John Ferguson (Auth.) - Mathematics in Geology-Springer Netherlands (1988)

John Ferguson
Mathematics
in Geology
Mathematics
in Geology
TITLES OF RELATED INTEREST
Analytical and computational methods in engineering rock

mechanics
E. T. Brown (ed.)
Boundary element methods in solid mechanics
S. L. Crouch & A. M. Starfield
The boundary integral equation method for porous media flow
J. Liggett & P. L-F. Liu
The dark side of the Earth
R. Muir Wood
Field mapping for geology students
F. Ahmed & D. Almond
Hemispherical projection methods in rock mechanics
S.D. Priest
Image interpretation in geology
S. Drury
Introduction to X-ray spectrometry
K. L. Williams
Numerical methods in science and engineering
G. de Vahl Davis
The outcrop quiz
J. B. Wright
A practical approach to sedimentology
R. Lindholm
Rock mechanics
B. H. G. Brady & E. T. Brown
Statistical methods in geology
R. F. Cheeney
Mathematics
in Geology
JOHN FERGUSON
Imperial College of Science and Technology,
University of London
SPRINGER-VERLAG LONDON LTD.

Springer-Verlag London 1988
Originally published by Allen & Unwin in 1988
First published in 1988
British Library Cataloguing in Publication Data
Ferguson, John
Mathematics in geology.
1. Geology--Mathematics
I. Title
510'.24553 QE33.2.M3
ISBN 978-94-015-4011-7 ISBN 978-94-015-4009-4 (eBook)
DOI 10.1007/978-94-015-4009-4
Library of Congress Cataloging-in-Publication Data
Ferguson, John.
Mathematics in geology.
Bibliography: p.
Includes index.
!. Geology - Mathematics. I. Title.
QE33.2.M3F47 1987 510'.24553 87-14336
ISBN 978-94-015-4011-7
Preface and acknowledgements
During 15 years of teaching mathematics, computing and statistics

to undergraduate students studying for a first degree in geology, a
number of difficulties have become apparent, not least amongst
which is the absence of a suitable textbook setting out the basic
mathematics which is required to understand many aspects of
modem geology. Another problem, particularly in British universit-
ies, is the wide disparity between levels of mathematical achievement
among undergraduates starting out on Earth science degree courses.
For example, about 25 per cent of a typical class of around 50
students could have no real working knowledge of calculus and at
least one-half of these will not have studied the subject at all. It is
also normal to find that perhaps only 10 per cent of the class will
have encountered matrix algebra before coming to university, while
only a similar percentage will have been taught statistics. If we add
to this the fact that about one-half of any class can be considered to
be non-numerate (although this level is falling as geology becomes
more-widely recognized as a numerical subject), there are many
problems involved in teaching mathematics.
The difficulty for the teacher lies in trying to encourage those
students who are numerate to develop their skills in the context of
geology, while trying to bring the knowledge of the remainder to a
satisfactory level. In the opinion of the author this minimum level of
achievement should consist of a well-developed understanding of the
language of mathematics which should enable the student to formu-
late a geological problem in mathematical terms to facilitate its
solution, with help and advice from a professional mathematician;
in other words, to equip the potential geologist with a language of
communication.
Experience of teaching mathematics courses suggests that the level
of application to the topic is greatly enhanced if the student can see a
direct geological application of what is being taught. Thus, although
there are many excellent mathematics textbooks that are easily and
cheaply available, there are very few which are totally oriented
towards geology. Looking at the geological literature, there is much
good basic mathematics scattered throughout textbooks, not least of
which are books such as John Davis' Statistics and data analysis in
geology or Harbaugh & Bonham-Carter's Computer simulation in
Vlll PREFACE AND ACKNOWLEDGEMENTS
geology, or journals such as the Journal of Mathematical Geology.

However, although lists of these and other similar works can be
given to students as recommended reading, to locate all of the
necessary books in the library, as well as finding the relevant
sections, would present the average student with a considerable
challenge, added to which must be the consideration that a
mathematics course may only be a very small fraction of the total
study that the student is expected to make.
This book is offered in order to achieve some of the above
objectives, and to attempt to bridge the gap between mathematics
and geology. If, by chance, it stimulates fellow geologists to think
more frequently in mathematical terms, then it will have achieved its
purpose.
I must at this point recognize the help and encouragement of my
colleagues in the Department of Geology at Imperial College, and in
particular Dr John Cosgrove, who kindly read the text and made
many suggestions for its improvement. I must also thank the
reviewers, who spent time reading the manuscript and giving the
benefit of their experience with helpful suggestions, and the many
students who have pointed out the shortcomings of my attempts to
teach them mathematics, which hopefully has enabled me to avoid
similar errors in the presentation of this work. Finally, thanks are
due to my wife, without whose encouragement and bullying the
writing would never have been completed.
John Ferguson
Contents
Preface and acknowledgements Vll
List of tables XI
Introduction
1.1 Solution of geological problems - are mathematical
methods necessary? I
1.2 Mathematical formulation of a geological problem 3
1.3 Computational aids 7
1.4 Speed and accuracy II
1.5 Computer literature 12
2 Useful mathematical ideas 13
2.1 Algebra 13
2.2 Geometry and trigonometry 43
2.3 Probability theory 51
3 Sequences and series 56
3.1 Definitions 56
3.2 Some common series 59
3.3 Geological applications 63
4 Rates of change - differentiation 76

4.1 Functions: some further definitions 76
4.2 Differentiation 78
4.3 Maxima and minima 87
4.4 Optimization and the method of Lagrange multipliers 94
4.5 Differentiation and least-squares methods 96
5 Areas and volumes: integration 105

5.1 Integration 105
5.2 Analytical methods 110
5.3 Volume calculations 115
5.4 Summation, integration and probability distributions 119
5.5 Fourier series and the analysis of sequential data 121
X CONTENTS
6 Number arrays, the algebra of matrices 129

6.1 Introduction and definitions 129
6.2 Basic matrix operations 133
6.3 Eigenvalues and eigenvectors 149
6.4 Geological applications and examples 157
7 Introduction to numerical methods 176

7.1 The need for numerical methods 176
7.2 Iterative methods for the solution of equations 181
7.3 The solution of simultaneous linear equations 187
8 Geological applications of differential equations 216

8.1 Definitions and examples 216
8.2 Numerical solution: the method of finite differences 219
9 Beyond traditional mathematics 238

9.1 Catastrophe theory 238
9.2 Information theory 246
Appendix 1 An introduction to the methods of

differentiation 255
A 1.1 The rules of differentiation and their application 255
Al.2 Standard forms and their use 258
A 1.3 Implicit differentiation 259
Appendix 2 An introduction to the methods of

integration 261
A2.1 The rules of integration 261
A2.2 Standard forms 262
A2.3 Substitution 262
Appendix 3 Computer programs 265
References 285
Index 289
List of tables
2.1 Measurements of (a) Epithyris oxonica (Buckman),

Jurassic, Bathonian, Blisworth Limestone and (b)
Turitella sp., Eocene, Bracklesham Beds 14
2.2 Data relating to the depletion of Mgl+ in 0.02 M
NaCl solution before visible aragonite-calcite
transformation was observed 16
2.3 Hypothetical grain-size data after a sieve test 18
2.4 Measurements of specimens of Gigantoproductus
inf/atus (Sarycheva) 29
2.5 Data from Table 2.4 expressed as logarithms 31
2.6 Data relating to the increase of strontium in sea
water during the cementation of aragonite ooids 33
2.7 Hypothetical data relating to the amounts of two
unique mineral components in three samples of rock 43
2.8 The ranges of values of the three common
trigonometric functions for the four quadrants of the
circle 47
3.1 The relationship between the arithmetic and
geometric means for grain-size data 63
3.2 Mineral composition of a quartz dolerite 66
3.3 Possible numbers of pieces of mineral to be found in
a subsample 69
3.4 Theoretical numbers of pieces of mineral in a
subsample 70
3.5 Data relating to the distribution of Epithyris oxonica
on a bedding plane 72
3.6 Data relating to a subfossil Scrobicularia community 74
4.1 Rate of change of growth in Homoeorhynchia acuta 82
4.2 Predicted y-value and its deviation from the
measured values 97
5.1 Values of the function y = 4x + x? and its integral 113
5.2 Power spectrum values for two series 125
5.3 Power spectrum values for two series 125
5.4 Power spectrum values for a series 126
5.5 Power spectrum values of sampled data 127
6.1 Data from Figure 6.1 recorded as a transition matrix 130
6.2 Data from Figure 6.1 sampled at 50-cm intervals 131
XII LIST OF TABLES
6.3 Conventional symbols used in matrix notation 133

6.4 Transition matrix for a sequence 157
6.5 Transition probability matrix for the data of Table
6.4 158
6.6 Result of powering the probability transition matrix
of Table 6.5 161
6.7 Transition probability and cumulative probability
matrix for a sequence 161
6.8 Calculation of an expected random matrix and
difference matrix 164
6.9 Raw data matrix relating to trace element analysis
of flint 166
6.10 Calculation of the dispersion matrix from the data
of Table 6.9 167
6.11 Part of the raw data set from Table 6.9 168
6.12 Details of calculation of the correlation matrix 169
6.13 Co-ordinates of points on the perimeter of an
outline of a spirified brachiopod before and after
shear 173
7.1 Evaluation of the series for -.}2 using a hand-held
calculator 178
7.2 Comparison of Gauss-Jordan and Gaussian
elimination in diagrammatic form 195
7.3 Percentage methane in samples of limestone 213
8.1 Values obtained for the finite-difference solution of
the equation d 2y/<fx2 = 5x after 99 iterations 223
Al.l Some common standard differential forms 258
A2.1 Some standard integral forms 262
Mathematics
in Geology
1 Introduction
1.1 Solution of geological problems- are mathematical

methods necessary?
A question which is often asked is whether it is necessary for

geologists to know and to use mathematics in the practise of their
science. There is no simple answer to this question, and it is true that
many geologists have had successful careers without ever needing
to get involved in anything other than simple mathematics, and all
the indications are that this is likely to continue into the future.
However, in many branches of the subject the trend has been
towards using a numerical approach for the solution of suitable
problems. The extent to which this occurs depends on the nature of
the area being studied; thus, in structural geology, which is con-
cerned in its simplest aspects with the geometrical relationships
between various features, there are many problems which are easily
solved. More recently the use of analytical methods has allowed the
solution of more-difficult problems. In another area, geochemistry,
two things have happened. On the theoretical side there has been a
greater integration with physical chemistry, which itself is a highly
mathematical subject; and on the practical side there is the need to
analyse and interpret the vast quantities of data which modem
instrumentation produces.
Within geology the application of numerical methods has been
given various names, so we have numerical geology, geo-
mathematics, geostatistics and geosimulation. Most university geo-
logy courses which concern themselves with numerical methods for
geologists largely deal with statistical applications in the subject,
with perhaps a little on simulation and computing thrown in for
good measure. Another trend which in some ways is extremely
useful is the use of relatively cheap and easily available statistical
packages, such as Microtab, published by Edward Arnold (Higgin-
botham 1985), which allow the rapid processing of large data sets, to
give perfectly satisfactory results in most cases. However, experience
has shown that in a limited number of situations things can go
wrong and, unless the operator is well versed with the particula:~
package being used, the chances of sorting things out for that
particular data set can be very slim.
2 INTRODUCTION
Again, there is the unexpected answer. The author remembers

well, as a postgraduate student, looking at results from a trend
surface mapping program and commenting 'but I thought that the
anticline drawn by the computer was a syncline!'. In this situation a
little more knowledge of the 'mathematics' of the algorithm used by
the computer program in question would have cleared up the
problem. The structure which was being mapped by the program
was, indeed, a syncline, and the opposite result produced was a
mathematical artefact, due to a problem of round-off error encoun-
tered during a matrix-inversion routine. Many more examples could
be given of misleading results being produced and accepted, and in
some cases published. There is no real excuse for this, but as long as
it happens doubts are going to be cast on the validity of using some
numerical methods in geology. Not only should we all remember the
truth of the old adage 'garbage in equals garbage out', but we should
also be aware that the converse of this statement is not always true!
The author firmly believes that some basic training in mathemat-
ics can prevent the recurrence of these sorts of problems. This book,
then, is offered as one means of rectifying this lack of basic
knowledge, and is largely directed towards considering the numeri-
cal methods most frequently used in geology today. Since knowledge
of basic mathematics is so fundamental to the understanding of the
methods, it is humbly suggested that the study of the subject should
take precedence over the teaching of most of what passes as
numerical geology. If, due to lack of time, a choice has to be made
between, say, a course in mathematics or a course in geostatistics,
then the former should be favoured. Also, in the field of computer
applications in geology there are many who think that the solution is
to let students loose with statistical packages. In reality this can be
no substitute for hands-on experience gained from writing programs
and running data sets.
This book therefore tries to present, in an easily digestible form,
some basic mathematics, knowledge of which will be useful to
anyone wishing to pursue numerical applications in geology. It is
assumed that the basic level of mathematical ability of those using
the book includes a knowledge of basic trigonometry, geometry and
the manipulation of simple algebraic equations. The work is not
rigorous, in the sense that it does not contain theorems and proofs;
indeed, there are many cases where phrases such as 'it can be shown
that .. .' are used. For those who wish to follow up individual
algorithms, sufficient references are given. As far as possible, the
examples used to illustrate the applications are geological in nature,
FORMULATION OF A GEOLOGICAL PROBLEM 3
using the author's own teaching material or material gleaned from
other sources and, in particular, from the geological literature. In the
latter case, full references are given.
Although there has been no intention to produce another statistics
or simulation text, these topics have been included where it has been
necessary to demonstrate the link between the mathematical ideas
being discussed and the areas of application. These factors can be of
particular importance in areas where geological decisions can have
economic consequences- the difference between profit and loss may
hang on a few parts per hundred million of a potentially interesting
deposit.
1.2 Mathematical formulation of a geological problem
As has already been mentioned, some aspects of geology are more

explicitly mathematical than others. Thus, for example, if we are
interested in rates of change the problem can, if it is reasonably
simple, usually be formulated in such a way as to allow the use of
differential calculus. This will give an analytical expression whose
solution can give an accurate answer. Similarly, in the calculation of
volumes, provided some mathematical expression can be found
describing the geometry of the object in question, integration can
lead to an accurate answer. Even in complex cases where it is
difficult to derive an equation describing the object, the use of the
principles of integration, using a numt;rical approach, can often lead
to a satisfactory, although not absolute, answer.
In other areas such as sedimentology - where grain size studies,
for example, can lead to numerical data - it is generally much more
difficult to formulate equations describing large-scale sedimentary
processes. However, it is possible to arrive at general expressions
which will allow further possibilities to be explored. We can take as
an example Elliott's Episodal Theory for coal measure deposition
(Elliott 1970). The following simplified resume has been adapted
from the introductory section of Elliott's paper:
(1) Deposition takes place during uninterrupted episodes of sub-

delta advance and retreat.
(2) During episodes of retreat, the major distributaries and asso-
ciated depositional environments give way to more extensive
smaller scale distributary systems and environments. This pro-
4 INTRODUCTION
cess increases the chance of intra-deltaic deposition at any

particular locality.
(3) Retreat leads to reductions in sub-delta topographical relief,
which gives rise to anaerobic conditions and perhaps later to the
spread of conditions favouring the development of seat earth
and peat in areas of suitable relief.
(4) Four types of sedimentation are present: an upwards developing
clastic succession of faunal mudstones to flaser silt sandstones
which are typical of the interdistributary areas; clastic succes-
sions of massive siltstones, complex silt sandstones, layered
sandstones and rippled sandstones laterally and terminally asso-
ciated with distributaries; washout and fill successions of sand-
stones which fill the distributary channels; and lastly, hydrologi-
cally controlled seat earths and coals in the swamp areas.
(5) Individual episodes of sedimentation are often terminated by
geographical changes associated with delta switching up-stream.
In areas outside this influence characteristic deposition is unin-
terrupted. When episodes are terminated by delta switching,
individual cyclotherms can show markedly differing boundaries
from area to area.
(6) Major environment sub-groups recognized during each episode
are: swamp, prodeltaic-interdistributary and interdistributary.
Geographic changes referred to in (5) above, are considered to
give rise to a random ordering of the deposits. It follows from
this that a coal seam found at one locality may arise from several
episodes of deposition within the swamp region or, similarly, a
prodeltaic environment could persist during several episodes.
(7) The thickness of successions which accumulate during one
episode can be influenced by contemporaneous compaction of
underlying deposits which can give rise to thicker than normal
deposition; this mechanism is particularly dominant where
intradeltaic deposits overlay thick peats.
(8) Variations in thickness of deposits accumulated during indivi-
dual episodes can also be influenced by regional subsidence.
(9) The geometry of some sedimentary bodies, for example levees,
can also influence local variations in thickness.
In other words, throughout the delta, characteristic sedimentary

sequences are being deposited at any particular geographical
locality. Such deposition continues until it is interrupted by changes
brought about by delta-switching due to natural phenomena. For
example, big storms or hurricanes can cause sediment to block the
FORMULATION OF A GEOLOGICAL PROBLEM 5
major distributary channel, and can bring about extensive changes
in the delta geography. Thus characteristic sequences are built up
beeause of random changes in geography.
A mathematical model should take into account at least the more
important elements of this theory. Consider the processes which
have controlled the deposition of a typical coal measure sequence at
a single point in space, through which we have drilled an exploratory
bore hole. The relationship between sediment and process can be
expressed in the form of a function:
Y= f(A, B, .. .)
(functions and function notation will be discussed in detail in

Section 2.1.5). Putting this expression into words: Y, the sediment
deposited during the interval under study, is equal to a combination
of equations which describe the various sedimentological processes
concerned in the deposition of the sediment. They are:
A, subdelta advance and retreat (leading to changes in the local

environment)
B, typical sequences developed in relation to geography;
C, local sedimentation episodes (controlled by A);
D, changes in the thickness of units (due to compaction of the
underlying peat);
D',changes in the thickness of units (due to regional compaction);
E, effect of the geometry of the individual bodies; and
F, major changes brought about by natural catastrophes.
Thus, we can say that
Y= f(A, B, C, D, D', E, F)
In his original paper, Elliott (1970) presents a simulation model

based on the nine elements of his theory, which allow the simulation
of a stratigraphic sequence by combining individual sedimentary
episodes, based on the following.
(a) A schematic map showing the subdivision of the delta into 100
sedimentation zones. For purposes of the simulation, the co-
ordinates of the corners of each zone are used as data to enable
the correct sedimentation sequence to be assigned, for each
depositional episode.
6 INTRODUCTION
(b) A series of 45 sedimentation sequences cross-referenced to the

zones of the map, each of which depicts a facies or sequence to
be deposited during one episode.
(c) Changes in delta geography are simulated by selecting co-
ordinate pairs (eastings and northings) at random, and by
assigning episodes according to (a) and (b), above.
(d) The thickness of sequences is assigned by generating random
numbers with known mean and standard deviation. Two fre-
quency distributions taken from actual stratigraphic sequences
are used (Elliott 1970, p. 326, text fig, 4), one relating to the
thickness of seat earths and the other to all other depositional
episodes. Coal thickness is derived from the second of these,
dividing the selected thickness by 12.
Using this approach it was not necessary to define any of the

contributing functions A-F, mathematically. However, it would be
possible to define these functions precisely, and to build a totally
mathematical simulation model. To show how this might be
achieved, we can consider the effect of the compaction of peat. In a
recent paper Elliott (1985) has shown that the compaction ratio for
peat has a linear relationship with moisture content:
where Mv = moisture content by volume and Rc = compaction ratio,

calculated as ~Z/ ~Z' as in Figure 1.1.
Another expression relating to the compaction of peat has been
presented by Falini ( 1965, p. 1336), where the pore index p is related
to the loading stress cr by
where e is the base of natural logarithms and cr is expressed in

kg m -z. The compressibility, K, is given by
K= dz/z
dcr
where the thickness passes from z to (z- dz), i.e. from ~Z to ~Z' in
Figure 1.1.
These two equations can be related back to that given by Elliott
(1985), through the pore index p, which for original peaty material
COMPUTATIONAL AIDS 7
_________
-rr-

