Edwin E Moise Calculus PDF

CALCULUS
SECOND EDITION
EDWIN E. MOISE
Harvard University
ADDISON-WESLEY PUBLISHING COMPANY

Reading, Massachusetts · Menlo Park, California · London · Don Mills, Ontario
This book is in the
ADDISON-WESLEY SERIES IN MATHEMATICS
Consulting Editor: LYNN H. LOOMIS
Copyright© 197.2 by Addison-Wesley Publishing Company, Inc. Philippines copyright 1972

by Addison-Wesley Publishing Company, Inc.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical, photocopying,
recording, or otherwise, without the prior written permission of the publisher. Printed in the
United States of America. Published simultaneously in Canada. Library of Congress Catalog
Card No. 76-150576.
Author's Note on the Second Edition
The preface to the first edition was an explanation of the author's intentions, and
since these intentions have not changed, the original preface is reprinted after this one.
But the present edition is a thorough revision of the first, with many major changes
and even more minor ones. Some of these are as follows.
1. The use of language has been simplified throughout. Excessive colloquialisms
have been eliminated.
2. Many problems have been added, most of them being easy. In cases where
several problems form a sequence, they have been combined into a single problem,
with parts (a), (b), (c), .... Thus it is now safe to assign all odd-numbered problems,
without checking to make sure that problem-sequences are not being broken up.
3. Various long sections have been divided into two parts.
4. The classical definition of a limit has been restored. Exploratory problems,
dealing with limits and continuity in terms of "boxes," have been eliminated, and
this material has been inserted in later portions of the text. ·
5. Section 5.8, on the derivative of one function with respect to another, has been
completely recast, following suggestions of Professor Hugh Thurston, of the University
of British Columbia. The new version is mathematically straightforward, and it
bridges the gap between the modern concepts of function and derivative and the
"fractional" notation du/dv commonly used in physics.
6. Chapter 8, on the conic sections, has been shortened and simplified, by
omitting various topics not ordinarily covered in a first course in calculus. In
particular, the section on the geometry of the ellipse has been omitted. In a way it is
a pity to leave this out, because it is very good mathematics, but in a first course in
calculus we barely have time for essentials.
7. In the chapters on vector spaces, the standard use of the terms "vector space"
and "inner product space" has been restored.
8. The old Chapter 10, on number theory and partial fractions, has been omitted.
The above remarks about the geometry of the ellipse also apply here.
9. The chapter on infinite series has been completely recast. In the first edition,
the idea of uniform convergence was built into the presentation, almost from the
outset; it was used, in various special cases, to justify term-wise integration, long
before the general definition of uniform convergenc� was stated. This treatment had
advantages, for some students, but it had a serious tlisadvantage: it meant that the
hardest part of the study of infinite series could not be skipped, or even postponed.
The chapter has now been arranged in such a way that the hardest parts of it come
iii
iv Author's Note on the Second Edition
last. Term-wise integration and differentiation of power series are introduced early,
and play a central part throughout the chapter; but the justification of these processes
, is saved for the end.
The construction of the complex numbers (using congruence classes of real
polynomials modulo 1 + x2) has been moved to an appendix.
10. The chapter on linear transformations, matrices, and determinants has been
recast and simplified in various ways. For example, the idea of isometries between
subspaces has been omitted (from the text and therefore from the problems.)
11. In the chapter on functions of several variables, the Leibniz notation for
partial derivatives has been introduced, in parallel with the subscript notation
f.,,fv, . . .. The former notation is of course universal in physics, and it cannot be
denied that it makes the chain rule easier to remember.
These examples should make it plain that this is not a perfunctory revision.
The intent of the revision is to make the book more teachable and more flexible,
without weakening its mathematical content. Some sections (as indicated above)
have been omitted outright. Some chapters have been recast in such a way that more
topics can be omitted at the teacher's discretion. But the main substance of the book,
and the conception of calculus that it attempts to teach, have not been changed. All
the hard problems in the first edition have been retained (except for one very embar
rassing case, in which I asked the student to prove a false theorem).
Most of the calculus books now in print are of one of the following three types:
1) Some are written on a high plateau of austerity and rigor, and the Devil take
the hindmost.
2) Some are "quick calculus" books. A typical device, in this sort of book, is to
use the Fundamental Theorem of Integral Calculus as a definition of the definite
integral. This enables the student to imitate the behavior of mathematicians, in
calculating definite integrals, without sharing the mathematicians' conception of
what the problem meant in the first place.
3) Some are a combination of types 1 and 2, with exact theoretical material

included in the text but not in the problems. A course based on a book like this is
evidently intended to play a double game, in which some of the students learn the
text, but most of them study, in effect, a "quick calculus" course based on the same
sort of problems used in books of type 2.
The present book is of none of these three types. It is addressed to all students
who ordinarily take calculus courses and pass them, and it is designed to teach the
ideas of calculus, at least in some form, to these students. In a sense we are playing
a double game, because the teacher has a great deal of choice. If the entire text is
taught, then the course is logically complete, and nothing that it includes needs to be
taught again later. But many of the theoretical sections, especially at the ends of
chapters, giving proofs of "foundational" theorems, can easily be omitted. For
example, if we omit Sections 5.6, 5.7, and 5.8, then something is subtracted from the
course, but nothing is disrupted or confused. And in Chapter 10 we can stop at almost
any point. But even if such omissions are made on a large scale, the remaining hard
core of the book conveys the ideas of calculus in conceptual forms.
Author's Note on the Second Edition v
The rock-bottom minimum, in any good mathematics course, is for the student
to attach conceptual meanings to the problems that he solves and to the "answers"
that he computes. If we settle for less than this, we are making a bad bargain. The
hard fact is that "practical" calculus courses are not practical. In real life it seldom,
if ever, happens that a mathematical problem takes the form of a homework exercise
which can be solved by copying the pattern of the "solved problems" that immediately
precede it. In physics it is the conceptual definite integral that is crucial, and numerical
valuations are often done by computers. Thus the art of setting up integrals is often
more useful than the art of calculating them by elementary methods. The same
principle applies very widely. When people put their m�thematical training to
practical use, they seldom need the logical refinements that appear in a thorough
treatise, but they nearly always use their conceptual grasp-at some level-of mathe
matical ideas. Obviously, many techniques are needed, and in this book we have
worked hard to teach them. But as the same time we have tried to produce the sort
of conceptual grasp of mathematics that can be put to work in real life.
New York City, N. Y. E.E.M.

October 1971
Preface
The mathematical content of the first ten chapters of this book is familiar and easy
to describe. These chapters present, more thoroughly than is customary, the material
normally covered in one-year introductions to college calculus, and end with a chapter
on infinite series. (This portion of the book is being published separately, under
the title Elements of Calculus.) In the last four chapters of the complete edition, the
choice of material is nowhere nearly so traditional. In particular, we have laid heavy
stress on the methods of linear algebra.
In the latter portion of this preface, we explain the considerations on which the
selection of topics in the last few chapters is based. Most of the novelties in the
Elements are in the style of treatment; and the ideas underlying them may best be
explained by means of numerous examples.
1. THE SPIRAL PROCESS
The central concepts of the calculus are deep. It is not to be expected that they can
be learned ali at once, in the forms in which a modern mathematician thinks of them.
Therefore, in this book, the more difficult ideas are presented in a series of different
forms, in ascending order of difficulty, generality, and exactitude. Thus the idea of
the definite integral makes its first and simplest appearance in Section 2.10; it is
generalized in Section 3.7; and it is not presented in final form (using Riemann sums)
until Section 7.1, where Riemann sums are needed, in the calculation of arc length.
Similarly, the chain rule for derivatives appears first in Section 3.6, for powers
and square roots of functions; it is proposed, in more general forms, in Problem
Sets 3.6, 3.8, 4.3, and 4.5; and it appears in final form only in Section 4.6.
The mean-value theorem is first stated, in geometric terms, in Section 3.2, before
any formal definition of the derivative. It is used freely thereafter. Finally, in Section
5.7, it is proved, after the ideas needed in the proof have been used and motivated
in other ways.
The idea of the limit of a function appears first in Section 2. 7. The formal defini
tion is in Section 3.3. Earlier sections include a lengthy preparation for the
formal definition, designed to eliminate in advance as many of its difficulties as
possible. This purpose is served by the text of Sections 1.4 and 2.5. Thus the style of
treatment is such that an inspection of isolated sections of the book is likely to lead
to an overestimate of the difficulty of the course. The point is that the sections are
not isolated: difficult discussions have been provided with elaborate foundations, in
the text and especially in the problems.
vi
Preface vii
The spiral treatment, in which concepts appear in various forms as the theory
develops, is intended to make the concepts easier to learn. But this is not its only
purpose. The processes by which special ideas are generalized, and heuristic ideas
are made concrete and exact, are part of the substance of what we ought to be
teaching. Thus the heuristic treatment of exponentials and logarithms, in Section
4.9, is not given merely in order to make the student's life easier. The transition from
Section 4.9 to Sections 4.10 and 4.11 (in which the theory is based on the definition
In x = n (dt/t)) is valuable in itself, as an illustration of a recasting process which
is essential both in the growth of mathematics and in the growth of the people who
use it.
2. MOTIVATION
The desire to solve interesting puzzles is very strong; there. is no maturity level at
which it disappears; and we should appeal to it continually. Most of the time,
however, when new ideas are introduced, they ought to be motivated by a sense of
power, and by the light that they throw on ideas already regarded as significant.
For example, if we present Riemann sums, in full generality, long before we deal
with problems in which they are needed, it is not reasonable to expect the student to
master their complications. Similarly, the completeness of the real number system,
in the sense of Dedekind, is not needed at all in the theory of pointwise limits: this
theory takes exactly the same form in the rational domain as in the real domain.
If we postpone the idea of completeness until the point where it is needed, in the
study of functions continuous on an interval, it is more likely to be understood,
partly because it is more likely to get the student's attention.
The problem of motivating the idea of the limit of a function involves a peculiar
difficulty. The only cases in which limx-+a f(x) is easy to calculate are those in which
f is a continuous function, described by a simple formula. In these cases, the formula
works just as well for x = a as for other values of x; in practice, it turns out that
the limit isf(a); and the student is likely to get the idea that the expression limx-+a f(x)
is merely a devious and pretentious description of j(a). If we avoid this trouble by
starting with significant cases, such as
1. sin x
1m--,
X-+0 X
then the technical difficulties are formidable, and workable problem material is hard
to come by. If we choose, instead, to discuss limits of sequences, then we have
evaded the issue by changing the subject: in the differential calculus, limits of
functions are what we need.
But there is a fourth alternative: we can introduce the idea of a limit not as a
subject in its own right, but as a device for solving a problem. In Section 2.7 we
mention limits for the first time, in finding the limit of a linear function. To pass
to the limit, in this case, we merely plug the hole in a punctured line. This process
has no intrinsic significance. But in the context of Section 2.7, it has an extrinsic
viii Preface
significance, because it is used to solve a nontrivial problem, namely, the problem

of finding the slope of the tangent to a parabola. Similarly, in Section 2.10, we use
the idea of the limit of a sequence, in a technically simple case, in order to find the
area of a parabolic segment. (A formal definition of the limit of a sequence finally
appears in Section I 0.1 ). There are many other points at which ideas are introduced,
in simple forms, in connection with a discussion of something else.
3. BLACK BOXES
It is generally agreed that in a physics laboratory the student should build as much as
possible of his own equipment. Nobody learns very much by watching the per
formance of the proverbial "black box." In mathematics the situation is similar:
we do not learn mathematical principles by hearing them mentioned once, no matter
how elegantly; we need to live with them and use them. Therefore, in this book
certain extremely powerful theorems have been proved long before being stated.
That is, the proof has been presented, in the form of a method of solving a certain
class of problems; and after the student has learned the idea by using it on many
problems, we have summed up the situation by stating the general theorem that the
proof proves. This scheme costs very little time, even in the short run; and in the
long run it is likely to save a great deal of time. The point is that if we allow recipes
to take the place of ideas, in a first course, then the ideas need to be taught all over
again later; and the second attempt may be harder, because the problem-solving
motivation for these particular ideas has already been used up.
There are good reasons for not giving examples of this technique.
It should be understood that the avoidance of black boxes has no particular
connection with the pursuit of logical rigor. Indeed, if we have to choose, it is better
to master an idea in an heuristic form, by using it repeatedly, than to listen once to a
rigorous exposition, and then forget it.
4. PROBLEMS
In a quick examination of a textbook, it is not a good idea to read the text and skip
the problems; it is better to read the problems and skip the text. The problems
represent the life that the student leads when he studies the course; and any ideas
that do not appear in them are unlikely to be learned, no matter how much preachment
may be devoted to them.
In this book, a variety of problems are used for a variety of purposes. There are:
I) Technical problems, as, for example, in the chapter on the technique of integration.
These are carefully graded, and often they form sequences, in which the answer
to one problem can be used in the solution of the next.
2) Theoretical problems, some easy, some hard. Vigorous attempts have been
made to find easy ones, so as to avoid a dichotomy between techniques (which the
student really uses) and "theory" (of which he is intermittently a spectator).
Preface ix
3) Puzzle problems.
4) Sketching exercises, in which the student is asked to translate back and forth
between analytic ideas and visual images.
5) Discovery problems, which anticipate, in special cases, ideas which will later
be explained in the text.
There is wide general agreement on the content of the first year course in college
calculus; and in writing the Elements, the author was in the happy position of
working on the basis of a consensus with which he was fully in sympathy. But there is
no such general agreement on the content of a course in intermediate calculus. In
the past decade, calculus courses have tended to grow, by including various topics
from advanced calculus and linear algebra. But it is not easy to decide which of these
topics should be included, and what relative stress should be placed on them; and in
fact there is no reason to suppose that such questions have unique answers.
On the other hand, every book and every course must make some choices, and
then stick to them long enough to permit a valid learning process. If the pursuit of
flexibility turns an intermediate calculus book into an anthology, then its little pieces
are unlikely to have any lasting effect. For example, if the treatment of infinite series
is sketchy, then its residuum in the mind of the student may include hardly more than
the ratio test. And the dangers presented by brief treatments of linear algebra are
worse.
Modern algebra is modern because its motivations and its applications came late.
Today, there are very good reasons for studying groups, rings, fields, vector spaces,
normed vector spaces, inner product spaces, linear transformations, matrices, and so
on. But the logical simplicity of the rudiments of these theories is misleading. For
example, the manipulative process of multiplying matrices can be taught to almost
anybody, at almost any level; but the significant applications of this process are
another matter entirely. In a short treatment of axiomatic and linear algebra, at the
freshman or sophomore level, we cannot presuppose knowledge of the significant
applications, and we have no time in which to present them. Thus we may fall into a
peculiar form of use-mention confusion: the reader hopes that the ideas of modern
algebra are going to be used, but in the end he sees that they have merely been
mentioned.
For these reasons we have tried, throughout, never to state an algebraic definition
until the reader already knows at least one important instance of the idea that the
definition describes; and once an algebraic idea has been introduced, we have tried,
throughout, to put it to work for the purposes that it is good for. Thus, for example,
matrices are introduced as a shorthand for handling linear transformations; and
thereafter the treatment of the two is closely tied together. The Schwarz inequality is
first introduced (on page 521) as a theorem in Cartesian three-space, and for this
case it is proved by the trivial observation that cos2 f) � 1 for every fJ. Later, on
page 536, it is proved in the general case, and thereafter it is used in a great variety
of ways, to trivialize problems which would not otherwise be trivial. It appears in
disguised forms in many problems (which should not be listed here). These examples
are typical of the style of Chapters 11 through 13. It appears to the author that the
x Preface
nature, the purposes, and the power of algebraic methods are hot likely to be under
stood unless they are conveyed to the student by some such extended experience.
The most impressive, but also the most difficult, of these applications occurs in
Chapter 12, on Fourier series. This topic is not ordinarily included in intermediate
courses; and if something must be omitted, in teaching a course from this book,
Chapter 12 is an excellent candidate for omission. (None of the material in it is used
later.)
Chapters 1 through 13 amount to more than 600 pages; something had to be
shortened; and so the treatment of functions of many variables is shorter than might
have been expected, and there is no separate chapter on differential equations. It
should be noted, however, that there is a substantial treatment of linear differential
equations at the end of Chapter 13, and that the viewpoint of differential equations
has been stressed throughout. (Recall, for example, the treatment of the fundamental
theorem of integral calculus, and of the elementary functions, in the Elements.). In
Chapter IO, the standard method of showing that a given series converges to a given
function is first to show that the series and the function satisfy the same differential
equation, and then to show that the differential equation (with initial condition) has
only one solution. Usually, the series is derived from the differential equation, and so
the student is not likely to be surprised when the same process is applied later to
equations whose solutions were not previously known. For this sort of reason, the
book conveys much more of the spirit and methodology of differential equations
than the table of contents would suggest.
Moreover, it appeared to the author that the natural sequels of the material in
Chapter 14 would grow exponentially more difficult, and that they rightly belong in
an advanced calculus course. The hard fact is that multivariate calculus, once we get
past its beginnings, is not an elementary subject; and if we try to make it seem elemen
tary, we are likely to give up both intuition and logic in favor of a bewildering
formalism. Thus it appeared, at the end of Chapter 14, that we should say either
much more, or no more at all; and since every book-even a calculus book-has got
to end somewhere, the choice was clear.
The above discussion is an attempt to indicate some of the author's objectives,
and some of the methods used in pursuing them. Obviously no such discussion can
prove anything about the extent of the contribution that the text makes to the
achievement of these objectives. A great deal has happened, in the teaching of cal
culus, in the past decade, and it remains to be seen how much more can be accom
plished, and how.
New York City, N. Y. E. E. M.

October 1971
Contents
Chapter 1 Inequalities
1.1 Introduction 1
1.2 Products which are equal to zero 2
1.3 Order . 3
1.4 Absolute values. Intervals on the number line 9
Chapter 2 Analytic Geometry
2.1 Introduction . 16
2.2 Coordinate systems. The distance formula 16
2.3 The graph of a condition. Equations for circles . 21
2.4 Equations of lin�s. Slopes, parallelism, and perpendicularity 26
2.5 Graphs of inequalities. And, or, and if ... then 33
2.6 Parabolas . 38
2.7 Tangents . 43
2.8 A shorthand for sums 49
2.9 The induction principle and the well-ordering principle . 51
2.10 Solution of the area problem for parabolas . 57
Chapter 3 Functions, Derivatives, and Integrals
3.1 The idea of a function . 63

3.2 The derivative of a function , intuitively considered 69
3.3 Continuity and limits 75
3.4 Theorems on limits . 82
3.5 The process of differentiation 89
3.6 The process of differentiation: roots and powers of functions . 97
3.7 The integral of a nonnegative function 102
3.8 The derivative of the integral . 109
3.9 Uniformly accelerated motion . 119
*3.10 Proof of the formula for the derivative of the integral 124
Chapter 4 Trigonometric and Exponential Functions
4.1 Directed angles. Trigonometric functions of angles and numbers . 128

4.2 The law of cosines and the addition formulas 135
4.3 The derivatives of the trigonometric functions; the differences tlx and
fl/; the squeeze principle 139
xi
xii Contents
4.4 The approximation of differences by differentials 148

4.5 Composition of functions 154
4.6 The chain rule 159
4. 7 Invertible functions. The inverse trigonometric functions 165
4.8 Simpson's rule. The computation of 1T 176
4.9 Exponentials and logarithms 185
4.10 The functions In and exp 191
4.11 Exponentials and logarithms. The existence of e 197
Chapter 5 The Variation of Continuous Functions
5.1 Intervals on which a function increases, or decreases 206

5.2 Local maxima and minima, direction of concavity, inflection points 211
5.3 The behavior of functions at infinity . 216
5.4 The introduction of functions into geometric problems; the use of
existence theorems as shortcuts 223
5.5 The use of functional equations as shortcuts . 232
5.6 The completeness of R and the existence of maxima 238
5.7 The mean-value theorem and the no-jump theorem . 246
5.8 The derivative of one function with respect to another 250
Chapter 6 The Technique of Integration
6.1 Introduction 254

6.2 Independent variables and indefinite integrals 255
6.3 Integrals leading to the logarithm and the inverse secant. Algebraic
devices 265
6.4 Integration by parts . 273
6.5 Integration of powers of trigonometric functions 278
6.6 Integration by substitution . 284
6.7 Algebraic substitutions 291
6.8 Algebraic devices: completing the square and partial fractions 297
Chapter 7 The Definite Integral
7.1 The problem of arc length 303

7.2 The definite integral, defined as a limit of sample sums . 308
7.3 The calculation of volumes, by the method of disks . 315
7.4 The general method of cross sections, and the method of shells 321
7.5 The area of a surface of revolution 327
7.6 Moments and centroids. The theorems of Pappus 335
7.7 Improper integrals 344
*7.8 The integrability of continuous functions . 350
Chapter 8 The Conic Sections
8.1 Translation of axes 356

8.2 The ellipse 360
8.3 The hyperbola 366
8.4 The general equation of the second degree. Rotation of axes 372
Chapter 9 Paths and Vectors in a Plane
9.1 Motion of a particle in a plane 381

9.2 The parametric mean-value theorem; l'Hopital's rule 385
9.3 Other forms of I'Hopital's rule . 393
9.4 Polar coordinates 397
9.5 Areas in polar coordinates 402
9.6 The length of a path . 405
9.7 Vectors in a plane 409
9.8 Free vectors . 415
9.9 Velocity, acceleration, and curvature 422
9.10 Concluding remarks on vector spaces and inner product spaces 430
Chapter 10 Infinite Series
10.l Limits of sequences 431

10.2 Infinite series. Convergence. Comparison tests 437
10.3 Absolute convergence. Alternating series 445
10.4 Estimates of remainders. 448
10.5 Termwise integration of series. Power series for Tan-1 and In 453
10.6 The ratio test for absolute convergence. Applications to power series 457
10. 7 Power series for exp, sin, and cos 463
10.8 The binomial series . 468
10.9 Taylor series . 473
10.10 Taylor's theorem: Estimates of remainders 477
10.11 The complex number system 479
10.12 Sequences and series of complex numbers. The complex exponential
function 484
10.13 De Moivre's theorem 489
*10.14 The radius of convergence. Differentiation of complex power series 493
*10.15 Integration and differentiation of real power series . 499
Chapter 11 Vector Spaces and Inner Products
11.1 Cartesian coordinate systems in three-dimensional space 508

11.2 Direction cosines. The directed normal form 512
11.3 Three-dimensional space, regarded as an inner-product space 518
11.4 The dimension of a vector space. Various ways to form a basis 526
11.5 Orthonormal bases . 530
11.6 The Schwarz inequality. More general concepts of norm and distance 533
Chapter 12 Fourier Series
12.1 Projections into a subspace, trigonometric polynomials and Fourier

series . 541
12.2 Uniform approximations by trigonometric polynomials 549
12.3 Integration of Fourier series. The uniform convergence theorem . 556
xiv Contents
Chapter 13 Linear Transformations, Matrices, and Determinants
13.1 Linear transformations . 563

13.2 Composition of linear transformations and multiplication of matrices 570
13.3 Formal properties of the algebra of matrices. Groups and rings 577
13.4 The determinant function 582
13.5 Expansions by minors. Cramer's rule and inversion of matrices 590
13.6 Row and column operations. Linear independence of sets of functions 596
13.7 Linear differential equations 601
13.8 The dimension theorem for the space of solutions. The nonhomo
geneous case . 607
Chapter 14 Functions of Several Variables
14.1 Surfaces and solids in R3 614

14.2 The quadric surfaces 620
14.3 Functions of two variables. Slice functions and partial derivatives 626
14.4 Directional derivatives and differentiable functions . 634
14.5 The chain rule for paths 641
14.6 Differentiable functions of many variables. The chain rule. 644
14.7 Directional derivatives and gradients 648
14.8 Interior local maxima and minima, for functions of two variables
Level curves . 651
14.9 Double integrals, intuitively considered 660
14.10 Cylindrical coordinates in space. The definition of the integral 666
14.11 Moments and centroids of nonhomogeneous bodies 674
14.12 Line integrals 680
Appendix A The Shorthand of Logic and Set Theory . . 687

Appendix B Algebraic Operations with Limits of Functions 690
Appendix C Algebraic Operations with Limits of Sequences 695
Appendix D The Error in the Approximation 11/ ""' df . . 697
Appendix E The Continuity of Composite Functions 700
Appendix F The Error in Simpson's Rule . . . 702
Appendix G The Idea of a Measurable Set . 705
Appendix H Proof of the Northeast Theorem . 707
Appendix I Proof of the Formula for Path Length . . . . 711
Appendix J A Method for Constructing the Complex Numbers 713
Appendix K Iterated Limits. Mixed Partial Derivatives . . 717
Appendix L Possible Peculiarities of Functions of Two Variables 721
Appendix M Maxima and Minima for Functions of Two Variables 725
Appendix N An Exact Definition of the Idea of a Function 727
Selected Answers . 733
Index . 759
1 Inequalities
1.1 INTRODUCTION
In this book it is assumed that you know elementary geometry and the algebra of the
real number system. Theorems of plane geometry will be used only occasionally, and
there is no need to reexamine the subject as a whole.
Inequalities, however, are another matter. We shall be using them constantly,
and they are tricky. We shall therefore handle them with care. To derive the laws that
govern them we first need to recall the elementary laws of the number system. These
are as follows.
We have given the set R of real numbers, with the operations of addition and
multiplication. Thus the number system is a triplet
[R,+-
, ].
Addition and multiplication are subject to the following laws:
Closure. For every a and b in R, a+b and ab are in R.
Associativity. For every a and b,
a+(b+c)= (a+b)+c,
and
a(bc)= (ab)c.
Commutativity. For every a and b,
a+b=b+a and ab=ba.

Distributive Law. For every a , b, and c,
a(b+c)=ab+ac.
Existence of 0 and 1. There are two different numbers 0 and 1 such that
a+O=a and a· 1 =a
for every a.
Existence of Negatives. For every a there is a number -a such that a+(-a)= 0.
Existence of Reciprocals. For every a ¥: 0 there is a number I /a such that a · I /a=1.
1
2 Inequalities 1.2
These laws are called the field postulates; and any number system which satisfies
them is called afield. There are many such number systems: the real numbers form a
field, and so do the complex numbers. For a long time to come, however, we shall be
working only with the real numbers. Therefore, when we speak of numbers, we mean
real numbers, unless the contrary is stated.
We shall assume not only the field postulates but also the familiar laws based on
them. For example, we know that (a - b)(a + b) = a2 - b2, and that a· 0 = 0
for every a.
1.2 PRODUCTS WHICH ARE EQUAL TO ZERO
.When we perform calculations, we shall not stop to justify them on the basis of the
field postulates. But the following principle is worth special mention, because it is
used in reasoning processes which don't involve calculations:
Theorem 1. If ab = 0, then e ith er a = 0 or b = 0.
Proof
1) If a 0,= there is nothing to prove.
2) If a � 0, then a has a reciprocal. Therefore
1
-1 (ab) = - ·
0' 1. b = 0,
a a
and
b = 0.
Thus either a = 0 or b = 0.
Obviously it is possible that a and bare both 0. In Theorem 1 (and everywhere
=
else in mathematics) when we say either ... or . . , we allow the possibility of both.
.
PROBLEM SET 1.2
1. Show that if x2 =0, then x=0.

2. a) Obviously the numbers 1 and -1 are roots of the equation
(x - l)(x + 1) =0.
How do you know that no other number is a root of the equation?

b) Show that 2 and 3 are the only roots of the equation
x2 - 5x + 6 =0.
3. If 0 had a reciprocal, then its reciprocal would be a root of the equation
O·x=l.
Show that this equation has no root.

4. a) If ab =ac, does it follow that b = c? Why or why not?
b) If ab =ac, and a ;t 0, does it follow that b = c? Why or why not?
1.3 Order 3
5. a) Show that if ab c = 0, then a = 0 or b = 0 or c = 0.

b) Show that 1, 2, and 3 are the only roots of the equation
x3 - 6x2 + llx - 6 = 0.
6. a) If a2 = b2, does it follow that a = b? Why or why not?

b) If a2 = b2, what can you conclude about the relation between a and b? Why?
7. Under what conditions (if any) is it true that
1 1 1
- +- =-- ?
x a x +a
8. a) Under what conditions (if any) is it true that
(a + b)2 = a2 + b2?
b) Under what conditions (if any) is it true that
(a + b)3 = a3 + b3?
*9. Consider the "number system" which has only two elements 0 and 1, with addition and
multiplication defined by the following tables:
+ 0 1 0 1
----- ---
0 0 0 0 0
1 0 0 1
Which of the field postulates hold true, in this system? Which, if any, fail to hold?
(The answer to this question suggests that the field postulates are not, in themselves,
a very adequate description of the real number system.)
*10. Consider the number system in which the "numbers" are 0, 1, 2, and 3, with addition
and multiplication defined by the following tables:
+ 0 2 3 0 1 2 3
0 0 1 2 3 0 0 0 0 0
1 2 3 0 0 1 2 3
2 2 3 0 1 2 0 2 0 2
3 3 0 1 2 3 0 3 2
Exactly one of the field postulates fails to hold in this number system. Find out which
one. [Hint: Don't bother to test the Associative and Distributive Laws; in fact, they
hold true in this system, although the verifications are extremely tedious.]
Does Theorem 1 hold true in this system? Why or why not?
1.3 ORDER
We think of the real numbers as being arranged on a line, like this:
-V3 11'
f �
- 3 - 2 -1 0 2 3
4 Inequalities 1.3
When we write a<b, this means (roughly speaking) that a lies to the left of b on the
number line. Thus what we have in mind is a system
[R, +, ·, <],
where<is a relation having the following properties:
0.1. (Trichotomy) For every a and b in R, one and only one of the following
conditions holds:
a<b, or a= b, or b<a.
0.2. (Transitil'ity) If a a to mean a<b; a � b means
that either a b or a= b. A number
=
a is positive if a > O; a is negative if a<0. Zero is neither positive nor negative.

But 0.1 and 0.2 do not, by themselves, enable us to handle inequalities. We
need to know how<is related to + and · . The laws are the following:
MO. If a > 0 and b > 0, then ab > 0.
AO. If a<b, then a +c <b+c for every c.
These four laws, in combination, tell the whole story: all of the elementary laws
of inequalities can be derived from them. You will carry out this process, in the
following problem set. Meanwhile we state the theorems without proof.
Theorem 1. lf a > 0, then -a<0.
Theorem 2. If a<0, then -a > 0.
Theorem 3. If a<b, and c<d, then
a+c<b+d.
Theorem 4. An inequality is preserved if both sides are multiplied by the same positive
number.
That is, if a<band c > 0, then ac<be.

Similarly,
Theorem 5. An inequality is preserced if both sides are divided by the same positive
number.
That is, if ac<be and c > 0, then a< b.
Theorem 6. An inequality is raersed if both sides are multiplied by the same negative
number.
That is, if a be.

1.3 Order 5
Theorem 7. An inequality is rerersed if both sides are divided by the same negative
number.
That is, if be<ac, and c<0, then b > a.

Consider now an inequality involving an unknown number x, for example,
3x+4<5x + 7.
An expression like this, involving a variable, is called an open sentence; in an open

sentence, x marks the spot where numbers are to be inserted. Some numbers, when
substituted for x, may give true statements, and other numbers may give false state
ments. For example,
3·2+4<5·2+7
is true, because 10<17; but
3(-5) + 4<5(-5) + 7
is false, because -11 > -18.

In simple cases like this, it is easy to find out what numbers satisfy the inequality.
If
3x + 4<5x + 7, (1)
then
4<2x + 7, (2)
by AO. (We have added -3x to each side of the inequality.) Therefore
-3<2x, (3)
by AO; and so
x > -t, (4)
by Theorem 4. (We have multiplied, on each side, by t, and then written the inequality
backwards, to put x on the left.)
Thus every number which satisfies (1) also satisfies (4). And all of our steps can
be reversed. If
x > -t, (4)
then
-3<2x, (3)
by Theorem 4; therefore
4 <2x + 7, (2)
by AO; and so
3x + 4<5x + 7, (1)
by AO. Therefore every number which satisfies (4) also satisfies (1). We can sum all
this up briefly by writing
3x + 4<5x+7 <=>- x > -t.
Here the symbol <=>- is pronounced "is equivalent to." When we write<=:>- between
two inequalities (or any two open sentences of any kind) we mean that whenever one
of them is satisfied, so is the other.
6 Inequalities 1.3
We use a single-headed arrow to indicate that one condition implies another.

For example,
x > 0 => x2 > 0.
This is true. (Why?) But
2
(?) x > 0 <=> x > 0 (?)
is false, because x = -1 satisfies the second inequality but not the first. Similarly,
a = b => a2 = b2 is true, but
( ?) a = b <=> a2 = b2 (?)
is false, because if a ":/= 0 and b = -a, then the second inequality holds, but the first
does not.
The shorthand symbols<=> and => are worth learning and using. The reason is
that when we write down strings of formulas, in solving a problem, we ought to
indicate what the connection between them is supposed to be. We are more likely to
do this if we have a way of doing it briefly.
Using the symbols=> and<=>, we can restate some of the theorems of this section
in a more efficient way. For example, AO says that
a a + e a< b. (6)
These fit together to give:
The Addition Law of Order. a a + e 0, a ae< be.

Theorem 5 says that
fore > 0, ae< be => a< b.
These fit together to give:
The Multiplication Law of Order. Fore> 0, a ae<be.
This will be referred to as MLO. Theorems 6 and 7 say that
fore< 0, a ae > be, (7)

and
fore< 0, be<ae => b > a. (8)
And (8) can be rewritten in the form
fore< 0, ae > be => a< b. (8')
Thus Theorems 6 and 7 fit together to give:
Reversal of Order. For c<0, a ae > be.
We sum all this up in the short form on the next page. The meanings of the
abbreviations should be plain.
Order
Trich. For every a and b in R, one and only one of the following
conditions holds:
a<b, or a = b, or b <a.
Trans. a<band b < e � a< e.
MO. a> 0 and b> 0 � ab> 0.
AO. a 0 � -a< 0.
[Theorem 2. a<O � -a> 0.
[Theorem 3. a<band e< d � a+ e a+ e<b+ e.
MLO. Fore> 0, a ae<be.
RO. Fore< 0, a ae be
The last three of these are convenient in solving inequalities; they enable u
e <=> at each stage, instead of working first forward and then backward.
nple, the solution of the illustrative problem above can now be written like t
3x+ 4< 5x+ 7

<=> 4<2x+7 by ALO
<=> -3 <2x by ALO
<=> -t < x byMLO
<=> x > -� by definition of >.
A linear inequality is said to be solved when we find an equivalent inequalit

form x< a or x > a.
)BLEM SET 1.3
Solve the following inequalities, by writing a chain of equivalent inequalities, and gi

he right the reason for each step, as in the text.
5 - 3x > 17+x 2. 5x - 3<17x+ 1

5x+ 3 > 17x+ 1 4. 5+ 3x<17+x
-3x - 7 <x +5 6. -4x - 8 <2x+6
6x - 10 > 5x+ 3 8. 3 - 2x<4 - 3x
2x - 6 <2 - 2x 10. 6x - 2<3+x
2x+6<3+x 12. 6(x - 2) > x -3
In the following problems, we d evelop the theory in which all of the results of
ion are derived from Trich., Trans., MO, and AO. Therefore, at the start, thest
8 Inequalities 1.3
the only statements that can be given as reasons in proofs. In each problem, however,
you may assume that the results given in the preceding problems are known and you may
cite them as reasons.
13. Following are the steps in the proof of Theorem 1. Complete the proof by giving a
reason for each step.
a) a> 0 => 0 <a b) 0 <a => - a + 0 < -a+ a

c) -a+ 0 < -a+ a => -a< 0 d) a> 0 => -a<0
14. Following is an outline of the proof of Theorem 2. Complete the proof by giving a
reason for each =>.
a< 0 => -a + a< -a + 0 => 0 < -a

=> -a> 0.
15. a) Give a reason for the statement
a a+ c b + c b - a> 0.
(More than one step is needed here.)

b) Give a reason for the statement
c> 0 and b - a> 0 => (b - a)c > 0.
c) Prove Theorem 4.
17. Show that

2
x f;; 0, for every x.
[Hint: By Trich., there are three cases to be considered: x > 0, or x = 0, or x < 0.

2 2
Show that in each of these cases we have either x > 0 or x O.] =
18. Show that

y2 - 2y + 1 f;; 0, for every y.
19. a) Everybody knows that 1 > 0. Prove it, on the basis of the theory that we have
developed so far. (You may assume, of course, that 1 ,,e 0.)
b) Show that
1
a> 0 => - > 0.
a
That is, the reciprocal of every positive number is positive. [Hint: By Trich., it
will be sufficient to show that the conditions 1/a = 0 and 1/a< 0 are impossible.
Remember that a · 1/a = 1.]
20. Show that

c> 0 and ac < be => a<b.
CThis is Theorem 5.)
1.4 Absolute Values. Intervals on the Number Line 9
21. Give the reason for each step in the following proof of Theorem 6.
a b - a >0 and e <0
=> b - a >0 and -e> 0
=> (b - a)(-e) >0
=> ae - be> 0
=> ae> be.
22. Give the reason for each step in the following proof of Theorem 7.
be < ae and e <0

=> be < ae and -e >0
1
=> be < ae and -> 0
-e
1
=> ae - be >0 and - >0
-e
1
=> -(ae - be) >0
-e
=> b-a>O
=> b >a.
23. Is there a positive number which is smaller than all other positive numbers? Why or
why not?
24. Is there a negative number which is larger than all other negative numbers? Why
or why not?
*25. ls it possible to define, for the complex numbers, a relation < which obeys the laws
0.1 and 0.2? (That is, can an order relation be defined for the complex numbers?)
Why or why not?
*26. ls it possible to define, for the complex numbers, a relation < which satisfies not only
0.1 and 0.2 but also MO and AO? [Hint: Since i ;f:. 0, we must have i >0 or -i >0.)
The language in which these problems are stated ought to suggest what the answers are.
The answer to Problem 26 indicates why it is that arranging the complex numbers in an
order is not a useful proceeding. In the complex number system, no theory of inequalities
can be made to work.
1.4 ABSOLUTE VALUES. INTERVALS ON THE NUMBER LINE
The absolute value lxl of a number xis defined by the following two conditions:
1) If x � 0, then lxl = x.
2) I f x < 0, then lxl = -x.

Thus under Condition (1) we have
12 1=2,
and under Condition (2) we have
1-21 = -(-2) = 2,
10 Inequalities 1.4
Thus the operation I I leaves positive numbers unchanged, and replaces each negative
number by the corresponding positive number. On this basis it is easy to see that the
following theorem holds.
Theorem 1. For every x,
lxl � 0.
Proof There are two cases to consider.

Case 1. x � 0. Here lxl = x, by definition of lxJ. Therefore Ix\ � 0 in Case I.
Case 2. x < 0. Here lxl -x, by definition of lxl; and -x > 0, by Theorem 2
=
of the preceding section. Therefore lxl > 0 in Case 2.

Thus in each case we have lxl � 0.
\x\2 = x2.
This is true because \xi is either x or -x, and ( -x)2 = x2•
A number x is a square root of a number a if x2 a. For each a > 0, -J� is the

=
positive square root of a. Thus, for example, 9 has two square roots, 3 and -3;
and /9 is 3, which is the positive square root. We define Jo 0. Here and hereafter, =
we are assuming that positive numbers have roots of all orders-square roots, cube
roots, and so on.
\xi= -J�.
Proof By Theorem 2, lxl2 = x2, and so \xi is a square root of x2• By Theorem 1,
lxl � 0. Therefore lxl = -J x2, by definition of -J� .
1-x\ = I x!.
This is true because 1-xj = -J (-x)2 = -J x2 = Ix!.

Theorem 5. For every x and y,
\xy\ = jx\ \yj.

·
This is true because lxyl = -J (xy)2 = -J x2y2 = -J x2 -J y2

• =
Jxl IYI.
·
x�Ix!.
Here, as in the proof of Theorem 1, we need to consider two cases.
Case 1. x � 0. Here x�jxj, because !xi = x.

Case 2. x < 0. Here !xi = -x, and -x > 0. (Why?) Thus x < 0 < -x =
jxj, and so x < jxj.
Theorem 7. (The triangular inequality) For every x and y,
Ix+ YI� lxl+ jyj.

The trouble with this theorem, if we try to prove it by brute force, is that there
are too many cases to consider: each of the numbers x and y may or may not be
negative; and if x and y have different signs, x + y may or may not be negative. It
turns out, however, that we can get a proof by examining only two cases:
Case 1. x+ y � 0. In this case
Ix+ YI= x + y.
Since
x �lxl , and y�jy l,
we have
x+ Y�lxl+ IYI,
and so
Ix+ YI�lxl + jyj.
Case 2. Suppose that x+ y < 0. Then -x - y > 0. Therefore
Ix+ YI= 1-x - YI= 1(-x)+ (-y)I�I-xi+ 1-yj,

by the result of Case 1. Since I-xi = !xi and 1-yl = Jyl, we have
Jx+ YI� Jxl+ JyJ,

which was to be proved.
Theorem 8. Given d > 0. Then
Jxl < d <::::>- -d < x < d.
lxl <d
This is geometrically obvious: lxl is "the distance between 0 and x, on the number
line"; and the points that lie within a distance d of the origin are the numbers between
-d and d. We get a more general result by using any given point a instead of the
origin.
12 Inequalities 1.4
Theorem 9. Given d > 0, and any number a. Then
Ix - al <d <=> a- d< x<a+ d.
lx-aj<d
Proof In Theorem 8, substitute x- a for x. This gives

Ix - al <d <=> -d<x- a<d.
And
-d<x-a<d <=> a- d<x<a+�
(Reason?)
If a<b, then the set of all numbers between a and b is called an open interval,
and is denoted by (a, b).
(a, b)
0 a b
There is a shorthand for this sort of statement:
(a, b) = {x I a< x<b}.

The expression on the right denotes the set of all objects that satisfy the condition
following the vertical bar. This is called the solution set of the open sentence a<x<
b. Similarly, the set of all positive numbers is the solution set of the open sentence
x > O; this is denoted by {x I x > O}. Thus two open sentences are equivalent if
they have the same solution set.
Sometimes it turns out than an open sentence never gives a true statement, no
matter what we substitute for x. In such cases, the solution set is empty. The empty
set is denoted by { }. For example,
{x I .Jx2 + 1 = x- 1} = { }.
The notation { } is designed to suggest its meaning: we describe sets in the brace
notation; and when there is nothing written between the braces, this means that the
set has nothing in it.
If we add to the open interval (a, b) the endpoints a and b, we get a closed interval,
denoted by [a, b].
[a, b]
0 a b
Thus
[a, b] = {x I a � x � b}.
We shall also be dealing with "infinite intervals." In the first figure below,
the "infinite interval" is
(a, oo) = {x I a < x}.

Similarly,
( - oo, a) = {x I x < a},

as shown in the second figure below.
(a, oo)
0 a
(- oo, a)
0 a
This notation, in which "oo" is used as if it denoted a number, is not very logical,
but it is convenient. To keep track of the notation, you should think of fictitious
"numbers" - oo and oo as the "ends" of the number line, as shown below.
We also use "half-open" intervals:
[a, b) = {x I a � x < b} and ( a, b] = {x I a < x � b},
[a, b)
0 a b
(a, b]
0 a b
and "closed infinite" intervals:
[a, oo) = {x I x � a}, and ( - oo, a] = {x I x � a}.

[a, oo)
0 a
(, oo, a]
a O
14 Inequalities 1.4
Finally, we may refer to the whole real number system Ras the interval ( - oo, oo).
Thus we have a total of nine kinds of interval:
(a, b), [a, b], [a, b), (a, b],

(a, oo), [a, oo), ( - oo, a), (- oo, a],
(-oo, oo).
In some of the problems below, you may find it convenient to use the following:
Theorem 10. If lxl =

lyl, then x =
y or x = -y.
Proof
Jxl = IYI => Jxl2 IYl 2
=
=> x2 = y2
=> x2 - y2 = 0
=> (x - y)(x + y) = 0
=> x y = or x = -y.
(The converse is obvious.)
PROBLEM SET 1.4
Describe each of the following sets in the interval notation. Your answers should be in
a form like the following:
{xi 3x+4 < 5x+7} = (-f, oo .)

(This is the example discussed in Section 1.3 of the text.)
1. {x J 3x+4 > 4x+5} 2. {x I 3 - x > x - 3}
3. {x J lxl < 1} 4. {x I Ix - 31 � 2}
5. {x I Ix - 51 < 5} 6. {x I 2x+3 � 6x - 4}
7. {x I 11 - xi � 2} 8. {x I 12+xi < O}
9. a) Is it true that Yx2 = x for every x? Why or why not? Describe the set
. {x I Yx2 = x},
in the interval notation.
b) Describe the set
{xi Y(x+1)2=x+1},
in the interval notation.
Find out for what numbers x (if any) each of the following conditions holds. In each
case in which the solution set is an interval, the answer should be given in the interval
notation.
10. Yx2- 2x+1 = x - 1 11. lx2- 5x+61 = Ix -31 Ix - 21·
12. lx2 - 5x+61=x2 - 5x+6 13. Ix - 51 = l2x - 31

14. Ix+11 =11 - xi 15 . vx2+1 = x
16. Yx2-l=x 17. l2x -lj +Ix+31 � l3x+21
18. 1 7x +31 +1 3 - xi � 6 Ix + ll 19. ,12x - x2 = 1

20. l2x - x21 = x +2x2 21. Ix + ll +l2x +31 > 2
22. Ix - ll lx2 +xi = lx3 - xi
Indicate graphically, on a number scale, the places where the following conditions hold;
describe the graphs in the interval notation if possible.
23. lxl <2 24. Ix - 21 <1 2
25. l2x - 31 ;;;; i 26. Ix - 11 < -� and (also) Ix - 21 ;;;; 1

27. 13 - 2xl ;;;; i 28. Ix 21 < :l- and x �2
29. 1 1 - xi ;;;; 2 30. l2x - 41 < 1
3 1. Ix 11 < 2
- and lxl � �- 32. 1 2x - 1 1 � 1 and l2x - 11 ;;;; 1
33. a) Show that if b � 0, then
l�I 1�1= ·
(By definition, the reciprocal of lbl is the number y such that lbl · y = I. Therefore
it is sufficient to show that lbl 1 1/bl
· l.)=
b) Show that if b � 0, then
l,bl
a lal
=
fbl.
34. a) Show that for every a and b,
la - bl � lal - lbl.
(There is a short proof.)
b) Show that for every a and b,
la +bl � lal - lbl.
(The proof is short.)
35. For what numbers a is the fraction a/lal defined? What is this fraction equal to, for
various values of a?
36. Sketch
{x I Ix - 21 + 1 7 - xi = 5}
on the number line, and describe this set in the interval notation.
2 Analytic Geometry
2.1 INTRODUCTION
This chapter includes various topics which serve as a preparation for calculus. Some
of these topics are familiar to you, at least in some form. In such cases you should
still read the text carefully, in order to learn the terminology that will be used hereafter.
2.2 COORDINATE SYSTEMS. THE DISTANCE FORMULA
We shall now apply algebra to the study of geometry. We start with a plane, in the
usual sense of Euclidean geometry; and we suppose that a unit of distance has been
chosen, once for all, so that the distance between two points Pand Q is a well-defined
nonnegative number. The distance between the points P and Q is denoted by PQ.
(We say merely that PQ is nonnegative, rather than PQ > 0, because we are allowing
the case P = Q, and in this case PQ = 0.)
To set up a coordinate system in a plane, we first need to assign number-labels to
the points of a line. We choose a point 0 as the origin; it is given the label 0.
Each pointP1 to the right of 0 is labeled with the distance x1 = OP1, which is positive.
And each point P2 to the left of 0 is labeled with the number x2 = -OP2, which is
negative. Thus we have a matching scheme, under which each point of the line is
matched with exactly one real number.
p Q 0 RS T
I I
-2 -1 6 I I I
1 V2 2
For the points marked in the figure, the matching pairs are
P�-2, Q �-1, R�l,
S� ../2, T � 2, U � 71'.
Here the double arrow� is pronounced "is matched with." Every such pair has the
form P� x, where Pis a point and xis a number. A one-to-one matching scheme,
16
2.2 Coordinate Systems. The Distance Formula 17
between the elements of one set and the elements of another, is called a one-to-one
correspondence between the two sets.
If the correspondence is set up in the way that we have just described, then we
can compute the distance between any two points by means of the formula
Here P1 +--* x1 and P2 +--* x2• This distance formula holds no matter how the points
P1 and P2 are situated on the line:
0 0
0 0
0 0
0 0
and so on; in every case, P1P2 = lx2 - x11. Thus we have a one-to-one corre
spondence P � x, between the points of the line and the real numbers, such that the
distance formula holds for every pair of points. Such a correspondence is called a
coordinate system for the line. If P +--* x, then x is called the coordinate of P.
These ideas are summed up in the following postulate.
The Ruler Postulate. Every line has a coordinate system. And given any two points
0 and P of the line, there is a coordinate system in which the coordinate of 0 is 0
and the coordinate of P is positive.
0 p
0 x>O
On the basis of the ruler postulate, it is easy to set up a coordinate system in the
plane. We take two perpendicular lines Xand Y, intersecting in a point 0. On each
of the two lines we set up a coordinate system, in such a way that 0 +--* 0; that is, the
coordinate of 0 is zero on each of the lines Xand Y. Xis called the x-axis, Y is
called the y- axis, and the point 0 is called the origin.
Given any point P of the plane, we drop a perpendicular from P to the x-axis,
ending at a point M. The point M has a coordinate x, on the line X. If M +--* x,
then x is called the x-coordinate of P.
y
y N--------- -,p
I
x I
M x
x
Ml
I
L--------- y
p N
18 Analytic Geometry 2.2
Similarly, we drop a perpendicular from P to the y-axis, ending at a point N.

If N +--+ y, theny is called they- coordinate of P. Thus we have a matching scheme
p +--+ (x, y)
between the points P of the plane and the ordered pairs (x, y) of real numbers. The
order in which we write the numbers makes a difference. In the left-hand figure below,
and Q +--+ (2, 1).
We may speak of "the point (1, 2)" or "the point (x, y)," meaning "the point
which is matched with (1, 2)" or "the point which is matched with (x, y)." Thus we
may write P = (x, y), meaning P+--+ (x, y).
y
p
2 -- -
-,
I
I N -----
p'
tt -
I Q y
I
--- ----,
+ I
I I I
I I I
I I I
M
x x x
0 2 0
I
Obviously x and y are determined when Pis known. And Pis determined when
x and y are known, because the vertical line through M and the horizontal line
through N intersect in exactly one point. Thus we have a one-to-one correspondence
p +--+ (x,y)
between the points of the plane and the ordered pairs of real numbers. Such a corre
spondence is called a coordinate system for the plane. We need to see how the algebra
in this situation is related to the geometry.
Consider first the question of distance. If we know the coordinates (x1, y1) and
(x2, h) of two points P and Q, then the points are determined, and so the distance
between them is determined. The following theorem gives a formula for the distance.
2.2 Coordinate Systems. The Distance Formula 19
Theorem 1. If
and
then
PQ= .J(x2 - X1)2 + (Y2 - Y1)2.
Proof Draw the vertical line through Q and the horizontal line through P, meeting
at the point R. Let S and T be the feet of the perpendiculars to X, from P and R
respectively. Then
PR= ST,
because opposite sides of a rectangle have the same length. And
ST= lx2 - Xii,

by definition of a coordinate system on a line. Therefore
PR = lx2 - Xii·
For the same sort of reason,
RQ= UV= IY2 - Y1I·

But t:,.PQR is a right triangle, with its right angle at R. Therefore, by the Pythagorean
theorem,
PQ2 = PR2 + RQ2
= /x2 - X1l2 + IY2 - Y1l2.

Therefore
PQ= .J1x2 - X1l2 2
+ IY2 - Yil -
This is not quite the formula given in Theorem I, because it uses absolute-value
signs instead of parentheses. But this makes no difference, because
/X2 - X112= (X2 - X1)2,

and
2 2
IY2 - Y1l = (Y2 - Y1) •
(Why? We need a theorem from Section 1.4.)
In the previous figures, we have shown the x-axis going positively from left to
right, and the y-axis going positively from bottom to top. Logically speaking, we
could equally well have put the axes in any of a number of other positions:
But fhe axes are usually drawn as shown on the right above. This figure shows the
minimum that must be indicated when graph paper is used for drawing pictures of
coordinate systems. That is, the axes must be labeled, and the number scale must be
shown on each axis, by indicating the coordinate of at least one point.
The two axes separate the plane into four parts, called quadrants. The quadrants
are numbered I, II, III, IV. That is, the first quadrant is the set of all points (x, y)
of the plane for which x > 0 and y > O; the second quadrant is the set of all points
(x, y) for which x < 0 and y > O; and so on.
y y
1
II I
--+---X
-'--
III IV
We have used the letters X and Yin order to have convenient names for the x
and y-axes. The axes are more commonly labeled as on the right above.
PROBLEM SET 2.2
Calculate the distances between the following pairs of points. Then plot the points and
check the plausibility of your answers.
1. a) (1, 2) and (3, 4) b) ( -2, -4) and (4, 2)

c) (7, -5) and (5, -7) d) (1, 0) and (0, 1 )
2. a) (3, 7) and (-3, -7) b ) ( 1, 3 ) and (--2, 7)

3. Obviously, PQ = QP for every pair of points P, Q; the distance between two points
does not depend on the order in which the points are named. Therefore any correct
distance formula has the property that when we interchange the two points, the formula
gives the same answer. Check algebraically that our distance formula has this property.
4. Find out whether or not the points ( - 1 0 , 10), (14, 3), and (38, -4) are vertices of an
isosceles triangle. Is the triangle equilateral?
5. Find all points (x, y) such that (0, 0), (2, 2), and (x, y) are the vertices of an equilateral
triangle.
6. Find out whether the points ( -2, 3), (0, 1), and (3, 4) are the vertices of a right triangle.
Then plot the points and check for plausibility. (This problem can and should be
worked by the use of distances alone. The use of slopes is not necessary.)
7. Find the coordinates of the point which is equidistant from (0, 0), (1, 2), and (3, -1).
Find the radius of the circle which passes through the three given points.
8. What point on the y-axis, if any, is equidistant from ( -1, -2) and (2, 3)?
9. a) Give a formula for the perpendicular distance between (x, y) and the x-axis.
b) Give a formula for the perpendicular distance between (x, y) and the y-axis.
10. Find out whether the points ( -1, -1), (0, 1), and (2, 5) are collinear. Then plot and
check for plausibility. (The remarks following Problem 6 also apply here.)
11. Find a point on the x-axis which is collinear with the points (1, 2) and (0, 3). (The
remarks following Problem 6 also apply here.)
2.3 The Graph of a Condition. Equations for Circles 21
The following problems are a review of the main theorems of elementary geometry that
we have been using so far.
12. Show that an exterior angle of a triangle is greater than either of its remote interior
angles.
l><i_.
A
6
B C D
. .,
B C D
That is, show that in the left-hand figure we have LACD > LA. The proof is
based on the figure on the right. [Query: If you know that LACD > LA, how do
you infer that LACD > LB?]
13. Show that there is only one perpendicular to a given line, from a given external point.
That is, show that the left-hand figure below is impossible for A � B. (We needed this
in order to explain what was meant by the x-coordinate of a point; A must be determined
when P is known.)
.. A
A B
•L
a
a
14. Write the proof of the Pythagorean theorem suggested by the figu,re on the right above.
15. The proof of Theorem 1 of this section was incomplete: it discussed only the most
significant case and neglected to mention two other cases. The point is that if P and Q
lie on the same horizontal line, or the same vertical line, then there is no such thing as
t-,PQR, and so the Pythagorean theorem cannot be used.
Show that the distance formula holds in the case x1 = x2, and also in the case
Yi =
Y2·
2.3 THE GRAPH OF A CONDITION. EQUATIONS FOR CIRCLES
Given a point Pand a positive number r, the circle with center Pand radius r is the
set of all points of the plane whose distance from Pis equal to r. That is, a point Q
is on the circle if PQ = r.
This is the first and simplest example of the idea of the graph of a condition. If
we state a condition which every point of the plane either satisfies or doesn't satisfy,
then the graph of the condition is the set of all points of the plane that satisfy it. (Thus
the graph is simply the solution set of an open sentence; we use the word graph
when the solution set is a set of points.) In this language, we say that the graph of the
condition OQ = r is the circle with center at the origin and radius r.
-r
The interior of the circle with center P and radius r (r > 0) is the set of all points
Q such that PQ < r. Thus the interior is the graph of the inequality PQ < r. We
indicate such graphs in figures by means of shading or cross-hatching.
Sometimes the condition takes the form of an algebraic equation. For example,
if Q +--+ (x, y), then the distance formula tells us that
OQ = Jx2 + y2.
Therefore the condition
OQ = r (1)
can be written in the equivalent form
Jx2 + y2 = r, (2)
or
x2 + y2 =
r2. (3)
The point (x, y) is on the circle if and only if x and y satisfy (2). And
Jx2 + y2 r <=> x2 + y2 r2
=
(r > 0). =
Thus the circle with center -at the origin and radius 2 is the graph of
J x2 + y2 =
2 <=> x2 + y2 = 4;
and the interior of this circle is the graph of
J x2 + y2 < 2 <=> x2 + y2 < 4.
Similarly, the first quadrant is the graph of the condition x > 0 and y > 0.
x>O
y>O
x>O
IV y<
O
The fourth quadrant is the graph of the condition x > 0 and y < 0.
We found that the circle with center at the origin and radius r is the graph of the
equation
x
2
+ y2 =
2
r .
Consider, more generally, the circle with center at Q =

(a, b) and radius r. By
definition, the circle is
{PI QP = }
r .
P(x, y)
r,,....
,,
.,.,....-
Q(a, b)
Using the distance formula to express QP algebraically, we get

QP = r
. <=> )( x - a)2 + (y - b)2 = r
<=> (x - a)2 + (y - b)2 = r2•

Thus:
Theorem 1. The circle with center at (a, b) and radius r is the graph of the equation
(x - a)2 + (y - b)2 =
r2•
An equation written in the above form is easy to interpret. For example, given
(x + 2)2 + (y - 5)2 = 4,
we see by Theorem 1 what the graph is. On the other hand, if such an equation
is "simplified" algebraically, it may look like this:
2
x + y2 + 4x - lOy + 25 = 0.
5
4
3
2
To find out what the graph is, we first "unsimplify" by completing the square:
x2+ 4x+ y2 - IOy = -25

<:::> x2+ 4x+ 4+ y2 - lOy+ 25= -25+ 4+ 25
2
<=> (x+ 2)2+ (y - 5) = 4.
In the general case, for equations of the form
x2+ y2+ Dx+ Ey+ F = 0,
there are three possibilities for the graph. In some cases, the graph is a circle. But
x2+ y2 = 0
is also an equation of this form, and its graph is not a circle, but a single point, namely
the origin. And the equation
x2+ y2+ 1 = 0
is never satisfied, for any x and y. Its graph is therefore the empty set { }.
By completing the square, starting with the general form, we shall show that these
three possibilities-a circle, a point, and the empty set-are in fact the only ones:
x2+ y2 + Dx+ Ey+ F= 0
2 (D)2 (E)2 (D)2

x + Dx+ 2 + y2+ Ey+ z = - F+ 2
(E)2
+ 2
( D)2 ( E)2 D2 + E2 - 4F
<:::> x+-+ y+-= .
2 2 4
If the fraction on the right, in the last equation, is positive, then it is= r2 for some
positive number r, and so the graph is the circle with center at (-D/2, - E/2) and
radius r. If the fraction on the right is = 0, then the equation takes the form
and the graph contains only the point ( -

D/2 ,
-
E/2) . Finally, if the fraction on the
right is negative, then the equation is never satisfied, for any x and y, and so the graph
is the empty set { }.
To sum up:
Theorem 2. The graph of an equation of the form
x2 + y2 + Dx + Ey + F = 0
is a circle, a point, or the empty set.
PROBLEM SET 2.3
Problems 1 through 6.
In the illustration below six figures are drawn. For each of these figures, state a condition
which has the given figure as its graph. In the figure, the arrowheads merely indicate that the
line is supposed to go infinitely far in the indicated direction. Thus (1) and (2) are entire
lines;(3) is a ray, going infinitely far on the right, but stopping at the point (0, 4) on the left;
and (6) is a segment, with endpoints (I,'-3) and (4, -3).
y (1) (4)
Problems 7 through 10.

