P. G. Francis (Auth.) - Mathematics For Chemis

Mathematics for Chemists
Mathematics
for Chemists
P. G. Francis
Department of Chemistry,
University of Hull
LONDON NEW YORK

Chapman and Hall
First publis~d 1984
by Clulpmml and Hall LId
I I New Feller !..mIe. London EC4P 4EE
in Ihe USA by
Publis~d
Chapman and Hall
733 Third Avenue. New York NY 100/7
10 1984 P. G. Francis
$Qftcover reprint of the hardcover 1st edition 1984
ISBN-13: 978-94-01()..8950-O e-ISBN-13: 978-94-1109-5552-3

001: 10.10071978-94-009-5552-3
This litle is available in bolh hardbound and paperback
editions. The paperback edition is sold subjtcllO t~ condition
tlull it slullillDI, by way of trade or ot~rwise, be lenl, re-sold.
hired aul. or olherwise circulated without the publisher's prior
consent in any form of binding or caverOlher titan tha.t in which
il is published and withoul a similar condition including this
condition being imposed on the subseqlUmt purcluJser.
All rights reserved. No part of this book may be reprinted, or
reproduced or utilized in any form or by any electronic ,
mechanicol or other meOllS. now known or hereafler inVf'nted.
including pholocopying and recording. or in any information
storage and retrieool system. wilhoul permission in writing
from Ihe Publisher.
Brhiala Library Cataloauill, ill PliblicatiOil Data
Francis, P.O.
Mathematics for chemists.
I. Chemistry- Mathematics
I. Title
510'.2454 QD42
Francis. P. G. (Patrick Geoffrey), 1929-

Mathematics for chemists.
Bibliography: p.
Includes index
1. Mathcmatics-I961 2. Chemistry-
Mathematics. I. Title.
QA37.2.F73 1984 510'.2454 84- 1802
Contents
Preface ix
1 Algebraic and geometrical methods 1

1.1 Natural numbers 1
1.2 Units and dimensional analysis 4
1.3 Functional notation 8
1.4 Quadratic and higher-order equations 8
1.5 Dependent and independent variables 10
1.6 Graphical methods 12
1.7 Some geometrical methods 14
1.7.1 Similar triangles 14
1.7.2 Triangular graph paper 14
1.7.3 Three-dimensional geometry 15
1.7.4 Circle, ellipse, parabola and hyperbola 17
1.7.5 Plane polar coordinates 21
1.8 Factorials and gamma functions 22
1.9 Probability 23
1.10 Complex numbers 24
2 Differential calculus 29
2.1 Significance and notation 29
2.2 The calculus limit 30
2.3 Differentiation of simple functions 34
2.4 The use of differentials; implicit differentiation 36
2.5 Logarithms and exponentials 38
2.6 The chain rule and differentiation by substitution 39
2.7 Turning points: maxima, minima and points of
inflection 41
2.8 Maxima and minima subject to constraint; Lagrange's
method of undetermined multipliers 45
vi Contents
2.9 Series 50
2.9.1 Geometric series 51
2.9.2 Power series and Taylor's theorem 51
2.9.3 Maclaurin's theorem 54
2.9.4 Inversion of a series 58
2.9.5 Empirical curve fitting by power series 60
2.10 The evaluation of limits by L'Hopital's rule 61
2.11 The principles of Newtonian mechanics 63
3 Differential calculus in three or more dimensions; partial

differentiation 73
3.2 An alternative approach to calculus 77
3.3 The total differential 78
3.4 General expression for a total differential 80
3.5 Exact differentials 82
3.6 Relations between partial derivatives 84
3.7 Extensive and intensive variables; Euler's theorem 86
3.8 Taylor's theorem in partial derivatives 88
3.9 Vectors 89
3.9.1 Scalar and vector products 92
3.9.2 Scalar and vector fields 95
4 Integration 98
4.2 Standard methods of integration 101
4.2.1 Simple functions 101
4.2.2 Reciprocal functions 101
4.2.3 Integration by parts 102
4.2.4 Integration by substitution 104
4.2.5 Expansion of algebraic functions 104
4.2.6 Integration by partial fractions 105
4.3 Standard forms of integral and numerical methods 106
4.4 Multiple integration 107
4.5 Differentiation of integrals; Leibnitz's theorem 108
4.6 The Euler-Maclaurin Theorem 109
Contents vii
5 Applications of integration 115
5.1 Plane area 115
5.2 Plane elements of area 117
5.3 Elements of volume; polar coordinates in three
dimensions 120
5.4 Line integrals 122
5.5 Curve length by integration 123
5.6 Applications of multiple integration 124
5.6.1 The pressure of a perfect gas 124
5.6.2 Interactions in a real fluid 127
5.7 The calculus of variations 129
5.8 Generalized dynamics 132
6 Differential equations 138

6.2 Equations of first order, first degree 140
6.2.1 Separable variables 140
6.2.2 First-order homogeneous equations 142
6.2.3 Exact equations 144
6.2.4 Linear equations of first order 145
6.3 Linear differential equations 147
6.3.1 Homogeneous linear equations with constant
coefficients 148
6.3.2 General linear equation with constant
coefficients 156
6.3.3 Linear equations of second order 159
6.4 Integral transforms 160
6.4.1 Laplace transforms 160
6.4.2 Fourier transforms 163
7 Experimental error and the method of least squares 165

7.1 Significance 165
7.2 Root-mean-square error 166
7.3 Distribution of error 168
7.4 The statistical analysis of experimental data 169
7.5 Propagation of error 170
7.6 Small-sample errors 171
viii Contents
7.7 The normal distribution of error 174
7.8 The method of least squares 176
7.8.1 Linear relation between two variables 176
7.8.2 Covariance and correlation coefficient 179
7.8.3 Uncertainty in the slope and intercept of the
least-squares straight line 180
7.8.4 Least-squares straight line with both variables
subject to error 183
7.8.5 Weighting of observations 183
7.8.6 Multivariable and non-linear least-squares
analysis 184
Appendix SI units, physical constants and conversion factors;

the Greek alphabet and a summary of useful
relations 185
Index 190
Preface
This text is concerned with those aspects of mathematics that are

necessary for first-degree students of chemistry. It is written from the
point of view that an element of mathematical rigour is essential for a
proper appreciation of the scope and limitations of mathematical
methods, and that the connection between physical principles and their
mathematical formulation requires at least as much study as the
mathematical principles themselves. It is written with chemistry
students particularly in mind because that subject provides a point of
view that differs in some respects from that of students of other
scientific disciplines. Chemists in particular need insight into three-
dimensional geometry and an appreciation of problems involving
many variables. It is also a subject that draws particular benefit from
having available two rigorous disciplines, those of mathematics and of
thermodynamics. The benefit of rigour is that it provides a degree of
certainty which is valuable in a subject of such complexity as is
provided by the behaviour of real chemical systems. As an experimen-
tal science, we attempt in chemistry to understand and to predict
behaviour by combining precise experimental measurement with such
rigorous theory as may be at the time available; these seldom provide a
complete picture but do enable areas of uncertainty to be identified.
Mathematical rigour has been provided for us by generations of
professional mathematicians, who continue to give support and advice
both in the application of established techniques and in the develop-
ment of new approaches. Experimental scientists have added to this a
rigour of a different kind, which is based upon a small number of
premises, or axioms, deduced from seemingly disconnected experimen-
tal observations. These provide the experimental laws of thermody-
namics and quantum theory whose justification lies in the absence of
disagreement with experiment. When these are expressed in mathema-
tical terms we can add mathematical rigour to produce theoretical
structures in which only the basic premises, often corresponding to
a simplified physical model, leave room for uncertainty. These
x Preface
considerations are reflected in this text by frequent reference to physical
principles, by the inclusion of a number of mathematical proofs and by
the inclusion of a discussion of the treatment of experimental data.
It is assumed that the reader has previous mathematical knowledge
extending to the elementary applications of calculus. The subject
matter is developed during each chapter, so that an initial reading of
only the earlier sections in each chapter is possible. The later sections
contain brief outlines of the principles and scope of more advanced
techniques and are intended to point the way to more specialized texts.
The examples are, in the main, given with their solutions; this is in
order that the text may be used as a reference source, for which purpose
worked examples are more valuable than exercises. Readers are
strongly recommended to attempt their own solutions to these
examples before consulting the one that is given.
A proper acknowledgement of the many sources from which the
author has drawn cannot be justly be made for a text in which the
material was largely developed during the last two- or three-hundred
years. It is a subject on which many admirable texts have been written,
the only justification for adding to their number being changes in
emphasis.
CHAPTER 1
Algebraic and geometrical

methods
1.1 Natural numbers

Quantitative measurements in science are usually made in terms of
appropriate, but arbitrary, units (Section 1.2), so that the numerical
values obtained depend on the size of the units and are not significant in
themselves. Only when there is a natural unit, such as the gas constant
R, do numerical values assume absolute significance; for example, the
heat capacity at constant volume of an ideal monatomic gas at a high
enough temperature is 3R/2, where the number 3 corresponds to the
three degrees of freedom of the particles.
There are some pure numbers, however, that arise naturally, the ones
most commonly met in science being 7t and e. These can be regarded as
the mathematical equivalent of natural units and occur so often in
scientific formulae and equations that we consider how they originate.
Our first natural number, 7t, could be defined as the ratio of
circumference to diameter for a circle. This ratio is a pure number that
is independent of the size of the circle, and it is also used as the natural
measure of angle. The definition of angle is
angle = (length of arc)/radius, (1.1a)
and for a given angle, this has a numerical value that is independent of
any conventional scale of measurement. The circumference of a
complete circle of radius r being 27tr, the angle corresponding to a full
rotation is 2nr/r = 2n natural units of angle, this unit being called the
radian. We also have the conventional definition of such a full rotation
2 Algebraic and geometrical methods
as 360 degrees (360°), so that
1t radian = 180°.
The definition of angle in two dimensions as (arc length)/radius can
be extended into three dimensions; if we take a conical segment from a
sphere we can define a three-dimensional solid angle as
solid angle = (area of segment of spherical surface)/(radius)2.
(1.1 b)
For a complete sphere of surface area 41tr 2,
complete solid angle = 41tr2/r2 = 41t,
and since this expression does not contain r it is also a pure number that
is independent of the size of the sphere. The unit of solid angle is called a
steradian. It is a useful geometrical concept that (area)/(radius)2 is
constant for a given solid angle, and this is used in Section 5.7.
Our second natural number, e, arises in a different way. It is useful to
divide the mathematical methods used in science into two categories,
geometrical methods and analytical methods. The above definition of 1t
is geometrical and belongs in the field of diagrams and models. The
second category is algebraic and numerical, and does not depend on
our ability to construct models and diagrams. Analytical methods are
developed by strictly logical deductions from first principles, called
axioms. This is generally the more powerful of the two mathematical
methods and, as such, is the one preferred by mathematicians and is
often the one needed to solve the more difficult problems. Geometrical
methods, on the other hand, often seem simpler but are sometimes
deceptively so. The wise rule is to use geometrical methods with care
and to defer to analytical methods when in doubt.
This division into geometrical methods and analytical methods is
arbitrary since they are parts of the same whole and a combination of
the two is frequently used in practice. Sometimes the connection
between the two methods is not obvious, an example being provided by
the trigonometric functions sin (J, cos (J, tan (J, .•.. The geometrical
definition of sin (J is
. (J opposite side
SIn = -"'-=----- in a right-angled triangle, (1.2)
hypotenuse
whereas the analytical definition is
(J3 (J5 (J7
sin (J = (J - 3! + 5! -71 + .... (1.3)
Natural numbers 3
We can see that these two definitions are the same in a particular case
by taking the angle eto be, say, 90° = nl2 radian. Then sin e = 1 by the
geometrical definition, and if we put e = nl2 = 1.5708 into the series
(1.3) it is easily shown that successive terms become rapidly smaller and
smaller, and that the sum tends towards a limit of 1.000.
The fact that the two definitions of sin e are always the same can be
proved only by means of extensive analytical argument. One way is to
use the definition of cos e as the series
e2 e4 (J6
cos (J = 1 - 2! + 4! - 6! + .... (1.4)
This is then the derivative (Chapter 2) of sin (J. The derivative is used to
define the slope of the tangent to a curve, and the angle between two
straight lines can be obtained as the inverse of the cosine series. In
this way, geometry can be developed by the analytical approach
leading eventually to showing that (1.2) and (1.3) are equivalent to
each other.
As mentioned above, our second natural number, e, belongs in
analytical methods. It is defined by the series
x2 x3
e = 1 + x + 2! + 3! + (1.5)
so that when x = 1,
1 1
e = 1 + 1 + 2! + 3! + ... = 2.71828 . . . . . (1.6)
This arises naturally in science because e, and only eX, gives exactly the
same quantity when it is differentiated (Chapter 2). This corresponds to
the quite common situation where the rate of change of a quantity is
proportional to the quantity itself, as in first-order reaction kinetics.
More fundamentally, exponential relations arise as a result of the
simple probability that determines the Boltzmann distribution which
underlies many physical phenomena (Section 2.8).
Putting x = 0 in (1.5) gives eO = 1. This is a particular case of the
general rule that any quantity raised to the power of zero is unity, which
follows from the laws of indices:
and
so that
and when x = y,
1.2 Units and dimensional analysis
The units used in the SI system are given in the appendix together with
numerical values of the fundamental physical constants and conversion
factors. Notice that the symbols for units are written only in the
singular, or they cannot be cleanly cancelled and we would also have
absurdities such as 0.99 g but 1.01 gms. This need not conflict with
colloquial usage; we may speak of a temperature difference of
10 kelvins but this is written as 10 K.
Algebraic symbols are used to denote physical quantities, such as p
for pressure; the symbol denotes the quantity, not the units. A
particular value of the quantity is then the product of a pure number
and its units, and this product follows the normal rules of algebra. Thus
2 atm is the product of the pure number 2 and the unit atm, so that if
p = 2 atm we have p/atm = 2, or a quantity divided by its units is a pure
number. An important particular case is when taking logarithms; since
the definition of a logarithm only applies to pure numbers, the value of
any quantity must first be divided by its units, such as In(p/atm). The
value obtained will then depend on the units being used since, for
example,
In (p/atm) = In(p/101.325 kPa)
= In(p/kPa) -In(101.32S)
= In(p/kPa) - 4.618.
The dimensions of a physical quantity are defined as the appropriate
combination of powers of the fundamentals mass (M), length (L) and
time (T); thus velocity has dimensions LT- 1 and force (mass
x acceleration) has dimensions MLT - 2. The dimensions are precisely
analogous to the SI units of kilogram, metre and second, so that force
has units of kg m S-2.
Dimensional analysis is a method of checking and predicting
relations between physical quantities based simply on the principle that
the dimensions must balance in any equation. A standard example of
the technique is to predict the form for Stokes's law for the drag on a
sphere moving in a viscous fluid. Since
force = mass x acceleration
it has dimensions MLT - 2. The viscosity '1 of a fluid is defined by
force per unit area
'1 = velocity gradient
force distance
= - - x --:--::--
area velocity
Units and dimensional analysis 5
with dimensions
M~ 1 ;t
T'l., x j22 x~ x L = ML -IT-I.
We then assume that the drag (resisting force) on the sphere will depend
on the size of the sphere (radius a), its velocity (v) and on the viscosity of
the fluid. Dependence on temperature will be taken care of through
change in the viscosity of the fluid. The relation that is assumed is
written as
and dimensional analysis will enable us to find the numbers oe, p and y.
The assumed relation is rewritten in terms of the dimensions (or units)
of the quantities, thus
MLT- 2 = L"(LT-I)P(ML -IT-I)y.
Finally, we equate powers of each of the primary dimensions M, Land
T to obtain
for M, y = 1,
for L, oe + p-y = 1,
for T, -p-y = -2,
so that oc = p = y = 1 and the required equation is
force = ka"v.
Physical, rather than dimensional, arguments are needed to show that
the proportionality constant k is 61t.
The same principle can be applied when using the practical SI units
of kilogram (kg), metre (m) and second (s), so that in any equation
connecting physical variables the units must cancel on the two sides of
the equation. This has two valuable and practical uses: unless the units
cancel, the equation being used is wrong; and when we wish to change
to other units, we can do so by a simple algebraic method called
quantity calculus. The latter is based on the principle that multipli-
cation by unity leaves any quantity unchanged, and we can construct
'unity brackets' by writing, as a fraction, the new units over the old
using appropriate conversion factors. As a simple example, if we write
(1 week/7 days) this has the value unity. To convert, say, 42 days into
weeks we multiply by the appropriate unity bracket and cancel the
units:
1 week
42 days = 42.eays- x 7..days" = 6 weeks.
Example 1.1
The gas constant R in the equation p V = nRT has the units needed for
the equation
R=pV
nT
to have the same units on both sides. Thus if the pressure p is in

atmospheres (atm), volume V in dm 3 (also called litres), and tempera-
ture T in kelvins (K), and amount of substance, n in moles (mol)
R = 0.08205 litre atm mol- 1 K - 1.
If we wish to use SI units instead, we make use of unity brackets based
on the conversion factors
1 atm = 101.325 kPa,
IJ =INm,
Idm 3 = 1O- 3 m 3 ,
IPa = INm- 2 ,
so that, using four significant figures,
R=
0.082057 ~..atnr 10- 3~ 101.325 X 10 3.Pt! X J
molK x ~ x ;mti. x ~# x N.m
Example 1.2
Show that Planck's constant has the dimensions of momentum
x length.
From the appendix,
h = 6.624 X 10- 34 J s.
The units of momentum x length are those of mass x velocity x length.
The unity brackets are based on the conversion factors
11= INm,
1 N = 1 kgms- 2 ,
so that the units of momentum x length are
kg x sm x m = )g'~ )4's'i. J
- , - x ~}If x ~}Ii = Js
Units and dimensional analysis 7
Example 1.3
Suppose that we are given that the spacing between the lines in the
microwave spectrum of HeI is 20.7 cm- 1 and that we need to calculate
the moment of inertia I, but can remember only that the spacing is 2B
where B is either h 2 /8n 2 I or h/8n 2 I c. This difficulty can be resolved by
adopting the strongly recommended rule that we substitute into
formulae not just numerical values but the units as well.
We have that either
h cm
or
I = 8n 2 c x 10.35'
where h == 6.626 x 10 - 34 J sand c = 2.998 X 108 m s - 1, so that either
1=
6.626 x 10- 34,J',8'.,.Qftl$ }(m kgm
x--x--x---
.m
8n 2 x 2.998 x 10 .m x 10.35
8 .J N~ 102..ern
= 2.70 X 10- 47 kgm 2 •

The second expression gives the correct units for the moment of inertia.
Example 1.4
Show that a diver at a depth of66 ft of sea water (density 1.03 gcm -3) is
under a pressure of about 3 atm.
The pressure exerted by a column ofliquid of density p and height h
is given by
p=pgh= 1 ; ; x9.81pf ~ x - - x12jn

- - x 66j(x-- -- J</
II 103$ ~W it
2.54..en1 104 ~ Pa~
x )rf x ~ x~
= 2.033 x lOs Pa = 203.3 kPa.

Thus the water exerts a pressure of about 2 x 101 kPa (2 atm), and so
the total pressure on the diver is about 3 atm.
1.3 Functional notation
When the value of a quantity z depends on the values of one or more
quantities x, y, ... , we call z, x, y, ... variables and show z to be a
function of x, y, ... by writing either
z = z(x, y, ... ), (1.7)
or
z =f(x, y, . .. ). (1.8)
The first form, equation (1.7), is used to show only that a dependence
exists, whereas the second labels a particular function as f(x, y, ... ),
and this would not be the same function as, say, g(x,y, ... ).
Variables are usually denoted by letters from the end of the Greek or
Roman alphabets, and constants are usually denoted by letters from the
beginning of the alphabets, but this is a useful guideline rather than a
universal rule.
We denote the value of a function at particular values of the variables
by f(a, b), which is the value off(x, y) when x = a and y = b. Thus the
enthalpy H of a substance, which depends on the temperature Tand the
pressure p, is shown as H (1', p), and at the particular values of the
variables T = 298 K and p = 1 atm we write H (298 K, 1 atm).
1.4 Quadratic and higher-order equations

An algebraic equation in a single variable, say x, is said to be of degree n
when the highest power of x in the equation is x" after negative or
fractional powers have been removed. Thus the equation
2
x-- = 1
x
must first be multiplied through by x to give
x 2 -x-2 = 0 (1.9)
and so it is a second-degree (or quadratic) equation. An equation that
contains x 3 is third degree (or cubic), and so on.
Such equations will be satisfied by particular values of x called the
roots of the equation, and an equation of degree n has n roots. An
equation is solved by finding the roots by any convenient method. In
general, if x = a is a root of the equation f(x) = 0, then (x - a) is a
factor, which means that the equation can be written as (x - a)g (x) = O.
The simplest solution will be when the factors can be seen by
Quadratic and higher-order equations 9
inspection. Thus equation (1.9) gives
(x -2)(x + 1) = O.
This will be satisfied if either (x - 2) = 0 or (x + 1) = 0, giving x = 2
and x = -1 as the roots of the equation.
The factors of a difference between two squares are particularly
useful, as
x 2 -4 = (x+ 2)(x -2).
The roots of a cubic equation such as
x 3 -2x 2 -x+2 = 0 (1.10)
may be found by trial and error. By assuming small whole-number
values of x we find that x = 1 satisfies the equation; this means that
(x - 1) is a factor. Long division of the expression by (x - 1) gives the
other, quadratic, factor, which in this case is the left-hand side of (1.9),
so that the roots of (1.10) are x = 1, x = 2 and x = -1.
A quadratic equation that cannot easily be factorized is solved by
using the formula for solution of a quadratic, which is obtained by
'completing the square'. Given the equation
ax 2 +bx+c = 0
we first divide through by a to obtain
b c
x 2 +-x+-
a a
= o.
The terms containing the variable x can be obtained by writing
so that the original equation can be written as
(x
b )2 b2 C b2 - 4ac
+ 2a = 4a2 - ~ = -4-a-O;2;--
We now take square roots of both sides, recognizing that this
introduces ambiguity of sign, so that
X = - -b ± 1.J
- (b2 -4ac). (1.11)
2a 2a
Corresponding formulae can be written for the solution to a cubic
equation, but these are too cumbersome to be useful; it is better to find a
numerical factor by trial and error and to use the formula (1.11) for the
remaining quadratic factor. Newton's method (Section 2.9.2) can be
used in the initial numerical approximation.
Real solutions of a quadratic equation will not exist when b2 < 4ac.
We can then use complex numbers to write solutions containing real
and imaginary parts as in Section 1.10.
Example 1.5
In physical applications we may expect physical conditions to restrict
the choice of roots. Thus the dissociation constant K of a solution of a
monobasic weak acid is related to its degree of dissociation tX at
concentration c by
so that
ctX 2 + KtX - K = 0,
tX= - -K ± -1 V/ (K 2 + 4Kc).
2c 2c
But K, c and tX are all positive, so that only the positive sign before the
root has physical meaning (see also Example 2.25).
1.5 Dependent and independent variables

When we say that a variable is independent we mean that it can assume
any value. A simple case is if we were concerned with, say, the length of a
given bar of metal as a function of temperature. We can fix the
temperature at any value we please, and this is an independent variable.
Once having fixed the temperature, the length is fixed by the physical
properties of the metal, so that length is not independent once the
temperature is fixed. This is an example of one dependent and one
independent variable.
If we were considering the volume of a sample of gas, this depends
not only on temperature but also on pressure. We then have three
variables altogether (volume, temperature and pressure) of which two
are independent (e.g. temperature and pressure) and one dependent
(volume). Notice that we could equally well say that the pressure
depends on the volume and the temperature, which would imply that
volume and temperature are independent variables and pressure the
dependent one. Thus we cannot specify which particular variables are
necessarily dependent or independent; what we can specify clearly is
Dependent and independent variables 11
that two are independent and the third (whichever one of the three we
choose) is dependent.
To determine how many variables in a given problem are in-
dependent we need to know how many relevant variables there are
(three for our sample of gas) and how many equations exist connecting
these variables (one, p V m = RT for a perfect gas). The number of
independent variables is then the total number minus the number of
equations (3 -1 = 2 for our sample of gas). The mathematical
significance of this is that each equation (simultaneous equation if more
than one) can be used to eliminate one variable from the problem; if we
have as many simultaneous equations as variables we can solve the
equations to obtain numerical values of the variables, meaning that the
number of independent variables is then zero. This need to have as
many simultaneous equations as variables in order to obtain a unique
solution is usefully borne in mind, particularly with problems involving
chemical equilibria.
A chemical example of the limitations on the number of independent
variables set by the existence of relating equations is the phase rule.
Each phase of C components requires C -1 composition variables (the
1 is a relating equation that the sum of the mole fractions is unity), so
that P phases require P(C -1) composition variables. The temperature
and pressure are two more variables, giving P(C -1)+2. The equa-
tions relating the variables are equality of chemical potential for each
component in each pair of phases, giving C(P -1) equations. The
number of independent variables is then
P(C-1)+2-C(P-1) = C+2-P.
Alternatively, instead of composition as a variable, we may choose to
use the C values of the chemical potential of each component, giving,
with temperature and pressure, C + 2 variables. In that case we have a
Gibbs-Duhem equation relating the chemical potentials in each phase,
giving P equations and so, again, C + 2 - P independent variables.
The number of independent variables is also called the number of
degrees of freedom of the system, and is the number of variables that
must be specified to define the state of a system completely. It is a
characteristic of chemical problems that we have more than this
minimum number of possible variables, resulting in a freedom of choice
which can cause confusion. Thus in the simplest case of a single phase of
a pure substance, the two independent variables may be chosen from p,
Vm and T, we might choose to use either (p,l) or (Vm,T) as the
independent variables and obtain correct results in either case. It often
happens that the relevant formulae are simpler in terms of one choice of
independent variables than another; for example if the volume is
subsequently to be held constant this should be chosen as one of the
variables in setting up the necessary equations because the simplest
form of the result will then be obtained. Experimental measurements
on chemical systems are often conducted at a constant pressure of one
atmosphere, so the best choice of independent variables is then (p,T).
Theoretical treatments, on the other hand, are often simplest in terms
of a fixed volume for the system and the independent variables used are
(Vm,T). Thermodynamic properties are expressed in terms of the
enthalpy H and Gibbs function G when (p, T) are chosen as the
variables; the total energy U and Helmholtz free energy A give the
simplest formulae in terms of (Vm' T).
1.6 Graphical methods

Graphical methods of presenting and analysing data are used because
they reveal experimental scatter of points, because curvature can be
detected or the choice of a best straight line can be made, and because
they facilitate interpolation (reading intermediate values between the
known points) and extrapolation (the projection of curves or lines
beyond the range of the points). In precise work, graphical methods
should be regarded only as a prelude to the analytical treatment
described in Chapter 7.
The following terms and conventions are used in graph plotting. The
vertical or y-axis is the ordinate and the horizontal or x-axis is the
abscissa. The scales should be chosen so that the given points fill the
graph, apart from short extrapolations, which implies that any straight
line will have an angle of slope near to 45°. A long extrapolation is
sometimes required, such as to absolute zero of temperature, but this
should be calculated, not drawn; similar triangles (Section 1.7.1) are
then particularly useful.
The scales on a graph are pure numbers so that only pure number
quantities should be plotted. The significance of this is that if the graph
were to show, say, length in metres as .a function of absolute
temperature, the labels on the axes of the graph are length/m and TfK
because length divided by metre and temperature divided by kelvin are
both pure numbers.
Graphs are always plotted with the dependent variable as ordinate
and the independent variable as abscissa. As pointed out in the previous
section, there is no mathematical distinction between two variables as
Graphical methods 13
to which is the dependent and which the independent variable, so that
physical criteria are required. The proper choice is often obvious in that
the independent variable is deliberately fixed and the dependent
variable is then the result of an experiment. Time, temperature and
composition are common independent variables and so should be on
the x-axis. When the proper choice is not obvious, precedent should be
followed; all graphs of the same quantities will then have the same
appearance.
The best graph to plot is a straight line because visual estimation is
then most precise and because simple analytical treatment is then
available. The equation of a straight line is conventionally written as
y = mx+c. (1.12)
When x = 0, y = c, which is the intercept on the y-axis, and when y = 0,
x = - clm, the intercept on the x-axis. The slope, or gradient, of the
graph is m, this being the scaled tangent of the angle between the line
and the x-axis and the value of the derivative dyldx (Chapter 2).
The best straight line through a set of points can be fairly precisely
estimated by eye using a tight cotton thread so as not to obscure any
points. The best line passes through the 'centre of gravity' of the points
and the sum of the squares of the deviations of the points from the line
should be a minimum, so that large deviations are much more
important than small ones. Such a visual choice of line will be
subjective, so that when the quality of the data merits it the analytical
method of least squares should be used as described in Section 7.8.
When a graph is used to test the agreement between experimental
data and a theoretical form of equation, or when numerical values of
constant parameters are required, a straight-line graph is much better
than a curve. Many equations that are not originally in the linear form
of equation (1.12) can be manipulated into that form before the scales
of the graph are chosen. Thus the first-order decay of quantity x with
time t is given by
This is converted into linear form by taking natural logarithms of both

sides
In x = - kt + In Xo
and a straight line should be obtained by plotting In x against t (the
independent variable, so abscissa or x-axis), with slope - k and
intercept In Xo on the ordinate (y-axis). When a logarithmic scale is
used the numerical values of any physical quantity must be divided by
their units as described in Section 1.2; changing these units will change
the intercept of a graph,but will not affect the slope.
1.7 Some geometrical methods
1.7.1 Similar triangles

If two triangles have corresponding angles equal, then they are said to
be similar and the lengths of corresponding sides are all in the same
ratio to each other. This is a particularly useful geometrical theorem.
In Fig. 1.1 a straight line passing through points A and B fits
experimental data plotted on paper shown in broken outline. The
intercepts e and b would be most easily found by fitting the two points
to y = mx +e by m = BN/AN and then e = YA -mx A • Then by similar
triangles e/( -b) = BN/AN = m, hence b = -elm as before.
I
I
I
I
I
~'-----;N :
I I
IL.. _ _ _ _ _ _ _ ---II
Fig. 1.1
1.7.2 Triangular graph paper

In an equilateral triangle, the sum of the perpendicular distances from
any point onto the sides is equal to the height of the triangle. This is
Some geometrical methods 15
used to represent the composition of three component mixtures on
two-dimensional paper.
In Fig. 1.2 point 0 is anywhere within the equilateral triangle ABC of
side a and height h. The sum of the perpendiculars OP + OQ + OR can
be seen to be equal to h by noting that the total area of the triangle is the
sum of the areas of AOB, BOC and COA, each area being (a/2) times
the corresponding height.
If a scale is chosen so that the height of the triangle is either unity (in
mole fraction) or 100 (in percentage composition), the amount of each
of three components can be plotted from zero on a side to 1 (or 1(0) at
the opposite apex. Triangular graph paper is available on which lines
parallel to the sides are drawn so that such scales can be used. In Fig.
1.2, the point 0 is where the percentage of A is QO, that ofB is RO and
ofC is PO; the scale is such that h is 100. The line BX in the figure is the
locus of points that would be obtained by adding pure B to a mixture of
composition X of A and C.
A
B C
~------------------~
a
Fig. 1.2
1.7.3 Three-dimensional geometry
Many problems in science, and particularly in chemistry, involve more
than two variables. These correspond to diagrams in three or more
dimensions. They are often treated by taking two-dimensonal sections
in planes perpendicular to each other. When lines or planes are
perpendicular to each other they are called orthogonal.
In Fig. 1.3 the point P is distance r from origin O. The coordinates of
Pare (a,b,c), these being the projections of the line OP on to the x-, y-
and z-axes. This is conveniently shown by drawing perpendicular, or
normal, lines onto the z-axis and into the (x, y) plane. The projection
ON in the (x, y) plane then gives projections OA = a and OB = b on
the x- and y-axes.
z
c
A-~~~~------~---y
B
A~ ____________ ~
x
Fig. 1.3
Pythagoras's equation in three dimensions then follows because

r2 = ON 2 +PN 2 = a2 +b 2 +c 2.
An alternative approach to direction in space is to use the angles IX, p
and y that the line OP from the origin makes with the X-, y- and z-axes,
the angles being taken in the planes containing the line and the axis as
also shown in Fig. 1.3. The projections of the line oflength r onto the X-,
y- and z-axes are then a = r cos IX, b = r cos p and c = r cos y; the factors
cos IX, cos p and cos yare called the direction cosines of OP and are
denoted by I, m and n. Then by Pythagoras's equation, [2 + m2 + n 2 = 1.
An equation containing two variables represents a line or curve in
two dimensions, and if it contains only first powers of the variables and
no cross-terms it can be written in the form of equation (1.12) and
represents a straight line. Similarly an equation in three variables
represents a surface in three-dimensional space, and if it contains only
the first powers of the variables and no cross-terms it represents a plane.
The general equation of a plane is
~+~+~ = 1, (1.13)
abc
as may be most easily checked by noting that the partial derivatives are
then constants (Chapter 3). Equation (1.13) is that of a plane through
the points A, Band C in Fig. 1.3, as can be seen by putting y = z = to
show that the intercept on the x-axis is x = a, and so on.
°
The direction of a plane in space is often required in crystallography,
when it is denoted by the Miller indices (h, k, I). These are the smallest
whole-number ratios of the reciprocals of the intercepts on the axes.
Thus the plane through A, Band C in Fig. 1.3 is denoted by
(l/a, l/b, l/c), multiplied by a common factor if necessary so as to give
the smallest whole numbers. The use of reciprocals here avoids infinite
values because a plane parallel to an axis is defined to have a
corresponding Miller index of zero.
1.7.4 Circle, ellipse, parabola and hyperbola

