual particles. Probability waves within the magneto-
tactic bacterium interfere and entangle in a state that is affected by the
earth’s magnetic field. Then, the resulting probability wave can influence
a chemical reaction at the molecular level. Such chemical reactions
provide the bacterium with the ability to navigate. The same type of behav-
ior is present in all animals that use magnetic navigation. As expressed by
Shakespeare, “In nature’s infinite book of secrecy, A little I can read.”

Diffraction
Interference refers to any combination or superposition of two or more
waves. In constructive interference, the superposition of two waves pro-
duces a greater disturbance than either wave acting alone. In destructive
interference, the superposition of two waves produces a disturbance
smaller than either wave acting alone. For example, if two sinusoidal
waves of the same frequency are in phase, then they add so that the wave
disturbance at each point in space is increased (constructive interference).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 297

If two sinusoidal waves of the same frequency are 1808 out-of-phase, then
the wave disturbance in the propagating medium is reduced (destructive
interference). If the magnitudes of two out-of-phase waves are the same,
the wave disturbance is completely canceled. A special type of interference
is produced by two waves of the same frequency and magnitude traveling in
opposite directions. The resultant wave pattern is stationary in space and is
called a standing wave. In a given structure, standing waves can exist only at
specific frequencies, which are called resonant frequencies.
Refraction refers to the change in direction of a propagating wave, such
as light or sound, in passing from one medium to another in which it has a
different velocity. In contrast, diffraction refers to various phenomena that
occur when a wave encounters an obstacle or a slit. It is defined as the
bending of waves around the corners of an obstacle or aperture into the
region of geometrical shadow of the obstacle. In classical physics, the dif-
fraction phenomenon is described as the interference of waves according
DOI:10.1190/1.9781560803737

to the Huygens–Fresnel principle that treats each point in the wave-front

as a collection of individual spherical wavelets. The amount of diffraction
depends on the wavelength: the longer the wavelength, the greater is the
spreading of the wave. Significant diffraction into the region behind the
obstacle occurs only if the size of the obstacle is smaller than the wave-
length. For example, if you sit behind a pillar in an auditorium, you still
hear the music because the long-wavelength sound waves spread behind
the pillar. On the other hand, you cannot see the performance. Your view
is obstructed because the wavelength of light is much smaller than the
pillar, so the light does not significantly diffract into the region behind the
pillar. Objects that are smaller than the wavelength do not produce signifi-
cant reflection. This is due to a lack of a substantial reflecting surface, so that
essentially all of the response can be attributed to diffraction. In other words,
the lack of reflection is due not to diffraction but to the inferior size and
shape of the object. The waves simply diffract around the small obstacle,
similar to flowing water spreading around a small rock.
Both diffraction and refraction are wave properties. Both represent some
sort of bending of waves. Due to the refraction of light waves, a spoon in a
glass of water appears to be bent. Water waves in a pond bend when they
encounter an obstacle. Generally, diffraction effects are most pronounced
when the dimensions of the obstacle are equal to or smaller than the wave-
length of the wave. When light waves are diffracted by a single slit, the result
is a diffraction pattern with bright and dark fringes. The central bright fringe
has the maximum intensity and width. Intensity of the fringes decreases as
we move along either side of the central maxima. Sound waves refract when
they cross two media. Refraction of light waves is the most common
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

298 Basic Wave Analysis, Part 3: Waveform Analysis

observation, because they produce strange optical illusions. The formation

of rainbows, splitting of white light by a glass prism, and mirages are
some examples. Both refraction and diffraction depend upon wavelength.
Thus, both can split white light in to its component wavelengths. In
summary, refraction is the bending of waves due to a change of velocity,
while diffraction is bending or spreading of waves around an obstacle.
Christiaan Huygens writes, “One may conceive light to spread succes-
sively, by spherical waves.” Huygens’ principle, as he put forward in
1678, is a geometrical construction in three spatial dimensions. The spheri-
cal wavelet is the essence of Huygens’ construction. The word “sphere”
refers to the spherical shell, and not to the insides of the shell. It is that infi-
nitesimally thin shell that reinforces the significance of Huygens’ principle.
We want to address two basic extensions. Both make full use of the analytic
tools of calculus, not yet developed in the time of Huygens. One extension is
the modification of Huygens’ principle to include the mathematics of dif-
DOI:10.1190/1.9781560803737

fraction. Essentially a weighting function (specifying amplitude and

phase) is assigned to the points on the spherical shell to take into con-
sideration its wave-like nature. This extension is sometimes called the
Huygens–Fresnel principle. The other extension is the modification of
Huygens’ principle to include the case of one spatial dimension and the
case of two special dimensions. The main attribute of the 3D Huygens’ prin-
ciple is that it permits the sharp transmission of a signal; that is, it does not
blur a signal by spreading it out into an unfamiliar form. In two dimensions,
we change the sphere to a circle. However, in two dimensions, the weighting
function is attached to the entire disc (i.e., to the circle itself and to the space
inside the circle). The result is that the main attribute of Huygens’ principle
is gone. More specifically, blurring occurs in the transmission of a signal in
two dimensions. In this sense, Huygens’ principle is not valid in 2D space. In
another sense, the wavelet is the essence of Huygens’ principle. We do not
want to mend the nature of Huygens’ principle, but we only want to change
the characteristics of the wavelet to meet the conditions at hand. The wavelet
itself represents the nature of wave propagation. Huygens’ principle is a con-
volutional model based upon the wavelet. With the appropriate wavelet,
Huygens’ principle is valid in any dimension. In three dimensions, the pro-
pagated signal is sharp. In two dimensions, it is not. That is the way of nature.
For example, Hadamard gives a solution in two dimensions in terms of
cylindrical waves such that Huygens’ principle (i.e., convolutional model)
still can be applied (see Baker and Copson, 1987).
Diffraction occurs with all waves, including seismic waves, sound
waves, water waves, and electromagnetic waves such as radio waves,
visible light, and x-rays. Diffraction is the bending of waves around small
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 299

obstacles and the spreading out of waves passing through small openings.
Francesco Grimaldi (1618–1663) first observed diffraction in the form of
the spreading of sunlight passing through a tiny pinhole in a screen.
Sir Isaac Newton maintained the theory that light is composed of tiny
particles. In 1678, Christiaan Huygens proposed that light is composed of
waves. Huygens’ principle states that:

Every point on a wavefront may be considered a source of second-

ary spherical wavelets which spread out in all directions with a
speed equal to the speed of propagation of the waves. The new
wavefront is the tangential surface to all of these secondary
wavelets.

In other words, the wavefront may be viewed as creating a series of

spherical wavelets. These wavelets combine in a linear fashion to continue
DOI:10.1190/1.9781560803737

the propagation of the wavefront. The new wavefront is the envelope of

these wavelets. More precisely, there are two envelopes; i.e., one in the
forward direction and the other in the backward direction. This arrangement
allows for the explanation of all possibilities, such as transmission, reflec-
tion, refraction, and diffraction. However, the rigorous mathematical
description of these possibilities could not be attained until much later by
Fresnel, Kirchhoff, and others. Often in scientific practice, many special
cases are found, from which the general case is developed. In the case of
wave propagation, Huygens proposed the general case, from which the
special cases are derived.
In the 18th century, most people, in agreement with Newton, did not
accept the theory that light is a wave. In 1800, Thomas Young (1773–
1829) presented a paper to the Royal Society which argues that light is
wave motion. His idea was greeted with a certain amount of skepticism
because it contradicted Newton’s corpuscular theory. Young believed that
a wave model could much better explain many aspects of light propagation
than the corpuscular model. In 1801, Young presented a famous paper to the
Royal Society, On the Theory of Light and Colours, which describes various
interference phenomena. One was his double-slit experiment. He sent light
through two closely spaced vertical slits and observed the resulting pattern
on the wall behind them. The pattern consisted of many vertical lines spread
out horizontally. Without interference, the light would simply make two ver-
tical lines on the screen. The wave characteristics of light caused the light to
pass through the slits and interfere with itself, producing the bright and dark
areas on the wall behind the slits. The double-slit experiment showed that
light is a wave. His work on the interference of waves changed the nature
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

300 Basic Wave Analysis, Part 3: Waveform Analysis

of light away from particles to waves. It was a momentous breakthrough in

the history of science. From that time forward, Young’s work was character-
ized by a single word—interference.
Huygens was correct, but he did not provide a laboratory experiment
such as that of Young. Instead, Huygens provided an astronomical test of
the validity of the wave theory of light. Huygens used observations of Io
(Jupiter’s third biggest moon, which is slightly larger than Earth’s moon)
to establish the optical (or relativistic) Doppler effect, which shows that
light is a wave. This result was confirmed by Albert Einstein in the develop-
ment of the special theory of relativity in 1905.
Augustin Fresnel (1788–1827) believed in the wave theory and used
diffraction to support it. Because illuminating a diffracting object will
produce a distinctive set of colored bands in the shadow, Fresnel
pasted a sheet of black paper to one edge of a diffractor to produce
changes in the shadow, i.e., the bright bands from the light vanished.
DOI:10.1190/1.9781560803737

This experiment enabled him to devise mathematical formulae for those

bright and dark bands; and subsequently his equations were useful for
predicting the interference patterns of light reflected from two mirrors.
In 1819, he submitted Memoir on the Diffraction of Light to the French
Academy of Sciences; unfortunately, the Academy mainly consisted of
Newton’s supporters who challenged Fresnel’s point of view in this way.
If light were indeed a wave, then diffraction from the edges of a sphere
would cause a bright area to occur at the center of the shadow of the
sphere. Ironically, this effect was observed later and became known as the
Fresnel Bright Spot or Poisson’s Spot (after Poisson, who was most critical
of Fresnel’s wave theory but later had to acknowledge its correctness).
Fresnel’s work was a momentous breakthrough in the history of science.
From that time forward, Fresnel’s work was characterized by a single
word—diffraction.
Actually, the experimental work of both Young and Fresnel involve
both interference and diffraction. According to Fresnel, “Nature is not
embarrassed by difficulties of analysis.” Diffraction occurs whenever
there is a discontinuity in the medium. Fresnel correctly interprets diffrac-
tion as the mutual interference of the secondary wave emitted from parts
of the wavefront which are not obscured by the diffraction screen.
However, interference also can occur where there is no diffraction, for
example, any occasional superposition of waves from different directions.
Diffraction is the bending of waves as they pass around the edge of an
object. The amount of bending depends on the relative size of the wave-
length compared to the size of the object. Often, the size of the object is
close in size to the wavelength; in such cases, the diffraction is noticeable.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 301

Diffraction is produced through the interference of the waves. If water

waves are incident upon an anchored boat, the boat would bounce up and
down in response to the incident waves, producing waves of its own. As
these waves spread outward in all directions from the boat, they interact
with other water waves. If the crests of two waves combine, an amplified
wave is produced (constructive interference). However, if a crest of one
wave and a trough of another wave combine, they cancel each other out
to produce less vertical displacement (destructive interference).
The study of diffraction can be categorized into two different types.
They are Fresnel diffraction and Fraunhofer diffraction. Fresnel diffraction,
or near-field diffraction, is a process of diffraction which occurs when a
wave passes through an aperture and diffracts in the near field. The near
field refers to the short distance through which the diffracted waves propa-
gate. When the distance is increased significantly, Fraunhofer diffraction
occurs. The outgoing diffracted waves become approximately plane
DOI:10.1190/1.9781560803737

waves and the rays become approximately parallel.

The Fresnel diffraction integral for Fresnel diffraction at point (x, y, z) is
given by
   
exp (ikz) 1 ′ ′ ik ′ 2
E(x, y, z) = E(x , y , 0) exp (x − x ) + (y − y) 2
dx′ dy′ .
ilz −1 2z
(10.5)
The propagating field originates at the aperture and moves along z. This inte-
gral is obtained by various approximations. In particular, the integral is valid
in a small area close to the origin, where the values of x and y are much
smaller than z. The Fresnel diffraction integral expressed in terms of a con-
volution is
E(x, y, z) = E(x, y, 0) ∗ h(x, y, z) , (10.6)
where the wavelet (i.e., the impulse response function) is
 
exp (ikz) ik  2 
h(x, y, z) = exp x +y 2
. (10.7)
ilz 2z
In other words, the propagation can be expressed as a convolutional
model. The geometric spherical wavelet originally specified by Huygens
in 1678 is now modulated (i.e., weighted) by amplitude and phase
characteristics.
Far-field diffraction is named Fraunhofer diffraction to honor Joseph
von Fraunhofer (1787–1826), a German physicist and optical lens innovator.
The far field refers to the long distance through which the diffracted waves
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

302 Basic Wave Analysis, Part 3: Waveform Analysis

propagate. When observed, the image of the aperture from Fresnel diffrac-
tion will change in terms of size and shape, i.e., the edges become more or
less jagged. In the case of Fraunhofer diffraction, the aperture image
observed only alters in terms of size due to the more collimated or planar
nature of the waves. In other words, in Fraunhofer diffraction, the light
source and an observation screen are effectively far enough from a diffrac-
tion aperture (e.g., a slit) that the wavefronts arriving at the aperture and the
screen can be considered to be collimated, or plane. Fresnel diffraction, or
near-field diffraction, occurs when this is not the case. Thus, the curvature
of the incident wavefronts must be taken into consideration. In order to
achieve agreement with experimental results, Fresnel asserts that the indi-
vidual contributions from the secondary wavelets have to be multiplied by
a constant and by an inclination factor. The various assumptions made by
Fresnel emerge automatically in Kirchhoff’s diffraction formula to which
the Huygens–Fresnel principle can be considered to be an approximation.
DOI:10.1190/1.9781560803737

Huygens’ principle says that each part of a wavefront acts as a source of

spherical wavelets which combine to yield the disturbance at the next
location. Huygens’ principle supports the use of geometric optics or geomet-
ric seismology. But, Huygens did not recognize diffraction! It is to the credit
of Fresnel to have completed Huygens’ principle and demonstrate its use for
diffraction. In effect, Huygens’ principle says the propagation of waves can
be expressed as superposition in a linear system. Fresnel’s essential contri-
bution is the demonstration of how the wavelets interfere with each other in
order to produce the wave motion. Let the given initial wavefront be the
input x and let the subsequent wavefront be the output y. The wavelet is rep-
resented by A. The specific form of A depends upon the considerations
imposed by the problem in question. When the wavefront is partially
obstructed, only those wavelets which belong to the exposed parts superpose
in such a way that the resulting wavefront has a different shape. This situ-
ation permits bending of light around the edges.
A good thing about free space propagation is that it is space-invariant.
Consequently, the system can be expressed by convolution. Huygens’ prin-
ciple takes the original wavefront x and convolves it with Green’s function A
to obtain the new wavefront y. The process is mathematically described by a
convolution given by
y=A∗x . (10.8)
In other words, the output y arises from the convolution (represented by
the asterisk) of the input x and the wavelet A. In microscopy, this convolu-
tion process mathematically explains the formation of an image that is
degraded by blurring and noise. Primarily, the blurring is due to
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 303

diffraction-limited imaging by the instrument. Typically, the noise is photon

noise, a term that refers to the inherent natural variation of the incident
photon flux. The degree of spreading (blurring) of a single point-like
object is a measure of the quality of an optical system. The 3D blurry
image of such a single point light source usually is called the point spread
function (PSF). Huygens’ principle formulates the propagation problem as
a linear system. This greatly simplifies the model and helps in the under-
standing of waves. In essence, the wavelet determines the form of propaga-
tion. The wavelet is known by many names, such as convolution kernel,
Green’s function, impulse response function, and point spread function.
Convolution implies replacing every original point on the wavefront by
its correspondent wavelet to produce the ensuing wavefront. Deconvolution
implies a restoration procedure that goes in the opposite way, i.e., gathering
the ensuing wavefront and putting it back into the original form. In the case
of known convolved image y and known wavelet A, deconvolution involves
DOI:10.1190/1.9781560803737

solving equation (10.8) in order to obtain the original image x, which is a

representation of the original state of affairs. In summary, the essential
knowledge of the transition is born by the convolution kernel (the
wavelet). Each particular representation (such as Fresnel diffraction and
Fraunhofer diffraction) has its own kernel.

Seismic convolutional model

Figure 10.3 appears in Huygens (1690) and shows how Huygens applied
his principle to reflection. Each point on the reflector AB acts as a source of a
spherical wavelet. For example, wavelet SN originates at reflector point A.
Whereas Huygens considers only one reflecting interface, the seismic
model considers many interfaces. Each interface has a reflection coefficient.
In a 3D earth, the waves travel in all directions; but, to keep the present
discussion simple, we consider the case of upgoing waves and downgoing
vertical waves. The simplest example of a traveling wave is primary reflec-
tion. Primary reflection is composed of the downgoing path from the source
to the reflector, and the returning upgoing path from the reflector to the re-
ceiver. Multiple reflection refers to an event that bounces back and forth
among various interfaces as it proceeds on its trip. Here, we consider the
most basic model, i.e., the model that admits only primary reflection and
neglects all transmission coefficients.
In geophysics, a digital wavelet is a signal that has finite energy. In other
words, a wavelet is a waveform with the bulk of its energy confined to a
finite interval on the time scale. The present time instant n ¼ 0 represents
the critical point in the classification of a wavelet. Past times would be all
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

304 Basic Wave Analysis, Part 3: Waveform Analysis

DOI:10.1190/1.9781560803737

Figure 10.3. Huygens’ reflection model. In the case of isotropic media (as in this
book), the curved lines represent spheres.

of the instants n , 0 before the present time. Future times would be all of the
instants n . 0 after the present time. A wavelet that has non-zero coeffi-
cients only for present and future times is said to be causal. A causal
wavelet can be written as
w = (w0 , w1 , w2 , w3 , · · ·) , (10.9)
where wn is the coefficient at discrete time n.
The theoretical division of the earth into thin layers produces a stratified
medium characterized by the interfaces between the layers. A time unit is
chosen, and the thickness of each layer is chosen as one-half of this time
unit. Each time that a downgoing wavelet encounters a horizontal interface,
a portion of the energy in the wavelet is reflected and the remainder is trans-
mitted. For the case of normal incidence (which we are treating here), the
reflected and transmitted waves have the same shape and breadth as the inci-
dent wave, but they differ in amplitude. The ratio of the amplitude of the
reflected wave to that of the incident wave is termed the reflection
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 305

coefficient. The sequence of reflection coefficients is called the reflectivity,

which is denoted by
r = (r1 , r2 , r3 , . . .) . (10.10)

The seismic trace is recorded at the surface. The trace is denoted by

y = (y1 , y2 , y3 , . . .) . (10.11)

The convolutional model consists of three components: the reflectivity

r, the wavelet w, and the reflection seismogram x (which has primary reflec-
tion only). The convolutional model is
y=r∗w . (10.12)

In analogy with Huygens’ principle, the reflectivity r acts as the source of

the reflected wavelets and the seismic trace y is the recipient of the wavelets.
DOI:10.1190/1.9781560803737

The convolutional model represents a linear system, with r as input, w as

impulse response function, and y as output. The important point to be
remembered is the analogy between Huygens’ diagram in Figure 10.3 and
the convolutional model shown in Figure 10.4. The inputs (reflecting
horizon versus reflectivity) correspond. The outputs (reflected wavefront

Figure 10.4. Convolutional model.

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

306 Basic Wave Analysis, Part 3: Waveform Analysis

versus reflection trace) correspond. The impulse functions (spherical

wavelet versus seismic wavelet) correspond. The wavelet is the winged mes-
senger traveling from input to output. Like the Greek god Mercury, the
wavelet can only travel forward in time.

