Mathematicalanal033535mbp PDF

OU162017>m
LONDON SCHOOL OF ECONOMICS AND POLITICAL SCIENCE

(UNIVERSITY OF LONDON)
STUDIES IN STATISTICS AND SCIENTIFIC METHOD

EDITED BY
A. L. BOWLEY AND A. WOLF
No. 3. MATHEMATICAL ANALYSIS FOR ECONOMISTS

MATHEMATICAL ANALYSIS
FOR ECONOMISTS
BY
R. G. D. ALLEN
The general science of mathematics is concerned with

the investigation of patterns of connectedness, in abstrac-
tion from the particular relata and the particular modes
of connection.
ALFRED NORTH WHITEHEAD, Adventures of Ideas
To connect elements in laws according to some logical

or mathematical pattern is the ultimate ideal of science.
MORRIS R. COHEN, Reason and Nature
MAQMILLAN AND CO., LIMITED

ST. MARTIN'S STREET, LONDON
1938
PRINTED IN GREAT I3KITAIN
FOREWORD
TEUS book, which is based on a series of lectures given at the London
School of Economics annually since 1931, aims at providing a course
of pure mathematics developed in the directions most useful to
students of economics. At each stage the mathematical methods
described are used in the elucidation of problems of economic theory.
Illustrative examples are added to all chapters and it is hoped that the
reader, in solving them, will become familiar with the mathematical

tools and with their applications to concrete economic problems.
The method of treatment rules out any attempt at a systematic
development of mathematical economic theory but the essentials of
such a theory are to be found either in the text or in the examples.
I hope that the book will be useful to readers of different types.
The earlier chapters are intended primarily for the student with no
mathematical equipment other than that obtained, possibly many
years ago, from a matriculation course. Such a student may need to
accustom himself to the application of the elementary methods
before proceeding to the more powerful processes described in the
later chapters. The more advanced reader may use the early
sections 'for purposes of revision and pass on quickly to the later
work. The experienced mathematical economist may find the book
as a whole of service for reference and discover new points in some
of the chapters.
I have received helpful advice and criticism from many mathe-
maticians and economists. I am particularly indebted to Professor
A. L. Bowley and to Dr. J. Marschak and the book includes numerous
modifications made as a result of their suggestions on reading the
original manuscript. I am also indebted to Mr. G. J. Nash who has

read the proofs and has detected a number of slips in myconstruction
of the examples.
R. G. D. ALLEN
THE LONDON SCHOOL OF ECONOMICS
October, 1937
CONTENTS
CHAP.
FOREWORD
A
----------
SHORT BIBLIOGRAPHY ..... -
PAGE
v
xiv
THE USE OF
ANALYSIS ......
GREEK LETTERS IN MATHEMATICAL
xvi
-------
- -
I. NUMBERS AND VARIABLES

1.1 Introduction ---------
------
1
1.2
1.3 -------
Numbers of various types
The real number system
...
3
6
1.4
1.5
Continuous and discontinuous variables
Quantities and their measurement
1.0 Units of measurement - -
.....
- - - -
-
-
7
9
13
1.7 Derived quantities - - - - - - - - 14
1.8 The location of points in space - - - - - 1G
1.9 Va viable points and their co-ordinates 20
methods ---------23
EXAMPLES 1 The measurement of quantities ; graphical
.
JpOJ ACTIONS
AND THEIR DIAGRAMMATIC REPRESENTATION 28
2.1 Definition and examples of functions 28
2.2 The graphs of functions - - - - - - - 32
2.3 Functions and curves - - - - - - - 3(5
2.4 Classification of functions - - - - - - 38

2.5 Function types - - - - - - - - 41
2.6 The symbolic representation of functions of any form - 45
2.7 The diagrammatic method - - - - - - 48
2.8 The solution of equations in one variable 50
2.9 Simultaneous equations in two variables 54
EXAMPLES
/
II
tions .........
Functions and graphs ;
the solutionjof equa-
57
III.
3.1 Introduction
3.2
.........
ELEMENTARY ANALYTICAL GEOMETRY
.....
The gradient of a straight line
61
61
03
3.3 The equation of a straight line - - - 66
viii CONTENTS
CHAP.
3.4 The parabola 09
3.5 The rectangular hyperbola - - - - - - 72
3.6
3.7
The circle
Curve classes and curve systems . - ... 75
76
3.8 An
economic problem in analytical geometry 80
EXAMPLES III-- The straight line curves and curve systems
;
82
IV. LIMITS AND CONTINUITY OF FUNCTIONS 85

4.1
4.2
The fundamental notion of a limit
Examples of the limit of a function
4.3 Definition of the limit of a
----- -
single- valued function

- - -
-
-
-
85
88
91
4.4 Limiting and approximate values 95
4.5 Home properties of limits
4.6 The continuity of functions
-
------ - - - - - 97
98
4.7
4.8 M u ti- valued
EXAMPLES IV
I functions ..----.
Illustrations of continuity and discontinuity of functions
Limits of functions ; continuity of functions

-
-
100
102
103
V. FUNCTIONS AND DIAGRAMS IN ECONOMIC THEORY - 107

5.1 Introduction - - - - - - - - -107
5.2 Demand and curves
functions - 108
5.3 Particular demand
functions and curves - - - - 112
5.4 Total re vemie functions and curves - - - - - 116
5.5 Cost functions and curves - - - - - -117
5.6 Other functions economic theory
arid curves in - - 121
5.7 Indifference curves for consumers' goods - - 124
5.8 Indifference curves for the flow of income over time - - 127
EXAMPLES V Economic functions and curves - - 129
VI. DERIVATIVES AND THEIR INTERPRETATION - - 134

6.1 Introduction 134
6.2 The definition of a derivative - - - - - 137
6.3 Examples of the evaluation of derivatives - - 140
6.4 Derivatives and approximate values - - - 142
6.5 Derivatives and tangents to curves - - - - - 143
6.6 Second and higher order derivatives - - - - 148
6.7 The application of derivatives in the natural sciences - 149
6.8 The application of derivatives in economic theory - - 152
EXAMPLES VI Evaluation and interpretation of derivatives - 1 57
VII. THE TECHNIQUE OF DERIVATION 160

7.1 Introduction 160
7.2 The power function and its derivative - - - - 161
7.3 Rules for the evaluation of derivatives - - - - 163
CONTENTS ir
CHAP. PAGE
7.4 Examples of the evaluation of derivatives - - 166
7.5 Tho function of a function rule 16S
7.6 The inverse function rule 171
7.7 Tho evaluation of .second and higher order derivatives - 172
EXAMPLES VII Practical derivation - - - - 175
VIII. APPLICATIONS OF DERIVATIVES - - - - 179

8.1 The sign, and magnitude of the derivative - - - 179
8.2 Maxirmim and minimum values - - - - 181
8.3 Applications of the second derivative - - - 184
8.4 Practical methods of finding maximum and minimum
values - - - - - - - -
-J8G
8.5 A general problem of average and marginal values - 190
8.6 Points of inflexion 191
8.7 Monopoly problems in economic theory - - -
196
8.8 Problems of duopoly 200
8.9 A note oil necessary and sufficient conditions
- - 204
EXAMPLES VIII General applications of derivatives ; eco-
nomic applications of derivatives - 205
IX. EXPONENTIAL AND LOGARITHMIC FUNCTIONS - 211

9.1 Exponential functions - - - - - -
211
9.2
9.3
Logarithms and their properties
Logarithmic functions ------- 213
217
9.4
9.5
Logarithmic scales and graphs
Examples of logarithmic plotting -----
-------
- - - - 219
223
9.6
*9.7
Compound interest
Present values and capital values ... -
228
232
9.8 Natural exponential and logarithmic functions - - 234
EXAMPLES IX Exponential and logarithmic functions ;
compound interest problems 238
X. LOGARITHMIC DERIVATION 242
10.2 Logarithmic derivation

10.3 A
problem of capital and interest
------
10.1 Derivatives of exponential and logarithmic functions
-
- 242
246
248
10.4 The elasticity of a function -
10.5The evaluation of elasticities
10.6 The elasticity of demand
-
----- - - - 251
252
254
10.7 Normal conditions of demand 257
10.8 Cost elasticity and normal cost conditions - - - 260
EXAMPLES X Exponential and logarithmic derivatives ;
elasticities and their applications - - - 264

x CONTENTS
CHAP. PAGE
Two on MORE VARIABLES
XI. FUNCTIONS OP
11.1 Functions of two variables ------ - - - 268
268
11.2 Diagrammatic
variables ---------
representation
a surface -
11.3 Piano sections of
of functions
- -
of
-
two
270
272
11.4 Functions of more than two variables - - - - 275
11.5 Non -measurable variables - - - - - -276
11.6 Systems of equations 278
11.7 Functions of several variables in economic theory - 281
11.8 The production function and constant product curves - 284
11.9 The utility function and indifference curves - -
289
EXAMPLES X F Functions of two or more variables 1
; eco-
nomic functions and surfaces - - - - 292
XII. PARTIAL DERIVATIVES AND THEIR APPLICATIONS - 296

12.1 Partial derivatives of functions of two variables - 296
12.2 Partial derivatives of the second and higher orders - 300
12.3
12.4
The signs of partial derivatives
The tangent plane to a surface
12.5 Partial derivatives of functions of
----- more than two
303
305
variables 309
12.6 Economic applications of partial derivatives - - 310
12.7 Homogeneous fmictions - - - - - -315
12.8 Eulcr's Theorem and other properties of homogeneous
functions 317
12.9 The linear homogeneous production function - - 320
EXAMPLES XII Partial derivatives ; homogeneous func-
tions ; economic applications of partial deriva-
tives and homogeneous functions - - - - 322
XIII. DIFFERENTIALS AND DIFFERENTIATION - 326

13.1 The variation of a function of two variables - - 326
13.2 The differential of a function of two variables - - 328
13.3 The technique of differentiation 330
13.4 Differentiation of functions of functions - - - 332
13.5 Differentiation of implicit functions - 334
13.6 The differential of a function of more than two
variables 339
13.7 Thesubstitution of factors in production
13.8 Siibstitution in other economic problems
13.9 Further consideration of duopoly problems
...
-
-
-
-
-
-
340
344
345
EXAMPLES XIII Differentiation ; economic applications
of differentials 347
CONTENTS xi
CHAP. PAGE
XIV. PROBLEMS OF MAXIMUM AND MINIMUM VALUES -
351
14.1 Partial stationary values 351
14.2 Maximtun and minimum values of a function of two
or more variables - - - - - - - 352
14.3
J4.4
14.5
Examples of maximum and minimum values
Monopoly and joint production
Production, capital and interest
----- - - 356
359
362
14.0 Relative maximum arid minimum values - - - 364
14.7 Examples of relative maximum and minimum values 367
14.8 The demand for factors of production - - -
309
14.9 The demand for consumers* goods and for loans - -
374
EXAMPLES XIV General maximum and minimum prob-
lems ; economic maximum arid minimum problems 378
XV. INTEGRALS OF FUNCTIONS OF ONE VARIABLE - - 384

15.1 The definition of a definite integral
15.2 Definite integrals as areas - - - -
- ... -
384
387
15.4 The technique of integration

15.5 Definite integrals and approximate integration
-----
15.3 Indefinite integrals and inverse differentiation
-
- -
-
390
393
396
15.6 The relation between average and marginal concepts - 400
15.7 Capital values 401
15.8 A problem of durable capital goods - 404
15.9 Average and dispersion of a frequency distribution -
406
EXAMPLES XV Integration ; integrals in economic prob-
lems 408
XVI. DIFFERENTIAL EQUATIONS - - - - - - 412

16.1 The nature of the problem - - - - - - 412
16.2 Linear differential equations and their integration - 417
16.3 The general integral of a linear differential equation - 422
16.4 Simultaneous linear differential equations
16.5 Orthogonal curve and surface systems ...
- - 425
439
430
16.6 Other differential equations
16.7Dynamic forms of demand and supply functions - 434
16.8The general theory of consumers' choice - - - 438
EXAMPLES XVI Differential equations economic appli- ;
cations of differential equations - - - 442
XVII. EXPANSIONS, TAYLOR'S SERIES AND HIGHER ORDER

DIFFERENTIALS
17.1 Limits and infinite series ..... .
446
446
17.2
series)
.---.----
The expansion of a function of one variable (Taylor's
449
xii CONTENTS
CHAP. PAGE
17.3 Examples of the expansion of functions - - 454
17.4 The expansion of a function of two or more? variables 456
17.5
17.6
values ---------
A complete criterion for maximum and minimum
Second and higher order differentials - - -
459
461
variables ........
17.7 Differentials of a function of two
17.8 Differentials of a function of two dependent variables

independent
463
4(55
EXAMPLES XVII Infinite series ; expansions ; higher

order differentials - - - - - -469
XVIII. DETERMINANTS, LINEAR EQUATIONS AND QUADRATIC

FORMS - 472
18.1 The general notion of a determinant - - 472
18.2 The definition, of determinants of various orders - 473
18.3 Properties of determinants 477
18.4 Minors and co -factors of determinants - - - 478
18.5 Linear arid homogeneous functions of several vari-
ables 481
1 8.6 The
solution of linear equations -
482
18.7 Quadratic forms in two and three variables
18.8
18.9
Examples of quadratic forms
Two general results for quadratic forms
----- -
-
-
-
-
485
489
491
EXAMPLES XVIII Determinants ; linear equations ;
quadratic forms - - - - - - -492
XIX. FURTHER PROBLEMS OF MAXIMUM AND MINIMUM

VALUES - 495
19.1 Maximum and minimum
several variables
19.2 Relative
-------
maximum and minimum values
values of a function of
- - -
495
498
19.3 Examples of maximum and minimum values - - 500
19.4 The stability of demand for factors of production - 502
19.5 Partial elasticities of substitution - - - 503
19.6 Variation of demand for factors of production - - 505
19.7 The demand for consumers' goods (integrability
ease) 509
19.8 Demands for three consumers' goods (general case) - 513
EXAMPLES XIX maximum and minimum prob-
General
lems ; economic maximum and minimum problems 517
CONTENTS xiii
CHAP. PAGE
XX. SOME PROBLEMS IN THE CALCULUS OF VARIATIONS - 521
20.1 The general theory of functional* - - - -521
20.2 The calculus of variations ,523
20.3 The method of the calculus of variations - - - 524

20.4 Solution of the simplest problem - 52(>
20.5 Special cases of Kulcr's equation - - - 529
20.6 Examples of solution by Euler's equation - - 530
20.7 A dynamic problem of monopoly - - - 533
20.8 Other problems in the calculus of variations - - 536
.EXAMPLES XX Problems in the calculus of variations - 540
INDEX :
......
------
MATHEMATICAL, METHODS 543
AUTHORS ----------
ECONOMIC APPLICATIONS 540
548
A SHORT BIBLIOGRAPHY
IT is essential to keep in mind, at all times, what mathematical analysis
isabout and the way in which it has developed over the centuries. Not
only beginners, but also accomplished mathematicians, can profit by
reading such short introductions as
Whitehead : An Introduction to Mathematics (Homo University
Library, 1911).
Brodetsky : The Meaning of Mathematics (Benn's Sixpenny Library,
1929).
Rice : The Nature of Mathematics (An Outline of Modern Know-
ledge, 1931) ;
and longer accounts such as

Dantzig : Number (1930).
Forsyth : Mathematics in Life and Thought (1929).
Hogben : Mathematics for the Million (1936).
The history of mathematics has been the subject of many books :
good short treatments are

Rouse Ball ^4 Short Account of the History of Mathematics (Gth Ed.
:
1915).
Sullivan : The History of Mathematics in Europe (1925).
The reader wishing to revise his knowledge of " matriculation algebra

"
and geometry and to obtain a " practical " introduction to the calculus
can consult
Durell :
Elementary Geometry (1919).
Durell, Palmer and Wright Elementary Algebra (1925) (including
:
an introduction to the calculus).
Thompson : Calculus made Easy (2nd Ed., 1919).

Irving Fisher : A Brief Introduction to the Infinitesimal Calculus
(3rd Ed., 1909).
General text- books on mathematical analysis vary a good deal in their
scope and in their treatment. The following selection includes works
ganging from the elementary to the advanced in content and from the
BIBLIOGRAPHY xv
practical to the rigidly theoretical in method. They are to be regarded
as complementary rather than competitive commodities.
Hardy : A Course of Pure Mathematics (3rd Ed., 1921).

Griffin : An Introduction to Mathematical Analysis (no date).
Osgood : Introduction to the Calculus (1922) ;
and Advanced Calculus
(1925).
f
de la Vallee Poussin : Cours d Analyse Infinitesimals (5th Ed.), Vol.
1 (1923), Vol. II (1925).
Conrant :
Differential and Integral Calculus (English Ed.), Vol. T
(1934), Vol. II (1936).
Pavate and Bhagwat : The Elements of Calculus (2nd Ed., 1932).
THE USE OF GREEK LETTERS IN MATHEMATICAL
ANALYSIS
THE English alphabet provides insufficient material for the notation ol
mathematical analysis and greater range and flexibility are acquired by
using letters from the Greek alphabet. There are 110 fixed rules governing
the mathematical use of Greek letters but the following table indicates a
practice which is fairly general. It should be noticed that the English
letters given are those which correspond to the Greek letters in mat he
matical, not literal, usage.
In trigonometry, a, j8, y can denote constant angles and 0, <f),

r
varia
angles.
CHAPTER I
NUMBERS AND VARIABLES

1.1 Introduction.
IT is conventional to divide mathematics into two separate studies,
geometry and analysis. In geometry, the study of space and of
spatial relationships, we investigate the nature and properties of
curves, surfaces and other configurations of points in space. Mathe-
matical analysis, including arithmetic and algebra, deals with
numbers, the relationships between numbers and the operations
performed on these relationships. The distinction between the two
"
studies lies in the difference between the basic raw materials ",
the spatial points of geometry and the numbers of analysis.
As the mathematical technique is developed, however, we find
that the distinction between geometry and analysis becomes less
clear and less relevant. The intended applications remain very
different but the methods are seen to be abstract and essentially
similar in nature. Mathematics involves the definition of symbols
of various kinds and describes the operations to be performed, in
definite and consistent ways, upon the symbols. The distinction
between geometrical and analytical symbolism is more or less arbi-
trary. Further, we shall see that it is a simple matter to devise a
method of connecting points of space and numbers, geometry and
analysis. That such a connection is possible is seen by the use of
graphical methods in algebra and by the fact that elementary
trigonometry is an application of algebraic methods to a study of
spatial configurations.
The mathematical technique is abstract and must be developed
apart from its applications. also essentially logical in character.
It is
Elementary geometry, for example, is largely an exercise in formal

logic,the deduction of the consequences of consistently framed
assumptions. The development of mathematical methods, however,
A M.A.
2 MATHEMATICAL ANALYSIS FOR ECONOMISTS
soon necessitates the introduction of a concept foreign to formal
"
logic, the concept of infinity ", the infinitely large and the in-
definitely continuous. It can be

speaking broadly, that the
said,
methods of mathematics form a branch of formal logic extended in
a particular direction to include the infinite as well as the finite.
The only requirements of mathematics, additional to those of logical
reasoning, are connected with the introduction of the infinite.
The popular belief that logical and mathematical methods are in
conflict has, therefore, no foundation. Mathematics, as a symbolised
and extended form of logic, can only be regarded as an alternative
to logic in a special sense. Some problems are adequately treated
by relatively simple logical reasoning and the introduction of

mathematical symbolism is then only destructive of clarity. But
formal logic, with its limited range, is clearly insufficient for the
elucidation of a vast number of problems and this provides the case
for the development and application of mathematics.
Though an abstract development, mathematics is not just a

fascinating game of chess played in 7i-dimensional space with pieces
of fantastic design and according to rules laid down in an arbitrary
way. The methods of mathematics are designed primarily for actual
or prospective application in the field of science, in the interpretation
of phenomena as observed, abstracted and classified. The nature
of the application must not, however, be exaggerated. The patterns
of scientific phenomena, the laws obeyed and the uniformities dis-
played, are not provided by mathematics. They are assumed into
the system and mathematics can only express and interpret them,
help us to deduce their consequences, or to forecast what will happen
if they hold, and tell us where to look for verification or contradiction
of our hypotheses.
The methods of mathematics apply as soon as spatial or numerical
attributes are associated with our phenomena, as soon as objects
can be located by points in space and events described by properties
capable of indication or measurement by numbers. The main object
of the following development is the description of certain mathe-
matical methods. At most points, however, the nature and field of
application of the various methods will be noticed and discussed,
the applicability of mathematical methods in economic theory
receiving special consideration.
NUMBERS AND VARIABLES
~\
In concluding these introductory remarks, we can note that \
fundamental bases of the mathematical technique have received

much attention in recent years. Jt is indeed a fascinating set of
questions that are posed in what can be described, for want of a
better phrase, as mathematical philosophy. The nature of some of
these questions can be observed at various stages in the following
treatment, but the temptation to pursue them must be resisted. The
further we probe into the fundamentals of mathematics, the more
shaky does the whole familiar structure appear. This need not
worry us unduly the crazy system works as no other system can.
;
1.2 Numbers of various types.

Numbers are usually taken as self-evident or undefinable entities
which do not require examination. It is easily seen, however, that
the number system is far from simple and that numbers of very
different kinds are freely used in arithmetic and algebra. It is con-
venient, therefore, to spend a little time sorting out the various types
of numbers and showing how they have developed side by side with,
and by reason of, the growth of mathematical analysis from the
simplest notions of arithmetic. Some light is then thrown on the

nature and applicability of mathematical analysis in its present
highly developed form.
We all begin, in arithmetic, with some idea of the integers, the
natural or whole numbers, as concepts intimately connected with
the process of counting or enumeration. It is not generally recog-
nised that counting, and hence the integers, involves two quite
distinct notions, the ordinal and the cardinal notions. The funda-
mental property of integers is that they can be written down in
succession without end :
1, 2, 3, 4, 5, ... .
The integers can thus be used to indicateany order or sequence and

this is one of the objects of counting. We have here the ordinal
aspect of counting and of the integers.. But the integers do more
than indicate order, for we can speak of four men, of four hats and
of four sticks without any notion of order being necessarily involved.
We simply mean that we have a collection of men, a collection of
hats and a collection of sticks, and that the collections have some-
thing in common expressed by the number four. There is, in fact,
a complete correspondence between the members of the three collec-
tions each man, for example, may be wearing a hat and carrying
;
a stick.This is the cardinal aspect of counting and of the integers,

an aspect which enables us to say how many men, hats or sticks
there are and which will later provide us with the basis of measure-
ment.
Developing the arithmetic of integers, we define the processes of
addition and multiplication, the sum and product of two or more
integers being themselves integers. Our first difficulty arises when
we consider the converse process to multiplication. The^ division of
one integer by another does not, in the vast majority of cases, pro-
"
duce an integer. Wecan only say, for example, that three into
" "
seven won't go exactly or that we put down two and carry one ".
In order to introduce uniformity, the number system is extended by
defining the fraction as a new type of number. ...The, wider set of
ryunêrs, integers and fractions, istermed the system of rational
numbers^. Like the integers, the rational numbers can be written
down in a limitless sequence of ascending magnitude, the order being
indicated most clearly in the decimal method of writing the numbers. *
The ordered rational numbers, however, display a property not
shown by the ordered integers. The sequence of rationals is not
"
only of limitless extent but also of limitless density ". As many
fractions as we like can be written down to lie between any two
numbers of the order, between J and 1 for example. The processes
of addition, multiplication and division can be extended to apply
to the whole system of rational numbers.
Even with the rational numbers, however, it is found that the
processes of arithmetic still lack uniformity. Consider the process
of extracting the square root of a given number. The square roots
of a few numbers, such as 36, 169 and 6, are found at once as
rational numbers. But, if rational numbers are alone admitted, we
can only say that the square roots of most numbers, e.g. 2, 3 and 5,
do not exist and we have again reached the unsatisfactory situation
in which we have to ,owii that a simple arithmetic process breaks
down. For uniformity, therefore, a further type of number is intro-
duced, the irrational number. ^rational numbers include the square
* A certain amount of duplication must bo eliminated by writing },},$, ... ,
for example, in the simplest equivalent form J.

NUMBERS AND VARIABLES 5
" "
roots of numbers which are not perfect squares and also the
solutions of many equations and such numbers as those denoted
by TT and e which play extremely important parts in mathematical

analysis. The definition and nature of irrational numbers will not
be discussed here but it can be assumed that they fit into their
appropriate places in the order of numbers according to magnitude.*
Further, with some difficulty, we find it possible to extend the
arithmetical processes to apply to irrational as well as to rational
numbers.
The next extension of the number system is useful in arithmetic
but only becomes vital when algebra is developed. We have not yet
referred to subtraction, the process converse to addition. When a
smaller (rational or irrational) number is taken from a larger one,
the result is an ordinary number of the same kind. This is not true
when the numbers are reversed. In arithmetic it is and
desirable,
in algebra essential, that the difference between any two numbers
should be a number of the same system. This uniformity can only
" "
be achieved by doubling the number system so far described by
the distinction between positive and negative numbers and by adding
"
a new number zero indicating nothing ". The difference between
any two numbers is now a number (positive, negative or zero) and
subtraction is a uniform process. By the adoption of further con-
ventions, arithmetic processes can be made to apply to positive
all
and negative numbers and (with the exception of division) to the

number zero.
An
important practical fact can be noticed in passing. The last
extension of the number system makes possible a great simplification
of algebraic work. In algebra we deal frequently with equations such
as the following :
" "
We recognise these as examples of the same form of equation
and (with the aid of zero and negative numbers) we can include all
ofthem in one uniform expression. We write the equations :
(-3)=Q; 2x* + -3)x +1=0

( ;
+(-5)=0.
* See 4.1
(footnote) below.
In the symbolism of algebra all equations of these kinds can be
included in the general quadratic form :
where and c stand for some definite numbers (positive, negative

a, b
or zero). The use of symbolic letters, such as the a, 6 and c in the
quadratic equation above, will be described more fully at a later
stage.
1.3The real number system.

The number system as now extended consists of the neutral
number zero and of all positive and negative numbers, rational or
irrational. This is the system of real numbers, as commonly used in
algebra and mathematical analysis. The rational and irrational
numbers can be arranged in an order of limitless extent and density.
The same is true of the real numbers except that the order is of
limitless extent in two directions instead of one. All negative
numbers appear before the positive numbers in the order of magni-
tude (with zero separating them) and we can proceed through larger
and larger negative numbers as well as through larger and larger
positive numbers. The real number system has also another pro-
" "
perty, i.e. the property that there are no gaps left in the order
" "
of the numbers. gaps in the order of the rationals but
There are
these are completely filled by the insertion of irrational numbers.*
The familiar rules of arithmetic and algebra apply, therefore, to a
system of real numbers with the properties :
(1) The numbers can be arranged in a definite order, the order of
magnitude, of greater and less.

(2) The number order is of limitless extent in both directions,
i.e. the numbers form a doubly infinite system.
"
(3) The number order is of limitless density and without gaps ",
i.e. the numbers form a perfectly continuous system.
The development of the real number system from the basic idea
of a sequence of integers marks the end of a certain line of evolution
and provides a convenient stopping place. One guiding principle
stands out in what we have said. The introduction of each new
number type was designed to impose uniformity where uniformity
* This is a statement we cannot justify here. It is based on the work of
Dedekind and Cantor ; see Hardy, Pure Mathematics (3rd Ed., 1921), pp. 1-31.
was previously lacking and, at the same time, to preserve all the
essential rules of arithmetic and algebra. This is a most important
principle which will be found at work throughout the development

of mathematical analysis, the constant striving after generality,
uniformity and simplicity. Herein lies much of the fascination of
the study of mathematics.
Even with the real number system, however, uniformity is not
completely attained and, sooner or later, analysis reaches a stage
where new extensions of the number system are required. The pro-
cess of division, for example, still fails in one case since a real number
divided by zero does not give a real number. We must turn a more
or less blind eye to this deficiency though it will be partly side-
" "
tracked later when the idea of a limit is introduced. more A
important lack of uniformity occurs when the solution of quadratic
equations is considered in algebra (see 2.8 below). Tf real numbers
are alone admitted, some quadratic equations have two solutions
(distinct or coincident) and others have no solution at all. It is
clearly desirable, from the point of view of uniformity, to be able

to say that all quadratic equations have two solutions. This
uniformity is, in fact, achieved by introducing what are termed

" " " "
complex imaginary numbers, a step which opens up entirely
or
new fields of mathematical analysis. But we content ourselves here
with the real number system which is sufficient for most of the
mathematical methods with which we are concerned.
Finally, the fact that the real number system is of limitless extent
and density makes it only too easy to speak of infinitely large
numbers and of numbers whose differences are infinitely small.
Right at the beginning of our development, therefore, we find our-
"
selves approaching the difficult country of the infinite ", of the
infinitely large and the indefinitely small and continuous. A more
determined expedition into this dangerous country must be left
until asomewhat later stage it is sufficient here to remark on the
;
need for such an expedition.
1.4 Continuous and discontinuous variables.
and of algebra into

TijCL generalisation of arithmetic into algebra
modern analysis is
largely dependent on the extension of the
symbolism. The number system remains the basic element but its
uses are made more by the symbolism. The first and most
flexible
obvious distinction between algebra and analysis on the one hand
" "
and arithmetic on the other lies in the use of variable numbers,
denoted by symbolic letters, instead of particular numbers.
Avariable number is any number, an unspecified number, from a
certain given set of real numbers and it is always symbolised by a
letter such as x y or t. Particular numbers for which the variable
y
can stand are called the values of the variable, and the whole set of
possible values makes up the range of the variable. The use of a
variable number necessarily implies a range of variation, a set of
numbers from which the values of the variable can be selected at
will. Many variables have a range including all real numbers and,
in this case, no explicit reference to it need be made. Other variables
have more restricted ranges, e.g. the range of positive numbers, of
numbers between zero and unity or of integers only. Here the range
should be clearly indicated in the definition of the variable.
One particular case merits separate notice. If a variable x can
take, as its values, all real numbers lying between two given numbers
a and 6, then its range is called the interval (a, 6). We write a <x <6
or a<#<6 according as the values a and b themselves are excluded
or included. For example, 0<x<l or 0<#<1 indicate that the
variable x has a range consisting of the interval of positive proper
fractions, the numbers zero and unity being excluded in one case and
included in the other.
The notion of an
interval provides a concept which later proves of
great convenience. We often make statements about the values of

" " " "
a variable x near to, or in the neighbourhood of, a particular
value a. The statements are made in this vague way but their mean-
ing can be made precise and should be borne in mind. neighbour- A
hood of the value x a is defined as an interval of values of x having
the value a as its middle point. In symbols, if k denotes a given
positive number, a neighbourhood of x a is defined by the interval
(a
- k) <x <(a +k). Here, (x - a) is numerically less than k and the
" "
total length of the interval is 2k. Though k is, in general, any
positive number we care to usually convenient to take k as
fix, it is
small, i.e. to take a small neighbourhood of x a.

An important distinction can now be made. A variable is con-
tinuous if its range is either the whole set of real numbers or any
" "
interval of the set. The
adjective continuous here is of exactly
the same connotation as the same adjective applied to the number
system itself the values of a continuous variable can be ordered so
;
that they are indefinitely dense and without gaps. It is often con-
venient, for example, to think of a continuous variable as taking
values successively in increasing order of magnitude and we describe
"
this by saying that the variable increases continuously in value
over such and such an interval Notice, however, that any
".
changes in the value of a variable are essentially timeless the ;
values of the variable make up a range of variation which must be

considered as a whole, though it may be convenient to pick out or
arrange the values in any definite way we like.
A discontinuous variable, on the other hand, has a range which IB
neither the whole set of real numbers nor any interval of the set.
The values of such a variable cannot be arranged in order of magni-
tude without gaps. Any set of numbers of a particular type, e.g. the
set of integers or of multiples of -|, provides a range of variation
which is discontinuous. For example, if the price of a commodity
is x pence quoted only in halfpennies, then # is a discontinuous
variable taking values which are positive multiples of J .
1.5 Quantities and their measurement.

It convenient, at this stage, to consider the use of our number
is
system in the interpretation of scientific phenomena. One use

springs to the mind at once, the use in the enumeration of discrete
physical objects or events. In all such uses, whether we are counting
(e.g.) men, road accidents or sums of money, the only numbers
" "
required are the integers themselves. The unit of counting is
here the number one corresponding to a single object or occurrence.
" "
For convenience, however, larger units can be adopted we can ;
count men or accidents in hundreds or thousands instead of in ones,

sums of money in pounds, shillings and pence instead of in the
smallest coin (e.g. halfpennies). The result of enumeration then
appears as a fraction or a decimal instead of as a simple integer.
Thus we can say that the number of road accidents is 53-7 thousands
instead of specifying the number as the integer 53,700. Or we can
say that a sum of money is 1 11s. 4|d. or ljf instead of specifying
that it contains 753 halfpennies. Hence, even in the use of numbers
in enumeration, the.range of the numbers employed can be quite
" "
extensive and the question of the unit of measurement arises as
it does in more cases.
complicated
Enumeration is not sufficient for the description of scientific pheno-
mena. We derive, from our observations, what can only be called
abstracted properties shapes, colours, temperatures, pitches of
and so on and it is essential to
notes, lengths, masses, time-intervals,
connect these properties with numbers if possible. The two aspects
of number, the ordinal and the cardinal, now become important.
The ordinal aspect of number applies at once if the property con-
sidered is capable of arrangement in some order, and this is
the essential prerequisite for any association of numbers with the
property. Some of the properties abstracted from scientific pheno-
mena (e.g. temperatures and lengths) can be presented in order
according to a criterion of greater or less, higher or lower or what-
ever it may be other properties (e.g. shapes and colours) cannot be
;
ordered in this way.* Hence we arrive at our first distinction. Pro-

perties not capable of arrangement in order can be termed qualities
and numbers are not directly applicable to them. Properties to
which a natural order is attached can be termed magnitudes and
numbers can be associated with them at once. The number attached
to a magnitude simply serves to indicate its position in its appro-
priate order and the association is loose and by no means unique.

The is, however, of considerable importance in the
association
the Centigrade and Fahrenheit readings of tempera-
scientific field,
ture providing a good example, and more will be said about it later
on.
A second distinction can be drawn amongst magnitudes according
as the cardinal aspect of number can or cannot be applied in addition
to the ordinal aspect. If a magnitude displays the additive property
that any two of its values can be taken and added together to give
a third value of the same magnitude, then numbers can be used in
their cardinal aspect and the processes of arithmetic apply. Magni-
tudes with the additive property can be called directly measurable
quantities and we shall show how numbers can measure, and not only
* It
may be possible, in the construction of physical theories, to analyse a
quality such as colour and to connect it indirectly with numbers (e.g. by
means of "wave lengths"). This merely indicates that our distinctions are
not hard and fast as, of course, they cannot be.
order, quantities. Lengths, masses, time-intervals and electric

charges are examples of quantities temperatures and pitches of
;
notes are instances of magnitudes which are not quantities.

In order to us consider the quantity length
fix ideas, let what ;
follows applies, with suitable modifications, to other quantities such

as mass or time-interval. The additive property of length makes it
possible to construct a scale for measuring length. definite length A

is chosen as standard and a number of such lengths are placed end
to end to make the scale. This is possible since we know what is

meant by equal lengths and since we can add lengths together to
get other lengths. The standard length, definable in any convenient
way, will be assumed here to be that length known as an inch. Any
given length can be compared now with the scale of inches and
defined as covering so many scale intervals, so many inches. All that
isrequired the process of counting and we get lengths of 2 inches,
is
10 inches, 143 inches, and so on. In other words, any given length
is measured by a number, such as
specified or 2, 10 or 143, when
compared with the standard inch scale.
A difficulty now presents itself. In almost all cases, the scale
reading of any given length must be taken to the nearest number of
inches. As a result, two lengths may have the same measure though
direct comparison shows them to be unequal or the sum of two or
;
more lengths may give a scale reading which is not the sum of the
separate scale readings. The fundamental ordered and additive

properties of length appear to have been lost in the process of
measurement, a state of affairs which cannot be tolerated. The
measurement of length must be made to preserve the fundamental
properties.
The first step is to admit fractional as well as integral numbers in
our measure of length. We can subdivide the scale unit by taking
(e.g.) tenths or hundredths of an inch and then compare the given
length with the subdivided scale. Alternatively, we can apply a

given length repeatedly to the fixed inch scale until the multiple
length coincides with a scale reading. Thus, the length of 2f or 2-75
inches is such that it covers 275 of the hundredth inch units, or such
that a hundred such lengths together cover 275 of the inch units.
In practice, this process will suffice since there must be some
physical limit to the comparisons that can be made between a
given length and any actual inch scale. But the measurement of
lengths by means of rational numbers fails in our theoretical require-
ments of length. Pythagoras' well-known theorem, for example,
requires that the diagonal of a square of one inch side should be
represented by J2 inches, and J2 is not a rational number. We
must, therefore, conceive of lengths measured by irrational numbers,
even though actual measurement with a scale of inches does not
support us.
What has happened now is that we have associated the concept
of the infinitely small and continuous with the observable property
of length and, like the abstract number system itself, we assume that
length is continuously variable. In other words, a new and abstract
concept of length is introduced ;
we assume that there is a " true "
length which is continuous but incapable of actual measurement.
This assumption of continuity, which must be recognised as some-
thing outside the sphere of actual observation, is imposed upon
length for purposes of mathematical convenience, in order that the
measure of length should display the additive property and obey
the rules of arithmetic exactly as we desire. The assumption of
continuity into the phenomena we study is of frequent occurrence
and the present instance is only the first of many. The only differ-
ence between this and some later instances is that we have grown
used to the assumption in this case, but not in others, so that we
" "
think it reasonable here but not there.
In the description of scientific phenomena, therefore, there appear
certain directly measurable quantities. Only three of these quan-
tities are required in mechanics, length for the description of space,
mass for the description of weight or inertia, and time-interval for

the description of temporal changes. Other fundamental quantities
must be added as the range of physics is widened. In electricity,
for example, we need the new fundamental quantity of electric
charge to describe the phenomena then included. Having selected a

standard scale unit, we can express each quantity as a variable
number. The recognised units of theoretical physics for measuring
length, mass and time are the centimetre, gramme and second
respectively. Other units are possible and are often used in more
mundane descriptions the units of a foot and a pound can replace,
;
for example, the standard centimetre and gramme units.

Finally, the property of continuity is imposed upon the measure

of each of the fundamental quantities. When we speak of a length
of a; centimetres, a mass of y grammes or a time-interval of t seconds,
the numbers x, y and t are assumed to be continuous variables subject
to the rules of arithmetic and
algebra. This seems reasonable
enough, but the remarks above serve to indicate, though inade-
quately, the serious abstractions and assumptions that lie behind
bald statements about the measurement of fundamental quantities.
1,6 Units of measurement.

The measurement of a quantity, as defined above, is only unique
if the standard scale unit is given. The choice of the standard unit
of each quantity open and can be made in a largo number of ways.
is
As we have noted, the scale unit of length adopted in theoretical

work is almost invariably the centimetre. But inches, feet, yards,
miles, millimetres, kilometres and rods or poles provide other recog-
nised examples of length units. It appears, therefore, that there is a
large element of arbitrariness in the measure of any quantity and

the question arises whether there is any real distinction between
quantities, with measures arbitrarily dependent on the choice of
scale unit, and magnitudes in general, indicated by numbers without
any pretence at uniqueness. It will be shown that the distinction is
real, that the arbitrary element in the measure of a quantity is much

more limited and easily dealt with than in the indication of non-
measurable magnitudes.
There exists a very simple rule connecting the measures of a given
quantity on two different scales. Let the second scale unit be
measured and found to contain A of the first scale units. If a
quantity has measures x and y in the first and second scale units
=
respectively, then we have x Xy, a relation which holds for any one
quantity whatever of the kind considered.
Hence, the numerical measure of a given quantity changes when
the unit of measurement is changed, but the change in the measure
is quite simple and follows the familiar rule of proportions. The
proportion is fixed by A, the measure of one unit in terms of the other.
For example, a length can be expressed either as x feet or as I2x
inches, whatever the value of x. The length measure changes
according to the fixed proportionality rule independently of the
length we happen to take. The connection between units may
be expressed in fractional, or even irrational, form. One inch is
measured as 2-54... on the centimetre scale and a length of x inches
is thus (2- 54...) x centimetres.
The measures of a series of quantities of the same kind on one

scale are exactly proportional to the measures on a second scale.
For example, the same set of lengths can be measured by
6, 12, 18, 24, 30, ... inches ;
I 1, f, 2, f, ... feet;
or , , , f, f, ... yards.
The limitation on alternative measures can be expressed by saying
that the ratio of the measures of two quantities of the same kind is
independent of the units chosen. The ratio of the measures of the
first two lengths instanced above is always | the first length is half ;
the second, a statement independent of scales.

The arbitrary element in the indication of a magnitude which is
not a directly measurable quantity is subject to no such limitation.
A set of magnitudes of the same (non-measurable) kind arranged in
ascending order can be indicated by any rising set of numbers
whatever :
1, 2, 3, 4, 5, ... ;
1, 3, 5, 7, 9, ... ;
1, 4, 9, 16, 25, ... ,

and so on.
There is no connection between the various numbers that can

indicate any one magnitude ;
the question of changing units does
not arise.
1.7 Derived quantities.
From the limited number of fundamental quantities necessary

phenomena can be obtained a large
in the description of scientific
number of secondary or derived quantities. The two most frequent
ways of defining derived quantities are by the multiplication and
division of fundamental quantities of the same or different kinds.
It is sufficient to quote, as examples of derived quantities defined
as products, the cases of superficial area and of cubic volume. Area

is obtained as the product of two, and volume as the product of three,
length measures. Probably more important, and certainly more

frequent, are the derived quantities obtained by dividing one funda-

mental quantity by another. The quotient of distance travelled by
an object divided by the time of transit gives the derived quantity
known as speed. If the mass of a body is divided by its volume, we
obtain the derived quantity of density. The quantity of electric
charge passing a given point in a wire divided by the time it takes to
pass provides the derived quantity of electric current. Further, to
quote a simple example from the economic sphere, a sum of money
divided by the number of units of a commodity bought with the
sum gives the price of the commodity.
From their definition it follows that derived quantities can be
measured, but only in terms of two or more standard units. A scale

unit for each fundamental quantity used in defining a derived
quantity must be chosen before the latter can be expressed in
numerical terms. This is made clear by the way in which derived
quantities are commonly specified speeds are indicated by measures
;
such as 30 miles per hour or 25 feet per second, densities by so many

pounds per cubic inch and prices by so many pence per pound. The
measures of derived quantities, like those of the fundamental quan-
tities, are numbers which are taken as continuously variable.
It will be seen later that derived quantities include two distinct
" " " "
types, the average type and the marginal type. Average
speed illustrates the first type if a train travels 45 miles in an hour
;
and a half, then we say that its speed, on the average, is .30 miles per
hour. The second type is illustrated by velocity or instantaneous
speed ;
the velocity of a train is 30 miles per hour at any moment,
if
this implies that the distance travelled in a very short time-interval

from the given moment, divided by the time-interval, is approxi-
mately equivalent to 30 miles in an hour. The price of a commodity,
as usually defined, is an average price. The corresponding marginal
"
concept can be defined, however, and given the name marginal
revenue ", a concept of considerable importance in economic theory.
Both corresponding types of derived quantities are measured in the
same way and in terms of the same pair of scale units.
The various scales necessary for the measure of a derived quantity
are at choice and, if any one unit is changed, then the numerical
measure of the derived quantity is changed. But the change is
governed by a rule of proportions similar to that already described.
Suppose, for example, that a derived quantity is obtained by the
division of one quantity A by another quantity B. The units of
measurement are changed so that one unit of the second scale of A
equals A a units of the first scale, and one unit of the second scale
of B equals A & units of the first scale. A given value of the derived
quantity is measured by x on the first scales and by y on the second
scales. Then we have
For example, the average speed of a train is 30 miles per hour. Now,
30 miles in an hour implies 30 x 5280 feet in an hour, and so
feet in a second. The speed is thus 44 feet per second. In general,

a speed of x feet per second corresponds to y miles per hour if
Again, if the average price of sugar is 3d. per pound, what is the
price in shillings per cwt.? A

cwt. or 112 pounds costs 336d. or
28 shillings. The required price is 28s. per cwt. In general, a price
of xs. per cwt. corresponds to j/d. per pound if
So, one measure of a derived quantity a constant multiple of an

is
alternative measure obtained by changing the units of measurement.

When the derived quantity is the quotient of two quantities, as
often happens, the multiple is simply the ratio of the scale multiples
of the separate quantities.
1.8 The location of points in space.

Our main purpose is to develop the methods of mathematical
analysis. Any introduction of geometrical or spatial considerations,
therefore, will be subsidiary to, and illustrative of, the analytical
development. It will not be necessary to enlarge upon the funda-
mental geometrical problems concerned with space as an abstract
conception and defined in any consistent way we please. We avoid
difficulties by using only the space of our senses, taken as possessing
the properties of being infinite, continuous and metrical. Primary
sensations give us a conception of space and spatial configurations
and tell us that we can assume that distance between points of space
ismeasurable (in definite units) and continuously variable. Further,

space can be regarded as of limitless extent in one, two or three
dimensions as the case may be.
The geometrical considerations given here can, therefore, be illus-
trated by drawing actual diagrams in the plane of the paper (two
dimensions) or by sketching plane representations of spatial models
(three dimensions) It is possible, by giving up visual representations,
.
to extend our geometry to spaces of more than three dimensions

with the same properties as those given above. But this would
require a precise re-definition of all the geometrical concepts which
we are taking as self-evident, a task which it is scarcely appropriate
to undertake here. Occasionally, however, we shall proceed by
" " "
analogy and refer to ^-dimensional distances ", flats and
" "
hypersurfaces as the analogues of three-dimensional distances,
planes and surfaces. It must be remembered that such geometrical

terms are introduced only as illustrative of the analytical methods.
The spatial properties we have assumed here are clearly analogous
to the properties of the real number system. This analogy makes it
possible to locate points in our metrical space by means of numbers,

the method being derived from the way in which we measure
distances in space and depending on the number of dimensions of
the space.
Space of one dimension consists of a continuous and indefinitely
extended straight line which we can denote by L. A practical
method of fixing the position of a point on a line is to state the
distance, in definite units, of the point from a given base point the :
position of a train on a railway track, for example, can be defined

in this way. We
can easily express this method in a precise form
suitable for the line representing our one-dimensional space. con- A
venient point O is selected on L and a definite scale of measurement,
e.g. one unit | inch, is taken. Taking for convenience of wording
"
the line L horizontal ", a convention is adopted whereby all
as
distances measured from to the right are positive and all distances
to the left negative. Then each point on the line L is located without
ambiguity by a number, the number of units (e.g. inches) of the
distance of the point from 0. The number can be positive (if the
point is to the right of 0) or negative (if the point is to the left of 0).
So, the point located by the number -f 3 is 3 units of distance to the
B M.A.
right of and the point located by - 1 is a half unit of distance to
the left of 0. Since the number system and the points on L are con-
tinuous, each point on L is represented by one definite number and
each number corresponds to one definite point on L. The method
depends entirely on the fixing of O and of a definite scale if a ;
different base and scale are chosen, an entirely different correspond-
ence, still unique within itself, is set up.

The next step is to locate points in two dimensions, i.e. in a
plane which can be visually represented by the plane of the paper.

Experience again provides a practical indication of the appropriate
method. A position in a flat country is conveniently located, as on
an Ordnance map, by giving the number of miles the position is
north or south of a given base point, and the number of miles it is
east or west of the same point. The essential fact is that two
numbers or measurements are required. To make the method pre-
cise, fix two straight lines L and L 2 in the plane which cut at right
angles in a point 0. In the visual representation, and for convenience

of wording, the first line L^ can be taken as horizontal and the second
line L 2 as vertical, the positive direction of L: being from left to right
and of L 2 upwards. A definite unit is chosen which serves for
measuring distances along both lines. From a point P in the plane,
perpendiculars PM
and PN are drawn to the lines L and L 2 The .
point P is then located without ambiguity by two numbers, the first

being the number of units in the distance and the second the OM
number of units in the distance ON. In Fig. 1 P is located by the ,
number pair (4, 3), Q by (3, 4), R by (f, -2) and S by -4, -f). (
Of the two numbers which locate the position of a point P, the first
is the distance of P from parallel to L l9 i.e. to the right or left of
L2 ;
if this number is positive, P is to the right of L
2 and if negative,,
P is to the left of L 2 . The second number is the distance of P from

parallel to L2 ,
i.e. the height of P above or below L^\ P is above Ll
if the number
positive and below L^ if the number is negative.
is
The lines L and 2 divide the whole plane into four " quadrants ".
L
In the N.E. quadrant both numbers are positive, in the N.W.
quadrant the first number is negative and the second positive, in the
S.W. quadrant both numbers are negative, and in the S.E. quadrant
the first number is positive and the second negative. The order of
writing the numbers is important an interchange of the numbers ;
(ifunequal) alters the position of the point (see P

and Q of Fig. 1).
From the continuity of the number system and of the plane, each
point of the plane is represented by a definite pair of numbers and
each pair of numbers by a definite point of the plane. Relative to
the fixing of L: and L 2 and of the scale of measurement, the corre-
spondence between number pairs and points is quite unambiguous.
Suppose a table is given showing corresponding values of two
variable quantities. Taking squared paper and ruling two perpen-
dicular lines along which values of the two quantities are respectively
measured, each pair of values can be plotted, by the above method,
as a point on the squared paper. The result is a graph, showing a
Scale: One unit = % inch

FIG. 1.
group of plotted points, from which the relative changes in the

quantities can be traced. It is not necessary here to take the same
scale of measurement for each quantity the units in which the
;
quantities are measured are more or less arbitrary and each scale
can be chosen for convenience of plotting. Examples I provide
practical instances of graphical representations of this kind.

The location of points in space of three dimensions can be carried
out in an exactly similar way. Three straight lines Ll9 L 2 and L$ are
fixed to intersect at right angles in a point 0. With definite positive
directions along the lines and a definite scale of measurement for
all distances, the position of a point P is located in space as follows.
" "
A rectangular box is constructed with three adjacent sides OM ,
ON and OL along the lines L^ L 2 and L 9 respectively, and with P
as the vertex opposite to 0. The position of P is then located by the
set of three numbers which represent respectively the number of
units in the distances OM ON and OL, due regard being paid to sign.
,
Remarks similar to those given in the case of two dimensions apply

to this three-dimensional method of locating points.
1.9 Variable points and their co-ordinates.

The importance of the methods of locating points in space is seen
when they are applied, with suitable modifications of terminology
and notation, to connect variable points and variable numbers.
A variable point in space is a point which can take any position from
As a point varies in position, so
a given range of possible positions.
the numbers which serve to locate the point vary in value. From
this simple fact, we can construct
a most powerful theoretical tool
applying to both analysis and
geometry.
We can start with the two-
dimensional case and, since we
are considering variable points
in general, we can give up any
>- attempt at scale drawings and
rely upon geometrical reasoning
illustrated by simple diagrams.
Fio. 2. Denote two given perpendicular
lines by Ox and Oy drawn with 9
positive directions as shown in Fig. 2 and with the line Ox taken

horizontally and the line Oy vertically. The point O, in which
the lines intersect, is called the origin of co-ordinates and
the lines Ox and Oy are called the co-ordinate axes or the axes of
reference. Any point P is taken in the plane Oxy and perpendiculars
PM and PN are drawn to the axes. With a definite scale of measure-
ment for all distances, let OM contain x units and ON y units of
distance. The point P is fixed by the number pair (x, y), where the
numbers must be written in this definite order. The point P is said
to have co-ordinates (x, y) the first being the a>co-ordinate and the
9
second the y-co-ordinate. Alternatively, the co-ordinate along the

horizontal axis (Ox here) is called the abscissa and the co-ordinate
along the vertical axis is called the ordinate of P.
Once the axes and the measurement arc fixed, all points
scale of
are located uniquely by their co-ordinates. As the point P varies
in the plane, so the co-ordinates become variable numbers with
certain definite ranges of variation. Tf P varies over the whole plane,
then the variables x and y each have the whole set of real numbers
as their range of variation. If P varies in a more restricted way,
then the variables x and y have ranges which may be continuous or
discontinuous and related in all kinds of ways. To a variable point
in two dimensions, therefore, there corresponds a pair of variable
numbers much use of this will be made later.
;
There are several ways of interpreting the co-ordinates of a point.

The point (x, y) can be regarded as distant x units perpendicularly
from the ?/-axis and y units perpendicularly from tho .r-axis. In
other words, the point is a distance x units to the right or loft of the
vertical base line and a height y units above or below the horizontal
base line. From another point of view, the point (x, y) is reached
by going a distance x units from O along Ox and then a distance
y units parallel to Oy, or conversely.
The co-ordinates of particular points should be noticed. The
origin is itself a point in the plane and its co-ordinates arc seen to
be (0, 0). Any Ox must have zero for its ?/-co-ordinate and
point on
it appears as
(x, 0), where the (positive or negative) distance of the
point from is x units. In effect, therefore, a point on the line
which is taken as the #-axis need have only one co-ordinate, and a
variable point on this line is represented by one variable number x.
Similar remarks apply to a point with co-ordinates (0, y) lying on
the ?/-axis. Finally, points in the various quadrants of the plane
have co-ordinates whose signs are determined by the rules already
given (1.8 above).

The extension of the representation of points by co-ordinates to
the case of three-dimensional space presents no essential difficulty.
Three perpendicular co-ordinate axes Ox, O\j and Oz must now be
fixed, intersecting in the origin and with positive directions as
shown in Fig. 3. For convenience, the plane Oxy can be taken as
horizontal, the axis Oz then being vertical. Dropping perpendiculars

from any point P in space to the planes Oxy, Oyz and Ozx (called
" "
the co-ordinate planes) and completing the rectangular box with
sidesOM ON and OL along the axes, let
,
OM
contain x units, ON
y units and OL z units of distance. Then the position of the point P
is fixed by three co-ordinates (x, y, z) written in this definite order.
For fixed co-ordinate axes and a fixed scale of distance measure-
ment, the variation of a point in space is described by three variable

numbers, the variable co-ordinates of the point. Two of the variables,
x and y, determine points in the horizontal plane Oxy while the third
variable z determines the height of the variable point in space above
or below the corresponding point in the horizontal plane.
There no need, of course, to adopt only the point of view taken

is
so far in which variable points are represented by variable co-

ordinates. Since the connection between points and co-ordinates
is quite unambiguous (relative to fixed axes and a fixed scale of
measurement), we can equally well start with variable numbers and

represent them by a variable point in space. For example, given
two variables x and y related in any definite way, we can fix two
axes Ox and Oy in a plane together with a scale of measurement and
so obtain a corresponding variable point P (x, y) in two-dimensional
space. All this is independent of the symbols adopted for the

variables concerned ; to a pair of variables denoted by p and g,
there corresponds a variable point (p, q) in a plane referred to two
perpendicular axes Op and Oq.
The perfect " two-way " connection we have described between
variable points and variable numbers clearly provides the link we
require between geometry and analysis. This link, as we shall see,

enables us to translate an analytical problem into a geometrical or
diagrammatic one, and conversely. We
have, in fact, established
one of the most useful methods of the mathematical technique.
EXAMPLES I
The measurement *
of quantities
1. Explain how a scale of measurement for mass, considered as a funda-

mental quantity, can be constructed. How is a given mass compared with
the scale? Must continuity be imposed upon the moasuro of mass? Consider
the other fundamental quantity, time -interval, on the same lines.
2. An has two equal sides of one inch enclosing an angle

isosceles triangle
of variable size. Show how
the length measure of the third side can give a
scale for measuring angles of less than two right angles, thus deriving an
angular measure from length measure. What is the measure, on this system,
of angles of 60 and 90?
3. Assuming that lengths can be measured along the circumference of a

circle, obtain an alternative length measure of angles, showing that angles up
to four right angles can now be measured. How can this be extended to angles
of any size? What is the relation of this longth measure to the familiar
measure in degrees?
4. Illustrate the change of measure of a quantity by expressing 32 1 inches

in feet ; 1-63 yards in inches ; 25 \ cwts. in pounds 4897 poxinds in tons and
;
0*285 hours in seconds.
5. Given the connection between a foot and a centimetre, find an approxi-

mate measure of a centimetre in inches. Express 5 inches in centimetres,
3 1 yards In metres and 1 mile in kilometres.
6. From the measure of a pound in grammes, find the approximate

measure of a gramme in ounces. Express 2 ounces in grammes and 12 pounds
in kilogrammes.
7. Express 5072 square inches in square yards and 0-038 cubic feet in
cubic inches. One acre is 4840 square yards how many acres are there in a
;
square mile?
8. Express a speed of 20 feet per second in miles per hour. What is the
measure of a speed of 60 miles per hour in. feet per second and in metres per
second ?
9. A uniform body weighs 6 cwts. and has a volume of 5 cubic feet. Find
the density in pounds per cubic foot and in ounces per cubic inch.
10. A cube of uniform substance has a side of 2 inches and weighs 3 pounds.
Find the density in pounds per cubic inch and in grammes per cubic centi-
metre.
* In these examples, take 1 foot = 30-48 centimetres and 1

pound 453 -6
grammes.
11. For a particle of mass m and velocity v, wo define
Momentum m . v and Kinetic Energy = \m . v2 .
How are the measures of momentum and kinetic energy expressed in terms
of longth, mass and time units ?
12. A body of mass 2 moving at a uniform speed of 30 feet per

pounds is
second. Find its momentum and
kinetic energy in pound -feet -second units
and also in centimetre -gramme -second units.
13. Illustrate the fact that the derivation of secondary quantities can be
carried beyond the first stage by expressing acceleration (the change of velocity
over time) in distance and time units. The acceleration due to gravity is
approximately 32 in feet-second units ; what is its measure in centimetres
and seconds?
14. A retail tobacconist buys a brand of tobacco at 5 9s. lOd. per 10 pound
bag wholesale. Adding 15% for his expenses and profit, what should bo the
retail price per pound (to the nearest penny) and per ounce (to the nearest
halfpenny) ?
Tho number of acres of land used in wheat production, tho amount of
15.
wheat obtained in bushels and the total cost of production are all known.
Explain tho derived nature of tho concepts of average product per acre and
of average cost per bushel of wheat.
16. From tho data given in tho table :
lind the average product per acre and the average cost per bushel of wheat
in each case. How
can these derived figures bo used to compare wheat
production in the different regions named?
17. Assuming that a bushel of British wheat, produced under the condi-
tions of tho previous example, is sufficient for 60 pounds of bread, calculate
the cost of production of the wheat content of a two pound loaf. The wheat
producer makes a profit of 7 and the loaf sells at 4d. ; what proportion of
J%
the selling price goes in baking and distribution ?
18. A small chocolate firm employs 15 hands at an average wage of 45s.

per week ; overheads and materials cost 41 5s. per week and the output of
tho firm is 34 tons of chocolate per year. Find the average product in pounds
per hand per week and the average cost of a J pound block of chocolate in
pence.
19. Show that the rate of simple interest is a number expressed only in
time units, varying inversely with time. How is the rate of interest per year
reduced to a rate of interest per week? What is the amount of 100 at
r% simple interest per year at the end of n years?
20. A man pawns his watch for 18s. and is charged Jd. per week by way
of interest. Calculate tho rate of (simple) interest per week that is implied by
this charge. If 12% per year is taken as a standard rate of interest for this
type of loan, is the pawnbroker extortionate in this case ?
Graphical methods
21. Selecting suitable rectangular axes in a plane, plot the points with co-
ordinates (4,
- 1) ; ( -|, 3) ; (2, \) ; (2-3,
- 1-7) and (
- 2-8, 0).
22. Show that the three points whose co-ordinates referred to rectangular
axes in a plane are (5, 7), A
(9, 3) and O ( B
~ 2, - 4) form an isosceles
triangle
ABC. Which are the equal sides? By direct measurement, determine
whether the other side is longer or shorter.
show that the points with co-ordinates - 3, -
23. By plotting a graph, ( 1)
;
(6, 2) ;-3) ( -3, 5)

(1, ;(5, -1) ; all lie on a certain circle. Loco to tho
centre and measure the radius of the circle.
Four points O, / Q and 7? are marked in a plane country. OR is a line

24.
J
,
in a W.E. direction and R is 12 miles from O. Tho position of P is 3 miles

E. and 4 miles N. of O, and that of Q is 4 miles W. and 5 miles N. of R. What
axes and scales are suitable for locating the points graphically? Kind the co-
ordinates of the points in the system you choose, plot the points on grapli
paper and measure the distances OP, PQ and QE. Estimate the length of tho
shortest route from to R calling at P arid Q and check your result by using
Pythagoras' Theorem.
26. Illustrate the arbitrary element in the location of points in a plane by
rectangular co-ordinates by considering the following.
Three points A, B and C lie in a plane in which lines L L and L. (L l hori-
zontal and L 2 vertical) are fixed. A is 3 inches above L v and 2 inchos to tho
left of L 2 B is 6 inches above L and 10 inches to the right of L 2 ; O is 1 inch
;
below L! and 8 inches to the right of L 2 Find the co-ordinates of the points
.
referred to L l and L 2 as axos when the scale unit is equal to (a) one inch,
(6) ono-teiith of an inch, and (c) one foot.
A point O' is now taken 2 inches below L l and 1 inch to the right of L 2 .
Straight lines L/ and L 2 are drawn through 0' parallel to L^ and L 2 respec-
'
tively. Find the co-ordinates of the three points on each of the above scales
when LS arid L 2 are axes.
26. Three mutually perpendicular axes L 19 L 2 and L 3 are fixed to intersect
in a point in space (L 3 being vertical). Find the co-ordinates of the point A
which lies 3J inches from O on 2 ; of the point L
in the plane of l and z B L L
which is 5 1 inches from parallel to L l and 2 inches parallel to L 2 ; and of the
points C and D which are 4 inches above and below B respectively. Draw
a rough diagram to illustrate.
27. Choosing convenient scales, plot a graph of the pairs of values of the
variables x and y given below :
Draw a freehand curve through the plotted points. Is there any evidence of
a " law " connecting changes in x and yl Find an approximate value of y
when x = 3-6.
28. The following pairs of values of x and y are given :
Plot the corresponding points on graph paper, draw a smooth curve through
the points and find the values of y when is - 6 and 5. What can bo deduced
it*
about the variation of y when x varies ?

29. Show, by a graphical method, how the pressure on a piston of a steam
engine varies at different positions of the stroke when the following facts are
given :
30. The following table (data from Schultz Statistical Laws of Demand
:
and Supply) gives p the yearly average N.Y. wholesale price of sugar, and q,
t
the yearly U.S.A. consumption of sugar, for the period from 1900 to 1914.
Plot a graph, one point for each year, to show the related variations of/) and q.
"
(Such a graph is called a scatter diagram ".) Is there any evidence that a
rise in consumption accompanies a fall in price ?
31. Points in a plane can be located by means of co-ordinates with reference
to axes which are not perpendicular (oblique co-ordinates). If lines LI and L 2 ,
intersecting in O at any angle, are taken as axes, the co-ordinates of a point P

are (x, y), where P is reached from O by going a distance x units along L l and
then a distance y units parallel to L z illustrate with a diagram.
.
32. The axes are at an angle of 45 plot the points whose co-ordinates are
;
given by the following pairs of numbers :
(2,2); (0,3); (-4,0) and (-3, -2).

(Scale one unit =one inch.)
:
33. A parallelogram has a horizontal side OP of 4 inches and a side OR of

3 inches at an angle of 60 to OP. Draw the parallelogram on graph paper
and find the co-ordinates of the corners and of the point of intersection of the
diagonals when the axes are taken as (a) OP and OR, and (b) horizontal and
vertical lines at O.
34. An entirely different method of locating points in space is by means of
polar co-ordinatos. A
point O and a direction Ox (taken horizontally) are fixed
in a plane. The polar co-ordinates of a point P
are defined as (r, 0), whore the
distance of P from O is r units and where OP makes an angle of 0, in the
counterclockwise direction, with Ox. Illustrate with a diagram and by
plotting a graph showing the points whoso polar co-ordinates are
(3, 60) ; (1, 300) ; (f, 0) and (f, 135).
36. Referred to rectangular axes, four points have co-ordinates
(1,1); (-2, -2); (3,0); (0, -2).
Fixing the point and tho direction Ox, find the polar co-ordinates of the
points. Illustrate by plotting the points on graph paper.
36. An observer, standing at a position in a plane country, locates four
landmarks A, B, C and I). A is JO miles away in a N.N.E. direction B is ;
duo S. and at a distance of 7 miles C is W.N.W. and 8 miles away and D is

;
distant 15 milos in a S.E. direction. Plot a graph showing the four landmarks
und find tho polar co-ordinates of tho plotted points referred to O and. tho "K.
direction through O. (Scale 5 miles one inch.)
:
CHAPTER II
FUNCTIONS AND THEIR DIAGRAMMATIC

REPRESENTATION
2.1 Definition and examples of functions.
IK mathematical analysis we are concerned with variable numbers.
The important thing about variables is not the way in which each
varies by itself, but the way in which different variables are related
one to another. Mathematical analysis, in short, is the study of
relationships between variable numbers. A simple technical term
is used to describe and symbolise a relationship between variables,
"
the term function ".
The notion of a function is an abstract and general one but
essentially very simple. In order to emphasise the simplicity of the
notion, we can conveniently introduce at once a few examples of
functions as applied in scientific work and everyday life. In applying
mathematical analysis to actual phenomena, the physical and other
quantities of the are expressed by variable numbers.
phenomena
Any observed or assumed relation between the quantities then
corresponds to a functional relation between the variables. The
following examples make this clear.
If a lead pellet is dropped from a height, we know that the distance
it has fallen (in feet or some other unit) depends on the time-interval
(in seconds or some similar unit) since the moment of dropping.
There is a relation between distance and time and the distance
" "
fallen is a function of the time-interval. Similarly, if a gas is
subjected to pressure at a constant temperature, its volume varies
" "
with, i.e. is a function of, the pressure. To take a more everyday
example, the amount paid in inland postage on a parcel (in some
" "
monetary units) depends on, and is a function of, the weight of
the parcel (in pounds or a similar unit). Finally, in economics, the
amount of a certain commodity demanded by a given market is
FUNCTIONS : DIAGRAMMATIC REPRESENTATION 29
taken as connected in some way with the market price of the com-
" "
modity, i.e. demand is a function of price.
In each of these examples, there are two variable quantities repre-
sented, in specific units, by variable numbers. The quantities, and
the associated variables, do not change independently of each other ;
there is a connection between corresponding values, a dependence

of one quantity upon the other. The idea of a function, therefore,
involves the concepts of a relation between the values of two vari-
ables and the dependence of one variable on the other.
Formal two technical terms can now be given.
definitions of
There is an implicit function or functional relation between two
variables x and y, with given ranges of variation, if the values that x
and y can take are not independent of each other but connected in
some definite way. If the value of x is known, then the value or
values that y can take are fixed and not arbitrary, and similarly
conversely. An implicit function is thus a mutual relation between
" "
two variables and either variable determines the other. The
variable y is an explicit function of the variable x if the value or
values of y depend in some definite way upon the value which is

allotted arbitrarily to x. In this case, it is the variable x which
" "
determines the variable y. In the same way, x may be given as
an explicit function of y.
It follows at once that a given implicit function between x and y
fixes two explicit functions- y as an explicit function of x and x as
an explicit function of y. These two explicit functions are said to be
inverse to each other. Conversely, each given explicit function must
arise from some implicit function and has its corresponding inverse
function. The difference between implicit and explicit functions is
mainly one of point of view or emphasis. If the relation between x

and y is regarded as mutual, then the term iniplicit function is
appropriate the variables are on an equal footing. If the relation
;
is regarded from a definite angle, e.g. y as depending on x, then we
use the term explicit function. Here, the variables are arbitrarily
separated and y is called the dependent and x the independent variable.
It is to be noticed, however, that a function is not a causal relation
even from this latter point of view. There
no discrimination
is
between the variables except for convenience, and one variable does
Ha/iisal rpln firms nppnr rvnlv VAt,wp>pn
quantities of actual phenomena and, when such a relation is inter-
pretedby a function, it merely happens that one view of the function

isdominant and the other neglected. Every mathematical function
can be regarded from either point of view every function has its
;
inverse.
The concept of a function is an extremely wide one. The follow-
ing examples serve to show this and will, also, help us to classify
functions in a convenient way for detailed consideration.
Ex. 1. The variables x and y take any numerical values which are
such that the value of y is double that of x. In symbols
2x-y = Q
is the expression of this
implicit function. The two explicit and inverse
functions are
y = 2x and x \y.
The range of variation of x and y consists of all real numbers.
Ex. 2. The variable y depends on the variable x, which can take any
numerical value whatever, according to the algebraic processes indicated
by the symbolic expression
Here we have y as an explicit function of x and the range of variation of x

consists of all real numbers. The implicit function from which it arises
is expressible
The inverse function is more difficult to derive, but it is seen that, to each
given value of y, there corresponds either a pair of values of x or no value
of x at all. Solving the above quadratic equation for a given value of y,
we find (see 2.8 below) that
This gives x as an explicit function of y in which the range of values of y,

for a real value of x, must be limited so that negative values numerically
greater than 4| are excluded.

Ex. 3. The variables x and y take any values whose product is 3. In
symbols, the implicit function is
and the two explicit

3,3
and inverse functions are
y=-* and #=-.

y
Again, the range of x or y consists of all real numbers.
EX. 4. The variable y is defined as the result of the process
y -100 (I -05)*
where the range of values of x is taken as consisting of positive integers
only. The inverse function can be expressed only in terms of logarithms
(see 10.2). This is an example of a function of a discontinuous variable
and it can be noticed that it expresses the amount y that results when
100 is left for x years at 5% interest compounded yearly.
Ex. 5. The variables x and y are related by the equation
an implicit function between x and y. The explicit functions arc
and x-=
The range of values of x for a real value of y consists of all numbers not
numerically greater than 4, and similarly for the range of y. Except
when x or y is 4, there are two values of y corresponding to each given
value of x, and conversely.
Ex. 6. An implicit function between x and y is defined by
3
x3 -ft/ 3^-0.
The two here cannot be expressed in any direct
explicit functions
algebraic way. This does not mean that the functions do not exist.
" "
Given any value of x, a cubic equation determines the corresponding
values of y and we know (see 2.8 below) that there is either one value or
three values of y, according to the value originally allotted to x. The
dependence of y on x is definite and so is that of x on y.
Ex. 7. Corresponding values of x and y are defined by the table :
2/^9 2/^10 2/-12
Here, y is a function ofx and the range of x consists of all positive numbers
not greater than 15. On the other hand, when x is considered as a function
of y, the range of the latter is discontinuous and consists only of the
positive integers between 6 and 12. This function expresses the inland
postage, y pence, as dependent on the weight of a parcel, x pounds (see
P.O. Guide, 1936). A function of this nature is called a step function.
Ex. 8. The variable y is number of marks obtained by

defined as the
the candidate whose official number was x at the June Matriculation
Examination of London University, 1936. Here, a table is necessary
showing the numbers of all candidates with the number of marks obtained
entered against each case. The range of x consists of those positive
integers allotted as numbers to candidates who sat the examination, and
the range of y consists of those positive integers corresponding to total
marks obtained by at least one candidate. Both variables have discon-
tinuous ranges.
2.2 The graphs of functions.

We have shown (1.8 above) how points in a plane can be located
by means of numbers. Direct use can be made of this method in the
graphing of functions. The practical process is described in text-
books on algebra and we need give here only a general account and
a few examples of the process.
From the definition of a given function, a table of corresponding
values of the two variables x and y can be constructed and made as
detailed as we wish. a matter of giving definite values
It is usually
to one of the variables and solving the formula of the function for
the corresponding value or values of the other variable in each case.
If the range of variation is continuous, there is no limit to the
number of entries that can be inserted.
Co-ordinate axes Ox and Oy
are ruled on a sheet of graph paper
(with, e.g., Ox and
horizontalOy vertical) and a definite scale
of measurement is selected. Each pair of values from the table
obtained from the function is plotted as a point on the graph paper.
As more and more points are plotted, it is found, in the case of any
ordinary function, that a freehand curve can be drawn through them.
Either the collection of plotted points, or the freehand curve, is
called a graph of the function. The following examples fully illus-
trate the process.
Ex.1. 2/=
The following table of values of x and y is obtained from the formula
of the function :
Plotting the corresponding points, it is seen that a U-shaped curve can

be drawn through them (Fig. 4). Between the points where x equals -2
and -1, more detail is required than elsewhere and one intermediate
point has been plotted. It is found that the lowest point of the curve is
at or very near to the point ( - 1^, -4J). The graph represents a curve
known as the parabola (sec 3.4 below).
Ex.2. xy=3.
The graph of the function can be plotted from the table :
It is seen (Fig. 5) that a smooth curve can be drawn through the set of
plotted points, except that there is some doubt about its course when x
is small. When x is actually zero, the formula of the function provides
no corresponding value of y at all. It is found, however, that the value of

y increases rapidly as x is given smaller and smaller positive values and
decreases rapidly for smaller and smaller negative values. The graph of
the function thus takes the form indicated it is a representation of a
;
curve known as the rectangular hyperbola (see 3.5 below).
Ex.3. *2 +*/2 = 16.

The following table is derived for values of x from - 4 to + 4 :
the values of y being taken to one decimal place. Outside the range
shown, there exist no values of y at all. At x = 4 there is a single value
c M.A.
of y while at other values of x there are two values of y. The graph of the
function takes the form shown (Fig. 6) it is recognised as a circle of
;
radius 4 units with centre at the origin of co-ordinates.
Ex.4. x*+y*-3xy=Q.
The table of corresponding values of x and y, from which a graph of the
function can be drawn, now appears :
Here, in order to obtain the values of y corresponding to any given

value of x, it is necessary to solve a cubic equation. The values above
O 2
. .
4 Jf
are correct to one decimal place and are obtained by a graphical

method described later (see 2.8 below). The set of points plotted from
" "
the table can be joined by a smooth curve (Fig. 7) consisting of a loop
"
and two tails ". The graph represents a curve, known as the Folium
of Descartes, of a more unusual type than those obtained in the
previous examples.
Ex. The step function defined by the table of 2.1 (Ex. 7) can be
5.
plotted in a similar way. No smooth curve can be drawn through the

plotted points which can, in fact, be connected only by a set of seven
disconnected lines parallel to the #-axis (Fig. 8). This is a function which
will later be described as discontinuous.
A graph is clearly of great service in the consideration of the

properties of particular functions it provides, at least, a visual repre-

:
sentation of the related variation of the two variables. For example,

the varying height of the curve shown 011 the graph (above or below
the horizontal axis Ox) indicates the way in which y varies as x is
changed in a definite way. The graph can also be used to replace
numerical calculations, provided
that approximate results suffice.
In plotting the graph, only certain
convenient (e.g. integral) values
of the variables are used arid,
once this is done, corresponding
values of the variables at inter-

mediate positions can be read off
approximately from the curve
shown. This method of graphical
8.
"interpolation" is used in the sol-
ution of equations (see 2.8 below) and in many other connections.*
The method of graphing we have described illustrates the relation
between two variables connected by a given mathematical function.
But the same method can be used when the variables represent
physical or other quantities given by observed phenomena and when
we know that no " perfect " functional relation can exist between
them, either in the nature of things or because some suspected rela-
"
tion is disturbed by random deviations or errors ". Here we obtain
"
what can be called a statistical ", rather than a functional, graph.
Two cases can be profitably distinguished. One of the observed
quantities x may be ordered in space or time and so that a single
value of the other observed quantity y corresponds to each value
of x. We take x along the horizontal and y along the vertical axis
and the graph shows a proceeding from left to right,
series of points
varying height illustrating the variation of the observed y as the

observed x varies according to its ordered progression. The graph
* For
example, the graph of y
~ Jx can bo plotted by giving x values
which are perfect squares. The values of square roots such as x/2 or s/3 can
then be obtained approximately as the heights of the graph corresponding to
(e.g.) x
= 2 or 3. Alternatively, tho graph of y x'1 can be plotted, using
integral values of x, and the approximate values of (e.g.) ^2 and %/3 can then
bo road off the points on tho axis Ox corresponding to heights of 2 and 3 on
the graph. See Hogben, Mathematics for the Million (1936), p. 415.
may show, for example, the varying observed volume of a gas at
constant temperature as the observed pressure increases over time.
From this graph, we could see whether, apart from errors and
random deviations, a definite functional relation between volume
and pressure can be reasonably assumed. Again, a graph can be
used to show the value of imports into the United Kingdom over a
period of years, or the variation in some other statistical time
In these cases, no functional relation (e.g.
series (see 10.4 below).
between imports and time) can exist and the graph is mainly
descriptive.
On
the other hand, neither of the two observed quantities may
possess a natural spatial or temporal order. We may have, for
example, a series of observed values of the market price (p) of a
commodity and the corresponding market consumption (q). Such
a correspondence can be obtained by talcing a succession of yearly
quotations of p and q (sec Examples I, 30). Each pair of values of
p and q can be plotted as a single point referred to two axes Op and
Oq and the whole set of observations is then represented by a
" " "
graphical cluster of points. Such a graph is called a scatter
"
diagram and its uses, at least for purposes of illustration, are
evident. It may be suspected, for instance, that there exists a
definite relation between p and q, a relation which is disturbed by
errors and other deviations in the actual observations. The correct-
ness of this supposition can be examined and, if a relation is found
to exist, its nature can be discovered.
2.3 Functions and curves.

The graphical method of representing functions can be extended
to establish a general connection between functions and curves. If
we are given a functional relation between two variables x and y of
continuous variation, then there is no limit to the number of entries
that can be inserted in the table of corresponding values of the
variables and, hence, no limit to the number of points that can be
plotted in the plane Oxy. gives rise, therefore, to an
The function
indefinitely large number of points in the plane and only a few of
them are shown in any actual graph. Just as the variables are
related in an ordered the function, so the corresponding
way by
points must display a definite characteristic, i.e. must make up a
locus or curve in the plane Oxy. Hence, to each given function relating
variables x and y, there corresponds a set of points comprising a
curve in the plane Oxy the analytical property defined by the
;
function is reflected in the geometrical property common to all points

on the curve.
Conversely, a curve, as a geometrical concept, is a collection of
points in a plane with a common Fixing co-ordinate
characteristic.
axes and attaching co-ordinates to each point of the plane, the
property defining the curve can be translated into an analytical
relation between x and y satisfied by
all points (x, y) on the curve, a
relation which is called the equation
of the curve. Consider, for example,
a circle with radius 4 units and
centre at a fixed point of a plane,
i.e. the curve consisting of all points
distant 4 units from 0. Fix co-

ordinate axes to pass through O, let
P with co-ordinates (x, y) be any
point on the circle and drop per-
pendiculars and PM to the axesPN Fm. 0.
(Fig. 9). From Pythagoras' Theorem
since OMx units and ON ~y units. But OP must equal 4 units for
all positions of P on the circle. Hence,
is the relation satisfied by all points (x, y) on the given circle, i.e. the
equation of the circle (cf. 2.2, Ex. 3, above).
The correspondence between the functions of analysis and the
curves of geometry is thus a perfect one. To each function relating
variables x and y there corresponds a definite curve in the plane
Oxy, and conversely. It is important to remember, however, that
the uniqueness of the connection depends entirely on the fixing of
co-ordinate axes in the plane. A
given curve has different equations
when different axes are selected.
It is now graph of a given function is simply an actual
clear that a
representation on squared paper of the curve which corresponds to

the function. The function is an abstract concept relating two
variables x and y ;
the curve is an equally abstract concept relating
points in two-dimensional space. If the variables are continuous,
there an indefinitely largo number of pairs of variable values and
is
of points on the curve. A graph, on the other hand, consists of a

finite collection of points selected arbitrarily from the indefinitely
numerous set possible and designed to indicate clearly the shape
of the curve. But the curve itself remains an abstract concept,
of which the graph is only a visual, and more or less imperfect,
representation.
Finally, notice that a curve, as defined above, need not be
"smooth" in the sense that it can be shown graphically as a
continuous line without breaks or angles. Nor need the curve be

described in any temporal way by a moving point. It is often
convenient to think of a point moving along a curve, but this is
only a device to facilitate exposition. A curve, like a function, is
defined and considered as a whole, as a collection of points in a plane
united by a common characteristic.
2.4 Classification of functions.

We now proceed to distinguish functions of different kinds
can
according to the nature of their definitions or of their symbolic ex-
pressions. It is necessary, for this purpose, to introduce a number
of technical but very useful terms.
Afunction is analytical if it can be expressed in symbols by means
of a single formula connecting the, variables, a formula which is the
" "
general law behind the function. From the formula, as we have
seen, values of one variable, corresponding to given values of the
other, can be derived by carrying out the algebraic or other
operations indicated. The first six examples of 2.1 give analytical
functions the other
;
two functions are not analytical.
The next distinction is an extremely important one. Here, we
regard functions as explicit, taking either y as a function of x or x as
a function of y. Explicit functions can be divided into two classes,
those which are single-valued and those which are multi-valued. A
function such that one and only one value of y corresponds to each
given value of x is said to define y as a single-valued function of x.
A function for which this is not true defines y as a multi-valued
function of and we have double-valued, triple-valued and higher

x,
valued functions. The inverse function, x as a function of y, can he
considered in the same way.
The functions of the first four examples of 2.1, as well as those of
the last two examples, all give y as a single-valued function of x.
In the fifth example, y is a double-valued function, and in the sixth
example y a triple-valued function of x. On the other hand, only
is
in the first, third and fourth examples is x a single-valued function

of y . This introduces a very important point the inverse of a single-
;
valued function is not necessarily itself a single-valued function. This

is clear from the inspection of even such a simple function as that of
the second example of 2.1.
In graphical terms, if y is a single-valued function of x, then the

curve corresponding to the function is cut by any line parallel to Oy
in only a single point. But, if y is multi-valued, then the curve can
be cut by such lines in two or more points. Tn the same way, # is a
single-valued (or multi-valued) function of y if the curve is cut by
lines parallel to Ox in only a single point (or in several points). There
is no necessary connection between these two ways of finding the
intersections of a line and the curve a given curve can be cut by a

;
line parallel to one axis in only a single point while lines parallel to
the other axis cut it in two or more points. The single-valued
property does not necessarily apply to both the inverse explicit
functions.
One reason
for the importance of single-valued functions is clear.
When y a single-valued function of x we can take the function
is 9
as described by values of x %hich increase steadily in value the ;
independent variable can then be completely ignored and attention

concentrated upon the variation in y. The value of y may be found
to increase, to decrease, to have alternating increasing and decreasing
" "
stretches or to vary in a more or less erratic way. Graphically,
we take x along the horizontal axis and describe the curve from left
to right. The varying height of the curve above or below the hori-
zontal shows the variation of y and we can see when the curve rises
y
(y increasing) and when it falls (y decreasing).

The remarks lead to the definition of a particularly useful
last
sub-class of single-valued functions. If y is a single- valued function
of a continuous variable x, and if y increases in value as x increases,
40 MATHEMATICAL ANALYSIS FOB ECONOMISTS
then y is called an increasing function of x. Similarly, if the value
of y decreases as x increases, we have a decreasing function of x. The
class of increasing and decreasing functions, taken together, com-
prise what are called monotonic functions. The graph of a monotonic
function of x, the axis Ox being taken as horizontal, either rises or
fallswithout interruption from left to right, rising for an increasing
function and falling for a decreasing function.
" "
Subject to minor qualifications concerning the continuity of the
function (4.6 below), a significant property of a monotonic function
can be established. Thea single-valued and monotonic
inverse of
function is also single-valued and monotonic in the same sense as the

original function. For example, if y is a single- valued increasing
function of x, it follows that only one value of x can correspond to
each value of y and that this value must increase as y increases.
This is quite clear graphically, since the curve representing y as a
single-valued increasing function of x (axis Ox horizontal) rises
throughout from left to right and so the distance of the curve from
the vertical axis increases as we proceed upwards. It is thus only
in the case of monotonic functions that we can always say that both
inverse functions are single-valued.
The and third examples of 2.1 show functions which are
first
monotonic, the first being an increasing function and the other a
decreasing function. In each case, what is true of y as a function

of x is also true of # as a function of y. The second example of 2.1
provides a case of a single-valued function which is not monotonic.
As x increases from large negative values up to x - f the value of
,
y decreases, whereas y increases as x is increased beyond x~ -f.

The graph falls and then rises. The inverse function, x as a function
of y, isnot single-valued, but double-valued, here.
Returning to the case of multi- valued functions, it is sometimes
"
possible to divide such a function into two or more branches ",
each branch being a single- valued function. The corresponding
curve is also divided into two or more distinct sections, each cut by
lines parallel to the appropriate axis in only a single point. The
function of the fifth example of 2.1 is a case in point. The explicit
expression, giving y as a double- valued function of x, is
which can be divided into two single-valued functions,
--
x2 and y
The corresponding curve, a circle, is likewise divided into two single-
valued sections, one above the axis Ox and the other below. In the
same way, the second example of 2.1 gives x as a double- valued
function of y divisible into the two branches
a = |(/4y+17-3) and 0:= -(>/4y +17 + 3).

The corresponding curve, a parabola, is divided into two sections by
a line parallel to Oy at x - each section
ij, being cut in only a single
point by lines parallel to Ox (see Fig. 4). The first branch of this
function is an increasing function of y, the corresponding section of
the curve rising from left to right the second branch is a decreasing
;
function of y and this section of the curve falls from left to right.
Such divisions of multi-valued functions into single-valued branches
are clearly very useful.
Finally, it can be noticed that some functions can be described as
symmetrical. Symmetry can be defined in various ways, of which

the following is
perhaps the most important. An implicit function
relating variables x and y is said to be symmetrical in x and y if an
interchange of these variables leaves the function unaltered in form.
This implies that y as an explicit function of x is of exactly the same
form as x as an explicit function of y. The function xy 3 is sym-
3 3
metrical on this definition and the inverse functions, y =- and x - ,
x y
are seen to be of identical form. Graphically, such a symmetrical
function is represented by a curve symmetrical about lines bisecting
the angles between the axes. The part of the curve on one side of the
line is the reflection in the line of the part of the curve on the other
side. The graph of xy = 3 illustrates this fact (see Fig. 5).
2.5 Function types.

So far we have considered only a number of particular functions
such as appear in the simplest operations of algebra. Our next task
is to group functions into types and to devise a symbolic notation
to include all functions of one type in a single formula. The grouping

of functions into types proceeds by means of the notion of analytical
" " "
form ". This notion, which must be sensed rather than defined,
is essential to the extension of the processes of algebra into the more
powerful methods of analysis, and its nature is indicated most easily
by giving concrete examples.
The explicit function y 2x is clearlybut one example of a wide
"
range of functions of the same form ". Other examples are
y=-3x; y-5#+2; y #-3 and y = l-2x.

All such functions can be included in the single formula
y = ax+b,
where a and b denote any definite numbers, positive, negative or zero.
" "
The function type, represented by this portmanteau formula,
is described as an explicit linear function of x.* In the same way,
we can define # as a linear function of y. Further, putting the linear
function into implicit form, we can write
ax+by +c~ 0,
where a, b and c stand for any three numbers whatever. The implicit
linear function gives rise to two explicit linear functions :
and *=
(-
the coefficients of which are expressed in terms of the three coeffi-
cients of the original implicit function, f

As a second example, we see that the explicit function
is one example of the function type, called the explicit quadratic

function, which can be symbolised
y=ax 2 +bx+c,
where a, 6 and c denote any definite numbers. In the same way, we
can have x as an explicit quadratic function of y. Putting either of
" "
these functions into implicit form, it appears as a polynomial
expression, containing no powers or products of the variables of
* " "
Tho term linear is used since, as wo see later, such a function is repre-
sented graphically by a straight line.
t The cases where a or 6 equals zero are a little troublesome (see 3.3 below)
but need not delay us here. Further, it may seem odd that there are three
coefficients in the implicit form and only two in the explicit form. This is,
however, only a matter of convenience (see 3.7 below).
higher degree than the second, equated to zero. Other implicit

functions are also of this form. The function type of which xya,
2 2
xy 3 is one case, and the function type ax +by c, of which
x*+y* lft is an instance, are examples. We derive, therefore, a

wider function type, symbolised by the formula
ax 2 +2hxy +by* +2gx +2fy -f c = 0,

which can be called the implicit quadratic function relating the
variables x and y. This includes, for certain values of the coeffi-
cients a, 6, c, /, g and the explicit quadratic function and also
h,
many others. The grouping of functions into types is not necessarily

unique ;
there are several overlapping types possible according to
the particular point of view adopted.
The number of different function types that can be distinguished
" "
is indefinitely
large. Amongst algebraic functions, we can go
" "
on to separate cubic functions of explicit and implicit form,
" "
quartic functions, and so on. There are also numerous functions
of forms we have not yet introduced and some of these types will
be considered at a later stage. The exponential, logarithmic and
trigonometric functions are notable instances.
The grouping of functions into types, largely a matter of con-
venience, is characterised by the use of symbolic letters other than
those standing for the variables themselves. The familiar algebraic
device of representing unspecified numbers by letters has been signi-
ficantly extended and, before proceeding, it is essential to get some
idea of this extension. The symbolic letters, such as a, b and c in the
" "
above formulae, must be variable since they stand for unspeci-
fied numbers, but there must be some difference between them and
the original variables x and y. This difference, on which the whole
grouping of functions into types depends, must be made clear.
To fix ideas, consider the quadratic function type which we can
write y=ax 2 +bx+c. Two points of view can be adopted. From
the first
point of view, we investigate the relation between the vari-
ables x and y and the numbers a, b and c must be taken as fixed, as
having the same numerical values whatever they may be. From
the other point of view, the function is treated as a whole and the
type includes a whole set of particular functions united by the
"
common property of being quadratic ". The formula, in fact,
represents any function of quadratic form. Here, x and y cease to
be of importance and the whole function is made to vary by changing
the values of a, b and c. These latter numbers are fixed within one
function of the quadratic type but variable from one quadratic
function to another.
Hence, the characteristic of letters such as a, b and c is that they
"
lead double lives ; they are variable constants ". To suggest this
double use, we describe such symbolic letters as parametric constants
and one of two shorter terms can then be used according to the point
of view. When attention is directed to the relation between the
variables of the function, the symbols denote fixed numbers and can
be termed constants. When
the important thing is the variation of
the function as a whole, the variables of the function fading out of
the picture, we describe the symbols as parameters.*
We have tried to show how the functional notation is made
more general and by the use of parametric constants. The
flexible
quadratic function type, for example, becomes a particular function

once numerical values arc allotted to the parameters. But, taking a
more general view, we can attempt to derive properties, not of one
quadratic function, but of all quadratic functions. Or we may know
that some quadratic function satisfies a given condition and then
proceed to determine the values of the parameters for which the
condition holds. This, for example, is the idea behind the fitting of
" "
a parabolic trend to a statistical time-series.
A simple algebraic instance can be quoted. It is required to find
the height of an open box of square base of side 4 inches so that the
surface area is 48 square inches. If the height is x inches, the surface
area (y square inches) is a function of x :
Putting ?/ = 48, the value of x is found to be 2 inches. This problem

can be generalised and solved for all cases by the use of parameters.
If y square inches is the surface area of a box of square base of
given side a inches and of variable height x inches, then
y=a 2 +4ax and x

4a
(y-a 2 ).
* The term parameter comes from Trapa/Acrpcco which means " to compare
"
or "to measure one thing by another ".
This a functional relation between x and
is y. If the surface area is
known to be b square inches, then

b-a*.
x=
,
inches.
4a
We have solvedproblems of this type, and the solution of any
all
particular problem is derived by allotting the parameters a and 6

the appropriate numerical values.
Finally, the use of parameters is carried over into the mathe-
matical sciences, as is seen, for example, by the parametric role of
"
market prices in economic theory. A
firm (under conditions of pure
competition ") must take as given the market prices that happen to
exist and we can assume its output is known for each given price
system. In combining the output decisions of all firms, we treat the

price system as variable and attempt to discover a set of prices con-
sistent with equilibrium (however the latter is defined). Prices are
parametric constants, constant in the examination of the decision
of a single firm, parameters when we combine the separate decisions
of a whole group of firms.
2.6 The symbolic representation of functions of any form.

Afurther generalisation of the functional notation now takes us
outside the limited field of algebra, to which we have been confined so
far, into the wider territories of modern analysis. The idea involved
is simple. Just as a symbol
is introduced in algebra to stand for
any unspecified number, so now a symbol is introduced to stand

for any function. In algebra, we deal with the properties of functions
of particular types ;
with the new notation, mathematical analysis
goes far beyond this and introduces more powerful processes, such
as those of the calculus, to deal with functions of any kind whatever.
The development of mathematics, here as elsewhere, is dependent on
the development of its notation.
An implicit function relating two variables x and y can be repre-
sented, whatever its form may be, by the notation
/(*,y)=o.
This is certainly appropriate for any analytical function, the sym-
bolic f(x, y) standing for some expression involving x and y. To
obtain, from the general notation, a particular function, we need
only specify the exact form of f(x, y) ; we take/(#, y)
for example, if
as 2x - /, x 2 -f-j/ 2 - 16 or x z +y 3 - 3xy, wo obtain a particular function
y = 2x, # +?/ 16 or # 3 -fz/ 3 - '3xy = Q. The notation can also be
2 2
applied, without confusion, to include even non-analytical functions.

Again, if y is an explicit function of x of any form whatever, we
can denote it by the notation
y =/()
This notation is clearly applicable to the case of a function which is
single-valued and analytical.* In such a function, y is defined as
equal to the value of a certain expression involving x, and f(x) can
be taken as a convenient way of representing this expression in x,
no matter what its form may be. To derive a particular function,
we supply a definite form for f(x) if we take (e.g.) f(x) as 2x,
;
3
x 2 +3# - 2 or -
^ we obtain three well-known single- valued functions.
x
The notation can also be extended to apply to functions which are
not single-valued or even analytical in its broadest use, y ~f(x)
;
signifies only that we are taking y as some function of x. Notice that,

in this notation, we need not refer to the variable y at all we can ;
say that f(x) is a function of x. For example, we often describe

2
(x + 3x
- 2) as a function of x. This is a matter of convenience only ;
a function always relates the values of two variables whether we

care to suppress reference to one of them or not.
Since the importance of a good notation, both general and flexible,
is evident, we can insert here a number of remarks on the symbolism
of mathematical analysis. Letters are used to denote both variables

and parameters, and, in order to distinguish one from the other, it
is usual to reserve the later letters of the alphabet for variables and
the earlier and middle letters for parametric constants. The letters
x, y z, u, v and t are most frequently used for variables, and, if these
t
do not suffice, the Greek letters , 77

and f are called into service.
When parametric constants are regarded primarily as constants,
* An
alternative, and in some ways a superior, notation is occasionally used.
Instead of writing y=/(x) to denote the single- valued function, wo write
y=y(x). This economises letters and enables us to write the dependent
variable by the single letter y when the dependence on x is not stressed, and
by the same letter with x in brackets when the dependence on x becomes
impdrtant. For an actual use of the notation in economics, see Frisch,
New Methods of Measuring Marginal Utility (1932).
the letters a, 6, c, ... ,

or the corresponding Greek a, /3, y, ... ,
are
conventionally used. When the parametric property is the more
important, it is often useful to take &, I, m, n, p, q and r,
or *, A, ft, p, to
v and
denote the parameters. These remarks
serve only as general guides the context should be sufficient, in
;
any particular case, to make clear the nature of the symbols used.
The letters denoting functions are of different nature from the
variable or parametric letters they do not denote anything to
;
which numerical values can be allotted but stand for the complex
notion of the form of a function. It is usual to reserve the letters
/ and g, and the Greek <f>

and /r, together with the corresponding
capitals, for the denotation of functions. It is also possible, for
greater variety, to add suffixes to the functional letter, fl9 /2 /a , ... ,
and fa, fa,^ being examples.

3 , ...
It is, of course, essential to denote different variables, parameters

or functions by different letters. For example, if we wish to say that
any implicit function gives rise to two explicit and inverse functions,
we must symbolise somewhat as follows. From the given implicit
function f(x, y) 0, we derive the two functions y = <f>(x) and x~*/j(y)
which are inverse to each other.
Functions refer to operations performed on variables, and it is
thus necessary to denote a function, not only with its operational
or functional letter, but also with an indication of the variable or
variables* towhich the operation applies. The notation f(x) for an
explicit function makes this clear, the letter / being the functional
letter and the letter x in brackets denoting the variable to which
the operation applies. Other operations, such as those of derivation
and integration, will appear later and will be denoted by further
symbolic devices of the same kind.
One more notational device remains for consideration, i.e. the
method of introducing particular values of the variables into a
general function. Suppose y f(x) is some explicit function in which
we give, to the variable x, definite numerical values such as
x 0, a? = J, x = l. The corresponding values of y are then written
/(O), /(|), /(I). Further, we can allot values to x which are fixed
but not specified. Such values of x can be denoted by a, 6, c, ... ,
or
by the letter x with suffixes or primes added :
*î> *^2> *^3>

or x x x , , ,
... .
The value of y when x isgiven the fixed value (e.g.) a or xt is written
f(a) or /(a^). These fixed, but unspecified, values of the independent
variable are, of course, instances of parametric constants. To illus-
trate with a particular function, if

= x 2 + 33-2,
f(x)
2
then /(0)-(0) +3(0)-2--2;/a)-(J)2+3(i)-2^-J,
/(l)
= (l)a+3(l)-2=2; and so on.
Further, f(a)
= a 2 + 3a - 2 .
Again, if/(#, y)~ is an implicit function, then /(a, y) = Q indicates

an equation to be solved for the values of y corresponding to # a,
and f(x, b)=0 an equation giving the values of x corresponding to
y =b. Specific numbers can, of course, be substituted for a or b.
As an example in the use of the functional notation, consider the
following operations of which much use will be made later. If a

and (a +h) are two particular values of a variable x, where a and h
are any and f(a f-A) are the
positive or negative values, then f(a)
corresponding values of a function f(x). The change (or increment)
in the value of the function corresponding to the given change in x is
f(a+h)-f(a).
The change in the function per unit change in x is then
f(a+h)-f(a)
h
and this is defined as the average rate of change of the function when
x changes from a to (a -f-A). The actual changes shown in x and
f(x) can be positive or negative according as they are increases or
decreases.
2.7.The diagrammatic method.

We
have seen that a definite curve referred to selected axes in a
plane corresponds to each particular function we care to define, and
conversely. Further, when we are dealing with definite functions
and curves, the graphical method, in which the graph of a function
or curve is plotted on squared paper, is of great service. (But it v
is between functions and curves

easily seen that the connection
"
remains of the greatest importance when we pass from particular

functions to function types. It is reasonable to suppose, and it is in
fact true, that the curves corresponding to a set of functions of the
same type form a class of the same general nature. To a function
type there corresponds a curve class, and properties common to
all functions of the type are paralleled by properties common to all
curves of the class. This correspondence \s investigated in the branch

of mathematics known as analytical geometry, where we consider
what function types correspond to particular curve classes and what
analytical properties of the functions reflect various geometrical
properties of the curves^ Some account of analytical geometry will
be given in the following chapter.
The function-curve correspondence would appear to be useless,
" "
however, when we come to arbitrary functions of unspecified
form. AIL that can correspond to an arbitrary function is an arbi-
trary curve, and neither graphical methods nor analytical geometry
avail. But it always happens, in our analysis, that functions,
though otherwise arbitrary, are limited by certain general properties.
It may be, for example, that a function is limited by the conditions
thatit is single-valued, positive and decreasing, or that it decreases
up to a certain point and then increases. The curve which corre-

sponds to the function, though still largely arbitrary in position,
now displays sufficient properties to make the drawing of an illus-
trative diagram worth while. In the first of the above instances,
the diagram would show a curve lying completely above the hori-
zontal axis and falling steadily from left to right.
It is here that the diagrammatic method of illustrating analytical
arguments becomes important. The least service diagrams can

perform is to illustrate and check our analytical development. It is
possible, in somewhere the properties of our functions are very

cases
definite, to make more positive use of diagrams by employing them

in conjunction with simple geometrical arguments. But we must be
careful to avoid mistaking casual properties of the curves in our
diagrams for essential general ones, and to prevent ourselves accept-

ing apparently self-evident results without formal proof.
^ It is to be noticed
that one curve serves to represent a given
function, whether the latter is expressed in implicit form or in either
of the inverse explicit forms. We need examine the shape of only
one curve to illustrate properties of a functional relation between
x and of y as a function of x and of # as a function of y. To see,
y,
for example, how many values of y correspond to a given value of x,
M.A.
and conversely, we look for the number of points of intersection of
the curve and lines parallel to the axes. The use of the diagram-
matic method on these lines is illustrated in 2.4 above.
A minor diagrams concerns the fact that
difficulty in the use of
we can fix the co-ordinate axes in two different ways, even when we
adopt the established convention of drawing one axis horizontal.
A given curve may look quite different when one variable is measured
along the horizontal axis than it does when the other variable is so
measured. Since it is important to be able to recognise the same
curve however the diagram is drawn, we can indicate here a way in
which a curve shifts when the axes are transposed. Fig. 10 shows
FIG. 10.
the curve corresponding to a given function relating x and y, firstly

when Ox is horizontal and secondly when Oy is horizontal. The two
" "
positions of the curve are such that one position is the reflection
of the other in the bisector of the angle between the axes, i.e. in the
broken line shown in the figure. Hence, if we start with the curve
in the first position and rotate the plane containing
through 180
it
about the bisector indicated, the new position taken up by the curve
is the position with axes interchanged.
2.8 The solution of equations in one variable.
As anexercise in the use of analytical methods supported by
graphs or diagrams, we can consider here the important question of

the solution of equations, to which many mathematical problems
reduce. An equation in a single variable x can be written, in general,
=
as/(#) 0, where/(#) denotes some given expression or function of re.
Our problem, in solving the equation, is to find the value or values

of x which make the value of f(x) equal to zero.
The simplest type of equation in one variable occurs when f(x) is
a polynomial, i.e. the algebraic sum of terms vhich involve only
various powers of x, and such equations can be classed according to

their degree, as determined by the highest power of x involved.
There are linear equations (of the first degree), quadratic equations
(of the second degree), cubic equations and so on. A solution of such
an equation, a value of x which makes the polynomial zero, is
usually called a root of the equation. The process of finding roots
of polynomial equations, at least in simple cases, is fully described
in text-books on algebra.
The general linear equation is ax+b = Q and this has one root,
x -- ,
in all cases. The general quadratic equation is
a
of which the solution is known to be
2a
It follows that there are two real roots if 6 2 > 4ac, and no real root
if 6 2
<4ac. 4ac, then the two real roots coincide and there is
If 6 2
only one value of x satisfying the equation, which reduces, in this

case, to a perfect square equated to zero. The formula provides the
roots indicated in the following examples of quadratic equations :
2# 2
-#-l=0, x-if and # 1,
o;
2
-3a+l=:0, a = (3+V5) and x^fi-JS),
x2 +x + 1 0, no real roots.
In the example, the roots can also be found by a simple process
first
of factorising the quadratic expression.

Cubic and higher degree equations present much more difficulty.
It is occasionally possible to solve such an equation by factorisation,
as in the case of the cubic equation
x*-2x*-2x+IÔ.
By trial and error we see that x 1 satisfies the equation. So one
root is -1 and (x+l) must be a factor of the cubic expression.
Taking out this factor, the equation becomes
The other roots of the cubic are thus those of the quadratic
The complete solution of the cubic then gives three roots,
a=-l, a? = J(3+V5) and a? = j(3-/ 6 )-
In general, however, analytical methods of solving cubic and

higher degree equations are difficult to devise and will not be con-
sidered here. Instead we can indicate a graphical method of
solving,
not only a polynomial, but also any given equation. By its nature,
the method can only produce approximate results, the approxi-
mation being closer the more accurately the graphs are drawn. The
following examples suffice to illustrate the method.
Ex. 1. - 2a; + l-0.

#3 -2.*; 2
The graph of the function ?/-#3 ~2x z -2x + 1 is constructed as shown

in Fig. 1 ]
,
where (for convenience) different scales for x and y are adopted.
x -2.v -2X + 1
The roots of the equation are those values of x which make y zero, and so
must be obtained from the points where the curve of the graph cuts Ox. In
this case, it is seen that the plotted curve cuts Ox in three points and the
approximate values of the three roots are read off as

x=-l, z=0-4 and z = 2-6.
These are the values, to one decimal place, of the exact roots already
found.
Ex.2. z3 -3z-3=0.
Fig. 12 shows the graph of the function y = x - 3x - 3, and the curve
3
plotted is seen to cut Ox in one point only where the value of a; is a little
greater than 2. The cubic equation has thus only one root which is
approximately equal to 2.1.
FUNCTIONS DIAGRAMMATIC REPRESENTATION
: 53
Ex. 3. A graphical method slightly different from that of the previous
examples can also be used. Taking the cubic equation of the last example,
we can write it in the form
Fig. 13 shows the graphs of the two functions ?/ =# and y -- 3x + 3, plotted

3
on the same graph paper and using the same scales. The first function
gives a well-known curve and the second a straight line. At any point
where, the curves intersect, the
values of y are equal, i.e. the corre-
sponding value of x satisfies the
given cubic equation. Fig. 13
shows that there only one point
is
of intersection and that the only

root of the equation is approxi-
mately 2-1 (as before).
Hence, to solve
any given
equation f(x)=0 approximately,
we need only plot the graph of
y=f(%) and determine where it
cuts the axis Ox. Or, we can
split the equation so that it appears in the form (f>(x)î/j(x), and

then find the x-co-ordinates of the points where the graphs of
y^=<f>(x) and y $(x) intersect. In either of these ways two things
are determined, the number of the roots of the
equation and the
approximate value of each root.*
Notice that the first piece of information
is often of use quite apart
from the second. If this

require, the
is all we
graphical method can
be generalised into a diagrammatic method of determining the
number of roots possessed by an equation of given type. For
example, any cubic equation can be written in the form
x*ax 2
+bx +c,
where a, band c are constants. The curve y ~# 3 has been plotted in
Fig. 13. The quadratic function y~ax 2 +bx +c (a ^0) is represented
by a U-shaped curve (known as the parabola, see 3.4 below), which
* Various
numerical methods of improving upon the approximate roots
obtained graphically have been devised. The best known is that of Homer ;
see Burnside and Panton, The Theory of Equations, Vol. I (8th Ed., 1918),
pp. 225 et seq.
cuts the curve either in one point or in three points.* The cubic
first
equation, therefore, has sometimes one real root and sometimes three
real roots according to the values of the coefficients.
2.9 Simultaneous equations in two variables.

From a single equation in one variable, we pass now to the next
simplest case where two equations are given connecting two variables
x and y. In general, the equations can be written
/i (*> V)
= and /2 (x, y) = 0,
where/! and/2 denote two given functional expressions. Such simul-
taneous equations are said to be solved when we have determined
a pair, or a number of pairs, of values of x and y satisfying both
equations. The simplest analytical device of solving simultaneous
equations is to obtain, from one equation, an expression for one
variable y in terms of the other variable x, and to substitute this
expression into the second equation. The result is an equation in
x only, which can be solved in the ways already indicated. The
following examples illustrate the method.
Ex. 1. a+y-3=-0 and #-3?/4 1=0.
From first equation, we obtain y =
the and, on substituting in 3x,
the second equation, we find that x = 2. The corresponding value of
y = 3-o: = l. The simultaneous equations have the unique solution
x = 2 and y 1.
The result can be generalised. The general pair of simultaneous linear
equations
a1a;+61y+c1 =0 and a 2#
has a unique solution :
"6
'
and -
Ex.2. 2z+2/-l=0 and a3 +y3 -3#t/ =0.

The first equation is linear and gives y = 1 - 2x. Substituting into the
second equation and simplifying, we obtain
This cubic equation is found, by graphical methods, to have three roots

which are approximately equal to
x =0-2, g=0-45 and x = 1-8.
* A similar result holds when a 0. Here the second curve is a straight line
representing y=bx+c which

9 again cuts the first curve in one or in three
points. (Cf. Fig. 13.)
The corresponding values of y, from y = 1 - 2x, are
y=-O6, y=0-l and y-= -2'6.
There are, therefore, three solutions of the given simultaneous equations.
A direct graphical method can be used to give the approximate

solution of any specified pair of simultaneous equations, fi(x, y) =.
and/2 ( x > y) 0- We plot the graph of each of these implicit functions,
using the same pair of axes Ox and Oy and the same scales in each
case. At any point of intersection of the curves of the graph, the
co-ordinates give values of x and y satisfying both the equations,
i.e.give a solution of the simultaneous
equations. The graph thus tells us
how many solutions there are and
gives us approximate values of x and y
for each solution.
To illustrate the graphical method, the

equations of Ex. 2 above can be used.
Fig. 14 shows the graphs of the implicit
functions
2x + y - 1 = and + y* - 3xy =- 0.
x*
The first curve is a straight line and the

second a looped curve which has been
plotted already (in 2.2 above). There
are three points of intersection, P P219
and P3 ,
which provide, by reading off
the co-ordinates, the three pairs of values FlG u

of x and y given above.
If the simultaneous equations are not given in specified form, then

the graphical method breaks down. But a diagram can still be
drawn, in many cases, to indicate the number of solutions, but not
their numerical values. For example, the solution of any linear
equation ax+by+c = Q and the equation x* +y*-3xy = Q can be
indicated by the points of intersection of some straight line (repre-
senting ax +by +c = 0, as shown in 3.3 below) and the looped curve
of Fig. 14. There is either a single solution, or three solutions, of the
simultaneous equations according to the values of a, b and c and the
position of the straight line.
It appears that, in general, two equations in two variables x and y
" "
provide a determinate solution, i.e. there is only a definite
number of pairs of values of x and y satisfying the equations. This
is a question which will be taken up at a later stage (see 11.6 below).
We can, however, consider here one objection to the method of solv-

ing equations as applied (e.g.) in economic theory. It is said that
" "
a circular argument is involved when we write two equations in
two variables x and y the first equation gives y as a function of x,
;
the second x as a function of y and these are inconsistent. This is an

argument that has no justification.
If no relation connects the variables x and y, then each can take
any value independently of the other and, if they have continuous
" "
ranges, we have a doubly infinite set of values of the pair (x, y).
If one relation is known to exist between x and y, then the number
of possible pairs (x, y) is much restricted. There is still an indefin-
"
itely large number of such pairs but the range
only singly is
infinite ". Finally, if another relation is known and added to the

first, the possibilities are still further restricted and, except in odd
cases, there is only one or a finite number of possible pairs (x, y).
In this case, x and y are said to be determined. In diagrammatic
terms, a pair of values of x and y is represented by a point P in a
plane. If no relation between x and y is given, P can move at will
over the plane. Jf one relation exists between x and y, then P can
move only in a restricted way, i.e. along the curve representing the
relation. If two relations between x and y are given, the possible
positions of P are confined to the points of intersection of two curves

and they are, in general, only of finite number.
To take a concrete " applied " example if the output x of an :
industry is known to be related to the price p of the commodity by

a demand relation (giving the amounts the market would take at
various prices), then the possible values of x and p are limited but
still indefinitely large in number. But, if we also know that the out-
put isrelated to the price by a supply relation (giving the amounts
the industry would offer at various prices), the values of x and p are
much more limited and there is, in general, only one or a few pairs
of values possible. Here we say that output and price are deter-
mined. The two relations of demand and supply are independent
of each other, and neither alone determines both output and price.
There is no circular argument.
EXAMPLES II
Functions and qrapJis
1. A railway A ROD with sides AH =2

ticket is in the form of a roc tangle
inches and BC 3 inches. The corner D by cutting along the
is clipped off
lino EF, where E lies on CD and F on AD so that GE ~\AF. Denoting CK
by x inches, find the area of the clipped ticket as a function of a? and represent
it graphically.
2. A variable point P is taken on a semicircle drawn on the diameter
AB4 inches. PN is perpendicular to AB and N is x inches along A B from A .
Find the length of PN in terms of x. Deduce the area of the triangle APN as
a function of #, and plot a graph of the function for ^x ^4.
3. A beam
has a rectangular cross-section of sides x and (#4-2) feet and
its length is y feet. Find y as a function of x if the volume of the beam is
100 cubic, feet, and represent graphically.
The appended table shows a simple way of obtain-
4. x 1
irig values of the quadratic function y 2x z - 2x 4- I .
Extend the table to give the values of the function for 2x 2 2 8

- 2x -2 -4
integral and half-integral values of x from 3 to 4- 3,
and plot a graph of the function. Find an explicit ex- 4-1 1 1
pressioii for x as a function of y and deduce that two

values ofx correspond to each value of y >\. Examine y 1
the cases where y ^ \ r.
5. I'lot a graph of the function y 5x - 2x z Show that y has a greatest .
but no least value. Locate the greatest value as accurately as you can from
the graph. Between what values of x is y positive?
6. Graph y2x - for positive values of a:, and show how the graph
1 4 -
x i
can be obtained by the addition of those of y=2x - 1 and y =- Establish .
that y ie always positive.

x
2x
7. Plot a graph of y = for positive values of x and show how the y
X 4~ A
graph for negative values of x can bo deduced. What are the greatest and
least values of ?/? Explain what happens to the value of y as x is increased
indefinitely.
8. A single man earns x a year. His taxable income, obtained by
deducting one-fifth of the total income plus his personal allowance of 100,
is taxed at 4s. 6d. in the except that the first 175 is taxed at only half the
,
full rate. If the tax is equivalent to y shillings in the of original income,

show graphically the variation of y for incomes between 100 and 1000.
New regulations are introduced whereby the first 135 of taxable income
is taxed at one-third of the standard rate and the rest at the full rate. Plot a
new graph (on similar lines) and compare the new values of y with the old
values.
2x 4- 1
Find the function inverse to y = - -
9. and show that it is single-valued.
x 1
Represent graphically, and give some account of the behaviour of the graph
in the neighbourhood of x 1 and of y = 2.
10. A chord of a circle of radius 3 inches has length x inches and is distant
y inches from the centre. Find a relation between x and y. Express y as an
explicit function of x and x as an explicit function of y. Are these functions
single-valued or not ?
11. Obtain an explicit expression for the function inverse to y=x + x and
show that it is not single-valued.
12. 13y selecting a sufficiently large number of values of x and finding the
corresponding values of y, indicate that the function y 3cc -j- 3# -f x - I is
3 2
mono tonic. Does y increase or decrease as x increases ? Illustrate graphically.