An equation in two variables, say x and y, which contains terms of the
second degree (x 2, xy or y2) but not higher, represents a circle, ellipse,
parabola or hyperbola. The particular type of curve is readily identified
from the relative values of the numerical coefficients.
Writing the general form of second-degree equation as
ax 2 + 2hxy+by 2 +2gx+2fy+c = 0, (1.14)
the conditions for the various kinds of curve are
(a) Circle a=b and h = 0,
(b) Ellipse ab > h 2,
(c) Parabola ab = h 2,
(d) Hyperbola ab < h 2.
At least one of the second-degree terms x 2, xy or y2 must be present, but
any other coefficients may be zero.
Example 1.6
A sample of perfect gas at fixed temperature obeys the relation p V = a
positive constant. Substituting p for y and V for x we have a = b = 0
and h is positive, so that h 2 > ab and the curve obtained is a hyperbola
(curve E in Fig. 1.4).
Example 1.7
The potential energy V for simple harmonic motion (Section 6.3) is
given by V = tkx 2. Substituting V for y gives a = kj2 and b = h( = 0),
so that ab = h2 (= 0) and the curve is a parabola (curve F in Fig. 1.4).
There are standard forms of equation for each kind of curve. These
are
(a) Circle x 2 + y2 = r2, (1.15)
(b) Ellipse x 2ja. 2 + y2jp2 = 1, (1.16)
(c) Parabola y2 = 4a.x, (1.17)
(d) Hyperbola x 2ja. 2 _ y2jp2 = 1. (1.18)
Example 1.8
Show that these standard forms satisfy the general conditions stated
above.
Having identified the type of curve represented by any particular

equation, its position and orientation relative to the coordinate axes
may be found by looking for symmetry and for points of intersection.
From equation (1.15) when x = O,y = ± rand wheny = O,x = ±r,so
that the equation represents a circle of radius r centred at the origin
(Curve A in Fig. 1.4).
An ellipse has major and minor axes with two foci on the long axis.
Equation (1.16) gives x = ± a. when y = 0 and y = ± Pwhen x = 0, so
that the axes of the ellipse are the x- and y-axes. When a. > p the x-axis
is the longer one and so contains the two foci (curve B in Fig. 1.4).
The parabola represented by equation (1.17) passes through the
origin (y = 0 at x = 0) and is symmetrical about the x-axis (y =
± 2J(a.x». Thus y2 = 4a.x gives curve C in Fig. 1.4 whereas x 2 = 4a.y
would give curve F.
A hyperbola lies between two intersecting straight lines called
asymptotes. Curve E in Fig. 1.4 (xy = constant) has the x- and y-axes as
asymptotes, these being identified by seeing that as x tends to zero y
tends to infinity, and as y tends to zero x tends to infinity. When, as here,
----+-~------r-~~~_+------+_~---x
Fig. 1.4
the asymptotes are perpendicular to each other, the hyperbola is said to

be rectangular. The general form, equation (1.18), cuts the x-axis at ± a:
and does not cut the y-axis. It is symmetrical about the x-axis and has
asymptotes given by yiP = x/a: and yiP = -x/a: (curve D in Fig. 1.4);
these are perpendicular to each other if a: = Pand the hyperbola is then
rectangular (see also Example 2.30).
Two operations may be necessary to convert any given equation into
standard form: displacement of the origin and rotation of the axes. If
the cross-product xy is present, axis rotation is needed to eliminate that
term. The effect of this may be calculated analytically but can be seen by
inspection in the simple cases usually met in physical applications.
Example 1.9
The equation
x 2 + y2 -2x+4y = 4
represents a circle because h = 0 and a = b. The presence of terms in x
and in y means that the centre of the circle is not at the origin. The
equation is reduced to standard form by moving the origin to eliminate
these linear terms; the term in x can be obtained by using
(x - 1)2 = X 2 - 2x + 1
and that in y from
(y+2)2 = y2+4y+4.
We can therefore write the equation as
(x-l)2+(y+2)2 =9
so that it is a circle with centre at (1, - 2) and radius 3.
The standard forms of equation are useful when experimental data

are to be fitted to such curves. As remarked in Section 1.6, a linear graph
is preferred when choosing the best fit to scattered points.
Example 1.10
Identify the curve
y = ocx(l-x)
and suggest two alternative linear forms.
This equation has the form
ocx 2 - ocx + y = 0
so that ab = h 2 and the curve is a parabola. Differentiation (Section 2.8)
shows the curve to have a maximum at x = 1/2, y = oc/4, and also y = 0
at both x = 0 and x = 1. The equation is reduced to standard form by
writing
x 2 - X + y/oc = 0,
and using
(x - 1/2)2 = x 2 - X + 1/4
we obtain
(x-l/2f = 1/4- y/oc
= -(y-oc/4)/oc.
This becomes standard form by interchanging the variables and
moving the origin to (1/2, oc/4).
A linear plot may be obtained either by plotting y against x(1 - x),
which is obvious by inspection, or from yagainst (x -1/2)2.
Such a relation is often a good approximation to the properties of
mixtures, such as the heat and volume changes on mixing liquids or on
mixing gases. Departure from symmetry about x = 1/2 can be allowed
for by using a series of terms in powers of (x - 1/2). A similar approach
is used to allow for anharmonicity of molecular vibrations in
spectroscopy.
Example 1.11
Show that, neglecting air resistance and the curvature of the earth, a
projectile moves in a parabola under the action of gravity.
Given velocity of projection u at an angle (X to the horizontal, the
vertical component of the initial velocity is u sin (X. The vertical height y
at time t is given by
y = ut sin (X _!gt 2 •
Simultaneously, the horizontal velocity is u cos (X so that the distance

travelled in the horizontal direction is x = ut cos (X. Eliminating t
between these two equations gives
x
t=---,
U cos (X
gx 2
y=xtan(X-22 2'
U cos (X
so that a = -g/(2u 2 cos 2 (X) and b = h = o. We therefore have

ab = h2 (= 0) and the curve of y against x is a parabola.
1.7.5 Plane polar coordinates

Coordinates such as x and y, which are plotted as lengths on graphical
scales, are called cartesian coordinates, in contrast to rand 0, a length
and an angle, which are called polar coordinates. When the x- and y-
axes are perpendicular to each other, they are called rectangular
cartesian axes.
Point P in Fig. 1.5 may be defined either by (x, y) or by (r,O), the
relations between the coordinates being
x = rcosO,
y = r sin O. (1.19)
Problems that possess circular symmetry will be solved most
conveniently by using polar coordinates. The radius r is defined always
to be positive so that x and y assume their proper signs in the various
quadrants due to the signs of cos 0 and sin O. The angle 0 is generated by
rotating the radius vector r from the x-axis in an anticlockwise
direction; positive rotation is therefore conventionally anticlockwise.
y
Fig. 1.5
1.8 Factorials and gamma functions

The factorial of a positive integer n is here denoted by n! and is the
product
n! = n(n-1)(n-2)(n-3) ... 1. (1.20)
Thus
6! _ 6 x 5 x 4 x 3 x 2 x 1 _ 2
3! - 3x2x 1 - 1 O.
It is conventional to define O! = 1 so that the general relation

(n + 1)! = (n + l)(n!)
is still true when n = O.
Since a graph could be plotted of n! against n and a smooth curve
drawn through the points for integral values of n, we wish also to be
able to find the factorial of non-integral numbers. We anticipate
subsequent chapters on integration to mention that this is achieved by
use of the gamma function r(x) defined in Section 6.4 as
r(x)=S;'tX-le-tdt (x >0), (1.21)
Probability 23
from which
x! = r(x + 1) = xr(x). (1.22)
Tables of this function are available, values being necessary only in the
range 1 < x < 2 since when x > 2 we use equation (1.22), thus
2.72l! = r(3.721)
= 2.721 r(2.721)
= (2.721) (1.721) r(l.721)
= 4.274 from the tables.
When n is very large, being typically a number of molecules in a
sample of substance, an approximation that is often used is Stirling's
theorem for In n! in the form
In n! ~ n In n - n, for large n. (1.23)
1.9 Probability
We express the probability of an event as a fraction between 0 and 1,
where zero means that the event will not happen and unity means that
the event is certain. A fundamental principle is that the probability of
an event is proportional to the number of ways in which that event can
occur. Thus in throwing two dice simultaneously we may obtain a total
of5 in four ways; by scoring any of (4, 1), (1,4), (2, 3)or (3, 2). Each of the
dice may fall in anyone of six ways, and for any particular score on one
we have six possibilities for the other; the total number of possible
results is therefore found by multiplying 6 by 6 to give 36. The
probability of scoring 5 is therefore 4/36 = 1/9.
This argument applies to independent events, which are common in
scientific applications. An important quantity is the number of ways of
arranging N objects in line. For this we may choose the first as anyone
of the objects, and so in N ways assuming them to be distinguishable.
The second is chosen from the remaining N - 1, and so in N - 1 ways.
Since each choice of the first may be followed by anyone of the choices
of the second we multiply N(N -1) to find the number of ways of
choosing the first two in the line. By repeating the argument we see that
the required number is NL
This is then extended to find the number of ways of placing N objects
into boxes that contain nl, n2, ... in successive boxes. We begin by
placing the objects in line in N! ways, and then divide the line into
groups of n 1 , n2, . .. . The objects in a group of nl may be rearranged
in n 1 ! ways and these have so far been included in the total of N!. Since
these rearrangements are not to be distinguished, we must remove them
from the product by dividing by nl!, hence the required number is
N!/(nl ! n2! ... ). If the objects are placed at random into the boxes, the
probability of obtaining particular numbers in the boxes is then
proportional to this factor; this is used in Example 2.17.
Example 1.12
The equilibrium constant for the reaction
H2 +D2 = 2HD
may be calculated from simple probability theory. We suppose that a
catalyst surface, on which dissociation into atoms occurs, is exposed to
a gas mixture containing equal numbers ofR and D atoms. We assume
that molecules are formed by random pairing of atoms on desorption
from the surface.We then have equal probability of any particular atom
being H or D. We may form H2 (or D 2) only by both atoms being the
same, but HD can be formed in two ways; either one site on the catalyst
may be occupied by H and the other by D, or the other way round. We
therefore form twice as many HD molecules as H2 or D 2 , giving for the
equilibrium constant
which is a good approximation to the actual value of K at a high

enough temperature.
1.10 Complex numbers

When equations are solved by simple algebraic methods we meet cases
where no solution exists, that is to say no real solution. Thus the
equation
x 2 +4 = 0, x=±j(-4)
has no real roots. We can define i as the imaginary square root of -1

and write the solution as x = ± 2i, which is the product of the real
number ± 2 and the imaginary number i. This can be extended to
include numbers having both real and imaginary parts, thus
x 2 -4x+8 = 0, x = 2±tj( -16) = 2±2j(-1),
which is written as the complex numbers 2 ± 2i.
Complex numbers 25
The definition and use of complex numbers in this way gives an
advantage in the sense that an equation of degree n then always has n
roots, some of which may be imaginary. These imaginary roots do not,
however, have any physical meaning in the real world; complex-number
theory cannot provide the physical scientist with any new concepts, but
it does permit neater solutions to some problems. If, say, we are
concerned with finding a boundary beyond which real solutions no
longer exist, complex numbers can enable us to approach that
boundary from the imaginary side as well as from the real one, and this
has applications in stability problems. Complex numbers are valuable,
moreover, as a general analytical tool.
In the general form of complex number a + ib both a and b are real
numbers; a is called the real part and ib the imaginary part. This should
be regarded as an ordered pair of real numbers which have no relation
between them. We must consider real and imaginary parts separately.
Thus
(a +ib) - (c +id) = (a -c)+i(b -d). (1.24)
When we have an equation relating complex numbers this really
constitutes two simultaneous equations, because we can then equate
the real parts and, separately, equate the imaginary parts. Thus
if a+ib = c+id then a = band c = d.
The imaginary number i follows the normal rules of algebra with the
additional rule that j2 = - 1. Thus
(1.25)
This also illustrates another important point, that we can obtain real
numbers from combinations of imaginary ones, again only real
numbers having physical meaning. When the sign is reversed in the
complex number z = (a + ib) to give z* = (a - ib) we call z* the
complex conjugate of z, and zz* is the real number given by (1.25). This
is used in wave mechanics, where the wavefunction I/J may be complex
but I/JI/J* is a real probability.
It is sometimes useful to give a geometrical interpretation to complex
numbers. This is done by plotting a two-dimensional graph showing
real parts as abscissa and imaginary parts as ordinate. This is called the
Argand diagram and is shown in Fig. 1.6 where point A represents the
complex number a + ib. The complex conjugate A * is then the
reflection of the point A in the real axis. The complex number A can be
regarded as the directed line, or vector, OA, and the addition of complex
numbers A and B can be shown as if we add vectors 0 A and AC to give
Imaginary
axis
C(o+c+i(b+d))
,,
,Q Real axis
I
,
'A*(o-ib)
Fig. 1.6
~C. The length OA is called the modulus IA Iofthe complex number A,

given by
(1.26)
and this does not contain i and so is a real number.
When we choose to use polar coordinates we use the angle 8, called
the argument of the complex number, and denote the modulus by r. We
then have an alternative notation for complex numbers since
a = rcos8 and b = rsin8, so that
a + ib = r(cos 8 + i sin 8). (1.27)
The analogy between complex numbers and vectors (Section 3.9) must
not be pursued too far because other allowed operations with vectors
do not apply to complex numbers. Thus multiplication of complex
numbers gives
(a+ib)(c+id) = (ac-bd)+i(bc+ad) (1.28)
or, in polar notation,
r1 (cos 81 + isin ( 1) x r2 (cos 8 2 + isin ( 2)
= r1r2[cos(8 1 +( 2)+isin(8 1 +( 2)]. (1.29)
Complex numbers 27
This means that on the Argand diagram we multiply by adding
arguments (0 1 + O2) and multiplying the moduli (rl r2), so that
multiplication and division involve rotating the vectors and changing
their length.
There is a particularly useful relation between the exponential of a
complex number and the trigonometric functions. By definition
x2 x3
eX = 1 + x + 2! + 3! + ... ,
and replacing the exponent by the complex number x + iy gives

e x + iy = eXeiy
and
. . y2 iy 3
elY = 1 + ly - 2! - 3! + ...
This series has alternating real and imaginary terms; the real terms are
those in the series for cos y and the real coefficients of the imaginary
terms are those for sin y. Hence
eiy = cosy + isiny (1.30)
and similarly
e- iy = cosy-isiny. (1.31)
This gives us a third, exponential, form of notation for complex
numbers. By comparing (1.30) and (1.31) with (1.27) we obtain
a + ib = r(cos 0 + isin 0) = reiO,
r = .j(a2 + b 2 ), 0 = tan -1 (b/a). (1.32)
This notation is valuable because exponential functions are particularly
easy to manipulate. Thus the multiplication expressions (1.28) and
(1.29) become
(a +ib)(c +id) = r l eiO , x r2eiO, = r1r2ei(O,+O,).
One of the principal advantages of complex numbers is that they
enable trigonometric functions to be rewritten in terms of exponentials;
trigonometric identities can then be readily established.
Example 1.13
cos 2 y+sin 2 y = (cosy+isiny)(cosy-isiny)
= eiYe- iy = eO = 1.
Addition or subtraction of (1.31) from (1.30) gives the useful
substitutions
eix+e- ix . eix -e- ix
cos x = ----:2:--- and SIn x = ---:-2-:-i- (1.33)
Example 1.14
Use equations (1.33) to show that cos 2 x - sin 2 x = cos 2x.
The application to complex numbers of the calculus techniques

described in the following chapters is straightforward, remembering to
treat real and imaginary parts separately.
CHAPTER 2
Differential calculus
2.1 Significance and notation

Calculus refers repeatedly to behaviour in the limit of infinitesimal
changes; a clear understanding of this concept, discussed in the next
section, is essential. Calculus is valuable in science in general, and in
chemistry in particular, for many reasons. Amongst these are that it
enables us to study the effects of infinitesimal changes in conditions,
effects which are often quite simple even though the effect oflarge finite
changes may be quite complicated. In this way, problems can be
separated into parts by examining separately the behaviour at a point
from the behaviour over a wider range. Calculus also provides a clear
and precise notation by which we can resolve problems that involve
changes in many variables into ones involving changes in only one
variable at a time. The techniques of differentiation and integration
give connections between the values of properties and their rates of
change, which are invaluable in the interpretation of experimental
measurements. An appreciation of these uses of calculus requires
not just the usual skills at the operations of differentiation and inte-
gration but an insight into the reasons why particular techniques are
effective.
There is need to distinguish between the notation:
L\x, which means a large, finite or macroscopic change in x,
dx, which is the differential of x,
!5x, which denotes an infinitesimal change in x, and
ax which is the partial derivative sign.
30 Differential calculus
We use the following functional notation:
y =f(x) means that y is a function of x, or that given x,
y is uniquely determined by an equation that
contains both x and y but no other variables,
only constants; another function of the same
variables would be written as, for example,
g(x).
y = f(u, v, w) means that y is a function of three variables u, v
and w, any other terms being constants; it does
not imply that u, v and ware independent.
dy
and y'(x) both denote the total derivative of y with
dx
respect to x.
usually denotes differentiation of y with re-
spect to time, or dy/dt.
f'(x) is the total derivative of the functionf(x).
2.2 The calculus limit

The idea of a limit is fundamental to calculus and will be expressed both
geometrically and analytically. In Fig. 2.1 a curve y = f(x) is drawn
with two points A and B on the curve separated by differences Ax in x
and Ay in y. The line AB is a chord having a slope defined as Ay / Ax.
When point B is moved down the curve to approach A, the slope of the
chord will change until, when B coincides with A, we obtain the slope of
the tangent to the curve at point A. The difference in x between points A
and B is denoted by Ax and as B is moved to approach A this difference
becomes infinitesimally small, being then denoted by (jx; the slope of
the chord becomes that of the tangent in the limit as (jx --+ o.
We are quite deliberately considering what happens as (jx --+ 0, not at
(jx = 0; a way of interpreting that is to point out that when A and Bare
coincident we cannot obtain the tangent to the curve because the slope
ofa line through a single point can have any value. To find the slope of
the tangent we need to take account of the shape of the curve on either
side of the point. The limiting value of (jy/(jx as (jx --+ 0 is called the
derivative of y with respect to x, and is written as dy/dx, so that
dy = lim (jy. (2.1)

dx 6x-O (jx
The calculus limit 31
y
y = '(x)
I
I
I
6x - __ ..JI
r------
I
I I
Fig. 2.1
The alternative analytical approach is to define by as the change in y

for change bx in x. Since y = f(x) we can write
(y+by) =f(x+bx)
and
by = (y+ by) - y = f(x + bx) -f(x),
so that
dy = limf(x+bx)-f(x). (2.2)
dx /)x-+O bx
An alternative notation is to denote a difference in x by a new symbol
such as h, and then allow h to tend to zero to give
dy = lim f(x + h) - f(x).
dx h-+O h
The definition of the derivative given by (2.2) is used in the next
section to obtain the rules for differentiation.
There are conditions that must be satisfied for calculus methods to be
applicable to particular functions. In analytical terms the limit (2.2)
must exist, the function then being said to be differentiable. A less
severe condition concerns continuity; a function is said to be continu-
ous if
lim f(x ± bx) = f(x). (2.3)
6:%-0
Condition (2.3) means that we can approach any point on the curve
from either side and obtain the same value. An example of a dis-
continuous function is y = l/x as x -+ 0; when we approach zero in x
from the positive side, l/x -+ + 00 whereas from the negative side
l/x -+ - 00, as shown in Fig. 2.2. Condition (2.3) is not then satisfied
and we say that l/x shows an infinite discontinuity at x = o. A
differentiable function is one for which the slope of the line is
continuous; for example a line drawn in the shape of a square is
continuous, but a function representing such a shape would not be
differentiable at the corners. This example allows us to introduce
another concept, that of a single-valued function; a circle or square
could give two possible values of y for a given value of x and vice versa.
y
y=t
Fig. 2.2
Fig. 2.3 is drawn to illustrate other terms that are used to describe the
behaviour offunctions. The straight line y = f(x) = x has the property
The calculus limit 33
Fig. 2.3
thatf( -x) = -f(x), meaning that changing the sign of x changes the
sign of y, and this is called an odd function of x. On the other hand, y
= x 2 isa parabola (Section 1.7.4) for whichf( -x) = f(x);changing the
sign of x does not change the sign of y and this is called an even function
ofx.
Most of the simpler functions used in physical applications are
continuous, differentiable and single-valued, and the methods of
calculus can then be used. A function such as y = l/x is 'well behaved'
except as x -+ 0, so that such functions can be used within certain
ranges, which may have to be specified.
Having performed the operation of differentiation once we can do so
again on the result of the first differentiation, assuming that also to be
differentiable. This gives the second derivative of y with respect to x,
which is written as d 2 y/dx 2 , and we can go on to higher derivatives
d 3 y/dx 3 , .••• A straight line will have the same slope at any point on
the line so that the first derivative is a constant; this means that the
second derivative, which measures the rate of change of slope with
change in x, is zero. The second derivative is thus seen to measure
curvature. When the second derivative d 2 y/dx 2 is positive, the slope is
increasing with increase in x and the curve lies above its tangent as in
Fig. 2.1; such a curve is called concave upwards. Conversely, a curve
with decreasing slope is called concave downwards.
The notation for higher derivatives can be explained by regard-
ing dy/dx as the product of the differential operator d/dx and the
variable y. Successive differentiation is then denoted by
(d/dx)2 times y = d 2y/dx 2.
2.3 Differentiation of simple functions

The definition of the derivative given in the previous section can be used
to obtain the rules for differentiation, examples of which will now be
given.
Example 2.1
Put y = x 2 , then
(y + <5y) = (x + <5X)2
= x 2 + 2x<5x + (<5X)2.
Hence
dy r (y + <5y) - Y
dx = ,,~~o <5x
= lim 2x<5x + (<5X)2
",,-+0 <5x
= lim (2x + <5x) = 2x
",,-+0
so that
dy d
- = _(x 2) = 2x. (2.4)
dx dx
Example 2.2
Put y = x", then
(y + <5y) = (x + <5x)"
= (x + <5xHx + <5x)(x + <5x) ..•
= x" + nx"-l<5x + terms in (<5X)2, etc.
Hence
dy r (y + <5y) - Y
dx = ,,~~o <5x
= lim (nx"-l + terms in <5x, etc.).
",,-+0
When we take the limit all terms except the first contain <5x and so tend
Differentiation of simple functions 35
to zero, giving
dy d
_ = _(x") = nx"-l. (2.5)
dx dx
The formulae for differentiating a product or quotient may be
derived from these first principles. If y = uv where u and v are both
functions of x
y + l5y = (u + l5u)(v + I5v)
where l5u and I5v are the increments in u and v produced by an increment
I5x in x. Hence
dy = lim (U + l5u)(v + I5v) - uv)
dx b-+O I5x
. (I5U
= lIm I5v I5Ul5v)
v-+u-+--.
6,x-+0 I5x I5x I5x
Now the limit of a sum is the sum of the limits, and that of a product is
the product of the limits. The first two terms contain the definitions of
du/dx and dv/dx and the third gives
( I·1m
6,x-+0
l.:)(
uU I·1m -I5v)
I5x
6,x-+0
the first term of which is zero. Hence

dy d du dv
dx = dx (uv) = v dx + udx· (2.6)
The quotient y = u/v gives
dy = lim [(u + l5u - ~)/ I5x],

dx 6,x-+0 v + I5v v
which may be written as
dy = lim 1 (v(u+I5U)-U _u(v+I5V)-v)

dx 6,x-+OV(V+l5v) I5x I5x
so that
du dv
v--u-
dy = ~(~) = dx dx (2.7)
dx dx V v2
Example 2.3
Show that if
x
then
dy 1
- = 1 - -2
dx x '
2.4 The use of ditTerentials; implicit ditTerentiation

The derivative dy /dx can be interpreted not only as the product of the
differential operator d/dx arld the variable y, but also as the ratio of the
differentials dy and dx. For this interpretation to be justified we must
be able to apply the usual rules of algebra to differentials, and this we
can in fact do. Introducing the notationf'(x) for the result of applying
the rules of differentiation derived in Section 2.3 to the function f(x) we
can write
dy = f'(x)
dx
and can then multiply both sides by dx to obtain

dy =f'(x)dx
and we call this the total differential of y. Notice that this is not called
the differential with respect to any particular variable; the variable x
appears only when we go on to write down the right-hand side of the
equation, and indeed it can be changed if we so wish without affecting
the left-hand side.
The rules for differentiation may then be written in the form
if y = x" then dy = nx"-l dx, (2.8)
if y = uv then = udv + vdu,

dy (2.9)
if y = u/v then dy = (vdu - udv)/v 2 • (2.10)
This notation is especially important when we cannot conveniently
write y as an explicit function of x. Thus
y2 -2xy =4
implies that y is a function of x, but an inconveniently complicated one.
The derivative may be obtained most easily by differentiating each term
The use of differentials; implict differentiation 37
to give
2ydy -2xdy -2ydx = 0
(y -x)dy = ydx
dy y
dx y-x
When this method of obtaining derivatives is used the result will
often contain both x and y, whereas when we can begin by obtaining y
as an expliCit function of x the result will be a function of x only. The
original equation can be used to show that the results are the same.
Example 2.4
Show that implicit differentiation of xy - x 2 = 1 gives the same result
as Example 2.3.
Differentiating the expression gives
xdy+ ydx -2xdx = 0
xdy = (2x - y)dx
dy 2x - Y 2X2 -xy x 2 + (x 2 -xy)
dx X x2 x2
But x 2 -xy = -1 so that, as before,
dy = 1-~
dx x2 •
Implicit differentiation may also be applied successively so as to

obtain higher-order derivatives, given that differentiation of dx gives
d 2x, of xdx as a product gives xd 2x + (dX)2 and, since dx/dx = 1,
d 2x/dx 2 is the derivative of 1 and so is zero. From Example 2.4 we have
dy/dx = 1 _1/X2 so that d 2y/dx 2 = 2/x 3 • Ifwe choose to use this as an
example of successive implicit differentiation we also have
xdy+ ydx -2xdx = 0
and differentiating again gives
xd 2y+dxdy+ dydx+ yd 2x -2xd 2x -2(dx)2 = 0,
and division by (dX)2 gives
d2y dy d 2x
X dx2 + 2 dx + (y - 2x) dx2 - 2 = o.
But d 2x/dx 2 = 0 so that
d2y 2dy 2
dx 2 = - xdx +~.
But dy/dx = 1 -1/x 2 so that, as before,
d2y 2
dx 2 = x 3 '
1.5 Logarithms and exponentials

The exponential function eX, or exp(x), is defined by the series
for all x. (2.11)
Differentiation of this series can be seen to give the same series as a

result, so that when we differentiate eX with respect to x we get eX. Thus
dy
if y = eX then dy = eXdx, dx = eX. (2.12)
The natural logarithm of y is defined by the relation

then x = Iny, (2.13)
and since eX is positive (or zero) for aU values of x, In y is defined only for
positive values of y. The notation In for natural logarithm is used, with
loge as an alternative, to distinguish it from the Napierian or decadic
logarithm log (or 10glo) which is defined by
if y = 10% then z = logy. (2.14)
Since In 10 = 2.3026 (or 10 = e2.3026), the logarithms to the two
bases are related by
(2.15)
so that
In y = 2.3026z = 2.3026 log y. (2.16)
The derivative ofa logarithm follows from the definition (2.14) and
from the fact that the reciprocal of dy/dx is dx/dy. Thus
dy
if y = eX, - = eX = y
dx
dx 1 1
dy = dy/dx = y'
The chain rule and differentiation by substitution 39
which may also be written as
1 dy
if x = lny, dx = d(lny) = -dy, - = d(lny). (2.17)
y y
Example 2.5
The rate of change of vapour pressure p of a liquid with temperature T
is related to the heat of vaporization !l.H by
dp !l.H
dT= RT2P.
This can be transformed by using both dp/p = d(lnp) and
d(I/T) = - (I/T2)dT into
d (In p) dpT2 !l.H
d(l/T) = -p dT = -R·
Example 2.6
Show that if the rate coefficient k of a chemical reaction depends on
temperature T according to
k=Ae- E / RT
where A and E are constants, then
dInk E
d(I/T) = - R
Example 2.7
Show that
2.6 The chain rule and differentiation by substitution

The chain rule is used when introducing a new dependent variable into
a function that contains only one independent variable. Thus if y is
known as a function of x, a two-dimensional graph can be drawn and
dy/dx is defined. We may choose to introduce a new variable, say u,
which is also a function of x. We then have three variables x, yand u but
only one is independent and it is still a two-dimensional problem. If the
new variable were not a function of only the existing ones, we would be
moving into three dimensions, having then two independent variables,
and would have to resort to partial differentiation as discussed in the
next chapter.
Given the three variables x, y and u of which only one is independent,
we can define three derivatives such as dy/dx, dy/du and du/dx (du/dy
is simply the reciprocal of dy/du). These are related by the chain rule
dy dydu
(2.18)
dx dudx'
This follows from the definition of a derivative as a limit and from the
fact that the product of two limits is the limit of the ,product.
The introduction of a new dependent variable is called substitution,
and its purpose in differentiation is to convert an otherwise com-
plicated function into one to which the simple rules can be applied. This
is best illustrated by examples.
Example 2.8
Let
y = (2X2 _1)3.
The right-hand side could be evaluated algebraically and then dif-

ferentiated term by term. A neater solution is to make the substitution
u = 2X2 -1,
then
du/dx = 4x
and
dy/du = 3u 2 = 3(2x-l)2.
Hence
dy dydu 2
-=--= 12x(2x-l).
dx dudx
This can be expressed as a general formula, that

if y = f(x)" then dy = nf(x)"-lf'(x)dx. (2.19)
Example 2.9
Let y = eax , where a is a constant. Substitute
u = ax,
then
du/dx =a
Turning points: maxima, minima and points of inflection 41
and
y = e", dy/du = e".
Hence
dy dydu
- = - - = ae" = ae DX •
dx dudx
In general
if y = e/(x) then dy = e/(X)f'(x)dx. (2.20)
Example 2.10
Find the derivative with respect to temperature T of the expression
Z = 1/(e hv /kT -1).
Substitute
u = (e hv /kT -1)
then
du _~ehv/kT
dT kT2
and
1 dz 1 1
Z=-, - u2
u du (e hv /kT _1)2
Hence
dz dzdu hv e hv /kT
dT dudT kT2 (ehv/kT _1)2 .
2.7 Turning points: maxima, minima and points of inflection

Fig. 2.4 is drawn to illustrate the various points A to E on a curve which
can be detected by differential calculus and one point F which cannot.
Since the derivative dy/dx measures the slope of a curve, when
dy/dx = 0 we have a horizontal tangent and y is said to be stationary.
We then have a maximum as at A, a minimum as at B or a horizontal
point of inflection as at C or D. These may be distinguished by
successive differentiation.
The second derivative d 2 y/dx 2 is the rate of change of slope. If this is
negative the slope is decreasing as at A, where it is positive on the left
and negative on the right, and point A is a maximum. Conversely, if
d 2 y/dx 2 is positive we have a minimum, as at B.
When both dy/dx and d 2 y/dx 2 are zero we continue with successive
y
x
Fig. 2.4
differentiation. The conditions d 2 y/dx 2 = 0 and d 3 y/dx 3 positive are

those for a minimum in the slope dy/dx. This corresponds to
horizontal point of inflection C, where the slope decreases to zero and
then increases again as we pass through the point. Point of inflection D
is the converse, where the slope has a maximum value of zero being
negative on either side, so that dy/dx = d 2 y/dx 2 = 0 and d 3 y/dx 3 is
negative.
In general, we continue with successive differentiation until a non-
zero derivative is found. When this is the nth derivative and n is even,
the conditions for a maximum (negative derivative) or minimum
(positive derivative) apply. When n is odd, the corresponding con-
ditions for a horizontal point of inflection are used.
The conditions so far described are familiar ones, but points E and F
also occur in physical chemistry. At E the slope is not zero but is
nevertheless a minimum, as can be seen by noticing that from D
onwards the slope becomes increasingly negative towards E and then
increases (becomes less negative) beyond E. Point E is therefore
characterized by dy/dx negative, d 2 y/dx 2 zero and d 3 y/dx 3 positive.
The endpoint of an acid-base titration is at the corresponding point of
inflection on a rising curve of pH against quantity, m say, of added base.
These conditions are therefore d(pH)/dm positive, d 2 (pH)/dm 2 zero
and d 3 (pH)/dm3 negative.
Point F is where the curve suddenly terminates; this is called an
endpoint maximum. Since the definition ofthe derivative dy/dx requires
that a limit is taken of the slope of a chord as the endpoints move closer
together, such a limit is possible when approaching F from the left but
Turning points: maxima, minima and points of inflection 43
not from the right. Such a one-sided derivative cannot be found
analytically. This occurs when the vapour pressure ofa liquid is plotted
against temperature; the curve vanishes at the critical point.
Example 2.11
Given y = x4+4,
dy = 4x3 = 0 at x = 0,
dx
d 2y = 12x2 = 0 at x = 0,
dx 2
d3y
-=24x=0 at x = 0,
dx 3
d4y
dx 4 = 24.
Hence the fourth derivative is the first one to be non-zero at the turning
point x = O. Since n is even and the derivative is positive the point
x = 0, y = 4 is a minimum.
Example 2.12
Show that the curve In y = x 3 + 1 cuts the y-axis at a horizontal point of
inflection.
Differentiation of the equation gives
dy
- = 3x 2 dx,
y
dy
dx = 3x 2 y.
The second derivative may be obtained either by differentiating the
equation a second time to give
or by differentiation of the expression for dyldx to give, as before,
d2 y d (dY) dy
dx2 = dx dx = 6xy + 3x 2 dx = 6xy + 9x 4 y,
and similarly
:~ = :x (:~ ) = (6x + 9x 4 ) :~ + y(6 +36x 3)
= 54x 3 y+ 27x 6 y+ 6y.