Huygens’ principle
The term secondary wave dates back to Christiaan Huygens in 1688,
but the discourse evolved and the word “wavelet” came into use instead.
Huygens’ work on perfecting timepieces to determine accurate longitude
at sea yielded the pendulum clock. Huygens’ combination of mathematics,
mechanics, and optics within the realm of astronomy yielded his design and
construction of the telescope he used to discover Saturn’s fourth moon.
Beyond these practical instruments, Huygens established the wave theory
of light. Huygens’ principle, which appears in every elementary physics
DOI:10.1190/1.9781560803737

book and deserves repeating in this section, states: “Every point on a

primary wavefront serves as the source of secondary spherical wavelets.
These secondary wavelets advance with a speed and frequency equal to
that of the primary wave at each point in space. The primary wavefront at
some later time is the envelope of these secondary wavelets.” Huygens’
principle, in fact, states that the wavefront at some later time (i.e., the
output) is equal to the convolution of the wavefront at the initial time
(i.e., the input) with the wavelet (i.e. the impulse response). In other
words, Huygens’ principle is a spatial convolutional model in which the
wavelet plays the key role.
Modern mathematics and physics began in the early 1600s with the
works of Galileo and Descartes. A new picture of how the universe works
was emerging. This revolution was happening thanks to the work of a few
other brilliant scientists and thinkers. One was the Dutch mathematician
and scientist Christiaan Huygens. A prolific researcher and inventor, he
made critical contributions to a variety of scientific fields during his lifetime.
One of his greatest achievements was his report to the French Academy
written in 1678, which was later published as Treatise on Light (Huygens,
1690). In the late 1600s, Newton and Leibniz disclosed their work on the
systemic system of infinitesimal analysis universally known as calculus.
Jean le Rond d’Alembert, Leonhard Euler, Daniel Bernoulli, and Joseph-
Louis Lagrange used calculus to study the properties of a vibrating
string of a musical instrument. In 1746, d’Alembert discovered the one-
dimensional wave equation, and within ten years Euler discovered the
three-dimensional wave equation. In the 1830s, George Green developed
the mathematical concept now known as Green’s function. However, in
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 307

his book, Treatise on Light, Huygens geometrically had already solved the
three-dimensional scalar wave equation by use of the Green’s function. The
Green’s function used by Huygens was his spherical wavelet.
Let us turn to geophysics. Diephuis (2001) writes,

The paper which most immediately comes to mind is that in

which seismic migration was lucidly defined (Hagedoorn, 1954).
The oft-made statement that Hagedoorn, in his scientific oeuvre,
was inspired by the works of Christiaan Huygens is rather common-
place, and does not properly honour the depth to which Hagedoorn
tried to grasp the intentions and thoughts of his great compatriot. As
far as possible, Hagedoorn studied the works of the great scientists
such as Newton and Huygens in their original language. For once in
his life he deplored not having followed a classical education, and
thus being unable to read the Principiae in its original language.
DOI:10.1190/1.9781560803737

But he did study the Traité de la Lumière (Huygens, 1690) in

French. Coincidence or not, the scribbles Hagedoorn as professor
made on the blackboard bear a certain resemblance to Huygens’
notes.

Hagedoorn did not use numerical solutions. In 1954, Hagedoorn solved

the seismic migration problem geometrically. Seismic wavefront charts
show the successive positions of a wave emanating from a point source.
The Green’s functions used by Hagedoorn were obtained from seismic
wavefront charts. In the course of time with the ascendency of the digital
computer and with the great advances in seismic acquisition, it was inevita-
ble that numerical solutions would take precedence. The geometrical
methods of Huygens and Hagedoorn were put aside. By the year 2000,
three-dimensional seismic exploration became standard. The three-dimen-
sional wave equation in its various forms could be readily solved numeri-
cally on the computer.
Even so, the basic problem is that a computer is essentially a black box.
A black box is any complex piece of equipment, typically a unit in an elec-
tronic system, with contents that are mysterious to the user. Certainly, the
computer codes used in three-dimensional seismic processing are myster-
ious to the geophysicist. The geophysicist sees the input and output of the
computer. The geophysicist also should see pictures of the assorted wavelets
(i.e., Green’s functions) that are contrived by the numerical solutions of the
wave equations within the computer. Appreciation and understanding of
geometrical concepts are essential from a research point of view. In this
book, we give the basic concepts of Green’s functions. Such knowledge
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

308 Basic Wave Analysis, Part 3: Waveform Analysis

can be helpful in the construction of computer programs to produce pictures

of what the mathematics is doing inside the computer.
Let us give a preliminary analysis of Huygens’ principle with reference
to 3D space and the wavelet (i.e., Huygens’ secondary wave) as a sphere. A
sphere is a perfectly round geometrical object in 3D space that is the surface
of a completely round ball. Similar to a circular object in 2D space, a sphere
is defined mathematically in 3D space as the set of points that are all at the
same distance r from a given point. In other words, a sphere is a perfectly
round 2D surface in three dimensions. We sometimes use the term “spheri-
cal shell” instead of “sphere” in order to emphasize that the sphere is not a
solid ball but a perfectly round 2D surface. Let the waves move at velocity v.
At time t, the sphere will have radius vt. Let us draw a sphere of this radius
about each point on the wavefront that emitted a wave at a previous time t.
We obtain a pattern of wave motion. According to Huygens, the envelope of
this pattern gives the new wavefront. This construction may be called the
DOI:10.1190/1.9781560803737

heuristic version of Huygens’ principle. Its use is instructive in any dimen-

sion. It is the version used in geometrical optics for pedagogical purposes.
The strong version of Huygens’ principle (Huygens–Fresnel) has a con-
volutional representation in which the wavelet is the convolutional kernel
(i.e., Green’s function). As will be shown in Chapter 11, the Green’s func-
tions (i.e., weighting functions) for three dimensions, two dimensions, and
one dimension are, respectively,

d(vt − r) u(vt − r) u(vt − r)

, √ , and , (10.13)
4pvr 2pv v2 t2 − r 2 2v

where u is Heaviside step function. Note that only the 3D Green’s function
involves the impulse function d. Hence, the wavelet does not spread out but
remains sharp. The 2D and 1D Green’s functions involve the step function.
They spread out and do not stay sharp. They permeate into the interior of an
expanding bubble. The case of three dimensions is special—it allows an out-
going spherical wave to retain its sharp form as it travels.
The sharp propagation of a wave means that the leading and trailing
edges propagate at the same speed, so the wave does not get thicker. For
example, if a light bulb is illuminated for one second, someone viewing
the light from 1 km away will see it glow for only 1 s. We see sharp
images of stars. These images would be somewhat smeared out if the
wavelet were not a spherical shell in a space of three dimensions. If a
stone is dropped into a still pond, a 2D circular wave travels outward. The
surface of the water inside of the expanding wave does not remain perfectly
calm but undulates. The wavelet is sharp for waves propagating in three
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 309

dimensions, but it is not sharp in two dimensions or one dimension. In fact,

the wavelet is sharp in any odd dimension (except one dimension), and it is
not sharp in any even dimension.
For simplicity, consider the case of a 3D homogeneous isotropic
medium, so all of the secondary spherical wavelets have the same radius.
In this case, the medium represents a linear space-invariant system. In engin-
eering terms, the wavefront at the initial time is the system’s input, the
wavelet is the system’s impulse response, and the wavefronts at some
later time are the system’s output. There are two wavefronts: the forward
traveling wavefront and the backward traveling wavefront. Huygens
demonstrates the possibility of two waves traveling in opposite directions,
as shown in Figure 10.5. The solution at a given point P depends only on
the data on the sphere of radius v Dt. It does not depend upon data in the
interior of this sphere. Thus, the interior of the sphere is a lacuna for the sol-
ution. Indeed, in 1678, Huygens introduces the possibility of waves travel-
DOI:10.1190/1.9781560803737

ing in opposite directions. This result is in keeping with d’Alembert who

shows that a solution to the wave equation is a superposition of two
waves traveling in opposite directions.
JeanLe Rond d’Alembert was born in Paris in 1717, and died there in
1783. Nearly all of his major mathematical works were produced during
1743 to 1754. He enunciated the principle known by his name, that the
“internal forces of inertia” (that is, the forces which resist acceleration)
must be equal and opposite to the forces which produce the acceleration.
By use of d’Alembert’s principle, the differential equations of motion of
any rigid system can be obtained. In 1747, he applied differential calculus
to the problem of a vibrating string, arriving at the one-dimensional wave
equation and showing that it was satisfied by
u(x, t) = f (x − vt) + g(x + vt) , (10.14)

Figure 10.5.
Forward traveling
wavefront and
backward
traveling
wavefront.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

310 Basic Wave Analysis, Part 3: Waveform Analysis

where f and g are arbitrary functions. This solution u could be interpreted

as two waves with constant velocity moving in opposite directions along
the x-axis.
The numerical solution of partial differential equations is a vast topic.
Partial differential equations are at the heart of many computer analyses
or simulations of physical systems. Partial differential equations are classi-
fied into the three categories: hyperbolic, parabolic, and elliptic. A typical
hyperbolic equation is the wave equation. A typical parabolic equation is
the diffusion equation. A typical elliptic equation is Poisson’s equation.
The solution u(x, t) of the hyperbolic and parabolic equations are func-
tions of time. Such problems are initial value or Cauchy problems: if infor-
mation on u (which may include time derivatives) is given at some initial
time t0 for all x, then the equations describe how u(x, t) propagates
forward in time. In other words, hyperbolic and parabolic equations describe
time evolution. The objective of numerical analysis is the tracking of that
DOI:10.1190/1.9781560803737

time evolution with a desired accuracy.

By contrast, the elliptic equation has a single static solution u(x, y). In
order to find the solution, we must specify a desired behavior on the bound-
ary of the region of interest. Such problems are called boundary value prob-
lems. In general, it is not possible to simply integrate from the boundary in
the same sense that an initial value problem can be “integrated forward in
time.” Therefore, the goal of a numerical code is somehow to converge
on the correct solution everywhere at once. In seismology, we are concerned
with the wave equation, which is of the hyperbolic type. The problem is an
initial value (time evolution) problem with boundary values at the edges
marking the region under consideration.
For example, consider the bottom row shown in Table 10.1, in which we
specify initial values I on one time slice. The objective it to compute succes-
sive rows of upward arrows " in their direction to advance the solution in
time. Boundary conditions B at the left and right edges of each row also
must be supplied, but only one row at a time. Only a few previous rows
need be maintained in memory. The stability of the algorithm is essential,

Table 10.1. Initial value problem with edge boundaries.

B " " " " " " B

B " " " " " " B
B " " " " " " B
B " " " " " " B
I I I I I I
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 311

because many algorithms which may appear to be reasonable are actually

numerically unstable. Stability may be difficult to achieve for initial value
problems but more easily produced with boundary value problems.

One-dimensional wave equation

The Cartesian space coordinates are x, y, z, the time coordinate is t, and
the velocity is v. The position vector is defined by r = x ex + y ey + z ez .
The symbol ∇ stands for the vector differential operator “del” or “nabla”
(named after a Phoenician harp-like musical instrument) given by
 
∂ ∂ ∂ ∂ ∂ ∂
∇= , , = ex + ey + ez . (10.15)
∂x ∂y ∂z ∂x ∂y ∂z

The Laplace operator or Laplacian is the differential operator for the

DOI:10.1190/1.9781560803737

divergence of the gradient of a function. It is denoted by the symbols

D, or ∇2 , or ∇ · ∇. It is

  
∂ ∂ ∂ ∂ ∂ ∂ ∂2 ∂2 ∂2
D =∇ =∇·∇=
2
, , · , , = 2+ 2+ 2 .
∂x ∂y ∂z ∂x ∂y ∂z ∂x ∂y ∂z
(10.16)
Let u denote the amplitude of the wave. The wave equation (in one, two,
or three dimensions) can be written as

1 ∂2 u
∇2 u = . (10.17)
v2 ∂t2
The wave equation (10.17) arises in acoustics, geophysics, elasticity, and
electromagnetics, and its basic properties and solutions were developed
originally in the areas of classical physics. The wave equations considered
in this book are linear. We should add that our wave equations are perfectly
elastic, have no damping term (proportional to ∂u ∂t ), and do not describe
attenuation. Often, the solution of a real-world problem requires the linear-
ization of some nonlinear partial differential equation. For example, in
acoustics, one starts with a set of nonlinear equations for a compressible
fluid. Then, the equations are linearized by assuming that all wave disturb-
ances are explained by small perturbations about an equilibrium state. It is
these small disturbances that satisfy a linear wave equation of the form
equation (10.17). Despite its limitations in geophysical modeling, wave
equation (10.17) provides insight into the complexities of wave propagation.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

312 Basic Wave Analysis, Part 3: Waveform Analysis

An intrinsic complexity results from the use of realistic velocity functions

v(x, y, z) which represent the subsurface in all of its geologic intricacies.
This embodies a first step. The wave equation (10.17) is called an acoustic
wave equation because it only accounts for compressional (P) wave propa-
gation. Realistic geophysical models require more elegant elastic wave
equations, which also include shear (S) waves together with surface
waves. Wave functions are coupled through appropriate boundary con-
ditions. The wave equation (10.17) represents the most basic approach to
geophysical analysis.
Waves satisfying the simple relation
v
= constant (10.18)
k
are called nondispersive waves. This is the case usually assumed in explora-
tion seismology. When v/k depends on wavelength (and hence on the fre-
DOI:10.1190/1.9781560803737

quency), the waves are called dispersive. In this case, the waveform changes
during propagation.
In one dimension (the x-axis), the (homogeneous) 1D nondispersive
wave equation is

∂2 u 1 ∂2 u
− = 0 with v = positive constant . (10.19)
∂x2 v2 ∂t2
Note that the factor 1/v2 can be memorized by comparing units, i.e., m−2 , for
both terms in equation (10.19). This equation arises in many problems,
examples of which are linearized supersonic flow, sound waves in a tube
or pipe, long water waves in a straight canal, and the transmission of elec-
tricity along an insulated, low-resistance cable. The wave velocity v is
taken as a constant.
The sounds of music and speech have been deeply rooted in our
evolutionary past. The study of sound began in ancient times. As early as
the 6th century B.C., Pythagoras conducted experiments on the sounds pro-
duced by vibrating strings. His most significant contribution to acoustical
theory was establishing the inverse proportionality between pitch and the
length of a vibrating string. The one-dimensional wave equation describes
waves traveling on a stretched string.
In order to see more clearly the mathematical aspects of wave pro-
pagation, let us proceed to solve (10.19). First, we introduce the change
of variables
a = x − vt , b = x + vt . (10.20)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 313

The quantity a represents the value of the x coordinate in a stationary

frame of reference at time t measured by an observer moving toward the
positive values of x with a speed v. The two coordinate systems must
have coincided at time t = 0. Similarly, the quantity b represents the
value of the x coordinate in a stationary reference frame at time t measured
by an observer moving toward the negative values of x with a speed v. These
two coordinate systems also must have coincided at time t = 0.
Solving equations (10.20) for x and t in terms of a and b, we obtain
a+b b−a
x= , t= . (10.21)
2 2v
Now, let us take derivatives of function u = u(x, t) with respect to a and b.
We find that
 2 
∂2 u ∂2 u ∂2 u ∂2 u ∂2 u 2 ∂ u ∂2 u
= + + and =v + . (10.22)
DOI:10.1190/1.9781560803737

∂x2 ∂a2 ∂a ∂b ∂b2 ∂t2 ∂a2 ∂b2

Substitution of equations (10.22) into the wave equation (10.5) yields

∂2 u
=0. (10.23)
∂a ∂b
From equation (10.23), we deduce that u does not contain any term which
depends on both a and b. Note that ∂u/∂b is independent of a and that
∂u/∂a is independent of b. Thus, we may write the solution to equation
(10.23) directly as
u = g(b) + f (a) . (10.24)

In order to explain the significance of the arbitrary functions f (a) and

g(b) and the physical meaning of the solution equation (10.24), let us
change back to the variables x and t. Thus, using the relations (10.20), the
solution to equation (10.24) becomes
u(x, t) = f (x − vt) + g(x + vt) (10.25)

as the general solution to the basic wave equation (10.17). The component
f (x − vt) represents a displacement wave of arbitrary shape traveling in
the positive x direction at a constant speed v without altering its shape.
Similarly, the component g(x + vt) is a displacement wave traveling in
the negative x direction with the same speed and it also has an unaltered
wave shape. Hence, every motion of the string consists of a superposition
of two waves traveling in opposite directions.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

314 Basic Wave Analysis, Part 3: Waveform Analysis

An important case of equation (10.25) is the sinusoidal (or harmonic)

wave
2p
u(x, t) = A cos (x − vt) (10.26)
l
traveling in the positive x direction at a constant speed v. Here, l is the
wavelength, which is the distance between two successive crests. Mathemat-
ically, the wavelength is
2p
l=v = vT , (10.27)
v
where v is the angular (or circular) frequency of the wave and T is the period
necessary for a complete cycle to pass a given point. The cyclical frequency
is f = 1/T, which is related to the angular frequency v by the equation
v = 2pf . The number of waves in a unit distance is called the cyclical
DOI:10.1190/1.9781560803737

wavenumber and is given by 1/l. However, we usually make use of the

angular wavenumber k. The angular wavenumber k is 2p times the cyclical
wavenumber. Equation (10.26) can be rewritten as
u(x, t) = A cos (kx − vt) = A cos (vt − kx) . (10.28)
Next, consider what happens when two sinusoidal waves of equal
amplitude A and angular frequency v travel in opposite directions. The
superposition of the two waves is

u(x, t) = A cos (vt − kx) + A cos (vt + kx) , (10.29)

where A cos (vt − kx) is a sinusoidal wave traveling in the positive x direc-
tion and A cos (vt + kx) is a sinusoidal wave traveling in the negative x
direction. Both waves have the same amplitude A and travel at the same
speed v. However, by using the trigonometric identity

cos (a + b) = cos a cos b + sin a sin b , (10.30)

we can rewrite equation (10.29) as

u(x, t) = 2A cos vt cos kx . (10.31)

Equation (10.31) is significant because it is not a function with argument

x − vt or x + vt. Thus, we conclude that the combination of these two sinu-
soidal waves does not produce a traveling wave. Thus, for this case, the two
traveling waves combine into a stationary or standing wave. From equation
(10.31), we observe that, at points where cos kx = 0, the two sinusoidal
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 315

traveling waves cancel each other, giving a zero displacement u at these

points. These points are called nodes. At points where cos kx = +1, the
two traveling waves add, yielding the greatest amplitude. These points,
which are halfway between any two successive nodes, are called loops or
antinodes.
Now, we can recapitulate our development of wave motion for the
stretched string. Suppose that one presses down at a certain point on a
stretched string. An equilibrium state is reached such that the string’s
elastic forces at the pressure point exactly balance the applied force.
Then, if we remove the applied pressure, the pressure point will no longer
be in equilibrium, although at that instant all of the neighboring points are
unchanged. However, as the material in the string at the former pressure
point moves to restore equilibrium, the material at the neighboring points
is disturbed. As a result, the neighboring points in turn move and the disturb-
ance propagates outward. Consider the type of motion that we expect to
DOI:10.1190/1.9781560803737

obtain as a result of this disturbance. We might expect the disturbance to

be a wave of some sort that can travel outward with a fixed velocity and
be unchanged in form from point to point. In mathematical terms, the
wave form at time t and position x is the same as at time zero and position
x − vt, where v is the wave velocity. Thus, we have
u1 (x, t) = u1 (x − vt, 0) = u1 (a, 0) = f (x − vt) where a = x − vt .
(10.32)
Differentiating equation (10.32) with respect to x and t, respectively,
we obtain

∂u1 ∂u1 ∂u1 ∂u1

= , = −v . (10.33)
∂x ∂a ∂t ∂a
A second differentiation produces the two equations
∂2 u1 ∂2 u1 ∂2 u1 2
2 ∂ u1
= , = (−v) . (10.34)
∂x2 ∂a2 ∂t2 ∂a2
This combines into
∂2 u1 2
2 ∂ u1
= (−v) , (10.35)
∂t2 ∂x2
which is the basic wave equation. The reason why the basic one-dimensional
wave equation (10.35) is known as the nondispersive wave equation is
because the solution equation (10.32) shows that the form of the wave
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

316 Basic Wave Analysis, Part 3: Waveform Analysis

does not change as the wave progresses, that is, because there is no dis-
persion of the wave shape during propagation.
Alternatively, a wave can travel in the opposite direction with the same
speed. That is,

u2 (x, t) = u2 (x + vt, 0) = u2 (b, 0) = g(x + vt) where b = x + vt .

(10.36)

If we use the same steps as previously, we find that

∂2 u2 2
2 ∂ u2
= (−v) , (10.37)
∂t2 ∂x2

which is the same basic wave equation. Thus, the general solution of the
DOI:10.1190/1.9781560803737

basic nondispersive wave equation is the sum

u(x, t) = u1 (x − vt, 0) + u2 (x + vt, 0) = f (x − vt) + g(x + vt) . (10.38)

In equation (10.38), f (x − vt) is simply a delayed version of f (x), where

the delay is given by vt (see Figure 10.6). Likewise, g(x + vt) is an advanced
version of g(x), where the advance is given by vt. Thus, in the general sol-
ution equation (10.38), the two terms f (x − vt) and g(x + vt) are interpreted
physically as waves of arbitrary but permanent shape, traveling with speed v
in the direction of the positive or negative x-axis, respectively.