hs hl
~~~~ _jl_
I..I. .Ill
I........... ..,"z.----____,.J.h;-
... .. t ~
'../.z.i.i.i.
........ . ,_z.i. ..::I._Jf__
Figure 1.1 Diagrammatic representation of the reduction in thickness of peat, from

original thickness !l.Z to !l.Z', due to overlying sediment column of thickness h, and
water column height h, (after Falini 1965).
with 95% moisture content is taken as 20. Brown coals with

moisture contents of between 50 and 70% have pore indices of 1-2
and bituminous coals with moisture contents of less than 10% have
pore indices of less than unity (Falini 1965, p. 1336). For the
purposes of our model we need only consider the early effects, i.e.
forest loading and sediment loading (Elliott 1985, p. 169, fig. 6).
Between these limits the compaction ratios are between 1 and 3.7,
pore indices are between 20 and 4, and moisture content (by volume)
between 97 and 82%.
Thus, we could, if we so desired, build one of these equations into
Elliott's model to replace the present method of assignment of
thickness for coal deposition. It would not be difficult to develop a
similar equation relating the compaction of other sedimentary types
in response to overburden pressure. This process of finding
mathematical expressions to replace our unknowns can, of course,
be continued, but it should be borne in mind that even without doing
so the approach adopted by Elliott has led to a useful and valid
model, albeit based on geological rather than mathematical data. If
such an exercise were undertaken, the outcome would be to lead us
to a deeper understanding of the geological processes which we are
studying.
1.3 Computational aids
The traditional aids to computation, the slide-rule and the book of

logarithmic tables (and other functions) are rapidly becoming
historical relics. In the mid-1970s hand-held calculators with exten-
sive mathematical function keys were relatively expensive. Today,
for the same financial outlay a desk-computer can be purchased and
hand-held calculators have become so cheap that when a fault
occurs in one, it is thrown away and a new one is bought. Also, until
relatively recently, in the field of education, computers were re-
8 INTRODUCTION
stricted to universities, but now they are in infants' classrooms.

Geology, like many other scientific subjects, moved uneasily into the
computing age during the 1960s, but having got the bit between the
teeth it has steadfastly refused to let it go. Gone are the days of
tediously working through long columns of figures by hand. Data
are now automatically filed using some suitable interface between
machine and computer, so that the data are ready for analysis
almost as soon as they are produced.
Concurrent with advances in computational aids and their availa-
bility is a decrease in numeracy among many students. A very
noticeable feature among classes of undergraduate students today is
that if a ban is put on the use of a calculator, then their ability to do
even the simplest numerical operations drops dramatically. Gone
are the days when the result of ll x 12 came out automatically.
Even the facility of using logarithmic tables has been lost, and blank
faces greet the teacher who has the temerity to introduce logarithms
into his lecture. Of course, these are extremes, but the point should
be made that there is no substitute in the learning process for sitting
down and working through a problem by hand. Indeed, if test data
are not treated in this way, how is one to assess the validity of the
machine-generated output in the first instance?
Thus, although computer programs which are designed to solve
many of the problems considered in this book are given, it is
suggested that students work through some of the examples by
hand. This will enable the intermediate steps in the calculation to be
seen, as well as providing data which will enable the working of the
computer program concerned to be checked.
1.3.1 Computers and computer languages

The technology of computer production is advancing so rapidly that
any comments other than the most general are likely to be out of
date before any book on the subject is published. However, some
comments are in order. There are currently two distinct groups of
machines: the mainframe computer and the microcomputer (usually
referred to as the micro). The former is usually static and physically
very large. It is very fast computationally, and is capable of
supporting many different input and output devices. The micro, on
the other hand, is small, often mobile, and usually supports a limited
number of input and output devices. Again, roughly, micros can be
divided into two groups by their initial cost: one group is priced in
the range 2000-10 000, while the other group ranges from around
100 to 1000. Of these two groups the cheaper ones are often
COMPUTATIONAL AIDS 9
regarded as mere toys, although many have been used for scientific
application with great success.
Most students, ~t least initially, are more likely to be involved
with micros rather than mainframes although, depending on the
facilities available, the graduation from one to the other can be
relatively rapid, particularly since many micros can now be easily
adapted for use as remote terminals to mainframe computers. The
mode of operation of the two types need not be considered here,
although the differences in how the two types go about their business
can be confusing to the uninitiated, as can the terminology used. At
this stage we need only note that hardware refers to the machine
itself and the peripheral devices such as printers, tape-drives, etc.,
whereas software refers to the 'programs' (note that it is conven-
tional to use the American spelling of the word in this context) which
are used to instruct the machine on how to perform the task at hand.
Programs are written in the form of instructions, which the
computer can translate into a code which will enable it to perform
the task required and, when all of the necessary instructions are at
hand, to carry them out and present the results, in a form which can
be understood. As human beings communicate with one another via
language, so we communicate with the computer via language.
There are many languages available for use in this area; some are
quite specific to a given machine, whereas others are designed with a
particular purpose in mind. Thus, the choice of a language will
depend on machine availability and the problem being tackled.
Many computer languages- like their human counterparts- have
dialects, which have been developed to suit either a particular
purpose or a particular machine. Languages also have rules which
relate to their use, as well as syntax. The objective of these languages
is to enable the user to communicate with the computer in an easily
understood fashion (from the user's point of view), to enable it to
perform the operation required.
Finally, we must always remember that today's computers are
idiots; they can interpret instructions as they are given, but cannot
yet interpolate in the sense that human beings can or supply missing
instructions. In general, a computer needs an explicit instruction at
every point.
1.3.2 Computers in geology

There is an extensive literature on the application of computers to
solve geological problems, and many programs have been published.
Notable contributions have come from Krumbein & Graybill
10 INTRODUCTION
(1965), Harbaugh & Merriam (1968), Harbaugh & Bonham-Carter

(1970) and Davis (1973). Some of these books contain computer
programs which enable the reader, if he or she so desires, to solve a
vast number of geological problems, but are mainly concerned with
either statistics or simulation. The programs are written in Fortran
(FORmula TRANslation) programming language, a high-level lan-
guage which allows easy translation of mathematical expressions
into program statements. Indeed, until comparatively recently most
published geological programs have been written in this language
and its use has been widespread in geology since the mid-1960s.
Some very early programs were written in Algol (ALGOrithmic
Language), but these often appeared translated into Fortran later.
Several versions of Fortran exist but currently most computers use
Fortran 77, a standard version of the language. Most published
geological programs are in the version of the language known as
Fortran IV, but it should not prove difficult to update them if
necessary.
More recently, because of the easy availability of micro-
computers, many geological programs have been written in BASIC
(Beginners All-purpose Symbolic Instruction Code). This change
has occurred because of technical difficulties in implementing the
Fortran language on a micro. There are many versions of BASIC
and, in the UK at least, much geological work has been done using a
version known as BBC-BASIC, which is implemented on the BBC
microcomputer. The language is not unlike Fortran, and conversion
from one to the other is relatively simple. However, there are some
facilities which are unique to both languages. Another language
which has received much attention in scientific work generally is
Pascal, which can be implemented on most micro computers. Both
BASIC and Pascal languages are also available for use on main-
frame computers.
Throughout this book a number of programs are referred to
which enable some of the more-complex mathematical operations to
be carried out more easily; listings are given in Appendix 3. The
programs are written in BASIC, because of its relatively wider
availability on microcomputers, although it should be stressed that
it is BBC-BASIC which has been used and some of the programs
contain statements or functions which are specific to the BBC
operating system. A number of programs are taken from the
author's own collection of material, while others are taken from the
literature. A few have been written specifically for this book, and
these are the only ones which are not adaptations of Fortran
SPEED AND ACCURACY 11
originals. Thus, if the majority of programs look like Fortran
programs, this is the reason. All have been extensively tested and
should prove to be relatively robust in use. Unfortunately, there is
no way of guaranteeing robustness under every circumstance, and it
is always possible to find a data set which will cause a program to
crash.
Although the programs, as written, will stand alone, they are
probably most useful if they are used as subroutines as part of
larger, more-complex packages designed with a specific task in mind.
Indeed, one of the virtues of having to learn a programming
language is that it not only teaches geologists to write their own
programs, but it also enables them to bring together and use parts of
already-published programs in such a way that the parts perform
specific tasks, obviating the necessity of writing complete programs
from scratch, to suit each individual problem as it arises.
1.4 Speed and accuracy
Speed and accuracy are two problems which, although they may
appear peripheral to the majority of matters discussed in this book,
turn out to be important in practice. Needless to say, they affect
some algorithms more than others. Later, and particularly in the
context of numerical analysis, some of the detail will be discussed
more fully. At this point a few general comments can be made. Since
it is fairly safe to say that these are usually not factors of concern to
the average mainframe-computer user, the comments are particu-
larly directed at users of micros. Some of the comments, especially
those concerned with accuracy, also apply when using hand-held
calculators.
Problems related to speed and accuracy arise in many ways, but
most importantly they relate to the word-length and the speed of the
machine being used. In some applications the efficiency of the
program itself will play an important part, particularly with regard
to the speed of execution. Since accuracy is related to word-length,
i.e. the number of significant figures which can be carried as a
number, it is fairly easy to check. It is possible to write a program to
evaluate a number such as 1/3, which cannot be expressed as a
rational number, and then successively to multiply the result by the
fractions 1/10, 1/100, etc., for say 50 increments. Performing such a
calculation on the BBC micro using the BASIC compiler, it is found
that the dynamic range of the exponent is down to 10- 39 , while the
12 INTRODUCTION
mantissa is expressed to seven-figure accuracy. Beyond 10- 39 the

program is terminated with a floating point underflow. Another area
which can be troublesome, and which is worth looking at, is the
conversion from a real number to its logarithm and back again.
Some quite surprising results can be obtained by checking these and
other functions. It is recommended that similar checks be made if
other compilers are being used. For example, A. R. Miller ( 1981) has
shown that there are considerable differences in the numerical
accuracy of different commercially available Pascal compilers.
1.5 Computer literature
Although computer programming is an important aspect of numeri-

cal geology, it is not proposed to follow the example of some earlier
texts by including some simple instruction on programming. It is
suggested that those interested in following up this aspect of the
work use some of the excellent books already available on the
subject.
A number of books which can be recommended include the
following. For an introduction to computers generally, the work by
Hollingdale & Toothill (1975), in the Pelican Series is excellent. The
user-guides issued with individual computers are generally good, but
several do assume much more knowledge than the beginner nor-
mally has, and are not really written with the simple application
aspect in mind. The BBC User Guide by Coli (1982) is essential for
those using a BBC micro. This book is clearly written, and straight-
forward to use. Of a more-specific nature and giving listings of
useful programs are A. R. Miller's (1981) text on the use of Pascal in
scientific programming generally, and Unwin & Dawson (1985), on
geographical programming applications. Shelley's (1984) text on the
essentials of Fortran 77 should prove invaluable for those using
mainframe computers.
2 Useful mathematical
ideas
To provide the basic level of mathematics needed to work through

this book, it is necessary to start by looking at some simple
mathematical ideas in the fields of algebra, geometry and trigo-
nometry. For most students this will be revision. However, since
throughout the book we are attempting to link mathematics to
geology, it is not intended that this chapter should be skipped on the
basis of having covered the ground at an earlier stage. Indeed, to
establish firmly the idea of thinking of geological problems in
mathematical terms, the topics covered in this chapter start with
geological data or problems which are then expressed mathemati-
cally.
2.1 Algebra
2.1.1 Linear equations
(a) THE EQUATION OF A STRAIGHT LINE

A number of specimens of the Jurassic brachiopod Epithyris oxonica
(Buckman) were measured, and the length and width oT the pedicle
valve were recorded. The data are shown in Table 2.1 a and
graphically in Figure 2.1.
As can be seen from Figure 2.1, a straight line passing through the
origin of the graph which honours most of the points, can be fitted to
the data. The equation of this line will be of the general form
y=ax
where y = the length of the pedicle valve, x = the width of the

pedicle valve and a = the slope of the line. The slope can be
conveniently found by either of two methods. If the graph is drawn
so that the scale for the two axes is the same and the slope of the line
is no steeper than about 60, then measuring the angle that it makes
with the horizontal axis and finding its tangent will suffice. This
angle is normally denoted by the symbol 9. Alternatively (and
14 USEFUL MATHEMATICAL IDEAS
Table 2.1 Measurements of (a) Epithyris oxonica (Buckman), Jurassic, Bathonian,
Blisworth Limestone, collected from Irchester, Northamptonshire, and (b) Turitel/a
sp., Eocene, Bracklesham Beds, collected from Bracklesham Bay, Hampshire.
(a) Epithyris oxonica (b) Turitella sp.
Length of pedicle Width of pedicle Length of shell Diameter of last

valve valve (mm) chamber
(mm) (mm) (mm)
18.0 16.0 10.0 3.0

19.0 17.0 13.0 3.5
20.0 19.0 16.0 5.0
21.5 19.0 22.0 6.0
26.5 23.5 24.0 6.5
29.0 25.0 28.0 6.0
33.5 29.0 35.0 8.0
36.0 31.0 39.0 8.0
1.0
30
E
E
I
f- 20 0
l.?
z
L.U
a= 49 o
_J 10
10 20 30 1.0
WIDTH mm
Figure 2.1 Plot of length against width for the measurements given in Table 2.la, of
the Jurassic brachiopod Epithyris oxonica (Buckman). Length = length of pedicle
valve; width = maximum width of shell at right angles to length.
particularly if the two axes are not equally scaled), the vertical rise
per unit length increase along the horizontal axis can be calculated.
Whichever method is used, the sign for the slope is determined by its
direction, being positive if the slope is upwards from left to right,
and negative if the slope is upwards from right to left. In this
example the measured angle e = 49 and tan e = 1.15.
Alternatively, we can measure; thus for a rise of 36 units on the
(vertical) y-axis, we have 31 units on the (horizontal) x-axis, so
ALGEBRA 15
36/31 = 1.16
Thus, we can express the growth of the terebratullid shell in terms of

the length and width of the pedicle valve:
length = 1.15 x width
Plotting the data relating to Turitella sp. from Table 2.1, which is
illustrated in Figure 2.2, we find that the best-fit line no longer passes
through the origin of the graph, but has an intercept on they-axis, at
-8.5 mm. Any equation describing the line must take this factor
into account, so the general equation is
y=ax+b
where y = the length of the spire, x = the diameter of the last

chamber, a= the slope of the line and b =the intercept of the line on
the vertical axis. Proceeding as before,
0=79, tan0=5.145
or
39/8 = 4.875
40
E 30
E
I
f- 20
lJ
zw
_J
10
9= 79
0 1+----,-1'c-o-~2o
Figure 2.2 Plot of length against dia-
DIAMETER mm meter for the measurements given in
-1 0 Table 2.1 b, of the Eocene gastropod
Turite//a sp. Diameter = diameter of last
preserved whorl; length = length of pre-
- 20 served spire.
Note here that a small inaccuracy incurred in measuring the angle

has made a significant difference in the result; this follows since, for
angles greater than 60, the value of the tangent increases very
rapidly for very small changes in the steepness of the slope.
Since the intercept is - 8.5, we can express the growth of our
gastropod as
length of spire= (4.875 x last chamber diameter)- 8.5

(measurements in mm).
Another example can be studied using results from experiments
designed to examine the phase change of aragonite to calcite, and
the effect of magnesium-ion inhibition on the reaction. Bischoff
(1968, fig. 6) presented data relating to the removal of magnesium
ions from solution, before the visual detection of calcite precipi-
tation, in the form of a diagram. The data in Table 2.2 are taken
from this diagram, and are also replotted as Figure 2.3.
Table 2.2 Data relating to the

depletion of Mg2+ in 0.02 M NaCI
solution before visible aragonite-
calcite transformation was
observed.
Time (h) [Mg2+] (ppm)
0.0 5.5
1.5 5.1
3.0 5.375
7.0 4.75
21.5 3.75
32.0 2.5
36.0 1.75
:r
E
a.
a.
,---, g_ 6"
+
N
0'
6
10 20 30
TIME h
Figure 2.3 Plot of the concentration of Mg2+ in aqueous solution against time,
recorded during an experimental study of the phase change aragonite-<:alcite in the
presence of magnesium (after Bischoff 1968).
ALGEBRA 17
The first thing which is apparent is that the slope of the line is in
the opposite sense to the previous examples. If we follow the rules we
have already used, we find
8 = -6, tan8 = -0.105
remember, since the line slopes from left to right the sign of the angle
e is reversed; alternatively
8 = (1.75- 5.5)/32
= -0.117
The intercept on they-axis is 5.5. Therefore, we can express the rate

of reduction of Mg 2 + in solution during the experiment as
ppm Mg2+ = ( -0.117 x time)+ 5.5

(time measured in hours).
From these three examples, we can now give a general equation
for a straight line:
y= ax+ b
where a = the slope of the line

positive if slope is upwards from left to right
negative if slope is upwards from right to left
b = the intercept on the y-axis
zero if intercept is at the origin of the axes
positive if intercept is above the origin
negative if intercept is below the origin
In these three examples we have 'eyeballed' in the best-fit line to

our data, attempting to honour as many data points as possible.
Clearly, this is not the most objective way to achieve this, and later
(Section 4.5.1) we shall be looking at methods of fitting these lines
mathematically.
Unfortunately, not all of the parameters in which geologists are
interested turn out as a data plot which can be honoured by a
straight line. Take, as another example, data from grain-size analysis
of sediments. Table 2.3 shows an hypothetical data set representing
the cumulative weight percentage of various grades of sand retained
on a set of British Standard sieves, which split the sample into
Table 2.3 Hypothetical grain-size data after a sieve test using British Standard
sieves at approximately half-<p intervals
Particle diameter
Size class Cumulative weight
percentage
mm <p units
2.00 -1.00 9
very coarse 1.40 -0.49 18
1.00 0.00 27
coarse 0.71 0.49 36
0.50 1.00 45
Sand medium 0.355 1.49 54
0.25 2.00 63
fine 0.18 2.47 72
0.125 3.00 81
very fine 0.09 3.47 90
0.063 3.97 100
fractions according to the accepted size classes in the sand grade.

Two scales of measurement are shown, these are mm and q> units
(explained below).
If we plot cumulative percentage against the mm scale, then the
best-fit line is curved, whereas if we plot the same data against the q>-
value we get a straight line. These plots are illustrated in Figure 2.4.
The straight-line q>-value plot is much easier to deal with than the
curve obtained using the mm plot. The reason for this becomes
apparent if we look at the relationship between the boundaries
of the size classes (very coarse, coarse, etc.) in terms of the sieve size
in mm. The classes are based on a geometric scale, i.e. 2, l, 0.5, 0.25,
0.125, ... , so going from the largest to the smallest size class the
ratio between each interval is one-half. In order to arrive at a linear
scale for the class boundaries with this sort of scale we can use the
formula
q> = -logz d
where d = diameter in mm and q> is a dimensionless number. It

should be remembered that to calculate the logarithm to base z of
any number x
logzx = lnxjln z
The convention used to denote logarithms in this book is that the

ALGEBRA 19
abbreviation 'In' denotes a logarithm to the base e (natural logar-
ithm), 'log' denotes logarithm to the base 10, and 'log/ denotes
logarithm to any other base, specified by the numerical value of z.
The principal advantages of using the <p scale in grain-size studies
are the following.
(a) If the data are plotted on ordinary graph paper a straight line is
produced, i.e. the plot is linear.
(b) The calculation of various statistical parameters used in sedi-
ment grain-size analysis is much simplified.
'""
-.../
v'<
'&'""'
2
V1
r-
z
:J
-B-
1:L
:L
sa 100
CUMULATIVE%
Figure 2.4 Plot of the data given in Table 2.3 relating to

a hypothetical grain-size analysis, showing the difference
- 1 between the mm and <p unit plots.
For further information on this important topic, students are

referred to a standard sedimentology textbook such as that of
Friedman & Sanders (1978, pp. 642ff.).
Another example involving curved plots can be taken from details
of other experimental work on the aragonite-calcite inversion
problem, published by Bischoff & Fyfe (1968). In these experiments
they found that plots for the rate of precipitation of calcite against
time in their experiments were non-linear. However, a plot of the
percentage of calcite against the square of the time gave a straight
line. This was an extremely significant result, as one model which
was thought to describe the rate of precipitation is
where gu> =the growth rate, kg= Boltzmann's constant for the
reaction and t = time. This equation simply means that the rate of
calcite precipitation will be dependent on the surface area available
for nucleation of the crystals. Clearly, since the plot of t 2 is linear,
and not t 3 as expected, then the model described by the equation is
not satisfactory. Thus, the authors were able to conclude (Bischoff &
Fyfe 1968, p. 74) that the growth rate of calcite, under the conditions
of the experiment, is independent of the surface area available for
nucleation.
These last two examples have shown that the relationship between
two measured variables x and y is not necessarily linear, although
there are many areas of geological interest, where simple measure-
ments can be made, which follow the linear equation. It is suggested
that the student spends some time trying out a few ideas of his or her
own, measuring and plotting pairs of variables of geological interest,
to see what sorts of plots are obtained. In order to assist in this
exercise, Program 2.1 is a listing of a graph-plotting program, which
will plot a graph on standard (22 x 28 em) listing paper. It should be
noted that it is intended solely for printer output, and the plot which
appears on the VDU during the running of the program will not be
easily interpreted.
So far in our discussion of linear equations we have considered
them as a means of expressing the relationship between two mea-
sured variables, one of which is, or is arbitrarily designated as, the
dependent variable, the other being the independent variable. Going
back to the earlier example, illustrating the growth of Turitella sp.,
the fitted line had the equation
y = 4.875x- 8.5
ALGEBRA 21
where x =the diameter of the last chamber (the independent vari-
able) andy= the length of the spire (the dependent variable). In this
example the coefficient of the x-value is defined as being a measure
of the slope of the line, the constant - 8.5, is the intercept on they-
axis, and the coefficient of they-value is unity. We can rewrite this
more conveniently, with the unknowns x and y on one side of the
equation (rounding-off in the process) as
5x- y = 8.5
Now if we arbitrarily alter the coefficient of the y-value in the

equation, then we will change the slope of the curve as well as the
intercept on the y-axis. However, the intercept on the x-axis will
remain the same; some examples are shown in Figure 2.5a. On the
other hand, if we hold the coefficients of the x- and y-values con-
stant and change the intercept value, then we will obtain a series of
parallel lines cutting both axes at different points, as in Figure 2.5b.
It follows from these two simple exercises that the slope of a curve
is governed by the ratio of the coefficient of the x-value to that of the
y-value, and the intercept on the x-axis by the ratio of the intercept
on the y-axis to the coefficient of the x-value.
If a general expression for the linear equation is
ax+ by= c
where a, band care numerical constants, then in general the slope of

the line is given by aj b. Also, if the intercept on they-axis is given by
c, then the intercept on the x-axis is given by cja. Thus, in the
examples studied earlier, where y is unity, the coefficient of the x-
value is the slope of the curve.
Returning to Figure 2.5a, it is apparent that if we draw the line
x = 5, the corresponding y-values for each of the equations plotted
can be read directly from the points of intersection. Thus, if
5x- y = 8.5
y = 16.5 (when x = 5)
Similarly, using Figure 2.5b and drawing the line y = 10, if
5x- y = 0
x= 2 (when y = 10)
y
20 (a)
2
10
-10
"r ,.1y
10
X
5 10
-10
Figure 2.5 Plot of families of functions related to the linear equation 5x- y = 8.5.
(a) Curves produced by altering the coefficient of they-value
I. 5x-y = 8.5 2. 5x- 3y = 8.5
3. 5x-5y=8.5 4. 5x-10y=8.5
(b) curves produced by altering the intercept or constant value.
I. 5x-y=O 2. 5x-y=8.5
3. 5x-y= 17
ALGEBRA 23
or, if
5x- y = 17
x = 5.4 (when y = 10)
These results are the same as those obtained by solving the corres-
ponding pairs of simultaneous equations. For example,
5x- y = 8.5
X =5
Multiply the lower equation by 5 and subtract:
5x- y = 8.5
- 5x =25
-y=-16.5
Similarly, the point of intersection of the curves in Figure 2.5a can

be found by solving any pair of equations simultaneously:
5x- y = 8.5
- 5x- lOy= 8.5
9y=O
thus: y = 0 and x = 8.5/5 = 1. 7. This demonstrates a link between

the points of intersection of curves on a graph and the solution of
simultaneous linear equations, which turns out to be useful in a
number of areas, not least of which is linear programming, which
will be considered in Section 2.1. 7.
2.1.2 Polynomial equations

Polynomial equations are equations of the general form
24 USEFUL MA TIIEMATICAL IDEAS
The most common forms of the polynomial are:
Quadratic
y = a:il + bx + c
which has one point of inflexion which is either a maximum or a

minimum (Fig. 2.6).
Cubic
y = ax? + b:il + ex + d
y
(a) 40
X
-5
y
(b) 80
X
- 5 5
Figure 2.6 Examples of the graphs of

quadratic equations.
(a)y= -6x2 + 10x+24;
(b) y = 9r- 18x- 27. -4 0
ALGEBRA 25
which can have two points of inflexion; a maximum and a minimum
(Fig. 2.7).
Quartic
y = ax4 + bx3 + ex? + dx + e
which can have three points of inflexion; either one maximum and
two minima, or two maxima and one minimum. The example
illustrated (Fig. 2.8) shows only one minimum over the range
plotted.
y
(a) 80
X
-5
-20
y
(b) 40
X
-5
Figure 2. 7 Examples of the graphs of

cubic equations.
(a) y = 2x' + 5x2 + 24x;
-80 (b) y = 3x3 - 9x2- 27x + 10.
y
2 50 0 0
-10000
Figure 2.8 Example of the graph of a quartic equation: y = O.OOlx" - 0.08x' +
0.5x2 - x + 15.
In applications of numerical methods in geomorphology, the use

of quadratic and cubic polynomials has been suggested for slope
analysis, the equations providing a means of describing the shape of
a profile (Doornkamp & King 1971, pp. l4lff.). The equations are
obtained by calculating a line of best fit using a least-squares
method. In geological simulations quadratic equations have also
been used combined with other functions, for beach profiles (Fox &
Davis 1971). In both of these applications the numerical value y (the
dependent variable) is the height of the profile above some suitable
datum level, at a point at a distance x (the independent variable)
from an origin measured along the datum.
2.1.3 The solution of a quadratic equation

We have already mentioned the quadratic equation in the previous
section, where the general form was given as
y= ax?+ bx+ c
Of particular significance, as we shall see later, are values of x when

ALGEBRA 27
y = 0, which are known as the roots of the equation. Thus, we need
to solve for x when
ax? + bx + c = 0
Extracting the factors is, in many situations, a relatively simple

matter. For example, find the roots of the equation
y = 2:x? - 3x + l
The coefficient of the :x?-term as a pair of factors is 2 and l; thus, the

first factors of x are
(2x ) (x )
The factors of the constant term are I:
(2x I) (x 1)
Now all that remains is to find the appropriate signs. If we return to

the constant term +I, we can obtain this either as +I x +I; or as
-I x -I. However, in order to make the x-term negative, then the
factors of the constant in this equation must also be negative, thus
(2x- 1) (x- 1) = 0
therefore x = 1 or x = l.
The rules for determining the sign of the factors of the constant
are:
(a) if both the x-term and the constant term are positive, then the
factors of the constant term will also be positive;
(b) if the x-term is negative and the constant term is positive, then
the factors of the constant will both be negative;
(c) if the x-term is positive and the constant term is negative, then
the factors of the constant term will be positive and negative.
It is wise to check the factors by multiplying out. Indeed, when the

signs are different it is essential in order to ensure that they are in the
correct order. The simplest way is to use the following:
2x-l=-x
X
x-l=-2x +
-3x
the arrows indicate multiplying and the two products should add up
to the middle term of the original expression, the sign will also match
if the factorization is correct.
Not all examples are so simple that they can be factored in this
way, and a formula which allows solution has to be used. If a and b
are the coefficients of the r-
and x-terms, respectively, and cis the
constant, then

- b -J(b 2 - 4ac)
X= _ _---..:.,o------'-
2a
For example, solve 3r- 7x- 6 = 0:
x = 3 or -2/3
Another example; solve r - 2x + 5 = 0:

2../(4-20)
X = ---'-----::2:-----'-
- 2 -J- 16- 2 4-J- 1

- 2 - 2
Since it is not possible to find the square root of a negative number,

we must write ..j - 1 as i, where i is an imaginary number (see
Section 2.1.6 for an explanation) then
x = (1 + 2z) or (1 - 2i)
Fortunately, in most situations where the solution of a quadratic

occurs in a geological example (see later, particularly Section 6.3) the
ALGEBRA 29
roots are real. However, we should remember the tests, using the
discriminant b2 - 4ac:
if~ - 4ac > 0 the roots are real and distinct;
if b2 - 4ac = 0 the roots are real and equal;
if b2 - 4ac < 0 the roots are complex.
We can interpret these rules as follows: if we graph the function, and

if there are two real roots, then the curve will cut the x-axis twice; if
there is one root, then there will be only one point at which the curve
cuts the x-axis; finally, if the roots are complex, then the curve will
not cut the x-axis at all.
It is also worthwhile to remember that there will be an axis of
symmetry for the curve, which can be found from
x = -bj2a
The problem of solving the roots of a quadratic is very common,

therefore a listing of a suitable program for their solution is given as
Program 2.2. In its present form the program can be run on its own,
removal of REM from statement 10 and changing statement 390 to
read 390 ENDPROC would allow it to be used as a procedure within
another program, provided a suitable call is made.
Table 2.4 Measurements of specimens of Gigantoproductus injlatus (Sarycheva),

from the Lower Carboniferous, Crook Bum Limestone, Crook Bum,
Northumberland. (a) Adult forms; (b) juvenile forms.
(a) Width Curved length (b) Width Curved length

(mm) (mm) (mm) (mm)
140 208 40 57
100 180 55 78
76 127 44 61
100 180 26 36
86 150 50 69
ll2 195 60 80
IIO 180 52 65
150 174 64 82
90 162 55 70
77 140
2.1.4 Other non-linear equations

Table 2.4 gives measurement data for a collection of the productid
brachiopod Gigantoproductus inflatus (Sarycheva), from the Lower
Carboniferous, Visean, Crook Bum Limestone, collected from a
locality in Northumberland. The data have been divided into two
sets; Table 2.4a refers to adult specimens, while Table 2.4b refers to
juveniles. These are plotted as a graph, Figure 2.9, and it is clear that
there are two distinct curves relating to the two groups. Also, the
distribution of the data points for each collection is honoured by a
200

Adults
1 50
E
E
I
f-
lJ
z
L.U
....J
100
..
L.U
>
a:
:::>
u
50 Juveniles
50 100 150
WIDTH mm
Figure 2.9 Plot of measurements given in Table 2.4, relating to the Carboniferous
brachiopod Gigantoproductus injlatus (Sarycheva). The curved length is measured
around the perimeter of the pedicle valve starting at the tip of the umbo, at right
angles to the hinge; the width is measured along the hinge line.
ALGEBRA 31
different form of curve. That for the juveniles is a straight line, while
that for the adults is a curve. The form of the adult data plot is
typical of this genus (Prentice 1956; Ferguson 1978) and illustrates
the principle of allometric growth, i.e. the relationship between the
size of one part of an organism and the size of another is not linear
but follows the general equation
(see, for example, Reeve & Huxley 1945, p. 121).

If we convert the data for the adult members of the population to
logarithms, as in Table 2.5, and replot as in Figure 2.1 0, the points
can be honoured by a straight line. If we estimate the slope and
intercept as discussed in Section 2.1.1, then we find that
log CL = 0.75log W + 0.72
This equation is the linear form of the allometric growth equation,

i.e.
log y = log B + a log x

Thus we can rewrite in the form of the allometric growth equation
by taking antilogarithms as appropriate. This gives an estimate of
the growth equation for the adult population
CL = 5.25 W0 75
Table 2.5 Data from Table 2.4a

expressed as logarithms.
log (Width) log (Curved

length)
2.15 2.32
2.00 2.26
1.88 2.10
2.00 2.26
1.94 2.18
2.05 2.29
2.04 2.26
2.00 2.24
1.95 2.21
1.85 2.15
2.5
....J
u
"" 2.25
0
2.0 2.2
log W
Figure 2.10 Plot of the logarithms of measurements of adult members of the
population of Gigantoproductus inflatus (Sarycheva) (data from Table 2.5).
Another extremely important equation in a geological context is

the exponential equation, which can be written in the general form
which describes growth
or as
which describes decay
Here, N, = number in population at time t, N 0 = number in popula-

tion at time t = 0, r = a constant, t = time, e = the base of natural
logarithms, numerically equal to 2.718282 (to six decimal places).
This equation has been used as a descriptor in such diverse topics as
population growth, the decay of radioactive isotopes (and hence is
used in calculations of geological time), the growth of crystals from a
melt and the rate of decay of a slope in stream development.
Another significant application is in the determination of rate laws
for chemical reactions from experimental data. In the case of the
important group of first-order reactions, the equation is
C =Co e -kt
or
InC= In C0 - kt
ALGEBRA 33
where C = concentration at a time t during the reaction, C0 = con-
centration at the start of the reaction, k = the rate constant and
t =time.
One method of finding k for reactions is to determine experimen-
tally the changes of concentration of a reactant with time and to
solve the problem graphically (Lasaga 1981, pp. 13ff.). Table 2.6
gives data from experimental cementation of aragonite ooids, show-
ing the rate at which strontium is released from the aragonite during
its conversion into calcite. A plot of this data, Figure 2.lla, shows
the characteristic exponential curve. Replotting the data using
logarithms, as in Figure 2.11 b, reduces the curve to two straight
lines, which show a break after about 20 days. This result is
interpreted as indicating a major change in reaction rate- a slowing
down - after the time in question. Based on other evidence, this is
thought to be the time of onset of cementation when rim cements
start forming around the ooids (Ferguson et a/. 1984). The rate
constants deduced from the graph are 1.017 and 0.5, indicating a
drop in rate of just over one-half.
Table 2.6 Data relating to the increase of strontium in sea water during the
experimental cementation of aragonite ooids at 200"C and I 500 lb in.-'
Time In (time) [Sr>+J In [Sr>+]

(days) (ppm)
0 7.7 2.04
7 1.95 30 3.4
12 2.49 93 4.53
15 2.71 125 4.83
19 2.95 153 5.03
27 3.30 156 5.05
51 3.93 230 5.44
63 4.14 310 5.74
78 4.36 300 5.70
2.1.5 Functions and function notation

So far we have used simple equations to express relationships
between variables, writing them out explicitly. However, as we saw
in Section 1.2, it is not always possible to be explicit, and we used the
notation
y=f( )
where the parentheses contained a list of variables. This is handy

300
Ia)
2 00
E
a.
a.
.---.
+
N
'-
(f)
'--'
100
2 4 b
In TIME days
Figure 2.11 Plot of experimental data showing the release of Sr2 from aragonite
during its conversion into calcite, plotted against time. (a) Plot of raw data; (b) plot of
logarithms of data. (Data as in Table 2.6.)
ALGEBRA 35
shorthand notation which is extensively used in mathematics, and it
is now convenient to explain its meaning and use.
The original meaning of the word 'function' in mathematics was
that some quantity, denoted by the symbol y, is related to another
quantity, denoted by x, by a simple formula or rule. A function is
written symbolically as
Y = f(x)
The idea of a function was closely related to the idea of a scientific

law: e.g. Boyle's Law, which states that at constant temperature the
pressure of a fixed mass of gas is inversely proportional to its
volume. Returning to an earlier example, the relationship between
the length of the pedicle valve and the width for the brachiopod
Epithyris oxonica could be expressed in this form, i.e.
Y = f(x)
where y =the length of the pedicle valve andf(x) = 1.15 x width.

As mathematics developed, the idea that the relationship was a
simple one was dropped. This followed because it was recognized
that some graphs which were originally thought not to represent
single functions, but rather to be combinations of different func-
tions, could be described by a single series, the Fourier Series which,
as will be discussed in Section 5.5, involves sines and cosines.
Further difficulties arose, particularly concerning the calculus of
variations where problems involving curves or surfaces are met, so it
is now accepted that x andy in the original definition, can stand for
any object whatever. In modern mathematics the term 'mapping' is
often used in place of 'function'. Their meaning is interchangeable.
A geological example of the definition of a function, which is very
similar to that used earlier in Chapter 1, can be taken from Sloss
(1962), where conceptual models for near-shore sedimentation are
considered. If S represents the gross geometry of a body of sedimen-
tary rocks, Q is the quantity of material supplied at the site of
deposition, R is the rate of subsidence at the site, D is the rate of
dispersal and M is the nature of the material supplied, then
S = f(Q,R,D,M)
In other words, the geometry of a sedimentary body is defined by a

function which relates the amount of material supplied, the rate of
subsidence, the rate of dispersal and the nature of the material. Also,
although the nature of the relationship, i.e. the function, is not
defined, then the concept has allowed the development of a simple
computer simulation model (Harbaugh & Bonham-Carter I970, pp.
373ff.).
2.1.6 Complex numbers

It was noted during the solution of some quadratic equations
(Section 2.1.3) that there were occasions when it was necessary to
consider the problem of finding a value for the square root of a
negative number. As an example we were asked to find the roots of
the quadratic
x2-2x+5=0
Using the equation it was found that the roots are given by:
2 4.J- I
2
and by writing .J - I as i, where i is an imaginary number, then the

roots are
x = (l + 2z) or (I - 2z)
A complex number can be defined as the sum of a real number and

the square root of a negative number.
To recap, the square of any real number, be it positive or negative,
is always positive, hence the square roots of negative numbers must
be a new type of number. These numbers are referred to as
imaginary numbers. Also, since a complex number can be defined as
the sum of a real number and the square root of a negative number,
and every negative number is the product of (- l) and the corres-
ponding positive number, it is sufficient to introduce a new type of
number for .J - I. This is the imaginary unit denoted by the symbol
i, hence the sum of a real and imaginary number is referred to as a
complex number. The solution to the quadratic equation, the roots
given above are, therefore, complex numbers. A complex number
whose imaginary part is equal to zero is a real number. As with other
numbers, there are rules of representation and manipulation for
complex numbers. These are not dealt with here, since most geologi-
cal problems have real solutions.
ALGEBRA 37
2 .1. 7 Linear programming
Linear programming is a method of finding the best solution to a
problem in maximization (or minimization), in the presence of
constraints. Thus, we might be asked to maximize the production
from an enterprise taking into account certain operational restric-
tions (constraints) or, in the opposite sense, we could be asked to
minimize production costs taking into account financial constraints.
This method of solution depends on expressing the problem in the
form of a linear function which is to be maximized (or minimized),
and combinations of the constraints in the form of linear inequali-
ties. The method is best understood by considering some simple
examples.
A mine produces two grades of ore, A and B. Because of
operational restrictions, the maximum production of ore grade A is
twice that of ore grade B, whose maximum daily output is 15 tonnes.
There are two smelting companies who buy ore, one is situated
10 km away and requires to be supplied with at least 10 tonnes of
grade B ore. The second is 25 km away and requires at least 16
tonnes of grade A ore. Unfortunately, the mine transport resources
are only 750 tonne-km day- 1 If the profit on grade A ore is
25 tonne- 1 and on grade Bore is 20 tonne- 1, obtain the tonnage
of each grade of ore per day which yields the greatest profit.
Let x = tonnes of grade A ore andy = tonnes of grade B ore. The
constraints on production are
16 <X< 30
10<y<l5
which relate requirements to production. The constraints on deliv-

ery are
25x + lOy< 750
which relate distance to the smelters to transport availability.

Finally,
25x + 20y= z
relating profit from each grade of ore to the overall profit from the
operation should be maximized. This is known as the objective
function.
In order to solve the problem the constraints are plotted on a

graph, as in Figure 2.12. The curves represent the maximum and
minimum of each grade of ore as well as the function relating to
distance and transport availability, which are
X= 16; X= 30
y= 10; y= 15
and
25x+ l0y=750
Notice that the inequality signs have now been replaced by equals
signs. Working through each, it should be noted that:
(a) x must be greater or equal to 16, therefore the area to the left of
x = 16 cannot be considered;
(b) x must be less or equal to 30, therefore the area to the right of
x = 30 cannot be considered;
(c) y must be greater or equal to 10, therefore the area below y = 10
cannot be considered;
y
X= 16 X =30
30
25x 10y" 750
20
1 6,1 5 24,15
y =15
0 ::::::::::.c
I :::::::::::.
10 i:------------,1'"""'6-,,1c-:~:-r-::::...:::..::..:.::..;::::...:::..::..:.:.:..:::=r=~:;;6-:;;,1o;-t Y =10
X
10 20 30
Figure 2.12 Graphical solution of first mine scheduling problem. The objective
function is maximized at x = 24 andy= 15.
ALGEBRA 39
(d) y must be less or equal to 15, therefore the area above y = 15
cannot be considered; and
(e) since the distance travelled cannot be greater than 750 tonne-km
day- I, then the area to the right of the line 25x + lOy= 750
cannot be considered.
Thus, we have defined the boundary of an area marked ABCD on

Figure 2.12, within which solutions are feasible. If we consider just
the corners of the quadrilateral and substitute the values for x any at
each into the objective function, we get
x,y 25x +20y = z
16, 15 700
16, 10 600
26, 10 850
24, 15 900
Thus, we find that the maximum profit of 900 is made if we supply

daily 24 tonnes of grade A ore and 15 tonnes of grade Bore. It can be
shown that a maximum (or a minimum) will occur at one of the
corners of the inscribed polygon defined by the constraints in a
linear-programming problem - the shaded area on the figure.
If we change this example slightly and, instead of having the
output of grade A ore controlled by the maximum output of grade B
ore, replace these constraints by saying that the maximum total
amount of ore produced of either type is 45 tonnes, then our
constraints are
16 <X IO<y
which relate to requirements, and
x+ y< 45
which relates to mining capacity, with the delivery constraints and

the objective function as before. The graph of their constraints is as
in Figure 2.13, and the area of feasible solutions is delineated as
before. In this situation the objective function is maximized at the
point C on the graph, where x = 20 and y = 25, which leads to a
daily profit of 1000 for the operation.
A graphical solution to a linear programming problem such as
y
X= 16
30
16,2 9
20
X
10 20 30 40
Figure 2.13 Graphical solution of second mine scheduling problem. The objective
function is maximized at x = 20 and y = 25.
those illustrated is quick and reasonably foolproof. However, if

there are more than two variables to consider, then, because of the
limitations of drawing graphs, the method is inappropriate. We have
shown that the possible solution to any two-dimensional linear
programming problem lies at one of the points of intersection of the
curves representing the constraints. It was shown in Section 2.1.1
that the point of intersection of two curves also represent the
solution of the two equations solved simultaneously, this leads us on
to an algebraic method of solution for these problems. We start with
an example.
A small company produces three types of fireplace, made out of
varying combinations of Cotswold limestone and Delabole slate.
Fireplace type A needs 10 cwt of Cotswold stone and 2 cwt of
Delabole slate, and sells for a profit of 30. Fireplace type B needs
3 cwt of Cotswold stone and 8 cwt of Delabole slate, and sells for a
profit of 50. The third type, C, needs 6 cwt of Cotswold stone and
4 cwt of Delabole slate and sells for a profit of 25. If the maximum
amount of Cotswold stone available is 200 cwt week- 1 and the
maximum amount of Delabole slate is 150 cwt week -I, how many of
each type of fireplace should be produced to maximize the profit,
assuming that all of the production can be sold. We can tabulate the
data as follows.
ALGEBRA 41
Cotswold stone Delabole slate Profit

(cwt) (cwt) ()
fireplace A 10 2 30
fireplace B 3 8 50
fireplace C 6 4 25
maximum rock
available
per week 200 150
Let us denote the number of fireplaces of types A, B, C by x~' x 2

and x 3, respectively, and the profit by z. Then the constraints are
The objective function is
which is to be maximized. First we need to get rid of the inequality

signs in our constraint equations. To do this we add a 'slack
variable' to each equation:
=200
= 150
We now have two linear equations in five unknowns. We cannot

solve these directly, since the problem is underdetermined, i.e. there
are more unknown variables than there are equations. However, we
can choose to set three variables to zero and solve for the remainder.
Set x 3, x 4 and x 5 = 0:
Solving these equations, we find
x 1 = 15.54, x 2 = 14.86, x 3 = 0
This is a basic feasible solution. Rounding-off (we cannot produce

fractions of fireplaces), so that x 1 = 16 and x 2 = 15, we can substitute
into the objective function, which gives a value of 1230 for the
profit.
We can also set x 2 , x 4 and x 5 = 0, in which case
x 1 = - 3.56, x 2 = 0, x 3 = 39.28
However, this is not a basic feasible solution, since a constraint is

that x 1 ~ 0 and we have found that x 1 = - 3.56.
We can also set x 1, x 4 and x 5 = 0, in which case
x 1 = 0, x2 = 2.78, x 3 = 31.94
which is a basic feasible solution. Substituting these values (after

rounding) into the objective function we find that the profit is 950.
We have now found that by producing 16 fireplaces of type A and
15 fireplaces of type B the company will maximize its profits, which
will be 1230 per week. As we noted earlier, the figures used in
maximizing the objective function have been rounded, since we
assume that we cannot produce fractions of fireplaces. Also, in this
example, since the problem was underdetermined, it is impossible to
calculate production figures for all three types of fireplace simulta-
neously. At an early stage we chose to set some variables to zero in
an apparently arbitrary fashion. This procedure is legitimate, and it
GEOMETRY AND TRIGONOMETRY 43
is possible to show that at each corner of the feasible space, only two
variables will have non-zero values.
Problems where the objective function is to be minimized are
solved in exactly the same way, except that the minimum value for
the objective function is sought. For further information on the use
of linear programming in a geological context, textbooks such as
Harbaugh & Bonham-Carter (1970, ch. 8) and Koch & Link (1971,
ch. 14.2) should be consulted. Both give a number of examples, and
references for further reading.
2.2 Geometry and trigonometry
2.2.1 Pythagoras' Theorem

Let us suppose that three samples of rock have been collected, in
each sample we have found two unique minerals which can be
labelled 1 and 2. By point-counting thin sections of the three rocks,
the percentage of the two minerals has been determined. On the
basis of this information, is it possible to express the closeness of the
relationship between the three rocks, numerically? A hypothetical
data set is listed in Table 2. 7.
Table 2~7 Hypothetical data relating to the

amounts of two unique mineral components in three
samples of rock.
Mineral I Mineral2
(%) (%)
rock A 10 10
rock B 30 15
rock C 25 45
If we plot these data on a graph, as in Figure 2.14, then it is

apparent that one method of quantifying the relationship between
the three rocks is to calculate the distances DA,B DA_cand Da.c using
Pythagoras' Theorem, which states:
The square of the hypotenuse of a right-angled triangle is equal to

the sum of the squares on the two adjacent sides.
From Figure 2.14 we can see that the distances (in geometrical
terms) between the minerals are given by:
so
#.
B
_j
4:: 2 s
a::
UJ
-z
:r:
A
2S so
MINERAL 2 %
Figure 2.14 Plot of the percentage of two minerals for three rock samples A, B and
C, showing their relationship using the Pythagorean distance coefficient (data as in
Table 2.7).
D A,B = J (Ax) + (Bx) 2 2
D A,C = J (Az) + (Cz) 2 2
Thus, for example, the distance between the two rocks A and Busing
this is
DA,B = J (lO- lW + (10- 30) 2
= ~425
= 20.62
Similarly,
DA.C = 38.08
Da.c = 30.4
An advantage of tackling the problem using Pythagoras' Theorem

is that if we increase the number of variables, then
where D A.B = the distance coefficient, A;= the ith measured variable
for A, B; = the ith measured variable for B and n = the number of
variables. Note that here, as elsewhere in this text, the conventional
symbol Lis used to indicate summation. The indices on it indicates
the range of values to be summed.
The measure D A.B has been termed the distance coefficient, and
has been used extensively in classification procedures. The smaller
the numerical value of the coefficient is, the closer the relationship
between the two samples under consideration (see Harbaugh &
Merriam 1968, pp. 163ff.). As we shall see in later sections of this
book, there are many other mathematical applications of this
important theorem.
2.2.2 Trigonometric ratios

In the right-angled triangle ABC Figure 2.15a the ratio of one side to
another does not depend on the size of the triangle, but on the size of
the angles. These are the 'trigonometric ratios'. They are:
sin 9 = AB/AC
cose = BC/AC
tane = AB/BC
When the triangle does not have a right angle the above ratios do
not hold, and neither does Pythagoras' Theorem. Thus, in the
triangle ABC shown in Figure 2.15b which does not have a right
angle, then to find the unknowns we use the sine rule:
sin A sin B sin C

-a-= -b- = -c-
A c
(a)
B [ 8 c.
A
Figure 2.15 The relationship between the length of the sides and the angles of a
triangle. (a) A right-angled triangle; (b) an obtuse-angled triangle.
This rule is normally used when one side and two angles are known.
If we are given two sides and an angle or three sides, then the cosine
rule should be used:
cr = b 2 +2 - 2bc cos A
The most useful applicativns of these ratios and rules in geology are
in solving problems related to geological mapping and section
construction. To illustrate their use in this important field, we will
consider three simple, but frequently occurring, problems.
(a) Find the true thickness of a stratigraphic unit given its apparent
thickness at outcrop, its dip and the angle of slope on which the
apparent thickness was measured. In Figure 2.16 CA is the
apparent (or outcrop) thickness, b is the true dip and a is the
slope of the surface. The true thickness AB is given by
AB= ACsin (a+ b)
(Barnes 1981, p. 74, fig. 6.1). If the surface is horizontal, then

a= 0, and
AB =AS sin b
(Ragan 1968, pp. 10-11, fig. 2.5).

(b) Find the depth D to the top of a stratigraphic unit from a point