The point at which the curve cuts the y-axis is when x = 0, so that
In y = 1, y = e = 2.718. From the above, when x is zero both dyldx and
d 2Yldx 2 are zero, but d 3 Yldx 3 = 6y = 16.3 and so positive, and so we
have a horizontal point of inflection of type C in Fig. 2.4.
Example 2.13
The velocity v of a chemical reaction is given by
v = kl(a _X)2 + k2x(a -x),
where kl' k2 and a are positive constants, x is variable. Find the
condition for a maximum in velocity, and the relation between kl and
k2 such that this occurs at x = a14.
The required condition for a maximum is dvldx = 0, d 2vldx 2
negative, where
dv
- = -2kda-x)+k 2(a-2x)
dx
=0
which is negative if k2 > k 1 •
So that
kl a-2x
for maximum velocity.
k2 = 2(a-x)
Put x = a14, then
kl _ al2 _ 1
k2 - 3al2 - 3
for the maximum to be at x = i.
This is the equation for a reaction which is second order in the reactant
together with catalysis by-product.
Maxima and minima subject to constraint 45
Example 2.14
Find the ratio of height to diameter for a cylinder so that the surface
area is a minimum for a given volume.
Put V = 7tr 2 h and A = 27trh + 27tr2. We have two simultaneous
equations to satisfy, that V is constant and A is a minimum. This may be
solved in various ways. We could insert Vas a constant and eliminate h
by writing
to give
A = 2V/r+27tr2
as an equation in the variables A and r, and use the condition
dA/dr = o. A neater solution is obtained by differentiating the
expressions both for V and for A and putting d V = 0 for constant
volume and dA = 0 for minimum surface area. Thus
dV = 27trh dr + 7tr2 dh
r
=0 if dr =
dh 2h
and
dA = 27th dr + 27tr dh + 47tr dr
=0 if :~ = - h : 2r
and these simultaneous equations give

2h = h+2r, h = 2r
as the required condition. Thus the most economical shape to make a
cylindrical tin is with the height equal to the diameter.
2.8 Maxima and minima subject to constraint; Lagrange's method

of undetermined multipliers
Lagrange's method is a general one for solving problems in maxima
and minima in the presence of constraints that can be expressed as
simultaneous equations. A simple example is to find the point on a
known curve that is nearest to the origin; the distance r will be a
minimum when dr = 0, which, with the equation of the curve as a
constraint, gives two simultaneous equations. The unknowns are the
two coordinates of the required point so that the problem has a unique
solution.
A related problem is to find not the maximum or minimum value of a
given function, but the function itself such that a stationary condition
applies. This is called the calculus of variations and is discussed in
Section 5.7.
The procedure in Lagrange's method is to write each of the
simultaneous equations in differential form, and to eliminate variables
in the usual way by multiplying equations by factors and adding them
together, the factors being the Lagrange multipliers. This provides an
alternative method of solving problems in which we have as many
simultaneous equations as variables, when numerical values can be
obtained for the multipliers as in the following Examples 2.15 and 2.16.
The method is particularly useful, however, when we have more
variables than equations, a solution then being obtained in terms of
multipliers whose values remain undetermined as in Example 2.17.
Example 2.15
Find the minimum distance from the origin to the curve y = 2 - x 2 •
The first equation is obtained by expressing the distance r from the
origin to any point (x, y) as
r2 = x2 + y2
and minimizing by differentiating and putting dr = 0, to give
2rdr = 2xdx+2ydy = 0
so that
xdx+ydy=O. (2.21)
The second equation is the constraint that the point (x, y) must lie on
the curve. We express that also in differential form by writing the
equation of the curve with zero on the right-hand side and differentiat-
ing. Thus
y_2+X2 = 0,
dy+2xdx = O. (2.22)
We solve these two simultaneous equations in the usual way by
multiplying equation (2.22) by a constant factor IX (the Lagrange
multiplier) and adding to equation (2.21). This gives
xdx+ydy=O
and
2lXxdx+lXdy = 0
so that
x(l + 21X)dx + (y + lX)dy = O. (2.23)
We now choose ex so as to eliminate one of the differential variables,
such as by putting
(1 +2ex) = 0, ex = -1/2.
The other term in (2.23) must then also be zero, so that
(y+ex)dy = (y-t)dy = 0
and, since dy may take any value, y = t. The equation of the curve then
gives
x2 = 2- y =!
and the required distance is obtained from
Since in this case we have as many equations as unknowns, an

algebraic solution of the simultaneous equations would give r as a
function of x, so that
r2 = x 2 + y2 = x 2 + (2 _ X 2)2
r2 = X4 - 3x 2 + 4.
Differentiation then gives
2rdr = (4x 3 -6x)dx
so that when dr/dx = 0, 4x 3 = 6x, x 2 = 3/2, r = fil2 as before.
Example 2.16
Find the shape of a right circular cone having minimum surface area for
a given volume.
If the cone has base radius r, vertical height h and slant height a, we
have a 2 = r2 + h 2 and the volume V and surface area A are given by
V = nr 2hl3
A = nr2 + nra.
For minimum A, dA = 0 giving
2nr dr + nr da + na dr = 0,
(2r+a)dr+rda = O. (2.24)
For a given volume, d V = 0 giving
t nrh dr + !nr2 dh = 0,
2hdr+rdh = O. (2.25)
and from r2 + h 2 - a 2 = 0
rdr+hdh-ada = O. (2.26)
We now take (2.24) + ex (2.25) + P(2.26) to give
(2r + a + 2exh + pr)dr + (r - pa)da + (exr + fJh)dh = O.
Choosing ex and p to eliminate the terms in da and dh, we put
r-pa = 0, p = ria
and
exr+ph = 0, ex = - phlr = -hla,
so that
2r+a-2h2Ia+r2Ia = O.
But h2 = a 2 - r2, so that
3r 2 + 2ra - a 2 = (3r - a) (r + a) = 0,
and since only positive values have physical significance, a = 3r.
An algebraic solution of the equations is possible but Lagrange's
method gives the neater solution.
Example 2.17 The Boltzmann distribution

To determine the most probable distribution of molecules amongst
quantum states: the Boltzmann distribution.
This is concerned with the most probable way in which a given
amount of energy will be distributed amongst molecules. It combines
the probability theory of Section 1.9 with maximization subject to
constraints, and also uses Stirling's theorem, equation (1.23).
The physical model is to suppose that we have a large number of
identical particles, each of which may be in anyone of a large number of
possible quantum states, each state having a characteristic energy. At
any instant, we suppose a particular number nj of particles to be in each
quantum state i, and this is called a distribution. Probability enters into
the argument by giving the number of ways of obtaining a particular
distribution (particular numbers of particles in each state); it is this
number of ways, or probability, that we seek to maximize subject to
constraints.
This is an example of the use of Lagrange's method when the number
of variables (the numbers n 1 , n2, ... ) is very large. The argument turns
on how many of these variables are independent, coupled with having
an equation in which the sum of a series of independent terms is zero. In
simplest form, if say x + y = 0 and if x and yare independent, then we
must have x = 0 and y = 0; otherwise any non-zero value of x would
require an equal and opposite value of y, but independence forbids that
because the value of y is not determined by the value of x.
Suppose that we have a fixed total number N of particles, and each
one has the same accessible set of quantum states, which have energies
81> 82> ••• ,8;, •••• We may regard the particles as objects and the
quantum states as boxes; from Section 1.9 the number of ways of
arranging N objects in boxes containing nl' n2, ... , nj, ... in each box
is the probability P of that arrangement, given by
N!
P = '"
nl·n2.·.· ni ... ·
(2.27)
The most probable arrangement is the one that can occur in the
maximum number of ways; this is when dP = 0 or, more usefully, when
d In P = dP / P = 0
so that
d InP = d InN! -d Innl! -d Inn2! - ... = o. (2.28)
If it were simply a matter of placing objects at random into boxes the
most probable arrangement would be an equal number in each box.
The problem is complicated by constraints, that the total number of
particles is fixed and that the total energy is also fixed. These constraints
give
N = nl + n2 + . . . + ni + ... = constant (2.29)
and since the ni particles in the state with energy 8i contribute ni8i to the
total energy,
E = nl81 + n282 + ... + ni8i + ... = constant. (2.30)
The variables in this problem are the numbers nl, n2' ... , n;, ....
Since we have three simultaneous equations, all but two of these
variables will be independent. We apply Lagrange's method by adding
(2.28) to at times the differential form of (2.29) and p times the
differential form of (2.30). The numbers nl, n2, ... that we seek are
average values, and so are treated as continuous variables rather than
integers. Hence
dN = dnl +dn2 + ... +dni+ ... = 0 (2.31)
and
dE = 81 dnl + 82 dn2 + . . . + 8i dni + ... = o. (2.32)
Before doing the addition, we modify (2.28) by use of Stirling's
approximation, equation (1.23)
In N! = N In N - N for large N,
giving
d InN! = In NdN.
Also, since N is constant, dN = 0, so that the term in N disappears from
(2.28). This leaves
(2.33)
We now add (2.33) + (X(2.31) + P(2.32) to obtain
(In nl + (X + peddn 1 + (In n2 + (X + pe2)dn2 + ...
+ (In ni + (X + pe;) dn i + . . . = O. (2.34)
We now choose the multipliers (X and p so as to eliminate n 1 and n2'
Equation (2.34) is then a sum of terms, each term containing only one of
the independent variables ni so that in order for the sum to be zero, each
term must be zero. Hence
(In ni + (X + pei)dni = 0,
The solution contains the undetermined multipliers (X and p. The term

e -cz may be written as a constant A, and physical reasoning identifies p
with lila', giving the Boltzmann distribution law in the form
ni = Ae-e.!kT.
2.9 Series
The use of series provides a valuable extension to algebraic methods.
We can use a convergent series of terms to represent quantities that
cannot be expressed in a simpler closed form, such as 1t, e or sin 0, and
also problems that cannot be solved by simple algebra often yield to a
solution in terms of series. The subject of series arises at this stage in our
discussion because the series with most physical significance can be
based upon Taylor's theorem, which is discussed in the next section.
A mathematical series contains terms that are related to each other,
so that a formula exists for calculating any particular term. Such series
are useful in practice only when successive terms become smaller and
smaller so that the series converges. We may then use the series either to
calculate its sum or as a source of successive numerical approximations.
Series 51
2.9.1 Geomecric series
This is a series of increasing powers of a variable in the form
s = 1 + x + x 2 + x 3 + .... (2.35)
Successive terms will become smaller and smaller, so that the series
converges, if -1 < x < 1. The sum s of an infinite number of terms
may be obtained by the device of multiplying each term in the series by
x and subtracting the result from the original series, thus
sx = x + x 2 + x 3 + X4 + .... (2.36)
Subtracting (2.36) from (2.35) we obtain
s(1-x) = 1, s = 1/(1-x).
Example 2.18
The theory of the heat capacity of a crystal gives rise to the series
1 +e- hv / kT +e- 2hv/ kT +e- 3hv / kT + ....
This can be seen to be a geometric series by making the substitution
x = e- hv / kT,
so that the sum of the series is
s = 1/(1-e- hv/ kT ).
2.9.2 Power series and Taylor's theorem

A power series is a general case of the geometric series discussed in the
previous section, each of the successively higher powers of the variable
being multiplied by a coefficient. In a mathematical power series the
coefficients themselves form a series, and the convergence of the series
depends on the ratio of successive coefficients being less than the
reciprocal of the variable. In general terms a power series has the form
(2.37)
A convergent power series is one of the most useful ways of
expressing many mathematical functions, and is often a useful and
significant way of expressing the values of physical properties. If the
series is stopped at x", it is then called a polynomial of degree n.
The values of the coefficients in a power-series expansion of a
function can be expressed in terms of successively higher derivatives of
the function. This is Taylor's theorem, which may be derived as follows:
Given a continuous functionf(x) whose valuef(a) is known at x = a
and which is successively differentiable; we denote the successive
derivatives with respect to x by f'(x),f"(x), ... ,and the values of these
derivatives at x = a by f'(a),f"(a), .... We then assume that when x is
increased from a to a + h the increase inf(x), which isf(a + h) - f(a), can
be expressed as a power series in h in the form
f(a+h) =f(a)+alh+a2h2+a3h3+.... (2.38)
When x is expressed as a + h the derivatives become
f '( ) =f'( h) = df(a + h) = df(a + h) dh

x a+ dx dh dx'
but x = a+h so that dh/dx = 1 and so
f '( ) = df(x)
x dh'
The technique is then to differentiate (2.38) successively with respect

to h, and then to put h = 0 to give expressions for the coefficients al ,
a2' .... Thus
f'(a + h) = :h (f(a) + al h + a2h2 + a3 h3 + ... )
= al + 2a2 h + 3a3 h 2 + ... ,
and when h = 0,
Next
f"(a+h)=2a2+2x3a3h+ ... ,
so that at h = 0
f"(a) = 2a2'
Hence Taylor's theorem can be written in the form
f(a + h) = f(a) + hf'(a) + ~~ f"(a) + ~~ f"'(a) + . . . . (2.39)
When this series converges it provides successive approximations to the

functionf(x) in terms of the values of the function and its derivatives at
the point x = a.
This form of Taylor's theorem applies when we have only one
independent variable x. With more than one independent variable the
theorem is expressed in terms of partial derivatives, and this is discussed
in Section 3.8.
Series 53
A geometrical interpretation of Taylor's theorem is shown in Fig. 2.5.
Point A is on the curve y = f(x) at x = a. For increment h in x we obtain
point B on the curve. The series (2.39) then provides successive
approximations to an extrapolation of the curve from a known point A
in terms of the known slope, curvature and successively higher
derivatives, all at point A. The first approximation is
f(a + h) = f(a) + hf'(a), (2.40)
y
y= f(x)
f(o+h)
I
I
I hf'(o)
A
_________________ JI
f(o)
h I
I
I
I
o o+h x
Fig.2.S
which is point C on the tangent drawn from A, at x = a + h. This

neglects any curvature, and so will be a poor approximation if
curvature is present so that the slopef'(x) changes with x. This change
in slope is given by the second derivative and appears as the next term in
the series. Higher derivatives allow for higher orders of curvature and
give better approximations to the true shape of the curve.
Equation (2.40) is the basis of Newton's method of successive
numerical approximation to a solution of the equationf(x) = O. Given
an approximate solution x = xo, suppose (xo + h) is the true root so
that
f(xo +h) = O.
We can find an approximate value of h by using the first approximation
from Taylor's theorem that
f(xo + h) = f(xo) + hf'(xo)
so that
h = _f(xo)
!'(x o )
and our improved estimate of the root is
f(xo)
Xl = xo+h = x o - - - (2.41)
f'(xo)"
This can be repeated to improve upon the value Xl.
Example 2.19
Given that x = 1.5 is an approximate root of the equation
In X = x 2 - 2, use Newton's method to find a solution correct to four
decimal places.
Put
f(x) = lnx-x 2 +2
f'(x) = l/x - 2x.
f(1.5) = 0.1555, 1'(1.5) = - 2.3333
Xl = 1.5 - 0.1555/( - 2.3333) = 1.5666
f(1.5666) = - 0.0054, 1'(1.5666) = - 2.4949,

X2 = 1.5666 - ( - 0.(054)/( - 2.4949)
= 1.5666 - 0.0022 = 1.5645,

f(1.5645) = - 5.6 X 10- 6 ,
so that rapid convergence has been obtained. Hence x = 1.5645, correct
to four decimal places.
2.9.3 Maclaurin's theorem

Maclaurin's theorem may be regarded as a special case of Taylor's
theorem. Since Taylor's theorem gives the value of a function in terms
of its value at a point and the derivatives at that point, and since the
point at which the value of a function is often simplest is when x is zero,
it is particularly useful to write the series for a = 0, so that x = h. This
gives Maclaurin's theorem, which is
x2
f(x) = f(O) + x!,(O) + 2! f"(0) + .... (2.42)
This may be used to develop series expansions of functions that can

be successively differentiated. The following examples illustrate the
expansion of particular functions.
Series 55
Example 2.20
Let
f(x) = eX, f(O) = eO = 1,
then
f'(x) = eX, 1'(0) = eO = 1,
f"(x) = eX, f"(0) = 1,
so that the expansion becomes
(2.43)
This gives the useful approximation that, for small x, eX ~ 1 + x, the

series being rapidly convergent for small values of x.
This is an example of the use of the theorem rather than a derivation

of the series because eX is defined as the series, as a result of which the
derivative is eX.
Example 2.21
Let
f(x) = sin x f (0) = sin 0 = 0,
then
f'(x) = cos x, 1'(0) = cos 0 = 1,
f" (x) = - sin x, f" (0) = 0,
f'''(x) = -cos x, !"'(O) = -1,
so that
x3 x5
sinx = x - 3! + 5! - ... = x for small x. (2.44)
Similarly we can obtain

x 2 X4
cos X = 1 - 2! + 4! - .... (2.45)
The hyperbolic functions sinh x and cosh x are defined by

sinh x = (eX - e - X)/2
x3 x5
= x+ 3! + 5! + ...
coshx = (eX +e- X)/2
x 2 X4
= 1 + 2! + 4! + ...
and are useful because the terms do nl)t alternate in sign as do those of
sinx and cosx.
These examples show that the expansions of the functions are
consistent with the expressions for the derivatives. Maclaurin's
theorem gives a new series, however, in the following example.
Example 2.22
Let
f(x)= In(1 + x), f(O) = In 1 = 0,
f'(x) = 1/(1 + x), 1'(0) = 1,
f"(x) = -1/(1 + X)2, 1"(0) = -1,
so that
This has been written as the variable y because the definition of the
logarithm gives
y = In(1 +x) if (1 +x) = e Y
so that we may also write
y2 y3
e Y = (1 + x) = 1 + y + 2! + 3! + ...
so that
y2 y3
X = Y + 2! + 3! + .... (2.47)
Now equations (2.46) and (2.47) describe the same relation between x
and y; they differ in that the first gives y as an explicit function of x, and
the second gives x as an explicit function of y. This is an example of the
inversion of a series, which is discussed further in the next section.
Example 2.23 The binomial series

This is a series expansion of (1 + x)":
f(x) = (1 + xr, f(O) = 1,
f'(x) = n(1 + xr- 1 , 1'(0) = n,
f"(x) = n(n-l)(1 +xr- 2 , 1"(0) = n(n - 1).
Hence
"_ n(n -1) 2 n(n - l)(n - 2) 3
(1 + x) - 1 + nx + 2! x + 3! x + .... (2.48)
Series 57
The successive coefficients of the powers of x are called binomial
coefficients and have characteristic values, for example
(1 +X)4 = 1 +4x+6x 2 +4x 3 +X4.
The binomial series converges only when - 1 < x < 1. When x is
sufficiently small for x 2 not to be significant, we can neglect the third-
and higher-order terms so that
(1 + x)" = 1 +nx for small x. (2.49)
Example 2.24
The binomial expansion gives a useful approximation when an
expression contains an algebraic denominator, such as
1 1
(a + X)3 - a 3(1 + x/a)3
1 (
= a3
)-3
1 +~
= :3 [1 -3: + 6(~r -10 (~y + ...J

which will be convergent so long as x < a, and terms beyond the second
can be neglected if x/a is small enough compared with unity.
Example 2.25
Given the relation between the equilibrium constant K and the degree
of dissociation 0: for a solution of a weak acid at concentration c
0:2C
K=-1-'
-0:
if we assume (1 - 0:) ~ 1 we obtain 0: = j(K/c). Use the binomial

expansion, assuming K ~ c, to show that successive approximations
are
0: = j(K/c) - (K/c)j2 + (K/C)3/2/8 - ....
Equation (2.49) is a quadratic equation for 0: in terms of K and c;
co: 2 +Ko: -K = 0,
so that the exact solution is given by formula (1.11) for solving a
quadratic equation as
K 1 j 2
0: = - 2c ± 2c (K +4Kc).
It is not possible by simple algebraic manipulation to separate out from
this the term J(K/c). We therefore resort to a series expansion by
writing
~ = -~±J(~)(1+ ~yI2,
so that the expression in the right-hand bracket is nearly unity, which
means that the binomial expansion will converge rapidly. Hence
~= -~ + J(~)(1+~ -~ 1~2+ .. .)
as required.
2.9.4 Inversion of a series

Given a convergent power series
y = x+a2x2+a3x3+ ... (2.50)
we can calculate y for a given value of x. Suppose, however, that we wish
to find x for a given value of y. We could do so in particular cases by
numerical trial and error, or we can invert (or revert) the series so as to
obtain an expression for x as a function of y.
Assume that a power series exists for x in the form
x = y+b 2y 2 +b 3y 3 + ...
and substitute into equation (2.50)
y = (y+b 2y 2+b 3y 3+ .. .)+a2(y+b 2y 2+b3y3+ ...)2
(2.51)
For this to be true, the coefficients of the second and higher powers of y,
which appear on expanding the right-hand side, must all be zero. Hence
b2 +a2 = 0, b2 = -a2'
b3 + 2a2b2 + a3 = 0, b3 = 2a~-a3'
and so on. The series produced is
x = y - a2y2 + (2a~ - a3)y3 + (- 5a~ + 5a2a3 - a4) y4
+ (14a~ - 21a~a3 + 6a2a4 + 3a~ - as) yS + . . . . (2.52)
Series 59
The rate of convergence of the series (2.52) will not be the same as
that of the original series (2.50) so that different numbers of terms will
be necessary to obtain the same precision.
Example 2.26
Given the series (2.46), obtain the series (2.47).
By comparing (2.50) with (2.46) we obtain a2 = -1/2, a 3 = 1/3,
a4 = -1/4, ... , so that (2.52) becomes
x = y+ y2/2+ y3/6+ y4/24 + y5/120+ ....
which is (2.47).
An alternative method of inverting a series is by successive approxi-
mations. We write equation (2.50) in the form
x = Y -a2x2 -a 3x 3
and substitute successive approximations for x into the right-hand side.
The first approximation is x = y, which we substitute, retaining powers
of y up to y2, so that
This is then the second approximation. We next substitute the second

approximation and retain powers of y up to y3
X = Y -a2 (y - a2y 2)2 - a3y 3
= Y - a2y2 + (2a~ -a3)y3,
and this is the third approximation. We continue by substituting the
third approximation for x 2 , the second approximation for x 3 and the
first for x\ retaining powers up to y4, and so on. This method can be
used for a convergent series of any kind.
Example 2.27
The equation of state of a real gas is a relation between pressure p,
molar volume v and temperature T. The virial equation of state is a
power series in the density l/v
pv = RT[1 + B(l/v) + C(1/v)2 + D(I/v)3 + .. .].
An alternative is to use a power series in pressure in the form
pv = RT+B'p+Cp2+D'p3+ ... ,
and series inversion is used to obtain relations between the coefficients
B', C, D', ... and the virial coefficients B, C, D, ....
Write the series in the form
p = RT(ljv) + B'p(l/v) + C'p2(1/v) + D'p3(1jv) + ...
and the successive approximations are used to substitute for p on the
right-hand side.
First approximation,
p = RT(l/v).
Substitute, retaining powers up to (1/v)2.
Second approximation,
p = RT(l/v) + B'RT(1/v)2.
Third approximation,
p = RT(l/v) + B'(l/v)(RT/v + B' RT/V2) + C'(1/v)(RT/v)2
= RT/v + B'RT/V2 + (B'2 + RTC')RT/V3
Fourth approximation,
p = RT/v+ B'(l/v)[RT/v + B'RT/V2 + (B'2 +RTC')RT/v 3]
+ C'(l/v)(RT/v + B'RT/V 2)2 + D'(1/v)(RT/v)3
= RT/v+ RTB'/v2 + (B,2 + RTC')RT/V3
+ [B'3 + 3B'C'RT+D'(RT)2]RT/v4 •
The required relations between the coefficients are
B = B',
e= B'2 + RTC',
D = B'3+3B'C'RT+D'(RT)2,
etc.
2.9.5 Empirical curve fitting by power series

We may wish to fit a curve to experimental points so as to produce an
economical expression for the data, or so as to interpolate and
extrapolate in a way that is not subjective.
In the absence of any theoretical guidance as to the form of equation
to be used, the normal practice is first to choose variables, and so scales
for plotting the data, so as to obtain as near to a straight line as possible.
A common practice is then to use a power series, the numerical values of
the coefficients of which are found by curve fitting. This may be done in
The evaluation of limits by L'Hopita/'s rule 61
two ways. Given an equation
y =f(x) = ao+a l x+a2 x2 + ... ,
we can plot y against x and extrapolate to x = 0, the intercept being ao.
We can then calculate and plot (y - ao)/x against x, when the equation
becomes
(y -ao)/x = al + a2x + a3 x2 + ... ,
so that the extrapolation to x = 0 gives a l ' and so on. This geometrical
method has advantages in that the effect of successive terms is readily
apparent and the increasing scatter of points which is inevitably
produced shows when higher terms become irrelevant. The principal
difficulty is that uncertain extrapolation is involved, the method
depending on the estimation of limiting tangents.
The corresponding analytical procedure is to calculate values of the
coefficients by the method of least squares, discussed in Section 7.8. In
that case it is important to use the smallest possible number of terms.
This is because a power series does not necessarily converge, and if
more terms are used than are necessary the series can become
oscillatory; a curve of complex shape that passes through every point
merely reflects the accidental scatter of the data, a set of duplicate
measurements producing a different curve.
2.10 The evaluation of limits by L'Hopital's rule

We often need to find the limit of a function when the variables are
taken either to zero or to infinity. This can usually be done by simple
algebra, but difficulties arise when this leads to a product of zero and
infinity, or to the ratios % or 00/00, all of which are indeterminate.
Such difficulties may be resolved by the use ofL'Hopital's rule, which
is that if two functions f(x) and g(x) are both zero at x = a then
lim f(x) = lim f'(x) (2.53)

g(x)
x ..... x ..... g'(x)'
This can be shown by using Taylor's theorem to expand the two

functions
f(x) f(a) + (x - a)f'(a) + (x - a)2J"(a) + ...
g(x) = g(a) + (x -a)g'(a) + (x -a)2g"(a) + .. .'
so that when f(a) = g(a) = 0 we may divide numerator and
denominator by (x - a) to give
f(x) f'(a) + (x -a)f"(a) + .. .

g(x) g' (a) + (x - a)g" (a) + .. .
and as x -+ a the rule follows. It is most often used either when a = 0 or
when a = 00, and it applies equally well when the functions f(x) and
g(x) both tend to infinity.
Example 2.28
. sin x
11 m--
x-+O X
would give % at x = 0, but differentiation of both the numerator and

the denominator gives
· cosx - 1
11m -1-- .
x-+O
Example 2.29
Evaluate
lim (x In x).
This gives a product of zero and - 00 at x = O. We therefore write the

expression as a quotient and apply L'Hopital's rule:
lim (x In x) = lim 1ln/ x = lim 11//X 2 = lim (-x) = O.

x-+O x-·O X ",-+0 - x "''''0
Example 2.30
Find the condition for the hyperbola
x2 y2
---=1
a2 b2
to be rectangular (asymptotes perpendicular to each other).
We have
b2 x 2 _ y 2 a 2 = a 2 b2
and the asymptotes are straight lines of slope equal to the value of
dy/dx as x -+ 00 and y -+ 00. By implicit differentiation
2b 2 xdx -2a 2 ydy = 0,
dy b 2 x
dx = a 2 y'
The principles of Newtonian mechanics 63
which gives 00/00 as x and y tend to infinity. Differentiation of
numerator and denominator with respect to x gives
. m
11 dy l'
- = 1m 2
b2
X"" 00 dx X"" 00 a dy/dx
so that in the limit,

(dy/dx)2 = b 2/a 2, dy/dx = ± b/a.
The asymptotes will be perpendicular if they have slopes of ± 1, so that
the condition for a rectangular hyperbola is b = a.
Example 2.31
An alternative way of evaluating limits is by the use of series
expansions. Example 2.10 gives an equation that occurs in the theory of
the heat capacity of a monatomic solid. The high-temperature limit, as
T -+ 00, of the expression
hv e h./kT
kT2 (e h./kT _1)2
gives 0/0. Expanding the exponentials in the form

e h ./kT = 1 + hv/kT + (hv/kT)2 + ...
shows that for very large T,
(e hY/ kT -1) -+ hv/kT and e h./kT -+ 1,
so that the limit as T -+ 00 becomes k/hv.
2.11 The principles of Newtonian mechanics

Mechanics is concerned with force, mass, momentum and energy. The
subject is deduced from a minimum number of first principles, the
choice of which distinguishes one approach from another. In the
Newtonian approach we regard force as fundamental, and this is
usually the one adopted in introductory treatments because it can be
applied directly to particular cases. An alternative is to regard energy as
fundamental, as in the treatment of Hamilton and Lagrange, which is
discussed in Section 4.7.
Force is a vector quantity, and the conditions for equilibrium under
the action of forces is one application of the vector analysis discussed in
Section 3.9.
Newtonian mechanics can be based on three principles: that when
bodies interact with each other the forces of action and of reaction are
equal in magnitude and opposite in direction; that a body changes its
state of rest or uniform velocity only if acted upon by a force; and that
force = mass x acceleration. (2.54)
This will apply in any direction in space; if the velocity of a body in the x
direction is v = dx/dt, a force F in the same direction will produce
acceleration dv/dt = d 2 x/dt 2 given by
d2x
F = m dt 2 '
(2.55)
Example 2.32
The conventional order of definition of mechanical properties is, first,
velocity as rate of change of distance, then acceleration as rate of change
of velocity and, finally, force as mass x acceleration. This order can be
reversed with advantage in physical applications; we can use a
transducer to measure the force that is exerted on a sufficiently massive
body as a result of acceleration of a containing vehicle. Knowing force
and mass we can calculate the acceleration; knowing acceleration as a
function of time we can calculate velocity, and knowing velocity as a
function of time we can calculate the distance travelled. This is the basis
of inertial navigation systems.
In a time interval dt, the product F dt will be

d2x dv
Fdt = m dt 2 dt = m dt dt = mdv = d(mv).
Anticipating our discussion of integration, this gives
f t. Fdt = flnV. d(mv) = mV2 -mvl'

tl mV l
The product mv is called momentum and the integral on the left-hand

side is called impulse. If the force is constant the impulse becomes
F(t2 -td, so that
impulse = force x time = change in momentum. (2.56)
We could equally well regard momentum as fundamental and obtain
(2.54) as a consequence of (2.56).
When two or more bodies interact with each other the forces, and so
the impulses, acting will be equal in magnitude and opposite in
direction, so that the same applies to the changes in momentum. The
total momentum will therefore be constant, which is the principle of
conservation of momentum.
Energy may be defined as the result of doing work. When force F
produces displacement dx in mass m,
dv dx
Fdx = m-dx = m-dv = mvdv.
dt de
Integration then gives
I x> F dx = IV> mvdv = -!mv~ --!mvI.

Xl VI
The expression -!mv 2 is called kinetic energy and the integral on the left
is called work. If the force is constant the work done becomes
F(X2 -Xl) so that
work = force x distance = change in kinetic energy. (2.57)
The forces with which we are concerned are often due to interaction
with a field, as with mass in a gravitational field or electric charge in an
electric field. The field strength is defined as the force on unit mass or
charge. In the earth's gravitational field the force exerted on a body of
mass m is called the weight wand the field strength is denoted by g, so
that
w=mg.
By comparison with (2.54) we see that g can be identified as the
acceleration of free fall under gravity.
The corresponding force F exerted by an electric field of strength E
on a charge e is
F= Ee.
When a body is given a displacement dx in the direction opposite to
the force F exerted on it by a field, work F dx is done on the body and its
energy increases. Energy that changes with position in a field is called
potential energy and is denoted by V. Since force is in the direction of
decreasing potential energy we have
dV
dV= -Fdx, F= - dx'
so that
force = - potential gradient. (2.58)
If a body falls freely under gravity, or a charge is moved by an electric
field, the decrease in potential energy is equal to the work done, which
by (2.57) will be the increase in kinetic energy. Denoting kinetic energy
by T and potential energy by V we have
T+V = constant. (2.59)
For this to be true, all of the assumptions that have been made above
must be true. In particular, the only effect of the action offorce must be
to increase the velocity of a body. This will be so only in the absence of
friction. Movement through a resisting medium, such as a fluid, or
sliding on a rough surface introduces a resisting force which reduces the
acceleration of the body; some of the work done is converted into heat.
At the molecular level, linear motion is converted into random thermal
motion of the body and its surroundings, which is an irreversible
process in the thermodynamic sense. Thus, although the conservation
of momentum always applies to real processes, conservation of energy
in the simple mechanical sense does not unless some formulation which
includes heat is adopted.
The motion of a structured body in three-dimensional space, such as
that of a molecule in a fluid, is resolved into linear motion (translation)
of the centre of mass and rotation about the centre of mass. Rotational
properties are expressed in terms of polar coordinates (Section 1.7.5).
Our discussion of mechanics has, so far, neglected structure, as if the
masses concerned were point particles. The arguments used apply to
translation of the centre of mass of an extended body when it is treated
as if all the mass were concentrated at that centre of mass, and all the
forces were moved parallel to themselves so as to pass through that
centre of mass. An applied force not passing through the centre of mass
will tend to produce rotation of the body. We therefore introduce the
principle of moments; the moment of a force about a point is defined as
the product of the force and the perpendicular distance from the point
to the line of action of the force. For an extended body to be in
equilibrium under the action of forces, not only must the sum of the
components of the forces in any direction in space be zero, but also the
sum of the moments of the forces about any point must be zero,
clockwise and anticlockwise moments being given opposite signs.
A molecule may be regarded as a set of discrete masses at the
positions of the nuclei. Fig. 2.6 represents a diatomic molecule
consisting of masses m1 and m2 rotating about an axis perpendicular
to the paper through point O. Mass m1 moves with tangential velocity
V 1 in a circle of radius r 1 about O. In time interval dt the radius vector r1
Fig. 2.6
moves through angle dO and mj moves through an arc of length

dS j = r 1 dO. Then
(2.60)
where OJ is the angular velocity, dO/dt, measured in radian/second.