Figure 10.6. Motion of a nondispersive wave shape as a function of x for a

sequence of time t = 0, t = Dt, t = 2Dt. Here, the wave shape is moving to the right
at a constant speed v.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 317

Initial value problem for the wave equation

Now, we can address our first important problem concerning the one-
dimensional basic wave equation. It is the initial value problem. The quan-
tity u(x, t) denotes the displacement from equilibrium of a particle of the
string at position x and time t. Thus, ∂u/∂t represents the time rate of
change of this displacement, or, in words, particle velocity at position x
and time t. Particle velocity ∂u/∂t should not be confused with wave
velocity v.
Next, consider a mathematical statement of the initial value problem.
We are given that the wave motion u(x, t) of an infinitely long string satisfies
the one-dimensional nondispersive wave equation

∂2 u 1 ∂2 u
− = 0 with v = constant . (10.39)
∂x2 v2 ∂t2
DOI:10.1190/1.9781560803737

We do not know the solution u = u(x, t). However, we are given the initial
displacement shape

u(x, 0) = f(x) (10.40)

and the initial particle velocity shape

∂u(x, t)
= c(x) (10.41)
∂t t=0

for all values of x . The quantity on the left in equation (10.41) is a little cum-
bersome, so we abbreviate it as

∂u(x, t) ∂u(x, 0)
= .
∂t t=0 ∂t

The notation ∂u(x, 0)/∂t does not mean the derivative of u(x, 0) with
respect to t, but instead it is a shorthand notation for the value of
∂u(x, t)/∂t evaluated at t = 0. Here, we consider a string of infinite length
in order to avoid the complicating effect of end conditions, so the domain
of x is −1 , x , 1.
Now, we seek a solution u(x, t) of the wave equation for 0 , t , 1 and
−1 , x , 1 that depends on the initial conditions f(x) and c(x) instead of
the two arbitrary functions f (x) and g(x) in the general solution

u(x, t) = f (x − vt) + g(x + ct) . (10.42)

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

318 Basic Wave Analysis, Part 3: Waveform Analysis

The use of two arbitrary functions f (x) and g(x) is an awkward way of
describing the initial shape u(x, 0), because, when we observe the initial
shape of the total wave, we cannot disentangle the wave moving to the
right from the wave moving to the left. Mathematically, it is better to
describe the solution in terms of the two initial conditions (10.40) and
(10.41), that is, the initial shape (i.e., displacement) f(x) and the initial ve-
locity of each particle of the string c(x). To understand how knowledge of
these initial conditions allows us to obtain the waveform u(x, t) without
reconstructing f (x) and g(x), first note that

u(x, 0) = f(x) = f (x) + g(x) . (10.43)

If

a = x − vt, b = x + vt , (10.44)
DOI:10.1190/1.9781560803737

then

u(a, b) = f (a) + g(b) . (10.45)

Hence,

∂u(a, b) ∂u ∂a ∂u ∂b ∂u ∂g
= + = (−v) + v, (10.46)
∂t ∂a ∂t ∂b ∂t ∂a ∂b

so

∂u(a, b) ∂f ∂g
= c(x) = v − + (10.47)
∂t t=0 ∂a t=0 ∂b t=0

or

∂u(a, b) ∂f (x) ∂g(x)

= c(x) = v − + . (10.48)
∂t t=0 ∂a ∂b

Integrating equation (10.48) from 0 to x, we obtain

x
1
−f (x) + g(x) = c(x′ )dx′ + g(0) − f (0) . (10.49)
v 0
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 319

Now, if we add equations (10.43) and (10.49), we obtain

x
1 1 g(0) − f (0)
g(x) = f(x) + c(x′ ) dx′ + . (10.50)
2 2v 0 2

Substitution of equation (10.50) into either equation (10.43) or (10.49)

produces
x
1 1 g(0) − f (0)
f (x) = f(x) − c(x′ )dx′ − . (10.51)
2 2v 0 2

In principle, knowledge of the initial conditions f(x) and c(x) has

allowed us to reconstruct the functions f (x) and g(x), provided we know
f (0) and g(0). By comparing equations (10.42), (10.50), and (10.51), we
deduce that
DOI:10.1190/1.9781560803737

x−vt x+vt
1 1 1 1
u(x, t) = f(x − vt) − c(x′ )dx′ + f(x + vt) − c(x′ )dx′
2 2v 0 2 2v 0
(10.52)

or

x+vt
1 1 1
u(x, t) = f(x − vt) + f(x + vt) − c(x′ )dx′ , (10.53)
2 2 2v x−vt

which does not depend on either f (0) or g(0). Equation (10.53) expresses the
solution u(x, t) for 0 , t , 1 and −1 , x , 1 in terms of the initial con-
ditions (or functions) f(x) and c(x). This equation is known as the d’Alem-
bert solution of the one-dimensional wave equation (10.39). At this point,
we have solved the initial value problem for wave motion on an infinitely
long string.
Now, we can interpret d’Alembert’s solution equation (10.53). We
recall that u(x, t) includes two linear waves: one moving in the positive
x direction with wave speed v and the other moving in the negative x direc-
tion with wave speed v. For ease of presentation, we assume that v is posi-
tive. Then, the amplitude u(x, t), at a typical point (x, t) of space-time,
involves the initial amplitude only at the two places, x − vt and x + vt,
from which waves of wave velocity v would arrive after time t. The ampli-
tude u(x, t) also depends on the initial particle velocity shape c throughout
the interval (x − vt, x + vt). Thus, in the x, t plane, we have two families of
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

320 Basic Wave Analysis, Part 3: Waveform Analysis

lines, i.e.,
x − vt = constant and x + vt = constant . (10.54)
These lines are known as characteristics of the wave equation. Through
any fixed point, e.g., (x1 , t1 ), of space-time, there pass two characteristic
lines, i.e.,
x − vt = x1 − vt1 and x + vt = x1 + vt1 . (10.55)
The coordinates a, b, given by
a = x − vt, b = x + vt , (10.56)
are called the characteristic coordinates. Considering Figure 10.7, we
observe that the general solution

u(x, t) = f (x − vt) + g(x − vt) (10.57)

DOI:10.1190/1.9781560803737

can be written as

u(x, t) = f (a) + g(b) , (10.58)

where each term on the right side of equation (10.58) is constant along the
characteristic line of one of the two families. If we compare this with our
previous description of traveling waves, we see that features of the wave
shape, such as a peak or a trough, are extended along characteristic lines

Figure 10.7. Characteristics of the basic wave equation. Through any fixed point
(x1 , t1 ) of space-time, there pass two characteristic lines x − vt = x1 − vt1 and
x + vt = x1 + vt1 . These lines are known as characteristics of the wave equation.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 321

in the space-time diagram. That is, the locus of a peak, for example, is a
characteristic line.
Next, consider d’Alembert’s formula in terms of the characteristic
lines. Through the space-time point (x1 , t1 ) pass two characteristic lines,
expressed in equations (10.55). They intersect the initial locus t = 0 in the
places

x = x1 − vt1 and x = x1 + vt1 . (10.59)

These places define the domain of the datum functions f and c upon which
u(x1 , t1 ) depends in d’Alembert’s formula:
x1 +vt1
1 1 1
u(x1 , t1 ) = f(x1 − vt1 ) + f(x1 + vt1 ) − c(x) dx . (10.60)
2 2 2v x1 −vt1
DOI:10.1190/1.9781560803737

Thus, the domain of dependence of u(x1 , t1 ) on f(x) is the set of two points
x1 − vt1 and x1 + vt1 and the domain of dependence on c(x) is the interval
between the two points. It is apparent that the domain of dependence is
bounded by the characteristics.
This general property has the physical interpretation that the waves
cannot travel faster than v. For example, suppose that we want to use
the string as a signaling device. We impose some sort of local deformation
f on the string and then release the string from rest at t = 0, so c = 0. The
velocity that our signal moves along the string is exactly v, so this represents
the speed of the signal transmission. Now, let us start with another type
of signal, e.g., one for which f = 0 and for which c is prescribed. Thus,
we have
x1 +vt1
1
u(x1 , t1 ) = − c(x) dx . (10.61)
2v x1 −vt1

This formula means that u(x1 , t1 ) is affected only by those portions of

the x-axis from which signals traveling no faster than those with a velocity
v can reach the point x1 . Thus, v represents the maximum velocity for signal
propagation. In this manner, we have illustrated a very important aspect of
the one-dimensional nondispersive wave equation—solutions (i.e., signals)
travel with finite constant speed v and they do not progressively change in
shape as they travel (i.e., there is no dispersion of the wave shape).
At this point, we could ask other questions. For example, do such signals
travel no slower than those with a velocity v? How do we see from equation
(10.61) that v is not only the maximum velocity but the velocity itself?
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

322 Basic Wave Analysis, Part 3: Waveform Analysis

The answers to such questions can be obtained by examining the Green’s

function of the wave equation in question. For example, the Green’s func-
tion of the deconvolution operator is the seismic wavelet. Such a seismic
wavelet can be computed numerically. The seismic wavelet is a valuable
resource in its own right. Because this book deals with basic wave analysis,
we only consider the scalar wave equation. The mathematical theory
becomes quite difficult for more general wave equations. However, there
is recourse. The Green’s function (i.e., wavelet) for any numerical wave
equation can be computed numerically.
In summary, the general solution, equation (10.42)—u(x, t) =
f (x − vt) + g(x + ct)—of the one-dimensional nondispersive wave equation
∂2 u 1 ∂2 u
(10.39)— 2 − 2 2 = 0, with v ¼ constant—is the sum of two terms:
∂x v ∂t
f (x − vt), which describes a wave moving toward larger values of x, and
g(x + vt), which describes a wave moving toward smaller values of x.
DOI:10.1190/1.9781560803737

In this section, we have shown how these waves also can be expressed
in terms of the initial displacement shape given in equation (10.40)—
u(x, 0) = f(x)—and the initial particle velocity shape given in equation
∂u(x, 0)
(10.41)— = c(x)—of each point of an infinitely long string. That
∂t
is, d’Alembert’s solution equation (10.53)—u(x, t) = 12 f(x − vt) +

1 x+vt
2 f(x + vt) − 2v
1
c(x′ )dx′ —which is valid for 0 , t , 1 and −1 ,
x−vt
x , 1 solves the initial value problem for the one-dimensional nondisper-
sive wave equation associated with an infinitely long string.
We remind the reader that the velocity v of propagation of the wave
(which is not to be confused with the particle velocity of each point of the
string) depends on the medium and, in this case, is equal to

tension
v= . (10.62)
mass density

This expression for velocity is more general than one would think. In
all media, it consists of the square root of the quotient of two terms; the
numerator is a quantity describing the elastic properties of the medium
and the denominator describes the inertial or mass properties of the
medium. The one-dimensional nondispersive wave equation results from
the assumption that each of these two terms is constant in the string so
that the velocity v of the wave is constant along its length. However, if
either or both terms would vary along its length, then the velocity of the
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 323

wave would be variable. In fact, this is the more realistic case, because
wave propagation in stratified media can be described by a one-dimen-
sional wave equation with variable velocity v(x). For example, we can
examine the more complicated situation in which the velocity v depends
on position x, i.e., the case of an inhomogeneous medium. To solve the
variable velocity wave equation, typically v(x) is considered to be piece-
wise constant or stratified. This allows the original wave equation to be
considered as a set of constant velocity wave equations. Much effort has
been expended upon this approach.

Two-dimensional wave equation

The two-dimensional wave equation involves spatial dimensions x, y;
that is,
∂2 u ∂2 u 1 ∂2 u
DOI:10.1190/1.9781560803737

+ = . (10.63)
∂x2 ∂y2 v2 ∂t2
It is often convenient to consider the solution of the two-dimensional wave
equation in polar coordinates r, f instead of Cartesian coordinates x, y. For
this case, as shown in Figure 10.8, the transformation equations are
x = r cos u and y = r sin u . (10.64)
In polar coordinates r, f, the Laplacian operator becomes
 
1 ∂ ∂ 1 ∂2
∇2 = r + 2 . (10.65)
r ∂r ∂r r ∂u2
The two-dimensional wave equation for u(r, u) is
 
1 ∂ ∂u 1 ∂2 u 1 ∂2 u
r + 2 = . (10.66)
r ∂r ∂r r ∂u2 v2 ∂t2

Figure 10.8. Two-

dimensional polar
coordinates.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

324 Basic Wave Analysis, Part 3: Waveform Analysis

Now, consider the special case in which the wave function u(r, u) in
two-dimensional polar coordinates is independent of u. Because u does
not depend on u, this equation reduces to
 
1 ∂ ∂u 1 ∂2 u
r = 2 . (10.67)
r ∂r ∂r v ∂t2
In this case, we say that the two-dimensional wave propagation is circularly
(or cylindrically) symmetric. Expanding the partial derivatives in equation
(10.67) produces

∂2 u 1 ∂u 1 ∂2 u
+ = , (10.68)
∂r 2 r ∂r v2 ∂t2
which has the appearance of a one-dimensional wave equation. Based on our
previous experience with the one-dimensional wave equation, we might try
DOI:10.1190/1.9781560803737

the coordinate transformations

a = r − vt , b = r + vt (10.69)

and obtain a wave equation in a, b coordinates. It is easy to show that this

procedure results in the wave equation
 
∂2 u −1 ∂u ∂u
= + . (10.70)
∂a ∂b 2(a + b) ∂a ∂b

Unlike the case for the basic one-dimensional nondispersive wave

equation, the right side of equation (10.70) is not zero. As a result, we do
not obtain the desired solution of the form

u(a, b) = g(b) + f (a) , (10.71)

which would give the superposition of two linear waves that do not change
their shape. Thus, for the two-dimensional case, circularly symmetric wave
propagation is dispersive.
At this point, we have come to an enigmatic result in physics. The prop-
agation of waves in a space with an even number of dimensions differs in a
profound way from wave propagation in space with an odd number of
dimensions. In the remainder of this book, we will examine this result in
terms of the classical scalar wave equation. A mathematical wrap up is
the following. Look ahead at Figure 11.11, “Huygens’ wavelet for r ≥ 0
with v = 1, t = 1, so vt = 1.” In the 3D case of the figure, the wavelet is
non-zero only in an infinitesimal neighborhood of the sphere r = vt. As a
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 325

result, in three dimensions, Huygens’ principle says that information from a

point source travels in the form of a sphere. The wavefront is thus sharp, with
a sudden onset at the start, and sudden cutoff at the end. In the 2D case of the
figure, the wavelet is non-zero in the entire interior r ≤ vt of the sphere. As a
result, in two dimensions, the behavior is different. Wavefronts do have a
sharp onset, but they decay with a long tail. As will be shown,
Figure 11.13 depicts the two-dimensional Huygens’ wavelet.
An often-cited example is water waves in a pond, because they will
leave a train of waves behind them. Of course, when the wave propagation
depends on f, we can expect an even more complicated wave motion. The
water wave is a surface wave, whose propagation is two dimensional, but
the particle motions are in fact three dimensional, with a damping amplitude
as the particle’s position gets deeper below the water surface. Therefore, the
particle motion is not exactly described by equation 10.63 or 10.70.
However, the particle motion described by equation 10.63 will be given
DOI:10.1190/1.9781560803737

in Chapter 12.
The story is told that Huygens discovered his principle by dropping
pebbles into the still water of a canal. The problem is that waves do not
stay sharp in even dimensions. They smooth out, as energy put into the
interior of an expanding ring. How did Huygens know that, in three dimen-
sions, all of the energy is concentrated in the infinitesimally thin outgoing
spherical wave?
As commonly assumed in exploration, seismic waves are not dispersive.
On the contrary, water waves as observed on the surface of a still pond
obviously are dispersive. With seismic waves, all frequencies travel at the
same speed in a homogeneous and isotropous medium. This means that
any waveform you create will keep its shape as it travels. The shape will
be the same whether the source is 100 or 1000 m away. With water
waves, the frequencies do not travel at the same speed, and the water
waves in a pond leave a train of waves behind them.

Three-dimensional wave equation

The three-dimensional wave equation involves spatial coordinates
x, y, z; that is,
∂2 u ∂2 u ∂2 u 1 ∂2 u
+ + = . (10.72)
∂x2 ∂y2 ∂z2 v2 ∂t2
Similarly, as shown in Figure 10.9, we may consider the three-dimensional
wave equation in spherical coordinates r, u, f. The transformation
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

326 Basic Wave Analysis, Part 3: Waveform Analysis

Figure 10.9. The three-dimensional

wave equation.

equations are
x = r sin u cos f, y = r sin u sin f, z = r cos u . (10.73)
DOI:10.1190/1.9781560803737

In spherical coordinates, the Laplacian is

1 ∂ 2 ∂ 1 ∂ ∂ 1 ∂2
∇2 = r + sin u + . (10.74)
r 2 ∂r ∂r r 2 sin u ∂u ∂u r 2 sin u ∂f2

Now, we can address to the three-dimensional wave equation

   
1 ∂ 2 ∂u 1 ∂ ∂u 1 ∂2 u 1 ∂2 u
r + 2 sin u + 2 = ,
r 2 ∂r ∂r r sin u ∂u ∂u r sin u ∂f2 v2 ∂t2
(10.75)

which describes the wave function u(r, u, f) in three-dimensional spherical

coordinates. If we assume that the wave function u is independent of the
angles u and f, the three-dimensional wave equation (10.75) reduces to
 
1 ∂ ∂u 1 ∂2 u
2
r = 2 , (10.76)
r 2 ∂r ∂r v ∂t2

which is equivalent to

∂2 u 2 ∂u 1 ∂2 u
+ = . (10.77)
∂r 2 r ∂r v2 ∂t2

Equation (10.77) describes a three-dimensional wave motion which is

spherically symmetric, because u is independent of u and f. Again, if we
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 327

attempt to obtain a wave equation in a, b coordinates, we produce

 
∂2 u −1 ∂u ∂u
= + . (10.78)
∂a ∂b (a + b) ∂a ∂b

Unfortunately, we do not obtain the desired form

∂2 u
=0. (10.79)
∂a ∂b

Equation (10.68) of the foregoing section and equation (10.77) here

describe the simplest types of two-dimensional and three-dimensional
wave propagation, respectively.
To this point, we have considered only a transformation of the
independent variables, that is, the transformation (10.69) of the foregoing
section mapped r, t into a, b. As we have shown, this is a convenient
DOI:10.1190/1.9781560803737

transformation for one-dimensional wave propagation. However, in

two- and three-dimensional wave propagation, it does not appear to be
useful. So, now we can consider the idea of transforming the dependent vari-
able. That is, we consider the functions F(a, b) and G(a, b) such that

u(a, b) = F(a, b)G(a, b) . (10.80)

Thus,

∂u ∂G ∂F ∂u ∂G ∂F
=F +G , =F +G . (10.81)
∂a ∂a ∂a ∂b ∂b ∂b

Further differentiation produces

∂2 u ∂2 G ∂F ∂G ∂2 F ∂G ∂F
=F + +G + . (10.82)
∂a ∂b ∂a ∂b ∂a ∂b ∂a ∂b ∂a ∂b

Substituting equations (10.81) and (10.82) into the transformed three-

dimensional spherically symmetric wave equation (10.78) produces

∂2 G ∂F ∂G ∂2 F ∂G ∂F
F + +G +
∂a ∂b ∂a ∂b ∂a ∂b ∂a ∂b
 
−1 ∂G ∂F ∂G ∂F
= F +G +F +G , (10.83)
(a + b) ∂a ∂a ∂b ∂b
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

328 Basic Wave Analysis, Part 3: Waveform Analysis

which we choose to write as

 
∂2 F −1 F ∂G ∂G ∂F ∂F 1 ∂G ∂F
= + + + −
∂a ∂b (a + b) G ∂a ∂b ∂a ∂b G ∂a ∂b
1 ∂F ∂G F ∂2 G
− + . (10.84)
G ∂a ∂b G ∂a ∂b

If we can find a function G(a, b) where G . 0 such that the right side of
equation (10.84) is zero, then there will exist a function F(a, b) such that
F(a, b) = p(a) + q(b) , (10.85)

which produces
F(r, t) = p(r − vt) + q(r + vt) . (10.86)
DOI:10.1190/1.9781560803737

Although equation (10.86) describes a wave that does not change its shape
as it travels, the combined function u(r, t) = F(r, t) G(r, t) still will produce
an amplitude modulated wave. It can be shown that, if
2 2 1
G(a, b) = = = , (10.87)
a + b r − vt + r + vt r
then equation (10.84) reduces to the desired result

∂2 F
=0. (10.88)
∂a ∂b

Thus, for the three-dimensional spherically symmetric wave equation, we

have a solution of the form

1 1
u(r, t) = p(r − vt) + q(r + vt) . (10.89)
r r
Hence, in three dimensions, the wave does propagate in an undistorted
way except for an amplitude factor of 1/r. Because the energy is pro-
portional to the square of the amplitude, the corresponding energy factor
is 1/r 2 . Notice that the area of successive spheres that the wave equation
(10.89) passes through increases as r 2 . Therefore, in order for energy to
be conserved, the energy of the wave in the radial direction must decrease
as 1/r 2 , which it does. Unfortunately, the same line of reasoning does not
produce the same kind of result in the case of the two-dimensional circularly
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 10: Waves Traveling in Opposite Directions 329

symmetric wave equation. In two dimensions, there is no breakthrough sim-

plification in the case of symmetry.
In conclusion, there are basic differences among one-dimensional, two-
dimensional, and three-dimensional wave propagations. When there are no
frictional forces present, energy must be conserved. As a result, the wave
energy that is transported on a one-dimensional string is the same at all
points of the string. The result is an undistorted propagation in one dimen-
sion. In two and three dimensions, the energy of the wave is spread out over
a larger arc or surface as the wave propagates. In order for that energy to be
conserved, there is a diminution of the amplitude of the wave as it travels
outward.
The definition in the SEG Wiki states: Spherical divergence is the
decrease in wave strength (energy per unit area of wavefront) with distance
as a result of geometric spreading (https://wiki.seg.org/wiki/Dictionary:
Spherical_divergence). A spherical wave traveling through the body of a
DOI:10.1190/1.9781560803737

medium continually spreads out so that the energy density decreases. For
a point source, the energy density decreases inversely as the square of the
distance the wave has traveled; this means that the amplitude deceases lin-
early with the distance traveled. For energy which travels along a surface,
the analogous term is cylindrical divergence, where the energy varies inver-
sely as the distance and the amplitude as the square root of the distance.
Other mechanisms by which a seismic wave loses energy involve absorption
and loss at interfaces by reflection (including diffraction, mode conversion,
and scattering).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms
DOI:10.1190/1.9781560803737
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 11

Green’s Function for the

Wave Equation

“Nature is an infinite sphere of which the center is

everywhere and the circumference nowhere.”
DOI:10.1190/1.9781560803737

—Pascal
Wavefront cones
First devised by Huygens (1678), the concept of the light cone (or wave-
front cone) is depicted in Figure 11.1. Figure 11.1a illustrates how a wave-
front of light spread out as a spherical wave (the circles represent spheres).
In Huygens’ diagram, three wavefronts are depicted as contour lines

a) b)

Figure 11.1. (a) Huygens’ depiction of the light cone and (b) light cone including
the time axis.