on the surface at a given distance M from the outcrop, measured
at right angles to the strike and also given the true dip b. The
depth D is given by
D = Mtanb
(Ragan 1968, p. 12, fig. 2.1 0)
Figure 2.16 To find the true

thickness of a stratigraphic unit
from its apparent thickness and
the angle of dip and the slope of
the surface.
(c) To find the dip and strike using three points on a map. The strike
is determined by joining two points of equal elevation, where the
contour is cut by the top or bottom of the same unit, using
points as far apart as possible, to give maximum accuracy. The
true dip is found using any third point (again, as far away from
the others as is reasonably possible) and the equation
tanb = VjH
where b = true dip, V = difference in height between the line of

strike and the third point, and H =the horizontal or map
distance between the two. The strike can be measured directly
from the map (Ragan I 968, p. 19).
Many more examples of the use of the basic geometric and

trigonometric concepts can be found in the geological and structural
geology literature. A most useful and comprehensive survey of their
uses with large numbers of examples is given in Ragan (1968).
Since there are many occasions when geologists are required to
measure angles greater than 90 and then convert them to one of the
trigonometric functions, it is worthwhile reminding ourselves of the
range of values for the functions in the four quadrants of the circle.
These are summarized in Table 2.8. It should be remembered that
tables of sines, cosines and tangents are normally given only for the
first quadrant (0-90). Thus, if we require, for example, the value for
the tangent of I 10 we look up the tangent of (110- 9or, i.e. 20,
and reverse the sign. Similarly, the cosine of 290 is found by getting
the value for (360- 290)", i.e. 70. In this case the sign is not
reversed.
Table 2.8 The ranges of values of the three common trigonometric functions for the
four quadrants of the circle. Note that whereas the sine and cosine distributions are
continuous, that for the tangent is not.
Circular
Angle sin cos tan measure
(degrees) (radians)
0 to 90 0 to + 1 + 1 to 0 0 to+ oo 0 to 1t/2
90 to 180 + 1 to 0 0 to -1 - oo to 0 ~t/2 to 1t
180 to 270 0 to -1 -1 to 0 0 to+ oo 1t to 3~t/2
270 to 360 -1 to 0 0 to + 1 - oo to 0 37t/2 to 21t
48 USEFUL MATIIEMA TICAL IDEAS
2.2.3 Angular measure

In the examples so far discussed in this chapter, and particularly
those which are related to problems of geological mapping, the
angular measure used is degrees. This is known as sexagesimal
measure. Fractions of degrees are measured in minutes (I'= I/60
degree) and seconds (I"= I/60 minute). An alternative to sexagesi-
mal measure is circular or radian measure. The measurement of an
angle in radians is defined as the length of an arc cut off by the angle
on a unit circle (i.e. radius I unit), with the centre of the circle at the
vertex of the angle. The radius and arc are understood to be
measured on the same scale. Thus, we have
360" = 27t rad ~ 6.283I9 rad
I80" = 1t rad ~ 3.I4I59 rad
90" = 7t/2 rad ~ 1.57080 rad
or I rad =57" I?' 44.8"
Throughout the remainder of this text the angular unit will be used
only in cases where an angle can be directly measured, elsewhere it
should be understood that the measure is in radians. Students
should also note that in computer programming, functions involving
angles (i.e. sin, cos, tan, etc.) use radian measure, and therefore
angles given or calculated in sexagesimal measure should be con-
verted accordingly.
2.2.4 Trigonometric identities

There are a number of common trigonometric identities which are
used throughout this text and which occur in a number of mathema-
tical applications in the geological literature. Although it is not
intended to prove or use them at this stage, a list of the standard
identities follows.
(a) Basic identities:
(i) sin 2A + cos 2A = I
(ii) I + tan2 A = sec2 A
(iii) I + cos 2A = cosecl A
(b) Sum and difference formulae:
(i) sin (A + B) = sin A cos B + cos A sin B
(ii) sin (A - B) = sin A cos B- cos A sin B
(iii) cos (A+ B)= cos A cos B- sin A sin B
(iv) cos (A- B)= cos A cos B +sin A sin B
tan A+ tan B
(v) tan (A+ B)= 1 - tan A tan B
(v1.) tan (A - B) tan A - tan B

= ..,.-------,----=
1 +tan A tan B
(c) Double- and triple-angle formulae:
(1.) . 2A 2 . A
cos
A 2 tan A
Sin = Sin = 1 + tan 2A
(ii) sin 3A = 3 sin A = 4 sin3A

(iii) cos 2A = cos2A- sin2A = 1 - 2 sin 2A = 2 cos2A- 1
1 - tan 2A
1 + tan 2A
(iv) cos 3A = 4 cos3A - 3 cos A
2tanA
(v) tan2A = 1 -tan2A
( .) t 3A _ 3 tan A - tan 3A
VI an - 1 - 3 tan 2A
(d) Half-angle formulae:
(1.) . A _ 2 tan tA
sm - 1 + tan 2tA
(1.1.) A _ 1 - tan 2!A

cos - 1 + tan2tA
( ... ) A_ 2 tan tA
m tan - 1 - tan2tA
(e) Factor formulae:

(i) sin A + sin B = 2 sin t(A + B) cos t(A +B)
(ii) sin A - sin B = 2 cos t(A + B) sin t(A - B)
(iii) cos A + cos B = 2 cos t(A + B) cos t(A - B)
(iv) cos A - cos B = 2 sin t(A + B) sin t(B - A)
(v) tan A+ tan B= sin (A+ B)

cos A cos B
sin (B- A)
(vi) tan A- tan B= -=.----'-:------::-
smA cosB
Consideration of the angular relationship between shear planes in

rock provide a good, simple, geological example of the value of
trigonometric identities. For simplicity the two-dimensional case
can be considered; that is, the biaxial stress field as illustrated in
Figure 2.17b. In this situation there are two principal stresses, cr 1 and
cr 3, mutually at right angles to one another. From the biaxial stress
analysis, equations can be derived that give the normal stress cr and
the shear stress -r along any plane inclined at 9 to cr 1 Then
Using the trigonometric identities
sin 29 = 2 sin 9 cos 9
we can express the biaxial stress equations in terms of the angle 29,
which is the angle between the shear planes (Fig. 2.17a)
As will be demonstrated in Section 4.2.5, we can use this pair of

equations to calculate a theoretical value for the angle 29, which can
then be compared with the observed angle in rock.
(a) (b)
la1 a, '-. //
j,)-A
'/l;..O"
/
/
/
-----<> <>--- 0"3-----<> / 0"3
0"3 0"3
l
/
/ '
/
o-,
jcr,
Figure 2.17 (a) The biaxial stress field acting on a small block of rock, producing
shear failure. (b) The biaxial stress field showing the relationship between the normal
stress (cr) and the shear stress ('t).
PROBABILITY THEORY 51
2.3 Probability theory
In some of the chapters which follow, the concepts of probability

theory are used or alluded to in a number of important applications.
Therefore, although this is not intended as a statistical textbook, it is
necessary to include a brief introduction to the basic concepts, for
completeness. For more-detailed accounts of this topic the student is
referred to statistical works. However, the following summary,
based on Gray (1967), together with the examples to be found in this
section and in later chapters, should provide an introduction to the
basic concepts and its potential use in geology. We shall start with
the basic definitions.
There are situations in which, when an operation is performed on
a system, there are a number of different possible outcomes and
where a particular outcome is not under the control of the operator.
Probability theory should provide a method of defining how likely a
specified outcome is, as a result of the operation. Also, the theory
should be able to define how the probabilities that certain basic
events occur can be combined to give the probability of the
occurrence of complex events. In an ideal situation it should be
possible to arrive at a measure of probability from a description of
the system and trial; unfortunately, this ideal is only achieved in
special situations.
In general, the probability of an event p(E}, is measured on the
scale 0-1.0 (0-100%), i.e.
0 ~p(E) ~ 1
The measure p(E) can be regarded as the limiting value of the

relative frequency of E. This follows since it can be shown that, when
repeated independent trials can be made, the relative frequency of E
converges to a limit as the number of trials increases. Thus, if we
spin a coin the relative frequency of the occurrence of heads tends
towards 50 per cent as the number of trials increases. Impossible
events have zero relative frequency, and hence zero probability,
while events that are certain have unit relative frequency, and hence
unit probability. However, it should be remembered that if the
frequency of E is 0 (or 1), then it does not mean that E is impossible
(or certain). The following axioms, sometimes known as the 'laws of
chance', are not proven, but can be justified by the agreement
between practical trials and theoretical calculations using them.
2.3.1 The addition axiom

If 1 and 2 are two mutually exclusive events (i.e. 1 and 2 cannot
occur together), then the probability that 1 or 2 happens is the sum
of their individual probabilities:
In general,
In systems which are symmetrical, such as a coin or a six-sided die

all possible outcomes of a trial can be considered to be equally likely.
In this situation, ifF is the event which occurs, then the probability
of it occurring is given by
F = number of equally likely favourable outcomes

p( ) total number of equally likely possible outcomes
2.3.2 Multiplication axiom

If two events 1 and 2 are independent, such that the outcome of
one does not affect the outcome of the other, then the probability
that both 1 and 2 happen is the product of their individual
probabilities:
In general,
p(E1 and 2 and ... and En) = fJ p(E;)

i=l
Note the use here, as elsewhere in this book, of the symbol fl, which
is the conventional shorthand for multiplication. The indices indi-
cate the range of values to be multiplied together.
2.3.3 Complementary events

When an event 1 does not happen as the result of a trial, then the
complementary event 2 happens. Since 1 and 2 are mutually
exclusive,
Further, since 1 or 2 is certain with unit probability, it follows that
The result of throwing a pair of dice a large number of times (60),

plotted as a histogram, is shown in Figure 2.18. The theoretical
distribution is indicated by the dotted lines on the figure. These show
how the axioms operate when there is a combination of independent
events (i.e. two dice) and mutually exclusive events (i.e. the numbers
given by the face of a single die). Thus, the probability that a six will
result from one throw of a single die is given by
p(6) = l/6
20
15
>-
u
z
w
::> 10
aw
a:.
u..
2 4 6 7 8 9 10 11 12
TOTAL VALUE
Figure 2.18 Histograms showing the frequency distributions of values obtained
from a pair of dice. , Distribution obtained after throwing a pair of
dice 60 times; , the theoretical distribution for throwing a pair of dice 60
times.
whereas the probability of getting two sixes, throwing two dice is
p(6 and 6) = t t (the multiplication axiom)

= l/36
The result of getting two sixes gives a total of 12. On the other hand,
if we wish to calculate the probability of getting the total value five,
then there are four ways from which this may arise:
(a) die l can show 4 and die 2 can show l,

(b) die l can show l and die 2 can show 4,
(c) die l can show 2 and die 2 can show 3, and
(d) die I can show 3 and die 2 can show 2.
Since these are complementary events, and since the probability of

two dice giving any particular pair of values is I/36, then the
probability of getting the total value 5 is
= 5)
I + I + I + I (th e a dd.1t10n
= 36 . axiOm
. )
P(total 36 36 36
The theoretical probabilities of getting other values (in the range 2-

I2), can be calculated similarly. To convert these values to the
number expected in our experiment, we simply multiply by 60, the
number of trials made. The close agreement between this limited-
trials experiment and the theoretical results (Fig. 2.I8), gives weight
to the earlier statement that the axioms are justified.
2.3.4 Conditional probability

Another important concept in the area of probability which has had
extensive geological application is that of conditional probability
and, in particular, the use of the Markov chain or process. For
example, in numerical stratigraphy Markov chains have been used
in attempts to understand some of the factors which control the
order of deposition of sedimentary sequences. Fundamental to the
ideas behind the Markov chain is that the outcome of an event is
related to previous events. They can be relatively simple, in that only
the immediately preceding event has any influence; or may be
complex, where a number of previous events in different combi-
nations have influence. Examples of the use of Markov chains will be
given in Section 6.4.1, along with references which will allow the
interested reader to follow up the matter.
3 Sequences and series
3.1 Definitions
3.1.1 The idea of infinity

If n takes successively the positive integer values (1, 2, 3, etc.), then
there is no limit to the values which n can assume. It does not matter
how large a number we may think of, n will eventually exceed it. If n
increases in this fashion we can say that n tends to infinity, i.e.
n-+oo.
Also, if f(n) = rr, where x is a positive integer and n increases
without limit, f(n) also increases without limit and will exceed and
remain greater than any positive number which n takes, however
large. Here, as n tends to infinity, so does f(n), i.e. f(n)-+ oo as
n-+oo.
If, for example,J(n) = n + Ijn, thenf(n) will still approach infinity
as n-+ oo, provided that n takes on positive integer values as before,
On the other hand, if the function is q>(n) = Ifn- n, then the series
will increase without limit through the negative values as n
approaches infinity. Thus, we can write q>(n)-+- oo as n-+ oo.
3.1.2 Definitions relating to series

A sequence is a set of ordered terms with a rule for obtaining each
term. It can be written as
A series is formed when the terms of the sequence are added
the sequence u 1, u2, u3, . , u. +
gives rise to the series u1 + u2 + u3 + ... + u. +

sn is the sum of the first n terms of a series
set:) is the sum to infinity of a series
L~=a u, is the sum of all the terms u,, where r is an integer in the
range a to b, inclusive
DEFINITIONS AND THEOREMS 57
n~=a u, is the product of all terms u,, where r is an integer in the
range a to b inclusive
A series may be finite or infinite:
the series U1+ U 2 + U 3 + . . . + Un iS finite

the series u1 + u2 + u3 + . . . + un + is infinite
and its sum is
denoted either by 2:~ 1 u, or, when it would not be ambiguous, by
l:u,.
3.1.3 The idea of a limit

Taking, as an example, the function f(n) = l + 1/n, if n takes
successively the values l, 10, 100, 1000, 10 000, etc., then the values
for the function are
2, 1.1, 1.01, 1.001, 1.0001, ...
Thus, as the value of n increases, the value of the function gets closer
to 1.0, although it never quite reaches it. We can describe this by
saying that f(n) tends to l, as n tends to infinity, i.e. f(n)-+ I as
n-+ oo. Alternatively, the limit of f(n) is I as n tends to infinity,
symbolically
limf(n) =I
n-+oo
As we shall see in the following two chapters, one of the more

important applications of the limit is in calculus. Here some small
but finite increment, usually denoted by L\x, is made infinitely small,
i.e. in the limit L\x-+ 0. This then allows the analytical solution of
various problems related to rates of change (differential calculus,
Ch. 4) and areas and volumes (integral calculus, Ch. 5).
3.1.4 Convergence of a series

If we have a series whose sum to n terms is given by
where n tends to infinity (i.e. increases without limit) and, if the sum
58 SEQUENCES AND SERIES
of the terms of the series tends to a finite limit S as n approaches

infinity, then the series is said to be convergent and S is its sum to
infinity.
On the other hand, if the sum tends to infinity as n tends to
infinity, then the series is said to be divergent to + oo. Also, if the
sum tends to - oo as n tends to infinity, then the series is said to be
divergent to - oo.
If we consider the series
1-1+3-3+5-5+ ...
then the sum s2n = 0 and the sum s2n+ I = 2n + 1. Thus, when n--+ 00'
then S2n --+ 0 and S2n +I --+ 00. This SerieS OSCillates infinitely.
If u1 + u2 + u3 + ... is a convergent series, then un--+ 0 as n--+ oo,
and the value of any term is given by
Also,
lim un = lim sn - lim sn-1
n--+ oo n--+ oo n--+ oo
=0
This is a necessary condition for convergence, but does not guaran-

tee it absolutely. It does not work for example, with the harmonic
series.
3.1.5 The time series

In geology, so-called time series play an extremely significant role in
many areas. Such series are frequently true time series, i.e. data
recorded at set time intervals, or a series relative to some parameter
which may be related to time. An example of the first category is the
measurement of the release of sr+ as aragonite is converted to
calcite, already used in Section 2.1.4. In the second category we
might include data recorded at set intervals, throughout a strati-
graphic sequence. In this situation, since the measurement made is
relative to thickness and it is likely that there is some relationship
between thickness and time, the measurements are also implicitly
related to time.
The timespan of measurement will always be finite, whereas the
number of possible measurements can be either finite or infinite.
SOME COMMON SERIES 59
This follows since, however small the timespan may be, it is possible
to sample at increasingly smaller intervals. Thus, if tu is the sample
interval and n is the number of data points sampled, then
lim L\x-. 0 as n -+ oo
i.e. we have a continuous record.

Many time series of interest in geology oscillate at approximately
regular intervals and with differing amplitudes. The first example
mentioned does not fall into this category since, as we have already
seen, the release of strontium follows an exponential function. On
the other hand, stratigraphic sequences frequently show cyclicity,
and many parameters which can be measured for such a sequence
reflect this and give rise to oscillating data sequences of great
complexity. For the analysis of such time series it is more convenient
if the data are sampled at regular intervals, the size of the interval
and hence the number of points usually being controlled by either
the phenomena being measured or by economic or time consider-
ations. Fortunately, there are very few situations where equal
spacing is not possible. We shall consider one method of time series
analysis of oscillating data in Section 5.4.
3.2 Some common series
3.2.1 Arithmetic series

An arithmetic series is of the general form
a+ (a+ d) + (a+ 2d) + (a+ 3d) + ...
where a = the first term and d = the common difference - to find d

subtract any term from the next one. The series a, b, c forms an
arithmetic series if the result
b = t(a+ c)
holds, and b is called the arithmetic mean.

In general, the arithmetic mean is given by
3.2.2 Geometric series

A geometric series is of the general form
a + aR + aR 2 + aR 3 + ...
where a = the first term and R = the common ratio term - to find R
divide any term by the previous one. The series can be either
convergent or divergent. The nth term un is given by
and the sum of the first n terms by
The series x, y, z forms a geometric series if
y = .J(xz)
andy is called the geometric mean of x and z.
In general, the geometric mean is given by
3.2.3 The harmonic series

A harmonic series is of the general form
a a
a+ a+ ((n)- 1) +a+ ((n)- 1) +
where a= the first term and (n) =the position of a term in the series.
The series can be either convergent or divergent.
The harmonic mean is given by
n
nf L (lfx;)
i=l
3.2.4 The binomial series

The expansion of (a+ xY for all x and positive integer values of n is
given by
SOME COMMON SERIES 61
n(n -1) (n -2)d'-3 3 + + Xn

3! X . . .
The general (r + 1) term is given by
n(n-1) (n- 2) ... (n- r + 1) d'-ry;

r!
In some texts the series is denoted by:
)-(n-l) ... (n-r+l)

w here ( r - r.'
As n is a positive integer, the series has a finite number of terms.

We can demonstrate this by expanding simple equations:
(a+ x) 1 = la+ lx
(a+ x) 2 = lal + 2ax + lr
(a+ x) 3 = 1~ + 3alx + 3axl + lr
The pattern of coefficients of the expansion is most usefully summar-

ized in the form of Pascal's triangle:
n Pascal's triangle
1
2 1 1 2
3 3 1
3
4 4 6 4 1
5 5 10 10 5
6 etc.
Individual terms in the triangle are obtained by adding the two

terms on either side on the previous line:
e.g. n =I I I
v I
n=2 I 2 I
yy
n=3 3 3
with the outermost pair of coefficients being set to unity.
3.2.5 Exponential series

An exponential function is one where the variable is in the exponent,
e.g. d'. In the special case eX (called the exponential function), e is the
number such that the gradient y =eX at (0,1) is unity. Using the
binomial series we can expand f(x) =ex, where x is a rational
number:
this is the exponential series. If x is negative and rational, then
We can calculate the value of e to any required degree of accuracy by

substituting x = 1 in the series ex as above. This gives
l }2 p }4 }9
e = e1 = I + TI + 2! + 3! + 4! + + 9!
whence e = 2. 71828, to six significant figures. It also follows that

eo= 1.
The logarithmic series is based on the function eX, which can take
any value between 0 and + oo, and corresponding to any real
positive number y there is a number x such that
y = eX or x = In y
i.e. x is the natural logarithm of y. The series is

GEOLOGICAL APPLICATIONS 63
which will converge, provided x is in the range - 1 to 1.
3.3 Geological applications
3.3.1 The estimation of the mean for a given data set

In statistics there are a number of measures of central tendency
which can be used in the statistical description of measurement data.
The most commonly used is the mean. However, as we have seen
during our consideration of simple series, there are three means
which can be calculated, associated with the arithmetic, geometric
and harmonic series. In geological data-processing the arithmetic
mean (or average) is usually used. However, in some situations it
may be more appropriate to use one of the other means. As an
example, we can take the data given in Table 2.3 to demonstrate the
differences in the arithmetic and geometric mean, by using the data
to calculate values for the mean grain size of the sediment sample. If
we use the arithmetic mean and the data for the rom-scale, then we
find that the mean is 0.607 rom (equivalent to about 0.75 units on the
<p-scale), and using the <p-scale data the mean is 1.49 <p (equivalent to
0.355 rom). When the geometric mean is used with the rom-scale,
then the mean is found to be 0.355 rom, which is equivalent to 1.49 <p.
That is, the geometric mean of the rom-scale gives the same value in
<p units, as the arithmetic mean of the <p-scale. This is summarized in
Table 3.1.
Table 3.1 The relationship between the arithmetic and geometric means for the
grain-size data of Table 2.3.
Arithmetic mean Geometric mean
mm equivalent q~ mm equivalent q~
mm-scale 0.607 0.75 0.355 1.49
Arithmetic mean
q~ equivalent mm
q~-scale 1.49 0.355
From the above it is clear that we need some rule which will
enable us to decide which mean is appropriate in any particular
situation. Distinguishing between the arithmetic and geometric
means is relatively straightforward, since things which grow expo-
nentially (such as the mm-scale in our grain-size problem) require
the use of the geometric mean. In this context we should also notice
that the geometric mean of the lognormal distribution is equal to the
median of the distribution. Also, if the natural logarithmic mean
Jl 1 {y) (throughout this text the Greek symbol Jl is used to signify the
mean of a data set; unqualified this implies the arithmetic mean, a
subscript will signify the mean other than the arithmetic mean, but
in this case the subscript l indicates the logarithmic mean) of a set of
lognormally distributed data is given by
n
Jl 1 {y) = 1/n L In Y;
i=l
and the geometric mean Jl8 {y) is given by
where y = ln(x- ~) and ~ is the addition constant, and is equal to

zero for small samples. In this case, if x is the measured variable,
then
y=x
As an example, we can take data relating to the measurement ofsr-2+

on a number of brines from experiments on the early diagenesis of
carbonates already referred to earlier (Section 2.1.4). The data
quoted in ppm, comprises some 36 points, and is illustrated in
Figure 3.1. Although this is not a perfect lognormal distribution, the
calculated statistics are as follows:
arithmetic mean= 57.31 ppm

raw data geometric mean = 39.21 ppm
median = 38.50 ppm
In data arithmetic mean = 3.67
and
exp 3.67 = 39.25 ppm
10
>-
u
z
w
:::>
as
w
a::
u.
Figure 3.1 Frequency distribution
of the data used in the example
calculation relating the geometric
and arithmetic means. The data are
25 65 105 145 taken from experiments carried out
45 85 125 165 simulating the early diagenesis of
CLASS MID-POINT carbonate ooids, and represent the
[Sr 2 +] ppm concentrations of Sr' + in the brines.
For these data there is a close agreement between the value of the
geometric mean of the raw data and the exponential of the arith-
metic mean of the transformed set. The difference between the
median and the geometric mean probably reflects the nature of the
data distribution. For further discussion of these relationships see
Rendu (1981, pp. 7ff.).
The use of the harmonic mean is not so clear-cut in geological
studies. It may be that in some applications of stratigraphic data
analysis the harmonic mean would be more suitable than the
generally used arithmetic mean. One important application of the
harmonic mean is in the estimation of the tectonic strain ratio. Lisle
(1977) was able to show that the strain ratio estimated from the
mean shape of deformed elliptical markers varied according to the
mean used (Lisle 1977, p. 144, table 1), and the best estimate (when
the results were compared with another method of analysis) is given
by the harmonic mean. To facilitate calculation of the various
means, a computer program, Program 3.1, is listed.
3.3.2 The mean in geothermics

An important factor which controls the temperature gradient of the
Earth's crust at a particular point is th_e thermal conductivity of the
constituent minerals of the rocks within the crust, near to the point
of measurement. The thermal conductivity, denoted by K, is a
measure of the energy flow (heat transfer) per unit area and
the temperature gradient, in a one-dimensional heat conductor
(Buntebarth 1984, p. 9). Values of K for rocks and minerals at the

Earth's surface range from 1 to 6 W m -I K -I and is dependent on
both the type of rock or mineral and the crystal structure. The
thermal conductivity of a rock can be estimated using the thermal
conductivities of the individual mineral constituents and their rela-
tive occurrence in the rock under consideration.
Thus, a maximum value for thermal conductivity can be calcu-
lated from the weighted arithmetic mean
n
~ax= I
i~l
piKi
and a minimum value from the weighted harmonic mean
1/~in =I Pi Ki
i=l
where Ki =the thermal conductivity of the ith mineral species,

pi = the probability of the occurrence of the ith mineral species and
n = the number of minerals present. The calculation of the thermal
conductivity of a rock using these equations can best be illustrated
by reference to an actual example.
A quartz dolerite has the composition given in Table 3.2. Assum-
ing that the pyroxene is augite and the plagioclase labradorite, the
thermal conductivities (K) of the minerals are as listed. We are asked
to calculate the maximum and minimum thermal conductivities for
the rock. The values of K given are taken from Horai & Simmons
(1969).
Table 3.2 Mineral composition of a quartz dolerite.

The thermal conductivities (K), of the minerals are also
given.
Mineral Percentage K
meal/em sec oc
pyroxene 34 9.13
plagioclase 46 3.65
hornblende 4.5 6.07
ilmenite 8 5.25
quartz 5.5 18.37
calcite 2 8.58
~ax= (0.34 X 9.13) + (0.46 X 3.65) + (0.045 X 6.07) +
(0.08 X 5.25) + (0.055 X 18.37) + (0.02 X 8.58)
= 6.6583
1/~in =(0.34/9.13) + (0.46/3.65) + (0.045/6.07) + (0.08/5.25) +

(0.055/18.37) + (0.02/8.58)
= 0.1912
and Kmin = 5.2289.
Because K is also dependent on the crystal structure, some rocks

can be anisotropic with respect to thermal conductivity. In this
situation there are three independent components for K in the
directions of the three perpendicular co-ordinate directions x, y and
z. Conventionally, Kz is vertical, and Kx and KY are horizontal and
are assumed to be equal.
A value for ~ is given by the arithmetic mean:
a value for ~in is given by the harmonic mean:
while the geometric mean
which lies in the region of
3.3.3 Discrete distributions

(a) THE BINOMIAL DISTRIBUTION
Let us suppose that we are sorting a pile of material formed of pieces

of mineral and gangue, and that the constituents have been well
mixed; in other words, the distribution of the two types within the
pile is random. Also, because of a coating of mud on each piece, we
cannot see what we have chosen until it has been cleaned.
If p is associated with picking a piece of mineral and q is
associated with picking a piece of gangue, then because only two
events can occur and no others, from the basic concepts of prob-
ability theory,
p+q=1
Now, using the multiplication axiom of probability theory, if two

trials are made, then the probability that we will choose two pieces
of mineral is
p(p in 2 trials) = P2
and that we will choose two pieces of gangue is
p(q in 2 trials) = q2
The remaining alternative is that with our two trials we will pick a
piece of mineral and a piece of gangue. Now, from the addition
axiom, the probability that either p or q will happen twice is
Thus, the probability that we choose one of each in two trials is
This expression is not very convenient, particularly in complex cases.

Thus, an alternative way of approaching the problem is needed. We
can argue as follows:
either: mineral can be chosen in trial 1 and gangue in trial 2
or: gangue can be chosen in trial 1 and mineral in trial 2
applying the multiplication axiom, the probability of the first

alternative happening is pq, and of the second alternative happening
is qp. Combining these, and ignoring the order of events, the
probability of choosing a piece of mineral and a piece of gangue in
two trials is 2pq.
Having covered all of the possible outcomes from a two-trial
situation, the alternatives can be summarized as follows:
(a) mineral chosen twice: [l,
(b) either mineral or gangue chosen: 2pq,
(c) gangue chosen twice: i
which are the terms of the binomial expansion
(p + q)2 = [l + 2pq + q2
If, instead of picking two pieces of rock (two trials), we pick a
number of subsamples of n pieces (n trials), then by extending the
above argument it is possible to show that the probability of getting
0, 1, 2, 3, ... , n pieces of mineral in our subsamples will be given by
the successive terms of the expansion of
reading from left to right (note the order of p and q in this last
expression).
Returning to our example, we may wish to test the hypothesis that
the mineral and gangue are mixed in equal proportions. To test this
we take 50 subsamples of four pieces each, counting the number of
pieces of mineral in each subsample. The data can be tabulated as in
Table 3.3.
Table 3.3 Possible numbers of pieces of mineral to be found in a

subsample of size 4, and the corresponding number of subsamples
found to have those number of pieces of mineral, when 50 subsamples
were selected.
No. of possible pieces No. of subsamples

of mineral in subsample
0 2
I 11
2 22
3 12
4 3
The probability P of finding 0, I, 2, 3 or 4 pieces of mineral in a

subsample are given by successive terms of the expansion of
which are
numbers in brackets indicate the corresponding number of pieces of

mineral. Since our hypothesis was of equal mixing,
p=q=0.5
Substituting the values for p and q into each term of the expansion
and multiplying the resulting probability values by 50 (the number
of subsamples selected), we get theoretical probabilities, and the
equivalent expected number of subsamples as given in Table 3.4.
Table 3.4 Theoretical numbers of pieces of mineral in a subsample of size 4, for 50

subsamples, assuming equal mixing of mineral and gangue.
Probability of getting the number Expected no. of subsamples

of pieces indicated of mineral
P[O] = 0.0625 3
P[l] = 0.25 12
1'(2] = 0.375 19
1'(3] = 0.25 12
1'(4] = 0.0625 3
If we compare the results given in column 2 of Table 3.4 with the

original data in column 2 of Table 3.3, then it can be seen that there
is a close similarity, which would support our original contention
that we had a 50: 50 mix of mineral and gangue in our original
sample. It should be noted that in a practical application of this
method, a statistical test would be used to quantify the similarity
between the observed and expected data.
One common use of the binomial distribution is in the situation
where there is no a priori knowledge as to the values for the
probability of an occurrence or non-occurrence. In this situation the
required probabilities are estimated from the data. Let us suppose
that we have superimposed a grid over an area of interest, and count
the number of occurrences of a particular item or event in each grid
square. We can tabulate the data to show the number of possible
occurrences of the item in any grid square and the number of times
each possible occurrence is encountered.
In Figure 3.2 the possible occurrences of the item are: 0, 1, 2 and 3,
and there are

.
Figure 3.2 Contiguous grid sam-
.

. pling scheme for the binomial dis-
tribution. 4 x 3 layout correspond-
ing to the hypothetical example
discussed in the text.
3 grid squares with 0 occurrences

5 grid squares with 1 occurrence
1 grid square with 2 occurrences
3 grid squares with 3 occurrences
total grid squares counted= 12
total number of items observed= 16
We can now estimate the probability of an occurrence and the

probability of a non-occurrence from the data. The equations are:
mean number
of events per
no. of events observed
grid square
no. of grid squares counted
_ mean no. of events per grid square

P(oa;urrence) - maximum no. of events in any grid square
q(non-occurrence) = 1 - P(oa;urrence)
It is now possible to use the expansion of (q + py (where n =maxi-

mum number of events which occur in any grid square) to give the
theoretical probability of the occurrence of 0, 1, 2, 3, ... , n events,
calculated by substituting the values for p and q into the successive
terms of the expansion. To convert the theoretical probabilities to
numbers of occurrences, multiply them by the total number of grid
squares counted.
As an example, we can look at data relating to the distribution of
the brachiopod Epithyris oxonica on a bedding plane in the Blis-
worth Limestone (Jurassic, Bathonian), at a locality in Northamp-
tonshire. Using a 1-m square grid, the area of exposure was
subdivided into 21 squares, and the number of individuals occurring
in each square was counted; the data obtained are given in Table 3.5.
Table 3.5 Observed and calculated (assuming a
binomial distribution model) data relating to the
distribution of Epithyris oxonica on a bedding plane in
the Blisworth Limestone, Northamptonshire. Observed
data based on a 7 x 3 contiguous 1-m grid.
No. of occurrences Frequency
Observed Calculated
0 3
I 4 5
2 6 7
3 4 6
4 2 2
5 2 0
no. of grid squares counted = 2I

no. of individuals observed = 46
Mean no. per grid square = 46/2I = 2.I9
p = 2.I9/5 = 0.44
q = I - 0.44 = 0.56
Using the expansion (q + p) 5 and substituting q = 0.56 and p = 0.44

into the individual terms of the expansion, the probability of
occurrence of 0, I, 2, 3, 4 and 5, individuals in a grid square were
calculated. Multiplication of these probabilities by the total number
of individuals observed gives the expected number of grid squares
for each class. The values are tabulated, and inspection of Table 3.5
shows that there is a reasonably close agreement between the
observed and calculated distributions, which may be of some
palaeoecological significance.
The binomial series has been used extensively in statistics as a
probability model, and has proved extremely useful in describing
certain types of discrete distribution. For applications in geology the
reader is referred to Cheeney (1983, pp. 39-40) and Davis (1973, pp.
247ff.).
(b) POISSON DISTRIBUTION

In the previous section, the use of the binomial distribution as a
probability model depended on two important points. First, since n
tends to be a small number there is no difficulty in using the
expansion, since there will be only (n + 1) terms. Secondly, by
making the two constants q and p add to unity, the sum of the terms
of the expansion will also be unity. Let us now consider the
exponential series, which might also be used as a probability
distribution if it could be arranged for the successive terms to add to
1 and then to assign a meaning to each term.
Since an exponential function has the property
and if a = e, x 1 = z and x 2 = - z
then e-z x ez = e0 = I see Section 3.2.5. This relationship can also be

written as
or
This arrangement fulfils our first requirement that the successive

terms will add up to unity. Unfortunately, the series is infinitely long
but, as will become clear from the example, the values of the
successive terms quickly become very small. Therefore, we can say
that when the number of terms being used is greater than five, the
sum approximates to unity. All that remains to be done is to assign a
meaning to each term. This is achieved by finding the expectation of
the occurrence of an event in a particular situation and assigning this
value to z in the last of the above equations. Now, as in the example
of the binomial, the first term will record the probability of the
occurrence of the zeroth event and the following (n + 1) terms of the
expansion will give the probabilities of occurrence of the 1st to nth
event, as follows:
'7" -z
~[n]
n!
If we use the data in Table 3.6 as an example, the total number of

Scrobicularia counted in all of the samples is 42, and the total
Table 3.6 Data relating to a subfossil Scrobicularia community from Gibraltar

Point, nr Skegness, Lincolnshire, sampled using random 50-cm' quadrats, counting
number of Scrobicularia in each. P" calculated according to the Poisson distribution.
The expected number of quadrats is calculated by multiplying P" by the total
number of quadrats counted in the original sample.
No. of Scrobicularia No. of quads pexp Expected no. of

quads
0 67 0.646 62
1 22 0.282 27
2 4 0.062 6
3 1 8.99x I0- 3 1
4 9.82 X J0- 4 0
5 8.58 X JO-l 0
number of quadrats thrown was 96, so the value of z (the expec-

tation or average number of Scrobicularia per sample) can be
calculated:
z = 42/96 = 0.437
and
e-z = 2.7183- 0437 = 0.646
i.e.
e-z = ln- 1( - 0.437) = 0.646
Substituting the values of z and e-z, into each term of the expansion,
the expected probabilities can be calculated, and from these the
expected number of quadrats.
As can be seen from the results in Table 3.6, there is a close match
between the observed and expected frequencies of occurrence (as
with the binomial distribution, a statistical test of significance
should be made, to test the degree of similarity). Used in this way,
the Poisson distribution is a useful method which is often used to
describe events in a continuum, on the basis of a number of discrete
samples. It is normally recommended that this distribution be used
when the event is relatively rare, while the binomial distribution
should be used when the event is common. Again, the reader is
referred to the geological literature for further examples (see Davis
1973, pp. 305-7).
Before leaving the discussion of the use of probability distribution
models, based on simple mathematical series, it should be mentioned
that several others have been used by ecologists to describe relation-
ships within a community, particularly relating the number of
individuals of a species to the number of species present in the
community. One particularly important model is the so-called
'Fisher's a distribution', which is a special case of the Poisson
distribution where the zero term is not included. This distribution
may have important applications in palaeoecology, and is discussed
in Section 7.4. I, where an example of its use is also given.
4 Rates of change
differentiation
4.1 Functions: some further definitions
In Section 2.1.5 functions and function notation were defined.

Because of the importance the function and its associated notation
in calculus, it is necessary at this point to give some further
definitions.
4.1.1 Rationa/functions
A rational function is a function of the form
P(x)
Q(x)
where P(x) and Q(x) are polynomials in x. If the degree of

P(x) <the degree of Q(x}, then
P(x). f .
Q(x) IS a proper raction
If, on the other hand, P(x) ~ the degree of Q(x), then
P(x) . . fi .
Q(x) IS an Improper raction
4.1.2 Continuous functions

If the graph of a function turns out to be a continuous curve, then it
is reasonable to assume that the function is continuous. For example,
we can draw the graph of
f(x} = x+ 2
as in Figure 4.la and produce a continuous straight line, whereas if

we draw the graph of
FUNCTIONS: SOME FURTHER DEFINITIONS 77
y
y 60
6 (b)
(a)
40
4 20
-1.0 -0.5
X
0.5 1.0
-20
-40
~---L----~2----~--~4--~S x -60
Figure 4.1 Graphs of f(x). (a) The function is continuous, f(x) = x + 2; (b) the
function is discontinuous,f(x) = 1/x.
f(x) = 1/x
we find that it is continuous except at x = 0 (Fig. 4.1 b). In the latter

case we can say:
as x-+ +0, 1/x---+ + oo and as x-+ -0, 1/x---+- oo,
thus when x = 0, f(x) is not defined and the function is discon-

tinuous.
More formally:
(a) A function is said to be continuous for x =a if f(x) tends to a

limit as x- a, from either side and each of these limits is equal to
f(a).
(b) A functionf(x) is continuous throughout an interval of values
of x if it is continuous for all values of x in that interval. It is
continuous everywhere if it is continuous for all values of x.
(c) It can be shown that a polynomial P(x) is continuous for all
finite values of x and the rational function N(x)/D(x) is con-
tinuous except at values of x where D(x) = 0.
78 RATES OF CHANGE- DIFFERENTIATION
4.2 Differentiation
4.2.1 Definitions
(a) Differentiation is the process of finding the derivative of a
function. The derivative is called the derived function or the
differential coefficient. The derivative of y with respect to x is
usually written as
dy
dx
The derivative f(x) with respect to xis written as either
f'(x) or d[f(x)]
dx
(b) The gradient of a curve at any point is the gradient of the tangent
to the curve at that point, and is given by the derivative dyjdx at
that point.
In other words, the derivative of a function defines a rate of change

and, having found the derivative, it is relatively easy to find the rate
of change at any given point on the curve. A function which is
continuous is differentiable, whereas one which is discontinuous is
not.
4.2.2 Differentiation from first principles

If we start with some function y = f(x), then the gradient of the
chord PQ, joining P to a point Q, near to it on the curve as in Figure
4.2 is
grad PQ = oy
bx
In the limit as Q gets closer toP so that ox becomes infinitely small

and eventually disappears, then the gradient of the chord becomes
the gradient of the tangent to the curve at P, i.e.
gradient of the tangent at P =limit~~ as ox--+ 0
or the rate of change of y with respect to x at point P is

DIFFERENTIATION 79
Figure 4.2 Graph of any function

y = f(x). The slope of the chord PQ
is given by Oy/Ox. As ox-+ 0 then
the chord becomes the tangent to
the curve at the point P.
.. (8y)
Itmtt - dy
8x 8x=O- dx
To find dyjdx when y = x2

(l) In order to find the slope of the chord near any point on the
curve defined by some value of x, we need to add a small
increment 8x to the right-hand side of the equation and a
corresponding increment 8y to the left-hand side of the equa-
tion, thus we get
y + 8y = (x + 8x)2
(2) Since y = x 2 , we can eliminate y from the right-hand side of the

equation as follows:
i.e. we have subtracted x 2 from each side. This gives the

increment Sy in terms of x.
(3) Expanding the terms in the parentheses, we get
8y = .~ + 2x8x + (Jx? - x2
80 RATES OF CHANGE - DIFFERENTIATION
(4) Simplifying,
oy = 2x+ ox
Ox
(5) We have already noted that as Ox-+0, the chord becomes the
tangent at the point on the curve for the value of x, i.e.
dy =limit (oy) ox-+0

dx ox
which is
~=limit (2x + OX)sx=O = 2x
Thus,
dy = 2x when y = x2
dx
We can now consider the general case where the exponent can be any
value n. Thus, we find the derivative whenf(x) =X'.
To find dyfdx when y =X'

Proceeding as before, we get
y + oy = (x + oxY
Expanding the terms in the parentheses using the Binomial

Theorem, where
n(n- 1)
(X+ OX)n =X'+ nx!'- 10X + l X l X'- 20.x2 + ...
so that
Eliminating y from the right-hand side of the equation:

DIFFERENTIATION 81
= x"-ls: +n(n-l)x"_ 2 s: __2 +

n uX l x 2 ux- ...
But
.
dy= l'Imtt
-
dx
Therefore
dy = nx"- 1 when y = x"

dx '
We can now illustrate this procedure using a simple example. Figure

4.3 shows a plot of data obtained by measuring the length and
thickness of specimens of Homoeorhynchia acuta (J. Sowerby). The
best interpretation of the raw data is the curve as drawn, which
30
.
,., '
/
/
.
20
,
/
E / '
E
Vl
Vl 6.y
UJ
z ,
:><: , /
,,
10
u ,
I
f-
10 20 30
LENGTH OF PEDICLE VALVE mm

Figure 4.3 Graph of measurements of the Jurassic brachiopod Homoeorhynchia
acuta (J. Sowerby), showing two simple methods of estimating the rate of growth.
shows that throughout the life-cycle of the animal there are changes
in the rate of growth relating the two measurements concerned. If we
join the two ends of the curve as indicated by the broken line we can
estimate the overall or average rate of growth by finding the tangent
of the angle e, which is
e = 4o
tan e = 0.839
i.e. the average rate of growth of the thickness of the shell relative to
its length is 0.839.
If we wish to know the rate of change for the stage of growth
represented by any particular thickness, then we can estimate the
value by constructing a chord to the curve near the point by taking a
small increment on either side of it and finding the corresponding
increment on the other axis. We can then calculate the slope of the
chord using these values or, more directly, measure its slope and find
the tangent of the angle, using tables. This is also illustrated in
Figure 4.3, for the length measurement 20 mm. The results for
lengths 10, 20 and 30 mm are given in Table 4.1, which show that
during the life-cycle of the animal there is an increase in the rate of
change of length relative to thickness.
Even a casual inspection of Figure 4.3 shows the limitations of
this approach, particularly since the chord is only an approximation
to the tangent to the curve, leading to inaccuracies in the estimated
growth values. On the other hand, if an equation could be fitted to
the raw data in the form of
thickness= /(length)
Table 4.1 Data relating to the rate of change of growth in the brachiopod
Homoeorhynchia acuta. Estimated rate of change was found from Figure 4.3, while
the calculated rate of change was found as described in the text.
Estimated Calculated
Length (mm) rate of change rate of change
Angle Tangent
(degrees)
5 29 0.554 0.503
10 34 0.675 0.726
15 44 0.966 0.900
DIFFERENTIATION 83
and if the derivative is then found, then the exact value of the rate of
change for any given length could be calculated.
Using a statistical method, the constants a and Bof the allometric
growth equation (Section 2.1.4) were estimated from the raw data.
The equation is
thickness= 0.14length 153
Applying the rule for differentiation to this equation, we get
~i2 = (1.53 X 0.14)L(I5J-I)
= 0.2142 L 053
By putting the values 10, 20 and 30 mm for length into the equation,
we can calculate the exact rate of change. The results are given in
Table 4.1, and we can make a comparison between those found
analytically and the estimates arrived at using the chord method.
The differences in the values obtained by the two methods result
from difficulties involved in constructing a chord so that it is close to
the tangent, and then using the slope of that chord to estimate the
rate of change.
Simple functions can be differentiated from first principles, but
unfortunately the majority of functions normally met are too
complex for such treatment. To facilitate the differentiation of
complex functions, two methods are available: (a) the use of rules
and (b) the use of standard forms. The application of these two
methods is described, with simple examples, in Appendix 1. Readers
are reminded that reference should be made to mathematical texts
for a more detailed treatment. For example, the books by Hogben
(1967) or Graham eta/. (1984) provide good simple introductions to
the topic. More-advanced texts which are useful include those by
Gow (1960), Sherlock et a/. (1982) or Stephenson (1973). Also, since
familiarity with the techniques can only be achieved by practice,
examples which are to be found in these books should be worked
through.
4.2.3 Successive differentiation

If y is a function of x, then in general dy/dx is also a function of x.
The derivative of dy/dx, i.e.
d(dyjdx) d2
denoted by d.:;
dx
is achieved by successive differentiation. This is the second differen-

tial coefficient, which can also be denoted by: f"(x), the first-order
coefficient being written as f'(x) in this notation, and higher-order
derivatives follow the same pattern. During the procedure of suc-
cessive differentiation the rules and standard forms (Appendix I) are
used as required. The process is illustrated best by a simple example.
Find the third derivative of 4.x3 + 2r + x;
dy = l2x 2 + 4x + 1
dx
dJy = 24
dxJ
Thus, the third derivative of 4.x3 + 2r + x is equal to 24.

There are a number of important uses of successive differentia-
tion, one of which will be dealt with in some detail in Section 4.3.
4.2.4 Partial differentiation

A function frequently contains two or more independent variables,
each of which can be differentiated in succession, holding the others
constant. In other words, if y = f(a,b,c), then we can differentiate
a with respect to y, holding b and c constant;

b with respect to y, holding a and c constant; or
c with respect to y, holding a and b constant.
The symbol ois generally used to indicate partial differentiation.

Example y = .x2 + 2z4 - 3x + z2 - 4
Differentiate with respect to x, with z constant:
oy = 2x- 3
ox
DIFFERENTIATION 85
Differentiate with respect to z, with x constant:
Example w = ar + 2bxy + cy2

Differentiate with respect to x, withy constant:
ow
-= 2ax+ 2by
ox
Differentiate with respect toy, with x constant:
ow 2bx+ 2cy
oy
-=
No other rules are required, and partial differentiation is carried out

using the rules and standard forms as described in Appendix 1.
4.2.5 The orientation of shear failure planes in rock

Before leaving this part of the introduction to differentiation, it is
instructive to consider a geological example. For this we return to
the problem of shear-plane orientation which was considered in
Section 2.2.4. In that section we considered the derivation of two
biaxial-stress equations, which were in terms of the angle (29)
between the shear planes. The equations were
where cr =the normal stress, cr 1 =the stress in the y-direction,

cr 3 = the stress in the x-direction and T = the shear stress.
The Navier-Coulomb criteria for shear failure states that shear
failure occurs when the shear stress (T) along a plane, is equal to the
cohesive strength of the material (C0) plus the resistance to frictional
sliding on that plane ( = normal stress multiplied by the coefficient of
frictional sliding, 11). This was shown diagrammatically in Figure
2.17b. Thus, we have
86 RATES OF CJL<\NGE - DIFFERENTIATION
by definition, ).1 = tan a, where a is the angle of sliding friction and crn
is the normal stress. This equation can be rewritten in terms of the
angle of sliding friction and the normal stress:
-r = C0 +tan a [( -cr -+cr

1
2-
3)
- (cr- -cr
1
2-
)J cos29
3
The shear stress -r has also been defined in terms of the angle
between the shear planes, therefore we can write
(-cr -2-cr
1- 3)
sin 29 = C0 +tan a [ (cr- -+
1
2-
cr 3) J - [ ( -cr -crJ)
1
2-
Jcos29
Now, differentiating with respect to 9,
2 - 2- ( cri- crJ) cos 29 =(tan a) x 2

which simplifies to
cos 29 = tan a sin 29

or
sin 29
I =ta n a --
cos29
or
tan 29 = 1/tan a= cota
Therefore,
tan 29 =cot a= tan(90- a)

thus,
29 = 90- a
Now, for many rocks the angle of sliding friction, a, is approxi-

mately 30, so
9 = 45 - 15 = 3o
and 29 = 60
MAXIMA AND MINIMA 87
When the angle 29 is measured in rock (see Fig. 2.16a), it is usually
found to be close to 60. Thus, we have demonstrated analytically a
result which is in agreement with that observed in rock, showing the
power of mathematical analysis applied to this type of problem.
4.3 Maxima and minima
Figure 4.4 is a sketch of the curve of a cubic function. The curve has
two turning points at p and q, p being a maximum and q a minimum.
We wish to find, analytically, the values of x at these turning points,
and which is a maximum and which a minimum.
To find the position of the turning points p and q, we differentiate
and put
dy =0
dx
the solution of which gives the turning points required. In order to
find which is a maximum and which is a minimum, we need to
differentiate with respect to x a second time, i.e. find the second
differential coefficient of the original equation. In order to find
which is which, we substitute the values of x for the maxima and the
minima in this equation and use the rule:
at a maximum positionf"(x) has a positive value
at a minimum positionf"(x) has a negative value

Example Find the turning points and identify the maxima and
minima for the equation
q
Figure 4.4 Sketch of the curve generated by a
cubic function, showing the turning points.
3xl- 9r- 27x + 10
dy = 9r- I8x- 27
dx
Setting this to zero,
0 = (9x + 9) (x- 3)
x= -1 or 3
Substituting,
when x = - I d 2y/dr = - 36, so the function is at a maximum
when x = 3 d 2y/dr = 36, so the function is at a minimum

Substituting in the original equation, the corresponding values of y
are
y = 25 and -71
This is shown diagrammatically in Figure 4.5, which represents

graphs of the function and its first and second derivatives. Since
finding maximum and minimum values for a cubic function is a
relatively common problem, a suitable computer program is listed as
Program 4.1.
It should be remembered that a cubic equation need not have
turning points, and in this situation there will be a point of inflexion.
If a sketch of the curve is made this will be apparent, however if no
sketch is made and the procedure outlined is followed, then the roots
of the quadratic obtained by differentiation will be complex, i.e.
there are no real roots, also f"(x) will be zero. We can test whether
the curve is concave up or concave down as it approaches the point of
inflexion, by substituting values on either side of the point of
inflexion into the second differential. The rule is
y = f(x) is concave up at x = 0 whenf"(x) > 0
y = f(x) is concave down at x = 0 whenf"(x) < 0

y
10 0
-1 0 0
Figure 4.5 Sketch of the cubic function y = 3x3 - 9x2 - 27x + 10. The first and
second differentials are also shown. Note that the turning points of y = f(x) have
values of x equal to the values of the points where the curve y = f'(x) cuts the x-axis.
Also, when y = j(x) is negative the turning point of y = f(x) is a maximum, and
when y = j(x) is positive the turning point is a minimum.
The point of inflexion is given by
(2~) (-b)
where a= the coefficient of the .xl-term in f'(x) and b = the coeffi-

cient of the x-term inj'(x).
Thus, having calculated the value for the point of inflexion [a
value of x such thatf"(x) = 0], values of x either side are substituted
into f"(x). An example of the use of this will not be given, as the first
geological problem covers just this situation.
4.3.1 Geological applications

In a geological context maxima and minima have been used by
geomorphologists in the analysis of slope form, particularly in the
situation where estimates of the true position of valley floors are

required when the valley has a thick infill of alluvium (Doornkamp
& King 1971, pp. 148ff.). Two examples, applying the method to
simple geological problems will be given. The first of these relates to
the growth of the Carboniferous productid brachiopod Semiplanus