Follow the same directions as in the previous problems for the illustration below.
y
Sketch the graphs of the following conditions, using cross-hatching to indicate regions.
11. x2 + y2 = 1 12. x2 + y2 < 1 13. x2 + y2 > 1
14. x = 2 and 0 � y � 2 15. x = - 3 16. y = 2 17. y = x/lxl, x >= 0
18. x2 + y2 � 1 19. x y � 0 20. x > 0, y = 3
21. a) Sketch the graph of the condition "(x, y) is equidistant from the points (0, 1) and
(1, 0)."
b) Write this condition in the simplest possible algebraic form.
22. Write the simplest equation that you can get, for the set of all points that are equidistant
from (1, 2) and (0, 3). What sort of a figure is this graph? How is it related to the
segment from (1, 2) to (0, 3)?
23. Same problem, for the set of all points that are equidistant from (1, 2) and (2, 2).
24. Same problem, for the set of all points that are equidistant from P i = (xi, Yi) and
P2 = (x2, J2).
*25. Describe and sketch the graph of the equation
v (x - 1)2 + (y - 2)2 + v (x - 4)2 + (y - 7)2 = v34.

[Hint: If you do a lot of algebra, you will probably get the wrong answer; the graph is
notan ellipse.]
*26. Describe the graph of the equation
v x2 + (y - 1)2 + v (x - 2)2 + y2 = 1.
[The same hint as for Problem 25 applies here.]
27. Draw the graph of the equation
x3y + y3x - xy = 0.
28. Draw the graph of the equation
x2y + xy2 - xy = 0.
29. Consider the set of all points that are twice as far from the origin as from the point
(3, 0). Find an equation for this graph, and sketch.
2.4 EQUATIONS OF LINES. SLOPES,

PARALLELISM. AND PERPENDICULARITY
Every line is the graph of an equation of the form
Ax+ By+ C = 0,
where A and B are not both = 0. The proof is as follows.

Every line is the perpendicular bisector of some segment. If Lis the perpendicular
bisector of the segment from Q +----'> (ai, bi) to R +----'> (a2 , b ), then
2
L = {P j PQ = PR} .
2.4 Equations of Lines. Slopes, Parallelism, and Perpendicularity 27
y
L
(Remember your geometry.) Therefore Lis the graph of the equation

2 2 2 2
.J(x - a 1) + (y - b1) .J(x - a2) + (y - b2)
=
x2 - 2a1x + ai + y2 - 2b1y + b� 2
<::?- x - 2a2x + a: + y2 - 2b2y + b�
=
<::?- 2(a2 - a 1)x + 2(b2 - b1)y + a i + bi - a � - b� = 0.
This has the desired form
Ax + By + C = 0,
with
and
The numbers A and B cannot both be 0, because a2 - a1 and b2 - b1 cannot both

=
be =0; the number pairs (au b1) and (a2 , b2) are the coordinates of Q and R, and
Q =;tf R, because Q and R are the endpoints of a segment.
An equation of this type is called a linear equation in x and y. Thus we have
Theorem 1. Every line is the graph of a linear equation in x and y.
y
L
If the line is not vertical, we can say more. In this case, the perpendicular segment
from Q to R is not horizontal, and this means that b2 - b1 =;tf 0. Therefore B =;tf 0,
and we can divide by B and solve for y. This gives
which has the form

y = mx + k,
where
A
m =
B b2 - bi
In the figure, the label k on the y-axis is correct, because k is they-coordinate of

the point where L crosses the y-axis (m · 0 + k k). The number k is called the
=
y-intercept of the line.

The number m also has a geometric meaning, as we shall soon see. If Pi�
(xi, Yi) and P2 � (x2, y2) are any two points of a nonvertical line, then the slope of
the segment from P1 to P2 is defined to be the fraction
The denominator x2 - Xi is marked 6.x in the figures below; it is pronounced

"deltax," and stands for the difference inx. Similarly, Yz - Yi is marked 6.y, which
stands for the difference in y. Here 6.y and 6.x are not necessarily distances in the
sense of elementary geometry, because they may be negative.
y
y
P1 P1
P, � t:.x O
" <
y O y
" < O� t:.x>O

P,
<
Slope = Y2-Y1 = t:.y

x2-x1 t:.x
We shall show that all segments of the same line have the same slope, and that
this slope is the number m which appears in the equationy = mx + b.
Given two points Pi� (x1,y1) and P2 �> (x2, Yz), on the line
y = mx + k,
then
m
Yz = x2 + k and
Therefore
and
Y2 - Y1 m.
=
X2 - X1
(In this calculation we do not care whether y2 - Yi and x2 - x1 are positive or

negative. The algebra takes care of all cases at once.)
The number m is called the slope of the line. And we have proved the following
theorem:
Theorem 2. The graph of the equation
y=mx+k
is the nonvertical line with slope m and y-intercept k. All segments of this line have
slope=m.
The equation given in this theorem is called the slope-intercept form of the equa
tion of the line.
y
y=x
A line can be described by many different equations. For example, the bisector
of the first and third quadrants above is the graph of each of the following equations:
y=x
<=> x-y=O
<=> 3x - 3y=0
2
<=> (x- y) =0
<=> (x - y)177=0,
and so on. But there is only one equation, in the slope-intercept form, for every
nonvertical line, because when the line is named, its slope and its y-intercept are
determined.
Often a line will be described by its slope m and the coordinates x1, Yi. of one of
its points. We can then find an equation for it in the following way. If (x, y) is any
other point of the line, then
because all segments of the line have the same slope m. Therefore
Y - Y1=m(x - )
X1 .
The graph of this equation contains (x1, y1), because 0=m · 0. And the graph is a
line with slope=m, because the equation has the form
y=mx + (y1 -mx1)=mx + k.

Thus:
Theorem 3. The graph of the equation y - y1 = m(x - x1) is the line which has
slope = m and contains the point (x1, Yi).
For example, the graph ofthe equationy -3 =-2(x + l)is the line which has
slope = - 2 and passes through the point ( -1, 3). Solving for y, we get the slope
intercept form y = - 2x + 1.
Theorem 4. Two nonvertical lines are parallel if and only if they have the same slope.
Given:
we need to prove two things: (1)If the slopes are the same, and the lines are different,
then the lines are parallel. (2) Ifthe slopes are different, then the lines are not parallel.
1) If m1 = m2, then k1 -¥- k2, because the lines are different. Therefore the lines are
parallel, because the two equations are inconsistent: they take the form
Since k1 -¥- k2, these equations have no common solution.

2) If m1 -¥- m2, the lines cannot be parallel, because the equations always have a
common solution. By subtraction we get
and we now find they-coordinate of the point of intersection by substituting in either

of the original equations.
Theorem 5. If two nonvertical lines are perpendicular, then their slopes are negative
reciprocals of each other.
y
Proof Given Li with slope mi and L2 with slope m2, intersecting at right angles at T.
Let (ai, bi) and (a2, b2) be points of L2 which are equidistant from T. Then Li is the
perpendicular bisector of the segment between these points. As we found earlier, the
slope of Li is
But we can calculate the slope m2 of L2 by the slope formula, using the points (ai, bi)
and (a2, b2). This gives
Obviously m2 = -1/m1.
This also works the other way around:
Theorem 6. Given two lines L1, L2, with slopes m1, m2. If
then Li and L2 are perpendicular.
\
\
\
\
\
\
\
\
\
\
\
\ Lz
\
\
L?.
Proof First we observe that the lines cannot be parallel, because m2 cannot be = m1.
(Why?) Let T be the point where they intersect. Let L� be the line through T,
perpendicular to L1. Then L� has slope m2 = - l/m1. But through a given point
there is only one line with a given slope. (Why?) Therefore L� is L2, and L2 is
perpendicular to L1.
Probably you have seen these theorems proved before, in different ways. The
treatment given above is intended to avoid repetitions and also to furnish some
practice in drawing geometric conclusions by algebraic methods.
PROBLEM SET 2.4
Find point-slope equations, and slope-intercept equations, for the Jines containing
the following pairs of points.
1. (-3, 2), (2, 1) 2. (3, -4), (1, 2) 3. (I, 0), (3, 3) 4. (-1, I), (2, -2)
5. Find an equation for the tangent to the graph of
x2 + y2 = 25,
at the point (3, 4).
6. Given thatP1 � (x1, y1) lies on the circle
x2 + y2 =
a2,
with
LetP2 be the point where the tangent atP1 crosses the x-axis. Find the distanceP1P•2
[Warning: Geometric distances are never negative.]
-a
7. Find the points P on the circle x2 + y2 2 so that the tangent line to the circle atP
=
passes through the point (2, 0). (You may use the fact that, at any pointPon a circle,
the tangent and the radius are perpendicular.)
8. Sketch the graph of the equation
x2 + y2 + I + 2xy + 2x + 2y = 0.
x2 + 4y2 + I - 4xy + 2x - 4y = 0.
10. Sketch the graphs of the following equations.
a) y = lxl b) y = -l2xl c) y = I - Ix - II
For this problem we offer a hint which applies equally well to a very large number of other
problems. If you didn't know the meaning of the symbol lxl, you would have no hope of
sketching the graph. This suggests that you should recall the definition of lxl, and use it.
2.5 Graphs of Inequalities. And, Or, and If ... Then 33
11. Sketch the graph of

!xi + lyl = 1.
[Hint: As a first step, sketch the portion of the graph that lies in the first quadrant.]
y = x +!xi + 1.
!xi - lyl = 1.
v(x -1)2 + (y - 3)2 + v(x - 4)2 + (y - 2)2 = v10.
15. Let C be the set of all points P such that the segment from ( - 1, 0) to Pis perpendicular
to the segment from P to (2, 1). What sort of figure is C? Sketch. (In answering this
one you should bear in mind that the endpoints of a segment are always different. That
is, there is no such thing as the segment from P to P.)
*16. Let A = ( -2, 0), let B = (2, 0), and let G be the set of all points P such that LAPE
is an angle of 60°. What sort of figure is G? Sketch. (You will have to remember and
use some plane geometry, to do this one. If you have suitable drawing instruments, you
ought to be able to do a good sketch.)
2.5 GRAPHS OF INEQUALITIES. AND, OR, AND IF ... THEN
We have found that the graph of the equation
(x - 1)2 + (y - 1)2 = 1
is the circle with center at A = (1, 1) and radius 1. The interior of the circle is the
graph of the condition AP < 1. This is the region marked R1 in the figure. It is the
graph of the inequality
(x - 1)2 + (y - 1)2 < I.
Similarly, the exterior R is the graph of the condition AP > 1, so that

2
R2 = {(x,y) I (x - 1)2 + (y - 1)2 > l }.

y
The graph of the equation
y = l - x
is a line L. The points lying above L form a set H1, called a halfplane. Evidently H1
is the graph of the inequality
y > 1 - x.
The points lying below L form a half-plane H2; and H2 is the graph of the inequality
y< 1 - x.
Consider now the double inequality
t< x < %.
The graph is an infinite vertical strip R1, lying between the lines
.l.. Ji
X -
- 2 and X -
- 2·
Similarly, the graph of
t<y< 1
is an infinite horizontal strip, as shown on the left below.
I I
.I I
y I I
I I
1 ___ ! __ �---- I
-�- --� � --- -
1 I
Consider next the condition
or t<y<l.
The graph of the condition using or is an infinite cross-shaped region. This region R'
is the union of an infinite vertical strip R1 and an infinite horizontal strip R2; it contains
all points of the plane that belong to R1 or to R2.
(In mathematics, when we say that one condition holds or another condition
holds, we allow the possibility that both conditions hold. If we mean " ... but not
both," we have to say so.)
Similarly, the graphs of the conditions
y>x, y > -x
are two half-planes H1 and H2• They are respectively to the left of the line y x =
and to the right of the line y = -x, as shown in the figure on the left below. The graph
of the condition
y> x and y > -x
is the intersection of these two half-planes. This is the interior R1 of LAOB.
y
y
y= - x y=x
The graph of the condition
y >x or y > -x
is the union of the two half-planes.
y y
S1
R
I- I I
R '
I R I
'
I I
�-
I I x
0 l
1 2 2
R
2 2
R
S2
Let us now see what sort of graph we get when we combine two inequalities by
"if ... then." Consider the condition
i<x i<y<l.

This says that
if t<x< }, then i<y<l.
Let R be the graph. We assert that R looks like the drawing on the right above. That
is, R contains all points that do not lie in either of the two vertical strips marked
S1 and S2. The reason is as follows:
1) If (x, y) is a point of R, and t < x< !, then we must have t <y< I. There
fore the part of R that lies between the lines x = t and x = i must be the interior of a
rectangle, as indicated by the dashed lines in the figure.
2) On the other hand, if xis not between t and i, then the condition for the graph
imposes no restriction on y at all. Therefore R contains all points to the left of the
line x = t and all points to the right of the line x = l R also contains these two
vertical lines, for the same reason.
The reasoning in (2) may seem a little tricky, but may be clarified by an analogy
from everyday life. The law in most places requires that if a person has seriously
defective vision, then he must wear corrective glasses when driving a car. A person
with normal vision automatically obeys this law; its restrictive clause does not apply
to him. In the same way, the "law"
i<x<�- => t<y<l

imposes a restriction only on points (x, y) for which l <x< -�; all other points
automatically obey the "law," because its restrictive clause does not apply to them.
Thus the "law" holds under each of the following three conditions:
1) t <x< ! and t<y< 1,

2) x � t,
3) x � !.
The graph of (1) is the rectangular region in the middle of the figure; the graph of (2)
is the infinite region to the left of the line x = t; and the graph of (3) is the infinite
region to the right of the line x = i.
y
x<O x>O
Yes Yes
y>O y>O
x<O x>O
Yes No
y<O y<O
Similarly, the graph of the condition
x>O =>- y�O

contains all of the plane except for the fourth quadrant. It is only in the fourth
quadrant that x> 0 holds and y� 0 does not hold; and the possibility x> 0,
y < 0 �s the only possibility that is ruled out by the condition x> 0 =>- y� 0.
In each of the following cases, the shaded region is the graph of the condition
appearing below it.
y y y
x;;;o
y y y
r- --
_ +
1 -
� -t-��.,.x
-1
There is no need to use graph paper in the following problem set. Reasonably
neat freehand sketches, with cross-hatching used to indicate regions, are sufficient.
PROBLEM SET 2.5

Sketch the graphs of the following conditions:
1. ! < x < t 2. Ix - 1 1 < l

3. } g < y < ii 4. ly - 21 < 110
5. Ix - 11 < t and ly - 21 < /0 6. lx - lj < t =>- ly - 21 < /0

7. !y - 2i < /0 =>- fx - 11 (x - 1)2+ y2 � 1

13. (x - 1)2+ y2 � 1 => x2+ y2 � 1
14. x2+ y2 � 4 and (x+ 1)2+ y2 � 1
15. x2+ y2 � 4 or (x+ 1)2+ y2 � 1
16. x2+ y2 � 4 => x2+ y2 � 1 17. x2+ y2 � 1 => x2+ y2 � 4
18. x2 + y2 � 1 and y ;;;; x 19. x2+ y2 � 1 and x � lyl
20. x+ y � 1 21. x - y � 1
22. -x+ y � 1 23. -x - y � 1
24. Ix!+ lyl � 1 25. !xi - ly l � 1
26. Ix+ yl � 1 21. Ix - yl � 1
28. x =
t => y = 1 29. Ix - 21 <Yo => ly - 11 < t
30. Ix - 31 < t => l y - 21 < t
31. Supposeyou know that (a)P => Qand (b)Pis false. What, if anything, can you infer
about Q?
32. Supposeyou know that (a)P => Qand (b) Qis true. What, if anything, can you infer
aboutP?
33. Supposeyou know that (a)P => Qand (b) Qis false. What, if anything, can you infer
aboutP?
34. Supposeyou know thatP => Q. Which of thefollowing arepossible?
a) P is trueand Qis true. b) Pis trueand Q is false.
c) Pis false and Qis true. d) Pis falseand Qis false.
35. Suppose you know that P <=> Q. Which of the combinations (a), (b), (c), and (d)
in Problem 34 arepossible?
2.6 PARABOLAS
The distance from a point to a line is the length of the perpendicular from the point
to the line . Given a point F and a line Dnot containing F, the parabola with focus F
and directrix Dis the set of all points of the plane that are equidistant from Fand D.
The parabola is the graph of the condition
FP =MP,
where Mis the foot of the perpendicular from P to D. The perpendicular line to D
2.6 Parabolas 39
through F is called the axis of the parabola. The point where the axis crosses the
parabola is called the vertex. (There is only one such point, because any such point
is midway between the focus and the directrix.)
The first step in the study of parabolas is to get equations for them.
��::::-+-:��+l��-. x
D..----------- _______ n_ ____
Mly=-'E.
2
In setting up our axes, we take the vertex as the origin, and the x-axis parallel to
p
the directrix, in such a way that Dis below the x-axis and the focus is above it. The
number is the distance from the focus to the directrix. Now let P � (x, y) be a
point of the parabola. Then
FP = Jx2 + (y �r _
and
Therefore
FP = MP
Jx2 + (y - �r = J (y + �r
x2 + (y - �)2 = (y + �r
x2 + Y2 py + 2 y2 + py + 2
_ p__
4
= p__
4
x2 = 2py
r
. - 21p ..
-
.,
x
This has the form
a = l/2p,
p 2a1
where and
=-.
Thus we have proved the following theorem:
y= ax2
is a parabola, with focus at (0, 1/4a) and directrix
y=-1/4a.
y
x
I
______ _ti_ ____
x
M
-4li
FP=MP <=> y=ax2.
If a parabola is situated like this, relative to the axes, then the parabola is said
to be in standard position. The use of standard position simplifies the equation
considerably. For example, if Fis the point (2, -1) and Dis the line y= 3, then the
parabola is the graph of the equations
FP=MP
<=> .Jex - 2)2 + (y + 1)2 = .J(y - 3)2
<=> x2 - 4x + 4 + y2 + 2y + 1 = y2 - 6y + 9
<=> x2 - 4x - 4 =-8y
<=> y = -tx2 + tx + t.
It is not hard to check, in general, that if the directrix is horizontal, then the equation
always takes the form
y = Ax2 + Bx + C, A :;i: 0.
And if the directrix is vertical, we get
x= Ay2 + By + C, A ,t: 0.
If the directrix is neither horizontal nor vertical, then the equation involves, in
general, terms in x2, y2, and xy, as well as linear terms and a constant. In this case it
is hard to derive the equation when the focus and directrix are given; and it is even
harder, when the equation is given, to see that the graph is a parabola. This case will
be discussed in Chapter 8.
For a long time to come, however, we shall deal only with the simplest case, in
which the directrix is horizontal.
Parabolas arise in a variety of contexts which appear at first to be unrelated.
Following are a few.
2.6 Parabolas 41
1) If a right circular cone is cut by a plane parallel to an element of the cone, the
resulting curve is a parabola. This was the viewpoint from which the Greeks studied
parabolas; and it is for this reason that a parabola is one of the conic sections. There
are other kinds of conic sections, obtained by slicing cones by planes in various
positions.
y
--------------- D
2) If a theoretical projectile is fired from the surface of the earth, in any direction
other than straight upward, the path that it moves along is a portion of a parabola.
In the figure on the right above, the x-axis lies along the surface of the earth, the
y-axis is vertical, L Cl.. represents the angle at which the gun is aimed, and Tis the
point where the projectile hits the ground. We say, "a theoretical projectile," because
to get this result you must assume both that the weight of the projectile is independent
of its altitude and that the air makes no resistance. These assumptions are false, but
they are good approximations to the truth, if the projectile is not going very fast or
very high. For high-speed, long-range projectiles, both assumptions are quite
unrealistic, and the situation is more complicated.
3) If you rotate a parabola around its axis, you get a surface which is called a parab
oloid of recolution. The mirror in a reflecting telescope is a paraboloid of revolution,
as is the reflector in an automobile headlight. The reason is that if a ray of light
travels along a line parallel to the axis, and is reflected in the usual way, it always hits
the focus. And conversely, if a ray of light starts at the focus, hits the surface and is
reflected, it always continues along a line parallel to the axis. The first of these prin
ciples is used in telescopes, and the second in headlights.
i i
I I
I I
l"1F
4) Suppose that you fire a "theoretical projectile'" vertically upward. It moves up a

vertical line, for a certain distance lz, and then comes down again along the same line.
Thus the path of motion is simply a segment. Suppose now that we label our
horizontal axis as the t-axis; we measure time starting at the moment of firing; and
we plot, for each time t, the height of the projectile at time t.
h ----- -:..:-...--�
The resulting graph is a portion of a parabola. In the figure above, a is the

time at which the projectile hits the ground. Note that the graph that we have been
discussing is not all of the parabola: one minute before firing, the projectile was in
the gun; it was not underground. And at t = a the projectile hits the ground, and
the motion stops. Therefore, in the figure there is a solid arc, which indicates the
portion of the parabola that is related to the physical problem; the irrelevant part
of the curve is indicated by dashed arcs to the left and right.
This example indicates that geometric ideas come up in physics in unexpected
ways; the uses of geometry are not limited to the study of figures in space.
PROBLEM SET 2.6
1. Take a full-size sheet of graph paper; draw the y-axis in the center; and draw the x-axis
near the bottom of the paper. Then choose the largest uniform scule that you can,
on the axes, in such a way that x ranges from - 2 to 2 and y ranges from -! to 4.
Now sketch the graph of y =x2• First plot the points corresponding to the following
values of x:
x=O, x=0.1, x=0.2, ... , x=0.9, x=l,
x= 1.2, x = 1.4, ... ' x = 1.8, x= 2.
Then draw the curve, freehand, as smoothly as you can. If this is done carefully, it will
really look as if FP = MP at every point of the curve.
One of the reasons for doing this is that it will give you an accurate idea of what a
parabola really looks like.
2. Show that
0 < x1 < x2 => xi < x� .
y=ax2 , a>O
Xz
Y1 <yz
2.7 Tangents 43
This means that the right-hand half of a parabola in standard position rises as we go
from left to right along the curve.
3. Show that
What does this tell us about parabolas in standard position?

4. Find the focus and the directrix of the graph of the equation y = x2•
5. Same problem, for the equation y = 3x2•
6. Same problem, for the equation y = tx2•
7. Show that the graph of the equation y = x2 + 1 is a parabola. To show this, you must
find its focus F and directrix D. You can then check by deriving the equation of the
parabola with focus F and directrix D.
8. Show that the graph of the equation y = (x - 2)2 is a parabola.
9. Same problem, for the equation y = (x - 2)2 + 1.
10. Show that the graph of y = x2 - 2x is a parabola.
11. Show that the graph of the equation y = (x + 1)2 is a parabola. (Find the focus F
and directrix D.)
12. Same problem, for y = (x + 1)2 - I.
13. Same problem, for y = (2x + 1)2•
2.7 TANGENTS
In geometry, tangent lines to circles are defined as follows.
Definition. A tangent to a circle is a line (in the same plane) which intersects the
circle in one and only one point. This point is called the point of contact.
It is then shown that a line is tangent to the circle if and only if the line is per
pendicular to the radius drawn to the point of contact. (In fact, the latter condition
is probably the one that you used to find the slopes of tangent lines to circles, in
Problem Set 2.4.)
y
L
z2 y2
�+b2 = 1.
Tangency can be defined in the same way for an ellipse. Ellipses will be studied
in Chapter 8. Meanwhile we observe that an ellipse is an oval curve, of the sort shown
in the right-hand figure above, and the tangents to it are the lines that intersect
it in one and only one point.
But for some curves, tangents cannot be described by the definition that we use
for circles. Consider, for example, a parabola, as shown in the figure below. The
tangent to the parabola, at the point (x1, y1), intersects the curve only at (x1, Ji). But
the vertical line through (x1, y1) has the same property; and the vertical line is not
a tangent.
We may try to get around this trouble by providing that the tangent line must
only touch the curve, without crossing it. But for many curves, this won't work
either. The graph of y = x3 is shown below. The tangent to this curve at the origin
turns out to be the x-axis; and the x-axis crosses the curve, at the point of tangency.
Jn other cases a tangent line may cross a curve in many points.
y
y
3
y =x
The geometric idea of tangency is obvious in all these cases. But the above
examples indicate that the mathematical definition that works for circles does not
work in general. To find the tangents to other curves, we need a better definition.
Consider first the graph of y = x2, and the fixed point (I, 1) at which we want to
find the slope of the tangent. For every other point (x, x2 ) of the curve, let Lx be the
2.7 Tangents 45
secant line through (1, 1) and (x, x2). Then the slope of L., is
x2
- 1
m,,, =
---
(x -:F- 1).
x - 1
Here the restriction x -:F- 1 reflects the geometric fact that it takes two different
points to determine a line. It also refers to the algebraic fact that fractions with
denominator 0 have no meaning.
y
y
Lx
!/'�m.
I
I
I
I
I
I
I
·X x
2
We shall now draw the graph of y = mx (x -:F- 1). We have
y = mx = x+ 1 (x -:F- 1).
The graph is a line from which one point has been deleted. For x = 1, there is no
such thing as "the secant line through (1, 1) and (1, l2"
) ; and for x = 1, there is no
such thing as the "fraction" m1 = 0/0. But this causes no trouble, because it is easy
to see that mx is very close to 2 when xis very close to 1. We express this by writing
lim mx = 2.
x->1
This is read: "The limit of mx, as xapproaches 1, is equal to 2." Later we shall give a
general definition of the idea of a limit. But in the present case, the meaning of the
limit is clear, and so we use it in the definition of the tangent to the parabola.
Definition. The tangent to the graph of
y = ax2 + bx + c,
at a point (x0, y0) of the graph, is the line through (x0, y0) with slope
Sx0 = lim 111.,,

x�x0
where mx is the slope of the secant line passing through the points (x0, y0) and
(x, ax2 + bx + c) (x -:F- x0).
Even in the general case, the slope is easy to calculate on the basis of this defini
tion. We have
ax2 + bx + c - ax� - bx0 - c

m.,,= (x � x0)
X - Xo
a(x2 - x�) + b(x - x0)

= (x � x0)
X - Xo
= a(x + x0) + b (x � x0)
= ax + (ax0 + b). (x � x0)
The graph of y = m.,, is a line with one point missing. The line from which the point
is missing is shown on the left below. The graph of y = m.,, is on the right.
y y
I
I
I
----- �/ I
I
I
I
I
I
��-4-��--+�--- x
1·ro
I
y=ax+(ax0+b) y=m.;
Here again, the limit of m.,, is simply they-coordinate of the point that is missing from
the graph. Thus we have:
Theorem 1. Let (x, y) be a point of the graph of
y = ax2 +bx + c.
Then the slope of the tangent to the graph, at (x, y), is
S.,, = 2ax +b.
For some curves, there is no tangent. Consider, for example, the graph of
y = lxl, at the point (0, 0) . For each x � 0,
.
lxl - IOI lxl

m = =-
"' x 0 x -
Thus:
m.,, = 1 for x > 0, mx = - 1 for x < 0.
(Remember the definition of Ix!.) Therefore the graph of y = m,, looks like the
2.7 Tangents 47
y k?
y=l,x>O
k?
y = -1,x<O
k?
y=fxf
drawing on the right above. For this graph there is no one number that y is close to,
whenever x is close to 0. Therefore, there is no such thing as
(?) Jim mx (?);

x->O
for every number k, the statement
(?) Jim mx = k (?)

x->O
is false.
Geometrically it is obvious that the origin is the only point of the curve at which
things go wrong; at every other point (x, lxl), the curve has a tangent; the slope of the
tangent is I for x > 0 and -1 for x < 0.
PROBLEM SET 2.7
1. You already have a carefully drawn graph of the equation J = x2• At each point (x, J)
of the graph, the slope of the tangent ought to be 2x. Check this graphically by drawing
lines of the proper slope at the points where x = 0.2, 0.4, 0.6, 0.8, and 1.
2. Given J = x2 - 4x + 4. Find the slopes of the tangents at the p oints where x = -
2,
x = 0, and x = 2 and sketch, showing all three of these tangents.
,
3. Same problem, for J = x2 + x + I, using the points where x = 0, x = L and x = 1.
4. By completing the square, show that
J = ax2 +bx + c
can be expressed in the form
y = a(x - A)2 + B.
For a > 0, this means that the point where x = A is the lowest point on the curve.
Find the slope of the tangent at this point.
5. a) Given the graph of

y = ax2
and a point (x0, Jo) of the curve. Show that the tangent at (x0, Jo) is the only non
vertical line which passes through (x0, Jo) and has no other point in common with the
parabola. That is, show t hat, if the graph of
(y - Yo) = m (x - x0)
intersects the parabola only at (x0, y0), then
m = 2ax0•
b) Prove the corresponding theorem for the graph of
y = ax2 + bx + c.
6. a) Get a plausible answer for the slope of the tangent to the graph of y = x3, at the
point (I, 1). Sketch the graph of y = mx, explain what sort of graph it is, and explain
as well as you can why your value for the slope is plausible.
b) Do the same for y = x3, at an arbitrary point (x0, xg).
7. a) Show that, if m < 0, then the line through the origin with slope m meets the graph
of y = x3 at precisely one point.
b) Show that, if m > 0, then the line through the origin with slope m meets the graph
of y = x3 at precisely three points.
8. Sketch the graph of
y = x lxl,
and describe this curve in terms of types of curve that we already know about. At
which points does this graph have a tangent? What is S2? What is S_2? Give, if
possible, a general formula for Sx. Is there such a thing as S0?
9. Consider the graph of
y = x3 - 4x.
Where does this cross the x-axis? At which points is the tangent horizontal? What is
the slope of the tangent at (0, 0)? For what values of x is y > 0? For what values
of x is y < 0? Use this information to draw a reasonable sketch of the graph, plotting
onlyfive points.
10. Carry out the steps of Problem 9 for the equation
y = 2x3 - 6x.
11. Show that every parabola has the reflecting property. In the figure, Tis the tangent at P,
and you need to show that Cl = /3. The key to the proof is that the quadrilateral FPRQ
is a rhombus. (That is, all four sides have the same length.)
y
2.8 A Shorthand for Sums 49
2.8 A SHORTHAND FOR SUMS
An arithmetic series is a sum of the form
Sn =a + (a + d) + (a + 2d) + · · · + (a + [n - l]d).
A geometric series is a sum of the form
Tn = a + ar + ar2 + ar3 + · · · + ar"-1•
There is a shorthand for sums, which makes them easier to handle. Given a sum
we write
n
Sn= I a;.
i=l
(This is pronounced: "The summation from 1 to n of a;.") That is, when we write
I��i and follow it with an expression involving i, this means that we are to substitute
all (integral) values of i, from 1 to n, and add the results.
1) The geometric series
Sn = a + ar + ar
2
+ · · · + arn-1
can be written as
The shorthand can be checked by substituting the values of i from 1 to n.
i = 2 3 4 n
2 1 a 1 4 l
ar - ar - ar -
2 3
ar ar ar
When we add these, we get the geometric series.
2) Consider the sum

Rn = l2 + 22 + 32 + ... + 2
n .
In the short form,

n
.o
Rn = '
L., z-.
i=l
3) An arithmetic series can be written in the form
Sn = L [a + (i - l)d].
i=l
This can be checked by means of a table of the sort that we gave above for the case
of the geometric series.
In each case, the formula after I gives the ith term; i = 1 gives the first term,
i= 2 gives the second term, and so on. This will always be true so long as we are
taking the sum from i = 1 to n. However, we also write such sums as
5
�
.::.., i· 3•
i=2
Here we take all values of i from i = 2 to i = 5 inclusive and add the results. There
fore
5
I i3 = 23 + 33 + 43 + 53 = s + 21 + 64 + 125 = 224.
i=2
In general, for m � n,
n
I a; =
am + a.,,+1 + + a n.
i=m
· · ·
Thus
4
I (af + 1) (a�+ 1) + (a�+ 1) + (a�+ 1)

i=2
=
= a�+ a�+ a�+ 3 I a; + 3.

1:=2
=
Note that I applies only to the expression immediately after it; in the last line, we
are told to add the numbers a� (from i = 2 to i = 4) and add 3 to the result. The
parentheses in the formula !t=2 (a� + 1) indicate that 1 is part of every term of the
sum.
PROBLEM SET 2.8
Find each of the following sums numerically:
3 3 5 4 3
1. �>2 2. I u-1)2 3. I u2 - 1) 4. "2i2 s. I u3 1)
i=l i=l i=2 i=l i=2
-
Each of the sums below is of the form Lf=m a;. Write each of them in the long form
am + Gm+l + ·
· ·
+ an.
3 n
s. I (3b� + d) 9. !i7
i=3 i=2 i=m
Convert each of the indicated sums to the short form:
10. 12 + 22 + 32 + · · · + n2 + (n + 1)2 11. 32 + 42 + ... + k2

12. k2 + (k + 1)2 + (k + 2)2 + · · · + (n - 1)2
I 1
13. 21 + .13 + ... + +-
.
k-1 k
--
14. Is is true that

n n n
I (a; + b;) = I a; + I b;?
i=l i=l i=l
Why or why not?
2.9 The Induction Principle and the Well-Ordering Principle 51
15. Is it true that

n n
L kai = k 2, ai?
i=l i=l
Why or why not?
16. Is it true that

n
L..
( -) · -
hi 2 h
=
h3
-a
n
L.. iz?
i=1 n n n i=l
Why or why not?
17. For 0 (�)is the number of subsets with exactly k elements, in a given set with
� k � b,
n elements. <mis the number of possible 13-card bridge hands; (552) is
For example,
the number of possible 5-card draw poker hands. Show that (�) <D· =
*18. Show that

GD G�). =
*19. Show that
2.9 THE INDUCTION PRINCIPLE AND THE WELL-ORDERING PRINCIPLE
Consider the following game. We have three spindles, of the sort used as targets in
quoits. On the first spindle is a stack of wooden disks, diminishing in size from bottom
to top. (See the figure.) The disks are numbered 1, 2, 3, . .. , n, from top to bottom;
in the figure, n = 5.
A B c
A legal move consists in taking the topmost disk from one spindle and placing it
on one of the other spindles, providing that we must not, at any stage, place a disk
above a smaller disk.
At the start, all the disks are on spindle A. The object of the game is to get all
the disks onto spindle B, by a series of legal moves.
For example, we might begin by taking disk 1 off spindle A and putting it on
spindle B. There would then be three possibilities for the second move: ( 1) Put disk 1
back on spindle A, (2) put disk 1 on spindle C, and (3) put disk 2 on spindle C. It
would not be legal to put disk 2 on spindle B, because disk 2 would then be above
disk 1, which is smaller.
We shall see that the game can always be completed, no matter how large the
positive integer n may be. For each positive integer n, Let Pn be the proposition that
the game can be completed, starting with n disks. What we need to show is that all
of the propositions P n are true.
Lemma 1. P1 is true.
(A lemma is a sort of subtheorem, used as a step in the proof of a harder theorem.)
Proof of Lemma I. Move the one and only disk from spindle A to spindle B. Then
the game is over.
Lemma 2. P 2 is true.
Proof of Lemma 2. (I) Move disk 1 to spindle C. (2) Move disk 2 to spindle B.
(3) Move disk I to spindle B. Then the game is over.
Lemma 3. P3 is true.
Proof of Lemma 3. By Lemma 2, disks 1 and 2 can be moved to spindle C. (Lemma

2 really means that any two disks at the top of a stack can be moved to any other
spindle.) Do this. Then move disk 3 to spindle B. By Lemma 2, disks 1 and 2 can
then be moved to spindle B, whereupon the game is over.
A pattern is now appearing, suggesting the following lemma. This lemma states
that if the game with n disks can be completed, then the game with n + I disks can
also be completed.
Lemma 4. For each n, Pn => P n+1·

Proof of Lemma 4. We are given n + I disks on spindle A, and we are given by
hypothesis that P n is true. Therefore the stack consisting of disks 1, 2, . . . , n can be
moved to spindle C by legal moves. Do this. (Disk n + 1 causes no trouble; it can
be regarded as the base of the spindle on which it lies, because it is larger than any
of the disks being moved.) Then move disk n + 1 from spindle A to spindle B.
By Pn> we know that disks 1, 2, ... , n can be moved from spindle C to spindle B.
Then the game is over.
Lemma 4 gives us an infinite chain of implications:
And Lemma 1 tells us that the first statement in the chain is true. Therefore all of
the statements P1, P2, • • • are true. This idea is conveyed mathematically as follows:
The Induction Principle. Let Pl> P2, • • • be a sequence of propositions (one

for every positive integer). If
a) P1 is true, and
b) Pn => Pn+i for every n,
then all of the propositions P1, P2, • • • are true.
The problem of the disks is probably the clearest illustration of what the induction
principle means. The principle is used continually, in all branches of mathematics.
In this section, we shall use it to get short formulas for certain sums.
Theorem 1. For every n,

n
n
�> = -(n+ 1).
i=l 2
Proof For each n, let Pn be the proposition that

n
n
Ii= -(n+ 1).
i=l 2
a) P1 is true, because
1
Ii = 1 = tCl + 1).
i=l
b) Pn =?- Pn+I for every n, because

n
n
Ii= - en+ 1)
i=l 2
n
n
=?- L i+ (n+ 1) = -(n+ 1)+ (n+ 1)
i=l 2
=?- �: i= G+ 1) en+ 1)
n +l
n+ 1
=?- I i = -- (n+ 2).
i=l 2
In this chain of implications, the first equation is Pn and the last is Pn+i Therefore ·
Pn =?- Pn+i· By the induction principle, Pn is true for every n, which was to be proved.
In fact, there is a simpler way of getting this result. If
Sn = 1 + 2+ 3+ · · · + (n - I)+ n,
then
Sn = n+ (n - I)+ (n - 2)+ · · · + 2+ I;
and adding terms in pairs, we get
2Sn = (I+ n)+ (I+ n)+ · · · + (I+ n)+ (I+ n),
to n terms. Therefore
n
2Sn = n(n+ 1) and Sn = -(n+ 1) '
2
as before. This device is neat but very special. Consider now the problem of calculat
ing
n
Sn = L i2 = 12+ 22+ 32+ ...+ n2.
i=l
We have just found that the sum of the first n positive integers is a polynomial in n,
of degree 2. This suggests that S n is a polynomial of degree 3. That is, we conjecture
that
Sn = An3+ Bn2+ Cn+ D,
for some numbers A, B, C, D. The problem is to find A, B, C, and D, and prove by

induction that they work. Let P n be the proposition that
n
Pn: L i2=Ana+ Bn2+ Cn + D.
i=l
Then Pn+I asserts that

n
Pn+I: L i2+(n+ 1)2=A(n+ l)a+ B(n+ 1)2+ C(n+ 1)+ D.
i=l
We want P n � P n+l• to make the induction proof work. This means that
Ana + Bn2+Cn+ D+ (n+ 1)2=A(n+l)a + B(n+ 1)2+ C(n+ 1)+ D.

If this equation holds, then P n � P n+i· (Check the algebra.) Collecting coefficients,
we get the equivalent equation
Ana + (B+ l)n2+ (C+ 2)n+ D+ 1

=An3+ (3A+ B)n2+ (3A+ 2B+ C)n+A+ B+ C+ D,
or
(1 - 3A)n2+ (2-3A - 2B)n+ 1-A-B-C= 0.
This holds if
A= t, B = t(2 - 3A)= t, C=l-A-B=i.

This gives
n
Pn: L i2=t na+ tn2+ i n+ D.
i=l
Thus, for any D, Pn � P,.+i· For D= 0, P1 is true. We take D= O; and we know

by the induction principle that
n
L i2=tna+ tn2+ tn,
i=l
for every n. Taking a common denominator on the dght and factoring, we get:
Theorem 2. For every n,

n
n
L i2=- (n + 1)(2n+ 1).
i=l 6
For some purposes, the following idea is easier to use thanthe Induction Principle.
The Well-Ordering Principle. Every nonempty set of positive integers has a

least element.
(See, for example, Problems 10 and 12 below.) The Well-Ordering Principle and the
Induction Principle are equivalent. (See Problems 14 and 15 below.)
PROBLEM SET 2.9
1. Prove by any method that for every n, the sum of the first n odd numbers is n2• That is,
n
I c2i - 1) = n2•
i=l
This can be shown by induction, but there are at least two other ways.
2. Prove by induction that

rn+l - 1
1 + r + r2 + · · · + rn (r � 1).
r - 1
3. Prove by induction that
4. Find by any method a formula for

n
I C3i - 1).
i=l
5. Find by any method a formula for

n
I C4i - 2) .
i=l
6. Find a formula for
n
2
Iu + i + 1).
i=l
7. Find a formula for
i (i2 - i).
i=l
8. Assume that if A1, A2, A3 are points, then
AiA2 + A2A3 � A1A3.
Prove that for every n � 3 we have
A1A2 + AzA3 + · · · + An_1An � A1A,,.
This is known as the polygonal inequality.
9. a) Let Pn be the number of moves required to complete the game with n disks. Show
that for every 11,
Pn+l = 2pn + 1.
b) Let Pn be as in (a). Show that for each 11,
p,, = 2" - 1.
(Since 210 = 1024, this means that the game with 20 disks requires over a million
moves. Thus, if you want to verify that P20 is true, the easiest way to do it is to
show by induction that Pn is true for every 11, and then set 11 = 20.)
*10. Throughout this problem, the numbers under discussion are positive integers. If
a =be for some c, then bis called a factor of a (or a divisor of a). If p > 1, and the
only positive factors of p are p and 1, then p is a prime. Obviously every prime has a
prime factor, namely, itself. Prove that every number greater than 1 has a prime factor.
[Beginning of the proof: "Let K be the set of all numbers Which are greater than 1 and
have no prime factors. We need to show that K is empty. If K is not empty, then . .."]
* 11. Following is the beginning of Euclid's proof that there are infinitely many primes.
Suppose that there are only a finite number of primes, say
Consider the number

N =
P1P2Pa · · 'Pn + 1.
Complete Euclid's proof, by showing that this situation i s impossible.

*12. Show that every rational number can be expressed as a fraction in lowest terms. [Hint:
Try the Well-Ordering Principle.]
13. In the song "The Twelve Days of Christmas," gifts are sent on successive days according
to the following scheme:
First day: a partridge in a pear tree.
Second day: another partridge, and two turtledoves.
For each i, let G; be the number of gifts sent on the ith day.Then
G; = G;_1 + i.
(Which we have just observed for i 2.)=
Let Tn be the total number of gifts sent on the first n days of Christmas. Get a
formula for Tn, in the form
?(? + ?)(? + ?)
?
As a check, the final value is T12 = 364. (I am indebted, for this problem, to Professor
Thomas F. Banchoff.)
* 14. Show that, if the Well-Ordering Principle is taken as a postulate, then the Induction
Principle can be proved as a theorem. [Start of the proof: Suppose that not all of the
propositions Pn are true, and let
K = {n I Pn is false.}.
Then K ""- { }. Therefore ...]
* 15. Show conversely that, if the Induction Principle is taken as a postulate, then the Well
Ordering Principle can be proved as a theorem. [Start of the proof: For each n, let Pn
be the proposition that none of the integers 1, 2, . .. , n belongs to K .... ]
The diagram below is related to one of the problems in this section.

2.10 Solution of the Area Problem for Parabolas 57
2.10 SOLUTION OF THE AREA PROBLEM FOR PARABOLAS
If a line intersects a parabola in two points, then it cuts off a region called a parabolic
sector. In the left-hand figure below, the sector is the region lying above the parabola
and below the line. In the third century B.C., Archimedes discovered a method for
finding the area of a parabolic sector. In this section we shall give an easier solution
of the problem.
The problem will be solved if we can find the area of a "curvilinear triangle" of
the type shown on the right above. If we can do this, then we can find the area of the
trapezoid in the other figure, and subtract the areas of the two curvilinear triangles.
The result will be the area of the sector.
We shall attack the area problem, for the graph of y = x2, by approximating the
region with rectangles, like this:
We cut the closed interval [O, h] into n little intervals of equal length, using the
di.vision points
O, � , 2h , ... , (i l)h , ih , ... , (n

- - l)h , h.
11 11 11 11 n
This gives a sequence of closed intervals
�l [� ' 2: 1 ' [(i l)h ' �l ' [(11 � l)h ' hl

[0,
� . . . ·
.
.
With each of these intervals as base, we construct a rectangle, using as altitude the
height of the parabola at the right-hand endpoint. The right-hand endpoint of the
ith interval is ih/n. Therefore the altitude of the ith rectangle is (ih/n)2•·
Therefore
the area of the ith rectangular region is
Let Rn be the union of all these rectangular regions. Then the area of Rn is
n n h3i2 n
h3
An = £.
�a·i= £.
�- = - � i2
£.
i=l n3 n3 i=l
·
i=l
We want to find out what limit An approaches as n becomes very large. If we find this
limit, then our problem is solved, because the limit is the area of the region R that
we started with.
We found, in Theorem 2 of Section 2.9, that
n n
I i2 = - (n + 1)(2n + 1).
i=l 6
(1 + 1-) ( 1 + -1 )
Therefore
h3 n h3
An =- - (n + 1)(2n + 1)
· = - .
n3 6 3 /1 2n
As /1 becomes large without limit, it is easy to see that
1 1 1
- 1
-- o, 1 + - -1, -o' and 1 + --1,
n n 2n 2n
so that
and
h3 ( 1 1
An =- 1 + - 1 + -
)( ) -
h3
-.
3 n 2n 3
Therefore the area under the parabola, from 0 to h, is
h3
A=-.
3
It would have been equally natural to approximate the area from the inside. We
shall see that this procedure leads to the same answer as before. Here we have cut
up the interval [O, h] into the same little intervals as before; but on each little in
terval we have set up a rectangle whose altitude is the height of the parabola at the
ah
--...,
I
I
I
I
I
I
I
I
I
I
I
I
I
I
h � .. (i-l)h fl: ...<n-l)h nh_

.
n n -
/
n n n n n
left-hand endpoint. Therefore, on [O, h/n] our "rectangle" is merely the base inter
val, with area 0; and thereafter the area of the ith rectangle is
a� = [(i - l)h] 2· � = h3 (i - 1)2.

n n n3
Let R� be the union of these rectangular regions. Then the area of R� is
To see why the last equation holds, observe that each of the indicated sums is the
sum of the squares of the integers from 1 to n- 1. Therefore
h3 n [ h3 .
A� = 3 (n + 1)(2n + 1) n2
J An -
=
- -
-
n 6 n
As n increases, An---+ h3/3 and h3/n---+ 0. Therefore A�---+ h3/3, and we get the same
limit as before. To sum up:
Theorem 1. Let
R y) I 0 � x � h y � x2}.
=
{(x, and 0 �
Then the area of R is h3/3.
y
It is easy to extend this result to the case in which the parabola is the graph of
y = kx2, k > 0.
y y=kx2
.h
i
n
Ui
. - (ih)2h
- -
n n
When we multiply y by k, this multiplies the area of each approximating rectangle

by k. Thus, if A11 = ,2;'.,.1 ai, as before, and
Bn = .2 a;,
-i=l
we have
n n
Bn = ,2 ka; = k ,2 G; = kAn.
i=l i=l
Since An--+ h3/3, we have

kh3
Bn -+-
3
Therefore we have the following theorem:
Theorem 2. Let
R = {(x, y) I 0 � x � h and 0 � y � kx2},
with k > 0. Then the area of R is kh3/3.

In general, for a < b let
be the area of the region under the graph of y = kx2, from x = a to x = b. Then
we have the following:
Theorem 3.
A�kx2 = � (b3 - a3).

3
(Proof? There are three cases to consider: a < b � 0, a < 0 � b, 0 � a < b.)
PROBLEM SET 2.10
Find the area under the graph of y = 5x2, between the following limits.
1. From 0 to 4 2. From 0 to 2 3. From -2 to 0
4. From 2 to 4 5. From -2 to 2
Find the area under the graph of y = 2x2 + 1, between the following limits.
6. From 0 to 4 7. From -1 to 0 8. From -1 to 3
9. Find the area of the parabolic sector between the graphs of y. = 2x2 and y = x + 1.
10. Same problem, for y = x2 and y = x.

11. Find the area of the sector between the graphs of y = x2 - 1 and y = -x2 + 1.
12. Same problem, for y = x2 and y = 2x2 - 1.
13. Solve, for the general case, the problem of Archimedes, stated at the beginning of
this section.
14. a) For each n, let

1
An = 1 + ----= .
v'n
Obviously An > 1 for every n. Under what condition for n can you be sure that
b) Under what condition for n can you be sure that
1
An - l < ?
10 '000 '000
c) Let E be any positive number. Under what condition for n can you be sure that
An - 1 < E?
15. a) For each n, let

2n - 2
En= --
n -1
3 -.
Under what condition for n can you be sure that En < lo?
b) Under what condition for n can you be sure that
c) Given any positive number E, under what condition for n can you be sure that
E,,. < E?
16. For each n, let
Obviously C,,. > 4 for every n. Given a positive number E, show that Cn - 4 < E
whenever n is sufficiently large.

+
n 1
17. a) For each n, let Dn n2 + + 2 . Under what condition for n can you be sure
3n
1
that Dn <
102 ?
b) Given any positive number E, under what condition for n can you be sure that
Dn < E?
18. Given an ellipse, find its area.
y
y
-b
- a
x2 y2
-+- = 1
a2 b2
This can be done by a method somewhat similar to one used in the preceding section of
the text. [Hint: In the figures, what is the relation between y and k?]
19. In the discussion preceding Theorem 1, we fo und that A� = A n - h3/n. Verify this
statement geometrically, without using a formula for either An or A�.
Hint: Draw a figure showing both the inner and outer rectangles, and explain why
h
An - A� = - . 1z2.
n
*20. a) Find a formula for
n
""' •
£.., l 3
•
i=l
b) Find the area of the region under the graph of y = x3, from 0 to 1.
*21. a) Let
as in the text, and let
Thus En is the error in the approximation An ""' '13/3, and En > 0 for each n.
Calculate En and show that En < h3/n for each n.
b) Show that for every E > 0, En < E when n is sufficiently large. That is, find a number
N such that En < E whenever n > N.
Functions,
3 Derivatives, and Integrals
3.1 THE IDEA OF A FUNCTION
Roughly speaking, a function is a law of correspondence under which to each element

of one set there corresponds one and only one element of another set. Consider
some examples.
I) Suppose that we have set up a coordinate system in a plane £. Then to each

point P of E there corresponds a number x which is the x-coordinate of P. Thus
we have a function
E--+R
which matches points P of E with elements x of R.
2) Similarly, every point P has a unique y-coordinate y. Thus we have another

function
E--+ R.
To distinguish these two functions, we give them different names, say, X and Y.
Thus
X: E--+ R,
: PHX
is the "x-coordinate function," and
Y: E--+R,
PHy
is the "y-coordinate function." When we write PH x (with the vertical bar on the
left-hand end of the arrow), this means that each point P is matched with its x
coordinate x. Thus we write ---+ between sets and H between elements of the sets.
3) If the real number x is known, then x2 is determined. Thus we have a function
f: R--+R,
2
X H X .
4) Every nonnegative real number has one and only one nonnegative square root.
Thus we have a function
g: R+--+R,
1-
: x H X,
-Y
where R+ denotes, as usual, the set of all nonnegative real numbers.
63
64 Functions, Derivatives, and Integrals 3.1
5) If x � 2, then x - 2 is nonnegative, and so has one and only one nonnegative

square root. Thus we have a function
h: [2, oo)-+R,
:XH �.
6) The absolute value /xi of xis defined by the conditions
x
/ / = x for x � 0 and x
/ / = -x for x < 0.
In either case, if xis known, then /x/ is determined. Thus we have a function
i: R-+R,
XH /x/.
In each of these six cases we have a function
f: A-+B,
where A and B are sets of some kind. The elements of A are the objects to which
things are going to correspond. The set A is called the domain of the function f
In each case, B is a set which contains all of the objects which correspond to elements
of A. The set B is called the range of the function/ Finally, to have a function/, we
must have a rule under which to each element of A there corresponds a unique element
of B. Under these conditions, we have a/unction of A into B.
We can sum up the preceding examples in the following table.
Example Function Domain Range Rule
1 x E R PH x
2 y --
E
I
R PHy
3 I R R XH x2
4 g R+ R XH V�
5 h [2, co) R xH Vx - 2
6 i R R XH/x/
It is not required that all the elements of the range actually get used. Thus, in
Example 3, x2 � 0 for every x, and so we could equally well write
/: R-+R+,
XHX2,
using R+ as the range instead of R.
Often functions are defined by algebraic formulas, but some of the most important
functions are defined in other ways. Consider the following example.
7) Given the parabola, shown below, which is the graph of the equation y x•2 =
For each point P of the parabola, the arc of the curve from the origin 0 to P has a
certain length. If to each xwe let correspond the length of the arc from 0 (0 , 0) =
to P = (x, x)2 , then we have a function

j: R-+ R+.
3.1 The Idea of a Function 65
(Here we are talking about simple geometric length, independent of direction, and
so the length of the arc is never negative.) Later we shall find that this function can
be described by a formula. But we don't need to know this, let alone find the formula,
to know that we are dealing with a function.
y y
8) Given the same parabola. To each number k � 0 there corresponds a number A

which measures the area of the shaded region in the right-hand figure. To be exact,
the region is
R2 = {(x, y) I 0 � x � k, 0 � y � x2}.
Thus we have a function
/2: R+---+ R+,
kHA.
In Chapter 2 we got a formula for this function:
kH ik3
for every k � 0.
9) Given the graph of y = x", for x � 0.
y
(The rest of the graph goes upward when n is even and downward when n is odd.)
To each k � 0 there corresponds a number A which measures the area of the shaded
region
Thus for each n we have a function
: kl---* A .
Only for the cases n = 1 and n = 2 do we know how to calculate the values of A.
But for n = 3, we nevertheless have a well-defined function/3. Later in this chapter,
you will see how this function can be calculated.
Given a function/: A -+ B, for each a in A we denote by f(a) the element of B
which corresponds to a. For example, if f is the function which squares things
(x � x2), then
f(l) = 1, /(2) = 4, /(3) = 9, f(J2) = 2;
and
/()-;;) =x for every x � 0.
In Example 9 above,f3(1) is the area under the graph of y = x3, from 0 to I ; and
so on.
If the domain A and the range B are sets of real numbers, then we can draw
pictures of the function. The graph of a function f: A _,. B is the set of all points
of the coordinate plane that have the form (x,f(x)). In other words, to draw the
graph of the function, we plot the point (x,f(x)) for each x in A.
r-Y!
I I
I I
I I
In the case shown in the left-hand figure above, the domain is a closed interval
[a, b]. Consider, next, the function g in Example 4, which extracts nonnegative
square roots:
g: R+-+ R,
: x 1--4 )"'-;,
The graph of g (the right-hand figure above) is the graph of the equation y = )°'-;.
To see that this graph is approximately right, observe that
y = )� <=> x � 0, y � 0, x= y2.
We get x = y2 by interchanging x and yin the equation y = x2. Therefore the graph
3.1 The Idea of a Function 67
of x = y2 is a parabola with directrix x = - t and focus (t, 0). And the graph of
y = ..j--; is the upper half of this graph.
A curve which is the graph of a function is called a function-graph. It is easy to
see what sort of curve is a function-graph: A set of points in a coordinate plane is a
function-graph if it intersects every vertical line in at most one point.
y y
I
I
I
y I
I
Yes No
Ordinarily, we make no distinction between a function-graph in a coordinate plane

and the corresponding function.
For example, in the figure above,jis a set of points, and is a function-graph. We use
the same symbol /for the corresponding function. Thus we say that the domain off
is the closed interval [ -1, 7], and the range off is R. (Obviously some smaller set
could be used as the range, but it is not obvious from the figure just what the smallest
possible range is.) We write f(O) = 2, f(l) = 1, J(2) = 3, and so on, because
0 H 2, 1 H 1, 2 H 3, under the action of the function/
Given a function
f: A-+B.
If bis = f(a) for some a in A, we say that bis a value of the function. For example,
2
4 rs a value of the function x H x , but -1 is not. The set of all values of a function
is called the image. If you reexamine Examples 1 through 6 above, you will find
that in 1 and 2 the image is all of R, and in the remaining cases the image is R+.
(You should check these cases.)
Similarly, for
f: [O, 1]-+ R, X H ../I - x2.

Here the graph is a quadrant of a circle, as shown on the left below, and the image
is the closed interval [O, 1].
1
f
A word of caution: it often happens that a figure looks like a function-graph if

you look at it sidewise. In the right-hand figure above, C is not a function-graph;
but sidewise it looks like one. More precisely, if you reflect C across the line y x =
you get a curve which really is a function-graph. We often use this device, to study
various curves C for which the reflection C' is a function-graph. But this does not
mean that C was a function-graph in the first place. Therefore, in the following
problem set, when you are asked whether certain curves are function-graphs, you
must look at the curves right side up. For the curve C shown in the above figure, the
answer is "No," even though for C' the answer is "Yes."
In some of the problems below, you are asked to find the image. In some cases,
the image is not an interval; and you may find it convenient to use the notation
{a, b, c, . ..}
for the set whose elements are a, b, c, .... Thus

N = { l, 2, 3, . . .}
is the set of all positive integers;

z = { ... ' - 2 , -1, 0, 1, 2, ...}
is the set of all integers; and
{O, l}
is the set whose only elements are 0 and 1.
PROBLEM SET 3.1
1. Givenf(x) = x2 + x + 1, for every x. Findf(O),f(l), and [(2).