13. Show that x*+bx+c can be written in the form (x + %b) 2 - J(6 2 -4c),
and deduce that y =x z -f bx -f c has a least value when x = - \b but no greatest
value. Hence indicate the general shape of the curve y=x 2 +bx+c. What
is the condition that the curve lies entirely above the axis of x ? Consider the
curve y - x* -f bx + c in a similar way.
14. Combine the results of the previous example to indicate the nature of
tho curves represented by the general quadratic equation y ax z +bx + c.
Show that tho curves are of the same type and can be divided into two groups,
one group consisting of curves with a lowest but 110 highest point, and the
other group of curves with a highest but no lowest point.
15. A ball, thrown vertically into the air with a given velocity v, reaches
a height x after time t. It is known that x is the difference between two
independent factors. The first is the offect of the initial velocity v and is
measured by v times t. The second is the effect of gravity and is represented
z
by a constant %g times t Express x as a function of t involving the two
.
parameters v and g. Taking g = 32, find the height of the ball after 4 seconds
if the initial velocity is 75 feet per second.
The volume of a rectangular box of sides x, (#4-1) and (x -f 2) inches

16.
is Of what type is y as a function of x ? Generalise to give the
y cubic inches.
volume when the smallest side is x inches and tho other sides are respectively
a inches and b inches longer. Arrange tho function to show its type and its
dependence on the parameters a and b.
17. Write down the general symbolic form for y as a cubic polynomial
function of x, and show that functions such as
and y = (x + 1 ) (x + 2) (x + 3) are all included.
s
y = l -re , y=x(l +#*)
8
yx 9
18. How many parameters are needed to describe the general cubic
functional relation between x and yl Show that the general expression
includes such simple functions as
1 +l _ . 1
and
*=? y= (sn) *=*-*
19. Show that the function type represented by y = ex + d
-=, where a, 6, c
and d are parameters, is a particular case of the general implicit quadratic

function. Deduce that this type is such that both inverse functions are
single- valued and of the same type.
20. Find the values denoted by /( - 1), /( - J), /(O), /(J) and /(I), when
(a) /() = 2x + 3, (b) J(x)

Zx ~H 1
=
i- and
(c) f(x)
= 2x* - 4cc + 3 +
x Explain - .
why
f(
- i) cannot be defined in the second case and/(0) in the last case.
,8 I
Q/y. O
21. If f(x) =- , express /(2a) in terms of a.
22. Show that, for the function f(x) = l - 2x + 3# , wo have /( -a)=f(a) 2 4
for all values of a. Indicate that this is true of any polynomial containing
only even powers of x.
Further, if/(#) =2x + 5x* -x*, show that/( -a) = -/(a) for any value of a.
For what general polynomial function of x is this relation valid?
23. Find the value of /(-) in terms of a, when f(x)~x* \ 3x-2 and
when/(#) =
â/
(%x--\}(x 2)
-
f- -~s~~\~ X T~
A 'JC
*n ^le ^ a^ er flas^f
/1\
show that /( - } =/(a) for any
\Cl/
value of a.
Find the value of f(a +h) in terms of a and ft when /(x) =- f(x) x*,
24.
x 1
-
/(#)=! -2# HI id /(a?) 2x
ratio'
2
5 l
-+
.
^
Deduce an expression for the " incrementary
,
f(a+h)-f(a)
h
in terms of a and h in each case.
25. Find y as an explicit function of I when y x z +3.r -2 where x~\ +t

and when y 1 ~x 2 where x ~~ - .
^wf ~i~ JL
26. Obtain a relation between x and ?/ when it is given that x=t z and
2/
~ 2i. Plot a graph of the relation by giving various values to t.
27. The variables x and y are both given functions of t and, by eliminating
t9 an
implicit function relating x and y is found. Obtain the relation in the
following cases :
(a) *-, y =
j^; (6) a
28. A
wooden letter tray is in the form of an open rectangular box, of which
the base sides are respectively twice and three times the depth. Express the
volume* of the tray and the area of wood used as functions of the variable
depth of the tray. Deduce a relation between the volume and area of wood.
What is the volume if 4 square feet of wood are used in the construction of the
tray?
The solution of equations
By writing factors, find the roots of each of the equations x - 3x + 2 =0,

2
29.
+ 5 + 2~0, 6x 2 +x-2=:Q, x 9 + 6z 2 + ILc + 6 ^0 and 2# 3 -5x* +a; + 2 =0.
2a 2
In the first three cases, check your results by means of the formula for the
general solution of a quadratic equation.
30. From the general formula, solve the quadratic equations x 2 + 2x - 3 = 0,
^0 and 3x* -Ix -3 =0.
81. A beam has a rectangular cross -section of sides x and (# -fa) feet and a
volume of b cubic If the length of the beam is c feet, show that algebra
feet.
"
indicates that there are two possible values of #, one of which does not "fit
the problem. Find an expression for the solution in terms of the parameters
o, 6 and c.
32. Show that each of the cubic equations a 3 + 5# a + 4x -4 =0, a 8 + 1 =0,

and 2# 8 +x 2 - 9x - 2 =0 has one integral root. Hence, complete the solution
of each equation.
33. By graphical methods, find the number and the approximate values of
the roots of each of the equations :
(a) xs + x 2 -2x-l =0, (6) x s - 2x - 5 =0, (c) x* - 3x 2 - 1 =0.
34. Verify the approximate solutions of the previous example by plotting

graphs of
y - x and y + x-2, y ^x* and y=2x =x and =

- 2
(a) ---x* (b) 5, (c) y y .
:
x ~~ o
35. Show graphically that the equation # -7# + 7

3
has two positive
roots almost equal in value, and evaluate them approximately.
36. Find the approximate values of the roots of #* - I2x -f 7 and of the
- I2x* + I2x - 3 -=0.
negative root of x*
37. By considering the forms of the general curves representing
y a &* + a 2x* + a&c + a 4 and y ~- -
x
where the coefficients are constants, show that the general quartie equation
must have either four, or two, or no real roots.
38. By algebraic methods, solve each of the following pairs of simultaneous

- 2x-y + 3=0 and xy=2;
equations: 3oH6</~ll and 6x 3Qy + 41 -0 ;
x - 2y + 1 -0 and x 2 f y z ~9.
39. Find the solution of the general pair of linear equations

a& + b t y G! = and ax 4- b zy + c z 0.
-f-
Show that the solution fails only when =~ . Illustrate this case of failure
ct z 2
by attempting to solve the equations 2x - y 3 { and x - 2y + 5 = 0. What

is the graphical reason for the case of failure ?
40. How many solutions are to be expected in each of the types of simul-
taneous equations (a) xy 1 and ax by c = 0, (b) xy = l and y ~ax 2 +bx+cl
: -I- -!
Find, by graphical methods, the approximate solutions of

xy I and y = 5x - 2x
2
.
41. The prices per bushel of wheat and rye are p and p 2 respectively. The t
market demand for wheat is given by x l 4 - Wp t +lp^ and for rye by

x 2 3 + Ipi - 5y> 2 The supply oif wheat is related to the prices by the relation
'
x l ~ 1 +pi - p z and the supply of rye by the relation x 2 = - 27 ~p l + 2p 2 Find .
the pair of prices which equate demand and supply both for wheat and
for rye.
42. In the problem of the previous example, a tax of ^ per bushel is im-
posed on whoat producers and a tax of t z per bushel on rye producers. Find
the new prices for the equation of demand and supply (substituting p^ - 1 for
p l and p z - 1 2 for p z in the supply relations above). Show that the wheat price
increasesby an amount(9 a ^
^) and the
ryo price by an amount ( I4t2 - 3^). ^
Deduce that
(a) a tax on wheat alone reduces both prices ;
(b) a tax on rye alone increases both prices, the increase in the rye price
being greater than the tax.
(See Hotelling: EdgewortWs Taxation Paradox, Journal of Political
Economy, 1932, especially pp. 602-3.)
CHAPTER III
ELEMENTARY ANALYTICAL GEOMETRY

3.1 Introduction.
WE propose, in the present chapter, to follow up a line of develop-

ment already indicated (2.7 above) and examine, in some detail, the
relation between function types and curve classes. Our investi-
gations belong to what is termed "analytical geometry" and our
method is to take certain well -known classes of curves in turn, deter-
mining what type of function corresponds to each class. We have
then an analytical method of treating the geometrical properties of
the curves and a diagrammatic method of illustrating analytical
properties of the functions.
Three simple, but essential, formulae must iirst be established.
Two fixed points have co-ordinates P(xl9 y^) and Q(x 2 y 2 ) referred ,
to rectangular axes fixed in a plane Oxy. Then :
The distance between P and Q is N/(# 2 -# 2

-f
~ 2
(1) 3 ) (y 2 ?/i) -
(2) The mid-point between P and
Q has co-ordinates
+2*2 2/i
(3) The point which divides PQ

in the ratio p q has co-ordinates
:
fWi
\ p +q p +q I
The proofs of these formulae, in-
N M
volving quite simple geometrical FIG. 15.
notions, make use of the notation
of Fig. 15 and apply no matter what positions the fixed points P
and Q take up in the plane.
(1) Pythagoras' Theorem, applied to the right-angled triangle
PQN, gives
PQ* =PK* +KQ* = (x 2 - xtf + (y, - &),
since PK ^LM = OM-OL^x Q
- xlt
and
PQ = V (z - ^) 2 + (y a - &)*.
'
Hence, 2 Q.B.D.
(2) If E
the mid-point of PQ, then a well-known property of
is
parallel lines tells us that N

is the mid-point of LM, i.e. =NM. LN
If the co-ordinates of Eare (a?, i/), then
LN=ON-OL=X-XI and
So x-xl =x 2
i.e. 2x xL
-y
Hence, a? = Xx +X%
-
^, and
J "I
similarly
I 2^1
y--^^
+2/2
^ T, ^
4 Q.T^.D.
(3) If jR divides PQ in the given ratio ^ :

g, then
and, if the co-ordinates of B are (x, y) so that

LN = x xt and
x - a?!
then =. >
# 2 -x q
Simplifying, we obtain x ~
and similarly
J y=
P+2 Q.E.D.
Formula a particular case of formula (3) with p~q~[. It
(2) is
is interesting to notice the connection between these results and the
notion of an average. The mid-point between P and Q is a kind of

simple average point and each of its co-ordinates is the simple
average of the corresponding co-ordinates of P and Q. In the same
way, the point E dividing PQ in the ratio p q is in the nature of a :
weighted average point between P and Q. Each of its co-ordinates

is the weighted average of the corresponding co-ordinates of P and
Q P'& co-ordinates being weighted with q and Q's with p. In fact,

9
the larger is q relative to p, the nearer is E to P, and the larger is the

weight of P's co-ordinates in those of J?.
ELEMENTARY ANALYTICAL GEOMETRY 63
The co-ordinates of the origin O are (0, 0). If P (x, y) is any fixed
point, the following are particular cases of the above results :
(1) The distance OP^Jx*+y*.

(2) The mid-point of OP has co-ordinates (\x, \y).
(3) The point dividing OP in the ratio p :

q has co-ordinates
/ p* py
p p
3.2 The gradient of a straight line.

The simplest curve class is the class of straight lines.* A straight
line is fixed iftwo points on it are specified. We can, therefore,
speak of the straight line PQ, P and Q representing two fixed points
on the line, but we must always remember that the straight line
extends indefinitely in both directions.
y
\J
M M \x
FIG. 16.
A most important property of a given straight line is its direction

relative to a pair of fixed axes Ox and Oy, defined by means of the
angle a (e.g. in degrees) that the line makes with the positive
direction of the horizontal axis Ox. If the angle a is acute, the line
slopes upwards from left to right if a is obtuse, the line slopes down-
;
wards from left to right. The angular measure of the direction of

a line, however, is not metrical. It is, therefore, inconvenient
for most analytical purposes and a metrical indicator of direction is
needed. Such an indicator is readily provided by making precise
* A straight line is a locus of" points in"a piano and thus a curve. It can,
however, be regarded as the limiting case of a curve which does not
" "
curve at all.
"
our everyday notion of a gradient ". We say, for example, that a
railway track has a gradient of 1 in 10 if it rises one yard vertically
for every ten yards horizontally. With the notation of Fig. 16,
the straight line PQ "rises" a distance NQ over a "horizontal"
NO
distance PN. The ratio ^rc measures the amount of the rise per
r
PN
unit horizontal distance and can be called the gradient of the line.
Hence :
DEFINITION : The gradient of the straight line PQ referred to the

axis Ox is the ratio of to PN.
NQ
Here P and Q are any two points on the straight line and is N
the point where the parallel to Ox through P
cuts the parallel to
Oy through Q.*
It is clearly essential that the gradient should depend only on the
direction of the straight line
and not on the positions of
P and Q selected to
the points
express it.This point needs
investigation. Let P' and Q'
be any other pair of points
on the given line. Then, from
Fig. 17, PQN and P'Q'N' are
similar triangles, and a well-
known geometrical property of
M '
M such triangles gives
FIG. 17. NQ N^Q' '
PN P'N'
The ratio is thus the same for all selections of P and Q. Exactly the
same result follows if P' and Q' are any points on a line parallel to
PQ. Hence, the gradient of a line depends only on its direction and
all parallel lines have the same gradient.
It must be stressed that the lengths NQ and PN of the gradient

ratio must be given signs according to the usual conventions. It is
easily seen that the gradient of a line sloping upwards from left to
* The " "

gradient is sometimes referred to as the slope of the line. If a is
the angle the line makes with the axis Ox, then the gradient is the trigono-
metric tangent of a, i.e. gradient tan a.
right is positive and that the gradient of a downward sloping line
is negative. In the first diagram of Fig. 16, where the line slopes
upwards, NQ PN and are both positive and the gradient is positive.
In the second diagram, where the line slopes downwards, NQ is
negative arid PN positive, i.e.the gradient is negative.

Finally, wê can show how the magnitude of the gradient indicates
the steepness of the line. Since all parallel lines have the same
gradient, we can draw lines through
O for convenience and 18 shows a y>
Fig.
number of such lines sloping upwards
with increasing steepness. The points
Pl9 P<2 P 3 ... are taken on the lines
, ,
with the same abscissa OM. Then
OM < ~OM < dM

i.e. the gradient of the line increases
as the line becomes steeper, fn the M x
same way, a line slopes downwards,
if
Fio. 18.
its gradient is negative but increases
numerically as the line becomes steeper. The following results are

thus established :
(1) The gradient of a line is a metrical indicator of its direction, all
parallel lines having the same gradient.

(2) A line sloping upwards from left to right has a positive gradient,
and a downward sloping line a negative gradient.
(3) The steeper the line, the larger is the numerical value of its
gradient.
Two
limiting cases arise when a given line is parallel to one or
other of the axes of reference. In Fig. 18, as the line becomes less
steep and tends to coincide with Ox, the gradient decreases and tends
to zero. Again, as the line becomes more steep and tends to coincide
with Oy, the gradient increases indefinitely.* Hence, the gradient of
a line parallel to Ox is zero and the gradient of a line parallel to Oy is
indefinitely large.
The angle between two given straight lines is independent of the
* In the terminology of the following chapter, the gradient '
tends to
infinity ", and the gradient of a line parallel to Oy is " infinite ".
M.A.
positions of the lines as long as the gradients are fixed. In particular,
the conditions that two lines are parallel or perpendicular involve
only the gradients of the lines. If m w
and 2 are the gradients, then
the lines are parallel if l m ~m
2 and perpendicular if m^xm^ -1.
The first of these results needs no further proof we have seen that ;
parallel lines have the same gradient. The proof of the second result
proceeds :
In Fig. 19, P and Q are two points with the same abscissa OM on
two perpendicular lines drawn, for convenience, through the origin.
The triangles OMP and QM are similar and, since P and Q are M M
necessarily lengths of opposite sign, the ratio
of M M
P to O equals minus the ratio of O M
to MQ. Hence,
MP x MQ = ,
or x wi 2
.
~~ *
^ ,, TTV? 1 wii
OM. OM
The condition can also be written
1 1
wij
= or m = 2 9
i.e.perpendicular lines are such that the

gradient of one is minus the reciprocal of
the gradient of the other.
FIG. 19.
3.3 The equation of a straight line.

A straight line is fixed in position if two things, e.g.
two points on
the line or one point on the line and the gradient, are known about
it. Our problem now is to find the equation of the line referred
to some selected axes of reference, i.e. the relation between the
co-ordinates of a variable point on the line.
A
line passes through the fixed point with co-ordinates (xlt P yj
and its gradient is known to be m. Then, from Fig. 16,
__
~~
m PN~OM-OL
x-xi
for any position of a variable point Q (x, y) on the line. Hence,
The equation of the straight line with gradient m passing through

the fixed point (a^, j/J is
y-yl =m(x-xl )
............................ (1)
Next, suppose that the line witli equation (1) also passes through
a second fixed point with co-ordinates (x 2 y 2 ). These co-ordinates >
must, therefore, satisfy the equation, i.e. equation (1) holds when
x2 is substituted for x and y 2 for y :
v-u^m(x~x)
& y & j. \ a i/ or m =?*^.
X X ,.,
2
Hence,
The gradient of the straight line passing through the two fixed
points (#!, t/i) and (x 29 y 2 ) is (y 2 ~2/i)/(^2~î)j an(i e equation of ^
the line is
The equation (2) can also be shown to be algebraically equivalent to

the alternative form
Attention drawn to the values w, XL and x 2 appearing in the

is
equation (1), and to the values xl9 x 2 yl and y 2 in the equation (2).
,
For any given line, these values must be fixed but, by varying the
values, different lines are obtained. The values are, in fact, para-
meters and the equation (1) or (2) is a function type representing the
whole Various sub-classes of the complete
class of straight lines.
class of lines are obtained by varying the parameters in defined
ways. For example, if the point (x1} yx ) is kept fixed and the para-
meter ra varied, the equation (1) represents a set of lines of varying
gradient passing through a fixed point, a set technically known
all
" "
as a pencil of lines. Again, if m
is fixed and the point (xlt yj
varied, we obtain a set of parallel lines with a given gradient m.

The equation of the straight line, in form (1) or (2), is seen to be
" "
of the linear type, i.e. x and y appear only to the first degree.
The converse is also true and any linear relation between x and y
represents a straight line. The relation
ax+by +c~0 .............................. (3)
can be written*
a
* This arrangement of the relation holds, provided that 6 4=0. When 6 =

we have a straight line parallel to Oy (see below).
On comparison with the equation (1), we see that (3) must represent
-
a straight line of gradient
-j and passing through f -,
Oj. So,
The equation ax +by -|-c = always represents a straight line and

the equation of any line can be written in this form.
There is, therefore, a perfect correspondence between the class of

straight lines and the linear functional relation.
We have shown, incidentally, how the gradient of a straight line
can be determined from the equation of the line. The gradient of
the line with equation (3) is given by -T, i.e. the gradient is minus
the ratio of the coefficients of x and y when the equation is written

with both terms on the same side. It follows that
*%# + b^y + q and a%x + b 2y + c 2
represent parallel lines if -- 1 = - -, i.e. if
and perpendicular lines if f

-~ f
- =- 1, i.e. if
J -^ J
oâ +&J&2 0-
For completeness, we must examine the cases where a line is

parallel to one or other of the axes. It is easily seen that the equation
~
of a line parallel to Ox is of the form y constant, and the equation of a
line parallel to Oy is of the form x= constant. This follows since the
y-co-ordinate of any point on a line parallel to Ox has the same

value as the j/-co-ordinate of any other point on the line, and simi-
larly for the ^-co-ordinates when the line is parallel to Oy. The
results can also be derived from the equation (1). A line parallel to
Ox has gradient m equal to zero and so its equation is
y-y^ or y~y\ constant.
A line parallel to Oy has an indefinitely large gradient. The equation

(1) can be written in the form
*-ai=-(y-yi)
and, when the value of m is increased indefinitely, the equation tends

to assume the form
x - #! = or x = Xj = constant,
which the equation of a line parallel to Oy.
is Notice, also, that the
equation (3) represents a line parallel to Ox when a = and a line
parallel to Oy when 6 0. The axes themselves are straight lines

with equations y = Q (the #-axis) and x = (the y-axis).
Finally, the equation of a straight line passing through the origin
with gradient m
has equation y=mx. So, the equation (3) represents
a line through the origin if c = Q, when it appears in the form
ax
3.4 The parabola.

A parabola is a curve defined as the locus of a point which is
equidistant from a given point S and a given line d in a plane. The
point S is called the focus, the line d is called the directrix and the
d
FIG. 20.
line KS perpendicular from S to d is called the axis of the parabola.

The distance of S from
d(SK) denoted by 2a, where a is some given
is
constant. The general form oft/he parabola is shown in Fig. 20. The
curve is symmetrical about its axis KS and must pass through the
mid-point of KS, called the vertex V of the parabola.
A whole set of different parabolas are obtained by varying the
positions of S and d the general shapes of all parabolas are similar
;
but the particular position and shape of a parabola depend on the

location of S and d. The most important thing distinguishing one
parabola from another is the distance of S from d, i.e. the value of

the parameter a. If a is small, then S is close to d and the parabola
iselongated as shown in the first curve of Fig. 20. If a is large, then
S is distant from d and the parabola is flat as shown in the second
curve of Fig. 20.

The equation of a given parabola can be obtained when the axes
of reference are selected. We propose, here, to select the axes in such
a way that the axis of the parabola is vertical, i.e. parallel to Oy.
The position of the origin is
still at choice and it is found
that the simplest result is
obtained when the origin is
taken at the vertex of the

parabola. In this case, the
equation of the parabola is
derived as follows.
Take axes of reference as
shown in Fig. 21 and suppose
FIQ. 21.
that the focus S is above the
vertex of the parabola. The
point S has co-ordinates (0, a) and the line d is parallel to Ox and a
distance a below it. If P is any point (x, y) on the parabola, then
the definition of the curve gives SPMP. But
SP = Jx 2
+(y-a)* and MP=NP +MN=y -fa.
2 2 2
So, a;
+(2/-a) -(2/+a)
and this, on reduction, gives the equation of the curve in the form
y=x4<x
2
.
On the other hand, if S below the vertex, the co-ordinates of S

is
are (0,
- a) and d is parallel to Ox and a distance a above it. An
argument exactly similar to that given above then shows that the
equation of the parabola is
The following result is thus established :
The equation of a parabola with axis vertical and vertex at the

origin of co-ordinates is y=âx* where a is some constant,
The sign of the constant a indicates the position of the parabola.

If a is positive, the axis of the parabola points upwards (as in Fig.
20) ;
if a is negative, the axis of the parabola points downwards.
The numerical value of a is the reciprocal of twice the distance of S
from d. If a is large, then S is near d and the parabola is elongated ;
if a is small, S is distant from d and the parabola is flat.
Suppose, now, that the axes are selected so that the vertex of
the parabola has co-ordinates (, 77), the axis of the parabola being
parallel to Oy as before. Then, if S lies above the vertex, the co-
ordinates of S are (, 77 + a) and the directrix d is parallel to Ox and
a distance (77
- a) above it. If P is
any point (x, y) on the parabola,
the condition that P is equidistant from S and d gives
On squaring each side of this equation and expanding, we obtain

2 2 - 2a(y 2
a ~(i/ -y) 2 2
(x -f) +(y -?7) -77) -f -f 2a(y -77) -fa ,
Similarly, if S lies below the vertex, the equation is
So,
The equation of a parabola with axis vertical and vertex at the

2
point (, 77)
is (y-rj)=a(x -) .
The constant a is to be interpreted, in sign and magnitude, as

before. The only difference between this general form of the equation
of a vertical parabola and the previous simpler form lies in the sub-
stitution of (x - ) for x and (y - 77) for y. When 77
= 0, the vertex
is at the origin and the general case reduces to the special case.
The most important feature of the general equation of a vertical
parabola is that it gives y as a quadratic function of x. The converse
of this is also true. The general quadratic function
y=ax 2 +bx +c
can be written (by completing the square in x) in the form
b \2 6 2 - 4ac
k
2
i.e
Comparing this with the equation of the parabola above, we deduce :
The equation yax 2 +bx +c, for any values of the constants a, b
and c(a7^0), represents a parabola with axis vertical.
The constant a is to be interpreted exactly as before and the vertex

of the parabola is at the point with co-ordinates
b b2
There is, therefore, a correspondence between the quadratic function

type and the class of parabolas with axis vertical.*
3.5 The rectangular hyperbola.

A rectangular hyperbola is a curve which can be defined as the
locus of a point the product of whose distances from two fixed per-
pendicular lines is a positive constant a 2 The fixed lines are called
.
the asymptotes and their point of

intersection the centre of the rect-
angular hyperbola. A convention

is required regulating the signs
to be allotted to the distances
of a point on the curve from its
asymptotes. Here, we draw one

M x
asymptote horizontally and the
other vertically, taking distances
above the first asymptote and to
the right of the second as positive.
The general form of the curve is
FlG . 22. shown in Fig. 22. There is one
portion of the curve in the positive
(orN.E.) quadrant of the plane, and, as a point P moves along this
portion to the right, the distance PM
from the horizontal asymptote
* The
quadratic function and its graphical representation have been con-
sidered in the previous chapter (e.g. Ex. 1 of 2.2). From our point of view,
parabolas are important because they represent such a simple and useful
function typo as the quadratic.
decreases in the same proportion as the distance PN from the

vertical The area of the rectangle ONPM
asymptote increases.
remains constant (a 2 ), and the curve approaches but never cuts the
asymptotes. There is a second portion of the curve, exactly similar
in shape, in the S.W. quadrant.
A whole set of different rectangular hyperbolas is obtained by
varying the positions of the asymptotes and the value of the constant
a2 The general shape of all curves is the same but the particular
.
shape of one curve, when the

y*
asymptotes are fixed, is de-
4-
termined by the value of
a2 .
Fig. 23 (where, for con- xy*
venience, only the portions 3--
of the curves in the positive
quadrant are drawn) shows
a set of different rectangular
hyperbolas corresponding to
the values of a 2 indicated.
Iri obtaining the equation
of a rectangular hyperbola
here, we take the co-ordinate
axes parallel to the asymp-
totes of the curve, only the
position of the origin being at choice. The simplest case arises

when the origin is taken at the centre of the curve, i.e. when the axes
are the asymptotes. If P is any point (x, y) on the rectangular
hyperbola in this case, then in Fig. 22
But, =:OM=x and MPÔN^y.

So,
The equation of a rectangular hyperbola referred to its asymptotes
xy = a where a2
2
as axes is ,
is a constant.
a2 a2
The equation can be written y~-~ or x , showing that y is a
x y
single-valued decreasing function of x, and conversely.
In the general case, where the asymptotes are parallel to the axes
and the centre of the curve is at the point (f 77), the distances of ,
any point (x, y) on the rectangular hyperbola from the asymptotes

are seen to be (x - ) and (y - 77) respectively. The equation of the
curve is then provided by the condition that
The equation of a rectangular hyperbola with asymptotes parallel

to the axes and centre at the point (, 77) is (x - g) (y - 77) = a 2 .
This general form only different from the previous special case
is
in that (x - g)
replaces x and (y - 77) replaces y. When 77
= 0, the
centre is at the origin and the special form is obtained.
The general equation of the rectangular hyperbola in the form just
obtained found to give y as the ratio of two linear expressions
is
in x, y being a single-valued decreasing function of x. Similar

remarks apply to # as a function of y. The converse is also true.
The function type
a{x -f 6X
y
can be written in the form
2 1
i.e. xy
d-i
-x -\
-
y - ^ 0.
a2 a% a2
On factorising, this reduces to the form
Comparing this with the above equation of the rectangular hyperbola,

we deduce :
The equation y~ - -
~, for any values of the constants
represents a rectangular hyperbola with asymptotes parallel to the

co-ordinate axes.
The centre of the curve is at the point

r (
-
,
1 . The class of
\ a8 <V
rectangular hyperbolas, with asymptotes parallel to the axes, is in
correspondence with a simple algebraic function type.*
3.6 The circle.
A circle is the locus of a point which moves at a constant distance

from a fixed point. The fixed point is the centre and the constant
distance the radius of the circle. The circle is a curve discussed in
allelementary geometries and needs no description here. It can be
stressed, however, that there is a whole class of different circles
obtained by varying the position of the centre and the length of the
radius.
A circle is symmetrical about any diameter. The choice of the
origin, therefore, is more important than the direction of the axes
in obtaining a simple equation of a given circle. If the origin is
taken at the centre of the circle, the equation of the circle is found
(by the method described in 2.3) to be
x*+y*=a*,
where a is the radius. This
the simplest equation possible for the
is
circle. Similarly, if the centre has co-ordinates (, 77) referred to the

selected axes, the equation of the circle is
This is tjie general equation of a circle.
If the squares in this latter equation are expanded, the equation

of a circle is seen to be a quadratic relation between x and y of such
form that the coefficients of x 2 and y 2 are equal and the term in xy
absent. The converse of this result is also true. By completing the
squares in both x and y, the quadratic relation
x 2 -f y 2 +ax +by +c -0
can be written in the form
+(y +\b)* = l(a* +b* 4c).
2 -
(x 4-la)
* Our expression of this result is not quite complete. The function typo
~
only represents a rectangular hyperbola of the form considered if (a a &i #1^2)
is positive. If that quantity is negative, the curve represented by the function
type is of the form (x f ) (y -rj) = a
- - 2 This is a rectangular
.
hyperbola lying
in opposite quadrants of the plane to that shown in Fig. 22.
The function type represented by a rectangular hyperbola (i.e. the ratio of
linear expressions) will be considered particularly in the following chapter (see
Ex. T of 4.2).
The equation thus represents a circle with centre at the point with
co-ordinates (
- |#,
- ^b) and with radius |>/a 2 +6 2 - 4c.
This statement, however, needs some qualification. It is only
when c<J(a 2 +& 2 ) that the radius has a definite and positive value.
(a + 6 ), the radius is zero and the circle reduces to a point.
2 2
If c
2
If c> J(a 2 -f-6 ), the square of the radius is a negative quantity,
i.e. no real radius or circle exists at all. In this last case, the locus
represented by the equation contains no points whatever. Hence,
If c<(a 2
+6 2 ), the equation x 2 +y 2 -{-ax +by +c represents a
definite circle with centre at the point (
-
|a,
- Ib) and with radius
3.7 Curve classes and curve systems.

We can now pass to a number of more general considerations in
the field of analytical geometry. The curves we have discussed
belong to particular curve classes, the class of straight lines, of
parabolas, of rectangular hyperbolas or of circles. In each case, an
equation has been derived to represent the whole curve class and the
equation a particular function type involving certain parameters.

is
Analytical geometry, in fact, is simply the study of curve classes in

relation to the corresponding function types, and, if a definite curve
of a certain class is given, we need only substitute the appropriate
values of the parameters in the corresponding equation or function
type. Our results can be summarised :
Curve class General equation
Straight lines
Parabolas (axis vertical)
......
.... ax + by+c=Q
yax 2
+ bx+c
Rectangular hyperbolas (asymptotes parallel al x+b
to the axes)
Circles x2 -t-i/
2
-{-ax + by +c =0
A number of useful comparisons can be made. The general

equation of a straight line or rectangular hyperbola gives y as a
single-valued monotonic function of x, and conversely. The general
equation of a parabola gives y as a single-valued (but not monotonic)
function of x while x is a double-valued function of y. The general
equation of acircle gives y as a double-valued function of #, and con-
versely. These results correspond to obvious geometrical properties

of the curves themselves. For example, any parabola is cut by a
line parallel to its axis in only a single point, whereas it is cut by
lines perpendicular to its axis either in points. no point or in two
The number of parameters in the equations of the different curve
classes is significant. The general equation of a straight line con-
tainstwo parameters, the ratios of a, b and c.* This corresponds to
the fact that two things (e.g. two points on the line) are needed
to define the position of the line. On the other hand, the general
equations of the parabola (axis vertical), the rectangular hyperbola
(axis parallel to the asymptotes) and the circle are function types
containing three parameters. Three things are required to fix the
position of one of these curves a definite vertical parabola can be
;
drawn, for example, to pass through three points known to lie on the
curve.
Special sub-classes of a complete curve class can also be con-

sidered. In a sub-class, the number of parameters in the corre-
sponding function type or equation is less than the number in the
complete class, and the most useful sub-classes are those defined
by a single parameter. A sub-class of curves of this latter kind

is
described as a system of curves. We can define, therefore, various

systems of straight lines, parabolas, rectangular hyperbolas or
each system corresponding to a functional equation with one
circles,
parameter. The following examples are instances of some of the

many ways in which systems of curves can be specified.
Ex. 1. The two most important systems of straight lines have been
referred to already (3.3 above). They are the system (or pencil) of straight
lines a
through given point and the system of parallel straight lines with
a given gradient.
*In the linear equation ax + by + c = 0, both a arid 6 cannot be zero. If
6 ^ 0, we can write y = ocx + where a = - and fl = - = ; if a =fc 0, we can write

/?
x a?y+p' whore <x' = --b and 0'

= c
. There are only two independent
d ci
parameters, either a and j3

or <x' and /?'.
The same reasoning shows that the
equation of the rectangular hyperbola y =~ r^ contains only three inde-
Qi^K 4* 02
pendent parameters, the ratios of a 19 b lt a a and 6 2 have only to divide . We
numerator and denominator by # 2 (which cannot be zero) to see this.
Ex. The system of parabolas with axis vertical and vertex at the
2.
by the equation y ax with one parameter a. A less

2
origin is defined
obvious system of parabolas is given by the equation
where h and k are fixed positive numbers (h> Jk) and a is a positive para-
meter. Any parabola of the system has vertex at a point {h (a - 1), -k}
lying on a fixed line parallel to Ox, and its focus is a distance |a from the
2
directrix. As the parameter a

increases, the vertex moves to the
right and the parabola becomes
flatter. A parabola of the system
cuts Ox where
2/=0 and xh(a -1)
It cuts Ox, therefore, at only one
point, at the end of the range of

values of x considered, i.e. at the
xa(h- >Jk) - h. The
point where
curve also cuts Oy at one point
/ 2
1\
where # = and yh 2
[l--j
\ QS
-k.
As a increases, the parabola cuts

the axes at points farther and
farther from the origin. Notice
that the range of x is so chosen,
FIG. 24. in each case, that only the part
of the parabola in the positive
quadrant is taken. The system of parabolas is illustrated in Fig. 24,
where certain curves of the system obtained when h = 10 and k = 4 are
drawn.
Ex. 3. The system of rectangular hyperbolas with asymptotes as axes

is defined by the equation xyâ? with one parameter a. This system is
illustrated in Fig. 23.
A
rather different system of rectangular hyperbolas is defined by
where h and k are fixed positive numbers and a is a parameter. Here

(x+h)(y+k)=a,
i.e. the rectangular hyperbolas have fixed asymptotes parallel to the axes
and with centre at ( -h, -k). The range Q^x^j--h limits each curve
/c
of the system to the positive quadrant. Certain curves of the system in

the case h = 2, k = 1 are shown in Fig. 25.
FIG. 25.
Ex.4. The equation x2 +y2 =a 2 represents a system of concentric

circles, a being a parameter. A different system of circles is defined by
the equation
x+y '{-*J%xy=a, O^xâ,
where a again the parameter.
is (The square root is assumed positive.)
The equation can be written
and, on squaring and collecting

terms, we obtain
The circle of the system with

parameter a thus has centre
(a, a) and radius a; as the
parameter a increases the centre
of the circle moves away from
O along a line bisecting the angle
between the axes and the radius
of the circle increases. From
the equation, it follows that
0<#+2/<a and, since x is
limited to the range <#<:#, both x and y are positive and less than a.
This limits each circle of the system to one quarter of its circumference,
as shown in Fig. 26.
Any system of curves defined by a single parameter a can be
represented by the equation
/(a?, y\ a)-0,
where, for convenience, theparameter is included in the functional
expression. An important kind of curve system arises when the
curves do not intersect each other, one and only one curve of the
system passing through each point of the relevant part of the plane.
Here, each pair of values of x and y is associated with only one value
of the parameter a, i.e. the value defined by the curve of the system
passing through the point (x, y). The parameter a can, therefore,
be separated off explicitly and the equation of the curve system can
be written in the form
/(a?, y)=a.
Each of the curve systems instanced above is of this type.* In
particular, Figs. 24, 25 and 26 show non-intersecting curve systems
in the positive quadrant of the plane and their equations can be
written
x +h
: a, (x+h)(y+Ic)=a and x+
fl
/
Vl/ +Kr
respectively, i.e. in the form f(x y)a. Non-intersecting curve
t
systems are of particular interest in economic theory (see 5.7 below).
3.8 Aneconomic problem in analytical geometry.

The consumers of a certain commodity are distributed over a geo-
graphical area which can be considered as a two-dimensional plane.
The commodity is produced by two firms situated respectively at
the points A 1 and A 2 in the plane. The problem is to determine how
the whole area is divided into two parts by a curve, the consumers of
one part obtaining their supplies of the commodity from the firm at
A! and the consumers in the other part buying from the firm at ^ 2 -t
The solution of the problem depends on what is assumed about the
" "
price at works charged by each firm, and about the nature of
the transport costs from factory to consumer. It is assumed, here,
* If the curve
systems of Figs. 24 and 26 were not confined to the positive
quadrant in the way shown, they would cease to be of the non-intersecting
type.
t The solution of this problem, as given here, is based on the work of Prof.
Schneider. See Bemerkungen zu einer Theorie der Raumwirtechaft, Econo-
metrica, 1935.
that the factory price p per unit of the commodity is the same for
both firms. Further, it is taken, that the transport cost varies
directly with the distance of the consumer from the factory con-
cerned, the distance being measured in a straight line. The transport
cost per mile is, however, different for the two firms, being tt per unit
of the commodity for the firm at A l9 and t 2 for the firm at A 2 To fix .
ideas, we can take t2 >tly so that it is more expensive, other things

being equal, to buy from the firm at A 2 .
if a consumer is st miles from A and s 2 miles from A2 ,

then he
pays (p +trs1 ) per unit of the commodity if he buys from A t and
(P ^ '2*2) if his supply comes from
A 2 Assuming that consumers
.
buy from the firm for which total

price cheaper, the boundary
is
between the two areas required is

defined by the points in the plane
where the total prices for the two
supplies are equal :
Choose axes of reference as shown

in Fig. 27, where the origin is F IO . 27.
taken as the mid-point of AA 2 .
If the firms are 2a miles apart, A L is the point ( -a, 0) and A 2 is the
point (a, 0). If P
is any point (x, y) on the boundary, then
"
f y2 and s9 = >/teô) 2 -l-vi

and (z+a)
2
+2/
2 - 2 V(z-a) 2 +7/ 2 ,
i.e. t^(x
2 2
+y +2ax -fa 2
)
= t 2 2 (x z -f y 2 - 2ax fa 2 ).
Collecting terms and dividing through by (/ 2

2
-/j.
2
),
we have
as the equation of the boundary referred to our selected axes. Write
The boundary is then seen to be a circle with centre at (6, 0) and with
radius s/6 2 -a 2 The circle has centre beyond A 2 on the line AÂ^
.
and encloses the point A 2 as shown in the figure. On the assumptions

M.A.
we have made, therefore, the firm at A 2 supplies all consumers
within this definite circle, and the firm at A supplies all consumers
l
outside the Many other problems of this kind can be devised

circle.
by varying the assumptions the method of solution is similar in

;
all cases.
EXAMPLES III
The straight line
1. Find the distance between the points (1, 2) and ( -2, 1)and the co-
ordinates of the mid-point between them. Verify your results by drawing an
accurate graph.
2. Obtain an expression for the distance between the points (a^ 2 2a^) ,
and (aV, 2aJ 2 ), where a, ^ and 2 2 ar constants.
3. Show that the points (1, 1), (

- 3, - 1) and (
- 4, 1) form a right-angled
triangle.
4. Show that the point (2,

- 1 )
forms an isosceles triangle, and the point
(is/3, 1 - V3) an equilateral triangle, with the pair of points (1, 2) and ( - 1, 0).
5. A quadrilateral is formed by the points A (5, 2), B C (3, 8) and
(20, 10),
D (
-
12, 0). Write down the equations of the sides and show that the quad-
rilateral is a parallelogram. Show that the diagonals AC and BD have the
same mid-point, i.e. bisect each other.
6. By considering an isosceles right-angled triangle, show that the gradient

of a lino at 45 to the horizontal is unity.
7. An equilateral triangle has a side of 2 inches ; show that the perpen-

dicular from a vertex to the opposite side is of length ^3 inches. Deduce that
the gradients of lines making angles 30 and 60 with the horizontal are
and s/3 respectively.
"*
8. Find the equations of the lines passing through the point (2, 1) (a) with
a gradient of 2, (6) making an angle of 60 with Ox.
9. Show that the line joining the points (a, 6) and (ka, kb) passes through
the origin whatever the values of a, 6 and k.
10. Find the equation of the lino joining the points (4, 3) and (
- 2,
- 1) and
deduce that the point (1, 1) is collinear with these two points.
11. Show that the line joining the points ( - 1, -J) and (2, 1) passes
through the origin, which is one point of trisection.
12. Find values of a, b and c so that the line ax 4- by + c passes through

the points (1, J) and ( 1, 2). Verify your result by writing down the equation
of the line joining these points.
13. Find the gradient of the line 2x - 3y -f 1 =0 and the co-ordinates of the
points where it cuts the axes. Hence, describe the simplest method of
plotting the line on squared paper.
14. By finding the condition that the point (#, y) is equidistant from the
points (3, 1) and (
- 1, 2), obtain the
equation of the line which is the perpen-
dicular bisector of the line joining these points.
15. Writo down the gradients of the lines x-y + l 0, x + y 7=0 and I
2x - $y + 1 = 0, and show that they form a right-tingled triangle. Plot a graph

to illustrate this fact.
16. Obtain the equations of the lines which pass through the point (2, 2)
and are respectively parallel and perpendicular to the lino 2#+y-3 0.
Verify graphically.
Find the co-ordinates of the vertices of the triangle formed by the
17.
lines x 1, x-3y-I and #+2/-5=0. Show that the line joining tho
point (2, 1) to any vertex is perpendicular to the opposite sido. What is
the geometrical meaning of this result?
18. Show that tho lines 3x +y - 4 =0, x - 2y -f 1 and x + 5y - = all

pass through the same point. Show that the equation of tho third line can be
written in tho form (3x -\-y - 4) - X(x - 2y + 1) for some value of A. Why
should we expect this?
Curves and curve systems

19. Use tho
definition of the curve to find tho equation of tho parabola
with focus at the point (2, 1) and directrix y~ - J.
20. Plot a graph of the curve y x . Mark the position of tho focus and
z
directrix of this parabola and verify that points on tho curve are equidistant
from focus and directrix.
and the directrix of each of tho parabolas
21. Locate the focus
-2, y=2x z -2x + l and y~Kx-2x
z
y=x + '3x
2
.
Compare the parabolas as regards size and position.

22. By considering tho parabolas which represent the function, show that
yax*+bx+c has a greatest value if a is negative and a least value if a is
positive. At what value of x do these greatest and least values occur? (Cf.
Examples II, 14.)
23.Write down the equation of the rectangular hyperbola with asymptotes

parallel to the axes, with centre at (
- 1, 2) and with a 2 = 3.
24. Find the centre and the asymptotes of the rectangular hyperbola
xy - 2x - y - I = 0. (Cf. Examples II, 9.)
26. If a point moves so that the difference of its distances from the fixed
points (a, a) and ( - a, - a) is always 2a, show that the curve described is tho
rectangular hyperbola xy = a
2
.
26. If thetwo fixed points of tho previous example are (>/2a, 0) and
(
- show that the equation of the rectangular hyperbola is x* -y z = a 2
N/2a, 0), .
Plot a graph of tho curve when a = 1 and verify that it is tho same curve
as in tho previous example with different axes of reference. Where are tho
asymptotes now?
27. Write down the equation of tho circle with centre at the point (2, f)
and with radius f . Where does the circle cut the axes ?
28. Find the co-ordinates of the centre and the radius of the circle whose
equation is x 2 -f y 2 - 3x = 0.
29. Show that (x + 1) (x - 2) +t/ 2 - 1 =0 is the equation of a circle with the

fixed points ( - 1, 1) and (2, - 1) at the ends of a diameter.
30. From (ho it 'suits of Examples Fl, 38, deduce the length of the chord
cut off the lino 2.T-//+3 by tho rectangular hyperbola xy 2, and the
length of tho chord cut off the line x - 2y + I
z 2
by the circle x +?y =9.
31. Re-examine the problems of Examples II, 40, in the light of the results
of the present chapter.
32. Draw a diagram to illustrate the system of curves yâ^-x*, where

^.T <a and where the positive square root is taken. What are the curves ?
33. If h and k are fixed positive numbers and a a positive parameter, show
that y k -
(x + h) 2 where
9 ^ x ^ aV& - h, represents a system of parabolas,
only the parts in the positive quadrant being taken. Draw a graph of certain
curves of the system when. h~5 and k -~ 25, and verify that each curve is
concave to tho origin. Compare with the system of parabolas of Fig. 24, where
each curve is convex to the origin.
CL
34. Show that (x - h) (y - k) =a, where ^ x ^ h - Kr and where h and
, k are
fixed positive numbers and a a positive parameter, represents a system of
rectangular hyperbolas confined to the positive quadrant. Taking h~2 and
k I, plot a graph showing the four curves of the system corresponding to
a i> ! * ftn( i 2- Compare with tho curve system of Fig. 25.
system y = *J(a -#) where a is a positive parameter

3
35. Consider the curve
and Q^jL-â. Draw a diagram showing the curves for a 1, 2, 3 and 4,
verifying that the curves are 11011- intersecting arid convex to the origin.
36. In tho problem of 3.8, two firms are 10 miles apart, the price of their
product at works is the same for each and transport cost per mile per unit of
the commodity is throe times as high for one firm as for tho other. Show that
the former firm supplies an area within a circle of radius 3J miles. Draw a
graph to illustrate the distribution of the market.
37. In the problem of 3.8, the two firms are 2c miles apart, tho prices at
works are p x for one firm and p a for tho other and transport costs are t per
mile per unit of tho commodity for each. Show that the curve separating the
areas supplied by the firms has equation (referred to the same axes as in 3.8)
~-^- = l, where a^-~^ and 6s c 2 -a*. Plot a graph to illustrate tho
distribution when p t ~ 32, pz 40, =l and 2c = 10 miles.

CHAPTER IV
LIMITS AND CONTINUITY OF FUNCTIONS

4.1 The fundamental notion of a limit.
WE now in a position to begin the promised discussion of the

are
" "
concept of the infinite so essential for the full appreciation both
of the power and of the limitations of mathematical analysis. We
must first go back to the number system itself and elaborate the
associated notions of order, continuation and limit.*
Real numbers are capable of arrangement in order of magnitude
and we can select, from this order, a particular set of numbers
according to any rule we care to specify. For example,
1
j, 23456
23456.
*j, o, r, *t, \j
}
... j
1
2> <?' '4> 5? 6? ?>>
are two instances of increasing sequences of real numbers arid it is
clearly possible to quote many other examples. All such sequences
display the property that the numbers of the sequence can be
written down one after the other, by the rule of selection, without
coming to an end. This is the property of indefinite or endless con-
tinuation.
Increasing sequences can be divided into two classes. In one class,

of which the first sequence above is a member, the numbers of the
sequence get larger and larger without limit as we proceed through

the order. Given any number we like to name, no matter how large,
we can always find a number of the sequence which exceeds it. The
numbers of an increasing sequence of this class are said to tend to
infinity and here we have the idea of the infinitely large, the endless
"
continuation of an order of larger and
* MuclTof the detail of tho
present chapter is, perhaps, unsuitable at a first
reading. In any case, only general ideas concerning limits and continuity
need be obtained at first and, after the development of the differential calculus
in subsequent chapters, the reader can return profitably to a further study of
the present chapter in conjunction with the examples set at the end.
In the other class, of which the second sequence above is a member,
the numbers increase as we proceed through the order but not with-
out limit. On the contrary, the particular example above is such
that the numbers steadily approach nearer and nearer to the value 1 ,
without ever attaining this value. We can get a number as close as

we like to 1 by going far enough through the order. In this case,
the numbers of the sequence are said to tend to 1, and 1 is the limit,
or limiting value, of the sequence. Each increasing sequence of this
second class tends to some definite number as a limit and, since the
interval between any member of the sequence and the limit is sub-
divided ever more and more finely by later members, we have
the basis of the idea of the indefinitely small or continuous, an idea
again dependent on endless continuation.*
In the same way, we can select decreasing sequences of real numbers,
i.e.sequences, in which the order of magnitude of the numbers is
reversed. Two examples of decreasing sequences are
1, 0, -1, -2, -3, -4, ... ,
9 3 4 5 6 7
^ 2~> 3^ 4> J5J 6">
Again, there are two classes, one class illustrated by the first and
the other by the second example. In a sequence of the first class,
the numbers grow indefinitely larger numerically, but through
negative values instead of positive values. Such a sequence is said
to tend to minus infinity and we have the idea of the infinitely large
" " "
and negative, of minus infinity as the counterpart to infinity ".
In the second class of sequence, the numbers tend to a limiting value,
decreasing to the limit instead of increasing to it. In the second
sequence above, the numbers decrease and approach the value 1,
i.e. the sequence tends to 1.
* It
may be noticed that we have given no justification for the statement
that there are only two classes of increasing sequences, those tending to infinity
and those tending to a finite limit. We have not excluded the possibility of
other cases of increasing sequences. It is, however, a fundamental property of
the real number system that cases other than the two here mentioned do not
exist. We must assume the result without proof here. In passing, we can
remark that the same property is not true of the system of rational numbers
only. An increasing sequence of rationals may tend neither to infinity nor to a
rational number as limit. This is, in fact, the way in which irrational numbers
can be defined as the limits of increasing sequences of rationals not tending
to infinity or to a rational limit. For a discussion of these fundamental points,
based on the work of Dedekind and Cantor, see Hardy, Pure Mathematics
(3rd Ed., 1921), pp. 1-31.
LIMITS AND CONTINUITY OF FUNCTIONS 87
Finally, sequences of real numbers can be written down in which

the order does not proceed according to magnitude at all, sequences
which are not increasing or decreasing. Any of the three possibilities
already described may obtain for a general sequence of this kind,
i.e. the sequence may tend to infinity, to minus infinity or to a
numerical limit. For example, the sequence

113
2> 2> 4> 4>
5
6s
7.
C>
" "
(obtained by crossing two previously quoted sequences) is
neither increasing nor decreasing, but tends to the limit 1. An
additional complication is here introduced. The sequence may be
such that none of the three possibilities holds ;
the numbers of the
" "
sequence may oscillate in value without displaying any of the
three regular progressions discussed above. An example of such a
sequence is
1> s?> 3, , 5, TJ-,

... .
Here the sequence said to oscillate without tending to a limit.

is
A given sequence of numbers, therefore, can behave in one of four

ways. The sequence may
(1) tend to infinity,
or (2) tend to minus infinity,
or (3) tend to a numerical limit,
or . (4) oscillate without tending to a limit,
as we proceed through the order of the sequence.

All concepts of the infinite in mathematics, whether of the infinitely
large or of the indefinitely small or continuous, are based on the idea

of an endless sequence of numbers. In general terms, the infinitely
large corresponds to a process of multiplication without end, the
indefinitely small to a process of division without end. The indefin-
itely large and small are thus closely related (as multiplication is
to division) and they are merely two aspects of the same essential
notion of order and continuation. Further, the notions of the
number system itself. The system is
infinite are implicit in the real
of infinite extent since the order of the numbers can be continued
indefinitely it is continuous or indefinitely divisible since we can
;
continue indefinitely to insert numbers between any two given

numbers. In the present chapter, the ideas of a limit, of the infinite,
88 MATHEMATICAL ANALYSTS FOR ECONOMISTS
and of continuity will be developed. But the development must not
be allowed to hide completely the fundamental ideas we have
attempted to set out briefly above.
4.2 Examples of the limit of a function.

Our next consideration is the extension of the concept of a limit,
as described above, to apply to functions. The most convenient way
of introducing the variety of cases of a limit of a function is by
considering a number of actual examples. A careful study of the
examples below is recommended they will be found to cover most
;
of the possible cases of limits. We can then proceed to more precise

definitions.
1
Ex.1. y = l- x -
Giving x in turn the sequence of positive integral values and the

sequence of negative integral values, the corresponding sequences of
values of y can be obtained from the function as shown below :
The curve representing the function,

a rectangular hyperbola, can be
graphed from these tables (Fig. 28).
The tables show four sequences of
numbers which heave been met before.
The two sequences of the first table
are associated in such away that,
J as the ^-sequence increases indef-
-4 -2
initely through integral values, the
-2- ^/-sequence increases and approaches
the limit 1. The variable x is
actually continuous but it is clear
FIG. 28.
that the insertion of further values
into the ^-sequence makes no essential difference the ^/-sequence always
;
increases to the limit 1. This idea of a correspondence between x- and

1
^-sequences is expressed by saying that the function y^-l-- tends to 1 as

x
x tends to infinity. The graph of the function illustrates this fact. As
x increases through positive values, the curve shown in the graph rises
from left to right and approaches the horizontal line drawn at unit
distance above the axis Ox.
In the same way, the second table provides a further pair of associated
sequences, the ^-sequence tending to minus infinity and the //-sequence
decreasing to the limit 1. This enables us to say that the function y = \~

x
tends to 1 as x tends to minus infinity. The graph again illustrates the
tendency to the limit, the curve falling from right to left and approaching
the same horizontal line as before as x decreases through negative values.
2
Ex.2. ?/-o; +3#-2.
The following tables of corresponding values arc obtained :
Jt follows that, when x is given an increasing sequence of values tending

to infinity, the corresponding sequence of values of?/ also tends to infinity.
A similar result holds when x is given a decreasing sequence of values tend-
2 -
ing to minus infinity. Hence, we say that the function y^x + 3x 2 tends
to infinity as x tends to infinity or to minus infinity. Fig. 4 illustrates these
facts, the curve shown (a parabola) rising indefinitely to the right and left.
In the same way, the function y =x* - 3x - 3 tends to infinity as x tends
to infinity. But as x is allotted a decreasing sequence of values tending
to minus infinity, the corresponding values of y are seen to form a similar
3
sequence. Hence, the function y x 3x 3 tends to minus infinity as
x tends to minus infinity. Fig. 12 illustrates this limiting tendency, the
curve shown falling indefinitely to the left.
Ex.3. =>!.
The following tables of corresponding values are constructed :
and the graph shown in Fig. 29 is obtained.

In this case, it is immaterial whether positive or negative values of x
are taken, the value of y being independent of the sign of x. As x is given
an indefinitely increasing sequence of values, the corresponding sequence
of values of y gets smaller and smaller. Hence, the function y~z

x*
tends
to zero as x tends to infinity or minus infinity. The value of y is not defined

at all when x is zero. But, as x
given a decreasing sequence of values
is
tending to zero, the corresponding sequence of values of y is seen to in-
crease indefinitely. Hence, the function y =x

-^ tends to infinity as x tends
to zero. The graph of the function illustrates these limits. The curve
falls and approaches the axis Ox as
we move to the right or to the left
and it rises indefinitely
indefinitely,
and approaches the axis Oy as
x decreases through smaller and
smaller values to zero.
Q
The function y ~ - shown graph-
,
x
ically in Fig. 5, behaves in a similar
way, except in one important par-

ticular. As x decreases in numerical
-4 4 X value, the numerical value of y
increases indefinitely, but y is
positive when x is positive and
negative when x is negative. Hence, the function y=-x tends to infinity as
x tends to zero through positive values, and tends to minus infinity as x

tends to zero through negative values.
Ex.4. y=
The graph of this function is shown in Fig. 33 below, the curve repre-
sented being a rectangular hyperbola. It is only necessary to construct
suitable tables of corresponding values of x and y, particular attention
being paid to the approach of x to the value 1 (where y is not defined). It

follows that the function y tends to 2 as x tends to infinity or to minus
infinity. Further, though not defined when z = l, the function y tends to

infinity as x tends to 1 (through values greater than 1) and to minus infinity
as x tends to 1 (through values less than 1).
(1+ 1
Ex y=
. 5.
y~ -
In this case, no value of y corresponds to zero value of x since, if we

substitute #=0, the expression for y takes the meaningless form of zero
divided by zero. The value of y is, however, obtained for any non-zero
value of x and the following tables can be constructed :
From these tables,and from the corresponding graph of the function

shown in Fig. 30, it evident that the value of y approaches the limiting
is
value 2 as # is given a sequence of smaller and smaller values, whether

positive or negative. Though x can never assume the value zero nor y
attain the value 2, we can say that the, function y= -{(1 + #) 2 -l} Lends
to 2 as x tends to zero.
Ex. 6. Consider the step-function of

Ex. 7 in 2.1 above. At the value x ^-3,
the corresponding value of y = 6 Further,
.
as x approaches the value 3 through values

smaller than 3, the corresponding value of
y remains 6. But, as x approaches 3

through values greater than 3 (e.g. as x
decreases from 4 to 3), the corresponding
value of y is fixed at 7. Hence, we must
say that, for this step-function, y tends to
the value 7 as x tends to 3 (through values

greater than 3) and tends to the value ftasx Fio. 30.
tends to 3 (through values less than 3). The

nature of the graph of the function (Fig. 8) near the point where x = 3
illustrates this unusual limiting case.
4.3 Definition of the limit of a single-valued function.

jThe examples of the previous section show that the notion of a,
from that of the limit of a sequence
limit of a function is to be derived
of numbers. a given single- valued function of #, a sequence of
If y is
values can be allotted to x according to some prescribed rule and the

limit of the function is then dependent on the behaviour of the
corresponding sequence of values of y. The limit is essentially a
relative concept, ^h vafee approached by the sequence of y as the.
independent variable x is made to change in some given way. We have,
therefore, to
"
y tends to
The second blank is
,
-
fill in the two blanks in the stock
as x tends to
at our choice
phrase
."
the first blank

- is then determined
;
from the behaviour of the function.

The following definitions include all the cases of the limit of a
function y~f(x) which is explicit and single-valued.*
(1) The limits off(x) as x ~> oo .
For the limit of the function as x tends to infinity, we observe the

behaviour of the sequence of values of the function corresponding to
a selected sequence of indefinitely increasing (positive) values of x.
There are four possibilities :
(a) The values of f(x) may be positive and becoming larger and
larger as x increases. The value off(x) can be made to exceed any

given number, no matter how large, simply by allotting a sufficiently
"
large value to x. Here, we say that f(x) tends to infinity as x tends
to infinity ", a phrase which can be symbolised
/(#)-><*> as #->QO or Lira f(x) = <*>.

X-+CQ
In this case, the curve yf(x) rises indefinitely to the right (the axis
Ox being horizontal) as shown in curve (a) of Fig. 31.
(6) The value of f(x) may be negative and numerically larger and
larger as x increases. This case is similar to the first and we say that
"
f(x) tends to minus infinity as x tends to infinity ", the symbolic
expression for which is
/(#)->
- oo as #->co or Lim /(#)= - oo .
re-*- oo
The curve y~f(x) now falls indefinitely to the right (see curve (6)
of Fig. 31).
(c) The values off(x) may form a sequence approaching a definite
value A as # increases. The value off(x) can be made to diifer from A
by as little as we please simply by allotting to a; a
sufficiently large
value. Here we say that " f(x) tends to the limit A as # tends to
"
infinity and we write
f(x)-+X as #->oo or Lim /(#)=A.
* The in a general way, are not

definitions given here, though clear enough
sufficiently rigorous from a purely analytical point of view. For more precise
definitions, see Hardy, op. cit. 9 pp. 162-8. See also Examples IV, 19 and 20.
The curve y=f(x) now approaches nearer and nearer (from one side
we move indefinitely
or the other) the fixed horizontal line y=X as
far to the right (see curve (c) of Fig. 31).
(d) The values of f(x) may behave none of the previous three
in
" "
ways, oscillating without settling down to any trend as x in-
creases. In this case, the function f(x) has no limiting value as x
tends to infinity and the curve y f(x) continues its finite oscillations
as we proceed indefinitely to the right (see curve (d) of Fig. 31).
x O
FIG. 31.
The following examples illustrate this class of limits :
1 , 2x+l
1 as as
x-l
as as
as as
x
The limit off(x) as x tends to minus infinity is defined and denoted
in similar ways. For example :
1 as - <x> as
x
(2) The limits off(x) as x-*a.
The problem of definition here is to trace the behaviour of

the values of f(x) as the variable x is allotted a sequence of
values approaching nearer and nearer to the given value x~ a. In
no case does x ever actually attain the value # a in the limiting
process.
Suppose, first, that the ^-sequence approaches x~a only through
values greater than a. The corresponding sequence of values off(x)
is subject to the same four possibilities as before, i.e. the function
f(x) can tend to infinity, to minus infinity, to a finite limit A or to no
limit at all as x approaches a through values greater than a. The
symbolic notation here is similar to that adopted above, the phrase
" "
x tends to a through values greater than a being used and written
"
x-*a + ". For example,
/(#)->A as xâ+ or Lim f(x)=X

Z-HJ +
indicates the case where f(x) tends to the limit A.

when the ^-sequence is taken through
Similar remarks apply
values than a. Here, x tends to a through values less than a
less
-
(written x^-a ) and we have, for example,
f(x)->X as x->a- or Lim f(x)~\.

x-+a
Finally, if it is found that f(x) tends to the same limit as x-*a -f

and as x->a ,
we say that f(x) tends to this limit as x tends to a
(written x->a). So, for example,
f(x)->\ as x->a or Lim
implies that f(x) tends to the limit A as x tends to a through values

both greater and less than a. This is the most important case in this
class of limits. #->a + and as x-*a - are different,
If the limits as
it is not possible to combine them in one notation. The diagrammatic
significance of limits as x tends to a is shown in Fig 32, which exhibit

the curve y=f(x) in the neighbourhood of the point In the xa.
cases of curves (a) and (6), f(x) has a single limit as x~>a, the limit
being finite in one case and infinite in the other. In the cases of
curves (c) and (d), the limits as x->a + and as x-*a - are different,
both being finite in one case and both being infinite in the
other.
The following examples illustrate the possibilities :
3
-- - >00 as x >0 >2 as
;
x
as x-*Q -f as 4-
;
x-\
> GO as x >0 ;
> oo as
The step-function of Ex. 7 of 2.1 above gives
2/->7 as #->3 + and y~>6 as #->3 - .
All the cases of Fig. 32 are illustrated by these examples.
(b)
(c)
4.4 Limiting and approximate values.

It isimportant to point out and emphasise the connection between
limiting and approximate values. If a sequence of numbers tends
to a finite limit, then, by numbers of the sequence
definition, the
approach nearer and nearer to the limiting value without ever at-
taining it. It follows that any number of the sequence is an approxi-
mate value of the limit, and conversely. Further, the approximation
becomes closer the farther advanced in the sequence is the number
taken. Thus, no number of the sequence
1 2 3 4 5. j6
JJ> 3> 49 5> 6) 7>
isever equal to the limit 1. But 1 is an approximate value of any

number of the sequence, and the farther we proceed through the
sequence the closer is the approximation.
Similar remarks apply when we come to the case of a limit of a
function. To suppose the function f(x) tends to a finite
fix ideas,
limit A as x tends to infinity, i.e. the value of f(x) approaches, but
never attains, the value A as # increases indefinitely. It follows that
A is an approximate value of the function f(x) when x is large arid
the approximation becomes closer as we allot larger values to x.
For example, since we know that
as x -> <x>
x-l
it follows that this expression approximately equal to 2 when the
is
value of x is large. And the larger the value of x taken, the better
is the approximation. To check this statement, put x~ 100 :
2x+l 201
V-~99n
x-l -2-0303...,
which is within 0*04 of the limiting value 2.

The limit of a function is thus an abstract notion in the sense that
the function never attains the other hand,
its limiting value. On
this abstract notion is precise and does correspond to a mathemati-
cally significant property of the function. One of the main uses of

the limiting concept is derived from the definition the limiting value ;
serves as an approximate value of the function for large values of x

(if the limit is obtained as x tends to infinity) or for values of x near
a (if the limit is obtained as x tends to a). In short, the precise
notion of a limit covers the vaguer notions of approximate values.
In conclusion, it is hardly necessary to stress the fact that the
" oo "
symbol which appears in the symbolic expression of certain
"
limits does not stand for a number infinity ". The symbol means
nothing by itself and must never be detached from its context. The
" "
phrase x tends to infinity ", or its symbolic expression x->cx> ",
is to be taken as a whole ;
it is simply a convenient way of expressing
the fact that the variable x is being allotted a sequence of larger and
larger finite values and that the corresponding variation in the value
of the function is under observation. Similarly, when we use the

"
phrase f(x) tends to infinity ", we must remember that the valuer
of /(#), however large they may be, arc always finite values. The
development here given of infinite limits involves only finite numbers
together with the idea of processes which can be applied to finite
numbers repeatedly and without end.
4.5 Some properties of limits.

Two
single-valued functions (x) and $(x) are given and it is
<
found that (x) -> A arid ifj (x) ->//, as #->co where A and ju, are finite
</> ,
values. The following results can then be established :
(1) <f>(x) + $(x) ->A-ff* as #->o>.

(2) (f(x) -ifj(x)~>X~fji as
(3) (>(x) .
^(#)->A/^ as
as
From our definitions of limits, it appears that these results are

too obvious to need formal proof. Since the value of (z) approaches </>
nearer and nearer to the value A, and t/j(x) nearer and nearer to the
value /z, the value of (e.g.) the sum of these two functions must
approach nearer and nearer to the value (A +/x) as x increases
indefinitely. Hence, (A +/LC) must be the limit of (<f>(x) +i(f(x)} as
a;tends to infinity. Exactly equivalent results hold for limits as x
tends to minus infinity or to a finite value a. Further, the results
extend, in an obvious way, for cases of sums, differences and pro-
ducts or quotients involving more than two functions.*
We are now able to evaluate the limit of a complicated function,
making use only of the limits of the simpler functions included in its
expression and the general results set out above. The method is
illustratedby the following examples :
Ex. 1. The constant value 2 has, of course, the limit 2 as x-*<x>.

3
The function >0 as a? >oo .
Hence, by result (1),
x
3
2+-->2 as #->oo .
x
* With more strict definitions of limits
(see 4.3, footnote), the results
here given are rigidly proved only with some difficulty. See Hardy, op. cit.,
pp. 125-30.
G M.A.
2x + 1 1
Ex.2. We know that _-~>2 and 1 >l as #-^<x>. But
x-1 x
x*+3x-l
X [X JLy X ~~ JL X X ~~~
J.
From result (2), it follows that
rr >L as X~>OO .
2x + l 1
Ex. 3. Since ~->2 and >0 as x ->oo the result , (3) shows that
x 1 x
~^\j as S/'^QO
It to be noticed that the results do not necessarily hold when

is
the limit of either of the functions <f>(x) and ift(x) is infinite. If, for
example, both functions tend to infinity as x tends to infinity, then
we can deduce nothing about the limit of the quotient of the
functions. Numerator and denominator both increase indefinitely
as x increases and the quotient can vary in diverse ways according to
the forms of the two functions. If (x) 2$ (x) the quotient has the <j>
limit 2, while if <

(x)
= |0 (x) the quotient has the limit |. And there
are manyother possibilities. Care must be taken, therefore, in the
use of the above results to avoid reaching fallacious conclusions
about limits of functions made up of simpler functions with infinite
limits.
4.6 The continuity of functions.

The definition of a function is based on the correspondence be-
tween values of two variables to each value of one variable there
;
corresponds a definite set of values of the other variable. The

variables themselves may be continuous (in the sense, that the
number system is continuous) or they may be discontinuous. But
the definition of the function quite independent of any idea of
is
continuity. The substitution of a value of one variable in the func-

second variable, is quite
tion, to give the corresponding values of the
independent of the substitution of other values.
The step we now propose to take is to extend the idea of con-
tinuity to qualify functions, to distinguish between functions which
are continuous and functions which are discontinuous. The distinc-

tion can be expressed, in a rough preliminary way, by saying that a
function is only continuous if the related variables vary continuously
together. The variables must be separately capable of continuous
variation. But this is not enough the continuous variation of the
;
two variables must be synchronised.

The discussion is again restricted to the case of a single -valued
function of a continuous variable: y=f(x). The general idea of a
continuous function applies to the function as a whole and involves
the continuous variation of y as x varies continuously. When we
attempt to formulate a precise mathematical definition, however, we
find it necessary to define first the concept of continuity at a point ,
i.e. at one value of x. Point continuity is mathematically more
significant than general continuity.
The for the continuity of the function at the point
first requisite
x=a is that/(#) should be defined both for the value x ~a and for
all values of x near x~a. There are now several possibilities con-
cerning the variation of f(x) as x approaches x a in relation to the
value /(a) assumed at x=a. Firstly, f(x) may tend to no limiting
value or to two different limiting values as x tends to a from one side
and from the other. Secondly, f(x) may tend to a definite limiting
value as x tends to a from either side but this limit is different from
/(a). Thirdly, f(x) may tend to the value /(a) as x tends to a from
either side. It is only in the third case that the function is described
as continuous at x=a.
In diagrammatic terms, the function yf(x) is represented by a
curve in the plane Oxy which is cut by lines parallel to Oy in single
points. Two conditions are needed if the curve is to be continuous
at the point where xa. Firstly, there must be a definite point P
on the curve at x=a
there can be no gap in the curve. Secondly,
;
as we move along the curve towards the value # a, we must arrive

at this point P
whether we approach from one side or the other ;
there can be no jump in the curve as we pass through P. A curve is

continuous at a point x=a, therefore, if there is neither a gap nor a
jump in the curve at the point. In Fig. 32, the curve (a) is the only
one which is continuous at xa. Curves (6) and (d) are discon-
tinuous because there a gap in the curve at the point, and the
is
curve (c) because there is a jump in the curve.

The formal definition of continuity of a function at a point is :
DEFINITION : The function yf(x) is continuous at x=a if
(1) /(a) exists and is finite ;
(2) f(x) exists and is finite at all values of x near xa in such a

way that Lim /(#)/(#).
The definition is equivalent to the identity of limit and value off(x)

at the point, and it is essential to see the necessity of the two con-
ditions that/(#) assumes a definite value f(a) at the point and that
f(x) tends to the same value as x approaches a from either side. The
conditions correspond to the diagrammatic conditions that the curve
y=f(x) should have neither gap nor jump at the point in question.
It is now a
simple matter to extend the definition to include con-
tinuity of a function in general. A function is continuous over a
range of values of x if it is continuous at every point of the range.
On the other hand, a function is not continuous if it exhibits at least

one point of discontinuity in the relevant range.
4.7 Illustrations of continuity and discontinuity of functions.

is continuous over a range, it is continuous at every
If a function
point of the range and can be represented by a curve which has no

gaps and no jumps in the range concerned. This implies, roughly
speaking, that the curve can be drawn graphically on squared paper
without taking pencil point from paper.. It is important, however,
to stress the fact that a continuous curve is not necessarily a
" "
smooth curve (using the everyday meaning of the term
"
smooth "). Smoothness is, in fact, more limited than continuity.*
In Fig. 33, the curves (a) and (6) are both continuous over the whole
range. The of these, the parabola y~ \ (x*

first +3x ~ is smooth
2),
as well as continuous. The second is the curve drawn from the
equation y = I +/(x- I) and it is continuous since it can be drawn
2
without taking pencil from paper. But it is not smooth in the usual
" "
sense since it has a sharp point at the point where rc = l.
If a function is discontinuous in a range, it must have at least one
point of discontinuity where one or other of the conditions for con-

* Smoothness of a curve is a " "
concept connected with the derivative of
the function (see 6.5 below).
tinuity fails to obtain. In diagrammatic terms, the curve represent-

ing the function has at least one gap or jump somewhere in the range
concerned. The most important example of a discontinuity arising
from the non-existence of some point on the curve, i.e. by reason of
a gap in the curve, is that of an infinite discontinuity. The value of
the function y=f(x) at such a point x=a is infinite in the sense that
" "
f(x) -> GO as x->a. The curve
goes off to infinity as we approach
the point from either side. This kind of discontinuity is of frequent
appearance even when we consider quite ordinary algebraic functions.
3 X
12"
9- (d)
-4 -2 2 4 6 8 X
o -12 x
FIG. 33.
The function which represents a rectangular hyperbola with asymp-

totes parallel to the axes (i.e. the function which is the ratio of two
linear expressions in x) always includes one infinite discontinuity, at
the point where we have the vertical asymptote of the curve. The
curve (c) of Fig. 33 shows a rectangular .hyperbola witli an infinite
discontinuity at x = l. Infinite discontinuities may be troublesome
but they can be allowed for, in mathematical analysis, by the device
of infinite limits.
The case where the discontinuity arises because of a jump in the
curve very different. The curve (d) of Fig. 33, corresponding to
is
the step-function of Ex. 7 of 2.1 above, displays six points of dis-

" "
continuity of the jump kind. Such a discontinuous function is
mathematical analysis, and it cannot be
clearly of little service in
represented by a simple analytical formula of the kind we usually
consider in mathematical work. But, as we shall see in the following
chapter, it is the kind of function that applies most closely to many
problems of economics and other sciences. We are here faced with
the difficulty that the functions most suitable for the description of
economic phenomena are often those which are most inconvenient
from the point of view of mathematical analysis. Some way
out of this difficulty of the discontinuous element in economics
must be found if the full power of mathematical analysis is to be
applied.
4.8 Multi-valued functions.

Limits and continuity have been discussed above only in relation
to single-valued functions. It is possible, however, to extend the
definitions to apply to functions which are multi- valued and given
in implicit form. Since no essentially new ideas are involved, it is
not necessary here to consider the extended definitions in detail. A

brief general account is all that is required.
An implicit relationbetween x and y usually gives y as a multi-
valued function of x. The ideas of limits and continuity, as applied
to such a function, necessitate the separate treatment of each of the
several values of y that correspond to any given value of x. If the
" "
function is easily divisible into a number of distinct branches each
of which single-valued (see 2.4 above), there is little difficulty.
is
" "
The limit of each branch of the function, as x tends to infinity
or to a definite value a, is defined as before and so the limits of
" "
the whole function are described. Further, each branch can be
considered separately from the point of view of continuity. All
66 " " "
branches may be continuous or one or more branches may
show points of discontinuity.
In diagrammatic terms, taking the axis Ox as horizontal, the curve
"
representing y as a multi- valued function of x shows several sec-
" " "
tions in vertical line with each other. Each section corresponds
"
to one single-valued branch " of the function. All sections may be
continuous and fit on to each other in such a way that the' whole
curve can be drawn without taking pencil from paper. On the other
" "
hand, points of discontinuity (whether of the infinite or jump
kind) may appear on one or more sections of the curve.
To take an
actual example, the relation x* y 2 9 defines y as a
double-valued function of x with the single-valued branches
y = Jx - 9 - Jx 2 -
2
and y 9,
where we can consider x as taking values numerically greater than 3.

Each branch is found to be continuous at all points. For the first
branch, y->vo as #->o>. For the second branch, y->-ao as
x-+ oo Fig. 34 shows the graph of the function and illustrates the
.
fact that it is a continuous function

with the specified limiting ten-
dencies.The curve is an example
of a rectangular hyperbola but the
asymptotes of the curve (shown
by broken lines in Fig. 34) are not
parallel to the axes of co-ordin-

ates. The double-valued function
x2 4-= 16, corresponding to the
y
2
circleof Fig. 6, can be shown to be

continuous in the same way.
When *a multi-valued function Fia. 34.
cannot be divided easily into dis-

"
tinct branches ", some difficulties emerge. But the ideas of limits
and continuity still apply in general. We can instance the triple-
valued function x 3 3 - 3xy ~ 0. The
-f j/ graph of this function (Fig. 7)
indicates that it is continuous, being a looped curve with three
sections fitting continuously on to each other.
EXAMPLES IV
Limits of functions
1. By giving x an arbitrary sequence of increasing positive values, find

the limit, as #->oo of each of the following functions
, :
1 1 -x 1 -x 2
2.P 8
+a? + l 1 + 2x + 3.E 2
Illustrate graphically. Are the same limits obtained as x -> - oo ?

2. What are the limits, as a;--oo , of the functions :
2 2s; 2 + 3a; + 1
Noticing that (c) is the sum of (a) and (6), and that (d) is the product of
(a) and (6), verify the general results that the limit of a sum or product is
the sum or product of the separate limits.
x+ 1
3. Explain why tho limit, as #-oo, of cannot be obtained by
j
oX 1
multiplying the limits of (x + 1) and - .
A 01
Show 4-1,4.
that
4.
a; +2
Why cannot the limit, as #->oo

of tho other two expressions ?
, of -
x
x+2,
1
- be deduced as tho sum of tho limits
5. Show that -
tiC
- -
o
->- as
4
x-+<x> . What percentage error is involved by
taking J as the value of this expression in ;r when x = 10 and when x 100?

How large must x bo for the expression to differ from J by less than -005?
6. Establish the general result that, as #->oo ,
by writing numerator and denominator as x f a^ 4-- ) and a; f a 2 -f-~ ) respec- 7
tively and by using the general results for the limits of sums, products and
quotients. Check the validity of the general result by reference to tho
particular cases given in previous examples.
7. Find an expression for the difference between ~ -

a z x+b 2
=-
1
and -
a2
and show
that it tends to zero as x ->oo . Hence deduce the general result of the previous
example.
By methods similar to those of tho previous two examples, show that

8.
z
a^x - + b~tx
a^c* +bzX-^c 2
- -f GI*
-->!
*
at
az
as #->oo .
What modifications are necessary when % or a 2 is zero? Check this result

by reference to tho particular cases of the first two examples above.
9. If y ax 2 + bx -f c, show that i/->oo as#->oo. How does the sign of a
affect the result?
10. If x is positive and 2/

= 2o;-3+-, show that 2/->oo as x-+Q and as
x
a;->oo .
Verify that these results can be obtained by adding tho limits of
(2a;
- 3) and - evaluated separately.
x
11. A between x and y is defined by x t 2 y2t. Show that

relation ,
x -ôo as and conversely. Can an exact meaning bo attached to the

2/-VQO ,
statement that x tends to infinity " faster " than y?

12. Show that -{(1 +x)* - 1}->3 as x-+Q. Generalise by finding the limit
x i
as #->0 of tho expression -{(a +#) 3 -a 3 }.
x
13. Show that -I-

X \L
--
X l)->-l
~{~ f
as #->0, and generalise by finding the
limit of - (
---- as x-*Q.
x\a+x a/)
14. What error is involved by taking - 1 as tho value of - - - 1 r when

x^ 0-1 tod when x ^0-01? x\l+x
{
)
1
15. Verify that -{1 - s/1
long as x ^-0.
x
Deduce
-x}
that, as x ^0,
is
{1
-
_
equal to the reciprocal of
N/i -x}-+\. By
{1 + s/1 -x}
a similar method,
as
show that -{vT+a? - \/l -#}-^l as #->0.

#
16. A lead shot is allowed to fall in a vacuum. It is known that the shot
falls Hie 2 foot in x seconds. Find an expression for the average number of
feet fallen per second between the rath and the (a+x)th second. Find tho
limit of the expression as #->0 and give a meaning to this limit.
17. P is the point with abscissa a, and Q the point with abscissa (a +x) on
tho curve y=x*. The line through P
parallel to Ox cuts the line through Q
parallel to Oy in the point L. The ratio of LQ to PL is tho gradient of the
chord PQ. Find an expression for this gradient in terms of a and x, evaluate
its limit as x-rO and explain the meaning of the limit.
(The results of this and the previous example illustrate problems of funda-
mental importance in tho development of the following chapters.)
18. A right-angled triangle has two equal sides of 1 inch. One of these
sides is drawn horizontally and divided into (n + 1) equal portions. On each
of the portions after tho first a rectangle is formed with height equal to the
vertical distance from the left-hand end of the portion to the hypotenuse of
tho triangle. Find an expression for the sum of the rectangle areas and 5
evaluate tho limit of the sum as n-ôo . What is tho meaning of the limiting
value?
19. Show that a more rigid definition of the idea of a limit of a function is
provided by tho following :
Tho function /(#) tends to tho limit A as a: tends to a if, for any selected
small quantity c, we can find a small interval a-S<x<a + 8 such that
{/(#) ~M i g numerically less than e for any x of the interval.
20. In the same way, examine the more rigid definition :
Tho function/ (a;) -> oo as #->oo if, for any quantity 6 no matter how large,
we can find a quantity a so that /(#) >b for all values of x >a.
Continuity of functions
21. Examine tho function y =

illustrate graphically.
1
-
1
-
x
+a?
from the point of view of continuity and
22. From the function xy + 2x + y - 1 = 0, find tho limit of y as #-* - 1, and

the limit of x as y -> 1 What restriction must be added to tho statement that
.
y is a continuous function of x and conversely? 9

23. In the general relation type axy + bx + cy + d 0, show 'that y is a single-
valued function of x which is continuous except for an infinite discontinuity
at one point. Show also that x is a similar function of y. Relate the result
to the properties of the rectangular hyperbola.
24. For what values of x does y = - - - X 4- 1

tend to infinity ? Indicate
the form of the graph of the function and describe its discontinuities.
25. What are the discontinuities of the function
x-x +
Illustrate graphically.
26. Show that the function M = -i

-
-^ - -
never tends to infinity
J if
<
6 2 2 4a 2 c a and tends to infinity for two values of x if 6 2 2 >4a 2 c 2
, Examine .
also the case where 6 2 2 4o a c a Interpret in terms of the continuity of the

.
function.
x 2 - -1
27. Show that y - is continuous except at x = 1. What is the nature
x 1
of the discontinuity at this point? Show that the function becomes com-
pletely continuous if the value y =2 at x 1 is inserted.
28. Describe the discontinuities of the step-function, illustrating with the

function defined by the table :
29. Themotion/ (x) is defined as taking the value x when x is positive and
f i
the value -x) when x is negative. Deduce that the function is continuous
(
but has a " sharp point " where x 0.

30. If a is a fixed positive number and if 7&->oo through integral values,
show that an ->0 if a< 1 and that an ->oo if a>l. What can be said of the
case a = 1 ? Deduce the limit of as n->oo for various fixed values of a.
l+an
31. The function f(x) is defined as taking its value from the limit of
as n tends to infinity. Describe the variation of the function for positive

values of a; and illustrate graphically. Show that tho function is discontinuous
at x = 1. What is the nature of the discontinuity?
32. Draw tho graph of the function 4x z + 9^/ 2 =30 and show that it repre-
sents a continuoiis oval curve.
33. The relation (1 -x) 3 -xy 2 defines y as a double-valued function of x.
Obtain the two single-valued " branches " and draw the graph of the function,
" "
showing that tho curve is continuous with a sharp point at (1, 0). What
are the limits of y as x -> ?
34. Show that the curve representing the double-valued function
y* = (a;-l)(a-2)
can be taken as continuous, except that there is a point (1,0) which must be
regarded as a part of the curve but completely separate from all other parts
of the curve.
CHAPTER V
FUNCTIONS AND DIAGRAMS IN ECONOMIC THEORY
5.1 Introduction.
ECONOMICS isan analytical study, concerned with the relations that

exist, or can be assumed to exist, between quantities which are
numerically measurable. As instances of variable quantities in
economics, we need refer only to prices, interest rates, incomes,
costs of production, amounts of goods bought or sold on a market
and amounts of factors of production employed by a firm or industry.
Some of these quantities can be measured in physical or " natural "
" "
units, others only in money or value units. The relevant point
here is that they are measurable in some units. There are, however,
other concepts of considerable importance in economics, concepts
which are ordered magnitudes rather than measurable quantities and
" " "
which can only be represented by indicators or index-numbers ".
" " " "
All concepts of utility or satisfaction are of this nature, as
are notions of the general level of prices or production. But, for the
moment, we content ourselves with the fact that economic measur-
" "
able quantities exist and leave the ordinal concepts for later
consideration.
no doubt that mathematical methods
It follows that there can be
and that economic relations are expressible
are possible in economics
by means of mathematical functions. Whether it is helpful to intro-

duce the mathematical technique is another question and it is not
proposed to devote any space here to a discussion of this purely
methodological matter. It is sufficient that mathematical analysis
is applicable and can be called into service when convenient.
An
important point that the relations of economics, and the
is
functions which express them, are usually of unspecified or unknown

form. It is not often that we can say that an economic function is
(e.g.) of linear or quadratic form, though it is sometimes convenient
to assume that can be approximately represented in one of these
it
ways. Even in the most general case, however, the economic con-
ditions of the problem impose certain limitations on the form of the
functions. By considering the economic nature of our problem, we
are usually able to say that the function concerned has the mathe-
matical property (e.g.) of being single-valued and decreasing, or of
being represented by a U-shaped curve. And this is enough for the
profitable use of mathematical analysis.

Analytical methods, then, are directly applicable to economic
problems. But diagrammatic methods are also of great service.
Any economic problem capable of symbolic representation can, in
general, be illustrated with the aid of diagrams. There is a curve
corresponding to each function we use to interpret a relation between
two economic variables ;
we have demand or cost curves as well as
demand or cost functions. Diagrams displaying the properties of
such curves and relating one curve to another are extremely useful.
At least they throw into prominence just those points we are attempt-
ing to make in our main argument.
We can now pass to a consideration of some of the functions and
curves commonly employed in economic analysis, stressing their
general forms and giving an indication of the more limited forms to
" "
be assumed in normal cases. Each problem must be considered
on its own merits and allowed to dictate what is or is not to be
"
regarded as normal ". Mathematics is the servant of the sciences.
not thejna&tfil. The present chapter is divided into two main
sections. In the first section, we are concerned with those economic
functions which are expressible directly in explicit and single-valued
form, with demand and cost functions in particular. In the second
section, we pass to more complicated cases where the economic
functions employed appear in implicit, and not necessarily single-
valued, form. Here we are concerned with function types and
curve systems (see 3.7 above) as applied in the interpretation of
"
indifference maps ".
5.2 Demand functions and curves.

Our first problem is to express, in symbolic and diagrammatic
form, the conditions of demand on a consumers' goods market. We
limit our discussion to the case where there is pure competition
FUNCTIONS AND DIAGRAMS IN ECONOMIC THEORY 109
amongst consumers, where (to use Professor Frisch's term) each

consumer acts only as a " quantity adjuster ". This case occurs
when no individual consumer has any direct influence on market
prices (which are the same for all consumers), each consumer being
able to choose only the amounts of the various goods he will take at
the ruling prices. In other cases (e.g. when consumers have some
direct influence on prices), the construction of the demand relations,
on the lines described below, automatically breaks down.
Consider the amount of a definite good X
demanded by a market
consisting of a definite group of consumers. In order to obtain a
simple representation of the demand for X, we must assume that
(1) the number of consumers in the group,

(2) the tastes or preferences of each individual consumer for all
the goods on the market,

(3) the income of each individual consumer, and
(4) X
the prices of all goods other than itself
are fixed and known. The amount of X each consumer will take can
then be considered as uniquely dependent on the price of X ruling
on the market. By addition, it follows that the total amount of X
demanded by the market depends uniquely on the market price
of X. The demand for X can only change if its market price varies.
This expression of market conditions can be translated at once
into symbolic form. Let p denote, in definite units, the market
price of X, and let x denote, in definite units, the amount of X
demanded by the market. Then x is a single-valued function of
p, which can be written, in general, in the symbolic form :
x <f)(p) the demand function for X.

The variables x and p take only positive values.
The demand function can be represented as a demand curve in a
plane referred to axes Op and Ox along which prices and demands
are respectively measured. Since w e have taken price as the inde-
r
pendent variable, we should, according to the mathematical con-

vention, draw the axis Op horizontally. Since the days of Marshall,
however, the economic convention has been otherwise and the axis
Op has been taken vertical. We shall here adopt the economic con-
vention and all the demand curves we use are referred to a vertical
price axis.
It must be emphasised, at the very outset, that the demand
relation is a completely static one and does not refer to changes in
demand over time. The demand function and curve assume a fixed
situation in which a set of alternative and hypothetical prices are
given. These prices cannot be taken as those actually ruling on a

market at successive points of time.
The demand function and curve can only be specified if the four
factors given above are known. The form of the function and the
shape of the curve depend on these factors if changes occur in any

;
of the factors, then the whole form of the demand function and curve
changes. We can describe this by saying that a change in any

one of the factors causes a shift in the position of the demand curve,
the new demand curve being different, in shape and location, from
the old in a way dependent on the nature of the change assumed in
the factor. A 'shift in demand, corresponding to a change in the
demand situation as a whole, must be carefully distinguished from a
move along a given demand curve. The former implies a change of
data and may easily take place over time the demand curve can
:
shift over time. The latter implies a hypothetical and non-temporal
change in price in a given situation. Another way of putting the

point is to say that the demand function x-<f>(p) has form depend-
ent on a number of parameters, e.g. the number of consumers, their
incomes and tastes and the prices of other goods. The parameters
are fixed for one demand function they assume different values
;
only for a shift of demand, for a change in the whole demand

relation.
To take an illustrative case,suppose the market demand curve for
sugar has been fixed. The population now increases, the consumers'
tastes change in favour of sugar, the prices of tea and coffee fall and
the income distribution changes so that the poor have more and the
rich less to spend. We can say that the demand for sugar, at any
one price whatever, is greater now than it was before. The whole
demand curve for sugar has shifted to the right.
We must now examine the general nature of the demand function
and curve in the light of the economic conditions of demand. The
point that arises here concerns the continuity of the variables
first
x and p and the demand function itself and it is worth while treating
this point at some length since our remarks apply, with little modifi-
cation, to other functions introduced later. It would appear that

the demand for (e.g.) sugar can only be defined for a limited number
of isolated prices, for (say) prices at halfpenny intervals from Id. to
" "
6d. per Ib. Given the relevant data, the demand curve for sugar
then consists of an isolated set of points in the plane Oxp. The
adoption of such a discontinuous representation is clearly incon-
venient for a theoretical and mathematical development of the
demand problem. Can price be taken, for theoretical purposes, as
a continuous variable There is certainly nothing logically incon-
?
sistent in the notion ofa continuously variable price. Further, an

examination of actual practice shows that prices are less discontinu-
ous than they appear at first sight. Sugar may be quoted at 2d.
or 3d. per Ib. But it is possible to insert intermediate prices by
quoting sugar at two Ibs. for 5jd. at three Ibs. for 8d. or 8|d.
; ;
at
five Ibs. for Is. Id., Is. lid., Is. 2d. or Is. 2jd. ;
and so on. Hence,
unless there are strong reasons to the contrary, it can be safely
assumed that the price in a demand relation is a continuous variable
all positive values. The
taking
demand is then uniquely deter-
mined for each price of this
continuous range.
But, even when p varies con-
tinuously, it does not neces-
sarily follow that x varies
continuously with p, i.e. that
the demand function is con-
tinuous. On the contrary, it
is reasonable to assume that __
demand varies by a series of Flo

"jumps" as price increases
continuously. Demand may be insensitive to small price changes and

" "
jumps only when price has increased by a definite amount. In
" "
this case, the demand function has jump discontinuities at
certain prices and a step-function of the land already considered.
is
The demand curve takes the form illustrated in Fig. 35.

It is clearly convenient, for theoretical purposes, to assume that
the demandfunction and curve are continuous, that demand varies
continuously with price. There is, at least, nothing that violates our
logical sense in this assumption. And it is reasonable to suppose that
any actual demand conditions, for a large market at least, can be
" "
fitted sufficiently well by a continuous demand curve for the
latter tobe assumed. Fig. 35 illustrates a hypothetical fitting of a
continuous demand curve. It is not proposed, however, to consider
" "
here the fit of a continuous demand curve to actual data, even
if such data can be obtained. We need only notice the convenience
of the continuous demand function for theoretical purposes and point
out the limitations of this continuity assumption. We proceed, there-
fore, on the assumption that demand is a continuous function of a
continuously variable price. The assumption is made for purposes

of mathematical convenience it can be given up, if necessary, at
;
the cost of considerable complications in the theory.
5.3 Particular demand functions and curves.

We come now to an assumption of very different nature from that
of continuity. It is assumed that the larger is the price the smaller
is the market demand good concerned. The validity of this
for the
assumption must be examined, in the light of actual conditions, in

each particular case. In some cases of demand, as Marshall has
pointed out following Giffen, it is certainly not valid. But it is
" "
surely appropriate to all usual or normal cases of demand. Our
main development can thus proceed on this assumption of falling
demand, and the theory can be modified where necessary to meet
any exceptional cases that arise.
The assumption implies that x decreases as p increases, i.e. that
the demand function is monotonic decreasing. It follows that the
inverse function is also monotonic decreasing. We can take cither
demand as dependent on price, or conversely :
x (j>(p) and p = $(x).

Here, both functions are single-valued, continuous and monotonic
decreasing. In diagrammatic terms, the demand curve falls con-
tinuously from left to right and it has the same general appearance
whether we take Op horizontally or (as we do here) vertically.
Further restrictions on the form of the demand relation can now
be added according to the particular requirements of the demand
problem considered. Fig. 36 shows three general demand curves in
certain hypothetical and illustrative cases. The first curve D
l is
taken to represent the demand of a single family for sugar.* The

price is measured in pence per Ib. and the demand in Ibs. per month
per adult member of the family. No sugar is demanded at prices
above Is. 3d. per Ib. As the price falls, the demand increases slowly
at first, then rather rapidly and finally (for prices below 2d. per Ib.)
SUGAR SUGAR
5 10 15 20 20 4O GO 8O X
Ibs. per month Mn. Ibs. per month
P\
9- CIGARETTES
* 3--
-4-
O 30 60 90 120 X
Mn. packets per month
FIG. 36.
slowly again. The family never demands more than the " satura-
"
tion amount of 17 Ibs. per adult per month. The demand curve
cuts both axes and is first convex and then concave to the price axis.
The shape of such a family demand curve can vary, of course, over
a wide range for different families and different goods. The form Dl
is given merely as a fairly typical example.
D
The demand curve 2 is illustrative of the of a large demand
market for sugar, taken as typical of a good with no closely related
substitutes. The market demand becomes zero at Is. 9d. per Ib.,
no family making any purchases at this or any higher price. The
demand then an ever increasing rate as the price falls
increases at
* An " individual " demand curve of this kind can be
regarded as the
limiting case of a
"
market " demand curve when only one consumer is taken.
A demand curve, in fact, can be defined for the whole group of consumers or
for any particular group of them we care to specify.
H M.A.
until approximately 80 per month are consumed at very low
Mn. Ibs.
prices. The curve cuts both axes and is convex to the price axis at
all points. As an approximate representation of such a demand, we
"
could take a demand curve approaching the axes " asymptotically
in the manner of the rectangular hyperbola (see 3.5 above).
Sugar is a commodity which is well defined and clearly demarcated

from its nearest substitute (say saccharine or jam). But most com-
mercial commodities are such that there is, in each case, a number
of closely allied commodities.* The variation can be obtained in
many ways, e.g. by change of quality or by trade-marks. The
demand curve D 3 is drawn to illustrate the demand for Player's
cigarettes,taken as typical of such a commodity. Assuming that
the price of competing cigarettes is 6d. per packet of ten (a fixed
price by our assumptions), the demand for Player's cigarettes is very
small at 9d. per packet and increases rapidly as the price falls,
becoming very large when the price is 5d. per packet or less. The
demand curve cuts the price axis and, since a small price fall results
in a large increase in demand, it is nearly parallel to the demand axis
for most of its course. The curve must cut the demand axis at some
point, but only when the demand is very large compared with that
at the price of (say) 6d. per packet.
It is often convenient to represent a demand law by a definite type
of function, at least as an approximation for a certain range of prices.
The demand curve is then of definite class, e.g. a straight line or a
parabola. The following function types can be given as examples of
suitable demand laws :
(1)'
v
x ~~, p=a-bx.
* (2) x = ----- 6, p~ x+bT
* -c.
b p+c
pW^te. (4) *= -,
(5) #=
yÎjP } p=a-bx*. (6) x = bp-*+c, p= -~
(7) x=aer**, p = oJlog-.
x
(8) x^pe-*^+ c \
" "
Each of these demand laws satisfies the normal monotonic de-
* I.e. commodities of the kind considered in the economic theory of " mono-
"or
" "
polistic imperfect competition.
creasing condition.* The variables x and p take only positive values

and a, 6 and c are positive constants in each case. If actual numerical
demand data are available, it may be possible to fit actual values to
the constants and to draw a definite graph of the demand curve.
The demand curve of the first law is a downward sloping straight
line cutting the price axis at the point A where p = a and with
gradient (referred to the price axis) numerically equal to b. If the
values of a and 6 are
changed, then the demand P >
line shifts in position. If 24-
a alone varies, the line SUGAR

is transposed parallel to
itself. If b alone varies,
the line is rotated about
the point A on the price
axis. If both a and b
vary, the line shifts by
transposition and rota- 60 x
40 100
tion. The second demand Mn. Ibs, per month
law is (x +b)( p +c) a. FIG. 37.
The demand curve is now
a rectangular hyperbola with asymptotes parallel to the axes and
with centre at the point where x = - b and p ~ -c. The curve is
restricted, of course, to that part of the rectangular hyperbola in the
positive quadrant. The demand curve shifts in position when the
values of a, 6 and c are varied. Variation in the values of b and c
alone transposes the curve without change of shape. If b decreases,
for example, then the curve is shifted bodily to the right. Variation
in the value of a, however, changes the shape of the demand curve.
Fig. 37 shows actual cases of linear and hyperbolic demand curves

drawn from the demand laws :
or ~
600 600
and -20 or p=
' - 5-
p+5 x+2Q
In each case, x is the market demand for sugar, in Mn. Ibs. per
month, when the price of sugar is p pence per Ib.
* The last two cases involve exponential and logarithmic expressions (see
9.8 below). In case (4), x must be limited to the interval <x^ T .

o
An
important instance of the use of parametric constants is thus
obtained. If a demand relation is known to be of one of the general
types quoted above, then the position of the demand curve is fixed
only when the parameters are given definite values. The values a,
b and under given demand conditions. But a shift
c are constants
in demand results when the parameters are changed in value, the
nature of the shift depending on the particular change adopted.
Demand laws such as those given above suffice, not only for the
description of some given demand situation, but also for the shifting
of the demand situation over time.
5.4 Total revenue functions and curves.

The demand for a good X is represented by the two inverse func-
tions, x <f>(p) and p = $ (x), of the demand law. When the demand
isx and the price p, the product R xp is called the total revenue
obtainable from this demand and price. It represents the total
money revenue of the producers supplying the demand and the total
money outlay of the consumers composing the demand. Now
i.e.R can be expressed either as a function of the price or as a function

of the output ( demand). The latter is the more convenient ex-
pression in most cases (e.g. in relating demand to cost of production)
and the function E~x^(x) is called the total revenue function of the
given demand This function, like the demand function itself,

law.
assumes pure competition amongst consumers. The function is
represented by the total revenue curve referred to a horizontal axis
Ox and a vertical axis OB.* The height of the total revenue curve
thus measures the total revenue obtainable from the output indicated.
The total revenue function, being obtained from a continuous
demand function, must itself be continuous. But it is not possible,
" "
at this stage, to indicate the normal form of the total revenue
function consequent upon the assumption of a monotonic decreasing
demand function. It will be seen later that the form of the total
revenue function depends on the concept of elasticity of demand (see
10.7 below). We can simply notice here that E increases or decreases
* The fact that Ox is the obvious axis to take as horizontal for the total
revenue curve is one of the reasons for taking Ox as horizontal for the demand
curve itself.
with increasing output according as the demand is (in a very general

sense) elastic or inelastic. By an elastic demand, for example, we
mean a demand which increases in greater proportion than the
corresponding decrease in price. In this case, the revenue obtained
from the demand must increase as the demand increases.
If the form of the demand law is given, however, it is a simple
matter to deduce the form of the total revenue function and curve.
For example, in the case of the linear demand law bx, we have pa-
R=ax-bx*.
This expression can be written in the form
The total revenue curvethus a parabola with axis vertical and

is
pointing downwards (see 3.4 above). The highest point of the curve
a
occurs where #=^T' Total revenue increases as output increases at
2fy
a
first, reaches a maximum value -
f
at the output x ~a and then
46 26
decreases as output increases further. A total revenue curve of this
kind, obtained from a linear demand law, is shown in Fig. 44.''

'
The total revenue curve contains no reference to the price of the

good. But the price can be read off the curve quite easily. If P is
the point on the curve at output x OM, then

R MP ,.
fnT>
P = xÔ~M =8
.
dieni f '
Price can be regarded, in fact, as average revenue, i.e. as the revenue
per unit of the output concerned. y/As such, it is measured by the

gradient of the line joining the origin to the appropriate point on the
total revenue curve. Since price decreases as demand increases, the
line OP becomes less and less steep as we move to the right along
the total revenue curve.
5.5 Cost functions and curves.

The following assumptions are needed to obtain a simple expres-
sion of the problem of cost of production. given firm produces aA
single uniform good X
with the aid of certain factors of production.
Some of the factors are employed in fixed amounts irrespective of
the output of the firm, and it is taken that the expenditure on these
factors is known and fixed. The remaining factors are variable and
the conditions of their supply are assumed to be known. For
example, the factors may be obtainable at fixed market prices.
Further, the technical conditions under which production is carried
out are taken as known and fixed. Finally, the firm is assumed to
adjust the employment of the variable factors so that a given output
x of the good X is obtained at the smallest possible total cost 77.
Then U is determined when x is given and assumes different values
as x is changed, i.e. 77 depends uniquely on x :
II~F(x) the total cost function.
The variables x and 77 are necessarily restricted to positive values.

The function is shown by a total cost curve referred to a horizontal
axis Ox and a vertical axis O77 drawn in a plane. The varying
height of the curve shows the changing total cost of an increasing

output.
The total cost functionand curve are static constructions relating
to various alternative and hypothetical outputs produced under
given conditions. They do not describe the variation of cost and
output over time. The form of the function F depends on such con-
ditions as the technique of production and the conditions of supply
of the variable factors of production. The function and curve are
fixed only for given conditions and are completely changed in form if
any of the conditions are varied. For example, if the technique of

production is improved by an invention or if the wages of the labour
employed fall, then the cost function is changed so that the cost of
any given output is less than before and the total cost curve shifts
to a new position below the old curve.
It should be noticed that the total cost function and curve are
" "
minimal concepts. By varying the employment of the variable
factors, a given output can be produced at various total costs of
which only the smallest is used to define the total cost function. In
diagrammatic terms, the ordinate of a point on the total cost curve
represents the least cost of producing the output concerned and the
same output could be produced at larger costs shown by points
vertically above the point taken on the curve. The positive quadrant
of the plane OxU is thus divided into two areas by the curve ; points
in the areaabove the curve correspond to total costs at which the

outputs concerned could be produced while points in the area below
the curve represent total costs too small for production of the out-
puts. The total cost curve is the lower boundary of the area repre-
senting possible cost situations. But, since we usually need consider
only the lowest cost of any output, we are interested in the cost curve
rather than in the area above it.
For purposes of mathematical convenience, the single-valued cost

function can be taken as continuous. Further, under "normal"
conditions with at least one factor of production fixed, it is assumed
" "
that there are fixed overhead costs, incurred even when no out-
put is produced, and that variable costs increase as output increases.*
The total cost curve, therefore, cuts the vertical axis at a point above
the origin and rises continuously from left to right. Fig. 38 shows a
20 40 M 60 80 X
Tons per week
FIG. 38.
" "
total cost curve of normal form appropriate to a hypothetical
case of a sugar refinery overhead costs are 250 per week and total
;
cost increases with output, reaching a value of 1250 per week when
the weekly output of sugar is 80 tons.
It is sometimes convenient, at least for approximate represen-
tations of cost over a limited range of outputs, to take the total cost
* If we consider an
industry instead of a firm and take all the factors of
production as variable, then overhead costs are zero. Or, we may consider
only the total variable costs of a firm and neglect overhead costs. In either
case, the normal form of the total cost curve is similar to that given here
except that it now starts from the origin.
function of some particular form. The following are types of cost
" "
functions appropriate to the normal case :
(1) II=ax+b. (2) II=ax 2 +bx+c.

(3) Il-Jax+b+c. (4) nâx*-bx 2 +cx+d.
X
(5) IJâx
x+c
^+d. (6) nâx*
x+c
+d.
(7) nâe bx .
(8) II=x ae bx + c +d.
The parameters a, 6, c and d are positive in each case. In case (2),
the total cost curve is a parabola with axis pointing vertically up-
wards and with vertex to the left of the origin at #= - . Since

JLCL
the curveis limited to the positive quadrant, it consists of an arc of
the parabola rising continuously from left to right. The curve

graphed in Fig. 46 below represents the cost curve
n=->x*+5x + 200,
where 77 per week is the total cost of an output of x tons of sugar
per week in a sugar refinery.
The parameters in cost functions of types such as those above
enable us to allow for shifts in cost. A cost curve given by (2), for
example, is fixed if a, b and c have given values. As the parameters
are changed, so the whole cost curve shifts in position corresponding
to some change in the conditions under which the firm operates. An
increase in c represents a rise in overhead costs and the cost curve
shifts upwards without change of form. Changes in a and 6, on the
other hand, correspond to modifications in variable cost and the
cost curve is changed in shape as well as in position.
At any output, the ratio of total cost to output defines average cost
or cost per unit of output. Denoting average cost by TT and using
Fig. 38, we have
f
and average read off the total cost curve as the gradient of the
cost is
line joining the point on the curve to the origin. Since average cost
varies with output, we have the average cost function TT /(#) where
f( x )= x
= ,
and a corresponding average cost curve referred to
x
axes Ox and OTT in a plane. The form of this function is to be
obtained from that of the total cost function. For example, if total
cost is given by the function (2) above, then
, c
7T=ax +h -f- .
x
Here TT->OO as x-^Q and as #->oo , i.e. the average cost curve must
fallas output increases at small outputs and rise at large outputs.
In the particular case of the sugar refinery already quoted, the
average cost curve is of the form shown by the curve A.c. of Fig. 46.
The average cost function, unlike the total cost function, is not
necessarily monotonic.
5.6 Other functions and curves in economic theory.

