Mathematics For Business, Finance, and Economics

2nd edition
1
i f ij i il^ 1 ■T3
■ ■ IH 11 1 B m VB
MkGHAEL WILKES INTER

Mathematics for
business, finance
and economics
Second Edition
E M. Wilkes
INTERNATIONAL
THOMSON
BUSINESS PRESS
INTERNATIONAL THOMSON BUSINESS PRESS

I(T)P ® An International Thomson Publishing Company
London • Bonn • Boston • Johannesburg • Madrid • Melbourne • Mexico City • New York • Paris
Singapore • Tokyo • Toronto • Albany, NY • Belmont, CA • Cincinnati, OH • Detroit, MI
Mathematics for business, finance and economics
Copyright © 1999 F. M. Wilkes
First published 1994 by Routledge
A division of International Thomson Publishing Inc.

I T P' The ITP logo is a trademark under licence
All rights reserved. No part of this work which is copyright may be reproduced
or used in any form or by any means - graphic, electronic, or mechanical,
including photocopying, recording, taping or information storage and
retrieval systems - without the written permission of the Publisher, except in
accordance with the provisions of the Copyright Designs and Patents Act
1988.
Whilst the Publisher has taken all reasonable care in the preparation of this
book the Publisher makes no representation, express or implied, with regard to
the accuracy of the information contained in this book and cannot accept any
legal responsibility or liability for any errors or omissions from the book or the
consequences thereof.
Products and services that are referred to in this book may be either
trademarks and/or registered trademarks of their respective owners. The
Publisher/s and Author/s make no claim to these trademarks.
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library
Second edition 1999
Typeset by LaserScript Limited, Mitcham, Surrey

Printed in China
ISBN 1-86152-241-X
International Thomson Business Press

Berkshire House
168-173 High Holborn
London WCIV 7AA
UK
http://www.itbp.com
Contents
Preface ix
Chapter 1 Straight lines and linear equations 1

1.1 Introduction 1
1.2 Co-ordinates 2
1.3 Straight lines: preliminaries 7
1.4 Identifying straight lines and equations 11
1.4.1 Obtaining the equation of a straight line from a literal
description of the relationship between the variables 12
1.4.2 Plotting a straight line with a given equation 12
1.4.3 Finding the equation of a line with known slope going
through a given point 14
1.4.4 Finding the equation of a line through two known
points 15
1.4.5 Establishing on which side of a line a given point lies 16
1.5 Two applications 18
1.5.1 Forecasting 18
1.5.2 Inventory control 18
1.6 Point-slope and general form 20
1.7 Straight lines and simultaneous equations 23
1.7.1 Solving simultaneous equations 25
1.8 Elementary row operations 32
1.9 Other linear systems in up to two variables 33
1.9.1 One equation in one unknown 34
1.9.2 One equation in two unknowns 35
1.9.3 Whole-number requirements 35
1.10 Linear systems of equations in more than two variables 36
1.11 Gaussian elimination 42
1.12 Identities 45
1.13 Eurther applications 47
1.13.1 Breakeven analysis 47
1.13.2 Market equilibrium 49
chapter 2 Linear inequalities 53
2.1 Introduction 53
2.2 Linear inequalities in one variable 55
2.2.1 Addition of a constant 55
2.2.2 Multiplication by a positive constant 55
2.2.3 Multiplication by a negative constant and reversal of
direction 56
2.2.4 Solution sets 56
2.3 Linear inequalities in two or more variables 60
2.4 Convex solution sets 65
2.5 Linear programming 67
Chapter 3 Matrices 77
3.1 Introduction 77
3.2 Some fundamentals 78
3.2.1 Labelling of matrices 78
3.2.2 Dimensions of matrices 78
3.2.3 The elements of matrices 79
3.2.4 Sub-matrices 79
3.2.5 The principal diagonal 80
3.2.6 Triangular matrices 80
3.2.7 Diagonal matrices 80
3.2.8 Zero matrices 81
3.2.9 Matrix notation 81
3.2.10 Equal matrices 82
3.2.11 Transposition 82
3.3 Addition, subtraction and scalar multiplication 85
3.3.1 Matrix addition 85
3.3.2 Subtraction of matrices 86
3.3.3 The identity element for matrix addition 87
3.3.4 Scalar multiplication 87
3.4 Matrix multiplication 89
3.5 Matrix inversion 97
3.5.1 The determinant 99
3.5.2 The adjugate matrix 100
3.5.3 The inverse 101
3.6 Simultaneous equations (i) 105
3.7 Rank 111
3.8 Higher-order systems: determinants and the inverse matrix 115
3.9 Simultaneous equations (ii) 119
3.10 Cramer’s rule 123
3.11 Concluding remarks 125
Chapter 4 Functions and turning points 126

4.1 Functions 126
4.2 Quadratic functions 132
iv Contents
4.2.1 One linear and one quadratic equation 142
4.2.2 Two quadratic equations 143
4.2.3 Inequalities 145
4.2.4 General form 147
4.3 Cubics and quartics 149
4.3.1 Cubics 149
4.3.2 Quartics 153
4.4 Polynomials 156
4.5 Rational functions 157
4.6 Other functions 163
4.6.1 Composite functions 163
4.6.2 Inverse functions 164
4.6.3 Continuous and discontinuous functions 167
4.6.4 Discrete functions 169
4.6.5 Homogeneous functions 170
4.6.6 Monotonic functions 171
4.6.7 Odd and even functions 172
4.6.8 Periodic functions 172
4.6.9 Absolute value functions 173
Chapter 5 Derivatives and optimization 175

5.1 Introduction 175
5.2 Slope and turning points 176
5.3 An approach to the derivative 177
5.4 The power function rule 180
5.5 Differentiating polynomials 183
5.5.1 Application to the marginal analysis of the firm 184
5.6 The product and quotient rules 189
5.6.1 The product rule 189
5.6.2 The quotient rule 190
5.7 The chain rule 192
5.8 The inverse function rule 196
5.9 Implicit differentiation 198
5.10 Higher-order derivatives 200
5.11 Local maxima and minima 202
5.12 Global maxima and minima 207
5.13 Concavity, convexity and points of inflection 213
5.13.1 Concavity and convexity 213
5.13.2 Points of inflection 215
f
Chapter 6 Functions of more than one variable 220

6.2 Linear functions of several variables 222
6.3 Quadratic functions 226
6.4 Slopes and first-order derivatives 229
6.5 Higher-order partial derivatives 236
6.6 Local maxima and minima 239
6.6.1 Conditions for a maximum 239
6.6.2 Conditions for a minimum 242
6.7 Saddle points 243
6.8 Stationary values: resume 247
Chapter 7 Constrained optimization 250

7.1 Introduction to constrained optimization 250
7.2 The method of substitution 256
7.3 Lagrange multipliers and equality constraints 263
7.3.1 Minimization problems 267
7.3.2 Interpretation of Lagrange multipliers 270
7.4 Sign-restricted variables 272
7.4.1 An interior optimum 273
7.4.2 A boundary optimum 274
7.4.3 A degenerate case 275
7.5 Lagrange multipliers and inequality constraints 289
7.6 Inequality constraints and sign requirements 296
7.7 The Kuhn-Tucker conditions and mathematical
programming 299
7.8 Economic application: multi-product monopoly 308
Chapters Integration 312

8.2 Rules for integration 314
8.2.1 The constant function rule 314
8.2.2 The power rule 315
8.2.3 The sum-difference rule 316
8.2.4 The constant multiple rule 316
8.3 Application to the marginal analysis of the firm 317
8.4 Differential equations 321
8.4.1 General considerations 321
8.4.2 First-order linear differential equations with constant
coefficients 322
8.4.3 Higher-order differential equations (special cases) 324
8.4.4 A first-order, second-degree example 327
8.5 Integration by substitution 329
8.6 The definite integral 333
8.7 Numerical integration 344
8.8 Concluding remarks 348
\
Chapter 9 Exponential and logarithmic functions 350

9.1 Exponential functions: introduction 350
9.2 The natural exponential function and its derivative 356
9.3 Integration of natural exponential functions 361
9.4 Natural logarithmic functions 366
VI Contents
9.5 The derivative of natural logarithmic functions 373
9.6 Elasticity 377
9.7 Natural logarithmic functions and integration 383
9.8 Integration by parts 386
9.9 Logarithmic and exponential functions to bases other than e 390
9.10 Aggregate sales curves 396
Appendix 401
A.l Numbers 401
A.2 Sets 402
A.2.1 Operations on sets 404
A. 3 Exponents 406
A.3.1 The product rule 406
A.3.2 The power rule 409
A.3.3 The quotient rule 409
A. 4 Absolute values 411
A. 5 Place value notation 412
A. 6 Precedence in arithmetic operations 415
A. 7 Scales of measurement 417
A. 8 Summation and product notation 420
A.8.1 Summation notation 420
A.8.2 Product notation 423
A.9 Sequences and series 423
Solutions to exercises 430
References and further reading 525
Index 527
Contents VII
»lObMn >ji< 4tiaui>^ Vv ^ 'j4’I F .V 259
.y *».<* 239
m>(Ur!%a?til h<\«i C.O .241 9
AKf i
V 243
»iwd; Toflkt »«iiKt o^ 4J»iuUwi^4 h:>' i yvoa * ‘'''‘*' i»,V 247
<W»F
_ r T ^:. ^^
•,r4 r ' ■•-• 250
lOK - I t . •«. .. - . ... ..., •( KiKiw^qA 2M>
’ ^ ,1^ i* ’.v* I KA 258
^04* r.*» ■‘- Â;r^'> v.v ; 2M
r*<)f •:Jj f.t.A 2^»7
•iKH f.i i . - Ji*.::.'-i.A., 270
iVM ‘4 17 A 272
«» ♦ T f ^ ^ . ■^*'^'A 275
4 ? ,4iu>W) •••,‘%ti / ' / 274
m 4 k ., . . r. Ht vhîjÂ. 275
*a- 7; ■ . K, -. 289 »
298 F4
•Hi rfrf- T n’ uS^

«u '‘•Mjir ^ at>:»iifTmi$K ' %,A 299
Otf . yt , i-- ^ M 5fJt
i«i *■ . tl»ilJ»fK)tvly«4v»’f i H.A
^ _ WklH <?«|| ,iut*tl?p^
312
oe^ MHu-.i'Xr. Hi ill
}J4
F» ikSu b.jn KHiAirvîfl t14
ppK(#V art if.-»: 911^
il5
rjkKd
i4 >18
• / I r ,, ti4* rrt ilN
S.) ««•»» (J Ann 317
fi.4 ■• 32)
14.) > m * III
1,^4-2 •!!(*»* rtuS:.'. '«•* -nilNi oonefiiM'
322
S.4l Tl4*v. nwci) 124
• 84 *1 A H ,4* .^p.^ 327
8 5 !nlrpi*«njt» > v*.. ^ 329
'"r*W 86 iUU At'iVlte i .r ^ ,v, ;. 333
8.7 Nufnerfc'iif iniA^ .j • 344
8.8 CttiucUidlitiK iMIiorHv 348
■^1'
, j-, CHop4®r 9 ExpofvonfidI an*J *6cj> X.jn %* 350.
9 1 f jn;)Ol»i»tkff ,' «■• M *r 350
;i 9JI l*h«, ntiloi*^ »**r jukI ‘t* cbnvAdw 358
9,2 Inl«J|l'k^ r?4l r||^J4iivt»lial 181
^ ■’I. - 94 fiMitmt ii.^- '’■*'^' ihh
a
UPtNWJli
I fl
.Ik
1
Preface
Mathematics for Business, Finance and Economics adopts a progressive

and user-friendly approach to mathematical methods and concepts used
in the social and management sciences. Use in the study of quantitative
approaches to management, finance, business studies and economics is
emphasized. The claim of being on the side of the student reader is not
unique, but this book has distinctive features designed to maximize the
accessibility of a fairly broad syllabus, and these are set out below:
1 To provide a thorough treatment of some of the most important

and relevant mathematical methods in the fields of management,
finance, economics and the social sciences in general.
2 To provide both academic depth and a wide breadth of coverage in
the area of constrained optimization.
3 To make the material accessible to the widest possible group of
students and to develop individual confidence and a positive
outlook in the use of the methods.
The principles and methods are described as far as possible in narrative

terms, and intuitive as well as technical explanations are offered.
Progression is graded throughout each chapter, and each chapter
contains numerous examples and end-of-section exercises.
Comprehensive worked answers are provided at the end of the book.
The approach has a number of other distinctive features. Differential
calculus, including functions of several variables, and integration are
first of all presented through polynomial and rational functions, with
exponential functions introduced at a later stage. The purpose of this
ordering is to allow the principles and essence of the methods to be
conveyed with a minimum of distraction.
Wherever possible, formality is eschewed and intuitive understanding
is emphasized. The level of difficulty is treated consistently, progressively
increasing throughout each chapter, then easing down again at the
start of the succeeding chapter.
The book places emphasis on constrained optimization, with detailed
discussion of both equality and inequality constraints and sign
restricted variables. The appendix provides self-contained basic material
and is designed to ensure that students have an essential and
reasonably homogeneous background for the main part of the text. A
summary of learning outcomes is given for each chapter.
Where applications assist in the development of the material they are
woven into the text, with further applications in separate sections.
When a new concept is introduced, although one or two terms only
will subsequently be used in the text, other terms that are used to
describe the concept are given. In conjunction with this feature, the
detailed index should enable reference usage.
Attention is drawn to the capabilities, usefulness and relative ease of
use of today’s mathematical software packages - in particular Derive.
The minimum background for students using the book is UK GCSE
mathematics. This is not saying a great deal these days, and, while
those coming from GCSE level will find much that is new and
challenging, the manner of approach has been designed for you, and
your sights can be raised accordingly. Those with additional or A-level
mathematics will, depending on the widely varying syllabuses, find
some familiar material, but there will also be much that is new in
either substance or approach, and the management, economics and
social science orientation will give new significance to, and insights
into, the methods presented.
The structure of the book has been designed to allow for flexible use
depending on course requirements, individual preferences of lecturers or
the degree of modularity in a degree programme. Consistent with the
progressive nature of the material, chapters are as self-contained as
possible.
There are several ways in which the book could be used in a
business or social science degree programme. Eor example, an
introductory module or the mathematics component of a combined
quantitative methods course might cover Chapters 2 to 5 and part of
Chapter 9. Not all parts of each chapter need to be covered or given
equal emphasis. There are ninety-one sections in the book and the
emphasis given to each section or subsection can be varied without
losing continuity. In a quantitatively oriented programme the book as a
whole would form a comprehensive syllabus for a full one-year course,
or two single modules spread over two teaching periods.
Whether you are a student on a specialist degree programme or are
taking quantitative options in this area, I hope that you will find this
book relevant, interesting and genuinely approachable throughout.
I welcome feedback from readers, and if you have specific comments
or suggestions for improvement I can be reached on e-mail at
F.M.Wilkes@bham.ac.uk.
E M. Wilkes
June 1998
Preface
CHART E R
1 Straight lines and linear

equations
This chapter introduces important concepts concerning straight lines

and linear equations. Sections 1.1 to 1.7 bring in fundamental ideas
and sections 1.8 to 1.12 build on these to introduce more advanced
concepts and methods. Sections 1.5 and 1.13 consider specific
applications.
LEARNING OUTCOMES
By the end of the chapter you should be able to plot lines with given
equations, express the lines in the most suitable form of equation and
find the equation of any line with two known points or other known
properties. You will have covered material useful in linear
programming, the calculus and other contexts and will have laid
foundations which will be built on in later chapters. You will also be
able to manipulate and solve systems of linear simultaneous equations
in two, three or more variables. You will have seen a number of linear
models, learnt of their use in business and economics, and seen how
the models can be adapted to changing circumstances.
1.1 The simplest ideas are usually the most valuable and most widely used.
Introduction This is only partly due to the fact that simpler concepts are likely to be
more widely known. It also results from the fact that basic techniques
are likely to be of more practical value than their complex cousins. This
principle is nowhere more evident than in the case of straight lines and
linear equations, which are widely used in the social sciences and in
management science and business studies. Among the variety of
applications that we will examine are breakeven analysis, depreciation,
stock control, forecasting, investment income generation, and market
supply and demand. Linear expressions occur frequently in these and
other contexts. But we begin with the representation of algebraic
expressions in diagrammatic form. It is of great value to be able to
show equations in an understandable way and also to have the option
of using algebraic techniques on a problem originally seen in
geometrical terms.
Straight lines and linear equations 1

1.2 The real numbers can be matched with points on the real line, a
Co-ordinates straight line of unlimited extent. First establish an arbitrary point to
represent zero and mark off a distance, also discretionary, on one side
of zero to represent one. With the real line laid out horizontally, points
to the right of zero correspond to positive numbers - again a
conventional choice. Once you have established the unit interval, the
positive integers follow as the unit distance is repeatedly added on the
right. By the same process the negative integers are established to the
left of zero. Rational numbers are located by subdivision of the unit
intervals, while the irrational numbers are located by geometric
construction. The real line with the construction for \/2 is shown in
Figure 1.1.
Figure 1.1
The real line contains all the real numbers, with no gaps unfilled. There
is a one-to-one correspondence between the points on the real line and
the real numbers which are the co-ordinates of the points. The real
line is a single co-ordinate axis which can be used to make
comparisons between numbers in terms of algebraic size, the larger
numbers being to the right. For example, 3 is to the right of 2, and —2
is to the right of —3. The line can also be used to show intervals (the
numbers lying between specified limits) and other subsets of the real
numbers. All of this work relates to a single dimension.
The usefulness of co-ordinates is greatly expanded when a second
real line is introduced at right angles to the original and passing
through zero. The two lines are co-ordinate axes and the resulting
system of rectangular or Cartesian co-ordinates allows all points in
two dimensions (the plane) to be uniquely located by two co-ordinates.
All points on the plane could be labelled using any intersecting pair of
straight lines as axes, but when the lines intersect at right angles we
speak of a Cartesian co-ordinate system. The x axis is sometimes called
the abscissa and the y axis the ordinate. The co-ordinates represent
an ordered pair of numbers in which the horizontal or x co-ordinate
value is stated first and the vertical or y co-ordinate value is stated
second. These are matters of convention rather than logical necessities.
The two axes do not have to be measured to the same scale or in the
same physical or monetary units. In Figure 1.2 the rectangular co¬
ordinate system is shown and points Pi, P2, P3 and P4 are indicated,
with the following co-ordinates:
Pi (2,1) P2(4,3) P3(-2,1) P4:(2,-3)
2 Mathematics for business, finance and economics

The first co-ordinate value is the distance to be moved from zero in the
X direction (taking account of sign) and the second co-ordinate
represents the vertical displacement from zero (against the y co-ordinate
axis). The co-ordinates will be written as {x, y), with subscripts as
needed to identify particular points. Results would not be invalidated by
switching the labelling of the axes or by reversing the positive and
negative sides of either axis.
Figure 1.2
Once you have identified points in the plane, three questions arise:
1 What is the straight-line distance between two points.?

2 What is the slope of the line connecting the points.?
3 What is the equation for the straight line passing through two
given points.?
>
The straight-line distance between two points with given co-ordinates

can be found with Pythagoras’ theorem. So the distance between Pi
and P2 is:
d=^(4-2)'+(3-l)'
= ^8

while the straight line distance separating the points P4 and P3 is
d=^[2-(-2)]2 + (-3-l)'
= ^32
In general, the distance formula for the straight-line distance between

two points in the plane with co-ordinates {x2, Hi) and (xi, yi) is
d = yix^-xif+iyi - yif (1-1)
Two special cases are included in equation (1.1). In the case where the
two points are on the same vertical line the formula simplifies to give
the difference between the y co-ordinates, and where the two points are
on the same horizontal line the distance is the difference in the x co¬
ordinates. The results for distances between the points are as follows:
Points Distance
P1P2 2.828427
P1P3 4
P1P4 4
P2P3 6.324555
P2P4 2
P3P4 5.656854
All lines other than vertical lines have a defined slope. The slope of a
straight line connecting two points is the ratio of the change in y
values to the change in x values. The slope of the line joining P2 and
Pi is
, 3-1
slope = = +1
while the slope of the P3 and P4 line is:
-3-1
slope = = -1
2-(-2)
If the co-ordinates of the two points are (xi, yi) and (x2, yj), then the
slope is
slope =
yi - yi
X2 - Xi
It does not matter which of the points is taken to be first - the result is
the same for distance and slope. For distance the differences are squared,
and for slope both numerator and denominator of the ratio are reversed
in sign if the order of the points is changed. In any slope calculation
there must be consistency as to which point is considered as first in
taking values for x and y. The slope of P1P3 is zero since there is no
change in the y value, while the slope of P1P4 is undefined as division by
zero would be involved (the line is vertical). Full results are as follows:

Points Slope
PiPz 1
P1P3 0
P1P4 undefined
P2P3 0.333333
P2P4 3
P3P4 -1
The slope of a line depends on the units of measurement on the axes.

For example, if the unit of measure along the y axis was originally
pounds sterling and if this is now changed to pence, with no change in
the units of measure along the x axis, the slope of all lines is increased
a hundredfold. Slope is not a dimensionless concept, which is why in
some applications (such as demand curves) it can be better to work in
terms of proportionate changes in the variables rather than the absolute
changes involved in measures of slope. The two co-ordinate axes divide
the plane into four sections called quadrants, labelled as shown in
Figure 1.3.
Figure 1.3
2 1
3 4
The first quadrant, where neither variable is negative, is often called the
positive quadrant. In the first and third quadrants the two variables
have the same sign, while in the second and fourth quadrants the signs
are opposite. Points in three dimensions can be identified uniquely if a
third axis, z, is introduced at right angles to the x, y plane. This is
usually shown as in Figure 1.4.
Note that there is a right angle between each pair of axes - as at the
cbrner of a cube. The co-ordinate system divides three-dimensional
space into eight orthants. Practical wbrk is usually confined to the
non-negative orthant. Although manual sketches in three dimensions
can sometimes be worthwhile, precise geometry is not normally
attempted by hand. Wire-frame diagrams of functions in three
dimensions can be produced by software such as Derive, Maple or
Mathcad, the use of which can enhance understanding of general

Figure 1.4
concepts and particular functions. Points in three space have co¬

ordinate triples (x, y, z). The Pythagorean distance formula applies, so
the distance, d, between two points P2 and Pi in three dimensions can
be thought of as the corner-to-corner diagonal through a cube and
where d is given by
d = \l ix2 - xi)^ + (1/2 -yif + (Z2 - Zif
For example, the straight-line distance between the points
Pi ;(8,9,10) and P2 : (6,5,4)
is given by
d = -^(8 - 6)2 + (9 - 5)^ + (10 -
= V4+16 + 36
= ^56 7.48
The concept of slope in three dimensions is more complicated in that it

is necessary to specify the direction of movement. We will take up this
question in Chapter 6.
Exercises 1.2
1 Given the following points:
Pi: (2,5) P2:(5,2) >3: (3,-2)
P4:(-2,-2) P5:(-1,3)
find the distance between the following pairs of points:
(i) Pi and P2 (ii) P2 and P3 (iii) Pi and P5
(iv) Pi and P4 (v) P3 and P5
2 For the points Pi to P5 in question 1, find the slopes of the

straight lines connecting:
(i) Pi and P5 (ii) Pi and P2 (iii) Pi and P3
(iv) P5 and P2 (v) P4 and Pi
3 Suppose that the point P6: (2, 2) is added to the five points
above. What can be said about the slopes of the lines
connecting:
(i) P6 and P2 (ii) P6 and Pi (iii) P6 and the origin
4 Find the straight line distance between the following pairs of
points in three space:
(i) Pi: (9, 5, 8) and P2: (5, 3, 4)
(ii) Pi: (3, 6, 12) and P2: (8, 6, 6)
(iii) Pi: (7, -2, -8) and P2: (8, -1, 1)
1.3 The equation of a straight line can be written in several ways. The
Straight lines - most convenient of these is often the slope-intercept form:
preliminaries
y = mx + c (1-2)
The slope-intercept form is usually the preferred algebraic expression

of a straight line in applications where there is a clear subject of the
equation or dependent variable, i.e. when the value of y clearly
follows from the value of x. In equation (1.2) m is the slope and c is
the intercept - the point where the line cuts the vertical (y) axis. In
Figure 1.5 the line L has negative slope (m < 0) and positive intercept
(c > 0), while line M has positive slope (m > 0) and negative intercept
(c < 0).
Figure 1.5
Line L might represent the equation
y = —0.5x -I-10

while M might represent
y = 1.2x — 5
Horizontal lines (parallel to the x axis) have zero slope (m = 0) and

so have the form
y = constant (1-3)
So the straight line defined by the equation
y = 10
is a horizontal line ten units above the x axis, while
y = -5
is a horizontal line five units below the x axis. Even this most simple
form of equation has its uses. Vertical lines have an undefined slope
(informally referred to as infinite). For such lines the x value does not
change, and in the ratio that would otherwise define the slope, division
by zero would be involved. Since this is not permitted, the slope is
undefined. A vertical line can be written as
X = constant (1-4)
Thus X = —10 defines a vertical line that cuts the x axis ten units to
the left of the origin. Equations (1.3) and (1.4) themselves have special
cases of interest. The lines defined by
y = 0 and x = 0
are the equations of the x axis and the y axis, respectively. Equation
(1.2) has a further important special case where the intercept c is zero,
so the line goes through the origin. An equivalent way of expressing
this is to observe that when c is zero x and y always remain in the same
proportion, the factor of proportionality being m - as evidenced by the
fact that equation (1.2) can now be written as
In the general case of a straight line in slope intercept form:
y = mx + c
there are two parameters, m and c, each of which can be

independently selected and which together uniquely identify a
particular straight line. If the value of c is varied while m remains
constant a family of lines is generated, with the common value of
slope as the familial characteristic. Some members of the family of lines
given by m = +0.5 are shown in Figure 1.6.

Figure 1.6
If y represents costs and x represents output level, then a family of lines

defined by slope is created if fixed costs, c, are changed while unit
variable cost, given by m, remains the same. Note that lines belonging
to the same family defined by slope are parallel. Thus the straight lines
defined by
y = 4:X+10
y = —7 + 4:X
are parallel. A family of lines can also be defined by the common

characteristic of passing through a given point. Such lines are
concurrent. Figure 1.7 shows members of the family of lines passing
through (0, c).
One of the lines passing through (0, c) is the y axis itself (with the
equation x = 0). In the context of a linear cost function, concurrent
lines would be generated if fixed costs, c, were constant while unit
variable cost, m, changed. Note that it is not necessary that the
common point is on an axis for the lines to be concurrent. The point
could be anywhere in the x, y plane. As we have seen, lines which
belong to the same family defined by slope are parallel. In contrast,
perpendicular lines intersect at right angles and, except for the case of
horizontal and vertical lines, have slopes of m and — 1/m, respectively
(where m ^ 0). Perpendicular lines are also described as normal or
orthogonal. The equations of two perpendicular straight lines can be
written as
y = mx + c
and
y = —x + k
m

Figure 1.7
where c and k are the y intercepts of the lines. So for example:
y = 2x — 4l
and
y = 3 — 0.5x
are perpendicular. These lines are shown in Figure 1.8.

Note that i/ = 3 — O.Sxis not the only straight line perpendicular to
y = 2x — 4:. Any straight line with equation in the form
y = —0.5x + k
is perpendicular to the line y = 2x — 4. So the lines
y = 100 — 0.5x
and
y = —0.5x
are also orthogonal to y = 2x — 4 (and to any line having a slope of 2).

Similarly, any line the equation for which satisfies
y = 2x + c
is orthogonal to the line y = 3 — 0.5x. Another way that orthogonality

can be expressed is that the product of the slopes of perpendicular lines is
minus one (neither being vertical).

Exercises 1.3
1 What can be said about the slope of the straight 1 ines:
(i) y = 3x + 5 (ii) y == 10-0.5X
(iii) y = 50 (iv) y = \/2x (v) x = 25
2 For the ollowing straight lines:
(i) / = 5x + 4 (ii) y == 4x + 5 (iii) y = 1 Ox + 4
(iv) y = 5x- 17 (v) y == -x +4
(vi) y = 5x (vii) y = 8x
identify those lines which are concurrent and those which are
parallel
3 Which Df the following lines are orthogonal:
(i) y = 5x+ 10 (ii) y == 0.2x-0.1
(iii) y = 20-0.2> (iv) y == —5x +6 (vi) y = —5x
1.4 To use straight lines effectively some basic skills are necessary. In
Identifying straight relation to an individual straight line these include:
lines and equations
• obtaining the equation of a straight line from a literal description of
the relationship between the variables:
• plotting a straight line with a given equation;

• finding the equation of a line with known slope going through a
given point;
• finding the equation of a line through two known points:
• establishing on which side of a line a given point lies.
1.4.1 The starting point for building models in business applications is often a
Obtaining the equation literal description of part of the overall problem and it is important to
of a straight line from a be able to make use of information in this form. For example suppose
literal description of the that the following statements are made;
relationship between
(a) Total costs are the sum of fixed and variable costs. Variable costs
the variables
accrue at the constant rate of £6 per unit produced and there are
inescapable costs of £10 000.
(b) The maximum price that can be charged for the product is £500
less lOp for each unit sold.
(c) Sales revenue is price times quantity. The market currently
determines price at £10 regardless of the number of units that we
sell.
These statements correspond to linear relationships described by the

following equations:
(a) y = 10 000 + 6x
where y is total costs and x is production volume:
(b) y = 500-0.lx
where y is price and x is quantity sold:
(c) y = lOx
where y is sales revenue and x is volume of sales.
1.4.2 A sketch of relationships is a great aid to understanding and a good

Plotting a straight line way to prompt useful questions. A unique straight line is identified if
with a given equation any two distinct points on the line are known. The points are
connected and the line is extended beyond each of them. Given the
equation of the line, the points that are usually most convenient to find
are the intercepts with the axes. We will encounter one exception, but
consider for the moment the line with equation
y = —4x + 20
The point where the line crosses the y axis is easiest to find. This is
done by setting x = 0 in the equation and solving for y. The result is
y = 20 and the line is known to pass through the point (0, 20). Next,
the X intercept is identified by setting y = 0 and solving for x. The
result is
0 = -4x + 20

So that:
4x = 20
i.e.
X= 5
So the point (5, 0) is on the line. Note that the intercept with the x axis
is a solution of the equation
mx + c = 0
and the x intercept therefore occurs at the point x = —cjm. Once you
have identified two points the straight line passing through them can be
drawn in. For the current example this is shown in Figure 1.9.
Figure 1.9
The exceptional case is where the straight line passes through the
origin so that x and y intercepts are the same. In this case a further
point is generated by putting a value for x into the equation, and the
point with the resulting co-ordinates is joined to (0, 0). For example, if
the line is
y = —4x
a convenient value of x is x = 1, for which the corresponding value of

y is -4. To identify the line the origin should be connected with the
point (1, -4). Note that lines which pass through the origin will have
points in only two of the four quadrants.

1.4.3 This is a useful skill where the rate of change is known and where a
Finding the equation of single observation of corresponding x and y values is available. Suppose
o line with known slope it is known that a line has slope +2 and passes through the point (3,
going through o given 5). The coefficients in the straight-line equation must fit the data given.
point Thus in the slope-intercept template
y = mx + c
the fact that the slope is 2 produces
y = 2x + c
so the one remaining unknown is the value of c. The co-ordinates of the

point on the line {x = 3, y = 5) inserted into the template mean that
5 = 2(3)+ c
5 = 6 -I- c
c = —1
So the equation of the line is
y = 2x — 1
For further practice we will find the equations of:
1 the line with slope —3, passing through the point (3, 8);
2 the line with zero slope, passing through the point (4, 5):
3 the line with undefined slope passing through (5, —6).
In example (1), the equation must take the form
y = —3x -f c
and, given that the point (3,8) must satisfy the equation, c must be
such that
8 = -3(3)-Fc
so that
c= 8 9
= 17
and the full equation is therefore
y — —3x +17
In example 2 the slope is zero, so the equation of the line must read
y = 0x + c
in which the value of x is immaterial (as the line is horizontal). With

the given point (4, 5) the value of c must be such that
5 = 0(4) -f- c
c= 5

so the equation of the line is simply:
y = 5
In example 3 the line is vertical. We can get at the equation of the

line by realizing that for vertical lines it is the value of y that is
immaterial (x will always be the same). In other words,
X = Oy + c
which with the given point (5, —6), means that
5 = 0( — 6) + c
c= 5
so
X = 5
1.4.4 If two points on the straight line are given the co-ordinates of these
Finding the equation of points inserted into the straight-line equation produce simultaneous
o line through two equations in m and c - one equation corresponding to each point. The
known points equations are then solved for m and c. Methods for solving
simultaneous linear equations will be discussed later; for now we will
consider two examples solved by elimination. The problem is: what
line is determined by the points (2, 13) and (—1, 4)?
The two parameters m and c must fit the data given. Thus with
y = mx + c
the two points produce, respectively,
13 = m(2) + c
that is
13 = 2m + c
and
4 = m( —1) + c
that is
4 = —m + c
If, term by term, the second equation is subtracted from the first c
disappears, leaving
(13 — 4) = [2m — ( — m)] + (c — c)
which simplifies to
9 = 3m
m = 3

The value m = 3 can now be inserted into either of the equations, with
the result that c = 7. The equation of the line is therefore
y = 3x +7
Consider another example. What is the equation of the straight line

passing through (3, 9) and (-1,-15).^ The equations produced are
9 = 3m + c
and
-15 = -m + c
subtraction gives
24 = 4m
m = 6
Substitution of this value into either of the original equations yields the
result that c = -9, so the equation of the tine is
y = 6x — 9
Note that this procedure is convenient in these circumstances, as the

coefficient of c will always be 1 in each equation.
1.4.5 This is important when the line represents a resource limitation - for
Establishing on which example the maximum amount of capital available - or a measure of
side of o line o given performance such as return for given risk in portfolio theory. We will
point lies usually want to know if the y co-ordinate of the point is greater or
less than that given by the equation. Thus given the point (1,8) and
the line
y = 4iX + 2
putting the x co-ordinate of 1 into the equation produces y = 6, so

the point is above the line. In contrast, the point (7, 27) is below the
line since the y value given by the equation when x = 7 is 30. The
two points and their relation to the line are illustrated in Figure
1.10.
In an economic context, suppose that x is output and y represents
total costs of production (when resources are used efficiently) within
the current technology. An output of seven units produced at a cost of
£2 7 would be super-efficient and must correspond to either an
improvement in production technology or a reduction in resource
prices.

Figure 1.10
Exercises 1.4
1 Find the equations of the following lines:
(i) slope = 4, passing through the point (5, 25);
(ii) slope = —5, passing through the point (18, 10);
(iii) horizontal, passing through the point (7, 10);
(iv) vertical, passing through the point (8, 4).
2 (i) For o particular straight line passing through the origin, if
X changes, then y changes by half the amount but in the
opposite direction. What is the equation of the line?
(ii) What is the equation of the line that passes through the
points (2, 10) and (—1,1)?
3 (i) What is the equation of the straight line passing through
the points (1,7) and (2, 1 1)?
(ii) For a given line, if x changes in value, then y changes by
the same amount in the opposite direction. The line passes
through the point (4, 6). What is the equation of the line?
4 (i) What is the equation of the straight line passing through
the points (2,-1) and (4, 9)?
(ii) What is the equation of the straight line passing through
the point (1,3) and which is orthogonal to the line:
y = 4 - 0.5x?

5 Given the straight line: y = 5x - 2 establish whether the
following points are on, above or below the line:
(i) (2,9) (ii) (3,13)
(iii) (-2,-11) (iv) (1,2)
1.5 In forecasting with simple linear regression a straight line of best fit
Two applications is calculated using observations of corresponding values of two
variables. The equation can be written as
1.5.1
Forecasting Y = a + bX
The intercept term, a, and the slope coefficient, b, are set at values
which together match the data as closely as possible. With X as the
independent variable, there should be grounds for supposing that
changes in X bring about changes in the dependent variable Y - there
should be a causative relation between the variables. For example, X
may represent income and Y expenditure, or X may be the outlay on
sales promotion, while Y represents total sales revenue. The
independent variable often represents time, either abstract time
(t = 0, 1, 2 ...) or calendar date (2000 etc.). To illustrate, suppose that
observations of the level of a company’s turnover (sales revenue) are
related to time in the following manner:
Y = 45 + 3.8X
where Y is turnover (in fmillion), and X = 0 represents 1987, X = 1

represents 1988 and so on. The regression equation can be used to give
a single-figure forecast of turnover in future years on the assumption that
the underlying relationship is unchanged. For example, to forecast a value
for 2001 X = 14 is put into the equation, resulting in a predicted value
of Y of £98.2 million. For 2002 the forecast value of Y passes the
hundred million mark for the first time at £102 million. Putting
X = — 3 into the equation would, assuming the relationship was valid
at that time, indicate that turnover in 1984 was £33.8 million. The
regression line is plotted in Figure 1.11.
If the regression equation is estimated on the basis of observations
from 1987 to 2000 inclusive, predictions for 2001 and 2002
represent extrapolation, as does estimation of the 1984 value.
Using the equation to generate values within the period 1987 to
2000 is interpolation. These cases are distinguished as there is
generally more confidence in the model structure being valid if the X
value stays within the range of sample values. The further outside
this range X goes, the less likely it is that the model structure still
applies.
1.5.2 Consider the classical model of inventory or stock control. In this

Inventory control model a single item of stock is withdrawn from inventory at a known
and constant rate. Suppose that a department has an opening

Figure 1.11
inventory of 525 reams of paper, which are used at the rate of nine
reams per day. Due to delay between ordering and delivery - lead-time
- orders for replenishment of stock are placed when stock on hand falls
to 150 units. The relationship between inventory level, I, and the
number of working days, t, is linear, with the intercept term being the
opening inventory and where the slope is given by the rate of
withdrawal from stock. Thus
I = 525 - 9t
is the equation of stock against time. The equation can be used to find
the number of days before an order for replenishment is placed. This is
t such that:
1 = 525 - 9t = 150
so that
9t = 375
t = 41.67
So the order should be placed during the forty-second working day. The
graph of inventory against time is shown in Figure 1.12.
Note that, ignoring any practical problems involving fractions of
days, a stockout would occur if lead-time exceeded 150/9 = 16.67
days. The level of stock which triggers a reorder is known as the
reorder level or reorder point (150 here).

Figure 1.12
Exercises 1.5
1 A company's profits in year t are forecast by linear regression
to be 7r(t) where: 7r(t) = 80 + 5t and t = 0 corresponds to
1992.
(i) What profit is given by the equation for
(a) 1994 (b) 1997
(ii) Assuming that the relationship was valid at the time, what
would be the estimate of the profit figure for 1989?
2 A company has an opening stock of 1000 units of an item,
the stock level being depleted by forty units per day's trading.
An order for replenishment of stock is placed when the level of
inventory is sufficient to cover six days' demand.
(i) Write down an equation showing stock level against the
number of days' trading.
(ii) Given the current policy, on what day should a
replenishment order be placed?
1.6 The point-slope form of equation can be written as

Point-slope and
y - yi = rn{x - xi) (1.5)
general form
where (xi, yi) are the co-ordinates of a specific point which it is known
lies on a line of slope rn. For example, suppose we require the equation
of the line of slope 2 passing through (1, 5). Putting the co-ordinates of
the point and an m value of 2 into equation (1.5) produces the
following result:
y~5 = 2{x-l) (1.6)

which could then be rearranged into slope-intercept form as
y = 2x + 3 (1.7)
Note that while the point-slope form is as in (1.5) the details as in

(1.6) will depend on the point selected, but these should rearrange into
the same slope-intercept expression. For example, suppose that instead
of (1, 5) another point on the same line, say (—5, — 7), had been given.
Insertion of the data into (1.5) produces
y-{-7) = 2[x-{-5)]
that is
y -\- 7 — 2{x 5) (1,8)
So, as before.
y = 2x + 3
The point-slope form arises from the fact that given one point (xi, yi)
on a particular line, any other unspecified point (%, y) on the same line
must be such as to result in the same slope as given by
X — Xi
Equation (1.9) is known as the gradient form of the equation of the

straight line through the point Xi, yi with slope m. The situation is
shown in Figure 1.13, where the slope, m, is (as always for straight
lines) the ratio of the change in y and the change in x between any two
points on the line.
Figure 1.13

The point-slope form (1.5) represents cross multiplication from
equation (1.9). Straight-line equations can also be written in general
form. This is
Ax + By + C = 0 (1-10)
so the line y = 2x + 3 would appear in general form as either (1.11a)

or (1.11b):
2x — y + 3 = 0 (1.11a)
-2x + y-3 = 0 (1.11b)
In (1.10) no one variable is singled out for special treatment as the

dependent variable (the subject of the equation). One advantage of
general form is that it extends readily to linear expressions in more
than two variables. For example, a linear relationship between three
variables x, y and z can be written as
Ax + By + Cz + D = 0 (1-12)
which corresponds to a plane surface in three dimensions.

A variant of the general form is that in which all of the variable
terms are on the left-hand side (LHS) of the equation and the constant
term is on the right-hand side (RHS), as in
ax + by = c
This is often the way that a linear equation arises, for example as a
constraint. The equation written in slope-intercept form as
y = 2x + 3
would appear in one of the following ways:
2x — y = —3
—2x + y = 3
If the equation is presented in this form it is less obvious on which

side of the line a particular point lies. For example, with the line given
by
—2x + y = 3
does the point (—4, — 6) lie above or below the line.? The point is below
the line since when x = — 4 is put into the equation the resulting value
of y is —5. Since this is greater than the y co-ordinate of —6, the point
lies below the line. Equations in which the constant term and no
variable terms are on the right-hand side often occur in problems with
resource limitations. For example, suppose that x and y represent the
number of units purchased of each of two types of security, the prices of

which are £5 and £4 per unit, respectively. If a budget of £2000 is to
be invested, the linear equation that results is
5x + 4:ij = 2000
In this context, a point above the line represents expenditure over

budget, while a point below the line would correspond to an
underspend.
Exercises 1.6
1 Use the point-slope form to express the equations of the
following lines;
(i) slope 3 passing through the point (5, 8);
(ii) slope -6 passing through the point (8, 12);
(iii) slope —0.5 and y intercept at —20;
(iv) a 45° slope passing through the origin.
2 Express, in general form, the equation of the straight line that
passes through the points (5, 5) and (-5, 35).
3 A straight line with slope 4 passes through the point (5, 25).
Express the equation of the line in:
(i) point-slope form;
(ii) slope-intercept form;
(iii) general form.
4 For the line given by: —5x + 2y = 20, is the point P(8, 32)
located on, above or below the line?
1.7 For two straight lines, L and M, exactly one of the following
Straight lines and relationships holds:
simultaneous
1 L and M have all their points in common;
equations
2 L and M have none of their points in common:
3 L and M have exactly one point in common.
These mutually exclusive cases are graphed in Figure 1.14. In Figure

1.14(a) L (dashed line) and M (solid line) are superimposed. For
practical purposes the lines are the same, although their equations may
arise in such a way that this is not obvious at the outset. For example,
the equation of the line L might be
L: y = 7.5 - 0.25x
while the line M might arise in the following form:
M: 2x + 8^ — 60 = 0
Of course, it must be possible to convert, by permissible operations, the

expression for L into the expression for M and vice versa.

Figure 1.14
In Figure 1.14(b) the lines are parallel. In slope-intercept form the m
value is the same but the c values are different. The lines might be
L: y = —2,x 5
M : y = -2x + 7
In Figure 1.14 (c) the lines have different slopes. This is the important
distinction between this case and cases (a) and (b). Straight lines of
different slope must cross once and once only. For example, the
equations of the lines in Figure 1.14(c) might be
L: y = -2x + 10
M: 1/ = -X + 8
The one point that the lines have in common when both equations
are satisfied is (2, 6). This point of intersection corresponds to the
simultaneous solution of the equations defining the lines. For a unique
solution of linear equations to exist, the equations have to be both
independent and consistent. Two equations are independent if it is
not possible to obtain one equation from the other by multiplication
through by a non-zero constant. Thus cases (b) and (c) represent lines
with independent equations. Consistent equations are those which can
be simultaneously satisfied. Thus cases (a) and (c) represent lines with
consistent equations. Clearly then, only case (c) corresponds to
equations which are both independent and consistent. Table 1.1
summarizes the properties of the three cases above.
Table 1.1
Case Independent Consistent No. of solutions
(a) * infinite
* 0
(b)
* * 1
(c)
1.7.1 There are several ways in which systems of linear equations can be
Solving simultaneous solved. At this point we will introduce two simple methods:
equations substitution and elimination.
Substitution The method will be introduced by solving the following simultaneous

equations:
3x + y = 11 (1.13)
X + 2y = 12 (1.14)
First express equation (1.13) with y as the subject of the following

equation:
y = ll-3x (1.15)

Now substitute the right-hand side of equation (1.15) for y in
equation (1.14). The result is
3C-f2(ll - 3x) = 12
X -f- 22 — 6x = 12
and therefore
-5x = -10
X = 2
Now substitute 2 for x in equation (1.15). The result is:
y = ii-3(2)
y = 5
The solution of the simultaneous equations (1.13) and (1.14) is the

unique pair of values x = 2, y = 5. This outcome could equally well
have been found by using equation (1.14) to express x in terms of y:
X = 12 — 2y
which when substituted into equation (1.13) produces

3{12-2y) + y=n
so that
36 - 6y + y = 11
36-5y = 11
-Sy = -2S
and therefore
y=5
and from either (1.13) or (1.14) x = 2.

Consider another example in the use of substitution. Suppose the
equations are expressed in general form as
7x + 4:y — 51 = 0
—3x + 7y + 4:8 ==0
From the first equation
4y = 51 - 7x
y = 12.75 - 1.75x
which when substituted into the second equation gives
-3x + 89.25 - 12.25X + 48 = 0
so:
15.25X = 137.25
X =9

and y emerges as
i/= 12.75 - 1.75(9)

y = -3
Elimination In elimination one equation is multiplied through by a constant such

that the coefficients of x or y become the same in each equation. Thus
if equation (1.13) is multiplied throughout by 2 the result is
6x + 2y = 22
If equation (1.14) is subtracted from the modified equation (1.13) the

terms in y will cancel, eliminating y, and one equation in x alone
results. In detail;
6x + 2y = 22
X + 2y = 12
5x = 10
X = 2
Substitution of x = 2 into either equation produces y = 5. The result

could equally well be obtained by multiplying equation (1.14) by 3 to
give equal coefficients of x in each equation; x would then have been
eliminated as follows:
3x + y = 11
3x + 6y = 36
-5y = -25
y = 5
In this case the inconvenience of negative values following subtraction

could be avoided by taking equation (1.13) from the modified equation
(1.14). The two equations are graphed in Figure 1.15, and the solution
values of x and y are indicated.
Substituting x = 2 into either equation is equivalent in geometric
terms to moving up the line x = 2 until it hits either equation and then
moving parallel to the x axis to find the value of y as shown by the
arrows. The figure shows that it does not matter into which equation
X = 2 is substituted or whether i/ or x is found first. Now consider four
examples using substitution and elimination. These will be followed by
an illustration from financial management.
Example I 5x — 2y = 26
—X — 4i/ = 8
Here, substitution is suggested by the second equation, in which x

appears with unit (negative) coefficient. Rearrangement produces
X = —4i/ — 8

Figure 1.15
5{ — 4y — 8} — 2y = 26
so that
-20y - 40 - 2i/ = 26
—22y= 66
therefore y = —3, and from the rearranged second equation
x = -4(-3)-8 = 4
In the light of the value of y here, recall that there is nothing to

prevent negative values from arising in the solution.
Example 2 3x + 8y = 64
—2x + 2y = 16
Elimination is convenient in this case. Multiplication of the second

equation by 4 produces the following result:
3x + 8y = 64
—8x + 8y = 64
llx = 0
so X = 0. Insertion of this value into either equation produces y = 8. In

this case the intersection of the lines represented by the equations
occurs on the y axis.
Example 3 3x + y = 11
6x + 2y = 24

Multiplying the first equation by 2 and subtracting the second gives
6x + 2y = 22
6x + 2y = 24
0 = -2
This impossible requirement to equate two different constants means

that the equations are inconsistent - they have no simultaneous
solution. The lines represented by the equations are parallel.
Example 4 3x + y = 11
— 12x — 4i/ = —44
Using substitution in this case, from the first equation
y = 11 — 3x
insertion into the second results in
-12%-4(11 - 3%) = -44

— 12% — 44 + 12% = —44
therefore:
0 = 0
which means that the two equations are not independent (the second is
—4 times the first). So there are an infinite number of solutions in this
case, but not anything will do in terms of % and y values; ‘most’ points
are not solutions. Only points having co-ordinates satisfying the
relationship y = 11 — 3x will fit. All other points are off the line. Now
consider a practical illustration of simultaneous equations in personal
financial management.
Financial management On retirement a manager expects to receive a pension plus a lump sum
example of £60 000. He wants to obtain an annual income of £7000 by
investing the lump sum in loan stock yielding 13% and a deposit
account yielding 9%. What amounts should be invested in each case.?
The problem produces a pair of simultaneous equations which can be
solved by elimination or substitution to give the required amounts. If %
is the sum invested in loan stock and y is the amount placed on
deposit, then, assuming nothing is held as cash,
x + y = 60 000
The requirement of an annual income of 7000 produces the following

equation:
0.13% + 0.09z/ = 7000
In this example substitution is more convenient. From the equation

governing the total sum to be invested we obtain
y = 60 000 — %

which on substitution into the annual income equation produces
0.13x + 0.09(60 000 - x) = 7000

0.04X+ 5400 = 7000
0.04x = 1600
and so
X = 40 000
so
y = 20 000
Two-thirds of the sum available should be placed in the loan stock

and one-third put on deposit to generate the required return. Practical
considerations not represented in this model, such as brokerage or
management charges, could be included in an expanded model. But
even this simple form has potential. For example, nothing stands still in
the real world; circumstances change - for example interest rates.
Suppose in the context of this example that interest rates are reduced
by one percentage point. What should be the response of our investor,
given the stated objectives.? First, note that if no changes are made the
investor’s income will suffer, since
0.12(40 000) + 0.08(20 000) = 4800 + 1600 = 6400
To retain the original level of income, the equations now become
x + y = 60 000
0.12x + 0.08z/ = 7000
and, again using substitution, from the first equation,
y = 60 000 — X
insertion of this relationship into the second equation produces
0.12x + 0.08(60 000 - x) = 7000

0.12x + 4800 - 0.08x = 7000
from which
0.04x = 2200
4x = 220 000
X = 55 000
and the investor is faced with a fairly radical adjustment of the

portfolio, with £55 000 to be invested in loan stock and just £5000
remaining on deposit.
Now suppose that interest rates are reduced by another percentage
point. The equations are now
X + y = 60 000
0.1 lx+ 0.071/ = 7000

Using elimination, multiplying the first equation by 0.07 results in
0.07x4-0.07y = 4200
O.llx+ 0.071/ = 7000
-0.04x = -2800
so that
4x = 280 000
X = 70 000
which requires that
y = -10 000
While the algebra does not rule out this outcome, an implicit
assumption in the use of this model would be that neither x nor y is
negative. So unless the investor has an unlikely facility to borrow at
7%, the income target can no longer be achieved even if the entire
lump sum was invested in the potentially more lucrative (and higher-
risk) security.
Exercises 1.7
Solve the following simultaneous equations using elimination:
(i) 4x + 3y = 110
2x +y =:50
(ii) 5x — y =:20
3x + 2y = 12
(iii) 6x — 5y = 16
4x + 2y = 0
(iv) 0.4x + 5y = 7
2x- 0.5y = 4.4
Solve the following simultaneous equations using substitution:
(i) 2x + y =:20
7x + 3y = 66
(ii) X- 1 ly = 40
2x- 7y = 35
Solve the following systems using the method of your choice:
(i) y = 4x-35
2x + y = 25
(ii) X + 25 = 3.5y
x = 4.5y-33
4 Where possible, find solutions to the following systems of
simultaneous equations. Where there is not a unique solution,
say whether there are ony solutions and give the reason why
the situation occurs.
(i) 0.4x + 0.2y = 3.6
0.1x + 0.3y = 2.9

(ii) -0.5x + y = 9.5
x-2y = -19
(iii) 2x - y = -3
-2x + 0.5y = -0.5
(iv) X - 2y = -9
-2x + 4/ = 19
1.8 It is often useful to change an equation into a more convenient form

Elementary row without altering the information it contains, or to combine or
operations interchange equations in such a way that the solution to the system as
a whole is unaffected. These permissible changes are called elementary
row operations and can take the following forms:
1 Multiplication throughout of any equation by any non-zero

constant (either positive or negative).
2 Addition to any equation in the system of a constant multiple of
any other equation in the system.
3 Interchange of the positions of equations in the system.
Repeated application of (2) means that any linear combination (a

weighted average) of equations in the system can be added to any
equation in the system. In fact (1) can be seen as an instance of (2)
where multiples of an equation are added to itself. To see that the
elementary row operations are legitimate, consider some examples. The
following system:
3x + y = 11
X + 2y = 12
has the solution x = 2, y — 5. So does the system
3x + y = 11
7x + 4:y = 34
Here the second equation of the original system has been replaced by
the second equation plus twice the first equation. The following system
also has the solution x = 2, y = 5:
X — 3y = —13
X + 2y = 12
In this case the original first equation has been replaced by the first less
twice the second equation. As a final example, the following system:
X — 3i/ = —13
7x + 4y = 34
also solves ior x — 2y = 5. This case uses both earlier substitutions.

The elementary row operations generate, as it were, a family of
systems of two linear equations having the same point as the solution.

Any two distinct equations passing through the point (2, 5) can be
operated on by elementary row operations to produce all linear
equations passing through (2, 5). The situation is graphed in Figure
1.16, which shows the four equations used.
Figure 1.16
Note that any three or more of these equations in two variables have
the solution (2, 5). In general, for a system of three equations in two
variables to have a solution any one of the equations must be
obtainable from the other two by elementary row operations.
Elementary row operations are implicit in the simplex method of linear
programming, which is closely related to Gaussian elimination. Further
operations on equations are possible. For example, both sides of an
equation can be raised to the same power. Consider the following:
2x — 1 = x + 2
The same value of x (3) will satisfy
(2x- 1)^ = ix + 2f
but squaring has introduced another ‘solution’, x = —1/3, which is not

valid for the original linear equation. If the problem-solver is moving
from the more complicated quadratic to the linear form a possible
solution will be lost. However, the ‘lost’ solution can be retrieved once
the other solution has been obtained.
1.9 Here we consider the cases of:

Other linear
• one equation in one unknown:
systems in up to
• one equation in two unknowns:
two variables
• whole-number requirements.

1.9.1 The simplest case of all is one linear equation in one unknown, but
One equation in one even this leads to an important point. Suppose that in a given case
unknown
2x + 7(x - 4) + 3(6 - 2x) - 20 = 20
which simplifies to
3x-30 = 0
X = 10
Here the single equation has a single solution. Now suppose that
(5x - 103) + 9(2x + 5) + 7(10 - 3x) = 2(6 + x)
Simplification produces
2x -i" 12 = 12 “1- 2.x
with the result that 0 = 0, which, although accurate, is not

enlightening! Whenever the result of simplification is 0 = 0, the
original expression represents an identity - a relationship which all
values of the variables satisfy - rather than an equation, which has to
be solved for the specific values of variables for which it is true. Now
suppose the equation is
6(2x + 10) + X = 5(20 + 3x) - 2(x - 10)
which reduces to
13X + 60 = 120 + 13x

60 = 120
The requirement to equate two unequal constants means that the

original expression cannot be satisfied for any values of x. The two parts
of the expression are inconsistent (they plot as parallel lines against x)
and produce a contradiction. If this occurs in practice (assuming error-
free calculation) it may mean that the model has been mis-specified or
that the desired solution, for example an equilibrium point, does not
exist. But this outcome can itself be useful information. The problem of
finding the x intercept of a straight line produces one equation in one
variable:
mx + c = 0
which solves for x = —c/m, a unique value except for the cases where
m = 0, c = 0 (where the line is the x axis) and m = 0, c ^ 0 (where no
solution exists). The value of x = —c/m represents a zero of the
following function:
y = mx + c
and is the root of the equation (an x value for which the corresponding
y value is 0).

1.9.2 The case of two or more variables in one linear equation also brings
One equation in two out some interesting points. Take as an example the equation:
unknowns
5x + 2y = 24: (1-16)
This equation has an infinite number of solutions but this does not
mean that any pair of values satisfies the equation. Unlike an identity,
some X and y combinations are excluded. The equation imposes a
restriction that not all pairs of values satisfy. Only those pairs
corresponding to points on the line defined by the equation are
solutions. Some non-solutions can be usefully categorized: for example,
in this case there are no solutions in quadrant 3, where x and y are
both negative. In the case of lines going through the origin, apart from
the origin x and y are always the same sign for a line with positive
slope or always of opposite sign if the slope is negative. In finding
solutions to equation (1.16) either x or y may be freely chosen and a
specific value of the other variable results. There is one degree of
freedom in the setting of variables. In one linear equation in n
variables there are n — 1 degrees of freedom.
1.9.3 At times, only solutions satisfying a number of practical conditions may

Whole-number be acceptable - for example those in non-negative integers. In the case
requirements of equation (1.16) there are three such solutions:
X y
0 12
2 7
4 2
Models expressed in rational coefficients in which only integral, sign-

restricted solutions are acceptable are common in practice and more
difficult to solve. Such problems often arise with an inequality
relationship - such as a budget constraint - and their solution may
require the use of integer programming methods.
Exercises 1.9
1 Find the value of x which satisfies the following equations;
(i) 5x + 3(x-8)+4(12-0.5) +16 = 100
(ii) 20 - 4x + 3(2x - 5) + 5(4x + 68) = 2(4x + 5) + 1 10
2 Comment on the following equations:
(i) 3(5x+10)-(8x + 50) = -5(8-2x)-3x
(ii) 6(4x + 2)-7(3x+1) =8(5-f 3x)-3(7x+11)-2
3 Given the equation: 8x -f 5y = 40 which of the following are
solutions, or are consistent with solutions to the equation:
(i) The point X = 2.5, y = 4
(ii) The point x = 0, y = 8
(iii) The point x = —2, y = 11.2
(iv) The point x = 5, y = 8
(v) Points on the line 0.5x + 0.3125y — 2.5 = 0

(vi) Points on the line y = 8 — 1.6x
(vii) Points on the line 20x + 12.5y = 50
1.10 When a decision-maker has more than two controls or courses of

Linear systems of action, models will involve more than two variables. Our purpose here
equations in more is not to detail methods of solution, but to describe conditions in which
than two variables solutions exist and the number of solutions to be expected. There will
be a unique solution if there is the same number of consistent
equations as there are unknowns and if these equations are
independent (no equation being a weighted average of the other
equations). Such a system with one solution is said to be determinate.
An example of a three-equation three-unknown system is the following:
3x + 2y — z = 0
z + X +y = 34:
z— X + y = 26
The solution to such systems can be obtained in several ways, for

example by use of Gaussian elimination matrix inversion or Cramer’s
rule. Here we will use substitution or a combination of substitution and
elimination. From the first equation, state z in terms of x and y:
z=3x4-2y
Substitution for z in the second and third equations produces
3x -\- 2y 4- X 4- y = 34
3x 4- 2y — X 4- y = 26
which simplify to
4x4- 3y = 34
2x4- 3y = 26
to which elimination or substitution can be applied. Elimination rapidly

produces the result
X = 4 y = 6
and insertion of these values into the expression for z gives the result
z = 24. The approach used above is as follows:
1 Use one equation to state one variable in terms of the other two.
2 Substitute for the chosen variable in the remaining two equations.
3 Use substitution or elimination to solve the two-variable system
that is left.
As a further example consider the following equation:
3x — 2z/ -f 4z = 110
2x -f 2i/ — 3z = —30
—2x + 6i/ -F 4z = 220

The most convenient substitution uses the third equation to state x in
terms of y and z:
X = 3z/ + 2z — 110
which when substituted into the first equation gives:
7y + lOz = 440
In the second equation, the result is
8y + z= 190
So we now need to solve the 2x2 system:
7y + lOz = 440
8y + z= 190
Using substitution, from the second of the remaining equations
z = 190 — 8y
which when substituted into the first equation gives
-73y = -1460
y = 20
and by back substitution, first into the relationship between y and z:
z = 190- 8(20) = 30
and then into the relationship stating x in terms of y and z:
jc = 3(20)+ 2(30)- 110 = 10
The complete solution to the system is therefore:
X = 10 y = 20 z = 30
Not all 3 X 3 systems have unique solutions. For example, in
2x + 3i/ + 4z = 90
4x + 6i/ + 8z = 150
5x + 2y + 3z = 100
the left-hand side of the second equation is twice that of the first, but
the right-hand side is 150 rather than 180. So the equations are
inconsistent and the system has no solution. But if the right-hand side
of the second equation had been 180 the equations would have been
consistent but not independent (the second equation now being twice
the first) and the system would have a continuum of solutions.
Similarly, a system that begins as one of two equations in three
variables will not have a unique solution. For example, if the following
equation:
z— X + y = 26

had been omitted from the system solved earlier on we should have
been left with
4x + 3^ = 34
and, of course,
z = 3x + 2y
for which there is a continuum of solutions. However, not just any

values of x, y and 2 will do. If, as here, the two equations are
independent and consistent the solution points will lie along a straight
line. In general, a consistent system in which there are more variables
than there are independent equations is said to be indeterminate. An
example of an indeterminate system is illustrated in Figure 1.17.
Figure 1.17
The case shown in Figure 1.17 is one of three variables and two
independent and consistent equations, and shows the two planes
corresponding to the equations intersecting in the positive orthant. The
planes are not limited to the positive orthant, but only those parts of
the planes are illustrated. The points common to both planes, lying on
a straight line, are shown. If a third equation is added, and if this is
independent of the original two equations and consistent with them,
the third plane that this equation produces will intersect the line
common to the first two planes at a single point. The co-ordinates of
this point will be the values of the variables which uniquely solve the
three-equation system.
A system of three independent equations in two variables cannot be
consistent and therefore cannot have a solution. In general, a system in

which there are more independent equations than there are variables is
said to be overdeterminate. In the three-equation two-variable case
the values of x and y which satisfy any two of the equations will
contradict the third, as for example in the following:
2x + y = 13
x+y = 9
X + 2y = 11
This system is graphed in Figure 1.18, where it is evident that

whichever line is last to be drawn will not pass through the intersection
of the other two lines. The system can be made determinate by the
deletion of one of the equations or by modification of at least one of the
equations to include a term in an additional variable.
Figure 1.18
A system of three equations in two variables can have a unique

solution. A minor amendment to the inconsistent case above produces
such a system:
(i) 2x + y = 13
(ii) x+y = 8
(iii) x + 2y = 11
In this case the solution is
X = 5 y = 3
This solution is produced by any two of the equations, and, although in

any such system that has a unique solution the equations must be
consistent, they cannot be independent. This means that any one of the
equations can be obtained by combining the other two. In the system
above, for example, the second equation can be formed by adding (i)
and (iii) and dividing the result by 3. The interaction between the
equations is shown in Figure 1.19.

Figure 1.19
A system of two equations in three variables will not produce a

unique solution - it must be indeterminate. But, provided the equations
are independent and consistent, a ‘solution’ for two of the variables can
be given in terms of the third - but not necessarily any two of the
variables. For example, the system
5x “1“ ^ “1“ 42 ” 100
2x + 2y + 8z = 64
can be solved for x and y in terms of z, but not y and z in terms of x.

In fact, X must be 17 for the system to be consistent.
Now consider
(i) 5x + 1/ + 4z = 100
(ii) 2x + 2y — 8z = 64
From the first equation, y can be stated in terms of x and z as
y = 100 — 5x — 4z
which on substitution into (ii) produces:
2x + 200 - lOx - 82 - 8z = 64
so that:
—8x — 162 = —136
or:
(iii) X = 17 — 2z
and substitution for x into (ii) produces:
(iv) y = 15 + 62

Acceptable values for x and y do not give a unique point but rather
form a set of points, the solution set. Definite values for x and y can
be obtained only by specifying a value for z, or by the addition of a
third equation consistent with (i) and (ii) and independent of them. For
example, if we specify that z = 4, then
X = 9 y = 39
emerge as the solution. The same unique solution results if either of the
following equations is added to the original system:
X 2y -(- 5z =11
3x — 4i/ + z = 67
Note that both of the above additional equations include the point
(9, —9, 4) and another of the family of planes associated with this
point would have done as well. The addition of any particular
equation says more about the necessary relationships between x, y
and z than does the selection of an individual value for z. The
usefulness of this arrangement, in which x and y are stated in
terms of z, is that when one of the variables, say z, is determined
outside the system the corresponding values of the in-system
variables can be worked out. Variables determined outside a system
are said to be exogenous, while variables determined within the
system are endogenous. Treating some variables as exogenous is
one way that an otherwise indeterminate system can be made to
work out. The equations in which each of the endogenous variables
is stated in terms of the exogenous variables are said to be reduced
form equations. Thus (iii) and (iv) above are reduced form
equations.
Economics provides a number of examples of systems with fewer
equations than unknowns. For example, in barter economies a system
with n goods can be solved for the n — 1 exchange ratios between n — 1
of the goods and an arbitrarily selected nth good or numeraire. If
‘accounting money’ is introduced and a specific price set for the
numeraire, equilibrium prices result for the remaining n — 1 goods.
A linear equation in three variables produces a plane in three
dimensions. But how can a straight line in three dimensions be
represented in algebraic terms.? One way is to note that a straight line
in three space satisfies two linear equations:
z = aix + biy + ci
z = ajx + b2y + C2
which do not represent parallel or identical planes. Straight lines in

three space may also be specified by parametric equations or by
vector equations.

Exercises 1.10
1 Use substitution to solve the following systems of three linear
equations in three unknowns:
(i) 4x + 3y-z = 23
z + 2x + 5y = 23
z+X- y = 6
(ii) 5x-2y-z = 0
7z - 1 Ox - 4y = 10
z + 5x + 8y = 70
(iii) 8x + 5y - z = 0
z - 3x - 2y = 74
z + X - 5y = 90
(iv) 4x + 5y - z = 0
z-X+y = 30
z + 2x + 2y = 40
2 What can be said regarding solutions to the following systems
and the relationships between the equations?
(i) 3x + 2y + 4z = 45
6x + 4y + 8z = 95
2x + 5y + 3z = 50
(ii) 2x + y + z = 35
3x + y+ z = 40
4x 4- 2y + 2z = 70
(iii)1 2x + 3y + 5z = 115
4x + 2y - 2z = 10
4x 4- 4y + 3z = 105
3 State X and y in terms of z for the following systems:
(i) X + 4y — 5z =40
8x + 2y + 5z =20
(ii) 2x + 3y + 2z =54
4x + 7y + 2z =116
1.11 One of the most efficient ways to solve simultaneous linear equations is
Gaussian Gaussian elimination. This method uses the fact that the solution of
elimination the system, if there is one, is unaffected by the use of elementary row
operations. The method will be explained through numerical examples,
starting with the following system:
2x + 5i/ + z = 46
x + 3y + 2.52 = 45
2x + 4.5i/ + 42 = 74
First divide the first row by 2 in order to produce a unit coefficient for
X, giving
X + 2.5y + 0.5z = 23

(If the coefficient of x had been zero in the first equation, the position of
this equation should be exchanged with one in which the coefficient of
X is non-zero. This switch is an elementary row operation which leaves
the solution of the system unchanged.) For each subsequent row,
subtract a multiple of the modified first row to eliminate x. For the
second row, subtract the modified first row as it stands, giving
0.5y + 2z = 22
Subtracting twice the new first row from the third gives
—O.Sy -I- 3z = 28
The system is now
X + 2.5y + 0.5z = 23
0.5y + 2z — 22
—0.5y -t- 3z = 28
The first row will remain as it now is, and the process is repeated for the
second and subsequent rows. Multiplying the new second row by 2 gives
1/ -f 4z = 44
and adding half of this to the third row gives
5z= 50
Now complete the process by dividing the third row by 5 (there are no
further rows to deal with) to give a unit coefficient for z;
z = 10
The system is now
X + 2.5y + 0.5z = 23
y -|- 4z = 44
z= 10
For a system with a unique solution the value of the last variable is
given in the last row. If in eliminating y from the last equation z also
disappears, then if the right-hand side is not also zero the equations are
inconsistent. If the right-hand side value is zero, the equations are not
independent and there are infinitely many solutions.
In the current problem the values of the remaining variables are
obtained by back substitution. Putting z = 10 into the second equation
gives
y + 40 = 44
y = 4
Substituting y = 4 and z = 10 into the first equation gives
X -F 10 -|- 5 = 23
so X = 8

and the complete solution to the system is
x = 8 i/ = 4 z=10
This is the method of Gaussian elimination with back substitution, a

very efficient method for solving simultaneous linear equations. Gauss-
Jordan elimination, or Jordan elimination, is an elaboration of
Gaussian elimination where when any row has been divided through to
give a 1 as its first non-zero coefficient a multiple of this row is
subtracted from preceding as well as subsequent rows to eliminate the
corresponding variable from all equations. So in our case, having got
the second row to
1/ -f 42 = 44
and adding half of it to the third row, in Gauss-Jordan elimination we

now also subtract 2.5 times this row from the first to eliminate y in all
equations but the second, giving
X — 9.5z = —87
1/ -I- 42 = 44
5z= 50
The third equation is now divided by 5; 9.5 times the result is added to
the first equation and four times the result is subtracted from the
second equation, to give the neat result
X= 8
y = 4:
z= 10
Now consider another example. We have already encountered the

following system:
X — 1/5-2 = 15
2x 5- 2i/ 5- 2 = 20
5x -T 4i/ — 2 = 0
Using Gaussian elimination the-first equation needs no division, and x

is eliminated from the second and third equations by subtraction of
twice and five times the first equation, with the following result:
X- y + z = 15
4i/ - 2 = -10
9y-ez = -75
Dividing the second equation by 4 and subtracting nine times the

result from the third equation gives the last two equations as
y — 0.25z = —2.5
-3.75z = -52.5

Dividing row three by —3.75 gives
z = 14
and the full system is
X — y + z = 15
y — 0.25z = —2.5
z= 14
Back substitution gives
y - 0.25(14) = -2.5
y=1
Substituting the obtained values of z and y into the first equation gives
X - 1 + 14= 15
so X=2
and the solution to the system is therefore
X = 2 y = 1 z=14
Gaussian elimination is an efficient means of solving simultaneous

linear equations and can be used for inverting matrices. It also
underlies the simplex method of linear programming.
Exercises 1.11
1 Use Gaussian elimination to solve the following system:
X + 3y + 2z = 29
2x + 8y + 6z = 78
3x + 7y + 7z = 85
2 Solve the following system using Gaussian elimination:
2x + y + 2z = 19
4x + 3y + 6z = 37
2x + 2y + 3z = 26
1.12 An identity is an equation which is true for all values of its variables.
identities For this reason it is sometimes called an identical equation, as
distinct from a conditional equation (the form we have been dealing
with up to now), which is true for only certain values of its variables.
For an equation to be an identity it must be possible to cancel out all
terms involving variables leaving only the equation of two equal
constants. Where it is necessary to distinguish identities from equations
the symbol = is used in place of the equation symbol =. Identities may
involve expressions in several variables, and they arise in economics,
finance and management science. Any manipulation permitted for an
equation can also be applied to an identity. So elementary row

operations could be carried out on identities. In economics the best-
known examples of identities are in national income accounting. Let Y
represent the value of gross domestic product (GDP). One way of
measuring Y is to take the total value of household incomes. Let C
represent the total expenditure by households on consumption, while S
represents the level of savings - defined as that part of income which is
not spent on consumption. With these definitions, the identity that
results is
y= C+S
Now GDP can also be defined and measured from the expenditure side
as
Y = C+l
where C is consumption expenditure, as before, and I represents the

total expenditure on investment goods. In theory the two measures of
income should be the same (in practice an adjustment is necessary to
make them equal because of measurement imperfections). As a result of
the accounting definitions, it follows that
y= C+S
= C+I
thus
S=I
This identity states that actual savings always equal actual investment.
This does not mean that the levels of intended savings and investment
are always equal - the equality is a property of equilibria only. When
working with models of national income accounting of this kind the
identity sign is not always used. This will not lead to errors, but care is
necessary in drawing conclusions.
In production planning x and y might represent daily production
levels of two goods manufactured by a firm. A feasible production plan
(represented by a pair of values of x and y) will not call for the use of
more resources than are currently available to the firm each day. If
each unit of good x requires 20 minutes of machine time and each unit
of good y requires 13 minutes of machine time, then the total machine¬
time requirement for any production plan is given by
20x + I3y
Suppose that on any day a maximum of 480 minutes of machine

time are available. The amount of unused or slack machine time is s,
defined by
s = 480-20x- 13i/

The overall allocation of machine-time manufacturing usage and idle
time will then be stated by the identity
2Qx -b 13^ + s = 480 (1.17)
In practice the identity sign is usually replaced by the equality sign

in expressions like (1.17), but it is important not to overlook the fact
that an identity is involved. Relationships, such as (1.17), linking
resource requirements and availability are important in solving
production planning problems by linear programming.
1.13 Here we consider two further examples of straight lines in business and
Further economics: breakeven analysis and market equilibrium. First
Applications consider the business application.
1.13.1 Breakeven analysis uses simultaneous equations to find the output at

Breakeven analysis which revenue (income) for a period covers costs (outgoings). Cost is
given by
c= F+ y
where C represents total cost, F is fixed costs (overheads such as rent,

lease payments, interest charges), and V is variable costs (wages,
materials, fuel, etc.). Breakeven analysis takes variable costs to be
proportional to output level, q. Thus
V = bq
where b is the unit variable cost of production. It follows that
C = F + bq
The firm’s total revenue, R, is also proportional to output level:
R^pq
where p represents product price (unit revenue, here assumed

constant). Breakeven output is achieved at the point where total
revenue covers total costs, that is, where
R = C
pq = F-hbq
ip -b)q = F
so that
(1.18)
where q* is the breakeven output level. The model is graphed in Figure

1.20, where qc represents plant capacity.

Figure 1.20
For the revenue and cost lines to cross in the positive quadrant
product price must be strictly greater than unit variable cost. If this is
not the case, no breakeven output level exists. The relationship of q* to
qc is also important. If the breakeven output level is a high proportion
of plant capacity the financial viability of the enterprise will be at the
mercy of minor changes in parameter values such as unit variable cost
For a numerical example let p = £10, b = £6, F = £10 000 and
qc = 4000. Substitution in equation (1.18) yields
*
10 000
Q. = 2500
10-6
so the breakeven output level represents use of the plant at 62.5%
capacity. Note that if price had been p = 8, while this value of unit
revenue covers unit variable cost, the theoretical breakeven level,
q* = 5000, is not reached by the time that the firm runs into the
capacity limit. The breakeven formula can also be used to deduce the
consequences of changes in the parameter values, p, b and F. For
example, suppose that interest-rate changes, a wage agreement and
inflation of raw material costs result in new values of b = £6.80 and
F = £10 400. Two questions of interest would be:
• What would be the new breakeven output level?

• What new level of price would leave the original breakeven output
level unchanged?
To address the first question substitute the changed values of the

parameters into equation (1.18). This produces
*
10 000
= 3250
3.2

so at the original price the output level required to break even increases
by 30%. As regards the second question, the new level of price required
to maintain the original breakeven output level will be p where
10 400
2500 =
p- 6.8
so that:
2500p- 17 000 = 10 400
27 400
2 500
= £10.96
Note from the formula that equal proportionate changes in F, b and

p leave the breakeven output level unchanged.
1.13.2 Analysis of the market for a good often uses linear supply and demand
Market equilibrium curves relating price to quantity supplied or demanded. A market is in
equilibrium at a price where the quantity of the good consumers wish
to buy is the same as the quantity that the price stimulates producers
to offer. In other words the market reaches equilibrium where the
supply and demand equations are solved simultaneously. Suppose that
in the market for a product supply and demand are given by the
following linear equations:
Demand: p = 1000 — 5q
Supply: p = 200 + 3q
where p = price and q = quantity. The equations define the demand

curve and the supply curve, respectively. The market is in equilibrium
for values of p and q satisfying both equations. Thus
1000- 5q = 200+ 3q
800 = 8q
q= 100
and from either the supply or demand equation:
p = 500
which is the equilibrium price or market clearing price. At a price

above equilibrium, supply exceeds demand and there is excess supply.
For example, at a price of p = 650, suppliers wish to offer q = 150
units but consumers would only be prepared to talce q= 70 units, so
excess supply is 80 units. At a price below the equilibrium level,
demand exceeds supply, producing excess demand. For example, at a
price of p = 350, consumers wish to purchase q = 130 units, while
producers would only be willing to supply q = 50 units. Figure 1.21
graphs the situation.

Figure 1.21
Equilibrium is at E and the excess supply and excess demand

disequilibria are illustrated. This model can be expand in a variety of
ways. For example, suppose that a tax of £t per unit (a specific
sales tax or excise duty) is imposed by the government. This will
have the effect of shifting the supply curve upwards by t at all
points. This is because for any value of q the consumer must now
pay for each unit the price given by the supply curve plus the tax of
ft. Various questions arise. For instance, does the presence of the
tax mean that the equilibrium price will rise by the amount of the
tax.? Analysis shows that this will not be the case. For example,
suppose that the tax levied was t = 40. Market equilibrium is now
given by
1000-5q = 200 + 3^ + 40
760 = 8q
Therefore:
q = 95 p=525
so the market clearing price has gone up by £25 rather than the full
£40 of the tax. While consumers pay an extra £25 per unit, the firm is
receiving £15 less.
These relative magnitudes, 25 and 15, depend on the slopes of the
supply and demand curves, and give an indication of the comparative
burden of the tax born by consumers and producers. A further question
is the level of tax that would maximize tax revenue - a problem best
addressed by calculus methods.

Exercises 1.13
1 A firm has fixed costs of £4800 per week. It mokes o single
product, for which the unit variable cost of production is £3.
The product sells at a price of £5. Plant capacity is 3600 units
per week.
(i) Find the breakeven output level.
(ii) What percentage of plant capacity is used at the
breakeven point?
(iii) If fixed cost changes to £5400 and variable cost becomes
£3.25 what will be the new breakeven output level?
(iv) With the new cost data of part (iii), what selling price
would leave the original breakeven output level
unchanged?
2 A firm's monthly fixed costs are £13500. The firm makes a
single product, for which the unit variable cost is £8. The
product sells at a price of £12. Plant capacity is 6750 units
per month.
(i) Find the breakeven output level.
(ii) What percentage of plant capacity is used at breakeven
level?
(iii) Find the effect on breakeven output of the following
changes:
(a) Cost and selling price both increase by £1.
(b) Cost and selling price both increase by 12.5%.
(iv) With other data at their original values, what percentage
reduction in unit variable costs would be needed in order
to reduce the breakeven output level to 2250 units?
3 The market demand and supply curves for a product are given
by
D: p = 200 - q
S: p = 50 + 0.5q
(i) Find the excess demand at a price of p = 80.
(ii) Find the excess supply at a price of p = 130.
(iii) Find the equilibrium (market clearing) price and quantity.
(iv) Now suppose that the government imposes a tax of t per
unit supplied, so that the new tax-inclusive supply curve is
S: p = 50 + 0.5q + f
Assuming market clearing, find:
(a) The government's receipts from the tax (T) for t = 15 and
t = 30.
(b) What would be the costs to the exchequer of a subsidy of
7.5 per unit supplied?
4 The market demand curve for a product is given by
D: p=120-0.4q
and the supply curve with existing plant is given by
Si: p = 0.6q-f20
(i) Find the market clearing price and quantity.
Straight lines and linear equations

Suppose that new manufacturing plant is installed, resulting in
the supply curve
S2; p = 0.1q + 40
(ii) From what level of output does the new plant result in a
lower supply price than that resulting from the old plant?
(iii) With the new plant, what is the excess demand or supply
at the old equilibrium price?
(iv) Find the equilibrium price and quantity with the new
supply curve.
(v) What would be the excess demand or supply at the new
equilibrium price using the old plant?

C H R
2 Linear inequalities
This chapter explains the nature of linear inequalities, operations that

may be carried out on them and some of their many areas of
application. Sign requirements on decision variables are also introduced.
Section 2.5 builds on the basic knowledge of linear inequalities
acquired in earlier sections to introduce linear programming problems
and a graphical solution procedure.
LEARNING OUTCOMES:
By the end of the chapter you will have seen the usefulness of linear
inequalities in formulating management science and economic models.
You will have learnt how to simplify individual and simultaneous linear
inequalities and how to obtain the set of solutions. You will also be able
to solve small-scale linear programming problems using a graphical
method.
2.1 Introduction In linear models it is highly desirable that relationships between

variables are, wherever possible, expressed as equations. This is because
equations are highly convenient for purposes of manipulation and
solution. However, in business, economic and financial problems first
descriptions of relationships in the model frequently involve statements
that certain quantities can be equal only in the extreme - for example
expenditure and budget. The most fundamental of all economic laws,
the law of scarcity, is an inequality and states that the material wants
of a society, W, are greater than the total volume of goods and services,
0, that the economy can produce in order to satisfy these wants. This
elemental truth is stated as a strict inequality thus:
. W>Q (2.1)
In (2.1) the symbol > means ‘is strictly greater than’. In a society
^ where the law of scarcity did not apply the study of economics would
be reduced to an abstract exercise. Needless to say, no such society is
known to exist. In management science a frequently encountered
inequality relationship is that between the quantity of a resource
needed by a production plan and the amount of that resource available.
Linear inequalities 53
Consider an example in which x and y represent the daily output levels
of two products made by a firm. Each unit of the first product requires
20 minutes of machine time, while each unit of the second product
requires 13 minutes of machine time. There are 8 hours of machine
time available each day, and the amount of machine time called for in
any production plan should not exceed the available total. This
requirement can be stated as an inequality;
20x + 13i/< 480 (2.2)
where < means ‘less than or equal to’, which has exactly the same
meaning as ‘should not exceed’ or ‘not greater than’. ‘Less than or
equal to’ is an example of a weak inequality, where the equation of
the two sides is permissible as well as the allowable inequality. Weak
inequalities arise much more commonly than strict inequalities and
they are also much more convenient. Whenever there are budgetary
constraints, limited physical resources, minimum targets or variables
which cannot be negative, a weak inequality is defined. Other inequality
symbols in common use are:
• < ‘strictly less than

• > ‘greater than or equal to'.
The symbol < could be read as ‘neither greater than nor equal to’,
while > could equally well be read as ‘not less than’. If any of the
symbols are struck out the opposite meaning to that of the original
symbol is taken. For example, the symbol represents ‘not greater
than’ (and is equivalent to <,) while ^ means ‘not less than or equal
to’ (i.e. greater than). Most importantly, the struck-out equality symbol
7^ has the meaning ‘not equal to’, i.e. necessarily greater than or less
than. Taken together, the equality and inequality relationships are
called relational operators, and in computer programming the
relational operators are used to compare collections of symbols
(‘strings’) as well as numbers. In total there are six possible
relationships:
Relational operator Meaning

equal to
7^ or <> not equal to
< strictly less than
> strictly greater than
< or <= less than or equal to
> or >= greater than or equal to
The notion of being ‘approximately equal to’ is also useful, although

what is meant by ‘approximately’ has to be defined in context. The
symbol is ~ and a common use is in rounding numbers - as an
example of an approximate equality we may write
3.1389 3.14

Exercises 2.1
1 Give an appropriate symbol for the following literal
descriptions:
(i) Not greater than.
(ii) Strictly greater than.
(iii) Neither less than nor equal to.
(iv) Not equal to.
2 Replace the followinq symbols with simpler alternatives:
(i) ^ (ii)
(iii) ^ (iv) ^
2.2 Linear Model-building and analytical work involving inequality relationships

inequalities in one frequently require the rearrangement, simplification or manipulation of
variable inequalities. Any such re-expressions must preserve the relationships
involved. The following operations are permissible and may be carried
out on an inequality:
1 Addition or subtraction of the same constant term to or from each

side of an inequality.
2 Multiplication or division of both sides of the inequality by a positive
constant.
3 Multiplication of both sides of an inequality by a negative number
and reversal of the direction of the inequality.
2.2.1 If it is required that

Addition of a constant
X < 10
the ability to add a constant to both sides means that it is valid to write
X + 4 < 14
or, since the constant may be any real number, it will also be true that
X — 71 < 10 — TT
The constant may in fact be an unknown, since in
x + b < 10 + b
all that is required is that both sides of the inequality are varied by the
same amount.
2.2.2 Given the inequality

Multiplication by a
X < 10
positive constant
after multiplication of both sides by (say) 20 it will be true that
20x < 200
or after division by 20 it will be true that
0.05x < 0.5
If the original relationship had been
-0.5y > 100
then after multiplication by 4 it will be true that

• -2y > 400
and for any value, x, known to be positive
—0.5x1/ > lOOx
2.2.3 The validity of this rule is most clearly illustrated by the comparison of
Multiplication by a two real numbers, for example
negative constant and
6 < 7
reversal of direction
But when this inequality relationship is multiplied through by —1, the
valid statement is
-6 > -7
since of two negative numbers, the value further to the right on the
real line is the greater of the two. Thus with an initial statement
2x — 4 > X — 3
multiplication by — 1 throughout the inequality produces the equivalent

statement that
—2x 5“ 4 ^ —X “I" 3
while
2 — 2x < 4x — 10
divided by —2 becomes
X — 1 > 5 — 2x
or for any value, b, known to be negative
b{2 - 2x) > fe(4x — 10)
2.2.4 A linear inequality involving one unknown can often be rearranged so

Solution sets that the range of values of the unknown which satisfy the relationship
are more clearly seen. For example, given that
2x — 3 < X + 5
the solution set can be found - the set of values of x for which the
relationship is true. The solution set, if not empty, will be infinite. This
is shown in the relationship above, where the addition of 3 to both
sides produces
2x < X + 8

and subtraction of x from both sides produces
X < 8
which is the solution set in this case.

Consider another example. Suppose that the initial inequality is
10-2x<22 + x
Subtraction of 10 from both sides results in the following:
—2x < 12 + X
Subtraction of x from both sides gives
-3x < 12
and division by —3 produces the solution set as
X > —4
However, suppose that the original relationship had been
6 - 4x + 5(x - 2) > 3(x - 2) - (2x - 3)
This expression simplifies to
X—4 > X— 3
and, to be satisfied, would therefore require that
0 > 1
The solution set in this case is the empty set cj). This inequality
relationship therefore cannot be satisfied by any value of x. In contrast,
suppose the original inequality relationship had been
10(x + 5) - 3(2x + 5) > -2x + 40 - 2(5 - 3x)
This simplifies to
4x + 35 > 4x + 30
in which all of the terms in x are seen to cancel, leaving the

‘requirement’ that
5 > 0
which means that the original relationship is true for all values of x.
This is so, since it is a truthful relationship that did not really involve
f
X at all, and therefore cannot not be true for any x. So far we have
considered cases involving weak inequalities. Similar principles apply
to strict inequalities (you could rework the above examples using <
and > in place of < and >, respectively). In either case, when just one
inequality is involved, the resulting solution set, if not empty, is
defined by an upper or lower end-point for permissible values of x.
When more than one relationship must be satisfied there may be both
upper and lower bounds on x. Consider the following simultaneous
inequalities:
(a) 2x - 3 < X + 5
(b) x + 1 < 2x — 3
We have already seen that (a) can be simplified to x < 8. Requirement

(b) can be rearranged as x > 4, so that, taken together, the solution set
is
4 < X < 8
which is a solution set with both upper and lower end-points. There is
an important distinction between linear equations and linear
inequalities. As inequalities, (a) and (b) can be satisfied simultaneously
and produce a non-empty solution set. But if (a) and (b) were equations
they would be inconsistent - as must always be the case for
independent equations in a single variable. But inconsistency can occur
with inequalities too. For example, the system
(c) 20-3x<10-x
(d) 7x — 12 < 3 -I- 2x
requires that x > 5 (from (c)) and simultaneously that x < 3 (from (d)).
Thus the solution set is empty. To further point up the distinction
between simultaneous equations and inequalities, there is no limit to
the number of distinct linear inequalities that can simultaneously hold.
For example, suppose that
(e) 2x — 3 < X -f 5
(f) 3x — 4 < 2x -f 2
(g) X + 1 < 2x — 3
(h) 1 — 2x < —X — 4
The individual solution sets of the inequalities are
(e) X < 8
(f) X < 6
(g) x>4
(h) X > 5
Note that (e) and (f) give upper bounds on x. The more restrictive
of the two is (f), which is the least upper bound (LUB) or
supremum. If the least upper bound, (f), is satisfied, then (e) must also
hold. In general, if the least upper bound is fulfilled all other upper
bounds will be satisfied and therefore do not need to be referred to. We
can then concentrate on a single numerical value - the LUB. Lower
bounds are given by (g) and (h), and, of these, (h) is the greatest
lower bound (GLB) or infimum. If (h) is satisfied (g) must also be
fulfilled. Of all the lower bounds, only the greatest lower bound need be

explicitly considered. This is because if the greatest lower bound is
satisfied, all of the other lower bounds will automatically be fulfilled.
Therefore, for a system of simultaneous inequalities the solution set will
be the range between these tightest limits, that is,
GLB < X < TUB
in this case
5<%<6 (2.3)
Solution sets of this kind, defined between the infimum and the
supremum, often result when the stability of the solution to a model is
being investigated in relation to the values taken by certain key
parameters. Work of this nature is called sensitivity analysis, and the
range of values of a variable for which a solution holds good (expressed
as in (2.3)) is sometimes called a tolerance interval or a range of
feasibility. A solution set for one variable can be identified as a set of
points on the real line - a section of it - which may or may not include
the end-points. A range such as (2.3) which involves weak inequalities
at both ends of the interval is said to be a closed interval on the line,
as the end-points, 5 and 6, are included in the set. This is shown in
Figure 2.1, the solid circles at the limits of the interval indicating that
the end-points are included. If the solution set (2.3) had involved strict
inequalities at both ends, i.e.
5 < X < 6
then an open interval on the real line would have been defined. This
would appear as shown in Figure 2.1(b), with the end-points as ‘hoUow’
circles - the convention to signify that they are not included in the interval.
Figure 2.1
X
(a)
-0-e-►
5 6
(b)
- 0-•-►
’ 5 6
(c)
An open-closed interval is defined if the lower bound is not

included but the upper bound is within the range. This would be the
case if the solution set had been
5 < X < 6
which would have appeared as shown in Figure 2.1(c). Obviously when
this situation is reversed, as for
5 < X < 6
the solution set defines a closed-open interval on the real line.
Exercises 2.2
1 Find the solution set for each of the following linear
inequalities;
(i) 4x-8 <2x+12 (ii) 10x + 8 >12x-9
(iii) 6x + 25 > 3x + 10 (iv) 4(5 + 2x) < 20 + 6x
2 Comment on the solution sets in the following coses:
(i) 6(x- 10) + 3(25-x) <7(4 + 2x)-(llx+ 14)
(ii) 7(x- 10) + 5(16-x) <6(x-8)-4(x- 15)
3 Find the solution set for the following systems:
(i) 3x + 5 < 6x — 10
X — 15 < 33 - 3x
(ii) 4x-20<3x+10
3x + 15 < 8x — 30
(iii) 6x — 5 > 5x + 5
7x — 5 < 2x + 20
4 Find the solution set for the following:
(i) 3x + 2<2x+10
3x - 4 < 4x - 8
5x + 6 < 4x + 12
6 — X < 2x — 3
(ii) X + 20 > 3x + 6
4x + 12 < 8x + 4
X + 2 > 11 - 2x
5x — 5 < 4x + 4
(iii) 2x + 6 < 4x H- 2
4-x<x-2
5x - 1 < 4x + 8
3x + 2 < X + 16
(iv) 5x + 5 < 4x + 14
3x — 2 < X + 12
X + 3 < 2x + 1
2-X < x-4
2.3 Linear inequalities usually involve more than a single variable as, for
Linear inequalities example, when the condition relates to an expenditure limit on
in two or more resources. Consider the following inequality:
variables
2x + y < 13
The solution set is all points on or below the line defined by the
equality part of the relation and is shown in Figure 2.2.