latissimus (J. Sowerby), and the second to finJing the thickness of
sediment infill in a valley.
Examples (a) Measurements of the hinge width and curved length of

the pedicle valve, of a number of individuals of the brachiopod S.
latissimus were plotted on graph paper, and best-fit lines were
eyeballed in. These curves were sigmoidal, indicating that the
underlying function was probably a cubic polynomial. The data
were then used to generate cubic equations using the method of least
squares (see Section 4.5). A plot of one data set, which is typical of
those obtained, is shown in Figure 4.6. The equation for this data set
was found to be
y = 0.00027xl- 0.0388x2 + 2.lx- 14.4
where x = the hinge width and y = the curved length of pedicle

valve, measurements in mm.
~ = 0.0008lx2- 0.0776x + 2.1

d2y
dx 2 =0.00162x- 0.0776
The solution of the quadratic is
X= 47.9 17.26i
Ignoring the complex part, we can substitute x = 47.9 into the

second differential, which gives
Since f"(x) = 0, we have a point of inflexion. Using the rule:
(i) setting x = 45.0 (just less than the point of inflexion),

f"(x) = (0.00162 x 45.0)- 0.0776
= -4.7 x w-J
90
60
E
E
I
.....
0
~
30
20 40 60
CURVED LENGTH mm
Figure 4.6 Graphical plot of the growth data measured from a single individual of
the productid brachiopod Semip/anus latissimus (J. Sowerby). The best-fit curve has
been eyeballed in.
(ii) setting x = 50.0 (just greater than the point of inflexion),

f"(x) = (0.00162 x 50)- 0.0776
= 3.4 x w- 3
Thus, when x = 45,/"(x) is negative and the curve is concave down,
and when x = 50, f"(x) is positive and the curve is concave up.
From this analysis we can conclude that for the specimens of S.
latissimus examined, shell increments are laid down in such a way
that the change in hinge width relative to the change in valve length
(of the pedicle valve), can best be described by a cubic equation. At
the point of growth where the hinge width is approximately 48 mm,
there is a change in growth pattern (corresponding to a point of
inflexion on the curve). Before this the hinge width grew faster than
the length, after which the opposite was the case. Consideration of
the shape and relative proportions of the shell of many productid

brachiopods shows that this is a reasonable interpretation, since
there seems to be a point after which an increase in the length of the
valves is not accompanied by a corresponding increase in width.
(b) The second example is concerned with trying to estimate the
depth of sediment infill in the valley of the River Greta in Kingsdale
Beck (near Ingleton, North Yorkshire), above the waterfall Thorn-
ton Force. About 50 m to the west of Thornton Force the old valley
is blocked by a plug of glacial till which appears to have caused, at
some point during the history of the river, the formation of a lake in
what is now called Kingsdale Beck. This has drained following the
cutting of a new course and deepening to its present level, leaving a
broad, flat floor to the valley. Using contour data from the Ord-
nance Survey (6 in. to I mile) map, a cross section was constructed as
illustrated in Figure 4.7, and the data were used to obtain a cubic
equation using least squares. This is
y = (- 2.6507 X J0- 5)r + (1.204 X J0- 2)_x2- 1.572X + 91.0079
where y = the height and x = the distance from the origin, both
measured in yards x 10- 1
It should be noted that the measurements were in feet (height) and
yards (distance), and both were converted to the scale indicated for
convenience of calculation.
~~ = ( -7.952} X J0- 5)_x2 + (2.408 X J0-2)x- 1.572
~~ = (1.59042 X J0- 4)X + (2.408 X }Q- 2)
w valley floor E
~ :::r~
=::-L-------::-'--:--'~---:-'~
600 1200 1800 2"100
DISTANCE yards
Figure 4.7 Cross section constructed from contour data across Kingsdale Beck, Nr
Ingleton, Yorkshire. Note the flat valley floor, which is probably formed from flat-
lying sediments, deposited at the bottom of a glacial lake.
The solution of the quadratic is
x = 95.2 or 207.6
Substituting these values into the second differential, we get, when

X= 95.2
~;, = 8.93 x to- 3
and, when x = 207.6
~= -8.93 x to- 3
Thus, we have:
when x = 207.58 the function is maximized at y = 46.4
and
when x = 95.23 the function is minimized at y = 27.6
For the present purpose, only the minimum value is of interest.

Converting our values back to their original units, x the distance in
yards from the origin of the plot (west edge of the section):
distance= 952.3 yd
andy the height in feet from Ordnance Datum (O.D.):
height = 828.1 ft
Since the measured level at the base of the valley from the map is
850ft, then the thickness of infill is 21.9 ft.
Using the data given by Wilson (1980), the base of the old river
channel at the northern end of the plug of glacial till is at approxima-
tely 775ft O.D. The level of the valley infill at Kingsdale Beck is at
850ft O.D., therefore assuming no change of level in the old valley
floor upstream, the maximum thickness of sediment is 75ft. If we
allow for changes in the valley upstream based on changes down-
stream, we estimate a change in level of 50 ft to give a base level for
the old valley at 825ft O.D. Thus, a minimum figure for the infill is
25 ft. These maximum and minimum figures should be compared
with the calculated value of 21.9 ft.
4.4 Optimization and the method of Lagrange multipliers
In Section 2.1.5 we considered the problem of optimization (either

maximization or minimization) of a linear function, subject to
certain linear inequality constraints. However, in many situations
the functions and constraints may not be linear, but in particular the
constraints can be equalities. The method to be described is designed
to cater for just this situation.
As an example, let us suppose that we manage two mines A and B
and that the total output we require is 100 000 tonnes week- 1 We
wish to minimize production costs for the two mines. What is the
optimum output for each mine when the function relating the cost of
the operation is
C= 5A 2 + 8B2 - 3AB
subject to (production)
A+ B = 100 000 tonnes
which can be written as
A= 100-B
(A and Bin tonnes x 103). One method of solution is by substitution.

The rewritten constraint for A can be substituted into the cost
function:
TC = 5(100- B) 2 + 8B2 - 3(100- B)B
=50 000- 1300B + l6B2
Taking the first derivative of the constraint equation and setting to

zero,
d(TC) = -1300 + 32B = 0

dB
B = 40.6 x l 03 tonnes
and the second derivative of the equation:

OPTIMIZATION AND LAGRANGE MULTIPLIERS 95
Since the value of the second derivative is positive, the calculated

value of 40.6 x 103 tonnes calculated for B is a minimum value.
Substitution of this value into the constraint equation gives a value
of 59.4 x 103 tonnes as the optimum output for mine A. Substituting
these values into the cost function we find that the weekly operating
cost is 23 593.76.
This method works well, provided the constraints are not too
numerous nor too complex. An alternative method which will work
equally well in complex cases is the method whereby artificial
unknowns, known as Lagrange multipliers are added. As many
dummy unknowns can be added as there are constraints.
Returning to the previous problem, we can rewrite our constraint
equation as
A+ B- 100 = 0
Multiplying this by the unknown constant A.,
/...(A+ B- 100) = 0
and then adding it to the cost function,
C~.. = 5A 2 + 8B2 - 3AB +/...(A+ B- 100)
Now, if the constraint is satisfied, then the last term in the Lagran-
gian cost function will be zero. Therefore, whatever values of A and
B minimize the cost C, the same values will minimize c~... However,
c~.. contains three unknowns: A, Band A.. We can solve the system by
partial differentiation and setting the results equal to zero:
oC1.. =lOA- 3B +A.= 0

aA
ac,
- " = 16B- 3A + A.= 0
aB
ac~.. = A + B - 1oo =o
at...
We now have three simultaneous linear equations, which on solu-

tion give
A = 59.4 x 103 tonnes
B = 40.6 x 103 tonnes

and
A.= -472.2
The values for production from the two mines are as we calculated
using the substitution method, but we also have an additional value
A.. The value of A. is significant, and represents the reduction in the
total operating cost if only 99 x 103 tonnes were produced instead of
100 x 103 tonnes per week. Thus, the 472.20 is known in economics
as the marginal cost of the last unit of production. It should be noted
that, having set ac,joh = 0, we automatically ensure that the con-
straint equation is embodied in the system. Additional constraints
are handled simply by multiplying each by a different Lagrange
multiplier, and including each in the system. For further information
on this method students are referred to economics texts such as
Glass (1980).
4.5 Differentiation and least-squares methods
4.5.1 Regression ofx against y

We have already seen (Section 2.l.l) that if x andy are related by a
linear law, then the equation can be expressed in the form
y=ax+b
In that section we showed that the unknowns a and b in this

equation could be estimated from a graph formed by plotting the
values for x and y, using a suitable scale and fitting a line by eye,
which best honoured the data. We will now consider an analytical
solution of the problem using the method of least squares. In this
method the sums of squares of the differences between the observed
and the calculated values will be minimized. Using the data in Table
2.1 a, relating to measurements of the brachiopod Epithyris oxonica,
we can work from first principles following the method outlined by
Moroney (1951, p. 280), as follows.
DIFFERENTIATION AND LEAST-SQUARES METHODS 97
Table 4.2 The predicted y-value and its deviation from the measured values,
assuming a linear relationship between x andy and that the x-values contain no error;
i.e. y is the dependent variable.
~easured width, Predicted width, ~easured length, Deviation

X y y
(mm) (mm) (mm) (mm)
16 16a+ b 18 16a + b- 18
17 17a+ b 19 17a+b-19
19 19a+ b 20 19a + b- 20
19 19a+b 21.5 19a+b-21.5
23.5 23.5a + b 26.5 23.5a + b- 26.5
25 25a+ b 29 25a+ b- 29
29 29a+ b 33.5 29a + b- 33.5
31 31a + b 36 3la+b-36
First, we set out the data showing how the pair of measurements
are related, using the equation for the straight line, as in Table 4.2.
The deviations can be squared as follows:
(16a + b- 18)2 = 256al- 576a + 32ab + b2 - 36b + 324
(17a + b- 19)2 = 289al- 646a + 34ab + b2 - 38b + 361
(19a + b- 20) 2 = 361al- 760a + 38ab + b2 - 40b + 400
(I9a + b- 21.5)2 = 361al- 817a + 38ab + b2 - 43b + 462.25

(23.5a + b- 26.5)2 = 552.25al- 1245.5a + 47ab + b2 - 53b + 702.25
(25a + b - 29)2 = 625al - 1450a + 50ab + b2 - 58b + 841

(29a + b- 33.5)2 = 841al- 1943a + 58ab + b2 - 67b + 1122.25
(3la + b- 36)2 = 96lal- 2232a + 62ab + b2 - 72b + 1296
which, on summation, gives
I:= 4246al- 9669.5a + 359ab + 8b2 - 407b + 5508.75
This equation can be written in the form of a quadratic, in terms of

either a orb, to give in terms of a
~= 4246.25a2 + (359b- 9669.5)a + (5508.75- 407b + 8b2) (4.1)
or, in terms of b,
~ = 8b2 + (359a- 407)b + (5508.75- 9669.5a + 4246.25a2) (4.2)
We have already seen in Section 4.3 that if we wish to make a value,

such as ~ in Eqns 4.1 and 4.2, a minimum then, in general, if
S= px2 + qx + c (4.3)
where S = ~ in Eqn 4.2, we need to set the first differential to zero,

i.e.
dS =O
da
Thus, differentiating Eqn 4.3 and setting to zero, we get
dS
-=2px+ q=O
da
which we can rewrite in terms of x, as
x= -qj2p (4.4)
In Eqn 4.1, and by analogy with Eqn 4.3, we find that the values for
p and q are
p = 4246.25 q = 359b- 9669.5
Substituting these values into Eqn 4.4,
(359b- 9669.5)
a= - 8492.5
which can be rewritten as
8492.5a + 359b- 9669.5 = 0
which is the partial differential il[jca of the sums of squares of the

deviations, holding b constant. Similarly, using Eqn 4.2 and the
procedure just outlined, we find that
16b + 359a - 407 = 0
which is the partial differential iJ'fjob of the sums of squares of the

deviations, holding a constant. Thus, we have a pair of equations
with unknowns a and b:
8492.5a + 359b = 9669.5
359.0a + 16b = 407.0
which, on solution, give
a= 1.229 b = -2.129
Thus, the equation is
length= 1.229 (width)- 2.129
However, we should remember that we made the assumption at the

beginning of the calculation that there was no error in the measure-
ment of the width (denoted by x) hence, by implication, length is the
dependent variable. This was a purely arbitrary decision, and we
could equally have assumed that the length measurement (denoted
by y) contained no error, in which case the width would be the
dependent variable. In this case the sums of squares would be:
~ = 5508.75al- 9669.5a + 407ab + 8b1 - 359b + 4246.25
which, on solution, gives
a= 0.809 b = 1.859
Thus, the equation is
width = 0.809 (length)+ 1.859
The first pair of values obtained in the calculations is commonly

termed the regression of x on y, while the second is the regression of y
on x. It should be noted that in most situations the dependent and
independent variables are known and are not interchangeable, as in

this example. Figure 4.8 shows the two regression lines, as well as the
estimated fit from the graphical solution of Section 2.1.1. This
allows a visual comparison of the results, and it can be seen that the
graphical solution runs between the two regression lines, and all
three cross at about the same point, the mean of the two measure-
ments.
Let us now look in more-general terms at the rather long-winded
calculation we have just completed. Each deviation squared, gave
the following terms related to the unknowns a and b in the linear
equation
y=ax+b
which are
ti -a ab b2 - b constant
which can be expressed in terms of the two measurements x andy, as
r -2xy 2x
The sums of squares in terms of x andy, summed from 1 toN, where

N is the number of pairs of measurements, were
:Er -:E2xy :E2x N -:E2y :Ey
which can be written as a quadratic in terms of the unknown a, as
40
E
E
I
f-
CJ 20
z
w
Figure 4.8 Graphical represen- ...J
tation of the data from Table 2.la
relating to the Jurassic brachio-
pod Epithyris oxonica (Buckman), /
showing the best-fit line (broken) 20 40

and the two calculated regression
lines. WIDTH mm
(Lr)al + [(I:2x)b- I:2xy]a + (N- I:2y + I:y)
and as a quadratic in terms of the unknown b, as
(NW + [(L2x)a- (L2y)]b + (I:r + 2x + I:y)
which lead to the two partial differentials
oi:foa = 0 = 2(I:r)a + (I:2x)b- I:2xy
and
oi:fob = 0 = 2(N)b + (I:2x)a- I:2y
These can be written as a pair of simultaneous linear equations
2{L.x2}a + (L2x)b = I:2xy
( E2x)a + 2(N)b = I:2y
that simplify to
(Lr)a + (Lx )b = I:xy
(I:x)a + (N)b = Ey
which can be solved for the unknowns a and b.

This method can be extended to enable higher-degree polynomials
to be fitted to suitable data. Thus, to fit a quadratic equation we can
expand our linear equations to give
c(N) + bi:x + ai:r = I:y

ci:x + bi:.XZ + ai:xl = I:xy
ci:x + bi:r + ai:x4 = I:ry
Note that the equations are written slightly differently from those
for the linear case; this is simply a matter of convenience, and the
pattern can be extended to enable higher-order equations to be fitted
as required.
4.5.2 Multiple regression: trend surface analysis

The technique of trend surface analysis, which has had extensive
application in geology (see, for example, Harbaugh & Merriam
1968, ch. 5; Davis 1973, pp. 322ff.), is a form of multiple-regression
analysis. The method produces surfaces in the form of contour
maps, which are described by integer series polynomials which
range, in most geological applications, from linear to sixth degree.
The method is claimed to provide an objective means of preparing
contour maps, and to facilitate the rapid comparison of a number of
variables for a given area. Having said this, it should be noted that
there are many drawbacks to the method. Some are due to the
nature and distribution of the types of data acquired by geologists,
but some arise from mathematical problems. Unfortunately, what-
ever the cause, it is possible to produce surfaces which are purely
mathematical artefacts and bear no resemblance to the actual
structure being mapped (see Section 7.3 for some discussion of the
problems, and particularly Section 7.3.4). Nevertheless, it is a 'quick
and dirty' method, which can be particularly useful if one of the
following is an objective of the study.
(a) To find the best-fit surface assuming that the differences between
the observed and calculated values are due to a random compo-
nent such as measurement error.
(b) The isolation of components within the data leading to the
separation of the regional and local trends.
In general, the surfaces fitted will obey the least-squares criteria for a
best fit, where
where z is the dependent variable - the geological property

measured.
Let us consider a simple case, a flat surface which exhibits simple
dip and strike. If x and y are the geographical co-ordinates and z is
the distance to the surface from some datum such as sea level, then
the surface can be described by the linear equation
z=A +Bx+ Cy
given the constants A, B and C, we can for any given point (x,y)
calculate
zcaic = A + Bx + Cy
and the deviation between zobs and zcalc is given by
deviation = zobs - A - Bx - Cy
If the sum of the squared deviations is given by F(A,B,C), then
F(A,B,C) = 1:(zobs- A - Bx- Cy) 2
If F(A,B,C) is to be minimized, then the partial differentials should

equal zero:
using the values of zobs the partial derivatives are
oF
oA = 1:2(z- A- Bx- Cy)( -1) = 0
-
oF
- = 1:2(z- A - Bx- Cy)(- x) = 0
oB
oF
oc = 1:2(z- A - Bx- Cy)(- y) = 0
-
which, on multiplying out and summing the individual terms, gives
An+ b1:x + c1:y = 1:z
A1:x + B1:r + c1:xy = 1:zx
A1:y + B1:xy + c1:yl = 1:xy
where n = the number of data points and the summation signs imply
summing from 1 to n. The solution of these equations is achieved by
rewriting them in matrix form as
J
1:x
1:y
1:xy
1:xy 1:yl
which can be simplified to
[A] [b) = [c)
and the unknown vector [b), found by
[Ar 1 [c) = [bJ
The constants for higher-order surfaces are obtained in a similar

fashion.
The use of matrix algebra in the solution of simultaneous linear
equations will be considered in Section 7.3.3, and an example of the
use of trend surface analysis and the calculation of the linear
regression coefficients will be given in Section 7.4.2.
5 Areas and volumes:
integration
5.1 Integration
Integration is the inverse of differentiation, the process of finding a

function when its derivative is given. Theoretically, we should be
able to retrace the steps of differentiation to obtain the integral. In
practice, this is not easy, as in many cases we cannot put the
derivative into the proper form. We saw in Section 4.2.1 that
differentiation is concerned with rates of change, integration or
integral calculus, on the other hand, is concerned with areas or
volumes. Thus, we can say that if a function f is positive and
continuous over x, when
then the area under the curve is
b
A= ff(x)dx = [F(x)]~ = F(b)- F(a)
a
where fis the sign indicating integration, f(x) is the integrand of

ff(x)dx, and b and a are the limits of integration. Alternatively,
n
A = lim
Llx-0
L1f(xk)Axk
k~
n-co
In words, the limit of the sums of the areas of inscribed rectangles,

when their width Ax approaches zero and their number becomes
infinite, will be equal to the area under the curvef(x) between x =a
and x =b. This is the fundamental theorem of integral calculus.
5.1.1 Numerical integration

The area AA'BB', under a curve y = f(x), as for example in Figure
5.1, can be found by dividing the interval AB into N equal parts,
106 AREAS AND VOLUMES: INTEGRATION
y
B'
Figure 5.1 The area under the

curve y = f(x), between the limits
A and B. A B
each of width Llx, and summing the area of theN strips (the smaller
Llx is, then the more accurate the answer will be).
b
~ Jf(x)dx
a
i.e. the sum of the areas of the rectangles approximates the area
ABCD, as illustrated in Figure 5.2.
For practical purposes we can replace the rectangles with trape-
zoids, as in Figure 5.3. By this method we calculate
P' P.' p'

2 ~.1. 4D
. 7.;.:::: ~r"'"<:
~ ""'y=flx)
Figure 5.2 The area under the

curve y = f(x), between the limits
A and B, approximated by rec-
tangles of width Llx and height p,
p',, p 2 p' 2, etc.
INTEGRATION 107
area 1 = !(y0 + Y1)L1x
area 2 = !(y 1 + y~.dx
Obviously, the smaller .dx is, then the greater the number of strips
and the more accurate the final answer. In general,
n-1
area= .dx{[HYo + Yn)] + L Y1}
i= l
One of the best-known methods of numerical integration is Simp-

son's Rule, where we do not have to integrate or draw a curve of the
function. Consider three points
on the curve y = ax2 + bx +c. Now, our three points give us
Yo = axl + bx + c
v ~--
/
~ y= f (x)
Yo Y1 Y2 Y3 Yt. Ys
Figure 5.3 The area under the curve
y =fix), approximated by trapezoids of
A B width L1x and ordinates y 0 , y,, etc.
If x 1 - x = h, then
Yo = a>! + bx + c
Y 1 = a(x + W+ b(x +h)+ c
and the area of the two strips is
A = hf3(y0 + 4y 1 + y:J
Thus, if an area under a curve is divided into an even number of

equal parts with distance h between any two successive divisions, the
total area will be approximately equal to the sum of the partial areas.
The area of the strips is
(n- 1) and n = h/3(yn_ 2 + 4Yn-l + Yn)

Summing these,
So far, we have labelled the first y-value as y 0 ; however, this is not

very convenient and we shall now return to the normal practice of
labelling our values I, ... , n, where n is an even number. The last
equation can now be written as
Thus,
b
L ~ ff(x)dx
a
provided that
h is small, given by (b- a)/n;

n is positive and even; and
b and a are the limits of integration.
INTEGRATION 109
2.5
Examples Evaluate Jdxjx using six strips
I
It is convenient to tabulate, as follows
X 1/x Weight Product

[weight x f(x))
1.0 1.0 1.0

1.25 0.8 4 3.2
1.5 0.667 2 1.334
1.75 0.571 4 2.284
2.0 0.5 2 1.0
2.25 0.444 4 1.766
2.5 0.4 0.4
I= 10.994
2.5
J
A= dxjx = (1/12)(I0.994)
I
= 0.916 sq. units

I
J
Evaluate In (4x + 5)dx using four strips
0
I- 0 I
h=--=-
4 4
X (4x + 5) In (4x + 5) Weight Product
0.0 5 1.61 1.61

0.25 6 1.79 4 7.16
0.50 7 1.95 2 3.9
0.75 8 2.08 4 8.32
1.0 9 2.20 I 2.20
)=23.19
I
A= Jln (4x + 5)dx = (1/12)(23.19)
0
= 1.9325 sq. units

Note that the weights or multipliers in the tables come from the
general equation for the sum, ie, the first and last strips have a
weight of 1 while odd numbered strips have a weight of 2 and even
numbered a weight of 4.
Program 5.1 is a computer program for evaluating Simpson's
Rule.
5.2 Analytical methods
In the introductory paragraph to this chapter it was mentioned that

integration could be considered as the reverse of differentiation. This
was followed by a discussion of the integral in terms of finding the
area under a curve, and in particular how to approximate this area
numerically. It is now necessary to discuss integration analytically.
In this context the topic divides naturally into two areas, the
indefinite integral and the definite integral.
5.2.1 The indefinite integral

Let us start with a simple equation
y =ax"
Now,
dy = nax"-1
dx
If we integrate by reversing the process,
_ na .x;<n-1)+ 1
y- (n- I)+ I
y =ax"
However, suppose that
y = 3x3 + 4
then
ANALYTICAL METHODS 111
Integrating,
Jdy = J(9r)dx
y 9 _ _x(2 +I}
= __
(2 + 1)
Using this approach the constant is lost, and we must take this into
account. It should be remembered that the constant must always be
provided for, so it is normal to write
y = 3x3 + constant
or, more usually,
y= 3x+ C
This is called the indefinite integral. If further information is given, it

is possible to evaluate the constant. For example, suppose we are
given that x = 2 when y = 40. Then
40 = 3(2)3 + c
C= 16
y = 3~ + 16
Thus, it is possible to show that two functions which have the same
derivative differ only by a constant. In general, there are three
methods of integrating:
(a) reversing the rules of differentiation;

(b) formulae; and
(c) reference to a table of standard integral forms.
The basic rules of integration and table standard forms can be

found in Appendix 2, along with simple examples illustrating their
use. Again, as with differentiation, practice is essential and the
reader is referred to mathematical texts for more examples and
details.
5.2.2 The definite integral

The alternative to the indefinite integral is the definite integral, where
the limits of integration are given. We have already discussed this
topic in the context of a numerical solution (Section 5.1.1), where the
problem of finding the area under the curve y = f(x) was outlined.
We can, of course, proceed analytically. Given y = f(x) we can
integrate to obtain
ff(x)dx = F(x) + C
Further, if the interval of integration is between a andb, then we can
find:
F(x)J: = F(b) = F(a)
which is the area between a and b provided a ~ x ~ b, i.e. it is a

proper integral. Since, if F(x) is continuous in the interval a~ x ~ b,
b
then f f(x)dx is also continuous.
a
F(x)J: means that the upper value of b is substituted for x to obtain
F(b) and the lower value of a is substituted for x to obtain F(a).
Example Given y = r, find the area under the curve from 0 ~ X ~ 4

4
fr dx = l.x3Jri
0
= 21.3 sq. units

4
Note that f r dx can also be written as
0
frdxJri
It will have been noticed that the constant of integration was not
used in this example. This follows since it automatically drops out in
the calculation. Thus, given
+ CJ:
b
ffix)dx = F(x)
a
ANALYTICAL METHODS 113
However,
F(x) + C]~ = [F(b) + C]- [F(a) + C]

= F(b)- F(a)
= f(x)J~
As a second example which will be worked through in slightly

more detail, we are asked to find the area under the curve
y = 4x + x2 when 4 ~ x ~ 8
4xl + .xl neg1ectmg

( 1) I ntegrate T . t h e constant
3
(2) Place brackets 2xl + .xl /3]!
(3) Substitute:
area= [2(64) + 512/3]- [2(16) + 64/3]

= 245.33 sq. units
It is now convenient to consider the problem in geometric terms.

First, the values of y for y = f(x) andy= Jf(x)dx can be calculated,
for values of x in the range 1 to 8 as in Table 5.1.
These data are plotted on a graph (Fig. 5.4), from which it can be
seen that the area under the curve y = 4x + x2 between the limits
x = 4 and x = 8 (shaded on the figure) is given by the difference in
the height of the ordinate for the curve 2x2 + tx3 at x = 8 (labelled A
on the figure) and the ordinate at x = 4 (labelled B). We can
generalize, and say that the area under any curve y = f(x) between
the limits A and B (A > B) is given by the height of the ordinate at A,
to the curve y = Jf(x)dx less the height of the ordinate at B. It should
Table 5.1 Values of the function y = 4x + x 2 and its integral.
X 2 3 4 5 6 7 8
f(x) 5 12 21 32 45 60 77 96
SJ(x)dx 1.33 10.66 27.0 53.33 91.66 144.0 212.33 298.66
y
300
200
y= Jt(x) dx A
.....
10'0
. .
.:::::
........
..........
................... ,
........................
B I:::::::::::::::::::::::::::
:::::::::::::::::::::::::::I
1 a a a a a a a a a a a a a a a a a a al
~ X
2 4 6 8
Figure 5.4 Geometric representation of the definite integral. The area outlined in
dots, under the curve y = f(x), is given by the difference in height of the ordinates to
the curve y = Jf(x)dx, labelled A and B.
be noted that if the curve cuts the x-axis between the limits, then the
areas above and below the axis should be calculated separately.
5.2.3 The mean value theorem

The mean value, c;, of y with respect to x, when y = f(x) over the
closed interval a ~ x ~ b is given by
VOLUME CALCULATIONS 115
1 b
=-Jydx
b-aa
The derivation of this is as follows. Let y 1, y 2, , Yn be values for the

function y = f(x) corresponding to the values x = a, x = a + Jx, ... ,
x =a+ (n- l)Jx, where nJx = b- a. We can calculate the mean
value for the function by
(y, + Y2 + + Yn) _ y,Jx + Y2Jx + + YnJx

n - nJx
However, since nJx = b- a, we can rewrite our expression as
If this expression tends to a limit as n-+ oo (or as Jx-+ 0), then the
limit is
For example, given y = r, find the mean value of yin the interval
O~x~4.
4
J
area = x 2 dx = tx3]~
0
= 21.3
~ = ( 4~ 0) (21.3) = 5.325
5.3 Volume calculations
In this chapter integration has been treated in the context of finding

the area under a curve, given limits in terms of values of one of the
unknowns (usually x). This use of integration can be extended to the
calculation of volumes, provided the third dimension can be defined.
The simplest form of volume calculation involves finding volumes
produced by rotating the curve around one or other of the axes.
More formally, the volume generated by rotating the curve

y = f(x) once around the x-axis, between the limits a and b is given
by
b
frcydx
a
Similarly, the volume generated by the curve x = g(y) around they-

axis, between the limits c and dis given by
d
frcrdy
c
This can be illustrated by an example where the volume of a solid

formed when the curve y = 3r is rotated once around the x-axis,
between the limits x = 4 and x = 5, is required. For simplicity, the
answer will be left in multiples of rc.
s
V= frcydx
4
= ssrc(3r) dx
4
2
s
= frc9x 4 dx
4
= rc9x5
5
J 5
= ~(28 125- 9216)
= -18 5909
-rc cu b"tc umts
.
In geology problems of volume estimation are more frequently

expressed in terms of a three-dimensional Cartesian co-ordinate
system. Typically such a situation might occur as a result of an
exploration programme, where an estimate of material in place is
VOLUME CALCULATIONS 117
needed. In this case it may be possible to define the surface
mathematically by the use of some technique such as trend surface
analysis (as discussed in Section 4.5.2), where the surface is described
by a polynomial of the form z = f(x,y), where x and y represent
geographical co-ordinates and z represents thickness. Here the
volume above some region R in the (x,y)-plane is given by double
integration:
JJ z dx dy = SJJ(x,y) dx dy
R R
In many instances it is simpler to find a numerical approximation

than to integrate the function. A numerical solution can be found by
dividing the area of interest into a number of conveniently sized cells
and calculating the value of z at the centre of each cell, by
substituting the co-ordinates into the function. The volume is then
found as the sum for all of the cells, of the product of cell area and
the thickness z. This technique can also be used where the data have
been hand-contoured or where no mathematical equation is avail-
able for the surface. In this situation the value of z is estimated at the
centre of each cell by a suitable method of interpolation, and the
volume is calculated by summation as before.
As an example, we can consider data relating to a thin bed of
material of economic interest. The thickness of this unit has been
measured at outcrop over an area of around 200 km 2, and from this
information a second-degree polynomial has been generated by the
method of trend surface analysis. The equation of the surface, in
terms of the thickness z (measured in em) and the geographical co-
ordinates x and y (kilometre-grid), is
z = 0.32 + O.l04x + o.0002y- o.0024r + 0.0013y
The problem now is to calculate the volume of material in place for

the area between the eastings 5 to 20 (x-values) and the northings 1
to 11 (y-values) inclusive. It is assumed that the base of the bed is
flat.
An exact analytical solution can be obtained by integration. First,
to simplify the integration the coefficients in the equation are
replaced by symbols
a= 0.32 P= 0.104 y = 0.0002
J =- 0.0024 A.=0.0013
which gives
z = a + Px + yy + r5x 2 + A.y2
Now the volume is given by
II 20
V= J J (a+ Px + yy + r>r + A.y)dxdy
y= I x=5
II
= J [ax+ Pr/2 + yxy + r5x /3 + A.lxU dy
y=l
3 0
II
= J [15a + {P/2)(375) + yy(15) + (r5/3)(7875) + A.y(15)]dy
y-1
= [15ay + (pf2)(375)y + (yy/2)(15) + (r5/3)(7875)y + (A.l/ 3)(15)Jll

= 150a + 3750P/2 + 1850y/2 + 78 750r5/3 + 19 950A.f3
= 150a + 1875P + 900y + 26 250r5 + 6650A.

Putting the coefficient values into this equation, we get
V= 48 + 195 + 0.18- 63 + 8.645
= 188.465
and, after converting scales, we find that the volume of material in
place is equal to 1 884 650m 3
The following results were obtained for cells of different sizes,
using a numerical approximation method. For this the thickness z
was found at the centre of each cell, and the cell volume calculated.
Summing the volumes for all of the cells gave the total material in
place:
cell side I 000 m, volume 2 208 800m 3
cell side 500 m, volume 2 045 632m 3
cell side I 00 m, volume 1 883 762 m3
In this example there is good agreement between the result

SUMMATION AND DISTRIBUTIONS 119
obtained for the analytical method and those obtained by numerical
approximation. This follows as a consequence of the simple surface
used to describe the data. Clearly, the more complex the surface is,
then the greater the differences between the estimates obtained using
different cell sizes. Also, in order to get a value comparable with that
obtained by integration, the cell size would need to be reduced
further.
5.4 Summation, integration and probability distributions
In Section 4.2.1 it was noted that only continuous functions can be

differentiated. Since integration can be considered as the inverse
of differentiation (Section 5.1 ), it follows that only continuous
functions can be integrated. We also noted that the area under the
curve between the specified limits can be found by one of two
methods:
(a) integration using the definite integral, leading to an exact

analytical solution or
(b) summation using a numerical method such as Simpson's Rule,
which leads to an approximation to the true value.
Thus, for a continuous function, the area under the curve can be
found by either integration or summation. Functions which are not
continuous, on the other hand, are not integrable and the summation
method is used.
In statistics there are a number of fundamental distributions
which are used to describe the probability of the occurrence of
events of different types. These distributions can be continuous or
discrete. The two which were introduced in Section 3.3.3, the
binomial and Poisson Distributions, belong to the latter group,
whereas the extremely important Normal Distribution, for example,
is a continuous function. Implicit in statistical theory is the relation-
ship between area and probability; thus, since the probabilities sum
to unity, the area under the curve or histogram described by the
distribution will also be unity. From the above discussion it is clear
that there will be differences in the way in which properties, such as
the mean or standard deviation for the two types of distribution, are
derived. To illustrate the differences let us consider the derivation of
the arithmetic mean. We define the mean as follows.
If xis a random variable, the mean of x, denoted by f.l.x is given by
(a)
for a discrete distribution with points x 1, x 2, x 3, , xj and fix) is

the probability that x = jth value also, L fix)= I.
allx
(b)
J xfix)dx
<Xl
f.l.x =
-oo
for continuous x, with a probability density functionfx(x) and

x2
Jfix)dx = 1
xl
Let us now consider the derivation of the mean for the uniform
distribution. This distribution, sometimes termed the random
distribution, can be either discrete or continuous. For the discrete
uniform distribution, the mean is given by
n .1 N+l
f.l.x= L J - = - -
j=l N 2
where x takes all integer values in the range 1 toN, inclusive. For
the continuous uniform distribution, the mean is given by
b 1
f.1.
X
= Jx--dx
a b- a
b 2 -li a+b
2(b- a) --2-
a and b are the interval throughout which the variable x is

distributed.
Although the results for the two distributions appear different,

and are arrived at by different paths, the observant student will have
noted that they have the same meaning. Although many other
examples showing the differences in treatment between the two types
of distribution in statistics could be given, we shall only emphasize
that whereas continuous data can be approximated by a discrete
FOURIER AND TIME SERIES 121
distribution, the converse is not the case. Thus, for example, the
binomial distribution is often used as an approximation to the
normal distribution, but never the other way round. For further
reading on this most important topic, see Cheeney (1983, ch. 5, pp.
58ff.) or any good statistical textbook.
5.5 Fourier series and the analysis of sequential data
In Section 3.1.5 we considered briefly the idea of the time series in

geology. It was noted that there were a number of possible types of
serially recorded data, some of which exhibit more- or less-compli-
cated oscillations, which could be either explicitly or implicitly
related to time, and could be sampled at regular or irregular
intervals. Because this information is complex, the underlying
trends, which may be related to some geological process of interest,
are not always immediately apparent. A large number of techniques
have therefore been developed to analyse or filter the data to
facilitate the recognition of these trends. In particular, data filtering
has become increasingly important in topics such as geophysics. It is
not intended to look in detail at these methods in this book.
However, since the use of the Fourier series is fundamental to many
of these techniques, the basic mathematics will be considered. To
simplify matters, we shall only consider the application of the series
to data which is sampled at regular intervals, also we shall treat the
topic numerically rather than analytically. The method discussed
follows closely that described by Harbaugh & Merriam (1968) and
Davis (1973). Stephenson (1973, ch. 15, pp. 254ff.) provides an
excellent, although mathematical, treatment of the topic, which is
useful for students wishing further information, not usually found in
geological texts.
In essence the Fourier series provides a means of breaking an
oscillating time series (or related series) down into a number of
simple harmonics which are represented numerically by a sequence
of sine and cosine terms, one of each representing individual
wavelengths, which in turn are related to the length of the series
sampled. The series can be progressively decomposed into har-
monics of progressively shorter wavelength, or can be built up by
combining harmonics to create a series which approximates the
original observed data. The importance of each harmonic in the
series can be judged by the numerical value of a parameter calcu-
lated from the sine and cosine terms, and known as the power
spectrum. As this value is related to the variance, it also provides a

means of comparing a number of similar time series statistically.
One general form of the infinite Fourier series can be written as
a "' [
z = 2o ."'5;. 1 a. cos (mr.x)
L . (mr.x)
+b. sm L J
The angular relationships are expressed in radians, and:
z = the dependent variable, i.e. the height of the measured

parameter on a graph,
a0 = the coefficient of the zeroth-degree cosine term,
a. = the coefficients of the cosine terms, n= 1, 2, ... , co

b.= the coefficients of the sine terms, n = 1, 2, ... , co
x =the points of equal interval along the horizontal axis (there

must be an even number) and
L = half of the fundamental sampling length, i.e. if we have k

points sampled, then L = kL1xj2, where Ax= x.+ 1 - x .
This series can be used to approximate certain types of mathema-

tical functions of the form
z = f(x)
In this situation the coefficients of the sine (b,) and cosine (a,) terms
can be approximated by
z z L
+_k + k-I zcos
2 o_
a = __ (nnLx,.) n = 0, 1, 2, .. . ,k/2
n k 2 i=l I
n =I, 2, 3, ... , k/2
where k =the maximum index for points at equal intervals (k should

be an even number and the number of points is k + I, as the index
goes from 0 to k) and i = the index of points: 0, 1, 2, ... , k. There is
no zero-degree sine term.
The power spectrum values related to each harmonic are calcu-
lated as
PS = ~(ti, + b~)
If we wish to retrace our steps we can do so, using the calculated
coefficients a. and b., and calculating the values of our original
series, at each sampling point, using
where z; = the observed value at the ith sample point, F(x;) = the
approximating function and N = the maximum degree of the series.
Now this is extremely useful, particularly should we wish to filter our
data, since we can use the equation to combine harmonics of our
choice. Thus, we can use only those which, from inspection of the
power spectrum values, are thought to be important, and can
construct a smoothed version of the original sequence. One of the
chief advantages of applying this method is that, unlike moving-
averages and related methods, points are not lost at the beginning
and end of the series. Before looking at a practical application, it is
convenient to start with some artificially constructed series, made up
of data of known amplitude and wavelength.
The series used were constructed using sine tables, the actual
angular value in the range 0-360 being used to represent distance in
metres and the sine value for each 5-m increment being used to
calculate the amplitude of the wave at the corresponding depth. In
its simplest form this gives a wavelength of 360m and, by adding
unity to each sine value, gives an amplitude of I. Changes in
wavelength were achieved by sampling the amplitude values already
generated at constant intervals according to the wavelength chosen.
For example, to generate a wavelength of 180m the series is sampled
at 10-m intervals. The amplitude, on the other hand, can be altered
by multiplying each value by a constant. In this way three series were
generated of 360-, 180- and 45-m wavelengths, the first of these
having an amplitude of 1m, the remainder an amplitude of 10m.
The series generated in this way are shown in Figures 5.5a and band
Figure 5.6b. By combining these fundamental series, two other
sequences were formed as illustrated in Figures 5.5c and 5.6c.
The numerical data relating to these series (using 19 points) were
analysed using the computer program given as Program 5.2, which
metres
7 0.0
52.5
35.0
17.5
10 20 20 30
AMPLITUDE
Figure 5.5 Artificially generated oscillating data series. (a) Wavelength 180m; (b)
wavelength 45 m; (c) combined 180 + 45 m. Amplitude constant 10m.
metres
(a)
52. 5
(b)
3s 0
1 7. 5
10 20 10 20 40
AMPLITUDE
Figure 5.6 Artificially generated oscillating data series. (a) Wavelength 73 m,
amplitude I m; (b) wavelength 36.5 m, amplitude 20m; (c) combined.
Table 5.2 Power spectrum values for two artificial series of wave-
length 360 and 180m, and the combined values. The series are
illustrated in Figure 5.5.
Power spectrum
Harmonic 360m 180m Combined
0 0 0 0
I 1.003 0 0.996
2 0 10.014 10.024
3 0.004 0 0.014
4 0 0.020 0.029
5 0.001 0 0.010
6 0 0.021 O.o28
7 0.001 0 0.011
8 0 0.01 0.018
9 0 0 0.010
has been written using the equations already discussed. The results
of the analysis of this data are summarized in Tables 5.2 and 5.3,
which give the power spectrum values related to each harmonic.
Remember that since we started with 19 data points, the series will
give rise to ( 19 - 1)/2 harmonics, plus the zeroth harmonic. Since the
zeroth harmonic has no sine term, it does not have a corresponding
power spectrum value.
Examination of the data in Table 5.2 shows that for the 360-m
wavelength the dominant harmonic is the first, that for the 180-m
Table 5.3 Power spectrum values for two artificial series of wave-
length 180 and 45 m, and combined values. The series are illustrated in
Figure 5.6.
Power spectrum
Harmonic 180m 45m Combined
0 0 0 0
I 0 0.044 0
2 10.014 O.o38 10.040
3 0 0.044 0
4 0.020 0.045 0.029
5 0 0.044 0
6 0.021 0.023 0.056
7 0 0.044 0
8 0.010 10.030 10.004
9 0 0.044 0
wavelength the dominant harmonic is the second, and that the value
of the power spectrum is equal to the amplitude of the series being
analysed. The result of combining the two series shows that, as
expected, the dominant harmonics are I and 2. Similarly, in Table
5.3 the dominant harmonic for the 45-m wavelength is 8, and in the
combined data the second and eighth harmonics dominate. Finally,
combining all three wavelengths gives the results shown in Table 5.4,
where the dominant harmonics are I, 2 and 8. In all of these
examples the values for the other harmonics have values less than
1.0 and result from noise introduced by the inaccuracy of calculating
the amplitude values, where only two significant figures were used.
Table 5.4 Power spectrum values for a com-
bined series of 360-, 180- and 45-m wavelengths.
Harmonic Power spectrum
0 0
I 0.908
2 10.137
3 0.111
4 0.129
5 0.111
6 0.131
7 0.110
8 10.100
9 0.111
Having looked at an example of the use of the Fourier series using

artificially generated data, we shall now use some experimental data.
These were generated during experiments designed to simulate the
early diagenesis of carbonate ooids (Ferguson et a/. 1984) by
constantly monitoring and recording temperature and pressure at
preset intervals. Data related to changes in sample thickness due to
compaction effects were also recorded by monitoring the behaviour
of a loaded piston which was in contact with the surface of the
charge being treated. Recorded data relating to the period of I h,
sampled at 2-min intervals for temperature, pressure and change in
piston height, have been analysed using the Fourier analysis pro-
gram. The results are given in Table 5.5, and the raw data are shown
graphically in Figure 5.7. Note that the values measured are relative,
not absolute.
There is a clear link between temperature and pressure in both the
figures and the data analysis results. The dominant wavelength of
Table 5.5 Power spectrum values of data sampled at 2-min intervals over a period
of I h, relating to temperature, pressure and the change in height (L1h) of a loaded
piston resting on the top of the charge in the cell. The original data are shown in
Figure 5.7.
Power spectrum
Wavelength
Harmonic Temperature Pressure L1h (min)
0 0 0 0
0.22 0.09 3.84 60
2 0.22 0.10 1.00 30
3 0.17 0.04 1.75 20
4 0.05 0.14 3.03 15
5 0.15 0.30 0.82 12
6 0.3 0.27 3.18 10
7 2.17 1.70 2.18 8.5
8 0.23 0.15 !.85 7.5
9 0.12 0.25 3.13 6.7
10 0.09 0.10 6.95 6
ll 0.13 0.11 2.06 5.45
12 0.11 0.09 2.19 5
13 0.17 0.26 5.40 4.6
14 0.40 0.10 4.96 4.25
15 0.05 0.22 4.38 4
60
E
UJ 30
I:
I-
5 1 0 15 90 100 110 120 130

Press. Height
ARBITRARY UNITS
Figure 5.7 Experimental data sampled at regular intervals over a period of 60 min.
See text and Table 5.5 for details.
8.5 min is controlled by the cutting in and out of the thermostat

which is used to keep the equipment at a constant temperature (in
this case 2ooq. For the change in height of the piston there are
many apparently significant harmonics, the largest of which (mea-
sured by the value of the power spectrum) is 6 min, with others at 4,
4.25 and 4.6 min. The 8.5-min harmonic related to temperature is
still present, but is not dominant as in the case of temperature and
pressure.
Although it is not immediately apparent why the piston oscillates
as it does, this example illustrates the use of this type of analysis in
simplifying large and complicated data sets. In the context of this
particular analysis, it is easy to see the effect of the cutting in and out
of the temperature control on other parameters, which will enable
filtering to be carried out to remove the effect, making the subse-
quent analysis of the data set much less difficult. As mentioned at the
beginning of this section, the theory of filters and its application is of
fundamental importance in the interpretation of many forms of
geophysical data.
For those interested in studying this field further, the classical text
is Blackman and Tukey (1959). However, as a further introduction
to the topic, at a more-advanced level than that given in this chapter,
the discussion of the topic by Camina & Janacek (1984) is recom-
mended.
6 Number arrays, the
algebra of matrices
6.1 Introduction and definitions
If we consider the written representation of a number, we can see

that it relies on the principle of position in one dimension only, the
straight-line power series. Arithmetic operations, on the other hand,
employ the same principle in a two-dimensional arrangement. So,
for example, if we were to formalize the operation of multiplication
in a two-dimensional set-up, it would be easy to demonstrate that
the position of individual numbers within the layout has a specific
meaning (Hogben 1967, p. 521). Such an arrangement is known as a
matrix, and is an economic way of representing complex operations
by assigning meaning to a number with reference to its cell position
within the grid.
A geological illustration of the meaning of the above statement
can be given using an example from stratigraphy. Figure 6.1 shows a
typical diagrammatic method of recording stratigraphic data in one
dimension in the form of a stratigraphic column. An alternative
representation would be as a matrix recording the relationships
within the data in terms of the number of transitions from one
lithology to another. If we use the main lithologies in the sequence,
then we can record the data as transition relationships in matrix
form, as in Table 6.1. The matrix is compiled by starting at the
bottom of the sequence and recording the number of times each
lithology passes into another, recording the transition from row to
column.
This has converted the conventional one-dimensional layout into
two dimensions, achieving a considerable economy of notation by
its use. Although this representation allows a quick appraisal of how
lithologies relate to one another, we have lost important information
in the form of the thickness of each recorded unit. This can be
rectified by choosing an appropriate sample interval and recording
transitions at each sample point for the chosen interval. The result of
sampling Figure 6.1 at 50-cm intervals is given in Table 6.2. The
significant differences between this and Table 6.1 are:
(a) we now have numbers in the principal diagonal which repre-
sent self-transitions, the actual numerical value representing
Shale marine
Limestone
Shale nan marine
Sandstone
Coal
Shale nan marine
Sandstone
Shale nan-marine
llllk;Shale marine
10m
Shale marine
Limestone
Shale nan-marine
Figure 6.1 Stratigraphic sequence 5

measured in the Lower Carboniferous,
Weardale, Co. Durham. Used to com- Sandstone
pile the transition data matrix given as
Tables 6.1 and 6.2. 0
Table 6.1 Data from Figure 6.1 recorded as a transition matrix.
Shale (m) Shale (nm) Sandstone Limestone Coal
shale (m) 0 2 0 0 0
shale (nm) 0 0 4 2 0
sandstone 5 0 0
limestone 2 0 0 0 0
coal 0 0 0 0
m, marine; nm, non-marine.

INTRODUCTION AND DEFINITIONS 131
Table 6.2 Data from Figure 6.1 sampled at 50-cm intervals.
Shale (m) Shale (nm) Sandstone Limestone Coal
shale (m) 3 2 0 0 0
shale (nm) 0 27 2 2 0
sandstone 3 36 0 0
limestone 2 0 0 2 0
coal 0 0 0 0
m, marine; nm, non-marine.
the relationship between the sample interval and bed thick-

ness; and
(b) there are differences in the transition numbers for shale (nm)/
sandstone and sandstone/shale (nm), again due to the rela-
tionship between bed thickness and sample interval.
From this, one of the principal difficulties in applying this method

can be seen; that is, the problem of how to choose the sample
interval. If the stratigraphic sequence to be sampled is made up of
units which are generally similar in thickness, then there is no
problem, but if, as in this example, there are wide differences in
thickness between units, then problems can arise. As can be seen, if
we reduce the sample interval so that it is small enough to include all
recordable transitions, then the values in the principal diagonal will
become extremely large. However, despite these limitations, matrix
notation of stratigraphic data has received considerable attention
from geologists, in their attempt to understand the underlying
processes which control the deposition of stratigraphic sequences. A
good introduction to this work, which includes the application of
aspects of probability theory to stratigraphy, is given by Harbaugh
& Bonham-Carter (1970, ch. 4).
Most algebraic and statistical methods have their equivalents in
matrix algebra, which allows for great economy in both manipula-
tion and representation of many problems. Also important is the
fact that there is a very high degree of compatibility between matrix
methods and the way in which computers can be most easily
programmed to perform arithmetic operations on large arrays of
data. Thus, stratigraphy is not the only field in geology where
knowledge of matrix algebra is useful. Indeed, one can say that since
the application of a wide variety of mathematical and statistical
techniques in geology, matrix algebra has become an essential tool
of the mathematical geologist. Before we consider specific applica-
132 NUMBER ARRAYS, THE ALGEBRA OF MATRICES
tions, we need to establish the accepted conventions of notation and

the basic rules of the algebra.
6.1.1 Scalars, vectors and matrices

In matrix algebra a scalar is a single number and is normally
represented by a lower-case Greek letter, e.g. A. or a. A matrix is a
rectangular array of numbers or symbols which represent numbers,
which are known as its elements. Individual elements can normally
have any finite numerical value, and can be negative, positive or
zero. A matrix can have any number of rows and columns and is
usually denoted by an upper-case letter in bold type, e.g. M. The
order of a matrix is its size expressed in terms of the number of rows
and columns: thus, a matrix of order n x m has n rows and m
columns, in the case of a square matrix its order is simply the number
of rows (or columns). A matrix with only one row is known as a row
vector, and one with only one column is a column vector. It should be
noted that a vector has magnitude and direction. If we have a vector
with elements x andy, then, by Pythagoras' Theorem, the magni-
tude is given by
and its direction by
A vector is represented by a lower-case letter in bold type, e.g. a.

Matrices are commonly square; other forms do exist, but because of
the rules of the algebra they are not as easy to manipulate.
In general, an operation performed on a matrix will involve all of
the elements of the matrix. However, to identify individual elements,
subscripts are used. If A is a 3 x 3 matrix, then the elements are
given as:
A
col 1 col2 col3
row 1 ai.I al.2 ai.J
row 2 a2.I a2,2 a2.3
row 3 aJ.I a3,2 aJ.J

BASIC MATRIX OPERATIONS 133
Table 6.3 Conventional symbols used in matrix notation.
Symbol Meaning
A any matrix
a any vector (row or column)
A. any scalar
an element of matrix A: the subscripts i and j give row and column
position, respectively
the identity matrix
the transpose of matrix X
the determinant of matrix M
the inverse of matrix A
If X is a matrix of size n x n, then each element can be denoted by

X;J where i and}= I, ... , nand i and} denote the row and column
position, respectively. Elements whose subscripts are equal form the
principal diagonal of the matrix. One important square matrix is the
identity or unit matrix, denoted by I, whose elements are zero except
for those of the principal diagonal, which are unity. As we shall see
later, this matrix has special properties and uses.
The transpose of a matrix is the matrix written with its rows as
columns. The transpose of matrix A is normally denoted by A' (or
A"f). A square matrix can be reduced to a simple form by calculating
its determinant, a single number denoted by 1 A J. Finally, the inverse
of a matrix is denoted by A- 1 and, as will be shown later, a matrix
whose determinant is zero has no inverse. The conventional matrix
notation is summarized in Table 6.3.
6.2 Basic matrix operations
6.2.1 Addition and subtraction

Provided two matrices are of the same order, they can be added or
subtracted simply by performing the required operation using each
corresponding element
A + B c
[
5 11 +
3 2j
134 NUMBER ARRAYS, TilE ALGEBRA OF MATRICES
6.2.2 Transposition
Writing the rows of a matrix as columns and columns as rows:
M M'
[-: :] [~
- l
-3 3
Row and column vectors can also be transposed.
6.2.3 Multiplication
(a) BY A SCALAR
To multiply a matrix by a scalar, each element is multiplied by the
scalar. For example, we might need to convert a matrix of measure-
ments in one scale to another, in which case we simply multiply each
element of the matrix by the required constant:
A . A. B
. 6
121
l8J
(b) BY A VECfOR
If we consider the algebraic equations
which can be written in matrix form as
y=Ax
this simply means that the vector of y-values, is equal to the matrix
of the coefficients of the x-values (denoted by a 1, etc.), multiplied by
the vector of the x-values. Thus, ifwe are given values for a 1, , a4