331
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

332 Basic Wave Analysis, Part 3: Waveform Analysis

(i.e., level lines) BG, dd, and CE, which occur

at succeeding instants of time. If we include
the time axis, we obtain the cone shown in
Figure 11.1b.
Rather than referencing time and space
separately, the typical convention is to reference
an entity called spacetime. Imagine a vehicle
moving at a constant velocity in a straight line.
Figure 11.2 describes its movement through
spacetime. In a spacetime graph, space is drawn
Figure 11.2. Movement horizontally and time vertically. The three dimen-
in spacetime. sions of space are length, width, and height. There
is only one dimension of time, which runs from
past to future. Because it is difficult to represent four dimensions, instead
we draw two dimensions of space on a horizontal plane, and draw time
DOI:10.1190/1.9781560803737

vertically.
Imagine a flash (i.e., impulse) of light. The light spreads out in every
direction, first hitting objects nearby, and then later distant objects. In two
dimensions, as shown in Figure 11.3, this movement resembles ripples in
a pond.
Now, we can draw time vertically, and stack the images shown in
Figure 11.3 above each other, so that the result is the cone shown
in Figure 11.4. This is called the light cone. A light cone is a flash of light
moving through spacetime. The entire light cone consists of two single
cones. The bottom cone shows light collapsing into a single event, and
the top one shows light exploding out again. It explains how the past can

a) b) c)

Figure 11.3. Light represented similar to pond ripples from (a) initial flash of
light to (b) wavefront at a given moment of time to (c) wavefront at a later moment
of time.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 11: Green’s Function for the Wave Equation 333

Figure 11.4. The

past light cone and
the future light cone.
DOI:10.1190/1.9781560803737

influence the present and the future. When light cones overlap, it means two
objects or events interact with each other. Every event in the universe has an
associated light cone. We deal with seismic waves instead of light waves. As
a result, we speak of wavefront cones. They are the mechanical-wave
counterparts of light cones.

Three spatial dimensions

Huygens’ principle says that all points of a wavefront may be regarded
as new sources of wavelets that expand in every direction at a rate depending
on their velocities. In a homogeneous isotropic medium, the velocity v is
the same constant at every point in all directions. In such a case, the
wavefront is spherical. For the sake of definiteness, we assume that v is a
positive number, not a negative number. As a result, we do not have to
keep account of whether v is positive or negative. The three spatial dimen-
sions are x,y,z. The relevant notation is shown in the Table 11.1.
A sphere is a perfectly round geometrical surface in 3D space. It is
defined mathematically as the set of points that are all at the same distance
r from a given point. The volume enclosed by a sphere is called a ball.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

334 Basic Wave Analysis, Part 3: Waveform Analysis

Table 11.1. Notation for three spatial dimensions.

Item Notation

Space-vector (also known r = (x, y, z)

as position-vector) 
Length (or magnitude) r = |r| = + x2 + y2 + z2
of vector
Laplacian operator ∂2 ∂2 ∂2
2
+ 2+ 2
∂x ∂y ∂z
 
Wavenumber vector k = kx , k y , kz

Wavenumber k = |k| = + kx2 + ky2 + kz2
Dot product k · r kx x + k y y + kz y
Source function f (r, t) f (x, y, z, t)
DOI:10.1190/1.9781560803737

Solution u(r, t) u(x, y, z, t)

Space-time impulse d d( x)d(y)d(z)d(t)
Huygens’ wavelet g(r, t) g(x, y, z, t)
Volume integral: dV dx dy dz

Accordingly, a sphere is the surface of the ball. The distance r is the radius
of the ball as well as the radius of the sphere. To ensure that we are
speaking of a sphere and not the associated ball, we sometimes use the
term “spherical shell” instead of the term “sphere.” In a homogeneous iso-
tropic 3D medium, the Huygens’ wavelet is a spherical shell; that is, the
wavelet is non-zero only on a sphere of radius r ¼ vt. Let the function
g(x, y, z, t) represent the wavelet. The spherical wavelet is not simply a func-
tion of the radius vector r. More exactly, the wavelet is a function of its
length r; that is,

g(x, y, z, t) ; g(r, t) = g(r, t) . (11.1)

At time t, the radius is r ¼ vt. The sphere can be represented by d(vt 2 r).
Because of spherical spreading, we include the spreading factor given by
1/r. We also include a constant scaling factor given by 1/4pv. As a
result, we can represent the Huygens’ wavelet by

 
1 1
g(r, t) = d(vt − r) (for t . 0) . (11.2)
4pv r
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 11: Green’s Function for the Wave Equation 335

Here, d is the Dirac delta function. Equation (9.16) in the Dirac delta
function and Heaviside step function section of Chapter 9 provides
d(t)
d(vt) = . (11.3)
v
From this, it follows that
r
d t−
d(vt − r) = v . (11.4)
v
Thus, the Huygens’ spherical wavelet also may be written as
r
1 d t−
g(r, t) = d(vt − r) = v . (11.5)
4pvr 4pv2 r
DOI:10.1190/1.9781560803737

Now, we can transform to spherical coordinates

kx = k sin b cos a, ky = k sin b sin a, kz = k cos b

(11.6)
x = r sin u cos f, y = r sin u sin f, z = r cos f .

Without loss of generality, we can choose the north pole b ¼ 0 in the

direction of the vector k ¼ (kx, ky, kz). Thus, sin b ¼ sin 0 ¼ 0, cos
b ¼ cos 0 ¼ 1. For this choice of north pole, we have

kx = 0, ky = 0, kz = k . (11.7)

Thus, the dot product of the vectors k ¼ (0, 0, k) and r ¼ (x, y, z) is

k · r = (0, 0, k) · (x, y, z) = kz = kr cos u . (11.8)

Now, let us now take the spatial Fourier transform of equation (11.1).
We have
1
d(vt − r) i k·r
G(kx , ky , kz , t) = e dx dy dz . (11.9)
−1 4pvr

In spherical polar coordinates, this Fourier transform is

p 2p 1
d(vt − r) i k r cos u
G(k, t) = sin u d u df r 2 dr e . (11.10)
0 0 0 4pvr
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

336 Basic Wave Analysis, Part 3: Waveform Analysis

The integral over f gives 2p, and the integral over r removes the delta func-
tion. Thus, we obtain
p 
1
G(k, t) = sin u d u [2p] (vt)2 e i kvt cos u
0 4pv(vt)
t p i kvt cos u
= e sin u d u . (11.11)
2 0
In this equation, we see that G(k, t) depends only on k only via its
magnitude k. Thus, we write G(k, t) simply as G(k, t). To evaluate the
above integral, we let
u = ivkt cos u , (11.12)
so that
du = −ivkt sin u d u . (11.13)
DOI:10.1190/1.9781560803737

The result is
−ivkt
1 1
G(k, t)O = G(k, t) = − eu du = − eu |−ivkt
+ivkt
2i vk +ivkt 2i vk
1
=− (e−ivkt − e+ivkt ) (11.14)
2i vk
1
=− [i sin(−vkt) − i sin(vkt)] .
2i vk
Thus,
sin vkt sin vkt
G(k, t) = =t . (11.15)
vk vkt
Equation (11.14) provides an important result. To explore it, first we should
introduce the cardinal sine function.
In mathematics, physics, and engineering, the cardinal sine function or
sinc function, denoted by sinc(x), has two slightly different definitions.
In mathematics, the historical unnormalized sinc function is defined by
sin x
sinc x = . (11.16)
x
Equation (11.16) is the definition that we will use. In digital signal pro-
cessing and information theory, the normalized sinc function is commonly
defined by
sin px
sinc x = . (11.17)
px
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 11: Green’s Function for the Wave Equation 337

In either case, the value at x ¼ 0 is defined to be the limiting value

sinc(0) = 1. The definite integral of the unnormalized sinc function over
all x is equal to p. The definite integral of the normalized sinc function
over all x is equal to 1. Additionally, all of the zeros of the normalized
sinc function are integer values of x. The normalized sinc function is the
Fourier transform of the rectangular function with no scaling, which is
used when reconstructing a continuous bandlimited signal from the
signal’s uniformly spaced samples. The two definitions differ only in the
scaling of the independent variable (the x-axis) by a factor of p.
Equation (11.15) may be written as
sin vkt
G(k, t) = t = t sinc(vkt) . (11.18)
vkt
In conclusion, the spatial Fourier transform of the three-dimensional
wavelet represented in equation (11.2) is t times a sinc function, as given
DOI:10.1190/1.9781560803737

by equation (11.18). Figures 11.5 and 11.6 show a graphical representation.

What have we discovered? A function g(x, y, z, t) of the three space
coordinates x, y, z would ordinarily have a spatial Fourier transform
G(kx , ky , ky , t) that depends upon the three wavenumber coordinates
kx , ky , ky . However, the spherical wavelet g(r, t) depends only upon the
radius r. Also, the spatial Fourier transform G(k, t) depends only upon the
wavenumber k. As will be discussed next, the same result holds in both
the 1D and 2D cases.

Figure 11.5. The circles

represent spheres; the
solid curve represents
g(r, t) for v = 1, t = 1;
and the dashed curve
represents g(r, t) for
v = 1, t = 2.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

338 Basic Wave Analysis, Part 3: Waveform Analysis

Figure 11.6. The

solid curve
represents G(k, t)
for v = 1, t = 1;
the dashed curve
represents G(k, t)
for v = 1, t = 2.
DOI:10.1190/1.9781560803737

One spatial dimension

Recall that the Heaviside step function, or the unit step function, usually
denoted by H (but also by u), is

0, t , 0
u(t) = . (11.19)
1, t . 0
It is a discontinuous function for which the value is 0 for the negative argu-
ment and 1 for the positive argument. It is an example of the general class of
step functions, all of which can be represented as translations of linear com-
binations of this version. Now, we want to think of the similar Heaviside step
function,

0, t , c
u(t − c) = , (11.20)
1, t . c
as a switch that is off until t = c. At this point, the switch turns on and takes a
value of 1. Heaviside step functions u(t − c) can only take values of 0 or 1,
but we can use them to derive other kinds of switches. For instance,
4u(t − c) is a switch that is off until t = c and then turns on and takes
a value of 4. Now, suppose that we want a switch that is on (with a value
of 1) and then turns off at t = c. We can use Heaviside functions to represent
this as well. The following function exhibits this kind of behavior:

1 − 0 = 1, t , c
1 − u(t − c) = u(c − t) = . (11.21)
1 − 1 = 0, t . c
The function (11.21) for t , c has value 1. When we hit t = c the func-
tion takes on value 0. We also can modify this so that it has values other than
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 11: Green’s Function for the Wave Equation 339

Table 11.2. Notation for one spatial

dimension.

Item Notation

Space vector r = ( x)
Length of vector r = | x|
Laplacian operator ∂
∇2 =
∂x2
Wavenumber vector k = ( kx )
Wavenumber k = |kx |
Dot product k · r kx
Source function f (r, t) f (x, t)
Solution u(r, t) u(x, t)
Space-time impulse d d( x)d(t)
DOI:10.1190/1.9781560803737

Huygens’ wavelet g(r, t) g(x, t)

Volume integral dV dx

1 when it is on. For instance, 3 − 3u(t) is a switch that has a value of 3 until
it turns off at t = c. In fact, we can use Heaviside functions to represent even
more complicated switches.
Next, consider the wavelet for one spatial dimension x. The relevant
notation is shown in the Table 11.2.
In Figure 11.7, the Huygens’ wavelet (for a given value of t) is rep-
resented by the line AB. In that time, the wavefront travels a distance of

Figure 11.7. Wavefront cone for one spatial dimension.

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

340 Basic Wave Analysis, Part 3: Waveform Analysis

vt to the right and left. Outside of the cone, the wavelet is zero. The
Huygens’ wavelet for t . 0 has the representation
⎧
⎨ 1 , −vt , x , vt
g(x, t) = 2v . (11.22)
⎩
0, x . vt
Figure 11.8 displays three switch representations. The curve shown in
Figure 11.8a is the Heaviside step function

1, vt + x . 0 or x . −vt
u(vt + x) = . (11.23)
0, vt + x , 0 or x , −vt

The curve in Figure 11.8b is the Heaviside step function


1, vt − x . 0 or x , vt
u(vt − x) =
DOI:10.1190/1.9781560803737

. (11.24)
0, vt − x , 0 or x . vt

The curve in Figure 11.8c represents the product of both Heaviside functions
(11.23) and (11.24); that is,

1, −vt , x , vt
u(vt + x)u(vt − x) = (11.25)
0, x , −vt or x . vt .

a)

b)

c)

Figure 11.8. (a) Switch that turns on at x = −vt; (b) switch that turns off at x = vt;
and (c) switch that turns on at x = −vt and then turns off at x = vt.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 11: Green’s Function for the Wave Equation 341

Figure 11.9. The

one-dimensional
Huygens’ wavelet.

(Note, because of its shape, this curve is referred to as boxcar-function.)

Therefore, the Huygens’ wavelet represented in equation (11.22) can be
written alternatively as
1
g(x, t) = u(vt + x)u(vt − x) . (11.26)
2v
Figure 11.9 depicts a graph of the Huygens’ wavelet represented with
equation (11.26). In terms of the radius r, it is
DOI:10.1190/1.9781560803737

1
g(r, t) = u(vt − r) (with r = |x| . 0) . (11.27)
2v
Now, we can find the Fourier transform with respect to x of the wavelet.
Making use of wave equation (11.22), we have
+vt
1
1 +vt ikx 1 eikx  sin vkt
G(k, t) = g(x, t)e dx =
ikx
e dx =  = .
−1 2v −vt 2v ik −vt vk
(11.28)
The resulting equation (11.28) states that the Fourier transform of wavelet
equation (11.22) is t times the sinc function of vkt; that is,
 
sin vkt
G(k, t) = t = t sinc(vkt) . (11.29)
vkt

Two spatial dimensions

Next, consider two spatial dimensions x and y. The relevant notation is
shown in the Table 11.3.
In the case of two spatial dimensions x and y, the Huygens’ wavelet
(defined for t . 0) is
⎧
⎪ 
⎨ 1 
1
, x2 + y2 , vt
g(x, y, t) = 2p v v t − x2 − y2
2 2 . (11.30)
⎪
⎩ 
0, x2 + y2 . vt
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

342 Basic Wave Analysis, Part 3: Waveform Analysis

Table 11.3. Notation for two spatial

dimensions.

Item Notation

Space vector r = (x, y)


Length of vector r = x2 + y 2
Laplacian operator ∂2 ∂2
+
∂x2 ∂y2
Wavenumber vector k = (kx , ky )

Wavenumber k = kx2 + ky2
Dot product k · r kx x + ky y
Source function f (r, t) f (x, y, t)
Solution u(r, t) u(x, y, t)
DOI:10.1190/1.9781560803737

Space-time impulse d d( x)d(y)d(t)

Huygens’ wavelet g(r, t) g(x, y, t)
Volume integral dV dx dy

Using Figure 11.10 for reference, the spatial Fourier transform of equation
(11.30) is
1 1
G(kx , ky , t) = g(x, y, t)ei(kx x+ky y) dx dy . (11.31)
−1 −1

Now, we can transform to the polar coordinates

kx = k cos a , ky = k sin a
(11.32)
x = r cos f , ky = r sin f .
Without loss of generality, we can choose the polar axis a = 0 in the direc-
tion of the vector k = (kx , ky ). Thus,

cos a = cos 0 = 1 and sin a = sin 0 = 0 . (11.33)

Thus, for this choice of polar axis, we have kx = k and ky = 0. Then, the dot
product of the vector k = (kx , ky ) = (k, 0) and r = (x, y) is

k · r = kx x + ky y = kr cos f . (11.34)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 11: Green’s Function for the Wave Equation 343

DOI:10.1190/1.9781560803737

Figure 11.10. The Huygens’ wavelet equation (11.31) with v = 1 plotted against
the radius r. The solid curve represents the case of t = 1, the dashed curve represents
the case of t = 2, and the dotted curve represents the case of t = 3.