Figure 2.2
In Figure 2.2 the inequality part of the relationship defines the

permitted (hatched) side of the line. The solution set is bounded only by
the line and extends without limit below the line. The solution set in
this case is called a half-space or half-plane, and the set could be
literally described as ‘the half space on or below the line y = 13 — 2x’.
Now consider the weak inequality
^ + 3iy < 19
The solution set is as shown in Figure 2.3.
Figure 2.3
In Figures 2.2 and 2.3 there is no exclusion of negative values for x

and y. In practice, however, it is often the case that only non-negative
values of the variables are meaningful - as, for example, with
production levels. Negative values of the variables are excluded by the
introduction of the sign requirements
X > 0 and y >0
in addition to the existing inequalities. In practice it is almost always
the case that sign requirements will take the form of non-negativity
requirements (rather than non-positivity requirements). These conditions
restrict the solution set to that part of the original set lying in the
positive quadrant, as shown in Figures 2.4(a) and 2.4(b).
Figure 2A
In one sense sign requirements are no different in character from

any other weak inequality. The non-negativity requirements on x and y
could equally well have been written in the following manner:
lx + Oy > 0
and:
Ox ly > 0
The equality part of the requirements - as in any other equation - will

produce a straight line in the x-y plane. In this case, the lines
correspond to the y and x axes, respectively. Now consider the system
produced by the two linear inequalities and the sign requirements taken
together. The full system is then
2x -i- y < 13
x + ^y < 19
> 0 y > 0
The solution set for this system is illustrated in Figure 2.5.

In Figure 2.5 the solution set is the perimeter and interior of the
polygon OCEB. For the moment note that the shape of the solution set
for simultaneous linear inequalities can vary considerably. Figures
2.6(a) to 2.6(e) illustrate the solution sets to the following systems:
(a) 2x3-1/<13 (b) 2x-hy<13

x-i-3y>19 4x-t-2i/>20
X>0

Figure 2.5
(c) 2x4-1/>13 (d) x + 3i/<19

x + 3y<19 3x + 9y>57
y>2
(e) 2x + y > 180

3x + 6y < 540
X 4- y = 100
> 0 y> 0
Note that in Figure 2.6(a) no sign restriction on x has been stipulated.

In Figure 2.6(b) the solution set is that part of the strip between the
parallel lines that does not involve negative values of x. In Figure 2.6(c)
the solution set is the hatched triangle, which, as it happens, does not
include any negative values of x even though these have not been
expressly ruled out. In Figure 2.6(d) the solution set is the straight line,
representing the only way that the requirements can be simultaneously
fulfilled - as strict equalities. In Figure 2.6(e), where the requirements
include an equality relationship, the solution set is that part of the line
given by the equation that is included in the region defined by the
inequalities. This is the line segment AB. In practice a constraint set of
this nature is not uncommon. For example, in batch production problems
X and y might represent the amounts produced of a uniform product by
different processes or in different plants. The inequalities would represent
resource-availability constraints, while the strict equality x + y = 100,
would represent a customer order for 100 units. Simultaneous
inequalities in several variables do not necessarily have solutions. For
example, if in case (d) above the inequalities involved had been
X -h 3y < 18
3x -I- 9y > 57
the intersection of the individual solution sets would have been empty,
since they would have been on conflicting sides of the parallel lines.
Figure 2.6
Where the equality part of the relationships produce non-parallel lines

the solution set may still be empty if sign requirements are present. For
example, the system
2x -f- ^ < 18
X + 2y > 40
X > 0 y > 0
restricted as it is to the positive quadrant by the sign requirements, has

no solution. A sketch of the system without the sign requirements
shows that the solution set lies entirely within the second quadrant.

Exercises 2.3
1 Describe the nature of the solution set in the following coses:
{') 2x + y < 25
(ii) x + y < 10
X>0 y >0
(iii) 5x — 2y > 20
y > 2.5x -13
(iv) 2x + 4y > 40
5x + 9y < 80
X>0 y >0
(v) x + 2y < 120
X + y = 80
X>0 y >0
2.4 Whatever the shape of the solution set for a system of simultaneous
Convex solution linear inequalities, it always has the important property of convexity.
sets In two dimensions, none of the interior angles of the solution set
polygon should be reflex (over 180°) and hence the boundary of the
feasible region should not be a reflex polygon. That is to say, the feasible
region will not be re-entrant - a convex set is one without
indentations. In Figure 2.7 since the angle at point A is reflex, a
feasible region of this nature could not result from a system of
simultaneous linear inequalities. Convexity means that for any two
points P and Q in the set all points on a straight line connecting P and
Q must also be in the set. So all the solution sets of Figure 2.6 are
convex. However, the set illustrated in Figure 2.7 is not convex, since
because of the inward pointing vertex it is possible to select points
within the set (for example P and Q as shown) such that some of the
points on the line segment connecting them are outside the set.
Figure 2.7
Convexity of the feasible region is an important property in problems
where the optimal value of a function is sought for points l5hng within
the feasible region. If the function to be optimized satisfies certain
conditions, then if the feasible region is convex we can be sure that a
point that is better than any other point in its locality will be the best
overall. An important concept in relation to convex sets is that of the
extreme point. An extreme point in a convex set does not lie on a line
segment joining any two other points in the set. Extreme points are
important in the simplex method of linear programming, an adjacent
extreme-point solution procedure, since it works towards the best
position by moving between successively better adjacent extreme points.
A non-convex set cannot be produced by linear inequalities which must
all be simultaneously satisfied. Each linear inequality produces a solution
set which is convex (a half-plane in the two-dimensional case, as we
have seen) and the intersection of convex sets must itself be a convex set.
A non-convex feasible region involving only linear inequalities could be
produced if there are alternative constraint systems. For example, suppose
that the permitted values of x and y must satisfy either (i) or (ii):
(i) 2x + y <3
x + y >2
(ii) 2x + y>3
x+y < 2
The region satisfying this somewhat uncommon arrangement is shown

in Figure 2.8.
Figure 2.8
While Figure 2.8 illustrates a possible situation, in practice if such a

problem did arise it could be addressed as two sub-problems
corresponding to the alternative, convex, regions defined by (i) and (ii)
separately, linear programming being the probable method of solution.

2.5 Convexity is important in problems that involve finding the maximum
Linear or minimum of a linear function subject to the condition that the
programming values of its independent variables lie in the solution set of a system of
linear inequalities. This is the situation in linear programming
problems. If the solution set is convex, any point which is the best point
in its own locality is the best point overall; a local optimum must be a
global optimum. This property means that highly efficient solution
procedures can be employed.
Consider an example. A firm can produce two products in amounts x
and y. These output levels are the firm’s decision variables. Typically,
management will wish to set production levels (choose values of x and
y) so as to maximize some objective, for example to make as much
profit as possible. Suppose that the most convenient time-frame for the
problem is one day. The accounts department estimates that each unit
of X produced makes £2 profit, while each unit of y makes £5. Thus the
firm’s objective is to choose values of x and y so as to maximize profit,
z, where
z = 2x + 5y (2-4)
Equation (2.4) is the firm’s objective function. Since (2.4) is a

linear function, unless there are some restrictions on the values taken
by the decision variables the objective function will have no finite
maximum. Restrictions are provided in practice by the fact that the
resources (both physical and financial) needed to produce the products
are not available in limitless amounts. If each unit of x requires 4
labour hours and each unit of y requires 3 hours a total daily
availability of 48 labour hours defines the following inequality;
4x + 3i/<48 (2.5)
(2.5) is known as a constraint. The left-hand side of (2.5) gives the

total labour-time requirement of any production plan (a pair of values
of X and y). For feasibility, the production plan must not call for more
labour time than is currently available. Suppose also that each unit of x
requires 1 kilo of a certain material and each unit of y requires 2 kilos.
An overall daily availability of 22kg of material defines a second weak
inequality as a constraint:
X + 2y < 22 (2-6)
In practice there may be a variety of constraints relating to other

scarce resources such as floor space, machine time or finance at various
times, but we shall limit ourselves to two constraints here. A crucial
word in this example is each. It is this word that tells us that the
problem is linear. If the resources required to produce a further unit of
a product depended on the amount already produced a non-linear and
much less tractable problem would be defined. In addition to the
resource constraints (2.5) and (2.6), there will normally be a
prohibition on negative values of the decision variables, the sign
requirements of earlier acquaintance. The complete problem in
algebraic form is then to
maximize z = 2x + 5y
subject to 4x + 3iy < 48
x + 2y <22
> 0 y >0
A problem of this size can be solved either by use of an algorithm

or by graphical means. An algorithm is a step-by-step procedure
whereby either the solution is reached in a finite number of steps or it
is shown that no solution exists. The principal algorithm in use is the
simplex method, which uses the fact that the solution set to the
constraints is convex. The simplex method involves the repetition of a
cycle of calculations (iteration). Microcomputer software using the
method is available, and it can be used to solve problems with many
variables and constraints. Here we will outline a graphical approach
which will illustrate the main features of the problems. The first step is
to draw a graph of the constraints and the solution set they produce.
This is shown in Figure 2.9.
Figure 2.9
The solution set is the interior and perimeter of the polygon OABC.
The solution set is known as the feasible set, feasible region or
opportunity set. In a production context it is also referred to as the
feasible production set, and the outer edge of the set - the line
segments connecting A, B and C - is the production possibility
frontier, representing the feasible production combinations which, with
the given resource levels, are Pareto optimal. Pareto optimality, an
important concept in economics, means that the output of one of the
products cannot be increased without that of the other being reduced.
Here we wilt refer to Figure 2.9 as the production possibility diagram.

Which of the points in OABC maximizes the value of the objective
function? Prior to linear programming a variety of rules of thumb and
conventional nostrums were employed to give answers (usually imperfect)
to problems of this kind. One of these prescriptions was:
1 Produce as much as possible of the most profitable product(s).
but the chief yardstick was often equipment utilization as reflected in

this ‘rule’:
2 Make sure that the production plan adopted makes full use of the
scarce resources.
Rule (1) needs spelling out in detail to be made operational and

rule (2) is not always feasible. More than this, neither rule, despite
their plausibility, can be relied on to produce the best position
(although either may, coincidentally, be satisfied at the optimum). The
chief shortcoming of ruie (1) is that it does not take sufficient account
of the constraints and relative resource levels. Rule (2), for its part,
fails to take account of the objective function! A correct solution
procedure must take due note of both aspects of the problem. The
provision of the technical means to do this was the great advance
made by linear programming. In the graphical method simultaneous
consideration is achieved by introducing the objective function into
the production possibility diagram. A contour of the objective
function is a line such that all x, y combinations on the line
correspond to the same value of z. The sense of ‘contour’ is the same
as in an Ordnance Survey map, with z representing height, and x and
y corresponding to the easterly and northerly directions, respectively.
In a profit-maximizing context the contours are also known as iso¬
profit lines. In Figure 2.10 the contours for z = 24, z = 40, z = 52
and z = 55 are shown as broken lines.
Figure 2.10
The further a contour is from the origin, the greater is the
corresponding value of z. The point within OABC lying on the highest
contour is the optimal solution to the linear programming problem. The
graphical procedure is implemented by first drawing a specimen
contour through a convenient point - usually one of the intercepts.
This could be C, for example, where the labour constraint line cuts the
X axis. At this point y = 0, so that from the constraint
4lX = 48
x=12
Insertion of x = 12 and y = 0 into the objective function gives z = 24.

The y intercept of the contour is then found by solving for the value of
y, which gives z = 24 when x = 0 Thus:
24 = z = 2(0) + 5y
from which
i/ = 4.8
With the intercepts obtained, the sample contour can be drawn in the
diagram and used as a guide for other contours. A value of z greater
than 24 produces a contour parallel to the original but further out
from the origin. For example, the z = 40 iso-profit line has its intercepts
at (20, 0) and (0, 8). The specimen contour can now be moved
upwards and to the right until there is just one point in common with
the feasible region - in this case point A. This will be the profit-
maximizing position. In linear programming problems the optimum
point will always be at a corner of the feasible region, that is, at an
extreme point. The optimal point may not be unique, but all optima
will give the same value of the objective function. In linear
programming the optimal solutions must always include corner points.
So in two variable problems a place where two of the constraint lines
intersect will be an optimum. Note that one of the intersecting lines
may be the x or y axis, representing the sign requirements satisfied as
equalities. Once an optimum has been identified in the diagram, the
approximate values of x and y can be read from the axes. More
accurate values for the variable at the optimum can be obtained by
solving the intersecting constraints as equations (these will be the only
requirements that hold in equality form at this point). Thus at A, where
the materials and x sign requirement both hold as equalities,
X + 2y = 22
X = 0
from which y = 11 which in turn produces z = 55, the highest possible

level of profit.
There is another value at the maximum position which is of interest.
With X = 0 and y = 11 only 33 labour hours are required. There are

therefore 15 hours of labour time unused or slack. As the optimum
slack in the labour constraint amounts to over 30% of the total
availability, rule (2) clearly does not work. It is worth examining what
would have happened had the production plan been set so as to make
full use of both resources. The only point in the feasible region which
achieves simultaneous full resource utilization is B. The profit made here
is found by first solving for the x and y values and then substituting into
the objective function. The requirements fulfilled as equalities at B are
4:X + 3y = 48
x + 2y = 22
Using elimination to solve for x and y gives
4:X + 3y = 48
4:X + 8y = 88
-5y = -40
so that
y= 8 and x= 6
which values produce z = 52. £3 less profit is made at B than at the

optimum - there is an opportunity loss of £3, the difference between
the profit levels at A and B. This is the price paid for the condition that
all scarce resources are fully used. Although rule (1) is fulfilled at the
optimum in the present example, this is not always the case. For
example, suppose that the unit profit on x had been £4 rather than £2.
The objective function would then have been
z = 4iX + 5y
Contours through C, A and B are shown in Figure 2.11. The highest

contour is now through B with z = 64. In comparison with A, 3 units
fewer of y are produced at B, but the crucial economic point is that the
relative reduction of 3 units in y production allows a 6-unit rise in x
output. This two-to-one trade-off or rate of substitution between x
and y, which was not worthwhile when the unit profit of x was £2, is
worthwhile when x’s unit profit is £5. The slope of the relevant section
of the production possibility frontier (given by the materials constraint
between A and B) shows the rate of substitution between products so
that resource limits are not exceeded. The slope of the iso-profit contours
shows the rate of substitution between the products that maintains a
given level of profit. If at any point a more favourable rate of substitution
exists (for movements in either direction) in terms of resources than for
the maintenance of profit, the point cannot be optimal. Note that at
each of the corner points A, B and C there are two such rates of
substitution on the resource side, dependent on the direction of
movement. At A and C the rates of substitution of y for x are oo and 0,
Figure 2.11
respectively. The optimal position depends on the relative slopes of the

iso-profit lines and the constraints. If the objective function contours
have, in absolute terms, lesser slope than either constraint, then point
A will be optimal; if they are steeper than both constraints C is best,
and if intermediate in slope, B is the superior position. Consider the
values of the slopes in the present example. For the labour constraint,
where
4:X + 3y = 48
y can be made the subject of the equation by writing
3y = 48 — 4x
so that
y = 16 — 1.33x
and the slope in absolute terms is 1.33. For the materials constraint we
have
x + 2y = 22
and to show up the slope we express y in terms of x as
2y = 22 — X
so that
y = 11 — O.Sx
Again disregarding sign, the slope is 0.5. With the original unit profit
figures the objective function was
z = 2x + 5y
which can be written as
5y = z — 2x

i.e.
y = 0.22 — OAx
where z takes a constant value (e.g. 24, 40) for any particular contour.
The absolute value of the slope is 0.4, which is less than either
constraint; hence A is optimal. With the constraints in explicit form,
note that the production possibility frontier (PPF) is an example of a
piece-wise defined function:
ll-O.Sx 0<x<6
16-1.33X 6<x<12
So long as the iso-profit contours have a slope different from that of

any constraint, one of the corner points of the feasible region will be
uniquely optimal (if both products showed a loss then the optimum
would be at the origin, 0). This is an important result, and it means
that, of all the points in any feasible region, only the corner points need
to be considered. If the slope of the objective function contours is the
same as that of one of the constraints the optimum will be non¬
unique, the highest contour will be superimposed on the constraint
edge with the same slope and two corner points will be equally good, as
will all points in between them. For example, suppose that
z = 8x + 6y
The contours would be as shown in Figure 2.12, and B, C and

intermediate points produce z = 96.
Figure 2.12
/A
>■
O X
In practice non-uniqueness at the optimum is usually a desirable
property, since the decision-maker is offered a costless choice between
the equally profitable joint optima. Consider another example:
maximize z = 12x + 2y
subject to 4iX + y < 800
2x+ 3y < 900
X < 180
^ > 0 y >0
In this problem the third constraint is an upper bound on the variable x

(it may correspond to some practical sales limit on the product). The
problem is graphed in Figure 2.13 and the feasible region is OABCD. A
specimen contour is shown through A. The optimum is at point C, where
4:x + y = 800
X = 180
from which it rapidly follows that y = 80 and that the value of the
objective function is maximized at z = 2320. At C there are 300 units
of slack in the second constraint. At B, which corresponds to full
resource utilization, the level of profit is z = 2200, revealing an
opportunity loss of £120 in comparison with the optimum. Now
consider minimization of the objective function and the following
problem:
minimize z = 3x + 5^
subject to X + 4y > 60
2x + > 50
^ > 0 y >0
Figure 2.13

Notice that in this problem the constraints take the form of > weak
inequalities. In a problem in which the objective is to minimize (such as
with a function expressing costs) and where, as in the present case, the
coefficients of the decision variables are positive, at least some of the
inequalities will be in this direction. The problem is graphed in Figure
2.14. The feasible region here is the area bounded only from below by
DBF and the axes. A specimen contour is shown through point D, with
the arrow indicating the direction of improvement. The optimum occurs
at B, where
X + 4y = 60
2x + y = 50
which solve for
x = 20 y = 10
Figure 2.14
The minimum value of z is z = 110. An advantage of using linear

programming is that the program output (or the result of manual
calculations in smaller problems) will provide more information than
just the optimal values of the decision variables. If a resource is in short
supply at the optimum (i.e. its constraint holds as an equality),
management will certainly seek to secure more of the resource and
relax the constraint to increase profit. But the extra resources will be
secured only at a price. The optimum position can be investigated to
reveal whether this price is worth paying. The most that is worth
paying is the increase in the objective function if a further unit of
resource was available or if an upper bound was relaxed. These
marginal values, dual values or shadow prices are crucial
75
Linear inequalities
management information in deciding whether a possible increase in
resource is worthwhile. We can also find out how the optimal
production plan responds to changes in product profitability as well as
variations in resource levels. This is the realm of sensitivity analysis,
which in small problems can be carried out in a graphical context but
is more efficiently conducted using the simplex method.
Exercises 2.5
1 Use a graphical approach to find the optimum for the
following linear programming problem:
maximize F = Sx] + 10x2
subject to 3xi + 2x2 < 54
2xi + 5x2 < 69
X] > 0 X2 > 0
Indicate the location of the optimum and state the optimal
values of the decision variables and the objective function.
2 With the constraints of problem 1, indicate the optimum
position(s) and find the optimal values of xi ,X2 and F for the
following objective functions:
(i) F = 8xi + 5x2
(ii) F = 3xi + 10x2
(iii) F = 8xi + 20x2
3 Solve the following linear programming problem:
maximize F = 35xi + 25x2
subject to 4xi + 3x2 < 92
xi + X2 < 3 8
X] < 20
X2 < 20
X] > 0 X2 > 0
4 Solve the following problem using the graphical method:
minimize F = 7xi + 3x2
subject to xi + 2x2 > 55
4xi + X2 > 80
xi > 0 X2 > 0

C H AFTER
Matrices
Sections 3.1 and 3.2 introduce basic concepts and matrix notation.
Sections 3.3-3.5 consider operations on matrices. Determinants are
introduced and linear dependence is discussed. Use of matrices in
solving linear systems is explained, and section 3.10 introduces
Cramer’s rule.
LEARNING OUTCOMES:
By the end of this chapter you will be able to add, subtract and
multiply matrices, evaluate determinants, find the inverse of a matrix,
and use matrices and determinants to solve systems of simultaneous
linear equations.
3.1 Introduction Quantitative models often use many unknowns and numerical values.
Masses of data could be handled ad hoc but this is inefficient. Error
tracking is difficult, and systematizing procedures and computer
implementation are more difficult. It is often desirable to use a
rectangular array, so conventions are needed for handling such arrays.
As an example, suppose that a firm can use two scarce resources to
produce two products, the Alpha model and the Omega model. Table 3.1
states the resource requirements per unit of each product produced.
Table 3.1 Alpha Omega
Materials (kg) 4 1
Labour (hrs) 2 3
Stripped of row and column labels, these resource requirements form

a rectangular array with two rows and two columns. This can be
shown as:
'4 1'
2 3
Conventionally enclosed in brackets, the array is a matrix. Some, but

not all properties of real numbers also apply to matrices (which have
77
Matrices
additional properties). Matrices are notation. In principle there is
nothing that can be done with matrices which could not be done
without them. But in practice matrices make a big difference because, as
with all good notation, matrices:
• are convenient for stating large and complex problems;

• facilitate the deduction of general properties:
• allow a focus on the numerical values in applications:
• are a suitable basis for the use of computers.
3.2 It is often necessary to operate on several matrices, so the first concern

Some is to ensure that the matrices are clearly identified and distinguished.
fundamentals
3.2.1 Matrices are usually identified by an upper-case letter, optionally

Labelling of matrices underlined or, as shown here, in bold italic. For example, the matrix of
resources might be written as
3.2.2 The dimensions are stated as the number of rows times the number of
Dimensions of matrices columns. The dimensions give the order of the matrix. Thus R is a
matrix of order 2x2. The matrix:
6 3 10
A = 4 2 5
21 3 9
is of order 3 x 3. In some applications matrices may have thousands of

rows and columns. A and R above each have the same number of rows
as columns and are square matrices, a category having special
properties. But matrices need not be square. For example.
with two rows and three columns, is of order 2x3, while
'1 5'
__ 2 6
^ 3 7
4 8
is a 4 X 2 matrix. An m-row, n-column matrix is of order m x n. So
D =

is a 3 X 1 matrix. A matrix with one column is called a column matrix
or, more commonly, a column vector. A matrix of order 1 x n such as
E = [1 2 3 5 4]
is called a row matrix or a row vector. By convention, when a vector

is not specified as a row or column vector it is assumed to be a column
vector. A matrix can be seen as a row vector of column vectors or a
column vector of row vectors. The matrix F = [6] is 1 x 1; 1 x 1
matrices have the properties of real numbers for operations such as
addition and multiplication, and no material additional properties. For
many purposes an individual number can be seen as a special case of a
square matrix where m = n = 1. The term ‘scalar, is used to
distinguish a single number in the context of matrices.
3.2.3 The elements of (or entries in) a matrix are the entities of which it is
The elements of composed. They could be integers - as in matrices A, B, C, D, E and F
matrices - or other numbers, including irrationals, as in
6.2 -4
G = 4/3 10^
TT —e
The elements of a matrix can also include variables, as in
3 —X
3.2.4 Starting with the matrix H, where

Sub-matrices 9
1 7 -3
2.5 4 0 3
H
6 8 -7 4
3 TT 6 V2
then
1 7
L = [6 8 -7 4]
7 2.5 4
1 -3
R M= [ttI
3 6
are examples of sub-matrices - obtained by deleting rows and/or

columns of H. Sub-matrices are used in expressing some optimization
conditions. Sub-matrices are also useful when forming a partitioned
matrix. For example, the matrix (with a particular partition indicated)
20 32 16 15 12
-17 44 27 13 41
15 34 26 57 12
22 19 20 35 17
79
Matrices
may conveniently be considered as being made up of sub-matrices
A B'
C D
where the sub-matrices are
■ 20 32 161 '15 12'

A = B =
.-17 44 27J .13 41.
15 34 26‘ '57 12'
C = D =
22 19 20 35 17
An m X n matrix can be partitioned into m-row vectors or n-column

vectors and the matrix can be analysed in terms of these vectors.
Partitioning helps to simplify operations on large matrices with special
structure.
3.2.5 The principal diagonal of a square matrix consists of the elements in

The principal diagonal the top-left to bottom-right diagonal. Alternative terminology is main
diagonal or leading diagonal. Thus, in the matrix N, where
12 3 4
5 6 7 8
9 10 n 12
13 14 15 16
underlined elements form the principal diagonal. Remaining elements

are ofT-diagonal. The sum of the main diagonal elements is the trace.
For the matrix N the trace, r=l-(-6 + ll-fl6 = 34.
3.2.6 A triangular matrix is a square matrix where only the elements on,
Triangular matrices or on one side of, the main diagonal are non-zero, an example being:
■ 1 0 0 O'
5 6 0 0
9 10 11 0
13 14 15 16
P is a 4 X 4 lower-triangular matrix. All the elements above the main

diagonal must be zero in a lower-triangular matrix. In contrast,
■-5 -4 -3 -2'
0-101
0 0 6 3
0 0 0 5_
is upper-triangular. Some elements on or above (below) the diagonal

can be zero in upper/(lower)-triangular matrices.
3.2.7 A diagonal matrix is a square matrix in which all the elements except
Diagonal matrices those on the leading diagonal are zero. For example.

10 0 0
0 4 0 0
0 0 7 0
0 0 0 ^2
is a 4 X 4 diagonal matrix. Note that diagonal matrices satisfy the

conditions for both upper and lower triangularity. A special case of a
diagonal matrix in which all of the elements on the main diagonal are
the same is called a scalar matrix. For example,
4 0 0
T = 0 4 0
0 0 4
is a scalar matrix, of which an important special case is an identity

matrix, where all main diagonal elements are one, as in the 3 x 3
example:
1 0 0
1 = 0 1 0
0 0 1
3.2.8 In a zero matrix all elements are zero. But zero matrices, unlike
Zero matrices identity matrices, need not be square. So, for example.
is a 2 X 3 zero matrix.
3.2.9 Elements can be referenced by their row and column number, e.g. row
Matrix notation 1, column 4, or row i, column j. To refer to the elements in a general
way, matrix notation is employed. Lower-case letters are used —
usually fl, b or c - in conjunction with a double subscript or index to
indicate the row and column of the element, as in the m x n matrix
flu fll2 fll3
«21 «22 dlS ^2n
fl31 fl32 «33 ^3n
_ ^ml ^m2 <^m3 ^mn _
where aij is the element in row i and column j. In square matrices

principal diagonal elements have the same row and column subscript
values and appear as an. It is sometimes convenient to refer to matrices
by their typical elements. The typical element for A would be fly, for
some other matrix B it would be by. In this manner, A could be
described as
A = (fli;) I ~ 1) 2, ■ • • 5 ni; j 1)2, • • ■,
81
Matrices
In this notation the ranges of the subscripts are not explicitly stated
when the order of the matrix is understood. Vectors require only one
subscript to identify their elements, as in the following column and row
vectors:
bi and [ci C2 C3 • • • c„]
bi
bm
3.2.10 Two matrices are equal when (1) the matrices have the same dimensions
Equal matrices and (2) corresponding elements in the matrices are equal. So if
3 3
A = and
6 6
the matrices are equal if and only if x = 9 and y — —1. Note that
3 3 0
A = and
6 6 0
do not satisfy the conditions for equal matrices.
3.2.11 A transpose is found by interchanging rows and columns, so that

Transposition fly = fly. The transpose of A is indicated by A' or AT For example, if
3 7-4
then A'
2-1 6
and if
7
8
then B' = [7 8 9 10]
9
10
The transpose of a column vector is a row vector and the transpose of a

row vector is a column vector. In A' above note that the matrix
(A')' = A: that is, the transpose of the transpose is the original matrix.
There is no analogous operation in the arithmetic of real numbers.
Consider two special cases involving transposition. First, for
1 3 4
A = 3 7 8
4 8 2
the transpose is A itself, so in this case A^ = A. For this to be so the

elements in the mth row of the matrix must be the same as elements in
the mth column, i.e. fly = fly for all if. Such symmetric matrices must
be square, with their elements symmetric about the main diagonal.

Now consider the transpose of A where:
The transpose is
Here the elements of the transposed matrix have the same absolute
value as the elements in corresponding positions in A but are of
opposite sign. For this to be so, aij = —aji and the matrix must square.
A matrix with this property is skew-symmetric or anti-symmetric.
The principal diagonal elements in a skew-symmetric matrix are zero.
Transposition can also be used to describe the interchange of two
rows or of two columns of a matrix. Such an operation leaves certain
essential properties unchanged and is an elementary matrix
operation.
Exercises 3.2
1 State the dimensions of the following:
3 2
B=[l 5 6 7]
6 1
-3
-2
-1 D = [0]
-2
-3
f = (o//)
where: ; = 1 to t;
/ = 1 to s.
2 Given
6 3 7
2 1 4
5 9 3
which of the following are sub-matrices of A?
3 7'
6 7
B 1 4
2 4
9 3
2 4
6 3
5 9
83
Matrices
3 Find the trace of
10 13 26 18
16 27 13 22
14 30 35 23
20 19 21 29
4 Are the matrices below upper-triangular, lower-triangular,

diagonal, scalar or in none of these categories?
'2 3 4' '2 0 O'

0 2 3 B = 0 2 0
0 0 2 0 0 2
"4 0 O' '4 0 O'
0 4 0 D= 0 3 0
0 2 4 0 0 2
'4 0 4' '1 0 1'
0 4 0 F = 0 1 0
4 0 4 0 0 1 _
5 What conditions are necessary for the following matrices to

be equal?
'9 X 7' '9 0 7'

y 3 8 and 3 3 8
2 4 z 2 4 1
(ii) 10 10 10
10 10 10
B = (b,v)
where; b,y = 10 for / = 1 to m and / = 1 to n.
6 Find the transpose of the following matrices:
4 -1
B= [2 TT 3-10]
-3 8
-8
0
V2 D = [100]
-1
4
7 Are the following symmetric, skew-symmetric or not symmetric?
9 4 13'
4 12 7
12 3 5
3 7 3 6

- 0 5 -1 -61
-5 0 3 0
B =
1 -3 0 9
L 6 0 -9 0.
r 0 1 2 11
1 0 3 5
C=
2 3 0 4
1 5 4 0.
r 8 8 1 7 -
8 TT 2 0
D=
1 2 0 6
7 0 6 e
3.3 We cannot march straight in and start adding and multiplying

Addition, matrices. First we have to say what is meant by these and other
subtraction and operations.
scalar
multiplication
3.3.1 Where matrices have the same dimensions, if elements in corresponding

Matrix addition positions are added the result is a matrix which is the sum of the
original matrices. Thus, if A = (ay) and B = (bij) are two matrices of
order m x n, the matrices can be added (they are conformable in
addition) and the sum A + B is the m x n matrix with elements
(at; + bij). So, if
_1
\
6
to
3‘
A = 3 0 and B = -2 -1
-1 2_ 4 9
then
8 7
A+B = 1 -1
3 11
More than two matrices may be added, as in

1
1_1
1_1
00 o^
(N O
'7 9‘ -3 2'
1
+ + =
'X)
3 2 5 3
1
Unknown elements may be involved in matrix addition, as in
b 4a 11' a -1 7
A = -b a 9 B = a c^ —c
a+b c 2c _-2b b b—c
the sum of which is
Matrices 85
b+a 4fl — 1 18
A+B = a—b a + c^ 9 -c
a— b c+b c+b
and if
15 -8 2x 7 25 -1
+ 7 2y z 31
10 3x
then from the elements in the (1, 1) positions it must be that
15 + 2x = 25
so X = 5, and from the elements in the (2, 1) positions it must be that
10 + 7 = z
so z = 17, and from the elements in the (2, 2) positions it must also be
that
3x + 2i/ = 31
therefore;
15 + 2y = 31
so:
2y = 16
i.e.
1/ = 8
The following properties hold for matrix addition:
• Matrix addition is commutative. The order in which the matrices

are added is immaterial:
A+B=B+A
• Matrix addition is associative. Grouping together for the purpose

of addition is immaterial, that is:
A + {B + C) = {A+B)+C={A + C)+B
• The transpose of the sum of (or difference between) matrices is the

sum of (difference between) the transposes:
{A±Bf =A'^ ±B^
3.3.2 If A = (Uij) and B = (by) are both matrices of dimension m x n, then B

Subtraction of matrices may be subtracted from A, and we may write
A B = (djj bjj)

which is the resulting matrix difference. For example, given
2 3 4 6 2-1
A = and B =
10 2 4-7 0
the matrix difference is
-4 1 5
A-B =
-3 7 2
It is obvious from these numerical values that where A and B are

matrices
A-B^B-A
Here B — A is
-1 -5
B
-7 -2
If the elements of A - B are (Cy) then the elements of B - A are (-Cy).

Matrix subtraction, unlike matrix addition but like the subtraction of
scalars, is not commutative.
3.3.3 The zero matrix is the identity matrix for the operation of matrix
The identity element for addition (requirements needed to fulfill an identity role depend on the
matrix addition operation being performed): that is, for conformable matrices A and 0:
A +0 =A for all matrices A
Note that for A = (ay):
0-A = -A (3.1)
where the matrix -A = (-ay) is the negative of A. Thus if
2
A = then
-4
The matrix difference A — B is the negative of the difference B — A.

Finally, note that as a rearrangement of equation (3.1)
A + (-A)=0
that is, a matrix plus its negative is a zero matrix of the same order.
3.3.4 First consider the addition of equal matrices. For example, start with a
Scalar multiplication matrix A where
Matrices 87
and add A to itself
4 1 8 2
2 3 4 6
which sum can be represented in general as
A+A = 2A
where the matrix 2A = (laij) is a scalar multiple of A. Not

surprisingly,
■4 1‘ ■4 1' ■4 1' 12 3'

+ +
2 3 2 3 2 3 6 9
and, in general, for c equal matrices which are added the result is
CA = (Cfly)
In scalar multiplication, for matrices A and B and scalars k and m it is

the case that
(i) {k ± m)A = kA ± mA
(ii) ikA±B) = kA±kB
(hi) m{kA) = {mk)A
(iv) 0A = 0
(v) -1A = -A
In (iv) 0 is the zero matrix, and in (v) —A is the negative of A.
Exercises 3.3
1 Find the sum of the matrices in the following coses;
2 3 4' ■-3 6 2'

1 -7 3 + 7 9 --1
-2 6 — i4 8 --3 8
10 -8‘ 2 16' -3 12
+ +
4 11 - 9 14 2 -7
4 2a _ b' —a 6 0
a b 6 + 9 —c 24
9 0+ c 1 9 c —a
2 Given that
5x 6 4-
—4y —z ■ z 4 ■
1_
2x 2y 24 12
1
find the values of x, y and z.

3 Given the matrices
7
and
2
find the scalar multiple k(A + B] when

(i) k = 2 {\\)k = -]
3.4 In the example of the two-product manufacturer the per unit resource
Matrix consumption data formed the following 2x2 matrix:
multiplication
Suppose we now want to find the unit costs of production for the
two goods, that materials cost £5 per kilo and that the wage rate for
labour is £6 per hour. The prices can be written as a 1 x 2 row vector
thus:
P=|5 6]
Given the price and resource-consumption information, the unit costs of

the Alpha and Omega models are
Alpha: 4 kg @ £5 +2 hrs @ £6 = £32

Omega: 1 kg @ £5 +3 hrs @ £6 = £23
These unit costs could be written as the following 1x2 row vector
C = [32 23]
where to arrive at the 32 element of the unit cost vector we perform an

operation combining the row vector of prices and the first column of
the resource requirement matrix R. Successive elements in the row
vector
[5 6]
are multiplied by successive elements in the column:
'4'
2
and the products are then summed. This process is indicated by

positioning the 2 x 1 column vector immediately after the 1x2 row
vector and showing the result as a 1 x 1 vector:
[5 = [32]
This outcome - in which a 1 x n row vector multiplies an n x 1

column vector, producing a scalar - is called the inner product, dot
Matrices 89
product or scalar product of the two vectors. Workings for the unit
cost of the omega model - the second element of the unit cost vector -
can be laid out in a similar way:
[5 = [23]
Taken together, these workings can be presented as follows:
4 1 (3.3)
5 6 32 23]
2 3
Arrangement (3.3) represents matrix multiplication, and is written in

terms of the matrices involved as
PR=C (3-4)
Now review what’s involved. The first element in C is the vector P

multiplied into the first column of R. The second element of C is P
multiplied into the second column of R. In setting out the results again
we will specify the dimensions of the matrices:
r
[5 [32 23]
3
Dimensions: 1x2 2x2 1x2
For this to work, the number of elements in the row vector P must be
the same as the number of elements in each column of R. This means
that the number of columns of P is the same as the number of rows of
R.
If the firm finds another supplier charging £7 per kilogram for
materials and £4 per hour for labour, what would be the unit costs of
the products.? The new prices can be included by adding a second row
to P The result is a second row in the product matrix C. First we set
out the results and then we describe them. The outcome is
■4 ■32
1
1
IT)
-
_
1' 23'
1_
.2 3. .36 19.
1
The unit cost of the Alpha model at the new prices, £36, is in the
second row, first column of the product matrix, and is the second row of
P multiplied into the first column of R. The new unit cost of Omegas,
£19, is in the second row, second column of the product matrix, and is
the product of the second row of P into the second column of R. The
workings show that a 2 x 2 matrix multiplied by a 2 x 2 matrix
produces another 2x2 matrix. Now imagine a third source for
materials and labour charging £3 per kilogram and £8 per hour. The
further expanded matrix operations are:

5 32 23
1
7 36 19
3
3 28 27
where the Alpha cost, £28, is element C31 of C and is the third row ofP
multiplied into the first column of R. The Omega cost, £27, is element
C32 of C and results from multiplying the third row of P into the second
column of R. Note the dimensions of the matrices;
P X R = C
3x2 X 2x2 = 3x2
It is essential that the number of columns in P is equal to the

number of rows in the second matrix, R. In terms of the dimensions, it
is as if the number of columns of P and the number of rows of R cancel
to give the dimensions of the product matrix C. If P had m rows and
two columns the dimensions of the product matrix C would be m x 2.
Now suppose there is a third model with per unit resource
requirements of 3.5 kilos of material and 1.5 labour hours. With these
details added, R becomes:
and there will now be three sets of unit costs for the three products,
represented in matrix form as
■5 6' ■32 23 26.5‘

■4 1 3.5
7 4 = 36 19 30.5
2 3 1.5.
3 8_ 28 27 22.5_
The unit costs in the third column of the expanded C matrix are
successive rows of P into the last column of R. If P had dimensions
mx 2 and R was 2 xn then C would he m x n. If the number of
columns of P and the number of rows of R are both p, then the
matrices are conformable in multiplication, and the dimensions of
the product matrix will be m x n. Notice that the product of the
matrices:
A B
'5 6 2
■4 1'
7 4 4
2 3
3 8 6
Dimensions: 3x3 2x2
cannot be formed as the number of columns of A is not equal to the

number of rows in B. When a matrix has the same number of rows as
91
Matrices
columns it can be multiplied by itself. The result is the square of the
original matrix:
AA=A^
For example, if
4 1
A =
2 3
then
'4 1' '4 1‘ 18 7'

2 3 2 3 14 11
Notice that the elements oi A^ are not the squares of the elements of
A[{aij)^]. A square matrix can be raised to higher powers: for example,
'4 1‘ '4 1' ■4 1' '86 39'

A^ =AAA
2 3 2 3 2 3 78 47
Notice that A^ can be obtained as AÂ or as AA^; the result is the

same in either case. But in terms of matrix multiplication this instance
of the order of multiplication being immaterial is the exception rather
than the rule. Consider the following 2x2 square matrix:
Now find B^:
a matrix B with the property that = B is idempotent. Note that the

identity matrix is idempotent, and that it follows from B^ = B that
B" = B for idempotent matrices. For square matrices the following rules
apply:
1 The product rule for the exponents of scalars is paralleled in

terms of square matrices A"'A" = A"'
2 The power rule for exponents carries over to square matrices:
(A"’)" = A'"".
In 1 and 2 aboye m and n are non-negative integers and, by definition,

A°=/.
In matrix multiplication order is important and usually affects the
result. For two square matrices A and B of the same order, in general,
AB^BA

For example, if
4 1 6
and
2 3 4
then the product AB is
■4 1‘ ■5 6' '27 28'

2 3 _7 4 31 24
In this case the matrix A is said to premultiply the matrix B, or B is

said to postmultiply A. Now consider the result when B premultiplies A:
■5 6‘ ■4 1' ■32 23'

7 4 2 3 36 19
This is an outcome different from AB. Where the product matrix is to

be transposed, it can be helpful to note that
{ABf = BÂ^
i.e. the transpose of the product is the product of the transposes — but
in reverse order. The matrices A and B above can be used to
demonstrate this result:
B^ A^ {ABf
■5 7' ■4 2' '27 31'
.6 4. .1 3. .28 24.
It can happen for particular matrices that the product AB is the

same as BA, for example if the two matrices are
5 4 [3 2"
and B= ^ ,
-4 5
with the result that in this case
22
= BA
7
When AB = BA the matrices are commutative. AB is commutative

with either A or B. While there are examples of commutative matrices,
matrix multiplication - unlike multiplication of real numbers - is not
commutative. Other properties of matrix multiplication differ from the
properties of scalar multiplication. For example, in scalar arithmetic if
the product ab is zero, then either a or b, or both, must be zero. But
with matrices A and B, if
AB = 0
it does not follow that A or B must be a zero matrix. For example, if
4 6 0 0
and then AB =
2 3 0 0
93
Matrices
In the arithmetic of real numbers for given scalars a, c and d, if ac = ad
for aÔ, then c must equal d. But in the case of matrices A, C and D, if
AC = AD where D^C
it does not follow that A is necessarily a zero matrix. For example,

given matrices
4 6 2 3 5 3
A = C= D
2 3 4 1 2 1
then
32 18
AC = = AD
16 9
A further surprising property is that the square of a matrix which is

not itself the zero matrix can be a zero matrix, that is
A^ =0 for AÔ
For example, if
1 2 0 0
then A^ =
-0.5 -1 0 0
These phenomena occur only when some of the matrices have a

hidden aspect of ‘zero-ness’ about them {A and B are singular, as
defined in section 3.3). The parallels between matrix multiplication and
multiplication in common arithmetic are limited, and there is no matrix
equivalent for the division of real numbers. Matrix multiplication may
involve more than two matrices. For example, three matrices - A, B
and C - can be multiplied together to form the product ABC if the
dimensions allow multiplication in this sequence. If A is of order m x n,
B is n X p and C is p x q, then ABC has dimensions m x q. Consider
some examples:
(i) A B C
'2' '2'
1 4 1'
2 3] 3 = [11 8 17] 3
.3 0 5
_1
rH
1
= [63]
(ii) A B C
2 1
'3'
1 3 '4 2 [4 2:
2
0 5 10
32 16
36 18
40 20

(iii)
2 1 3 5 4
2 3 O'
-1 4 0 2 3
1 -1 2
3 1 2 1 0
15 11
12 3 O'
3 8
4 1-12
19 15
59 41 34 22
35 14 1 16
79 53 42 30
However, in the following case the matrices A and B are not

conformable, so the product cannot be formed:
A B C
3
'4 5'
2 [6 9]
1 7
1
Although not commutative, matrix multiplication is associative.

Thus for conformable matrices A, B, C and D
(AB}CD = A(BC)D = AB(CD)
i.e. so long as the ordering A, B, C, D is preserved it does not matter

how the matrices are bracketed together. For conformable matrices A, B
and C
A(B + C)=AB+AC
(B + C)A = BA+ CA
Matrix multiplication distributes over matrix addition. For example, if the

matrices involved are
A B C
5 '3 7' '2 11'
+
-1 .0 5. .4 6..
the postmultiplication of A by the sum of B and C gives the result
A (B+C)
■ 5 4' '5 8' '41 84'
.-1 9. .4 11. .31 91.
and AB + AC works out as
AB + AC
■ 15 55' 26 29' '41 84'
+
-3 38 34 53. 31 91
95
Matrices
In the multiplication of real numbers 1 is the identity element for
multiplication, leaving unchanged any number by which it is
multiplied. In matrix multiplication the equivalent role is filled by the
identity matrix, I, such that for a matrix A
AI = A and lA = A
where I is of appropriate order in each case. Identity matrices are

diagonal matrices in which the non-zero elements an = 1. Examples
are as follows:
'1 0 0 o'
1 0' '1 0 o' 1 0 0
0
0 1 0 =h 0 0 1 0
0 1
0 0 1 0 0 1
0
Note that the order of the identity matrix can be shown by a
subscript and /i = [1]. To show that I is the identity element in
multiplication suppose that
and observe that if the 2 x 3 matrix A is premultiplied by an identity

matrix, I must be 2 x 2:
1 o' ■4 1 6‘ ■4 1 6'
0 1 2 3 5 2 3 5
If A is postmultiplied, I3 is used (the subscript is often omitted):
10 0
4 16 4 16
0 10
2 3 5 2 3 5
0 0 1
Exercises 3.4
1 Carry out the following matrix multiplications:
(i) 8 3
[3 5]
-1 4
2 1 ■ '6 9'
3 -2 1 4
(iii)
8
1
-1
(iv)
2 5
■5 4 3'
-3 4
1 -2 6
6 -1

Given
2 1
A
-1 3
find: (i) A^ (ii) A^ (iii) A^

Which of the following matrices are idempotent?
(i) 2 4
-0.5 -2
(ii) 8 7
-8 -7
4 Show that the following are commutative:
3 7 -2 -4
and B=
-7 3 4 -2
5 Carry out the following matrix multiplications:
2 5 -1
[4 3
-2 6 4
(ii) r3 2
1 3 3 6
4 -1
(iii) r 3 2 01 ri 31 r
1 0
1 2 -1 2 1
3 5 -1
2 5 1 0 4 -
3,5 For two square matrices A and B of the same order, if and only if
Matrix inversion
AB = BA = I
the matrix B is said to be the inverse of A, and A is the inverse of B.

The inverse relationship is written thus:
B=A~^ and A=B~^
It follows that
so a matrix is the inverse of its own inverse. A matrix and its inverse
are commutative. Consider two examples of inverses. If
2 3 0.8 -0.6
A = then is A ^
1 4 -0.2 0.4
Matrices 97
As a check, the multiplications AA~^ and A~Â may be carried out to
confirm that the result in both cases is /2- In the former case the
workings are
'2 3' 0.8 -0.6' ’1 O'

1 4 -0.2 0.4 0 1
For the second example consider the following: if
2 1
A = then is
1 1
If the inverse of A exists, then it is unique and A is invertible. But, in

apparent contrast to real numbers, not all matrices have inverses. As
may be deduced from the commutative property of inverses, only square
matrices have inverses, but not all square matrices are invertible: certain
conditions need to be satisfied. Consider the problem of finding the
inverse of a 2 x 2 matrix A from scratch. The inverse, B, with elements
bij, must be of order 2x2 and such that
All ai2 ■fall fal2 1 o'

^21 «22 .fa21 fa22 _ 0 1
Each element of the product matrix, I, gives a constraint on the

elements of A and B that make it up. So the elements of the inverse bij
must satisfy the following:
+ (^12^21 = 1
diibn + «i2l^22 = 0
^^211^11 + (^22^21 = 0
(i2lbl2 + «22l’22 = 1
When these requirements are solved to state the by in terms of the fly,
the results are
, —fll2
b,,
”11 —
Allfl22 — <^12^*21 flllA22 — «12fl21
, —^^21 , flu
b21 = b22 =
«11«22 — ^^12^21 flllfl22 — «12«21
The denominator aiifl22 — 12^21 in all the expressions is vital to the

inverse, and an early task is to find this value. But note that the by will
be undefined and the inverse will not exist if this number is zero. The
number fliia22 — fli2<i2i is the determinant of the 2x2 matrix,
because its value (crucially, non-zero or zero) determines whether or not
the inverse exists. The B matrix could be shown as the scalar product of
the reciprocal of the determinant and a matrix comprising the
numerators of the ratios above thus:
4->=B =-1-[ ““
^11^22 ~ ^*12^*21 [~^21 ^*11

When the information is presented in this way two new concepts
used to find the inverse are distinguished: the determinant and the
matrix of rearranged and re-signed elements of A making up the
remainder of B, called the adjugate matrix or adjoint of A. Let us
consider these concepts in more detail.
3.5.1 The determinant of

The determinant
ail <^12
A =
«21 «22
is written as either
1A| or det A
and is the product of the main diagonal elements less the product of the
two remaining elements. It requires more substantial calculations to
obtain determinants of matrices of order higher than 2x2, and these
will be considered later. For the moment we will develop the idea and
illustrate its uses in the context of 2 x 2 matrices. Consider some
examples:
2 3'
(i) A = A| =2(4) - 1(3) = 5
1 4
■ 1
(ii) A = |A1 = 1(3)-(-!)(-!) = 2
-1
0.5
(iii) A = detA = 0.5(8) - (7)(3) = -17
3
2 3
(iv) A = detA = 2(6)-(3)(4) = 0
4 6
9 4.5
(v) A = detA = 9(-2) - (4.5)(-4) = 0
-4 -2
The determinant can be positive, negative or zero. In cases (iv) and

(v), where |A| = 0 the matrices are singular. In contrast, cases (i)-(iii)
represent non-singular matrices. This is an important distinction since
only non-singular matrices have inverses. In singular matrices of order
2, one row is always a multiple of the other (an example of linear
dependence, which can take more varied forms in larger matrices), and
the same must also be true of columns. So if, in terms of the rows.
ail fll2
A =
kail kai2
then the determinant of A must be zero, since
detA = aiikai2 — kaiiai2 — 0
99
Matrices
3.5.2 Given the matrix A, where:
The ad jugate matrix
All «12
A =
an ^22
the matrix
«22 —fll2
—fl21
is the adjugate matrix or adjoint of A and is written as
adj A
We shall look at the adjoints of larger matrices later on.

Consider some examples of the adjoints of 2 x 2 matrices:
3' -3‘
(i) A = adj A =
4 2 .
-1 1'
(ii) A = adj A =
3 1
The role of the adjugate matrix becomes clear in forming the inverse
when we examine the products A adj A and adj A A. The order of
multiplication is immaterial as A and adj A are commutative so the
products are always the same. Take the matrices above as illustrations.
In the case of example (i) A adj A works out as
'2 3" 4 -3' '5 O'

1 4 -1 2 0 5
The product adj A A gives the same result:
4 -3' '2 3' '5 O'

-1 2 1 4 0 5
In the case of (ii), the products A adj A and adj A A work out as:
1 -1' '3 1' '2 0'

-1 3 1 1 0 2_
The product of a matrix and its adjoint is a diagonal matrix with

elements an = |A|. In the case of matrices of order 2x2:
A 0
A adj A =
0 A
Confirmation of this outcome is obtained by forming the product:
flu fll2 fl22 —fll2 «llfl22 — ^112^*21 0

_fl21 «22 fl21 flu 0 «ll^l22 — ^12^*21

3.5.3 The onty modification needed to A adj A to obtain the inverse is
The inverse division of its elements by | A |. The inverse is the scalar product of the
reciprocal of the determinant and the adjugate matrix:
âd]A=A-^ (3.5)
|A|
We can now complete the inversion process for the two examples above:
(a) If
since
4 -3
\A\ = 5 and adj A =
-1 2
then
4/5 -3/5
-1/5 2/5
(b) If
since
3 1
A| = 2 and adj A =
1 1
then
-1 1 . .
A ^ = -—r adj A =
3/2
1/2
1/2
1/2
|A|
In using this method for obtaining the inverse of a matrix it is advisable

to begin by finding the determinant rather than the adjugate matrix,
since if the determinant is zero the inverse matrix does not exist.
Consider some further examples:
(c) If
since
3 -4
|A| = 10 and adj A =
-2 6
then
1 , . 0.3 -0.4'
\A\ ’ " -0.2 0.6
Matrices 101
(d) If
7 1
A =
9 2
since
2 -
\A\ = 5 and adj A =
-9
then
1-1
1 , 0.4 -0.2'
— adj A =
|A| -1.8 1.4
(e) If
2 1
A =
5 2
since
|A| = —1 and adj A =
then
1 1 , ■-2 1'
A ^ = 7—7 adj A =
|A| 5 -2
(f) If
4 6
A =
2 3
since |A| = 0, A is singular and the inverse matrix in this case does not
exist.
Now consider two special cases. A square matrix is orthogonal if
A'A=I
that is, if the transpose of the matrix is the inverse. For example,
0.8 0.6'
- 0.6 0.8
is orthogonal. For orthogonal matrices, |A| = ±1, and if A and B are

orthogonal matrices of the same order the product AB is also
orthogonal. An intriguing special case is
\
'5 -4' 1 1 , . ■5 -4'

A = A ^ = T—1 adj A =
6 -5 A 6 -5
Here, A is its own inverse and A is involutional. For involutional

matrices A^ = I, and the rule for real numbers that if = 1 then

a = ±1 does not apply to matrices (although there is an element of
‘oneness’ about involutional matrices, as we shall see). The following
properties of inverses can be helpful in analysis and in minimizing the
extent of manual calculations;
• The inverse of the inverse is the original matrix. That is:
=A
• The determinant of the inverse is the inverse (reciprocal) of the

determinant:
\A-^\ = {\A\)-^
• The inverse of the transpose of a matrix is the transpose of the

inverse:
(A^)-i = {A-^f
• The inverse of the product of two matrices is, in reverse order, the
product of their two inverses:
Only square matrices can have an inverse, but non-square matrices

may have a left inverse or a right inverse. B is the left inverse of A if
BA = I
C is the right inverse of A if
AC=I
Both left and right inverses exist only if A is square, in which case
B = C=
We will use inverse to refer to the multiplicative inverse defined above,

but an inverse can be defined in relation to other operations. A given
matrix when combined with its inverse produces the identity matrix for
that operation (the zero matrix is the identity element for matrix
addition and the matrix -X represents the additive inverse of X).
Exercises 3.5
1 Establish whether or not B is the inverse of A where
(i) 3 0.8 -06 . '
B
4 .-1 1.
(ii) 5 ■ 0.4 -0.5

B
4 -0.2 0.5
2 1 -0.4
B
5 -1 0.2
Matrices 103
Find I A| for the following;
(i) 9 7
A
2 3
(ii) -3 -2
-1 -4
2 5
3 4
0.4 0.3
0.3 0.2
3 tt"
27r -2.
(vi) 8 2
12 3
3 Find the odjugote matrices for the following:
3 2
A
1 5
(ii) 2 -6
A
3 -7
(iii) 3 — TT
A
2 e
(iv) 5 0
A
0
4 Find the inverse, if it exists, for the following:
4 5
A
2 3
4 2
A
3 4
3 2
A
10 5
(iv) 1.5 -0.5
A
-0.4 0.2
8 -4
A
-6 2
5 Which of the following ore orthogonal or involutional?
(i) 7 -6
8 -7
0.6 -0.8
0.8 0.6

0.8 0.21
II
.1.8 -0.8.
■-0.6 0.8'
II
0.8 0.6.
3.6 Consider the system:

Simultaneous
lx\ + 1%2 = 34
equations
9xi + 2x2 = 48
which in matrix form is
■7 1' Xi ■34'
9 2 _X2_ 48
That is, a 2 X 2 matrix of coefficients (A) premultiplying a two element

column vector of unknowns (x), the result being the 2x1 vector of
right-hand side values (b). A system of simultaneous linear equations in
matrix terms is
Ax = b
in which b ^ 0. Suppose the system was one equation in one

unknown:
ax = b
The solution for x is
b
X = -
a
that is
X = a~^b (3-6)
To show the equivalent of (3.6) for larger systems, given
Ax = b
premultiply both sides by the inverse of A, giving
A~Âx = A~^b (3.7)
Since the product of a matrix and its inverse is the identity matrix (3.7)
becomes
Ix = A~^b
Since I = x we can leave out I, so
X = A^^b (3.8)
Matrices
105
In terms of our current example, (3.8) is
'xf ■7 r -1 ■34'
.^2. .9 2. .48.
0.4 -0.2' ■34‘
-1.8 1.4 .48.
~ .6 .
That is, Xi = 4 and Xj = 6. The two main steps are:
1 Find the inverse (if it exists) of the coefficient matrix.

2 Premultiply the right-hand side vector by the inverse.
Consider some examples.

(i) The system
2xi + 3x2 = 16
Xi -I- 4x2 = 13
can be expressed as
'2 3' 'xi 16'

1 4 .^2. 13
-1
Xi '2 3' 16'
X2 1 4 13
and so
Xi 0.8 -0.6' 16' '5'

X2 -0.2 0.4 13 2
(ii) If
■3 -1' Xi '23'
4 2 X2 24
then
0.2 0.1‘ '23' 7'

0.4 0.3 24 -2
(iii) If
0.4 -0.2" Xi ■-50'

-1.8 ' 1-4. _X2. 250
then:
Xi ■7 1 '-50' -lOO'
X2 9 2 250 50

(iv) But if the system is
■3 1' Xi '20'
6 2 _X2_ 40
the determinant of the coefficient matrix is zero, so the inverse does not
exist. The equations as a whole are either inconsistent or, as here,
linearly dependent. The coefficient matrix has linear dependence in
either case. If the right-hand-side elements are in the same relationship
as the rows in the A matrix the equations are consistent (but linearly
dependent). Otherwise the equations are inconsistent - in the two-
variable case producing parallel straight lines. Here the second equation
is twice the first and there are infinitely many solutions.
In the case of
(V)
'2 3' Xi lO'

6 2 _X2_ 40
the A matrix is again singular and so is non-invertible. In this case the

equations are inconsistent, the relationship between the rows of A is
not replicated between the elements of b. The straight lines
corresponding to the equations are parallel and there are no solutions.
The value of the inverse in the solution of linear equations is clear
when the system is solved more than once with different right-hand
side values. For example, suppose that solutions are required as follows:
■7 1' Xi [341 3' lO'

and
21
for b =
9 2 _X2_ 48 11 0 27
The inverse need be obtained only once, solutions being found by

postmultiplying the inverse by the respective right-hand-side vectors.
The results are
0.4 -0.2' ■34' '4'

-1.8 1-4. .48. .6.
0.4 -0.2’ ■ 3' r
-1.8 1.4. .11. . 10.
0.4 -0.2 ■ 10" 4
-1.8 1-4. 0. -18
0.4 -0.2' ■21' '3'
-1.8 1-4. .27. .0.
Use of the inverse matrix as outlined above is not the most rapid
means of obtaining a once-only solution to a linear system, but
advantages follow if the inverse of the coefficient matrix is available. For
example, it can be useful to know the range of values of the right-hand
side for which the solution values are non-negative - a situation which
arises in linear programming. Suppose that Xi and Xj are production
levels and the right-hand side values are resources. Suppose that 48
units of the second resource are obtainable, but that availability of the
first resource is variable: for what range of values of the first resource
are both Xi and X2 non-negative? To answer this question, let the level
of the first resource be bi. The system is then
■7 r 'xi' 'h'
9 2 X2 48
and the solution values of Xi and X2 - along with the statement of sign
requirements - will be
0.4 0.2' 'bi' 0

- 1.8 1.4 48 0
We can read directly from this that
= OAbi -0.2(48) > 0
which means that
bi > 24
Similarly, as regards X2
X2 = -l.Sbi + 1.4(48) > 0
which means that
bi < 37.33
So, with other data at original levels, for the solution values of both
variables to be non-negative the amount of the first resource must lie in
the range
24<bi< 37.33
This range of values is called the tolerance interval or range of

feasibility for the parameter bi. As a check on this result, note that
■ 0.4 -0.2' '24' ■ O'

-1.8 1-4. .48. .24.
■ 0.4 - 0.2' '37.33' '5.33'
-1.8 1-4. 48 0
Similar calculations can be carried out for the second resource.

Suppose there are 34 units of the first resource. Then permissible values
for the second resource must be such that
Xi' 0.4 -0.2' ■34 O'

>
X2 -1.8 1.4 bi 0
that is
= 0.4(34) - 0.2/72 >0

from which
b2 < 68
The required non-negativity of X2
X2 = -1.8(34) + 1.4^2 > 0
implies that
1.8(34)
b2 > 43.71
1.4
In summary:
43.71 < ba < 68
which is the tolerance interval for ba.

We have considered the scope for variation in bi and ba alone in
examples of single parameter analysis. It is also possible to conduct a
joint analysis, in which statements about the relative sizes of bi and fea
result. It is required that
Xi 0.4 -0.2' 'bi' O'

>
X2 -1.8 1.4 ba 0
from which it emerges that to ensure non-negativity of Xi, bi and ba

must be such that
OAbi -0.21)2 > 0
that is:
ba < 2bi
To make sure that X2 does not fall below zero it is necessary that
-1.8bi + 1.41)2 > 0
or (approximately)
1)2 > 1.291)1
Neither Xi nor X2 will be negative so long as the relative sizes of bi and

ba satisfy
1.29bi < ba < 2bi
Figure 3.1 shows this result.

The two limiting relationships between the values of ba and bi are
shown as lines through the origin with respective equations
ba = 2bi and ba = 1.29bi
The permitted region is on and between the lines. The individual range
of variation for bi given a value of ba is found by taking a horizontal
line into the diagram at the given height. Points of intersection with
the rays then give the end-points of the individual range for bi.
Figure 3.1
Exercises 3.6
1 Solve, if possible, the following systems of simultaneous linear
equations using the inverse of the matrix of coefficients:
(i) 4 ^1 55
1 ^2 20
(ii) 2 120
1 3j [x2_ 10
(iii) 5 9irxi 45
24j [x2 20
(iv) 9 61 [xi 30
6 4j [x2_ 20
(v) 2 -71 rxi -18
2 -2} [X2 -8
(vi) 9 -6' xi 30
-6 4j [x2_ 20
(vii) 8 10 xi 189
4 6 X2 106
(viii) -6 0.4
0.8 -0.2j[x2.
2 For
'4 7' ’-^1 b]'

2 6 .^2. .bi.
solve for the following right-hand-side values:
(i) b] 67
L^2 J 46

(ii) 'b]' '37'
>2. .36.
(iii) 'bi ■ ■ 8'
.bi. .-6.
(iv) 'b]' 143
-
.b2_ .109
(v) b]' 133
>2. 114
(vi) 'bi 1 -230
>2. .-340
(vii) 'b]' '220 ■
>2. 16C
3 For the following system:
■4 5‘ bi
2 3 .^2. .b2_
(i) Solve the system for b] = 90 and 62 = 50.

(ii) With b2 = 50, for what range for bi are both X] and X2
non-negative?
(iii) With b] = 90, for what values of b2 are X] and X2 non¬
negative?
(iv) For what range of relative sizes of bi and 62 are the
values of xi and X2 both non-negative?
3.7 The system of equations:

Rank
'2 3' Xi 'h'
4 6 _%2 _ .^2.
does not have a unique solution, since the matrix of coefficients
is singular. But if the equations are consistent it may be possible to

solve for one variable in terms of the other. For example, if
'2 3‘ Xi '12'
4 6 _X2. 24
we can write
,2
X2 = 4 - 2
Matrices 111
But note that in the case of
’0 r 'xi r
0 1 X2 1
Xi /Xj can have any value. The original system of equations is, in a
sense, not as big as it seems. While the matrix A is of order 2 x 2, it
has rank 1. The rank of a matrix is the largest number of its rows or
columns which are linearly independent. If the rank of a square
matrix is less than its order, there will not be a unique solution to the
corresponding linear equations. Consider some examples of rank.
(i) The matrix
'2 -1 4‘
6 5 3
_8 4 7_
is of order 3x3 but has rank = 2. The linear dependence is most easily
seen in terms of the rows; the third row is the sum of the first two. This
does not mean that it is the third row in particular which is dependent.
Any one of the rows can be expressed as a linear combination of the
other two (row 1 = row 3 - row 2) and can be dropped from the
system without losing information. There will be the same degree of
linear dependence between the columns as between the rows of a
square matrix. In case (i) the link between the columns is less obvious,
but inspection reveals that:
col 3 = ^ col 1 — ^ col 2

16 16
If the rows (or columns) of a matrix are represented by the vectors
® 2) ■ ■ ■
and there can be found m numbers
Wi, W2,---Wm
not all of which are zero, and such that
IViOi + W2fl2 3-1- = 0
then the vectors making up the matrix are linearly dependent. If no such
numbers Wj exist, then the rows (or columns) are linearly independent. A
collection of linearly independent vectors forms the basis of a set of
vectors if every vector in the set is a linear combination of vectors in the
collection. This may seem obscure, but the idea of a basis is important in
linear programming. The columns of coefficients of variables in a basic
solution can be used to generate the columns of coefficients of variables
currently zero. For example, given the three vectors:
3 1 4
2 0 2
4 15

then the two vectors
3 and 1
2 0
4 1
form a basis for the three-vector system as a whole, since the third
vector can be formed from a linear combination of the first two (it is
their sum in this case). Of course, bases are also formed by the other
possible pairings of vectors. Thus, if the vectors
3 and 4
2 2
4 5
are selected, the second vector is the difference between the third and
the first. If the vectors selected are
1 and 4
0 2
1 5
then, from this basis, the first vector is obtainable by taking the
difference between the third and the second vectors. Consider two
further examples of matrices where the order is greater than the rank.
(ii) The matrix
■ 2 1 7 4'
6 8 4 6
-15 9 2
3 4 2 3
is of order 4 x 4 but has rank = 3. In this case the last row is half the
second row. Given consistency, it would be possible to state three of the
variables in terms of the fourth. This situation arises quite commonly in
economics, where the variable used as the parameter is determined
outside of the system (it is exogenous). The in-system (endogenous)
variables can then be given specific values.
(iii) The matrix
' 9 7 8 6‘
2 10 3
7 6 8 3
11 8 8 9_
is of rank 2. In this case any collection of more than two of the four
rows (or columns) shows linear dependence. The row relationship can
be expressed as
row 3 = row 1 — row 2

row 4 = row 1 -|- row 2
Two of the variables could, given consistency, be expressed in terms of
the other two. For a system to have a definite solution rank must equal
order, but there is no easy test to establish the rank. Without prior
information, the fact that the rank of a matrix is less than its order is
often found through the determinant being zero. Rank does not only
apply to square matrices. For example, the matrix
’2 1 4 0.5 3'
4 2 8 1 6
is of order 2x5 and rank 1. In terms of the rows, the second row is
twice the first. This is the row rank. The column rank must be the same
as the row rank, which means that in this case there is only one
independent column, of which all others are multiples. The matrix
■ 2 3 4 1.5'
-1 7 -2 3.5
4 6 8 3
has order 3x4 and rank 2, since there are only two linearly
independent rows or columns.
Exercises 3.7
1 In the following systems, state the rank of the matrix of
coefficients, A. What can be said about the relationship
between the solution values, if any, of the variables in each
case?
■
■
(i) 2 -1 'xi ■ 20'
.-1 0.5. .X2. .-10.
(ii) ■9 16" 'Xl 1r
1-
1_
CM
.X2 5.
-6 12' 'xi ■ • 24'
4 -8. .X2. .-18.
2 What is the rank of the following:
'1 2 3'
2 4 6
4 8 12
'2a a -0.5 a
2b b -0.5b for a b,c^0?
2c c -0.5c
a b c
2a 3b 4c for 0, b, c^ 0?
2a 2b 2c

3 (i) Do the vectors
6 and 1
-2 4
3 -1
form a basis for
6 1 7 5
A= -2 4 2 -6
3 -1 2 4
(ii) What is the rank of A?

4 What is the rank of the matrix;
'42-1 0.5 -2
-8 -4 2-1 4
3.8 Applications may involve systems with anything from three to several
Higher-order thousand equations. Computer software is available to solve systems of
systems: linear equations and to obtain matrix inverses. While it is quicker and
determinants and less error prone to use packages such as Derive or Maple for larger
the inverse matrix matrices than to work out the inverse by hand, there is value in
knowing something about larger systems. Consider, for example:
'3 2 9' Xi ■49'

3 15 X2 = 33
2 14 X3 26
We will find the solution to this system using the inverse of the matrix
of coefficients (A). In obtaining the inverse of A we shall first find the
determinant. There are various ways in which the determinant can be
found, including a special procedure for 3 x 3 matrices. Here we will
use Laplace expansion, which evaluates a determinant as a weighted
sum of lower-order determinants. To use this approach we will need
two new concepts. A minor is the determinant of a submatrix of A. We
will need the minors associated with the individual elements, fly, of A.
For the element fly, this is the determinant of the submatrix obtained by
striking out the ith row and the jth column of A. For example, the
minor associated with an is mu where:
mu = -1
The minors associated with the elements uu and au are, respectively.
mu = 2 and mu = = 1
A cofactor is a signed minor. To obtain the cofactor, the minor

associated with an element is multiplied by -1-1 or -1 depending on the
position of the element in the matrix, the sign for element a^ being
Matrices 115
given by: ( - 1)''^^ Call the cofactors of elements Cij. The cofactors then
are
Cj; = ( - ly^'niij
The point of this is the following result:
The determinant of a matrix is the sum of products of the

elements in any row or column, and their associated cofactors.
So in the case of matrix A, using the first row
\A\ = fliiCii + auCij + fli3Ci3

= fliiuiii — aumu + ciisniis
= 3(-l)-2(2) + 9(l)
so the determinant is
|A| = 2
We can gain practice and confirm this result by expanding the

determinant using another row or column. For example, using the
second-row elements and cofactors the resulting calculations are
2
m2i = so C21 (-l)^^^m2i = l
1
3
^22 = so C22 (-l)' + 'm22 = -6
2
3
^23 = so C23 ( - l)^ + ^m23 = 1
2
So:
A\ = fl2lC21 + ^22^22 + (^23^23 — 3(1) "F 1( — 6) -|- 5(1) — 2
The same value results from using elements of the third row and their
cofactors. Similarly, the value of the determinant could have been
obtained by using elements and cofactors in one of the columns. But
note that the sum of products of the elements of one row and the
cofactors of another is zero. The amount of work involved is reduced if
there are zeros in a row or column. For example, the determinant of
2 4 0
B= 5 7 0
3 6 5
is most easily found by use of the third column:
\B\ = 5( - 1)^ + ^

The adjugate matrix is the transpose of the matrix of cofactors. For a
matrix of order 3x3 the adjoint is
cn C21 C31
C12 C22 C32
Cl3 C23 C33
The inverse is the scalar product of the reciprocal of the determinant

and the adjugate matrix. That is
Cii C2I C3I

A-^ C12 C22 C32
Cl3 C23 C33
In the present case the only missing information is the cofactors of

elements in the third row of A. These are obtained as follows
m3i = = 1 so C31 = ( - = 1
17132 = -12 so C32 = {-l)^^^m32 = 12
17133 = =-3 so C33 = ( - l)^'^^m33 =-3
So the inverse should be
■-1 1 1‘ ■-0.5 0.5 0.5'

-2 -6 12 -1 -3 6
2 -3 0.5 0.5 -1.5
1 1
-1.
Confirmation is obtained by postmultiplication of A by A
1-
■3
0
2 9' -0.5 0.5 0.5 ’1 0

3 1 5 1 -3 6 = 0 1 0
-1
0
2 1 4 0.5 0.5 -1.5 0

1
We can now solve the original equations. Using the inverse
Xi -0.5 0.5 0.5 '49' ■5'

X2 -1 -3 6 33 = 8
X3 0.5 0.5 -1.5 26 2
The inverse matrix is valuable when solutions are needed for various
right-hand-side values. Incidentally, note that for the order 2 matrix:
uii ai2
> A =
CI21 ^22
the matrix of cofactors is
0-22 —^21
—ai2 dll _
the transpose of which produces the adjoint.

Determinants have several properties it is useful to know:
1 The determinant of a matrix is unchanged if the matrix is

I TI
transposed: that is, |yl| = |i4 |.
2 Switching two rows or two columns of a matrix leaves the absolute
value of the determinant unchanged but changes its sign.
3 If any row (column) of a matrix is a weighted average of other
rows (columns) the determinant is zero.
4 Multiplication of all elements in a row or column by a constant
multiplies the determinant by that constant.
5 If to any row of a matrix is added a multiple of any other row, the
determinant is unchanged - so also for columns.
6 The determinant of an upper- or lower-triangular matrix is the
product of the elements on the main diagonal.
7 If A is of order n, then the determinant of the scalar product of the
matrix is the determinant of A multiplied by the scalar raised to the
nth power: \kA\ =
8 For square matrices of the same order, the determinant of the
product of the matrices is the product of the individual
determinants: \AB\ = |A||B|.
9 Where A is invertible = (|A|)~^.
10 If the elements of any row (column) are multiplied by the cofactors
of any other row (column) the result is zero.
Exercises 3.8
1 Given the matrix:
'7 5 9
A= 3 8 4
6 2 1
(i) Find the minors of the following elements:
(o) 033 (b)aii (c) 022 (d) 032 (e) 021
(ii) Find the cofactors of the minors in (i).

2 (i) Using the elements and cofactors of the first row, find the
determinant of
‘2 4 3
A = 4 5 1
9 2 4
(ii) Confirm the outcome of (i) by expanding the determinant

by the third column.
3 Find det A for
(i) rZ 9 8‘
A= 1 0 2
5 3 4

(ii) 3 2
5 8
-1 4
4 Find the inverse of

'1 12
A= 5 4 9
3 2 3
5 Find the inverse of

2
0
1.5
3.9 In this section the use of determinants and inverse matrices in the
Simultaneous solution of linear equations is taken further. For a numerical example,
equations (ii) we shall use the inverse matrix approach to solve the system
4xi + 7x3 = bi
7x\ + 3X2 + 5X3 = ^2
4xi + 2X2 + 3X3 = ^3
for the following sets of right-hand-side values;

_1
-1
■34'
1
'
'h'
bo
LO
00
bi 19 and
1
.^3. 35 12 -2_
The matrix of coefficients is
4 0 7
7 3 5
4 2 3
There are no obvious dependencies between the rows of A and we begin

by finding the determinant. It is most convenient to expand by the first
row, which contains a zero. The determinant is
|A| = 4cii -|- 0ci2 -F 7ci3
The minors and cofactors are
3 5
mil = = -1 so Cii = ( - l)^^^mii = -1
2 3
7 5
^ni2 = so Cl2 = { - l)^^^mi2 = -1
4 3
Although it is not needed for the determinant, if |A| is not zero we

shall need Ci2 later on.
7 3
mi3 = = 2 so ci3 = ( - l)^'^^mi3 = 2
4 2
Matrices 119
So the determinant is
I A| = fliiCii + ai2Ci2 + — 4( — 1) + 0( — 1) + 7(2) — 10
The matrix is non-singular and each vector of right-hand-side values

gives a unique solution. Remaining cofactors are now found. For the
second row
mi = = -14 so C21 = ( - = 14
m2 = — 16 so C22 = ( l)^ + ^m22 -16
mi = so C23 = ( l)" + '^n23 = -8
and for the third row
mil = -21 so C31 — 1 = -21
mu = = -29 so C32 = 1i-D^^^mu = 29
mu = = 12 so C33 = *( - l)^ + ^m33 = 12
So the inverse is
■-1 14 -21‘ ■-0.1 1.4 -2.1'

4-1 = J_
-1 -16 29 -0.1 -1.6 2.9
10
2 -8 12 0.2 -0.8 1.2
The result can be checked by pre- or postmultiplication of 4 by 4 \ In

the latter case
■4 0 7' -0.1 1.4 -2.1 '1 0 o'

7 3 5 -0.1 -1.6 2.9 — 0 1 0
4 2 3 0.2 -0.8 1.2 0 0 1
We can now use the inverse to solve the system of linear equations
for the three sets of right-hand-side values. For the first right-hand side
'xi' -0.1 1.4 -2.1 '47' ■3'

X2 = -0.1 -1.6 2.9 58 = 4
Xi 0.2 - 0.8 1.2 35 5
For the second set of right-hand-side values the solution will be
'xi' -0.1 1.4 -2.1 '34' ■-2'

= - 0.1 - 1.6 2.9 19 1
0.2 - 0.8 1.2 12 6
while for the third group of values the resulting x, work is
Xi - 0.1 1.4 - 2.1 ■ 14' O'

X2 - 0.1 -1.6 2.9 -2 -4
.^3. 0.2 -0.8 1.2 -2 2

The inverse matrix is valuable in assessing the impact of changing
circumstances. The right-hand-side values may be resource availabilities
and Xi may represent production levels. If current levels of the resources
limit the achievement of the firm’s objective it will be someone’s job to
increase availability. In many cases the model will give information on
the profit impact of changes in resource levels. In this context it is
important to know how wide a variation in the bi the current solution
can tolerate without some variables becoming negative. Consider the
first set of right-hand-side values in the example, in particular bi. The
range of values of bi for which all x,- are non-negative satisfies
0.1 1.4 -2.1' ■ bi' 'O'

0.1 -1.6 2.9 58 > 0
0.2 -0.8 1.2 35 0
The requirement on bi arising from the non-negativity of Xj is found by

multiplying the first row of the inverse by the right-hand-side vector
including the unknown bj. This produces the requirement from Xj
-0.1bi + 1.4(58)- 2.1(35) >0
-O.lbi + 81.2 - 73.5 > 0
which implies
bi < 77
and from X2
-O.lbi - 1.6(58)+ 2.9(35) > 0
i.e.
-O.lbi - 92.8 + 101.5 > 0
which implies
bi < 87
and from X3
0.2bi - 0.8(58) + 1.2(35) > 0
i.e.
0.2bi -46.4 + 42 > 0
which implies
bi > 22
Of the two upper bounds, 77 is the least, and so gives the tightest
restriction, and in conjunction with the one lower bound the overall
requirement on bj for sign feasibility of all variables is that
22<bi<77 (3.9)
Matrices
121
The range (3.9) is the tolerance interval or range of feasibility
for bj. The tolerance interval for X2 can be found by similar means from
the second row of the inverse into the right-hand-side vector, in which
b2 is substituted for the original specific value of 58. The resulting
tolerance interval is
55.86 <b2< 60.5
Note that this is a much narrower range than for bi. The feasibility of
the solution is more sensitive to variations in b2 than in bi. This is
likely to be an important concern for management if ^2 is not a control
variable - being, perhaps, the subject of random variation, external
regulation or the actions of competitors. For the right-hand-side
element bj, the tolerance interval works out as
33.62 <b3 < 36.43
Here again, the permissible range of variation of the resource level is

much narrower than for bi.
Exercises 3.9
1 Solve the system
'2 1 O' ’xi ■ ■ 35'

3 2 4 140
0.5 1 5 .^3. 120
2 Given
'1 1 3‘ ’xi ■ b]'

4 7 2 ^2
= b?
3 5 4 .^3. .bs.
solve for the values of the x; when
(i) b] 85
62 185
b3 185
(ii) bi 70
62 50
bs 90
(iii) ibi -4
62 139
bs 78
(iv) b] 9
b2 63
bs 45

3 For the data of problem 2(i) find the ranges of variation for
each right-hand side element individually for which xi, X2 and
X3 are all non-negative.
3.10 Another way of finding the solution to a system of linear equations is

Cramer's rule by use of Cramer’s rule, in which the values of the x,- are found as the
ratios of determinants. The method proceeds as follows. For the system
flu fli2 fll3 Xi 'bi'
fl2i fl22 <^23 X2 = b2
_fl31 fl32 t?33. X3
the values of the Xj are given by
bi <^12 fll3
b2 ^22 ^23
bz ^32 fl33
Xi
1^1
flu bi fll3
fl21 b2 «23
fl31 bs ^*33
X2
1A|
flu fll2 bi
fl21 «22 b2
fl31 ^^32 bz
X3
|A|
In each case the numerator is the determinant obtained by replacing the

column of coefficients of the variable to be obtained by the vector of right-
hand-side values. The denominator in each ratio is the determinant of A
itself. Now apply Cramer’s rule to the system of section 3.9. We have
already found that jAj = 10. Using the first set of right-hand-side
values, the value of Xi is given by Cramer’s rule as
47 0 7
58 3 5
35 2 3
Expanding the numerator by the second column, this works out as
7
5
= 3( - 104) - 2( - 171)
= 30
123
Matrices
so:
Xi = 30/10 = 3
and Xi will be
4 47 7
7 58 5
4 35 3
Again expanding by the second column, the numerator determinant is
7 5 4 7 7
-47 + 58 - 35
4 3 4 3 5
= - 47(1) + 58( - 16) - 35( - 29)
= 40
so
X2 = 4
and X3 is
0 47
3 58
2 35
X3
10
Using the second column to expand, the determinant works out as
4 47 4 47
4 35 7 58
= 3(-48) 2(-97)
= 50
so X3 = 5 and the solution to the system is
Xi = 3 X2 = 4 X3 = 5
Cramer’s rule is equivalent to use of the inverse matrix but does not
require the complete inverse to find values for particular variables. The
rule is a useful means of solving systems of simultaneous linear
equations. Because it finds solution values for variables individually,
Cramer’s rule is efficient when values of all unknowns are not required.
Exercises 3.10
1 Use Cramer's rule to solve the following linear equations:
\
-9 1 -
■^1 ■ ■53-
.6 2. .^2. .46.
■4 5' ■î ■ ■26'
.1 3. .^2. 3.

2 Solve the following system by Cramer's rule:
2 3 r 'xi ‘ ■16‘
4 1 5 X2 = 34
2 6 3 .^3. 30
3 Use Cramer's rule to find the value of X2 satisfying the

following simultaneous equations;
X] - 22x2 + 7x3 = 107

xi + 29x2 + 8x3 = 61
-2x1+ 12x2+4x3 = -92
3.11 In this brief introduction to matrices the focus has been on systems of
Concluding simultaneous linear equations. In this chapter and in Chapter 2 we
remarks discussed other methods of solving such systems. No single method is
the most appropriate for all applications, but there are circumstances in
which each technique is likely to work to advantage. Less systematized
approaches, such as elimination or substitution, are quick ways to solve
2x2 systems. They can also be practicable in 3 x 3 cases, where a
convenient feature of the system can be exploited. But informal
approaches are not recommended for larger systems. The use of
determinants, as in Cramer’s rule, has advantages if the values of only
some unknowns are required. Gauss-Jordan elimination, or a variant
making use of elementary row operations, is highly efficient if a one-off,
complete solution is required. If a solution is required for more than
one set of right-hand-side values, or if tolerance intervals for right-
hand-side values are needed, then the inverse matrix is valuable.
The number of arithmetic operations needed to find the solution to a
system of n equations in n unknowns varies roughly with the cube of n.
In comparison to a system of order three, over twice as much work is
involved in solving a system of order four, and about five times as much
work is required to solve a system of order live. So it is as well that
there are software packages (such as Derive, Maple, MathCad, etc.) for
evaluating determinants, finding inverse matrices and solving systems
of linear simultaneous equations. On a modern PC inversion of a
100 X 100 matrix takes but a few seconds. Mathematics software is
well worth investigating if you do much work in this area.
125
Matrices
CHART E R
4 Functions and turning points
In this chapter we introduce the concept of a function and study the

properties of a range of functions used in the management and social
sciences. We focus on polynomials, especially the simpler forms -
quadratics and cubics - and functions which are the ratio of
polynomials. Exponential functions, which are also very important in
finance and economics, are considered in Chapter 9.
LEARNING OUTCOMES
After reading this chapter you will be able to solve quadratic equations
and mixed systems of one linear and one quadratic equation. You will
learn principles of curve sketching, and consider business and economic
applications of quadratic, cubic and rational functions.
4.1 Managers and economists deal with measurable entities - for example
Functions prices, quantities, rates of interest, inflation and growth, profits and
rates of taxation - and try to understand the relationships between
them. There are many possible objectives, including predicting the
consequences of decisions, influencing or controlling outcomes, or
ideally optimizing the values of key decision variables. Where an
optimization model is used the objective will be to achieve the best
value of some measure of performance or effectiveness. For example, a
firm may wish to maximize its profits, or a contractor may wish to
minimize cost or the completion time of a project. The concept of a
function arises in most theoretical or empirical studies of relationships
between variables whether or not the relationships are initially
expressed in symbolic terms.
A function can be seen as a rule which links together a given
number, ac, and another single number, the function of x - written as
f(x) and read as f of x’. Formally, a function is a relation between two
sets that associates a unique element of the second set with each
element (or n-tuple of elements in the case of functions of several
variables) of the first set. If the rule represented by /(x) gives more than
one value for a given x, as for example when a square root is taken, in
strict terms it does not satisfy the definition of a function - which

requires a unique outcome. But the formal requirement can often be
met by focusing on a particular part of the relationship.
The rule or rules for the function could in principle be set out in
terms of literal instructions for a series of operations to be performed on
X to obtain the number/(x). For example, the instructions could be to
double X and add 5, which would be written in function notation as
/(x) = 2x + 5
or to ‘square x and subtract 10’, which would appear as
/(x) = x^ — 10
But a function such as
/(x) = x^ — 8x^ + 7x^ — 2.5x^ + 2x — 4
would be very cumbersome to spell out. We frequently use another

symbol in place of /(x), most often y. In the function
y = — 3x — 10 (4-1)
X is the independent variable or argument and y is the dependent

variable or the subject of the equation. In graphing a function, the
convention is to measure the dependent variable against the vertical
axis. Classification of variables as independent or dependent is natural
when there is a causal link between x and y. Where no such
relationship exists the explicit form (4.1) may still be used for
convenience, although the implicit expression of the relationship
between the variables
x^ — 3x — 10 — 1/ = 0
makes no distinction as to dependency, and removes the asymmetric

treatment of x and y. Where a causal relationship exists, x may be
called a decision variable or control variable in management
science or, in macroeconomics, an instrument if its value can be
varied by the decision-maker. The subject of the equation, y, would be
referred to as the objective, while the function itself (which one may
wish to optimize) is the objective function.
There may be limits on the values which the dependent variable can
take. These limits may be defined by the context of a particular
application. For example, output may be limited from above by plant
capacity and from below by the need at least to satisfy the most
important customers. In other cases the variable x may have no
meaning if negative, or it may be impossible to carry out the operations
required by the function for certain values of x. The set of permissible
values of x is called the domain of the function. The set of values of y
which result when the function is applied to values of x in the domain
is called the range. For the relationship between x and y to satisfy the
definition of a function, each x value in the domain must produce a
Functions and turning points 127

unique value of y in the range. But it is allowable for different values of
X in the domain to result in the same y value in the range. The
function relationship can be shown as an input-output process in the
following way:
INPUT PROCESS OUTPUT
An X value from The operations A unique y value in

the domain required by the the range
function f{x)
Consider some examples of possible domains and ranges. Three cases

are shown in Figure 4.1. In the first case the function is
y = x^
for which the domain is unlimited (minus infinity, shown as — oo, to

plus infinity). Real numbers must produce a non-negative square so
that the range is restricted to all positive numbers and zero. This can be
shown as
Domain — oo < x < oo

Range y >0
This case is illustrated in Figure 4.1(a). The domain is the real numbers
for which the operation of squaring can be performed - all real
numbers. Since the square of a real number cannot be negative, the
range of the function is, by its own nature, restricted to the positive real
numbers and zero.
In the second case the function is
y — +V4: — 2x
(meaning that only the positive square root is involved in the function).
The domain and range are
Domain — oo < x < 2

Range y >0
This case is illustrated in Figure 4.1(b). The operation required by the

function cannot be performed on numbers in excess of +2. So here it is
the domain which is restricted by the form of the function. The range
will also be limited by the function, being the set of non-negative real
numbers.
In the third case the function is
y = 1.2x + 7
and the domain and range are
Domain (specified) — 5 < x < 20

Range 1 < 1/ < 31

Figure 4.1
^ Range
0
>■
X
Domain: -oo < x < +=»

(a)
(b)
This case is illustrated in Figure 4.1(c). In this case a restricted domain

has been ‘artificially’ imposed. This may reflect realistic values in an
application. For example, x might be a change in stock level. If the
current inventory is 5 units, then the fall in stock cannot be greater than
5. If the maximum storage capacity is 25 units, then the rise in stock

cannot exceed 20 units. When the domain of a function is restricted in
this way, if the function is always increasing or always decreasing (is
monotonic), the end-points of the range can be obtained by substituting
the end-points of the domain into the function. In practice there is often a
restricted or target range for the output of the function — as, for example,
when a y value must be achieved within specific limits of accuracy. For
example, a non-profit organization aims to break even on average but in
any year may wish its excess of income over outgoings to be within the
range -£10 000 to +£10 000. A target range for a function will in turn
imply a restricted domain for the independent variable.
A piece-wise defined function is an interesting form that reflects
situations where the nature of the relationship between the variables
changes at certain points in the domain. One example of a piece-wise
defined function is
which is graphed in Figure 4.2. Solid disks show points in the range
where the form of the function changes. An unfilled disk indicates a
point where the value of /(x) is not included in the range. So with the
uppermost line segment the end-point at x = 15 is excluded.
Figure 4.2 fM A
>■
X
Although a piece-wise defined function at first looks strange, it fulfils

the requirements for being a function. Indeed, far from being contrived,
it is a form which corresponds to many practical situations in business.
For example, the function may represent a manufacturer’s long-term
costs of production for various output levels. The firm may have a
choice between two plant sizes or may choose not to operate at all. If

the decision is made not to produce, long-term costs will be zero.
Where production does occur the 2x -\-IQ component of the function
shows the costs arising in the smaller of two plant sizes the
manufacturer may use. This size has relatively low fixed costs of £10
which are inescapably associated with any positive level of use and a
comparatively high unit variable cost of £2 up to the plant capacity of
15 units. The larger of the plant sizes has relatively high fixed costs of
£30 but only half the unit variable cost of the lower-capacity plant.
Note that this function exhibits step discontinuity against the y axis
at X = 0 and x = 15.
A further example of the relevance of piece-wise defined functions is
where a curvilinear expression is approximated by a series of linear
segments, as shown in Figure 4.3. Piece-wise linear approximation
represents a possible way of making problems that involve the
optimization of non-linear functions more manageable.
Figure 4.3 f(x)k
Exercises 4.1
1 In the absence of other restrictions, find the domain for the
following functions: ^
(i) y = (x+10)^ (ii) y = (40-2x)2
2 What is the range for the following functions:
(i) y = x^ (ii) y = 2x2 - 10
3 With the given domains, find the range for the following
functions:
(i) y = 3x-15 domain:-5 < X < 20
(ii) y=100 —5x domain:—10 < X < 100
(iii) y = 250 - 1 Ox domain: -200 < x < -100
4 Do the functions below have step discontinuities against the y
axis?
f/ 1 / X + 10 for X < 30
^ 1 2x — 20 for X >30
131
Functions and turning points
5x + 5 for 0 < X < 15
2x + 60 for X > 15
4.2 If the dependent variable i/ is a quadratic function of x, this can be

Quadratic functions written as:
y = ax^ + bx + c (4.2)
in which a, b and c are real constants and a, the leading coefficient,

is non-zero. Where y is the dependent variable the exponents of the
terms in x will be 2, 1 or 0 (in the case of the constant term). Later we
state the general equation of the second degree, but will concentrate on
expressions of the form (4.2). The graph of a quadratic equation is
called a parabola. Parabolas are U-shaped curves and, with y as the
dependent variable, the parabola makes an upright or inverted U shape
depending on the sign of the leading coefficient.
Figure 4.4 shows the effect of varying the parameter a in the
parabola y = ax^. High and positive values of the leading coefficient
produce a steeply sloping parabola which opens upwards. In this case
the extreme point of the parabola, the vertex, represents a minimum
for the function. Negative values of a mean that the parabola opens
downwards and the vertex is a maximum. The parabolas of Figure 4.4
are symmetric about the y axis - the axis of symmetry of the
parabola. All parabolas given by the equation
y = ax^ +bx-\-c
have axes of symmetry parallel to the y axis.
Figure 4.4

Figure 4.5 shows the effect of varying the value of the constant term
c in the parabola
y = ax^ + c
Here, c gives the y intercept, and the effect of varying it is to shift the
whole parabola up if c increases or down if c is reduced.
An important effect of such variations in the constant term is to

change the points (if any) where the curve cuts the x axis. Note that
the slope of the curve at any point is unaffected by the changes in c.
Changes in the value of the x coefficient, b, in the quadratic
y = ax^ + bx + c
move the vertex of the parabola relative to both of the axes, although
the parabola remains upright. As b is decreased, the algebraic value of
the slope is decreased. So if at a point the slope is positive to begin
with, the curve becomes less steep. If the slope is negative, then the
curve will decrease more rapidly at the given point. While variations in
the value of c move the parabola relative to the vertical axis, and
changes in b move the parabola relative to both axes, the curve can be
shifted to the left or to the right alone by replacing x with x - k (which
produces a shift to the right) or with x + k (to produce a shift to the
left), where k > 0 in both cases. Making such a change to the
independent variable affects both b and c in such a way as to give a
horizontal movement only.
These shift effects can be confirmed with the parabola
y = ax^ + bx + c

Substitute x - k for x to give
y = a{x — k)^ +b{x — k) + c
which must represent a rightwards shift, since any original value of x

must now be increased by k in order to maintain the original value of
y. Expansion of the brackets gives
y = ax^ — 2akx + ak^ + bx — bk + c

= ax^ {b — 2ak)x + c+ (ak^ — bk)
which shows the changes to b and the constant term needed to effect a
shift of the whole curve k units to the right. Consider the simplest
parabola, y = x^. A one-unit shift to the right is represented by
= (x — 1)^ = x^ — 2x -I- 1
Figure 4.6 illustrates the shift as a two-part process. Figure 4.6(a) showing
the original function, y = x^. Figure 4.6(b) shows the effect of the
subtraction of 2x: this moves the parabola across and down. In Figure
4.6(c) the constant. 1, has been added, with the original function indicated
by the broken line. The overall effect is a lateral movement only. The effect
on the slope of the function for any given value of x can be seen in Figure
4.6(c). For any value of x less than i, the function y = x^-2x+lis
further from its vertex than is y = x^, and is steeper in absolute terms. For
any value of x greater than ^ the function y = x^ — 2x + 1 is nearer to its
vertex than is ^ = x^ and has a lesser slope.
The X intercepts are important in most applications. These are the
points where y = 0 and are the zeros of the function, and the
corresponding values of x are the roots of the equation
ax^ + bx-h c = 0
One way to obtain the roots of a quadratic equation is by factoring. This

can be the best approach when the coefficients a, b, and c assume
convenient values. To factor a quadratic expression equated to zero is to
express it as the product of two linear terms, at least one of which must
therefore be zero if the equation is to be satisfied. For example, the
equation
x^ - lOx + 21 = 0
factors as
(x - 3)(x — 7) = 0
so the roots are
X = 3 and x = 7
In other words, it is these values of the independent variable which

produce the zeros of the function
y — X^ — lOx + 21

Figure 4.6
Where a quadratic equation does not factor in a fairly obvious way,

the quadratic formula is usually the best way to find the roots. The
formula is
-b ± (b^ - 4ac)2 ^
x =--- 14. J
2a
The quadratic formula can always be used to solve for the exact values
of the roots, and can be slightly rearranged as
—b (b^ — 4(ic)2
^ ^-
2a 2a
which shows more clearly that the two roots of the quadratic will be
equal distances of
(b^ — 4flc)2
2a
on either side of the vertex. The vertex itself, as suggested by the fact
that it lies on the axis of symmetry, is located centrally between the
roots at -~b/2a. The expression
b^ — 4ac
is the discriminant.