and x 1 and x 2, we can calculate y 1 and y 2 simply by following the
rules of number algebra. Translated into matrix form, this is the
multiplication of the first row of the matrix by the values of the
vector x and summation to give the first element of the unknown
vector y, followed by multiplication of the second row of the matrix
by the vector of x and summation to give the second element of the
vector y. This is the row-by-eolumn rule for matrix multiplication:
which is
In geometric terms we can consider the vector x as representing

the co-ordinates of a point relative to a pair of axes, then the column
vector y is the co-ordinates of the points after multiplication. The
matrix A represents an operation which transforms the point P 1 to
some new position P 2 , as in Figure 6.2a.
Several common transformations are recognized.
(a)
A reflection of the plane in the lint: through the

origin at 45 to the x-axis, see Figure 6.2b.
(b)
An elongation or compression parallel to the x-

axis, see Figure 6.2c.
(c)
An elongation or compression parallel to the y-

axis.
(d)
A shear parallel to the y-axis.

(b) (c)
/ 7
[: :] /'- ---y
2.4/~
0 7 0 10
7
[0-11 56
-6\,5 -----1--~----------------
2 .47/~:~
_____ -, (d)
''
' ', .' ,'
-u -- --- - - - - - - - - - - - - - - - - - -~.
-6 0 6
12 r(1-:_:-:: ~ :] (e)
o- '
32
Figure 6.2 Graphical representation of some simple transformations.
(e)
A shear parallel to the x-axis, see Figure 6.2d.
A change in sign in any of these gives a reflection in the appropriate

axis, as for example illustrated in Figure 6.2e. Figure 6.3 shows two
examples of the transformation of the unit square, showing shear
and reflection. It should be noted that these ideas can be extended
into n dimensions, where n is the order of the matrix of transforma-
tion, and hence the number of rows in the vector. For most practical
y
6
[~ -~]
d1
Da,
'1
2
1>,
X
-2 2 4 6
Figure 6.3 Transformation of the unit square abed. Showing a reflection in the y-
axis, a,b,c 1d, and shear, a,b2c2tf,.
geological purposes three dimensions are sufficient, and higher-

order matrices appear mainly in statistics.
The row-by-column rule for multiplication of a matrix by a vector
can be extended to any size matrix, provided the vector has the same
number of rows as the matrix has columns.
Find the product Xy = z, where
Then
X=
[-: :l andy=
[:]
[- 2 X 2) + (2 X 4)
z= (Q
(1
X
X
2) + (4
2) + (3
X
X
4)
4)
=
[::]
(c) BY ANOTHER MATRIX
This operation also follows the row-by-column rule as above, where
a row of the first matrix is multiplied by a column of the second

matrix and each resulting product summed.
Find the product AB = C, where
andB= [41 26]

Then
C= [(3
(6
X
X
1) + (4
I)+ (8
X
X
4)
4)
(3
(6
X
X
2) + (4
2) + (8
X
X
6)]
6)
[3819 6030]
In general, we can say that a matrix of order p x q can be multiplied
by a matrix of order r x s only if q = r, and the resulting matrix will
be of order p x s.
In matrix algebra the product AB does not equal the product BA,
if either of these products exist- i.e. multiplication is not commuta-
tive as in number algebra. For this reason post-multiplication and
pre-multiplication are distinguished. Thus, for the product AB, A is
post-multiplied by B, and for the product BA, A is pre-multiplied by
B. A computer program is listed as Program 6.1, which will multiply
a matrix by a scalar, a vector or another matrix (within the rules of
the algebra). The general case is given by
A matrix remains unchanged if it is multiplied by the identity matrix.

This operation is equivalent to multiplying by unity in number
algebra, thus
AI=A
Also, the following result holds:
AI=IA=A
6.2.4 Matrix raised to any power

In some circumstances it is necessary to raise a matrix to some
predetermined power. In this case the rules of matrix multiplication
are obeyed as given above. Thus, if we take a simple example
AA=B or A2 =B
where
A~ [: ~J
then
B= [(3 X 3) + (4 X 5) (3 X 4) + (4 X 1) l
(4x3)+(lx5) (4x4)+(lxl)j
From the rules, it follows that only square matrices can be treated in
this fashion.
6.2.5 Determinants
The determinant of a 2 x 2 square matrix is calculated as the
product of the NW and SW elements minus the product of the other
two:
A IAI
[: :]
I (4 X 5) - (2 X 6) I = 8
The determinant of a matrix is zero if any two rows or columns are

identical, or if one row is a multiple of another row. A matrix with a
zero determinant is said to be singular.
The determinant of a 3 x 3 matrix is found as in the following
example.
5 4 2
6 4 3
0 2 5
Stage 1. Multiply the determinant of the 2 x 2 minor formed by the

four elements of the SE corner by the NW element of the top row:
4 3
5 = 5(20 - 6) = 70
2 5

outer four elements by the central element of the top row, and
reverse the sign:
6 3
- 4 = - 4(30- 0) = - 120
0 5

four elements of the SW corner by the NE element of the top row:
6 4
2 = 2(12- 0) = 24
0 2
Stage 4. Sum up the values resulting from the calculations of stages

1-3:
I D 3 1 = 70 - 120 + 24 = - 26
or
5 4 2
ID 3 1 = 6 4 3
0 2 5
4 3 6 3 6 4
=5 -4 +2
2 5 0 5 0 2
= 70 - 120 + 24 = - 26
BASIC MATRIX OPERATIONS I4I
The determinant of a 4 x 4 matrix:
+ +
a b c d
e f g h
j k I m
n p q r
f g h e g h e f h e f g
ID4 1=a k I m - b j I m +c j k m - d j k I
p q r n q r n p r n p q
and the determinant of each 3 x 3 minor is calculated as outlined

above. The determinants of higher-order matrices can be calculated
by extension of the above method and, as can be seen, this is an
extremely cumbersome and long-winded process. Fortunately, there
are other methods for finding the determinant of a matrix, and one
of these will be considered in the next section.
The determinant can be interpreted geometrically, as follows.
Consider the matrix R:
whose determinant IR I= 0- (- 1) = I, and the matrix S:
whose determinant IS I = 4- I = 3. Use these to operate on the co-

ordinates of a unit square a, b, c and d as in Figure 6.3, represented
by the vectors
a= b= c= d=
The multiplication operations are
Ra= r- ll =al Sa= 13] = az

,
L 1
_j
I
L3
r- 21 14 l
d
Rb= =bl Sb= = bz
L LsJ
and
Rc=
r:J =cl Sc= 1lJ
6
L6
= Cz
Rd=
[- ~] =dl Sd= [ :] ~d,
and a, a 1 and a2, etc., are as plotted in Figure 6.3. As noted earlier,
the multiplication of the co-ordinates for each point by the matrix R
represents a rotation and reflection in the y-axis, but without a
change in area; while multiplication by the matrix S involves a shear
to the right and a stretch, resulting in an increase in area. Measuring
the area shows that there has been an increase, by a factor of 3, from
the original. Thus, the ratio of the new area to the old is equal to the
value of the determinant of the matrix used in the transformation. If
a singular matrix had been used, then the transformation would be
to a straight line; in other words, we would lose one dimension. This
is not reversible, whereas transformations brought about by non-
singular matrices are. These ideas can be extended to matrices of
higher order. For example, in the case of a third-order matrix the
vectors will represent the comers of a solid, and after multiplication
by a singular matrix, the transformation will be from a volume to an
area, i.e. again one dimension is lost.
Determinants are the basis of Cramer's method for the solution of
simultaneous linear equations. We can write the equations in matrix
form as demonstrated in Section 6.2.3:
[A][x] = [b]
In words, the matrix of the coefficients of the equations multiplied

by the vector of unknowns, is equal to the vector of values on the
right-hand side of the equation. Cramer's Rule states that
xn = Z/IAI
where Z is the determinant of the matrix of coefficient values with

the nth column replaced by the vector [b], i.e. we replace the
coefficients of the unknown, which we are determining, by the vector
of values from the right-hand side of the equation, provided that
IAI#O.
This method will be dealt with in detail in Section 7.3.3, but is
noted here since the idea of using determinants to solve simulta-
neous linear equations is used in Section 6.3.
6.2.6 Matrix inversion

The inverse (or reciprocal) of a square matrix is the matrix which
when multiplied by the original, gives the identity or unit matrix:
For a 2 x 2 matrix:
Stage I. Reverse the signs of the NE and SW elements.

Stage 2. Interchange the NW and SE elements. -
Stage 3. Divide each element produced by the determinant of the
original matrix.
Note that if the original matrix (of any order) is singular, i.e. has a
determinant of zero, then there is no inverse. Let us now consider an
example:
A A-'
la bl
I
I
I d -b
I i iAT TAT
dj
I'
Lc
I
I!
I
-c a
LiAT IAI
If
A= and IAI = 8
5 - l
8 4 1o.625 - o.25l
-3
4
l
2 l- 0.75 0.5
We can check our arithmetic:
[: :] [l OJ
0.625
[ - 0.25]
-0.75 0.5 0 l
Another method, known as the method of synthetic elimination,

will be described, as this can be extended to deal with square
matrices of any order. It also allows the calculation of the determi-
nant of the matrix at the same time. The method involves augment-
ing the matrix whose inverse is to be found by an identity matrix of
the same order, and by simultaneous operations converting the
former into the identity matrix and the latter into the inverse sought.
We will now use the matrix of the last example to illustrate this
method for the 2 x 2 case:
Stage I. The matrix A is augmented by an identity matrix on the

right:
Stage 2. Row l is divided by the value of the first element (4), to

give l at a 11:
BASIC MATRIX OPER-ATIONS 145
Stage 3. Row I is multiplied by a value (6) such that when it is
subtracted from row 2, we get 0 at a 21:
Stage 4. Row 2 is divided by a value (2) to give I at a 2,2 :
Stage 5. Row 2 is multiplied by a value (!) such that when it is

subtracted from row I, we get 0 at al,2:
[01 OJ
1
[
-;
2- t]
t_
Thus, we have arrived at the same answer as before. We can, by
using the transformation rules relating to determinants, calculate the
determinant of the original matrix. The rules invoked are:
(a) if each element of any row i in a matrix A is multiplied by a

constant a, then the determinant becomes aA;
(b) if two rows of a matrix A are interchanged, then the determi-
nant becomes - A; and
(c) if a multiple of one row is added to another row, then the
determinant remains A.
Thus, the determinant is found as the product of the divisors used to

reduce the pivots to unity at each stage. In the example above, the
divisor at stage 2 used to reduce element a 11 to unity is multiplied by
the divisor at stage 4, which was used to reduce element a 22 to unity.
Thus
IA1=4X2=8
This method can be applied to matrices of any size and, in

practice, because of the large number of arithmetic operations
involved, this is normally performed by computer. A suitable
computer program based on the method is listed as Program 7.3,
where it is used as the basis of a method for the solution of
simultaneous linear equations. However, before leaving the topic we
will work through an example using a 3 x 3 matrix.
Find the inverse of the matrix
5 4 2
6 4 3
0 2 5
We start as before, augmenting the matrix:
5 4 2 0 0
6 4 3 0 0
0 2 5 0 0
Divide row 1 by 5, to give I at au:
0.8 0.4 0 0
6 4 3
10: 0
I
0 2 5 L0 0
Multiply row 1 by 6 and subtract from row 2, to give 0 at a 21:
row 2 6 4 3 0 0
6 x row I 6 4.8 2.4 1.2 0 0
subtract 0 -0.8 0.6 - 1.2 0
which gives
0.8 0.4l 0.2 0 0
:6J
0 -0.8 - 1.2 0
0 2 0 0 0
Divide row 2 through by - 0.8, to give 1 at a 22 :
0.8 0.4 0.2 0 0
0 -0.75 1.5 -0.8 0
0 2 5 0 0 0
Multiply row 2 by 0.8 and subtract from row 1, to give 0 at a 1,2:
row 1 0.8 0.4 0.2 0 0
0.8 x row 2 0 0.8 -0.6 1.2 -0.64 0
subtract 0 - 1 0.64 0
which gives
0 - 1 0.64 0
0 -0.75 1.5 -0.8 0
0 2 5 0 0
Divide row 3 through by 5, to give 1 at a 33:
0 - 1 0.64 0
0 -0.75 1.5 -0.8 0
0 0.4 0 0 0.2
Multiply row 3 by - 0.75 and subtract from row 2, to give 0 at a 23 :
row2 0 -0.75 1.5 -0.8 0
-0.75xrow3 0 -0.3 -0.75 0 0 -0.15
subtract 0 1.3 0 1.5 -0.8 0.15

which gives
0 - l 0.64 0
0 1.3 0 1.5 - 0.8 0.15
0 0.4 0 0 0.2
Divide row 2 by 1.3, to give l at a 2_2 :
- l 0.64 0
0 l.l5 -0.6 0.115
0 0 0.2
Multiply row 2 by 0.4 and subtract from row 3, to give 0 at a 32 :
row 3 0 0.4 0 0 0.2
0.4 x row 2 0 0.4 0 0.46 -0.24 0.046
subtract 0 0 -0.46 0.24 0.154
which gives
ll 0 ll - l 0.64 0
l:
l.l5 -0.6 0.115
0 :J -0.46 0.24 0.154

--'
Subtract row 3 from row I, to give 0 at a 1,3:
row l 0 - l 0.64 0
row 3 0 0 -0.46 0.24 0.154
subtract 0 0 -0.54 0.4 -0.154

EIGENVALUES AND EIGENVECTORS 149
which gives, finally,
II 0 0 -0.54 0.4 -0.154
l: 0
0 1.15
-0.46
-0.6
0.24
The determinant is given, as before, by the product of the divisors:

0.115
0.154
IA I = 5 X (- 0.8) X 5 X 1.3 = - 26
The arithmetic can be checked by multiplication of the matrix by its

inverse to give the identity matrix. Unfortunately, in this example,
because only a maximum of three significant figures has been carried
in the calculation, the result of this operation will not be as expected.
The method of matrix inversion will be looked at later in the context
of the solution of simultaneous linear equations (Section 7.3.3b),
where the problem of accuracy will also be discussed.
6.2.7 Division in matrix algebra

We should note that there is no direct equivalent to the arithmetic
operation of division in matrix algebra. Thus, if we need to perform
this operation on two matrices, we multiply by the inverse of the
divisor, the relationship is:
A/S = s- 1A
It is also worth noting that, since we can only find the inverse of
square matrices, it is only possible to perform this particular
operation if the divisor is a square matrix with the same number of
rows as the other has columns.
6.3 Eigenvalues and eigenvectors
The topic of eigenvalues and eigenvectors is regarded by most authors

as probably one of the most difficult in matrix algebra, a fact which
is borne out by a quick check of the majority of textbooks. Fortunately
for those who would like to understand and use these proper-
ties without becoming too involved in the detailed mathematics,
a paper by Gould (1967) explains the topic in an extremely clear and

understandable fashion. Thus, following the example of King (1969)
and Davis (1973), the discussion of eigenvalues and eigenvectors
which follows relies heavily on Gould's paper. For the student who
would like a more mathematical treatment, Sawyer's A path to
modern mathematics (1966, ch. 5) provides an excellent introduction.
First, we must establish the idea of the characteristic equation
associated with a square matrix A, of order n. Consider a hypotheti-
cal simultaneous linear equations set
[A][x] = l[x]
which is the same as
[A][x] = [b]
except that
[b] =l[x]
What is now required is to find values of l, which satisfy this

relationship. We can rewrite the equation as
([A] - l[I])[x] = [OJ
This relationship can be written in the form of simultaneous linear

equations. Suppose, for simplicity, that we have two unknowns. We
rewrite as
A solution for the vector [x] is given by
[x]=~
IAI
This is Cramer's Rule, where the values of x are given as the ratio of
two determinants; however, the determinant of the numerator is
zero, since it has one row made up of zeros. We can rewrite the
equation as
IAI[x]=O
Now, if the vector [x] is not equal to zero, then it follows that the
determinant I A I must be zero:
I[A]- .l.[I]I = =0
The equation
IA-.l.ll =0
is known as the characteristic equation for any square matrix A of

order n, and the vector ). represents the characteristic roots or
eigenvalues for that matrix.
We can expand the determinant, i.e. write it as we would calculate
it (Section 6.2.5), hence
Multiplying out,
A,-;.
B2-).
A 1B2 - B2.l.
- A,.l. + ).2
- A I A + A I B2 - B2). + A2
which is
Since in the practical situation we know the numerical values of A,,

B 1, etc., we can collect up the terms in the form of a quadratic
A.2 + aA. + fJ = 0
Finding the roots of this equation will give the values of .l., the
eigenvalues of the matrix.
The use of a simple example will help to clarify the calculation
procedure. Consider the equation set
4x+ 2y= a
6x+ 5y=b
The coefficient matrix is
[A]=
[ 46 25]
Subtracting A. times the identity matrix,
[A] _ A.[I] = l4- 6

A. 2 ]
5- A.
Equating the determinant to zero,
[
4-A. 2 l =0
6 5-A.j
Expanding the determinant,
1(4- ..1.)(5- A.)- (2)(6)1 = 0
Multiplying out,
20- 9A. + A. 2 - 12 = 0
Collecting terms,
A. 2 - 9A. +8= 0
Factorizing,
(A.- 8)(A. - 1) = 0
thus A.= 8 or 1, which are the eigenvalues of the matrix A.

Several things should be noted at this stage. In the example just
calculated there are two real roots, but for other matrices this is not
the only possibility. Thus, going back to the problem of the solution
of quadratic equations, it should be remembered that there are
occasions when the roots are complex numbers (Section 2.1.3).
EIGENVALUES AND EIGENVECTORS I 53
Although matrices with complex eigenvalues do occur, m most
practical applications the matrix is symmetric, and symmetric
matrices have real roots. Also, the number of roots will be equal to
the order of the matrix. Finally, since the sum of the eigenvalues is
equal to the sum of the elements in the principal diagonal, this can
be used to check the arithmetic.
Returning to an earlier equation,
(A- A.l)x = 0
which can be rewritten more fully, as (for a 2 x 2 matrix)
Now, if the eigenvalues A. have been found, we can solve for x 1 and
x 2, which are the eigenvectors associated with the particular eigen-
value substituted in the equation. There will be one pair of eigenvec-
tors for each eigenvalue. Thus, for the example which we have been
using:
for A.= 8
2 l
5-8 J
which is analogous to the simultaneous linear equation set
whence x 1 = I and x 2 = 2
for A.= I
2 l
5- I _I
J54 NUMBER ARRAYS, THE ALGEBRA OF MATRICES
which is
whence x 1 = 2 and x 2 = - 3.
To summarize, we have now calculated:
(a) an eigenvalue of8 with an associated eigenvector of(l,2); and

(b) an eigenvalue of I with an associated eigenvector of (2,- 3).
In geometric terms we have reduced our matrix to two vectors whose

magnitude is given by the eigenvalues, and whose directions are
given by the eigenvectors. We are now in a position to consider a
geological example, which we will interpret graphically after calcu-
lating the eigenvalues and their associated eigenvectors.
We start with the following matrix (the row and column labels are
appended for clarity)
Mg K
Mg 1.0 0.45]
[
K 0.45 1.0
whose elements are the product moment correlation coefficients for

the trace elements magnesium (Mg) and potassium (K), based on
data from ten samples of Cretaceous flint, collected from Neolithic
flint mines at Grimes Graves, Norfolk. The values of unity in the
principal diagonal of the matrix represent self-correlation. The
coefficient ranges from + I .0 (perfect positive correlation) to 0.0
(complete disassociation) to - 1.0 (perfect negative correlation).
Students who have not already met this coefficient are referred to
either Cheeney (1983, pp. 87ff.) or Moroney (195I, pp. 286ff.), or
any good statistics textbook for further details. Proceeding as in the
previous example,
r' I - )c
0.45 l
(A]- )c[I] =
~45 I- )c.JI
Equating the determinant to zero,
Expanding the determinant, multiplying out and collecting terms,
A2 - 2A + 0.7975 = 0
Using the formula for the solution of a quadratic,
A= 1.45 or 0.55
which are the eigenvalues of the correlation matrix.

Substituting A= 1.45, we get
l'- ~::
I.
0.45l
1 - 1.45 J
which gives
- 0.45x 1 + 0.45x2 = 0
0.45x 1 - 0.45x2 = 0
whence x 1 = 1 and x 2 = 1. Similarly, substituting A= 0.55, we get
0.45x 1 + 0.45x2 = 0
0.45x 1 + 0.45x 2 = 0
whence x, = 1 and x 2 = -1.

The eigenvalues and eigenvectors can be plotted relative to a pair
of axes x 1 and x 2, where x 1 represents the element Mg and x 2
represents K, as in Figure 6.4. The eigenvalues define the lengths of
two vectors, whose slopes are defined by the eigenvectors. These are
mutually orthogonal (i.e. at right angles) and form the major and
minor axes of an ellipse (it should be noted that not all matrices give
vectors which are mutually orthogonal). From the figure it can be
seen that the vector represented by the eigenvalue 1.45 (plotted using
K
1.0
M
1.0 g
0.55
Figure 6.4 Diagrammatic representation of a correlation matrix for the elements

potassium and magnesium, trace elements in Cretaceous flint from Grimes Graves,
Norfolk. Showing the eigenvectors and eigenvalues. See text for details.
the same scale as the axes, which are of unit length), corresponds to
half the length of the major axis of the ellipse and has a slope of 45,
while the vector represented by the eigenvalue 0.55 is half the length
of the minor axis and has a slope of 135 (both slopes are measured
from the vertical axis of the plot). The values of the correlation
coefficients for the two metals represent the tangents of the angles
which two vectors make with the appropriate axis, and can be
represented by plotting on the perimeter of the ellipse as shown. The
origin of the axes for this type of plot is always at the centre of the
ellipse. To summarize, we have taken a matrix representing the
correlation between two variables, and have reduced it to a pair of
vectors whose magnitude and direction are given by the eigenvalues
and eigenvectors of the correlation matrix. With more-complex data
sets, where many variables are involved, the method can be
extended. It is an extremely useful technique, and is the basis of a
number of methods in multivariate statistics. Another application,
as will be seen in Section 6.4.4, is in structural geology.
During the consideration of the problem of determining the
eigenvalues and eigenvectors, the characteristic equation for a
square matrix of order n has been defined, and a mathematical
GEOLOGICAL APPLICATIONS AND EXAMPLES 157
technique has been described for its solution. The method outlined
can be used to solve square matrices of any order, but it proves to be
too cumbersome for the solution of examples where the order is > 3.
Several alternative methods of solution have been developed to
facilitate the solution of higher-order matrices. One method, specifi-
cally designed to find the eigenvalues and eigenvectors for real
symmetrical matrices, known as the Jacobi method, is described by
Cheeney (1983, pp. ll9ff.). This method is used as the basis of
Program 6.2, which is a development of a Fortran program given by
Davis (1973, pp. 168-9).
6.4 Geological applications and examples
6.4.1 Stratigraphic data analysis

In the introduction to this chapter the idea of recording stratigraphic
data in matrix form was introduced. The particular matrix is known
as a transition matrix, and has two principal uses in geological data
processing. In one application the matrix is transformed into a
transition probability matrix and used in simulation studies. In the
other it is used without alteration in conjunction with its randomized
form, to deduce cyclic associations. Armed with the basic techniques
of matrix algebra, we can look at both of these applications.
(a) TRANSITION PROBABILITY MATRICES: CALCULATION AND

USES
The calculation of a transition probability matrix is relatively simple
once the transition matrix for the sequence has been compiled. The
row sums for each lithology are found, and the individual transition
frequency are divided by the appropriate row sum. This gives values
for the probability of one lithology (row label) passing into another
lithology (column label). Tables 6.4 and 6.5 show these operations
Table 6.4 Transition matrix for a sequence measured in the Jurassic, Blisworth
Limestone. Sample interval 20 em. Lithologies simplified.
Limestone Marl Clay Row sum
limestone 38 4 8 so
marl 4 9 2 15
clay 7 2 6 15
Table 6.5 Transition probability matrix for the data of Table 6.4.
Limestone Marl Clay Row sum
limestone 0.76 0.08 0.16 1.00

marl 0.27 0.60 0.13 1.00
clay 0.47 0.13 0.40 1.00
for data compiled from a sequence m the Blisworth Limestone

(Jurassic, Bathonian). Thus,
=
P
{i.j) s
TU.J)
(1)
where Pu.12 = the probability of lithology i passing into lithology j,

T(i,J) = the transition frequency for lithology i passing into lithology j
and s(l) = Li= I T(iJ)' n being the number of lithologies.

Most uses of the transition probability matrix have been based on
the idea that stratigraphic sequences are not random, and that each
lithology deposited is in some way dependent on earlier events. Such
sequences, which show some dependence on earlier events, occur in
many natural time series and have been called Markov processes,
after the Russian mathematician who was the first to make a
comprehensive study of them. Thus, stratigraphic data can be said
to exhibit a Markov property, which in the simplest case means that
the lithology being deposited at any particular time is dependent on
the immediately preceding depositional event (one-step depen-
dence). However, in most cases it is likely that more-complicated
relationships occur. The notion of the Markov property implies that
the sequence has a 'memory' for previous events. One simple method
of discovering the extent of the 'memory' is to find the probabilities
associated with more than a single step.
Let us consider what happens if we use the data in Table 6.~,
starting, for example, with the lithology marl. For step l the
probabilities of going to any lithology will be as given by the values
in the corresponding row. For subsequent steps the probabilities will
be governed by the axioms of probability theory, as already dis-
cussed in Sections 2.3.1 and 2.3.2. The sequence starts
p(marl-+limestone) = 0.27 (step I)
p (limestone-+ clay) = 0.27 x 0.16 = 0.0432 (step 2)
this follows from the multiplication axiom.

limestone 0. 76 X 0.27: 0.2052
~
i
limestone~ marl 0.08 X 0.27:0.0216
1
cay 0.16 X 0.27: 0.0432
0
~limestone 0.27 X 0.60: 0.162
m~"' m"l~:::
0.60 X 0.60:0.360
060
0.13 X 0.60:0.078
0
~limestone 0.47 X 0.13:0.0611
clay 0.13 marl 0.13 X 0.13: 0.0169
~clay 0.40 X 0.13:0.0520
p (marl->limestone) (step 2): 0.2052 + 0.162 + 0.0611
:0.4283
p (marl-> marl) (step 2) : 0.0216 + 0.36 + 0.0169
:0.3985
p (marl-> clay) (step 2) = 0.0432 + 0.078 + 0.052
:0.1732
Figure 6.5 Tree diagram for the three-state transition probability matrix of Table
6.5, to show the successive probabilities of reaching a given lithology after two steps,
starting with marl.
However, we can also get to clay at step 2 via either
p (marl-+ marl) = 0.60 (step l)
p (marl-+ clay)= 0.60 x 0.13 = 0.078 (step 2)
or
p (marl-+ clay) = 0.13 (step l)
p(clay-+clay) = 0.13 x 0.40 = 0.052 (step 2)
Thus, the probability of being in lithology clay at step 2 is the sum of

the probabilities for each possible path:
p (marl-+ clay) (step 2) = 0.0432 + 0.078 + 0.052 = 0.1732

which follows from the addition axiom of probability theory. The

complete sequence of possible events for two steps is shown as a tree
diagram in Figure 6.5. The diagram summarizes how, starting with
the lithology marl, we can get to any of the three lithologies in the
sequence in two steps, the probability values for each possible path
are given.
If we complete similar calculations starting with each of the other
lithologies, then we will find that the probability values are the same
as those which would have been obtained if we had performed the
matrix multiplication operation
where M is the original transition probability matrix. Similarly, if we

wish to find the probabilities associated with each state after three
steps, we calculate
MMM = MJ = p<step 3)
Successive powering of a transition probability matrix generally

leads to a limiting matrix T, where the elements of each column are
equal; at which point the probability of passing from one state to
another is independent of the starting state. Table 6.6 shows the
result of powering the data of Table 6.5, from which we conclude
that for the data set under consideration, transitions after step 8 will
be independent of the starting state, i.e. the sequence has no
'memory' of the starting lithology after eight depositional events.
Also, since the rows of the matrix are equal, this must represent the
proportions of each lithology in the original sequence on which the
matrix was based. To facilitate this calculation, Program 6.3 is
given, which will power a transition probability matrix to any
predetermined level.
The transition probability matrix is also used as a basis for the
simulation of stratigraphic sequences. For this, the matrix is used in
its cumulative form, where the probabilities in each row are cumu-
lated from left to right. The two matrices in Table 6. 7 show (a) the
transition probability matrix and (b) the cumulative transition
probability matrix for data taken from a sequence in the Jurassic,
Blisworth Limestone, measured at Irchester in Northamptonshire.
The sequence is illustrated in Figure 6.6a.
A simulation is started by choosing, at random, a starting
lithology (say, limestone), picking a random number in the range 0--
Table 6.6 The result of powering the probability transition
matrix of Table 6.5, correct to two decimal places. Row and
column labels have been omitted, as well as some of the
individual steps. Note that the probabilities recorded for
row 2 of the step 2 matrix are the same as those calculated in
Figure 6.5.
step 2 0.67 0.13 0.20

0.43 0.40 0.17
0.58 0.17 0.25
step 4 0.62 0.17 0.20

0.56 0.24 0.20
0.61 0.19 0.21
step 6 0.61 0.18 0.20

0.59 0.20 0.20
0.61 0.19 0.20
step 7 0.61 0.19 0.20

0.60 0.19 0.20
0.61 0.19 0.20
step 9 0.61 0.19 0.20

0.61 0.19 0.20
0.61 0.19 0.20
1.0 (using either random-number tables or a computer). Then, if the

random number falls in the range 0-0.74, the original lithology is
deemed to have passed into itself, if it falls in the range 0.74-0.84
then a marl follows. However, if it is in the range 0.84-1.0 a clay
follows. Suppose our first random number was 0.82, then the
sequence starts
limestone-+ marl
Table 6.7 Transition probability and cumulative probability matrix for the se-
quence illustrated in Figure 6.6a. Sample interval 20 ern.
(a) Transition probability (b) Cumulative transition

probability
Lithology Limestone Marl Clay Limestone Marl Clay
limestone 0.74 0.10 0.16 0.74 0.84 1.00

marl 0.25 0.59 0.16 0.25 0.84 1.00
clay 0.43 0.43 0.14 0.43 0.86 1.00
(a) (b)
Clay
Blisworth Cloy
Limestone
Limestone - Cross Bedded
Marl
Limestone
Morl- Epdllyris Bed
Clay
Limestone Marl
Cloy
Morl Cloy
Limestone
Clay
Limestone- fossiliferous
Limestone
Marl
Clay
Limestone Marl
Cloy Clay
Marl Limestone
Clay Clay
Marl
Clay Limestone
Limestone- fossiliferous-oolitic
Clay
200 Limestone- cross bedded- oolitic Marl
Limestone- d!gl.'noides Bed- oolitic

Cloy
Limestone- oolitic Marl
Clay
em Clay
@ Marl
Limestone -oolitic
@
@
Marl- K.shorpi Bed
0
Figure 6.6 (a) Sequence measured in the Bathonian, Blisworth Limestone at
Irchester, Northamptonshire. (b) Simulated sequence using the Markov simulation
method described in the text and the data of Table 6. 7b.
We select a second random number, say 0.215. Going to the row
giving the transition probabilities for marl, and since our number is
in the range 0-0.25, the marl is followed by a limestone and the
sequence becomes
limestone -+ marl -+ limestone
This process of picking (or generating) a random number and

assigning a new lithology is repeated as many times as required,
building up a simulated stratigraphic sequence.
As the probability transition matrix was based on a 20-cm
sampling interval, it then follows that the lithology generated at each
stage is equivalent to 20 em thickness of rock. The resulting sequence
will not be exactly the same as the original, but will show a close
resemblance. Figure 6.6b shows part of a generated sequence using
the data of Table 6. 7b, alongside the original measured sequence, to
illustrate this point. It must be remembered that there is no objective
method of comparing a simulation of this nature with the real
geological situation. One can only pose the question: is there any
similarity between the two? If the answer is yes, then it is likely to
imply that some of the ideas behind the model may be valid. On the
other hand, a negative answer need not imply that the model is
invalid.
In this section we have presented the notion of event dependence
in the form of the Markov process as a method of analysing
stratigraphic sequences, without recourse to statistical proof or
verification of the hypothesis, but simply as an example of the use of
matrix algebra. Those who would like to pursue these ideas further
should consult the textbooks by Harbaugh & Bonham-Carter (1970)
and Davis (1973), both of which deal in some detail with Markov
models in geology, and contain numerous references to papers on
the subject.
(b) ELUCIDATION OF CYCLICITY IN STRATIGRAPHIC DATA

The validity of the Markov Model as applied in stratigraphic
analysis has been questioned by Selley ( 1970), who proposed a
simple alternative. The model is designed to isolate and define
cyclicity within a stratigraphic sequence starting with the transition
matrix. Selley ( 1970, p. 559) suggests that the transition matrix will
be objective and reproducible if the matrix is based on a record of
bed transitions only. Having found the transition matrix (as shown
in Section 6.1) for the sequence being studied, the next step is to
calculate the equivalent expected random matrix, i.e. the transition

matrix expected if the same number of beds had been deposited at
random. This matrix will have the same total number of transitions
as the original, as well as having the same row and column sums.
Each element of the expected random matrix is found by calculating
E . = RSUMi CSUMj
'' TSUM
where EiJ is the expected frequency for each element iJ, RSUMj is
the row sum of the jth row of the observed matrix, CSUMi is the
column sum of the ith column of the observed matrix, TSUM is the
sum of all the elements in the observed matrix, and i and j range
from 1 to n, n being the number of different lithologies in the
sequence. In other words, if we wish to find the expected frequency
for any element iJ we multiply the row total for row j by the column
total for column i and divide by the total number of transitions.
Having obtained the expected random matrix, this is then subtracted
from the original to give the difference matrix. These calculations are
shown in Table 6.8, which is based on a simple example, recording
Table 6.8 Calculation of an expected random matrix and difference matrix, based
on a transition matrix relating to a sequence in the Lower Carboniferous (see text for
details).
Limestone Sandstone Shale Row sum
(a) Observed transition matrix 0

limestone 0 3 4 7
sandstone 2 0 4 6
shale 5 3 0 8
column sum 7 6 8 N= 21
(b) Expected randomized matrix E

2.3 2.0 2.7 7
2.0 1.7 2.3 6
2.7 2.3 3.0 8
column sum 7 6 8 N=21
(c) Difference matrix D = 0- E

-2.3 1.0 1.3
0 - 1.7 1.7
2.3 0.7 -3.0
transitions only, from a borehole core through strata in the Car-
boniferous, Visean of Upper Teesdale, Durham.
Selley's argument (Selley 1970, p. 560) is that we should pay
particular attention to transitions which occur more frequently than
we would expect if the sequence were random. Thus, by inspecting
the difference matrix we should be able to construct a facies
relationship diagram for the sequence being examined. Looking at
the difference matrix in Table 6.8, the largest difference (numeri-
cally) is for the shale-+ limestone transition, if we now look at the
row corresponding to limestone, the largest value is for lime-
stone-+ shale transition. Thus, the transition
shale -+ limestone -+ shale
represents the major cyclic sequence in the data. Further minor

cycles can be found by looking for the largest value which has not
previously been accounted for. In this example we note that there is
a minor cycle
sandstone -+ shale -+ sandstone
Although these ideas have not received universal acceptance, it is an

easy method to apply, allowing the apparent cyclicity in data sets to
be analysed objectively. Also, as Selley (1970) has shown, the results
can be intriguingly different from those expected.
6.4.2 The matrix and its transpose in statistics

(a) THE DISPERSION MATRIX
In some multivariate statistical methods a matrix known as the
dispersion matrix is used as a starting point in an analysis. The
matrix for each population is made up as follows (in the simple two-
variable case).
Element I, I is the sum of the squared deviations from the mean
(the use of the unqualified mean in this section implies the arithmetic
mean), for variable I
Elements I ,2 and 2, I are the sum of the product of the differences of

the means for each variable
Element 2,2 is the sum of the squared deviations from the mean for
variable 2
which is
D=MM'
where D is the dispersion matrix which will be square and of order i,

and i is the number of variables, M is the data matrix where each
element is expressed as a deviation from the appropriate mean and
M' is its transpose.
The matrix will have i columns (number of variables) and n rows
(number of samples). Details of this calculation are set out in Table
6.1 0, using data for trace elements in ten samples of flint from the
Neolithic flint mine site at Grimes Graves, Norfolk, as listed in
Table 6.9.
Table 6.9 Raw data matrix R, relating to trace element analysis of ten samples of
flint, from Grimes Graves, Norfolk.
Sample no.
2 3 4 5 21 22 23 24 25 Mean
aluminium (ppm) 636 616 605 586 495 668 648 512 635 503 590.4
magnesium (ppm) 29 24 26 26 90 21 21 19 28 18 30.2
potassium (ppm) 240 247 220 247 202 217 250 203 234 172 223.2
(b) THE CORRELATION MATRIX R

Another important matrix in multivariate statistics is the correlation
matrix R, where each element is the product moment correlation
coefficient relating the variables. In this case the data are pretreated
so that the measured variable for each sample has a mean of zero
and a standard deviation of unity. Thus, if N is the normalized data
matrix, then each element ni.j' where i is the sample index and j is the
variable index, is given by
where xi.j is the ith sample of the jth variable, aj is the standard
Table 6.10 Calculation of the dispersion matrix, using the flint mine data of Table
6.9.
(a) The data expressed as the deviation from the mean for each element, matrix M
aluminium 45.6 25.6 14.6 - 4.4 - 95.4 77.6 57.6 - 78.4 44.6 - 87.4
(ppm)
magnesium - 1.2 - 6.2 -4.2 - 4.2 59.8 - 9.2 - 9.2 - 11.2 - 2.2 - 12.2
(ppm)
potassium 16.8 23.8 - 3.2 23.8 - 21.2 - 6.2 26.8 - 20.2 10.8 - 51.2
(ppm)
(b) Data matrix M written as its transpose M'
AI Mg K
45.6 - 1.2 16.8

25.6 -6.2 23.8
14.6 -4.2 -3.2
-4.4 -4.2 23.8
-95.4 59.8 -21.2
77.6 -9.2 26.8
-78.4 - 1!.2 -20.2
44.6 -2.2 10.8
-87.4 - 12.2 - 5!.2
(c) Details of calculation of the dispersion matrix, where MM' = D
D,_, = (45.6 X 45.6) + (25.6 X 25.6) + ... + (- 87.4 X - 87.4)

D,_, = (- 1.2 X45.6) + (- 6.2 X 25.6) + ... + (- 12.2 X - 87.4)
D,_, = (16.8 X 45.6) + (23.8 X 25.6) + ... + (- 5!.2 X - 87.4)
D,_, = (45.6 X - !.2) + (25.6 X - 6.2) + ... + (- 87.4 X - 12.2)
D,_, = (16.8 X 16.8) + (23.8 X 23.8) + ... + (- 5!.2 X 5!.2)
(d) The dispersion matrix D

AI Mg K
AI
l37182.4 - 5358.8
1~92 J
Mg - 5358.8 4099.6 - 884.4
K 10849.2 - 884.4 5777.6

Table 6.11 Part of the raw data set from Table 6.9, and the calculated mean a
standard deviation for each element.
Sample no. Mean Standard

deviation
2 3 4 21
aluminium 636 616 605 586 668 622.2 28.1

(ppm)
magnesium 29 24 26 26 21 25.2 2.6
(ppm)
potassium 240 247 220 247 217 234.2 13.1
(ppm)
deviation of the jth variable and xj is the mean of the jth variable.
The correlation matrix R is given by
R = (l/n)(NN')
where N is the normalized data matrix, N' is its transpose and n is the
number of samples. The matrix R will be square and of order m,
where m is the number of variables. Details of this calculation, using
the data set given in Table 6.11, are given in Table 6.12 as an
example.
In both of the example calculations given, only three variables
have been used. However, the method can be extended to take into
account any number of variables. For details of the use of these
matrices in multivariate statistics, the reader is referred to a statisti-
cal text such as that of Davis (1973). Program 6.4 is a computer
program which allows the multiplication of a matrix by its transpose
and, if required, will also calculate either the dispersion matrix D or
the correlation matrix R.
6.4.3 Vectors and matrices in structural geology

If we consider the case of simple shear in two dimensions, as
illustrated in Figure 6.7, then the vector joining an initial point with
co-ordinates x,y to its final position x',y' can be resolved into
components u, which is parallel to the x-axis, and v, which is parallel
to the y-axis:
u = x'- x v= y'- y
which leads to the displacement equations

Table 6.12 Details of calculation of the correlation matrix R.
(a) The normalized data matrix N (each value has a mean of zero and standard
deviation of unity)
Sample no.
2 3 4 21
aluminium (ppm) 0.49 -0.22 -0.61 - 1.29 1.63

magnesium (ppm) 1.46 -0.46 0.31 0.31 - 1.62
potassium (ppm) 0.44 0.98 - 1.08 0.98 - 1.31
(b) The transpose matrix N'
AI Mg K
0.49 1.46 0.44

-0.22 -0.46 0.98
-0.61 0.31 - 1.08
- 1.29 0.31 0.98
1.63 - 1.62 - 1.31
(c) Multiplication of N by its transpose, where NN' = Rn
5.0 - 2.422 -2.741

- 2.422 5.0 2.283
2.741 2.283 5.0
(d) The correlation matrix R, where R = (1/n)(NN')

AI Mg K
:l
i
AI 1.00 0.48 - 0.55
Mg 0.48 1.00 0.46
K -0.55 0.46 1.00

y y
x.y x~y'
L __ _ _ _ _ _L __ _ _ _ ~ X
X
Figure 6.7 Simple shear showing the transformation of the point (x,y) to the new
position (x' ,y') and the angular shear strain 'I'.
u = - yy v= 0
where y =tan 'fl. 1f1 is the angular shear strain, and is negative.
The co-ordinate transform equations are
x' = x- yy y'=y
or
x' = x- yy y' =Ox+ y
which is the strain matrix:
-y
[:]
Strain equations can be written either relating the final position to
the original or vice versa. Thus, for body translation we have the
following.
(a) Lagrangian specification
x' =ax+ by y' =ex+ dy
in matrix form
[: y
(b) Eulerian equation
x=Ax'+By' y= Cx' +Dy'
in matrix form
[xj [A BJl [x'y'Jl

y C D
If the values for the coefficients of the matrix for the Lagrangian
specification are known, then the values of the coefficients of the
Eulerian form can be calculated as follows:
A = df(ad- be)
B= - bf(ad- be)
C=- ef(ad- be)
D = af(ad- be)
Then, applying the rules of matrix algebra and remembering that the
above equations represent the inverse of a 2 x 2 matrix (Section
r; :J [: :r
6.2.6),
The displacement vector field is given by
u = x'- x =(a- l)x +by
v=y'- y=ex+(d-l)y
which is
[u] v
[a-
e
1 b
d- I
J IlyXJ
L
If the transformation equations are linear, then homogeneous strain

results. One important property of homogeneous strain can be
derived from calculating the effect of distorting a circle of unit radius
with its centre at the origin of the linear displacement equation. It is

deformed into a new shape, an ellipse, and the x' ,y' co-ordinate pairs
representing the perimeter of the deformed circle are given by
[: :] [:J [:]
and the lengths of the major and minor semi-axes of the ellipse are
given by the eigenvalues of the shear matrix, and the principal finite
extensions e 1 and e2 , by the lengths of the eigenvalues minus one.
That is, since the lengths of the major and minor semi-axes are given
by (1 + e1) and (1 + eJ, respectively (Ramsay & Huber 1983, p. 31),
then e 1 = (largest eigenvalue- 1.0) and e2 = (smallest eigen-
value- 1.0). Also, since the eigenvectors associated with each eigen-
value give the vector direction, we can use them to define the
orientation of the strain ellipse. Any area change or dilatation is
given by the determinant of the matrix and the ellipticity by the
square root of the ratio of the largest to smallest eigenvalues. It
should be noted that the strain ellipse concept is a Lagrangian
concept. Students wishing more information on this application of
matrix algebra are referred to the textbook Strain analysis applied to
structural geology by Ramsay & Huber (1983).
Before leaving this topic, it is worthwhile considering two simple
examples. Table 6.13 lists co-ordinate pairs defining the simplified
outline of the Palaeozoic brachiopod genus Spirifer and the resulting
pairs after shear represented by the two matrices
and
[: :J
new co-ordinate pairs being calculated as outlined in Section 6.2.3.
For the first of these matrices, which represents simple shear, the
angular shear strain is given by
tan- 1 1.5 ~ 56
Table 6.13 Co-ordinates (x,y) representing points on the perimeter of a simplified
outline of a spiriferid brachiopod, and the corresponding pairs (x;,y; and x;,y;) after
shear represented by the two matrices given in the text. These data are plotted in
Figure 6.8.
X 3 5 6 7 9 II 3 9 6 5.5 6.5
y 4 4 4 4.5 4 4 4 2 2 1.5 I
x; 7 9 II 12.75 13 15 17 6 12 8.25 7 8
y; 4 4 4 4.5 4 4 4 2 2 1.5 I I
x; 6 10 14 16.5 18 22 26 8 20 13.5 12 14
y; 9 II 13 15 15 17 19 7 13 9 7.5 8.5
We can verify this by reference to Figure 6.8, where a line joining the
point representing the umbo of the shell (a) and the sulcus (b) is
parallel to they-axis in the original undeformed shape. However, in
the deformed shape (outline at lower right on figure) a line joining
the same two points in their transformed positions (a 1,b 1) makes an
angle of about 55 with the y-axis. Also, since the determinant is
unity there is no area change. This can also be verified by reference
to the figure.
The second matrix represents homogeneous shear, and it follows
y
20
10
X
10 20 30
Figure 6.8 An illustration of the effect of shear on the outline of a spiriferid

brachiopod. The lower figure shows the effect of simple shear (it has been moved to
the right in the figure for clarity of illustration), while the upper shows homogeneous
shear.
that the changes will be more complex for this transformation. First
we calculate the determinant of the matrix, which is
2
= I(2 X 2)- (1 X 1) I = 3
2
There will therefore be an area change. The transformed figure will

be three times greater than the original, which can again be verified
by reference to the upper figure in Figure 6.8. The eigenvalues are
given by
Expanding,
. F-4..1.+ 3 = o
(..1.- 3)(..1.- 1) = 0
. 1. = 3 or 1
Thus, the principal extension is 2 and the ellipticity is .J(3/1) = 1.73.

The principal extension can be verified by comparing the horizontal
distance c-d (Fig. 6.8) with the horizontal distance c2-d2, which are
10 and 20 units, respectively.
The eigenvectors are given by
2-..l.x+ y=O
x+ 2-..l.y=O
For the eigenvalue . 1. = 3 we get
-x+y=O
x-y=O
whence x = I and y = 1.
GEOLOGICAL APPLICATIONS AND EXAMPLES 17 5
For the eigenvalue A.= 1 we get
x+y=O
x+y=O
whence x = 1 and y = - 1.
The eigenvectors show that the principal axis of the strain ellipse
for this matrix is at 45 to the x-axis, whereas the minor axis is at
right angles to this. Returning to Figure 6.8 we can verify this by
observing that lines joining corresponding points, for example a,a 2
or b,b 2 will also be at 45" to the x-axis, i.e. parallel to the principal
axis of the strain ellipse.
So far we have only considered simple examples of the shear
matrix order 2, dealing with changes in two dimensions; we can,
however, take into account the third dimension, and more-complex
shear matrices of order 3 and vectors of the same order locating
points in three dimensions can be used. The mathematical pro-
cedures are the same as those outlined above, only the interpretation
will be slightly different. The determinant will give the change in
volume (if any), while the eigenvalues (3) of the shear matrix will
give the lengths of the principal axes of the ellipsoid. Students should
also be aware of the existence and uses of other matrices in structural
geology, often referred to as tensors. These matrices provide a
convenient method of specifying completely the three-dimensional
state of stress at a point. In the context of structural geology, simple
tensors are square matrices of order 3, with the elements represent-
ing the nine stress vectors acting across the positive faces of a small
cube whose edges are parallel to the co-ordinate axis as:
x-axis y-axis z-axis
(J'x.y (J'x.y (jx,y
(J'y.x (J'y,y (J'y.z
(J'z,y (J'z.y (}' z.z
The rules of tensor algebra are the same as those of matrix algebra,
and for their application to problems in structural geology students
are referred to texts such as that of Means (1976).
7 Introduction to numerical
methods
7.1 The need for numerical methods
Implicit in much elementary mathematics is the idea that to a

particular mathematical problem there exists a solution whicfi is
exact, and which can be found by algebraic manipulation. Thus, for
example, an equation of the form
aX 2 +bX+ c= 0
can be solved if the factors can be found; or by the application of the

formula
X=- b .J(b2 - 4ac)

2a
as explained in Section 2.1.3. Unfortunately, such simple algebraic

manipulation is not always possible, and many authors have classi-
fied equations according to the difficulties encountered in solving
them (Harbaugh & Bonham-Carter 1970, p. 170). Thus, even ordin-
ary linear differential equations are relatively easy to solve, whereas
many partial differentials are difficult and, in the case of non-linear
equations, solution may be impossible by traditional algebraic
techniques.
Another difficulty which can arise is that of writing down an exact
decimal representation of the answer to a particular problem.
Because of the need to overcome these difficulties, a branch of
mathematics known as numerical analysis has developed. Thus, as
Noble (1964, p. 1) has commented, 'the aim of numerical methods is
to provide practical procedures for calculating solutions of problems
in applied mathematics to a specified degree of accuracy'. In
geology, since many of the problems which can be tackled using
mathematical methods are complex, often involving 'difficult' equa-
tions, it is important to have some knowledge of basic numerical
techniques. What follows is an introduction to the most commonly
used numerical methods in mathematical geology. Before getting
TilE NEED FOR NUMERICAL METIIODS 177
involved in the methods themselves, it is worth while looking in
some detail at the various problems involved in obtaining numerical
solutions.
7.1.1 Accuracy and error

The accuracy to which an answer needs to be known will depend on
the nature of the problem being tackled. Thus, for example, it would
not normally be acceptable to calculate the value of a particular
parameter to 3 decimal places if that value is going to be involved in
further calculations requiring that it be multiplied by a large number
(say, in excess of 103). On the other hand, common sense suggests
that, if for one reason or another, errors accumulate during a
calculation which cannot reasonably be avoided, then it may not be
worth while carrying out the calculation to a high degree of
accuracy. In general, errors are usually under the control of the
operator, and as the theory relating to how errors accumulate is well
understood, necessary action can be taken.
There are four sources of error.
(a) Round-off error. This is due to the need to represent a number in

decimal notation, using a finite number of significant figures,
and will be dependent on the facilities available for calculation.
Thus, we may be required to find .)2, which can be represented
by the series of fractions: l/1, 3/2,7/5, etc., as in Table 7.1, each
subsequent term becoming closer to the true value of .)2, but
never quite reaching it. If we evaluate each successive term of
this series, using a hand-held calculator (Table 7.1), the follow-
ing is apparent: if we require an answer accurate to l decimal
place, then the value given by the third term is sufficiently
accurate; for 3 decimal place accuracy, rounding-off the fifth
term suffices, and so on. However, on evaluating the ninth and
subsequent terms, we have come to the limit of the accuracy of
our calculator. Therefore, if we require more significant figures,
we will require a more accurate calculating machine or a
computer.
As was noted in Section 1.3.1, it is important to remember
that different computers will give answers to different degrees of
accuracy, since accuracy depends on the word-length of the
machine, and occasionally on the particular compiler being
used. It has been known for two different compilers on the same
computer to give different answers in complex calculations, due
178 INTRODUCTION TO NUMERICAL METHODS
Table 7.1 Evaluation of the series for ../2, using a
hand-held calculator. Numerators and denominators of
successive terms are evaluated by doubling those of the
previous terms and adding the term before that: e.g. term
4 numerator = (2 x term 3 numerator) + term 2 numer-
ator= (2 x 7) + 3 = 17; similarly, term 4 denomina-
tor= (2 x 5) + 2 = 12.
Fraction Decimal
1/1 1.0
3/2 1.5
7/5 1.4
17/12 1.416666
41/29 1.413793
99/70 1.414285
239/169 1.414201
5/7/408 1.414215
1393/985 1.414213
3363/2378 1.414213
to the way in which round-off error is handled. In general,

mainframe computers have a larger word-length than micros,
and the latest generation of micros have a larger word-length
than older models. In this context one must be particularly
careful when adapting programs to run on different machines.
(b) Mistakes. We are all prone to making these. There is only one
way to prevent them; by checking and rechecking. It cannot be
overemphasized how essential it is when developing computer
programs, to have ready a test data set for which the correct
answer is known, to the degree of accuracy required, so that the
program can be thoroughly checked. Also, it is good practice,
when developing a program, to include sufficient PRINT or
WRITE statements throughout the program to ensure that you
know exactly what the machine is doing while it is carrying out
the instructions you have given it.
(c) Errors arising because of the nature of the formulae being used.
For example, a continuous function may have to be represented
by a series of discrete numbers of limited accuracy.
(d) Error inherent in constants. It may be that constants which are
being used have their own inherent error, or are only known
approximately.
It must be emphasized that it is extremely important, in numerical

THE NEED FOR NUMERICAL METHODS 179
work, to bear in mind all of the sources of error during a particular
computation, to make every effort to minimize them and to take
note how they accumulate and affect the final results. It is also good
practice to make a statement of the maximum or probable error
when presenting the results of a calculation. Most good textbooks
contain a statement on error theory; a useful summary for those
requiring an introduction to the subject is given by Cheeney (1983),
or in a more mathematical text by Atkinson ( 1985, ch. 3, pp. 33ff.).
7.1.2 Speed of calculation

The speed of calculation can be very important, particularly since
the advent of microcomputers. For example, a program designed to
perform calculations solving the Laplace equation for two-dimen-
sional potential flow, originally developed and operated on a
mainframe computer requiring only a few minutes of execution time,
took 1t h when programmed to run on a micro. This particular
program, which depends on the use of an iterative method, points to
a problem which is common to all such methods: they may require
long execution times because of the large number of iterations
required to produce solutions which converge to an acceptable
value. Techniques for speeding up calculations are therefore of
paramount importance. Sadly, most methods of speeding up calcu-
lations on microcomputers, such as using the resident integer
variables or integer arithmetic, lead to severe restrictions in accur-
acy. In general terms, we can say that as the number of unknowns
increases, or as the complexity of the equations to be solved
increases, then the time required to solve them also increases.
Frequently these times increase exponentially.
The speed of computation for a particular machine or compiler
can be checked relatively easily using one of the many published
bench tests. These are frequently published in popular computer
magazines, and allow timings to be made in a variety of circum-
stances.
7.1.3 Suitability of method

Care must be exercised in the choice of method, not only because of
the factors discussed earlier, but simply because the methods are not
universal in their application. Many instances could be quoted to
illustrate this problem; one example will suffice at this stage.