Hence, equation (11.31) becomes

1 2p
G(k, f, t) = r dr g(r, f, t) eikr cos f d f . (11.35)
0 0

Noting that r2 = x2 + y2 , we can write equation (11.31) as

⎧
⎨ 1 √
1
, r , vt
g(r, f, t) = 2pv v2 t2 − r 2 . (11.36)
⎩
0, r . vt

Circular symmetry (as used in mathematics) is a type of continuous

symmetry for a planar object that can be rotated by any arbitrary angle
and map onto itself. Radial symmetry (as used in biology) is a type of sym-
metry in which identical body parts are distributed in a circular arrangement
around a central axis. A starfish is radially symmetric but not circularly
symmetric. To summarize, we can say that radial symmetry (biology) or
rotational symmetry (geometry) relates to rotational invariance about a
fixed finite angle, but circular symmetry relates to rotational invariance
about any angle.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

344 Basic Wave Analysis, Part 3: Waveform Analysis

The cone function g(r, f, t) given in equation (11.36) does not depend
upon f. In other words, we have verified the cone function is circularly sym-
metric as required. As we will demonstrate next, G(k, f, t) also is circularly
symmetric. Thus, we shall write these functions as g(r, t) and G(k, t).
Equation (11.35) becomes
1 2p
G(k, t) = r dr g(r, t) eikr cos f d f . (11.37)
0 0

The Bessel function of the first kind and order zero is

2p
1
J0 (kr) = eikr cos f d f . (11.38)
2p 0

Substituting equation (11.38) into equation (11.37), we obtain

1
DOI:10.1190/1.9781560803737

G(k, t) = 2p g(r, t) J0 (kr) r dr . (11.39)

0

The Hankel transform fH (k) of a circularly symmetric function f (r) is

defined to be
1
fH (k) = f (r) J0 (kr) r dr . (11.40)
0

Thus, G(k, t) can be written as

G(k, t) = 2p gH (k, t) , (11.41)

where gH (k, t) is the Hankel transform of g(r, t). Thus, the Fourier transform
of the circularly symmetric function g(r, t) is itself circularly symmetric and
can be performed by the single integral (11.39). Now, let us evaluate G(k, t).
Substituting equation (11.36) into equation (11.39) produces
vt
1 1
G(k, t) = 2p √ J0 (kr) r dr . (11.42)
0 2pv v t2 − r 2
2

By using formula 6.554 (2) in Section 6.55 “Combinations of Bessel

functions and algebraic functions” in Gradshteyn and Ryzhik (2007), we
obtain

1
dx sin y
x J0 (xy) √ = , y.0. (11.43)
0 1−x 2 y
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 11: Green’s Function for the Wave Equation 345

Let x = r/vt, y = vtk. Then, equation (11.43) becomes

vt  
r 1 1 sin vkt
J0 (kr)  dr = = sinc(vkt) . (11.44)
0 vt r 2 vt vkt
1−
vt

Multiplying this equation by t and simplifying, we have

vt
1 1 sin vkt
√ J0 (kr) r dr = = t sinc(vkt) . (11.45)
v 0 v t −r
2 2 2 vk

Equating equation (11.42) with equation (11.45), we obtain

sin vkt
G(k, t) = = t sinc(vkt) . (11.46)
vk
DOI:10.1190/1.9781560803737

This is the result we wanted in two dimensions. It is the same result as we

obtained in both three dimensions and one dimension.
Importantly, in this process, we employed Bessel functions. Friedrich
Wilhelm Bessel was born in 1784 in Minden, Westphalia and died in
1846 in Königsberg, Prussia. At the age of 14, he left school to become
an apprentice to a commercial import-export firm, which inspired his
research in navigation and the use of astronomic observations to determine
longitude. After writing his 1804 paper which calculated the orbit of
Halley’s comet, Bessel concentrated his work on astronomy and mathemat-
ics. In 1806, Bessel accepted the post of assistant at the Lilienthal Observa-
tory, from which he was able to begin his examination of James Bradley’s
18th century stellar observations at Greenwich and eventually produce
precise positions for 3222 stars.
In 1809, at the age of 26, Bessel was appointed director of Königsberg
Observatory and professor of astronomy. Because a doctorate degree was
required for this position, Carl Friedrich Gauss, who had met Bessel in
Bremen in 1807, encouraged the University of Göttingen to award the
degree to Bessel. Despite his lack of a university education, Bessel was a
major figure in astronomy during his lifetime, and he had a very significant
impact on university teaching.
Bessel’s concentration at Königsberg Observatory was to determine
the positions and proper motions of 50,000 stars. During this project,
he calculated the distance to 61 Cygni, a dim star near the Sun, by employing
the principles of parallax. From noting the deviations in the motions of
Sirius and Procyon, he deduced the cause to be the gravitational attraction
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

346 Basic Wave Analysis, Part 3: Waveform Analysis

of unseen companions. Indeed, this “dark companion” of Sirius was the

first correct claim of a previously unobserved companion by positional
measurement; the discovery of Sirius B was confirmed 18 years later
in 1862.
Importantly, Bessel realized that, before a positional observation could
be relied upon for accurate results, quantitative knowledge of every possible
error that might enter into the finished result should be determined. By
eliminating all sources of error (optical, mechanical, and meteorological),
Bessel produced strikingly delicate results which could be used to derive
even more information (Fauvel et al., 1993).

Green’s function for the wave equation

Our focus is the nonhomogeneous wave equation given by
DOI:10.1190/1.9781560803737

∂2 u(r, t)
− v2 ∇2 u(r, t) = f (r, t) with v = positive constant . (11.47)
∂t2
The wave equation is called nonhomogeneous because of the non-zero
source term f (r, t) on the right side of the equation. Because, we are
working in a homogeneous medium, velocity v remains constant. The 1D,
2D, and 3D solutions all can be derived from the three-dimensional wave
equation, if we define initial conditions at a plane (for 1D), line (for 2D),
and point (for 3D). The line solution for two dimensions is Hadamard’s
solution.
Green’s function g is defined as the solution of the wave equation with
an impulsive source function d, that is, with the space-time impulse d on the
right side of the equation:

∂2 g(r, t)
− v2 ∇2 g(r, t) = d(r, t) with v = positive constant . (11.48)
∂t2
In order to solve this equation for g, we can make use of the spatial
Fourier transform, designated in this section by F , which is defined by
1
F g(r, t) ; G(k, t) = g(r, t) ei k·r dv . (11.49)
−1

The spatial Fourier transform of ∂2 g/∂t2 is

∂2 g ∂2 G
F = 2 . (11.50)
∂t2 ∂t
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 11: Green’s Function for the Wave Equation 347

The spatial Fourier transform of ∇2 g is

F (∇2 g) = −k2 g . (11.51)

The spatial Fourier transformation of d(r, t) leaves only the time delta func-
tion; that is,

F (∇2 g) = F[d(x)d(y)d(z)d(t)] = d(t) . (11.52)

Thus, the spatial Fourier transform of equation (11.50) is

∂2 G
+ v2 (k2 G) = d(t) . (11.53)
∂t2
Because the spatial variables do not appear in equation (11.53), we may
replace the partial derivative by a total derivative, so equation (11.53) is the
DOI:10.1190/1.9781560803737

ordinary differential equation

d2 G
+ v2 (k2 G) = d(t) . (11.54)
dt2
Because the right side of equation (11.54) is a delta function, we recog-
nize G as the Green’s function for the ordinary differential equation (11.54).
We require that Green’s function G be causal (i.e., one-sided in time), so
G = 0 for t , 0. For this reason, the Laplace transform method is ideally
suited for the solution of equation (11.54).
Let GL (k, s) be the Laplace transform of G(k, t); that is,
1
GL (k, s) = G(k, t)e−st dt . (11.55)
0

Then, the Laplace transform of d 2 G/dt2 is

1
d 2 G −st
e dt = s2 GL , (11.56)
0 dt2

where, as customary, we assume zero initial conditions for Green’s function

G and its derivative. Also, we know that the Laplace transform of d(t) is 1.
Thus, the Laplace transform of the ordinary differential equation is the
algebraic equation

s2 GL + v2 k2 GL = 1 . (11.57)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

348 Basic Wave Analysis, Part 3: Waveform Analysis

Solving this algebraic equation, we obtain the solution

1
GL (k, s) = . (11.58)
s2 + v2 k 2
From tables of the Laplace transform, we immediately find that

sin vkt
G(k, t) = . (11.59)
vk
Thus, Green’s function G for the ordinary differential equation (11.54) is the
sinc function.
Let us summarize. In the previous three sections, we define three
Huygens’ wavelets: the one-dimensional wavelet, the two-dimensional
wavelet, and the three-dimensional wavelet. We find that the spatial
Fourier transform of each wavelet is the same, i.e., sin vkt/vk. In this
DOI:10.1190/1.9781560803737

section, we define three Green’s functions: the one-dimensional Green’s

function, the two-dimensional Green’s function, and the three-dimensional
Green’s function. We find that the spatial Fourier transform of each function
is the same, i.e., sin vkt/vk. Therefore, we conclude that the Huygens’ wave-
lets are the same as the Green’s functions. The reason that the results of
Huygens and Green are the same is that both are working with impulse
response functions. In other words, Huygens‘ wavelets, Green’s function,
and impulse response functions are different names for the same thing in
the case of traveling waves. Tables 11.4 and 11.5 summarize the results
of the previous three sections.
The inverse spatial Fourier transform of G(k, t) provides Green’s
function g for the wave equation. Thus, the required Green’s function g is

Table 11.4. Coordinates.

Spatial coordinates Wavenumber coordinates

1D r = ( x) k = (k )
r = | x| k = |k |
2D r = (x, y) k = (kx , ky )
 
r = x 2 + y2 k = kx2 + ky2
3D r = (x, y, z) k = (kx , ky , kz )
 
r = x 2 + y2 + z 2 k = kx2 + ky2 + kz2
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 11: Green’s Function for the Wave Equation 349

Table 11.5. Huygens’ wavelet and its spatial Fourier transform.

Wavelet g(r, t) for Fourier transform

r ≥ 0 and t ≥ 0 G(k, t)
u(vt − r ) sin vkt
1D
2v vk
u(vt − r ) sin vkt
2D √
2 p v v2 t 2 − r 2 vk
d(vt − r ) sin vkt
3D
4pvr vk
DOI:10.1190/1.9781560803737

Figure 11.11. Huygens’ wavelet for r ≥ 0 with v = 1, t = 1, so vt = 1.

the same as the Huygens’ wavelet g given in Table 11.5 (for one, two, and
three dimensions). Diagrams of the Green’s functions for one, two, and three
dimensions at a fixed instant of time are shown schematically in
Figure 11.11. It is apparent that the one-dimensional and two-dimensional
Green’s functions are non-zero throughout the light cone (i.e., for r ≤ vt),
whereas the three-dimensional Green’s function is non-zero only on the
boundary of the light cone (i.e., for r = vt). As shown in Figure 11.12,
the one-dimensional Green’s function is a rectangular function. As shown
in Figure 11.13, the two-dimensional Green’s function is a bowl-shaped
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

350 Basic Wave Analysis, Part 3: Waveform Analysis

Figure 11.12. The

one-dimensional
Huygens’ wavelet
with v = 1, t = 1.

Figure 11.13. The two-

dimensional Huygens’ wavelet
with v = 1, t = 1, so vt = 1.
DOI:10.1190/1.9781560803737

function. The three-dimensional Green’s function is an infinite spike

(i.e., Dirac delta function) on the boundary of the light cone. All three
Green’s functions are, of course, zero outside of the light cone (i.e.,
for r . vt).
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 12

Initial and Boundary Conditions

“Newton has shown us that a law is only a necessary relation

between the present state of the world and its immediately
DOI:10.1190/1.9781560803737

subsequent state. All the other laws since discovered are

nothing else; they are in sum, differential equations.”

—Henri Poincaré

The convolution integral

Let g(r, t) be the Green’s function (i.e., impulse response function) for
the wave equation. Let the source function f (r, t) be causal. Of course, the
Green’s function is also causal, so both g and f vanish for t , 0. A black
box is any complex piece of equipment (for example, a unit in an electronic
system) with contents that are mysterious to the user. In our case, the black
box is the subsurface of the earth. On the surface, we can measure the input
signal that we send into the subsurface, and we can measure the resulting
output signal received at the surface. Here, we also are considering an
input-output problem, but is a much simpler case. It is the case of one-
way transmission through a homogenous isotropic medium. The black
box is represented by its impulse response (also known as the Green’s func-
tion). The output w(r, t) is the convolution of the impulse response g(r, t)
and the input f (r, t). If we let the asterisk represent the convolution oper-
ation, then we can write

w(r, t) = g(r, t) ∗ f (r, t) . (12.1)

351
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

352 Basic Wave Analysis, Part 3: Waveform Analysis

Because both g and f vanish for t , 0, the output w also vanishes for
t , 0. The output w is also known as the retarded (or delayed) potential.
Now, we can verify that the convolution w is a solution of the non-
homogeneous wave equation

∂2 w
− v2 ∇ 2 w = f . (12.2)
∂t2
Into the left side of the wave equation (12.2), we substitute
t 1
w=g∗f = dt0 dx0 f (x0 , t0 ) g(x − x0 , t − t0 ) . (12.3)
0 −1

We note that
t 1
∂2 (g ∗ f ) ∂2
DOI:10.1190/1.9781560803737

= 2 dt0 dx0 f (x0 , t0 ) g(x − x0 , t − t0 ) , (12.4)

∂t2 ∂t 0 −1

which is
t 1
∂2 (g ∗ f ) ∂2
= dt0 dx0 f (x0 , t0 ) g(x − x0 , t − t0 )
∂t2 0 −1 ∂t2
 2 
∂g
= ∗f . (12.5)
∂t2

Similarly,
t 1
∇ (g ∗ f ) =
2
dt0 dx0 f (x0 , t0 ) ∇2 g(x − x0 , t − t0 )
0 −1
= (∇2 g) ∗ f . (12.6)

Thus, we obtain
 2 
∂2 ∂g
(g ∗ f ) − v ∇ (g ∗ f ) =
2 2
−v ∇ g ∗f .
2 2
(12.7)
∂t2 ∂t2

The Green’s function g in question is defined by

∂2 g
− v2 ∇ 2 g = d . (12.8)
∂t2
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 12: Initial and Boundary Conditions 353

Hence,
∂2 w
− v2 ∇ 2 w = d ∗ f = f , (12.9)
∂t2
which shows that w indeed satisfies the non-homogeneous wave equation
with source f. The constant velocity v can be either positive or negative. It
is our custom to assume that velocity v is positive.
What is the physical meaning of convolution? The physical sense of the
convolution (12.1) may be described as follows. Here, we use one dimension
x, but the same argument apples in the case of more dimensions. The func-
tion f (x, t) represents the input. It follows that the impulsive point input at
(x0 , t0 ) can be represented as
f (x0 , t0 )d(x − x0 , t − t0 ) . (12.10)

The entire input f (x, t) can be obtained by summing (more precisely,

DOI:10.1190/1.9781560803737

integrating) all of these impulsive point inputs; that is, the input has the rep-
resentation
t 1
f (x, t) = dt0 dx0 f (x0 , t0 )d(x − x0 , t − t0 ) = d ∗ f . (12.11)
0 −1

Because Green’s function g is the impulse response, each impulsive

point input (12.10) yields the response

f (x0 , t0 )g(x − x0 , t − t0 ) . (12.12)

In other words, the d in expression (12.10) is replaced by g in expression

(12.12). Accordingly, we replace the d in expression (12.11) in order to
obtain the output w(x, t) given by
t 1
w(x, t) = dt0 dx0 f (x0 , t0 ) g(x − x0 , t − t0 ) . (12.13)
0 −1

This equation is called the convolution integral. It is the superposition of the

point responses.
The 1D case. First, we address the case of one spatial dimension. Into
equation (12.13), we substitute the one-dimensional Green’s function
⎧
⎨1
, −vt ≤ x ≤ vt
g(x, t) = 2v . (12.14)
⎩ 0, otherwise
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

354 Basic Wave Analysis, Part 3: Waveform Analysis

The result is the 1D convolution

 x+v( t−t0 )
1 t
w(x, t) = dt0 dx0 f (x0 , t0 ) . (12.15)
2v 0 x−v( t−t0 )

The 2D case. Similarly, using the two-dimensional Green’s function,

the result is the 2D convolution
 
1 t f (x0 , y0 , t0 )
w(x, y, t) = dt0 dx0 dy0  ,
2pv 0 v2 (t − t0 ) − (x − x0 )2 − (y − y0 )2
2

(12.16)
where the double integral is over the area of a circle with center (x, y) and
radius v(t − t0 ).
The 3D case. Because we want to integrate over time, we use the
three-dimensional Green’s function g in the form
DOI:10.1190/1.9781560803737

r
d t−
g(r, t) = v . (12.17)
4pv2 r
Thus, in the 3D case, the convolution integral is
 
r − r0 
t  d t − t0 −
v
w(x, y, z, t) = dt0 dx0 dy0 dz0 f (x0 , y0 , z0 , t0 ) ,
0 4pv r − r0 
2

(12.18)
where

r − r0  = (x − x0 )2 + (y − y0 )2 + (z − z0 )2 (12.19)

and where the triple integral is over the volume of the sphere in which the
source function is defined. Because of the sifting property of the delta func-
tion, equation (12.14) becomes the so-called Lorentz equation
 
r − r0 
 f x0 , y0 , z0 , t −
1 v
w(x, y, z, t) = dx0 dy0 dz0 ,
4pv2 r − r0 
(12.20)

where the triple integral is over the volume of the sphere with center
r = (x, y, z) and radius vt. The Lorentz equation expresses the output at
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 12: Initial and Boundary Conditions 355

(x, y, z, t) as a sum of contributions from all points in the given sphere. For
example, the contribution from

r0 = (x0 , y0 , z0 ) (12.21)

was emitted at a time r − r0 /v earlier, and is attenuated with distance

by the factor

1
. (12.22)
4pv2 r − r0 

Source signature
There are many applications of the convolution integral. As a special
case, we can consider a source initiated at the point r = 0 with source
signature given by the causal wavelet h(t), i.e., h(t) = 0 for t , 0. Thus,
DOI:10.1190/1.9781560803737

the source function is

f (r, t) = d(r) h(t) . (12.23)

In order to solve this problem, we could make use of the Lorentz

equation given as equation (12.20), however, we shall solve this problem
from another point of view. The homogeneous 3D wave equation is

∂2 u
− v2 ∇ 2 u = 0 . (12.24)
∂t2
Let us assume that u has polar symmetry. In spherical coordinates with
polar symmetry, equation (12.24) becomes
 
∂2 u 2 1 ∂ ∂u
− v r =0. (12.25)
∂t2 r2 ∂r ∂r

Let us use the following method to find a solution for equation (12.25).
If we introduce the substitution

1
u(r, t) = c(r, t) , (12.26)
4pv2 r
we find that c(r, t) is governed by the 1D homogeneous wave equation

∂2 c 2
2 ∂ c
− v =0. (12.27)
∂t2 ∂r 2
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

356 Basic Wave Analysis, Part 3: Waveform Analysis

However, we know that d’Alembert’s general solution of the 1D

equation includes the form of two waves going in opposite directions; that is,
r r
c(r, t) = h t − + f t + , (12.28)
v v
where h and f are two arbitrary functions. Thus, equation (12.26) becomes
1 r 1 r
u(r, t) = 2
h t− + 2
f t+ , (12.29)
4pv r v 4pv r v
which represents a solution of the three-dimensional wave equation (12.25).
The first term on the right represents a wave outgoing from r = 0, whereas
the second term represents a wave incoming to r = 0. We only want to
consider outgoing waves, so we consider a solution of the form
1 r
u(r, t) = 2
h t− , (12.30)
4pv r v
DOI:10.1190/1.9781560803737

where h(t) is the source signature. From its derivation, we know that the sol-
ution equation (12.30) satisfies the homogeneous wave equation (12.24)
provided r . 0. However, it is important to know what happens at r = 0.
Substituting equation (12.30) into the left side of the homogeneous wave
equation (12.24), we have
 
∂2 1 r 1 r
h t− −v ∇
2 2
h t− =0. (12.31)
∂t2 4pv2 r v 4pv2 r v
Now, near r = 0, the first term is finite, and the second term behaves like
 
h(t) 2 1
− ∇ . (12.32)
4p r
It can be shown that
 
1
∇2 = −4pd(r) . (12.33)
r
Thus, near r = 0, the second term behaves like
h(t)
− [−4pd(r)] = h(t) d(r) . (12.34)
4p
Therefore, the function u(r, t) given by equation (12.30) is the solution of
the non-homogeneous wave equation
d2 u
− v2 ∇2 u = d(r) h(t) . (12.35)
dt2
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 12: Initial and Boundary Conditions 357

Figure 12.1. A
representation of
equation (12.30).