At constant angular velocity w, the linear velocity Vj will be constant
in magnitude but changing in direction. From the triangle of velocities
in Fig. 2.6, in time dt the mass mj gains velocity Vj dO in the direction
towards 0; this corresponds to an acceleration towards the centre given
by
.
centnpeta I acce I ·
eratton = Vj dO
dt = VjW = r l w 2 . (2 .61)
If we suppose the masses to be connected by a weightless string, this

must exert a force on m j , or the mass exerts a force on the string given
by
(2.62)
If the string were elastic, or for a chemical bond, the bond length would
increase with increase in w, this being called centrifugal stretch. Since
the string, or bond, must exert equal and opposite forces on the two
masses we have
and the position of the axis of rotation will be where

(2.63)
This is the condition for finding centre of mass, so that a molecule that
is freely rotating in space does so about its centre of mass.
A general motion of a rigid body in space, which includes both
translation and rotation, can be expressed as the sum of translation of
the centre of mass and rotation about the centre of mass. This may be
seen by considering some particular line fixed in the body and
perpendicular to an arbitrary axis of rotation; rotation through the
same angle about any parallel axis will then produce the same angle of
rotation of the line, together with translations of the line parallel and
perpendicular to itself. We may therefore choose any parallel axis to
describe the rotational part of the motion, for which we choose the one
through the centre of mass, and the remainder of the motion will be
translation. Thus when an athlete throws the hammer, the centre of
mass of the hammer travels in a parabola (Example 1.11), the hammer
meanwhile rotating about its centre of mass.
The kinetic energy T of our rotating diatomic molecule is the sum of
the kinetic energies of the two masses, so that
T = tm 1 vI +tm2v~
= !(m1ri +m2d)co 2.
The quantity in brackets is called the moment of inertia I of the
molecule about an axis through the centre of mass and perpendicular to
the axis of the molecule. Then
kinetic energy of rotation = t I (1)2. (2.64)
If the internuclear distance is , = '1 +'2 then (2.63) gives
and (2.65)
so that
1= m1 m 2 ,2.
m1 +m2
We define the reduced mass m* by
1 1 1
.=-+-,
m m1 m2
(2.66)
so that
(2.67)
In the general case of a molecule composed of masses m 1 ,
m2, ... held by chemical bonds at various positions in space, the centre
of mass is a point such that for any axis through that point
where 'i is the perpendicular distance from mi onto the axis. The
moment of inertia about a particular axis is then
1= 'Lmi'f.
When the molecule is planar we can choose rectangular coordinate axes
OX, Of through the centre of mass as in Fig. 2.7. Mass ml at point
P(x, y)will contribute m1 (PR)2 = m1 y2 to the moment of inertia about
the x-axis OX. Ifwe then choose arbitrary axes OX', Of' by rotating
the axes through an angle (), the point P will assume new coordinates
(x', y') relative to OX', OY' given by
y' = PN = PM -MN = PM -QR = ycos()-xsin(),
x' = ON = OQ + QN = OQ + RM = x cos () + y sin ().
y
yl\
\
\
\ P(X,y)
\
\
\
\
\
\
\
\
\
\
\
\
\ M
\
\ ,,.,,
, ' ....0' \ \ x
\
Fig. 2.7
The mass ml then contributes m 1 (y cos () - x sin ()2 to the moment of

inertia about OX'. The total moment of inertia about the new axis OX'
is then
I' = L mi (Yi cos () - Xi sin ()2.
The moment of inertia of our arbitrary planar molecule will therefore
vary with the angle (), and will have a stationary value if dJ' Id() = O. This
is when
:~ = - 2L mi (Yi cos () - Xi sin () (Yi sin () + Xi cos ()
= -2Lmix;y; = o.
The expression LmixiYi is called the product of inertia about the x- and
y-axes, and we see that if this is zero the moment of inertia will be a
maximum or a minimum. The axes for which this applies are called the
principal axes of the molecule, and the moments of inertia about the
principal axes are called the principal moments of inertia of the
molecule.
In three dimensions we find corresponding conditions, that when the
axes are chosen so that the products of inertia about each pair of axes
are zero,
the moments of inertia will have maximum or minimum values. When

we choose these principal axes as the ones about which we express the
rotational properties of the molecule, we obtain the simplest
expressions.
When a force acts to cause rotation of an extended body about an
axis, the moment of the force about the axis is called the torque. Thus if
force F is applied to mass m1 in Fig. 2.6, in a direction perpendicular to
the axis of the molecule, the torque about 0 is Fr 1 • The effect of such a
single applied force, as in a molecular collision, is complicated by the
fact that it produces both rotation about the centre of mass and
translation of the centre of mass. We therefore consider a simpler case,
which is when the molecule is polar and the force is produced by
interaction with an electric field in a direction perpendicular to the axis
of the molecule, as in Fig. 2.8. If the charges on m1 and m2 are + e and
- e respectively, then in field of strength E each mass experiences a
force Ee. These forces will be equal in magnitude and opposite in
direction, and this is called a couple. The total moment of the two forces
about 0 is then
Eer 1 + Eer2 = Ee(r1 + r2).
This moment depends only on the perpendicular distance r = r 1 + r2
between the parallel forces, and so will be the same about any axis
Ee
E m1+6 r1
0I
r2
-6
I~
~
... Ee
Fig. 2.8
perpendicular to the plane containing the forces. If the molecule were
inclined at an angle e to the direction of the field, as in Fig. 2.9, the
perpendicular distance between the forces would be r sin e so that, in
general,
torque = Eer sin e = J.lE sin e, (2.68)
£e
+e
£
ma
r----------~-i-n~9------~-e
£e
Fig. 2.9
the product re = J.l being called the dipole moment of the molecule.
Such a pair of equal and opposite forces will have no effect on the
position in space of the centre of mass of the molecule because, when
they are translated so as to pass through that centre, their resultant is
zero. The effect of the field is therefore to apply torque and produce
pure rotation of the molecule.
For linear motion equation (2.56) applies, so that
force = rate of change of linear momentum. (2.69)
In Fig. 2.8 the forces and the momenta are both vectors (Section 3.9)
passing through the mass points and perpendicular to the axis of the
molecule. We define angular momentum as the moment of the
momentum, so that mass m1 has linear momentum m1 V1 and angular
momentum m 1V1 r1 about an axis through O. From this definition and
(2.69) we obtain
moment of force = torque
= rate of change of angular momentum, (2.70)
the total angular momentum of the two masses about 0 being
angular momentum = m 1v 1 r 1 +m2v2r2 = (m 1 d +m2rDw = Iw.
(2.71)
The rotational kinetic energy of our molecule is simply the sum of the
kinetic energies of the masses, so that
kinetic energy = tml vf + tm2v~ = t(m1rf + mzd)w Z = tIw 2. (2.72)
The relations used to describe angular motion are thus the same as
for linear motion if we replace force by moment of force, mass by
moment of inertia and velocity by angular velocity.
The motion of a structured body in three-dimensional space is
resolved into translation of the centre of mass and motion relative to
the centre of mass. For a rigid body we need to consider only
translation and rotation, but for molecules we must include also the
vibration of the chemical bonds. This vibration is discussed in Example
6.9, and makes additive contributions to both potential energy and
kinetic energy.
Problems involving collisions between structureless particles are
solved by using the principle of conservation of linear momentum.
When rotation is possible, we have a corresponding conservation of
angular momentum. When structured particles collide, equal and
opposite forces act on each one at the point of contact. If we calculate
the total angular momentum of the two bodies both about the same
arbitrarily chosen axis, the forces acting on impact contribute equal and
opposite moments about that axis. By equation (2.70) these produce
equal and opposite rates of change of angular momentum, so that the
total angular momentum about any axis remains unchanged.
In problems involving real molecules it may be necessary also to take
account of the energies of allowed quantum states; these impose
additional constraints, but the principles outlined above still apply.
CHAPTER 3
Differential calculus in
three or more dimensions;
partial differentiation

Typical three-dimensional problems are the (p, v, T) properties of a gas
or the vapour pressure of a two-component mixture. Here we have
three variables and an equation connecting them, so that two variables
are independent (see Section 1.4). In such cases the geometrical method
is to draw a three-dimensional diagram and then to consider two-
dimensional sections, which apply when the third variable is held
constant. Thus we may consider p against vat constant Tfor a gas, or p
against mole fraction x at constant T for the mixture. The correspond-
ing analytical problem is: Given a function z = z (x, y), how does z vary
with x when y is kept constant, and with y when x is kept constant?
The general principle of taking two-dimensional sections of multi-
dimensional space enables us to use the methods of differential calculus
developed in the previous chapter. Given a relation z = z (x, y), which
defines a surface in three dimensions, constant y means a plane parallel
to the (z, x) plane as in Fig. 3.1. Ifthe surface cuts the plane in the curve
shown, we can apply the methods of two-dimensional calculus to this
curve. A tangent at point P will have a slope which is 'dz/dx in the plane
at constant y'. This is written (oz/ox)" to be read as 'partial dz by dx at
constant y'. The analytical definition of this partial derivative is:
If z is a continuous function of the independent variables x and y,
then
(0 z/ uX,
!'l ) _ I· z(x+~x,y)-z(x,y)
- 1m i: (3.1)
6" .... 0 ux
74 Three or more dimensions; partial differentiation
z
x
z,x plane at constant y
Fig. 3.1
where only the variable x is changed from x to x + {)x, y being kept

constant.
Analogously, we can define partial derivatives in the two other
planes, at constant x and at constant z, denoted by (iJz/iJy)x and
(iJy/iJx)z' The derivatives in the previous chapter, where only two
variables were concerned, are called total derivatives, and there are
important differences between the properties of total and of partial
derivatives. One property that they have in common is that the
reciprocal of, say, (iJy/iJx)z is (iJx/iJy)z just as l/(dy/dx) = dx/dy.
When we have more than two independent variables, we define the
partial derivative with respect to anyone of the independent variables
by keeping the others constant. Thus if z depends on x, y, u and v we can
write
z = z(x,y,u,v)
and can define partial derivatives such as (iJz/iJx),.u.v, (iJz/iJy)x,u,v' etc.
The rules for partial differentiation are the same as in the previous
chapter, it being necessary only to remember that variables that are
Significance and notation 75
being held constant are treated in the same way as any other constants.
For example, if
z = 2X2 - 3xy + 7
then
(OZ/ox)y = 4x - 3y
and
(OZ/oy)x = - 3x.
The method of implicit differentiation described in Section 2.4 can
also be used. In the above example we can find (ox/oY)z by differentiat-
ing the equation at constant z, which gives
0= 4xdx-3xdy-3ydx at constant Z
so that
(4x - 3y)dx = 3xdy at constant z
(ox/oY)z = 3x/(4x - 3y).
This example shows that the chain rule of Section 2.6 does not apply
to the partial derivatives when a different variable is held constant in
each case; with partial derivatives a minus sign appears, thus
(oz/ox)y(ox/oY)z = 3x = - (oz/oY)x. This relation is derived analyti-
cally in Section 3.6.
When a function is defined in terms of particular independent
variables, such as z (x, y), it will be clear that partial differentiation with
respect to one variable implies that the other is to be kept constant. We
can then use the abbreviated notation oz/ox to imply constant y
without ambiguity. In physical applications, however, we normally
have a wide choice of independent variables. Thus for some property X,
if we write oX / aTit may mean at constant v or at constant p, these being
different quantities (the relation between them is discussed in Section
3.6). We therefore need either to use the full notation, as (oX/aT)v and
(oX/aT)p, or to begin with an explanatory statement such as 'using
independent variables v,T' so that oX/aT then means (oX/aT)v and not
(oX/aT)p.
The higher-order differential coefficients such as (0 2 z/ ox2 )y are a
straightforward extension of the rules for total differentiation, but a
new possibility also arises in the form of a mixed second derivative. This
may be written in the form 0 2 z/oxoy or more explicitly as
o:;y == (:x (:; )x)y == (:x)y (!;)x·

These alternative forms all show that z is first to be differentiated with
respect to y at constant x, the result then being differentiated with
respect to x at constant y. From the above example we have
oz 02 Z
oy = -3x, oxoy = -3.
If, on the other hand, the order of differentiation were reversed by

differentiating first with respect to x at constant yand then with respect
to yat constant x, the same result would be obtained. This illustrates an
important principle, that under proper conditions the order of
successive differentiation may be reversed. The properties of partial
derivatives are so important in chemistry that we give alternative proofs
of that principle, here and in the next section.
A geometrical interpretation of a mixed second derivative is that it
measures the rate of change of slope as we move in a perpendicular
direction over a surface. Using x and y as independent variables, the
rate of change of slope oz/ox with change in y is
o oz
oyox'
The derivative in reverse order is the rate of change of oz/oy with
change in x. We can move from a point z (x,y) to a nearby point
z (x + ~x,y + ~y) by changing one variable at a time, corresponding to
moving along two sides of a rectangle. Fig. 3.2 shows a projection of
this into the (x, y) plane, the z-axis being perpendicular to the plane of
8y
z(x,Y)
O Z(X+h,Y+6Y )
fix
Fig. 3.2
An alternative approach to calculus 77
the paper. By changing x first and then y the change in z is
oz (oz 0 oz )
~z = ox ~x + oy + ox oy ~x ~y,
whereas change first in y and then in x gives
oz (oz 0 oz )
~z = oy ~y + ox + oy ox ~y ~x.
In each case the expression in the brackets is the slope at z (x, y) plus the
increase in slope on moving by either ~x or ~y. The two expressions will
be the same so long as the order of successive partial differentiation can
be reversed. The expressions must be the same if a unique surface exists
over which the alternative movements are made, and the conditions for
that are that the variables shall be independent and the function and its
derivatives shall be finite, continuous and single-valued.
3.2 An alternative approach to calculus

The interpretation of calculus as the behaviour of functions in the limit
of infinitesimal changes that was adopted in the previous chapter is the
more conventional one. An alternative approach, which has some
advantages, is to consider the behaviour in terms of changes whose
small magnitude is actually assessed at each stage of the argument. This
regards the derivative not just in the limit but as differing from a ratio of
infinitesimal quantities by a term that is zero to a number of decimal
places. We write
( oz) = z(x + h,y) - z(x,y) + O(P) (3.2)
ox y h
where O(P) means a term which is zero to p decimal places. As the
increment h in x is made smaller, so p becomes larger. The advantage of
this interpretation of a derivative is that we can then multiply both sides
by h to obtain
h(!:)y = z(x+h,y)-z(x,y)+hO(p).
We are now able to write the mixed second derivative by applying the
same rule; increase y by a small increment, for which we again choose h,
and write the increase in the function again with an error O(P). Hence
2 0 OZ 2 02 Z [
h oyox = h oyox = z(x+h,y+h)-z(x,y+h)]
-[z(x+h,y)-z(x,y)]+h2 0(p). (3.3)
When written in this form, the same result is obtained by partial
differentiation in either order, because
82 z
h2 8x8y = [z(x+h,y+h)-z(x+h,y)]
- [z(x, y+h)-z(x, y)]+h 2 0(P), (3.4)
and simple algebraic manipulation shows the right-hand sides of (3.3)

and (3.4) to be the same. For this to apply, the variables x and y must be
independent, so that we can increase either variable by h without
change in the other variable.
3.3 The total differential

We have seen that if we have only two variables, and so two-
dimensional diagrams, all derivatives are total derivatives, but with
three or more variables we use partial derivatives. We have also seen
that although the total derivative dy/dx denotes the operation of
differentiation on the function y = f(x), we can in fact separate this as
the ratio of dy to dx. We shall now see how we can use a total
differential of this kind even with more than two variables.
The geometrical interpretation of the total differential dz is that it is
an increment in z, presupposing that it is such a small increment that
any conclusions we draw are actually valid only in the usual calculus
limit. We have seen in the previous section that we can if necessary place
this on a firmer footing by stating explicitly that we are in error by a
quantity that is zero to a certain number of decimal places.
The important thing about a total differential is that it is a real
quantity. If we have an equation for a surface in three dimensions in the
form z = z(x, y), the meaning of dz is the increase in z when we move
over the surface from one point on the surface to another one which is
very close to it. It could be regarded as the increase in altitude when we
walk obliquely across a hillside. This geometrical interpretation is here
very useful because it is easily seen that we can move from one point to
another on such a surface by a variety of routes, Consider a particular
path which moves first in the plane at constant x, and then turns
through a right-angle to move in the plane at constant y so as to reach
the desired finishing point. The gradients in these two planes are given
by the partial derivatives at constant x and at constant y respectively.
In two dimensions, the slope of a tangent is dy/dx, which is the
increment in y divided by the increment in x. We then have the trivial
The total differential 79
relation
meaning that the increment in y is the derivative times the increment in

x. When the same relation is applied to a two-dimensional section of
multidimensional space it becomes
dz = (!:)y dx.
It is important to notice that this expression contains a mixture of total

and partial differentials. The total differentials dz and dx are potentially
real increments; dz is, so far, the increase in z when we move distance dx
in the plane at constant y.
Combining this expression with the principle that we can make a
movement over a surface by movements in perpendicular planes, the
geometrical argument shows that we then add the increments in z
produced by the component moves. Thus
dz = (!:)y (:;)x
dx+ dy. (3.5)
This is the definition of the total differential of z = z (x, y). When we

have more variables a term is included for each, so that if
z = z(x,y,u,v, ... )
then
dz = (!:) y,u,v, ... dx + (:;) x,u,v, ... dy +

(3.6)
Example 3.1
The pressure of a gas depends on the temperature Tand molar volume
v. Thus p = p (T, v) so that
dp = (Z)v (::)T
dT+ dv.
This equation means that the differential change in pressure produced

by simultaneous differential changes dT in T and dv in v is given by the
right-hand side of the equation. These differential changes do not have
any physical significance because they need to be integrated (Chapter 4)
before they represent real changes, and integration will be possible only
when we know the physical properties of the gas. We cannot, of course,
predict such properties by purely mathematical means.
If we make the physical assumption that the gas is perfect we then
have pv = RT, in which case
R p
and
v T
so that in this case
p P dp dT dv
dp = -dT--dv or =---
T v p T v
The definition of a total differential is quite general, and applies

whether or not the variables are independent. This will be shown in the
next section since it provides useful practice in handling differentials.
The definition of the total differential is the basis from which other
useful relations are derived, so that we can use these relations even
when we do not use only independent variables, which may happen in
physical applications either by accident or by design.
3.4 General expression for a total differential

The geometrical justification of the definition of a total differential
given in the previous section was based upon the assumption that a
surface could be drawn in space of dimensions equal to the number of
variables. The existence of an equation for the surface means that all
but one of the variables can be independent. It was also stated that the
definition applies even when the variables are not independent and this
will now be shown.
Suppose a quantity z depends upon the independent variables x and
y,
z = z(x, y), x and y independent. (3.7)
Suppose we now choose to introduce another variable wand write
z = z(x, y, w). (3.8)
We now have four variables, z, x, y and w, but still only two of them are
independent. This means that a relation must exist between w, x and y,
which we write as
w = w(x, y). (3.9)
Since z may now be expressed as a function of any two of the variables
we may also write
z = z(x,w) (3.10)
General expression for a total differential 81
and
z = z(y, w). (3.11)
Since equations (3.7), (3.9), (3.10) and (3.11) all express w or z as a
function of two variables, we can write the corresponding expressions
for the total differential dz:
from (3.7)
dz = (!:), (:;)x
dx+ dy, (3.12)
from (3.9)
dw = (~:), dx+ (~;)x dy, (3.13)
from (3.10)
dz = (!:)w dX+(!:)x dw, (3.14)

from (3.11)
dz = (:;) w dy + (!: ), dw. (3.15)
Each of these expressions contains the total differential of three of the

four quantities z, x, y and w. In particular, (3.14) and (3.15) will remain
true if we choose to hold the fourth quantity constant. Dividing (3.14)
by dx at constant y we obtain
and dividing (3.15) by dy at constant x gives
(3.17)
Substitution of (3.16) and (3.17) into (3.12) and using (3.13) gives
dz = (!:) w,'
dx+ (:z)
y w,X
dy+ (!:) "X
dw, (3.18)
which would be obtained if we had assumed (3.8) in the first place and
written the total differential to include the superfluous variable.
The number of independent variables that properly apply to a
particular application is a physical problem, not a mathematical one.
The significance of the above is to show that we do not make any
mathematical errors as a result of a physical error in choosing that
number.
3.5 Exact differentials
This term applies to an expression of the form X dx + Y dy, and means
that a quantity, z say, exists whose total differential is of that form, so
that
dz = Xdx+Ydy. (3.19)
This will be so if the relation (3.19) is identical with
dz = (!:)y dX+(!;)x dy (3.20)
and the test for exactness is to see if this is so. This can be done by taking
the mixed second derivative and checking that the order of differenti-
ation can be reversed. Thus to confirm that
and Y= (oz)
oy x
we calculate
(~~)x = O~2:X = (::); (3.21)
This operation is called cross-differentiation, and is true for exact

differentials.
Example 3.2
Show that (2xydx + x 2 dy) is an exact differential, whereas
(x 2 dx + 2xy dy) is not.
Applying the cross-differentiation test, equation (3.21),
o
oy (2xy) = 2x
o
ox (x 2 ) = 2x
and the test succeeds with the first expression whereas from the second
expression
o
_(x 2 ) =0
oy
o
ox (2xy) = 2y
and the test fails.

Exact differentials 83
Example 3.3
The volume v of a sample of gas depends on the temperature T and the
pressure p. Thus v = v(T, p) and
dv = (!;) dT + (:;)T dp.

p
(3.22)
If the volume is changed by dv at constant pressure, the work done on

the gas is - pdv. Substitution for dv from equation (3.22) gives
work = -pdv = _p(ov) dT_p(ov) dp.

or p op T
We can show that work is not an exact differential by applying the
cross-differentiation test, equation (3.21)
(:p[ -p(!;)J)T -(!;t-p O~2~,

=
(:r[ -p(:;)J)p O~2~. = -p
These would be identical only if (ov/OT)p were zero. This is not

physically possible because the coefficient of expansion (1/v) (ov/op)Tis
always finite and positive.
The geometrical significance of this is that for an exact differential a

unique surface exists when z is plotted against x and y. Ifwe move from
one point (Xl' YI)on the surface to another (X2, Y2) by changes in X and
y, the change in z is uniquely defined by the initial and final values of X
and y. Conversely, an inexact differential does not define a unique
surface and the change in z will depend on the particular path followed
from (Xl' YI) to (X2' Y2). A thermodynamic function of state is one that
can be expressed as an exact differential, so that changes in the function
are independent of path. The above example shows that work is not a
thermodynamic function of state, which is very significant.
Example 3.4
Cross-differentiation of an exact differential is used to obtain the
Maxwell relations in thermodynamics from the Gibbs equations.
Given the exact differential
dG = vdp-SdT,
cross-differentiation gives
When we have more than two independent variables, a straightfor-

ward extension of the above argument gives, for z = z(u, v, w, ... )
(&)
dz = -
au V,w,...
du + -
av
(&) U,w,...
(&)
dv + -
aw U,v, ...
dw +
and the cross-differentiation test for an exact differential

Adu+Bdv+Cdw+ ...
becomes
( ~~ ) u, w, ... = (~:) v, w, .. .
( :~ ) u, v, . .. = (~~) u, w, ... '
and so on, for any pair of terms.
3.6 Relations between partial derivatives

We can use the definition of a total differential given in Section 3.3 to
obtain useful relations between partial derivatives. If z = z (x, y) then
dz = (!:)y (:;)x dx+ dy. (3.23)
At constant z, dz = 0, so that
(!:)y dx = -(:;)x dy, at constant z.
Dividing by dy at constant z
(3.24)
which is the chain rule for partial derivatives mentioned in Section 3.1.
Example 3.5
Equation (3.24) can be used to obtain purely mathematical relations
between experimental properties that are defined as partial derivatives.
Relations between partial derivatives 85
The coefficient of expansion IX = (l/v) (av/OT)p, isothermal compress-
ibility ,,= - (l/v) (av/ap)T and thermal pressure coefficient y =
(op/OT)vare related by
ICy = _!(av) (a p ) = !(av) =
V ap T or v v or p
IX,
so that any experimental measurements of these three quantities which

did not obey this relation would certainly be wrong. The relation is
useful also in that anyone quantity can be obtained by measuring the
other two.
In the same way, the heat capacity at constant pressure
C p = (aH/aT)p, isenthalpic Joul~Thomson coefficient p = (aT/ap)H
and isothermal Joul~Thomson coefficient ¢ = (aH/ap)T are related
by¢= -pCp.
If a relation exists between the variables x and y in equation (3.23),

only one variable is independent and we can write the total differential
of z with respect to either one of the variables. For example, dividing
equation (3.23) by dy we obtain
dz _
dy -
(az) dx + (az)
ax dy ay
y x·
(3.25)
The variables most often used in physical applications are p, v and T,

two of which are independent due to the existence of the relating
equation of state. If we choose to introduce a new variable, u say, into
our relation z = z (x, y), the new variable must be related to the existing
ones to be useful. Dividing equation (3.23) by dy at constant u we
obtain
(3.26)
This equation relates partial derivatives when the variable being held
constant is changed.
Example 3.6
The molar heat capacity at constant pressure Cp = (aH/OT)p, and that
at constant volume Cv = (aU /OT)v are related because H = U + pv, so
that
To develop this into a relation between C p and C v we need to relate
(oH/O/)vand (oH/O/)p. From equation (3.26)
(o;)v (o;)p +(~~)T(~)V

=
so that
The term (OH/OP)Tcan be interpreted using the Gibbs equation

dH = TdS+vdp
and dividing by dp at constant T to obtain
(~~)T =T(:!)T+ V.
Using the Maxwell relation of Example 3.4, we obtain
( OH) = v _T(OV) ,
op T or P
so that
Cp = Cv+T(::)v (;; )p.

This is as far as thermodynamic and mathematical relations can take us.
For a perfect gas, pv = RTso that (ov/O/)p = R/p and (op/iJT)v = R/v.
Hence (oH /oPh = 0 and
perfect gas.
3.7 Extensive and intensive variables; Euler's theorem

The terms extensive and intensive are used in a physical context to
distinguish between variables that are proportional to amount of
substance (extensive) and those independent of amount of substance
(intensive). Extensive variables include mass, volume, energy and
entropy, and intensive variables include pressure, temperature and
refractive index.
In mathematical terms this is a distinction between functions of first
degree, and those of zero degree, in the amount of substance. A
functionf(x) is of degree m in variable x if when x is multiplied by a
constant, say A, the value of the function is multiplied by Am,
f(AX) = Am f(x).
Extensive and intensive variables; Euler's theorem 87
A function of first degree is one that is proportional to the variable
f(AX) = Af(x) for m = 1, extensive.
An intensive property, on the other hand, is a zero-degree function
f(AX) = f(x) for m = 0, intensive.
In physical problems we usually have more than one variable to
consider, some of which will be extensive and some intensive. If an
extensive variable z(a, b, x, y) depends on extensive variables a, band
on intensive variables x, y, multiplying the amount of substance by A
means multiplying the extensive variables by A but keeping the
intensive variables constant. Then
Z(Aa, Ab, x, y) = Az(a, b, x, y).
Partial differentiation with respect to Aat constant a, b, x and y gives
OZ d(Aa) OZ d(Ab)
o(Aa) ~+ o(Ab) ~ = z(a, b, x, y)
oz OZ
a o(Aa) + b o(Ab) = z(a, b, x, y).
This is true for any A, so put A = 1 to give

OZ OZ
a oa +b ob = z. (3.27)
In general, a functionf(xl, x 2 , x 3 , ••• ) is homogeneous of degree m

if
(3.28)
and Euler's theorem is
of of of
Xl ~+X2~+X3~+ ... = mf (3.29)
uX I uX2 uX 3
In the above example the function is first-degree in the extensive
variables but zero-degree in the intensive ones, so that only the former
appear in equation (3.27)
When we have an equation in differential form for a function of state
in terms of both extensive and intensive variables we may use Euler's
theorem in the form (3.27).
Example 3.7
For a mixture of substances 1 and 2 which is able to gain or lose
numbers of moles dn l and dn2 of substance, the total energy U is given
in differential form by
dU = TdS - pdv + J.ll dn l + J.l2dn2.
The variables S, v, n l and n2 are extensive. We write
U = U(S, v, nl' n2)
and Euler's theorem gives
au au au au
S as +vav+nlanl +n2an2 = U.
But from the given equation
au au au
as =T, av= -p, and -;--- - .-2,
II
un2
so that
U =TS-pv+nlJ.ll +n2J.l2.
On the other hand, for the Gibbs function G we have
dG = v dp - S dT + J.ll dnl + J.l2 dn2
where nl and n2 are extensive but p and Tare intensive. Euler's theorem
therefore gives
terms in p and T not appearing.
By using Euler's theorem in this way we are, in effect, integrating

differential equations.
3.8 Taylor's theorem in partial derivatives

Taylor's theorem is derived for functions of a single independent
variable in Section 2.9.2. The treatment for more than one in-
dependent variable is analogous but with total derivatives being
replaced by partial derivatives with respect to each independent
variable. Thus if a function I(x, y) is known at point
(a, b) we assume that for the point (a + x, b + y) the increment in the
function may be written as a power series in the form
I(a+x, b+ y) -/(a, b) = alx+a2y+a3x2 + a4XY + asy2 + ....
We denote the partial derivative with respect to x, of/ox, by I" and
higher derivatives a2f/ax 2 by I"", a2f/axay by I", and so on. Then
I,,(a+x, b+ y) = al +2a3x+a4Y+ ...
Vectors 89
and putting x = y = 0 gives f,,(a, b) = at. Similarly fy(a, b) = a2.
Taking second derivatives and then putting x = y = 0 gives
a3 = f",,(a, b)/2 and a4 = f"y(a, b). In this way we obtain Taylor's
theorem in the form
f(a +x, b+ y) = f(a, b)+ xf,,+ yfy+ 2!1 (x f""+2xyf,,y+ y fyy) +
2 2
... ,
(3.30)
where the derivatives are evaluated at the point (a, b). This is used in
Section 7.5.
3.9 Vectors
A vector quantity is one that has magnitude, direction and sense, in
contrast to scalar quantities, which have magnitude only. Mass and
time are examples of scalar quantities, whilst displacement, velocity and
force are vectors. A vector quantity may be represented by an arrow of
length proportional to the magnitude of the quantity and in the
appropriate direction and sense; they are then added together by
joining the arrows head to tail as in Fig. 3.3, where displacements OP
and PQ are added to give OQ.
o~------~------~
A
Fig. 3.3
We denore a vector quantity by a letter in bold-face type, such as A,

and the (scalar) magnitude of Aisdenoted by A or IAI. ThenifOP, PQ
and OQ represent the vectors A, Band C we can write the vector
equations
A+B= C, A+B-C=O, (3.31)
so that - C is a vector with the same magnitude and direction as C but
opposite sense. When C is obtained by adding together A and B, it is
called the resultant of the two vectors. Conversely, we may resolve
vector C into its components, A and B, in the directions of A and B.
The magnitude s of vector s is a scalar, and may be operated upon like
any other scalar quantity. Thus we may multiply vector s by scalar a to
obtain a vector of magnitude as and in the same direction and sense as s.
We may also differentiate with respect to a scalar quantity; differenti-
ation of the scalar magnitude s of displacement s with respect to time t
gives the velocity vector v in the same direction as s and of magnitude
ds/dt. Arguments of this kind may be used to show that momentum,
acceleration and force are also vector quantities.
A useful analytical device is to define a set of vectors i, j and k in the
directions of rectangular coordinates in space and each of unit length.
In Fig. 3.4 the line OP represents vector A and gives scalar projections
Ax, A, and A., on the x, y- and z-axes. The product iAx is then a vector of
magnitude Ax in the direction of the x-axis, and this is the component
of A in the x direction. We may then add the components to obtain
A = iAx+jA,+kA z. (3.32)
x
Fig. 3.4
Addition or subtraction of vectors is then achieved by adding or

subtracting the components giving, for example,
(3.33)
When a body is free to rotate in space it does so about its centre of mass
Vectors 91
(Section 2.11). We then define three rectangular coordinates through
that centre of mass as origin, so that any arbitrary axis of rotation
makes angles (x, f3 and y with the X-, y- and z-axes. It may be shown that a
rotation through angle w about that arbitrary axis is then the same as
the sum of rotations wi, wm and wn about the three coordinate axes,
where I, m and n are the direction cosines (Section 1.7.3) of the axis of
rotation. It follows from this that if we represent angular rotation by an
arrow drawn in the axis of rotation, of length proportional to the angle
of rotation and in the conventional sense that the rotation is in the
direction of a right-hand screw, then the component rotations about
the axes obey Pythagoras's equation
(w1)2 + (wm)2 + (wn)2 = w2 W+ m2 + n2) = ro 2 •
and the arrows may be treated as any other vector quantity.