We are asked to solve:
which, on solution using the method of Gaussian elimination (see

Section 7.3.1) gives
X 1 = 3.5 X2 = - 3.25 X 3 = 2.25 X4 = 0.25

By back-substitution using the calculated values in the original
equations, we can calculate new values for the right-hand sides of
the equations, which should tally with those given originally. This
operation, which will be explained later, is often referred to as
calculating the residuals, and can be written in matrix form as
32 10 7 8 7l 3.5 0
-1
23 7 5 6 5 -3.25 0
R= I
33 8 6 10 9 2.25 -0.251
31 7 5 9 10 0.25 0
I
J
These residual values do not appear too bad, but it is easy to verify
that
using, for example, the Gauss-Seidel iteration method (Section

7.3.2). This particular example emphasizes the difficulties encoun-
tered during the solution of some problems of this type. As in this
case, the fact that the answer is incorrect is not always apparent at
first sight.
ITERATIVE METHODS 181
7.2 Iterative methods for the solution of equations
This section introduces one of the most important classes of

numerical method, that of iteration. The basic technique is best
illustrated by describing the procedure for solving an algebraic
equation x?- 5x + 2 = 0 and a transcendental equation In x = l + Ifx
(transcendental equations are those containing transcendental func-
tions such as In or tan, for example).
7.2.1 Simple iteration

(a) We are asked to solve the equation
x?= 5x+ 2=0
by a simple iterative method.
Step 1. Rewrite the equation in the form x = f(x):
X = 0.2:i'- + 0.4
Step 2. Make initial guesses of the values of x:
x= 0 and x= 10
Step 3. Putting x = 0 on the right-hand side of the equation,

x= 0.4
Putting this value on the right-hand side of the equation,
X = 0.2(0.42) + 0.4 = 0.432
This process is repeated until, on the fourth occasion, we

find that
x = 0.43825 (lst root)

Step 4. Putting x = 10 on the right-hand side of the equation,
XI= 20.4
x 2 = 83.6
x 3 = 1398.2
etc.
Step 5. Make another guess at the value of the second root: say
x = 5, then put this value on the right-hand side of the
equation:
x 2 = 6.232
x 3 = 8.1676
etc.
The value of the roots of the equation can be evaluated by an exact

analytical method, and are
x = 0.438447 and 4.561553
In this example, any choice of value for x which is just lower than the
value of the largest root, will give answers which will converge on the
lowest root. A choice of x which is higher than the largest root will
always diverge. This illustrates one of the fundamental problems of
iteration, that of convergence.
(b) Find the positive root of the equation (correct to 4 decimal
places),
lnx=l+1/x
First we need to find an approximation x 0 to the positive root of the

equation. The simplest way of achieving this is to sketch the curves,
by plotting the two functions
y =In x and y = 1 + (1/x)

on the same axis, as in Figure 7.1. The point of intersection of the
two curves will give the root of the equation. Inspection of Figure
7.I indicates that the point of intersection lies somewhere just to the
right of x = 3
x 0 = 3.0
We now rewrite our equation in the form
x= f(x)
which is
x = exp[l + (1/x)]
and set up the recurrence relationship
In other words, having written the function with the unknown

required on both sides of the equation, we attach subscripts to the
unknown such that an estimated value can be inserted on the right-
Y
Figure 7.1 Sketch of the curves of

the two functions y = lnx and
y =I+ 1/x. The point where the
two curves cut, near x = 3, repre-
sents an approximation to the value
of the root near 3, of the equation
-3 lnx= I+ 1/x.
hand side of the equation and an updated estimate calculated. This

new value can then be used to produce further updates until
successive estimates differ by less than some preset tolerance.
Students should note that it is not normally necessary to calculate
values for the initial steps, using this method, to the accuracy
required in the final answer.
Setting x 0 = 3.0, we get
x 1 = exp[l = (1/3.0)] = exp(l.3) = 3.7

x 2 = exp[l + (1/3.7)] = exp(l.27) = 3.56
x 8 = exp[1 + (1/3.5913)] = 3.59107
x 9 = exp[l + (1/3.59107)] = 3.59114
Since the successive approximations are alternatively above and

below the root, the root lies between 3.59107 and 3.59114. Hence,
x = 3.5911, correct to 4 decimal places.
We can now summarize the method as used in the two examples,
to arrive at a general procedure for simple iteration.
Step 1. Find approximate values for the root or roots required.

The simplest method is to sketch a graph y = f(x), points
where the curve cuts the x-axis give approximate values of
the real roots. Sometimes (as in the second example,
above) it is convenient to write the equation in the form
f..(x) = fz(x) and plot the two functions y = f..(x) and
y = fz(x). Any real roots are given by the abscissae of the
points of intersection.
Step 2. Write the equation in the form x = f(x).
Step 3. Set up the recurrence relationship.
Step 4. Solve the equations, starting with the initial approxima-
tion of the value (or values) of the root, iterating until
successive estimates are within the accuracy required.
(c) Testing for convergence. In the first example above, using the
method of simple iteration, we could only find one of the two roots
for the equation. For one root successive estimates converged to give
the correct answer, while for the other, the estimates diverged,
successive estimates becoming larger (i.e. x, tends to infinity as r
increases). From a practical standpoint it is important that conver-
gence should occur, and that it should happen as rapidly as possible.
To test for convergence, it can be shown that an equation will
converge if the following result holds:
In other words, if, after differentiating the equation and putting into
the differential the initial guess for the unknown root, the result is
less than unity, then the iteration will converge, giving a real root of
the equation. Returning to our two examples:
For the quadratic x = o.2.x2 + o.4,
f'(x) = 0.4x
when x = O,f'(x) = 0 will converge
when x = 4, f'(x) = 1.6 will not converge
For the equation x = exp[1 + (1/x)],

f'(x) = - (1/r)exp[1 + (Ijx)]
f'(x 0) = - (1/9)exp[1 + 1/3]

~- 0.4
therefore the equation will converge. This test can be extended to

include higher-order equations, as required.
In the simple cases considered, convergence has been quite rapid,
but this is not always the case. However, it can be said that if a real
root or roots exist and the iteration converges, then an answer will
eventually be obtained using the method as outlined. It is also true
that a correct answer will eventually be obtained, even though there
are arithmetic errors in the intermediate calculation.
7.2.2 The Newton-Raphson iteration formula

The Newton-Raphson iteration formula states that if xn is an
approximation to the root a of the equation
f(x) = 0
then a better approximation is generally given by xn+ ,, where
Also, convergence will tend to be faster by this method than by the

simple iterative procedure just described.
We are asked to find, correct to 4 decimal places, the root near 0.5
of the equation
e-x- sin x =0
f(x) =e-x- sin x
f'(x)= -(e-x+cosx)
Hence the Newton-Raphson formula becomes
e-xn- sin X
X
n+ I
=x
n
+e Xn + COS Xnn
If x 0 = 0.5, then
x, = 0.5 + 0.128/1.485 = 0.586
x 2 = 0.586 + 0.0035/1.3897 = 0.5885
x 3 = 0.5885 + 0.0005/1.3693 = 0.5885

Hence
x = 0.5885 correct to 4 decimal places
This is also an excellent method for calculating the nth root of a

number. For example, calculate .J11 correct to 4 decimal places:
f(x) = r- a
f'(x) = 2x
The Newton-Raphson formula becomes

SIMULTANEOUS LINEAR EQUATIONS 187
x2 - a
-xn--2-
xn+l- n
x.
Choosing x 0 = 3.0,
XI= 3- ((9- 11)/6) = 3.33
x 4 = 3.31667 - [(3.31667 2 - 11)/6.63334] = 3.31662
Hence .J 11 = 3.3166, correct to 4 decimal places.

Unfortunately, space does not permit a more extensive treatment
of iterative procedures, and students requiring further information
should consult a text on numerical analysis, such as Noble (1964) or
Atkinson (1985). The iterative solution of simultaneous linear
equations will be discussed in the next section.
7.3 The solution of simultaneous linear equations
The solution of simultaneous linear equations is a problem of great

practical importance in many fields of science, particularly since
many problems in applied mathematics can be reduced to such a set
of equations. In geology a number of applications spring to mind: in
optimizing problems using linear programming; in contour mapping
using trend surface analysis; in problems involving the solution of
the Laplace equation for modelling the flow of liquids or gases in
rock; in a number of multivariate statistical methods; etc. Many of
these we have either already considered, or will consider later. Thus,
in general, we are required to find
when
188 INTRODUCTION TO NUMERICAL METIIODS
Given values for
The methods of solution can be divided into three separate groups.
(a) Developments of traditional algebraic techniques:
(i) Gauss-Jordan elimination; and

(ii) Gaussian elimination.
(b) Iterative methods- the Gauss-Seidel method.

(c) Matrix methods:
(i) determinants - Cramer's Rule; and

(ii) matrix inversion.
The choice of method will depend on a number of factors, including

how large the system of equations is and what type of calculating
equipment is available. Both of these factors lead back to the
problems mentioned earlier, those of speed and accuracy. For
instance, suppose we were to use Cramer's Rule in a situation where
the number of unknowns was 100: we would need 1.6 x 10 160
operations and, if our computer does I million operations per
second, it would take 5 x 10 146 years to complete the calculation! We
have already shown that although it is possible to test the accuracy
of a solution by back-substitution and calculating residual values
(the difference between the given and calculated values from the
right-hand side of the equation), the small numerical size of the
values does not guarantee the accuracy of the solution. It may also
transpire that a particular problem may not be soluble using the
technique chosen, if at all. As an example, we are asked to solve
89x2 = 144
89x 1 + (144 + a)x2 = 233
when a= 0.018 and a= 0.02. On solving, we find

(a) When a= 0.018,
x 1 = - 159.2 and x 2 = 100
(b) When a= 0.02,
x 1 = 18.8 and x 2 = - 10.0
Also, if we are asked to solve these equations when a= 0.01818,

there is no solution, the reason for this being that the first equation is
almost exactly a multiple of the second. An example such as this
serves to illustrate the problem of conditioning, in which the matrix
of coefficient values is said to be 'ill-conditioned'. This is actually a
misnomer, as it is the problem which is ill-conditioned, not the
matrix. Since problems of this nature arise frequently in many
practical situations, the matter will be discussed more fully in
Section 7.3.4.
7.3.1 Methods of solution based on algebraic techniques

(a) GAUSs-JORDAN ELIMINATION
In Gauss-Jordan elimination we systematically eliminate unknowns
from the left-hand side of the equation to leave the pivots, or
principal diagonal, of Xp x 2, , x". We are asked to solve
2x 1 - x 2 + 3x3 - x 4 = 7 (1)
x 1 - x 2 + 4x 3 - 2x4 = 5 (2)
3x 1 + 2x2 + x 3 + 4x4 = 31 (3)
4x 1 - 3x2 + 3x3 - 3x4 = - 5 (4)
First we need to find multiples ofEqn I which will eliminate x 1, from

Eqns 2, 3 and 4. Appropriate values are - t, - f and -f.
For convenience we can note these multipliers, denoted by m, in
the left-hand column. Also, for clarity we can leave out Xp x 2, etc.,
and tabulate:
190 lNTRODUCTION TO NUMERICAL METHODS
m x, x2 X3 x4 b
2 - I 3 -I 7 (I)
-t - I 4 -2 5 (2)
-.l
2 3 2 4 31 (3)
2
4
4 -3 3 -3 -3 (4)
To calculate these multipliers the coefficient (2) of x 1 in Eqn 1, is

used as a divisor; this is known as the lst pivot and Eqn 1 is known as
the lst pivotal equation. Therefore, to find the multiplier we divide
the value we are eliminating by the pivot and reverse the sign. We
then multiply Eqn 1 by the appropriate multiplier, and subtract
from this the equation which contains the value we wish to elimi-
nate, as shown below:
XI x2 X3 x4 b
2 -1 3 -I 7 (1)
-t
- 1 t -.l
2
~
2
7
2
add -I 4 -2 5 (2)
-t 5
2
-.l
2
3
2 (6)
By repeating this operation, using Eqns 3 and 4 and the appropriate

multipliers, we get
XI Xz X3 x4 b
2 - 1 3 - 1 7 (5)
-t 5
2
-.l
2 2
3
(6)
7 -22 II 41
2 2 2 (7)
-1 -3 - I - 19 (8)
Note that we have recorded the pivotal equation (now labelled Eqn
5), unchanged after these operations. The next step is to eliminate
the x 2 coefficients from Eqns 5, 7 and 8. In this process the x 2 value of
- t in Eqn 6 is used; this is the 2nd pivot and Eqn 6 is known as the
2nd pivotal equation. The multipliers are calculated using - t as the
divisor:
m x, xl X3 x4 b
-2 2 - 1 3 -I 7 (5)
-t 5
2
-1.
2
3
2 (6)
7 7 II 41
7 2 2 T T (7)
-2 - 1 -3 -I - 19 (8)
and, performing the operations as before, using the new set of
multipliers, we get
x, xl X3 x4 b
2 -2 3 6 (9)
5
-l
2 2
3
2
3
2 (10)
14 -5 31 ( 11)
-8 2 -22 (12)
We now need to eliminate the x 3 coefficients from Eqns 9, 10 and 12.

The 3rd pivot is x 3 in Eqn 11, and Eqn 11 is the 3rd pivotal equation.
The multipliers are calculated as before, using the coefficient 14 as
the divisor:
m x, x2 XJ x4 b
I
7 2 -2 3 4 (9)
5 5 3 3
2
-l
2 2 2 2 (10)
14 -5 31 ( 11)
4
7 -8 2 -22 (12)
So we have
x, x2 xJ x4 b
2 7
9 59
7 (13)
-t -!l
28
-ill
28 (14)
14 -5 31 (15)
-.2.
7
-N7 (16)
By back-substitution we can eliminate the x 4 -values from Eqns 13,

14 and 15, and find that
For this method, ifn is the number of unknowns, then the number of
operations N is given by
(b) GAUSSIAN ELIMINATION

The Gaussian elimination method is similar to the Gauss-Jordan
method just described, but instead of eliminating all terms except the
pivots, leaving the principal diagonal, only the bottom half of the
matrix is eliminated.
We are asked to solve
2x 1 - x 2 + 3x3 - X4 = 7 (I)
x 1 - x 2 + 4x 3 - 2x4 = 5 (2)
3x 1 + 2x2 + x 3 + 4x4 = 31 (3)
4x 1 - 3x2 + 3x3 - 3x4 = - 5 (4)

We start as before, by eliminating x 1 from Eqns 2, 3 and 4, to give
XI x2 x3 x4 b
2 - 1 3 - 1 7 (5)
-t T
5 -1.
2 T
3 (6)
_]_ II
T
7
2 T
41
T (7)
-1 -3 -1 - 19 (8)
Next we eliminate x 2 from Eqns 7 and 8 only, leaving the 1st pivotal
equation unchanged:
m XI x2 x3 x4 b
2 -1 3 -1 7 (5)
-t T
5 -1.
2 T
3 (6)
7 T
7 _]_
2 T
II 41
T (7)
-2 -1 -3 - 1 - 19 (8)
to give
XI x2 x3 x4 b
2 -1 3 - 1 7 (9)
-t 5
T
_]_
2
3
2 (10)
14 -5 31 ( 11)
-8 2 -22 (12)
Finally, we eliminate x3 from Eqn 12, leaving the other equations

unchanged:
m x, Xz xJ x4 b
2 - 1 3 - 1 7 (9)
-t 2
5 _]_
2
3
2 (10)
14 -5 31 ( 11)
8
T4 -8 2 -22 (12)
to leave
x, x2 x3 x4 b
2 - 1 3 - 1 7 (13)
_]_ 3
-.l
2 2
5
2 2 (14)
14 -5 31 (15)
--2- -30
(16)
7 T
By back-substitution we find that
x, = 1 x2= 2 x 3 = 4 x 4 = 5
For this method, if n is the number of unknowns, then the number of

operations N is given by
We can compare the two methods as shown in Table 7.2.
(c) GAUSSIAN ELIMINATION WHERE THE COEFFICIENTS ARE

INEXACT
In the previous examples, exact coefficients obviated the necessity of
checking the calculation for accuracy at each stage. When the
coefficients are inexact it is necessary to check the calculation, to
ensure that errors do not accumulate, and that the final answer will
be sufficiently accurate.
We are asked to solve the following sets of equations by Gaussian
elimination:
Table 7.2 Comparison of Gauss-Jordan and Gaussian elimination in diagramma-
tic form. In the Gauss-Jordan method we reduce the matrix of coefficients to the
elements of the principal diagonal, whereas in Gaussian elimination we reduce it to
the upper triangular form.
Gauss-Jordan Gaussian elimination
stage I X X X X X X X X
X X X X X X X X
X X X X X X X X
X X X X X X X X
stage 2 X X X X X X X X
X X X X X X
X X X X X X
X X X X X X
stage 3 X X X X X X X
X X X X X X
X X X X
X X X X
stage 4 X X X X X X
X X X X X
X X X X
X X
stage 5 X
X
X
X
2.4759x 1 + 1.6235x2 + 4.623lx3 = 0.0647 (I)
l.4725x 1 + 0.9589x2 - l.3253x 3 = 1.0475 (2)
2.695lx 1 + 2.8965x 2 - l.4794x 3 = -0.6789 (3)
First we set out the equations as follows:
m XI Xz x3 b L
2.4759 1.6235 4.6231 0.0647 (l)
1.4725 0.9589 - 1.3253 1.0475 (2)
2.6951 2.8965 - 1.4794 -0.6789 (3)

196 INTRODUC110N TO NUMERICAL METHODS
Notice that we now have one extra column in our layout, the check-
sum column, labelled L
this will contain the sum of the numerical
values for each equation:
.L = x, + x 2+ ... xn + b
We form our multipliers as before, dividing the coefficient that we
are eliminating by the pivot, and reversing the sign. We perform this
operation to at least one more decimal place than the original
coefficients:
m x, x2 X3 b .L
2.4759 1.6235 4.6231 0.0647 8.7872 (1)
-0.59473 1.4725 0.9589 - 1.3253 1.0475 2.1536 (2)
- 1.08853 2.6951 2.8965 - 1.4794 -0.6789 3.4333 (3)
which gives
m x, x2 X3 b .L
2.4759 1.6235 4.6231 0.0647 8.7872 (4)
-0.00664 -4.07480 1.00902 -3.07241 (5)
1.12927 - 6.51178 -0.74933 -6.13183 (6)
Notice that we have two new check-sum values for Eqns 5 and 6. We
can now check our calculation as follows.
Take the check-sum value for the 1st pivotal equation, multiply by
one of the row multipliers as in (a) below. Add this to the check sum
for the row chosen as in (b) below. This value should agree with the
new check-sum value for that row, i.e. subtraction should give a zero
value, as in (c) below.
For example, to check Eqns 2 and 5:
(a) 8.7872 x -0.59473 = - 5.226011

(b) - 5.226011 + 2.1536 = - 3.072011
(c) - 3.07241 - (- 3.072011) = 0.000399
Again, to check Eqns 3 and 6:
(a) 8.7872 x - 1.08853 = -9.56513

(b) -9.56513+3.4333= -6.13183
(c) -6.13183-(-6.13183)=0
We now eliminate the x 2-value from Eqn 6:
m b
2.4759 1.6235 4.6231 0.0647 8.7872 (4)
- 0.00664 - 4.0748 1.00902 - 3.07241 (5)
170.07078 l.l2927 -6.51178 -0.74933 -6.13183(6)
to give
XI x2 XJ b I
2.4759 1.6235 4.6231 0.0647 8.7872 (7)
-0.00664 -4.0748 1.00902 - 3.07241 (8)
- 699.51619 170.85548 - 528.6588 (9)
Back-substitution leads to
XI = 1.8408 X2 = - 2.072 X3 = - 0.2442
Now, since we are dealing with inexact coefficients, it is wise to

check our calculations by calculating a vector of residual values r.
Using matrix notation, we can write
r=b-d
where b is the vector containing the original values from the right-
hand sides of the equations and dis the vector containing calculated
values for the right-hand sides of the equations, calculated as:
where Ax is the matrix of coefficients for the x-values and c is the

vector of x-values:
(2.4579 X 1.8408) + (1.6235 X ( - 2.072)) +

(4.6231 X ( - 0.2442)) = 0.0648
(1.4725 X 1.8408) + (0.9589 X ( - 2.072))-

(1.3253 X ( - 0.2442)) = 1.0467
(2.6951 X 1.8408) + (2.8965 X ( - 2.072)) -

( 1.4794 X ( - 0.2442)) = - 0.6821
b d r
0.0647 0.0648 - 0.0001
1.0475 1.0467 0.0008
-0.6789 -0.6821 -0.0032
The size of the third residual is probably due to the loss of significant
figures during the formation of the 2nd pivot and its use in
subsequent calculations. This suggests that the equations may be ill-
conditioned, and it may be that we would only be justified in quoting
our answer correct to 2 decimal places.
In the methods described we have been systematic in our choice of
pivots, and have also used the values in the order given and without
any attempt at scaling. However, there are variants of the methods
discussed which involve scaling where it is practicable, and involve
choosing pivots according to their numerical value. This last method
has the advantage that, if we use the numerically largest element in
each row as the pivot, then inaccuracies in calculation due to using
excessively small pivots, as in the last example, can be avoided.
A suitable computer program for the solution of simultaneous
linear equations is listed as Program 7.1. It is based on the Gaussian
elimination method as described in the text, except that the pivots
are chosen on the basis of the largest element in each row. Results
from the program include the value of the determinant of the
coefficient matrix and the values of the pivots used in the calculation.
7.3.2 Methods of solution based on an iterative technique
Algebraic methods of the solution of simultaneous linear equations
and those based on matrix algebra are direct methods, whereas the
iterative method, known as the Gauss-Seidel method, is an indirect
method.
For example, we are asked to solve the equations
We can rewrite the equations in terms of x 1 and x 2 :
x 2 = (4- 2x 1)/3 (2)
We can now guess a value for x 2, say 3, and solve Eqn 1:
XI = (2 - (2)(3)]/3 = - 1.33
This value can be used in Eqn 2:
x 2 = [4 - (2)(- 1.33)]/3 = 2.22
Putting the new value of x 2 into Eqn I,
XI=- 0.81
Putting this value into Eqn 2,
x 2 = 1.87
We can repeat this procedure as many times as we wish, to give an

answer to any required degree of accuracy. The results for this
problem are tabulated below, for seven iterations.
Iteration number
2 3 4 5 6 7
x, - 1.33 -0.81 -0.58 -0.51 -0.44 - 0.41 -0.41

x, 2.22 1.87 1.72 1.67 1.62 1.61 1.60
The true values are
X 1 =- 0.40
The solution of a pair of simultaneous linear equations by this

method is shown graphically in Figure 7.2. Note that the solution of
the equation set is where the two curves intersect, and the iterative
solution is indicated as a path between the lines representing the
equations.
In general we have
If there are more than two unknowns, then we can proceed by one of
several different methods, only two of which we need to consider
here.
(a) The method of simultaneous corrections, where the estimates of

x 1, , xn remain constant during each cycle of iteration, only
In 1 t i a I
guess
Figure 7.2 Graphical illustration

of the iterative solution of simulta-
neous linear equations. Each suc-
cessive iteration produces values of
x andy which converge on the true
value as indicated. X
being updated at the end of the cycle. The recurrence relation-
ships for the case of three unknowns are
(b) The method of successive corrections, where the estimates of x 1,

... , x. are updated as each new estimate is available. In the
situation where three equations are involved the recurrence
relationships are
- c2y;l()
3
+ z v\
Either method is satisfactory, although the second generally con-

verges much quicker than the first. The starting estimates for x\(),
etc., can be set to zero or equal to the value from the right-hand side
of the equation divided by the coefficient corresponding to the x-
value being estimated.
Iterative methods for the solution of simultaneous linear equa-
tions have a number of advantages over others. They are relatively
fast computationally, and are particularly useful where the matrix of
coefficients is diagonally dominant. In the specific situation where
there are a large number of zero values among the coefficients, it is
probably the best method to use, since problems of ill-conditioning
are not so important.
A computer program based on the method of successive correc-
tions is listed as Program 7.2. This program allows a maximum of
1000 iterations with the tolerance for an acceptable solution being
set at 0.0001, both of these can be altered by changing statements
410 and 190, respectively. If an acceptable solution is not found after
the maximum number of iterations, then the current estimates x\(),
. , x~l) are listed. Also, the initial guesses are set at zero in this
version; again alteration of statement 220 can change this assign-

ment. No specific test has been included to ascertain whether the
solution will converge and execution will be terminated by the

program error 'TOO LARGE', if the solution is divergent.
7.3.3 Methods of solution based on matrix algebra

There are two principal methods for the solution of simultaneous
equations based on matrix algebra, using determinants and matrix
inversion. The first of these, known as Cramer's Rule, which we have
already noted in passing (Section 6.2.5), is extremely time-consum-
ing when large numbers of unknowns are involved, and is not
recommended for large systems of equations. The other method,
matrix inversion, is much less cumbersome, but it can be affected by
ill-conditioning induced by problems of round-off error. Both
techniques will be discussed.
(a) THE METHOD OF DETERMINANTS: CRAMER'S RULE

Taking as a starting point two general equations:
We can rewrite as
and
provided that
Now, from our earlier consideration of matrices (Section 6.2.5), the

determinant of a 2 x 2 matrix is
IAI =ad- cb
where
A~ [: :]
Therefore we can write our equations in matrix form as
z, b, a, z,
z2 b2 a2 7'
~2
X=
I
X2=
a, b, a, b,
a2 b2 a2 b2
provided that
The method can be extended to solve sets of equations with as many

unknowns as required. We proceed as follows:
(l) Calculate the determinant of the coefficient matrix. If this is not

zero, proceed as below.
(2) To find xn, replace the vector containing the xn coefficients with
the vector containing the values from the right-hand side of the
equation, calculate the determinant and divide by the determi-
nant of the coefficient matrix.
(3) Repeat (2) until all xn have been determined.
For example, solve the following:
3x+y+4z=l2
2x + 4y + 2z = 8
y + 3z = 16
First, find the determinant of the 3 x 3 coefficient matrix (Section

6.2.5):
3 1 4
4 2 2 2 2 4
IDI= 2 4 2 = 1 - 1 +4
1 3 0 3 0
0 3
= 32
Since D -=1= 0, we can continue
12 4
x=.!..
32 8 4 2 = 3~6 =- 3
116 3
3 12 4
y=fi 2 8 2 =#=1
0 16 3
3 12
z=.!..
32 2 4 8 ='3~=5
0 16
The principal drawback to this method is the amount of simple

arithmetic which has to be carried out, so it is not a method
recommended for the solution of systems of equations with more
than three unknowns for hand-calculation, or more than six
unknowns using a micro.
(b) SOLUTION BY MATRIX INVERSION
It is possible to take our original general form of the equations and
rewrite them in matrix form:
A=
X1l -Z 1 l
x=
~21 b= z2
I
~"J
I
Zm J
An exact analytical solution is given provided that IA I #- 0, using
X= A- 1b
If we consider the previous example, then the inverse of the

coefficient matrix is calculated using the method of synthetic elimi-
nation outlined in Section 6.2.6, to give
0.3125 3.125 x w- 2 -0.4375
A-1 = -0.1875 0.28125 - 6.25 x w- 2
6.25 x w- 2 -9.375 x w- 2 0.3125
and
12 -3-1
[A -I] 8 l I
I
16 5J
A program using this method is listed as Program 7.3, and is based

on a matrix-inversion program given by Davis (1973, p. 142). Some
indication of the condition of the original equations can be gained
by comparing the size of the individual terms of the inverse matrix
with the roots of the equations. If these terms appear large com-
pared with the roots, then it is likely that the equations were ill-
conditioned.
7.3.4 Ill-conditioned equations

It is important to be able to recognize whether a particular set of
simultaneous linear equations is ill-conditioned, particularly since, if
they are, it may be that the solution obtained using a particular
method is not correct and that any subsequent calculations based on
the answers obtained may be misleading. There are no formal,
simple tests for conditioning, and therefore the following notes are
for guidance only. There is no guarantee that they will always work.
(a) If one equation is nearly a multiple of the other, then the system
of equations will be ill-conditioned and problems of round-off
error will become very important. Small changes in the coeffi-
cients lead to large changes in the roots of the equations (cf.
example in general discussion at start of Section 7.3). Test by
calculating the determinant which should be small compared
with the expansion of the terms that make up the determinant.
Also note that if the determinant is small in the sense just
described: it is likely that an iterative procedure will take a long
time to converge. Thus, the rate of convergence can be taken as
a rough guide to condition.
(b) If the determinant is zero, then the equations are exact multiples
and plot graphically as parallel lines.
(c) Estimates of roots which differ appreciably from the true roots
can produce small residuals (cf. earlier comment on small
residuals not guaranteeing good solutions, Section 7.3.lc).
(d) If individual terms of the inverse of the coefficient matrix are
large compared with the roots of the equation.
(e) If the ratio of the largest to the smallest eigenvalue is large.
The opposite to ill-conditioned is well-conditioned, and occurs

when the values of the diagonal elements of the coefficient matrix are
large relative to the others. When all of the non-diagonal elements
are zero, then the equations are perfectly conditioned. Some graphi-
cal representations of these problems where ill-conditioning can
arise, are shown in Figure 7.3. Although systems of equations
involving only two unknowns are shown, because of the limitations
of graphical illustration, the principles involved are the same no
matter how many unknowns are involved. It is hoped that the above
guidelines will enable the problem to be identified and, in most cases,
overcome if at all possible.
As was seen in Section 4.5.2, one practical use of the solution of
simultaneous linear equations, which has been extensively used by
geologists, is trend surface analysis. This is a technique based on the
method of multiple regression; unfortunately, it is particularly prone
to problems of ill-conditioning. There are three principal causes:
(a)
89 1~4.011= -0.45
155 89
-6
(b)
- 3
- 5
Figure 7.3 Examples of ill-conditioned equations. (a) Graph of the equations
89x 1 + (144 + a)x, = 233 and 55x, + 89x2 = 144, where a is a small increment. The
two equations plot almost exactly on top of one another. (b) Graph of the pair of
equations 3x, + 2x, = 2 and 6x, + 4x2 = 10, which plot as two parallel lines.
(a) round-off error during the calculation of the coefficient matrix;

(b) badly distributed data (in the geographical sense) and
(c) rogue values.
Of these, the most common is the first and, in practice, it often

proves to be the easiest to overcome. In most cases it is simply a
matter of recalculating the geographical co-ordinates with their
origin at the centre of the area being studied.
7.4.1 Iterative methods

(a) THE CALCULATION OF THE DIVERSITY INDEX, FISHER'S a
In palaeoecology it is often useful to be able to express by a single
number the relationship between the number of individuals in a
population and the number of species which go to make up that
population. Such a number is generally termed the diversity index.
One useful index is 'Fisher's a' (Williams 1964), which is a special
case of the Poisson distribution, where the zero occurrence is not
taken into account. This particular diversity index is important
since, as well as expressing diversity as a unique number, it also
allows the calculation of the theoretical numbers of species with 1, 2,
3 ... , n individuals in the population.
The diversity index, a, is given by
S= a In ( 1 + ~) (1)
where S = the number of species and N = the number of individuals

in the population. Now, if
x= N/(N+ a) (2)
then the theoretical number of species with 1, 2, 3, ... , N individuals

is given by the successive terms
ax, a:il/2, ax3/3, .. ., ax"/N (3)
Suppose we wish to calculate a (correct to 1 decimal place), when

the sample size (representing the population), is 350 individuals and
the number of species in the population is 30.
(a) Notice that Eqn 1 cannot be solved analytically.
(b) Rewrite Eqn 1 as a recurrence relationship:
(c) Make an initial guess of a= 0.1
a 1 = 30/ln(l + 350/0.1] = 3.676
a2 = 30/ln[l + 350/3.676] = 6.5695
a 3 = 30/ln[l + 350/6.5695] = 7.5111

a 4 = 30/ln[l + 350/7.5111] = 7.7664
a 5 = 30/ln[l + 350/7.7664] = 7.8327
Therefore the value of a required= 7.8

Having calculated the required value for a, we can now find
the number of species represented by I, 2, 3, 4 and 5 individuals.
We now have to calculate the value of x in Eqn 2 above.
Substituting the required values, we get
X= 350/(350 + 7.8) = 0.978
and, from Eqn 3,
no. of species represented by 1 individual

= 7.8 X 0.978 = 8
no. of species represented by 2 individuals

= 7.8 X 0.957/2 = 4

= 7.8 X 0.935/3 = 2

= 7.8 X 0.915/4= 2

= 7.8 X 0.895/5 = 1
Generalizing from the results obtained, we can say that Fisher's a

distribution tells us' that, in a sample of a population, there are
normally more species represented by a small number of individuals
than are represented by a large number. In the example above, just
over 50 per cent of the species are represented by four or less
individuals, forming about 8.5 per cent of the total sample size. It is
instructive to compare these generalizations, based on this hypothe-
tical example, with the data plotted on Figure 7.4. This shows a plot
of number of individuals against number of species, for a collection
of invertebrate fossils, from the Visean, Scar Limestone, at its
outcrop in Swinhope Burn, Nr Stanhope, Co. Durham. This collec-
tion, which is assumed to represent a living population, contained
304 individuals of 33 species. The figure shows both actual and
calculated distributions, which match quite closely. Also, the distri-
bution is similar in pattern to the example calculated.
11
o Observed
Calculated
9
7
Vl
LU
u
LU
~ 5
Vl
u..
0
a:: 3 0
LU
co
L:
::J
z
2 4 6 7 8 10
NUMBER OF INDIVIDUALS
Figure 7.4 Plot of palaeoecological data showing the number of individuals repre-
senting a species against the number of species represented by that number of
individuals; (0) observed; (e) calculated. For a community collected from the
Lower Carboniferous of Weardale, Co. Durham. Fisher's a= 8.802; no. of spe-
cies = 33; no. of individuals = 304. See text for details.
(b) CALCULATION OF THE DISSOLUTION RATE OF ANHYDRITE
The problem of the dissolution of gypsum and anhydrite in the
foundations of dams and other large hydraulic structures is ex-
tremely important. A study by James & Lupton (1978) has shown
that, although both materials are relatively insoluble, significantly
large amounts of them can be removed by dissolution in periods of
between 50 and I00 years. Such a process, if unchecked, can lead to
structural failure, perhaps of catastrophic proportions. The mathe-
matical solution of one particular equation involved in the calcula-
tion provides a good example of the use of simple iteration. If we
consider the dissolution of anhydrite, then the amount dissolved is
governed by a second-order kinetic reaction, where
dM
-=K(S-c)
dt
This equation can be interpreted as meaning that the rate of

dissolution (dMfdt) is equal to the difference between the concentra-
tion of a saturated solution (S) less the concentration of the final
solution (c), after time t, multiplied by the solubility constant (K).
Integration of this equation gives
KAt_ c
V- S(S-c)
where M is the mass (kg), K is the solubility rate constant

(m4 kg- 1 s- 1), A is the area of mineral exposed (m2), tis the time (s),
Vis the volume of solution (m 3), Sis the concentration of a saturated
solution of the mineral (kg m- 3) and c is the concentration of the
final solution (kg m- 3).
Suppose we let KAtfV= x, and a typical value of x could be 0.12,

then
x = c/S(S- c)
and the recurrence relationship is
C; = x(S(S- C;_ 1)]
If the solubility of anhydrite is 3.5 kg m- 3 and the required accuracy

of the calculation is 0.0001, and if we make an initial guess for c0 = 0,

then
c 1 = 0.12[3.5(3.5- 0)] = 1.47
c2 = 0.12[3.5(3.5- 1.47)] = 0.8526
c3 = 0.12[3.5(3.5 - 0.8526)] = 1.1119
c 11 = 0.12[3.5(3.5- 1.03503)] = 1.03529
c 12 = 0.12[3.5(3.5- 1.03529)] = 1.03518
c 13 = 0.12[3.5(3.5- 1.03518)] = 1.03522
i.e. the concentration of the final solution is 1.0352 x 10- 4 kgm- 3

The result as calculated, although small numerically, is highly
significant, since the constants used to calculate the rate x were
based on relatively small values derived from experimental data. In
particular, because of limitations in acquiring the data, the time
factor is small. Hence, extrapolating to a 'real-life' situation where
the volumes of water circulating are very large and time is measured
in years, the result implies that small fractures in the anhydrite will
be enlarged by dissolution relatively quickly, leaving sizeable voids.
7.4.2 Solution of simultaneous linear equations

It has already been noted that the solution of simultaneous linear
equations is basic to many numerical techniques of interest to
geologists, and a number of examples appear elsewhere in this book.
In Section 4.5.2 we developed the basic mathematics for multiple
regression applied to trend surface analysis and, because the topic of
simultaneous linear equation solution had not been tackled, we left
the matter without looking at an example. We are now in a position
to tackle a simple example.
The data for this problem are presented in Table 7.3, and relate to
the percentage of methane relative to two other light hydrocarbon
gases, ethane and propane, detected in a number of samples of
limestone. These are from the Carboniferous, Lower Namurian,
Great Limestone, in Newlandside Quarry, Nr Stanhope, Co. Dur-
Table 7.3 Percentage methane in samples of limestone col-
lected on a regular grid from a quarry face at Newlandside
Quarry, Nr Stanhope, Co. Durham.
Vertical, Horizontal, Percentage

X y methane
x to-'.
z
0.5 4.5 0.61

0.6 3.6 0.64
0.7 2.9 0.47
1.6 4.5 0.46
1.5 3.6 0.59
1.5 2.8 0.41
2.5 4.4 0.78
2.3 3.6 0.84
3.2 3.6 0.87
3.1 2.8 0.95
3.2 2.0 0.87
3.2 1.4 0.80
4.2 1.2 0.78
4.3 2.0 0.71
4.5 3.8 0.76
ham. They were collected on an approximate 1-m grid on either side

of a vertical joint which has been mineralized. By using the tech-
nique of trend surface analysis and fitting a linear surface to the
data, it is hoped to show the existence of a relationship between the
mineralization and the percentage of methane in the rock. The
hypothesis is that, if any relationship exists, then the linear trend of
the data will be parallel to the joint plane.
The sums, sums of products and sums of squares required to solve

the partial differentials calculated for the data, are
2:>= 36.9 v= 46.7
Iz = 10.54
L.l = 161.63 LXY = 104.34
:Lzx = 27.958 LZY = 32.106

with n = 15 (the number of data points), we can substitute in the

appropriate equations to give:
15.0A + 36.9B + 46.7C = 10.54
36.9A + 116.41B + 104.34C = 27.958
46.7A + 104.34B+ 161.63C= 32.106
The solution of these, using Gaussian elimination, is
c = 0.0105928
B = 0.0835258
A = 0.4633817
Back-substitution into the original equations gives the residual

vector, which is
I
0.04379 i
I
I
R= 0.03072
0.03888
The equations were solved by hand-calculation following the

method outlined in Section 7.3.lc (for inexact coefficients). Check-
ing the solutions obtained, using either Program 7.2 or Program 7.3,
gives closely comparable results, except for the value for A, which
differs by 0.00083; this difference is probably due to using a value of
46.7/15, which cannot be expressed exactly, for one of the second
multipliers. However, since the residuals are small and the determi-
nant large (Program 7.3 gives a value of 4576), the solutions found
are acceptable and can be used to calculate values for the required
surface.
Substituting the values for A, B and C into the equation:
Z(calc) = A + Bx + Cy
we can calculate values for z(caic) for any pair of co-ordinate values x
andy. The results are presented as a linear trend map in Figure 7.5.
5
61 46

(55.3)

(64.5)
78

(71 .9)
4
76
64 S4 87 (88.0)

(55.2) I

(69.4)

(76.9)
3 4 7:
95

I
(55. 3)
_J I (75.2)
<
u
f-
a:: 50/o 80/o 7 1
I.U (84.4)
>
80

(74.6) 78

(82.7)
2
HORIZONTAL
Figure 7.5 Linear trend surface map, calculated using the multiple regression
method discussed in Section 4.5.2. The map represents a small section of a quarry
face, showing a contoured plot of percentage methane (data from Table 7.4). Each
marked sample location gives the original data value and the calculated trend value.
The trend value is in parentheses. , Linear-trend contour line, values
are percentage methane; - - - - - , mineralized joint plane.
From this map it can be seen that although there are discrepancies
between the original data and the surface, the linear trend is almost
parallel to the mineralized joint plane, which fits in with the
hypothesis that there may be a relationship between the mineraliza-
tion and the occurrence of light hydrocarbons in these samples.
8 Geological applications of
differential equations
Having looked at the basic methods of differentiation and integ-

ration in Chapters 4 and 5, we now come to the consideration of
differential equations. Initially we shall give some definitions and
simple examples. Following this, consideration will be given to some
geological examples. As many of the examples chosen are compli-
cated, it is not planned to look at their complete mathematical
derivation or proof, but rather to look at the geological implications
of their use. It is hoped that the references given will be adequate for
those who wish to explore these particular applications further. The
use of differential equations in geology is probably one of the most
important topics in mathematical geology, and this is likely to
continue to be so in the future.
8.1 Definitions and examples
An ordinary differential equation is any relationship between the

variables x andy, and the derivatives dyfdx, d 2yfdr, etc. The term
'ordinary' serves to distinguish them from partial differential equa-
tions, which involve partial derivatives. The order of a differential
equation is that of the highest-order derivative occurring, so
~~ = f(x) is a first-order differential equation
d y + dy = f(x)
2
is a second-order differential equation
dr dx
The solution of a differential equation is an equation relating the

variables involved which does not contain any differential coeffi-
cients. In other words, we integrate to eliminate the differentials:
dy = 3x
dx
3
fl.x) = J3xdx = 2.r + C
DEFINITIONS AND EXAMPLES 217
Note that here we include the constant of integration, which leads to
the general solution. Thus, the equation gives rise to a family of
curves, which will have the same slope for a particular value of x, as
shown in Figure 8.1. If a particular solution is required, then an x-
and corresponding y-value must be given. These are referred to as
boundary conditions, and enable the constant of integration to be
calculated. Thus, in the example we have just considered, if x = 0
when y = 2, then
3
y=-r+2
2
We have already seen that if the equation is in the form
dy =f(x)
dx
then we integrate directly, so
y = .fj(x)dx
y
30
-10
Figure 8.1 The family of curves of the differential equation dyjdx = 3x, i.e.
f(x) = tx' + c for c = - I, 3 and I 0.
218 GEOLOGICAL APPLICATIONS
However, if the equation is of the form
dx = f(y)
dy
then the equation is rewritten in the form
_l_dy =I
f(y)dx
and then integrated
I
J-dy= Jdx
'f(y)
Another possibility is the form
dy _f(x)
dx- g(x)
In this, the variables are separable, so we can rewrite it in the form
g(y) dy = f(x)
dx
in which each side can be integrated separately:
fg(y)dy = .[t(x)dx
An important equation of this type is
dy =ky
dx
k being a constant. We can separate the variables:
_!_dy =k
ydx
which allows us to integrate:
I
f-dy = fkdx
y
NUMERICAL SOLUTION: FINITE DIFFERENCES 219
which leads to
lny= kx+ C
or
y = exp(kx +C)
= expkxexpC
8.2 Numerical solution: the method of finite differences

8.2.1 Differential equations with one independent variable
Starting with basic differentiation, it should be remembered that the
first derivative of a function at any point on the curve is given by
dy = lim./{x + L\x) - j{x)

dx L\x
L\x-+ 0
The derivative at any point x = P, is given by
where E is an error term which approaches zero as, in the limit,

.:\x-+0. Therefore, in the limit we can write
which can be rewritten as:
( dy) :::::,f(x0 +Llx)- f(x 0 )

dx p Llx
If x 1 = x 0 + Llx, x 2 = x 0 + 2.:\x and ./{x1) represents the value of the

function at x 1, etc., then we can rewrite the equation as
This equation can be expressed in words as meaning that the slope of

the tangent to the curve at a point P half-way between x 0 and x 1 is
equal to the difference in value of the function at these points divided
by llx. In the limit, as llx--. 0, the finite-difference quotient (the right-
hand side of the equation) becomes equal to the derivative. It is more
usual to express the derivative in the form of the difference between
points separated by a distance of 2\x. Thus, we can write, for the
derivative at x 0 :
If we wish to obtain the finite-difference quotient for the second

derivative, we proceed similarly. First we introduce a second point
on the curve, Q. Then, as before, the approximation to the first
derivative is
The second derivative can be written as
Thus, we can express the second derivative as an approximation to

the difference between the first derivative at points P and Q:
(d
dr
y)
2 ~ (dy/dx) 0
~
-
lx
(dy/dx)p
0
We replace the derivatives on the right-hand side of the equation by

their finite-difference quotients:
Thus,
We can derive the finite-difference quotients for higher-order deriva-
tives similarly.
To use the finite-difference approximation we can drop the
y = f{x) notation and use values of y themselves. The finite-differ-
ence approximations with the y = j{x) notation replaced by the
corresponding y-values, for differential equations up to order four,
are
Having developed our general finite-difference approximations, how

can we use them to solve any particular differential equation? The
method illustrated has three steps.
( 1) Translate the derivative into a finite-difference quotient.

(2) Formulate a set of simultaneous linear equations which will
enable the unknowns to be evaluated. The number of equations
will equal the number of unknowns.
(3) Solve the simultaneous equations.
We shall now tackle an example. Consider the linear differential

equation
We translate this into a finite-difference quotient:
We are asked to evaluate the differential for the range of values

x = 0-0.5. If we choose an interval of Llx = 0.05, we will have 11

values of x indexed with j, as follows:
These values can now be inserted into the finite-difference equation,

using the same index for y:
Yo- 2y 1 + Y2 = 0.000625
Y1 - 2y 2 + Y3 = 0.00125
Y2- 2y3 + Y4 = 0.001875
Y8 - 2y9 + Yw = 0.005625
which are a set of nine equations with 11 unknowns (y-values). In

order to solve these equations, we need to set boundary conditions.
This will provide two values of y, y 0 and y 10 , which will make the
number of unknowns nine. For the present purpose we will set y 0 = 0
and y 10 = 10.0. The equations will now be
- 2y 1 + y 2 = 0.000625
Y1 - 2y 2 + y 3 = 0.00125
Y2 - 2y 3 + Y4 = 0.001875
=- 9.994375
These equations can be solved iteratively using Gauss-Seidel

iteration (Section 7.3.2), where the general form is
To illustrate the computer solution of the finite-difference equa-
tions, Program 8.1 is given. It should be noted that the boundary
conditions are set in statements 80 and 90, and the general form of
the equation for iteration is in statement 230. Similar examples could
be run by changing these statements appropriately. The results are
summarized in Table 8.1.
Table 8.1 Values obtained for the finite-differ-

ence solution of the equation d 2y/dr = 5x after
99 iterations using Program 8.1. Values marked
with an asterisk are the boundary values.
x(i) YoJ x(i) YoJ
0.00 o.oooo 0.30 5.9596

0.05 0.9895 0.35 6.9625
0.10 1.9797 0.40 7.9698
0.15 2.9712 0.45 8.9820
0.20 3.9646 0.50 10.0000*
0.25 4.9605
8.2.2 Finite-difference solution with more than one

independent variable
In many practical applications there is more than one independent
variable. For example, the flow of a liquid through a porous medium
can be considered in either a two- or a three-dimensional context.
There will, accordingly, be two or three independent variables.
Suppose we have an arbitrary grid, each cell being defined by an
easting of size Llx and a northing of size Lly, and denote easting by
the index i and northing by j. In each cell some parameter of interest
is measured, which we can call z. This value can be represented on a
third axis, and the resulting surface is described by the function
z = J(x,y)
x andy being the independent variables and z the dependent one.

Now, the derivatives of z will be partial derivatives with respect to
either x or y, written as
oz oz
ox and oy
Higher-order partial derivatives can be denoted in the usual way.

Following from Section 8.2.1, we get
OZ = Z;+ l . j - Z;-l,j
ox 2L\.x
and
OZ= z.l,[' + I - z.l,J-
- 1
oy 2Ay
Also, the second partial derivatives are
o
_ z=
2 z.+ 1 .-2z.+z
I l,}
,}l - 1,}
.
of Ax2
and
-
o2z = z .+ 1 - 2z . . + z. 1
oy AI
l,) '' 1,{-
It should be noted that there are now two indices in the equations;
when we hold y constant the index j remains constant and, con-
versely, when xis constant i remains constant.
Now consider the partial differential
where C is some numerical constant. Using our finite-difference

quotients we have
z.+
1
I .-
,)
2z.1.) + z. 1-
I .
,) + z . .+I
1,)
- 2z1,). + z.1,)-
. I = c
A.'~? Ay2
If we use a regular grid where Ax= Ay, then this equation becomes
z.+
l 1,). +z , - 1.j. + .. + 1 + z1,)-
zl,) .. 1- 4z l,). = c(Ax) 2
Inspection of Figure 8.2 shows that there are 25 cells involved, nine
of which are internal (i.e. have no external boundary), four have
external corners and 12 contain an external edge.

i. j 1

i -1, j i,j i +1,j

r - - - -,- - - - ,
I o I

I 0
I e e 1
L
z !j) i, j -1 : I :
t- - - - -- - - - -.
I o
e II e e '
I
- -- - _,_ - - - -'
I I I
I 0 I
x (o)
Figure 8.2 Method of indexing points for the finite-difference quotients, with two
independent variables. Points are at the centre of a regular grid of .1.x = Ay.
We can arbitrarily set our boundary condition at zero, in which

case all cells containing external edges or comers have the boundary
value
z 11 = zl.2 = z1,3 = ... = 0.0
and we have to find values for the nine internal cells:
The equations are

Rearranging,
which can be solved using the Gauss-Seidel method. A good

exposition of this method, which is commonly applied to problems
of groundwater flow and associated areas is given in a number of
books. These include Harbaugh & Bonham-Carter (1970), Verruijt
(1970) and Wang & Anderson (1982).
As mentioned at the end of the previous section, differential

equations in geology are commonly in use in the study of ground-
water flow. Therefore, since there are a number of good texts on this
application (see references at end of last section), the applications to
be considered now are taken from other areas.
8.3.1 The exponential-growth equation

The exponential-growth equation is one of the simplest differential
equations, yet has had the widest application throughout science.
Geology is no exception, and in this section only a few of its many
applications are reviewed.
There are two forms of the equation, one relating to growth
dN
-=rN
dt
and the other, relating to decay
dN
- = -rN
dt
where N =the population size, t =time and r =constant. Both
equations can be solved analytically, and lead to
and
where N, = the size of the population at time t and N 0 = the size of

population at time zero. The equations may also be written as
InN,= lnN0 + rt
and
lnN, = lnN0 - rt
As we saw in Section 2.1.4, a graph of population size against time,

according to these equations, has a characteristic exponential form.
(a) THE RATE LAW FOR FIRST-ORDER CHEMICAL REACTIONS

The exponential equation is used to describe the rate of change in the
amounts of reactants in first-order chemical reactions. The equa-
tions are either
dC = -kC
dt
InC= In C0 - kt
where C0 = the concentration at the start of the reaction, k = the

reaction rate constant, C =the concentration at time t, during the
reaction, and t = time
or
dC = k(C- C)
dt s
In c -c)
(C,- C 0
= - kt
where C, = the steady-state concentration, other symbols as above

(Lasaga 1981, p. 28).
In a laboratory study of the thermal conversion of lipids extracted
from sedimentary rocks into a kerogen-like substance, Shioya and
Ishiwatari (1983, p. II) give the following in the interpretation of
their results:
In order to estimate the rate of conversion of soluble lipids into a

kerogen-like substance, the lipids were heated at 6 different
temperatures, in the range 125-200"C for periods ranging from I
to 7 days. At the end of each experiment the residual lipids, any
gases generated and the kerogen-like substance formed, were
measured. Assuming a first order reaction:
dA=kS
dt
where A = ratio of the percentage kerogen-like substance to

original lipids
S = ratio of the percentage residual lipids to original lipids
Since no gases were generated for the range of temperatures

chosen, the amount of kerogen generated equals the amount of
lipids which disappeared. Therefore:
combining the two equations:
InS= - kt + In I 00
The value of k, the rate constant for each of the 6 different

temperatures used in the experiments, were estimated from plots
of InS against time. By plotting these constants against tempera-
ture, they concluded that the half-life of lipids in sediments is
about 104-10 5 years at 0-3oc.
(b) DISTRIBUTION OF SEDIMENTARY ROCK MASSES

Garrels and Mackenzie (1971, pp. 255ff.) developed two simple mass
balance models which can be used to help describe the distribution
through time of sedimentary rock masses preserved in the Earth's
crust. These are the constant-mass model and the linear-accumula-
tion model (Garrels and Mackenzie 1972, p. 262). The two models
have different starting points. The first assumes that the total mass
of sediment has remained constant throughout geological time, and
that there has been a balance between the metamorphism of
sediments and the erosion of igneous rock, whether of primary or
secondary origin. The implication of the constant-mass model is that
at some time early in the Earth's history a large mass of sediment
was formed almost instantaneously. The linear-accumulation
model, on the other hand, assumes that the erosion of igneous rock
has exceeded the rate of incorporation of sediments into the igneous
mass. For the present purposes, only the constant-mass model will
be considered.
If we start with a large mass of sediment M, at time t 0 , then the
processes of erosion and the incorporation of material back into the
igneous mass will commence, so that by time t 1 the original mass will
have decreased by an amount !!Me. However, since the process of
deposition will also be taking place, a 'new' set of sediments will be
deposited, equal in amount to !:J.Md. At time t 2 both 'old' and 'new'
sediments will be affected by the processes in operation, giving rise
to 'new' igneous rock and 'new' sediments. However, throughout
the whole process, the total amount of sedimentary rock (of all ages)
remains the same.
Harbaugh & Bonham-Carter (1970, p. 367) have formalized this
model, suggesting that the amount of material lost by erosion, etc.,
per unit time can be described by the exponential decay equation
dM
--=aM
dt
where M = the mass of sediment, t = time and a= a constant

controlling the loss of sediment. This gives
where M 0 = the starting mass and M, = the mass remaining at time

t. Now, since the amount of sediment deposited is equal to the mass
removed, the amount deposited is given by
=Mo-Me-a'
where Md = the amount deposited.

If we divide geological time into a number of equal intervals, each
!:J..t long, then this equation becomes
where M = the total mass of sediment.

If geological time is divided into N equal parts, and we use the
subscript i to label each increment, with i = I being the oldest, then
the age of rocks being deposited during the ith time increment is
(N- l)!:J..t
The mass deposited during any time increment is Md. The mass
remaining today M,, from sediments deposited at time i, is
Thus, given values for M, a, N and lit, we can calculate the mass
remaining for values of i, where i goes from 1 to N. A Fortran
program based on both of the Garrels and Mackenzie models is
given by Harbaugh & Bonham-Carter (1970, pp. 367tf.). They also
include a comprehensive discussion of the mathematics of the two
models.
8.3.2 Gravity measurement

In geophysics the use of the torsion balance to measure gravity is of
interest, since, although its use has been superseded by more-modern
instruments, its theory is important. The torsion balance measure~
the distortion of the gravitational field, by measuring the rotation of
a torsion balance beam, due to the gravitational forces acting on two
masses at the ends of the beam. Since these forces which produce the