It is appropriate, in some problems, to represent the conditions
of supply of a good in a way analogous to those of demand. Suppose,
for example, that a market consists of individuals who bring fixed
stocks of various goods to a market-place for exchange amongst
themselves. If there is pure competition amongst the individuals,
the market prices (in money terms) of goods being given, then
all
each individual determines by how much he will increase or

decrease his stock of any one good. If he wishes to increase his
stock, he forms part of the demand for the good if he decreases his ;
stock, he contributes to the supply of the good. Then, if all prices

other than that of the good concerned are fixed, the total demand
and the total supply of the good are defined, by addition over the
whole market, as dependent upon the price. The demand and
supply functions are exactly similar, supply being negative demand,
and it can be taken that the functions are single-valued, continuous
and monotonic, demand increasing and supply decreasing as price
decreases.
A different representation of demand and supply is appropriate to
simple problems of international trade.* Two countries A and B
exchange two goods X and F, country A producing only X and
country B only 7. It is assumed that the conditions of production
and the supply of resources are given in each country. Then country
A can determine uniquely the amount of X it is willing to offer in
exchange for any given amount of Y, i.e. the amount of X offered
* See The Pure Theory
Marshall, of Foreign Trade (1879).
is a single-valued function of the amount of Y taken, a function
which appears as the demand and offer curve of country A (when
referred to axes Oxy in a plane). A similar function and curve is
defined for country B
the good Y being offered and
9
demanded. X
These curves differ from those previously defined in that they are
referred to axes along which amounts of two goods are measured.
The do not appear explicitly. But, if P is any
prices of the goods
point on the demand and offer curve of the country A, then the
gradient, referred to Ox, of the line OP gives the amount of Y taken

in exchange per unit of X, i.e. it gives the price of in terms X
of Y. A similar price can be defined, as a ratio of exchange, for the
country B.
We assumed, in 5.5 above, that only a single good was produced
by the given firm. A first approach to a more general problem can
be made by assuming that a firm produces two goods and Y underX
given technical conditions and making use of fixed supplies of certain
factors of production. The total cost of production is now given
and the interest lies entirely in the varying amounts of the two
goods that can be produced. If a given amount x of the good is X

produced, then the fixed resources of the firm can be adjusted so as to
produce the largest amount y of the other good Y compatible with

the given production of X. Here y is a single-valued function ofx
which can be taken as continuous and monotonic decreasing the ;
larger the amount of X

required, the smaller is the amount of Y
obtainable. Conversely, if x is the largest amount of X that can be
produced jointly with a given amount y of Y, then x is a single-
valued, continuous and decreasing function of y. The two functions
must be inverse to each other, i.e. two aspects of a single relation
between the amounts of X and Y produced, a relation imposed
by the condition of given resources. In symbols, we can write the
relation
giving y=f(%) and x

where/ and g are single-valued and decreasing functions to be inter-
preted in the way described. The relation, F(x, y) -0, can be called
the transformation function of the firm and it serves to show the
alternative productive possibilities of the given supplies of the
factors. The corresponding transformation curve in the plane Oxy
is by parallels to either axis in only single points and is downward

cut
sloping to both Ox and Oy* Further, it can be taken, in the
" "
normal case, that the production of Y decreases at an increasing
rate as the production of X is increased, and conversely. The trans-
formation curve is thus concave to the origin. Fig. 39 shows a
STEEL
20 40 60
Grade B, tons per week
FIG. 39.
"
normal " transformation curve in the hypothetical case of a firm
producing two grades of steel with given supplies of labour, raw
materials and equipment. The transformation curve, it should be
noticed, is a
"
maximal " concept and forms the outer boundary of
the productive possibilities. Any point within the curve corresponds
to productions of X
and Y possible with the given resources while
productions represented by points outside the curve cannot be
obtained no matter what adjustments of the given resources are
made.
The analysis can be generalised to allow for the use of given re-
sources in the production of different goods, not at the same time,
but in different periods of time. In the simplest problem, it is
assumed that a firm, with given technical conditions and given pro-
ductive resources, can arrange production in two years in various
* "
The curve
sometimes called a
is production indifference curve ", a
terminology which is not to be recommended since the curve has little in
common with the indifference curves described below. The term " trans-
formation curve" is used (e.g.) by Hayek, Utility Analysis and Interest,
Eaonnmir* .Tmirnnl.
ways. convenient here to take incomes rather than physical
It is
production, i.e. to take the resources of the firm as producing,

according to their allocation, various incomes this year and various
incomes next year. Then, for a given income of x this year, the
firm can determine the largest income y obtainable next year from
its resources. Conversely, the largest present income (#) can be
found if next year's income (y) is given. There is thus a transfor-
mation function relating incomes in the two years, a function very

similar to the transformation function described above and siibject,
" "
in the normal case, to similar conditions. The corresponding
transformation curve can be taken as downward sloping and concave
to the origin as the income desired this year increases, the income
;
obtainable next year decreases continuously and at an ever increasing

rate.*
5.7 Indifference curves for consumers' goods.

The economic problems to which we turn now are interpreted, not
in terms of a single function or curve, but with the aid of a function
type or a curve system. The first of these problems is concerned

with the choice of an individual consumer (e.g. a family) in respect
of the whole range of consumers' goods available on a market. In
order to obtain a simple representation, we take the case where only
two goods X and Y
appear in the consumer's budget and, to this
extent, the problem is over-simplified. The basic assumption now
is that the consumer distributes his expenditure on the two goods
" "
according to a definite scale of preferences ". His tastes ", on
this assumption, are such that he can arrange all possible purchases
of the goods in ascending order of preference and, given any two
alternative sets of purchases, he can either tell which purchases are
preferable or say that they are indifferent to him.

The precise expression of this assumption can be most easily given
first in Measuring purchases of the two goods
diagrammatic terms.
along axes Ox and Oy selected in a plane, any set of purchases (so
X
much of and so much of Y) is represented by a point P with co-
ordinates which measure the respective amounts of
(x, y) X and Y
purchased. If we start with a given set of purchases represented by
* This transformation curve is "
called, by Professor Irving Fisher, the in-
vestment opportunity line ". See Fisher, The Theory of Interest (1930),
pp. 264 et seq.
a point PQ(X O ?/ ), then all other possible purchases can be classed

,
into three groups, those preferable to, those which are less preferable
than, and those which are on a level of indifference with the basic
purchases (x y ). The third group of purchases is represented by
,
a set of points, P Pl9 P 2 ... making up a certain curve. This is

, > ,
called the indifference curve, corresponding to the level of prefer-

ence associated with the original purchases (x y ). Now take any ,
'
purchases (x Q y$) not included in the indifference group already

9
defined. A second group of indifferent purchases can be defined at

the level of preference of (XQ, y ') and represented by a set of points,
PO> PI P** > making up another indifference curve. This process
?
can be continued until possible sets of purchases have been in-

all
cluded, and the final result is a whole system of indifference curves,

each curve consisting of points at one level of indifference. Further,
by the basic assumption, the indifference curves themselves can be
arranged in ascending order of preference, i.e. according to the pre-
ferences shown by the individual for the various levels of indifference
attached to the curves. The indifference, map, consisting of the whole
set of indifference curves, serves to describe completely the indi-
two goods. To compare any two
vidual's scale of preferences for the
purchases as regards preference, we need only determine whether

" " "
the corresponding points lie on higher ", lower or the same
indifference curves.
The basic assumption tells us only that some indifference map
can be assumed. The continuity assumption is now added for
purposes of theoretical convenience. It is assumed, firstly, that the
individual can vary his purchases of each good continuously and,
secondly, that the variation from one set of purchases to indifferent

purchases is continuous. This assumption has, of course, no justifi-
cation except as an approximation and on grounds of expediency.
Each indifference curve can now be taken as continuous and the
whole set of curves appear in a continuously variable order of
ascending preference.
Finally, we add three assumptions limiting the form of the in-
" "
difference map in the normal case of consumer's choice. The
first isthe very reasonable assumption that no indifference curve
intersects itself or any other indifference curve. This means that
there is only one direction of variation from a given set of purchases
which leaves the consumer indifferent and that no set of piirchases
can be at two levels of indifference. The second assumption is that
the level of preference increases as we move across the indifference
map in the N.E. direction in the plane Oxy. This fixes the order
of ascending preference of indifference curves and implies that the
" "
consumer deems himself better off whenever he increases his
purchases of both goods. From this reasonable assumption, it follows
that the level of preference can only remain unchanged if an increase
in the purchase of one good is offset by a decrease in the purchase of
the other, i.e. the indifference curves are downward sloping. The
third assumption is that each indifference curve is convex to each axis
of reference, implying that ever larger and larger increases in the
purchase of one good are required to compensate (i.e. to preserve
a given level of indifference) for a steadily decreasing purchase of the
other good. It may be objected that this last assumption is not
reasonable as a universal, or even usual, condition. This may be so,
in which case the indifference curves must be assumed to be of other
forms. But the assumption can be retained to cover " normal "
cases until we find that our theory based on it fails to account for
observable facts.
Since the consumer's preferences for two goods are represented by
a system of curves, the analytical expression of the problem follows
from the considerations of 3.7 above. The indifference map has the
property that one and only one curve of the system passes through
each point of the positive quadrant of the plane Oxy. The complete
map corresponds, therefore, to a functional relation between x and y
involving one parameter a which can be separated off as shown in
the general expression :
</>(x, y)=a.
Here <f>(x, y) is a continuous expression in the amounts x and y of

the two goods purchased by the consumer. The variables x and y,
and the parameter a, take only positive values. If the value of a is
fixed, we obtain the equation of one definite indifference curve of the
system. Different indifference curves arise when the value of a is
changed. The parameter a is thus associated with the particular
level of preference which corresponds to the purchases x and y of the

"
two goods ;
it is an indicator " of the ordinal and non-measurable
"
concept of preference or utility ".
The following are actual examples of functional relations suitable

" "
for the expression of a normal scale of preferences for two con-
sumers' goods
(I)
:
- x
h - vy
-\-h
+k
=a, where 0<x<a(A - Jk) - h.
(2) (x+h)(y +k)â, where O

1
(3) x+y + \/'2xy ~-

a, where <# <;#.
Here a the parameter of each preference scale and h and k are
is
positive constants. Figs. 24, 25 and 26 show indifference maps

drawn from relations of the above types and illustrates the fact
" "
that the preference scales are of normal form.
It must be emphasised that the whole indifference map, given by
<f>(x, y)-â, is needed to describe the preferences of the consumer.
The parameter a is an essential part of the description. To allow for

changes in the consumer's tastes we must modify the form of the
whole indifference map. It is the function <f>(x, y) which reflects the
particular tastes of the consumer the form of (/> ;
is fixed for given
tastes and is modified whenever tastes change in any way. These
remarks can be illustrated by the relations (1) and (2) quoted above.
The constants h and k have fixed values for a given indifference map
but, by changing their values, we modify the whole map and allow
*
for changes in the complex of preferences of the consumer. Hence,

in representing a scale of preferences by a relation such as (1) or (2),
the variation of the parameter a gives the different curves of the
indifference map while the constants h and k, given in value for a
given complex of tastes, are changed only when we wish to consider
the results of varying tastes.*
5.8 Indifference curves for the flow of income over time.

A system of indifference curves, very similar in nature to that for
* It is
important to bo clear on tho use of parameters here. The parameters
of a demand or cost curve relate to external factors (e.g. the number of con-
sumers or the prices of the factors of production). In a given situation, there
is only a single demand or cost curve. As the situation changes, the parameters
take different values and the curve shifts in position. But a given situation as
regards consumers' preferences is described by a whole system of curves and
u parameter is needed to take us from one curve to another. If the situation
itself changes, we must add other parameters in order to change the whole
system of curves. The varying tastes of a consumer are described by the
shifting of a system of curves, i.e. by a system of a system of curves.
consumers' goods, can be defined to describe a rather wider economic
problem. An individual consumer can be assumed to act according
to a scale of preferences, not only for goods purchased in one period
of time, but also for goods bought in successive periods. But the
" "
discontinuous flow of purchases over time can be analysed on
exactly similar lines to purchases at one time. In order to obtain
a convenient representation of the wider problem, severe simplifying
assumptions must be made. We
assume that the consumer takes
into account purchases of goods made in only two successive years
and that his income can be taken as an indicator of his preferences
for the actual collection of purchases made in either year. We can
then speak of the consumer's preferences for various groupings of
this year's and next year's incomes, so much to spend now and so
much to spend next year.
On the basic assumption of a scale of preferences for this and next
year's income, the consumer's time preferences for income can be
represented by a system of indifference curves in a plane referred to
axes along which we measure the amounts of this year's income (x)
and of next year's income (y) One income grouping (so much this
t
year and so much next year) can be compared with another grouping
by the position of the corresponding points on the indifference map.
The groupings are indifferent to the consumer if the points lie on
one indifference curve one grouping is preferred to the other if it
;
" "
lies on a higher indifference curve. The income indifference map
corresponds to a function type of the form
0(a?, y)=a,
where t/t(x,
a given relation between x and y and where a is a
y) is
positive parameter indicating the level of preference for the incomes
x and y. The indifference map can be assumed to be continuous,
the curves being non-intersecting, downward sloping and convex to
the axes. Preference for income combinations, as indicated by a,
must increase as we move across the indifference map in the N.E.
direction. These limitations on the form of the indifference map are
" "
intended to apply to the normal case of a consumer's preferences
forincome over time and they are similar to the limitations imposed
in the previousproblem of purchases of goods made at one time.
Further properties of the indifference map for incomes can also be
assumed in " normal " cases. If neither income is large, the indiffer-
ence curves must be steeply inclined to Ox (the axis of present

income) since some of this year's income will only be sacrificed in
return for relatively large additions to next year's income. As the
combined income of the two years increases, the curves tend to
become less steep and, when both incomes are large, they approxi-
mate closely to straight lines inclined (negatively) at an angle of 45.
The addition to next year's income will then be little larger than the
loss to this year's income for which it compensates. It is seen that
the curve system of Fig. 24 is suitable for the representation of such
an income indifference map. The analytical expression is
h-*Jy+k
_,
where a is the parameter indicating the level of preference for the
incomes x and y in the two years. Fixed positive values must be
given to h and k if the consumer's preferences for income are known.
These values are only changed when the whole scale of income pre-
ferences is modified.
EXAMPLES V
Economic functions and curves
1. The price of a certain brand of tea is p pence per pound and it is found
that the market demand is x thousand pounds per week where
Represent these demands graphically and show that the demand law is
approximately of the linear form x = 120 - 2p. Then graph the total revenue
curve and find the largest revenue obtainable.
2. A cheap gramophone sells at p and it is known that the demand,
x hundred machines per year, is given by
Plot a graph of the demand curve. At what price does the demand tend to
vanish? Draw a graph of the total revenue curve and determine at what
price and output the total revenue is greatest.
3. A firm selling toothpaste investigates the form of the demand for its
product by fixing different prices in four successive periods. It is found that
the necessary outputs are
M.A.
Choosing suitable scales, plot the total revenue for these outputs graphically.
Show that total revenue can be taken approximately as a linear function of
output. Deduce the demand law.
4. What type of demand curve corresponds to the demand law
whore a and c are positive constants? Show that there is some demand 110
matter how largo tho price. Is there any limit to the oxtent of the demand
for small prices? Show that total revenue falls steadily as output rises and
compare with the case of the previous example.
5. Examine the demand curve p 9 where a and b aro positive

constants. Show that tho demand increases from zero to indefinitely large
amounts as tho price falls. What type of curve is tho total revenue curve ?
Show that revenue increases continuously to a limiting value and contrast this
case with that of the previous example.
6. From a consideration of the total revenue curves obtained in the

previous two examples, why can we say that these demand laws are not
typical of ordiiiaiy demands, except as approximations for limited ranges of
prices ?
7. Of what type is the demand curve p 4-c, whore a, b and c are

a
x -\-b
positive constants? If
j
is small, show that a relatively small fall in price
demand from zero to a large amount.

increases tho Is this a suitable approxi-
mate demand law for (e.g.) Player's cigarettes?
8. A speculative builder of working-class tenements rents each tenement
at p shillings per week. It is known that
p = "s/225 - 9#,
where x the number of tenements let per week. Graph this demand law.
is
Is it thokind of law you would expect hero ?
9. Find x as a function of p from the demand law p = Va - bx. Show that
tho demand curve is an arc of a parabola with its axis parallel to Ox. Locate
the vertex and indicate tho shape of tho curve.
10. The number (x) of persons per day using a motor coach service to
Southend is related to the fare ( p shillings) charged by the law
Show that the demand curve is a parabola and locate its vertex. Also graph
the total revenue curve, showing that revenue rises rapidly to a maximum
before falling off slowly. Generalise by considering the demand law
p = (a - foe) 2 , whore a and 6 are positive constants.
11. If x is the number (in hundreds) of business men travelling by the
8.15 a.m. train from Southend to London when the return fare is p shillings,
the demand law is
Draw the demand curve, show that it is a parabola and locate the vertex.
Also draw the total revenue curve and find at what fare the revenue is greatest.
Contrast the forms and economic interpretations of the demand curves in this
and the previous example.
12. At a charity performance at the local cinema, it is known that the
attendance at a uniform charge of p pence will be x = 6, where a and 6 are

constants. It found that the cinema, which has 3000 snafcs, is half-filled
is
when Is. is charged but that only one-sixth of the seats are empty at 9d.
Find the constants a anci 6. Deduce tho charge that fills the cinema and show,
if all the proceeds go to the charity, that the latter benefits most in this case.
constant " costs 11= ax + 6, show that the average cost

"
13. In the case of
curve is a rectangular hyperbola, average cost falling continuously towards
the value a as output increases. yV. builder of small bungalows has fixed
overheads " of 5000 per year and other costs are always 300 per bungalow.
"
Graph the curve showing average cost per bungalow when a variable number
x is built per year. \
14. A fixed plant is usod to manufacture radio sets and, if a; sets are turned
out per week, the total variable cost is (3.c + 2V# )'
- 2
Show that average
variable cost increases steadily with output.
15. If the overheads of tho plant of the previous example are 100 per week,
find the average cost in terms of the output and draw the average cost curve.
What is the least value of average cost ?
16. A coal retailer buys coal at a list price of 40 shillings per ton. He is
allowed a discount per ton directly proportional to the monthly purchase, the
discount being 1 shilling per ton when he takes 100 tons per month. His over-
heads are 50 per month. Obtain total and average cost as functions of his
monthly purchase (x tons) and draw the corresponding curves for purchases
up to 1000 tons per month.
17. The electricity works in a small town produces x thousand units per day
at a total cost of (2\/40:e - 175 + 90) per day. Draw the total cost curve.
Express the average cost, in pence per unit, as a function of x, draw the
average cost curve and show that, as output increases above a certain
minimum, average cost decreases rapidly.
18. Tho total variable cost of a monthly output of # tons by a firm producing
a valuable metal is (-&B S - 3# 2 + 50#) and tho fixed cost is 300 per month.
Draw tho average cost curve when cost includes (a) variable costs only, and
(6) all costs. Find the output for minimum average cost in each case. Show
that the output giving least average variable cost is less than that giving least
average (fixed and variable) cost.
19. A
firm produces an output of x tons of a certain product at a total
variable cost given by
II=ax 3 -bx 2 +cx.
Show that the average cost curve is a parabola, find tho output for least
average cost and the corresponding value of average cost.
20. A tobacco manufacturer produces x tons per day at a total cost of
x (x + 200)
4(a; + 100)'
Graph the and average cost curves and show that average cost decreases
total
continuously from 10s. per Ib. towards a lower limit of 5s. per Ib.
21. Generalise the previous example by taking the total cost as
X+b
U
rr
ax -- ,
X+C
where a, b and c arc positive constants (6 >c). Show that total cost increases,
and average cost decreases, as output increases. Of what type is the average
cost curve ?
22. If the tobacco manufacturer produces at a total cost of
when his output is x tons per day, draw the graphs of the total and average
cost curves. Describe the way in which total cost increases with output and
find the output at which average cost is least.
Generalise to show that the total cost function
Ilax --- +d
2
x c +
gives total and average cost curves of " normal " form.
23. On the market of a country town, butter is brought for sale from the
surrounding district weekly. It is found that the weekly supply (x Ibs.)
depends on the price (pd. per Ib.) according to
a = 100\/p~r 12 + 150.
From a graph of the supply curve, show that there is no supply at any price
less than Is. per Ib. and that the supply increases continuously as tho price
increases above Is. per Ib.) Generalise by considering x=a>Jp -6+c, where
a, b and c are positive, as a typical supply law.
24. English cloth is exchanged for German linen. Tho amount of cloth
(y million yards per year) offered by England in exchange for a given amount
(x million yards per year) of linen is given by
~~
2400
The amount of linen offered by Germany for a given amount of cloth is
~
10
Represent these conditions diagrammatically and show that, for equilibrium,
10 million yards of cloth exchange against 15 million yards of linen.
25. With a given plant and given supplies of labour, raw material, etc., a
" "
chocolate firm can produce two lines in various proportions. If x thousand
Ibs. of one lino are produced per year, then
( * <60)
thousand Ibs. per year is the output of the other line. Represent the con-
ditions graphically. What kind of curve is this transformation curve ?
26, Aman derives &x from his business this year and y next year. By
alternative uses of his resources, he can vary x and y according to
Graph the transformation curve. If he wants 750 next year, how much can
x2
he take out of tho business this year? by showing that y =6 --
Generalise
"
represents a normal " case in this type of problem if the constants a and b
are such that b>a. What is the transformation curve?
27. If a coal mine works x men per shift, the output per shift is
25 V
F2/
tons of coal. Draw a graph to show the way in which output varies with tho
number of men and find the size of the shift for maximum output. Express
the average product per man as a function of x, draw tho corresponding curve
and show that it is a parabola. When is average product greatest?
28.
Draw a graph showing

-
A consumer's indifference map for two goods X and Y is defined by
4 - *Jy -+
#-H4
1
_d
the five indifference curves for the values 2, 3, 4, 5, 6
" " form.
of the parameter a. Verify that they are of normal
29. Tho relation (x + l)(y f-2) 2a defines a system of curves. Draw a
graph showing the curves, in the positive quadrant, for a 2, 3, 4, 5, 6. Are
these curves suitable indifference curves for the case- whore an individual can
derive incomes of x arid ?/ in two years ?
30. Consider tho form of the indifference map given by tho rotation
where h and k are fixed positive constants. Draw graphically a selected

number of the curves in the case h=2 k 9 l, arid in the case h = 1, k 1 Hhow .
that one case is suitable for the indifference map for two consumers' goods
and tho other for incomes in two years.
CHAPTER VI
DERIVATIVES AND THEIR INTERPRETATION
6.1 Introduction.
HAVING made clear the nature of some of the fundamental ideas of

the mathematical technique, we find ourselves in a position to attack
one of the central problems of mathematics, the problem with which
"
we are concerned in the differential calculus ". If y is given as a
function of #, then the value of y is determined by, and changes with,
the value allotted to x. It is clearly important to devise a method
of comparing the changes in y with those in x from which they are
derived, of measuring the rate at which y changes when x changes .^
In the differential calculus, we make precise what we mean by the
rate of increase or decrease of a function, we set out a method of
evaluating these rates for various functions and we systematise the

problem by introducing convenient symbols, processes and rules.
In the words of Professor Whitehead, the differential calculus
"
is the systematic consideration of the rates of increase of
functions ".
In the present chapter, it is to treat the problem of
proposed first
defining and measuring from the purely analytical

rates of change
point of view. It is then found that a very important diagrammatic

interpretation of the problem can be given. Finally, since the im-
portance of any mathematical method depends on its use in inter-
preting scientific phenomena, we shall consider quite generally the
applicability of our new concepts in a wide range of scientific studies.
The development of the practical technique of dealing with our
problem will be given in the following chapter.
We can begin by considering special functions of simple form. If,
in the function y x z we increase the value of x from one definite
,
value #! to a larger value x 2 the value of the function increases from

,
x^ to # 2
3
The variable x is increased by amount (x 2 - x ) and the
-
DERIVATIVES AND THEIR INTERPRETATION 135
3 - 3
corresponding increase in the function is (# 2 a^ ). The average rate
of increase of the function per unit increase in # is then
/y
l>
3 _ "^/y 3
an expression which depends on both xt and x%. The average rate of

increase is dependent on the value of x from which we start and on
the amount of increase allotted to x.
Consider, now, only those increases which start from a definite

value of x. Adopting a different notation, let x be the fixed original
value of x and h the amount of the increase allotted. As x is
increased from x to (x +h), the function increases by an amount
3
{(x +h) -x*} and the average rate of increase is
h
which istaken as dependent on the variable increase h in x.
As the value of h becomes smaller, the average rate of increase
is seen to approach the limiting value 3x 2 which we term the
(instantaneous) rate of increase of the function at the point x.
Hence, by definition, the rate of increase of the function at
the point x is
Lim ~ .
"
The average rate of increase of the function is a perfectly definite

concept for any actual increase h in x however small. On the other
hand, the instantaneous rate of increase is a limiting and abstract
concept, the limiting value approached by the average rate of
increase as smaller and smaller increases in x are allotted. It will be
found that, although scientific phenomena are expressed directly
by means of average rates of increase, mathematical analysis finds

it more convenient to treat the related abstract concept of the
instantaneous rate of increase. The mathematically significant fact

is that a function istending to increase at a certain rate at a definite
point, not that it actually increases at a certain average rate over a
definite range of values of the variable.
The instantaneous rate of increase has a value for any value of x
we care to select. The rate of increase of x 3 is 3x 2 when we measure
So, the rate of increase of x at the point x = 1 is 3,
3
it at the point x.
at the point # 2 it is 12, and so on. In the first case, the value
of x 3 is tending to increase at the rate of 3 units per unit of x and
3
similarly for the other cases. The rate of increase of a: at the point
x changes as x changes, being always given by the expression 3# 2 .
The rate of increase of a function of x is itself a function of x.
As a second example, it is seen that the function y = - decreases

X
by an amount
(I
I
--- I
7
\
1 as x increases by
J an amount h from x to
\x x+hJ
(x + h). The average rate of decrease of the function is thus
_._ __
h \x x+hJ x(x+h)
The instantaneous rate of decrease of the function at the point x is
T im
Lim
.
r
h -+o
1/1----
* \x[
M!
%+h'
=
)
- o
x2
The function - decreases at various rates as x increases from different

x
i
values, the rate being always given by the expression x~
The process described above can be extended and generalised to
apply to any single-valued function whatever. Suppose the inde-
pendent variable of the function y=f(x) changes in value by an
amount h from x to (x +h). We can regard h as taking positive or
negative values, a positive value indicating that x has increased and
a negative value that x has decreased. The corresponding change
in the value of the function is then of amount {f(x +&) -/(#)} and
this, again, can be positive or negative according as the value of the

function has increased or decreased. The expression
f(x+h)-f(z)
'
then indicates the average rate of change in f(x) per unit change in x.
The sign of this expression is important. If the sign is positive, then
f(x) changes in the same sense as x itself, increasing as x increases and
decreasing as x decreases. If the sign is negative, then f(x) changes
in the opposite sense to x, increasing when x decreases and decreasing
when x increases.^The important question is whether the average
rate of change tends to approach any definite limiting value as h gets
smaller through positive and negative values. If a limit exists, then

it represents the instantaneous rate of change off(x) at the
point x,
i.e. the rate at which/(#) is to increase or decrease
tending (according
to the sign of the limit) as x changes from the point x. If no limit
exists, then the value of the function changes in an erratic way as
smaller and smaller changes are made in the variable x, changes
which cannot be described by any definite instantaneous rate of
change. The whole problem of the existence and value of the rate
of change of a function y f(x) thus turns on the existence and value
" "
of the limit of the incrementary ratio T{/(# + A) -/(#)} as h tends
to zero at a definite point x.
In the first example considered above, the rate of change of x*
at the point x is given by 3x* and, since this is positive, we know that
x 3 increases as x increases. On the other hand, the rate of change of
the second function y = - at the point x is

x
T . 1 / 1
T 1\ f 1 1
r\r= ~
.
}
Lun v 7--xl -Lim^
1
*
A-X) h\x+h
,
h_ Q [ x(x +A)J x*
Except for the sign, this is identical with the rate of decrease we
obtained previously. In fact, the negative sign of the rate indicates
that it is a rate of decrease as x increases and the numerical value,
i.e. , then measures the actual rate of decrease at any point. A

x2
rate of change, having sign as well as magnitude, is a wide concept
including both rates of increase and rates of decrease.
6.2 The definition of a derivative.

If y /(#) is a single- valued function of a continuous variable, and
if the incrementary ratio T{/(X +h) -f(x)} tends to a definite limit

fi
as the value of h tends to zero through positive or negative values,

then we say that the function has a derivative at the point x given
in value by the If the incrementary
limit of the incrementary ratio.
ratio has no limiting value, then the function has no derivative at
the point x.
A number of different notations for the derivative of a function

are in current use. The two following notations are the most common
and they be used throughout the present development.
will The
derivative of j/ f(x) at the point x is written either as
d d
or as y' =/'(*)
Since the function itself can be written as y or as f(x), each of these
notations appears in two equivalent forms. Hence :
DEFINITION : The derivative of the function y~f(x) with respect

to the variable x at the point x is
h
if the limit exists.
It is convenient, at this stage, to set out a number of observations

on the definition of the derivative :
(l)The concept of a derivative, from its very nature, applies
only to a function of a continuous variable. Further, the defini-

tion given here applies only to functions which are single-valued
and the extension of the concept to cover multi-valued functions is
postponed.
(2) The incrementary Tatio used in the definition is not defined
when h is actually zero in this case it assumes the meaningless form
;
of zero divided by zero. Thus the derivative cannot be regarded as

the ratio formed by putting A~0 in the incrementary ratio. The
derivative, in fact, the result of an operation performed on the
is
function, the operation of obtaining the limit of the incrementary
ratio. In the notation -

or -7- fix), the -;- must be regarded as an
ax ax ax
operational symbol applying to the function and its use is very
similar to that of the symbol / in the functional notation f(x). In
particular, it must be noted that the symbol -~ does not imply that
ax
the derivative is a ratio of one thing (dy) to another (dx).
(3) The incrementary ratio can be expressed in an alternative
"
form which is sometimes useful. Corresponding changes or incre-
"
ments in the variables x and y connected by the function can be
denoted by Ax and Ay respectively. The incrementary ratio is then
Ay and
-j
, the derivative appears as the limit of this ratio as Ax tends
to zero. Both Ax and Ay take finite values and, if a value is allotted

to Ax, the corresponding value of Ay is found from the function
itself.
(4)The value of the derivative must be expressed as so many units

of y per unit of x. If we change the units in which either variable
is measured, then we must also change the numerical value of the
derivative at any point. The rule for change of units, considering

the derivative as a quantity derived from x and yf(x), is given in
1.7 above. If an old unit of x equals A new units and an old unit of y
equals /*,
new units, then the new value of the derivative is
Â times
the old value. The fact that the derivative value depends, according
to the familiar proportionality rule, on the units of measurement of
both x and y must always be remembered.
(5) It is important to be clear about the use of the value x
that appears in the definition of the derivative. The derivative is
obtained by finding the limit of the incrementary ratio for a constant
value of The limit, once found, has a value dependent on the
x.
particular value of x selected, and different limiting values result

when different constant values are given* to x. Here we have
another example of a parameter x is a constant in finding the
:
limit (when h is the only variable) but parametrically variable from

one limit to another. The value of the derivative thus depends on
the value fixed for x, i.e. the derivative is itself a function of x. There
are, in fact, two points of view ;
the derivative may be obtained as
a definite value for a given x (the derivative at a given point) or it
may be taken as a function of x (the derivative at a variable point).
From the latter point of view, the notation f'(x) is particularly
" "
appropriate and an alternative term derived function can be
used. From the original function /(#), we derive a second function
f'(x) by the process of finding derivatives of f(x) for all the various
possible values of x.
(6) It must not be assumed that the derivative of /(#) necessarily
exists at any point. In fact, the derivative cannot be used unless
we have determined that the incrementary ratio has a limit at the
point in question. A function may have a derivative at some points
and not at others some functions may have a derivative at a point
;
and others not. It will be seen (6.5 below) that the non-existence
of a derivative is exceptional and of very special significance.
(7) The derivative notations we have given apply to any function
y~f(x). The notations must be suitably modified when a particular

3
form is given for a function. So, -r-(# )
and ^- (-) denote the
CvX _ CvX \X/
derivatives of the particular functions x 3 and - respectively. The
notation also needs modification when we wish to write the value of

the derivative, not at a general point x, but at some definite point.
The following examples illustrate the flexibility of the notation.
and
denote the derivatives off(x) at the particular points and x x0

respectively. More generally, the derivative off(x) at x = a is
For a particular function, say y=x*, the notations
and
[/(*')]
Ufa
v
'Jô
,
[^3)1
Ldx^ "J^j [(*)
Lefar
'
indicate the derivative value at x 0, x J and x~ a respectively.

(8) Finally, it is scarcely necessary to note how convenient it is to
have a special notation and terminology for the derivative concept.
Limits of the kind indicated by the derivative are of frequent appear-
ance in mathematical theory and applications. The short and
flexible derivative notation, in replacing the clumsy limit notation,
makes it possible to deal with these limits with great ease. In
particular, the process of systematising the operation of finding the
limit, to be discussed in the next chapter, is facilitated. Without
this labour-saving device, scarcely any progress is possible.
6.S Examples of the evaluation of derivatives.

The considerations of 6.1 show that
-=~- \x I *jX and -= \ /

- *
ax ax \x/ x2
The following examples are further illustrations of the way in which
derivatives of particular functions are obtained from the definition :
Ex. 1. The derivative of (2#-l) at any point x is obtained as the

limit as h tends to zero of the ratio
__
~~
A
So ~(2x-l) = 2.
dor
The derivative, in this case, is a constant. A similar result holds in the

case of the derivative of any linear function :
d .
-j-(ax+b)-=a.
ax
Ex. 2. The incrementary ratio for the function (a;

2
+ 3x - 2) is
l[{x+h)* + 3(x+h)-2}-(x*+3x-2)]
=l[(x* +2hx + A 2 ) + (3x + 3h) -2 - (a* +3x -2)]

it
2hx+3h+hz
as A->0.
So t(
In the same way, the derivative of the general quadratic function is
2
-=-
(ax +bx+c) =2ax+b.
2x +
Ex. 3. For the function -- 1
,
the incrementary ratio is
X JL
1 +1 2x + - + 2h + 1) - + 1) (x +h -
{2(x +h) (2x (2x 1)
1] = (a? 1)
l x-l) h(x-l)(x-
-3
So __ _ 2
(a;-!)
The result, as given, holds for any value of x not equal to 1 . If x = 1 the ,
function itself has an infinite discontinuity and the question of the
existence of the derivative at this point needs further consideration (see
6.5 below).
6.4 Derivatives and approximate values.

The
analytical interpretation of the derivative as the measure of
the rate of change of the function is clear from 6.1 above. Further,
since the rate of change is the limiting value of the average rate of
change over a small interval, the derivative is also an approximate

value of the average rate of change, the smaller the interval of the
average rate the closer being the approximation. Hence :
If f(x) has a derivative /'(#) at the value x, then /'(#) measures

the rate of change of f(x) at the point x, and is an approximate
measure of the average rate of change of f(x) for any small interval
from the point x.
As already noted, the

rate of change can be a rate of increase or a
rate of decrease according to the sign of the derivative at the point.
In either case, the derivative as the rate of change must be expressed
as so many units of the function per unit of x. Two units of measure-
ment are involved in the derivative and, if either unit is changed, so
isthe expression of the derivative.
The use of the derivative as the approximate value of the incre-
mentary ratio or average rate of change is an important one. If
Ax is any small increment in the value of x, and if Ay is the corre-
sponding small increment in the value of yf(x), then the deriva-
is the limiting and approximate value of the quotient of
tive f'(x)
Ay divided by Ax. Hence

Ayf'(x)Ax approximately when Ax is small.
In other words, if we multiply a small increment in x by the rate of
change /'(#), we obtain an approximate value of the corresponding
small increment in the function y~f(x).
Again, with a slightly different notation, the value of the derivative
f'(a) at the point x=a is an approximate value of the expression
r {/(a 4-A) -/()} for a small value of h, i.e. x
fi
f(a +h) =/(a) +hf'(a) approximately when h is small.

" "
This result enables us to trace the effect of small errors or devia-
DERIVATIVES AND THEIR INTERPRETATION 14^
tions in the variable x. The value of x

is erroneously taken as d
"
instead of the correct value (a +A), where h is a small error ".
" "
The above result shows that the error in the value of the function,
resulting from the insertion of a instead of (a 4- A), can be taken
approximately as hf'(a). For example, the volume of a sphere of
radius x inches is f TTX* cubic inches. If the radius of a given sphere is
estimated as a inches, then the estimate of the volume is |7ra 3 cubic
inches. The error involved in the volume estimate is approximately
,
2
4:7ra h cubic inches, where h is the small error in the original estimate
of the radius, f -7- (f 7r#

3
)
= 4?r^ 2 .
j
It is often convenient, for a definite small range of values of x, to
replace the value of a given function f(x) by an approximate linear

expression in x.* The result we have given enables us to do this. If
the approximation is to serve for values of a: near x a, we substitute
h x - a in our result and write
f(x)=--f(a)+(x-a)f(a)
approximately for small values of (x -a). For example, if
.,
. 2x+l , tl
. 3
/(*)= and /(*)=-(ar-ir
r
then /(3)=| and /'(3)=-|.
So =f+(*-3)(-f),
--
-
approximately when (x 3) is small. In general, therefore, it is
possible to represent any given function by an approximate linear

expression, f(x)~a +/8#, for a small range of values of x. The
constants a and j8 are determined in the way described.
6.5 Derivatives and tangents to curves.

One
of the important problems of geometry is that of finding the
tangent at any point on a given curve. The tangent describes the

direction of the curve at the point and, to define it, we must introduce
the idea of a limiting process. The line joining any two points P and
* und Rente
See, for example, Wicksell, Uber Wert, Kapital (1893, reprinted
1933), pp. 31-2.
Q on a curve is called a chord of the curve and its position varies as
either P or Q is varied. Suppose that P is fixed and that Q moves
along the curve and approaches P from one side or the other. It is
clear that it is usually, but not always, the case that the chord PQ
tends to take up a definite limiting position PT as Q approaches P.
Fig. 40 illustrates a case where this happens. It is possible, however,
X
FIG. 40.
that no such limiting position PT

exists at certain points, e.g. at
points where the curve is discontinuous. If the limiting position

PT of the chord exists, then it is called the tangent to the curve at
P. If no limiting position exists, then the curve has no tangent
at P.
The instantaneous direction of the curve at any point, i.e. the
direction in which the curve
tending to rise or fall, is indicated by
is
the tangent to the curve at the point. A convenient measure of the

direction of the curve is thus provided by the gradient of the tangent.
This gradient varies as the tangent varies from point to point on the
curve and is always given as the limiting value of the gradient of any
chord PQ as Q approaches P.
Our discussion is now limited to the case of a curve represented by
a single-valued function y=f(x), there being only one point on the
curve above or below each point on the horizontal axis Ox. Let P
and Q be two points on the curve with abscissae x and (x -f h) respec-
tively, x being fixed and h any positive or negative amount. The
gradient of the chord PQ is the ratio of the difference between the

ordinates of Q and P
to the difference between the abscissae of these
points (see 3.3 above). So
Gradient of
The gradient of the tangent at P is the limiting value (if it exists) of

this expression as Q approaches P, i.e. as A->0. Hence
Gradient of tangent at P^Lim ^ x + ^ ~^X- -=f'(x).

The problem of finding the tangent and its gradient at any point on
a curve thus equivalent to the problem of finding the derivative

is
of the function yf(x) which represents the curve. So
The tangent to the curve yf(x) at the point with abscissa x

exists if the function has a derivative at the point x and the tangent
gradient =f'(x).
From a straight line through a fixed point

3.3 above, P (xly yt ) on
the curve y ~f(x) and with gradient m has equation
y-y =m(x-x
1 l ).
This line is the tangent at P if the gradient m is equal to/'(#i). So
The equation of the tangent to the curve y -f(x) at the point with
co-ordinates (a?!, y) is y
- yl -
=f'(xl (x xt ). )
Since the point P

lies on the curve, we have yl /(xj). The tangent
equation thus involves only one parameter, i.e. the abscissa xl
of the selected point on the curve.
Taking the parabola yx 2
as an example, we can show that
2 2x at the point
T- (x ) 2x, i.e. the tangent gradient of the parabola is
CLX
with abscissa x. The tangent equation at (xl9 yt ) is
y - j/! = 2xt (x - xx ) where y^ = xf,

y ZxyX xf +2x^ = 0,
i.e.
- -
i.e. y-2xlx+xl 2 =Q.
For example, the point (1,1) on the parabola 2
the tangent
lies yx ,
at the point has gradient 2 and the equation of the tangent is

y - 2x +1 =0. The tangent gradient and equation at any point on
K A M.A.
a given rectangular hyperbola can be found in an exactly similar
way. It is only necessary to evaluate the derivative of the function
which represents the curve.
The diagrammatic interpretation of the derivative as a tangent
gradient is a very important one and much use of it will be made
later when we come to the applications of the derivative in such
problems as that of maxima and minima. In the meantime, it can
be used to supplement points already discussed.
It has been seen that the process of derivation enables us to
derive, from a given function y~f(x), a second function yf'(x).

The curves representing these two functions can be plotted with
reference to the same axes and the connection between them is
interesting. The ordinate of the derived curve yf'(x) at any point

is equal to the tangent gradient at the corresponding point on the
original curve y^f(x). The original curve shows the way in which
X
FIG. 41.
the value of the function f(x) changes the derived curve shows the
;
variation of the rate of change of the function. For example, the

curve y = ^x 2 and the corresponding derived curve y \x are shown
in Fig. 41. The curve shows that x 2 increases at an increasing
first
rate as x increases. The second curve, which is a straight line, shows

that the rate of increase of x 2 actually increases at a steady rate.
The nature of the approximation involved in writing f'(x)Ax for
the increment in the function f(x) is clearly seen in Fig. 42. If P is
the point (x, y) on the curve y=f(x) and Q the point with abscissa
x+Ax, then Ax=LMPN. The increment in the function is

NT
Ay=MQ - LP=NQ. Now, f'(x) ^gradient of the tangent PT = ,
i.e. f(x)Ax=NT. The approximation of f'(x)Ax to Ay is thus

represented by the approximation of NT to NQ. The difference
between these lengths clearly decreases as Ax decreases and Q
approaches P.
It is obvious that there is some
between the continuity of
relation
a function and curve and the existence of a derivative and tangent.
It can be shown, in fact, that the existence of a derivative and
tangent is a more severe restriction on the function and curve than

the condition of continuity Of the two continuous curves of Fig. 33,
.
the first has a tangent at all points while the second has one point
(a sharp point) where no tangent exists. The fact that a curve is
continuous does not imply that a tangent exists at all points. A
continuous curve can be
drawn without taking pencil y\
from paper ;
a curve which
has tangents at all points is 2
not only continuous but also

" "
smooth in the ordinary
sense of this term. All 1-
the ordinary functions of

mathematical analysis are
continuous and have deriva-
tives (except perhaps at
very special points) and FIG. 43.
they are represented by

continuous and smooth curves. Further, when we introduce a
function such as a demand or cost function, it is usual to assume, not
only that it is continuous, but also that it has a derivative, excluding
the case where the corresponding curve has a sharp point.
Finally, nothing has been said so far about the possibility that the
incrementary ratio defining the derivative has an infinite limit. If
j;{f(
x as & point where f(x) exists, we say
" "
that the derivative/ '(x) is infinite at the point. In diagrammatic
terms, there is a definite tangent to the curve yf(x) at the point P
with abscissa x and the gradient of the tangent is infinite, i.e. the
tangent is parallel to Oy. The case is illustrated by the curve
y l + Ux - 1, shown in Fig. 43, which has a vertical tangent at the
point where x 1 Infinite derivatives can be obtained even in cases
.
of ordinary continuous and smooth curves. Further, if /(#) is itself

infinite at the point in question, we can still say, in many cases, that
the derivative exists and is infinite at the point. The rectangular
hyperbola of Fig. 33 is a case in point. Both the function -

2x
X~
-fl
-~ and
its derivative -- -
(x
3
1)
are infinite at x = l. The curve " goes off
" "
to infinity at this point and has a vertical tangent at infinity ",
i.e. the asymptote x 1. To deal with many infinite discontinuities,
therefore, we have only to allow for the possibility of both the
function and its derivative being infinite.
6.6 Second and higher order derivatives.

The derivative of a function of a variable x is also a function of x.
It is apply the process of derivation all over
possible, therefore, to
again to the derivative and to obtain what is called the second de-
rivative of the original function. The second derivative, if it exists,
can be obtained for various values of x and again a function of x. is
We can thus proceed to find the derivative of the second derivative,

i.e. the third derivative of the original function. In the same way,
derivatives of the fourth, fifth and higher orders can be obtained in
succession. Extending the notations already adopted for the de-
rivative, the sequence of successive derivatives of a given function
yf(x) can be written
dy
~
d*y d*y .
d_ d^ d^
or y, t/', t,", </", ... ; f(x), /'(*), /"(*), /'"(a), ....

The sequence of derivatives can be extended indefinitely, with the
sole limitation that the derivatives at and beyond a certain order
may fail to exist on account of the breakdown of the limiting process
which defines them. For all ordinary mathematical functions, how-
ever, the derivatives of all orders exist.* But it is found that the
first and second derivatives are sufficient for the description of the
main properties of functions, and, in the applications of mathe-

* It often
happens, of course, that the derivatives beyond a certain stage
2
become zero, as in the case of the function y =aj This is not the same thing
.
as saying the derivatives do not exist.

DERIVATIVES AND THEIR INtERPRETATION 149
matical analysis, derivatives of higher order than the second are

seldom needed.
The analytical interpretation of the second derivative is easily
provided. The value of /'(#) measures the rate of change of f(x) as
x increases. The value of /"(#), being the derivative of f'(x), thus
indicates the rate of change of /'(#)> i- e the rate of change of the rate
-
of change of f(x), as x increases. For example, the function x 2 in-

" "
creases at an increasing rate as x increases and this acceleration in
the value of x z is measured by the second derivative of the function.
The value of /"(#)> it should be noticed, must be expressed in so
many units off'(x) per unit of x, i.e. in so many units off(x) per unit
of x, per unit of x. As in the case of the first derivative, a change in
the units of measurement necessitate a change in the value of the
second derivative.
In diagrammatic terms, the value of f"(x) at any point measures
the tangent gradient of the derived curve yf'(x) at the point. A
more useful interpretation, however, connects the second derivative
with the curvature of the original curve y =f(x), curvature being
defined by means of the
rate of change of the tangent gradient as we
move along the curve. This connection will be investigated at a
somewhat later stage (see 8.3 below).
6.7 The application of derivatives in the natural sciences,

In the interpretation and explanation of scientific phenomena, we
attempt to throw into prominence the causal or other relations
between the concrete quantities or attributes with which we deal.
Mathematics is designed as a tool to aid us in this attempt. Now, a
glance at the problems of the natural or social sciences shows that
they are largely, but not exclusively, concerned with the question of
change, with the way in which the variable quantities change together
or in opposition. The practical method of expressing change is
by means of the average rate of change of one quantity as another

quantity varies. Hence, when we come to express any assumed con-
nection between quantities by means of a functional and mathe-
matical relation between variable numbers, we find that our main
concern is with the average rates of change of these functions.
These remarks apply even if discontinuous variables and functions

are used to represent the necessarily discontinuous quantities of
scientificphenomena. But it is found mathematically convenient
to assume, wherever possible, that the variables and functions are
continuous and then to deal with instantaneous rates of change as
the limiting and approximate values of average rates of change.
Sometimes it is necessary to retain discontinuous variables, but the
theory then becomes mathematically much more difficult and much
less elegant: We are led, therefore, to the study of the derivatives
of continuous functions and their use in the interpretation of the
scientific phenomena under consideration.
The differential calculus is thus of almost immediate application
in all scientific studies. Any relation between variable quantities is
represented by a function, and the derivative of the function measures
the instantaneous rate of growth of one quantity with respect to the
other and approximates to the actual average rate of growth when
the variation considered is small.
Examples of the use of derivatives can be provided from most
scientific theories. In the science of physics, derivatives are em-
ployed to express physical movement of any kind. As a simple
example, consider a train moving along a railroad track. The dis-
tance travelled by the train from a fixed point depends on the time
that has elapsed, i.e. the distance travelled is a function of the
time-interval.* All that can be actually observed about the train's
motion is its average speed (over some time-interval, long or short)
" "
and this can be connected with the incrementary ratio of the
distance function. But, from the observable concept of average
speed, we
derive the limiting concept of the velocity of the train at
a certain time. Velocity is simply the limiting value of average
speed as the time-interval of the latter is made shorter. Hence, we
need the derivative of the distance function to measure the velocity
of the train at any moment and to approximate to the average speed
of the train over any short time-interval from the given moment.
The importance of the limiting idea involved in the derivative is
seenby the way in which we have come to express physical motion.
" "
Velocities are not natural concepts ; they can be derived only
by a process of abstraction, of finding limiting values. When we say
* For
example, if x is the distance travelled in time t, then x at represents
motion at a uniform speed and xat 2 motion at a uniform acceleration, a
being a constant in each case.
that a train travelling at 60 m.p.h., we are expressing a rather

is
complicated idea. We
do not mean that the train will actually
travel 60 miles during the following hour, one mile during the next
minute or even 88 feet during the next second. The speed of the
train will generally change during any time-interval however short.
The statement means that the distance travelled by the train per
unit of time will correspond more and more closely to 60 m.p.h., or
any equivalent, as the time-interval is made shorter and shorter. In
physics, therefore, we have accepted the mathematically significant
limiting concept of velocity as replacing the practical concept of
average speed.
The second derivative of distance travelled as a function of time
is the derivative of the first derivative and so measures the rate of
change of velocity over time. The rate of change of velocity is
commonly described as the acceleration and this isagain a limiting

concept, the limiting rate at which velocity is changing. Accelera-
tion is, in fact, another concept to which we have grown accustomed
in the description of movement and its abstract nature has ceased to
be a difficulty.
The measurement of both distance and time must appear
units of
in the derivatives which measure velocity and acceleration, a fact
which is clear from the way in which velocities and accelerations are
specified. We say, for example, that the velocity of a train is 44 feet
per second (30 m.p.h.) and that its acceleration is 2 feet per second
per second at a given moment. This means that the train is tending
to cover distance at a rate proportional to 44 feet in one second and
that its tending to increase at a rate equivalent to 2 addi-
velocity is
tional feet per second in a second. In the same way, when we say
that the constant acceleration of a body moving under gravity is
32 feet per second per second, we imply that the velocity of the body
is tending to increase always by 32 feet per second for every second
that elapses.
In other branches of mechanics or physics, the derivatives of
functions are used to interpret motion, whether the motion is of
liquids in hydro-mechanics, of air and gases in acoustics and the
theory of sound, of electric currents in electro-mechanics or of heat
" "
in thermo-mechanics. Further, in a less exact science such as
biology, derivatives again have their uses. For example, the

changes in the proportion of dominant to recessive genes through
successive generations can be interpreted, in a broad view, as the
derivative of the proportion as a function of time.*
6.8 The application of derivatives in economic theory.

" "
Finally, we come to economics as the most exact of the social
sciences. We have seen that a relation assumed between economic
variables, e.g. the relation between demand and price, can be ex-
pressed by means of a function and curve. It is usually assumed
that the function and curve are continuous and smooth, i.e. the
function has a derivative and the curve a tangent at all points. The
derivative of the function and the tangent gradient of the curve must
now be considered in some detail and their economic meaning made
clear.
We are accustomed, in economics, to describe the variation of one

quantity Y with respect to another quantity X by means of two
concepts, an average concept and a marginal concept. The average
concept expresses the variation of Y over a whole range of values of
X, usually the range from zero up to a certain selected value. Thus
average cost is the relation of total cost to the whole of the output
concerned. Marginal concepts, on the other hand, concern the
"
variation of Y on the margin ", i.e. for very small variations of X
from a given value. So the marginal cost at a certain level of output is
the change in cost that results when output is increased by a very small
unit amount from this level. It is clear that a marginal concept is
only precise when it is considered in the limiting sense, as the varia-
tions in Xare made smaller and smaller. It is then to be interpreted
by means of the derivative of the function which relates X and Y.

Average and marginal concepts are not new or peculiar to economics.
In describing the motion of a train, the average concept is the
average speed of the train over (e.g.) its first hour's run while the
marginal concept is the velocity of the train at the end (e.g.) of the
hour's run.
Some actual examples can be taken to illustrate the average and
marginal concepts. We
have seen that the demand for a good
can be expressed, under certain conditions, by the relation p~^(x) 9
giving price as a continuous and decreasing function of the demand.

* See The Causes of Evolution
Haldane, (1932), Appendix.
But the price p can be regarded as the average revenue obtained from
the demand x, i.e. total revenue divided by the amount demanded
or produced. The ordinary demand function and curve are thus
equivalent to the average revenue function and curve. Total revenue
Rxift(x) is also a continuous function of demand. From the total
revenue curve, we read off average revenue (or price) as the gradient
of the line OP joining to the appropriate point P on the curve.
If output is increased by a small amount Ax from a certain level x,
suppose that total revenue is found to change by an amount AR.

There is an increase or a decrease in revenue according to the sign
of AR. The added revenue per unit of added output is then the
" "
ratio of AR to Ax, i.e. the average revenue for outputs from
x to x -f Ax. As the output change becomes smaller, we obtain the
rate of change of revenue on the margin of the output x as the limit
An
of as Ax->0. This is termed the marginal revenue at the output x
-p
and it is measured by the derivative of R as a function of x :
Marginal revenue = -^ = -^-{xibtx)}.

T
dx dx
Marginal revenue is thus an abstract concept only definable for
continuous variations in revenue and output. But it is always
approximately equal to the added revenue obtained from a small unit
increase in output from the level #.* In diagrammatic terms, the
marginal revenue at an output x is measured by the tangent gradient

to the total revenue curve at the point with abscissa x.
Marginal revenue is itself a function of the output x, its value

depending on the particular margin of output considered. We can
thus draw the marginal revenue curve to show the variation in
marginal revenue as output increases. As in the cases of the total
and average revenue curves, the outputs are measured along the
horizontal axis. Since marginal revenue is the tangent gradient of
the total revenue curve, the form of the marginal revenue curve can
be deduced at once from that of the total revenue curve.
* The term "
marginal "increment of revenue ", which is sometimes used, is
not to be confused with marginal revenue ". The former is the increment
AR in revenue resulting from any increase Ax in output from the level x. Now
Tr>
AR = Ax approximately when Ax is small. Hence the marginal increment
dx
of revenue for a small increase in output is approximately equal to the mar-
ginal revenue times the increment in output.
Two particular types of demand law can be taken for purposes
of illustration. The linear demand law gives pa-bx
R=ax-bx 2
and ~j-=a-2bx (see 6.3, Ex. 2).
dx
The total revenue curve a parabola with axis pointing vertically
is
downwards. The average and marginal revenue curves are straight

lines sloping downwards and the gradient (referred to Ox) of the
latter is twice that of the former. Fig. 44 shows the three curves in
the particular case where the demand law for sugar is p~ 15- \x
(see 5.3 above). Here, as in all cases, it is found convenient to plot
SUGAR
25 50 75 x
Mn. Ibs. per month M.RN
FIG. 44.
the average revenue (or demand) curve on the same graph, and
same axes and scales, as the marginal revenue curve.
referred to the
It is to be noticed that the marginal revenue curve cuts the axis Ox
(marginal revenue zero) at the same output at which total revenue

is This is a general property which follows since, at this
greatest.
output, the tangent to the total revenue curve is horizontal with
zero gradient.
From the demand law p r - c, we derive

x +6
ax dB ab
En r-cx and -r =
"" ~"~
,
-c (see 7.4, Ex. 15).
-
~x +6 dx (x+b) 2
In Fig. 45 are shown the total, average and marginal revenue curves
obtained in the particular case where the demand law for sugar is
600
*p
x s?L-5 (see 5.3 above).
-{-20
Again the marginal revenue curve
falls continuously, lies under the average revenue (or demand) curve
and cuts Ox at the output where the total revenue is greatest.
The three concepts of total, average and marginal cost can be

defined, under the conditions of the cost problem of 5.5 above, in
a very similar way. The total and average cost functions and
curves have been sufficiently described already, average cost being
read off the total cost curve as the gradient of the line OP joining
to the point P with the appropriate abscissa on the total cost curve.
SUGAR
25 50 75
Mn. Ibs. per month
FIG. 45.
If output is increased by an amount Ax from a certain level x and if

the corresponding increase in cost is J77, then the increase in cost per
unit increase in output is .

Marginal cost is defined as the limiting
~j
value of this ratio as Ax gets smaller, i.e. marginal cost is the deriva-
tive of the total cost function IJ=F(x). It measures the rate of
increase of total cost and approximates to the cost of a small addi-
tional unit of output from the given level. Further, the marginal
cost ofany output is given as the tangent gradient of the total cost
curve at the appropriate point. Since marginal cost varies with the
output at which it is measured, we have a marginal cost function and
a corresponding marginal cost curve.*
If the total cost function is II~ax 2 +bx +c, then
Average cost TT = ax -f b -h - and Marginal cost -= = 2ax + b.

x ax
The total cost curve is the rising portion of a parabola with axis
pointing vertically upwards. The average cost curve is found to be
U-shaped and the marginal cost curve is a straight line sloping
*
Marginal cost is described as, or equated to, " marginal supply price "
by some writers.
See, for example, Bowley, The Mathematical Groundwork
of Economics (1924), p. 34.
upwards with gradient 2a. The three curves are shown in Fig. 46 in
the particular case where /7=T^c 2 +5x +200 is the cost relation for
a sugar refinery (see 5.5 above). It is to be noticed that the marginal
cost line passes through the lowest point of the average cost curve
at an output OM where the tangent to the total cost curve goes
through the origin. At such an output, average cost is smallest and
equal to marginal cost.
n
25 50 25 50
7*0/75 per week Tons per weeh
FIG. 46.
Average and marginal concepts can be usefully defined in other

economic problems. The transformation problems of 5.6 above
provide examples. If P is a point on the transformation curve for
the production of two goods X and Y with given resources, then the
co-ordinates of P
represent a possible production. The gradient of
OP (referred to Ox) is the average output of Y per unit output of X,
the amount of Y produced as a proportion of the amount of X

produced. The tangent gradient of the transformation curve at P
(referred to Ox),i.e. the derivative of y with respect to x, is negative
and numerical value measures the marginal rate of substitution

its
of Y production for X production, the rate at which the production

of Y is increased when the production of X is decreased. In the same
way, the tangent gradient of the transformation curve for incomes
in two years is used to define what Professor Irving Fisher calls
the marginal rate of return over cost.* Marginal concepts are also
fundamental in the use of indifference curves for choice between two
" "
goods or two incomes. The marginal rate of substitution of one
good for another in consumption is defined as the numerical value
*
See 9.7 below and Examples VI. 31.
of the tangent gradient at the appropriate point on an indifference

curve for the two goods. Again, the numerical value of the tangent
gradient of an indifference curve for two incomes is used to define
" "
the marginal rate of time preference for present over future
income in the case of the individual concerned. These concepts will
be considered in detail at a later stage.
EXAMPLES VI
Evaluation and interpretation of derivatives *
1. Find, from the definition, the derivatives of x 2 and (I +x)

z
.
2. Show that the derivative of a constant is zero and that the derivative
of x is unity for any value of x. Interpret these results in terms of rates of
change.
3. Use the definition of the derivative to establish the general result that
the derivative of (ax z + bx + c) is (2ax + b).
4. Evaluate (x(l +x)} and -j-{x(l -x)}

from the definition. Show that
ax ax
these derivatives are respectively the sum and the difference of the derivatives
of a; and x 2 Does this suggest a general rule?
.
From the definition, show that - = -^ and hence

5.
verify the derivative of

-
2a?
x-
+-i
1
-r- (
dx \a 2x + 6 2 /
r^ )
(a 2x
T^T;2
+ 62 )
6. Show that the derivative of a stop -function is zero except at certain

points where it does not exist at all. Illustrate the statement that a function
has no derivative at a point of discontinuity.
7. Find the average rate of increase of x 2 when x increases from 1 to 1*1.
What percentage error is involved when the derivative of x* at x 1 is used to
represent this average rate of increase ?

8. Find the average rate of decrease of
2x + 1
--
r- as x increases from 2 to each
x 1
of the values 3, 2-5, 2*1, 2-05 and 2-01. Find, from the derivative, the in-
stantaneous rate of decrease at x = 2 and compare with each of the average
rates.
9. Evaluate the derivatives of

la;-
and Show that
1
= 3 - 2x and
- 7 .
-
-
=i(o? + 1) approximately when x is nearly equal toWhat percentage
1.
errors are involved in the use of these approximations when x = 1 1 ?

10. If x is small, show that f(x) =/(0) + xf'(Q) approximately. Find the
derivative of r
1
- +x
and show that this expression can be represented approxi-
mately by (1 -a;) when x

small. Express the difference
is between the ex-
pression and its approximation in terms of x.
* Further
examples on the interpretation of derivatives are given in
Examples VII, 24-37 below.
11. Show that/(a -l-#) =f(a) +xf'(a) approximately when x is small. From
the derivatives of x* and -
, show that
x2
<!+*)' .= 1+3* and
approximately when x is small.

3
Check the first result by multiplying (1 +#)
out as a cubic expression in x. How good is the second approximation when
a=0-5?
12. Use the derivative of x to show that the approximate increase in the
z
area of a circle is 2nah square inches when the radius is increased by a small
amount h from the value a. Calculate the actual increase in area when the
radius increases from 10 to 10-1 inches and compare with the approximate
value given by the above formula.
13. A circular ink-blot grows at the rate of 2 square inches per second.
Show that the radius is increasing at the rate of inches per second at the
7rr
time when its length is r inches.
14. surface. area and volume of a sphere of radius r are S = 47rr 2 and
The
V =f 77T 3 Find approximate expressions for the increases A$ and AK when
.
the radius is increased from r by an amount Ar. Deduce that the propor-
tionate increase in S is approximately twice, and in V approximately three
times, the proportionate increase in radius.
15. The pressure of a gas at constant temperature is p Ibs. per square inch
and the volume v cubic inches where pv 10. Find the rate of decrease of
volume as pressure increases and the approximate decrease in volume when
the pressure is increased from p by an amount Ap.
At what point on the parabola y x 2 does the tangent make equal

16.
2
angles with the axes? Use the derivative of (1 +x) to obtain the equation of
the tangent at (0, 1) on the parabola y =x z 2x + I. -f-
17. the equation of the tangent to the parabola y=x 2 + 3x -2 at

What is
the point with abscissa x =a? Show that there is one point on the parabola
with tangent parallel to Ox. What is this point ?
18. From the derivative of (ax 2 +bx +c), show that the tangent at the* point
(#i 2/i) on tne parabola y =ax 2 + bx+c has equation
y+y l =2axx +b(x+x l l )+2c.
19. Show that the tangent at the point (1, 1) on the rectangular hyperbola
xy 1 cuts equal lengths off the axes.
20. Show that xy l + yx l =2<x.* is the tangent at the point P(x lt yj on the
rectangular hyperbola xy = a
2
If the tangent cuts the axes in
. and B and A
if PM and PN
are perpendicular to the axes, show that bisects OA M
and Nbisects OB. Show also that the area of the triangle OAB is a constant
independent of the position of on the curve. P
21. From the derivative of -
x
-
JL
, find the equation of the tangent at the
-
point (x l9 2/ x ) on the rectangular hyperbola (x l)(y -2) =3. Show that the
tangent gradient tends to become infinite as x l -+l. What is the relation of
this fact to the vertical asymptote of the curve ?
22. Find the derivatives of ax and ace , where a is a constant.

2
A
ball rolls at feet in t seconds. Show that the average speed over any timo-
interval and the velocity at any time are both constant.
A
lead pellet falls at* feet in t seconds. What is the average speed between
the jth and the 2 th seconds? Find the velocity after t seconds. How good
an approximation is the velocity after 3 seconds to the average speed between
times t ~3 and t 3-5 seconds?
23. Show that the acceleration of the lead pellet of the previous example
is constant over time.
24. What is the derivative of ( 1 4- 2#) ? Show that the electric current in a
wire is constant if (1 4- 2) units of electricity pass a point hi the wire in t
seconds.
25. Draw the total revenue curve on one graph and the demand and
marginal revenue curves on another graph in the case of the demand for
gramophones of Example V, 2. Verify that the marginal revenue is zero at
the output giving greatest total revenue.
26. If the market demand for tea is given by the law of Examples V, 1,
draw the total revenue curve and read off the tangent gradient to give the
marginal revenue at demands of 50, 60 and 70 thousand Ibs. per week. Check
these values from the expression for marginal revenue. Draw on one graph
the demand arid marginal revoiiue lines. At what price does marginal revenue
vanish ?
27. Evaluate the derivative of (ax + bx*). If WOx passengers travel on tho
train of Examples V, 1 1 find an expression for the marginal revenue derived
,
by tho railway company. Draw the marginal revenue curve, showing that it
is a part of a parabola falling to the right. Find the fare at which marginal
revenue is zero and verify that this fare produces the greatest total revenue.
28. A firm produces

radio sots according to the conditions of Examples V,
15. Draw two
graphs, one showing the total cost curve and the other the
average and marginal cost curves. Verify that average and marginal cost are
equal when the former assumes its loast value.
29. If the cost of a coal merchant's supply of coal is of the form described
in Examples V, 16, show that the marginal cost decreases steadily as the
monthly purchase increases.
30. From
the derivative of (ax 3 +bx z + cx), find tho marginal cost of any
output in the case of the firm of Examples V, 18. Show that the marginal cost
curve is a parabola and locate tho output for least marginal cost. Draw the
average and marginal cost curves on one graph and show that the output for
least average cost is greater than the output for least marginal cost.
31. A business produces an income of x this year and % next year, where
x2
these values can be varied according to the relation y = 1000 -^^ Explain
~ -
I
Kdy\ - J
1 > can bo interpreted as the marginal rate of return over
cost. Show that the value of this marginal rate is when this year's
income is #.
CHAPTER VII
THE TECHNIQUE OF DERIVATION

7.1 Introduction.
THE object of the differential calculus is to systematise the

first
process of finding the derivatives of functions, to make the writing

down of a derivative an almost mechanical matter. Text-books on
the subject necessarily devote a considerable amount of space to
but it must be remembered that the systematic evaluation
this stage,
of derivatives is not an end in We
evaluate derivatives only
itself.
because they are useful to us in the application of mathematical

methods in the natural Or social sciences. These applications would
be extremely laborious were it not for the technique of the differential
calculus we should find ourselves repeating over and over again the
;
same kind of algebraic process, the process of finding the limits of

certain expressions, whenever we need the derivatives of particular
functions. We
have, therefore, to learn to evaluate derivatives easily
before we can apply them fruitfully in economics or any other scien-
tific study.
The systematised technique of derivation involves two steps. The
first consists of the evaluation (from the definition) of the derivatives
"
of the simplest functions, the results, which are called standard
forms ", being set out in tabular form and memorised. The table
of standard forms, once obtained, is taken for granted and the
derivatives it contains simply quoted whenever they are required.
All the functions considered in the present chapter are actually
" "
derivable from one simple function, the power function. Our
table of standard forms, therefore, need contain only a single entry
at the moment, i.e. the derivative of the simplest power function.
Many other functions, such as the trigonometric, exponential and
logarithmic functions, are used in mathematical analysis and some
of these will be introduced at a later stage. It will then be necessary
THE TECHNIQUE OF DERIVATION 161
to extend the table of standard forms by evaluating the derivatives

of the simplest examples of the new functions.
The second step is to frame a set of rules which serve for the
derivation of more complicated functions. A given function, no
matter how involved, is reduced to a combination of simpler functions
the derivatives of which appear in the table of standard forms. The
rules for derivation simply tell us how the derivative of a combination
of this kind can be obtained in terms of the simpler standard form
derivatives. The combinations may involve sums, differences, pro-
ducts or quotients. But, as we shall see, they may also include a
very different kind of configuration. In all cases, the rules for deriva-
tion are relatively simple. With the table of standard forms and
sufficient skill in themanipulation of the rules, we can set about
evaluating derivatives with the greatest confidence. This confidence
should be acquired in the course of the present chapter.
7.2 The power function and its derivative.

In elementary algebra we deal with expressions of the power
n
type a where n is called the index (or exponent) of the given base a.
,
The meaning of the power varies, however, with the nature of the
number n. For example, a 2 =a xa, ab Ja, a~

2
and aT% -$-=2
a va
In general terms, the power a n is to
be interpreted as follows :
If n is a a n denotes the result of

(1) positive integer, multiplying
a by itself n times.
(2) If n is a positive fractional value, a n denotes the positive value
r
r
of a certain root : aq '(/a .
(3) If n is a negative integral or fractional value, a n denotes the

-- 1
reciprocal of the corresponding positive power : a 9 = -- .
va r
(4) If n a n stands for unity a
is zero, 1. :
n
is an irrational number, the power a is more complicated
If n
(5)
in meaning. We can write n, in this case, as the limit of a sequence
of integral or fractional numbers n l9 n z n s ... The power a n is then
, ,
.
defined as the limit of a n r as r~>oo .
"
In all cases, the power a n obeys the familiar "index laws developed
in elementary algebra.
L M.A.
162 MATHEMATICAL ANALYSIS FOE ECONOMISTS
One of the
simplest of all functions isthe power function y =# n ,
where the index n is a fixed number. The variable x takes values

continuously from the whole set of numbers except that only positive
values can be considered in certain cases (such as y Jx). dis- A
tinctive characteristic of the power function is that both the function
and its inverse are of the same type. For example, the power func-
tion y x 2 gives the inverse function x = Jy, which is a power function
of y.
The derivative of x n , for any value of n, is now required as a
standard form. A number of simple cases can be taken first :
d i \ T-
T-(*)=:Lim
ax
(x+h)-x =Lim
* T h
= 1, - .
X -
( T T
=Lim /O i\ O
(2x+h)~2x,
d .
^
3 T .
(x
-
d m
_. i/
r -(-J=Lun rl -7--J=
i i\
:
T.
Lim -- -
;
i
rr= --5,
i
dx \a;/ h \x h ^Q +h xl A_> x(x +h) x2
.e.
~ "
.e.
A uniform law is now apparent. In each of the above cases the

derivative of x n is found to be of the form nx n ~l This is, in fact, .
true for any given value of n whatever and,

to provide a general
proof, we need a result which will be established at a later stage (see

17.3, Ex. 1, below). The result can be expressed quite simply :
where A some quantity involving x which even when
--
is is finite
(X+k) -X ^Lnn X n f/
d n n n
mu
Then , ^ T-
=Lim T .
+-k\ - ^}
T~(^ X )
-r- 1 1 1 1 1
h _+ h [\ xJ j
=Lim # n (- +A 2 ]
xn -~
h -+Q \X X / X
Hence, we have the standard form :
(x
n
)
=nx n ~ l for any fixed value of n.
This result can now be quoted whenever required. For example,

we are entitled to write at once, not only the particular derivatives
evaluated above, but also many others such as
d fl
- = i
N
v i
'
j**/
~
.
/Y/*
U'^x />4
iC
_ 2 _t _
d_ | _2_
fJL^ = A(Z-t) = _!-*=

O L=. '
ax
,//>. V / //>.
\jjc/ ax V
/7/y.
/
& /- / i^
Many rather different kinds of derivatives are thus included in the

one standard form. The method of treating fractional and negative
indices is to be noticed.
7.3 Rules for the evaluation of derivatives.

We come now to the second stage in the technique of derivation,
to the framing of a set of rules for the derivation of combinations of
functions. The first three rules relate to combinations which are
sums (or differences), products and quotients respectively. In each
of these rules, as set out below, it is assumed that u and v are two
given functions of x with known derivatives at the point concerned.
164 MATHEMATICAL ANALYSIS JgJOR ECONOMISTS
RULE I. The derivative of a sum^df difference.
The derivative of the sum (orîfference) of two functions is the
sum (or difference) of the separate derivatives :
j
*
d du, dv
. . d du dv , . .
(u+v) -r- + -^- and

dx (t4-v)=-
ax
r-
dx dx
.
, dx dx
RULE II. The derivative of a product.
The derivative of the product of two functions is equal to the
firstfunction times the derivative of the second plus the second
function times the derivative of the first :
d dv du
RULE III. The derivative of a quotient.

The derivative of the quotient of two functions is equal to the
denominator times the derivative of the numerator minus the
numerator times the derivative of the denominator, all divided by
the square of the denominator :
du dv
d (u\ _ dx dx
dx U/ v2
A
As a
.
,
special case :
d
- - fl\
- I 1
~ - - Idu
- -~ .
dx \uJ u 2 dx
Formal proofs of the three rules proceed directly from the defini-
tion of the derivative. The proof of the first rule is :
If u = (/>(x) and v ifj(x) are the two given functions, then
d. .
_.
-r~(u +v)=Lim
A->0
_du dv
~dx dx'
and similarly for the difference of u and v. use here the fact We
(see 4.5 above) that the limit of a sum (or difference) is the sum (or
The proofs of the other two rules,
difference) of the separate limits.
though more complicated, can be given in essentially the same way.*
These rules can also be deduced from the first rule by a method to
be described at a later stage (see 10.2 below).
*
See, for example, Hardy, Pure Mathematics, (3rd Ed., 1921), pp. 203-4.
Important deductions from Rules I and II concern the behaviour

of constants in the process of derivation. A constant can be
regarded as the (limiting) case of a function of x which does not
change in value as x varies. A constant must thus have a zero
rate of change, i.e. the derivative of a constant is zero. Rules I and
II now give
d ,
. du , d . . du
-r (u +0) -r- and -=- (au) a ,
dx dx dx dx
where u is any function of x and a is any constant. Hence, an
additive constant disappears when the derivative is taken and a
multiplicative constant remains unaffected by derivation. These
facts are of very great service, as the examples of the following
section show.
The three rules are set out above for the case of a combination of
two functions only. But, if more than two functions appear in a
combination, the rules can be applied several times in succes-
sion to give the derivative. It can be noticed, however, that the
sum and difference rule extends at once to give the derivative
" "
of the algebraic sum of a number of functions as the similar
" "
algebraic sum of the derivatives of the separate functions. For
example,
r d
,
\_û
777 v aw
'
dx dx dx dx
where u, v and w are three functions of x with known derivatives.
The
derivative of a product or quotient involving more than two
functions is less easily obtained, as is shown by the following example :
Taking the product (uvw) as (uv) times w, we obtain

d dw d dw I dv du\
.
~j-
,
(UV)
^
= UV-^- +W \U +V I
dx dx dx dx \ dx ax/
du dv dw
vw ^- +uw -^- +uv -=
dx dx dx
In an exactly similar way, we can show that
d fuv\ 1 / du dv dw\
= +uw~r -uv-j-]
w*\[vw dx
I I .
dx\wJ dx dxJ
It is possible to put the product and quotient rules in a somewhat

different form suitable for the extension to the case of more than
two functions. The product rule for the derivative of the function
y~uv can be divided through by y=uv and written

1 dy _ I du Idv
ydx udx vdx
ni
In the same way, the quotient rule for y=- can be arranged
1 dy __ 1 du Idv
ydx~udx vdx
Further, the derivative of y=uvw obtained above becomes
1 dy
_ I du Idv I dw
ydx~udx vdx wdx
9
and the derivative of v

y becomes
w
1 dy __ 1 du Idv I dw
ydx~udx vdx wdx
We now have a common form of the product and quotient rules
which can be extended, in an obvious way, to the case where any
number of functions are involved. The meaning of this alternative
expression of the rules will be made more clear later (see 10.2 below).
7.4 Examples of the evaluation of derivatives.
The examples below illustrate the way in which the rules for
derivation are used in practice. The derivative of a fairly compli-
cated function is to be obtained by several applications either of the
same rule or of different rules. The method of dealing with constants
in the derivation process is particularly to be noticed. It is also
clear that some derivatives can be obtained in two or more different
ways. The result is always the same and all that matters is to get
it in some way or other.
Ex.1. (5x)=5-j-(x) = 5.
d
. 2
Z. j~\~~/""J~\
ax \x/ ax \x
Ex. 3.
x
Ex.4. -(
Ex.5. (-
Ex.6.
Ex. 7. For any fixed values of the constants a, 6, c, ... ,

we have
-T- (ax + 6) =a ;
-^~ (ax
2
+ bx+c) = 2ax + b ;
-y- (ao;
3
-f bx2 +cx+d)=3ax2 +2bx+c; and so on.
cte
Ex. 8. +a;
2
)}=a:~(l +a;2 ) + (l
jj-{x(l +x*)^(x)=x(2x) +(1
= l+3a;2 ,
2 =A
^{a;(l +a: )}
2
Ex.9. ^-(l+a;) =^(
ax ax
or
ax
Ex. 10. + (l+2a;)

^{a:(l+a:)(l-f2a:)}=a:(l4-a;)^(l+2a;)
+60; + 6a;2 .
Ex.11,
jfc
d d
Ex.12, "f
9 "*2 -1 ^
d /2xz + l\ d / 1\ 1 2xa -
or 2* + - 1=2 ;=
dx\ x /)=-r
j-( (
2 2
<ic\ a;/ a; a;
3
" (l-2s)+2(l+s)
2
= 2
*
(l-2a;) (l-2a;)
Ex. 14. ~( {f ^ A
//>
UtJU I
^ I Ju \ I Î t 'V
I */
I.
~r V
1
I I
J I
2
ic 2
Ex. 15.
f d -
dx\x+br-cx)=a^-\
-=-[
/
ax
(
1
dx\x + b/
\ d - x \
)
d
-c-f-(x)
dx
__ (x + 6) - x "
~ __
ab
Ex. 16. ~
X ~
dx f(x) f(x) dx (x) _af'(x)-f(x)
d_ (f(x)\ = '
dx\~ x f x* x*
where /(^) is any single-valued function of x.
7.5 The function of a function rule.
The three rules we have given are not in themselves sufficient for
the derivation of all the functions met with in ordinary mathe-
matical analysis. On the contrary, there are many relatively simple
functions the derivatives of which cannot be found by means of the
three rules. Consider, for example, the function
" "
which is a of the simple quadratic function (2# 2 - 3) and
mixture
the simple square root function. On attempting to evaluate the
derivative, however, it is found that the function does not break
" "
up into sums, differences, products or quotients of standard form
"
functions. The mixture ", in fact, is not as before. We have a
function which is fundamentally different in nature and we need
some method of evaluating its derivative.
The problem is solved by introducing a new variable :
THE TECHNIQUE OP DERIVATION 169
Here, u is a function of x and its derivative is
du d /f^ n ^
srs <*'-s)=te.
The original function now appears as y = Ju, i.e. as a simple function
of the variable u with a known derivative with respect to u :
The function y \/2# 2 -3 has been arranged, therefore, in such a

way that we can say that y is a function of u, where u is a function
of x. Both these functions have known derivatives. Can we now
deduce the derivative of y with respect tox\ The answer is provided
by the very simple rule given below.
The way in which we have arranged the function y N/2# 2 -3
"
shows that it can be described as a function of a function of x ".
It clearly only one example of a wide range of such functions to
is
which the following rule for derivation applies :
RULE IV. The derivative of a function of a function.

If y is a function of u where u is a function ofx, then the derivative
of y with respect to x is the product of the derivative of y with

respect to u and the derivative of u with respect to x :
dy ~~ dy du
dx du dx
The formal proof of the rule proceeds :
Ify~f(u), where u = (f>(x), then y=f{<f>(x)} and
dx h-+Q h
Let k = <>x + h)-<>x)->0, as
So
dx h ^Q
T
= L/im
.
-- - - j
*
k
f(u + k)-f(u) T .
.L/im
A-*O
h
<f>(x + h)-<j>(x)
7
h
= dy
-7
du
-7-
du dx
We use here the result (see 4.5 above) that the limit of a product is
the product of the separate limits.
The rule clearly extends to the case where there are several " inter-
"
mediate functions. For example, if y is a function of u where u
is a function of v which, in its turn, is a function of x, then
dy _ dy du dv
dx du dv dx
This follows from two successive applications of the simple function
of a function rule given above.
The
success of the function of a function rule in practice depends
" "
largely on the introduction of an intermediate function u in such
a way that the derivative is most conveniently obtained. The
method of breaking up functions for the application of the rule is
illustrated by the following examples.
and so
dx
Since u is only introduced for convenience of working, the final step
" "
is to get rid of this intermediate variable in terms of x.
y = (l+x)
2
Ex.2. ,
i.e.
So .(.
y u* where u = 1 + x.
This result has been obtained before (7.4, Ex. 9) by other methods.
i.e. 2/=- where M =
which has also been obtained before.

Ex.4,
i.e. y=un where u=ax + b.
*
(ax b)=~
fT+x
y = V^ where w =
I1.
i.e.
** A A ~2X
__
NOW
Tk.T
^.jM=;__ = _y
I /J-
)
T
3
and -J.(^.) =7 r (7.4, Ex. 13).
So A /LE-i
/
5i V"fT2S~2 V~l~+^
A -2s 33
(1^2^~2
Each of these five examples is an instance of one general form
which can be derived at once from the function of a function rule :
d , n\ . du
--( u ' nun-l ,
dx dx
where u is any given function of x. In the case n - 1, we have
d fl\
dx
__du>
"" ~
u 2 dxW '
which is the special case of the quotient rule already given.
7.6 The inverse function rule.
From the function of a function rule we can derive the fifth, and
last, This rule provides the derivative
of the rules for derivation.
of the function inverse to a given function and it applies only if both
the inverse functions are single- valued.
RULE
V. The derivative of an inverse function.
The
derivative of an inverse function is the reciprocal of the
derivative of the original function :
dx_l_
dy~~dy'
dx
provided that both functions are single-valued.
The proof is as follows. Suppose that the single- valued function
y= {>
j (x)
=
gives the single- valued inverse function x *fj(y). Then
<l*{<f>(x)} must equal x for any value of x. So
But W(*)>=*(y) *(*) (by Rule IV)
dx dy
dy dx
So, the product of -=- and

-j-
is unity and the rule is proved.
As an illustration of the inverse function rule, we can take the

case of the derivative of >Jx. If y = Jx, then x y
z
The inverse
.
function rule then gives
d . . . 1 I I
This result agrees, of course, with that obtained from the standard
form giving the derivative of x n for any value of n.
The process of evaluating derivatives can now be regarded as
complete. The five rules, taken in conjunction with a suitable table
of standard forms, are sufficient for the derivation of all single-
valued functions, no matter how
complicated is their analytical
expression. Even when functions of entirely new types are intro-
duced, it is only necessary to extend the table of standard forms
by including the derivatives of the simplest instances of the
new function types, as obtained directly from the definition of
the derivative. Examples of this extension of the table of standard
forms will be given later.
7.7 The evaluation of second and higher order derivatives.
The practical method of obtaining the second and higher order

derivatives of a function introduces nothing that is new. Having
obtained the first derivative by means of the rules set out above,
the second derivative is obtained by a further application of the rules,
this time applied to the first derivative considered as a function
of x. The third, fourthand higher derivatives are then obtained in
succession in the same way.
It is possible, however, to extend the sum and product rules so
that they apply directly to second and higher order derivatives.
If u and v are two single-valued functions of x, we have
d _d Idu dv\
rv ~~
d*^ '~dx\ dx dx \dx dx)
_d u
2
d fdu\ d fdv\
__
"" ~ 2
dx \dxl dx \dxl dx
In general, for any value of r which is a positive integer,
d r
, v
d u r
d'v
The extension of the product rule is more difficult. Prom

d , . dv du
9
dx dx dx
i d2 d ( dv\
.
v d I du
we have -^(uv) =T- (u-^ }+ -T- \v -
2
dudv\ fdudv
~ / U d v *~~ +
d*u
+ V~ ~
/ --
d v 2
L n dudv
----- d 2u -
I-T; .
dx*
The third derivative is then found by a further derivation process as
du d 2 v d 2u dv
A general result for the rth derivative of the product is obtained, in

this way, by repeated derivation :
_d
r
'
.
I
(
v
nii%j \
'~
ni
r
- d__ v
+
dud
'
\ A* __ r ~l
'~ 1+
v I
__ _ ___
rfr-
x
f
z '- z
I
I
r(r-l)d v
__
2
'
_ __
dx
r- z
r~2
ud 2v
dx 2
(_ ___
d r ~ l u dv
dx r ~ l dx
|_ 7;
d ru
_
. _
dx r
This is known as Leibnitz's Theorem and the coefficients of the

successive terms on the right-hand side follow the well-known
" "
binomial law. Prom the point of view of the present develop-
ment, however, only the first and second derivatives are normally
required and these can always be obtained by direct application of
the simple rules given above. A number of examples are given for
illustration :
Ex.l. (#) =4*';

J^)
Ex.2. )
Ex.3. -(*) = *0-i; jp(a) =n(n -!)*-;
-7-= (#
n
)
= ft (w - 1) (w- - 2)# n ~3 and
; so on.
oar*
The general result for the rth derivative of this standard form can be
written down by inspection :
r
d n - n ~r
T~r (x =n(n 1) (n -2)
)
... (n-r + l)x .
The previous two examples are particular cases of this standard result.
If n a positive integer, as in the first example, then the derivatives after
is
the ?&th order are all zero. If n has any other value, the process of finding
successive derivatives can be extended indefinitely without producing
zero derivatives.
Ex4 .--
dx\3x+2
18
d* 1
Ex.5,
2* - 3
and similarly for higher order derivatives.
Ex. 6. If f(x) denotes any single- valued function, then
{*/(*)} =*/+/(*),
and {xf(x ) }
=
and = '
~ - {xfl (x) ~ *Z)
{~?} ^ \f i&xf (x} f(x)} f(x)}
JL
(
]
i
" /^ f ^^ f ^^j
rH 9T/W ^.c^
f
r
â/^ -t-y
-l- /i-l
y
f
^*c^c/
f i-t
j/Vi-Hl
\JC)j}
EXAMPLES VII
Practical derivation
1. Write down, from the standard form, the derivatives of Va?, -- and
x
2. Obtain the derivatives of l+2ce a -3:e 4 ; (l-x)

3
; (x -2)(2a; + 1)
^
;
1
9
l-x* (a; + 2)(a;4-3) l-
3. If 0, 6 and c are constants, show that
d / 1 \_ a d / 1 \ 2ax+b
2
"" "" &n ~
di + bJ
\aa;+6/
da; \aa; (ax +
4- 6)
2
da; \aa; a + 60? + c/ (aa;
2
+ + c) 2
fea;
4. Show that
5. If /(a;) =az+b, find the derivatives of a;/ (a;) and "
6. Verify that
deduce that
1 +a;+a; 2
+ ... 4- a;
n = -
^
-
l ~x .
By means
a/
of derivatives,
1 xn nxn
(1 x) 1 x
7. It is given that y = 1 +w* where w = 1 -a;

2
Find y as a function of x. .
Write down the derivatives of y with respect to u and of y and u with respect
to a;. Verify that = -~ -=-
-
da; dw da;
i
1
8. Find the derivatives of ... and
Jx 4- 1 v a; + 1 - vx - I
9. Use the function of a function rule to find the derivatives of
and
Deduce the last derivative from the other two by the quotient rule.
10. From the function of a function rule find the derivatives of
Multiply out these powers and deduce the derivatives by using the sum, pro-
duct and quotient rules only.
11. Find the derivative of each of the general expressions :
'
a*+te+c'
12. If f(x) is a single-valued function of x, express the derivatives of *Jf(x)
and its reciprocal in terms of the derivative of
13. If x and y satisfy the relation xy =a, show that

*?=_ and *?=_*,
ax x dy y
so verifying the inverse function rule in this case. Consider the relation
n a in the same
xy way.
d / dy\
14. If y a single-valued function of x, show that -r- (x 2 -f y 2 = 2 x + y -f-
is )
( I
dy x
Deduce that -~ = - - if x and y satisfy the relation x 2 + y 2 = 1
Deduce --
, Verify by .
dx y
finding the derivative of y =
15. Evaluate the second derivatives of and
16. Find the first and second derivatives of

ijx
-
ax + b
- and of -
ax + 6
~ Deduce
that successive derivatives of those functions differ only by a constant factor.
#-~ + 2 dy =0.
dîi
17. If xy =a, show that Show that the same result holds
dx dx
foTxy=a+bx.
18. If y x2 H , show that x 2 -j~ + x ~ - 4w =0. Does the same result
x* ft
dx* dx
hold for the relation y =ax* + ?
19. Show that all derivatives of ax

2
+bx+cof higher order than the second
are zero.
20. If f(x) is a single-valued function of x, obtain the third and fourth

derivatives of x/(x) in terms of the derivatives off(x). In general, deduce that
and verify by Leibnitz's Theorem.

21. Draw graphs of y =x* and y=f/x for a range of positive and negative
values of x and verify that each function is monotonic increasing. How is one
graph to be obtained from the other ? Find the derivative of f/x and check that
the tangent gradient to the curve y =â?, as estimated from the graph, equals
this derivative. Show that the derivative becomes infinite at x=Q and so
that the tangent to the curve is Oy at this point.
22. Write xx and x* as products and deduce their derivatives from the
2
,
3
by the product rule. From the derivatives of x and X use

s
derivative of x 2
,
the quotient rule to obtain the derivatives of 2 and and the inverse - -
,
x x*
function rule to obtain the derivatives of *Jx and tyx. Find the derivatives
of x* and x" a by the function of a function rule from the derivative of *Jx.
23. Generalise the results of the previous example to show how the standard
~
form n nxn l can be deduced, for any rational value of n, from the rules
dx (x )
d
of derivation and from the fact that (x) 1 .
u.'X
24. Find the derivative of (a - bx)x 2 If a hemispherical bowl of radius

.
10 inches is filled to a depth of x inches with water, it is known that the volume
(10 -x)x cubic inches. Find the rate of increase of volume as
2
of water is 77-
the depth increases and an approximate value of the volume increase when
the depth increases from 5 to 5-1 inches.
25. If water is poured into the bowl of the previous example at the rate of
one cubic inch per second, show that the depth increases by r-r
~
-
inches
'
7rX
per second when the depth is x inches.
'
26. Find the equation of tho tangent to the curve y = ax f 6 + ~ at the point
with abscissa x^ Where is the tangent parallel to Oxl x
27. A projectile travels (1 + ) feet in t seconds. Show that it moves with

2
increasing velocity and acceleration. What is the velocity after 10 seconds?

Find an approximate value of the distance travelled during a small time A2
from the tth second.
28. After t seconds a body is x at + %bt 2 feet from its starting-point. What
are the velocity and acceleration then ? Show that tho velocity is \/ 2
-f 2bx
feet per second when a; feet have been covered.
29. If the demand law is p c, show that total revenue decreases as

x
output increases, marginal revenue being a negative constant.
30. In tho case of the demand law p (a
- bx) 2 show that the average and ,
marginal revenue curves are both parabolas, tho former lying above the latter.
Show that marginal revenue falls to a negative value and then rises to zero as
output increases. Plot the total, average and marginal revenue curves in the
case of the demand of Examples V, 10.
31. Show that the demand law p=a-bx 2 gives parabolic average and
marginal revenue curves of similar shape, both falling continuously as output
increases. Where is marginal revenue zero ?
32. What is the marginal revenue function for the demand p = Va - bx ?
Under the conditions of Examples V, 8, how many tenements must be rented
per week before marginal revenue falls to zero ?
33. Show that marginal revenue can always be expressed as p +x
Deduce that the gradient of the demand curve is numerically equal to
p at the
x
output where marginal revenue is zero. Check this result in the cases of the
particular demand laws of the two previous examples.
M M.A.
34. The total cost function is 77 = */ax + b + c. Find an expression for
marginal cost and show that it decreases as output increases. Plot the mar-
ginal cost curve for the electricity output of Examples V, 17.
X 4- 6
36. If IJ ax \-d is the total cost of an output x, show that the mar-
ginal cost of the output is a -j

*
1 4- :
r^
\ If 6 >c, deduce that marginal
\X ~\~ Cf J
cost falls continuously as output increases. Draw the marginal cost curve
of the tobacco manufacturer of Examples V, 20.
36. If a firm can produce two chocolate " lines " according to the conditions
of Examples V, 25, interpret the value of the derivative At what output ~ .
dx
is a small decrease in the production of one line accompanied by an equal
increase in the other line ?
37. The output of a coal-mine is given in terms of the number of men

working per shift according to the relation of Examples V, 27. Draw a graph
showing variation of output. Find expressions for average and marginal
output when x men per shift are worked and draw the corresponding curves.
Show that both "curves are parabolas and that average output only equals
marginal output when the former has its greatest value.
CHAPTER VIII
APPLICATIONS OF DERIVATIVES
8.1 The sign and magnitude of the derivative.
WE have seen that the value of a derivative can be interpreted in
two ways. The
derivative of a single- valued function f(x) measures
the rate of change of the function and the tangent gradient of the
curve yf(x) at the point in question. In the present chapter we
base certain very important applications of the derivative on these
interpretations of its value. It will be noticed, however, that our
" "
results depend largely upon diagrammatic intuition and are not
rigorously established. The less strict development is sufficient at
this stage, but it can be added that it is possible to provide formal
analytical proofs of our results in all cases.*

The value of the derivative off(x) at the point x a is f'(a). The
meaning of the sign of f'(a) is evident. If f'(a) is positive, then the
rate of change of f(x) is positive, i.e. f(x) increases as x increases
through thd value x=a. The tangent gradient of the curve y=f(x)
is positive and the tangent and curve slope upwards from left to
right at the point with abscissa a. The converse properties hold if

/'(a) is negative. Hence,
(1) f'(a)>implies that f(x) increases as x increases and that the
curve y=f(x) rises from left to right at the point x=a.
/
(2)/ (a)<0 implies that f(x) decreases as x increases and that
the curve y~f(x) falls from left to right at the point x=a.
The numerical magnitude of the derivative f'(a) then measures how
rapidly the function f(x) increases or decreases, and how steeply the
curve y=f(x) rises or falls, at the point x=a.
The results can be extended to indicate the nature of the function
or curve over a whole range of values of the variable x. In order to
* Some of the when we
analytical proofs are given at a later stage, consider
the important result known as Taylor's series.
see when the function increases or decreases, or where the curve
rises or falls, it is only necessary to examine the sign of the derivative
of the function. For all ranges of values ofx in which/'(#) is positive,
we know that/(#) increases continuously as x increases and that the
curve y=f(x) rises continuously from left to right, and conversely
for ranges in which /'(#) is negative.
One case of special interest now presents itself, the case where the
derivative is zero at a given point. If/'(a) = 0, then f(x) is neither
increasing nor decreasing, and the curve y=f(x) is neither rising nor
falling, at the point x=a. The value of the function is momentarily
stationary and the curve has a tangent parallel to Ox. The value
of the function at such a point is called a stationary value and much
of the following development is concerned with these values. Before
proceeding, it is instructive to examine some particular functions and
to obtain a general idea of the properties holding at their stationary
values.
Ex. 1.
Here =4-2* = -2(a?-2).

ax
The derivative is positive when x<2, zero when x = 2 and negative
when #> 2. The value of the function increases at first, becomes station-
ary at x = 2 and then decreases as x
increases beyond this value. The
corresponding curve is a parabola
rising to a peak at the point where
x = 2, as shown in Fig. 47. In this
case, we say that the function has a
"maximum" value y =4 at the point
where x 2.
Ex.2. y=
x TT dy
Jtiere -=- =
dx
FIG. 47.
The sign of the derivative
exactly is
opposite to that found in the previous case. The value of the function is
=
again stationary at x 2, but the function decreases as x increases up to
the value 2 and increases as x increases beyond 2. The curve, graphed in
=
Fig. 48, is a parabola with a lowest point at x 2. In this case, we say
" "
that the function has a minimum value t/t=4 at the point where x = 2.
APPLICATIONS OF DERIVATIVES 181
Ex. 3. 2/
= 6x-3z2 + ia;3 .
Here J
ax
= 6 -6*+f* 2
=f (z-2) 2 .
The derivative is positive at all points except that it becomes zero at

the single point x = 2. The function is mono tonic increasing but has a
stationary value y = 4 at x = 2. The graph of the curve is shown in Fig. 49,
FIG. 48.
from which it is seen that the curve always rises except at the single point
where the tangent is parallel to the axis Ox. The stationary value, in this
"
case, is neither a maximum " nor a " minimum ". The point concerned
"
is an example of what is called a point of inflexion ", a name derived
from the fact that the curve crosses over the (horizontal) tangent as we
pass through the point.
8.2 Maximum and minimum values.

It is now necessary to make more precise the meaning of the terms
maximum and minimum values, which, we have seen, are connected
with the stationary values of a function.
DEFINITION The function f(x) has a maximum (minimum) value

:
at a point where the value off(x) is greater (less) than all values in the
immediate neighbourhood of the point.
The maximum and minimum values together can be termed the

extreme values of the function.
It is assumed, in the following development, that the function and
itsderivative are finite and continuous at all points. The corre-
sponding curve is then smooth, being free of discontinuities and

"
sharp points ". Fig. 50 shows a hypothetical, and rather artificial,
curve of this nature, drawn for the purpose of indicating the various
possibilities that can arise.
It is clear, in the first place, that a maximum minimum value

or
off(x) can occur only at a stationary point where the curve yf(x)
has a horizontal tangent. If the tangent slopes upwards at any
point, there are larger values off(x) immediately to the right of the
point. If the tangent slopes downwards, there are larger values of
f(x) immediately to the left of the point. Neither of these cases is
x
FIG. 50.
by definition, at a point where f(x) is a

possible, maximum. The
tangent can, therefore, only be horizontal at a maximum point
and (similarly) at a minimum point. Thus all the maximum and
minimum values of a function are included amongst the stationary
values.
In the second place, ofthe points where the tangent to the
all
curve yf(x) horizontal,

is there are some points (such as and D) A
giving maximum values and other points (such as and E) giving B
minimum values of f(x). But there remains the possibility of a
" "
third kind of point, such as the inflexional point C, where the
function has neither a maximum nor a minimum value. Stationary
values, therefore, include cases other than maximum and minimum
values.
A method for distinguishing different kinds of stationary values is
suggested by Fig. 50. The tangent gradient of the curve y =/(#) is
positive (the tangent being upward sloping) immediately to the left

and negative downward sloping) immediately to
(the tangent being
the right of a point where f(x) has a maximum value. Hence, the
derivative f'(x) changes from positive, through zero, to negative
values as x increases through a value giving a maximum off(x). At
a point where the function has a minimum value, the tangent
gradient and derivative change sign in the opposite sense. Finally,
at a point where there is a stationary value which is neither maxi-
mum nor minimum, the zero value of the tangent gradient and
derivative does not mark a change in their sign at all they have ;
the same sign on each side of the point.

These results, indicated by inspection of the curve of Fig. 50, can
be set out in precise analytical form :
CRITERION FOR MAXIMUM AND MINIMUM VALUES

(1) All maximum and minimum values of the single- valued
function f(x) are stationary values and occur where /'(#) is
zero.
(2) If f'(x) changes in sign from positive to negative as x increases

through a value a where /'(a) = 0, then /(a) is a maximum
value of the function /(#).
If f'(x) changes sign from negative to positive under
the same conditions, then /(a) is a minimum value of the
function.
If f'(x) does not change sign under the same conditions, then
/(a) is neither a maximum nor a minimum value of the

function.
"
There are thus two conditions to consider. The first is a neces-
"
sary condition for an extreme value, while the second adds the
"sufficient" condition which enables us to distinguish maximum,
minimum and other stationary values. The conditions together make
up "necessary and sufficient" conditions (see 8.9 below). We can
notice an alternative, but less exact,
way of expressing the necessary
condition. If y=f(x) has a stationary value (e.g. a maximum or
*
minimum value) at a point, then y is momentarily steady in value

and Ay = approximately for any small change in x from the point.
Thus when treating the function at a maximum or minimum position,
we can regard it as a constant for small variations about the position.
In any case, the value of the function always changes very slowly
from such a position.*
It must be emphasised that a maximum value of a function is
not necessarily the " greatest " value of the function. A maximum
value is greater than all neighbouring values but there may be still
greater values of the function at
more remote points. Similarly, a
minimum value need not be the
" "
smallest value. There is only
one greatest and only one least
value of the function but there
may be several alternating maxi-
mum and minimum values. This
is from Fig. 50. Further, it
clear
important to realise that every

is
stationary value is not an extreme

value and we should always be
on the look-out for "inflexional"
stationary values. Finally, our criterion may break down and fail
to show a maximum or minimum value if the function or its
derivative is Fig. 51 illustrates the fact that a

not continuous.
maximum value, occurring at a " sharp point " of the curve, may
not be indicated by the criterion we have given.
8.3 Applications of the second derivative.

The second derivative f"(x) of a function, being the derivative of
the first derivative /'(#), measures the rate of increase or decrease
of f(x), i.e. the rate of increase or decrease of the tangent gradient
to the curve y ~f(x) as we pass through the point concerned. The
sign of f"(x) at any point xa

provides some useful information.
If f"(a) is positive, then f(x) is changing at an increasing rate as
x increases through a and the tangent gradient to the curve yf(x)
increases as we pass through the point with abscissa a. The tangent
to the curve turns in the anticlockwise direction and the curve is
convex when viewed from below at this point. Conversely, if /"(a)

* The device of
keeping a maximised or minimised variable constant is
often adopted by Wicksell. See, for example, Lectures, on Political Economy
(Ed. Robbins, 1934), Vol. I, p. 181.
isnegative, then/(#) changes at a decreasing rate as x increases, the

tangent to the curve turns in the clockwise direction and the curve
isconcave from below at the point where x =a. These results, which
are fully illustrated in Fig. 52, are independent of the value of the
derivative f'(a), of whether the tangent to the curve slopes upwards,
downwards or is horizontal at the point where x~a^ Hence :
(1) /"(a)> implies that the function f(x) changes at an increasing

rate as x increases through the value a and that the curve
y~f(x) is convex from below at the point x a.
(2) f"(a) <0 implies that the function f(x) changes at a decreasing
rate as x increases through the value a and that the curve
y=f(x) is concave from below at the point x a.
Fia. 62.
The numerical magnitude of f"(a) then indicates how rapidly the

" "
value of f(x) is accelerating and how great is the curvature of
the curve yf(x) at the point xa. All questions relating to the
nature and extent of the curvature of a curve are answered by an
* But tlip results, which are obvious
enough when f(x) increases and the
tangent slopes upwards, should be interpreted with care when f(x) decreases
and the tangent slopes downwards. When we say that f'(a)< and /"(a) >0
implies that f(x) is decreasing and changing at an increasing rate at x = a, we
mean that the numerical rate of decrease off(x) is getting less (and not greater)
as x increases. A negative quantity, which is getting less in numerical magni-
tude, is increasing.
examination of the value of the second derivative of the function
concerned.
An alternative criterion for extreme values of a function can be
given in terms of the second derivative. It is assumed that the
function f(x) is finite, continuous and possesses continuous first and
second derivatives. If f'(a) is zero and f"(a) negative, then f'(x)
must decrease through zero as x increases through a, i.e. the change
in sign of /'(#) is from positive to negative and x= a must give a
maximum value of the function. In the same way, we see that a
minimum value of the function is obtained at x=a when f'(d) is
zero and /"(&) positive. Hence :
ALTERNATIVE CRITERION FOR MAXIMUM AND MINIMUM

VALUES
(1) All maximum and minimum values of f(x) occur where
/'(*)=o.
(2) If f'(a) =0 and f"(a) <0, then f(a) is a maximum value of the
function. If f'(a) = and /"(#)> 0, then f(a) is a minimum
value of the function.
" "
The condition is, as before, a
first necessary condition. The
" "
second is a sufficient condition indicating situations in which only
a maximum (or only a minimum) value off(x) can occur. But it is
" "
not complete, i.e. both necessary and sufficient (see 8.9 below),
since the case where f"(a) is zero is not covered. This case includes
all inflexional stationary values and may include other maximum or
minimum values as well. The alternative criterion, though often
more useful, is somewhat more limited in its scope than the first
criterion.
8.4 Practical methods of finding maximum and minimum values.

In practical problems, we require the maximum or minimum
values attained by one variable (y) as the value of another variable
(x) is changed. If the variables are related by a single-valued
function y =/(#), we proceed :
(1) The derivative f'(x) is obtained.
(2) The equation f'(x)

= is solved to give a number of solutions
(3) Each solution (e.g. x=a) is taken in turn and examined.

Either: the change in sign of f'(a + h) as h changes from
negative to positive is determined. If the change is from
positive to negative, /(a) is a maximum value off(x). If the
change is from negative to positive, /(a) is a minimum value
of f(x). If there is no change of sign, /(a) is an inflexional
value of /(#).
Or f"(a) is obtained and
:
sign determined.
its If f"(a) is
negative, /(a) is a maximum value of f(x). If f"(a) is
positive, /(a) is a minimum value of /(#).

If this process is carried out in the cases of the three simple
functions quoted in 8.1 above, it is found that one stationary value
y = 4 occurs at x = 2 in each case. This is a maximum value, a mini-
mum value and an inflexional value in the three cases respectively.
The following examples provide further illustrations.
Ex.1. -a:3 -3a; 2 +5.
Here =3z2 -6x -

ax ax2
The stationary values of y occur where 3x(x - 2) =0, i.e. at x =0 and at
x =2. The corresponding values of y are y =5 and y = 1. To distinguish
between maximum and minimum values,
at z = 0, ^4= -6<0;
A
The function has a maximum value 5 at # = and a minimum value

1 = 2. The graph of Fig. 56 below illustrates.
at x
Ex. 2. y=#4
Here = 4a3 - 12z2 + 16 - 12z 2 - 24*
ax ax*
There are stationary values y=-llato;=-l and y = 16 at x =2. At

x = - 1, the value of the second derivative is 36. The function thus has
a minimum value of - 1 1 at this point. At x = 2, the second derivative
is zero and the nature of the stationary value here remains undecided.
To settle this point, the first derivative at x =2+ h has value 4A2 (A+3).
This is positive for all small (positive or negative) values of h. The
function has an inflexional stationary value at x = 2. The graph of
Fig. 57 below illustrates.
2x + l
Ex.3.
x-1
Here ^=--
which is negative for all values of x. The function has no stationary,
and so no maximum or minimum, values. The curve representing this
function is a rectangular hyperbola with asymptotes parallel to the axes,
a curve which clearly has no tangent parallel to Ox.
2-x
Ex.4. y-
x*+x-2
dy
Here 2
dx (x*+x-2)
There are stationary values y=-lata:=0 and y= -J at #=4. We
now examine the sign of the derivative near these points :
dy_ h(h-4) ^ __,,.

at x =h -i
;
dy_
--- at z =
The first expression changes from positive to negative as h is given

small values changing from
negative to positive. The
second expression changes in
I
the opposite sense as k is varied

1+ '
similarly. The function thus

has a maximum value - 1 at
x=0 and a minimum value
2 -1
J^ \
j_^g. The curious feature of this

case is that the maximum
value of the function is smaller
than the minimum value . This
apparently paradoxical result
is due to the fact that the
function has infinite values, at
FlG 53 # = 1 and at x = - 2. (At each

of these values the denomin-
ator of y is zero.) The graph of the function, shown in Fig. 53, illustrates
how the presence of infinities influences the maximum and minimum
values.
Ex.5. y=ax2 +bx + c.
3-*-
62 4ac
There is one stationary value y -j- at x --b This value is
a maximum if a is negative and a minimum if a is positive. The curve

corresponding to this function is a parabola with axis vertical. The
stationary value occurs at the vertex of the curve and is a maximum or
minimum according to the direction of the axis as determined by the sign
of a (see 3.4 above).
^
Ex. 6. y=ax + b +-, where a, 6 and c are positive constants and x is
x
assumed to take only positive values.
Here dy'-a- C
~dx ~tf*
^--
dx*~x*'
Since x is positive, there is only one stationary value and this occurs
at the point x /v/-- . The second derivative is positive. So, the function
has a minimum value y = 2 Vac + 6 at this point.
Ex. 7. y=ax*-bx+c+-x , where a, 6, c and d are positive constants
and x is restricted to positive values.
dy == d , dzy 2d
Here 2a*-&-~2
_|
;
-**+#
The stationary values occur at points where the derivative is zero, i.e.
where x satisfies the relation :
It can be shown that this cubic equation has only one positive root, and
that the value of this root is greater than x

=
The second derivative
oa
is seen to be positive. The function has thus a single minimum value for
the range of positive values of x.
Ex. 8. An open box consists of a square base with vertical sides and
has a volume of 4 cubic feet. What are the dimensions of the box for a
minimum surface area ?
Let x feet be the side of the square base and h feet the depth of the
box. Then A# 2 =4 since the volume is given as 4 cubic feet. From this
we obtain h in terms of x. The surface area is
y = x + 4Ao? = x H
2 2
square feet.
So, y is written as a function of x only and its minimum value is
required for variation in x (which alone governs the dimensions of the
box). The values of x for stationary values of y are given by
Hence x = 2 gives the only stationary value of y. Since
-
r --r when x =2,
ax* x*
the stationary value is a minimum. The minimum surface area is thus

1 2 square feet and the dimensions of the box are then side of base :
= 2 feet
and depth of box = 1 foot. The depth of the box is half the side of the
square base.
8.5 A general problem of average and marginal values.

Prom a given single-valued function f(x) at a point x, we derive
the pair of values :
Average value off(x) =J-^

f(x)
; Marginal value off(x) =f'(x).
x
The average value is taken over the whole range from zero to the
" "
given value x the marginal value refers to the
; margin at the
given value x. In diagrammatic terms, if is the point with P

abscissa x on the curve y~f(x), then the average value of f(x) is
" "
represented by the gradient of the radius vector OP and the
marginal value by the gradient of the tangent at P. The application
of these general concepts in the economic problems of demand and
cost have been considered at an earlier stage (6.8 above).
It is now assumed, both x and f(x) take only
for convenience, that
positive values. It is required to find the values of x which corre-

f/ x
\
J-^
spond to maximum or minimum values of the average x So
'
dx x x*
- 2xf'(
d?_ [/(*)! __x*f"(x)
dx*\~x j~ x3
f(x)
-^ - occur where
Stationary values of xf'(x) -f(x)=0, i.e. where
X
x
At such a point, the value of the second derivative reduces to
dx 2 x x
and, since x is taken as positive, this has the sign of /"(#). The
average value of f(x) is thus stationary at any point where the
average and marginal values off(x) are equal. The stationary value
is a maximum /"(#) is negative at the point
if it is a minimum if
;
f"(x) is
positive at the point.
In diagrammatic terms, a stationary value of the average value
of/(#) is shown by a point P
on the curve y-f(x) where the tangent
coincides with the radius vector OP, i.e. at a point where the tangent
passes through the origin 0. The average value is a maximum if the

curve is Concave from below at the point P it is a minimum if the
;
curve is convex from below at the point. These facts are clear from
the two cases illustrated in Pig. 54.
8.6 Points of inflexion.