Three cases are distinguished in terms of the roots by the relation of
the discriminant to zero. These are as follows:
Case 1: — 4flc > 0
Here the discriminant is positive and the quadratic has two distinct real
roots. In geometric terms, the parabola crosses the x axis at two
separate points. As an example, consider the quadratic equation
y = — 3x — 10
On equation to zero, this factors as
(x + 2){x — 5) = 0
so the roots are located at x = — 2 and x = 5. The function is graphed

in Figure 4.7.
Figure 4.7
Case 2: — 4ac = 0
A zero value for the discriminant means that the two roots are equal
and, as can be seen from the quadratic formula, will be at the vertex
—b/2fl. This is the case of double roots. In an equation of higher
degree, a root occurring three times is a triple root. In general, a root
occurring n times, where n > 2, is said to be a multiple or repeated
root. In contrast, a single root is sometimes called a simple root. A
case of a quadratic with double roots is illustrated in Figure 4.8, which
is drawn for the equation
y = x^ — 6x + 9
Factoring and equating to zero gives
(x — 3)(x — 3) = 0
There is a single zero of the function corresponding to the double root

of X = 3. The parabola just touches the horizontal axis, rather than
intersecting it.

Figure 4.8
Case 3: — 4ac < 0
When the discriminant is negative there is no solution to the equation

in real numbers. The parabola has no point in common with the x axis.
This case is illustrated in Figure 4.9, which is drawn for the equation
y = + 6x + 13
Figure 4.9
Application of the quadratic formula in this case produces the x values
-6± v^(36 - 52)

x =-^-
that is
X = — 3 + 2i

and
X = —3 — 2i
where: i is the imaginary number So the roots of this quadratic,

while distinct, are complex numbers (which are made up of a real
part and an imaginary part). Note that the real component of the
complex roots, which will be —b/2a (—3 in this case), corresponds to
the vertex of the parabola, which is always located at x = -bj2a
whether the roots of the equation are real or complex.
It is often useful to make a sketch of a function as part of the process
of analysis. For quadratic functions, points on the graph which are
helpful for curve sketching can be produced by several criteria, of which
we will use five for the time being. These are:
1 The y intercept. This is the point where the graph cuts the
vertical axis, and is found by setting x equal to zero.
2 The X intercepts. These are the points, if any, at which the graph
cuts the horizontal axis (so that y is zero) and can be found by use
of the quadratic formula.
3 The turning points. The graph of a quadratic has a single
turning point, the vertex of the parabola at x = —b(2a. This result
is convenient for quadratics, but calculus methods are usually
required to locate the turning points of other functions.
4 Extreme x values. These are used to show the ultimate directions
in which the function is heading. The function is worked out for
very large positive and negative values of x.
5 Other x values. Used as necessary, depending on the particular
case and the accuracy required, a selection of x values not covered
by points 1-4 above is plotted.
An additional step of value in curve sketching (the identification of any

asymptotes) is mentioned later. Also, use can be made of any known
properties of symmetry possessed by the function. However, steps 1-5
above, appropriately employed, are sufficient in most business and
economic applications. Now consider the use of these five steps to sketch
1/ = — 4x — 21
1 When x = 0,^ = —21, so that the y intercept lies below the x axis.
The co-ordinates of this point are (0, —21).
2 When y = 0 the quadratic formula (or factoring) results in
X = —3 and x = +7
The co-ordinates of the points are (-3, 0) and (+7, 0).

3 The X value at the vertex is
-(-4)
— +2
2(1)

and the corresponding value of y is —25. The co-ordinates of the
vertex are therefore (2, —25).
4 When x is sufficiently large (positive or negative) for any parabola,
the overwhelming influence will be from the term in x^. This is
called the dominant term, and in sketching any function it is
useful to look for dominant terms. Here x^ has a positive coefficient,
and, since x^ is positive for high x values at either end of the
domain, the function as a whole must be large and positive in these
regions. In terms of the sketch, this means that the curve will open
upwards.
5 ‘Moderate’ or obviously significant values of x are likely to be
useful. When the roots are not too close together, a selection of
points for x values between the roots will usually suffice. In the
current example a possible selection of moderate x values and the
corresponding values of y could be
X y
-2 -9
-1 -16
-|-3 -24
+5 -16
which should be sufficient for sketching purposes. The appearance of

the function is as shown in Figure 4.10.
Figure 4.10
For parabolas in general much can be learnt from the value of the
leading coefficient.
139
1 If a > 0 the parabola opens upwards and the vertex is a minimum
of the function.
2 If fl < 0 the parabola opens downwards and the vertex corresponds
to a maximum.
In historic terms, a quadratic equation of particular interest is
— X — 870 = 0
Using the quadratic formula, the x values produced are
X = 30 and x = —29
The interest of this equation is that its solution was first obtained 3900
years ago in Mesopotamia. It arises from a problem stated on a
Babylonian tablet dated 1900 bc. The problem was expressed as that of
finding the side of a square the area of which exceeds the side by 870.
Babylonian algebra could find the positive root or roots of a limited
number of quadratics. This amazing civilization was also able to solve
some equations of higher degree.
Consider an example of quadratic equations and parabolas in
business management. Suppose that a firm’s costs, C, are given by
C = 4q + 450
in which q represents both output level and sales. The product is sold at
a price, p, under demand conditions given by
p = 50 — O.lq
First consider the firm’s turnover. Turnover, R, is total receipts from

sales of the product and is also known as total revenue, sales or
income. Turnover will therefore be
R = pq
= 50q — O.lq^
As a quadratic equation this is written out in full as
R =-O.lq^ + 50q +0 (4.4)
in which form it is evident that
a = —0.1 b = 50 c= 0
Since the leading coefficient here is negative, the parabola opens

downwards. The y intercept is at (0, 0). The x intercepts are at (0, 0)
and (500, 0), while the vertex has an x co-ordinate given by
-V2fl = -50/2(-0.1) = 250
It is at this output that turnover is at a maximum. The value of R will

then be
R = -0.1(250)^ + 50(250) = 6250

But the firm’s objective may well be to maximize profits rather than
sales. Profit, tt, is the difference between income and costs and will
therefore be
7r = R-C
= 50q-0.1q^ -4q-450 (4.5)
= —O.lq^ + 46q — 450
Step 1 of the sketching procedure shows the y intercept at (0, -450)

and the quadratic formula shows that the zeros of the profit function
occur at
q = 10 and q = 450
These are the intercepts with the horizontal axis. The lower of these
levels (with co-ordinates (10, 0)) is the breakeven output level, while the
other zero is a point of theoretical interest only, being well past the level
at which a firm would choose to operate in practice. The vertex of the
profit parabola is at q = 230, at which the maximum profit of
TT = 4840 is made. The revenue and profit parabolas are shown in
Figure 4.11.
Figure 4.11
The profit curve always peaks at a lower output than the revenue
curve (so long as costs always increase with output level). This will not
be proved here, but it is a useful exercise to see how the subtraction of
the cost expression affects the values of b and a (and hence —b/2a) in
the profit equation as compared to the turnover expression.
We have already considered the solution of systems of linear
equations. Systems which contain even one equation with a single non¬
linear term are non-linear systems and are much less tractable than
linear systems — statements relating the number of solutions to the
of independent equations and unknowns no ionger apply.
141
Non-linear systems are less widely used in applications than linear
alternatives and we will confine the presentation here to simple cases.
4.2.1 Consider the solution of a mixed system of one linear and one quadratic
One linear and one equation. For example, suppose that
quadratic equation
4:X + 2y = 20
and that simultaneously it is required that
y= - 7x + 16
A solution procedure useful with mixed systems of this kind is first to

solve for y in terms of x in the linear equation. Thus we obtain
y = 10 — 2x
The next step is to substitute for y in the quadratic equation. This

produces
10 — 2x = x^ — 7x + 16
which on rearrangement becomes
— 5x -1- 6 = 0
which is satisfied for x = 2 and x = 3. The corresponding values of y

(obtained from either the quadratic or the linear expression) are i/ = 6
and 1/ = 4. Thus the two solutions to the mixed system are
X= 2 and y= 6
X= 3 and y = 4:
As a further example of the solution of a mixed system consider
2x + y = 7
and
y — x^ +4:X + 20
The result to be obtained is not one which could occur with two linear
equations. Substitution of 7 — 2x for y into the original quadratic
equation produces
x^ -1- 6x + 13 = 0
which has no solution in real terms. The solution values (as may be
confirmed) are the complex numbers
x = —3-|-2i and y=13—4i

X = —3 — 2i and y = 13 + 41

4.2.2 Systems that involve two simultaneous quadratic equations of the form
Two quadratic
y = ax^ + bx + c
equations
y = dx^ + ex+f
can also be solved, since they can be reduced to a single quadratic in x

by subtraction. For example, if
y = 3x^ — lx-2
and
1/ = — 2x + 16
subtraction gives
0 = 2x^ - 5x- 18
which solves in the normal way for
X = —2 and x = 4.5
with the corresponding values of y (obtained from either of the

quadratics) being
y = 24 and y = 27.25
Consider market equilibrium with a linear demand curve and a

quadratic supply function, with demand given by
D : p = 90 — 2q
where p = price and q = output (sales), and with market supply given by
S: p = 20 + 0.5q^
The graph of this system is shown in Figure 4.12.
Figure 4.12
Equating the value of price given by the supply and demand

equations produces
p^90-2q
= 20 + 0.5q^

so that
0.5q^+2q-70 = 0
+4^-140 = 0
which factors as
{q-10){q+14) = 0
Therefore, market equilibrium is reached when
q — 10 and p = 70
Note that the other solution to this mixed system (which does not
correspond to a practically realizable situation) occurs at the values
q — —14: and p = 118
A case where simultaneous quadratic equations can arise is in

breakeven analysis, where costs are a quadratic function of output level
and where a straight-line demand curve is faced. Suppose that demand
is given by
D : p = 200 — 2.5q
and costs are
C= 1600 + 1.5g^
Breakeven is the output level where sales revenue first covers costs. This
is where
R = 200q-2.5q^
= 1600 + 1.5g^
= C
thus
4q^ - 200^+ 1600 = 0

q^ - 50q + 400 = 0
so that
iq-10)iq-40) = 0
giving
q = 10 and q = 40
So the breakeven output level is 10. The situation is graphed in Figure

4.13. ..

Figure 4.13
4.2.3 In a quadratic inequality there will be a solution set of values of x

Inequalities which satisfy the inequality. For example, consider the relationship
> 3x + 10
Rearrangement gives
x^-3x-10>0 (4.6)
Taken as an equation, (4.6) solves for
X = 5 and x = —2
The function
/(x) = x^ — 3x — 10
is graphed in Figure 4.14. /(x) is non-negative provided that x < —5 or

X > 2, so the solution set will be
— oo < X < —5
2 < X < oo
Figure 4.14
As a further illustration, the relationship
4x -F 35 > 14 + x^

rearranges to
-x^ + 4x + 21 > 0
with the solution set
-3 < X < 7
Consider an economic example in quadratic inequalities. Suppose

that a firm’s average costs of production, AC, are given by
AC = 0.05x^ -8X + 350
where x is output level. The firm sells in a market where price is set at
p = 210. Management regards the product - one of several it makes -
in strategic rather than immediate profit-making terms, but it is not
company policy to sell anything at a loss. Given these conditions, what
range of output levels is available to management.? The requirements
mean that the range must be such that unit costs do not exceed unit
revenue: that is
O.OSx^ - 8x + 350 < 210
so that
0.05x^ — 8x + 140 < 0
which is more conveniently stated as
x^ — 160x + 2800 < 0
As an equality, this factors to
(x-20)(x- 140) = 0
so the range of output levels available is
20 < X < 140
This example is shown in Figure 4.15.
Figure 4.15

4.2.4 Just as the equation of a straight line could be expressed in general
General form form, so there is a general form of the equation of a parabola. When
the axis of the parabola is parallel to the y axis this is
Ax^ + Dx + Ey + F — 0 (4-7)
in which A ^ 0, E ^ 0 and A, D, E and F are real numbers. The general

form of the parabolic equation itself is a special case of the general
quadratic equation, which includes possible terms in y^ and xy:
Ax^ + Bxy + Cy^ +Dx + Ey + F = 0 (4.8)
in which not all of A, B and C can be zero, and in which all coefficients
are real numbers. The general form of a parabola with axis parallel to
the y axis emerges when B and C are zero. Parabolas symmetric about
axes not parallel to the y axis are generated when
B^ - 4AC = 0
and where the values of A, B and C are not all zero. For example, if
A = 0, B = 0 and D = — 1 the resulting equation can be written as
X = Cy^ + Ey + F
which describes parabolas with axes parallel to the x axis, and which
open to the right or to the left. The simplest such parabola is where
E = F= 0 and C=1
in which case
x = y^ or y = ^/x
The graph of ^ is shown in Figure 4.16. By limiting the range

to the positive root only, the requirements of a function are satisfied and
the curve as a whole can be seen as a plot of the functions
y = —^yx and y = +y/x
Note that in the general quadratic equation a circle results if
A = C and B = 0
an ellipse is produced if
B^ - 4AC < 0
and the outcome is a hyperbola if
’ - 4AC > 0
We discuss hyperbolas later under rational functions.

There are some very special cases of the general quadratic equation.
If the only non-zero coefficient is A the equation reduces to
Ax^ = 0

Figure 4.16
and the y axis is produced - a conic section can produce a straight line
rather than a curve, but such a section is said to be degenerate. In
the event that A, B and C are all zero, the equation (4.8) reduces to the
general linear equation - which is contained as a special case.
Exercises 4.2
1 Find the roots of the following quadratic equations;
(i) x2 - 7x + 12 = 0
(ii) x^+x-30 = 0
(iii) x^ + 9x + 14 = 0
(iv) x^ - 16x + 64 = 0
(v) x^ + 1 8x + 81 = 0
(vi) x^ - 7x + 0 = 0
2 Find the roots of the following quadratic equations:
(i) 2x2-14x +24 = 0
(ii) x2-2.3x+1.32 = 0
(iii) 2x2 _ 3^ _ 9 _ Q
(iv) x2 + 99x - 100 = 0
3 Find the roots of the following equations and comment on the
results:
(i) x2 -20x+ 100 = 0
(ii) x2+9 = 0
(iii) x2 - 10x + 29 = 0
4 Find the solutions to the following systems of one linear and
one quadratic equation:
(i) 4x + y = 20
y = x2 - 25x +118
(ii) 6x + 2y = 100
2x2-y-68x +550 = 0

5 Find solutions to the following systems with two quadratic
equations:
(i) y = 4x^ - 20x + 60
y = 3x^ — 5x + 10
(ii) 6x^ — 1 Ox + 20 — y = 0
4x2 - 5x + 18 - y = 0
6 Find the solution sets to the following quadratic inequalities:
(i) x2-2x-24<0
(ii) x2 + 3x-10>0
(iii) 3x2 + 10x + 50<2x2 + 21x + 26
7 A firm's demand curve is given by
p = 700 — 5x
where p = price and x = quantity produced and sold. Costs

are given by
C= 150 + 40x + x2
(i) What output level maximizes the firm's revenue from sales,
and what is the maximum revenue level achievable?
(ii) What output maximizes profit, and what is the maximum
profit made?
8 If market demand is given by
p = 720-4q
while market supply curve is given by
p= 160 + 0.25q2
find the market clearing values of price and quantity.

9 A firm's product sells at the price:
p = £10500
Its average costs are given by
AC = 0.1x2 -80x+ -17500
Within what range of output levels does revenue at least cover

costs?
4.3 If the dependent variable y is a cubic function of x, this can be written as

Cubics and quartics
y = ax^ + bx^ +cx + d (4.9)
4.3.1
Cubics where in (4.9) the leading coefficient, a, is not zero, and all the coefficients
- a, b. c and d - are real numbers. The simplest form of cubic equation is
y = ax^
the appearance of which is shown in Figure 4.17.

Figure 4.17
Figure 4.17(a) is drawn for a > 0, while Figure 4.17(b) shows the
appearance when a < 0. If a = 1 the resulting function is
y = x^
which is the cubing function and, if plotted accurately, it is evident

how steeply the graph rises. So it is usual to plot cubics using a
different unit on the vertical axis. The cubing function has a single zero
at X = 0, where the curvature of the function changes and, at the same

time, the slope becomes zero. All cubics change ‘curvature’ at some
point (they inflect) and have two turning points or none. A cubic
cannot have a single turning point. The cubing function can be seen as
a limiting case where two turning points have coalesced. Figure 4.18
graphs a cubic with two turning points. The equation is
y = — 2x^ — 24%
Note that with no constant term the graph of this cubic passes through
the origin, which is one of the three zeros of the function.
Figure 4.18
There is an analytical method for finding the roots of cubic equations

- a much more complicated version of the formula approach used for
quadratics - but it is rarely used and will not be detailed here. In
Chapter 5 the turning points of cubics will be found using calculus. But
there are special cases for which the roots of a cubic equation are easily
obtained. For instance, if one root is already known, then only a
quadratic equation needs to be solved to find the other two. This is the
case when a function passes through the origin, i.e. when the constant
term is zero. For example, in the function
y = x^ — 2x^ — 24x
because of the absence of the constant term the right-hand side factors
as
y = x{x^ — 2x — 24)
the bracketed part of which also factors to produce
y — x{x -I- 4)(% — 6)

Thus if ^ = 0, it must be the case that
X = 0 or X = —4 or x = 6
The appearance of this function, with the (local) high point before the
low point, is a property of all cubics in which the leading coefficient is
positive. If a < 0, the cubic with turning points will appear as in Figure
4.19, although the actual positions of the turning points need not be as
shown. The cubic in Figure 4.19 has one real positive root (where the
curve crosses the x axis) and two complex roots.
Figure 4.19
In economics, cubics are often used to represent cost functions,

relating total costs of production to output level. With appropriate
coefficients, cubic cost functions have reasonable properties (the cost of
producing a marginal unit falling at first and then rising later on). To
illustrate, suppose that a firm’s costs are given by
C = 0.2x^ - 18x^ + 800x+ 100 (4.10)
That part of 4.10 in the positive quadrant is shown in Figure 4.20.

This cubic inflects but has no turning points. The presence of a local
maximum would mean, implausibly, that total costs would actually go
down as output expanded over a certain interval. The intercept with
the vertical axis, 100, represents fixed costs.

Figure 4.20
C(x) ^
4.3.2 Where the dependent variable is a quartic function of x, this can be

Quartics written as
y = ax‘^ + bx^ + cx^ + dx + e (4.11)
in which the leading coefficient, a, is non-zero, and where a, b, c, d and

e are real numbers. Quartic functions have three turning points or one.
They cannot have two turning points, but turn they must. This is due
to the fact that the leading term (by which the value of the function is
dominated for values of x at either extreme) is an even power, so the
function must eventually assume similar y values if x is large and
negative to those when x is large and positive. If the function turned
twice the values at the extremes would be of opposite sign. For a
numerical example of a quartic equation, consider the function
1/= x^ - 125x^+ 2500 (4.12)
The function is sketched in Figure 4.21 (where larger units on the

vertical scale are implied). The function has zeros at
X = —10, X = —5, X =+5 and x =+10
Equation (4.12) is interesting in two respects. First, it is symmetric

about the y axis. This will be so if there are no odd powers of x.
Second, although (4.12) is a quartic it can be solved as a quadratic.
First make the substitution
z = x^
The original quartic equation in x then becomes the quadratic equation

in z:
z^ - 125Z + 2500 = 0

Figure 4.21
which factors as
(z- 100)(z- 25) = 0
so that
z = 100 or z = 25
and therefore
= 100 or = 25
thereby producing the stated results.

A special case of a quartic equation has only even powers. This is
known as a bi-quadratic - which equations can be solved using the
quadratic formula. The roots of quartics in general could be found
through formulas, but the method is very long-winded and it is much
better to use mathematical software such as Derive to obtain precise
values. For an economic example, suppose that a firm has fixed costs of
£7.29m and an excess of revenue over variable costs of
90003C^ -
where x = the level of output and sales.

Profit, TT, will then be given by
7r= 9000x^ - - 7 290 000 (4.13)
for the domain x > 0. The profit function is as shown in Figure 4.22.
Substitution confirms that breakeven is at x = 30. The other zero in
the domain is at x = 90. (4.13) can be expressed as a quadratic in
z = x^. This is
-z^ + 9000z - 7 290 000 = 0

Figure 4.22
which produces a curve opening downwards, with the vertex

representing a maximum. The vertex occurs when z = —bj2a, so the
maximum of profit is found at
-9000
z= = 4500
-2
So in terms of the original independent variable, the maximum
occurs when
X = v/4500 « 67
at which output level the profit made is £12.96m.
Exercises 4.3
1 Find the x intercepts of the following cubic functions:
(i) y = — 5x^ — 50x
(ii) y = x^ - 2x^ + X
2 Find the x values which satisfy the following quortic equations:
(i) x^ - 13x2 + 36 = 0
(ii) x^-37x2 + 36 = 0
3 A firm's total variable costs are given by:
VC = x^ - 90x2 ^ 4000x
Find total variable costs when average variable costs are at a

minimum.
4 A company has fixed costs of £1 250000 and its sales
revenue exceeds variable costs by
5000x2 _ ^4
where: x = the output level.
At what output is the company's profit maximized, and what is the

profit made?

4.4 Linear, quadratic, cubic and quartic functions are all examples of
Polynomials polynomials in a single variable. A polynomial expression can also be
described as multinomial in that it contains more than one term, and
is therefore distinguished from a monomial, or single-term expression,
such as 4flx. In general, a polynomial in x can be written as
f{x) = Unx'' + an-ix''~^ + an-2X^~^ H-h a^x^ + uix^ +aix + ao

(4.14)
where all coefficients Ui are real numbers and where n, a non-negative

integer, is the degree of the polynomial so long as the leading
coefficient is non-zero. In the general form of a polynomial both x
and y may be raised to integral powers, but we will focus on
polynomials where only the independent variable is raised to powers
other than one.
A cubic is a polynomial of the third degree, a polynomial of the
fourth degree is a quartic, and a straight line is produced by a
polynomial of the first degree. In (4.14), aq is described as the constant
term. Where a polynomial/(x) = 0 a polynomial equation results. The
notation of (4.14) is very flexible, the subscript form for the coefficients
allowing polynomials of any degree to be conveniently represented. For
example, a polynomial of the fifth degree, a quintic, is written as
fix) = asx^ -f -t- a^x^ + a2X^ -[■ a\x-\- oq
in which the leading coefficient, as must not be zero. The roots of

quintics and polynomials of higher degree can’t generally be found by
use of formulas. The only exceptions are special cases where some of
the terms have zero coefficients. For example, the quintic
2x5-486 = 0
solves for x = 3, which is the only real root of the equation. The sixth-
degree equation:
x^ - 9x5 + 8 = 0
is a quadratic in x^ which solves for x^ = 1 and x^ = 8, so the real

roots are x = 1 and x = 2.
Note that in (4.14) the exponents of x must be non-negative
integers. Therefore, the function
y = 4x5 _|_ _ 2jj3 ^y.2 _
is not a polynomial due to the negative exponent in the final term.

Neither is
y = 7x^ + 3x5 5 + 2x -f 6
due to the exponent 1.5. Nor is the following case a polynomial:
y = 3x’^ - 2x^^ -f x'^^ 7

But the coefficients of the terms in x are not limited to whole or
rational numbers. For example,
y = TTx;^ — y/2x^ + \/3x + 7
is a valid polynomial. Polynomials with complex coefficients can be

considered, but these are beyond the scope of this book.
Exercises 4.4
1 Which of the following ore polynomials?
(i) y = - 4x^ + 2x + 1
(ii) y = x^ - TTX^ + 2x^ + \/2x
(iii) y = x^ - 5x^ + 3x“^ 4- 2x + 1
(iv) y = 4x - 1
(v) y = x^ + x5
(vi) y = x'00
2 State the degrees of the following polynomials:
(i) f(x) = 4x^ - 6x^ + 2x
(ii) fjxj = 1000-2x3-0.001 x^
(iii) y = 100 - X
(iv) y = 7
4.5 A rational function is the ratio of two polynomials. Thus f(x) is a

Rational functions rational function of x if it can be expressed in the following way:
P(x)
fix) =
Q(x)
a„x'' + a„-ix''~^ +an-2x'‘ ^-h aix^ + aix + ap
(4.15)
bmX"’ + br„-lX'”-^ + br„-2X'^-^ ■■■ + b2X^ + biX + bg
where P(x) and Q(x) are polynomials and where f(x) is defined for
Q(x) 7^ 0. So, for example, the following are rational functions
— 8X3 _|_ yj
(i) f(x) =
6x4 _ 7
x^ — 1
(ii) fix) =
X — 1
(iii) fix) = ^
X
(iv) fix) = ax^ + bx + c
Example (i) illustrates a rational function in which P(x) is a quintic

and Q(x) a quartic (there is no requirement for P to be of greater
degree than Q). In Example (ii) note that the numerator expression
x^ - 1 factors as
(x + l)(x - 1)

so the rational expression apparently simplifies to x + 1, but there is a
subtle difference, in that for the rational function
X = +1, for which the denominator is zero is, by definition, excluded

from the domain. Thus the graph of (ii) appears as the straight fine
y = x + 1, with one point excluded, and is illustrated in Figure 4.23.
Figure 4.23
0 X
Example (iii) is the reciprocal function and is graphed in Figure

4.24. This is the simplest case of a rational function that is not a
polynomial. Note that the curve separates into two branches, for x > 0
and X < 0. The origin, x = 0, is not in the domain. The reciprocal
function is an example of a rectangular hyperbola. For this function
the X and y axes are asymptotes - straight fines which the function
approaches but never quite reaches. Hyperbolas always plot in two
sections with asymptotes, and in the case where the asymptotes are
orthogonal the hyperbola is said to be rectangular. Note that the
reciprocal function can, by cross-multiplication, be written as
xy=l
or, apparently rather obscurely, as
(x -|- 0){y -|- 0) — 1=0
A rectangular hyperbola with asymptotes located at x = -a and y = -b

can be written as
ix + a){y + b)-c = 0 (4.16)

So, for example, the rectangular hyperbola defined by the equation
(x + 2)(y + 3) — 1 = 0
which expands as
xy + 2y + 3x + 5 = 0
has asymptotes given hy x = —2 and y = — 3. This function is graphed

in Figure 4.25.
A sixth step can now be added to the curve-sketching procedure:
identification of any asymptotes of the function. The general equation of
the hyperbola is given by the general quadratic equation (4.8) for
values of A, B, and C such that:
> 4AC
The reciprocal function is a special case in which
A = C = D = E = 0, B = +l and F=-l
Example (iv) illustrates the fact that a polynomial is also a rational

function in the case where Q(x) is of degree zero.
Rational functions produce a wide variety of graphs. Indeed, rational
functions going no further in complexity than the ratio of linear and
quadratic expressions can have very varied appearances. An expression
which has both a rational part and an integral part is called a mixed
expression, an example of which is
7 4
3x^ -F Sx -f-

Figure 4.25
In a mixed expression the term which dominates the value of the

function may vary depending on the value of x. For example, in the
function
f(x) = ax + -
X
in which a 0 and b ^ 0, when x is large the dominant term is ax, but

when X is sufficiently close to zero the term b/x will dominate. To
illustrate rational functions in economics consider hyperbolic demand
curves. If the relationship between price and quantity is given by
P=l {4.17)
where /c is a positive constant.

The demand curve produced by (4.17) is a rectangular hyperbola
with the axes as asymptotes. The curve has the special property that
since pq = k the total expenditure of consumers is constant for all
values of q in the domain {q> 0). This means that demand is unit
elastic throughout the entire length of the curve. It may be argued
that the demand curve (4.17), while plausible for moderate values of q,
is unrealistic in that there is no finite limit to either price or quantity.
These reasonable points can be met by use of a demand curve based on
(4.16). This produces a hyperbolic curve which has definite maxima for
both price and quantity. Such an example is displayed in Figure 4.26,
for which a> 0 and b > 0.

Figure C26
The domain of the function is defined for
0 <q<
where: = the market saturation level of demand if the good were

free.
With the equation linking price and quantity in this case being
{q + a){p + b) = c
it will be seen that setting p = 0 produces
n c — ab
“ =—
while the lowest price that will choke off all demand is where
n c — ab
p a
The fact that consumer expenditure varies along this curve can be
deduced from the fact that, while revenue is zero at and it is
strictly positive at points in between. Suppose that the demand curve is
given by
(g + 5)(p + 8) = 240
then the market saturation level of demand will be
240 - 40
4 = = 25
8

while the lowest price that eliminates all demand will be
0 240 - 40
P = 40
5
A business example is the graph of overhead per unit output, average

fixed costs (AFC). This concept is important, especially for firms which
produce capital-intensive products. Similar concepts are also used as
measures of efficiency or performance indicators. The function
_ 1 000 000
AFC =-
<1
might represent average fixed costs for an organization, and the graph
of the branch of the hyperbola for <? > 0 is shown in Figure 4.27.
Figure 4.27
Low levels of output mean that the price of the product must reflect
a disproportionate element of overhead, particularly if variable costs are
not great. A classic example is oil production, and other examples are
found in utilities such as railways. In these instances, increasing output
level will mean that prices can be sharply reduced or high short-term
profits made if the producer is a monopolistic provider.
Exercises 4.5
1 Find the location of the asymptotes for the following
rectangular hyperbola.
xy-y-2x + 2 = 0
2 For the demand curve given by
(q+ 10)(p + 20) =5000

(i) Find the market saturation level of demand.
(ii) Find the lowest price that will choke off all demand.
3 If a firm's fixed costs are £500000, what minimum level of
output is needed to ensure that fixed costs per unit do not
exceed £2.50?
4.6 This section introduces other types of function that can arise in a
Other functions quantitative study of the management and social sciences and to which
we may later need to refer. These are
• composite:
• inverse:
• discontinuous:
• discrete:
• homogeneous:
• monotonic:
• odd and even:
• periodic:
• absolute value functions.
Exponential functions are the subject of Chapter 9.
4.6.1 A composite function is a function of a function, rather than a

Composite functions function of an independent variable. For example, if we begin with the
relationships
y = 2x + 1 (4.18a)
z = 3y - 5 (4.18b)
then z is a composite function of x. Note that z ultimately depends on

an independent variable, and in the present example this relationship is
easily worked out. Substituting for y in (4.18b) from (4.18a) yields
z = 3(2x + 1) — 5
= 6x — 2
As another example, consider z as a function of x where
y = 5x + 4:
z = 3y^
so by substitution
z= 3(5x + 4)^

This two-stage relationship between z and x can be expressed in the
alternative notational scheme as
z=f{g{x))
where the function g{x) is
g{x) = 5x -h 4
and where the function / is ‘squaring and multiplying by 3’. Composite

functions can involve more than one function of a function. Thus, if
w = 5x -t- 4
y = 2w — 6
z = 0.5y + 7
it follows that
y = lOx -h 2
and therefore that
z = 5x -I- 8
To check this result, take a value of x, work the original equations

through and verify the outcome using the composite function. So if
X = 2, then from the original equations w = 14, so y = 22 and z = 18,
which is confirmed by substituting x = 2 into the composite function.
4.6.2
Inverse functions
The inverse of a function is the function which ‘undoes’ the
operations of the original function. For example, if the operation of the
original function is ‘doubling’, the inverse function will be ‘halving’; if
the original function is ‘squaring’, the inverse function is ‘taking the
square root’. Suppose the original function is doubling and adding 5.
The inverse function is then halving and subtracting 2.5. These
operations can be written as
dependent variable = 2(independent variable) + 5 (4.19a)

dependent variable = 0.5(independent variable) — 2.5 (4.19b)
Clearly (4.19a) and (4.19b) are mutual inverses. Inverse functions

can be used in two ways. By far the most common use in business and
economics is to describe a given relationship between two variables
with, as may best serve the purpose to hand, an exchange of roles
between the dependent and the independent variables. Suppose that in
(4.19a) the dependent variable is cost, C, and the independent variable
is output, q, so that the cost function is
C = 2q+5 (4.20)

which is the most convenient form for finding the cost associated with
a specific output level. However, suppose that the relationship relates to
output at a particular plant. Suppose also that the plant manager is
concerned that head office may set a budget implying a particular level
of output. For the purpose of knowing the output consequences of any
given level of cost, the inverse function is more convenient. This is
q = 0.5c-2.5 (4.21)
So a budget of c = 500 indicates output of 247.5.

Inverse functions are frequently used in the analysis of demand and
supply. The demand function states quantity purchased as a function
of price charged (and in general other factors too). Suppose the demand
function is:
q= 1000-0.25p (4.22)
The demand curve is conventionally written with price as the

dependent variable. This will therefore be the inverse function obtained
by rearranging (4.22) with p as the subject of the equation. Thus
p = 4000-4q (4.23)
In terms of notation, if the original function is
y=f{x) (4.24)
the relationship can be expressed with x as the subject by using the

inverse function and writing
*=r‘W (4.25)
where in (4.25) f~^ represents the inverse function. This does not mean
the reciprocal of the original function [l//(x)]. In (4.20) and (4.21), as
in (4.22) and (4.23), the same underlying relationship is being
described. However, if the roles of dependent and independent variables
are exchanged and the function is inverted as a separate step, the two
variables retain their original roles and a different functional
relationship results. For example, starting with
y = 2x + 5 (4-26)
first substitute x for y and y for x giving
X = 2i/ + 5 (4.27)
Now invert
' y = 0.5x-2.5 (4.28)
The function (4.28) involving, as it does, halving the dependent

variable and subtracting 2.5 is the inverse of the function (4.26). In
fact (4.28) will be the mirror image of (4.26) in the 45° line {y = x).
This is illustrated in Figure 4.28, in which the lines intersect where
y = x-= -5.

Figure 4.28
Wx
y = 0.5x-2.5
Also shown in Figure 4.28 is the relationship between y = (for

X > 0) and y = +^yx. In this case it will be recalled that y = ^Jx (with
roots of both signs allowed) is not technically a function, but by
restricting the domain o{ y = x^ to non-negative values of x we ensure
that the inverse relationship meets the function criterion. In terms of
the notation for inverse functions, if equation (4.26) is a specific case of
y =/(x), then in (4.28) we are setting out y = /~^(x) as distinct from
X = f~^{y). A test that two functions are inverses is provided by taking
the composition of the two functions. For inverse functions
/ = X
that is, if the inverse function is applied to the y value produced by a

given X, then the original x value will be returned. If this is not the
case the two functions are not inverses. For the inverse function/~^(x)
to exist, each element in the range of /(x) must correspond to a unique
element in the domain. This is in addition to the requirement for/(x) to
be a function that each x value in the domain gives a unique output
from fix). Such functions are one to one, a property possessed by
always-increasing functions and always-decreasing functions, which are
the only continuous functions for which inverses exist.

4.6.3 A continuous function of a single variable is one for which the graph
Continuous and of the function does not display any vertical or horizontal gaps. It is
discontinuous functions possible to draw the graph of a continuous function without lifting
pencil from paper. This means that the function must have no gaps
when viewed from either axis. Continuous does not necessarily mean
smooth, since a continuous function can have sharp corners or cusps,
with possible awkward consequences. Important results follow if a
function is continuous. Of special significance in the study of business
and economic problems are the following points:
1 If a function f{x) is continuous at every point in a closed interval

[a,b], then the function must have an absolute maximum and an
absolute minimum within the interval.
2 Where a and b are particular values of x, if f{a) and fib) are of
opposite sign, then the function must cross the x axis between a
and b.
The import of point 1 is that the dependent variable must, in the

interval, be contained within certain limits, while point 2 means that
there must be at least one real root between a and b. It can be shown
that all polynomial functions are continuous at all points and rational
functions are continuous at all the points for which they are defined.
Sums, differences, products and quotients of continuous functions are
also continuous. Indeed, a continuous function of a continuous
function is continuous. Unless otherwise specified, we will presume
continuity (at least in the regions of interest) for functions used in later
chapters. This is not a sweeping assumption, as many business and
economic applications use polynomials. Functions which are not
continuous at all points are often perfectly manageable for certain
purposes. This is just as well, since many interesting and realistic
situations in business and economics involve discontinuity. A
discontinuous function is one in which at certain points there is an
abrupt jump in the value of, or the absence of a value for, one variable
as the other variable is continuously changed. Figure 4.29 illustrates a
case of a step discontinuity against the y axis at x = x°.
Many examples of step discontinuities occur in economics and
business. For example, ‘welfare traps’ represent step discontinuities in
the tax-benefit system. Any arrangement in which at a threshold
level the rate of benefit or deduction changes, and where the changed
rate applies to all income (rather than that beyond the threshold level
or between levels), will build in a step discontinuity. Figure 4.29 could
illustrate a severe welfare trap. Let x represent wages and y represent
total income inclusive of benefit. In the case shown, the full benefit
(OC) is paid up to the wage level x° and is then abruptly withdrawn.
A further example of a step discontinuity in economics is the
marginal revenue function in the kinked demand curve model of
oligopoly.

Figure 4.29
A business example of a step discontinuity is price breaks. A price

break occurs when the unit cost of an item changes at a particular
level of sales. In practice the new price usually applies to all items
purchased rather than only to those units beyond the threshold. The
graph of total cost of purchases against quantity bought will then
exhibit step discontinuities at the price-break levels. The diagram that
results has the appearance of Figure 4.30.
Figure 4.30
In Figure 4.30, C represents cost, and two price breaks are shown at
sales levels qo and qi. A step function is one which takes different

constant values for successive intervals in the domain. Such a pattern
arises in the graph of marginal cost of production for a firm with a
resource-constrained linear technology. At first the cheapest possible
production process is used, but as the required level of production is
increased the first of the resource constraints is met. This will require
the introduction of a less economical process which requires less of the
resource at the limit of its availability. The marginal cost jumps at this
point. A further jump occurs when the next resource constraint
becomes binding - calling for the introduction of a still less economical
process. The resulting graph is that of a step function, and is illustrated
in Figure 4.31.
Figure 4.31
4.6.4 A discrete function is an example of a non-continuous function and

Discrete functions is defined for discrete points in the domain. Such a function is
discontinuous everywhere. Discrete functions occur in business and
economics, but are often adequately approximated by continuous
functions. Examples of discrete functions are shown in Figure 4.32.
Figure 4.32(a) might represent a product demand function for an
item which, although sold in individual units, has sufficient volume to
make a continuous approximation acceptable. This is often the case
with consumer goods. In contrast, if the commodity was a capital good,
a continuous approximation might be inappropriate. An extreme
example in this category is a power station. But there are low-volume
consumer goods too. As an example. Figure 4.32(b) might show the
costs of production of a specialized car in terms of the number of units

Figure 4.32
->■
X X
(a) (b) (c)
per week. Even when the product is mass produced and costs only
pence, the practice of packaging items for sale in quanta other than
one unit will present the customer with a discrete unit cost function.
For example, Figure 4.32(c) might represent the average cost to a
consumer of items retailed in packets of two, four or six units.
4.6.5 A homogeneous function is one in which a given proportionate

Homogeneous functions change, p, in the value of the independent variable produces a
proportionate change in the value of the dependent variable of p".
Homogeneous functions are important in the theory of production and
cost. We shall introduce the concept of homogeneity here, but will take
up the topic again when functions of several variables have been
introduced. Consider a straight line through the origin given by the
equation
y = mx
If X is now multiplied by the factor p, the right-hand side of the

equation becomes
m(px) = pmx = py
so y has increased by the factor p^.

A straight line through the origin is said to be homogeneous of
degree one or linearly homogeneous. Note that the quality of
homogeneity applies only to straight lines which pass through the
origin. If the y intercept is not zero, the function will not be
homogeneous to any degree. In the simplest quadratic equation
y = ax^
if X is replaced by px the right-hand side becomes
a{px)^ = ap^x^ = pâx^ = P^y
so y = ax^ is homogeneous of degree two. In contrast, the function
y = x^ — 2x

is not homogeneous. In this case, although there are specific values of
y, p and n that will satisfy
ipxf - 2(px) = p"y (4.29)
for the function to be homogeneous the relationship must hold for

arbitrary positive p and x.
The degree of homogeneity of a function need not be a whole
number. For example, in the operation of taking a square root, a
function results if only the roots of a specific sign are taken. Thus
y = +y/x is a function. Now multiplication of x by the factor p produces
on the right-hand side
+ ^px = ipx)^ = pi = piy
so, with the positive root of p also exclusively taken, the square root
function is homogeneous of degree one half. The degree of homogeneity
need not be a positive number. For instance, y = 1/x is homogeneous of
degree —1. In the case of functions of several variables, degree-zero
homogeneity can be a convenient property. To illustrate, consider the
breakeven formula
and let all of the parameters F, p and b be scaled by a factor t. The

result is
tF _ tF
tp — tb tip — b)
F 0
=-r = q = qt°
p— b
so that breakeven output level is unchanged. So breakeven output level

is homogeneous of degree zero in fixed cost, unit cost and price when
the cost function is linear and the demand curve is horizontal.
4.6.6 A function which is always increasing or always decreasing is

Monotonic functions monotonic. If the function always increases the function is said to be
isotonic. When a function is isotonic, for two values of x, X2 and Xi,
whenever
X2 > Xi
it is always the case that
f{X2)>f{X\)
A distinction is drawn between strictly isotonic functions for which

the strict inequality holds as above, and weakly isotonic functions for
which
f{x2)>f{xi) for all X2 > Xi

Functions for which
fixi) < whenever X2 >
are monotonic decreasing or antitonic. A similar distinction may be

drawn between weakly and strictly antitonic functions. Monotonicity
can be defined over parts of the domain and different monotonicity may
be exhibited in different parts of the domain. So we speak of functions
which are locally or globally or piece-wise monotonic. The possible
monotonicity of a function can be checked by investigating the function
describing the slope. The slope function must not change sign
anywhere in the domain if the original function is to be monotonic.
Monotonicity is an important property in a number of respects - for
example in the context of transformations of variables in particular
scales of measurement or in terms of the existence of the inverse
function.
4.6.7 A function is an odd function if it changes sign but not its absolute
Odd and even functions value when the sign of the independent variable is reversed. So, for any
odd function
f{-x) = -fix)
An example of a function with this property is the cubic
y = 7x^ + 6x
As with all odd functions, this is symmetrical about the origin.

Polynomials which have only odd powers of x are odd functions. A
function which is unchanged in value when x is reversed in sign is said
to be an even function. Thus for even functions
/(- =fix)
and the graph of an even function is symmetrical about the y axis. An
example of an even function is the bi-quadratic
y = —— 3x^ — 10
This property is possessed by any polynomial in which all powers are

even and where all powers of x have coefficients of the same sign.
4.6.8 A function is said to be a periodic function if, starting from any

Periodic functions point, its value is unchanged when x is increased by a specific
increment or any whole multiple of that increment. For a periodic
function, therefore,
fix) =fix + na)
where n is an integer and a is the (constant) period of the function. The

trigonometric functions y = sin x and y = cos x are examples of
periodic functions - indeed all continuous and differentiable periodic

functions can be expressed as algebraic combinations of sines and
cosines. In economics, businesses such as travel agencies or agricultural
producers will have strongly periodic elements within their cash flows.
In management science a good example of periodicity is provided by the
track of inventory level against time in the classical static model of
stock control, as shown in Figure 4.33.
Figure 4.33
The variable 1 represents the level of inventory, t is time and the

period of the function, a, is the length of the inventory cycle. Note that
the cycle may be taken from peak to peak, trough to trough, or from
any given level of stock to that same level - which will re-occur a units
of time later. Economic time series data also frequently contain a
periodic or cyclical component.
4.6.9 An absolute value function states the magnitude of an expression

Absolute value regardless of sign. The form of such functions is obtained by taking the
functions modulus of f{x), and the function will appear as
y = \f{x)\
Absolute value functions are plotted only in quadrants one and two,
and will change abruptly when the function would otherwise have
extended below the x axis into quadrants three and four. Figure 4.34(a)
shows the graph of y = \ x\, while Figure 4.34(b) represents
y = \ x^ — c\.
» Absolute values are involved in calculations of distance and are an
important concept in computer science. In statistics, absolute values
are used in one measure of the deviation of a variable from its average
value - the mean absolute deviation (MAD). In management science,
such measures of dispersion themselves find use in adaptive
forecasting methods - as applied, for example, in some models in stock
control.

Figure 4.34
Exercises 4.6
1 Express z as a function of x where
(i) y = 5x - 6 z = 3y + 4
(ii) y=100 —4x z= 100 + 4/
2 Express z os a function of x where
(i) w = 3x + 4 y = 2w + 5 z = 4/— 8
(ii) w=10 —2x y = 3w —50 z= 100 — 4/
3 Express z os a function of x where
(i) / = 2x + 5 z = y^-4/+10
(ii) / = x-2 z =/^+6/^ + 12/ +8
4 Find the inverse function stating x as a function of / where
(i) / = 4x+12
(ii) / = 20x — 50
(iii) / = 0.025x - 0.5
(iv) y = +Vx
(v) / = 200-10/x
5 Establish whether or not the following functions are
homogeneous and, if so, state the degree of homogeneity in
each case:
(i) / = 6x
(ii) / = 6x + 5
(iii) / = 3x^
(iv) / = x(3x + 4)
(v) / = X2
(vi) / = 1 /x2
6 Are the following odd functions, even functions or neither?
(i) f(x) = 4x^ + 2x^ + 1 Ox
(ii) f(x) = x"* + 3x^ + 10
(iii) f(x) = x*^ + x^ + 10

C H A P TER
Derivatives and optimization
In this chapter the concept of the derivative - an expression from

which the slope of the function can be found and which helps to
identify any turning points - is introduced. Rules for finding derivatives
for functions of a single variable are studied. Derivatives are then
applied to optimization problems in economics and business - where
the best value of functions is sought - and points of inflection are
considered.
LEARNING OUTCOMES
By the end of the chapter you should be able to find derivatives for
many polynomial and rational functions, understand their uses in
management and economic applications, and find local and global
optima in a variety of cases.
5.1 In normative models in business, finance and economics the decision-

introduction maker’s aim is to find values of the decision variables that optimize a
function representing their objectives. Many possible goals may be
relevant. For example, a manufacturer may wish to know:
• the level of production that minimizes the firm’s unit costs;

• how to combine processes to make a given output at minimum cost;
• how best to apportion a scarce factor between products;
• the volume of product sales at which company turnover peaks;
• how to maximize company profits or return on capital.
Addressing these questions through analysis is not just a matter of

finding solutions but also involves finding properties characterizing
optima or equilibria. The analytical results make clear important
concepts such as marginal revenue, marginal cost and elasticity of
demand, and bring out the relationships between them. Although
decision-makers may seek merely satisfactory performance - such as a
target rate of return or the sales growth or profit norm for the industry
- there are few convincing reasons for not going for an optimum once
all relevant factors have been taken into account. This usually
corresponds to a maximum or minimum of a function, although there
Derivatives and optimization 175

are many different cases - for example the presence or absence of
constraints.
5.2 Consider the function of Figure 5.1. The expression, if any, that would
Slope and turning produce a function of this appearance would be of daunting complexity,
points but it is only the graph that is of interest here. For values of x which are
large in absolute terms the function has the x axis as an asymptote and
approaches this from above. The function has seven turning points, at
values of x from xi to X/. It is clear that Xi, X3, X5, and Xy produce
values of the function which are the largest in their own neighbourhood
- if not necessarily overall. The values of the function at these points are
said to be local maxima. The values of x = X2, X4, and xe produce the
lowest value of the function in their immediate vicinity. The values of the
function at these points are local minima. Turning points can be either
maxima or minima and are called stationary points or extreme
points. The term ‘stationary point’ highlights the property common to
aU turning points - the fact that at the precise point the function is
stationary - it has zero slope. If a function does not have zero slope at
a point it must increase for a change in x in one direction (either an
increase or a decrease) and fall as x changes in the other direction. So
the function cannot have either a maximum or a minimum at the point
itself.
Figure 5.1
Zero slope is a necessary condition for a local maximum or

minimum. For a maximum, zero slope is essential but is not enough on
its own - it is not a sufficient condition. All the values /(xi)./(x3),
/(xs) and/(x7) are local maxima, but only /(xs) is the global maximum.
This is the highest value taken by the function at any point in the
domain. Similarly, /(xy) is the global minimum. For a point to be the
global maximum of a function with an unrestricted domain, the value of
/(x) at this point must be greater than at the other local maxima and be
greater than/(x) as x increases or decreases without limit. That is, as
X —> +00 and as x ^ —00

Where the domain of the function is restricted it is necessary that
the turning point which is a potential global maximum must not be
less than the value of f{x) at the end-points of the domain. The usual
purpose is to find the global maximum or minimum. In Figure 5.1 the
global maximum is unique, but it may be the case that the overall
highest value of the function occurs at more than one x value. Such
non-uniqueness gives the decision-maker an extra choice, and it may
be clear that one of the points is better on grounds not included in the
function itself. A three-part strategy can be used to locate a global
maximum as follows:
1 Find all turning points of the function.

2 Distinguish between turning points giving local maxima and those
producing local minima.
3 Find the global maximum from amongst the local maxima.
The first step requires a means by which to determine all values of x for
which the function has zero slope. This could be done if
• an expression could be obtained which gives the slope of the

function at each value of x;
• if this expression were set equal to zero and solved for x.
The values of the function corresponding to these x values include all

local maxima and minima produced by turning points. The
procedures can be applied to business and economic models - for
example finding revenue or profits maximizing output levels for a
firm, the level of use of plant giving a minimum of unit cost, or many
other applications.
5.3 To move to finding an expression which gives the slope of the function,
An approach to the consider the most convenient non-linear function, the squaring
derivative function:
y = x^
Take any point P on this function. As shown in Figure 5.2, the slope of
the function at point P is approximately equal to the slope of the
straight line connecting P and the nearby P° (a wide separation of P
and P° is shown for illustrative purposes).
Let the value of the independent variable at P° be x + Ax, where
Ax, read as delta x, means a small change in x. The value of the
function at P is x^ and at P*^ it is (x + Ax)^. Thus the change in the
value of the function as x changes to x + Ax will be Ay, where
Ay = (x -F Ax)^ — x^

Figure 5.2
The slope of the line PP'^ (which approximates the slope of the function
at P) is
pOn
slope of PP° =-
^ PQ
change in value of function

change in value of x
_^f{x +Ax) -fix)

Ax
Ax
This ratio will now be used to describe the slope of the line segment
PP°. From what we have already established we can write
Ay (x + Ax)^ — x^
Ax Ax
x^ + 2xAx + (Ax)^ — x^
Ax
2xAx + (Ax)^
Ax
so that
Am
— = 2x + Ax (5.1)
So 2x + Ax is the precise slope of the line PP°. Now as P° gets closer to

P so Ax becomes smaller and (5.1) becomes a better approximation to
the slope of the curve itself at P. The value of (5.1) as Ax becomes
smaller approaches 2x, and no leap of imagination is needed to see that

the slope of the function itself at P is 23c. Note that 2x is a formula
which states the slope of the function y = at any point in the
domain - this expression is the derivative of y = x^. So when x = 1
the derivative gives the slope of y = as 2; when x = — 4 the slope is
—8; and when x = +150 the slope is +300. Figure 5.3 plots the
derivative dy/dx = 2x on a graph set below that of the function y = x^.
The same horizontal scale is used in both cases. Note that the height of
the derivative graph gives the slope of the original function.
Figure 5.3
We need new notation to distinguish the derivative itself from the

slope of the chord PP°. The two most commonly used forms of notation
for the derivative are shown in equation (5.2).
^ = 2x=r{x) (5.2)
dx
Each scheme has its advantages and we will use them both. The
dy/dx notation for the derivative compares most directly with the ratio
Ay/Ax, which gave the slope of the chord. This is Leibniz’s notation.
In a similar vein, the derivative of /(x) can also be indicated by
d/ d/(x)
dx dx

The f'{x) representation of the derivative is called function notation.
The remaining notation that is used to indicate the derivative is
D/(x) or DJ{x)
The above approach to obtaining the derivative of y = x^ could be

used, laboriously, to find derivatives for higher powers of x and establish
that for
y = x^ dy/dx = 3x^
y = x'^ dy/dx = (5.3)
y = x^ dy/dx = 5x^
while moving in the other direction we can obtain
y = x^ dy/dx = 1
n (5.4)
y — x^ dy/dx = 0
so the derivative of a constant is zero.
Exercises 5.3
1 Using the Ay/Ax approach above, find the derivative of
y = 5x^.
5.4 Our results so far invite the conclusion that to obtain the derivative of x
The power function raised to any power, the power is reduced by one and the ‘old’ power
rule becomes the coefficient of x in the derivative. That is, if
1 / = x"
n-l
(5.5)
(5.5) is an important formula on which is based the power function

rule for differentiation - the name given to the process of obtaining
the derivative. Consider examples of (5.5) in the following cases:
When n is a positive integer
y = x^
For example, the slope of the function y = x^ when x = 3isl458, and

the slope when x is —2 is —192.
When n is a negative integer
= -2x-^
dx

We have found here the derivative of l/r^, to which it was not obvious
that the rule applied. To use it, convert the reciprocal to the form where
X is raised to a negative power. In this way we have found that for
dy ^ -2
dx:
When a is a rational number
y=
dv
Here we have found the derivative of = i/x. Again the first step is to
convert the expression to a form to which the rule can be applied. So for
y = yjx
dx 2^Jx
As with reciprocals, to obtain the derivative, terms involving radical

signs should be converted to power function form. As another example,
f{x) =
Remembering that the radical sign is in effect an instruction to divide

the exponent by 2, we can re-express this function as
f{x) = x^/^
In the original notation this is
= I.5y/X
When n is an irrational number
fix) =
fix) = \/2x'^^~^
so the rule applies without variation when n is an irrational number.

, Minor generalization of (5.5) produces the power function rule. A
power function takes the form y = cx'\ where the base, x, is variable
and the exponent, n, is a constant. The rule is
The power function rule Where y = cx"
dx

The rule applies where c and n are any real valued constants. So, for
example, if
y =
—■= 12x^
ax
and
fix) = -27.5%-^
fix) = llOx-5
Consider the special case of the power function rule where n = 0. For
y = cx°
to which application of the power function rule produces a result of

zero. This outcome is itself a named rule:
The constant function rule Where /(x) = c
fix) = 0
This confirms that the slope of a horizontal line is zero!
Exercises 5.4
1 Find the derivative for the following functions:
(i) y = x^
(ii) y = xioo
(iii) y = X-3
(iv) y = 1/x^
(v) y = xl/3
(vi) y = -X-2/3
(vii) f[x)
= X^3
(viii) f(x] = -l/x^5
(ix) f{x) = x2^
(x) y = 5x8
(xi) y = 0.25x^
(xii) y = -0.2/xi
(xiii) f[x) = lOxO
2 Find the derivatives for the following:
ID f|x| = x’
|ii| f|x| = X-’
(iii) f(x) = -x“®
(iv) y = 1 /x'°
(v) y = 1 /1 Ox
(vi) f(x) = x^"^
(vii) f(x] = -x~'^'^
(viii) y =-10/10x10
(ix) f(x) = x^’^

(x) /■(x) = 1
(xi) f[x) = -0.05x^°
(xii) y =0.1/x° '
3 Find the derivatives of the following functions, in which o, b,
m, n, p and tt ore constants:
(i) f{x) = X
(ii) f(x) = x^
(iii) f(x) = x~P
(iv) f(x) =-x“("~')
(v) y =-l/x"
(vi) f(x) = ox^
(vii) f(x)=x^/m
(viii) y = x^"
(ix) fix) = x^^
4 Find the derivatives for the following:
(i) y = 1 /x'
(ii) y =l/xO
(iii) y =(^2-l)x'/2+i
(iv) f(x) = -0.2x0 2
5.5 We now consider multi-term expressions. The sum-difference rule

Differentiating allows the power function rule to be extended to functions where
polynomials several such terms are separated by plus and minus signs. The rule is
The sum-difference rule Where y =f{x) ± g{x)
^=f\x)±g'{x) (5.6)
In (5.6) the separated components are shown as functions. They may

be individual terms or themselves comprise several elements. In either
case the expression as a whole consists of a number of terms each of
which can be differentiated individually. This is the nub of a powerful
rule. No matter how many functions are linked by the plus and minus
signs it always boils down to a number of distinct terms which can be
tackled one at a time. As an example of the rule, consider
y = 8x^ + x^
32x^ + 3x2
dx
Now apply the rule to the difference between terms:
y = 7x^ — 9x2
35x^ - 18x
dx

Both sums and differences may be involved, as in
y = — 2x^ + 3x^ — 7x^ + 4x^ + 25
- 12x5 + 15x^ - 2Lv^ + 8x

dx
Now use the rule to differentiate a non-polynomial:
= x5-4x^-5+ V2.x^^-x“^
^ = 5x^ — 18x5'5 -f yj2x'^^~^ + 2x“5

dx
y is not a polynomial as not all exponents are positive integers. The

sum-difference rule can be applied when x is raised to any real power.
5.5.1 In the marginal analysis of the firm in Chapter 4 we considered a firm

Application to the with quadratic revenue and profit functions and used the properties of
marginal analysis of the parabolas. The same data can be used to show the effectiveness of
firm derivatives. The revenue function was
R=50x-0.1x2 (5.7)
where R is sales revenue and x represents physical sales volume

(= output in this model). We know that (5.7) produces a parabola
which opens down - so the turning point (the vertex) is a maximum
(later on we will distinguish maxima from minima using derivatives).
Applying the power function and sum-difference rules, the derivative is
dR
_=50-0.2x (5.8)
dx
which shows the rate of change of revenue as output (= sales here) is

varied. This is the important economic concept of marginal revenue
(MR). Defined as a derivative, marginal revenue is a rate of change at a
point, but it is often helpful to think in terms of small practical
variation, usually one unit. This finite approximation to MR is
defined as the extra revenue from one more unit of output or, as
convenient, the loss of revenue resulting from the sale of one less unit.
Consider an output of 100 units. Setting x = 100 in (5.8) produces
^ = 50 - 0.2(100) = 30 = MR
dx
So the rate of change of revenue at x = 100 is 30.

Now consider finite approximations. We can use (5.7) to calculate
revenues when x = 100, 101 and 99. Results are shown in Table 5.1.
These approximations to MR are within one-third of a percent of the
rate of change figure. In absolute terms the finite change in revenue
goes down by 0.2 for each unit increase in x. This constant rate of
change (—0.2) is the slope of the marginal revenue line. The value
—0.2 is the slope of the slope of the revenue function.

Table 5,1
X TR TR(x) -TR(IOO)
100 4000 0
101 4029.9 -f29.9
99 3969.9 -30.1
Now repeat the calculations for a level of output where the unit
change is a greater proportion of the initial output level. At x = 2
equation (5.8) gives the marginal revenue as 49.6. Table 5.2 shows the
finite change calculations around the initial x = 2 position. The finite
measures are an even better approximation accurate to one-fifth of a
percent of the rate of change figure. The significance of this high
precision is that where the derivative is easy to calculate it can be used
to assess the consequences of finite changes. Where the derivative is
hard to find, finite measures may approximate well enough to the rate
of change at a point. As we have seen, to give a stationary value of the
function the derivative must be zero. In the present example the one
stationary value is a maximum taken for a value of x satisfying
dR
= 50 — 0.2x = 0
dx
so that:
X = 250
Sales of 250 units per period will produce revenue of 6250 (by
substitution of x = 250 in (5.7)). Figure 5.4 graphs the problem.
Table 5.2
X TR TR(x) - TR(2)
2 99.6 0
3 149.1 -649.5
1 49.9 -49.7
Note that MR starts at 50 and is a straight line (being the derivative

of a quadratic). Under normal circumstances a firm would not choose
to operate at a level of output at which MR was negative. A revenue-
minded firm would set output level so that MR was zero, while, as we
will see, the prq^t-minded firm chooses an output where MR is positive.
The equilibrium condition for maximum revenue is MR = 0. This
has a commonsense appeal, but this is not always so. If a condition
seems counter-intuitive, to understand its validity confirm that an
optimum would not be achieved if the condition did not hold. In the
present case, if MR was positive revenue could be increased by
increasing output, while if MR was negative revenue could be increased
by reducing output. So it is essential that MR be zero for a maximum of
total revenue. Now consider profits again. Given the cost function
C = 4x -f 450, profits (represented by the symbol tt) are
7r = 46x-0.1x^-450 (5.9)

Figure 5.4 TRA
Using the knowledge that the profits parabola opens down, a maximum
of profit occurs when the derivative of (5.9) is zero. That is
dTT
— = 46-0.2% = 0 (5.10)
from which
X = 230
which confirms the result found in Chapter 4 using the properties of

parabolas. The derivative d7r/dx is marginal profit, which is zero at a
maximum. However, the condition for maximum profit is usually stated
in terms of marginal revenue and marginal cost, dc/dx. By definition.
n = R-C
dTT dR dC
(5.11)
d% d% dx
SO
(5.12)
dx dx
(5.12) states that, for a maximum of profit, marginal revenue should

equal marginal cost:
MR = MC

To confirm this in the current example:
MR = 50-0.2(230) =4
MC = ^ = 4
Appreciation of this fundamental economic principle can be

reinforced by supposing that the firm operated where MR ^ MC. In
particular, suppose that the output level was such that
MR = 2 MC = 4
so the firm could increase profit by reducing output, as the rate of loss
of revenue (2) is less than the rate of reduction of cost (4), so the excess
of revenue over cost must rise. An alternative description of marginal
cost as escapable cost comes to mind. Again, if an output level (225
units) was set such that
MR = 5 and MC = 4
the firm could increase profits by stepping up production, as the rise in

costs would be more than offset by the increase in revenue.
Now consider an example with a function to be minimized. Suppose
that the average total cost (unit cost) for the manufacture of a product
is:
-1
y = x^ -90x + 4000 + 25 OOOx (5.13)
where y is average cost and x is output level. The domain of this

rational function is x > 0, and the graph is illustrated in Figure 5.5.
The function asymptotes to the y axis and for high values of x it
approaches unit variable costs as given by
x^ — 90x + 4000
as unit fixed costs
25 000
become ever smaller. The firm wishes to know the level of output, x*,
for which unit cost is lowest. This is the output where its most
competitive price can be charged without making a loss. While this is
not likely to be the most profitable point in the short term, it can be
important in aggressive competition. In theory a minimum unit cost
position such as this should result under conditions of pure competition
in the long run. Knowing the general shape of the curve, the minimum
unit cost position is found by setting the derivative of (5.13) to zero.
That is
dy T
90 - 25 000X-2 = 0 (5.14)
— = 2x
dx

Figure 5.5
Multiplying both sides of (5.14) by 0 produces the cubic
2x^ - 90x^ - 25 000 = 0
This equation has one real solution at x* = 50, the other two roots
being complex. An output of 50 units minimizes unit cost at £2500.
It is a useful exercise to compare this value of x = 50 with the
output level at which unit variable cost reaches a minimum. Given
the shape of the unit fixed cost hyperbola, unit variable cost always
reaches its minimum at a lower output level than unit cost as a
whole. In the present case unit variable cost reaches its minimum
value at X = 45.
Exercises 5.5
1 Use the sum-difference rule to find the derivative for the
following:
(i) y =
(ii) y =
(iii) y = -f x^ — x^
(iv) f(x) = 5x^ + 3x^ - 7x^

(v) f{x) = 9x^ - 7x^ + 5x^ - 8x^ + 14x - 100
2 Find the derivative for the following expressions, in which a,
b, c, d, k, m and n are constants:
(i) f(x) = ex'" + bx"
(ii) f(x) = cx'”'^^ — c/x"~'
(iii) f{x] = ax^ + bx'^ + cx + d
XH+l
x'”+^
H y +k
m+ 1
(v) y = OqX^ + O 1 x'^ + 02X^ + OaX^ + 04X + 05

3 Find the derivative for the following functions, and the points
at which the functions have stationary values.
(i) / = 6x2 - 156x+100
(ii) y = 5x2 + ioOx-20
(iii) y = 2x2 -9x2 -60x+ 100
5.6 It is often necessary to find the derivative of expressions which are ratio
The product and or the product of two or more functions, each of which may be easier
quotient rules to differentiate than the expression as a whole. So if the derivative of
the whole can be set out in terms of the derivatives of its parts a more
convenient method will result. Here we consider two rules for
differentiation: the product rule and the quotient rule.
5.6.1 For a function which is the product of two functions, the product rule
The product rule states:
The product rule Where y =f{x)g{x)
p-=f{x)g'{x) + g{x)f'{x) (5.15)

dx
(5.15) is usually written in more economical form as
fg' + af
Begin with an expression for which the result is already known. The
cubing function y = x^ can be represented as the product of two
functions,
f{x) = x^ and g{x) = x
Expressed in this way, our task is to find the derivative of
y = x^x
Applying the product rule, as expected,
^ = x^l + x2x
ax
= + 2x2 = 3x2
Now consider the derivative of
y = 4x2(x2 + 2x^)
This expression can be seen as the product of
/(x) = 4x2
g{x) = (x^ + 2x^)

Application of the product rule produces:
— = 4iX^('ix^ + lOx^) + + 2x^)8%

dx
= 12x^ + 40x^ + 8x^ + 16x^
= 20x^ + 56x^
which could be checked by multiplying out the bracket in y and

differentiating the result using the sum-difference rule. As may be
confirmed, the slope of the function when x = 3 is 42 444.
The more complicated are /(x) and ^(x), the greater is the value of
the product rule. For example, consider
y = {x^ + 2x^ — 5x)(x^ — x^ + 4x^ — 7x^ + x)
applying the rule
^ = (x^ -I- 2x^ — 5x)(5x^ — 4x^ -(- 12x^ — 14x + 1)

dx
+ (3x^ + 4x — 5)(x® — x^ + 4x^ — 7x^ + x)
which could be simplified to
x(8x® + 7x5 _ ^ 3q^3 _ 3323^2 + _ iq)
The unsimplified form of the derivative may well suffice. If the aim is to
find the numerical value of slope it is usually quicker to use the
unsimplified form. The product rule also applies to expressions involving
the product of more than two functions. For example,
y = f{x)g{x)h{x)
^=fgh'+fg'h+f'gh
Thus for
y = 2x(3x - l)(4x + 2)
dll
~ = 2x(3x - 1)4 + 2x3(4x + 2) + 2(3x - l)(4x + 2)
QX
which simplifies to 72x^-|-8x-4, which can be confirmed by

expanding y and using the sum-difference rule.
5.6.2 As its name suggests, the quotient rule applies to the differentiation of
The quotient rule the ratio of two functions:
fix)
The quotient rule Where y =
aix)
dt/ _ gf -fg'
dx g^ (5.16)

First apply the rule to the cubing function, but where this is
expressed as
Applying the rule:
dy x^5x^ — x^2x
dx
5x^ — 2x^
x"^
_ 3x^
x^
= 3x^
The value of the rule is in its application to more involved expressions

such as
X^ + X
y = 3o,+ x^2
for which the quotient rule produces
di^ (3 + x^)(3x^ + 1) — (x^ + x)2x

dx (3 + x2)^
Using this expression, when x = — 2 the slope of y is

dy/dx = 51/49 1.04. Now consider
x^ — 2x^
4-x7
Here
dy (4 — x^)(5x^ — 6x^) + 7x^(x^ — 2x^)

dx (4 - x^)^
A special case of the quotient rule arises when the numerator is 1.

This is the reciprocal rule, giving the derivative of the reciprocal of a
function. If
1 .u
y= then — =-Y
m dx
For example if /(x) = x^, then by the reciprocal rule
1 j dy —2x —2
y = —r and — = —^ = —r
x^ dx x^ x^
which confirms the result already obtained by other means.

Exercises 5.6
1 Use the product rule to find the derivative for the following:
(i) y = (x^+x)(x^ - x^)
(ii) y = (Sx"^ — 4x^)(l Ox^ + 7x^)
(iii) y = (4x3 ^ 3^2 ^ 8x)(2x + x2 + x^)
(iv) y = (x+ l)(x- 1)
2 Use the quotient rule to find the derivative for the following:
(5x^ + 2x7)
(ii) y (3x4 _ 2x3)
X
(iii) fN (7x3 + 5)
r \ fi \ Tlx^ - 5x + 3
H Hx =-3-
3 (i) Use the quotient rule to find the derivative of
f(x} = (x^ - 1 )x“'
Can your result be confirmed by proceeding in a different
way?
(ii) Find the derivative of f(x) = (x^ - 2x^)(x3 + 2x)(3x - 5)“^
and evaluate the slope of the function f(x) at
(a) x=l
(b) x = 2
5.7 The chain rule applies to expressions which are a function of a

The chain rule function. This is an alternative name for the rule which is also known
as the composite function rule. Consider
y = (x^+x)^ (5.17)
Currently our only means of finding the derivative is by expanding the

bracket. Expansion is practical here but if the exponent of the bracket is
53 instead of 3 it is clear that a more efficient technique is required.
The expression in the bracket is itself a function, which we will call
g(x). That is
x^ + x = g{x)
To obtain y, g{x) must have a further operation applied to it - cubing.

So for (5.17) we write
y=
So (5.17) is a function (cubing) of a function (adding its square to the
original value) and is thus a composite function. The second operation
we will refer to as f{x), and the chain rule can be stated as

The chain rule Where y = f[g{x)]
(5.18)
dx dg dx
Begin by applying the rule to a case for which the outcome is

known. The function
y = x^
could be expressed as
y=
for which, in terms of the chain rule,
gix) = x^ and y =f[gix)] = [g{x)]^
Applying the rule
~ = 2x and ^ = 3[g{x)]^
dx dg
so from (5.18)
dy
= 3[x2]^2x
dx
= 3x^2x
= 6x^
Applying the rule to the original case of y [x^ + x] ^ gives
dy
= 3(x + x)
d^
and
dg
= 2x + 1
dx
So from the rule
^ = 3(x^ + x)^(2x + 1)
dx
Now consider the chain rule as applied to
y = 4(x^ — 7x^ + 8x)^
In this case the derivative is
^ = 20(x^ — 7x^ 4- 8x)‘^(3x^ — 14x + 8)

dx

The chain rule can be used with the sum-difference rule, being
applied at a number of points. As an example consider
y = {x^ + 4x)^ - {x^ +
^ = 3(x^ -b 4x)^(3x^ + 4) — 2{x^ + + 2x)

ax
The chain rule is applied separately to each term in y. If each of several

functions can be differentiated individually, then in principle sums,
differences, products, quotients and powers of the functions are also
differentiable. The chain rule and the product rule can also be used
together. If
y = {x^ — xf{x + x^)^
^ — xf[4:{x + x^f‘{l + 2x)]
+ (x + - x)^(2x - 1)]
where the chain rule is applied within the product rule. It may also be
necessary to use the chain and quotient rules together. As an example,
consider
(2x- 1)^
^ “ 7x2
for which
dy _ 7x2[3(2x - 1)^2] - (2x - l)^14x

dx 49x^
The chain rule also applies to functions where bracketed terms are
raised to non-integral powers. Consider
y = (2x + 1)3
The rule gives
dy
1
(2x + 1)2
Now consider the rule applied to a case where the relationship

between y and x is mediated through another variable, z:
y = 3z^ and z = 2x^
As usual
d^ _ dy dz
dx dz dx

where
= 62
dz
and
dz
— = 43C
dx
so that
624x = 24zx
dz
into which, substitution for 2 gives:
24(2x^)x = 48x^
This result is confirmed by substitution for z in the original function:
y= 3(2x^)^ = 12x^
so that
^ = 48*’
dx
The chain rule can also be applied to find the derivative dy/dx where
there are several ‘intermediate’ variables. For example, if
y=/(z) and z = y(w) and w = h{x)
then by the chain rule
dy dy dz dw
dx dz dw dx
Given the particular functions
y = z = w^+w w = 5x^ + 1
the derivative of y with respect to x is
dy
= 2z(2w + l)(10x)
dx
which after substitution for z in terms of w gives
dy
= 2(w^ + w)(2w + l)(10x)
dx
which after substitution for w in terms of x gives
= 2[(5x2 + if + (5x^ + 1)][2(5x2 + 1) + l](10x)

dx
which simplifies to
20x(5x^ + l)(50x^ + 35x2 ^

so when x = 1, d^/dx = 10 920. As more functions are interposed between
the variables y and x, more work is involved in finding the derivative dy/dx
even with the use of the chain rule and the eventual expression becomes
more complicated, but the principle of the rule still applies.
The chain rule, the power function rule, the sum-difference rule, the
product rule and the quotient rule are a useful toolkit of techniques
that allow derivatives to be obtained for a wide range of functions in
business and economic problems.
Exercises 5.7
1 Use the chain rule to find the derivative for the following:
(i) f(x) = (x3 + 1)^
(ii) f(x) = (x4+x2)^
(iii) f{x] = (4x^ + 3x^ - 7x2 ^ 2xf
(iv) f(x) =5(1 -x^)^
(v) y = (x2 -h x)^ -f (x^ + 2x)^
(vi) y = (6x^ — 5x'^ -K 6x2)*^ - (x® — x^)^
2 Use the chain rule with the product rule and/or the quotient
rule to find the derivatives for the following:
(i) y = (x®-1-x'^l^lx^ - x®)*^
(4x® -I- 6x)

(ii) y =
(X2-1)'
(3x2 _ 2x)®
(iii) y =
8x4
3 Use the chain rule to differentiate the following:
(i) f(x) = (5x® - 7x2)5
(ii) f(x) = (4x® + 8x)®'2^
4 Find the derivatives of the following:
(i) f(x) = (ox2 + fax + c)"
1)[(m+ 1)x + m](^+^)

(ii)
(m+ 1)
5 Use the chain rule to find the derivative dy/dx for the following:
(i) y = 5z® and z = 4x^
(ii) y — z^—z + 2 and z = 2x2 — 2x
6 Given the functions
y = z^ + 2z z = 2w w = x®-x
(i) Find the derivative dy/dx.
(ii) Evaluate the derivative at x = 1.
5.8 Suppose that x and y are related by the following equation:

The inverse
X = y^+ 9y^ + 39y + 100 (5.19)
function rule
How can the derivative dy/dx be found? Recall the definition of a
function. The relationship y =f{x) is a function if each value of x in

the domain yields a unique corresponding value of y. If it is also true
that each value of y results from a unique value of x, then the inverse
function exists. The inverse function rule is
The inverse function rule Where y =f{x)
and X =f~îy)
%^ 1
(5.20)
dx dx/dy
where dx/dy 7^ 0.
Take an example that can be checked by other means:
X = 0.25y — 3 (5.21)
As the inverse function always exists for straight lines (with defined,
non-zero slope) we can find the derivative dx/dy = 0.25 and apply the
rule to produce
dy 1 1
= 4
dx dx/dy 0.25
This result can be confirmed by obtaining the inverse function for

(5.21) and obtaining dy/dx directly. The inverse function will be
y = 4x -h 12
for which
The one-to-one relationship between x and y needed for the inverse

function rule can be expressed in terms of slope. The inverse function
exists if the slope of x = y(y) does not change sign in the domain - in
other words, if the function is monotonic. If the function does not
change slope, then for any value of y(x) there is only one real value of
x(y). An example where this does not apply is y = x^, where because of
the change in slope there are two values of x for any (non-zero) y in
the range. So to use the inverse function rule in the case of
,2
X = y
the domain should be restricted to either y > 0 or y < 0. This done, y

expressed in terms of x will satisfy the requirements for a function.
Therefore the inverse function rule will apply, and we can write
dx dx/dy
1
2y

The inverse function rule comes into its own with more complicated
expressions where the inverse function cannot be explicitly stated. A
good example is provided by (5.19), so given
X = + 9y^ + 39y + 100
for which
^=3y^ + 18y + 39
dy
the inverse function rule produces:
_1_
dx 3y^ + 18z/ + 39
Before using the rule we should be satisfied that it applies. That the
slope of (5.19) is always positive follows from the fact that the roots of
dx/dy are complex - the graph does not cross the axis so the derivative
(the slope of the original function) does not change sign.
Exercises 5.8
1 Use the inverse function rule to find the derivative dy/dx of the
following:
(i) x = 4 + 0.2y
(ii) 0.25x = 0.125/-2.5
(iii) X = y^
(iv) X = 2y^ - 3y^ + 6y - 100

2 Find the derivative dy/dx for the following:
(i) x= (2y3+y)(y4 + y2)
(ii) x = (5y3 + 12y- 10)^
3 Find the derivative dx/dy for y = x^ + 2x^ + 5x and evaluate
the result for x = 1.
5.9 Much work using functions of a single variable involves the form where
Implicit one variable is the subject of the equation, as is y in
differentiation
y=fix) (5.22)
Another way of describing this relationship is to say that y is stated as

an explicit function of x. If (5.22) is rearranged as
y -f{x) = 0
fix, y) = 0
then the relationship is stated as an implicit function in which all

terms are on the same side of the equation. This form is not
asymmetric in the way the variables are treated. Consider:
fix,y) = xy-Sx-6y + 4: = 0 (5.23)

The procedure is to differentiate each term with respect to x and, where
a term includes y, to treat y as an unknown function of x. Take the first
term, xy. To differentiate this with respect to x the product rule is used.
So, bearing in mind the way we intend to treat y, the product rule gives
d(xy) dy dx
dx
dy ,
x^ + y
dx
— 5x differentiates with respect to x as —5 and the term +4 has zero as
its derivative. The derivative with respect to x of the term —by is
dy
-6
dx
So the derivative with respect to x of (5.23) as a whole is
dy , ^ dw
xj- + y 6^ = 0 (5.24)
dx dx
(5.24) is an implicit function involving dy/dx which can be rearranged
to state dy/dx explicitly as
y
(5.25)
dx X
Note that dy/dx is expressed in terms of both x and y. In (5.23) we

chose an example that could be rearranged to state y as an explicit
function of x. We now make the rearrangement and differentiate the
results to confirm (5.25). Equation (5.23) can be written as
y(x — 6) — 5x + 4 = 0
from which
5x — 4
y (5.26)
X — 6
from which dy/dx can be obtained by use of the quotient rule:
dy (x — 6)5 — (5x — 4)
dx (x — 6)
-26 (5.27)
(»: - 6)^
At first glance, (5.27) does not look a great deal like equation (5.25),
but substitution for y from (5.26) into (5.25) results in
^ 5x — 4
dy_ ~ x-6
dx (x — 6)
-26
“ (x - 6)^
which confirms our original result.

Now try implicit differentiation on a function of higher degree:
f{x, y) = x^y^ + 3y — 4:X = 0 (5.28)
Differentiation of the first term with respect to x produces
x^2y^ + y^2x (5-29)

dx
where the chain rule is employed in producing the derivative of y^ with
respect to x as
T dy
2y^
dx
The derivatives of the other terms follow and the overall result is
2x^y ^ + 2xy^ + 3 ^ — 4 = 0
dx dx
which rearranges as
dy 4 — 2xy^
dx 2x^y + 3
Exercises 5.9
1 Find the derivative dy/dx for the following implicit functions:
(i) 4x-5y+10 = 0
(ii) 1 Ox + 5xy -2y - 100 = 0
2 Rearrange function 1 (ii) above to state y as an explicit
function of x and obtain an expression for dy/dx using the
quotient rule. Confirm that the result agrees with 1 (ii) by
evaluating the expressions at
(i) x=10 y = 0
(ii) X = 1 y = 30
3 Use the implicit function rule to find the derivative dy/dx for
the following:
(i) f(x, y) = 5y - 2x + xy^ = 0
(ii) f{x, y] = 2x^y^ - 3x^ + xy - 7y = 0
5.10 The derivative of a function of x is in general a function of x which can

Higher-order be differentiated to obtain the second derivative - the derivative of
derivatives the derivative. The second derivative, also typically a function of x, can
be differentiated to obtain the third derivative. This process can be
continued to find still higher-order derivatives where these exist. Just as
the first derivative is the expression giving the slope of the original
function, so the second derivative gives the slope of the slope. To
illustrate higher-order derivatives, consider the quartic
y = 2x‘^-7x^+4:X^ + 8x + 20 (5.30)

A list of the derivatives is:
First derivative:
o 3
— = 8x 21x^ + 8x + 8
dx
Second derivative:
d^^ 7
^ = 24x^ -42x + 8
dx^
Third derivative:
d^y
^ = 48x-42
dx^
Fourth derivative:
d^y
= 48
dx^
Fifth and all higher-order derivatives:
d^y
= 0 for n> 5
dy’^
In function notation the second derivative of /(x) with respect to x is

represented as
rix)
while the third derivative is shown as
f"ix)
Although higher-order derivatives can be represented in this manner,
the notation soon becomes cumbersome. It is more economical to
write
/(")(*)
for the nth derivative of /(x) with respect to x. For example, consider
/(x) = x^° — 2x® + 4x^ -f x^
for which derivatives down to the fourth are
f'{x) = lOx^ — 16x^ -h 20x^ -f 4x^

/"(x) = 90x^ - 112x^ + 80x^ + 12x^
/"'(x) = 720x^ - 672x5 + 240x^ + 24x
fix) = 5040x^ - 3360x^ + 480x + 24
A polynomial of degree four such as (5.30) has non-zero derivatives

down to the fourth. For an /ith degree polynomial, all derivatives
beyond order n will be zero. For other classes of function - such as

rationals - there may be no limit to the order of derivative that can be
found. For example, with
y = x-^
the second derivative is:
d y
= 2x
dx^
and, for example, the sixth derivative is
d^y
= 720x“^
dx^
Of the higher-order derivatives, the second is the most often used. A

rather surprising implied use of a third derivative was made by US
President Nixon in 1972 when he announced that the rate of increase
of inflation was decreasing. However, applications in business and
economics rarely go beyond the second derivative, so we shall
concentrate on the properties and uses of the second derivative.
Exercises 5.10
1 Find the second derivative for the following functions:
( ) f(x)=x5
( i) f(x)=7x^
( ii) f{x) = 5x^ + 3x2 ^ 2x
( v) f{x] = 8x5 _ 9x4 + 7^3 _ 8x2
2 Find the third and fourth derivatives for the following:
( ) X = x‘
( i) y = 2x5 _ 3^4
( ii) y = 0.01x10-0.005x20
3 Find the second derivative for the following:
( ) f|x| = x-'
( i) f(x) = 1 /x2
( ii) f[x) = xO 5
( v) f{x) = (x2 -1- sf
4 Find the second derivatives for the following:
(i) y = ox"* -f bx^ -f cx^ + dx + e
( i) y = —mx + nx "
5.11 An important use of the second derivative is in distinguishing local

Local maxima and maxima from local minima (for which the slope changes in a different
minima way). Through a maximum the slope of the function diminishes, so the
slope of the slope - the second derivative - is negative. Through a local
minimum, the slope of the function increases with x - the second
derivative is positive (strictly speaking, at a maximum the slope of a
function will not be increasing and at a minimum it will not be

decreasing). The second derivative test is used to identify the nature
of stationary points in functions of a single variable:
1 If for X giving /'(x) = 0 it is also the case that /"(x) < 0, then the x
value produces a local maximum of /(x).
2 If for X giving /'(x) = 0 it is also true that f''{x) > 0, then the x
value produces a local minimum of/(x).
3 If for X giving f'(x) = 0 it is also true that /"(x) = 0, further
investigation is needed to determine the nature of the stationary
point.
We will consider case 3 later - the point could be a maximum, a

minimum, or the function may inflect. In the vast majority of cases 1 or
2 applies and the second derivative test is decisive. Consider the quadratic
fix) = 50x — O.lx^
for which
/'(x) = 50-0.2X
which is zero when x = 250. To establish whether x = 250 produces a

maximum or a minimum, consider the second derivative:
f"(x} = -0.2
Since the second derivative is negative regardless of x, any turning

point is a maximum. Figure 5.6 shows the graph of the function and
the first and second derivatives (a constant function at the level —0.2).
Now suppose a maximum is sought for
y = 500-x^- 20x
Setting the first derivative to zero gives
-2x- 20 = 0
dx
which solves for x = —10. The second derivative is
meaning that x = —10 produces a maximum, at which point y = 600.

This example illustrates the fact that there is no guarantee that the
maximum producing value of x will be positive.
Now find the maximum of
y = 40x — x^ — 500
Setting the first derivative to zero gives
— = 40-2x = 0
dx

Figure 5.6
meaning that x = 20 produces a maximum. At this point y = —100.