2. Givenf(x) = 2x2 - x + 3. Findf(-1),f(O), andf(2).
3. Givenfas in Problem 2. Get a general formula forf(2 + h).
4. For what positive integers n (if any) is the graph of y xn a function-graph? = For
each such case (if any), what are the domain and the range?
5. Same question, for the graphs of the equations x = yn.
3.2 The Derivative of a Function, Intuitively Considered 69
6. For what positive integers n (if any) is the graph of y = lxln a function-graph? For
each such case (if any), what are the domain and range?
7. Same problem, for the graphs of the equations lyln = x.
*8. Same question as 6, for y3 + ny = x.
9. Is the graph of x = Vy a function-graph? If so, what are the domain and the ima ge?
Sketch.
10. Same question, for y = lxl/x.
11. Same question, for Jyl = x.
12. Same question, for y = lxl + x.
13. Same question, for y = x2 + x + 1. (Here, of course, the only trouble is in finding
the image. The image is an interval, and should therefore be described in the interval
notation.)
14. The postage rate for airmail letters within the United States is now (1971) ten cents
per ounce or fraction thereof Thus we have a function
amp: R+ ___,. R+,
where amp xis the airmail postage (in cents) for a letter of weight x (in ounces.) Thus
amp t = 10, amp 1 = 10, amp 1T = 40, amp 0 = 0, and so on. Sketch the graph of
this function. What is the image?
15. The roundoff function r: R ---+ R assigns to each number the nearest integer (with a
half-integer assigned to the next highest integer). Thus r(2) = 2, r(2!) = 2, r(2t) = 3,
r (2! ) = 3. Sketch the graph of this function from 0 to 3. What is its image?
16. Under what conditions is a semicircle a function-graph?
17. Under what conditions is a parabola a function-graph? (To solve this one, you will
need a theorem from a problem in Chapter 2.)
3.2 THE DERIVATIVE OF A FUNCTION, INTUITIVELY CONSIDERED
In Section 2.7 we solved the tangent problem for parab olas. Given the graph of
y = ax2 + bx + c,
we found that for each x0, the slope of the tangent at the point (x0, y0) of the graph
was
Obviously a parabola with its axis vertical is a function-graph; its equation expresses
yin terms of x. Thus we have a function
f: R-R
x H ax2 + bx + c.
Now at each point of the graph off there is one and only one tangent; and this
tangent has a certain slope. Thus we have another function
f': R-R
: X H S"' = 2ax + b.
For each x,f'(x) is the slope of the tangent to the graph ofjat the point (x,f(x)).
To see how this works, consider the simplest example, in which
j(x) = x2•
Here the parabola is the graph of the function
j: X H X2,
and the line is the graph of the function
f': X H 2X.
For each x, the value off' is the slope of the tangent to the graph off For example,
at the point where x I the slope of the tangent tofis 2; and f'(l)
=
, 2 ·I 2. = =
At x %, we getf'(i)
=
t; and tis the slope of the tangent to the parabola. Where
=
x - 1, the slope of the tangent to the parabola is -2; andf'(-1) 2(-1)

= -2.
= =
y
In general, suppose that we have a function

f: R ___.... R.
If the graph ofjhas a nonvertical tangent at each point (x,f(x)), we letf'(x) be the
slope of this tangent. This gives a new function
f': R ___.... R.
The new function f' is called the derivative off Consider another example.
y
A careful inspection of the figure above indicates that f' is (at least approximately)
the derivative of/ Thus, at x = 0, the tangent tofis horizontal; andf'(O) = 0, as it
should be. At x = 1, the tangent to f seems to have slope = -1; and f'(1) = -1.
At x = -1, the tangent to f has slope = I; and /'(-1) = I. For x > 0. the
tangent to f has negative slope; and/' (x) < 0 for x > 0. For x < 0, the tangent
to f has positive slope; and/'(x) > 0 for x < 0.
It may be that at some points f has no tangent. At such points,/' is not defined.
Thus, in some cases, the domain off' is a smaller set than the domain of/ Consider,
for example, the function/: x H Ix/.
f'
-1
f'
-1
For every x > 0, the slope of the tangent is 1; and for every x < 0, the slope of the
tangent is -1. Therefore the graph off' looks like the figure on the right above.
Drawing both/ and/' on the same set of axes, we get the left-hand figure be low.
You should carefully inspect the figure on the right below, to conv ince yourself that
f' is the derivative off, at least approximately. Heref has a t angent at x = 0, but
the tangent is vertical, and therefore there is no such thing as f'(O). When x > 0
and x is small, thenf' (x) is large, becausef is rising steeply. When x > 1, f' (x) is
small. It looks as iff'(2) = O; and the g raph off has a horizontal tangent at the
point ( 2,/(2)).
y
f
f f'
f'
-1
0
A function which has a derivative at every point of its domain is called differ
entiable. The following theorem describes a fundamental property of differentiable
functions:
The Mean-Value Theorem. Every chord of a differentiable function is parallel to the

tangent at some intermediate point.
y y
Here by a chord we mean a segment joining two points of the graph. The theorem
says that if/ is differentiable on [a, b], then there is some i between a and bat which
the slope of the tangent is the same as the slope of the chord. As indicated in the
right-hand figure above, there may be more than one such point.
The situation with regard to this theorem is awkward. It is geometrically
obvious. Also it is important and we shall need it soon. On the other hand, the
proof of the theorem is hard, and involves ideas which belong in the later portion of a
calculus course. We shall therefore postpone the proof, but use the theorem whenever
we need it.
The theorem can be stated in a form which looks more algebraic. If f is defined
on an interval [a, b], then the slope of the chord joining the endpoints is
f(b)-f(a)
b-a
and the slope of the tangent at x is f' (.X). Thus the theorem states that
f(b)-f(a)
f'(x)
=
b - a
for some x between a and b. In this style we can restate the theorem as follows:
The Mean-Value Theorem. Suppose that f is differentiable on the closed interval

[a, b ]. Then for some x between a and b we have
f(b)-f(a) .
f'(x)
=
b - a
Note that, if we merely required that the graph have a tangent at every point, the
theorem would become false. The graph shown below has a tangent at every point,
but one of these tangents is vertical. Therefore the function f is not differentiable on
[a, b]. And no tangent line is parallel to the chord from P to Q.
.... Q
,,,.,,,,, l
.,,.,,. I
,,,,,. I
p
I
I
x
a b
Hereafter, the mean-value theorem will be referred to as MVT.
PROBLEM SET 3.2
In each of the figures below, a function-graph/is given. Do a tracing of each graph on a

sheet of writing paper, and then draw a plausible sketch of the graph off'. Obviously your
sketch off' cannot be exact. But f' should be 0 at points where the tangent to f is hori
=
zop.tal; /'(x) should be > 0 where the original graph slopes upward;j'(x) should be < 0
wheref slopes downward; and so on. In some cases, you may find that the values off' are
so large that there is no room for them on the paper. In such cases, draw as much of the
graph of/' as space permits.
Some but not all of the functions shown below satisfy the conditions of the mean-value
theorem. For each such function, draw the chord between the endpoints of the graph,
draw a tangent line which is parallel to this chord, and drop a dashed line from the point of
tangency to the point x on the x-axis. (See the figures in the text, illustrating MVT.)
1. y 2. y 3. y
1 4 4
f, 2 2 f
x
-1 -4 -2 2 4
-2 -2
4. y 5. y 6. y
1
f
f
x x
-1 1 -1
y y y
7. 8. 9.
y y
y
10. f 11. 12.
f
f
x x -x
-1 -1 1
-1
-1
y y y
13. 14. 15.
1
-1 -1 -1
3.3 Continuity and Limits 75
16. 17. y 18. y

1
-1 1 x -1
y y y
19. 20. 21.
1
1
3.3 CONTINUITY AND LIMITS
Let/be a function defined on an interval, or defined for every x. Roughly speaking,

f is continuous if we can draw the graph without lifting the pencil from the paper.
For example, the function f(x) = J1 - x2 is continuous, because it is the upper
half of the circle x2 + y2 = 1. Most functions that arise naturally are of this type,
and in this book we will rarely deal with any other kind of function. But some very
simple functions are not continuous. Consider, for example, the airmail postage
function defined in Problem 14 of Problem Set 3.1. The graph looks like this:
40
30
20
IO
Here y = amp x, where amp x is the airmail postage on a letter weighing x ounces.
The values of this function make sudden jumps at integral values: the graph cannot be
drawn without lifting the pencil from the paper, and so the function is discontinuous.
Functions of this kind are used in physics. For example, the so-called Heaviside
function is defined by the conditions
h(x) = {� if
if
x < 0
x � 0.
The graph looks like the figure below. It makes a sudden jump at x = 0.
We shall now make the idea of continuity more exact, in several stages. Given
a point x0, in the domain off, we want to explain what it means to say that f is
continuous at x0. First we try the following:
1) Whenever xis close to x0,f(x) is close to f(x0). In symbols,
x � X0 => f(x) � f(x0.

)
This is the idea, but it is not good enough; the question is how close things are
supposed to be to each other. As xgets very close to x0,f(x) is supposed to become
very close to f(x0). This suggests:
2) We can makef(x) as close as we please tof(x0), by taking xsufficiently close to x0•
This is better, but it can be improved. We measure the closeness of two numbers
by taking the absolute value of their difference. Thus if E is a positive number, and
/f(x) - f(xo)/ < E,
then we say thatf(x) is £-close tof(x0). In these terms, we can restate (2) as follows:
3) For each E > O,f(x) is £-close to f(x0) whenever xis sufficiently close to x0•
If o > 0 and Ix - x01 < o, then we say that x is a-close to x0• The idea of
"sufficiently close" can be described by taking a positive number o. This gives:
4) For each E > 0, there is a o > 0 such thatf(x) is £-close tof(x0) whenever xis
0-close to Xo·
We can now draw a picture.
/(xo)+E
/(xo) - - - - - -
In the figure, the solid rectangular region is called an EO-boxfor the functionfat the
point (x0,f(x0)
) . When we call it a boxfor the function, we mean that no point of the
graph lies above the box or below it. If the function is continuous, then for every
positive number E, no matter how small, we can find a o > 0 that gives an EO-box.
We now restate (4) as follows.
Definition. Let x0- be a point in the domain of the function f Suppose that for
every E > 0 there is a o > 0 such that
/x - x01 < O => /f(x) - f(xo)I < E.
Thenf is continuous at x0•

This definition applies very simply to the function/(x) = 2x, at the point (I, 2).
Given any E > 0, we can find an EO-box, as shown in the figure; we simply take o =
E/2. Then, algebraically,
Ix - II< o => Ix - II < E/2

=> l2X - 21 < E
=> lf(x) - /(I)I < E;
and this is what we need, to show that the function/ is continuous at the point x = I.
Of course, the function is also continuous at all other points, and we can show in
exactly the same way that the definition of continuity applies, taking o = E/2. (What
would we do for f (x) = 3x, at any point x0 ?)
We shall now apply the definition of continuity to the function f(x) = x2, at
the point (I, I).
y
1-------
d I d'
Given an E > 0, we want to find an ED-box for f, at (I, I), so that
Ix - 11 < D => lx2 - I I < E.
To find the desired number D directly requires clumsy calculations, but there is an
easier way. Let d and d' be the numbers such that
f (d) = d2 = I - E; f(d') = d'2 = 1 + E.
(See the lower figure on page 78.) The graph rises from left to right. Therefore,
d < x < d' => f (d) <f(x) < /(d')

<=> d2 < x2 < d'2
¢:> I - E < X2 lx2 - II < E.
Thus the dotted rectangle in the figure boxes in the graph, in the same way that an
ED-box does. We call such a rectangle a dd'-box. Obviously a dd'-box is just as
good as an ED-box. And, in fact, given a dd'-box, we can always get an ED-box that
lies in it. Let D be the smaller of the positive numbers 1 - d and d' - 1. (Jn fact,
d' - 1 is the smaller, but we don't need to use this.) Then
Ix - 11 < D => d < x < d',

and so
Ix - I I < D => lx2 - II < E,
which is what we wanted. We are going to use this method again, and so we record
it as a theorem.
Theorem 1. Let x0 be a point in the domain of the function f Suppose that for every
E > 0 there are numbers d and d' such that d < x0 < d' and
d < x < d' => f (x0) - E < f (x) < f(x0) + E.
Then/ is continuous at x0.

y
f(xo)+.- - - - -.------�-�--
f(xo) - - - - - - - - - - -- - - - -
8 8
d
Proof Let o be the smaller of the numbers x0 - d and d' - x0. (In the figure,
o = d' - x0.) Then
Ix - x0I < o => d < x < d' => f (x0) - € < f(x) < f (xo) + €
<=> If (x) - f(xo)I < E.
We shall now reexamine the idea of a limit, which we used in defining the slope
of the tangent to the graph of a function.
To find the slope of the tangent at the point (x0,f(x0)), we let m(x) be the slope of the
secant line through the points (x0,f(x0)) and (x, f(x)), where x -¥: x0. Thus the slope
of the secant is a function, and we are now describing it in functional notation. By
definition, the slope of the tangent is
f'(x0) = Jim m(x),

x-xo
if such a limit exists. We shall now give a definition of the limit. The idea is that
limx�x m(x) L if the function m becomes continuous at x0 when we insert the
0
=
value L as the value m(x0). Thus we want to use L as m(x0) in the definition of
continuity. This gives the following:
Definition. Let m be a function defined at each point of an interval /, except at the

point x0• Suppose that for every E > 0 there is a o > 0 such that
0 < Ix - x0I < o · => lm(x) - LI < E.
Then
Jim m(x) = L.
For example, for f(x) = x2, x0 = l, we have
x2 - 1
m(x) = -- = x + 1 (x -¥: 1).
x - 1
When we insert the point (1, 2) on the graph of the function m, we get a continuous
function (which is equal to x + 1 for every x ). Thus lim,,,_"'0 m(x) 2, not just =
intuitively but also in terms of our definition of a limit.

There is one more problem to consider. What if j(x0) is defined? In this case,
what do we mean by lim.,�.,J(x)? The answer is that we ignore the value off at x0,
and investigate how the rest of the graph behaves. To be exact:
Definition.Let/be a function defined on an interval I, except perhaps at the point x0•
Suppose that for every e > 0 there is a b > 0 such that
0 < Ix - x0I < b � If (x) - L I < e.
Then
lim f(x) = L.
x-+x0
Note that here we have simply copied the preceding definition, using f for m: all
along, the value x x0 was ruled out by the condition 0 < Ix - x01. The left-hand
=
figure, showing the eb-box, looks the same as before, except that there is no point in
plotting /(x0) (which may not be defined, and which will not in any case be used).
f(Xo) �/
-r?:
-- - ---
�� -+-��--<�:--��-x
I
Xo
I
The definition of a limit applies in peculiar ways to certain peculiar functions.

In the right-hand figure above, lim.,�.,JCx) = L, but L :;i:. f(x0). The following
theorem shows that the strange situation shown in the figure cannot happen if the
function is continuous.
Theorem 2. f is continuous at x0 if and only if
lim f(x) = f(xo).
Here the displayed formula says three things at once:

1) f(x0) is defined. That is, x0 is in the domain off
2) lim.,�.,J(x) exists. That is,f approaches a limit, as x--+ x0•
3) f(x0) and lim.,�.,0 f(x) are the same number.
PROBLEM SET 3.3
1. How close to 3 does x need to be, for 2x to be within 0.001 of 6? (Answer in the form
Ix - 31 < · · · � 12x - 61 < 0.001. Sketch the graph of j(x) = 2x, and sketch your
.8-box (• = 0.001).)
2. Find numbers d and d' such that d < x < d' � lx2 - 321 < 0.0001. Sketch the graph
ofj(x) = x2 and sketch your dd'-box. In your sketch, you will have to distort the scale
grossly, because of the small size of your •.
3. Show that the function f(x) = x2 is continuous at the point x0 = 3. Use the method
that was used in the text for the same function at the point x0 = 1 and apply Theorem I.
Thus your answer will include statements in the form: "Let d = . . . , and let d' = . .. .
Then d <x < d' => 32 - E < x2 < 32 + €." Sketch the function, showing your
dd'-box.
Answer as in Problem 3 for the following functions, at the following points x 0•

4. j(x) = 2x2, Xo = I. 5. j(x) = V�, Xo = 4.
6. f(x) = V':;, x0 = 8. 7. f(x) = x3, x0 = 2.
8. f(x) = xn, where n is any positive integer and x0 is any number.
9. f(x) =v:;, x0 is any positive number.

* 10. A function f is called Lipschitzian if there is a number m > 0 such that for every two
points x, x0 we have
l/(x) - f(x0)1 � m Ix - x0I.
Show that every Lipschitzian function is continuous.
3.4 THEOREMS ON LIMITS
In this section we shall give the elementary rules that we use in dealing with limits.
These rules are much easier to learn and to use than they are to prove, and so many of
the proofs are omitted from this section. (You will find the missing proofs in Appendix
B.) But some of the proofs are easy, and they throw some light on the idea of a limit.
Theorem 1. If limx-xJ(x) = L, then limx-x0 [-j(x)] = -L.
Proof To get -/from/, we flip the graph of/ across the x-axis. We know that for
every e > 0, fhas an eb-box at (x0, L). If we flip the box across the x-axis, in the
same way that we flipped the graph, this gives us a box for -fat (x0, -L).
This theorem can also be proved algebraically. The hypothesis means that:
(1) for every e > 0 there is a o > 0 such that
0 < Jx - x01 < O => If (x) - LI < e;
3.4 Theorems on Limits 83
the conclusion means that: (2) for every" > 0 there is a o > 0 such that
O<lx-xl
o <o => 1-f(x)-(-L)I<"·
Since 1-f(x) -(-L)I = lf(x) - LI, it is obvious that (1) => (2), and thus the
theorem holds.
Theorem 2. If limx-xJ(x) = L, then limx-xo [f(x) - L] = 0.
!I
I � f
J---�- 1
I
I
51
rl f-L
Proof Given f, we get f - L by moving the graph up or down a certain distance

(down or up, according as L is positive or negative). We move the box along with the
graph; and this gives a box for the functionf - L.
Theorem 3. If Iimx-xo [j(x) - L ] = 0, then limx- x0 f(x) = L.
To prove this, merely use the previous proof in reverse; move the box along
with the graph.
Theorem 4. If limx_.,0 f(x) = L, and k is any number, then limx-xo kf(x) = kL.
That is, the limit of a constant times a function is the same constant times the
limit of the function.
y
Proof
1) Fork = 0, this is easy: kf(x) = 0 for everyx, and so lim.,_x0 kf(x) = 0 = 0 · L.
2) Suppose that k > 0. For every E > 0, the graph off has an EO-box at (x0, L).
Therefore, for every E > 0, the graph of /has an (E/k) o-box at (x0, L). Thus
0 < Ix - x01 < O => 1/(x) - LI < E/k.
Multiplying by k, in the inequality on the right, we get:
0 < Ix - x01 < o => lkf(x) - kLI < E,
and so lim,H,,0 kf(x) = kL, which was to be proved.

3) Proof for k < O?
These proofs illustrate the way we work with boxes, to prove things about limits.
The following theorems will be proved later (unless you find a way to prove them
for yourself).
Theorem 5. If lim x�x f(x) =Land 1imx�x0 g (x) = L', then

0
Jim [ (x) + g(x)] L + I.:.
x-+x f
=
0
That is, if each of the functions/and g has a limit, as x--+ x0, then the sum also
has a limit, and the limit of the sum is the sum of the limits.
Theorem 6. If limx-•xo f(x) = Land limx-•xo g(x) = L', then
Jim [f(x)g(x)] = LI.:.

x-xo
Theorem 7. If limx�x f(x) =Land limx�x0 g (x) = L', and L' � 0, then
0
Jim
f(x) =
!=._
.
x-xo g(x) I.:
Caution: The preceding theorem says nothing about what happens when L' = 0.
anything can happen, even in very simple cases. If
And in fact, for L' = 0
f(x) = 2x, g(x) = x,

then
f(x) =
2x
and Jim
f(x) = 2.
g(x) x x-o g(x)
And any number k can be used in place of 2. Therefore, if f (x) --+ 0 and g (x) --+ 0,
the quotient//g can approach any number whatever as a limit. This should not sur
prise us, because every time we calculate a derivative we are finding the limit of a
quotient
f(x) - f(xo)
X - Xo
whose numerator and denominator are approaching 0.

In the preceding section we showed that f is continuous at x0 if and only if
lim f(x) = f(x0).
(This was Theorem 2.) Hereafter, we shall regard the above formula as interchange
able with the definition of continuity. Thus every theorem on limits automatically
gives us a theorem on continuous functions. Some of these are as follows:
Theorem 8. If f is continuous at x0, and k is any number, then kf is continuous at x0•

Proof We are given that
lim f(x) = f(xo).
By Theorem 4,
lim kf(x) = kf(x0),
and this means that kf is continuous at x0•
Theorem 9. If f and g are continuous at x0, then so also are f + g and Jg.
Proof? (Use Theorems 5 and 6.)
Theorem 10. If f and g are continuous at x0, and g(x0) >'6 0, then fig is continuous
at x0•
(Use Theorem 7.)
Most of the time we shall apply these results not just at one point x0 but through
out the domain of the functions f and g. For these cases, we can state our theorems
more briefly as follows:
Theorem 11. Letf and g be functions with the same domain. Iff and g are continuous,
then so also are kf, f + g, and Jg. And fig is continuous at every point x0 where
g(x0) ¥=: 0.
Thus, for example, given that/(x) x2 + 1 and g(x)

= x4 + 4 are continuous,
=
on the entire real number system, we can infer immediately that kf,f+ g,fg, andfig
have the same property. Here g(x) ¥=: 0 for every x. Given
h(x) = x2 - 1,
we can infer that/+ h andfh are continuous everywhere, and thatflh is continuous
except at 1 and -1. Of course, at x 1 and x
= -1 it is not just continuity that
=
breaks down: the quotient function is not even defined at these points, because the
denominator ofj/h becomes 0.
Finally, a trivial observation.
Theorem 12. Every constant function is continuous.
k -- -- --
---
Xo X
Proof Given/(x) k, for every x in a certain domain. We need to show that for
=
every " > 0 there is a o > 0 such that
Ix - Xol < o => If (x) - kl < E.
Obviously any positive number can be used as o.
PROBLEM SET 3.4
In solving the following problems, you need not base your work directly on the definition
of continuity or the definition of a limit; you are free to use all the theorems stated in this
section. Note that the later problems are not based on this section at all; they are extensions
of the theory.
1. Show that if/(x) = kx, then/is continuous.
2. Same, for f(x) kx2•=
3. Same, for f (x) xn (n a positive integer).

=
4. Same, for f (x) x3 - 3x2 + x - 4.

=
5. Apolynomial of degree n is a function of the form
where an � 0. (The function which is = 0 for every x is a polynomial of degree 0.)

Show that every polynomial is continuous.
6. Show that if
xn
f(x) -
1 + x2'
then f is continuous.
A function f is bounded if there is a number Msuch that - M � f (x) � Mfor every x

in the domain of M. If the above inequalities hold, then we say that Mis a bound off.
Obviously, to show that a function f is bounded, you have to name a bound of f; and to
show that a function is unbounded, you have to show that no number Mis a bound off.
Find out which of the following functions are bounded, on the given domains, and justify
your answers:
1
7. f(x) = -- , - oo < x < oo 8. f(x) = x2, - oo < x < oo
1 + x2
x2
9. f(x) = x2, 0 ;;:; x ;;:; 1 10. f(x) = --2 , - w < x < w
1 + x
x
11. f(x) = x3, 0 ;;:; x ;;:; 2 12. f(x) = -- 0 � x � 1
1 + x2' - -
x x
13. f(x) = , 1;;:; x < oo 14. f(x) = -- , - oo < x < -1
1 + x2 1 + x2
x 1
15. f(x) = -- , - oo < x < oo 16. f(x) = -- , 1 < x < oo
1 + x2 1 + x3
1 x4
17. f(x) , -1 < x < 1 18. f(x) , 0 < x < oo
1 + x3
= = --
1 + x3
x + 1
19. f(x) = -- , -1 < x < 1
1 + x3
20. Show that if f is bounded, then so also is kf for every k.
21. Show that if f and g are bounded (on the same domain), then so also is f + g.
*22. Show that if/and g are bounded (on the same domain), then so also is/g. You may
find it convenient to write the condition for boundedness in the form If(x)I ;;:; M.
Can you infer also that fig is bounded? Why or why not?
*23. a) Show that if f is bounded and
Jim g(x) = 0,
x-xo
then
Jim [f(x)g(x)] = 0.
x-xo
(First try proving this for the case M 1.)

=
b) Show that if
Jim f(x) 0, =
then
Jim [f(x)sin x] = 0.
x-xo
(Here it makes no difference whether degrees or radians are being used.)

c) In Problem 23(a), can we get along without the hypothesis that/is bounded? That
is, if
Jim g(x) 0, =
x-xo
does it follow that

Jim [f(x)g(x)] = 0,
no matter what kind of function f may be? Why or why not?

d) Show that if
Jim f(x) = 0,
x�o
then
Jim
x�o
[t(x) �]sin
X
= 0.
(Query: Can Theorem 6 be applied to this problem?)

e) Show that
1
Jim x2 cos - = 0.
x-o x
t 5 I
M � -
-M -----
*24. a) A function fis locally bounded at x0 if there are positive numbers M and o such that
0 <Ix - x01 < o=> lf(x)I < M.
b) If fis bounded, does it follows that f is locally bounded at each point of its domain?
c) Conversely, if f is locally bounded at each point of its domain, does it follow that f
is bounded?
d) If f is locally bounded at each point of the open interval (0, 1), does it follow that f
is bounded on (0, 1)? Why or why not?
e) Show that if
Jim f(x) = L,
x-xo
then fis locally bounded at x0. (This result does not require that x0 be in the domain
of f If you draw a picture of what you have, and a picture of what you want, and
.
compare the two, this proof may become obvious.)

f) Show that if f is locally bounded at x0, and
Jim g(x) = 0,
x-xo
then
Jim f (x)g(x) = 0.
x-xo
3.5 The Process of Differentiation 89
3.5 THE PROCESS OF DIFFERENTIATION
The theorems in the preceding section tell us enough about limits to give us some
information about derivatives.
To make some formulas easier to write, we introduce an alternative notation for
the derivative: we write DJ to mean the derivative off Thus
Df=f',
by definition. Similarly, if
h(x) =f(x) + g(x)
for every x in a certain domain, then
D(f + g) = h'.
Similarly, when we write
D(x2 + 2x + 5)
we mean the derivative of the function
x H (x2+2x+ 5).
We know already what this derivative is:
D(x2+2x+ 5) = 2x + 2.
Here we are merely rewriting the result which we got quite a while ago: for each x,
the slope of the tangent to the graph of
y= ax2+bx+c
is given by the formula
S,, = 2ax+b.
We recall that a functionfis differentiable at a point x0 if it has a derivative at x0.
When we say thatf is differentiable, we mean that it has a derivative at every point of
its domain. For example, if f(x) = lxl, then f is differentiable at 1, but not at 0.
But ifj(x) = x2, thenf is differentiable ( without qualification).
Theorem 1. The derivative of a constant function is 0.
Here by a constant function we mean a functionffor whichf(x) = k for every x.

Obviously the slope of the tangent is 0 everywhere. Algebraically,
f(x) - f(x o)
!'(Xo) - 1.lm
X - x0
_
rr;->x0
k
k
- = 1·1m 0 = 0 .
= !.im ---
x->x0 X - X0 x->x0
Theorem 2. Jf f is differentiable, then so also is kf for every k, and
D(kj) = kDf
Proof Take any x0 in the domain off Then
1. f(x) - f(xo) _ '

Im - f (x0),
x-->x0 X - x0
by the definition off' (x0). Therefore, by Theorem 4 of Section 3.4,
1. kf(x) - k f(x0) _ '

Im - kf (Xo)
X - Xo
•
"''"'"'o
Therefore, at each point x0 , the derivative of kf is k times the derivative off We

have seen an example of this:
Dx2 = 2x,
and
D(kx2) = 2kx,
as it should be.
Theorem 3. If f and g are differentiable, then so also is f+ g, and
D( f + g) = DJ+ Dg.
Proof. Given an x0 in the domain of f and g, we have
f(x) - f(xo)
l im = f '(xo)
"'"""'o x - x0
and
1. g(x) - g(x0) '

Im =
g (x0).
"''"'"'o X - Xo
We want to prove that
1. [j(x) + g(x)] - [f(x0) + g(x0)] _ f'( '

Im - x0) + g (x0) .
"''"'"'o x - Xo
Since we know by Theorem .5 of Section 3.4 that the limit of the sum is the sum of the
limits, the result follows immediately; the big fraction in the third formula is the sum
of the two fractions in the preceding two formulas.
We shall now show that
Dx3 = 3x2•
For each x, let
f(x) = xa;
and take any x0. Then
· f(x) - f(xo) r· x3 - x�
' i
f (x0) = Im = Im
o:->o:0 x - x0 o:->o:0 X - x0
. (x - x0)(x2 + x0x + x�)

= IIm
.,_,.,o x - Xo
= lim (x 2 + x0x + xg)

!t:�:t:o
by Theorems 6, 4, and 5 of Section 3.4. (Why?) Therefore
for every x0, and
which is what we wanted.

To extend this result to j(x) = x", for every positive integer n, we merely need
to know the general factorization formula
This is easy to check by multiplication.
Th�orem 4. Dx" = nx"-1, for every positive integer n.
Proof. Let f(x) = x" for every x, and take any x0• Then
. f(x) - f(x0) 1. x" - xg

IIm = lffi
1. (x - x0)(x11-1 + x"-2x0 +
= Im
z-+ :Z:o X - Xo
· ·
= lim (x"-1 + x
11-2
x0 + · + xxg-2 + xg-1)
x-xo
(to n terms)
Thus/' (x0) = nx�-1 for every x0, and so
Dx" = nx"-1,
The preceding theorems, in combination, enable us to differentiate any poly

nomial. For example,
D(17x29 + 7TX17 - 7x5) = Dl7x29 + D7Tx17 - D7x5,
because the derivative of the sum is the sum of the derivatives. This is
= 17 29x28 + 177Tx16 - 35x4•

·
Theorem 5. If f is differentiable at x0, then f is continuous at x0.

y
I
I--/'
I
I
I
I
---1I
I
I
·x
---+---�xc!--·o----
This is easy to see, as a matter of common sense. By definition, f'(x0) is the

slope of the tangent. Therefore f must have a nonvertical tangent at x0. But if f
"made a jump" at x0, there surely couldn't be a nonvertical tangent. The secant
lines would get too steep for their slopes to approach a limit. Pictures are useful, but
it is hard to be sure that the pictures we draw allow for all possibilities. Jn any case,
the theorem is easy to prove algebraically. Given
. f(x) - f(x0)
Itm - f'(x0),
_
x-xo X - Xo
then
. . [
f(x) - f(xo)
hm [f(x) - f(x0)]
x-+x0
=
hm
x-x0 X - x0
· (x - x0)
J
=
f'(x0) · 0 = 0.
(Theorem needed, for the second step?) By Theorem 3 of Section 3.4, lim.,_,, f(x) =
0
f(x0), which was to be proved.
The differential calculus would be simpler if the derivative of the product were
equal to the product of the derivatives; but this is not so. For example, take
f( x) = xa, g (x) = x2.

Then
f'(x) g'(x)
· = 3x2 • 2x = 6x3,
which is not the same as

Dx5 = 5x4•
A correct formula can be derived as follows. Take any x0, and suppose, as usual,
thatfand g are differentiable. Then
1. f(x)-f(x0) '
-f (X0) (1)
_
lffi
.,_,.,o X - Xo
and
1. g(x) - g(x0)
tm = g '(Xo) . (2)
.,_,.,0 X - x0
To find the derivative of the product, we need to find
.f(x)g(x)-f(x0)g(x0) .
lim (3)
x->x0 X- x0
In a similar situation, when we were finding the derivative off+ g, there was no
problem: we looked at the fraction whose limit we wanted to find, and observed
that it was the sum of the two fractions
f(x)-f(xo) g(x ) - g(x0)

X - Xo X - Xo
whose limits we knew. If these fractions appeared in (3), then we could use the fact
that their limits are given. Since neither of them appears, we use a trick: we simply
put one of them there, fix up the rest of the fraction so that its value is unchanged,
and hope for the best:
f(x)g(x) - f(x0)g(x0) f(x)g(x)-f(x0)g(x) + f(x0)g(x) - f(x0)g(x0)

x-� x -�
f(x) - f(xo) + g(x) - g(xo)

f(xo).
=
g(x)
x - x0 X - x0
Now we can see what happens as x ->- x0:
1. f(x)-f(xo) '
-f (Xo,
)
_
lill
x->x0 X - X0
lim g(x) = g(x0),
1. g(x) - g(x0)
lffi = g'(Xo) .
:z:->:z:o x - Xo
Therefore, by our theorems on limits of sums and products, we have
!. f(x)g(x) - f(x0)g(x0)
1m -f'(x0)g(x0) + f(X0)g'(x0)
_
•
:z:->:z:o X - Xo
In words:
Theorem 6. The derivative of the product of two differentiable functions is the

derivative of the first, times the second, plus the first times the derivative of the
second.
More briefly:
D(fg) = f'g + g'f
Let us try this one out for
f(x) = x3, g(x) = x2•
Now
/'(x) = 3x2, g'(x) = 2x.
Therefore
f'(x)g(x) + f(x)g'(x) = 3x2 x2 • + x3 2x • = 5x 4,
as it should be.
Next we want to find the derivative of the reciprocal g = l// of a function f
As usual, we take a fixed x0; and we assume that f is differentiable at x0. We must
have/ (x0) � 0, or g(x0) would not be defined. Now
1. g(x) - g(x0) 1. I //(x) - 1 /J(x0)

g (x0) - Im ,
' . _ _
1m -
x�x0 X - Xo x�x0 X - Xo
if such a limit exists. Algebraically,
1/f(x) - l/f(x0) =
f(x0) - f(x) f(x) - f(x0) -1
x - X0 f( x)f(xo)(x - x0) x - X0 x
f( )f(x0)
As x-->- x0, the first fraction approaches f'(x0), and the second fraction approaches
- 1/ [/(x0)] 2 , because/(x0)� 0. Therefore
, -f'(xo)
g (x ) =
o [f(xo)J2 '
and so
at every point x where f (x) � 0. In words:
Theorem 7. The derivative of the reciprocal of a differentiable function is equal to

minus the derivative of the function, divided by the square of the function (wherever
the function is different from zero).
Using the preceding two theorems, we get
D
(f); D f
( ;1) f' ·
1
; + fD
(1); !'
;- + f.
-g' =
7 f'g - g'f
g2
= ·
· = =
In words:
Theorem 8. The derivative of the quotient of two differentiable 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
(wherever the denominator is not = 0).
More briefly, at every point x where g(x) ¢ 0, we have
D (�) = gf' �Jg'

Let us try this one out in the case
f(x) = x4, g(x) = x2,

where we .already know the answer. By Theorem 8 we get
D (-xx42) = x2 • 4x3 - x4 2x
(x2)2
· 2x5
=- =
x4
2x.
This is right, because Dx2 = 2x.

For convenience of reference, we list our differentiation theorems as short for
mulas:
(i) Dk 0,
=
(ii) D(kf) kDf,

=
(iii) D(f + g) DJ + Dg, =
(iv) Dxn =nxn-i (n a positive integer),

(v) D(fg) f' g + gf,
=
(vi) n(�) f[_2 _

(wherever f ¢ 0),
(vii) �) = gf' - Jg'

D (J (wherever g ¢ 0).
g2
Theorem 5 did not involve a formula. It said that if a function is differentiable
at a point, then it is continuous at the same point.
Finally, some remarks on the notation used for derivatives. When a function is
named by a letter, such as/, then the notation DJ is unambiguous: it means f'.
But when we describe functions by formulas, it is not always obvious what function
we mean. When we speak of "the function x2 - x + 1,'' it is obvious that we mean
x H x2 - x + 1; and when we speak of "the function t2 + 2t -
3," we mean
t H t2 + 2t - 3. But if we speak of the "function"
t2 - tx2 + x,
we might have either of two things in mind:
a) xis regarded as a constant; and our function is
j: t H !2 - tx2 + X.
b) tis regarded as a constant, and our function is
g: x f-> t2 - tx2 + x.
In such a case, it would hardly do to indicate a derivative by writing
(?) D(t2 - tx2 + x) (?),
because nobody could tell whether we meantf' or g'. To eliminate the ambiguity,
we write D1 or D,,, to indicate which letter does not represent a constant. Thus
D1(t2 - tx2 + x) = f'(t) = 2t 2

- x ,
while
D,,,(t2 - tx2 + x) = g'(x) = -2tx + 1.
Similarly,
Dx(ax3z + z2) = 3ax2z,
D.(ax3z + z2) = ax3 + 2z,
Da(ax3z + z2) = x3z.
PROBLEM SET 3.5
All of the following are differentiation problems. Most of them can be worked by the
standard formulas that we have just derived. But in some cases you will need to start with
the definition of[' (x0) and then use various algebraic strategems.
1 x
2.
0
1. D(7x1 - x8) D-- 3. D --
x + 1 x + 1
1 y
4. D -- 5. D -- 6. D(?y4 _ y2 + 7T)
x2 + 1 y3 - 3
7.
n (�) 8.
(
n 2 � x) 9. D(l + x)3
10. a) Dx(x3y + ay3 + xy 2) b) Dv (x3y + ay3 + xy 2) c) Da (x3y + ay3 + xy2)

11. a) Dx(3axy + x2 + a3) b) Da(3axy + x2 + a3)
x2 - x + 1
12. D[(x2 - x + l)(x2 + x + 1)] 1 3. D
x2 + x + 1
x + 1
14. D-- 15. D(x2 + x)2
x3 - x
16. If you worked Problem 9.of Section 3.3, you know that f(x) Vx is continuous in its
=
entire domain R+. Assuming, in any case, that this is true, find['. [Hint: Set up the
fraction whose limit is/'(x0), rationalize the numerator and hope for the best.]
17. Given/(x) = Vx + 1 (x;;; -1),findf'. Here you may assume thatfis continuous.
But you should mention this fact, at the stage where you need it.
18. Given/(x) = Vx2 + I, find/'. (Assume that/ is continuous.)
19. Givenf(x) = VJ - x2 (-1 ;;:; x ;;:;: 1), find/'.

3.6 The Process of Differentiation: Roots and Powers of Functions 97
20. Gi veng(x) = x2Vl - x2, findg'. 21. Find D(I/Vx) (x > 0).
22. Find D(l/Vl - x2). 23. Find D(x/Vl - x2) (-1 < x < 1).
24. Now solve Problem 19 by the methods of Chapter 2, without using limits or differentia-
tion formulas.
Find out whether the following formulas are correct, and give your reasons.
25. D(x2 + 1)2 = 2(x2 + 1) (?) 26. D(x2 + 1)2 = 2(x2 + 1) · 2x (?)
27. D(x2 + 1)3 = 3(x2 + 1) (?) 28. D(x2 + 1)3 = 3(x2 + 1) · 2x (?)
29. D(x2 + osoo 500(x2 + 1)499 (?) [Hint: In fact, this formula is wrong. And it is
=
possible to prove that it is wrong, without finding out what the derivative of the given
function really is.]
30. Prove by induction that Dxn = nxn-1, without using either a factorization formula
or the binomial theorem.
31. Same problem, for Dx-n = -nx-n-1•

32. Find DVx2 + x + 1.
3.6 THE PROCESS OF DIFFERENTIATION:

ROOTS AND POWERS OF FUNCTIONS
Some of the answers that you got in the preceding problem set deserve to be regarded
as standard differentiation formulas. For example, you found that for
f(x) =
Jx
we have
f'(x) =
2 vlrx .
This problem is going to come up again. We had, therefore, better add it to the list
of formulas at the end of Section 3.5:
- 1
(viii) D.jx =
2,/x .
You found also that
> - 1),
1
Dy'X+I (x
2.Jx + 1
=
D J x2 + 1 =
x ,
J x2 + 1
DvI1 - x2 =
-X
--=== (-1 < x < 1).
J1 - x2
In each of these cases, we have the problem of finding the derivative of the positive
square root Jj of the differentiable positive functionf If we can solve this problem
in the general case, then we can get a formula
n.J]= ?;
and we can then apply the formula hereafter. Suppose, then, that we are given
g(x) == .JJ(x),
on a domain where f (x) > 0; and suppose that f has a derivative at a certain point
x0 of the domain. We want to find g'(x0). By definition,
1. g(x) - g(x0) = 1.lm JM - Jf(XJ

g (Xo) =
'
lm
x->x0 X - x0 x->x0 X - x0
= lim
x->a:o
[.JI(X5 -- .Jf(XJ
XXo
.JTW .J!!S?__
.jf(X5 Jf(xo)J
·
+
[f(x) - f(xo) . 1
JJ(x) JJ(xo)J
= lim
a:->a:o X Xo
- +
1
= f'(x0) lim ,
-> x a:o .jf(X5 + JJ(xo)
provided that the latter limit exists. It is easy to see that this limit exists, provided that
lim .jf(x) = JJ(x0). (1)
Since the limit of the sum is the sum of the limits, it then f ollows that
lim [.jf(X5 + JJ(xo)] = 2Jf(xo);

and since the limit of the quotient is the quotient of the limits, we get:
Theorem 1. If f is positive and differentiable, then
DJ = f' .
] 2J]
Let us try this on the function x H Jx + 1. Here f(x) =x + 1, and f' (x) = 1
for every x. Therefore
1
DJ.X+l = J.X+l '
2 x+ 1
which is the right answer. For x HJ x2 + 1, we have
f(x) = x2 + 1,
f'(x) = 2x,
x) x
DJx2 + 1 = DJJ(x) = f'( = �
2JiW 2Jx2 + 1
which is the right answer. Formula (viii), of course, depends on
lim.JJ(x) = .JJ(x0), (1)
which we haven't proved. We postpone the proof, observing meanwhile that (1) is
reasonable. Since f is continuous, we have
x """' x0 => f(x) """'f (xo) .

Since .J is continuous, we have
f(x) """'f(xo) => .J f (x) """' .J f(xo).

Fitting these two statements together, we get
x """' x0 => .Jf(x) """' .J f(x0),

which is what we want.
Consider now the following function:
g(x) = (x5o _ xl7 + 1)247.

If you want to find g' (x), it is not helpful to observe that g is a polynomial, of degree
12,350, or to recall the binomial theorem. In fact, the right approach is to solve first
a more general problem. Given a function g which is a power of a function f Thus
g(x) = f"(x),
where by f"(x) we mean [f (x)]n. Then
1. rCx) - rCxo)
(Xo)
'
g Jill
X - X0
=
x-+x0
Iim
([f(x) - JCxomr-1Cx) + r-2(x)f(xo) + · · · + r-1cxo)l )
X - x0
=
x-+x0
We see that this limit is

g'(x0) = nj"-1(x0)f'(x0), (2)
provided that in the brackets on the right we have
lim fk(x) = fk(x0) (3)

x�x0
for each positive integer k. Equation (3) states that the kth power of a continuous
function is always continuous. And this is true: given that
Jim f(x) = f(x0),

x�x0
it follows that
lim f2(x) = lim [f(x)f(x)] = f(x0)f(x0) = f2 (x0),
because the limit of the product is the product of the limits. For the same reason,
Ink - 1 such steps, we get Eq. (3). Therefore Eq. (2) is correct; and we have:
Theorem 2. If f is differentiable, and n is a positive integer, then
nr = njn-1!'.
Let us try this on our polynomial of degree 12 ,350:
D (xso _
x17 + 1)247 =Df247 = 247f246f'
= 247(xso x17 + 1)246(50x 49 - 17x16).
-
Note that our use of the shortcut formula Dfn =nfn-1.f' has two advantages
over the method based on the binomial expansion. First, the calculation is possible,
as a practical matter. Second, it gives the answer not merely in a correct form, but
also in a factored form, which is easier to handle than the binomial expansion of the
derivative.
Since we know how to differentiate fractions, we know how to differentiate
functions of the form
1
f(x) =---;;,
x
where k is a positive integer. We have
-Dxk -kxk-l
f'(x) = =
(xk)2 x2k
wherek + 1 = 2k - (k - 1). If we express 1/x k as x-k, and make the same change
in the formula for f', then we get
Dx-k = -kx-k-1•
This has the same form as our previous formula Dxn =nxn-1, with n = -k . What
is needed, to take care of all such cases, is the following:
Theorem 3. If n is a positive integer, and f is differentiable, then
Ifn is a negative integer, then the same formula holds at every pointx wheref(x) � 0 .
The last condition is necessary: ifn < 0 , thenf(x) appears in the denominator of
rcx) = l/J-n(x), and r is therefore not defined at points where f (x) = 0.
Theorem 3 has already been proved for the case in which n is a positive integer.
For the case in which n is a negative integer, = -k, the proof is as follows:
kp-1f'
nr =Df-k = D _!_k = k
-
f f2
= -kf-lc-lf ' = nr-l f'.
For convenience of reference, we list all the differentiation formulas that we have
so far:
(i) Dk = 0, (ii) D(kf) = kDf,

(iii) D(f + g) = DJ + Dg, (iv) Dxn = nxn-i (n :rt= 0),
(v) D( fg) = f' g + gf, (vi) ()

D ! =
f
- [_
!2'
(vii) ()
D L = gf' - Jg' ,
2
...) DyX
(Vlll r=
1
2Jx
---=,
g g
(ix) DJ]= �J' (x) nr = njn-1.f' (n :rt= 0).
Each of these formulas holds for every x for which its right-hand member is defined.
PROBLEM SET 3.6
Find, by any method:
1 x
Dv'(x +l)(x +2) 2. D D
(x2 +x +1)2 (x2 +2x+1)2
I. 3.
x2 +1
4. DVx4 +5x2 +2 5. D(x3 +x2 - x +7)�2 6. D- --
x2 - 1
v'x - I
7. DVx(x - 2) 8. DVx3 +2x +1 D --
x2 + 1
9.
x2
J�
1
10. D- - -- 12. DYv'�
-
x2 +1 +x
11. D
1
13. DVx2 (Warning: Don't "simplify" yourself into a wrong answer.)
x3y2
14. D,, v' x2 +a2 15. Dv(x3y2 - x2y3)3 16 · Dx
(x2 +y2)2
17. Recall that for x > 0,
by definition. Show that
.{;xv = ( �:X)1' (x > o) .

_
[Hint: By definition, v
.a,
xv is the number which, when raised to the qth power, gives x'l>.
Therefore you need to show that the qth power of the right-hand side of the above
equation is xv.]
This is an instance of a frequent phenomenon: often, a problem becomes easy if
we rewrite it, using the definitions of the ideas that the problem involves.
18. Find Dx312, and write the answer in a form which brings out the analogy with the
formula Dxn nxn-1•
= You may assume that x312 is continuous.
19. Find D.,[V x3 + x(x3 + x)].

20. Find D.,(x4 + 2)312•
*21. Get a general formula for D/312, where f is any positive function (differentiable, of
course). The answer should be written in such a form as to bring out the analogy with
the formula D fn =
nfn-lf'.
*22. Find a formula for D/512, where f is differentiable and positive.
23. Find D.,(x2 + 3x + 1)512•
24. Find a formula for Dx-312 and write it in a form which brings out the analogy with
the formula Dxn = nxn-1•
25. a) Simplify
a - b
a3 - b3.
(Obviously, the word "simplify" can mean many different things. In this case, it
means to get rid of the numerator, so as to get a formula which will be useful to you
in solving the next part of the problem.)
b) Find DflX: assuming that fix is continuous.
26. a) Simplify, as in Problem 25,

a - b
(Hereq is a positive integer.)

q_ q _
b) Find DV x, assuming that V x is continuous.
*27. Given positive integers p, q. Find a formula for Dxp/q, valid for x > 0, and write the
answer in a form which brings out the analogy with the formula Dxn nxn-1•
=
3.7 THE INTEGRAL OF A NONNEGATIVE FUNCTION
Given the graph ofy = kx2. In Section 2.10, we calculated the area A" of the shaded
region
R = {(x,y) \ 0 � x � h, 0 � y � kx2}.
We found that
Ah -
- �3 h3.
y y
2
y=kt
3.7 The Integral of a Nonnegative Function 103
In this situation, we regarded h as a fixed number. On the other hand, it is plain

that h can be any positive number, and that when his named, Ah is determined. Thus
Ah can be regarded as a function.
To discuss Ah as a function, without confusing the notation, let us relabel our
horizontal axis as the t-axis, as shown on the right. Thus our parabola becomes the
graph of the equation y = kt2• (Here t is acting like
x.) For each x � 0, let F(x)
be the area of the region under the parabola, from 0 to x. Thus F(I) is the area under
the parabola from 0 to 1 ; this is
k k
F(l) =- · 1 3 =- .
3 3
F( 3) is the area under the parabola from 0 to 3; this is
F(3) = � · 33 = 9k.
3
And so on. Thus we have a function F: R+-+ R+. And we have a formula giving
the values of F:
F(x) = � x3•
3
Here we have replaced h by x in the area formula
k
3
A"=- h.
3
Thus, starting with the nonnegative function
we have defined a new function
. 3 k
-x.
F.x1--+
3
For each x, the value of the new function is the area under the graph of the old one.
We can generalize this scheme in the following way.
y
y=f(t)
n/l
I I
I
I
I
I I
I
I
I
I I I
a x b
Given a continuous nonnegative function/, defined on an interval, [a, b]. As before,

we label the horizontal axis as the t-axis, because we want to use the symbol x for
another purpose. For each number x on the interval [a, b], let R., be the shaded region.
Thus
R.,= {(t, y) I a � t � x, 0 � y � f(t)}.
And let F(x) be the area of the region R.,.
This is the scheme that we used for the parabola. For the parabola, we had
f(t)= kt 2, and we used a= 0. But we can go through the same proceeding starting
with any number a and any continuous function/ Let us look at some more examples.
Consider
f( t )= t + I, t � 1.
This is a function
f: [1, oo)--+ [2, oo).
y
y=t+l =f(t)
4
(x, x+I)
I
I
I
3 I
I
1
I
2- I
/I I
// I I
1 / I I
/ I I
// I
I
I
/ I
t
a=l x
For every x on the infinite interval [ I, oo ), let F(x) be the area under the graph, from
1 to x. We now have a new function
F: [l, oo) --+ [O, oo).
In this case, it is easy to write a formula for F. For x= I, the area is 0. Therefore
F(l)= 0. For x > I, F(x) is the area of a trapezoid lying on its side, with its "bases"
vertical. The altitude is h= x - I, and the lengths of the bases are b1= 2 and
b2= x + 1. Therefore
F(x)= i(b1 + b2)h H2 + x + l)(x - 1)

=
= Hx + 3)(x - I)= t(x2 + 2x - 3).

Let us now try the parabolay = t2, taking a= -2:
y = f(t) = t2, t � -2.

Here we have
f: [-2, oo)--+ [O, )
co .
·y
-2 x
For each x � -2, F(x) is the area under this graph, from -2 to x. By our old for-
mula,
F(x) tx3
= t(-2)3 = tx3 + l
-
In all these situations, F(x) is determined by (a) the given function f, (b) the
number a, and (c) the number x. All this is conveyed by the notation
F(x) = Li(t) dt.

That is, if/ is continuous and nonnegative, and a � x, then
is the area under the graph of f, from t = a to t = x.
y
y
fi �J(t)
I
I
1x f(t) dt I
I
I a I
I I
a x
Here it should be understood that f is continuous on an interval containing a

and x. The expression
is called the integral from a to x of the function f The number a is called the lower
limit of integration (or, briefly, the lower limit) and x is called the upper limit of
integration (or simply the upper limit). The function f is called the integrand. The
notation for the integral may look formidable at first, but it is not hard to learn and
is convenient: it shows at a glance that we are taking the integral, of a certain function,
between certain limits.
We proceed to generalize these ideas in two ways.
a) Suppose that f is negative, for some values of t. In this case, areas below the
t-axis are counted negatively. For example, in the left-hand figure below, Ai and A2
are positive numbers, representing the areas of the two shaded regions. We count
Ai positively; it is the area of a region above the t-axis. We count A2 negatively;
it is the area of a region below the !-axis.
y y
Thus we have
Similarly, in the figure on the right,
b) So far, we have required a < x. If a > x, we first find
J:f(t) dt,
and then reverse the sign. Thus, in the figure on the left below,
under our old definition. And
under our new definition.

(x, 2-x)
y 4
y=f(t)
x?
a
Consider, for example, f(t) 2 - t, defined for every t (in the right-hand
x ;:;;; 2
=
figure above). Take a = - 1 . Then, for -1 � we have
L: f(t) dt = J�(2 - t) dt = t(x + 1)(3 + 2 - x) = Hx + 1)(5 - x),
by the formula for the area of a trapezoid.

For x�2 we have
J"'(2
-1
- t) dt = t · 3 · 3 - t(x - 2)(x - 2)
For x;:;;; -1 we have
J"'c2
-1
- t) dt = -r-\2J.,
- t) dt
= -[t(-1 x)(2 - x - + 3)]
= Hx - x).
+ 1)(5
You should check that these are the right answers for the three cases. In each
case, we have computed areas of triangles and trapezoids by elementary area formulas,
and then attached the correct sign to the area of each region.
PROBLEM SET 3.7
1. a) Consider f(t) = It!. Get a formula for
f ltldt,
valid when x ;;;; 0. Sketch.

b) Now get a formula for
valid when x � 0. Sketch.

c) Now get one formula for the same integral, valid for every x.
d) Let
F(x) = f' t
l l dt.
Get a formula for F'(x), valid when x > 0.

e) Now get a formula for F'(x), valid when x < 0.
f) Finally, get a formula for F'(x), valid for every x.
2. Do the same six things for
f'(t2 + 1) dt.
3. Consider the function defined by the graph below.
This function is called the signum. Algebraically,
when < 0,
sigt when = 0,
when > 0.
Obviously sig is not continuous at 0. But we define
f' sig t dt
in the same way as for continuous functions. For example,
i1 sig t dt = I;
here the integral is the area of a square of edge I. Similarly,
1-i sigtdt = - ( - !) = 1,
13 sigtdt = 3,
3.8 The Derivative of the Integral 109
f1 sig t dt -1 + 1 O;
J_l
= =
and so on.
Do the same six things for f (t) sig t that you did forf (t)
= = It/ and f (f) = t2 + 1.
4. Do the same six things, for f(t) t It!.
=
5. a) Explain why
( t3 dt
1 0.
J_l
=
b) Explain why
(1 t273 dt 0.
J_l
=
c) Let f be a cubic polynomial, and suppose that

I (-a) = I (0) = I (a) = 0.
Show that
fa f(t) cit = o.
6. Explain why
0 <
J3 -dt 1 14
< - .
1
t 12
7. Explain why
3 1---+t2 dt
J
1 3
0 < < 5.
1
3.8 THE DERIVATIVE OF THE INTEGRAL
In Problems 1 and 2 above, you found that if
F(x) = {xf(t) dt,

then
F'(x) = f(x)
for every x. That is, at each point x the derivative of the integral function Fis simply
the value of the integrand function f.
In fact, it is not hard to convince ourselves that if f is a continuous function, this
is what always happens. Consider first the case in which f is positive.
y
y y=F(x)
y=f(t)
Take a fixed x0• By definition,
F(x) - F(x0)
F'(x0 ) -_ 1.tm .
o:->o:0 X - Xo
Now
F(x) = Lf(t) dt and F(x0) =
f"•f(t) dt.
a
Therefore
F(x) - F(x0) =
.
Cf(t) dt - L f(t) dt.
"
For x0 < x, as in the figure below,
"
F(x) - F(x0) ( f(t) dt.
J..
=
Since f is continuous,
F(x) - F(x0) � (x - x0)f (xo).
Here"�" means "is approximately equal to." We are claiming that the area under
the curve, from x0 to x, is closely approximated by the area of a rectangle with base
x - x0 and altitude f(x0). Therefore
F(x) - F(x0)
� f(Xo·'
) ·
X - Xo
and the approximation gets better as x gets closer to x0•
y
y=f(t)
a x
For the situation shown in the figure, in which the graph rises to the right of x0,
it is easy to see why the approximation is good. Here we have
F(x) - F(x0) = (x - x0)f (x0) + E,
where E is the area of the little curvilinear triangle at the upper right. Now
---
E
E < e(x - x0) and < e,
X - X0
where e is the altitude of the curvilinear triangle. Thus, when we write
F(x) - F(x0) !"'3 (x - x0)f (xo),

the error in the approximation is E; and when we write
F(x) - F(x0) ,,,..,

"" f(Xo) '
X - Xo
the error in the approximation is Ej(x - x0) , which is less than e.
a x x0
If x < x0, the same approximation formula holds, although the reasons are
slightly different. Here the area under the curve from x to x0 is
F(x0) - F(x);
the area of the rectangle is
( x0 - x)f (x0);
these are approximately equal. Changing the sign of each, we get
F(x) - F(x0) ,,,_,
F(x) - F(x0) !"'3 (x - x0)f(x0), and rv f(Xo'

)
X - Xo
as before.
But the fraction on the left is the slope of the secant line to the graph of F. The
limit of this fraction is F' (x0); and since the fraction is close to f (x0) when xis close
to x0, we ought to have
F'(x0) = f(xo).
y
F
If this is true, then we have:
Theorem 1. If f is continuous on an interval containing a, then
D., f'f(t) dt = f(x)
at each point x of the interval.

In fact this is true, and can be proved by a more careful use of the ideas that we
have just been describing informally. But let us postpone the proof until the end of
this chapter, and see, in the meantime, what the theorem is good for. Consider the
following problem:
Problem I. Calculate the area under the graph of y = x4, from x = 0 to x = I.
To solve this problem, the first step is to realize that whoever proposed the
problem has asked the wrong question: the answer to his question is a number, and
there is nothing about this number that is easy to see.
y y
y=x4 y=f(t) =t4
The easiest way to solve Problem 1 is to consider instead the following:
Problem 2. Find a formula for the function
F(x) = f t4 dt.
It might seem that Problem 2 must be harder, but this is not true. The point is
that, while information about the number F(l) is hard to come by, Theorem 1 tells us
something about the function F, namely,
F'(x) = x4•
We now ask ourselves what sort of function has x4 as its derivative. We get powers
of x by differentiating powers of x, using the formula
Thus
This is like F'(x), except for the factor 5. But the 5 is easy to get rid of: we divide
x5 by 5, getting
D(tx5) = t . 5x4 =x4.
Thus we have found a function

G(x) =-kx5,
which resembles the given function
F(x) = f't4 dt.