A
single-valued function yf(x) is defined to have an inflexional
value at a point where the corresponding curve crosses from one side
of its tangent to the other. The point itself is described as a point
of inflexion. The most important property of a point of inflexion is
that marks a change in curvature, the curve changing from convex
it
to concave from below as we pass from left to right through the
point, or conversely.* This property is clear from the inflexional

* If the function is
single-valued and the curve smooth, then all changes of
curvature occur at points of inflexion. For multi-valued functions, on the
"
other hand, a change in curvature may occur where the curve turns back
on itself " and the tangent is vertical. Such a point is not a point of inflexion.
cases shown in Fig. 55. There are two classes of points of inflexion.
A point of one class (as illustrated in the first two cases of Fig. 55)
marks a change of curvature from convex to concave from below as
we move from left to right along the curve. A point of the other
class marks a change of curvature in the opposite sense (as shown in
the second two cases of Fig. 55). The actual tangent at the point of
inflexion is not restricted in any way it can slope upwards or down-
;
wards with any numerical gradient whatever. Upward and down-

ward sloping tangents are shown in Fig. 55 for each of the two classes
of inflexional points. Further, as limiting cases, the tangent can be
parallel to Ox with zero gradient or parallel to Oy with infinite gradient.
x O
Fio. 55.
In addition to the change of curvature property, another charac-

teristicof points of inflexion is evident from Fig. 55. A point of
inflexion always corresponds to an extreme value of the tangent
gradient of the curve. At a point of inflexion of the first class, the

tangent gradient is a maximum, the gradient increasing as we move
from the left towards the point and decreasing as we move to the right
away from the point. At a point of inflexion of the second class, it is
seen, in the same way, that the tangent gradient is a minimum.
Assuming that the single-valued function f(x) is finite and con-

tinuous with continuous first and second derivatives, it is a simple
matter to express the properties of an inflexional value of f(x) in

analytical terms. The derivative/'^), as the measure of the tangent
gradient, must have an extreme value at any point of inflexion. It
is necessary, therefore, that the second derivative /"(#), being the
derivative of /'(#), has zero value at the point. Further, the value
off"(x) must change in sign as x increases through the point con-
cerned, the direction of the change determining the class to which
the point of inflexion belongs. If f"(x) changes from positive to
negative, the derivative f'(x) has a maximum value and the curva-
ture changes from convex to concave from below, i.e. the point of
inflexion is of the first class. The converse change of sign of /"(#)
indicates a point of inflexion of the other class. So
CRITERION FOR POINTS OF INFLEXION

(1) An inflexional value of the function f(x) can only occur at a
point where f"(x) 0.
(2) If /"(a) =0, and if /"(#) changes in sign as x increases through

the value a, then /(a) is an inflexional value of the function
The direction of the change of sign of f"(x) indicates
f(x).
the class of the point of inflexion.
" "
The first condition is necessary for points of inflexion. The
" "
second adds the sufficient condition and, altogether, we have a
" "
criterion which is complete, i.e. necessary and sufficient (see
8.9 below).
If the function is assumed to have a continuous third derivative,
an alternative form of the criterion can be given in which this
derivative is used. If /"'(a) is negative at a point where /"(#) is
=
zero, then/'(#) is a maximum at the point x a, i.e. we have a point
of inflexion of the first class. Similarly, if f'"(a) is positive at the
point where /"(a) is zero, we have a point of inflexion of tjie second
class. Hence, if/" (a) = and/'" (a) ^0, there is a point of inflexion
at xa and the sign of the non-zero third derivative indicates to
which class the inflexional point belongs. This alternative form of
"
the criterion is not complete (not necessary and sufficient ") since
the case where the third derivative is zero is not considered.
A point of inflexion, as we have remarked, is in no way dependent
on the value assumed by the first derivative of the function at the
point. It may happen, however, that the first derivative is zero at
the point of inflexion, the value of the function being stationary
as well as inflexional. The second, and incomplete, criterion for
stationary values (8.3 above) can now be extended slightly by the
addition of the condition :
If /'(a ) =/"(a) =0 and if f'"(a) ^0, then /(a) is a stationary and

inflexional value of the function f(x).
The criterion is still incomplete since it takes no account of cases

where the third derivative is zero.*
As long as the third derivative of the function f(x) is not zero at
the point xa,
the following scheme indicates all the possible cases
of stationary and inflexional values off(x) :
Stationary value Inflexional value
/ \ / \
Extreme Stationary and Inflexional and
value inflexional value non-stationary value
/'(a) = /"(a) ^0 f'(a)=f"(a) = f'(a) ^0 f"(a) =
Two examples will illustrate the method of locating inflexional
values in the cases of particular functions :
Ex.1.
Here ^-
d
y
'*
dx*~
- ''
dx*
There is thus only one inflexional value of the function, i.e. y = 3 at
x = 1. The third derivative is positive and the second derivative changes
sign from negative to positive as we pass from left to right through the
point of inflexion. The point is thus an inflexion of the second class, the
curvature of the curve changing from concave to convex and the tangent
gradient ( -3) being a minimum. Fig. 56 indicates the point of inflexion
P and also the maximum and minimum points A and B on the curve
representing this function.
Ex. 2.
Here
g = 4(a;-2)*(* + l) ;
g-12*(*-2) ;
g-24(*-l).
* The criterion is
fully completed in 17.6 below, where extreme and in-
flexional values are determined by the first non-zero derivative of f(x) of
whatever order it may be.
There are two points of inflexion. At x = there is the inflexional value
y =0 at x =2 there is the inflexional value y = 16. At the first of these
;
points the third derivative is negative. The point is an inflexion of the

first class, the curvature changing from convex to concave and the tangent
6--
4--
3--
2--
'-1 1
FiO. 56.
gradient (16) being a maximum. At the second of the points the third
derivative is positive. The point is an inflexion of the second class, the
curvature changing from concave to convex. Further, the first derivative
isalso zero at the point and we have here a stationary as well as an
FIG. 67.
inflexional value of the function. The zero tangent gradient at the point
is a minimum
value of the tangent gradient. Fig. 57 indicates the graph
of the curve representing the function and shows the two inflexional
points in addition to the single minimum point.
8.7 Monopoly problems in economic theory.

Many of the problems of physics and other mathematical sciences
can be reduced to the determination of positions of maximum or
minimum. It is sufficient to instance the importance of maximum
or minimum potential, energy, action and entropy. An exactly
similar situation is found in certain branches of economic theory. In
static problems, we find it convenient to assume that the individual
" "
consumer seeks the highest position on his preference scale
consistent with market conditions, that the individual firm fixes
output or price to produce the largest net revenue and organises its
factors of production to give the largest output at a given cost or the
smallest cost for a given output. Similar assumed principles, in
more complicated forms, are to be found in dynamic problems. If
functions of a single variable suffice to interpret the phenomena
concerned, then the methods of the present chapter apply at once in
the solution of our problems. Some simple examples, intended to
illustrate the method, are given in the following paragraphs.*
As a first problem, suppose that a firm produces a good under X
known cost conditions represented by the total cost function n=F(x).
The demand of the firm's market for the good X is assumed to be
known and represented by the demand function x = (p) or p (x),<f> *fi
where x is the demand at price p. Within the limits set by this

demand relation, the firm is assumed to act as a monopolist with the
object of maximising net revenue. Two alternative points of view
can be taken. Either the firm fixes its output and leaves the price
:
to be determined by the demand conditions. Or the firm fixes its :
price and the demand conditions determine the appropriate output.

The analysis of the problem is different in the two cases but, as we
shall see, the results obtained are effectively identical.
Assuming that the firm fixes output, then the price to clear any
output x must be p = *f>(x) as given by the demand conditions. The
gross revenue from output x is S=x*f/(x), the total cost is IIF(x)
and the net revenue is (R-II) given as a function of x. The output
* For an account of the importance of maximum or minimum positions,

" loose "
particularly in relation to indefinite relations between economic
variables, see Edgeworth, Mathematical Psychics (1881, reprinted 1932), pp.
1-15 and pp. 83-93.
x fixed by the firm for maximum net revenue must satisfy the two
conditions :
~(B-U) =
^ '
and (JR-
^
m , n A TJ_- .
--- = 0. . dR -. A ,,, .
, ,.
The first condition is -r z i.e. At the output for
ax ax ax ax
equilibrium, therefore, marginal revenue must equal marginal cost.
The second condition, for a maximum as opposed to a minimum, is
_ d tdR dU\ __
d 2R d 2TI
~~ ~~
dxdx dx \dx dx 1 dx 2 dx 2
'
i.e. ^
dx 2 dx 2
Hence, at the equilibrium output, marginal revenue must be in-
creasing less rapidly than marginal cost. This second condition is
100 150 200 250 30O X

Tons per month
Fio. 68.
automatically satisfied if, for example, marginal revenue is decreas-

ing while marginal cost is increasing as output increases from the
value at which these marginal concepts are equal.
The position can be represented on a diagram in two different
ways. The and cost curves can be drawn on one dia-
total revenue
gram, taking output along the horizontal axis and revenue or cost
x
along the vertical axis. Suppose that, on the two curves, points in
the same vertical line can be found such that the tangents to the
respective curves are parallel f -= = -= ) . The output common to
such points is a monopoly equilibrium output provided only that the

total revenue curve is less convex (or more concave) from below than
the total cost curve (

-p^ < -r j
. . When the total revenue and cost
" "
curves are of normal form (see 10.7-8 below), the position is
illustrated by the curves of Fig. 58. The net revenue obtainable
from any output is shown by the vertical distance of the total revenue
curve above the total cost curve. This is clearly a maximum in the
position PQ, the tangent to the total cost curve at P being parallel
to the tangent to the total revenue curve at Q. The monopoly out-
put OM is thus uniquely determined in this case and the monopoly
price is read off as the gradient of OQ.
M.C.
.A.C.
50 100 150 200 250 N 300 X

Tons per month
FIG. 69.
The monopoly situation can also be represented on a diagram show-

ing the average and marginal revenue and cost curves referred to the
same axes and scales. Fig. 59 exhibits these four curves as obtained
from the total curves of Fig. 58. Three of the curves happen to be
"
straight lines, but this is not an essential feature of the "normal
case here taken. The unique monopoly output is now given as the
abscissa OM P
where the marginal curves intersect.
of the point
The monopoly price is M
R, the average cost of the monopoly output
isMQ and the maximised net revenue is QR times OM, i.e. the
APPLICATIONS OP DERIVATIVES 199
rectangular area shaded. The condition for a maximum, as opposed

to a minimum, is automatically satisfied here since the marginal
revenue curve falls while the marginal cost curve rises.*
The analytical solution of the monopoly problem can be illustrated
by assuming particular cost and demand functions of simple type
" "
appropriate to normal conditions. If the cost function is quad-
ratic in form 77 ax 2 + bx + c and if the demand function is the linear
form p=/3-ax (all the constants specified being positive), then the
first condition of equality of marginal cost and revenue is
There a unique equilibrium output, provided that /?> 6. This last

is
condition is simply that marginal revenue is greater than marginal
cost at zero, or very small, output. If it is not satisfied, the firm

never makes a positive profit and its losses are least when it produces
nothing. The second condition for equilibrium

^ (-r-s<T-s-) is
\dx 2 dx 2 1
always satisfied. Figs. 58 and 59 are drawn for cost and demand
functions of these types. A
sugar refinery produces an output of
x tons per month at a total cost of (^# 2 4-15a;-f 800) and mono-
on a market with a demand law p = 50 - -^x,
polises the.sale of sugar
where p is the price (per ton) of sugar. The monopoly equilibrium
output is just under 150 tons per month and the monopoly price is
approximately 35 per ton or rather under 4d. per Ib.
Taking the second view of the monopoly problem, the firm fixes
the price, and its output to meet the demand must be x </>(p).
Here
=xp=p(f>(p) and II=F(x)=F{<f>(p)}.
The net revenue at the price p is (R - 77) and this is a maximum if
-72-7J-0 and
* It is to be noticed that, as far as our formal solution is concerned, there is

no reason to suppose that net revenue is positive even when it is maximised.
The total cost curve may lie completely above the total revenue curve. In
this case, the firm either goes out of business or stays in and cuts its losses.
Maximum net revenue is then minimum net loss.
The first condition is
d ( j., v,
dn n
5(p*(p)>-5=o.
... ATI I dn dHdx\
+ (p) +pm --<f,>(p)=0
.,. .
.,,
. . .
^smce -^=-j- Tp )
The equation to be solved for the equilibrium price is thus
The limiting condition for maximum net revenue is
*I* R-n)\<o,
(
dp\dp^ j
-
f)
Any value of p and inequality is a
satisfying the above equation
possible monopoly price and the corresponding output is x = <f>(p).
The second analysis of the monopoly problem, which is due to
Cournot, can be shown to lead to the same equilibrium price and
output as is obtained in the first analysis. This fact can easily be
checked in particular cases, e.g. with a quadratic cost function and
a linear demand function. Or, in general terms, the above equation
for the monopoly price gives
dB dB dx
dfl _(f>(p)+p</>'(p) _dp dx dp dB
~ ~~ ~ ~~ '
dx <f>'(p) dx dx dx
dp dp
i.e. the condition of the first analysis that marginal cost and marginal
revenue are equal is satisfied also in the condition of the second
analysis.
8.8 Problems of duopoly.

The demand of a market for a good X is represented, as before,
by the demand relation p ^(x) connecting the price p and the
demand x. The production of the good is shared between two duo-
polist firms selling at the same price p. The first duopolist produces
an output xl at a total cost of J71 =-P1 (o:1 ) and the second duopolist
produces an output x 2 at a total cost of U

2 =F 2 (x 2 ). The solution
of the problem of the distribution of the market between the two
duopolists depends entirely upon what is assumed about the reaction
of one duopolist to any action on the part of the other.
It is assumed, in the simplest duopoly problem, that each duo-
polist expects the other to make no change in current output no

matter what changes he makes in his own output. Subject to this
expectation, each duopolist then aims at fixing his output for
maximum net revenue. If outputs xl and x 2 are fixed by the two
duopolists, the price of the good is determined by p = t//(x), where
x =X! + x2 is the total output. The net revenue of the first duopolist
is (x^p - IIJ and, for this to be a maximum, xl must be chosen so
that
d .
j-,
d .
This is the familiar equality between marginal revenue and mar-

ginal cost. The difficulty here is to express marginal revenue in a
suitable form. Since x1 p=x1 î(x), where x=x 1 -{-x 2) we have
making use of the assumption that the first duopolist considers x 2

as fixed. Hence, for any given output x 2 of the second duopolist,
the equation* which determines the first duopolist's output x is
In the same way, given any output xl of the first duopolist, the
second duopolist fixes his output x2 so that
These two equations are together sufficient, in general, to determine

the outputs of the two duopolists. The total output and the price
at which it is sold then follow at once.
The equation gives the output of the first duopolist in terms
first
of whatever output the second duopolist is producing, i.e. it gives xi

" "
as a function of x 2 It can be taken, under
. normal conditions,
that an increase in x 2 results in a decrease in x^ of smaller amount.
The dependence of x on x 2 can be represented by a " reaction
curve
"
C-L in the plane OxjX 2 The " normal " form of Cl is shown
.
in Fig. 60. The curve must be considered in relation to the axis Ox 2

and its gradient to this axis negative but numerically less than
is
unity. Similarly, the second equation gives x 2 as a function of x1

and a second reaction curve C 2
is obtained. This curve is
related to the axis Ox1 and, in

" "
the normal case, its gradient
to this axis is negative and
numerically less than unity.
The two curves O and G2 thus
intersect in a single point P and
the co-ordinates of P give the
equilibrium outputs x1 and x 2
of the two duopolists.
FIG. 60.
In the particular case where
the duopolists have the same
total cost function TIF(x), the equations giving their outputs are
The reaction curve Gl when

viewed from the axis Ox2 is now of
exactly the same form as the reaction curve C 2 viewed from the
axis Oxlf It follows that x and x 2 must be equal at the point of
intersection. Thus, as we expect, the total output is shared equally
between the two duopolists : xl x 2 = Ix. The value of x is given by
Further, if each duopolist produces at constant total cost, the total

output shared equally between them is given by
i.e. iff (x) + {iff (x) + xtft'(x)} = 0,

i.e.
The output is such that the sum of the average and marginal
total
revenue from the total demand p = if/(x) is equal to zero. If the
average and marginal revenue curves are as shown in Fig. 61, the
total output under duopoly is given by ON. This output can be
compared easily with that under pure monopoly. The monopoly

output of a single monopolist with constant total cost is such that
marginal revenue is equal to marginal cost, i.e. is equal to zero. In

Fig. 61, the monopoly output is OM
and this is less than the duopoly
,
output ON.
In conclusion, we can indicate the way in which the
general
duopoly problem can be analysed. It is assumed that the first
duopolist, when he varies his own output xlt expects the second duo-
polist to react and vary his output x 2 according to some definite law
x
M.R.
FIG. 61.
x2 f(x1 ). Thus, if he changes his output from a level xlt he expects

his rival's output to expand or contract at a rate indicated by the
dx
derivative Following Professor Frisch,* this derivative
-^=f(xl ).
can be termed a " conjectural variation " and it may be positive or

negative in value according to circumstances. For a maximum net
revenue, the first duopolist's marginal cost must equal
The equation, which gives x^ as a function of x2 and determines the

reaction curve Cl of the first duopolist, is now
In the same way, it is assumed that the second duopolist expects

the output of the first duopolist to vary according to a definite law
dx
xl g(x 2 ). The derivative -r^=g'(x 2 ) is again the conjectural varia-
*
Frisch, Monopole-Polypole, Nationalokonomisk Tidsskrift, 1933.
tion. The equation giving x 2 as a function of xt and defining the
reaction curve C2 of the second duopolist is
The two reaction curves, by their point or points of intersection,

determine the duopoly distribution of output between the firms.
Their forms depend on the nature of the conjectural variations of
the two duopolists and, by making various assumptions about
these conjectural variations, a whole series of duopoly problems can
be defined and analysed. The problem will be elaborated at a later
stage (13.9 below).*
8.9 A note on necessary and sufficient conditions.

In the above analysis we have had occasion to draw distinctions
" " " "
between necessary conditions, sufficient conditions and con-
"
ditions which are necessary and sufficient ". The following
observations serve to explain more fully the nature of these dis-
tinctionsand to illustrate their importance.
To start with a simple example, we can examine conditions under
which a four-sided figure is a rectangle. First, if the figure is a
rectangle, then one of its angles must be a right angle. This is a
necessary condition. While all rectangles have a right angle for one
angle, there are also other figures with the same property. If, how-
ever, one angle of the figure is a right angle and all sides are of the same
length, then the figure must be a rectangle. This is a sufficient con-
dition. All figures with the property stated are rectangles, but there
are some rectangles (i.e. those not squares) which do not display the
property. The condition is thus not complete, not necessary and

sufficient. Finally, if the figure is a rectangle, then one angle is a
right angle and opposite sides are of equal lengths. Conversely, if a
figure hasone angle a right angle and opposite sides of equal lengths,
then the figure is a rectangle. We have here a necessary and suffi-
cient condition it is complete, including all rectangles and no figures
;
other than rectangles.

In general, a necessary condition for a certain property is such
* The
analysis given above, in the case where the conjectural variations are
based on the work of Cournot. For the general problem, see Hicks,
zero, is
The Theory of Monopoly, Econometrica, 1935.
that, if the property holds, then the condition is satisfied. The con-
dition is satisfied by all things with the property but may be satisfied
also by things without the property. A sufficient condition is such
that, if the condition then the property holds. The con-
is satisfied,
dition is satisfied by no thing without the property but may not be

satisfied by some things with the property. A necessary and sufficient
condition one which holds in both the ways described if the pro-
is
perty holds, then the condition is satisfied if the condition is ;
satisfied, then the property holds. The condition is complete and

includes things with the property and no others.
all
An important instance of the distinctions here drawn

is provided
by the conditions for maximum and minimum values of a function

f(x) which has a finite and continuous derivative. A necessary con-
dition for a maximum value at x a is that f'(a) 0. This condition
is satisfied at all maximum positions but also at other positions (e.g.
minimum or some inflexional positions). A sufficient condition for

a maximum value is that f'(a) = Q and /"(a) <0. We have
a maxi-
mum value whenever this condition is satisfied but some maximum
values can occur even when the condition is not satisfied. A neces-
sary and sufficient condition for a maximum value is that f'(a)
and that f'(x) changes sign from positive to negative as x increases
through the value a. A maximum value must satisfy this condition
and we have a maximum value whenever the condition is satisfied.
EXAMPLES VIII
Oeneral applications of derivatives
1. Write down the derivative of 3# 3 -f 3# 2 + x - 1 and show that this

dt/
function is monotonic increasing. If y ax* -f bx 2
4- ex + d, express as a
square plus a constant term. Deduce that the function is monotonic if

6 2 < Sac and that it then increases or decreases according to the sign of a.
2. Show, by means of derivatives, that y ~ --

1
~
7 and y
1
-z
A
- x
are both
-
JL ~r~ *C
monotonic functions. Generaliseby showing that y is always a
monotonic function. When does it increase and when decrease? Illustrate

these results by considering the shape of the rectangular hyperbolas which
represent the function.
3. Show that y x -f - has one maximum and one minimum value and that
the latter is larger than the former. Draw a graph to illustrate.
4. Show that y = 2x - 1 + - has a single minimum value and is positive for
x
all positive values of x.
5. Find the maximum and minimum values of y = x 3 - 3x - 1 and of
y = 3rc* - lOz 3 a
+ 6 + 6. Illustrate graphically.
6. If y = x* - 4# 4- 6z -
8 2
4a; - 3, show that ^=4 (a
- 1 )8. Deduce, from
criterion for maximum and minimum values, that y has a minimum
value at
case?
7. Show
x=l. Why does the second criterion
that each of the functions t/

1
X
= -I--
_
\
-|-
and
x
t/
_
fail to give the result in this
2
= V3 - # a has a single
maximum value. Draw rough graphs of the functions to illustrate.

Show that y = x\/l +# has a minimum and y= -x*Jl+x a maximum
8.
value. Draw the graphs of the functions for x > - 1 and deduce that the
double-valued function y 2 = x 2 1 + x) is continuous with a maximum and a
(
minimum at the same value of x. Then consider y = + s/a; 2 (l +a?) and

y - */#* 1 + x) as the two single-valued branches of this function, showing
(
that each branch is continuous but without a derivative at x = 0. Illustrate

the difficulty of defining derivatives for multi- valued functions.
9. Show that y
2
x(x*
- 1) can be divided into two single-valued branches,
one with a maximum and the other with a minimum value. Deduce that the
curve representing the double-valued function is continuous, denned only for
certain ranges of x and shows a vertical tangent at throe points.
10. Find the derivative of y = \fx* and show that it is infinite at x 0.

Draw a graph of the function and indicate its behaviour in the neighbourhood
of the origin. Deduce that y has a minimum value at the origin which is not
a stationary value. Contrast this function and its graph with y = $x (see
Examples VII, 21).
11. Show that the perimeter of a rectangle of area 16 square inches is least
when the rectangle a square of side 4 inches.
is
12. A rectangular area is to be marked off as a chicken run with one side
along an existing wall. The other sides are marked by wire netting of which
a given length is available. Show that the area of the run is a maximum if
one side is made twice the other.
13. A
cricket field consists of a rectangle with a semicircular area at each
end. The perimeter
is to be used as a quarter-mile running track. Find the
dimensions of the field so that the area of the rectangular portion is the largest
possible. (Take IT = 3-14159.)
14. A
tinned soup manufacturer uses tins which are circular cylinders
closed top and bottom. Find the most economical dimensions of the tin
(i.e. minimum surface area) when the volume is given.
If the top and bottom
of the tin are cut from square sheets and the surplus wasted, find the new
dimensions for greatest economy.
15. Express the distance of O from a point on the line 2x + y 6 as a function
of the oj-co-ordinate of the point. Find the point on the line nearest O and
deduce that the shortest distance from O to the line is perpendicular to the
line and of length \/5. Generalise to show that the shortest distance from
dXt + by-i + c
(x l9 yj to the line ax + by + c = is
APPLICATIONS OP DERIVATIVES 207
16. Find, on the part of the rectangular hyperbola xy = 4 in the positive
quadrant, the point which is nearest to O and show that the shortest distance
is perpendicular to the tangent at this point. What is the shortest distance of
the point (0, 2) from the parabola y = x 2 ?
17. Show that the curve y 2x - 3 + - is convex from below for all positive
x c
values of x. Is the same true of the curve y ax + b + -1
x
18. Show that the curve y = ax* + bx* + ex can have only one point of in-
flexion. If a is positive, show that the curvature changes from concave to
convex from below as we pass through the inflexional point from left to right.
Deduce that the point of inflexion is also a stationary point if b z = 3ac.
--
%x
19. Show that the curve y - has three points of inflexion separated
x + 1
by a maximum point and a minimum point. Verify these facts by drawing a

graph of the curve.
20. Prove that the curve y = x* has a single stationary point which is a point
of inflexion. Are there any other points of inflexion ?
21. Show that the curve y = $x is convex from below for negative values
of x and concave from below for positive values of x. Deduce that the origin
is a point of inflexion. Why is this point not given by the criterion that the
second derivative is zero ? Check the result by considering the function as the
inverse of y = x 3 .
22. From the second derivative, verify that the rectangular hyperbola
xy 1 is concave from below for negative values of x and convex from below
for positive values of x. In what sense is x a point of inflexion ? Contrast
this case with that of the previous example.
23. Iff(x) is a single-valued function of x, find where (xf(x)} attains maxi-

mum and minimum values and interpret in terms of the curve y /(#). If
x and f(x) are' both positive, show that {xf(x)} can only be a maximum at a
point where the curve y ~f(x) is downward sloping with a curvature less than
a certain amount.
Economic applications of derivatives
-
x+b
downward sloping and convex from below. Do the same properties hold of
the marginal revenue curves? Show further that, for each of the demand
laws p = *Ja-bx and p a- bx z f the demand and marginal revenue curves are
downward sloping and concave from below.
25. Show that the demand curve o? = 4a 3 - 3ap 2 +p 3 whore a is a positive
,
constant, and p is lessthan 2a, is downward sloping with a point of inflexion.

How does the curvature change?
26. With the aid of derivatives, check the positions of maximum total
revenue obtained in the cases of the demand laws of Examples V, 2, 10 and 11.
27. It is given that a demand curve is convex from below f

^ >0j at all
points. Show that the marginal revenue curve is also convex from below
either if ~~ is positive or if is -
negative and numerically less than -r~ .
-^
If the demand curve is always concave from below, does a similar property
hold of the marginal revenue curve?
28. By examining the signs of certain derivatives, show that each of the
total cost functions
/j
i A
77=vaa? + & + c and 77= ax t-d (b>c)'
v
te +c
gives average and marginal cost curves which fall continuously with increasing
output.
29. From the sign of the second derivative, show that the transformation
curve of Examples V, 25 is concave from below at all points.
30. If the supply of a good is related to its price by the law x=a>Jp-b + c,
where o, b and c are positive constants, show that the supply curve is upward
sloping and concave to the axis Op at all points. Illustrate with the case of
Examples V, 23.
31. An indifference map is defined by the relation (x + h)*Jy + k a, where
h and k are fixed positive numbers and a is a positive parameter. By
expressing y as a function of x and by finding derivatives, show that each
indifference curve is downward sloping and convex from below.
32. If U=ax 2 + bx + c is the cost function of a monopolist and if p = )3 - ocx

is the demand law, find the monopoly price and output when the monopolist
is assumed to fix the price. Verify that this is the same result as when the
monopolist fixes the output.
33. A radio manufacturer produces x sets per week at a total cost of
(gx z + 3# + 100). He is a monopolist and the demand of his market is
x = 75 - 3p, when the price is p per set. Show that the maximum net revenue
is obtained when about 30 sets are produced por week. What is the monopoly
price? Illustrate by drawing an accurate graph.
34. If the manufacturer of the previous example, with the same costs,
-
produces for a demand of x 100 20 *Jp sets per week, show that he should
produce only 25 sets per week for maximum monopoly revenue. What is the
monopoly price now ?
35. In the case of Example 33, a tax of k per set is imposed by the gover-
ment. The manufacturer adds the tax to his cost and determines the
monopoly output and price under the new conditions. Show that the price
increases by rather less than half the tax. Find the decrease in output and
monopoly revenue in terms of k. Express the receipts from the tax in terms
of k and determine the tax for maximum return. Show that the monopoly
price increases by about 33 per cent, when this particular tax is imposed.
36. Generalise the taxation problem of the previous example by finding the
effects of a tax of k per unit of output when a monopolist's total cost is
II ax 2 + bx + c and the demand law is p p - ocx. Show that the tax brings
in the maximum return when & = (0 - 6) and that the increase in monopoly
price is always less than the tax.
37. If a monopolist has a total cost of n=ax 2 + bx + c and if the demand
law is p = j8 - ax 2 show that the output for maximum revenue is
,
How many sets per week should the radio manufacturer of Example 33
produce when the demand is x 10^25 -p sets per week?

38. Tho demand of a monopolist's market is p - ax and ho
j8 produces an
output of x units at a total cost of 77 ax - bx + cx + d. Show that tho output
3 z
for maximum net revenue is the positive root of
3ax*-2(b-<*)x-(p-c) = Q.
Taking the cost conditions of the firm of Examples V, 18, and the demand law
p = 50 - fJG, show that the firm must produce 3^ tons of its product per month
for maximum monopoly revenue. What is the monopoly price per ton?
39. A
firm with a total cost function 11 F
(x) sells on a perfectly competi-
tive market, the market price being fixed at p. Show that the output of tho
firm for maximum net revenue is such that marginal cost equals p, provided
that total costs are covered. Deduce that there is a supply relation for the
firm, giving the output as a function of the market price p.
40. If F (x) = ax 2
in the previous example, show that tho supply
+ bx + c
relation is linear . Show
that p must exceed b -f 2 Vac if total costs are to bo
covered but that, if only variable costs are to bo covered, p need only exceed 6.
Illustrate the determination of the supply curve by drawing a diagram show-
ing the average and marginal cost curves.
41. A
sugar refinery has total cost equal to (-rjtf + 5x + 200) when x tons
2
of sugar are produced per week. The fixed market price is p per ton. What
is tho supply curve of the firm? What is tho lowest price to cover total costs?
At what price will 150 tons be produced?
42. A plant produces x tons of steel per week at a total cost of
3x* + 50x + 300).
If the market price is fixed at 33 6s. 8d. per ton, show that the plant produces
16f tons per week.
43. In the problem of the previous example, show that the plant's output
at the fixed market price of p per ton is tho root of
which is greater than 10 tons per week. What is tho smallest price for total
costs to be covered? Show that tho supply curve is
for values of p greater than this minimum amount. Connect the supply curve
with the marginal cost curve and illustrate graphically.
44. The market demand for a good is given by p - <x.x. The market is
supplied by two duopolists with cost functions "
1
" are
U
= a^x^ + b lx l + c l and
77 2 = a 2 2 2 + &2a; 2 + c 2- Assuming that the
.
conjectural variations zero,
show that the reaction curves are straight lines. Deduce the equilibrium
output of each duopolist.
46. The
duopoiists of the previous example are radio manufacturers pro-
ducing identical sets. The total cost of an output of x sets per week is
(Aa + 3-7: + 100) in each case. When the price is p per set, the market
*
j
demand is x = 75 - 3p sets per week. Show that the total equilibrium output
is approximately 41 sets per week. Compare with the monopoly output of
Example 33.
O M.A.
46. If the market demand of the previous example is #=10N/25-> sets
per week, show that the reaction curve of the first duopolist is
2x 2 - 4)
and second duopolist. Draw a graph showing the two re-
similarly for the
action curves and deduce that the total output is now approximately 32 sets
per week. Compare with the monopoly output of Example 37.
47. A with fixed plant, supplies of raw materials, etc., produces two
firm,
goods X
and Y in amounts related by the transformation function yf(x).
The market prices of the two goods are fixed at p x and p y Show that the
.
outputs for maximum total revenue are such that -f'(x) equals the ratio of
Px to p y .
48. If the firm of the previous example monopolises the sale of both A"
and F, the demand curves for andX Y
being given and independent, show
that the outputs for maximum total revenue are such that -/'(#) equals the
ratio of the marginal revenues from the X
and Y
demands.
49. A steel plant is capable of producing x tons per day of a low grade steel
40 5x
and y tons per day of a high grade stool, where y = -^.--- If the fixed
AU ~ X
market price of low grade steel is half that of high grade steel, show that
about 5 tons of low grade steel are produced per day for maximum total
revenue.
50. The steel producer of the previous example monopolises the sale of both
quality steels. If the prices of low and high grade steel arc p x and p v per
ton, the demands are p x 20 - x and p y 25 - 2y. Find an equation giving
the output x of low grade steel for maximum total revenue. Show, by a
graphical method, that just under 6 tons of this steel are produced por day.
CHAPTER IX
EXPONENTIAL AND LOGARITHMIC FUNCTIONS
9.1 Exponential functions.
WE have been concerned, so far, with power functions and with the
variety of functions that can be derived, in one way or another, from
power functions. Such functions are of great practical importance
and they serve to describe, accurately or approximately, many of
the ways in which one variable depends on another. It is now con-
venient to extend the range of our function concept by the definition
of an entirely new function type.
The power function is represented by the general form y~xn ,
where n is any given number. A new function can be defined by the
simple process of taking the base of the power as a fixed number and
the index as variable. So, instead of writing a fixed power of a
variable number, we write a variable power of a fixed number. The
new function so obtained is called an exponential function and we
write
ya x
,
where a is Since a power can be

the fixed base of the function.
defined for all values of the index, the exponential function is a
function of a continuous variable x.
found convenient, in general, to limit the value that can be
It is
allotted to the base a ofan exponential function. If a is negative,
then many values of a x (e.g. when x is or ~-|) are not defined in
terms of the real number system. To avoid this difficulty, we always
take a as a positive number. If a is a positive fraction (between
and 1) and if we write b = -> 1, then
i.e. any power of a number less than unity can be reduced to a power
of a number greater than unity, the sign of the index being reversed.
There can be no objection, therefore, if we take the base a as a
number greater than unity. This is assumed throughout the following
development.*
The graph of the exponential function can be plotted once the
base is given a definite value. For the particular function y = 2 x a ,
table of values of x and y can be obtained :
the value of y being taken, where necessary, to two decimal places.

It is seen, from Fig. 62, that the points plotted from this table can
be joined by a smooth line which is the curve representing the func-

tion y = 2 X The graph shown is constructed for selected rational
.
values of x only. But, for an irrational value of x, the value of 2 x is

defined as the limit of a set of rational powers and the corresponding
point must fit into the graph in a continuous way. The function
y=2
x thus continuous.
is
The graphs of other examples of the exponential function type

y=a x are of exactly similar shape. It will be seen later that the
graph of one exponential function (e.g. y = lO x ) can be obtained from

* Notice that the cose a 1 is trivial. The exponential function is then the
constant y 1.
EXPONENTIAL AND LOGARITHMIC FUNCTIONS 213
" " "
that of any other (e.g. y = 2 x by a process of stretching or
)
con-
"
tracting in the direction of Ox. The general shape of the curve
y=a x is always the same its steepness or gradient varies with the
;
value allotted to a and that is all.

We conclude that y =a x is a single-valued and continuous function.
monotonic increasing, y increasing over the
It is also seen that it is
whole range of positive values as x increases from infinitely large
negative to infinitely large positive values. It follows that, given
any positive number p, we can find a unique value q so that p=a

q .
An approximate value of q, for given values of a and p, can be read

off the graph of the function y=a x as the abscissa corresponding to
the ordinate p. We have, therefore, the important result that any
positive number can be expressed as a power of a given number greater
than unity.
It can be shown that, in general, irrational values of q correspond
to rational values of p, and conversely. To illustrate the nature of
the correspondence between p and q, we can take the useful base 10
and write, for example,
34 = locals and 7240 = 10 3 8597

'
the indices being irrational numbers written correct to four decimal

places. Again
10^ = 2-1528 and 10* = 31-6228,
these numbers being irrational and given to four decimal places.
The result we have just given leads us at once to the consideration
"
of what are called logarithms ".
9.2 Logarithms and their properties.
The logarithm of a positive number p to the base a (>1) is defined

as the index of that power of a which equals p. In symbols :
DEFINITION : If p=ag ,
then q=log a p.
Since a logarithm is simply the index of the power to which the base
must be raised to obtain the given number, it follows, from what has
been said above, that any positive number has a unique logarithm
to a definite base greater than unity. On the other hand, a negative
number has no logarithm to any such base.
The logarithms of practical work have the convenient number 10
as their base. Since, to take two examples already quoted,
34 = 1C 1 7240 = 10 3
'
5315 '
8597
and ,
it follows that
Iog10 34- 1-5315 and Iog10 7240 = 3-8597.

These logarithms are correct to four decimal places. The two nota-
tions, one in power form and the other in logarithmic form, are
simply two ways of saying exactly the same thing.

The logarithms to the base 10 are called common logarithms and
their values have been calculated and set out in tables of logarithms
to four, five or more decimal places. Hence, to find the common
logarithm of a given number,we have only to look up the number in
the tables. The numbers 1-5315 and 3-8597 given above were, in fact,
obtained in this way. No attempt is made here to give an account
of the way which common logarithms are used in practice to
in
facilitate There are many practical devices to
numerical work.
remember and it is assumed that the technique is familiar or can be
obtained by reference to a text-book on algebra. The general laws
of logarithms, given below, provide the basis of the practical work.
Returning to logarithms to any base a greater than unity, two
particular cases are derived from a
Q
1 and al =a :
log a 1~0 and log a a = 1.
Three general laws are obeyed by all logarithms :
LAW I.The logarithm of a product.

The logarithm of a product of two numbers is the sum of the
separate logarithms :
lg (Pi
'
#2) = lga#i + Iog a p 2 .
LAW II. The logarithm of a quotient.

The logarithm of a quotient of two numbers is the difference of
the separate logarithms :
LAW The logarithm of a power.

III.
The logarithm of a power is the index times the logarithm of the
base of the power :
The laws can be extended to the case of products or quotients of
more than two numbers and they can be combined to give the
logarithms of complicated products, quotients or powers. Since the
logarithm of 1 is zero, we obtain a particular, and useful, case of
Law II :
Notice, also, that the powers of p to which Law III applies include
fractional powers. For example, we can write
Since logarithms are indices, the laws of logarithms are simply

translations of the index laws. For example, since indices are added
when powers are multiplied, we expect that the logarithm of a pro-
duct is the sum of the separate logarithms. Formal proofs of the
three logarithm laws, based on the index laws, are as follows :
In Laws I and II, let
logaPi - q1 and Iog a p 2 q2

so that pl aft and p 2 = aq *.
Then pl .
p 2 ~ aft aq = a qi+92
.
*
y
Pi_aft_ aqi _ qt _
p 2 ~a**~
So log a (pi .
p 2 ~ql + q 2
)
~ log a ^! + Iog a p 2 ,
and log a = ql - q 2 = log aPl - Iog a p 2 .
\Qj
In Law III, let log p =qso
a that p ~a q .
n
Then p
n
i.e. lga(p )
:
The use of logarithms, both and practice, is evident from

in theory
the three laws. Expressions involving sums or differences are not
easily treated by the use of logarithms. The logarithm of a sum or
difference is not reducible and it must be stressed, in particular, that
loga (#1+^2) is not equal to the sum of log a and Iog a p 2 On the ^ .
other hand, by taking logarithms, an expression involving products

and quotients is much simplified. The use of logarithms here
replaces multiplication and division by the simpler processes of
addition and subtraction. Further, logarithms are equally useful in
dealing with powers and provide, for example, the only simple means
of solving an equation in which the variable appears as an index. As
a very simple instance, consider the equation :
3 2*-4:=:0 or 2*=|.
Here log (2*)= log (f),

i.e.
It*
.6.
x log 2
<Y
*/
-
i
_
-
log 4 log
log 4 - log
log
~
2
3
3,
The logarithms can be to any base. Taking the base as 10 and look-
ing up the logarithms concerned (to four decimal places) in tables,
thon 0-6021 -0-4771
*-
__ _ 125
6*010 "30l"~' 4IO>
correct to three decimal places. Other instances of equations most
readily solved with the aid of logarithms arise in problems of com-
pound interest (see 9.6 below).
Since the base of a logarithm can be chosen arbitrarily and for
convenience, it is useful to have the following law which connects
logarithms to one base with those to another base :
LAW IV. The charge of base in a logarithm.

The logarithm of a number to the base a is the logarithm of the
number to the base 6 times the logarithm of b to the base a :
log a p=log a blog b p.

As a particular case, put p a. Then, since Iog a = 1, we have
Iog 61og 6 a =
fl l, i.e. Iog a 6
The change of base law can thus be written in two alternative ways :
lo go ^ = log a b log b p =
The formal proof of the law proceeds :
Let log&# q so that p bq .
Then =log a (6) =q log a b=^log a b log b p.

log a p
The change of base law shows that the logarithms of a set of
numbers to one base are simply constant multiples of the logarithms
of the same set of numbers to another base. To change the base of
a whole set of logarithms is only a matter of multiplying each
logarithm by a constant. For example, logarithms to the base 2

can be obtained from common logarithms :
Iog 2 p
= (Iog 2 10) loglo p =
From tables of common logarithms, therefore, it is possible to con-
struct tables giving logarithms to any base other than 10.
9.3 Logarithmic functions.

If the variable y isthe value of the logarithm of the variable x to
a given base a, we obtain the logarithmic function
y=log a x.
The between x and y defined by the logarithmic function is
relation
not, however, a new one. The logarithmic function, in fact, is simply
the inverse of the exponential function. This follows from the
definition of a logarithm; if y~log a x, then x~a v Since the .
exponential function is single-valued, continuous and monotonic

increasing (a>l), the logarithmic function, as its inverse, possesses
exactly similar properties.
For any definite value of a (>1), the graph of ylog a x is the
x
graph of the exponential function y a with the axes Ox and Oy
transposed. Alternatively,
the graph can be obtained
directly from tables of log-
arithms. The graph of
y=log2 x
is shown in Fig. 63. It can
be derived from the graph of
y = 2 shown in Fig. 62, by
X
,
transposing the axes.

Logarithms to one base
are constant multiples of
logarithms to another base.
-2-
It follows that the graph of FIG. 63.
y=logb x can be derived

from that of y = log a x by multiplying all ordinates by a constant
" " " "
amount, i.e. by stretching or contracting the graph in the
direction of Oy. All logarithmic functions have the same general
shape when represented graphically ;

the steepness of the graph
varies with the value given to the base but the general shape of the
graph remains. Further, since the graph of an exponential function

is the graph of the corresponding logarithmic function with axes
transposed, one exponential graph is obtainable from another ex-

" " " "
ponential graph by a process of stretching or contracting in
the direction of Ox. This fact was stated above it has now been ;
justified.
The graph of a logarithmic function shows the following properties.
There is no logarithm of a negative number the logarithm of a ;
positive number less than unity is negative the logarithm of the ;
number 1 is zero the logarithm of a number greater than unity is

;
positive, increases and tends to infinity as the number increases.

These properties are obtainable, of course, directly from the defini-
tion of a logarithm but they are particularly clear from the graph.
It is interesting to compare the three function types we have now
considered. In their simplest forms, we can write
where a and a are greater than unity and where n is taken as positive.
As x tends to infinity, so does the value of each function. But it
can be shown that log o; tends to infinity more slowly than xn and the
tt
x
latter, in its turn, tends to infinity more slowly than a So, for a .
large value of x, the functions ascend in order of magnitude :
loga x<x <an x

.
The proof of this statement can be given as follows.

If h is positive and if k is a positive integer, the Binomial Theorem
(proved in text-books on algebra) gives
Let Ja = l+h where h is positive and suppose that the number x lies
between the integers k and (k + 1), i.e. k <#<& -f 1. Then

la* _(l+hy (!+&)* l+lch
*
~ jch_
x
ax
Hence, is greater than %h*k and the latter tends to infinity as k,
x
and so x, tends to infinity. So
a*
--*oo as #-><*> .
x
X
'
T- a x ~ (a\ n (la*\
n . x
Now 1
[ ) where yv -,
xn \xJ \ny] n
and this expression tends to infinity as y, and so as x, tends to
infinity.
xn
Finally n where z=nlog
*"a x,
loga x z
and this tends to infinity as z, and so as x, tends to infinity.
a a* j x"
bo ><x> and ,
>-oo as x-+<x>
Xn
,
10ga#
n and a x must be
and loga #, x in ascending order of magnitude if x
is given any large value.
The result can be illus-
trated by comparing the
three functions
graphically. Fig. 64 shows

the graphs of the three
functions plotted on the
same scales. It is clear that
the graphs rise to the right
at different rates and that
-2 10 x
for any large value of x.
9.4 Logarithmic scales and graphs.

The introduction of logarithms enables us to extend the process
of representing a function graphically by measuring the variables,
" "
not on the familiar natural or numerical scales, but on what are
" "
called logarithmic scales. The advantages and applications of
this radical change in the graphical method are described in the
present and following sections.

A
variable x is measured according to some definite scale and a
by points on an axis Ox. The usual

series of its values is represented
method to take distances along Ox from a base point
is as equal,
or proportional, to the values of x plotted, a method which gives the
natural scale for x. Now suppose that the points are plotted at
distances along Ox which are equal, or proportional, to the logarithms
of the values of x concerned, the point marked x on the axis being a
distance log x from the base point. It is convenient, but not
essential, to take logarithms to the base 10 from this purpose. A

logarithmic scale for x is thus obtained.
The important characteristics of the logarithmic scale as compared
with the natural scale are best introduced by considering some
numerical examples. Of the two sequences of numbers
100, 150, 200, 250, 300,
100, 150, 225, 337-5, 506-25,
the shows a regular increase of 50 units and the second a regular

first
increase of 50 per cent, from one number to the next. On a natural

scale, the points representing the first sequence appear at equal dis-
tances from each other and those representing the second sequence
at increasing distances along the axis. The logarithms are
2, 2-176, 2-301, 2-398, 2-477, ... ,
2, 2-176, 2-352, 2-528, 2-704, ... .
Hence, on a logarithmic the second sequence that gives

scale, it is
points at equal distances from each other and the first sequence
shows points at decreasing distances along the axis. In the same
way, the two decreasing sequences of numbers
100, 80, 60, 40, 20,
100, 80, 64, 51-2, 40-96,
show decreases of 20 and of 20 per cent, respectively. The first

units
sequence is represented by points at equal distances on a natural

scale and at increasing distances (to the left) on a logarithmic scale.
The second sequence corresponds to points at equal distances on a
logarithmic scale and at decreasing distances (to the left) on a
natural scale.
It appears, therefore, that equal distances between points on a
natural scale indicate equal absolute changes in the variable, and
equal distances between points on a logarithmic scale indicate equal
proportional changes in the variable. This property is easily verified
in general. If xl9 x 2 x z are values shown by points at equal distances
,
on a natural scale, then

and the variable increases by equal absolute amounts. The same

property on a logarithmic scale implies that
log # 3 - log x 2 ^log x 2 - log xl9
*^3 *^2
log --= log-?,
i i
i.e.
and we have equal proportional changes in the variable. The obverse

of this property is that equal proportional increases are shown on a
natural scale by points at distances from the base point which
increase more and more rapidly, while equal absolute increases are
shown on a logarithmic scale by points at distances along the axis
which increase more and more slowly.
If points are plotted from a function y=f(x) by taking x along the
horizontal axis Ox on a natural scale and by taking y along the
vertical axis Oy on a logarithmic scale, we obtain the semi-logarithmic
graph of the function. The graph shows a curve drawn through the
plotted points and the varying height of the curve shows, not the
variation of y with x but the variation of log y with x. The curve
9
isobtainable, of course, as the natural graph of log y as a function

of x. If the curve is seen to rise through equal heights over certain
ranges of values of x, then the value of y is subject, not to equal
absolute increases, but to equal proportional or percentage increases.
The use of semi-logarithmic graphs is thus clear. If we wish to
compare percentage changes in the value of one variable as the

other variable increases steadily, we plot the relation between the
variables on a semi-logarithmic graph. If it is absolute changes
that are important, we plot the relation on a natural graph.
Many statistical time-series are appropriately plotted as semi-
logarithmic graphs. Time-intervals are taken, on a natural scale,

on the horizontal axis. The variable of the time-series, to be plotted
on the vertical axis, may be such that proportional changes in its
value are important. Trade, consumption or employment figures
for an increasing population are cases in point. In such cases, a
logarithmic scale is used for the vertical axis and a semi-logarithmic
graph is drawn. An example is given in the following section.
Another application can be quoted in which the logarithmic scale is
used for the independent variable measured along the horizontal
axis. Suppose y represents the number of families with incomes
" "
over x. Then the relation of y to x (the cumulative frequency
distribution of incomes) is best plotted with y on a natural and x on
a logarithmic scale.The characteristic that is most important is
the variation of numbers for proportional changes in income.*

The logarithmic graphing of a function can be carried one stage
further by taking both variables on logarithmic scales, in which case
a logarithmic graph is obtained. The curve representing the function
y=f(x) now shows the variation of log y as log x changes and corre-
lates proportionalchanges in y with proportional changes in x. So,
ifa curve on a logarithmic graph shows equal rises for equal moves
to the right, then the result of increasing x by equal percentage
amounts is to y by equal percentage amounts. The
increase
logarithmic graph is when proportional changes in both

thus useful
variables are important. For example, variables which are related
for very small and, at the same time, for very large values are not
easily shown on natural scales. Taking logarithmic scales, how-

ever, reduces the large variations to reasonable proportions while
magnifying the small variations. f
Logarithmic graphs are extremely useful, though perhaps not well
understood, in the statistico-economic field. Pareto's Income Law,
for example, asserts that a logarithmic graph shows a straight line
relation between x and y, where y is the number of persons with
incomes over $Lx.\ As income rises proportionally, the number of
persons with that or higher income falls off in proportion. The main
use of logarithmic graphs, however, is in the statistical correlation
of economic variables by means of scatter diagrams (see 2.2 above).
When we correlate, for example, the populations and unemployment
rates in urban and rural areas of varying size, the variations are
great and percentage changes are more significant than absolute

changes. It is thus appropriate to plot the scatter diagram by
logarithmic graphing. An example of a correlation scatter diagram

* See
Gibrat, Les Intgalites iSconomiques (1931) and Allen and Bowley,
Family Expenditure (1935), pp. 63-5 and Diagram XIII.
f This follows since log x increases more rapidly than x at first and then
more slowly than x (see the graph of the logarithmic function).
See Pareto, Coura d'tconomie politique (1897), Book 3, Chapter
J 1 and
Bowley, Elements of Statistics (4th Ed. 1920), pp. 346-8.
plotted on logarithmic scales is given in the following section.

Another important example of the use of a logarithmic scatter
diagram arises in the statistical determination of demand or supply
relationsfrom the correlation of market data of prices and quantities
demanded or supplied. The important thing here is the proportional
change in price or quantity from one year to another, or in one year
"
as compared with the norm ".
9.5 Examples of logarithmic plotting.

The methods of logarithmic graphing, described above in general
terms, are most clearly appreciated by giving actual examples of
their use. The two cases considered in detail below illustrate very
different applications of the methods.
Ex. 1. Value of total imports, U.K., 1820-1897.
The variation in the value of U.K. imports, as given in the above table,
isrepresented graphically in Fig. 65 by two different methods. The first
graph takes a natural scale for the measurement of import values and the
second graph a logarithmic scale. In each case, the years are measured
along the horizontal axis on a natural scale and each import value is
plotted at the centre of the averaged period.

It is seen that changes in import values are almost imperceptible, on
the natural scale, in the early years and very marked in the later years.
But this hides a very important fact. The relative changes from one year
to another are roughly of the same importance in the early years as they
are in the later. The logarithmic scale brings out this fact, the variations
IMPORTS, U.K., 1820-97.
imports j i
Mn.
500
400
300
200
100
1820 1830 1840 1850 1860 1870 1880 1890 Years
Imports
Mn. *
500
400
300
200
100
75
50
25
1820 1830 1840 1850 1860 1870 1880 1890 Years'
FIG. 65.
in the height of the second graph being of comparable magnitude at the
beginning and end of the period. These variations represent relative

changes in import values. For example, imports were valued at f 65Mn.
per year in the period round 1842 and 77Mn. in the period round 1845.
The increase was 12Mn. or 18-5 per cent, of the earlier figure. In the two
periods centred at 1887 and 1890, the increase was 61Mn. or 166 per
cent, of the 1887 figure. The natural graph shows the second increase as
roughly 5 times the first. The semi-logarithmic graph shows the second
percentage increase as less than the first. For most purposes, the percen-
tage increases are the more interesting and the semi-logarithmic graph is
to be preferred.
Ex. 2. Number of insured workers in employment, U.K., 1929-1935.
The above table provides a picture of the decline in employment be-

tween 1929 and 1932 in various important industries and of the corre-
p M.A.
spending recovery between 1932 and 1935.* A graphical representation
can be given by taking per cent, changes in the first period along the
horizontal axis and per cent, changes in the second period along the
vertical axis. One point is then plotted for each of the 25 industries, the
height of the point indicating the per cent, change 1932-5 and the hori-
zontal distance the per cent, change 1929-32. For convenience, the
changes are referred to 100 as base in each case. Thus the decline in the
coal-mining employment in the first period is represented by 70-4 (29-6

per cent, decrease) and the recovery in the second period by 105-1 (5-1 per
cent, increase). The point representing the industry has co-ordinates
(704, 105-1). It remains to determine whether natural or logarithmic
scales are the more appropriate. It is clearly convenient to represent
an industry which has recovered completely (1935 employ mental 929
employment) by a point with distance to the left of the 100 mark equal
to height above the 100 mark. This is achieved by taking logarithmic
scales. Suppose, for example, that employment falls to four-fifths its
1929 value in 1932 and then recovers completely by 1935. This implies a
20 per cent, decline and a 25 per cent, recovery in the two periods, i.e. the
plotted point is (80,125). Now
log 80
- log 100 = - -
i.e. (log 125 log 100).
So, on logarithmic scales, the point (80,125) is as much to the left of the
100 mark as it is above the 100 mark, as required.
A logarithmic scatter diagram of the data is shown in Fig. 66, the points
representing the decline and recovery of the various industries. Most
industries are shown by points in the N.W. quadrant, a decline followed
by a recovery. The broken line, sloping downwards at 45, corresponds

to complete recovery, so that industries shown by points below the line
have failed to recover completely and industries shown by points above
the line have more than recovered their employment position. Six of the
industries have increased employment in both periods, as shown by the
points in the N.E. quadrant. The cotton industry, shown by a point in

the S.W. quadrant, has a progressive decline in employment. Finally,
Public Works employment has changed, as we should expect, in the
opposite sense to the majority of industries and is shown by a point in
the S.E. quadrant. Since the point is above the broken line, the later
decline was not sufficient to reduce employment to the 1929 level.
* No correction has been made for the increase in the total

insured popu-
lation over the period 1929-35. The picture thus shows the recovery in a
somewhat too favourable light.
It is we have marked our logarithmic scales

to be noticed that
according to equal intervals of the variable. In the vertical scale

of Fig. 65, for example, the markings are at 10, 20, 30, ... (Mn. of
imports). The characteristic appearance of the logarithmic scale is
then that the markings get closer and closer as we proceed up the
scale.* Another point of importance is that a logarithmic scale
avoids any difficulty about choice of zero mark or of units, a difficulty
which is evident in using natural scales. In fact, a fixed distance
on the logarithmic scale represents a given per cent, change no
matter where we are on the scale and we have never to refer distances
NUMBER OF INSURED WORKERS IN EMPLOYMENT, U.K., 1929-35.
40 50 60 70 80 90 10O 110 12O 13O

FlQ. 66.
(as in the case of the natural scale) to the fixed origin or zero mark.
The logarithmic has a clear advantage over the
scale, in this respect,
natural scale. A minor difficulty of the logarithmic scale is that only
positive values of the variable can be plotted (since negative numbers
have no logarithms). An adjustment can always be made, however,
to avoid negative values, as is seen in Ex. 2 above, where negative
percentages are converted to positive numbers below 100.

A functional relation between x and y can be represented, there-
ways on a diagram, as a natural, a semi-

fore, in three different
logarithmic or a logarithmic graph. The differences between the

*
'
Special graph paper, marked in this way either for semi-logarithmic or
for logarithmic graphing, is provided commerically. But ordinary graph
paper can bo used and adapted for the purpose by plotting distances directly
from the logarithms.
three methods are well brought out by considering what relation is
represented by a straight line in each case. On natural scales,
is the relation represented by a straight line graph. If y is plotted

on a logarithmic scale against x on a natural scale, the relation
y=ab x
is represented by a straight line graph. For, we can write the
relation as log j/=log a + xlog fc, i.e. logy is related linearly to x.
Finally, on a logarithmic graph, the relation
y~ax b
appears as a straight For, the relation can be written as
line.
log y= log a + b log x and logy is related linearly to logx. So, an

exponential function appears as a straight line on a semi-logarithmic
graph and the line's gradient is the logarithm of the base of the
exponential. A
power function is shown as a straight line on a
logarithmic graph and the gradient is the constant index of power.
If y is the number of incomes over x, for example, Pareto's Law is
y~ y/ ,
where a and m are constants. Plotting on a logarithmic
graph, we downward sloping straight line with a numerical

obtain a
gradient equal to the constant m.
9.6 Compound interest.
A sum of 100 is invested and accumulates at compound interest

at the rate of 4 per cent, per year. If interest is added yearly, then
100 + 4 = 104 = 100(1-04)
is the amount at the end of the first year. Again,

104 + y|ol04 = 104 (1-04) = 100 (1-04) 2
is amount at the end of two years. Similarly, 100(1-04) 3 is the

the
amount after three years, and so on. In general, if y is the amount
of the investment after x years, then
y = 100 (1-04)-.
Suppose now that interest is added twice a year. Then, with a rate
of 4 per cent, per year, 2 per cent, is added in each first half-year
and another 2 per cent, in each second half-year. It follows, as
before, that 100 amounts at the end of successive half-years to

sums given by the sequence
2 3
100(1-02), 100(1-02) , 100(1-02) ,
... .
Hence, at the end of x years, the amount is y where
These results can be generalised at once. If &a is invested at

compound interest at lOOr per cent, per year compounded yearly,
then the amount after x years is y where
x.
y=a(l+r)
If the interest is added n times a year, then
In this result, which includes the previous one as a particular case,

it is to be understood that x is a discontinuous variable, taking values
which are multiples of - .

Discontinuity is an essential feature of
IV
this compound interest problem.

The dependence of y on the parameters indicating the interest rate
and the frequency of compounding interest is to be noticed. The
amount y, after any period, is clearly larger the higher is the interest
rate. Further, the amount is larger when interest is compounded
twice a year than it is on yearly compounding. For,
i.e. a(l+r) 2x >a(l+r) x .
In general, the more frequently is interest added, the larger is the

amount of a given sum at the end of any period.
The amount of an investment increases over time in what is called
"
a geometric progression ", each amount being a fixed multiple of
the previous year's amount. Analytically, we can express this
growth at an ever increasing rate by the exponential function
T
yab nx ,
where b 1+-
IV
is a constant greater than unity. The
growth is shown by the heights of

successive points on a certain
exponential graph, the points being spread out at equal distances

along the horizontal time-axis. For example, the heights of points
on the graph of y 2 (shown in Fig. 62) at abscissa aj = l, 2, 3, ...
= x
represent the amounts of 1 after successive years when interest is
added yearly at 100 per cent. It is clear that semi-logarithmic graph-
ing is appropriate to this problem. The compound interest growth
curve is then a straight line with gradient log 6= log (l + -) . The
\ til
percentage rate of growth is a constant fixed by r and n.

A
simple problem way in which logarithms are to
illustrates the
be used in dealing with compound interest growth :
ANational Savings Certificate (1935 issue) costs 15s. and realises 20s.
at the end of 10 years. Assuming that interest is added four times a year,
it is required to find the rate of interest
represented by this growth. If
the interest rate is lOOr per cent., then
is the amount (in shillings) after x years. But y = 20 when x = 10.

20 = 15(l+Jr) 40
'
So, ,
i.e. 40 log(l + Jr) =log 20 -log 15-0-12494,

i.e.
log(l+Jr) =0-0031235,
i.e.
l+Jr = 1-00722.
=
So lOOr 2-89 approximately and the rate of interest per year is 2-9 per
cent, correct to one decimal place.
The growth of an investment when interest is added at definite

intervals is a function of a discontinuous variable. It remains to
consider what meaning can be attached to a notion of growth at
continuous compound interest. Our problem now is to examine the
result of letting n, the number indicating the frequency of compound-
ing interest, take larger and larger integral values.

/ l\ n
To start with a simple case, 11+-) is the amount of 1 at the
\ nj
end of a year when interestis compounded at 100 per cent, per year
n times in the year. The values of this expression for certain values
of n are
the values being given correct to three decimal places. It is clear

n
that
/
-l\ tends to a definite limit, in the
( 1 -f )
neighbourhood of 2-718,
as n tends to infinity. It can be shown, by rigid methods, that this
is correct * and the limiting value is denoted by the letter e. So,
/ i\n
DEFINITION: e = Lim (in 1 .
n->oo \ nl
show that e, which must be a pure number, is very
It is possible to
number TT and cannot be expressed in fractional
similar to the familiar
form or as a terminating decimal. The value of e can be found,
however, correct to any given number of decimal places by giving
/ n
l\
na sufficiently large value in ( 1 +- 1 .
So, to five decimal places,
\ Tft/
6 = 2-71828.
Our definition of e is such that the amount of l at the end of one
year, when the interest at 100 per cent, isadded more and more
frequently, approaches the value e. This is only one of the
many uses of the number e which of very great importance in is
mathematical analysis.
Returning to the general case of compound interest growth,
r \nx (/ r\ r
}
rx
(f l\ mV*
= a<[l+-) =aUl + -} ->ae rx as TI->QO
( 1+-) }
V ,
nl [\ n/ j [\ ml }
I L\ m
since 1 1 -f- >e as m, and so as n=rm, tends to infinity. Also,
\ m/ )
since the discontinuous variable x is a multiple of -, it tends to

n
become less and n increases. Hence, as n tends
less discontinuous as
to infinity added more and more frequently, the
and interest is
rx
compound interest formula tends to assume the form y~ae and
the variable x tends to become continuous. We have now derived
a concept of continuous compound interest as the result of a limiting
process in which interest is compounded more and more often. Our
result is :
The amount of a after x years when interest is compounded con-
tinuously at the nominal rate of lOOr per cent, per year is given by
rse
y=ae .
* See Hardy, Pure Mathematics (3rd Ed., 1921), p. 137.

This compound interest formula, dependent on a continuous variable
x, is an abstract one. But, though interest can never be compounded
continuously in actual practice, the formula can be taken as a con-
venient and approximate representation of the actual state of affairs
when interest is compounded frequently.
9.7 Present values and capital values.

A sum of a due x years hence and the rate of interest and the
is
frequency of its compounding are known. Then we can determine

the sum to invest now so as to produce the given sum of fa at the
end of x years. This sum is called the present value, or the discounted
value, of a available x years hence.
If interest is compounded once yearly at lOOr per cent., then y
is the present value of a available x years hence provided that
So y
Similarly, if interest is added n times a year at lOOr per cent., we have

a
m^*m
* I ~\nx
'
^ n)
Finally, if interest is added continuously at lOOr per cent., then

TX
ye =a,
y = ae~
rx
i.e. .
The present value y of a given sum at a given date depends on

the parameters r and n indicating the interest rate and the frequency
of compounding. The present value is smaller, the higher is the
interest rate and the more frequently is interest compounded.
The uses of a computation of present value are fairly obvious. If
a dealer has a claim on a in x years' time, he can sell his claim now
to another person, not for the full a, but for the present value of
this sum calculated at the current rate of interest. Further, we can
find the total present value of a claim on a whole series of sums due
in successive years. The series
fa , !,
a2am ,
...
may represent the values in the current and in

successive years of m
the crops obtainable from a piece of land, of the outputs of a given
machine or plant or of the incomes due to a given individual. We

" "
have, in fact, a given stream
of crop values, of output values or
of incomes over time. The present value of the stream, if interest
is added yearly at lOOr per cent, per year, is given by the sum
This sum can be called the capital value of the land, machine or
income stream in question. It represents the sum which must be
invested now to produce incomes of a %, a 2 ... am in successive , ,
years. It is to be noticed that the capital value of an output or

income stream depends, not only on the items of the stream and on
the number of years that the stream flows, but also on the interest
rate that is taken. One and the same income stream has different
capital values when different interest rates are current. Simple
examples can be given to illustrate this fact.
Shares in a mining company are expected to produce dividends of 40,

32, 24, 16 and 8 in the present and in the four following years, and
to be worth nothing thereafter. If interest is added once yearly at 5 per
cent., the present or capital value of the holding is
the calculation being made with the aid of logarithms. A similar com-
putation shows that the capital value is 116-2 if the interest rate is
only 2i per cent. These are the two sums which can be invested now
to produce, at the respective interest rates, the given income stream over
the five years.
Shares in a new trading company arc expected to produce dividends
of 1, 2, 4 and 6 in the present and in the three following years. In the
fourth year, it is expected that the shares can be sold for 120. The
capital value of this holding, taking interest as added once yearly at 5 per
cent, per year, is given by
6 120 \ .
i + i
=110-4.
+ T-î )=*
1 -05 1 -05 2 1 -05 3 1 -05V
But, if is 2| per cent., the capital value is found to be

the interest rate
121 !. This holding worthisless than the previous one if the rate of
interest is 5 per cent, but worth more if the rate of interest is 2J per cent.
The relative capital valuation of the two holdings is reversed by the
change in the rate of interest.
In conclusion, we can consider a simplified economic problem in
the field of capital and interest. Let x be the income derived by
an entrepreneur this year and y his income next year. The relation
between these two incomes is given by technical considerations as
y=f(x) (see 5.6 above). The entrepreneur wishes to arrange his

resources to give incomes which correspond to the largest possible
present value. If the rate of interest, compounded yearly, is fixed
at lOOr per cent., the present value of x now and y next year is
1+r l+r
an,-
Ihis a maximum
.
f = n andj * - -
f
is if -r -r <0, i.e. if
l +
dx
m=0
l+r
and
dx 2
&
1-fr
<0.
The first condition shows that r equals the expression { -f'(x)} - I

which, as the numerical gradient of the transformation curve
y=f(x) reduced by one, is called the marginal rate of return over

cost.* At the incomes for maximum present value, therefore, this
marginal rate must equal the fixed rate of interest. The second
condition is that /"(fl?)<0, i.e. the transformation curve must be
" "
concave from below, a condition which is satisfied in the normal
case.
9.8 Natural exponential and logarithmic functions.

/ i\n
The introduction of the number e-Lim (1+-) =2-71828...
\ nl n->oo
provides us with the most important cases of the exponential and

logarithmic functions. When the base of the function is taken as e,
" "
it is called a natural exponential or logarithmic function. The
standard forms of the natural functions are y = e x and y=log B x and
the curves representing them are similar to those already incjicated
(in Figs. 62 and 63) for the base 2.
A more general form of the natural exponential function is
y=ae*,
where a and b are constants. But this function can be derived quite
easily from the standard form y=e*. The method of derivation is
* See and pp.
Fisher, The Theory of Interest (1930), pp. 159 et aeq. 514-5.
See also, Examples VI, 31.
best described in a particular case in diagrammatic terms. The

and
curve y=e* is obtained by stretching the curve y = e x in the hori-
x
zontal (Ox) direction so that each point on the former is twice as far
from Oy as the point with
the same height on the lat-
ter. This is illustrated in Fig. fy-*e** /
67. The curve y = 2e* x is
then obtained from the curve
y $ x by stretching in the
vertical (Oy) direction so that
all ordinates are doubled,
as in Fig. 67. Two
shown
stretching processes are thus
needed to transform the
standard curve y ex into
the curve y 2e* x . In gen-
eral, if a and b are positive,
the curve y=aebx can be
derived from the curve y e x in two stages. The standard ex-
ponential curve is first reduced in the Ox direction in the
ratio 6 :
1, i.e. if b> 1 and stretched if b <1.
contracted The curve
so obtained then expanded in the Oy direction in the ratio a 1,
is :
i.e. stretched if a > 1 and contracted if a < 1 As a result, the ordinate

.
/y
of the curve yaebx at abscissa - is a times the ordinate of the curve

x
?/ at abscissa #.* Exactly similar remarks apply to the relation
e
between the natural logarithmic function y a log e bx, where a and b

are constants, and the standard form y=log e x.
All problems of growth at compound interest added continuously
are described by means of a natural exponential function. An
investment of a at lOOr per cent, continuous compound interest
increases over time according to the law yae rx The particular.
* If the constants a and b are allowed to take negative values, a further

modification of the standard exponential curve is needed. The curves
y = ae~ y = ae only in that the abscissa ( - x) gives the same
bx and bx differ
ordinate of one curve as the abscissa x of the other. One curve is the reflection
of the other in Oy. One curve falls and the other rises from left to right.
- ae bx is the reflection of the curve y ae bx in Ox.
Similarly, the curve y
Negative values of a and 6, therefore, require the reflection of the standard
exponential curve in one or other of the axes.

shape of the curve representing this growth depends, as we have
seen, on the values of the constants a and r. The larger the
interest rate, the steeper becomes the curve and the faster becomes
the growth of the investment. The larger the original sum in-
vested, the higher becomes the curve and the larger the amount of
the investment at any time.
A somewhat different use of the curve y=ae rx can be noticed.
Any point (x, y) in the positive quadrant of the plane represents a

situation in which y is available x years hence. If this point lies
on the curve y~ae rx then ,
= 50 the present or discounted

value of this sum is a if the
60--
interest is reckoned continu-
ously at lOOr per cent. The
curve thus connects all points
40--
representing situations with
the same present or dis-
20-^ counted value at a given

rate of interest and, for this
reason, it can be described
8 12 16 as a discount curve.
FIG. 68. A system of discount
curves yae rx is derived by
fixing the value of r and taking a as a parameter. Certain curves of
the system are shown in Fig. 68, in the case where the interest rate
is 1 per cent, (r = 0- 1 )
. One curve of the system is obtained from any
other by stretching or contracting in the vertical (Oy) direction and
one curve passes through any given point in the positive quadrant
of the plane. The discount curve system provides a simple means of
comparing the present values of different sums available at different
future dates. Suppose that two situations are given (certain sums
available at certain dates) and represented by points in the plane Oxy.
If the points are on the same discount curve, then the present values
are equal and given by the parameter of the discount curve. If one
point is on a higher discount curve than the other, then the present
values are different and the first situation corresponds to the greater
present value. It is necessary, of course, that the rate of interest
should be known and fixed the discount curve system is given only for
;
one rate of discounting. The whole system changes when the rate is
changed, the curves becoming steeper for larger rates of interest.*
A discount curve y=ae rx becomes a
straight line log y= log a + rx
when it is
plotted on a semi-logarithmic graph in which the vertical
scale for y is based on natural logarithms. The line slopes upwards
with a gradient equal to the interest rate r. The discount curve
system of Fig. 68 then reduces to a set of parallel straight lines.
In conclusion, it is easily shown that an exponential or logarithmic
function to a base other than e can be expressed as a natural ex-
ponential or logarithmic function. If b is a positive number, the
definition of a logarithm enables us to write
b = e}oe *b .
The general exponential function y=ab GX then appears

x
y=ae(cloe *W ,
i.e. as a natural exponential function. Further, since
the general logarithmic function y a logb cx can be written
y= log - cr
Ep '
i.e.as a natural logarithmic function.

There is, therefore, no need to consider exponential or logarithmic
functions other than those to the natural base e. In practical work,
it is convenient to take common logarithms and powers of the base
10. But, in theoretical work, it is always found preferable to use
the natural base e. In this case, for example, the derivatives of the
exponential and logarithmic functions appear in their simplest form
(see 10.1 below). Natural exponential and logarithmic functions are
thus assumed in all our theoretical work and, if we write a logarithm
without specification of the base, it is to be understood that the base
ise. The step from the practical logarithms to the base 10 to
theoretical logarithms to the base e is, however, a very simple one :
log,* = log. 10 lo glo z

. =1 ^^ = (0-4343) log1Q x,
lo gio e
the numerical multiplier being given correct to four decimal places.
system of discount curves is obtained when a is fixed and r is

* Another
taken as a parameter. This sytem enables us to determine at what interest
rate y available in x years has a given present value of a.
EXAMPLES IX
Exponential and logarithmic functions
1. Draw a graph of y 10*. Read off the values of x for 7/-0-5, 1-2 and
8-9 and check from tables of logarithms.
2. Use logarithmic tables to construct a graph of y Iog 10 #. Add the

graph of y = x - 2 and so solve Iog 10 # + 2 = x approximately.
Find, with the aid of logarithmic tables, approximate solutions of the
3.
equation 2 3~ = 10 and of the equation 2** = 7.

X x .
4. Illustrate the fact that the logarithm of a non-prime number can be

reduced to logarithms of primes by showing that
Iog 10 1176 = 3 Iog 10 2 + Io glo 3 + 2 Iog 10 7.
Check the result by looking up the logarithms in tables.
5. By expressing the left-hand side as a single logarithm, show that

i log v- logy +$iogv-o.
Express y explicitly in terms of x if it is given that
l + 2x
6. Express the logarithm of as sums and differences of
+ 3x
logarithms of simpler expressions. Hence, find the value of this expression
when # 2-4.
7. Show that (x + N/a?

2 - 1) (x
- *Jx z - 1) = 1 and deduce that
log(x + \^^T) = - log(# - *Jx 2 - 1).
- and both tend to zero as x tends to

8. Indicate why Iog 2 # x* 3"* infinity.
Deduce that
9. Yearly production of bricks, U.K., 1816-1849.
Data from Shannon, Bricks A Trade Index, 1785-1849, Economica, 1934.

Represent the above time-series on two graphs, one showing the number
of bricks on a natural scale and the other on a logarithmic scale. Which
graphical representation is to be preferred?
10. If y is the number of persons with incomes over #, the following table
is obtained for super-tax payers, U.K., 1911-2 :
Plot the data on a logarithmic graph and show that the plotted points lie
close to a downward sloping line. Deduce that Pareto's Income Law is
m
approximately satisfied (y a/x ) and estimate the values of the constants
a and m. (See Bowley, Elements of Statistics, 4th Ed. 1920, p. 347.)
11. Plot the price and consumption data of Examples I, 30 as a scatter

diagram on logarithmic scales for both variables.
12. Number of Insured Workers in Staffordshire, at July, 1935, and percentage

Unemployed, at May, 1936.
Plot a scatter diagram on logarithmic scales. Is there any evidence of a

correlation between unemployment and the size of the area ?
13. Given \og e 10 2-3026, use tables of common logarithms to plot a graph
of y - loggir. Put y 1 and read off the value of e.
y = fe and y 2e~l x
x zx
Show how these curves
14. Plot graphs of y e~ , .
can be obtained from the curve y = e x .
15. If y = %(e -e~ x ), show that e zx - 2yex - 1 - 0. Solve this quadratic

x
equation in e show that x = log (y + *Jy 2 + 1). Why must the other root of
x to
the quadratic be neglected ? In the same way, show that y = % (e x + e~ x and )
x = log (y 4- */y
2 - 1) are inverse functions.
a _ Q~~ + I/
16. Show that y and x - are inverse functions.
Jlog-1
-~^
e + ez^ y
17. Given Iog 10 e = 0-4343, use tables of common logarithms to evaluate
y = xe~ approximately for x 0, |, 1, f 2, 3, 4 and 6. Hence plot a graph of
x
,
the curve y xe~ x for positive values of x. (This is one of Pearson's system
of curves for fitting to statistical frequency distributions. See Eldorton,
Frequency curves and correlation.)
18. Using the method of the previous example, plot a graph of the curve
y = e~\& for positive and negative values of x. (This is the normal curve of
error of statistical theory.)
19. Indicate why x e~ x and x n e~l x must tend to zero as x tends to infinity
n *
for any finite value of n. From the graph of y = xe~ x , illustrate that xe~ x -*Q
as #->oo .
Compound interest problems

20. What is the amount of 100 after 6 years and what amount must be
invested to realise 100 after 10 years when interest at the rate of 3 per cent.
per year is added (a) yearly, (b) twice yearly, (c) continuously?
21. In how years will an investment double itself when interest at
many
2J per cent, per year is added (a) yearly, (6) twice yearly, (c) continuously?
What are the periods when the interest rate is 6 per cent. ?
22. A National Savings Certificate costs 15s. and realises 20s. after 10 years.
Find the rate of interest involved when it is added (a) yearly, (6) twice a year,
(c) eight times a year, (d) continuously. Show that the nominal rate is
smaller, the more frequently is interest added. Take Iog 10 e 0-4342945 and
use Chambers' seven-figure logarithmic tables.
23. The certificate of the previous example produces 17s. 3d. after 5 years.
What rate of interest, added yearly, does this represent ? Is it less than the
rate over the complete 10 years?
24. In a previous issue (1933), a certificate cost 16s. and realised 20s. after
8 years, 21s. 4d. after 10 years and 23s. after 12 years. Find the interest rate,
added yearly, for eacli of these periods. Is it true, as claimed, that the present
issue bears the same rate of interest as the 1933 issue over the 10 years period?
26. Interest at lOOr per cent, compounded yearly is equivalent to interest

at 100s per cent, compounded n times a year, a given sum producing the same
/ s\ n
amount after any period at the t D rates. Show that / (l+ ) -1.
by an amount equal approximately when

2
Deduce that s is less than r to fs
and n = 4.
s is small
26. If lOOr per cent, compounded yearly and lOOp per cent, compounded
continuously are equivalent interest rates, show that p = log e (l + r). Plot a
graph of this relation to show that p is always smaller than r but approxi-
mately equal to r when r is small.
27. A sinking fund is formed by investing x at the end of each year for
m years. Show that the final amount of the fund is
when interest is added yearly at lOOr per cent, per year. By writing the sum
of this geometric progression (17.1 below), show that the fund will amount
finally to a if ar
Show that approximately 79-5 must be set aside each year if the sinking
fund (interest at 5 per cent.) is to replace a machine costing 1000 after

10 years. (See Fowler, Depreciation of Capital, 1934, p. 131.)
28. If the rate of interest (added yearly) will be 100?*! per cent., 100r 2 per
cent., ... 100r
TO per cent, during
m
the next years, show that a will amount
to a ( 1 + r x ) ( 1 + r 2 )
.
( 1 +
. . rm ) at the end of the period. What is the present
value of 6 duo in m years ?
The interest rates in successive years from now will be 4 per cent., 3 per
cent., 2 per cent., 2J per cent., 3J per cent., .... Find the amount of 100
after 5 years and the present value of 100 due in 4 years.
29. Find the present value of 100 due 10 years hence when interest at
2 per cent, per year is compounded (a) yearly, (b) continuously.
30. A
mine-owner derives an income of 2000 this year and his income
falls in each following year until no income results. Find the present
by 200
value of the income stream when interest is added yearly at (a) 4 per cent,
per year, (6) 5 per cent, per year.
- bo taken as the
31. Why can present value of an income stream of a
a year for ever? (Interest at lOOr per cent, compounded yearly.)
32. A plantation brings its owner nothing this year or next year. In
fir
the two following years the incomes are 300 and 400 and the income is
thereafter 500 a year for ever. What is the discounted value at the beginning
of the fifth year of the constant income stream when interest is added yearly
at (a) 4 per cent, per year, (b) 5 per cent, per year? Find the present value
of the whole income stream now at the same interest rates. Compare those
values with those of Example 30 above. (See Fisher, The Theory of Interest,
1930, pp. 133 et seq.)
33. Interest is added yearly at 3 per cent, per year. What is the present
value of a perpetual income of 100 beginning two yoars from now? This
income can be produced by investing 2000 in a business this year and 1200
next year. What is the present value of the investment? Is it a profitable
investment ? *
34. Draw a graph of certain curves of the discount curve system when the
rate of interest is fixed at 5 per cent. Use the graph to determine whether
220 due in 10 years has a larger or smaller present value than 150 due in
3 years, interest being added continuously at 5 per cent, per year.
35. If r a parameter, what is the form of the curve system y = 100era: orx a
is
semi-logarithmic and on a natural graph ? Draw certain curves of the system

using a logarithmic scale for y. Hence estimate at what continuous rate of
interest 200 due in 8 years has a present value of 100.
M.A.
CHAPTER X
LOGARITHMIC DERIVATION
10.1 Derivatives of exponential and logarithmic functions.
THE introduction of the exponential and logarithmic functions makes
itnecessary to extend the list of standard derivative forms by the
addition of the derivatives of the simplest of these functions, y e x
and y~ log x. The rules for derivation then apply, exactly as
before, more complicated exponential

to give the derivatives of
and logarithmic functions and of combinations involving these
functions.
Standard form derivatives are found from using first principles,
the definition of the derivative and the properties of the function

concerned. But the exponential and logarithmic functions are in-
verse to each other and we have only to find the derivative of one
of them from first principles, the other derivative following from the
inverse function rule. The derivative of the logarithmic function
y log x is found more easily from the definition :
log (x + h) -log x 1.
* ---s
x
where n denotes j-
and tends to infinity as h tends to zero. But
nt
I l\ n
the expression (1 f ~ 1 tends to the limit e as n tends to infinity. So
T .
log
^-v(x + h) -log x 1 T . /_ IV 1. 1
Lim ^-=-Lim .
+-
A^ e
log 1 =-]oge=r-~ ,
h-+o x _>*, \ nl x x
d . 1
i.e. r logx = x-.
-
ax
dx I
The function inverse to y = e x is x= logy with derivative -=-.
dy y
LOGABITHMIC DERIVATION 243
Hence, the inverse function rule gives the derivative

d r
dx
gOJ
d
^ ,
l
-
l
Sy
dy
d
i.e. -~e x = e x .
ax
VUJb
The derivative of the exponential or logarithmic function to a base

a other than e can be deduced at once :
Since log a x = log a e log, x,
we have x = log a x = - log a e.

^ log a e
^ log,
Again a* = (^8e a ) = (
where u x log a a and the function of a function rule is used. So

d
-~a x =a x log
hee a.
dx
The derivative of the logarithm of any function of x or of any
functional power of e can be deduced from the function of a function
rule. If u =f(x) is a single-valued function of x, then
d __
d . du _ 1 du
'
dx du dx u dx
^. d u d du du
~r & ~r- = e -r~
., ,
u
Similarly -7- e >
ax du dx dx
or
d
An important special case of the latter result is
d
e a* = ae ax-
dx
d x=
6
e.g. -j- e~ -e~*.
dx
The list of standard form derivatives, as now constituted, is
d "
j-xnnx* 1
dx
1
,
e* =e *> aX = a * lo Z> a '

dx -d x
d . 1 d . 1,
The following generalisations of the standard forms are also estab-

lished :
d n du
u nu n-l
dx dx
d u u
du d Idu
'
dx dx' dx u dx
where u is any single-valued function of x.
The following examples illustrate the practical method of evalu-
ating derivatives when the functions concerned involve exponential
and logarithmic expressions :
Ex. 2. -e* - e* ~
(a; '
)
2 - 2xe**.
dx dx
~e-#=e-**^-(-x*)=
dx dx^
'
2) = (2x
In general,
LOGARITHMIC DERIVATION 245
Ex. 3. -j- x*e* = 20:6* + x2 e x = x (x + 2) ex .
ax
-j- (x
2
+ 3x - 2) e x = (2x + 3) e x + (x2 + 3x - 2) e x = (z2 + 5z + 1 ) e*.
ctx
In general,
-,- (a^
2
4- bx + c)e* = {ax 2 4- (2a -f-
6) a; -f (6 + c)} ex .
(tx
n A
Ex.4. _
1-2*
In general,
d .
T~ log(ax
.
--
dx 6 ; ,
ax + b
In general,
^ ,
-=- log (aa:2 + ox + c)y
dx 6V
Ex - 6 -
Ex. 7. -
-r- ^ n log x =x n ~1 (l -f n log a;).
EX.8.
Ex. 9. Derivatives of the second and higher orders are to be obtained
by successive applications of the derivation process. Thus
d, I d* I d* 2
In general, the following results can bo established :
dr
__
dx*
dr
10.2 Logarithmic derivation.
Ifyf(x) is any single-valued function of x, then

^
~
dx ydx'
an important result with
practical uses. many The proportional
change in the value of the function as x increases from x to (x 4- h) is
-f(x)
The rate of proportional change in the value of the function at the
point x is thus
~
Lim = Lim
hf(x) f(x) A _>0 h ydx dx
The derivative ,
=- thus serves to measure the rate of
dx ydx
proportional change of the function. In diagrammatic terms, just
as -~ measures the tangent gradient of the curve y=f(x) plotted on

dx . ,-. *
natural scales, so measures the tangent gradient of the curve

dx
when plotted as a semi-logarithmic graph. In fact, whenever we
consider proportional changes in the function y=f(x) as x changes,
we use a semi-logarithmic graph and the appropriate derivative
djlogj/)
dx
The logarithmic derivative provides the simplest method of
finding the derivative of a product or quotient. In general terms,

if y uv, where u and v are two single-valued functions of x, then
log y = log (uv) = log u + log v,

and
1 dy _ 1 du Idv
ydx udx vdx'
T u
y=
\
it - i
we have
Again, ,
log y = log = log u - log v,

^ J
and
j Q - Idu ___
___dy
1
*^_ .
Idv__ _
ydx udx vdx*

These two results are equivalent to the product and quotient rules
for derivation given above and we have now established the rules
with the use only of the sum or difference rule and the notion of
logarithmic derivation.
A completely general form of the product and quotient rules can
_.
If y = -
be given by adopting the same method of logarithmic derivation.
23...
^1^2^ 3 *
single-valued functions of x, then
,
where t^, u2 uz
, ,
... and vl9 v 2 v 3
, ,
... are any given
=
log y log u 4- log u 2 +log u 3 + ... - log vl - log v 2 - log v 3 - ... .
1 dy 1 dui 1 du z 1 du 3 1 dv 1 dv z 1 dv 3
'"
ydx u^dx u z dx u 3 dx v^ dx v 2 dx v 3 dx
Further, it is often more convenient,

in practice, to evaluate a
derivative by writing the logarithm of the function before proceeding
to differentiate. In fact, we prefer to carry out the steps of the
logarithmic derivation process rather than quote the rules for the
derivatives of products or quotients. This remark holds particularly
of the derivation of a function which involves complicated products,
quotients or powers. The following examples illustrate :
Ex. 1. y =
Here logy = log x -log (x + 1) -log(#
, Idy 1 1 1 x*-2
and
ydx x x+ l x+2
dy -x x*-2 x 2 -2
i.e. -j-
dx
Ex.2.
Here log y
= 2 log x + J log (2&
- 1) - \ log (x + 1),
and
, Idy
--
2
--
1
-
2 11
a;
rl
e " X /2aT
^-a? ^ x+1 2o:(z + 1) (2s -
x~ a
y = x e~M
n
Ex. 3. ^, where a and r& are constants.
log y = n log a? 1 (#
Here - ~ 2
= . . .
,
and -_j*
(a; -a) =-(
a; a:
-
i.e.
dx
10.3 A problem of capital and interest.

The exponential and logarithmic functions are important, as we
have seen, in economic problems of capital and interest. A simple
case can be considered here as an example of the type of analysis
involved.* A good is produced at a given moment of time ( = 0) at
a fixed and known The good is not sold at once but is stored
cost.
for a variable length of time. The value of the good increases in a

known way as time goes on, the selling price per unit being f(t)
after t years. The function f(t) is assumed to be known and to
increase with t. As concrete, though necessarily highly simplified,
examples of the kind of problem considered, we can quote the
obvious cases of wine and timber. New wine is bought by a dealer
and laid down for sale at a later time. Or a given piece of land is
put under timber at a given time and the timber is to be cut for sale
* The development given here is based on the work of Wicksell, Lectures

on Political Economy, Vol. I (Ed. Robbins, 1934), pp. 172-84, and of Fisher,
The Theory of Interest (1930), pp. 161-5.
at a later time. The problem

is to determine the optimum time for
selling the wine or cutting the timber.

It is assumed that interest is added continuously at the fixed
market rate of lOOr per cent, over the periods concerned. Then, if
x denotes the present value per unit of the good sold after t years,
rt
x=f(t)e~ (see 9.7 above).
It is assumed that the storage period of t years is chosen to maximise

the net profit as discounted to the base time ( = 0), i.e. to maximise
the value of x since the cost of production is constant. The optimum
storage period is given by
dx ^ d 2x
dT and
^ <0 "
log x = F F = log / (t)

Now (t)
- rt where (t) ,
and
d(ldx\ ld*x_(dai\*_ (> '
dt \xdil x'dt* x*\dt)
The conditions for the optimum storage period are thus
F'(t)=r and *"'(*) <0.
The first condition is that f"(t)=-rlogf(t)=~- is equal to the

(tt
J \tf
fixed rate of interest. At the optimum time of sale, therefore, the

rate of proportional increase in the value of the good over time is
equal to the market rate of interest. We

find that it is the marginal
variation of value that is the important factor. Further,
i.e.the rate of proportional increase of value must be decreasing at

the optimum time of sale.
Fig. 69shows the solution of the problem in diagrammatic terms.
The curve represents the function y=f(t) on a semi-logarithmic
diagram, log y being taken against t. The zero point on the
horizontal axis Ot is the time ( = 0) when the initial cost of
production is incurred and that on the vertical axis is taken, for
convenience, as y=a, where a is the fixed cost of production.
The gradient of the tangent to the curve at the point P (at time t)
is given by
dt f(t)
Interest is reckoned continuously at the given rate of lOOr per cent.

per year and y = xe curves, x
rt
represents a system of discount
being a parameter. Here,
x is the discounted value
(at = 0) of ,y available
at time t. The system
appears on the diagram
as a set of parallel straight
c^' ^ -" lines (log y log x + rt)
with gradient r and cut-
ting the vertical axis at
y = x. One line of the set
passes through the point

P and cuts the vertical at
>. Q where
X
FIG. 69. OQ =log x - log a =log - ,
since the origin is at ya. x is now the discounted value of a unit

of the good sold at time and the optimum time of sale (t~tm is
t
)
obtained when x or OQ is maximum, i.e. it is given by the point A

a
where a discount line touches the curve yf(t). At t tm the ,
tangent gradient of the curve equals the gradient of the discount
line, i.e.
-' =r. Further, for a genuine maximum, the curve must
/w
be concave from below, i.e.
2
This condition
t
.*ffiQ}<0.
dtf(t)
" "
is satisfied in the normal case shown, the value of f(t) increasing
rapidly at first and then more slowly. The point JB, where the
tangent at A cuts the vertical axis, must lie above if the dis-
counted sale value of the good is to be greater than the cost

incurred.
10.4 The elasticity of a function.
We have seen that -~j measures the rate of proportional

doc
change of the function yf(x), the proportional change in y being

related to the absolute change in x. Suppose, now, that we relate
proportional changes in both variables. If x is changed from x to
(x + H)y the proportional changes in x and y are - and 77 \
respectively. The average proportional change in y per unit propor-

tional change in x is
/(x) h
If the derivative of the function exists, the rate of proportional
change in y for proportional changes in x is

X
-- _ ~~j~f~
f(x)
x
- -
ydx
Denoting the logarithms of x and y by u and v, we can write
v = log ?/, where y / (x) ,

where x eu ,
and
dv dvdydx
= *
du dydxdu
=#=-
Idy
ydx
xdy
ydx
*.
The rate of change of y=f(x), when both changes are expressed in
proportional terms, is thus measured by
d(logx) ydx
This rate is termed the elasticity of the function at the point x and
"tpti jy
can be denoted by
Mix
= -/(#).*
MiX
^ Hence :
DEFINITION The elasticity of the function yf(x) at the point x

:
is the rate of proportional change in y per unit proportional change

in x :
Ex d (log x) y dx
adopted here is that suggested by Champernowne, A

* The notation
Mathematical Note on Substitution, Economic Journal, 1935. An alter-
native notation is E x (y) E x {f(x)}. No established notation for elasticities
is in current use.
A
logarithmic diagram is the appropriate representation of the
function y=f(x) when proportional changes in both x and y are
considered. The gradient of the tangent to the curve shown on the
diagram is then -A~^; *
Hence, the elasticity of a function at
various points is given by the tangent gradient of the corresponding

curve when logarithmic scales are taken for both variables the :
elasticity can be read off the logarithmic graph of the function just
as the derivative is read off the natural graph.
The important property of the elasticity of a function is that it is
a number which is independent of the units in which the variables
are measured. This is clear since the elasticity is defined in terms of
proportional changes which are necessarily independent of units.

More formally, if units are changed, so that the new measures of x
and y are x' =Xx and y' py respectively, then
x' dy'
^Xxd (//#) _ Xx JJL dy __ x dy
y' dx' p,yd(Xx) pyXdx ydx*
i.e. the elasticity of the function is unaltered.
10.5 The evaluation of elasticities.

fi QJ
Since the elasticity of a function -~ with

is simply the derivative
dx
x
a multiplicative factor -, which makes it independent of units, we
y
can obtain elasticities from the corresponding derivatives. It is
interesting, however, to translate the derivative rules into elasticity

form and to examine the nature of the elasticities of simple functions.
If u and v are single-valued functions of x, then
Eu Ev
Ex uv
E(uv) _Eu Ev
Ex "Ex Ex'
Jf [ 1
\vJ Eu Ev
Further, if y is a function of u, where u is a function of x, then
Ey
'
As an example of the method of proving these results, we have

x d(uv)
E(uv) ___ __L =
x I du
__ % _
dv\ xdv Eu __
xdu ___
_ Ev __
^ '
>y
x
_ ____ =
I
[_
1 1 I .
Ex uv dx uv \ dx dxj udx vdx Ex Ex

The other results are established in a similar
way.
It is to be noticed that the elasticity rules are simplest in the
cases of product and quotient and more complicated for sum and
difference. This completely reverses the position obtained in the
case of the derivative rules. But it is no more than we expect since
an elasticity is a derivative in which the variables are expressed as
logarithms and since logarithms are designed to deal conveniently
with products and quotients. Further, since the elasticity of a
constant is zero,
E(u+a)__
~ u Eu E(au)_Eu
an
Ex ~u + aEx Ex ~lx'
i.e. it is the multiplicative constant that disappears in an elasticity.
As examples of the elasticities of simple functions, we have
-
:
E .
, . ax -E- . .
, E .
ax = OLX.
=r (ax + 6)' .-
, (ax?)' a and (ae )
'
Ex^ ax + b' Ex^ Ex^
Hence it is the power function, and not the linear function, which
has a constant elasticity. If a is a positive constant, then the
function with the same positive elasticity a at all points is y ax",
and the function with the same negative elasticity (
- a) at all points
is yx?a. In particular, the function yax has elasticity 1 at all
and the function xy=a has - at all In
points elasticity ( 1) points.
diagrammatic terms, the function yax represented by an
11
is
upward sloping straight line with gradient a on logarithmic scales

and the gradient measures the constant elasticity of the function.
In the same way, the function yxa =a is shown as a downward
sloping linewith numerical gradient a on logarithmic scales. The
rectangular hyperbola xy=a is a particular case, shown on a log-
arithmic diagram by a line sloping downward with gradient and
elasticity numerically equal to unity.
The elasticity of a function varies, in general, with the value of x
taken. Certain points may be found where the numerical value of
the elasticity is unity and these points are of particular interest.
E
If {/(#)} 1 at a point x, then a proportional increase in x from
iLx
this point gives rise to an equal proportional increase in f(x). On
E
the other hand, if vr {/(&)}> 1 at the point x, then a proportional
litX
increase in x results in a greater proportional increase in/(#). Con-.

pi
versely, if -=-{f(x)}<l at the point x, the proportional increase in

E -
f(x) is less than that in x. Similar remarks apply when -^-{/(x)} 1
JjjX
at a point x, but here proportional increases in x correspond to
proportional decreases in f(x).

Further, at a stationary vahie of a function, the derivative, and so
the elasticity, is zero. As particular cases of the product and quotient
rules for elasticities, we have
" "
It follows that a maximum or minimum value of the total ex-
pression (xf(x)} can occur only at a point where the elasticity of

- 1 and that a maximum or minimum value of the " "
f(x) is average
expression J^ -I can occur only where the elasticity of f(x) is 1.

I
x J
Finally, it is clear, from the definition, that the elasticity of a

" "
function f(x) is the ratio of the marginal value of f(x) to the
" "
average value of f(x) at the point in question, i.e. the ratio of
f'(x) to
^ Hence, at a point where the
"
average
"
value is a
x
maximum or minimum and the elasticity is equal to 1, we have also
" " " "
that the average and marginal values of the function are
equal (see 8.5 above). Points where the elasticity of a function is
numerically equal to unity are thus of considerable importance.

This is particularly evident, as we shall show, when we consider the
application of the elasticity concept to the demand and cost problems
of economic theory.
10.6 The demand.

elasticity of
The market demand for a good can be represented, under certain
conditions, by the monotonic decreasing function % = <f>(p). The

elasticity of this function defines the elasticity of demand at any

price. Since itsnegative at all prices, it is found convenient
value is
to make it positive by the introduction of a negative sign. This step

does not affect the elasticity concept in any material way. So :
ni ,. A pdx dilogx)
rj= - - - = --]. -
.
/. 7 7
Elasticity
y of
J demand ' ^
xdp d(logp)
The value of 77,
which is independent of both price and quantity
units, varies from point to point and always measures the rate of
proportional decrease of demand for proportional increases in price

from the price and demand in question.*
When weconsider proportional changes in demand and price, it
isa great advantage to plot the demand curve on a logarithmic
diagram, taking both price and demand on a logarithmic scale. The
tangent to the demand curve is then downward sloping with a
numerical gradient equal to the elasticity of demand. For example,
the demand law x = ap~ a is shown, on a logarithmic diagram, as a
- a). The
straight line with gradient ( elasticity of demand, in this
and equal to a at all points.
case, is constant
If demand
curves are drawn on natural scales, as is usual in
economic works, it becomes more difficult to estimate and compare
the elasticities at various points on the same or on different demand
curves. It is tempting to estimate the elasticity from the gradient
of the demand curve, to say that a demand curve steeply inclined to
the price axis has a large elasticity. This is incorrect. Demand
and - and value cannot be

elasticity is the product of
-
-
( its
J
judged solely from the second of these two factors. For example,
the lineardemand curve has a constant gradient but its elasticity is
not constant. The elasticity, as is easily seen, decreases as the price
decreases and the demand increases.
Two
diagrammatic methods have been devised by Marshall for the
purpose of estimating the elasticity of a demand curve drawn on
* The demand and the with
inverse function is p^(x) elasticity of price
respect to demand is ^-, the reciprocal of 77. Some writers use the term
p ax
flexibility of price to indicate the reciprocal
concept (see Moore, Synthetic
Economics, 1929, p. 38). Notice also that the elasticity of demand and
flexibility of price are sometimes written without the negative sign (see
Schultz, Statistical Laws of Demand and Supply, 1928).
natural scales.* The first method makes use of a simple geometrical
property of the demand curve. With the notation of Fig. 70, tPT
is the tangent at any point P on the demand curve and the
elasticity
of demand at P is
_p I dx\ _MPMT _MT
A well-known geometrical property of parallel lines MT PT
-_. =
gives 5-
PT ON
(since MP is parallel to Op) and -^-^- (since NP is parallel to
Ox). Hence :
_MT _ON _PT

^~0jf~~lv7~~lp~'
We have thus three convenient length ratios as measures of the
demand elasticity at any point on a demand curve.
M T x B x
FIG. 70. FIG. 71.
The second method depends on the fact that any curve with
equation xpa is a demand curve with unit elasticity at all points.
Such a curve can be called a constant outlay curve, since the outlay
(xp) of consumers is constant at all prices. On logarithmic scales,
the curves for various values of a form a set of parallel straight lines
sloping downward with unit gradient. On natural scales, we have
the system of rectangular hyperbolas shown in Fig. 71. Through
each point P of a given demand curve AB, there passes one of the
constant outlay curves. The elasticity of the demand curve AB at
P can be compared with the unit elasticity of this constant outlay
curve by comparison of the gradients of the two curves referred to
*
Marshall, Principles of Economics (8th Ed. 1927), p. 839.
LOGARITHMIC DERIVATION 25?
the price axis. This - in the elas-

is possible since the other factor
ticity expression is the same for both curves at P. If the demand

curve AB is steeper to the price axis than the constant outlay curve
at P, then the demand elasticity is greater than unity at this point on
AB. If AB
is less steep than the constant outlay curve at P, then
the demand elasticity of AB

is less than unity. Finally, if AB
touches the constant outlay curve at P (as shown in Fig. 71), then
AB has unit elasticity of demand at P. Hence, the path of a given
demand curve across the constant outlay curves gives us a good idea
of the variation of demand elasticity along the demand curve.
10.7 Normal conditions of demand.

For any demand law p = $(x), we have Exp~xtfj(x) and y
dR d dp I.
xdp\
.
( P) ~P +
~T~ T~ X X ~T~ P I H T~ \ }
*
dx dx dx \ pdxl
But the elasticity of demand 77
= - - -7- and so
djK
This an important result. From it we deduce

is :
(1) 77
> 1 at a given price and demand, a small decrease in price
If
results in a.more than proportional increase in demand, marginal
revenue -, is positive and total revenue increases as output (

= de-
dx
mand) increases. This is the case of elastic demand.
(2) If rj
1 at a given price and demand, the small decrease in
price and increase in demand are proportionally equal, marginal
revenue is zero and total revenue has a stationary (usually a maxi-
mum) value.
(3) If 77 <1 at a given price and demand, a small decrease in price

is accompanied by a less than proportional increase in demand,
marginal revenue is negative and total revenue decreases as output
increases. This is the case of inelastic demand.
elastic and inelastic demands (see 5.4 above)
Our general idea of
isthus in agreement with the precise measure of demand elasticity.
We have assumed, in the normal case of demand, that price
decreases are associated with demand increases, and conversely, i.e.
that the demand curve falls from left to right. A further property
of demand is now assumed to hold in the normal case. It is taken
that, as the demand increases, the elasticity of demand decreases
continuously from values greater than unity at small demands to
values less than unity at large demands. As demand increases it
becomes continuously more inelastic. It follows that there is one
definite demand x = a where the demand elasticity is equal to unity.
This normal property of the demand curve is illustrated by the
curve AB of Fig. 71. The curve touches a constant outlay curve at
one definite demand, is steeper to the price axis than the constant
outlay curves for all smaller demands and less steep than the constant
outlay curves for all larger demands.
In the normal case of demand, therefore, total revenue R increases
with output at first (77 > 1), reaches a definite maximum value at the
FIG. 72.
output x=a (77

= 1) and
then decreases as output increases further
(rj <1). Average revenue, of course, decreases continuously as output
increases. Now, from the expression for (tx

above,
marginal revenue = average revenue ( 1 )
Since 77
is positive, marginal revenue than average revenue at
is less
all outputs. Further, as output increases, rj decreases and becomes
unity at the output xa. Hence, for outputs less than a, ( 1

j
is
positive and decreasing. So marginal revenue must decrease as

output increases until it becomes zero at the output x = a where 77 = 1 .
For further increases in output, marginal revenue is negative but

need not decrease continuously. The normal forms of the total,
average and marginal revenue curves are indicated in the two
diagrams of Fig. 72. The diagrams must be consistent in the sense
that the ordinates of the average and marginal revenue curves are
respectively equal to the gradient of the vector from O and the
gradient of the tangent at the corresponding point on the total
revenue curve.
The linear demand law, p=a-bx, and the hyperbolic demand
law, p~ X'T -c,

-f-
are both of the normal form here described.
* *
In
the first case, 77

=6 f b\ and, in the second case,
77
1
zJ (l+-)(l-&C-C20.
a \ xl
Each of these expressions decreases continuously as x increases.
The total, average and marginal revenue curves in particular cases
of linear and hyperbolic demands are graphed in Figs. 44 and 45
above. It is clear, from these graphs, that the demand curves are of
normal form.
In conclusion, an inter- p >
esting connection between

the average and marginal
revenue curves can be
established.* If the tan-
gent at any point P on

M
the average revenue curve
cuts the price axis in A A.R.
and if A
is joined to the
point Q on
the marginal
_ M.R.
revenue curve at the same
\
output as P, then the
FIG. 73.
gradient of AP
referred to
the price axis is twice that of AQ. In Fig. 73, PM and QN are per-
pendicular to the price axis. Then (10.6 above) :
_OM_ p .
-..p
'
MA
* See Robinson, The Economics of Imperfect Competition (1933), pp. 29 etseq.
where 77
is the demand elasticity at P. But marginal revenue equals
average revenue times ( 1

J,
i.e.
and
So
The gradient of AP
MP = NO NO
to the price axis is -^-r T ^=-r = 2 -=~%, i.e. twice
the gradient of AQ to the price axis.
This result provides a method of tracing the marginal revenue
curve from a given demand (or average revenue) curve. Select any
point P on the demand curve, draw the tangent to cut the price axis
inA, draw the line AQ with gradient to the price axis half that of
AP and find the point Q on AQ with the same output as at P. Then
Q is a point on the marginal revenue curve and the whole marginal
revenue curve is traced by repeating the process for different points
on the given demand curve.
10.8 Cost elasticity and normal cost conditions.