This example illustrates the fact that there is no guarantee that the
maximum value of the function will be positive.
Now consider the cubic
y = x^ + 3x^- 24:X + 20 (5.31)
for which
du
— =3^ +6^- 24 (5.32)
and
d^y
y4= 6x + 6 (5.33)
dx^

Setting (5.32) to zero to identify the stationary values of the
function, we have
3x^ + 6r — 24 = 0
Dividing throughout by 3
+ 2x — S = 0
so
(:x: + 4)(x — 2) = 0
so the two turning points of the function are at x = —4 and x = +2.

As with all cubics the second derivative is linear. Inserting the values of
X = —4 and x = 2, at x = —4
= -18
which by the second derivative test means a maximum. But when

X = 2
d y = +18
dx2
so X = 2 produces a minimum. The function is graphed in Figure 5.7.
Figure 5.7
■ Note that (1) the local maximum here does not produce the highest
value of the function - which is unlimited if x becomes arbitrarily large
and positive, and (2) that the sign of x at a turning point is usually
important in practice. These points will be considered later.
Now consider
fix) = x^ - 15x2 + 87x- 100

for which
f'ix) = 3x2 _ 30^ + 87
Setting /'(x) to zero produces only the complex roots
X = 5 + 2/ and x = 5 — 2i
The absence of a solution in real numbers means that this cubic has no
turning points. It appears as graphed in Figure 5.8.
Figure 5.8
Exercises 5.11
1 Find the stationary point of the functions below and use the
second derivative test to establish if the point is a maximum or
a minimum;
(i) y=100x-2.5x2
(ii) y = 4x2 - 72x + 100
2 Find the stationary points of the following functions and
establish their character:
) f(x) = 0.5x^ - 2.25x2 - 60x
') f(x)= 1000-2x3+6x2+48x
ii) f(x) = - ] 2x^ + 28x2
, ,, , x-^ 5x^
v) f[x] = ^-— + 4x + 2
3 (i) The stationary points of y = x^ - 8x^ + 22x2 - 24x + 100

occur at X = 1, x = 2 and x = 3. Use the second derivative
test to establish the nature of each stationary point.
(ii) The function y = 6x^ - 67.5x'‘ - 320x^ + 2700x2
+ 12000x + 500 has stationary points at x = —4,
X = —2, X = 5 and x = 10. Use the second derivative test
to establish the nature of each stationary point.

4 Do the following functions hove turning points? Provide os
much information os possible in each cose.
(i) f(x) = 3x3-18x2 _ 1053x + 1000
(ii) f(x) = x3 - 3x2 + 15x + 100
(iii) f(xj = x^ - 12x3 + 80x2 _ 400x + 20
5.12 The second derivative identifies points as maxima or minima in their

Global maxima immediate vicinities but says nothing about which of the maxima gives
and minima the greatest value overall. So, unless there is only one local maximum,
further work is required to identify the global or absolute maximum
(for which there may be other possibilities). This may take the form of
enumeration - calculating the value of the function at each local
maximum - quite an efficient method with straightforward expressions.
In business or economic modelling meaningless outcomes need to be
excluded - for example negative output levels or those beyond the
maximum capacity of plant. The decision-maker expects the model to
provide sensible values of the decision variables - and ideally those
values which produce an optimum. Irrelevant values can be excluded
in a variety of ways; for example:
1 Using a function for which extrema occur only for realistic values
of the decision variables.
2 Placing specific limits on the domain of the function.
3 Discarding extrema associated with unrealistic values of the
decision variables.
A risk involved with approach 3 is the assumption that among the

associated x values there will be some within a reasonable range. If this
is not so, then either approach 2 must be used or the function must be
redefined. In approach 1 quadratics and quartics with negative leading
coefficients (for maximization) and positive (for minimization) produce
finite maxima and minima, respectively. With appropriately selected
coefficients, the extrema can also be confined to the positive quadrant.
For example, a quadratic profits function will follow from a linear
demand curve and a linear or quadratic cost function. Suppose that the
monthly profit function for a small firm assembling a personal
computer is
7r= -1000 + 200x-x2
where the —1000 term represents fixed costs and where the first-order
condition for a maximum
dTT
— = 200-2x = 0
dx

is satisfied within the positive quadrant for x = 100. The second
derivative test
confirms that the stationary value is a maximum. As the one candidate

local maximum, this will also be the global maximum. We presume
that the level of profit achieved, tt = £9000 (representing a margin of
£90 per unit output) is acceptable and that the output level is within
the firm’s capacity. So in terms of a reasonable value, it happens that
things have worked out well here. Now consider a quartic case. The
objective function to be maximized is
fix) = -0.75x^ + 7x^ - 21x^ + 24x
for which the first derivative is
fix) = + 21x^ - 42x + 24
and for which the second derivative is
f'ix) = -9x^ + 42x - 42
When the first derivative is set to zero, the solutions obtained are
X= 1 X = 2 and x = 4
Evaluating the second derivative at these x values shows that x = 1 and

X = 4 produce local maxima, while x = 2 gives a local minimum. The
global maximum is found by evaluating /(x) at x = 1 and x = 4. The
superior point is x = 4, which produces fix) = 16, in comparison to
fix) = 9.25 at X = 1. This function is graphed in Figure 5.9.
Figure 5.9
The function itself can be used to rank local maxima, which involved
the evaluation of two possibilities in this case. But a function may
produce values greater than that for any stationary value. An obvious
case is where the function increases without limit as x moves without

restriction in one or other direction. This is so for all cubics, for
example with
y = 2x^ - 45x^ + 300x
for which
- 90X + 300
dx
which, when zero, solves for x = 5 and x = 10. The second derivative is
12X-90
which is negative at x = 5 and positive at x = 10, so there is a local

maximum at x = 5, producing y = 625, while the local minimum at
X = 10 produces y = 500. The function appears as in Figure 5.10.
Figure 5.10
But the value of x = 5 does not give the global maximum of the
function, and x = 10 does not correspond to the overall minimum of y
The function is unbounded in either direction. Other functions may take
unlimited values at particular points in the domain. This is so for rational
functions where the denominator is zero for certain x values. Consider
1
(5.34)
(x- 10)
for which
’ -1
dx (x - 10)^
This derivative is not zero for finite x. The function has no stationary
value and is graphed in Figure 5.11. It can be seen that y becomes
arbitrarily large and positive as x approaches 10 from above, and
arbitrarily large and negative as x approaches 10 from below.

Figure 5.11
To take such cases into account, an exhaustive process for

identifying the global extremes of a function would need to include a
search for vertical asymptotes. There are also more esoteric examples of
functions for which the value of the expression oscillates between ever
larger extremes - both positive and negative - as some crucial value of
X is approached. Also, piece-wise defined functions can be given
exceptional properties at individual points. But these cases are not our
main interest. Rather, we will consider the principles of approach 2,
where explicit account is taken of the function's domain. Suppose a
maximum value is sought for
1/ = 500 + 22.- 150x -
for which
^ = 45x- 150 - 3x^ = 0

dx
which solves for x = 5 and x = 10, and
— 6x
which is positive for x = 5 and negative for x = 10. So x = 5 produces

a local minimum, while x = 10 gives a local maximum. The function is
graphed in Figure 5.12.

Figure 5.12
Now suppose the function represents a desirable factor such as net

income, y, against the decision variable, x. For example, x may be the
output of a possible new product to be added to the existing range, so
negative values of x would have no meaning. What is the optimal
output for the new product.? In the context of the model, the optimal
decision is the value of x within the domain x > 0 giving the highest
value for y. Inspection of Figure 5.12 reveals this to be x = 0 (where
y = 500) rather than at the local maximum, where x = 10 (ly = 250).
This is an example where the global optimum occurs at an end-point
of the domain - which means that the new product should not be
made.
Figure 5.13
A maximum at an end-point is a possibility that must be investigated

when there is a restricted domain. The function should be evaluated at
end-points as well as at turning points in the domain. Consider the
quartic illustrated in Figure 5.13. There is a fixed lower end-point, L,
for the domain but there are three places where the upper end-point, U,
might be set: Ui, Uz and U3. The function itself increases without limit
for values of x below L or above U3. The decision-maker seeks the
global maximum in a domain where L is the lower end-point and
where the upper end-point is at one of the U values. It is clear from
Figure 5.13 that with the upper end-point at the various U values the
overall maximum is as follows:

• Ui - the overall maximum is at the lower end-point x = L
• U2 - the overall maximum is at the turning point where x =
• 1/3 - the overall maximum is at the upper end-point x = U3
In addition to turning points and end-points, if the function is not a

polynomial, any values of x for which the function is not defined must
be taken into account. As an example, consider finding the global
maximum of
1/ = 3x^ - 124x^ + 1380x^ - 2400x + 2000
with domain restricted to
(i) X = 0 to X = 2
(ii) X = 0 to X = 4
(iii) X = 0 to X = 12
(iv) X = 0 to X = 30
(v) X = — 10 to X =
First, we need the turning points to locate the local maxima. Thus
^ = 12x^ - 372x^ + 2760x - 2400 = 0

dx
This cubic solves for x = 1, x = 10 and x = 20. The second-order

condition is
d^y 9
^ = 36x^ - 744x + 2760 < 0
dx^
The second derivative is positive at x = 1 and x = 20 and negative at

X = 10, which is therefore the only local maximum. Now consider the
position of the global maximum in the specified domains.
(i) Since there is no local maximum in this interval, the overall

maximum must be at one of the end-points. Checking the value of
the function at these points,
at X = 0 y = 2000
at X = 2 1/ = 1776
so the global maximum in this domain is 2000 at x = 0.

(ii) Again there is no local maximum in the domain. So the value of y
must be at a maximum at one of the end-points. At x = 4.
y = 7312, which is the global high with this domain.
(iii) In this case the local maximum is in the domain so there are three
points at which the function should be evaluated:
x = 0 X = 10 and x=12
At X = 10, 1/ = 22 000, while at x = 12, 1/ = 19 856, so this time

the overall maximum occurs at the turning point.

(iv) Here the upper end-point of case (iii) is raised, and the value of the
function at x = 30 must now be compared with the figure 22 000
at X = 10. At X = 30, y = 254 000, so this is definitely the global
maximum.
(v) In this case the overall maximum occurs when x = —10, where
y = 318 000. Note how rapidly the function increases for negative
X values since all terms will now be positive.
We have seen that a restricted domain can arise because of the

nature of the independent variable. One instance is where the decision
variable (x) is probability where the domain is: 0 < x < 1. The domain
may also be narrowed if there are other considerations - such as
minimum and maximum levels of operation of plant and equipment.
Exercises 5.12
1 With the specified turning points, identify the global maximum
for the following:
(i) f(x) = 1000- 3x^ + 56x3 _ 335^2 ^
(turning points at x = 2, x = 4, x = 8)
(ii) f{x] = 250 + 12x - 10.5x2 + 7^3 _ 1.5^4
(turning points at x = 0.5, x = 1, x = 2)
2 Given turning points at x = 3, x = 6 and x = 12, find the global
minimum of f(x) = x'^ - 28x3 _|_ 252x2 _ 864x + 5000.
3 Which of the following functions have finite global maxima?
(i) f(x) = 100 000 + 2500x - 0.0005x2
(ii) fix) = 0.01x2-5000x-10000000
(iii) fjx) =0.01x3 - 1 000000x2
(iv) f(x) = 75 + 20x + 30x2 _ q. 1 x3
(v) f(x)=x~’-x'°
(vi) f(x) = 4000x - x^
4 Find the global maximum of f(x) in each of the following cases:
(i) f(x) = 1500 - x3 + 39x2 - 360x
(with domain restricted to 5 < x < 25)
(ii) f(x) = 1500 - x3 1 33x2 - 360x
--

(iii) f(x) = 1500 - x3 - 1 8x2 _
(with domain restricted to: 0 < x < 3)
(iv) f{x) = x3 - 45x2 _(_ (500x
5.13 In the second derivative test we saw how the value taken by the second
Concavity, derivative allowed local maxima to be distinguished from local minima.
convexity and The test can be regarded as distinguishing those stationary values at
points of inflection which the function’s curvature (we use the term informally) makes it
5.13.1 locally ‘hollow from below’, or concave, from those points where the
Concavity and convexity function is locally ‘hollow from above’, or convex. Local concavity

means that a stationary value will be a local maximum, while local
convexity implies that the turning point produces a local minimum.
Figure 5.14 illustrates this distinction for a cubic.
Figure 5.14
A function is concave over an interval of x if for any two points in

the interval the function lies on or above the straight line connecting
the corresponding points on the graph. If the function is always above
the straight line, it is strictly concave over the interval. The function is
convex over the interval if the function lies on or below a straight line
connecting the points on the graph corresponding to any two points in
the interval. The function is strictly convex over the interval if it is
always below the straight line.
Figure 5.14 gives examples of concavity or convexity over an
interval. The function is concave between points a and c because no
matter how closely the end-points of the interval are approached, the
function lies above the straight line connecting those points. However,
care is required. For instance, the function of Figure 5.14 cannot be
said to be concave over the entire interval a-e even though the
function lies above the straight line connecting a and e. The reason is
that points can be selected near to e where this is not the case.
The second derivative being negative (f"{x) < 0) is sufficient for strict
concavity. This corresponds to the slope of the function diminishing and
the graph curving down. The second derivative being positive
(/"(x) > 0) is sufficient for strict convexity. This corresponds to the slope
of the function increasing and the graph bending up. Note that in a
concave or convex section the slope of the curve itself could be either
negative or positive.
Some functions, such as quadratics, have the same curvature
throughout, while others, such as cubics, change curvature as x varies.
A function for which /"(x) < 0 is globally concave (the curve always
bends downwards - where it bends at all) is a concave function. A
function for which /"(x) > 0 is globally convex (the curve always bends
upwards - where it bends at all) is a convex function. A linear
function satisfies both of these conditions (since/"(x) = 0) and so is

both a concave function and a convex function. But a function such as
that of Figure 5.14 is neither a concave function nor a convex function.
In maximization problems it is very useful to know if a function is a
concave function. If this is so, a local maximum, if any, will also be a
global maximum, and all that is required for a global maximum at
% = is the first order condition
f'{x) = 0 at X =
This is both necessary and sufficient for a global maximum of a

concave function. In the above, we are saying ‘a’ rather than ‘the’
because of the possibility of linear sections in a function which is
concave but not strictly concave. If the function is strictly concave
throughout, then x° produces the unique global maximum.
In minimization problems it is of similar value to know that a
function is convex over the whole domain. In such a case a necessary
and sufficient condition for a global minimum of the function at
X = is
/'(x) = 0 at X = x*^
Here again, if the function is strictly convex throughout, then x°

produces the unique global minimum. Three further points on concave
and convex functions are worth noting:
1 If/(x) is a concave function g{x) = —/(x) is a convex function.

2 The sum of convex (concave) functions is a convex (concave)
function.
3 Points on or below a concave function form a convex set. Points on
or above a convex function form a convex set.
Point 2 means that in a function with several terms, if the terms each
produce a concave function the overall function is also concave.
5.13.2 Where a function changes from strict concavity to strict convexity (or
Points of inflection vice versa) over an interval, there will be a point in the interval where
the second derivative is zero. Such a point is an example of a point of
inflection. In Figure 5.14 point d is an inflection point. A distinction is
drawn between inflection points such as d where the function is
changing in value and those points - stationary points of inflection -
where the function is static, and where the first derivative is also zero.
Point d is therefore a non-stationary point of inflection. The usual
way in which a stationary point of inflection is discovered is if in the
search for a maximum or minimum of a function a value of x is found
which produces/'(x) = 0 and for which/"(x) is also zero. However, not
all points where /"(x) = 0 are points of inflection.
Consider some examples. First find the inflection point of
/(x) =-0.5x^ + 15x^+ 20x+ 100 (5.35)

Here we have the advantage of knowing that all cubics have a point of
inflection. We shall now locate this point. The first derivative is
fix) = -l.Sx^ + 30X + 20
Setting the second derivative to zero produces:
f'ix) = -3x+ 30 = 0
which solves for x = 10. This is a possible inflection point. It will be an

inflection point if the second derivative also changes sign as it passes
through X = 10. This possibility can be checked by finding the value of
f"(x) on either side of x = 10 and close to this value. For example,
taking a one-unit variation
x = 9 gives /"(x) = 3 while x=ll gives /"(x) = —3
So in this case the function does change sign through x = 10, which
therefore produces a point of inflection. Finally, substitution of x = 10
into f'(x} produces a non-zero result (170), so that x = 10 gives a non¬
stationary point of inflection.
To understand the requirement that f'ix) change sign through the x
value for which f'ix) is zero, recall that on one side of the point of
inflection the function is concave if'ix) < 0), while on the other side it
is convex if'ix) > 0). If the third derivative is non-zero at the candidate
point of inflection, this is sufficient to ensure that the second derivative
changes sign. Such is the case for (5.35), for which
fix) = -3
The sign of/"'(x) does not matter - the important fact is that it is non¬
zero. However, the third derivative could also be zero, and yet the point
could still be a point of inflection. Where f'ix) is easily calculated, this
is a quicker check than working out fix) either side of the point in
question.
Now consider another example. Let
/(x) = x^ - 60x^ + 1200X+ 10 000 (5.36)
for which
fix) = 3x^ - 120x3-1200
and
fix) = 6x - 120
From the equation of the second derivative to zero, the value of

X = 20 emerges. The fact that the third derivative
fix) = 6/0
confirms that x = 20 gives a point of inflection. In this case at x = 20

the value of fix) is also zero, so a stationary point of inflection is

produced. It is possible for a function to inflect at a point even if the
third derivative is zero. This is confirmed by the quintic
fix) = x^ (5.37)
which inflects at x = 0 (a stationary point of inflection), at which point

all derivatives down to the fifth are zero. These discussions will lead to
a comprehensive rule for sorting out the character of the turning and
inflecting points of functions, but first consider whether the quartic
fix) = 0.25x^ - + 150x^ - lOOOx (5.38)
has a point of inflection. The first derivative is
fix) = x^ - 30x^ + 300x - 1000
and the second derivative is
fix) = 3x^ - 60x + 300
Setting fix) = 0 gives the equation
3x^ — 60x + 300 = 0
that is
x^ — 20x + 100 = 0
which factors as
(x- 10)^ = 0
so there is a double root, and the second derivative is zero only at

X = 10. Checking the third derivative at this point,
f"\x) = 6x — 60
= 0 at X = 10
which does not resolve the issue. Trying fix) for one-unit departures
from X = 10, setting x = 9 gives
fix) = 3(81) - 60(9) + 300 = +3
and, when x = 11
fix) = 3(121) - 60(11) + 300 = +3
so the second derivative does not change sign as x passes through 10.
It follows that X = 10 does not produce a point of inflection. Note in
this case that
fix) = 0 at X = 10
so what kind of a stationary value is x = 10.^ In fact it produces a

minimum of (5.38). This can be deduced from the following general rule
for establishing the nature of stationary values;

If all derivatives as far as fîx) are zero at x = but /"(x) 7^ 0
then if n is even, P{x) < 0 means that the point is a local
maximum, while /"(x) > 0 implies a local minimum. If n is an odd
number, then x = x^ produces a point of inflection.
Using this rule in the case of (5.38) the first non-zero derivative is
= 6. Thus, since the fourth (even) derivative is positive, the value
X = 10 produces a minimum. Using the rule on the quintic (5.37), the
first non-zero derivative at x = 0 is P’ = 120. Since it is the fifth (odd)
derivative that is the first to be non-zero, we conclude that there must
be a point of inflection at x = 0.
Another way of finding the nature of a point where several higher
derivatives are zero is to find out if /'(x) changes sign as x increases
through the point in question. If it does, the point will be a maximum if
the change in sign is from positive to negative, and a minimum for a
negative to positive change. If /'(x) does not change in sign a stationary
point of inflection is identified.
In the microeconomic theory of production and cost, a cubic cost
function is often used for its plausible economic properties. For example,
suppose that daily costs of a car manufacturer are given by
C(x) = O.lx^ - 30x2 + 3100x + 10000 (5.39)
where C(x) is the total cost of producing x cars per day. Marginal cost is
the rate of change of total cost, C'(x), and is given by
C'(x) = 0.3x2 _ 60x + 3100
The point of inflection of the total cost function (5.39) is the daily
output level producing minimum marginal cost. This is given by
C"(x) = 0.6x - 60 = 0
which solves for x = 100. Beyond this output marginal costs begin to
rise - even though average costs continue to fall for some time. The
example illustrates the fact that to find a point of inflection is to find
the maximum or minimum of the first derivative.
Exercises 5.13
1 Find the point of inflection for the following:
(i) f(x) =1000- 2.5x3 ^ 60x2 ^ 300x
(ii) f(x) = 4x3 _ 48x2 + 25x 4- 100
(iii) f(x) = 0.25x^ — 9x3 + 108x2 -4 20x
(iv) f(x] — 0.15x^ — 0.75x^ -I- x3 -I- 15x + 300
2 Are the points of inflection for the following functions
stationary or non-stationary?
(i) f(x) = x3 - 1 8x2 + 108x + 100
(ii) f(x) = 3x3 _ 27x2 + 1 OOx - 50
(iii) f(x) = x^ - 14x3 + 60x2 _ 50x + 100

3 Do the following functions hove points of inflection?
(i) f{x) = 0.25x^ - 5x3 ^ 37 5^2 _ i Qx + 50
(ii) f{x) = 0.5x^ - 3x3 _ 30x2 + 5x + 20
(iii) f(x) =3x^-1 OOx^ + 1000x3

CHART E R
6 Functions of more than one

variable
This chapter considers the concepts of functions and differentiation

where there are two or more independent variables, and areas of
application in management and social science are indicated. Partial
derivatives are introduced and used to identify stationary points of
functions of several variables.
LEARNING OUTCOMES
By the end of the chapter you will be familiar with the concept of
partial derivatives and how they’re used to identify stationary points of
functions of several variables. You should also be able to differentiate
polynomials in two or more variables and distinguish between types of
stationary value.
6.1 Economic problems can often usefully be discussed in terms of the

Introduction relationship between two key variables. The classical Marshallian
demand curve is an example of such a relationship. But there are
limitations to the conclusions that can be drawn from two-variable
models. Most quantitative problems in business and economics involve
more than two variables. Consider demand again. The quantity
demanded, x, of a product per period time might be expressed as
x=f(p, Pc, Ps, Y, i,...) (6.1)
where (6.1) is the demand function. Many factors influence demand

for a product. Those identified in (6.1) are:
• p = the product’s own price

• Ps = price of a substitute
• i = a measure of interest rates
• Pc = price of a complement
• Y = disposable income
In (6.1), X is shown as a function of five variables. The demand-

curve relationship between quantity demanded and a product’s own
price is a special case of (6.1) - one curve being defined for each set of
values of Pc, Ps, Y, i. As another example of a function of several

variables, the production function states output per period time, the
dependent variable, as a function of the amounts used of factors of
production, the independent variables or arguments of the function.
At a high level of aggregation we can write
q = f(L. K, M) (6.2)
where: q = product output;
L = labour input;
K = capital applied:
M = materials used.
The form of the function, the way that quantities of input translate
into output, represents the technology of the manufacturer. In this and
later chapters we will consider a number of forms that the production
function might take. For the most part we will concentrate on the
relationship between one dependent variable and two independent
variables. This will illustrate most of the important and relevant
properties of functions of several variables. We will use the notation
z=f{x, y) (6.3)
where x and y are sometimes referred to as input variables, as well as

the more familiar terms independent or decision variables. In this vein,
the dependent variable z is an output variable. The domain is the set
of X and y combinations to which the function is restricted or for which
it is defined. The range of the function is the corresponding set of
output values of z. For a function of a single variable the domain will
be the real line or those sections of it for which the function operation
can be performed, or to which it is otherwise restricted. Domains for
functions of two variables, if not the whole of the x, y plane, are most
commonly areas on the plane such as triangles, rectangles, other
polygons, disks or irregular regions. But the domain might also be a
line, curve (or parts thereof) or discrete points. Commonly, the domain
is the first quadrant where both independent variables are non¬
negative. This may be further restricted with one or more functions of x
and y (often linear in practice). The function (6.3) produces a surface,
rather than a solid, in three-dimensional space.
Consider a particular instance of equation (6.3):
z = 5x^ + 3xy - y^ - 2x+ lOy + 100
This function is a quadratic in x and y, all terms being of degree two,

one or zero. The constant means that the origin is not a point on the
surface, although (0, 0, 100) clearly is. Insertion of any other pair of
values of x and y into the function gives the corresponding value of z.
For example, when x = 2 and y = 3, then:
z = 20 + 18 - 9 - 4 + 30 + 100
= 155
Functions of more than one variable 221

Manual curve sketching in three dimensions calls for skills that few
possess and it is rarely worth attempting more than an outline by
hand. Computer software is available that will produce wire-frame
diagrams of functions of two variables. Such software can be of great
assistance in understanding the properties of a function, and several
diagrams in this book are generated using Derive. Sophisticated
packages such as Mathematica, MathCad or Maple are available in
addition to intermediate range products such as Derive.
6.2 By no means all work with functions of two variables requires

Linear functions of computer assistance. The simplest form of (6.3) is a linear function of
several variables two variables, which is useful in much practical work. Consider
z = 40 —X —0.5i/ (6.4)
Linear functions in two independent variables produce plane (flat)

surfaces in three dimensions. That part of the plane corresponding to
(6.4) in the positive quadrant is shown in Figure 6.1.
Figure 6.1 zk
Three axes are required. By convention the dependent variable, z, is

measured against the vertical axis, while the independent variables, x and
y, are at right angles (and to z) in the horizontal plane. The axes can be
thought of as the bottom-left corner of a cube. The x, y and z axes are the
three fundamental directions, which it may help to think of as North
iy) East (x) and Up (z). Points on the plane are identified by co-ordinate
triples of x, y and z satisfying (6.4). For example, the points:
P = (20, 30, 5) and Q = (10, - 30, 45)
lie on the plane (though Q would be outside the region shown in Figure
6.1). The point
R = (20, 20, 35)

is above the plane, while
S = (-20, 40, 35)
lies below it. Contours of the surface (lines on the x, y plane above
which the surface is at a constant height) may be obtained by slicing
through the surface for given z in a plane parallel to the horizontal.
The resulting edge is then projected down to the x, y plane to form the
contour. This is illustrated for z = 15 in Figure 6.2. The figure shows
the part of the contour lying in the positive x, y quadrant.
Figure 6.2
In Figure 6.2 ABC represents the original plane, which is intersected at

height H = 15 by a plane parallel to the x, y axis. With attention confined
to the non-negative orthant, the line DE represents the edge where the two
planes given by (6.4) and z = 15 meet. The contour itself is FG. Note
that the contour for z = 0 is the line BC (which is contour and edge
simultaneously, as the plane for the given value of z is the x, y plane itself).
Note that the contour for z = 40 is the origin. This is because the only
point in the non-negative orthant having a point in common with the
plane at z = 40 is A, which is directly above the origin, 0. The equation of
any contour above is found by solving the simultaneous equations for the
planes that intersect in each case. Thus where z = 15 the equations are
z = 40 — X — 0.5y
z= 15
So:
0 = 25 — % — O.Sy

The equation of this contour could be expressed explicitly as
y = 50 — 2x
and by rearrangement the equation of the 2 = 0 contour is
y = 80 — 2x
and the equation of the z = 40 contour is
y = —2x
Consistent simultaneous equations in three variables correspond to

intersecting planes. Where two equations are given, the intersection is a
straight line and to get specific values for the variables a third equation
needs to be added. In geometric terms, the line representing the
intersection of the first two equations would penetrate the third plane
at a single point. The co-ordinate triple of this point is the solution of
the simultaneous equations. For example, suppose that the third
equation added to the system comprising (6.4) and 2 = 15 is
z=55-x-2y (6.5)
and therefore since 2=15
15 = 55 — X — 2y
so that
X + 2y = 4:0
and this equation and the equation of the contour already obtained
y = 50 — 2x
can be solved simultaneously as
2x + y = 50
X + 2y = 40
which by elimination or substitution produce the solution
X = 20 and y = 10
Thus the co-ordinate triple of the one point that lies on all three planes
is (20, 10, 15). The straight-line distance between two points in three
space can be found by use of Pythagoras’ theorem.
The concept of a simplex arises in linear programming where the
main method of solution is the simplex method, and in decision theory
where the idea of a simplex is useful in the theory of games. A simplex
may have any number of dimensions. In geometric terms a two-
dimensional simplex is the smallest region without indentations (a convex
figure) that contains three points which do not lie on the same straight
line (are not co-linear). So a two-dimensional simplex is a triangle. A
one-dimensional simplex is the smallest convex region containing two

points which are not superimposed - a straight-line segment. A zero¬
dimensional simplex is a single point. A three-dimensional simplex is
the smallest non-reentrant solid that will contain four points which do
not lie in the same plane (are not co-planar). This is a tetrahedron. A
simplex is regular if its vertices (the points) are evenly spaced.
Simplices of up to three dimensions are illustrated in Figure 6.3. An n
dimensional simplex is the smallest convex region containing n -|- 1
points in independent locations. Independent locations means that the
points are not all contained in n — 1 dimensions. In the case of three
points not all must lie on the same straight line (one dimension).
Figure 63
(a) A Iwo-dimensional simplex (b) A one-dimensional simplex
Exercises 6.2
1 Given the plane defined by z = 200 - 5x - 2y, do the points
below lie on, above or below the plane?
(i) (20,50,0) (ii) (-10,25,210)
(iii) (-20, -10, 330) (iv) (40, 125, -240)
(v) (-80, 500, -440) (vi) (60, -60, 20)
2 Find the equations of the contours of z = 3x -I- 0.5y — 20 for
the following:
(i) z=100 (ii) z =-20 (iii) z = 50
3 Find the point in common between the three planes defined by
the following:
(i) z = 9 - 4x - y
x + y-|-z — 6 = 0
3y -F 4z - X = 17
(ii) 5x — 2y -f z = 140
3x -I- y — 0.5z = 40
0.25x -F y -F z = 30

6.3 A quadratic function in two independent variables is one in which the
Quadratic functions degree of every term (add the powers to which each variable in the
term is raised) is 2, 1 or 0, there being at least one term of the second
degree. As an example consider
2 = lOOx + 60y — 2xy — 2x^ — — 300 (6-6)
A function of this form could well arise where a firm makes and sells
two products in amounts x and y, has a quadratic cost function and
faces linear demand equations for the two products. The value of z
would represent profit. Later in the chapter we shall show how (6.6)
might be arrived at from basic demand and cost information, and we
shall also find the best overall profit position. For the time being,
consider the general features of the function as shown in Figure 6.4.
Figure 6.4
The function has a single turning point which is a local and the
global maximum. This peak occurs in the positive quadrant -
important if x and y are physical measures such as output. An
elevation of the function is shown in Figure 6.5. Figure 6.5 shows a
vertical section, profile or ‘elevation’ of the surface when one variable is
set at a specific value. The vertical section corresponding to i/ = 5 is
shown. By substitution into (6.6) the equation of the elevation curve is
^{y=S) ~
If X is set at a specific value, a vertical section of the surface results, for

which 2 is a function of y alone. For example, with x = 10
^(*=10) = 500 + 401/-i/^

Figure 6,5
All elevations of this function are quadratics with the same general
appearance, but with different intersection points with the axes and
corresponding to different values of z at their highest points. In
microeconomics when the function is a production function (showing
output level as a function of input levels) a vertical section of the function
shows output graphed against variations in the level of one input only.
This is a total product curve or productivity curve. Horizontal
sections through the surface z projected to the x, y plane generate
contours of the surface as in the linear case. Here, the contours resemble
those on an Ordnance Survey map - indeed ‘contour’ is interpreted in
just the same way as on a map. The contours of the function for
z = —300 and z = +500 are shown in Figure 6.6. All contours of (6.6)
have the same general shape. The equation of any one is given by:
lOOx + 60y - 2xy - 2x^ - y^ - 300 = constant
So for the z = 800 contour the equation is
lOOx + 60y — 2xy — 2x^ — y^ — 1100 = 0
In microeconomics, contours of functions often represent important

economic concepts. For example, contours for a production function
show input combinations that produce the same level of output. These
are isoquands. If the objective is profit the contours are iso-profit
)
lines (‘iso’ means equal).
The number of independent variables in a quadratic function can
exceed two. As we have seen, there are many independent influences
on the demand for a product, and there are numerous variable inputs
in a typical production process. The function
z = 3xy + 2y^ — wx + + 5yw + x + y + w — 20

Figure 6.6
is of degree two in the three independent variables x, y and w. Note

that if we take only the terms of degree two in the function above
z = 3xy + 2y^ — wjc + + 5yw
the result is a function homogeneous of degree two. The homogeneity

means that if all of the independent variables are scaled by a factor p.
then z will be scaled by a factor of p^. Homogeneous functions play a
role in the theory of production.
Exercises 6.3
1 Given the production function
z = 120x + 50y — 2xy — 4x^ — 0.5y^
find the total product curves for the following:
(i) y = 10 (ii) x = 25
2 Find the equation of contours of
z = 1 Oy — 3xy — 100 + 6x — 9x^ for the following:
(i) z = 900 (ii) z = 0
3 Establish whether the following functions are homogeneous:
(i) z = y^ + 2xw — 3wy + w^
(ii) z = 3x^ + 2y^ + xyw - w^
2
y
(iii) z = wx - xy H--
w
(iv) z = wx + ^ + (x - 1

6.4 ^ Consider a linear function of two independent variables:
Slopes and first-
z = 0.6x + 0.2y + 20
order derivatives
The function is graphed in Figure 6.7 and an arbitrary point P is
indicated. What is the slope of the surface at the point P.? The first
observation marks an important distinction between functions of
several variables and functions of a single independent variable: the
slope of a function of several variables at any point depends on the direction
of movement. In the linear case, wherever a point is located on the
surface a one-unit increase in x alone increases z by 0.6, while a one-
unit increase in y alone increases z by 0.2. These values, 0.6 and 0.2,
are the slopes of the surface in the fundamental x and y directions.
Movement in other directions changes z by an amount which can be
easily determined in the linear case, but the pivotal role in terms of
optimization is played by the slopes in the fundamental directions. So
we will concentrate on these slopes which are given by the partial
derivatives of the function. A partial derivative is an expression for the
rate of change of a function when one variable only changes. The partial
derivative of a function z with respect to x is written as
Figure 6.7
where dzjdx is read as ‘partial dee z dee x’ and is obtained by

differentiating the function in the ordinary way, except that ah non-x
terms and parts of terms are treated as constants. So, for the linear function
z = 0.6x -F 0.2y

when partially differentiating with respect to x, the term 0.2y is treated
as a constant and so disappears in the process. The result is
|^ = 0.6=/,
OX
Similarly,
Consider the non-linear function
z = 2xy
When differentiating partially with respect to x, the ‘non-x’ part of the

term {2y) is treated as a constant. In this context, 2y is simply regarded
as the coefficient of x. Thus with respect to x:
dz
= 2y =fx
dx
Similarly, with respect to y
These expressions give the slope of the function z in the two

fundamental directions. Thus at the point (3, 5, 30)
/, = 2(5) = 10 /, = 2(3) = 6
so the function slopes more steeply in the x direction at this point.

Consider
fix, y) = xy^
for which the partial derivative with respect to x is
fx =
while the partial derivative with respect to y is
fy — x2y = 2xy
In the case of the partial derivative with respect to x, the x part of the
term is linear, with the coefficient y^, which therefore ‘comes down’
into the derivative as would any constant. In differentiating partially
with respect to y, the non-y component, x, comes down into the
derivative unaltered to multiply the derivative of y^ with respect to y.
Now consider
3 2
z = x^y^

When differentiating this function partially with respect to x, the non-x
component, y^, is treated as a constant and multiplies the derivative of
with respect to x (3x^). Thus
dz 2..2
3x y
dx
The derivative with respect to y has the x^ component treated as a

constant, so
^2
— = x 3-7 0 3
2y = 2x y
dy
Now consider the partial derivatives of a function with more than

one term:
z = 2xy + xy^ — x^y^
The sum-difference rule applies to functions of several variables. Dealing

with each term in turn produces
dz 2 2..2
3x y
and the partial derivative with respect to y is:
dz
= 2x + 2xy — 2x y
dy
Consider again the profits function of section 6.3. This was
2 = lOOx -I- 60y — 2xy — 2x^ — y^ — 300
and the partial derivatives will be
dz
= 100 — 2y — 4:X (6.7a)
dx
and
dz
= 60 — 2x — 2y {6.7b)
dy
Partial derivatives are important in equilibrium conditions - statements

that need to be true for an optimum. Furthermore, the derivatives usually
have economic significance, and an understanding of this import gives
insight into economic problems and a feel for the meaning of the symbolic
manipulation. But there is also an immediate practical use of the
derivatives. The slope of the function in the fundamental directions is the
marginal profit from expansion of x or ^ production alone. For example,
if the firm is initially operating at output levels of x = 12 and y — 8 the
rates of change of profit at the margin can be found by substitution into
(6.7a) and (6.7b), respectively. The results are
dz dz
= 36 and = 20
dx dy

These values for marginal profit are instantaneous rates of change at
the point given by the original output levels. Since the partial
derivatives are themselves functions of x and y, any change in x or i/
brings about a change in the value of the derivative. The values of the
partial derivatives are useful in obtaining an approximate value of the
actual change in z for finite increments in x or y. Consider an example.
Suppose that the output of x is increased by one unit, to 13. What will
be the actual change in profit.^ This can be found by substituting x = 13
and y = 8 into z and subtracting the value of z at x = 12, y = 8. At
the new output levels the value of profit is
z = 100(13) + 60(8) - 2(13)(8) - 2(13)^ - 8^ - 300 = 870
while when x = 12, overall profits are
z = 100(12) + 60(8) - 2(12)(8) - 2(12)^ - 8^ - 300 = 836
So the actual change in profit level following a one-unit increase in x is
870 - 836 = 34
Consequently, the value of the partial derivative, 36, would in most

cases be a good enough approximation to the actual change in profit.
As a practical illustration, suppose that a way could be found to
increase output of x by one unit but at a cost (over and above the costs
implicit in z) of £100. The proposal should be rejected - the figure
given by the derivative is sufficient to do this.
It is possible to be more specific in many cases about what the values
of dzjdx and dzjdy represent in terms of finite increments in x or y. In
the present context dzjdx is an upper bound on the change in z
following a unit increase in x. For an increase of n units in x an upper
bound would be given by n{dzldx\ The upper-bound property arises
because z is concave in the x direction. This can be inferred in the
present case without explicit reference to the second derivative, because
the linearity of (6.7a) and the negative coefficient for x make it clear
that the derivative is a diminishing function of x - as x goes up by any
amount dzjdx, the instantaneous rate of change, always goes down. So
we can conclude that the derivative will overstate, to some extent, the
change in z for any finite increase in x; the smaller the increase in x,
the less will be the extent of the overstatement. The same upper-bound
property is true of the derivative with respect to y. If y is incremented
by one unit to 9 while x remains at 12, the actual change in z is 19,
compared to the value of 20 for dzjdy at x = 12, y = 8.
Now consider another example in partial differentiation. Let
z = 3x‘^y^ - 2x^ + 4y^ = /(x, y) (6.8)
In finding the partial derivatives in this case, we shall highlight the

constancy under partial differentiation of those parts of terms that are
not expressions in the variable with respect to which the differentiation

is being carried out. For partial differentiation with respect to x,
rearrange the terms as follows:
z = - [2]x^ + [4/]
Those terms/parts of terms in square brackets are either constants or

are treated as constants when differentiating with respect to x. So in
dzjdx, [3i/^] will multiply the result of differentiation of The term
[4i/®] disappears under differentiation with respect to x. Thus
~ = 4:x^[3y^] - [2]3x2
= 12x^y^ — 6x2
Approaching the derivative with respect to y in the same way, z can be

written as
z = [3x^]y^ - [2x2] + [4]/
and the partial derivative with respect to y will be
-=[3x^]5y^ + [4]6y^
= 15xV + 24/y2
Consider an example function with more than two independent

variables. Let
/(x, y, w) = Ix^y^w^ — 8x^y^ + 9w^y^ + 6xw^
The three first-order partial derivatives are found in just the same way
as for functions with two independent variables. The derivatives are
fx = 14xi/2w^ — 24:X^y^ + 6w^

fy = 21x2^2^4 _ 27w^y^
fw = 28x^y^w^ + 45wî/2 I2xw
The product, quotient and chain rules can also be used for finding
the partial derivatives of functions of several variables. For example,
given
fix, y) = i2 + 3x + 4i/)2
using the chain rule, the derivative with respect to x is
^ /» = 2(2 + 3x + 4i/)(3)
= 12 + 18x + 24y
and the derivative with respect to y is
fy = 2(2 + 3x + 4^)(4)
= 16 + 24x + 32y

These results can be checked by expanding the original function:
fix, y) = 4 + 12x + 161/ + 9x^ + 24xy + 16y^
So
fx = 12 -T 18x -T 24i/
fy = 16 + 24x + 32i/
The product rule could also be used here, since the function can be
written as
fix, y) = (2 + 3x + 4y)(2 + 3x + 4y)
so that by the product rule
/x = (2 + 3x + 4y)(3) + (2 + 3x + 4y)(3)
= 12-|-18x-)“ 24y
/, = (2 4- 3x + 4y)(4) + (2 + 3x + 4y)(4)
= 16 + 24x + 32y
Now consider another example using the chain rule. If
fix, y) = (3x^ - 7y^ + 5xy^ - 9x^yf
then
fx = 4(3x^ - 7y^ + 5xy^ - 9x^y)^(9x^ + 5y^ - 18xy)

fy = 4(3x^ - 7y^ + 5xy^ - 9x^y)^( - 28y^ + lOxy - 9x^)
A further example of the use of the product rule is given by
fix, y) = (4x^y^ + 6x^y^)(3x^y^ - 7xy)
for which
fx = (3x^y^ - 7xy)(8xy^ + 18x^y'^) + (4x^y^ + 6x^y'^)(9x^y^ - 7y)

fy = (3x^y^ - 7xy)(8x^y + 24x^y^) + (4x^y^ + 6x^y^)(6x^y - 7x)
Now consider the quotient rule in partial differentiation. Given
9xy + 2x^y^
fix, y) =
x3 + y4
then
(x^ + y^)(9y + 4xy^) — (9xy + 2x^y^)3x^

fx
ix^+y^f
_ (^^ + y^)i9x + 6x^y^) - (9xy + 2x^y^)4y^
^ (x3+y4)^
The rules can be applied to functions with more than two

independent variables. As an example, consider the use of the chain
rule to find the partial derivative where
fix, y, w) = (x^y^w^ + x^w^)^

making use of the rale
fw = 7{x^y^w^ + x^w'^)^{2x^y^w + 4x^w^)
Exercises 6.4
1 What is the slope of the plane z = 1000 — 2x + 0.5y in the
fundamental directions?
2 (i) Find the first-order partial derivatives of z = — 4xy + /•
(ii) Find the slope of the function in the fundamental directions at:
(a) (10, 2, 22) (b)(-2,7,67)
3 For f(x, y) = x^y^, find the partial derivatives and the slope of
the function in the fundamental directions at the point (5, 4,
1600).
4 Find the first-order partial derivatives of the following functions:
(i) F(x, y) = x^+ xy^+ x^y^
(ii) f(x, y) = 100x99yioi
(iii) f(x, y] = 4x^ + 2y^ - 3x^y^ + 4x^y^
(iv)
(v) q = l°k^ (for a and b as constants)
(vi) z = nx^y^~^ (for m and n as constants)
(vii) z = ^
(viii) z = x^y + yx“^

5 Find the first-order partial derivatives for the following:
(i) f(x, y, w) = 1 Ox^y'^w^ — 2x^y^ — 5x^w^ + 4y^ w^
(ii) Q = (Where A, r, s and t are constants)
6 Given
z = 500x + 400y + 4xy - 3x^ - 2y^ - 1000
where z represents profits and x and y are output levels, find
expressions for the marginal profits on the two products.
7 Find the partial derivatives for the following:
(i) f(x, y) = {6x^y^ + 2xy^)(3x^y - 8x^y^)
(ii) F(x, y) = (1 + Axy)(l + 6xy)
(ii) f(x, y) = (2x9y2 - x^)(y^ - 5xy)“'

(i) ^(-^z y) = (6x'^y^ - dx'^y^)*^
(ii) f{x, y] = (ox^ + bxy + cy^)^^
10 Find the partial derivatives for the function
f(x, y, w) = (x^ - x^y^w9)(y^ - w^x^)

6.5 Partial derivatives of a function will in general be functions of the
Higher-order independent variables and can usually be differentiated. For example,
partial derivatives consider again the function (6.8); this was
2 = 3x^1/^ - 2x^ +
We had found that
^ — 12x^ — 6x^ (6-9)

ox
If (6.9) is itself differentiated partially with respect to x the result is
^^2 = 36x^1/^ 12x=fxx (6.10)
Note the two forms of notation. This derivative is found in the same
way as for the first derivative and is the second direct partial
derivative of z with respect to x (‘direct’ will be clarified shortly).
Consider the second direct derivative with respect to y. At the first order
we had
f)7
— =15xV + 24y5 (6.11)
dy
Differentiation of (6.11) partially with respect to y produces
= 60xV + 120i/^ =fyy (6.12)

oy
It will have occurred to the reader that it should be possible to

partially differentiate a function first with respect to one variable and
then with respect to another. For example, there is no reason why the
first derivative with respect to x should not then be differentiated with
respect to y. The result is written as
The process by which 60x^y^ was arrived at should be familiar, and the
function notation for this derivative, fxy, is that which common sense
suggests. (6.13) is a second-order cross-partial derivative or mixed
derivative. The somewhat counter-intuitive notation
d^z
dydx
may appear to have the denominator the wrong way round. But when
viewed as the derivative with respect to y of the derivative with respect
to X or:
d dz
dy dx

the ordering of dy and dx in the denominator looks more reasonable.
Now find the cross-partial derivative of (6.8) first with respect to y and
then with respect to x. At the first order we had found
fy = 15x‘^y'^ + 24i/^
and partial differentiation with respect to x produces
d^z
fyx = eOx^y'^ = (6.14)
dxdy
Notation aside, (6.14) is identical to (6.13). This is no coincidence.

For a very wide range of functions of two independent variables second
order cross-partial derivatives are equal. Exceptions have curious
continuity properties and will not concern us here. All this means that
there is no need to fuss over the order in which dx and dy should
appear in the denominator. This is more than just a technical point and
often leads to interesting insights in particular applications. Also, in
obtaining the mixed derivatives it may be easier to use one sequence of
derivations rather than another. For example, given
(f/" + 4)
— 4:xy^
- 7)
to obtain:
d^z
dxdy
it is best to differentiate first with respect to x. This eliminates the

awkward quotient at the first stage, differentiating which is wasted
effort (it contains only y). Proceeding in the most efficient manner.
and so
^=-12 2^-^
dydx ^ dxdy
As further practice in partial differentiation, consider
2 = 4x^ — 5xy^ + 4x^1/ — 2y^
We will find the first order partial derivatives, the slopes of the function
in the fundamental directions at the point (2,3, —64) and the second-
order direct and cross-partial derivatives. At the first order
dz 7 9
— = 12x — Sy -h 8xu
dx
and
dx
— = —lOxy + 4x^ — 6y^
dy

The slope in the x direction at (2, 3, -64) found by substitution in
dzjdx is
=12(2)2-5(3)^+ 8(2)(3) = 51
ox
and the slope in the fundamental y direction at this point will be
— = -10(2)(3) + 4(2)2 _ ^(3)2 ^ _9g

oy
The direct second-order partial derivatives are
52 z
^-24x + 8y
-lOx- I2y
dy^
Second-order cross-partial derivatives work out as
dh dh
— lOy -h 8x =
dydx dxdy
Higher-order partial derivatives could be taken. For example, the third-

order cross-partial derivative
dh
d^xdy
involves partially differentiating z twice with respect to x and once with

respect to y (the result here is 8). As with second-order mixed
derivatives it does not matter in what sequence the differentiation is
carried out. Similarly, the fourth-order mixed derivative
cl^z
d^xd^y
requires partial differentiation twice with respect to both x and y. Here

this derivative is zero. Partial derivatives of higher order than the
second are rarely required, and our focus is on first- and second-order
derivatives.
Exercises 6.5
1 Find the second-order direct derivative, d^zfdx^, for the
following:
(i) z =
(ii) z = 7y'2 - 8x^y^ - 2x2y
(iii) z = 5x^y^ — 6x^y^w^ — xyw
2 For the functions in problem 1 above, find the second-order
direct derivative with respect to y.
3 Find second-order cross-partial derivatives for the functions in
problem 1 above.

4 For z = 2x^ + — 4xy^ + 5x^y + — 3y^ + 7xy — 3x
+ 8y + 9
(i) Find the slope in the x direction when x = 1 and y = 2.
(ii) Find the slope in the y direction when x = 1 and y = 2.
(iii) Confirm that d^z/dydx = d^z/dxdy and evaluate at
X = 1, y = 2.
(iv) Evaluate d^z/dx^ at x = 1, y = 2.
(v) Show that the slope in the y direction falls for increasing y
at X = 1, y = 2.
(vi) Find d^z/dx^ and d^zjdy^
5 (i) Given
f{x, y) = 2x^y^ — 3x^y^ + 2x^y^
find all of the second-order partial derivatives.
(ii) For: f{x, y, w) = x^y^w^ find:
(^) fxy (b) fxw (c) fyw
6 For z = x^y'^ — 4x^y^ -f 6x^y^ find:
d^z
(ii)
dxdy^
d^z
(iii)
dx^dy^
d^z
(iv)
dxdy^
7 For f(x, y, w) = 3x"‘y'^w - 4x^ w^ - Sy'^w'^ find:

(i) fvv (ii) fyw (iii) fxyw • (iv) fwww
6.6 At the first order, for a local maximum or a local minimum of

Local maxima and
z =fix, y)
minima
6.6.1 the function must have zero slope in the fundamental directions, i.e.
Conditions for a
dz ^ dz ^
maximum (6.15)
These first-order conditions mean that z cannot be changing in the

fundamental directions at a maximum. Conditions (6.15) do not
distinguish between maxima and minima (or another possibility we will
consider later). Any point satisfying (6.15) is a stationary point and it is
at the second order that its character is identified. The conditions are in
two parts. The first is to be expected from the single-variable case:
dh
< 0 < 0
dx^ dy^

that is, second-order direct partial derivatives must be non-positive for a
maximum. However, we will work with the strict inequalities
d^z
<0 (6.r6a)
dx^
and will not consider cases where the second derivatives are zero. But
there is another condition to be fulfilled involving the cross partial
derivatives. This (expressed as a strict inequality) is
d^z d^z ( d^z \ ^

(6.16b)
dx^ dip- \dxdy) ^
The left-hand side of (6.16b) is the discriminant (we will not

consider cases where the discriminant is zero). Later, we will see that if
(6.16a) is satisfied but (6.16b) is not, a different kind of stationary
value is produced. Second-order conditions as a whole (6.16a) and
(6.16b), if satisfied, mean that the function z is locally concave. If the
second-order conditions hold for all values of x and y, then z =f{x, y) is
a concave function.
Consider a maximization example. For
z = 35 -I- lOx — Sy + 2xy — x^ — 2y^ (6-17)
we will find the stationary point and establish if it is a maximum. First-

order conditions are
= 10 + 2y — 2x = 0
ax
— = —8 -F 2x — 4i/ = 0
dy
The second-order direct derivatives are
d^z dh
-2 and -4
dx^ dy^
while
d^z
dxdy
The direct second-order derivatives being negative meet the

requirement (6.16a). In general, the second-order derivatives are functions
of X and y and must be evaluated for the values of these variables that
satisfy the first-order conditions. But when z is quadratic the second
derivatives will be constants. Condition (6.16b) is also satisfied since
(_2)(-4)-( + 2)2 >0
So the discriminant has the appropriate sign and the stationary point is
a maximum. The location of the point is determined by the first-order

conditions, which can be solved as simultaneous equations. Note that
adding the two equations in this case eliminates x, leaving
2 ~ 2y = 0
so y = 1, and substitution into either equation produces x = 6. Note

that since there is no other solution to the first-order conditions the
function has a global maximum ofz = 61 atx = 6,^=l.
Consider a further example. Show that the function
z = 1152x — 84i/ + 2xy^ -1- x^y — — 8y^
has a maximum atx=10,i^ = 2. The first-order conditions are
dz
= 1152 2y^ + 2xy — 12x^ 0
dx
dz
^ = —84 -|- 4xw -\-x^ — 24i/^ = 0
dy
Insertion of x = 10 and y = 2 shows that the equations are satisfied

and so the point is a stationary value. At the second order the
requirements are that
d^z
= 2y — 24x < 0
dx^
d^z
= 4x — 48^ < 0
dy^
Note also that
dh
4:y + 2x
dxdy
and that condition (6.16b) must also be satisfied. In this example, with
the second derivatives being functions of x and y, the conditions must
be evaluated at the point in question. Thus
d^z
-236 < 0
dx^
d^z
-56 < 0
dy^
so (6.16a) is satisfied. Furthermore, since
dh
= +28
dxdy
condition (6.16b) will produce
(- 236)(- 56)-(28)2 > 0
and, as all conditions are met, the point (10, 2, 3968) is a maximum.

For functions of more than two independent variables, the first-order
conditions for a maximum are that all first-order partial derivatives are
zero. Second-order conditions are expressed using determinants and will
not be considered here.
6.6.2 Consider the conditions for a local minimum of a function of two

Conditions for a variables. First-order conditions are the same as for a local maximum, i.e.
minimum
dz . dz
—= 0 and — 0
ox oy
At the second order, condition (6.16a) is replaced by
dh
> 0 > 0 (6.18a)
dx^ dy^
and there is the further condition on the discriminant that
dh dh ^ " 2
_ > 0 (6.18b)
dx^ dy^ \dxdy)
This is the same as (6.16b). The only difference between the

conditions for a maximum and for a minimum are the signs of the
direct second-order derivatives. Consider an example. Find the
minimum value of
z = 2x — 3y — O.Sxy -f 0.5x^ -I- 0.25y^
The first-order conditions for a minimum are
dz
= 2 — 0.5y + X = 0
dx
dz
= —3 — 0.5x -f 0.5y = 0
dy
These conditions solve for x = 2, y = 8, which gives the only stationary

value. Is this a minimum.? The second-order derivatives are
d^z
^=1 î = 05 and = -0.5
Bx^ ■ dxdy
The direct derivatives are positive as required, and condition (6.18b) is

also satisfied since
(l)(0.5)-(-0.5)^ =0.25 >0
so the point x = 2, y = 8 is a minimum, giving z = —10.
Exercises 6.6
1 For the following functions, does the stationary value satisfy
the conditions for a local maximum?
(i) f{x, y] = 20x + 64y -|- 8xy — x^ — 32y^
(ii) f(x, y] = 2x^ +y^ — 16x — 20y — 2xy
(iii) f(x, y) = 5x — 4y + xy — 0.25x^ — 0.5y^

2 Find the maximum value of the following functions:
(i) f{x, y] = 100 — 4x + 5y + X/ — — 0.5y^
(ii) f (x, y) = 100 + 5x - 16y + 2xy - 0.25x2 - 8y2
3 Confirm that
z = 1000 — 21X + 576y + x2y + xy2 — 2x^ — 8y2
has a local maximum when x = 2 and y = 5.
4 Find the minimum value of the following functions:
(i) f(x, y] = 100 - 14x + 4y — 2.5xy + 1.5x2 2y2
(ii) f{x, y) = 126 + 6x — 54y — 3xy + 0.5x2 9^2
5 Show that
f(x, y) = x^ — 288x - 0.5x2y + 42y - 2xy2 + 16y^
has a local minimum value when x = 10 and y = 1.
6.7 We have examined maxima and minima but there is another form of
Saddle points stationary point - a possibility that does not exist for functions of one
variable - a saddle point. A saddle point is relative maximum in one
direction and a relative minimum in the orthogonal direction. Saddle
points are important in constrained optimization and the theory of
games. First-order conditions for a saddle point are the same as for
other stationary values:
dz ^ , dz
—= 0 and ^= 0
ox oy
But at the second order a sufficient condition for a saddle point is that
the discriminant is negative; That is,
dh dh fd^Y
dx^ dy^ \dxdy) ^
Saddle points have the appearance of a saddle or a mountain pass, with

the route reaching its highest point at a local low between the hills.
The appearance of a saddle point is illustrated in Figure 6.8.
It might be thought that for a maximum in one direction and a
minimum at right angles, in addition to (6.18c) one of the direct
second-order derivatives ought to be negative and the other positive.
This is not required but does arise when the relative maximum and
minimum are in the fundamental directions. Imagine such a case. Now
rotate the surface through 45°. This will not change the fact that a
saddle point exists, but it will alter the directions relative to which the
extrema exist and what happens in the fundamental directions. Unlike
a maximum - for which movement in any direction reduces 2 - what
happens to z as we move away from a saddle point depends on the
direction. So at a saddle point the direct second-order derivatives may
have the same sign or opposite signs, as the numerical illustrations will
confirm. But first show that the stationary value of
2 = 25 + 12x — 30y + 6xy + 1.5x:2 — 3y^

Figure 6.8
— 12 + 6y + 3x — 0
and
TT- = —30 + 6x — 6y — 0
dy
which solve for x = 2 and i/ = -3. At the second order
d^z _ d^z _ d^z

dx^ dy^ dxdy
Since the direct partial derivatives are of opposite sign, there is no need
to evaluate the discriminant, as (6.18c) must be fulfilled. So the
stationary value of z at (2,-3) is a saddle point. The function is
graphed in Figure 6.9, where it can be seen that the minimum is in the
X direction and the maximum in the y direction.
Now consider
f(x, y) = lOxy + x^ + y^
For any stationary value of this function
fx = lOi/ + 2x = 0
fy = lOx + 2y = 0
which conditions solve for x = 0 and y = 0. so the stationary point is

at the origin. At the second order
fxx — 2, fyy = 2 and f^y = 10
so
fxJyy - iixyf = (2)(2) - 10^ = -96 < 0

Figure 6.9 z
Centre x:0 y: 0 Length x: 10 /: 10 Derive 3D-plot
and a saddle point occurs as shown in Figure 6.10. In this case the saddle
is oriented at 45° to the x, y axes. The fact that both direct second-order
partial derivatives are positive means that the function, while zero at the
stationary value, is increasing in both fundamental directions.
Now consider
f{x, y) = Sxy — 0.25%^ — 0.25y^
Figure 6.10
Centre x:0 /: 0 Length x: 10 /: 10 Derive 3D-plot
At the first order
fx = 5y — 0.5x = 0
fy = 5x- 0.5y = 0

and at the second order
= -0.5, fyy = -0.5 and f^y = 5
so that again the discriminant is negative, ensuring a saddle point. This

function is graphed in Figure 6.11. The appearance of the function is
similar to that of Figure 6.10. But in this case the function, while
stationary at (0, 0, 0), is decreasing in both fundamental directions.
Figure 6.11
Exercises 6.7
1 Identify the stationary point of each of the following functions
and confirm that the points are saddle points:
(i) f(x, y) = 8 x - 5y -F 2xy - 2x^ - 0.25y^
(ii) f(x, y) = 15x -F 12y — 3xy — 0.75x^ -F 1.5y^
(iii) f(x, y) = 2x2 ^ o.5y2 -f 8xy - 200
2 Find the stationary point of the functions below, and establish

whether the point is a maximum, a minimum or a saddle
point.
(i) f(x,y)= 1000-220x + 310y- 14xy + 5x2 + 10y2
(ii) f(x, y) = 15x -F 17y — 7xy — 3x2 _ 4^2
(iii) f(x, y) = 34x - 1 Oy - xy - 2.5x2 ^ 2y2

3 Locate and establish the nature of the stationary points of the
following:
(i) z = 20x -F 68 y + 2xy + 4x2 __ 5^2
(ii) z = 17x — 15y -F 7xy — 4x2 _ 3^2
(iii) z = 8 x— 12y— 2 xy + 2x2 y2

(iv) z = 8x - 20y + 4xy + x2 - 2y2
(v) z = 500 — 2.5x2 _|_ _ ^y2
(vi) z = 4x2 _ o.5y2 + 4xy -F 32x - 20y

6.8 ^ It may help to show second order conditions for the various types of
Stationary values: stationary point in the form of a table. This is done in Table 6.1. The
resume bottom-left box is an impossible combination. Two further matters are
worth raising at this point. First, note that where the discriminant is
zero, i.e.
fxjyy - (Ixyf = 0
there can still be a stationary value of/(x, y). Second, in the numerical
examples we concentrated on quadratics and in all cases a stationary
value of some kind existed. But other outcomes are possible. For
example, when z is a polynomial of higher degree than two or is not a
polynomial at all, the number of stationary values may be a finite
number greater than one. Even where z is quadratic there are
possibilities other than one stationary value. Outcomes can be classified
as
1 A single stationary value

2 No stationary value
3 A continuum of stationary values
Table 6.1
Discriminant
> 0 < 0
Negative Maximum Saddle point

Direct
Derivatives Positive Minimum Saddle point
Opposite sign * Saddle point
What cannot occur in the quadratic case is a finite number of

stationary values greater than one. Where z is quadratic, first-order
conditions are simultaneous linear equations - which can have one
solution, no solutions or an infinity of solutions. We have seen
examples of category 1, so let us consider a function of two variables
with no stationary values. Consider
z = 6x^ + l-5y^ + 6xy — 18x
for which the first-order conditions for a stationary point are
dz
— = 12x -I- 6z/ — 18 = 0
ox
dz
— = 3y + 6x = 0
dy
But the equations here are inconsistent. They can be written as
12x + 6^ = 18
6x + 3y = 0

for which multiplication of the lower equation by 2, and subtraction
from the upper equation produce the unfulfillable condition
0 = 18
So there are no values of x and y which can satisfy both necessary

conditions and consequently there is no point where the function is
stationary. In this case the graphs of dz/dx = 0 and dz/dy = 0 produce
parallel lines in the x, y plane, as shown in Figure 6.12(a).
Figure 6.12
A minor modification to z results in a case of infinitely many

solutions to the first-order conditions. If
z= + l-5i/^ + 6xy — 18x — 9y + 13.5
then
— = 12x -F 6i/ — 18 = 0
ox
^=3i/ + 6x-9 = 0
dy
Each condition can be rearranged as the equation:
y = 3 — 2x
This corresponds to the single-line locus of dz/dx or dzjdy = 0 shown

in Figure 6.12(b). The graph of z is shown in Figure 6.13.
The surface produced by this function resembles a bent sheet of
paper touching the x, y plane along the straight line y = 3 — 2x. Note
that at the second order in this case
d^z , d^z
=3 and ——=6
dx^ dy^ dxdy

Figure 6.13 z
The direct second-order derivatives satisfy the conditions for a

minimum and are also consistent with a saddle point. But the test to
distinguish these possibilities fails as the discriminant itself vanishes:
(12)(3) - 6^ = 0
Figure 6.13 reveals points with co-ordinates related by ^ = 3 — 2% as

local minima. This can be confirmed at any point on the line by
calculation of z for variation in x or y alone.
We can now find local maxima and minima of polynomials if there
are no effective restrictions on the independent variables. We have
assumed that whatever values occur are sensible, i.e. unconstrained
optimization where values of variables may be freely chosen. Such cases
do arise - for example in problems involving rates of change. But we
also need to consider another important class of problems where there
are limits to the decision-maker’s freedom of choice - a topic introduced
in Chapter 7.
Exercises 6.8
1 For the following functions find out if a stationary value exists.
Where a stationary value does exist, establish its nature.
(i) f{x, y) = 12x2 + 12xy + 3y2 ^ 8x - 1 Oy
(ii) f{x, y) = 60x + 64.5y - 2.5x2 + 3.5y2 0.5xy
(iii) f(x, y) = 1000 — 20x2 _ ^^2 20xy — 50x -h 30y
2 Establish whether there are many or no stationary values for:
(i) f(x, y) = 2.25x2 + 9y2 _ 9xy - 9x + 1 8y + 100
(ii) f(x, y) = 0.5x2 2y2 2xy — 5x — 8y + 100

CHAPTER
Constrained optimization
Here we consider optimization models where the values of decision

variables are restricted - singly or jointly, for example by limited
supplies of key resources, budgetary restrictions, finite capacity of plant,
the fact that negative values may be meaningless and other realistic
requirements.
LEARNING OUTCOMES
After reading this chapter you will see what is involved in the solution
of a range of problems and be able to tackle many directly. You will be
able to interpret key parameters and extract valuable management
information from a range of models.
7.1 In practice the values of decision variables must meet certain conditions
Introduction to if they are to be allowable. In manufacturing, production levels should
constrained be positive or zero - sign restrictions apply. Other requirements may
optimization relate to resource bottlenecks, required order sizes and other values set
externally. Internal management decisions such as the allocation of
output or investment capital between plants may need to be observed.
Consider some examples.
A firm may want output levels that maximize profits subject to the
condition that available supplies of materials for the production process
are not exceeded. Or the firm may wish to minimize the cost of
producing an order of a given size - a problem of minimizing cost
subject to an output constraint. In terms of service provision,
management may find that present waiting times for its customers are
unacceptable and may seek an economical rearrangement of its services
to give better expectations of waiting times.
In fact the structure of constrained optimization problems is
characteristic of a wide range of business, economic and financial
problems. There is much more to the subject than we will have space to
consider, and we will focus on representative problems, the properties of
optimal solutions and their interpretation. Insight can be gained from
numerical examples, and in some cases we will use complete
enumeration (essentially a ‘non-method’) to get results, enable

consideration of optima (and other solutions) and assess the impact of
constraints.
Important though it is, it is not simply the solution, or the method of
solution that is significant, there is value in knowing the equilibrium
conditions, what they mean in any problem and the management
information that they contain. Current values of decision variables such
as production levels can be checked against the conditions to find out if
the position is optimal.
The problems can be approached in more than one way - for
example through mathematical programming, of which linear
programming (LP) is one instance, and numerical methods. We will
examine the following three methods in some detail:
• tangency:
• substitution:
• Lagrange multipliers.
To begin to see how this sort of problem can be approached we will

start with a particular function and find the maximum without
constraints. We will then see what difference is made by the
introduction of a single joint constraint on the independent variables.
The function is
z = 60x + 80y — 2xy — — 2y^ (7.1)
With no restrictions on the choice of values of x and y, for a maximum

it is required that
^ = 60 — 2i/ — 2x = 0
ox
and
dz
— = 80 — 2x — 4y = 0
oy
These first-order conditions solve for x = 20 and y = 10. At the second

order:
d^z d^z d^z

dx^ dy^ dxdy
All conditions are fulfilled, since the direct second-order derivatives are
negative and the discriminant condition is satisfied too, since:
(_2)(-4)-(-2)2=4>0
Thus the point with x and y co-ordinates (20, 10) is a maximum, and
by putting these values into z we find the maximum value of z to be
1000. The point (20, 10, 1000) is the one turning point of the
function (there being no other solutions to the first-order conditions)
and is the global, as well as a local, maximum. The point is the
Constrained optimization 251

unconstrained or free maximum of the function, which is illustrated
in Figure 7.1.
Figure 7.1
Centre x:0 /: 0 Length x: 10 y: 10 Derive 3D-plot
Now suppose that the decision-maker does not have complete

freedom to set values for the independent variables. Specifically, suppose
that the values of x and y must satisfy the requirement:
2x + 1/ = 40 (7.2)
which might be a budget constraint where x and y are the amounts of

two raw materials, the prices of which are £2 and £1 per unit,
respectively, and where £40 is the budget. The function (7.1) would be
profits expressed in terms of materials used. The aim here would be to
secure a maximum of (7.1) for values of x and y which can be bought
within budget (i.e. which satisfy (7.2)). There is a structural similarity
to linear programming (LP) problems, except that in this case the
objective function, (7.1), is non-linear. The problem is indeed a
programming problem - a quadratic programming (QP) problem if
certain other conditions are satisfied, and which we shall address here
in terms of the calculus.
First consider the tangency approach. In Figure 7.2 contours of z are
crossed by the line given by the equality constraint (7.2). The contours
crossed in this case correspond to higher values of z as the free
maximum, P, is approached. The constrained optimum is the point of
tangency, R, between a contour and the constraint line. Why should
this be so? Consider a point such as Q. This cannot be the maximum
because contours cross the constraint line at this point, implying that
contours inside (and higher) than that through Q can be reached by
moving along the constraint towards R with x increasing at the expense
of y. If the original point is Q' higher contours can be reached by

substituting y for x in movement across higher contours towards R. It
is only at point R that movement in either direction along the
constraint line would necessarily mean movement to a lower contour.
Figure 7.2
For an analogy, regard the problem as that of gaining the highest

point reached by a road (the constraint line) on an Ordnance Survey
map with topographical contours. The y direction represents movement
north, while the x direction corresponds to east. Movement towards R
from above is in a direction somewhat east of south. The highest point
of the road is not usually on the brow of the hill (which roads normally
avoid) but, rather, where it is tangent to a contour of the hill.
Although feasible in principle, the purely geometric approach of
drawing contours and finding the tangency point is long-winded and
inaccurate, but it does give a number of useful insights. Tangency
means that slopes are equal, which means equality of rates of change
or rates of substitution - for which there is usually an economic or
financial interpretation. In the present case the optimum is reached
when the rate at which x can be substituted for y maintaining cost is the
same as the rate at which the substitution can be made maintaining
profit. Advantage can be taken of any difference to increase profit, to
reduce cost, or both.
The tangency property of the optimum, spelt out in terms of the
expressions for slope, can be used to solve for the optimal values of x
and y. This is superior to measuring these values off the axes. The slope
of the constraint is found by writing the constraint with y as a function
of x:
1/ = 40 — 2x

So the slope of the line is
How is the slope of a contour of z found? The answer involves the

differential of z, dz, which is the approximate increment in z following
increments, dx and dy, in the independent variables. The differential of z is
j dz ^ dz .
dz = —dx + —di/ (7.3)
dx dy
In (7.3) the singular treatment of dz, di^ and dx differs from the use
of a term such as dy/dx as a composite symbol representing the value
towards which the ratio Ay/Ax tends. But there are occasions when
the separate interpretation of dx, dy and dz is valuable. In (7.3) dz is
the change in height of the tangent plane to the surface at a point, and
this approximates to the actual change in z. With a contour there can
be no change in z so long as any movement from the original point is
along the contour through the point, so we can write
dz dz
dz = T^dx + —dy = 0 (7.4)
dx dy
Rearrangement gives
dz dz
— dy = -^dx
dy dx
which can be re-expressed as
dzjdx
(7.5)
dx dz/dy
(7.5) is the implicit function rule, the usefulness of which in the

present context is in finding the slope of the contours of a function of
two independent variables. Given the requirement that tangency must
occur between a contour and the constraint line at the constrained
optimum, the implicit function rule can be used to find the co-ordinates
of the optimal point. We have already seen that the slope of the
constraint line is —2. Now in terms of the function, z, since
— = 60 — 2i/ — 2x
dx
and
dz
— = 80 — 2x — 4y
dy
the slope of the contours at any point is found using (7.5). Thus
dy —(60 —2y —2x)

dx 80 — 2x — 4y

Tangency requires the equation of slopes. Thus
_ -(60 -2y - 2x)

80 — 2x: — 4i/
2(80 — 2x — 4i/) = 60 — 2i/ — 2x

160 — 4x — 8y = 60 — 2i/ — 2x
and therefore
2x + 6y = 100 (7.6)
where (7.6) is fulfilled at any point of tangency. Our particular point of

tangency is located on the constraint line given by 2x + i/ = 40 and the
co-ordinates of the constrained optimum are obtained by simultaneous
solution of this with (7.6). Using elimination,
2x + 6y = 100
2x + y = 4:0
5y = 60
so
y = i2
and by back substitution into either equation
X = 14
The constrained maximum of z (obtained by substitution of the optimal

values of x and y into (7.1)) is
z = 980
This procedure, establishing and applying the conditions for tangency,

is used in the microeconomic theory of the firm to find expansion paths
for output combinations as the level of available resource increases. It is
also used in portfolio theory to find the critical line - which shows the
ideal balance between investments that ensures minimum risk for each
value of return. We have been assuming that the tangency point - the
solution of first-order conditions - gives a maximum rather than a
minimum. The nature of the function, the context of the application or
both may make it obvious that the required extreme is produced, but, in
general, second-order conditions need to be considered.
Exercises 7.1
1 Use the tangency approach to find the maximum value of:
f{x, y) = 22x 33y 3xy — 1.5x^ — 2.25/^
subject to X -t- 3y = 55

2 For:
f[x, y) = 20x + 16y - 0.1 - 0.05y^
by how much is the maximum of f(x, y) reduced when x and y

must satisfy 5x + 2y = 325?
7.2 The method of substitution is considerably easier to use than

The method of tangency and is often the method of choice in the case of the
substitution optimization of a function subject to a single linear constraint. We shall
use the current example to illustrate both the method and the useful
economic and financial information that can be extracted from
constrained optimization problems. The current problem is
Maximize z = 60x + 80y — 2xy — — 2y^

subject to 2x + ^ = 40
To solve the problem the constraint is rearranged to express y in

terms of x:
y = 40 — 2x
Now substitute for y in the objective function. This produces
z* = 60x + 80(40 - 2x) - 2x(40 - 2x) - x^ ~ 2(40 - 2xf
Because of the substitution, the problem has been reduced to one in a

function of a single variable to which the procedure for unconstrained
maximization can be applied. Also, we have called the function z*
rather than z to remind us that what we shall be optimizing is the
value of z provided that the constraint is satisfied. A different z* expression
would obviously have resulted if a different constraint had been used.
The expression for z* in this case simplifies to
z* = 140x -
and at the first order
dz*
— = 140 - lOx = 0
dx
Second-order conditions are satisfied here since
From the first derivative x = 14, implying y = 12, so that z = 980 is

the maximum value achievable given the constraint. This confirms the
result obtained with the tangency approach.
The plot of z* against x is shown in Figure 7.3. The figure shows
that the z* parabola has its vertex above x = 14. It was purely for
convenience that we selected the variable y to be substituted out. As an
exercise the reader should rework the calculation making the
substitution 20 — 0.5y for x.

What would happen if the right-hand side (RHS) of the constraint
was increased by one? You can be sure that if the problem models a
real business situation it will be a managerial priority to relax any
constraints that restrict the achievement of company objectives. In the
present case it is clear that an increase in the RHS will allow an
improved position, so an increase is a relaxation of the constraint (the
direction representing easement is not always clear for equality
constraints). If the RHS had been 41, re-working the problem
y = 4:1 — 2x
and after simplification
z* = 146x- 82
dz*
— = 146 - lOx = 0
dx
so
X = 14.6, y — 10.8 and z = 983.8
The increase in z of:
983.8 - 980 = 3.8
is attributable to the extra unit. This is a marginal valuation of the

resource represented by the RHS and is the most that is worth paying
to secure the extra unit. This important management information
allows judgement to be made on costed proposals to relax the
constraint. The marginal value will depend on the amount of resource
available and typically will be diminishing - as is the case here. Table
7.1 illustrates this changing valuation at the margin. Note that since
the unconstrained maximum occurs when the RHS is at 50 units,
being ‘forced’ to use an extra unit of resource (the constraint must be

Table 7.1 Right-hand-side z Marginal value
35 955 5.80
36 960.8 5.40
37 966.2 5.00
38 971.2 4.60
39 975.8 4.20
40 980 3.80
41 983.8 3.40
42 987.2 3.00
43 990.2 2.60
44 992.8 2.20
45 995 1.80
46 996.8 1.40
47 998.2 1.00
48 999.2 0.60
49 999.8 0.20
50 1000 - 0.20
51 999.8 - 0.60
52 999.2 - 1.00
53 998.2 - 1.40
54 996.8 - 1.80
55 995 - 2.20
satisfied as an equality) at this point would reduce z by 0.2; hence the

negative marginal value of the resource at this point. In the present
example, with quadratic z*, the local maximum is also the global
maximum. If the function z* is cubic or higher, because of either a
cubic z or a non-linear constraint, then despite the constraint there
may be no finite maximum or minimum of z. For example, the problem
Maximize z = — ISOy -f x^y

subject to 2x -\- y = 12
on substitution for y produces
z* = -2160 + 360x + I2x^ - 2x^
and, while the first-order condition solves for x = — 6 (minimum) and

X = 10 (maximum), these values give only local maxima and minima.
Without further restriction, z* is unbounded. As may be seen from the
sign of the dominant cubic term, it becomes arbitrarily large and
positive for negative x, and arbitrarily large and negative for positive x.
In practice, of course, there will be further restrictions on the values
that the decision variables can take. These restrictions often take the
form of sign requirements on the variables, which may also be
individually bounded from above. While allowance for sign
requirements can be tacked onto the substitution approach, the
Lagrange multiplier framework is more appropriate for this purpose. In
the present context we will simply remain on informal alert for
unbounded cases.

Consider a further example using the method of substitution with a
single linear constraint. This example also considers the useful
economic and financial information that can be found from constrained
optimization problems. The problem is
Maximize z = xy
subject to X + y — 12
Note that in this case if there was no effective constraint the function
would have no finite maximum. To solve the problem the constraint is
first rearranged to express y in terms of x. This gives
y = 12 — X
Now substitute for y in the objective function
z* = x(12 — x) = 12x — x^
from which
dz*
= 12 — 2x = 0
dx
The second-order condition is satisfied since
d^z*
-2
dx^
From the first derivative, x = 6, implying y = 6, and z = 36 is the

maximum achievable given the constraint.
What would happen in this case if the RHS of the constraint was
increased by one unit.? Reworking the problem with the increase
y = 13 — X
z* = x(13 — x) = 13x — x^
dz*
2x = 0
so that
X = 6.5, y = 6.5 and z = 42.25
The increase in z of 42.25 — 36 = 6.25 is attributable to the extra unit

on the right-hand side, and 6.25 is a marginal valuation of the
scarce resource (or whatever is represented by the RHS). But there is an
important difference between this case and the first example. Here,
because of the non-concave nature of z, the marginal value of the
scarce resource is increasing. For example, it may easily be found that
with 14 units on the RHS the maximum of z is 49, implying a value of
6.75 for the fourteenth unit of resource. In practice, at least beyond a
certain point, marginal values decline. This situation is also implied by
the concavity assumption of quadratic programming.

The most common type of constraint used in business and economic
models is linear, to which the method of substitution is well suited. But
it is possible to use the method with a non-linear constraint which can
be expressed as an explicit function. For example, suppose that we need
to find the maximum of the function
z = 30x -1- 5y
subject to
+ 2iy — 20 = 0
First, the constraint is written in explicit form. In this case it is more

convenient to state i/ as a function of x. That is,
^ = 10 — 2x^
Substitution into the objective function produces
z* = 30x+ 50 - lOx^
for which the first- and second-order derivatives are
30 — 20x = 0
dx
d^z*
-20 < 0
dx^
So the solution to the first-order condition produces a maximum. The

resulting values are
X = 1.5, y = 5.5 and z = 72.5
Care may be needed in general as only a local optimum may be

produced.
The method of substitution can also be used in some cases where
there is more than one constraint. ‘Using up’ the information in any
one constraint by making a substitution in the objective function
reduces the number of freely determinable variables by one. So when
we have a function of two variables to be maximized with respect to a
single constraint the z* function is of a single variable. In the case
where a function of two variables is to be maximized with respect to
two linear constraints there will be 2 — 2 = 0 variables over which the
decision-maker has control if the constraints are independent and
consistent. So the decision-maker is restricted to the values of the
variables that satisfy the simultaneous equations given by the
constraints. The decision-maker faces a fact rather than a problem. But
if z is a function of more than two variables, then two constraints will
still leave one or more variables to be freely chosen.

Consider the following problem:
Maximize z = wx + wy + xy + 6x
subject to 2x + y + w = 16
X + 3y — 2w = 23
Using the first constraint to express w in terms of x and y, the RHS of

the result
w = 16 — 2x — y (7.7)
is now substituted for w in the second constraint. The outcome is
X + 3^ — 2(16 — 2x — I/) = 23
5x + 5y — 32 = 23
X + y = 11
that is
^ = 11 — X (7.8a)
which when substituted into (7.7) gives
w= 5 — X (7.8b)
When there is one more variable than there are equations we can
solve for all but one of the unknowns in terms of the remaining
variable - the numeraire - which in this case is x. Now substituting
(7.8a) and (7.8b) into z gives
z* = (5 — x)x + (5 — x)(ll — x) + x(ll — x) + 6x

= 6x — x^ + 55
At the first order it is required that
dz*
— = 6-2x = 0
dx
At the second order the condition for a maximum is fulfilled since
Therefore solution of the first-order condition and substitution of the

result into (7.8a) and (7.7) gives the outcome
x~3,y = S,w — 2 and z* = 64
and in this case since z* is a quadratic we can be confident that the

solution is a global maximum of z subject to the constraint.
In the case of a function of three variables optimized subject to a
single constraint the use of the method of substitution reduces the
problem to that of unconstrained optimization of a function of two
variables. Consider the following problem:
Maximize z = 6wx — '+ xy

subject to 5x + 2w + i/ = 20
The most convenient explicit statement of the constraint is
y = 20 — 5x — 2w
and substitution for y in z produces
z* = 6wx — + x{20 — 5x — 2w)

= 4wx + 20x — 5x^ —
for which
dz*
= 4x — 2w = 0
ow
^ = 4w + 20- 10x = 0
ox
and
d'^z* _ d^z* _ d^z*

dx^ dxdw
so that the second-order conditions for a maximum will be fulfilled.

Solving the first-order conditions produces
X = 10, w = 20, y — —70 and z = 100
which represents the global optimum as well as a local one.

The method of substitution illustrates a valuable problem-solving
principle. When there appears to be no means of solving a problem it is
worth asking the following question:
Can the problem be converted, without altering its essential

character, into a form for which techniques of solution are
available.^
In applying this principle, note the proviso ‘without altering its essential
character’, the problem-solving art being to find a modification that
achieves this end.
A typical business problem involves more than one constraint. For
example, when looking at an investment plan over the medium term a
company will usually face limited investment funds in each year. The
manufacturer will usually face, at least in the short term, limitations on
capital equipment and floor space, as well as limited availability of some
materials and components. While some multi-constraint problems can
be addressed by the method of substitution, this would not as a rule be
the technique of choice - the Lagrange multiplier method provides a
better framework for multi-constraint problems. Substitution is a useful
technique, with the advantage of being simple, but it has limitations. It
can become impractical or unusable if
• there are many constraints:

• a constraint cannot be stated as an explicit function;

• the constraints are inequalities rather than equations;
• there are sign restrictions on the decision variables.
We now turn to a method that can accommodate these factors, at least

for formulation and characterization of the solution.
Exercises 7.2
1 Find the maximum value of
z = 1 OOx + 50y + 2.5xy
subject to the constraint that 8x + 5y = 200

2 Given the condition
-2x + y = 100
find the maximum value of the function z = 20y — 0.1 x^

3 Use the method of substitution to find the maximum value of
z = 1650x2 - 80x2 _ 4^2y2 ^ 1000
subject to 0.5x + y = 20
4 Find the maximum value of z where
z = wx + 4xy + 1 Owy + 30w
subject to
x + 10y +5w = 80
-2x + 30y + 2.5w= 1 15
y) = ^ 000 - 5 wx + 6xy - 3y2
subject to 5x + 6y — 15w = 30
6 For the following problems, find the value at the margin of an
extra unit of the resource represented by the right-hand side:
(i) Maximize f(x, yl = xy
subject to 0.25x + 0.25y = 120
(ii) Maximize f|x, y) = 1 Ox + 0..25xy + 20y
subject to 2.5x + 4y = 100
(iii) Maximize f(x, y) = 100 + 200x - y2
subject to 2x-y = 100
(iv) Maximize fix, yl = 100x2 _ 5pxy + 200y
subject to 2x+y = 600
7.3 In the substitution method the constraint was included within a revised
Lagrange objective function to convert a problem in constrained optimization into
multipliers and a simpler unconstrained problem. The constraint was included in a way
equality that made it implicit in a modified objective function and is asymmetric
constraints in that it singles out one variable for special treatment. For it to be able

to do this it must be possible to solve for one variable in terms of the
others. There will be cases in which this cannot be done, but it would
be useful to keep the idea of including the constraint in a revised
objective function and seeking an unconstrained stationary point. This
is done in the Lagrange multiplier method.
Consider the original problem again. The function to be maximized is
z = 60x + 80y — 2xy — — 2y^
with the constraint
2x + ly = 40
First express the constraint implicitly as
40 — 2x — ly = 0
and incorporate the left-hand side into a revised objective function:
L = 60x + 80y — 2xy — x^ — 2y^ -\- j(40 — 2x — y)
This is the Lagrangian function, in which the new unknown, j, is

the Lagrange multiplier. Note that L is a function of x, y and ;, and
that so long as the term
;(40 -2x-y)
is zero L will have the same value as z. To see how the method works,
differentiate L partially with respect to x, y and j, and set the derivatives
to zero to produce a stationary value of L. The first-order conditions are
then
dL
L;t = t;- = 60 — 2i/ — 2x — 2; = 0
ox
dL
Lu = — = 80 — 2x — 4y — j = 0
dy
dL
Li = — = 40 — 2x — y = 0
d]
Note that the derivative with respect to the Lagrange multiplier

means that the constraint must be satisfied, and that the stationary
value of L and the corresponding value of the original function are the
same. It can also be shown that, subject to certain conditions, these
first-order conditions give a maximum of z subject to 2x -F i/ = 40.
While this is not obvious, solving the first-order conditions confirms the
result from substitution. From the partial derivative of L with respect to
X, we obtain
i = 30 — y — X
Substitution into the derivative of L with respect to y gives
50 — X — 3y = 0

which can be solved simultaneously with the derivative Lj to obtain the
optimal values of x and y. After rearrangement we have
2x + 6y = 100
2x + y = 40
5y = 60
so
y = 12, X = 14 and j= 4
and the value of the objective function is F = 980.

The value of the Lagrange multiplier is the approximate change in z
(actually 3.8) if an additional unit of resource was available.
What guarantees that the result is a constrained maximum of z?
There are two questions here. Is the point a local maximum of z subject
to the constraint and is the point a constrained global maximum.? For a
local maximum, second-order conditions require concavity of the
objective function at the stationary value. A sufficient second-order
condition for a function of two variables maximized subject to one
equality constraint proves useful. Given the problem
Maximize f{x, y)
subject to g(x, y) — 0
the Lagrangian for which will be
h(x, y, j) =f{x, y) +j[gix, y)]
then at the second order it is sufficient that the value E, defined by

2 2
^ — l^xxSy 2hxygxgy T l^yydx
should be negative. In E
_dg %
~ dy ■
dy
where all components of E are evaluated at the x, y and j values that

satisfy the first order conditions. Because of the way the two constraint
derivatives gx and gy enter the condition (either squared or multiplied
together) it does not matter whether we write g{x, y) as:
40 — 2x — y
or as
2x + y — 40
If E turns out to be positive a constrained minimum is produced; if

E = 0 no statement can be made on the basis of this test about the
nature of the stationary value. In the current example
i-'xx — 2, Lxy = 2, Lyy = 4, gx = 2, gy = 1

so
E = ( - 2)( - 1)" - 2( - 2)( - 2)( - 1) + ( - 4)( - 2)2

= -2 + 8-16 = -10<0
and the point is a constrained maximum.