To be exact:
G'(x) =F '(x) for every x, (1)
G(O)=F(O). (2)
To see why (2) holds, we observe that
G(O) = t05 = O
and that
F(O) = f t4 dt = 0.
Equations (1) and (2) ought to guarantee that
G(x) =F(x) for every x;

that is,
G=F.
The functions F and G start with the same value, at x= 0. And (1) tells us that F
and G always change at the same rate. This suggests the following:
Theorem 2. (The uniqueness theorem.) Let F and G be differentiable functions,

defined on the same interval I, and let a be a point of I. If
F(a) = G(a) (3)

and
F' (x) = G' (x) for every x in I, (4)
then
F(x) = G(x) for every x in I. (5)
Here we call the interval I because we want to allow intervals of all kinds, including
[a, b], [a, b), [a, oo), ( -oo, a], and so on. We also allow the case I= ( - oo, oo).
This is the case for the functions
F(x) = f 't4 dt, G(x) = tx5
that we have been discussing in the last few pages.

The uniqueness theorem is a consequence of the mean-value theorem (MVT)
of Section 3.2. The proof is as follows.
Suppose that F(b) � G(b)for some bon the interval I. For each xon [a, b], let
H(x) = F(x) - G(x).
Then H(a) = 0, but H(b) � 0.
The slope ofthe chord joining the endpoints ofthe graph of His
H(b) - H(a)
� O.
b-a
By MVT there is an x between a and bsuch that
= H(b)-H(a)
H'(x) .
b-a
Therefore H' (x) � 0. But this is impossible: for every xwe have
H'(x) = F'(x) - G'(x) = 0.
Theorems 1and 2,in combination,enable us to solve some difficult area problems.
Example 1. Find the area ofthe region above the xa

- xis and below the graph of
f(x) = 1 -x2•
y
Obviouslythis area is
Let
F(x) = 1:(1 - t2) dt.

Then
F'(x)= I - x2,
by Theorem 1. It is easy to find another function with this derivative, namely,
G(x) = x - tx3•
Now F(-1)=0, and
G(-1) = -1 + t = -i. (?)
But this is easy to fix: we change our minds and write
G(x) = x - tx3 + i.
Then
G(-1)=0,
as it should be; and by the uniqueness theorem it follows that
G(x) = F(x)
1: (1 - t2) dt = F(l) = G(l) = 1 - -} + i = f.
Example 2. Find the area under the graph of y = x2 + x + 2, from x = -1

to x = 2.
: 1
-1 1 2
Here the area is
fy2 + t + 2) dt.
Let
F(x) = J�y2 + t + 2) dt.

Then
F'(x) = x2 + x + 2.
As our first guess, let G(x) = }x3 + }x2 + 2x, so that G'(x) = F'(x). We find that
G(-1) -t + t - 2
= = -1-l-. Therefore we really want G(x) = tx3 + t x2 +
2x + ll, so that G( -1) = 0. This gives the answer in the form
r�t2 t + + 2) dt = F(2) = G(2)

.
=
l 8 + t .
4 + 2 . 2 + 1-l =
v- =
J,_l
The same scheme can be used to calculate integrals in which the integrand is
negative and hence does not represent an area. Given
F(x) =
f'f (t) dt,
we first find (if we can) another function G such that
G'(x) = F'(x) = f(x).

We then arrange for G(a) to be 0, by adding a suitable constant to the first G that
!Jf(t) dt
we tried. We then know that G(x) F(x) for every x. Therefore
=
J a = F(b) = G(b),
in the same way as for positive functions, and for the same reasons.
PROBLEM SET 3.8
I. By the methods of this section, find the area under the graph of y x - x3, from =
x 0 to x
= 1, and sketch. (Here, and hereafter in this problem set, you should
=
explain what functions you are using as the functions/, F, and G.)
2. Find the area of the region lying below the x-axis and above the graph of y x4 I. = -
Note that here the function f is negative, so that the area and the integral are different;
the area A is positive, and
J_� f (t) dt = -A.
3. a) Find the area under the graph of y = x10, from 0 to b.

b) Find the area under the graph of y = x10, from a to b. (Q11e1y: Do you need to give
separate discussions for the cases 0 < a < b, a < 0 < b, and so on?)
c) Same as Problem 3(b), for the graph of y x100. =
d) Now find a general formula for
f' tndt,
valid for every positive integer 11.
4. Find
a) fu2 + 2t + 5)dt b) f cx2 + 2x + 5)dx c) J

0
1
(z2 + 2z + 5)dz
d) A general formula for
F(x) J:(t2 + 2t + 5) dt.

=
5. a) Find
f (t3 + t - 1) dt.
b) Get a general formula for
F(x) f + t - 1) dt.
= (t3
6. Get a general formula for
f'(t5 - 2t3 + 1) dt.

7. a) Find the area under the graph of
x
v' x2 + 1 '
y=
from 1 2.
to Unless you happen to remember a function whose derivative is
x/ v'x2 + 1, you are going to have to figure out how this function might arise as
the answer to a differentiation problem. The radical in the denominator suggests
that somebody has been using the formula
f'
Dv'f- =
2v'J'
-
b) Find
Jl t2t + 1 dt.
-1
__
v'
__
Then sketch the graph of
y = /(!)
v' t2 + 1 '
as well as you can, and explain how the numerical value that you got for the integral
could have been predicted, without any calculations at all.
8. Let
F(x) f(1 +
= v'()0 dt.
Express F'(x) by an elementary formula (that is, by a formula not involving integrals
or differentiations). Note that you are not being asked to express F(x) by an elementary
formula.
9. Same as Problem 8, for
F(x) r(1 + v'()500 dt.

=
10. Same as Problem 8, for

1
11. Find the area under the graph of y = _ 1 __ (and above the x-axis) from 0 to 1.
v x + 1
12. Same, for the function y = --

.
vz -x
13. Find
[Hint: This problem is easier if you forget the binomial theorem.]
14. Find
15. Find
2 x2
16. For n 7" 0, we have

Ji (1 + xa)2
dx.
Since n 7" 0, the function

1
f(x) = x-1 = -
never appears as the derivative of a power of x. If we allowed n = 0, then xn = xo 1 =
for x > O; the derivative is O; and l/x still does not appear. Thusf(x) l/x is not the
=
derivative of any integral power of x.

Question: ls there any function at all which has f(x) = 1/x as its derivative, say,
for x > 0? If so, what function?
* 1 7. Consider
If you attempt to evaluate this integral by applying the methods of this section in a
mechanical sort of way, you will get an "answer." If you try to interpret your answer
geometrically, you will see that your answer cannot possibly be right. What went wrong?
(Evidently we must have been trying to apply a theorem in a case in which its hypothesis
is not satisfied. The question is what theorem and what hypothesis.)
*18. In Theorem 1, suppose that we had omitted the hypothesis that/ is continuous. Give
an example to show that the resulting theorem would not have been true. [Hint: You
have already seen cases in which a function of the type
F(x) = ft (t) dt
fails to have a derivative at some point x0; and surely we cannot have
if there is no such thing as F' (x0).]

3.9 Uniformly Accelerated Motion 119
3.9 UNIFORMLY ACCELERATED MOTION
Suppose that a particle is moving, according to some given law, along a line . If we
think of the line of motion as the y-axis, then the motion can be described by a
function
f: I->- R;
for each time ton the interval,f(t) is they-coordinate of the moving particle at time t.
Thus, for example, in the figure below, the total time interval I is the closed interval
[t1, t4]. The figure tells us that, at the start of the motion, the time is t1 and the particle
is at the pointy = 1; in the time interval [t1, t2], the particle rises from 1 to 3; in the
time interval [t2, t3], the particle falls from 3 to -1; and in the time interval [f 3, t4],
the particle rises from -1 to 4.
-1
The figure shows a finite time interval I= [ti. t4]. More generally, the function
f may be defined on an infinite time interval I= [t1, oo) or I= R = (-oo, oo).
But most of the time, on or near the earth, the motion begins at some time t0, and
eventually the motion stops. The velocity is the function
v=f': I_.. R,
provided that f is differentiable. The acceleration is the function
a= v': I->- R,
provided that vis differentiable. Thus the acceleration is the derivative of the derivative
off. We call this the second derivative off, and denote it by f". Thus we can sum up:
v=/', a= v =f"I
(by definition).
Finally, there is a fourth function associated with the motion. This is the function
F: l->-R
which gives, for each time t, the force F(t) acting on the body at time t.
We shall now see what form these functions take when/ describes the motion of a
freely falling body. Before we can work mathematically on the problem, we have to
state our physical assumptions in mathematical form.
1) Newton's second law asserts that acceleration is proportional to force divided by

mass; that is,
F(t)
a(t) = k1- (k1 = const).
m
2) For a freely falling body (or a body projected vertically upward), the force is the
resultant of the weight (which acts downward) and the air resistance (which acts up
ward when the body is falling and downward when the body is rising). If the speed is
moderate, then the air resistance can be neglected. Hereafter, we shall assume that the
weight is the only force, so that
F(t) = W(t) < 0 for every t on [c, d],

where W(t) is the weight at time t. W(t) < 0 because weight pulls things downward.
3) Evidently, the weight will not change merely with the passage of time, but it will
depend on the altitude; the greater the altitude, the less the force of gravitation. But
if the altitude is not very great, then the weight will be very nearly constant. We shall
assume hereafter that the weight of a freely falling body is constant. Therefore
F(t) = W(t) = k2 < 0. Therefore
k2
a(t) = k1 · - < 0 for every t;
m
and
k3
a(t) = - < 0.
m
This last equation says that for each falling body there is a constant which is equal to
the acceleration, independently of the time.
4) There remains, however, a question: is there one constant which works for all
falling bodies, or does the constant acceleration depend on what sort of body is
falling? Conceivably, the law governing the free fall of heavy bodies (such as cannon
balls) might be different from the law governing the fall of light bodies (such as BB
shots). Jn fact, until the time of Galileo, everybody thought that heavy bodies fell
faster. The story goes that Galileo proved them wrong by dropping two iron balls
of different sizes off the leaning tower of Pisa: they hit the ground at the same time.
Since k3/m is independent of m, there is a constant -g = k3/m which gives the
acceleration of every freely falling body, regardless of its mass. The number
k3
g= --
m
is called the acceleration of gravity. If distance is measured in feet and time in seconds,
then numerically
g � 32, measured in ft/sec2•
The above discussion can be summed up as follows:
If we neglect air resistance and neglect the variation of weight with

altitude, then the acceleration of a freely falling body is given by the formula
a(t) = -g,
where g is a constant and
g � 32 ft/sec2•
We now consider the problem of finding the functions that satisfy the equation
a(t) = J"(t) = -g.

Problem. The function
f: R---+R
has the following properties:
a) f"(t) = -g for every t,

b) f'(O) is a given number v0,
c) f(0) is a given number Yo·
What is f ?
Using the notation v for f' and a for v' = f", we write these conditions in the
form:
a) = -g for every t,
a(t)
b) v(O) v0,
=
c) f(O) =Yo·
Thus our data consist of (a) the constant acceleration -g, (b) the initial velocity
v0 = v(O), and (c) the initial position y0 = f(O). The solution is as follows:
a) We know that
v'(t) = a(t) = -g
for every t. The function
u(t) = -gt (?)
has -g as its derivative; the only trouble is that u(O) is 0 instead of v0• But this is
easy to fix: we change our minds and let
u(t) = -gt + v0•
Our function u then has the same derivative as v, and has the same value at t = 0.
By the uniqueness theorem, u and v are the same function, and so
v(t) = -gt + v0•
b) We know now that

f'(t) = v(t) = -gt + v0•
We want to find/ Now the function
z(t) = - � t2 + v0t (?)

2
has -gt + v0 as its derivative; the only trouble- is that z(O) is 0 instead of y0• But
this is easy to fix: we change our minds and let
g
z(t) = - - t2 + v0t + Yo·
2
The function z then has the same derivative as f, and has the same value at t = 0.
By the uniqueness theorem, f and z are the same function, and so
f(t) = - � t2 + v0t + Yo·

2
This completes the solution. We sum up in the following theorem:
Theorem 1. Let f be a function R __,,. R. If

a) f"(t) = -g for every t,
b) f'(O) = v0, and
c) f(O) =Yo·
then
d) f(t) = (-g/2)t2 + v0t +Yo for every t.
Thus the mathematical problem defined by (a), (b), and (c) has only one solution.
This fact is important in applications, because, if our mathematical problem had two
solutions, we would have to find out which of the two solutions applied to the
But, if f"(t)
physical situation that we started out to investigate. -g, f'(O) =
=
v0 = 10, andf(O) = v0 = 5, thenfmust be the functionf(t) = (-g/2)t2 + lOt + 5.

Theorem 1 can be stated in a more general form. If I is any time interval whatever
(finite or infinite) and t0 is any point of I, then we can consider a function
f: /_,,.R,
such that
a) J"(t) = -g for every t,
b) f'(to) = Vo ,
c) /(to) =Yo·
The uniqueness theorem applies to our problem in exactly the same way as before,
and the algebra is only slightly more complicated. We are given
v'(t) = -g.
We try
u(t) = -gt (?);
we observe that u(t0) = -gt0 instead of v0; to fix this, we let
u(t) = -gt + gt0 + Vo·
Now u has the same derivative as the unknown function v, and has the same value
at t = t0• By the uniqueness theorem, u(t) = v(t) for every t, and so
v(t) = - gt + gt0 + v0•

This solves half of our problem.
We know that/'(t) = v(t). We therefore try
z(t) = - � !2 + gt0t + V0t (?);

2
we observe that z(t0) = (g/2)t� + v 0t0 instead of y0; and we fix this by letting
g g
z(t) = - - t2 + gt0t + v0. t - - t20 - v0t0 + Yo
2 2
- � (t - t0)2 + Vo(t - to) + Yo·

2
Then z has the same derivative as/, and has the same value at t0. By the uniqueness
theorem it follows that z{t) = f(t) for every t. Therefore
f(t) = - £ (t - t0) 2 + v0(t - t0) + Yo·

2
None of these formulas should be learned. What you need to learn is the process by
which they were derived; if you remember the method, you can use it. For example:
Problem. Given f"(t) = 3,/'(3) = 1, and/(3) = 2, what is/?
Solution. Let v = f'. Then v'(t) = 3. This suggests that v(t) = 3t. Adding the
appropriate constant, to get v ( 3 ) = 1, we obtain
v(t) = 3t - 8.
Now
f'(t) = 3t - 8.
This suggests f(t) = ft2 - 8t. Adding the appropriate constant, to get f (3) = 2,
we have
f(t) = ft2 - 8t - t . 32 + 8 . 3 + 2 = it2 - 8t + ¥ . - -
This is the answer. (Two differentiations verify that it is an answer; and two applica
tions of the uniqueness theorem tell us that it is the only answer.)
PROBLEM SET 3.9
Find formulas for the unknown functions, under each of the following sets of conditions.
In all but one of these problems, the conditions are enough to determine the function.
In three cases, however, there are infinitely many possibilities; and in these cases you should
try to explain what the possible functions are.
1. /'(t) = 3t + 4,/(0) 4 = 2. f'(x) = x3 - 7x + 5,/(0) = -1
3. f"(t) = -1,/'(0) 2,/(0)

= = 3 4. f"(x) = 3x2,f' (1) O,f(1)
= = 1
3 , 1
5. J"(t) = t ,f'(O) = 1,/(1) = 0 6. f (x) = 2 ,/(1) 2
x
=
, x
7. g (x) = _ / ,g(O) = -1 8. g'(x) = x(x2 + 1)2, g(3) = 1
v 1 - x2
, 1
9. I (t) = t ,/(2) = 5
V
10. f'(t) = t2(1 + t3)10,f(O) = 2 (By all means, do not use the binomial theorem on this
one.)
11. j'(t) = t2 + 1,/(1) = 2 12. f"(x) = x,/(1) = O,f'(I) = I
t2 1
13. f'(t) = + 3 ,/(1) = 1 14. g"(t) = + l 3,g(O) = l,g(I)= 1.
(I t )2 (t )
15. A "theoretical projectile" is fired vertically upward, from the surface of the earth, at
time 0, with initial velocity 10 ft/sec. When will it hit the ground again? For what time
interval is its motion described by the condition a(t) = g?
- (Following the advice
given at the end of this section, you should solve this problem with your book closed,
using the methods but not the results given in the text.)
I 6. A "theoretical projectile" is fired vertically upward, from the surface of the earth, and
hits the ground again ten seconds later. What was the initial velocity?
17. A "theoretical projectile" is fired vertically downward from the top of a 200-foot
building and hits the ground 2 seconds later. What was the initial velocity?
18. We state this problem in a nonmilitary form. A billiard ball is raised to a certain height
y0 and simply dropped, so that it begins its free fall at velocity v0 = 0. Five seconds
later it hits the ground. What was y0?
19. Free fall near the surface of the moon works the same way as free fall near the surface
of earth, except that the constant acceleration -gL (L for lunar) is different; the smaller
mass of the moon makes the difference. Suppose you went to the moon, dropped a
billiard ball as in Problem 18, and found that it dropped 3 feet in one second. What
could you conclude aboutgL?
*3.10 PROOF OF THE FORMULA

FOR THE DERIVATIVE OF THE INTEGRAL
We shall now prove Theorem I of Section 3.8. We have a continuous function f;

we let
F(x) = ixf(t) dt;

we take a point x0; and we want to show that
Let e be any p ositi v e number. Since f is continuous, we know that the graph off
has an eb-box at the point (x0,f(x0)).
3.10 Proof of the Formula for the Derivative of the Integral 125
f(xo)+l---------�--+--�---,
a x0-5 Xo
Thus
Ix - x0I < a => If(x) - f(xo)I < E
<=> f(x0) - E < f(x) < f(x0) + E.
We are going to use these inequalities to get information about the function
1
m(x) = -- [F(x) - F(x0)].
X - Xo
Here m is the slope function for the function F, so that lim,,_.,0 m(x) = F'(x0).
Evidently
F(x) - F(x0) = i "'f(t) dt - f f(t) dt
= f !(t) dt,
:ro
and so
m(x) = __ l
x - X0
[F(x) - F(x0)] = _ l_
X - Xo
r
"'o
f(t) dt.
If f is positive and x0 < x, as in the figure, then F(x) - F(x0) is the area of the
shaded region.
Case 1. Suppose that Xo < x < Xo + a, as in the figure. Then
f(xo) - E < f(t) < f(xo) + E (x0 ;;; t ;;; x).

Therefore
and so
[f(x0) - E:](x - x0) < ff(t) dt

'
"' •
< [f(x0) + E:](x - x0).
Dividing by the positive number x - x0, we get
f(x0) - E < m(x) < f(x0) + E.

Thus we have shown that
x0 < x < x0 + o => If (x0) - m(x)I < E. (.1)
Case 2. Suppose that x0 - o < x < x0• Then
f(x0) - E < f(t) < f (x0) + E (x � t � x0),
just as in Case 1. Therefore
"'0 "'0
i [f(xo) - €] dt J:f(t) dt i
< < [f(x0) + €] dt.
(We are integrating from left to right.) Therefore
[f(x0) - €](x0 - x) < I:f(t) dt < [f(x0) + E:](X0 - x).
Since x0 - x > 0 in Case 2, we can divide by x0 - x, preserving these inequalities.

This gives
f(x0) - E < --
1
X0 - X "'
J"'•f(t) dt < f(x0) + €.
When we interchange x and x0, this changes the sign of each of the factors in the
middle of this expression. Therefore
To sum up:
f(x0) - E <
X -1 Xo l"'f(t) dt
--
a:0
< f(x0) + €.
X0 - o < x < X0 => /(x0) - E < m(x) < f(x0) + E => lm(x) - f(x0)1 < E,
(2)
exactly as in Case 1. Fitting together our results in Cases 1 and 2, we get
0 < Ix - x01 < o => lm(x) - f(x0)1 < E.
Therefore
1
lim m(x) lim [F(x) - F(x0)]
a::-.a::o
=
a::-t-a:o
--
X - Xo
This proof is not easy, but it might have been worse. It was made simpler by the
fact that for each E > 0, the o > 0 that we get from the hypothesis lim.,_,.,0/ (x) =
f(x0) is precisely the o that we need, to conclude that lim.,_,.,0m(x) f(x0). =

3.10 Proof of the Formula for the Derivative of the Integral 127
PROBLEM SET 3.10
Find the first and second derivatives of the following functions.
1. f(x) = r t2dt 2. g(x) f' (t3 1) dt

= +
4. {"' � dt
3. h(x) =
J("'2,, (t4 - t)dt f(x)

Jo t
= +
l
5. g(x) = r4 + t8dt
Vl 6. h(x) i v t2 + 1dt
=
7. f(x) =
J("'4,, 1dt
v( 8. g(x) f (1 + t3)100dt
=
9. {"' dt 10. f(x) J: v2 + tdt

Jo 1 + t2
h(x) = -
2
=
11. g(x) = f (t2 + 1) dt

v! 12. h( ) x
f "'
-x=
1 t 4dt
l
+
� dt
-
13. 14. J1x t3 dt+ 1

15.
iJ,,x 4dt 16.
Jx -1t2++t110dt
2 t +1 1T
2
17. J"' J 1 + tdt 18. {"' J 1 + t dt
-1 1 t
- Ja 1 + t4
19. If you know that
D., f /(t)dt = f(x),
for every continuous function f, this does not immediately enable you to find the
rJo2x
derivative of
f(x) = Vl + t8dt.
But find the answer f', by any method.
*20. Findg'(x), given
g(x) =
(
Jo
"' ' vl + t8 dt.
Trigonometric and
4 Exponential Functions
4.1 DIRECTED ANGLES. TRIGONOMETRIC

FUNCTIONS OF ANGLES AND NUMBERS
In elementary geometry, when we speak of an angle we simply mean a geometric

figure, that is, a set of points:
--+ -+
A
v
If AB and AC are rays which have the same endpoint A, but do not lie on the same
line, then their union is the angle LBAC. (In the figure, the arrowheads remind us
that the sides of an angle are rays rather than segments.) Some authors define the
word angle in such a way as to allow "zero angles" and "straight angles."
A C B B A c
In any case, in elementary geometry the idea of an angle does not include the idea
of order; the sides of an angle are not arranged in an order, any more than the sides
of a triangle are.
Terminal Initial
OL.
LAOBA 0 L.
LBOA A
In trigonometry, however, the order of the sides of an angle makes a difference.
Henceforth, whenever we speak of an angle we shall mean a directed angle. Thus,
-+ --+ -+
in the figures above, LAOB is an ordered pair of rays (OA, OB); OA is
the ray the
--+
initial side, and OB is the terminal side. Thus LAOB is different from LBOA.
128
4.1 Directed Angles. Trigonometric Functions of Angles and Numbers 129
Wf suppose that a coordinate system is given in the plane. The counterclockwise

direction is the direction from the positive x-axis to the positive y-axis, as shown
below. The counterclockwise direction in a coordinate plane is regarded as positive;
and the clockwise direction (running the other way) is regarded as the negative direc
tion.
A new coordinate system is called right-handed if it gives the same counterclock
wise direction; otherwise it is called left-handed. In the figure below, the right-handed
coordinate systems are marked R, and the left-handed ones are marked L.
·+
x
y
Lx y
x x
R
x x
"X
x
L
We can now define the trigonometric functions of an angle LAOB. The procedure
--+
is as follows. We set up a right-handed coordinate system, in which the initial side OA
--+
is the positive half of the x-axis. On the terminal side OB we choose a point P ¢ 0.
P has coordinates (x, y), in the coordinate system that we have set up, and the distance
OP is a positive number r.
It is easy to show (by similar triangles) that the ratios xfr, yfr, yfx, xfy, r/x, r/y are
independent of the choice of P; they depend only on the angle that we started with.
Thus we can define the trigonometric functions of LAOB as follows:
sin LAOB = y/r, cos LAOB = xfr,

tan LAOB = yfx (for x ¢ 0), cot LAOB = x/y (for y ¢ 0),
sec LAOB = rfx (for x ¢ 0), csc LAOB = r/y (for y ¢ 0).
130 Trigonometric and Exponential Functions 4.1
We have defined six functions. Note that the domains of these functions are not
sets of numbers, but sets of angles.
Consider now the unit circle C, with center at the origin, in the xy-plane.
-1 0
P, %
P0 1
x
-1
Let P0 be the point (1, 0), as in the figure. To each real number fJ there corresponds a
point P0 of C, under the following rules:
1) Given fJ > 0, we start at P0 and move around C in the counterclockwise direction

until we have traced out a path whose total length is fJ. The point where our path
ends is P0•
2) Given fJ < 0, we start at P0 and move around C in the clockwise direction, until
·
we have traced out a path whose total length is lfJI. The point where our path ends.
is P0•
These rules define a function
w: R-+ C
fJ f-'>P0 == w(fJ),
under which to each real number fJ there corresponds a point of C. The function w
is called the winding function. Note that the values of the function w are points
rather than numbers. Note also that
Po+2" ==Po,
for every e. The reason is that when we add 27T to fJ, this merely means that we
take another round trip around the circle, ending at the same point P0 where we
began. Similarly,
and
for every integer n, positive, negative, or zero.

We shall use the winding function to define trigonometric functions of numbers,
in the following way. For each number fJ, let
LO== LP00P0•
The symbol L() is pronounced "angle () ; " L() is the angle which corresponds to the
number e. We now define
sin()= sin L()= sin LP00P0,

cos() = cos L()= cos LP00P0,
and so on.
We have defined these functions in terms of L() because we want to emphasize

their geometric meaning. But for some purposes, it is simpler to forget about angles,
and merely use the coordinates of P0• If
Po= (xo, Yo),

then
sin() = y0,
COS()= Xo,
tan() =
Yo (whenever x0 � 0),
Xo
Xo 0),
cot() = (whenever Yo �
Yo
1
sec()= - (whenever x0 � 0),
Xo
csc () = _!_ (whenever Yo � 0).
Yo
Using these definitions, we can derive the usual formulas. Since P0 is on the unit
circle C, we know that OP0= 1. Therefore
x� + y�= 1,
and we have:
Theorem 1. For every (),

cos2 () + sin2 e= 1.
If the sign of 6 is changed, this sends us around the circle C in the opposite
direction. Therefore the points P8 and P_8 are symmetric across the x-axis, as in the
figure.
y
Pe
/1
/ I
/ I
/ I
/
Therefore
Y-o = -Ye·
This gives:
Theorem 2. For every 6,

sin (-6) = -sin 6, cos c -6) = cos 6.
Plotting the points P0, P1112 and P11, we get the following:
,
Theorem 3.
sin 0 =
0, cos 0 =
1,
•
7T '!!.
Sln -
=
1, cos = 0'
2 2
sin 7T = 0, cos 7T = -1.

Theorem 4. For each 6,
sin (7T + 6) = -sin 6, cos (7T + 6) = -cos 6.
Proof For each (), the points P9 and P1T+9 are symmetric across the origin. This
holds in all quadrants. Therefore
and the theorem follows, by definition of the sine and cosine.
In the kind of trigonometry that we are dealing with now, the relation between
angles and numbers is a little tricky. If () is known, then P9 is determined, and so
L() is determined; L() is LP00P8•
y y
But if the angle is known, the number() is not determined. In the figure on the right,
LP00Q is given, but for this angle we may have
() = !7T, or () = 21T + !7T = \1 7T, or () = t7T - 21T = -i1T.
In fact, for every integer n, positive, negative, or zero, we may have
() = !7T + 2n1T.
If an angle LAOB corresponds to a number(), under the rules that we have been
giving, then we shall say that LAOB has measure (), and we shall write, for short,
LAOB = L().
(We have seen that every angle LAOB has infinitely many measures e. For
this reason, it would be misleading to speak of "the measure of an angle.")
So far, we have used the notation L() only for angles "in standard position,"
that is, angles with the positive half of the x-axis as initial side. But it will be con
venient to use the same shorthand for angles in general. Thus
LP00Q = L �, and LP 0Os =

3
L 7T.
4
R x'
'
But if we set up new axes x', y , we can also say that
LQOS = L 7T - 2
, and LQOT = LTT.
Using other axes, not shown in the figure, we see that
LSOP0 = L (- ) 37T ,
4
and so on.
PROBLEM SET 4.1
Derive the trigonometric identities given or suggested below. The derivations should
be based on the definitions and theorems given in this section of the text.
1.
sin 8
2.
cos 8
3.
tan x
4.
cotx
5. --1
secy
sin 8 cos 8 sec 8 csc 8
6. 7. 8. 9. 10.
csc z cos 8 sin 8 csc 8 sec 0
11. 1 + tan2 8 12. cot2 8 + 1 = 13.

cscx
sec x
14. sec2 8 - 1 =
1 5. tan( -8) = 16. cot (-8) = 17. sec(-6) 18. csc (-8) =
19. tan(7T + 8) 20. cot(7T + 6) 21. sec(7T + 6) 22. csc(7T + 8)

-
=
23. sin(7T - 6) = 24. cos(7T 8) = 2 5. tan(7T - 6) = 26. cot(7T - 8) ='
27. sec ( 7T - 6) = 28. CSC(7T - 6) =
29. a) Show that for every 8, 6 0, we.have
!sin 8 - sin 801;:;; 10 - 001.

b) Show that the sine is a continuous function.
30. a) Show that for every 8, 80, we have
jcos fJ - cos fJ01 ;:;; lfJ - e01.

b) Show that the cosine is a continuous function.
4.2 The Law of Cosines and the Addition Formulas 135
4.2 THE LAW OF COSINES AND THE ADDITION FORMULAS
Jn the figure on the left below, We have X9 =COS 6, andy9 =sin 6, by definition Of
the sine and cosine.
y y
P(x, y)
More generally, we have:

---+
Theorem 1. Let P be any point of OPe, and let OP=a. Then the coordinates of P
are
x=a cose, y=a sine.
Proof By similar triangles,
- lxel
lxl =- and
'
a 1
Therefore
lyl =a IYol·
In these equations x xe also agree in
and sign, and similarly for y and y9• Therefore
x=axe= acose, y=aye=a sine,

which was to be proved. From this we get immediately:
Theorem 2. (The law of cosines). If LACE = L8, then

c2 =a2 + b2 - 2ab cos e.
(The notation is that of the following figure.)
y
Proof By the preceding theorem,
B = a
( cose,a sine).
And obviously
A = (b, 0).
Therefore, by the distance formula,
c2 = (a cose - b)2 + (a sin e - 0)2

=a2 cos2e - 2ab cose + b2 + a2 sin2e
= a2( cos2e + sin2e) + b2 - 2ab cose

= a2 + b2 - 2ab cose'
Theorem 3. For every e and cp,
cosce + cp) = cose cos ·
Then
A= 1
( ,0),
c =(cos e
( + rp), since + rp)),
and so by the distance formula
AC2 = [cos e
( + cp) - 1]2 + sin2 e
( + cp)
= cos2ce + rp) - 2 cos e
( + rp) +1 + sin2 e
( + cp)
= 2 - 2 cosce + r/J).
4.2 The Law of Cosines and the Addition Formulas 137
---+
We now set up a new coordinate system, with OP6 as the positive x'-axis.
In the new coordinate system,
A=P_8 = (cos (-8), sin (-8)) =(cos 8, -sin 8),

C =Pq, = (cos cf>, sin cf>).
Therefore, by the distance formula,
AC2 =(cos 8 - cos c/>)2 + (-sin 8 - sin c/>)2
= cos2 8 - 2 cos 8 cos cf> + cos2 cf> + sin2 8 + 2 sin 8 sin cf> + sin2 cf>
= 2 - 2(cos 8 cos cf> - sin 8 sin cf>).
But the distance AC is independent of the coordinate system. Therefore
2 - 2 cos (8 + cf>) = 2 - 2(cos 8 cos cf> - sin 8 sin cf>),

and
cos (8 + cf>) = cos 8 cos cf> - sin 8 sin cf>,
Once we have the addition formula for the cosine, it is easy to get similar formulas
for the other trigonometric functions.
Theorem 4. For every 8 and cf>,
cos (8 - cf>) = cos 8 cos cf> + sine sin cf>.
Proof Using -cf> for cf> in the preceding theorem, we get
cos (8 - </>) = cos e cos (-cf>) + sine sin (-cf>).
But we know that
cos (-cf>) = cos cf>, and sin (-cf>).=-sin cf>.
Using these, we get the desired formula for cos (6 + cf>).

Theorem 5. For every 8,
cos (� - ) e =sine, and sin (� ) -e =cose.
Proof. By Theorem 4,
cos (� )
-e =cos 7!. cose
2
+ sin 7!. sine
2
=0. cose + 1 . sine,
by Theorem 3 of Section 4.1. Therefore
cos (� )-e = sine.
Using 7r/2 - 8 for 8, we get
cos [� - (� - ) J e =sin (� ) - e .
Therefore
sin (� )
-e = cose.
(The name of the cosine is a reference to this theorem; the word cosine is from
the Latin complementi sinus, meaning sine of the complement.)
Theorem 6. For every fJ and cp, sin (fJ + cp) =sin fJ cos cp + cos fJ sin cp.
Proof.
sin(fJ + cp) =cos [� - (fJ + J cp)
=cos (� - fJ) cos cp + sin (� ) - fJ sin cp
=sin e cos cp + cose sin cp.
PROBLEM SET 4.2
tan A +tan B
1. tan (A + B) = 2. tan (A - B) =
l - tan A tan B
cote cot</> - 1
3. cot (8 + </>) = 4. cot (A - B) =
cote + cot</>
5. sin 28 = 2 sine cose 6. cos 28 = 2 cos2 e - 1
7. cos 2li = 1 - 2 sin2 IJ 8. cot (Ii - </>) =

4.3 The Derivatives of the Trigonometric Functions 139
31T
9. a) sin-=
2
b) sin (3 7T + o)
2
31T
10. a) cos2 = b) cos (3 7T + o
2
) =
31T
11. a) tan-=
2
b) tan (3 7T + e
2
) =
o (}
-
0 l + cos 20 =
12. 2 sin - cos- =
2 2
13. 2 cos2
2
- 1 14.
J 2
l + cos(} l - cos 20
�
15.
J 2
= 16.
J 2
= 17. =
(} sin(} [Hint: Let <P = (}/2, so that 8 2</i, and rewrite the formula
18. tan -
=
2 + cos(} in terms of <fi. Then prove it.]
(} 1 - cos(}
19. tan - =
2 sin(}
*20. Show geometrically (without using any of the theory developed in this section) that the
formula in Problem 18 holds whenever 0 is between 0 and TT. Discuss the problem of
extending the formula from this special case to the general case.
21. Show that there is no formula which expresses sin (0/2) in terms of sin e. That is,
show that sin ((}/2) is not determined if only sin(} is known.
2 2. Find a formula which expresses !sin ((}/2)1 in terms of cos 0.
23 . Show that there is no formula which expresses sin(}in terms of tan e. That is, show that
tan(}does not determine sin 8.
24. Show that there is no formula which expresses sin ((}/2) in terms of sin (}and cos e.
25. Show that if Pe is known, then P3e is determined. [Hint: If Pe = P4,, what is the relation
between 0 and <P? In this case, what is the relation between 3(} and 34'? Between P3e
and P3q,?]
26. It is a consequence of Problem 25 that, if sin 8 and cos(}are known, then P38 is deter
mined, and therefore sin 3(}is determined. How ? That is, find a formula which expres
ses sin 3(}in terms of sin (}and cos e.
27. Can cos 3(}be expressed in terms of cos(}? If so, derive such a formula. If not, explain
how you know that no such formula exists.
4.3 THE DERIVATIVES OF THE TRIGONOMETRIC FUNCTIONS;

THE DIFFERENCES /lx AND llf; THE SQUEEZE PRINCIPLE
If we try, in a straightforward way, to find the derivative of the sine, we get into
trouble. By definition,
!'(
x0)
_
-
1. f(x) - f(xo)
1m ,
x->x0 X - Xo
if the indicated limit exists. For f (x) = sin x, this definition says that
.
Sln
, Xo =
1.
!ill
sin x - sin x0
,
x->x0 X - X0
if the indicated limit exists. In fact, the limit does exist. But it is not obvious what we
ought to do to this expression
sin x - sin x0
X - Xo
in order to find its limit. For functions f which were defined algebraically, we found
ways to cancel out x - x0 in fractions of the form
f(x) - f(xo)
X - Xo
using various algebraic tricks. Evidently some new device is needed for the sine.
It is as follows. Let
Ax= x - x0•
The symbol Ax is all one symbol. It is pronounced "delta x," and the Greek delta
is supposed to remind us that Ax is the difference in x. Obviously, x = x0 + Llx.
Similarly, let
Llf = f (x) - f (xo).
Here A/ is the difference in/, as we pass from x0 to x.

Geometrically, the use of the differences Ax and A/ is indicated by a new set of
axes, with the new origin at the point (x0,f(x0)).
y
t:.f=f(x)-f(xo)
1J
I
I
I I
I I
x0 x=x0+t:.x
Geometrically, it is easy to see that the expressions
Jim
f(x) - f(xo)= '
f (xo)
x->x0 X - x0
and
Llf xo Ax) - f(xo)

= lim f(
+
lim = f'(xo)
ax->O Llx ax->O Ax
are merely two different ways of describing the same limit f' (x0) .
The point of this procedure, in finding the derivative of the sine, is that it enables
us to apply the addition formula for the sine. For f (x) =sin x, we have
f(x0 + Ax) =sin (x0 + Ax)

=sin x0 cos Ax + cos x0 sin Ax.
Thus we have
1. sin (x0 + Ax) - sin x0
=Sill Xo = Jill
. ,
f'(Xo)
il.x->O Ax
. sin x0 cos Ax + cos x0 sin Ax - sin x0

=IIm
il.x->O Ax
. [ . cos Ax - 1] 1. [ sin Ax . ]
=I!ill Sill Xo + Jill cos Xo --
il.x->O Ax il.x->O Ax
We are going to show that

cos Ax - 1
Jim =0 (1)
il.x->O Ax
and
sin Ax
Jim = l. (2)
il.x->O Ax
It will then follow that
sin' x0 =cos x0,
and
D sin x =cos x.
The unknown limits ( 1) and (2) have curious forms. Since cos 0 = I, the first
limit has the form
. cos (0 + Ax) - cos 0

IJill =cos , 0. (3)
il.x->O Ax
And since sin 0 = 0, the second limit is
. sin (0 + Ax) - sin 0 .

I1m =sm , 0. (4)
il.x->O Ax
Thus we have found that if cos' 0 =0 and sin' 0 = 1, then sin' x = cos x for every x.
To simplify the notation, in the theorems that follow, we use e in place of Ax,
and state the theorems that we need in the following way:
sine
Theorem 1. Jim -8- = 1.
8->0
. cose - 1
Theorem 2. IIm = 0.
9-o e
Theorem 1 is the hard part; given that Theorem 1 holds, Theorem 2 follows from
it. To see this, we first observe that
cose - 1 [ cose - 1 . cose + �.

lim
o-+o e
= Jim
9-+o e cose + J
Simplifying on the right, we express this as
r [ cos2 e - 1 [ -sin2 e
J J
r
1 .
/:::i e(cose + 1) = 9� e(cose + l)
The last formula can be factored into three parts, giving
lim
cose - 1
= [ -lim
i_n_e
s_ ][ 1im sine
0
][ 1im --1-- ] .
o-+o e 8-+o e 8-+ o-+o cos e + 1
Given that Theorem 1 holds, this gives
cos e - 1
lim = -1 . 0 . l = 0.
o-+o e
(Query: How do we know that lim0_,0 sine = 0, and that lim _0 cose
9 1 ?) =
It remains to prove Theorem 1. First we observe that only positive values of e

need to be considered, because when we replace e by -e, the value of the fraction
(sine)/O is unchanged. Thus if (sine)/e-+ 1 ase approaches 0 through positive values,
it follows that (sine)/e ---+ 1 as () -+ 0 through negative values.
We shall show that for 0 < e < 1Tj2 we have
sine � e � tane.
-1
In the figure, () is the length of the arc from Q to P. Since
RP= sin() and QS = tan(),
the inequalities that we want take the form
RP�()� QS.
To prove this, we have to go back to the definition of arc length.

The figure below shows a broken line inscribed in the arc from Q to P, with n
segments of equal length a1 = a2 = · · · = an. Thus the length of the broken line is
(In the figure, 11 = 3.) We extend the radii of the circle until they intersect the vertical
line through Q; and for each segment of our broken line we let b; be the length of the
corresponding segment on the vertical line through Q.
y s
It is a matter of elementary geometry to check that
and that
for each i.
Therefore
RP<A11< QS,
and so
sin()<A,, <tan 8.
As n -+ oo, A11 -+ () . In fact, this is the definition of the length of a circular arc.
Therefore
sin()�()� tan().
(When we pass to a limit, a "weak inequality'' a � An or A11 � b is always

preserved, but a "strong inequality" a < A,. or A,, 1 for every n,

11
but we cannot conclude that
(?) lim /1 + 1 > 1. (?)

n-oo 11
In fact, the limit is 1, which is � I, but not >I. Hence the overcautious weak in
equalities that we have written above. The strong inequalities sin 8 < 8 < tan 8
always hold for 0 < 8 < 7r/2, but we are not stopping to prove it.) Therefore
::::; _e_ ::::; 1_

.
_
1
- sine - cose
As 8-+ 0, cos 8-+ cos 0 = 1. (You proved this in Problem 30(b) of Section 4.1.)
Therefore I/cos 8-+ 1, because the limit of the reciprocal is the reciprocal of the limit.
Thus the picture must look something like the figure below.
y= cos e
e
----y = sine
That is, the graph of y = 8/sin 8 is "squeezed into l , " and
e
1.1m -- = 1.
9 ... 0 sin e
(This is an instance of a general "squeeze principle," to be discussed further at the

end of this section.) Therefore
because the limit of the reciprocal is the reciprocal of the limit. As we have seen,
this means that:
Theorem 3.
D sin x =cos x.
Once we know how to find D sin x, the derivatives of the other trigonometric
functions are easy.
cos (x0 Ll.x) - cos

X0
cos' x0 =lim
+
----'-
--"------'-----
- -=
Ax->O Ll.x
1. cos x0 cos Ll.x - sin x0 sin Ll.x - cos x0

=Im -
Ax->O Ll.x
cos Ll.x - 1 . sin Ll.x

= cos Im - Im --
1. 1.
Xo Sill Xo
Ao:->0 Ll.x Ao:->O.6.x
= (cos x0) • 0 - (sin x0) · 1 = -sin x0.

Thus:
Theorem 4.
D cos x = -sin x.
By simpler methods, we get
D tan x =sec2 x,
D cot x = -csc2 x,
D sec x = sec x tan x,
D csc x = -csc x cot x.
You will be asked to derive these, in the problem set below.

In finding the limit of &/sine, we used the following idea:
Theorem 5. (The squeeze principle). Letf and g be functions defined at every point
of the interval I, except perhaps at the point x0• If
Jim f(x) = L,
and for each x, g(x) is betweenf(x) and L, then
lim g(x) = L.
y y
g (}
L Lt------
(}
Two illustrations of the theorem are shown above. The theorem is geometrically
clear, and is also easy to prove. The point is that since g(x) is betweenf(x) and L,
any box for fat (x0, L) is automatically a box for g at (x0, L). Since fhas an EO-box
at (x0, L), for every E > 0, it follows that g does also. Therefore
lim g(x) = L,
x�x0
by definition of a limit.
f
-+--
Ll--.,--.c.+-----,,.>.£'-- g
(} -.:-.-�/ : :
I I
j L------l-----J
I
I
The same idea also works when two functions approach the same limit, and a
third function lies between them.
y
f f
(} g
If
g(x) � h(x) � f(x),
and
limf(x) = lim g(x) = L,
x-+xo
then it follows that

limh(x) = L.
Similarly for the following situation:
All of these ideas are very closely related, and we shall refer to all of them as the
squeeze principle.
PROBLEM SET 4.3
Derive formulas for the following:
1. D tan x 2. D cot x 3. D sec x 4. D csc x
5. DVl - sin2 x [Warning: It is very easy to get a wrong answer to this one.]
6. DVl - cos2 x [Same warning.]
7. D cos2 x - :2 � ce>J�" 8. D(cos2 8 + sin2 8)
9. D 2 sin x cos x 10. Dv'l + tan2 8
sin x
11. D(csc2 8 - cot2 8) 12. D
1 + cos x
cos x
13. D l . 14. D(x2 sin x)
+ smx
Show that the following differentiation formulas are correct:
15. D sin 2x =(cos 2x)2 1 6. D cos 2x =(-sin 2x)2
17. D tan 2x =2 sec2 2x 18. D sin ( -x) = [cos ( -x)](-1)
19. D cos (-x) = [-sin ( -x)]( -1) 20. D cot 2x = -2 csc 2x cot 2x
21. D tan (-x) = [sec2 (-x)](-1) 22. D sin 3x = (cos 3x)3
23. D cos 3x = (-sin 3x)3 24. D tan 3x = 3 sec2 3x
*25. Make a plausible guess for D,, sin ax, and verify it if you can.
*26. Same, for D., sin v'x.

*27. If f(x) = sin x and g (x) = cos x, then
(a) f' =
g, (b) g' = -f, (c) f(O) = 0, (d) g(O) = 1.
Is it possible that there is another pair of functions satisfying the same four conditions?
[Hint: Suppose that the pairs fv g1 and /2, g2 satisfy (a) through (d). Consider the
function
What sort of function is F? From what you learn about F, what can you conclude
about /1,/2, g1, and g2 ?]
The answer to this problem has a rather curious significance: it means that all
properties of the sine and cosine are contained, implicitly, in conditions (a) through (d).
That is, the sine and cosine are completely described by the conditions
• I
Sill = COS, cos' = -sin, sin 0 = 0, cos 0 = 1.
4.4 THE APPROXIMATION OF DIFFERENCES BY DIFFERENTIALS
We recall, from the preceding section, the apparatus which we set up in order to
calculate the derivative of the sine. Given a function
/: J-..R,
where I is an interval, and a fixed point x0 of I. For each point x of I, we let .6.x =
x - x0, so that x = x0 + .6.x.
/.,
6}L
I Llx I
I I
We let
IJ..f = f(x) - f(x0) = f(x0 + .6.x) - f(x0).
In the old notation,
' 1. f(x) - f(xo)

f ( Xo) -
_
lffi ,
x--+x0 X - Xo
by definition. In the new notation, this takes the form
f'( 1. f(xo + .6.x) - f(xo) 1. .6.f

x0 ) = im = 1m - .
ax->O 6.x ax->O ilx
4.4 The Approximation of Differences by Differentials 149
When Llx is small, LlfLl

/ x is close to f'(x0). Thus
when Llx::::::; 0,
where ::::::; stands for the phrase "is approximately �qua! to." This ought to mean that
when Llx::::::; 0.
Let us interpret this last statement geometrically.
y
f
Xo X
In the figure, the line Tis the tangent to the graph off at the point (x0,f(x0)). Thus
the slope of Tis f' (x0). If S = (x, y), then
Y f (xo) -
f'( Xo'
)
-
X - X0
because the slope of the segment from P to S is the slope of the line T. This gives
Y -
f(xo) = f'(x0) Llx.
This quantity is called the differential off at x0, and is denoted by elf (See the label
in the figure.) To repeat:
df = f' ( x0) Llx,
by definition. Since x0 is regarded as fixed, throughout this discussion, dfis a function,

whose value is determined when Llx is named. The differential is often convenient for
purposes of numerical approximation. We have observed that
when Llx::::::; 0.
In our new notation, this says that
Llf::::::; df when Llx::::::; 0.
Let us try this on some numerical examples, and see how good the approximation
looks.
Example 1. Let
f(x) = .J� (x � O);

and take
Xo= 25, bx = 0.4.
y
5
4
3
2
�---f�---'�---'�---'�---'�---'�---'�---'��'--�'--�'---�'---�'---!--,__ X
0 2 4 6 8 10 12 14 16 18 20 22 24 I 26
Xo=25
Then
f(x0) =
.J25 = 5,
f'(x) =
1;- (x > 0),
2yx
and
df= lo bx · = lo (0.4)
· = 0.04.
The approximation formula
df�bf
suggests that
.-/25.4 = f(x0 + bx) = f(x0) + bf� f(x0) + df= .J25 + 0.04;

.-/25.4 � 5.04.
The actual value of .-/25.4, correct to six decimal places, is
.-/25.4 = 5.039841.
Thus the error in our approximation is 0.000159, which is not bad. Note also that the
approximation b f� df wasn't supposed to be good except when bx is small; and
Ax= 0.4 is not very small. Using Ax= 0.1, we get
1
df ;- (0.1) = 0.01;
2y25
=
.J25.l �f (5) + dj = 5.01.

The correct value is

)25.1 = 5.00999,
so that our error is 0.00001, which looks better. Using t:.x = 0.01, we get
dj = 11-o(0.01) 0.001; =
)25.01 f::::! 5.001.

Using five-place common logarithms, we get
)25.01 f::::! 5.0010.

Thus, in this case, the differential is as accurate as five-place tables.
It is natural to ask why the approximation
b.f f::::! df =
f'(xo) b.x
should be as good as it is. The reason is as follows. We know that
On this basis, we wrote

b.f
f'(x0) f::::! when b.x f::::! 0. (1)
b.x
Multiplying by b.x, we got
b.f f::::! f'(xo) b.x when b.x f::::! 0. (2)

The second of these approximations is much better than the first. The point is that if
and b.x f::::! 0,
then the product
[b.b.xf - f'(x0)] b.x f::::! O;
when you multiply two numbers each of which is small, the product is even smaller.
We shall now express these ideas in a more exact form. For each b.x, let
E(b.x) f (x 0 + b.x) - f(x0) j'(xo)

_
(b.x -¥: 0).
b.x
=
Then
lim E( b.x) = 0,
6.x-+O
because
1. /(xo b.x) - f(xo) -

!ID
+
- f'(Xo) .
ax-+O b.x
Thus the graph of the function Elooks like the figure on the left below.
y y
// y =E(t.x)
(D.x7'0)
/
To this graph we add the origin. That is, we define
E(O) = 0.
The graph of the extended function Eis shown on the right above. We now have
lim E(tu) = E(O) = 0.

Ax-+O
Note that Eis defined on some open interval containing 0.
y y
x
�---+- -xo- �--a ��Xo�X-o�+-�
a -
The reason is that if the open interval (x0 - a, x0 + a) lies in the domain off,
then the open interval ( -a, a) lies in the domain of E. An open interval containing
a given point will be called a neighborhood of the given point. In this language, we
can sum up the above discussion in the following theorem.
Theorem 1. Letf be a function defined in a neighborhood of x0, and suppose that f

is differentiable at x0• Then there is a function E, defined in a neighborhood of 0,
such that
i) 6.J= f'(x0)6.x + E(6.x)6.x, and

ii) lim�x�o £(6.x)= E(O) = 0.
(Proof Use the function Ewhich we have just defined.)
This theorem explains why

d
f = j'(xo) 6.x
is a good approximation of
6.j = J(x0 + 6.x) - j(x0)
when !1x � 0. The reason is that
11/- df= !1f-f1(x0)!1x = E(!1x)!1x.

Thus when !1x � 0, the error in the approximation !1f � df is doubly small, being a
small multiple of the smaU number 11x.
In many cases, it is possible to estimate the largest possible error that can result
when you use df as an approximation for !1f For a discussion of this problem, see
Appendix D.
PROBLEM SET 4.4
Following is a partial table of the sine and cosine functions, for ready reference in solving
some of the following problems:
x sin x cos x
0.0814 0.0814 0.9967

0.1222 0.1219 0.9925
0.2094 0.2079 0.9781
0.3840 0.3746 0.9272
1. Find sin 0.1251 approximately (sin 0.1251 = 0.1248, correct to four decimal places).
2. Find cos 0.0844 approximately (cos 0.0844 = 0.9964).

3. Find sin 0.2123 approximately (sin 0.2123 = 0.2108).
4. Find cos 0.3869 approximately (cos 0.3869 = 0.9261).
5. How do you account for the first two entries in the first line of the above table?
6. Without the use of tables of any kind, get the best approximation you can for
sin 0.5235988. (This is a trick question.)
7. Same question, for cos ( -6.2832).
Without using tables of any kind, get numerical approximations for the following. The
answers given are the "right answers," correct to the indicated number of decimal places;
it is not to be expected that an approximation process based on the differential will give them
exactly.
8. >'Y27.1 [Answer: 3.004] 9. V25.2 [Answer: 5.0200]
10. {116.3 11. -V' -7.9

12. One of the standard approximation formulas used in mathematical physics says that
sin x ""'x when x ""'0.
Explain how this formula is related to the ideas in this section of the text. [Hint:
Consider the general approximation formula
b.f"" df when b.x ""0 .
. What form does this take, for j(x) = sin x, x0 = O?]

13. Same question, for
cos (1 - x) "" 1 when x ""1.
14. Another standard approximation formula says that
(1 + x)" ""' 1 + nx when x ""'0.
Interpret this in terms of the theory that we have been developing, and justify it. [Hint:
Surely the given formula is equivalent to
(1 + 6 xr - 1 ""' n 6 x when 6 x ""'0.
Here, what isf? What is x0? What are 6fand df?]

15. Same question, for the formula
x
v1 + x""' 1 + - when x""' 0.
2
16. Same question, for the formula
1 x
- ""'1 when x""' 0.
.3; - 3
v1 + x
17. Without using calculus at all, justify the approximation formula
1
R:Jl-x when x""' 0.
1 + x
--
Is this a "doubly good" approximation in the same sense in which 6f""' df is "doubly
good"? Why or why not?
4.5 COMPOSITION OF FUNCTIONS
In calculating derivatives, we have often found it convenient to regard one function

as a power of another. For example, given
<f>(x) = (x2 + 3x + 5)5,

we let
g(x) = x2 + 3x + 5,
so that
We can then get <P' in the form
<P' = 5g4g', </>'(x) = 5(x2 + 3x + 5)4(2x + 3).
Similarly, we have found it convenient to regard one function as the positive

square root of another. For example, given
�(x) = Jx2 + 1,
we let
g(x) = x2 + 1.
We then get <P' in the form
</>'(x) = � = __
x_
2Jx2 + 1 Jx2 + 1
4.5 Composition of Functions 155
The idea that we have been using is that of composition of functions. In the
first case, the action of ¢ is described by
¢: xH (x2 + 3x + 5)5.
We split this operation into two steps, like this:
x H x2 + 3x + 5 H (x2 + 3x + 5)5.
The first of these steps represents the action of the function
g: x H x2 + 3x + 5.
The second step raises things to the fifth power. It can thus be described by the
function
In this situation, g is called the inside function; it represents the first step. The function
f is called the outside function; it represents the second step. And ¢ is called the
composition off and g. The reason for the use of the terms inside and outside is that
we can write
cp(x) = f(g(x)).
To get cp(x), we should substitute g in the formula for f
Diagrammatically:
x � x2 + 3x + 5 � (x2 + 3x + 5)5•
�
Our second example fits the same pattern. We have
cp(x) == .Jx2 + 1, g(x) = x2 + 1, f(u) = .J�,

cp(x) =f(g(x)) )g(x) = .Jx2 + 1.
=
Diagrammatically:
f ;-
-
g
x H x2 + 1 H vx2 +
1.
�
Algebraically, to get the values of the composite function ¢ = f(g), we substitute
g(x) for u in the formula forf(u). This is why we described the "square-root function"
f by the formula
f(u) = .J�
instead of the equally logical formula f(x) = ,J--;. We want to form the composite
function by setting
u = g(x) = x2 + 1,
and it would hardly make sense to set ( ?) x = g(x) = x2 + 1 ( ?).
We sum all this up in the following definition:
156 Trigonometric and Exponential Functions 4.S
Definition. Given two functions
g: A-+ B, f: B-+C,
the composition
f(g): A-+ C
is the function whose values are given by the formula
f(g)(x)=f(g(x)).
Here, for each x, f(g)(x) denotes the value of the functionf(g) at the point x.
Diagrammatically:
g f
AHBHC.
�
Let us consider some more examples.
Example 1. Let
f(u)=sin u, g(x)=2x+ 1.
Then
f(g(x)) = sin g( x) = sin (2x+ 1).
(In this example, what is A? What are B and C?)
Example 2. Let
f(u)=u2+u+I, g(x) = J;.
Then
f(g(x))= c);)2 +J; + 1 =x + J; + I.
(What are A, B, and C?)

Example 3. Let
f (u)= sin u, g(x)=x2 - 5.
Then
f(g(x))=sin (g (x))=sin (x2 - 5).
Example 4. Let
f\14 dt, g(x)

f(u) = - 1) =
x2•
Then
<i(X)
f(g(x)) J (t4
=
dt=J(x2o (t4
0 . - 1) - 1) dt.
Thus, for example,
f(g(3))=f 2(t4 dt=f(t4 - 1) - 1) dt.

In Examples I through 4 above, we supposed that f and g were given, and we
then proceeded to form the composite function ¢=f(g). More often, however, we
are given a function ¢, and in order to investigate the function ¢, we express it as the
4.5 Composition of Functions 157
composition of two other functions, each of which is simpler than </>. For example,
f2(t4
to investigate the function
</>(x) = - 1) dt,
we first observe that it has the form
(u(x)
</>(x) =
Jo
(t4 - 1) dt,
where
g(x) = x2•
Thus
</> = f(g),
where
f(u) = lu(t4 - 1) dt, g(x) = x2.

Similarly, in the preceding three examples, if</> is given by the final formula, we shall
for many purposes need to set up a pair of functions f and g in, such a way that
</>=Jw.
The derivative of a function is also a function; and so we can form composite
functions of the type/' (g) and f (g'). Consider, for example,
f(u) = ua, g(x) =sin x.

Then
f'(u) = 3u2, g'(x) = cos x.
Therefore
f (g(x)) =sin3 x, f'(g(x)) = 3 sin2 x, f'(g(x))g'(x) = .3 sin2 x cos x.

These formulas are significant, because it will turn out that
Df (g(x)) = D sin3 x = f'(g(x))g'(x) = 3 sin2 x cos x.

Similarly,
g(f(u)) = sin u3, g'(f(u)) = cos u3, g'(f(u))f'(u) = (cos u3)3 u2,
which will turn out to be the derivative of cos u3. (Here cos u3 is the cosine of u3,
not the cube of cos u.) Let us try one more example:
f(u) =cos g(x) = .JX:, g'(x) = lr,

2"x
u,
f(g(x)) =cos x, f'(g(x)) = -sin .J-;,

f'(g(x))g'(x) =(-sin -jx) 11_.
2, x
In dealing with composite functions, we shall need the following:
Theorem 1. The composition of two continuous functions is continuous. That is, if
lim g(x) = g(x0) = u0

and
limf(u) =f(u0),
u-+uo
then
Jim J(g(x)) = J(g(x0)).
x-+xo
The idea here is that
x R:o! x0 => g(x) R:o! g(x0) => f(g(x)),::::; f(g(xo)).