A less generally recognised, but extremely useful, application of
the elasticity concept is found in the analysis of the cost problem
(5.5 above). This application provides a striking contrast to the
use of demand elasticity. The latter refers to the average quantity
(average revenue) and enables us to determine, by relating its value

to unity, whether the corresponding total quantity (total revenue)
increases or decreases. The position is completely different in the
cost problem. Here the total quantity (total cost) increases for all
outputs and its elasticity is defined and used to deduce properties

of the average quantity (average cost).
If a firm produces an output re at a total cost II=F(x), then
g
J total cost /c=
Elasticity of
U dx ~ d(logx)
,, 1 ,
The value of /c is independent of both cost and output units and

measures the proportional rate of increase of total cost for propor-
tional increases in output from the level in question.* From the

definition K is equal to the ratio of marginal to average cost. Further,
the elasticity of average cost TT=
From
irdx~ndx\x/
these results, it
x
can be expressed
FIx*\
follows that
dx
:
__ I
~
U dx
:
l
_
~~~
(1) If *<! at a given output, we have the case of increasing

returns in which a small increase in output is obtained at a less than
proportional increase in cost, average cost is greater than marginal
cost and average cost decreases as output increases.
(2) If K= 1 at a given output, the case is one of constant returns

in which the small increase in output is proportionally equal to the
increase in cost, average cost equals marginal cost and average cost
has a stationary (usually a minimum) value.
(3) If K> 1 at a given output, we have decreasing returns and the

situation is exactly the reverse of that when 1. K<
been assumed, in the normal case of cost conditions, that
It has
total cost increases continuously from a positive value (representing
overhead costs) as output increases from zero. A further property
of the normal case is now added, the property that cost elasticity
increases continuously from values less than unity at small outputs
to values greater than unity at large outputs. Returns become
increasingly unfavourable as output increases. There is one definite
output x=a, where cost elasticity equals unity and returns cease to
be increasing and become decreasing.!
It follows, in the normal case, that average cost falls with increas-
ing output at first, reaches a minimum value at the output xa and
then increases as output increases further. Marginal cost is less
than average cost for outputs less than a and greater than average
* The
elasticity of total cost is described by Professor Moore (Synthetic
" " and
Economics, p. 77) as the coefficient of relative cost of production its
"
reciprocal as the coefficient of relative efficiency of organisation ". The
reciprocal of K has also been used by W. E. Johnson (The Pure Theory of
Utility Curves, Economic Journal, 1913, p. 508) and by Professor Bowley
(The Mathematical Groundwork of Economics, 1924, p. 32).
f As a slightly less severe normal assumption it can be taken simply that
K = 1 at some definite output x a, K< 1 for outputs x < a and K > 1 for outputs
x>a.
cost for outputs greater than a. At the output x=a, marginal cost
is rising and
equal to the (minimum) average cost. The normal
forms of the total, average and marginal cost curves are shown in
the two diagrams of Fig. 74. The forms of the average and marginal
cost curves can be checked from the form of the total cost curve,
and conversely. If P
is the point on the total cost curve at a
given
output, then average and marginal cost are to be read off as the
gradients of OP and of the tangent at P respectively.
The simplest cost function satisfying the normal cost conditions
is the quadratic form
U a
where a, 6 and c are positive constants. Here
_ x(2ax
The derivative of K can be shown to be positive, i.e. K increases as x

increases. The total, average and marginal cost curves in the case
of a particular quadratic cost function are graphed in Fig. 46 and
they are seen to be of normal form.
It be convenient to take the total cost curve with a point of
may
inflexion and the marginal cost curve with a minimum point at some
output less than that at which average cost is a minimum. This is

shown in Fig. 74. The quadratic cost function does not exhibit this
additional property and the simplest normal cost function to do so
is the cubic form
U=ax3 - bx + cx + d
2
where a, b, c and d are 2

positive constants (6 <3ac). Here
Average cost = ax
2 -
x
Marginal cost 3ax 2 - 2bx +c
and -: -
=6ax-2b.
dx 2
Since -= is 2
positive for all values of x (since 6 <3ac) and since
-7 2 = at the single
^ output
L x~ ---
,
the total cost curve rises but
dx 3a
has a point of inflexion at the output x~- . The average cost curve
,
has a single minimum value at an output greater than x -

(see
<dd
8.4, Ex. 7). The marginal cost curve is a parabola with a positive
minimum
3 c
value - -
3a
-b2
at the output
L x
b
3a
- . The curves are thus
of the normal form shown in Fig. 74.

As a final illustration of the elasticity concept, we can consider a
problem very similar, in some respects, to the cost problem just
discussed. The output of a firm, under given technical conditions,
is determined by the amounts of the various factors of production
employed. It is now assumed that the factors are always employed
in the same fixed proportions. The output x of the good pro-
duced is then uniquely dependent upon the proportion A by
which the factors are increased (A> or decreased (A<1) from a
1)
given basic position. We can, therefore, take x as a function of A

and define
. .
Elasticity of productivity
. .
= -Xdx = rf(loerx)
If >1, we have the case of increasing returns in which a small

increase in all factors employed results in a more than
proportionate
proportional increase in output. In the same way, e = 1 and e I <
correspond to cases of constant and decreasing returns. In the
normal case, decreases continuously as A, and the amount of the
factors employed, increases. It is to be noticed that the reciprocal
of e is analogous to K in that it measures the elasticity of cost with
respect to output, cost being expressed in terms of the amount of the
factors used instead of in money terms. But c is limited, as K is not,
to the case of factors used in fixed proportions.*
EXAMPLES X
Exponential and logarithmic derivatives
~ ex , e 1+a! , x 2 e~ x ,
1. Obtain the derivatives of the functions 2X , e l x , : ,
3. 1 +x 2
xe~**, log(l-a;), log (a 4 l)(# + 2), log\/l-a 2 Iog(l4x 3 ), log-
X
, -,
X log--
L X ~~~
1
and log
- - -
2. By writing eax +b = eb (ex )

a
, verify that eax + b = aenx -
b
.
Find the derivatives of (ax 4- b)e and (ax 4 b)e~ x x.

3.
4. Show that ~
ax
{ l(e
x
+ e~ x )*} = e 2X - e~ 2X .
5. Show that -=- (x logo?) ~1 4- log x, -

( -- = (1
-
logrr) and
ax ax \ x 'J x
d - 1
,
vrc 2 - a a ) - ~r=F^- "- '
dx log
(x 4-
vx 2 ~ a 2
6. Establish the general result
~lo
dx
7. Find the second derivatives of e~ x , e ax+b , log (ax + b),x log x and x 2 log x.
8. Show that x = (x + r)ex and 00

)
=--
(
- r
l) (x
- r)e~ x .
-j-j.(xe
)
-^(xe-
9. By logarithmic derivation, evaluate the derivatives of
and
10. Find the derivatives of e ax9+bx+c and (ax* + bx + c) ex by taking logarithms

before derivation.
11. By logarithmic derivation, show that
n -
~(x
dx
x -
r (x e ax+b )--=(ax + n)xn
)^xx (l^\ogx) and dx l
e ax+b .
12. Find the maximum and minimum values of x 2 ex Show that (log x - x) .
has only one maximum and (x log x) only one minimum value.
13. Show that the curve y = xe~ x has one maximum point and one point of
inflexion. Verify that y and its derivatives tend to vanish as x -> oo . Illus-
trate from the graph of the curve (Examples IX, 17).
* The concept of elasticity of productivity (Ergiebigkeitsgrad) is used by
Prof. Schneider (Theorie der Produktion, 1934). It describes one aspect of the
" "
production function which will be discussed at a later stage.
14. Show that, on the curve y = e~W, there are two points of inflexion
separated by a maximum point. Hence, indicate the shape of the curve and
describe its behaviour as #-> (Examples IX, 18).
<
15. If y ?(e
x
+ e~ x ), show that y and -~^ are always positive while -~
changes sign at x~Q. Deduce that the curve representing this function
roughly resembles a parabola with a minimum at x = 0. How can the curve
be derived from the curves y = ex and y e~ x ?
16. Ify = x ,showthat

=
ax 2/(l-7/)and-2/(l-2/)(l-22/).
Deduce
i +e dx*
that y increases continuously between and 1 as a; increases. Show that the
curve representing the function has a point of inflexion at x = and indicate
its shape. (This is the logistic curve, see 16.2 (3), Ex. 3 below.)
17. The selling value of a good is /() after t years from the time when the
fixed cost of production of a was incurred. If /() represents lOOr per cent.
(reckoned continuously) on the outlay a, find r as a function of t. It is
assumed that the entrepreneur aims at maximising r. Show that the optimum
storage period t and the (maximised) value of r are given by f(t)=ae and
rt
f'(t) = rf(t). Illustrate diagrammatically and contrast this problem with that
of 10.3.
18. The cost of planting a piece of land with timber is 272. The value of
the timber after t years is 100eiV*. Show that the present value of the
timber is greatest (the rate of interest being 5 per cent, compounded continu-
ously) if it is cut after 25 years. Show also that the maximum return (reckoned
continuously) on outlay is 6J per cent, after 16 years. Hence illustrate the
difference between the problem of 10.3 and that of the previous example.
(Take log,, 2-72 = 1.)
19. It is assumed, at given prices of consumers* goods, that the utility of

an individual .consumer is measurable and dependent on his income x. How
is the marginal utility of income then defined ? If it is known that marginal
utility of income decreases towards zero as income increases, show that
u = a log - and u = axa are two possible forms for the utility function.
Elasticities and their applications
20. Show that the inverse function rule is of the same form for elasticities
as for derivatives. Verify the rule by finding the elasticities of e x and log x
separately.
x x and
21. Evaluate the elasticities of xe , xe~ yper*l*+*\
tt x \
22. If the elasticity of /(#), then the elasticities of xf(x)
is and"^ - are
and (e - 1) respectively. Check with/(#) ax a x
(c + 1) .
23. Find the elasticity of demand TJ

when the demand law is x -
20 -
and
p = 3. Plot an accurate graph of the demand curve, draw the tangent at the
point P
where p = 3 and locate the points T and t where it cuts the axes.
Hence verify that vj^MT Nt, where : OMÔN
and are perpen-
: PM PN
diculars to the axes.
24. If p cents per bushel is the price of corn and x bushels the yearly con-
sumption per head of corn in the U.S.A., it is found that
By plotting on a logarithmic diagram, show that the elasticity of demand

for corn can be taken as approximately constant for all prices.
25. The following table gives the average N.Y. price ( p cents per Ib.) and
the yearly consumption (x th. short tons) of sugar, U.S.A., 1904-6 :
Plot on a logarithmic diagram and draw three parallel lines, one through each
plotted point, so that the second line is equidistant from the other two. It is
assumed that the elasticity of demand for sugar is constant and that the
logarithmic demand line has shifted downwards by equal distances over the
three years. Estimate the elasticity of demand for sugar. (See Pigou, The
Statistical Derivation of Demand Curves, Economic Journal, 1930.)
26. If A.R. and M.R. denote the average and marginal revenue at any
output, show that w = *
at this output. Verify for the linear

A.rv. M.rv.
demand law p a-bx.
27. Find the elasticity of demand in terms of x for each of the demand
laws p *Ja -~bx, p (a- bx) 2 and p a- bx 2 Show that 77 decreases as x.
increases and find where rj equals unity in each case.

28. If the demand law is x ae~ bp express demand elasticity and total,
,
average and marginal revenue as functions of x. Show that the demand is

of normal form. At what output is total revenue a maximum ?
29. A monopolist radio manufacturer produces x sets per week at a total

_3jp
cost of 2
(^c + 3# + 100).
w sets per week when the
The demand is x e
price is jp perPlot the marginal cost and the marginal revenue curves
sot.
on the same graph for outputs up to 50 sets per week and deduce that
approximately 25 sets per week should be produced for maximum monopoly
profit. (Take log e 10 - 2-3026.)
30. If the demand law is x-p a e~ b ^ +c \ show that the demand increases as
the price decreases, becoming large as the price approaches the value . Find
j-
the elasticity of demand for any price greater than 7 Is the demand of
normal form?
31. If a firm produces an output # at a total cost of II=ax* + bx, find an
expression for *, the elasticity of total cost. Show that K is always greater
than unity and increases as x increases.
32. Show that, for the total cost function 17= \fax -t- 6, the elasticity of total
cost increases but remains less than unity as x increases.
y b -4-
33. The total cost function of a firm h d. Show that the

is II ax
x+c
average and marginal cost curves are of normal form if rz, 6, c and d are
-
positive and if d< a(c b), the former having a minimum point and the latter
rising continuously.
34. If II = ax z \-d is the total cost function, where a, 6, c and d are

x+ c
positive (&<c), express average and marginal cost as functions of x. Show
that -=( ) increases with x, being negative for small x and positive for large
dx\x/ fijj
x. Show, also, that -7 increases from zero to large values as x increases.
ctx
Deduce that the cost curves are of normal form (see Examples V, 22).
35. (y) of a good Y depends on the amount (x) of a single

The production
variable factor of production JC according to the law y = ax*(b x)* whore a
and b are positive constants. Express average and marginal product as
functions of x and indicate that the forms of the corresponding curves can be
taken as typical of the normal case of production (see Examples V1T, 37).
36. is given by tho law x ^= a*Jp - 6, where p (which is

The supply of a good
greater than the price and a and b are positive constants. Find an
6) is
expression for e, tho elasticity of supply as a function of price. Show that e
decreases as price and supply increase and becomes unity at the price 26.
What property of the supply curve holds at the point where e = 1 ?
CHAPTER XI
FUNCTIONS OF TWO OR MORE VARIABLES
11.1 Functions of two variables.
OUR study of relations between variable numbers has been limited,
so far, to the case of two variables, one of which is taken as a function
of the other. This restriction can now be relaxed by defining
relations involving many variables and by applying such relations
in the interpretation of scientific phenomena. It is evident that all
sciences are concerned with large numbers of inter-related variable
quantities and that only by a process of severe simplification can

functional relations between two variables be applied at all. For
example, the volume of a gas depends on the pressure, on the tem-
perature and on other factors. In economics, when an individual
considers his purchases on a market, the demand for any good
depends, not only on the price of the good, but also on his money
income and on the prices of related goods. Again, to quote an every-
day example, the sum of money extracted by a London taxi-driver
from his passenger depends on the distance travelled, on the time
taken for the journey and on the proportion expected by way
of a tip.
The formal extension of the function concept presents no difficulty.
Suppose, first, that we have three variables, x, y and z, each with its
appropriate range of variation. The variables are related by an
implicit function if the values taken by them are not arbitrary but
connected in some definite way. In general, we write
f(x,y,z)=0.
Further, the variable z is an explicit function of the variables x and y
if the values of z depend in a definite way on those allotted arbitrarily
to x and y. In this case, we write
s=/te y)'
A given implicit function between three variables gives rise to three
FUNCTIONS OF TWO OR MORE VARIABLES 269
distinct explicit functions ;

each variable can be taken as a function
of the other two. For, when values are allotted arbitrarily to two
of the variables, the functional relation determines the values of the
third variable. When one particular explicit function is considered,
say z as a function of x and y, the variable z is called the dependent
and the variables x and y the independent variables.
The terms already used (2.4 above) to distinguish different kinds
of functions extend at once. It is important to distinguish, in
particular, between single-valued functions, to which the notation
z f(x, y) particularly applies, and multi-valued functions. Of single-
valued functions, we can note the class of monotonic functions where
the dependent variable increases (or decreases) steadily as the
independent variables increase. Again, the function z =/(#, y) is said
to be symmetrical in the variables x and y if an interchange of these
variables leaves the function unaltered in form. Finally, the con-
cepts of limits and continuity extend easily to the case of a single-
valued function of two variables. If, for example, the function
z=f(x y) approaches the limiting value A as x approaches the
y
value a at the same time that y approaches the value 6, we write
f(x, j/)->A as x>a and j/-~>6.
The function continuous at the point xa,

is yb
if z takes the
definite value /(a, 6) at the point and if z tends to the same value as
x and y approach a and 6 respectively from either side. All ordinary
functions are continuous except, perhaps, at a few isolated points
where the variables assume infinite values.
The following examples show some common cases of functions of
two variables and their grouping into function types :
Ex. 1. The implicit function 2x + 3y -z - 1 =0 is a particular case of

the general linear functional relation
ax+by+cz+d=Q.
Each of the explicit functions obtained here is single-valued. For
example, from the particular function quoted, we have
z = 2z+3y-l; y=4(l-2#+z) and z = i(l -3y+z).
Ex. 2. The three explicit functions
z=3xy; z=x2 +y2 and x=x2 -2xy-y2

" "
are instances of the explicit and homogeneous quadratic type
z = ax 2 + 2hxy + by 2
.
In the of these instances, the other two explicit functions are also
first
single-valued. In the other two instances, x and y are double-valued

functions. For example, in the third case, we have
z and y= -x\/2x2 -z.
Ex. 3. The explicit function z

= -
x2
-
appears in implicit form as
x2 - xz + yz - z =--
0,
and an instance of the implicit quadratic function type.

this is
Here, y is also a single-valued function of x and z but # is a double-

valued function of y and z. Another example of the same type is
#2 -f 2/
2
+z 2 = 16,
where each variable is a double-valued function of the other two. The
general implicit quadratic function type can be written
ax2 + by2 + cz2 + 2fyz + 2gxz + 2hxy + 2ux + 2vy + 2wz + d = 0,
where all the coefficients are constants.
z
Ex.4. The explicit functions z=ex2 ~~v t z=log(x +y
2 2
)
and z=x2ev
illustrate the fact that exponential and logarithmic expressions can
appear in functions of two variables. In the second case, for example,
the expression (x 2 + y 2 ) takes a definite value when values are allotted to
x and y and the logarithm of this value then defines z.
11.2 Diagrammatic representation of functions of two variables.
If three axes Oxyz are fixed to intersect at right angles at the

origin O and if a definite scale of distance measurement is
in space
selected, then (from 1.9 above) a point P in space can be located

by three co-ordinates (x, y, z). It will be assumed throughout the
following development that Oz is drawn vertically upwards and that

Ox and Oy are drawn, in a horizontal plane, in the W.E. and S.N.
" "
directions respectively. The co-ordinate z thus represents heights
above or below the horizontal plane Oxy while the co-ordinates x
and y represent distances E. and N. of the origin in the plane Oxy.
These conventions are adopted solely for ease in description they ;
are not essential to the argument.

From a given single-valued function z=f(x, y), a table of values
can be constructed by giving x and y arbitrarily selected values and
by finding the corresponding values of z. To each pair of values of
x and y there corresponds a point Q in the plane Oxy (OMx and
ON = y). A perpendicular QP is then erected of length equal to the

value of z from the table (Fig. 75). The point P (x, y, z) is thus
obtained in space with co-ordinates related by the given function.
" "
The whole table provides a finite cluster of such points in space
and, if the variables x and y are continuous, there is no limit to the
number of points that can be obtained. The infinite cluster of points
makes up what is called a surface in space. The surface has the
property that it is cut by lines parallel to Oz in no more than a single

point the height of which above or below Oxy represents the value
of the function z=f(x, y) for the values of x and y concerned. An
easily visualised example of such a surface is provided by an open
hemispherical bowl resting with its rim in a horizontal plane.
FIG. 75.
More = is a given implicit function defining

generally, iff(x, y,
z)
z as a multi-valued function of continuous variables x and y, we can
stillrepresent it by a surface drawn in space. The heights of
various points on the surface above Oxy represent values of z
obtained in the function from given values of x and y but, since the
function is multi-valued, the surface can be cut in more than a single
point by a parallel to Oz. A sphere can be taken as a simple example
of the kind of surface obtained when z is a double-valued function
of x and y.
To a
functional relation between three variables x y and 2, there-
y
fore, there corresponds a surface referred to axes Oxyz in space.

Conversely, if axes and a scale of distance measurement are fixed,
to any surface in space there corresponds a definite functional
relation between three variables, the equation of the surface. The
concept of a surface, it should be noticed, is not limited to the case
" " " "
of a surface without jumps or gaps and with an equation
which is a continuous function of the variables.
We can construct an actual three-dimensional model of any surface
of which the equation is given (e.g. the surface z=x* + y 2 ). Axes
Oxy are drawn in a plane board and a set of points is selected and
marked to correspond to a set of values of x and y and to cover the
plane fairly uniformly. At each point, a vertical pin is erected with
Height equal to the value of z given by the function. As more and
more pins are erected, their heads are seen to describe a definite
surface, the surface representing the function. Such a model is the
three-dimensional equivalent of the graph of a function of one
variable but, on account of the labour involved, its construction is
rarely a practical proposition.

Again, if a function type is defined with the aid of certain para-
meters (e.g. the type z=ax 2 + 2hxy + by z ), tl\gn there corresponds a
whole class of surfaces in space. Each surface of the class corre-
sponds to the function obtained by giving definite values to the

parameters. The systematic study of function types and surface
classes issimply an extension of analytical geometry from two to
three dimensions. For example, it is found that the class of planes
in space is represented by the linear function type and that a
class of surfaces known as paraboloids corresponds to the function
type zax 2
+ by 2 It is not proposed, however, to elaborate this
.
interesting study here.
11.3 Plane sections of a surface.

Without constructing an actual model of the surface in three
dimensions, it is difficult to visualise the form of a surface corre-
sponding to a given function. This difficulty can be overcome, to a
large extent, by a method of great practical and theoretical service,
"
the method of plane sections ". The essentials of the method
are contained in certain expedients familiar in everyday life. An
" "
architect, for example, seldom constructs a working model of a
house he has designed he draws ground and floor plans (horizontal
;
sections) and cross-sections or elevations (vertical sections). Again,

" "
motorists and hikers do not carry relief models of Devonshire to
guide them about that county they refer to Ordnance maps which
;
reduce the country to plane representations by means of a system of

contours or horizontal sections. The methods underlying these and
other practical devices are not difficult to generalise.
A given plane cuts a surface in points lying on a curve called a
plane section of the surface. Of all the possible sections of the surface,
the most convenient are the horizontal sections by planes perpen-
dicular to Oz and the vertical sections by planes perpendicular to
Oxy. Each of these sets of sections can be considered in turn.

A horizontal section of the surface z =f(x, y) is a curve lying in the
surface at a given and uniform height above the Oxy plane. Taking
Fio. 76.
different horizontal sections at various heights and projecting them

on to the Oxy plane (by reducing the heights to zero), we obtain a
set of curves in Oxy which can be called the contours of the surface.
Each contour, to which can be attached the height (or value of z)
which defines it, shows the variation of x and y for the given value
of z.If the variables are continuous, the set of contours consists of
a limitless number of curves all included in the general equation
f(x, y) = constant
by allotting different values to the constant. In Fig. 76 are shown
certain contours of the "paraboloid" surface with equation z= x -fj/ ,
2 2
at heights z = l, 2, 3 and 4. The contours are here concentric

circles with centre at the origin.
The shape of the surface, i.e. the variation of z with x and y, is
shown in a very convenient way by the contour system. As the
s M.A.
independent variables change, a point (x, y) moves in the plane Oxy
and the variation of z (the height of the surface) is seen by tracing
how the point moves in relation to the contours. The value of z
increases, remains fixed or decreases according as the point crosses
from lower to higher contours, remains on one contour or crosses from
higher to lower contours. Further, the values of z attached to the
contours crossed indicate the actual changes in z. If, as is usual,
the contours are for values of z at constant intervals, then z changes
more rapidly and the surface is steeper when the contours are closer
together. Acurve across the contour system (e.g. the line AB of

Fig. 76) implies a definite change in x and y and the contours show
how the corresponding value of z changes. The curve, in fact, is the
projection on to Oxy of an actual path on the surface. The markings
of roads, railways and other features of a country on a contoured
Ordnance map are made on this principle.
Vertical sections of the surface can be defined in a similar way.
Fixing a value of y, we have

z^ a section of the surface by a
plane perpendicular to Oy
showing the variation of z
as x changes only. Such a
section,which is conveniently
projected on to the plane
" "
Oxz is an
y
elevation of
the surface showing the up
and down movement of a
path on the surface in the
W.E. direction. The section
FIQ. 77. can be traced, at least rough-

ly, from the contour map
already obtained. The curve of Fig. 77 is the elevation of the

surface z=x* + y 2
for y 1, i.e. the parabola z l+x 2 . It can be
traced by observing how the line AB (t/
= l) crosses the contours of
Fig. 76. By taking different planes perpendicular to Oy, a whole

set of vertical sections can be obtained in the plane Oxz, all sections
being included in the general equation

z=f(x,b)
for different values of the parameter b. Other systems of vertical
sections can be derived by taking a set of planes perpendicular to

Ox (on each of which x has a constant value) or by taking a set of
vertical planes through Oz and at various angles to Ox (on each of
which x and y vary in proportion).
A system of sections by planes perpendicular to one or other of
the axes shows, therefore, the related variation of two of the variables
of the function for a given value of the third variable. Further, by
arranging the sections of the system in order and relating them, we
can even allow for the variation of the third variable. This idea of
holding one variable constant while observing the variation of the
other two is at least a first step in the examination of the complex
variations of the complete function it lies behind most of the
;
analytical and diagrammatic expedients which are described below.
11.4 Functions of more than two variables.

There is little difficulty in completing the extension of the function
concept and notation to the general case where any number of
variables is taken. An implicit function relating four variables
x, y, z and u is denoted by f(x,y> z,u)~Q and implies that one vari-
able a (single-valued or multi-valued) function of the other three.
is
Further, to denote u as an explicit function of three independent

variables x, y and z, we write, in general, u f(x, y, z). This notation
applies particularly, but not exclusively, to the case where the
function issingle-valued.
More generally, a functional relation between n variables
/(#!, Z 2 Z3
, ,
... X n )=0
- 1) variables. Or,
gives one variable as a function of the other (n
if a variable y is given as an explicit function of n independent
variables x l9 x 2 #3
, ,
... xn then we write
,
y=f(*i, x 2 , #3, ...x n ).
The different variables related, or the different independent variables,

are here conveniently represented by a single letter x with different
suffixes. It is easy to see, in this notation, the number of the variables
concerned. The notation should not be confused, however, with the

practice we have sometimes adopted of adding suffixes to indicate
particular values of one variable.
Pure geometry as an abstract study is just as possible in four or
more dimensions two or three dimensions, and the correspond-
as in
ence between the functions of analysis and the concepts of geometry
persists no matter how many variables are related. But actual
diagrammatic or graphical representations are no longer possible
when more than three variables are involved. We should need to
fix more than three axes intersecting at right angles and, unfor-
tunately, the resources of concrete space are limited to three axes or

dimensions. In the present development, therefore, we consider
first the analysis of three-variable relations with concrete diagram-
matic representations used as illustrations. We can then extend the

results in a purely formal and analytical manner to the general case
where many variables are related. Diagrammatic illustration finds
no place in this extension, nor is it needed. Analysis, as we shall see,
finds it little more difficult to deal with the general case than with
the three-variable case.
It is sometimes convenient, however, to use certain geometrical
terms even in the formal analysis of functions of many variables.
The use of the terms is merely descriptive and made by analogy with
the three-variable case. So, a set of values of n variables
(#1? #2> ^3> %n)
" "
is called a point in n-dimensional space referred to n mutually
perpendicular axes, and any relation between the variables is
" " ' ' " "
described as a hyper-surface ", a locus of a point in n dimen-
" "
sions. In particular, the hyper-surface representing a linear
"
relation between the variables is termed a flat ", a concept which
includes a straight line (two dimensions) and a plane (three dimen-

sions) as particular and concrete cases.
11.5 Non-measurable variables.

It was seen (1.5 above) that many properties of scientific pheno-
mena can be described by the measurable quantities with which our
analysis has been concerned. There remain other properties capable
of order without the additive character necessary for direct measure-
ment and described by ordered but non-measurable magnitudes.
We have yet to investigate whether, and under what conditions,
mathematical analysis can be applied in such cases.
A clue to the solution of our problem is to be found in the use of a
contour map of a function. As x and y vary continuously, a function

z=/(#, y) increases, remains constant or decreases according to the
way in which the variable point (#, y) crosses the contour map of the
function. There is no reference here to actual numerical changes in z ;
only the order of the values of z and of the contours is needed. It is

possible to proceed to trace actual numerical changes in z by examin-
ing how close together the contours are as we cross them. But this
second and quantitative use of the contour map is quite independent
of the first use. It should be possible, therefore, to take a non-
measurable magnitude as dependent on measurable variables and
represent it by a contour map used only in the first of the ways
described. This is the point we must now develop.
Suppose that the level assumed by a non-measurable magnitude
depends uniquely on two measurable quantities x and y. As x and y
vary, therefore, we can observe simply whether the magnitude
increases, decreases or remains at the same level. A set of numbers z
is now associated with the ordered set of levels of the magnitude so
that z and the magnitude increase, decrease or remain constant
together. Then z must be some function of x and y, zf(x, y), a
function which is, however, by no means unique. The magnitude
can be represented just as well by any other variable number or
function which changes always in the same direction as z=f(x, y)
when x and y vary. For example, the variation of the magnitude
2
might be indicated by {f(x, y)} by e/tev) or by log /(a?, y). In
,
general, the dependence of the magnitude on the variables x and y

can be expressed :
=F(z) where =/(*, y),
i.e. =F{f(x,j/)}.
Here, z=f(x, y) denotes any one variable number which indicates the
variation of the magnitude and then denotes any other variable
number, or the whole class of possible variable numbers, indicating
the magnitude's variation. F(z) is any function restricted only by
the fact that F'(z) must be positive, so that and z move always in
the same direction.
In diagrammatic terms, the variable magnitude is shown by the
height above Oxy of any one of a large number of surfaces. The
points on the different surfaces for a given pair of values of x and
y are at all kinds of heights, but they move up and down together as
x and y vary. In particular, if x and y vary so that the height of one
surface is unaltered, then the same is true of all surfaces. The
different surfaces have one feature in common they all have ;
identically the same contour map. The values of attached to the

contours of the map change according to the choice of the function F,
but this affects in no way the form and order of the curves of the
map. Since tt F(z) is the equation of the contour
constant if z is,
map can be written, in a form quite independent of F, as
f(x, y) constant.
For example, the same non-measurable magnitude can be shown by

z = x2 + y2 ,
z = (x 2 + y 2 )
2
,
z = ex*+v* or z=log(x + y )
2 2
and the contour map of each function is the same system of circles
x2 + y2 constant
shown in Fig. 76. The value of z attached to the various circles
changes from one function to another but that is all.

The dependence of a given non-measurable magnitude on two
variable quantities x and y is described by the function F{f(x, y)},
where f(x, y) is any one form of the dependence and where F is any
monotonic increasing function. The dependence can be represented,
without ambiguity, by a system of curves in the plane Oxy defined
by the relation /(#, y) constant. As x and y vary, the path of the
point (x, y) across the curve system tells us whether the magnitude
increases, decreases or remains constant, and the curve system must
be used only in this non-quantitative sense. It follows that mathe-
matical analysis can be applied, subject to the limitations indicated,
in problems concerned with non-measurable magnitudes. The
development given here can be extended, of course, to cases where
the magnitude depends on more than two measurable quantities.
11.6 Systems of equations.

It has been shown above) that two equations provide, in
(2.9
general, a determinate solution for two variables. This result can

now be taken up and generalised. If one relation
/i(*,2/)=0
is given between two variables x and y, there is an indefinitely large
number of pairs of values of x and y possible and the solution of the

"
equation for x and y is said to be indeterminate ". If two relations
A(*,y) = and f2 (x,y)=V
are given, there is, in general, only a finite number of possible pairs
"
of values ofx and y and the solution of the equations is then deter-
minate ". The solution may be unique (as in the case of two linear
equations) or it may be multiple (as in the case of a linear and a
quadratic equation). If we are given three relations
A(s, y)=0, /,(*, y)=0 and /,(*, y) =

then the two equations give a determinate set of solutions for
first
x and y and these do not, in general, also satisfy the third equation.
There is, in fact, no solution consistent with the three equations.
The same result holds, a fortiori, if more than three relations between
the variables are given. In diagrammatic terms, one equation is
shown by a curve in the plane Oxy and the co-ordinates of any one of
the infinitenumber of points on the curve satisfy the equation. Two
equations are represented by a pair of curves which intersect, in
general, in only a finite and determinate number of points. The
co-ordinates of any point of intersection provide a solution of the
equations. Three equations are represented by three curves and,

since there is in general no single point lying on all curves, the
equations have no solution.

Very similar results hold of a system of equations in three variables.
If the number of equations is less than three, an infinite number of
sets of values of the variables is possible and the solution of the
equations is indeterminate. If there are three equations, the number

of sets of values of the variables is finite (unique or multiple) and the
solution of the equations is determinate. If there are more than
three equations, then the solutions of any three of them do not, in
general, satisfy the others and the equations have no consistent
solution. These solutions are supported by considering the three
dimensional surfaces which represent the equations.
A general result is now evident. A system of equations is given
between n variables. Then :
(1) If there are less than n equations, the system is indeterminate

in the sense that an infinite number of sets of values of the variables
is possible.
If there are exactly n equations, the system is determinate in
(2)
the sense that only a finite number of sets of values of the variables
is possible, the solutions of the equations.
(3) If there are more than n equations, there is no one set of values
of the variables consistent with all equations.
One on the generality of these results must be noticed.
restriction
It is system should be mutually
essential that the equations of the
consistent with each other and independent of each other. In fact,
it is not permissible to include an equation in the system either
if it cannot hold if one of the others does or if it automatically holds
if the others do. For example, the equation 2.x + 2y - 4z 5 cannot

be included in the same system as x + y - 2z = 3 they are not;
consistent. - =
Nor can the equation 2x + 2y 4z 6 be included with
x + y - 2z = 3 they are not independent equations and the one tells
;
us no more about the variables than the other. Hence, before the
solution of a system of equations can be examined, we must see
whether the system is mutually consistent and independent. It
often happens, for example, that the number of equations in a given
system one more than the number of variables, so that it would

is
appear that no solution exists. On examination, however, it is found

that one of the equations is derivable from (i.e. not independent of)
the others. When this equation is discarded from the system, the
number of equations equals the number of variables and the system
is determinate.*
The
solution of a system of equations can also be approached from
a different angle. One equation between n variables determines one
variable as a function (not necessarily single-valued) of the other
(n- 1) ;
two equations determine two variables as functions of the
other (n - 2) ;
and so on. In general, if there are ra equations in
n (m<n), any m of the variables can be determined as
variables
functions of the other (n-m). Then, if further equations between
the variables are found, the m
selected variables can be eliminated
by the substitution of the functions already obtained, and the new

forms of the equations involve only (n - m) variables. Thus each
*
This case often arises in the analysis of economic equilibrium. See, for
example, Pareto, Manuel d'fconomie politique (2nd Ed. 1927), pp. 591-3 and
pp. 610-5; Bowley, The Mathematical Groundwork of Economics (1924),
pp. 20-2 and p. 51.
equation of a system can be used to determine and eliminate one of

the variables in terms of the others. As long as the number of
equations is less than the number of variables, there remain some
variables which cannot be determined and eliminated and to which
values can be allotted at The equations are indeterminate.
will.
But when the number of equations and variables is equal, we can

eliminate the variables one by one until a single equation in one
variable is left to complete a determinate solution. For example,
x + y + z+u- 1=0 and 2x-y-2z + u + l
are two equations from which we find
z = ^(2+x-2y) and u = %(l -x- y).
If, now, a third equation is given
we have on and y only

substitution an equation in the variables x :
2
I3x + 2xy + ly - 2x 5y 15 = 0.
- - 2
From the system of three equations, two have been used to eliminate
two variables and the other gives an equation in the remaining
variables.
11.7 Functions of several variables in economic theory.

It is clear that functions of many variables must play an important
part in any* precise and general interpretation of economic pheno-
mena. This fact has been implicitly recognised already when we
found that functions of one variable could only be applied when the
problems were simplified and abstraction made of the inter-relations
of the many variable quantities involved. We
are now in a position
to reconsider the problems in a more general light.
To generalise the problem of demand, we suppose that n consumers'
goods Xl9 X 2i JT 3 ... X n are sold at uniform prices pl9 p 2 p& ... p n
, ,
on a competitive market consisting of a fixed number of consumers

with given tastes and incomes. Then the amount x r of any one good
Xr demanded by the market is uniquely dependent on the prices of
all the goods on the market. We can thus write
Zr = <t>r(Pl,P2>P3> Pn)
as the demand function for the good X r, a function which, for con-
venience, can be assumed continuous in all the variables.
The number of the variables overcrowds our picture of market
demand. It is possible, however, to select a few of the prices accord-
ing to the particular aspect of the problem considered and to assume
that all the other prices are fixed. In particular, we can study the
inter-relations of the demands for two goods X and X 2 by assuming
that the prices of all other goods are fixed. Then
*i
= & (Pi> Pz) and
2 (p ly p 2
x 2 == <f> )
are the demand functions, each dependent on the two variable

prices. Each function can be shown as a demand surface with heights
above the horizontal plane Op^ 2 representing the varying demand
for the good. The vertical sections of such a surface are particularly
interesting. The section of the surface x 1

= l (pl ^2) by any plane <f> ,
perpendicular to Op 2 (on which p 2 has a fixed value) is an ordinary

demand curve showing the variation of xt as p varies. There is one
such demand curve for each fixed price p 2 of the other good and the
whole system of demand curves shows the way in which demand
shifts as the price p 2 is changed. All these demand curves are down-
ward sloping in the normal case. The section of the surface by a
plane perpendicular to Op^ is a curve showing the variation of xl as
p 2 varies for a given value of pv At any point where this section is
upward sloping, a rise in the price of X 2 results in a rise in the demand
"
for Xl and the goods can be called competitive ", at least in a rough
sense. If the section is downward sloping, the converse holds and
" "
the goods can be called complementary at the prices concerned.
It is often convenient to assume, as an approximation for certain
ranges of the prices, that the demand functions are of definite types.
The following are examples of demand functions of relatively simple
type in the case of two related goods :
(1) a?i=o 1 -o 11 ft+a 12 y a ,

X2 =a 2 + a 2l pl -a 22 p 2 .
P2+<*>22
(3) xl = ax A -H pf* 9
x2 = a 2 pf
a^ a ~a
(4) xl p{~ e " p* +ai , x 2 =p 2
In the normal case of demand, the constants an and a 22 as well as ,
a^ and a 2 can be taken as positive. But the constants a12 and a 21 can
,
" "
be of either sign. The goods are competitive if they are both
positive and " complementary " if they are both negative. It can
be noticed that the demand functions (3) and (4) become linear if
the demands and either both or one of the prices are taken on
logarithmic scales.
A slightly different presentation can be given to show the demand
of an individual consumer instead of that of the whole market. If a
consumer has given tastes, his demand for a good X r is a function
of his income /x and of all the prices :
%r = <l>r(p>,Pl,P2,P3, -..Pn)-
As a problem of particular interest, we can trace the effect on the
demand for X r of changes in income and in the price of X r If all .
other prices are taken as fixed, we have a demand surface

r
= ^r(^Pr),
showing the demands for X r for various incomes and prices. The
section of the surface by a plane perpendicular to Op is an ordinary
demand curve for a fixed income level. As the income level is
changed, the section shifts in position. We can thus observe the

shifting of demand consequent upon the variation of individual
income. On the other hand, if the price of the good is fixed, the
variation of demand as income increases is shown by the section of

the surface by a plane perpendicular to the axis Op r .
A generalised cost problem provides a second example of the use

of functions of several variables. If a firm produces different goods
or different qualities of the same kind of good, and if the technique
of production and the supply conditions of the factors of production
are given, then definite amounts of the various goods can be obtained
at a determinate minimal total cost :
n=F(x ,x 2 x 3 ...xn ),
1 , ,
where xl9 x 2 x 3 ... x n are the amounts of the goods Xlt X 2 X 3 ... X n
, , , ,
produced. If only two goods X and Y are produced, the cost

function TI=F(x, y) can be represented by a cost surface with heights
above the horizontal plane Oxy showing the costs of producing
various combinations of outputs. The vertical sections of the
surface represent the cost of producing a variable output of one good
in conjunction with a fixed output of the other good. A particular
form of the joint cost function, such as
n=ax* + 2hxy + by 2 + 2gx + 2fy + c,
can be assumed for convenience in certain cases.
The transformation
functions of 5.6 above can also be generalised
in an obvious way. The technique of production and the pro-
ductive resources of a given firm or country are fixed. Then, if
#i> #2> #3> x n are the amounts of various goods l9 2 X^ ... X X ,

Xn ,
that can be produced in a given year, or the incomes obtainable

in a series of years, there must be some relation between these
variables :
F(xly x 2 # 3
, ,
...# n ) = 0.
For, given the outputs desired of all goods but one, or the incomes
desired in all years but one, then technical considerations determine
the amount of the remaining output or income. When there are
three goods or incomes, we have a transformation surface
F(x,y,z)=0
referred to axes Ox, Oy and Oz along which amounts of the goods or
incomes are measured. In the normal case, we can take the surface
as downward sloping and convex to the origin at all points. A simple
case of the normal type of transformation function is
ax 2 + by z + cz 2 = d,
where a, 6, c and d are positive constants.
11.8 The production function and constant product curves.

An important problem, to which only incidental reference has so
far been made,* concerns the conditions under which factors of
production are combined in the production of a given good by a firm

or industry. With given technical conditions of production, the
amount of a good X
produced depends uniquely on the amounts of
the variable productive factors, A lt A 2 A^, ... n used. If x is the
,
A
amount produced when amounts %, a 2 a 3 ... a n of the factors are
, ,
employed, we can write the production function
It will be assumed, throughout the present development, that the

factors of production are continuously divisibleand that the pro-
ductive process is continuously variable. The production function
is then a continuous function of continuous variables. It is to be
noticed that our presentation of the problem does not exclude the
* See 10.8 above.
possibility of the variable factors being used with certain fixed factors
of production. A definite production function can be assumed in
all cases, no matter how many factors are fixed and how many
variable. Only the appropriate form of the function depends on
considerations relating to the nature, the variability or otherwise,
of the factors.
If there aretwo variable factors A and B, the production function
x=f(a, b) can be represented by a production surface referred to axes
of which Oa and Ob are taken horizontally and Ox vertically. The
method of plane sections is now of great advantage. The contours
of the production surface consist of a system of curves in the plane
Oab which can be termed constant product curves and are defined by
f(a, 6) = constant.
One curve of the system, corresponding to a given value xl of the
constant, includes all points (a, b) representing amounts of the
factors giving a definite product x. The curves form a continuous
arid non-intersectingsystem covering the positive quadrant of the
plane Oab in such a way that one, and only one, curve of the system
passes through each point. As the amount of the factors used are
changed in any way, the corresponding point (a, b) moves in the
plane Oab and its path across the constant prodiict curves determines
the resulting variation of the product obtainable.
The vertical section of the production surface by a plane perpen-
dicular to Ob (on which the value of b is fixed) is a curve showing the
variation of product as various amounts of the factor A are used with
a given amount of the factor B. Thus
*=/(! W= ^L(O), a=/(a, &.)
= #2(0),
are the equations of the vertical sections for fixed amounts b l9 b^ ...
" "
of the factor B. They are diiferent elevations of the production
" "
surface. A
similar set of sections or elevations is obtained by
fixing the value of a and taking planes perpendicular to Oa. A third

" "
set of vertical sections, corresponding to elevations of the surface
of a different kind, can also be used to advantage. The sections are
obtained by taking planes through Ox and at various angles to Oa
and 06, and they show curves giving the variation of product as the
factors are varied in proportion. Thus, if a fixed point on one of the
sections corresponds to amounts a^ and b of the factors, then any
other point on the section is given by a = Aax and b = Xb where A is l9 a
variable proportion. The equation of the section is
a function of the variable A (see 10.8 above). Different sections of

the complete set are obtained by allotting different values to the
constants and b % .
" "
In the normal
case of production, the production surface can
be subjected to certain restrictions apart from that of continuity.
" "
It can be assumed, in the simplest normal case, that product can
only be maintained, when less of the factor B is used, by increasing
the use of the factor A .
Further, as the substitution of the factor
FIG. 78.
A for the factor B

proceeds, increasingly larger additions of A are
needed to compensate (i.e. to maintain product) for a given reduction
of -B.* The constant product curves are then downward sloping and
convex to the origin at all points of the positive quadrant of Oab,
i.e. they form a system of curves similar in form to that of the
" "
normal indifference map described in 5.7 above.
" "
In a more general normal case, the properties assumed above
apply only to certain ranges of employment of the factors while, out-
side these ranges, product is maintained only by increasing the use
* " "
This is the principle of increasing marginal rate of substitution (see
13.7 below).
of both factors. The form of the constant product curve system is
then that shown in Fig. 78. In the area of Oab between the curves
OA and OS, the curves are downward sloping and convex to the
origin ; outside this area, the curves turn back and slope upwards.
Here, when one factor is used in much greater proportion than the
other, an increase in the factor's use requires more of the other factor
even to maintain product.
The normal form of the vertical sections of the production surface
can be traced from that of the constant product curves. If a line
distant 6 X from the axis Oa is drawn in the plane Oab, the way in which
it cuts the constant product curves shows the variation of product
FIG. 79.
on the section perpendicular to Ob corresponding to the fixed amount

&x of the factor B. In the simpler normal case, the line cuts higher
and higher constant product curves from left to right, i.e. the vertical
section rises as a increases. In the more general normal case of
Fig. 78, the product increases as a increases at first, attains a maxi-
mum value when a equals a certain value % (where the line cuts the
curve OA and touches a constant product curve) and then decreases
as a increases further. The vertical sections of the production
surface, for fixed amounts b l9 6 2 6 3 ... of the factor B, are of the
, ,
" "
form shown in Fig. 79. Each section has a product corre-
peak
sponding to a combination of factors shown by a point on the curve
OA of Fig. 78. The greater the fixed amount of B used, the more
of Ado we need before we obtain the peak product and the greater
is the peak product. The curve OA of Fig. 78 is thus of great interest ;
" "
itcorresponds to a ridge line of the production surface and indi-
cates the peak product obtainable from any given use of the factor B.
The vertical sections of the production surface for fixed use of the
factor A are similar in form and lead to the definition of the curve
OB of Fig. 78 as another " ridge line " of the surface.*
A particular production function of the simpler normal type is
where a a positive fraction.! The constant product curve, for the

is
given product x = xl9 is given by
A A>
b
=^>
JL
m = --- are rpositive

*
where A'=[-?} and constants. So
\AJ 1-a
db mA' d 2 b_
and -
and the constant product curves are downward sloping and convex
to the origin at all points. In the particular case where a |
the
curves are rectangular hyperbolas with Oa and Ob as asymptotes.
Further, the vertical section for the fixed amount b of the factor B is
x=A"a; (A'ÂbS-).
Since a a positive fraction, the section rises from left to right at a
is
" "
decreasing rate and never attains a peak value. Notice that
log x = log A + a log a + (1 - a) log 6.
On taking logarithmic scales for all variables, the production surface
becomes a plane and the constant product curves straight lines.
A production function of the more general normal type is
x = 2Hab-Aa*-Bb*,
where A, B and H
are positive constants such that 2
>AB. It can H
be shown that the constant product curves are of the form of Fig. 78
* The normal form of the vertical sections described
is by Knight, Risk,
Uncertainty and Profit (1921), pp. 100-1.
t See, for example, Douglas, The Theory of Wages (1934).
with OA and OB as straight lines (see Examples XIII, 32). The

vertical section for which 6 b has equation
x= -
i.e.
('-
i.e. the section is a parabola with axis vertical. The peak product
Z72 A K ff
6j is obtained when the amount &! of the factor A is used
2
-j
with the fixed amount b of the factor B. It is to be noticed that a

zero (or negative) product is obtained if the amount of the factor A used
is less than -
*
(H - \/H - AB)
2
or greater than ~
A A.
For intermediate amounts of the factor, the product rises from zeio
*>
1500
8 16 24 a 8 16 24
WO men-hours 100 men -hours
FIG. 80.
to a peak and then falls to zero again. Fig. 80 shows certain constant
product curves and vertical sections when the product of wheat in
bushels is
when lOOa men-hours of labour are employed on b acres of land. It

appears, for example, that no wheat is obtainable from 10 acres when
less than 400 men-hours are employed and that the wheat product
rises to a maximum of 640 bushels when 1200 men-hours are em-
ployed.
11.9 The utility function and indifference curves.

An individual consumer has a scale of preferences for two con-
M.A.
Burners' goods X and Y
represented, according to the conditions of
5.7 above, by the indifference map with equation
<f> (#> y)
= constant.
Denote u = <f>(x, y),
a continuous function of two continuous variables. The original
indifference map is then the contour map of the surface representing
this function. The variable u takes a constant value on any one
indifference curve and increases as we move from lower to higher
indifference curves. Hence, as the purchases of the individual
change, the value of u increases, remains constant or decreases
according as the change leaves the individual better off, indifferent
or worse off, i.e. the value of u indicates the level of preference or the
utility of the purchases x and y to the individual consumer.

The concept of preference or utility, however, is an ordered but
non-measurable one and the remarks of 11.6 above are of direct
application. The function u ^>(x, y) is only one way of indicating
the dependence of utility on the purchases. In general, if F(u) is
any monotonic increasing function of u, then
serves asan indicator of utility and can be termed the general index-
function of utility.* It is quite immaterial which utility function,
e.g. u = </>(x,u=log <f>(x y), ... , is used to represent the depend-
y), 9
ence of utility, as a non-measurable magnitude, on the individual's

purchases. The surfaces representing the various functions, referred
to axes of which Ou is vertical, are different but rise and fall
together
and have the same contour map, the indifference map of the indi-
vidual as originally defined.
The problem is generalised by assuming that the individual has
a definite preference scale for the goods lt 2 3 X X X , , ... Xn appearing
in his budget, a scale described by the indifference map
<f>(xl9 x 29 x3 ,
... xn ) = constant.
For a given value of the constant, this relation connects all purchases
which are indifferent to the individual while, for increasing values of
the constant, we move from one set of indifferent purchases to
another set at a higher level of preference. If there are three goods
a translation of the term " fonction-indice " introduced by Pareto,

* This is
Economic math6matiqu& (Encyclopedie des sciences math&natiques), p. 596.
X, Y and Z, the indifference map is represented by a system of

surfaces in Oxyz space which, in their normal form, fit into each other
as a series of shells, all downward sloping and convex to the origin.
For more than three goods, no diagrammatic representation is
possible, but the indifference can be described, by analogy, as

map
a system of hyper-surfaces in many-dimensional space. The indi-
vidual's utility is again a non-measurable concept dependent on the
purchases according to the general index-function of utility
where F (u) is any function such that F r
(u) > 0. The particular utility

function u = <f> (xl9 x 2 # 3, only one of many possible forms.
,
. . . xn ) is
A similar representation applies when we consider, not the pur-

chases of the individual of different goods at one point of time, but
the flow of his purchases or income over time. If xl9 x 2) x 3 ... xn ,
represent incomes in successive years according to the conditions of

5.8 above, we have a definite indifference map
(
Xi9 x zi x & xn) constant,
and we can write a corresponding index-function of utility
where F(u) is any monotonic increasing function.

As an approximate representation of a normal indifference map
for two goods or incomes, we can sometimes take the particular form
where a, 6, a and /? are positive constants. Any monotonic increasing

function of u, for example the function
u' log u = a log (x + a)+fl log (y + 6),
serves equally well as an index of utility. The indifference map,
common to all forms of this particular utility function, is shown in
Fig. 25 in the case where oc=p~l and a

= 2, 6 = 1. It is seen to be
of normal form. A rather different utility function
u =ax + by + c*/xy
also corresponds to a normal case of an indifference map for two
goods or incomes. Any monotonic increasing function of u, such as
serves equally well as an index of utility. The corresponding

indifference map is shown in Fig. 26 in the case where a =6 =1
and = /2.
EXAMPLES XI
Functions of two or more variables
1. *
Express the function z = -- Xt/
- in implicit form and show that each
variable a single-valued function of the other two of the same form in each
is
case. Deduce that the sections of the surface representing the function by
planes perpendicular to an axis are rectangular hyperbolas. In what sense is
this a symmetrical function?
2. Illustrate graphically the form of the contours, and of the other sections,
of the surface z = *Jxy.
3. that the sections of the surface x + y + 2z = 3 by planes perpen-

Show
dicular to parallel straight lines and deduce that the surface is a
an axis are
plane. Locate the points where the plane cuts the axes. Generalise to show
that ax + by -f- cz + d = is always the equation of a plane.
4. Examine the sections of the surface x* + y 2 rz a = a 2 and show that

-J
it is
a sphere of radius a.
5. Show that x 2 4- y* - z 2 = is the equation of a circular cone with its axis

along Oz.
6. Show that the contours of the surface x* -f t/ a + z 2 - 2xz - 2yz = form
a system of and deduce that the surface is a cone.
circles
7. By a geometrical show that ( -Sr-

construction, > , *s tno
^ "o"/
mid -point between two given points (x l9 y lt z x ) and (x 2 , yz , z 2 ) referred to
rectangular axes in space. Show
that the mid-point between any two points
on# + + z+l 2/
Ois also in the surface and deduce that this is the equation
of a plane.
8. If (x l9 y l9 zj and (x 2 y z z 2 ) are two given points referred to rectangular

, ,
axes in space, show that the distance between the points is
y,)
1
+ (*!-*,).
Deduce that the equation of a sphere with centre (a, 6, c) and radius r is
(x
- a) 2 + (y - &)* + (z - c) 2 = r 2 .
9. Show that any point on the surface x +

2
1/
2
+ z2 - 2z = is such that its
distance from is twice its distance from Oxy.
10. By examining the sections of the surface by various planes, show that
z = x* + y* is a surface obtained by revolving the parabola z x* about the axis
Oz in the plane Oxz.
11. Show that a
z=/(fl5 + 2/ ) is a surface obtained by revolving the curve
2
z f(x about the axis Oz in the plane Oxz. Illustrate by considering the
2
)
surfaces z = a? 2 y*, z = e&+& and z = log (a: 2 + y 2 ).

-f-
12. If z = - xt
x-y
, show that z ->
"
oo as x ->a
"
and y ->o, whore a is any constant.
Deduce that the surface represented by the equation has infinities at all
points above the line x-y Q in the plane Oxy. Check from the vertical
sections of the surface.
13. Show that each section of the surface z = (x + y)e-(*+) by planes

perpendicular to Oy has a single maximum point and that the sections are
the same curve translated varying distances parallel to Ox.
Show that,
14. for each of the functions
z = &x z y, z = Vic 2 - y* and z -x -f-

y
,
the value of z changes from z to Az when the values of a; and y are changed in
proportion from x to A# and from t/ to \y. Deduce that the surface repre-
senting any of these functions has a straight lino section by a vertical plane
through and a given point on the surface.
15. For any of the functions of the previous example, show that z = X(/>(r),
/v
aj
where r , and that z = y$(s), where s = .
16. If <f(x) and if*(y) are two functions of single variables, show that the
sections of the surface z ^(x) + ^(y) by planes perpendicular to Ox or Oy are
of the same shape but variable height, and that similar sections of the surface
" " " "
z~(/>(x)*fi(y) consist of a curve stretched or contracted by a variable
amount in the direction Oz.
17. In what sense can it bo said that the function u x 2 + y 2 + z 2 has con-
tours consisting of a set of concentric spheres in Oxyz space ?
18. Show that x + y + z=l, x+3y-z = 3 and 2x~y -\-z-l have a unique
solution in x, y and z. Illustrate by drawing tho planes which represent these
equations.
19. Have the equations x+2y-z=I, 2x-y + z = 3 and x + 7y - 4z = any
solution? Verify that one equation can bo deduced from the other two,
i.e. that the equations are not independent.
20. Eliminate z and u from the equations

2xy + z-u+l-Q, x + y*-2z + u~2 and
2
4x*y* + z*- u*+ 1 =
and obtain an equation in x and y only.
Economic functions and surfaces

21. If the present value of y available x years hence, interest being
z is
reckoned continuously at lOOr per cent., show that z ye~ rx Show that the .
contours of the surface z = ye~~ rx , for a given value of r, form a system of

discount curves. What are the vertical sections?
nr\
ftf)
22. The demand for tea is x, 40 and for coffee # a = 10 thousand Ibs.
Pi P*
per week, where p l and p^ are the respective prices of tea and coffee in pence
per At what relative
Ib. prices of tea and
coffee are the demands equal?
Draw a graph to show the shifts of the demand curve for tea when the price
of coffee increases from 2s. to 2s. 6d. and to 3s. per Ib.
23. Show that x = a t and a?t eP* are two simple examples of a demand
law for a good X^ in competition with a good X 2
and that
and #.= e-ft
PiP* Pi
are corresponding laws when X t
and X 2 are complementary.
24. The demands x 1 and a? 2 of a consumer for two goods at market prices
p l and p 2 are related by the condition *

= , where the a's and
the are constants. If p, = a?1 ^ 1 + x 2 p 2 is tho consumer's income, find the
6's
demands as functions of /x, p^ and p z If the prices are fixed, show that each
.
demand is a linear function of /*.

~
25. It is known that x = Aa a b l a is the product (in bushels) of wheat when
lOOa men-hours of labour are employed on b acres of land. Find the constants
A and a, given that 1500 bushels can be obtained from 100 acres when 10,000
men-hours of labour are employed and that 2120 bushels can be obtained from
the same area when 20,000 men -hours of labour are employed. Draw a graph
of the variation of product as varying amounts of labour are applied to 100
acres. What is the product when 20,000 men-hours of labour are employed
on 150 acres?
26. The production function is x AaP-W, where A, a and ]8 are constants.

If the factors are increased in proportion, show that the product increases in
greater or less proportion according as (a + jS) is greater or less than unity.
How is this property shown on a vortical section of the production surface
through O and a given point on tho surface ? What is the special property of
tho case a 1 - j8?
27. The following data are taken from Douglas, The Theory of Wages :
Manufacturing Industries, U.S.A., 1900-20.

x = Day's index of physical volume of manufactures (1899100),
a = Average number of wage-earners in manufacture (1899 100).
6 Volume of fixed capital in manufacture (1899 100).
Find the value of x given each year by the production function x d*b* and
express as a percentage of the actual value of x. Show that the deviation
never exceeds 10 per cent, and find the average percentage deviation (neglect-
ing signs). Plot a graph on a logarithmic scale of the variation of the actual
and estimated values of x over the whole period.
It is given that x = 40 a+b

bushels of wheat are obtained
when lOOa men-hours of labour are employed on b acres of land. If 10 acres

are cultivated, draw a graph to show the variation of product as a varies.
Show that the greatest product is obtained from this area when approximately
1050 men-hours of labour are employed.
29. Consider vertical sections of the production surface
- Aa* - Bb*
_2Hab
X~
Ca + Db
and show that a proportional increase in the factors results in an equal pro-
portional increase in product.
30. Show that the production surfaces
Bb 2 and x = ^2Hab - Aa z - Bb*
have constant product curves of identical form but that the variation of
product from one curve to another is different. How does product increase
in the two cases as the factors are varied in proportion from any combination?
31. If a and b are positive constants, show that the utility function
gives an indifference map of normal form for purchases x and y in tho ranges
6, the curves being arcs of concentric circles.
32. Examine the utility function u -- .

, where a, 6 and c are positive
c - Vy + b
constants, and show that tho indifference map is a set of parabolic arcs and
of normal form for certain ranges of values of the purchases x and y.
CHAPTER XII
PARTIAL DERIVATIVES AND THEIR APPLICATIONS

12.1 Partial derivatives of functions of two variables.
IN a function of two variables z=f(x, y), the two variables x and y
can be varied in any way quite independently of each other. In
particular, one of the variables can be allotted a fixed value and the
other allowed to vary. The function, in such a case, reduces to a
function of a single variable only. Two functions can be obtained
in this way, z as a function of x only (y fixed) and z as a function
of y only (x fixed). The derivative of each of these functions can be
defined at any point and evaluated according to the familiar tech-
"
nique. The derivatives so obtained are called the partial deriva-
" " "
tives of the function z=f(x, y), the term partial implying that
they are defined only for very special variations of the independent
variables. One partial derivative follows when x is varied and y kept
constant, the other when y is varied and x kept constant.
If the function z=f(x, y) is single-valued, its partial derivative
with respect to x at the point (x, y) is the limiting value of the ratio
*
^ as the arbitrary increment h tends to zero
33 and the
partial derivative is =
written either as ^- r as z x ~fx ( x
ox /(#>
^- y)-
2/)> >
ox
The notations are similar to those for ordinary derivatives. But, to
we have the partial derivative with respect to x (i.e. that
indicate that
" "
the other variable y is regarded as fixed), we use the symbol d
" "
instead of the previous d in the first notation and add a suffix x
in the second notation. An exactly similar definition and notation
can be given for the other partial derivative. Hence :
DEFINITION : The partial derivative ofz=f(x, y) with respect to x

at the point (x, y) is
PARTIAL DERIVATIVES AND THEIR APPLICATIONS 297
and the partial derivative with respect to y at the same point is
9 -, v , ,, v -r . f(x, y -\- k)
- f(x, y)
It is sometimes convenient to use slightly different notations. For

3f Clf 3 3
example, dx and
-- can be used instead of and .
Further, when
dy dx dy
no ambiguity is likely to arise, it is possible to usefx and/v as simple
notations for the two partial derivatives.
The meaning of partial derivatives is most clearly seen when they
are interpreted in diagrammatic terms. In Fig. 81, P is the point on
FIG. 81.
the surface z by the values x and y of the independent

f(x, y) defined
variables. Two vertical sections of the surface can be drawn through
P, one perpendicular to Oy and the other perpendicular to Ox. The

former is the curve on the surface passing through P in the W.E.
direction and showing the variation of z as x varies. On this section,
the value of y retains throughout the constant value allotted initially
at P. The tangent PT X to the section at P has gradient measured
by the derivative of z as a function of x (y constant), i.e. by the
partial derivative evaluated at (x, y). Hence, is to be interpreted

c/x
as the gradient of the W.E. section of the surface at P or, more

shortly, as the W.E. gradient of the surface at P. In the same way,
the value of at (x, y) measures the gradient of PTy ,
the tangent
dy
at P to the vertical section of the surface perpendicular to Ox, i.e. the
S.N. gradient of the surface at P. Hence :
The partial derivatives and measure the gradients of the
surface z=f(x, y) in two perpendicular directions at the point P

defined by (x, y), the directions in the planes perpendicular to Oy
and Ox respectively.
It is to be noticed that we have not yet obtained a measure of the
gradient of the surface in any other direction.

The remarks (6.2 above) on the meaning of an ordinary derivative
apply, with obvious modifications, to partial derivatives. One
additional point, however, must be stressed. The value of ~ at a
uX
point (x, y) depends, not only on the value of x from which we start
in the limiting variation defining the partial derivative, but also on
the value of y which is kept constant during the limiting process.
The partial derivative, as the W.E. gradient of the surface z=f(x, y)
at the point P, varies in value when the point P is moved in any way,
i.e. when either or both of the variables x and y are varied. Similar
remarks apply to the partial derivative .

Hence, the value of each
partial derivative at a point (x, y) is a function of both x and y. To

stress this point, we can refer to the partial derived functions and use
'
the particularly appropriate notations fx (x, y) and /'(#, y).
No essentially new idea is involved in the definition of the partial
derivatives and they can be evaluated exactly as ordinary deriva-
tives. The
variable other than the one directly concerned must, of
course, be treated as a constant in the actual process of derivation.
In particular, the rules for finding the partial derivatives of composite
functions are similar to those for derivatives. The modified form of
the function of a function rule, however, merits separate notice. It
can be expressed :
If z is a single-valued function of u where u is a single-valued

function of x and y, then z is a function of x and y with the partial
derivatives
_
dz dz du
___ _ ji
.,
n fi
dz ~~
dz du
dx du dx dy du dy
In particular, if u is any single-valued function of x and y, we have
^
w^rw 11 - 1 du
i
&
e = eu <^ j
and
9 i
* ^
ox log^--
;
ox ox dx ox uox
and similar forms for the partial derivatives with respect to y.
The practical technique of partial derivation is fully illustrated by
the following examples :
Ex.1. ~(2tf-3y + l)=2; ^.(2a;-3y

+ l)= -3.
Ex.2. ~(x* + 2xy-y*)=2(x + y)' y
j^(x*+2xy-y*)=2(x-y).
Ex. 3. For any fixed values of the coefficients a, b, c, ... ,
^-(ax
+ by + c)=-a\ -=-(
2
r(ax* + 2hxy + by
2a;(a;-y + l) -a; 8 _a;(a:-2y+2)
dy\x-y +
EX.5.
. -
oa; ay
Ex. 6.
Ex. 7. If z=x*-
v #-2/ ,
then
log z
=2 log x +\ log(# + y) - 1 log (x -y).
Idz
-- 2 I 1 11 ~~
2x 2 -xy-2y*
x 2x+y 2x -y x (x2 - y 2 )
dz ~ x(2x*-xy-2y*)
i.e.
-p-
Similarly ~=
oy
Ex. 8. definition of the partial elasticities of a function of two

The
variables z =f(x y) is similar to that of the elasticity of a function of one
y
variable :
Ez__xdz_d(logz) Ez ydz_d(logz)
a
Ex~~zdx~d(logx) Ey~ zdy~ d(logy)'
So, if zx y^ a
)
where a and ft are constants, then
log z = a log x + ft log y

Ez = Ez =
and -=- a and -=.- p.
â: jEJy
12.2 Partial derivatives of the second and higher orders.
The two partial derivatives of a function zf(x, y) are themselves

functions of x j/.
The partial derivation process can thus be
and
repeated and we obtain two partial derivatives from each of the
partial derivatives of z, i.e. four second-order partial derivatives of the
original function z. By an extension of the notation, we can write
the new second-order partial derivatives as
~ _
'
~ '
_ ~~ ~~
dx* dx \dx By dx dy \dx dx~dy dx \dy dy* dy \dy
Alternatively, if //(a;, y) &ndfy '(x, y) denote the (first-order) partial
derivatives, the second-order partial derivatives appear as
or, more shortly, as

fxx> fvx', fxv and fyv .
The order of the suffixes indicates the order in which the partial
9 2z
derivations are carried out. Thus, =/^ (x, y) =/ ya? is the
second-order partial derivative obtained by partial derivation first
with respect to x and then with respect to y.

" "
It is to be noticed that the two cross partial derivatives
and -
^ are quite
^ & and there is no a ^
distinct in meaning priori
dydx dxdy
reason to assume that they have the same value at any point. A
" "
result can be established, however, stating that the two cross
partial derivatives are identical in value, provided that certain con-

ditions relating to the continuity of the function are satisfied.*
This result, known
as Young's Theorem, will not be proved here but
it is below in the cases of two particular functions. It is, in
verified
fact, safe to assume that, for any ordinary continuous function, the
order of partial derivation is immaterial :
dy dx dx dy
The
Ex. T
1.
following examples illustrate
From
d (
^-
-)
dx\x-y + \J
x*
-
\
~-
= x(x-2y
:
+ 2)
^-~
(x-y + 1)
2
we ,
'
. .
derive
32 / a;
2
\_ 1
~
dx 2 \x - y + 1) (x
- y + 1) 4
d* / x* \ 1
~~
- - 4
dy dx \x y + 1/ (x y+ 1)
From ^- ~ =
d /
dy\x-y + lJ (x-y + l)
^2 we
(
- - xz \
)
-
x2
,
'
derive
(x-y + 1)*
srjr -l
dxdy\x-y + l/
-
;
(x-y + l
T
2x(y-l)
(x-y
1
See Courant, Differential and Integral Calculus, Vol. II (1936), pp. 56-7.
Ex.2. Since llo(^ + ^ and
,
wehave
X + ^-2*/
los^
10 " Zx ~ 2y 1
-
~
Ex. 3. The partial derivatives of a function <f>(x, y) are <f> x and <f> y. It
isgiven that the partial derivatives of the ratio of x to v considered as <f> <f> ,
a function of x and y, are respectively negative and positive :
^ 9 (I*
and ^r T5
It is required to express these inequalities in terms of the first and

second-order partial derivatives of <f>(x, y).* We have
xv
The inequalities are, therefore, equivalent to the forms
<t>v<l>xx
-
<f>x<tfr
< and <t>x<l> V v
-
<f> v <f>
Each of the four second-order partial derivatives is a function of

x and y and, by a further partial derivation process, we obtain eight
partial derivatives of the third order. can then proceed to We
partial derivatives of the fourth and higher orders. The notations
we have given clearly extend to partial derivatives of order higher
than the second. Young's Theorem can also be extended to show
that the order of partial derivation is immaterial in a partial deriva-
tive of any order whatever. Partial derivatives of the third and
higher orders are, however, of little practical use and we shall seldom
have need of them in the present development.
* The
problem is taken from Pareto, Manuel d'iconomie politique (2nd Ed.
1927), p. 676, where the mathematical steps given here are omitted. The
function <f>(x t y) is a utility function of an individual for two goods and Y. X
12.3 The signs of partial derivatives.
The definition of partial derivatives shows that -r- and ~ must

dx dy
measure the rates of change of the function z =f(x, y) as x varies
(y constant) and as y varies (x constant) from the point (x, y).
Further, as we have seen, these partial derivatives measure the
gradients of the surface z =f(x, y) in two perpendicular directions.
The meaning of the signs assumed by the partial derivatives is now
clear. Wehave, for the partial derivative with respect to x :
If at the point then the function z=f(x,y)

(1) ^->0 (a, 6),
increases as x increases from the value a (y remaining equal

to the value b) and the surface zf(x, y) rises from W. to E.
at the point where xa and y = b.
(2) If <0 at the point (a, 6), then the function decreases and
ox
the surface falls in the direction named above.
The sign of at the point (a, b) is to be interpreted in a similar way

oy
and refers to the variation of y from the value b (x remaining equal
to the value a) and to the path on the surface through the point
where xaand y = b in the S.N. direction. The signs of the partial
derivatives thus determine, at various points, the increasing or
decreasing nature of the function, the rising or falling nature of the

surface, in two particular and important There remain,
directions.
for later detailed consideration (14.1 below), the cases where one or
both partial derivatives assume zero values.
" "
The values of the two direct second-order partial derivatives
of z=f(x y) have also quite simple interpretations.
9
The partial
d 2z dz
derivative -^-^2
dx 6 of dx as x increases from
measures the rate of change
the point in question (y remaining constant), i.e. it measures the

" "
acceleration of the function z=f(x,y) as x increases and y
d 2z
remains constant. In diagrammatic terms,
^ measures the rate
of change of the W.E. gradient of the surface z=f(x, y) as we move

from the point concerned in the same W.E. direction, i.e. it deter-
mines the curvature of the surface in the W.E. direction. So :
2
d z
(1) If
^->0 at the point (a, 6), then the function z=f(x,y)
uX
changes at an increasing rate as x increases from the value a
(y remaining equal to b) and the surface z=f(x, y) is convex
from below in the W.E. direction at the point (a, 6).
2
3 z
(2) If
ox
TJ
< at the jpoint (a, 6), then the function changes at a
decreasing rate and the surface is concave from below in the

directionnamed above.
dzz
The interpretation of the value and sign of -. is exactly similar.
y
" "
It is more difficult to interpret the value of the cross second-
2
9 z
order *
partial derivative. The value of . at any point measures
dz
dxdy
both the rate of change of as y increases (x remaining constant)
OX
and the rate of change of -- as x increases (y remaining constant).
In diagrammatic terms, the value indicates the way in which the

W.E. gradient of the surface z=f(x, y) changes as we move in the
S.N. direction, and the way in which the S.N. gradient changes as
d 2z
we move in the W.E. direction. So, if is positive at a point P
dxdy
on the surface, the W.E. gradient increases as we move N. from P
and the S.N. gradient of the surface increases as we move E. from P.
It is difficult to visualise this state of affairs. For example, if the
surface rises to the right and falls to the left as we look N. from P,
9 2z
then the positive sign of implies that the surface tends to rise
more rapidly to the right and to fall more steeply to the left as we
actually move N. from P. The S.N. path on the surface through P
moves on an increasingly precipitous incline to right and left. The
9 2z
opposite result holds if is negative at P. These properties of
the surface are quite different from, and independent of, the separate
curvatures of the surface in the W.E. and S.N. directions.
One important point must be stressed at this stage. The various

partial derivatives of a function serve to indicate the rising and
falling nature and the curvature of the surface representing the
" "
function but only in two particular directions, the fundamental
directions in planes perpendicular to Ox and Oy. Nothing is said
about the other directions on the surface, directions in which both x
B 2z B 2z
and y vary. For example, the fact that both ^- and ^ -
are positive
dx 2 ay*
at a point P on the surface zf(x, y) implies that the surface is
convex from below in the W.E. direction and in the S.N. direction
at P. It does not imply that the surface is also convex from below in
any other direction (say the S.W. to N.E. direction) at P. In fact,
" "
it is quite possible that there is a dip in the surface between the
two fundamental directions, so that it appears concave from below
in the S.W. to N.E. direction.
The results obtained above clearly extend to allow for the cases
where any or all of the variables are measured on logarithmic,
instead of on natural, scales. For example,
d(logz)_l dz
dx zdx
measures the rate of proportional change of z=f(x, y) for actual
changes in x (y constant) and is shown by the W.E. gradient of the
surface obtained when z is taken on a logarithmic and x and y on
natural scales. Again, the partial elasticity
Ez_d(logz)_xdz
Ex d (log x) z dx
measures the proportional change in z for proportional changes in x
(y constant) and shown by the gradient in the W.E. direction of
is
the surface obtained when all variables are plotted on logarithmic

scales.
12.4 The tangent plane to a surface.

The simplest two-dimensional curve is a straight line, represented
by a linear equation in two variables. The simplest surface in three
dimensions is the plane and we can show that it is represented, as
we expect, by a linear equation in three variables. Geometrical
considerations (similar to those of 3.1 above) establish the simple
u M.A.
result that, if two points in space have co-ordinates (xl9 yly zj and
(#2> #2> ^2) referred to rectangular axes Oxyz, then the mid-point
1 2
between them has co-ordinates ( > "Hi~" Suppose,
\ 2 2 2 /)
now, that (a^, y^ ZJ and (# 2 2/ 2 ^2) are any two points on the surface
> >
(whatever its form may be) corresponding to the linear equation
where a, 6, c and d are constants. Then

axl + byl + cz1 =d and a# 2
Adding, and dividing through by 2, we find
i.e. the mid-point of the two selected points also lies on the surface.
This is true of any selected pair of points on the surface and it follows
FIG. 82.
that the latter must be a plane. The general linear equation in three
variables thus represents a plane. The actual location in space of
a plane whose equation is given is easily determined by finding the
points A, B and C where the plane cuts the axes (see Fig. 82). On
the axis Ox, we have y=z = Q. Substituting in the equation of the
plane, ax=d. So
a
Similarly 0JB=?b and 06' = -.

c
The cases where one (or more) of the constants a, b and c has a zero
value clearly correspond to planes parallel to one of the axes.
A plane is given as passing through the fixed point (xl9 y^ zj.
If the equation of the plane is
then axi + byl -t- czl d.
It follows that d can be eliminated so that the equation of the plane

appears
or
-i=-7(-i)-r(y-i)
G Lf
(
dz a dz
TT
Hence = and
T
= --b
dsc c oy c
are the gradients of the plane (referred to the horizontal plane Oxy)
in the two fundamental directions perpendicular to Oy and to Ox
respectively. The following result is thus obtained :
The equation of the plane passing through the point (xl9 yl9 zj
with gradients a and j8 in the fundamental directions is
In general; a plane through a given point on the surface z =/(#, y) P

cuts the surface in a curved section on which the point lies. If it P
is possible to vary the position of the plane so that the part of the
section including P
encloses a smaller and smaller area and finally
tends to close down on P
itself, then the limiting position of the plane
can be described as the tangent plane to the surface at P.* Suppose
that the tangent plane exists at P and that a section of the surface
and its tangent plane is taken by a plane through P perpendicular
*
The possibilities are rather more complicated than this brief statement
indicates. It may happen, of course, that no such limiting plane exists at all
and the surface may have no tangent plane at a point where there is a " sharp
" " "
point or an edge of the surface. Further, it may happen that the section
tends, not to a point, but to a straight line through P. The limiting plane is
then a tangent plane touching the surface at all points on the straight lino
through P. In any case, it should be noticed that the section of the surface
by the tangent plane can consist of a curve as well as the isolated point P.
Only one part of the section need close down on the point P.
to Oy. We
obtain, from the surface, the curved path in the W.E.
direction and, from the tangent plane, the tangent PT
X to the path
at P (see Fig. 81). It follows that the tangent plane, containing the
line PT X has gradient in the direction of the plane perpendicular to
,
dz
Oy given by the value of ox at P. In the same way, the gradient of
the tangent plane in the other fundamental direction is given by the
value of -- at P. If the point P has co-ordinates (x lt yl9 %), the

dy
fundamental gradients of the tangent plane at P can be denoted :
So:
The equation of the tangent plane at the point (x1; ylt z,) on the
surface z f(x, y) is
This is an obvious extension of the form given for the tangent line
to a two-dimensional curve (6.5 above).
As an example, the paraboloid surface z=x 2 + y* gives
dz , dz
-~-2x and = 2y.
ox dy
The tangent plane at (xlt yl9 zj has equation
.e. z = 2xxt + 2yVl + zl -

Since the point must lie on the surface, we have zl x^ + y^ and
is the equation of the tangent plane.
An interesting result can now be deduced. If and ~ are both

^-
positive at a point P
on the surface zf(x y), then the tangent 9
plane at P
is positively inclined to the plane Oxy in the two funda-
mental directions, and so in any direction in which x and y increase
and ~ are both

dz dz
together. Hence, in the particular case in which
~
positive, the function z=f(x, y) increases whenever x and y both

increase. The converse result holds in the particular case when both
partial derivatives are negative.
12.5 Partial derivatives of functions of more than two variables.

The definition of partial derivatives extends readily to cases of
functions of more than two variables. If u=f(x, y, z) is a single-
valued function of three variables, then a function of one variable
isobtained whenever two of the variables are given fixed values.
This can be done in three different ways and we have u as a function
of x only (y and z fixed), of y only (z and x fixed) and of z only
--
(x and y fixed). Hence, we can define three partial derivatives,
du du du
r , TT- an d -jr j
of the junction at any point (x, y, z), e.g.
(/x oy oz
du .
f(x + h, y, z)-f(x, y, z)
=
OX h ^ fl
Then three second-order partial derivatives are obtained from each of

the first-order partial derivatives, making a group of nine in all. For
all ordinary continuous functions, however, Young's Theorem asserts
that the order of partial derivation is immaterial and only three of

" "
the six cross second-order partial derivatives are distinct. So we
have only three "direct" partial derivatives (~-, ~-^ and 1
Û
'
A ^ ( d U S U
* *
and three cross 4-- 1 J *'

partial derivatives I
\dxdy
.
-^
dydz
A
and ^-1.
dzdx/
\
The alternative notations for partial derivatives also extend, in an

obvious way, to this three -variable case. Finally, if necessary, we
can obtain partial derivatives of higher order than the second by
further derivation processes.
In the general case of a single-valued function of n variables,
yf(xl9 x 2 ... xn ), there are n first-order partial derivatives
,
dy
dy_
y '
'" jty '
dx1 dx 2 dxn
Each corresponds to the variation of y as one of the independent

variables changes, the other (w-1) variables remaining constant.
There are then n 2 second-order partial derivatives but this number is
reduced, in the case of ordinary continuous functions, since Young's
Theorem shows that the order of partial derivation is immaterial.
" "
There are, in fact, n direct partial derivatives
" "
and \n(n- 1) cross partial derivatives
The analytical interpretation of the partial derivatives extends at

once. The first-order partial derivatives, evaluated at a given point,
measure the rates of change of the function as one of the variables
increases from the given value, the other variables remaining
constant. Thus, if is positive at the point (a, 6, c), then the

uX
function u=f(x, y, z) increases as x increases from a, y and z having
" "
the fixed values 6 and c respectively. Further, the direct second-
" "
order partial derivatives measure the various accelerations of the
function, one of the variables increasing from the given value and
the others remaining constant. It is not possible, of course, to give
any concrete diagrammatic interpretation of partial derivatives
when there are more than two independent variables.
12.6 Economic applications of partial derivatives.

Under the conditions of 11.7 above, the market demand for any
good X r is a function of all market prices :
r
= ^r(A^, -Pn) (r
= l, 2, ... tt).
The partial derivatives of this function indicate the variations of

demand as one of the prices varies, other prices remaining constant.
The partial derivative

-
,
which must be negative in the normal
dp r
case, shows the rate at which the demand for X r decreases as its
price increases. It usually convenient to put this rate into elas-

is
ticity form the partial elasticity of demand for r with respect to X

its price :
xr dp r
This expression, independent of demand or price units, shows the

rate of proportional decrease of demand for proportional increases
in price. It is an extension of the ordinary demand elasticity
obtained when demand is regarded as dependent only on the price
of the good concerned. The ordinary demand elasticity is defined

only when the prices of other goods are fixed and known if these ;
prices change, the whole demand law shifts and the elasticity must
be evaluated anew. But our generalised demand law and the
demand elasticity -r\ rr obtained from it automatically allow for these
shifts in demand. In fact, rj rr is a function, not only of the price p r ,
but also of all other prices ;

its value is thus seen to change when
any of the prices change.

dx
The partial derivative
r
measures the rate of change in the
Ps
demand for X r when the price of another good X 9 is increased, and it
is best considered in relation to the corresponding partial derivative

dx
^
dp r
, measuring the rate of change in the demand for X a for increases
in the price of X If both partial derivatives are positive, then the

r.
demand for X r increases with the price of s and the demand

for X, X
increases with the price of r The goods r and X
s are then . X X
competitive, using this term in a broad sense. If both partial deriva-
tives are negative, the demand for one good changes in the sense
opposite to that of the price of the other good and the goods can be
called, in the same broad way, complementary goods.* It is often
convenient, again, to use the elasticity forms

3 (log ay) p s dx r d(logx 8 )^ = p r dx 8
i.e. the partial elasticities of demand for one good with respect to
the price of another good.
If the demand laws for two goods X l and X% are linear :
then the price elasticities of demand are
and ?
i?n=0ii;r;
#!
i?i=-0ii$;
Xl ^n=-n?
X 2
1
?22=22^
X2
and all depend on the prices assumed for both goods.
*
There remains the possibility that the partial derivatives are of opposite
signs.In this case, the goods, on the present definition, are neither competitive
nor complementary. See Schultz, Interrelations of Demand, Journal of
Political Economy, 1933, and 19.7 below.
On the other hand, if the demand laws are of the form
then we have linear relations when all variables are taken on log-
arithmic scales and the price elasticities of demand are constants :
'nn
= an ; 112
= - ^12 ; ^21
= - <% and 7/22
= a 22 .
It is to be noticed that the two elasticities of demand for one good

with respect to the price of the other are of the same sign, in each of
the above cases, only if a12 and a 21 are of the same sign.
The production function x=f(alt a 2 a 3 ... an ) shows the depend- , ,
ence of the output of a good X

on the amounts of the variable
factors AI, A 2 A^, ... A n used in the production. Suppose that a
,
and
dx
da
A. P.
a O
FIG. 83.
particular combination (o 1 a 2 a 3 ... an ) of the factors is used and

, , ,
A
that all factors except 1 are regarded as fixed in amount. The ratio
f
(the amount of product per unit of the factor AJ can be called the
ai
dx
average product and the partial derivative the marginal product
oa^
of the factor A^ at the combination (al5 2, a3 ,
... an ). The latter
measures the rate of increase of output as the factor A l is increased,
the amounts of other factors remaining unchanged. Both average
and marginal products are functions of %, a 2 a 3 ... an i.e. depend , , ,
on the grouping of the factors considered. Average and marginal

products of the other factors can be defined in similar ways.
With two variable factors A and JS, average and marginal products
can be shown diagrammatically on the vertical sections of the pro-
duction surface xf(a t b). The section of the surface by the plane
6 = 6! shows the variation of product for a variable use of the factor A

with a fixed amount b of the factor B. If P (a, b^ is a point on this
section, then the average product of A at (a, 6 X ) is shown by the
gradient (referred to Oa) of the line OP and the marginal product of A

by the gradient (to Oa) of the tangent PT. In the normal case of
production (see 11.8 above), the section, which can be called the
product curve for the factor A, takes the form shown in the first
diagram of Fig. 83. A second diagram can now be constructed
showing the variation of average and marginal products (see Fig. 83).
Average product increases to a maximum at a definite point a~a 2
where the tangent to the product curve passes through the origin.
Average product equals marginal product at this position. Further,
at some larger amount a~al9 product is a maximum and marginal
product zero. All three curves depend on the fixed amount 6 t of the
factor B used and change in position and form when this amount
b l is changed.
In the particular case where the production function is
x^2Hab-Aa*-Bb* (H*>AB),
T h2 dr
then ~ = 2Hb-Aa-B- and ~ = 2(Hb -Aa).
a a da
So
da \a/ a* oa* \aj a3
Hence, for .the fixed amount 6 X of the factor J?, average product of A
VgA ~~
A
b l9 and both average and marginal
- "
products are then equal to 2(H jAB)h^. In this case, the opti-
"
mum use of the factor A and the maximised average product
increase in proportion to the fixed amount of the factor B used. The
marginal product curve for A, when b=b l9 is a downward sloping
TT
straight line cutting Oa where a = ^bly product here being a maxi-

ô.
mum. The average and marginal product curves are of the normal
form of Fig. 83.
If u= cf> (#!, x 2y #3 ,
... xn ) is one form of an individual's utility
function for consumers' goods X X X19 2, 3 , ... X n) it might appear
,1 , ,, ,. T i .
x. du du du du
that the partial derivatives = = - - the
r dxl
,
dx 2
,
dx 3
,
. . .
dxn
represent
r
" "
marginal utilities of the various goods to the individual at a
set of purchases (xlt x 2 xs, ,

... xn ). For example, measures the
(/X-^
rate of increase of utility when the purchase of l is increased, the X

purchases of other goods remaining unchanged. But we have seen
that utility is a non-measurable magnitude to be indicated in
general by
where F(u) is any function such that F'(u)>0. Can any meaning
"
be attached, therefore, to increments of utility and so to marginal
"
utility ? We have
i.e. marginal utilities involve the arbitrary function and lose their F
definiteness when utility is regarded as non-measurable. But
d d d dU_duLf 3u du
t
'
t
'
t
.....
m
'
du_ t
'
u
.....
m
dx dx z dx$ dxn dx dx z dx s dxn
and the ratios of the marginal utilities are definite concepts quite
independent of the non-measurability of utility. This fact will be
developed in the following chapter (13.7 below).
Suppose, for example, that one form of the utility function is
where only two goods X and Y are considered by the individual.

Then
^ <_. \tns i
w i
\if | vx cllHU ,_
ox oy
du du -_- a B
i 'Q ________ '
_______
dx' dy x+a y+b
Another form of the same utility function is
du' a du'
,
whence - = and
,
-r = j8-
L .-
dx x +a ay y+b
The ratio of marginal

B utilities, 9
- : --=- ,
is thus * J definite
perfectly
y + b'
12.7 Homogeneous functions.

Whenthe variables x and y are increased or decreased in a fixed
proportion from given values, the corresponding increase or decrease
in the function zf(x, be in greater, in equal or in less pro-
y) may
portion. In the very special case where zf(x, y) increases or de-
creases always in the same proportion as x and y, the function is said
to behomogeneous of the first degree, or to be linear and homogeneous.
Such a function has the property that its value is doubled whenever
the values of the independent variables are simultaneously doubled
from any position whatever. And a similar result holds for any
ratio other than that of doubling. Hence :
DEFINITION: zf(x,y) is a linear homogeneous function if

f(Xx, Xy) =Xf(x, y) for any point (x, y) and for any value of A what-
ever.
To quote particular examples, each of the following function types

is linear and homogeneous :
(1) z=ax + by, (2) z~
(3) Z
= ^ + 2kXy + W, (4) ^
where the coefficients, a, 6, c, ... ,
and the index a are constants.
The fact- that z changes proportionally for proportional changes in
x and y can be verified in each case.
If P is a given point on the surface representing a linear homo-
geneous function, then any point with co-ordinates proportional to

those of P also lies on the surface. The points (x, y, z), (\x ly, \z) y
and on the Surface if one of them does. But all

(2x, 2y, 2z) all lie
points of this nature lie on the straight line OP joining the origin to
the given point P. In fact, the line joining to any point on a linear
homogeneous surface lies entirely in the surface. The surface is
completely described by lines passing through the origin and is a

particular case of what is called a ruled surface. It also follows that
the tangent plane at P
touches the surface, not only at P, but also
at all points on the line OP. A
cone with vertex at the origin is an
easily visualised example of the type of surface corresponding to a
linear homogeneous function.
The sections of a linear homogeneous surface by planes perpen-
dicular to an axis also reflect its special properties. If P is a point
(x, y) on the contour by the plane z
= z1? then the point Q (2x, 2y)
must lie on the contour by the plane z = 2zv But the points P and Q
lie on a line through so that OQ = 20P. Hence, any radius
through O in the plane Oxy cuts the zl and 2zx contours in points of
which the second is twice as far from as the first. The contour 2z1
is of exactly the same shape as the contour zx but radially double in
size. A similar result holds for any pair of contours :
All contours of the linear homogeneous surface are similar in shape

and vary in size radially according to the ratios of the fixed values
of z defining the contours.
" "
Any one contour is a radial
projection of any other and the whole
system of contours can be derived given any one contour. Further,
the tangents to the various contours at points where they are cut by
a given radius through must all be parallel.
Exactly similar results hold of sections of the linear homogeneous
surface by planes perpendicular to Ox or Oy. The various sections
of such a system are radial projections of any one section of the
system and vary in size according to the fixed values of x or y
defining the sections. Also, from the ruled surface property, the
vertical section of the linear homogeneous surface by a plane through
and a point P on the surface consists of a straight line.
The linear homogeneous function is the simplest case of a wider
class of homogeneous functions. More generally, if
for any point (x, y) and for any value of A, then the function z =/(#, y)
is homogeneous of the rth degree. For example, z = - j~ is
homogeneous of zero degree, z=ax + 2hxy + by is homogeneous of

2 2
the second degree and z = ax y is homogeneous of degree (a-f/J).

a p
The diagrammatic properties of a general homogeneous function can

be illustrated in the quadratic case (r = 2). The vertical section of a
quadratic homogeneous surface by a plane through and a given
on the surface is always a parabola with vertex at and axis
point
along Oz. The contours of the surface are radial projections of each
other and vary in size according to the square roots of the values of z
defining them. A
similar property holds also of the vertical sections
of the surface by planes perpendicular to Ox or Oy.
The definition of homogeneous functions extends without diffi-
culty to the case of functions of more than two variables. The
function yf(xl} x 2 x& ...) is homogeneous of the rth degree if
,
for any point (xly any value of A. The case r = 1

x z x3
, , ...) and for
gives the linear homogeneous function, and here the proportional

change in the values of all the independent variables produces an
equal proportional change in the value of y.
12.8 Euler's Theorem and other properties of homogeneous functions.

If zf(x, y) is a linear homogeneous function of two variables,
then the following properties can be shown to hold at any point
(x, y) whatever :
(1) The function can be written in either of the forms
z~s
where <f>
and are some functions of a single variable.
(2) The partial derivatives

--- and are functions of the ratio of
x to y only.
dx ^
(3) Euler's Theorem : x -
-f y =- z.
% dx By
(4) The direct second-order partial derivatives are expressed in
terms of the cross second-order partial derivative
d 2z y d z
2
. d 2z x d 2z
and 2
dx 2 x dx dy dy y dx dy
The proofs of these results can be set out formally :
Since /(A#, Xy)=Xf(x 9 y) for any value of A,
/ y\ 1 !
we have f I - - fix. y} for A = -
xj
[ , ,
^
\ x) x
i.e.
-/( 9--* CO-
since / ( 1,
-1 is a function of - only. Similarly, for A=~ ,
\ x/ x y
Finding the partial derivative of z with respect to x :

x<f) (-]
where <f>' (-) denotes the derivative of the single variable function
/ % \x/
-
-) with respect to x Further,
and -
Hence, both appear as functions of the ratio x only.
ox oy
which is Euler's Theorem. Finally, to prove the last result, we

notice that Euler's Theorem holds for any values ofx and y whatever,
i.e. x~ +y^~ is identically equal to z. This identity is maintained
no matter how x and y are varied and it follows that the partial
derivatives of one expression are equal to the corresponding partial

derivatives of the other.* Hence :
dz dz\ dz
dz d*z d 2z dz . d*z y
*** *N I " > n I
iJ *\ <> -\ J *.xy. _ o _ _
ox ox 2 ox oy ox ox 2 x ox oy
A similar result is obtained by partial derivation with respect to y.
* The equation of the

derivatives of each side of an identity is a device of
wide application 14.8 below).
(see, e.g.,Before applying the device, it is
essential to establish that the relation used holds identically for all the
variables with respect to which the derivation is carried out. relation A
holding only for certain values of the variables cannot be treated in this way.
Two examples make this clear. For all values of a?, we have
Hence -r-(
dx
But a2 + 2 ~ 3ic
holds only for the particular values x = 1 and x 2. Hence
as can be verified at x 1 or x = 2.
The truth of Euler's Theorem is also clear in diagrammatic terms.

The tangent plane (12.4 above) at the point (xl9 y l9 zj on the surface
z=f(x 9 y) has equation
dz\ . dz\
It passes through the origin if x~y = z= satisfies the equati
Euler's Theorem asserts that this condition is satisfied at all points

on a linear homogeneous surface. The tangent plane thus passes
through the origin at all points, a fact which agrees with the ruled
surface property of linear homogeneous surfaces.
The four properties generalise to the case of a function z=f(x, y)
which is homogeneous of degree r :
- and -
(2)
ox
are homogeneous of degree (r I).
oy
*
/\ oz z
(3) x + y~^rz.
ox oy
d 2z
The first of these extended results

established, exactly as before, is
by using the definition of a homogeneous function. The second and

third results then follow by derivation. The fourth result is obtained
from the third result, which
identically true for all values of
is x and
y, by finding the separate partial derivatives :
d dz dz\ _ dz
f
'
dx\dx dy/ dx
d*z d*z dz
i.e.
2
dx dxdy dx
a- - (r - ix
w
-i /
Similarly
i
x +y 1 -
.
J
dx dy *w=
dy*
(r l))
dy
Multiplying by x and y respectively and adding,
and extended forms, for
Finally, the properties hold, in modified
homogeneous functions of more than two variables. It is not
necessary to set out the properties in the completely general case
here the nature of the extensions is sufficiently obvious.*
;
12.9 The homogeneous production function.

linear
The case where the production function of a good X with respect
to variable factors A lt A 2 A 3 ... A n is linear and homogeneous is
, ,
one of particular interest. From the definition and results (1) and (2)
of 12.8 above, the case is characterised by the fact that a propor-
tional increase in all factors leads to a proportional increase in
product and leaves the average and marginal product of each factor
" "
unaltered. We
have, in fact, the case of constant returns to scale
where only the relative amounts of the factors used is important and
not the actual scale of production. For example, if wheat is produced
with land and labour under constant returns to scale, then the wheat
product is (e.g.) doubled when twice the number of men are employed
on twice the area of land. Further, the product of wheat per man
or per acre depends only on the number of men employed per acre,
as does the marginal product of wheat per man or acre.
In the case of two factors A and B and constant returns to scale,
the production surface is subject to restrictions additional to the
normal ones already indicated (11.8 above). The surface is ruled by
straight lines through the origin and any section through Ox consists
of a straight line. The constant product curves in the plane Oab are
now radial projections of each other and vary in size according to
the constant products which flefine them. In particular, any radius
through cuts the curves in points where the tangents are parallel.
So, in the normal case of Fig. 78, the curves OA and OB, marking
the points with tangents parallel to an axis, reduce to straight lines.
The sections of the production surface by planes perpendicular to
(e.g.) Ob are also radial projections of each other and,

in the normal
case of Fig. 79, the maximum points all lie on a straight line through
0, i.e. as the amount used of the factor B
increases, the maximum
product obtainable by varying A increases in proportion. Further,

there is one line through which touches all sections, i.e. the
* See Examples XII, 24-6, and 18.5 below.

maximised value of the average product of A is the same no matter

what fixed amount of B is used.
Euler's Theorem states that, for any combination of factors,
dx dx
%=a
,
da +b^dbl ,
and this can be illustrated diagrammatically (Fig. 84). P is the point

on the production surface for the combination (a, b) of factors and
the tangent PQ to the section perpendicular to Ob is drawn. Then
RP = RP
dx
a;
= gradient of PQ = ^ -=j. ,
i.e.
dx
a =- = JSP.
So ~
db da
M
FIG. 84.
The total product MP is divided, MR = b db and RP = a~-

-
da
At the
point A on the section where total product is a maximum, =

i da
oor or
and so - = j> At such a point, the average product of B is a
maximum on the other section perpendicular to Oa. Similarly, at

the point B on the section where the average product of A is a
flf fy* ^T
maximum. = - and so -^ = 0. i.e. the total product is a maximum
da a db
on the other section perpendicular to Oa. There is a correspondence
between the maximised total product for variation of one factor and
the maximised average product for variation of the other factor.
X M.A.
Thefollowing are simple examples of linear homogeneous pro-
duction functions corresponding to constant returns to scale :
(H
2
>AB in the two latter cases.) In each case, it is easily verified
- and 'i\
|n and the marginal products
(xd uI
f and are functions of - only, i.e. depend only on the ratio of

-^7 j
the factors used. Case (1) corresponds to the simpler case of normal
production and cases (2) and (3) to the full normal case illustrated
by Figs. 78 and 79.

EXAMPLES XII
Partial derivatives
1. Find the first and second-order partial derivatives of each of the

functions x -f y* - 3xy
3
; ===.
1
;
~
ex y and log --x
.
Verify Young's Theorem
v# 2 + y* x +y
that the order of partial derivation is immaterial in each case.
2. Evaluate the partial elasticities of x 2 e v and (x + y)ex+ v .
3. Obtain the partial derivatives of -* ~- -

azx + b2y + ca
- -
4. By logarithmic derivation, evaluate the partial derivatives of
(x-y)(x-2y)
and show that + y) n
e*+v} = + y)4**} = (x + y + n)(x
-j^{(x ^-{(a?
3n
5. Show that -{(X + y)e?*v} = (x + y + n)e*+*
and that *+ {(x
~
y) ^ V} = (x ~ y + 2r ~ n
X2
6. Show that z and its partial derivatives of all orders become
x-y
infinite at any point where x = y.
7. If 2 =/(w) where u is a function of x and y, show that
Verify in the case where f(u) = log u.

8. If z= ^ (u) + ^ (v), where ^ (w) is a single-valued function of u x -f ay and
^ (v) is a single-valued function of v = x-ay, show that = a8 Verify -
f
%
when ^ (w) = w 2 and ^ (v) = ev .
y
9. Evaluate the first and second-order partial derivatives of
u (x
z
+ 2xy-y z )ez
and verify that the order of partial derivation is immaterial.
10. Show that the surface z = \lxy (x and y positive) rises and is concave
from below in each of the fundamental directions.
11. If >0
<0 at a point (x, y), show that z increases whenever x
and
ox
oy
increases and y decreases from the point (x, y).
12. Find the equation of the plane passing through the points with co-
ordinates - 1, 0, 4), and (1, 2, 1) referred to
(2, 1, 0), ( rectangular axes.
Where does the plane cut the axes ?
13. If A is a parameter, show that the equation x + y + Az = 1 represents a

system of pianos intersecting in a given lino in the co-ordinate plane Oxy.
14. Show that the tangent plane at the point (x 19 y l9 zj on the surface
z \fxy has equation xy + yx v - 2zz 1 ~ 0.
15. Find the equation of the tangent plane at the point (2,
- 1, 2) on the
surface z = - - Where does it cut the axes? Deduce that z decreases
whenever x and y increase from the values x 2 and y= - 1.
Homogeneous functions
16. Graph the contours of the surface z \xy given by z 2 and z 4 and
show that the latter is radially double the size of the former.
17. that the contours of the surface z =

Show
circles which are radial projections of each other.
-
/>2 i A/2
form a system of
18. Show that the following are linear homogeneous functions :
Express each in the form z Obtain and --

and so verify Euler's
Theorem in each case.
x</> (
'). dx ^
19. Verify property (4) of 12.8 in the case of z = *Jxy.
20. Show that z (a t x + b^y) (a 2 x 4- z 6 2 2/) 1 ~ a and
aîy^--^ -f a^x^y^-^ are
linear and homogeneous and verify Euler's Theorem in each case.
- xy +
For each of the quadratic homogeneous functions 2y and
2
21. z x*
^3 _l_
-j/3
z _L 9 show that the partial derivatives are linear and homogeneous and
* that x
verify +y - = 2z.
dx dy
22. For the homogeneous function z = axayP, show that
23. If 2 is a function of x and y homogeneous of degree zero, show that z
can be written as a function of - only and that ~= -r- . Verify

J by con-
x dy ydx
...
sidenng
6 z = log& x-y
.
-
- and z =
x+y -^-^
, x* - xy -f
2x*
- -
2t/
+ xy + y*
- a
24. If y f(x lt # a x 9 ,
, ... xn ) is homogeneous of degree r, show that
2/1 T
at rft
_
V'
.
ft._2* x
fx I_a a? n \
'
9 '"
"* 1
'
where </>
is some function of (n - 1) variables. Deduce that the partial deriva-
tives are homogeneous of degree (r - 1) and that
26. If y is a linear homogeneous function of x l9 # a , ... a? n , show that
and similar expressions for the other direct partial derivatives.
26. Show that the relation between the second-order partial derivatives of
a function of n variables, homogeneous of degree r, is
Economic applications of partial derivatives and homogeneous functions

27. If the demand laws for two goods are given by (4) of 11.7, show that
" "
the direct price elasticities of demand are independent of the prices
" "
while the cross price elasticities are determined in sign by the constants
a 12 and a al respectively.
28. The employment of lOOa men-hours on b acres of land gives

x~2(l2ab- 5a 8 - 46 8 bushels of wheat. Graph the average and marginal
)
product curves for labour when 10 acres are cultivated. Compare with the
product curve of Fig. 80.
29. Draw the average and marginal product curves for labour employed
on (a) 10 acres and (6) 100 acres when x = -40

?(12ab - 5a - 46 )
a a
is the
wheat product for lOOa men-hours on 6 acres. Verify that the maximum
average product is the same in the two cases, ten times as much labour being
required in the second case as compared with the first.
OA ^ v ^ X- r x-
30. For the linear homogeneous production function x
2Hab-Aa*-Bb*
-^
Oci
=rr ,
-- - -\-JLJo
show that the average and marginal products of the factors depend only on
the ratio of the factors and verify that the product is always a times the
marginal product of A plus b times the marginal product of B.
31. The production function x - */2Hab - Aa 2 Bb 2 is linear and homo-

geneous show that the maximum value of the average product of A is a
;
/Tyg A j>
constant A/ independent of the fixed amount of B used.

32. The production function is x Aa where a + 0< 1. Show that there a bP
are decreasing returns to scale and deduce that the total product is greater
than a times the marginal product of plus 6 times the marginal product of B.A
33. If a men are employed in planting 6 acres with timber, the amount of
timber cut after t years is x=f(a, b, t). What meanings can be attached to
~
-
, and -r- ? If x = Aaa b l atP, where a and Q are fixed positive fractions,
da db dt
show that there are constant returns to land and labour after a fixed time t
and that, for given employments of land and labour, the timber product
increases, but in a decreasing proportion, as time goes on.
34. Find the ratio of the marginal utilities for two goods when the utility
function is u = ax 4- by + c*fxy. Verify that the same result is obtained when
the utility function is written u' log (ax 4- by 4- c^/xy).
36. If U F(u) is the index-function of utility where u axy, show that

(a)
-
= when F(u) = u; (b) --<0, when F(u) = *Ju\ and (c) -->0,
36. If u= <f> (x, y, z, . . .

)
is one form of the utility function for goods
X, Y 9 Z, ... and ifU = F(u) is the index -function of utility, show that
and similarly for the other second -order partial derivatives. Deduce that the
sign of each of these derivatives varies, in general, with the choice of the
arbitrary function F. Illustrate with tho results of the previous example.
CHAPTER XIII
DIFFERENTIALS AND DIFFERENTIATION

13.1 The variation of a function of two variables.
THERE only one way in which the value of a function of a single

is
variable can change, i.e. by changing the value of the independent

variable. The variation of the function is then adequately described
by means of the derivative. The rate of change of ?//(#) from any

point x is measured by f'(x) and Ay~f'(x)Ax is the approximate
increment in y for an arbitrary small increment Ax in x.

The expression of the variation of a function of two variables is
much less simple. The value of zf(x, y) changes when x changes
(y remaining constant) at a rate measured by ox .

Hence, if A xz
denotes the increment in z due to an arbitrary small increment Ax
in the variablex from a point (x, y), then
AgZ=~ Ax approximately.
In the same way, if A y z denotes the increment in z due to an

arbitrary small increment Ay in the variable y from the point (x, y) >
then
A vz ~Ay approximately.
But these are only two special ways in which the value of the function
can change and there remains the important problem of expressing
the variation of the function when the independent variables vary
together in any way whatever. A single partial derivative is not
sufficient here and an addition to our mathematical equipment is
needed.
It is assumed that the function z=f(x, y) possesses continuous
partial derivatives at the point (x, y). If h and k denote arbitrary
DIFFERENTIALS AND DIFFERENTIATION 327
increments in the variables x and y from the point (x, y), the corre-
sponding increment in the value of the function is
h
But, from the definition of a partial derivative,
-. _
9
fl OX
as
since the partial derivative is continuous. It follows that
-- --
fl
- - -
d ,,
- --
--/(*,
uX
y) + *, where eÔ
. _
as A and &->0.
Again,
, where ,-0 t-,0.
Hence, writing the partial derivatives in an alternative notation,
Jz~ U" + e ^+ (^ + ^j
) ^ J
where e and 7^->0 as h and &->0.
The expressions and 77 must be small when both h and & are small.
So, changing the notation for the increments in x and y, we have :
The increment in the function z f(x, y) corresponding to arbitrary

small increments Ax and Ay in x and y is approximately
dx
-. dy
This is a result of the first importance. The increment in z when

x varies alone is represented approximately by ox Ax and the
increment when y varies alone approximately by Ay. It is now

established that the increment in z when x and y vary together is ex-
pressed approximately by the sum of the two increments in the simple
directions. The two
partial derivatives can be used, therefore, to
describe the general variation of a function of two variables. But
it is clearly convenient to have a precise
symbolism for this general
variation and a technique for operating upon the symbols. The
" " " "
concept of a differential and the process of differentiation are
introduced to serve just this purpose.
Our fundamental result can be written in a variant form as
/(a + *,6 +*)=/(, b)+hfx '(a &)+*/,>, y b)
approximately when h and k are small. x~a and y~b are

Hence if
inserted in f(x, y) instead of the correct values x = a 4- h and y = b + k,

then the error in the value of the function approximately is
hfx '(a,b)+kfy '(a,b).

Again, putting x a + h, y b + k, we have
- a) x '(a, b) + - 6) y '(a, b)
f(*> V) =-/(, b) + (x f (y f
approximately when (x-a) and (y-b) are small. This result
enables us to replace f(x, y) by an approximate linear expression in
x and y for small ranges of values about the values x a, y~b
(cf. 6.4 above).
13.2 The differential of a function of two variables.

The increment in the value of a function z=f(x, y) for small
increments Ax and Ay in the independent variables can be reduced
to two parts, a precise expression, ~-Ax + ~~Ay, which is of the same
ox ay
order of smallness as the increments Ax and Ay themselves, and an
expression, cAx-i- yAy, depending on values e and 77 which are small
when Ax amd Ay The second expression is thus of the

are small.
order of the squares of Ax and Ay. The precise and first-order part
of A z is termed the differential of the function zf(x, y) :
DEFINITION : The differential of a function z=f(x, y) with con-

tinuous partial derivatives at a point (x, y) is
. dz A dz .
az = ^r Ax+-~ Ay
y 9
'
dx dy
where Ax and Ay denote arbitrary increments in the independent
variables from the point (x, y).
For convenience, we now denote the arbitrary increment Ax by

dx and call it the differential of the independent variable x. Simi-

larly, Ay is written dy and called the differential of y. The differ-
dependent variable z is then defined in terms of the
ential of the
independent by the formula

differentials
dz =
7
dZ 7
ax
^-
- 7
-f- ay.
9# 9/y
It must always be remembered, in using this formula, that dx and dy

are no more than arbitrary increments in the independent variables.
An alternative notation for the differential of z f(x y)9
is
df( x > 2/) --/x'fa > y) *B -h/V fa, ?/) <fy =/x
* +/ dy.
The interpretation of the differential of a function follows from
the definition. The statement that dz-^-dx +

dx
^dy implies
r that
dy
the increment in the value of zf(x. y) is approximately
. dz A dz A
Az^---Ax + ~-Ay
dx dy
when small increments Ax and Ay are allotted to x and y. The differ-
ential dz replaces the increment Az in much the same way that the
derivative replaces the average rate of change. So z=f(x, y) tends
to increase (or to decrease) whenever x and y are varied by small
amounts so that the corresponding differential dz is positive (or
negative).
The expression dz dx +
^ dy is often called the "complete"
differential of the function z f(x y) and from it can be derived two

y
partial differentials. When x is varied by an amount dx while y is
held constant, the differential becomes

dz
dz =
,
-
dx
dx
and the partial derivative is thus the ratio of the differential dz

dx
(y held constant) to the differential dx. Similarly, the partial
derivative is the ratio of the differential dz (x held constant) to

^-
y
the differential dy. A partial derivative can be regarded as the ratio
of two differentials.
It can be noticed that the definition of a differential applies
automatically to the case of a function of one variable. If y=f(x)
has a continuous derivative /'(#), then the differential of y is
dy=f'(x)dx
for an arbitrary increment dx in #,* This implies simply that
Ay~f'(x)Ax approximately
when the arbitrary increment Ax is small. It follows that the
derivative f'(x) is the ratio of the differential dy to the differential dx.
dij
The notation -
is now very useful since it can be regarded either as
dx
a derivative or as a ratio of differentials. The differential notation,
however, adds nothing to the derivative notation in the case of a

function of one variable. It is only when at least two independent
variables are involved that the differential notation becomes essential
in the description of variation.
13.3 The technique of differentiation.
The process of obtaining the differential of a function is called

differentiation. The most obvious method of differentiating z -/(#, y)
is to evaluate the partial derivatives and to substitute in the funda-
mental formula dz = ~dx + dy. But other methods are often

dx dy
more convenient. It can be shown quite easily that the rules obeyed
by differentials are exactly similar to those obeyed by derivatives.
If u and
7 /
d(u
v are
XT
v)=du
two functions of x and
7
dv ;
J I
d(uv)
X 7
udv -\-vdu
y, then
7
;
7
a I-
fU \
I
-
= V dU
~
-
Further, if z=f(u) where u is a function of x and y, then

dz f'(u)du.
dii
So d (un ) = nu*~ l du ; d (eu ) eu du\ d (log u) = --
As examples of the method of proof, we have
d d
d (uv) = ^- (uv) dx + (uv) dy
ox dy
du\ 7 / dv du
U
(dvX
d-
dv - \ Idu , du , \
u (dv y ~
,
^-dx +
\ dy +v \-z-dx + dy] ]
\dx dy J \dx dy /
*The derivative /' (x ) is sometimes called the "differential coefficient"; it is

the coefficient of the differential dx in the differential of the function y =/(#).
Again, if z=f(u), then

dz .du , dz du
=f'(u)
J and =f(u) =-
dx dx dy By
. dz _ 9z _ ... /du 9% \ ...
dz = r-dx + =-dy =Jf
.
7 v 7 7 7
and -
t (u) \ =-dx + dy] j (u)du.
dx dy \dx dy /
A good practical method of differentiation is to split up the given

function into groupings of functions u, v, w, ... each of which in- ,
volves only one of the variables x and y. The differential of the

function is expressed in terms of the differentials of u,v,w,... by
the above rules. The latter differentials are then given by ordinary
single-variable derivatives, e.g. the standard forms
dx
d (xn ) nx"- 1 dx ;
d (e x ) ~e*dx\ d (log x) ,
x
and similar results in the variable y. Finally, it is to be noticed that
(J11
the result d (log u) a method of logarithmic differen-
u provides
tiation. For example, if u, v and w are functions of x and y and if
QI9J
z~ ,
then on taking the logarithm
log z =log u H- log v - log w

dz du dv dw
and =
W
1
Z U V
The practical technique of differentiation is illustrated by the

following examples. Each result can be checked by using the
partial derivatives already obtained (12.2 above).
Ex. 1. z=x2 + 2xy-y2 .
Here dz = d (a: 2 4- 2d (xy) - d (y*)

)
= 2xdx+2(xdy+ydx)-2ydy
Ex.2. z=-
x-y + l
Here dz=~
-y + l)2xdx-x 2 (dx-dy)
2
(x-y + l)
2
Ex.3. z=log(* +#2 ).
So z log u where u = x 2 + y2 ,
du
and
tillU.
,
dZ
d(x*+y*)
^r
= ^xdx+ydy
u .
u x z +yz xz + y z
Ex.4. z=t
'x-y
So log z =2 loga; + |log(a; + 2/) -ilog(#-y),
T~ -
Id(x-y)-
a; 2 # + ?/ 2 x-y
-y z )dx+x(x-y)(dx -
+ dy) -x(x + y)(dx -dy)
~ (2x
2 - x?y - 2 y2 ) <fa + a:2
t % '
-
^^^^2)
i.e. dz = -^ - _= ^
{(2a;
2 -
a;y
- 2
2y ) dx
13.4 Differentiation of functions of functions.
The differential dz=-z-dx + ~dy is defined for a function of two

dx dy
independent variables, dx and dy being then arbitrary increments.
But exactly the same formula can be shown to give the differential
of a function in which the variables x and y are not independent
but dependent on some other set of variables. In this case, dx and dy
are no longer arbitrary increments they are proper differentials ;
depending on the arbitrary increments in the actual independent

variables defining x and y. The proof of this extension of the
differential formula is too involved to be given here.* It can be
assumed to hold at least for ordinary continuous functions. Hence,
The differential of the function z =/(#, y) is given by

dz dz
dz ~dx + T
dy
J
7
dx dy
whether the variables x and y are independent or not.
The extended formula is of importance in the evaluation of the

differentials,and hence of the derivatives, of various types of
functions of functions. The simplest case of a function of a function
*
See, for example, Phillips, A Course of Analysis (1930), pp. 229-31.
has already been considered. If zf(u) where u is a function of

x and y, then we have seen that
dzf'(u) du
-i
dz /.// v du dz ,, v du
and -~=f'(u)~
J v
and ~-=/(tO;T'
dx dx dy dy
We can now
pass to the two most useful cases of more complicated
functions of functions.
In the first case, z is a given function of two variables x and y
which are themselves given functions of a single independent
variable t. Hence, z is a function of t and the derivative -=- is

at
required. Write z=f(x 9 y) with partial derivatives ~ and .
dx y
Further, write x = (f>(t) with derivative -~- = <f>'(t)
and y = i/j(t) with
.. (M>
derivative - =$ r
(t). Then
Cut
dz dz
dz = ~dx + ~dy,
J
dx dy
where dx = <f>'(t)dt and
So *
The ratio of the differentials dz : dt is to be interpreted at once as the

derivative of z as a function of t. With a small notational change,
the derivative can be written
dz _ dz dx dz dy
dt dx dt dy dt
This result is a very simple one and, in practice, all we need do is
to write the fundamental formula dz=~dx + ~dy, divide through

to
dx dy
by the differential dt and interpret the ratios of differentials as
derivatives according to the definition of the problem. As a par-
ticular case, if z=f(x, y), where y is a given function of x, then
dz __
dz dz dy
dx dx
^}y dx
In the second case, z is a given function of two variables x and y
which are themselves functions of two independent variables u and v.
It is required to find the partial derivatives of z considered as a
function of u and v. Write z =/(#, y) where x = <f>(u, v) and y = *l*(u 9 v) 9
all functions having known partial derivatives.
mi
Then
t
dz= ~
dx dy
dx dx dy dy
.
where dx=du
,
du
+ ~dv
ov
and dy = ou
^du+~dv.
dv
,
~ 7 fdzdx dzdy\ fdzdx dzdy\ .

So dz= [
Veto cw
+ J /
oyouj\dxdv
7
du + U- 3- +3- / )
oyovl
dfo-
The partial derivative of z with respect to u is the ratio of the differ-

entials dz and du when t; is kept constant, i.e. it is the first bracket in
the above expression for dz. The partial derivative of z with respect
to v is obtained, in the same way, as the second bracket in the ex-
pression for dz. Hence,

dz __ dz dx dz dy dz
__
dz dx dz dy
du dx du dy du dv dx dv dy dv
These results are again very simple. In practice, we divide the
formula dz ~dx-i--dy through by either du or dv and interpret
the resulting differential ratios as partial derivatives.

The basic form of the differential formula is thus capable of
dealing with various types of functions of functions it is merely ;
a matter of making the interpretation appropriate to the case under

consideration. This is a striking illustration of the flexibility of the
differential notation. Our results are still limited, however, to the
case of single-valued functions and it remains to extend the appli-
cation of differentials in the construction of a theory of multi- valued
and implicit functions.
13.5 Differentiation of implicit functions.

Two x and y are related by a given implicit relation
variables
f(x, j/) = 0. In general, we now have y as a multi-valued function

of x and # as a multi- valued function of y. The simplest method of
dealing with such a relation is to introduce a third variable z which
takes its value from /(#, y) for any values of x and y whatever. The
single-valued function z=f(x, y) has the given implicit function as

its zero contour by 2 = 0, i.e. the given implicit function can be
studied by relating those values of x and y which make 2 equal to
zero in the explicit function 2=/(#, y).
As x and y vary in any way, independently or not, the variation
of 2 /(#, y) is given by the complete differential
dz =fx dx +fv dy
where /^ and/y are the partial derivatives of f(x, y). If (x, y) are
values making 2 = and if dx and dy are variations from these values
keeping 2 = 0, then dz = and
fx dx +fy dy = Q
is the relation between the differential increments satisfying the
given implicit function f(x, y)=0. Hence,
dy - ** dx - &.
7- and
<1I1U -7- -jr
dx fv dy fx
Starting from a given point satisfying the relation /(#, y)
(x, y)
=
and remaining on one branch of y as a function of x, let Ay be the
increment in y corresponding to a given increment Ax in x. Then,
by the definition of differentials, the ratio dy dx written above
: is
the approximate and limiting value of the ratio Ay Ax when Ax :
is small. Hence, the ratio dy dx can be interpreted as the derivative

:
of y as a function ofx given by f(x, y) = Q, taking only one branch of

this function through (x, y). In this way, the notion of a derivative
is extended to apply to multi-valued, as well as to single-valued,
functions. The ratio dx dy is interpreted, in the same way, as the

:
derivative of x as a function of y given by the same implicit function.

Hence,
The implicit function /(#, j/)=0 gives y as a function of x with

derivative -~-= -^ at the point (x, y), and it gives a; as a function
**
dx f
of y with derivative -=- = -^ at the point (x, y).
dy Jx
The two functions are inverse to each other and it is seen that one
derivative is the reciprocal of the other.
One point must now be stressed most strongly. The derivatives
/JtJ ny
p and -7- , being expressed in terms of the partial derivatives of
f(x, y) = 0,
are themselves functions of both x and y. The derivatives
differ, therefore, from those of single-valued functions and the reason
for this is not difficult to see. If y is a single- valued function of x,
then it is sufficient to fix a value of x, the corresponding value of y
fid/
and of the derivative -~ being uniquely determined. The derivative
CLX
of a single-valued function of x depends only on x. But, if y is a

multi- valued function of x, it is not sufficient to fix a value of x since
this still leaves several equally possible values of y at choice. The
branch of the function, and hence the derivative ~ ,

can only be
dx
determined when the starting-point is specified as regards both x
and y. The derivative of a multi-valued function is essentially
dependent on both x and y.
These remarks can be illustrated by the simple relation
Here f(x, y)
= a2 + y - 16 and fx = 2x, fy - 2y.
So -j-
= at the point (x, y).
ax y
The values (x, y) must, of course, satisfy the given relation. In this
case, y is a double-valued function of x and it is possible to separate
the two branches :
yWIcTâT2 and y=-N/l6â

On the positive branch, we have the single-valued function
which has derivative, at any value of x, given by

dy x _ x
" '
dx~
On the negative branch, y~ - \/16 -x 2 with derivative
x _ x
dy__ ~
dx~~
For a given value of x, therefore, there are two different values of
~ , one for each branch of the function. But both values are included
dx .
(1 tJ *!*
in the general result = : the implicit form of the function and

dx y
of the derivative expression has thus great advantages even when

the branches of the function can be separated.
The derivative of an implicit function given by a relation f(x, y) =
is evaluated at once from the partial derivatives of the expression
f(x, y). In practice, however, it is often better to differentiate

f(x, y) Q as it stands according to the method indicated in the
following examples.
Ex. 1 . The relation x2 + y2 - 2x + ty + 1 = gives y as a double-valued

function of x. From
/(Z, y)-z 2 + */ 2
we have fx = 2(x-l) and
SO
dy =
jT-
ax
--
x-l
+2 rt
'
y
Alternatively, differentiating the relation as it stands.
x-l
dy = ---
dx
-, .
dx
_
i.e. -j- ~ as before.
Ex. 2. The relation x* +y3 - 3j:?/ = gives as a triple-valued ;<y

function
of x. From /(x, ?/)
= # + y 3 - 3xy, we have
3
/ = 3a^-3y and /y -3?/ 2 -3;r.

B
OO
^
"7
dx
" ==
^ -?/
-- o
y -x
2
-
2
"
Alternatively (*
3
) +( M 3
) -3(x) =0,
which gives the same value of the derivative.
The second derivative of an implicit function is defined and

obtained from the first derivative. If/(#, y) =0 gives y as a function
of x, then from ~- = --J- ,

we obtain
dx Jy
M.A.
I (jj) - and
fjfl
Substituting these expressions and the value of - -

,
we have
ctx
<Py = f
The second derivative thus involves the first and second-order

partial derivatives of f(x, y) and is a function of x and y. For
/
x7^y
example, the relation ^

2
+ j/ 2 = 16 gives
- -= and
y
In diagrammatic terms, the implicit function f(x, y) = Q is repre-
sented by a curve in the plane Oxy. In general, one branch of the
curve passes through a point (x, y) satisfying the relation and there
isa definite tangent to the branch at the point. The tangent gradient,
referred to Ox, is then given by the value of the derivative
S= -**>.
Further, the curvature of the branch at the point is indicated by the
value of -r- at (x, y). Our previous discussion of the applications of
ctx
derivatives now extends at once to the case where we have a

curve with an implicit and multi-valued equation. It must be
remembered, however, that the results refer only to a single branch

of the curve, the branch passing through a specified point (x, y).
The equation of the tangent at a point (xl9 y ) on the curve
f(x,y)=Q is readily obtainable. The tangent is the line passing
through the point (xly yj with gradient to Ox given by the value of
-~ = -
dx y
jv
at this point. The tangent equation is thus
where fXl and fVl are the values of the partial derivatives of f(x, y)
at the point (x^ y^). This is a generalised form of the tangent
equation previously obtained (6.5 above).
For example, in the case of the circle with equation
we have /*,
= 2(^-1) and .^
The tangent at (xl9 y) on the circle has equation
-
1) (x
- x,) + ( + 2) (y - yi ) =0,
(*i yi
i.e. (Xi-l)x + (y l + 2)y=xl *
Since the point (xl9 yj lies on the circle, we have
i.e. a?!
2
+ y^ - xl + 2^ =^ - 2^ - 1.
The equation of the tangent is thus
13.6 The differential of a function of more than two variables.

The definition of a differential extends at once to the case of a
function of more than two variables. The
variation of u=f(x, y, z)
as the independent variables vary is given by the differential
du 7 du du
. ..
du
,
-- ax + ay 4- - dz.
dx By dz
This implies 'that, if Ax, Ay and Az are arbitrary small increments

in the independent variables, then the increment in u is approxi-
mately
A du
A A du, A du
Au~~-Ax+~~Au+~
dx dz
Az.
dy
The complete sum of the separate (approxi-
differential is again the
mate) variations due to changes in x only, in y only and in z only.

The differential formula again extends to the case where x, y and z
are no longer independent variables and it can be used to give the
derivatives of functions of functions of various types.
The variation of variables x, y and z connected implicitly by some
relation /(x, j/, z)=0 must satisfy
where fX9 fv and/z are the partial derivatives off(x, y, z). Regarding
z as a multi-valued function of x and y defined by this relation, we