Second-order conditions often have a revealing interpretation - for
example in the neo-classical theories of the firm and the consumer -
but for actually finding optimal values they can be more trouble than
they are worth. It can be a major task to solve first-order conditions.
We will usually assume that the objective function has at least local
concavity for maximization or convexity for minimization so a solution
to the first-order conditions is at least a local optimum.
The Lagrange multiplier j gives the rate at which the optimal value
of z changes as the constraint right-hand side (RHS) varies. As a rate of
change at a point, this is the marginal valuation of the resource and is
a useful guide to the price worth paying at the margin for further units.
How this works out in the present case can be seen by re-solving the
problem with the RHS at 41. The equations to solve become
2x + 6y = 100
2x -\-y = 4:1
5y = 59
It follows that
y = 11.8, X = 14.6 and z = 983.8
confirming that the change in z resulting from the extra unit of

resource is 3.8. In this example ;, as a rate of change at a point,
represents an upper bound on the change in z following a unit increase
in resource. The value of; in this case rules out a proposal costing more
than j to relax the constraint by one. But without carrying out the
calculations we cannot evaluate proposals costing less than j. It is
instructive to see the pattern of results as the RHS changes. Details are
shown in Table 7.2. Note how x, y, j and AF move by constant
increments as the RHS value [b) goes up. The x and y values at b = 50
are those at the unconstrained maximum. When the RHS is increased
beyond 50 the objective function is forced down (since with an equality
constraint resources cannot be left unused), so j becomes negative.
Caution is needed when, in maximization problems, the objective
function is not globally concave. With second-order conditions satisfied,
the Lagrange multiplier method gives constrained local maxima but the
objective function may be unbounded despite the constraint. But
problems where profit or revenue forms the objective function do not
have such outcomes! This may mean that the function used to model
the real situation has all the properties it is convenient to assume. But
it may mean that although the objective function itself has no finite

Table 7.2
b X y i F
35 11 13 6 955
36 11.6 12.8 5.6 960.8
37 12.2 12.6 5.2 966.2
38 12.8 12.4 4.8 971.2
39 13.4 12.2 4.4 975.8
40 14 12 4 980
41 14.6 11.8 3.6 983.8
42 15.2 11.6 3.2 987.2
43 15.8 11.4 2.8 990.2
44 16.4 11.2 2.4 992.8
45 17 11 2 ' 995
46 17.6 10.8 1.6 996.8
47 18.2 10.6 1.2 998.2
48 18.8 10.4 0.8 999.2
49 19.4 10.2 0.4 999.8
50 20 10 0 1000
51 20.6 9.8 -0.4 999.8
52 21.2 9.6 -0.8 999.2
53 21.8 9.4 -1.2 998.2
54 22.4 9.2 -1.6 996.8
55 23 9 -2 995
maximum there are other restrictions on the values of x and y which

keep the function finite - sign requirements, for instance.
7.3.1 Minimization of a function of several variables can also be approached

Minimization problems through Lagrange multipliers. In problems with equality constraints
and variables which are not sign restricted, first-order conditions for a
constrained minimum are the same as for a constrained maximum. At
the second order a constrained minimum requires that
^~ ^^xySxSy T-^yydx ^ ®
As an example of minimization consider the problem
Minimize F = 0.5x^ + 8y^ + 4:xy — 80x — 240i/ + 5000

subject to 2.5x + 5y = 75
The Lagrangian function is
L = 0.5x^ + 8y^ + 4x1/ — 80x — 240i/ + 5000 +/(75 — 2.5x — 5y)
and the first-order conditions are
(i) L;j; = X-b 4i/— 80 — 2.5; = 0

(ii) Ly = 16y + 4:X — 240 — 5] — 0
(iii) Lj = 75-2.5x-5y = 0
From (i)
2.5; = X + 4i; — 80

so that
5; = 2x + 8^ — 160
which in (ii) produces
16i/ + 4x — 240 — 23c — 8^ + 160 = 0
i.e.
8/y + 2x — 80 = 0
with the result that
X = 40 - 4zy
substitution into (iii) produces
75 - 100 + lOi/- 5i/= 0
so that
5^ = 25
and therefore
y= 5
It then follows from substitution into the earlier equations that
X = 20 and j = —16
Now check the satisfaction (or otherwise) of the condition on E at

the second order. In this case we have
kjcjc — 1) Lxy = 4, Lyy = 16, Qx — 2.5, 0y = 5
so that E evaluates as
E = (1)( - 5)2 - 2(4)( - 2.5)( - 5) + 16( - 2.5)^

= 25 - 100 + 100 = 25 > 0
which is the required sign for a constrained minimum of F. With

X = 20 and y = 5, the minimum value of E is F = 3000. The negative
value of the Lagrange multiplier means that as the RHS of the
constraint is increased a lower value of F is achievable.
An example shows the caution necessary when the objective
function is not globally concave. Here the Lagrange multiplier method
produces a constrained local maximum (and a constrained local
minimum) but the objective function is unbounded despite the
constraint. The problem is
Maximize z = lOOOx + 200^ - x^y — 55x2

subject to 2x + y = 20

The Lagrangian function and first order conditions are
L = lOOOx + lOOy — x^y — 55x^ + j(20 — 2x — y)

(i) Lx = 1000 - 2xy - IlOx - 2] = 0
(ii) Ly = 200 — x^ — i = 0
(iii) Lj = 20 — 2x — y = 0
From condition (ii),
i = 200 — x^
Substitution into (i) produces
(i)' 1000-2x1/-IlOx-400 +2x^
From (iii),
y = 20 — 2x
and substitution into (i)' results in
1000 - 2x(20 - 2x) - IlOx - 400 + 2x^ = 0
which reduces to
x^ —25x + 100 = 0
which factors as
(x-5)(x-20) = 0
when X = 5, y = 10, j = 175 and z = 6365. To find the nature of this

point, consider the second-order condition. We have
Lxx — 2i/ 110, Lxy — 2x, Lyy —0
and
Qx — ~2, Qy = —1
so that, given L^x and L^y, E will be a function of x and y:
E = i-2y- 110)( - 1)' - 2( - 2x)( - 2)( - 1)
When X = 5 and y = 10, E evaluates as
E = -130 + 40 = -90 < 0
so the point is a local constrained maximum. Now, when
X = 20, y = —20, j = —200 and z = 2000
the value of E is
E= -70 + 160 = 90 > 0
so this point is a local minimum of z subject to the constraint. But the

function’s highest point is not at x = 5, nor is its lowest point at

X = 20. For higher x values beyond the constrained local minimum at
X = 20 the function increases without limit. For x values below x = 5
the function is unbounded from below. These outcomes can be shown
by use of substitution. Putting
y = 20 — 2x
into the expression for z gives
2* = 2x^ - 7Sx^ + eOOx + 4000
in which the leading term is the key since x can be arbitrarily large and
positive (no sign requirement on y) or negative (no sign requirement on
x). The graph of z* is shown in Figure 7.4.
Figure 7.4
7.3.2 For the maximization of a function of two variables subject to a linear

Interpretation of equality constraint, we can write:
Lagrange multipliers
Maximize z =/(x, y)
subject to ax + cy = b
Hence
L = fix, y) + jib - ax - cy)
dl Of . ^
(i)
dx~ dx ~ °
dL df
(ii) ^ ^ - ;c = 0
oy ay
(iii) b — ax — cy = 0
dy

Solving (i) for j,
a dx
while from (ii):
, _ 1 ^
^ c dy
The derivative df /dx is the rate at which the objective function rises
as X is increased. The ratio 1/a is the rate of change of x as the
resource level is increased (since the constraint is linear, 1/a is also the
number of units of x from an extra unit of resource, but it is the rate of
change that matters here). For example, if a = 2, then x can be
increased at the rate of ^ a unit per unit of resource transferred. The
product
1 ^
a dx
can now be interpreted as the rate at which the achievement of the

objective changes, as further resource is applied to x.
Condition (i) sets the value of the Lagrange multiplier to this rate.
Similarly, in (ii) the Lagrange multiplier is set at the rate at which the
objective changes as further resource is applied to y. Equalizing these
rates through j at the optimum means that the rate of change is the
same if a marginal variation in resource is applied through either
variable. The necessity of this condition can be seen when the rates
differ. Suppose that at some point the following values obtained
while
a dx
These rates apply in either direction at the margin, and it may help to
think in terms of a small finite change. If a unit of resource is
withdrawn from use in x, then on this account the value of the objective
function falls by approximately 5. But if this same unit of resource is
now applied to increase y, then on this account the value of the
objective function increases by about 7. The net effect is therefore an
improvement of 2, so the original point could not be optimal.
There is always scope for gains when there is a difference between
marginal rates. For optimality there must be a uniform rate - and j is
that rate. The multiplier can be seen as a mediator between competing
claims on the scarce resource. In any one equation j carries information
about the value of the resource in an alternative use. The effect of j is
to ensure that at the margin the scarce resource is allocated optimally
between possible uses. In other words, the use of resource is maximally
efficient - it is economized. (A common misuse of ‘economize’ is ‘to
minimize the use of’. The true meaning is ‘to make the most efficient
use of’.)
Exercises 7.3
1 Using the Lagrange multiplier method, find the maximum value
of
f(x, y] = 20x + 15y — xy — — 0.5y^
subject to the constraint x + 2y = 20. Use the sufficient

second-order condition to confirm the maximum.
2 For
fix, y) = 1200x + 400y - 20xy - 40x2 _ 2.5y2
and given the resource constraint 1 Ox + 5y = b:

(i) Find the maximum value of f{x, y) when the value of b is
150.
(ii) Find the value at the margin of an extra unit of the
resource.
3 Minimize
y] ~ 2x2 ^ 4.5y2 - 1 OOx - 200y
subject to 2.5x + 1.25y = 40 and find the effect on the

objective of a unit increase in the right-hand side.
4 Find the values of x and y which maximize
^(•^/ y) = + 101 y + 9xy -f 5x2 _ Qy2

subject to 3x + 2y = 29.
What is the effect on the objective of a unit increase in the
right-hand side?
5 Find the maximum value of the function
f(x, y) = xV
subject to the constraint x + y = 20.
7.4 Problems in which negative values of the decision variables are

Sign-restricted meaningful can arise in a number of ways. There are variables such as
variables profit (loss), net present value (plus or minus) or the balance of payments
(surplus or deficit), which can be of either sign. There are models
expressed in terms of changes in the values of non-negative variables - for
example changes in stock levels, price indices or a measure of money
supply. But many models use physical quantities such as manufacturing
output, gross domestic product (GDP) or prices - measures which cannot
be negative. In other models practical upper and lower limits for variables
need to be recognized. For example, the maximum production level is set
by current plant capacity. If one decision variable is probability, p, the
value of p must be in the range 0 <p < 1.
We begin with the most common practical limitation on decision
variables - non-negativity. Introducing sign requirements makes any

model less convenient and it is useful to check whether, in a particular
case, sign requirements can be ignored or circumvented. One way is to
use for the objective a function for which negative values of the
decision variables will not arise. Alternatively, any negative values could
be discarded. This is often reasonable - as in the square-root model of
stock control. But avoidance devices are not always enough, and sign
requirements may need to be built in directly.
Consider the problem of maximizing a function of one, sign-
restricted, variable with no other constraints:
Maximize y=f{x)
subject to X > 0
Sign requirements could be seen as a form of inequality constraint.

Also, a sign requirement is an instance of a lower bound. Problems
with variables limited from below at levels other than zero can be
reformulated in terms of sign-restricted variables. We will assume that a
finite maximum of /(x) exists for a finite value of x within the domain,
and that the function is continuous and differentiable for all values of
X > 0. On this basis, cases such as the rational function
1
fix) =
(1 - x)
which is unbounded as x approaches unity, are ruled out. Given these

assumptions, three cases can be distinguished
7.4.1 Figure 7.5 shows a case where /(x) has a maximum for strictly positive
An interior optimum X. The optimum here is a turning point in the positive quadrant, so the
sign requirement does not affect the optimum. The maximum occurs
when X = x°. Two qualitative properties of this interior optimum should
be noted:
dw
X > 0 and —= 0
dx
Figure 7.5

Our assumptions rule out the interior situations shown in Figure 7.6.
Figure 7.6(a) shows a cusp, while Figure 7.6(b) shows an asymptote in
the positive quadrant. We shall not be concerned with such possibilities.
Figure 7.6
7.4.2 In Figure 7.7 the function has negative slope for all permitted values of
A boundary optimum X. It follows that the maximum of the function in the domain must
occur at the (lower) boundary, i.e. at x = 0.
Figure 7,7
Figure 7.8 shows a case where the optimum is at the boundary, but
where there is also a local maximum at a turning point in the domain.
However, the end-point produces the global maximum given the
domain. Two qualitative properties of the boundary maximum case
should be noted:
X = 0 and ^ < 0
dx
When we have considered the third case we will draw these results
together.

Figure
7.4.3 ‘Degenerate’ implies a limiting, coincidental or singular state where

A degenerate case special conditions apply - as in the merging of two normally distinct
situations. Figure 7.9 shows a case with a turning point at the end¬
point. Here the turning-point maximum and the boundary maximum
coincide. The qualitative conditions are
X = 0 and ^= 0
dx
Figure 7.9
The results can now be summarized. In all cases dy/dx is either zero or
negative. The derivative is negative only if x = 0. At the sign-restricted
optimum x is either zero or positive. The positive outcome occurs only if
dy/dx is zero. So in all cases
du
(a) < 0
dx
(b) X > 0 (79)
dy „
(c) X-— = 0
dx
The properties (7.9) are necessary conditions for a sign-restricted
local maximum. Relatively simple problems can be solved using the
following four-step procedure:

1 Find the derivative di//dx.
2 Test the satisfaction of the conditions for x = 0. With x = 0, (b) and
(c) must be satisfied and what remains is to check the sign of dy/dx
at X = 0. If all the conditions are satisfied record the values of X
and y.
3 Test the satisfaction of the conditions for dy/dx = 0. If di//dx = 0,
then (a) and (c) must be satisfied and any non-negative roots of
dy/dx = 0 should be found. Record any such values of x and the
corresponding y values.
4 Select the position with the best value of y recorded under 2 and 3
above. Given the properties assumed for the objective function, this
position is the sign-restricted optimum.
Note that in any one problem both steps 2 and 3 can give results
satisfying all conditions. So there may be several candidates for the
overall optimum. In step 4 we distinguish between these points by
enumeration, but higher-order derivatives could be used to eliminate
local minima and points of inflection. We could also confirm the
behaviour of the function for extreme positive values for x. Examination
of the sign of the leading coefficient should provide the assurance
required.
Consider an example:
Maximize y = 20 + 8x — x^
subject to X > 0
First find the derivative as required by step (1). It will also help to set
out the conditions in full for ease of reference. It is required that
du
~=8-2x<0
dx
X > 0
Now go to step 2 and try x = 0. To ‘try’ this value means to see if

condition (a) is met at x = 0. It is not, since dy/dx = 8 > 0. Step 3 is
next. Here we try dy/dx = 0. Since with dy/dx = 0 conditions (a) and
(c) must be satisfied, to try this possibility means to look for non¬
negative X for which dy/dx = 0. Since the only solution is x = 4, this
value satisfies condition (b) and the point x = 4, y = 36 is recorded.
Step 4 here is trivial as only one point was identified under 2 and 3. So
(4, 36) is the sign-restricted maximum - as a sketch of the parabola
confirms.
For a less straightforward exercise consider the following:
Maximize y = 21x^ - 2x^ - 60x -|-100

subject to X > 0

For step 1 the first derivative is
~ = 4:2x-6x^ - 60 < 0
dx
Moving to step 2 and testing the sign of the derivative dy/dx at x = 0

gives
-60 < 0
dr
so the condition on the derivative is satisfied and r = 0, i/ = 100 is

recorded as a possibte optimum. Now move to step 3 and try
dy/dx = 0. Satisfaction of (a) and (c) is ensured, so the only question is
the sign of r. The quadratic solves for x ~ 2, and r = 5, which both
satisfy r > 0. Second-order conditions could be used to show that x = 2
gives a relative minimum and r = 5 represents a local maximum. But
in step 4 inserting all three candidate values into the expression for y
gives
r y
0 100
2 48
5 75
so the sign-restricted maximum is when r = 0. The sign of the leading

coefficient or a sketch of the function, as in Figure 7.10, will confirm
this result.
Figure 7.10
Maximize z = 100 — 25r-f 1.25r^

subject to r > 0

for which
dz
— = -25 + 2.5x
dx
At X = 0, dz/dx = -25, sox = 0isa candidate for a sign-restricted
maximum. When dz/dx = 0, x = 10, so this is also a candidate. We
choose between x = 0 and x = 10 on the basis of the value of z:
X z
0 100
10 -25
It might appear that x = 0 gives the sign-restricted maximum. But

this is not so, as in this case our original assumptions about the function
are not met, since z is unbounded for large positive values of x. This is
revealed by the positive leading coefficient. Second-order conditions would
have identified the turning point at x = 10 as a minimum but would
not, of course, have applied at x = 0. If there is reason to suppose that
the objective function may continue indefinitely to change in the
direction of optimization, the method should be used with care. The
context in which the method is being applied may make this clear.
For sign-restricted minima we must also allow for boundary as well
as interior optima. Sketches will confirm that the only difference from
(7.9) in necessary first-order conditions is that the slope of the function
at a boundary minimum should not be negative. If the end-point is to
give a minimum the function must not be going down as we move into
the feasible region. Conditions at the first order are
X > 0 (7.10)
Minimize y = 2x^ — 294x -I- 100

subject to X > 0
The conditions for a sign-restricted minimum are
-^= 6x^ - 294 > 0

dx
X > 0
First try x = 0. At this point dz/dx = —294, so the derivative condition

is not satisfied. Now try dz/dx = 0. The quadratic solves for
X = 7 and x = —7

where —7 breaks the sign requirement. So the sign-restricted minimum
is at X = 7, when the value of y is —1272. The graph of the function
appears as in Figure 7.11.
Figure 7.11
As a further example, consider finding a minimum of the quartic
y = 3x^- 28x^ + 24x^ + 144x + 500

subject to X > 0
Necessary conditions are
~ = 12x^ - 84x^ + 48x + 144 > 0

dx
X > 0
At the end-point x = 0, the derivative is positive, so this is a possible

sign-restricted minimum. Setting the derivative to zero under step 3
means solving a cubic. The roots emerge as
X = — 1, X = 2 and x = 6
After eliminating the negative root, the function is now evaluated at

permissible values of x:
X y
0 500
2 708
6 68
»
so the sign-restricted minimum is at x = 6.

Now consider maximizing a function of two independent variables
with sign requirements on each variable and no other restrictions, i.e.
Maximize /(x, y)
subject to X > 0, y > 0

Necessary conditions here are a generalization of the conditions for a
function of a single variable:
fx<0
xfx = 0
X> 0
(7.11)
fy<0
yfy = 0
y>0
Since each variable, the derivative of the function with respect to the
variable, or both must be zero, a possible procedure to find solutions to
conditions (7.11) would be to try allowable combinations in a
systematic way. We could, for example, proceed as follows:
Try: Check for:

1 X = 0, y = 0 fx<0fy<0
2 X = 0, fy = 0 fx<0y >0
3 A = 0, y = 0 X > 0 fy <0
4 /, = 0,/, = 0 X > 0 y > 0
Maximize f(x, y) = 15x — lOy 5xy — 2x^ — 4y^

subject to X > 0, y > 0
for which
/* = 15-5y-4x /y = -10-5x-8y
First try X = 0 and y = 0. If it turns out that /* < 0 and fy <0, then
this is a candidate for the sign-restricted optimum. When x = 0 and
^ z= 0, /c = 15 and fy = —10. The outcome for fx means that this point
cannot be optimal. The results for this, and the other, combinations are
shown in Table 7.3. The conditions are only fulfilled in case 3, and the
sign-restricted maximum is /(3.75, 0) = 28.125. Note that case 4
corresponds to the unrestricted maximum (for which/(x, y) = 150)
and that the function does not increase without limit for x and y in the
non-negative region. Since neither x nor y can be negative, the term
— 5xy cannot be positive. Neither can the terms —2x^ and —4y^. Since
these three are all the terms of degree two, the sign-restricted optimum
in case 3 must be the global optimum. In general the number of cases
to be examined in this approach is 2", where n is the number of sign-
restricted variables.
Table 7.3
1 X =0 y = 0 fx= 15 fy ~ -10
2 X = 0 fy = 0 fx= 21.25 y = -1.25
3 fx =0 y = 0 X = 3.75 fy — -28.75
4 fx =0 fy = 0 X = 10 y = -5

In problems with several independent variables some variables may
be restricted in sign, while others are not. If you are required to
maximize
f{Xl, X2.Xn)
then if some variable Xi is sign restricted it will be required that
fxi<0
îfxi — 0
Xi > 0
If some variable Xj is not sign restricted the normal condition
fxi = 0
is all that is required at the first order. Following the procedure of Table
7.3, if m of the n variables in the problem are sign restricted there will
be 2"^ cases to enumerate.
Consider an example. Find the maximum value of the function
f{x, y) = 40x — 50y + 5xy — 3x^ — 2.Sy^
when the following apply
(i) X is sign restricted;

(ii) y is sign restricted;
(iii) both X and y are sign restricted;
(iv) neither x nor y is sign restricted.
The derivatives are
= 40 + 5i/ - 6x fy = -50 + 5x- 5y
and for (i) the necessary conditions are
fx<0
xfx = 0
X> 0
fy = 0
Trying x = 0 we have
fx^4:0 + 5y<0 fy =-50-5y = 0
so from fy = 0, y = —10 and therefore fx = —10, and the point where
x = 0, i/ = —10 and f{x,y) = 250
is a possible optimum. Now trying fx = 0, we then require
fx = 4:0 + 5y-6x = 0 fy = -50 -\-5x-5y = 0

which solve for
x=—10 and y =—20
which is not feasible in problem (i). So the optimum here is
x = 0, i/ = -10 and f{x, y) = ISO
Now consider problem (ii). The requirements here are
fx = 0
fy<0
yfy = 0
y>o
Trying y = 0, it is then required that
f^=40-6x = 0 fy = -50 + 5x<0
From fx = 0, the value of x is x = 20/3 and fy = —50/3, so this is a

possible optimum. We have seen that when fy = 0 and fx = 0 negative
values of both variables result. So the optimum for (ii) is
x = 20/3,y = 0 and /(x, i/) = 400/3
Now consider case (iii). The results are shown in Table 7.4. So the
optimum is as for case (ii). In problem (iv) we have already seen that
the unrestricted optimum is at
x = —10, i/=—20 when f{x,y) = 300
Table 7.4
X = 0 y = 0 fx = 40 fy ~ -50
X = 0 fy = 0 fx = -10 y = -10
fx = 0 y = 0 X = 20/3 fy = -50/3
f. = o X = -10 y =
-20
A lower bound may be other than zero, as when a manufacturer

wants to ensure that the order of a favoured customer is fulfilled. In
such cases the variable can be transformed, without distorting the
problem, to one for which the lower bound is zero. The problem can
then be solved in the new sign-restricted variable and the results re¬
expressed in terms of the original variable. For an example, consider the
following:
Maximize z(x) = 160x — 2x^

subject to X > 50
First define a new variable, w, such that
w= X — 50
so that
% = vv + 50

Now re-express the problem in w:
Maximize z{w) = 160(w -f 50) - 2(w -f 50)^

subject to w -|- 50 > 50
which simplifies to the following:
Maximize z(w) = 3000 — 40w — 2w^

subject to w > 0
for which the conditions for a sign-restricted optimum are
dz
= -40 - 4w < 0
dw
w > 0
When w = 0, dz/dw = —40, so the conditions are met. When

dz/dw = 0, w = —10, which breaks the conditions. So the optimum is
at w = 0, which corresponds to x = 50 and z = 3000. Figure 7.12
graphs the problem.
Figure 7.12
The unrestricted maximum at % = 40 corresponds to w = —10.

Variables can be restricted from above as well as from below. If a
variable is restricted only from above, then (excepting the cases
mentioned earlier) the maximum of a function of this variable will
occur either at the end-point or at a turning point below the end-point.
This parallels the sign-restricted case. We can find necessary conditions
by transforming the problem to one in a sign-restricted variable.
Consider the following:
Maximize z{x) = 160x — 2x^

subject to X < 30

Define a new variable, v, such that
V = 30 — 3C
Clearly
V > 0
Now, since
3C = 30 — V
substitution of this relationship into the original problem gives
z(v) = 160(30 - v) - 2(30 - v)^

30 - V < 30
which simplifies to
z(v) = 3000 - 40V - 2v2

V> 0
for which the necessary conditions are
^ = -40 - 4v < 0
dv
V> 0
which conditions are fulfilled for v = 0, giving a constrained optimum at

the end-point, where jc = 30. Our purpose was to give conditions in
terms of the original variable. To find these conditions, first consider the
requirement dz/dv < 0. What does this mean in terms of dz/dx.? The
composite function rule can answer this question. Since v is a function of x
dz dz dv
dx dv dx
but since
V = 30 — X
it follows that
so:
dz dz
dx dv

Since the conditions in terms of v require that dz/dv < 0, this means
that, in terms of x,
dz
— > 0
dx:
Now consider the requirement that
= 0
We can write this as
dz
that is
dz
which means that
(30 — x)^ = 0
dx
Thus the full conditions in terms of x are
dz
d^-
(30 — = 0
dx
(30 - x) > 0
In general, given the problem
Maximize z(x)
subject to X < b
it is necessary that
dz
—> 0
dx
dz (7.12)
{b-x)— = 0
dx
(b — x) > 0
which is the form of conditions we require.

^ Now suppose that a variable is restricted both from above and from
below. Without loss of generality, we can take the lower bound to be
zero, so in terms of a single variable the problem is as follows:
Maximize z(x)
subject to X > 0
X < b

Necessary conditions at the first order include elements of both (7.9)
and (7.12). Here we will describe an implementation that enables the
solution of problems on a modest scale. There are three possible
outcomes:
1 An optimum at the lower bound.

2 An optimum at the upper bound.
3 An interior optimum.
Recall that a turning point may occur at the boundary and we are
excluding functions with such properties as a vertical asymptote in the
domain. The requirement on the derivative (non-positive, non-negative
or zero) will vary between outcomes 1, 2 and 3. For a local optimum at
the lower bound
whereas at the upper bound it is required that
and, of course, at a turning point
Subject to the usual provisos, a five-step procedure is useful:
1 Find dz/dx.
2 Try x = 0. If dz/dx < 0, record x = 0 as a possible optimum.
3 Try x = b. If dz/dx > 0, record x = fc as a possible optimum.
4 Solve dz/dx = 0. Roots in the interval 0 <x <b should be recorded
as possible optima.
5 Evaluate z at all points recorded under steps 2-4. The greatest
value of z identifies the optimum.
Consider an example. Find the maximum value of
z= 500-240x + 39x^ -2x^

subject to X > 0 and x < 10
Applying the procedure, first find the derivative
-240 + 78x- bx^

dx
At X = 0, dz/dx = —240. So x = 0 is recorded. At x = 10,

dz/dx = —60, ruling out the upper bound as a possibility. Solving
dz/dx = 0 gives x = 5 and x = 8. The function is now evaluated at the
three following positions:

X z
0 500
5 25
8 52
So the maximum is when x = 0.

Two variable problems with sign requirements on both variables but
also involving a linear equality constraint are of interest because many
examples of constrained optimization in microeconomics take this form.
The usual contexts (consumption, production or resource levels) imply
that negative values of the decision variables are meaningless. There is
not much extra work involved explicitly to take account of realistic
signs for the variables, since such problems can be re-expressed as
problems in a single variable restricted from above and from below.
Consider
z = lOOx — — 112y — 4i/^ -|- 2xy
for which a maximum is sought with respect to the linear constraint
5x + y = 90
Without sign requirements the maximum value of z is 1920 when

X = 20 and y = —10. Now add sign requirements on both x and y. So
it is also required that
X > 0 and y > 0
From the constraint
1/ = 90 — 5x
but now since y >0 it must be the case that
90- 5x > 0
that is
X < 18
So after substitution for y the problem is now to maximize
z = lOOx - x^ - 112(90 - 5x) - 4(90 - 5x)^

+ 2x(90 — 5x)
= 4440x- lllx^ -42 480
to which the method above can be applied. The derivative is
dz
-- = 4440 - 222x
dx
At X = 0, the derivative dz/dx = 4440 > 0, so this is not a possible

maximum. At the upper bound, x = 18, dz/dx = 444, so this position

is a possible constrained maximum. Setting dz/dx = 0 gives x = 20, so
this point is ruled out. The constrained maximum is therefore
X = 18, 1/= 0, z = 1476
The total effect of the sign requirements can be costed. It is the

difference between the values of the objective function achieved without
and with the sign requirements. That is
1920- 1476 = 444
If there is discretion as to whether variables are or are not sign

restricted (as for example with dividend changes), then the price of not
exercising the discretion can be calculated as above. This is valuable
management information.
Exercises 7.4
1 (i) Maximize z = 24x - 1.5x2 ^ 120
subject to
X > 0
(ii) Maximize z= 100-21x-3.5x2
subject to X > 0
2 (i) Maximize z = 1000 + 120x + 27x2 _ ^3
subject to X > 0
(ii) Maximize z = 8000 - 1440x2 + 128x2 - 3x^
subject to X > 0
3 (i) Minimize z= 1.2x2 _ 12x + 30
subject to X > 0
(ii) Minimize z = 7x2 + 84x + 50
subject to X > 0
4 (i) Minimize z = x2 - 13.5x2 - 30x + 650
subject to X > 0
(ii) Minimize z = 0.75x^- 110x2 + 1500x2
subject to X > 0
5 For
y) = 100 — 300x + 40y + 1 Oxy — 30x^ — y^

find the maximum value of f[x, y) subject to the conditions that
(i) X>0
(ii) y > 0
(iii) X > 0 and y > 0
(iv) Both X and y are unrestricted in sign.
(i) Maximize z = 500 + 150x - x2
X < 50
subject to
(ii) Maximize z= 1000 + 2000x-8x2
subject to X < 100
(iii) Maximize z = 1600x - 2x2
subject to x<450

7 (i) Maximize z = 2500-528x +45x2 -x3
subject toX > 0 and x < 20
(ii) Maximize z = 97.5x2 - 900x - x3
(iii) Maximize z = 2x3 - 105x2 ^ 1500x - 1000
8 For
f{x, y) = 100 - x2 - 2.5xy - 2/2 + 1 Ox + 2y
find
(i) Tfie unrestricted maximum.
(ii) Tlie maximum subject to tine constraint 2x + y = 20.
(iii) The maximum subject to the condition 2x + y = 20 and
the sign requirements x > 0 and y > 0.
7.5 Weak inequalities are the most common form of limitation in business,
Lagrange financial and economic applications. Resource constraints are almost
multipliers and always ‘less than or equal to’ rather than ‘strictly equal to’. Sign
inequality requirements and upper bounds can be seen as a form of weak
constraints inequality. For example in a problem involving the non-negative
variables Xi and X2 the sign requirements could be written as
— Ixi -b 0X2 < 0
Oxi — 1X2 < 0
Conditions for optima with variables restricted both from above and
below provide a route into the conditions required for more general
inequality constraints. Consider a problem in a single variable (not sign
restricted) with a single inequality constraint. Suppose that
F = 600 + 72x - 3x2
is to be maximized subject to
4x < 40
F may be profit expressed in terms of a variable input, x, on which

maximum expenditure is 40. Set up the Lagrangian as for an equality
constraint
L = 600 + 72x - 3x2
for which
= 72 — 6x — 4;
dx
and
^ = 40 — 4x
dj

What are the requirements on these derivatives at an optimum? Clearly,
since dLjd] is a rearrangement of the constraint, this derivative must
be non-negative. But what of dLjdxl To approach an answer, let us
look at the sketch of the problem as shown in Figure 7.13.
Figure 7.13
The problem is one of maximizing a function subject to an upper

bound on the decision variable. What happens at the end-point is
crucial in determining whether this value produces a maximum or
whether an interior point is optimal. Suppose x was set at its end-point
value; F could then be expressed in terms of the end-point. The
behaviour of the function as the end-point varies can be found from its
derivative. Let b represent the value of the right-hand side (RHS) (with
the original data b = 40), then at the end-point x = bj4:. Now
substitute this value into F, The result is
F = 600 + 72(V4) - 3(V4)^

= 600 + 18^-317^16
Differentiating,
dF
-=18-6V16
Equating dF/db to zero gives b = 48. This value of b maximizes F given

X = b/4. If b is less than 48, dF/db will be positive, indicating an
increase in F if the constraint is relaxed. The value of dF/db is the rate
of increase of F given b. The rate of increase of the maximum value of
F as the constraint is relaxed is also given by the Lagrange multiplier.
Now consider various values of b. With b = 48 (i.e. x = 12),
18-6(48)/16 = 0=;
and, with x = 12 and ; = 0,
=72-6(12)-4(0) = 0

Now suppose that b is set at a value less than 48, say 40,
corresponding to x = 10. Then
|^=18-6(40)/16 = 3=;
and, with j = 3 and x = 10,
dL
— =72-6(10)-4(3) = 0
In either case the derivative dL/dx is zero. If a value of b greater

than 48 was selected and if it was required that x = b/4 with no option
to satisfy the resource constraint as an inequality, then the resulting
value of i would be negative, but still this value of j, along with the
corresponding value of x = b/4, would produce dLjdx = 0. With the
option to satisfy the resource constraint as an inequality, the minimum
value of j is zero, consistent with the fact that marginal variations in
the RHS would have no effect on the optimum. In this case we are able
to set dF/dx = 0 (corresponding to the interior and unconstrained
optimum position) and, with j at zero, dLjdx = 0 in these
circumstances too. So in all cases we may equate this derivative to zero.
How can it be determined whether the resource constraint should
hold as an equality or an inequality.? Introduce a slack variable into
the constraint to produce an equality for all values of x and write
4x + s = b
where the slack variable s is any excess of resource availability (the

constraint RHS, b) over the resource requirement (4x). For the original
wealc inequality to be satisfied the slack variable s must be non¬
negative. Recall that at the optimum the Lagrange multiplier j will be
either positive or zero. With the inclusion of a slack variable the
conditions become
72 — 6x — 4; = 0
40 — 4x — s = 0
given the requirements that
s > 0 and j > 0
The conditions are two equations in three unknowns, but an infinity of

solutions does not follow because of an important relationship between
j and s at any optimum: at least one of s and j is zero, j is the value (the
effect on the objective) of extra resource (increased b) at the margin.
But such effect must be zero if not all the resource was used (i.e. if
s > 0). On the other hand, if at the optimum j > 0, then it must be the
case that s = 0. This is because, if s was not zero and some resource
was unused, with j > 0 it must be possible to increase F by using more
of the resource, in which case the original position could not be a
maximum. Both s and j could be zero if the optimum was such that the
turning point of F occurred at the boundary. In this event it is only a
relaxation of the constraint (an increase of resource) that is correctly
valued by j. Any tightening of the constraint (decrease of resource)
would reduce F. Care is needed in interpreting multipliers in such
degenerate situations. It is now required to find a solution to
72 — 6x — 4j = 0
b — 4x — s = 0
s; = 0
s > 0, j > 0
in which b is the value of the constraint RHS. Note that the product
term sj makes this a non-linear system, and the sign requirements on s
and i mean that linear inequalities as well as simultaneous equations
are involved. In a problem of this size a solution procedure is as follows:
1 Set up the Lagrangian, L.

2 Obtain the derivatives of L.
3 Set out the conditions including slack variable and sign
requirements.
4 Try s = 0. If feasible values of x and j emerge, this is a possible
constrained optimum; if not, the position is ruled out.
5 Try j = 0. If feasible values of x and s result record the position.
6 If more than one possibility has been recorded evaluate the
objective function at the x values in the potential optima.
For an exhaustive process to be practicable, the list of possible optima

must not be too long. Let us apply this enumerative approach to the
present problem for three different values of the RHS:
1 b = 40
2 b = 56
3 b = 48
In case 1, with b = 40, first try s = 0. This means that
40 — 4x — 0 = 0
so that X = 10. Inserting x = 10 in the derivative condition produces
72-6(10)-4; = 0
so ;■ = 3 and all conditions are satisfied, so the position is a possible

optimum. The value of the objective function here is F = 1020. Now
try j = 0. The x derivative condition is then
72 - 6x - 4(0) = 0
which solves for x = 12 and on substitution into dL/dj produces
40-4(12)-s = 0

which solves for s = — 8. As this is not a feasible outcome, at b = 40
there is only one possibility for the optimal position:
X = 10, j = 3, s = 0 and F= 1020
which must therefore be the optimum.

In case 2, with b = 56, setting s = 0 in the resource constraint
requires that
56 — 4x — 0 = 0
so X = 14. Putting this value of x into the dLjdx equation results in
72 - 6(14) - 4j = 0
which solves for ; = — 3 and rules out this position as a possible

optimum. Next, trying j = 0 gives
72 - 6x - 4(0) = 0
which solves for x = 12. Putting this x value into the second constraint
produces
56-4(12)-s = 0
so s = 8. This is the only combination of values satisfying all

requirements and so must be the optimum which, in full, is where
X = 12,; = 0, s = 8 and F = 1032
In case 3, with h = 48, trying s = 0,
48 — 4x — 0 = 0
so s = 12, which when put into the x derivative condition produces
72 - 6(12) - 4; = 0
so ; = 0. This is a permissible outcome and produces the one candidate

for optimality:
s = 0, j = 0, X = 12 and F = 1032
The reason that both s and j are zero here is that with the RHS set at
48 there is just sufficient resource to reach the unconstrained optimum.
Note that, although resource increases are of no value, any reduction
in resource would cut into F.
The procedure is summarized as follows. Given the single-variable
problem
Maximize z(x)
subject to ax + c < b
in which a, b and c are real valued constants, the Lagrangian is
L(x) = z(x) + i(b — ax — c)

and conditions at the first order can be expressed as
Zx — aj = 0
b — ax — c — s = 0
sj = 0
s >0, j >0
Solutions to these necessary conditions for a local constrained

maximum are then sought, using steps 1-6 or other means. We now
extend this approach to two decision variables and we will broaden the
form of constraint into one which does not translate directly into upper
bounds. Consider the following problem:
Maximize F = 80x — 2.5x^ + 40^ — 4^^ + 2xy

subject to 4:X + 2y < 80
The Lagrangian is
L = 80x — 2.5x^ + 40y — 4i/^ + 2xy + ;(80 — 4x — 2y)
and the first-order conditions are
dJ-j
(i) — = 80 - 5x-I-2i/- 4; = 0
ax
(ii) — = 40 — 8i/ -f 2x — 2; = 0
ay
dTj
(iii) — = 80 — 4x — 2z/ > 0
d]
In (iii) a slack variable can be added to give the requirement as an

equality:
80 — 4x — 2i/ — s = 0
To these three conditions should be added
sj = 0
and the requirements that
s > 0, i > 0
Trying s = 0, it emerges from (iii) that
y = 40 — 2x
Substitution of this expression for y into condition (i) produces
80 - 5x + 2(40 - 2x) - 4j = 0
i.e.
160 — 9x — 4j = 0
while substitution for y into (ii) results in
40 - 8(40 - 2x) + 2x-2j = 0

which in turn simplifies to
-280 + 18;c-2; = 0
The resulting two simultaneous equations in x and j are
18x- 2; = 280
9x + 4; = 160
which solve for
X = 16 and j = 4
with the consequence that y = 8, and from substitution of x and y

values into the objective function, F = 960. This is a possible optimum.
Now try setting the Lagrange multiplier j to zero. Conditions (i) and
(ii) then require that
5x — 2y = 80
—2x + 8i/ = 40
which solve for x = 20 and y = 10. Substituting these values into the
constraint in equality form produces
80-4(20)-2(10)-s = 0
so s = —20, indicating violation of the original constraint and the fact

that the position is not feasible (the x and y values associated with
j = 0 give the unconstrained optimum and s is the shortfall from the
minimum RHS that will achieve this position). Thus the optimum is
X = 16, = 8, j = 4 and F = 960
We have confined discussion to the case of one linear constraint in one

or two variables with a ‘friendly’ quadratic objective function. The
conditions can be broadened to more general cases with several
constraints, more variables and a wider range of objective functions and
constraints. But in more complex cases it is one thing to describe and
interpret the solution in terms of the conditions that apply, and quite
another to obtain the values of variables in the solution.
Exercises 7.5
1 Find the maximum value taken by the function
z= + 7.5x2 _ 150x + 1000

subject to 5x + 40 < 100.
F(x, y] = 40x — 8x2 ^ ^ QQy _ ^y2
subject to 2x + 4y<40.

f(x, y) = 120x - 2.5x^ + 3xy - 60y - 3.5y^

subject to 3x + 2y < 35.
f(x, y) = 1 OOx - 3x^ + 4xy - 55y - 2.5y^

subject to 2x + 5y < 90.
7.6 The ideas of previous sections are now brought together. We will look
Inequality at the maximization of a quadratic function of two variables subject to
constraints and a linear constraint and sign restrictions. A second linear constraint will
sign requirements be added later and a solution procedures will be discussed.
First consider the following problem:
Maximize F = 100 + 8x + IGy —

subject to X + 2y < 15
X > 0, y > 0
The Lagrangian is formed in the usual way
L = 100 + 8x + 16y — x^ — y^ + j{15 — X — 2y)
and the first-order conditions can be written as
Lx = 8 — 2x — i < 0
Ly = 16 — 2y — 2j < 0
Lj = 15 — X — 2y > 0
xLx — 0, yl^y ~ 0, ]Lj — 0
X > 0, y > 0, j > 0
The inequality form for the derivatives with respect to x and y follows
from the sign restrictions on the variables. The derivative with respect
to the Lagrange multiplier restates the constraint and so is also an
inequality. The zero value for the products of the variables and their
corresponding derivatives follows from the sign restrictions. Finally, the
sign requirements themselves are restated. For solution purposes, the
addition of slack variables makes a systematic search for solutions
somewhat easier. With sign-restricted slack variables t and u
introduced, the conditions can be written as
8 — 2x—;-l-t = 0
16 — 2y — 2] + u = 0
15 — X — 2y — s = 0
tx = 0, uy = 0, s; = 0
X > 0, y > 0, j > 0, t > 0, u > 0, s > 0

The requirement that the products of variables and corresponding
slacks should be zero has the same implications as the derivative-times-
variable form - in each case at least one member of the pairing must
be zero. These are complementary slackness conditions.
In small problems the combinations of variables to be set at zero can
all be considered. In the present case, with three pairs of linked
variables, there are 2^ = 8 possibilities. But in a problem with n
variables and m constraints, the number of combinations under
complete enumeration is 2"'“'“", which soon becomes impractical. Table
7.5 shows the values of non-zero variables and the objective function
value for each set of zero variables.
Table 7.5 solution values (non-zero) Objective Comment

zero group
X t y u s ; F
(a) t u s 3 6 2 175
(b) t u i 4 8 -5 180 Infeasible
(c) t y s 15 -60 22 -5 Signs
(d) tyj 4 -16 11 116 Signs
(e) XUS -7.5 7.5 0.5 163.75 Signs
(f) X u j -8 8 -1 164 Infeasible
(g) X y s ♦ * *
100 No solution
(h) xyj -8 -16 15 100 Signs
Cases (b), (c), (d), (e), (f) and (h) all violate sign requirements: in fact
(b) and (f) are infeasible in terms of the resource constraint. Case (g)
requires 15 to equal 0. Case (b) gives the unconstrained optimum since
; = 0, so the constraint RHS does not restrict achievement of objective.
Moreover, since t and u are also zero, the derivatives with respect to x
and y are the derivatives of F, and their satisfaction as equalities is
necessary for the unconstrained optimum.
The only feasible solution here is the constrained optimum, case (a).
The Lagrange multiplier j is positive (s = 0 so the resource constraint is
satisfied as an equality) and gives the instantaneous rate of change of
the maximum of F as the constraint is relaxed. It is revealing to match
the cases of Table 7.5 with a graph showing the kind of point to which
the case corresponds. Figure 7.14 shows the location of the points as
each would be in a problem in which they were feasible. Also shown
are the contours that would result in the point being optimal (only one
such set of contours applies in any particular problem).
Case (a) at the tangency point marked as T1 shows the optimum in
the current problem. Case (b) shows how an interior optimum
(corresponding to an unconstrained maximum) would appear. As
shown by the negative value of s, in the current problem (b) lies outside
the feasible region above and to the right of Tl. Case (c) is a corner-
point solution since the resource constraint and the sign requirement
on y are satisfied as equations. So the corresponding straight lines
intersect and result in a corner of the feasible region. Corner-point
297
Figure 7.14
optima can be seen as ‘frustrated tangency’ - the nearest feasible

position to a point of tangency which cannot be reached due to sign
restrictions.
Case (d) gives a point such as T3 where the contours are tangential
to an axis. At T3 the axis is the x axis (with equation y = 0). Case (e)
is corner A of the feasible region. In the current problem the
negativity of t means that this point cannot be the constrained
optimum. Case (f) corresponds to an axis tangency point such as T2.
In the current problem (not shown as such in the diagram) this
tangency position lies above the corner A and so is infeasible in terms
of the constraint - as evidenced by s = — 1. Case (h) is the origin. This
is also a corner point which could be optimal, but not in the current
problem. Case (g), while feasible in terms of x and y, is not a solution
to the conditions.
The information in Table 7.5 can be mobilized in another way. All
cases in which t = 0 are those for which x is not sign restricted (if
t = 0, Lx = 0, as required for non-sign-restricted variables). In all
solutions where t = 0 ((a)-(d) inclusive) x is positive. We can therefore
conclude that a sign restriction on x has no effect in this problem.
Cases where u = 0 show the effects of removing the sign restriction on
y - these are (a), (b), (e) and (f). Again (in this problem) the sign
requirement on y makes no difference as y is positive in each case.
Cases where ; = 0 correspond to not using the constraint. So if we also
want the best sign-restricted solution unconstrained by resources Table
7.5 should be inspected for the best feasible solution from the four
combinations with j = 0 ((b), (d), (f) and (h)). The unconstrained
optimum (b) is achievable in this case. The difference in the value of
the objective function between (b) and (a) is 180 — 175 = 5. This is
attributable to the constraint, and represents the maximum price worth
paying for its removal as an effective restriction.

Exercises 7.6
1 Given
f{x, y) = 32x - + 16y - + 150
(i) Find the maximum value of the function subject to

2x + y < 30 and the sign requirements x > 0, y > 0.
(ii) What improvement v/ould be possible if the constraint was
removed?
2 For
f(x, y) = 63x - 0.8x^ - 35y - 0.4y^ + xy
(i) Find the maximum value of the function subject to

X + 0.8y < 35 and the sign requirements x > 0, y > 0.
(ii) What improvement in f(x, y) is possible if the constraint
was removed? What is the 'cost' of the constraint to the
decision-maker?
(iii) With the constraint retained, what improvement in the
function f{x, y) would be possible if the sign requirement
on X was removed?
(iv) With the constraint and the sign requirement on x still in
place, what improvement in f{x, y] could be gained if it
was possible to remove the sign requirement on y?
7.7 Objective functions are not always quadratic and constraints will not
The Kuhn-Tucker always be linear. But if the objective function and the left-hand side
conditions and (LHS) of the constraints are differentiable and certain regularity
mathematical conditions are satisfied, then the conditions developed below will apply.
programming The constraints should not produce singularities such as an outward¬
pointing cusp at the optimum. Constraint qualification conditions
could be introduced to provide against such situations. We will present
optimality conditions in a two-variable two-constraint model, but the
conditions generalize to larger problems. Examples will be confined to
quadratic objective functions and linear constraints.
The two-variable two-constraint problem can be expressed as
Maximize F = F(xi, X2)

subject to gi{xi, xj) <bi ^
g2(xi,X2)<b2
xi > 0, X2 > 0
(7.13) is the non-linear programming model. Necessary conditions

are obtained by forming the Lagrangian
L = F(Xi,X2)+jllbl -gi(Xi,X2)] +}2[b2 - g2(Xl,X2)]
299
At the first order
dL dF . dgi . %2 . ^
—72 — < 0
dxi dxi dxi dxi
dL dF , dgi . dgj
r\ 7l r\ 72 r\ <0
dX2 dX2 dX2 dX2
dL dL ^
Xi — +X2^ = 0
uXi UXj
Xi > 0, X2 > 0
(7.14)
dJj
^ = h -giixi, X2) > 0
571
bi -giixi, X2) > 0

dji
. dL dL
7i ^ + 72 — 0
d]i d]2
7i > 0* 72 > 0
These conditions characterize local maxima of F subject to the

constraints and sign requirements, and are the Kuhn-Tucker or
Karush-Kuhn-Tucker conditions, derived independently by Karush
in 1939 and Kuhn and Tucker in 1951. Excepting constraints
producing extraordinary circumstances at end-points, the Kuhn-Tucker
conditions are necessary for a local maximum, but are not in general
sufficient. However, if a solution procedure identifies all points satisfying
the Kuhn-Tucker conditions, then, at least in principle, the global
maximum can be found by running the points through the objective
function. Finding any solutions to such conditions can be a major task.
There may well be a large, possibly even infinite, number of solutions,
in which case enumeration is out. But the conditions can check the
credentials of a claimed optimum. Limited usefulness at high generality
does not condemn an approach - other ways of seeking exact, global
optima may also be difficult. Simpler circumstances can often be
assumed. For example, if:
1 the objective function is concave or

2 the constraint functions g are convex,
then the Kuhn-Tucker conditions are both necessary and sufficient for a
global optimum. This means that ‘all’ that is required is a feasible
solution to the system (7.14). Fortunately, in economic and business
models it is often acceptable to assume 1 and 2 above. Such problems
are called convex programming problems.
The Kuhn-Tucker conditions are necessary for a maximum of the
original function F subject to the constraints, but this does not
correspond to a maximum of the Lagrangian function in all

directions. The constrained optimum of F gives a saddle point of
the Lagrangian — a maximum in the x, directions and a minimum
with respect to the multipliers. This is shown by the required signs
for the partial derivatives. Those with respect to the x variables must
be non-positive, while the derivatives with respect to the multipliers
must be non-negative. Let us see how the Kuhn—Tucker conditions
turn out in some simple cases. Earlier, we considered single-variable
problems with an upper and lower (zero) bound. We can now set
out the required conditions as a special case. Consider the following
problem:
Maximize z{x)
subject to X > 0 and x < b
The Lagrangian is
L = z(x) +i(b - x)
and conditions are
dz . ^
L,=^-j<0
ax
xLx = 0
X > 0
Lj = b-x>0
jLj = 0
j>0
Note that the product of x and Lx is required to be zero, and not x and
dz/dx.
Consider a numerical example:
Maximize F = x^ — 10x-|-50
subject to X < 8
X > 0
The Lagrangian and Kuhn-Tucker conditions are
L = x^ - 10x+ 50+;(8 -x)

L;c = 2x-10-j<0
xL^ = 0
X > 0
Lj = 8-x>0
jLj = 0
;>o
301
With slack variables introduced in complementary slackness form it is
required that
2x — 10 —; + t = 0
:>jt = 0
X > 0, t > 0
8 — X —s= 0
js = 0
; > 0, s > 0
The enumeration results are shown in Table 7.6. The constrained

maximum is at the x = 0 end-point. The interior turning point (b) is a
minimum, the objective function in this case being convex. The problem
is graphed in Figure 7.15.
Table 7.6
zero group solution values (non-zero) objective
00
SO
(a) t s F= 34
II
II
(b) ti X= 5 s=3 F = 25
(c) X s no solution
(d) xj t = 10 s = 8 F= 50
Figure 7.15
One of the advantages of enumeration is that significant business

decisions are often taken by an executive or committee on the basis of
professional advice. Quite frequently in this process the board is
presented with a list of possibilities from which one is to be selected and
acted upon. This will not inevitably be the course of action showing the
greatest value of the objective function (although this option should be
on the list), as wider considerations may need to be taken into account.
An enumeration approach has the advantage of providing a
comprehensive listing from which options may be drawn.
Some possibilities highlighted by enumeration which do not satisfy
the Kuhn-Tucker conditions provide better values of the objective
function than some points for which the conditions are satisfied. This

allows a value to be placed on the relaxation or removal of binding
constraints. Such points may not show up in a procedure
concentrating exclusively on points fulfilling the Kuhn-Tucker
conditions.
Now consider a more substantial problem in two decision variables,
two linear constraints and a quadratic objective function:
Maximize F = lOx — O.lx^ + 20y — O.Iy^

subject to Sx + y < 200
X + 2y < 90
X > 0, y > 0
Before solving this problem we will solve the unconstrained problem to

gain a point of reference. We will then look at the feasible region,
discuss the possible locations of the constrained optimum and set out
the Kuhn-Tucker conditions. In the unconstrained case the first-order
conditions are
F^ = 10 - 0.2x = 0
Fy = 20- OAy = 0
so
X = 50, y = 50 and therefore F = 750
Second-order conditions for a maximum are satisfied at this position.

Consider the constraints as shown in Figure 7.16.
Figure 7.16
Possible positions of the optimum in the feasible region are the

corner points 0, A, B and C, a point of tangency between a contour of
the objective function and constraint one (shown as Tl), a point of

tangency between a contour and constraint two (shown as T2),
tangencies with the axes corresponding to T3 and T4, and an interior
optimum marked as I. Not all these possibilities relate to the objective
function in the current example. The tangency position for one
constraint may be infeasible in terms of the other, and the tangencies
within the axes may lie outside the feasible region. With two
constraints, two multipliers will be needed. The Lagrangian and Kuhn-
Tucker conditions are
L = lOx - O.lx^ + 20y - 0.2y^ + j(200 - 5x-y) + k(90 -x-2y)

Lx = 10-0.2X-5;-k< 0
Ly = 20 - 0 Ay - j - 2k < 0
xLx = 0, yLy = 0
X > 0, y > 0
Lj = 200- 5x-y>0
U = 90-x-2y>0
jLj = 0, kLk = 0
}>0,k>0
In complementary slackness form, the conditions produce the following

equations, of which the optimum will be a feasible solution:
10 — 0.2x — 5] — k + t = 0
20 - OAy - j - 2k + u = 0
200 — 5x — i/ — s = 0
90-x-2y-r = 0
tx = 0, uy = 0, sj = 0, rk = 0
X, t, u, y, s,;, r, /c > 0
Enumeration can be organized around the 2^ = 16 sets of values of the

zero variables using a table. The optimal solution that emerges is
X = 30, y = 30, k = 4, s = 20
t = 0, u= 0, r = 0, j = 0
These values produce a constrained optimal value of the objective

function of F = 630. The location of the optimum, a tangency position
in this problem, is shown in Figure 7.17.
Tangency is with constraint two and is below constraint one
(corresponding to j = 0). So, even though the amount of resource one
is less than the amount needed to reach the unconstrained optimum,
further units of resource one are not worth having without a significant
increase in resource two. This is important information and could not
occur in linear programming. Incidentally, the (corner) solution at the
intersection of the resource constraints corresponding to full utilization
of both resources would give F = 627.04.

Figure 7.17
Consider a further example:
Maximize f{x, y) = 120x — x^ + 200y — 2y^

subject to X + 2i/ < 130
2x + y < 152
X > 0, y > 0
The Lagrangian is
L = 120x - + 200y - 2y^ +;(130 - x - 2y) + k{152 -2x-y)
for which first-order conditions in complementary slackness form are
120 - 2x - j - 2k + t = 0
200 -4:y-2j-k + u = 0
130 — X — 2y — s = 0
152-2x-i/-r = 0
tx + + }S + kr = 0
X, y, t, u, j, k, r, s > 0
The sixteen combinations of zero-valued variables and resulting values

of the other variables are shown in Table 7.7.
The constrained optimum (and the only complete solution to the
conditions) is case (b), which results in f(x, y) = 8300. The optimal
position is at tangency between the objective function contours and the
first constraint. The sign-infeasible ‘solutions’ of Table 7.7 contain three
noteworthy cases. Case (c) is tangency with the second constraint. This
gives an infeasible resource-one requirement as shown by the negative s.
Case (a) represents full use of both resources - the constraints are
satisfied as strict equalities. Case (d) is the unconstrained optimum. The
problem is graphed in Figure 7.18, with the optimum shown as point T.

Table 7.7 solution ]>alues (non zero)
zero group
X t y u j s k r
(a) t u s r 58 36 36 -16
(b) tusk 50 40 20 12
(c) t u j r 52 48 -18 8
(d) t uj k 60 50 -30 -18
(e) tysr no solution
(0 ty s k 130 -480 -140 -108
(g) tyjr 76 -216 54 -16
(h) tyjk 60 -200 70 32
(i) X u s r no solution
(j) X u s k -150 65 -30 87
(k) X uj r -1036 152 -174 -408
(1) X uj k -120 50 30 102
(m) X y s r no solution
(n) X y s k no solution
(o) X yj r no solution
(P) X yj k -120 -200 130 152
Figure 7.18
Enumeration nears a practical limit for two sign-restricted variables

and two linear constraints. For larger problems in linear constraints, if
a quadratic objective function is retained, a solution can be found using
quadratic programming (QP). In QP the first-order conditions in
complementary slackness form and the sign requirements make up the
constraints. A means can be found to provide for complementary
slackness {xt = 0, i/u = 0, etc.) and a modified form of objective is used
solely to find a solution to the Kuhn-Tucker conditions. This can be
done on the basis of the simplex method.
For problems involving more general forms of objective function or
constraints, methods such as the gradient search procedure can be

used. But, as with linear programming, non-linear programming
problems do not have to be solved manually. Personal computer software
enables problems of significant size to be solved. For example, the
accompanying software package to Hillier and Lieberman (1995) can
solve problems with up to six linear constraints, three decision variables
and a polynomial objective function in which the exponent for each
variable in each term is a non-negative integer up to 9. The general non¬
linear programming problem is an n-variable m-constraint version of the
model (7.13). While the Kuhn-Tucker conditions characterize exact
solutions to these problems, there is no general algorithm that will find
the global optimum in all problems. Algorithms do exist for particular
cases (including linear programming and quadratic programming) and
there are procedures which give good approximate solutions.
Exercises 7.7
1 For the following problem:
Maximize f{x, y) = 40x — + 60y — 2y^

subject to 5x -f 4y < 120
1 Ox + y < 140
X > 0, y > 0
(i) Set out the Kuhn-Tucker conditions in full.

(ii) Confirm that the set of values
88 _100 ._192 . 632

/ “4^' ^“490
satisfies the conditions. Verify that this position, which is

the optimum, is a corner point of the feasible region.
(iii) What increase in the value of f{x, y) would be possible if
the constraints were completely removed?
2 Set out the Kuhn-Tucker conditions for the following problem:
Maximize f(x, y) = 12 OOOx — 0.2x‘^ + 2 000 OOOy — y®

subject to 3x-|-7y<21
2x2 + y < 20
X > 0, y > 0
3 For the following problem:
Maximize f(x, y) = 1 OOx — 2x2 ^ _ y2

subject to 2x + y < 65
0.5x -F y < 38
X > 0, y > 0
(i) Give the Kuhn-Tucker conditions in complementary

slackness form.

(ii) Find the constrained optimum.
(iii) What is the unconstrained optimum? What, therefore,
could be said to be the cost of the constraints as a whole?
7.8 A model in which an extreme value of a quadratic function is sought

Economic subject to linear restrictions on decision variables is that of the multi¬
application: multi¬ product monopolist. Suppose that a firm makes two products, for
product monopoly which demand conditions are given by
p{x) = 7S - X, p{y) = 74: - 2y (7.15)
where x and y are the quantities produced, and p{x) and p{y) are the
prices. The products have independent demand functions, as shown
in (7.15) (no x term in the p{y) equation, no y term in p{x}). We will
assume a linear technology, although a quadratic programming (QP)
approach requires only the weaker condition that the products are
technically independent {y does not appear in the production
function for x and vice versa) and give quadratic or linear cost
functions. The link between production levels is through common
consumption of scarce resources.
Suppose two resources have limited supply. Table 7.8 gives
availabilities, prices and per unit requirements.
Table 7.8 Availability Requirements Price per unit

PC y
resource one 120 5 4 £3
resource two 140 10 1 £2
From the data in the table the unit cost of production for each
product can be found. Each unit of x requires 5 units of resource one
@ £3 per unit and 10 units of resource two @ £2 per unit. Thus
unit cost of % = 5(3) + 10(2) = 35

unit cost of y = 4(3) + 1(2) = 14
So for any x and y the production costs for the two products are
C(x) = 35x and Ciy) = 14y
The demand conditions imply that revenues from the products, R{x)
and R{y), will be
R(x) = px = 75x — x^
R(y) =py= 74y - 2y^
Unit profits from the two products, 7r(x) and n{y), will be
Tx{x) = R(x) — C(x) = 40x — x^
Tîy) = R(y) - C{y) = eoy - 2y^

The firm wishes to select values of x and y to maximize overall profit
from the two products, F, where
F = 7r(:>:) + 7r(i/)
Negative product is meaningless and resource limits mean that the

firm’s problem is to set x and y so as to
Maximize F = 40x — x^ + 60y — 2y^

subject to Sx + Ay < 120
lOx A-y < 140
X > 0, y > 0
The Lagrangian is
L = AOx — x^ + 60y — 2y^ + j(120 — 5x — Ay) + fc(140 — lOx — y)
and the Kuhn-Tucker conditions in complementary slackness form are
40 — 23c — 5; — 10k + t = 0
60 — 4y — 4; — k + u = 0
120 — 5x — 4i^ — s = 0
140 — lOx — y — r = 0
tx = 0, uy = 0, sj = 0, rk = 0
X, t, u, y, s, j, r, k > 0
The following values meet the Kuhn-Tucker conditions and so give the
constrained optimum
88 100
— ^ 12.57, y = 14.29
7
192 632
-Ri 0.39, k = « 1.29
490 490
t = 0, u = 0, s = 0, r = 0
Each resource is fully used and the optimal solution is the extreme
point given by the two resource constraints as equalities. The optimal
value of the objective function is
38 896
F= 793.80
49
For comparative purposes, the unconstrained maximum occurs when
x = 20, ^ = 15 and F = 850
The cost imposed by the constraints as a whole is therefore
850 - 793.8 = 56.2
The problem is graphed in Figure 7.19. The constrained optimum is P

and the unconstrained optimum is U.

Figure 7.19
As the objective function is strictly concave, the values of the

multipliers,; and k, are upper bounds on the increases in the objective
function following unit increases in the resources. So they can be used
to rule out proposals to increase resources. For example, if there was an
opportunity to take an extra 5 units of the second resource for £17
(that is, at a cost of £7 over and above the normal cost of £10), this
would not be worthwhile, since the increase in profit if the resources
were obtained at the original prices is less than the cost premium for
the additional units. That is.
632
5 = 6.45 < 7
4^
But a proposal to secure the extra resources for £15.50 (a total
premium of £5.50) could neither be rejected nor be accepted on the
basis of the value of k. The actual changes in F following unit increases
in one resource alone are as follows:
279
resource one —>^121 AF = 0.2278
1225
1514
resource two —>141 AF = 1.2359
1225
At the optimum each multiplier is an increasing function of the level

of the other resource. This cannot occur in linear programming. It
can also be shown that as the result of the equality of second-order
cross-partial derivatives each multiplier is affected in the same
(linear) way by the level of the other resource. If either resource is
increased by one unit, the multiplier for the other resource increases
by
83.2
0.17
4^

Exerci^ 7.8
1 A two-product firm faces demand conditions given by
p(x) = 95 - 0.5x, p(y) = 85 - 0.5/
in which x and / are the quantities produced and sold at

prices p(x) and p(/). The following resource constraints must
also be observed:
X + / < 100 (resource one)

x + 2y < 140 (resource two)
Resources costs are £5 and £10 per unit, respectively.

Negative production levels are meaningless. The firm's
objective is profit maximization.
(i) Set out a constrained optimization model of the problem.
(ii) Find the optimal production of each good and the
maximum profit.

CHART R
8 Integration
This chapter considers integration - which can be seen as

differentiation in reverse - the process by which a function with a
given derivative may be found. We present the concept of an
antiderivative, give fundamental rules for finding indefinite and definite
integrals, and introduce integration by substitution. We apply the
methods to cost and revenue functions in the theory of the firm.
Differential equations and numerical methods are introduced.
LEARNING OUTCOMES
By the end of the chapter you will be able to find the indefinite integral
for polynomial, selected rational and other functions. You will know
how to evaluate the definite integral in these cases, how to find areas
under curves and you will be able to solve some simpler types of
differential equation.
8.1 In differentiation we used a suitable rule to find the derivative of a

Introduction given function, but this function could itself be a derivative. At times it
would be useful to recover the function giving the derivative. To do this,
we will need the derivative and the value of the function itself for given
X. For example, knowledge of the marginal revenue function and the
fact that the total revenue function goes through the origin will allow
us not only to reconstruct the total revenue function but also to
uncover the product demand curve. The process by which an original
function is recovered from its derivative is integration. In addition,
integration can be seen as the process of finding areas under the curve
defined by the derivative - the evaluation of the definite integral. We
begin with the idea of recovering a function given a derivative - for
example 3x^. This is not too difficult a task, as we have frequently come
across terms such as this in differentiation. The solution might appear
to be x^, but further thought reveals other possible functions with
as their derivative. For instance, + 1 would do just as well and so
also would
x^ + 74, x^ - 20, x^ + TT, - 2\/2 (8.1)

In fact, x^, + 1 and all the functions in (8.1) are antiderivatives of
3x^. But an expression such as + x is not an antiderivative of 3x^,
all of which must have the form x^ + k, where fc is a constant - in
graphical terms differing only by the location of the intercept. The
antiderivatives are graphed in Figure 8.1 and it is evident that at any
given value of x, say x° all antiderivatives have the same slope.
Figure 8.1
If in an application it is known that the antiderivative takes a specific

value for given x, this information can be used to identify the relevant
value of k. For example, if the antiderivative of 3x^ was known to take
the value 950 when x was 10, then k must be —50 and the original
relationship must therefore be
y= — 50
The expression
x^ +k
which covers all antiderivatives (and is sometimes called the general

antiderivative (of 3x^)), is the indefinite integral of/(x) = 3x^. The
indefinite integral uses the following notation:
y 3x^dx = x^ + k (8.2)
The integral sign
■ /
was chosen by Leibniz as the symbol for integration. It is an archaic ‘S’,
the initial letter of ‘summa’ and reflecting the area under the curve
aspect of integration (the area can be found as the limit of a process of
summation). In (8.2) the symbol ‘dx’ means ‘with respect to x’ (as in
integration 313
differentiation), so the whole left-hand side of (8.2) reads ‘The integral
of 3x^ with respect to x’. The expression to be integrated, here Sx^, is
called the integrand. Using Leibniz’s notation (for both the derivative
and the integral), the steps can be summarized as follows. Starting with
and interpreting dy and dx as distinct entities write
dy = 3x^dx
from which the original function y is recovered by taking the integral of

both sides:
y — J <iy = J 3x^dx = +k
Where no specific function is stated, the integrand is usually written as

/(x) and the indefinite integral as F(x) T- k, as in
J/(x)dx = F(x) -t- k
Construction is usually more difficult than separation. So it is no

surprise that integration is harder than differentiation, but many of the
rules for differentiation have their counterparts in integration.
Exercises 8.1
1 Which of the following ore antiderivatives of 4x^?
(i) x^+100 (ii) x^-TT (iii) x^-O.Olx
|iv|(x2-1)" |v| |x2-1)(x2 + 1|
2 Which of the following are antiderivatives of x^ — x?
,i| o.2x^-0 5x^ |ii|
(iii) 0.4x^ — x^ (iv) 0.2x^ — 0.5x^ — IOOtt
8.2 Here we introduce four useful rules for integration.

Rules for
integration
8.2.1 This rule applies when the integrand is a constant. The constant
The constant function function rule is
rule
J cdx = cx + k (8.3)
where the constant c is any real number. As an example, consider the

integral of the number 7 with respect to x:
J 7dx = 7x + k
This rule is a special case of the more general power rule.

8 2.2
. ^ ^ The power rule applies when the integrand is a variable raised to a
The power rule given power, i.e.
/ x”dx
n+ 1
+k (8.4)
where n is any real number except —I. The rule can be confirmed by
differentiation. We can get a long way with this straightforward rule, so
we now give several examples. Consider the integral of the cubing
function. Here, n = 3 and the power rule gives
(i) j x^dx =
This illustrates the power rule when u is a positive integer. The rule
also applies if n is any negative integer other than —1. For example,
f X~^
(ii) / x“^dx = —
In this example the power rule is applied directly, but note that what
we have shown is that
and it is easier to handle a negative exponent than a reciprocal.

The questions arise as to why n = — 1 is excluded and what the
integral of is. An attempt to apply (8.4) to this case would result in
division by zero, the resulting ratio being x°/0. Here we merely point
out that x“^ (i.e. 1/x) does have an integral, but the integral is not a
power function (we will cover it in the next chapter).
Now consider the use of the power rule with a rational exponent.
Letting n = j,
(iii) J x^dx = + k
Again the rule is applied directly, inserting ^ for n. Note that this
example shows that
/ -y/x dx
3
+k
It is far easier to work out the integral with a rational exponent than to
try to operate with radical signs. The power rule applies to irrational
exponents. Suppose n = \/2:
r /
(iv) / x’^^dx = -h k
J V2 + 1
The power rule is a simple, important and frequently used rule for
finding indefinite integrals.
Integration 315
8.2.3 This rule is equivalent in integration to the sum-difference rule for
The sum-difference rule differentiation, and is just as easily applied. The sum-difference rule for
integration can be stated as
J [fix) ± g{x)]dx = J f{x)dx± j ^(xjdx (8.5)
The rule reads as follows: the integral of the sum of, or difference
between, functions is the sum of (or difference between) the integrals.
As an example, consider
/(4.3 + 5x^)d. = /4xM.+ /5/dx
= x'^ + ki + x^ + k2
= x'^ + x^ + k
The power rule is applied to individual terms and only one constant
of integration is needed, as the sum of or difference between two
arbitrary constants {ki and kj) is an arbitrary constant {k). The sum-
difference rule greatly expands the range of the power and constant
function rules, and applies regardless of the number of terms. An
important consequence of the rule is that polynomials can be integrated
term by term. For example.
/(5x‘-2x’ + 9x^-14x+13)dx
= x^ — 0.5x^ + 3x^ — 7x^ + I3x + k
A multi-term integrand needn’t be a polynomial: powers of x which are

not positive integers may be involved. For example,
[ (3x0-5 _ _ 2x-3)dx = 2x^-5 - TTX^ + x-^ + k
8.2.4 This rule states that the integral of a constant multiple of a function is
The constant multiple that multiple of the integral of the function. That is.
rule
J cf(x)dx (8.6)
The usefulness of the rule is that a constant may be taken outside

the integral sign. As an example, consider
/(45x^ + 60x3)dx
The number 15 (the highest common factor of 45 and 60) can be

taken outside the integral sign, and you are therefore required to find
15 J (3x^ -I- 4x5)dx = 15(x5 -f .x'^) + k= 15x5 + 15x^ -|- k
and there is no need to multiply the constant of integration by 15.

1 Find the following indefinite integrals:
(i) /3dx
(ii) /-7.5dx
(iii) f \/3dx
(iv) /rrdx
(i) f x'^dx
|ii| /x-’dx
|iii| j ^dx
(iv) f x^/'^dx
(v) f x°^dx
(vi) /x-o^dx
Iviil f±dx
(viii) Jx'^^dx
(ix) f x^^dx
3 Integrate the following expressions with respect to x:
(i) 4x^ + 5x^
(ii) 6x^ - 2.5x'^ + x^
(iii) 20x^ - 3x“'^ + 4x“^
(iv) 5xi^+x-’i
(v) ■K^X^~^ + {27T + ])x'^'^ + k
4 Find the following:
(i) J12x^dx
(ii) /(9250x^ - 125x9)dx
(i) / (ox^ + bx + c)dx
(ii) f mx^dx
(iii) /[(n + l)x" - (m - l)x'”“^]dx
(iv) /(3f2 + 2t)df
(v) -3 f x~‘^dx
8.3 Marginal cost, MC, is the rate of change of total costs as output is
Application to the varied, so MC(x) is the first derivative of cost with respect to output
marginal analysis level X. Either total variable costs, VC(x), or the total including fixed
of the firm costs, TC(x), can be used. The relationship is
dTC(x)
^ = MCW
dx
Using total variable costs, the indefinite integral of the expression for
marginal cost will be total variable costs plus the constant of integration:
I MC(x)dx = VC(x) + k
Integration 317
A value can be chosen for the constant of integration to correspond to
the level of fixed costs. Suppose that a firm finds that its marginal costs
are given by
MC(x) = 4x + 20
from which it follows that
(8.7)
Now use one observation for total cost to find the value of k relevant
to the firm. If at an output level of x = 100 total costs are £23 000,
then k must take a value such that
2(100)^ + 20(100) + k = 23 000
So
k= 1000
and costs in full can be written as
Total costs = 2x^ + 20x + 1000
Similar analysis can be carried out on the revenue side. Marginal

revenue, MR, is the rate of change of total sales revenue as output level
is varied. So MR(x) is the first derivative of total revenue, R(x), with
respect to output, x. That is,
So total revenue is the integral of marginal revenue with respect to

output:
J MR(x)dx = R(x) + k
However, as there are no ‘fixed revenues’ the value of k in this context

is zero. If in a particular case
MR(x) = 100 - 6x
then total revenue is given by
= lOOx- 3x^
Since total revenue is the product of price, p, and volume of sales, x, the
equation of the demand curve for the product can be deduced:
R = px = lOOx — 3x^

so that
p = — = 100 — 3x (8.8)
Integration has allowed the demand curve to be found from

observations on the revenue consequences of output variations. For a
further example, suppose that cost and revenue information was
available for a firm as functions of output level, x, as follows:
Marginal cost = 140 + 2x

Marginal revenue = 320 — 4x
Total cost = £2000 when x = 10
and that the profits maximizing price is required. The profits

maximizing output could be found by equating marginal revenue and
marginal cost. But to find the level of profit and price at the optimum it
is necessary to obtain the cost, revenue and profit functions in addition
to the demand-curve relationship. On the income side, total revenue,
TR(x), is given by
J MR(x)dx = J (320 —4x)dx
= 320x — 2x^
Dividing by x gives
price = 320 — 2x
In terms of costs, total costs, TC(x), are given by
= J (140 + 2x)dx
= 140x + x^ -I- k
and since TC(x) = 2000 when x = 10, the level of fixed costs, k, must
satisfy
140(10) +(10)2 + k = 2000
so that
k= 500
Thus
» TC(x) = 140x + x2 + 500
Profits, 7r(x), are then given by
7r(x) = TR(x) - TC(x)

= 320x — 2x2 — 140x — x2 — 500
= 180x- 3x2 _ 5QQ
Integration 319
Maximizing 7r(x) with respect to x:
— = 180-6x = 0
ax
which solves for x = 30. At the second order
which satisfies the condition for a maximum. The level of profit at

X = 30 is
TT = 180(30) - 3(900) - 500 = 2200
and from the demand curve the profit-maximizing level of price is
p= 320-2(30) = 260
Exercises 8.3
1 A firm's marginal revenue function is
MR(x) = 2000- lOx
What is the relationship between price and quantity in this

case?
2 With X representing output and the following information on
costs:
Marginal cost =100-l-x-|- 0.3x2

Total cost = 4000 when x = 20
find the firm's total cost function.

3 Given the following information on a firm's costs and revenues:
MC(q) = 210+ 5q
MR(q) = 480- lOq
TC(q) = 2780 when q=10
where
MC(q) = marginal cost

TC(q) = total costs
M(q) = marginal revenue
q = output level
(i) Find the product price that would produce maximum

revenue.
(ii) Find the maximum profit at the corresponding price.

8.4 In dynamic models the natural way to express relationships between
Differential variables is often through rates of change. Where change is continuous,
equations rates of change are represented by derivatives, which are included in the
equations defining the model. For example, models involving steady rates
of growth (or contraction) - perhaps of the gross domestic product (GDP)
of an economy, its total population, money supply or pollution level, may
involve derivatives. Here we consider only the simplest differential
equations. Remarks are grouped under the following heads:
• general considerations:
• first-order linear differential equations with constant coefficients;
• higher-order differential equations (special cases):
• a first-order, second-degree example.
8.4.1 An equation containing one or more derivatives of a function is called a

General considerations differential equation. Even an expression such as
is a simple example of a differential equation. The equation may appear

as above, but could also arise in implicit form as
fix) - ^
dx
= 0
So the equation
6x^ -^ + 10 = 0 (8.9)
dx
is a simple example of a differential equation in implicit form. There are

many different types of differential equation. Where the function is of a
single variable, the equation is an ordinary differential equation -
in contrast to the partial differential equations which arise from
functions of several variables. A comprehensive type classification of
differential equations would include many extra categories to those
mentioned here - for example separable, exact and homogeneous
differential equations.
In general, ordinary differential equations can be written in the
following way:
/ dx ’ dx^ ’
(8.10)
For example.
(8.11)
is an ordinary differential equation of order three, since it contains the

third derivative and no higher derivatives. The degree of the
Integration 321
differential equation is the power to which the highest-order derivative
is raised (after any rationalization to remove fractional powers). So
(8.11) is first degree despite the fact that the second derivative is
squared. The degree is given by the term involving the third derivative,
and this is raised only to the first power. By contrast,
7x^ + 2x + = 0
is of the third degree, since the highest-order derivative is cubed. A

differential equation of the first degree in which the coefficients are
either constants or functions of the independent variable is called a
linear differential equation. The derivatives in the equation may be
of any order.
dy d^y d^y
dv’ dx2’ dx^’
but note that these are not themselves raised to a power other than
one.
The equation
5x'
Ay , -
— + 7x
dx
is an example of a linear differential equation, in this case having

constant coefficients. In contrast, equation (8.11) is a non-linear
differential equation because of the squaring of the second derivative
term. A first-order differential equation, as its name implies,
contains only the first derivative. The equation
is an example of a first-order differential equation of the second degree.

The equation
2
7x^ + 3
is a second-order differential equation of the first degree. Differential

equations are used in studying dynamic processes in both the natural
and social sciences - for example radioactive decay, meteorology,
economic growth and continuous compounding of interest.
8.4.2 Equation (8.9) is an example of this type, which constitutes a special

First-order linear category for which there are simple solution procedures. The solution of
differential equations an ordinary differential equation takes the form:
with constant
y = F(x) + k
coefficients

in which there are no derivatives, and which must be consistent with
the original equation.
Consider (8.9). This is a linear first-order differential equation in
constant coefficients and can be written more conveniently as
= 6^ + 10
dx
Once this rearrangement is carried out it is clear that the solution can
be obtained by taking the integral. We can write
y = F(x) + k = J {6x^ + 10)dx:

= 2x^ + lOx -h k
and the equation
y = 2x^ + lOx -I- k
is the general solution - it includes all antiderivatives. In a practical

application there will usually be at least one observation of the value of
y for given x. For instance, in the current example if we know that
when X = 4:, y = 183
then for the equation to be satisfied it must be the case that
183 = 2(4)^+ 10(4) +k
so k= 15. We can therefore write
y = 2x^ + lOx + 15
which is the particular solution to (8.9) where k takes a specific

value as a result of a value for y linked to a value of x. Such data is
called a boundary condition, or in problems where the independent
variable is time, t, and where a value of F(t) + k is given at t = 0 the
boundary condition is called an initial condition.
Now consider some further examples. Find the particular solution to
1.5x^ — 4x^ — 8x + ^ + 10 = 0
dx
given
y = 300 when x = 4
Reorganizing the original equation,
— = 4x3 - 1.5x2 + 8x- 10

dx
so, with integration,
y = f (4x3 _ 1 5^2 4- 8x — 10)dx

and the general solution is
1/= — 0.5x^ + 4x^ — lOx + k
When X = 4, y = 300, so
300 = (4)^ - 0.5(4)^ + 4(4)^ - 10(4) + k

= 256- 32 + 64-40 + ^
so that
k=52
and the particular solution is therefore
1/ = x^ — 0.5x^ + 4x^ — lOx + 52
As a third example, find the particular solution to
16x^ + 12x^ - 40x - 2^ - 100 = 0

dx
given that
y= 8 when x = 2
rearrangement gives
^ = 8x^ + 6x^ - 20x - 50

dx
So
» = /(8.^+6.^-20x- 50)dx
= 2x^ + 2x^ — lOx^ — 50x + k
which is the general solution. Now, using the boundary condition,
8 = 2(2)^ + 2(2)^ - 10(2)2 _ 5Q(2) + k

= 32 + 16-40- 100 +k
so = 100 and the particular solution is
y = 2x^4- 2x^ - 10 x2 _ 5q^ ^ jqq
8.4.3 Here we briefly consider problems involving second derivatives. For

Higher-order example.
differential equations
(special cases) 12x + 8 = 0 (8.12)
dx^
is a second-order differential equation of the first degree. But because the

equation contains only the second derivative it is much easier to deal
with than second order differential equations including both first- and
second-order derivatives. Nevertheless, two arbitrary constants will be

included in the general solution, and to find a particular solution
observations for two values of x are needed. In general, the number of
arbitrary constants involved is the same as the order of the differential
equation. As we will see, these observations can take various forms.
Suppose that the boundary conditions are
when X = 2, y = 20
when x= 3, y = 44
Now consider the solution to (8.12). Rearrangement produces the more

convenient form
= 12x — 8
dx^
Now, to begin finding the solution we use the fact that
fd^y , du
/ —rdx = — +constant
J dx^ dx
and in the present case we have
/ (12X-8)d* = 6*^ +constant
which result we write as
^=6x2-8x + ki (8.13)
dx
In (8.13) the constant of integration is ki since a second constant is

also needed. (8.13) is not a solution of the differential equation, since it
still contains a derivative. Integration is again needed to eliminate all
derivatives and obtain the solution. The result is
y = J (6x^ - 8x + ki)dx
so that
^ = 2x^ — 4x^ + kix + ^2 (8-14)
which is the general solution. The conditions can now be used to find
the particular solution. These give two simultaneous linear equations in
ki and ka- Substitution into (8.14) when x = 3 produces
44 = 54 - 36 + 3ka + ka
and when x = 2,
20 = 16- 16 + 2ki +k2
Integration 325
Simplification and rearrangement results in the simultaneous equations
3h+k2 = 26
2ki -j-kz = 20
so, using elimination,
ki = 6
Thus kj = 8 and the particular solution is
y = 2x^ — 4:X^ + 6x + 8 (8.15)
(8.15) could have been obtained with data for just one value of x,
provided that in addition to y the value of dy/dx at this point was also
given. For example, in the present case the information
dy
y = 20 when x = 2, and — = 14
dx
would have produced (8.15). The problem of recovering the original

function given only the form of its nth derivative could be solved in a
similar fashion. The n boundary conditions might, for given x, be the
values of the original function and its derivatives down to the {n— l)th
derivative.
Consider an example when n = 3:
d^y
12x 1.5 = 0
dx^
where when x = 1,
d y
= -1.5
dx^
^ = 0.25
dx
y = 22.25
Rearrangement gives
dî/
12x- 1.5
dx^
and integration produces
—r = 6x^ — 1.5x + ki
dx^
which when x = 1 is -1.5, from which it follows that ki = -6. Thus
d^y
= 6x — 1.5x — 6
dx^

Integration gives
^ = 2x^ — 0.75x^ — 6x + k2
ax
which when x = 1 takes the value 0.25, and therefore kj = 5. Thus
~ = 2x^ -OJSx^ -6x + 5

ax
and integration produces
y = 0.5x‘^ — 0.25%^ - 3x^ + 5x + ^3
which when x=l is 22.5, so ks = 20 and the particular solution is

therefore
y = 0.5x^ — 0.25x^ — 3x^ + 5x + 20
8.4.4 We conclude with a very simple example of a non-linear differential

A first-order, second- equation. Consider
degree example
-8x^ + 112x- 392 = 0
which is an ordinary differential equation of the first order and second

degree. First the general solution. Rearrangement produces
= — 56x -I-196
and factoring the quadratic gives
I') =4(.-7)^
Taking the square root of both sides produces
^ = ±2(x- 7)
dx
and so the integrals of 2x — 14 and 14 — 2x represent general

solutions. That is,
(i) y = x^ — 14x -h k
(ii) y = 14x — x^ + k
are both general solutions. Context may suggest which of (i) and (ii)
best fits the application. For example, if y is total cost of production and
if X is output level, with typical values around 50 (say), then (i) is
appropriate. Variable costs (14x - x^) would unrealistically be given as
negative by (ii) for values of x over 14. On the assumption that (i) is
most suitable, then given the condition that
y = 2000 when x = 45
Integration 327
the constant of integration (representing fixed costs) emerges from (i) as
k = 325. So the appropriate particular solution is
y = — 14x + 325
In presenting this illustration of a higher-degree differential equation,

it must be emphasized that non-linear differential equations are usually
much more difficult to solve.
Exercises 8.4
1 Give the degree and order of the following differential
equations: 3 % 2
10x3 -3x , /d y = 0
^dx^
i) 4^-7 = 0
' dx2 dx
ii) 20x
^dx^y \^dx^
(iv) 6x7- 12^+ 100 = 0

dx
(v) 100^^+ (^) -17x^ = 0
VI ) 27x3
dV^dV 10
dy
0
dx^ ' dx^ ^ ' dx
2 Find the general solutions to the following differential
equations:
dy 8x3 ^ 9^2 ] Q
I 3;
dy
i) -2x3-4.5x27 0
dx
ii) 100-$^+! Ox’ 0

dx
(iv) 27x2 -x3 -0.5^+ 17 = 0

dx
3 Find particular solutions to the following differential equations
with the given boundary conditions:
(i) ^ = 5x^-4x3 + 3x2 +1

dx
Boundary condition: y = 300 when x = 3
dy
(ii) x3 - 1.5x2 —+ 4x 10 = 0
dx
24x3 - 90x2 + 3 - 6x - 15^0

dx

(iv) 2.5x^ + 3x3 _ 0.25 - 5x + 5 0
dx
Boundary condition: y = 1 12 when x = 2
Find particular solutions in the following cases:
(i) 10 + 2x -3 3x-2 = 0
dx
Boundary condition: y = 35 when x = 0.2
(ii) 10.5x6-0.5^-10x-6 X + 16
dx
Find particular solutions to the following differential equations:
(i) 10-^-30x = 0
dx^
Boundary conditions: y = 0 when x = 3
y = —210 when x = 5
(ii) 6x-0.25^ = 3
dx^
Given that when x = 6, y = 672 and dy/dx = 362.
Find the particular solution to the following third-order
equation:
6 0.1 dV
- 0
dx^
Given that when x 1,
dV
70
dx2
^ = 60.
dx
y = 50
7 Find the appropriate particular solution for the following

second-degree equation:
0.9x2 -0.1 -18x + 90 = 0

dx
given that (a) the variable y represents the costs of production

with output, X, typically in the range 50 to 150 units and (b)
when the output level is 100, costs in total are 20000.
8.5 Integration can sometimes be simplified by a change of variable in the

Integration by integrand. An appropriate substitution is found, the simpler expression
substitution in the new variable is integrated and the substitution is then reversed.
This is the essence of the method of substitution, which has its roots
in the power and chain rules for differentiation.
Integration 329
Consider an example. It is possible to work out the indefinite integral
J — 7)^2xdx
by expanding the bracket and using the sum-difference rule. But the
solution is easier if an appropriate substitution — a change of variable —
is made. Suppose we let
u = x^ — 7
With this change only, the integrand would then be
JuÎxdx
which looks worse than the original! But what is needed is an

expression entirely in u. To achieve this the substitution must be carried
through to any remaining parts of terms in x, and du and dx need to
be related. Now, since u. — x^ — 7,
^ = 2x (8.16)
dx
Interpreting du/dx in (8.16) as a ratio, we can write
da = 2xdx (8.17)
In (8.17) 2xdx is the differential of a. The differential of a function

u =/(x), designated da, is defined as/'(x)dx. The usefulness of the
differential here is that (8.17) can be used to express the integrand
solely in terms of a. When 2xdx is replaced by da, the simplicity of the
resulting integral is evident:
(8.18)
The final step is to convert the right-hand side of (8.18) back to an

expression in x, the role of a as an intermediate device being complete.
Making this substitution, the result is
Consider a further example. Suppose we need the indefinite integral
/(.^+2>:^4)(.+ l)d*
We attempt as the substitution setting a equal to the larger bracket in

the integrand. The word ‘attempt’ is used with justification. The first, or
subsequent, attempts may not work, and judgement can be necessary
in choosing effective substitutions. In doing this, the hope is that the
remainder of the expression will relate conveniently to da. This turns
out to be the case here. Making the substitution,
a = x^ + 2x — 4

and, following the steps of the first example above, the integrand
initially becomes
u(x + l)dx
Taking the derivative
du
= 2x + 2
dx
we can write
du = (2x + 2)dx
or, more to the point,
/ -.xj du
(x + Ijdx = —
The integrand and indefinite integral are
which on re-expression in x becomes
(x^ + 2x — 4)^ ,
-i-
The integrand is simple enough for the result to be checked by
expansion and integration of the original expression in terms of x.
We now consider two examples in which the method of substitution
comes into its own. Find the indefinite integral
J 0.5x(x^ -F 10)^dx
Let
u = x^ -F 10
so that
du
du = 2xdx or 0.5xdx =
Substitution in the integrand produces

i
1
ui0.5xdx = J ^du = — J wdu
uV2
+k
(x2 + 10)^/^
Integration 331
For the second example, consider
/ (xO-^ - 1)^
^0.8
dx
The integrand can be re-expressed as

^-0.8(^0,2 _ 1)4
As the substitution try
u = (x°'2 - 1)
from which will result
du = 0.2x“° ^dx
so that
x~°'®dx = 5du
In terms of the substitute variable, the required integral has now

become
j Su'^du = + k = (x°'^ -1)^ +k
The method is only useful if the substitution simplifies the integrand.

Not all possible substitutions have this effect. The approach is likely to
be productive when the integrand can be expressed as a constant
multiple of the product of a function of the substitute variable and its
derivative; that is,
km^
dx
(8.19)
In the simpler cases k = 1. The examples we have considered are of the

form (8.19). The substitution approach is less likely to be of help if the
substitution results in a variable multiple (for example x or x^) in place
of k. Finally, note the following points:
1 More than one possible substitution may be effective.