In Appendix E it is shown that this idea can be used to get a proof.
PROBLEM SET 4.5
For each of the functions ¢, given in the problems below, find formulas for functions
f andg, such that</> =j(g). Then get formulas for f',g',f'(g), and ¢'.
1. <f>(x) =sin2 x 2. <f>(x) =cos2 x 3. <f>(x) = (sin x + cos x)2

4. <f>(x) =sin 2x 5. <f>(x) = tan 2x 6. <f>(x) =cos 2x
7. <f>(x) = Vl - x2 8. <f>(x) =sin6 x 9. <f>(x) = f't + x
10. <f>(x) = fin:r (t2 + 1) dt 11. <f>(x) = fos.r (t2 + 1) dt
(Note that the function
f(u) = f' (t2 + 1) dt
can be expressed without the use of integral signs; f can be calculated as a polynomial.)
sin u - sin u0 sin x2 - sin xg

12. a) Find Jim . b) Find Jim------.:.
U-4>1lo Ii - Uo x2 - xfi
(It is not hard to see a very plausible answer to Problem 12(b). To prove, in an orderly
way, that your answer is right, you should express the function
sin x2 - sin xfi

<f>(x) = --=---
x2 - xfi
as the composition /(g) of two functions f andg, and then apply Theorem 1.)
sin x3 - sin x8
13. Find Jim -----
x-+xo X Xo -
14. G iv en <f>(x) =sin x2, proceed as in Problems 1 through 11.
15. Do the same, for <f>(x) =sin x3•
16. Do the same, for <f>(x) = sin ·<X:·
4.6 The Chain Rule 159
17. Given
<f>(x) = fin� I + ,2 dt.
On the basis of the theory that you know so far, you are in no position to calculate
f
(u)
=
f v'l+f2 dt.
And you have, so far, no general formula for
D[f(g)] =
?
On the other hand, you ought by this time to be able to make a good guess about
sinx v'l+f2 dt
D[f(g)], and then use your guess to write some kind of formula for
<f>'(x) = D
J
0
.
[Hint:
As a start, what is/'(u)?]
sinx - I
18. Find lim
:r-1Tf2 x - 7r12
.
[Hint: If you can figure out what the geometric meaning of this limit is, it will then be
easy to find its numerical value.]
cos x + 1 tanx - I
19. Find lim ---- [Same hint.] 20. Find Jim .
x-;; x - 1T :r-"14 x - 14
1T
sin 2x - I secx - 1
21. Find Jim 22. Find Jim ---
.x�"/4 X - 14
1T :r�o .X
4.6 THE CHAIN RULE
You may have observed, in the preceding problem set, that the formula
Df(g) = f'(g)g'
held in a number of cases. For example, if
f(u) = u",
then
f' (u) = nu"-1;
and
Df(g) = Dg" = ng"-1g' = f'(g)g'.
Similarly, if
f(u) = Jli,
then
f' ( ll) = ----!,: :'

2,,. ll
Df(g) = D,/g =
-1=
2 Jg
· g' =
f'(g)g'.
The same formula seems to hold for

f(u) = sin u,
at least in the cases where we can test the formula by calculating DJ(g) = D sin g.
For example, it turns out that
D sin 2x 2 cos 2x,
=
and this has the form

DJ(g)= f' (g)g''
where
f(u) =sin u, f ( u) cos u,
' =
g(x) = 2x, g'(x) = 2,

f'(g)g' = (cos 2x) 2 · = 2 cos 2x.
The formula
Df(g)= f'(g)g'
is called the chain rule. In fact, it always holds, whenever the right-hand side has a
meaning, that is, wheneverf'(g) and g' are defined. We shall prove this at the end of
this section. First, we give some illustrations of its use.
Example 1. Consider .\ ( ? \ '

cp(x) = sin (3x + 1).
This is a composite function
cp(x) = f(g(x)),
with
f(u)= sin u, f'(u) = cos u,
g(x) = 3 x + 1, g'(x )= 3.
By the chain rule,
cp'(x) = D sin (3x + 1) = [cos (3x + l)]D(3x + 1)

= 3 cos (3x + 1).
Example 2. Consider
cp(x) = sin (k + x) .
By the chain rule,
f (x) = [cos (k + x)]D(k + x) = cos (k + x).

Note that if the chain rule is known, and the formula
D sin= cos
is known, we can find D sin (k + x) without using the addition formula.

To give new applications of the chain rule, we should not be talking about cases
where the outside functionf is u" or)�. For these outside functions, we have known
for a long time that the chain rule held. After the trigonometric functions, the next
outside functions to consider are integrals:
Example 3. Consider
cp(x) =
f,kx 1
- dt (k, x > 0).
1 t
u
Here
J
1
cp(x)= f(g(x)), f(u) = - dt (u > 0),
1 t
f '(u) = 1:' g(x) = kx, f'(g(x))= : ' g'(x) = k,

u x
Dcp = Df(g) = f '(g)g' = J... k = _!.

kx x
lkx "'
This is a curious result:
J
1 1 1
D - dt = - = D - dt.
1 t x 1 t
What does it tell us about the functions?
Example 4. The chain rule can be applied several times in the same problem. For
example, we know that
D sing= (cosg)g',
whatever g may be. We can then apply the formula in cases where g' itself needs to
be calculated by the chain rule:
D sin sin x = (cos sin x)D sin x

= (cos sin x) cos x.
Here sin sin x is the sine of the sine of x, which is different from sin2 x.
Therefore
D sin sin sin x = (cos sin sin x)D sin sin x
= (cos sin sin x)(cos sin x) cos x.
Example 5. Similarly,
D{[(x3 + 1)2 + 1]2 + 1}3 = 3{[(x3 + 1)2 + 1]2 + 1}2D{[(x3 + 1)2 + 1]2 + I}
= 3{ }2 2 [(x3 + 1)2 + I]D[(x3 + 1)2 + I]
•
3{ }2 2 [ ] 2(x3 + l)D(x3 + I)
= · · ·
= 3{ }2 • 2[ ] 2( ) · · 3x2•
Here we have left braces, brackets, and parentheses empty, in the intermediate
stages, to make the steps easier to follow. The final answer is
3{[(x3 + 1)2 + 1]2 + 1}2 2[(x3 + 1)2 + • l] · 2(x3 + 1) 3x 2,

·
which can be simplified slightly by collecting constants.

We shall now prove the chain rule. Given
rp(x) = f(g(x)),
we want to show that for each x0 we have
rp'(xo) = f'(g(xo))g'(xo).
Obviously, we must assume that
a) g has a derivative g' (x0) at x0, and

b) f has a derivative/'(g(x0)) at g(x0).
A differentiable function is continuous. Therefore
lim g(x) = g(x0).
For convenience of reference later, we write this in the form
c) lim [g(x0 + �x) - g(x0)] = 0.

t.x->O
By definition,
rp(x) - rp(xo )
rp'(x0) = lim
x->x0 X - Xo
rp(x0 + �x) - rp(x0)
= lim
t.x->O �x
. f( g(x o + �x)) - f(g(xo))
=hm .
t.x->O �X
Let
�u = g(x0 + �x) - g(x0) = g(x0 + �x) - u0,
so that
g(x0 + �x) = u0 + �u.
Then
"'' 1. f(uo + � u) - f(u0)

� (Xo) -
_
lm .
t.x->O �x
Here the numerator is a difference
�f = f (u0 + Liu) - f(u0)

between two values of the function f
Now comes the crucial idea: we apply to f the theorem stated at the end of
Section 4.4. We need to change the notation of the theorem, using u in place of x,
to fit the notation of the present discussion. The theorem then says that there is a
function E, defined in a neighborhood of 0, such that
�f = f'(u0) �u + E(�u ) �u,

and
Jim E(�u) = E(O) = 0.
t.u->O
Therefore
D.f D.u D.u

= f'(u0) + E(D.u)
D.x D.x D.x
g(xo D.x) - g(xo) (x0 �x) - g(x0)

E(D.u) g
+ +
= f'(g(xo)) + .
D.x D.x
It is now easy to see what the limit is. By definition of g'(x0), we have
. g(x0 + D.x) - g(x0)

I1m = g'(x0)•
<ix-o D.x
As D.x ---+ 0, D.u---+ 0. (Remember the definition of D.u, and recall condition (c),
at the beginning of the proof.) Therefore, by Theorem 1 of Section 4.5, we have
lim E(D.u) = 0.
Ax-+O
This gives
.
D.f
ef/(x0) = Jim - = f'(g(x0))g'(x0) + 0 · g'(x0).
<ix-o D.x
We therefore have:
Let f and
Theorem 1.
g be functions. Then
Df(g) = f'(g)g',
at every point x0 at which the right-hand member has a meaning.
That is, the formula holds at every point x0 such that (a) g is differentiable at
x0 and (b) f is differentiable at g(x0). These conditions illustrate the normal pattern
of theorems involving differentiation formulas: the equation holds whenever the
quantities mentioned in the right-hand member exist.
PROBLEM SET 4.6
In this problem set, your main job is to learn to use the chain rule. In each odd-numbered
problem, from 1 to 19, you should indicate the logic of your work by writing formulas for
f,g,f',f'(g),andg', before writing the answer in the form D[(f g)] = f'(g)g'. For example,
given the function
<f>(x) = sin (x2 + 1),
your solution should be written in the form
f(u) = sin u, g(x) = x2 + 1,

f'(u) =COS!I, f'(g(x)) = cos (x2 + 1),
'
g'(x) = 2x, ¢ (x) = [cos (x2 + 1)]2x = 2x cos (x2 + 1).
If you go through this routine for one day, you are Jess likely hereafter to omit the factor
g' followingf'(g) in calculating Df(g).
The parentheses and brackets in the expression [cos (x2 + 1)]2x look clumsy, but to
eliminate the brackets we have to change the order of the factors, as in the last expression
above. It would have been simpler to write
(?) ¢'(x) =cos(x2 + 1 )2x (?)

but this is the wrong answer: the function on the right is the function whose value, for each
x, is the cosine of 2x3 + 2x. If you write this formula for ¢',you are relying on the reader to
remember what the problem was and to realize that you must not mean what you are saying.
In some cases you may not feel sure whether brackets are necessary. When in doubt,
use them.
Now find the derivatives of the following functions:
1. 2. 3. 4. 5. 1) 6.
-1
sinx2 sin2 x cosx3 cos3x tan(t2 + tan t2 +
7. 8. 9. 10.
2
x
sin(x3 + x) sinx3 +x cos ,i:x ,1 cosx I I. tan --
12. 13. 14. 15.

2
tanx -
(vx)2 [x[2 a) sec2 x b) secx2
16. a) tan2 x b) tanx2 17. cos4x - sin4x 18. cos 2x 19. cos2 x - sin2 x
20. 21. 22. 23.

2
x sinx
sec 2x sin x cos2x + sin3 x tan -
+ cosx
24. sec vx2 + 1 25. cos vx2 + 1 26. a) '\!tanx b) tan ,;:;.:
27. 28. 29.

\
3·- ,,--
30. 31. 32. t 33.

a) sin x b) \sinx sin sinx sin2 sinx + cos2 sinx
cos:c
cos cosx sin cosx sin sin sin tan sinx
34. 35. L t3Jt 36. dt

sin (I + sinx)
ix' 0 cos f
37.
f,k:J; -1 dt f,x 1 f,k - dt.
Let k be any positive number; and for each positive number x, let
- - dt -
t t
1
¢(x) =
1 f l 1
Find the simplest possible formula for ¢'(x). Then do the same, for the functions ¢(x)
f,x' - dt J,"'' 1- dt - 2 Jx 1- dt x3 1
defined by the following formulas.
38. 0) 39. 40. l - dt

t t t
1
(x >
V;; 1 1 "' 1
t
lx3�dt-3 x�dt 42. J\;�dt

1 1 1 1
41. J t 43. J - dt - - - dt
t t 2J t
sin 1
1 1
f 1 1
- dt (0 < <
1
44. J
t
.t
X TT )
45. For each x > 0, let

1
=f,x �dt.
t
f(x)
1
Show that for every pair of positive numbers a and b, we have
[Hint:
/(ab) = f(a) + f (b).
When we try to attack this problem by the methods of calculus, the obvious
introduce a function into the problem.]

trouble is that the problem does not appear to involve any functions. Therefore our
first step should be to
4.7 Invertible Functions. The Inverse Trigonometric Functions 165
46. Let ef>(x) =f(xn), where x > 0 and/is as in the preceding problem. Find ef>'(x).
*47. Given
D sin =cos, D cos = -sin,
sin 0 = 0, cos 0 = 1,
and given no other information whatever about the sine and cosine, prove that
sin (k + x) =sin k cos x + cos k sin x,

cos (k + x) = cos k cos x - sin k sin x,
for every k and x. [Hint: Let f be the function which is 0 if the first equation holds;
=
let g be the function which is = 0 if the second equation holds, and investigate the
function
F =12 + g2.]
This result tends to confirm a claim that was made in Problem *27 of Problem Set
4.3. The claim was that all properties of the sine and cosine are contained, implicitly,
in the properties that we have just used to prove the addition formulas. Later we shall
find further confirmation of this.
*48. Let/be a function, defined for every x, such that
(a) f" =
-f , (b) f (0) =0, (c) f' (0) = I.
Show that f(x) =sin x for every x.

*A9. Let g be a function, defined for every x, such that
(a) g" = -g, (b) g(O) = I, (c) g'(O) =0.
Show that g(x) = cos x for every x.
4.7 INVERTIBLE FUNCTIONS. THE INVERSE TRIGONOMETRIC FUNCTIONS
A function f is called invertible if its graph intersects every horizontal line in at most
one point. Thus f(x) = x3 is invertible, but f (x) = x2 is not.
y y
y=f(x)=x3. y=f(x)=x2.
Iff is invertible, then for each number yin the image of/there is exactly one number
x in the domain of/ such that/(x) y. =
Thus to every invertible function f there corresponds a new function 1-1, called
the inverse off. (This is pronounced f inverse. The symbol -1 is not an exponent,
really;andJ-1is not 1//) The inverse is defined by the condition that
1-1(x) = y ifJ(y) = x.
If fis invertible, this condition defn

i es a function, because for each xin the image
offthere is exactly one such y. It is not hard to see what this relation between fand
J-1 means geometrically. The point (x, y)is on the graph of J-1if the point (y, x)
is on the graph of/ Therefore,to get the graph ofJ-1 from the graph off,we should
reflect the graph offacross the line y x. =
Let us see what this means algebraically. Consider

f(x) =
a
x .
The graph offis the graph of the equation

a
y= x . (I)
The graph ofJ-1is the graph of the equation
x= � m
Here we have simply interchanged and yin Eq. (I). Now (2)is equivalent to
x
y = fl;: .
Thus
J-l (x) = -{Y;:,
as we would expect: the inverse of cubing is the extraction of cube roots.
y
f
y
Theorem 1. Let f be an invertible function. Then
J(!-1(x)) = x,
for every x.
Proof For each x, lety =J-1(x). Then/(y) = x, by definition ofJ-1. Therefore,
J(!-l(x)) = f (y) = x.
We can use this idea to calculate the derivatives of inverse functions, assuming
that the inverse function has a derivative.
Example 1. The function f(x) = x3 is invertible, and its inverse is J-1(x) = \o/;:.
Thus
(i1'�)3 = x.
We take the derivative on each side of this equation, using the chain rule for the
composite function on the left. This gives:
3(\o/x)2 D\o/x = 1,
D\o/x = --1= (x ¥- 0).

3\o/x2
You may have calculated this by another method, in Problem Set 3.6, but the present
method is easier.
Example 2. A function of the form j(x) = xq (where q is a positive integer) is

not necessarily invertible; in fact, it never is when q is even. We therefore restrict x
to positive values. This gives an inverse function
y
r1<x>=vx
We calculate D�/x in three steps, as follows:
1) (::Jx)q = x,
2) q(::Jx)q-l D�/x = 1,
- 1
3) D�x = ---== (x > 0).
q::}xq-1
When we use this method, the equations that we write have the following general
form:
1) j(J-1(x)) = x,
2) 1
f'(f- (x))Df-1(x) = 1,
3) Df-1(x) = \
f'(r (x))
(f'(f-1(x)) 9'6 0).
(You should check this against the preceding examples.) The method assumes that
our problem has an answer, that is, that1-1 has a derivative. Thus we need to show
that this holds, in every case in which the fraction at the last stage has a meaning.
This is easy to see. Consider I, 1-1, as in the figure below, with
Yi = 1-1(x1), X1 = l(Y1),
as the labels indicate.
If I has a tangent line L, at (y1, x1), then1-1 has a tangent line L,' at (x1, y1): to get
this, we reflect both the graph and the tangent line across the line y x. The slope
=
of Lis
m =
f'(y1) =
l'(f-1(x1)).
If m 9'6 0, then Lis not horizontal. Therefore L' is not vertical, andl-1 has a deriva
tive at x1. Thus we have completed the proof of the following theorem.
Theorem 2.
1
D r-1( x)
1 f'(r1Cx))'
=
wherever the fraction on the right has a meaning.

In most cases, the method used in deriving this formula is easier to use than the
formula itself. To find n1-1, we write 1(!-1(x)) 1, differentiate, and solve for
=
D l-1, as in Examples 1 and 2.

We shall now discuss the so-called "inverse trigonometric functions." This
involves a slight difficulty, because the fact is that no trigonometric function is
invertible. The reason is that every trigonometric function satisfies the identity
I(x + 27T) = I(x),
for every x for which the trigonometric function l(x) is defined at all. Therefore
every value that a trigonometric function takes on at all is taken on for infinitely
many values of x. For example, the graph ofj(x) = sin x looks something like this:
-1
y=f(x)=sinx
If we restrict x to the interval [-7T/2, 7r/2], then we get a new function whose graph
includes some, but not all, of the original graph. This new function is denoted by
Sin, and the graph of y = Sin x looks like the left-hand figure below.
y
y
Sin
The graph looks as if Sin ought to be invertible; and in fact this is not hard to see.
In the right-hand figure above, we have switched the notation to fit the definition of
the sine, so that y = sine. Every point of the semicircle corresponds to exactly one
() on the interval [ -7T/2, 7r/2]; and every horizontal line intersects the semicircle in
exactly one point.
As always for inverse functions, we get the graph of Sin-1 by reflecting the graph
of Sin across the line y = x. Therefore the graph of Sin-1 looks like this:
y
Similarly, we define Cos x to be equal to cos x, on the interval [O, 7T], and we
show that Cos is invertible. The graphs of Cos and Cos-1 look like this:
y y
To find the derivative of Sin-1, we write
sin Sin-1 x = x,
(cos Sin-1 x)D Sin-1 x = 1,
l
D Sin-1 x =
cos Sin-1 x
We want to simplify the expression cos Sin-1 x on the right, and, while we are at it,
we shall get a formula for sin Cos-1 x. Since
cos2 u + sin2 u = 1,
we can now solve, getting
cos u = ±.JI - sin2 u, (1)
sin u = ±.JI - cos2 u. (2)
For
u = Sin-1 x,
this gives
sin u = sin Sin-1 x =x,
and so from (I) we get
cos Sin-1 x = ±.J 1 - x2• (3)
Similarly, for
u = Cos-1 x
we have
cos u =cos Cos-1 x =x
· ,
and so from (2) we get
sin Cos-1 x = ±.JI - x2. (4)
Formulas (3) and (4) are correct, but they are not good enough for our purposes.
In fact, the double signs can be omitted, and the formulas still hold:
Theorem 3.
cos Sin-1 x == .J 1 - x2,

sin Cos-1 x = .J 1 - x2•
To see this, we merely need to remember that
On this interval, the cosine is � 0. Therefore, in (3), it must be the plus sign that
applies. Similarly,
0 � Cos-1 x � 7T.
On this interval, the sine is � 0. Therefore, in (4), it must be the plus sign that
applies.
We now substitute .J l - x2 for cos Sin-1 x, in the formula that we got for
D Sin-1 x. This gives:
,--
Theorem 4. D Sin-1 x = 1/...; 1 - x2 (-l<x<l).
Note that D Sin-1 x is always >0, just as the graph suggests that it ought to be.
At the endpoints of the graph, the tangents are vertical.
The proof of the following theorem is like that of the preceding one:
Theorem 5. D Cos-1 x = -1/.J1 - x2 (-l<x<l).
Note that D Cos-1 xis always <O, as it should be.

For tan x, the process is simpler. The graph of y = f (x) = tan x looks something
like this:
To get an invertible function Tan, we take the portion of the graph that lies between
x = -7T/2 and x = 7T/2. We could verify by brute force than Tan is invertible, but
it is easier to prove first the following theorem:
Theorem 6. Letfbe a differentiable function on an interval/. Ifj'(x) � 0 for every

x in J, thenfis invertible.
The proof is based on the mean-value theorem. If f is not invertible, then the
graph intersects some horizontal line in more than one point. Thus
f(a) =
f(b),
for some a and b in /. Therefore the graph has a horizontal chord. By MVT, this
means that the graph has a horizontal tangent; that is,f'(x) = 0 for some x, which
contradicts the hypothesis for f.
Now the domain of Tan is an open interval ( -1T/2, 1T/2). On this interval,
Tan' x = sec2 x � 0. Therefore Tan is invertible.
The graphs of Tan and Tan-1 look like this:
y
y
y = Tanx
7r
2
--------�---------
2
Theorem 7. D Tan-1 x = 1/(1 + x2).
The derivation is easier than the preceding ones, because it turns out that there
are no double signs to be eliminated.
For the secant, the situation is trickier, and some handbooks contain formulas
that are wrong. The reason is that the graph of the secant looks like this:
y
I IY = secx I
I I I
I I I
I I I
I I I
I I,,.
3.,,. : 3.,,.
1
-2 _,. - �: 2
1r
12
x
I I I I
'(\ :(\
-1
I I I I
I I I I
I I I I
I I I I
I I I I
(Remember that sec x = I/cos x wherever cos x ¥- 0.) This graph consists of
infinitely many connected pieces, but none of these connected pieces is the graph of an
invertible function. We therefore cannot use all of any one of the pieces. Everybody
agrees that we ought to use the part of the graph where 0� x < TT/2, but there is no
general agreement on what else we ought to use. To be safe, we define Sec x only for
0� x < TT/2. (See the graphs below.)
y y
Jy=Sec x
I
I .,,.
---------
I 2
I
I
I
I
--f---
.,,. --L---x
2
(Query: How do you know that the secant never takes on the same value twice,
on the interval [O, TT/2) ?)
On the basis of the definition of Sec-1, it is plain that the equation
y = Sec-1 x
means two things:

.1: = secy (5)
and
7T
O:Sy
- < -. (6)
2
We now calculate the derivative. We have
Sec Sec-1 x = x, (7)

and
Sec' u = Sec u Tan u,
for every u from 0 to TT/2. (Why are we justified in using capital letters on the right?)
Therefore, by the chain rule,
(Sec Sec-1 x)(Tan Sec-1 x)D Sec-1 x = 1,

and
x(Tan Sec-1 x)D Sec-1 x= 1. (8)
Therefore
x = -----
D Sec-1 1 (9)
x Tan Sec-1 x
We now need a formula for Tan Sec-1 x, analogous to the formulas for sin Cos-1 x
and cos Sin-1 x. We know that
1 + tan2 u = sec2 u
for every u. Therefore
tan u= ± ) sec2 u - I.
For u= Sec-1 x, this says that
tan Sec-1 x= ± ) sec2 Sec-1 x -
Since 0 � Sec-1 x < 1T/2, we have
tan Sec-1 x = Tan Sec-1 x � 0,
and so it must be the plus sign that applies on the right. Therefore
Tan Sec-1 x = )x2 - 1,

and we have:
Theorem 8. D Sec-1 x = 1/x)x2 - I.
For convenience of reference, we repeat these differentiation formulas:
D Sin-1 x = ---1 , D Cos -1x = �

) 1 x2 - )1 x2 -
1
D Tan-1 x= , D Sec-1 x= _1
x2 )
__
1 + x x2 - 1
We now have a new set of functions arising as derivatives: none of these four functions
has appeared before as a result of differentiation. This means, for one thing, that we
can use our new functions to solve certain area problems that we couldn't solve
before. Later we shall see that the process by which we find a function whose deriva
tive is a given function has many other applications.
You will also need to remember
cos Sin-1 x = )1 - x2, sin Cos -1 x = )1 - x2.

PROBLEM SET 4.7
For each of the following functions, calculate the derivative.
1. Sin-1 (x - 1) 2. Cos-1 (x - 1) 3. Tan-1 (x + I)

4. Sec-1 (x + 1) 5. Sin Sin-1 (x + 1) 6. Cos Sin-1 x
7. Sin Sin-1 x2 8. Cos Sin-1 (2x) 9. Sin-1 \1 I -

x2
10. Cos-1 Vl - x2 11. Tan-1 (x2 + I) 12. Tan-1 (sec2 x - J)
v1 - x2
13. Sec-1 x2 14. See1 (I + tan2 x) 15. Tan-1
x
I 1 I
16. Sec-1 17. Cos-1 � 18. Sin-1-
x x
19. Sec Tan-1 x 20. Tan Sec-1 x 21. Sin Tan-1 x

22. Cos Tan-1 x 23. Tan Sin--1 x 24. Tan Cot-1 x
25. Sin-1 (2x + 1) 26. Tan-1 (I - x) 27. Sin-1 x2
28. Show that
l
See1 x Cos-1 - ,
x
=
for every x on a certain interval. What interval?

29. Show that
11'
Sin-1 x + Cos-1 x
2'
for every x on the interval [ I, 1 ]. - (A very short proof is possible. Remember the
uniqueness theorem of Section 3.8.)
'/�
- ? dt. 2dt.
I I
30. Find f -1 -
+ r-
Sketch. 31. Find
L l--
+ t
Sketch.
0
-1
1v2 r
f/2 dt. dt.
1
32. Find
0 VI - t2
33. Find f o v1 - t2
34. Try to get the right answer for the area under the graph of y I /(I + x2) on the whole
= ,
interval ( - w, w). You need not justify your answer, so long as it is right.
35. Given
0 � x � l,
find a formula for f-1(x). Then explain how your answer might have been predicted
without a calculation.
(2 1
dt.
2 1
dx.
36. Find
J 21v3 iVi2 1
37. Find
J xVx2
__
_
_
l
-
1 -
11v3 t
38. Find 1 --- v t2 + 1
dt.
0
39. In Theorem 6 we required that f' (x) be different from 0 everywhere on the interval I.
This hypothesis was satisfied by Tan on the open interval ( -Tr/2, Tr/2), and so we
could conclude that Tan is invertible. But Theorem 6, as it stands, does not apply
to Sin on [ - Tr/2, Tr/2] or to Sec on [O, Tr/2), because the derivatives of these functions
vanish at the endpoints - Tr/2, Tr/2, and 0. To take care of such cases, we need the
following:
Theorem. If f is differentiable on an interval I, and f'(x) ¥- 0 at every interior point of

I, then f is invertible.
Here by an interior point of I we mean a point of I which is not an endpoint.
Reread the proof of Theorem 6 and see whether it proves this more general theorem.
If so, say so and explain. If not, furnish whatever additional reasoning is necessary.
40. It might also be convenient to have the following generalized form of the uniqueness
theorem (of Section 3.8). Here we require that F'(x) = G'(x) at all interior points of
the interval I.
Theorem (?). Let F and G be differentiable functions, defined on the same interval I, and
let a be a point of I. If (i) F(a) G(a) and (ii) F'(x)
= G'(x) for every interior point
=
x of I, then (iii) F(x) = G(x) for every point x of I.

Reexamine the proof of Theorem 2 of Section 3.8, and see whether it proves the more
general theorem above. (If not, complete the proof.) Then name a case in which the
more general theorem is more convenient to use.
4.8 SIMPSON'S RULE. THE COMPUTATION OF 11
In Section 3.8 we developed a method for evaluating definite integrals. To find
f !(x) dx,
where f is continuous, we first set up the function
F(x) = f'f (t) dt.
Then F'(x) = f(x), for every x. We find another function G, such that G'. = f
Then F and G have the same derivative f; and by adding a constant to G, we· get a
function, say H, such that H' G' = = f and H(a) = 0. Since F(a) = 0, we know
by the uniqueness theorem that F(x) = H(x) for every x. Therefore
f f(t) dt = H(b).
It is possible to write a theorem which sums this up very briefly:
Theorem 1. If f is continuous, and
G'(x) = f(x) (a � x � b),

then
f !(x) dx = G(b) - G(a).

4.8 Simpsons' Rule. The Computation of TC 177
Proof For each x, let
F(x) =if(t) dt.

Then
F'(x) = f(x), and F(a) = 0.
Let
H(x) = G(x) - G(a).
Then
H'(x) = G'(x) = f(x), and H(a) = 0.
Therefore F(x) = H(x) for every x, and so F(b) = H(b). Therefore
ff(t) dt = H(b) = G(b) - G(a).
The proof reproduces the procedure that we have been using all along. G is the
first G that we try, with G' = f; and His the function that we get when we adjust the
constant.
But in many cases it is hard to find a known function which has a given function
fas its derivative. For example, if we had never heard of tan, Tan, or Tan-1, then we
would have had no chance at all of finding a known function G such that
G'(x) 1-.
1 + x2
= -
Later, we shall learn more and better methods for attacking such problems. But no
method, and no system of methods, works all the time. Therefore we often need to
use numerical methods, to calculate definite integrals approximately.
One way is the following. Suppose that we didn't know anything about deriva
tives, but we needed to find H (1 - x3) dx approximately. We might divide the inter
val [O, 1] into 10 subintervals of length 0.1, and add the areas of the circumscribed
rectangles.
y
i= xi= Yi= ai =
0 0 1 0.1
1 0.1 0.999 0.0999
2 0.2 0.992 0.0992
3 0.3 0.973 0.0973
4 0.4 0.936 0.0936
5 0.5 0.875 0.0875
6 0.6 0.784 0.0784
7 0.7 0.657 0.0657
8 0.8 0.488 0.0488
9 0.9 0.271 0.0271
0.7975
Here the areas of the ten circumscribed rectangles are
and their total area is 0.7975. This gives
A = f(1 - x3) dx � 0.7975 = A1.
The approximation A � A1 is not very good: by an easy calculation based on Theorem

I, we get the exact answer
f(1 - x3) dx = 0.7500.
We might also have used inscribed rectangles. Their total area would be
0.7975 - 0.1000 == 0.6975 = A2•

(Why?) The approximation A � A2 is not very good either. But their average is
considerably better.
Aa =
t(A1 + A2) == 0.7475 � 0.7500.
The sum A3 has a geometric meaning: it is the sum of the areas of the inscribed trape
zoids.
CS:J
I I
I I
I I
Over each of the little intervals, the area of the trapezoid is the average of the areas
of the inscribed and circumscribed rectangles; and it is not hard to check that the
same is true of the sums. This helps to explain why the approximation A � A3 is
4.8 Simpson's Rule. The Computation of 7t 179
reasonably good; we have approximated the graph of f by an inscribed broken line,

and used the area under the broken line as an approximation of the integral.
In practice, however, nobody uses the approximation A I:::::! A3, because there is
another method which gives better results without any extra work. This method is
Simpson's rule The scheme is as follows.
..
g y
Suppose that we have a functionf, whose values we can compute, on an interval

[a, b]. We find a quadratic function
g(x) = Ax2 + Bx + C
which agrees wlthf at a, at b, and at the midpoint of [a, b]; and we use the approxima
tion
f!(x)dx fg(x) dx.

�
Here by a quadratic function we mean a function given by a formula A,x2 + Bx +

C. We allow the case A = 0, and so the graph may turn out to be a line instead of a
parabola. In any case, the integral on the right is easy to calculate: if
G(x) = :i x3 + !!. x 2 + Cx ' ·
3 2
then
G' = g,
and so
fg(x)dx = G(b) - G(a).
In the figure above, the approximation looks good, because the errors on the
two halves of [a, b] seem to cancel each other out. Most of the time, we cut [a, b]
into a certain number of little intervals [a;, ai+1]; we then use Simpson's rule on each
of the little intervals, and add the results.
· We shall now develop a shortcut formula for Simpson's rule, in a special case.
Theorem 2. Let g be a quadratic function, and let k be a positive number. Then
Jk g(x)dx
-k
=
�(Yo
3
+ 4y1 + Y2),
where Yo = g(-k), y1 = g(O), and y2 = g(k).

Before proving that this formula is true, let us first check it, in a simple case, to
make sure that it is not absurd. One of the possibilities is that g(x) = l for every x.
In this case, the integral on the left is equal to 2k. Thus our fom.Jla says that
2k = � (1 + 4 + 1),
3
which is correct. Any time you wonder whether you have remembered Simpson's
rule correctly, you should check by this method; the check uncovers the most common
errors in recollection.
We proceed to the proof. We have
g(x) = Ax2 + Bx + C.
Let
A B
G(x) - x3 + x2 + Cx'
2
= -
3
so that G' = g. Then
fkg(x) dx =
G(k) - G(-k) = iAk3 + 2Ck.
(The algebra here is straightforward.) We need to express A and C in terms of

y 0, y1' y2, and k. Evidently C is no problem:
C = g(O) = y1.
To find A, we use
2
Yo = Ak - Bk + Ak2 + Bk +
C, Y2 = C,
Yo + y2 = 2A'k2 + 2C, Yo + y2 = 2Ak2 + 2Ji.
We can now solve for A:
Our expressions for A and C now give
f g(x) dx
k
= iAk3 + 2Ck =
k
- (y0 - 2Yi + Y2) + 2kYi
-k 3
k
= 3 ( Yo + 4y1 + Y2),

Let us try Simpson's rule on the function
1
f(x) -1 � x � 1.
x + 2
= --
,
Here we have
k = 1, Yo= 1,
y
3
1
2 f(x) =
x+2
--
- 2�1 -�
- 1 -+ --''-- '---
2 --X
I
� � � � �
I
I
The rule gives
Jl-1 �
x
� t(l
+ 2
+ 2 + t) � 1.11.
Later, we shall find ways to calculate this integral as exactly as we please. It will then
turn out that the right answer, correct to four decimal places, is 1.0986. In this case,
the approximation is good, in spite of the length of the interval [ -1, 1], because the
portion of the graph off that we are dealing with is very close to its approximating
parabola.
y
-1
Let us now try

1
f(x) = , -1 � x � 1.
1 + x2
--
Here we have
k = 1, Yo= i, Y1 = 1,
The rule gives
1 dx � Ht
J-11 + 4 + t) t � 1.67.
x2 +
--- =
Since
-1 x = 1
D Tan --- ,
1 + x2
the right answer is
f1 ---
dx
= Tan
-1 1 - Tan-1 ( l)
- =
7T
- -
-11 + x2 4
= !!...�
. 1.57.
2
Here the error is about 0.10, which is not very bad. To get better results, we need to
cut up our intervals into smaller pieces. The first step in deriving the necessary
formulas is to generalize Theorem 2, to take care of the case in which the origin is not
necessarily the midpoint of the interval over which we are integrating.
Theorem 3. Let g be a quadratic function, and let k be a positive number. Then

ia+2k k
g(x) dx = -
(y0 + 4yi + J2),
a 3
where
Yo = g(a), Yi= g(a + k), Y2 = g(a + 2k).
The easiest way to see this is to move the graph a + k units to the left, so that the
point (a + k, 0) falls on the origin. When a parabola (or a line) is moved in this way,
it is still a parabola (or a line); the integral does not change, and neither do the
numbers k, y0, Ji, and y2• Therefore Theorem 3 is a consequence of Theorem 2.
Consider now a functionf, on an interval [a, b]. We cut up the interval [a, b]
into an even number 2n of little intervals, each of length
b-a
k= .
2n
The division points are x0, Xi, . . . , x2n, as shown in the figure for n = 2.
Yt f
Yo
On the interval [x0, x2) = [a, a + 2k], Simpson's rule gives

ia+2k k
f(x) dx R::i (y0 + 4yi + J2), -
a 3
where Yi =/(xi); for each i. On the interval [x2, x4] = [a + 2k, a + 4k] we get
ia+4kf(x) dx R:i
k
- (Y2 + 4y3 + y4).
a+2k 3
For the 2n little intervals we have
Ja(bf(x) dx R:i � (y0 + 4y1

3
+ 2Y2 + 4y3 + 2y4 + · · ·
+ 4Y2n-1 + Y2n).
This formula is the final form of Simpson's rule. Let us try it, with k = 0.2, to get a
better approximation of
11 d x
-1 x + 2
.
The computation looks like this:
i= xi= Yi=
0 -1.0 1.0000 1 1.0000

1 -0.8 0.8333 4 3.3332
2 -0.6 0.7143 2 1.4286
3 -0.4 0.6250 4 2.5000
4 -0.2 0.5555 2 1.1110
5 0 0.5000 4 2.0000
6 0.2 0.4545 2 0.9090
7 0.4 0.4167 4 1.6668
8 0.6 0.3846 2 0.7692
9 0.8 0.3571 4 1.4285
10 1.0 0.3333 1 0.3333
16.4795
This gives
1
f � 0· 2
(16.4795) 1.0986.
J_1 x + 2
R:i R:i
3
This answer is actually correct, to the fourth decimal place. Obviously, however,
we must have been lucky: Simpson's rule is not supposed to be exact, and, besides,
we were carrying only four decimal places in the calculation.
When you use Simpson's rule, it is a good idea to use a table like the one shown
above. Make sure that the last entry in the fourth column of your table is 1 and not 2.
We have postponed until now the presentation of Simpson's rule, because this
is the first point at which we can do something interesting with it. The interesting
thing is as follows. We know that
11---
dx = Tan-1 1 0 =-
o1 + x
2 - Tan-1 7T .
4
Therefore
Applying Simpson's rule, we can thus get a numerical approximation of Tr. This is
Problem 1 below.
PROBLEM SET 4.8
1. Apply Simpson's rule to the function
1
f(x) = (0 � x � 1),
1 + x2
with k = !. Check your answer against what people have been telling you about 'TT.
If you want to use k = 0.1, to get a more exact approximation, it might occur to you
to use a slide rule to calculate they/s. Would this be a good idea? Why or why not?
How about five-place log tables?
f(x) = 3x3 - 5x2 + 1 (-2 � x � 2),
with k = 2. Then calculate J:2/ (x) dx exactly, and compute the error in your approxi
mation.
f(x) = x3 + x2 - 17 (-100 � x � 100),
with k = 100. Then calculate the integral exactly, and compute the error in the
approximation.
4. Apply Simpson's rule to
f(x) = x3 - 2x + 3, (-1 � x � 1),
with k = l Compute the error.

5. Apply Simpson's rule to
f(x) = x4 - 2x + 3,
over the same interval as in Problem 4, using the same k, and compute the error.
6. There ought to be a theorem which accounts for some of the results that you have been
getting. State and prove the theorem.
.
f(x) =1 - x3,
on the interval [O, 1], using k = 0.1. (This is the integral which we investigated in the
text above, using inscribed rectangles, circumscribed rectangles, and finally trapezoids.)
*8. Given a positive number k and numbers y0, y1, and y2, write an explicit formula for a
quadratic function g such that g( -k) y0, g(O) =Yi. and g(k)
= y2 That is, write
= •
an expression of the form

g(x) = Ax2 +Bx + C,
in which the coefficients A, B, and C are expressed algebraically in terms of k, y0,

Yi• andy2•
4.9 Exponentials and Logarithms 185
*9. Does the theorem that you proved in Problem 6 hold only on intervals of the type
[ -k, k] or does it hold on any interval [a, b]? Proof or refutation?
After finishing Problem 1, you may want to try a smaller k, to get a better approximation
of 11'. As a check,
11' = 3.14159265,
correct to eight decimal places.
In Appendix F, at the end of the book, you will find a theorem which enables us, under
some conditions, to set a limit on the error in Simpson's rule.
4.9 EXPONENTIALS AND LOGARITHMS
For the case in which the exponents are positive integers, exponentials are part of
elementary algebra. We begin with:
Definition. For each positive integer n,
xn = xxx . . . x (to n factors).
It is then easy to see, simply by counting factors on the left and on the right
that the familiar laws of exponents hold:
(A)
(xmr = xmn. (B)
If n is a negative integer, then -n > 0, and for x :;if 0 we define
Thus, for example, for n = - 3 we have
3 1 1
x- = --
-
x-C-3l x3 ·
For x :;if 0, we define
It can be shown that, if x :;if 0, then formulas (A) and (B) hold for all integers m and n.
When the exponents are allowed to range over all real numbers, exponentials
cease to be part of elementary algebra. In this section we shall state the facts about
exponentials and logarithms, but will make no attempt to verify them. (In the follow
ing two sections, we shall see how these facts fit together to make a logical theory.)
We begin with a positive base and a rational exponent.
1) Suppose that a > 0, and that x is a rational number p/q (where p and q are
integers and q > 0). We want to define ax = a'IJfa in such a way that (A) and (B) will
continue to hold. For (B) to hold, we must have
(a'IJfo)q = aP.
That is, aPfa must be the qth root of aP. Hence the following:
Definition. If a > 0 and q > 0, then
a'Pfa= ,:;a'P.
Here we cannot allow the case a < 0. For a = -1, we would get
( -1)1/3 = � -1 = -'- 1,
( -1 )2/6 = � ( -1)2 = �1 = 1.
Thus, for a < 0, a'Pfa would depend not merely on the number that we use as an
exponent but also on the notation in which the number is expressed. This would lead
to nothing but trouble.
2) It is a fact that for a > 0, and x and y rational, the following laws hold:
a"'. av= a"'+v, (A)

(ax)v =
a xv, (B)
aO = 1. (C)
3) The rational numbers on the x-axis do not fill up the x-axis, because every interval,
however short, contains irrational numbers. Therefore the set Q of all rational
numbers forms a sort of infinitely dotted line. So far, the function/(x) =ax has been
defined only for rational values of x. Therefore/is a function Q-+ R+, and the graph
is an infinitely dotted curve, as in the figure below. Note that/(x) > 0 for every x,
because a"' = a'flfq, which is the positive qth root of the positive number aP.
a>l
\ I
\ I
\ I
\ I
' I
' I
' /
'
',1 "' ,,/
f(x) =a", x in Q
............
a<l
--
.......... __
j: Q-->R+
It is a fact that the definition of this function can be extended so as to give a new
function:
/: R-+ (0, oo)
X H ax> 0,
defined for every x, such that f is continuous and satisfies (A), (B), and (C). For
a = I, we have f(x) = l'" = 1 for every x. But for a > 0 and a � 1,f is invertible.
4) We now define loga as the inverse of f(x) =

a'" (a � 1). That is,
y = lo�x ¢>- av= x,
by definition. The image of the exponential function includes all positive numbers.
Therefore the domain of its inverse includes all positive numbers, and we have a
function
log,,: (0, co ) __,.. R.
Logarithms to the base a obey the following laws:
loga xy = loga X + logay, (A')
loga b x
= X loga b (b > 0, b � 1), (B')
loga 1 = 0. (C')
In fact, these are derivable from (A), (B), and (C).

Since the logarithm and exponential are inverses of each other, we have
loga ax= x,
alogax =
x.
And the graph of either of these functions is the reflection of the graph of the other
across the line y = x.
y
y=f(x)=ax, a>l
5) We now want to calculate the derivative of the logarithm. By definition,

1. loga (x0 + 1:1x) - loga x0
, x0
I oga = lm . .
t:.x-o 1:1x
Using the laws (A'), (B'), and (C'), we express this in the form
!:1x 1/
t:.x
1 x0 + 1:1x
Jim - loga = lim loga ( 1 + - )
t:.x-o 1:1x x0 t:.x-o x0
f:1x "'o-'
= lim - loga
t:.x-o
[
1
x0
( 1 + -
x0
) t:.x] .
Let
Ax
h = .
Xo
Since x0 is fixed, h --+ 0 as �x --+ 0. This gives
log� x0 = _!_ lim loga (1 + h)11h.

Xo h-+O
If loga has a derivative, then the limit on the right-hand side exists, and conversely.
Suppose that (1 + h)1'" approaches a limit, and call this limit e. Thus
e = lim (1 + h)11".
h-+O
Suppose that loga is continuous, so that the limit of the logarithm is the logarithm of
the limit. Then
log� x0 = l. loga lim (1 + h)11h = l. loga e.

Xo h->O Xo
Thus
D loga x = -x1 loga e.
Since e1 = e, we have
log. e = 1;
and so for a = e our differentiation formula takes the form
D log. x = -
1
x
.
Considering the complications of the preceding discussion, the simplicity of this

formula is surprising; it is also 'important (see the next section).
6) On the basis of the preceding formulas, it is easy to find Da", for a ¥- 1. (For
a = 1, we have no problem.) Let
f(x) = loga x,
so that
J-1(x) =a'".
The general formula
J(j-1(x)) = x
thus takes the form
loga a'" = x.
Since
Du loga u = -1
u
loga e,
the chain rule gives:
Cx loga e) Da" = 1.
Therefore
Da" = --
1
loga
e
a'".
In particular, for a = e we have

De"'= e"'.
A final simplification: we assert that
1
-- =log. a.
loga e
Proof. Let
x = loga e, y =loge a.
Then
a"'= e,
by the definition of the logarithm. Therefore
(a"') 11 = e11, and '"11

a = a.
This holds when xy = 1. Since the exponential function is invertible, it cannot take
on the same value twice. Therefore the equation can hold only when xy = I. There
fore
1
-= y,
x
which was to be proved. Thus we can write
'"
Da'" = a log. a.
This is better, not just because it avoids a fraction, but also because e is one of the
two bases for which tables of logarithms are published.
Throughout the following problem set you may assume that the statements
made in this section are true. (They will be proved in the following two sections.)
For convenience of reference, we give a summary.
Laws of Exponentials (a > 0)

a) a'" · a11 = a'"+11, b) (ax) v = axv,
c) a0 = 1, d) a'" > 0 for every x,

e) Da'" =a"' log. a, f) e = l im1i-o (1 + h)1f1i.
Laws of Logarithms (a > 0, a � 1)
g) loga xy =lo� x + logay, h) log0 bx = x loga b (b > 0),
i) loga 1 = 0, j) D loga x = l/(x log. a) ,
k) loga a x= x, 1) aloUa x = x.
PROBLEM SET 4.9
Find the first and second derivatives of the following functions.
1. x loge x 2. xex 3. xe2"'

4. e"' cos x 5. e'" sin x 6. iex(sin x +cos x)
7. log. x2 8. [loge x]2 9. log. xsoo
10. esin a: 11. 10"' 12. log. e"'

13. [log. e]"' 14. elog, "' 15. x2e "'
2
16. e "'• 17. xe"' 18. log. (x3)
19. log. (1 - x) 20. log. (e"' + 1) 21. e"'-1
22. el-X 23. log. sec x 24. e•"'
!h
25. log. sin x 26. log. cos x 27. log. (sec x + tan x)
28. log.
J 1 -x
29. log. (csc x + cot x) 30. log. (x + v x2 + 1)
l
31. Show that for every x > 0,
log. x =
i1
"' -dt.
t
32. Show that if a and b are positive and different from 1, then
1
logb a =
1--
.
oga b
33. Show that, under the..same conditions,
logb x = (loga x)(logb a).
34. Show that, if a and b are positive and different from 1, then
[Hint: What is a10g• b, and why?]

35. The function
f(x) = e "'
has the property of being its own derivative. But it is not the only such function,
because for every k,g(x) = ke"' has the same property. We have, however, the following
theorem.
Theorem. If g ' (x) = g(x), on ( - oo, oo), then there is a constant k such that
g(x) ke"'
= for every x.
That is, g(x)/e"' is a constant.
Prove this.
36. Show that the function f(x) = e"' is completely described by the conditions
f'(x) = f(x) ( - oo <x < oo ), (1)
f(O) = 1. (2)
That is, show that (1) and (2) imply that f(x) = e"' for every x.
37. Show that the function f(x) = e-"' is completely determined by the conditions
f'(x) =
-f(x) ( -oo <x < oo ) (1)
f(O) = 1. (2)
That is, show that (1) and (2) imply that f(x) = e-"' for every x.
4.10 The Functions In and exp 191
4.10 THE FUNCTIONS In AND exp
In the preceding section, we gave a sketch of the way that logarithms and exponentials
ought to behave, postponing both the proofs and also the basic definitions. We
shall now fill these gaps.
If you review the formulas of the preceding section, you will see that after con
siderable complications in the middle, we got a formula that looked simpie:
1
D loge X = - .
x
This enables us to write a f ormula for loge:
f,"' t dt.
1
loge x = -
1
If the theory works, then this formula must be right: the functions on the two sides
of the equation have the same derivative (namely, 1/x), and they have the same value
at x = 1 (namely, 0); and so it follows by the uniqueness theorem that they are the
same function.
We shall use the function f: (l/t) dt as the foundation of the theory of exponentials
and logarithms. The scheme is to investigate the function ff (1/t) dt, learn its proper
ties, and then define all our other functions in terms of it. Thus, at the beginning,
we shall investigate f: (1/t) dt without assuming that we know anything about
logarithms, or about exponentials, or about the number e. To emphasize that we are
starting afresh, we give the function a new name In. (Here In is suggested by
natural
logarithm.) And the official theory
. begins with the definition of In in terms of an
��:
For each x > 0,

"' dt
Definition.
ln x =
J1
-
t
.
Soon we shall show that every real number y is equal to In x for some x. For this
purpose we shall need:
Theorem 1 (The no-jump theorem). If f is continuous on [x1, x2], then f takes on

every value betweenf(x1) andf(x2).
192 Trigonometric and Exponential Functions 4
That is, if f(xi) < k < f(x2), then there is an x, between Xi and x2, such ti
f(x) = k. And ifj(x2) < k < f(xi), then the same conclusion follows. This theor1
will be proved in the next chapter.
Our first few theorems on the function In are easy.
Theorem 2. D In x = I/x.
Proof. This follows from the definition of In and the formula for the derivative
the integral.
Theorem 3. In 1 = 0.
This is obvious.
Theorem 4. For every k, x > 0, D,,, In kx = I/x.

Proof. By the chain rule,
1 1
D"' Inkx = - · k =-
kx x
Theorem 5. For every a, b > 0, ln ab = ln a + In b.
Proof. The trouble with this theorem is that it does not appear to involve any fu1
tions. To prove it, we first restate it, using k for a and x for b. It then says that 1
every k, x > 0,
Inkx =Ink + In x.
---- �e proof is now as follows. Let
f(x) =Inkx, g(x) =Ink + In x.

Then
f'(x) = _! = g'(x), j(l) =Ink,

x
and
g(l) = Ink + In 1 =Ink + 0 =Ink.
By the uniqueness theorem,f(x) = g(x) for every x.
We now want to show that the graph of In looks approximately like the drawi
above. The figure suggests that In 1 = 0 and In' 1 = 1. These things we alrea
know. Other things suggested by the figure are conveyed by some of the following
theorems.
Theorem 6. In is invertible.
Proof We know that In' x = 1/x; and I/x -:F- 0 for every x. Therefore In' x -:F- 0 for
every x. By Theorem 6 of Section 4.7, In is invertible.
Theorem 7. For every x > 0, and every positive integer n,

In xn = n In x.
Proof. Obviously this formula holds when n = 1. And if it holds for any particular
integer n, then it holds for the next integer n + 1. Proof:
In xn+i = In (x xn) = In x + In xn
·
= In x + n In x = (n + 1) In x.
Therefore, by induction, we have In xn = n In x for every x, which was to be proved.
A number M is called an upper bound for a function f if

f (x) � M for every x.
If there is such a number M, then we say that f is bounded above. (For example,
the sine is bounded above, because sin x � 1 for every x.) If no such number M
exists, then we say that/ is unbounded above. (For example, ifj(x) x, for every x, =
then/ is unbounded above.)
Theorem 8. In is unbounded above.
Proof In Theorem 7, take x = 2.
Then
In 2 n = n In 2 ,
for every n. And In 2 > 0, because In 2 is the area of a region. Therefore In cannot
have an upper bound; no number M is greater than or equal to all of the numbers
n In 2 , because
M
n ln 2 > M whenever n > - .
In 2
Theorem 9. In (1/x) = -In x.
Proof
In � + In x = In e x)
· = In 1 = 0;
and from this the theorem follows.
Theorem 10. In 2-n = -n In 2.

Proof By Theorems 7 and 9.
A number mis called a lower bound of a function/ if m �f(x) for every x. If

there is such a number m, then we say that f is bounded below. (For example, the
sine is bounded below, because -1 � sin x for every x.) If no such number m
exists, then we say that f is unbounded below. (For example, if f(x) = x, then f is
unbounded below.)
Theorem 11. The function In is unbounded below.
Because no number mis less than or equal to all the numbers In 2-n = -n In 2.
Theorem 12. Every real number is a value of the function In. That is, every number y
is equal to In x for some x > 0.
Proof Since In is unbounded both above and below, it follows that every number
y lies between twoYalues of In. If In x1 < y < In x2, then it follows by Theorem 1
that y = In x for s�e x. Thus the image of the function In is the entire interval
R = (-oo, oo).
We know by Theorem 6 that In is invertible. Its inverse will be denoted by exp.
That is:
Definition. exp = ln -
1.
Since In x will turn out to be log. x, this means that exp x will turn out to bee"'.
But we should not use the notation e", at this stage, because we have not yet defined e
in the present treatment.

The graphs of In and exp are shown in the figure opposite. Since exp and In are
inverses of each other, we have:
Theorem 13. exp In x = x.

Theorem 14. In exp x = x.
These are instances of the general rule
1-1(f(x)) = J(f-1(x)) = x.
As always, for functions which are inverses of one another, the image of exp is the
domain of In. Therefore
Theorem 15. exp x > 0 for every x.

exp
y=x
/
/
/
/
/
,,
/ In
/
/
/
/
This theorem is also easy to see graphically, in the figure above. The graph
of In lies to the right of the y-axis. Reflecting this graph across the line y = x, we get
the graph of exp. Therefore the graph of exp lies above the x-axis.
Theorem 16. exp 0 = 1.
Because In 1 = 0.
Theorem 17. exp (k + x ) = (exp k)(ex� x) for every k and x.
Proof. Both sides of the equation have the same In:
ln exp (k + x) = k + x,
because In and exp are inverses of each other. And
In [(exp k)(exp x)] = In exp k + In exp x = k + x.

Since In never takes on the same value twice, the theorem follows.
Theorem 18. exp' = exp.
Proof. We know that ln exp x = x. Since In' u = 1/u, the chain rules gives
1
(In' exp exp' -- exp x = 1, exp' exp x,
I
x) x = 1, x =
exp
x
for every x. Therefore exp' = exp, which was to be proved.
We now have functions ln x and exp x which have the properties that the functions
log. x and e"' are expected to have. A natural next step is to find a number e, and
define the exponential function e"', in such a way that exp x = e"'. We shall do this in
the next section. Meanwhile we give a quick summary of this one.
f,.,
Definitions
dt
a) In x = - (x > 0),
1 t
Laws for In
c) In 1 = 0, d) In xn = n In x (x > 0),
e) In kx = In k + In x (k, x > 0), f) In' x = I/x (x > 0).
Laws for exp

g) exp 0 = 1, h) ·(exp k)(exp x) = exp (k + x),
i) exp x > 0 for every x, j) exp In x x = (x > 0),
k) In exp x = x for every x, I) exp' = exp.
PROBLEM SET 4.10
Some of the problems below are to be solved by any method that works, including
methods based on the unproved results of Section 4.9. Some, however, are supposed to
be worked strictly on the basis of the theory developed in this section; and these are stated
in the notation of In and exp. Thus, if the problem uses the notation a"', Ioga x, then the
solution may use the theory in Section 4.9; but if the problem uses In, exp, then the solution
should also.
Find the derivatives of the f ollowing functions:
1. ln2 x 2. In In x 3. In (x2 + 1) 4. In x2 + 1
5. exp x2 6. [exp x]2 7. exp (2 In x) 8. In (exp x2)

9. exp sin x 10. sin (exp x) 1 1. exp (x In x) 1 2. e"' log, x
13. In sin x 14. sin In x 15. X"' (X > 0) 16. (sin x)sin x (sin x > 0)
17. We found that the function
ln x =
lx dt
-
1
t
was unbounded above. Is this true also for
f(x) =
Jx dt
�?
1 vt
Why or why not?
18. How about the function
g(x) =
lx dt
2?
1 t
19. Given
h(x) =
l"'2 dt
-- (0 < x < oo),
. 1 vr
find h'(x), by any method. Note, however, that you are not being asked to calculate h;
you are being asked only to calculate h'.
20. Given
rsinx
f(x) = Jo vl+t2 dt,
findf'(x).
4.11 Exponentials and Logarithms. The Existence of e 197
1
2 . Given
f(x) =
fanx v1 + t2dt,
0
find f'(x).
22. Given
g(x) = fxdtt'
i
findg'(x).
23. Given
h(x) =exp (f' �dt),

find h'(x).
24. Find
sinx + 1
lim
x�3"/2 X-37TI2 .
(By far the easiest way to solve this problem is to think of a geometric meaning for it.)
tanx-1 lnx In x2
25. Find Jim 26. Find Jim . 27. Find Jim 1. --
x�1X - 1
--
X-+1if4 X-7T14 . x-..1X -

tanx + 1 exp ( 2x)- l
28\� Find Jim 29. Find Jim
i x�-rr/4 X + I4 .
1T x�o x
30'. Using Simpson's rule, compute an approximation of In 2. To four decimal places, the
right answer is 0.6931; and if you cut up the interval [1, 2] into ten parts, you get a
good approximation.
31. Show that for x � l,
lnx�x-1.
Show that for 0 < x < 1, the same inequality holds.
32. Given f (x) = x-1, find a formula for 1-1(x), and sketch both functions on the
same set of axes.
Show that expx � x + 1, for every x. [Hint: Try to use a known property of In.]
33. Let k = Jn-1 l. Show that k > 2.
34. Show that k < 4.
4.11 EXPONENTIALS AND LOGARITHMS. THE EXISTENCE OF e
In Section 4.9, we wanted to define e as the limit of the function
f (h) = (1 + h)lfh,
as h � 0. To investigate this limit, we first need a proper definition of the function
(1 + h)1'"· Since the exponent 1/h varies continuously through real values, we need
a definition of the exponential ax, where a > 0 and x is not necessarily rational.
The right definition is not hard to find. We know that if n is a positive integer, then
In n
a = n In a.
Therefore
a" = exp (n ln a) .
If the laws of logarithms hold for all exponents, then
ln a"'= x ln a,
which gives
"
a = exp (x ln a) .
We take this last equation as our definition of the exponential function a". Thus:
Definition. For a > 0, a" = exp (x ln a) .
This gives:
Theorem 1. ln a" = x ln a.
We are now ready to show that (1 + h)1f1i approaches a limit, as h ---+ 0. Let
f(x) = (1 + /
x)1 x.
Then
1
lnf(x) = - ln (1 + x).
x
We now replace x by �x, and observe that
\ 1
ln f
, (�x) = - ln (l + �x)
�x
!
ln (1 + �x) - In 1
�x
This last fraction is the fraction whose limit is ln' 1, by definition of the derivative.
Therefore
lim lnf(�x) = ln' 1 = t = l,
Ax-o
and
limf(�x) = lim exp lnf(�x) = explim lnf(�x)
Ax-o Ax-o Ax-o
= exp 1 = ln -1 1.
Replacing �x by h again, we have
lim (1 + h)111i = exp 1 = ln-1 1.

h-+O
Now that we know the limit exists, we can use it as a definition of e:
Definition. e = lim1i�o (1 + h)1fh.
And we know:
Theorem 2. e = exp 1 = ln-1 I.
This theorem has a geometric interpretation.