In equation (12.35), the source function is shown to be d(r)h(t). Thus,

DOI:10.1190/1.9781560803737

u(r, t) given by equation (12.30) is the solution to the problem that we

planned to solve. In other words, the solution equation (12.30) is the
output for the source-signature input equation (12.23). Solution equation
(12.30) shows that a point at distance r from the origin is at rest from
time t = 0 to time t = r/v. Then, a point at distance r experiences a response
of the same shape and duration as the source signature h(t), but with the
attenuation given by the factor 1/4pv2 r.
Considering Figure 12.1, if the source is located at the point r0 instead of
the origin, the wave equation (12.35) is modified to

d2 u
− v2 ∇2 u = d(r − r0 ) h(t) (12.36)
dt2
and the solution becomes
 
1 r − r0 
u(r − r0 , t) = f t− , (12.37)
4pv2 r − r0  v

where

(r − r0 ) = (x − x0 , y − y0 , z − z0 ) (12.38)

and

r − r0  = (x − x0 )2 + (y − y0 )2 + (z − z0 )2 . (12.39)
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

358 Basic Wave Analysis, Part 3: Waveform Analysis

If the source signature h(t) is a delta function applied at t = t0 ,

i.e., h(t) = d(t − t0 ), then the corresponding response is seen to be
 
r − r0 
d t − t0 −
v
g(r − r0 , t − t0 ) = . (12.40)
4pv2 r − r0 
The response due to a distribution of sources f (r0 , t0 ) can be obtained
by adding the response due to each elementary portion of the source. The
response to the source f (r0 , t0 ) is f (r0 , t0 )g(r − r0 , t − t0 ). Thus, the total
response is the integral over all (r0 , t0 ); i.e.,
   
1 t f (r0 , t0 ) r − r0 
w(r, t) = dt0 dx0 dy0 dz0 d t − t0 − ,
4pv2 0 r − r0  v
(12.41)
where the triple integral is over the volume where the source function
DOI:10.1190/1.9781560803737

f (r0 , t0 ) is defined and non-zero. The above integral can be simplified to

 
r − r0 
 f r0 , t −
1 v
w(r, t) = dx 0 dy 0 dz 0 , (12.42)
4pv2 r − r0 
where the triple integral is over the volume of a sphere with center at r
and radius vt. More precisely, this triple integral is over the support of
f (the region where f = 0). In physical terms, the effect at r due to a
source at r0 needs a finite time r − r0 /v to reach the position r, and it
suffers an attenuation 1/(4pv2 r − r0 ) with distance r − r0 . Thus, by
this different path we have arrived at the same results as before, as equations
(21.41) and (12.42) are the same as (12.18) and (12.20) in the previous
section.

Surface convolution
The time function d(t) represents a spike at time t = 0. Let c(r) be a
function of the position vector r. Then, c(r)d(t) represents an impulsive
space-time function at time t = 0. The function c(r)d(t) is zero both
before and after time t = 0. Such a function can be realized physically as
follows. We position explosive charges at locations r and let the strength
of the charges be given by c(r). Then, the function c(r)d(t) represents the
experiment of igniting all of the charges at t = 0. This experiment starts
the wave motion in the medium.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 12: Initial and Boundary Conditions 359

Because Green’s function g(r, t) represents the impulse response of the

medium, we can find the resulting wave motion q(r, t) by convolving the
input c(r)d(t) with the Green’s function; that is,

output = impulse response ∗ input

or

q(r, t) = g(r, t) ∗ [c(r)d(t)] . (12.43)

The output function q is called a surface retarded (or delayed) potential.

This expression can be simplified as follows. For simplicity, we can
address the 1D case, so r becomes x. We have

q(x, t) = g(x, t) ∗ [c(x) d(t)] , (12.44)

DOI:10.1190/1.9781560803737

which may be written as

t 1
q(x, t) = dt0 dx0 g(x − x0 , t) c(x0 )d(t0 ) . (12.45)
0 −1

By the sifting property of the delta function, this equation becomes

1
q(x, t) = dx0 g(x − x0 , t) c(x0 ) = g(x, t) ∗ c(x) . (12.46)
−1

Thus, because of the delta function d(t), the convolution over space and
time becomes merely a convolution over space. Although we derived this
result in the 1D case, it holds in the 2D and 3D cases as well. Hence, we
can write

q(r, t) = g(r, t) ∗ c(r) . (12.47)

Into this equation, now we will substitute, respectively, the one-

dimensional, two-dimensional, and three-dimensional Green’s functions.
In the 1D case, we substitute the one-dimensional Green’s function into
equation (12.47) to obtain
x+vt
1
q(x, t) = c(x0 ) dx0 . (12.48)
2v x−vt

Let us interpret this equation. As depicted in Figure 12.2, the line

segment x − vt ≤ x0 ≤ x + vt may be considered to have center x and
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

360 Basic Wave Analysis, Part 3: Waveform Analysis

Figure 12.2. A
graphical aid for
interpreting the
surface retarded
potential.

radius vt. The length of the line segment is 2vt. Thus,

x+vt
1
c = c(x0 ) dx0 (12.49)
2vt x−vt

represents the average value of c over the line segment with center x and
radius vt. That is, c is the mean explosive charge within the given line
DOI:10.1190/1.9781560803737

segment. Equation (12.49) states that the wave motion is given by time t
multiplied by this mean; i.e.,

q(x, t) = tc . (12.50)

In the 2D, case we substitute the two-dimensional Green’s function into

equation (12.48) to obtain

1 c(x0 , y0 )
q(x, y, t) = dx0 dy0  , (12.51)
2pv v t − (x − x0 )2 − (y − y0 )2
2 2

where the integration is over the area of a circle with center (x, y) and
radius vt.
In the 3D case with r = (x, y, z), we have

q(r, t) = dx0 dy0 dz0 g(r − r0 , t) c(r0 ) , (12.52)

in which the triple integral is over all space where c is non-zero. Now, we
substitute the expression for the three-dimensional Green’s function into
equation (12.52) to obtain
 
r − r0 
 d t−
v
q(r, t) = dx0 dy0 dz0 c(r0 ) , (12.53)
4pv r − r0 
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 12: Initial and Boundary Conditions 361

where the integration is over the volume of a sphere with center r = (x, y, z)
and radius vt. However, because the integrand contains the delta function,
this volume integral reduces to a surface integral over the surface of the
sphere. That is, only points on the spherical shell r − r0  = vt contribute,
so this integral becomes

1
q(r, t) = dS c(r0 ) , (12.54)
2pv2 t

where the surface integration is over the surface area S of a spherical shell
with center r = (x, y, z) and radius vt. This result, i.e., that the volume inte-
gral reduces to the surface integral over the spherical shell, has profound
physical implications concerning the nature of the 3D space in which we
live. Christiaan Huygens was the first to perceive this deep result, and it is
the familiar Huygens’ principle. Because the area of the spherical shell is
4pv2 t2 , we see that the mean value of the explosive charge on the spherical
DOI:10.1190/1.9781560803737

shell is

1
c = dS c(r0 ) , (12.55)
2pv2 t

where the surface integration is over the spherical shell S. Thus, equation
(12.55) states that the wave motion is given by
q(r, t) = tc . (12.56)

Next, we will consider another physical experiment, which is the same

as the foregoing example except that the explosion produces an impulsive
doublet d′ (t) instead of the impulse d(t). Considering Figure 12.3, we rep-
resent the explosive distribution by f(r), so the input is f(r) d′ (t). Then,
the resulting wave motion is
output = impulsive response ∗ input

or
p(r, t) = g(r, t) ∗ [f(r)d′ (t)] . (12.57)

This expression can be simplified. Again, we will address the 1D case.

We have
t 1
p(x, t) = dt0 dx0 g(x − x0 , t − t0 ) f(x0 )d′ (t0 ) . (12.58)
0 −1
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

362 Basic Wave Analysis, Part 3: Waveform Analysis

Figure 12.3. The

impulse d(t) and
its first derivative,
called the
doublet d′ (t).
DOI:10.1190/1.9781560803737

The sifting property of the doublet is defined as

1
∂
dt0 g(x − x0 , t − t0 )d′ (t) = g(x − x0 , t) . (12.59)
−1 ∂t

Thus, we have
1
∂
p(x, t) = dx0 g(x − x0 , t)
−1 ∂t
1 (12.60)
∂ ∂
= dx0 g(x − x0 , t) = g(x, t) ∗ f(x) .
∂t −1 ∂t

In the 1D case, we have

x+vt 
∂ 1 1
p(x, t) = f(x0 ) dx0 = [f(x + vt) − f(x − vt)] . (12.61)
∂t 2v x−vt 2

Likewise, in the 2D case we have

  
1 ∂ f(x0 , y0 )
p(x, y, t) = dx0 dy0  , (12.62)
2pv ∂t v2 t2 − (x − x0 )2 − (y − y0 )2
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 12: Initial and Boundary Conditions 363

where the area integral is over the area of a circle with center (x, y)
and radius vt. In the 3D, case we have
 
1 ∂ 1
p(r, t) = f(r 0 ) dS , (12.63)
4pv2 ∂t t

where the surface integral is over the surface area of a spherical shell with
center r and radius vt.

Wave equation with initial conditions

The spatial Fourier transform of the non-homogeneous wave equation

∂2 u
− v2 ∇ 2 u = f (12.64)
DOI:10.1190/1.9781560803737

∂t2
with initial conditions

∂u(r, 0)
u(r, 0) = f(r), = c(r) (12.65)
∂t
is, respectively, the non-homogeneous ordinary differential equation

d2 u
− v2 k2 U = F (12.66)
dt2
with specified initial conditions

∂U(k, 0)
U(k, t) = F(k), = C(k) . (12.67)
∂t
From the theory of ordinary differential equations, the solution of equation
(12.66) with initial conditions (12.67) is
t
∂G(k, t)
U(k, t) = G(k, t − t0 ) F(k, t0 )dt0 + F(k) + C(k)G(k, t) ,
0 ∂t
(12.68)
where G is the Green’s function

sin vkt
G(k, t) = . (12.69)
vk
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

364 Basic Wave Analysis, Part 3: Waveform Analysis

We can write equation (12.68) as

U(k, t) = G(k, t) ∗ [F(k, t) + F(k)d′ (t) + C(k)d(t)] , (12.70)
where the convolution is with respect to only t. Inverse Fourier transform-
ation changes the multiplications in this expression to convolution with
respect to r. Thus, we obtain
u(r, t) = g(r, t) ∗ [ f (r, t) + f(r)d′ (t) + c(r)d(t)] , (12.71)
where now the convolution is with respect to both r and t. Thus, the solution
of the non-homogeneous wave equation (12.64) with initial conditions
(12.65) is in the form of the sum of volume retarded potential w and
surface retarded potentials p and q:

u(r, t) = w(r, t) + p(r, t) + q(r, t) . (12.72)

We have already found expressions for these retarded potentials in the

DOI:10.1190/1.9781560803737

preceding two sections. In the 1D case, we have

t x+v(t−t0 )
1
u(x, t) = f (x0 , t0 )dx0 dt0
2v 0 x−v(t−t0 )
x+vt (12.73)
1 1
= [f(x + vt) − f(x − vt)] + c(x0 )dx0 ,
2 2v x−vt

which is d’Alembert’s solution for the non-homogeneous wave equation.

Wave propagation
Next, we can provide a physical interpretation of the solution of the
homogeneous wave equation
∂2 u
− v2 ∇ 2 u = 0 . (12.74)
∂t2
That is, the source term f is zero, so we only have the two initial con-
ditions f and c. These initial conditions make up the initial perturbation

f(r) d′ (t) + c(r)d(t) (12.75)

which produces, for t . 0, the perturbation

∂
u=p+q= (g ∗ f) + g ∗ c . (12.76)
∂t
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 12: Initial and Boundary Conditions 365

This convolutional formula says that the perturbation u at the point r at

an instant of time t . 0 is a superposition of the elementary perturbations

∂g(r − r0 , t)
f(r0 ) and c(r0 )g(r − r0 , t) (12.77)
∂t
produced, respectively, by the initial perturbations

f(r0 ) d(r − r0 )d′ (t) and c(r0 )d(r − r0 )d(t) (12.78)

located at the point r0 . The point r0 ranges over the set in which the initial
perturbations f and c are located. This result constitutes the principle of
superposition of waves. Concrete realization of this principle depends
upon the structure of the Green’s function and, consequently, on the
number of space dimensions. As a result, we will show that there are funda-
mental differences in the character of wave propagation in space (i.e., 3D),
DOI:10.1190/1.9781560803737

in a plane (i.e., 2D), or on a straight line (i.e., 1D).

Considering wave propagation in 3D space, the three-dimensional
Green’s function

d(vt − r)
g(r, t) = (12.79)
4pvr
represents the perturbation from a momentarily acting point source d(r)d(t).
At an instant of time t . 0, this perturbation occupies a spherical surface
(i.e., spherical shell) of radius vt with a center at the point r = 0, as
shown in Figure 12.4. This means that such a perturbation propagates in
the form of a spherical wave on the shell r = vt, moving outward from

Figure 12.4. For

t . 0, the causal
three-dimensional
Green’s function
resides on a sphere
(i.e., spherical shell)
of radius vt.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

366 Basic Wave Analysis, Part 3: Waveform Analysis

the center with speed v. After the wave has passed a point, there is no longer
any disturbance. In this case, we say that Huygens’ principle takes place
in space.
From the principle of superposition, it follows that the perturbation
u(r, t) for t . 0 from the initial perturbation

f(r) d(t) + c(r)d′ (t) (12.80)

is completely determined by the values f(r) and c(r) over the spherical
surface with center r and radius vt. Now, let the initial perturbation be con-
fined to a region D. Let r be outside the region D, and let d1 and d2 be the
minimum and maximum distances from the point r to the points in the
region D, as shown in Figure 12.5. The perturbation will reach the point r
at an instant of time t1 = d1 /v and the perturbation will continue to reach
r until the instant t2 = d2 /v. Thus, the point r will be acted upon throughout
DOI:10.1190/1.9781560803737

a time interval of length

d2 − d1
t2 − t1 = . (12.81)
v

However, for t . d2 /v, there will be no more disturbance at point r. That

is, at time t1 , the front edge of the wave passes through point r. At time t2 ,
the rear edge of the wave passes through this point.
At an instant of time t, the front edge is the external envelope of the
spherical shells with center r0 (within the region D) and radius vt. Similarly,
the rear edge is the internal envelope of these spherical shells, as shown in
Figure 12.6. At any instant of time t, the perturbation occupies the region of
space enclosed between the front and rear edges of the wave.

Figure 12.5. A graphical

description of the initial
perturbation confined to a region D.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 12: Initial and Boundary Conditions 367

Figure 12.6. Huygens’

principle and the envelope
of spherical shells.
DOI:10.1190/1.9781560803737

Figure 12.7. The causal

two-dimensional Green’s
function occupies the
area of a circle for t . 0.

Next, consider the propagation of waves over a plane (2D propagation).

The Green’s function (defined for t . 0)
⎧
⎨ 1 1
 , r ≤ vt ,
g(r, t) = 2pv v 2 t 2 − x2 − y2 (12.82)
⎩
0, r . vt

represents a perturbation from the impulsive point source d(r)d(t). This per-
turbation at any instant of time t . 0 occupies the area of a circle with center
at the space origin and radius vt.
Figure 12.7 shows that the front edge of the perturbation moves over the
plane with a speed v. However, in comparison from what occurs in 3D prop-
agation, the perturbation at r continues to be present after the front edge has
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

368 Basic Wave Analysis, Part 3: Waveform Analysis

DOI:10.1190/1.9781560803737

Figure 12.8. The point source solution in two dimensions.

passed, so there is no rear edge. Therefore, we say that wave diffusion takes
place.
Let us explain why wave diffusion occurs over a plane. Figure 12.8
portrays 2D propagation from a point source d(x) d(y) d(t). Figure 12.9
portrays 3D propagation from a line source d(x) d(y) 1(z) d(t), where the

Figure 12.9. The line source solution in three dimensions.

Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 12: Initial and Boundary Conditions 369

function 1(z) is defined to be unity for all values of z. This line source is
concentrated on the z-axis. The solution arising from the line source must
be independent of z, because it is unique and also is unchanged by translation
parallel to the z-axis. Thus, the z-derivative in the 3D wave equation
vanishes, and the solution satisfies the 2D wave equation with right-side
term d(x) d(y) d(t). Therefore, the point-source solution in two dimensions
is the same as the line-source solution in three dimensions.
From a line source in 3D space, the perturbation propagates as a cylind-
rical wave with cylinder

x2 + y2 ≤ v2 t 2 . (12.83)

The front edge of the wave is the surface x2 + y2 = v2 t2 of the cylinder,

and this front edge moves with a speed v perpendicular to the z-axis. After
the front edge has passed, the perturbation is retained for an infinitely long
DOI:10.1190/1.9781560803737

time due to contributions coming in from points farther and farther down the
line source.
This last sentence can be discussed in greater detail. Because of
Huygens’ principle, a perturbation from the source d(x) d(y) 1(z) d(t) will
reach a given point r0 = (x0 , y0 , 0) at an instant of time t . 0 from those
two points given by the intersection of the line source with the spherical
shell

(x − x0 )2 + (y − y0 )2 + z2 = v2 t2 . (12.84)

These two points are

 
r1 = 0, 0, v2 t2 − x20 − y20 , r2 = −r1 . (12.85)


When t , t0 where t0 = x20 + y20 /v, there will be no disturbance at the
point (x0 , y0 , 0). At time t0 , the front edge of the wave will pass through this
point; this point edge comes from the origin (0, 0, 0). At all succeeding
instants of time t . t0 , perturbations from the line source at r1 and −r1
will reach this point, and so a composite perturbation due to these two
source points will be observed at this point at time t . t0 . Because of the
line source, the perturbation will continue indefinitely and, hence, the rear
edge of the wave is absent.
Because wave diffusion occurs in 2D propagation (r = (x, y)), in
the case of the initial impulsive source d(r)d(t), it follows that wave
diffusion also will be observed for an arbitrary initial perturbation
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

370 Basic Wave Analysis, Part 3: Waveform Analysis

Figure 12.10. The initial perturbation concentrated within a region D. This

diagram is ostensibly the same as the as the diagram in Figure 12.6. The difference is
that a circle in Figure 12.6 represents an actual sphere, whereas in this diagram a
circle represents an actual circle.
DOI:10.1190/1.9781560803737

f(r)d′ (t) + c(r)d(t). In fact, the response u(r, t) due to such an initial pertur-
bation is completely defined by the values f(r0 ) and c(r0 ) for r0 within the
circle with center r and radius vt. Thus, if the initial perturbation is
concentrated within a region D, then the perturbation u(r, t) propagates
over a region which is the union of the areas of the circles with center r0 ,
(within D) and radius vt. As shown in Figure 12.10, the front edge is
observed passing an arbitrary point, but there is no rear edge.
Finally, we can discuss wave propagation on a straight line (1D propa-
gation). The Green’s function (defined for t . 0)
⎧
⎨1, −vt ≤ x ≤ vt
g(x, t) = 2v (12.86)
⎩
0, |x| . vt

represents the perturbation from a momentarily acting point source d(r)d(t).

This perturbation at an instant of time t . 0 occupies the segment
−vt , x , vt. Thus, two front edges, occurring at x = vt and x = −vt,
will be observed moving on the straight line with speed v to the right and
left, respectively. As in the case of 2D propagation, the perturbation will
continue after the front edge has passed. In the present case, the perturbation
is constant and equal to 1/2v. Wave diffusion occurs.
Again, we can explain the reason for wave diffusion by interpreting
the one-dimensional Green’s function in three dimensions. The one-
dimensional Green’s function in 3D space represents the response to a
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 12: Initial and Boundary Conditions 371

DOI:10.1190/1.9781560803737

Figure 12.11. Interpretation of the one-dimensional Green’s function in three

dimensions.

plane impulsive source d(x) u(y) u(z) d(t) concentrated over the plane x = 0.
Here, u is the Heaviside step function. One might ponder the question, “Is it
correct that there would be edges y = z = 0 meeting in (0, 0)?” From such a
source in 3D space, the perturbation propagates in the form of a plane wave
|x| , vt.
In Figure 12.11, the leading edge |x| = vt moves with speed v, perpen-
dicular to the source plane z = 0. Note that the leading edge consists of
two planes x1 = vt and x2 = −vt, moving with speed v to the right and
left, respectively. After the front edge of the wave passes, the perturbation
continues at the same value 1/2v for an infinitely long time.
Now, we can apply Huygens’ principle. The perturbation from the
plane impulsive source will reach a given point (x0 , 0, 0) at an instant of
time t . 0 from those points which represent the intersection of the spheri-
cal shell (x − x0 )2 + y2 + z2 = v2 t2 with the source plane x = 0. The locus
of these points of intersection, therefore, is the circle C0 given by
y2 + z2 = v2 t2 − x20 , x = 0 . (12.87)
Thus, for t , t0 where t0 = |x0 |/v, there will be no disturbance at the point
(x0 , 0, 0). At time t0 , the front edge of the wave will pass through this point;
this front-end disturbance originated at the origin (0, 0, 0). At a succeeding
instant t . t0 , we observe a perturbation composed of component pertur-
bations originating on the circumference of the circle C0 . Because these
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

372 Basic Wave Analysis, Part 3: Waveform Analysis

circles spread out over an infinitely large plane (plane z = 0), these succeed-
ing perturbations continue indefinitely, so the rear edge of the wave is
absent.
Because there is 1D wave diffusion in the case an initial impulsive per-
turbation d(x)d(t), there is also 1D wave diffusion for an arbitrary initial per-
turbation c(x)d(t). However, this is not the case for the arbitrary initial
propagation f(x)d′ (t), as will be shown next.
For this case, we can examine a doublet point source d(x)d′ (t). Because
d(x)d(t) produces the perturbation g(x, t), the doublet source produces the
perturbation

∂g(x, t) ∂g(x, t) 1 ∂H(vt − |x|) 1

∗ d(x) = = = d(vt − |x|) . (12.88)
∂t ∂t 2v ∂t 2
Thus, this perturbation is concentrated at only the two points x = vt
and x = −vt. Hence, after the front edge of the wave passes, there are no
DOI:10.1190/1.9781560803737

more disturbances. In this situation, the propagation of sharp signals occurs.