2 Complicated expressions may need a sequence of substitutions.
3 There may be no substitution that is effective.
There are many deceptively simple-looking expressions for which the

- 1
integral is not an algebraic function - for example (x^ + 1)^, the
integral of which is an elliptic function.
Exercises 8.5
1 Use integration by substitution to find the following integrals:
(i) /(10 + x2)22xdx
(ii) J (2x^ + 6x -f 15)^ (2x + 3)dx

(iii) /4x(x2 - 15)5dx
(iv) /(6x2 -8x + 20)(3x-2)dx
(i) f 3x(5x^ — 19)2dx
(ii) /-4(5+xr^dx
dx
(iii)
(x - 5)^
-6
(iv) dx
(3x + 4)^
3 For c/ 7^ 0, b 7^ 0, n 1 confirm that
r d . d
dx = n— 1
+k
[bx + c)" b(l — n)(bx + c)
(i) ^ +^ dx
(x2 + 7x - 8)^
6x^ + 4x + 2
(ii) -dx
(x^ + x2 + X + 1)"
8.6 The indefinite integral of a function of x is a function: the definite

The definite integral is a number - the difference between the values of the
integral indefinite integral for two values of x. These values are end-points of
an interval in the domain and are the limits of integration. A value
for the definite integral may be all that is required - an advantage
when the expression has no antiderivative in terms of elementary
functions (those that can be built up from polynomial, rational,
exponential and trigonometric functions, and their inverses). In such
cases the definite integral may be sought by numerical methods. The
value of the definite integral of a function of a single variable
corresponds to the area under the curve of the integrand between the
limits of integration. It was the problem of finding the area enclosed
by a curved boundary that gave rise to the integral calculus. The
definite integral of a function /(x) between limits of integration a and b
is written as
px = b
/ /(x)dx
Jx = a
for which
rx = h
for X = b for x = a
f(x)dx = (F(x) -I- k) - (F(x) + k) (8.20)
/
Jx = a
where F(x) + k is the indefinite integral a is the lower and b the upper
limit of integration. The constant of integration, k, cancels as the
Integration 333
integral is evaluated. Consider the definite integral oif{x) — 2x from
X = 2 to sc = 5:
r=5 for X = 5 for x = 2

2xdx = {x^ + k) - (x^ + k)
Jx = 2
= 25 + /c-(4 + k)
= 21
The definite integral is the difference between the heights (distances
above the x axis) of the curve F(x) = x^ + at x = 5 and x = 2. This is
illustrated in Figure 8.2, which is drawn for = 0.
Figure 8.2
In the theory of the firm, iff(x) = 2x is marginal cost, then F(x) is

total variable cost. The definite integral represents the change in total
variable costs when output rises from x = 2 to x = 5. The output
increase could be justified in profit terms only if there was at least an
equal increase in revenue between 2 and 5.
Consider some examples:
l'X = i
(i) / (x^-l)dx = (0.25x^-x)^=^-(0.25x^-x)’'=2
Jx = 2
= (0.25(81) - 3) - (0.25(16) - 2)
= 17.25 - 2
= 15.25
px = ‘i
(ii) / (6x^ — 4x + 4)dx = (2x^ — 2x^ + 4x)‘'‘^^
Jx=l
— (2x^ — 2x^ + 4x)*^ ^

= 128 - 32 + 16 - (2 - 2 + 4)
= 108
The lower limit of integration is often zero, as in the next example:

l'X = 8
(iii) / 0.625xMx = (0.125x5)* = ^-0
Jx = 0
= 0.125(32 768)
= 4096

Neither the limits of integration nor the definite integral have to be
positive. This is illustrated in the following case:
l'x=5
(iii) / (4x — 20)dx = {2x^ — 20xY^'^ — {2x^ — 20xy~^

Jx = -3
= -50-78
= -128
Figure 8.3 shows the graph of F{x) = 2x^ — 20x.
Figure 8.3
F(x) is positive at x = — 3 and negative at x = 5, so the difference

will be negative as the upper limit of integration is taken first. Now
consider an integrand with a rational exponent:
l'X = 4 i'X = 4
(v) / 3(-\/x)dx = / 3x^dx
Jx=l Jx=l
= 16-2
= 14
Where the upper or lower limit of integration is infinite, we speak of an

improper integral. For example, if the definite integral
px=U
/(x)dx
Jx — a
approaches the value D as the upper limit of integration, L7, increases

without limit, then the improper integral
/(x)dx
Integration 335
exists and its value is equal to D. For example
(Vi)
= 50
Improper integrals in which one or both limits of integration are

infinite often arise in applications involving probability. An integral is
improper if the value of the integrand becomes infinite at or between
the limits of integration. For example
(vii) f ^dx = {20^/x + — i20y/x +

Jx = 0 V^
X = 40
Although 20^x has a definite value at x = 0, the integrand, 10/v^x, is

not defined at the lower limit. If the integrand becomes infinite at a point
between the limits of integration, the improper integral, where it exists,
can be obtained by dividing the range of integration into two parts - in
which the problematic value of x is the upper limit of integration in one
part and the lower limit in the other. The required overall value is the
sum of the two (still improper) integrals. The definite integral can be seen
as the area under the integrand between the limits of integration. We
can illustrate this property in the case of the function /(x) = 2x between
the limits x = 2 and x = 5. We saw from Figure 8.2 that the definite
integral was the difference, 21, in the height of x^ at x = 5 and x = 2.
In terms of the integrand, 2x, 21 is the area between the interval x = 2
to X = 5 and /(x) = 2x. This is illustrated in Figure 8.4.
Figure 8.4
In Figure 8.4 the area of interest is that of the trapezium made up of

the rectangle, with area R, and the triangle, with area T. The area of
the rectangle is base times height, so therefore
R = (5-2)/(2) = 12

The triangular region has an area of half the base times height, so
r = i(5-2)[/(5)-/(2)]
= 1.5(10-4)
= 9
The area between the line and the x axis above the interval is
Area = R + r= 12 + 9 = 21
The fact that, whatever the limits of integration, the definite integral is
the area under the curve can be confirmed in the case off{x) = 2x.
With limits of integration a and b the definite integral is
r %xdx = (X^)" = '’ - (x2)" = «

Jx = a
= b^-a^
The rectangular area is
R = {b — a)f{a) = (b — a)2a = 2ba — 2a^
The triangular area will be
T = l{b-a)\f{b)-M]
= j{b — a){2b — 2a)
= (b- a)^
= b^ — 2ba +
So the trapezoidal area as a whole is:
Area = R + T
= 2ba — 2a^ +b^ — 2ba +
= b^-a^
In the theory of production, total variable cost at an output level is

the area under the marginal cost curve from the origin to that output. The
total area can be seen as the sum of successive trapezoidal columns
with unit base - the incremental costs of successive units of output.
We now set out rules for, and properties of, definite integrals.
(i) f c/(x)dx = c f f{x)dx (8.21)

Jx=a Jx=a
^ The constant multiple rule applies to definite integrals. Consider

px = 20 _ /'X = 20 ^
[ ^xdx = n [ ^dx
Jx = l0 2 2* = io 2
= 7r[(0.25x2 + fc)f°'-’= = 20 _ (0.25x2 +fc)f°^’^=10]
= 7r[(100 + k)-(25 + /c)]

= 75tt
Integration 337
The next rule is also familiar from rules for indefinite integrals;
(ii) [
Jx=a
\f{x) ±g{x)]dx = f
Jx=a
f{x)dx± f
Jx=a
g{x)dx (8.22)
The sum-difference rule is applied in the following example:

|•x = 2 i‘X = 2 nx = 2
/ (2x + 3x^)dx = / 2xdx+ / 3x^dx
Jx = 0 Jx = 0 Jx = 0
After integration of both sides, the definite integral is evaluated as:
{x^ + x^f^^ - {x^ + = [{x^f^^
i.e.
12 = 4 + 8
The third rule is

nX -x = b0 px = a
(iii) / f{x)dx = — f{x)dx (8.23)

Jx=a Jx=b
It follows from this fairly obvious property that
f fix) = 0
We now turn to a result mentioned in the context of improper

integrals. For a point c in the range of integration (that is, a < c < 1?)
the definite integral can be broken down into two (or more) parts:
px = b j-x = c rx = b
(iv) / f{x)dx— / f{x)dx+ / fix)dx (8-24)
Jx=a Jx=a Jx=c
In addition to use with improper integrals, we will see two further ways
in which this rule is useful. But first consider two examples of
integrands that could be evaluated without the rule. Given the
requirement to find
.x=2
2xdx
Jx=\
fx=l
which is
,.2\x= 3
ix^)^ = ^ - (x^)^=^ = 9-1 = 8
This outcome could be obtained by selecting (say) c = 2 and evaluating
the definite integral in two parts as follows:
rx = 2 px = 2
/ 2xdx+ / 2xdx
Jx = 1 Jx = 2
2\x = 2 r^2\x=l
+ [(^2)* = ^ 2\j: = 2i
(X^)
= (x^)
= 3+5
= 8

Now find
(0.75x^ + 6x+ 10)dx

x=2
= (0.25x^ + 3x2 + 10xr=* - (0.25x3 + Bx^ + lOxf = 2

= 400 - 34
= 366
With, for example, the value of c set at 6, the result is
(0.25x3 + 3x2 ^ loxf = ^ - (0.25x3 + 3x2 ^ iQxf = 2
+ (0.25x3 + 3x2 ^ ioxr=3 - (0.25x3 + 3x2 ^ 20xr = ®

= 230 - 34 + 400 - 230
= 196 + 170
= 366
The rule comes into its own in two types of problem, examples of
which follow. The first case is piece-wise defined functions. The value of c
would coincide with the point where the definition of the function
changes, as in
Jx = 2
where
r/ N j 2x + 5 for X <6
\ 5x — 10 for X > 6
Figure 8.5 shows the graph of the function and the area under the
curve corresponding to the definite integral. The break is set at c = 6.
Excluding the constant of integration, the indefinite integral will be
^ f x^ + 5x for X < 6
\ 2.5x^ — lOx for X > 6
so the definite integral is
(x2 + 5x)*^^ - (x2 + 5x)*^^3 (2.5x2 _ 20x)*’^3
- (2.5x2 _ 20xf=^
= 66-24 + 80- 30
= 92
»If the graph of the integrand crosses the x axis between the limits of
integration (8.24) is again useful. Consider
(x - 5)dx = (0.5x2 _ 5^yv = 6 _ (Q 5^2 _
= (18 - 30) - (8 - 20)

= 0
Integration 339
Figure 8.5
This result is correct and may be the relevant value (for example when
the surpluses and deficits of business units are to be aggregated). But as
is clear from Figure 8.6, there is some space between the integrand and
the X axis.
Figure 8.6
The problem is that a negative sign is given to areas under the x

axis. The positive and negative parts offset each other and in this case
cancel altogether. But we may need to know the total area regardless of
sign. This value is found by working out the integral either side of the
zero of the function and ignoring sign. Here, this is found as follows:
(x — 5)dx = (0.5x^ — 5x)^ ^ -(0.5x^ - 5x)* = ^
= -12.5-(-12)
= (-)0.5

Where the function is above the axis:
''X = 6
[ {x- 5)d3c = (0.5x^ - 5xf = ^ - (O.Sx^ - 5x) x=5
Jx = S
= -12 - ( - 12.5)
= ( + )0.5
The absolute values of the two areas are added to give the total area
enclosed between the function and the x axis. This is one unit. A further
example illustrates the fact that (8.24) can be extended to involve as
many subdivisions of the range of integration as are needed. We shall find
i'X = 16
/ (3x^ — 36jc + 60)dx
Jx = Q
If one or both roots of the quadratic fall between the limits of
integration, subdivision of the interval will be required. The roots are
X = 2 and x = 10, and the absolute value of the area is found by
dividing the range of integration into the sections
0-2,2-10 and 10-16
The indefinite integral (constant omitted) is
(3x^ — 36x + 60)dx = x^ — 18x^ + 60x
So
(x^ - 18x2 + 60x)*=2 _ _ 18x2 + 60x)*=°
56-0
56
|•x = 10
(x2 - 18x2 + 60x)^ = 20 _ (^3 _ 50x)^ = 2
Jx = 2
-200 - 56
-256
x= 16
I = 10
(x^ - 18x2 + 60x)*=26 _ (x^ _ 18x2 + 60x)^=^°
448 - ( - 200)
648
So the total area between the curve and the x axis regardless of sign is
. 56 + 256 + 648 = 960
If sign had been taken into account, the result would have been
'^x=l6
rX = J
= (x^ - 18x2 + bOx)^^^'’ - (x2 - 18x2 50x)’= = o

Jx = 0
= 448-0
= 448
Integration 341
Where it is required to find the area enclosed between two curves,
rather than the area between one curve and the x axis, the area can be
obtained with the use of definite integrals if the functions concerned are
non-negative over the relevant interval. If the two curves are those of
y = fix) and y = gix), and the limits between which the area is required
are a and b, then the definite integral needed is
fx = h
/ I \fix) - g(x)]d X
provided that f{x) > gix) between a and b. If it is more convenient, the
integral could be expressed as
rx^b px = b
0(x)dx (8.25)
/
Jx = a
fix)dx
/
Jx = a
■
Figure 8.7 graphs a possible situation for parabolic/(x) and linear

^(x). The enclosed area is shaded. In Figure 8.7 it is presumed that the
required area is the entire region between the curves, although the
definite integral (8.25) is not restricted to this case. For a numerical
exercise, suppose that
fix) = 125 + 20x-x^
and
gix) = 80 -(- 8x
Figure 8.7
Apart from the steepness of the curves, the problem graphs as shown in
Figure 8.7. We assume that the entire region is required, so the first
task is to find the points of intersection of the curves. These points give
the limits of integration, a and b. Begin by equating the values of the
functions:
fix) = gix)
I25 + 20x-x^ = 80 + 8x

so that
- 12X-45 = 0
which factors as
(x+ 3)(x- 15) = 0
so the limits of integration are
a = —3 and b = +15
Carrying out the integration in the form of (8.25),
= ( - x^/3 + + 45x)’' - x^/3 + + 45%)’' ^

= 900 - ( - 72)
= 972
So the area between the curves is 972.

In problems involving areas enclosed between curves, care must be
taken if the curves cross within the range of integration. This is the
reason for the stipulation that/(x) > g{x) between a and b. If this is not
the case, the range of integration can be broken up into intervals
within which one function is never less than the other.
Exercises 8.6
Evaluate the following definite integrals:
(i)
(ii) [ (6x — 3)dx

-5
20
(0.9x^ + X + 10)dx
2 Find the following definite integrals;
0
(iv)
Integration 343
3 Find the area between the x axis and the following curves
between the given limits:
(i) f(x) = 3x2 + 2x + 5
between the limits x = 0 and x = 10.
(ii) f(x) = x^ + 1 Ox
between the limits x = — 10 and x = —2.
(iii) f(x) = 1.5x2+ 12x +30
between the limits x = —5 and x = 5.
4 Find the total area under the curves given by (i) and (ii) below,
between the given limits and regardless of sign:
(i) f(x) = X - 10
between x = 5 and x = 15.
(ii) f(x) = 3x2 _ 1 8x + 15
between x = 0 and x = 8.
5 Evaluate the following:
^40
(i) / f(x)dx
Jo
where
fi^\ = / 20 for X < 20
\X for X > 20
(ii) ^ ^(^)dx
where
f 4x — 20 for X < 10
\ X + 10 for X > 10
8.7 We evaluated a definite integral by finding an antiderivative, working

Numerical out its value at the end-points of the range and taking the
integration difference. But this cannot always be done. Some indefinite integrals
are difficult to obtain and other expressions have no elementary
formula for the antiderivative. Numerical methods compute an
approximation to the definite integral, and we will examine two
methods: the trapezium rule (in US terminology the trapezoidal
rule) and Simpson’s rule.
We will start with the trapezium rule. Suppose we require the
definite integral for a function for which a section of the curve is shown
in Figure 8.8. The limits of integration are a (lower) and b (upper). The
interval a, h is divided into a number of sections or steps - here
arbitrarily six. The section width or step size is w, given by
In each step a trapezium is formed, the base of which is the step width;
the parallel sides are the ordinates at the beginning and end-points of
the section, while the fourth side is the straight-line segment
connecting the tops of the ordinates. The area of the trapezium

Figure 8.8
>-
a w w w X
approximates to the area under the curve in each section, the accuracy
of the approximation in each step being better the smaller is the step
size. With f{x) being the value of the function (the height of the
ordinate) at any point, the areas of the trapezia are
Ti =^{fia)+f(a-\-w)]
w
^2 = ^ [/■(« + w) +f{a + 2w)]
w.
T3 = ^ [/”(« + 2w) +f{a + 3w)]
T4 — — \f{a + 3w) + f{a + 4w)]
75=^ \f{a + 4w) +f{a + 5w)]

w
'^6 = 2 + 5w) +f{a + 6w)]
So the total area of the trapezia and the approximation to the value of
the definite integral is T, where
T = 0.5w[/'(a) + 2f{a + w) + 2f{a + 2w) + 2f{a + 3w)

+ 2f{a + 4w) + 2f{a + 5w) + f{a + 6w)]
The trapezium rule for approximating definite integrals is as follows:
Trapezium rule
w
/ fix)dx ^ — \f{a) + 2f{a + w) + 2f{a + 2w) + ...+/(« + mw)]
Jn ^
where m is the number of steps. Let us use the rule to approximate the
area under the curve y = x^ between x = 1 and x = 4. We will
arbitrarily divide the interval into six steps, so the step size, w, will be
w=
Integration 345
Using the rule,
r = ^ IfiD + 2/(1.5) + 2/(2) + 2/(2.5) + 2/(3) + 2/(3.5) +/(4)]
= 0.25[(1)^ + 2(1.5)^ + 2(2)2 ^ 2(2.5)2 + 2(3)2 ^ 2(3.5)2 + (4)2]

= 0.25(1 + 4.5 + 8 + 12.5 + 18 + 24.5 + 16)
= 0.25(84.5)
= 21.125
Here we can find the exact value of the integral by analytical means:
So in this case the trapezium rule is out by just under 0.6%. Given the
shape of ^ = ^2, since the area of each trapezium is slightly greater
than the true area under the curve in its section, the result must be an
overall overestimate. With other functions, the result may be either an
overestimate or an underestimate. Accuracy increases as the step size is
reduced so long as the function has a continuous second derivative, and
up to the point where rounding errors overtake the theoretical
improvement from reduced step size.
Table 8.1 gives the results of the trapezium rule for the current
example and for a range of step sizes. The increase in accuracy as the
number of subdivisions increases is evident. The error is less than 1%
for 5 steps, and less than 0.1% if 15 steps had been used. But in most
cases it will not be possible to calculate the precise percentage error.
Now use the trapezium rule with five subdivisions to estimate the
definite integral
With
w = (b — a)/5 = 1
the trapezium rule gives
T = 0.5[f(l) + 2/(2) + 2/(3) + 2/(4) + 2/(5) +/(6)]

= 0.5(16 + 40 + 36 + 32 + 40 + 36)
= 100
The exact value of the integral can again be obtained in this case:
= (0.25x^ - 3x^ + 12x2)*=^ - (0.25x^ - 3x^ + 12x2)’' = ^

= (324 - 648 + 432) - (0.25 - 3 + 12)
= 108 - 9.25
= 98.75

Table 8.1
Steps T error % error
1 25.50000 4.5000 21.4286
2 22.12500 1.1250 5.3571
3 21.50000 0.5000 2.3810
4 21.28125 0.2813 1.3393
5 21.18000 0.1800 0.8572
6 21.12500 0.1250 0.5952
10 21.04500 0.0450 0.2143
15 21.02000 0.0200 0.0952
20 21.01125 0.0112 0.0536
25 21.00720 0.0072 0.0343
50 21.00180 0.0018 0.0086
100 21.00045 0.0004 0.0021
200 21.00011 0.0001 0.0005
300 21.00005 0.0000 0.0002
If the upper limit of integration had been 5 instead of 6, then with

the interval divided into four sections the trapezium rule gives precisely
the correct value of 72 for the definite integral.
Simpson’s rule uses a parabolic rather than a linear function to
approximate the curve in each section of the interval. The formula is
Simpson’s rule
/ f{x)dx [/'(a) + 4/(a + w) + 2f{a + 2w) + 4(fl + 3w) + 2f{a + 4w)
+ ... + 2f{a + {m- 2)w) + 4/(a + (m - l)w) +/(b)]
where w is defined as for the trapezium rule and where the number of
steps, m, must be even.
We will now use Simpson’s rule to approximate the integral
'‘X= 3
x^dx
i:
lx = 0
Using m = 6 subsections, we have:

(3-0)
w = 0.5
The area given by the rule is
S = (0.5)/3[0 + 4(0.5)^ + 2(1)^ + 4(1.5)^ + 2(2)^ + 4(2.5)^ + (3)^]

= (0.5)/3(0.25 + 2 + 20.25 + 32 + 156.25 + 81)
= (0.5)/3(291.75)
= 48.625
For comparative purposes, the exact area can again be calculated:
x'^dx = (0.2x^)* ^ - (0.2x^)’'-° 48.6
So the error in this case is 0.025, or just over 0.05% of the true figure
- and this for a division into only six sections.
Integration 347
Now estimate the area under the same curve using the trapezium
rule. This is
T = 0.5/2[/’(0) + 2/(0.5) + 2/(1) + 2/(1.5) + 2/(2) + 2/(2.5) +/(3)]

= 0.25(0 + 0.125 + 2 + 10.125 + 32 + 78.125 + 81)
= 0.25(203.375)
= 50.84375
an error of just over 4.6%, given the chosen number of divisions of the
interval.
Usually Simpson’s rule gives better results than the trapezium rule,
though this is not true in every case.
Exercises 8.7
1 Using m = 6 steps, estimate the value of the definite integral
(6x2 ^ 2)dx
(i) Using the trapezium rule.

(ii) Using Simpson's rule.
(iii) Calculate the exact value of the integral by analytical
means.
2 For the definite integral
0.5x2 + l)dx
and with m = 4 steps:

(i) Estimate the value of the integral using the trapezium rule.
(ii) Estimate the value of the integral using Simpson's rule.
(iii) What is the exact value of the integral?
8.8 The purpose of this chapter was to introduce the concept of an integral,
Concluding and to enable polynomial and rational expressions to be integrated
remarks using simple rules. More advanced devices for integration include the
following:
• further analytical techniques (such as integration by parts and

integration by partial fractions):
• use of standard formulas (where the form of the integrand is such
that the indefinite integral is already known;
• use of Taylor series approximation to an expression:
• use of more sophisticated numerical methods, such as variants of
Simpson’s rule, to obtain the value of definite integrals:
• use of software such as Mathematica, MathCad, Maple or Derive,
which not only evaluate definite integrals but also find expressions
for indefinite integrals where these exist.

The definite integral has many uses: in finance in the calculation of
present values under continuous compounding and in microeconomics
in estimating producer or consumer surplus. We will revisit integration
with exponential and logarithmic functions in Chapter 9.
Integration 349
CHART E R
9 Exponential and logarithmic

functions
In this chapter exponential and logarithmic functions are introduced,

their relation to growth and appreciation, decay and depreciation, and
elasticity is shown, and a marketing application is given. We show how
exponential and logarithmic functions can be differentiated and
integrated: integration by parts is introduced and further examples of
differential equations are given.
LEARNING OUTCOMES
By the end of the chapter you will be able to use calculus on simple
exponential and logarithmic functions. You will gain useful insight into
business, financial and economic concepts, and be aware of important
properties of models based on exponential and logarithmic functions.
9.1 Exponential functions are widely used in economic and business

Exponential models where change and growth are involved. In polynomials the
functions: exponents of the variables are constants. Exponential functions have
introduction variable exponents. For example, with x as the variable,
y = a^ (9.1)
is the simplest example of an exponential function. In (9.1) a is the

base and is a positive constant not equal to one. The variable x is the
exponent. The behaviour and appearance of a* depends on whether the
base, a, is greater or less than one. Consider the case where a > 1,
specifically
y = 2^
Consider some values ot y = 2^ for various x, as shown in Table 9.1.

Successive unit increases in x produce successive doubling of the value
of the function. This factor of multiplication is the value of the base.
Conversely, unit decreases in x halve the function’s value - the multiple
here being the reciprocal of the base. The function z/ = 2^ is graphed in
Figure 9.1.

Table 9.1
X 0123 456 -1 -2 -3 -4
1 2 4 8 16 32 64 0.5 0.25 0.125 0.0625
Figure 9.1 ^WA
0 X
Figure 9.1 is typical of exponential functions with a base greater

than one. Features of the function and its graph are that it is:
• positive everywhere:
• always increasing:
• increasing at an increasing rate.
Such curves are growth curves and are used to model expansion and
increase - for example of populations, atmospheric CO2 or outstanding
debt at a compound rate. As one example, for a developed economy
an arguable sustainable annual rate of growth in real gross domestic
product (GDP) is around 2.5%. Thus
GDP(1993) = GDP(1992)(1.025)
or, with reference to an initial year zero, GDP in some year, t, would be
given by
GDP, = GDPo(1.025) (9,2)
In (9.2) the base a = 1.025. The base of the exponential function is one
plus the rate of growth as a decimal. If in general the rate of growth
had been 100r%, then the relationship would have been
GDPt = GDPo(l + r)
Now consider an exponential function where the base is less than

one:
Table 9.2 shows values of the function for various x. As x increases

by unit increments the value of the function is progressively halved.
Conversely, as x is decremented the value of the function doubles. The
graph oi y = {jY is shown in Figure 9.2.
Exponential and logarithmic functions 351

Table 9.2 X 0 1 2 3 4 -1 -2 -3 -4 -5
(if 1 0.5 0.25 0.125 0.0625 2 4 8 16 32
Figure 9.2
The graph is the reflection in the y axis of the graph of 2^^ and it
typifies exponential functions with a base less than one. That is,
0 < a < 1. Properties of the graph are that it is:
• positive everywhere;
• always decreasing;
• decreasing at a decreasing rate.
Exponential functions with base less than one produce decay curves
and are used to model processes of contraction. It is a characteristic of
decay curves that, if x represents time, the time required to reduce y to
half its value is the same regardless of the initial value of y. Note also
that the process of exponential decay could just as well be modelled
using the form of function
y=
where a > 1. This follows from the fact that
In terms of the examples we have been using,
(if = i/ = 2-'
There is an obvious relevance of decay curves to physical processes such

as radioactive decay. In economics and finance, decay curves can be used
to model the depreciation of assets, economic output during periods of
recession or, for the domain x >0, demand. For example, suppose that
during a protracted recession GDP is contracting at 1.5% per annum in
real terms. After t years of recession, GDP will be given by
GDPt = GDPo(0.985)‘

In finance, suppose that each year the retail value of a car depreciates
by 20%. The value of the car after t years will then be
= Vo(0.8)‘
where Vo is the original value of the car. So if the car cost £7500
initially, then after four years its value would be V4, where
V4 = 7500(0.8)^ = 7500(0.4096) = 3072
Consider growth curves again. An important financial application of

exponential functions is in compound interest. Suppose that a bank
deposit account pays 10% interest annually compounded. Suppose that
the sum deposited initially, the principal, is £1. By the end of one year
the investor wilt have £1.10 in the account - the amount on deposit at
the start of the year, £1, plus the interest on this sum, 0.1(£1). Now
suppose that the whole £1.10 is left on deposit. There is no reason to
treat this investor less well than a new customer making a deposit of
£1.10 at this time, so for the second year interest is earnt on the entire
sum - principal and first-year interest - of £1.10. This is the key point
in compounding - that interest itself earns interest. At the end of the
second year the investor will have what the year was started with,
£1.10, plus the interest on this sum, 0.1(£1.1). The total is £1.21.
Table 9.3 records the growth of the original £1 invested.
Table 9.3
Principal One year Two years Three years
1 1 +0.1(1) 1.1 + 0.1(1.1) 1.21+0.1(1.21)
= 1.1 = 1.1(1 +0.1) = 1.21(1+0.1)
= (1 +0.1)2 = (1 + 0.1)^
= 1.21 = 1.331
After n years the original £1 becomes £(1 + 0.1)” at 10% interest,

and £1 invested for n years at a rate of 100 x% would become
£(1 + x)”. The future value, FV^, of a principal sum £P after n years
at 100 x% is then
FV„ = P(l+x)" (9.3)
In writing (9.3) as an exponential function of n, we have used x to

represent an unspecified component of the base - the reason for this
will become evident later. Consider an example using (9.3). The future
value of £75 invested for 12 years at an 8% compound interest rate is
■J
FV12 = £75(1 + 0.08)^^ = £75(2.51817) = £188.86
If £75 had been invested for 8 years at 12% the result would have been
FVs = £75(1 + 0.12)® = £75(2.47596) = £185.70
In practice financial institutions pay interest more frequently than

once a year on many accounts. Let us see how the formula can be

adjusted to allow for more frequent payments. If the quoted rate of
interest was 100x% but in fact interest was paid half-yearly at
100(x/2)%, the result is two compounding periods a year and (for
investors) a rare example of benign understatement! In these
circumstances, £1 on deposit for one year at 100x% nominal becomes
X T X
(l+-)^ = l-l-x + ->l-l-x
So future value is increased using half the nominal interest rate

compounded twice as often. For example, if a nominal 12% annual interest
is in practice compounded twice at 6%, the end-of-year value of £1 is
(1+0.12/2)^ = 1.1236
equivalent to an annually compounded rate of 12.36%. What if interest

was compounded quarterly? At 12% the result would be
(1 + 0.12/4)^ = 1.1255 or 12.55%
If interest was compounded monthly the result would be
(1 + 0.12/12)^^ = 1.1268 or 12.68%
Credit cards usually have interest quoted as a monthly rate. The

equivalent annual rate is more than twelve times the monthly rate; 2%
per month is equivalent to 26.82% p.a. (given by (1.02)^^ = 1.2682).
In some major financial transactions interest may be charged on a daily
basis, so the equivalent annual rate in this case would be given by
(1 + 0.12/365)^^^ = 1.1275 or 12.75%
For example, £850 borrowed initially at a nominal 12% but with daily
compounding would, by the end of the year, have grown to
850(1 +0.12/365)^^5 958
What would happen if interest was compounded more frequently

still? Is there a limit to the value that the equivalent annual rate would
take? The answer is provided by the limiting value of
as n increases without limit and approaching continuous

compounding. In working towards a value for (9.4), we start with the
simpler case where x = 1. Table 9.4 shows values of (1 + l/n)” for
selected n.
Table 9.4 0 1 2 10
n 5 25 100
(1 + l/«)" 1 2 2.25 2.48832 2.59374 2.66584 2.70481
n 500 1000 10 000 100 000 500 000 1 000 000
(1 + 1/n)" 2.71557 2.71692 2.71815 2.71827 2.71828 2.71828

As n increases, the value of (1 + l/n)” approaches the figure
2.71828 (to five decimal places). When stated as a limit (a value that
an expression approaches ever more closely as the independent variable
increases), this is written as
lim (1 + l/n)” = 2.71828 18284 59045 ... = e (9.5)

n oo
e, an irrational number, known as the natural number, is an important

special constant. What of (1 + x/n)"? It can be shown that the limit of
this term as n increases is e raised to the power of x. That is,
lim fl+-)''=e" (9.6)
y = is a most important exponential function. Before considering

properties of e*, we can answer the question of where the process of
ever more frequent compounding was leading. 12% continuous
compounding gives an end-of-year result for a principal of £1 of
Pii 1.1275 or 12.75%
Two years of continuous compounding at 100x% p.a. produces a future

value of So with x = 0.12, after two years £1 would become
£g2(0.12) _ £1.2712
In general, if a principal of £P is invested at 100r% p.a. continuously

compounded, with no withdrawals, the future value after £ years is
PV(£) = Pc'"'
So, for example, a principal of £500 deposited for two years at 12%
continuously compounded would produce a future value of
FV(2) = 500eO-^^ = 635.62
while over five years the future value would be
FV(5) = 500e°^ = 911.06
For any continuously compounded rate there is a unique equivalent

discretely compounded rate. For example, the equivalent rate to 12%
continuously compounded with once-yearly compounding is 100//%,
where:
(1+ //)" = 1.271 2
which solves for y = 1.1275 or 12.75% per annum.
Exercises 9.1
1 In a certain country the level of GDP in 1998 was £350bn.
Growth of GDP is at the rote of 3% per annum.
(i) Express this relationship as an exponential function.
(ii) Assuming that the relationship remains valid, use the
function to state GDP in (a) 2000 and (b) 2004.

2 A car is bought new for a price of £ 10 000. By the end of
each year the car has lost 25% of its value at the start of that
year.
(i) Express this relationship as an exponential function.
(ii) Find the value of the car after (a) 3 years and (b) 6 years.
3 (i) £750 is deposited in an account in which interest is
compounded at 8% per annum. With no withdrawals,
what is the value after, (a) 5 years, (b) 10 years and (c)
20 years.
(ii) Now suppose that the interest on the account was 16%.
Find the value of the account after (a) 5 years, (b) 10
years and (c) 20 years.
4 A finance company charges a nominal rate of interest of 24%
p.a. If £5000 is borrowed, calculate the outstanding debt
after a year with the frequency of compounding:
(i) Once a year (ii) Twice a year
(iii) Four times a year (iv) Monthly
5 With interest compounded continuously at 10%, find the value
of £800 invested for:
(i) 2 years (ii) 10 years (iii) 20 years
9.2 The function y = is the natural exponential function. An

The natural alternative notation for e* is exp (x), which is easier to print,
exponential particularly where the exponent is a function. Although a much larger
function and its base than e could be selected, following an analogy in Finney and
derivative Thomas (1990) we can use the values of to illustrate the
phenomenal growth shown by exponential functions. In Figure 9.3
both axes are measured in centimetres, and co-ordinates for x = 1 and
X = 2 are shown. It is not quite possible to continue the graph to
X = 3, when its height would be just over 20cm. Table 9.5 gives the
height reached against the y axis (by for selected values of x. As can
be seen, it is unlikely that values of x in excess of 65 will have much
practical significance!
Table 9.5
X e* Comment
5cm 1.48 metres Shoulder height
10cm 220 metres Height of respectable skyscraper
25cm 720 000 km Well beyond the orbit of the moon
50cm > 5000 light years 1 /6 of the way to the galactic rim
65cm > 15 bn light years To the edge of the universe
Returning to earthbound discussion, the more general form
y = Ae"’* (9.7)

Figure 9.3
describes a wide variety of processes of continuous change. The value of

A locates the y intercept while the value of m modifies the slope. Values
for A and m are chosen to fit the situation under study. For example,
the formula for calculating the future value of the principal invested
under continuous compounding is an application of (9.7).
Now consider the derivative of the function y = e^. Recall that e* was
the limit of (1 + x/n)" as n increased. We can use the chain rule to find
the derivative of (1 + x/n)" and then find the limit of the expression
obtained. If
then by the chain rule

dy
-= 1+-
But as n increases indefinitely, the fact that the exponent of (1 + x/m) is

n — 1 rather than n, makes no difference, so
dje^) lim / x\'>-i
(1 + -) (9.8)
dx n ^ oo \ nJ
The function y = has the unique property of being its own
derivative. It follows that all derivatives of e* are also e*. Now consider
the derivative oi y = e'”^. It can be shown that
mx
= me (9.9)
dx

The chain or product rules can be used to establish (9.9). Using the
product rule to confirm the derivative when m = 2,
for which, using the product rule,
d{e’‘e^) 2x
= = 2e
ds:
Note that the exponent will not alter as higher derivatives are taken,
but the coefficient increases by a factor of m each time. Thus for the
nth derivative
^ ^ ^n^mx (9.10)
dx"
Consider some examples for specific values of m. First take rn as a

positive integer. For
fix) =
the derivative is
f'(x) =
Note that the second derivative here is
f"(x) =
Now apply the rule to the case where m is a negative integer:
m = e-"'
f'(x) = -2e-^‘
In this case the second derivative is
f"(x) = 4e-^^
Note that where the exponent is negative, derivatives of progressively

higher order will alternate in sign. Now consider a rational exponent;
fix) =
f'ix) =
and irrational exponents
fix) =
f'ix) = V2e^^'^
fix) =
fix) =

The derivative of y = is mAe"’^, Thus for
6x
y = 15e
6x
90e'
dx
In general, a function may contain terms involving both x” and e^. The
simplest case of such a combination is
y = xe*
for which the derivative is obtained by use of the product rule:
^ = xe" + g*
dx
= e’'(x + 1)
And given
S 2x
y = xe
the derivative may also be obtained using the product rule:
dx
The derivative of the function
y = e^*(2x + x^)^
may be obtained by combined use of the product rule and the chain rule:
^ = 4e^*(2x + x^)^ + e^^2(2x + x^)(2 + 3x^)

dx
This unsimplified form of the derivative is sufficient for many purposes

- for example when the slope of the function is required for specific
values of X. At X = 1 the derivative evaluates as:
= 66e^
(*=i)
If further analysis is needed (e.g. finding higher-order derivatives) the time

required to simplify dy/dx may be well spent. It can be rearranged as
^ = 2x(x^ -I- 2)(2x^ -f 3x^ -f 4x + 2)6“^*

dx
Considerable manipulation is required to achieve this simplification.

Programs such as Derive can make such rearrangements in less than a
second of processing time and free of error! The exponential function
fix) = can be seen as a special case of the exponent being a function
of X, as in
(9.11)

in which it will be assumed that the exponent gix) is a differentiable
function of x. The first derivative of (9.11) is
(9.12)
Note that the exponent does not change in the derivative. As an

example, consider a linear exponent
y = gmx + c
for which, since g'{x) = m, from (9.12) the derivative is
Now consider the simplest form of non-linear exponent, where
g(x) =
so the function to be differentiated is
In this case g'(x} = 2x and
Now consider
y = e-,x^ + 7x + 6
for which, using (9.12), since g'{x) = 2x + 7, the first derivative is
^ = (2x+7)e"' + ^’'+^
dx
Obtaining the derivative of
y = {x^ - 5x^)6"'
requires the use of the product rule and is
^ = 2x(x^ - 5x^)6^^^ -I- e*^(3x^ - 20x^)

dx
and, as usual, the slope of the function at any point can be found by
evaluating the derivative. So at x = 1 the slope will be
2(1 - 5)e + e(3 - 20)

dx
= -25e
The derivatives of exponential functions with bases other than e

involve logarithms and will be considered after logarithmic functions
have been introduced.

Exercises 9.2
1 Find the derivative dy/dx for the following exponential
functions:
(i) y = e^^
(ii) y = e“'^^
(iii) y = e°
(iv) y = — 10e“°
(v) y =
(vi) y = ?e°^^^
2 Find the second derivative, f”(x), in the following cases:

(i) f{x]=0.25e^^ (ii) f(x) =-200e-°i^
(iii) F(x) =
3 Find the required derivative in the following cases:
(i) f'"(x) where f(x] = \/2e^^
(ii) f"(x) where f(x) = —e“^
(iii) f'''(x) where f(x]=^^
4 Find the first derivative where:

(i) f(x) = 2xe^^ (ii) f(x)=xê'‘^
(iii) F(x) = e^(x + e^)
5 (i) Find the derivative dy/dx for y = e^^(3x + x^)^ and
evaluate your result at x = 1.
(ii) Find the first derivative of y = (2e^^ — e^^)(x^ + x)^ and
evaluate the result at x = 2.
6 Find the first derivative for the following:
(i) y = e20”10x
(ii) y=
(iii) y=
(iv) y=
(v) y = e'/^
(vi) y=
7 (i) Find the first derivative of f{x) = (2x^ + 4x-^)e°'^^^ and
evaluate the expression at x = 1.
(ii) Find an expression for the second derivative of f(x) = e^"
and evaluate the expression where n = 3 and x = 2.
9.3 This section considers the integration of exponential functions in which

Integration of the base is e. We begin with the simplest case:
natural
exponential
functions
J eÂx = + k
This result would be expected from what we have already learnt about
e^. Given that c* is its own derivative, it must also be its own indefinite
integral (disregarding the arbitrary constant).

Now consider the modest generalization: i/ = e'"*. The indefinite
integral is
/ e'^^dx = — + k
m
so, for example,
e^^dx = ^ + k
As further examples consider the following:
/ 6e^’^dx = ^ + k
_p-2x
e~^^dx = —--h k
I e^^dx = — +k
TT
It follows from the differentiation rule (9.12) that
/aWWdx = ^W + )c
So, for example.
J Ixe^^dx = e" +fe
and
J (^SX^+4,X + 9)e^' + 2.^ + ^ ^x^ + 2x^+9x^^
and
Jo,5xe^*"dx = 0.125e2*‘ + k
Consider definite integrals involving exponential functions, for

example
„2x 7,\x=l
I
r ^ e^^dx = (O.Se^" + kfi^^ - (0.5e^" + kfi
Jx=l
= 0.5e^ -0.5e^
= 23.605
and
^x=1.5
/ e^'-^dx = (0.25e^"-4)" = ^-^ - (0.25e^"-^)’'=^
Jx=l
= 0.25e^ - 0.25e°
1.597

while
(0.6x + 0.+ o.7x -1 =i

_ + 0.7* — 1 ^5c = 0
= 1 - 1/e
= 0.6321
As an example of the use of definite integrals of exponential functions,

consider the exponential probability distribution. This is useful in
management science and describes, among other things, the distribution of
service times for customers in a simple queueing system. The distribution is
/(f) = tie-'*' (9.13)

where the single parameter p > 0 and the domain is t > 0. In the
queueing theory context, t is the time taken to serve a customer and p
represents the mean rate of service. (9.13) is also referred to as the
negative exponential distribution. The graph of the distribution is
shown in Figure 9.4.
Figure 9.4
Note that, given the domain, /(£) begins at (0, p) on the vertical axis
and asymptotes to the t axis as t increases. Areas under the curve
correspond to probabilities, and we can use definite integration to
confirm that the total area under the curve is one. The total area is
given by the improper integral
1
rt = oo
/ pe-^“d£ = ( - =°
Jt = o
= (0 + k) - ( - 1 + fc)
= 1
The area under the curve between two values of t is the probability
that a customer’s service time lies between those limits. If in a

particular case the average service time is 2 minutes, this corresponds
to the mean rate of service p = 0.5 per minute. We can use definite
integration to answer quality of service questions such as:
1 What percentage of customers are not served in under 3 minutes.?

2 For what percentage of customers will the service time take longer
than the average.?
The answer to 1 is given by the area under the curve beyond t = 3.

This is
ft —CO
0.5t^t=3
Jt=i
= (0)-(-0.2231)
= 0.2231
so the percentage of customers not served in under 3 minutes is

22.31%. Now consider question 2. The mean service time is 1/p, so the
probability of a service time taking longer than 1/p is
ft = CO
/ pe-^^'dt = ( - = Vm
= 0 + e~^
= 0.3679
So in any simple queueing system 36.79% of customers experience a

longer than average service time. Exponential functions allow a wider
range of functions to be used in the types of differential equation
discussed in Chapter 8. For example, consider:
12x^ + ^- 3e~^* = 0
dx
given the boundary condition that y = 11 when x = 0. Rearrangement

produces
dx
so that
y= (3e-3* - 12x2)dx
Thus the general solution is
y = - 4x^ + k
At the value of x = 0,
y = -l + k= 11

so k= 12, and the particular solution is therefore
z/ = -g-3^-4x^ + 12
As a further example, consider the particular solution to the second-

order linear differential equation
given that when x = 0, i/ = 6 and dy/dx = 5.1. The equation

rearranges as
which equals 5.1 when x = 0, so that ki must equal 5. Insertion of this

value for ki into the derivative and subsequent integration results in
y = x^ + + 5x + k2
and, since y = 6 when x = 0, the value of ^2 must be 5. The particular
solution is therefore
y = x^ + + 5x + 5
Exercises 9.3
(i) J e^^dx
(ii) /-e“°-^'^^dx
(iii) f ne^^dx
(iv) /(dx^ + 3e^ ■^^)dx
(i) / xe^^dx
(ii) /(5x-2)e’0^'-^^-6dx
3 Evaluate the following definite integrals:

x=4
(i)
L x=2
e°-^^dx
Jx = 0

4 With the given conditions, find particular solutions to the
following differential equations:
(i) 5x + 0.5$^-2e2^ = 0
dx
where y = 2e^ when x = 1.
(ii) -0.1 8x2+$^ + 20e-° 2^ = 0

dx
where y = 100(e“2 + 1) when x = 10.
9.4 An exponential relationship between variables x and y may take the form
Natural
x = ay (9.14)
logarithmic
functions which could be read either as
X is the number that results when the base a is raised to the

power y
or as
y is the power to which the base a must be raised to give the

number x
A shorthand way of making the second statement is
y is the logarithm to the base a of the number x
This is written as
t/ = logflX (9.15)
and is defined for « > 0 and a ^ 1. (9.14) and (9.15) describe a single
relationship between x and y. They view different aspects of the same
connection and are inverse functions. Which form it is best to use
depends on the task in hand. In this section we will be concerned with
logarithmic functions and their relevance to business and economic
concepts.
Consider some examples of logarithmic relationships between two
variables, and their restatements in exponential form:
y = l0g2 1 means 2» = 1 therefore y = 0

y = logs 3 means = 3 therefore y= i
y = logio 100 means IQy = 100 therefore y= 2
y = log2 0.25 means 2« = 0.25 therefore y = -2
y = logg 31.5 means = 31.5 therefore f/ = 3.45
y = log! 8 means f = 8 therefore y = -3
Logarithms (logs) to the base e are natural logarithms and there

are two slightly different forms of notation, y = log,, x is also written as
y = In X

Logs to the base 10 are common logarithms, which were often used
in manual arithmetic before electronic calculators.
Now consider the graphs of logarithmic functions. Figure 9.5 shows
the graph of y = log„ x for a> 1. Note that the domain of the function is
X > 0. The necessity of this can be seen from (9.14) - given positive a,
there is no number, y, to which power a can be raised to give a negative
outcome (x). The graph is always increasing, but at an ever-decreasing
rate. Although there is no limit to the value that y can take, the
declining slope is such that ‘astronomic’ increases in x are needed to
produce even small changes in y In the case of the natural logarithm
function, with a scale of centimetres on each axis, in order that y
reaches a height of 50 cm the corresponding x co-ordinate is in excess of
5000 light years along the horizontal axis! This phenomenal ‘flatness’ is
the view through the inverse function of the steepness of ^
Now consider the graph of y = logg x when 0 < a < 1. This is shown
in Figure 9.6. Note that the log function when 0 < a < 1, defined only
for X > 0, is always-decreasing at an always decreasing rate. The
following properties of logs are an alternative way of stating properties
of exponents. Where m and n are positive numbers,
(i) logg(ni.n) = loggm + loggM (9.16)
(9.16) states that the log (to any base) of the product of two numbers is
the sum of the individual logs. So with the product of 4 and 8,
y
log2 (4.8) = log2 4 + log2 8

= 2 +3
= 5
which is another way of saying that
2^ 2^ = 2®

Figure 9.6
Equation (9.16) was the basis of calculation with logs - using the
properties of exponents to transform multiplication into addition.
YYi
(ii) log„ - = logfl m - log„ n (9.17)
Equation (9.17) states that the log of the quotient is the difference
between the logs of the two numbers and enables division to be
transformed into subtraction. For example,
243
logs = logs 243 - logs 27
= 5-3
= 2
which is another way of saying that
(hi) log^m^ =plog„m (9.18)
Equation (9.18) can be verified from (9.16). For example,
log,, XY^ = logfl mmm

= logfl m + log„ m + log„ m
= 3 log„m
This rule, in conjunction with (9.16) is useful in re-expressing a non¬

linear relationship in linear form. For example, suppose that x and y
are related by
y = x^

Taking logs of both sides (to the base e) gives
In 1/ = In (x") = n In X
so the cubing function
y = x^
can be replaced with the linear relationship
In = 3 In X
The simplification achieved shows up in a side-by-side comparison of

the two graphs, as shown in Figure 9.7.
Figure 9.7
The connection between whatever is represented by x and y is the

same in each case, but the log-linear form is more convenient for
measurement and testing. This point is reinforced for more complicated
functions such as
z = AxV (9.19)
This a Cobb-Douglas production function, in which z represents

output, X represents labour input and y is a measure of capital input at
a high level of aggregation (sector or national level). While (9.19) has
degree a + P, it is linear in the logs of the variables. Taking logs on
both sides.
In z = In A -F a In X -1- /3 In y
which is more convenient for estimation since linear regression can be

used.
Exponential and logarithnnic functions 369

The last property of logs considered here is:
loghtn
(iv) Change of base: log^ m = (9.20)
logfcfl
The change of base formula is useful in differentiating exponential

functions with bases other than e. An example of the use of this formula is
log, 100 4.6052

logio 100 =
log, 10 2.3025
Now consider the application of logarithms to compound growth. If a

population is growing at 2% p.a., how many years does it take for the
population to double.^ The answer is not fifty years. Suppose that the
original population is N. After x years of growth at 2% the population is
N(l+0.02)^ (9.21)
and the required value of x - the doubling time - is found by

equating (9.21) to 2N. So
N(1 +0.02)* = 2N
and after division by N,
(1 + 0.02)* = 2
Using property (iii) and taking the logs of both sides with (say) a base
of 10,
logio (1 + 0.02)* = logio 2
so that
X logio (1 +0.02) = logio 2
that is,
logio 2
^ logio (1.02)
0.3010
“ 0.0086
« 35
A population growing at an annual rate of 2% compound doubles

every thirty-five years. Note that doubling time is independent of
population size (although very small populations will not show
proportional year-on-year growth). At the same rate of compound
growth, a population of 5 billion will add 5 billion people in the same
time that a population one-tenth the size adds 500 million. Similar
questions can be addressed in other areas. If an economy has a sustained
real growth rate of 3%, with a stable population the per-capita gross
domestic product will double in x years, where
(1.03)* = 2

Using base e logarithms,
In 2
X =-
In 1.03
_ 0.6931
“ 0.0296
^ 23 A years
Suppose that an investor places a sum on deposit at a guaranteed

net 10%. How long will it take for the investment to double.? This will
be X years, where
In 2
X =-
In 1.1
0.6931
“ 0.0953
7.27 years
Now consider a borrower who takes out a capital loan at 30%

compound p.a. In the absence of repayments (and further borrowing)
the outstanding liability will double in x years, where
In 2
X =-
In 1.3
0.6931
“ 0.2624
^2.64: years
So at such an interest rate a repayment ‘holiday’ of under 2 years 8

months sees a doubling of indebtedness. Far higher rates are applied to
personal loans from fringe financial companies and ‘loan sharks’. Similar
calculations can be made under continuously compounded interest. For
example, doubling time at an interest rate of 15% is given by
gO.isx ^ 2
Taking logs to base e,
O.lSx = In 2
so that
0.6931
X = = 4.62 years
0.15
The calculations can also be applied to decay models. For example,

suppose that a company currently has 20% of the market for a
particular product range, but each year its market share is 9 7% of the
share in the preceding year. How long will it be before the company’s
market share is down to 10%.? The relevant equation is
20(0.97)* = 10

so that
(0.97)* = 0.5
xln0.97 = ln0.5
In 0.5
In0.97
-0.6931
^ “ -0.0305
22.7 years
So in the unlikely event of stable market conditions for that length of

time, the company’s share of the market would take almost 23 years to
halve. Note that the time to halving at 3% per annum contraction is
not the same as the time to doubting at 3% per annum expansion.
Suppose that the quantity of a radioactive contaminant in an
environment decays exponentially at the rate of 0.01% per annum.
How many years must elapse before only half of the original quantity of
pollutant remains.^ The answer is the half-life of the substance, and is
given by x, where:
g-o.oooix ^05
Taking logs to base e,
-0.0001x = ln0.5
so that
-0.6931
" = 300001 =
So the half-life of the radioactive material is almost 7000 years. Note
that in contrast to exponential growth, exponential decay to half of an
original quantity takes significantly longer than under a linear process,
where a fixed percentage of the original amount is subtracted each
period - as in straight-line depreciation. For example, in the above
case, if the volume of the contaminant was reduced by 0.01% of its
original quantity each year, then the linear equivalent ‘half life’ would
be given by
n = —— = 5000 years
0.0001
Exponential (true) half-life is therefore longer than the linear equivalent
by the ratio
— logg 0.5
1.3863
05
that is, by some 38.63%.

Exercises 9.4
1 Express the following logarithmic functions in terms of on
exponential relationship between the variables:
(i) y = logiox (ii) x = log y (iii) x = log^y
5
2 Express the exponential relationships below as logarithmic

functions:
(i) y= ^ (ii) x = ay
10 (iii) p = ^ 2
3 Find the value of y when:

(i) y = log (ii) y = log
525 (iii) y = Iog20.125
48
(iv) y = log 1 (v) y = logo, 1

7 25 6 (v') y = In
4 Evaluate:
(i) log + log l
28 2 6
Iii]
(iii) logo ,582
5 What is the doubling time for:

(i) a population growing at 4% per annum;
(ii) an investment returning % p.a. compound;
8
(iii) a debt on which interest is charged at % p.a. 20
6 What is the doubling time under continuous compounding at:

(i) % (ii) 10 % 20
7 How long would it take for a company's market share to halve

when in any one year its market share as a percentage of the
previous year is:
(i) 98% (ii) 95% (iii) 90%
8 What is the half-life of a radioactive substance which decays
exponentially at the rate of one-thousandth of per cent per 1
annum?
9.5 We shall work towards the derivative oi y — logg x by making use of the
The derivative of facts that:
natural logarithmic
• We know how to differentiate the natural exponential function.
functions
• The inverse function rule applies since the logarithm function is
the inverse of the exponential function.
Now y = logg X can be re-expressed as x = e^, for which
dx
From the inverse function rule,
dy ^ 1
dx dx/dy
__ 1
~ ey

so
d( loge X) ^ I
(9.22)
dx X
It is surprising that the derivative of log^ x is the reciprocal function,

But the result answers the question of what is the integral of 1/x?
We now know that
/.->dx = log,* + ^.
We will return to integration later. For the moment we take up some

consequences of and variations on (9.22).
(i) Higher derivatives of ly = In x. Since
dx X
dx^ x^
Higher-order derivatives are then obtained in the normal way.

(ii) For y = m logg x,
dy m
dx X
Thus, recalling the alternative notational schemes, given
/(x) = 171nx
fix) = 17x-^
(iii) For y = \n x"’,
dy m
dx X
This follows from property (9.18). Therefore, since log^x'" = mlog^x,

their derivatives must also be identical. Consequently, if
then
dy 3
dx X
Now consider the derivative of
y = x\nx (9.23)
This can be obtained by use of the product rule.
dx X
= 1 + In X

Equation (9.23) is a special case of the product
y = fix) In X
for which
^ =f{x)- +(inx) fix) (9.24)

ax X
so, for example, if
y = x^ In
then the derivative will be
^y 31 T 21 I
— = X — + 3x In X
dx X
= x^(l + 3 Inx)
The quotient rule can also be used to find the derivative of

expressions with a logarithmic component. For example, if
logfX
y = ^—
2x
Using the quotient rule,
dy 2x(l/x) — 2 logg X
dx 4x2
_ 1 — logg X
2x2
Now consider an expression having both an exponential and a

logarithmic component:
y = e('"lnx)
Since this function is a particular example of the form y = for

which the derivative is f'{x).^^^\ we could correctly write
dx X
But a dramatic simplification is possible. Consider the original function.

Taking logarithms of both sides produces the result
Iny = ln[e("’
7
= m Inx
= ln x"’
so the relationship between y and x is, in fact,
1/ = x'”
for which the derivative is mx"'“^ and to which our original

derivative simplifies. Note that the case of m = 1 shows that = x.

Consider the derivative of composite functions involving logarithms:
for example
y = loge0(x)
Using the chain rule,
dx dg dx
1
g (x)
g{x)
g'jx) (9.25)
g(x)
Note that whatever the complexity of the expression of which the

logarithm is taken, the derivative of the ‘log/ part is the reciprocal of
the expression itself. Then, as in any instance of the chain rule, this
derivative is multiplied by the derivative of the ‘inner’ function g(x).
Consider an example. Suppose that
y = loge
dy 3x^ _ 3
dx x^ X
which confirms what we already knew about the derivative of

y = logg x"* and had obtained by other means. Now consider the
derivative of
y = logg 6x
Application of the chain rule produces
^=A
dx 6x
_ 1
X
It is clear from this example that the derivative oi y = logg mx is 1/x

for any real value of m. But the fact that the function y = logg 6x has
the same rate of change throughout the domain as y = logg x is not
altogether obvious. It may help to give an intuitive feel for the result by
using property (9.16), from which we may write
logg 6x = logg 6 + logg X
of which the derivative of the right-hand side is
As a further example, consider
1/ = In (x^ + 1)

for which
dy 2x
dx x^ + 1
The ratio of the derivative of a function to the function itself, as in

(9.25), is called the logarithmic derivative. Given y = g{x),
g'jx) ^dy/dx dy _d{\ny)

g{x) y y dx dx
There is an important relationship between logarithmic derivatives

and the concept of the elasticity of a function. This is important in
economics, and we will consider elasticity in the following section.
Exercises 9.5
1 Find the derivative dy/dx where:
(i) y = 20 log^ X (ii) y = log^ x^
(iii) y = x^ In X (iv) y = (4x — 2x^) In x
2 Find f'(x) where:
(i) f|x) = In x^ (ii) f{x) = In lOOx
(iii) f(x) = logg (5x^ — 6x^ — 7x^)
3 (i) Find the derivative of
logg X
y = (x2+x)
(ii) Fine the derivative of

X
y= loQe X
4 Find the required derivative in the following coses:

(i) The third derivative of y = In x.
dV
(ii) dx2 given that y = x logg x
(iii) f'|x given that f(x) — (logg xf
9.6 The derivative of a function measures its rate of change in terms of x

Elasticity and y. The elasticity of a function measures the proportional rate of
change. In finite terms, the elasticity of a function is the ratio of the
proportionate change in the dependent variable to the proportionate
change in the independent variable. That is, if x changes by an amount
Ax and the resultant change in y is A^, then the proportionate
changes in the two variables are
Ax Ai/
and
X y

so the increments in the two variables are taken as a proportion of
their current values. Elasticity is then calculated as the ratio of these
ratios:
elasticity = ^
which two-tier ratio reduces to
elasticity = (9.26)
y/\x
The point elasticity of the function replaces the ratio of increments in

(9.26) with the derivative:
X dy
E yx — (9.27)
y dx
The main advantage of elasticity as a measure of the responsiveness

in y to changes in x is that, unlike the slope, it is independent of the units
of measure of either x or y. This is most easily seen in the ratio (9.26).
For example, any change in the units of y by some factor c also scales
the increment in y by the same factor, which therefore cancels out in
the elasticity ratio. It would not cancel in the ratio of increments alone,
the limiting value of which is the slope of the function. The elasticity
ratio (9.26) is homogeneous of degree zero in x and its increment, and
in y and its increment. The value of the dimensionless property of Eyx is
evident in respect of the measurement of the sensitivity of the quantity
demanded of a product, y, to changes in the product’s own price, x.
This is the elasticity of demand, for which it is the usual, though not
universal, practice to attach a minus sign to the formula. This is done
so that a positive value results for Ed, the own-price demand elasticity.
With p — price and q = quantity, the formula is
p^
Ed = - (9.28)
q dp
It is meaningless to describe demand for a product as price-sensitive

because the slope of the demand function y =f{x) is f'{x) = —50. The
value of slope could be reduced to —0.5 by a change in units of
measure of price from pounds to pence. The absolute value of slope can
be set to any value by selection of units. It is meaningful, however, to
say that the quantity demanded of a product is sensitive to price
because the elasticity of demand is greater than one - which is the
watershed value for elasticity. One is a significant figure for Ed since
values greater than one mean that the proportionate change in
quantity demanded exceeds the proportionate change in price. Consider
own-price elasticity in the case of a linear demand curve. There is a
minor inconvenience here, as the relationship between price and

quantity is conventionally stated with price as the dependent variable
and we start with the equation
p a — bq
in which a> 0 and b > 0. Note that when q = 0, p = a, and when

q = a/b, p = 0. Demand elasticity at any point on the curve is given by
p ^
Ed
q dp
where
-1
dp \dqj b
Thus
Ip {a-bq)
= —u-
(9.29)
b q bq
Note from (9.29) that, while slope is invariant, elasticity is not constant
along a straight-line demand curve. The range of variation is from oo
when q = 0 to 0 when p = 0. The graph of the demand curve is shown
in Figure 9.8(a) and that of demand elasticity in Figure 9.8(b).
Figure 9.8
The graph of the elasticity function is one branch of the

rectangular hyperbola with asymptotes at x = 0 and y = —1. Elasticity
' of demand is a valuable concept in assessing the effect of volume or
price changes on the value of sales. For example, with quantity as the
decision variable
R=pq

where R = value of sales (total revenue) and from which
(9.30)
From (9.30) it is clear that the necessary condition for a maximum of

revenue is satisfied at the output level where = 1, i.e. where demand
is unit elastic. Using price as the decision variable gives:
— —Q Ed + q
= q{l-Ed)
from which it is clear that where demand is elastic (Ed > 1) a price
rise will, other things being equal, result in a revenue decrease. But
when demand is inelastic (Ed < 1) the strategy of generating more
revenue by increasing price will succeed. The question arises as to what
point on the linear demand curve corresponds to unit elasticity. From
(9.29) this will be where
a — bq = bq
i.e. where
q = a/2b
which is the value of q half-way to the q intercept and which

corresponds to a point half-way along the demand curve.
Consider a numerical example. Let demand conditions be given by
p = 100 — 2q
for which we will now find the following:
1 The elasticity of demand at q = 20.

2 The point where elasticity = 0.25.
Addressing task 1, first note that when q = 20, p — 60. Restating the
demand equation with q in terms of p gives
q = 50 — 0.5p
from which

so that
Ed ( - 0.5) = 1.5
20
The most convenient way to address task 2 is to use equation (9.29),
from which
100 - 2q
Ed 0.25
2r~
so that
100-2q = 2q(0.25)
2.Sq= 100
q = 40
We saw that a straight-line demand curve had varying elasticity. It is

interesting to note that constant unit elasticity corresponds to the
demand curve given by the rectangular hyperbola
A
P =-
q
From the rearrangement of this relationship we obtain:
dq A
dp
so that
p A
Ed
qp^
PQ
Therefore
Ed = l
Along such a demand curve, price and quantity decisions have no

bearing on total revenue, which is fixed at £A.
What is the relationship of logarithmic derivatives and elasticity? To
establish the link, first observe that given the relationship
P = /(g)
we know that
dlnp _f'iq) _ dp/dq

dq f{q) P
and that
ding 1
dq q

So, as a cumbersome ratio,
ding 1
dq ^ q
dlnp dp/d(j
dq p
for which the left-hand side reduces to
d\nq
d\np
while the right-hand side becomes
p ^
q dp
Thus the equation of the two means that
_ ding
(9.31)
^ dlnp
This result is useful since there are demand curves for which it is
easier to find elasticity through (9.31) rather than using (9.28). For
example, if
p = Aq^
taking logs of both sides produces
log,p = loggA - 21oggg
from the derivative of which the elasticity of the original function is

readily obtained, since
dIogeP ^ _2
dlog,q
the negated reciprocal of which is demand elasticity. Therefore
Ei=i
This outcome may be confirmed by using (9.28).
Exercises 9.6
1 Given the demand-curve relationship p = 500 - 2.5q where
p = price, find the value of elasticity of demand where:
(i) p = 50 (ii) p= 100 (iii) p= 150
2 Given the demand relation p = 1000 — 4q:
(i) Find the level of output for which elasticity of demand is 3.
(ii) Find the price at which the elasticity of demand is 1.5.
(iii) Find the value of price for which total revenue is
maximized.

3 Find the elasticity along the demand curve
100
4 Suppose that a demand curve is given by

_ 1000
^“q+10
Find the elasticity of demand when:

(i) q = 2.5 (ii)q=10 (iii)q = 40
9.7 Here we examine straightforward cases of integrals involving

Natural logarithmic expressions in the antiderivative or the integrand. We begin
logarithmic with a result familiar from earlier work:
functions and
integration (i) -dx = logg x + k
J X
for which further elaboration is not required. We know also that
/ —dx = m log„ x + k
= logg x"" + k
So, for example,
r9
/ —dx = 9 In x + k
J X
= In x^ + k
For a rational function in which the numerator is the derivative of the

denominator, the integral is the logarithm of the denominator:
(iii) dx = In/(x) + k
So, for example.
f (Sx^ - 4x + 4) 3 2 ^ ^ 7
Now consider the indefinite integral of the natural logarithm function.

This is
(iv) J logg xdx = X logg X — X + k
= x(logg X — 1) + k
We shall not prove the above result, but it can be confirmed by

differentiation. Starting from the function
y = X logg X — X + k

and using the product rule on the first term, the derivative is obtained
as
^ = X- + (log,x) 1-1
dx X
= log, X
which confirms the result. An extended form of (iv) can be obtained as

follows:
/log,x”d* = /„log,xdx
= „/log,xd*
The rule may therefore be stated as
(v) J log, x"dx = nxflog, X - 1) + k

For example,
J In x^dx = 7x(ln x — 1) + k
Note that
/ dx
—— = In (In x) + k
xlnx
which can be confirmed by differentiation using the chain rule. Letting
y = In (In x) and In x = u
so that
y = In (u)
and
dy dy du
dx du dx
so therefore
dx u X
1
xlnx
For example, using (vi),
[ — dx = 5 In (In x) + k
J xlnx

We will work through two examples of definite integrals. First
consider the definite integral
The indefinite integral is
logp (3x^ — lx) + k
so that the definite integral evaluates as
[logg (3x^ — — [logg {3x^ — 2x)Y^^ = logg 280 — logg 21
Note, by way of revision, that this result will simplify to
Now consider
8x(logg X — 1) + k
so the required definite integral is
[8x(loggX- 1)]^=^ - [8x(loggX- l)]*=2-5

= 40(logg5-l)-20(logg2.5-l)
« 24.3775 - (- 1.6742)
26.0517
There are far more complicated cases of integration involving

logarithmic functions than these examples. Section 9.9 discusses the
integration of logarithmic functions to bases other than e.
Exercises 9.7
Find the following indefinite integrals:
2 Find:

3 Obtain the following indefinite integrals:
(i) Jin x*^dx
(ii) Jin x'^dx
4 Find:
10
(i)
/ X log eX
dx
dx
(ii)
2x In
5 Evaluate
f*x = 5 (6x2 _ 5^1
I'l L
^x= 10
(ii) / log^ x^°dx
Jx = 4
9.8 In this chapter we have seen how integrals can be found for certain
Integration by exponential and logarithmic functions. In Chapter 8 we concentrated on
parts the integration of power functions. In this section we consider a technique
of integration that is useful for functions containing terms which are the
product of power and exponential components, although the usefulness of
the rule is not confined to this situation. The method is equivalent in
integration to the product rule for differentiation, and is known as
integration by parts. The product rule stated that if the dependent
variable, y, could be expressed as the product of two functions of x, as in
y = f{x)g{x)
then one way to carry out the differentiation was to use the fact that
dx dx ^ dx dx
To see how an equivalent approach can be useful in integration, first

integrate both sides with respect to x. The result is
This relationship is used to best advantage in the rearranged form
/0d/=/fl- //da (9.32)

which is known as the by-parts formula. The expression to be
integrated is interpreted as f g df, and the by-parts rule is used to
express this in a way that is easier to integrate. Examples will illustrate
the use of the rule. Consider
This expression will be taken as the g df component of (9.32), the use

of which will also require expressions for / and d^. Thus, with
it is clear that
fix) = e*
and with g(x) = x,
^= 1
dx
So, using (9.32),
dx
= xe* - e* + fc
= e*(x - 1) -I- k
Now consider
j 0.5xê*dx
for which we again set
dx
It follows from this choice that
/(X) =
and with
g{x) = 0.5x^
it is the case that
dg
= X
dx
So, in the form of (9.32) we now have
/0.5AM. = 0,5.^.'-/..-d.
= 0.5xê^ ~[e’‘(x-l)] + k
= e"(0.5x^ -x + l) + k

Now consider
J xe^ ^^dx
for which we set
^ _ pO.Sx
dx~
As a result
fix) =
and since
g{x) =
it is the case that
The integral can now be written as
/ xe^'^’^dx = 2xe'
,0.5x
Jle- Sx
dx
2xe,0.5x - +k
„0.5x
- 4) + /c
Now consider an example in which integration by parts is used

twice. For
J 2xê°'^^’'dx
we again set the df derivative equal to the exponential component of

the integrand
d/ _ 0.25x
dx“
Thus
fix) =
and with
gix) = 2x^
it follows that

The integral can therefore be written as
in which integration by parts is applied again to the ffdg term on the

right. In this case
= - 256e®-2^*
so the required integral is
8jc2g0.25x _ 64xe0-25x ^ 256e°-^5* + k
which is more neatly rearranged as
8(x^ — 8x + 32)e°'^^* + k
Now consider an example of definite integration:
/'X = 12
lOxe 0.4X-0.8 dx
Jx = 2
First find the indefinite integral. Setting
„0.4x-0.8
dx
then
fix) = 2.5e°-^*-°-*
and, with
gix) = lOx
dg
= 10
dx
The indefinite integral is then
J 10xe°'^’'“° ®dx = 25xe®-^’'"°-®25e0-^’'-0-^dx

0.41-0.8
25xe' - 62.5e'0.4j:-0.8 + k
Ignoring the constant of integration, the definite integral is

J
(25xe“'^’'““'® - 62.5e®-^*"°-®f- (25xe“-^’'“°'^

- 62.5e0'^^-°-^r=^
= 300e^ - 62.5e^ - (50e° - 62.5e°)
= 237.5e^ + 12.5
12979.56

Exercises 9.8
1 Use integration by parts to find the following integrals:
(i) f 5xe^ dx
(ii) f2xê^dx
2 Find:
(i) f xe^^ dx
(ii) f 5xê°-^^ dx
3 Evaluate the following definite integral:
200xe°^^-°^dx
9.9 It is convenient to consider logarithmic functions first. Given

Logarithmic and
exponential y = loga
functions to bases There are two stages in obtaining the derivative of a logarithmic
other than e function to a base other than e. These are:
1 Change the base to e.

2 Differentiate
Putting this process into effect, from the change of base formula
(9.20)
logfl X =
loggfl
It follows that, since 1 / logg a is simply a constant.
d( loga ^ 1 1
dx loggfl X
So, for example, given
y = log2 X
then
dy ^ 1
dx X In 2
I
“ 0.6931X
The second derivative can be found in the following way: writing the
first derivative as
(—V-i
dx \ln 2 )

from which
dx^ \ln 2 )
_ 1
x^ In 2
Now consider
y = logio
^ loge
log, 10
so that
di/ _ 1 3x^
dx logg 10 x^
3
X logg 10
1.3
X
In the case of
/(x) = logs (8x^ + 7x + 6)

_ logg (8x^ + 7x + 6)
logg 8
so the derivative is
1 16x+7
/'(X) =
logg 8 (8x2
Integration of logarithmic functions to other bases can proceed along

similar lines to differentiation - first change the base to e, then
integrate. For example,
= 0.4343x(logg X — l) + k
As an illustration involving a definite integral, consider

Changing the base produces
i:
15 ^
log,
fx= 5 log, 2
which will be
x = 15
• 5
4x(log, X - 1) ---4x(log,x - 1)
log, 2 log, 2
/20x(log, X - 1)V
log, 2
300(log, 15 - 1) - 100(log, 5-1)
log, 2
451.4713
0.6931
651.34
For exponential functions to bases other than e, consider
y = a’'
How can the derivative be found.? It will be recalled that
means that
X = log« y
and from what we have already seen
dx 1
dy ylog,a
Invoking the inverse function rule,
(9.33)
= fl* log, a
So that, as with the natural exponential function, the exponent of the

base remains unchanged in the derivative, and each successive higher
derivative would simply multiply (9.33) by a further log,a. Thus
So, applying this result to the case of
i/ = 2^
= (0.6931)2^

and
d^M T
3-| = (0.6931)^2’=
= (0.4805)2’^
As further examples, consider
fix) = 10"
for which
/'(x) = 10"log,10
^ (2.3026)10"
and for which
f"ix) ^ (2.3026)^10"
(5.3019)10"
Consider also
fix) = (1)'
for which
f'ix) = (l)"log,(l) ^ -0.6931(1)"
from which it can be seen that the slope of the function is always
negative but is, in absolute terms, a decreasing function of x. Note that
where the base, a, is less than one, the higher derivatives will alternate
in sign. In the present case
rW = [log, (1))"(!)“ » 0.4805(1)*
and
/"'(x) = -0.3330(1)"
Now we consider an example of a derivative that requires the use of

the product rule. If
fix) = (x^ + 5x^)4"
then
f'ix) = (x^ + 5x^)(ln4)4" + 4"(3x^ + lOx)
Now consider the derivative of an exponential function with a base other

than e, in which the exponent of the base is a function of x. That is
y=
The derivative can be approached in the following way. First take

logarithms to the base a of both sides:
logfl y = fix)

Using the change of base formula for logarithms,
ôgeU
logfl y = = f{x)
log, a
so that
logeh = {log, a)f{x)
Differentiating with respect to x,
5!*!^=i^=(iofc«rw
dx y dx
so that
^ = {\og,a)f'ix)y
dx (9.34)
= (log,
So, with the exception of the coefficient logg a, the outcome is as for the
derivative of e^. Therefore, given
. r45C^ -7x + 8
y — ^
using (9.34),
^ = (logg 5)(8x- +^
dx
= (1.6094)(8x - 7)54*"-7x + 8
= (12.8755X- 11.2661)5^*'“^^ + ®
A derivative that requires the use of the product rule is
y =
for which
^ = (x5)[( In 4)(2x)4’='] + (5x^)4^'

dx
Now consider integration of exponential functions to bases other than e.

To begin with, take the simplest form for the integrand:
a*
+k (9.35)
logg a
This result can be confirmed from the differentiation formula (9.33).

For an example of the use (9.35), consider the indefinite integral
f 10*
/ 10*=--—+fc = (0.4343)10*+ /c
J logg 10
Putting (9.34) into reverse gives the conclusion
J /'(x)(/^*^dx =
(/(*)
loggfl
+k

So, for example,
265C^ +8x —9
/ {I2x + + = -^-— + k
logg2
= (1.4427)2^"' + ®"-^+ k
And, as an example of a definite integral

=2
(4x-
/
»/x
'x == 0
a2x^ —x+1
+k
In 4
and the definite integral will be
^(47 _ 4) ~ 11816
In 4
In earlier sections we saw examples that highlighted one sense of the

term ‘exponential growth’. Let us look more closely at what exponential
growth really implies. Exponential growth means that the rate of
change of a quantity is proportional to the value of the quantity at any
point. That is
^ % (9.36)
dx
Equation (9.36) is a differential equation, the solution of which,

constant of integration apart, is an exponential function. For the
natural exponential function k= 1, while for y = a’‘, k = logâ. So long
as k> 0, not only does y increase with x, but the rate at which it
increases also rises with x. It can be shown that any quantity growing
exponentially will ultimately outgrow any quantity growing in
proportion to fixed powers of x. It is an important property of an
algorithm that the time taken to solution grows as a power of the
problem size. That is, it is desirable that a computation procedure
should be a polynomial time algorithm rather than an exponential time
algorithm.
Exercises 9.9
1 Find the first derivative of the following functions:
(i) y = log5X (ii) y = logo2X
2 Find the required derivative in the following cases:
(i) f'(x) where f(x) = logôX
(ii) f"(x) where f(x) = logio->f
(iii) f'(x) where f{x) = logg X 2
3 Find f'(x) where:

(i) f(x) = loggx^ (ii) f(x) = logo, X°-^
5

4 Find the derivative of the following functions:
(') y=
(ii) y = log2(2.5x^ - + 5x^)
(i) /log2xdx
(ii) /12logi5X^dx
6 Evaluate:
/'X= 12
/ 2log,o x^dx
Jx = 4
7 Find the required derivative for:

(i) f'(x) where f(x) = 10-^
(ii) P'(x) where f(x] = 5^
(iii) f'(x) where fjx) = (2x^ - x^)
8 Find the first derivative of the following functions:
(i) y=10^' (ii) y = (iii) y = (x2)5^'
(i) / 5^dx
(ii) /(7r)6^dx
(iii) /(10x-8)105^'-s^dx
10 Evaluate:
rx = 2
/ (4x-1)12^^ -^dx
Jx= 1
9.10 Many models in economics and business are based on exponential

Aggregate sales growth or decay, of which the simplest form is
curves
y = a’‘
where a value of a greater than one corresponds to growth, while

values in the interval 0 < a < 1 represent decay. But the function
y = fl* itself has only a single parameter, a, to adapt to the situation
being modelled. This does not give sufficient flexibility to match
conditions where the rate of growth itself changes over time. Take a
marketing example. In the case of a new consumer product from an
established company with a well-resourced launch, sales may get off to
a rapid start, after which growth slackens. For fashion items, beyond a
certain point, sales per period may approach zero. Examples of products
likely to achieve a rapid start to sales are the latest recording by an
established pop group and an ‘instant’ book on a sensational but not
lasting issue. Figure 9.9 shows the graph of possible aggregate sales of
such products, plotted against time since launch.
Figure 9.9 is a modified growth curve and has the equation
y = r(l-fl-'’") (9.37)

Figure 9.9
r > 0, fl > 1 and b > 0
In (9.37) X could represent time from launch or a measure of

marketing effort. The two interpretations can coincide - as when x
represents the number of days for which a marketing campaign runs.
As can be seen from the figure, the parameter T is an upper bound on
total sales and, since
^-bX ^ Q y. ^
y has this value as an asymptote. T is known as the market potential

or saturation level of demand. Given the ceiling set by the value of T,
the values of a and b determine how slowly or rapidly this ceiling is
approached. A selection of values of (9.37) is shown in Table 9.6.
While for each pair of values of a and b the curves have the same
general shape as in Figure 9.9, the initial steepness and the sharpness
of the ‘bend’ depend on the values of a and b (and on the effectiveness
of the company’s marketing). When the base a = e the function is
sometimes called the modified exponential. The values of the
parameters a and b would be selected to match as closely as possible the
sales characteristics of the product under study. But even with careful
selection of parameters, the modified growth model will not fit a
common pattern of aggregate sales known as the S-curve demand
function, as shown in Figure 9.10. Incidentally, the term ‘demand
function’ is used in the general sense of quantity dem’anded expressed
as some function of any independent variable, rather than the specific
economists’ usage of q =/(p) in a given time period.
The S-curve pattern of demand sees sales at first responding slowly
to increases in the independent variable, but with, up to a point, an
increasing rate of growth. The maximum rate of growth of aggregate
sales occurs at the point of inflection. Subsequently, growth of
aggregate sales declines, with the market potential as asymptote - as
with the modified growth curve. If x measures sales promotion effort
(length of campaign, expenditure, etc.), the y intercept is sales with zero
effort (or default level). Then for positive x the first part of the S curve

Table 9.6 X
a b 1 10 50 100 1000
1.02 0.2 0.40 3.88 17.97 32.70 98.09

1.2 0.02 0.36 3.58 16.67 30.56 97.39
1.2 0.2 3.58 30.56 83.85 97.39 100
1.2 0.6 10.36 66.51 99.58 100 100
1.2 1.0 16.67 83.85 99.99 100 100
1.6 0.2 8.97 60.94 99.09 99.99 100
2.0 0.2 12.94 75.00 99.90 100 100
e 0.2 18.12 86.47 100 100 100
e 1.0 63.21 100 100 100 100
Figure 9,10
shows increasing marginal returns to sales promotion input. Diminishing

returns set in after the inflection point. With x representing time, an S-
curve pattern of aggregate sales against time might be exhibited by a
good popular music recording by a relatively less-well known group.
Initially, the record-buying public becomes aware of the record
gradually, but sales begin to take off after sufficient ‘plays’. With x as
time, the intercept may represent orders taken prior to launch. If this is
not applicable, the whole function could be shifted down. The
characteristic S-curve shape can be produced in a number of ways. In a
marketing context two relevant expressions are the Gompertz
function and the logistic curve. The Gompertz function is
y = Ta^^ (9.38)
in which the exponent of a itself has the exponent x, and where in

order to generate the S-curve shape
0 < fl < 1 and 0 < b < 1
Sensible results are produced when the value of b is near the upper
end of its range, while a is less than 0.5. The upper asymptote is at T
(representing market saturation, so T > 0) and the y intercept occurs at
y = Ta. The function has the x axis as an asymptote as the value of x

becomes large and negative. In view of the fact that the decision
variable, x, is the exponent of an exponent, it may be more convenient
to work with the log of sales. The result of taking logs on both sides of
(9.38) would be
log, y = log, T + b* log, a
from which the derivative dy/dx could be more easily found in the
following way:
dx
=i^
y dx
so that
^ = Tfl*’" [b*(ln b)(ln a)]
Suppose that a marketing department is considering the length of an

advertising campaign for a new product. With x being the length of
campaign in weeks, it is estimated that aggregate sales, y, are given by
y = 20 000(0.25)^°”^'
Suppose that the department has narrowed its options to five possible
lengths of campaign, including no campaign at all. Aggregate sales as
given by the Gompertz function would be as shown in Table 9.7.
Table 9.7
length of campaign aggregate sales % market potential
(x) iy)
0 5 000 25.00
13 11659 58.29
26 16210 81.05
39 18429 92.15
52 19 373 96.87
While the example hasn’t been set in the wider context of profits
maximization, from Table 9.7 we can see that it is unlikely that, with
the market already at 92% of saturation level, the campaign would run
beyond 39 weeks. S-shaped patterns of aggregate demand are also
produced by the logistic curve. In this model aggregate sales, y, are
given by
T
(9.39)
1 + ae-^^
T is market potential or saturation (the curves are also known as

saturation curves), so T > 0 and a > 0 and b > 0. The y intercept is
at r/(l + fl). As with the Gompertz curve (for the parameter ranges
specified), the lower asymptote is the x axis. The inflection point of

(9.39) is where sales are growing most rapidly, and is found by setting
the second derivative of (9.39) to zero. It emerges that the value of x at
the inflection point is
x = -^ln(l/a) (9.40)
Consider an example. For purposes of comparison with the Gompertz

curve, let
T = 20 000 and a = 3
These values give the same upper asymptote (20 000) and y intercept
(5000) as in the Gompertz example. A value of b about 0.1 produces
similar overall results. Table 9.8 shows corresponding values from the
logistic and Gompertz curves.
Table 9.8
X logistic curve Gompertz curve

0 5 000 5 000
13 11004 11659
26 16356 16210
39 18 855 18429
52 19 674 19 373
The inflection point of the logistic curve is found from (9.40) as
x = -^ln(l/3)^ll
So the rate of growth of aggregate sales peaks in week eleven. Logistic

curves are used in many other situations: one example is the diffusion
of information through a population. The ‘S-curve’ profile occurs in a
variety of business and financial contexts: the graph of the total project
expenditure against time typically follows this pattern.

Appendix
The Appendix reviews fundamental concepts, terminology and notation

that will enable you to use the book most effectively.
A.1 Effective use of numbers in business is an essential skill. The aspiring

Numbers financial analyst or economist must begin with a grasp of the
fundamental operations involving numbers, a classification of which
gives a useful starting point.
The counting numbers are the positive whole numbers represented
by the sequence of Arabic numerals
1, 2, 3, 4, 5, 6, 7, 8, 9, ...
in which zero is not included. Counting numbers are also called

natural numbers, but in actual use the natural numbers are
sometimes taken to include zero. The counting numbers form part of
the larger set of whole numbers, the integers
... - 5, - 4, - 3, - 2, - 1, 0, 1, 2, 3, 4, 5, ...
The integers comprise all positive and negative whole numbers and
zero. The negative integers are given by subtracting the counting
numbers from zero. The integers (think of integral meaning whole) can
be seen as part of a broader category - the rational numbers. A
rational number is the ratio of two integers and so may be an integer or
a fraction. When expressed in decimal form, a rational number either
terminates or repeats. For example,
17^8 = 2.125
is a rational number which terminates, while
> 16-^ 7 = 2.285714...
is a rational number which repeats ad infinitum the six digits after the
decimal point. A rational number terminates only if its denominator
has no prime factors other than 2 or 5. Irrational numbers cannot
be expressed as the ratio of two integers. ‘Irrational’ is not used in the
sense of ‘unreasonable’, but rather as ‘non-ratio-nal.’ The real
numbers constitute the rational and irrational numbers taken
Appendix 401
together. The real numbers contain all decimal representations, but
there are equations which have no solution in real numbers. The
simplest case is
+ 1 =0
for which the solution is ± i, where
Numbers of the form ki, in which k is a non-zero real number, are

imaginary numbers. Imaginary numbers obey normal rules of
arithmetic and are useful in practical contexts, including economic
growth models. Complex numbers have both a real and an imaginary
component, and are written in the form
a + bi
where a and b are real numbers. Examples of complex numbers are
5 + 4i, 6 - 2i, - 3.5 + 7.71 V2 + 0.4i
In the complex number a + bi the product bi is the imaginary part,

while a is the real part. Complex numbers come in pairs which have
the same real part, and where in one the imaginary part is added and
in the other it is subtracted. These are complex conjugates and take
the form
a + bi and a — bi
Addition and subtraction of complex numbers is straightforward:
(a -1- bi) -1- (c -(- di) = (a -1- c) -1- (b -I- d)i

(a + bi) - (c + di) = (a - c) + (b - d)i
For example,
(7+5i) + (-3 + 8i) = 4 + 13i

(10 - 2i) - (8 - 5i) = 2 + 3i
A.2 It is often convenient to refer to several items taken together in a

Sets collective entity called a set. The objects that make up the set are called
the elements or members of the set. If x is a member of a set A we
write:
X e A
which is read as
‘x is a member of A'
or as
‘x belongs to A’.

Where an element does not belong to a set, the ‘member of’ symbol is
struck out. So if y does not belong to A this is written as
y /A
If the number of elements is not large, then all elements can be

listed: for example the set A may be
A = {1,2, 3,4, 5, 6, 7}
Note the convention of enclosing the list in braces (curly brackets).

Alternatively, the properties required for membership of the set can be
described. In the case of set A we could write
A = {x: X is a counting number less than 8}
where the inside of the bracket reads
‘members of the set are x where x is a ...’
The description is the defining relation for the set. A, with a finite
number of elements, is a finite set. Sets which have the same number
of elements are said to be equivalent even though their elements may
differ. So the sets
(1 2 3}, (5 6 7} and (X Y Z}
are equivalent. Sets are equal when the membership of the two sets is
identical. So
(1 2 3} and (2 3 1}
are equal; the order in which the elements are listed does not matter. A
subset of a given set has as its members some (or all) the elements of a
set and no other elements. Thus, for example,
B = {1, 2, 3, 4} C = {2, 4, 6} D = {1, 3, 5, 7}

E = {5, 6, 7} F = {3}
are all subsets of A. The subset relationship is represented by
B C A
which reads
‘B is a subset of A’
Counting numbers can be seen as a subset of the integers, which in

turn can be seen as embedded within the real numbers. If B is a
proper subset of A it contains some, but not all, of the elements of A.
The notation is
Be A
The set
G = {2, 4, 6, 8}
Appendix 403
is not a subset of A since it contains an element which does not belong
to A. The set G not being a subset of A is indicated by
G%A
A superset has a given set as a subset. A is a superset of B, C, D, E

and F but not of G. D is a superset of F. The set of counting numbers up to
and including 10 is a superset of A. ‘Superset’ is a term used in marketing
when speaking of product improvements. If version 4(v4) of a piece of
software is said to be a superset of version 3.5(v3.5), what is meant is that
v4 has all the features of v3.5 and some additions. ‘Being a subset of’ is a
relationship which can be applied successively. So, in general, if
F C E and E C A, then E C A
The subset relationship is a transitive relation. Transitivity is an

important property of relations in economics. In consumer behaviour, if
among three groups of goods A is preferred to B and B is preferred to
C, then if the preference relation is transitive it can be inferred that A is
preferred to C. There is a special set which is a subset of all other sets.
This is the set containing no elements, the null set, and it is given the
symbol 0. The universal set, U, consists of all elements studied in a
given context, so in that context all sets are subsets of U. All subsets of
a set can be put into four categories:
1 Each individual member of the set.

2 All possible combinations of the members.
3 The set itself.
4 The null set.
The number of subsets for a set with n elements is 2”. The set of all
subsets of a given set is called the power set, written as 2*. So set A
with seven elements has 2^ = 128 subsets, while a set with 30
elements has a power set with 1 073 741 824 members!
A.2.1 We may wish to refer to the aggregate membership of two or more sets,
Operations on sets This is the union of the sets, and for sets E and F is written as
EUF
where the union symbol U is read as ‘cup’ and where
EUF= {3, 5, 6, 7}
As another example
BUC= {1, 2, 3, 4, 6}
Note that where an element is in both sets it appears once only in the
union. The order of elements does not matter and its immaterial
whether we write E U C or C U B. Union is an operation that can
include more than two sets. For example,
CUEUF = {2, 3, 4, 5, 6, 7}

Union is the elements of sets taken together. The intersection of sets is
the elements the sets have in common. The intersection symbol is an
inverted union symbol, n, which is read as ‘cap’. So
Bnc
refers to the elements belonging to both B and C. Thus
Bnc = {2, 4}
Note that where two sets have no elements in common, the intersection
of the sets has no members, i.e. it is the null set. So
Fnc = 0
Sets having no elements in common are said to be disjoint. The

intersection of more than two sets can be found. The result is those
elements found in all of the sets. For example, if the set
H = {2, 3, 4, 5}
then
BncnH = {2, 4}
Both union and intersection may be included in an expression. For

example,
En{BUC}
First find B U C
BUG = {1, 2, 3, 4, 6}
Now find the intersection of E with this union:
En{BUC} = {6}
We may wish to refer to all items that do not belong to a set. This
concept of the complement of a set allows contradictions and it is
better to consider complementarity as relative to another set. This is
not restrictive since the other set could be the universe of discourse.
Any set and its complement are disjoint. We shall here be concerned
only with the universal complement - all elements in the universal set
that do not belong to A. The complement of A is written as
A or C{A} or A'
For example, if the universe of discourse was counting numbers up to

10,’ then the complement of the set A would be the numbers 8, 9 and
10, written as
A = {8, 9, 10}
The set difference between two sets is the elements contained in

one set but not in the other. The set difference B — C is
B-C = {1, 3}
Appendix 405
while
D-F= {1. 5, 7}
and
F- B = 0
The symmetric difference is the set of elements belonging to

precisely one of two given sets. So the symmetric difference is the union
of the sets from which the intersection is excluded, and it is denoted by
© or V. Thus the symmetric difference of the sets B and C is:
B0C = {1.3,6}
Sets and operations on them can be represented in Venn diagrams,

where the universal set is shown as a square, with individual sets as
circles. In Figure A. 1(a) the union of B and C is the shaded area
covered by the two sets together. It is meaningful to consider the union
of disjoint sets - which in a Venn diagram would appear as two non¬
overlapping circles, as in Figure A. 1(b). Figure A. 1(c) shows the union
of a set with one of its proper subsets. This is the set itself. The
intersection of sets is shown as the overlap between circles, as in the
shaded lenticular region in Figure A.2(a). The complement of a set D is
shown by shading all the universal set except D, as in Figure A.2(b).
Venn diagrams are particularly helpful where more than two sets are
involved, with complicated combinations of operations. For example, in
Figure A. 3 (a) the shaded area shows B n C n H, while in Figure A. 3(b)
the shaded area shows the intersection of the complement of C with the
intersection of B and H, i.e. {B fi H} n C'.
Venn diagrams are useful in counting problems, where the
number of elements in a subset having certain properties (in relation to
the set or sets of which it is a part) is needed. For example, suppose
that in Figure A.3 B, C and H represent subscribers to The Financial
Times, Time magazine and the Economist, respectively. The shaded area
in Figure A. 3 (a) shows subscribers to alt three publications, while
Figure A. 3(b) shows subscribers to The Financial Times and the
Economist but not Time.
A.3 In x", X is said to be the base, while n is the exponent. It is helpful to

Exponents recall exactly that x” is x multiplied by itself n times:
\ JL • t\> t n\i
t\r • t\i • iX' • tX' • A, » • • 1H1 "VC
There are a number of rules for handling exponents and we will use
this simple definition to reinforce the rules in various examples.
A.3.1 When x” is multiplied by x'", the result is x raised to the sum of the
The product rule powers:
.yn m _ yH + m
tX iX - X

Figure A.l
The'validity of the rule can be confirmed by selecting particular values

for n and m, and setting out the multiple in full. So
= (xx:x)(»:x) =
The product rule can be applied no matter how many individual

terms of the form x" are multiplied together. So, for example,
x^x^x^ = x^^
Appendix 407
Figure A.2
(a)
(b)
Figure A.3
(b)

A.3.2 When x" is itself raised to a power, m, the result is x raised to the
The power rule product of the two powers. Thus
(x")'" = x""’
This rule can be confirmed in a simple example. For instance,
(x^)^ = (xx)(xx)(xx) = X®
As with the product rule, more complicated cases can be dealt with
under the power rule. Consider the case of repeated raising to a power:
((x")"’f = x""'^
So, for example,
((X^)')" = X24
In the context of raising a bracketed term to a power, note also that

given two variables, x and y,
{xyf = x"y"
A special case here is where only one variable, x, is involved but where
X has a coefficient, a, other than one. It is the case that
(ax)" = a"x"
So, for example,
(2x)^ = 2x2x2x = 2^x^ = 8x^
Where there are two variables each already raised to a power, as in
(x’^yPy = x^^yP^
it is again helpful in fixing this in one’s mind to apply it to a simple

example:
= {x^y){x^y){x^y) = x^y^
A.3.3 Where a term consists of the ratio of x raised to differing powers, the
The quotient rule result is x raised to the difference of the powers, i.e.
— = x"-'"
X"'
For example:
X5 •yi'V'V'■V’
•A-A/•'V A* (A- o
_ ■'
—_
■ i\i
'ytj,
►V
/Y''Y'
yX/A/
The quotient rule is the product rule applied to the case where one
of the powers is negative. There are several interesting special cases of
the quotient rule:
(i) m = n — = x''~" = x° = 1
X”
Appendix 409
This provides a revealing demonstration of the fact that — 1
(ii) m = n+ 1 +1
This shows that the reciprocal of x is x to the power minus one.
(iii) n = 0 = = x-"'
x^
Since x° = 1 the quotient rule can be used to confirm that x raised to

the power -m is the reciprocal of x"’.
Now consider rational exponents, first where the numerator is one and
where the denominator is an integer. The simplest case is X2. This is a
convenient way to represent the square root of x, and includes both the
positive and the negative root. Note that the radical sign, refers to the
positive root. So 4^ = ±2, while \/4 = 2. For an integer, m, when x is
raised to the power 1/m the result is the mth root of x. This is
represented as x^/'" or, using radical signs, as -y/x. It is easier to work
with rational exponents than radicals, so an early step is to convert
radical signs to rational exponents. The power rule can be used to
confirm that x^/"" is the mth root of x. The mth root of a number is the
value that when multiplied by itself m times produces the original
number. Therefore, applying the rule.
(xi/'")"’ = = X^ = X
As examples of the real roots corresponding to rational exponents

consider:
36^/2 = ±6
125^/^ = 5
256^/^ = ±4
243^/5 = 3
64^/^ = ±2
Note that where m is even there are two real roots of opposite sign.
Now suppose the numerator of the exponent is a number other than
one. If X is raised to the power n/m, the result is the mth root of x", i.e.
For example, x^/^ is the cube root of x squared:
so when x = 8, x^/^ = 4 and x raised to the power 3/2 is the square

root of X cubed:
x^!^ = ^,/x^
So if X = 16, then x^!^ = ±64.