That is, e is the number such that

le -1 dt= 1.
1 t
Later we will find an efficient method of calculating e. In fact,

e= 2.7182818,
correct to seven decimal places. It will turn out that
1 1 1
e=l+-+-+ ..·+-+ ··· '
1! 2! n!
where
n!= 1 · 2· 3 · · · n.
The series on the right is infinite, but the terms diminish so rapidly as n increases
that we get good numerical approximations by using the first few terms.
We expected exp x to bee"'. We can now show that this is true:
e"' = exp (x In e),
by definition of e"'; and since In e= I , we have:

Theorem 3. e"' = exp x, for every x.
As before, we define the logarithm as the inverse of the exponential. That is
y = loga x <=> aY = x.
Since e"' and exp x are the same function, they have the same inverse. Therefore
we have:
Theorem 4. log. x = In x, for every x > 0.

Thus exp really is an exponential, and In really is a logarithm. Once we know the
laws governing e"' and log. x, it is easy to derive the laws governing other positive
bases. The first step is to express Ioga in terms of In. We recall that for a > 0,
aY = exp (y ln a),
by definition. Therefore
v
In a = In exp (y In a) ,
and
In av= y In a.
Since a"' and loga x are inverses of each other,
a
log.x
= x.
In this equation, we take the In of each side, getting
(loga x) In a= In x.
This gives:
Theorem 5. For every x > 0,

x
In
Iogax = - .
In a
Thus the function Ioga is a constant times the function In; and this means that the
extension of the theory from In to loga is easy.
Theorem 6. For every a > 0, � 1,
Ioga xy = Ioga x + Ioga y (x, y > 0), (g)
Ioga b"' = x loga b, (h)
Ioga 1 = 0, (i)
1
Ioga/ X = , (j)
x log, a
Ioga a"' = x, for every x, (k)

log."'
a = x, for x > 0. (I)
Here the formula designations are those of the summary at the end of Section 4.9.
The proofs are as follows.
Proof For a = e, the first three formulas are known to hold, because in this case
Joga = log, = In. If we divide every term by In a, then we get loga throughout, and
the equations still hold. To get Eq. (j), we observe that
In x 1
log� x= D loga x= D - - --
In a x In a
"'
Equations (k) and (1) merely remind us that loga x and a are inverses of one
"'
another. Similarly, we get the laws governing the exponential a by using the fact that
"' Ina
a"' = exp (.JC In a) = e
.
Theorem 7. For every a > 0,
a"'· a11 = a'*11, (a)

(a"')v =a"'11, (b)
aO = 1, (c)
a"' > 0 for every x, (d)
Da" =a"'Ina. (e)
The proofs are as follows.
a) We have
In (a" · a") =In a" + In a11 =xIn a + y In a
= (x + y)Ina = In a"'+v.
Since a"· a" and ax+v have the same In, they must be the same; In is invertible,
and so In never takes on the same value twice.
b) By definition, b11 = exp (yIn b). Therefore
(a")11 =exp (yIn a") = exp (yxIn a)
=exp (xyIn a) = a"Y.
c) a0 =exp (0In a) =exp 0 =1.

d) a"' =exp (xIn a) > 0.
e) Da"'= D exp (xIn a) = [exp (xIna)] In a, by the chain rule. Therefore Da"' =
a"'In a.
This completes the program that was sjs,ewked_jn Section 4.9. There are, however,
some things that we still need to check. In the elementary theory, we stated:
Definition 1. For every positive integer n, and every real number x,
(to n factors).
In the new theory, we stated:
Definition 2. If x is positive, and n is any real number, then
[xn] =exp (nIn x).

We have used ( ) and [ ] to distinguish the two definitions. If n is a positive
integer, and x > 0, then both definitions apply; and we need to know that they give
the same answer. In fact, this is true:
In (xn) = nIn x,
by Theorem 7 of Section 4.10. Therefore
(xn) = exp [Jn (xn)] = exp (n In x) = [xn],
by definition of [x"].
Similarly, we now have two definitions of avla.
Definition 1. If a > 0, and x is a rational number p/q, then
Definition 2. If a > 0, and x = p/q, then
a
"'
= a111q = exp (x In a) = exp (� )
In a .
These definitions agree if it is true that
�a" = exp (� a) .
In
The proof is as follows. Let
Then
and
q lny = p ln a.
Therefore
In y =!!.in a,
q
and y = exp (� ) In a ,
We have found that the differentiation formula
Dxk = kx"-1
holds true in certain cases. We first proved it for the case in which k was a positive
integer. Later we found that it held true when k was a n7gative integer. For k = t,
and x > 0, it says that
/
Dx112 _
-
ix112-1
2
_
-
ix-112
2 -
__ 1_
2.Jx,
which is correct. We can now prove the following:
Theorem 8. For every x > 0, and every real number k,
Dx" = kx" -1.

Proof By definition,
x" = exp (k In x).
Therefore
Dx" = D[exp (k In x)] = [exp (k In x)] D[k ln x]
= xk · k · ! = kxk-1•
x
In this section we have presented no new results, except for Theorem 8; we have
merely furnished proofs for the theory sketched in Section 4.9. You therefore have
no new material for problem work. Hence we give the definitions of a new set of
4.ll Exponentials and Logarithms. The Existence of e 203
functions, the hyperbolic functions, and list various identities which they satisfy.
In the following problem set you will be asked to derive these. The theory is simpler
than the theory of trigonometric functions. In fact, once you know about the expo
nerltial function, most of the following formulas have straightforward derivations.
The functions are called the hyperbolic sine, hyperbolic cosine, hyperbolic
tangent, and so on.
Definitions
. e"' - e-"'
smhx = ---
2
e"' + e-"'
coshx = ---
2
e"' - -"'
sinhx .
= --
e
tanhx =
e"' + e-"' coshx
e"' + e-"' coshx

co thx = = -- .
e"' - e-"' sinhx
2 1
sechx = - --
e"' + e-"' cosh x
2 1
--
cschx = -
e"' - e-"' sinh x
Identities
sinh ( -x,) = -sinhx. (1)
cosh (-x) = coshx. (2)

tanh ( - x) = -tanhx. (3)
cosh2 x- sinh2 x =:===,,(. (4)
1 - tanh2x = sech2 x. (5)
coth2 x- 1 = csch2 x. (6)
sinh (x + y) = sinhx cosh y + coshx sinh y. (7)
cosh (x + y) = coshx cosh y + sinh x sinh y. (8)
tanhx + tanh y
tanh (x + y) = . (9)
1 + tanhx tanh y
sinh 2x = 2 sinhx coshx. (10)
cosh 2x = cosh2 x + sinh2 x. (11)
e
"'
= coshx + sinh x. (12)
e-" = cosh x - sinhx. (13)
Derivatives
sinh' x = cosh x. (14)
cosh' x = sinh x. (15)
tanh' x = sech2 x. (16)
coth' x = -csch2 x. (17)
sech' x = -sech x tanh x. (18)
csch' x = - csch x coth x. (19)
PROBLEM SET 4.11
Verify the following. (The numbers in parentheses refer to the numbered formulas
above in the text.)
1. (12) 2. (13) 3. (1) 4. (2)

5. (3) 6. (14) 7. (15) 8. (16)
9. (17) 10. (18) 11. (19)
12. Find the derivative of
F(x) = cosh2 x - sinh2 x.
13. Verify (4). 14. Verify (5). 15. Verify (6).

16. Let
A = sinh (x + y) - sinh x cosy - cosh x sinhy,
B = cosh (x + y) - cosh x coshy - sinh x sinhy.
Show that A + B = 0. (It is not necessary to go back to the definitions to show this.
Try Identities (12) and (13).)
17. Let A and B be as in Problem 16. Show that A - B = 0.
Now verify the following.
18. (7) 19. (8) 20. (9) 21. (10) 22. (11)
23. Express cosh 3x in terms of cosh x.
24. Show that
x > 0 => sinh x > 0.
25. Show that
x < 0 => sinh x < 0.
26. Show that, on the interval [O, oo) cosh is increasing.

,
27. Show that, on the interval ( - oo, OJ, cosh is decreasing.

28. Show that
· cosh x � 1 for every x.
29. Show that sinh is invertible.

30. Show that
cosh x = VI + sinh2 x,
for every x. Note that there is no double sign in this formula; if your derivation leads
to a"±" sign, you musf find a way to get rid of it.
31. Find cosh sinh-1 x.
32. Find D sinh-1 x.

33. Find D sinh-1 2x.
34. Show that cosh is not an invertible function.
35. Let
Cosh x = cosh x (0 � x).
(Compare with the definition of Cos: Cos x = cos x (0 � x � 7T .) ) Show that Cosh is
invertible.
36. Show that
.
Sillh X =
{v cosh2 x - 1, for x � 0,
-Vcosh2x - 1, for x < 0.
37. Show that
sinh Cosh-1 x = v x2 - 1.
38. Find D Cosh-1 x.
39. Firid D Cosh-1 x2•
40. Show that tanh is invertible.

41. Show that
sech x = VI - tanh2 x,
without any double sign in front of the radical.
42. Find D tanh-1 x.
43. Solve for x:

e2"' + e"' - 6 = 0,
and explain why this equation has only one root.
44. Solve for x:

e"' + 2 - 35e-"' = 0.
45. Solve for x, in terms of y: \
e"' + y - .6y2e-"' 0.
I
=
�
46. Find a formula which express s sinh-1 x as the logarithm of an algebraic expression.
Hint: The graph of sinh is the graph of the equation
y = i(e"' - e-"') . (1)
Therefore the graph of sinh-1 is the graph of the equation
x = i(e11 - e-i-). (2)
Here we have reflected the graph across the line y = x, by interchanging x and y in
Eq. (1). Now solve for y in (2), getting
y = ( " · ) .
Then
sinh-1 x = ( · · · ) .
47. Analogously, get a formula for Cosh-1 x.

48. Analogously, get a formula.for tanh-1 x.
The Variation of
5 Continuous Functions
5.1 INTERVALS ON WHICH A FUNCTION INCREASES, OR DECREASES
The function f is increasing if
x < x' => f(x) <f(x').

Similarly, f is decreasing if
x < x' => f(x) > f(x').
y
y
Here x and x' are any points in the domain off Some simple functions are neither
increasing nor decreasing. For example, f(x) = x2 satisfies neither of the above
conditions.
Often, however, we can get a good description of a function by cutting up its
domain into subintervals, in such a way that on each subinterval the function is either
increasing or decreasing. For example, the domain might be a closed interval [a, b],
and the graph might look like this:
206
5.1 Intervals on Which a Function Increases, or Decreases 207
This function is increasing on the interval [x2, x3]. Similarly, f is increasing on

[x0, x1] and [x4, x5], and is decreasing on the interval [x1' x2].
y
f y
I
: I
I
I
I
I
I
I
I
�--+-��--x
J
X2 X X1 X3
Similarly,f is decreasing on [x3, x4].

If a function is differentiable, then we can find out where it is increasing or
decreasing by examining the derivative.
Theorem 1. If f'(x)> 0 at every interior point of I, then f is increasing on I.
We recall that an interior point of an interval is any point which is not an endpoint.
Theorem 1 is a consequence of the mean-value theorem (MVT).
If we had
(?)a,b in l, a< b, f(a)> f(b) (?),
as on the left of the graph above, then the slope of the chord would be
f(b) -f(a)< 0,
b -a
and this would give f'(x)
< 0 for some x between a and b. This is impossible,
because such an x would be an interior point of I. If we had
(?) c, din/, c< d, f(c)= f(d) (?),

as on the right, then the chord would be horizontal, and we would have f' (x)= 0
at an interior point x of I.
In Theorem 1, we allow the possibility that I is an infinite interval. Consider
f ( x)= xz, I= [O, co).

Here
f'(x) = 2x,
208 The Variation of Continuous Functions 5.1
and so /'(x) > 0 at every interior point of I. Therefore f is increasing on I. Note

that when we allow the possibility that/'(x) = 0 at endpoints of I, we are not splitting
hairs; if we required that/'(x) be >O everywhere in I, then the theorem would not
apply in the simple case f(x) x2, I= [O, oo), or to the case f(x)
= cos x, I= =
[7T, 27T]. We don't need theorems to be as general as possible, but we want them to be
general enough to be usable. And it is not unusual to find that/'(x) = 0 at an end
point; in fact, this is what usually happens, when we break up the domain of our
function into the largest possible subintervals on which the derivative does not change
sign. Here f is increasing on 11 and /3, and decreasing on 12; and the derivative
vanishes at the endpoints x1 and x2•
Consider another example:
f(x) = x2 - x (0 � x � 2).
Here
f'(x) = 2x - 1, f'(t) = 0, and f'(x) > 0 on ( t, 2].

In the left-hand figure below, we have used this information, and have plotted
f(t) and/(2). Obviously /(1) = 0, and so we have plotted this point exactly also.
y y
2
f 2
x x
3
-1 -1
The same principle works in the opposite case:
Theorem 2. If f' (x) < 0 at every interior point of I, then f is decreasing on I.
(Proof Apply Theorem 1 to the function g = - f By Theorem 1, g is increasing on

I. Therefore f is decreasing on I.)
5.1 Intervals on Which a Function Increases, or Decreases 209
To apply this theorem to the same function f(x) = x2 - x, on the interval

I= (0, !], we observe that/ is decreasing on this interval, because
f'(x) = 2x - 1 < 0 for 0 < x < t.
We use this information to complete our sketch.

This example doesn't look impressive, because we already knew how to sketch
parabolas. It is not so obvious, however, how to sketch the graph of a cubic function
taken at random, say,
f(x) = x3 + 2x2 - 3x - 4, -2 � x � 2.
This is not a put-up job; it is a "real-life" problem, and nothing is going to come out
even. We need to find out where f' > 0 and where f' < 0. Now
f'(x) = 3x2 + 4x - 3,
so that
-2 ± JTI
f'(x) = 0 when X=
3
Since J13 � 3.6, the roots of the equationf'(x) = 0 are
-6
Since the graph off' is a parabola opening upward, it must look like the drawing
on the left above. Thus
f'(x) > 0 when x < x2,
f'(x) < 0 when x2 < x < x1,
f'(x) > 0 when x > x1.

Therefore f is increasing on [ -2, x2]; f is decreasing on [x2, x1]; and f is increasing

on [x1, 2]. We calculate
f(-2) = 2, f(x2)::::::; 2.1,
f(x1) ::::::; - 4. 9 , /(2) = 6.
This gives us our sketch on the right. (The problems in the following problem set are
not this awkward.)
To apply this method, you need to know how the derivative behaves; and we
may use the same method in investigating the derivative. For example, in the pre
ceding problem we had
2
f'(x) 3x + 4x - 3.
=
If we let
2
g(x) = f'(x) = 3x + 4x - 3,
then
g'(x) = 6x + 4.
Therefore g is increasing for x > -i, and is decreasing for x < -i· Plotting g
exactly, at the points -2, x2, 0, and x1, we get the sketch of f' which is given
above. We know thatf'(x) > 0 for x1 < x < 2, because f' increases, starting at the
value f'(x1) = 0. Similarly, f'(x) > 0 for -2 < x < x2, because on the interval
[-2, x2], f' decreases toward f'(x2) = 0. Similarly in the middle interval [x1, x2].
This idea is simple enough, but it is so useful that we had better record it as a theorem:
Theorem 3. Iff is increasing on [x1, x2], thenf(x) > f(x1) for every x on (x1, x2].
__/L_ I I
We recall that (x1, x2] is a half-open interval;
(x1, x2] = {x I x1 < x � x2}.
We have been talking about·the casef'(x1) = 0.
Theorem 4. Iff is decreasing on [x1, x2], then f(x) < f(x1) for every x on (x1, x2].
PROBLEM SET 5.1
For each function given, state on what intervals the function is increasing, and on wha
intervals it is decreasing; and sketch the graph.
5.2 Local Maxima and Minima, Direction of Concavity, Inflection Points 211
1. f(x) = sin x,
2. f(x) = Sin-1 x, -1;:;:; x;:;:; 1
1
3. f(x) -2;:;:; x;:;:; 2
=
- --2
1 +x '
x
4. f(x) -2;:;:; x;:;:; 2
-1 --2
=
+x '
5. f(x) = x3 - 3x, - 2;:;:; x � 2
6. f(x) = x3 + 3x2 - 2, -1;:;:; x ;:;:; 3
7. f(x) = (sin x + cos x)2 - 1,
8. f(x) = x3 + 6x2 + 9x + 3, -1 ;:;:; x;:;:; 2

'
9. f(x) = ew , 0;:;:; x ;:;:; 1
1 0. f(x) = x In x, 1 ;:;:; x;:;:; 5
11 . f(x) =cos x, 0 ;:;:; x;:;:; 27T
12. f(x) = sin 2x 0 ;:;:; x ;:;:; 1T
13. -2;:;:; x;:;:; 2

4
f(x) = -x +2x2
14. f(x) = xe-w -1;:;:; x;:;:; 2
1
15. f(x) -1 ;;;; x ;;;; 1
-1 --
=
+x 4
x
6
1 . f(x) ---4 -1 ;;;; x ;:;:; 1
1 +x
=
7
1 . f(x) = x cos x - sin x
18. f(x) = x/2 + sin x, 0 ;;=;x;:;:; h
19. f(x) = e"' - 2x, 0;:;:; x;:;:; 2
(Here you are not going to be able to get answers in an exact numerical form. The figure
should indicate plausible approximations.)
20. Investigate the converse of Theorem 1. That is, find out whether the following state
ment is true:
Theorem(?). If (i) f is continuous on [a, b], (ii) f is differentiable on (a, b), and (iii) f is
increasing on [a, b], then (iv)/'(x) > 0 for every x of (a, b).
2 1 . Is the following true?
Theorem(?). If f is differentiable at x0 and f '(x0) > 0, then some chord of the graph of
f has a positive slope.
22. Investigate:
Theorem(?). Let f be a function satisfying (i), (ii), and (iii) of Problem 20. Then (iv')
f'(x) � 0 for every x of (a, b).
5.2 LOCAL MAXIMA AND MINIMA,

DIRECTION OF CONCAVITY, INFLECTION POINTS
Again we consider a continuous functionf, defined on a closed interval [a, b]. In the
figure,f(x2) M; and Mis the largest value off
=
We say thatf has a maximum at x2; and we say that Mis the maximum value off
Similarly,f(x3) = m; and mis the smallest value off We say that/has a minimum
at x3 ; and we say that mis the minimum value off
Here when we speak of maxima and minima, we mean maxima and minima on the
whole domain of the functionf; in this case the domain is [a, b]. Before you know
what is a maximum or minimum, you must first know the domain of the function.
In the figure above, f(x1) is not a minimum value, becausef(x3) <f(x1). But
f(x1) is the smallest value thatf takes on when x is close to x1. We say thatf has a
local minimum at x1. This is abbreviated as LMin. Local minima can occur in three
ways:
y y y
�
I
I
I
I
I
(fV
I
I
I
I m I
I I
I I I I
I
I
I I
I
I
I I I
I I I I I I
x x x
X1-0XJ X1+0 X1 X1+0 X1 -o Xj
1) x1 may lie on an open interval (x1 o, x1 + o), in the domain off; and /(x1)
-
may be the smallest value of the function on the interval (x1 - o, x1 + o). In this
case, we say that/has an interior local minimum at x1. This is abbreviated as ILMin.
2) x1 may be the left-hand endpoint of the domain of/; and/(x1) may be the smallest
value Of j on an interval (X1, X1 + 0).
3) x1 may be the right-hand endpoint of the domain of/; and/(x1) may be the smallest
value off on an interval (x1 o, x1].
-
Thus, for the function/whose graph is sketched at the beginning of this section,
we have local minima at x1 and x3. Note that every minimum is automatically a local
minimum, just as the tallest man in the world is automatically the tallest in his own
neighborhood.
Local maxima are defined similarly. Local maximum is abbreviated as LMax.
A local maximum can occur in three ways:
5.2 Local Maxima and M inima, Direction of Concavity, Inflection Points 213
y y y
� I
I
I I
0
I
I
I
I
!\): I
I
I
I
I
I
I I I I I I I
x x x
Xi-5 Xi xi+5 X1 xi+5 X1-5 Xi
In the figure on the left, f has an interior local maximum at x1• This is abbreviated
ILMax.
There are simple conditions under which a function has an ILMax or an ILMin
at a given point.
Theorem 1. If/is increasing on an interval [x1 - o, x1] and decreasing on an interval

[X1, X1 + o], then/ has an ILMax at X1·
Theorem 2. If/is decreasing on an interval [xi - o, Xi], and increasing on an interval
[xi. X1 + o], then/ has an ILMin at Xi.
In applying these, we use the derivative. If f' > 0 on (xi - o, x1) and f' < 0
on (xi, x1 + o), we can apply Theorem 1. Similarly for Theorem 2. In fact, if you
find out where a function is increasing and where it is decreasing, it is always obvious
where the interior local maxima and minima are; they are at the turning points, where
the graph stops behaving in one way and starts behaving in the other way.
Most of the time, for functions defined on a closed interval, the endpoints of the
interval give either local maxima or local minima. Therefore, if we are investigating a
function for local maxima and minima, we always investigate the endpoints. Of course,
interior local maxima and minima may occur anywhere in the interior of the interval.
In searching for them, we use the theorem suggested by the figure below. If the f unc
tion is differentiable, then at an interior local maximum the derivative must be = 0.
�-+-�---�x��--
X1
f'(x1) =0
Theorem 3. If f has an ILMax at Xi, and f is differentiable at Xi, then f' (xi) = 0.
This is geometrically obvious, and a logical proof is also easy. Let
f(x) - f(xi)
m(x) =
,
X Xi -
so that
Jim m(x) =
f'(x1).
1) Suppose thatf'(xi) > 0. Then the function m(x) must be >0 when x �Xi.
Take x !::::! Xi, with x > Xi. Then
x �Xi and x > Xi => m(x) > 0 and x - Xi > 0
=> m(x)(x - Xi) > 0 => f(x) - f(xi) > 0
=> f(x) > f(xi),
which is impossible, becausef has an ILMax at Xi.
2) Suppose thatf'(xi) < 0. Then the function m(x) must be <0 when x �Xi.
Take x �Xi, with x < Xi. Then
x !::::! Xi and x < Xi => m(x) < 0 and x - Xi < 0
=> m(x)(x - Xi) > 0 => f(x) - f(xi) > 0
=> f(x) > f(xi),

which is impossible.
Since (1) and (2) are both impossible, it follows thatf'(xi) = 0, which was to be
proved.
5.2 Local Maxima and Minima, Direction of Concavity, Inflection Points 215
This is the standard method for finding an ILMax. Given a differentiable function
f, we find the pointsx wheref'( x) 0. Usually there are only a finite number of
=
such points. These are the only possible places where interior local maxima can occur.
Therefore we have only a finite number of values of x to investigate; and when we are
done, our list of interior local maxima is complete.
Note, however, that the converse of Theorem 3 is false: iff' (x1) = 0, it does not
follow that f has a local maximum ( or a local minimum) at x1• For example, if
(fx) = x3, -1 � x � 1, thenf'(O) = 0, butf is increasing on the whole interval
[-1, 1]
y
We have a similar theorem for interior local minima:
Theorem 4. Iff has an ILMin at x1, andf is differentiable at x1, then f'(x1) = 0.
Proof Let
g(x) = -f(x).
Theng has an ILMax at x1. Thereforeg'(x1) = 0. Thereforef'(x1) = -g'(x1) =
-0 = 0, which was to be proved.
y
y
�
I I
I
I
I I
If f' is increasing, on an interval [xi. x2], thenf is concave upward on [x1, x2].
(You ought to be able to convince yourself that this is a reasonable use of language.)
IfJ' is decreasing on [x2, x3], thenf is concave downward on [x2, x3]. In the figure on
the right, x2 is the point at which the direction of concavity changes. Such a point is
called an inflection point. Of course, the direction of concavity can change from up to
down or from down to up. Hence:
Definition. An inflection point of a function f is a point at which f' has either an

ILMax or an ILMin.
Note the way in which these definitions fit together. If you know how to investi
gate (a) increasing, (b) decreasing, (c) interior local maxima, and (d) interior local
minima, then automatically you know how to investigate direction of concavity and
inflection points. The reason is that f', once you get it, is a function, and can be
investigated in the same way as any other function, with the aid of its derivativef".
Wheref' increases,fis concave upward; where/' decreases,fis concave downward;
and where f' has an interior local maximum or minimum, f has an inflection point.
Most of the time, we investigate local maxima and local minima because we
want to find the maxima and minima. We find the maxima and minima, on the whole
domain, by looking to see which local maximum value is the largest and which local
minimum value is the smallest.
Finally, we observe that a function may easily have a local maximum or minimum
at an endpoint at which it is not differentiable. For example, the function f(x) =
x2/3 (0 � x < )
oo has a minimum (and hence a local minimum) at x = 0. The
theory takes care of this case. Since the derivative jx-1/3 is positive in the interior
of the interval [O, )
oo , it follows that the function is increasing, and so it has a
minimum at the left-hand endpoint.
PROBLEM SET 5.2
1 through 19. For each of the functions described in Problems 1 through 19 of the
preceding problem set, find the local maxima, the local minima, the maximum, the minimum,
the inflection points (if any), and the image. (The image will always turn out to be a closed
interval.) Tell where each of the functions is concave upward and where it is concave down
ward.
20. Consider the function defined by the following conditions:
1 1
f(x) = x sin for 0 < x � -
x 'TT
/(0) = 0.
An exact sketch is not practical, because the ILMax and ILMin points are hard to
calculate. Give a rough sketch, however, indicating as well as you can how the function
behaves. Is it continuous at O? Does it have a local maximum or minimum at O? Is it
differentiable at 0?
*21. Suppose that/is both continuous and differentiable on [O, 1]. Does it follow that/has
an LMax or an LMin at 0? Why or why not?
5.3 THE BEHAVIOR OF FUNCTIONS AT INFINITY
So far, we have been discussing functions on closed interva!S: In this section, we shall
consider larger domains, including infinite intervals, such as ( - oo, oo), [O, oo ),
and so on, and also intervals with holes in them. For example, the domain of the
tangent is
D = {x I x � TT/2 + nTT};
and this is an infinite interval ( - oo, )
oo with infinitely many holes in it.
5.3 The Behavior of Functions at Infinity 217
Most of the ideas that we shall be investigating are illustrated by a simple function,
whose domain has a hole in it at 0.
y
x
_) -1
!"'---
1l
I
x
-1 I
I
I
I
1 I
f(x) = x (x�O) I
I
f(x) = --
1 , x�±I
x2-1
A carefi:I inspection of the left-hand graph above will give you an idea of the
meanings of the following statements:
limf(x) = 0, (1)
x--+ oo
lim f(x) = 0, (2)

x-+-co
lim f(x) = oo, (3)

x-+o+
lim f(x) = - oo. (4)

x-+O-
Definitions will come later. Meanwhile, let us look at another example. The
function whose graph is shown on the right above has the following properties:
i) f has an interior local maximum at 0. (At x = 0, the denominator is -1.

Everywhere else near 0, x2 > 0, x2 - 1 > -1, and 1/(x2 - 1) < -1.)
ii) li�f(x) =
-oo.
o:--+1
iii) lim f(x) = 00.

ai-+1+
lim f(x) = - oo.

iv) x-+-1
+
v) lim f(x) = oo.

x-+-1-
Here statements (ii) through (v) mean the things that the figure suggests. An
examination of the formula shows why the figure is right. For example, if x > 1,
and x � 1, then x2 - 1 > 0, and x2 - l � 0. Therefore 1/(x2 - I) is positive and

very large. This is shown in the figure and stated by (iii). Similarly, if x < I and
x � I, then x2 - 1 < 0 and x2 - 1 � 0. Therefore 1/(x2 - 1) is negative and is
numerically very large. This is shown in the figure and stated by (ii).
Let us now make this precise, by stating definitions that we can work with.
Definition. limx�00f(x) = L means that for every E > 0 there is an M such that
x > M => L - E < f(x) < L + E.
L-< --------- --- --------
This is like the definition of limx-xof(x) = L. Roughly,
lim f(x) = L means x � x0 => f(x) � L,
and
limf(x) = L means x � oo => f(x) � L.

x-+oo
In the definitions, the condition x � x0 is expressed by 0 < Ix - x01 < o, and the
condition x � oo is expressed by x > M.
Let us see how our definition of limx_,00 applies to the function
1
f(x) =
-
(x ¥: 0).
x
We claim that
lim ..! = 0.
X-Jof"fJ X
Under the definition, given E > 0, we are supposed to find an M such that
·l
-E < - < E whenever x > M.
x
This is trivial: take M = 1/E. When x > 1/E, obviously 0 < 1/x < E. Similarly:
Definition. limx�-00/(x) = L means that for every E > 0 there is an M such that
x < M => L - E </(x) < L + E.

L-•
In the same spirit:
Definition. limx�x0+ f(x) = oo means that for every M there is a (J > 0 such that
x0 < x < x0 + o => f (x) > M.
y
I
I I
I I
I
M t- _J
I I
------ -
I I
I I
I I
I I
I I
I I
I I
�--L-��--'-�-'--��x
x0 x0+o
That is, you can make f(x) as big as you want (i.e., > M) by taking x to the right
of x0 and very close to x0 (i.e., between x0 and x0 + b.)
We need to talk about one-sided limits (as x � x � or x � x�) because these
often turn out to be different. In some cases, however, the one-sided limits have the
same value. In such cases, limx�x. f (x) must exist, and must be their common value.
Thus
1.im 2
1 = 00.
x-+O X
The following two theorems justify the remarks that were made above about
f(x) = 1/(x2 - 1).
Theorem 1. Suppose thatf(x) > 0 on an interval (x0, x1). If limx�x.� f(x) = 0, then
1.lill - 1 = 00.
x�,,0+ j(x)
Proof M > 0. We need to find a b > 0 such that l/f(x) > M whenever
Given
x0 < x0 + b. Let E = I/M. By the definition of the statement lim,,,_." f(x) = 0,
< x
•
we know that there is a o > 0 such that
f ( x) < E whenever x0 < x < x0 + b.

This is theo that we want: when x0 < x < x0 + o, we have
f(x) <
1,
M
and hence
1
__ > M.
f(x)
(Remember that M > 0, and f (x) > 0 for the values of x that we are interested in.)
Similarly, we have:
Theorem 2. Suppose that f(x) < 0 on an interval (x0, x1). If lim,,_,.,0+ f(x) = 0, then
1
I.lm - = -oo.
+
x->xo f(x)
Proof Given M < 0. Let E = -1/M > 0. Leto be a positive number such that
-E <f(x) whenever x0 < x < x0 + o.

When x0 < x < x0 + o, we have
1 1 1
f(x) > -E, 1 < -€ ' -- > - , -<M.
f(x) E f(x) f(x)
(Here we have been reversing inequalities, because we have been dividing by negative
numbers.)
Following the analogy of the above definitions, you ought to be able to write
your own definitions of the following statements:
Jim f(x) = oo, lim f(x) = - oo,

l.�-+O') x--oo
Jim f(x) = - oo, lim f(x) = oo.

X4-00
Consider now the question of
(?) lim
x-+oc:
(1 !)x (?)
+
X
We use question marks, because it is not obvious that the indicated limit exists at all:
as x---->- 1/x---->- 0, and 1 + 1/x---->- 1.
oo, Therefore we have an "indeterminate form
of the type 100." We recall, however, a similar situation before where we got an
·
answer like this:
Jim (1 + h)lf1i = e = 111-1 l.

h-+0
This was also of the form "1 a:i." And the two are related: if we let
1
f(u) = (1 + u)11u and g(x) = - '
x
then
( �r
1 + = f(g(x)),
and we want to find

limf(g(x)).
x-+ co
This is like the situation in Section 4. 5. There we found:
Theorem. If lim., �xo g(x) = u0 = g(x0) and limu-.uo j(u) = j(u0), then
limf(g(x)) =f (u o ) .
For the case in which x---+ ro, instead of x---+ x0, this theorem is still true, and
the proof is virtually the same. That is:
Theorem 3. If lim.,_.00 g(x) = u0 and limu_.,,0 f(u) = L, then
limf(g(x)) = L.
x-+ oo
Roughly speaking, the reason is that
x � ro � g(x) � u0 � f(g(x)) � L.
In fact, the same result holds if lim.,�00 g(x) = ro.
Theorem 4. If lim.,�00 g(x) = ro and limu� oo f(u) = L, then
lim f(g(x)) = L.
x� oo
These theorems give quick answers to some rather hard-looking problems.

Returning to our discussion of
1
f(u) = (1 + u)llu, g(x) = -
x
, f(g(x)) = (1 ;r+
we get immediately:
Theorem 5. lima:-.oo (1 + 1/x)"' = e.
This limit is used as a definition of e, in some treatments of exponentials and

logarithms. In such a treatment, the formula e = limh�o (1 + h)1fh appears as a
theorem.
PROBLEM SET 5.3
Investigate the following functions for maxima, minima, local maxima, local minima,
direction of concavity, and inflection points. Then investigate for limits of the sort defined
in this section.
1
1. f(x) = (x >'6 0, x >'6 2)
x(x - 2)
2· f(x) =
(x - l)(x -
1
3) ( x ;= l, x ;= 3)
3. f(x) = 2
x -x -6
1
(x ¢ -2, x ¢ 3)
1 1
4. f(x) --
2 - 5. f(x) (x ¢ 0)
=
2
x + 1 x
=
x xz
6. I <x) = --
2 - 7. f(x) = --
2 -
x + 1 x + 1
8. f(x) =
3
x + 1
1
( -1 ¢ x) 9. f(x) = -
x + 1
-3
x +
1 (-1 ¢ x)
IO. f(x) = 3
x -x
1
(x ¢ 0, x ¢ 1, x ¢ -1) 11 . f (x) =
-
1-
x3
+x
-3 (-1 ¢ x)
)"'2
Investigate:
(
1
12. lim 1 + 2 13. lim (1 + cosx)secx 14. Jim o + v':X)11v'x
x-+oo X x-,,12 x-o+
15. Jim (1 + x4)11x' 16. Jim (x)1f(x-ll

x-o x-1
( Ir
Investigate the following, for lim x-oo .
17. f(x) = 1 +-
2x
18. f(x) =
( 2
1 +-
x
r2
19. f(x)(1 L)"'
= +
20. f(x)( �r 21. /(x) ( -rx

l +
l
3. 3r2
= 1 + = (1 + e-xy' 22.
ln x
1 + -
(
2
x
24. Discuss as in Problems 1 through 11
f(x) = (In x)/x (x > 0).
(Here the sticky point is limx-• er,· You ought to be able to figure out what this limit is,
and convince yourself that your answer must be right. But to prove that the answer is
right is an unreasonably hard problem, at this stage.)
25. Find lim x-o+xIn (1/x). You need not prove that your answer is right.
26. Is there such a thing as limx - 'Y) sinx? Why or why not?
27. Is there such a thing as Jim"'_"' (l/x) sinx? Why or why not?
28. Prove the following:
Theorem (The squeeze principle). If
f(x) � g(x) � /z(x) (x �a),

and
Jim f (x) = Jim h(x) = L,
x--oo x-�
5.4 The Introduction of Functions into Geometric Problems 223
29. If you borrow a dollar for a year, at 100 % simple interest, then at the end of the year
you owe 2 dollars. (A certain Marcus Junius Brutus lent money at this rate, in the first
century B.C. He was also an assassin.) If interest is compounded semiannually, then
at the end of the year you owe
(1 + t)2 = t = $2.25.
If the interest is compounded n times a year, then you owe
Suppose now that interest is compounded continuously: the bank passes to the limit,
as n increases without limit, and at the end of the year they charge you the limit. How
much do you owe?
30. Suppose that the basic interest rate is 6%, but interest is compounded continuously,
as in Problem 29. How much do you owe? (To get a numerical answer to this one,
you will need to use one of the tables at the end of the book.)
5.4 THE INTRODUCTION OF FUNCTIONS INTO GEOMETRIC PROBLEMS;

THE USE OF EXISTENCE THEOREMS AS SHORTCUTS
On several occasions already we have been confronted with problems which did not
appear to involve functions, and have solved them by introducing functions.
For example, in Section 3.7 we wanted to find the area under the graph of y =
4
x ,
from o to 1.
!I
y
y=t4
y=x4
F(x)
We solved this problem by attacking the more general problem of calculating the
function
We found that
xs
F(x)
5
and then set x == 1 to get the answer t.

Similarly, in Section 4.10 we wanted to show that
In (ab) == In a + In b,
for every pair of positive numbers a, b. To use the methods of calculus, we had to
introduce functions into the problem. Given k > 0, we set
f(x) =In kx (x > 0),
g(x) = Ink + In x (x > 0).
We then found that/'(x) = g'(x) for every x, and/(1) = g(l). It followed that
f = g; and this proved our theorem.
We use the same kind of method to attack problems in maxima and minima which
may be stated in geometric or physical terms. Consider some examples.
Problem 1. A segment of length 1 has its endpoints on the sides of a right angle.
What position for the segment gives maximum area for the resulting triangle?
y
The first step is to introduce a coordinate system, as shown on the right above.
The endpoints of the segment now fie on the positive ends of the axes.
Let x be the x-coordinate of the endpoint that lies on the x-axis; and let the other
endpoint be (0, y). When x is named, y is determined. Thus there is a function f
which gives y in terms of x. Since
x2 + y2 = 1,
we have
f(x) = .JI - x2 (0 � x � 1).
And for each x, the area enclosed is
A(x) = txf(x) = tx.J1 - x2.
We need to investigate the function A for maxima. Now
A'(x) = [
.! x � + .J1 - x2
2 .Jr x2
] =
.! -x2 � x2)
2
- .J 1 - x2
1 2x2 - 1
= (0 � x � 1).
2 .Ji - x2
Therefore A'(x) = 0 when x = ±/212. Since we are concerned only with numbers
on the interval [O, l], only x = Ii.12 is of interest to us. Here A = t. Any maximum
of A is surely an ILMax, because A(O) = A(I) = 0, and A(x) > 0 for 0 < x < 1.
Therefore our problem is solved, with A = !, if we know the following theorem:

Theorem 1 (Existence of maxima). Iff is continuous on [a, b], then/has a maximum
value on [a, b].
y
y
--(1\1 --�
M
I I
I I
I I I I I
x x
a x b a b=x
The maximum may be an ILMax, as on the left above, or it may be at an endpoint,

as on the right. But in many cases, like the one we have just been discussing, it is
plain that the second of these possibilities does not arise. In such cases, we can infer
immediately that the maximum is an ILMax. If the derivative vanishes at only one
point, then this point must be the maximum.
We shall prove Theorem 1 in Section 5.6. Meanwhile, let us look at some more
applications of it. In the preceding example, there are other functions that we might
equally well have introduced.
y
If the angle at P has measure e (0 � e � TT/2), then

y =sine, x =cose,
and
A(O) = txy =t sine cose =! sin 20.
Therefore
A'(O) = !(cos 20) 2 = t cos W. ·
The only point 0 on the interval [O, TT/2] where A' (0) = 0 is the point where
o = !!. .
4
We claim, without further investigation of derivatives, that this must be where the
maximum occurs. (As in the previous discussion, there must be a maximum some
where; this is not at an endpoint 0 or TT/2; it is therefore an interior local maximum;
at an ILMax, A' (0) = 0; and 0 = TT/4 is the only point of the interval at which
A'(O) = 0.)
Setting 0 = TT/4, we get the maximum value of A as
A(�) =
i sin ( �)
1 · = i sin � = i,
as before.
On reflection, you may find a way to solve this problem by purely geometrical
methods, without taking any derivatives or even introducing any functions. The
geometric method is easier if you think of it. Even in cases where elementary methods
can be made to work, however, calculus does the same job methodically.
Problem 2. In a coordinate plane, let A (0, 1) and B

= (3, 2), as shown in the
=
figure. What is the length of the shortest path from A to the x-axis to B? And where
should the path touch the x-axis, for this minimum to be attained?
In other words, for what choice of P (x, 0) is the sum of the distances AP
=
and PB as small as possible?
B
2
Solution. Let
f(x) =AP+ PB
= .J12 + x2 + ·./(3 - x)2 + 22
=.Ji + x2 + .Jx2 - 6x + 13.
Then
x x -3
f'(x) =
i +
J + x2 Jx2 - 6x + 13
xJx2 - 6x + 13 + (x - 3)-,/�
J 1 + x2 Jx2 - 6x + 13
•
Thereforef'(x) = 0 when
x2(x2 - 6x + 13) = (x2 - 6x + 9)(x2 + 1),

or
x4 - 6x3 + 13x2 = x4 - 6x3 + 9x2 + x2 - 6x + 9,
or
3x2 + 6x -9 = 0,
or
x2 + 2x - 3 = 0,
or
(x + 3)(x - I) = 0.
To examine second derivatives looks hard. Let us try to use reasoning instead.
I) When x decreases past 0, AP increases, and so does PB. The same is true when x
increases past 3. Therefore, in searching for a minimum, we can restrict the search to,
say, the interval [ -1, 4].
2) Suppose that we know that the function has a minimum, somewhere on the
interval (-1, 4]. Then the minimum must be an ILMin, at whichf'(x) = 0. There
is only one such point on our interval, namely, x = 1. Therefore the minimum
must be at x = I.
Obviously, to complete this discussion, we need the following theorem:
Theorem 2 (Existence ofm in ima). If/is continuous on [a, b], then/has a minimum
value on [a, b].
The proof is easy, granted that Theorem 1 is true. Since -fis continuous, it has
a maximum; and any maximum of -fis a minimum off
-M
:
-� b l �-- x
1 _
�
�
��-+--+'a'--- ·��
�
-- ��x
I I I
M -t- ---- :
I -f I
I
Here again, once the problem is solved, you may be able to think of a simpler
attack on it. But the methods of calculus work in any case.
Problem 3. Find the right circular cylinder of largest volume, inscribed in a sphere
of radius 1.
-1
To avoid a difficult drawing problem, we show not the three-dimensional figure

but merely a plane cross section of it. One way to introduce a function into this
problem is to express the volume of the cylinder as
(0 � x � 1).
This gives
V'(x) = (
2TT 2x ·Ji - x2 + x2 ·
Ji -
-
x
)
x2
Therefore
V'(x) = 0 -¢> 2x - 3x3 = 0 -¢> x(3x2 - 2) = 0 .
Since x must lie on [O, 1), we find that V'(x) = 0 only when x = 0 or x = Ji
Now V has a maximum, because V is continuous on [O, 1]. And this must be an
ILMax, because V(O) = 0 = V(l), and V(x) > 0 everywhere else on [O, 1]. There-
fore V'(x) = 0 at the maximum. Therefore the maximum occurs at x = J'i. Hence
the maximum volume is
There is another function that we might have used to solve the same problem. We
might have written
V(y) = TTr2h = TTX2 2y• = TT(l - y2)2y = 2TT · (y - y3).

This would give
V'(y) = 2TT(l - 3y2),
so that
V' (y) = 0 <=> 3y2 = 1 <=> y = Jt or y = -/i.
Here again only the positive number applies, because y must be on the interval [O, 1].
As before, we conclude that the maximum value occurs at y = J};
The second method is simpler. This sort of thing happens often. It is therefore a
good idea to have a quick look at all of the functions that it seems natural to try,
before doing any hard work with any one of them. If the first function that you try
looks simple, there is no point in examining others.
Our third problem shows a danger which should be remembered hereafter. We
might have supposed that the inscribed cylinder attains its maximum volume at the
stage where the inscribed rectangle (in the cross section) attains its maximum area.
But this is false: it is easy to show that the inscribed rectangle of maximum area is a
square; and the cross section of the maximal cylinder is a rectangle of base 2.Jj and
altitude 2.Jt. Therefore we should never assume without proof that two maximum
minimum problems are equivalent.
A further word of caution: In establishing that a certain x0 gives a maximum or
minimum, you may use the theorems of the preceding sections. Under certain con
ditions, you may avoid these theorems (and the calculations that they require) by the
sort of reasoning that we have used in the problems above. But in any case, you must
use either the theorems of the preceding sections or a reasoning process which justifies
your conclusions. To find a point x0 where a derivative vanishes and hence infer that
your problem is solved is a mistake. For one thing, x0 may give a minimum when you
were looking for a maximum, or vice versa. For another thing, x0 may give a point
of inflection.
PROBLEM SET 5.4
1. Find the area of the largest rectangle than can be inscribed in a semicircle of radius a.
2. Find the area of the largest rectangle that can be inscribed in an equilateral triangle
whose sides have length a.
3. Find the area of the triangle with the smallest area which contains a square with side a.
4. Find the perimeter of the triangle with the smallest perimeter which contains a square
with side a ..
· 5. A rectangular field has one side along a river and a fence along the other three sides.
If the total length of the fence is k, what is the maximum possible area of the field?
6. Given a rectangular field with one side along a river, as in Problem 5. If the area of
the field is A, what is the minimum possible length of the fence?
7. If a rectangular wooden beam is supported horizontally at its ends, then the maximum
weight that it can support at its midpoint is proportional (at least approximately) to its
width, and to the square of its thickness. That is, W = k · x · y2, where xis the width,
y is the thickness, and k is a constant depending on the wood (and on the units of
length and weight).
Suppose that such a beam is to be cut from a cylindrical log of radius a, in such a
way as to maximize W. What should be the width and the thickness?
8. An open pan is to be made out of a square metal sheet, by cutting out the square pieces
from the corners of the sheet and folding up the sides of the metal that is left. (The
square pieces are to be thrown away.) If the sheet has edges of length a, what is the
volume of the pan of largest volume that can be made in this way?
9. An open pan, of the sort described in the preceding problem, has a total surface area
of 128 sq. in. What is the largest possible volume?
10. Find the closed circular cylinder with volume 10 cu. in. and surface area as small as
possible.
11. Solve the same problem, given that the cylinder is open at one end.
12. Solve the same problem, given that the cylinder is open at both ends. (It sits on a flat
table and holds flour.)
13. A piece of sheet metal, n feet long and w feet wide, is to be bent so as to form a trough
n feet long, with open top, open ends, and triangular cross sections. What is the greatest
possible cross sectional area?
14. A trough is to be made with isosceles right triangles as endpiecesand congruentrectangles

as sides, as shown in the figure. If the total surface area is to be 100 sq. in., what is
the maximum volume?
15. In a rectangular parallelepiped, with a square base, the total length of the edges is k.
What is the largest possible volume?
16. A rectangle is to be inscribed in the region above the x-axis and below the graph of
y = 1 - x2• Find the area of the rectangle of maximum area.
17. Same problem, for y = 1 - x4•
18. Find the rectangle of maximum area contained in the region above the line y = t,
to the right of the line x = 1, and under the graph of y = 1/x.
19. A rectangle is inscribed in the region R = {(x, y) I ixl + [y\ � 1}, in such a way as to
maximize the area. Find the area of the rectangle.
Find the values of x at which the following functions take on their maximum values,
and j ustify your answers. You need not find the maximum values of the functions.
x
=
x
20. f(x) 21. g(x)
1
= --
+ x2 1 + x4
22. h(x) J"' Sin-1 t dt 23. =

finx V1 + t8 dt
</>(X)
=
-1 0
Problems 24 through 27. Investigate the preceding four functions for minimum values.
28. An isosceles triangle has base d and. altitude h. Find the area of the rectangle of largest
area that can be inscribed in it.
29. Given a triangle with angles of 30°, 60°, and 90°, there are three plausible ways of
inscribing in it a rectangle of maximum area; the rectangle may have a side lying along
any one of the three sides of the triangle. Show that all three of these "maximal"
rectangles are really maximal; that is, show that they all have the same area.
30. Show that there are some triangles for which the conclusion in Problem 29 does not hold.
31. Show, however, that the conclusion of Problem 29 holds for a class of triangles which
includes more than the 30°-60°-90° triangles.
32. Consider the curve which is the graph of the equation x2 + 4y2 = 4. Find the area
of the rectangle of largest area that can be inscribed in this curve.
33. A right circular cone has a base of diameter d, and altitude h. Find the volume of the
largest right circular cylinder that can be inscribed in it.
34. Find the area of the isosceles triangle of maximum area that can be inscribed in a circle
of radius r.
35. Find the volume of the right circular cylinder of maximum volume that can be inscribed
in a sphere of radius r.
36. Suppose that in Problem 34 the word "isosceles" is omitted. Is the solution of the
resulting problem the same as before?
37. Similarly, discuss the problem obtained by omitting the word "right" in Problem 35.
38. Find the length of the longest ladder than can be carried (in a horizontal position)
around the corner shown on the left below. The segment from P to Q shows a possible
position of the ladder.
39. In the right-hand figure above, the circle (of radius r) is inscribed in the right angle
LBAC. What is the minimum possible area of 6ADE?
**40. Suppose that in Problem 39 we do not require that LBAC be a right angle. Given
that LBAC has measure ix, find the minimum possible area of 6ADE, in terms of r
and ix. (This is much harder than Problem 39.)

5.5 THE USE OF FUNCTIONAL EQUATIONS AS SHORTCUTS
In the preceding section, we found that under some conditions we could locate maxi
mum and minimum values merely by finding a point where the derivative vanishes.
We shall now see that in some cases we can locate maximum and minimum values
without calculating the function. Consider first a simple problem, from Section 5.4.
Problem 1. A segment of length 1 has its endpoints on the sides of a right angle.
What position for the segment gives maximum area for the resulting triangle?
As in Section 5.4, we set up the axes as shown. Let x be the x-coordinate of the
lower endpoint of the segment; and for each x from 0 to 1, letf(x) be they-coordinate
of the other endpoint. Note that we are entitled to use functional notation: f(x)
really is determined when x is named. And for each x, we have
x2 + [j(x)]2 = l2,
because x2 + [ f(x)]2 is the square of the length of the segment. Therefore the function
f satisfies the equation
x2+f2=l (0 � x � 1). (1)
The area of the triangle is
A(x) = tx ·f(x). (2)
Now in (1), the left-hand member is a function, whose derivative is 2 · x+ 2 ·ff'.
But this function is known to be a constant, equal to 1 for every x from 0 to 1.
Therefore
x +ff' = 0 (0 < x < 1). (l')
Here, of course, we are assuming that f has a derivative, for 0 < x < 1, but this
must be true, because the graph off is a quadrant of a circle. Obviously
A'(x).= ! · x ·f'(x) +t ·f(x). (2')
The maximum of A(x) must be an ILMax; and so, at the maximum of A(x), we have
xf' +f = 0. (2")
We now know:
x
!' = - -
on (0, 1),
f
!' = _i at the maximum.

x
5.S The Use of Functional Equations as Shortcuts 233
Therefore, at the maximum, both of these equations hold, and
and x = f(x).
f x
That is, the maxim um is achieved when the triangle is isosceles.
This discussion has been long, because ideas needed to be explained; but once the
ideas are understood, the calculations are simple:
x 2 + f2 = 1, 2 · x + 2 ·ff'= 0, f' = - J;
A(x) = tx · f(x), A'(x)= tx · f'(x) + tf(x);
and hence
A'= 0 <::?- f'= - j

x
Therefore, at the maximum,
- -= _J
x
and x = f(x).
f x
In this case, of course, it was not much trouble to find a formula for f and use it.
But in many cases, equations like
x2+12= 1
are more convenient than formulas for the function f These are called functional
equations. Obviously every trigonometric identity is a functional equation. Usually,
however, we use the word identity when the function is known, and the termfunc
tional equation when the equation itself is being used as a working definition of the
function.
Consider another example, Problem 3 in Section 5.4.
Problem 3. Find the right circular cylinder of largest volume, inscribed in a

sphere of radius a.
y
a
As before, we show a vertical cross section of the figure. Let x be the radius of the
inscribed cylinder, and letf(x) be half the altitude. Then
x2 +f2= a2, 2 · x + 2 ·ff' = 0,
and
x
f' = (0 < x < a). (3)
f
234 The Variation of Continuous Functions s.s
Now the volume is

V(x) = 7TX2 2f(x),
•
so that
V'= 27T (x2.f' + 2xf).
•
At the maximum, V' = 0, and so
- 2xf= - 2!
f' = (at Max). (4)
x2 x
Therefore, at the maximum, both our formulas for f' must hold, and so
2
- �
=- f
f x
and
./2
f=-·x. (5)
2
For a = 1, this tells us that
X - '\/1;,_
-
3 '
as before.
Note, however, that in a way the most natural answer to a problem like this is a
shape, rather than a size. And the solution based on the functional equation ordinarily
gives the answer in the form of a shape, that is, in the form of a ratio between two
measurements. For example, in the preceding problem the constant a, which deter
mines the size of the whole configuration, disappeared immediately when we differ
entiated in the equation x2 + f2 = a2. Our final equation (5) means that at the
maximum,
2f(x) = /2 x,
that is, the altitude of the maximum cylinder is equal to /2 times the radius of its base.
y
-a
5.5 The Use of Functional Equations as Shortcuts 235
The answer is also a shape when the problem is to find the rectangle of maximum
area in a given circle:
x2 + f2 = a2, 2x + 2ff' = 0,
f' = - � (0 < x < a);

f
A(x) = (2x) 2f(x) = 4

· · x f,
·
A'(x) = 4(xf' + f),
A'(x) = 0 <=> f' = -f

x
Therefore at the maximum,
- �= f '= - 1 and x = f,
f x
because x and f are both positive. This is a qualitative answer, as it should be: it
says that the maximum rectangle is a square. The constant a has disappeared,
because the shape of the maximum rectangle is the same for all circles.
In the following problem set, you will find more cases in which maxima and
minima can most conveniently be found by using functional equations. Meanwhile
let us look carefully at what happens when we take the derivative on each side of a
functional equation. The ideas here are illustrated by a simple case. When we write
x2 + 12 = a2 (6)
=> 2. x + 2 . ff' = 0, (7)
we are claiming that every differentiable function which satisfies Eq. (6) also satisfies
Eq. (7). It often happens that there is more than one such function/ For example,
consider
Here
-x -x
f{(x)= ,
..Ja2 - x2 fi( x)
and
x -x -x
f�(x) = =
--;===
..J a2 - x2 -..Ja2 - x2 f2(x)
Therefore
fi(x)f{(x) = -x, and
Therefore
{2. x + 2. fd{ = 0,
(8)
2 . x + 2. ! !�
2 = 0.
That is, both/1 and/2 satisfy (7). A figure makes it obvious what is going on here.
At each of the labeled points, we have

-1 -x
fi(x)= =
f;(x)/x f;(x) '
because the tangent is perpendicular to the radius.
The same sort of thing goes on in more complicated cases. The graph of
y= x3 - x (9)
looks like the left-hand figure below.
Therefore the graph of
x= ya-y (10)
looks like the right-hand drawing below.
We have interchanged x and y in Eq. (9), and reflected the graph across the line
5.5 The Use of Functional Equations as Shortcuts 237
y = x. This gives the curve C which is the graph of (10). C is not a function-graph.
But C is the union of the graphs of three functions fi,f2,f3, as indicated in the figure.
And each of the functions ft> h, and /3 satisfies the functional equation
x
=/3 - f
Therefore each of these functions satisfies the differential equation
1
= 3 ·Pf' - !'.
This is what we are claiming when we differentiate the functional equation, and write
x =/3 - f => 1 = 3 . !21' - f'.
PROBLEM SET 5.5
In Problems 1 through 10 below, the notation 5.4.n refers to Problem n of Problem Set
5.4. In each of these cases, the indicated ratio is to be found by the method based on func
tional equations.
1. In 5.4.1, find altitude/base, at the maximum.
2. In 5.4.2, same.
3. In 5.4.5, same, using the side parallel to the river as base.
4. In 5.4.7, findy/x, at the maximum.
5. In 5.4.14, let l be the length of the rectangular side and let w be the width. Find w/l
at the maximum.
6. In 5.4.15, let h be the altitude and let e be the length of each edge of the base. Find
h/e, at the maximum.
7. In 5.4.16, find altitude/base, at the maximum.