Using the same arguments about scalar manipulation as above, it
follows that angular velocity, angular momentum (the moment of
linear momentum) and torque (the moment of a force) may be
represented by vectors in the same way. There is, however, one
difference between the vectors used to represent these angular pro-
perties and the vector quantities previously considered, namely that the
vector is localized in the axis of rotation. Vectors based on displacement
of an extended body, on the other hand, are non-localized in the sense
that they may be freely moved parallel to themselves.
Example 3.8
When a gyroscope spins with angular velocity c.o about a horizontal axis
it has angular momentum 1c.o, represented by a vector located in the
axis of rotation as shown in Fig. 3.5. If it is supported at one end only,
mg
I 'I
,, '
,
~ __~x____+T~____~~/~
mg
_ - - \ - -1. mgxdf
~
Fig. 3.5
the weight mg of the gyroscope is balanced by upthrust mg at the
support; these forces generate a couple mgx represented by a vector in
the horizontal plane perpendicular to the axis of rotation. From
equation (2.70).
couple = rate of change of angular momentum,
so that in time interval dt angular momentum mgxdt is generated,
represented by a vector into the plane of the paper in Fig. 3.5 in the
direction perpendicular to the axis of rotation. The two angular
momentum vectors are drawn in the figure as when viewed vertically
downwards; the resultant, and so the axis of rotation of the gyroscope,
rotates in the horizontal plane in an anticlockwise direction from that
point of view. This rotation of the gyroscope about the vertical axis is
called precession, the direction of precession being the opposite of what
would occur if the gyroscope were a wheel running on a horizontal
table.
This also applies to the effect of a magnetic field on a spinning
magnetic dipole; the field imposes a couple on the spinning dipole,
which therefore precesses about the direction of the magnetic field.
Nuclei that behave in this way are detected by the absorption of
radiation of frequency equal to the precession frequency in a nuclear
magnetic resonance spectrometer, the precession frequency depending
on the magnitude of the magnetic field strength at the particular
nucleus.
3.9.1 Scalar and vector products

The product of two vectors, A and B, is defined in two ways, as the
scalar (or dot) product A. B and as the vector (or cross) product A x B.
The scalar product is defined as
A.B = ABcos(}, (3.34)
where () is the angle between the two vectors. This product is a scalar
quantity and is the magnitude of one vector multiplied by the
projection of the other upon it. The sign will be positive or negative
according to whether the angle () is acute or obtuse. When the two
vectors are parallel it is simply the product of their magnitUdes, and
when they are perpendicular to each other the scalar product is zero;
such perpendicular vectors are called orthogonal.
When the vectors are expressed in component form, the unit vectors
i, j and k are defined to be orthogonal so that i.i = j.j = k.k = 1 and
Vectors 93
i.j = j.k = k.i = 0 and the scalar product becomes
A.B = (iA" + jAy + kA.). (iB" + jBy + kB.)
= A"B" + AyBy + A.B.. (3.35)
The vector product or cross-product A x B is defined as a vector of
magnitude IA x BI given by
IA x BI = ABsinO (3.36)
in the direction perpendicular to the plane containing A and B and in
the sense that a right-hand screw brings one from A to B through an
angle of less than 1[. When the order of writing the product is reversed,
the direction of the vector reverses so that B x A = - (A x B).
The vector product represents the magnitude of one vector multi-
plied by the component of the other perpendicular to the first, which is
the area of the parallelogram based on the two vectors as in Fig. 3.6.
AxB
Fig. 3.6
This illustrates another useful principle, that plane surface area may be
represented by a vector of length proportional to the area and in the
direction normal to the surface. When plane surface area A is viewed
obliquely from the direction making an angle 0 to the normal to the
surface, the area seen is A cos O. If we define the line of sight by unit
vector r this may be. expressed in the form of a scalar product:
effective area = A.r = A cos O. (3.37)
The moment ofa vector about a point, such as that of force F about
o in Fig. 3.7, can be expressed as the vector product of F and the radius
o,
\
\ rsin 9
Fig. 3.7
vector r of any point A on the line of F, since

moment of force = F x r = Fr sin 0 (3.38)
in the direction perpendicular to both F and r and in the sense of the
right-hand screw convention.
An electric dipole consisting of charges ± e separated by distance r
has dipole moment p, which is a vector on the axis pointing from the
negative to the positive end and of magnitude p. = reo Electric field
strength E, being defined as the force exerted on unit positive charge, is
a vector directed from the positive to the negative of the two charged
plates that generate the field. When a dipole is subjected to a field,
the torque on the dipole (Section 2.11) and its potential energy are the
vector and scalar products respectively of the dipole moment and
the field strength. From equation (2.68) we have
torque = JJE sin 0 = JJ x E. (3.39)
The potential energy V will be a minimum when the dipole is aligned
with the field, which is when 0 = O. The increase in potential energy
when the dipole is rotated through an angle 0 as shown in Fig. 3.8 is the
work done against the forces ± Ee acting at the ends of the dipole. Each
end moves a distance x = (r - r cos 0)/2 against the direction of the
field so that, anticipating our discussion of integration,
dx = !rsinOdO,
V=2 Sa" Eedx = Eer Sa 8

sin 0 dO = - p.E cos 8,
so that
V= -p.E (3.40)
The condition that a scalar product of zero means that the vectors are
Vectors 95
E
Ee
..".
x:
+I----*+e
-e
Ee
Fig. 3.8
orthogonal (perpendicular to each other) may be used to explain the

general definition of orthogonal functions. We may extend equation
(3.35) by writing
(3.41)
where Ai is the component of A in the ith coordinate direction. If we

then allow the number of dimensions to tend to infinity we are
supposing A to be made up of an infinite number of components, or A
may be replaced by a continuous functionf(x). Under these conditions
we may replace the summation in (3.41) by integration over the
continuous variable, to give
f: f(x)g(x)dx =0 (3.42)
as the condition for the functionsf(x) and g(x) to be orthogonal in the

interval (a, b). This is used in wave mechanics when we replacef(x) by
the wavefunction .p, and g (x) by the complex conjugate .p * (Section
1.10) so as to obtain the condition for .p and .p * to be orthogonal.
3.9.2 Scalar and vector fields

A region of three-dimensional space in which a quantity varies from
point to point is called a field. If the quantity is scalar, such as density,
temperature, pressure or concentration, we have a scalar field and the
normal methods of calculus apply. A gravitational, electric or magnetic
field, on the other hand, is a vector field because it has direction and
sense, as well as magnitude, at any point.
When we differentiate a scalar quantity with respect to a coordinate
direction in space we obtain the gradient in that direction. We can
express the gradient of a scalar field as a vector by writing, for example,
ioq,/ox for the gradient of q, (x, y, z) in the x direction, where i is a unit
vectQr in that direction. In three dimensions we may write
d '" .oq, .oq, k0q,

gra 'I' = I ox + J oy + Tz· (3.43)
This may also be written in operator form; we define 'nabla' as the

operator
V = i~+ j~+ k~, (3.44)
ox oy oz
so that (3.43) becomes
V q, = gradq,. (3.45)
The significance of this is that by adding together the gradients in the

three coordinate directions as vectors we obtain the magnitude and
direction in which the rate of change in q, is greatest; grad tP is a vector
that is normal to the contour surfaces over which q, is constant. Thus if
tP is the (scalar) potential of mass m in the earth's gravitational field, and
z is the vertical axis, we have oq,/ox = oq,/oy = 0 and oq,/oz = mg, so
that grad cP = kmg, meaning that the maximum, in this case the only,
potential gradient is mg in the vertical direction. This notation becomes
useful in more complicated cases; the direction and strength of the
electric field at a point due to a surrounding distribution of charges in
space can be found as - grad", where", is the total potential due to the
superimposed charges. Gradients of temperature, concentration or
electric potential will produce a flow of heat, diffusion of material or an
electric current, so that this notation is useful in three-dimensional
problems of these kinds.
When we are concerned with vector quantities in three-dimensional
space, new possibilities arise. In a fluid (liquid or gas) a flow of material
can occur either in parallel streamlines or in the form of eddies or swirls.
If we consider the work done in moving from one point to another in
space, when this is independent of the path taken between the points we
say that we have a conservative field. This applies to mass in a
gravitational field if we neglect air resistance.
If the field exerts force F on a particle, the work done in a
displacement dr is the scalar product F. dr given by
work = F.dr = F"dx+F),dy+Fzdz. (3.46)
Vectors 97
This work will be independent of path and the field will be conservative
if (3.46) is an exact differential and the cross-differentiation identity
applies (Section 3.5), or if
and
oFz of,, (3.47)
oy = oz·
We may therefore express departure from the condition of the work
being independent of path as departure from (3.47). This is called the
curl of the vector field, defined by
_ I_
curI F - (OFz OF,,) _(OF" OFz) k(OFy
- - - +J - - - + - -OF,,)
oy oz oz ox ox -
oy .
(3.48)
The first bracket in (3.48) is the x component of the vector curl F, and it
depends on the changes in F in the (y, z) plane, and in fact measures
rotation of the field in that plane. The significance of curl F is that it
measures eddies or swirls in the field: if we move from one point to
another along a path that is opposed to the eddies we do more work
than if we move along a path that is in the same direction as the eddies.
Equation (3.48) contains the derivatives of the components of F in
perpendicular directions, such as ofz/ ox. The derivatives in the parallel
directions, such as of,,/ox, give the components of the increase in
magnitude of the vector. This is a scalar quantity called divergence,
defined by
d· F _ of,, of,, of.
IV - ox + oy + OZ ' (3.49)
and detects sources or sinks in the field; this would apply if heat were
being supplied at a point in a flowing fluid because thermal expansion
would then create increase in velocity and so a positive divergence of
the velocity field.
These concepts provide concise notation in problems involving
three-dimensional fields. The nabla operator (3.44) can be used for
vector fields by noting that it can be regarded as a vector, so that both
scalar and vector products of V with F can be written. It can be seen
from the above definitions that
V.F = divF and V x F = curiF. (3.50)
CHAPTER 4
Integration

Integration can be defined as the inverse of differentiation or as
the limit of a sum. If the differential of a functionf(x) isf' (x) dx, then
the integral off' (x) dx is f(x):
If' (x) dx = f(x) + C. (4.1)
This is called an indefinite integral because the addition of any constant

c to the integral f (x) will give the same differential f' (x) dx. Also, if we
multiply f(x) by a constant, K, thenf' (x) dx is multiplied by the same
constant; this means that constants can be taken outside the integral
sign:
I Kf'(x)dx = K If'(X)dX+C. (4.2)
When the integration is between particular values of x, say from
r
x = a to x = b, we have a definite integral and the integration constant
I
cancels out.
f' (x) dx = [f(X) = f(b) - f(a). (4.3)
This definition of integration as the inverse of differentiation gives the

rules for integrating simple functions, and these are discussed in the
following sections.
A definite integral can also be defined as the limit of a sum, which is
the reason for the use of the integral sign:
f f(x) dx = 6~~0 :t: f(x) bx, (4.4)
where 1: denotes the sum produced by adding successive increments bx

to x, starting at x = a and ending at x = b.
The proof of the equivalence ofthese two definitions requires careful
mathematical analysis, but a geometrical interpretation is shown in
Fig. 4.1. The area of the shaded rectangle is ybX and the sum of the areas
of such rectangles becomes the area under the curve as bx -+ o.
area abBA = f y dx = f f(x) dx. (4.5)
a x b
Fig. 4.1
Another interpretation is illustrated in Fig. 4.2. We can follow the

curve y = f(x) starting from the point A, which is (a, f(a», by making
small steps bx in x. A tangent to the curve has slope dy/dx, so that the
increment 0 y in y along the tangent for increment bx in x is
dy
by = dx bx. (4.6)
In the limit as bx -+ 0, a succession of such steps along tangents will
i 1/
follow the curve, and the sum of the steps between a and b becomes
dY dy
L
x=b b (b)
lim d bx = -d dx = dy = f(b) - f(a). (4.7)
6s-0 x=a X a X I (a)
100 Integration
y
y = f(x)
feb) ------------------
f(o)
a b x
Fig. 4.2
Thus
or
f(b) = f(a) +
i dx
abdY dx (4.8)
f(b) = f(a) + s: f' (x) dx. (4.9)
Example 4.1
If a quantity X depends on temperature T and pressure p, we write
X = X (T, p). The change in X from temperature Tl to temperature T2 at
constant pressure is
X(T2 ,p)-X(T1 ,p) = f:'(~)' dT.

If X is the enthalpy H then (oH/OT)p is the heat capacity Cp and
H(T2,P)-H(Tl'P) = r' CpdT.

JT 1
These interpretations of integration may be summarized in the form

that: if we begin by knowing the properties at a point (the values of Xo
and dy/dx), integration produces a curve; if we know a curve (y = f(x»,
integration produces an area; and if we know an area, integration will
produce a volume.
Standard methods of integration 101
It is assumed here that the function to be integrated is well-behaved in
the sense that it is continuous and single-valued. Continuity usually
applies in physical applications, but functions that are not single-valued
are often met; these are discussed in Section 5.1.
4.2 Standard methods of integration
4.2.1 Simple functions

The integral of a function is that quantity which, when differentiated,
produces the original function. Thus
f X"
X,,+l
dx = n + 1 + C, n + -1,
fdX = x+C,
f eX dx = eX + C,
feaxdx = ~eax+c,
f [j(x) + 9 (x)] dx = ff(X) dx + f 9 (x) dx.
4.2.2 Reciprocal functions

When n = - 1 in the first example above, the rule does not apply. This
particular case is evaluated by the following device:
put y = In x, then x = eY ,
dx
dy = eY = x.
Taking reciprocals
dy 1
dx x
Hence
102 Integration
4.2.3 Integration by parts
This applies to the integral of a product of two functions. From the
differential of a product we have
d (uv) = u dv + v duo
We can integrate term by term to give
f d(uv) = f u dv+ fv duo

But
f d(uv) = uv,
so that
f u dv = uv - f v duo
If we denote a functionf(x) by u and a derivative 9 (x) dx by dv this gives
ff(X)g(X) dx = f(x) f g(x) dx - ff' (x) ( f 9 (x) dx )dX.
It is useful also to express this in words, as 'the integral of a product

of two functions is the first function times the integral of the second,
minus the integral of (derivative of first function times integral of
second)'.
Example 4.2
This can be used to illustrate an important principle, that the factors of

the integrand must be chosen so that differentiation of one part and
integration of the other shall yield a simple integral. This may not be the
case if the parts are incorrectly chosen; we might wrongly regard the
above example as a product of terms taken in the opposite order, as
thus producing only a more complicated integral. If this happens, the
rule is to try again with the factors the other way round.
Example 4.3
Integration by parts can sometimes yield the same integral again, which
can then be evaluated by the solution of an algebraic equation.
Denoting the integral by the symbol I, we could have
1= f sin 2 0 dO = - f cos 0 dO
sin 0 cos 0 + 2
- sin 0 cos 0 + f (1 - sin 0) dO 2
- sin 0 cos 0 + f dO -I.

Hence
21 = 0 - sin 0 cos 0 + c,
1= ~ - ~ sin 0 cos 0 + c
2 2
i
so that, for example
"/2 7t
sin 2 0 dO = -.
o 4
When a definite integral is integrated by parts the application of the

limits is straightforward so long as we remember that the limits mean
r I -r
taking the value of the integral at the upper limit minus that at the lower
limit. Thus
u dv = [ uv v duo
Example 4.4
The integral of In x is obtained as the product of In x and unity:
fIn xdx = f(ln x).l dx
= x In x - f ~ x dx
=xlnx- fdX
= x In x -x +c.
104 Integration
4.2.4 Integration by substitution
This applies when the integral can be regarded as the product or ratio of
two parts, one part being the differential of the other part.
Example 4.5
I = fX~!X4
Since d (x 2 + 4) = 2x dx, the factor of 2 being immaterial, we substitute
U = x2+4, du = 2x dx.
Then
I ="21 fdU
-;- = 1ln U + C
= 11n(x2 +4)+C.
Example 4.6
1= fnxx dx.
This is the product of l/x and In x, where d In x = (l/x) dx. We

therefore substitute U = In x, du = dx/x,
I = f Udu = 1u 2= 1(In X)2 + C.

When substitution is used to evaluate a definite integral, the change
in variable means that the limits of integration must also be changed to
the values of the new variable at those limits. Thus if Example 4.5 were
between limits of 0 and 2 for x, the limits for u become the values of
u = x 2 + 4 at x = 0 and x = 2, or u = 4 and u = 8. This can, of course,
be avoided by returning to the original variable before taking the limits.
4.2.5 Expansion of algebraic functions
f
When
f(X)
1= g(x) dx,
wheref(x) is of higher degree (contains higher powers of x) than g(x),

the expression can be expanded by division of the denominator into the
numerator.
Example 4.7
I f f
= x:: 2 dx = (X - X22: 2 ) dx
f f
= x dx - X22: 2 dx
x2
= 2- ln (x 2 +2)+C.
4.2.6 Integration by partial fractions

Algebraic functions with denominators that can be factorized can be
expanded by resolution into partial fractions. This is the inverse of the
simple addition of fractions; since
1 1 2x+ 1
- - + - - = -:------:.,.......,-----,-,-
x+2 x-I (x+2)(x-l)
we can equally well resolve the right-hand side of this equality into the
left. We define parameters A and B by writing
2x+1 A B A(x+1)+B(x+2)
(x+2)(x-l) = x+2 + x-1 = (x+2)(x-l) ,
which will be true if
A(x+ 1)+ B(x+2) = 2x+ 1.
Since this must be true for any value of x, we can find A and B either by
suitable choice of x to eliminate one of the parameters, or by equating
constant terms and the coefficients of x. Thus
if x = 1, 3B = 2, and so B = 1,
if x = - 2, - 3A = - 3, and so A = 1.
The rules for obtaining partial fractions depend on the factors in the
denominator of the expression;
(a) for factor (x + a)
A
use
x+a
(b) for factor (x + a)2
A B
use - - + -:-------:-.,.
x+a (x + a)2
106 Integration
(c) for factor (ax 2 + bx + c)
Ax+B
use
ax 2 +bx+c
Example 4.8
I=f x 8dx
-4x
3
Put
8 8 ABC
--:=---=
x 3 - 4x x (x + 2) (x - 2)
=-+--+--
x x+2 x- 2
Then
A (x + 2) (x - 2) + Bx (x - 2) + Cx (x + 2) = 8.
Equating coefficients of x 2, of x and constant terms gives
A+B+C = 0, B=C and A = -2.
Hence
B = C = 1,
and so
1= -2 dX f --+
f -+ dx f -dx-
x x+2 x-2
X2
=In ( ~
-4) +c.
4.3 Standard forms of integral and numerical methods.
Although it is generally possible to differentiate a continuous
function, integration only gives a closed expression in particular cases.
Extensive tables are available giving the integral of many standard
functions; a particularly comprehensive set is included in the
Handbook of Chemistry and Physics, published at intervals by the
Chemical Rubber Publishing Company, USA. One important stand-
1
ard form is
00
.J 1t
e -",'x'd x·=--
o 21X
and this is discussed in Section 5.2.
For integrals which cannot be evaluated in closed analytical form,
geometrical and numerical methods are available at least to obtain an
Multiple integration 107
approximate solution. An integral can be interpreted as the area under
a curve, so that we can draw the curve and measure the area between
particular limits to evaluate the definite integral. This applies also when
we want to integrate a quantity which appears as a curve on a recorder
trace. The area can be found by counting squares or vertical strips, or by
cutting out the shape and weighing it. Numerical methods are based on
the same principle, and details are given in texts on numerical analysis.
The Euler-Maclaurin formula is particularly useful and this is
discussed in Section 4.6.
4.4 Multiple integration

Just as we may repeat the operation of differentiation to obtain higher-
order derivatives, so we may repeat integration to evaluate multiple
integrals. The notation used is to show the required number of integral
signs together with the variables of integration. Thus
f ff(X) dx dx or f dx f f(x) dx,
both of which mean integrate with respect to x and then integrate the
result again with respect to x. This is a double integral and the solution
will contain two integration constants.
Multiple integration is most useful when we have more than one
independent variable. If we write
ff(X,Y)dX
with two variables x and y in the integrand, it cannot be evaluated

without a known relation between x and y so as to be able to substitute
for y to obtain an integrand containing x only. An alternative,
analogous to partial differentiation, is to integrate with respect to one
rf rf
variable whilst regarding other variables as constants. We do this when
defining the multiple integral
f(x, y) dx dy or dy f(x, y) dx,
both of which mean integrate first with respect to x, keeping y constant,

and then with respect to y, keeping x constant. The limits c and d apply
to x and the limits a and b apply to y. Multiple integrals are evaluated
from the inside to the outside in the first form shown, or from right to
108 Integration
left in the second form. This order of integration does not, in fact,
matter so long as the limits of integration for one variable do not
contain the other variable.
Example 4.9
Jl1 Jo12 x 0dy= Jl1 "3 ydy

f3 [X3 YJ2
3 38
2
ydxdy= 1 3
=~ [~J: = 332 .
Notice that the limits of integration must be applied successively, not
simultaneously.
= [i:I 15'
Since the variable y occurs in the limits of the integration with respect
to x it is essential that the inner integral be evaluated first. If the integral
were written in reverse order we would have
y2 Jl
JoI' Jo11 x 2
ydydx = JoI' [X2 I' x 2
-2- 0dx = Jo 2 dx
4.5 Differentiation of integrals; Leibnitz's theorem

Differentiation is the inverse of integration so that the operations of
differentiation and integration, both with respect to the same variable,
f
will cancel.
d
dx f(x) dx = fdf(X) f
~ dx = df(x) = f(x).
If, however, we have a function of more than one independent variable

we may differentiate with respect to one variable and integrate with
The Euler-Maclaurin theorem 109
respect to another. Since the variables are independent, these oper-
ations are independent of each other, so that the order of differenti-
ation and integration may be reversed.
oy
a f f(x,y) dx = fof(X,y)
oy dx. (4.10)
This will have meaning when the integral is between limits, and
complications arise if the limits of integration contain the variable
of differentiation in which case Leibnitz's theorem is used, which
is
a ib(Y) ib(Y) of(x, y) ob oa
a
y a(y)
f(x,y)dx=
a(y)
aY dx+f(b'Y)a-f(a'Y)a·
Y Y
(4.11)
Example 4.11
Leibnitz's theorem can be used to obtain an alternative form for an
integral by differentiation with respect to a parameter other than the
variable of integration. Thus if we differentiate the integral
1
00
o
2 1
xe- ax dx = -
2a
with respect to a we obtain
- 1 00
0 x 3 e -ax dx
2
=
d 1 1
da 2a = - 2a 2 •
4.6 The Euler-Maclaurin theorem

When we are concerned with a function for which no simple closed
form of integral exists, we may be able to expand the function as a series
and then integrate term by term. This will apply so long as the function
and the series used both converge sufficiently rapidly in the interval
considered. The Euler-Maclaurin theorem can then be used, and is
derived here by the use of Taylor's theorem.
We wish to obtain a series expansion of the definite integral
f~' f(x)dx
so that the integral may then be evaluated. We first expandf(x) as a
110 Integration
Taylor series about the value f(a) at x = a in the form
(x a)2
f(x) = f(a) + (x - a) I' (a) + 2! I" (a) + ... , (4.12)
and then integrate to obtain

r
Jxo
l
f(x)dx = (Xl -xo)f(a)+
~-~-~-~
1 2! ° I' (a)
(Xl - a)3 - (xo - a)3 f" ( )

+ 3! a + ... ,
Ifwe now choose a to be the mid point of the interval (xo, Xl) the terms
containing derivatives of odd order will cancel; we put h = (Xl - x o), so
that (Xl - a) = h/2 and (xo - a) = - h/2, giving
to[Xl f(x) dx = hf(a) + 223!

h 3 hS
I" (a) + 2 5! f4 (a) + . . . .
4 (4.13)
We now use Taylor series again to expressf(a),f" (a),f 4 (a), ... in terms
of the values at the end points of the interval; we adopt the device of
taking the sum of the values of the function at the end points, but the
differences of the values of the derivatives. Again using (4.12), we have
f(a) = f(x) - (x - a)I' (a) (x ;t)2 I" (a) - ....
When we substitute x = Xo and x = Xl and add the resulting expreS"-

sions, the terms containing odd powers of (x - a) cancel since (Xl - a) =
- (xo - a), and we obtain ...
and this is used to eliminate f(a) from (4.13). We also write Taylor
expansions off' (X),!3 (x), .. .in the form
f' (x) = I' (a) + (x - a)f" (a) +

(x
;! a)2
f3 (a) + ... ,
f3 (x) = f3 (a) + (x - a)f4 (a) +

(x
;! a)2
fS (a) + ....
By taking the differences between the values of these expressions at the
limits, the constant terms and those containing even powers of (x - a)
will cancel to leave
!"(rx) = ~ (I' (xd -1'(Xo») - 2~:!f4(rx) - ... , (4.15)
4 3 3) 2
f (rx)=};1 ( f (xd-f (Xo) -2h2 3!f 6 (rx)- ... , (4.16)
and so on. Substitution of (4.14), (4.15) and (4.16) into (4.13) gives
f~If(X)dX = ~h (f(Xd+f(Xo»)
1 (2!1- 1
- 22 3! )h2 (f ' (xd -I'(xo) )
+214 (1
3! (1 1) - (1
2! -3! 1)) h
4! -5! 4(f 3(xd-f 3)
(xo)
-terms in (P(Xd-f 5 (xo»).etc. (4.17)

This formula can be used to evaluate the integral in terms of the values
of the functionf(x) and of the derivatives of odd order I' (x),!" (x), .. ,
all at the end points of the integration range x = Xo and x = Xl' The
advantage of this form of equation is that we can divide the range of
integration into small parts, which we show symbolically as
rxof(x) dx = rXI
+ r
X2
+ ... + rxo ,
Jxo Jxo JX 1 JX__ 1
and the right hand side of (4.17) then simplifies. The first term can be
expressed in terms of the sum of the values of f(x) at the n points; in
particular, if h = 1 this gives
~ ito (f(X;) +f(xi+ I») = Jo f(x;) -~ (f(Xo) +f(x,,) ).

and in the terms containing the derivatives, the intermediate terms
cancel to leave only the values at the ends of the whole range. Hence we
obtain the Euler-Maclaurin formula in the form
(4.18)
112 Integration
and the numerical coefficients of subsequent terms can be shown to be
1/30240, 1/1209600, ....
The two most common applications of this formula are to the
numerical evaluation of convergent, but intractable, integrals which do
not possess a solution in a closed form (numerical quadrature), and to
the replacement of the sum of a convergent but intractable series by an
integral (e.g. in statistical mechanics).
The formula as written in (4.21) contains the derivatives at both ends
of the range of integration; if this range is from zero to infinity and if the
values of the derivatives approach zero as x tends to infinity, the
formula simplifies to
roo 1 1 1
Jo
00
f(x)dx = i~of(xJ-2f(0)+ 12f'(0)-720f3(0)

1 5
+ 30240 f (0) - .... (4.22)
Example 4.12
Use the Euler-Maclaurin theorem to evaluate r:
e-«' x' dx when
a = 0.5, and compare the result with example 5.1.
We putf(x) = e-«2 x2, from which we obtainf(O) = l,J(oo) = O,and
all of the derivatives of odd order are zero at both limits. Hence by
1
(4.22),
00
~ e _«'n' --.
e _«'X2 d x = £..., 1
o "=0 2
The integral is shown in example 5.1 to have the value .j 'It/2a. In order
to obtain rapid convergence we must have a small enough value of IX.
Putting IX = 0.5 we obtain
fooo e-«2x2dx = ~: = 1.772454,
and direct summation of the series gives
L e-«2n2= 1+e-O.S+e-l+e-2.2S+e-4+
00
"=0
The sum of this series to eight terms is 2.272454, so that we have
agreement with the Euler-Maclaurin formula.
This expression arises in calculating the contribution from trans-
lation in a particular direction to the partition function in statistical
mechanics. We then require to evaluate the sum
L e -n'h'j8ma 2kT = L e -n2h2j8ma 2kT -1.

00 00
n=1 n=O
1
The Euler-Maclaurin formula gives
~ e -n2h2 j8ma 2kT -_

.t...
00
e
-n2h2j8ma2kT
+-1
n=O 0 2
_ (2nmkT)1/2 ~
- h2 a+ 2'
so that
~ _n 2h2j8ma 2kT = (2nmkT)1/2 _~

.t... e h2 a 2'
n=1
Since the value of the integral is usually a very large number, the
correction of - 1/2 given by the Euler-Maclaurin formula may usually
be neglected and we say that the sum may be replaced by the integral for
the purposes of evaluation.
Example 4.13
Following the previous example, the relation which corresponds to the
rotational contribution to the partition function is
L
00
(21 + 1)e -J(J+l)8/~
J=O
(where () = h 2 /8n 2 Ik). This is also an intractable series, whereas the
corresponding integral can be evaluated in closed form.
When 1(1) = (21 + 1)e- J (J+l)8/T the terms up to «()/T)2 are
1(0) = 1, 1'(0) = 2 -()/T,
f3(0) = -12()/T+12«()/T)2- ... ,
f5(0) = 120 «()/T)2 - ... ,
and all relevant terms become zero as 1 -+ 00. The integral may
be evaluated by making the substitution u = 1(J + 1W/T to
give
1 00
o (21 + l)e- J (J+l)8/TdJ =
T
0'
114 Integration
The Euler-Maclaurin formula (4.22) then gives, to terms in (8jT)2,
f: (2J + l)e-
J=O
J (J+1)6/T =!.- +! - ~(2 -~)
8 2 12 T
+_1_ (-12 ~+ 12 (~)2)

720 T T
- 30~40 ( 120 (~ y)
+ ...
T(
=fi
18 1(8)2
1+ 3Y +15
4(())3 )
Y +315 Y + ....
In this case, the sum may be replaced by the integral when T;}> (). This
applies except for small molecules at low temperatures, for which the
series expansion can be used.
CHAPTER 5
Applications of integration
5.1 Plane area

As shown in Section 4.1, the integral
ff(X)dX
gives the area between the curve y = f(x), the x-axis and the ordinates
x = a and x = b.
If, instead, we required the area between the curve and the y-axis, we
sum horizontal strips of length x and thickness dy to obtain
if
f(a)
(b)
xdy,
wheref(a) andf(b) are the values of y =f(x) at x = a and at x = b

respectively.
The sum of these two areas must be the difference between the areas
of the rectangles a xf(a) and b xf(b), as can be seen by reference to
if
Fig. 4.2. This can be proved analytically as follows:
i b
f(x)dx+
(b)
xdy
f.b
a
ib Xdxdy dx
f(a)
= a f(x)dx+ a
= f [j(x) + x!'(x)] dx
116 _Applications of integration
= r:: d(xf(x»
X-b
[
= xf(x) ] x=a
= bf(b)-af(a).
The various manipulations used here should be carefully noted.
These relations give the area between a curve and one of the axes.
The area between two curves can be found as the difference between
the area under one curve and the area under the other.
The conditions necessary for this interpretation of integration to be
valid are that the function should be continuous and single-vaiued; this
means that for any given value of x, the function y = f(x) shall have a
single, unique, value. This condition will not be met if the equation
contains quadratic or higher-degree terms in y. Thus the equation
x 2 + y2 = r2
represents a circle ofradius r, centred at the origin, as shown in Fig. 5.1.

y
Fig. 5.1
If we simply integrated
f: ydx = f: ± ..)(r2 -x 2)dx
the result would be zero, as the sum of equal positive and negative areas;
this is because area has the signs shown in the quadrants of Fig. 5.1.
Plane elements of area 117
When a function is not single-valued, we must restrict the range of
integration, the area in the first positive quadrant of our circle being
obtained by choosing the positive root:
area of quadrant = IJ (r2 - x 2) dx.
This integral may be evaluated by using the substitution x = r cos 0,

dx = - r sin 0 dO, so as to take advantage of the relation
1 - cos 2 0 = sin 2 0:
area of quadrant = r
J"/2
rsinO(-rsinOdO)
= _r2 r
J"/2
sin 2 0dO
= 1[r 2 /4
by using the result of Example 4.3. The area of a circle drawn entirely
within the positive quadrant is then 1[r2.
5.2 Plane elements of area

In Section 4.1, the use of integration to obtain the area between a curve
and the x-axis as
area = [YdX (5.1)
is justified as the sum of areas of vertical strips of height Y and width dx.
In the same section, integration is interpreted as giving an increment in
Y by moving along the curve y = f(x), so that
y = s: dy. (5.2)
rf:
If we now combine equations (5.1) and (5.2) we obtain the area as a
double integral
area = dy dx. (5.3)
The geometrical interpretation of equation (5.3) is that we first obtain

the ordinate y by the inner integral and then sum the areas of vertical
strips by the outer integral. Alternatively, we can interpret the double
integral by regarding dy dx as a differential element of area, being that
118 Applications of integration
of a rectangle of sides dy and dx. Then
dA = dydx
and
area = fdA = ff dYdX .
In the same way, we can obtain the area of a circle, using plane polar
coordinates, by summing the areas of annular strips of radius rand
thickness dr as shown in Fig. 5.2. Given that the circumference of a
circle is 2nr the area of an annular strip becomes, as dr --+ 0,
dA = 2nrdr.
dA=271"rdr
Fig. 5.2
The area of a circle of radius a then becomes
fdA = f: 2nrdr = na 2•
The differential element of area in plane polar coordinates, shown in

Fig. 5.3, is obtained by increasing r by dr and () by d(}. This requires the
element of arc length ds = r de, which follows from the definition of
angle in polar coordinates as angle = (arc length}jradius.
Then
dA = rd(}dr
and
area of a circle = Jor Jor

a 21t
r d() dr
= 2n f: r dr = na 2 •
Plane elements of area 119
dA=rdOdr
Fig. 5.3
By using polar coordinates we have avoided the difficulty met in

Section 5.1 where the area of only a quadrant of the circle could be
found. This is because the polar equation of a circle centred at the origin
is simply r = a, which is a single-valued function.
Example 5.1
An integral that arises frequently in physical applications has the form
1=
1
000 e -IX
2
x
2
dx. (5.4)
This may be evaluated by the following device. As a definite integral (i.e.

between limits), the variable x is immaterial, so we may rewrite the
integral in terms of variable y as
1 =
1
000 e -IX
2
Y
2
dy.
We now multiply the two integrals (5.4) and (5.5) together. Since they
(5.5)
have common limits, this may be written as
This double integral is over the positive quadrant of the (x, y) plane. We
now transform to polar coordinates, using x 2 + y2 = r2; the element of
area dx dy becoming r dO dr. The limits for the same positive quadrant
become 0 to n/2 in () and 0 to 00 in r, so that
[2 = foOO f"/2 e _",2,2 r d() dr = foo -n e _",2,2 r dr.