obtain the partial derivative ~ as the value of the ratio dz dx for :
y constant. Putting dy = Q in the above equation, we have
.e. -
\dx/y constant fz
^2
cv -i i
= fdz\
U- = ~ fy
Similarly ) T-
dy \dyj x constent /,
In diagrammatic terms, a three-dimensional surface with equation
form f(x, y, z) = has gradients (referred to Oxy) in the
in implicit
two fundamental directions given by the above ratios of partial
derivatives. The tangent plane at (xl9 yl9 2j) thus has equation
(V
- 2/i) (see 12.5)
-g
i-e. / (x +/, (y - yj) +ftl (z - zj = 0.
- xj)
More generally, for a function y=f(xlt x a ... xn of n independent
, )
variables, we define
The complete differential contains as many additive terms as there

are independent variables. The interpretation and application of
this formula are exactly similar to those already discussed in the
simpler cases above.
13.7 The substitution of factors in production.

The production function x=f(a, b) for two variable factors of
production gives a system of constant product curves

/(a, 6)
= constant.
In the normal case, one curve of the system passes through each
point (a, 6) of the positive part of the plane Oab and is downward
sloping and convex to the origin (at least over a certain area of
the plane). Differentiating and denoting marginal products by
. dx dx
and
/=ai f*=db>
fa da+ft db=0
is the (approximate) relation between increments da and db in the
factors along the constant product curve through (a, b). This
relation holds for any point (a, 6) we care to select. Hence, the
tangent gradient of the constant product curve through (a, b) is
**=J
da fb
when referred to the axis Oa.In the normal case and over the
relevant area of Oab, this gradient is negative and its numerical
value is termed the marginal rate of substitution of the factor B for
the factor A in the production of the good X :
r=-*U-
da
fb
The value of r depends on the combination of the factors considered,
i.e. is a function of both a and b. It represents the additional amount
of the factor B, from the given combination of factors, necessary to
maintain product unchanged when a small unit reduction is made
in the use of the factor A .
Since the constant product curves are convex to the origin in the
normal case, the value of r must increase as b increases (and a
decreases) along the constant product curve. The convexity con-
dition thus the expression of the principle of increasing marginal
is
rate of substitution, of the assumption that it becomes increasingly

more difficult to substitute B for A as the substitution proceeds.
The interesting point now is to determine how fast r increases, i.e. to
" "
measure the elasticity of substitution. For any change from
(a, b) along the constant product curve, d (-) represents the in-
\a/
crease (or decrease) in the use of B as compared with that of A and
dr=d (^) the corresponding increase (or decrease) in the marginal
\Jb'
rate of substitution. The ratio of these differentials, expressed in
proportional terms to make them independent of units of measure-
ment, is defined as the elasticity of substitution between the factors
at the combination of factors considered. Hence,
DEFINITION : The elasticity of substitution between A and B is
= ~'
!(-)
b \al
where the differentials correspond to a variation along the constant

product curve through (a, 6).
In diagrammatic terms, a appears as the ratio of the relative
increase in the gradient of OP to the relative increase in the tangent
gradient of the constant product curve at P as the point P moves

along a constant product curve in the plane Oab.
The value of a can be written in terms of the partial derivatives
of r or of the production function itself. We have
~
a db - b da 1 7 dr T dr 7,
2
and dr da -f == db.
a da db
But db - ~da = -r da, and

J
so
Jb
fb\ ar~\-b , . / dr dr\

ar = -
.
W-
T - da/
1
a I I
-
2
da and I r -
) da.
a \ db
TT r ar +b
Hence,
ab dr dr
db da
Evaluating -
& da = da ( ) and-
(
-
)
~
in terms of the first and
\fb j db db \fb j
second-order partial derivatives of /(a, 6), it is found that
abT
where T= -
(faa fb
2 -
2fab fJ> +fbb fa
2
).
From appears that, although a has been defined

this last result, it
for the substitution of B for A, the same value is obtained when
the substitution is of A for B. The elasticity of substitution is
symmetrical with respect to the two factors.
The curvature of the constant product curve at (a, 6) is
d*b .
d tdb\ f -_
d , , - _ (dr
I I
dr db\ - - &
dr
_
dr
j /y\ j
2
da da \da/ da \da db da) db da
d zb
Hence, a is a positive multiple of the reciprocal of ,
i.e. it is
aa
positiveand inversely proportional to the curvature of the constant
product curve. The larger is the value of or, the flatter is the constant
product curve and the more slowly does the marginal rate of substi-
tution increase as B is substituted for A. The magnitude of CT is
thus an indication of the ease with which product can be maintained

by substituting There are two limiting cases. If A and B
B for A.
are perfect substitutes, so that product is maintained by increasing
B in proportion as A is decreased, then the constant product curve

2
d b
is a straight line, -=-^ is zero and a is infinite. If A and B are entirely
incapable of substitution, being needed in a fixed proportion, then

an increase in one of the factors from this proportion must leave
product unchanged. The constant product curve has a right angle
at the point concerned, -=-^

is infinite and a is zero. As a increases
from zero to infinity, substitution between the factors becomes
increasingly easier.
So, the elasticity of substitution a varies from one combination of
factors to another, is independent of the units in which factors and
product are measured, is a symmetrical relation between the factors,

is positive for all normal combinations of factors and varies from zero
to infinity according to the ease with which one factor can be sub-
stituted for the other in production.
When the production function is linear and homogeneous (constant
returns to scale), a becomes of simpler form. Since
/aa =-/.> and /w (12.8 (4) above)

=-J/
we have r=
and a=
' , ,
*
= -
(by Euler's Theorem).
Using the alternative notation for partial derivatives,

dx dx
da db
x-
dadb
When there are constant returns to scale, cr is inversely proportional
to the cross second-order partial derivative of the (linear and homo-
geneous) production function.* In the particular case where the

~a
production function is x = Aa b
a l
,
it is easily seen that r~-
and that cr= 1 at all points.
* It
was in this form that cr was first defined by Hicks, The Theory of Wages
(1932), pp. 117 and 245. See, also, Robinson, The Economics of Imperfect
Competition (1933), pp. 256 and 330, Lerner and Kahn, Notes on the Elasticity
of Substitution, Review of Economic Studies, 1933.
13.8 Substitution in other economic problems.
Exactly similar definitions can be given for the marginal rate and
elasticity of substitution for an individual consumer's scale of
preferences. A system of indifference curves for two goods X and Y
is defined by the equation
<f> (x, y)
= constant.
The tangent gradient to the curve through the point (x, y) is
dy = _^ '
dX <f)y
In the normal case, the curves are downward sloping and convex to
the origin. The marginal rate of substitution of the good Y for the
good X in consumption, defined as

_^ _ x
~dx~fa
isthus positive and increases as the substitution of for proceeds. Y X
" "
It is to be noticed that E
is the ratio of the marginal utilities of
X and Y and independent of the form we assume for the individual's
utility function (see 12.8 above). But there is no need to consider
" "
marginal utility at all ;
is sufficient for the
the expression R
description of the indifference map. The elasticity of substitution
between X
and Y is then defined as in the production case, taking
values between zero and infinity according to the ease with which X
and Y can be substituted in consumption to maintain a given level
of indifference.*
A similar analysis applies for a consumer's preference scale for
incomes in two years. If x and y are the amounts of this and next
year's incomes, we have an indifference curve system given by
if; (x, y)
= constant.
The marginal rate of time-preference is defined as the numerical
gradient of the indifference curve through (x, y), reduced by unity f

<
**'
dx
* See Hicks and Allon, A Reconsideration of the Theory of Value, Economica,
1934.
The reduction of - -~ made since we wish to define a zero time-

f J
f by 1 is
preference as the case where an equal addition to next year's income compen-
sates for a given reduction of this year's income, i.e. where ( - -~ ) = 1. This
is only a matter of the scale on which time-preference is measured.
The elasticity of substitution between the two incomes is then
defined exactly as before.
The same concepts are also of use in other problems, e.g. in
problems involving the notion of a transformation function (see
11.7 above). If two goods are produced with given resources so that
F(x,y) =
is the relation between the outputs x and y, then the expression
gives the marginal rate of substitution of Y production

(--r-jj-r
CLXj y
for Xproduction. Similarly, if x and y are this year's and next

year's incomes obtainable from given resources, the expression
written is the marginal rate of return over cost. The corresponding
elasticities of siibstitution then follow as before except that a
negative sign must be added, the transformation curves being

concave, and not convex, to the origin in the plane Oxy.
13.9 Further consideration of duopoly problems.

In the duopoly problem with conjectural variations zero, as
analysed in 8.8 above, the reaction curve Cl of the first duopolist
gives #! as a function of x 2 defined by
f(xl9 x 2 ) = ifi(x)+x 1 *li

l
(x)--r- = Q. (where aj= l +a: a )
The equilibrium outputs are then determined by the intersection of

C and the similar curve (7 2 for the second duopolist. For stable
equilibrium, we require that C be downward sloping with a gradient
(to the axis Ox 2 less than unity and similarly for C 2
)
It is now .
possible to investigate the implications of these conditions.

Differentiating the above implicit relation,
Xx
where p =-
dx ^
Hence, is negative and numerically less than unity provided
that p is positive. It is assumed, in the normal case, that the

p = \ft(x)
*
demand function is decreasing and that marginal cost
is increasing. It follows that the numerator of p is positive. The

condition we require then reduces to
which is satisfied either if )

is negative or if ^"(x), though
- These alternative conditions apply
positive, is less that { $'
to the convexity of the demand curve and one or other is satisfied

in many cases of demand laws, e.g. in the case of the linear demand
law pft-ax.
A rather more complex duopoly problem is the following. The
first duopolist is assumed to act as before, expecting his rival's
output to be unaffected by his own actions. The reaction curve C
FIG. 86.
dx
is then obtained with gradient -=-^ as given above with value lying
between - 1 and in the normal case. The second duopolist is
assumed to forecast correctly the effect of his own actions on his

dx
rival's output, i.e. his conjectural variation is -=-^ as obtained from
Cv The second reaction curve gives x 2 as a function of xl defined by
dn z . .
=
-j
a# (where a?
2
Since $'(x) < and < 1 + -=-^ < 1, the left-hand side of this equation
dx%
is greater than the form {^j(x)-\-x^'(x)} in the case where his con-
jectural variation is zero. Hence, for any given xl9 ~~ must be
greater than before and, since this is assumed to be an increasing

function of x 2) the value of x 2 must be greater than before. The
new reaction curve G 2 is thus farther to the right than the old curve
'
(7 2 Ox 2 being horizontal as in Fig. 85. If xl and x 2 are the old

,
'
equilibrium outputs, and #/ and x 2 the new, it follows that #/ <x^

and x 2 >x 2. The second duopolist gains by being "alive" to his
rival's reactions. Further, since Cj has a numerical gradient (to Ox 2 )
less than unity, the sum of #/ and x% is greater than the sum of
Xi and # 2 at least for a small shift in C 2 The joint output is greater,
, .
and the common price charged by the duopolists smaller, than in the
previous case.
EXAMPLES XIII
Differentiation
1. If h and k are increments

in x and y from the point (x, y), find the
increment in z = x 2 4- 2xy - y 2 in terms of x, y, h and k. What is the difference
between this increment and dz? Show that the difference is of the second
order in h and k.
Write down the equation of the tangent plane at (a, 6, c) on the surface
2.
zf(x, y) and show that the increment in the height of the plane when incre-
ments dx and dy are given to x and y from (a, 6, c) is dz. Hence illustrate the
approximation of dz to the increment in z.
- = ^/x 2 -\-y 2 ~
3. Evaluate the differentials of z x3 4- y
3
3xy ; z ; z ex v ;
x
z = \og
4. Obtain the differential of z = (x + y) (x - y) by logarithmic differentiation

and check by 2 2
differentiating z x -y directly.
5. Take logarithms and differentiate z (x + y) *Jx - y and z = (x + y)ex~v .
6. Find the differential of u - (x 2 + - z z

2xy y )e .
/
7. Show that f(x, y)=f(Q, 0) + .T/a (0,
,
0) -f 2// '(0,
y 0) approximately when
x and y are small.
--
Deduce
1
that, for small values of
-= 1 -x+y
-\-x-y
approximately.
x and y,
Find the difference between these expressions in terms of x and y.

8. Show that - X
x-\-y
= %(x-y + 2) approximately when x and y are each
nearly equal to 1 . What error is involved in the use of this approximate value
when x = and y = 0-9
1-2 ?
9. The volume of a right-circular cone of base radius x and height y is

z
%TTX y. If x is estimated as a with an error of h and y as b with an error of k,
show that the error in the estimate of the volume is %7ra(2bh + ak) approxi-
mately.
10. For the cono of the previous example, show that the proportional
Ax
-- Ait
increase in volume is approximately 2 f-
- - when x and y are increased by
Ax and Ay. Deduce that a 1 per cent, increase in base radius increases volume
by the same percentage amount as a 2 per cent, increase in height.
11. If at is the relation between the pressure (p) volume (v) and tem-
pv 9
perature of a gas, a being a constant, show that the proportional increase

(t)
in volume is approximately equal to the difference between the (small) pro-
portional increases in temperature and pressure. Deduce that a simultaneous
1 per cent, increase in temperature and pressure leave the volume approxi-
mately unaltered.
dz
12. From the expression for dz, find - when z x* + y 2 where x 1 + 1 and -
Cut
y~l-t. Express z as a function of t and check your result by direct

derivation.
13. Evaluate if z ---- , whore x e* and y ~ e~ l *
dt x fy
14. Find -
and ~ if z log (x
- y), where x = - and y .
Express z as a
du d'V u v
function of u and v and check by direct partial derivation.
15. If z -/(#, y), where x ~ a + at and y-b + fit, show that
dz
-^ dz n dz
and
_ d zz
a
9 d zz
+2
_ _ d zz n9 d*z
16. If z =f(x, y), where x au + <*v and y~bu + fiv, show that
dzdz dz dz
dz dz
^'
du
~- = a ---f n
= (i~- +
dv dxdx
b~
,
and
_
.
dy dy
Generalise by finding tho partial derivatives off(x, y, z, ...) as a function of
u and v if x au + <xv, y bu + ftv, z cu + yv, ... .
17. Find --- when z =/(#, y), where x and y are functions of t and show that
dt
dx dy dz
z dz d*x
fdy\*
dt dt dy z \dx/ dx dt z dy dt^
A homogeneous of the rth degree iff(u, v) = \ f(x, 2/)
r
18. function f(x, y) is
where u = Xx and v=--Xy. Differentiate each side with respect to A, put A= 1
and show that
x a +y a
rf( x > y) (Eider's Theorem)
ox oy
aad
19. If '
zf(x,y), where x~Xx and y
Xy vary in a fixed proportion,
f .. dz I
( dz. dz\
, Xdz ,,
f
consider z as a function of A and show that -^ = -?{x~-y~
,
y } If - .. ,
.
-
dX A \ dx dy/ z d\
is the elasticity of z =f(x, y) for proportional variation in x and y show that t
x-
ox y~
+ = ez. If the function is homogeneous of degree r, show that e r
oy
and so deduce Euler's Theorem.
20. Find ~
dx
and -~^ from the implicit function
CLX
(a;
- a)
2
+ (y
- b)
2 = c*.
21.. If -, show that dx

ax 2 + 2hxy + by 2 + 2gx+2fy + c-Q, ~
~= - ax+
hx b
y+g
evaluate the second derivative. + V +J
dy~ d u 2i
22. Find
-y
and -y-^ when a; 3 -f 2/ 3 - 3xy 0. Show that y is a maximum
dx dx
when x = ^/2. Interpret in terms of the graph of the function.
23. If x 3 y + z - Sxy and and
3 3
4- defines z as a function of x y, find
-
.
ox oy
Deduce the second-order partial derivatives of z.
24. Given f(x, y, z)

= 0, find the first and second-order partial derivatives
of z as a function of x and y.
26. If x 2 + y 2 - 2ax -1-1 = defines y as a function of x, a being a constant,
show that 2xy ^ + x* - y 2 - 1 = 0.

dx
26. Show that the tangent line at (x l9 y^ on the curve ax* + by
z
c is
axx l -\-byy 1 ~c, and that the tangent plane at (sc l9 y lt zj on the surface
ax 2 + by 2 + cz 2 d is axxl + byy l + czz^ d.
27. Explain the moaning of each of tho six partial derivatives, of one
variable in terms of another, obtainable from the implicit function /(a;, y, z) 0.
Express each in terms of the partial derivatives of/ and deduce that
dz dy dx_
_
dy dx dz
28. Two relations <f>(x, y, z) = and i/r (x, y, z)
= define y and z as functions
~
of x. Differentiate each relation and obtain expressions for - -
and in terms
of the partial derivatives of and i/j.
the relations xy + yz-z = Q and # y-
z 2
29. Differentiate -f- z to give
~ and ~ . Solve the equations to obtain y and z explicitly in terms of x and
obtain tho derivatives directly.
Economic applications of differentiate

30. If x=f(a,a linear homogeneous production function, show that
b) is
the marginal rate of substitution between the factors depends only on the ratio
:
___
6 a of the factors used. Verify in the particular cases
x = v 2Hab - Aa 2 - Bb 2 and x -
2Hab - Aa* - Kb* - Ca + Db
-- "
31. Show that the property of the previous example is also true of the
quadratic homogeneous production function and illustrate with tho function
x 2Hab - Aa 2 - Bb 2 . Is the property true for any homogeneous production
function, e.g. for x = Aaab& ?
32. If the production function
point (a, 6)
Deduce that the first

_
db -
daT
__
on a constant product curve
Hb-Aa
Ha-Bb
derivative is
is x=2Hab- Aa 2 - Bb
OT-H, _ - (H(Ha-Bb)*'
d 2b
da 2
-AB)x
_ _2
'
negative and the second positive provided

2
,
-
show that at any
that
JL>
a>6> l
a. Deduce that the constant product curves are of normal
form in the area between the lines Hb - Aa = and Ha - Bb = 0.
33. Show that, for the production function x *J2Hab - Aa 2 - Bb 2 9
Hx 2
2
-I.
'(H ~AB)ab
34. Show that the production function x = Aa a b& gives cr 1 for all com-
binations of factors.
35. Show that the elasticity of productivity e (10.8 above) is tho elasticity
of the vertical section of the production surface through tho axis Ox and the
given point on the surface. Use the results of Example 19 above to show that,
for any production function,
dx dx
36. If ,u log(x + a) a (y + b)P is one form of a utility function, find the

marginal rate of substitution between tho goods and Y and deduce that the X
elasticity of substitution is
box + apy
ff x -} .
37. If ax 2 -H by
2= constant is the transformation function for two goods
X and F, show that the marginal rate of substitution of Y production for
X production is =-- and that the elasticity of substitution is always unity.
38. Two
goods X
and Y are produced jointly with the aid of two factors
A and B. Technical conditions define a relation f(x, y, a, b) = 0, so that
(e.g.) the production of Y is determined if tho amounts of the factors used and
the production of X
are given. Show how a and 6 can bo used as parameters
to describe the shifting of a transformation curve between and Y pro- X
duction. What do the partial derivatives ~ ~ and ~ represent
1 * , ?
dx da db
39. If, in the problem of tho previous example, technical conditions give
two production functions x 0(a, b) and y$(a, 6), show that
/da\ fa
\dx/V constant â0&~"^Z>fAfl
and interpret the meaning of this derivative.
40. Of the duopolists of Examples VIII, 45, the first acts (as before) on the
assumption that the second does not change output while the second correctly
estimates the first's output changes consequent upon his own changes. Show
that the total equilibrium output is now approximately 44 sets per week. How
is this output divided?
CHAPTER XIV
PROBLEMS OF MAXIMUM AND MINIMUM VALUES
14.1 Partial stationary values.
IN our discussion of the partial derivatives of a function of two
variables (12.4 above), we postponed consideration of the cases
where one or both of the derivatives assume zero values. Suppose
now that -- and ~- 7^0 for a function z=f(x, y) at a point (a, b).
ux (jy
Then z has a stationary value for variation in x from the value a, y
d 2z
retaining the constant value b. The sign of at (a, b) determines
OX"
whether z is a maximum or a minimum.* The vertical section of the
surface z=f(x, y) which passes through (where x=a and y P b) in
the W.E. direction has a maximum or minimum point at P.
dz
In general,, the condition ^-
= is an implicit relation between
ox
x and y represented by some curve in the plane Oxy. Points on the
surface z=f(x,y) above this curve correspond to maximum or
minimum values of z in the W.E. direction. Some of these points
may correspond to maximum values (satisfying the condition

9 2z \
- < ) and make up what can be called a ridge line of the surface
in the W.E. direction. Others of the points may correspond to
minimum values
(^>o) and define a trough line of the surface in
the W.E. direction. The curve in Oxy giving the ridge and trough
lines of the surface is clearly located on the contour map of the
*
Throughout the present chapter, the possibility that stationary values
can include points of inflexion, though not overlooked, will not be taken into
account.
surface. The tangent gradient of the contour through the point
(x, y) is
dz
dy _ dx
dx~ dz
9
which is zero whenever = 0. The ridge and trough lines of the

(/X
surface thus lie above the locus of points on the contour map where
the tangents are parallel to the axis Ox. For example, in the case
shown in Fig. 87 below, the line AS
across the contour map corre-
sponds to a ridge line of the surface in the W.E. direction.
A similar analysis
J holds if ^-7^0 and
dx
-=0 at any point.
J r The
dy
function then has a partial maximum or minimum value (according
2
d z\
to the sign of -~ for variation of y alone. The ridge and trough
j
lines of the corresponding surface are then obtained in the S.N.

direction. They lie above the points on the contour map where the
tangents to the contours are parallel to Oy. The line CD across the
contour map of Fig. 87, for example, represents a ridge line of the
surface in the S.N. direction.
The analysis extends without difficulty to functions of more than
two variables. In general, if y=f(xl x 2 ... xn ), there are n different
, ,
kinds of partial stationary values, each corresponding to the vanish-

ing of one of the n partial derivatives of the function. Further,
though the analysis is expressed in terms of explicit functions, it
applies equally well to functions written in implicit form.
14.2 Maximum and minimum values of a function of two or more

variables.
If both the partial derivatives of the function zf(x, y) are zero
at the point (a, 6), then z has a stationary value for variation in
x alone and for variation in y alone from (a, b). Further,
dz dz
,
dz=~dx+~dy =Q
dx dy
for any values of dx and dy, i.e. z has a stationary value for any
PROBLEMS OF MAXIMUM AND MINIMUM VALUES 353
variation whatever from the point (a, 6). Such a point is said to
define a stationary value of the function. Ignoring the possibility of
inflexional points, there are three cases to distinguish :
(1) The value of z is a maximum for any variation of x and y from

(a, 6), the surface zf(x, y) having a "peak" at the point
where x = a and y b.
(2) The value of z is a minimum for any variation, the surface

" "
having a hollow with lowest point where x=a and y b.
(3) The value of z is a maximum for some variations and a mini-
mum for other variations from (a, 6), the surface having a
"saddle point" where x=a and y b, i.e. a point where
" " " "
there is a pass over a ridge of the surface.
The three cases are illustrated in Fig. 86.
Fio. 86.
Stationary values, occurring where the tangent plane to the

surface is horizontal (parallel to Oxy), are located on the contour
map of the surface at points where the locus of points with tangents
parallel to Ox ( =0 J
cuts the locus of points with tangents parallel
to Oy ( = OJ. There is a stationary value wherever the ridge or

M.A.
trough line in the W.E. direction cuts the similar line in the S.N.
direction. The lie of the contours around such a point determines
ibhe nature of the stationary value. If the contours tend to close
down to a single point as their height increases or decreases, then the
stationary value is a maximum value of type (1) or a minimum

value of type (2). If a contour crosses itself at the point and there
isno tendency to close down on the point, then the stationary value
occurs at a saddle point of type (3). The cases are fully illustrated
by the particular contour maps of Figs. 87 and 88.
In more rigid analytical terms, the function z =f(x, y) has a
maximum value at (a, b) if z is a maximum for any variation of x
and y from the values a and b respectively. A minimum value of z
isdefined similarly. Maximum and minimum values together make
up the extreme values of the function. It is assumed that the function
and its partial derivatives are finite and continuous.* A maximum
(minimum) value can occur only if the function is a maximum
(minimum) in each of the two fundamental directions in which x
and y vary alone, i.e. only if the two partial derivatives are separ-
ately zero and the function has a stationary value. To distinguish
maximum from minimum values, or either from other stationary
3 2z 3 2z
values, is a matter of some difficulty. If and are both negative
at a stationary point, then z must be a maximum in each of the

fundamental directions and it is likely that z is a maximum in all
d 2z d 2z
directions. Similarly, if ^-r and - -
are both positive at such a point,
ox* oy*
then have a minimum value for all directions of varia-
z is likely to
tion. But these by no means certain. We must exclude

results are
the possibility that z behaves differently in some directions as com-
pared with the fundamental directions. It is shown below (14.7,
Ex. 3 and 19.1) that
dxdy
is a condition sufficient to exclude this possibility. Hence :
* As in the one-variable
case, a maximum or minimum value of z f(x y) t
can occur at a point where the partial derivatives are not continuous. Such
a point is not shown by the criterion below.
CRITERION FOR MAXIMUM AND MINIMUM VALUES

(1) All maximum and minimum values of z=f(x, y) occur where
dz = Q for all variations of x and y, i.e. where
*?=*? =
dx By
*
/o\
(2) If ^
d z
and
*
^ d z are
&
*
negative and u 4.u
A such that*
a *2 a2 *
> f^ 1
d *z Y
1
*.
at
2 2 2 2
dx dy dx dy \dxdy /
a point # a, j/
= 6, then /(a, 6) is a maximum value of the
d 2z d 2z
function. If and -= n
2
are positive subject to the same
ox 2 dy
inequality at a point xa, y = b, then /(a, 6) is a minimum
value of the function.
The a necessary condition. The second condition when added

first is
to the first is a sufficient condition and defines situations in which
only a maximum (or only a minimum) value can occur. But the
condition is not complete it is not necessary and sufficient.
; It
d 2z d 2z d 2z 2
is possible to show that, whenever ^ 2 ^-r
dx dy 2
< I --^- \
(
\dxdyj
1
, then the
stationary point concerned is a saddle point. But the case where

B 2zd 2 z I d 2z \ 2
- -
is still open
* and the stationary
J value concerned
dx 2 dy 2 \dxdy/
may be a maximum or minimum value or it may not.
Thecriterion holds, in exactly the same form, for functions given
implicitly. If f(x, y, defines z as a function of x and y, we have

z)Q
only to write
-- , and
ctliU -~-^ ~jT
fa Jz dy /,
and the criterion applies. The necessary condition for a maximum
or minimum reduces to fx ==/ = 0.
value of z
The criterion of maximum and minimum values can be extended

to functions of more than two variables, but only as regards the
necessary condition at this stage. All the maximum and minimum
values of y=f(xl9 # 2 ... #J occur at points where dy =
,
for all
variations in the independent variables, i.e. where
dy^^dy^ '" = dy_ =Q

dxt dx 2 dxn
Similarly, if f(xl9 x 2 ,
... xn> y) =0 defines y as a function of the other
variables, then the maximum and minimum values of y occur where
A=/a%= =/n = 0-
Sufficient conditions are given later (19.1).
An
important feature of the results here given is that, since the
differential of a maximised or minimised variable is zero at the point
concerned, we can treat such a variable as a constant in the process

of differentiation. Suppose that, in the relation
the variable y is to be maximised for variation in xl9 # 2 ,

... xn the
,
variable z remaining fixed. Differentiating and putting dy = (since

y is maximised) and dz = (since z is fixed), we have
for any variations in the variables xly x 2 ,

... xn . So
Exactly the same conditions obtain if z is maximised while y is

held constant. There is no distinction, in differentiating, between
variables which are maximised (or minimised) and variables which
are held constant.
14.3Examples of maximum and minimum values.

Maximum and minimum values of a function z =f(x, y) are
located in practice as follows. The equations ~ = ~~ = are solved
for x and y and each solution examined in turn. In the majority of

cases, the sufficient conditions (involving the second-order partial
derivatives) enable us to distinguish maximum and minimum
values. In some cases where the conditions fail, we can examine the
contour map of the function in the neighbourhood of the point
concerned. And, even if the sufficient conditions apply, the contour
map provides a useful illustration of the situation. The following
examples illustrate the method :
Ex. 1. z=x2 + y*.

Here ;r- = 2# and ^-=2y.
dx dy
There is one stationary value (2=0) occurring where x=y=0 and this
value must be a minimum since all other values of z are positive. The
sufficient conditions are found to support this, since
3-2=^ =2 and a-|-=0.

ox*oy ox oy
The contour map of the function has trough lines along Ox and Oy and
is shown in Fig. 76 above.
Ex. 2. z=2x+y-x 2 +xy-y2 .
Here ~- = -2x+y* + 2 and ~-=x-

ox oy
and stationary values of z occur where
2x-y-2=Q and x-
There is a single stationary value (z=f), where #=^f and y = J.
Suice - - 2, - 1, and ,
^-5a = - 2,
dx*
the sufficient conditions show that z =-J is a maximum value. The contour
FIG. 87.
map of the function is shown in Fig. 87, rising contours closing down
on the point where x =f y =| The line AB (2x - y - 2 =0) is a ridge line
,
.
in the W.E. direction and the line CD (x- 2^ + 1=0) is a ridge line in
the S.N. direction.
Ex. 3. z=4x2 -xy+y 2 -x 3 .
TT Sz n _ dz
f-4 fvf*o
licit* =-- X'**
&jc /- ox
y-~ -%i" *
,
.
TT~
...
= 2(4-3*); -1 and S-2.
Stationary values of z are given by
'3x
z
-8x+y = Q and #=2z/.
On solving (e.g. by eliminating y) we find two stationary values, 2=0
y
where # = 2/ = 0, and 2= Jj^ where #=f, 2/=f. The first is a minimum

value of z since, when x ==y =0, we have
and -?i ^ -x
.)'='
But when ic=f and y =f the two direct partial derivatives of the second
. y^
5--
Fia. 88.
order are opposite in sign ( - 7 and 2) and the stationary value here occurs
at a saddle point. Fig. 88 shows certain curves of the contour map of the
parabola y = 8# 3x ) represents the

function. The curve ABC - 2
(the
ridge and trough lines of the surface in the W.E. direction. The part
f\y 7$"y \
#<o and is ^ ne line and the
(4. ,
^- '~r\>v) trough part
4.
;: ,
/}-?
~=0 and
^52^
-<
\
is the ridg
p.
Une - The
32
line
\
^ fc
= 2y) is
the trough line in the S.N. direction (;r-=0 and ^-i>0)

2
. The lie of
\dy dy /
the contours about the two points of intersection of the line and the
parabola show that thereis a minimum value of z at the origin and a
saddle point at x =f y =f ,
.
Ex. 4.
Taking z as a function of x and y, we find
dz
~ x -I dz
"" y
The only stationary values of z, for variation of x and y, occur at x 1,

y =0. On substituting in the original equation, two values of z are found,
z - 1 and z = J-. Further, we find
dx* (82 + 1) 2
when x = 1, y = 0. Similarly, at the same point,
=0 and ~ 5
= - -
So, when z = - 1, the direct second-order partial derivatives are positive

and the cross derivative is zero. By the sufficient conditions, z = - 1 is a
minimum value of 2. In the same way, z J isfound to be a maximum
value of z. The solution can be illustrated by a contour map of this
double- valued function (see Examples XIV, 9).
14.4 Monopoly and joint production.

A monopolist produces amounts x l and x 2 of two goods X1 and X 2
at a total cost of
IJ=F(xlt The two goods are related in con-
x 2 ).
sumption and the demands of the market are Xi = <h(pi,pt) and
#2 = ^2(^,^2), where p l and p 2 are the prices charged. The net

revenue of the monopolist is
y=x pl +x 2 p 2 -II.
1
Making use of the demand relations, this is a function of the two

prices. It is taken that the monopolist fixes prices to maximise net
revenue. The necessary condition for this is
i.e.
... sn\ dx,
/ a/i\ a^ + / a/7\ a# 2 ^
and x +l p1 }.* Ip J
-
=0.
a
r a^!/ dp 2 r a# 2 /) a^ 2
There are two relations to determine the prices. Subject to the

prices obtained satisfying the conditions for a maximum (rather than
a minimum) value of y, the problem is solved.
To illustrate the solution, a simple case can be considered in
which the cost and demand functions are linear :
TTâ^ + aoO^ ;
'
xi = al ~~a n p l -a l2 p 2 and
X 2 =a 2 -a12 pl -a 22 p 2 .
The average costs of producing the goods can be taken as constants,

<%! and
a 2 respectively. Denote
" "
i.e. the demands when the competitive prices (equal to average
costs) are charged. Then
x = x10 - an (P! - ai - a12 (p 2 - a 2
) )
and x 2 =x 2Q - al2 (pI - ai - a 22 (p 2 - a a

) )
and the monopolist's net revenue is
- a a ).
=a:10 - 20H (pl - ai ) - 2a12 (# 2 - a 2 )
and 2a 12 (ft - ax) - 2a 22 (p 2 - a 2 ).

-^L^XM-
Putting these expressions equal to zero for maximum net revenue and
solving, we find the monopoly prices
Further, ~= 2
- 2au ;
- ~- = - 2a12 and ~= 2
- 2a 22 .
If On and 22 are positive and such that i2

2
<a ii a 22> ên *ê prices
give a maximum value of net revenue. These limitations on the

values of the constants are assumed below.
Suppose, now, that two independent monopolists produce : and X
X 2 with costs J^
= a^
and IJ2 a 2# 2 each assuming that the other
,
willnot change his price and maximising net revenue. The net
revenue of the first monopolist is
where the demand gives x as before. For a given value of p 2) the
maximum net revenue is obtained when ~=0, i.e. when

dpi
x1Q - 2au (pi - ax ) - a12 (p 2 - a 2 ) = 0.
Similarly, for the second monopolist,
#20
~ i2 (Pi
- i)
~ 2a 22 (p 2 - a 2 ) = 0.
Solving these two equations, the prices charged by the independent

monopolists are found to be
~
Since ~^\ -2an <0 and J for -r-^f,

similarly *
the prices correspond
to positions of maximum net revenue.

If Ojg is negative, so that the goods are (in the broad sense) com-
petitive in consumption, it is seen that the joint monopoly prices are

greater than the separate monopoly prices ;
the effect of joint mono-
polistic production is to restrict output and raise price. If a^ 2 is
and the goods are complementary, there are two cases *

positive :
(1) ii<aii~ s and
Here, for the joint monopoly prices, (PI-&I) and (p 2 -^) are
positive and both prices are above cost as represented by o^ and a 2 .
It may or may not be that both prices are above those of separate
monopoly. This case holds when, as in the majority of situations,
* Notice that the case where a la >n #10 and o 12 < 22

#20
* s exactly parallel
and that impossible, since a 12 < a u a 2 2
a
with and adds nothing to case (2) it is
for o l8 to be greater than both o n

" and a 22 .
#10 #20
the relation between the goods (as shown by a12 ) is weak as compared
with the direct price effects on demand (as shown by au and a 22 ).
(2) ii*ii and
Here, (pl - ax ) is negative while (p a - 2 ) is positive, i.e. the good 1 X

is sold by the joint monopolist at a price below cost a x and certainly
,
below the separate monopoly price. It pays the monopolist now to

" "
use one good as a loss leader in order to promote the sales of the
second good. The conditions for the case are (roughly) that the goods
are strongly complementary (a12 large) and that the demand for the
" "
loss leader is sensitive to changes in its price (a n large). For
example, a monopolist manufacturer of razors and blades may gain
by marking down the price of razors (for which the demand is
elastic) below cost to promote the sales of blades (for which the
demand is less elastic and strongly complementary with that for
razors).*
14.5 Production, capital and interest.
To generalise the problem of 10.3 above, suppose that a variable

time elapses between the beginning of a production process and the
final output of a good X
produced with two factors A and B. The
amount of output x depends, according to technical conditions, on
the amounts a and b of the factors used and on the length t of the
period of production :
x=f(a, 6, t).
It isassumed that the prices of the factors are given as p a and p b and
the market rate of interest as lOOr per cent, per year compounded
continuously, all taken (for convenience) in terms of the product x.
Further, taken
it is that the factors are paid off at the beginning of
the production. So
xe~ rt
* (a ' M) (say)
isthe product discounted to the beginning of production as a ratio

of the cost then incurred. It is assumed that the entrepreneur is a
* For an " "
analysis of the loss leader problem, see Roos, Dynamic Eco-
nomics (1934), pp. 128-47. On problems of joint production, see also Stackel-
berg, Grundlagen einer reinen Koatentheorie (1932) and Hicks, The Theory
of Monopoly, Econometrica, 1935.
monopolist arranging production to maximise y. The variables to

be chosen are a, b and t and we must have
^ = ^
da db dt
log y = log x
Since - rt - + bp b
log (ap a ),
, 1 dy
y 1 dx pa
^
we have '
y da x da ap a -f bp b
1 dy __ 1 dx pb
ydb~x db ap a
1 dy
__ Idx
_- ____
y.
p
2/3^ a; dt
dx Idx
- ~ and r
* ai' x 36 ~*a*
are the necessary conditions for a maximum value of y. These
equations, with xf(a b, t), determine the equilibrium values of
y
a, 6, t and x as dependent on the given values of p a p b and r. ,
Suppose now that there is pure competition among entrepreneurs

so that the market rate of interest adjusts itself to make (discounted)
receipts just cover cost. The given value of r is such that the
maximised value of y is unity and
=e ~ rt
So P a = e~ da
* T-
rt
and ^
Pt> ^-
db
Production is extended until the discounted marginal products of the

factors are equal to the fixed prices of the factors.
For example, suppose that the production function is
where A is' constant and a and j3

are fixed positive fractions. For a
fixed period of production and variable factors, there are constant
returns to scale. The necessary conditions for monopoly equilibrium
are
Pa =^--^, ^ = _,__ (1 and r=-,

Idx dx -a 1 dx 8
= a-, 1 1 ,
since -
-^1 = 7 and ---='-.
x da a x db b x dt t
B
The period of production t=- is here determined solely by the rate
of interest and not by the prices of the factors. The first two con-
ditions are equivalent and imply that (1 -<x.)ap a equals
.e.
61 -<xp a
.
'
Only the ratio of the amounts of the factors used is determined ;

the
actual scale of production is immaterial (constant returns to scale).
The ratio, further, is fixed by the prices of the factors and not by
the rate of interest. Notice also that
~ --
- constant,
1 a
i.e.the relative share of the two factors in the product is constant,
unaffected by changes in their prices or in the interest rate.
Under competition, with the particular production function now
assumed, we have the added condition
i.e.the combined share of the factors in the product is constant,

independent of their prices or the interest rate. The share of capital
(as represented by the entrepreneurs) is then constant also. In the
competitive case, in fact, the relative shares of the factors and of
capital in the total product are constant, no matter how the prices
of the factors or the interest rate change.*
14.6 Relative maximum and minimum values.

A different problem, that of relative maximum and minimum
values, arises when we seek the maximum or minimum values of a
function zf(x, y) where the variables are not independent but
related by some given relation <f>(x, 2/)
= 0. The given relation is
often termed a side relation. The side relation determines one

variable (say y) as a function of the other. Substituting,
z=f(Xyy) = *l*(x) (say)

and our problem effectively one of maximising or minimising a
is
function of one variable. In simple cases, the side relation gives y

*
Compare the results here obtained with those given by Edelberg, An
econometric model of production and distribution, Econometrica, 1936.
as a single-valued function of x, in which case $(x) also appears in
single-valued form and our problem is solved without difficulty.

In general, the side relation is multi-valued and a more subtle
method of attack is needed. But the essential idea is still the re-
duction of z to dependence on one variable only. Taking x as the
independent variable and assuming that all partial derivatives are
finite and continuous, the function and side relation give
and
S =/ +/.|
,
^,|.0.
=-/-/-
Equating this expression to zero, we obtain :
The function z=f(x, y) has maximum or minimum values

relative to the side relation <f> (x, y) = only at points where
.
= .
</>x <f>y
This a necessary condition for maximum or minimum values.

is
Adding the side relation, there are two equations to be solved for x
and y. Each solution is a possible maximum or minimum situation
and must be examined further. We find
_
dx*~dx
on reduction. The sign of this expression at a point given as a

solution of the necessary condition determines whether the point
corresponds to a maximum or minimum value of z.

In diagrammatic terms, instead of locating the peaks and hollows
of the surface z=f(x, y), we consider only a definite path on the sur-
face, defined by the side relation </> (x, y)
= 0, and determine where
the path attains highest or lowest height above the zero horizontal
its
plane Oxy. Using the contour map of the surface, our path is a
given curve crossing the contours of the map and we seek those
points on it where the highest or lowest contours are cut. It is clear
that, at these points, the given curve touches a contour (see Fig. 89
below). The tangent gradient of the given curve <f>(x, y)
= is then
equal to that of the contour f(x, y) = constant, i.e.

( -7^) equals
f f \
\""r)
*
which is the necessary condition above.
Jv'
For functions of several variables, we seek the maximum or
minimum values ofy=f(xl9 x% xn ) subject to a given side relation
9
...
^(#i> #2> x n)
= Q- The side relation gives x1 as a function of the
-
other (n 1) variables for which
Then y is a function of (n - 1) independent variables and
and similar expressions for the other partial derivatives. At a

maximum minimum position, all these partial derivatives
or must be
zero. It follows that
Jx^ __ J_xj_ __ _JXn

^x, <*2 i>x n
isthe necessary condition for a maximum or minimum value of y.

There are (n - 1) equations here and these, with the side relation,
determine certain sets of values of xl9 # 2 ,
... xn amongst which are all
those giving maximum or minimum values of y. Sufficient con-

ditions for distinguishing maximum and minimum values are given
later (19.2).
In a more general problem, the maximum
minimum values of y or
are required subject to r given side relations between the variables
(where r <n). An alternative method of approach can now be used
to advantage.* In the case of two side relations
<f>(xl9
x2 ,
...xn ) = Q and $(xl9 x 2) ... xn ) = 0,
write z =f(xl x* ...,
xn ) - X<f> (xl9 x^ ... xn ) - pf* (xl9 x29 ... xn ).
Then z must have a maximum or minimum value for any values of
A and fi whatever (since the coefficients of A and p, are zero). But
* The
method, which also applies in the simple case (r= 1) already analysed,
is known as Lagrange's method of solution by " undetermined " multipliers.
See, for example, Osgood, Advanced Calculus (1925), pp. 180 et seq. and de la
Vallee Poussin, Cours d'analyse infinitesimal, Vol. I (5th Ed. 1925), pp. 147-9.
Take the two side relations as giving xl and # 2 as functions of the

other (n - 2) variables which are independent. Then we can choose
A and u, so that the expressions for -

and - are zero. For a
dxl dx 2
maximum or minimum value of z as a function of the (n - 2) inde-
pendent variables, all the partial derivatives , ,

. . . must be
OXg OX^
zero. In this way, we have altogether n equations :
These, together with the two side relations, determine certain values
of A and /z and of the n variables xl9 x 2 ... x n Amongst these values
,
.
are those which give the relative maximum or minimum values of y
required. Sufficient conditions cannot be written here.
14.7 Examples of relative maximum and minimum values.

The following examples illustrate the practical solution of relative
maximum or minimum problems. It is important to note, in each
case, which variables are taken as independent.
Ex. 1. To find extreme values of z = x2 + y 2 relative to
From f(x,y)=x 2 +y 2 we have fx ~2x and f y =*2y. Again, from

,
4(x 2/)=z + 2/ -42;--2,v + 4, we have ^. = 2 (a -2) and v = 2(y -1). At

2 2
9 <f>
a relative maximum or minimum value of z, we must have

x y
x-2~y-l'
Hence, x = 2y. Substituting in the side relation,
5y
2 - l(ty + 4=0.
So ,-li and *=
The corresponding values of z are
z = x 2 +2/2 = 5l - = 2(3
By (examining the sign of the second derivative of z as a function of x,

it can be shown that the plus sign above corresponds to a relative maxi-
mum and the minus sign to a relative minimum value of z. Or, Fig. 89
shows certain contours of the surface 2
zx
+y2 (i.e. circles centred at 0)
and the curve representing the side relation (a circle centred at a: = 2,
y = l). It is seen that this latter circle touches two of the contours. For
one point of contact the value of z is a relative minimum
2=6- 2^/5 1*53 approximately,
which is the square of the radius of the smaller contour touched. For
the other point of contact
2 = 6+ 2^5 = 10-47 approximately,
the square of the radius of the larger circle touched.
,#=16.0
Ex. 2. To find the dimensions of the rectangular block of maximum

volume which can be cut from a sphere of radius a inches.
Let the sides of the block be 2x, 2y and 2z inches. Simple geometrical
considerations show that, for maximum volume, the corners of the block
must lie on the sphere's surface with diagonals as diameters of the sphere.
Hence, # +y +z =a
2 2 2 2
We require the maximum volume V = 8xyz
.
subject to this relation.

From f(x, y, z) = Sxyz, we have fx = 8yz / = 8xz fn = 8xy. ; ;
From <f>(x, y, z) ^x2 +y2 +z 2 -a 2 we have x = 2x y>

= 2y </> ; cf> ; <f> z
= 22.
Hence, for the relative maximum of F,
4yz 4xz 4xy

x y z
It follows that x2 =y = z 2
2
.
Since x 2
+y 2 +z 2
a 2 and x, y and z are positive, we find
a
These values clearly correspond to a (relative) maximum volume
The block of maximum volume is a cube of side -

,~ inches.
v*
Ex. 3. The function z =/(#, y) has a stationary value at (a, 6), so that
fx ^fy =0 at this point. The relation
y-b=m(x-a),
where m is a fixed value, indicates a particular direction of variation of
x and y from (a, 6). To find the nature of the stationary value of z in this
definite direction :
The side relation gives y as a function of the independent variable x

and we find
-jL
= mj (from the side relation)
f
-Jyy (
\
m + 7~
(.i-J xv \
Jyy/
+ fxxjyy
)
.
f
--
Jyy
~
Jxy
(from the original function).

The sign of this last expression, for the given value of m, determines
whether z is a maximum or minimum.
Iffxx an(i/vy are both negative and such that
JxxJyy^'Jxy
the expression shows that is negative whatever the value of m, i.e. z

,^
ax
has a maximum value in all directions from (a, 6). These are sufficient
conditions that /(a, b) is an (unrestricted) maximum value of the function.
Iffxx smdfyy are both positive subject to the same condition, it follows
similarly that /(a, 6) is a minimum value of the function.
14.8The demand for factors of production.

The output of a good X when amounts a and 6 of two factors A
and B are used is given by the production function x=f(a, b). It is
assumed that the prices of the factors are fixed at p a and p b and that
2A M.A.
a given output x is required at minimum cost. We choose a and 6
to minimise II~ap a + bp b subject to /(a, b) = x. Taking a as the
independent variable, the side relation gives
^=_/.
da /'
So _ =J
dU
,
(|+1 , t
_ =1 ,.- f ft= o
db a.
/
is the necessary condition for minimum cost,
i.e. *=&.
Ja Jb
The factors are employed in amounts such that their marginal
products are proportional to their given prices. Further,

2
cZ /7
So, if y->0, then the position concerned corresponds to a minimum

cLa
cost. The condition implies that the constant product curve at the
point concerned is convex to the origin. The equilibrium position
is thus stable at all points where the constant product curves are
convex to the origin.

The necessary condition with the side relation determines the
demands a and 6 for the factors in terms of the given values of
p a p b and x. The (minimised) cost is then determined in the same
,
way. As the given output x is changed, the prices of the factors

remaining constant, the demands for the factors vary and the cost
of production varies to give the. ordinary total cost function.
In diagrammatic terms, the contour map of the function
consists of a set of parallel straight lines in the plane Gab. The
contour for fixed cost 771 is a line with gradient equal to f - -- and
J
77 TI1
^
cutting the axes at distances and from 0. As /I, is increased
P* P*
the line moves away from If a given product x
parallel to itself.
is to be obtained, then all
possible combinations (a, 6) of factors lie
on the corresponding constant product curve. We seek that com-
" "
bination with minimum cost, i.e. lying on the cost line of the
parallel set which is nearest 0. The combinationgiven by the is
co-ordinates of the point P where the constant product curve

touches one of the cost lines. If the constant product curve is of
normal form (downward sloping and convex to 0), there is only one
such point P and it corresponds to minimum cost. At P, the
--- and that of the
gradient of the constant product curve f 1 cost
(\-
)
are equal.
As the value of x is changed,

This is the necessary condition above.
we obtain a series of different

constant product curves and a
series of points P P
l9 2 , ... where
they touch cost lines
AÂ^,....
This is shown in Fig. 90 where the
values of x(xly x 2 ,
... )
increase.
As x changes continuously, the
points describe a curve in the
plane Oab. The varying co- A4 a
ordinates of a point on the FIG. 90.
curve show the variation pf
demand for the factors as the output x varies. Further, the varying
cost corresponding to the cost lines touched defines the total cost
function for varying output x.
If the production function is linear and homogeneous (constant
returns to scale), then Euler's Theorem gives
So
XX
where A denotes the common value of p a :fa and p b :fb
X
. Now
da
te =p <
dx
da db ,
since &omf(a,b)=x.
'j^+fojZ^ 1
We consider here a variable equilibrium position with output x

changing and p a and p b fixed. Hence, average and marginal cost are
equal at all outputs. Average cost must thus be constant irrespective
of output (see Examples XVI, 22, below), a natural result in this
case of constant returns to scale. In Fig. 90, the constant product
curves are now of the same shape (radial projections of each other)
and the locus of P is a straight line. The factors are used in the same
proportion and their amounts and total cost increase in proportion
as output increases.
If the good X, produced under constant returns to scale, is sold on
a competitive market at a price p equal to the constant average
cost, then
So Pa=P/a and p b =pfb .
" "
This is the law of
marginal productivity the price of a factor ;
equals the marginal product of the factor valued at the selling price
of the product. Suppose, further, that the demand law for is X
x = <f>(p) with elasticity ?j
=-*- . To determine the amounts
(a and of the factors and the selling price (p) in terms of given
6)
factor prices (p a and p b ), we have now three conditions :
(!)/(, 6) = *(!>); V)pa=pfa', (3)p=p/-

The output x-f(a, b) </>(p) and total cost IIpx=ap a + bp b are
also determined. Full competitive equilibrium, with constant
returns to scale, is thus determinate.
We
can trace the variation of the demands for the factors of pro-
duction, given above in terms of p a and p b , as one of the prices
(say p a ) changes while the other (p b ) remains constant. For the
linear homogeneous function x=f(a, 6), we have
fbb = - fab
and -
faa=~-a fa (12.8, (4), above).
The between the factors cr= -

elasticity of substitution is
*^J db
The equations (1), (2) and
hold for any values of p a and
(3) p b and
can be differentiated partially with respect to p a :
da db dp x dp
Making use of (2), (3) and (4), we have :
a *
db
and .
By simple algebraic methods,* the values of these partial
derivatives can be found and written :
dp a
da
- --
rr
a (ap a
Pa \ XP
I-n - H--
bp
XP
b
(7
\
I
'
A
and
.
56
^
= --
a6,
(rj
- N
a).
dp a xp^'
TTT Ea =~- Eb p a db
p a da and -= =V ^
,
Write ^= ^- * or ^he elasticities of demand for
Ep a a dp a Ep a b dp a
the factors with respect to p a and Ka = and /c 6
-^(/<- + /c 6 = l)
xp xp
for the proportions of total receipts going to the factors. Then
Ea
Eb
and
The interpretation of these results is clear. If the price of the
* For example, by eliminating ~- we obtain two equations in

,
- and -
dPa dPa
" " tya
which are easily solved. Or, the method of solution by determinants can
be utilised (see 18.6, Ex. 2, below).
factor A rises,
the demands for both factors are affected, and in two
ways. In the first place, the cost of production is increased, the

product is dearer and (for a decreasing demand law with 77 positive)
less of the product is bought. There is then a proportional decrease
in the demand for both factors. This is shown by the negative term
(
- K arj) in the expressions for both elasticities of demand. Secondly,
the factor B is now cheaper relatively to the factor A and it pays to
substitute B for A in production as far as possible. The demand for
B thus increases at the expense of that for A, as is shown by the
positive term (/c a) in the expression for one elasticity and the
-
negative term ( fc 6 a) in the other. The total effect on demand is
found by addition. The demand for A falls in any case, but that for
B may rise or fall according as the substitution effect is stronger or
weaker than the effect through the demand for the product.
14.9 The demand for consumers' goods and for loans.
A
similar analysis applies to the demand of a consumer for two
goods andX Y
obtainable at given market prices p x and p y The .
consumer has a fixed income p and makes his purchases to attain the
highest position on his preference scale, as represented by an index-
function of utility u= <f>(x, y).
We have to find x and y, the individual's purchases, so that u is a

maximum subject to the relation expressing the balance of the indi-
vidual's budget
%p x + yp y = /*.
:
This expression of the problem is independent of which particular

form of the index-function of utility is selected all forms of the ;
index-function increase and decrease together and so attain their

maximum values together, at the same values of x and y.
The side relation gives y as a function of x for which
~ _
py
'
dx~ Pv
so =h + +*=h-jl+'
d*u
and - I +- + + . .
dy\
+ ( ,
.
dy\ dy
xPv + <I> XX P V Z ).
The necessary condition for a relative maximum value of u is thus
^* = ^v
Py' Px
i.e. the marginal rate of substitution between the goods in consump-
tion (R = x (f> y ) is equal to the ratio of the given market prices.

:
<f>
This condition, together with the side relation, is sufficient to deter-

mine the purchases of the individual in terms of the given values of
p x> p y and /x, i.e. to determine the demand functions of the individual
(see 11.7 above). A sufficient condition for a relative maximum

value of u is
() yy px - <> xv p xp y + rxPy < .
Using the necessary condition, this implies that

fad* - 2Â0. + 0*A 2 >.
In diagrammatic terms, the purchases of the individual are given

by the co-ordinates of the point P in the Oxy plane where the fixed
line with equation xpx + ypy p*, expressing the budget balance,
touches one of the indifference curves of the system
(f)(x, y)~ constant.

The point P corresponds to a maximum value of u if the indifference
curve at P is convex to the origin. The condition of tangency is
equivalent to the necessary condition and the condition of convexity
" "
to the sufficient condition written above. So, in the normal case
where the indifference curves are downward sloping and convex to
the origin, a unique and maximal pair of individual purchases are
" "
defined for any given market prices or given income. The normal
case, in fact, corresponds to the case of stable demand functions.
The variation of the individual's demands
two goods as one for the
of the market prices or the income varies is shown by the changing
position of the point P as the line xp x + yp ~p< moves in conformity
}J
with price or income variation.*

We can consider, as a second problem, the conditions under which
an individual regulates the flow of his income, and hence of his con-
sumption, over time. This is a problem of importance in the theory
* For more
complete accounts of the problem of demand for consumers'
goods, see Hicks and Allen, A Reconsideration of the Theory of Value, Kco-
iiomica, 1934, and Allen and Bowley, Family Expenditure (1935). Expressions
are heie found for the elasticities of demand with respect to price or income.
The methods used are extensions of those first introduced by Pareto,
of interest.* assumed that the individual has given resources
It is
which can produce various incomes x this year and various incomes
y next year subject always to a given transformation relation
F(z,y)=0.
Interest is reckoned at the fixed rate of lOOr per cent, compounded
yearly. We determine, first, those incomes which maximise the
present value y
1+r
subject to the transformation relation. We have
^ = 1+JL^ and #2 = J
dx 1+rdx dx 2 l+rdx 2
where the derivatives of y with respect to x are given by F(x, y) =
Hence, for maximum present value, we must have
i.e. the incomes in the two years are such that the marginal rate of
return over cost equals the given rate of interest. The sufficient
condition for a maximum present value ~7<o *s satisfied in the
" "
normal case where the transformation curve in the plane Oxy is
concave to the origin (see 9.7 above).

The necessary condition above and the transformation relation
determine the incomes X Q and y Q with maximum present value, both
incomes depending on the interest rate r. It is now assumed that
the individual can modify his incomes or lending
by borrowing
money this year to be repaid next year with interest at the fixed
market rate. This process leaves the present value of income un-
i.e. y
changed, v
where x and y are any incomes as modified from X Q and y Q by loans.

For example, if the individual borrows (x - # ) this year, then
is the amount he must repay next year, and this is equivalent to the
* The based on that given by Fisher, The Theory of
analysis here given is
Interest (1930).
relation written above. It is assumed, further, that the loans made

by the individual are such that the utility function
w /
ij* \** j yi
of the individual for incomes in two years is maximised subject to
the relation above. The particular form of the utility function
taken is quite immaterial. The relation gives
dy^
dx
So \b x + 0i/ ~
The necessary condition for relative maximum utility is thus
,-*_! '
*,
i.e.loans must be made so that the marginal rate of time-preference
of the individual becomes equal to the market rate of interest. This
condition, together with the constant present value relation, deter-
mines the incomes x and y in terms of the given interest rate. In
particular, (x-x )
is deter-
mined as a function of r, the
demand function of the in-
dividual for loans. The value
of (x x ) may be positive or
negative according to the
interest rate, i.e. the individ-
ual may borrow or lend this
"
year to reach his optimum"
income stream. As in the
case ofthe consumption goods
problem, if the indifference

= constant, are Q A'
curves, \jj (x, y)
FIG. 91.
convex to the origin at all
points, then the positions obtained are stable and correspond always
to maximum, rather than to minimum, values of utility.
The problem can be put in a slightly different way capable of
simple diagrammatic representation. Suppose that the individual,
' '
before borrowing or lending, has incomes x and y in the two years.
Represent these incomes by a point P '

in the plane Oxy (Fig. 91).
By borrowing or lending, the individual can modify his incomes to
x and y where the point (x, y) must lie on the line A'B' with equation
1"' + r
'
I) 1 J
1+7-
i.e. the line through P '
with fixed gradient (to Ox) which is negative
and numerically equal to (1-f-r). The incomes selected by the
individual are given by the co-ordinates of the point P' where A'B'
touches an indifference curve of the system
iff (x, y)
= constant.
If the indifference curves are everywhere convex to the origin, then
P is uniquely determined and corresponds to a maximum of utility.
f
' '
The incomes X Q and y Q can be varied according to the given
initial
transformation relation F (x, y)=0 and the point P can take any
'
position on the corresponding transformation curve QR. As P '
varies, so does the point P' and we require that position of P' giving
the largest possible utility level. The position required is given by
P where an indifference curve is touched by the line (parallel to AB
A'B') which touches the transformation curve at some point P .
In other words, we seek a with fixed gradient -(l+r) which

line
touches the transformation curve at P and an indifference curve
at P. The incomes first derived by the individual are given by the
co-ordinates of P and the
incomes, as modified by loans, are
final
given by the co-ordinates of P. In this position, the numerical

gradient of AB
(1 -f r) is equal to that of the transformation curve
(F x F y):and to that of the indifference curve (ifj x y ), i.e. r equals :

ifj
the marginal rate of return over cost and the marginal rate of time-
preference. Notice that, for a line with fixed negative gradient
-(1 the present value of the incomes represented is constant.
-t-r),
The maximum value of this present value attainable is given by the

particular line AB. Hence, as in the previous analysis, we can first
locate P for maximum present value of incomes on the transfor-
mation curve, and then locate P for maximum utility of incomes
obtainable by borrowing or lending from those at P .
EXAMPLES XIV
General maximum and minimum problems
the contour, ridge and trough lines of each of the functions
1. Indicate
z = x*-\-y 2 -z=l-x 2 -y 2 z = x*-y 2 +l. Show that x y = Q gives a
1 ; ;
stationary value of z in each case. Determine the nature of the stationary

value in the various cases.
2. Show that z ax 2 + by 2 + c always has a single stationary value which
is a minimum, a maximum or a saddle point according as a and b are positive,
negative or of opposite signs.
3. Show that z = x + xy + y has a minimum x 2 + xy - y 2 a
2 2
value and z
saddle point at x y 0.
4. Show that z = ax 2 + 2hxy + by 2 + 2gx + 2/?/ + c has two straight ridge or

trough lines and deduce that the function has only one stationary value.
5. If z = (x - y)ex+v show , that there are ridge and trough lines which are
straight and parallel and that z has no stationary value.
6. Show that z = (x + y )ex*-y* has a minimum value at x = y =

2 2
and
saddle points where x 0, y 1 .
7. Indicate the form of the contour map of z x 3 -f y 3 - 3xy and show that
the ridge and trough linos are arcs of parabolas. Deduce that z has a minimum
value at x = y = 1 and a saddle point at x = y 0.
8. Locate the stationary values of z 2 (x - y) 2 - x* - y*.
9. If z is a function of x and y denned by x + y + 3z 2 - 2x

2 2
2z = 0, show -I-
that the contours arc circles centred at x = 1, y = which expand as z increases

- l<z< --J-) and then contract as z increases further (
( -i<z<i). Deduce
that z - 1 is the minimum and z~ the maximum value of z. (See 14.3,
Ex. 4 above.)
Show that the only stationary points on the surface z x 2e v form a line
10.
of minima lying along the axis Oy. Verify by examining the nature of the
sections of the surface by planes perpendicular to Oy.
11. If (ax + by + c)<f> (x, y) and - = (ax + by + c)i/t (x, y), show that z has
ox c)y
a line of stationary values corresponding to ax + by + c on the contour map
in the plane Oxy. Illustrate with z = x 2ey .
12. If u~f(z) where z <j>(x, y), show that the stationary values ofuomdz
occur at the same values of x and y. If/'(z)>0 always, deduce that corre-
sponding stationary values of u and z are also of the same type.
13. The variable p is defined as a function of four variables a, 6, t and r by
the relation ale pt + bre pT /(a, 6, t r), where I and r are constants and/ is a
t
given function. Find four equations to determine a b,t and T for maximum p. 9
(See Wicksell, Lectures on Political Economy, Vol. I (English Ed. 1934), p. 181.)
14. By eliminating y, show that z e x*-\ v* has a minimum value e 2 relative
to x - y + 2 = 0. Check by means of the results of 14.6.
15. If x and y are positive, show that the maximum value of xy subject to
x 2 -{-y 2 a 2 occurs where x = y = - . Illustrate by drawing the contour map

N/2
of z = xy and the circle x 2
+ y = 4.
2
/j*2 >tj2
16. The and y are related 2 + 2 = 1, where a>6. Show that

variables x
a b
(x + y ) has then a maximum value a and a minimum value 6.
2 2 Illustrate
diagrammatically when a = 5 and 6 = 3.
17. Establish the necessary condition for an extreme value of z=f(x, y)
" undetermined "
relative to <f>(x, y) Q by Lagrange's method, using an
multiplier A.
18. Find the stationary values of z x 3 4- y 8 - 3xy subject to y mx where
m is a constant. For what values of m is z (a) a maximum and (6) a mini-
mum at # = 2/ = 0? Deduce that x y Q is a saddle point of the surface
z = x 9 + y 3 - 3xy (see Example 7 above).
19. A
rectangular brick has a given volume ; show that its surface area is
least when it is a cube. Show that the same result holds if the brick has a
given surface area and maximum volume.
20. An open box of given volume 4 cubic feet has a rectangular base and
vertical sides. It is made of wood costing 1 shilling per sq. foot. Show that,
for least cost of construction, the base of the box is a square of side double
the depth. What is the least cost?
" "
21. An ellipsoidal block of wood has a surface with equation
& y* *
a* b* c*
referred to axes Oxyz in space. Show that - is the volume of the largest
rectangular brick that can be cut from the wood.

22. If axes Oxyz are fixed in space, what is the distance of the point (x, y, z)
from O ? Express the distance from O to any point on the plane 3x 4- 4y + 5z = 10
in terms of the x and y co-ordinates of the point. Deduce that J2 is the
shortest and perpendicular distance from O to the plane.
23. Given z mf(x) + n<f>(y) where mx -\-ny-a and + n = 6, find z as a m

function of x and y for given values of a and 6. Deduce that z has extreme
values for variation of x and y where
(See Wicksell, op. cit., p. 140.)
Economic maximum and minimum problems

24. A monopolist firm produces two chocolate lines X t
and 2 at constant X
average costs of 2s. 6d. and 3s. per Ib. respectively. If p l and p 2 are the prices
charged (in shillings per Ib.), the market demands are
thousand Ibs. per week of the two lines. For maximum joint monopoly
revenue, show that the prices are fixed at nearly 4s. 6d. and 4s. 9d. per Ib.
respectively.
25. In the problem of the previous example, find the prices of the two lines
fixedby independent monopolists and show that they are less than the joint
monopoly prices.
26. A monopolist produces cheap razors and blades at a constant average
and Is. per dozen blades. The demand of the market
cost of 2s. per razor
per week is x, - thousand razors and x,. = - thousand dozen blades

when the prices are p (shillings per razor) and p 2 (shillings per dozen blades).
1
Show that the monopoly prices, fixed jointly, are 4s. per razor and 2s. per
dozen blades.
27. If the demands of the market

in the previous example are
x1 =U-2Pl -2p 2 and X 2 = sf-2p 1 -^p 29
show that the joint monopolist would give the razors away and sell the blades
at a high price.
28. The market demand law for a good X is p = jS - <*x. A monopolist

produces X at average cost (ax + 6) for output x and sells to a merchant at a
price TT which maximises his profits. The merchant is a monopolist with
constant distributive costs and maximises his profits by selling on the market
B-b
at price p. Show that the amount of X produced and sold is x = -~- 5. 2 (Ci -f- 4(X.)
and that TT = - 2<xx and p =

)3 j3
- ax. Find the output if the producer mono-
" "
polist sold direct to the market and show that bilateral monopoly hero
restricts output and raises price.
29. Generalise the problem of the previous example by taking a market

demand law p = </>(x) and producer's average coat /(a?). Show that
(x) + 3xp(x) + x*</>"(x) =f(x) + xf'(x)

<f>
gives the output x and that TT <f>(x) +x</>'(x). (See Bowley, Bilateral Mono-
poly, Economic Journal, 1928.)
30. In the problem of 14.5, assume that the product after t years represents
the original cost accumulated at some continuous interest' rate of lOOp per
cent. Given the factor prices, express p as a function of a, 6 and t. The
monopolist is assumed to arrange production to maximise p ; find equations
for a, 6 and t. If the maximum equals the market rate of interest r, show that
the result is equivalent to that of the competitive case of 14.5.
31. If the production function of 14.5 is xâbe^*, show that the
equilibrium period of production is rr- whore r is the market rate of

interest. In the competitive case, show that r is given by
32. If x A Jab is the production function, find the amounts of the factors
used at given prices pa and p b to produce an output x at smallest cost. In the
case of pure competition on the market for X
with the demand law x fi- up,
show that the demands for the factors are
-
-, and 6 ~
A
-7
. .
^4 Vpa A*
33. For any production function, if each output x is obtained at minimum
cost at given factor prices, show that the common equilibrium value of
/p fp
^~ and -~ is equal to the marginal cost of output x. If the firm fixes output
Ja Jb
to maximise profits at a given selling price p, show that the marginal produc-
tivity law pa =pfa and p b ~pfb holds. Why need not total cost equal total
receipts in this case?
34. A good
"
X
is produced with two
"
"
substitutional
" factors A and B and
a limitational factor G according to the relations x=f(a, 6) = ^(c). If
each output x is produced at minimum cost at given factor prices pa p b and p e , ,
show that the amounts of the factors used and the total cost are determined
as functions of x and deduce that
PaJPb^^H_ Pc_
fa h
?(C)"**
The selling price p is fixed and output chosen for maximum profits. Show that
and
"
In what sense can the factor prices bo said to be equal to the marginal
revenues" of the factors? (See Georgescu, Fixed coefficients of production
and marginal productivity theory, Review of Economic Studies, 1935.)
86. If u = xyP is an individual's utility function for two goods, show that
his demands for the goods are x ^^ -- and y *~~ where p and py
are the fixed prices and the individual's fixed income. Deduce that the
/LI
elasticity of demand for either good with respect to income or to its price is
equal to unity.
36. An individual's utility function is u 2gx + 2fy - ax* - 2hxy - by Show

2
.
that his demands for the goods are linear in the income the coefficients ^
depending on tho prices of the goods.
37. Differentiate the equilibrium conditions giving the demands for two
goods X
and Y
for an individual with utility function u <f (x, y) in the case
where = 0. Show that
where *-+ <Pxx <Pyy
and deduce that the demand for a good decreases as its price increases provided
that ^
and
<^ are negative. Why is this statement of tho problem
not
independent of the particular utility function selected ? (Seo Pareto, Manuel
d'&conomie politique 2nd Ed. 1927, pp. 679-84.)
t
38. The incomes of an individual in two years are x and T/ O and his utility
function for incomes is u = x*y&. Show that the demand (x - x ) for loans this
year decreases as tho given market rate of interest lOOr per cent, increases.
Deduce that the individual will not borrow this year at any (positive) rate of
Q
interest if y 9 < -a # .
39. In a self-contained community, an individual can produce, with his

given resources, amounts of goods X, Y, Z, ... according to the transformation
function F (x, y, z, ) and he can then buy or sell on a market at fixed
. . .
prices px p v p z ...
, , His utility function is u= <f>(x, y, z, ...). How are the
, .
amounts he demands or offers determined? Illustrate diagrammatically in

the case of two goods and show that the solution is exactly analogous to that
of the problem of the demand for loans (14.9 above).
40. Values (xt ) of a variable quantity X are given for a series of (2n+ 1)
Jars
t= -n, -(n-1), ...-1,0, 1, ...(w-1), n
taking = at the centre of the period. Write x = at + b + v
t t
where a and 6
are to bo determined so that the sum of all observed values of v t
*
is a minimum.
Show that
Ztx t _ _ Zx t
_
a -^,-~ and & ^
2t* 2n 1 -{-
where S implies summation over the (2n 4- 1 ) observed values. Iri what sense
does x at + b represent the " trend " of the series of XI
41. The Board of Trade Wholesale Price Jndex-mimber ^(1924^100) at
certain dates is given by the following table :
Find the trend x at + b by the method of the previous example. Compute

the trend values for each date arid compare with the actual value of the index
Represent the original values and the trend values graphically.
CHAPTER XV
INTEGRALS OF FUNCTIONS OF ONE VARIABLE
15.1 The definition of a definite integral.
" "
THE concept of an integral has two very different characteristics
and two correspondingly distinct applications. From one point of
view, an integral is the limiting value of a certain summation ex-
pression which is of frequent appearance in mathematical analysis
and which corresponds, in diagrammatic terms, to an area enclosed
by a plane curve or curves. The integral, viewed in this way, is
" "
called a definite integral. From another point of view, an
integral the result of reversing the process of differentiation. The
is
derivative of a function of one variable is itself a function of the

same variable.The inverse problem is to obtain, from a given
function, a second function which has the first as its derivative. The
" "
second function, it can be found, is called the indefinite integral
if
of the first.
It isessential, in the following development, to distinguish care-

" "
fully between the two aspects of an integral, the definite and
" "
the indefinite aspects. The actual definition of an integral can
be framed in one of two ways and, once one aspect is adopted as the
basis of the definition, it is then our business to establish the property
implicit in the second aspect. It is found that the sum-area concept

of an integral provides the better starting-point. The definition
given below, therefore, makes use of this concept and it is only at a
later stage that the integral is connected with the inverse differentia-
tion process. This is merely a matter of formal presentation the ;
sum-area and inverse-differentiation properties of integrals are

equally important in application.
It is assumed that yf(x) is a single-valued function which is
continuous for all values of x in the given interval from to xa
INTEGRALS OF FUNCTIONS OF ONE VARIABLE 385
x =&.* The interval of x, of length (6 - a), is divided in any way we

please into n sections by means of the dividing points :
a=x i3 x% # 3
9 ,
... ,
#n _i, xnt #n +i=0.
We form the sum :
where each term is obtained from a different section and consists of

the length of the section times the value of the function at the lower
(or left-hand) point of the section. For convenience, a special
notation is adopted in which we write the sum as
where the term written is the typical (rth) term of the sum and the
n
symbol 2 indicates that all such terms, from the first (r
= l) to the
r=l
nth (r~ri), are to be added together.f The number n of the sections
into which the given interval is divided is now increased in any way
such that each section becomes smaller. It can be shown that the
sum above then increases and approaches a definite limiting value.
No formal proof of this basic result is offered here but a diagram-
matic indication of its correctness is given below. The limiting
value, approached as n tends to infinity, is called the definite integral
of the function between the lower limit a and the upper limit b and
&
f
written in the symbolic form I
f(x)dx. So :
Ja
n
fb
DEFINITION: I
f(x)dx=lLim 2 f(xr )(x r+I -xr ).
Ja n K r=l
The process of finding the integral of a function is called integration.

From the definition, it is clear that the value of a definite integral
is simply a number which can be positive or negative and which
depends only on the form of the function and on the values of the
* The rigid assume that the function
definition of a definite integral does not
is continuous. development, following Riemann (after whom the
The full
integral is sometimes named), is somewhat involved see Goursat, Cours
d* analyse mathtmatique, Vol. I (4th Ed. 1924), pp. 169 et seq. The essentials
of the definition, however, are clearly indicated here.
f This 2 notation is found to be of great convenience in dealing with sums
of all kinds. The Greek capital 2, an alternative to the English S, is an
obvious letter to indicate a sum.
2B M.A.
limits (a and 6) taken. The variable x is not involved ;
it has been
"
integrated out ".
It can be shown, further, that sums other than that written above
have exactly the same limiting value, the definite integral, as n tends
to infinity. Such sums are formed by taking, for each section, the
length of the section times the value of the function at the upper
(or right-hand) point of the section, or indeed at any point within
the section. This fact, again not formally proved, is supported by
the diagrammatic illustrations below. Hence :
f
Ja
'
where xr can take any value from xr to x r+l inclusive.
Put more broadly, the definition is as follows. The interval from
x= a to x = b is divided into a number of sections of which Ax is a
typical one with x as one point contained in it. Then
where each section length Ax i& multiplied by the value of the
function f(x) at a point of the section, the whole set of products is

summed and the limit found as the number of sections into which
the given range (a, 6) is divided is increased indefinitely.*
The
following properties of definite integrals are derived at once
from the definition. If f(x) and </>(x) are single-valued functions
continuous over the relevant intervals, then :
(1)
(2)
J\*{-f(x)}dx=-\*f(x)dx.
a J a
Fb Fb
(3) I
kf(x)dx = k\ f(x)dx (k being a constant).
J a Ja
(4) f(x)
J\'f(x)dx=
a J\ a
* The
way in which the integral notation has arisen is clear from this rough
statement of the definition. In the limit, the sum 2f(x)dx is written Sf(x)dx,
S being an alternative symbol for 2- The S then becomes elongated into the
conventional sign I.
b
Jf
(5) {f(x)+4(x)}dx=ff(x)dx+\ t(x)dx.
a Ja Ja
The last of these properties can be considered as an illustration of

the formal method of proof. We have
w
2f\Jf try
V
x r,
r=l
n
2fJ\/~x \
r)
- Sr ) + 2 <
(r) ( X r+l
~
r=l
i.e. = f */(*) * + f
J a J
on taking the limit as n tends and using the property

to infinity
(4.5) that the limit of a sum is the sum of the separate limits. It
can be noticed that this result extends to the integral of a difference
and, in an obvious way, to the integral of sums or differences of any
number of separate functions.
15.2 Definite integrals as areas.

The diagrammatic representation of a definite integral which
follows makes the concept much more clear and, at the same time,
leads to important applications. Suppose, first, that b>a and that
the continuous function f(x) assumes only positive values between

x=a and x = b. Further, for simplicity of exposition, take the
function as monotonic increasing. The function is then represented
(Pig. 92) as a continuous curve lying above the horizontal axis Ox and
rising from left to right between the ordinates AP (at x=a) and
BQ (at x = 6). Let CD be a typical member (x r x r+1 of the n , )
sections into which we divide the range AB (n = 8 in the figure). The

n
contribution of this section to the sum 2 f(%r) (
x r+i - x r) is CD times
r=l
OR, the ordinate of the curve at C, the area of the rectangle
i.e. it is
with CD as base and CR as height. The whole sum is thus the sum
of the rectangular areas shaded, in Fig. 92, an irregular area lying
entirely under the curve between P and Q. As the number of
sections is increased, each section becoming smaller, the shaded
area, comprising an increasing number of rectangles becoming steadily
thinner, increase and approach a limiting value which we can
must
identify as the area under the curve, above Ox and between the
ordinates AP
and BQ. Hence, the sum written above has this
area as limit and
f
J a
,
ABQP under the curve yf(x).
It is also clear that the same result follows if we start from one of
the other sums used in the definition of the integral. The sum
n
2 /(#r+i) (
x r+i - x r) ^ represented by a sum of rectangle areas, a
typical rectangle having CD as base and DS as height. We have

now an irregular area enclosing the curve and, as the number of
sections is increased, the area decreases and approaches a limiting
value which is the area ABQP under the curve. Further, the sum
n
'
2 f( x r')( x r+i-%r)> where x r is any value between x r and x r+1 ,

is
r=i
shown as the sum of rectangle areas, a typical rectangle having base
CD and height equal to some ordinate of the curve above CD. This
area again approaches the area ABQP under the curve as the number
of sections increases. It does not matter which of the sums is taken ;
the limit is always the definite integral shown by the area ABQP
under the curve.*
* It can be noticed that we have
actually given here a definition of the area
Tinder a curve. The only areas known to elementary geometry are those of
rectangles. But, if we fit to a given curvilinear area a set of rectangles in one
of the ways indicated, and if the sum of the rectangle areas tends to a limit as
the number of the rectangles increases, then the limit can be defined as the
curvilinear area. Compare this definition with the graphical method of esti-
mating a curvilinear area in practice, by counting the number of small squares
of the graph paper covered by the area.
A argument applies when the function, though positive,

similar
is not monotonic increasing between x=a and x =6. The integral
fb
I f( x )dx can always be represented as the area under the curve
Ja
y=f(x), above the axis Ox and between the ordinates at x=a and
x =6. This is illustrated in Fig. 93.
B x
yv
o,
O a & X
FIG. 93. FIG. 94.
When the function f(x) assumes negative values between x=a

and x=b(b>a) complications arise which can only be overcome by
9
making conventional distinctions between positive and negative areas.

If f(x r ) is negative, then the contribution of the section (x r x r+l ) to ,
n
sum 2 - %r )
the f( x r) 0*Wi *s
negative and shown by a rectangle below
r=l ,
the axis Ox. It follows, when the limit is taken, that the integral
fb
Ja
f(x)dx is still represented by an area between the curve y=f(x),
the axis Ox and the ordinates at xa and x = 6, provided that any
parts of the area above the axis Ox are taken as positive and any
parts below the axis Ox as negative (see Fig. 94). The integral and
area as a whole may, of course, be positive or negative.*
So, if the function y~f(x) is positive in the interval (a, 6) and the
6
f
curve above Ox, then the integral f(x)dx is positive and measured
'a
Cb
* It must be emphasised that the is here not the sum, but
integral f(x)dx
'
a
the difference, of the numerical areas PAG and QBC of Fig. 94. If the
numerical sum is required, the areas PAG and QBC must be obtained separ-
rc
fb
- f(x)dx, where OO
ately as I
f(x)dx and- I
J
c.
Ja o
by an area under the curve. If the function is negative in the
interval and the curve below Ox, then the integral is negative and
numerically equal to the area below Ox and above the curve. If the
function changes sign in the interval and the curve crosses the axis
Ox, then the integral may be positive or negative and is represented
by the algebraic sum of a positive area (above Ox) and a negative
area (below Ox).
A further convention regarding the sign of an area is needed to
/b
interpret the integral f(x)dx as an area in the cases where a>b.
Ja
Each of the terms (x r+1

- x r appearing
any of the sums defining
)
in
the integral is now negative instead of positive and the corresponding
rectangle areas are described from right to left instead of from left
to right. We simply need the convention, therefore, that an area
described from right to left is numerically equal, but opposite in
sign, to the similar area described from left to right.
fb
In all cases, the integral f(x)dx is measured by the area be-
J a
tween the curve y=f(x), the axis Ox and the ordinates of the curve
at x a and x = b. Various parts of the area must be considered as
taking signs according to the conventions here indicated. The area
as a whole is the algebraic (and not the numerical) sum of the
separate parts and it may be positive or it may be negative.
15.3 Indefinite integrals and inverse differentiation.
In introducing the second aspect of integrals (as inverse to

derivatives), we can quote a simple example of the kind of problem
involved. The variable y is known to be a single-valued function
of x such that, for all values of x, we have
dy = ,
:r 1+a:
ax
" " ~~ of a
This is a differential equation involving the derivative
ctx
function so far unknown. Is it possible to deduce the actual form

of this unknown function from the equation? We know that
(*)
=! and
So ,-(a: + Jx 3 ) = l+a; 2 .
doc
One possible form of the function we seek is thus
This is obviously not unique since the addition of any constant to y

leaves the derivative unaltered. So, if c is any constant whatever,
y=zx + lx* + c
is a form of the function required. It is also fairly clear that we can
2
expect no other form of y to have the derivative (l+x ) at all
points. In a particular case such as this, it is possible to reverse the
" "
differentiation process, to solve a differential equation of the
above type. Notice that the additive constant, which disappears on
differentiation, reappears in the reverse process. We have now to
generalise this step and connect it with integration.
fb
The definite integral I
f(x)dx depends for its value on the form
Ja
of the function f(x) and on the values of the limits a and 6. If the
form of the function and the lower limit a are given, take a variable
number x as the upper limit and write the integral f(x)dx. The
J a
value of the integral now depends on the value allotted to x i.e. the y
integral is itself a function of x. It is called the indefinite integral of
the function f(x) and written simply I

f(x)dx. So :
DEFINITION : I
f(x) dx f(x) dx =& function of x.
J J a
By indefinite integration, we obtain, from a given function of x, a

second function of the same variable. In diagrammatic terms, the
indefinite integral is a variable area, i.e. the area between the curve
y=f(x), the axis Ox, the fixed ordinate at x a and a variable
ordinate at x.
The fundamental result connecting differentiation and integration

can now be stated :
Iff(x)a continuous function, then the derivative of the indefinite

is
integral off(x) is the function f(x) itself:

Without giving a completely rigorous proof, the correctness of this
result can be seen in diagrammatic terms. Let F(x) denote the
indefinite integral of the continuous function/ (a;). Then
rx+h r rx+h
-F(x) = J \ f(x)dx-\*a f(x)dx=\ f(x)dx
a '
by property (4) of 15.1. The last integral written is shown by the

area between the curve y =/(#), the axis Ox and ordinates at x and
(x + h). If h is small, the area differs by a small amount from that
of the rectangle with base h and height /(#), the ordinate of the
curve at x. So, if h is small,
F (x + h) - F (x) = hf (x) approximately,
i.e. - -- j-
-
=f(x)
.. . .
approximately.
. .
It follows that, as fe~>0, the ratio on the left-hand side tends to /(#),
i.e. the derivative of F(x) isf(x). Q.E.D.
The fundamental can be put in a more practical form.
result If
a function <f>(x)
can be found with f(x) as its derivative, then
and !/(#)<## and (f>(x) must be identical except for an additive and
arbitrary constant :
f(x) dx~<f> (x) + constant.
The integral here is between limits a and x. Putting x =a,

a
= + constant.
r f(x) dx
*<!
(f> (a)
The integral now is zero, representing an area on a base of zero

length. It follows that
constant = -<(a),
i.e. the arbitrary constant of the indefinite integral can be written in

terms of the arbitrary and constant lower limit of integration. Put
a: =6 in the indefinite integral and substitute for the constant. So
fb
f(x)dx = <f> (b) 4- constant = <f>(b)-<f> (a).
i:
The practical rule for integration is thus :
If f(x) is a continuous function and if -=- {</>(x)} =f(x), then

ctx
I
f(x) dx = <f> (x) + constant
and lf(x) dx=<f>(b)-<)> (a).

Ja
To integrate a given function f(x), it is only necessary to find, by

some means or other, a function <f>(x) with/(#) as its derivative.
The fundamental result can be put in yet a third way. If it is
/jilt
given that the variable y is a function of x such that -p =f(x) 9
where f(x) is a given continuous function, then
dy f( )-d
dx
and y and I
f(x)dx can differ only by a constant. Hence :
di/
The solution of the differential equation ~-f(x) is
ctx
y = \f(x) dx + constant.
We have -thus established the fact that integration is inverse

differentiation. If the function f(x) is obtained from <f>(x) by deriva-
tion, then the function </>(x) is obtained from f(x) by (indefinite)
integration. The only difficulty is that an arbitrary constant re-

appears in the integration process. The same fact is expressed by
saying that the integral off(x) provides the solution of the differential
equation -~=f(x). The concept of a differential equation will be

dx
greatly extended in the following chapter and the idea of integration
as inverse differentiation will be generalised to the idea of the
integration of differential equations of all kinds.
15.4 The technique of integration.

The practical method of evaluating a definite or indefinite integral
depends, as we have just seen, on finding a function with the function
to be integrated as its derivative. If -=-
{<f> (%)}=/(%), then we can
write at once
I
/(#)d# = ^(#)-f constant; I
f(x)dx = <f>(b)-<f>(a).
J Ja
In attempting to systematise the practical method on lines similar

to those that proved successful in the case of derivation, we first
obtain and table some standard forms. From the derivative results,
we have
d fxn+l \
-=a* (n^-1);
did
& '
y-(logs)
cfo
= -;
x :r (*)'
dx^
= "-
Reversing these results, we have the three main standard forms for
integrals :
xn+l
xn dx = 4- constant (n ^ - 1).
- dx =
(2) log x + constant.
J
J x
(3) \e x dx ex + constant.
Further, if f(x) is any continuous function of x, then
The standard forms for integrals can thus be generalised :
f f/7rrH n + 1
(!') /'(*) {/(*)}" d:c
= + constant (n ^ - 1).
J n -f-
,
1
(2')
f
J
-^ (fa = log f(x) + constant.
/v 35 )
(3
7
) /'(a;) e^<*> do;

= e'<*> + constant.
The second step to frame, if possible, a set of rules for the

is
integration of combinations of functions, functions of functions and

so on. Two such rules can be written down at once from properties
(3) and (5) of 15.1. For indefinite integrals, we have
and (Ji(x) +ft(x)} dx =
By an obvious extension and combination of these results, we have
There is no difficulty in integrating an expression consisting of sums

or differences of simple functions. But we can proceed no further ;
any attempt to frame other rules of general applicability, for the

integration of products, quotients or functions of functions, is
doomed to failure. The practical process of integration can be made
neither as systematic nor as complete as in the case of derivation.
" "
In integrating, we are driven back to tentative trial and error
methods, aided by a few more general devices (such as that of

"
integration by parts "). In each case, the most important point
to remember is that any suggested form <j>(x) of the integral of f(x)
can always be checked by derivation: {<f> (%)}=/(%). In the

d'Oc
following illustrative examples, the additive constant that appears

in all indefinite integrals is omitted for convenience of writing.
Ex. 1 .
^
I
Jx dx =
/
I x^ dx
22
#*= = N/# 3
_
, by (1) above.
Ex. 2. t(x
2
+3x + 2)dx =
^x*+^x*+2x
and I
(ax
2
+ bx+c)dx = - ax 3 + ^ bx2 + ex (in general) .
and I
(ax + b) n dx = - a (ax + b) n dx = - -
I
- -
(in general) .
f dor
Ex.4. by (2'),
J^-loga+a?),
and
, f dx = -
I r -
1 f adx
7=
]ax + b ajax + b a
1.
- --
&v(ax + b)
log
.
, v
/
,.
(in
\ e
general)/ .
lx
~
Ex -
_
5 -
xdx
-i-^ .
,_
14 -^
ov i
by(2) >
and " log (a* 2 + bx + c) (in Senera1 )-

J o^+te + c
Ex.6.
Ex. 7. fe 2* efo =i 2a;

ofo e 2 *, by (3'),
J2e =^
and feox+& da; =- oa; +& efa =- eaa!+ 6 (in general).
|ae
Ex. 8. lxe~** dx = - H( - 2) e-* ^ = ~ ^ e-^ 8

, by (3').
Ex.9. {x 2 e~**dx.
Since ^- (ê^^) = e~* + x - 2xe,-**) = (1 - 2x2

( )
e~ x',
(tx
so xe~ x
*
- f (1 - 2#2 ) e-^ ^= fe~* dx-2 \aPe~* dx,
f Ife -ar* 1
x2e~ x* dx dx - - xe~ x
*
i.e. I I .
^
The integral I e"** rfa;, though it is known to exist, cannot be evaluated
in terms of ordinary functions. have, therefore, reduced the given We

integral to as simple a form as possible.
15.5 Definite integrals and approximate integration.

The
integral of a function f(x) between given limits is evaluated
once the indefinite integral (x) is found as a function of x. The <f>
ft
f(x)dx =
formula <f> (b)
- <f> (a) suffices. For example :
J a
=log2.
Areas under plane curves can then be evaluated at once. The para-
bola with equation y =# 2 + 3x -f 2 lies above Ox for positive values of
x. The area under the curve between ordinates at x = and x = 3 is
'z=0
Wehave noted, however, that the indefinite integrals of some

functions, even of quite simple form, are often extremely difficult,
if not impossible, to evaluate.* If the definite integral of such
a function required, for example, to give the area under the cor-
is
responding curve, it is necessary to adopt methods other than that

described. A
number of formulae giving the approximate values of
have been devised and are of service in the present
definite integrals
problem. depend on replacing the area under the

All these formulae
actual curve considered by that under a simpler curve which approxi-
mates to the actual curve.
Suppose that the function f(x) is continuous and positive in the
ft
interval (a, 6) and that we require the value of f(x)dx. The
Ja
interval (a, 6) is divided into some even number (2n) of parts of equal
length h = b-a
-
This is shown, in Fig. 95, by the points A 19
2n
FIG. 95.
, ... A 2n9 A 2n+l on the axis Ox, where OA 1 a and OA 2n+l ~b and
AAA^.Ji. The ordinates = AP, 2, ... can be
b
f
read off the curve yf(x). The integral f(x)dx is the total area
Ja
* It is not
possible, for example, to find e-^da: in terms of ordinary func-
tions. The integral exists and must be a function of x. But the function is
not of algebraic form ;
it is, in fact, a " transcendental " function which can
x*
(xa e~ dx for various values of x. It
is a frequent device, in mathematical analysis, to define new functions as the

integrals of other functions.
under the curve and on the base ^4 1^4 2 n+iJ *- e *ê sum -
under the curve and on the bases AÂ^ A 2A Z ... ,

.
As a first approximation, the area under the curve on the base

-4^4 2 can be taken as the area of the trapezium A^P^PÂ^ :
Similarly, the approximate areas under the curve on the other bases
are
The approximate value of the total area, i.e. of the integral required,
is found by addition. Hence,
fb
f(x) dx =
a
approximately, the trapezoidal rule for approximate integration.

A closer approximation to the area under the curve on the base
AJÂÂS given by the area under the arc of the parabola (with axis
is
vertical) passing through the points P19 P 2 and P 3 To calculate .
this latter area, let the equation of the parabola be
when the origin for x is taken, for convenience, at 2 (This step A .
does not affect the value of the area required.) The curve passes
through Px ( -*, ft), P 2 (0, y a ) and P 3 (A, y s ). Hence,
Solving for a, j8
and y, we find
The area under the parabola on the base AÂÂ^ is
-h L* * ~*x=>h Lrf ^ -V h
= -aA 3 + 2y/i = -(
The approximate areas under the curve y f(x) on the bases
A BA QA 7 , ... are obtained similarly as
By addition, we have the approximate area under the curve on the

total base, i.e. the approximate value of the integral required.

Hence,
' u
3
approximately, Simpson's rule for approximate integration.

Two examples illustrate the rules obtained :
Ex. 1. To an approximate value of log 2

find I
^
,
we divide the
range (0, 1) into four equal parts by the points
The ordinates of y =- at the points

r of division are
1 +x
2/1
= 1, 2/2=1 2/3
= 1> 2/4 =$ and 2/5
= 1-
The trapezoidal rule (h
= l) gives the approximate value
O A -rX
Simpson's rule gives the approximate value
log 2
=P +*) +2(1) +4(|+f)} =tft& =0-6933.
JO --^-={(1
A +#
Five-figure tables give the correct value of log 2 as 0-69315. The

nature of the approximation of the two rules is thus seen. The rules give
closer approximations, of course, if the range (0, 1) is divided into more
than four parts.
Ex. 2. The area under the curve 2/ == e~ aS above Ox and between the
ordinates x = and x =2 is given by the integral
2
p.e-* dx.
Jo
Jo
Dividing the range (0, 2) into ten equal parts (h=\), the ordinates at
the eleven points of division are found, from tables, as
ft
= 1-0000, 2/2-0-9608, ya =0-8521, y4 -0-6977,
^5 =0-5272, */6 =0-3679, y7 =0-2369, 2/8=0-1409,
i/9 =0-0773, 2/10=0-0392 and 2/11 =0-0183.
The trapezoidal rule gives the approximate value
fV^fe =^{1-0183 +2(3-9000)} =0-88183,

Jo
and Simpson's rule gives the approximate value
Fe-**dx =^{1-0183 +2(1-6935) +4(2-2065)} =0-88209.
Jo
The required area, to three decimal places, is thus 0-882.
15.6 The between average and marginal concepts.
relation
The demand of a market for a good X
can be represented by three
functions and curves showing total, average and marginal revenue
respectively for various outputs of X
The relations between these
.
functions and curves can now be examined in more detail. In Fig.

96, P and Q are points on the average and marginal revenue curves
\M.R.
FIG. 96.
corresponding to a given output x OM. Denote average revenue

(price) by pMP and marginal revenue by pm =MQ. Then the
total revenue from the output x is
R=px = OM MP, .
i.e. the area of the rectangle OMPN. Further,

dR_
~~^ m>
dx
i.e. . dx 4- constant.
The arbitrary lower limit of the integral is here fixed at zero output.
Put x = then the constant must be zero since B is
;
zero at zero
x
output. Hence, ,
R = f Pm d9
I
Jo
the area under the marginal revenue curve, above Ox and between
i.e.
the ordinates at and M. As long as the marginal revenue curve is

above Ox, this area is positive and increases as x increases. When
the curve falls below Ox, part of the area must be taken as negative
and the total area decreases as x increases.
The from any output can be read off as an area

total revenue
eitherfrom the average or from the marginal revenue curve. We
have a relation between the two curves at any output. No matter
how the position of M
varies, the area of the rectangle is OMPN
always equal to the area under the marginal revenue curve on the
base OM .
A similar result holds for any pair of average and marginal curves.
For example, if TT and 7rm denote the average and marginal cost of an
output x of a good, TT and 7rm being functions of x, and if the cost of
zero output is zero,* then total cost is
I 7rm dx. n=xiT

Jo
Total cost is read off the average cost curve as a rectangular area
and off the marginal cost curve as an area under the curve.
15.7 Capital values.
It has been shown that the present or capital value of an income
stream (e.g. from land or a machine) consisting of a this year and
al9 a2 ,
. . . am in m following years is
where interest is reckoned yearly at lOOr per cent, (see 9.7 above).
This calculation can be generalised to allow for a continuous income
stream and. for interest computed continuously.
It is assumed that income is obtained continuously over time,
the rate at any time t years from the present being f(t) per year.
This implies that, in the small interval of time between t and (t + At)
years from the present, the income obtained is approximately
f(t)At. Interest is computed continuously at the rate of lOOr per
cent, per year, where r can vary over time as a function of t. The
present value of the income in the small time-interval (t, t+At) is
" "
rt
f(t) e~ At approximately. If the income stream dries up x years
from the present, the approximate present value of the stream is
where the summation extends over all small intervals At from =

* If the cost of zero is 77 it is found that the constant in the integral
output ,
expression for 77 is 77 . So 77 - 77 = m dx
ir t i.e. the area under the marginal
J
o
revenue curve, is the total variable cost of the output x.
2c M.A.
to t~x. As the number of time-intervals is increased, each interval
becoming shorter, this approximation becomes closer. The exact
present value is defined as the limit of the sum as the number of
intervals increases indefinitely, i.e. as
V =\*f(t) *"*<**
Jo
This is the capital value of the stream estimated at the present time
( 0). Thus, in passing from a discontinuous to a continuous
income stream, the capital value changes from the sum of a finite
number of terms to the limiting case of a sum as represented by an
integral. We have an immediate application of the sum-area aspect
of an integral.
The result obtained holds for an income stream and a rate of
interest varying in any way over time. In the particular case where
income is derived at a constant rate of a per year and where the
rate of interest is fixed at lOOr per cent, per year over time, we have
y=â\ e~ dt = a \- --e- rt
rt
e~ rt
Jo L r -W ~a\L
\
r
\
-U
i.e. y = ~(
T
The capital value thus depends, in a simple way, on the size of the
income stream, the number of years it flows and the rate of interest.
Problems of capital accumulation can be treated in a similar way.
If a alt a 2 ... am are invested in the present and in succeeding
, ,
years at lOOr per cent, compounded yearly, the total sum accumu-
lated at the end of the period is y where
For a continuous investment which, t

years from now, is made at the
rate of f(t) per year, the total sum accumulated at a time x years
from now is found, by an argument similar to that above, to be
where interest is added continuously at lOOr per cent, per year. In

the particular case where the investment a and the interest rate
lOOr per cent, are both fixed over time, we find
=a P
J
.e.
This expression is simply e raj times the previous expression giving

the present value (at = 0) of the same constant income stream,
i.e. the accumulated value of the constant income stream after
x years is the present value itself accumulated at the fixed rate of

interest. This result does not hold when the investment or the
interest rate varies over time.
As an
application of the results above, suppose that the production
process of 14.5 above is carried on continuously over time. A
constant expenditure at the rate of (ap a + bp b ) per unit of time is
made continuously and the product x of each investment is realised
after the production period of t years. The total accumulated value
of all investments in a period of t years is
where interest is reckoned at the fixed continuous rate of lOOr per

cent, per year. The sum k can be interpreted
either as the accumu-
lated amotint of loans needed to start the production process (be-
tween the original investment and the appearance of the first product)
" "
or as the total value of all intermediate product existing at any
" "
time. In any case, k represents the amount of working capital
needed to keep the production process running continuously. In
the competitive case, where the market rate of interest is such that
we have k
r
a 1 ~a
If x=Aa b fl is the production function, then (from 14.5)
and the capital needed varies inversely with the competitive level of
the market rate of interest.
15.8 A problem of durable capital goods.

The general problem of durable capital goods is complex but can
be simplified, at a first approach, by making the following assump-
tions.* A capital good (e.g. a machine) is produced at a cost of f(t)
dependent, in a known way, on the length t years of its active and

continuous life. An entrepreneur, buying the good at its competitive
(cost) price, obtains from its use a constant and continuous income
stream of a per year. At the time of purchase, the discounted

(capital) value of all future product obtained from the good during
its life of t years is
where lOOr per cent, per year is the constant and continuous market
rate of interest. It is assumed that the entrepreneur chooses the
length of life of the good to maximise the capital value as a propor-

tion of the cost :
rt
_a(l-e-
y~
)
'
rf(t)
TT dy =
Mere ,-
-*/'()"* fl
" + f(i} ]
<Pir
r{f(t)Y I /'(*)/
The necessary condition for the optimum length of life is thus
~ + f(t)
e'*-l+r
TWr
a relation giving t in terms of the given rate of interest.
In the normal case, the cost/() increases, but less than proportion-
ately, with the durability (t years) of the good. As an actual example
of a cost function of this type, take f(t) =bjt, where b is a constant.
Here f'(t)=-- and the equation giving the optimum t in terms

. ttJt
of r is
e rt = l+2rt.
* The
analysis is based on the work of Wicksell following Gustaf Akerman.
See Wieksell, Lectures on Political Economy, Vol. I (English Ed. 1934), pp. 274
et seq.
Write x=rt, and we have

ex = l+2x.
Solving by graphical methods (2.8 above), it is found that x has one

value (other than zero) which is approximately x 1-27. So
t -
= 1'27
T
approximately.
The length of of the capital good, as selected by the entre-

life
preneur, varies inversely with the rate of interest. If the market

rate of interest is 3 per cent., the optimum length of life is over
42 years, but at 4 per cent, the length of life is not quite 32 years.
Suppose now that production is carried on continuously, the
capital goods being brought into use continuously and uniformly at
the rate of n goods per year. After an initial period, it follows that
the goods are scrapped at the same rate, each having served its
optimum life of t
years. At any moment, there are nt goods in use
with ages varying uniformly from to t years. The number of
goods in use with between x and x + Ax years of life to run is approxi-
mately n Ax. The value of the future product of one such good,
discounted to the moment considered, is
-(!--").
r
v
The total present value of all goods in use is approximately
the sum extending over all intervals Ax from to t.

Taking the
limit for the continuous range of ages of the goods, the exact present
value of all capital goods in use at the given moment is
~
r
= na I
(
t
1
+ - e- f
.
*
1\
= na2 (e~ rt + rt- I).
/
.
'
r \ r r) r
This capital value, like the optimum life of each good, is dependent
on the given market rate of interest.
15.9 Average and dispersion of a frequency distribution.
Weare given N
items showing values of a certain attribute
according to the frequency distribution :
For example, the families in a certain town may be distributed

according to the number of rooms occupied. The attribute, number
of rooms occupied, takes the values = 1, x 2 ~2, x 3 = 3, ...
^ Then .
Hi represents the number of families occupying one room and so on.

The average value of the attribute is defined as
and the standard deviation of the attribute as a x where
These are familiar statistical figures, the average indicating a

" "
central value and the standard deviation indicating the spread
" "
or dispersion of the attribute in the group concerned.
Most of the variable characteristics considered statistically are
not of this discontinuous nature. They are so finely divisible as to
be measured by values which can be taken as continuous variables.
The ages of a set of school children, the rents paid by a group of
families and the death-rates in a number of districts are diverse
examples of continuously variable characters. In practice, a

number of items distributed according to such a characteristic is
shown in definite groups or grades of the variables. Families, for
example, may be arranged in grades of rent between 2 and 3 shillings,

between 3 and 4 shillings, and so on.however, a rough
This is,
method and the results obtained are affected by the nature of the
grading. Something more precise is required for a theoretical treat-
ment. assumed, therefore, that the frequency distribution of a
It is
number of items is represented by a continuous function y=f(x),

implying that the number of items with value of the characteristic

between x and (x + Ax) is approximately f(x)Ax the approximation 9
being closer as the value of Ax is made smaller. The total number

of items is
N = Xf(x)Ax
approximately. Taking the limit, we have exactly
N=\*f(x)dx,
Ja
where a and b are the lowest and highest values of x found.

The approximate values of the average and standard deviation
of the variable in the distribution are ~x and a x where
x=-lZxf(x)Ax and oy^-r-, (#

- x)t 2 fix)I Ax.
J\
N J \ ) x .
N \
Taking the limit for an exact representation, we have

rb
xf(x)dx
-_ *
P J a
NJa '
J a [f(x)dx
(\x-x)*f(x)dx
and r'-ilV
'*- N \(X X)
*) x)dx= -
J\*f(x)dx
a
We have here another direct application of the sum definition of an

integral.
For example, Pareto's income law asserts that
A
y (A and a positive constants)
X'
represents the number of persons with an income of r, a continuous

frequency distribiition of persons according to income. The number
of persons with income between a and 6 is
A7
N=\ ~-dx
,
---
a \aa
\
a &-* X!. a
The average income of these persons is
= -\ x 4,
_lf &
*dx = -^\
1 f
&
A,
dx=-=-r
1
-A
N oc-lW- -I
I 1
---
1
b
1
a~
Take and ax let b tend to infinity. The number of persons with
income over x is thus
N= a I
X*
and the average income over x is
The law has the property that the average income over x is a
constant multiple (
-
)
of x. In practice it is found that a = 1-5
\a I/
approximately. Hence, the average income over x is approxi-

mately 3#, a result which agrees well with observed data of income
distribution, at least for large incomes.
EXAMPLES XV
Integration
Cb
1. From the sum definition of an integral, show that I 1 dx = b - a. What
is the indefinite integral of 1 ? Hence check the result.
n /
xr _|_/ \ rb
sum 2 /( r+1 - x ) has
2. Iff(x) is continuous, the
A
) (xr+l r f(x)dx as
1 \ ' . .'(X
f&
its limiting value as n increases indefinitely. Deduce that xdx = %(b 2 a2)
and check from the indefinite integral of x.
rb re rb
3. If f(x) = xn , check that f(x)dx=\Ja f(x)dx+\Jc f(x)dx for any values
Ja
of a, b and c.
4. Show that ~| [ /(a;)^|

= -/(a;). (Use property (1) of 15.1.)
5. Evaluate the indefinite integrals of

ax + b
6. Use the results (l')-(3') of 15.4 to evaluate the integrals of
Cx P x 4- P X
7. Show that
f
- dx = log (
1 + c*) and
C
-~~-x dx = log (e -
j ^ j
8. From the integral of the general expression (ax + b) n show , that
+a;) rfa; = -(l+a;)

8 8
Expand (1+x)* in powers of a? and show
. that
( 1 -H a;)
2
dx = x + x* + - x 9 .
Why are the two results apparently different ?
j
dy =
9. If -T- log x, check that y - x (log x - 1) + constant.
a
F _Ks-ft) l(x-b)*
10. Show that \(x-b)e 2 z
<fo = -o ae 2 a
.
11. Verify that

^ ^
=-
(~TT
-- i)
and deduce that
dx 1, #-l
12. Express a* as a power of e and deduce that a* dx - ~ .

J
8a
log
(2
13. Find 2*6te. Check by drawing an accurate graph of y = 2* and esti-
J
mating (by counting squares of the graph paper) the area under the curve,
above Ox and between the ordinates at x- and x = 2.
14. Find, integration, the area under the straight lino y x+ 1, above
by
Ox and between the ordinates x and x 1. Represent graphically and
chock by calculating the area of a trapezium.
15. Show that
16. Find the area under the parabola y-x^ above Ox and between the t
ordinates at x = - 3 and x^=3. Verify that it is twice the area between the
ordinates at x = and x = 3.
17. Show that the area between the rectangular hyperbola xy = a 2 and the
axis Ox and between ordinates at x = a and x = b is a2 log - , if a and 6 have the

same sign. Why does the result not hold when a and 6 have opposite signs?
(2
Q
(x*-4x+ 3)dx and check that it equals the difference
between the area between the parabola y = x*-4x + 3 and Ox from x = to
x = I and the area between Ox and the parabola from x = 1 to x= 2. Illustrate
graphically.
19. By taking the difference between two integrals, show that the area
enclosed between the parabolas y x 2 and y = *Jx is -J-.
*
20. Show that xe~ x*dx = and |
a?V* dx = - \
e~&dx for any value
J a .'a 2 .'a
of a (see 15.4, Ex. 9, above).
21. Calculate approximate values of LT bv the and

2 trapezoidal
Simpson's rules, dividing the range (0, 1) into ten equal parts in each case.
The value of this integral is known to be JTT. Hence estimate the closeness of
the two approximations.
22. If the velocity of a body at time t is v =/(*), show that x= \

f(t)dt
is the distance travelled from zero time to time t. Find the distance when
v = a - bt and show that the body comes to rest after ^L
travelling a distance .
23. The acceleration of a moving body is a constant a over time. Show
that, after time t, the velocity is v u + at and the distance travelled x ut + \at*
where u is the velocity at zero time.
Integrals in economic problems

24. If the marginal revenue of output x is pm a- bx, find the total revenue
function by integration and deduce the demand function.
ab
26. If the marginal revenue function is pm = ----
+
- c, show that
a (x o)
p
^ ---- - c is the demand law.
x+b
26. Find the total cost function if it is known that the cost of zero output
is c and that marginal cost of output x is 7r
m = ax + 6.
27. If ir
m = j=z= is the marginal cost function and if the cost of zero
output is zero, find total cost as a function of x.
28. If interest is reckoned continuously at 4 por cont. per year, show that
the capital value of uniform income stream of 100 per year for 10 years is
approximately 824 and for 100 years approximately 2,454. (e^ 2-71828.)
29. Write down the capital value of a uniform income stream of a per year
for x years, reckoning interest continuously at lOOr per cent, per year. By
letting x -* oo , deduce that the capital value of such an income stream going
on for over is - .
30. An income stream decreases continuously over time for x years, the rate
bt
being a e~ per year at t years from now. Find the capital value at lOOr per
cent, continuous compound interest. Show that this equals the capital value
of a uniform income stream of a per year for x years if the rate of interest is
raised to 100 (r + b) per cent, per year.
31. An income stream decreases continuously for x years. At time t years

from now, the income obtained is at the rate of a(l - 2 at) per year and the
continuous rate of interest is 100r(l - at) per cent, per year. Show that the
capital value of the income stream equals that of a steady and continuous
income of a per year for x years at the constant rate of interest, computed
-
continuously, of 100r(l otx) per cent, per year.
32. In the problem of Examples XIV, 31, show that the value of all inter-
mediate products at any time is
-IO-D-
where - = 16 log A. In the competitive case (r being such that ap a + bp b = xe~ )
rt
show that
33. In the problem of 15.8, p is defined so that f(t) = -(l -e-p< ), i.e. the
P
product of the capital good represents a net rate of return of 100/> per cent.
(computed continuously) on its cost/(tf). Find an equation for the length of
life of the capital good so that p is a maximum.
34. In the competitive case of the problem of 15.8, the market rate of
interest is such that the present value of the product of the capital good
equals its cost. Show that the optimum length of life of the good as deter-
mined in 15.8 equals that determined in the problem of the previous example.
35. If f(t) 3a*Jt and the competitive case of the previous example holds,
show that the market rate of interest is approximately 4 J per cent, per year
and that the capital good has a life of approximately 28 years.
36. In the problem of 15.8, if f(t) b*Jt and the market rate of interest is
lOOr per cent, per year, show that the present value of all capital goods in use
at any moment is approximately k 0-55 - . If the rate of interest is 5 per
cent, per year, show that the present value is approximately 220 times the
value of the yearly product of the capital goods.
37. In the problem of 15.8, the cost of a capital good of life t years is
where f(t) b ( 1 ---- . Show that the optimum value of t is given by

J
er _
When the interest rate is 4 per cent, per year, find by graphical methods the
approximate value of t.
38. In a sample of 898 working-class families in Bolton, 1924, the families
were classified according to the number of rooms occupied :
Show that 3-65 is the average number of rooms occupied per family. Find
the standard deviation of the number of rooms occupied per family in this
group of families.
39. By differentiating both sides of the equation, show that
xre-* dx = xr~ l e~ x dx - xre~ x
f r^
(oo
Cb
any value of r. If Ir denotes " x e~ dx Lim \x e~ dx deduce that
r x r x
for j
|
b >oo- *
Ir rlr _ l and that / = 1. Hence show that the average and standard devia-
tion of the frequency distribution y xe~ x for the range from x to infinity
are x = 2 and a x J2.
CHAPTER XVI
DIFFERENTIAL EQUATIONS
16.1 The nature of the problem.
DATA provided by scientific observation usually relate to small
movements or changes in the quantities concerned. Information in
"
bulk", covering a wide range of events, is scarcely to be looked
for in practice. Translating the data into forms suitable for mathe-
matical expression, we obtain relations between increments in
variable quantities and, in the limit, the relations involve the
derivatives or differentials of the variables. Mathematical theories
"
in the sciences, if they are to be realistic ", must thus be built on
the basis of differential equations, relations between the derivatives

or differentials of varying quantities. It is then the business of the
theories to deduce the functional equations between the variables
which lie behind the differential equations, i.e. to express the
" "
general laws whose variations correspond to the given data. In
physics, for example, much of the data is expressed in terms of
"
variations in energy ", of one form or another, and leads to
certain fundamental differential equations from which the laws of
" "
conservation of energy are to be derived. Differential equations
thus occupy a fundamental position in most highly developed mathe-
matical sciences.
The nature of a differential equation can be examined first in the
case where only first-order variations in two variables x and y are
involved. The simplest case can be written
where f(x) is some given function. We have seen (15.3 above) that
this case is easily dealt with by means of the concept of an integral.
DIFFERENTIAL EQUATIONS 413
A more general form of the case is
!=/<*,*>
where the given function f(x, y) involves both x and y. Given the
values of x and y, then the value of the derivative of the relation
that must exist between x and y is also known. From this in-
formation, we attempt to deduce the actual form of the functional
relation, to find that particular function which gives a derivative
satisfying the above relation. This is the problem of the solution,

"
or integration ", of a differential equation, a problem which is
clearly an extension of that already solved by simple integration.
A still more general form of the type of differential equation here
considered is a relation involving the derivative in a non-linear
a relation including, for example, such terms as or log

f
~J [-/
The development of the following sections is concerned almost
entirely with the simpler linear differential equation, but it must be
remembered that more complicated forms can be considered and
solved.
diJ
The meaning of a differential equation of the form - -
is
&x f(x, y)
most clearly seen in diagrammatic terms. At each point (x, y) of
the plane Oxy, the equation provides a definite value of the derivative
-
This is the gradient of the direction in which the variables are
ax
allowed to vary from the gradient of the tangent to the
(x, y), i.e.
(at present unknown) curve through (x, y) expressing the relation

between x and y implied by the differential equation. Our problem
is thus that of constructing complete curves from knowledge only of
their tangent gradients. An example illustrates. The equation
dx 2xy
gives a tangent gradient at each point of the plane Oxy. At the
point (1, 1), for example, the tangent gradient is

-^
= J, a line sloping
ax
upwards with gradient | .through the point (1,1). A convenient set
of points, covering the plane Oxy fairly uniformly, is selected and
the tangent gradient evaluated at each point. When lines with the
appropriate gradients are drawn through the points, a graph of the
kind ^shown in Fig. 97 is obtained. It now appears that a system
of curves, one through each point of the can be drawn to
plane,
touch the lines of the graph. This curve system, some members of
Fid. 97.
which are shown in Fig. 97, corresponds to the solution of the

given
differential equation. In this particular case, the curve system
consists of circles with equation x 2 + y z -2ax + 1= where a is a
" "
parameter. This equation is the integral of the differential
equation.*
From a slightly different point of view, we can start from any
point of the plane Oxy and describe a curve through it by moving
always in the direction indicated by the differential equation.
Having obtained this curve, we can select a point not on it and
describe a second curve passing through this point. The
process
can be continued until a whole system of curves is obtained to
represent the solution of the given differential equation.
* The use of the term " integral " to describe the solution of a differential
equation is clear. The solution corresponds to a curve system which is built
up, or "integrated ", step by step, from the tangent gradients given by the
differential equation.
Animportant point now arises. The integral of a differential

equation of the type considered is shown by a curve system and
so it must involve a single parameter or arbitrary constant. This
is illustrated by the integral, x 2 + y 2 - 2ax -h 1 = 0, of the equation
above. The point is worth emphasising by putting the converse

problem. The following examples show that a single arbitrary
constant can be eliminated from a functional relation by differ-
entiation :
Ex. 1. From the relation x 2 -e y =a,

we obtain 2x dx - e y dy = 0,
i.e. the differential equation ~2xe~~ v .
dx
Ex. 2. From x 2 + y 2 - 2ax + 1 = 0,
we obtain 2x dx + 2y dy - 2a dx = 0,
ie
(xdx + ydy)
dx
Substituting in the original equation, we find
X2 +y 2 ~2x rt
-
x dx + y
J dy
^ _ ^
+ 1=0,
(tx
2
i.e. (x*-y -l)dx + 2xy dy=Q,
i.e. rfy as-ys-l
'
dx "2xy
a differentia] equation in which the parameter a does not appear.
y - 2ax + a = 0,
2 2
Ex. 3. Differentiating
we have 2y dy -2adx=0.
Substituting a =y ~- in the original equation we find a differential
dx .,
equation which is not linear in the derivative ~~ :
The
integration of a differential equation reverses the procedure
of these examples, taking us from a relation in the derivative to the
complete relation between x and y . So the integral of ~ 2xe~ v is

ctx
x*-ev~a, and similarly for the others. In each case the integral
involves one arbitrary constant :
The integral of a differential equation in the derivative of y with

respect to x is a relation between x and y which includes a single
arbitrary constant.
The result stated, though not formally proved, has been illustrated
sufficiently for our purpose.*
To extend more than two variables, suppose that we
to cases of
have a relation between the first-order variations in three variables
x, y and 2. In its simple form the relation can be written
f(x, y, z)dx+g(x, y, z)dy-dz = 9
giving the values of the partial derivatives
x >y> z) and x >y> z )'

-^=f( ^~=9(
In diagrammatic terms, we are given the tangent plane (by its
gradients in the two fundamental directions) at each point of Oxyz
" "
space and require to integrate the planes into a system of sur-
faces. The integral of the differential equation thus appears as a
between x, y and z, including one arbitrary constant, and
relation is
shown by a system of surfaces. An example makes this clear :
Ex. The relation x2 + y2 =az (a being a parameter)

gives 2x dx + 2y dy = a dz.
Substituting a =2 -
y dy i
-,
7/ di/
in the original relation, we find
2xz dx + 2yz dy - (x2 + y2 ) dz = 0,

dz -
_ _ 2xz _ ... ..
.
QYlf\ . __
_dz - 2yz
-
We have a differential equation in which the parameter a does not

appear. Conversely, the integral of the differential equation shown is the
relation x2 +y 2 =az including the arbitrary constant a.
More involved differential equations in three or more variables

can be written and some examples will appear later. The simple
type here considered suffices for the moment.
* It is to be noticed that the indefinite integral which is the solution of
conta
-=/(#) contains
-~-f(x) an additive constant, a particular case of the present
dx
17
general result.
16.2 Linear differential equations and their integration.

Of the first-order differential equations in two variables x and y,
the most the linear differential equation in which the deriva-
useful is
tive or differentials are related linearly, squares, cubes and other
expressions in the derivative or differentials being excluded. We

can write the equation either as
where P and Q are given functions of x and y, or as
5-'*>-
where /(x, y) is some function ofx and y. To pass from one form to
the other, it is only necessary to associate f(x, y) with the ratio
- ] It is to be noticed that, though the derivative appears
linearly, there is no such restriction on the variables x and y.

As in the case of simple integration discussed in the previous
chapter, there is no comprehensive practical scheme for integrating

differential equations, even of the linear type. Some particular
methods for certain cases are given below but there remain many
instances in which the integral can be obtained only after much
trial and error or which cannot be integrated by any practical
method.
The following are the four main cases for which definite practical
methods of integration can be given :
(1)'
v
The form ^
dx =/().
t/v '
By simple integration, the solution of this form is
where a is an arbitrary constant.
(2) The form -^=/(a?).

dx y
This form can also be solved by simple integration. Since
d n 1 dy
the equation is - -
y) =/(#)
dx (log
2D M.A.
Hence, log
t/=J
f(x)dx + a,
where a is an arbitrary constant. The variable y is found explicitly

in terms of x by taking the exponential of the right-hand side.
E$. 1. y(l-x)dx-xdy=Q.
Here 1
2/
JÎz.^1
dx x x
_i,
and logt/ = If 1 da: + constant = log x -x + 6.

J
So log
(f)-(* -6),
-6>
i.e. y =#e~ (a:
==e 6 (xe~ x ).
Writing a=e &

, where a is thus an arbitrary constant, we have
x
y=axe~ .
Ex. 2. To find y as a function of x when it is given that the elasticity

of y with respect to a; is a constant a. We have
x~~ dy _ . 1
*~
t% a
~5 ctj i.e. ~j
-
y dx y dx x
C dx
So log y=a I )- constant = a log x + constant,
~ ) = constant,
i.e. ~ = constant = a,
i.e. y^dx ".
Ex. 3. As x increases, the variable y increases at a rate always equal

to a multiple p of its current value :
-^ =/>y.
To find y as a function of x,
we write :
Idy
So
y = epaj+l> = e = aepx
6
and epx ,
where a is an arbitrary constant. Compound interest growth at the con-

tinuous rate of lOOp per cent, per year is a particular case of this result.
(3) The form <f>(x)dx + $(y)dy = Q.

"
This case, in which the variables are said to be separate ",
holds whenever the differential equation can be so arranged that the
coefficient of dx depends on x only and the coefficient of dy on y

only. The integration is then effected at once :
\<f> (x) dx 4- I
iff (y) dy
= constant.
It is only necessary, therefore, to evaluate two simple integrals.
Ex.1. e*^=2*.
dx
Here 2x dx -ev dy=* 0, i.e. 2 \xdx-\ev dy = constant.
So 32 -ey =a, i.e. ev =x2 -a,

2
or y=log(z -a),
Ex. 2. (y + k) dx + (x + &) dy = 0, where & and k are fixed constants.

dx
-- dy = C dx
-- C dy
-TT .
= constant.
x+h7 -^
Here + >
1 -e-
1 + ^7
y+k jx+h )y+K
So log (#+ A) + log (i/ + fc)= 6, i.e. log (a; + h) (y + k) = b.
Hence, writing =
a e b we have
t
where a is an arbitrary constant. The integral is represented by a system

of rectangular hyperbolas (see 3.7 above).
Ex. 3. The variable y increases as x increases at a rate given by
where L and />

are given constants.* To find y as a function of x :
LJ prtf
L dy *^ <
T
/T^Y*
.
If*
f
I
L dy = /V)* [ h
So -=-px-b, i.e. - =
log
where a =e" b is an arbitrary constant. After rearrangement,
L %
* This is a modified version of

growth at compound interest. The curve
representing y as a function of # is called a logistic curve and it is sometimes
used to express the growth of a population over time.
" "
(4) The exact form.
It may happen that the differential equation can be arranged,
after multiplying through by a suitable factor, in the form
where the left-hand side is the differential of some function <f)(x, y),
" "
i.e. Pdx + Qdy~d<f>(x, y). The equation is then said to be exact
and its integral is clearly
<f> (x, y)
= constant.
The following examples illustrate how equations can be arranged
" "
so that they appear in the exact form.
Ex. 1. y dx + xdy=0.
The equation is exact as it stands and can be written d(xy) =0.
The integral is xy=a, where a is an arbitrary constant.
Ex. 2. (x
2
-y2 - l)dx+2xy dy = Q.
2 z -
_ (x -y l)dx + 2xy dy
2tf
Hence, on multiplying through by 2,

the given equation becomes
and the integral is

-~-
y j. 2/2 a. 1
=a, where a is an arbitrary constant.

So
as represented by the system of circles of Fig. 97.
Ex. 3.
fjfj
-~
dx
= -2vyy -f k -
J* __ k/'J/
^=
-L. Jfi
:, where h and k are fixed constants.
We can write the equation in the form

dy /i ,
The is thus
integral
-
h-Jy + a,
,
where a
.
is
...
an arbitrary constant.
This integral is shown by a system of curves, each being the arc of a

parabola (see 3.7 above).
The differential equation of the linear type in three variables

x, y and z appears in the form
where P, Q and R are three given functions of x, y and z. This is

often termed the total differential equation. The integral is a relation
between the variables including a single arbitrary constant and
reproducing, on differentiation, the given differential equation.
Since wet can take z as a function of x and y, we have
\ _
~ P *i_W __#
~ '
dxly constant R ty~ \dy/ x constant R

An alternative expression of the differential equation is thus
3z
where the functions / and g are to be associated with ( - and )
/
\ El
- Q\
( respectively.
-p ) The cases where there are more than three
variables are exactly similar. Practical methods of the kinds indi-

cated above serve to integrate particular forms of the total differ-
ential equation.* The following examples illustrate the procedure :
Ex. 1 . The partial elasticities of z as a function of x and y are constants

a and /? respectively. To find z in terms of x and y :
xdz , y dz n
-~-=a and ---=,
z ox zoy
dx dz
dy --
i.e. a + =0.
x y z
So
aj x + j3J.J:^J_-
=^ (6 an arbitrary constant)
J J y J z
and alog#4-]81ogi/-logz = &, i.e. logf ^-J=6.
Writing a =e~
&
,
the integral of the differential equation is
z = axay^.
Ex. 2. xdx + ydy+zdz=Q.
So \xdx + ?/% +\zdz = constant,

I
i.e.
2
J# + %y* + lz* = constant.
* For a method which serves most ordinary
in cases, see Piaggio, Differential
Equations (1920), p. 139.
The integral required is thus represented by the system of spheres
Ex. 3. 2xzdx + 2yzdy - (x 2 + 2

/ ) dz = 0.
" "
Multiplying through by -^ , the equation is thrown into exact form
So
2
and the integral of the equation is x +y2 az, where a is an arbitrary
constant. This integral is represented
" " by a system of surfaces known as
paraboloids (see 11.2-3 above).
16.3 The general integral of a linear differential equation.

Two points arise in considering linear differential equations from
a general and theoretical, as opposed to a practical, angle. One
concerns the question whether an integral of a given equation exists
at all, and the other whether the integral, if it exists, is uniquely
determined or not.
The fundamental " existence "
results, which answer the first
question, can be stated, but not formally proved, here.* The linear
differential equation in two variables,
= or y)>
j|=/(*,
can be shown to possess an integral in all cases where certain con-

ditions are satisfied by the functions P and Q, or by the function /.
These conditions are satisfied in all ordinary cases, e.g. when / is a
function with continuous partial derivatives. In the two-variable
case, therefore, we can assume that all the usual linear differential
equations are integrable.
A complication arises when we pass to cases of three or more
variables. It can be shown that the linear differential equation
Pdx + Qdy+Rdz~Q fails to possess an integral even in quite
* See de la Valise Poussin, Cours d* analyse infinitesimale, Vol. II (5th Ed.
1925), pp. 133 et seq., pp. 292 et seq. ; Piaggio, Differential Equations (1920),
pp. 139 et seq.
ordinary cases. The integral exists only if the functions P, Q and JR,
having continuous partial derivatives, satisfy a certain relation
which is called the integrability condition. We cannot assume that

the equation has an integral at all without first imposing or verifying
the integrability condition. For example, the equation
2xz dx + 2yz dy - (x 2 + y 2 )dz=Q

is found to satisfy the integrability condition its integral has, in
;
fact, been found above. On the other hand, the equation

dx + zdy + 2ydz = 0,
though quite simple in form, fails to satisfy the integrability con-

dition and possesses no integral.
In diagrammatic terms, these important results imply that a set
-~
of tangent lines in the plane Oxy with gradients given by (tx =f(x, y),
where / is a given function with continuous partial derivatives, can

"
always be built up, or integrated ", into a system of curves. But
this isnot true of the corresponding case of a set of tangent planes
in space Oxyz. It may not be possible, even in quite simple cases,
to build up the planes into a system of surfaces, i.e. to find a
set of surfaces to the given tangent planes. It is not easy to
fit
see why the extra dimension spoils the simplicity of the previous
result.
Turning to the question of the uniqueness of integration, the

integral of a linear differential equation in two variables is an
equation in the variables including one arbitrary constant. This
constant must be involved linearly. If, for example, squares or
cubes of the constant appear, then the process of eliminating it must
lead to an equation involving squares or cubes of the derivative ,

(tx
i.e.to a non-linear differential equation. Example 3 of 16.1 illus-

trates this point. It follows that the constant can be separated off
linearly from the variables of the integral,* i.e.
* In
fact, the curve system representing the integral of the differential
equation must be such that one, and only one, curve of the system passes
through each point of the relevant part of the plane Oxy (see 3.7 above).
The integral of the linear differential equation Pdx + Qdy = Q can
X) y = constant,
be written
<f> ( )
where ^ is a function with partial derivatives proportional to P and Q.
Ex. The equation 2 - y 2 - 1 dx + 2xy dy=Q has an

(# ) integral
which can be arranged in the form
= constant.
x
Here </>(x,y)
=
which are proportional to the coefficients of dx and dy in the original

differential equation.
It is
easily seen that the form of the function (x, y) is by no means </>
unique. In fact, if </>(x, y) = constant is one form of the integral of

the given equation and if F is any function, then
F{(j>(x, y)}
=. constant
is form of the integral. For,
also a if <f>(x, y) has a constant value
for certain values of x and ?/, then F{</> (x, y)} also has a constant
value for these x and y. Hence :
If <f> (x, y)
= constant is one integral of P dx + Q dy = 0, then the
general integral is F{<f> (x, y)} constant, where F is an arbitrary
function.
This lack of uniqueness only affects the form of the analytical

expression of the integral it does not correspond to any lack of
;
uniqueness in the integral itself. The curve system which represents

the integral is quite unique, being obtained equally well from
F{<f>(x, y)}
= constant as from <f>(x> y) = constant. The contour map
of the function z=F{<f>(x, y)} does not depend on which form of F
isselected and always gives the integral of our differential equation.
For example, the integral of (x 2 -y 2 -I)dx + 2xydy = Q is shown by
the perfectly definite system of circles (Fig. 97). The analytical form
of the integral can, however, appear in various disguises, e.g.
--
x* + 2/ 2 + l =
x
const.
,
;
-(z - 2
+ 2/~2--
x2
+ l)-2 = ,
const. ;
,
^
log
- --
x
= const.
Exactly similar results hold for the total differential equation in

three variables P dx + Q dy + B dz = 0, provided that the integrability
condition is satisfied. The form of the integral appears
F{<l>(x, y, z)} constant,

where ^ (x, y, z)
= constant is one integral and
an arbitrary F is
function. The partial derivatives of $ are proportional to P, Q and R.

But the integral itself, as represented by a system of surfaces in
space, is quite definite. The result extends, of course, to cases of
more than three variables.
16.4 Simultaneous linear differential equations.

Two linear differential equations in three variables must give rise
to two functional relations between the variables, each including
one arbitrary constant, and represented by two systems of surfaces
in space. An actual example will illustrate the position.
Ex. By differentiating and eliminating the constants in
x+y+z = a and x 2 +y2 =bz,

we heave dx 4- dy + dz = and 2xz dx + 2yz dy - (x 2 + y2 )
dz = 0.
The two functional relations provide the solution of this pair of linear
differential equations. The latter can always be arranged to give the
ratios of dx, dy and dz in terms of x, y and z. Hero, on solving the differ-
ential equations for the differentials, we find
' 2 2
_ x + y + 2xz
dy _ ~~ dz _ x -y
dx~
~~~ 9Z
dx _ -dy dz
'
2z(x-y)
Further, the relations giving the differential equations can be arranged
to give any two of the variables (say, y and z) as functions of the third (x).
In the present case, we find
y= },>(b
2
+ 4ab)-4:bx-4x 2 - \b
and z = =F W(6 4- 4a6) - 4bx - 4x 2 - x + a + \b.