8. In 5.4.28, same.
9. In 5.4.33, let a be the altitude of the cylinder, and let r be the radius of the base. Find
a/r, at the maximum.
10. In 5.4.34, let h be the altitude, and let a be half the length of the base. Find hfa, at the
maximum.
11. In 5.4.35, let h be the altitude and let a be the radius of the base. Find hfa, at the
maximum.
12. We know of a function f, with domain [ -1, 1], which is a solution of the functional
equation sin f(x) = x. (Our "known function," of course, is f(x) = Sin-1 x.) What
other continuous solutions of the equation have the entire interval [ -1, 1] as domain?
Draw a figure.
13. Write a differential equation which is satisfied by all solutions of the functional equation
x4 + [/ (x)]4 = 1.
14. a) Let n = 101010• Sketch the graph of xn + yn l. [Hint: A commonly used drawing
=
instrument will give you an excellent sketch.]

b) Let n = 101010 + 1. Sketch the graph of xn + yn = 1. [Same hint.]
15. Find the functions/which satisfy the differential equation
x +ff'= 0.
(You need not show that the solutions that you describe are the only ones.)
16. Given that f and/' are continuous, let
F(x) = f'f (t)/1(t) dt.
Calculate F(x), in terms off
*17. Now show that your list of solutions, in Problem 15, is complete.
18. Let f be the function whose graph is the union of (a) the lower left-hand quadrant of
the circle with center at (0, 1) and radius 1 and (b) the upper right-hand quadrant of the
circle with center at (0, -1) and radius I. Show that f is a solution of the differential
equation
[f'(x)]2 = [x + f(x)J'(x)]2,
except, of course, at the endpoints x = ± 1, where the tangent lines are vertical and the
function- has no derivative. As a start, observe that at x 0, the tangent to the graph
=
is horizontal and the equation is satisfied: 02 = [O + 0 0]2• ·
*19. Consider the family of quadratic functions represented by the formula
f (x) = (x - a)2• (1)

Differentiating, we get
j'(x) = 2(x - a),
and squaring, we get [j'(x)]2 = 4(x - a)2, and
[j'(x)]2 = 4 · /(x) . (2)
Evidently (1) = (2). But the converse is false.
a) Show that one of the solutions of (2) is a linear function f

b) Show that (2) has some solutions which are neither quadratic nor linear; that is,
the differential equation has solutions whose total graphs are neither lines nor
parabolas.
5.6 THE COMPLETENESS OF R AND THE EXISTENCE OF MAXIMA
In Section 5.4 and later, we have used the fact that, if f is continuous on [a, b],
then f has a maximum value o:n [a, b]. In Section 5.4 this theorem was used as a
shortcut in finding maximum values, but this is only one of the uses of the theorem.
In fact, the theorem is part of the foundation of the calculus, as we shall see.
In proving it, we shall need to use, for the first time, the fact that the number line
has no holes in it. As a guide in giving an exact description of this property of the
number system, let us consider what happens when you remove a point from the
number line, thus getting a system which really does have a hole in it.
Let A be the set of all negative numbers, and let B be the set of all positive
5.6 The Completeness of R and the Existence of Maxima 239
numbers. We mean strictly positive and strictly negative, so that 0 belongs neither
to A nor to B. Then
1) B has no least element.
�1--"x'---x�-- ·1=----
� ---= 2��-=---
3 ----
- R+
2
The reason is that if xis a positive number, then so is x2

/ , and x/2 < x. Therefore
no positive number x is less than all other positive numbers, and so (1) holds.
Similarly,
2) A has no greatest element.
R- �-_
---.,;.
3 .. �� -�,...-
2 - � �l �X--:':-X-+-1 ---·
-
2
For if x < 0, then x/2 < 0, and x2

/ > x.
Now let
K = A u B = {xI x ¥: O}.
Then obviously:
3) K is the union of two nonempty sets A and B, such that (a) every number in A
is less than every number in B, but (b) A has no greatest element, and (c) B has no
least element .
Evidently this situation could not have arisen if we had not excluded 0: if we
put 0 in A, then 0 would be the greatest element of A; and if we put 0 in B, then 0
would be the least element of B. Thus the situation described in (3) can arise only in a
number system with a hole in it, and so the following statement conveys the idea that
there are no holes in R:
The Dedekind Cut Postulate (DCP). Suppose R is expressed as the union of two
nonempty sets A and B, such that every element of A is less than every element of B.
Then either A has a greatest element or B has a least element.
A B
I
Xo
In the figure, x0 must belong either to A or to B. Therefore x0 is either the

greatest element of A or the least element of B.
We have stated DCP as our first description of the completeness of R, because
it is the best known description, and in some ways the most natural. But for some
purposes, the following idea is easier to use. Given a sequence
[a1, b1], [a2, b2L ...

of closed intervals. If every interval in the sequence contains the next, then we say
that the sequence is nested. Algebraically, this means that
for every i.
For example, if
[a·i' b·]
i
-
-
( - ! !) •
1
' .
1
for every i,
then the sequence is nested. This sequence "closes down on O." That is, 0 lies in each
of the intervals in the sequence, and 0 is the only number that lies in all of them.
A more important example is as follows. Given a circle of radius 1, let Pn be the
perimeter of an inscribed regular (n + 2)-gon, and let qn be the perimeter of a circum
scribed regular (n + 2)-gon. Evidently
P1 < P2 < p3 < · · · ,
and
for each i.
Thus we have a nested sequence
of closed intervals. And this sequence "closes down on 27T." That is, 27T lies in all of
the intervals in the sequence, and no other number lies in all of them.
The following postulate says that every nested sequence of intervals closes down
on at least one point.
The Nested Interval Postulate (NIP). For every nested sequence of closed intervals
there is a number x which lies in every interval in the sequence.
This conveys the idea that the number system is complete. Suppose, for example,
that 27T were missing, so that the number system had a hole in it where 2rr ought to be.
Then no number at all would lie on all of the intervals [Pi, q1], [p2, q2], . • • that we
have just discussed. Similarly, if ./2 were missing, then there would be a nested
sequence of closed intervals closing down on no number whatever. (We could use
[,./2 - lfi, ,J2 + l/i] as the ith interval in the sequence.)
Using the nested interval postulate (NIP), we shall prove the following theorem:
Theorem 1. If/is continuous on [a, b], then/has an upper bound on [a, b].
That is, there is a number M such that/(x) � M for each x of [a, b].
Lemma. If/is unbounded above on an interval [c, d], then/is unbounded above on
at least one of the halves of [c, d].
By the halves of [c, d] we mean the intervals [c, (c + d)/2] and [(c + d)/2, d].
The proof of the lemma is immediate: if/has an upper bound M1 on [c, (c + d)/2]
and has an upper bound M2 on [(c + d)/2, d], then/has an upper bound on [c, d].
\\{e merely use the larger of the bounds M1 and M2•
y y
I
I
I
I
I
--1-------
1
I
I
I
I
�+-�
c��c+�d��-d,,...__
._ x
2
We proceed to prove the theorem. For short, we say that an interval is good
if f is bounded above on the interval; and we say that an interval is bad if it is not
good. Thus we need to prove that [a, b] is good. We start by supposing that [a, b]
is bad, and we shall show this assumption leads to a contradiction.
If [a, b] is bad, then it follows that at least one of the halves of [a, b] must be bad.
Let [a1, b1] be a bad half of [a, b]. For the same reason, [a1, b1] must have a bad half.
Let [a2, b2] be a bad half of [a1, b1]. Continuing this process to infinity, we get a
sequence
[a1, b1], [a2, b2], ...

of closed intervals, all of which are bad, and each of which is a half of the preceding
one. Therefore
and so
1
b; - a; = -: (b - a).
2'
By NIP, there is an x such that a; � x � b; for each i.

But f is continuous at x. Thus, for every E > 0, f has an EO-box at the point
(x,J(x)).
y
M = f(x) + • --------1--------,
I I
I
I
f(x)
Thus
lx-xl<o ::?- f(x)-E<f(x)<f(x)+E,
and so f (x) + E is an upper bound for f on the interval (x -o, x + o). But since
lim (b; - ai) = 0,

i-+ 00
we have
bi - ai < a
for some i. For such an i, the closed interval [ai, bi] lies inside the open interval
(x - a, x + a). That is,
as shown in the figure.
x-o a; x
(This is easy to see geometrically, because [a;, b;] contains the midpoint x of the open
interval, and is less than half as long.)
But this situation is impossible, because f is bounded above on (x - a, x + a)
and is not bounded above on the smaller interval [ai, b;]. This contradiction completes
the proof of the theorem.
One of the ideas that we have just used is going to be useful later. We therefore
record it as a theorem:
Theorem 2. Suppose that
for each i, and

lim (bi - a;) = 0.
i-+ 00
Then every interval (x - a, x + a) contains some interval [a;, b;].
This was proved in the preceding discussion. If (x - a, x + a) contains [a;, b;],

then of course it follows that (x - a,.\' + a) contains all of the later intervals
Given that a functionfis bounded, it does not follow thatfhas a maximum or a

minimum. Consider, for example,
f = Tan-1.
7r
---------------
2
When xis far to the right, Tan-1 xis close to TT/2, but Tan-1 xis never actually equal
to TT/2 for any x. Similarly, when xis far to the left, Tan-1 xis close to -TT/2, but
-TT/2 is not one of the values of the function. On the other hand, it is easy to see that
the numbers TT/2 and -TT/2 are related to the function Tan-1 in a special way: TT/2
is an upper bound of the function; and of all upper bounds of the function, TT/2 is the
smallest. We express this by writing
!!. =sup Tan-1.

2
Here sup is pronounced supremum. To be exact:
Definition. If k is an upper bound of a function f, and k is smaller than every other

upper bound of/, then k is called the supremum of/, and we write
k =sup/
More generally, we define the supremum for any set of numbers:
Definition. Let B be a set of numbers. If x � k, for every xin B, then k is an upper

bound of B. If k is smaller than every other upper bound of B, then k is called the
supremum of B, and we write
k =sup B.
Consider, for example, the case where B is an open interval (a, b). Every number
k ?; b is an upper bound of B. Thus the upper bounds of B form an interval [b, w ).
B
�
0 a b
Here b is an upper bound of B, and b is smaller than all other upper bounds of B.
Therefore
b =sup B.
Consider now
B-
_ {1 '3� '4'····
� n -
n
1
,
\
... f'
2
Here the upper bounds of B are the points of the interval [ 1, w), and sup B = 1.
4
5
��---'-��-----++<>---'-�- x
0 1 2 3 5
x =2 346
· In each of these cases, starting with a nonempty set B which is bounded above,
we have found that the upper bounds form an interval of the type [k, w), and k =
sup B. The following postulate says that this is what always happens:
The Least Upper Bound Postulate (LUBP). Let B be a nonempty set of numbers.
If B has an upper bound, then B has a supremum.
Using the least upper bound postulate, we shall show that no continuous function
can behave like Tan-1 if its domain is a closed interval:
Theorem 3 (Existence of maxima). If/is continuous on [a, b], then/has a maximum

value on [a, b].
y
M =f(x),f(x) �M for every x
Proof We know by Theorem 1 that/ is bounded. Let
k = supf
Thenf(x) � k for every x on [a, b]. We need to show thatf(x) = k for some x.
Suppose not, and let
1
(a � � b).
g(x)
= x
k - f(x)
Then g is continuous. But g is unbounded. For suppose that
g(x)� M for a� x � b.
Then
1
1 :$ M, � k - f(x),
k - f(x) - M
and
f(x) :$
- k - _!_ for a � x � b.
M
This is impossible, because k is the least of the upper bounds off
Thus, if/has no maximum, there is a continuous function g which is unbounded
on [a, b]. This contradicts Theorem 1, and so completes the proof of Theorem 3.
We have already observed, in Section 5.4, that the existence of maxima implies
the existence of minima. Therefore
Theorem 4 (Existence of minima). If/is continuous on [a, b], then/has a minimum

value on [a, b].
(This was Theorem 2 of Section 5.4.)
PROBLEM SET 5.6
1. Let B be the set of all rational numbers p/q for which p2/q2 < 2. What is sup B?
2. Consider a circle of radius 1. For each polygon P inscribed in the circle, Jet k(P) be the
perimeter of P. Let B be the set of all numbers k(P). What is sup B?
3. Consider the graph off (x) = sin x, 0 � x � 1T. Suppose that we cut up the interval
[0, 7r] into little intervals, in any way, using subdivision points 0 x1 < x2 < < = · · ·
xi < xi+l < · < x,. = 1T. Over each little interval [xi, xi+i l we set up the tallest
· ·
possible inscribed rectangle with [xi, xi+i l as base. Let s be the sum of the areas of the
rectangles. Let B be the set of all numbers s which are obtainable in this way. What is
sup B? (A numerical answer is called for here.)
4. Let B be any set of numbers. If b EB, and b is larger than every other element of B,
then b is called the greatest element of B, and we write b = Max B. Question: If B
has an upper bound, does it follow that B has a Max?
5. Suppose that we had defined bounds and suprema in the following way:
"Let B be a set of numbers, and let k be a number. If x < k, for every x in B,
then k is a strict upper bound of B. If k is a strict upper bound of B, and is smaller
than every other strict upper bound of B, then k = sup B."
a) What is the difference between this "definition" and the usual definition of upper
bounds and suprema?
Under the new "definition" of "supremum," which if any of the following statements
are true?
b) Every finite set has a "supremum."

c) No finite set has a "supremum."
d) Every open interval has a "supremum."
e) No open interval has a "supremum."
f ) Every closed interval has a "supremum."
g) No closed interval has a "supremum."
6. If B is a set of numbers, then -B denotes the set obtained when we replace every
element x of B by its negative -x. That is,
-B = { -x Ix EB}.
For example, if B = [l, 2], then -B = [ -2, -1]; if B = [ -1, oo ) , then -B =
(- oo, 1 ], and so on. Prove the following:
Theorem. If (a) k is an upper bound of B, then (b) -k is a lower bound of -

B And
.
conversely, (b) implies (a).
(This is easy; don't try to make it hard.)
7. If k is a lower bound of the set B, and k is greater than every other lower bound of B,
then k is called the infimum of B, and we write k = inf B. Show that if a set B is bounded
below, then B has an infimum.
8. Let B be a set which is bounded below, and let K be the set of all lower bounds of B.
Describe Kin the interval notation.
*9.· Let [av bi], [a2, b2], .. .be a nested sequence, and letA = {ava2, • • • }, B = {b1,b2, • • . }.
Show that (a) every number b; is an upper bound of A. Let x = sup A. Then show that
(b) ai � x � bi for every i.
This result means that the least upper bound postulate (LUBP) implies the nested
interval postulate (NIP).
*10. Let Kbe a (nonempty) set of numbers, bounded above. Let A be the set of all numbers
a which are not upper bounds of K. That is, a EA if a < k for some k in K.
Show that A cannot contain a greatest element.
* 11. Show that the Dedekind cut postulate (DCP) implies the least upper bound postulate
(LUB P) .
The results of Problems 9 and 11 mean that
DCP => LUBP => NIP.
Thus our only really new assumption, in this section, is DCP.
5.7 THE MEAN-VALUE THEOREM AND THE NO-JUMP THEOREM
The mean-value theorem was stated in Chapter 3, and we have been using it ever
since. We are now finally in a position to prove it. We need one preliminary result.
Rolle's Theorem. Iffis continuous on the closed i'nterval [a, b] and differentiable on
the open interval (a, b), and f(a) f(b) 0, then j'(x)
= = 0 for some x between
=
a and b.
y
Proof There are three cases to consider:

1) Suppose thatf(x) = 0 for every x on [a, b]. Then any number x between a and b
gives/'(x) = 0.
2) Suppose that/(x) > 0 for some x on [a, b]. Now fhas a maximum at some x,
and x is not a or b. Therefore fhas an ILMax at x. By Theorem 3 of Section 5.2 it
follows that f'(x) 0.
=
3) Iff(x) < 0 for some x, then the minimum of/is an ILMin. By Theorem 4 of
Section 5.2 we know that at an ILMin the derivative vanishes.
The proof of MVT is now easy.
f(b l- -�---------
g
a b
5.7 The Mean-Value Theorem and the No-Jump Theorem 247
Given that f is continuous on [a, b] and differentiable on (a, b), let g be the linear
function which agrees with fat a and at b. Thus
g(a) = f(a), g(b) = f(b).
We could write a formula for g, in the form g(x) mx + k, if we needed to, but
=
we don't need to. Since the derivative of a linear function is simply the slope of the
line which is its graph, we know that
g '(x) f(b) - f(a)

,
=
b - a
for every x. For each x of [a, b], let
c/>(x) = f(x) - g(x).

Then cf> is continuous on [a, b] (because fand g are), and cf> is differentiable on (a, b),
with
f(b) - f(a)
c/>'(x) f'(x) - g'(x) f'(x) -
= =
b - a
Since cf>(a)=c/>(b) =
0, we can apply Rolle's theorem. Therefore cf>'(x) =
0 for
some x. Thus
f(b) - f(a) o,
f '( x)
_ =
b - a
and
f(b) - f(a)
f' (x)
=
b a -
for some x, which was to be proved.

The no-jump theorem is harder. To prove it, we need to go back to first principles,
and we need some preliminary results.
Lemma 1. Let f be a continuous function, on an open interval containing x0• If

f(x0) > 0, then there is a o > 0 such that
x0 - O < x < x0 + c'J => f (x) > 0.
Proof Since fis continuous, we know that
lim f(x) f(xo).

=
In the definition of a limit, we take E = f(x0) > 0. There is a o > 0 such that
x0 - o < x < X0 + o => f(x0) - E <f(x) <f(x0) + E => 0 <f(x),
because f(x0) - E = 0. Therefore the o that we have is the o that we wanted.
Lemma 2. Letfbe a continuous function, on an interval containing x0• Iff(x0) < 0,

then there is a o > 0 such that
x0 - o < x < x0 + o => f(x) < 0.
Proof? (The proof of Lemma 1 can be adapted, to give a proof of Lemma 2. But
it is quicker to derive Lemma 2 from the statement of Lemma 1.)
A functionfchanges sign, on an interval/, iff(x) > 0 for some x in I and/(x') <
0 for some x' in /.
Lemma 3. If fis continuous, on an interval containing x0, and f(x0) ¥- 0, then there
is a 0 > 0 SUCh that j does not change sign on the interval (x0 - O, Xo + 0).
Proof. For f(x0) > 0, this follows from Lemma 1. For f(x0) < 0, it follows from
Lemma 2.
We are now ready to prove the following convenient special case of the no-jump
theorem.
Theorem 1. If/is continuous on [a, b], and/ changes sign on [a, b], then
f(x0) =
0
for some x0 in [a, b].
y
The proof is based on Lemma 3 and the nested interval postulate (NIP). We
suppose that/(x) ¥- 0 for every x in [a, b]. We shall show that this assumption leads
to a contradiction.
Given that f changes sign on [a, b] and that f(x) is never =
0, it follows that f
changes sign on one of the halves of [a, b]. We recall, from Section 5.6, that the
halves of [a, b] are [a, (a + b)/2] and [(a + b)/2, b]. Let [a1, b1] be half of [a, b],
such that f changes sign on [a1, b1]. Similarly, let [a2, b2] be half of [av bi], such
that f changes sign on [a2, b2]. Proceeding to infinity in this way, we get a nested
sequence
[a1, b1], [a2, b2], ...
of closed intervals, such that f changes sign on each of them. Evidently
b; - a; = 2 -i( b - a),
5.7 The Mean-Value Theorem and the No-Jump Theorem 249
and so
lim (bi - a;) = 0,
i-t> 00
as in the proof of Theorem 1 of Section 5.6. By NIP, there is an x0 which lies on all
of the intervals in the nested sequence. That is,
for every i.
By Lemma 3 there is a o > 0 such that/ does not change sign on the interval (x0 - o,
x0 + o). By Theorem 2 of Section 5.6, there is an i for which [a;, b;] lies in (x0 - 0,
Xo + 0), as indicated in the figure.
This is impossible, because f changes sign on [a;, b;], but does not change sign on
(x0 - o, x0 + o). This contradiction completes the proof of Theorem 1.
It is now easy to prove the no-jump theorem.
Theorem 2 (The no-jump theorem). If f is continuous on [x1, x2], then f takes on

every value betweenf(x1) andf(x2).
Proof Suppose first that
j(x1) < k < f(x2),

and let
g(x) = f(x) - k.
Then g changes sign on [x1, x2]. Therefore g(x0) 0 for some x0 on [x1, x2]. This
=
givesf(x0) - k = 0, andf(x0) = k.
Iff(x2) < k < f(x1), then the same function g still changes sign, and so the proof
is exactly the same.
This completes our reexamination of the foundations of calculus. It now appears

that the idea of a continuous function is adequately described by the EO-definition
of a limit and that the completeness of the number system R, in the sense of "no
holes," is adequately described by the Dedekind cut postulate (which implies the least
upper bound postulate and the nested interval postulate).
The theorems in this section and the preceding one are not news; it was obvious
at the outset that these theorems ought to be true. But the fact that these theorems
can be proved, on the basis of a single simple assumption DCP, is significant. It
means that mathematics hangs together in a special way.
Nobody expects that a doctor will write down a definition of the word man
and then write a few assumptions about men, in such a way that all medical science
can be derived by logical reasoning from the definition and from the assumptions.
Medicine is an empirical science: it depends on observations of fact, not just at the
outset but continually. Mathematics is different.
Moreover, in your study of mathematics you have already passed the point where
the truth can be relied upon to be obvious and where obvious things can be relied
on to be true. From now on, logic is going to be an important part of your mathe
matical equipment. This is partly due to recent developments. As late as 1800,
calculus was illogical, and very few people cared. In the last century, however,
mathematical ideas which require careful logical analysis have become more
important, in pure research and also in applications.
5.8 THE DERIVATIVE OF ONE FUNCTION WITH RESPECT TO ANOTHER
Let fandg be differentiable functions. Take a point x0, and form the differences
Af= f(x0 + Ax) - f(x0),

A g =g(x0 + Ax) - g( x0) .
If!if/Ag approaches a limit, as Ax---->- 0, then this limit is called the derivative off
with respect to g, and is denoted by dfd
/ g. That is,
Jim
!if= df
6.x-+O A g dg,
by definition. In fact, the limit always exists, wheneverg'(x0) "16- 0.
Theorem 1.
df f '
-=-,
dg g'
whereverg'(x) "16- 0.
Proof
lim
!if =
Jim
!if/fix = f'(xo)
.
6.x-+O fig 6.x-+O fig/fix g'(x0)
For the case in which g(x) = x for every x, the derivative off with respect tog
reduces to an ordinary derivative:
Theorem 2. Ifg(x) = x for every x, then
df df
= f'(x).
=
dg dx
Obviously,
df !if
= Jim f'(x0),
=
dg 6.x-+O fix
for each x0•
5.8 The Derivative of One Function with Respect to Another 251
Some examples are as follows:
d sin x cos x
-cot x, (wherever cos x ":/= 0)
=
---
=
d cos x
---
-sin x
dsin x
--- = cos x,
dx
de"' e"'
(wherever x ":/= 0)
dx2 2x
We often write
d df
-f(x) for
dx dx
Thus every derivative can be written in the form
d
f'(x) = -f(x)
dx
,
' df.
f
=
dx
The notation df/dx for derivatives is widely used, especially in physics, and it is
natural to use it when you are continually dealing with the derivative df/dg of one
function with respect to another. It has a disadvantage, however: there is no con
venient way to write the value of the derivative at a particular point x0• Sometimes
we denote this by
°fx 'X=Wo'
but the notation f' (x0) is more convenient.
We now want to prove a sort of cancellation law
df. dg =
df
dg dh dh
We can derive this from the equation
l:lf. !:l g - l:lf

!:lg l:lh - l:lh '
ta�ing the limit as l:lx--+ 0. Thus we need
l:lj df !:lg dg
---- _ _,,_ _
!:lg dg ' l:lh dh '

as l:lx --+ 0. This requires
g'(x0) ":/= 0,
as in Theorem 1. Hence the conditions in the following theorem:

Theorem 3. If f, g, and h are differentiable, then
df dg df
dg. dh dh'
=
'
wherever g ¥:- 0 and h' ¥:- 0.
Theorem 4.
dg -
- 1_
df df/dg'
_
wherever df/dg ¥:- 0.

(The limit of the quotient is the quotient of the limits.)
We shall now find a short-cut for calculating derivatives of the type df/dg.
Consider
d sin2 x 2 sin x cos x
2 SinX
. (cos x ¥:- 0).
d sin x cos x
--- = =
This has the form

du2
2u.
du
- =
This is like
df
f'(x) 2x.
=
dx
=
That is, to find du2/du (where u is a function), we treat u as if it were a dummy variable
x and differentiate in one step. This is an example of the following situation.
Definition.Let f and g be functions. If there is a function </> such that f </>(g),

then we say that f is a function of g.
=
For example, sin2 x is a function of sin x, with </>(u) u2• And cos2 x - 2 cos x
is a function of cos x, with </>(u) u2 - 2u. The easiest way to calculate df/du, in
=
each of these cases, is to write
d sin2 x du2
2u 2 sm x'
•
d sin x du
--- = - = =
d(cos2 x - 2 cos x) d(u2 - 2u)

2u 2
d cos x du
= = _
= 2 cos x - 2.
This procedure is justified by the following theorem.
Theorem 5. Let/be a function of g, = </>(g), where all the functions are differentiable.
Then
</> ( )
wherever g' ¥:- 0.
% =
' g,
5.8 The Derivative of One Function with Respect to Another 253
Proof
df f' gg
= = </>'( ) ' = f (g).
dg g' g'
Using this theorem, we can write immediately
dtan2x
=
2 tanx,
dtanx
instead of using Theorem 1 and writing
dtan2x 2 tanx sec2x

= 2 tanx·'
dtanx sec2x
there is no point in writing sec2x = D tanx in both numerator and denominator,

since we are about to cancel it out in any case.
PROBLEM SET 5.8
Calculate dffdg, given:

2
1. /(x) =e"' , g(x) = x2
2
2. /(x) = e'" ,g(x) =2x
3. f(x) =e"', g(x) =Tan x 4. f(x) =exsinz,g(x) =x

5. f(x) =exsinx,g(x) = x3 __
3
6. f(x) =e"' ,g(x) = x2
7. /(x) =e'"3, g(x) =x4 8. f(x) = sin x,g(x) = cos x
9. f(x) = x3,g(x) =Tan x

Problems 10 through 14. In Problems 1 through 5, first calculate the function
</> such
that f = </>( g). (Answer in the form </>(u) =.) Then calculate ¢' (u) =
· · · · . Finally,
· ·
calculate <f>'(g),and compare it with your previous formula for df/dg. (Or, if you worked
the problems this way in the first place, work them by the other method, and check.)
Calculate as for Problems 1 through 9.
15. f(t) =sin t,g(t) =et 16. f(t) =cost, g(t) =Tan t
17. f(t) =t6,g(t) =t3 18. /(t) = t6,g(t) =Tan t

19. f(x) = ln x, g(x) =e"'
Problems 20 through 24. Solve the preceding five problems by another method.
25. Given/2 + t 2 + 1 = 0, find df/dt.
26. Given /3 + t3 = 1, find df/dt. Then calculate f = f(t), find /' (t), and compare the
result with dffdt.
27. Same, for f4 + t4 = 1. (Here there are two functions f = f(t) to be considered.)
28. Now try to check your answer to Problem 25 in the same way that you checked your
answers to Problems 26 and 27. (It often happens that a formal process gives "answers"
in cases where there never was a question.)
The Technique
6 of Integration
6.1 INTRODUCTION
In Section 3.7 we found a way to solve certain types of area problem. To :find the
area under the graph of a continuous function/, from a to b, we introduce the area
function
y
A= f f(t) dt
rYf
I
a x
We know that
F'(x) = D f'f(t) dt = f(x) for every x.
To calculate the area function, we :find another function G such that

G' =
f
We then know that
G' = F '.
If it happens that G(a) = 0, then we have G(x) = F(x) for every x. If not, we let
H(x) = G(x) - G(a).

Then
H'(x) = G'(x) = F'(x) and H(a) = 0.
Therefore
H(x) = F(x) for every x,
and so
ff(t) dt = F(b) = H(b) = G(b) - G(a).
254
6.2 Independent Variables and Indefinite Integrals 255
To sum up:
G' = f => ff(t) dt = G(b) - G(a).
The notations t and G were introduced for the sake ofthe derivation. Once we have
the answer, it is natural to use x and F, and write:
F' = f => ff(x) dx = F(b) - F(a).

More formally:
Theorem 1 (The fundamental theorem of integral calculus). Iff is continuous on

[a, b], and F' = f, then
ff(x) dx = F(b) - F(a).
To apply the theorem, ofcourse, we need to find F when/is given. This process
is called antidijferentiation. We shall see later that the method of antidifferentiation
enables us to solve not only the sort ofarea problems that we have used it on so far,
but also a variety of problems which, offhand, don't look like area problems at all.
But these applications should be postponed. The point is that, to apply the method,
we need to know how to calculate a function F whose derivative is a given function/;
up to now we have been finding such functions F only by hit-or-miss procedures, in
simple cases; and it would not be good to reduce various problems to problems in
antidifferentiation, when we are unable to solve the antidifferentiation problems.
We should therefore first learn better methods for calculating functions when their
derivatives are given.
6.2 INDEPENDENT VARIABLES AND INDEFINITE INTEGRALS
The usual way of defining a function is to write an expression which gives the value
ofthe function for every number in the domain. For example, we may define functions
f and g by writing
f(x) = x2 (-oo < x < oo), g(x) = ..}� (x � 0).

In these formulas, the letter "x" is called the independent variable. It is simply a
dummy letter, marking the places where numbers are to be inserted. Logically
speaking, it makes no difference what letter we use as a dummy. For example, we
could have defined exactly the same functions by writing
f (t) = t2 (-oo < t < oo), g(t) = Ii (t � 0).

Ifwe have decided to use, say, x as the dummy, then we say that/is a function of x.
Thus, when we write g(t) = cos2 t, we are describing g as a function of t ; h(rJ..) =
a3 - 1 is a function of rJ..; and so on.

We return now to the problem ofantidifferentiation. We found long ago, from
the uniqueness theorem, that iftwo functions have the same derivative, on an interval,
256 The Technique of Integration 6.2
then they differ by a constant. Thus, if

j'(x) =x2 (-oo < x < oo),
then /must be a function of the form
x3
f(x) =3 + C,
where C is a constant. The converse is ,trivial: for every C, D(x3/3 + C) =x2•

Therefore
{F I F' =x2} = {�3 c}.

+
The set of all functions F for which F' =/is commonly denoted by
ff(x) dx.
This is called the indefinite integral off Thus
J x4 dx = {F J F'(x) =x4} -;= {h5 + C},
Jcos x dx ={F I F'(x) =cos x} = {sin x + C},
and so on. Any other dummy letter would have done as well:
Jt4 dt ={F I F'(t) =t4},

and
J cos t dt = {F I F'(t) =cos t}.

In each case, the braces on the right indicate that we are talking about the set of all
functions of the form given inside. The symbol dx (or dt) merely reminds us that
x (or t) is the dummy letter used in describing the function. In the examples above,
the reminder may seem unnecessary. Similarly, when we write
f(3x3 + 2x4) dx = {3:4 2;5 c},

+ +
we might have gotten along without the "dx," because the only constants involved
are the numerical constants 2 and 3. On the other hand, if we write
Jc!Xx3y + (3x2y2) dx,
the "dx" is needed; it tells us that a, (3, and y are to be regarded as constants, and that
the function which we are dealing with is
f(x) = ax3y + f3x2y2.
When the problem is understood in this sense, it is plain ·that the answer is
f(o:x3y + {Jx2/) dx= {o:x4y + {Jxa

-y2 + C} .
-4- 3- (i)
This should be compared with
(ii)
f(o:x3y + /Jx2/)do:= r2�3y + f3x2/o: + c) , (iii)
f(o:x3y + f3x2y2)d/3= {o:x3y/J + /32�2�2 + c). (iv)
In (ii), o:, x
{3, and are constants, and the function is
g(y) o:xay + fJx2y2.
=
In (iii), x, y, and fJ are constants, and the function is
h(o:)= o:x3y + {Jx2y2.

Similarly for (iv).
The process of calculating indefinite integrals is called indefinite integration, or
briefly, integration. Given a differentiation formula, we can get a corresponding
integration formula merely by writing the given formula "backwards," with minor
adjustments in some cases to take care of constants. In each case below, the formula
on the right follows from the formula or formu1as on the left.
Dxn= n xn- (n ¥- 0), l
Dxn+l= ( + l)xn ( n n ¥- -
1) ,
Dcx:11) = xn (n ¥- -1) => f xndx= { n � 1 xn+1 + .c} (n ¥- -1),

D-y;-x= 1;-x ' 2-y
-
D(2Jx) = /x 1
=> J)xdx= {2Jx + c},
D x= x sin cos => f x dx=

cos {sin x + C},
D x= cos -sin x,
D(-cos x)= x sin => f xdx=

sin {-cos x + C},
D In x = .! (x
x
> 0) => J � dx= {In x + C} (x > 0),
De"'=e" => fe"' dx= {e"' + C}.
We know many more differentiation formulas than this, and so we could have
written many more integration formulas. But we postpone the complete list until we
can write it in a better form, which we shall now explain.
Given a function/, if u is another function, then f (u ) is a composite function.
By the chain rule,
Df(u)=j'(u)u'.
It follows that
ff'(u)u'(x) dx =
{f(u(x)) + C}.
For example, if
f(u)= sin u, u(x)= x2 + 1,
then
, D[f(u(x))]= D[sin (x2+ 1)] =f'(u)u'(x) =

(cos u)2x
= [cos (x2 + 1)]2x.
Therefore
f [cos ( x2 + 1)]2x dx= {sin (x2 + 1) + C}.

More generally,
D sin u(x) = [cos u(x)]u'(x),
and so
f [cos u(x)]u'(x) dx = {sin u(x) + C}.
This works for any functions. If
F'=f,
so that
ff(x) dx =
{F(x) + C},
then
D[F(u(x))] =
F'(u(x))u'(x) =
f(u(x))u'(x),
so that
ff(u(x))u'(x) dx= {F(u(x)) + C}.

In such formulas, we abbreviate u'(x) dx by the symbol du. Thus we write
f cos u du= {sin u + C},

which means that for every differentiable function u(x), we have
f [cos u(x)]u'(x) dx = {sin u(x) + C}.

Similarly, we write
fe"du = { e" + C}�
which means that if u is any differentiable function, then
fe""">u'(x)dx = {eu!x> + C}.

This is true, because
Deu<x> = e"«x>u'(x).
Using different dummy letters, we can convert the above formula to any of the forms
or feu<t>u'(t)dt,
and so on. More often, however, we start with an integral described in the long
notation and observe that it is convertible to a short form. For example,
fex2+12x dx
has the form
where u(x) = x2 + 1. Therefore
fex2+12xdx fe"du = =
{ e" + C} =
{ ex2+1 + C}.
Similarly, f [sin (t2 + 1)]2t dt has the form f sin u du. Therefore
J [sin (t2 + 1)]2t dt = J sin u du = { -cos u + C} = { -cos (t2 + 1) + C}.
Note that the solution is not finished in the third formula above, because u is a
function. To complete the solution, we need to express the function u in terms of
che dummy letter t. To sum up:
F' =
f => D[F(u)] = f (u)u'.
Therefore
ff(x)dx = {F(x) + C} => ff(u(x))u'(x)dx =

{F(u) + C}.
In the abbreviated form, using du for u'(x) dx, we have
ff(x)dx =
{F(x) + C} => ff(u) du = {F(u) + C}.
Using this general idea, we can write all of our old integration formulas in the more
general form. The first few look like this:
JUndu={nU:ll + c}, J)udu= {2-Ju + c},

J cos u du= { sin u + C}, J sin udu= { -cosu + C},
J J:.udu={ ln u + C} (u > 0), J e

u
du= { e" + C}.
And of course we have
f [f(x) + g(x)]dx=ff(x)dx + Jg(x)dx,

J!<J(x)= k ff(x)dx, k 0, �
because
D[f+ g] =DJ+ Dg and D (kf) = kDf
Let us now consider how to apply such formulas as these, as a practical matter.
Example 1. Consider
Jcx2 + 1)7xdx.
This is almost, but not quite, in the form
J u7du.
If we take u(x) = x2 + 1, then
du=u'(x) dx=2x dx.
We therefore have
J(x2 + 1)7xdx=Jt(x2 + 1)72xdx = Jtu du '
= {t tu8 + C} = { 1\(x2 + 1)8 + C}.

·
This checks:
D[T\r(x2 + 1)8]=T1s 8(x2 + 1)7 2x=x(x2 + 1)7•

· ·
Example 2. Consider
cos -Jx
J - 1_ dx
"\ x
(x > 0).
The only form that might fit this integral is the form f cos udu. Thus we would have
1
u(x) =.jX, du = u'(x)dx =--dx.
2.jX
The only difference between what we have and what we want is a multiplicative
constant. Therefore
JX
cos 1 ;- 1
dx= (cos v x) x dx= 2 (cos v x) x dx
J J J
;-
Jx J 2J
= 2 Jcos u du= {2 sin u + C}= {2 sin Jx +

C}.
Example 3. Con:.ider
Jecos"' sin x dx.

So far we have only one integration formula involving the exponential function:
feu du= {eu +

C}.
If our problem fits this form, we must have
u(x) =cos x, du = u'(x) dx= -sin x dx.

Here again the multiplicative constant causes no trouble:
x) dx= - eu du
Je00sx sin x dx=
J{- -e008"'(-'-sin
J
= eu + C}= {-ecosx + C}.
Below we shall give a list of all the integration formulas that we can write, at this
stage, on the basis of the differentiation formulas that we know. Special explanations
are needed, however, in connection with the formula for f (lju) du. Given a function
u, defined on a domain where u(x) > 0 for every x, we know that
1
D ln u(x)= - Du(x).
u(x)
We need to know that u(x) > 0 on t� domain under consideration, because only
positive numbers have logarithms. Therefore we write
J �du= {In u + C} (u > 0).
But even where u(x) < 0, it makes sense to write
J �du;
that is, it makes sense to ask what functions f have (1/u)u' as their derivatives. The
answer is easy: if u(x) < 0, then -u(x) > 0. Therefore -u(x) has a logarithm, and
D[ln (-u(x))l = -1- D[-u(x)]= -1- (-u'(x))= --1 u'(x).

-u(x) -u(x) u(x)
This gives us
J �du= {ln(-u) + C} (u < 0).

Hence the two formulas for f (l/u) du in the list below.
Ikf(x) dx k ff(x) dx (k � O)
= (1)
f [f(x) g(x)] dx ff(x) dx f g(x) dx

+ = + (2)
(n � -1) (3)
f � du = { In u + C} (u > 0) (4)
f � du = {In (-u) + C} (u < 0) (5)
f u du { u C}
cos = sin + (6)
I u du { u C}
sin = -cos + (7)
f u du { u C}
sec2 = tan + (8)
f u du { u C}
csc2 = -cot + (9)
I u u du { u C}
sec tan = sec + (10)
I u u du { u C}
csc cot = -csc + (11)
f du { C}
e
"
= e
"
+ (12)
Ja" du {1:"a c} (a 0, a � 1)
= + > (13)
I�
J1 - u2 { u C} (lul 1)
= Sin-1 + < (14)
f�1 u +
{ = Tan-1 u + C} (15)
I u J�
u2 - 1
= { Sec-1 u
.
+ C} (u > 1) (16)
To solve the following problems, you will start by expressing the given integral
in the form f f(u) du. In each such case, you should (a) say what u and du are and (b)
state the general formula that you are applying. It is natural to write down the original
integral first, and after this it would be awkward to interrupt the solution with the
formulas for u(x) and du = u'(x) dx. But u and du can be filled in on the right, like
this, for example:
J(x3 + 1)10 x2dx = Jt<x3 1)1°3x2dx

+
{ u{x) = x3 + 1
= fiu10 du {t TI-u11
= · + C}
du = 3x2 dx
= h\(x3 + 1)11 + C}.

This form of the solution shows what we have in mind; writing formulas of the type
U(X) = X 3 + 1, du = 3X2dX, f tu16 du = a 1\-U11
' + C} Will help you tO avoid
mistakes. For example, you might write hastily
(?) f(x3 + 1)10x2dx = {-fi(x3 + 1)11 + C},
as if it were true that for u(x) = x3 + 1, du = x2dx. When we write formulas for u
and du, we uncover such errors.
Similarly for the following wrong solution:
(?) J<x2 + 1)2dx = {t(x2 + 1)3 + C}.
In full, the solution would begin like this:
(?) J<x2 + 1)2dx

{ u = x2 + 1
du = dx(?!)
The error is obvious, and so we start over again:
J<x2 + 1)2 dx = J<x' + 2x2 + 1) dx = {tx5 + fx3 + x + C}.
PROBLEM SET 6.2
Calculate the following integrals, and check by differentiation in each case. Some of
these problems fit together in sequences, in which the answer to one problem helps in the
solution of another; you should watch for such patterns.
1. f (1 + x2)3x dx 2. f (1 + t 3)t2dt 3. f (2 + u2)3u du
4. I (t4 + l)t8dt 5. f (x2 + t2)3 dx 6. f (x2 + t2}8tx dx

7. f(x2 + t2)3 txd t 8. f (l

- 1
+ y1x)3-=dx
v'x
9. f (t3f2 + 5)10.,/(dt
10. J(rs/2 - l)rs/2dt 11. f (I + sin x)2 cos xdx 12. f (1 + tan x)3i2 sec2 x dx
13 . fv'cos x sin xdx 14. f (e"' + 2)4 e"' dx 15. f (e"' - 2)3e-"' dx
16. f(e"' -l- e-"')2 e"'

( - e-"')dx 17. f (e"' + e-"')3 dx 18. f :
(1 x2)2
dx
19. f(I
x2
+ x3)3
dx 20.
f : 1 x2
dx
21. f x2
1 + xs
dx (There are two intervals to be considered in this problem.)
22. fl in xdx 23. f � In (x2)dx (Same comment.)
24. fsin x cos xdx 25. J .

sin2 x cos xdx 26. J sin3 x cos xdx
27. Jsin101 x cos xdx 28. J cos2 x sin x dx 29. J cos3 x sin xdx
30 . Jcos57 x sin xdx 31. J (1 + tan2 0)dO 3 2. J tan2 OdO
33. J(cot2 0 + 1) dO 34. J cot2 0dO 3 5. J sin 0

--do
cos2 0
3 6. Jcos 0
-- dO
sin2 0
37. J sin 20d0 38. J cos 20dO
39. J(cos2 0 - sin2 0)dO 40. J (cos2 0 + sin2 0)dO 41. J (2 cos2 0 - 1)dO
42. Jcos2 0dO 43. J (1 - 2 sin2 0)dO 44. J (2 sin2 0 - 1)dO
45. J
sin2 0dO 46. J sin2 20dO 47. J sin2 0 cos2 0dO
48. J
cos2 0 sin 0dO 49. J sin 0 (1 - sin2 0)dO 50. J sin3 0dO
51. J
cos (0/2) dO 52. fJl - cos 0
2
dO 5 3. J v' 1 - cos 0 sin 0 dO
54. J
x e-"'2 dx 55. J t2et3 dt 56. J xe"'2 dx
57. f
e2"'dx 58. J e5t dt 59. J e1
2
tdt
6.3 Integrals Leading to the Logarithm and the Inverse Secant, Algebraic Devices 265
60. J dt
et'+3t 61.
f dx
ein sec' x 62.
f cosxdx
esin x
63. J sin t dt
e008 t 64. J2x+idx 65. J xdx
10x•
66.
I dx(10x)2 67.
I (2 dt
+ o-3/2 68. J dt
t(2 +t2)-312
J (2 dt dt dt
I
tdt
69. +i-312) 70.
I .Y1 - t2
71.
.y 1 - t2
dt dt
t2 dt
I - (2t)2
t
72.
I ( .Y1· _ 12)s
73.
I\o/1 +t3
74.
.y 4
I exdx I -=dx
t3 ex ex
75.
I 1 +t
4 dt 7 6.
1 +
77.
Y] _ex
78.
I
ex
Yl - e2x
dx
osx
c--dx xdx
(There are different intervals to consider in Problems 79 through 84.)
79.
I x 80. --
I 2 x
sin
81. J xdx
tan
2 I x x x xdx csc x x xdx

sin cos
8 .
sec 2 +sec
secx +tan
tan
83.
I x x csc
+ csc
+cot
cot
84. J xdx
sec
6.3 INTEGRALS LEADING TO THE LOGARITHM

AND THE INVERSE SECANT. ALGEBRAIC DEVICES
In the preceding section, we got two formulas for J du/u, for the intervals (0, oo)
and (-co, 0).
d u
I -; ; = {In u + C} (u > 0) (4)
du
I - =
u
{ln ( - u) + C} (u < 0). (5)
Since lul = u when u > 0 and luJ = -u when u < 0, these two formulas can be
combined into one:
r- =
du
{ln Ju J + C} (on (0, oo ) or ( - oo, 0)). (17)
• Ll
Here the expression in parentheses on the right reminds us that the formula can
be used on an interval where u > 0, or on an interval where u < 0; it cannot be used
on an interval where u takes on the value 0. When u = 0, there is no such thing as the
"1/u" on the left or the "In lul" on the right. Thus, whenever we apply formula (17),
we might have used formula (4) or (5). The advantage of (17) is that it is easier to use.
Consider
r -l d
x
.
J-2 x
266 The Techniqu� of Integration 6.3
In the fundamental theorem of integral calculus, we take
1
f(x) =- F(x) = In !xi.
x
,
Then F'=f Therefore
l-1d =
2 F(-1) - F(-2) = ln 1-11 - ln 1-21=0 - ln2 = -ln2.
-2 x
This is negative, as it should be; the integrand is negative, and we are integrating
from left to right. The calculation might be confusing if we used formula (5):
r-i dx = [ln(- x)]=; =In [-(-1)] - In [-(-2)] =0 - ln2 - ln2.

L2 x
=
Hereafter, we shall use the following shorthand for this kind of calculation:
r-i dx [In Ix!]=�=In 1-11 - In 1-21.

L2 x
=
In general
[F(x)]�=F(b) - F(a),
by definition. Sometimes, where no confusion could result, we may omit the opening
bracket on the left. Thus
�3I= � � �- - =
We can convert various integrals to the form J du/u. For example,
sin u
J tan u du =
J --
cos u
du.
Except for sign, this has the form
v' v u, dv
J � =cosu du. = - sin
dv
Since
J --;- vl = (v
{ln l v + C} > 0 or < 0),
d:
we have
u du = du = -
J tan
J u
u
-
J
- sin
--
cos
={-In !vi+ C} ={-Jn lcosul + C};
J tan u du = {Jn !sec ui + C} (sec u > 0 or sec u < 0). (18)
This is a standard formula.

6.3 Integrals Leading to the Logarithm and the Inverse Secant. Algebraic Devices 167
Similarly,
cos u
This gives
J cot u du =
J -.
-
smu
du.
f cot udu= {ln Jsin uJ + C} (sin u > 0 or sin u < 0). (19)
By an ingenious device, we can find
J secxdx.
We multiply and divide by secx+tan x, getting
sec2 x+ secx tan x d

J secxdx =
J secx+ tan x
. x.
Since
D secx = secx tan x
and
D tan x = sec2 x,
the integral has the form
where
u = secx+ tan x,
du = (secx tan x+sec2 x) dx.
Therefore
J secxdx = {In u
J l + C} = {ln Jsecx+ tan xi + C}.
As always, the chain rule gives us a more general formula for J sec u du:
J sec u du = {In Jsec u+tan ul + C} (sec u+ tan u > 0 or < 0). (20)
Similarly,
cscx(cscx+cot x)
J cscxdx =
J csc x+cot x
dx = {-lnJcscx+cot xi+ C};
and this gives
J csc udu = { -ln Jcsc u+cot ul + C} (csc u +cot u > 0 or < 0). (21)
Consider now the formula

1
D Sec-1x = (x > 1).
xJx2 - 1
The graph of Sec-1 looks like this:
7r
2 ---- -- ------- --- -----
(See Section 4. 7.) Thus Sec-1 is defined on the interval [I, oo ). But at 1 its tangent is
vertical; and so the differentiation formula holds only for x > 1. It gives
dx
I .J
x x2 -
.
1 = {See1 x + C} (x > 1),
and more generally
I
du
U'\I U 2 - 1
= {See1 u + C} (u > 1). (16)
Notice, however, that the integral
-2 1
J-ax.Jx2-l dx
___
makes sense. We therefore need an integration formula which will apply to this
integrand on the interval ( - oo, -1). On this domain,
1
(-1) 1
Ix.Jx21 - 1
__ dx =
I -x.J( -x)2 - 1 dx=

f .J - 1 du, u u2
where
u(x)=-x, du=(-l)dx.
Therefore, for x < -1,
•
r I
x'./ x·
�x
-1
= {See1 u + C} = {S.ee1 (-x) + C}
= {See1 lxl + C} (x < -1),
because !xi = -x when x < 0. Fitting our two formulas together, and passing to
the general case (with a function u instead of x), we get
fu.Ju2 - 1
du
= {Sec-1 lul + C} (u > 1 or u < - 1) . (22)
There is a rough rule to help you decide which of our present list of formulas to
apply to a given problem: look in the integrand for functions which are the derivatives
of other functions. The point is that all our formulas have left-hand members of the
form ff(u) du; and we need to decide, in each case, what u is.
Example 1.
I: In x
dx.
ls there anything here that is the derivative of something else? Yes:
D In x = .!. .
x
Taking u(x)= In x, we have
du= u'(x) dx= .!. dx.

x
{u
Thus our integral has the form
I
ln3 x
� dx=
I (ln3 x)
1
� dx= I u3 du = l x;
= {: 4
}
+ c = {t ln4x + C}.
du= - dx
x
f
Example 2.
x dx
(1 + x2)7 .
Looking for functions which are derivatives of other functions, we observe that
D(l + x2)= 2x.
Multiplicative constants are no trouble:
x dx !· 2x dx ! u-7 du,
J (1 + x2)7
=
2 J (1 + x2)7
=
2 J
where
u(x) = 1 + x2, du= u'(x) dx= 2x dx.
Therefore the answer is
!. 1 u-7+l } {-l } {-l (1 }

{2 -7 + 1
+ c =
12
u-6 + c =
12
+ x2r6 + c .
Example 3. Sometimes we have to hunt harder:
x dx
J )1 - x4
•
There is no hope that l/J l -x4 is part of du. Either the problem is hard or du
must be x dx, or a constant multiple of x dx. Now 2x=Dx2; and x2 is what gets
squared under the radical sign in the denominator. This suggests
u=x2,
x dx 1 2x dx 1 du
JJ1 - x4=,2 JJ1-(x2)2 =2 JJ 1 - u2
={t Sin-1 u + C} ={t Sin-1 x2 + C}.
Example 4. Some obscure-looking integrals may be calculated algebraically:
�
Jx-1 =J(1 + -) dx.
1
x-1
(Here we have divided the denominator into the numerator, getting a quotient and a
remainder.) Therefore
xdx
= {x + In Ix-11 + C}.
J x-1
Example 5. Sometimes we need to find other algebraic devices, for such problems
as this:
dx
J1 + e-x
As it stands, this is hopeless: nothing in the integrand is the derivative of anything
r
else. But
dx � du
= =. u=ex+ 1
• 1 + e-x J
ex+ 1 u J
={ln lul + C}={ln (1 + e") + C}.
(No absolute-value signs are needed, because 1 + e" > 1 for every x.)
Example 6. Sometimes the same devices appear in more complicated forms:
r ---
dx - e" dx
-
e" dx
-
du --- (u=ex, du=e"dx)
•
-
e" + e-x J J
1 + e 2" - 1 + (e")2 1 + u2 J
={Tan-1 u + C} ={Tan-1 e" + C}.
Here we have used, in combination, the methods that worked in Examples 3 and 5.
Example 7. Often we need routine algebra and arithmetic:
_E_= ! dx
= !.2.
t dx
J + x2 J J
.
4 4 1 + (x/2)2 4 1 + (x/2)2 ·
Here u=x/2, du=tdx. This gives
{t Tan-1 � + c}.
x (1/J3) dx
Similarly,
I J3d-x x2 I/31 Ji -d(x/J3)

=
2
I Jt (x/J3)2
=
{ ;3 }= sin-1 + c .
There was nothing special about the numbers 4 =

. 22 and 3 = (/3)2• In the same
way, we get
Ia2 dx x2 {.!a
+
= Tan-1 � + c
a
} (a > 0),
IJa2dx- x2 { = sin-1 � + C)
a J
(a > 0).
Passin g from x to any differentiable function u, we get two more standard formulas:
du 1 {a }
I
u
1 -Tan- - + C (a > 0), (23)
a2 u2
=
+ a
IJa2du- u2 = {
sin-1 � + c
a
} (a > 0). (24)
PROBLEM SET 6.3
Calculate the following integrals, and check by differentiation in each case.
t
1. I � v' 1 x4
dx 2. J 4 v'1 - t2
dt 3. J � v'1 4y4
d
y
x
4. J y
v'4 - y4
dy J
5. 1 + 9x 4
dx 6. J 1
:39 4 dx
x
7. I 9 :sx4 dx 8. J � 5 t4
dt 9 J �
. 1 z6
dz
10. I � 2 zs
dz J 1�
11.
2zs
dz 12. J 1 �6dz
+ z
x
13. I � 5 zs
dz 14. J 1 :55zs
dz 15. J 7
Vl - x8
dx
x
16. J 3
Vl - x8
dx 17. J 1 � xs
dx 18. J � 1 xs
dx
1
19. J : e•
1
• dz 20. J 1 :t e2t dt 21. J e"' e-x +
dx
22. J e"'e"' - e-x

+
e-
,.dx
·
23. I e"'e"' - e-x
e-x +
dx 24. J (e"' e- )( + "' e"' - e-"') dx
2
x x
25. J (e"' e-•')(e2"'
+ + e-2"') dx 26. J 2
v'2 - x3
dx 2 7. J 2
v'2 - x6
dx
28. J v1 x-2x32 dx 29. J v1x-2x6 2 dx 30. f- 1

xln-2xdx
31. J v1Sin-1x
-x2dx 32. J Tan-Ix
+x2 dx
-1-- 33. f (xe"' + e"') dx
34. J xe"' dx 35. J In e"'' dx. 36. J ln2 e"' dx
sinx) dx+
37. J (xcosx 38. fxcosxdx 39. f (x sinx -cosx) dx
40. J x sinxdx 41.J (2x Inx+ x) dx 42. J xlnxdx
43. J (3x+2 Inxx2 44. J x21nxdx 45. J ln3x
--dx
x
) dx
1
46. J x( l + 1n2x) dx 47. J (x + x1)2In+(x2x2 +2x) dx 48. I tVt21 dt
- 1
49. I x2 V; - 1dx 50.f Ve2x dx 1

1
51 .J (sin-Ix+ Vl � x2) dx
J (Cos-Ix - Vl x_ x2) dx
-
52. J V 1� x2dx 53.f Sin-Ixdx 54.
55. J Cos-Ixdx 56.J Cos-I (2x) dx 57. f �du

1 + e2u
f� 1 + e4u du f du f(Tan-Ix+ 1 :x2) dx
59. l �:2u
58. 60.
6 .1 I 1 :x2dx 62. f Tan-1 xdx 63. Jv1 -z2dz

64. We know that 1
D-x Dx-I -ix-2
=
x2
Consider
= =
•
JI-Ix21 dx.
y
6.4 Integration by Parts 273
In the fundamental theorem of integral calculus, we take
1
f(x) F(x) =
x2' x
Then F' = f Therefore
fl -dx
1
= F(l) - F(-1)
-1
x2
Now we interpret the problem geometrically. We seem to have proved that the region
under a positive function has negative area.
a) What went wrong?
b) Show that the area in question is not only positive but infinite. (This does not
follow from the mere fact that the region is unbounded. Some unbounded regions have
finite areas.)
65. Let R be the region under the graph of f(x) = I/v'�, from x = 0 to x = 1. Show
·
that R has finite area.
6.4 INTEGRATION BY PARTS
By differentiation, we get
D[x sin x] = x cos x + sin x.
Since
D cos x = - sin x,
we have
D[x sin x + cos x] = x cos x + sin x - sin x = x cos x.
Therefore
Jx cos x dx = {x sin x + cos x + C}.
Thus, working backward, we have found the solution of an integration problem which
m ight have looked hard if we had approached it forward, starting'wifu_the unknown
integral f x cos x dx. We shall now describe a general method of solving problems of
this kind.
The formula for the derivative of a product is
D[u(x)v(x)] = u(x)v'(x) + u'(x)v(x).

Therefore
J [u(x)v'(x) + u'(x)v(x)] dx = {u(x)v(x) + C}.

T herefo re
Ju(x)v'(x) dx = u(x)v(x) - Jv(x)u'(x) dx.

Here we have dropped the constant, C, because each of the indefinite integrals on the
two sides of the equation carries its own constant with it; what the equation says is
that the two sides of the equation represent the same class of functions. Using the
short notation
du = u'(x) dx, dv = v'(x) dx,
we get the formula
Ju dv= uv - Jv du.
This is the formula for integration by parts; the word parts refers to the functions
u(x) and v'(x) in the integral on the left. Any time we apply the formula, we replace
one integral by another. The method is useful when the new integral is easier to
calculate than the old one.
Let us first try the method on
J x cos x dx.
Let
u =x, dv =cos x dx,
so that
du= dx and V = Sill X.
(We need not allow for a constant here; any function v whose derivative is cosx
will work. We will return to this point in a moment.) By the basic formula, we get
J
x cos J
x dx = u dv= uv - Jv du= x sin x - J sin x dx
= {x sin x +cosx + C}.
If we had used the seemingly more general
v =sin x + c,
we would have
Judv = x(sinx + e) - J(sinx + ;)dx
= {x sinx + ex + cosx - ex + C} = {x sin x + cos x + C},
exactly as before. The same happens in general:
u(v + c) - f<v + c)du = uv + uc - Jv du - uc = uv J

- v du.
In applying the basic formula, we made what may seem to be an arbitrary choice
of u and dv. We might have taken
u =cos x, du = -sin x dx,
x2
dv = x dx , v = -.
2
This would have given
2
J xcos x dx = J udv = uv - J v du =· � cos x + t x2sinxdx.
J
This is true, but is worthless as a method of finding f x cos x dx, because the new
integral is harder to calculate than the old one.
An equally bad choice would be
u= x cos x, du= cosx - x sin x,
du= dx,
which gives
xcosxdx = udv = uv -
f J J vdtt
= x2cosx -J(xcosx - x2sinx)dx.
Here again the new integral is harder than the old one. We remember also that no
term of the form x2cos x appears in the right answer. Therefore the term x2 cos x
cannot be the beginning of the solution, as we might hope: it must be a blind alley.
These examples indicate that integration by parts can be either a good or a bad
method, according to the skill with which we choose the parts. Practice is a help,
but there are general rules which help us to decide what choices are promising:
1) dv has got to be something that we know how to integrate. (If it isn't, we can't
apply the method at all.)
2) We want f vdu to be an. easier integral than f u du. Therefore we want du to be
simpler than u. At least, we don't want du to be more complicated than u.
3) For the same reason, we want v to be simpler than dv; at least we don't want it to
look worse than du.
These rules are not infallible, but they are a help. Let us try them on
J xe"' dx.
We can integrate both x and e". Therefore (1) gives us no guidance. Rule (2)
suggests that u= x and u = e
r
are both acceptable, but that u = x is to be preferred.
(De"'= e"', which is no worse thane·", but Dx =I, which looks good.) We therefore
use
u= x, du= dx, du= e"' dx, v = e"'.
This gives
J xe"'dx = J u du= uv - J v d u = f
xe"' - e"'dx = {xe"' - e"' + C}.
Note that (3) advises us not to try

"
u = e , dv = x dx,
because we would then get i: = x2/ 2, which looks worse than du. In fact, this choice
won't work. Consider next
f x2e'" dx.
Rule (3) tells us that we had better take du = e'"dx. We therefore take
u= x 2, du= 2xdx, du= e'" dx, v = e'".
This looks good under rule (2), and acceptable under rule (3). We get
f x2e'" dx = f u du= uv - f v du = x2e" - 2 f xe'" dx,
= {x2e" - 2xe'" + 2e"' + C},

by the result of the previous problem. If we hadn't known the answer to the previous
problem, it would still be easy to see that we had made progress, in replacing
J x2e"' dx by J xe"'dx; we would then attack the new problem'-by the same method.
It sometimes happens that integration by parts gives us not aB
' �pression for the
integral that we started with, but an equation that can be solved for, this integral.
Consider
f e"' sin x dx.
We take
u = e'", du= e"' dx, du= sin x dx, v = -cos x,
f e'" sin x dx = -e'" cos x + f e'" cos x dx.
We repeat the process, taking
u= e'", du= e'"dx, du= cos x dx, v =sin x.
For short, we wrf(�
I= f e" sin x dx.
We then have
I = -e'" cos x + f e"' cos x dx = -e"' cos x + e'" sin,x - f e"' sin x dx.
_
Here the last integral is simply the one we started with. Therefore
2I = {e"' sin x - e"'cos x + C},

and
f e'" sin x d/= {t[e'" sin x - e"' cos x] + C}.
Sometimes we need to make a strange choice, in which u is the whole integrand

and du is merely dx. This is what we need, to find
J in x dx.
Here we use
u= lnx,
1
du= - dx, dv = dx, v = x.
x
When we replace 1 by x, we seem to have lost somewhat, but the profit in passing
from In x to l/x more than makes up for it. In fact, this scheme works:
J inx dx = uv - Jv du = x Inx - J �
x · dx
= xlnx --J1 · dx = {xlnx - x + C}.
PROBLEM SET 6.4
Evaluate the following integrals. Each of them can be calculated by the method of
integration by parts. You should try to work these problems with the smallest possible
number of false starts. In each case, survey the situation and try to arrive at a conclusion on
the question of-what choice of u and dv is most promising. If you do this carefully,
you ought to be able to solve each of the problems below on the first try.
Each answer should be checked by differentiation.
1. J ln2xdx 2. J In (x2) dx 3. J (ax)e"'dx
4. J xe axdx 5. J x sin axdx 6. J xcosaxdx

7. J ea'" sin x dx 8. J eax COS xdx 9. J eax sin bxdx
10. J eax cos bxdx 11. J x2sin xdx 12 . J X " COS X dx

9 '
13. J x3e"'dx 14. J x2 ln xdx 15. J2 x In2 xdx
16. J x3 ln x dx 17. J Sin-1xdx 18. J Sin-1(2x) dx
19. J Tan�1xdx 20. J xTan-1xdx 21. J xex sin x dx
22. Derive and check:
J In" xdx = x In" x - n J lnn-l xdx.
Formulas of this kind are called reduction formulas. By n - 1 applications of the

formula, we can calculate the integral on the left.
23. Find J ln3 xdx.
24. Derive a reduction formula for J xn sin xdx.