torque in the beam are first derivatives of gravity, their rates of
change must be second derivatives of the potential (Nettleton 1976,
pp. 67ff.). The torque is given by
T= mhl [ (a~;z) cos a- (a~;z) sin a J

+ ml 2 [ 2 ( o 2
axay o
u ) cos 2a + ( (Jy2
2
iJ2u ) sm
u - ox2 . 2a J
where m =the mass of weights on the beam, h =the vertical distance
between the weights, 2/ = the horizontal distance between the
weights, T = the torque, a = the angle between the beam and the x-
axis (usually north) and o2ujoydz, etc. = the 2nd partial derivatives
of the gravitational potential u.
Since the torque produces a rotation of the balance, and if the
deflected position is n and the undisturbed position is n0 , then the
equation becomes
n- n0 = P(uyz cos a- uzx sin a)+ Q(2u_.Y cos 2a + ul'l sin 2a)
where P = 2NDmhl/-r and Q = 2NDml 2 j-r. P and Q are constants,
and depend on the physical parameters of the instrument,
and ul'l = uYY- uxx.

There are four unknown quantities in the equation, which are the
four second derivatives. Also, as there is no way of measuring n0 , so
that is also an unknown. In this application it is not necessary to
solve the differentials and, in practice, readings are taken in five
different directions (i.e. five values of a), which reduces the problem
to solving five simultaneous linear equations in five unknowns.
8.3.3 Diffusion of gases through rock

The use of organic geochemistry as a method of oil exploration
depends on locating anomalous amounts of the light hydrocarbon
gases methane, ethane and propane in samples of subsoil or near-
surface rock. The occurrence of such anomalous amounts of light

hydrocarbon gases are assumed to be caused by the diffusion of the
gases from gas or oil reservoirs. The model is based on the premise
that the gases 'leak' through the reservoir cap rock and diffuse
upward, and are approximately at equilibrium, at a depth of 2-3 m
from the surface. They enhance the naturally occurring gases
already present in the rock, and form an anomaly significantly above
the regional background (Horvitz 1939).
Subsequent to Horvitz's paper, there has been much discussion on
the assumptions behind the model, and consequently on its validity
as an exploration technique. A paper by Stegna (1961) tested the
validity of the assumption that gases could diffuse through the
overlying rock by developing a theoretical model. This model
quantifies the process of diffusion to see, under what conditions of
depth and time, steady-state diffusion could occur when taking into
account the nature of the rocks likely to be found above a reservoir.
Stegna (1961, p. 447), started by taking the Fick's Law equation
for diffusion as:
oc
at =DVC
which simply states that the change of concentration C with time tat
any given point is equal to the Laplacian of the concentration, VC,
times the coefficient of diffusion, D. To solve the differential
equation, so that values for the concentration at a given depth z at
time t could be found, boundary conditions are set using the
following assumptions.
(a) Reservoir formation was instantaneous, at
timet= 0
depth z = h
(b) The gas concentration in the reservoir remains constant during

time, i.e. the amount of gas lost by diffusion is very small
compared with the amount present.
(c) At the surface z = 0 and the concentration of gas from diffusion
is small enough to be neglected.
(d) The reservoir has a sedimentary cover of average diffusion
coefficient D.
(e) The lateral extent of the reservoir is significantly greater than its
depth. This means that the one-dimensional equation
can be used.
These assumptions, which are reasonably realistic, give the bound-

ary conditions
C(O,z) = 0
The equation and boundary conditions are satisfied oy the function
This does not, however, satisfy the first boundary condition. There-
fore Stegna (1961, p. 448), assumed that the concentration at timet
and depth z consisted of two parts:
C(t.zl = C'(t.zl + C"(l.z>

and the boundary conditions for the function C"(l.z)' are
C"(O,z) = - cO,z/h
C"(t.hl = 0
C"(I,O) = 0
and a particular solution is
C"(l.z) = A.e- alDt sin az
which will satisfy the third of the above boundary conditions, but
not the second, unless
a=nrc/h
where n is an integer. To satisfy the first condition, the constants A"

are chosen so that
n~l
co
An Stn h
(mt) Z = - C0
(z)fz
This leads to an equation which allows the calculation of the gas
concentration Cat a depth z and at time t, which is
C (t.z) = Co (:.
h
+ ~ L..
~ (- 1)"
e
-(n21tz/irl)Dt
sm h
n7tZ)
1tn=l n
where C0 =the concentration of absorbed gas (methane) in the

immediate vicinity of the reservoir, and is taken to be 0.1 (Stegna
1961) and D =the diffusion coefficient, and is taken as being
between 10- 6 and w- 5 cm 2 s- 1
An average diffusion coefficient can be calculated from
where a;= the thickness of the ith lithology and D; = the diffusion
coefficient of the ith lithology.
It is interesting to note that Stegna concluded that in some
situations sufficient time could not have elapsed from the formation
of the reservoir to the present for steady-state diffusion to occur.
Also, he argued that since the time required to reach steady-state
diffusion varies with the square of the depth 'the geochemical
method is a structure-prospecting rather than a direct oil-prospect-
ing method' (Stegna 1961, p. 450).
This example is only one of many ways of solving the differential
equation for Fick's Law. Harbaugh & Bonham-Carter (1970,
pp. 238-58) look at several examples with different objectives, and
hence different boundary conditions. Thus, if the material through
which diffusion takes place is anisotropic then the diffusion coeffi-
cients kx, kY and k=, for each direction x, y and z is
OC = k o2 c + k o2 c + k iJ2c
ot x8X2 yay =a?
which can be solved by finite differences.
8.3.4 Formation of crenulation cleavage
This example follows closely the arguments and discussion of
Cosgrove (1976), with some reference to the earlier work of Cobbold
eta/. (1971). There have been two proposed origins for crenulation
cleavage (Cosgrove 1976, p. 157): either it is the result ofmicrofault-
ing or the result of microfolding. Both types are well represented
geologically, along with a third type - crenulation cleavage in
conjugate sets. This last type has been considered as a consequence
of microfaulting. Cosgrove (1976) demonstrated that these three
types of crenulation cleavage are different expressions of buckling
instabilities developed in anisotropic materials.
Certain assumptions were made:
(a) the material is linearly elastic or Newtonian viscous;

(b) it is anisotropic and homogeneous;
(c) it deforms by plane strain;
(d) deformation is not accompanied by a volume change;
(e) the maximum principal compressive stress is parallel to the
mineral fabric; and
(f) it is in a uniform state of initial stress which gives rise to
homogeneous flattening within the material.
The co-ordinate directions x and y are chosen to coincide with the

symmetry axes of the material, as shown in Figure 8.3.
Figure 8.3 The relationship between foliation and the co-ordinate axes x and y
chosen for the mathematical analysis of the formation of crenulation cleavage (after
Cosgrove 1976, fig. 4).
The aim of Cosgrove's analysis is to investigate the types of

instability which might disrupt the initial state. The displacement
patterns which result are ordered and non-uniform, and are des-
cribed by displacement vectors U and V which are in the x- and y-
directions, respectively. By restating the above conditions mathema-
tically a single equation has been developed, solutions to which
satisfy the conditions specified. The buckling equation is a partial
differential equation of fourth order:
(Q - P/2)~ + 2(2N- Q) 0 ~:~ + (Q + P/2) + ~ = 0
where <p is a displacement function related to U and V:
u= o<pxy v= o<pxy
ay ox
where P =the differential stress (cr 1 - cr 3), Q =a measure of the
resistance to shear and N = a measure of the resistance to compres-
sion in the direction of the maximum compressive stress cr 1
The shear and compressive moduli M and L for any stress
condition can be given in terms of Q and N, as
M= N+ P/4 L= Q+P/2
Their ratio M/L is a measure of anisotropy, and when MandL are

very different, i.e. the ratio is significantly greater than or less than
unity, the material is highly anisotropic.
The general solution of the partial differential (Cobbold et a/.
1971, p. 29) is
<p = J;(x + ~nY) + fz(x + ~nY) + J;(x + ~zY) +fix+ ~z.Y)

where ~ 1 and ~ 2 are arbitrary constants, and J;, fz, J; and h are
arbitrary functions.
To obtain specific solutions (i.e. displacement patterns) of the
general equation, boundary conditions need to be defined. These
boundary conditions are such that there is no deflection of shear
stress at the edges of the material, which implies that buckling
disturbances are restricted to the central part of the material. As
Cosgrove (1976, p. 158) notes, in geological situations buckling is
not restricted to these boundary conditions. However, the simplifi-
cation is valid, and leads to the three types of the displacement
function <p which are of geological interest. Cosgrove (1976, p. 159)
has designated these as types I, 2 and 3, as follows
type I 2M>L
where the stress conditions are
P>L
type 2 2M<L
where the stress conditions are such that
4(M/L)(L- M) < P < L
type 3 2M<L
where the stress conditions are
P>L
From these it is easy to see that the type of buckling which develops
depends on the material properties, M and L, and the size of the
differential stress, P. It should be noted that:
(a) type 3 buckling solutions develop in the same types of materials

as type 2, but with a higher stress;
(b) structures associated with type 2 will develop before P is large
enough to develop type 3 structures: the previous development
of type 2 structures will generally inhibit the development of
type 3 structures; and
(c) the displacement pattern for type I and type 3 solutions is the
same.
In general, type I structures develop in materials with low aniso-

tropy, whereas type 2 structures develop in materials with high
anisotropy.
9 Beyond traditional
mathematics
In common with most other subjects, mathematics is making

advances in many areas, some of which are totally new or are new
developments of traditional mathematics. In the case of such
advancements, those involved may have applications in mind be-
yond the theoretical, but as in many fields of human activity there is
frequently a marked time-lag between theoretical development and
application. In this final chapter two topics have been chosen which
have had some application in geology. They are here as a reflection
of the author's interest, rather than their importance in absolute
terms.
The advantage of both is that they provide a framework within
which to re-think some old problems, rather than producing imme-
diate new answers. With this in mind, it is interesting to speculate on
what might have happened if, for example, one of the topics,
catastrophe theory, had been around when Darwin was developing
his theory of evolution. Catastrophe theory is concerned with the
mathematics of systems which, although they can show gradual
changes from one state to another, can also jump 'catastrophically',
without going through the intermediate stages. Such ideas give
credence to a theory of evolution in which change is not always
continuous, a factor which might explain why palaeontologists have
had so little success in their search for the so-called missing links.
In a book such as this it is not possible to go into the theory of the
methods in great detail. It is possible, though, to give sufficient
background to stimulate interest in the topics that the developments
already made, toward their application to the study of geology, are
not neglected. It is also worth bearing in mind that they are not the
only possibilities which the new mathematics has to offer.
9.1 Catastrophe theory
Catastrophe theory has been principally developed by Thorn (1972),

with notable contributions by Zeeman (see, for example, Zeeman
1976, for a readable and comprehensive account of the development
CATASTROPHE THEORY 239
of the topic to that date). The subject concerns the study of sudden
changes in a system, from a state of equilibrium. The theory derives
from differential topology, which is concerned with the properties of
surfaces in many dimensions. Topology allows the qualitative
constructions of models which closely mirror the behaviour of many
natural systems. When such surfaces are smooth, they are reflecting
a state of equilibrium, when they are not smooth it indicates that the
state of equilibrium has broken down, and catastrophes occur. At
present the model is qualitative rather than quantitative.
Since it is difficult to describe all possible equilibrium surfaces and
their associated catastrophes, matters have been simplified by the
recognition of elementary forms. For processes where there are no
more than four controlling factors, it has been shown that there are
seven elementary catastrophes. These can be understood and ap-
plied to problems in sciences such as geology. In geology two papers
are of note, those by Henley (1976) and Cubitt & Shaw (1976); in
other branches of science notable contributions to the theory of
evolution have been made by Waddington (1974) and Woodcock &
Poston (1974). The discussion which follows is largely based on the
references already cited, with the addition of Saunders (1980).
9.1.1 General theory

Considering catastrophe theory in its simplest form, we can say that
it will relate to any system which can be described by an energy
function of the form
where V = the energy (or potential) function, Q; = the finite set of

state variables a, b, c, ... ; i ranges from 1 to n: these control the
system in the form of generalized co-ordinates and IJ =the finite set
of state variables x, y, z, .... ; j ranges from 1 to k: these are
externally controlled parameters which describe the behaviour of the
system.
Stationary values of V with respect to Q; will give rise to
equilibrium, maximum values giving unstable equilibria whereas
minimum values give rise to stability. In other words, maximum
energy represents unstable equilibrium and minimum energy repre-
sents stable equilibrium. In general the positions and number of
equilibria are affected by variation of the control variables a, b, c,
etc. If we limit the control space to four dimensions (a, b, c, d), then a
240 BEYOND TRADITIONAL MATHEMATICS
smooth reversible change in the co-ordinates in the neighbourhood

of a given point, will produce only one of seven possible elementary
catastrophes. The number of elementary catastrophes is dependent
only on the number of dimensions of the control space (e.g. for two
dimensions only two are possible), whereas the type of catastrophe is
controlled by the energy function.
The simple 'cusp' catastrophe, described by Zeeman ( 1971 ),
requires a two-dimensional control space a and b, and the energy
function is
Since the first derivative with respect to x is a cubic, i.e.
dvjdx = _x-.3 +ax+ b
which will have either one or three real roots, it follows that there
will be either one or three equilibrium positions for any given pair of
values a, b. If for any position in the control space there is only one
equilibrium position which is stable, there will be an energy mini-
mum. Where there are three positions, the upper and lower are
minima and the intermediate will be a maxima. The maxima will
give rise to an unstable equilibrium.
The state of equilibrium is maintained on a topological surface
either by smooth changes or by catastrophic jumps. The theory can
be used to model and predict features of stable and unstable
equilibrium. Zeeman (1971) has suggested that the catastrophe
model could be appropriate if the system exhibits one or more of the
following:
(a) bimodality;
(b) divergence;
(c) catastrophic jumps; or
(d) hysteresis delays.
Before considering some geological applications of the theory, we

shall look at a simple physical model or machine, which shows all of
the phenomena associated with the theory, known as Zeeman's
catastrophe machine, and is illustrated in Figure 9.1. Most authors
who have written on this topic have suggested that the machine,
which is easily constructed, should be made and used to carry out
the experiments described. This view is iterated here.
BASE z
0
f-
<(f-
uw
a:::Vl
:J
Figure 9.1 A diagrammatic representation of Zeeman's catastrophe machine.
The machine consists of a cardboard disc whose diameter is equal

to the unstretched length of a rubber band. The disc is fixed to a
suitable base, with a pin acting as a pivot at its centre, so that it is
free to rotate. The rubber band which was used to establish the
diameter of the disc and another band of equal length are attached
at some point near the perimeter of the disc as in Figure 9 .1. The free
end of one band is fixed to the base at a distance of two diameters
from the disc pivot and in line with it. The machine is operated by
using the free end of the other band as the control point and moving
it in the plane of the disc. Movements of the control point generally
cause a smooth rotation of the disc, but in some situations the disc
jumps. If all of the positions at which sudden jumps occur are
marked on the base, a diamond shaped outline is generated (only
half the area is shown on the figure). The outline or curve has four
cusps, which form the bifurcation set of the cusp catastrophe
equilibrium position. If the control point lies inside the curve there
are two stable equilibria, if it is outside there is only one. The sudden
movement of the disc relates to movement from one equilibrium
position to another. This movement occurs at the point B if the
control point moves from A to B as in the diagram; or at A, if the
movement is in the opposite sense.
The phenomena exhibited by this machine can be visualized by
means of a three-dimensional diagram, as illustrated in Figure 9.2.
This model represents the cusp catastrophe, the simplest of the seven
elementary catastrophes. The diagram is produced, constructing a
behaviour surface over one segment of the bifurcation set, by
plotting the points for which the first derivative d V/dx = 0. Every
point on the top and bottom sheets of the behaviour surface (i.e. the
part of the surface directly over the bifurcation set) represents a
position of the disc where energy is a minimum, for the correspond-
ing position of the control point. Therefore, if the control point is
outside the bifurcation set only one mode of behaviour is possible,
Figure 9.2 The cusp catastrophe model for Zeeman's catastrophe machine.
whereas inside the bifurcation set two modes are possible. These
correspond to the two stable positions of the disc with two local
minima (one on the top sheet and one on the bottom), even though
the behaviour surface at this point is overfolded and has three
sheets. This follows since the middle sheet, corresponding to a local
maximum, is made up of points representing the least probable
behaviour of the system. These factors can be more easily under-
stood by plotting the energy function V(x) for values of u < 0 for
different values of v (the position of the control point). A set of plots
of the energy state are shown in Figure 9.3. The position of the ball
in the diagram shows, in a pictorial manner, how the catastrophe
occurs.
Returning to our discussion of Figure 9.2, the inclusion of the
middle sheet ensures that the behaviour surface is smooth and
continuous; however, it should be noted that the control point is
never represented on the middle sheet. Whenever the control point
crosses the fold, either at A or at B, the point jumps between the top
and bottom sheets catastrophically, since the middle sheet is inac-
cessible. The fold curve, marking the boundary between sheets,
represents the inflection points. This sort of behaviour is characteris-
tic of physical systems in which friction is important, as it allows
them to assume a state of minimum energy. In the case of Zeeman's
catastrophe machine, the energy is the potential energy stored in the
rubber bands, so the disc will rotate until the tension of the rubber
bands is minimized. Unless energy is added to the system, the
machine will remain in a position of stable equilibrium. The process
that keeps it there is called the dynamic.
u
lv \:
v
~
X
V
J I c) I d I \ l \. I
V W ~1um~
e f \J.
Figure 9.3 Energy states across the bifurcation set for Zeeman's catastrophe
machine (u < 0 is fixed, values of v vary: a, b, c, etc.).
The behaviour associated with the cusp farthest away from the
disc can be explained in a similar fashion, except that the energy is
inverted, i.e. there are two points with maximum energy and only
one where there is a minimum. As before, the dynamic compels the
machine to remain in the single stable position of minimum energy.
Also, on the behaviour surface the positions of maxima and minima
are reversed and, as the maxima are unstable, the behaviour point
can only lie on the middle sheet. This phenomena is known as a
dual-cusp catastrophe.
Features of this system, which are common to all systems that
follow the cusp model, are:
(a) bimodality over part of the range,

(b) hysteresis since the transition from one sheet to the other occurs
at two positions, and does not occur in the middle of the cusp
and
(c) divergence due to a small perturbation in the initial state of the
system can lead to large differences in the final state.
Since each of these characteristics is present in a number of

important geological systems, we are now in a position to look at
some applications.

For the purposes of illustration, two topics have been chosen. One
relating to faulting, which is based on Henley (1976, pp. 653ff.); the
other relates to evolution.
If we consider faults which are currently active, then we come to
the inescapable conclusion that, although the effect of their move-
ment is catastrophic in human terms, they are also catastrophic in
the sense of the foregoing discussion. There are quite long periods of
quiescence (hysteresis delays), followed by very sudden movements,
often of great magnitude (catastrophic jumps). These factors, as we
have seen, suggest that the catastrophe model might be appropriate.
We can set up a qualitative model of the system as indicated in
Figure 9.4a. The control variables are friction and shearing energy,
while the response variable is the rate of movement. If the values for
friction remain low, then it will require a small amount of strain
energy to initiate movement. Also, if the strain energy increases, so
will the movement. At high friction levels, which presumably occur
in nature, increased energy will not cause movement until some
threshold is reached, at which point there will be sudden jump to
high rates of movement, following the path ABCD on the figure.
Since the energy is released, due to the movement, there will be a
second jump, back to quiescence, as indicated by the path DEB.
Analysing the behaviour of a fault in this fashion shows that the
catastrophe model provides a reasonable quantitative interpretation
of reality.
To model evolution we can use as our control variables the
increase in variability of the organism and the increase in environ-
mental pressure. These factors are plotted as cusp catastrophe in
Figure 9.4b, the interaction surface being used to define the relation-
ship between the two control variables. Thus, at point A variability
is low but the environmental pressure is high. At point B variability
is high and environmental pressure is low. Now, Darwin's Theory in
its simplest form tells us that evolution occurs as a response to
environmental pressure, and that those organisms which can adapt
quickly can be expected to survive.
Using Figure 9.4b let us explore what might happen using the
model. Starting at a point A, where both environmental pressure
and variability is low, and moving toward B, since there is only an
increase in the potential toward variability, there will be a tendency
to remain the same. If, on the other hand, variability remains low
and environmental pressure increases, movement will be towards C
ressure
are a of
catastrophic
evolution
Figure 9.4 (a) The cusp catastrophe model applied to fault dynamics (after Henley
1976). (b) The cusp catastrophe model applied to evolution. The model relates
variability of the organism to environmental pressure.
and possible extinction. The remaining option is that as environ-

mental pressure increases, variability also increases, then there is the
possibility that the path will lead smoothly toward evolution (solid
line in figure), or with a small perturbation, the path will be diagonal
across the surface to a point where a catastrophic jump is possible
(broken line on figure). Thus, the analysis seems to indicate the
possibility of evolution occurring as an almost continuous process
or as catastrophic jumps, these jumps perhaps being the reason why

no intermediate forms are found for some lines of evolution.
9.2 Information theory
The basic tenets of information theory were largely developed by

Shannon and his co-workers in the Bell Telephone Laboratories in
the USA. The mathematical theory of communication was pub-
lished by Shannon (1948), and subsequently a large number of
applications of the theory have been made in many fields outside
communications, including geology. More recently a number of text
books on the topic have been published, perhaps the most compre-
hensive being that of Guiasu (1977). In geology, applications have
been published by Pelto (1954), Tasch (1965, 1980), Beerbower &
Jordan (1960) and Ferguson (1980, 1982, 1983).
Shannon's original work was concerned with designing communi-
cations systems which allow the transmission of selected informa-
tion from one point to another, where it could be reproduced almost
exactly. The mathematical theory developed allows the definition
and quantification of information both at the source and destina-
tion, thus giving an insight into what occurs in the channel used for
transmission. From these it is possible to predict the behaviour of
communications systems mathematically.
A simple communication system can be considered as follows:
information-+ transmitter-+ channel-+ receiver-+ destination

i
nmse
This can be easily adapted in a number of situations of interest to

geologists. Let us consider, for example, the process of fossilization.
The following schematic is after Tasch (1965):
fossilization collection
living organism fossil interpretation
i i
noise noise
In this, the information source is the living organism and the

received message is the fossil. Noise is added in many parts of the
system; for example, the effects of tectonism or poor collection (i.e.
INFORMATION THEORY 247
the fossil is damaged during its extraction from the rock). These
factors can produce distortion of the information, while the loss of
soft parts and damage can cause an irretrievable loss of information.
Another example is given by considering the problem of produc-
ing a map on the basis of limited data. In this case the use of either a
statistical or mathematical model is resorted to. We can use the
following schematic to represent the process:
mathematical model
information source map

(measurement) i
noise
Here the information is transmitted via the mathematical model

which, by its very nature, will impose its own characteristic structure
on the data, which can be considered as noise. The destination is the
map which is interpreted by the geologist.
Other concepts of information theory have also been applied in
geology. Pelto (1954) used the concept of redundancy in the map-
ping of multi-component lithological systems. Beerbower & Jordan
(1960) also used the same concept in an attempt to produce a
taxonomic diversity index. In the next section we will consider the
basic mathematics of the theory before exploring some of the
applications in more detail.
9.2.1 General theory

The measure of information proposed by Shannon (1948, p. 380) is
the bit or binary digit. Thus, if we have a device such as a switch,
which has two stable positions, on or off, then
log 2 2 = 1.0 bit
Any number of such devices can store N bits of information and, as

the total number of possible states is 2N, then
log 2 2N = n bits
There is a close relationship between information and uncertainty,

so the more uncertain an answer is, then the more information is
passed when the answer is given. If there are a number of alterna-
tives, then the amount of information is maximized when each is

likely to occur. In other words, we have maximum uncertainty.
If we consider a system with N alternatives, and the probabilities
of each alternative is given by
then the information H, is given by
n
H= -K L P; log2P;
i=l
In this equation K amounts to a choice of unit of measurement

(Shannon 1948, p. 393, Guiasu 1977, pp. Iff.). The equation is
central to information theory, and relates information to choice and
uncertainty. His the entropy of the system.
If we consider the case of two alternatives, each being equally
likely, then
H = -0.5 log2 0.5 + 0.5 log2 0.5

= -1.0
If pi= I and p 2 = 0 (or vice versa), then
H=O
or, if Pi= 0.25 and p 2 = 0.75 (or vice versa), then
H=0.8
Thus, His maximized when Pi = p 2 = p 3 = ... = Pn and His equal to

Iog2 n
The ratio
n
L P;log2pi
i=i
is known as the relative entropy (RE) (Shannon 1948, p. 398).

In information theory the idea of redundancy is simply concerned
with the repetition of the same message. Thus, if we haven possible
messages and only one is transmitted, say 10 times, then assuming
the channel is noiseless, only the first transmission of the message
has any meaning, the remainder merely repeat the first and are
therefore redundant. Redundancy (R), is measured as
R=l-RE
or, as a percentage,
R% = (1 - RE) X 100
The concept of redundancy is of considerable importance in the

study of languages; written English has a redundancy of about 50
per cent, which means that when English is written half is deter-
mined by the structure of the language and half is freely chosen.
There is a close link between redundancy and crossword puzzles. If
the redundancy is low, then large crosswords become difficult to
construct. A language with a redundancy of 33 per cent should
theoretically enable the construction of three-dimensional puzzles
(Shannon 1948, p. 399, Guiasu 1977, p. 130).
So far we have only considered single, discrete sources of informa-
tion. Let us suppose we have two dependent events x andy, with m
possibilities for x and n for y. If the joint probability of occurrence is
p(i,j), where i refers to the first and j to the second event, then
H(x,y) = -:E. p(i,j) log 2 p(i,j)

1,)
(Shannon 1948, pp. 394ff.) and
H(x) = - :E p(i,j) log 2 :E p(i,j)

1,) 1
H(y) = - :E p(i,j) log 2 :E p(i,j)

l,J l
Then
H(x,y) ~ H(x) + H(y)

It can also be shown (Shannon 1948, p. 395) that, if the two events
are independent, then
H(x,y) = H(x) + H(y)

In other words, for two or more dependent events, the information

content of the joint event is less than or equal to the sum of the
individual information contents, while for two or more independent
events the information content of the joint event is equal to the sum
of the individual information contents.
Finally, another useful and important property of the entropy
equation is its use in the case of successive choices. To illustrate this,
let us suppose we have three choices with associated probabilities of
0.5, 0.333 and 0.167. Then, if we can breakdown the choice further
into two successive choices, such that we have equal probabilities for
both, and then for one we have two further choices with probabili-
ties of 0.667 and 0.333, then
H(0.5, 0.333, 0.167) = H(0.5, 0.5) + 0.5H(0.667, 0.333)
In other words, the original His equal to the weighted sum of the Hs
for the successive choices (Shannon 1948, p. 393, Guiasu 1977,
pp. 57ff.).

(a)THE PROCESS OF FOSSILIZATION AS A COMMUNICATION
SYSTEM
In an early, thought-provoking paper, Tasch (1965) attempted to
define what happens in the process from living organism to collected
fossil, in terms of a communication system. He showed that the
amount of information passed (denoted by Sr) was a combination of
the original information (S) less the loss due to fossilization, i.e. the
loss of soft parts, loss due to distortion, breakage, chemical
alteration, etc. (denoted by N), plus information gained by multiple
repetition of the message, i.e. lots of individuals preserved (R) and
finally additional energy input (Sc) because of the requirement to
transform one information pattern to another. Thus, the equation
linking these individual attributes is
Sr = [S- D- N] + R + Sc (1)
There is also an additional input of information (St) which is derived

from other areas of study such as sedimentology, stratigraphy, etc.,
which adds to Sr to give enlarged data (Dd). Thus, we have
Sr + Dd = S' = k(S- D) (2)

where the maximum possible data retrieval after fossilization is S',
which is approximately equal to the original information in the
living animal (S) allowing for the irretrievable loss (D). The deriva-
tive data (Dd), Eqn 2, serve to compensate in part for the losses D
and N, Eqn I. Also, symmetry and consistency of organization allow
partial compensation for the effects of N, if N is distortion or
breakage. This is summarized in Figure 9.5.
Output
Sr + Dd at te
Input = s ORGANISM BURIAL

Irretrievable
at t 0 Loss D
Time, n x 10" yrs
Zoological Sedimentological Stratigraphical Ecological
Figure 9.5 Summary of the fossil communication system (after Tasch 1965).
(b) DIVERSITY INDICES IN PALAEOECOLOGICAL STUDIES

Invertebrate community studies have frequently led to attempts to
express the relationships between the number of organisms living in
the community and the number of species represented, as a single
number: the diversity index. We have already met one such diversity
index, Fisher's a, in Chapter 7. Beerbower & Jordan (1960) pro-
posed the use of the entropy equation for this purpose. Their
equation (p. 1185) is
H = I; = - I: P; In P;
where H (the diversity index) = 0 when P; = 1 and H = In k when

p 1 = p 2 = p 3 = ... = Pk That is, the maximum diversity possible is
Ink where k = the number of species represented in the community.
They also showed how the sample diversity related to the population
diversity as well as the use of statistical tests to compare indices. For
further details the reader is referred to the original paper.
252 BEYOND TRADmONAL MATHEMATICS
In a book concerned with the use of mathematical methods in

recent ecology, Pielou (1969, pp. 230-l) extended the ideas of
Beerbower & Jordan and, using Shannon's concept of weighted
entropy, proposed the use of generic and specific diversity. Generic
diversity is defined as
where pj (j = I, ... , g) is the proportion of individuals in the jth

genus. The specific diversity for a genus:
where qj.k (k = I, ... , S) is the proportion of individuals in the kth

species when there are Sj species in the jth genus. The specific
diversity for the whole population is given by
In words, the specific diversity for the whole population is the

entropy of the weights plus the sum of the weighted entropy for each
species.
(c) MEASURES OF MIXING

In what is probably the earliest application of information theory in
geology, Pelto (1954) used relative entropy as a measure of mixing in
a sedimentological study. This idea was further developed by R. L.
Miller & Khan (1962, pp. 425ff.) and applied to data relating to the
distribution of recent foraminifera (R. L. Miller & Khan 1962, p.
427), and to the sorting of sediments (R. L. Miller & Khan 1962, pp.
432ff.). As a result of this study they suggested that the use might be
extended into micropalaeontology, as a mixing descriptor. More
recently, Ferguson (1983) suggested the use of the measure as a
coefficient of elimination in the analysis of C-H-0 data in the study
of kerogen (kerogen is the product of the diagenesis of organic
matter in rock, and its characteristics are fundamental to the study
of petroleum source rocks).
In this application it was suggested that the redundancy equation
could be used to relate the proportions of the three elements carbon,
hydrogen and oxygen, which are the principal constituent elements
of kerogen, the equation is
fll.lrORMATION THEORY 253

coeffi ctent o f e11mmat10n
= [1 + (r.;=,,p;log 2 p;)J x 100
1. 585
The proposal was based on the idea that the generation of kerogen
started with an hypothetical organic compound which is a mixture
of equal proportions of the three elements, and which would have a
coefficient of elimination of 0 per cent. This hypothetical compound
would then undergo evolution due to geological processes which
could eliminate oxygen and hydrogen, leaving graphite, i.e. 100 per
cent carbon, with a coefficient of elimination of 100 per cent. In real
life most organic compounds which result from biogenic reactions
have values for the coefficient in the range 10-28 per cent, whereas
products resulting from the action of geological processes gave
considerably higher values. Indeed, it was suggested that kerogens
resulting from diagenesis had values in the range 28-48 per cent,
those resulting from catagenesis in the range 48-64 per cent and
metagenesis> 64 per cent.
(d) INDEPENDENCE OF DATA SOURCES

The independence of data sources is an area where the least work has
been done, since it does not often prove possible to study mixed data
sources in geological situations. However, one area which has been
considered briefly in this context is that of trend surface analysis.
Proponents of the technique have suggested that trend surface
analysis can, in certain circumstances, be considered to split the data
into two components - the regional and local trends. In this the
regional component is contained in the fitted values, and the local
component is in the residuals, along with any noise in the original
data. Unfortunately, this is a rather simplistic approach, and the
definition of what constitutes these components can be somewhat
arbitrary. In many cases trend surface analysis simply removes the
noise from the data.
In some simple cases it has proved possible to split the two
components from the data and then to examine the two trends
separately (Ferguson 1982). If a measure of the information content
of the original data and that of the two trend components can be
estimated, then we can use the relationships
H(x) = H(y) + H<:) (for independence)
H(x) # H(y) + H(z) (for dependence)

254 BEYOND TRADmONAL MATHEMATICS
where H<x> = the information content of the original data, H<Y> = the
information content of the regional trend component and H<=> = the
information content of the local trend component.
The calculation of the information content of the three compo-
nents is relatively simple. The range of the data values is subdivided
to give a number of suitable class intervals (the same for each
component), the width of the class interval in most applications
being the contour interval at which the final maps will be displayed.
The number of points which fall into each class interval is counted,
and the probability of points falling into each class interval is
obtained. Then the information content of the three sets can be
calculated:
From the resulting Hs we can calculate the entropy ratio for the data
sets
entropy ratio = H(y) + H(z)

H(x)
where the entropy ratio = I for independence or> I for dependence

(Ferguson 1982, p. 10).
In a number of practical situations the relationship between the
calculated values for the surface and the residual values was
examined. From this it was possible to show that as the complexity
of the surface fitted increased, so the entropy ratio approached
unity, indicating that the residuals were independent and probably
represented random noise. In fact, the method proved particularly
useful in isolating cases of overfitting of surfaces. In this situation
the surface generated is largely a mathematical artefact and the
entropy ratio values show a steady increase as more-complex
surfaces are fitted to the data. This follows since the model itself
(least squares) is generating information which is not present in the
original data (Ferguson 1982, p. 42).
Appendix 1 An introduction to the
methods of
differentiation
Al.l The rules of differentiation and their application
(a) The derivative of a function with a constant exponent. We have already

shown that if y = x", then
dy = nx"-1
dx
It follows from this that the derivative of a variable with respect to itself is
unity, i.e. if y = x, then
dy =I
dx
For example, differentiate y = 4x3 + 3x:
dy = (3 X 4)_x(l-l) + (3 X J)_x(l-1)
<ix-
= 12x2 + 3
We can take, as a further example, differentiate
y= ex
Write as
Then
=ex
256 APPENDIX 1
(b) If c is any numerical constant, its derivative is zero; i.e. the rate of
change of a constant is zero.
General form y = u + c, where u = f(x).
Example: y = 5x 2 + 6
dy =lOx
dx
(c) If cis any numerical constant and the equation is of the form y = cu,
where u = f(x), then
d(cy) _ dy
---c-
dx dx
Example: y = 4(5x' + 6)
dy = 4(l0x)
dx
or y = ujc, where u = f(x), then

dy _I du
dx-cdx
Example: y = (3x3 - 7x)j8
(d) Product rule: y = uv, where u and v = f(x)
d(uv) dv du
--=u-=v-
dx dx dx
Example: y = (x' + 4)(x 3 - I)
= [(x' + 4)3 x 2] + [(x3 - I )2x]
= (3x4 + 12x2) + (2x'- 2x)
= 5x4 + 12x2 - 2x
INTRODUCTION TO DIFFERENTIATION 257
(e) Quotient rule: y = u/v, where u and v = f(x)
du dv
v--u-
d(ujv) _ dx dx
~- y2
Example: y = (x 2 + 4)/(x 3 - 1)
dy _ (x 3 _ 1 )d(x~;4)_(x2 + 4)d(x::-l)
dx- (x3 - 1)2
_ [(x 3 - 1)2x]- [(x2 + 4)3x 2]

- .x6- 2x3 +I
-X'- 12x2 - 2x
.x6- 2x 3 +I
(f) Function of a function: y = f(v), e.g. y = v/(1 - v2) and v = f(x), e.g.
v = 1 - x 2 Then
dy _ dy dv
dx- dvdx
Example: y = v3 + 2, v = 2x
dy = 3v2 dv = 2
dv dx
Now,
dy _ dy dv
dx- dv dx
= 3v2 x 2
But v= 2x, so
dy = 3(2x)2 x 2 = 24x 2
dx
(g) The derivative of an inverse: y = ,j x
dx dy I
dy = I/ dx or dyjdx
258 APPENDIX 1
(h) The derivative of a variable function with a constant exponent:
y = (u)", where u = f(x)
dy = n(u)"- 'd(u)
dx dx
Example: given y = v3 + 3, v = 2x, then
n(u)"- 1 = 3(2x) 2 (note: constant disappears)
d(u) _ d(2x) =2
dx- dx
dy _ 3(2r) x 2 = 24x2
dx-
A1.2 Standard forms and their use
A list of some of the more commonly used standard forms is given in Table
A 1.1. Many others can be found in mathematics textbooks dealing with
calculus, to which the reader is advised to refer for a more comprehensive
list.
Table Al.l Some common standard differential forms.
Function Derivative
x" nx"-'
a' a' Ina
lnx 1/x
,jx tx-1;2
e-' e-'
sinx cosx
cosx - sinx
tanx (1 +tan' x) or
sec' x
The following simple examples show how the standard forms are used in
conjunction with the rules given in the previous section.
INTRODUCTION TO DIFFERENTIATION 259
(a) Differentiate y =(cos x sin x). Using the product rule
d(cosxsinx) d(sinx) + . d(cosx)

---'---..,d-x---'- = cos x _d_x_ sm x __ d_x_
Using the standard forms for the derivatives of sin x and cos x, we get
= cos x cos x + sin x(- sin x)
= cos 2x- sin 2x = cos 2x
(b) Differentiate y =tan x 3 Using the rule for a variable function with a
constant exponent and one of the standard forms for the derivative of tan x,
we get
(c) The use of logarithms in differentiation can often lead to considerable

simplification. Thus, using logarithms we can differentiate, for example,
by rewriting in the form
In y = In x + 3 In( I + x 2)
Using the standard form, we get
dy _I+ 6x
dx- x
(I+ x 2)
A1.3 Implicit differentiation
In general, an implicit function is of the form f(xy) =constant. For

example, 3x2 + y 2 = 12 is an implicit function. It can be rewritten in the
explicit form, i.e. y = f(x):
Either form of the equation can be used for differentiation; however, the
explicit form would present more difficulties than the implicit. For the
solution of the implicit equation we use the function of a function rule: i.e.
find dyjdx when 3x2 + y = 12:
260 APPENDIX 1
d(3x 2) d(Y) _ d(l2)
--+-----
dx dx dx
6x+ 2ydy = 0
dx
dy _- 6x _- 3x
dx-2_Y--y-
Thus, as a general rule it is better to use the implicit form of the equation
and differentiate using the function of a function rule.
Appendix 2 An introduction to the
methods of integration
A2.1 The rules of integration
In Appendix 1, outlining the methods of differentiation, we considered the

basic rules, along with some simple examples of their application. We shall
follow the same procedure for integration.
(a) IKydx=Kiydx
For example, we are asked to integrate 4x3 Thus
4x3+ 1
I 4x3 dx = 4Ix3 dx = - - + C = x' + C
3+ 1
Another example, we are asked to integrate 1.)x. Thus
5_x615
J1.)xdx= Ix 115 dx=-6-+ C
(b) I F(kx)dx = (1/k)JJ(kx)d(kx) + C

For example, we are asked to integrate cos 3x. Thus
I cos 3x dx = (l/3)Jcos 3x d(3x) = (l/3)sin 3x + C
(c) I(y + k)dx = Iydx + kx + C

We are asked to integrate (1 + 5x)jx. Thus
1 + 5x
J--dx = J(l!x + 5)dx =In x + 5x + C
X
Another example:
I 2xdx_ 3 = (l/2)ln(2x- 3) + C
(d) I (y 1 + y 2 + Y1 ... )dx = Jy 1dx + Jy 2dx + Jy 3dx ...

For example, integrate (4x 3 - 6x 2 + 4). Thus
IC4x 1 -6x2 +4)dx= I4x 1dx- I2x 2dx+ I4dx
4x' 6x 3
=---+4x+ C
4 3
= x' - 2x 3 + 4x + C
262 APPENDIX 2
(e) Juvdx=ufvdx- Jz(dujdx)dx, where z= Jvdx. Thus, the integral
sin X COS X is
Jsin x cos x dx = sin x Jcos x dx- Jsin x d sin xjdx dx

= J
sin'x- sin x cos x dx
A2.2 Standard forms
The most commonly used method of integrating is by using a table of

standard integral forms. If the equation in question does not correspond to
one of these, then it should be rewritten in a form that does. Some standard
integrals are given in Table A2.1.
Table A2.1 Some standard integral forms.
Derivative Integral
X"+ I
X"
n+ I+ C
e' e'+C
eh (1/k)e'-' + c
a' ~+C
Ina
1/x lnx+ C
sinx -cosx+C
COSX sinx+ C
tanx ln(secx) + C
sinkC - (l(k)cos kx + C
coskC (1/k)sin kx + C
I+ tan'x tanx+ C
sin-'x+ C
'-"CI - x')
I
I+ x 2 tan-'x+C
A2.3 Substitution
We have already noted that, if necessary, equations can be rewritten to

correspond with known standard forms, this is substitution. In other words,
the function is changed and the transformed integral evaluated.
INTRODUCTION TO INTEGRATION 263
Example: evaluate f x.j(5 + xZ)dx. Substitute
Then
dz
dx =2x
so that dz = 2x dx. By rule (b),
= t + Cz3!2
Returning to our original equation:
In another example, we are asked to evaluate f .j(l - r)dx. Substitute
x =sin B
Then
dx
-=cosO
dB
so that dx = cos B dB. The integral now becomes
f .j(l - sin 2B)cos B dB
Now y'(l - sin 28) =cos B (trigonometric identity), so we can rewrite as

f cos B cos B dB
= fcos 2BdB
= HO + cos 2B)dB
By rule (b)
y = !(I + cos B)d(2B)dB
By rule (d)
y = HB dB + Hcos 28 dB
y = !B + *sin 28 + C
264 APPENDIX 2
Since
sin 28 = 2 sin 8 cos 8
Then
y= tB + !(2 sin 8 cos 8) + C

Then
Appendix 3 Computer programs
The programs in this appendix have been developed to enable some of the
basic mathematical techniques to be explored further. It is suggested that if
a program is to be used, then the data from the appropriate example in the
text should be used as test data, to ensure that the program is working
correctly. All of the programs are specifically written for the BBC Micro
(Model B), and have been extensively tested. Students wishing to use the
programs on other systems should check some of the functions used. For
example, VDU statements are specific to the version of BASIC in which the
programs were written. As far as possible the programs have been kept
straightforward, and hopefully, with the REM statements, should be
relatively easy to follow. Also, prompts are given by the program for any
input required. They are presented here in the same order as they appear in
the text.
10 REM PROGRAM TO PLOT A GRAPH

20 REM Y AXIS VALUES LABEL A X AXIS VALUES LABEL B
30 REM INPUT N - NUMBER OF PAIRS OF VALUES
40 REM INPUT !DENT - A SUITABLE TITLE FOR THIS RUN
50 REM INPUT A AND B - LIST OF DATA TO BE PLOTTED
60 REM PRODUCES GRAPH ON LINE PRINTER DATA SCALED AUTOMATICALLY
70 REM VERTICAL AXIS 50 LINES
80 REM HORIXONTAL AXIS 70 CHARACTERS
90 REM RANGE FOR BOTH AXIS THE SAME- SMALLEST 0.01 TO 1.0 LARGEST
2.5 TO 250
100 REM OBSERVATIONS DO NOT HAVE TO BE ORDERED
110 REM DOES NOT PROCESS MULTIPLE DATA SETS
120 DIM A(500),B(500),BL$(3),Z$(70)
130 BL$(1)=" ":BL$(2)="*":BL$(3)="+"
140 REM SET PARAMETERS FOR PRINTING NUMBERS ON SCALE
150 @\=&01020105
160 INPUT"TYPE IN NUMBER OF PAIRS OF DATA POINTS",N
170 INPUT"TYPE IN TITLE FOR THIS RUN",ID$
180 INPUT "TYPE IN LABEL FOR Y AXIS",YAX$
190 INPUT "TYPE IN LABEL FOR X AXIS",XAX$
200 INPUT "TYPE IN UNIT OF MEASUREMENT",UN$
210 FOR !=1 TO N
220 INPUT"Y="A(I); "X="B(I)
230 NEXT I
240 PRINT "ALL DATA POINTS ENTERED"
250 REM MAKES WARNING SOUND WHEN ALL DATA ARE ENTERED
260 FOR pitch~52 TO (52+48) STEP 4
270 SOUND 1,-15,pitch,5
280 NEXT pitch
290 AR=A( 1)
300 FOR I=1 TO N
310 IF A(I)>AR THEN AR=A(I)
320 NEXT I
330 SCALE=O
340 SC=40
350 PROCScale
360 SCALA=SCALE
370 FOR I= 1TO N
380 A(I)=A(I)SCALA+O.
390 NEXT I
400 AR=B( 1)
410 FOR !=1 TO N
420 IF B(I)>AR THEN AR=B(I)
430 NEXT I
440 SCALE=O
266 APPENDIX 3
450 SC=70
460 PROCScale
470 SCALB=SCALE
480 FOR I=1 TO N
490 B(Il=B(I)*SCALB+O.
500 NEXT I
510 PROCSort
520 REM SWITCH ON PRINTER
530 VDU2
540 PROCGraph
550 PROCident
560 REM SWITCH OFF PRINTER
570 VDU3
580 PRINT'IF YOU REQUIRE ANOTHER GRAPH, TYPE RUN TO RESTART"
590 END
600 REM
610 REM
620 REM
630 DEF PROCSort
640 REM PROCEDURE TO SORT DATA
650 REM Y VALUES IN DESCENDING ORDER
660 REM ORDERS X VALUES TO CORRESPOND WITH Y VALUES
670 REM BECAUSE OF ROUND OFF PROBLEMS WITH HIGH VALUES SORT USES THE
INTEGER VALUE FOR X AND Y
680 REM SECONDARY SORT PUTS X VALUES INTO ASCENDING ORDER IF
REQUIRED
690 MIN=N-1
700 FOR I~1 TO MIN
710 L=I
720 IF INT(A(L))<INT1A(L+1)) GOTO 750
730 IF INT(A(L))>INT(A(L+1)) GOTO 830
740 IF INT(B(L))<=INT(B(L+1)) GOTO 830 750 DUM=A(L)
760 A(L)=A(L+1)
770 A(L+1)~DUM
780 DUM=B(L)
790 B(Ll=B(L+1)
800 B(L+1)=DUM
810 L=L-1
820 IF L>O GOTO 720
830 NEXT I
840 ENDPROC
850 REM
860 REM
870 REM
880 DEF PROCScale
890 REM SCALE VALUES FOR BOTH AXIS
900 REM FIRST CALL VERTICAL (Y) AXIS
910 REM SECOND CALL HORIZONTAL (X) AXIS
920 IF AR<0.01 GOTO 1110
930 IF AR>1 GOTO 960
940 SCALE=SC/1
950 GOTO 1130
960 IF AR>10 GOTO 990
970 SCALE=SC/10
980 GOTO 1130
990 IF AR>25 GOTO 1020
1000 SCALE=SC/25
1010 GOTO 1130
1020 IF AR>50 GOTO 1050
1030 SCALE=SC/50
1040 GOTO 1130
1050 IF AR>100 GOTO 1080
1060 SCALE=SC/100
1070 GOTO 1130
1080 IF AR>250 GOTO 1110
1090 SCALE=SC/250
1100 GOTO 11 30
1110 PRINT "DATA VALUES OUT OF RANGE" 1120 GOTO 590
1130 ENDPROC
1140 REM
1150 REM
1160 REM
1170 DEF PROCGraph
COMPUTER PROGRAMS 267
1180 REM START GRAPH PLOTTING LOOP
1190 L=1
1200 FOR I=1 TO 40
1210 II=41-I
1220 IF L>N GOTO 1270
1230 IF (II MOD 4)=0 GOTO 1580
1240 IF II=INT(A(L)) GOTO 1340
1250 PRINT" I"
1260 GOTO 1880
1270 IF (II MOD 4)=0 GOTO 1300
1280 PRINT" I"
1290 GOTO 1880
1300 SCALY=II/SCALA
1310 SC$=STR$(SCALY)
1320 PRINT RIGHT$(SC$,5)
1330 GOTO 1880
1340 M=L
1350 IF L=N GOTO 1400
1360 FOR K=L TO N
1370 IF INT(A(K))<>INT(A(K+1) ) GOTO 1400
1380 M=M+1
1390 NEXT K
1400 FOR J=1 TO 70
1410 Z$(J)=BL$(1)
1420 FOR K=L TO M
1430 IF J=INT(B(K)) GOTO 1450
1440 GOTO 1490
1450 IF Z$(J)OBL$(1) GOTO 1480
1460 Z$(J)=BL$(2)
1470 GOTO 1490
1480 Z$(J)=BL$(3)
1490 NEXT K
1500 NEXT J
1510 PR$=""
1520 FOR J=1 TO 70
1530 PR$=PR$+LEFT$(Z$(J), 1)
1540 NEXT J
1550 PRINT" I",PR$
1560 L=M+1
1570 GOTO 1880
1580 SCALY=II I SCALA
1590 IF L>N GOTO 1620
1600 IF II= INT(A(L)) GOTO 1640
1620 PRINT RIGHT$(SC$,5)
1630 GOTO 1880
1640 M=L
1650 IF L=N GOTO 1700
1660 FOR K=L TO N
1670 IF INT(A(K))<>INT(A(K+1) ) GOTO 1700
1680 M=M+1
1690 NEXT K
1700 FOR J=1 TO 70
1710 Z$(J)=BL$(1)
1720 FOR K=L TO M
1730 IF J=INT(B(K)) GOTO 1750
1740 GOTO 1790
1750 IF Z$(J)OBL$(1) GOTO 1780
1760 Z$(J)=BL$(2)
1770 GOTO 1790
1780 Z$(J)=BL$(3)
1790 NEXT K
1800 NEXT J
1810 PR$=""
1820 FOR J=1 TO 70
1840 PR$=PR$+LEFT$(Z$(J), 1)
1850 NEXT J
1860 PRINT RIGHT$(SC$,5),PR$
1870 L=M+1
1830 NEXT I
1890 PR$="
1900 FOR J=1 TO 70
268 APPENDIX 3
1910 IF(J MOD 7)=0 GOTO 1950
1920 Z$(J)="-"
1930 PR$=PR$+"-"
1940 GOTO 1960
1950 PR$=PR$+"."
1960 NEXT J
1970 PRINT PR$
1980 PR$="
1990 FOR J=7 TO 70 STEP
2000 SCALX=J/SCALB
2010 B$=STR$(SCALX)
2020 K=LEN(B$)
2030 L=5-K
2040 IF L=O GOTO 2080
2050 FOR M=1 TO L
2060 B$=" "+B$
2070 NEXT M
2080 PR$=PR$+B$+"
2090 NEXT J
2100 PRINT PR$
2110 ENDPROC
2120 DEF PROCident
2130 REM PRINTS TITLE AND SCALE
2140 PRINT' '"Graph of- ",ID$
2150 PRINT'"Y axis represents ", YAX$, ''"Scale factor for Y axl::>
is",SCALA
2160 PRINT''X axis represents- ",XAX$,' '"Scale factor for X axis
is",SCALB
2170 PRINT'"Original data follows"
2180 PRINT''Measurements in ",UN$
2190 PRINT''
2200 FOR I=1 TO N
2210 A(I)=(A(I)-0.5)/SCALA
2220 B(I)=(B(I)-0.5)/SCALB
2230 PRINT A(I);" ";B(I)
2240 NEXT I
2250 PRINT' ''GRAPH COMPLETE"
2260 ENDPROC
Program 2.1 Program to draw a graph of x against y, using a BBC Micro and an
Epson printer. The axes are scaled automatically to fit the printer paper size
(22 x 28 em). Not to be used as VDU output in its present form.
10 REM DEF PROCQUAD

20 REM REMOVE REM FROM PREVIOUS STATEMENT TO USE THIS AS A
PROCEDURE
30 REM PROGRAM TO SOLVE A QUADRATIC
40 REM FORM OF EQUATION Ax2 + Bx + C = 0
50 REM VALUES REQUIRED ARE COEFFICIENTS A & B AND THE CONSTANT C
60 REM MULTIPLE DATA SETS PROCESSED
80 VDU3
90 INPUT"TYPE IN COEFFICIENTS"'"REMEMBER A 0 TERMINATES THE
EXECUTION"' "A="A
100 IF A = 0 GOTO 390
110 INPUT "B="B
120 INPUT'TYPE IN CONSTANT"'"C="C
130 REM CALCULATE DISCRIMINANT AND TEST
140 REM SET PARAMETERS FOR PRINTING RESULTS
150 @.=&20206
170 VDU2
180 REM PRINT OUT DATA
190 PRINT'' '"Coefficients are,"'"A="A, ''B='B, '"Constant is,''"C="C
200 DISC=B2-4AC
210 REM IF DISCRIMINANT > 0 ROOTS ARE REAL AND DISTINCT
220 IF DISC>O GOTO 280
230 REM IF DISCRIMINANT 0 ROOTS ARE REAL AND EQUAL
240 IF DISC=O GOTO 320
250 REM IF DISCRIMINANT 0 ROOTS ARE COMPLEX
260 GOTO 350
270 REM CALCULATE AND PRINT ROOTS ACCORDING TO VALUE OF DISCRIMINANT
280 R1=(-B+SQR(DISC))/(2*A)
290 R2=(-B-SQR(DISC))/(2*A)
300 PRINT''' "Roots are real and distinct"'"X1 =",R1;" X2 =",R2
310 GOTO 70
320 R=B/(2*A)
330 PRINT'' '"Roots are real and equal"'"X ",R
340 GOTO 70
350 PR=-B/(2*A)
360 CN=SQR(-DISC)/(2*A)
370 PRINT'' '"Roots are complex"'"X ="PR"(+ or -)"CN"i"
380 GOTO 70
390 END
Program 2.2 Program to solve for the roots of a quadratic equation. Requires
coefficients of xs and the constant, as input.
10 REM THIS PROGRAM CALLED MEAN

20 REM CALCULATES 1. ARITHMETIC MEAN
30 REM 2. GEOMETRIC MEAN
40 REM 3. HARMONIC MEAN
50 REM STORES UPTO 500 VALUES
60 REM CAN CALCULATE ANY OR ALL OF THE ABOVE MEANS
70 REM DOES NOT PROCESS MULTIPLE DATA SETS
80 DIM X(500)
90 Y$="N"
100 CLS
110 PRINT'"This program calculates three means"'"ThPY arP:
Arithmetic Mean"'" Geometric MeanM'" Harmonir Mean"
120 PRINT'' "Which will you require"'"Typ~ ARITHMETIC, GFOMFTRTr nr
HARMONIC"
130 INPUT TH$
140 IF TH$"ARITHMETIC" GOTO 190
150 IF TH$="GEOMETRIC" GOTO 190
160 IF TH$="HARMONIC" GOTO 190
170 PRINT'' "BAD DATA!"
180 GOTO 120
190 IF Y$="Y" GOTO 300
200 INPUT'' "How many data points?" 'N
210 IF N>500 OR N< 2 PRINT' '"Too many or too few points!" ELSE GOTO
230
220 GOTO 200
2 30 PRINT' ' "Now input your data, one point at a time"
240 FOR I = 1 TO N
250 INPUT X(I)
260 NEXT I
270 SOUND 1,-15,53,20
280 PRINT'' "All data is entered"
285 VDU2
290 PRINT'' "NUMBER OF POINTS ENTERED ="N
295 VOU3
300 Y$="N"
310 IF TH$="ARITHMETIC" THEN PROCamn ELSE GOTO 360
320 VDU2
330 PRINT' '"THE ARITHMETIC MEAN ="AMEAN
340 VDU3
350 GOTO 450
360 IF TH$="GEOMETRIC" THEN PROCgmn ELSE GOTO 410
370 VDU2
380 PRINT'' "THE GEOMETRIC MEAN ="GMEAN
390 VDU3
400 GOTO 450
270 APPENDIX 3
410 PROChmn
420 VDU2
430 PRINT'' "THE HARMONIC MEAN ="HMEAN
440 VDU3
450 PRINT' '"Calculate another mean? Y/N"
460 INPUT Y$
470 IF Y$="Y" GOTO 110
480 PRINT' '"THIS EXECUTION TERMINATED"
490 STOP
500 DEF PROCamn
510 REM CALCULATE ARITHMETIC MEAN
520 XSUM=O
530 FOR I = 1 TO N
540 XSUM=XSUM + X(I)
550 NEXT I
560 AMEAN = XSUM/N
570 ENDPROC
580 DEF PROCqmn
590 REM CALCULATE GEOMETRIC MEAN
600 XPROD LOG(X(1))
610 FOR I = 2 TO N
620 XPROD = XPROD+LOG(X(I))
630 NEXT I
640 GMEAN = 10"(XPROD/N)
650 ENDPROC
660 DEF PROChmn
670 REM CALCULATE HARMONIC MEAN
680 RXSUM 0
690 FOR I = 1 TO N
700 RXSUM = RXSUM+(1/X(I))
710 NEXT I
720 HMEAN = N/RXSUM
730 ENDPROC
Program 3.1 Program to calculate the arithmetic, geometric or harmonic means of a

set of data. Accepts up to 500 values.
10 REM PROGRAM TO SOLVE A THIRD DEGREE POLYNOMIAL FOR MAXIMA AND

MINIMA
20 REM INPUT NEEDED ARE THE FOUR NUMERICAL CONSTANT"