Seismic navigation through the subsurface is akin to the steering a ship
in the day of sail. As Shakespeare’s contemporary, British playwright Ben
Jonson (1572– 1637), wrote:

Each petty hand

Can steer a ship becalm’d; but he that will
Govern and carry her to her ends, must know
His tides, his currents, how to shift his sails;
What she will bear in foul, what in fair weathers;
Where her springs are, her leaks, and how to stop ‘em;
What strands, what shelves, what rocks do threaten her.

Conclusion
The word “theoretical” refers to the mathematical and physical under-
pinnings of science. The word “empirical” refers to conclusions inferred
from data. In geophysics today, observational data are abundant and com-
puters are powerful. In antiquity, Aristotle divides the theoretical sciences
into three groups: physics, mathematics, and theology. Aristotle’s physics
is the study of nature. In this sense, it encompasses not only the modern
field of physics but also biology, chemistry, geology, and meteorology.
Experimentation is almost unknown in antiquity. In fact, the discipline of
physics as created by Aristotle has nothing to do with experiment or
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Chapter 12: Initial and Boundary Conditions 373

quantitative measure. Aristotle’s physics uses a priori arguments to explain

physical phenomena. In the early 1600’s, Galileo shows that Aristotle’s
description of motion is wrong, and Galileo’s correct description opens
the door to modern science.
An experimental tradition began in Western Europe during the Renais-
sance. There was also a tradition of applied mathematics. Jean Baptiste
le Rond d’Alembert argued that mathematics is basic to all of physics.
There was a belief that experimental physics had value only to the extent
that physical precepts could be made quantitative. The importance of
mathematics for experimental physics was questioned. Some scientists
even condemned the excessive use of mathematics, claiming that it led to
an inappropriate reliance on abstract systems. Even so, the general
opinion was that both experiment and mathematics were essential for arriv-
ing at knowledge. Knowledge about the physical world could not be
obtained from first principles without resorting to experiment. Reason dic-
DOI:10.1190/1.9781560803737

tated a middle course, combining experiment with quantitate measure to

allow a constant check on theory.
In the 17th century in the Netherlands, Christiaan Huygens found new
methods for grinding and polishing lenses, making telescopes more power-
ful. Also in the Netherlands, Antoni van Leeuwenhoek became the first
person to make and use a microscope. In 1688, William of Orange was
installed on the English throne, subsequently promoting increased scientific
contact between England and the Netherlands. In his Opticks, Newton
(1704) combined experiment and mathematics in an outstanding way.
Scientists on the Continent, especially in Holland, elaborated on the
experimental Newtonian method. At the University of Leiden in the early
18th century, Willem ‘s-Gravesande, Herman Boerhaave, and Pieter van
Musschenbroek followed Newton’s example in organizing experiments.
The work of ‘s-Gravesande emphasized the importance of mathematics in
experimental physics and, in effect, ‘s-Gravesande and the other Dutch
physicists redefined physics by making it the amalgam of both theoretical
and empirical methods.
Today, science finds itself in a similar situation. The computer has come
into science as a new entity that demands recognition. The previous years
can be considered a contest between theoretical and empirical methods.
Today, it is a contest between theoretical and computational methods. Let
us use exploration geophysics as an example. In the first half of the 20th
century, geophysics can be thought of as divided into two camps—theoreti-
cal and empirical. The theoretical side primarily is concerned with very
simple earth models. Despite their physical simplicity, the models require
very complex mathematics in the form of differential and integral equations.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

374 Basic Wave Analysis, Part 3: Waveform Analysis

Only in rare cases could an explicit solution be attained. In contrast, the

empirical side primarily is concerned with the recording of seismograms,
which they interpret mainly by eye. The two camps of the 20th century
have little in common.
In the first half of the 21st century, exploration geophysics remains
divided into two camps, but there are differences from the past. Now,
the theoretical is concerned primarily with the mathematics of very compli-
cated earth models. Because of their physical intricacy, the models require
the use of intricate mathematics and substantial computer programming on
the most powerful computers. On the empirical side, the concern primarily is
results. It involves the processing of vast amounts of seismic data with the
advanced computer programs in order to produce subsurface maps and sec-
tions, which they interpret mainly by eye and lifelong experience. The two
camps (theory vs. experience) have little in common. Many of the theoreti-
cal research papers involve such difficult mathematics that they become
DOI:10.1190/1.9781560803737

inaccessible to empirical people. Similarly, many of the empirical results

cover such enormous scope that they become inaccessible to theoretical
practitioners.
Amerigo Vespucci explored a new land and he was the first to perceive
that it was a continent. The only way possible at that time was by extensive
exploration coupled with well-grounded deductions in line with a remark-
ably almost-accurate conception of the circumference of the earth. In
1500, Amerigo wrote in a letter, “Rationally, let it be said in a whisper,
experience is certainly worth more than theory. I have told how far I have
sailed toward the south and toward the west.”
This situation of theoretical versus empirical methods is particularly
acute in the case of full waveform inversion (FWI). The papers of the jour-
nals have many articles on FWI, most of which involve advanced mathe-
matics. The methods of FWI follow the motion of seismic waves by use
of the wave equation. For that reason, we have included Part 3 in this
book, which serves as a basic introduction to the elusive but essential fea-
tures of the wave equation. Geophysics requires extensive exploration
coupled with well-grounded deductions in line with a remarkably almost-
accurate conception of the wave equation. As Johann Wolfgang von
Goethe wrote, “Man is not born to solve the problem of the universe, but
to find out what he has to do; and to restrain himself within the limits of
his comprehension.”
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

References

Al-Chalabi, M., 1973, Series approximation in velocity and traveltime

computations: Geophysical Prospecting, 21, no. 4, 783– 795, https://
onlinelibrary.wiley.com/doi/abs/10.1111/j.1365-2478.1973.tb00058.x.
Baker, B. B., and E. T. Copson, 1987, The Mathematical theory of Huygens’
Principle, third edition: Oxford University Press.
Bergler, S., E. Duveneck, G. Höcht, Y. Zhang, and P. Hubral, 2002,
Common-reflection-surface stack for converted waves: Studia Geophy-
DOI:10.1190/1.9781560803737

sica et Geodætica, 46, no. 2, 165– 175, https://link.springer.com/article/

10.1023/A:1019845818278.
Berkhout, A. J., 1987, Applied seismic wave theory: Elsevier.
Červený, V., 1974, Reflection and transmission coefficients for transi-
tion layers: Studia Geophysica et Geodætica, 18, no. 1, 59 –68,
https://link.springer.com/article/10.1007/BF01613709.
Červený, V., 1976, Ray amplitudes in a three-dimensional inhomogeneous
medium: Studia Geophysica et Geodætica, 20, no. 4, 401 –404, https://
link.springer.com/article/10.1007/BF01617653.
Červený, V., 2001, Seismic ray theory: Cambridge University Press.
Červený, V., 2002, Fermat’s variational principle for anisotropic inhomo-
geneous media: Studia Geophysica et Geodætica, 46, no. 3, 567–588,
https://link.springer.com/article/10.1023/A:1019599204028.
Červený, V., and M. A. de Castro, 1993, Application of dynamic ray tracing
in the 3-D inversion of seismic reflection data: Geophysical Journal
International, 113, no. 3, 776 –779, https://academic.oup.com/gji/
article/113/3/776/582174.
Červený, V. and I. Pšenčı́k, 2017, Elementary Green function as an integral
superposition of Gaussian beams in inhomogeneous anisotropic layered
structures in Cartesian coordinates: Geophysical Journal International,
210, no. 2, 561 –569, https://doi.org/10.1093/gji/ggx183.
Červený, V., I. A. Molotkov, and I. Pšenčı́k, 1977, Ray method in seismol-
ogy: Univerzita Karlova.

375
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

376 Basic Wave Analysis

de Bazelaire, E., 1988, Normal moveout revisited: Inhomogeneous media

and curved interfaces, Geophysics, 53, no. 2, 143– 157, https://doi.
org/10.1190/1.1442449.
Diephuis, G., 2001, Introduction: Geophysical Prospecting, 49, no. 6, 621–
623, https://doi.org/10.1046/j.1365-2478.2001.00301.x.
Dix, C. H., 1955, Seismic velocities from surface measurements, Geophys-
ics, 20, no. 1, 68–86, https://doi.org/10.1190/1.1438126.
Encyclopedia Britannica, 2019, Huygens’ principle: https://www.britannica.
com/science/Huygens-principle, accessed October 9, 2019.
Fauvel, J., R. Flood, and R. Wilson, 1993, Möbius and his band: Mathemat-
ics and astronomy in nineteenth-century Germany: Oxford University
Press.
Fresnel, A., 1818, Memoir on the diffraction of light, in: Classical and modern
diffraction theory, 2016, K. Klem-Musatov, H. C. Hoeber, T. J. Moser,
and M. A. Pelissier, Eds., Society of Exploration Geophysicists.
DOI:10.1190/1.9781560803737

Glogovsky V., E. Landa, S. Langman, and T. J. Moser, 2009, Validating the

velocity model: The Hamburg Score: First Break, 27, no. 3, 77 –85,
http://earthdoc.eage.org/publication/publicationdetails/?publication=
28832.
Gradshteyn, I. S., and I. M. Ryzhik, 2007, Table of integrals, series, and pro-
ducts, seventh edition: Academic Press (an imprint of Elsevier).
Green, G., 1828, An essay on the application of mathematical analysis to the
theories of electricity and magnetism: Nottingham.
Grimaldi, F. M., 1665, A physico-mathematical treatise on light, colors and
the rainbow, in: Classical and modern diffraction theory, 2016, K. Klem-
Musatov, H. C. Hoeber, T. J. Moser, and M. A. Pelissier, Eds., Society of
Exploration Geophysicists.
Hagedoorn, J. G., 1954, A process of seismic reflection interpretation: PhD
thesis, University of Utrecht, and Geophysical Prospecting, 2, no. 2,
85– 127, https://doi.org/10.1111/j.1365-2478.1954.tb01281.x.
Helmholtz, H., 1858, Theory of air vibration in pipes with open ends (Theorie
der Luftschwingungen in Röhren mit offenen Enden), in: Classical and
modern diffraction theory, 2016, K. Klem-Musatov, H. C. Hoeber,
T. J. Moser, and M. A. Pelissier, Eds., Society of Exploration
Geophysicists.
Hubral, P., 1977, Time migration—Some ray theoretical aspects: Geophy-
sical Prospecting, 25, no. 4, 738– 745, https://doi.org/10.1111/j.1365-
2478.1977.tb01200.x.
Hubral, P., 1979, A wave-front curvature approach to computing ray
amplitudes in inhomogeneous media with curved interfaces: Studia
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

References 377

Geophysica et Geodætica, 46, no. 1, 13– 19, https://link.springer.com/

article/10.1023/A:1024841509612.
Hubral, P., 1980, Wave-front curvatures in three dimensional laterally
inhomogeneous media with curved interfaces: Geophysics, 45, no. 5,
905– 913, https://doi.org/10.1190/1.1441095.
Hubral, P., 2003, Traveltime formulas of near-zero-offset primary reflec-
tions for a curved 2D measurement surface: Geophysics, 68, no. 1,
255– 261, https://doi.org/10.1190/1.1441100.
Hubral, P., and T. Krey, 1980, Interval velocities from seismic reflection
time measurements: Society of Exploration Geophysicists.
Hubral, P., and J. Schleicher, 1996, Seismic image waves: Geophysical
Journal International, 125, no. 2, 431– 442, https://doi.org/10.1111/j.
1365-246X.1996.tb00009.x.
Hubral P., J. Schleicher, M. Tygel, and C. Hanitzsch, 1993, Determination
of Fresnel zones from traveltime measurements: Geophysics, 58, no. 5,
DOI:10.1190/1.9781560803737

703– 712, https://doi.org/10.1190/1.1443454.

Huygens, C., 1690, Treatise on light, in: Classical and modern diffraction
theory, 2016, K. Klem-Musatov, H. C. Hoeber, T. J. Moser, and
M. A. Pelissier, Eds., Society of Exploration Geophysicists.
Khaidukov, V., E. Landa, and T. J. Moser, 2004, Diffraction imaging by
focusing-defocusing: An outlook on seismic superresolution: Geophys-
ics, 69, no. 6, 1478 –1490, https://doi.org/10.1190/1.1836821.
Kirchhoff, G., 1862, On the ray theory of light, in: Classical and modern dif-
fraction theory, 2016, K. Klem-Musatov, H. C. Hoeber, T. J. Moser, and
M. A. Pelissier, Eds., Society of Exploration Geophysicists.
Klem-Musatov, K. D., H. C. Hoeber, T. J. Moser, and M. A. Pelissier, 2016a,
Classical and modern diffraction theory: Society of Exploration
Geophysicists.
Klem-Musatov, K. D., H. C. Hoeber, T. J. Moser, and M. A. Pelissier,
2016b, Seismic dffraction: Society of Exploration Geophysicists.
Krey, T., 1952, The significance of diffraction in the investigation
of faults: Geophysics, 17, no. 4, 843 –858, https://doi.org/10.1190/1.
1437815.
Krey, T., 1980, Straightforward derivation of Hubral’s wavefront curvature
differential equation in inhomogeneous isotropic media: Geophysics,
45, 5, 964–967, https://doi.org/10.1190/1.1441100.
Landa, E., 2013, Goldin’s legacy: Geophysical Prospecting, 61, no. 6,
1080– 1083, https://doi.org/10.1111/1365-2478.12069.
Landa, E., S. Fomel, and T. J. Moser, 2006, Path-integral seismic imaging:
Geophysical Prospecting, 54, no. 5, 491–503, https://doi.org/10.1111/j.
1365-2478.2006.00552.x.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

378 Basic Wave Analysis

Lawyer, L., C. Bates, and R. Rice, 2001, Geophysics in the Affairs of

Mankind: A personalized history of exploration geophysics: SEG.
Louboutin, M., P. Witte, M. Lange, N. Kukreja, F. Luporini, G. Gorman, and
F. J. Herrmann, 2017, Full-waveform inversion, Part 1: Forward model-
ing: The Leading Edge, 36, no. 12, 1033 –1036, https://doi.org/10.1190/
tle36121033.1.
Louboutin, M., P. Witte, M. Lange, N. Kukreja, F. Luporini, G. Gorman,
and F. J. Herrmann, 2018, Full-waveform inversion, Part 2: Adjoint
modeling: The Leading Edge, 37, no. 1, 69–72, https://doi.org/10.
1190/tle37010069.1.
Mallet, R., 1846, On the dynamics of earthquakes; Being an attempt to
reduce their observed phenomena to the known laws of wave motion
in solids and fluids: The Transactions of the Royal Irish Academy, 21,
51– 105, https://www.jstor.org/stable/30078999.
Mallet, R., 1862, Great Neapolitan earthquake of 1857. The first principles
DOI:10.1190/1.9781560803737

of observational seismology as developed in the report to the Royal

Society of London of the expedition made by command of the
Society into the interior of the kingdom of Naples, to investigate the cir-
cumstances of the great earthquake of December 1857, Chapman &
Hall.
Mallet, R., J. Mallet, J. William, and British Association for the Advance-
ment of Science, 1858, The earthquake catalogue of the British
association, with the discussion, curves, and maps, etc., Taylor and
Francis.
Mann, J., R. Jäger, T. Müller, G. Höcht, and P. Hubral, 1999, Common-
reflection-surface stack—a real data example: Journal of Applied
Geophysics, 42, no. 3/4, 301– 318, https://doi.org/10.1016/S0926-
9851(99)00042-7.
Moser, T. J., 1991, Shortest path calculation of seismic rays: Geophysics,
56, no. 1, 59 –67, https://doi.org/10.1190/1.1442958.
Moser, T. J., 1994, Migration using the shortest-path method: Geophysics,
59, no. 7, 1110 –1120, https://doi.org/10.1190/1.1443667.
Moser, T. J., and V. Červený, 2007, Paraxial ray methods for anisotropic
inhomogeneous media: Geophysical Prospecting, 55, no. 1, 21 –37,
https://doi.org/10.1111/j.1365-2478.2006.00611.x.
Oettler J., V. S. Schmid, N. Zankl, O. Rey, A. Dress, and J. Heinze, 2013,
Fermat’s principle of least time predicts refraction of ant trails at sub-
strate borders: PLoS ONE, 8, no. 3, e59739, https://doi.org/10.1371/
journal.pone.0059739.
Rayleigh, L., 1897, On the passage of waves through apertures in plane
screens, and allied problems, in: Classical and modern diffraction
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

References 379

theory, 2016, K. Klem-Musatov, H. C. Hoeber, T. J. Moser, and M. A.

Pelissier, Eds., Society of Exploration Geophysicists.
Ricker, N., 1940, The form and nature of seismic waves and the structure of
seismograms: Geophysics, 5, no. 4, 348 –366, https://doi.org/10.1190/1.
1441816.
Robinson, E. A., 1957, Predictive decomposition of seismic traces, Geo-
physics, 22, no. 4, 767 –778, https://doi.org/10.1190/1.1438415.
Robinson, E. A., 1977, Geophysical exploration for petroleum and natural
gas: IEEE Transactions on Geoscience Electronics, 15, no. 1, 3 –11,
https://ieeexplore.ieee.org/document/4071823.
Robinson, E. A., 1981, Time series analysis of geophysical inverse scatter-
ing problems, Applied Time Series Analysis II: Elsevier, 101 –167,
https://doi.org/10.1016/B978-0-12-256420-8.50010-1.
Robinson, E. A., 1982a, Inverse problems for reflection seismograms: IFAC
Proceedings, 15, no. 4, 939–944, https://doi.org/10.1016/S1474-
DOI:10.1190/1.9781560803737

6670(17)63115-X.
Robinson, E. A., 1982b, Migration of seismic data as W.K.B. approxi-
mation: Geoexploration, 20, no. 1/2, 7 –30, https://doi.org/10.1016/
0016-7142(82)90004-7.
Robinson, E. A., 2000, Wavelet estimation and Einstein deconvolution:
The Leading Edge, 19, no. 1, 56 –59, http://dx.doi.org/10.1190/1.
1438456.
Robinson, E. A., and S. Treitel, 2008, Digital imaging and deconvolution:
The ABCs of seismic exploration and processing: Society of Explora-
tion Geophysicists.
Sherwood, J. W. C., P. S. Schultz, and J. R. Judson, 1976, Some recent
developments in migration before stack: Digicon Inc.
Tarantola, A., 2006, Popper, Bayes and the inverse problem: Nature Physics,
2, no. 8, 492 –494, https://www.nature.com/articles/nphys375.
Weglein, A., 2013, A timely and necessary antidote to indirect methods and
so-called P-wave FWI: The Leading Edge, 32, no. 10, 1192– 1204,
https://doi.org/10.1190/tle32101192.1.
Wiechert, E., 1896, Die Theorie der Elektrodynamik und die Röntgen’sche
Entdeckung: Schriften der Physikalisch-Ökonomischen Gesellschaft zu
Königsberg, 37, 1– 48.
Witte, P., M. Louboutin, K. Lensink, M. Lange, N. Kukreja, F. Luporini,
G. Gorman, and F. J. Herrmann, 2018, Full-waveform inversion, Part
3: Optimization: The Leading Edge, 37, no. 2, 142– 145, https://doi.
org/10.1190/tle37020142.1.
Yilmaz, Ö., 1979, Pre-stack partial migration, PhD thesis, Stanford
University.
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

380 Basic Wave Analysis

Yilmaz, Ö., 2008, Seismic data analysis: Society of Exploration

Geophysicists.
Young, T., 1803, Experiments and calculations relative to physical optics in:
Classical and modern diffraction theory, 2016, K. Klem-Musatov, H. C.
Hoeber, T. J. Moser, and M. A. Pelissier, Eds., Society of Exploration
Geophysicists.
Young, T., 1804, A reply to the animadversions of the Edinburgh Reviewers
[i.e. Lord Brougham], on some papers, published in the Philosophical
Transactions, Longman, Hurst, Rees, and Orme, 17 –18.
DOI:10.1190/1.9781560803737
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Index