2
These functions involve the two arbitrary constants a and 6. We can

now say that the derivatives of these two functions must correspond to
the values of ~ and v - given from the differential equations.
^ In dia-
dx dx
grammatic terms, the solutions of our differential equations is shown by
a pair of systems of surfaces in space. But the intersection of each pair
of surfaces (one from each system) is a curve in space. The solution is thus
shown by a whole system of curves, not of the plane kind, but of the spatial
kind. In the present case, the curves are parallel plane sections (by
x+y+za) of paraboloid surfaces (a; 2 +y*bz). These curves are in
space ; they do not lie in one and the same plane.
In general, two given linear differential equations in three variables

can be arranged in the form
dx __dy _^dz
P"<p:fl'
where P, Q and R are some functions of x, y and z. So, if we con-
sider y and z as functions of x we can write 9
dy , v , dz
and
~=f(x,y,z) ^ =g(x, y, z),
O R
where the functions / and g are equivalent to the ratios
p and p.
It can be shown that the integral of these equations exists in all
cases subject to the usual conditions concerning the continuity of
the functions P, Q and R, or the functions / and g.* The integral
appears in the form of two relations between the variables, each
involving an arbitrary constant which can be separated off linearly :
<f>(x,y,z)=a and ifj(x, y, z)=b.

As before, the analytical forms of (f>
and iff
are not unique. In fact,
the integral can be expressed in general as
F{<f>(x 9 y, z)}
= constant and G{*f*(x, y, z)}
= constant,
where F and O are any functions. Alternatively, the integral can
be written, on solving the two relations, as giving (say) y and z as
functions of x. Each function involves two arbitrary constants and
its analytical form is not unique.
The integral described can be shown, in diagrammatic terms, as

a pair of systems of surfaces in space, i.e. as a system of curves
in space. This curve system is perfectly definite and not affected
by the lack of uniqueness of the analytical form of the integral.

The nature of this diagrammatic representation can be examined
further.
* See de la Vall6e Poussin,

9
Cours d analyse infinitesimale. Vol. II (6th Ed.
1925), pp. 141 et seq.
P be any point on a curve in space, and draw the tangent line

Let
" "
to the curve at P (Fig. 98). The projection of the curve and its
tangent on to the plane Oxy gives a plane curve passing through a
point Q with a definite tangent, Q being the foot of the perpendicular
from P to Oxy. The plane curve can be regarded as the " shadow "
of the spatial curve on the
plane Oxy (taken as horizon-
tal). The curve in space is
represented by two functions,
y and z as functions of x.
The shadow plane curve is
then represented by the first
of these functions, y as a
function of x, and the grad-
ient of its tangent at Q is
indicated by the derivative
~
Ct'X
. In the same way, the
" "
space curve has a shadow FIG. 98.
on the plane Oxz, a curve

representing z as a function of x and with tangent gradient given
by -j-
. The curve in space and its tangent at P can thus be
ctx
" "
investigated by means of two shadow curves and their tangents.
Now, the given differential equations
dy /./ dz
v
=/(#> y> *>) and = g (x, y, z) -
^ ^
-~ and at each point P(x, y, of space, and so
give values of z)
CLX (tX
"
determine two shadow " tangents, one in the plane Oxy and the
" "
other in Oxz. These shadows can be combined to give a definite
tangent line in space at P. The integral of the differential equations
isthen obtained by building up the tangent lines into a system of
curves in space, one curve passing through each point of space.
The practical process of integrating a pair of differential equations
of linear form is often extremely difficult.* The following examples
illustrate the practical integration.
* See Piaggio, Differential Equations (1920), pp. 133 et seq.
~ " *
x y z
One of these equations

^ can be written - =- or -~=-.
dx x ydx x
On integrating this by ordinary methods (16.2 above), we have
y =a# (a being an arbitrary constant).
Similarly, from the other equation -- = - we find ,

ax x
z = bx (b being an arbitrary constant).
The integral of the pair of differential equations is thus
y=ax and z = bx t
shown by a system of straight lines in space, all passing through O.
X ' '
==
2r 2
'
^2xz^2yz 'x
+y
It can be verified that these differential equations can be arranged in
the form (amongst many others)
-=0 and xdx+y dy + 2zdz=Q.

We can now integrate at once :
fdx Cdy = constant and f T f T f 7

= constant,
I
- - I x dx + y dy I -f 2 z dz
1
J x y J J J J
i .e.
log x - =
log y constant and lx 2 + \y 2 + z 2 = constant.
The integral of the equations can then be written
y=ax and x2 + y 2 -f 2z 2 = 6,
tt.,a and 6 are arbitrary constants. The two surface systems are now
= ax) and a system " "
a systetp of planes (y of surfaces known as ellipsoids
(x
2
+ y2 +~2z 2 = 6). The integral is shown by a set of plane sections of a
"
system of ellipsoids ".
Ex .3.
^ =
~ dy
= dz
.
2z (x - y)
. 2
x 2
+y 2
-f 2yz x +y2 + 2xz
It can be verified, in this case, that the two relations
dx+dy + dz=Q and 2xz dx+2yz dy - (x2 +y2 )dz =0

hold. The integral of the first relation is found at once as
x+y+z=a
and that of the second (16.2 above) as
where a and 6 are arbitrary constants. The complete integral is thus

obtained in the form already considered above.
16.5 Orthogonal curve and surface systems.

The linear differential equations in two variables
=0 and =
give two tangent lines at each point (x, y) of the plane Oxy, one with
/
p\
gradient (to Ox) equal to (
-^) and the other with gradient equal
to
Q Since the product of these gradients is - the tangent lines
^ 1,
are always perpendicular. Hence, the two curve systems repre-

senting the integrals of the equations must be such that, wherever
a curve of one system cuts a curve of the other system, tljef'two

curves are at right angles. Such curve systems are described as
orthogonal systems. An example of differential equations of the
orthogonal type is given by y dx -f- x dy =

and xdx-ydyÔ. It is
easily seen that the integrals are respectively and x 2 ~y 2 =b,xya
where a and b are arbitrary constants. The curve system, in each
case, is a system of rectangular hyperbolas, one system being
obtained from the other by rotation through 45. Fig. 99 shows
certain curves of the systems in the positive quadrant of Oxy and
clearly indicates the orthogonal property of the systems.
The total differential equation in three variables
gives a tangent plane at each point (x, y, z) of space, and its gradients
are determined by the ratios of the functions P, Q and It. The pair
of linear differential equations
gives a tangent line at each point of space and its direction is deter-
mined by the functions P, Q and J?. It is a well-known result that
this tangent line isperpendicular to the previous tangent plane at
the same point. This can be verified by taking projections on the
co-ordinate planes. The tangent plane contains a horizontal line
Oxy) with gradient referred to Ox equal to dx = -^

(parallel to
-~ -
y
This projects directly on to the plane Oxy. The tangent line, as we
" "
have seen, has a projection or shadow on the Oxy plane with
gradient to Ox equal to -=^ =~ . These two lines in Oxy are perpen-

ax JL
dicular. A similar result holds for the projections on to one of the

other co-ordinate planes, and it is then clear that the tangent line in
space must be perpendicular to the corresponding tangent plane.

The total differential equation (if it is integrable) is shown by a
system of surfaces in space and the pair of linear differential equa-

tions by a system of curves These systems must here
in space.
possess the property that, wherever one of the surfaces is cut by one
of the curves, the intersection is at right angles. The surfaces and
curves are said to form orthogonal systems. For example, the total
differential equation xdx + ydy + zdz = has an integral which is
the system of concentric spheres (centre at 0) x* + y 2 + z 2 =c. The
pair of linear equations
^ =_^ = ~ has been found to integrate to
x y z
y=ax and z=bx, a system of straight lines all passing through 0.

The lines are the diameters of the spheres and the two systems are
clearly orthogonal.
16.6 Other differential equations.

There are many types of differential equations more complicated
than the simple linear equations so far considered. For example,
fdy\ 2 dy
is a non-linear equation involving only a
y(-r) -2#-p4-t/=0
first-order derivative and its integral (see 16.1) is /
2
-2ao; + a 2 =0,
where a is an arbitrary constant. Again, by differentiating the
relation y=zax +- twice and eliminating the parameters a and 6,
we obtain the differential equation x 2 ~r\ + x-~--y = Q which

ax
involves derivatives of the first and second orders. The integral of
the equation is the relation above and includes two arbitrary
constants.* The practical methods of integrating such differential
equations are not easy. The following general problem can be taken
as an illustration .f
" "
Ex. The dog and his master problem.
A man stands at a point and his dog at a point A in an open field.
The man begins to walk at a uniform
pace along a path at right angles
to OA, while the dog runs at a
uniform pace (greater than that of
the man) always in the direction
pointing to the position of the man.
When does the dog catch up with
the man ?
Take co-ordinate axes as shown

M
in Fig. 100, Oy being the direction of
the path taken by the man. Choose
the unit of distance as OA, so that A O
is (1, 0).Let the ratio of the man's FIG. 100.
speed to that of the dog bo u, where
u<l. After a certain time, suppose that the dog is at the point P(x, y)
and the man at Q on Oy. If the path taken by the dog is as shown with
equation y=f(x), then PQ is the tangent to the path at P with gradient
dy MQ_ OQ-y~'
dx MP x
So
*
This suggests an extension of the result of 16.1 above. It can, in fact, be
shown that, if the nth derivative is that of the highest order in a differential
equation in two variables, then the integral of the equation is a relation
between the variables which includes n arbitrary constants.
j*
For an account of this and similar problems, see Osgood, Advanced
Calculus (1925), p. 332 et seq.
Let s =arc AP of the dog's path
=- (distance travelled by man) = U = U
-
(y-x-/-)*
U \ dX/
Now s is a decreasing function of x with negative derivative
ds_ 1 /Ay d 2y dy\^ x d 2y

dx u\dx dx2 dxJ udx 2
The increment ds, being compounded of perpendicular increments dx
and dy along the dog's path, is given by ds 2 =dx 2 +dy 2 .
So
taking the negative root since the derivative is negative. Hence,
The equation, / =/(#), of the dog's path is thus the integral of

2
dy\ ^
iL
dx/)
^^
a differential equation of the second order. To integrate, write
so that -. Then
f dp = u fdx ,
.
i.e. I -, -
-
I h constant.
+*/l+p ) = t* log a; + const., +p 2 ~axu

2
So log(# i.e. p -h\/l .
Since p -j-
==
(gradient of path of dog at ^4)=0 when x l9 we find,
on substitution, that the constant a*=l. Hence,

M 2
a: =N/l +j9 +^>.
' ' " ^
SO 3.-!* -,
and
Hence, ~
1"" 14
1 /f u \ / cc1+u \
and 2/ =1; ( x I da; - f
I or"* da; ) -f constant =-1 (
ic
) -f- 6.
2\J J / 2\l+u l-u/
At A 2/=0 when x = l, and so we find &=-

y
g.
The equation of the - at
dog's path is thus

' '
2(1 +u) 2(T-u) :T^

The dog catches up with the man when x 0, in which case
The man walks a certain multiple ( ^ of the distance OA before the dog
\ l u^ /)
catches up with him. For example, if the dog's speed is twice that of the
man (u = \ ), the equation of the dog's path is y J (\/# 3 - 3\/# + 2) and the
man walks only two-thirds of the distance OA before the dog reaches him.
Similar complications arise in equations which involve the differ-
entials of three or more variables. Further, when the number of
variables exceeds two, we can
define equations of a rather different
type involving partial derivatives instead of differentials. The

simplest form of this type of equation occurs when the partial
derivatives are related linearly as in the following example :
_. dz dz
Ex. x ^~ J
-f?/^ = z.
dx dy
The equation is integrated when we have expressed z explicitly in
terms of x and y. The following trick is needed in the particular case
considered. %
Write u = so that
x
dz
From the original differential equation, we have z - x -~ - = _

y
ox **y
a
So
-
dzxdy -ydx =
dw-3 ^-3-
x2
dy
Hence the variation of u as a function of x and y depends only on the
variation of - and not on the separate variations of x and

, y. It follows
x
y
that u is some function of - only, and so
x
Z=i
where <f>
is any function whatever. This is the integral of the given
2E M.A.
differential equation. It is to be noticed that the arbitrary element in
the integral consists, not of an arbitrary constant, but of an arbitrary
function. This is a special feature of equations in partial derivatives.
The form of the integral shows that z is a linear homogeneous function
of x and y (see 12.8 above). The integral, in fact, provides the converse
of Euler's Theorem :
Ifx ~-+2/~-=z at all points (x, v), then z is some linear and
ox oy
homogeneous function of x and y.
As an application of the result, suppose that a good is produced by X

using two factors A and B
according to the production function x =/(#, 6)
and that the product is always divided (without excess or deficit) according
to the marginal productivity law. Hence, each unit of the factor A
receives a return equal to the marginal product ~- and similarly for the
factor B. It follows that
dx ,dx
a^-+b^r=x
da do
for all combinations of the factors. This is only possible, as our present
result shows, if the production function is linear and homogeneous.
16.7 Dynamic forms of demand and supply functions.

A market for a good X is made up of a competitive group of buyers
and a competitive group of sellers. Under static conditions, assum-
ing that the prices of other goods are fixed, the demand of the
buyers can be represented by a function x = <f>(p) and the supply of
the seller by a function xf(p), where p denotes the uniform price
of the good. A position of static equilibrium is said to be determined
at a price which equates demand and supply <f>(p)=f(p)- This
:
is an equation in one variable with a definite number of solutions,
each corresponding to a possible equilibrium price.

The static problem implicitly assumes that there is no " specu-
lative
"
demand or supply and that there is no " time-lag " in
production on the supply side. Our problem can be extended by
dropping these assumptions in turn. The market price of is now X
taken as varying over time, the current price being a function of t,
the time (in years or some other unit) since a fixed base period. We
have to find, under various assumed conditions, the form of the
price function p(t). The

problem gives a definite price and
static
" "
equilibrium position the extended dynamic
; problem must give
the course of prices over time and the path of a moving equilibrium.
The demand of the market (as a rate per unit <^ time) at the
current time t is taken as depending on the price p (t) and on certain
speculative elements, including the buyers' view of the future
course of prices. We consider here only the simple case in which
" "
the buyers' foresight is confined to an allowance for the rate at
which the price is changing at the time concerned, demand depending

on whether prices are rising or falling and at what rate they are
changing. We now have demand as dependent on p(t) and its
derivative p'(t) :
x = <f){p (t), p'(t)}, the dynamic demand function.

In the same way, assuming the same kind of speculation (but no
time-lag) on the supply side, we can write
x =f{p (t), p'(t)} , the dynamic supply function.
For equilibrium over time, the price function p (t) must be such that
demand and supply are equal at all times t :
This is a differential equation including the first derivative of p

with respect to t, and its integral gives p as a function of t involving
one arbitrary constant. The latter can be determined in terms of
the price ruling in the base period ( = 0).
p Hence, given the
the course of prices over time is determined.
initial price,
The problem can be illustrated by taking the particular case

where the demand and supply functions are linear. Under static
conditions, we write demand as xap+band supply as x = ap+{i.
" "
In the normal case, the constants a and j8 are negative and 6 and
a are positive. For equilibrium, demand equals supply, i.e.
giving one equilibrium price p =a - -

a
In the simple speculative case considered, we write :
x = ap (t) -f b + cp (t)
r
Demand .
Supply x = ocp (t) + ft + yp'(t),

where a, 6, a and j8 are constants with the signs indicated above and
where c and y are new constants which can be taken as positive.
This implies that a rising price, other things being equal, stimulates
both demand and production. It is possible, however, to take y as
if a rising price causes sellers to hold back their supplies.
negative
For moving equilibrium, the price function p (t) must be such that
a
ap (t) + b+ cp'(t) =*p(t)+p + yp'(t),
i.e. j/():= {p(t)~p} where p-

^
Write q(t)=vlt)-p and X=
c-y
mr - AQ
. 1
Ihen 3
r
(0~A5 r
(^), i.e. ~-=A.
This familiar differential equation has integral q =
So p(t)p+Ae.
Putting = p=p (the initial price), we find A
and p^-p. The
course of prices over time is thus uniquely given by
where p= a-a- ,
A=- -
c-y
and p Q is the initial price.
The important term in the expression for p (t) is eAe This term .
increases and tends to infinity, or decreases and tends to zero,

-
according as A is positive or negative. Since we have taken (a a)
as positive, the sign of A is governed by that of (c-y). Hence,
(1) If y <c, A is positive and the price steadily diverges from

the static equilibrium value p as time goes on.
(2) If y>c, negative and the price steadily approaches

A is
the static equilibrium value p as time goes on.
A stable course of prices is thus only possible if y>c, i.e. if the
speculative element in supply is stronger than that in demand.

If, in addition to the speculative element, there is a time-lag of
constant length years in production, then the supply of is X

dependent on the price and the price changes at the earlier time
(t
- 0) rather than at the current time t. The current supply was
arranged for years ago at the beginning of the production process
and was subject to the price then ruling. The general condition for
moving equilibrium is now
and with linear demand and supply functions
Such differential equations can be integrated only with the aid of

66
complex numbers ", entities designed for the description of
variables which oscillate in value. But the general is found
solution
to be very similar to that of one special case which we can easily
solve.*
When there is a production time-lag but no speculation, the
equation for p (t) ceases to be differential and takes the form
.e.
Here, linear demand and supply functions are assumed and p has
the same value as before. If the course of prices over the initial
period from t = to t = 6 is given, then the equation above gives the

price p(t) at any subsequent time t. In fact, measuring all prices
from the level p, the current price is just a constant multiple of the
price 6 years ago. The multiple is the negative number - . There
are three c&ses to consider. Suppose, for simplicity, that the prices
in the period (0, 0) oscillate about p, being as much below p at
t as above p at 0. Then :
(1) If a and a are numerically equal, then the price variation in

the period (0, 20) completes the oscillation of the period (0, 0), the
price at
= 20 =
being the same as that at t 0. In successive periods
t
of 20 years, the same cycle of prices is repeated. The course of price

over time is illustrated in diagram A of Fig. 101.
(2) If a> a numerically, then the ratio - is negative and numeri-

a
cally less than unity. The price again oscillates, describing a cycle
* On the problems considered hero, see Tinbergen, Utilisation des equations
fonctionelles, Econometrica, 1933, and Theis, A quantitative theory of industrial
fluctuations, Journal of Political Economy, 1933. The first work in this field
is due to Roos and Evans.
in every period of 20 years, but now the extent of the oscillation
diminishes as time goes on. The oscillations are "damped" as
shown in diagram B of Fig. 101.
(3) If a<a numerically, the case is similar to the previous one
O 20 30 40 50 t O 20 30 40 50 t
20 30 40 50 t
FIG. 101.
except that the oscillations increase in amplitude as time goes on.

" "
This case of explosive oscillations in price is illustrated by
diagram of Fig. 101.
16.8 The general theory of consumers' choice.

The analysis of the demand of an individual for consumers' goods
has proceeded, so far, on the assumption that the individual's scale
of preferences can be represented by an indifference map, i.e. by a
utility function (measurable or not). This assumption can now be
scrutinised.
Suppose that x and y are the individual's current purchases of

two goods X
and Y. It is now assumed simply that the individual
has a definite scale of preferences for small changes in these pur-
chases. He can distinguish that small increase Ay in his purchase of
Y which just compensates (i.e. leaves him indifferent) for a given
small decrease - Ax) in his purchase of X.
( Larger increases than
Ay are preferred by the individual whereas smaller increases leave
him in a position worse than before. Hence, for an indifferent
change, Ay is a definite multiple of (-Ax). Taking the limit for
smaller and smaller changes, the assumption defines the differential
dy as a definite multiple of ( -dx). The multiple (which is positive)
is defined for each initial set of purchases (x, y) we care to select and
varies as these purchases vary. The multiple is, in fact, a function

of the variables x and y and we can write it as R (x, y). Our assump-
tion thus gives the differential equation for indifferent changes :
or R(x, y)dx + dy = 0.
-^=B(x,y)
The function E
takes only positive values. It is further assumed
that the scale of preferences is continuous in the sense that has R
continuous partial derivatives.
The differential equation above can be integrated in all cases
(16.3 above) to give the general integral
F{<f>(x, y)}= constant.

Here F is any function and <f> (x, y) is a function such that
In diagrammatic terms, our assumption corresponds to the defini-

fjfi
" "
tion of a definite indifference direction given by --~B(x
dx
i y)
at each point
(x, y) of the plane Oxy. The indifference directions can
then be integrated to give a whole system of indifference curves
defined by F{(f>(x, y)}~ constant. The situation is exactly as
before but based on a different assumption. do not assume aWe
complete scale of preferences but a scale for small changes from any
given set of purchases. Finally, the utility function index
with a constant value along any one indifference curve, follows from
our new assumption.
From 13.8 above, the function R(x, y) here defined is to be inter-
preted as the marginal rate of substitution of Y for X. Previously,
we deduced this concept from an assumed utility function. The
position is now reversed the marginal rate of substitution
;
is the
fundamental concept and the indifference map and utility function
are deduced from it. This is an important change of approach.
The general case where any number of goods appear in the con-
sumer's budget is sufficiently represented by the case of three goods
X, Y and Z. The assumption
to be adopted asserts that the indi-
vidual has a scale of preferences for small changes from a given set
of purchases (x, y } z) so that he can distinguish that small increase
Ax in the purchase of X which compensates for given small decreases
(-Ay) and (-Az) in Y and Z. The individual prefers a larger
increase than Ax, but rejects a smaller increase. Taking smaller and
smaller changes and writing differentials for the increments, we have
dx as a multiple of ( -dy) plus
a multiple of ( -dz). The multiples
are positive and functions of the purchases (x, y, z). So
dx = R1 (x, y, z) (~dy) + R 2 (x, y, z)(- dz)
or dx + Rl (x ) y, z)dy + R 2 (x, y, z)dz~0
isthe differential equation for indifferent changes in purchases. The
functions S l and R 2 take positive values and are assumed to have
continuous partial derivatives.
The interpretation of Rl and R 2 is not difficult. The ratio of
compensating increments in the purchases of X and Y when the
purchase of Z remains unaltered can be written
dx\
= Ri(x,y,z).
-7-)
il
y' z constant
The function R
thus represents, at the purchases (x, y, z), the
l
marginal rate of substitution of for Y X

In the same way JR 2 repre- .
sents the marginal rate of substitution of for Z. Our assumption X

isequivalent to the definition of two marginal rates of substitution
as describing the individual's scale of preferences. In diagrammatic
terms, we have, at each point of Oxyz space, an indifference plane
given by dx-{-R1 dy + R 2 dz~0 and with gradients (-Ri) and (-R 2 )
in the Oy and Oz directions. The plane includes all the indifferent
directions of change from the purchases (x, y, z).
The equation dx + R^dy + R 2 dz = Q need not possess an
differential
integral (16.3 above). In general, therefore, we cannot integrate

the set of indifference planes into a complete set of indifference
surfaces, and we cannot assume that any utility function exists.

The assumption of a scale of preferences for small changes of pur-
chases does not imply that a complete scale of preferences exists.
The consumer can discriminate between small changes from his
established purchases but need not be able to discriminate between
widely different sets of purchases. The assumption of the marginal

rates of substitution as fundamental has shown up a new possibility.
The differential equation is integrable only in the special case
where R l and E 2 satisfy the integrability condition (16.3 above) :
dR 2 --
BR,
In this case we can write an integral in the form
F{<f> (x .
y .
z)}
= constant,
where F is any function and <f)(x, y, z) a function such that
y and $ z =R 2 (x,y,z).
^Ê^(x
<Px
9 9 z)
<Px
The integral is shown by a system of indifference surfaces touching

the given indifference planes and the function u F{<f)(x, y, z)} can
be taken as a function index of utility. The given marginal rates
of substitution are equal to the ratios of the partial derivatives of
u in any one of its forms. It must be remembered now that the
indifference map and utility function are deduced from the marginal
rates of substitution and not conversely.
In developing the theory of individual demand for three or more
goods, therefore, we proceed from the assumption of marginal rates
of substitution. As a special case, we can then add the integrability
assumption, take an indifference map and a function index of
utility and deduce certain results which hold only in this case. We
have a general theory and a special (integrability case) theory of
demand, a distinction which will be made clear later (19.7-8 below).
The results of 16.5 above are of direct application to the present
problem. The differential equations

1
-==-- define a tangent
Jttj -1*2
direction at the point (x, y, z) in space, a direction perpendicular to

the indifference plane given by dx-{-Rl dy + 2 dz = Q. If the indi- B
vidual increases his purchases from the given set (#, y, z) in the way
indicated by this direction, these are the increases he prefers above
all others. The direction can thus be termed the individual's
preference direction at (x, y, z).* The values of B^ and JR a at any

point thus indicate, not only the marginal rates of substitution for
indifferent changes, but also the ratios of the increases in purchases
which are most preferred by the individual.
Since R and R 2 have continuous partial derivatives, the differ-
ential equations for the preference direction can be integrated to
give a system of curves in space, the lines of preference of the indi-

vidual (16.5 above). One such curve passes through each point of
space and indicates the most preferred direction of change from the
point. The complete lines of preference exist even when there is no
indifference But, in the integrability case where indifference
map.
surfaces exist, it follows that the lines of preference and the in-
difference surfaces form orthogonal systems. A line of preference
cuts an indifference surface at right angles and indicates the
" "
quickest way from one indifference surface to the next, i.e. the
direction in which the individual's utility level is increased most
rapidly.
EXAMPLES XVI
Differential equations
1. Draw a graph showing a sufficient number of the tangent directions
-r- =- to indicate the curve system which is the integral of the equation.
Draw a similar graph of -~ -- Check by integrating the equations

.
by
dx x
" "
the variables separate method (see 16.5 above).
2. Illustrate graphically the differential equation
~
ax
=
x
and its
integral. From the equation, prove that each curve of the integral system
has a maximum point at the same value of x.
3. If y= 1 +
x
=- ,
i
show that (y
~ l)dx + (x
-
l)dy = 0. Conversely, show
"
that the differential equation has the integral shown (using the variables
"
separate method). What are the curves represented?
* Preferencedirections were introduced by Edgeworth, Mathematical
Psychics (1881, reprinted 1932) and by Irving Fisher, Mathematical investi-
gations in the theory of value and prices (1892, reprinted 1925). In the case of
two goods, indifference curves are defined by the relation ~ = -R(x, y) and
preference lines
~= _
dij
by the relation ax
1
. \\re
*^
have, in all cases, orthogonal

R(x, y)
systems of indifference curves and lines of preference. Either system defines
the individual's scale of preferences.
4. Show that y a(x a) -f is a certain system of straight lines and eliminate
a to obtain the differential equation of the system.
5. Obtain a differential equation of the second order, representing the

curve system with equation ax 2 4- by* = 1 (a and b parameters).
6. The equation y
ax* + b represents all parabolas with axis along Oy.
Find a equation to represent the curve system. Of what order
differential
is the equation? Consider, in the same way, the equation y = ax* + bx + c,
representing all parabolas with axis parallel to Oy.
7. Derive linear differential equations by eliminating a from the relation
X
(t
x + y~ ae*' and from the relation z =e v .
8. Find an integral of each of the differential equations :
~
9. If a is a given constant, obtain an integral of
cfry
= a-;c
dx x
Show that the curve system obtained can be represented in either of the forms
ye*x~* constant and x - a log x -f log y constant.
10. Show that y = aexn is an integral of

~
dx nyx
n" 1 .
11. Show that the integral of (y*-l)dx- 2dy = Q is y = -1 ^- .
CMJ
12. Integrate (2x + y)dx+(x + 2y)dy = by showing that the left-hand side
of the equation is an exact differential.
13. Show that ydx+ (2x + y)dy = Q becomes exact on multiplying through
by 3y. Hence integrate the equation.
__
14. Find the partial derivatives of and hence solve the differential
^
equation
y(2x - y )dx - x(x* 2y*)dy = 0.
3 8 -
-,* dx dy dz .
-r
15. Integrate
,
-f + = 0.
16. Show that a differential equation of the form
is always integrable. How is the integral obtained?

17. Which of the differential equations
Q; dx-zdy-
= Q and y(y + z)dx + x(x-z)dy + x(x + y)dz=Q
are integrable and which not ?
18. Integrate 2xdx + zdy + y dz = Q by showing that the left-hand side is
an exact differential.
x + *'
19. Find the partial
r derivatives of with respect to x, y and z.
x+y+z
Hence integrate (y -f s) dx + a?" dy
a
+ x*dz- 0.
20. Find the integral of the simultaneous differential equations
z dz.
Examine this integral and that of example 15 above in diagrammatic terms

and show that the curve and surface systems concerned are orthogonal.
Economic applications of differential equations

21. It is known that the elasticity of a demand law x </>(p) is
of the form
(a-bp) where a and 6 arc given constants. By integrating a differential
a
equation, show that the law is x p e~*>(v+ \ c being an arbitrary constant.
c
22. If H is the total cost of an output x, it is known that marginal cost
( )
always equals average cost f ) .
Integrate and show that 17 is a
fixed multiple of x, i.e. that average cost is constant.
23. If y is the number of incomes of r, it is found that y decreases as

x increases according to the law ~- - m - where m is a given constant.
ax x
Integrate and show that the dependence of y on a? is y =
a ~ .
24. The price of tea, initially 3s. per lb., is p(t) pence per Ib. after t weeks.
The demand 120 - 2p + 5 -~ and the supply x = 3p - 30 + 50 -f- thousand

is x
at at
Ibs. per week. Show that, for demand always to equal supply, the price of tea
must vary over time according to tho law p 30 + 6e-&. What are the prices
after 10 and after 50 weeks? Draw a graph to show the approach of the price
to the equilibrium value of 2s. 6d. per lb.
26. The demand (per unit of time) for a good is x ap + b and the supply
g oip + p whore p is the price.If there is (e.g.) an excess of demand over
supply in any period, it is assumed that tho price changes to decrease the
excess at a rate proportional to the excess.
If p(t) is the price at time t, show that the assumption implies that
t)
where k is some positive constant. Deduce that p (t) is
-
given by ~r + k (p p)
~
p= -
b B
where If p is the initial price (t = 0),
- show that p (t) = p + (p - p) e~kt
a a
and that tho price tends to the equilibrium value p. See Evans, Mathematical
Introduction to Economics (1930), p. 48.
26. An individual's preference scale for two goods X and Y is defined
by tho marginal rate of substitution of Y for X, R = ~ ~-. &

Show that
u = (x - a) 2 + (y - b) 2 is one form of the utility function. 2/
27. If R= -
-
px + a
is the marginal rate of substitution of Y for X, show that
one form of the individual's utility function is u (x + a)*(y + b)& whore a, b,
a and ft are given constants.
28. The marginal

fe
rate of substitution of Y for X R=is ~ where
+ fry + y ocx
tho coefficients are given constants. If a 6, show that one form of the
utility function of tho individual is quadratic in x and y. If a 6, show that =
tho utility function noed not be of this form. Illustrate by considering the
particular cases of the two previous examples.
" "
29. Throe goods X, Y and Z are said to be independent in consumption
ifthe ratios of the individual's marginal rates of substitution 1 JR 1 R 2 are : :
of the form </>(&) $(y) x( z ) where <f>,

: : and x are functions of tho single
i/i
variables named. Show that the integrability condition is always satisfied

and indicate the form of the utility function. Illustrate when
1 iîR^fr-a): (y-b):(z-c).
30. Tho marginal rates of substitution of for X Y and Z are known to be
ratios of linear expressions in x, y and z, so that
(djX -t- b y + c^z + djdx + (a z x + b.2 y + c 2 2 + d 2 )dy + (ajc + b 3 y + c 3 z 4- djdz

-=
is the differential equation of the indifference plane. What is tho condition for
the existence of indifference surfaces? Show that the condition is satisfied
if foj ~a 2 c 2
,
6 3 and a 3 = c,. In this case, show that the utility function can
bo written as a quadratic in a;, y arid z.
CHAPTER XVII
EXPANSIONS, TAYLOR'S SERIES AND HIGHER ORDER
DIFFERENTIALS
17.1 Limits and infinite series.
WE have seen (4.1above) that there are two broad classes of number
sequences. A sequence either tends or does not tend to a finite
limit,the second -case including sequences tending to infinity and
sequences which oscillate. For example,
1
2>
3.
4>
2.
8> 16>
5.
^.1 L
1, 4, 9, 16, ...->>.
- 2
The nth members of these sequences are f 1
~J and n respectively
and the results can be written in the alternative forms
1 ^l as n-><*> and w 2 ->o> as n->oo .
2n
Here n represents any positive integer.

A number sequence can be treated in a slightly different way by
forming from it a second sequence, a sequence of differences. The
nth member of the new sequence is the difference between the nth
and the (n
- l)th members of the old sequence. The new sequence is
written down with plus signs between successive members (to indi-
cate our intention to add them together), continued indefinitely and
called an infinite series. For example, the two sequences above give
rise to the infinite series
111 1
1 + 3 + 5 + .+(2n-l) + ...
.. ,
where the expression in n represents the nth term of the series in

EXPANSIONS, TAYLOR'S SERIES 447
each case (n being a positive integer). From the definition, the sum
of n terms of the infinite series equals the nth member of the original
So, for any positive integral value of n,
sequence.
111 11
+ .. .+(2ft-l)=ft 2 .
In general, if /(ft) is member of a given sequence, then the

the nth
infinite series of differences sum of n terms equal to /(ft).
has
If the original sequence has a finite limit, then the sum of n terms
of the derived series has the same limit. The series is said to be
convergent and the limit of the sumsum to infinity. If the
is the
original sequence has no limit, neither has the sum of n terms of the
series. The series is said to be divergent and there is no sum to
infinity.For example, the series first written above is convergent
with unity as the sum to infinity the second series is divergent,
;
the sum of n terms tending to infinity with n.

The limit of any sequence (if it exists) can thus be written as the
sum to infinity of a convergent infinite series. Any member of the
sequence and the sum of any number of terms of the series can then
serve as an approximate value of the limit, a fact which provides one
of the main uses of a convergent series. If A is the sum to infinity of
a certain series, then A is given by the limit of the sum of ft terms and
found approximately by adding together a sufficient number of
terms of the series. This may be the simplest way of getting an
approximate value of A if the latter is a complicated expression.
Again, the sum to infinity provides an approximate value of the sum

of a large number of terms of the series. The sum of (say) 100 terms
of the series (|-f i + i-f-...) is not easily calculated. But we know
that the sum to infinity is 1 and this can be taken as the approximate
value of the sum of the 100 terms.
Conversely, if we are given a certain infinite series, we can form
a sequence of sums, the nth member of the sequence being the sum
of ft terms of the series. By definition, the given series is convergent
if the derived sequence tends to a finite limit.
Two important cases of series can be considered in detail. A
geometric series consists of terms which increase (or decrease)
from one term to the next by a given factor, called the common
ratio. In general, if the first term is a and the common ratio r,
the series is
2 - 3 -
-f . . .
It is shown, in elementary algebra, that the sum of n terms is
The derived sequence is thus
n
If r numerically less than unity, the expression r becomes smaller
is
and tends to zero as n increases indefinitely. Hence,

a a a
---- ---- rn -> --
S - as
1-r 1 -r l-r
The geometric series is convergent with sum to infinity -- , pro-
vided that r numerically less than one. In other cases, the expression
is
rn does not tend to zero as n increases and the geometric series is

not convergent. The sum of n terms either tends to infinity or
oscillates.
_ 1 1 1 1
_,
Ex.1.
g
+ +- + ...
5
+- + ....
Here a= |, r \ and 8n - - ^2
"~
= 1 - ^->1 as TI~>OO .
** I 2 ^
The series is convergent with sum to infinity 1 (as obtained above).
Ex. 2. The recurring decimal 0-1 is a short way of writing the sum
to infinity of the series
+Jfa +J ^.tf + ... ^ .
1
This is a convergent geometric series with a =r j ^. The sum to infinity
is
Y ~~ Q* Hence, the decimal form of the fraction is (M.
A I /
Ex. 3. An income stream of a starts next year and continues for

?iyears. Reckoning interest yearly at lOOr per cent, per year, the present
value of the stream (see 9-7 above) is
a a a a
+ ______
/i ..\o
I
i
__________
/i ..va
\
'
L
^^
i
_
l+r
. . ,
a V1W af / 1 \1
1+r 1 r\ VI +r/ /
ÎTr
If the income stream goes on
for ever, the present value is obtained by
letting n->oo i.e. it is,

the sum to infinity of the convergent geometric
series above. Hence, the present value of a per year for ever, at interest
at lOOr per cent, reckoned yearly, is - For example, the present value
of 3 per year for ever is 100 if interest is reckoned yearly at 3 per cent.
per year.
An arithmetic series consists of terms which increase (or decrease)
from one term to the next by a given amount, called the common
difference. If the first term is a and the common difference d, the
series appears
It is easily shown that the sum of n terms is
not convergent and

f
Since S n -> oo as n-+<x> ,

the arithmetic series is
never possesses a sum to infinity.
Ex. 1. 1+2+3 + -+n + ... .. .
Here Sn = \n(n + 1) the sum of the first n positive integers.
Ex.2. l+3+5 + ...+(2n-l) + ... .
Here 8n = \n{% + (n - 1)2} =n 2 the sum of the first n odd positive

integers.
17.2 The expansion of a function of one variable (Taylor's series).
By a process of long division, we find
xn+l
1 -X I -X
If x has a value numerically less than one, the term containing x n+l
decreases and tends to zero as n increases indefinitely. Hence,
1 -x
if x is numerically less than one. We have thus expressed -
i
- x
as
the sum to infinity of a convergent infinite series. The result can

be checked in this case. The series written is a geometric series
with common ratio x (<1) the series is convergent with sum to
-
;
infinity equal to
- the formula of
X (by
~~"
17.1).
A
IP M.A.
The advantage of the series expression for -
l
- x
is clear. If x is
given a small value, then, instead of evaluating

-
j.
- x by long
division,
we find its approximate value (to any required degree of accuracy)

by adding a sufficient number of terms of the series. For example,
if x=
30
-
1
-= 1
X
1+ x + x* + ... = 1+0-033334-0-00111+0-00004 + ...
= 1-03448.
Hence to four decimal places, ff = 1-0345, as
can be verified by long
division. The
term of the series is too small to affect the
fifth
approximation and is not included. Again, if a; is a small number,

then to a degree of accuracy represented by x*
1 -x
i.e. for approximate purposes the expression
-
j.
-~
x
can be replaced
by the simpler quadratic expression (l+# + # ).

2
The problem can be generalised. Given a single-valued function

a convergent infinite series, the successive
/(#), it is desired to find
terms of which involve increasing powers of x, with f(x) as its sum
to infinity. In symbols, we wish to write
f(x)=a Q + a^x + a zx + a 9x +
2 3
(to infinity),
. . .
where the a's are certain constants. The problem is a little complex
since we have first to see whether the process indicated can be
carried out at all, then to determine for what values of x it is
valid and finally to allot the appropriate values to the constant
coefficients.
If the process is possible, the series obtained is called the expansion
of f(x) as a power series, i.e. as a series in ascending powers of x.
The particular case above indicates the uses of the expansion when
obtained. For a definite value of x, the accurate value of f(x) is
obtainable only by substituting directly in f(x) or by finding the
limit of the sum of n terms of the series. But, if approximate results
suffice, we need only add a limited number 'of terms of the expansion,
It is possible, for example, to replace f(x) approximately by a quad-
ratio or cubic in x when x is small, a great simplification if f(x) is at

all complicated.*
It is assumed, in the following, that f(x) has continuous deriva-
tives up to any desired order at the points concerned. We start
with a simple case where the problem is really solved in advance but
which will indicate the more general method. Suppose that
n
f(x)=a Q +a1x + a 2x + a zx* + ...+ anx
2
,
a given polynomial of the nth degree in which the coefficients have

known values. By successive derivation,
f'(x) =a + 2a 2x + 3a 3x* + ...+ nanxn ~l ,
f"(x)=2a 2 + 6a 3x + ... + n(n - l)an xn ~ 2 ,
=n(n-l)(n-2)...3.2. lan
Putting x 0,
and n
f( )(Q)n(n- l)(n-2) ...3.2. la n .
A convenient algebraic notation can be introduced here. The

product of the positive integers from 1 to n inclusive is called n
factorial and written In. Hence, from the results above,
i i i
*o=/(0), <h=J
The original polynomial can thus be written in the form

T3
f(x) =/(0) +f(0)x +/"(0)
/7<2
+/'"(0) + ... +/ ~ .
A function which is not a polynomial cannot be expressed exactly

in this way. But we can always find the difference between a given
function f(x) and the series written above and determine whether
the difference is large or small. To be quite general, let n be a fixed
positive integer and xa
a fixed value of the variable x. Then, for
the given function f(x), write
f(a + *) -/(a) +/'(a) x +/"(a)

~ + ...+/M (a) ~ + Bn (x),
[^ .
* This is an extension of the method of

approximations to the value of
a function by a linear expression (see 6.4 above).
"
where Rn (x) is the difference, or remainder ", between f(a + x)
and the series written. The fundamental result is that
xn+l
where 6 is a positive fraction. In other words, Rn (

x) is t>he term we
*K**
expect to follow / (w) (a) r except that the value of the
derivative is taken, not at a, but at some point (not otherwise speci-

fied) between a and (a ~f x).
Without giving a complete
proof of the result,* we can
illustrate (in diagrammatic
terms) the simple but im-
portant case where n is
zero. The curve y~f(x) is
continuous with a tangent

at all Let
points (Fig. 102).
A and P
be points on the
curve with abscissae a and
(a + x) and ordinates f(a)
x and /(a + x). The chord AP
FIG. 102. (a + x)-f(a)
has gradient
referred to Ox. Since the curve passes continuously from A to P, it
is clear that there must be a point Q on the curve between A and P

where the tangent is parallel to AP. Let be the abscissa of Q so
that is between a and (a + x), i.e. = a + Ox for some fractional value
of 6. The tangent gradient at Q is /'() =f'(a + Ox) So .
x
i.e. f(a+x)=f(a)+Ro(x) where R Q (x)^f'(a + 6x)x
and our result for 7i=0 is obtained.f
* Tho Seo Hardy, Pure Mathematics
rigid proof of this result is complex.
(3rd Ed. 1921), pp. 262 et seq.
"\
The result when n = is known as the Mean Value Theorem and its
meaning can be expressed as follows. Tho mean increment in the function
f(x) for the increment in x from a to (a + x)
x
is
ratio tends to/ (a). Our result shows that the ratio equals the derivative of
- - x
- - As #->a, this
f(x) at some point between a and (a + x) even when x has not a small value.
Consider, now, the result of letting n take larger and larger

integral values. The number of terms in the series for f(x) then
increases, i.e. we are deriving an infinite series. All now turns on
" "
the limiting tendency of the remainder Rn (x). If, for a certain
value of x, Bn (x) tends to zero as n tends to infinity, then the
infinite series we obtain is convergent and we have
f(a + x)=f(a)+f(a) X +f"(a)^+f'"(a)^+... (to oo ).
On if Rn (x) does not tend to zero, then the infinite

the other hand,
we obtain not convergent and cannot have f(a + x) as a sum to
is
infinity. The broad outline of the solution of the problem of ex-

panding f(a -f x) as a power series is now clear. The possibility of
the expansion turns on the behaviour of Rn (x) as n->o> Rn (x) .
may tend to zero for some values of x and not for others. The
former values of x are those for which expansion is possible. Once
this has been determined, the actual expansion is the series written
above.
Collecting our results :
The function f(x) has finite and continuous derivatives of all
orders at a: a. For a given positive integral value of n,

/y.2
f(a + x) =f(a) \ ft -f-
where 6 is some positive fraction. Further,
f(a + x) =f(a) +f'(a)x+f"(a)

~+ ... +/<>() -+ ...,
l L*
---
for any value of x such that /(n+1) (a + 0#)-jIn I~-> ~T~
as
The expansion of/(a-f x) shown is known as Taylor's series.

To summarise the position, the values of f(x) and its derivatives
are known at the point x
a. It is then possible, subject to conditions
named, to express the value of the function at any point (a -f x) in

the neighbourhood of xa
as a series in ascending powers of x, the
coefficients involving only values at xa. In short, the values of
f(x) and its derivatives at x=a provide also the value of f(x) at any
neighbouring point. The particular point #=a, in the neighbour-
hood of which the expansion holds, can be chosen at will. By
selecting different points we get different expansions, e.g.
f(i+x)=f(i)+f*(i)x+f '(i&+... (* =1 )-
Finally, it is to be stressed that each expansion holds only for certain

values of x. As extreme cases, it may be that f(a + x) can be ex-
panded for no values of x other than x = (the most unfavourable
case) or that f(a + x) can be expanded for all finite values of x (the
most favourable case).
17.3 Examples of the expansion of functions.

The most useful and frequently quoted cases of expansions are
considered in some detail in the following examples :
Ex. 1. If f(x) = xr (r any real number), then
r ~n
In general, fW(x) =r(r - 1) (r -2) ... (r-n + I)x
Taylor's series (a = l, n = l) gives
~
where X = \r(r -1)(1 +0x) r 2 is finite even when #=0. This gives a
r
convenient approximate expression for (1 -f x) when x is small. More
generally,
r(r-l)
r-- ... - -- - ~
provided that
tends to zero as n tends to infinity. If x is positive and less than unity,

then #n+1 -0
and (n + 1) (1 + ftc) n+l ->oo as w-ôo Hence, .Rn (#) as written
.
above must tend to zero. The same result holds, but is more difficult to
prove, if a; is negative and numerically less than unity. We thus have the
binomial expansion,
for any value of x which

r(r-l)'*2
v
is
... - --
r(r-l)(r-2)
-
numerically less than one.

p
...
i -
(r-n + 1)-
E.g.
To obtain an approximate value of such a binomial expression, we have

only to add a sufficient number of terms of the series. So,
=1 +0-05 -0-00125 +0-0000625 + ...
= 1-0488 to four decimal places.
The fifth term of the series is not needed here.
Ex. 2. All the derivatives of e x are equal to e x .

Hence, by Taylor's
series with a=0,
x2 3
xn
+ _ + ...,
x
_ + -- + ...
provided that Rn (
x) i =-e**->0 as n->oo .
The latter is seen to hold for any finite value of x whatever since
increases more rapidly than any power #n+1 as n tends to infinity. The
series written above thus holds for any finite value of x and it is called the
exponential series.
If x is small, the exponential ex can be replaced approximately by the
3
quadratic (1 + x + 1# ) or by the cubic (1 +x + \x* + &E ). The latter gives
2
ex toa degree of approximation represented by x4 =

E.g., if # 0-1, the .
cubic expression certainly gives ex correct to three places of decimals.

We must include more terms when the value of x is larger or when greater
accuracy is required.
Taking x = 1, it is found that the series
e " 1+1+ + + '''

Jl jl
gives e correct to five decimal places when ten terms are added. Noting
that each term can be obtained, as a decimal, by dividing the previous
term by 2, 3, 4, ... , the ten terms are easily evaluated each to six decimal
places as follows :
e = 2-5
+ 0-166667
+ 0-041667
+ 0-008333
+ 0-001389
+ 0-000198
+ 0-000025
+ 0-000003
= 2-718282
i.e. e =2-71828 correct to five decimal places.
Ex. 3. If /(a) = log z, then
/"(*)- -i* /'"(*) ..

A*)-? -p.
In general, 1
\n-l
*===
f(n) ( x )=(- 1)"- -
x
Taylor's series with a
=1 gives
a;-+-- + ...+( -I)"-1 +-,

- *)"
provided that En (x) =(-!) n+1 ^j = <
tends to zero as n-xx> This condition (exactly as in Ex. 1) is satisfied

.
if x is numerically less The series written above, the logarithmic

than one.
series, is valid for such values of x.
Here again we can replace log(l+z) by an approximate quadratic

or cubic expression if x is small. The series gives a method of finding
approximate values of particular (natural) logarithms. E.g.
= 0-1 -0-005 +0-00033 -0-000025 + ...
= 0-0953 to four decimal places.
17.4 The expansion of a function of two or more variables.

The function zj(x, y) is assumed to have continuous and finite
partial derivatives up to any desired order at (a, b).

Write <f>(t)=;f(a + xt,b+yt)=f(u,v) (u=a + xt, v = b+yt).
Then f(<)==+
r
du dt dv dt du
y
dv
(13-4
v above)'
*
and .,
> <
d
)=(a;-
I
df
+ /
df\
L dfdfL
=*T(
dt \du/ 5^ d
dt
d
d/u,
d*f dv\ I 3 2/ d/u dv\
~X _ ~ + +y
dt) \dudv ~dt dt)
and so on. In general, we find
d f n Sn
f
+ nxyy n ~ l ,
^^
dxdv"-
-
1
+ yn ^-
n)
dv
,
where the numerical coefficients follow the " binomial " form
indicated above (17.3, Ex. 1). By Taylor's series for <f>(t),
f-2 fn /n+1
Putting < = 1 and substituting the values of ^ (t) and its derivatives
(all evaluated at <
= 0, so that u=a and v = b), we have
f(a + x,b+y)=f(a,b)-
'
a,b
uju c/t*/ c/
y uy 'aft
where
being some positive fraction. The notation adopted here needs a

littleexplanation. The values a and b appearing as suffixes of
successive terms are to be inserted for x and y in the partial deriva-
tives of f(x, y) but not in the powers of x and y included in the
terms. Thus
'
+2' =xfx (a"
a b
and similarly for the other terms. Again, in the expression for
Rn (#> y)> the values a + Ox and b + dx are to be substituted for x and y
in the (n + l)th order partial derivatives concerned.
We have now obtained an extended form of Taylor's series and
provided that Rn (x, y)->Q as ?i->ao .
This is the expansion off(a + x, b + y) as a series in ascending powers

of x and y, the coefficients being the values of the function and its
partial derivatives at the point(a, b). The expansion holds only for
those values of x and y which make JRn (x, y)->0 as n-xx> It is to .
be noticed that the terms of the expansion are grouped so that each
containsall powers and products of x and y of a certain degree, the
term being linear in x and y, the second term quadratic and

first
so on. Successive terms are variations of higher and higher

orders.
The use of Taylor's series in expanding and finding approximate
values of given functions is exactly as before. If x and y are small,
then f(a + x, b + y) can be replaced approximately by a quadratic
or cubic expression in x and y by retaining only the first few terms
of the series and the complex later terms are not needed.
Ex. All the partial derivatives of ex+y are equal to e x+ v .
Fixing the point (0, 0) and applying Taylor's series,
y'
provided that Rn (x, y) = v
. 6 <*+*)->() as 7&->oo .
The condition holds for all values of x and y (as in 17.3, Ex. 2). So
\L 11
which agrees with the ordinary exponential expansion on substituting
(x +i/) for the index.
Taylor's series can be extended to cases where more than two

variables appear. As the number of the variables increases, each
term of the series involves more and more powers and products.
In general, it f(xlt x 2 ... xn ) is a function of n variables and

,
if
(%, a 2 ... an ) is a fixed point, then

,
a 2/
a 2 /" \
...+2xn _1 xn -
^ n / aiaa ... an
subject to conditions similar to those already given. Our results are

thus perfectly general and the only difficulty is the labour involved
in writing the appropriate series.
17.5 A complete criterion for maximum and minimum values.

In addition to its practical use in the expansion of functions,
Taylor's series is a powerful theoretical tool of service in such
problems as that of maximum and minimum values. We can, first,
complete the tests for extreme values of a function of one variable
(8.2, 8.3 and 8.6 above) and then, in a later chapter, use similar
methods to develop tests for functions of several variables.
A convenient form of the definition of extreme values is :
The function f(x) has a maximum (minimum) value at x=a if
-/(a) is negative (positive) for all small values of x.
It is again assumed that the function has finite and continuous

derivatives up to any desired order at #=a. Taylor's series gives
for any fixed positive integral value of n. This form of the series
is clearly of direct application to our problem.
(1) Suppose /'(a) ^0.

By the continuity of the derivative, /'(a + Ox) has the same sign
as /'(a) for a certain range of small (positive and negative) values
of x. Now
f(a+x)-f(a)=f'(a + 0x)x (n = 0)
and so f(a + x)-f(a) must change sign as x changes sign. The

point x =a cannot, therefore, give an extreme value off(x). Hence,
if x=a gives an extreme value of /(#), we must have /'(a) =0.
(2) Suppose /'(a) =0, /"(a)*0.

As before, /"(a + Bx) has the sign of f"(a) for a certain range of
positive and negative values of x. But
(n = l),
i.e. f(a + x) -f(a) has the sign of f"(a + Ox), i.e. of /"(a). Hence,
If /'(a)=0 and/"(a)<0, #=a gives a maximum value off(x).
If/'(a) = and/"(#)>0, #=# gives a minimum value off(x).
(3) Suppose f'(a) =/"(a)
= 0, /'"(a) ^0.
Again f'"(a + 6x) has the sign of /'"(a) for sufficiently small
positive and negative values of #. So
- f(a) =f'(a)x+/"(a) ~ +/'"(a + fa) ~

If If
=/"'( + fe)s (n = 2)
shows that/(a + a;) -/(a) changes sign as # changes sign. The point
xa, which is a stationary point, is thus a point of inflexion and
does not give an extreme value off(x).
- 1)
(4) Suppose/'(a)-/-...-/ (w ()-0, /
Taylor's series can now be written
Of the two terms on the right-hand side, the derivative has a constant
sign, that of / (n) ( a )> f r sufficiently small (positive or negative)
values of x. The other term ( j- ) changes sign with x if n is an odd

\ til
integer and is always positive if n is an even integer. In the former

case, f(a + x)~f(a) changes sign as x changes sign and x=a must be
a point of inflexion. In the latter case, f(a -f x) -f(a) has a constant
sign and x=a gives a maximum or minimum value off(x) according
as/ (n) (a) is negative or positive. Hence,
COMPLETE CRITERION FOR MAXIMUM AND MINIMUM VALUES

(1) If f(x) has an extreme value at x=a 9 then/'(a) = 0.
(2) If /'(a)=/
//
(a)
= ...=/<-D(a)=0, /<">(a)^0, then f(x)
has a stationary value at x=a which is an inflexional value if
n is odd, a maximum value if n is even and /<n) (a)<0 and a
minimum value if n is even and / (n) (a) > 0.
Subject to the condition that the derivatives involved are finite

and continuous, the criterion is complete and so both necessary and
sufficient. There is no case of failure ;
unless the function is a
constant (and so without maxima or minima) there must always be
some derivative which is not zero. The order and sign of this
derivative determines the nature of the point considered. The
practical method of finding maximum and minimum values (8.4
above) is scarcely affected a few doubtful cases
; (such as that of the
following example) are cleared up and that is all.
Ex. y = (x-\)*.
S
J-^-W S^ 12 ^ 1 )2'
g-^*- 1) and
g-24.
There only one stationary value, y =0, occurring at x 1. At this point,
is
the first three derivatives are zero and the fourth derivative is positive.
Hence, /==0 is a minimum value of the function. This can be checked
sincey is positive for all values of x except that it is zero when x = 1. A
graph of the function shows a ciirve very similar in shape to the parabola
and with vertex at (1, 0) on the axis Ox.
17.6 Second and higher order differentials.

The differential of a function of several variables describes the
" "
first-order variation of the function.
But, just as derivatives
of various orders are needed to describe completely the variation of
a function of one variable, so something more than the ordinary
differential is required in the case of a function of two or more
variables. The need is met by the definition of differentials of the
second and higher orders. In the following, the definitions and
results are given only for a function, z /(#, y), of two variables, but
they are easily extended to more general functions.

The function z =/(#, y) is assumed to have finite and continuous
partial derivatives of all orders required. The first-order variation
2 *5
is described by dz=~ dx + dy whether the variables x and y are

ox oy
independent or not. It is very important, however, to distinguish
carefully between the two cases. If x and y are independent, then
and are functions of these variables but dx and dy are arbitrary
and constant increments in the variables. But, if x and y are de-
pendent on other variables, then and are also functions of the

dx dy
genuine independent variables and dx and dy are expressed in terms
of the same variables plus their arbitrary and constant increments.
In either case, dz involves functions of the independent variables
(whether they are x and y or others) and certain constants, the
arbitrary increments in the independent variables. The differential
of dz, as a function of certain independent variables, can now be
defined in the ordinary way and is called the second differential :
d 2z=d(dz).
The process can be extended and, in general, the nth differential of

z is defined as the differential of the (n - l)th differential :
dnz=d(dn ~ l z).
As in the case of derivatives, however, it is seldom that we need

differentials of higher order than the second.
Our problem now to express higher order differentials in terms
is
of successive partial derivatives of the function. The importance of

distinguishing the independent variables becomes evident here.
Unlike the expression for dz, which is the same in all cases, our
results for the higher order differentials are different and simpler
when x and y are the independent variables than they are in other
cases. The reason for this is not far to seek. In differentiating dz
to obtain d 2 z, we must know how to treat the dx and dy appearing
in dz. If x and y are independent variables, dx and dy are constants
and do not worry us. In other cases, this is not so and due account
must be taken of the variation of dx and dy in the differentiation
process. Before beginning to differentiate, therefore, it is essential
to know whether x and y are independent or dependent
variables.
17.7 Differentials of a function of two independent variables.
Differentiating the first differential of a function of two inde-

pendent variables, we have
a dz\ , a dz\ . } _ fa /az\ . a tdz

a
dx\dxJ 5- } dy
dxJ
y\ dx +\^r *- dx +T' U-
r {*-)
dy\dxJ j \dx\dy
\dx\dy) ty\d
i.e. d 2z = 5-5 do;

7 O
2 77 + . 7 2O
rt
+ 2 ^-5- dxdy
y 2 dy
.
dx* dxdy dy
The essential point is that rfx and dy are treated throughout as
constant increments otherwise the ordinary rules of differentiation
;
(13.3 above) are used. It appears, therefore, that d 2z is a quadratic
expression in dx and dy, the coefficients being given by the second-

order partial derivatives of z at the point (#, y) in question.
In finding second differentials in practice, we can either evaluate
the second-order partial derivatives and use the above result, or we
can differentiate twice by rule in the particular case considered.
The latter method usually involves less labour and is adopted in
the following examples. In each case, however, the results can be
checked by means of the partial derivatives.
Ex.1. z=
Here dz = 2 (x + y)dx + 2 (# - y)dy
and d2
= 2(dx 2 +2dxdy-dy 2 ).
2
Ex.2. z= ,
x-y + I
Here dz ^(-
x -y + 1) 2
,2
- 2xy + 2x) dx
d
(x-y + 1)*
{(y-\)dx-xdyy
Ex.3. z=l
Here dz
x2 +y 2
2 2 2 2
Z &
d(xdx+ydy)(x +y )-d(x +y )(xdx + ydy)
U> 752 2 75
2
(x +y )
_ (y (dx
2
+dy
2
)(x
2
+y2)-2(xdx+ydy) 2
2 2 2
(x +y )
(x
2 - y2 dx 2 + 4xy dx dy -
) (x
2 - y2 ) dy2
The process of differentiation can be repeated to give
dx 2 dxdy
ty*
Tfiv /9^Z ^?
3Z /5^2
i.e. f/
3
z =- 3
Ac 3 H- 3 . . dx 2 dy 4- 3 -
=--; rfx dy + -
2
3 ^v
3
-
arr ox* oy ox oy 2 dy
This is a cubic expression in dx and A/ with coefficients given by the
third-order partial derivatives of z. Higher order differentials are

obtained in exactly the same way and it is to be noticed that the
" "
numerical coefficients of successive terms follow the binomial
law (17, 3, Ex. 1, above).
We can note here a connection between successive differentials of
a function of independent variables and the corresponding terms of
the Taylor expansion of the function. Assuming that Taylor's series
is valid, it can now be written in the form
d2-z d*z
y + k)-f(x, y )=
-
, r r
if li
where the be taken at (x y) and with respect
differentials of z are to }
to arbitrary increments dx~h and dy~k in the variables. Hence,

the increment Az in the function for increments h and k in the
variables from (x, y) is compounded of the successive differentials,
dz, d 2z t d*z, ... ,
of the function at (x, y). The complete variation of
the function is described only by using differentials of all orders But, .
the successive differentials involve higher powers of the increments

" "
h and k, i.e. they are of higher order of smallness if h and k are
small. For approximate results, only a few differentials (i.e. a few
terms of Taylor's series) are needed to describe the variation of the
function. As a first approximation Az dz\ as a second approxi-
mation Az dz + Mz 2
;
and so on.
The higher order differentials of a function of a single variable,
y =f(x )> appear as a special case of the general process of differen-
tiation described above :
dy=f'(x)dx, d*y=f"(x)dx*, d*y^f'"(x)dx\ ... .
In general, dny = / (n) (x) dxn .
The first derivative, as we have seen, can be interpreted as the

ratio of the differential dy to the differential dx. This result now
extends. The second the ratio of d 2y to dx 2 i.e. the
derivative is ,
ratio of the second differential of the dependent variable y to the
square of the differential (or arbitrary increment) of the independent

d 2y
variable x. The notation -,--2 for the second derivative is thus
dx
justified. Similar results hold for higher order derivatives.
17.8 Differentials of a function of two dependent variables.

In the function zf(x } y), the variables x and y are dependent on
a set of independent variables u, v, w, . . . . The process of obtaining
the second differential d 2z now proceeds :
=d **+ -d *> +d
dz ,, _ . . fdz\ 7 dz
a^2
dx ^T"
dxdy I
a~a~^
dx \dxdy
a^2
dy
J
I
y iT
dy
.e.
2
dx 2 dxdy dy dx dy
The expression for d 2z now includes two additional terms, the terms
in the second differentials of x and y as functions of the independent
variables u,v,w, ... Ifx and y happen to be independent variables,
.
these terms are zero (since dx and dy are constants) and the expres-
sion for d 2z reduces to that previously obtained. In general, however,
d 2x and d 2y must be retained and interpreted in the light of the
particular problem considered.
2G M.A.
Two simple applications of the result can be noted. If zf(x, y)
where x and y are functions of a single variable t, the partial deriva-
tives and derivatives of the functions being known, then from
dz=dx + ~dy, we derive (13.4 above) :
dz __ dz dx dz dy
dt~ dx dt dy~dt'
Prom d%=^dx* + 2^dxdy +

2 2
dx dxdy dy
we derive
2
dzdtx
<fady
+ d^zfdy]
2
* + dzdty2 '
dxdydtdtdy \dt dxdt 2 dydt

The ratio of d 2z to dt 2 is to be interpreted as the second derivative
of z as a function of t
(defined by means of x and y). We have thus
a simple formula for this derivative in terms of the partial derivatives
of z and the derivatives of x and y. The same result is obtained by
taking -=- as a function of t and by finding its derivative directly.

at
In particular, if z f(x, y) where y is a function of x,
dz __ dz dz dy
dx dx dydx
2
d 2z dy cPz /dy\ dzdfy
d?z__<Pz * + '
2 2
^dy \dx) dydx
In this case, d 2x does not appear in d 2z since x is the independent
variable.
y is an implicit function of x defined by the relation f(x,

If y) = 0,
we have seen (13.5 above) that
which gives the derivative -jr~ ~T Differentiating again,

ax jv
d(fx dx+fy dy) = Q,
i.e. /.. dx 2
+ 2fav dx dy +fyy dy 2 +/, d 2y = 0,
2 2
noting that d y, but not d x appears since y is taken as a function
9
of x. Dividing through by dx 2 9
This gives the second derivative of y as a function of x. If the value

fj4J
of ~j- previously obtained is substituted, the result already written
doc
(13.5 above) is again obtained.

If the same relation is regarded as giving # as a function of y,
dx f
then fxdx+fy dy Q gives the first derivative = -j-
and from
d (f* dx +/, dy) = 0, we find
dy Jx
fxx dx* + 2fxv dx dy +fvv dy* +/. d*x = 0,

which gives the second derivative,
dx dx
* +2/f"
+ +f ( Y\
+Jxx '
dy \dy) }
dx
This can be expanded, as before, on substituting the value of -=- .
dy dx ~
It is to be noted that, though the and -=- are first derivatives
dx dy
reciprocal, the same relation does not hold between the second
derivatives.
In practice, the derivatives of implicit functions are best obtained
by differentiating the relation between the variables as it stands,
always remembering which of the variables is dependent :
Ex.1. a
Taking y as a* double -valued function of x, we have
dx
=, .--.
dx y
Differentiating again,
d 2y If
l 'e ' L+ fo
d^~~y\ \dxJ J~ y
But, taking a; as a double-valued function of y,
dy
=. .- dy x
and
dy/
2
If, /dx\*} la: 2 + 16
Ex.2. x*+y*-3xy=Q.
Regarding y as a triple- valued function of x,
Differentiating again,
Similarly, if a; is regarded as a function of y, we find

dx ^ y
2
-x d?x __ 2xy
Returning to the general case where z=f(x, y) and x and y are

functions of certain independent variables, the third differential
d 3z can be obtained by differentiating the second differential d 2z :
. dx* +2 ~- dxdy + ^ dy* + ~d*x + ^2 ffiy)

*
dxdy dy dx dy I
52 32
on reduction. The expression of the differential has now become

very cumbersome. Higher order differentials are obtained by further
differentiation and are exceedingly involved.
As a particular case of the above results, let y=f(x) where x is
dependent on other variables. Then

= f"(x)dx*+f(x)d*x, ....
For example, if a; is a function of a single variable t,
dt
~ -
dt
'
dt*
-
dt*
'
dt I
This the ordinary function of a function rule for functions of one

is
variable, extended to the second and higher order derivatives.
EXAMPLES XVII
Infinite series
1. Write down the sequence of numbers with nth member (

--- )
i
\2 n+ 1/
and obtain the infinite series of which the sum of n terms is given by TEis
expression. Show that the series is convergent to i.
2. Find the infinite series with sum of n terms given by (a) n 2 (n-f- 1) and
(6) \n(n l)(2y& + 1). Show that neither series is convergent.
-t-
3. Write the series with sum of n tqrms - - and show that it is convergent.
4. It is known that TT is the sum to infinity of the series
*(i-i+t- ++*-A+...).
Find the value of tr to five decimal places.
5. Show that *J2 is the sum to infinity of the binomial series
t{i--H-i(i) 2 -A(i) 3 + -..}, (cf. n, 3, Ex. i.)
and deduce the value of ^/2 correct to four decimal places.
6. Express the recurring decimals 0-13 and 0-27 as fractions.

7. Write down the sum to infinity of each of the convergent series
Illustrate the approximation to the sum to infinity by finding the sum of ten
terms in each case.
8. A ball is dropped from height h feet on a hard floor, bounces, falls,

bounces again, and so on. Each bounce is to height e times the previous fall,
e being a positive fraction. Show that the distance travelled by the ball
before coming to rest can be represented as the sum to infinity of an infinite
geometric series. If the distance is 2h, show that e .
9. An
income stream, starting with a next year, is such that each year's
income half that of the previous year. Adding interest yearly at lOOr per
is
cent, per year, find the present value of the stream flowing for n years, and of
the stream flowing for ever.
10. An income stream starts next year with a and flows for ever so that
each year's income is a fixed percentage (100s per cent.) less than the previous
year's income. Find the present value of the stream, adding interest yearly
at lOOr per cent, per year.
Expansions
11. If 6 is some positive fraction, use Taylor's series (n = 0) to show that
log ( 1 + x) - and deduce that
1 -{ uX
Then, from Taylor's series (n= 1), show that
12. For values of x numerically less than one, show that
13. Write /s/6 = |- \/25 - 1 and use a binomial series to calculate its value
correct to five decimal places.
14. Find a quadratic expression which approximates to == when x is

vl + #
small. Put #= sfe and find ^5 correct to four decimal places.
x< 1 numerically, write expansions for s/l+o?, *Jl-x and *S]T-x z

15. If .
By algebraic multiplication of infinite series, verify that the expansion of a

function which is the product of two parts is the product of the expansions of
the separate parts.
16. Form an infinite series each term of which is the integral of the corre-
sponding term of the expansion of (1 4- a)" in ascending powers of x. What
1
isthe series ? Show that its sum to infinity is the integral of (1 4- a?)"
1
. What
general rule does this result suggest ?
17. By taking sufficient terms of the appropriate series, find the values of
Je and
- correct to four decimal places.
18. Show that

Given log 2 = 0-6931, find values of the natural logarithms of 3, 4 and 5
correct to three decimal places. (Note log 4 can be found without using
:
the infinite series.)
19. If interest at lOOr per cent, compounded yearly is equivalent to interest

at 100s per cent, compounded n times a year (cf. Examples IX, 25), show that
r exceeds s by approximately sa .
&n
20. If interest at lOOr per cent, compounded yearly and at lOOp per cent.
compounded continuously are equivalent, show that r - p = %r approximately
z
when r is small (cf. Examples IX, 26).

21. Find the amount of after n years at lOOr per cent, interest which is
1
(a) simple, (6) compounded yearly, and (c) compounded continuously. If

8
(m) is small enough to be neglected, show that the last amount exceeds the
t(rn) and the second
2
first by by Jr
2
n approximately.
22. A
man's income is *? and his (measurable) utility is then <f>(x) where
<f>(x) increases at a decreasing rate as x increases. The man makes a fair bet
on an event with chance p, laying pa against ( 1 - p)a that the event happens.
Show that his expectation of utility is now - -
{ptf)(x -hi- pa) + (1 p)<f> (x pa)}.
If the bet small, use Taylor's series to show that the expectation gives less
is
utility than the original income x. (See Marshall, Principles of Economics,

8th Ed. 1927, p. 843.)
23. By evaluating derivatives and using Taylor's series, show that
assuming that the series is convergent.
24.
pression
1 x y
-
Expand, by Taylor's
-
-\-
series for
as a series in ascending powers of x and y. Deduce
and quadratic approximations to this expression when x and y are both

a function of two variables, the ex-
linear
small.
(Cf. Examples XIII, 7.)
Higher order differentials

25. that neither y=(x~ I) 3 nor y (x - I) 8 has a maximum or mini-
Show
mum value and that each has an inflexional point at (1, 0).
26. Find the partial derivatives, and hence the first two differentials, of
x*
each of z~x* + 2xy 2
-y , z ---- and z log (x
2
+y 2
).
x y -f 1
x 3 + y* - 3xy 9 z = */x*
z
27.
~
ex v
Find the second
and 2 = log - differential of each of z
where x and y are independent variables.

+ y2 ,
28. If x y and z are independent

9 variables, find the second differential of
u=(x* + 2xy-y*)ez .
1 d 2z
29. If z = where x = e* and y e**, evaluate ~r-r
x+y cut*
Given that x* + y* - 3xy = 0, use the derivatives of y as a function of

30.
x and of a? as a function of y to show that y has a single maximum value
for variation of x and conversely. Illustrate diagrammatically. (See Example
XIII, 22.)
31. The relation f(x, y, z)
*
gives z as a function of x and y. Differentiate
the relation and obtain the first and second order partial derivatives of z in
terms of those of /. Illustrate by taking the relation x 3 -f t/ 3 4- z 3 - 3xy = 0.
(See Examples XIII, 23.)
32. If z =f(x, y) whore x and y are given functions of the independent

variables u and v, show that
dz d*x dz d*y
_
Ix dy du ou dy du* dy* \ou dx &u*
and similar results for the other two partial derivatives of the second order.
33. From the results of the previous example, find the second-order partial
derivatives of z = log (x-y) where x = - and y = ~. Express z explicitly in

terms of u and v and find the partial derivatives directly,
CHAPTER XVIII
DETERMINANTS, LINEAR EQUATIONS AND
QUADRATIC FORMS
18.1 The general notion of a determinant.
DESPITE its somewhat terrifying name, a " determinant " is a
mathematical tool of a very ordinary kind and involves no new
ideas of any description. Briefly, a determinant is a notation that is
found convenient in handling certain involved, but essentially com-
monplace, algebraic processes. Certain expressions of a common
form appear in algebraic problems such as that of the solution of
linear equations, expressions consisting of sums or differences of a
number of terms each of which is the product of a number of quan-
tities. The expressions (ab-h z ) and (abc-af 2 -bg 2 -~ch* + 2fgh) are
cases in point. Quite apart from other considerations, the labour of
writing out the more complicated of these expressions is severe and
there is every reason to welcome a compact and general notation
for them. The determinant notation is justified on these grounds
alone.
There however, more in the determinant notation than this.
is,
Once the notation is introduced, the expressions denoted by deter-

minants are seen to obey quite simple rules and the algebraic pro-
cesses in which they appear become simpler and more uniform than
before. As a consequence, results can be established which would
almost certainly be missed were it not for the new notation. It is
have become of general use no
for such reasons that determinants ;
notation can be retained unless saves labour and enables us to

it
carry out processes more easily and with greater generality than
before.
The foundation of a determinant is a square " block " of quantities
" "
written down in rows and columns in crossword form. There is
no restriction on the quantities except that each must be capable of
DETERMINANTS, LINEAR EQUATIONS 473
taking a single numerical value they can be numbers, constants,

;
variables, functions, derivatives of functions, and so on. The number

of the quantities must be a perfect square four quantities can be ;
written down in a square form of two rows and two columns, nine
quantities in a square form of three rows and three columns, and
so on. Examples are provided by the following arrangements :
9
du du x2 xy xz
dx dy ,
'
and ** y yz
Sv dv
3 -r -r
ox dy xz yz z2
A determinant of two rows and columns is called a determinant of

the second order, of three rows and columns a determinant of the
third order, and so on. In general, a determinant of the nth order
has n 2 quantities arranged in n rows and n columns. The quantities
themselves are called the elements of the determinant and we can
speak of the elements of the first row or column, of the second row
or column, and so on. Each element is allotted, of course, to one row
and to one column.
A general method of denoting the block of quantities making up
a determinant is required. From
the point of view of stressing the
arrangement in rows and columns, the best notation is
an a12 6^3 ... aln
where the integer n denotes the order of the determinant. The two
suffixes of an element denote the row and the column into which the
element is to be placed. Thus, a 53 is the element to be inserted in
" "
the cell at the intersection of the fifth row and third column. It
must be noted, however, that the notation is designed for dealing
with determinants in general in any particular case the elements
;
appear as actual numbers or quantities of one kind or another.
18.2 The definition of determinants of various orders.
A good working definition of determinants can be given in suc-

cessive stages as follows.* A determinant of the second order is
* For a more and general Die Determinanten

strict definition, see Netto,
(2nd Ed. 1925), pp. 8-13.
" "
defined as taking its value from the cross multiplication of the
four elements which compose it and we write
#11 #12
#21 #22
The determinant on the left is simply another way of writing the
algebraic expression on the right.
Ex. 1. 2 -1 = 2x3-(-l)xO
3
Ex.2. a h =axb- =ab -h2 .
h b
Ex. 3. x* 2xy = x*xy*-2xy*2xy= -

2
2xy y
Ex.4. du du du dv du dv
'
dx dy dx dy dy dx
dv dv
dx dy
where u and v are two given functions of x and y. This determinant,
which is often written g Uy v \ i
is termed the Jacobian of u and v with respect to x and y.
A determinant of the third order is defined in terms of those of the

second order by the rule :
#12 #22 #23 ~#12 a 21 ^23 + #13 #21 #22
#21 #32 #33 #31 #33 #31 #32

#31 #32 #33
- a12 (a 21a 33 ~ a 23a31 )
+ #! 3 (#21#32~#22#3l)
= #11#22#33 + #12#23#31 + #13#21#32
- #n# 23#32 - #12#21#33 ~ #13#22#31-
The second-order determinant multiplying au is derived from the
original determinant by omitting the first row and the first column
(the row and column intersecting in a^), and similarly for the other
determinants. The third-order determinant is thus seen to be a
short way of writing an algebraic sum of six terms, each term being
the product of three elements of the determinant so chosen that one
element comes from each row and one from each column. Further,
half the terms are added and half subtracted to form the algebraic
sum.
Ex. 1. 1 3 -2 -1 -2 -1
-103 -2 -2
-2 1
=
(0 +6) -3(- 1-0) -2(2-0) =6 +3 -4 = 5.
Ex.2.
Ex.3.
Ex.4.
d(x,y,z)
A determinant of the fourth order is then defined ;
This is an obvious extension of the rule for third-order determinants

and it remains to evaluate each of the determinants shown on the
right-hand side and collect terms. It is then found that the deter-
minant denotes the algebraic sum of 24 terms, each term consisting
of four of the total of sixteen elements so chosen that one comes from
each row and one from each column. Again, half the terms are
positive and half negative same kind of way as before. The
in the
determinant notation represents more and more lengthy expressions,
and so becomes more and more convenient, as we proceed.
In general, a determinant of the nth order is written down in terms
of determinants of the (n - l)th order by the rule :
%n a 22 a 23 a 2n
#22 a 2n
... a a nl an3 ... an
... a nn
The signs of the successive entries on the right-hand side are alter-
natively positive and negative until all the n elements of the first
row are taken. The (n- l)th order determinant multiplying alr is
obtained from the original determinant by the omission of the first

row and rth column. The determinant is now to be evaluated by
going, step by step, back to determinants of the second order and
so to the elements themselves. It is then found that we have the
algebraic sum of terms each of which is the product of n of the n 2

elements selected so that one element comes from each row and one
from each column. Half the terms have a positive and half a
negative sign. The number of terms in the sum is
\n=n(n-l)(n-2) ...3.2.1,
whichis very large when the order of the determinant is high.
The determinant notation as written above can be sometimes

abbreviated to
|
a r8 |
(r and s = l, 2, 3, ... n),
where a r8 is a typical element (in the rth row and 5th column).
A determinant, therefore, is a notation expressing an algebraic
sum of terms which are products of the elements of the determinant
and its value is of the same nature (e.g. numerical or a function of
certain variables) as the elements which compose it. The practical
way of finding the value of a given determinant is indicated by
the definition, i.e. by successive reduction to determinants of lower
and lower order until second-order determinants are obtained and
evaluated by cross multiplication.
18.3 Properties of determinants.

The following are simple properties of the determinant notation :
(1) A determinant is unchanged in value if its rows and columns

are transposed :
au nl
^23 ..'
^21 ^22 ^2i
#wl #n2 tt n3 nn #lw #2n #3n #wn
The value of a determinant is unaltered numerically but

(2)
changed in sign if two rows (or two columns) are interchanged :
.. "^
e.g. fif'i2 ^11 ^13 ^ln ^'11 ^12 #13 *
#lw
(3)The value of a determinant is increased Jfc-fold when each

element in one row (or column) is increased in this way :
e.g. kan ka lz ka13 ... kaln =k an al2 13 ... aln

% 21 a 22 a 23 ... a 2n
(4) A determinant has zero valuethe elements of one row (or if
of one column) are equal or proportional to the corresponding

elements of a second row (or of a second column).
(5) If the elements of one row (or of one column) of a determinant

appear each as the sum of two parts, then an additive rule applies :
e.g. a12 ln
anl 2n <*>
'
a12 aln
(6) The value of a determinant is unchanged when a multiple of

the elements of one row (or of one column) is added to the corre-
sponding elements of a second row (or of a second column) :
e.g.
2
nl
To make clear the properties, it is a useful
meaning of these
exercise to verify that they hold in the particular cases of second
and third-order determinants. It is also a relatively easy matter
to give general proofs, using properties of the values and signs of the
terms in the expanded form of a determinant (18.2 above).* The
practical evaluation of determinants is often simplified by using the
above properties, e.g. a combination of
(6) (4) sometimes shows and
that a determinant has zero value. The following examples illustrate :
Ex. 1. Show that 1 2 3 -0.

456
789
Take the elements of the first row from those of the second and third
rows, using property (6) with k = - 1. Then
1 2 3 = by property (4).
333
666
The same result holds if the elements of the determinant (reading by
rows from leftto right) form any set of integers in Arithmetic Progression.
Ex. 2. Express 1 a in factorial form.
1 b
1 c c2
If a = 6, the two rows of the determinant become identical and it has
first
zero value. Hence, (a - 6) must be a factor of the value of the determinant.

Similarly, (b
- c) and (c - a) are factors. Since the value of the determinant
is of the third degree in a, b and c, it must be a numerical multiple of these
2
factors. The leading term in the expanded value is 1 x 6 x c (from the
" "
diagonal of the determinant) and the multiple can only be unity. The
value of the determinant is thus (a -b)(b- c) (c - a).
18.4 Minors and co-factors of determinants.
Selecting an element of a given determinant, we delete the row

and column intersecting in the element and obtain a determinant of
* See
Netto, op. tit., pp. 13-9, and Courant, Differential and Integral Cal-
culus, Vol. II (English Ed. 1936), pp. 20-3.
order one less than that of the original determinant. The derived
determinant is called the minor of the selected element in the given
determinant. There are n 2 minors of a determinant of the nth order
and each is a determinant of the (n - l)th order. In
A \
a rs |
(
r an d 5 = 1, 2, 3, ... n)
the minor of the typical element a rs is denoted by A r8 ,
as obtained
from A by omitting the rth row and sth column.
The a selected element in a given determinant is the
co-factor of
minor of the element with a sign attached. The rule of signs is
quite simple. If the numbers of the row and column containing the
element add to an even number, then a plus sign is given to the co-
factor if they add to an odd number, then a minus sign is allotted.
;
The co-factor of a rs in A is denoted by A rs where

A rs =A rs if (r + s) is even
- A rs if (r + s) is odd.
It is important to distinguish between the closely related concepts
of minors and co-factors. The minors of various elements in
Our definition of determinants can be re-framed in terms of

minors or co-factors. We have
A= \
a r3 \=
or =#11^11 + #12^12
i.e. the value of A equals the sum of the elements of the first row
each multiplied by the corresponding co-factor. A similar result
can be shown, by property (2) of 18.3, to hold for the elements of
any row. For example, taking the second row, we have
#21 -^21 +#22 -^22 "+"#23 ^23 + +#2n^2n
= -(#21^21 -#22^22+ #23^23
- #2fAn)-
The bracket on the right-hand side is, by the definition, the value
of the determinant obtained by interchanging the first two rows of A.
The bracket thus equals ( - A) and so
A =a 21 A 21 + a 22 A 22 +a 23 A 23 + ... + a 2n A 2n .
Again, take the sum of the elements of one row of the determinant
each multiplied, not by the corresponding co-factors, but by the
co-factors of the elements of another row, e.g.
a 21 A n +a 22 A 12 +a 23 A 13 + ...+a 2n A ln .
This sum must, by the definition, equal A except that the first row
inA is replaced by the row of elements (a 21 a 22 #23* a 2n)> i- e the
, , -
sum represents a determinant with the first two rows consisting of

identical elements. By property (4) of 18.3, the sum is zero.
The following, results are now established :
The sum of the elements of any row of A a rs each multi- \
plied by the corresponding co-factor is equal to the value of the

determinant. The sum of the elements each multiplied by the
co-factor of the corresponding element of another row is zero.
In symbols, for unequal values of r and s,
The first result, an extension of the definition, is often called the

expansion rule and its use in evaluating determinants is evident. By
property (1) of 18.3, the results are true also for the elements of any
column of the determinant.
Minors of higher order than those discussed above can also be
defined. Selecting two elements not in one row or column and
omitting the two rows and two columns containing them, a second-
order minor is derived as a determinant of order two less than that
of the original determinant. The process can be continued by
selecting more and more elements and omitting the relevant rows
and columns. In particular, the principal minors of A = a r8 are |
the minors of various orders obtained by selecting the last, the last
" "
two, ... elements of the principal diagonal (an a 22 a 33 ... ann ) , , ,
and then carrying out the deletion process indicated.

18.5 Linear and homogeneous functions of several variables.

The determinant notation serves to express precisely the pro-
perties of linear and homogeneous functions of more than two
variables. Let y=f(x1 x 29 # 3 ... x n be a linear and homogeneous
, , )
function with continuous partial derivatives of the first two orders :
and / = -- (r and s l, 2, 3, ... n).

dx r dx s
Then, Euler's Theorem (12.8 above) can be extended to give :
atany point (xl9 x 2 x 3 ... xn ). Since the result is identically true,

, ,
it
can be differentiated with respect to any variable x r :
1A
O. 1* f - 1 O* f -i. 'Y1 f -1 I
/y -P f\ ( O\
IJ IT '
2i7 27* '" 3j ST ~* * * '
^71 ./ 7~w j l^'l
(r
= l, 2, 3,...n).
The n relations (2) arc extensions of the results (4) of 12.8 and give
each direct second-order partial derivative in terms of the cross-
partial derivatives.
We denote by F the determinant of the (n-fl)th order formed
" "
from the block of second-order partial derivatives of/, bordered
with the first-order partial derivatives :
A /2 /3 ..- fn
fl fll J12 /13 fin
f f f> f f
Jn Jin /an /3w Jnn

Let FQ denote the co-factor of the clement in F and F Qrs the co-
factor of the element frs in F Q (r and s = l 9 2, 3, ... n). Then
ô = (3)
and J 0ff =
-*^F (4)
The proofs of (3) and (4) are interesting exercises in the manipu-
lation of determinants :
Multiply the relation (2) by F Qlr (r = 1, 2, 3, ... n) and add together

the n equations obtained. Then
1 (Al^ Oll + - -
2H
482 MATHEMATICAL ANALYSIS TOR ECONOMISTS
By the results of 18.4, the coefficient of xt is F Q and the coefficients
of a? 2 #3 ... xn are all zero. Hence,
, ,
forany value of xl9 i.e. F ~ 0. This proves (3).

The determinant j? can be evaluated as follows
by 18.3, (3)
by 18.3, (6)
adding x 2 times the elements of the third row, x 3 times the elements
of the 'fourth row, ... to the elements of the second row and using
(1) and (2) above. Expanding in terms of the elements of the second
row and applying a similar manipulative process all over again (with
columns instead of rows), we find
!=-*
4* z
Hence, F oll = - F, which

l
is one of the results (4). The other
y
results follow by similar reductions of F.
18.6 The solution of linear equations.

The determinant notation is of particular use in solving linear
equations, operating with quadratic forms and making linear sub-
stitutions. We propose to give here some account of the use of the
notation in the first two problems.*
* For applications of the determinant notation to linear substitutions, see

Netto, op. cit., pp. 77 et seq. and pp. 100 et seq.
As we have seen (11.6 above), a system of exactly n equations

in n variables is sufficient to determine the values of the variables
provided that the equations are consistent and independent. But
this tells us nothing about the actual values obtained in the solution,
or even whether one or more alternative solutions exist. These points
can be cleared up at least in the simple case where all the equations
are of linear form. The solution of the pair of linear equations
=c and
in two variables is (by the method of 2.9 above) given by
and =
Using the determinant notation, the solution can be written

x y I
02
The solution is unique and appears in terms of second-order deter-

minants involving the coefficients of the equations.
To generalise the result, consider a system of n linear equations
in n variables, xl9 x 2 # 3 ... x n
, ,
:
11*1 + 12^2
21 1 '
022 2
The given coefficients in the equations form a square block of ele-

ments giving the determinant A =| a rs (r and 5
= 1, 2; ... n). Here,
a rs the coefficient of the sth variable (x s ) in the rth equation.
is
The constant terms (a l9 a 2 ... a n ) on the right-hand sides of the

,
equations are kept separate. Assuming that A ^ 0, the equations are

solved by the following device. Multiply the equations respectively
by A ll9 A 21 ... A nl the co-factors of the elements of the first column
, ,
of J, and add. Then
By the results of 18.3, the coefficient of x^ here is A and the

coefficients of the other variables are all zero. Further, let A l
denote the value of A when the first column of elements is replaced
by the column (e^, a2 ,
... an ), so that
A =at A + a 2 A 21 + ... +an A nl
llL
.
The equation then becomes Ax1 =A l9 i.e.
Similar processes give the values of the other variables and so :
The solution of the n linear equations written above is unique

and can be expressed in the form
where A is the nth order determinant a rs and A ly J 2 J 3 ... A n are

| \
> ,
the determinants obtained by substituting the column (a1? a 2 ... an ) ,
for the first, second, third, ... and nth columns of A respectively.
The result for two variables given above can be verified as a

particular case of this general solution.*
Ex. 1. The equations
give
1
3 1 -6
-134
1 1 2
Ex. 2. In considering the variation of demand for factors of production

(14.8 above), we had to solve the equations
+p a v --p w^=
Pa
for the three variables
dp and
* We can now see, at least roughly, why A = is a case of failure of our

result. If A
and J t J 2 J 3 ... J n ^=0, only infinite values of the variables
, , ,
can satisfy the equations, and the latter are thus inconsistent for all finite
values of the variables. If some of J 19 J |9 J ...^ w are also zero, then at least
,
?
one variable becomes indeterminate (zero divided by zero) and the equations
are not independent.
The solution can be written in determinant form
u
The first three determinants are easily expanded since one column in
each case contains two zeros. The other determinant, expanding in terms
of the first column, equals
- PaPb) - a -
-n (PaPb
~
( y- PaPb)
since xp = apa -f bp b .
Hence,
u v w I
xp - xpa (rj -a) a

-~
XP
and (w ==
) ^
db
= -- (n - a).
ab
18.7 Quadratic forms in two and three variables.
An expression which is the sum of a number of terms each of

the same (given) degree in certain variables can be termed a/orm.
Classifying forms according to the degree of the terms, we distinguish
linear forms, quadratic forms, cubic forms, and so on. The following
analysis is concerned with forms no more complicated than the

quadratic and makes their nature clearer than any general remarks
we can offer.
The general expression of a linear form is (ax + by), (ax + by + cz),...
when there are two, three, ... variables. Here a, 6, c, ... are constant
coefficients. forms present no algebraic
It is clear that linear diffi-
culties and no special theory need be developed.