The product, power and quotient rules also apply to irrational
exponents. As an example involving the product rule, consider
^ ^2v/2
Care may be required, as in this example, where the result, the sum of
the exponents, is not x^. As an example applying the power rule,
consider
= x^
An example of irrational exponents involving the quotient rule is
^ — y.l-V2 _ „-0.414 ...

/ — X — X
xî
In the above cases the base has been variable and the exponent
constant. In other circumstances the reverse may be true, or indeed
both base and exponent may be variables, as in x*. We will not
consider exponential functions (which result when the exponent is
variable and which are the subject of Chapter 9) at this point, other
than to point to a couple of examples. Note that in
3* = 81
the value of x emerges as 4, and in
-1024^ = -4
the value of x required to satisfy this equation is x = 0.2.

Note that exponents may themselves have exponents, allowing
astronomical numbers to be represented in a form involving just a few
numerals. For example, the number represented by 10 raised to the
power 10^® would be 1 followed by 100 zeros. A practical example is
the Gompertz function described in Chapter 9.
A.4 The sign of numbers can be important - as when a number represents

Absolute values profit, an account balance or the time remaining to complete a project.
But there are other situations in which the sign of a number is less
significant than its size. One instance is when numerical methods are
used to approximate a value such as project yield. What matters is the
absolute accuracy of the approximation rather than whether it is an over-
or underestimate. In computing, achieving an agreeable visual format of
output may require the separate representation of the magnitude of a
number and its sign. The magnitude of a number regardless of sign is
called the absolute value or modulus of the number. For some
' number, x, the absolute value or modulus is written as
ABS(x) or |x|
The former notation is often used in computing while the | | modulus

symbol predominates elsewhere. When x is positive, the absolute value
Appendix 411
is the number itself. When x is negative, the modulus is minus x. Thus,
for example,
|62| = 62 and | — 62| = 62

ABS( - 100) = 100 and ABS{ - tt) = tt
For two numbers, a and b, which can be positive or negative, the

following properties hold (inequalities such as < ‘less than or equal to
are discussed in Chapter 2):
(i) |a + b| < |fl| + 1^1
So, for example

|6 + 4| = 10 = 161 + |41
but
|6 + (-4)| = 2 < |6| + | -4|
(ii) |fl-1^1 > l|a| - |fe||
So that
|12 -9| = 3 = 1|12| - |9||
but
|-12-9| = 21>||-12|-|9|1 = 3
(iii) ABS(flb) = ABS(fl)ABS(&)
So that
ABS(( - 6)5) = ABS( - 6)ABS(5) = 30
Here we have looked at the absolute values of individual numbers.

The concept of absolute value also applies to functions in the form of
absolute value functions (which are discussed in Chapter 4).
A.5 The most common notation for numbers is place-value notation

Place-value using a base of ten. This is the denary system. But no one format is
notation the most suitable for all purposes. Variations on the place-value theme
use the fact that any positive number can be written as
+ an-ib''~^ + an-2b''~^ + ...

+ + aob^ + U-ib ^ a-2b ^ + ...
where b is the base or radix and the fli are non-negative integers less
than b. Therefore in the denary system in the number
8459067.123
each digit or numeral multiplies a power of 10 appropriate to its place

in the number. The number is
8(10^) 4- 4(10^) + 5(10^) + 9(10^) + 0(10^) + 6(10^) + 7(10°)

+ 1(10-^)+ 2(10-2)+ 3(10-3)

The numerals
0123456789
are symbols and each represents a number. The base in place-value

notation need not be 10. There are useful alternatives. For example,
octal numbers use a base of 8. So
654.3
in octal represents
6(8^) + 5(8^) + 4(8°) + 3(8“^) = 428.375 (to base 10)
Octal numbers are used to some extent in computing, where numbers

to the base of 16 are widely used. These are hexadecimal numbers,
which use as numerals the digits 0 to 9 and the letters A, B, C, D, E
and F. Thus in ‘Hex’ 3AE represents
3(16^) + 10(16^) -f 14(16°) = 942 (to base 10)

An important property of hexadecimal and octal numbers is that the
base is a power of 2. This facilitates conversion between hex, octal and
binary numbers. Binary notation uses base 2, so the only digits
employed are 0 and 1 - the binary digits or bits. A sequence of bits
(usually 8, sometimes 16) encodes a single piece of data - for example
a number or an alphabetic or other character - and is called a byte.
The binary number 111011 represents
1(2^) + 1(2^) + 1(2^) + 0(2^) + 1(2^) + 1(2°)

= 58 (to base 10)
A distinction is made between a binary number and a binary-coded

decimal (used in computing) in which the place values represent powers
of 10 but binary digits replace the usual decimal numerals. Thus the
binary-coded form of the decimal number 579 would appear as
0101 0111 1001

Less widely used bases are 5 (quinary) and 12 (duodecimal). In
duodecimal notation the letters A and B (or T and E) serve as additional
numerals. Fragments of a sexagesimal system (base 60) can be seen in
the subdivision of hours into minutes, and minutes into seconds. In
some examples we have used a point ‘.’to show where negative powers
of a base begin. More precisely, we have use fixed-point notation, in
which a number is written with the point separating the fractional and
integral parts. In fixed-point notation there are fixed numbers of digits
before and after the decimal point. For example, if the numbers
0.0054321 and 126.72

were both shown in fixed-point notation with ten digits, five of which
follow the decimal point, then the numbers would appear as
00000.00543 and 00126.72000
Appendix 413
Fixed-point notation (or an informal variant) is the most common
everyday use of the decimal system.
In floating point notation a number is shown as a multiple of a
base raised to a power. The ratio
250000
128
with four places before and after the point, has as its quotient in fixed-
point notation
1953.1250
This number can be shown in floating-point form in several ways. For

example;
19.53125 X 10^
0.1953125 x 10^
195312.5 X 10“2
1953.125 X 10°
With the 10° suppressed, the last instance is equivalent to fixed-point

form. The exponent can be positive or negative but cannot contain a
decimal point. In floating-point form the decimal point need not appear
in its ‘proper’, fixed-form, place. The base, here ten, is as appropriate to
the context. Where the coefficient of the base raised to a power is as
large as possible subject to still being less than the value of the base
itself, the result is called scientific notation or exponential notation.
In scientific notation the result of the division above would appear as
1.953125 X 10^
Note that in terms of the standard output used by some computer

software (for example Pascal or QuickBasic), numbers with more than a
certain number of significant figures would be likely to be presented in
exponential notation, where instead of writing 0.05960464 as
5.960464 X 10"^
this may appear as
5.960464E“^
where replaces 10“^.

So far we have discussed notation schemes for representing individual
numbers. There are also varied of ways of representing arithmetic
operations on numbers. The everyday manner of writing the sum
2+2=4 .
is with the operator - here addition, the symbol for which is '+’ -
placed between the addends or, in general, the arguments to which

the operation applies. This arrangement of arguments and operator is
called infix notation. By contrast, prefix notation places the operator
before the arguments. In prefix notation the above sum would be
written as
+ 22 = 4
Prefix notation is also called polish notation. In postfix notation the

operator is placed after the arguments and the sum would appear as
2 2 + = 4
Postfix notation is also called reverse polish notation, and is used

internally by computers and in some scientific calculators.
A.6 Rules of precedence govern the order in which operations are carried
Precedence in out. In the absence of such rules, in infix notation the expression
arithmetic
7-3x4
operations
could have different values depending on which of the operations of
subtraction or multiplication was done first. With subtraction first,
7-3x4 = 4x4=16
whereas, with multiplication performed first,
7-3x4=7-12 = -5
In fact multiplication takes priority over subtraction, so given this rule

of precedence the unambiguous result of the calculation is —5.
Arithmetic operations are conducted in the following order:
1 Exponentiation.
2 Multiplication and division.
3 Addition and subtraction.
Thus
7-3x22 = 7- 3x4 = 7-12 =-5
Infix notation alone does not always give the most economical
representation of a series of operations. For example, consider the
following set of instructions in four digits and three operators:
1 Subtract 3 from 7.
2’ Add 4 to 2.
3 Multiply the numbers produced by instructions 1 and 2 above.
Execution of these instructions is secured by use of parentheses (round

brackets such as those enclosing this remark). With infix notation and
parentheses, the expression is written as
(7 - 3) X (4+ 2) = 24
Appendix 415
which is more clearly an accurate statement of what is required than
the bald sequence of symbols (or string)
7x4+7x2-3x4-3x2
Where parentheses are used, precedence is given to their contents.

Parenthesized strings are evaluated first - they are treated as a whole.
This is called aggregation. Normal rules of precedence apply within
the parentheses. An advantage of reverse polish notation (shared by
prefix notation) is that there is no need for the use of parentheses. With
reverse polish notation, the instructions 1-3 above can be written
unambiguously as
7 3 -4 2 + X
Read from right to left, the string reads ‘multiply the sum of 4 and 2 by
the difference between 7 and 3’. Although they are unambiguous, ease
of reading is not a notable feature of either prefix or postfix notation,
and the legibility provided by parentheses is a worthwhile net gain.
More complicated sets of instructions than 1-3 above may require
several levels of bracketing, or nested brackets. There are rules of
precedence for brackets - innermost brackets are evaluated first, then
work proceeds outwards to the highest level of bracketing. By
convention, the innermost brackets are parentheses, which are worked
out before square brackets [ ], which are evaluated before braces or
‘curly’ brackets {}. Occasional use is also made of a vinculum, a solid
line,_, placed above or below terms to be aggregated, as in
ax + by — c
Where a vinculum is used, the terms so aggregated are evaluated first -

even before parentheses. Where there is no other prioritization, the
expression is evaluated by proceeding from left to right. As an example
of an arithmetic calculation involving nested brackets, consider
4 + 2{5[5 - 1] - [7(6 + 1) - 2(3 x 2^ - 8)]} - 10
Taking the innermost (round) brackets from left to right:
6 + 1 = 7, and since 2^ = 8 and 3x8 = 24, then 3x2^ — 8 = 16
so the expression becomes:
4 + 2{5[5 -l]-[7x7-2x 16]} - 10
The square brackets are
5-1=4 and 49 - 32 = 17
so the expression is now
4 + 2{5 x4-17}-10

Within the braces
20- 17 = 3
so we are left with
4+2 X 3 - 10
i.e.
4 + 6 - 10 = 0
We have discussed rules of precedence between the operations of

exponentiation, multiplication/division and addition/subtraction. Also
included in arithmetic operations are negation, modulo arithmetic and
integer division. The word ‘operator’ (which, strictly speaking, means
the symbol for the operation) can be used in other contexts where
mathematical or logical operations are performed on variables, values,
expressions or strings of letters or symbols. For example, the following
symbols are relational operators.
= < < > > ^
There are also logical operators such as NOT, AND or OR. A

binary operation applies to any two variables, values, strings,
expressions, sets, etc. For example, the operation of taking the
intersection of two sets is a binary operation. In its most general sense,
the word ‘operation’ can apply to any procedure as applied, for
example, to expressions or to strings of characters. Integration (for
which the operator symbol is J) and differentiation are examples of
operations in a wider sense.
A.7 We now consider appropriate scales of measurement given what it is

Scales of that the numbers represent. In economics, monetary variables such as
measurement prices, costs and revenues, and physical variables such as output or
consumption levels are measured on a ratio scale. A ratio scale has a
fixed zero, and both differences between and ratios of numbers are
meaningful. We speak of profits increasing by £500 000, investment
rising by DM2.4 million, costs doubling or manufacturing output falling
by 10%. But the units in which these variables can be measured are
not unique. Prices could be measured in pence rather than pounds,
profits in £k rather than pound units and investment in $ rather than
DM. But all of these possible conversions have the property of
preserving ratios. For example, if profits double when measured in
pounds, they also double in terms of £k and, given a steady exchange
rate, they also double in DM terms. On a ratio scale the unit of measure
is unique only up to an identity transformation. That is, if we
measure a variable in units x, then, so far as a model is concerned,
units y would serve just as well where
y = ax for any a > 0
Appendix 417
where, for example, if x is £ and y is DM, then (at the time of writing)
fl = 2.70. If X = grams and y = kg, then a = 0.001. Equilibrium
conditions derived from the model (such as marginal revenue equalling
marginal cost) are unaffected, although the numerical values at which
equilibrium is reached depend on the units chosen. Ease of use,
convention or transactions being conducted in an international unit of
account (e.g. the Ecu) or a single national currency (e.g. US dollars) are
common grounds for selecting one unit over another.
It is not always appropriate, or even meaningful, to measure a
variable against a ratio scale. An example is provided by utility
functions in consumer theory or portfolio theory, where alternatives
are ordered in terms of desirability and choices are based on individual
or collective preferences. For such purposes an ordinal scale is
appropriate. When a variable is measured against an ordinal scale no
significance is attached to ratios of numbers and differences between
numbers have no meaning apart from sign. This is because an ordinal
scale simply puts into rank order the sets of values of the arguments
of the function. Suppose that an investor prefers the combination (A)
of 20 units of return and 100 units of risk to the combination (B)
giving 30 units of return and 160 units of risk. This preference
ordering would be reflected by a function, /, which attached a larger
number to combination A than to combination B. That is, a function,
/, for which
/(20100) >/(30160)
If the values actually given by / are 55 and 54, respectively, that is

fine. If the numbers attached to the combinations were 1 000 000 and
3, respectively, or 1000.001 and 1000, that would serve just as well. It
cannot be inferred that because there is ‘only’ a difference of 0.001
between the two values there is not much to choose between the
combinations. All that matters is that A gets a larger number than B.
In this case if a function / accurately ranks sets of values of its
arguments, then any transformation of / that preserved the ordering
would do just as well. For a transformation to be order preserving all
that is required is that the transformation function has a positive slope
in the range of values to which it is to be applied. For example,
g=f or g = ^/f
both meet this condition and represent the ordering as well as /.

For other purposes an ordinal scale may be too weak and a ratio
scale too strong. An interval scale lies between them. In an interval
scale absolute differences between values are meaningful but ratios are
not (although ratios of differences are meaningful). There is no unique
zero for the scale. Temperature is measured on an interval scale in
degrees Celsius or Fahrenheit. Dates are also measured on an interval
scale. To speak of the delivery of stock being a week late makes sense.

but no meaning is attached to ratios of dates. It is meaningful to speak
of one temperature reading being 20 degrees higher than another, but
it is nonsense to say that a day on which temperature reached 30°
Celsius was twice as hot as one on which the maximum temperature
was 15°. A variable measured on an interval scale is unique up to an
order-preserving linear transformation. This is defined as
y = ax + b, where a > 0
where x represents the original units, y is the transformed unit and a

must be positive. The constant term b may be positive, negative or zero.
Note that with b = 0 the identity transformation is produced. An
example of such an order-preserving linear transformation is provided
by the well-known formula for converting degrees Celsius to degrees
Fahrenheit:
F = 32 + 1.8C
A nominal scale is even weaker than an ordinal scale. In a nominal

scale, observations are put into categories without ordering or
quantitative measurement. Official or commercial application forms
often call for nominal scale information - for example gender, marital
status or occupation. This information can be matched with other
records such as buying decisions to form a basis for targeted mailings.
The scales of measurement discussed do not stipulate particular units
of measure. Financial variables may be measured in £, $, DM, yen, etc.,
while units of volume may be tonnes, barrels, litres, etc. Within this
diversity there are important business and economic performance
indicators which take the form of ratios and so are independent of units
of measure - percentages are dimensionless in this sense. For
example, the fact that first-quarter turnover of Acme pic is already 35%
of last year’s total does not depend on whether turnover is measured in
£, £k or £m. Rates of change such as percentage growth in gross
domestic product (GDP) or charges such as the interest rate on a loan
are other examples. As a finite approximation, elasticity of demand is
given by the ratio
% change in quantity demanded

% change in price
Demand elasticity is used as a measure of the responsiveness of

quantity demanded to changes in price because, unlike the slope of the
deqiand curve, elasticity is independent of the units in which price and
quantity are measured. It can be important to distinguish between flow
variables and stock variables. A production function expresses output,
q, in terms of the inputs employed in its manufacture. For example:
q = /(labour, capital, materials)
All these variables - labour, capital, materials and output - are

measured as rates of flow per unit time. A time period of day, week.
Appendix 419
month or year is used as convenient. Optimal production plans relate
to resource availabilities and production levels expressed as rates of
flow per unit time. In contrast, a stock variable relates to the total,
number or aggregate value of items existing at some time. The
number of cars on the road, a firm’s finished goods inventory or the
nation’s currency reserves and available supply of capital goods are
examples of stock variables. Macroeconomic models use the concept of
stock variables. Where resources are limited it is important to
distinguish between two types of resource - those resources for which
unused supplies today can be carried over to tomorrow and those for
which non-use of today’s supply means that the productive potential is
lost. The former category is called a pool resource and the latter a
non-pool resource. Materials often represent a pool resource,
whereas labour is non-pool; unworked hours yesterday are not
available today.
A.8 We often need to refer to a list of numbers or expressions to be summed

Summation and - for example values of a variable over time such as returns on an
product notation investment, or values of different variables at a given time, say end-of-
A.8.1 year balances of budget centres. When evaluating an investment, we
Summation notation aggregate lifetime returns after appropriate weighting. Suppose that a
project gives returns in five successive periods. These could be
represented as
^1' ^2» ^5
This list could be referred to using a subscript notation as a set with

the elements
Xi, where i = 1, 2, 3, 4, 5
and where X; is the typical element. We may wish to know the sum of
the returns, S. This could be set out in full as
S= -f- X2 -I- X3 -I- X4 -I- X5
Such an exhaustive listing will not be usable if the number of elements

is large. A more compact way to represent the sum S is as
i=5
s=
i=l
This is sigma notation. Sigma (S) can be read as an instruction to

add all elements of the form xj, the terms of the summation, starting
from the subscript value given below S , f = 1, the lower limit of
summation, and concluding at the i value above E - here 5 - the
upper limit of summation. The subscript i is known as the index of
summation. Sigma notation has the great merit of compactness and is
also very flexible. For example, if we wanted to consider returns after a
project was up to speed and before it began to run down, we might

omit the first and last years from the sum. To do this, simply alter the
limits of summation:
i=4
S = ^ Xi = X2 + X3 + X4
i=2
If the upper and lower limits of summation are the same, just one
element is involved. For instance,
i=2
s = y^xi = X2
i=2
Another special case is where all terms in the summation are the same.
The result is simply the constant value multiplied by the number of
terms, as, for example, when each term takes the constant value k:
i=n
k = k + k + k + k + ... = nk
i=l
So, for example,

i=6
y^7 = 7 + 7 + 7+ 7+ 7 + 7 = 42
i=l
It is sometimes convenient to split a sum into two or more parts, for

example
3 33
i=l i=l i=l
where x,- might be the direct costs of a project in year i, with yi

representing indirect costs. This result is the sum rule. The result is
valid if the addition sign in the above is replaced by the subtraction sign.
To achieve economy of notation, the ‘/ =’ is edited out of the upper limit
of summation. A weighted sum may be required - as in investment
appraisal, where project returns are weighted by factors dependent on
the time of receipt and the interest rate. A weighted sum such as
WiXi + W2X2 + W3X3
can be written as
3
yy WiXi
1=1
A special case of a weighted sum is when all weights are the same,
for example:
5
y^ = 4:xl + 4x1 + 4^4 + “^^5
i=2
i=2
Appendix 421
Note that the constant can be taken outside the summation sign. The
following are examples of the ways in which the notation is used:
3
^ ^ x=1-)-2t3 = 6
X=1
This shows the sum of specific numbers rather than variables. In this
case the terms are the index of summation itself. A convenient formula
exists for finding the value of such a sum, given only the number of
terms, the first term and the common difference between the terms.
Other expressions can be calculated directly. Consider
20
J]](0.1x^-2x^-3x+10)
j:=1
This could be found through a computer program to be —1760. As a

further example, consider
5
V- . / 1111
^7/" = 2 + 3+4 + 5
n=2
30 20 15 12
77_
Summations may contain an unlimited number of terms. This arises

when there is no upper limit and/or no lower limit to the index of
summation. For instance, the sum
OO
2-'’ = 1/2 + 1/4 + 1/8 + 1/16 + 1/32 + ... = 1

fi = i
shows a sum to infinity. In this case there is a finite sum and the
series is said to converge. A financial example is provided by securities
such as Consols or War Loan, the returns on which are perpetual. The
sum of a finite number of terms of an infinite series (up to n = 100, say,
in the case above) is a partial sum. A summation may be taken over
more than one index. Two examples are provided by the problem of
finding the variance of the yield on a portfolio of many securities and
the problem of finding the variance of time taken to perform a sequence
of tasks with random durations. Consider the double sum
1=2 i=2
YY
i=i/=i
The most convenient way to take this sum is to take the first value of
the index i and run j through all its values, then take the next value of
i and again run theindex through all its values - and so on.
Adopting this approach, the sum set out in full is
CiiXiXi +C:2XiX2 +C21X2X1 +C22X2X2
= + (Ci2 + C2l)XiX2 + C22(X2)^

It is not necessary for indices to have the same upper and lower
values and there may be more than two indices over which to sum.
The sum may be of values of an expression which is a function of the
values of the indices. For example, consider the double sum
8 7
Y -xy-y^ +2x+ 3yf
.x = ly=l
This is easily evaluated (as 41 692) using a computer program.
A.8.2 If the product of terms is needed (as when finding discount factors if
Product notation year-to-year interest rates differ) product notation uses the symbol H
to indicate the product of terms in the same way that ^ means the
sum. Thus
k=n
JJ = Xi>;2^3 • --Xn-lXn
k=l
represents the product of n terms x^. The product may involve values of
the index itself as in the factorial of a number. This is
k=n
JJ/f=lx2x3x4x5x...xn = n!
k=l
where n! is read as ‘n factorial’. Thus where n = 6,
k^b
Y[k= 720 = 6\
k=l
Where there is no upper limit to the value of the index, as in

00
Y[
k=l
we speak of an infinite product. Note that the value of the product of

an infinite number of terms is not necessarily infinite; for example:
OO
Yl 4k^/i4k^ - 1) = 1.570796 ... = 7r/2

k=l
This is Wallis’s product.
A.9 A decision-maker is frequently presented with sequences of numbers. The

Sequences and numbers may represent the quarterly interest payments on an
series investment, successive values of GDP or annual profits for the last
decade. With their elements taken in succession, each of these collections
of numbers represents a sequence - an ordered set of numbers that is
either finite or denumerable (i.e. it can be matched one to one with the
counting numbers). The set of cash returns to a twenty-year project
Ki, R2, R^, ■ ■ ■ Rt< • • •' fî9. ^20
Appendix 423
is a sequence. The symbol Rt, indicating the return in year t, represents
the typical member of the sequence and is called the general term.
Where successive terms of a sequence are related in the same way, the
sequence is called a progression. The way in which pairs of terms are
related defines a particular kind of progression. Because of their
importance in finance and economics, we will be concerned with
arithmetic progressions and geometric progressions. In an
arithmetic progression, consecutive terms differ by a fixed amount.
Thus, for example, the sequence
50, 60, 70, 80, 90, 100
is an arithmetic progression. So is the sequence
13, 10, 7, 4, 1, -2, - 5,...
The natural numbers also form a sequence, as does the set of

integers. In the first of the progressions above, the common difference
between successive terms is 10 and the first term is 50. In the second
example the first term is 13 and the common difference is —3. In
general, if the first term is represented by a and the common difference
by d, any arithmetic progression containing n terms will have the
following form:
a, a + d, a + 2d, a + 3d,.. .a + (n - l)d (A.l)
The example progressions above have, respectively,
a = 50, d = 10, n = 6
and
a = 13, d = —3, n = oo
Using the first example to check the value of the last term as given by
(A.l), this should be
50+ (6-1)10 = 100
which is correct.
A financial example of arithmetic progression is simple interest,
where a fixed sum is paid in each period regardless of the amount
to which the original deposit has built up. So if £100 is banked on
1 January 1999 at 12.5% simple interest per annum the amount on
deposit increases by £12.50 each year, and a running record of
these amounts makes up an arithmetic progression shown as
follows:
1999 2000 2001 2002 2003 2004

100 112.5 125 137.5 150 162.5
The value of the investment after n years can be obtained using (A.l),
but care is required. The value after one year is the second term in the
progression and in general the value after n years is given by the

{n + l)th term in the progression: a + nd. So, since a = 100 and
d= 12.5, the value after 20 years will be
a + nd= 100 + 20(12.5) = 350
A further illustration of arithmetic progressions is provided by

straight-line depreciation, where the book value of an asset in any
year is its book value the year before less a constant amount. If an
asset is initially valued at £50 000 and is written off over ten years at
£5000 p.a. its book value over time forms an arithmetic progression:
Initial value One year Two years Three years

Book value £50 000 £45 000 £40 000 £35 000
The nth term of this progression gives the book value after (n — 1)
years. Thus, since a = 50 000 and d = —5000, the book value after
seven years (corresponding to term eight in the series) will be
50 000 + (8 - 1)( - 5000) = 15 000
It is often useful to know the sum of a number of terms in a

sequence. Such a sum is called a series, which can be written concisely
in sigma notation. For example.
t=20
t=l
represents the sum of a sequence of twenty terms where the general

term is Rf Where the sequence has a finite number of terms the series
is a finite series; otherwise it is an infinite series. The number of terms
is the length of the series.
Series taking the following form are important:
ao + aix + a2X^ + asx^ + ...
This is a power series. Power series are sometimes used to give a more
tractable representation of a function - a particularly important
example is the Taylor series. The sum, of an arithmetic progression
of n terms could be written in full as
Sn = + (î + d) + (tf + 2d) + (fl + 3d) + ... + [fl + (n — l)d]
A formula for this sum can be found by noting the result of adding
terms in a particular way. If the first and last terms are added, the
result is
, a + [a + {n— l)d] = 2a + {n— l)d
and if the second and second last terms are added.
(a + d) + [a + (n - 2)d] = 2a + {n- l)d
Similarly, if the third and third last term are added,
[a + 2d) + [a + (n - 3)d] = 2a + (n - l)d
Appendix 425
Terms can be paired off so that the sum of pairs is 2fl + (n — l)d. Since
there are n/2 pairs all told, the total is
S„ = j l2a + (n - l)d] ■ (A.2)
To find the sum to 100 terms of the arithmetic progression
10, 15, 20, 25, 30,...
note that a = 10, d = 5 and n = 100. So using the formula, the sum is
Sioo = 50[20 + (99)5] = 27 750
For
100, 95, 90, 85,...
the value of a = 100 and d = - 5. So from (A.2) the sum to 31 terms is
S31 = 15.5[200 + 30( - 5)] = 775
and the sum to 100 terms is
Sioo = 50[200 + 99( - 5)] = -14 750
The formula can be used in a different way to find the value of n

required to make the sum to n terms reach zero for the progression
-200, - 190, - 180, - 170,...
Here, a = —200 and d = 10. So the value of n is required for which
S„=|[-400 + (n-l)10] = 0
i.e.
^[-410 + lOn] =0
So, disregarding the trivial case in which n itself is zero, the sum is zero
when
-410 + lOn = 0
that is, after 41 terms.

A formula such as (A.2) stating the required sum in terms of known
parameters is called a closed-form expression. Where closed-form
expressions can be found they are very useful. Note that (A.2) will work
even if the number of terms is odd. In the case of odd n, the solitary
middle term will be the [{n + l)/2]th term, with a value of
+ {[(m + l)/2] — l}d
which is half of 2a + (n— l)d.

An application of (A.2) is in simple interest calculations. Suppose
someone borrows £100 annually at 5% simple interest. What would be
the total interest they would pay over twenty years.? The interest

payments in each year form an arithmetic progression, as can be seen
from the row in Table A.l. The total of interest payments over the
twenty-year period is found from (A.2), with the values a = 5, d = 5
and n = 20. The result is
20
— [2(5)+ 19(5)] = 1050
Table A. 1 Interest payable at

£100 invested at t= 1 t=2 t=3 £=4 £ = 5
t= 0 5 5 5 5 5
t= 1 5 5 5 5
t= 2 5 5 5
t= 4 5 5
t= 5 5
15 20 25
E 5 10
In a geometric progression the ratio of each term to the preceding

term is constant - terms differ by a constant multiplier. The following
progressions are geometric:
5, 10. 20, 40, 80. 160,...
128, 32, 8, 2, 0.5, 0.125,...
Successive terms in each of these progressions take the form
a, ad, ad^, ad^, ad'^, ad^,.... ad”^^
where the first term is a, the constant factor of difference (common

ratio) between adjacent terms is d and the nth term is ad"~^. The
number of terms in a geometric progression must be finite or
denumerable. In the first illustration a = 5 and d = 2, while in the
second a = 128 and d = 0.25.
In demography, if a population grows at a constant annual rate from
a base level a, annual population forms a geometric progression. For
example, if the annual rate of growth is 2%, then d = 1.02, and if the
initial population is 50 000 000, then population figures year by year
form the progression:
50 000 000 51000 000 52 020 000 53 060 400 54121608...
The inexorable nature of proportionate growth is apparent, as is the

fact that year-on-year increases, though constant in percentage terms,
mean increasing actual additions to the population.
In the declining-balance method of depreciation the first term a is
thd asset’s original book value, the factor of difference, d (which must
be less than one), gives the proportionate change in value from year to
year and the (n + l)th term of the progression gives the value of the
asset after n years. For example, if a car depreciates in value by 20%
each year, the common ratio is
d = 1 - 0.2 = 0.8
Appendix
427
and the value of the car after n years is given by the {n+ l)th term of
the progression. So after three years the value of a car bought for
£9000 would be
ad" = £9000(0.8)^ = £4608
Geometric progressions are important in compound interest, and the

sum to n terms is of particular importance. As was the case with
arithmetic progressions, it is possible to obtain a closed-form expression
for the sum, S„:
Sfj = a ud ad^ -|- ad^ -I- ad^ 4"... 4" ad"
Note that
dSfj — ud -|- ad^ -I- ad^ -I- ad^ 4-... -I- ad" ^ 4- ad"
so S„ and dS„ have all but their first and last terms in common. That is
S„ — dS„ = a — ad^
so
S„(l — d) = a — ad'^ = a(l — d")
so that
_ a(l - d")
(A.3)
“ (1-d)
An annuity consists of the receipt (or payment) of a fixed amount

each year. The present value of an annuity is the lump sum to which
the future receipts are presently equivalent. If the appropriate discount
rate is 100r%, then the present value of an annuity of £1 received at
the end of each of n years will be
111 1
5^(l + r)-‘ (A.4)
(1 + r) ^ (1 + r)2 ^ (1 4- r)^ ^ ^ (1 + r)" £=1
The constant difference in the annuity sum is 1/(1 -|- r) and the first
term, a, is also 1/(1 4- r). Use of this information in (A.3) produces
(l + r)-^[l-(l + r)-"]
1 -(1 + r)-^
(A.5)
1 - (1 4- r)-"
Expression (A, 5) allows rapid calculation of the present value of any

annuity. For example, with a discount rate of 10% (r = 0.1) an annuity
of £50 for six years would have a present worth of
£282.24
0.1
428
An important property of (A. 3) is that if d < 1 the series has a
bound on its value even if the number of terms is infinite - it is
convergent. In a sum to infinity, the index of summation may be
unlimited from above, from below or in respect of either extreme, as
with
oo t=b 00
Xt or Xt or ^ Xt
t=a —oo —00
Where there is no finite limit to the sum to infinity the series is

divergent, which may mean that the sum Sn —> oo or S„ ^ —oo or
that Sn oscillates. If in (A.3) the value of d is less than 1 it is clear that
as n increases without limit d” approaches zero. This is written as
follows:
as n —> oo d" —> 0
so
as M —> oo Sn ^ (^-6)
(1 - d)
Expression (A. 6) can be used to give the present value of undated

stocks (perpetuities) such as Consols or war loan. To illustrate: if in
the annuity formula (A. 5) we let n go to infinity, (1 + r) " goes to zero
and the formula reduces to
so the present value of a perpetuity of £1 is simply r With a payment

other than £1 it is the annual sum divided by the interest rate as a
decimal. If the annual sum is £S and the interest rate is 100r%, then
So at a 10% discount rate, a perpetuity of £50 has a present value of
500
and, if it can be sold, £500 should be its approximate price. For

instance, what should be the price of a unit of 3^% war loan stock if
the market rate of interest for the low-risk security is War loan is
quoted in units of £100 nominal value and the annual interest
payment is 3^% of this figure, i.e. £3.50. If for simplicity a single
annual payment is assumed, the price should be the present value
3.5
PV = £63.64
0.055
which approximates to the value of war loan stock at the time of

writing.
Solutions to exercises
Chapter 1
Exercises 1.2 1 (i) ^(5-2)2 + (2-5)2 = Vl8 (ii) V20
(iii) v^l3 (iv) \/65 (v) V4]
2 (,) slope = 2_j_^, = 3
(ii) -1
(iii) -1
(iv) -1/6
(v) 7/4
3 (i) zero (horizontal) (ii) undefined (vertical)
(iii) slope is +1
In case (i) the slope formula would involve division by zero.
4 (i) cl=[{9-5f + (5-3f + (S~4ff
= (16 + 4+ 16)5
= 365
=6
(ii) d=[(3-8)2 + (6-6)2 + (12-6)2]5
= (25 + 0 + 36)5
= 615
«7.81
(iii) c/=[(7-8)V(-2-1)2 + (_8-1)2]2
= (1 +9 + 81)5
= 915
^9.54
Exercises 1.3 1 (i) The slope is 3. (ii) The slope is —0.5

(iii) Zero slope - the line is horizontal, (iv) Slope = V2
(v) Undefined slope - the line is vertical.

2 The lines (i), (iv) and (vi) are parallel.
The lines (i), (iii) and (v) are concurrent.
The lines (vi) and (vii) are concurrent.
3 The lines (i) and (iii) are orthogonal.
The lines (ii), (iv) and (v) are orthogonal.
Exercises 1.4 1 (i) y = 4x + 5 (ii) y = —5x + 100

(iii) y = 10 (iv) X = 8
2 (i) y = -0.5x (ii) y = 3x + 4
3 (i) y = 4x + 3 (ii) y = 10 - X
4 (i) y = 5x - 11 (ii) y = 2x + 1
5 (i) Above the line, (ii) On the line,
(iii) Above. (iv) Below.
Exercises 1.5 1 (i) (a) Here f = 2, so n(t) = 90 (b) Here f = 5, so n(f) = 105.
(ii) In this case t = -3, so that, from the equation, n(t) = 65.
2 (i) / = 1000 — 40f, where / is the level of inventory and f is the
number of days'trading.
(ii) Stock falls to the predetermined reorder level on completion of t
days trading, such that / = 1000 — 40f = 240, so f = 19.
Exercises 1.6 1 (i) y = 3x — 7 (ii) y = —6x + 60

(iii) y = -0.5x - 20 (iv) y = x
2 y + 3x — 20 = 0 or —y — 3x + 20 = 0
3 (i) y-25=4(x-5) (ii) y = 4x + 5
(iii) 4x - y + 5 = 0 or -4x + y - 5 = 0
4 The point is above the line.
CN
o
Exercises 1.7 1 / = 10 (ii) x = 4 y = 0

X
II
(i)
(iii) x= 1 y = -2 (iv) x = 2.5 y=1.2
2 (i) X = 6 y= 8 (ii) x = 7 y = -3
3 (i) x= 10 5 y= (ii) X = 3
y=8
Substitution is more convenient in these cases since rearrangement is
not needed.
4 (i) Unique solution: x = 5, y = 8.
(ii) Infinite number of solutions (which must, however, satisfy
y = 9.5 + 0.5x). The 'two' equations describe the same line.
(iii) Unique solution: x = 2, y = 7.
(iv) No solutions. The equations are inconsistent; they generate
parallel lines.
Exercises 1.9 1 (i) x= 10 (ii) x = -12.5

2 (i) The equation simplifies to the statement —20 = —40, so the
equation cannot be satisfied for any values of x.
(ii) The equation simplifies to the statement 0 = 0, so the equation
represents an identity and will be satisfied for any value of x.

3 Cases (i), (ii) and (iii) satisfy the equation but case (iv) does not.
Cases (v) and (vi) give the same relationship between x and y as the
original equation, but case (vii) corresponds to a line parallel to the
original and so is inconsistent with solutions to the given equation.
Exercises 1.10 l(i)x = 5 y = 2z = 3

(ii) X = 4 y= 5 z=10
(iii) x=10 y = 8 z=120
(iv) x = 2 y = 4 z = 28
2 (i) The system is inconsistent (see the first and second equations)
and there are no solutions.
(ii) The first and third equations of this system are not independent
so there is not a unique solution to the system. Either the first
and second equations or the second and third equations can be
used to express any two of the variables in terms of the third
variable.
(iii) Here there is a unique solution x = 5, y = 10, z = 15 - so in
this case the three equations are both independent and
consistent.
3 (i) X =-z y=10+1.5z
(ii) x=15-4z y = 8 + 2z
Exercises 1.11 1 Using the first equation to eliminate x from the second and third
equations
X + 3y + 2z = 29
2y + 2z = 20
-2y + z = -2
from which the second equation is used to eliminate y from the third
equation, which is then divided through by 3. The result is
X + 3y + 2z = 29
y+ z = 10
z = 6
from which back substitution produces the following solution of the

system:
x= 5 y = 4 z = 6
2 Using the first equation to eliminate x from the second and third
equations
X + 0.5y + z = 9.5
y + 2z = -1
y+z= 7

from which the second equation is used to eliminate / from the third,
which is then divided through by -1, with the following result:
x + 0.5x + z = 9.5
/ + 2z=-l
z = -8
Using back substitution the solution of the system is
x= 10 x= 15 z = -8
Exercises 1.13 1 (i) Breakeven output level is
4800
= 2400 units
5-3
(ii) Two-thirds of plant capacity.

(iii) New breakeven output level is
5400
3086 units
^ 5-3.25
(iv) New value of price, p, such that
5400
2400
3.25
so p = 5.5
(i) Breakeven output level is
13 500
= 3375 units
12-8
(ii) 50% of plant capacity.
(iii) (a) Breakeven output level is unchanged,
(b) New breakeven output level is
13 500
q= = 3000
13.5-9
(iv) 25% reduction required.

3 (i) Excess demand of 30 units.
(ii) Excess supply of 45 units.
(iii) q= 100 p=100
(iv) (a) T = 1350 when t= 15
’ T = 2400 when t = 30
(b) Cost of subsidy (at t = —7.5) is 787.5.
4 (i) p = 80 q = 100
(ii) q > 44
(iii) 300 units of excess supply.
(iv) q = 160 p = 56
(v) 100 units of excess demand.
chapter 2
Exercises 2.1 1 (i) < (ii) > (iii) > (iv) 7^ or <>
2 (i) > (ii)< (iii) > (iv) <
Exercises 2.2 1 (i) x<10 (ii) X < 8.5 (iii) x >-5 (iv) x < 0
2 (i) The solution set is empty. The identity simplifies to the statement
1 <0.
(ii) All values of x satisfy the inequality. The terms in x cancel,
leaving the statement 10 < 12.
3 (i) 5<x<12
(ii) 9<x<30
(iii) The solution set is empty. The inequalities require that,
simultaneously, x > 10 and x <5.
4 (i) 4 < X < 6 (ii) 3 < x < 7 (iii) 3 < x < 7 (iv) 3 < x < 7
Exercises 2.3 1 (i) The half-space on or below the line y = 25 — 2x.

(ii) The interior and boundary of the triangular region defined by
the line segment y = 10 — x between and including the points
(0, 10) and (10, 0) and the x and y axes.
(iii) The strip bounded from above by the line y = 2.5x — 8 and
from below by y = 2.5x — 10.
(iv) The solution set is empty. Note that if the sign requirement on x
was removed it would be possible to satisfy the two constraints.
(v) The line segment bounded by the points (80, 0) and (40, 40).
Exercises 2.5. 1 For this problem the optimal solution is located at the intersection of
the two constraints. This is shown as point B in Figure S2.1. Optimal
values of the decision variables and objective function are
xi = 12 X2 = 9 F = 186
Figure S2.1

2 (i) The optimal solution is at C in Figure S2.1, where the values of
the decision variables are xi = 18 and X2 = 0, and where
F= 144
(ii) Point A is the optimum position here. At this point the decision
variable values are X] = 0 and X2 = 13.8, giving F = 138.
(iii) In this case, the contours of the objective function are parallel to
the second constraint. As a result, there is a non-unique
optimum. The corner points A or B, or any point in between, will
produce the highest value of F as F = 276.
3 In this problem the second resource constraint is redundant, as can
be seen from Figure S2.2. Constraint 1 always sets tighter limits for
both xi and X2. The feasible region OABCD is determined by the
first constraint and the upper bounds. The optimum is C, where
X] = 20, X2 == 4 and F = 800.
Figure S2.2
4 The problem is graphed in Figure 52.3. The feasible region is the

open area bounded from below by the axes and ABC. Point B is
optimal, being in contact with the lowest objective function contour.
At B the values of the decision variables and the objective function
are xi = 15, X2 = 20 and F = 165.
Chapter 3
Exercises 3.2 A is 2x2 B is 1x4

1 C is 5x1 D is 1x1 E is fxs
2 B, C and D are submatrices of A. But E cannot be obtained by
deleting rows and columns of A and is therefore not a submatrix of
A
3 The Trace, T, is the sum of the elements of the main diagonal of the
matrix, i.e. T = 10 + 27 + 35 + 29= 101.

Figure S2.3
4 A is upper-triangular; B is a scalar matrix; C is lower-triangular; D is

a diagonal matrix; E does not fall within any of the categories; F is
upper-triangular.
5 (i) X = 0, y = 3, z = 1
(ii) m = 2, n = 3
6 C^ = [-8 0 V2 4]
TT
3 = [100]
-1
0
7 A is not symmetric; B is skew-symmetric; C is symmetric; D is
symmetric.
Exercises 3.3 1 (i) ■-1 9 6'

8 2 2
6 3 4
(ii) 5 20'
15 18
(iii) 4-0 6-t-2a a-b

0 +9 b-c 30
.11 0 1 + c2
2 From the elements in the (1, 1) positions 5x — 4y = z, and from

those in the (1,2) positions 6 — z = 4; so z = 2. From elements in
the (2, 1) positions 4 + 2x = 24; so; x = 10. Hence 50 - 4y = 2,
giving y = 12 which is confirmed by the elements in the (2, 2)
positions.
436
5 3
3 A+B
2 -1
(i) Thus when k = 2
10 6
k[A + B] =
4 -2
(ii) and when k = — ]
-5 -3
k{A+B)--
-2 1
Exercises 3.4 1 (i) [19 29]

13 22‘
16 19
12 22
13 -3
7 -7
(iv) 15 -2 36
11 -20 15
29 26 12
3 5
-5 8
1 18
-18 19
16 55
(iii) A^ =
55 39
3 (i) In this case
2 0
.0 2
so / A and the matrix A is therefore not idempotent.
(ii) In this case:
8 7
-8 -7
so A^ = A and the matrix A therefore is idempotent.

4 With A premultiplying B, the result is
22 -26
AB
26 22
With A posfmultiplying B, the result is
22 -26
BA
26 22
so that the matrices A and B are commutative.

5 (i) 4
2 5
[4 3] 3
-2 6
2
4
[2 38 8] 3
2
[138]
' ']r4 [3 6]
4
221
19 [3 6]
11 J
66 132
57 114
33 66
3 2 0 '1 3'
1 0 1 •
1 2 -1 2 1
-
3 5 -1
2 5 1 0 4
7 11
1 0
5 1
3 5 -1
12 15
36 40 55 -4
12 8 5 4
54 57 75 -3
Exercises 3.5 1 (i) Since

C)
00
CD
o
5 3' 1 O'
1
5 4 -1 1 0 1
B is the inverse matrix to A (and A is the inverse of B).

(ii) Since
5 5 0.4 -0.5 1 O'

2 4 -0.2 0.5 0 1
B is the inverse matrix to A (and A is the inverse of B).

(iii) Since
p
bo
'3 2' 1 -0.4'

1
5 5 -1 0.2 0 -1
B and A are not inverses.

2 (i) |A| = 13
(ii) [Al = 10
(iii) |A| = -7
(iv) |A| = -0.1
(v) |A| = -(6 + 27r2)
(vi) |A| = 0
3 (i) Ad/A=
(ii) AdjA= t
(iii) Adj A =
(iv) AdjA =
■1.5 -2.5'
4 (i)
-1 2
0.4 -0.2
(ii) -0.3 0.4
■-1 0.4'
(iii)
2 - 0.6
2 5
(iv)
4 15
(v) No inverse: determinant is zero.
5 (i) Involutional: A“^ =A
(ii) Orthogonal: A^ = A“^
(iii) Involutional: A“^ =A
(iv) Involutional: A~^ =A
1-1
1_1
Oi o
0.4 -0.6' [55'

Exercises 3.6 1 (i) = =
.^2. -0.2 0.8 [20_

1-
1-1
CN ■—
o o
c> c>
-<1 0.3 40'

=
(ii)
CN
2^2 -0.1 -10

1
' 2 -4.5 ■ '45' O'

■^2, -1 2.5 20 5
(iv) The determinant is zero and the inverse does not exist. In this
case the equations are consistent; the first and second right-
hand-side elements are in the same relationship as the first and
second rows of the coefficient matrix. So there is an infinite
number of solutions.
’2Cl -0.2 0.7' ■-18' -2'
.^2_ -0.2 0.2 -8 2
(vi) The determinant is zero and the inverse does not exist. Here, the
equations are inconsistent; the first and second right-hand-side
elements are not in the same relationship as the first and second

rows of the coefficient matrix, and there are no solutions to the
system.
Xl 0.75 -1.25' '189' 9.25'
(vii)
.^2. 0.5 1 106 11.5
xi ■ 1 2' '2' '10'

(viii)
.^2. 4 3 4 20
L
Xl 8
2 (i)
L
5
Xl ■ -3'
.^2. .
7
Xl 9"
.^2. L
-4
Xl 9.5
.^2. .
15
’xi ^ O'
(v)
.^2. L
19
Xl ■ 100
(vi)
.^2. _
-90
Xl 20
(vii)
.^2. L
20
Xl ■ 1.5 -2.5' '90' '10'
3 (i) 2 50 10
.^2. 1
(ii) Using the inverse matrix, it is required that
^1 1.5 -2.5' b] O'

>
X2 -1 2 50 0
So from the sign requirement on X]: xi = 1.5b] — 2.5(50) > 0,

which implies that b] > 83.33; and from the sign requirement on
X2- X2 = -b] + 2(50) > 0, which implies that b] < 100; therefore
the tolerance interval for b] is: 83.33 < b] < 100.
(iii) In terms of b2, it is required that
XI 1.5 -2.5' '90' O'

>
X2 -1 2 .^2. 0
From the sign requirement on xi: xi =1.5(90) — 2.5b2 > 0, which

implies that 82 < 54; and from the sign requirement on X2:
X2 = — 1 (90) + 2^2 > 0, which implies that 82 >45; so the
tolerance interval for t)2 is: 45 < 62 < 54.
(iv) In the joint analysis, it is required that
Xl -2.5 b] 0
>
.^2. 2 bi 0
So that from the sign requirement on xi: xi = 1.5b] - 2.5^2 > 0,

which implies that 62 < 0.6fc>i; and from the sign requirement on X2:

X2 = —bi + 2^2 > 0, which implies that b] > 0.5b2', therefore the
range of relative sizes is: 0.5t)i < ^2 < 0.6b].
Exercises 3.7 1 (i) As revealed by the fact that |>l| = 0, the rank of A = 1, the
second row being the first row multiplied by -0.5. Since the
same relationship holds between the right-hand-side elements,
the equations are consistent, and any pair of values that satisfies
2xi - X2 = 20, that is X2 = 2xi - 20, will be a solution to the
system.
(ii) Here, the rows (columns) are linearly independent and the rank
of the A matrix is 2. The equations have the unique solution
given by
1 -1.6' ’1 r ■ 3'
-0.5 0.9 5 -1
(iii) Here, |A| = 0 and the rows (columns) of the A matrix are
linearly dependent (row 2 is —2/3 of row 1). But in this case
the right-hand-side elements do not stand in the same
relationship as that between the rows of A. So the equations are
inconsistent, and there is no solution to the system.
2 (i) The rank of A = 1. The second and third rows are twice and
four times the first row, respectively.
(ii) The rank of A = 1. Here, the linear dependency is more easily
seen in the relationship between the columns. Column 1 is twice
column 2, while column 3 is — ^ column 2.
(iii) The rank of A is 2. The third row is twice the first row so the
rank cannot be greater than 2. But the rank cannot be less than
2, as the second row is not a multiple of the first.
3 (i) Yes. The third column of the matrix is the sum of the two vectors
making up the basis, and the fourth column of the matrix is the
difference between the two basic vectors.
(ii) The rank of A is 2.
4 The rank of the matrix is 1. All columns can be seen as constant
multiples of any one given column.
Exercises 3.8 1 (i) (a) 7 5

^33
“38
= 56-15
= 41
(b) _ 8 4
2 1
= 8-8
= 0

(c) _ 7 9
m22
6 1
= 7-54
= -47
(d) _ 7 9
m3 2
“34
= 28-27
= 1
(e) 5 9
= 2 1
= 5-18
= -13
(ii) C33 = ( - 1 )^'^^m33 41

cii = (- =0
C22 = ( - l)^^^m22 = -47
C32 = ( - 1 = -1
C2i=(-l)2^'m2i = 13
2 (i) Using the first row
|A| = oi 1 cn + 012C12 + ai3Ci3
= 2cii +4ci2 + 3ci3
now
5 1
mil ^ 20-2 = 18
2 4
so
cii (-l)'^'mii = 18
4 1
mi2 16-9 = 7
9 4
so
C12 (-l)'^'mi2 = -7
4 5
mi3 = 8-45 = -37
9 2
so
ci3 = (-l)'*®m,3 = -37
So the determinant is
|A| = 2(18)+4(-7) + 3(-37)

= 36-28 -111 =-103

(ii) Using the third column:
|A| = 013C13 + 023C23 + 033^33

= 013^13 ~ 023^23 + 033^33
= 3mi3 - m23 +4m33
where
4 5
mi3 = 8-45 = -37
9 2
2 4
m23 = 4-35 = -32
9 2
2 4
^33 = 10- 16 =-6
4 5
so that
A1 = 3(-37)-(-32)+4(-6)
= 111 +32-24 = -103
which confirms the result in (i).

3 (i) Here, the most convenient row or column by which to expand is
the second row. The value of the determinant is then
I- + I - 1|^*^|2)m23 = -m2, - 2m23
where
9 8
m2i = = 36-24 = 12
3 4
and
7 9
m23 = 21 -45 = -24
5 3
|Al = -12-(2)(-24) = 36
(ii) Expanding by the third row
|A1 = |- l)m32 + (- 1|^+^(4|m33

= m32 + 4m33
where
4 2
m3 2 32-16= 16
8 8
and
4 3
0133 = 20 - 24 = -4
8 5

so
\A\ = 16 + (4)(-4) =0
The matrix is singular. The linear dependency in this case is evident

as the third row being row two less twice row one.
4 Begin by evaluating the determinant by the first row and note that
cii = mil C]2 = -mi2 Ci3 = mi3
so
|A| = mil - mi2 + 2mi3
where
4 9
mil = = 12 - 18 = -6
2 3
5 9
mi2 = = 15-27 = -12
3 3
5 4
mi3 = = 10- 12 = -2
1 2
so
|A| = -6-(-12) + 2(-2) = 2
The minors of elements in the second row are
1 2
m2i = = 3-4 = -l
2 3
1 2
m22 = = 3-6 =-3
3 3
1 1
m23 = = 2 -3 = -1
3 2
so that the cofactors for this row are
C21 =1 C22 = -3 C23 = 1

For the third row
1 2
^31=. o =9-8 = 1
4 9
1 2
^32 = ^ ^ =9-10 = -l
5 9
1 1
0133 = 4-5 = -l
5 4
So that the cofactors are
<^31=1 C32 = 1 C33 = — 1

The inverse matrix is then
J_ cn C21 C31
1-1
Cl 2 C22 C32
JA\
.Cl3 C23 C33
so
-6 1 r -3 0.5 0.5
1
12 -3 1 6 -1.5 0.5
2 1 -1 -1 0.5 -0.5
-2
Checking the result by multiplication

'1 1 2' -3 0.5 0.5 '1 0 O'
5 4 9 6 -1.5 0.5 = 0 1 0
3 2 3 -1 0.5 -0.5 0 0 1
5 In this case use the second row to find the determinant:
7 2
= 10.5- 10 = 0.5
5 1.5
The minors of the first row are

1 0
m^] = = 1.5 - 0= 1.5
1 1.5
0 0
mu = = 0
5 1.5
0 1
mi3 = = 0-5 =-5
5 1
So the cofactors for this row are

Cii = 1 .5 Ci2 = 0 Ci3 = -5
In the second row

-3 2
m2i = = -4.5 - 2 = -6.5
1 1.5
7 2
0122 = 10.5 - 10 = 0.5
5 1.5
7 -3
m2 3 = 7+15 = 22
5 1
So the second row cofactors are

C21 = 6.5 C22 = 0-5 C23 = —22
In the third row

-3 2
^31 = 0-2 = -2
1 0
7 2
m32 = 0
0 0
7 -3
0133 = =7+0=7
0 1
445
So the third row cofactors ore
C31 = —2 C32 =0 C33 = 7
The inverse is
■ 1.5 6.5 -2‘ ■ 3 13 -4‘

0 0.5 0 0 1 0
-5 -22 7 -10 -44 14
Checking the result by multiplication
'7-3 2' ■ 3 13 -4' "1 0 O'

0 1 0 0 1 0 = 0 1 0
5 1 1.5 -10 -44 14 0 0 1
Exercises 3.9 1 The determinant of the coefficient matrix is -1 and its inverse is
■-6 5 -4
A ’ 13 -10 8
-2 1.5 -1
so that the solution values for the variables are given by
■-6 5 -4' ■ 35' '10'

^2 13 -10 8 140 = 15
-2 1.5 -1 120 20
2 The inverse of the coefficient matrix is
3.6 2.2 -3.8

1-1
-2 -1 2
-0.2 -0.4 0.6
so the solution values for the variables are given by
’xi ■ ■ 3.6 2.2 -3.8' ■ 85 '10'

^2 = -2 -1 2 185 15
.^3. -0.2 -0.4 0.6_ L 185 20
r
3.6 2.2 -3.8' 70 20
^2 = -2 -1 2 50 -10
.^3. -0.2 -0.4 0.6 90 20
’Xl ■ ■
-
(iii) 3.6 2.2 -3.8' -4 -5
->^2 = -2 -1 2 139 25
.^3. -0.2 -0.4 0.6 78 -8
'xi ■
-
IV ' 3.6 2.2 -3.8' 9 'O'
->^2 -2 -1 2 63 9
.^3. -0.2 -0.4 0.6 45 0

3 The tolerance interval for b] is obtained from the requirements that
all Xj are non-negative. That is, b] must be such that
■ 3.6 2.2 -3.8' ’ b] 'O'

= -2 -1 2 185 > 0
^3. -0.2 -0.4 0.6 185 0
From the X] line: 3.6bi + 2.2(1 85) — 3.8(1 85) > 0, which implies
that b] > 82.22. From the X2 line: —2b] — 185 + 2(1 85) > 0,
which implies that b] < 92.5. From the X3 line: —0.2b] — 0.4(1 85)
+ 0.6(1 85) > 0, which implies that b] < 1 85. So the tolerance
interval for this parameter is 82.22 < b] < 92.5.
For 62 h is required that
' ■ 3.6 2.2 -3.8‘ ■ 85' 'O'
^2 = -2 -1 2 82 > 0
^3. -0.2 -0.4 0.6 185 0_
From the xi line: 3.6(85) + 2.2b2 — 3.8(1 85) > 0, which implies
that b2 > 1 80.45. From the X2 line: —2(85) — 62 + 2(1 85) > 0,
which implies that 62 < 200. From the X3 line: -0.2(85) — 0.4^2
+ 0.6(185) ^ 0, ^/hich implies that 182 ^ 235. So the tolerance
interval for this parameter is 180.45 < 82 < 200.
For 83 it is required that
xi' ■ 3.6 2.2 -38 . ‘ ■ 85 ' ’O '
-2 -1 2 185 > 0
3
^ . -0.2 -0.4 0.6 . ^3 0
From the xi line: 3.6(85) + 2.2(1 85) — 8.883 > 0, which implies
that 83 < 1 87.63. From the X2 line: —2(85) — 185 + 283 > 0,
which implies that 83 > 177.5. From the X3 line:
—0.2(85) — 0.4(1 85) + O.683 > 0, which implies that 83 > 151.67.
So the tolerance interval for 83 is 177.5 < 83 < 187.63.
Exercises 3.10 53 1
46 2
6 2
106-46
18-6

9 53
6 46
X2 =
9 1
6 2
414-318
18-6
= 8
So the solution is
xi = 5 and X2 = 8
(ii) 26 5
3 3
xi
4 5
1 3
78- 15
12-5
= 9
4 26
1 3
1 3
12-26
12-5
2 The determinant of the coefficient matrix (expanded by the first row)

is
2mii - 3mi2 + mia
where
1 5
mn = = 3 - 30 = -27
6 3
4 5
mi2 = = 12-10 = 2
2 3
4 1
mi3 = = 24-2 = 22
2 6
So
A| = 2(-27)-3(2) + 22 = -38

By Cramer's rule
16 3 1
34 1 5
30 6 3
in which the numerator determinant is
CO
16 1 5 34 5 + 1 34 1
1
6 3 30 3 30 6
= 16(-27)-3(-48) + (174) = -114
So
-114
><1 = 3
-38
For X2:
2 16 1
4 34 5
2 30 3
34 5 -16 4 5 +1 4 34
30 3 2 3 2 30
= 2(-48)-16(2)+ (52) = -76
So
-38
And for X3:
2 3 16
4 1 34
2 6 30
X3 =-
-38
1 34 -3 4 34 + 16 4 1
6 30 2 30 2 6
= 2( - 174) - 3(52) + 16(22) = -152
So
X3

The complete solution is therefore
X] = 3 X2 = 2 X3 = 4
3 The coefficient matrix is
■ 1 -22 7‘
A= 1 29 2
-2 12 4
and the value of X2 is given by
1 107 7
1 61 8
-2 -92 4
X2 =-
1 -22 7
1 29 8
-2 12 4
Expanding by the second column, the numerator determinant is
107(- 1)^+2^12 + 61(- 1)2 + 2^22 - 92(- l)^ + 2m32

= -107mi2 + 61m22 - 92 m32
where
1 8
mi2 = = 4+16 = 20
-2 4
1 7
0122 = = 4+14= 18
-2 4
1 7
0132 = =8-7=1
1 8
SO the determinant is
- 107(20)+ 61 (18)+92(1)
= -2140+ 1098 + 92 = -950
The denominator determinant also expanded by the second column

is
- 22(- l)^+2mi2 + 29(- 1)2 + 2^22 + 12(- l)^ + 2m32

= 22(20)+ 29(18)- 12 = 950
so that the value of X2 is:
X2 = -950/950 = -1

Chaptei^ 4
Exercises 4,1 1 (i) x>-10 (ii) x<20

2 (i) y>0 (ii) y>-10
3 (i) -30<y<45
(ii) -400 < y < 150
(iii) 1250<y < 2250
4 (i) No. While the slope of the function shows abrupt change at
X = 30, there is no step discontinuity in f{xj itself.
(ii) Yes, the function f(x) has a step discontinuity when x = 15.
Exercises 4.2 1 (i) x = 3, x = 4

(ii) X = —6, X = 5
(iii) X = —2, X = —7
(iv) X = 8 (double root)
(v) X = —9 (double root)
(vi) X = 0, X = 7
Expression (vi) is more conveniently written as x^ — 7x = 0, so that
x(x — 7) = 0, that is (x — 0)(x — 7) = 0
2 (i) x = 3, x = 4
(ii) x= 1.1, x= 1.2
(iii) X = — 1.5, X = 3
(iv) X = 1, X = -100
3 (i) X = 10 (repeated root)
The discriminant here is zero. The curve touches the x axis at one
point (from above, since the leading coefficient is positive).
(ii) X = +3/, X = —3/
The discriminant is negative in this case. Since the value of b in the
quadratic formula is zero, the roots are the imaginary numbers
3^—1 and —3.^—1. The parabola never reaches the x axis.
(iii) X = 5 — 2/, X = 5 + 2/
Here the discriminant is negative and the roots are the complex
conjugate pair 5 ± 2/. The parabola never reaches the x axis, and it
has its vertex at x = 5.
4 (i) X = 7, y = -8 and x= 14, y = —36
(ii) X = 20, y = -10 and x= 12.5, y = 12.5
5 (i) X = 5, y = 60 and X = 10, y = 260
(ii) X = 0.5, y = 16.5 and x = 2, y = 24
6 (i) —4 < X < 6
(ii) X < —5 X > 2
(iii) 3 < X < 8
7 (i) The revenue parabola is R = 700x — 5x^, with the vertex at
-700
X = = 70
-10

at which point revenue is
R = 700(70)-5{70f = 24500
(ii) Profits, H, are given by;
tt=R-C
= 700x - 5x2 - 150 - 40x _ ^^2
= 660x — 6x2 _ 1
with its vertex at
at which point profit is
7r = 36300- 18 150- 150= 18000
8 Equating the value of price given by the supply and demand

equations
720-4q= 160 + 0.25q2
which reorganizes as the quadratic equation
560-4q-0.25q2 =0
that is
q2 + 16q-2240 = 0
which factors as
(q-40)(q + 56) = 0
Discarding the negative value of q leaves q = 40, at which point

p = 560, obtained from either the supply or the demand equations.
These are the required market clearing values.
9 Equating unit cost and price gives
0.1x2-80x+ 17500= -10500
so that
x2 - 800X + 70000 = 0
which factors as
(x- 100)(x-700) = 0
so for costs to at least cover revenue, output must be in the interval
100 < X < 700

Exercises ^.3 1 (i) X = —5, X = 0, X = 10
(ii) X = 0, X = 1
Since there is no constant term in either case, x = 0 must be one root
of each equation. In case (i), taking xout, the remaining quadratic
x^ — 5x — 50 = 0
solves for
X = —5 and x = 10
In case (ii) x = 1 is a double root.

2 (i) X = ±2, X = ±3
(ii) X = ±1, X = ±6
These quartics arebi-quadratics, and can be solved as quadratics in
x^.
3 Average variable costs are
vr
_= - 90x + 4000
X
the parabola for which opens upwards and has its vertex, a
minimum, at
-(-90)
= 45
at which point total variable costs are
VC = (45)^ - 90(45)^ + 4000(45) = 88 875
4 Profit is given by the equation
TT = -x^ + 5000x2 _ 1 250 000
Since this is a bi-quadratic, letting z= produces
TT = -Z 2 +5000z- 1 250000
which is a parabola in z opening down so that the vertex is a

maximum. This is located at
-5000
2500
-2
so that, discarding the negative root, x = 50, at which point
7r= -(50)V 5000(50)^ - 1 250000

= 5 000000
Exercises 4.4 1 (i), (ii), (iv) and (vi) are polynomials. Recall that only non-negative
integers are allowed as exponents. So the term 3x~2 disqualifies (iii)
and the term x^ disqualifies (v).

2 (i) The degree is 5; the polynomial is a quintic.
(ii) The degree is 4; the polynomial is a quartic.
(iii) The degree of the polynomial is 1.
(iv) The degree of the polynomial is 0.
Exercises 4.5 1 The given equation for the hyperbola
xy — y — 2x + 2 = 0
can be re-expressed as
(x-l)(y-2) = 0
from v/hich the asymptotes are located at x = 1 and y = 2.

2 (i) Market saturation occurs v/hen p = 0, at which point
,0 = ^0°°-200 = 240
(ii) The lowest price that eliminates all demand will be
0 5000 - 200
480
10
3 Per unit fixed costs are given by the rational function
500000
AC =-
q
so an output level of at least q = 200 000 would be needed to
ensure that per unit fixed costs do not exceed £2.50.
Exercises 4.6 1 (i) z= 15x- 14 ii) z = 500 - 16x

2 z = 24x + 44 ii) z = 24x + 1 80
3 z = 4x^ -f- 12x -I- 15 ii) z = x3
4 (i) x = 0.25y-3
(ii) X = 0.05y -t- 2.5
(iii) X = 4y -I- 20
(iv) x = y2
10
(v)
(200 - y)
5 ( Homogeneous of degree one.
( Not homogeneous.
( ii) Homogeneous of degree three.
(iv) Not homogeneous.
(v) Homogeneous of degree one-half.
(vi) Homogeneous of degree minus two.

6 (i) f[x) is an odd function.
(ii) f(x) is an even function.
(iii) f(x) is neither an odd nor an even function.
Chapter 5
Exercises 5.3 1 The slope of the straight-line segment connecting tv/o points on
y = 5x^ for which the x co-ordinate differs by Ax is given by
Ay 5(x + Ax)^ — 5x^

Ax Ax
5(x^ + 2xAx + (Ax)^) — 5x^
5x^ -1- 1 Ox Ax + 5(Ax)^ — 5x^
1 Ox Ax -I- 5(Ax)^
= 1 Ox + 5 Ax
but as Ax approaches zero, so does 5Ax, so the derivative of

y = 5x^ is
^=10x
dx
Exercises 5.4 1 (i) ^ = 7x^

dx
i) ^ = 100x^9
dx
-3
ii) -3x“ i.e.
dx
(iv) i.e. -35

dx x-
1
(v) ^=(l/3)x-2/3 i.e.
dx 3x^/^
2
H ^ = 12/3)x-/^ i.e. 3^3,3
’(vii) f'(x) = V3x^3-1
(viii) P(x) = v/5x-^5-i

V5
(ix) f'(x) = 27rx^’'“ ^
(x) ^ = 40x^
dx

(xi| |=
2
(xii) ^ = 2x-’' i.e. ^
(xiii) f'[x] = 0
(i) f'(x) = 9x8
(ii) f'lxl =-9x-''> i.e. ^
(iiil f'lxl = 8x-’ i.e. ^
(iv) ^= i.e. -lOx-"

dx x^'
dy -1
i.e. -O.lx-2
'' dx"10x2
(vi) P(x) = v/4x^^-’
(viil f'lxl = V2x-'/2-' i.e. ^
I ■■■] dy 10 -11
I.e. lOx
(ix) f'(x) = Vnx^n-i

(x) f'(x) = 10000x99
(xi) f'(x) = -X^9
0.01
(xii) ^=-0.01x-' ’ I.e.
dx „i.i
3 (i) ^'(x) = 1
(ii) P(x) = mx'^-’
(iii) f'{x] = -px~P~^
(iv) f'(x) = (n - 1 )x“"
1
(v) I.e. nx,-n—
dx X■n-i-1
(vi) f'{x) = bax^ ^

(vii) f'(x) = x"’“^
(viii) ^ = >x>-^
(ix) — n2
f'{x) = n^x - 1
-« ... X
X^
‘I

Note that (V2 - 11(72 + 1) = 2 - 1 - 1.
(iv) f'(x) = -0.04X-0 8
Exercises 5.5 1 (i) ^ = 4x3+3x2

dx
(ii) ^ = 5x^-3x2
dx
(iii) $^ = 6x^+5x^-4x2
dx
(iv) P(x) = 20x2 + 9x2 _ -14^

(v) f'{x) = 45x^ - 28x2 + 15x2 _ ] ^ ]4
2 (i) f'{x) = max'^~^+nbx'^~^
(ii) P(x) = (m+l)cx'” - (n-l)c/x"~2
(iii) f'{x] = 3ox2 + 2bx + c
(iv) ^ = x" + x"’

dx
(v) ^ = Sanx^ + 4ai x^ + 3a2X^ + 203X 4- 04

dx
3 (i) $^=12x-156
dx
So a stationary value occurs at x-= 13 where the derivative is zero.
(ii) 10x+ 100

dx
So a stationary value occurs at x = — 10.
(iii) ^ = 6x2 _ 18x — 60

dx
So stationary values occur at x = —2 and x = 5.
Exercises 5.6 1 (i) (x2+x)(3x2 - 2x) + (x2-x2)(2x+1)

dx
(ii) ^ = (5x^ - 4x2)(30x2 + 35x^) + (10x2 + yx^^20x^ - 8x)

dx
(iii) ^ = (4x2 ^ 2x^ + 8x)(2 + 2x + 4x2)

> dx
+ (2x + x2+x^)(12x2+6x + 8)
(iv) ^=(x+l) + (x-l) = 2x

dx
^ ... dy (x^ - x"*)(3x2 + 2x) - (x2 + x2)(5x'^ - 4x2)
dy (3x^ - 2x^)(20x^ + \Ax^] - (5x^ + 2x^)(12x^ - 6x^)
(3x4-2x3)2
(7x3+5)-x(21x^)
(iii) V[x]
(7x3 + 5)2
, x3(8x - 5) - (4x2 _ 5x + 3)3x2

(iv) f'(x) =--
3 (i) The quotient rule is relevant, since the function is
Using the rule
f'W = ^(2^1 7-1)

2x2 x2 + 1
x2 + 1
This result could have been obtained without the use of the quotient
rule, since f(x) simplifies to
(ii) Using both the product and quotient rules, with the first two
brackets representing the numerator, the derivative is:
f'(x)= [(3x-5)[(x3-2x2)(2x + 2)
+ (x2 + 2x)(3x2 - 4x)] - (x^ - 2x2)(x2 + 2x)3]
^(3x-5)2
i.e.
p.. _ x2(12x3 _ 25x2 — 24x + 60)

(3x-5)2
23
(a) For which, at x = 1, f'(x) =
(b) and at x = 2, P(x) evaluates as 32.
Exercises 5.7 1 (i) P(x)=4(x3 + l)3j3x2)

(ii) P(x) = 5(x4 + x2)^(4x3 + 2x)
(iii) f'(x) = 7(4x^ + 3x4 _ y^2 ^ 2x)*^(20x4 + 12x3 - 14x + 2)
(iv) P(x) = -25(1 -x5)^(5x4) = -125x4(1 - x^)"

7 (v) ^==3{x^+xf(2x+]) + 2{x^+2x){3x'^+2)
ax
(vi) ^ = 6(6x^ - 5x^ + 6x2)^(42x^ - 20x3 ^ 12^]

dx
-9(x3-x^)®(8x^-7x6)
2 (i) (x3+xY6(x^-x3)^(5x‘^-3x2)
+ (x^ - x3)^5(x3 + x^)^(3x2 + 4x3)
dy (x2-1)3(12x2+6)-(4x3+6x)3(x2-1)22x
dx- (x2_1)6
dy 8x^3(3x2 _ 2x)2(6x - 2) - 32x3(3x2 - 2x)^

<"'> Tx =-643^5-
3 (i) f'(x) =1(5x3-7x2)-5(15x2_14x)
(ii) f'{x] = 0.25(4x3 ^ 8x)“°^^(l 2x2 + 8)
4 (i) f'{x) = n(ax2 + bx + c)"~' (2ax + b)
(ii) f'(x) = (m2 - l)[(m+l)x+m]”"
5 (i) ^=15z2(16x3)
dx
= 240z2x3
= 240(4x^)^x3
= 240(16)(x3)x3
= 3840x’^
(ii) ^=(2z-1||4x-2|
= (4x2 _ 4_j^ _ IU4^ _ 2)

= 16x3 _ 24x2 _j_ 4_^ 2
6 |i| ^ = |2z + z)(2)(3x2-1|
= |4w + 2)(2l(3x^-1)
= (8w + 4)(3x2- 1)
= 4(3x2 _ i)(2x3 -2x+ 1)
which expands as
24x^ -32x3 + 12x2 + 8x-4
(ii) When x = 1 the derivative dy/dx evaluates as 8.
Exercises 5.8 1 (i) ^ = 0.2
so that
^ = -L = 5
dx 0.2

(ii) After multiplication by 4 throughout the equation
^ = 0.5
dy
so that
±-±-7
dx 0.5
(Nil 1 = 5/
SO that
dy_ 1
dx 5/^
(iv) ^ = 8/3 - 9y2 + 8y
so that
dy _ 1
dx 8y2 - 9y2 + 8y
dy _ 1
2 (i)
d^ " (y^ + y2)(6y2 + 1) + (2y3 + y)(4y3 + 2y)
_1_
dx“3(5y3 + 12y- 10)^(15y2 + 12)
3 ^ = 5x^ + 6x^ + 5
dx
so that by the inverse function rule
dx_ 1
dy 5x^ + 6x2 ^ 5
and when x = 1
dx 1 ^
j- = TT = 0.0625
dy 16
dy _ 4
Exercises 5.9 1
dx 5
dy _ 5y + 10
(ii)
dx 2 - 5x
2 Having first expressed y as an explicit function of x, the derivative

dy/dx is then:
dy _ -480
^ ” (2 - 5x)2

which evaluates as follows
(i) atx=10 y=0 ^ = ^=-0.20833
(ii) atx=1 y = 30 ^ = -^ = -53.33
and in terms of the original derivatives obtained in 1 (i) and 1 (ii) th

corresponding evaluations are
|i) atx=10 y=0 ^ = ^ = -0.20833
(ii| atx=l y = 30 ^ = ^ = -53.33
which confirms the equivalence of the two results.
3 (i) jX-
^ dx”(2xy + 5)
dy 6x - y - 6xV^
dx 4x2y + X - 7
Exercises 5.10 1 (i) f"(x) = 20x3

(ii) P'(x) = 210x^
(iii) f"(x) = 60x^ + 6
(iv) P'(x) = 160x3 _ 108x2 42x - 16
2 (i) ^=120x=
= 360x2
= 120x2 - 72x
^ = 240x - 72
dx^
(iii) ^ = 7.2x^ - 34.2x'^
j4
^ = 50.4x^-581.4x1^
dx^
3 (i) f"(x) = 2x-3
(ii) f"|xl=^ i.e. -bx-^
(iii) f"{x) = -0.25x-i'5
(iv) Using the chain rule
f'(x] = 7{x^ + 5f2x = 14x(x2 + 5)^
To obtain the second derivative, the product and chain rules should
be used. The result is
f "(x) = 1 4x[6(x2 + 5)^2x] + (x2 + 5f]4
= 168x2(x2 + 5)^ + 14(x2+5)*

i2
^ (') 12ox^ + 6bx + 2c

dx^
j2
(ii) -j-y = -m^(m + 1 jx'”"’ “ ^ + n^(n + 1 )x“" “ ^

dx-^
Exercises 5.11 1 (i) For a stationary value of the function
$^=100-5x = 0
dx
So that X = 100/5 = 20 is a turning point. At the second order
So the turning point represents a maximum.
(ii) ^=8x-72 = 0
dx
So that X = 9 is a turning point. At the second order
So that a minimum of the function occurs at x = 9.

2 (i) f'(x) = 1.5x^ - 4.5x - 60 = 0 solves for x = -5, x = 8. At the
second order P'(x) = 3x - 4.5, which is negative for x = -5
and positive for x = 8, so x = —5 is a maximum, x = 8 is a
minimum.
(ii) f'{x) = -6x^ + 12x + 48 = 0 solves for x = -2, x = 4. At the
second order f "(x) = -12x + 12, which is negative for x = 4
and positive for x = —2, so x = 4 is a maximum, x = —2 is a
minimum.
(iii) f'{x] = 4x^ — 36x^ + 56x = 0, in which, since there is no
constant term, one root must be x = 0. We can therefore write
f'ix] as
(x - 0)(4x^ - 36x + 56) = 0
Now taking 4 out of the large bracket and factoring gives
4x(x-2)(x-7) = 0

so that the roots are x = 0, x = 2 and x = 7.
The second derivative is f”{x) = 12x^ — 72x + 56 and
when X = 0, f "(x) = 56 > 0
so that X = 0 corresponds to a minimum of f(x);
when X = 2, f"{x] = -40 < 0
so that X = 2 corresponds to a maximum of f(x);
when X = 7, f"{x] = 140 > 0
so that X = 7 corresponds to a minimum of f{x]

(iv) f'{x] =x^ — 5x^ +4 = 0
Note that the function f'(x) is a quadratic in x^, which factors as
f'(x) = (x2 - l)(x2 -4) = 0
in which each bracket in turn factors to produce
(x+l)(x-l)(x + 2)(x-2) = 0
so that the stationary values of the function are x = — 1, x = 1,

X = —2 and x = 2. The second derivative is
f"(x) =4x2 _ 10;^
for which
at X = — 1, f''(x] = 6
so that X = — 1 is a minimum;
at X = 1, f "(x) = —6
so that X = 1 is a maximum;
at X = —2, f"{x) = —12
so that X = —2 is a maximum;
at X = 2, f"(x) = 12
so that X = 2 is a minimum.
3 (i) ^ = 4x2 _ 24x2 ^ 44^ _ 24

dx
dV = 12x2 - 4Q^ ^ 44
dx2
at X = 1,
= 8
dx2
so that X = 1 produces a minimum;

at X = 2,
so that X = 2 produces a maximum;

at X = 3,
= 8
so that X = 3 produces a minimum.
(ii) ^ = 30x^ - 270x3 - 960x2 ^ 5400X + 12 000

dx
120x3-810x2 - 1920X + 5400
at X = -4,
dV
^ = -7560
dx^
so that X = —4 produces a maximum;

at X = —2,
= 5040
so that X = -2 produces a minimum;

at X = 5,
^ = -9450
dx-^
so that X = 5 produces a maximum;

at X = 10,
= 25 200
so that X = 10 produces a minimum.

4 (i) P(x) = 9x2 -36x-1053 = 0
for which the roots are x = -9 and x = 13. There are therefore two
turning points: -9 (maximum) and 13 (minimum).
(ii) f'(x] = 3x2 _ 6x + 15 = 0
the roots for which are the complex pair
X = 1 + 2/ and x = 1 - 2/
so that f{x] has no turning points.

(iii) f'[x) = 4x3 _ 36x2 + 160x - 400 = 0
We can deduce that this equation has either three positive roots or
one. In either case there is a real root, and the original quintic must
have a turning point. (This follows from Descartes' rule of signs,
in which the maximum number of positive roots is given by the
number of sign changes - here 3 - in the coefficients. But the
number of positive roots is equal in parity - odd or even - with the
number of sign changes. Therefore there must be three positive roots
or one). In fact there is one real root. The roots of P(x) are
x = 5, x = 2+4/ and x = 2 -4/
Exercises 5.12 1 (i) f"(x) = -36x2 ^ 236x - 672

so that
when X = 2, f"{x] = -144 (so this is a local maximum)

when X = 4, P'(x) = 96 (local minimum)
when X = 8, P'(x) = -288 (local maximum)
Since the leading coefficient of f(x) is negative and the power of the
leading term is even, the function decreases without limit at either
extreme and the global maximum will be at either x = 2 or x = 8.
Evaluating f{x] gives
x = 2, f{x] = 1592
X = 8, fix] = 2024
So the global maximum will be at x = 8.

(ii) P'(x) = -21+42x-18x2
so that
when X = 0.5, f"{x] = -4.5 (local maximum)

when X = 1, f"{x] = 3 (local minimum)
when X = 2, f"{x) = -9 (local maximum)
Evaluating f[x] at the local maxima gives
x = 0.5, f(x) = 254.15625

X = 2, f{x] = 264
So that X = 2 is the global maximum.

2 f"(x) = 12x2 - 168x +504
so that
when X = 3, f"(x) = 108 (minimum)

when X = 6, f"(x) = -72 (maximum)
when X = 12, P'(x) = 216 (minimum)

Since the leading term is positive and increases without limit for x
either positive or negative, then either x = 3 or x = 12 will be the
global minimum.
when X = 3, f(x] = 4001

when X = 12, f(x) = 3272
So that X = 12 constitutes the global minimum where f(x) = 3272.

3 (i) Since the leading coefficient of the quadratic is negative there is
a finite maximum.
(ii) No finite maximum - the leading coefficient is positive.
(iii) No finite maximum.
(iv) No finite maximum (f(x) increases without limit for negative x).
(v) No finite maximum. The function increases without limit as x
approaches zero.
(vi) There is a finite maximum; f{x] is unbounded from below at
either extreme. Note that for
f'(x) = 4000 -4x^=0
X = 10 is a multiple root, and since
f"(x) = -12x2 < 0
is satisfied for all x, then x = 10 is the single local maximum and the
global maximum.
4 (i) First establish where the turning points are, and whether a local
maximum lies within the permitted domain:
f'(x) = -3x2 ^ 73^ - 360 = 0
which solves for x = 6 and x = 20. The second derivative
f"(x] = -6x + 66
is negative atx = 20, showing that the local maximum lies in the
restricted domain. The procedure calls for the function to be
evaluated for the local maximum and the end-points of the domain.
The results are
X=5 f(x] - 550

X = 20 f(x) = 1900
X = 25 f{x) = 1250
So the global maximum over the restricted domain is 1900 at x = 20.

(ii) Establishing whether a local maximum lies in the permitted
domain,
f'{x) = —3x2 ^ _ 3^Q _ Q
which solves for x = 10 and x = 12. The second derivative

f "(x) = —6x + 66

is negative at x = 12, showing that the local maximum lies within
the domain. Evaluating the function for the local maximum and the
end-points of the domain. The results are
X=2 f(x) = 904

X = 12 f (x) = 204
x = 25 f(x)=150
So the global maximum over the restricted domain is 904 at the

end-point, x = 2.
(iii) The first derivative
P(x) = -3x2 _ 36x - 60 = 0
solves for X = — 10 and x = —2, both of which points lie outside the
domain. The function is then evaluated at end-points only, giving
X = 0 f(x] = 1500
x=3 f(x)=1131
So the global optimum in the restricted domain is at the end-point

X = 0.
(iv) The first derivative is
f'(x) = 3x2 _ 90^ + 600 = 0
which solves for: x = 10 and x = 20, which are both within the
domain. The second derivative
f "(x) = 6x - 90
identifies x = 10 as the local maximum, so evaluating the function

at this point and the end-points:
X=5 produces f(x) = 2000

X = 10 produces f{x) = 2500
X = 30 produces f(x) = 4500
so the end-point x = 30 gives the global maximum of f(x) in the

domain.
Exercises 5.13 1 (i) The second derivative
f"(x) = -15x+ 120
is zero for x = 8, and since the third derivative
f'"(x) = -15
is non-zero, x = 8 produces a point of inflection of f(x).

(ii) The second derivative
f "(x) = 24x - 90

is zero for x = 4. The third derivative
f'"(x) = 24
is non-zero, so that at x = 4 f(x) has a point of inflection.

(iii) The second derivative
f"(x) = 3x2-54x + 216
is zero for x = 6 and x = 12. The third derivative
f'"{x] = 6x - 54
is non-zero at both x = 6 and x = 12, which are therefore points of

inflection.
(iv) The second derivative requirement
f"(x) = 3x3 _ 9^2 ^ ^0
has X = 0 as a solution, and, with 3x taken out,
3x(x^ — 3x + 2) = 0
so
3x(x — 1 )(x — 2) = 0
from which it is clear that the other two roots are x = 1 and x = 2.
The third derivative
P"(x) = 9x2 - 1 8x + 6
is non-zero at all three values, which are therefore confirmed as

points of inflection.
2 (i) The first derivative is
f'(x) = 3x2 -36x+ 108
The second derivative condition is
P'(x) = 6x - 36 = 0
which solves for x = 6, giving an inflection point since f"'(x) = 6.

When X = 6 the first derivative is zero, so that x = 6 produces a
stationary point of inflection.
(ii) The first derivative is
P(x) = 9x2 — 54x + 100
The second derivative condition
f"{x) = 1 8x - 54 = 0
solves for X = 3 (which is an inflection point since f'"(x) = 1 8). At

X = 3, the first derivative takes the value 19, so that x = 3 produces
a non-stationary point of inflection.

(iii) The three relevant derivatives are:
f'(x) =4x2 -42x2 + 120X-50

f"(x) = 12x2-84x+120
P"(x) = 24x - 84
From the second and third derivatives, x = 2 and x = 5 are

confirmed as points of inflection. When x = 2 the first derivative is
104, and when x = 5, f'(x) = 0. So x = 2 is a non-stationary point
of inflection, while x = 5 is a stationary inflection point.
3 (i) The second derivative condition
f"(x) = 3x2 -30x +75 = 0
is satisfied only for x = 5, at which point the third derivative
P"(x) = 6x - 30
is zero. As may be confirmed, the second derivative does not

change sign through the point x = 5, so that the function does not
inflect.
(ii) The second-order condition
f"(x) = 6x2 - 1 8x - 60 = 0
solves for X = 5 and x = —2, which reference to the third derivative
f "'(x) = 12x - 1 8
confirms as an inflection point in both cases.

(iii) The second-order condition
f "(x) = 60x2 _ 1200x2 ^ ôOOx = 0
solves for X = 0 and x = 10 (double root). The third derivative is
P"(x) = 1 80x2 _ 2400x + 6000
is non-zero for x = 0, confirming this point as one of inflection. But

at X = 10, f'"(x) = 0, and, as may be borne out by checking the
sign of f"(x) at (say) x = 9 and x = 1 1 (positive in both cases),
X = 10 does not produce an inflection point.
Chapter 6
Exercises 6.2 1 (i) On the plane. The values of x, y and z satisfy the equation.
(ii) Above the plane. The z value in the co-ordinate triple is above
the value (200) given by the function using the x and y co¬
ordinates.
(iii) Above the plane.
(iv) Above the plane.
(v) Below the plane.
(vi) On the plane.

2 The equations of the contours are, respectively:
(i) y = 240-6x
(ii) y — -6x
(iii) y = 140 — 6x
3 By use of substitution:
(i) x=l y=2 z=3
(ii) X = 20 y = -5 z = 30
Exercises 6.3 1 (i) z = 1 OOx - 4x2 ^ 450
(ii) z = 500-0.5/2
2 (i) With / stated as an explicit function of x, the equation of the
contour is
1000 - 6x + 9x2
(10-3x)
(ii) With y as an explicit function of x, the numerator of the ratio

factors as (10 — 3x)^ with the result that the contour is the
straight line y = 10 — 3x.
3 (i) All terms are of degree two, so the function is homogeneous of
degree two.
(ii) Not homogeneous - due to the degree-three term xyw.
(iii) Homogeneous of degree two.
(iv) Not homogeneous - due to the —2x + 1 component of the
expansion of (x — 1 )^.
Exercises 6.4 1 The slope is —2 in the fundamental x direction and 0.5 in the y
direction.
2 (i) | = 2x-4y
|^=-4x+l
dy
(ii) (a) X direction: 2(10) - 4(2) = 12

y direction:-4(10) + 1 =-39
(b) X direction: 2( - 2) - 4(7) =-32
/direction:-4( - 2) + 1 =9
3 The partial derivatives are
fx = 2xy2
fy = 3x2/2
and at (5, 4, 1600) these functions evaluate as
fx = 640
fy = 1200

4 (i) fx = + Zx'^y^
fy = Zx/*^ + Zx^y'^
(ii) = 9900x9
fy = 10100x99yioo
(iii) fx = 16x^ - 9x^y^ + 16x^y^
fy = 1 Oy^ - 12xV^ + 12xV^
(iv) ^ = 0.7l~^-^k°^
ol
^ = 0.4P-^k-°‘'
^=bn>-'
ok
Bt
(vi) — =mnx^“V"~^
ax
^=n[n- l)xV^
dz
(viii) 5x^y — 2yx ^

dx
dz 5
-2
ay
5 (i) fx = SOx^y'^w^ - 8x^y^ - lOxw"*
fy = 40x^y^w^ — 6x^y^ — 12y^w^

fw = 20x^y^vv' — 20x^ + 8y^ w
(ii) rAL^-^K^M^
OL
dQ
= sAL^K^-^M*
W
dQ
= tAL^K^M^-^
m
6 The marginal profits on x and y are the first-order partial derivatives
with respect to these variables. So marginal profit on x is
500 + 4y — 6x; marginal profit on y is 400 + 4x — 4y.
Solutions to exercises 4Z1

7 (i) f, = (3xV-8xV)(12xy2+2y2)
+ (6x^y^ + 2xy^)(6xy - 24x^y^)
fy = (3x^y - 8x^y^)(l 8x^y^ + 4xy)
+ (6x^y^ + 2xy^)(3x^ - 24x^y^)
(ii) fx = Ay(] + Bxy) + By(l + Axy)

fy = Ax( 1 + 6xy) + 6x( 1 + Axy)
(x - y)'
r x(x - y) + xy
' (x - yf
x2
~ (x - yf
^ (y^ - 5xy)(4xy2 - 3x2) _ (2x^2 _ ^3)( _ 5^,
(y2 - 5xy)2
, (y^ — 5xy)(4x^y) — (2x^y^ — x^)(2y — 5x)

(y' - 5xy)2
9 (i) fx = 6(6x^y5 - 3x^y3)^(24xV^ - 12x3y^)

fy = 6(6xV^ - 3xV^)^(30xV^ - 9xV^)
(ii) fx = d(ax^ + bxy + (2ox + by)

fy = d(ax^ + bxy + cy^f~^ (bx + 2cy)
10 fx = {y^ - w'^x^)(3x^ - 2xy^w^] + (x^ — x^y^w^)[ - 2w^x)

fy = (y3 - w^x^)( - 2x^yw^) + (x^ - x^y^w^]{3y^)
fw = (y^ - w^x^)( - 2x^y^w) + (x3 - x^y^w^){ - 4w3x^)
Exercises 6.5 (i) The first order direct derivative with respect to x is
dx ^
so the required second derivative is
d^z
- 20x3y2
dx^
= — 160x3 y3 _ 84x^y
dx'^

(ii) = 30xy^ - 12y^w^
2 « Sl = 2x=
(ii) 0 = 5O4y7-48xV
(iii) = 30x^y - 12x^w^

oy^
d^z d^z
3 (i) lOxV
dxdy dydx
= -120x^y2 - 14x6 =
dxdy dydx
d^z a2;
= 45x^y^ - 24xyw^ - w =
dxdy dydx
dz
4 (i) ^ ~ 1 Oxy + 2x + 7y - 3
so at X = 1, y = 2 the slope is 25.