0 0 2
This has the effect of introducing a factor r into the integral, which can
now be evaluated by the substitution
Hence
f
so that
[=
OO
e -'"
22
x dx = --.
.In (5.6)
o 20:
Example 5.2
The previous example can be used in the integration of similar forms,
such as
Integrating by parts, choosing x as the first part, gives
where
[=1
- fOO e-'" 2x 2dx
2 20: 2 0 .
The first term in [1 is zero when x = 0, but when x = 00 we need to use

L'Hopital's rule (Section 2.10)
· x
IIm---=== I·1m 1 2 2 =0'
X _00 e'" x x _ 00 20: 2 xe'" x
so that
5.3 Elements of volume; polar coordinates in three dimensions

In rectangular coordinates, the differential element of volume is that of
Elements of volume; polar coordinates in three dimensions 121
a cube of sides dx, dy and dz, so that
volume = f dV = fffdXdYdZ. (5.7)
Polar coordinates in three dimensions may be defined in terms of

either one angle and two lengths (cylindrical polar coordinates, Fig. 5.4)
or two angles and one length (spherical polar coordinates, Fig. 5.5).
P(r,O,z)
z
y
Fig. 5.4 Fig. 5.5
Cylindrical polar coordinates Spherical polar coordinates

x = r cosO x = r sin 0 cos cfJ
y = r sin 0 y = r sin 0 sin cfJ
Z=Z z = r cos 0
dV = rdOdrdz d V = r2 sin 0 dO dcfJ dr
The corresponding elements of volume are constructed by increasing
r by dr, 0 by dO and cfJ by dcfJ. When 0 is increased by dO the point Pat
radius vector r moves in an arc of length rdO. The volume element in Fig.
5.4 is therefore r dO dr dz. When cfJ is increased by dcfJ in Fig. 5.5, the
radius to be considered is r sin 0, producing an arc of length r sin 0 dcfJ.
The volume element in Fig. 5.5 is therefore r sin edcfJ r dO dr =
r2 sin 0 de dcfJ dr. In each case volume will be positive above the (x, y)
plane and negative below.
Example 5.3
The volume of a cylinder of radius a and height h with base centred at
the origin is
Example 5.4
The volume V of a sphere of radius a centred at the origin must be
found from that of the positive hemisphere, so that the limits of
integration are 0 to n/2 for e and 0 to 2n for tjJ:
1V = f: f: xf:1\2 sin e de dtjJ dr = f: f:x r2 dtjJ dr
= f: 2nr 2 dr = j nr3.
V = 4nr 3 /3.
There is a general analytical method for obtaining the relations for

changing the variables in integration when changing the coordinate
system. This makes use of lacobians and is described in texts on
calculus.
5.4 Line integrals

It is shown in Section 3.5 that, if z is a unique function of x and y, the
expression dz = X dx + Y dy satisfies the cross-differentiation test and
is an exact differential. The geometrical interpretation of this is that we
can then plot a three-dimensional graph of z against x and yand obtain
a unique surface. A physical application of this is when z is a
thermodynamic function of the state defined by x and y. Change in z
then depends only on the initial and final states, being independent of
path.
We may write the change in z as the integral of dz between the
endpoints. In two dimensions, with one independent variable, this gives
equation (4.7). In three dimensions we obtain
f dZ = f(X dx + Ydy). (5.8)

Curve length by integration 123
The right-hand side of (5.8) contains both the variables x and y, so can
be regarded as an integral in the (x, y) plane. Between given endpoints
the left-hand side gives the change in z and the right-hand side is called a
line integral. Ifwe suppose that we move from one value of z to another
across the surface in three dimensions along a particular path we can
project this path into the (x, y) plane and so define a curve. The line
integral is then evaluated along that curve in the (x, y) plane.
To evaluate an integral containing two variables x and y we usually
need to know an equation connecting them so as to be able to express
the integral in terms of a single variable. This corresponds to choosing a
particular path. However, if X dx + Y dy is an exact differential it can be
written as single derivative dz and so integrated. The line integral is then
independent of path, the change in z depending only on the endpoints.
This also means that the line integral of an exact differential over a
closed path is zero because the endpoints then coincide. Thus we may
use the condition that the line integral is independent of path as an
alternative definition or test for a thermodynamic function of state.
A line integral can be defined and used even when it does not contain
an exact differential, and it need not contain both terms on the right-
hand side of (5.8). The general characteristic of a line integral is that it
contains two variables and so, if not an exact differential, can be
evaluated only for a known equation connecting the variables, which
means along a particular curve. An example is when the variable of
integration is the distance along a curve, which clearly depends on the
path taken between the endpoints, and this is discussed in the next
section.
S.S Curve length by integration

When a point moves along a curve, we denote distance travelled by s.
Then for increments dx in x and dy in y we can apply Pythagoras's
theorem to obtain
ds 2 = dx 2 + dy2.
The length of curve between points (Xl' yd and (X2' Y2) is then
s = i2 ds
where
ds = ..) (dx 2 + dy2)
= ..)[1 + (dy /dx)2] dx,
1
so that
s=
2
ds=
iX, J[1+(dy/dx)2]dx.
1 x,
This integral cannot be evaluated until we know dy/dx in terms of x

only, for which we need to know the equation connecting x and y.
Example 5.5
The circumference of a circle of radius r centred at the origin is found
from that of the positive quadrant. The equation of the circle is
x 2 + y2 = r2
2xdx+2ydy = 0
dy x x
dx y
The length of arc is therefore
X2
Jo
S=f"'(1+ 2
r -x2 )dX=f"'J(:
Jo r-x 2)dx.
t
Putx=rcos(),dx= -rsin()d(),
s= £J[1~(x/r)2]dX= [12 -~~i;()d()=r I2d ()=i r .
Hence the circumference of a complete circle is

s = 2nr.
5.6 Applications of multiple integration

Multiple integration is discussed from the analytical point of view in
Section 4.4 and in geometrical terms in Sections 5.2 and 5.3. It is
generally useful in physical applications, in terms of physical variables
of various kinds. This is because differentials can be combined to great
advantage in many applications, and finite solutions are then obtained
by multiple integration. This is illustrated in the following.
5.6.1 The pressure of a perfect gas

This is given by the kinetic theory of gases in the form of the well known
equation
pV=tNmu 2 •
Applications of multiple integration 125
It is derived in standard texts in ways that avoid mathematical
complexities such as multiple integration, but this is at the cost of
physical clarity. An alternative derivation is presented here in a form
which is extended in the next section to include the properties of real
fluids.
Since pressure is force per unit area, and force is rate of change of
momentum, we can calculate the pressure exerted by collision of gas
molecules on unit area of wall from the rate of change of the
component of momentum perpendicular to the wall.
In a perfect gas, molecules of mass m are assumed to move freely and
independently of each other. Any particular molecule will be in a
random position and moving in a random direction with a particular
velocity.
Random position and random direction lead to simple calculation.
For N molecules in volume V, the density is N / V and the number of
molecules in an element of volume d V is the density times that volume,
or (N / V)d V. Random direction in space can be dealt with using solid
angle, this being defined as area/(radius)2 (Section 1.1), and the total
solid angle for a sphere is 47t. The fraction of molecules that will be
moving towards an element of surface du on a sphere of radius r is
du/47tr2.
The speed of a particular molecule is a more complicated problem
because some molecules move quickly, some slowly, and there is a most
probable speed at a given temperature. We have a continuous
distribution of velocity amongst the molecules and this is expressed as a
velocity distribution v(u); the fraction of molecules with speed between
u and u + du depends upon u and is written as v( u) duo
We multiply together the above three factors to obtain the number of
molecules in a volume element d V which are moving towards an
element of area du on a sphere of radius r with a speed between
u and u+du as
N du
dN = V dV 47tr2 v(u)du. (5.9)
In Fig. 5.6 we consider molecules moving from an element ofvolume

at P in the gas to collide with an element of area dA on the surface of a
wall. Using spherical polar coordinates, with angle cjJ in the plane
perpendicular to the paper (compare Fig. 5.5), the element of area
dA can be resolved into an element perpendicular to the radius vector
r as
du = dA sin () sin cjJ. (5.10)
P(r,lJ,¢»
Fig. 5.6
The volume element at P is

d V = r2 sin 0 dO dl/> dr. (5.11)
The pressure of the gas is the force exerted per unit area on the wall
by elastic collisions of the molecules on the wall; it is the rate of change
of momentum perpendicular to the wall on unit area. When a molecule
from P collides elastically with the wall, its component of momentum
perpendicular to the wall becomes reversed in sign. The component of
velocity u perpendicular to the wall is u sin 0 sin l/>, and the correspond-
ing change in momentum is 2mu sin 0 sin l/>.
The factor that remains to be considered is the number of collisions
in unit time; a molecule from P will reach the wall in unit time if the
distance r is less than or equal to u.
The total pressure is now found by integration. The differential
expression for the pressure, dp, is obtained by mUltiplying the number
of collisions, from equations (5.9), (5.10) and (5.11), by the change in
momentum per collision:
N dA sin 0 sin l/>
dp =- r2 sin 0 dO dl/> dr 4 2 v(u)du 2mu sin 0 sin l/>
V nr
= N 2m sin30 sin 2 l/> uv(u) dO dl/> dr dA duo

V n
The differential variables show this to be a five-fold integral, and we
Applications of multiple integration 127
now consider the ranges of integration. We obtain unit area by
integrating A from 0 to 1, and unit time by integrating r from 0 to u.
Integration of e and ¢ from 0 to n/2 ensures that only single values of
sin () and of sin ¢ are used and we introduce a factor of 4 so that point P
moves through a complete hemisphere based on the wall. The velocity u
is integrated from 0 to 00. Using the notation of Section 4A we write
p = 4 N 2m (00 uv(u) du (,,/2 sin3 ed() ("/2 sin2 ¢ d¢ (u dr (1 dA.

V n Jo Jo Jo Jo Jo
It is important here to evaluate the integrals from right to left because
the variable u appears in the limits of the integration with respect to r.
f:
We have
dA = 1,
f: dr = u,
(,,/2
Jo sin 2 ¢ d¢ = n/4,
("/2
Jo sin 3 () d() = 2/3,
so that
p = --m
1N
3 V
10
00
u2 v(u)du.
The remaining integral is the average value of the square of the

velocity, which we write as u 2 , so that
pV=jNmu 2 •
5.6.2 Interactions in a rea/fluid

In a real fluid the molecules exert attractive and repulsive forces on each
other and, except at high temperature, the attractive forces are
dominant, so that the pressure exerted on the walls is less than it would
be for a perfect gas. We denote by w(r) the potential energy of
interaction of a pair of molecules separated by a distance r in the bulk
fluid; the attractive force is then -dw(r)/dr. The number of pairs of
molecules at separation r depends upon this potential and we denote by
g(r) the pair distribution function, which is the ratio of the number of
pairs at separation r to the random number.
Fig. 5.7 shows a plane Oz drawn within bulk fluid. Molecules on the
left-hand side interact with those on the right-hand side; by restricting
the interaction to a window of unit area we can calculate the total force
per unit area, perpendicular to the plane, exerted by the molecules on
one side on the molecules on the other.
z
Fig. 5.7
We use point P and unit area at 0 to define an element of volume on

the right-hand side within which molecules would be visible from P. An
element of area dA at 0 gives a component dA sin e sin 4> perpendicular
to the radius vector r. We define fraction oc of the radius vector as being
on the left-hand side of the plane, so that the solid angle subtended at P
becomes dA sin () sin 4>1(ocr)2. This gives an area visible at radius r of
r 2dA sin e sin 4>1(oc 2r2). The volume element d V R in the range r to
r + dr on the right-hand side is then
dVR = dA sine sin 4> dr/oc 2 • (5.12)
Point P is defined by spherical polar coordinates (-ocr, e, 4» and
we construct a volume element d VL at P by using oc as the variable at
constant r. Then
d V L= (ocr)2 sin ede d4> d(ocr)
= oc 2r3 sin ede d4> doc. (5.13)
The number of pairs of molecules at this separation r is g(r) times the
random number; the random number of molecules in d V R is (N 1V) d V R
and in d V Lis (N 1V) d V L. Since each molecule in d V Linteracts with all
The calculus of variations 129
molecules in d V R' the number of pairs is
(~y dVLdVRg(r).
The component of the attractive force - dw(r)/dr perpendicular to
the plane is - (dw(r)/dr) sin () sin ¢. The differential contribution to the
pressure from the intermolecular forces, dPinl' is then
dPint = dw(r). . (N)2. . . dr

- ~ sm () sm ¢ V (X2 r3 sm () d() d¢ d(X dA sm () sm ¢ (X2 g(r)
. 3 () . 2
= - ( N)2
V
3 dw(r)
sm sm ¢r g(r)~drd()d¢d(XdA.
We again have a five-fold integral and the limits are 0 to 1 in A for unit
area, 0 to 1 in (X to include all pairs for a given direction of the radius
vector r, 0 to n/2 in () and ¢, with a compensating factor of 4 to cover a
complete hemisphere on each side of the plane, and 0 to co in r:
Pint =
-4 ( V
N)2 l<Xl
0
dw(r)
r3g(r)~dr Jo("/2 sin 3 ()d() Jo("/2 sin2¢d¢ fl
0 d(X JoedA,
=
N)2
-1n ( V Jo(<Xl r3g(r)~dr,
dw(r)
which may also be written as
Pint = -61 (N)2

V Jo('" g(r)r~4nr2
dw(r)
dr.
This is an important equation in the theory of liquids anO gases.
5.7 The calculus of variations

The determination of stationary (maximum and minimum) values by
differentiation ofa known function is described in Section 2.7, and this
is extended to simultaneous equations in Section 2.8. Another class of
problems involving stationary conditions is when the function itself is
unknown. For example, we can show that the shortest distance between
two points in a plane is a straight line, and that between two points on a
sphere is by the great circle route, by finding the function which
describes the path of minimum length. We have seen in Section 5.6 that
curve length in two dimensions is given by the integral
s= rx, .J [1 + (dy/dx)2] dx. (5.14)

JX1
Here, as in other problems of this kind, we have a known integral which
depends upon an unknown function y = 4> (x). The integrand is a
known function of the unknown function y, and can be written in
general terms as
1= ff(X, y, y')dx (5.15)
where, again,fis known but y is not. We have introduced in (5.15) the

device of regardingf as dependent not just on the two variables x and y
but also on the third superfluous variable y' = dy/dx. As shown in
Section 3.4, we may still use the expression for the total differential in
the presence of such a superfluous variable, which we shall later do.
The problem now takes the form of finding the function y = 4>(x)
which makes the integral I stationary. For this we need to differentiate I
with respect to some parameter that produces variation in the function
y and so in the integral. This is achieved by the device of writing the
unknown function as the sum of two terms
y = 4>(x) + 8'1(X), (5.16)
where '1(x) is yet another function defined as arbitrary but for the
condition that '1(x) = 0 at the endpoints x = a and x = b, and 8 is an
adjustable parameter. If we now vary 8 for given 4>(x) and '1(x) we
produce a variation in the curve of y against x between fixed endpoints,
as in Fig. 5.8. When 8 = 0 we obtain the simple form y = 4>(x). Change
in 8 produces a change in y at constant x for given 4>(x) and '1(x).
The integrand in (5.15) is then a function of 8 and we can use the
stationary condition
01
08 = o.
As shown in the previous section we may interchange the order of
i
differentiation and integration so that
01 = b
of (x, y, y') dx = O. (5.17)
08 a 08
Now change in 8 at constant x, 4>(x) and '1(x) produces changes in y and
The calculus of variations 131
y
'. ,,
,
-"1= cp(x}+ £l'}(x)
...
y =cp(x) -'"
"', ...
,.
x
Fig. 5.8
in y' so that
oj ojdy oj dy'
-=--+--
oe oy de oy' de
and from (5.16)
dy dy' d dy d dy d'1 ,
de = '1(x), - = - - = - - = - = '1
de dedx dxde dx '
so that (5.17) becomes
(b( '1 oyoj + '1, oy'OJ) dx

Ja = O. (5.18)
Integration of the second term in (5.18) by parts gives
[OJ Jb J.b d oj
J.•b'1 ,oj
a'dx =
y
J.b 81'1
aY
oj ,
dx = 81'1 -
Ya
da''1 dx .
aXy
(5.19)
By definition, '1(x) = 0 at x = a and at x = b so that the integrated part

is zero and (5.18) becomes
(5.20)
Now equation (5.20) contains '1(x) which may be any function, so that
the only way in which the integral must be zero is when the bracketed
part is zero. Hence
af _~ af = 0 (5.21)
ay dxay' ,
which is called Euler's equation.
Example 5.6
Show that a straight line is the shortest distance between two points in a
plane.
Whenf(x, y, y') is defined by (5.14) we have
f (x, y, y') = y' (1 + y'2)
and since we are regarding y and y' as separate variables
af = o.
ay ,
and substitution into (5.21) gives

d y'
dx y'(1 + y'2) = 0,
or y' is independent of x. This is the equation of a curve of constant
slope, which is a straight line.
The application of this technique to three (or more) variables

requires only a straightforward extension of the argument. In place of
(5.15) an additional variable z will introduce also z' as
f = f(x, y, y', z, z').
We define two arbitrary functions in place of the one function '1 (x) and
two parameters in place of the single parameter e, one set for y and one
set for z. Equation (5.17) becomes the sum of two sets of terms and
(5.20) likewise. The condition for the integral in (5.20) to be zero is then
that we have an Euler equation of the form of (5.21) in each of the
variables x and y.
5.8 Generalized dynamics

The discussion of dynamics in Section 2.11 was based upon Newton's
laws of motion. This is not the only basis, however, on which the theory
can be built, a useful alternative formulation being due to Hamilton
and Lagrange. This regards kinetic and potential energy, rather than
Generalized dynamics 133
force, as fundamental and postulates that a system will move in such a
way that the difference between the kinetic energy T and the potential
energy V, integrated with respect to time, is stationary. This difference
is called the Lagrangian function L and the postulate is called
Hamilton's principle, so that
1= it2
t,
Ldt = it2 (T- V)dt
t,
(5.22)
is to be stationary, and we can apply the methods of the previous

section.
An abstract formulation of this kind is of course more difficult to
visualize than Newton's laws because it is not so clearly related to
everyday experience. Its great advantage is generality, which arises
because it is expressed in terms of the kinetic and potential energies,
both of which are scalar quantities. This in turn, means that we are not
confined to any particular system of coordinates because we need only
to express position in space, not direction as well, which means that we
can easily change coordinate systems.
Hamilton's principle is related to the little used concept of action,
which is the change in the product of momentum and length. If a field,
such as gravity, produces a force F on a particle in the x direction, it
moves so that V decreases and T increases by the same amount. The
difference in (T - V) is twice the change in potential energy, or twice the
work done by the applied force. Denoting momentum by p we have that
work = fFdX and p = f Fdt ,

so that
f L dt = 2 f f F dx dt = 2 ff F dt dx = 2 f p dx,
J
and p dx is action. An early formulation of quantum theory was in this
form, the Wilson and Sommerfeld rule being that p dx = nh over aJ
closed cycle of the motion. It was shown in ExaPlple 1.2 that such a
relation is dimensionally correct.
We can use the calculus of variations from the previous section to
show that the two formulations of the principles of mechanics are the
same. For motion of a particle of mass m in the x direction, the kinetic
energy T = ! mi 2 • When force F acts on the particle in the same x
direction the potential energy will change, so that we write that
potential energy as V(x). The Lagrangian function L then depends on x
(through V) and on x (through n and we have
L(t, x, x) = !mx 2 - V(x). (5.23)
Euler's equation (5.21) then becomes
aL _~ aL = 0 (5.24)
ax dt ax '
and from (5.23) we have
aL
-=
av.
--=F
ax ax '
which is the force acting on the particle, and
aL . d aL .
ax = mx, so that --=mx
dt ax '
and we obtain F = mx, which is Newton's law. Thus Hamilton's
principle (5.22) is the same as Newton's law.
The rotation of a diatomic molecule was considered on the basis of
Newton's laws in Section 2.11, where it was assumed that masses m1
and m 2 were rigidly held at a separation r from each other. By using the
more powerful technique of equation (5.24) we can consider not only
that case but the more general one, where the masses are attracted, or
repelled, by each other with a force that is a function of the separation r.
The potential energy may be written in the general form V(r); this will
apply either to a chemically bonded molecule or to the general case of
the interaction between two particles in three-dimensional space.
We use the polar coordinates shown in Fig. 5.9 to define the positions
of masses m1 and m2 at distances r 1 and r2 from the centre of mass 0 of
the pair of particles. We suppose the two particles to be moving in
space, and so relative to each other, so that both distance r and angle ()
change with time. We define the reduced mass m* of the two particles
when they are at separation r as in (2.66) by
and the rotational kinetic energy from (2.64) and (2.67) becomes in
terms of the angular velocity {j = OJ
rotational KE = !m*r 2 82 • (5.25)
Since the distance r between the particles is now a variable, we also
have radial kinetic energy given by
(5.26)
Fig. 5.9
But, from (2.65)

and
and (5.26) becomes

radial KE = -!m*r2 • (5.27)
Hence
(5.28)
We now have L in terms of the coordinates r, rand e, 8, and we can
write Euler's equation (5.24) both in terms of the angular coordinate
oL _~~ = -~(m*r20) = o. (5.29)

oe dt oe dt
and in terms of the radial coordinate
oL d oL oV d
.2
- - - - = m*re ----(m*r) =
or dt or or dt
o. (5.30)
In equation (5.29), m*r 20 = 10 is the angular momentum; this is shown

to be constant. In equation (5.30), m*r8 2 is the centrifugal force in the
direction of r opposite to the attractive force between the particles
-oV/or, and m*f is the radial momentum. When the attractive and
centrifugal forces balance, d(m*r)jdl = 0 and the radial momentum is
constant. In the simplest case of circular motion the radial momentum
is zero and r is constant.
The form of the general motion represented by (5.29) and (5.30)
depends on the form of V(r). In the important case of an inverse square
law attraction, corresponding to gravitational attraction of planetary
bodies or the electrostatic interaction between charged particles, the
equations represent conic sections (Section 1.7.4). By considering the
polar equations of conics, or otherwise, it may be shown that when
particles having the same charge approach each other the path of one,
relative to the other as focus, is a hyperbola, and for unlike charges we
may obtain also a parabola, ellipse or circle for decreasing values of the
relative velocity of approach, as illustrated in Fig. 5.9.
This approach to dynamics is called 'generalized' because the
equations are the same in any coordinate system. When we denote a
generalized position coordinate by qi we can also define a correspond-
ing, or 'conjugate' 'generalized momentum' by
aL a(T- V) aT
Pi = aqi = aqi aqi
if V is independent of qi' (5.31)
The significance of this may be seen from the simple case when qi is an x
cartesian coordinate. The momentum is then Pi = mx and the kinetic
energy T is -! mi 2, so that
aT = ~(-21mi2) = mi = p.
aqi di "
and if qi were an angular coordinate 0 we would have T = tIlP, and
again
aT d 1 '2
de '
aqi = (2 10 ) = 10 = Pi'
The condition in (5.31) that V is independent of qi means that we
have a conservative system (Section 3.9.2), or that the work done in any
displacement is independent of the velocity (no 'friction').
In these generalized terms we can define also the Hamiltonian
function H which is often used in quantum mechanics. The formal
definition is
(5.32)
Again for the simple case of motion in the x direction, this gives
pi=mi,
and
so that
H = mx 2 -(!mx2 - V)
= !mx 2 + V
= T + V = total energy.
This means that the Hamiltonian function has the simple meaning of
being the total energy so long as the system is conservative.
We can express Lagrange's equation in terms of coordinate qi as
which with (5.31) gives
We can then write the equation in terms of H by examining the

parameters on which H depends; we have in equation (5.32) that H
depends on Pi> qi' and on L. In turn L depends on qi (for T) and on qi
(for V). Thus
so that for the ith coordinate

" dPi-:;-
d H = Pi d qi+qi oLd qi-~
aLd'qi'
uqi uqi
But the generalized momentum Pi is defined by equation (5.31) so that
the first and last terms cancel to give
dH = qidpi - ~L dqi'
uqi
Hence
oH .
~=qi' (5.33)
uPi
and
oH oL .
- = - - = -po (5.34)
Oqi Oqi ,.
Equations (5.32) and (5.33) are simpler than the equation in terms of the
Lagrangian function L in that they have a degree of symmetry; in most
problems, however, the Lagrangian function is found first and then the
Hamiltonian. The expression (5.33) gives the rate of change of
generalized momentum and so may be regarded as a generalized force.
CHAPTER 6
Differential equations

Differential equations arise frequently in mathematical and physical
theory because the calculus limit so often gives simple relations. In
physical theory we may have simple relations for the rates of change
with time, distance or temperature of such quantities as energy (thermal
conductivity, heat capacity), momentum (force, viscosity) or amount of
substance (fluid flow, reaction kinetics, heat of reaction). These theories
give rise to equations containing derivatives, and comparison with
experiment requires the solution of such equations by integration, the
purpose of which is to eliminate the differentials.
The methods of integration discussed in the previous chapter can be
used only when the integrand can be expressed in terms of a single
variable, so the solution of a differential equation requires the
separation of the variables into terms that can be individually
integrated. Problems involving only two variables will give rise to total
derivatives and ordinary differential equations, whereas if more than
two variables are relevant, partial derivatives and partial differential
equations will be obtained.
A trivial example of a differential equation is
dy
dx = 2x.
This may be written with the variables on separate sides of the

equation, as
dy = 2xdx.
We may then integrate each side of the equation, to obtain
fdY = f 2XdX
Y = X 2 +c.
The integration constant C is important in differential equations.
A simple physical example is when the rate of change of a quantity x
with respect to time t is proportional to the quantity itself, so that
dx
-= -kx
dt '
giving
fd: = - f kdt ,
In x = -kt+C.
The value of the integration constant can be determined by the use of
boundary conditions, such as knowing the initial value Xo of x at t = o.
Then by putting x = Xo and t = 0,
In Xo = C,
(:J
and
In = -kt,
An alternative interpretation of this form of relation is in terms of a
time constant; if an equilibrium system is disturbed by a small amount
Ax it will return to equilibrium at a rate which to a first approximation
may be assumed to be proportional to the displacement. Then
giving
Ax = Axoe- t /<,
so that after time T the displacement will have decreased to lie = 0.37
of its initial value. This is called the time constant, or relaxation time, of
the system and is seen to be the reciprocal of the first-order rate
constant k in reaction kinetics.
Not all differential equations possess an analytical solution, but
standard methods are available for particular classes of equation. This
classification is based upon the concepts of order, degree, linearity and
homogeneity.
140 Differential equations
The order of a differential equation is that of the highest derivative
occurring in it, dy/ dx being first order, d 2 Y/ dx 2 being second order, and
so on. The degree is the power of this derivative. A linear equation is
one containing only the first powers of a variable and its derivatives,
with no cross-terms, so that
d 2y dy
ao(x) dx 2 +adx) dx +a2(x)y =f(x) (6.1)
is a second-order linear equation.

The term 'homogeneous' is used in various ways. A first-order
equation is homogeneous if each term contains the variables to the
same power (Section 6.2.2). A linear equation, on the other hand, is
homogeneous if all terms contain the dependent variable, so that
f(x) = 0 in (6.1).
The solution of a first-order differential equation contains one
arbitrary constant, that of a second-order equation contains two
constants, and so on.
In the following sections we consider first-order first-degree equa-
tions, classed as separable variables, homogeneous, exact or linear, then
higher-order linear equations.
The subject of differential equations, and their range of application,
is very wide and provides scope for considerable ingenuity in the
methods of solution. This chapter does no more than introduce the
techniques and some typical applications.
6.2 Equations of first order, first degree
6.2.1 Separable variables

If a differential equation can be written with all terms containing one
variable on the left-hand side and all terms containing the other
variable on the right-hand side, the solution is merely integration.
Example 6.1
dy
dx +4x = 1.
Multiply both sides by dx, giving

dy+4xdx = dx,
which may be written as
dy = (1 - 4x)dx.
Equations of first order, first degree 141
Integrate both sides
fdY = f(1-4X)dX
Y = X_2X2+C
Y = x(1-2x)+C.
The geometrical interpretation of this solution is a family of curves,
each of which corresponds to a particular value of the integration
constant C. Suppose we require the solution passing through the point
(1,2), then we substitute x = 1 and y = 2 into the equation so as to
eliminate the integration constant, giving
2 = -I +C, C=3
so that the solution to the differential equation becomes
y = x(I-2x)+3.
Example 6.2
An equation may sometimes be separated into integrable terms even
when the variables are not completely separated. Thus if
dy 2x-y
dx x+1
the above procedure gives
(x + I)dy = (2x - y)dx,
xdy + dy = 2xdx - ydx.
The variables are not separable as such due to the terms xdy and ydx,
but the combination xdy + ydx is actually the differential of the
product xy, so that
dy + d(xy) = 2xdx
which may be integrated term by term to give
f dY + fd(XY) = f 2XdX ,
y+xy = x 2 +C,
so that
x 2 +C
y=--.
I+x
Example 6.3
A chemical rea{;tion is described as 'first order' when the rate of
production of product is proportional to the first power of the reactant
concentration. This is a different use of the term 'order' from the
mathematical definition. For the reaction
A ~ B
(a-x) x
the rate coefficient k is shown, together with the concentrations of the

two species at time t. Then
dx
dt = k(a -x).
This is a separable variables equation, so that
dx
-(a-x
- ) = kdt,
-In(a -x) = kt + C,
and the integration constant C is found from the boundary condition
that x = 0 at t = 0, giving C = -In a and
In(_a) = kt,
a-x
x = a(l _e- kt ).
6.2.2 First-order homogeneous equations

A first-order equation contains only the first differential such as dy/dx.
An equation will be homogeneous if all terms are of the same degree,
the ratio dy/dx being zero degree, x and xdy/dx being both first degree,
x 2 and xy both second degree, and so on. The standard method of
solution is to replace one variable by introducing a new one that is the
ratio of the variables. If the variables are x and y we can eliminate y by
the substitution y = zx.
Example 6.4
Multiply by dx to give
xdy+2ydx = O.
We now substitute y = ZX, so that
dy = zdx + xdz,
Equations of first order, first degree 143
and
x(zdx + xdz) + 2zxdx = 0,
zdx + xdz + 2zdx = 0,
3zdx + xdz = 0.
The equation will now be of the separable variables kind, to be solved as
before. Thus we collect all terms in dx on one side of the equation and
terms in dz on the other, to give
3zdx = -xdz,
dx dz
x
In x = -! Inz+C,
In x 3 z = C',
x 3 z = C",
where C, C' and C" are constants, which are actually related to each
other by C' = 3C = In C", but the only point of interest is that a
constant has been retained in the correct place. We finally return to the
original variables by the substitution z = y/x, to give
x 2 y = C.
The correctness of the solution may be checked by differentiation and
substitution into the original equation.
Example 6.5
An equation that is homogeneous in the terms containing the variables
but also contains an additive constant can be reduced to a simple
homogeneous equation by using a substitution to eliminate the
constant. Thus if
dy
x-+y=3
ax
we substitute z = = dy, to give
y - 3, dz
xdy+ (y -3)dx = 0,
xdz+zdx = 0,
dx dz
x z
Inx = -lnz+C
Inzx = C,
zx = C',
so that
x(y -3) = C.
6.2.3 Exact equations

An exact differential is defined in Section 3.5 as being of the form
X dx + Ydy where a function z(x,y) exists with this as its differential:
dz = Xdx+Ydy = (!:)y dX+(:;)x dy,

so that
and Y- ( -az)
- ay x·
The test for such an exact differential is the cross-differentiation

identity
If a differential equation either has this form or can be converted into it,
direct integration is then possible.
Example 6.6
dz = 2(x + y)dx + (2x + l)dy.
We first test whether or not this is exact, as in Section 3.5, by seeing if
cross-differentiation applies, which means looking for equality of
a a
ay (2 (x + y» and ax (2x+ 1).
In this case both are equal to 2 and so it is an exact equation. It may be

possible to see by inspection what the function of x and y is that gives
the differential equation, but if not we may proceed as follows.
The solution may be obtained from the equations
2(x+ y) = (::)y and
From the first equation
z = f2(X + y)dx
= x 2 + 2xy + C(y),
Equations offirst order, first degree 145
this integration being the inverse of partial differentiation with respect
to x so that C(y) is a constant only so far as x is concerned; it may be a
function of y. Then
(aoz) .= a0 (x +2xy+C(y»
y x Y
2
dC(y)
= 2x+-- = 2x+ 1
dy
so that
dC(y) = 1, C(y) =y+C
dy
where C is now a numerical constant. The required solution is therefore
z = x 2 + 2xy + y + c.
6.2.4 Linear equations offirst order

These have the form
dy
dx + P(x)y = Q(x) (6.2)
where P(x) and Q(x) do not contain y but may be functions of x or

constants. Writing the equation in the form
dy + P(x)ydx = Q(x)dx (6.3)
the solution is obtained by multiplying both sides by the integrating
factor efP(x)dx.
This has the effect of converting the left-hand side of the equation into
an exact differential by making cross-differentiation apply (Section 3.5).
We show this by multiplying the left-hand side of (6.3) by a functionf(x)
to give
f(x)dy + P(x) yf(x)dx
and cross-differentiation gives
o~~) = OP(~:f(X) = P(x)f(x),

so that
f df(X)
f(x) =
f P(x)dx
lnf(x) = f P(x)dx,
f(x) = e fP(x)cIx,
omitting any integration constant so as to obtain the simplest solution.
Example 6.7
dy
dx +2xy = 4x.
Multiply by ef2xdx = e x2 gives

eX2 dy+ 2xyeX2 dx = 4xex2dx
d(ye X2 ) = 4xe x2 dx.
Integrating both sides
ye X2 = 2e x2 + C, y = 2+Ce _X2.
Example 6.8
Successive kinetically 'first-order' chemical reactions give linear dif-
ferential equations. An extension of Example 6.3 is
A~ B ~C
(a-x) (x-y)
where the rate coefficients kl and k2 are shown, together with the
concentrations of the various species at time t. The rate of production
of C is given by
This may be seen to be a first-order linear differential equation by

writing it in the form
dy
dt +k 2 y = k 2 x.
The integrating factor is e$k 2dt = e k2t, giving

e k2tdy + k2 ye k2 tdt = k2 xe k2tdt,
d(ye k2t ) = k 2 xe k2t dt.
The right-hand side contains the variable x, which may be eliminated
by applying the result of Example 6.3 to the first stage of the reaction.
Thus
Substitution and integration then gives
y = a-
k2 a e -k,t+ Ce -k2t.
k2 -kl
Linear differential equations 147
The integration constant C may be found from the initial conditions
that Y = 0 at t = 0, giving
Y = a[1-(k 2 e -k,t-k 1 e -k,t)/(k 2 -kdJ.
Fig. 6.1 shows typical curves of (a - x), (x - y) and y, each plotted
against time t.
A-B-C
Fig. 6.1
6.3 Linear differential equations

A linear equation is one containing only the first powers of a variable
and its derivatives, the general form of linear equation of order n being
dny dn - 1 Y
ao(x) dxn+adx) dxn- 1 + " .+an(x)y =f(x). (6.4)
There is a general method of solution for such an eqtUltion only when

the coefficients ao, a1> ... are constants, and this is discussed in Section
6.3.2. An equation in which all terms are functions of the dependent
variable Y is called homogeneous, which is whenf(x) = 0 in (6.4), and
this is discussed in Section 6.3.1.
An important consequence of an equation being linear is that linear
combinations of solutions may also be used. For example, if y = oc(x)
and y = fJ(x) are both solutions of the homogeneous equation
d 2y dy
ao dx2 +al dx +a2Y = 0,
then y = ex(x) + fJ(x) is also a solution. This follows because the
equation may then be separated into parts after making the substi-
tution, one part containing ex and the other part containing fJ, and both
parts are zero since ex and fJ are both solutions of the equation.
A physical system is called linear if its variables are linearly related,
meaning that multiplication of one variable by a factor increases
another by the same factor. Thus a linear amplifier is one for which if we
double the input we double the output. For a 'first-order' chemical
reaction the rate is proportional to concentration and a series·of such
reactions gives rise to a linear differential equation, as in Example 6.8.
The differential equations describing linear physical systems will be
linear not only in the dependent variable y and its derivatives but also in
the independent variable x. We may use linear differential equations
not only to describe physical systems that are truly linear, but also as a
first approximation to non-linear systems. This is because the effect of
small enough disturbances will be given by the first, linear, term in a
Taylor expansion.
6.3.1 Homogeneous linear equations with constant coefficients

These have the form
d"y d,,-l y
aO-+al--1 + ... +a"y = o. (6.5)
dx" dx"-
The substitution y = em,x reduces the solution of the equation to that of
the auxiliary. algebraic equation in m.
Example 6.9
Put
dy m,x
dx=me,
then
em" (m 2 + m - 2) = O.
This equation will be satisfied by solving the auxiliary equation
m2+m-2 = 0,
which factorizes to give either m = 1 or m = - 2, so that the differential
equation is satisfied by either y = eX or y = e - zx. It is also satisfied
by y = C1e x and by y = Cze-zX, where C 1 and C z are arbitrary
constants. Furthermore, since the equation is linear, we may also
write
y = Clex+CZe-zx.
This last solution contains the two arbitrary constants expected in the
solution of a second-order equation and is the most general form, the
earlier solutions corresponding to values of zero or unity for the
arbitrary constants.
This method gives solutions whenever the auxiliary equation has real
roots. If the auxiliary equation were to give equal roots, however, we
would obtain only a solution containing one arbitrary constant. In that
case, multiplication of that root by Czx will be found to produce the
required second solution.
A solution to the differential equation may exist even when the roots
of the auxiliary equation are imaginary since these appear as imaginary
exponents in the solution. From Section 1.10
e iy = cos Y + i sin y
and
e -iy = cos Y - i sin y.
Now, the arbitrary constants used in the solution of the differential
equation may also be complex numbers and these will then produce
real solutions, as in the following example.
Example 6.10
Put Y = em.>:, so that

emX(mZ + 1) = °
and the auxiliary equation becomes
mZ+ 1 = 0, m = ±i,
giving the general solution
y = Ae ix + Be -ix.
Taking A and B to be complex numbers, we put A = a + ib and
B = a - ib, giving
y = (a + ib)(cos x + i sin x) + (a - ib)(cos x - i sin x)
= 2acosx -2bsinx,
or
y = Ccosx+Dsinx,
which is wholly real.
Example 6.11
An equation of the form of Example 6.10 is obtained for simple
harmonic motion, when a particle of mass m experiences a restoring
force proportional to its displacement x. Then
d 2x
force = m- = -kx,
2 dt
d2 x k
d t 2 +-x = O.
m
Put
then
a = ±i.j(klm)
giving
x = Ae it.J (kim) + Be - it.J (kim),
and when A and B are allowed to be complex,

x = C cos [t .J (kim)] + D sin [t .J (kim)].
By using the trigonometric identity
sin(A+B) = sinAcosB+cosAsinB
we can write the solution in the form
x = A sin(wt + 4»,
where the constants A and 4> are simply related to the constants C and
D, and
w = .j(klm).
This equation suggests the interpretation in terms of polar coordi-
nates shown in Fig. 6.2; point P moves in a circle of radius A with
angular velocity w, and x is the projection ON of OP onto the vertical
axis. Point N moves up and down on the vertical axis with simple
harmonic motion. The amplitude of the motion is A, and the angle fjJ
gives an arbitrary zero from which the motion begins at time t = O. A
graph of x against t has the form of a sine wave as shown in the figure.
The time period T is that taken for one complete cycle of the motion
and the frequency v is l/T, so that
21t
T= - = 21tJ(m/k), v = (1/21t)J(k/m).
w
If OP is a vector (Section 3.9) representing the dipole moment of a
rotating polar molecule, or if x is the fluctuating dipole moment
accompanying vibration of a molecule, an oscillating electric field is
produced which will be the same as that of electromagnetic radiation of
the same frequency, so that interaction with the radiation can occur and
an absorption spectrum will be observable.
From equation (2.58),
force = -potential gradient = -dV/dx,
so that the potential energy for simple harmonic motion is given by
V= - f Fdx = !kx 2 ,
and the curve of V against x is a parabola (Example 1.7).