The general quadratic form is
(ax
2
+ by 2
4- 2hxy), + by 2 + cz 2 + 2fyz + 2gxz + 2hxy),
(ax
2
. . .
according to the number of variables. There is now more algebraic

difficulty in dealing with such forms and the main problem that
presents determine whether a given form has the same
itself is to
sign (i.e. positive always or negative always) for values of the

variables which are not all zero. The solution of many problems in
mathematical analysis depends on the exact conditions under which
this property holds.
With two variables, we can write the quadratic form
F(z, y)=
"
adopting a process (familiar in elementary algebra) known as com-
"
pleting the square in the variable x. It follows that F'(#, y) is only
positive for all values of x and y not both zero, if the two squares in
the above expression have positive coefficients, i.e. if a and (ab - h 2 )
are both positive. Similarly, F(x, y) is negative under the same
conditions only if the coefficients are negative, i.e. only if a is nega-
tive and (ab-h 2 ) positive. These conditions are easily seen to be
both necessary and sufficient. Writing the expression (ab
- h 2 ) in
determinant form, we have the result :
The quadratic form (ax 2 + by 2 H- 2hxy) is positive for all values

of x and y (other than x y 0) if, and only if,
a> and a h >0

h I
and negative for all values of x and y (other than x~y = 0) if,
and only if, the same expressions are negative and positive
respectively.
It maybe objected that the condition written is unsymmetrical

in the sense that the coefficient of x* rather than that of y 2 is 9 ,
selected for determination of sign. In fact, if we complete the

square in y instead of in x, we find that F(x, y) is always positive if
b and (ab - h 2 ) are both positive and always negative if b is negative
and (ab
- h2 ) The essential point, however, is that these
positive.
second conditions are equivalent to the first since (ah - h 2 ) is only

positive if ab is positive, i.e. if a and b have the same sign. Our con-
ditionis thus quite symmetrical. It may be objected also that the
determinant notation is not necessary here. This is true but the
notation is introduced to facilitate the generalisation of the result
to cases ofmore than two variables. It is to be noticed that the
determinant used is most easily written from the coefficients of the
quadratic form when arranged :
F(x, y)~ax + hxy

2
+ hxy -h by 2 .
The determinant is called the discriminant of the quadratic form.

In the case of three variables, we write
F (x, y, z) ax 2 by 2 + cz 2 + 2fyz -f 2gxz + 2hxy
-{-
g \
2
ab h 2 ac g 2 2 af gh
x
a
*u
(ha /
z ~\
a
1/ H ~\
a
~ z
2
\ -\ ~\~ &
a
yz
g \ ab h2 f af gh
(h
x+
a
y-\
a z]
/
-\
a
2
I
\
y -\ r
ab
,-;
n*
2
-ch 2 + 2fgh 2
+ abc-af -bg z '
ab~^h?
" "
where we have completed the square twice in succession. Now,
F (x, y, z) is positive for any values of the variables not all zero
if, and only if, the coefficients of the three squares in the above
expression are all positive, i.e. if
(ab-h
2
and - 2 - bg 2 - ch 2 4-
a, ) (abc af 2fgh)
are positive. Similarly, F(x, y, z) is negative for any values of the
variables not all zero if, and only if, the same coefficients are all
negative, i.e. if the three expressions written above are negative,
positive and negative respectively. The determinant notation is
again applicable and we have the result :
The quadratic form (ax 2 + by 2 + cz 2 -f 2fyz -h 2gxz + 2hxy) is positive

for all values of x~y and z (other than x y z 0) if, and only if,
a h >0 and a h g >0
h b h b f
ff f c
and negative for all values of x, y and z (other than x=y=z Q) if,
and only if, these expressions are negative, positive and negative
respectively.
It can be shown, as before, that the conditions set out here are
symmetrical and can be written in two equivalent ways by selecting
the coefficient b or c instead of a. The third-order determinant
written is called the discriminant of the quadratic form and is most
easily derived when the form is arranged :
F (x, y, z) ax 2
-f- hxy -f gxz
+ hxy + by 2 +fyz
2
+gxz+fyz -fez .
A more restricted problem of the same nature

that of finding is
the conditions under which a quadratic form preserves a given sign,

not for all values of the variables but for values satisfying a given
linear relation in the variables. The general method of
solving this
problem is quite simple. The linear relation gives one variable in
terms of the others and the quadratic form is then reduced to a form
in one less than the original number of variables. The previous
results apply at once. Suppose that the two variables of the quad-
ratic form (ax 2 -f by
2
-f- 2hxy) are related a# + /fy 0, where a and j8
are constants as well as a, b and Then - -x and the quadratic

c. y~
form becomes :
~-
p
2
p
which is x (and for the corre-
positive (negative) for all values of
sponding values of y given by the relation) and only if, the ex- if,
2 -
positive (negative) From an example
2
pression (6a a/? -f-
2Aa/J) is
given above (18.2), the condition can be written in determinant

form :
The quadratic form (ax 2 -f by 2 + 2hxy) is positive for all values of x

and y (other than x = y = 0) which satisfy <xx + /3y = Q if, and only if,
a <0
a a h
ft
h b
and negative under the same conditions if, and only if, the deter-
minant is positive.
Notice that the third-order determinant is obtained from the dis-
criminant of the original quadratic form by " bordering " with the
coefficients a and /J from the linear relation.
In the three variables case, the variables in the quadratic form

(ax
2
+ by + cz + 2fyz + 2gxz + 2hxy)
2 2
are related ax + fiy + yz = 0. Eli-
minating x= + yz), the form becomes

a (j8?/
where A=~(W = (ca

2
+ ay 2 - 2gray)
and r
= (afy-g*p- +/<x).
The form is positive for all values of y and z (x being given by the
linear relation) if and A> 2
AJ3-H >O y
i.e. if
and (be -/
2
)a
2
+ (ab - g*)fi 2 + (ah - A 8 )y a
- 2 (a/ - fl*)ft, - 2 (fy -
-/A) ay 2 (<* -fg) aj8 > 0.
The form is negative under the same conditions if the two expres-
sions given above are negative and positive respectively. It is
easily verified that the expressions are simple determinants and so :
The quadratic form (ax

2
4- by
2
+ cz 2 -f 2fyz + 2gxz + 2hxy) is positive
for all values of x y and
9
z (other than x = y z 0) which satisfy
=Q if, and only if,
a j8 |<0 and y <0

a a A h g
h b b f
f C
and negative under the same conditions if, and only if, the two
determinants are positive and negative respectively.
18.8 Examples of quadratic forms.

The following examples illustrate our results :
Ex. 1. (x 2 + 2y2 + z 2 +yz + 2xy) is positive for all values of x, y and z

which are not all zero ;
for
-1>0 and 1 =
f>0.
2
i
Ex.2. (x
2
-y2 -Iz 2 +xy) is negative for all values of x, y and z
subject to the relation x + y + 2z=Q since
-KO.
Or, eliminate z = - \ (x + y) and write the quadratic form
But (3#
2
+ lOxy + I ly2 ) is positive for all values of x and y since
3>0 and 3 5 = 8>0.

5 11
The original quadratic form is thus negative for all values of x and y
and for z = - J (# + y).
Ex. 3. 2/1? 2/2

an d 2/3 are given functions of three variables xlt x 2 and x3
such that
= =
3a? 2 9^' d#3 9^2
It is required to find the conditions under which
92/2
+ p
dxl
dx,l dx3z +
9a; 2
dx,^ dx3 +
9o:3
3
cfaJ
8
is negative for all values of dxl9 dx 2 and dx3 .* The conditions are
dx l
<0 9a: x 9x 2
>0 and
OXn OX*
OX-\
32/2 3?/ 2 3|/2 3^2 3^2

9^ 9^ 9# 3
3^3 3^3 3^3

dx l dx 2 dx3
i.e. the Jacobians ~~ ,

~ 2~
and ^ are negative, positive
and negative respectively.

* See
Ramsey, A Contribution to the Theory of Taxation, Economic Journal,
1927, p. 50.
18.9 Two general results for quadratic forms.

The problems considered above can now be generalised. The
general quadratic form in n variables xl9 x 2 # 3 ... xn is , ,
F(xv x 2 # 3 , ,
... a?n )=o ua1 a + a M a? a a + ...
+ 2a12xlx 2 -1- 2a13xlx3 -f . . .
which can be written more symmetrically as
ly
x 2 x^
,
... xn ) = anx^ + a12x x 2 + .
. + 0> 2nX 2Xn
K..+annsw '
where the a'sdenote given coefficients such that the order of the
suffixes is immaterial (a 2i = a 12 ...). The determinant J
, r |>
written down from coefficients of the quadratic form, is called the

" "
discriminant of the form. We also need the bordered determi-
nant
OC . . . OL n
OCj 2
where al5 a 2)k ... a n are some constants.

The quadratic form F(xly x 2 x a ... xn ) , ,
is said to be positive
takes only positive values for all permissible values of
definite if it
the variables which are not all zero. Similarly, the form is negative
definite if it takes only negative values. Two general results can
now be stated :
(1) The quadratic form is positive definite if, and only if,
A and all its principal minors are positive, i.e.
a n >0, an a12 >0

22 ^22
nl an2
and the form is negative definite if, and only if, the above
expressions are alternatively negative and positive.
(2) The quadratic form is positive definite subject to the

relation a^ + a 2# 2 + + <*nxn = between the variables if, and
only if, A' and all its principal minors are negative, i.e.
<0
aln
and the form is negative definite under the same conditions if,
and only if, the above expressions are alternatively positive and
negative.
The proofs ofthese general results follow lines similar to those
indicated in the particular cases already established and no further
details need be given here. The proof of (1) depends on the
reduction of the quadratic form to the sum of n squares, all the
coefficients of which must be positive (or all negative). The proof
of (2) is obtained when the quadratic form is reduced from one in
n variables to one in (n - 1) variables by means of the given relation.
EXAMPLES XVIII
Determinants
= abc + ab + ca + be.
Hence evaluate the determinants.

4. Show that a = 4abc
b
c c
by noticing that the determinant vanishes if (e.g.) a (
5. Show that 1 (18.3, (3) and (6)).

1
-1 -1
Deduce that the determinant equated to zero gives a cubic equation in x
with two roots x = a and a third root x = - 2a.
6. Prove that a a* a - 1
3
-(a&c-l) 1 a a 2
(18.3, (2) and (5)),
6 62 63 - 1 1 b 62
c c2 c3 - 1 Ice 2
and hence evaluate the determinant.
7. Factorise
8. Write down the co -factors of the elements of the second and third rows
of the second determinant of Example 1 above and verify the results of 18.4
in this case :
(1 + a)A 21 + A^ 1 23 and (1 +a)A 3l ^-A 32 + A 33 ^0.

9. Shdwthat u = ax c
'
(where a 4- /? + y l)isa linear and homogeneous

function for which
du_a,u u a(a-l)w " d zu ctflu

'" 9 " ' '
dx~~x' dx dy xy
Verify the determinant properties of 18.5.
Linear equations
10. Solve 2x - y = 2, 3y + 2z = 16 and 3z + 5x = 21.

11. If a, 6 and c are given constants, find the solution of
x+y+z a, x-y + z b and x -\-y-z c.
12. Use the determinant of 18.3, Ex. 2, to solve the equations
a? + 2/ + 3=-= 1, ax + by + cz = d, a 2 jc + b*y + c*z dz ,
where a, 6 and c are constants.
13. From the result of 18.6 verify that the equations
cvz + d^ - 0,
'3^0,
have a unique solution which can be written
where A 19 J 2 J 3 and J 4 are the co-factors of A 1? A 2 A 3 and A 4 in

, ,
A* A A3 A4
at 6j cx rfj
3 3 3
Generalise this form of the solution of linear equations.

14. Show that the solution of the three equations of the previous example
satisfies a fourth equation a tx + b^y + c t z + d 4 = provided that
aa 6a c2 d2
3 *>3
C3 dB
a o* CA d
This is the condition that four linear equations in three variables are con-
sistent. Generalise it.
Quadratic forms
15. that (4xy - 2x 2 - 3y 2 ) is negative definite and (x 2 + y 2 + z 2 - yz)
Show
positive definite for all values of the variables (not all zero).
and y take any values subject to 2x + y 0, show that (x 2 + y 2 + 3xy)
16. If x
isnegative definite. Is this quadratic form also negative definite for any
values of the variables ?
17. Show that (xy yz + xz- x -y - z )

2 2 2
-}- is negative definite for values of
the variables which are such that x + y + z 0.
18. If a, b, h, a andft
have positive values, show that tho quadratic form
by + 2hxy) always positive definite subject to ax - f$y 0.
2 2
(ax -f is
19. uf(x y, z) is a function of throe independent variables. Obtain the

t
second differential d 2u and find the conditions that it is positive dofiriite for
values of dx dy and dz, (a) without restriction and (6) subject to
9
du fx dx +Jy dy +fz dz 0.
Write the condition that (ax 2 + by 2 -f 2hxy) is positive definite subject
20.
to + fty 0. Show that the quadratic form is positive definite for any values
ocX
of a and under conditions identical with those required for tho form to
be positive definite for mirestrictod variation of x and y. Generalise and
indicate the relation of the conditions for relative definite quadratic forms
to those for unrestricted definite quadratic forms.
CHAPTER XIX
FURTHER PROBLEMS OF MAXIMUM AND MINIMUM
VALUES
19.1 Maximum and minimum values of a function of several variables.
THE extreme values of a function of any number of
definition of
variables can be put most precisely in the form (see 17.5 above) :
The function y=f(xl9 # 2 ... x n of n independent , )

variables has
a maximum (minimum) value at x1 =a l9 x 2 a 2 ,
... xn =a n if
/(^-fo^, a 2 -h# 2 ,
... <* n +xn ) -/(%, 2>
a n) negative (positive)
for all small values of x l9 x 2 ,
... x n .
The analysis which follows is set out, for convenience, in terms of

functions of two variables. It is, however, quite general in form.
assumed that the function z=f(x, y) has continuous partial
It is
up to any desired order at the point (a, b). Then Taylor's

derivatives
Theorem gives, for any integral value of n,
-
, b+y)-f(a, b)=df +-~d*f+d*f +...+ d"f+R n (x, y),
where the successive differentials of the function are taken at the

point (a, b) and with arbitrary increments dx x and dyy, and
where
-Rn (
x , y) =-\
Tfc
- T* 1
7 d n + lf at the point (a + Ox, b + Oy).
As before, 6 denotes some positive fraction.
(1) Suppose that df-Q for some variations in the variables from
the point (a, b). Then by the continuity of the partial derivatives
of the function, rf/^0 also for some variations from the point
(a + 0x, h + Oy), provided only that x and y are sufficiently small.
fl
o.,
Further, if +y r- happens to be positive for certain values

df=x^~
ox c/y
of x and y, then it must be negative
and y which are
for values of x
numerically equal but opposite in sign. Hence, df must take both

positive and negative values for different variations from the point
(a + 6x, b + 6y). Taylor's Theorem (n = 0) gives
f(a + x, 6 + */)-/(, b)=df at (a + 0x, b + 0y),
i.e.f(a + x,b + y)-f(a,b) is sometimes positive and sometimes
negative according to the values allotted to x and y. It follows, from
our definition, that /(a, b) cannot be an extreme (maximum or
minimum) value of the function. If z=f(x, y) has an extreme value
at (a, b), then df must be zero for all variations from (a, b).
Suppose that df=0 for all variations while d f^Q for some
2
(2)
variations in the variables from the point (a, b). By continuity again,
we can take d 2f^=0 also for some variations from (d-\-0x, b + 6y),
provided that x and y are sufficiently small. Taylor's Theorem,
with n 1, now gives
f(a, b) = ~d f
z
at (a + 0x, b + Oy)
and, using the definition of extreme values, we can distinguish three

cases * :
2
(a) If d f<0 for all variations from the point (a, 6), then z=f(x y) 9
has a maximum value at this point.
(b) If d zf> for all variations from the point (a, 6), then z =f(x, y)
has a minimum value at this point.
(c) If d 2f is positive for some variations and negative for others,

then z=f(x, y) has a stationary, but not an extreme, value at the
" "
point (a, 6), i.e. the point corresponds to a saddle point of the
function.
z
(3) If for all variations from the point (a, 6), it is
df=d f=0
necessary to proceed to the terms in the Taylor series which involve
the third and higher order differentials. The analysis, though
possible, becomes complex and, as the cases considered are not
often met with in practice, we need not examine them.
Collecting our results and extending them to the general case of a
function of any number of variables, we have
* It is
d*f< (or d />0) for all variations except some which
2
possible that
make d sf0. This isnot included in the three cases here distinguished and
further treatment is necessary to allow for such possibilities.
GENERAL CRITERION FOR MAXIMUM AND MINIMUM VALUES

(1) If y =f(x x ... xn has an extreme value at xl =a
ly 2) ) l9
# 2 a 2 ... xn ~an then dy = for all variations of the. variables

= , ,
from these values.
(2) If rfi/
= for all variations and d zy 7^0 for some variations
of the variables from the values (al9 a 2 ... a n ), then y has a ,
stationary value at this point which is a maximum value if

d zy<0 for all variations, a minimum value if d 2 y>0 for all
" "
variations and a saddle value if d y takes positive and
2
negative values for different variations.
The conditions given are respectively necessary and sufficient for

extreme values. But the criterion is not complete, i.e. not necessary
and sufficient, since no allowance is made for cases where d zy = for
allvariations from the values (a l9 a 2 ... a n ). ,
The
differential form of the criterion can be expanded to give
conditions relating to the partial derivatives of the function. The
necessary condition (dy 0) simply implies that all the partial
derivatives of y must vanish at an extreme value (see 14.2 above).
The sufficient conditions relate to the second differential
all the partial derivatives being evaluated at the point (al9 a 2 ... a n ). ,
A maximum (minimum) value of y is obtained if this quadratic form

isnegative (positive) definite for all variations dxl9 dx 2 ... dxn The ,
.
conditions for either case are given in 18.9 above.

In the case of two independent variables, a point where fx ~fv =
gives a maximum value of z~f(x, y) if
d 2z=fxx dx 2 +fxy dxdy
is negative definite, i.e. if
fxx <0 and fx

'.X
,
JXV
Jxv Jvv
The point gives a minimum value of z if the first inequality is

2 1 M.A.
reversed. The second inequality /which is the same in both cases
of extreme values, is
Jxxjyy^> (jxv)
The conditions are those previously given (14.2 above).
If there are three independent variables, then a point where
fx =fy =zfz = gives a maximum value of u f(x, y, z) if

2 2
d u -fxx dx +fxv dx dy +fxz dx dz
2
+fxv dx dy +/ dy +fvz dy dz
+/xz dxdz +fyzdydz +fZ9 dz*
is negative definite, i.e. if
/, /,xv >0 and Jxx Jxv Jxz

Jxv Jvv Jxv Jvv Jvz
Jxz Jvz Jz'.
The point gives a minimum value of u if the first and third inequal-
ities are reversed.
19.2 Relative maximum and minimum values.

The advantage of the differential form of the criterion for extreme
values is applies, with suitable restrictions, even when the
that it
variables are not independent. A function y=f(xl9 x 2 ... xn ) has ,
an extreme value relative to a number of given side relations,
<i (#15 # 2,
xn ) = 0,
. . .
2 (xl9 x 2 xn ) 0,
(f>
if the conditions previously
,
. . . . . .
,
written are satisfied, provided that the differentials of y are expressed

in terms of the independent variables only, the other variables being
eliminated by the side relations.
With a single side relation, we seek the extreme values of
y=f(xl9 x2 ,
... xn ) relative to (f>(xl9 x2 ,
... xn ) 0. The necessary
condition is that
dy fxi
dx -f/a;8 dx 2 +/a;a dx% + ... +fx n dxn =
for all increments which satisfy
Taking a^ as the dependent variable given by the side relation, we

can eliminate dxt and obtain
*
for all increments in x z xa> , ... xn .
Hence,
/&=/&=!=...=/..
These are the necessary conditions (see 14.6 above). To obtain
sufficient conditions, we write
f f /7 /y 2 _j_ f /7o* 2 tyf /it*
fj'y/ tviA/o
/"72/j/ ~~~"
if J 3Ci
/"72/yt
1.
I
'~J X Xi VvvUi "T" I {EoJCft ^/**

2 / "T~
i
* * *
i
'" J 5Pi JC*

^'* 1
\
i^ . . .
,
where ^ c?^ + <f> Xz

dx 2 + . . . + ^^ rfo; n
=
Eliminating d x^ by means of the last relation,

2
we have
where ^ a rf^ -f <^ a

.
i
,
2
da; 2 + . . . + (^^ rfo; n = 0.
A point which satisfies the necessary conditions gives a relative
maximum (minimum) value of y if d 2y is a negative (positive)
definite quadratic form for all increments dxly dx 2 ,
... dx n subject to
the relation written above. The conditions for either case, given in
18.9 above, indicate the appropriate signs for the principal minors
" "
of the determinant formed from the coefficients in d 2y, bordered
with the partial derivatives Xl Xz
... <f>y
n <f> , <f> ,
.
In the case of two variables, a point satisfying the necessary coii-
f f
~*
ditions <f>(x, y) and ^ gives a maximum value of z~f(x, y)
relative to <f> (x, y)

= if
fxy - dx dy
dx dy + / -
is negative definite subject to <f> x dx-\-(f> y dy

= Q, i.e. if
<* <
y
J f
/ so:
~J Jt ^ ~ ' x
1~
JL
<Pa; JT Y>ara: /a;i/ <?x\
<Px 9x
~ "
Expanding the determinant, the condition is
2
(/.-#..) ^,- (/-*) #.'- (/,-*) *,*><>,
i.e.
This agrees with the result previously obtained (14.6 above) when
f f^
we remember that * = by the necessary conditions.
19.3 Examples of maximum and minimum values.

The results of the previous two sections are illustrated by the
following examples :
Ex. 1. If u=x 2 +y 2 + z2 is a function of three independent variables,

then du = 2 (x dx -f y dy + z dz) and d*u = 2 (da;2 + dy* + dz 2 ).
Extreme values of u occur only where du=Q for all variations, i.e. where
x =y =z =0. Since d2u>Q for all variations from any point, these values
must give a minimum value of u. Hence, u has a single minimum value,
u = 0, which occurs where x y z 0. This is also clear since u is positive
except when x, y and z are all zero.
Ex. 2. If x, y and z are independent variables and
then du = (l -2x)dx + (z-2y)dy + (2+y-2z)dz

eZ w = - 2 (cfa 4- dy + dz ) + 2di/ dz.
2 2 2 2
and
Extreme values of u can occur where
i.e. s = i, t/
= f, 2=|, giving
But d ^ 2
is negative definite at all points since
-2<0, -2 =4>0 and -200 -6<0.
-2 -2 1
1 -2
Hence, u has a single maximum value ^f .
Ex. 3. To find the shortest distance of the origin of co-ordinates from

the plane with equation ax -f by +cz d :
Let u be the square of the distance from the origin to the point with
z) on the plane.
co-ordinates (x 9 y, Then
u=x2 + y2 +z2 where ax + by 4- cz = d.
PEOBLEMS OF MAXIMUM AND MINIMUM VALUES 501
The relative maximum of u is required. The necessary condition is
du = 2 (x dx + y dy + z dz) =0,
subject to a dx + b dy + c dz = 0.
So ?=|=?-
a b c
Since ax + by + cz d, these equations give

ad bd cd
and the corresponding value of u is
U=s
Taking the equation of the plane as giving a; as a function of y and z, then
d2x = 0, and
d2u = 2x d2x+2(dx2 +dy2 + dz2 ) = 2(dx 2 + dy 2 +dz2 )>0
for variations from any point. The positions obtained is thus one of
relative minimum. The shortest distance required is the square root of
the minimum value of u, i.e. it is
d
Ex. 4. To find the rectangular block of maximum volume we can cut

from a sphere of radius a (14.7, Ex. 2), we require the maximum value of
V = 8xyz relative to x2 + y2 + z2 =a2 The necessary condition is
.
dV = 8(yz dx+xz dy+xy dz) = subject to 2(x dx +y dy +z dz) =0.

yz xyxz
- == _=_^.
Hence,
x y z
2 2 2
=a2 we x =y =z V
Using x + y + z , find
=-^
, giving
=^
Now, d2 V = Syz d2x -f 16 (z dx dy + x dy dz -f y dx dz),
where xdx+ydy + zdz==Q
and xd2x + dx2 +dy2 +dz* =0,
where the given relation determines x as a function of y and z. On
eliminating d*x, we have
d2 V = 8{2z dx dy + 2x dy dz+2y dx dz-~(dx2 +dy 2
x
where xdx + ydy+z dz =0.
d
At the point where x=y^z=^-j^ ,
subject to dx 4- dy + dz = 0.
For a relative maximum of V, the quadratic form
- dx* + dxdy + dx dz
+ dxdy-dyz +dydz
+dx dz+dy dz-dz2
must be negative definite subject to dx+dy + dz=0. The conditions for
this are satisfied since
1 1 = 4>0 and 0111

1-111
1 1 -1 11-11
8a3
111-1
Hence, V '* is the relative maximum value of V.
19.4 The stability of demand for factors of production.

The demands of a firm or industry for two factors of production
(14.8 above) are stable if the constant product curves, obtained from
the given production function, are convex to the origin at all relevant
points. The conditions for equilibrium and stability of production
can now be examined in more general cases.
A good X
produced with n variable factors A l9 A 2) A 3 ... A n
is ,
according to the production function x~f(a1) a 2 a 3 ... an ), which , ,
is assumed to have continuous partial derivatives of the first two
orders :
-= and (rand =
Our problem is, given the market prices pl9 p 2 p& ... p n of the
first ,
factors, to find the grouping which produces a given output x at mini-

mum cost. We
have to minimise IIa
1 pl +a 2 p z + a.3 p 3 + ... +a n p n
relative to/(%, a 2 a 3 , ,
... a n )=x. The necessary condition is
dll=p1 da^ +p 2 da z + p 3 da 3 + ...+p n dan = 0,

subject to /! da,i +/, da 2 +/3 rfa 3 + +/n dan = 0,. . .
\- ?l^^_^3_ "' _^
A / /. fn'
There are (n-1) equations which, with the given side relation,
determine the equilibrium employments of the factors in terms of
the given output and the given prices of the factors. We thus have
the demand for each factor and we can write
a r ^<f> r (x,plt p Z) p 99 ...p n ) (r
= l, 2, 3, ... n).
Taking the side relation as giving ax as a function of a 2 a 3 , ,

... an ,
then, differentiating twice, we have

2 *
A +/22 da 2 + ..
2
d a +/u da
But d*II = d (P! dal +p 2 da 2 + ...+p n dan = )
For a minimum value of II, we have d 2II> and

2
Ai da>i +/i2 âi da 2 +/13 dal da 3 + . . . -f An dax dan
2
+/i2 d% da 2 -f /22 rfa 2 -f/23 da 2 da s 4- . . .
-h/2n da 2 da
+
is a negative definite quadratic form subject to
A dax 4-/2 rfa 2 -f /3 ofa 3 + +/n da n = 0. . . .
From 18.9, (2), a sufficient condition is
A
/X /a
I/A > ' A
I/X /a
/A /s
/ O
/i /ii Aa A /u /i2 /is
/2 /12 /22 /2 /12 /22 /23

/3 /13 J23 /33
Suppose that/!, /2 /3 ... /n are all positive and that the inequalities
, ,
above are satisfied for all combinations of factors within a relevant

range. Then, for any set of outputs and prices of the factors (within
a certain range), equilibrium is possible and the position determined
is stable. We have the stability conditions for the demand for
factors of production, conditions which are given as limitations on
the form of the production function. For a production function
satisfying the conditions, the demand functions (<f> r ) for the factors
are uniquely determined and stable. If there are only three factors,
it is easily shown that the stability conditions imply simply that
the constant product surfaces in factor space are downward sloping
and convex to the origin at all points, at least within a certain range.
19.5 Partial elasticities of substitution.

Werequire the following notations which represent certain
features of the production function xf(al a 2 a3 ... an ). For any
, , ,
combination of the factors, define
(r
= l, 2, 3,...n).
Hence, KI + * 2 -f s + ... + *
= !. /c
At an equilibrium position, fl9 /2 /3 .../n are proportional , ,

to the
prices pl9 p 2 p s ... p n of the factors.
, , Hence,
i.e. KT is the proportion of total cost (77) which is spent on the

factor Ar .
Let F= A /, /, /,
/I /ll /1 8 /13 /In
/2 /12 /22 /23 /2n
/n /In /2w /3w Jnn

and let Frs denote the co-factor of frs in F (r and s = 1, 2, 3, ... n).
Write (,and = l,2,3,...n).

For any unequal values of r and s9 the value of <r rs is called the
partial elasticity of substitution of the pair of factors -4 r and s (as A

against all other factors). Its value depends on, and varies with,
the grouping of factors employed. The interpretation of the partial
elasticities of substitution will appear in the following section and it
can be shown that they are related to the ordinary elasticity of
substitution betweentwo factors as defined in 13.7 above (cf.
Examples XIX, 9).
A number of relations exist between the values of a rs for various
values of r ands. If each element of the first row of F is multiplied
by the co-factor of the corresponding element of another row, then

the sum of the products is zero (18.4 above). So
/l^rl+/2^r2+/3^r3+-..+/n^rn = (f
= 1, 2, 3, ... n).
From the definitions above, we have
So /qor rl + fc 2 or
r2 -f K^<j rZ + . . . + Kn arn = (r
=1 , 2, 3, . . .
n).
We can write this relation in the form
The last two inequalities of the stability conditions (19.4) imply that
F and Fnn are of opposite sign. The same is true of F and F rr for
any value of r.* But a rr is proportional to the ratio of F rr to F

and so
cr rr <0 (r
= l, 2, 3, ...n).
Hence, for a production function satisfying the stability conditions,
for any value of r (r

= 1, 2, 3, ... n). In any case, there are n relations
limiting the values that can be taken by the partial elasticities of
substitution for any grouping of the factors. f
The values of the partial elasticities of substitution can be positive
or negative. But the limitations above show that the positive values
must be more numerous or important than the negative values. In
particular, the (n
- I) partial elasticities of substitution between any
one factor and the others cannot all be negative. If there are only
three factors (A l9 A and A then
2 3 ),
/c 2 o-12 H- 13 > 0,
/c 3 c7 Kior12 -H 3 23 > /c o- and /c or
1 13 + /c 2 (T23>^-
It follows either that all three partial elasticities of substitution
(<712 , and or 23 ) are positive or that one of the

o-
13 partial elasticities is
negative and the other two positive.
19.6 Variation of demand for factors of production.

It is now assumed that there are constant returns to scale in the
production of the good X, the production function x =/(! # 2 # 3 an) , ,
. . .
being linear and homogeneous with the properties set out in 18.5
above. Given the output (x) and the prices (pl9 p 2 p%, ,
... p n ) of the
factors, the demands for the factors are determined by
Pi == P^ =z P^ = JP*=\
/I /2 /3 Jn
and /(!, a 2 a 3 , ,
... a n )=x.
The average cost of the output x is
U
X X X
since a lt/i -f a a /2 + 3 /3 + . . . + an fn x by Euler's Theorem. Further,
* The order in which the factors are enumerated is immaterial ; we can
take the factor A r last just as well as the factor A n .
t The other conditions of stability give rise to further limitations on the

partial elasticities of substitution.
ifthe output varies while the prices of the factors remain fixed, the
marginal cost at the output x is
dU da*
--
da* da* dan
. da* f da* dan

-~ -f-/3 da* , - .
since /! -7-= H-/a +/n -~-

fl
- = 1 from the production
-f . . . lunc-
.
ax ax
doc ax
tion. Average cost and marginal cost are equal and, therefore,
constant for all outputs.
Extending the problem of 14.8 above, it is assumed that is sold X
on a competitive market at a price p equal to the constant average
cost. Then the common equilibrium value of ~, ~ ~

, ,
... ~ *8
equal to p, i.e.
Jl h h Jn
Pr=Pfr (r
= l, 2, 3, ... n),
"
which is the law of marginal prodiictivity ". Let the demand of
the market for X

be given by x (f>(p) with elasticity 77 ~~.
Then the position of competitive equilibrium is described by
f(a l9 a 2 , 3, ...an ) = <fi(p),
and
cWIU Pi_P*_P* ----
= -Pn-v
f f f f p.
Jl /2 /3 Jn
These equations determine the values of a1? a 2 a 3 ... a n and p in , ,
terms of the given prices of the factors. The output x = <f>(p) and
the total cost U=xp= a1 p 1 + a 2 p% + a 3 p 3 + ... +a n p n are also given
in similar terms.
The demands of the competitive market for the factors are deter-
mined, by the equilibrium conditions, as functions of the prices of
the factors. Let one of these prices (p^) vary while all the other
prices remain fixed. The
resulting variations in the demands for
the factors are then to be found by differentiating the equilibrium
equations (which hold for any prices of the factors) with respect
to ft. From /(<%, a 2 a 3 , ,
. . . an ) = <
(p),
dai Ba 2 dan _ xdp

From pf =pl9 pfz =p 2y ...pfn =p n ,
Hence
idp da, da, da

r=.
7
jF stands for the determinant already written (19.5 above) and
J^ is the co-factor of the element in F. Further, F rs and F 0r3
denote the co-factors of the element frs in F and ^ respectively
(r and a
= l, 2, 3, ... n). Then, by 18.5 and 19.5,
a
^0, F rs = fFK a rs =^FKrars
s ,
F^-Zgp,
x
Js Jr
where a rs is the partial elasticity of substitution of Ar and A s and
is the proportion of total cost ( = total receipts) spent on A r .
Using the determinant notation (18.6 above) to solve the linear
equations above for ^

dpi
,
we find
da-L
With the aid of 18.3, (5), the numerator can be written
2>
since pft =pt and ^=
Hence,
i.e.
Proceeding in exactly the same way, we find
and similar results for the demands for other factors.

The results, though expressed in terms of variations in the price
Pi, are clearly quite general. We can, in fact, write one formula to
express the effect of a change in the price of any factor on the demand
for any factor :
.^_J
Z_f = ! (QT "
-.-n) (r and s = l, 2, 3, ... n).
Ep r a 8 dp r
These expressions for the partial elasticities of demand for the
factors are of the same form as those obtained in the case of two
factors (14.8 above). Their interpretation proceeds as before but
some new conclusions are now reached. If the market price of one
factor A r rises, then the demand for this or any other factor is affected
in Firstly, the cost of production is now higher and the
two ways.
product dearer. For a decreasing demand law (^ 0), the amount of >
the product sold is less and there is an all-round and proportional
decrease in the demand for the factors. This effect is shown by the
negative term (
- K rrj) which appears in each of the elasticities of
demand with respect to the price of A r .
Abstracting from this effect,
the factor Ar is now relatively more expensive than other factors
and pays to substitute other factors for
it r in production. The A
demand for A r thus decreases on account of substitution, as shown
It/a
by the negative term (/c r o> r ) in
^
^ .
(By 19.5 above, a rr is negative
for stable demands.) The effect on the demand for one of the other
factors is more complex but the net result (on the demand for^4 s )
shown by the term -- There are two cases

is (K r a rs ) in =r . :
(1) If o> s >0, then the demand for A 3 increases on account of

substitution ;
A
takes part in the replacement of A r in
the factor s
production. In this case, the factor A s is said to be competitive

with the factor A r at the grouping of factors considered.
(2) If v rs < 0, then substitution results hi a decrease in the demand

for A s ;
the factor A s, like the factor A
has been partly replaced
r,
by other factors in production. Here, the factors A r and A s are

said to be complementary at the grouping considered.
The sign of o> s therefore, indicates whether r and
,
A A s are com-
petitive or complementary factors. From the limitations described

in 19.5, we see that competition between factors is, on the whole,
more general than complementarity. One factor, in any case, cannot

be complementary with all other factors. Where there are only
three factors, for example, either all factors are competitive with
each other or one pair of the factors is complementary while the
other two pairs are competitive.
19.7 The demand for consumers' goods (integrability case).

We have considered demands of a consumer with
(14.9 above) the
a given income when there are two goods obtainable on a market at
given prices. The demands are stable if the indifference curves of
the consumer are downward sloping and convex to the origin at all
relevant points. In extending the analysis to the general case where
there are more than two goods available, we assume first that the
scale of preferences of the consumer can be represented by an in-
difference map and a utility function. Complications relating to

" "
integrability are deliberately avoided in this first approach.
If there are n goods X X Xl9 2, 39 ...
Xn ,
let one form of the utility
function of the consumer be
X2 X3) ,
...Xn
with continuous partial derivatives of the first two orders :
+'= and += (
The consumer has a given income and can purchase the goods at
p,
given market prices, p l9 p 2 # 3 , , p n He makes his purchases so

that u has a maximum value subject to the condition expressing the
fact that he must balance his budget :
X1pl + X 2p 2 + X3p 3 + ...+ Xnp n =fJL.

Since each utility function increases and decreases with any other,
a maximum of u corresponds to a maximum of any form of the
utility function and our results are thus independent of the fact that
utility is not a measurable concept. The necessary condition for the
maximum value of u we seek is
du = ^ dx + (/> 2
dx 2 -f ^ 3 dx 3 + . . . 4- <j> n dxn = 0,
subject to pl dxl 4- p 2 dx 2 +Ps dx 3 -f . . .
+p n dxn = 0,
ie jk == ib = = "" = ^2.
! -
Pi Pz Ps Pn
These (n-I) equations, with the given side relation, determine the
equilibrium purchases of the consumer in terms of his given income
and the given prices, i.e. the consumer's demands as functions of
P>Pi>P*>P*> Pn-
The condition of " budget balance " gives xl as a linear function
of the other purchases and so d 2a?1 = 0. It follows that a sufficient
condition for maximum u is that
= u dxj 2 +
<f>
<
12 dxl dx 2 4- <
13 dx: dx3 + . . . + <f>m dxl dxn
+ <f>i*
&x\ dx^ -f $22 dx2 * + <f> 23 dx2 dx3 + ...+ <f> 2n dx 2 dxn
4- .........................................................
is a negative definite quadratic form subject to
i.e. to
(using the necessary conditions).

Hence, by 18.9, (2),
& <f> 2 fa 02
fa fal faz 01 0n 012
fa 012 022 02 fa*
03 013 0!23
The equilibrium position and the demand functions are stable for
any income and market prices if ly 2) 3 ... n are positive and if
<f> <f>
<
, <f>
the determinant inequalities above are satisfied for all purchases of

the consumer, at least within a relevant range. These stability
conditions, it should be noticed, are limitations on the form of the
utility function. When there are only three goods, the conditions
imply that the indifference surfaces of the consumer are downward
" "
sloping and convex to the origin at all points, i.e. the normal
form of the indifference map is sufficient for stability of demand.
The stability conditions appear in a form which involves a parti-
cular utility function 0, and it is not clear that they are independent
of the non-measurable character of utility. We
can, however, trans-
late the conditions so that they involve only the marginal rates of
substitution between the goods (1G.8 above). The ratios
express the marginal rates of substitution of t for 2 Jf 3 n X X , ,

. . . X
respectively. The necessary conditions for equilibrium show that
these ratios are equal to the corresponding ratios of market prices,
i.e. the equilibrium purchases of the consumer are such that each
marginal rate of substitution equals the ratio of the prices of the

goods concerned. Now
0i
and similarly for the partial derivatives of the other marginal rates
of substitution. The determinant inequalities of the stability con-
ditions, using these results, can be reduced to
03
dx2 a^ \0!
__
<f>
dx3
The stability conditions thus involve only the marginal rates of
substitution and are independent of the actual form chosen for the
utility function.
Exactly as in 19.5 above, we define
r- and o>.
\-xn </> n "*
(r and s 1
2, 3, ... n),
where <P rs is the co-factor of <f> rs in the determinant
X <p 2 03 0n
01 011 012 013 -
01n
02 012 022 023 02n
0n 01n 02n 03n 0*n

At the equilibrium position, K T represents the proportion of total
expenditure (p = xl PI + x2 p z + + x n p n ) which is spent on the good
. . .
X r .The value of a rs is defined as the partial elasticity of substitution

of the pair of goods r and X X
8 in consumption.
The demand of the consumer for each good is defined as a function

of ft, ply p 2 p 3 ... p n by the conditions
, ,
Pi P2 P3 Pn
and xlPl + x zPz + X3Ps + + X nPn I*"
The parameter A (dependent only on /x, pl9 p 2) p 3 ... p n is introduced , )
for convenience and often described as the "marginal utility of
money ". Following the method of 19.6 above, these equations can
be differentiated with respect to /* or to one of the prices to give the
variations of demand for any good :
Ex s jj,
dx s , Ex 8 p r dx a ,
Ep, xs dfi ffipr %s dpr

It is found (see Examples XIX, 17 and 18) that
Ex s
demand for each good is affected in two ways

It follows that the
by an increase in the price of any one good (X r ). Since the money

income of the consumer is fixed, the increase in the price results in a
lower real income and causes a change (usually a decrease) in the
\
- Kr
ux
X
(Ex^' -^ )
in the
expression for
~-
f .
Further, the good r is now relatively more
expensive than other goods and the consumer proceeds to substitute

other goods for r in consumption. X
The demand for r is thus X
decreased, as shown by the negative term (K r a rr ) in the expression
ri
for =-- .
(The value of cr rr is negative by the stability conditions.)
The effect of substitution on the demand for one of the other goods
(X s ) is represented by the term (K r a ra ) in

-- . The consumption of
&Pr
X s increases or decreases on account of substitution according as
this term is positive or negative. Hence, r and 8 are competitive
X X
in consumption if a rs >0 and complementary if cr rs <0. As in the
production problem, the stability conditions impose limitations on
the values of o> s (r and $ = 1, 2, 3, ...?&) which imply that the com-
petitive relations between the goods outweigh, on the whole, the
*
complementary relations .
19.8 Demands for three consumers' goods (general case).

In generalising the results of the previous section, we must assume
that the preference scale of the consumer is defined, not by a utility
function and a complete indifference map, but only by the marginal
rates of substitution between the various goods at different levels of
consumption. To simplify the exposition, we take the case where

there are three goods X, Y and Z and make use of the notation and
results of 16.8 above. Each possible set of purchases of the con-
sumer is represented by a point (x, y, z) in space referred to axes
* We have
given here definitions of competitive and complementary goods
which are more strict than those previously suggested (12.6 above). On the
ffr l^v
old definition, X r and X. are competitive goods if ? and -~ are both
T?<y / Ifr \ -*^Pr -^Ps
positive. From -^
- = KT ( ara - -- ) , wo see that the new definition (<rrs > 0)
&pr \ kiiL* jgx jgx
usually implies the same thing. But one or both of -=~ and -=~ can be
Ex Ex ?f P*
negative if -~~ or -=~ is positive and greater than ars , i.e. if the effect on
Z&/Z jfejLt
demand via changes in real income is stronger than the substitution effect.
Similarly, the two definitions of complementarity are roughly, but not exactly,
in agreement.
2K M.A.
Ox, Oy and Oz. The preference scale of the consumer is then
described by an indifference plane dx + E^y + B^dzÔ at any given
point of space. Here, S
1 and J5 2 are given functions of x, y and z
(the purchases of the consumer) which express the marginal rates

of substitution of Xfor Y and Z respectively. The functions are
assumed to have continuous partial derivatives. The differential
equation of the indifference plane is not necessarily integrable and

we are not at liberty to assume that a utility function or a complete
system of indifference surfaces exists.
With given market prices, p x p y and p z) of the goods and a given
,
income /z, the purchases of the consumer must conform to the con-
" "
dition of budget balance (xpx + yPv + zpzp*) an<* any variations
in the purchases from an established set (x, y, z) must satisfy
p x dx +p y dy +p z dz 0.
If the purchases of Y and Z are increased by dy and dz respectively,

the necessary decrease in the purchase of X is
The indifference equation shows that the compensating decrease is
dz.
If the necessary decrease is less than the compensating decrease,

then the consumer tends to increase his purchases of Y and Z as
indicated. Conversely, if the necessary decrease is greater than the
compensating decrease, the consumer will tend to decrease his pur-

chases ofY and Z. In fact, equilibrium of consumer's choice is
only possible if the necessary and compensating decreases are equal
for all values of dy and dz. Hence, the necessary conditions for
equilibrium are
Rl==^ and E 2 =^
Px Px
and the marginal rates of substitution, as before, are equal to ratios
of the market prices.
To examine the stability of demand, we suppose that increases
dy and dz in the purchases of Y and Z have been made from
the equilibrium position, together with the necessary decrease
dy + ~ dz } in the purchase of X. The new values of the marginal

(--
P* P '
rates of substitution are approximately (Rl + dRl and )
where
dR dR*
-^~
dx
dR-i
-dy
-
dy
,
+ --~dz
dz
dR-t ,
and
T -,-n
2
*
dRtt2
dx
,
^ dy
9 ,
Suppose, now, that an exactly similar change from the new purchases
iscontemplated. The necessary decrease in the purchase of is X
Px Px
but the compensating decrease in X is now
(
- dx) c = (Rl + dBj} dy + (B* + dR 2 ) dz.
The original purchases are stable provided that this second change
will not be made, i.e. provided that the compensating decrease in X
is less than the necessary decrease :
dy H- (R 2 + rf^ 2 ) dz<dy + dz =B l dy + R 2 dz,

Px Px
i.e. dRl dy + dR 2 dz < 0,
where the variations in purchases are related
p x dx+p y dy+p z dz Q or dx + Rl dy-{-R 2 dz Q t
using the necessary conditions. Substituting for dRl and dR 2 ,
9jR 2
i .e .
^ 1 n
dx 2 + - ~ -l ^
dx dy+-
T
-~ dx dz , ,
-
2 dx dy 2\dy dz
-
~z
2 dz %\dy dz J
y T
dz
must be a negative definite quadratic form, subject to
dx + R! dy + R 2 dz = 0.
Hence, from 18.9, (2),
1 R t
1
51*
2 dx dy
R,
IdRi
2 dx 2~dx
dR1
+
2\dy dz)
1 7l 7? 1 /
I .
\\ dR,
2 dx 2 \ By dz J dz
The demands of the consumer, given by the necessary conditions,

all prices and incomes if .Ka and R z are positive and if
are stable for
D>0 and A <0 for all purchases. These are the general stability
conditions. Since an indifference map need not exist, the conditions
cannot be interpreted as before.
A certain simplification can be made :
D=-
19B,
2 dx
dx
dx dy
Further, write
dx dz
dx dy dz
By manipulation or expansion of the determinants, it can be shown

that
where ~ a
" a ~ W_!
\ dz
The stability conditions (Z>>0 and A <0) thus reduce to

B, <0 and B, >A 2 .
dx dy dz
SB*
dx dy dz
These conditions are very similar to those of the integrability case

but the third-order determinant shown must not only be positive but
also greater than a certain positive amount A 2 The conditions of .
the integrability case must still hold but they are only necessary,
and not sufficient, for stability. Notice that the integrability con-
dition (16.8 above) is simply A and that the general stability
conditions then reduce to those previously written.
The demand of the consumer for each of the three goods is deter-
mined, as a function of /z, p x p y and p z by
, ,
the conditions :
9 2
Px Px
This is true of all cases. In the special integrability case, the forms
assumed by the marginal rates of substitution and by the stability
conditions are simpler than in the general case. The results (19.7
above) relating to the variations of demand and the partial elas-
ticities of substitution hold only in the simpler case and need
modification when we
pass to the general case. In particular, the
distinction between goods which are competitive and goods which
are complementary in consumption is found to be less clear-cut in
the general than in the special case.*
EXAMPLES XIX
General maximum and minimum problems
1. Show that u x 2 + y 2 4- z a + xy has a minimum value at x = y z = 0.
2. Show that u = x* -f y + xyz has minimum values at all points where
2
x y= and z< 2 numerically.

3. If u = x 4- y + xz - re 2 - y 2 + z 2 show that u has a single stationary value
,
which corresponds to a saddle point.

* An
analysis of the competitive and complementary relations between
goods and of the variations of demand in the general case are given by Hicks
and Allen, A Reconsideration of the Theory of Value, Economica, 1934.
The stability conditions given in this article are not fully expressed and the
present development can be taken as replacing the earlier work.
4. Show that u Slxyz, subject to x + y + z = 3a, has a maximum value
(u = a) when z x~y
a. Deduce that the geometric mean between three
positive quantities is always less than the arithmetic moan except in the case
where the quantities are equal. Generalise the result to apply to any number
of quantities.
5. Find the shortest distance from the origin of co-ordinates to any point
on the surface xyz = a in the positive quadrant.
6. If a is a positive constant, show that the maximum value of
u a (a - x) (a - y) (a - z)
relative to x + y 4- z 2a occurs where x, y and z are each equal to -fa. Inter-
pret this result in the light of the fact that the area of a triangle with sides
x, y and z is given by Va(a - x) (a - y) (a - z), where 2a is the perimeter.
Economic maximum and minimum problems

7. A
monopolist produces three goods lt 2
X X X
and 3 The total cost of .
outputs x 19 x z and x s of tho three goods is (x 1 x l + a 2 # 2 -f a 3 # 3 ), whore lf a 2 and

3
are constants. The monopolist charges prices p 19 p 2 and p 3 and the demands
of his market are
- a (p - ) - a (p - a ),
,) 12 2 2 13 3 3
where 10 # 20 x 30 and the Find the prices the mono-

ic , , a's are given constants.
polistmust charge for maximum joint revenue and show that the following
conditions must be satisfied by the a's :
an >0, a n a 12 >0 and
Extend the results of 14.4 above.
8. Find the conditions under which the equilibrium of tho general problem
of 14.5 above is stable, expressing them in terms of the partial derivatives of
the production function x~f(a, 6, t).
9. A good X
is produced with three variable factors according to the
production function x~f(a lt a 2 a 3 ). Show that cr l2 (as defined in 19.5 above)

,
is a negative multiple of the ratio
*dfe]
a3
\a > 2
where the apply to a variation of the factors A l and A 3 (the

differentials
factor A 2 being held constant) so that product is unchanged. Express this
result in terms of an elasticity of substitution of the kind defined in 13.7 above.
10. Three factors are obtainable at given prices, p lf p 2 and p 3 When .
amounts a l9 a 2 and a3 of the factors are employed, the output of a good X is

x = A$ aâ^. The good is sold at the competitive price p on a market with
demand law x Bp~*. Find the amounts of the factors demanded in terms
of p l9 p 2 and p 3 By differentiation, show that
.
Ea l
11. For the production function of the previous example, show that the
demands for the factors are stable and verify that the partial elasticities of
substitution are all equal to unity. Hence use the results of the previous
example to check the formulae of 19.6.
12. Agood X is produced with n factors of production, the production

function, x=f(a lt 2,
a 3 ... an ), "being of any given form. The equations
,
?1=|=?S=...=?5
= A and f(a v a v a v ...a n ) = x
Ji J* Ja Jn
give the demands for the factors in terms of the given output (x) and the given
prices (p lt p z , p a ...p n ) of the factors. By differentiating these equations,
,
show that the variations of the demands are given by
when output varies (prices of the factors fixed), and by
&"?--" <*= '.. >
when the price of one factor varies (output and prices of other factors fixed).
Here, s
Fis the co -factor of fs in the determinant and KT and ars have the F
meanings of 19.5. Interpret the second results in terms of the competitive
and complementary relations between the factors.
13. A monopolist sells a good X

at a price p on a market with demand law
x<f>(p) and fixes output so that net revenue is a maximum (i.e. so that
marginal revenue equals marginal cost). He uses n factors of production,
obtainable at given prices (p l9 p^ 9 p 99 ---Pn^ an(l hi s production function,
x=f(a l9 cr
2, 3,
... linear and homogeneous.
a n ), is Show that the price p and
the amounts of the factors employed are given by
and
i A /3 Jn ^ r/ i
where 77
is the elasticity of the demand for the good.
14. In the problem of the previous example, the price of the factor l
is A
varied while the prices of other factors are fixed. Differentiate the equilibrium
equations and show that
Ea r p. da r _
ft
-
=rl 2>
EPl a r ^r-
dPl N
i lr
r
where ^ 1 and a lr have the meanings of 19.5. Deduce that, in the normal case
where rj > 1 and ~>0, a fall in the price of A, increases the demand for each
dp
factor less in this monopoly case than in the case of competition (19.6 above).
15.A consumer has a given income (^) and can buy three goods at given
prices (p lt p z and 3 ). One form of his utility function is u x^xfixf*, where
>
a x , a 2 and a 3 are positive constants. Find the demand of the consumer for
each good in terms of /x, p l9 p 2 and p 3 , and show that
with similar results for the other two goods.

16. Show
that the utility function of the previous example satisfies the
stability conditions and gives partial elasticities of substitution which are all
unity. Use the results of the previous example to verify, in this case, that the
general formulae of 19.7 hold.
17. The equilibrium equations of 19.7 give the demands of a consumer as
functions of his income (/*) and of the prices (p 19 p z , p^, ... p n ) of the n goods
obtainable. If income varies while prices remain fixed, differentiate tho
equations with respect to p and show that
Exr -- --
-= P dx
EH xr op,
- r _x -- --
1 <f> l
-\-x 2 </> 2
xr
+ ...+x n </> n &r
-=-
<P
\r i, ., o, ... n)
where the utility function and the notation of 19.7 is adopted.

18.In the problem of the previous example, the price of tho good l
varies X
while income and the other prices remain given. By differentiating tho
equilibrium equations with respect to p 19 show that
.
where K and a lr have the meanings defined in 19.7.
19. Tho scale of preferences of a consumer for n goods is defined by the

differential equation
(a x + a^xj dXi + (a z + a aa.r a dx 2 +

) . . . + (a n f a nn x n ) dx n =
whore x l9 x z a? 3 , , ... x n represent any set of purchases made by the consumer
and where tho Show that a utility function exists and
a's are constants.
verify that the demands of the consumer are stable, provided that c/ n or 22 , ,
33 ... a nn arc? all negative.

, Find the demand for each good as a function of
the given income of the consumer and of tho given market prices. Show that
the demand depends linearly on tho income (prices fixed) and is a ratio of
quadratic expressions in the prices (income fixed).
20. A firm with given resources (labour, plant, raw materials, otc.) produces
n goods and its outputs of tho various goods arc related by a transformation
function F(x l9 x z x a ... x n Q.
, Tho goods aru sold at given market prices,
, )
p lt p 2t p 3 ... p n and the firm fixes tho outputs to maximise its revenue
, ,
E = x p +# 2 /> 2 + ... +x n p n Find equations giving tho amount of each good

l l
.
supplied as a function of all the market prices. Under what conditions is the
supply stable ? How can partial elasticities of substitution between the goods
in production be defined and used to describe the variations of supply of
different goods as the market price of one good changes ?
21. Extend the analysis of the demand of an individual for loans (14.9
above) to the case where x lt x 2 # 3 ... x n are incomes obtained in n , ,
successive years, related by a transformation function F(x lf x 2 x 3 ... x n )^0. , ,
Assume that the individual has a utility function for incomes which can bo
written u$(x^ x 2 x3 , ... x n ) and that the rate of interest is I00r l por cent.
,
from the first to the second year, 100r 2 per cent, from the second to tho third
year, and so on.
CHAPTER XX
SOME PROBLEMS IN THE CALCULUS OP VARIATIONS
20.1 The general theory of functionals.
WE have been concerned so far with the theory of functions of a
finite number of variables, with the variations of quantities depend-
ing on one or more other quantities. There remains for brief con-
sideration an extension of the analysis, opening up many new fields
of application, which takes a quantity as dependent not upon a
finite set of other quantities but upon one or more variable functions.
In the simplest case, a variable quantity u is defined as taking
itsvalue from the form assumed by a function x ^>(t). To each
function <f> (t) there corresponds a definite value of u, and as the form
of the function is changed so is the value of u. The dependence of u
upon <f>(t) is called & functional and written u F{<f>}> The functional
symbol F is to be regarded in a manner analogous to the ordinary
symbol for a function. It is essential, however, that the functional
u = F{</)} should not be confused with the function of a function
u=F{(/)(t)}. The latter assumes that <f>(t) is a given function of t
and hence that u is also a definite function of t. The former takes </>
as a variable function, the functional F associating one value of u

with each whole function The variable t does not itself appear
<f>.
in the determination of u in the functional relation.*

In diagrammatic terms, the variable function x = <f>(t) is shown
by a variable curve C in the plane Oxt. As the form of the function
changes (</>l9 2 <^ 3 ...), the curve C shifts and takes up different
<f> , ,
positions and shapes (Clt <7 2 ^3> )

If ^ is a functional of
,
then <f>,
its value depends on the particular position taken by the curve (7,
* More u may be a
functional of several variable functions
generally,
x=<f>(t), y^t(t), write i* = F{fa <A X
z^xM We } Further, the
functions <, $, x, ... may be functions of several variables instead of one
variable t only. For tho general theory of functionals, see Volterra, Theory of
Functionals (English Ed. 1931).
and to the series of positions (Ol9 (7 2 , (7 3 , ...)there corresponds a
series of definite value of u (u^ u 2 u^, ...).
,
For example, Fig. 103
shows five positions of a curve limited to pass through two fixed
end-points A B
and and there will be five corresponding values of u.
An example of a functional can be taken from quite simple
economic theory. Each member of a competitive group of firms
producing a good X
with known cost functions fixes that output (x)
which makes marginal cost equal to the given market price (p) of
FIG. 103.
the good. Further, the output fixed by the firm varies when
different market prices are given. A
supply function, x=f (p)> can
thus be defined for each firm. But if the firm has monopoly control
and fixes output and price subject only to certain demand conditions,
then the output is such that marginal cost equals marginal revenue.
If the marginal revenue curve is given (given demand conditions
and given prices of all other goods), there corresponds one definite
monopoly output xm . As the marginal revenue curve shifts in position
(changing demand conditions or varying prices of other goods), the
monopoly output is determined afresh and varies in value. The
monopoly output xm depends on the form of the marginal revenue
function (f>(x) and we have the supply functional xm = F{<f>}*
The step from functions to functionals can be regarded as another
instance of the step from the finite to the infinite so characteristic
* See
Schneider, Theorie der Production (1934), p. 74.
SOME PROBLEMS IN THE CALCULUS OF VARIATIONS 523
of mathematical analysis. If u=f(xl} # 2 ... xn ) is a function of a ,
finite number (n) of variables,we can increase n indefinitely by the

introduction of more and more variables. The function u then tends
to involve an infinite number of variables. But this is just what is
implied by saying that u depends upon the form of a variable

function x = <f)(t). The function, in fact, comprises an infinite
number of particular values (xlt x 2 ... xn ...) and u depends, in
, , ,
the functional notion, on the whole of this infinite set of values.

The functional uF{(f>} implies that u depends on the infinity of
values making up the function x = <f>(t).
20.2 The calculus of variations.

The most important and frequent case of a functional u = F{<f>}
occurs when the form of F appears as an integral. Suppose that
f(t) is a function changing in form as (f>(t) changes. Write
where t and ^ are certain limits of integration. Then the value

of u depends on what particular form we take for <f> (t) and hence for
f(t), i.e. u is a functional of the variable function (f>.

The problem
that usually arises is to determine that function (f>
which makes u a
maximum or a minimum. The analysis of this problem is termed
the calculus^ of variations, one branch of the much wider theory of
functionals.* Many problems of importance in the varied applica-
tions of mathematical analysis are found to relate to the calculus
of variations. Two examples will illustrate this fact :
Ex. 1. The problem of the surface of revolution of minimum area.
A problem arising early in the development of the calculus of variations

is that of determining that curve of all curves joining two fixed points
A and B
on the same side of a given line L which forms a surface of the
smallest area when revolved about the line L. Axes Oxt are fixed in any
way so that Ot lies along L and so that the fixed points have co-ordinates
A (t Q , x ) and B (t ly xj which are all positive and t^> tQ . lfx = </> (t) is any
curve lying above Ot and joining A and J5, then it can be shown that the
analysis include an account

* Most text-books on advanced mathematical
of the calculus of variations. See, for example, Courant, Differential and
Integral Calculus, Vol. II (English Ed. 1936), Chapter VII; Osgood, Advanced
Calculus (1925), Chapter XVII ; de la Vall6e Poussin, Cours d'analyse infini-
tisimale, Vol. II (1925), Chapter X.
area of the surface formed by revolving the curve about Ot is propor-
tional to
Here u is a functional of the variable function and we seek that function

<f>
(subject to the limitations indicated) which makes u a minimum.

Ex. 2. The brachistochrone problem of John Bernouilli.
One of the first problems in the calculus of variations, posed by John
Bernouilli in 1696, is to fix a curve joining two given points and A so
that the time taken by a particle to
. slide under gravity along the curve
^ from to A is least. Axes Oxt are
chosen, as shown in Fig. 104, with
origin at the given point and with
Ox drawn vertically downwards. Let
A have co-ordinates (tlt xj which
are both positive. If x = <f>(t) is the
equation of any curve joining to
A so that<(0)^0 and <(i)=#i, then
FIG. 104. it can be shown that the time taken
by a particle to slide under gravity

from to A is proportional to
u-
Here, again, u is a functional dependent on the form assumed by the
variable function <f> and we seek that form of which corresponds to the
<
minimum value of u.
20.3 The method of the calculus of variations.

Two important preliminary points must be considered in devising
practicalmethods of solving problems in the calculus of variations-
The first point concerns what can be called boundary conditions.
The problem to be solved is usually framed so that only certain
arcs of the variable curve x = <f)(t) in the plane Oxt are needed,
definite conditions being imposed upon the points which mark the
ends of the arcs. In the most frequent case, the conditions are that
the arcs should start and finish at two fixed points A and B in the
plane. The boundary conditions, in such a case,' impose limitations
on the field of possible variation of the curve x = (t) and we have <f>
to consider, in fact, only those curves which can be drawn from

A to B. This is so in the two problems instanced above and it is
illustrated Fig. 103. In analytical terms, the function

by x=<
can only be selected provided that
<f>(to)=x g and
where A(t X Q ) and B(t^, x) are the two fixed end-points. Other
,
boundary conditions may be imposed in less usual cases. It may be

given, for example, that the variable curve x = <f>(t) should join two
points P and Q which lie one on each of two given curves in the
plane Oxt. Here, we have the boundary conditions <f>(t Q )=XQ where
(to, #o) are ^he co-ordinates of P and satisfy some given relation
*ti(t
x )=
,
and similarly for the co-ordinates (^, xj of Q.
;
Even when the boundary conditions are taken into account, the
field of possible variation of the function x <f>(t) is so large
that
analysis of the problem is practically impossible and it becomes
essential to limit the field of variation by some further device. An
obvious step to take only those functions, satisfying the boundary
is
conditions, which are continuous and possess continuous derivatives

up to any desired order. Further, the field can be more severely
limited particular type or curves of a
by taking functions only of a
particular class. take, for We may
example, only functions of the
quadratic type represented by parabolas with vertical axis. We
know that a function type or curve class can be represented by a
relation involving certain parameters a, /?, y, . . .
;
the larger the
number of the parameters the more general is the function type or
curve class. The result of this limitation on the field of the variable
function is to replace the function x <f>(t) of variable form by
x = <f>(t; a, y, ...), ,
where <f>
is now of fixed form and the variation of the function is
replaced variation of the parameters involved.

by the
It important to appreciate the nature of the step now taken.
is
We have given up the consideration of any comprehensive variation

in the function x~<f>(t) and have limited ourselves to a more re-
stricted variation described by parameters in a function of fixed
form. If the parameters are few in number, the restriction is very
severe. For example, using only three parameters, we may write
so that our variable curve limited to the class of parabolas with

is
their axes vertical (parallel to Ox). But, by taking more and more
parameters, we can make the field of variation of our function type
more and more general. If a sufficiently large (but finite) number
of parameters is selected, the restricted field of variation can be
made to few important respects from the complete field we
differ in ;
simply exclude from the latter the more unusual kinds of functions.
The problem to be solved is now greatly simplified. We seek the
extreme values of u = I
f(t) dt where f(t) depends on the variable
J /o
function x <f>(t). Limiting the latter in the way indicated, the

expression to be integrated becomes f(t a, /?, y, ...) where a, /?, y, ...
;
are parameters and where the form of the function / is fixed. It

follows that the value of the integral u depends only on the para-
meters a, /?, y, ... and we require its maximum or minimum value
for all variations in the parameters allowed by the boundary con-
ditions. The problem of the calculus of variations is thus reduced
to a problem of extreme values of an ordinary function of several
variables a,
/?, y, ... Functional are changed back into functions
.
and we proceed on familiar lines.* The extent to which the simplified

problem approximates to the original one depends on the number of
parameters we care to take. The most important thing about the
analysis which follows is that it is quite independent of how many
parameters there are, provided only that their number is finite.
The solution we obtain is not perfectly general but it provides an
approximation sufficient for all practical purposes.
20.4 Solution of the simplest problem.

The function f(t), which gives tlie variable u on integration,
depends in some definite way upon the variable function x = <f>(t).
The dependence usually includes, not only itself, but also the <f>
various derivatives of <f>.

Of the many possible cases, only the
simplest will be analysed here, the case where f(t) depends on the
variable t, on the function <f>(t) and on the first derivative <f>'(t) :
f (t,
*>
)
=/{, W), f (*)}.
step from functions to functionals is reversed. A functional can be

* The
regarded as a function of an infinity of variables. In solving problems of

functionals, we approximate by taking a function of a large number of variables
(the parameters 0, y, ...). This is in line with the methods always adopted
,
for dealing with the infinite and the step made here is, after all, not unusual.
This is a function of t
given in the function of functions form. But,
since the function <f>
included is of variable form, the form of the
function/ to be integrated is also variable. We assume, further, that
the boundary conditions are such that the variable curve x = <l>(t)
passes through two fixed points A (Z ,
x )
and B(tly xj. The problem
to be solved is thus :
/ dx\
f (
t, x, 1 dt are required
for all possible variations in the function x~</>(t), such that

(/>(t Q )x and <t>(ti)=xl where (t Q X Q ) and (tl9
x: ) are fixed points.
The problems instanced in 20.2 above are both of this form. In the
first problem of the surface of revolution, we have
and in the problem of John Bernoulli!
In solving the problem, we impose the limitations on the varia-

tion of <f> (t) already described, taking this function in the form
x = t(t; a,&y, ...)
where <f>
is a fixed function (with a continuous derivative) and
where a,are parameters.
/?, y, Allotting arbitrary differential
...
increments Sa, S/?, 8y, ... to the parameters, we derive first the
corresponding variations 8x and Sx' in the function x and its
, . ,. ,
dx
derivative x -=-:
at
dx dx dx
80;=: 8a + ^5 8^-fr-
r
9a 9j3 9y
, fdx\ d (dx\ - d fdx\ ^ d (dx
and &P =81-^
\dtJ
)= T (^ JSa + ^l-y-
dot\dt/
-
r
8j8 + f
dp\dt/ dy\
d ( 3x \* ^ d ( dx \*(*^ d ( dx \
= 1; l^)8a ^ + + r
1^5J8j8 ^- I 1
dt \docJ dt \df}/ dt \dyl
d dx dx dx ,
\ d
It should be remarked that all the variations here are ordinary

differentials and subject to the usual rules of differentiation. They
" "
are denoted by the symbol 8 in order to distinguish them from
the other differentials, dx and dt, we obtain when we consider x as a
function of the values of the parameters being given. Hence,
t,
" "
throughout our analysis here, d refers to variation in the variable
" "
t and 8 to variation in the parameters a, j8, y, ... .
The function f(t, x, x') and the integral u can now be considered
as dependent on the parameters a, /?, y ...and the variations in
their values are obtained as
\J1
J _ \J*J I ~ . v^ ~
dx ox ox
and
Now, using the result of 15.3 above,
Hence, except for the addition of an arbitrary constant,
The expression for the variation in u then becomes
Since,by the boundary conditions, the curve x <f>(t) always passes

=
=
through two fixed points at t t Q and t~t^ it follows that the varia-
tion of x is zero, i.e. 8^ = 0, at these points. Hence,
The integral u is to have an extreme value for variation in the

function x = <j>(t) as obtained by varying the parameters a, j3, y, ... .
The necessary condition for this is that 8^ = for all values of

8a, 8/J, Sy, ... ,

i.e. for all values of Sx. From the above expression
for 8u, this is only true if
f
dx dt \dx
This result, known as Euler's equation, determines that function
x == <f> (t) which maximises or minimises the value of u . Since -^-

and
df
-, are partial derivatives of f(t, x, x') and so functions of t, x and
ox ~
dx
x'
j- ,
the equation is a relation in x as a function of t which involves
fit
the derivative of x with respect to t, i.e. it is a differential equation

which must be solved to give the function x = <f>(t) we seek. Our
problem in the calculus of variations is thus reduced to the relatively
simple problem of integrating a differential equation. The methods
for completing the solution are set out in Chapter XVI above.
Enter equation, however, is
\s
only a necessary condition for extreme
values of u. There remains the problem of distinguishing between the
different extreme values. The number of variables (i.e. the para-
meters is here very large and, as we have seen in the
a, /?, y, ...)
previous chapter, a general criterion for separating maximum from

minimum values is not readily obtainable. We must content our-
selves with the fact that, in simple practical cases, we can usually
tell from general reasoning whether we have a maximum or a
minimum value of u.
20.5 Special cases of Euler's equation.

Two
particularly simple cases of the problem of the previous
section often arise in practice :
/ dx \
fit, j7) dt, Tihere the function
ri
to be integrated does not involve x explicitly.
=
^=
1'
Writing x and noting that 0, Euler's equation becomes

^-
giving 9
x ') =a"
fa'M'
where a is some constant. This is a differential equation involving
2L M.A.
only the first derivative x', i.e. a simple differential equation of the
first order which can be solved, in m'ost cases, without difficulty.
(2) To find extreme valuesjof u = \

f f x, -7- dt, where the fiction
J
to be integrated does not contain /

explicitly.
dx
Writing x' ==-=- , differentiating f(x 9 x') as a function of two vari-
d
ables each depending on t,
df
and using Euler's equation = ~ -
ox c
\
= .
X _
+ ^d^_d_l
~ ,df\
dt dx dt dx' dt dt \dx''J dx' dt dt V dx'J
r)f
Hence, / and x' -, , having equal derivatives with respect to t, can
ox
only differ by an arbitrary constant, i.e.
f(x, *')=*' (*,*')+

-jjf
where a is some constant. Again we have a differential equation of
the first order, involving only x and its derivative x'. The solution,
giving the function x = (t) we require, proceeds with little difficulty.
<f>
It is to be noticed that the two problems of 20.2 above are both of

this special form.
20.6 Examples of solution by Euler's equation.

The methods of solution developed here can be illustrated by the
following three examples :
Ex. 1. To find extreme values of u = \ t(-j-) dt subject to the

J< \u*J
boundary conditions that the curve x = <f>(t) passes through two fixed
points A(t Qy x )
and B(tl9 xt ).
We have the special case (1) of 20.5 with
2
/(*, z')=to' and -?-,f(t,x')=2tx'.
Hence, 2tx' = constant, i.e. =-

-^-
where a is some constant. The integral is obtained at once as
J
- d + constant, i.e. #=a
where 6 is a second constant. The extreme value of u is thus given by a
logarithmic curve.
The appropriate values of the constants a and b are to be found by using
the boundary conditions. Since the curve must pass through the fixed
points A and J5, we have
XQ =a log t
Q +b and xl =a log t + b.
On solving these linear equations, the constants are given as
_r - log t
and
log ^ - log t
Q
.
In this way we fix that particular logarithmic curve which passes through
A and B. The use of the boundary conditions in determining the relevant
values of the constants of the solution is to be noticed.
Finally, the extreme value of u determined is
U=\ ti-r-
It is not possible to indicate here whether this is a maximum or a minimum

value.
2
Ex. 2. To find the minimum M=
value of u == 1
I x V 1 4- (
<.c,^\
-
7
7 )
dt where
Jt*
the curve x~</>(t) is subject to the usual boundary conditions.

We have here the special case (2) of 20.5 with
a xx'
2
f(x,x')=xjl+x' and x')
gp/(*, ^^YTx7*
and Euler's equation is
_ ~~"2
jijj
On multiplying up, squaring and collecting terms, we find
a;
2
-a
~
2
dx
x ,, =
.
a2 dt a
dx ( dt b
TT
Hence,
.
I .
= 1 h constant
,
=t
J<s/a
2
-a 2 J a a
where both a and 6 are constants. It is easily verified that
and the solution of our differential equation becomes
log = i.e. x + v x2 a 2 = ae .
a a
2L2 M.A.
Now,
t-b t~b\
Adding, x = %a{e a + e a )
which is the function giving the minimum value of u. The boundary con-
ditions provide the values of a and 6 in terms of ( x ) and (tlt xj ,
:
and +e
The minimised value of u is then given by
u= A/
I7~7fof
1 +(-=-) M
eft
,
where -=- =
v \dt/ dt
i.e. w= dt = - a\ \e a +e
j
This integral can be evaluated fairly easily.

/ t-b <-fo\
The curves corresponding to the function type x = \a\e a +e a ) are
"
called catenaries ". The curve which gives the minimum value of u
is thus that catenary which passes
through the two fixed points A and B
given by the boundary conditions.
The shape of this curve is shown in
Fig. 105. The solution of the problem
of 20.2, Ex. 1, is now obtained ;
the
catenary is the curve which gives
a surface of revolution of smallest
area. It is to be noticed that the
nature of this geometrical interpreta-
Fio. 105. tion of the problem shows that the
value of u we have found is a
minimum and not a maximum.
=
Ex. 3. To find the minimum value of u
y- 1
^
1 + ( -=-} \ dt subject
to the usual boundary conditions.
This is again the special case (2) of 20.5 with
*"
t? and /<*,*>
and Euler's equation is
v?
We find, on multiplying up, squaring and collecting terms,
I-a*x . dx _1 ll-a?x
*/ i.e.
i.vy.
a?x
,
at a* x
The integral of this differential equation is
X dt t-b
*
(
J
VVI i -a*5"x 1x = J( a + constant =
i I
4. 4.
where both a and b are arbitrary constants. It is now only a matter of

evaluating the integral shown. This can be done only with the aid of
" "
trigonometric functions and we cannot proceed further here.* It can
be stated, however, that the relation between x and t obtained on com-
pleting the integration is represented by a curve of a well-known class,
"
the class of cycloids ". The solution of our problem, taking account as
before of the boundary conditions, is given by that cycloid which passes
through the two fixed points and A. It is this curve which gives the
minimum time of descent from to A (see the brachistochrone problem
of 20.2 above).
20.7 A dynamic problem of monopoly.

A monopolist sells a good X
on a market consisting of a com-
petitive group of consumers. If he produces an output x per unit of
time, the cost of production is given by H (x) and remains unchanged
throughout the period from t~t to t 1 which we consider. It is
assumed that the market price of X varies continuously over the
period and is represented by some function p (t). The demand of the
"
competitive group of consumers is assumed to involve a specu-
"
lative element and to be described by the demand law given in
16.7 above :
* = #{*>(), P'(t)}.
The market demand thus varies over the given period according to
the course taken by the price of X
The monopolist's profit per unit
.
of time when he produces an output x per unit of time and sells it

at the price p (t) is given by
This is the rate of profit at any time t and depends on the price and
* The solution of the problem can be written :
t = b + a'(e-sm 0), x = a'(l

-cos 8),
where a' is an arbitrary constant obtained from a. The required relation
between x and t is obtained by eliminating the parameter 0. The expressions
sin 8 and cos 8 are trigonometric ratios of the angle 8 radians.
the rate of change of the price at this time. The total profit in the
given period from t=t Q to t=^ is obtained by the addition of the

profits in successive units of time. Since the variation of price and
output is assumed to be continuous, the addition can be represented
by an integral, i.e. total profit is
For a given course of price over time, the demand x of the market is
determined at each moment and there is a definite value of u corre-
sponding. Different values of u are obtained for different courses of
price. Hence, the monopolist's total profit depends on the form
of the price function over time, on the course taken by price. It is
assumed, that the ruling price at the initial moment t t Q is
finally,
given as p and that the subsequent course of price is fixed by the
monopolist so as to maximise his total profit in the whole period
from t = t to t = t^. It is required to determine the course of price,
i.e. p (t), actually fixed by the mono-
the form of the price function
polist. The corresponding output at successive moments is then
derived from the given demand law.*
As a first problem, it is taken that the choice of the monopolist
is restricted by the fact that the final price (p at tt^ is given in
addition to the initial price. (This artificial assumption will be

dropped when it has served its purpose.) We have then a problem
in the calculus of variations to find the function p (t) which
:
maximises the value of u = - n(x)} dt, where x = 1

I
{xp <f>(p, p )
and
where the boundary conditions give p=p when t=t Q and p=p t
when t^t^. This is the special case (2) of 20.5 with
f(P,P')=xp-n(x) where x = <f>(p,p')

as the function to be integrated to give u.
* The
problem is based on the work of Roos and Evans, see Evans, Mathe-
matical Introduction to Economics (1930), pp. 143 et seq. Notice that, if prices
do not change continuously but are fixed at definite intervals, then the mono-
polist's total profit is a function of a finite number of prices (one for each of
the intervals in which prices are fixed). The problem of maximising total
profit is then relatively simple, a problem of the maximum value of a function
of several variables. It appears that here, as in many problems of capital
theory, the discontinuous form is both more realistic and more simple from
the mathematical point of view.
Hence,
dx dn dx f dIT\ dx
and Euler's equation gives

/ dn\ dx
p - -j- ^-7 + constant.
, x , A ,
xp-II(x)p I ]
This is a first-order differential equation in the function p(t) 9
3.-1D
remembering that x = </>(p, p') is a given function of p and p' = -f-.

dt
The solution of the equation gives the function p (t) required except
that it involves two arbitrary constants (the constant above and a
constant of integration). The appropriate values of the constants
are givenby the two boundary conditions fixing the initial and final
prices. The problem is thus theoretically determinate and the
following particular case shows how an actual solution can be
obtained when we know the forms of the demand and cost functions.
Suppose that the demand function is linear, x=ap + b+cp', and
the cost function quadratic, 77=a# 2 +/fo;4-y, where the six co-
efficients are all constants. The equation for p (t) is now
p(ap+b+cp')-IJ=cp' [p --T-) + constant,

\ (t x I
dlldp __
ap 2 + op - /Inconstant.
=- + 9 , , .
i.e. . c-r:
dx dt
Differentiating with respect to t and collecting terms,
2
d*nd
-
2
p d*IIdp (
d77 V A
z -
-
C ~z r-r- ~r Q'C 7- T" I ttp -{- (t I U.
2 2
dx dt dx 2 dt \ dx)
But, -
r = 2a^ + = 2ca-~4-2aajp
-
j8
dt
+ (26a-f-j8) and
dx
The differential equation for p (t) then reduces to
2c 2 a -^ - 2a (aa - l)p 4- (6 - 2a6a - a)8) =0,
,
where
~
p=
r
6-2a6a~aj8
-r rr-
2a(aa-l)
>
- ^^d
,
A2 =
N9 a(aa-l)-
-
C 2a
5
-
2
(A is positive if the constants have the signs appropriate to the
normal case, i.e. a < and a> 0). The solution of this simple second-
order differential equation is known to be
where A and B are constants of integration. The solution can be

checked by differentiation. The constants p and A are given in
terms of the coefficients of the demand and cost laws. The constants
A and B and final prices by
are to be found in terms of the initial
substituting p =p Q when t = t Q and p ~p1 when t = t^.

The problem is
solved and the course of prices over time, as determined by the
monopolist, is
given uniquely.
Returning to the general case, given the final pricepl9 we have
shown that the price function p (t) and the maximised value of total
profits u can be obtained. The total profits (as maximised) must
be a function of p ly say u~F(p1 ). To complete the problem, the
monopolist has only to fix that final price (of all possible final prices)
which maximises u as a function of pl9 i.e. which gives the largest
total profits of all. For this, we put
du ,
-i = .
subject to
, t
ri -
<
dpi dpi*
and determine p from the resulting equation.
20.8 Other problems in the calculus of variations.

The simple problem in the calculus of variations treated above is
capable of generalisation in a number of directions.* For example,

the function in the integral u may involve second and higher order
derivatives of the variable function <, or the variable function may
depend, not on one variable t, but on several variables Zl5 Z 2 3 ... , ,
.
Further, the extreme value of u may be required subject to a number

of given side relations in the variable function, the problem of
relativemaximum and minimum values. But perhaps the most
useful extension of the problem arises when the integral u depends
on several variable functions instead of on only one such function.
This extension can be illustrated by the case where there are two
variable functions, x = <f>(t) and y = if/(t) and where the integral to
)
* See
Courant, op. cit., pp. 507-20.
be maximised or minimised is of the simple form (analogous to that

of 20.4 above) :
subject to the boundary conditions that x and y both assume fixed

values when t = t Q and i t^.
It can be (following the argument of 20.4) that, when the
shown
parameters of the variable functions are changed so that 8x and Sy
are the resulting increments in the functions themselves, then the
variation of u is
where x'~ ~,~ and y'= -~ . The necessary condition for an extreme
Cit Cut
value of u is that 8u = for all variations Sx and 8y. Hence,
and -
dx~dt\dx'l dy~~dt\dy'
i.e. there are two differential equations (each similar to Euler's
equation above) to be solved for the functions x = <f>(t) and y = tft(t).

An economic example of the present problem can be taken from
a theory of saving developed by Ramsey.* There are two variable
factors of production in a community, i.e. labour B and capital C 9
each of which is taken as homogeneous in nature. If amounts b and c
of the factors are used, the total product (or income) of the com-
munity is given by the production function product =/(&, c). The
:
variations of labour and capital over time are assumed to be con-

tinuous and represented by the functions b(t) and c(t). The rate of
dc
growth of capital, or the saving of the community, is -=-
ut
at any time.
The consumption (or expenditure) of the community is then
(1)
It is assumed that the utility of the community's consumption is

measurable and dependent only on the amount of the consumption,
u </>(a). Similarly, the disutility (assumed measurable) of labour
* See Ramsey, A Mathematical Theory of Saving, Economic Journal, 1928.
A modified version of this theory is given here.
is taken as v = if/(b). Over a period of time (t ^), the total ,
net
utility of the community is
U=\ #()-
where future utilities are not discounted. This quantity U is a
functional of the variable functions b(t) and c(t), representing the
changes in the employment of labour and capital over time. The
function a(t) is expressed in terms of b(t) and c(t) by (1) above. It
is assumed that the amount of labour and capital employed at the
" "
initial time (t 1 ) are
given and that the optimum amounts to
be aimed at (t=t^ are also known. Subject to these boundary con-
community is taken as fixing the amount of work it does
ditions, the
and the amount of saving it makes over time to maximise total
utility. We have, therefore, to find the functions b(t) and c(t)
which maximise the value of U, subject to the usual boundary
conditions.
Write c' ~ dc and

at
where F (b, c, c') <f> (a)

- (/>) and a ~f(b, c)
- c' .
The conditions for the maximum of U are

dF d fdF\ dF d (dF\
= I
r; 10^ and
.
= (
,
)
The first condition gives
i e-
1>e
The second condition gives
The functions a(t), b(t) and c(t) are to be found from the differential
equations (1), (2) and (3), the boundary conditions being used to
evaluate the arbitrary constants that appear on integration. The

solution of the problem is determinate in general.
The partial derivatives ^ and ~ of the production function
measure the marginal products of labour and capital at any time and
can thus be associated with wages and the rate of interest respec-
tively. The
rate of wages, by (2), equals the ratio of the marginal
disutility of labour to the marginal utility of consumption. The rate

of interest, by (3), is equal to the proportional rate of decrease of the
marginal utility of consumption over time. Assuming that the rate
of interest never becomes negative, the marginal utility of con-
sumption falls over time either until it vanishes or until the rate of
interest becomes zero. Further, if (f>'(a) is a decreasing function of a
(decreasing marginal utility), then consumption rises over time until
its marginal utility or the rate of interest is zero.
An expression for the rate of saving can be found as follows.

Using the results (1), (2) and (3), we have
=#" <)
,, .
rf7
W ,. v
+ *'(
,,, .
~ y c)
- a}
- ~ - a}
jt
#"(a) {f(b, c)
= -- + f tf>'(b)
So <f>'(a)/(b, c) fa {<f>'(a)} dt
j dt + constant
f
â^'(rt)
-
(V(a)
~ dt + U'(6) -j dt + constant
- (a)
- (b)} + constant,
a<f>'(a) {<f> <fi
i.e. <f>' (a) {f(b, c)-a}=

constant - {<f> (a) -tfi(b)}.
* Since
v {a<^'(o)}
= a-T {^'(a)} + <f>'(a) -3- , we have
a</>'(a) a-?- {<f>'(a)}dt + \</>'(a)-- dt,
d da
{&' (o,)} dt ~
f ( ,
i.o. I ct -r- (i <ft
(d) I
d>'(fi} r (It*
j at j dt
Hence, using (1), we find
^C _^ ITV W / Y\"l) /A\
where A is a constant.
Since U is to be a maximum (as opposed to a minimum), the
excess of utility over disutility, cf>(a) - ^(6), must increase over time.
Further, it can be taken that capital increases at a decreasing rate
" "
until it becomes stationary at the optimum at t =^. So, from (4),
the excess of utility over disutility increases and tends to its maxi-
mum value A at t = t^. Then, from (4), the rate of saving multiplied
by the marginal utility of consumption at any time equals the

amount by which the excess of utility over disutility falls short of
the optimum amount A. As time goes on, consumption increases,
the marginal utility of consumption decreases and the rate of saving
decreases until it becomes zero.
EXAMPLES XX
Problems in the calculus of variations
1. Find the curve x </>(t) which corresponds to an extreme value of
fdx\ 2
* VTr) dt and passes through two fixed points at t = t and t t^
2
(tj
.t \dt/ r
tl /dx\ 2
Generalise by considering u= 1 t
n
dt, where n is a fixed number, on
j' ^ dt/j
{
similar lines.
fdx\ 2
x* ( ~r dt nas an extreme value when the function
(h to \dt/
)
is of the form x = a*Jt - b. How do the boundary conditions that the

curve passes through fixed points at t t and t t determine the constants
a and b ?
3. Show that an exponential curve gives the maximum or minimum
values of u= I
(
-
) dt, subject to the conditions that the curve passes
jtoX 2 \dts
through fixed points at t 1 and t = tlf
4. If
u=\ \ x /
+ dt, whore fixed values of x correspond to
\l \~:r) \
t~t and t t
19 find the function x=</>(t)
which gives u its extreme values.
Show that the corresponding curve is an arc of a parabola.
5. Light, travelling in a plane Oxy, follows a path y (x) which is deter- <f>
mined (by Format's principle) so that the time of transit is a minimum. Thia
Cxi i / fdy\*
equivalent to finding the function which minimises u = - + -
is <f>
| ^ 1 (
^ ) dx,
where v a function of x and y denoting the velocity of light at various

is
points. Assuming that v is a constant, show that the path of light between
two fixed points is a straight lino.
SOME PROBLEMS IN THE CALCULUS OP VARIATIONS 541
6. The
velocity of light, in the problem of the previous example, increases
in proportion to the height above the zero horizontal line Ox. Patv~ky
and show that the path of light between two fixed points is an arc of a circle.
2
f(x)\l
/
1 + (/dx\
-y } dt, where x has
(*i
.2o\dt/ V
fixed values at t tQ and t = t 19 show that an extreme value of u is given by the
function x = <f> (t) defined by
f dx
where a and 6 are constants to be determined by the boundary conditions.

Show that the previous three examples are particular cases of this general
result.
8. A radio manufacturer (see Examples V, 15, and Examples VIII, 33)

produces x sots per weok at a total cost of 77, where 77 = -^-a: 4- 3#+ 100.
2
-
The demand of the market is x 75 3p -f 275 sets per week at any time ~
when the price is p per sot. The manufacturer is a monopolist fixing the
course of price over time so that his total profit is a maximum. Tho initial
price ( = 0) is 16 per set and it is required to obtain a price of 20 per set
after 100 weeks (t~ 100). Show that, at any time t weeks after the initial
week, the price per set is
p= 15-179 + 0-171e- 333 * + 0-650e- *. m
Take all figures correct to throe decimal places and Iog 10 e 0-4343.
9. In the problem of the previous example, show that the price falls to a
minimum after about 20 weeks. What is this minimum? Draw a graph to
illustrate the course of price over time.
10, In the solution of the problem of 20.7, show that
~ (Pi-P)e**
A _ (Pi ~ P) e ~** -(Po-P) e ~* tl
pf__(Po-p)e*
tl
. eM*i-*o)-e- A <i-o)
'
e*<*i-W - e-*<i-fc>
11. Under the conditions of the previous example, show that the price p
falls inthe period (f tj from p Q to a minimum and then rises to p^ Verify
,
that the minimum price occurs when
12. In the problem in the theory of saving analysed in 20.8, the production
function /(&, c) is linear and homogeneous. Show that
and interpret this result. Show, also, that
ldc_df a<l>'(a)- "
c dt~^c c<f>'(a)
What is the rate of interest at the optimum time (i
= f when capital ceases
1)
to accumulate?
INDEX
[Numbers refer to pages; numbers in italics refer to examples appended
to chapters.]
MATHEMATICAL METHODS
Abscissa, 21 Co-factors of determinants, 479-80
Acceleration, 149, 151, 185, 303, 310 Co-ordinates, oblique, 26
Analytical functions, 38 polar, 27
geometry, 49, 61-82, 272 rectangular, 20-2
Approximate integration, 396-9 Cubic equations, 31, 51-4
Approximate values from derivatives, Curve classes and systems, 48-9, 61,
142-3, 146-7 76-80, 414, 423-30
from infinite series, 447, 450, 454-6 Curves and functions, 36-8, 48-50
from limits, 95-7 Cycloid, 533
Arithmetic series or progression, 449
Average and marginal concepts, 15, Definite integrals, 384-90
152, 190-1, 254, 400-1 quadratic forms, 491
Averages, 62, 406-8 Dependent and independent vari-
Axes, co-ordinate, 20-1 ables, 29, 269, 275, 332, 364-9,
461-9
Base, of logarithms, 213, 216 Derivatives and partial derivatives :
of power and exponential functions, and approximate values, 142-3,

161, 211-2 146-7
Binomial coefficients, 173, 457, 464 and extreme and inflexional values,
expansion, 454-5 180-95, 351-9, 364-9, 459-61,
Boundary conditions (calculus of 497-500
variations), 524-5 as rates of change, 142, 179, 246,
Brachistochrono problem, 524, 533 251, 303-5, 310, 326
as ratios of differentials, 329-30,
Calculus of variations, 523-33, 536-7 465
Catenary, 532 as tangent gradients, 143-8, 179,
Circle, 34, 37, 75-6, 79 246, 252, 297-8, 303-5, 308, 338
tangent to, 339 definition of, 137-42, 296-7, 309
Concavity and convexity of curves infinite values, 147-8
and surfaces, 184-6, 191-2, 303-5 in natural and social sciences, 149-
Cone, 292, 315 57
Continuity of functions and curves, of functions of functions, 168-71,
98-103, 147, 206, 269 298-9, 333-4, 466, 469, 471
of numbers and variables, 6-9 of implicit and multi-valued func-
of quantities, 12-3 tions, 206, 335-40, 466-8
of space, 16-7 of power, exponential and log-
Contours of surfaces, 273-4, 277-8, arithmic functions, 162-3, 177,
316, 352, 353-4, 357, 358, 368 242-6
Convergence and divergence of series, rules and evaluation, 163-75, 244-8,
447 298-300
544 INDEX
Derivatives and partial derivatives :
Exponential series, 465
second and higher orders, 148-9, Extreme values, 181, 194, 354, 459,
172-5, 246, 300-2, 309-10, 337-8, 495
466-9
sign and magnitude, 179, 184-6, Factorial notation, 451
193, 303-5, 310, 459-61 Format's principle, 540
standard forms, 160, 163, 244 Flats, 17, 276
Derived function, 139, 146 Frequency distributions, 406-8
Determinants, 472-80 Functional, 620-3
and linear equations, 482-5 Function of a function, 168-71, 298-9,
and linear homogeneous functions, 330, 332-4, 466, 469, 471
481-2 Functions, 28-50, 268-78
and quadratic forms, 485-92 approximate expression, 143, 328,
elements 473
of, 450, 454-6
expansion rule, 480 defined by integrals, 397
minors and co -factors of, 478-80 diagrammatic representation, 32-8,
Differentials and differentiation, 328- 48-50, 270-5
40, 461-9 limits and continuity of, 88-103
and extreme values, 495-502 Function types, 41-5, 76-80, 272
Differential coefficient, 330
Differential equations, 390-1, 393,
Geometric series or progression, 229,
412-34
447-8, 449
and arbitrary constants, 415-6, 431
exact form, 420, 422 Geometry and analysis, 1, 16-23,
in the calculus of variations, 529- 36-8, 48-9, 270-2, 275-6
Gradients of curves and tangents,
33, 537
144-5, 179, 246, 252, 338, 413-4,
linear, 417-25
423, 427
non-linear, 430-3
of straight lines, 63-6
simultaneous, 425-30
of surfaces arid tangent planes,
variables separate form, 418-9
297-8, 303, 305, 308, 340, 416
Discontinuity of variables and func- 221-8
Graphs, 19, 32-6, 52-6,
tions, 9, 31-2, 34, 100-3
infinite, 101, 148, 188
jump, 102 Homogeneous functions, 315-9, 324,
Discriminant of quadratic forms, 481-2
487-8, 491 Hyper-surfaces, 17, 276
Elasticities of functions, 251-4, 300, Identities, derivation of, 318, 348

305, 421 Implicit functions, 29, 268, 275, 334-
Ellipsoid, 380, 428 9, 466-8
Equations, graphical solution, 52-6 Incrementary ratio, 137-9
quadratic, cubic and polynomial, Increments in variables and func-
5-6, 7, 30, 31, 51-4 tions, 137, 142, 146-7, 326-8,
simultaneous, 54-6, 60, 483 464-5
systems of, 278-81, 483-5 Indefinite integrals, 384, 391
Equations of curves and surfaces, 37, Infinite derivatives, 147-8
66-76, 272 discontinuities, 101, 148, 188
Errors in variables and functions, limits, 85-98
142-3, 328 properties of numbers and space,
Euler's equation, 529-33, 537 6-7, 16-7
Theorem, 317-9, 348, 349, 434, 481 series, 446-9
Expansions of functions, 449-69 Inflexional values, 181, 182, 191-5,
Explicit functions, 29, 268, 276 460-1
Exponential functions, 211-3, 217, Integrals and integration, 384-99
218-9, 228, 234-7, 242-6 and areas, 384, 387-90
INDEX 545
Integrals and integration as inverse Orthogonal curves and surfaces, 429-

differentiation, 384, 390-3 30
in the calculus of variations, 523
of differential equations, 390-1, Parabola, 33, 69-72, 76-8
393, 412-34 tangents to, 145
standard forms, 394 Paraboloid, 272, 273, 308, 422
Interval, 8 Parametric constants, 43-5, 46-8, 76-
Inverse functions, 29, 171-2 80, 525-6
Irrational numbers, 4-5, 12, 86 Partial derivatives (see derivatives)
derived functions, 298
Jacobians, 474, 475, 490 differential equations, 433-4
elasticities of functions, 300, 305,
Lagrange's multipliers, 366, 379 421
Leibnitz's Theorem, 173 stationary values, 351-2
Light, path of, 540-1 Pearson's curve system, 240
Limits, of functions, 88-98, 105, 269 Pencil of lines, 67
of sequences, 85-8, 446 Plane, 272, 292, 305-7
Linear differential equations, 417-25 Planes, co-ordinate, 22
equations, 54, 60, 483-5, 493 Plane sections of surfaces, 272-6
functions, 42, 67-8, 76-7, 228, 269 Power function, 161-3, 211, 218-9,
homogeneous functions, 315-6, 317- 228
9, 481-2 series, 450
Logarithmic derivation and dif-
ferentiation, 246-8, 300, 331 Quadrants of a plane, 18
functions, 217-9, 228, 234, 237, 242-6 Quadratic equations, 5-6, 7, 30, 51
graphs, 222, 225-6, 228, 252 forms, 485-92
scales, 219-28 functions, 42-3, 71-2, 76-7, 269-70
series,456 homogeneous functions, 316
Logarithms, 213-7, 237 Qualities, 10
Logic and mathematics, 1-2 Quantities, 10-6
Logistic curve, 26 5, 419
Radius vector, 190, 191
Magnitudes, 10, 13-4, 276-8 Rates of change, 48, 134-7, 142, 149-
Maximum and 'minimum values, 180- 50, 179, 246, 251, 303, 305, 310,
91, 352-9, 364-9, 459-61,
205, 326
495-502, 523-4, 526-33, 536-7 Rational numbers, 4, 11-2, 86
Mean Value Theorem, 452 Real numbers, 6
Minors of determinants, 478-80 Rectangular hyperbola, 33, 72-5, 76-
Monotonic functions, 40, 269 9,103
Multi-valued functions, 38-41, 102-3, Relative maximum and minimum
206, 269, 334-9, 466-8 values, 364-9, 498-500, 501-2
Ridge and trough lines, 288, 351-2,
Natural exponential and logarithmic 357-9
functions, 234-7 Roots of an equation, 51
scales, 219, 227-8 Ruled surfaces, 315
Necessary and sufficient conditions,
183, 186, 193, 204-5, 355, 365-6, Saddle points, 353, 496
461, 497, 499 Scatter diagram, 26, 36, 222-3, 226
Neighbourhood, 8 Semi -logarithmic graphs, 221, 223-5,
Normal curve of error, 240 228, 237, 246
Number e, 230-1, 465-6 Sequences of numbers, 85-8, 446
Numbers, 3-7, 85-7 Shortest distance of point from line
and plane, 206, 380, 500-1
Ordinate, 21 Side relations, 364, 498
Origin of co-ordinates, 20-1 Sigma notation, 385
546 INDEX
Simpson's rule for integration, 399 Tangent planes to surfaces, 307-9,
Single-valued functions, 38-41, 269 340, 416, 423
Smoothness of curves, 100, 147 Tangents to curves, 143-8, 338-9, 413-
Space, location of points in, 16-23 4, 423, 427
Speed, 15, 150-1, 152 Taylor's series, 179, 453, 458
Sphere, 292, 368, 422, 430, 501 Time series, 36, 44, 221, 223-5, 382-3
Standard deviation, 406-7 Total differential equation, 421-2
Stationary values, 180, 194, 353 Trapezoidal rule for integration, 398
Step-functions, 31, 34, 91, 102
Straight line, 63-9, 76-7 Units, change of, 13-4, 15-6, 139
Surface classes and systems, 272, 416,
423-30 Variables, 8-9, 29-32, 43-5, 46-8, 98-9
Surfaces and functions, 270-5 Velocity, 15, 150-1, 152
of revolution, 292, 523-4, 532
Symmetrical functions, 41, 269 Young's Theorem, 301-2, 309
ECONOMIC APPLICATIONS
Average cost, 120-1, 155-6, 261-3, Cost functions, elasticity of, 260-3
401, 505-6 normal conditions of, 119, 261-3
product, 133, 178, 267, 312-3, 321-2 particular forms of, 120, 155-6, 199,
revenue, 117, 153-4, 257-60, 400-1 262-3, 535
shifting of, 118, 120
Bilateral monopoly, 381
Demand and offer curves, 121-2
Capital and interest, 234, 248-50, Demand functions, curves and sur-
362-4, 403, 404-5 faces (for consumers' goods), 28-
and saving, 537-40 9, 108-10, 121, 281-3, 374-5, 434,
values, 233, 401-3, 404-5 509-17
Coefficient of relative cost of produc- continuity of, 110-2
tion and of relative efficiency of dynamic forms of, 435-8, 533-6
organisation, 261 elasticity of, 117, 255-60, 266, 310-
Competitive and complementary 2, 375, 512-3, 520
goods and factors, 282, 311, 361- normal conditions of, 112, 267-9
2, 509, 513, 619 particular forms of, 112-6, 117, 154,
Compound interest problems, 31, 199, 259, 282, 311-2, 435-8, 535
228-34, 235-7, 248-50, 362-4, shifting of, 110, 115-6, 282, 283
376-8, 401-3, 404-5, 418, 419 stability of, 375, 510-2, 614-7
Conjectural variations, 203, 346-7 Demand functions (for factors of
Constant outlay curves, 256-7 production), 369-74, 484-5, 502-9,
product curves, 285, 320, 341-3, 519
370-1 elasticities of, 373-4, 508-9, 519
returns to scale, 320, 363, 371-2, stability of, 370, 502-3
505 Demand functions (for loans), 376-8,
Consumers' choice, theory of, 124-9, 520
289-91, 374-8, 438-42, 509-17, stability of, 377
520 Discount curves, 236-7, 250, 293
Cost functions, curves and surfaces, Disutility of labour, 537-8
117-9, 155-6, 262, 283, 370-2, Duopoly problems, 200-4, 345-7
506, 533-5 Durable capital goods, 404-5
average and marginal, 120-1, 155-6,
261-3, 371-2, 401, 444, 506-6, 522 Elasticities of cost, 260-3
INDEX 547
Elasticities of demand (consumers' Mathematical methods in economics,
goods), 117, 255-60, 266, 310-2, 107-8
375, 512-3, 520 Monopoly problems, 196-200, 359-64,
of demand (factors of production), 381, 518, 519, 522, 533-6
373-4, 508-9, 519
of productivity, 263, 350 National Savings Certificates, 230,
of substitution, 341-3, 344-5, 372-4, 240
504-5, 512-3, 520
of supply, 267
Pareto's income law, 222, 228, 407-8
Period of production, 262-4
Factors of production, 263, 284-9, Preference directions, 442
381-2, 537 lines, 442
demand for, 369-74, 502-9, 519 438-42
scales, 124, 290,
relative shares of, 364 Present or discounted values, 232-4,
236, 249-50, 362-5, 376, 401-3,
Increasing and decreasing returns, 404-5, 448-9
261, 263 Price, as average revenue, 117, 153
Indifference curves and surfaces (for as a discontinuous variable, 9, 111,
consumers' goods), 124-7, 156, 534
289-91, 344, 375, 439, 441-2, 511 variable over time, 434-8, 533-6
(for income over time), 127-9, 157, Production, capital and interest, 362-
291, 344, 377-8 4, 403
Indifference direction, 439 Production functions and surfaces,
piano, 440, 513 133, 178, 263, 267, 284-9, 312-3,
Investment, amount of, 228-32, 233, 362, 537
235-6, 402-3 linear and homogeneous, 320-2,
opportunity line, 124 343, 371-4, 434, 505-9, 541
normal conditions of, 286-9, 313,
Joint production, 350, 359-62, 518 320-2
particular forms of, 288-9, 313,
Linear homogeneous production func- 322, 343, 363, 403
Production indifference curve, 123
tion, 320-2, 343, 371-4, 434, 505-
9, 541
*
Loss leaders, 362 Rate of interest as marginal product

of capital, 539
Reaction curves, 201-4, 345-7
Marginal cost, 155-6, 261-3, 401, 505-
6, 522
Revenue, total, average and mar-
15, 116-7, 153-4, 257-60,
disutility of labour, 539 ginal,
increment of revenue, 1 53 400-1, 522
product, 178, 267, 312-3, 321-2, 539
productivity law, 372, 381-2, 434, Saving, rate of, 540
506 theory of, 537-40
rate of return over cost, 156, 159, Spatial distribution of consumers,
234, 345, 376-8 80-2
rate of substitution, 156, 286, 341- Speculative demand and supply, 434-
3, 344-5, 372-4, 439-41, 511-2, 7, 533-6
513-7 Stability of demand, 370, 375, 377,
rate of time preference, 157, 344, 502-3, 510-2, 514-7
377-8 of price over time, 436
revenue, 15, 153-4, 257-60, 400-1, of supply, 520
522 Substitution, 340-5, 372-4, 439-41,
supply price, 155 503-5, 509, 512-3, 517, 520
utility, 265, 314, 344, 537 Substitutional and limitational fac-
utility of money, 512 tors, 381-2
548 INDEX
Supply functional, .522 Transformation functions and curves,
Supply functions and curves, 121, 122-4, 156, 234, 284, 345, 376-8,
132, 209, 434, 520, 522 520
dynamic forms of, 43.5-8
Utility function*, 126, 289-91, 302,
elasticity of, 207
313-4, 439, 441, 471 9 537
stability of, 520
marginal, 265, 314, 344, 537
particular forms of, 127, 291, 314
Taxation, 60, 208 Wages as marginal product of labour,
Time-lag in production, 436-8 539
AUTHORS
Akarmaii, G., 404 Hogberi, L., 35
Allen, R. G. D., and Bowley, A. L., Hotelling, H., 60
222, 375
Johnson, W. E., 261
Bernoulli, John, 524, 527
Rowley, A. L., 155, 222, 239, 261, Kahn, R. F., 343
280, 381
Knight, F. H., 28S
Burnside, W. S., and Pant on, A. W.,
53 Lerner, A. P., 343
Cantor, G., 6, 86
Marshall, A., 109, 112, 121, 255, 256,
Champernowne, D. G., 251
411
Courant, R., 301, 478, 523, 536
Moore, H. L., 255, 261
Cournot, A, A., 200, 204
Dedekind, R., 6, 86 Netto, E., 473, 478, 482
do la Vallee Poiissin, ('. J., 366, 422,
426, 523 Osgood, W. F., 366, 431, 523
Douglas, P. H., 288, 294
Pareto, V., 222, 280, 290, 302, 375, 382
Edelberg, V., 364 Phillips, K. G., 332
Edgeworth, F. Y., 196, 442 Piaggio, H. T. H., 421, 422, 427
Elderton, W. P., 240 Pigou, A. C., 266
Evans, G. 0., 437, 444, 534
Ramsey, F. P., 490, 537
Fisher, Irving, 124, 156, 234, 241, Riemann, B., 385
248, 376, 442 Robinson, Joan, 259, 343
Fowler, R. F., 241 Roos, ( F., 362, 437, 534
1
.
Frisch, R., 46, 109, 203

Sehn-ider, E., 80, 264, 522
Georgescu-Roegon, N., 382 Schultz, H., 26, 255, 311
Gibrat, R., 222 Shannon, H. A., 238
Giffen, R., 112 Stackelberg, H., 362
Goursat, E., 385
Theiss, E., 437
Haldane, J. B. S., 152 Tinbergen, ,J., 437
Hardy, G. H., 6, 86, 92, 97, 164, 231,
452 Volterra, V., 521
Hayek, F. A., 123
Hicks, J. R., 204, 343, 362 Whitehead, A. N., 134
Hicks, J. R., and Allen, R. G. D., Wicksell, K., 143, 184, 248, 379, 380,
344, 375, 517 404
PRINTED IN GRKA1 BRITAIN BY ROBERT MACLKHOSE AND CO. LTD.
THE IJN1VKRSITY PRESS, M.ASMiW

Mathematicalanal033535mbp PDF

Uploaded by

Copyright:

Available Formats

Mathematicalanal033535mbp PDF

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

Mathematicalanal033535mbp PDF

Uploaded by

Copyright:

Available Formats

OU162017>m

LONDON SCHOOL OF ECONOMICS AND POLITICAL SCIENCE

STUDIES IN STATISTICS AND SCIENTIFIC METHOD

No. 3. MATHEMATICAL ANALYSIS FOR ECONOMISTS

The general science of mathematics is concerned with

To connect elements in laws according to some logical

MORRIS R. COHEN, Reason and Nature

MAQMILLAN AND CO., LIMITED

reader, in solving them, will become familiar with the mathematical

original manuscript. I am also indebted to Mr. G. J. Nash who has

I. NUMBERS AND VARIABLES

2.4 Classification of functions - - - - - - 38

IV. LIMITS AND CONTINUITY OF FUNCTIONS 85

single- valued function

Limits of functions ; continuity of functions

V. FUNCTIONS AND DIAGRAMS IN ECONOMIC THEORY - 107

VI. DERIVATIVES AND THEIR INTERPRETATION - - 134

VII. THE TECHNIQUE OF DERIVATION 160

VIII. APPLICATIONS OF DERIVATIVES - - - - 179

IX. EXPONENTIAL AND LOGARITHMIC FUNCTIONS - 211

compound interest problems 238

X. LOGARITHMIC DERIVATION 242

10.2 Logarithmic derivation

elasticities and their applications - - - 264

XII. PARTIAL DERIVATIVES AND THEIR APPLICATIONS - 296

XIII. DIFFERENTIALS AND DIFFERENTIATION - 326

XV. INTEGRALS OF FUNCTIONS OF ONE VARIABLE - - 384

15.4 The technique of integration

XVI. DIFFERENTIAL EQUATIONS - - - - - - 412

cations of differential equations - - - 442

XVII. EXPANSIONS, TAYLOR'S SERIES AND HIGHER ORDER

17.8 Differentials of a function of two dependent variables

EXAMPLES XVII Infinite series ; expansions ; higher

XVIII. DETERMINANTS, LINEAR EQUATIONS AND QUADRATIC

EXAMPLES XVIII Determinants ; linear equations ;

quadratic forms - - - - - - -492

XIX. FURTHER PROBLEMS OF MAXIMUM AND MINIMUM

20.3 The method of the calculus of variations - - - 524

and longer accounts such as

good short treatments are

The reader wishing to revise his knowledge of " matriculation algebra

an introduction to the calculus).

Thompson : Calculus made Easy (2nd Ed., 1919).

Hardy : A Course of Pure Mathematics (3rd Ed., 1921).

matical, not literal, usage.

In trigonometry, a, j8, y can denote constant angles and 0, <f),

NUMBERS AND VARIABLES

Elementary geometry, for example, is largely an exercise in formal

definitely continuous. It can be

by relatively simple logical reasoning and the introduction of

Though an abstract development, mathematics is not just a

In concluding these introductory remarks, we can note that \

fundamental bases of the mathematical technique have received

1.2 Numbers of various types.

simplest notions of arithmetic. Some light is then thrown on the

The integers can thus be used to indicateany order or sequence and

a stick.This is the cardinal aspect of counting and of the integers,

for example, in the simplest equivalent form J.

by TT and e which play extremely important parts in mathematical