A boundary value problems, 310–311

brachistochrone curve, 17
abstraction, 213
acoustic wave. See sound waves C
Age of Enlightenment, 8
allothetic, 234, 235 Candide, 8
alternating-current (AC) electric system, capacitors, 240
241– 242 Cartesian coordinate system, 122–126
AM. See amplitude modulation Cassini, Giovanni Domenico, 285
amplitude modulation (AM), 241– 242 Cauchy problem, 192, 310
Analyse des réfractions, 16 CDP. See common depth point
angle of emergence, 42 –43 chain rule, 160, 161
angular wave number, 135 CMP. See common midpoint
DOI:10.1190/1.9781560803737

anisotropic materials, 179 coherency, velocity estimation, 98–100

antinodes, 315 color spectrum, 285
Apollonius of Perga, 47 Columbus, Christopher, 295
astronomical unit (AU), 285 common depth point (CDP), 100
astronomy, 6, 13, 282 common midpoint (CMP), 82–98, 104–106
asymmetric ray, 212 with dipping interface, 92, 94– 95
AU. See astronomical unit with flat interface, 92–94
average velocity, 58 –64 hyperbola in dipping interface, 95 –98
normal ray perpendicular to interface,
B 89 –90
primary reflection events from interface,
Bacon, Francis, 234 84 –86
beams, 177 –178 single flat reflector, 86 –88
Gaussian, 178 single straight-line dipping interface,
light, 178 88 –89
particle, 178 stack, 100
Bell, Alexander Graham, 240 time distance curve for dipping interface,
bending of waves, 169, 192, 285, 297 –298, 90 –92
300, 302 common offset point (COP), 83
Benz, Karl, 290 common receiver point (CRP), 83
Bernoulli, Johann, 17 –20 common shot point (CSP), 81 –82, 90, 94
Bernoulli’s solution, 17 –20 compressional wave, 191
binary stars, 11, 98 Conduitt, John, 17
blocky model. See piecewise-constant conic section, 47, 84
model conjugate gradient algorithm, 35
body waves, 245 continuous waves, 241, 242
borehole geophysics, 40 –41 contour interval, 159–160

381
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

382 Basic Wave Analysis

convolution, 270, 303 dip correction, 78

convolution integral, 351 –355 dip moveout (DMO), 91 –92
Cooke, William Fothergill, 147 dip-slip fault, 8–10
coordinates, traveling waves, 121 –126 Dirac, Paul Adrien Maurice, 265
COP. See common offset point Dirac delta function, 265–271
Copernicus, Nicolaus, 281 definition, 266–267
corrections Fourier transform of, 275–277
dynamic, 77–78 rectangular approximation of, 267
static, 77 –78 sifting property, 268–270, 354
Cosmotheoros, 283 directional derivative, 160– 162
Cotes, Roger, 70–71 The Discourses and Mathematical
Cowdray, Lord, 244 Demonstrations Relating to
critical angle, 50–56 Two New Sciences, 282
critical angle of incidence, 51 dispersion, 285, 289
critical condition, 51 dispersive waves, 312
cross-hole tomography, 21 diversifiers, 291
CRP. See common receiver point diving waves, 171–174
CSP. See common shot point Dix formula for interval velocity, 106–113
DOI:10.1190/1.9781560803737

curvature, traveling waves, 126– 127 DMO. See dip moveout

cyclical wavenumber, 314 downgoing vs. upgoing waves, 152–158
cycloid, 19–20 dynamic corrections, 77 –78
cylindrical waves, 260 –261
E
D The Earthquake Catalogue of the British
Association, 79
DAB. See digital audio broadcasting earthquake seismology, 7–11
d’Alembert, Jean Baptiste Le Rond, Eckhardt, E. A., 245
154 –155, 248 eikonal equation, 15, 44– 45, 163–168
d’Alembert equations, 156, 248 –251 elastic deformation, 148, 191
damped waves, 241, 242 elasticity, 148– 149
datum plane, 41, 77, 191 electromagnetic (EM) radiation, 14–15
Death Valley, 29 electromagnetic waves, 12–13, 128, 240–243
de Broglie, Louis, 292 –293 electron microscope, 294
de Broglie hypothesis, 293 electrons, 128, 178, 284, 292–294, 296
deconvolution, 303 ellipse, 23, 47, 84
de Coulomb, Charles-Augustin, 11 elliptical orbits, 281
de Fermat, Pierre, 15 –16 emergence angle, 42 –43
DeGolyer, Everette, 245 energy signals, 287
depropagation, 210 –221 An Essay on Criticism, 36
depth point, 22–24, 44, 58, 69, 78, 80, 84, Euler, Leonard, 17, 132, 306
86, 92, 100, 159, 212 exploration geophysics, 3– 7, 374
derivative, Fourier transform of, 277 –279 extrema, 28
Dialogue Concerning the Two Chief World
Systems, 281 F
diffracted ray, 177, 195
diffraction, 296 –303 fathometer, 130
digital audio broadcasting (DAB), 242 faults, 8
digital wavelet, 303 dip-slip, 8 –10
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Index 383

oblique-slip, 8 Grimaldi, Francesco, 299

strike-slip, 8 group interval, 82– 83
Fermat’s derivation of Snell’s law,
177– 183
H
Fermat’s principle, 15–16, 180 –181,
185– 189 Hamilton, William Rowan, 167
Fessenden, Reginald, 240 –244 Heaviside, Oliver, 147, 268– 269
fixed-frame coordinate system, Heaviside step function, 266, 268–271
192– 193 Fourier transform of, 275–277
flat layered earth model, 56 –60 Heisenberg uncertainty principle, 293
flow line, 161 –163, 167, 170, 171 heliocentrism, 281
Fourier transforms, 266, 271–275 Herglotz, Gustav, 292
of delta and step functions, Herschel, William, 11
275– 277 Hertz, Heinrich, 240
of derivative, 277– 279 heterodyning, 241
Fraunhofer diffraction, 301 –302 heterogeneous material, 190
Fraunhofer, Joseph von, 301– 302 homogeneous material, 190
Fresnel, Augustin Jean, 153, 300 Hooke, Robert, 141
DOI:10.1190/1.9781560803737

Fresnel diffraction, 301 –303 Hooke’s law, 146, 149

Fresnel zone, 174 –177 horizontal phase velocity, 45
full waveform inversion (FWI), 15, 31 –37, horizontal slowness, 42 –43
171, 246– 247, 374 horizontal wavenumber, 253–254
FWI. See full waveform inversion Huygenian eyepiece, 282
Huygens, Christiaan, 240, 282–283, 285, 298
derivation of Snell’s law, 183–185
G
light cone, depiction of, 331–333
gamma rays, 12, 15 principle, 16, 185–189, 282, 289–290,
Gaussian beams, 178 298–299, 302–303, 306–311
geometrical optics, 14 reflection model, 303– 305
geometric spreading, 263, 329 wavelet, 334–335
geophones, 32, 82, 151 Huygens–Fresnel principle, 297–298,
geophysical models, 4 302, 308
building, 5–6 hyperbola, 47, 71–75
fundamental notions of, 5
geophysics, 146 I
controlled experiments, 4–5
empirical exploration, 6 idiothetic, 234
exploration, 3 –7, 374 clues, 235
theoretical investigation, 6 image ray, 89, 90, 212
Geussenhainer, Otto, 244 imaging. See migration
global positioning system (GPS), 235 impulse function. See Dirac delta function
Goethe, Johann Wolfgang von, 374 incident ray, 48, 177, 207, 209
GPS. See global positioning system raypath, 194, 198, 199
gradient, 159–160 inductive sensing, 296
gravitational redshift, 10 inertial navigation system (INS), 235
Green’s function, 291, 306 INS. See inertial navigation system
three-dimensional, 350 interference, 283– 291, 296– 297,
for wave equation, 346 –350 299–300
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

384 Basic Wave Analysis

interval velocity, 56 –64 Method of Exhaustion, 188

Dix formula for, 106 –113 Methodus Incrementorum Directa et
isotropic materials, 179, 190 Inversa, 70
iterative improvement method, 26– 27 Meucci, Antonio, 240
Mexican Eagle Oil Company (El Aquila),
K 244
Michell, John, 8–11, 79, 98
Karcher, J. C., 245 microorganisms, 291
Kelvin, Lord, 150, 240 microscope, 294
Kepler, Johannes, 281 microwaves, 12, 15
Khyber Pass, 30 migration, 24–26, 91–92, 100, 212, 213, 307
Kleist, Ewald Georg von, 240 Mintrop, Ludger, 244
Kronecker delta, 266 Mintrop seismograph, 244
modulus of elasticity, 149
L mollusks (Tochuina tetraquetra), 296
Morse, Samuel F. B., 147, 239–240
La Géométrie, 47 Morse code, 239–240
Laplace operator, 311 moving-frame coordinate systems, 193–200
DOI:10.1190/1.9781560803737

Laplace transform, 347 –348 Musschenbroek, Pieter van, 240, 373

laser, 178
law of refraction, 16, 56, 180, 183
least time, principle of, 162 –163 N
Leeuwenhoek, Antoni van, 291, 373
Leibniz, Gottfried Wilhelm, 7, 121– 122, Newton, Isaac, 284–285, 299
224, 286, 306 Newton–Cotes formulas. See quadrature
Leibniz, Gottfried Wilhelm von, 121 formulas
Leyden jar, 240 Newton’s law, 146, 149, 150
light, 14– 15, 284 –285 NIP. See normal incidence point
beam, 178 NMO. See normal moveout
cone, 331 –333 nondispersive waves, 312
velocity of, 286 nondispersive wave equation, 315–317,
linear approximation, 69– 75 321–322
Lisbon earthquake, 7–8, 79 normal incidence paths, 104, 106
lodestone, 10 –11 normal incidence point (NIP), 104, 106, 211
longitudinal wave. See compressional wave normal moveout (NMO), 84
loops, 315 correction, 78, 86, 92
The Ludger Mintrop Award, 244 velocity, 100– 106
normal ray, 210– 211
normal stress, 147
M
magnetic compass, 11, 294, 295 O
magnetism, 10 –11, 296
magnetoception, 295 –296 objective function, 27, 34 –36
magnetotactic bacteria, 296 oblique mean-square velocity, 64 –65
Mallet, Robert, 78–79 oblique root-mean-square (ORMS) velocity,
Marconi, Guglielmo, 240 –243 65 –66
matter waves, 291– 296 oblique-slip fault, 8
McCollum, Burton, 245 ocean bottom acquisition, 40
mechanical traveling waves, 12, 128 –129 offset velocity, 114
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Index 385

Olbers, Heinrich Wilhelm, 282 bending, 192

one-dimensional wave equation, 311 –316 classical, 189–200
one spatial dimension, 338 –341 depropagation, 210–221
Opticks, 373 fixed-frame coordinate system,
192–193
P interval velocities, 229–234
moving-frame coordinate systems,
Pappus of Alexandria, 47 193–200
parabola, 47, 197, 207– 208 normal moveout velocity, 221–229
partial derivative, 143, 160, 161, 170 problem, 192
particle beams, 178 propagation, 210– 221
path integration, 235 shooting method for, 192
phasor, 132 –133 time-distance curve
photons, 284 ordinate of, 201–210
physical optics, 14 second derivative of, 201, 203–210
Picard, Jean-Félix, 285, 286 slope of, 201–210, 231–232
piecewise-constant model, 190 –191 two-point, 192– 193
Pikes Peak, 29 verification, 234–236
DOI:10.1190/1.9781560803737

planets, 281 –282 wavefront curvature, determination of,

plane waves, 251 –259 200–210
plastic deformation, 148 receiver correction, 78
Poe, Edgar Allan, 282 –283 reciprocal Pythagorean theorem,
point spread function (PSF), 303 164–165
power, 287 reciprocity, principle of, 183
power signals, 287 reflection
primary reflected waves, 78 ray, 177, 210–212
principle of least time, 15 –16, 162– 163 seismology, 14, 40 –41
principle of reciprocity, 97, 183, 222, 226 time, 39
profiles/sections, 83 tomography, 21
propagation, waves, 210 –221, 364 –372 refraction, 297
propagation number. See angular wave of light waves, 297– 298
number path, 53
protons, 294 ray, 177
PSF. See point spread function seismology, 14, 41
Pythagorean equation, 255 –259 refractive index, 284
Pythagorean theorem, 102, 107, 164 The Reginald Fessenden Award, 243
remote sensing, 6–7
Q active, 7
passive, 7
quadrants, 122 resonant frequencies, 297
quadrature formulas, 70–71 reverse time migration (RTM), 15,
quantum electrodynamics, 293 24 –26
Ricker, Norman, 288
R Ricker wavelet, 288
radio waves, 15, 241 –243 Rømer, Ole, 286
ray equation, 168– 171 root-mean-square (RMS) velocity, 64– 66,
rays, 14–15, 139 –140, 168 73 –75, 110–118, 221
ray tracing, 14– 15, 168 –169, 171, 177 RTM. See reverse time migration
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

386 Basic Wave Analysis

S frequency, 136
period, 133–134
saddle point/pass, 28 –31 phasor, 132– 133
Sahara Desert ant, 295 spatial frequency, 137
Schrödinger, Erwin, 265 velocity, 136
Schwartz, Laurent, 271 wavelength, 133, 134–136
seismic convolutional model, 303 –306 slowness, 43–44, 46, 163, 169–170
seismic datum correction, 22 Snell’s law, 16, 17, 50, 56 –57, 61, 68 –69
seismic imaging, survey of Fermat’s derivation of, 177–183
depth points, 22–24 Huygens’ derivation of, 183–185
earthquake seismology, 7– 11 ray theory and, 192
exploration geophysics, 3 –7 Snell’s law of refraction, 182–183
full waveform inversion, 31 –37 Snell’s parameter, 60 –64, 66–69
geometrical seismology, 14–21 sound waves, 12, 130–131, 191, 311–312
iterative improvement method, 26–27 source correction, 78
physical seismology, 14–21 source signature, 355–358
reverse-time depth migration, prestack, space, 21
24–26 dimensions of, 145
DOI:10.1190/1.9781560803737

saddle point/pass, 28 –31 spatial function, 269

tomography, 21–22 specular reflection, laws of, 48
traveling waves, 12– 14 sphere, 333– 334
seismic reference datum (SRD), 21–23, spherical waves, 261–263
51–54 SRD. See seismic reference datum
seismic tomography, 21–22 stacking (NMO) velocity, 75, 100–106,
seismic waves, 7, 12, 245 –251 113–118
depropagation of, 210 –221 stacking time, 75, 103, 111, 113
propagation of, 210 –221 stacking velocity, approximation of, 113–118
velocity of, 39 standing waves, 12, 129, 297, 314
seismology, 13 –14, 145, 310 stars, 281–283
exploration, 14, 243 –245 static corrections, 77–78
geometric, 14, 21, 177 –178, 213 stationary phase argument, 187
physical, 14–21 stationary/standing wave, 314
reflection, 14, 39 –41, 106 steepest-ascent algorithm, 35
refraction, 14, 41 Stokes, George Gabriel, 288
teleseismology, 13 –14 strain, 148, 149
SEISMOS, 244 stress, 147–148
sensors, 32 strike-slip fault, 8
shear stress, 147 –148 subatomic particle, 294
shear wave, 191, 247 superposition principle, 290–291
shooting method, for ray tracing, 192 surface convolution, 358–363
single flat reflector, 86– 88 swiftness, 163, 167
single horizontal interface, 47– 50 symmetric ray, 211–212
singularity functions, 266 Systema Cosmicum, 281
sinusoidal waves, 132 –137
amplitude, 136 T
angular frequency, 137
angular wavenumber, 137 taut-string curvature, 127, 146–147
cycle, 133 Taylor, Brook, 70
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

Index 387

Taylor expansion, 204 two spatial dimension, 341– 346

Taylor series, 70 –71, 222, 223, 226 two-way image time, 212
telegrapher’s equations, 147– 152 two-way traveltime, 41 –42, 49
telephone, 239 –240
Tesla, Nikola, 241 –242
Tesla coil, 241 –242 U
test path, 162– 163
UFT. See unified field theory
theodicy, 7
unified field theory (UFT), 293
thermal expansion coefficient, 190
unifiers, 291
Thiessen, Alfred, 241
unit vector, 123–124, 161
Thornhill plaque, 9
universe, 281–283
three-dimensional wave equation,
325– 329
three spatial dimensions, 333 –338 V
time adjustments, 77 –78
time-distance curve, 39–45, 85 vectors, 123–126
alternative expression for, 45 –47, velocity
61–62 equation, 140, 163
DOI:10.1190/1.9781560803737

angle of emergence, 42 –43 estimation

critical angle, 50 –56 approximation of stacking velocity,
for dipping interface in CSP, 90 113–118
in exploration, 44– 45 coherency, 98–100
linear approximation and, 69–75 common midpoint, 84–98
for simple flat interface, 48– 49 corrections, 77–78
single horizontal interface, 47–50 Dix formula for interval velocity,
Snell’s parameter and, 66–69 106–113
and wavefront curve, 43 multiple coverage, 78–83
time function, 269, 271, 272, 358 stacking (NMO) velocity, 100– 106
tomography, 15, 21 –22 function, 17– 18, 20 –21, 39–40
torsion balance, 11 interval, 229–234
transverse wave. See shear wave of light, 286
traveling waves, 12 –14 normal moveout, 221–229
assumption, 154 oblique root-mean-square, 65 –66
coordinates, 121 –126 of particle, 141–142
curvature, 126– 127 of propagation, 246
electromagnetic, 12–13, 128 root-mean-square, 64 –66, 73– 75,
mechanical, 12, 128 –129 110–118, 221
motion, 128 –132 search procedure, 103
parameters, 13 of seismic waves, 39
property of, 12 –13 waves, 13, 45, 140– 144
sinusoidal, 132– 137 velocity-depth model, 22
traveltime, 159 –160 vertical phase velocity, 46, 258, 259
traveltime tomography, 21 vertical seismic profile (VSP), 40 –41
A Treatise of Artificial Magnets, 11 Vespucci, Amerigo, 374
Treatise on Light, 286, 306 –307 vibroseis, 14
tsunami, 7 Viviani, Vincenzo, 146
turning-ray tomography, 22 voice transmission. See radio waves
two-dimensional wave equation, 323 –325 VSP. See vertical seismic profile
Downloaded 09/02/20 to 132.174.252.179. Redistribution subject to SEG license or copyright; see Terms of Use at https://library.seg.org/page/policies/terms

388 Basic Wave Analysis

W shear, 191
sinusoidal, 132–137
water waves, 12, 40, 240 –241, 245 –246, sound, 130–131, 191, 311–312
282 –284, 297 –298, 301, 312, 325 spherical, 261– 263
wave equation, 15, 311– 316 standing, 12, 129, 314
with initial conditions, 363– 364 traveling, 12 –14
initial value problem for, 317– 323 coordinates, 121–126
one-dimensional, 311 –316 curvature, 126–127
three-dimensional, 325 –329 electromagnetic, 12– 13
two-dimensional, 323 –325 mechanical, 12
wavefronts, 140, 213 –221 motion, 128–132
cones, 331 –333 parameters, 13
curvature, 214– 221, 226 property of, 12–13
recursion for, 218 –221 with respect to distance, 45 –46
wavelength, 13 –16, 32, 36, 45, 128, with respect to time, 46 –47
133 –137, 139, 169, 174, 178, 213, sinusoidal, 132–137
246, 253, 255 velocity, 13, 45, 140–144
wavelet, 287 –288, 306 wave train, 245
DOI:10.1190/1.9781560803737

wave-particle duality, 293 Wheatstone, Charles, 147

waves, 14 Whittaker, Edmund, 8
basic properties of Wiechert, Emil Johann, 291–292
equation, 144–147
rays, 139 –140
telegrapher’s equations, 147 –152
X
velocity, 140 –144 x-rays, 15
wavefronts, 140
compressional, 191
cylindrical, 260– 261 Y
diving, 171– 174
Young, Thomas, 290, 299–300
downgoing vs. upgoing, 152 –158
frequency of, 246
matter, 291 –296 Z
plane, 251– 259
propagation, 210 –221, 364 –372 zero-offset reflection time, 211
seismic, 7, 245 –251 zero-offset traveltime, 86