The velocity at displacement x is given by
dx
v = dt = wA cos (wt + fjJ),
so that the kinetic energy T becomes

T= !mv 2 = !mw2A 2cos 2 (wt+fjJ)
= !mw2A2[1-sin2(wt+~)]
= !mw 2(A2 _x 2).
But w 2 = kim, so that

T= !kA 2 -!kx 2
giving total energy T + V = ! kA 2 , independent of x. Thus when x = 0
the energy is wholly kinetic, when x = A it is wholly potential, and we
have an interchange between kinetic and potential energies during the
motion.
The vibration of a diatomic molecule occurs by extension and
compression of a chemical bond and we can, to a first approximation,
assume that the restoring force is proportional to the displacement.
Given that we resolve the general motion of a molecule into motion of
the centre of mass and motion relative to the centre of mass, the motion
will be as shown in Fig. 6.3, each atom vibrating relative to the centre of
mass O. The restoring forces on the two masses are equal and opposite,
so that
r1 r2
14-----
I
- -------*------1 I
• ~ I • • m2
1 0
I 1
~-------------------~
ro
Fig. 6.3
But from equations (2.65) and (2.66)
and
so that
d2r
F= -m*-
dt 2
and the motion will be as described above with the usual substitution of
reduced mass m* for m, zero displacement corresponding to the
equilibrium separation r 0 so that x becomes (r - r 0).
This simple model of molecular vibration will apply only to small
displacements; in vibrationally excited states the restoring force will no
longer be proportional to the displacement and the curve of potential
energy against r will depart from the parabola corresponding to simple
harmonic motion.
Example 6.12
Simple harmonic motion will occur only in the absence of frictional
losses. The motion of real macroscopic systems will be subject to
resistance which, often to a good first approximation, is proportional to
the velocity. This applies both to the vibration of an object immersed in
a fluid such as air or a liquid, and to electrical oscillations in the
presence of electrical resistance. The process then becomes irreversible
and potential and kinetic energy are gradually con .. erted into heat, so
that the vibration decays at a rate depending on the resistance. This is
called damping. The restoring force is then given by
dx d2x
force = -kx -{3- = m -
dt dt 2 '
where the frictional, or damping, force is {3 times the velocity and acts in
the same direction as the restoring force when dx/dt is positive and so
the displacement increasing.
This is a homogeneous linear equation with constant coefficients and
the usual substitution x = eat gives the auxiliary equation
2 {3a k
a +-+-=0,
m m
with the solution
a= - 2~ [{3 ± ({32 - 4mk)J.
The behaviour then depends on the relative magnitudes of {32 and 4mk,
which means on the extent of the damping.
When heavily damped, {32 > 4mk and the auxiliary equation has real
roots,say -al and -a2; the solution then has the form of Example 6.9,
which corresponds to exponential decay of any imposed displacement.

This is illustrated as curve A in Fig. 6.4.
When {32 = 4mk, the auxiliary equation has equal real roots, giving
the solution
This also gives exponential decay, the damping force being just
sufficient to prevent oscillation, but allowing the decay to occur most
quickly (curve B in Fig. 6.4). This is called the critical damping
condition.
If {3 were zero we would have purely imaginary roots to the auxiliary
equation as in Example 6.11, and so simple harmonic motion. When
o < {32 < 4mk, less than the critical amount of damping is present and
the roots of the auxiliary equation will be complex, of the form
a = d±if,
x
Fig. 6.4
where d (for decay) andf (for frequency) are real numbers given by
d = - p/(2m) and f = [J (4mk - p2)J/2m
= J(~- :~2).
The solution is then the product of terms arising from the real and the
imaginary parts; the real part is exponential decay and the imaginary
part gives periodic oscillation of frequency f/(2n), the result being a
damped oscillation of form C in Fig. 6.4. We notice that the frequency
of the damped oscillation is lower than the value (1/2n)J(k/m) which
would be obtained in the absence of damping.
The critical damping condition is very important in practice because
that gives the fastest response without oscillation. This is achieved by
adjusting the value of p: in a simple moving-coil galvanometer system
by adjusting the circuit resistance to equal that for critical damping,
and in more complicated control systems by adjusting the control
signal input which depends upon the derivative dx/dt.
6.3.2 General linear equation with constant coefficients
This has the form
dRy dn - 1 y
ao dxn + a1 dx n- 1 + ... + anY = f(x). (6.6)
We again use the principle that the n adjustable constants expected in

the solution of an nth-order linear equation are obtained by adding
together the n separate solutions, as in Example 6.9. In this case,
however, we also take advantage of the' fact that only one of these
solutions need give the right-hand side of the equation; the rest can then
be solutions of the corresponding homogeneous equation, which is
withf(x) = O. To distinguish these two kinds of solution, the ones with
f(x) = 0 are called the complementary function and the solution with
the given value of f(x) is called the particular integral. The general
solution of the equation is then the sum of these two. We will see that
this separation of solutions has not only mathematical significance but
physical significance as well.
There are systematic methods for finding a particular integral for
equations of various types, details of which are given in texts on
differential equations. In simple cases, however, it is often possible to
find a solution by trial; in particular, whenf(x) is a polynomial, or an
exponential or trigonometric function, the same function with ar-
bitrary coefficients should be used as a trial solution. This is substituted
into the equation and, if successful, the values of the coefficients will be
found; if not, we try multiplying the trial solution by x.
Example 6.13
d 2y dy
dx 2 + dx -2y = 1-2x.
Heref(x) = 1-2x, which is a first-degree polynomial, so that we try
the general first-degree polynomial y = ax + b as a solution.
Substituting this into the equation gives
a-2(ax+b) = 1-2x,
which is satisfied by a = 1 and b = O. A particular integral is therefore
y = x. The complementary function for this equation was found in
Example 6.9 to be
and the general solution is the sum of the complementary function and
a particular integral, or
y = CleX+C2e-2x+x.
Example 6.14
Show that the method of this section gives the same solution as the
method of Section 6.2.4 by solving the equation
dy
dx +y = 3x.
We are here comparing the method that applies to a linear equation

of any degree, so long as the coefficients are constants, with the method
that applies to any linear equation of the second degree.
We write the linear homogeneous form of the equation,
dy
dx + y = 0,
and put y = emx,

emX(m + 1) = 0, m= -1
to yield the complementary function

y = Ce- x •
The right-hand side of the original equation is again a polynomial of
first degree, so we try y = ax + b to give
dy
dx + y = a+ax+b
= 3x if a = 3, b = - 3,
giving y = 3x - 3 as a particular integral, so that the general solution
becomes
y = Ce- x + 3x -3.
The solution corresponding to Section 6.2.4 is to multiply the
equation by eJPdx = eJdx = eX, and the equation becomes
dy
eX dx + ye X = 3xe x,
d(yeX) = 3xe xdx,
ye X = 3e X(x-1)+C
so that
y = 3x-3+Ce- x
as before. The same equation is solved by a third method in Example
6.16.
Example 6.15
We have discussed damped harmonic motion in Example 6.12 and have
interpreted the damping as a dissipation of kinetic and potential energy
as heat. A common physical situation is when such oscillation is
sustained by a forcing function in the form of a periodic input of energy
which may be applied mechanically, or as an applied alternating voltage
to an electrical circuit, or as incident radiation of a particular frequency
on a molecular system. In order to sustain oscillation the forcing
function must be 'in tune' with the oscillation; when we apply a
sinusoidal forcing function - F sin pt the equation becomes
dx d2 x .
force = -kx-Pdt-m dt 2 = -Fsmpt.
The complementary function will be as in Example 6.12. A particular
integral is sought by using a trial function of the same form as the right-
hand side of the equation, so we put
x = A cos pt + B sin pt,
dx
dt - Ap sin pt + Bp cos pt,
and the equation becomes

-mp2 x - pApsin pt + PBpcos pt + kx = F sin pt,
which is satisfied if
k = mp2, p= y(k/m),
PAp = -F, A = -F/(Pp) = -(F/p).j(m/k),
B=O.
The particular integral is then
The damped oscillation conditions are, in the notation of Example

6.12, 0< p2 < 4mk and a = d ± if, so that the complementary function
is (compare Example 6.10).
x = e- Pt /(2m)(C 1 cosft + C 2 sinft),-
and the general solution is
x = e- Pt /(2m)(C 1 cosft + C 2 sinft) - ~ J(;) cos [t J(~) J

The physical significance of this is that the motion is the damped
harmonic motion that would follow a displacement in the absence of
the forcing function, together with the final term in the equation. The
condition that was chosen to obtain the particular integral was when
P = .J (kim), which is the value of w for simple harmonic motion
(Example 6.11). This corresponds to applying a forcing function at the
frequency w/2n which the system would display if it were undamped, in
spite of the fact that the real damped oscillation frequency is lower.
If the system were displaced and then the forcing function applied, it
would display damped motion for a time until the contribution from
the first term (the complementary function) in the general solution had
decayed, followed by sustained oscillation given by the second term
only. The condition when the forcing frequency is equal to the natural
undamped frequency is called resonance; this is the completely
contrary condition to critical damping.
6.3.3 Linear equations of second order

The general second-order linear equation can be reduced to a first-
order linear equation if a solution can be found by inspection. Given
the equation
(6.7)
if Y = Yl is a solution to the corresponding homogeneous equation

withf(x) = 0, then put Y = Y1U(X), so that
dy du dYl
dx = Yl dx + u dx'
d2y d2u dYl du d 2Yl dYl du

dx 2 = Y 1 dx2 + dx dx + U dx 2 + ~ dx'
Substitution into (6.7), remembering that when Y = Yl the expression
on the left-hand side of (6.7) is zero, gives
d2u + 2 dYl du =f(x).

dx 2 dx dx Yl
This is a first-order linear equation, which may be solved for du/dx,
from which u(x) may be found by integration.
6.4 Integral transforms

Substitution methods are often used to transform from one in-
dependent variable to another. This may be achieved not only by
algebraic substitution but also by the use of substitutions in the form of
integrals. We consider two forms of such integral transformation:
Laplace transforms, which are useful in solving differential equations;
and Fourier transforms, which are used in instrumentation.
6.4.1 Laplace transforms

The Laplace transform L(f(x)) of the functionf(x) is defined by
L(f(x)) = I'O e - PXf(x)dx. (6.8)
The limits of integration with respect to x mean that the variable x

disappears after integration; we obtain instead an expression in terms
of the new variable p, which is why this is called a transformation. This
is useful in solving differential equations for three reasons: the first is
that the integral can be evaluated for simple functions and, indeed, is
given in tables. The second reason is that it is a linear operator, which
means that multiplying by a constant, or adding or subtracting terms in
f(x), simply means multiplying, adding or subtracting terms in L(f).
The third reason, which allows it to be useful in solving differential
equations, is that the Laplace transform of derivatives dy/dx,
d 2 y/dx 2 , ••• are also simple algebraic functions of the new variable p.
The terms in a linear differential equation can in this way be converted
into an algebraic equation in p, which can be solved to give the Laplace
transform of y in terms of p. The tables of transforms are then
consulted again in ·the inverse sense of: knowing the transform L(y),
find yand so the solution of the equation. Only a simple example will be
given to illustrate the method, the reader being referred to a text
containing a full table of transforms for more details. This technique is
particularly useful for the equations describing electrical circuits.
Integral transforms 161
A particular example of a Laplace transform is the gamma function
r(x), defined by
(6.9)
so that r(x) is the transform of t x - 1 • This may be evaluated by

integration by parts. We write the function for (x + 1) as
r(x+ 1) = to tXe-tdt
= -[tXe-t];+x LOO tX-le-tdt

= xr(x). (6.10)
Repeated integration by parts then gives
r(x + 1) = x!r(l)
where
so that
If we substitute n for x and px for t in (6.11) we obtain the form (6.8) as
pft foro xfte-PXpdx = pft+l L(xft) = n! (6.12)
In particular, when n = 1 and when n = 0

L(x) = 1/p2 and L(l) = lip. (6.13)
The Laplace transform of dyldx is given by
roo dy
L(dyldx) = Jo e- PX dx dx
= [ye- PX ]; + p LOO ye-PXdx
= - y(O) + pL(y), (6.14)

where y(O) is the value of y when x = O.
Example 6.16
As a simple example we use Laplace transforms to solve the same
equation as Example 6.14
dy
dx + y = 3x.
Taking Laplace transforms of each term gives

- yeO) + pL(y) + L(y) = 3/p2.
Solving for L(y) we obtain
1 3
L(y) = p+1 y(O) + p2(p+1)
1 3 3 3
= p + 1 yeO) - p+ p2 + P + 1
where we have used partial fractions for the last term. We now take the
inverse transform of each term. From a table of transforms, or by
evaluating the integral, we find
L(e- X ) = 1/(P+1)
which, together with (6.13), gives
y = y oe- x -3+3x+3e- x
= (Yo + 3)e- -3 + 3x,
X
which is the same solution as obtained by the complementary function

and particular integral approach in Example 6.14 when C = Yo + 3.
Example 6.17
The Schrodinger equation in one dimension has the form
a 2t/! 8n 2m
ax 2 = - ~ (E - V)t/!.
For free translation of a particle of mass m between walls separated by

distance a, the potential energy V does not change and so may be taken
as zero, and the wavefunction t/! must be zero at the walls, which are
defined by x = 0 and x = a.
This equation may conveniently be solved by using Laplace trans-
forms. From tables, or by integration similar to (6.14),
L(d 2t/! /dx 2 ) = p2 L(t/!) - pt/!(O) -1/1'(0).
Integral transforms 163
Taking transforms of the equation, with V = 0 and 1/1(0) = 0,
81t 2 m
p2 L(I/I) - 1/1' (0) = - ~ EL(I/I).
Putting 81t 2mE/h 2 = A2 we obtain

L(I/I) = I/I'(0)/(p2 +A2).
Again from tables
L(sin AX) = A/(P2 + A2)
giving the solution to the equation
1/1' (0) .
1/1 = -A- sm AX,
and at x = Q, 1/1 = 0 so that
sin AQ = 0, A.a = n1t.
Hence
6.4.2 Fourier transforms

Fourier transforms are of practical importance because their use
enables instruments to be designed and used in such a way that results
can be obtained very quickly. This is an advantage because measure-
ments of weak signals always have superimposed upon them the effects
of random disturbances called noise. Since noise produces signals of
random sign these average to zero for a large enough number of
repeated measurements, leaving the required signal less obscured by
noise.
Sine and cosine functions are periodic, so that a graph of sin wt
against t has the form of a wave of frequency w/21t. If we superimpose
sine and cosine functions of different frequencies we produce a
complicated pattern which, given enough terms, can be adjusted to fit
any particular curve as closely as we may wish. The inverse of this, the
resolution of a given function into a sum of sine and cosine terms, is
Fourier analysis, for which there are standard procedures. In the limit
when the number of terms becomes infinite we may replace the sum by
an integral; we then obtain the Fourier transforms of!(w) as
F(t) = -1 f+oo !(w)e-iwtdw

21t - 00
and
/(w) = f:: F(t)eiwtdt
which relate functions of frequency (w) and functions of time (t). The
complex exponential function is expressed in sine and cosine terms in
Section 1.10.
An appropriate text should be consulted for the derivation and use of
Fourier analysis and transforms. Their application in spectroscopy is to
transform measurements of the decay with time of the response
induced by an irradiating pulse into the required spectrum as a function
of frequency. The mathematical and physical principles may be
brought together by observing, first, that a pulse is equivalent to a
superposition of different frequencies with the major components near
the frequency of the pulsed radiation, so that a spectrum of absorption
frequencies can all be excited simultaneously and, secondly, that the
resulting decay, or relaxation in time, that is observed can be
transformed into a superposition of exponential decays in time of a set
of frequencies, the set being the required spectrum.
CHAPTER 7
Experimental error and the

method of least squares
7.1 Significance
Experimental measurements must be made and used with their
probable accuracy in mind. Some properties can and should be
measured more precisely than others; for example, it is as absurd to
measure the length of a piece of string to 0.01 mm as it would be to
weigh a block of gold to the nearest kilogram. The least that should be
done is to quote only figures that are believed to be reliable, so that a
result given as 2.1 implies that it is between 2.05 and 2.15, and 2.1234
implies that it is between 2.12335 and 2.12345. It is preferable to give an
explicit statement of precision, such as 2.123 ± 0.024, the result and the
uncertainty both being quoted to the same number of decimal places.
This implies that the results have been analysed with great care because
two figures are quoted in the uncertainty, which is the most that can
ever be achieved. It does not imply that it is not possible to obtain
greater precision in the measurement because improved equipment and
better instruments may well produce a more precise result; it does
imply, however, that such measurements will not give results outside
the bounds set on the existing measurement. This chapter is concerned
with the methods of analysing experimental data which are necessary to
achieve such a high level of confidence in the statement of results.
Experimental errors are of two kinds, systematic and random. A
typical systematic error would arise if length were measured using a
ruler calibrated at one temperature but used at a higher one. All
measurements would then be short by an amount proportional to the
166 Experimental error and method of least squares
length being measured. Systematic errors arise from inadequacy in the
calibration of instruments, or from inadequacy in the design of
apparatus or in the technique used.
The term 'random' is used in error analysis in a particular sense.
Instability, due to poor control, mechanical vibration or thermal or
electrical noise, produces a scatter in the experimental results. Small
errors occur more often than large ones, and an error of given size will
usually be equally often positive as negative. Such errors can be
analysed statistically in terms of the distribution of error about a mean
value; this distribution is a curve as in Fig. 7.1, which shows the number
of times a particular error occurs as a function of the size of that error.
Such a curve could be drawn if a large enough number of repeated
measurements were made.
The simplest case is the one already mentioned, when errors are of
random sign. The curve is then symmetrical about a maximum, which is
the arithmetic mean. This is the normal, or Gaussian distribution of
error, discussed in Section 7.7. If significant systematic error is present
in addition to such random error, the mean will not be the true value
and the curve may not be symmetrica1.
Fig. 7.1 shows the probability that a particular result will be obtained
in the measurement of a quantity x. We do not therefore expect
repeated measurements to give the same result, and the quality of a set
of measurements is assessed by the spread of results obtained. The
uncertainty in the final result needs to be known and the conventional
way of expressing that uncertainty is discussed in the next section.
Since the methods used to assess the size of experimental error are
statistical, the sample size, which means the number of independent
measurements made, is very important. The statistical theory is
developed by assuming at first an infinitely large number of measure-
ments. If we then add the results together, we sum over such a large
number that the differences between adjacent results becomes so small
that we can replace sums by integrals. For this reason sums and
integrals occur in this chapter, depending on whether or not we are
assuming an infinite number of measurements.
7.2 Root-meaD-square error

If the measurement of a quantity x is repeated n times, the results will
scatter about the arithmetic mean x given by
x = LXi, (7.1)
n
~
I
>c'
~
...r...:
~
s
and this is the best result obtainable from the measurements. The
experimental error in a particular result Xi is then the difference
between that value and the arithmetic mean, (Xi - x). The final result of
the experiment is to be expressed as x ± Ax, where Ax is a conventional
assessment of the uncertainty of the measurement x of x.
In order that Ax shall be a useful measure of uncertainty it should
reflect the spread in the values of Xi obtained. We cannot use for this the
average value ofthe error (Xi - x) because that is zero by definition ofx.
We therefore use instead the square of the deviation, (Xi - X)2, and
define the root-mean-square error, s(x), by
s(x) = -J II (Xi - X)2 /n]. (7.2)
The relation between this and the standard deviation q(x) is discussed
in Section 7.6.
The root-mean-square error can be calculated most easily from a set
of values of X by use of the relation
S(X)2 = (l/n) L (Xi - X)2 = (l/n) L (xf - 2XiX + x 2)
= Lxt/n-2xLxdn+x2
= Lxt/n-x 2. (7.3)
7.3 Distribution of error

For discrete variables, such as the number of matches in a box, we
define the probability p(x) of obtaining a particular result X as the
fraction of a very large number of samples which contain that number.
In physical measurements, however, we are usually concerned with
continuous, rather than discrete, variables. We can then define only the
probability of obtaining a result in an interval of X between X and
X + dx as p(x)dx, where p(x) is a continuous function of x. This is the
smooth curve shown in Fig. 7.1. We ensure that p(x) is a fraction
between 0 and 1 by normalizing the function to unity, which means
making the total area under the curve unity by
f +OO
_ 00 p(x)dx = 1. (7.4)
The mean value x of x is obtained by summing (integrating) the

product of x and its probability p(x) over all possible values of x,
x = _ f +OO
00 xp(x)dx. (7.5)
The statistical analysis of experimental data 169
The root-mean-square error O'(x) in x is then given by
0'(X)2 = L+oooo (x -x)2 p (x)dx. (7.6)
7.4 The statistical analysis of experimental data

The methods used in the statistical analysis of data can be classified
either from the mathematical or from the physical points of view, both
of which will now be considered.
The simplest case is when measurements of a single quantity are
repeatedly made. The results will scatter and, if a sufficiently large
number ofreadings is taken, the distribution curve can be drawn. Only
a small number of results will be available in real cases and there are
standard tests, such as the 'chi-squared' test not considered here, to
determine how closely a set of results follows a particular distribution.
This has value as a way of testing for systematic error of a kind that
distorts the distribution.
We are more often concerned with measurements of more than one
variable. Statistical methods can then be used either to test whether or
not the variables depend on each other, or to find the most probable
values of numerical coefficients in an equation that is assumed to fit the
data. The latter application is the more important one in physical
science. From the statistical point of view there is no difference between
these two applications because the uncertainty in the coefficients of a
fitting equation can be regarded as a measure of the degree to which the
assumed relation fits the data. Statistical texts discuss the concepts of
covariance and correlation coefficient (both considered here) and of
tests of significance and confidence limits (not considered) as measures
of the extent to which variables are related rather than as measures of
the uncertainty in the values of the fitting parameters.
Another way of comparing the various aspects of statistical analysis
is in terms of the relevant number of independent variables. This is a
physical concept, and for well defined chemical systems is easily
determined. So long as dl independent variables are identified and their
values controlled in the experiment, a true value of any experimental
quantity will exist, and results will be subject only to experimental
error. On the other hand, properties and materials may not be
rigorously defined and the independent variables may not be ident-
ifiable and controllable. A true value will not then exist and relations are
being sought between quantities which both show legitimate varia-
tions, and this form of correlation analysis is not considered here.
7.5 Propagation of error
Experimental results are often substituted into equations so as to
obtain values of related quantities. Knowing the uncertainty in the
results we then need to calculate the uncertainty in the related quantity,
for which we use the formula for propagation of error.
Suppose a quantity z(x, y) depends on experimental variables x and
y, and that measurements are repeated n times to obtain pairs of results
(Xl' yd, (X2' Y2),· .. ,(xn , Yn). We define the probability of obtaining a
result in the range (x to x + dx, y to Y + dy) as p(x, y)dxdy. The
f:: L+:
arithmetic mean value z of z is then
z= z(x, y)p(x, y)dxdy. (7.7)
We can also write the mean value of x in this form,
X= f _+OO f+oo xp(x, y)dxdy

00 _ 00 (7.8)
because the integration with respect to y is simply the integral of the

probability of obtaining any value of y for a particular value of x, which
is normalized to unity by the two-dimensional analogue of (7.4)
The root-mean-square errors cr(x), cr(y) and cr(z) in x, y and z
respectively are defined as
+OO f+oo
cr(X)2 = f_ 00 (x -_ 00 X)2 p(x, y)dxdy, (7.9)
+OO f+oo
cr(y)2 =f -00 -00 (y - y)2p(X, y)dxdy, (7.10;
cr(Z)2 = L+: L+: (z - Z)2 p(x, y)dxdy, (7.11)
and we require an expression for cr(z) in terms of cr(x) and cr(y).

Assuming an infinite number of measurements, the mean values will
be free from experimental error so that zis obtained from x and y. We
use Taylor's theorem (Section 3.8) to express a departure of z from the
arithmetic mean in the form
(z -Z) = (oz)
oX y
(x -x)+ (oz) (y _
oy x
y)+~2 [(02~)(X
ox
_X)2
+ 2 (o~:y) (x -x)(y - y) + (:;~) (y - y)2 ]
+ higher-order terms. (7.12)

Small-sample errors 171
We assume the errors to be small so that only the first-order terms in
the Taylor expansion are significant. Then
(z _Z)2 = (!:Y (x _X)2 + (;;Y (y - yf + 2 (;:)(!;) (x -x)(y - y).

(7.13)
Substituting (7.13) into (7.11), assuming higher derivatives to be
negligible so that the first derivatives are constants to be taken outside
the integrals, we obtain
U(Z)2 = (!:Y U(X)2 + (;; Y U(y)2 +2 (!:) (!;) u(x, y) (7.14)
where
u(x, y) = L+oooo L+: (x -x)(y - y)p(x, y)dxdy. (7.15)
The mean squares of the errors U(X)2 and U(y)2 are called the variances
in x and y, and the new term u(x, y) is called the covariance of x and y.
The significance of covariance is that it determines whether or not the
variables are independent of each other. If they are independent, the
integral in (7.15) separates into the product of two integrals each with
respect to one variable only, and both of these integrals are zero by
definition of the arithmetic mean. The covariance is therefore zero if the
variables are independent.
In general, if z is a function of independent variables x, y, u, ... the
formula for propagation of error is
U(Z)2 = (!: y U(X)2 + (;; y U(y)2 + (;~ Y U(U)2 + .... (7.16)
7.6 Small-sample errors

We aSSllme that a particular distribution curve such as Fig. 7.1 exists for
the results of any experimental measurement. This curve is character-
ized by the form of the distribution, the arithmetic mean and the root-
mean-square error. A particular form of distribution is considered in
the next section, before which we consider relations that apply to any
form of distribution.
The arithmetic mean has been defined for a series of n results by
equation (7.1) and the corresponding root-mean-square error as s(x) by
equation (7.2). These have also been defined in terms of integrals by
(7.5) and (7.6), These become the same onlv as the number of results
tends to infinity because we require, in principle, an infinite number of
points to establish the curve of distribution of error with complete
certainty. The mean xn of a sample of n results is only an estimate of the
mean Xoo of the distribution, so that a different sample of points will
give a different value of xn; only as n tends to infinity will the means of
different samples become the same and equal to the mean of the
distribution. In the same way, the root-mean-square error of a sample
of n results, denoted by s(x), is only an estimate of that of the
distribution u(x), which is called the standard deviation.
If we obtain n results and calculate the mean X n , and then repeat the
set of n measurements a large enough number of times, we obtain a set
of values of xn which will themselves follow a distribution about the
true mean Xco" We can then calculate the standard deviation of the
mean u(xn ), which is a measure of the uncertainty in xn due to the fact
that we have too few results. By definition
and since
_ Xl X2
Xn=-+-+
n n
the integral separates into the sum of n integrals of the form
n 1 f+oo _ n U(X)2 U(X)2
L L
_ 2 2
U(Xn) = 2 (Xi -Xoo) p(x)dx = -2- = --,
i=ln -00 i=l n n
so that
_ U(X)
U(Xn) = Tn· (7.17)
This relation shows that when n duplicate measurements are made of

the same quantity, each of which is subject to error with standard
deviation u(x), and the mean is calculated so as to obtain a more reliable
estimate, the uncertainty is thereby reduced by the factor 1/ In. Thus
the mean of nine duplicate readings will reduce the uncertainty by a
factor of one-third, but to improve the precision by one decimal place
we would need to take 100 readings. It is seldom worth while to repeat a
reading more than about 10 times, except when automatic recording is
available. When it is possible to take readings very quickly, automatic
repetition is especially useful because the effect of even high-frequency
noise can then be averaged out.
The effect of small-sample error on the estimate s(x) of the standard
deviation can also be calculated. Here we are seeking not the error in
Small-sample errors 173
the measurement, but the error in our estimate of that error, which is of
second order in small quantities. If the true mean of the distribution,
x<x>' were known we could obtain a good estimate of the standard
deviation from only n results by using the relation
2 L (Xi -X<x»2
U(X) = =------'-- (7.18)
n
We introduce X" into this expression by writing the algebraic identity
(Xi-X<x» = (xi-x,,)-(x,,-x<x»
so that
(Xi -X<x»2 = (Xi -X,,)2 -2(Xi -X,,)(X,,-X<x»+ (X,,-X<x»2,
and
L(Xi-X<x»2 = L(x i -x,,)2+ n(X,,-X<x»2, (7.19)
the middle term having disappeared because L (Xi - x,,) = 0 for the
arithmetic mean. We can then write the required relation between s(x)
and u(x); by definition
(7.20)
and combining (7.18), (7.19) and (7.20) gives

nu(x)2 = L(Xi-x,,)2+n(X,,-X<x»2
= ns(x)2 + nu(x,,)2
= ns(x)2 + U(X)2
so that
UX
n
( )2 =--SX
()2
=
L (Xi - l
X,,)2
' (7.21)
n-1 n-
which is the equation for the standard deviation of a set of measure-
ments ofa quantity x. The estimate s(x) of the standard deviation u(x) is
said to be biased by the smallness of the sample, this bias being
corrected by the factor nj(n -1). This can be interpreted by noting that
the calculation of the mean involves the use of an equation containing
the n variables, so that (n -1) variables remain independent, or we have
(n -1) degrees of freedom. By analogy, if we fit n sets of values of
several experimental variables to an equation containing k numerical
coefficients, we then have (n - k) degrees of freedom and the standard
deviation of the fit is
(7.22)
where (Zi - Zcalc) is the deviation of an experimental value of Z from the
fitting equation.
7.7 The normal distribution of error

We assume random sign error in a set of independent measurements of
an experimental variable x. This means that errors of given magnitude
occur equally often with positive as with negative sign, and the sum of
the errors is zero. When error is defined as departure from a mean value
m, this must be the arithmetic mean because
L(xi-m) = LXi-nm = 0
so that
m = Lxdn = x.
We can find an equation for the probability P(Xi - m) of obtaining
error (Xi - m) by maximizing the probability of obtaining a set of
independent errors subject to the condition that m is the arithmetic
mean.
The probability P of obtaining a set of results Xl' X2' ... , Xi' ... is
the product of the probabilities of obtaining each result,
P = np(xi-m)
which will be a maximum when dPjdm = 0, or
dlnP _ 1 dP _ 0
dm-Pdm-·
Introducing the condition that m is the arithmetic mean x, we have
(7.23)
and
din P = L din P(Xi - m) = L din p(xi - x) = 0
dm dm dx'
so that
L [d In P(Xi - x)jd(xi - x)] = o. (7.24)
Solving equations (7.23) and (7.24) simultaneously by adding A. times

equation (7.23) to equation (7.24) (compare Section 2.8) we obtain
dlnp(xi -x) _ dlnp(x2 -x) _
d( ) +A.(xl-x)+ d( ) +A.(x 2 -x)+ ... =0.
Xl - X x 2 -x
(7.25)
The normal distribution of error 175
Hence, by suitable choice of A,
din P(Xi - x) = - A(Xi - x)d(xi - x)
and integration gives
P(Xi - x) = K exp [ - A(Xi - x)2/2].
The integration constant K is determined by normalizing p(x i - x) to

unity, so that
00
5-+00 P(Xi - x)d(xi - x) = 1,
f:
or since P(Xi - x) is symmetrical about Xi = x,
00 P(Xi - x)d(xi - x) = !.
This gives an integral of the standard form discussed in Example 5.1,
from which
K f: oo
exp[ -A(X i -x)2/2]d(xi -x) =K J(n/2A) =!
K = J(A.j2n)
and
P(Xi - x) = J (A/2n) exp [ - A(Xi - x)2/2].
Writing the first and second derivatives of P(Xi - x) with respect to

(Xi -x),
dp(xi - x)/d(xi - x) = - J(A/2n)A(x i -x)exp[ -A(X i -x)2/2]
d p(xi -x)/d(x i _X)2 = J(A/2n)A[A(xi _X)2 -1] exp[ -A(Xi - x)2/2].
2
At Xi = x the first derivative is zero and the second is negative, so that

P(Xi -x) is a maximum at the arithmetic mean, as required. The curve
will have the shape shown in Fig. 7.1, and the halfwidth ofthe curve can
be expressed as the error at the point of inflection; this is when the
second derivative is zero, or
A(Xi - X)2 - 1 = 0
(Xi - x) = ± 1/ J A = ± 0', say.