9z
= 3y^ - 8xy + 5x^ - 6y + 7x + 8
9y
which evaluates to +4 at x = 1 and y = 2.
(iii) ^^=-8y+ 10x + 7 =

dydx dxdy
d^z
(iv) 12x+10y + 2 = 34 at x=l,y = 2
dx^
d^z
(v) = 6y - 8x - 6 = -2 at x = 1, y = 2
5y2
di^z
VI = 12
dx^
d^ z
5 (i) fxx = 60xV - 60xV^ + 4y^

, = 84x6y6 - 30x^y + 1 dxy^
fyy = 84x6y6 - 6x6 ^ 24x^y^
(ii) (a) fxy = 20x^y^w6

(b) fxw = 30x^y^w6
(c) fy^ = 24x6 yV6

6 01 0 = 6OxV‘-24y2
(ii| = 150x‘‘y'‘ - 24x^ + 144xy^

dxdy^
(Mi) = 600xV'‘ - 48x + 144y2
7 (i) = Sx"*/* - 20x^vv^ - SOy'^w^

(ii) fy^ = 12x^/3 - 1 80yV^
(iii) =48x2/^
(iv) fwww = -240x^w^ - 600y^w^
Exercises 6.6 1 (i) At the first order the conditions ore
fx = 20 + 8y — 2x = 0
= 64 + 8x — 64/ = 0
from which, at the second order, it works out that
fxx = -2 fyy = -64 fxy = 8
and, since the discriminant
(-2)(-64)-82 >0
is positive, the stationary value is a maximum of the function.

(ii) At the first order, the conditions are
= 4x — 16 — 2/ = 0
fy = 2/ - 20 - 2x = 0
^XX = 4 fyy = 2
No further work is needed since it is clear from the direct second-

order derivatives that the conditions for a maximum are not satisfied.
(iii) The first-order conditions are
fx = 5+y — 0.5x = 0
fy = -4 + x- x = 0
fxx='-0.5 fyy = -l
So far so good. But since , the discriminant evaluates as
(-0.5)(- 1)- 12 <0
Thus the conditions for a maximum are not fulfilled in this case.

2 (i) The first-order conditions
fx = -4 + X - 2x = 0
fy = 5+ x — y = 0
solve for X = 1 and y = 6. At the second order
fxx — 2, fyy = —1 and fxy = 1
and the condition on the discriminant
(-2)(-l|-l2>0
is satisfied, so the stationary value is a maximum, and at x = 1,

y = 6,
f(x, y) = l 00- 4-f 30+ 6-1-18=113
(ii) The first-order conditions
= 5 + 2y — 0.5x = 0
= -16 + 2x- 16y = 0
solve for X = 12 and y = 0.5. At the second order
— —0.5, fyy = — 'Id and fxy = ‘2.
and the condition on the discriminant
(-0.5)(- 16) -2^ > 0
is satisfied, so the stationary value is a maximum. At x = 12,

y = 0.5,
f(x, y) = 100 + 60 - 8 + 12 - 36 - 2 = 126
3 First-order conditions for a local maximum are
^=—21 + 2xy + y^ — 6x^ = 0

ox
^ = 576 + x2 + 2xy - 24y2 = 0

dy
Substitution of the values x = 2 and y = 5 in these conditions

confirms that they are satisfied. At the second order the direct
* derivatives
d^z
^ = 2y- 12x = -14<0
ox^
^ = 2x-48y = -236 <0

ay^

are the appropriate sign at x = 2 and y = 5. The mixed derivative
d^z
2x + 2x= 14
dxdy
at X = 2, y = 5. So the discriminant condition
(- 14)(-236)- 142 >0
is also fulfilled, and a maximum at x = 2, y = 5 is confirmed.

4 (i) The first-order conditions are
f^ = -14-2.5y + 3x = 0
fy = 4 - 2.5x-f 4y = 0
which solve for x = 8 and y = 4. At the second order:
ôc = 3 fyy = ^ fxy^ -2.5
These values fulfil the discriminant condition since
(3)(4) - 2.52 >0
So X = 8 and y = 4 produce a minimum of f(x, y). Substitution of

these values into f{x, y) shows the minimum value to be 48.
(ii) The first-order conditions are
^ = 6-3y + x = 0
fy = —54 — 3x -f 1 8y = 0
which solve for x = 6 and y = 4. At the second order
4x = 1 = 18 fxy = -3
These values fulfil the discriminant condition since
(11(18)-1-3)^ >0
So X = 6 and y = 4 produce a minimum of f{x, y). Substitution of

these values into f{x, y) shows the minimum value to be
f(x,y) =126 + 36-216-72-18+144 = 0
5 The first-order conditions are
fx = 3x2 _ 2Qg _ ^ _ 2y2 _ Q
fy = -0.5x2 + 42 - 4xy + 48y2 = 0
which solve for x = 10 and y = 1. For these values the second-

order conditions work out as
fxx = 6x - y = 59 > 0
fyy = -X - 4x + 96y = 46 > 0

so the conditions on the direct second-order derivatives ore fulfilled.
The value of the mixed derivative
4y = -X - 4/ = -14
leads to satisfaction of the condition on the discriminant since
(59)(46)-( - 14)2 >0
Exercises 6.7 1 (i) At the first order, for any stationary value of f(x, y)
fx = 8 + 2/ - 4x = 0
fy = —5 + 2x — 0.5/ = 0
These conditions solve for x = 3, / = 2. At the second order
fxx = —4 and fyy = —0.5
Since both direct second-order derivatives are negative the

stationary value could be either a maximum or a saddle point; a
minimum is ruled out. The cross-partial derivative is f^y — 2 and the
discriminant evaluates as
(-4)(-0.5)-22 = -2<0
Since this is negative the location is a saddle point.

(ii) The first-order conditions for a stationary value are
fx = 15 — 3/ — 1.5x = 0
fy= 12 — 3x + 3/ = 0
which solve for the values x = 6 and / = 2. At the second order
fxx = — 1 -5 and fyy = ^
which implies a saddle point, since, regardless of the value of f^y,

the discriminant is negative, as required for a saddle point.
(iii) The first-order conditions for a stationary value are that
fx = 4x + 8/ = 0
fy = / + 8x = 0
which equations solve for x = 0 and / = 0. And at the second

order
fxx = 4, fyy = 1 and fxy = 8
which implies a saddle point since the discriminant is negative:
(4)(1)-82 = -60<0

2 (i) At the first order
= -220-14/+ 10x = 0
fy = 310-14x + 20x = 0
which solve for x = 15 and / = —5. At the second order
^xx = 10 and fyy = 20
thus eliminating the possibility of a maximum. The cross-partial

derivative
fxy = -14
SO the discriminant evaluates as

(10)(20)- 196 = 4 >0
which rules out a saddle point and identifies the point as a

minimum.
(ii) At the first order
15-7x-6x = 0
fy = 17 - 7x - 8/ = 0
which conditions solve for x = — 1 and / = 3. At the second order
fxx = -6 and fyy = —8
eliminate the possibility of a minimum. The cross-partial derivative
f^ = -7
so the discriminant evaluates as
|-6)|^8|-|- + = -1 <0
which identifies the point as a saddle point.
(iii) At the first order
fx = 34 — X — 5x = 0
fy = —10 — x + 4x = 0
which conditions solve for x = 6 and / = 4. At the second order
fxx = -5 and fyy =
Since the direct derivatives are of opposite sign, the stationary value
in this case must take the form of a saddle point.
3 (i) Saddle point at X =-4, X = 6.
(ii) Saddle point at x = 3, / = 1 •
(iii) Minimum at x = 2, / = 8.
(iv) Saddle point at x = 2, / = 3.
(v) Maximum at the origin.
(vi) Saddle point at x = 2, / = -12.

Exercises 6.8 1 (i) For a stationary value of the function, at the first order it is
required that
= 24x + 12/ + 8 = 0
fj. = 12x + 6/ - 10 = 0
An attempt to solve these equations requires the equation of two

unequal constants. They are therefore inconsistent and no point
fulfils the conditions for a stationary value.
fx = 60 — 5x + 0.5/ = 0
fy = 64.5 + 7y + 0.5x = 0
solve for X = 11, y = — 10. At the second order
fxx = —5 and fyy = 7
which, being of opposite sign, imply that the discriminant is

negative, and so a saddle point is produced by the values x = 1 1,
/= -10.
(iii) At the first order
= -40x + 20/ - 50 = 0
fy = — 10/ + 20x + 30 = 0
Multiplying fy by 2 and adding the two equations produces the

result 10 = 0, so the equations are inconsistent, and no stationary
value of f(x, /) exists in this case.
2 (i) At the first order
fx = 4.5x — 9/ — 9 = 0
fy = 1 8/ - 9x + 18 = 0
which conditions can both be rearranged as the equation
/ = 0.5x — 1
so all points satisfying this condition produce a stationary value of

the function (which turn out to be minima).
fx = x + 2/-5 = 0
fy = 4/ + 2x - 8 = 0
which equations can be rearranged as
^ X + 2/ = 5
X + 2/ = 4
which are inconsistent, and f(x, /) therefore has no stationary

value.

Chapter 7
Exercises 7.1 1 The slope of the constraint line is —1/3, os may be seen from
rearranging the constraint as:
55 1
The derivatives of the objective function with respect to x and y are
fx = 22 + 3y - 3x
fy = 33 + 3x — 4.5/
Tangency between the contours of the function and the constraint

requires that
_4=_1
fy 3
SO
(22 + 3/ - 3x) _ 1
(33 + 3x - 4.5/) “ 3
Hence
66 + 9/ — 9x = 33 + 3x — 4.5/
which rearranges as
12x- 13.5/= 33
into which the constraint relationship x = 55 - 3/ may be

substituted, with the outcome that
38
y — X = 17
which values produce a maximum value for f(x, /) of 643.5.

2 First find the value of the function f[x, y) at the unconstrained
optimum. At the first order the conditions are
fx = 20 - 0.2x = 0
fy = 16-0.1/ = 0
which equations solve for x = 100 and / = 160. At the second

order
fxx = -0.2 fyy = -0.] fxy=0
so
fxxfyy - [fxyf > 0

and the stationary value is a maximum. At this point
f(100, 160) = 2280
With the constraint included, the equilibrium condition is then
20 - 0.2x 5
16-0.1x~2
which can be restated as
x= 1.25/- 100
From the constraint
325 2/ .
X= ^ ' =65 0.4/
so
1.25/ - 100 = 65 - 0.4/
and so, with y = 100 as a result, and x = 25
f(25, 100) = 1537.5
so the reduction in f{x, y] is
2280- 1537.5 = 742.5
Note that if the function f(x, y) represents profit, then £742.5 is the
maximum price worth paying in order to remove the constraint.
Exercises 7.2 1 The constraint; y = 40 — 1.6x, when substituted into z, produces
z* = 1 OOx + 50(40 - 1.6x) + 2.5x(40 - 1.6x)

= 120x - 4x2 ^ 2000
So the conditions for a constrained maximum are
dz*
,=120 8x = 0
dx
d^z*
so from the first-order condition x = 15, and therefore from the

constraint y = 16, which values produce z = 2900 as a maximum.
' 2 From the, constraint y = 100 + 2x, substitution for y in z yields
z* = 20(100+ 2x) -0.1x2
= 2000+ 40x-0.1x2

so the first and second derivatives are
dz*
= 40-0.2x = 0
dx
-0.2 < 0
dx^
From the first-order condition x = 200, so from the constraint

y = 500. The maximum value of z is 6000.
3 From the constraint y = 20 — 0.5x substitution into z gives
z* = 50x2 000
the derivatives of which are
dz*
1 OOx - 4x3 = 0
dx
d^z*
= 100 - 12x2 < 0
dx2
The first-order condition can be reorganized as
x(100-4x2) = 0
so
X = 0, X = 5 and x = —5
At the second order the condition for a maximum is satisfied for the
values X = 5 and x = —5, both of which give a maximum value of
z, subject to the constraint, of 1625.
4 From the first constraint:
X = 80 — 1 Oy — 5w (1)
Inserting this relationship into the second constraint gives
-2(80- 10y-5w) + 30y-f2.5w= 115
which simplifies to the relationship
w = 22 — 4y
Inserting this into (1) produces
X = 1 Oy - 30
Substituting into z for w and x in terms of y and simplification gives
z* = 320y - 40y2

so
320 - 80/ = 0
-80 < 0
The first-order condition solves for y = 4, giving a maximum os the

second-order condition is automatically satisfied. From the earlier
equations, if y = 4, then x = 10 and w = 6, giving a constrained
maximum of z of 640.
5 From the constraint
X 2y
w = —h — 2
3^5
which substituted into the objective function gives
f* = 1000-^ + 4xy+ 10x-3y2
which is to be maximized with respect to both x and y. At the first

order
fx* = -^ + 4x+10 = 0
fy" = 4x - 6y = 0
which equations solve simultaneously for x = 15 and y = 10 (so

that, by implication, w = 7). At the second order
to,' = -^<0 f„- = -6<0 fy,’ = 4
so that the discriminant condition
(- 10/3)(-6)-(4)2=4 >0
is satisfied and the function achieves a maximum value of
f(w, X, y)= 1000-5(7)(15) + 6(15)(10)-3(10)2 = 1075
6 (i) With the right-hand side = 120, the problem solves for
X = 240, y = 240, f{x, y) = 57600
' With the right-hand side = 121, the problem solves for
X = 242, y = 242, f(x, y) = 58 564
so that the marginal valuation of the resource represented by the

right-hand side is 964, since
58564-57600 = 964

(ii) With the right-hand side = 100, the problem solves for
x= 12, x= 17.5, f(x, /)=522
With the right-hand side =101, the problem solves for
X = 12.2, X = 17.625, f(x, y] = 528.25625
so that the marginal valuation of the right-hand-side resource is
528.25625 - 522.5 = 5.75625
(iii) With the right-hand side = 100, the problem solves for
X = 75, X = 50, f{x, x) = 12 600
With the right-hand side = 101, the problem solves for
x = 75.5, x = 50, f(x, x) =12700
so that the marginal valuation of the resource is 100.

(iv) With the right-hand side = 600, the problem solves for
X = 255, x = 90, z = 6975000
With the right-hand side = 601, the problem solves for
X = 255.425, X = 90.15, z = 6 998 269.375
so the marginal resource valuation is 23 269.375.
Exercises 7,3 1 The Lagrangian is
X/ /) = 20x + 15x - XX - x^ - 0.5x^ + /(20 - x-2y)
so that the conditions at the first order are
Lx = 20 - X - 2x - / = 0
Ly = 15 — X — X — 2/ = 0
L, = 20 - X - 2x = 0
From Lx,
/ = 20 - X - 2x
which when substituted into Ly produces
15 - x-x-2(20-x-2x) = 0
which can be rearranged as the following equation:
6x + 2x = 50 (Ij
Now writing the Lj constraint as
X + 2x = 20

and subtracting from (1) produces
5x = 30
so that X = 6, and by substitution back y — 7 and / = 1. The

maximum value of f(x, y] is therefore
20(6) + 15(7) - (6)(7) - 6^ - 0.5(7)^ = 122.5
The sufficient second-order condition v/as that
2 2
E = f-xx(Sfy) ~ '7-^xyQxQy + f-xy(Sfx) < 0
In the present case
f-xx = ~2, Lyy = — 1, Lxy = —1, Qx ~ ~ \ 9y ~
So E evaluates as
£ = -2(-2p-2(-1)(-l||-2) + (-1|(-l|^
= -8 + 4-1 =-5<0
as required at a constrained maximum.

2 (i) The Lagrangian is
yI /) =1200x + 400/ — 20x/ — 40x^ — 2.5/^

+ /(150- lOx-5/)
and conditions at the first order are
ly, = 1200- 20/ - 80x - 10/ = 0

= 400 - 20x - 5/ - 5/ = 0
/.,• = 150 - 1 Ox - 5/ = 0
which solve for
X = 5, / = 20, / = 40 and f{x, /) = 10 000
(ii) First-order conditions would be unchanged except for 151

replacing 150 in Lj. The solution values then are
X = 4.9, / = 20.4, /■ = 40 and f{x, /) = 10 040
so that the value of an extra unit of resource is 40.

3 The Lagrangian is
L = 2x^ + 6x/ + 4.5/^ — 1 OOx — 200/ + /(40 — 2.5x — 1.25/)
and conditions at the first order are
Ly, = 4x + 6y- 100-2.5/ = 0

Ly = 6x + 9/-200- 1.25/= 0
=40-2.5x- 1.25/ = 0

For the second-order condition
Lxx = 4, txy = 6, Lyy = 9,g^ = -2.5, gy = 1.25
so
F = 4( - 1.25)^ - 2(6)( - 2.5)( - 1.25) + 9(2.5)^

= 6.25-37.5 + 56.25 = 25
and therefore the solution of the first-order conditions corresponds to

a minimum. The conditions solve for
X = 5.25, / = 21.5, / = 20 and f(x,/) =-2012.5
4 The Logrongion in this cose is
£ = 44x+ 101x + 9xy + 5x2 - +/(29 - 3x - 2/)
and at the first order
Ix = 44 + 9/ + 1 Ox - 3/ = 0
Ly = 101 +9x- 16/-2/ = 0
Lj = 29 - 3x - 2/ = 0
which conditions solve for
X = 7, / = 4, / = 50 and f{x, y] = 1081
At the second order
/.XX = 10, /.xy = 9, Lyy = -16, gx = -3, gy = -2
so
E = 10( - 2)^ - 2(9)( - 3)( - 2) - 16( - 3)^

= 40- 108- 144 = -212 <0
so that a constrained maximum occurs. With 30 as the right-hand-

side value the results are (approximate values)
X = 7.311,/ = 4.033, / = 51.137 and f(x, y) = 1131.57
5 The Lagrangian is
L = x^x + j(20 - X - x)
so the first-order conditions are
Lx = 3x'^y - / = 0
Ly = x^ - / = 0
L, = 20 - X - X = 0
From Lx and Ly, eliminating / we obtain
x^ - 3x^x — ^ (1)

from Lj,
X = 20-x
Substituting this relation into (1) gives
-3x2(20-x) = 0
i.e.
4x2 _ ^0x2 = 0
which factors as
x2(4x-60) = 0
which is satisfied only for x = 0 and x = 15. At the second order
Lxx = Sxy, Lxy = 3x , Lyy = 0
also
gx = -1 and gy = -1
So for a maximum
E=|6xy|(-1|^-2(3x2l(-l)(-1)<0
when X = 0, f = 0 and the sufficient form of this second-order

condition is not fulfilled (the graph of x2(20 — x) has a point of
inflection at the origin). When x = 15, y = 5 and E evaluates as
£ = 6|151|5)|-I|"-2|3)(15)^|-1||-11
= 450- 1350 =-900 <0
so a constrained maximum is produced. The function has the value

16 875.
Exercises 7.4 1 (i) The derivative is
^ = 24 - 3x
dx
so trying x = 0, dz/dx = 24 > 0, so that x = 0 cannot produce the

sign-restricted optimum. Now trying dz/dx = 0, x = 8, so that the
conditions are fulfilled and z = 216.
(ii) The derivative is
^ = -21 -7x
dx
so trying x = 0, dz/dx = -21 < 0, as required, so x = 0 could

produce the sign-restricted optimum. Trying dz/dx = 0, the value of x
is 8. Thus the conditions are fulfilled only at x = 0, where z = 100.

2 (i) The derivative is
^ = 120 + 54x - 3x2

dx
When X = 0, dz/dx = 120 > 0, so that x = 0 cannot produce the
sign restricted optimum. Now try dz/dx = 0. The equation solves for
X = —2 or X = 20. Thus the conditions are fulfilled only at this point
and the sign-restricted maximum of z is 6200.
(ii) The derivative here is
-2880x +384x2 _
dx
At X = 0, the derivative dz/dx = 0, so that in this case a turning
point occurs at the boundary and x = 0 could produce the sign-
restricted optimum. The other values of x which satisfy dz/dx = 0
are found by solving
-2880+ 384x-12x2 = 0
which solves for x = 12 and x = 20. The possibilities to be

enumerated are then as follows:
X z
0 8000
12 -40384
20 -24000
so the sign-restricted maximum is at x = 0.

3 (i) The derivative is
^ = 2.4x- 12
dx
Trying x = 0, dz/dx = — 12, so that x = 0 cannot produce the sign-

restricted minimum. Now trying dz/dx = 0, this solves for x = 5,
which is acceptable and at which point
z= 1.2(25)- 12(5)+ 30 = 0
so that the constrained minimum is zero at x = 5.

dz
— = 14 + 84
dx
At X = 0, dz/dx = 84 > 0, as required, so that x = 0 could

produce the sign-restricted minimum. Now trying dz/dx = 0, the
resulting value x is of x = -6. Thus the conditions are fulfilled only
at X = 0, where z = 50.
4 (i) The derivative here is
^ = 3x2 - 27^ _ 30
dx

Trying x = 0, dz/dx = —30 < 0, thus ruling out the origin os the
sign-restricted minimum. Now trying dz/dx = 0, the equation solves
for X = — 1 and x = 10. Thus the conditions are fulfilled only at
X = 10, giving z = 0 as the constrained minimum of the function.
(ii) The derivative here is
j
= 3x3 - 330^2 _ 3ooOx
dx
At X = 0, dz/dx = 0, so this is a possible sign-restricted optimum.

The other values of x satisfying dz/dx = 0 are the roots of the
quadratic
3x2 _ 330^ _ 3000 = 0
i.e.
x2 - n Ox - 1000 = 0
which solves for x = 10 and x = 100. The possibilities can be

enumerated as follows:
X z
0 0
10 47500
100 -20000000
so the function takes its sign-restricted minimum atx = 100.

5 The derivatives are
fx = —300 +10/ — 60x

fy = 40 + 1 Ox - 2/
(i) We require that
f.<0
xfx = 0
X > 0
fy=0
Trying x = 0, fy = 0 solves for y = 20, at which value fx — — ] 00, so

this is a possible optimum. With fx = 0, fx = 0 and fy = 0 solve for
X = —10 and y = —30, which is infeasible. So the optimum here is
X = 0, / = 20, f(x, x) = 500
(ii) It is necessary that
^x = 0
fy<0
yfy = 0
X>0

Trying x = 0, fx = 0 solves for x = -5, which is acceptable and at
which value fy emerges as —10. So a possible optimum is / = 0,
X = -5, f(x, y] = 850. Trying fy = 0 along with fx = 0 produces
the infeasible position already identified in (i). So the optimum is
X = -5, X = 0, f{x, y] = 850
(iii) The outcomes are
1 x = 0 y = 0 fx = —300 fy = 40
2 x = 0 fy = 0 fx = -100 x = 20
3 fx = 0 y = 0 X = —5 fy = —10
4 f^Ô fy = 0 x = -10 x = --30
so the solution with both variables sign restricted is outcome 2, i.e.
X = 0, X = 20, f{x, x) = 500
(iv) The conditions required are
f^ = 0 fy = 0
which, as we have already seen, solve for
X = — 10, X = —30 when f{x, y) = 1000
which is the overall maximum of the function.

6 (i) The conditions are
dz
|50-x|^ = 0
dx
(50 - x) > 0
The derivative is
150-2x
dx
Trying 50 — x = 0, dz/dx = 50 > 0, so this point fulfils the

necessary conditions. Trying dz/dx = 0, x = 75, so 50 - 75 < 0,
and therefore this is not a possible constrained optimum (it is, of
course, the unconstrained maximum). The solution is therefore
X = 50 when z = 5500
^ = 2000 - 16x
dx
when 100 — X = 0, dz/dx > 0, so that this is a possible optimum.

When dz/dx = 0, 100 — x = —25, thus ruling out this possibility. The
constrained optimum is therefore at x = 100, when z = 121 000.

(iii) The derivative is
^ = 1600 - 4x
dx
At X = 450, dz/dx = -200, thus ruling out this position as a

possible optimum. When dz/dx = 0, x = 400, so that
450 - X = 50 > 0, and this is the constrained optimum (and is also
the unconstrained optimum).
7 (i) The derivative is:
^ = -528 + 90x - 3x2

dx
Follow the five-step procedure. At x = 0, dz/dx = —528, so this is

a possible optimum. At x = 20, dz/dx = 72, so this too is a
possible constrained optimum. Now trying dz/dx = 0, the roots are
X = 8 and x = 22. Clearly, x = 22 is infeasible. Evaluating z at the
points recorded:
X z
0 2500
8 644
20 1940
so that X = 0 is the optimum.

^ = 195x - 900 - 3x2

dx
At X = 0, dz/dx = -900, so this is a possible optimum. At x = 50,
dz/dx = 1350, so this is also a possible constrained optimum. Now
dz/dx = 0 solves for x = 5 and x = 60. Ruling out x = 60 and
evaluating zat the three recorded points:
X z
0 0
5 -2187.5
50 73 750
so that X = 50 is the optimum.

(iii) The derivative is
^ = 6x2-210x+ 1500
dx
At X = 0, dz/dx = 1500, so this is not a possible optimum. At
X = 30, dz/dx = 600, so this position is a candidate. When
dz/dx = 0, X = 10 or X = 25. Evaluating z at the three possibilities:
X z
10 5500
25 2125
30 3500

Therefore x = 10 is the constrained optimum.
8 (i) The derivatives ore
4 =-2x-2.5x+10
fy = -2.5x - 4/ + 2
which when equated to zero solve for
x = 20, y = —12, giving f(x,/) = 188
which is the optimal position.

(ii) The Lagrangian is
L= 100 - x^ - 2.5xy - 2/^ + 1 Ox + 2/ + /(20 - 2x — /)
and the first-order conditions are:
= -2x-2.5y+ 10-2/ = 0
Ly = —2.5x — 4y + 2 — j = 0
which solve for
x= 14.5, x = -9, /•= 1.75
giving f{x, y] = 181. At the second-order
Lxx = —2, Lyy = —4, Lxy = —2.5
and from the constraint
9x = -2, gy = -]
so that the second-order condition
E = LxxQy — 2Lxygxgy + l-yyg^

= -2(-]f-2(-2.5)(-2)(-])-4(-2f
= -2+ 10- 16
= -8 <0
as required for a constrained maximum.
(iii) Using the information from the constraint that
X = 20 - 2x
since x ^ 0/ follows that this can be expressed in terms of x as

X < 10. Substitute for y in f{x, y) to get
z = 116x — 4x^ — 660
for which the derivative is
^ = 116 - 8x
dx
492
Trying x = 0, dz/dx = 116 > 0, so this cannot be the optimum.
Trying dz/dx = 0, the equation solves for x = 14.5, which is (now)
infeasible. Trying x = 10, dz/dx = 36 > 0, as required for the
upper bound to be a possible optimum. Since this is the only
possibility, the solution is
X = 10, / = 0, f(x, y) = 100
Exercises 7.5 1 The Lagrangian is
L = x^+7.5x^ - 150x+ 1000 + /(100-5x-40)
and at the first-order the conditions are that
3x2 + 15x - 150 - 5• ^ 0

100 - 5x - 40 - s = 0
sj = 0
Trying s = 0, x = 12 and / = 92.4, so this point is a possible

optimum. Now trying / = 0, the equation solves for x = — 10 and
X = 5. At X = — 10, s = 110, so this point too is a possible optimum
(recall that there are no sign requirements). At x = 5, s = 35, so this
point also cannot be ruled out at the first order. Evaluating z at the
points obtained;
X z
12 2008
5 562.5
-10 2250
so the constrained maximum is at x = — 10.

2 The Lagrangian is
L = 40x - 8x2 ^ ^ QQ^ _ ^ _ 2x - 4y]
and the first-order conditions can be expressed as
40 — 16x -f 4/ — 2/ = 0
4x + 80 - 10/ - 4/ = 0
40 - 2x - 4/ - s = 0
sj = 0
s > 0, / > 0
Trying s = 0, from the third equation x = 20 — 2y, and substitution

into and solution of the first two equations for y and / leads to
X = 4, y = 8, / = 4, z = 480
Trying / = 0, the first two equations can then be written as
16x — 4y = 40
-4x + 1 Oy = 80

which solve for y = 10 and x = 5, implying that s = -10. The
constraint would be violated at this position. The constrained
optimum is then
X = 4, y = 8, / = 4, z = 480
3 The Lagrangian is
L = 120x - 2.5x^ + 3xy - 3.5y^ - 60y + /(35 - 3x - 2y)
120 - 5x + 3y - 3/ = 0
3x — 7y — 60 — 4/ = 0
35 — 3x — 7y — s = 0
5/ = 0
s > 0, / > 0
Trying s — 0, the conditions solve for
X = 15, y = -5, i — 10, f(x, y) = 1225
Trying / = 0, the first two equations solve for
x = 330/13, y = 30/13
but the constraint-related equation then produces s = —595/13, so

the constraint would be violated at this position. The constrained
optimum is then
X = 15, y = -5, / = 10, f(x, y) = 1225
4 The Lagrangian is
L = 1 OOx - 3x^ -f- 4xy - 2.5y^ - 55/ + /(90 - 2x - 5y)
100 — 6x + 4y — 2/ = 0
4x — 5y — 55 — 5/ = 0
90 — 2x — 5y — s = 0
sj = 0
s > 0, j > 0
Trying s = 0, the conditions solve for
X = 23, y = 8.8, but / = -1.4
meaning that this position cannot be optimal. Trying / = 0, the

conditions solve for
X = 20, y = 5 and s = 25

This is the only point fulfilling the conditions, and therefore produces
the constrained maximum of f(x, y) = 862.5.
Exercises 7.6 1 (i) The Lagrangian is
L = 32x + 16/ — + 150 + /(30 — 2x — /)
32-2x-2j + t = 0
16 — 2/ — /+u = 0
30 — 2x — / — 5 = 0
fx = 0, uy = 0, sj = 0
■X > 0, / > 0, / > 0, f > 0, u > 0, 5 > 0
Enumerating the various combinations of variables set to zero, the

results are as follows:
zero group solution values (non-zero)
X t X u s /
(a) tu s 12 6 4
(b) fuj 16 8 -10
(c) fys 15 -15 1
(d) fyi 16 -16 -2
(e) XUS -120 30 -44
(f) X u 1 -32 8 22
(g) Xys No solution
(h) xyi -32 -16 30
From the table, the optimum position is (a), at which point
X = 12, X = 6, /■ = 4, giving f{x, y) = 450
since this is the only complete solution to the conditions.

(ii) Here the unconstrained optimum (b) can be achieved (since the
sign requirements happen to be met at (b)) and f{x, y] = 470.
So the cost of the constraint is 470 — 450 = 20.
L = 63x - 0.8x^ - 35/ - 0.4/^ + xy + j(35 - x - 0.8/)
and the first-order conditions in complementary slackness form are
63 - 1.6x + X - / + f = 0
—35 — 0.8x + X — 0.8/ + u = 0
35 — X — 0.8x — s = 0
fx = 0, uy = 0, sj = 0
X > 0, y > 0, j > 0, t > 0, u > 0, s > 0
Enumerating the various combinations of variables set to zero, the

results are as follows (approximate values in case (a)):

zero group solution values (non-zero)
X t X u s /
(a) tu s 36.31 -1.64 3.27

(b) tu i 55 25 -40
(c) tX s 35 5.6 7
(d) fyi 39.375 -4.375 -4.375
(e) XUS -19.25 43.75 -87.5
(f) X u j -19.25 -43.75 70
(g) xys No solution
(h) xyi -63 35 35
The constrained optimum position is identified as (c), since this is the

only complete solution to the conditions. At this point
X = 35, / = 0, / = 7 and f(x, y) — 1225
(ii) Without the requirement to meet the constraint the best position
is (b), where the sign requirements happen still to be met and
where
X = 55, X = 25, f{x, y] = 1295
The constraint therefore costs 1295 — 1225 = 70.

(iii) No change in f{x, y).
(iv) Combination (a) now becomes feasible. At this point
fix, y] ^ 1229.58
So the benefit of the removal of the sign restriction on y is 4.58.
Exercises 7.7 1 (i) The Lagrangian is
L = 40x - x^ + 60/ - 2/^ + /(120 - 5x - 4y)

+ M140- lOx-y)
and the Kuhn-Tucker conditions are
= 40-2x-5/- 10k < 0

Ly = 60 — 4y — 4j — k < 0
xLx + yLy = 0
X > 0, X > 0
/., = 120 - 5x - 4x > 0
4 = 140 — 1 Ox — X > 0
I'Lj + /c4 = 0
j > 0, k > 0

(ii) When
88 TOO . 192 , 632

^ = T'>' = ^'' = 4TO''^ = 490
SO that the first four conditions are satisfied. We also obtain
Li = 0, 4 = 0
so that the last four conditions are also clearly satisfied.

(iii) The unconstrained optimum is when
4 = 40 - 2x = 0
4 = 60 - 4/ = 0
that is, where x = 20 and y = 15, giving f{x, y] = 850. At the

constrained optimum:
38 896
X) ^ 793.80
49
so that the possible improvement in f(x, y) is 56.2.

2 The conditions are
4 = 12 000 - 1.2x^ -3i-4xk<0

4 = 2 000000 -8y^ -7j-k<0
x4 + x4 = 0
X > 0, X > 0
4 = 21 -3x-7x>0
4 = 20 - 2x2 - X > 0
jLj + /c4 = 0
j >0, k >0
L = 1 OOx - 2x2 ^ ôy - x2 + /(65 - 2x - y) + i^(38 - 0.5x - y)
for which conditions at the first order in complementary slackness

form are
100-4x-2/-0.5k + f = 0
60 — 2y — / — /c + u = 0
65 — 2x — y — s = 0
38 — 0.5x — y — r = 0
tx + uy + js + kr = 0
X, y, f, u, /, k, r, s>0
(ii) The sixteen combinations of zero valued variables, and the

resulting values of the other variables, are as follows:

zero solution values (non-■zero)
group
X t X / s k r
(a) tu s r 18 29 18 -16
(b) tusk 20 25 10 3
(c) tu j r 24 26 -9 8
(d) fujk 25 30 -15 -4.5
(e) fysr no solution
(f) tysk 32.5 -75 -15 21.75
76 -468 -87 -408
(g) f X / »■
(h) tyjk 25 -60 150 25.5
(j) X u s k -280 65 -70 -27
(k) X u j r -108 38 27 -16
(1) X u j k -100 30 35 8
(m) X y s r no solution
(n) X y s k no solution
(0) xyjr no solution
(p) X / / k -100 -60 65 38
(iii) The optimal solution is (b), where f(x, y] = 2075. The

unconstrained optimum is (d), where f(x, y) = 2150. The cost of
the constraints as a whole is therefore 75.
Exercises 7.8 1 (i) The costs for the products are
unit cost of X = 1 (5) + 1 (10) = 15

unit cost of / = 1 (5) + 2(10) = 25
so
C(x) = 15x, C(y) = 25y
Revenues from the sale of the goods are
R{x) = xp(x) = 95x - 0.5x^
/?(/) = XP(X) = 85/ - 0.5/2
so that overall profit is
F = /?(x) - C(x) + /?(/) - C(x)

= 95x - 0.5x2 _ 15^ ^ 85/ - 0.5/2 _ 25^
= 80x — 0.5x2 ^ _ Q 5^2
so that the problem in full is
Maximize F = 80x — 0.5x2 _j_ _ o,5/2

subject to X + / < 100
X + 2/ < 140
X > 0, / > 0

for which the Lagrangian is
L = 80x — 0.5x^ + 60/ — 0.5/^

+ /(100 - X - x) + ic( 140 - X - 2/)
and the conditions in complementary slackness form are
S0-x-j-k+f = 0
60 — / — / — 2/c + u = 0
100 — X — / — s = 0
140-x-2/-r = 0
tx + uy + js + kr = 0
X, y, f, u, j, k, r, s>0
(ii) The results of applying the combinations of zero variables are

as follows;
zero solution values (non-zero)

group
X f y u / s k r
(a) fu s r 60 40 20 0
(b) tusk 60 40 20 0
(c) tu j r 68 36 -4 12
(d) tu j k 80 60 -40 -60
(e) ty s r no solution
(f) tys k 100 -80 -20 40
(g) fy/'' 140 -180 -40 -60
(h) fyjk 80 -60 20 60
(j) X u s k -120 100 -40 -60
(k) X u jr -85 70 30 -5
(1) xujk -80 60 40 20
(m) Xys r no solution
(n) Xys k no solution
(o) xyjr no solution
(p) xyjk -80 -60 100 140
Cases (a) and (b) produce the same solution to the conditions.
This simply means that the corner point at the intersection of
constraints one and two is also a point of tangency (with the first
constraint). The optimal production levels for the two products are
therefore:
X = 60 and / = 40
giving a maximum overall profit of F = 4600. Note that the

unconstrained optimum is case (c) (the values of / and k both being
zero), and profit here would be F = 5000.

Chapter 8
Exercises 8.1 1 (i) (ii) and (iii) and (v) are antiderivatives of 4x^ since each is x"‘±
constant and would give 4x^ on differentiation. Cases (iii) and (iv)
have a term in x with a non-zero coefficient which would not vanish
on differentiation.
2 Cases (i), (ii) and (iv) are antiderivatives of x^ - x. Case (iii) is not
an antiderivative of x^ — x, since its derivative is 2x^ — 2x.
Exercises 8.2 1 (i) 3x+ k

(ii) -7.5x + k
(iii) \/3x -f k
(iv) Ux + k
2 (i) +k
10
,-4
(iii) - k, i.e. + k
4x'^
4
(iv) ^x^/^-hk
D
.1.2
(v) +k
1.2
, 0.6
(Vi) +k
0.6
(vii) 2x5 + k, i.e. 2^yx + k
yV3+l
(viii) + k
\/3 + l
j.2n + i
H 2iTTT + ‘^
3 (i) x"* + x^ +
(ii) X* - 0.5x^ -f 0.25x^ + k
(iii) 4x'^ + x'^ - x“^ + k

(iv) 2x^-^ — O.lx""'® +/c
(v) n^x^ + ' +k
4 (i) ]2 J x^dx= 12y = 4x2
|ii| 125 1 (2x'‘ - x’idx = 185 -
= 50x5 _ i2.5x’°

a o t> o ,
5 2 + 2^ +CX + k
(iii) ^ ^ +k
(iv) P + + k
(v) x“^ + k
Exercises 8.3 1 J (2000 — 10x)dx = 2000x — 5x^ = total revenue
since in this context the appropriate value of k is zero. So
2000x - 5x2
price = = 2000 - 5x
X
2 Total cost, TC(x), is given by
J (] 00 + X + 0.3x2)cIx
so that
TC(x) = 1 OOx + 0.5x2 0.1 +k
Nov/ when x = 20, TC(x) = 4000, so that
TC = 100(20) + 0.5(400) + 0.1 (8000) + k = 4000
so
3000 + k = 4000
= 1000
and the total cost function is
TC(x) = 1 OOx + 0.5x2 ^ 0.1 x2 + 1000
3 (i) Total revenue, TR, is maximized when MR = 0, that is, when
480- 10q = 0
q = 48
To find the corresponding value of price, TR will be given by
J MR(q)dq = J (480— 10q)dq
, =480q-5q2
the constant of integration being zero in this context. So, since
TR = pq
p = 480 - 5q
and when q = 48, p = 240.

(ii) The firm's profit, IT, is given by
TT = TR - TC
Now total costs will be
TC = y MC(q)dq
= I (210 + 5q)dq
= 210q + 2.5q2 + i^
when q = 10,
TC = 210(10)+ 2.5(100)+ ^ = 2780
so
1^ = 430
Profit will then be
TT = 480q - 5q2 - 21 Oq - 2.5q^ - 430
= 270q-7.5q2 -430
and
drr
= 270- 15q = 0
d^
which solves for q = 1 8. Checking at the second order.
which confirms the maximum. The level of profit when q = 18 is
TT = 270(18) - 7.5(18)^ - 430 = 2000
and the profit-maximizing price is;
p = 480-5(18) = 390
Exercises 8.4 1 (i) Second degree, third order.

(ii) First degree (linear), second order.
(iii) Third degree, third order.
(iv) First degree, first order.
(v) Fourth degree, first order.
(vi) First degree, fourth order.

2 (i) y = 2x4+ 3x3+2.5x2 + 1 Ox + it
(ii) y = 0.5x4 ^ 5^3 j^j^ + k
(iii) y = 1 OOx + x'° + k
(iv) y = 1 8x3 _ 0,5x4 _|_ 34_^ ^
3 (i) The general solution is
y — J - 2x + 1 )dx
= x'^ - x4 + x3 - x2 + X + /c
and the condition requires that
300 = 3^ - 34 + 33 - 32 + 3 + it
so k = 117, and the particular solution is therefore
y = x^ - x4 + x3 - x2 + X + 117
(ii) The general solution is
y = 0.25x4 _ q + 2x2 + 1 Ox + it
Insertion of the boundary condition produces
90 = 84 + it
so /c = 6, and the particular solution is therefore
y = 0.25x4 — 0.5x3 ^ 2x2 ] Ox + 6
(iii) y = —2x4 _|_ ■] q^3

(iv) y = 2x^ + 3x4 _ i q^2 ^ 20x
and in this case note that k = 0.
4 (i) The general solution here is
y = 10x-x"2 + 3x“^ +it
that is
in 1 3 ,
y=10x-^ + -+ k
X^ X
When X = 0.2, y = 35. That is
35 = 20 -L+A+fc
0.4 ^0.2
i.e.
35 = 20-2.5+ 15 +it
so that k = 2.5, and the particular solution is
y = 1 Ox — x“2 + 3x“^ +2.5

(ii) The general solution is
y = 4x“^ + 3x^ - x^ - 32x + k
Inserting the given condition that when x = 1, y = 0,
0 = 4 + 3-1-32+ A:
so that k = 26. The particular solution is therefore
y = 4x“^ + 3x^ - x^ - 32x + 26
5 (i) Since
dV
^= 10-30X
dx^
lOx- 15x2 + Jî

dx
and
y = 5x^ - 5x^ + k]X + k2
The boundary conditions give two simultaneous equations in k-\ and

k2-.
3k^ +k2=90
5ky + /c2 = 290
-2k = -290
so k] = 100, hence /c2 = -210, and the particular solution is
y = 5x^ — 5x^ + 1 OOx — 210
(ii) The differential equation rearranges as
^=24x- 12
dx^
so
^ = 12x2 - 12x + /ci

dx
Using the boundary condition to find k-[,
362 = 12(36)- 12(6) + fci
so that k] = 2 and
^ = 12x2 - 12x + 2
dx
which on integration produces
y = 4x2 _ ^^2 2x + /c2

Using the boundary condition on y,
672 = 4(216) -6(36)+ 2(6) +ic2
means that ^2 = 1 2 and the particular solution is therefore
X = 4x^ - 6x^ + 2x + 12
6 The equation rearranges as
d X
= 60
dx^
and, integrating,
60x + k]
so, with the given information, k] = 10, and integration gives
^ = 30x2 + 1 Ox + i^2
dx
This derivative is 60 when x = 1, so /c2 = 20. Integration then gives
X = 1 Ox^ + 5x2 ^ 20x + k2
which is 50 when x = 1, so that /c3 = 15 and the particular solution

is
X = 1 Ox^ + 5x2 ^ 20x + 15
7 Reorganization gives
2
= 9x2 - 1 80 + 900
so that
^ = ±3(x- 10)
dx
and, with integration, either
X = 1.5x2 _ 20x + k
X = — 1.5x2 ^ 20x + k
^ With the output range given, the variable component (variable

costs) is only plausible if positive, so that in this context
X = 1.5x2 _ 20x + k
Now when x = 100, y = 20 000, so
20000= 15 000 - 3000 +

Thus k, representing fixed costs, is 8000, and the relevant particular
solution is
y = 1.5x2 20x + 8000
Exercises 8.5 1 (i) Trying the substitution
u = 10 + x2
so that du = 2x and the integral in terms of u is
j u^6u = 0.25u'^ + k
and therefore, in terms of the original variable, x.
J {]0 + x^f2x6x = 0.25(]0 + x^f+ k

(ii) Using the substitution
u = 2x2 ^ + 15
it follows that
du = (4x + 6)dx
that is.
du
(2x + 3)dx =
The integral in terms of u is
i3 u4
= 0.125(2x2+6x+ 15)
(iii) Using the substitution
u = (x2 — 15)
it follows that
du = 2xdx
so that
4xdx = 2du
and the integral in terms of u is
/■ i , 4u2/2
/ 2u2du = —2-h k

which in terms of x is
(iv) Using the substitution
u = (6x2 _ ^ 20)
it follows that
du = (12x — 8)dx
so that
0.25du = (3x — 2)dx
and the integral required is
0.25udu = 0.1 25u2 + k
which in terms of x is
0.125(6x2 -8x +20)^ +
2 (i) Setting
u = 5x2 - 19
it follows that
du = 1Oxdx
so that
3xdx = 0.3du
and the integral required is
j 0.3u2du = 0.2u2/2 ^ ^ 0.2(5x2 - 19)^/^ + k
(ii) Let
u= 5 +X
then du = dx, and the required integral is
-4u-^du =u-^ + k = {5+ x)“^ + k
(iii) The required integral is
Letting
u=X—5
then du = dx and the integral is
J + k
= -(x-5r^ +k
= -1-^ + k
(x-5)
(iv) Let
u = 3x + 4
then du = 3dx, so dx = du/3 and the integral is
/ -2u ^du = 2u ^ + k
= 2(3x + 4)“’ +k
2
+ k
(3x + 4)
3 Let
u = bx + c
so
du
dx
and the integral is
td/b}u,-n+ 1
+ k
(1 -n)
+ k
b{] - n](bx + c)"'
4 (i) Let
u = x^ + 7x - 8
then du = (2x + 7)dx and the integral is
/ ^=-0-25u-^ + i^
=---^-j+k
(x2 + 7x - 8)^

(ii) Let
u= +X+ 1
so 26u = (6x^ + 4x + 2)dx and the required integral is
J 2u~^6u = -u~^ + k
=-^-2 + *^
(x^ + x^ + X + 1
Exercises 8.6 1 (i) (2x2 + i^)f°^"=^-(2x2 + i^)f°^^=2 = 98-18 = 80

Note: from this point on the constant of integration will be omitted.
(ii) (3x2 _ 3^j forx=-l _ j3^2 _ 3^j forx = -5 ^ ^ - 90 = -84
(iii) (0.3x2 ^ 0.5x2 ^ I Ox) ^=20 _ 0 = 2800
(iv) (8x1-5) forx=16 _ jqî.5j forx = 4 ^ 5]2 _ 64 = 448
21" (,-—j - f-—j

= 0 - ( - 200)
= 200
(ii) (50v/x) - (50Vx) = 250 - 100 = 150

(iii) n[(x2)f°^"=^-(x2)f°^"=2] = 117n
(iv) (0.2x5 ^ ^3 ^ 20x)^=° - (0.2x5 ^ ^3 ^ 20x)^=-^°
= 0-(-20000- 1000-200)
= 21 200
3 (i) The required area is given by the definite integral

/•x=10
/ (3x2 2x -f- 5)dx
Jx = 0
= (x2 +x2 +5x)^^'° - (x2 +X^ +5x)'^“°
= 1000+100 + 50-0
= 1150
(ii) The area is given by the definite integral

fX = -2
/ (x2 + 1 0x)dx
7x = -10
= (0.25x^ + 5x2)^=-2 _ (o.25x^ + 5x2)^=“^°
= 24 - 3000
= -2976

(iii) The area is given by the definite integral
r ^ (1.5x2 +12x + 30)dx

Jx^-5
= (0.5x2 ^ 6x2 ^ 30x)^=^ - (0.5x2 ^ 6x^ + 30x) x = —5
= 362.5 -(-62.5)
= 425
(i) In this case the definite integral is split into two parts:
10 /-IS
[ (x - 10)dx + [ (x - 10)dx
Js J]0
the absolute value being taken for each part. So between 5 and 10
the integral evaluates as
x=5
(0.5x2- 10x)^=^°-(0.5x2 lOx)
= 50- 100-(12.5-50)
= - 12.5
and the area regardless of sign in this interval is 12.5. In the second
part the curve is above the axis and the area is given by
(0.5x2 _ 10;() ^ = ’^-(0.5x2- 10x)^=’°= 12.5
So the total area is 12.5 + 12.5 = 25.

(ii) First find the roots of
f (x) = 3x2 _ ] 3^ ]6
to find the points where the function crosses the x axis. f(x) = 0 for
X = 1 and X = 5. So the interval should be divided at 1 and 5; the
total area is the sum of the absolute values of the definite integrals:
[ f(x)dx + [ f(x)dx + [ f(x)dx

Jo j] Js
The individual areas are
(x2 -9x2+ 15x)^=^ -(x2-9x2 + 15x)^=° = 7
(x2-9x2+15x)^=^-(x2-9x2 15x)^=^ = -32
(x2-9x2 + 15x)^ = ®-(x2-9x2 15x)^=^ = 81
so the required total area is
7 + 32 + 81 = 120
(i) The definite integral required is
fx = 20 /.x = 40
/ 20dx + / xdx
Jx = 0 Jx = 20

which evaluates as
(20x)^=2° - (20x)^=° + (0.5x2)^=^° - (0.5x2)^=20

= 400 - 0 + 800 - 200
= 1000
(ii) The required area is
^10 ^15
/ (4x - 20)dx + (x + 10)dx
Js J]0
= (2x2 _ 20xr=^° - (2x2 _ 20xr=^ + (0.5x2 ^ i0x)^=i5
-(0.5x2 + 10xr=^°
= 50+ 1 12.5
= 162.5
Exercises 8.7 1 (i) Using the trapezium rule with m = 5 (and therefore with
w = 0.5), the definite integral is evaluated as
T = 0.25 [f(2) + 2f(2.5] + 2f(3) + 2f(3.5) + 2f(4) + 2f(4.5) + f(5)]

= 0.25(26 + 79+112+151 +196 + 247+ 152)
= 0.25(963)
= 240.75
(ii) By Simpson's rule,
S = 0.5/3[f(2) +4f(2.5) + 2f(3) +4f(3.5) + 2f(4) +4f(4.5) + f(5)]

= 0.5/3(26+ 158+ 112 + 302 + 196 + 494+ 152)
= 0.5/3(1440)
= 240
(iii) r ^6x2 + 2)dx = (2x2+2x)''=^-(2x2+2x)^=2

Jx = 2
= (250+10) - (16 + 4)
= 240
Note the precise accuracy of Simpson's rule for a quadratic function.

2 (i) Using the trapezium rule as specified,
m=4
and so
w = (3 - 1)4 = 0.5
and
T = 0.25[f(l) + 2f(l .5) + 2f(2) + 2f(2.5) + f{3)]

= 0.25(1.5 + 9.875 + 30 + 73.875 + 77.5)
= 0.25(192.75)
= 48.1875

(ii) Using Simpson's rule,
S = 0.5/3[f(l) + 4f(] .5) + 2f(2) + 4f(2.5) + 2f(3)]

= 0.5/3(1.5+ 19.75 + 30+ 147.75 + 77.5)
= 0.5/3(276.5)
= 46.08^
(iii) The exact value of the integral is 691 /15 = 46.066.
Chapter 9
Exercises 9.1 1 (i) GDP(f) = 350(1.03)', where t = 0 for 1998.

(ii) (a) For 2000 t = 2, so that
GDP(2) = 350(1.03)^ = 371.315
(b) For 2004 t = 6, so that
GDP(6) = 350( 1.03)* « 417.918
2 (i) Letting V(t) represent the value of the car after f years, the
exponential function is V(t) = 10000(0.75)^
(ii) (a) For t = 3, V(f) « 4219
(b) For t = 6, V(t) 1780
3 (i) Future value is given by FV(t) = 750(1.08)^, so that
(a) With t = 5, FV(5)«i 1102
(b) With f= 10, FV(IO) « 1619
(c) With t = 20, FV(20) « 3496
(ii) Here FV(t) = 750(1.16)^, so that
(a) With f = 5, FV(5) « 1575
(b) With t= 10, FV(10)« 3309
(c) With t = 20, FV(20) 14596
Note the dramatic difference in the results over the 20-year horizon
between the two interest rates.
4 Letting y — debt outstanding;
(i) y = 5000(1+0.24) =6200
(ii) y = 5000(1+0.12)^ =6272
(iii) y = 5000(1 +0.06)^ =6312
(iv) y = 5000(1 +0.02)^^ = 6341
5 With a principal of £800, the future value is FV(t) = 800e° so
that
(i) Whent = 2, FV(2) «£977
(ii) When t= 10, FV(IO) !:a£2715
(iii) When t = 20, FV(20)«£5911
Exercises 9.2 1 (i) ^ = 7e^^

dx
(iii ^=-5e--

(iii) ^ = 0.2e°'2^
dx
dy e-o.ix
(iv)
dx
dy 3e^3x
(v)
dx
dy gO.snx
(Vi)
dx
CNX
2 (i) f"(x)
0)
II
(ii) f"(x) = -2e-0'^
(iii) P'(x) = 16(e^^ —
3 (i) P"(x) = 4eV'2>‘
(ii) p'(x) = -e-^
Note that —e“^ is its own second derivative.
6.25s-»='<
4 (i) Using the product rule,
f'(x) = + 2x2e^^ = 2e^^(] + 2x)
(ii) Using the product rule,
f'{x) = 3x^6^^ + x34e^^ = xê^îa + 4x)
(iii) Using the product rule,
f'(x) = e^(l + e^) + (x + e^)e^ = e^(l + x + 26^^)
5 (i) Both the product and chain rules are involved. The derivative is
^ = e^^[3(3x + x^)^(3 + 4x3)] ^ 5e5x(3^ + ^4)3

dx
After expansion and rearrangement, this can be written as
x2(x3 + 3)^(5X^ + 1 2x3 + 1 5X + 9)g5x
which, at X = 1, evaluates as 656e^.

(ii) Again using the product and chain rules
^ = 2(2x + 1 )(x3 4 x)(2e3^ - e^^) + (x^ + x)^(6e3^ - 26^^)

-
which at X = 2 evaluates as 336e'^ - 1326"^.

This derivative can be reorganized as
2x(x + l)(3x3 + 7x + 2)e3^ - 2x(x + l)(x3 + 3x + l)e^^
6 |i)
dx

CM ^
|ii)
dx
iii) ^ = 4xe
dx
(iv) ^=(20x-30)e^°^'-2°^-^°
dx
M ^ = £^1
' ' dx 2v/x
(vi) ^ = (8x^ - 24x^)e^^“®^^

dx
7 (i) Using the product rule,
f'{x) = (2x^ + 4x^)xe°^^' + eO^^'(8x3 + 20x^)
which at X = 1 evaluates as 34e° '^.
(iii) f'(x) = nx"-’e^"

f"(x) = nx"- ’ (nx"-1 e^”) + e^"[n(n - 1 )x"-2]
and where n = 3 and x = 2
f"(x) = 3(2)2[3(2)ê®] +e®[6(2)] = 156e®
Exercises 9.3 1 (i) 0.2e^^ + k

(ii) 4e-°25x + j^
(iii) +k
(iv) 2x^ + 2e^ +k
2 (i) 0.5e^' + k
(ii) + k
(iii) 2e'^^ + k
3 (i) The indefinite integral is
2e° +k
so the required definite integral is
(2e0.5x)x=4 _ (2e05x)x = 2 ^ 2(e^ _ e) « 9.34
(ii) The definite integral is

|g0.1x2 + 0.8xjX = 3 _ |g0.1x2+0.8xjx=l ^ ^3.3 _ ^0.9 _ 24.65
(iii) The definite integral is
(0.256^^'+^)^ = ^ - (0.256^^'+^)^=° = 0.25e^" + 2 _ o.25e°

= 0.25(e^n + 2 _ ij
^529705

(iv) The definite integral is
(_ = e-' + 1 ^ 0.6321
4 (i) The equation rearranges as
^ = 4e2^ - 1 Ox
dx
so
/ = — 5x^ + k
and, since y = 2e^ when x = 1,
2e^ = 2e^ — 5 + k
therefore k = 5, and the particular solution is
X = 26^^ -5x2+5
(ii) The equation reorganizes as
-20e-°2^ + 0.18x2
dx
for which the integral is
X = 100e-° 2^ + 0.06x2 + k
and when x = 10,
X = 100e“2 + 60
so k = 40, and the required particular solution is
X = 1 OOe-O-2^ + 0.06x2 40
Exercises 9.4 1 (i) x=10>' (ii) y = 5^ (iii) y =

2 (i) x = logToX (ii)x = log^x (iii)q=log2P
3 (i) x= 2 (ii) x= 1.5 (iii) y =-3
(iv) X = 0 (v) X = -2 (vi) X = 2
4 (i) 3+4 = 7
(ii) log3 729 — log3 81 =6 — 4 = 2
(iii) 2 logo.5 8 = 2(3) =6
5 The doubling time is x, where:
In2 0.6931
(i) X = = 17.67 years
In 1.04 “0.0392
In 2 0.6931
(ii) X = = 9.01 years
In 1.08 “ 0.0770
In 2 0.6931
X = 3.80 years
(iii)
hrr2 “0.1893 ■

The doubling time is x, where:
In 2 0.6931
(i) x= = 6.93 years
0.1 0.1
In 2 0.6931
II x= = 3.47 years
0.2 0.2
The 'halving time' is x, where:
In 0.5 -0.6931
X = = 34.31 years
In 0.98 - 0.0202
In 0.5 -0.6931
(ii) x = = 13.51 years
In 0.95 -0.0513
In 0.5 -0.6931
X = : 6.58 years
In 0.9 -0.1054
8 The half-life is x years, where
In 0.5
69 315 years
- 0.00001
Exercises 9.5 , (i|
^ 4
(ii)
dx X
(iii) = x^ —-I-2xInX = x(l -f2lnx)

dx X
(iv) ^ = (4x - 2x^) — + (4 - 4x) In x

dx X
= 4 — 2x -f (4 — 4x) In x
^ ,, ,,, , 4x^ 4
2 (i| = X
100 1
|ii| f'M =
lOOx X
20x3 - 18x^ - 14x

|iii| I'M =
5x'^ — 6x3 _ 7x2
20x^ - 18x- 14
5x3 _ 5x2 _ 7^
3 (i) Using the quotient rule:
dy _ (x^ -t-x)(l/x) - (2x + l)loggX

dx (x2 + x)^
_ (x q-1) - (2x-f l)logeX
~ (x2 + xf
At X = 5, the value of the derivative is —0.013.

Again using the quotient rule,
dy log^x-x(l/x)
dx
(loge^)
1 1
logeW
4 (i) ^-1
dx X
so that
dx2
SO that
dV^_
dx^
= 1 + ln X
dx
so
^-1
dx2 X
(iii) Using the composite function rule,
f'(x) = 2(loggX)^ 1
so
f'M =
Exercises 9.6 = -2.5

dq
so that, using the inverse function rule,
^ = -!-=-04
dp -2.5
Elasticity of demand, Ej, in this case is
pdq 0.4p
E,= ,=-^1-0.4) =
qdp q q
(i) With q = 50, p = 375 and therefore
0.4(375)
Ed = = 3
50

(ii) With q = 100, p = 250 and therefore
. 0.4(250) _ ,
(iii) With q = 150, p = 125 and therefore
0.4(125)
= 0.33
150
2 Here
-4
dq
so that
^= -0.25
dp
and
E« = "|-025l = ^
0.25(1000-4q)
q
250 - q
q
(i) when
q
250 — q = 3q
250 = 4q
q = 65
(ii) when
q
250-q= 1.5q
250 = 2.5q
so q = 100 and therefore p = 600.

(iii) This occurs when = 1, i.e. where
250-q ^
q
so q = 125 and p = 500.

3 We have: p = 100q“^ ^, so
In p = In 100 — 2.5 In q
and
c/lnp
= -2.5
c/ln q
the negated reciprocal of which is elasticity of demand, so
4 Rearrangement gives
1000 - 1 Op
P
from which, using the quotient rule,
dq _ 1000
dp p2
so that elasticity of demand is given by
^ p/ 1000 \ 1000
pq
Therefore
(i) when q = 2.5, p = 80 and
^
1000
2.5(80)
.
(ii) when q = 10, p = 50 and
_ 1000
^ 10(80)
(iii) when q = 40, p = 20 and
^
c 1000
I _ 19^
^ ~ 40(20) ~
Exercises 9.7 1 (i) lOOloOeX+fc (ii) nlog^x + Zc (iii) - log^x + it:

2 (i) In (x"* - 1 Ox) + k (ii) 0.5 In (ox^ + bx + c) + k
3 (-) 6x(lnx - 1) (ii) nx(lnx - 1)
4 (i) lOlog^ (log^x) (ii) 0.5 In (In x)
0.5 In (4x3 _ 5 ^2) ^ ^

so that the definite integral is
[0.5 In (4x^ - 5x^1]^=^ - [0.5 In (4x2 _ 5x2)]^ = 2

= (0.5 In 375)-(0.5 In 12)
« 2.9635 - 1.2425
« 1.7210
(ii) The indefinite integral is
1 Ox(ln X — 1) + /c
[10x(lnx- l)]"*"^® - [10x(lnx-

= 100(ln 10- 1) -40(ln4- 1)
« 130.2585-15.4518
« 114.8067
Exercises 9.8 1 (i) With
then
f(x) = e"
With
g(x) = 5x
then
r
dx
/ 5xe^dx = 5xe^
-/
5xe^ - 5e^ +
5e^dx
k
5(x- l)e^ + it
(ii) With
then
fix} = e"
With
g(x) = 2x2
520 Mathematics far business, finance and economics

then
J 2xê^dx = 2xê'‘ - j 4xe''dx
= 2x2e^-4(x- ])e’^ + k
= 2(x^ - 2x + 2)e^ + k
2 (i) With
—-e^'
dx“
then
f(x) = 0.5e^^
With
g(x) = X
then
= 0.5xe2^ - 0.25e2^ + k
= 0.25(2x- l)e2^ + i^
(ii) With
dx
then
f{x) = 2e°^^
With
g(x) = 5x^
then
Applying the by-parts approach to the second integral gives
40xe° - 80e° +k

so the original integral is
10x^6° - 40xe° ^^ + 80e° = 1 Ofx^ - 4x + 8)e° + k
3 Setting
^ _ .0.5X-0.5
dx'
so
f(x) = 2e05^-°'^
and with
g(x) = 200x
and hence
^ = 200
dx
Thus the indefinite integral is
J 200xe° ^^'°^dx = 400x6° -^^~ - J -4006° -^^'°^dx
= 400x6°-^^-°^ - 800e° ^^-°^ + k

so that the definite integral is
(400x6°^^-°^ - 800e°^^-°^)^=^ - {400xe°^^-°^
= 20006^ - SOOe^ - (1200e - 800e)

= 12006^ - 400e
7779.55
Exercises 9.9 1 (i) ^ = J_ =_!_

" dx xln5 1.6094X
rt ^ ^ ^
dx X In 0.2 -1.6094X
1 1
2 (i) f'{x)
xln60 4.0943x
1
f"{x)
“x2|n 10
(iii) f'N 2x
5
3 (i) f'M
xlogg8 X
0.5 0.72
(li| f'(x)
xloggO.5 X

1 (lOx-4)
4 (i)
dx i^TTO (5x2 _ 4^)
1 (10x2-3x+10)
In 2 (2.5x3 - x2 + 5x)
1
In 2
J Inxdx 1.4427x(lnx — ]] + k
(ii) The indefinite integral is
^ ^ In, 5
6 The indefinite integral is
2 .^^Sxflnx - 1) + /c
/ fii 2.6058x(ln x — 1) + /c
[2.6058x(lnx- l)f=^^ -2.6058x(lnx- l)f=^

« 2.6058[12(ln 12- 1) -4(ln4- 1)]
2.6058(17.8189- 1.5452)
Pi 42.41
7 (i) f'(x) = 10^ In 10?^ (2.3026)10^

(ii) f"(x) =5^(ln 5)2 (2.5903)5^
(iii) f'(x) = (2x^ — x^)(ln 6)6^ + 6^(1 Ox^ — 3x^)
8 (i) ^=(lnl0)(2x)10^'
dx
« (4.6x)10^'
(ii) ^ (In 16)(1 Ox - 0.5) 10^^' -° ^

dx
(iii) ^ = x3[(ln5)(2x)5^'] + (3x2)5^'

dx
Note: the product rule is used here.
’ I'l
+k
1 q5x^ - 8x
(iii) +k
In 10

10 The indefinite integral is
j (4x-11122''’-'‘dx = I^^+(r
SO that the definite integral is
[(122"'-1/(ln 12)r = 2 - [(122"'-")/(ln 12)]"=’

1 201 648.36 - 4.83 1 201 643.53

References and further reading
1 Anderson, D. R., Sweeney, D. J. and Williams, T. A. (1997) An Introduction to

Management Science (8th edn). West Publishing Co.
2 Begg, D., Fischer, S. and Dornbush, R. (1987) Economics (2nd edn),
McGraw-Hill.
3 Bishir, J. W. and Drewes, D. W. (1970) Mathematics in the Behavioural and
Social Sciences, Harcourt, Brace & World.
4 Borowski, E. }. and Borwein, J. M. (1989) Dictionary of Mathematics, Collins.
5 Boyer, C. B. (1968) A History of Mathematics, Wiley.
6 Chiang, A. C. (1984) Fundamental Methods of Mathematical Economics (3rd
edn), McGraw-Hill.
7 Churchhouse, R. F. (1978) Numerical Analysis: A First Year Course,
University College Cardiff Press.
8 Clapham C. (1989) The Concise Oxford Dictionary of Mathematics, Oxford
University Press.
9 Courant, R. (1937) Differential and Integral Calculus (2nd edn), vol. 1,
Blackie.
10 Finney, R. L. and Thomas, G. B., Jr. (1990) Calculus, Addison-Wesley.
11 Hart, W. L. (1966) College Algebra (5th edn). Heath & Co.
12 Hillier, F. and Lieberman, G. J. (1995) Introduction to Operations Research (6th
edn) McGraw-Hill.
13 Hollingdale, S. (1989) Makers of Mathematics, Penguin.
14 Intriligator, M. D. (1971) Mathematical Optimisation and Economic Theory,
Prentice Hall.
15 Jeffrey, A. (1989) Mathematics for Engineers and Scientists (4th edn). Van
Nostrand Reinhold.
16 Mizrahi, A. and Sullivan, M. (1988) Mathematics for Business and Social
Sciences (4th edn), Wiley.
17 Samuels, J. M., Wilkes, F. M. and Brayshaw, R. B. (1995) Management of
Company Finance (6th edn). Chapman & Hall.
18 Walsh, G. R. (1975) Methods of Optimization, Wiley.
19 Weber, J. E. (1982) Mathematical Analysis: Business and Economic Applications
(4th edn). Harper & Row.
20 Wilkes, F. M. (1983) Capital Budgeting Techniques (2nd edn), Wiley.
21 Wilkes, F. M. (1995) Mathematics for Business, Finance and Economics,
Routledge.
22 Wilkes, F. M. and Brayshaw, R. E. (1986) Company Finance and Its
Management, Van Nostrand Reinhold.
References and further reading 525

Specific sources and further reading for each chapter are:
Chapter 1: 1, 4. 6. 16, 19, 22
Chapter 2: 1, 12, 16, 20
Chapter 3: 4, 6, 10, 16, 19
Chapter 4: 2, 4, 5, 9, 11, 13, 17, 19, 20
Chapter 5: 4, 6, 16, 19
Chapter 6: 6, 10, 15, 19, 20
Chapter 7: 4, 6, 10, 12, 14, 15, 18, 19, 20, 21
Chapter 8: 6, 7, 10, 16, 19
Chapter 9: 3, 6, 10, 19
Appendix: 4, :5, 6, 8, 11, 13, 17, 20

Index
abscissa 2 chain rule 192-6, 376

absolute values 173-4, 411-12 exercises 196
addends 414 in partial differentiation 233-4
adjacent extreme points 66 circle 147
adjoint 99 closed form expression 426
adjugate matrix 99, 100 closed interval 59
aggregate sales curves 396-400 closed-open interval 60
aggregation 416 Cobb-Douglas production function
algorithm 68 369
annuity 428 cofactors 115-17
antiderivatives 313 column vector 79
exercises 314 common difference 424
anti-symmetric matrices 83 common logarithms 367
antitonic function 172 common ratio 42 7
Arabic numerals 401 commutative matrices 93, 97
arguments of function 127, 221, commutative property 86, 95, 97
414-15 complementary slackness conditions
arithmetic progression 424 297
associative property 86, 95 complement of set 405
asymptotes 138, 158 complex conjugates 402
average fixed costs (AFC) 162 complex numbers 138, 402
axis of symmetry 132 composite function rule 192, 284
composite functions 163
barter economies 41 compound interest 351
basis 112 computer software 125, 222, 414
binary-coded decimals 413 concave function 213-14, 240
binary numbers 413 concurrent lines 9
binary operation 417 conditional equation 45
biquadratic equations 154 consistent equations 25
bits 413 Consols 428
boundary conditions 323 constant function rule 182
braces 416 constant multiple rule 316
brackets 416 constrained optimization 250-310
breakeven analysis constraint 67
exercises 51 constraint qualification conditions
and simultaneous equations 47-9 299
breakeven output level 141 consumer theory 418 '
by-parts formula 386-9 continuous compounding 354-5
bytes 413 continuous function 167
contours 69-73
Cartesian co-ordinates 2 control variables 127
convex function 65-6, 214-15 diagonal matrix 80-1, 96
extreme point 66 differential equations 321, 330,
convex programming problems 300 364-5
co-ordinates 2-7 exercises 328-9, 366
cost theory 218 ordinary 321
counting numbers 401 partial 321
counting problems 406 differentials 330
Cramer’s rule 122-4 direct partial derivative 236
critical line 255 discontinuities 131, 167
cross-partial derivative 236-8 discontinuous function 167-8
cubic cost function 218 discount rates 428
cubing function 150 discrete function 169-70
cubics 149-53 discriminant
exercises 155 for quadratic equation 13 5-7
cusp 274 in second order conditions 240
customer service times 363-5 distance formula, exercises 6-7
disjoint sets 405
decay curves 352-3 distance formula 4
decay models 371-2 divergent series 428
decision variables 67, 127, 175, 224, domain of function 127
250-1, 260 dominant term 139
declining balance depreciation 427 dot product 89-90
defining relation of set 403 double roots 135
definite integral 312, 333-43 double subscript 81-2
exercises 343-4, 365 doubling time 370
degrees of freedom 35 dual values 7 5
demand 220, 308
elasticity of 419 economization 271
market potential 397 elasticity 160, 377-8
saturation level 397 of demand 378-81, 419
demand curves exercises 382-3
elasticity of 378-81 point 378
exercises 382-3 elementary matrix operation 83
hyperbolic 160-2 elementary row operations 32-3
and market saturation 161-2 elements of matrix 79
as unit elastic 160 elimination 15, 25, 27-9
demand and supply 165-6 Gaussian 42-5, 125
curves 49-50 ellipse 147
denary system 412-13 endogenous variables 41, 113
denumerable sequence 423 equal matrices 82
dependent variable 7. 127 equal sets 403
depreciation 3 52-3 equality constraints 270-1
declining-balance 427-8 equation of a straight line 7-10
straight-line 372, 425 finding through a given point 14-15
derivatives 177-80 finding through two known points
exercises 182-3, 188-9, 192 15-16
higher-order 200-2 identifying 11-12
logarithmic 374-7 obtaining 12
and natural exponential function plotting 12-13
357-60 equations
partial 229-35 conditional 45
Descartes rule of signs 465 identical 45
determinants 98, 115 linear 34-42, 122-4
properties 117 linearly dependent 107
determinate system 36-8 parametric 41
528 Index
reduced form 41 dominant term 139
vector 41 exercises 131-3, 174
see also differential: quadratic; exponential 350-65
simultaneous; straight-line homogenous 170-1
equilibrium price 49 inverse 164-6
equipment utilization rule 69 logarithmic 366-77
equivalent sets 403 monotonic 130, 171-2
escapable cost 187-8 odd and even 172
even function 172 periodic 172-3
excess demand 49 piecewise defined 130-1
excess supply 49 quadratic 132-42
excise duty 50 exercises 148-9
exogenous variables 41, 113 one linear and one quadratic
expansion paths 255 142
explicit form of equation 127 two quadratic equations 143-8
explicit function 198 quartics 153-5
exponential function 350-5, 356-7 exercises 155
and derivatives 357-60 rational 157-62
exercises 355-6, 361 exercises 162-3
integration of 361-5 several variables 220
to bases other than e 390-5 fundamental directions 222
exponential notation 414 future value
exponential probability distribution 363
exponents 406 game theory 224
irrational 411 Gauss-Jordan elimination 44, 125
power rule 409, 411 Gaussian elimination 42-5
product rule 406-8, 411 exercises 45
quotient rule 409-11 general antiderivative 313
extrapolation 18 general form of straight line equation
extreme points 66, 176 22
general form of quadratic equation
factoring 134 147
family of lines 8 general solution to differential
feasible production set 68 equations 323
feasible region 68 general term of sequence 424
financial management, and geometric progression 427-9
simultaneous equations 29-31 global maxima/minima 176-7,
finite series 425 207-13, 262
finite set 403 exercises 213
fixed-point notation 413-14 Gompertz function 398-400, 411
floating point notation 414 gradient form 21
flow variables 419 greatest lower bound 58
forecasting, straight line application growth curves 351, 353
18
free maximum 252 half-life 372
function of a function rule 192 half space 61
function notation 180 hexadecimal numbers 413
functions 126-7 higher-order derivatives 200-2
absolute value 173-4 exercises 202
composite 163-4 higher-order partial derivatives 236-8
continuous/discontinuous 167-9 higher-order systems 115-17
cubic 149-53 homogenous functions 170-1
exercises 155 horizontal lines 8
discrete 169-70 hyperbolas 147, 158
domain 127-30 hyperbolic demand curves 160-2
idempotent matrices 92 inverse function 164-5
identities 34, 45-7 exercises 198
identity matrix 81, 96 rule 196-8, 366, 373
imaginary numbers 138, 402 inverse matrix 97, 116-17
implicit differentiation 198-200 involutional matrix 102-3
exercises 200 irrational numbers 401
implicit form of equation 127 iso-profit lines 69, 227
implicit function rule 254 isoquands 227
improper integral 335-7 isotonic function 171-2
income 140 iterations 68
inconsistent equations 29
independent equations 25, 38 joint analysis 109
independent variable 127 Jordan elimination 44
indeterminate systems 38
index of summation 420 Karush-Kuhn-Tucker conditions 300
inequality constraints 289-95, 296-8 Kuhn-Tucker conditions 299-307
exercises 299 exercises 307-8
inequality relationships
linear 53-76 Lagrange multipliers 264-7, 289-95,
quadratic 145 296-8
infimum 58-9 exercises 272, 295-6
infinite product 423 interpretation of 270-1
infix notation 415 minimization problems 267-70
inflection points 151, 215 Laplace expansion 115
initial conditions 323 law of scarcity 5 3
inner product 89 leading coefficient 132
input variables 221 leading diagonal 80
instruments 127 lead time 19
integer programming 35 least upper bound 58
integers 401 left inverse 103
integrand 314 Leibniz’s notation 179
integration 312-14 limits of summation 420
applied to marginal analysis 317-20 linear dependence 99, 107, 111
by parts 386-9 linear differential equation 322
by substitution 329-32 with constant coefficients 322-4
definite 333-43 linear equations
exercises 317, 332-3, 348 exercises 35-6, 42
limits of 333 in more than two variables 36-41
of natural exponential functions one equation in one unknown 34
361-5 one equation in two unknowns 35
of logarithmic functions 383-5 whole-number requirements 35
numerical 344-8 linear function of several variables
rules 222-5
constant function 314 exercises 225
constant multiple 316 linear inequalities
power 3 1 5 exercises 55
sum-difference 316 in one variable 55
interest rates 353-4, 371 addition of a constant 55
interpolation 18 multiplication by negative
intersection of sets 405 constant and reversal of
interval scale 418 direction 56-60
inventory control multiplication by a positive
exercises 20 constant 55-6
reorder level/point 19 simultaneous 58-60
straight-line application 18-20 strict 53-4
530 Index
in two or more variables 60-5 multiplication 89-9
weak 54 associative property 95
linear programming 47, 67-76, 224, conformability in 91-2
252 identity element for 9 6
exercises 76 postmultiplication 93
simplex method 66, 68 premultiplication 93
linear regression 18 negation 87
local maxima/minima 176, 202-6 notation 81-2
exercises 242-3 principle diagonal 80
first-order conditions 239-42 rank 111-14
logarithmic derivative 377 scalar multiplication 87-8
logarithmic functions 366-72 and simultaneous equations
exercises 373, 377, 385-6. 395-6 105-10, 118-21
and integration 383-5 submatrices 79-80
to bases other than e 390-5 subtraction 86-7
logical operators 417 transposition 82-3
logistic curve 399-400 triangular 80
log-linear functions 369 zero 81
lower bounds 58 maxima and minima
lower triangular matrix 80 global 176-7, 207-13
local 176, 202-6
main diagonal 80 membership of set 402
marginal cost 186-8, 318, 337 method of substitution 256
exercises 320 minors 115
and integration 317-20 mixed derivative 236-8
marginal profit 186, 231-2 mixed expression 159-60
marginal revenue 184-8, 319 model-building 55
equilibrium condition 185-6 modified growth curve 396-7
finite approximation 184-5 monetary variables 417-18
marginal values 75, 257, 259 monotonic functions 130, 171
market clearing price 49 multi-constraint problems 262
market equilibrium multinomial equation 156
exercises 51-2 multiple roots 135
and linear demand/supply curves multi-product monopoly 308-10
49-50 exercises 311
market potential 397
market saturation 161, 397 natural exponential functions see
matrices 77-8, 124-5 exponential functions
addition 85-6 natural logarithmic functions see
associative property 86 logarithmic functions
commutative property 86 natural numbers 355, 401
conformability in 85-6 negative exponential distribution 363
identity element for 8 7 nested brackets 416
applications 115-17 nominal scale 419
Cramer’s rule 122-4 non-linear differential equation 327-8
determinants 115-17 non-linear programming 299
diagonal 80-1 non-linear systems 141-2
difference 87 non-negativity requirements see sign
dimensions 78-9 requirements
elements 79 non-pool resource 420
equal 82 non-unique optima 73, 177
exercises 83-5, 88-9, 96-7, 103-5, normal lines 9
110-11, 114, 118, 122, 124 null set 404
inversion 97-9 numbers
labelling 78 complex 402
counting 401 piecewise defined functions 73, 130-1,
imaginary 402 210, 339
irrational 401 place-value notation 412
natural 401 point elasticity 378
rational 401 point of inflection 215-18
real 401-2 exercises 218-19
numeraire 41, 261 stationary/non-stationary 215
numerals 401 point-slope form 20-1
numerical methods for integration polish notation 415
344-5 pol3momials 156-7
Simpson’s rule 347-8 differentiating 183-8
trapezium rule 345-7 exercises 157
pool resource 420
objective function 67. 127 portfolio theory 255, 418
octal numbers 413 positive quadrant 5
odd function 172 postfix notation 415
off-diagonal elements 80 postmultiplication of matrices 93
one to one function 166 power function rule 180-1
open interval 59 power rule for exponents 92, 409
operators 414-15, 417 power series 425
binary 417 power set 404
logical 417 prefix notation 415
relational 417 premultiplication of matrices 93
opportunity loss 71 present value of annuity 428
opportunity set 68 price breaks 168
optimization under constraint principal 353
250-310 principal diagonal 80
order of matrix 78 production functions 221, 369
order-preserving transformations 418 production plan 67
ordinal scale 418 production possibility frontier 68, 73
ordinary differential equation 321 production theory 218, 221
ordinate 2 and integration 337
orthants 5-6 productivity curve 227
orthogonal lines 9 product notation 423
orthogonal matrices 102 product rule for differentiation 189-90
output variables 221 product rule for exponents 92, 406-8
overdeterminate system 38-40 progressions 424
proper subset 403
parabolas 132 Pythagoras formula 3, 6
parameters 8
parametric equations 41 quadrants 5
parentheses 415 quadratic equation
Pareto optimality 68 exercises 148-9
partial derivatives 229-35 general 147-8
cross 236 and one linear equation 142
direct 236 quadratic inequality 145-6
exercises 235, 238-9 two equations 143-5
higher-order 236-8 quadratic formula, 135
mixed 236 quadratic functions 132-5, 226-8
partial sum 422 discriminant 136-7
particular solution 323 exercises 228
partitioned matrix 79-80 parabola 132-5
periodic function 172-3 points on the graph 138-40
perpendicular lines 9-11 quadratic inequality 145
perpetuities 428 quadratic programming 252, 259, 306
532 Index
quartics 153-5 several variables 236
queuing systems 363-5 test 203-6
quotient rule for differentiation 190-1 sensitivity analysis 58-9, 76, 109
in partial differentiation 234 sequences and series 423-4
quotient rule for exponents 409-11 set
complement 404
radix 412 difference 404-5
range of feasibility 59-60, 108, 121 disjoint 404
range of function 127 elements or members 402-3
rank 111-14 intersection 404
rate of substitution 71, 253 null 403
rational functions 1 5 7-62 power 403
exercises 162-3 subset 403
rational numbers 401 superset 403
ratio scale 417 union 403-4
real line 2 universal 403
real numbers 401 Venn diagrams 405
reciprocal function 158 sexagesimal system 413
reciprocal rule 191 shadow prices 75
rectangular array 77 sigma notation 420-3
rectangular hyperbolas 158-9 index 420
reduced form equations 41 lower limit 420
re-entrant region 65 terms 420
reflex polygon 65 upper limit 420
regression, linear 18 sign-restricted variables 61, 250, 258,
relational operators 54, 417 272-3, 296
re-order level 19 simple interest 424
repeated root 135 simple root 135
resource constraints 289 simplex 224-5
reverse polish notation 415 method 68
right inverse 103 Simpson’s rule 347-8
roots of equation 34-6, 134, 465 simultaneous inequalties 58
roots of numbers 410 simultaneous linear equations 23-5,
row vector 79 58-60
rules of precedence 415 and breakeven analysis 47-9
elimination 27-9
saddle points 243-6, 301 exercises 31-2
sales 140 and matrices 105-10, 118-21
saturation curves 399 substitution 25-7
saturation level 397 simultaneous quadratic equations 143
scalar matrix 81 singular matrix 94, 99
scalar multiplication 87-8 skew-symmetric matrix 83
scalar product 90, 101 slack variables 46, 71, 291, 297
scalars 79 slope 4-6, 229-31
scales of measurement 417 slope-intercept form 7-10
interval 418-19 solution sets 41, 56-60
nominal 419 convex 65-6
ordinal 418 exercises 60
ratio 417-18 feasible 68
units 419-20 solutions to differential equations
scarce resources 67 general 323
scientific notation 414 particular 323
S-curve demand function 397-8 specific sales tax 50
second derivatives 200 square of matrix 92
exercises 206 square matrix 78
stationary points 176, 239-40, 247-9 Taylor series 425-6
stationary point of inflection 215 technical independence 308
step discontinuity 131 third derivative 200
marginal revenue function 167-8 tolerance interval 59-60, 108, 121
price breaks 168 total product curve 227
welfare traps 167 total revenue 140
step function 168-9 total variable cost 337
stock control see inventory control trace 80
stockouts 19 trade-off 71
stock variables 419, 420 transitivity 403
straight-line depreciation 42 5 transposition 82-3, 83
straight-line equation trapezium rule 345-7
exercises 17-18 triangular matrix 80
finding through a given point 14-15 turning points 138, 176
finding through two known points exercises 207
15-16 turnover 140-1
formula 7-10
identifying 11-12 undefined slope 8
obtaining 12 unique solution 36
plotting 12-13 union of sets 404
straight lines unit costs of production 89-90
applications 18-20 universal set 404
distance formula 3-4 upper bounds 58
establishing on which side a given upper triangular matrices 80
point lies 16 utility functions 418
exercises 11
and simultaneous equations 23-32 variables
slope 4-6 decision/control 127
strict inequality 53 dependent/independent 127
strip 63 endogenous/exogenous 41
subject of equation 7, 127 explicit/implicit 127
submatrices 79-80 in matrices 79
sub-problems 66 slack 71, 291, 297
subsets 403 vector equations 41
substitution method vectors 79
constrained optimization 256-63 Venn diagrams 405
integration 329-32 vertex of parabola 132, 140
linear equations 25-7 vertical lines 8
sum(-difference) rule 183-4 vinculum 416
exercises 188
integration 316 Wallis's product 423
sigma notation 421 War loan stock 428
summation notation see sigma notation weak inequality 54
sum to infinity 422 whole-number requirements 35
supersets 404
supply see demand and supply X intercept 138
supremum 58-9 X values 138
symmetric difference 406
symmetric matrix 82 1/ intercept 138
tangency method in constrained zeros of function 34, 134

optimization 252-5 zero matrix 81
exercises 255-6
534 Index
Mathematics for Business,
Finance and Economics Second edition
Michael Wilkes
Mathematics Michael Wilkes is Professor of Business Investment

for Business, and Management at the University of Birmingham.
Finance aitcl ^ The author of Capital Budgeting Techniques and
co-author of Management of Company Finance, he is
EfTonomics^^
an experienced teacher and researcher in the field of
investment decision making.
Mathematics for Business, Finance and Economics provides a thorough and

accessible introduction to the mathematical methods and concepts used in
quantitative analysis for management, finance, business studies and economics.
The second edition has been completely revised and edited to suit the needs of
students on contemporary courses.
Topics covered include linear equations and inequalities, functions, derivatives

and optimization of functions, and optimization under constraint. Rigorous in
the concepts, yet straightforward in its style. Mathematics for Business, Finance
and Economics concentrates on building student confidence in the tools before
they go on to the applications, making it the ideal text for introductory
Features of this edition include:

• coherent and consistent organization of the chapters offers 6
graduated approach to lead students through from a basic le\
more advanced applications.
• the adoption of a user-friendly approach, combining a detailed
description of the principles and methods with numerous exan
and end-of-section exercises
• comprehensive worked answers at the end of the book
Cover design/illustration:Terry Foley
-—fpt-
ISB -86152-241-X
THOMSON
BUSINESS PRESS

Mathematics For Business, Finance, and Economics

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mathematics For Business, Finance, and Economics

Uploaded by

Copyright:

Available Formats

2nd edition

MkGHAEL WILKES INTER

INTERNATIONAL THOMSON BUSINESS PRESS

Copyright © 1999 F. M. Wilkes

First published 1994 by Routledge

A division of International Thomson Publishing Inc.

British Library Cataloguing-in-Publication Data

Second edition 1999

Typeset by LaserScript Limited, Mitcham, Surrey

International Thomson Business Press

Chapter 1 Straight lines and linear equations 1

Chapter 4 Functions and turning points 126

Chapter 5 Derivatives and optimization 175

Chapter 6 Functions of more than one variable 220

Chapter 7 Constrained optimization 250

Chapters Integration 312

Chapter 9 Exponential and logarithmic functions 350

Solutions to exercises 430

References and further reading 525

•Hi rfrf- T n’ uS^

Mathematics for Business, Finance and Economics adopts a progressive

1 To provide a thorough treatment of some of the most important

The principles and methods are described as far as possible in narrative

1 Straight lines and linear

This chapter introduces important concepts concerning straight lines

Straight lines and linear equations 1

Pi (2,1) P2(4,3) P3(-2,1) P4:(2,-3)

2 Mathematics for business, finance and economics

1 What is the straight-line distance between two points.?

The straight-line distance between two points with given co-ordinates

Straight lines and linear equations 3

In general, the distance formula for the straight-line distance between

d = yix^-xif+iyi - yif (1-1)

while the slope of the P3 and P4 line is:

4 Mathematics for business, finance and economics

The slope of a line depends on the units of measurement on the axes.

Straight lines and linear equations 5

concepts and particular functions. Points in three space have co¬

d = \l ix2 - xi)^ + (1/2 -yif + (Z2 - Zif

For example, the straight-line distance between the points

Pi ;(8,9,10) and P2 : (6,5,4)

d = -^(8 - 6)2 + (9 - 5)^ + (10 -

The concept of slope in three dimensions is more complicated in that it

6 Mathematics for business, finance and economics

The slope-intercept form is usually the preferred algebraic expression

Line L might represent the equation

Straight lines and linear equations 7

Horizontal lines (parallel to the x axis) have zero slope (m = 0) and

So the straight line defined by the equation

is a horizontal line ten units above the x axis, while

In the general case of a straight line in slope intercept form:

there are two parameters, m and c, each of which can be

8 Mathematics for business, finance and economics

If y represents costs and x represents output level, then a family of lines

are parallel. A family of lines can also be defined by the common

Straight lines and linear equations 9

where c and k are the y intercepts of the lines. So for example:

are perpendicular. These lines are shown in Figure 1.8.

is perpendicular to the line y = 2x — 4. So the lines

are also orthogonal to y = 2x — 4 (and to any line having a slope of 2).

is orthogonal to the line y = 3 — 0.5x. Another way that orthogonality

10 Mathematics for business, finance and economics

Straight lines and linear equations 11