40 REM INPUT DATA
50 INPUT"VALUES FOR CONSTANTS"'"SET x3 VALUE TO ZERO TO TERMINATE
RUN"' "X"3 VALUE",A
60 IF A=O GOTO 490
70 INPUT"X"2 VALUE",B'"X VALUE",C'"CONSTANT",D' "TITLE (IN INVERTED
COMMAS)",XNAM$
90 @\=&20206
100 REM FIND FIRST DIFFERENTIAL
110 A1=A*3
120 81=8*2
130 C1=C
140 REM TEST FOR TWO REAL ROOTS
150 disc=B1"2-4*A1*C1
160 IF disc<O GOTO 460
170 IF disc=O GOTO 220
180 S=SQR(disc)
190 X1=(-81+S)/(2*A1)
200 X2=(-B1-S)/(2*A1)
210 GOTO 240
220 X1=-81/(2*A1)
230 X2~X1
240 REM FIND SECOND DIFFERENTIAL
250 A2=A 1*2
260 82=81
270 REM CALCULATE VALUES FOR FUNCTION USING ROOTS OF FIRST
DIFFERENTIAL
280 X=X1
290 J=1
300 XVAL2=A*X"3+B*X"2+C*X+D
310 J=J+1
320 IF J>2 GOTO 360
330 X=X2
340 XVAL1=XVAL2
350 GOTO 300
360 XA=A2*X1+B2
370 XB=A2*X2+82
380 REM ENABLE PRINTER
390 VDU2
400 PRINT' '"TITLE FOR THIS DATA SET "XNAM$' '"THE ORIGINAL
EQUATION WAS" ,A"X3" ,B"X2" ,C"X" ,D' '"SOLUTION OF SECOND DIFFERENTIAL FOR
X="X1" IS"XA''" AND FOR X="X2" IS"XB
410 IF XA>O GOTO 440
420 PRINT'"" WHEN X="X1" ,X IS MAXIMISED AT Y="XVAL1" "AND WHEN
X="X2" ,X IS MINIMISED AT Y="XVAL2
430 GOTO 470
440 PRINT""' WHEN X="X2" ,X IS MAXIMISED AT Y="XVAL2" "AND WHEN
X="X1" ,X IS MINIMISED AT Y="XVAL1
450 GOTO 470
460 PRINT' '"NO SOLUTION, ROOTS ARE COMPLEX"
470 VDU3
480 GOTO 50
490 END
Program 4.1 Program to solve problems of maxima and minima for a third-degree
polynomial. Input values required are the coefficients of the equation and the
constant.
10 REM PROGRAM CALLED SIMRULE

20 REM PROGRAM TO FIND THE AREA UNDER A CURVE
30 REM USES SIMPSONS RULE WITH N STRIPS
40 REM NUMBER OF STRIPS MUST BE EVEN
50 REM READ IN FUNCTION TO BE EVALUATED IN THE FORM OF A BASIC
STATEMENT
60 REM READ IN DATA
70 INPUT"NUMBER OF STRIPS, MUST BE AN EVEN NUMBER"' "ZERO TERMINATES
RUN",N
80 IF N=O GOTO 380
90 IF (N MOD 2) =1 GOTO 360
100 INPUT'"TYPE IN FUNCTION AS A BASIC STATEMENT"' "REMEMBER, NO
STATEMENT NUMBER"' ,X$
110 INPUT"LOWER LIMIT", LL' "UPPER LIMIT", UL
120 REM SET CONSTANTS
130 H~(UL-LL)/N
140 FACT=H/3
150 N1=N+1
160 SUM=O,X=LL-H
170 REM CALCULATE AND SUM AREAS OF EACH STRIP
180 FOR I=1 TO N1
190 X=X+H
200 XX=EVAL(X$)
210 IF I=1 OR I=N1 GOTO 250
220 IF(I MOD 2)=1 GOTO 270
230 SUM=SUM+4*XX
240 GOTO 280
250 SUM=SUM+XX
260 GOTO 280
270 SUM=SUM+2*XX
280 NEXT I
290 AREA=SUM*FACT
272 APPENDIX 3
310 VDU2
320 PRINT'' '"NUMBER OF STRIPS ="N' "LOWER LIMIT OF INTEGRATION
="LL' "UPPER LIMIT OF INTEGRATION ="UL' '"AREA ="AREA
330 REM DISABLE PRINTER
340 VDU3
350 GOTO 370
360 PRINT" ODD NUMBER OF STRIPS ENTERED, TRY AGAIN!"
370 GOTO 70
380 END
Program 5.1 Program to find the area under a curve between given limits, using
Simpson's Rule. The function to be evaluated is entered in the form of a BASIC
statement.
10 REM THE NAME OF THIS PROGRAM IS FORAN

20 REM PERFORMS FOURIER ANALYSIS ON SEQUENTIAL DATA
30 REM DATA SAMPLING MUST BE AT FIXED INTERVAL
40 REM ONE VARIABLE OR PROPERTY MEASURED AT EACH POINT
50 REM MAXIMUM NUMBER OF SAMPLE POINT IS 401
60 REM BASED ON EQUATIONS GIVEN BY HARBAUGH AND MERIAM ( 1968)
p.114ff.
70 REM USES NUMERICAL INTEGRATION METHOD
80 REM BASIC VERSION OF AN UNPUBLISHED FORTRAN PROG. BY FERGUSON
90 DIM ADEG(201),cnt(401),U(401), ACOSC(201),ASINC(201),PSPECT(201)
100 REM
110 REM CLEAR SCREEN AND SET NEW PAGE
120 REM MAIN PROGRAM READS DATA AND CALLS PROCEDURES
130 REM NUMBER OF DATA POINTS IS KMAX AND MUST BE ODD
140 REM VALUE FOR SAMPLE INTERVAL IS TH
150 REM MEASURED VARIABLE IS U(I), I= 1 TO KMAX
160 REM CUMULATIVE SAMPLE INTERVAL FOR KMAX POINTS IS cnt(I)
170 VDU2
180 CLS
190 VDU3
200 INPUT "TYPE IN NUMBER OF SAMPLE POINTS"'"MAXIMUM NUNBER OF
POINTS IS 401""'MUST BE AN ODD NUMBER'""TYPE O(ZERO) TO END RUN",KMAX
210 IF KMAX=O GOTO 600
220 INPUT"TYPE IN TITLE",T$
230 VDU2
240 PRINT"TITLE OF THIS DATA SET:"' ,T$, ' ' '
250 VDU3
260 INPUT "TYPE IN VALUE FOR SAMPLE INTERVAL",TH
270 PRINT"TYPE IN MEASURED VARIABLE"'"FOR EACH SAMPLE POINT"
280 FOR I = 1 TO KMAX
290 INPUT U(I)
300 NEXT I
310 REM CALCULATE THE CUM. INTERVAL VALUE FOR EACH SAMPLE POINT
320 cnt(1)=0
330 FOR K = 2 TO KMAX
340 cnt(K) = cnt(K-1)+TH
350 NEXT K
360 REM SET UP PARAMETERS TO PRINT NUMBERS
370 @'=&0002040A
380 REM CALL PROCEDURE TO CALCULATE BASIC STATISTICS
390 REM THE PROCEDURE IS CALLED stat
400 PROCstat
410 VDU 2
420 PRINT"RESULTS OF STATISTICS OF ORIGINAL
DATA" ' "MEAN", AVG' "VARIANCE", VAR' "STANDARD DEVIATION" , STDEV' "TOTAL SUMS
OF SQUARES",STOTSS
430 PRINT''' "THE ORIGINAL DATA SET IS AS FOLLOWS:"'" INTERVAL
VARIABLE"
440 FOR I = 1 TO KMAX
450 PRINT cnt(I),U(I)
460 NEXT I
470 VDU 3
480 REM CALL PROCEDURE TO CALCULATE SIN AND COS TERMS AND POWER
SPECTRUM
490 REM THE PROCEDURE IS CALLED forft
500 PROCforft
510 VDU 2
520 PRINT'''
530 PRINT "RESULTS OF FOURIER ANALYSIS"
540 PRINT" HARMONIC COSINE SINE POWER SPEC"
550 FOR I = 1 TO LK
560 PRINT ADEG(I),ACOSC(I),ASINC(I),PSPEC T(I)
570 NEXT I
sao VDU 3
590 GOTO 160
600 END
610 REM
620 REM PROCEDURE TO CALCULATE FOURIER COEFFICIENTS FROM TIME TREND
DATA
630 REM
640 DEF PROCforft
650 REM CALCULATES COSINE TERM (ACOSC)
660 REM CALCULATES SINE TERM (ASINC)
670 REM CALCULATES POWER SPECTRUM FOR EACH PAIR OF TERMS
(PSPECT)
680 REM CALCULATES EACH OF THE ABOVE FOR EACH HARMONIC (ADEG)
690 REM THERE ARE (ABS(KMAX/2))+1 HARMONICS
700 L\=KMAX/2+1
710 LK=L\
720 ADEG ( 1) =0
730 FOR L= 2 TO LK
740 II=L-1
750 ADEG(L)=ADEG(II)+1
760 NEXT L
770 AMAX1=cnt(KMAX)
780 SMCON=O
790 FOR L= 1 TO LK
800 SMCON=(U(1)+U(KMAX))/2
810 KMAXJ=KMAX-1
820 FOR K=2 TO KMAXJ
830 SMCON=SMCON+(U(K)*COS((ADEG(L)
*PI*cn~(K))/(AMAX1/2.0)))
840 NEXT K
850 ACOSC(L)=(2.0/(KMAX-1))*SMCON
860 NEXT L
870 ACOSC(1) = ACOSC(1)/2.0
880 FOR L = 2 TO LK
890 SMCON =0
900 FOR K=2 TO KMAXJ
910 SMCON=SMCON+(U(Kl*SIN((ADEG(L)
*PI*cnt(K))/(AMAX1/2.0)))
920 NEXT K
930 ASINC(L)=(2.0/(KMAX-1))*SMCON
940 NEXT L
950 ASINC(1)=0.0
960 PSPECT(1)=0.0
970 FOR L = 2 TO LK
980 PSPECT(L)=SQR(ACOSC(L)"2
+ASINC(L)"2)
990 NEXT L
1000 ENDPROC
1010 REM
1020 REM PROCEDURE TO CALCULATE STATISTICS FOR DATA
1030 REM
1040 DEF PROCstat
1050 REM CALCULATES MEAN (AVG)
1060 REM CALCULATES VARIANCE (VAR)
1070 REM CALCULATES STANDARD DEVIATION (STDEV)
1080 REM CALCULATES THE TOTAL SUM OF SQUARES (STOTSS)
1090 SUM=O
11 00 SUMSQ~O
1110 SUMSQD=O
1120 FORK= 1 TO KMAX
1130 SUM= SUM+ U(K)
1140 SUMSQ=SUMSQ+U(K)"2
1150 NEXT K
1160 AVG=SUM/KMAX
274 APPENDIX 3
1170 A=SUM"2
1180 B=A/KMAX
1190 STOTSS=SUMSQ-B
1200 VAR=STOTSS/(KMAX-1)
1210 STDEV=SQR(VAR)
1220 ENDPROC
Program 5.2 Program to perform Fourier analysis on sequential data. Maximum

number points 401 (must be odd) and sampled at set interval.
10 REM PROGRAM TO MULTIPLY A MATRIX BY

20 REM (1) A SCALAR
30 REM (2) A VECTOR
40 REM VECTOR MUST HAVE SAME NUM. ROWS (NR2\) AS THE MATRIX HAS
COLUMNS (NC1\)
50 REM NR2\=NC1\~NRC\
60 REM (3) A MATRIX
70 REM FIRST MATRIX HAS SAME NUMBER OF COLUMNS (NC1")
80 REM AS SECOND HAS. ROWS (NR2\)
90 REM NC1\=NR2\=NRC\
100 REM OTHER PARAMETERS - NO. OF ROWS OF FIRST MATRIX NR1\
110 REM NO. OF COLUMNS OF SECOND MATRIX NC2\
120 DIM AMAT(10,10),BMAT(10,10),
CMAT(10, 10),AVECT(10),CVECT(10)
130 REM READ IN OPTION AS ABOVE
140 INPUT"TYPE IN NUMBER TO INDICATE CHOSEN OPTION'''" 1 " MlJT.T. BY A
SCALAR"'" 2 = MULT. BY A VECTOR""' 3 = MULT. BY A MATRIX""' A VALUE
GREATER THAN 3 TERMINATES THE EXECUTION",IOPT
150 IF IOPT>3 GOTO 630
160 IF IOPT=3 GOTO 440
180 REM READ IN DATA
190 INPUT"VALUE OF SCALAR",SCALAR
200 INPUT"NO. OF ROWS OF MATRIX", NR 1"' "NO. OF COLUMNS OF
MATRIX", NC 1'.
210 PRINT"TYPE IN THE MATRIX ROW-WISE ONE ELEMENT AT A TIME"
220 FOR I\=1 TO NR1\, FOR J\=1 TO NC1\
230 INPUT AMAT(I\,J\)
240 NEXT J\, NEXT I\
250 PROCMscalar
260 GOTO 610
270 INPUT" NO. OF ROWS OF VECTOR", NR2\ "'NO. OF ROWS OF
MATRIX" ,NR1'.' "NO. OF COLUMNS OF MATRIX"NC1"
280 IF NC1\<>NR2\ GOTO 400
290 NRC\=NC1\
300 PRINT''TYPE IN THE MATRIX ROW-WISE ONE ELEMENT AT A TIME"
310 FOR I\=1 TO NR1\, FOR J\~1 TO NRC\
340 PRINT"TYPE IN THE VECTOR ONE ELEMENT AT A TIME"
350 FOR I\=1 TO NRC\
360 INPUT AVECT(I\)
370 NEXT I \
380 PROCMvector
390 GOTO 610
400 PRINT"BAD DATA"'"NO. OF COLUMNS OF MATRIX NOT EQUAL TO NO. OF
ROWS OF VECTOR"
410 SOUND 1,-10,97,10' SOUND 1,-10,105,10
420 SOUND 1,-10,89,10, SOUND 1,-10,41,10, SOUND 1,-10,69,20
430 GOTO 140
440 INPUT"NO. OF ROWS OF 1ST. MATRIX" ,NR1\, "'NO. OF COLUMNS OF 1ST
MATRIX", NC1'.' "NO. OF ROWS OF 2ND. MATRIX", NR2"' "NO. OF COLUMNS OF 2ND
MATRIX", NC?"
450 IF NC1\<>NR2\ GOTO 570
4 60 NRC'.=NC 1\
470 PRINT"TYPE IN THE 1ST. MATRIX ROW-WISE ONE ELEMENT AT A TIME"
480 FOR I\=1 TO NR1\, FOR J\=1 TO NRC\
510 PRINT"TYPE IN THE 2ND. MATRIX ROW-WISE ONE ELEMENT AT A TIME"
520 FOR I\=1 TO NRC\, FOR J\=1 TO NC2\
530 INPUT BMAT(I\,J\)
550 PROCMmatrix
560 GOTO 610
570 PRINT"BAD DATA"' "NO. OF COLUMNS IN 1ST. MATRIX NOT EQUAL TO NO.
OF ROWS IN 2ND."
5 80 SOUND 1 , -10, 9 7, 10, SOUND 1 , -10, 1OS, 10
5 90 SOUND 1 , -10, 8 9, 10, SOUND 1 , -10, 41 , 10' SOUND 1 , - 10, 6 9, 20
600 GOTO 140
610 PROCPrdata
620 GOTO 140
630 END
640 REM
650 REM
660 REM
670 DEF PROCMscalar
680 REM MULTIPLY A MATRIX BY A SCALAR
690 FOR I\=1 TO NR1\
700 FOR J\=1 TO NC1\
710 CMAT(I\,J\)=AMAT(I\,J\)*SCALAR
730 ENDPROC
740 REM
750 REM
760 REM
770 DEF PROCMvector
780 REM MULTIPLY A MATRIX BY A VECTOR
800 CVECT(I\)=0.0
810 FOR J\=1 TO NRC\
820 CVECT(I\)=CVECT(I\)+AMAT(I\,J\}
*AVECT(J\)
840 ENDPROC
850 REM
860 REM
870 REM
880 DEF PROCMmatrix
890 REM MULTIPLY A MATRIX BY A MATRIX
910 FOR J\=1 TO NC2\
920 CMAT(I\,J\)=0.0
930 FOR K\=1 TO NRC\
940 CMAT(I\,J\}=CMAT(I\,J\)t
AMAT(I\,K\}*BMAT(K\,J\}
950 NEXT K\, NEXT J\, NEXT I\
960 ENDPROC
970 REM
980 REM
990 REM
1000 DEF PROCPrdata
1010 REM TO PRINT RESULTS OUT
1020 REM SET PRINT PARAMETERS AND SWITCH ON PRINTER
1030 @\=&20206
1040 VDU2
1070 PRINT' '"Multiplication by a scalar~'"Scalar ~",SCALAR
''"Original data"
1080 FOR I\=1 TO NR1\, PRINT'' FOR J\=1 TO NC1\
1090 PRINT AMAT(I\,J\);
1110 PRINT'' NResulting matrix"
1120 FOR I\=1 TO NR1\, PRINT'' FOR J\=1 TO NC1 ..
1130 PRINT CMAT(I\,J\);
1150 GOTO 1440
1160 PRINT ''"Multiplication by a vector"'"Original vector"''
276 APPENDIX 3
1170 FOR I~=1 TO NRC~
1180 PRINT AVECT(I~)
1190 NEXT I.
1200 PRINT' '"Oriqinal matrix"
1210 FOR I~=1 TO NR1\o PRINT'' FOR J\=1 TO NC1\
1230 NEXT J~o NEXT I\
1240 PRINT'' "Resultinq vector"''
1250 FOR I\=1 TO NR1\
1260 PRINT CVECT(I\)
1270 NEXT I\
1280 GOTO 1440
1290 PRINT'' "Multiply by a matrix"' "Oriqinal matrices"''
1300 PRINT "Matrix 1"
1310 FOR I\=1 TO NR1\o PRINT', FOR J\=1 TO NC1\
1330 NEXT J\o NEXT I\
1340 PRINT'' "Matrix 2"''
1360 PRINT BMAT(I\,J\);
1380 PRINT'' "Resultinq matrix"
1400 PRINT CMAT(I\,J\);
1420 PRINT' '
1440 VDU3
1450 ENDPROC
Program 6.1 Program to perform multiplication operations on any matrix up to size

10 x 10. Will multiply by a scalar, a vector or another matrix.
10 REM PROGRAM EIGVAL

20 REM THIS VERSION IN BASIC, AFTER DAVIS (1973) FROG 4.10,
P166-67
30 REM USES JACOBI METHOD
40 REM TO CALCULATE THE EIGENVALUES AND EIGENVECTORS OF A
SYMETRICAL MATRIX
50 REM RESULTING EIGENVECTORS IN MATRIX A (IN DIAG)
60 REM EIGENVALUES IN MATRIX B (COLUMN-WISE)
70 DIM A(10,10),B(10,10)
90 @\=0.20206
100 REM INPUT DATA
110 REM INPUT ORDER OF MATRIX- 0\
120 INPUT"TYPE IN ORDER OF MATRIX"' "TYPE IN 0 TO TERMINATE
EXECUTION", o.
130 IF 0\=0 GOTO 1120
140 REM INPUT MATRIX
150 PRINT "INPUT MATRIX, ONE ELEMENT AT A TIME - ROW WISE"
160 FOR I\=1 TO 0\
170 FOR J\=1 TO 0\
180 INPUT A(I\,J\)
200 REM B IS SET AS IDENTITY MATRIX
210 REM SET UP CONSTANTS AND COMPUTE THRESHOLD
220 N~o.
230 INORM=O
240 FOR I\=1 TO 0\
250 FOR J\=1 TO 0\
260 IF (I\-J\)<>0 GOTO 290
270 B(I\,J~)=1.0
280 GOTO 310
290 B(I~,J\)=0.0
300 INORM=INORM+A(I\,J\)*A(I\,J\)
310 NEXT J\: NEXT I\
320 INORM=SQR(INORM)
330 FLORM=INORM*1.0E-08/N
340 THR=INORM
350 THR=THR/N
360 IND\=0
370 REM FIND OFF-DIAGONAL ELEMENTS WITH VALUES GREATER THAN
THRESHOLD
380 FOR I\=2 TO 0\
390 rr.=I\-1
400 FOR J\=1 TO II\
410 IF ((ABS(A(J\,I\)))-THR) < 0 GOTO 760
420 REM COMPUTE SIN AND COS
430 IND\=1
440 IlL- A(J'., I'\.)
450 AM=(A(J\,J\)-A(I\,I\))/2.0
460 AO=AL/SQR(AL*AL+AM*AM)
470 IF AM<O GOTO 490
480 GOTO 500
490 AO=-AO
500 SX=AO/SQR(2.0*(1.0
+SQR(1.0-AO*AO)))
510 SX2=SX*SX
520 CX=SQR(1.0-SX2)
530 CX2=CX*CX
540 REM ROTATE COLUMNS
550 FOR K\=1 TO 0\
560 IF (K\-J\)=0 GOTO 610
570 IF (K\-I\)~0 GOTO 610
580 AT=A(K\,J'.)
590 A(K\,J\)=AT*CX-A(K\,I'\.)*SX
600 A(K\,I\)=AT*SX+A(K\,I\)*CX
610 BT=B(K\,J\)
620 B(K\,J\)=BT*CX-B(K\,I\)*SX
630 B(K\,I\)=BT*SX+B(K\,I\)*CX
640 NEXT K\
650 XT=2.0*A(J\,I\)*SX*CX
660 AT=A(J\,J.)
670 BT=A(I\,I\)
680 A(J\,J\)=AT*CX2+BT*SX2-XT
690 A(I\,I\)=AT*SX2+BT*CX2+XT
700 A(J\,I\)=(AT-BT)*SX*CX+A(J\,I'\.)
*(CX2-SX2)
710 A(I\,J\)=A(J\,I\)
720 FOR K\=1 TO 0\
730 A(J\,K\)=A(K\,J\)
740 A(I\,K\)=A(K\,I\)
750 NEXT K\
760 NEXT J\
770 NEXT I \
780 IF IND\>0 GOTO 360
790 IF (THR-FLORM)>O GOTO 350
800 REM SORT EIGENVALUES AND EIGENVECTORS
810 FOR I\=2 TO 0\
820 J\=I\
830 IF(A(J%-1,J\-1)-A(J\,J%))<0 GOTO 850
840 GOTO 950
850 AT=A(J\-1,J%-1)
860 A(J\-1,J\-1)=A(J\,J\)
870 A(J'.,J'.)=AT
880 FOR K\=1 TO 0\
890 AT=B(K'.,J\-1)
900 8(K\,J\-1)=8(K'.,J\)
910 B(K\,J\)=AT
920 NEXT K\
930 J\=J'\.-1
940 IF (J\-1)>0 GOTO 830
950 NEXT I\
960 REM PRINT OUT EIGENVALUES AND EIGENVECTORS
980 VDU2
990 FOR I\=1 TO 0\
278 APPENDIX 3
1000 FOR J\=1 TO 0\
1010 IF J\<>I\ GOTO 1030
1020 PRINT"Eigenvalue "I\" ="A(I\,J\)
1030 NEXT J\
1040 FOR K\=1 TO 0\
1050 PRINT"Eigenvector"I\", "K\"
="B(K\,I\)
1060 NEXT K\
1070 PRINT'
1080 NEXT I\
1090 REM SWITCH OFF PRINTER AND RETURN TO START
1100 VDU3
1110 GOTO 120
1120 END
Program 6.2 Program to find the eigenvalues and eigenvectors of a real symmetrical
matrix up to size 10 x 10. Based on the Jacobi method.
10 REM DEF PROCMpow

20 REM TAKE REM OUT OF PREVIOUS STATEMENT TO USE AS A PROCEDURE
30 REM PROGRAM TO POWER A MATRIX
40 REM PRINCIPLE USE WITH TRANSITION PROBABILITY MATRICES
50 REM BASED ON HARBAUGH AND BONAHM CARTER (1970) P.119
60 DIM TM(25,25),PM(25,25), pmat(25,25)
70 REM SET PARAMETERS TO PRINT RESULTS
80 @\=&01020205
90 REM INPUT DATA
100 INPUT "TYPE NUMBER OF LITHOLOGIES, TYPE 1 TO TERMINATE
EXECUTION",M
110 IF M=1 GOTO 440
120 INPUT"TYPE NUMBER OF STEPS",N\
130 FOR I=1 TO M:FOR J=1 TOM
140 INPUT"TYPE RAW DATA, ROW WISE",TM(I,J)
150 NEXT J: NEXT I
170 VDU2
180 FOR I=1 TO M:FOR J=1 TO M
190 PM(I,J)=TM(I,J)
200 NEXT J:NEXT I
220 PRINT' '"INPUT MATRIX"
230 FOR I=1 TO M:PRINT' :FOR J=1 TO M
240 PRINT TM(I,J);
250 NEXT J:NEXT I
260 FOR !.=1 TO N\
270 L 1.=L+ 1
280 PRINT'" "INPUT MATRIX TO THE POWER" ,L1\
290 FOR I=1 TO M: FOR J~1TO M
300 pmat(I,J)=O
310 FOR K=1 TO M
320 pmat(I,J)=pmat(I,J)+TM(I,Kl
*PM(K,J)
330 NEXT K: NEXT J: NEXT I
340 FOR I=1 TO M: FOR J~1 TO M
350 PM(I,J)=pmat(I,J)
360 NEXT J
370 PRINT'
380 FOR J=1 TO M
390 PRINT PM(I,J);
400 NEXT J: NEXT I: NEXT L
410 REM SWITCH OFF PRINTER AND CLEAR SCREEN
420 VDU3:CLS
430 GOTO 100
440 END
Program 6.3 Program to raise a matrix to any given power. Handles matrices up to
25 x 25 in its present form.
10 REM PROGRAM TO MULTIPLY A MATRIX BY ITS TRANSPOSE
20 REM THIS PROGRAM CALLED MULTR
30 REM ALSO CALCULATES THE DISPERSION OR CORRELATION MATRH'ES
40 DIM X(50, 10),XMT(10, 10),XMEAN(10),XDEV(10)
50 REM SET PARAMETERS FOR PRINTING OUT RESULTS
60 @\=&0102020A
70 REM INPUT MATRIX SIZE, TITLE AND DATA
80 INPUT"TYPE NUMBER OF ROWS= NUMBER OF VARIABLES"' "VALUE GREATER
THAN 50 TERMINATES EXECUTION",N'.
90 IF N\>50 GOTO 610
100 INPUT"TYPE NUMBER OF COLUMNS NUMBER OF SAMPLES",C\
110 INPUT"TYPE TITLE"T$
120 PRINT"INPUT MATRIX ROW WISE"
130 NS=C\
140 FOR I\=1 TO N\
150 FOR J\=1 TO C\
160 INPUT"ELEMENT VALUE",X(I\,J\)
170 NEXT J\
180 NEXT I\
190 REM CHOOSE TYPE OF CALCULATION
200 REM 1 = USE RAW DATA ONLY
210 REM 2 = CALC. EITHER DISPERSION OR CORRELATION MATRICES
220 INPUT"DO YOU WISH TO USE RAW DATA ONLY?"'"IF YES TYPE 1, IF NO
TYPE 2",M
230 IF M=1 GOTO 270
240 PROCTRSF
250 GOTO 360
260 REM SWITCH ON PRINTER AND PRINT OUT INPUT DATA
270 VDU2
280 PRINT' '"TITLE:",T$
290 PRINT' '"INPUT MATRIX"' '
300 FOR I\=1 TO N\
310 PRINT'
320 FOR J\=1 TO C\
330 PRINTX(I\,J\);
340 NEXT J\
350 NEXT I\
360 REM MULTIPLY MATRIX BY ITS TRANSPOSE
370 FOR I\=1 TO N\
380 FOR J\=1 TO N\
390 XMT(I\,J\)=0
400 NEXT J\
410 NEXT I \
420 FOR I\=1 TO N\
430 FOR J\=1 TO N\
440 FOR K\=1 TO C\
450 XMT(I\,J\)=XMT(I\,J\)
+X(I\,K\)*X(J\,K\)
460 NEXT K\
470 NEXT J'.
480 NEXT I\
490 PRINT' '"RESULTANT MATRIX"' '
500 FOR I\=1 TO N\
510 PRINT'
520 FOR J\=1 TO N\
530 PRINT XMT(I\,J\);
540 NEXT J\
550 NEXT I\
560 PRINT'' "This data set finnished"
570 REM SWITCH OFF PRINTER AND CLEAR SCREEN
580 VDU3:CLS
590 PRINT"RETURNING TO BEGINING OF PROGRAM"'"FOLLOW INSTRUCTIONS TO
TERMINATE THE EXECUTION"
600 GOTO 80
610 END
620 REM
630 REM
640 REM
650 DEF PROCTRSF
660 REM PROCEDURE TO NORMALIZE RAW DATA MATRIX USING MEAN OR MEAN
AND STANDARD DEVIATION
670 REM SWITCH ON PRINTER AND PRINT OUT ORIGINAL DATA
280 APPENDIX 3
680 VDU2
690 PRINT' '"TITLE,",T$
700 PRINT' '"INPUT MATRIX"''
710 FOR I\=1 TO N\
720 PRINT'
730 FOR J\=1 TO C\
740 PRINTX(I\,J\);
750 NEXT J\,NEXT I\
760 VDU3
770 REM CALC MEAN FOR EACH VARIABLE
780 FOR I\=1 TO N\
790 XSUM~O.O
800 FOR J\~1 TO C\
810 XSUM~XSUM+X(I\,J\)
820 NEXT J\
830 XMEAN(I\)~XSUM/NS
840 NEXT I\
850 REM SUBTRACT MEAN FROM EACH DATA VALUE
860 REM DESTROYS ORIGINAL DATA MATRIX
870 FOR I\~1 TO N\
880 FOR J\~1 TO C\
890 X(I\,J\)~X(I\,J\)-XMEAN(I\)
900 NEXT J\:NEXT I\
910 REM CLEAR SCREEN AND SWITCH OFF PRINTER
920 CLS 'VDU3
925 PRINT"PROGRAM WILL NOW CALCULATE EITHER"'"THE DISPERSION MATRIX
OR"'"THE CORRELATION MATRIX, YOUR CHOICE"
930 INPUT"DO YOU WISH TO CALCULATE THE DISPERSION MATRIX?"' "IF YES
TYPE 2, IF NO {IMPLIES CORR. MATRIX), TYPE 1",R
940 IF R=1 GOTO 1050
960 VDU2
970 PRINT' '"DISPERSION MATRIX HAS BEEN CHOSEN"'" MATRIX WHICH
FOLLOWS HAS BEEN NORMALIZED, WITH A MEAN OF ZERO FOR EACH VARIABLE"
980 FOR I\~1 TO N\
990 PRINT '
1000 FOR J%~1 TO C\
1010 PRINT X(I\,J\);
1020 NEXT J\
1030 NEXT I\
1040 GOTO 1260
1050 REM CALCULATE STANDARD DEVIATION
1060 FOR I\~1 TO N\
1070 XSUM~O.O
1080 FOR J\=1 TO C\
1090 XSUM=XSUM+X(I\,J\)*X(I\,J\)
1100 NEXT J\
1110 XDEV(I\)~SQR(XSUM/NS)
1120 NEXT I\
1130 REM DIVIDE EACH VALUE BY THE CORRESPONDING STD. DEV.
1140 FOR I\~1 TON\
1150 FOR J\~1 TO C\
1160 X(I\,J\)~X(I\,J\)/XDEV(I\)
1170 NEXT J\, NEXTI\
1190 VDU2
1200 PRINT' '"CORRELATION MATRIX HAS BEEN CHOSEN"'" MATRIX WHICH
FOLLOWS HAS BEEN NORMALIZED, WITH A MEAN OF ZERO"'"AND STANDARD
DEVIATION OF UNITY FOR EACH VARIABLE"''
1210 FOR I\=1 TO N\
1220 PRINT'
1230 FOR J\=1 TO C\
1240 PRINTX{I\,J\);
1260 ENDPROC
Program 6.4 Program to calculate either the dispersion matrix or a correlation

matrix. Based on the transposition method outlined in the text. Maximum number of
samples= 50, maximum number of variables= 10.
5 REM PROGRAM TO SOLVE SIMULTANEOUS LINEAR EQUATIONS USING
GAUSSIAN ELIMINATION, MULTIPLE DATA SETS PROCESSED
10 DIM A(10,11),C(11),X(11)
15 REM READ IN DATA MAXIMUM NO. OF EQUATIONS IS 10, TO TERMINATE
RUN ENTER VALUE > 10
20 INPUT"NUMBER OF EQUATIONS"'"MAXIMUM ALLOWED IS 10"'"A NUMBER
GREATER THAN 10 TERMINATES THE EXECUTION",N\
30 IF N\>10 THEN END
40 M'.=N\+1
50 PRINT"INPUT COEFFICIENTS"'"ONE AT A TIME, ROWISE"
60 FOR I\=1 TO N\,FOR J\=1 TO M\
70 INPUT A(I\,J\)
80 NEXT J\, NEXT I\
90 VDU2
100 PRINT"' "DATA ENTERED IS"'"
110 FOR I\=1 TO N\,FOR J\=1 TOM\
120 PRINT A(I\,J\);
130 NEXT J\,NEXT I\
140 VDU3
145 REM FIND LARGEST PIVOT AND RE-ORDER EQUATIONS
150 FOR R\=1 TO N\-1
160 Q\=R\
170 P\=Q'.
180 IF Q\ = N\ GOTO 220
190 Q'.=Q\+ 1
200 IF A(P\,R\)<A(Q\,R\) GOTO 170
210 GOTO 180
220 FOR J\=R\ TO M\
230 C(J\)=A(R\,J\) ,A(R\,J\)=A(P\,J\)
,A(P\,J'.)=C(J'.)
240 NEXT J'.
245 REM CALCULATE UNKNOWNS USING GAUSSIAN ELIMINATION
250 FOR I\=R\+1 TO N\
260 FOR J\=R\+1 TO M\
270 A(I\,J\)=A(I\,J\)-(A(I\,R\)
*A(R\,J\))/A(R\,R\)
290 NEXT R'
300 X(N\)=A(N\,M\)/A(N\,N\)
310 FOR R\=N\-1 TO 1 STEP -1
320 D=O
330 FOR J\=R\+1 TO N\
340 D=D+A(R'.,J\)*X(J\)
350 NEXT J'r.
360 X(R\)=(A(R\,M\)-D)/A(R\,R\)
370 NEXT R\
375 REM FIND DETERMINANT AS THE PRODUCT OF THE PIVOTS
3 80 D=A ( 1 , 1 )
390 FOR I\ = 2 TO N\
400 D=D*A(I'., I \ )
410 NEXT I\
415 REM OUTPUT RESULTS
420 VDU2
430 PRINT''' "SOLUTION IS"'"UNKNOWNS"
440 FOR J\=1 TO N\
450 PRINT X(J\)' NEXT J\
460 PRINT''' "THE PIVOTS ARE"
470 FOR I\=1 TO N\
480 PRINT A( I\, I\) ,NEXT I\
490 PRINT'' '"DETERMINANT IS"
500 PRINT D
510 VDU3
515 REM RETURNS TO START TO PICK UP NEXT DATA SET
520 GOTO 20
530 END
Program 7.1 Program to solve simultaneous linear equations using Gaussian

elimination. The largest pivots are chosen at each stage. The maximum number of
equations is I0.
282 APPENDIX 3
10 REM PROGARAM TO SOLVE SIMULTANEOUS LINEAR EQUATIONS USING THE
GAUSS-SEIDEL METHOD
15 REM USES METHOD OF SUCCESSIVE CORRECTIONS
25 REM TOLERANCE (TLRl SET AT 0.0001, MAXIMUM ITERATIONS (KK%)
1000
30 REM BASED ON ORIGINAL FORTRAN PROGRAM BY FERGUSON
35 REM INITIAL ESTIMATES OF X SET TO ZERO
40 DIM X(10),COEF(10,10),D(10)
50 REM READ IN DATA
60 VDU3
70 INPUT "TYPE NUMBER OF UNKNOWNS"'"MAXIMUM 10"' "A LARGER VALUE
TERMINATES THE RUN",N\
80 IF N\ > 10 THEN END
90 CLS
100 INPUT"TYPE TITLE FOR THIS RUN",TITLE$
110 PRINT "INPUT COEFFICIENTS"'"1 AT A TIME, ROWISE"
120 FOR I\ = 1 TO N\: FOR J\ = 1 TO N%
130 INPUT COEF(I\,J\)
140 NEXT J.
150 INPUT"REM. THIS IS VALUE FOR"'"R.H.S.OF EQUATION",D(I"o)
160 NEXT I \
170 CLS
171 VDU2
172 PRINT''' "DATA ENTERED IS"'''
173 FOR I\ = 1 TO N\: FOR J\ = 1 TO N\
174 PRINT COEF(I\,J\);
175 NEXT J.
176 PRINT D(I\)'
177 NEXT I \
178 VDU3
180 REM SET CONSTANTS AND DUMMY LISTS TO ZERO
190 TLR = 0.0001
200 KK\ = 0
210 FOR I\ = 1 TO N\
220 X(I\)=0
230 NEXT I\
240 REM CALCULATE VALUES OF X(I\)
250 FOR I\ = 1 TO N\
260 DUM ~ 0.0
270 DUM=DUM+D(I\)
280 FOR J\ = 1 TO N\
290 IF I\ = J\ GOTO 310
300 DUM= DUM-(COEF(I\,J\)*X(J\))
310 NEXT J\
320 X(I\)=(1.0/COEF(I\,I\ll*DUM
330 NEXT I \
340 REM TEST IF ENOUGH ITERATIONS USING TOLERANCE
350 DD=O.O
360 FOR J\ = 1 TO N\
370 DO= DD+(X(J\)*COEF(1,J\))
380 NEXT J\
390 IF ABS (0(1)-00) < T~R GOTO 430
400 KK'-=KK'\+1
410 IF KK\ = 1000 GOTO 490
420 GOTO 250
430 VDU2
440 PRINT"TITLE IS ",TITLE$,' "SOLUTION OF EQUATIONS"'"NUMBER OF
ITERATIONS",KK\
450 FOR I\=1 TO N\
460 PRINT X(I\)
470 NEXT I\
480 GOTO 60
490 VDU2
500 PRINT"NO SOLUTION AFTER 1000 ITERATIONS"
510 PRIN'!" '"TITLE IS ",TITLE$, "'ESTIMATE OF X( 1) - X(N)"' "AFTER
1000 ITERATIONS" ' ' '
520 FOR I\=1 TO N\
530 PRINT X(I'.)
540 NEXT I\
550 VDU3
560 GOTO 60
570 END
Program 7.2 Program to solve simultaneous linear equations by the Gauss-Seidel
method, using successive corrections. The maximum number of equations is 10.
10 REM PROGRAM TO SOLVE SIMULTANEOUS LINEAR EQUATIONS BY MATRIX
INVERSION
20 REM MATRIX INVERSION BASED ON DAVIS (1973) P.142
30 REM COEF IS THE MATRIX OF COEFFICIENTS, ICOEF IS THE INVERSE
40 REM Z IS THE VECTOR OF VALUES FOR THE RHS OF THE EQUATION, DET
IS THE DETERMINANT
50 REM THE COEFFICENT MATRIX IS REDUCED TO AN IDENTITY MATRIX
60 REM THE MAXIMUM MATRIX SIZE IS 10*10
70 REM X IS THE VECTOR OF UNKNOWNS
80 DIM COEF(10,10),ICOEF(10,10),
Z( 10) ,X( 10)
90 REM READ IN AND PRINT OUT DATA
100 CLS
110 VDU3
120 INPUT"TYPE IN NUMBER OF UNKNOWNS"' "A VALUE > 10 TERMINATES THE
RUN", N\
130 IF N\>10 GOTO 670
140 PRINT"TYPE IN COEFFICIENTS"'"ONE AT A TIME, ROWISE"
150 FOR I\=1 TO N\, FOR J\=1 TO N\
160 INPUT COEF(I\,J\)
170 NEXT J\
180 INPUT"VALUE FOR RHS OF EQUATION", Z ( I")
190 NEXT I\
200 VDU2
210 PRINT' '"DATA IS"'''
230 PRINT COEF(I\,J\);
240 NEXT J\
250 PRINT Z(I\)
260 NEXT I'.
270 REM SET ICOEF TO IDENTITY MATRIX
290 ICOEF(I\,J\)=0
300 NEXT J.
310 X(I\J=O
320 ICOEF(I\,I\)=1.0
330 NEXT I\
340 DET=1.0
350 REM FIND INVERSE OF COEFF. MATRIX
360 FOR I\=1 TO N\
370 IDIV=COEF(I\,I\)
375 IF IDIV=O GOTO 515 ,REM THIS IMPLIES DET 0
380 DET = DET*IDIV
390 FOR J\=1 TO N\
400 COEF(I\,J\)=COEF(I\,J\)/IDIV
410 ICOEF(I\,J\)=ICOEF(I\,J\)/IDIV
420 NEXT J\
430 FOR J\=1 TO N\
440 IF(I\-J\)<0 OR (I\-J\)>0 GOTO 450 ELSE GOTO 500
450 IRATIO=COEF(J\,I\)
460 FOR K\=1 TO N\
470 COEF(J\,K\)=COEF(J\,K\)-IRATIO
*COEF(I\,K.)
480 ICOEF(J\,K\)=ICOEF(J\,K\)-IRATIO *ICOEF(I\,K\)
490 NEXT K\
500 NEXT J'.
510 NEXT I'.
512 GOTO 520
515 PRINT''' "DETERMINANT OF COEFFICIENT MATRIX ZERO! "THEREFORE NO
SOLUTION", GOTO 110
520 REM PRINT OUT INVERSE OF COEFFICIENT MATRIX
530 PRINT'' '"INVERSE OF COEFFICIENT MATRIX IS"'''
540 FOR I\=1 TON\, FOR J\=1 TO N\
550 PRINT ICOEF(I\,J\);
560 NEXT J\,PRINT'' NEXT I'.
5 70 PRINT ' " "DETERMINANT =", DET, " '
580 REM CALCULATE UNKNOWNS AND PRINT RESULTS
590 FOR I'.=1 TO N\,FOR J'.=1 TO N'.
600 X(I\)=X(I\)+ICOEF(I\,J\)*Z(J'.)
620 PRINT"UNKNOWNS ARE"'''
630 FOR I\=1 TO N\
284 APPEllo'DIX 3
640 PRINT X ( I \ )
650 NEXT I \
660 GOTO 110
670 VDU3
680 END
Program 7.3 Program to solve simultaneous linear equations by the method of
matrix inversion. The maximum number of equations is 10.
10 REM PROGRAM TO SOLVE THE EQUATION d2y/dx2 5X

20 REM FOR THE RANGE OF VALUES X = 0 TO 0.5
30 REM THE INTERVAL DELTA X = 0.05
40 REM USES THE METHOD OF FINITE DIFFERENCES
50 REM SIMULTANEOUS LINEAR EQUATIONS SOLVED BY GAUSS-SIEDEL
60 DIM Y(11)
70 REM BOUNDARY VALUES
80 Y( 1 )=0.0
90 Y(11)=100.0
100 REM REMAINING Y VALUES TO ZERO
110 FOR I=2 TO 10
120 Y(IJ=O.O
130 NEXT I
150 VDU2
160 REM SET UP TITLE FOR PRINTING RESULTS
170 TBTITLE$="ITER. Y1 Y2 Y3 Y4 YS Y6
Y7 Y8 Y9"
180 PRINT '' ,TBTITLE$,'
190 REM ITERATIVE PROCEDURE STARTS NEXT
200 FOR J = 1 TO 99
210 FOR I = 2 TO 10
220 K=I-1
230 Y(I)=(Y(I-1)+Y(I+1)-0.000625KJ
/2.0
240 NEXT I
250 IF J<70 GOTO 370
260 REM CONVERT CURRENT SOLUTION INTO STRINGS FOR PRINTING
270 A$="
280 IF J>9 GOTO 310
290 B$=STR$(J)+"
300 GOTO 320
310 B$=STR$(J)+"
320 FOR I = 2 TO 10
330 A$=STR$(Y(I))
340 B$=B$+LEFT$(A$,6)+"
350 NEXT I
360 PRINT B$,
370 NEXT J
380 REM DISABLE PRINTER
390 VDU3
400 END
Program 8.1 Program to solve the second-order differential equation d'y/dx' = 5x.
See text for details.
References
Atkinson, K. 1985. Elementary numerical analysis. New York: Wiley.
Barnes, J. W. 1981. Basic geological mapping. Milton Keynes: The Open University
Press.
Beerbower, J. R. & D. Jordan 1960. Application of information theory to palaeonto-
logical problems: taxonomic diversity. Journal of Palaeontology 43, 1184-98.
Bischoff, J. L. 1968. Kinetics of calcite nucleation: magnesium ion inhibition and
ionic strength catalysis. Journal of Geophysical Research 73, 3315-22.
Bischoff, J. L. & W. S. Fyfe 1968. Catalysis, inhibition and the calcite-aragonite
problem. (I) The aragonite-<:alcite transformation. American Journal of Science 266,
65-79.
Blackman, R. B. & J. W. Tukey 1959. The measurement of power spectra. New York:
Dover.
Buntebarth, G. 1984. Geothermics- an introduction. Berlin: Springer-Verlag.
Camina, A. R. & G. J. Janacek 1984. Mathematics for seismic data processing and
interpretation. London: Graham and Trotman.
Cheeney, R. F. 1983. Statistical methods in geology. London: Allen and Unwin.
Cobbold, P. R., J. W. Cosgrove & J. M. Summers 1971. Development of internal
structures in deformed anisotropic rocks. Tectonophysics 12, 23-53.
Coli, J. 1982. BBC microcomputer user guide. London: British Broadcasting Corpora-
tion.
Cosgrove, J. W. 1976. The formation of crenulation cleavage. Quarterly Journal of the
Geological Society of London 132, 155-78.
Cubitt, J. M. & B. Shaw 1976. The geological implications of steady-state mechan-
isms in catastrophe theory. Mathematical Geology 8, 657-{)2.
Davis, J. C. 1973. Statistics and data analysis in geology. New York: Wiley.
Doornkamp, J. C. & C. A.M. King 1971. Numerical analysis in geomorphology- an
introduction. London: Edward Arnold.
Elliott, R. E. 1970. Simulation of a productive coal measures sequence. The Mercian

Geologist 3, 319-35.
Elliott, R. E. 1985. Quantification of peat to coal compaction stages, based especially
on phenomena in the east Pennine coalfield, England. Proceedings of the Yorkshire
Geological Society 45, 163-72.
Falini, F. 1965. On the formation of coal deposits of lacustrine origin. Bulletin of the
Geological Society of America 76, 1317-46.
Ferguson, J. 1978. Some aspects of the ecology and growth of the Carboniferous
gigantoprodutids. Proceedings of the Yorkshire Geological Society 42,41-54.
Ferguson, J. 1980. Application of information theory in geological data processing.
Quarterly Journal of the Geological Society of London 137, 107-8.
Ferguson, J. 1982. The application of information theory to trend surface analysis.
Computer applications in geology iii. Geological Society of London Miscellaneous
Paper No. 15, 3--44.
286 REFERENCES
Ferguson, J. 1983. The analysis ofC-H-0 data in kerogen studies using a coefficient
of elimination. Journal of Petroleum Geology 5, 400--8.
Ferguson, J., P. R. Bush & B. A. Clarke 1984. The role of organic matter in the early
diagenesis of carbonate ooids- an experimental study. Journal of Petroleum Geology
7, 245-66.
Fox, W. T. & R. A. Davis, Jr 1971. Computer simulation model of coastal processes in
eastern Lake Michigan. Technical Report No. 5, O.N.R. Task No. 388-092/10-18-
68(414), Contract NOO 14-69--c-0151. Williamstown, Mass.: Williams College.
Friedman, G. M. & J. E. Sanders 1978. Principles ofsedimentology. New York: Wiley.
Garrels, R.N. & F. T. Mackenzie, Jr 1971. Evolution ofsedimentary rocks. New York:
Norton.
Glass, J. C. 1980. An introduction to mathematical methods in economics. New York:
McGraw-Hill.
Gould, P. 1967. On the geographic interpretation of eigenvalues: an initial explora-
tion. Transactions of the Institute of British Geographers 42, 53-86.
Gow, M. M. 1960. A course in pure mathematics. London: Unibooks, English
University Press.
Graham, D., C. Graham & A. Whitcombe 1984. A-level mathematics course com-
panion. London: Charles Letts.
Gray, J. R. 1967. Probability. Edinburgh: Oliver and Boyd.
Guiasu, S. 1977. Information theory with applications. New York: McGraw-Hill.
Harbaugh, J. W. & G. Bonham-Carter 1970. Computer simulation in geology. New

York: Wiley.
Harbaugh, J. W. & D. F. Merriam 1968. Computer applications in stratigraphic
analysis. New York: Wiley.
Henley, S. 1976. Catastrophe theory models in geology. Mathematical Geology 8,
649-55.
Higginbotham, P. G. 1985. Microtab - an all purpose statistical package ( BBC
version). London: Edward Arnold.
Hogben, L. 1967. Mathematics for the million, revised edn. London: Pan.
Hollingdale, S. H. & G. C. Toothill 1975. Electronic computers, revised edn.
Harmondsworth: Penguin.
Horai, K. & G. Simmons 1969. Thermal conductivity of rock-forming minerals.
Earth and Planetary Science Letters 6, 359-68.
Horvitz, L. 1939. On geochemical prospecting. Geophysics 4, 210-25.
James, A. N. & A. R. R. Lupton 1978. Gypsum and anhydrite in foundations of

hydraulic structures. Geotechnique 28, 249-72.
King, L. J. 1969. Statistical analysis in geography. Englewood Cliffs, Prentice-Hall.

Koch, G. S., Jr & R. F. Link 1971. Statistical analysis ofgeological data, vol. 2. New
York: Wiley.
Krumbein, W. C. & F. A. Graybill 1965. An introduction to statistical models in
geology. New York: McGraw-Hill.
Lasaga, A. C. 1981. Rate laws of chemical reactions. Reviews in Mineralogy,

Mineralogical Society of America 8, 69-110.
Lisle, R. J. 1977. Estimation of the tectonic strain ratio from the mean shape of
deformed elliptical markers. Geologie en Mijnbow 56, 140-4.
REFERENCES 287
Means, W. D. 1976. Stress and strain. Basic concepts of continuum mechanics for
geologists. New York: Springer-Verlag.
Miller, A. R. 1981. Pascal programs for scientists and engineers. Berkeley, California:
SYBEX.
Miller, R. L. & J. S. Khan 1962. Statistical analysis in the geological sciences. New
York: Wiley.
Moroney, M. J. 1951. Factsfromjigures. Harmondsworth: Penguin.
Nettleton, L. L. 1976. Gravity and magnetics in oil prospecting. New York: McGraw-
Hill.
Noble, B. 1964. Numerical methods: I. Iteration programming. Edinburgh: Oliver and
Boyd.
Pelto, C. R. 1954. Mapping ofmulticomponent systems. Journal of Geology 62, 501-

11.
Pielou, E. C. 1969. An introduction to mathematical ecology. New York: Wiley-
Interscience.
Prentice, J. E. 1956. Gigantoproductus edelburgensis (Phillips) and related species.
Proceedings of the Yorkshire Geological Society 30, 299--358.
Ragan, D. M. 1968. Structural geology: an introduction to geometrical techniques, 2nd

edn. New York: Wiley.
Ramsay, J. G. & M. I. Huber 1983. Techniques of modern structural geology. Vol.I:
Strain analysis. London: Academic Press.
Reeve, C. R. & J. S. Huxley 1945. Some problems in the study of allometric growth.
In Essays on growth and form presented to D'Arcy Wentworth Thompson, W. E.
LeGros Clark and P. R. Medawar (eds). Oxford: Oxford University Press.
Rendu, J.-M. 1981. An introduction to geostatistical methods of mineral evaluation.
South African Institute of Mines and Metals.
Saunders, P. T. 1980. An introduction to catastrophe theory. Cambridge: Cambridge

University Press.
Sawyer, W. W. 1966. A path to modern mathematics. Harmondsworth: Penguin.
Selley, R. C. 1970. Studies of sequence in sediments using a simple mathematical
device. Quarterly Journal of the Geological Society of London 125 (for 1969), 557-81.
Shannon, C. E. 1948. A mathematical theory of communication. Technical Journal of
the Bell System 27, 379-423.
Shelley, J. 1984. Essentials of Fortran 77. New York: Wiley.
Sherlock, A. J., E. M. Roebuck & M.G. Godfrey 1982. Calculus: pure and applied.
London: Edward Arnold.
Shioya, M. & R. Ishiwatari 1983. Laboratory thermal conversion of sedimentary
lipids to kerogen-like matter. Organic Geochemistry 5, 7-12.
Sloss, L. L. 1962. Stratigraphic models in exploration. Journal of Sedimentary
Petrology 32, 415--22.
Stegna, L. 1961. On the principles of geochemical oil prospecting. Geophysics 26,447-
51.
Stephenson, G. 1973. Mathematical methods for science students, 2nd edn. London:
Longman.
Tasch, P. 1965. Communications theory and the fossil record of invertebrates.

Transactions of the Kansas Academy of Science 68, 322-9.
288 REFERENCES
Tasch, P. 1980. Palaeobiology of the invertebrates, 2nd edn. New York: Wiley.
Thorn, R. 1972. Stabilite structurel/e et morphogenese. New York: Benjamin.
Unwin, D. J. & J. A. Dawson 1985. Computer programming for geographers. London:

Longman.
Verruijt, A. 1970. Theory of groundwater flow. London: Macmillan.
Waddington, C. H. 1974. A catastrophe theory of evolution. Annals of the New York

Academy of Science 231, 32-41.
Wang, H. F. & M. P. Anderson 1982. Introduction to groundwater modeling. New
York: W. H. Freeman.
Williams, C. B. 1964. Patterns in the balance of nature. London: Academic Press.
Wilson, C. D. V. 1980. A seismic survey of the former channel of the River Doe,
Ingleton. Proceedings of the Yorkshire Geological Society 42, 617-20.
Woodcock, A. E. R. & T. Poston 1974. A geometric study of elementary catas-
trophes. Lecture notes in mathematics 373. Berlin: Springer-Verlag.
Zeeman, E. C. 1971. The geometry of catastrophe. Times Literary Supplement, 1556-

7 (10 December).
Zeeman, E. C. 1976. Catastrophe theory. Scientific American 234, 65-83.
Index
Index
accuracy 1.4, 7.1.1, 149, 188 expansion 69

addition Series 3.2.4
axiom 2.3.1, 54, 160 Theorem 80
of a matrix 6.2.1 bit 247
algebraic equations 2 body translation 170
matrix form of 134-5 boundary conditions 217, 222-3, 225,
Algol 10 232-3, 234, 236
allometric growth 31, 2.1, 83, 4.3 Bracklesham Beds Table 2.1
equation 31 buckling 237
amplitude 123--{) equation 236
analytical methods I, 182 instabilities 235
of integration 5.2
analytical solution 117, 227
angle of sliding friction 86 calcite 16, Table 2.2, 2.3, 20, 33, 2.11
angular calculation, speed of 7.1.2
measure 2.2.3 calculator 8, 177, Table 7.1
shear strain 170, 6.7, 172, 6.8 calculus see differentiation and
anhydrite dissolution 7.4.1(b) integration
anisotropic 235 cartesian co-ordinates 116
with respect to thermal catagenesis 253
conductivity 67 catastrophe
with respect to diffusion 234 theory 9.1, 238; geological
anisotropy 236 application 9.1.2, 9.4
approximate value 184 dual-cusp 243
aragonite 16, Table 2.2, 2.3, 20, 33 catastrophic jumps 240, 244-5
ooids 33, Table 2.6, 2.11 change, rate of 78, 82-3
area characteristic
partial 108 equation 150, 151
surface 20 roots 151
arithmetic check-sum 196
mean 6, 58-9, 63, 64-5, 67, Table chord 80, 82, 83, 4.2
3.1, 119, 165--8, Program 3.1, gradient of 78
weighted 66 circular measure 48
series 3.2.1, 63 class interval 254
axis coal measures, Elliott's Episodal
of an ellipse 155, 6.4, 172 Theory for deposition 3-7
of symmetry 29 coefficient 21, 2.5A, 28, 190--1, 196
matrix 143, 197, 203-4, 208; inverse
back-substitution 180, 188, 192, 194 of 205
Basic 10, 265 of binomial expansion 61-2
basic feasible solution 42 of diffusion 232
Bathonian Table 2.1, 158 of elimination 252
BBC Basic 10 of frictional sliding 85
bench tests 179 second differential 84, 87, 4.5, 88
best-fit 17, 18, 4.6 column vector 132, 135
biaxial-stress 50, 2.17, 85 columns of a matrix 132, Table 6.3
bimodality 240, 243 Communication Theory see
binary digit 247 Information Theory
Binomial communication systems 246, 9.5
Distribution 3.3.3(a), 119--20 commutative 138
292 INDEX
compaction 7 applications 4.3.1
of peat 6 cumulative
ratio 6 percentage 18
complementary events 2.3.3, 54 probability matrix 160, Table 6.7
complex Crammer's Rule 143, 150, 202,
eigenvalues 153 7.3.3(a)
numbers 2.1.6, 152 crenulation cleavage 8.3.4
roots 29, 88, 90, 152 crystal structure 66, 67
compression 135 cusp catastrophe 240, 244, 9.4
computer cyclicity 6.4.1(b)
languages 1.3.1; Algol 10;
Basic 10, II, 265; BBC Darwin's theory of evolution 244,
Basic 10; Fortran 10; Fortran 9.4(b)
iv 10; Fortran 77 10, 12; decay
Pascal 10, 12 equation 32, 226, 229
mainframe 8-9 of radioactive isotopes 32
micro 8-9 definite integral II 0, 5.2.2, 119
programs Appendix 3 deltaic deposition 1.2
simulation models 36 dependence 253-4
concave dependent
down 88,91 events 250
up 88, 91 variable 21, 99, 223
conditional probability 2.3.4 derivative 78, 80, Ill, 219-20, 221,
conditioning 189, 206 223
constant mass model 8.3.1(b) first order 219, 220, 231, 241
constant higher-order 84, 221
error in 178 partial 103, 223-4
in equations 21, 42, 102 second order 94, 220, 231
of integration Ill, 112, 217 third order 84
constraints 37-42, 2.12, 2.13, 94, 95 of a function with a constant
equation 94-5, 96 exponent 255
continuous of a numerical constant 256
distribution 119 of a variable function with a
function 4.1.2, 4.1(a), 105, 119 constant exponent 258
uniform distribution 120 of a variable with respect to
control space 240 itself 255
converge 182, 7.2 derived function 78
convergence 58, 182, 185, 206 determinants 133, 6.2.5, 142, 145,
tests for 7.2.1(c) 151-2, 155, 172, 173-4, 175,202-4
convergent series 3.1.4 geometric interpretation 141-2, 6.3
co-ordinate transformation rules 145
directions 235 deviation
of a point 135, 142 squared 97, 103, 165
pairs 172, Table 6.13 diagenesis 126, 253
transform equations 170, 172 difference matrix 164, Table 6.8, 165
correlation differential calculus see differentiation
coefficient 154, 156, 166, 6.4 differential
matrix 155-6, 6.4.2(b), 168, Table coefficient 78
6.12, Program 6.4 equations 216, 8.1;
negative 154 applications 8.3; first order 216;
positive 154 order of 216; ordinary 216;
self- 154 partial 216, 236; second
cosine rule 46 order 216, Program 8.1
cost function 94-5 topology 239
cubic equation 24, 2.7, 87, 4.5, 88, differentiation 4.2
240 application of rules Appendix 1.1
INDEX 293
applications 82-3, Table 4.1, 4.3, ellipticity 172, 174
4.2.5, 4.3, 4.5, 4.3.1, 4.6, 4.7 elongation 135
definitions 4.2.1 energy function 239-40, 242, 9.3
from first principles 4.2.2 entropy 248
partial 4.2.4, 95, 99, 103 ratio 254
successive 4.2.3 weighted 252
use of standard forms Appendix Epithyris oxonica 13, Table 2.1 (a),
1.2, Table A.l.l 2.1, 35, 71, Table 3.5
implicit Appendix 1.3 equality constraints 94
diffusion equation see under type
coefficient 232, 234 equilibrium surface 239
gas in rock 8.3.3 erosion 229
dip and strike 46-7 error 7.1.1
discontinuous function 76, 4.l(b) arithmetic 246
discrete distributions 3.3.3, 119--21 measurement 102
dispersion matrix 6.4.2(a), Table 6.1 0, round-off 2, 177-8, 208
Program 6.4 sources of 177-9
displacement theory 179
equation 168, 171 estimation
function 236 of mean 3.3.1
vector field 171 of probability 70-1
dissolution rate 7.4.1(b) Eulerian equation 171
distance event
coefficient 45 complimentary 2.3.3
in geometric terms 43, 2.14 independent 52-3
distributions see under type mutually exclusive 53
diverge 182, 184 evolution 244, 9.4(b)
divergence, of a system 240, 243 exact coefficients 194
divergent series 58 expectation 73-4
diversity expected
generic 252 probability 74, Table 3.6
index 208; calculation of 7.4.1(a), random matrix 164, Table 6.8
7.4, 9.2.2(b) exponent II
of a population 251 exponential
specific 252 equation 32
division of matrices 6.2.7 growth equation 64, 8.3.1
dominant harmonic 125--{) series 3.2.5, 73
double extension
angle formulae 2.2.4(c) principal finite 172, 174
integration 117
dual-cusp catastrophe 243 facies relationship diagram 165
dummy unknowns 95 factor formulae 2.2.4(e)
dynamic 242 factorisation 2.1.3
factors 27
economics 96 feasible space 43
eigenvalues 6.3, 6.4, 172, I 74--5, Fick's Law 232, 234
Program 6.2 filter 121
complex 153 theory of 128
eigenvectors 6.3, 6.4, 172, 174, finite differences 8.2, Program 8.1,
Program 6.2 Table 8.1, 234
elementary catastrophes 239, 240 finite differences quotient 220
elements of a matrix 132 first -order
Elliott's Episodal Theory 1.2 derivatives 219, 220, 241
ellipse differential; coefficient 98;
major axes I 56, 6.4, 172 equation 216
minor axes 156, 6.4, 172 reaction 32, 8.3.1(a)
294 INDEX
Fisher's a distribution 75 graph 20, Program 2.1
use as a diversity index 7.4.l(a), 7.4 gravity measurement 8.3.2
flow of a liquid 8.2.2 ground water flow 226
Fortran 10 growth 15--16, 2.1, 2.2
Fortran iv 10 allometric 31, 2.9, 83, 4.3;
Fortran 7 10 equation 31
fossilization (as an information exponential 64, 8.3.1
system) 246, 9.2.2(a), 9.5 of a population 32
Fourier series 35, 5.5, Program 5.2 rate of 20, 82, 4.3
infinite 122
numerical approximation to 122 half angle formulae 2.2.4(d)
fraction, proper 76 half-life 229
function harmonic
continuous 4.1.2. 4.1 (a) mean 59, 65, 67, Program 3.1;
discontinuous 77, 4.1(b) weighted 66
linear 37 series 58, 3.2.3, 63
notation 2.1.5, 4.1 simple 121
objective 37-8, 39, 41 heat conductor 65
rational 76 Homoeorhynchia acute 81, Table 4.1,
function of a function rule Appendix 4.3
1.1 (f) homogeneous strain 172, 173, 6.8
hydrocarbon gasses 212, Table 7.3,
gas concentration 234 7.5, 231
Gaussian Elimination 180, 7.3.1(b), hystresis 240, 243-4
Table 7.2, Program 7.1, 214
with inexact coefficients 7.3.1(c) identity matrix 133, 143, 149, 152
Gauss-Jordan Elimination 7.3.1(a), ill-conditioned 189, 201, 7.3.4, 7.3
Table 7.2 imaginary number 28, 36
Gauss-Seidel Iteration 180, 7.3.2, 7.2, implicit
Program 7.2, 222, 226 differentiation Appendix 1.3
general solution 217 functions 259
generic diversity 252 independent
geographical components 67
co-ordinates 117 events 52-3, 249-50
geological variable 21, 26, 219, 223
processes 7 inexact coefficients 194, 214
time 32, 229-30 infinity 3.l.l
geometric inflexion, point of 88, 90-1
mean 59, 63-5, Table 3.1, 3.1, 61, Information
Program 3.1 theory 9.2; geological
scale 18 applications 9.2.2
series 3.2.2, 63 source 246-7
geometrical interpretation of integral
determinant 141-2, 6.3 calculus 105
eigenvalues and eigenvectors 155, definite 5.2.2
6.4 indefinite 5.2.1
integration 113, Table 5.1, 5.4 integrand 105
geometry 2.2 integration 5.1
Gigantopro"ductus injlatus 30, Table double 117
2.4, Table 2.5, 2.10 numerical 5.1.1; Simpson's
gradient Rule 107-10, l!9, Program 5.1
temperature 65 rules of Appendix 2.1
of a chord 78, 82, 4.2 standard forms Appendix 2.2;
of a tangent 78, 4.2 Table A2.1
grain size 17-19, Table 2.3, 2.4, 63, substitution Appendix 2.3
Table 3.1 intercept 15, 17, 21, 31
INDEX 295
intersection, point of 21, 183-4, 200, roots 151
7.2 correlation 155, 6.4.2(b), 166, Table
inverse of a matrix 133, 6.2.6, 171, 6.12, Program 6.4
205-6 determinant 6.2.5, 142, 145, 151-2,
iteration 155, 172, 173,202-4
simple 7.2.1 difference 164-5, Table 6.8
Newton-Raphson 7.2.2 dispersion 6.4.2(a), Table 6.8,
Program 6.4
Jacobi method 157, Program 6.2 division 6.2.7
expected random 164, Table 6.8
kerogen 252-3 form of equations 103
higher order 137, 142
Lagrange multipliers 4.4 identity 133, 143, 149, 152
Lagrangian specification 6.4.3(a), 172 inverse 133, 171,205-6
Laplace equation 179, 187 inversion; application of 202,
Laplacian 232 7.3.3(b), Program 7.3; method
laws of chance 51; see also probability of 6.2.6
theory minor 140, 141
least squares see also regression multiplication 6.2.3, 160, Program
criteria 102 6.1; by a matrix 6.2.3(c); by a
method of 90, 96 scalar 6.2.3(a); by a
limit 3.1.3, 77, 78, 80-1, 105, 115, 219 vector 6.2.3(b); row-by-column
limits of integration 105; see also rule 135
definite integral normalized 166, Table 6.12
linear notation Table 6.3, 197
-accumulation model 229 of coefficients 143, 198, 201, 203-4,
equation 2.1.1, 96, 102 208
function 37 order of 132, 168
inequalities 37 raised to a power 6.2.4
programming 2.1.7, 2.12, 2.13, 187 real symmetrical !57
trend map 215, 7.5 singular 139, 142, 143
local trend 102, 253-4 square 132, 144, 150, 168, 175
logarithm 12, 31, Table 2.5 strain 170, 17 5
to any base 19 subtraction 6.2.1
to base e 19, 32 transformation 136
to base 10 19 trans1t10n 129, 164, Table 6.1,
logarithmic series 62-3 Table 6.2, 164, Table 6.8;
lognormal distribution 64 probability 6.4.1.(a), Table 6.5;
cumulative 160, Table 6.7;
mainframe computer 8-9 powering 160, Table 6.6,
mantissa 12 Program 6.3
mapping 35 transpose 133, 6.2.2; applications in
marginal cost 96 statistics 6.4.2
Markov unit 133, 143
chains 54 maxima and minima 4.3, 240,
processes 54, 158 Program 4.1
mass balance 229 application of 4.3.1
mathematical model 5 maximisation/minimisation problems
see also; constant mass model; see; Lagrange multipliers; linear
diffusion of gas in rock; Elliott's programming
Episodal Theory mean see under type
matrices in structural geology 6.4.3 mean
matrix in geothermics 3.3.2
addition 6.2.1 value theorem 5.2.3
augmentation 144 measure
characteristic; equation 150, !51; angular 2.2.3
296 INDEX
measure-continued differentials 101, 103, 213
circular 48 differentiation 4.2.4, 95, 99
radian 48 Pascal (computer language) 10
sexagesimal 48 Pascal's triangle 61
measurement error 102 pattern of coefficients 61
micro computer 8 peat
microfaulting 235 compaction of 6
microfolding 235 pore index 6-7
Microtab I pivot 189-91, 196
mistakes (in calculation) 178 choice of 198
mixing, measure of 9.2.2(c) scaling 198
multiple regression 4.5.2, 212 pivotal equation 190-1, 193, 196
multiplication axiom 2.3.2, 54, 68, !58 point
multipliers 190-1, 196 of inflexion 88-9, 90-1
Lagrange 4.4 of intersection 21, 183-4, 200, 7.1
multivariate statistics 156, 165, 187 Poisson distribution 3.3.3(b), Table
mutually exclusive events 53 3.6, 119
polygon 39
natural logarithm 32 polynomial 76, 77, 101
natural logarithmic mean 64 equations 2.1.2
Navier-Coulomb criteria 85 integer series I 02
negative second degree 117
correlation 154 positive number 36
number 28, 36 positive
Newton-Raphson iteration 7.2.2 correlation 154
noise (in data) 126, 246--7 root 182
normal distribution 121 post-multiplication 138
normal stress 50 potential function 239
normalised data matrix 168, Table power spectrum 121, 123, 125, 126,
6.12 Tables 5.2-5.5
number pre-multiplication 138
complex 2.1.6, 152 principal
imaginary 28, 36 axis 175
positive 36 diagonal 129, 133, 189, Table 7.3
rational II extension 172, 174
real 36 stress 50
numerical probability
accuracy 1.4, 7.1.1, 149, 188 conditional 2.3.4
approximation 117, 118 density function 120
integration 5.1.1; Simpson's distribution 5.4
Rule 107, 110, 119, Program 5.1 estimation from data 70-1
methods in geomorphology 26 expected 74, Table 3.6
theoretical 70, 71
objective function 37, 41 theory 2.3, 158
optimisation 4.4 product moment correlation coefficient
order see correlation coefficient
of a matrix 132, 168 product rule Appendix l.l(d)
of differential equations 216 production, last unit of 96
orthogonal 155 programs, computer 9, II
oscillating series see Fourier Series proper fraction 76
Pythagoras' theorem 2.2.1, 132
IT 18, Table 2.3, 63, Table 3.1
palaeoecology 208, 251 quadratic equation 24, 2.6, 36, 97,
partial 100-1, 151
areas 108 solution of 2.1.3, Program 2.2, 155
derivatives 103, 223-4 quadrats 74, Table 3.6
INDEX 297
quartic equation 25, 2.8 infill 4.3.1(b)
sorting 252
radian measure 48 thickness of 90, 93
random 67, 102 sedimentary
numbers 16(}-3 episodes 5
rate rock mass model 8.3.1(b)
of change 78,82-3 processes 3
of dissolution 7.4.1(b) sedimentation, near-shore 35
of growth 20, 79, 4.3 sedimentology 3, 20
rational function 76 self-correlation 154
reaction Semip/anus latissimus 4.3.1(a). 4.6
first order 32, 8.3.1(a) sequence 56
rate 33 sequential data 121
second order 211 series 56
real number 36 applications 3.3
receiver 246 arithmetic 3.2.1, 63
reciprocal (of a matrix) 143 binomial 3.2.4
recurrence relationship 183, 184, 200, convergence of 3.1.4
209, 211 convergent 58
redundancy 249, 252 definition 3.1.2
reflection 136, 142 divergent 58
regional trend 102, 253-4 exponential 3.2.5, 73
regression 4.5.1 finite 57
coefficients I 04; calculation Fourier 35, 5.5, Program 5.2
of 7.4.2 geometric 3.2.2, 63
multiple 4.5.2, 212 harmonic 58, 3.2.3, 63
relative entropy 248, 252 infinite 57
resident integer variables 179 logarithmic 62-3
residuals 180, 198, 214, 253 sum to infinity 58
vector of 197 time 3.1.5, 121
roots sexagesimal measure 48
complex 29, 88, 90, 152 shear 135, 6.3, 172
of an equation 27, !52 failure 85
of a number (calculation) 186-7 matrix 170, 175
real 29, 152, 184-5 plane 50, 4.2.5
rotation 142 simple 168, 6.7, 172, 6.8
round off error 2, 177-8, 208 strain 170
row-by-column rule 135, 137 stress 50, 86
row vector 132 significant figures 177
rows (of a matrix) 132 simple iteration 7.2.1
Simpson's Rule 107-10, Program 5.1
sample interval 129-31, Table 6.2, simulation 5
163, Table 6.7 simultaneous linear equations 23, 41,
scalar 132, Table 6.3, 134 96, 101, 104, 142-3, 150, 153, 221,
scale, geometric 18 231
Scrobicularia 73-4, Table 3.6 applications 7.4.2
second solution of 7.3; Cramers'
degree polynomial 117 rule 7.3.1(a), Program 7.1;
derivative 95, 220, 231 Gaussian elimination 7.3.1(b),
differential coefficient 84, 87, 4.5, 88 Program 7.1; Gauss-Seidel
second order method 7.3.2, 7.2; simultaneous
differential equations 216, Program corrections 200; successive
8.1 corrections 201, Program 7.2;
kinetic reactions 211 matrix algebra 7.3.3.; matrix
sediment inversion 7.3.3(b), Program 7.3
grain size, analysis of 17-19 sine rule 45
298 INDEX
singular matrix 139, 142, 143 sums
size classes 18 ofproducts 213
slack variable 41 of squares 99, 100,213
slope surface area 20
of a tangent to a curve 220, see also symmetry, axis of 29
gradient synthetic elimination 144-9, 205,
of a line 13-17, 21, 31; sign of 14, Program 7.3
17
solution tangent 13, 16
general 217 gradient of 78, 4.2
particular 21 7 tectonic strain 65
specific diversity 252 temperature gradient 65
speed 1.4; of calculation 7.1.2 tensors 175
Spirifer 172, Table 6.13, 6.8 theoretical pro ba bili ty 70, 71
square thermal conductivity 65--6
matrix 132, 144, 150, 168, 175 of minerals Table 3.2
root 28, 36 thickness
standard deviation 6, 168, Table 6.11, of infill 4.3.1(b)
Table 6.12 of sediment 90, 93
statistical third order matrix 142, 175
applications time series 3.1.5
packages I topological surface 240
statistics, multivariate !56, 165, 187 topology 239
steady-state diffusion 232, 234 torque 231
straight line 18-19 torsion balance 8.3.2
equation of 2.l.l(a), 96 transformation 6.2.3(b), 142, 174
power series 129 equation 172
strain rules for determinants 145
homogeneous 172, 173, 6.8 transition
matrix 170, 175 frequency 158, Table 6.4
plane 235 matrix 129, Table 6.1, Table 6.2,
ratio 65 164, Table 6.8; powering 160,
tectonic 65 Table 6.6, Program 6.3
stratigraphic numbers 131
data 129, 6.1, Table 6.1, Table 6.2; probability matrix 6.4.1(a), Table
analysis 6.4.1 6.5
sequence 60, 131, 6.1, 163 relationship 129
simulation 160-3 translation 170
stress 175 transmitter 246
biaxial 50, 2.17, 85 transpose of a matrix 133, 6.2.2, 168,
compressive 235 Table 6.12
loading 6 tree diagram 6.5
normal 50, 2.17, 85--6 trend
principal 50 local 102, 253-4
vectors 175 regional 102, 253-4
strike 46, 47 trend surface
structural geology 47, 156, 168, 175 analysis 4.5.2, 117, 187, 7.4.2, 7.5,
structure prospecting 234 253
subscripts 132-3 trigonometric
substitution formulae; double and triple
method of 94-5, Appendix 2.3 angle 2.2.4(c); half
subtraction, of a matrix 6.2.1 angle 2.2.4(d); factor 2.2.4(e);
successive differentiation 4.2.3 sum and difference 2.2.4(b)
suitability of method 7.1.3 functions 47, Table 2.8
sum and difference formulae 2.2.4(b) identities 2.2.4
summation 119 ratios 2.2.2; applications 47
INDEX 299
trigonometry 43 vector 104, 132, 150, 154
Turite/la sp. 15, Table 2.1, 2.2, 20 column 132
turning points 87-8, 4.4 direction 132
magnitude 132
uncertainty 247 of residuals 197
underdetermined 41 row 132
uniform distribution 120 volume
unit calculation of 5.3; by double
matrix 133, 143 integration 117-19; by
square 141, 6.3 rotation 116
unstable equilibria 239 numerical approximation of 118
upper triangular form Table 7.2
wavelength 121-8
variability 244
well-conditioned 206
variable
dependent 21, 99, 223
independent 21, 26, 219, 223 Zeeman's catastrophe machine 240-3,
slack 41 9.1. 9.2. 9.3

John Ferguson (Auth.) - Mathematics in Geology-Springer Netherlands (1988)

Uploaded by

Copyright:

Available Formats

John Ferguson (Auth.) - Mathematics in Geology-Springer Netherlands (1988)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

John Ferguson (Auth.) - Mathematics in Geology-Springer Netherlands (1988)

Uploaded by

Copyright:

Available Formats

What is the book about?

What is the book about?

What topics does it cover?

What topics does it cover?

John Ferguson

Analytical and computational methods in engineering rock

SPRINGER-VERLAG LONDON LTD.

First published in 1988

British Library Cataloguing in Publication Data

Library of Congress Cataloging-in-Publication Data

During 15 years of teaching mathematics, computing and statistics

geology, or journals such as the Journal of Mathematical Geology.

Preface and acknowledgements Vll

2 Useful mathematical ideas 13

3 Sequences and series 56

4 Rates of change - differentiation 76

5 Areas and volumes: integration 105

6 Number arrays, the algebra of matrices 129

7 Introduction to numerical methods 176

8 Geological applications of differential equations 216

9 Beyond traditional mathematics 238

Appendix 1 An introduction to the methods of

Appendix 2 An introduction to the methods of

Appendix 3 Computer programs 265

2.1 Measurements of (a) Epithyris oxonica (Buckman),

6.3 Conventional symbols used in matrix notation 133

1.1 Solution of geological problems- are mathematical

A question which is often asked is whether it is necessary for

Again, there is the unexpected answer. The author remembers

1.2 Mathematical formulation of a geological problem

As has already been mentioned, some aspects of geology are more

(1) Deposition takes place during uninterrupted episodes of sub-

cess increases the chance of intra-deltaic deposition at any

In other words, throughout the delta, characteristic sedimentary

(functions and function notation will be discussed in detail in

A, subdelta advance and retreat (leading to changes in the local

Thus, we can say that

In his original paper, Elliott (1970) presents a simulation model

(b) A series of 45 sedimentation sequences cross-referenced to the

Using this approach it was not necessary to define any of the

where Mv = moisture content by volume and Rc = compaction ratio,

where e is the base of natural logarithms and cr is expressed in

Figure 1.1 Diagrammatic representation of the reduction in thickness of peat, from

with 95% moisture content is taken as 20. Brown coals with

1.3 Computational aids

The traditional aids to computation, the slide-rule and the book of

stricted to universities, but now they are in infants' classrooms.

1.3.1 Computers and computer languages

1.3.2 Computers in geology

(1965), Harbaugh & Merriam (1968), Harbaugh & Bonham-Carter

1.4 Speed and accuracy

mantissa is expressed to seven-figure accuracy. Beyond 10- 39 the

1.5 Computer literature

Although computer programming is an important aspect of numeri-

To provide the basic level of mathematics needed to work through

2.1.1 Linear equations

(a) THE EQUATION OF A STRAIGHT LINE

where y = the length of the pedicle valve, x = the width of the

(a) Epithyris oxonica (b) Turitella sp.