Then
1
P(Xi- X) = J exp[ -(Xi-X)2/20'2] (7.26)
0' (2n)
and 0' is called the standard deviation. This can be identified with the
root-mean-square error o-(x) by
0-(X)2 = L+",'" (x i -x)2 p (x i -x)d(x i -x)

= 2 L'" (x -x)2{1/[0-v'(27t)]}exp[ -(x
i i -x)2/20- 2 ]d(Xi- X )
= 0- 2 by Example 5.2.
Equation (7.26) describes the normal or Gaussian distribution of
error. We can expect experimental measurements to follow this
distribution so long as only random sign errors are present.
A theorem that is important in practice is the central limit theorem,
which states that if mean values are taken of samples having any
distribution, those means follow a normal distribution. Such averaging
can occur in the use of measuring instruments, so that we then obtain a
normal distribution of error even when the quantity being measured
does not itself follow that distribution.
7.8 The method of least squares

Given a normal distribution of error, the probability of obtaining a set
of results Xl' X 2 , •.. , Xi is
p = n[p(X i -x)]
=( 1 )n exp ( - L (X i2- X)2). (7.27)

0- -J (27t) 20-
This will be a maximum when
L (Xi - X)2 is a minimum.
This is the principle ofleast squares, that the sum of the squares of the
deviations is a minimum.
7.B.1 Linear relation between two variables

Given that experimental variables X and yare related by the equation
y = mx+c, (7.28)
we wish to obtain the most probable values of m and c to fit a set of
independently measured points (Xl' yd, (x 2 , Y2), ... , (Xi' Yi), ... ,
(Xn' Yn). In general, both X and Y will be in error. We assume a normal
distribution of error for each variable at each point, the means of which
The method of least squares 177
distributions lie on the straight line
Yi = mxi+c. (7.29)
In Fig. 7.2, curves illustrating the distributions parallel to the x- and
the y-axes have been drawn displaced for clarity; these curves should be
interpreted as being centred on (Xi' y;) and in the planes perpendicular
to the paper and parallel to the axes. An error in x displaces the point
from the centre of its distribution to, say, the position of the triangle,
and error in y to the position of the square. The observed point is then
found at X.
y
(ii' Jij )
I? X
6-- - - ~-x
xi
I
I
Fig. 7.2
When we assume independent random errors in x and in y, the

probability of obtaining point (Xi' y;) is the product of the probabilities
P(Xi - x;) and P(Yi - Yi). If we assume the standard deviation a(x) to be
the same at each of the n points, and similarly for a(y). we obtain
P(Xi -Xi' Yi - y;) = P(Xi -Xi)P(Yi - Yi)'
so that
p_( -
1
21ta(x)a(y)
)n exp [_(L(X2a(x)2
i -Xi)2 +L(Yi-Yi)2)]
2a(y)2 '
which will be a maximum when

a(y)2 L (Xi - XY + a(x)2 L (Yi - YY is a minimum. (7.30)
Further assumptions are necessary for the best line to be found because
so far we have more unknowns than data points.
If a group of n points from a chosen range of x is taken and the mean
(x, y) is calculated, this mean point will then have a smaller uncertainty
than the individual points by the factor 1/ J n (equation (7.17)). A
recommended procedure when the errors in both x and Y are significant
is, therefore, to divide the points into groups at the upper and lower
regions of the variables, calculate the mean point of each group, and use
these two points to define the straight line.
In real applications the errors in the two variables are often not the
same. We then assume the better known variable to be free from error,
this being used as the independent variable x. When we assume error to
be vested only in Y we have Xi = Xi' and condition (7.30) becomes
L (Yi - Yi)2 is a minimum
with
We put di = Yi - mX i - c and call this the residual; this is the deviation

of the point (Xi' Yi) from the line measured parallel to the y-axis. The
least-squares condition is then that the sum of the squares of the
residuals shall be a minimum,
D = Ld; = L(Yi-mxi-c)2 is a minimum.
We now regard D as a function of the two unknowns m and c, so that
dD = (:~)dm+(~~)dC = o.
A solution of this equation will be
(:~) = (~~) = 0
giving
and since we may now omit the subscript i without causing confusion,
Lxy-mLx 2 -cLx = 0, (7.31)
:C L(Yi- mxi- c)2 = -2L(Yi- mxi- c) = 0,

Ly-mLx-nc = O. (7.32)
Solving (7.31) and (7.32) simultaneously for m and c gives the equivalent
forms of equation
nLxy - LX~> L(x -x)(Y- y) L(x -x)y
m = = = .;;-------'-'=- (7.33)
nLx 2 -(Lx)2 I(X-X)2 L(x-xf
LYLx 2 -LxLxy _ _
c= nLx2 _ (LX)2 = y-mx. (7.34)
Equation (7.34) shows that the least-squares line passes through the
'centre of gravity' of the points, (x, y), and since (7.32) may be written as
L(Yi-mxi-c) =0
we see that minimizing the sum of the squares of the residuals
incidentally makes the simple sum of the residuals also equal to zero.
7.8.2 Covariance and correlation coefficient

The covariance u(x,y) of variables x and Y is defined in Section 7.5 and
shown to be zero if x and yare independent. On the other hand, if x and Y
are linearly dependent, the slope m of the least-squares line Y = mx + c
can be expressed in terms of u(x, y). For a finite number of points, m is
given by (7.33) and the estimates s(x, y)of u(x, y), S(X)2 of U(X)2 and S(y)2
of U(y)2 are
s(x,y) = [L(x -x)(y-y)]/n, (7.35)
S(X)2 = [L(x -x)2]/n, (7.36)
s(yf = [L(y - y)2]/n, (7.37)
so that the estimated value of m is
m = s(x, Y)/S(X)2. (7.38)
The values of m and c for the least-squares line will be uncertain because
only a limited number of points are available. As the number of points
tends to infinity the uncertainty in the line tends to zero, s(x, y) ..... u(x, y),
S(X)2 -+ U(X)2 and L (y - y) ..... m L (x - x), so that
S(y)2 = L(y_y)2 = m 2 I(x-x)2 = m2s(x)2.

n n
Hence for an infinite number of points
U(y)2 = m 2u(x)2
and (7.38) gives
u(x, y) = mu(x)2 = u(x)u(y). (7.39)
The ratio
u(x, y)
p= -----:--,--~ (7.40)
u(x)u(y)
is called the correlation coefficient. This will be zero if x and yare

independent since u(x, y) will then be zero. When x and yare linearly
related, equations (7.39) and (7.40) show that p = 1, and this provides a
test for such a linear relation.
When this is applied to a finite number of points we can calculate
only an estimate r(x, y) of the correlation coefficient, given by
r = r(x, y) = s(x, y)/s(x)s(y), (7.41)
which may be written in the alternative forms
2 [~::<x - x)(y - y)]2
r = ~"'--'---.,..,.:.~--=--,-":-",,
L(X -.x)2L(Y - y)2
nCLY+ nmLxy - (Ly)2
nLy2 _(Ly)2
(n LXY - LX Ly)2
(7.42)
This will not be unity even when a true straight line exists because
small-sample error means that expressions (7.35}-(7.37) are not the
same as (7.9), (7.10) and (7.15). A value of r that is less than unity is
therefore evidence either of limitations in the number and precision of
the points, or that the supposed linear relation does not apply; it is not
possible to separate these two possibilities for a given set of results.
A correlation coefficient that is much less than unity does not imply
independence of the two variables since it may be that a curve rather
than a straight line fits the data. This may be seen by plotting the graph,
and the variables should then be changed to convert a curve into a
straight line before applying the least-squares analysis.
7.8.3 Uncertainty in the slope and intercept of the least-squares

straight line
Equations (7.33) and (7.34) for m and c were derived assuming the
errors to be vested entirely in the variable y. The error bYi is the
difference between Yi and the value mX i + c given by the line,
(7.43)
From equation (7.33)
and by propagation of error (Section 7.5), with (x; - x) assumed to be

precisely known,
Writing the mean value of byf as (1';, and assuming this to be the same
for all y;,
~( -)2 2 2 2
«(jm)l = L. x;-x (1'y = (1'y = nay (7.44)
[I(X;_X)2]2 I(X;-X)2 nIx 2 -(Ix)2
where
-
(1 - r2)[ n ~>2 (Iy)2]
(7.45)
n(n -2)
The divisor (n - 2) in (7.45) arises because two numerical coefficients, m
and c, are derived from the data leaving (n - 2) degrees of freedom.
When the least-squares line is used to estimate the value Yo of y
corresponding to the value Xo of x we have
Yo = y+m(x o -x)
so that
(bYO)2 = «(jy)2 + (xo _x)2«(jm)2.
Again assuming a mean value (1'; for all values of y, the error in the mean
value of y is given by equation (7.17)
(by)2 = (1'; In
which, together with (7.44), gives
S:)2
(oyo = (1' 2y
(1-+
n
(xo - X)2)
I(X;-X)2
(1'; (xo - X)2 n(1';
= -+-=.-'--:---=---'-;;-
n nIx 2 -(Lx)2
(7.46)
The uncertainty in the intercept c is then found by putting Xo = 0;
(7.47)
with (1'; given by equation (7.45). By combining equations (7.44) and

(7.47) we obtain
Example 7.1
The result of applying this analysis to a set of points is shown in Fig. 7.3.
The mean point (x, y) has error bars at ± a y/ -J n. The least-squares line,
a, is drawn with alternative slopes (b and e) of ± 15m. The curves d and e
are drawn to show the errors ± 15Yo at particular values of Xo as given by
equation (7.46). The error ± be in the intercept on the y-axis is due
mainly to the second term in the bracket in (7.47), corresponding to
uncertainty in the slope, with an additional contribution from the first
term which corresponds to the uncertainty in y. The correlation
coefficient for the points shown is r = 0.775; it can be seen that as Iowa
value as this indicates a considerable scatter in the points.
x
Fig. 7.3
7.8.4 Least-squares straight line with both variables subject to error
The calculation of the least-squares line in Section 7.8.1 assumes that
the error is vested in only one of the two variables and equations (7.33)
and (7.34) for m and c assume the error to be in y. Experimental
conditions may justify this assumption, otherwise we might equally
well assume the error to be vested only in x; we would then obtain a
different straight line by interchanging x and y in the equations.
Since the least-squares line passes through the 'centre of gravity' of
the points, the two lines obtained by assuming the error to be vested
either in y or in x will intersect at that point. The ratio of the slopes of
these lines is related to the estimate r of the correlation coefficient.
From equation (7.38)
s(x, y) I(x -x)(y - y)
m = S(X)2 = I(x _X)2 (7.48)
We write the alternative line as x = m'y + c' and by interchanging x and

y we obtain
, s(x, y) I (x - x)(y - y)
m = - - - = =--,=-,--~--;;----'---'- (7.49)
S(y)2 I (y - y)2
Now m is the slope dyldx, whereas m' is dx/dy; we therefore compare m
with 11m'. The ratio of the slopes is then
m , s ( x , y)2 2
-- = mm = = r . (7.50)
11m' s(xf s(yf
As the number of points tends to infinity, the estimated correlation
coefficient r tends to the value p free from small-sample error. The
alternative lines will then move closer together as the number of points
is increased, and will coincide for an infinite number of points if the
linear relation applies so that p = 1. It is conventional to assume that if
the errors in the two variables are both significant the best fit to the data
passes through (x, y) and bisects the angle between the alternative lines,
with slope equal to the mean of m and 11m'.
7.8.5 Weighting of observations

We have assumed so far that all experimental points are equally
reliable, having used the same uncertainty (1 y for all values of y. There
may, however, be experimental reasons for regarding some points as
more reliable than others; this can be included in the analysis by
weighting the points. The simplest way of achieving this is to repeat the
entry of any particular point into the analysis w times, thus making that
point w times as significant as a single point. It is necessary to remember
only that these weighting factors must be included in an effective total
number of points.
The proper weight to be used in precise work can be seen from the
expression for the normal distribution, equation (7.27); the quantity to
be minimized is the sum of the squares of the deviations divided by (12,
so that each point should be weighted by the reciprocal of its standard
deviation, such as might be estimated by the formula for propagation
of error.
7.8.6 Multivariable and non-linear least-squares analysis

When fitting results to a more complicated equation than that of a
straight line, the analytical methods of this chapter are still applicable
and remain useful in simple cases. Assuming the error to be vested in a
particular variable, the difference between an experimental value of the
variable and that predicted by an assumed equation is the residual d j •
The principle of least squares is that L: (d f) is a minimum. Partial
differentiation of L: (d f) with respect to each numerical coefficient in
turn gives a set of normal equations, which can then be solved. When
these equations are linear in each coefficient the solution is straightfor-
ward; otherwise linearization may be achieved by using trial values of
the coefficients, for which a text on statistical analysis should be
consulted.
An alternative to the analytical approach is to use trial-and-error
searching by computer to find values of the coefficients that produce
the smallest value of L: (d j )2. Standard programs are available for this
purpose, some of which make use of expressions for the partial
derivatives with respect to the coefficients to accelerate convergence. It
is important when using numerical methods to allow only the smallest
possible number of search parameters; an excessive number will
produce an undulating curve or surface which may fit the data precisely
but be physically misleading. A polynomial fit in which the coefficients
alternate in sign may be evidence of this effect.
APPENDIX
SI units, physical constants

and conversion factors;
the Greek alphabet and a
summary of useful relations
SI units
SI stands for the agreed international system of units, which is based
upon the kilogram (kg), metre (m) and second (s), together with the
kelvin (K) unit of temperature, the mole (mol) for amount of substance
and the ampere (A). The candela (cd) unit ofluminous intensity is also a
basic unit. Other units are derived from these basic units.
The kelvin is defined as 1/273.16 of the thermodynamic temperature
of the triple point of water. This means any subsequent refinements in
absolute temperature measurement will alter the size of the degree and
so, say, the absolute temperature of the boiling point of water, without
changing the triple point. The ice point is 0.01 K below the triple point
so that O°C is 273.15 K.
The mole is the amount of substance containing as many elementary
units as there are atoms in 0.012 kg of 12c. This allows experimental
measurement of the Avogadro constant by, in effect, counting this
number of atoms, and of the gas constant from measurements of the
properties of gases.
The unit offorce is the newton (N), defined as that which produces an
acceleration of 1 m s - 2 in a mass of 1 kg. Pressure is then force per unit
area in N m - 2, this unit being called the pascal (Pa). Energy is measured
in joule (J), which is 1 N m, and power in watt (w) which is 1 J s - 1.
186 SI units, physical constants, conversion factors
The ampere (A) is that current which produces a force of
2 x 10- 7 N m -1 between long straight parallel conductors 1 m apart in
vacuum. The volt (V) is then the potential difference such that a current
of 1 A dissipates 1 watt, and the coulomb (C) is 1 A s. The faraday
constant (F) can then be measured as the number of coulombs of charge
carried by one avogadro of electrons.
Standard gravity is 9.80 665 m s - 2 and a standard atmosphere is the
pressure produced by a mercury column 0.76 m high under standard
gravity at ODC; this is 101.325 kPa. The preferred unit is the bar, defined
as 100 kPa. Other units that are still in use are the calorie in various
forms, usually the thermochemical calorie of 4.184 J, and the
Angstrom, A, which is 10- 10 m. The word 'litre' is now regarded as a
special name for the cubic decimetre, with the recommendation that
neither the word litre nor its symbol, 1, should be used to express results
of high precision. This is because the litre was previously defined as
1.000028 dm 3 , this definition being rescinded in 1964.
Physical constants
Recommended values of the fundamental physical constants are
published by the International Union of Pure and Applied Chemistry.
Some of the values given in Pure and Applied Chemistry, 51, 1 (1979)
are:
Avogadro constant L 6.022045(31) X 1023 mol- 1
Boltzmann constant k 1.380662(44) x 10- 23 J K- 1
gas constant R 8.31441(26) J K- 1 mol- 1
Planck constant h 6.626176(36) x 10- 34 J s
Faraday constant F 9.648456(27) X 104 C mol- 1
mass of proton 1.6726485(86) x 10- 27 kg
charge on electron e 1.6021892(46) X 10- 19 C
mass of electron 9.109534(47) X 10- 31 kg
speed of light in vacuum 2.99792458(1) X 108 m S-l
permittivity of vacuum 8.85418782(5) x 10- 12 C 2 N- 1 m- 2
normal atmosphere 1.01325 x 105 Pa exactly
zero of Celsius scale 273.15 K exactly
standard acceleration of
free fall 9.80665 m s - 2 exactly
The notation used is that the figure(s) in brackets is (are) the standard
deviation of the last figure(s) quoted.
Sf units, physical constants, conversion factors 187
Conversion factors for units
1 amp = I A = I J s - I V - I 1 inch = 2.540 cm
1 atm = 101.325 kPa 1 joule = I J = I N m
1 cal = 4.184 J 1m 3 = 106 cm 3
1 coulomb = 1 C = 1 A s = 1 J V-I 1 micron = 1 p. = I p.m = 10- 6 m
0.30 debye = 10- 30 C m I newton = I N = 1 m kg s - 2
1 dyne = 10- 5 N lpascaI=1 Pa=INm- 2 =lm- I kgs- 2
1 electron volt = 1 eV = 1.6021 x 10- 19 J 1 pound (UK) = 1 Ib = 0.4536 kg
1 erg = 10- 7 J I torr = 1 mmHg = 0.1333 kPa
1 farad = 1 F = 1 A S V-I 1 volt = 1 V = 1 J A -I S - I
= 1 m - 2 kg - I S4 A 2 = 1 m 2 kgs- 3 A-I
1 gallon (UK) = 4.546 litre I watt = 1W = 1J s - I = I m 2 kg s - 3
= 1.2009 gallon (USA)
The Greek alphabet
A IX alpha N v nu
B
r
p beta
y gamma
-
0
~
0
xi
omicron
n
,
i\ (j delta 1t pi
E E epsilon P p rho
Z zeta 1: (J sigma
H '1 eta T T tau
e 0 theta Y u upsilon
I iota <l> 4> phi
K K kappa X X chi
i\ ). lambda 'I' psi
M p. mu n '"
OJ omega
188 SI units. physical constants, conversion factors
Summary of useful relations
cos 2 A + sin 2 A = 1
cos 2 A - sin 2 A = cos 2A
sin 2A = 2sinA cos A
sin (A±B) = sin A cos B±cos A sin B
cos (A ±B) = cos A cos B+ sin A sin B
eXeY = eX+ Y
a-X = 1jaX
if y = eX then x = In y
In x = loge x = 2.3026 log x = 2.30261og 10 X
log 1o lO-x = -x
b 1
ax 2 +bx+c = 0; x = --±- .j(b2 -4ac)
2a 2a
.
SIn x
x3
= x -3-'. +,- ...
XS
5.
n(n - 1)
(1 + xt = 1 + nx + 2! x 2 + ...
n(n -1)
(a+xt=an+nan-1x+ 2! an- 2 x 2 + ...
x2
f(x) = f(O) + xl' (0) + 2! f" (0) + ...
· f(x)
11m - () = 1·x~a I' (x) ·ff() () 0
1m ~() 1 a = 9 a = or CX)
x~a 9 X 9 X
z = z(x, y); dz = (!:)Y dX+(;;)x dy
dy dy du
dx du dx
SI units, physical constants, conversion/actors 189
(;:)y = -(;;)x (;~)%

f udv = uv - f vdu
e iy = cos y + i sin y
e- iy = cos y -i siny
Index
Abscissa, 12 Conservative field, 96, 136

Action, 133 Continuous function, 31, 32, 33
Angular momentum, 71, 91, 135 Convergence of series, 50-61
Angular velocity, 67, 91 Coordinates
Anharmonicity, 21 cartesian, 21
Approximation, numerical, 53, 55, 57 cylindrical polar, 121
Arc length, 1, 67, 118 plane polar, 21, 118, 150
Area rectangular, 21
by integration, 99, 115, 119 spherical polar, 121
differential element of, 117 Correlation analysis, 169
Argand diagram, 25 Correlation coefficient, 180, 182, 183
Argument of complex number, 26 Couple, 70
Asymptotes, 18, 62 Covariance, 171, 179, 183
Axes Critical damping, 154
principal, 70 Critical point, 43
rectangular cartesian, 21 Cross differentiation, 82, 97, 122, 144,
145
Curl of vector field, 97
Binomial series, 56 Curvature, 33
Boltzmann distribution, 48 Curve fitting, 60, 173, 184
Cylindrical polar coordinates, 121
Calculus of variations, 46, 129
Central limit theorem, 176
Centrifugal force, 67, 135 Damped harmonic motion, 153
Chain rule Damping, critical, 154
partial derivatives, 75, 84 Dependent and independent variables,
total derivatives, 39 10, 12, 48, 80, 171
Circle, 17, 116 Derivative
Complex conjugate, 25 mixed second, 75
Complex numbers, 10, 24, 149, 154 partial, 73
exponential form, 27 reciprocal of, 40, 74
Concave upwards, 34 second, 33
Conjugate, complex, 25, 95 total, 30, 31, 74
Index 191
Differentiable function, 31 Factorial, 22, 161
Differential Field
exact, 82, 144 electric, 65, 94, 152
total, 36, 78 gravitational, 65
Differential element scalar and vector, 95
of area, 117 First order reaction, 13
of volume, 120 Fluids, theory of, 127
Differential equations, 138 Force, 63, 64, 66, 70, 89, 152
exact, 140, 144 generalised, 137
first order homogeneous, 140, 142 Fourier transform, 163
linear, 140, 145, 147 Functional notation, 8
linear, homogeneous, 148 Functions, even and odd, 33
order and degree of, 140
separable variables, 140
Differentiation Gamma function, 22, 161
by substitution, 39 Gas
cross, 82, 97, 122, 144, 145 real, 59, 127
implicit, 36, 75 perfect, 6, 11, 18, 80, 86, 124
of integrals, 108 Gas constant, 6, 186
rules for, 34 Gaussian distribution of error, 166
Dimensional analysis, 5 Generalised dynamics, 132
Dipole moment, 71, 94, 152 Generalised momentum, 136
Direction cosines, 16, 91 Geometric series, 51
Discrete variables, 168 Gibbs equations, 83, 86, 88
Distribution function, pair, 127 Gibbs function, 12, 83, 88
Divergence of vector field, 97 Gradient of scalar field, 95
Dynamics Graphical methods, 12
generalised, 132 Gyroscopic motion, 91
Newtonian, 63
Heat, 66, 154
Ellipse, 17 capacity, 63, 85, 100
End point maximum, 42 of vaporization, 39
Energy Hamiltonian mechanics, 132
heat, 66, 154 Harmonic motion
kinetic and potential, 65, 132, 152 damped,153
Enthalpy, 8, 12, 100 forced, 158
Equations simple, 18, 150
cubic, 9 Helmholtz free energy, 12
quadratic, 8 Hyperbola, 17,62
second degree, 17 Hyperbolic functions, 55
Equilibrium constant, 10, 24, 57
Error Imaginary number, 25
in arithmetic mean, 172 root, 10, 24, 149, 154
normal distribution of, 166, 174 Implicit differentiation, 36, 75
random sign, 166, 176 higher order, 37
root-mean-square, 166, 168, 169 Impulse, 64
small-sample, 171 Independent variable, 10, 12, 48, 80, 171
systematic, 165 Indices
Euler-Maclaurin theorem, 107, 109 laws of, 3
Euler's theorem, 86 Miller, 17
Euler's equation, 132 Inertia
Exact differential, 82, 144 moment of, 7, 68, 134
Exponential functions, 1,27, 38 principal moments of, 70
Extensive and intensive variables, 86 product of, 69
192 Index
Inflection, point of, 41 Modulus of complex number, 26
Integration, 98 Molecular properties, 48, 66, 68, 70, 72,
area by, 99, 115, 119 94,134,153
by partial fractions, 105 Moment of a force, 66, 70
by parts, 102, 120 Moment of inertia, 7, 68, 134
by substitution, 104, 120 principal, 70
by using polar coordinates, 119 Momentum
curve length by, 123 angular, 71, 91, 135
multiple, 107, 117, 124 linear, 64, 71, 125
replacing summation by, 113 Multiple integration, 107, 117, 124
volume by, 100, 120, 122
Intercept, 13, 14 Nabla operator, 95
Inversion of a series, 56, 58 Napierian logarithm, 38
Newtonian mechanics, 63
Joule-Thomson coefficient, 85 Newton's method, 10, 53
NMR,92,164
Kinetic energy Normal distribution of error, 166, 174
angular, 68, 72 Numerical quadrature, 112
linear, 65, 132, 152
Kinetic theory of gases, 124 Operator
Kinetics, reaction, 13, 44, 139, 146 differential, 34
nabla, 95
Lagrange's method of undetermined vector, 95
multipliers, 45, 174 - Ordinate, 12
Lagrangian mechanics, 132 Orthogonal, 16,92,95
Laplace transform, 160
Least squares, method of, 13, 176, 184 Parabola, 17,21
error in both variables, 178, 183 Partial derivatives, 73
straight line, equations for, 179 chain rule for, 75, 84
weighted, 183 reciprocal of, 74
Leibnitz's Theorem, 108 relations between, 84
Length Taylor's theorem in, 88, 170
of arc, 1, 67, 118 Partial fractions, 105, 162
of curve, 123 Partition function
L'Hopital's rule, 61, 120 rotational, 113
Limit translational, 112
calculus, 30 Parts, integration by, 102, 120
expansion in series, 63 Path, dependence on, 76, 78, 83, 122
L'Hopital's rule for, 61, 120 Perfect gas, 6, 11, 18, 80, 86, 124
of sum or product, 35, 40 pH, 42
Line integral, 122 Phase rule, 11
Logarithm, 38 Planck's constant, 6, 186
Logarithmic series, 56 Polar coordinates
cylindrical, 121
Maclaurin's theorem, 54 plane, 21, 118, 119, 150
Maxima and minima, 41 spherical, 121
calculus of variations, 129 Potential energy, 65, 96, 133, 152
subject to constraint, 45 Power series, 51,60
Maxwell's relations, 83, 86 Precession of gyroscope, 91
Mean Principal axes of molecules, 70
arithmetic, 166 Probability, 23, 25, 49, 168, 174
error in, 172 Projectile, motion of, 21
Miller indices, 17 Propagation of error, 170, 181
Index 193
Quantity calculus, 5 Stirling's theorem, 23, 49
Quantum theory, 72, 95, 136, 162 Stokes's law, 4
Straight line
Radian, I, 67 equation of, 13
Random sign error, 166, 176 least-squares, 176, 179
Reaction kinetics, 13, 44, 139, 146 Substitution
Real and imaginary parts, 25, 149, 155 differentiation by, 39
Rectangular coordinates, 21, 121 integration by, 104, 120
Reduced mass, 68, 134 Successive approximation
Relaxation time, 139 Newton's method, 53
Residual, 178, 184 substitution in series, 59
Resonance, 159 Symmetry, circular, 21
Resultant, 89 Systematic error, 165, 166
Root-mean-square error, 166, 168, 169, Taylor's theorem, 51,61, 109, 148
170, 181 in partial derivatives, 88, 170
Root of an equation, 8 Torque, 70, 94
imaginary, 10,24, 149, 154 Thermodynamic function of state, 83,
Rotation 122
dynamics, 72 Thermodynamic properties, 12, 66, 83
sign of, 21 Thermodynamic relations, 83, 85, 87
Time constant, 139
Scalar, 89 Total derivative, 30, 31, 74
product, 92 chain rule, 39
Series, 50-63 Total differential, 36, 78
binomial, 56 Transformation, integral, 160
convergence of, 50--61 Triangular graph, 15
cosine, 3, 55 Trigonometric identities, 27, 188
exponential, 3, 38, 55 Turning point, 41
geometric, 51
inversion of, 56, 58 Units, SI, 4, 185
logarithmic, 56
power, 51, 60 Variables
sine, 2, 55 dependent and independent, 10, 12,48,
sinh and cosh, 55 80, 171
SI units, 4, 185 extensive and intensive, 86
Similar triangles, 14 Variance, 171
Simple harmonic motion, 18, 150 Variations, calculus of, 46, 129
damped, 153 Vector, 89
forced, 158 components of, 90
Simultaneous equations, 11, 45 differentiation of, 90
Single-valued functions, 32 localized, 91
Solid angle, 2, 125, 128 operator, 95
Spectroscopy, 7, 152, 164 product, 92
Spherical polar coordinates, 121 Volume by integration, 100, 120, 122
Standard deviation, 168, 172, 173
Standard forms of integral, 106, 119 Weighting in least-squares analysis, 183
Stationary values, 41, 45, 129 Work, 65, 83, 96

P. G. Francis (Auth.) - Mathematics For Chemis

Uploaded by

Copyright:

Available Formats

P. G. Francis (Auth.) - Mathematics For Chemis

Uploaded by

Document Information

Original Description:

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

P. G. Francis (Auth.) - Mathematics For Chemis

Uploaded by

Copyright:

Available Formats

Mathematics for Chemists

LONDON NEW YORK

ISBN-13: 978-94-01()..8950-O e-ISBN-13: 978-94-1109-5552-3

Brhiala Library Cataloauill, ill PliblicatiOil Data

Francis. P. G. (Patrick Geoffrey), 1929-

1 Algebraic and geometrical methods 1

3 Differential calculus in three or more dimensions; partial

6 Differential equations 138

7 Experimental error and the method of least squares 165

Appendix SI units, physical constants and conversion factors;

This text is concerned with those aspects of mathematics that are

Algebraic and geometrical

1.1 Natural numbers

angle = (length of arc)/radius, (1.1a)

to have the same units on both sides. Thus if the pressure p is in

= 2.70 X 10- 47 kgm 2 •

p=pgh= 1 ; ; x9.81pf ~ x - - x12jn

= 2.033 x lOs Pa = 203.3 kPa.

1.4 Quadratic and higher-order equations

so that the original equation can be written as

1.5 Dependent and independent variables

1.6 Graphical methods

This is converted into linear form by taking natural logarithms of both

1.7 Some geometrical methods

1.7.1 Similar triangles

1.7.2 Triangular graph paper

Pythagoras's equation in three dimensions then follows because

1.7.4 Circle, ellipse, parabola and hyperbola

Having identified the type of curve represented by any particular

the asymptotes are perpendicular to each other, the hyperbola is said to

The standard forms of equation are useful when experimental data

Simultaneously, the horizontal velocity is u cos (X so that the distance

so that a = -g/(2u 2 cos 2 (X) and b = h = o. We therefore have

1.7.5 Plane polar coordinates

1.8 Factorials and gamma functions

It is conventional to define O! = 1 so that the general relation

which is a good approximation to the actual value of K at a high

1.10 Complex numbers

has no real roots. We can define i as the imaginary square root of -1

~C. The length OA is called the modulus IA Iofthe complex number A,

and replacing the exponent by the complex number x + iy gives

The application to complex numbers of the calculus techniques

2.1 Significance and notation

2.2 The calculus limit

dy = lim (jy. (2.1)

The alternative analytical approach is to define by as the change in y

2.3 Differentiation of simple functions

the first term of which is zero. Hence

The quotient y = u/v gives

dy = lim [(u + l5u - ~)/ I5x],

which may be written as

dy = lim 1 (v(u+I5U)-U _u(v+I5V)-v)

2.4 The use of ditTerentials; implicit ditTerentiation

and can then multiply both sides by dx to obtain

if y = uv then = udv + vdu,

Implicit differentiation may also be applied successively so as to

1.5 Logarithms and exponentials

for all x. (2.11)

f(a + h) = f(a) + hf'(a) + f"(a) + f"'(a) + . . . . (2.39)