25. Derive a reduction formula for J xnex dx.

26. Derive a reduction formula which reduces n in J x" Inn x dx.
27. J sin In x dx. (Here you should survey the situation, decide on the most promising
procedure, and then proceed with faith.)
28. J cos In x dx. (Same comment as for the preceding problem.)
6.5 INTEGRATION OF POWERS OF TRIGONOMETRIC FUNCTIONS
We shall find how to calculate integrals of the form
J sinnx cosmx dx,
where n and m are any integers, positive, negative, or zero, and integrals of the forms
f secnx tanmxdx,
and
f cscnx cotmxdx,
where n � 0 and m � 0. We shall discuss the various cases in the order of increasing
difficulty.
m odd and positive. (1)
For example, we might have
f sin2 x cos3 x dx.
The method in such cases is as follows. Since
cos2x + sin2x = 1, cos2x = 1 - sin2x,

and
J sin2x cos3 x dx = J sin2x(l - sin2x) cosx dx
= J sin2x cos xdx - f sin4 x cosxdx
= {t sin3x - t sin5x + C}.

This method works whenever in is odd. (In this case n need not be an integer; it may
be any real number.) For m = 2k + 1, our integral has the form
f sin"x cos2k+lx dx = f sinn.x(cos2x?cosxdx
= J sinnx(l - sin2x?cosxdx .
6.5 Integration of Powers of Trigonometric' Functions 279
We expand (1 - sin2 x)k by the binomial theorem. This gives us a sum of integrals
of the form
f� m du (u = sin x, du = cos x dx).
We integrate these one at a time and add the results.
f sin11 x cosm x dx, n odd and positive. (2)
This is like the preceding case. Here n = 2k + 1, and the integral has the form
J cosm x(sin2 x)k sin x dx = J cosm x(l - cos2 x?(-sin x) dx.

-
Expanding by the binomial formula, we get a sum of integrals of the type
fcos; x( -sin x) dx;

we evaluate each of these by the formula for J u1 du, and add.
J sin" x cosm x, m and n � and even. (3)
To handle this one, we recall that

cos 2x = cos2 x - sin2 x = cos2 x - (1 - cos2 x) = 2 cos2 x - 1.
Solving for cos2 x, we get

1 + cos 2x
cos2 x =
2
Similarly,
cos 2x = (1 - sin2 x) - sin2 x = 1 - 2 sin2 x,
and
1 cos 2x
sm 2 x =
• -
.
2
Making these substitutions in the integrand, we get a form in which the exponents
are divided by 2. For example,
1 cos 2x 1 + cos 2x
J sin2 x cos2 x dx
J dx
-
·
=
2 2
We now make the same sort of substitution again, getting
1 - cos 4x .
1
4J
dx
2
=
J
tx - t cos 4xdx = {tx - ;t2 sm 4x + C}.
When the exponents are large, this method is tedious, but at least we know that it
will work.
(n positive). (4)
For n = 1, we know that
Jtan xdx = {-In jcos xi + C} (cos x > 0 or cos x < 0). (4a)
For n = 2,
Jtan2 xdx J(sec2x - 1)dx.

=
(Remember that 1 + tan2 x = sec2 x.) This gives
Jtan2xdx = {tan x - x + C}. (4b)
For n > 2, we have
Jtann xdx Jtann-2x(sec2x - 1)dx,

=
and so
Jtann xdx =
1
J
-- tann-i x - tann-2 xdx.
n - 1
(4c)
This is called a reduction formula. By repeated applications of it, we can reduce the
integral to one of the forms (4a) and (4b).
Jcotn x dx (n positive). (5)
This is like (4). For n = 1,
cos x .
Jcot x J sm
-.-dx
x
= = {In Ism x I + C} (sin x > 0 or sin x < 0). (5a)
For n = 2,
Jcot2xdx J(csc2x - l)dx.

=
(Remember that cot2x + 1 = csc2x. ) Thus
Jcot2xdx = {-cot x - x + C}. (5b)

6.5 Integration of Powers of Trigonometric Functions 281
For n > 2,
fcotn x dx fcotn-2 x cot2 x dx fcor-2 x(csc2 x

= = - 1) dx;
and so
1
Icotn x dx = - -
n
-
- 1
f
coc-1 x - cotn-2 x dx. (5c)
By repeated applications of (5c), we can reduce our integral to one of the forms
(5a) and (5b).
fsec" x dx, n even and positive. (6)
For n = 2, J sec2 x dx = {tan x + C}. For n = 2k, k > 1,
Isec" x dx fsec2k x dx fsec2k-2 x sec2 x dx

= =
f(1 + tan2 x)"-1 sec2 x dx.

=
When we expand (1 + tan2 x)k-l by the binomial formula, we get a sum of integrals
of the form
Jtan; x sec2 x dx fu; du. =
We integrate each of these by the power formula and add the results. For example,
fsec6 x dx f(sec2 x)2 sec2 x dx f(1 tan2 x)2 sec2 x dx

= = +
f(1 tan2 x tan4 x) sec2 x dx

= + 2 +
= {tan x + i tan3 x + t tan5 x + C}.
fcscn x dx n even and positive. (7)
This is like (6). For /1 = 2,
fcsc2 x dx = {-cot x + C}.
For n = 2k, k > 1,
Jcsc2k x dx Jccsc2 x)k-1 csc2 x

=
Jccot2 1t 1 csc2 x dx
= x + -
Jccot2 x + l)k-1(-csc2 x dx).

= -
When we expand the binomial, we get a sum of integrals of the form
f cot' x( -csc2 x) dx = J uidu.

f seen xdx, n odd and positive. (8)
For n = 1, we found that
secx(sec +tan
f secxd x = f sec x
x
+tan x
x)
dx = fdu-u ,
where
u = sec x +tan x, du = (sec x tan x +sec2 x) dx.
Therefore
fsecxdx = {lnjsecx +tanxj + C}.
For n odd and greater than 1, we have a problem. For example, in f sec3 x dx
it does no good to write
f sec2 x sec xdx f

= (1 +tan2 x) sec xdx,
because the second term fits no standard form. The solution is obtained by integrating
by parts. We have
f seen xdx = f secn-2 x sec2 xdx.

Let
u = secn-2 x, dv = sec2 x dx,
du = (n - 2) secn-3 x sec x tan x dx = (n - 2) sec"-2 x tan x dx,
v =tan x.
This gives
f seen xdx = f udu = uv fvdu -
= secn-2 x tan x - f
(n - 2) secn-2 x tan2 xdx
:; secn-2 x tan x - (n - 2) f secn-2 x(sec2 x - 1) dx
= secn-2 x tan x - (n - 2) f seen xdx + (n - 2) f sec"-2 xdx.

6.5 Integration of Powers of Trigonometric Functions 283
Thus, if I is the original integral, we have
I = secn-2 x tan x - (n-2)1 + (n - 2) secn-2 x dx, J

and
n-2
J seen x dx = --
n-1
1
secn-2 x tan x + --
n-1
J secn-2 x dx.
There is a similar reduction formula which works for odd powers of the cosecant:
1 n-2
J cscn x dx = - --
n-1
cscn-2 x cot x + --
n-1
Jcscn-2 x dx. (9)
To derive this, we integrate by parts, taking u = csc"-2 x, dv = csc2 x, and proceed

as in the previous derivation to solve for J csc" x dx.
By these formulas and methods we can integrate products of powers of trigono
metric functions. There is absolutely no need to memorize the formulas which we have
just finished developing. You can handle simple cases by remembering the methods.
If you need to compute an integral, in one of the difficult cases-and this almost never
happens to people, in real life-then you look up the appropriate reduction formula.
Moreover, it isn't even safe to try to memorize complicated formulas: you are very
likely to misremember them and get wrong answers.
PROBLEM SET 6.5
Before starting to work on these problems, you should read Section 6.5 carefully, until
you understand what the methods are and why they work. In working the problems, you
should refer to the text as seldom as possible. You should try to avoid looking up even the
reduction formulas (8) and (9), unless a problem requires you to apply one of them more
than once. If only one reduction is required, you should integrate by parts, instead of using
the reduction formula. As you will see, the first few problems below are designed to remind
you of the methods that we have been using. Check by differentiation in each case.
1. J sin2 x cos3 xdx 2. J sin3 x cos2 xdx 3. J sin2 x dx
4. J cos2xdx 5. J sin2 x cos2 xdx 6. J tan4xdx
7. J cot4xdx; 8. J tan5 xdx 9. J cot5 xdx
10. J sec4 xdx 11. J csc4xdx 12. J sec3 xdx

13. x dx
f csc3 14. x dx
f sec5 15. x dx
J csc5
16. x x dx
J sin sec 17. x x dx
J cos csc 18. x x dx
J sin sec3
19. x 2x dx
J sin2 sin 20. 2x dx
J cos3 21. 2x dx
f cos2
22. 2x 2x dx
f tan sec 23. x x dx
f csc tan 24. I --dx
sinx
x
cos2
25. --dx cos 1

26. I --dx 1
27. I --dx
x
J
x x
x
sin2 cos4 sin4
2 . J x dx
8
sec
1
tan
29. x1 x dx
J
csc cot
30.
x
There is a reduction formula of the form
J sinn x dx = A sinn-l x x cos +BJ sinn-2 x dx,

where A and Bare constants, expressed, of course, in terms of n. Derive such a formula.
[Hint: It is no use trying to do this merely by the use of the elementary trigonometric
identities relating the sine and the cosine.]
31. There is a reduction formula of the form
J cos" x dx = A cos"-1 x x
sin + BJ cos"-2 x dx.
Derive it.
6.6 INTEGRATION BY SUBSTITUTION
In Section 6.2 we found that there was a close connection between certain simple
integrals and some more complicated ones. For example, if we know that
Jx2 dx = {tx3 + C},
then we know that
J sin2 e cos e de= {t sin3 e + C}.
(We are using a different dummy letter in the second problem, for reasons which will
soon be clear.) Thus we have two related integration problems:
Jx2 dx {tx3 + C}
l x�sino l ,,_,sin o
J sin2 e cos e. dB = {t sin3 0 +

(!)
C}.
6.6 Integration by Substitution 285
The ( !) at the bottom indicates that the equation in the bottom line is the final
conclusion. The pattern here is the following:
ff(x) dx = {F(x) + C}
l X-+U(O) l X-+U(O)
ff(u) du =ff(u(O))u'(O) d() = {F(u(O)) (!)

+ C}.
Thus, if we know how to find Jf(x) dx, we can use the result to find Jf(u) du.
It sometimes happens, however, that we want to move in the opposite direction;
sometimes we can see how to calculate
ff(u(e))u'(e) d(),
and we want to use the result to calculate Jf(x) dx. So as to give ourselves a simple
example to work with at first, let us suppose that we know about the functions
Sin and Sin-1, but do not know that l/.J 1 - x2 is the derivative of Sin-1 x. We then
consider
J dx
.J 1 - x2 •
We observe that it does not fit any form that we know. But perhaps it would be
manageable if we could extract the indicated square root. For x = Sine, the square
root can be extracted. (See below.) If we replace the dummy letter x by the function
Sine, then dx becomes Sin'() de, and we get the related integrals on the left in the
following diagram:
J dx
=?
l
.Ji - x2
X-+�inO
·J cos e de _ ?
.J1 - Sin2 e � .
The trigonometric integral is easy:
cos e de cos e de
J.J1 J .J =
J = {O C}.
_
_
1 d() +
- Sin20 cos2e
(Query: Why is it true that

.Jcos2 () = cos e'
for the values ofe that we need to consider? What values of() do we need to consider?)
286 l'he Technique of Integration 6.6
The above calculation enables u::. to complete our diagram:
fJldx- X2
=
(I)
{Sin-1 x + C}
1 x-Sino ·r o-Sin-1x
f cos e de
= {0 + c}
J1 - Sin2e
In this case, of course, the solution in the top line was known before we started.
But the same scheme works in general, whenever we can calculate the new integral
on the lower left:
ff(x) dx
l x-u(8)
ff(u) du ff(u(O))u'(O) dO
= = {G(O) + C}
We shall prove, at the end of this section, that this procedure is valid, whenever the
symbols u' and u-1 have a meaning; that is, whenever u has both a derivative and an
inverse. Meanwhile, we shall show how the scheme is used to solve problems
which would otherwise be hard.
Example 1.
f dx
x2J1 - x2 -
_?
•
(-1 < x < 1, x ¥: 0).
As in the preceding case, it seems to be the radical that is causing the trouble; and so
we get rid of it by the substitution
x--)-Sin 0 ( - :!. < e < :!.

2 2 '
)
dx --)-Sin' e dO = cos e de.
This gives
f J dx --)- f cos e dO
J csc2 0 dO {-cot 0 + C}.

=
=
x2 1 - x2 Sin2 0 cos e
(Throughout, -Tr/2 < e< Tr/2; on this interval, sin x = Sin x, and the usual
identities hold automatically.)
We now reverse the substitution, using e--)- Sin-1 x. This gives
fh
x2 1 - x2
=
{-cotSin-1x + C}.
A formula for cot Sin-1 x is easy to read off from a figure.

- 1
Here
Sine= x, e= Sin-1 x;
.
cot Sm 1
_
x= cot e= -k = J 1 - x2
-'-----
x x
Therefore
dx = {- J� + c }
J x2J1-x2 x .
Note that all the trigonometry has cancelled out of the problem. Our answer checks:
DJ� = l. x
x x2
-x
J1-x2
(
- J1 - x2 · )
= 1 [-x2-cJ1-x2)2J= -1 .
x2)1 - x2 x2)1 - x2
We can sum this up in a diagram as follows:
dx {- � }
Jx2J1-x2=(
+ c
x
l x�SinO I o�Sin-1 :v
J e de =
csc2 {-cot e+ C}
-.. e
- x2•
The substitution x Sin is the usual one to try, if the troublesome part of the
integrand is
J1 In other cases, x-... Tan e works in much the same way.
Example 2. Consider
To get rid of the radical, we use
x -+ Tane (- 2 :!!. < e < !!)

2 '
dx -+ sec2ede.
This gives
j..J1 + x2 dx->-j..J1 + Tan2esec2ede.

The domain of Tan is the interval (-TT/2, 7r/2), on which sece > 0. Therefore
sece= ..)1 + Tan2e, and
JVl + Tan2esec2ede= sec3 ede. J

We now use one of the reduction formulas of Section 6.5:
Jsec3ede= fsecnede (n = 3)
n
1
= -- secn-2etan e+
n-1
2
n-1
- Jsecn-2ede
= t secetane+ t secede J
= {tsecetane +tin lsece +tanel+ C}
= {G(e) + C}.
We complete the solution by letting e -+ Tan-1 x. This gives
J ..J1 + x2 dx = {G(Tan-1x) +C}

= {!(sec Tan-1x)(tan Tan-1x)
+ t In I sec Tan-1x + tan Tan-1 xi + C}.
Obviously tan Tan-1 x = x. The formula for sec Tan-1 x can be read off from a
figure.
y
In the figure, -7T/2 < e < 7r/2, bute may be positive or negative. We take OP = 1.
This gives
x =Tane, e = Tan-1 x, r = sece = secTan-1 x,
so that
secTan-1x =.Ji+x2•
Therefore the answer is
f.J1 +x2dx = {tx.Jl +x2+ tlni .J l +x2+xi+ C}.
This can be simplified slightly: since .J 1 +x2+ x > 0 for every x, we can omit the
absolute value bars, getting:
f.J1 +x2dx = {tx.J l +x2+tln(.J l +x2 +x) + C}
As before, we sum up in a diagram the process by which the problem was solved:
I.JI+x2dx = {tx.J l +x2+tln(.J l +x2 +x) + C}

(!)
l x-Tane Je-Tan-1 x
J 8d 8 =
sec3 {tsec8tane +tlnjsece + tan8j + C}
Such diagrams are worth drawing, especially the first few times you use the
substitution process; often the calculations are long, and it is easy to lose track of
what the process means.
The answer in Example 2 suggests that no method would have made the problem
seem easy. Note that the formulas of Section 6. 5 are turning out to be useful in solving
problems which do not appear, at first, to involve trigonometry at all.
We return to the general theory, to see why this method works. The pattern of
our work is described by the diagram:
ff(x) dx 0) {G(u-1(x)) + C}
l x- 1<(6) r e- u-1(x)
Jf(u(8))u'(8) de = {G(8) + C}
What we are claiming, when we use the method of substitution, is that, if the second
equation holds, so does the first. In terms of the definition of the indefinite integral,
this means the following:
Theorem 1. If u is differentiable and invertible, then

G' = f(u)u' => D[G(u-I)] =f
The proof is as follows. By the chain rule,
D[G(u-I)] = G'(u-I) Du-1.
By hypothesis, G' = f(u)u'. Therefore
G'(x) = f(u(x))u'(x)
G'(u-I) = f(u(u-1))u' (u-1),
and
D[G(u-1)] = G'(u-1) Du-I = f(u(u-1))u'(u-1) Du-1.
Now
f(u(u-1)) = f,
because u(u-I(x)) = x for every x. Therefore
1
f(u(u-1))u'(u-I) Du-I = f · u'(u-1) Du-I = f · u'(u-1) · -- ,
u'(u-1)
by the general formula for the derivative of the inverse of a function. Now u' (u-1)
cancels, and gives us D[G(u-1)] = f, which was to be proved.
PROBLEM SET 6.6
Calculate each of the following integrals, by any method. In most cases, but not all,
the easiest method is to use a substitution of the form x _,.Sin 8, x-+ Tan 0, or x -+ Sec 0.
In each case where you do use the method of substitution, you should sum up the process
of solution in a diagram as in Examples 1 and 2 in the text. Finally, check in each case by
differentiation.
1. J (1 - x2)-3!2dx 2. J v' x2
dx
+ 1
3.
J
dx
v' x2 - 1
4. J dx
x(l + x2)
dx 5.
J xv' x2 - 1
dx
dx 6. J x(l - x2)-3f2dx
7. J dx
x2(1 + x2)
8. J 1
xdx
+ x2
9.
J Vl - x2dx
10.
J
dx
x2v' x2 - 1
11.
J v' x2 - 1
xdx
12. J x2v'1 - x2dx
13. J x2dx
Vl - x2
14.
J
x3dx
v'1 - x2
15.
J 1 +
x2dx
x2
16.
J x2v'l + x2 dx 17. J� v'l - x2dx 18. J (1 : x2)3dx
19. J xv'l + x2dx 20.
J x3 v' 1 + x2dx 21.
J 1
x3
+ x2
dx
22. J x2v' x2 - 1dx

6.7 Algebraic Substitutions 291
Jb f(u(O))u' (0)dO iu(b)

23. a) Show that
= /(x) dx,
a ida>
whether u is invertible or not.

b) Show that if u is invertible, then
lcl f(x)dx J,"-l(d) f(u(O))u'(O)dO.
=
C u-l(c)
24. Obviously there is no point in writing this on a paper which is to be turned in and
graded; but for your own benefit, reproduce the proof of the following, without reference
to the text:
G' = f(u)u' => D[G(u-1)] = f
6.7 ALGEBRAIC SUBSTITUTIONS
It is a good rule, if you have a problem which you don't see how to solve, to try to
think ofan easier problem that resembles it. If you can solve the easier problem, and
bridge the gap between the two, then you have solved the problem which you started
J-::: =x2= == d x.
with. For example, consider
.J2x + 1
This does not fit any of the standard forms that we know about. There is no reason to
.)-;, or .J(, then

suppose that a trigonometric substitution would help; and in fact, none of them
would. We note, however, that ifthe denominator were ofthe form
the problem would become easier. Now
2x + 1 = t�x = Ht - 1 ) .
We therefore try the substitution
x - u(t) = Ht - 1),
dx-+ u'(t) dt = t dt.
x2 dx -+ J t(t - dt
Under this substitution,
.
1)� k
J.J2x +
·
1 .Ji
The latter integral is easy to calculate. It is
! J t2 - 2� + dt i ·J(t3/2 - 2tl/2 + t-1/2) dt

.jt
1
{i{i·t5;2 - 2 . ita12 + 2t112) + C}

=
h\ts12 - tta12 + it112 + C}

=
{G(t)
=
=
+ C}.
To get the answer to the problem which we started with, we use the inverse substitution
t-+ u-1(x) = 2x+ 1.

This gives
f x2 dx
= h�(2x+ l)s/2 - t(2x+ l)a/2+!(2x+ 1)112+ C}
-J2x+ 1
= {G(u-1(x))+ C}.
The scheme here is the same as in the preceding section:
x2
J-J2x+1 dx = {G(u-1(x))+ C}
<1>
lx�1i(t)
t2 - 2t+ 1
J dt {G(t)+C}
-
r -
8v t
The only differences are that (a) the functions u and u-1 are described algebraically,
and (b) the formulas for G(t) and G(u-1(x)) are too long to be conveniently written in
the diagram. In any case, we know that the method works: this follows from Theorem
1 of Section 6.6.
Often we can tell that a substitution is going to work, long before we know what the
answer is. As soon as we wrote
x2 dx
J-J2x+1 -+
JW - 1)� l dt
-Jt
·
,
it was evident that the numerator was a polynomial. We can integrate the quotient
of a polynomial and a power. Similarly, we know that we can integrate
f (x3 - 3x/ +4)2 dx'.

•
x2 3
the calculation will be tedious, but the outcome is not in doubt.
If one algebraic substitution works, there are usually others that also work.
In the preceding problem, we might have used
y2=2x+l, x = t(y2 - 1).

Thus we use
x -+ u(y) = t(y2 - 1),
dx-+ u'(y) dy = y dy,

x2-+ t(y2 - 1)2, -J2x+ 1 y,
- ; ay =i cy4
_,..
HY2 - 2. Y
J J - 2y2+ 1) dy
= {21oy5 - i-y3 +b + C}.

As usual, we now reverse the substitution, using
This gives the final answer
{�0(2x + 1)512 - t(2x + 1)312 + i(2x + 1)1'2 + C},

as before.
There are no rules which tell us the best substitution to try in every case. The
best approach is to look at the integrand, ask ourselves what feature of it is most
troublesome, and then choose a substitution which seems likely to remove the
troublesome feature. For example, if we want to calculate
j. dx
'
-Jx2 + 1
we want to extract the square root; we can do this if
x---+ Tan e ( - !!.2 < e < !!.)2 ,

x2 + 1 ---+ sec2 ().
This works:
J-Jx2dx + 1
---+ J sec2 e de
sec e
= J sec e de ,
which leads to a solution, as you found in Problem 2 of Problem Set 6.6.

We might also have tried
z2 = x2 + 1,
x ---+ u(z) = ,/ z2 - 1
This gives
dx =
\ z
/
2
- 1
dz.
J
dx
-Jx2 + 1
__..
J! z
.
,Iz2
z
- 1
dz = J-J__:!:___
-z2
, 1
which gets us nowhere, unless we happen to remember the solution of Problem 3 of
Problem Set 6.6.
Usually, to find out what algebraic substitution is going to work, we need to
solve an algebraic equation. For example, given
dz
J 1 + ./:'
we wish that the denominator had merely the form t. To get
1 +;-;� t,
we need
z� (t - 1)2•
We usually write this with"=" signs:
1 +;;=t, J;=t- 1, z = (t- 1)2•

This gives
z� u(t) = (t- 1)2,
dz� u'(t) dt = 2(t - 1) dt,
2(t- o dt = (2 - �) dt
- Ji :=J:Z � J t J
= {2t - 2 ln !ti + C}.
The reverse substitution
t� u-l(z) = 1 +)-;,
gives the final answer
{2(1 +,Jzj -
2 In (1 +-Jzj + C}.
(Query: Would it be all right to delete the "1" in the first parenthesis?)
This is probably the most efficient solution. If we hadn't thought of it, we might
have tried
,J; = t,
which gives the substitution
z� u(t) =t2,
dz� u'(t) dt = 2t dt,

dz 2t dt
J +Jz � J1+t.
l
Dividing 1 +t into t, we get
t 1
--=1---.
l+t l+t
Therefore
2J_!___E:!_ =2J(1 - -1-) dt= {2t- 2 Jn jl+ti. + C}.

l+t l+t
Finally we apply the inverse substitution
t---+ u-1 (z) = .jz,
getting
I 1 +dz.J
--- = r - 2 ln (1 + vz)
{2vz r + C}.
z
(Is this really the same as the previous answer? Why or why not?)
We have used the substitutions x---+ Sin8; x---+ Tan8, and x---+ Sec() to handle
integrands involving the radicals
and
Slight variations enable us to take care of more general cases, involving
.Ja2 - x2, .Jx2 - a2.

For example, to find
J Ja2 - x2dx (a > 0),
we use x---+ a Sin() = u(8), so that
a2 - x2 = a2(1 - Sin28), Ja2 - x2 = a cos().
Here x/a = Sin8, so that
() = Sin-1 �.
a
Thus
f Ja 2 - x2dx ---+ J (a cos8)a cos8dfJ = J

a2 cos 2fJdfJ
= J
a2 t ( cos 28 + 1) dfJ
2
= {: \in2fJ + � e + c .}
Now
sin 28 = 2 sin()cos8 = 2 · � ! Ja2 - x2•

·
a a
This gives
2
J .Ja2 - x2dx = { tx.Ja2 - x2 + � Sin-1 � + c} .
In the same way, we use
x---+ a Tan 8,
to get. rid of -Ja2 + x2; and we use

x ---+ a Sec e,
a
to get rid of -J x2 - a2.
There are miscellaneous substitutions which work on miscellaneous problems .
For example, in
dx
J x2(x 2 + 1)
'
the trouble seems to be that the integrand is concentrated in its own denominator.
We ought to be able to correct this by letting
1
x ---+ u(t) = - ,
-1
dx ---+ - dt.
t2
This gives
fx2(x� +
x
1)
---+ J( -1+
1
� t2) dt {-t +Tan-it= + C}
---+ {- � + Tan-1 � + c}.

Here, in the last step, we have applied the inverse substitution
-
t ---+ u 1(x) = x. 1
-
In writing up solutions of problems, in the following problem set, you need not
draw diagrams of the form:
ff(x) dx = {G(u-1(x)) + C} (!)
1x�11(t)
ff(u(t))�'(t) dt = {G(t) + C}
But whenever you use a substitution, you should explain what you are doing, by
writing formulas of the type
x ---+ u(t) = · · · ,
6.8 Algebraic Devices: Completing the Square and Partial Fractions 297
PROBLEM SET 6.7
Calculate the following, by any method.
dx vx
1.
J (I + V:x-)3
2.
J (1 + Vx)3
dx 3.
J (a2 _
x2)-af2dx
4. J (a2 + x2)-af2dx 5.
J
dx
v1 + e"'
6.
J
dx
v1 - e"'
dx dx dx
7.
J l + �x
8.
J (1 - �x)2
9.
J Yvx + 1
10.
J
z3dz
vz + I
11. J Vz2
z3dz
- I
12.
J
dx
YI + �'x
dx dx
13.
J � l + vx
14.
J x4(x - I)
15.
J (l - x2)4dx
dx dx
16.
J (I + vx)3vxdx 17.
J� 18.
J (1 + e"')4
19.
J
dx
Yl + e2"'
20. J Sin-1xdx 21.
J xln xdx
22.
J x Sin-1xdx 23. J Tan-1 xdx 24.
J xTan-1 xdx
25.
J
l
(1 + vx)2
dx 26. J 1
vx(l + vx)2
dx 27.
J
1
l + fix
dx
28. J v1 +fix
I
dx 29. J : (l
dx
e"')2
30.
J
dx
v 1 + e3X
31. J dx
x3(1 - x)
32. J x2 ln xdx 33. J x2Tan-1xdx
6.8 ALGEBRAIC DEVICES: COMPLETING

THE SQUARE AND PARTIAL FRACTIONS
In Section 6.6 we used trigonometric substitutions to calculate integrals involving

Ja2 ± x2 and Jx2 - a2. By completing the square, we can extend these methods so
as to take care of expressions of the form Jax2 + bx + c. For example,
x- + x +
9
1 x- + x + i + !
?
(x + t)- +
? (.)3)2
2
= =
Therefore
dx dx
J v x2 + x +
I
1
-
J '
I
(x + t)2 + (./3/2)2
'
JJuzdu+
which has the form
.
a2
We can calculate this by the substitution

u--+ a Tan e.
This gives
{lnl .Ju2 +a2 +ul + C} = {ln l .Jx2 +x + 1 +x +ti+ C}

.
Similarly,
x2 +x = x2 +x +t - t = ex +t)2 - ew,
and so
dx dx
J .Jx2 +x
= J .Jex +t)2 - eW'
which has the form
J du.Ju2 - a 2
.
Here we would use
u--+ a Sec(),
and proceed as in Section 6.6.
The following simple-looking problem has a curious solution:
J �=?
x2 - 1
We try
x--+ Sec() ex> 1)
so that
dx --+ sec() tan () d(),
giving
sec() tan() d()

J tan2 ()
= J csc () d() = {-In Iese e +cot e1 +C}
Now
J
x 1 x+ 1 x+ 1
--+ = = ex> 1).
x -1
-
.Jx2 - 1 .Jx2 - 1 .Jex + 1)ex - 1)

This gives the answer
J� { 1
x2 - 1
= t 1n � +
x +1
1 c} ex> 1).
We check by differentiation:
( l : �� I )
D tin = Detinlx - 1\ - tln\x +1\)
1 1 1 1 1 ex + 1) - ex - 1)
=-·
2 --
x -1
-
-·--
2 x+1
=-·
2 ex - l)ex + 1)
1 •
=
x2 - 1 ·
6.8 Algebraic Devices: Colfi'pleting the Square and Partial Fractions 299
This shows that our answer was right. But it also shows that our use of trigonometry
was unnecessary; the solution depends merely on the algebraic identity
_l
1 _l.
2_
_ 2_ +
__
x2 - 1 x - 1 x +1
This suggests that we should have a systematic method of breaking up rational

functions into sums of simpler functions. We call this the method of partialfractions.
Theorem 1. If a � b, then there are numbers A and B such that
ex+d A _B_
= + (x � a, b).
__
(x - a)(x - b) x - a x - b
Proof The obvious method works:
ex +d
= � +_!}__
(x - a)(x - b) x - a x - b
{
<=> ex +d = A(x - b) +B(x - a)
<=> A+B = e and
Ab+Ba= -d.
We solve for A and B by any method, getting the solution
ae+d be+d.
A= B=
a - b' b - a
These values satisfy both equations.

Note that since a - b appears in the denominators, we really needed the hy
pothesis a -¥- b. And for a= b, the theorem is false. That is, you cannot express
1/(x - a)2 in the form
A B A+B
-- +-- =--.
x - a x - a x - a
It might seem that we should have stated a stronger theorem, as follows:
Theorem 1. If a � b, then
ex +d ae+d . 1 be+d . _1_

=
_
+
_
.
(x - a)(x - b) a -b x - a b - a x -b
But nob'ody could remember this formula. The efficient way to handle such
problems is the following. Given
fcx 2��x
- - 5)
=
?
we know by Theorem 1 that there are numbers A and B such that
(x
1 A B
------ = -- +-- .
- 2)(x - 5) x - 2 x - 5
The only problem is to find out what they are, numerically. We first write
1 = A(x - 5)+ B(x - 2).

Since this equation holds for every x, it must hold for x =2 and for x = 5. Therefore
1 =A· (-3), 1 =B· 3;
A= -t, B =t.
This is another example of the efficiency of existence theorems: often, if you

know in advance that a problem has a solution, you can use a simple procedure to
find out what the answer is. Without Theorem 1, the shortcut calculation of
1 _!_ .__
l l __
1
=- + .
(x - 2)(x - 5) 3 x-2 3 x-5
would not have been valid. To see this, consider the following analogous procedure:
"Problem." Find the numbers a and b such that

sin x =ax+ b.
"Solution." Letting x =0, we get
0 =a· 0+ b.
Therefore b =0, and
sin x =ax.
Letting x =7r/2, we get
l=a·2:
2'
and a =2/7r. Therefore
2x
(?)
. Sill
. X =- for every x (?)
This is wrong: in fact, our formula is correct for only three values of x, namely,
x = 0 and x = ±7r/2. The fallacy was in assuming at the outset that the problem
had a solution, when in fact it has none. What the above line of reasoning really
proves is the following:
sin x is a linear function (1)
=> sin x is the linear function 2x/7r. (2)
The statement (1) => ( 2) is true, but it is not useful, because (1) is false.
The method that we used for quadratic denominators also works whenever the
denominator can be factored into linear factors all of which are different.
2x2+ 1 A B C
��-- = + + �� ,
(x+ l)(x+ 2)(x+ 3) x+ 1 x+ 2 x+ 3

2x2+ 1 =A(x+ 2)(x+ 3) + B(x+ l)(x + 3) + C(x+ l)(x+ 2),
x = -1, 3 =2A, A-- �·
2,
x =-2, 9 = -B, B= -9;
x = -3, 19 =2C, C=:!ll-.

This solution depends on the following existence theorem.
6.8 Algebraic Devices: Completing the Square and Partial Fractions 301
Theorem 2. If p(x) is of degree � 2, and a, b, and c are all different, then there are
numbers A, B, and C such that
p(x) = __ A + __ B + _ c
.
_
(3)
(x - a)(x - b)(x - c) x - a x - b x - c
With our present equipment, we could give only a brute-force proof. But we know
how to handle simple cases, and in the following problem set you will see how various
more difficult problems of the same type can be solved.
PROBLEM SET 6.8
Find:
dx dx dx
1.
J x2+2x+5
2.
J v'x2+2x+5
3.
J x2 - 4
dx dx dx
4.
J x2+x- 4
5.
J Yx2+x - 4
6.
J v2 - x2
dx
7.
J >12 - 2x - x2
dx
8.
J v -2x- x2
(Is this an impossible problem at the outset?)
dx dx d
9.
J x2+ 6x + IO
10.
J v-x2- 6x +10
11.
J x2 - ;
6 +10
dx x dx dx
12.
J (x - l)(x - 2)
13.
J (x - l)(x - 2)
14·
J x(x - l)(x - 2)
x dx
15.
J x(x- l)(x- 2)
Find the unknown coefficients A, B, C, . . . which satisfy the following equations:
1��x
1 A B C
16.
(x - I)2 (x - 2)
=
(x - 1)2
+
x- 1
+ --
x - 2
--
17. Find
J (x _ _ 2)
1 A B C D d
18.
x2 (x - l )2 = x - + (x - J)2 +x- I
2 +x --
19. Find
J x2(x
�
1)2
I A B c D
20. = + + +
x+ I (x- 2)3 (x - 2)2 (x - 2)
---
(x+ l)(x - 2)3

--
dx I A Bx+ C
21. Find
J (x+ l)(x- 2)3
22.
2
x(x + I)
=-+
x x2+ I
dx 1 A Bx+ C Dx+ E
23. Find
J x(xz + I)
24.
x(x2+1)2
=-+
x (x2+ 1)2
+
x2+1
dx
J J
sine
25. Find 26. Find d8
x(xz+1)2 1 + cos 8
27. Given 8 = 2 Tan-1 x, calculate sin 8 and cos 8, in terms of x.
28. Find J d8/(1 +cos 8). (One way to do this is to use the substitution
= 2 dx
8 ->- u(x) = 2 Tan-1 x, d8 _,. u'(x) dx
1 + x 2'
sin 8 _,. ? cos 8 ->- ?
But when you see the answer, you may be able to think of a simpler method of solution.)
Find:
f f f
d8 d8 dx
29. 30. 31.
1 +sin 8 cos 8 x2 +6x +9
f f J
d8 d8
32. 33. 34. sec3 8 cot 8 d8
sin 8 +cos 8 sec 8 +tan 8
f J
d8 d8
J
d8
35. 36. 37.
1 - sin 8 2 +cos 8 cos 8 - sin 8
The
7 Definite Integral
7.1 THE PROBLEM OF ARC LENGTH
-C.o_!l§ider the parabolic arc which is the graph of y = x2, 0 � x � 1. We shall

calculate its length. The ideas that we use to do this will apply to other curves, but the
general problem is no harder than the special one; if we were interested in arc length
only for parabolas, the ideas in this section would all be needed.
The length of an arc of a circle is defined (as in Section 4.3) as the limit of the
lengths of the inscribed broken lines. More generally, suppose that f is a continuous
function on a closed interval [a, b].
y
By a net over [a, b] we mean an ascending sequence N of numbers

a = x0 < x1 < · · · < xi < X;+i < · · · < Xn = b.
For each i, let
We join successive points P;_1, Pi with segments, getting a broken line as in the figure.
Such a broken line is said to be inscribed in the graph off Its length is
PoP1 + P1P2 + · · · + Pn-1Pn.
We denote this by p(N). That is,
n
p(N) = I P;-1P;.
i=l
We use the functional notation p(N), because when the net N is named, the broken
line is determined, and so also is its length.
The graph of a continuous function on a closed interval may have infinite length.
But if the length is finite, we ought to be able to approximate it by using a net N
which cuts up [a, b] into very small pieces. This idea is the basis of the following
definitions.
303
304 The Definite Integral 7.1
Definition. Let
be a net over [a, b]. The mesh of N is the largest of the numbers
The mesh of N is denoted by INI .
Definition. If p(N) approaches a limit L, as INI approaches 0, then fis said to be

rectifiable, and the number Lis called its length.
We need, of course , to explain what is meant by the statement
lim p(N) = L.
INl->O
Intuitively, this means that p(N) R:! L when IN I R:! 0. We define this idea by the same
method that we used to defn i e the limit of a function at a point . To make the analogy
clearer, we write the old and new definitions in parallel .
Definition. Let f be a function, on an Definition. Let f be a function, on an

interval [a, b]. Let x0be a point of [a, b]. interval [a, b].
Suppose that for every E > 0 there is a Suppose that for every E > 0 there is a
a > 0 such that if x is a point of [a, b], a > 0 such that if N is a net over [a, b],
then then
0 <Ix - x01 <a INI < 0
=> lf(x) - LI < E . => lp(N) - LI < E.
Then Then
limf(x) = L. Jim p(N) = L.
x--+xo INl->O
To calculate the arc length, we fri st express the length p(N) (of the inscribed
broken line) in terms of things that we know how to handle. By definition,
n
p(N) = I P;-1P;.
i=l
(See the figure on p . 303.) The segment from P;_1 to Pi looks like this:
Yi=f(xi)
Yi-I= f(Xi-1)
�-+-��--'---.L__��-'--�-x
Xi-I Xi Xi
7.1 The Problem of Arc Length 305
Thus
= [i (��:YJ Lix;.
+
Here the fraction Liyi/Lixi is the slope of the chord from Pi-l to Pi; and the mean-value
theorem says that this is the slope of the tangent line at some intermediate point.
Thus we have
Liyi
= f'(xi) (xi-1 < X; < X;).
Lixi
Making this substitution, and extracting the square root, we get
Taking the sum from i =1 ton, we get
n n
p(N) =I ;-1 =I J1
i P P; + [f'(xi)]2 Lixi.
i
•
=l =l
The problem is to find out what happens to the sum on the right as INI -+ 0. We can
find this by giving a geometric interpretation to the sum.
y y
g
� -t-�-�X i---
I X�i�X�.i��---x
For each x on [a, b], let

g(x) = J1 + [f'(x)]2.
In the figure,
y =
g(x), X;-1 <Xi < X; for 1 � i � n.
On each little interval [xi_1, xi], of length Lixi X; - x,_1, we have set up a rectangle
=
with [xi_1, xi] as base, and altitude g(xi). The area of this rectangle is then
g(xi) Lixi.
The sum of these areas is
i g(x;) Lix; i_i=l J1

i=l
=
+ [f'(i;)]2 Lixi = p(N).
If f' is continuous, then so is g; and I g(xi) Lixi ought to be close to the area
under the graph of g, when the mesh of the net N is small. That is, we ought to have
INI � 0 => i� g(xi) !ixi fg(x) dx,

�
which means that
lbg(x) dx.
n
lim L g(xi) !ixi =

INl->Oi=l a
This gives us a formula for arc length:
L = .J1f + [f'(x)]2 dx.
This holds whenever f' is continuous; and we will complete the proof later in this
chapter. Meanwhile, consider some examples.
Example 1. Let
f(x) = 1, O�x�l.
Thenj'(x) = 0, and
L = Jf1o .J1 -+ 02 dx = 1,
which is the right answer.
Example 2. Let
f(x) = kx, O�x�l.
Thenf'(x) = k, and
L f
= .J1 + k2 dx = .J1 + k2,
which is the right answer, by the Pythagorean theorem.

Example 3. Let
f(x) = .J1 -x2, O�x�.J2.

-
- 2
Then
-x x2
f'(x) = , [f'(x)]2 = '
.J1 1
-
- x2 -x2
1 + [f'(x)]2 =1 - x2 + x2 = __' 1
1 -x2 1 - x2
7.1 The Problem of Arc Length 307
and
i,1;10 i\!212
"-J1 + [f'(x)]2 dx
dx ,;-
[Sin-1 x]0 212
-Ji
L = = =
o o - x2
= Sin-1
-J2_ = !!. .
2 4
This is the right answer, because Lis one-eighth of the circumference of a circle of
radius 1.
Example 4. We return to f(x) = x2, 0 � x � 1. Here

/'(x) =
2x , 1 + [j'(x)]2 = 1 + 4x2,
L =
f-J1 + 4x2 dx.
Now
u = 2x.
This is
H-x-JI + 4x2 + t In 12x + -J1 + 4x21 + C}.
Therefore, by the fundamental theorem of integral calculus, we have
L =
ls+ t In c2 + -Js).
2
The answer suggests that no method would have made the problem look easy.
PROBLEM SET 7.1
Find the lengths of the graphs of the following functions, between the indicated limits.
I. f(x) = x312, 0 � x ;;; 2

2. f(x) = x, 0 � x � 7T/4
In cos
3. f(x) = x, 7r/4 � x � 7r/2

In sin
*4. f (x) = Jn x, 1 ;;; x ;;; 3

5. f (x) 1 + tx3i2, 0 ;;; x � 4
=
x
*6. f (x) = e , 0 � x � 1
"'
7. a) /(x) = t(ex + e- ) , 0 � x � 1 (You can solve this one, by an algebraic trick,
without using any of the standard formulas for hyperbolic functions. But the
problem is a little easier if you remember that sinh x = He"' - e-" ,
) cosh x =
}Ce" + e-x .) For the definitions of the hyperbolic functions sinh and cosh, and the
formulas governing them, see the end of Section 4.11.)
8. Let f be a function with a continuous derivative, on an interval containing x0•
"'
Let r(x) P0P,,/P0Px, that is, the ratio of the arc length to the length of the chord.
=
Show that lim,,_x0 r(x) I.

=
"'
To prove this, you will need to use the formula which expresses P0P0. as an integral.
In many books you will see a "proof" of the integral formula for arc length, based on
the assumption that r(x) --+ I. This is an example of a proof of infinite thinness: the
hole in it is as big as the proof, because to fill the hole you must first prove the theorem
itself by another method.
9. Consider the sequence of broken lines suggested by the figure below. Each broken line
forms a stairway from P to Q. The nth stairway has n vertical segments and n horizontal
segments . For each n, let B,. be the length of the nth broken line. Find limn-oo Bn.
10. Let/be any function on [a, b], and let x be any point of(a, b). Let mv m2, and mbe the
slopes of the chords over the intervals [a, x], [x, b], and [a, b]. Show that mis between
m1 and m2 . (Unless, of course, m1 = m2 = m.)
More precisely,
f(x) - f(a) f(b) - f(x)

x-a b-x
and
f(b)-/(a)
m
b - a
The theorem says that either (a) m1 � m � m2 or (b) m2 � m � m1.
7.2 THE DEFINITE INTEGRAL, DEFINED AS A LIMIT OF SAMPLE SUMS
In Section 3.7, we defined the definite integral in terms of area, with areas above the
x-axis counted positively and those below counted negatively. In the preceding
7.2 The Definite Integral, Defined as a Limit of Sample Sums 309
bg(x) dx
section, however, we regarded the integral as the limit of a sum:
n
J a
= Lim
JNJ...,Oi=l
2 g(.Xi) �xi.
Most of the time hereafter, the definite integral will be used in this way, and so we
shall redefine the integral, using the above formula as a definition. For this purpose,
we need to investigate nets, and sums of the type
n
2 g(.X;) �xi.
i=l
Consider first ail increasing continuous function/, on an interval [a, b].
y y
f(b) ------ ---- --- r --

'
f(b) --- --- -----,- -- -
1
I I
I I
I I
x0=a x1
I I IX3 · · · Xn=b X
iI �-II _____ _J
I
�- _J
I
_
I
f(a) I I
I I
�--+-��x f(a)
a=xo xi xz x3 Xn=b · · ·
--- _
.J
(a) (b)
On the interval [a, b] we form a net
N: a = x0 < x1 < · · · < xi-l < Xi < · · · < Xn = b.
The points of the net N cut up the interval [a, b] into little intervals [xi_1, X;]. For
each i from 1 to n,
m; be the minimum value off on the ith interval [x;_1, x;],
let
and let M; be the maximum value. Since f is increasing, we have m; = / (x;_1),
M; = f (x;)· As usual, �x; = X; - X;-i. so that �x; is the length of the ith interval
[x;_1, x;]. Iff is positive, as in part (a) of the figure above, then the sum
n
s(N) = 2 m; �xi
i=l
is the sum of the areas of the inscribed rectangles, and the sum
S(N) = L M; �X;
i=l
is the sum of the areas of the circumscribed rectangles. For functions which may be
s(N) and S(N) are sums of signed areas. In either
negative, as in part (b) of the figure,
case,s(N) is called the lower sum off over the net N, and S(N) is called the upper sum
off over N.
On each interval [xi�1, x;] we choose a sample point X;. Thus
(1 � i � 11) .
The sequence
is called a sample of the net N. The sample gives a sum

n
L (X) L f(xi) 6.xi.

i=l
=
This is called the sample sum off over the sample X.

As in the preceding section, the mesh of N is the largest of the numbers 6.x;.
The mesh is denoted by INI. Thus
INI = max {6.x;}.
Let R be the region between the graph off and the x-axis, from a to b.
Theorem 1. If/is continuous, and N1 and N2 are any nets over [a, b], then
s(N1) � S(N2).
That is, every lower �um off is less than or equal to every upper sum off For
positive functions this is obvious, because in this case s(N1) is the area of an inscribed
polygonal region (lying under the curve) and S(N2) is the area of a circumscribed
polygonal region. Jn general,
where A and D are areas of inscribed regions, and Band C are areas of circumscribed
regions. (To see how this works, see the figure (b) above.) Therefore,
A� C, B ?;;_ D, -B� -D, A - B� C - D,
and s(N1) � S(N2), as before.

Note that in Theorem 1 it is not required thatf be an increasing function.
Theorem 2. Iff is continuous and increasing, then
lim [S(N) - s(N)] = 0.

1s1-o
That is, the upper sums are close to the corresponding lower sums, when the mesh
is small. To prove this, we observe that the difference S(N) - s(N) has a geometric
interpretation.
y y
f(b) --
,
f(b) --------------
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
.-+---�]----�
I I
I I
I I
j(a) --,
I I
I I
This difference is
n n n
S(N) - s(N) = .L Mi �xi - L mi �xi= L (Mi - mi) �xi;

i=l i=l i=l
and this is the sum of the areas of the rectangles drawn solid in the figure. These
rectangles can all be moved to the left and stacked up inside a rectangle of altitude
f(b) - f(a) and base INI. (Remember that INI is the largest of the numbers �xi.)
Therefore
S(N) - s(N) � [ f(b) - J(a)] INI, ·
and so S(N) - s(N) -+ 0 as IN/ -+ 0.
Theorem 3. If f is continuous, and

Jim [S(N) - s(N)] = 0,
IXl->O
then the sample sums .L (X) approach a limit, as INI -+ 0.
That is, there is a number k such that

Jim L (X) = k.
l1Yl->O
Proof We have to start the proof by naming the number k. The numbers s(N) are
bounded above. (By Theorem 1, every upper sum is an upper bound of the lower
sums.) Let
k = sup {s(N)}.
Consider an interval [x;_1, x;]. For each i,
m; � f(i;) � M;.
Since !J.xi > 0, this gives

m; !J.xi�f (xi) !J.xi�M; !J.x i.
Therefore the sums from 1 to n rank in the same order:

n n n
I m; !J.xi�I f(x;) !J.x;�I M; !J.xi,

i=l i=l i=l
so that
s(N)� I (X)� S(N).
That is, every sample sum lies between the lower sum and the upper sum.
We are now almost done. Given E > 0, we want a c5 > 0 such that
INI < 0 => IL (X) - kl < €.

By Theorem 3 there is a o > 0 such that
INI < o => S(N) - s(N) < E.
Thus when INI < o, the interval from s(N) to S(N) has length less than E.
s(N) 2.::(X) k S(N)
This interval contains I (X). ( See the inequalities above.) And it also contains k:
s(N)� k, because k is an upper bound for the lower sums; and k� S(N), because
k is the least upper bound of the lower sums. Therefore IL (X) - kl < E, because
I (X) and k are squeezed together: they both lie on the same short interval.
We can now give the new definition of the integral.
Definition. fa f(x) dx = lim1N1-?0 I (X), if the indicated limit on the right exists.
If the limit exists, then we say that f is integrable on [a, b ].
Theorems 2 and 3 fit together to give:
Theorem 4. If f is continuous and increasing on [a, b], then f is integrable on [a, b ].

Later we shall see that all continuous functions are integrable, whether or not
they are increasing.
Our calculations of definite integrals have been based on the differentiation
formula
D f'f(t) dt = f(x),
where/is continuous. We need to know that this differentiation formula still holds,
under the new definition of the integral. This is the purpose of the following theorem.
Theorem 5 (The betweenness theorem for integrals). If f is integrable on [a, b ], and
m�f(x)�M (a� x� b),

then
m(b - f
a) � !(x) dx �M(b - a).
Proof Let N be any net over [a, b], and let X be any sample of N. Then
m �f (xi) �M for every i.

Therefore
m D.xi �f(xi) D.xi �MD.xi.
Forming the sample sum 'L (X) by addition, we get
n n
'L m D.xi � 'L (X) � 'L M D.xi.

i=l i=l
But
n n n n
'L m D.xi = m 'L D.xi; 'L M D.xi = M 'L D.xi,

i=l i=l i=l i=l
and
n
'L D.xi = b - a.
i=l
(Why?) Therefore
m(b - a) � 'L (X) � M(b - a);
and this holds for every sample sum, over every net N. Therefore the same inequalities
hold for lim 'L (X), and the integral lies between m(b - a) and M(b - a), which
was to be proved.
If you review the proof of the formula
D f'f(t) dt = f(x),
in Section 3.10, you will find that in this proof, all that we needed to know about the
integral was the information conveyed by the betweenness theorem for integrals.
Therefore the differentiation formula continues to hold, under our new definition,
wherever the integrand is continuous. It follows that the fundamental theorem of
integral calculus still holds true.
At the end of this chapter we shall prove that every continuous function is inte
g rable.
*PROBLEM SET 7.2
1. In Theorem 2 it was assumed that the function f is increasing. Does the same scheme
of proof work, for a decreasing function? If so, draw a figure illustrating the proof for
decreasing functions. If not, explain how the scheme breaks down, for the case of a
decreasing function.
2. In Theorem 2 it was assumed that f is both continuous and increasing. Suppose we
assume that f is increasing, but not that f is continuous. What changes (if any) do we
then need to make in
a) the definitions of mi and Mi, b) the definitions of s(N) and S(N), and
c) the proof of Theorem 2?
3. Prove the following:
Theorem A (The mean-value theorem for integrals). If f is continuous on [a, b], then
there is a point x, between a and b, such that
ff (x) dx = j(x)(b - a).
4. Consider the function f defined by the following conditions: /(!) !, /Ci) !, =

=
/Ci) = !; /(x) 0 for every other x on [O, I]. Is this function integrable? Why
=
or why not?
*5. Consider the following function, on [O, I]. If x is irrational, then/(x) 0. If x p/q, = =
in lowest terms, then/(x) I/q. At what points (if any) is this function continuous?
=
Is the function integrable? Why or why not?

6. Given a continuous functiong, on [a, b], and a net N over [a, b]. Show that there is a
sample x of N such that L (X) s� g (x) dx. =
7. In Section 7.1 we showed that for every net N we could choose a sample X in such a
way that the length of the inscribed broken line is equal to the sample sum L (X),
not just approximately but exactly. ls it always possible to choose a sample X' such
that L (X') is exactly equal to the arc length? (Here we are assuming, as usual, that
/' is continuous.)
*8. The following remarks are a very sketchy indication of an amusing proof of an important
theorem which is known to you in a slightly weaker form. Fill in the gaps, and state the
theorem which is proved.
F' f. f is known to be integrable on [a, b], but is not necessarily continuous.
=
n n
L [F(x;) - F(xi_1)] L j(x;) D.xi.

i=l
=
i=l
As INI ___,. 0, L'f=if (x;) D.xi ___,. ?; but Lf=1 [F(xi) - F(xi_1)] was something simple, all
along.
*9. Let/be differentiable on [a, b]. Show that if
j'(a) < k <

f (b � =� (a)
,
then k =j'(x), for some x between a and b.

[Hint: Remember the definition of j'(a). Do a sketch illustrating the definition.]
*IO. Theorem (171e no-jump theorem for derivatives). If f is differentiable on [a, b], and k
is between j'(a) and j'(b), then k j'(x) for some x between a and b.
=
Thus, for example, the function
for x ,e. 0,
for x = 0
cannot be the derivative of any other function f

7.3 The Calculation of Volumes, by the Method of Disks 315
*11. Theorem. Iff is differentiable at a, then
xl��- (f(x;� =��Xi)) =['(a).

x:-a+
More precisely, for every " > 0 there is a c5 > 0 such that
a - c5 < x1 < a < x2 < a + c5
I [Cx2) - f(x1) _ ' a

I
"·
[( )
*
 0 such that for every
x1 and x2 of [a, b],
[f(x1) - [Cx2)i � k [x1 - X2i·

Show that f(x) = sin x is Lipschitzian on the interval ( - ct:>, )
ct:> .
b) Show that every Lipschitzian function is continuous.

c) Give an example to show that a continuous function is not necessarily Lipschitzian.
d) Show that if[' is continuous on [a, b], then/is Lipschitzian on [a, b].
e) Show that if[ is Lipschitzian on [a, b], then/is integrable.
Since Theorem 3 is known, it will be sufficient to show that if/is Lipschitzian,
then
lim [S(N) - s(N)] = 0.

IN/�o
13. a) A function f is uniformly continuous on [ a, b] if for every " > 0 there is a c5 > O
such that
[x - x'[
*
< c5 I/(x) - f(x')[ < "·
(Here x and x' are any points of [a, b].) Show that if[' is continuous, then f is
uniformly continuous on [a, b].
b) Show that every uniformly continuous function is integrable.
7.3 THE CALCULATION OF VOLUMES, BY THE METHOD OF DISKS
The volumes of various solids can be expressed as definite integrals. In this process,
we shall assume that the following volume formulas are known.
I
I
I
I h
I
I
I
) ---- ---- -
,,....-- -----
-- ........,
/
/
b r
/
V =abc. V = 7rh(r2 - s2).

The first of these solids is a rectangular parallelepiped; the second is a right circular
cylinder; and the third is a cylindrical shell, that is, the portion of the larger cylinder
that lies outside the smaller cylinder.
We get a coordinate system in space by setting up a z-axis, perpendicular to the
xy-plane at the origin. Here, and throughout this chapter, we shall indicate only the
positive half of each axis, thus getting a picture of only the "first octant," in which
the points have nonnegative coordinates.
Consider now the function

f(x) = 1/x
on the interval [1, 2]. Let R be the region under the graph, in the xy-plane. We
rotate the region R about the x-axis. This gives a solid S.
y y
� I
x
Let vS be the volume of S. We shall express vS as a definite integral. First we form a

net
over the interval [1, 2]. For convenience, we use equally spaced points, so that
xi - xi-l = �x = 1/n
·
for each i. Over the intervals [xi_1 , xi ] we set up the circumscribed rectangles. These
form a region Rn which is an approximation of the region R. Then we rotate Rn
about the x-axis. Each of our rectangles then gives a cylinder (lying on its side), and
the cylinders form a solid Sn which is an approximation of S. In the figure on the
right below we show only the ith cylinder. Its altitude is �x xi xi_1, and the
= -
radius of its base is

y y
Therefore its volume is
V·
. =
xi-1
.
7T (-1-)2• �x
7Tr2 �x · =
'
and so the total volume of the circumscribed solid Sn is

n
vSn .2 vi .2 7T
n 1 2
= �x. =
(-) ·
i=l i=l X;-1

This is a sample sum of the function
1
g(x) = 7T
x2'
over the net
(In fact, it is the upper sum of g over the net N, because g(xi_1) is the maximum
value of g on the interval [x;_1, X;].) The mesh of N is
INI = �x =
.! ,
- -
n
and so IN! 0 as n co. Therefore
lim vSn
n-+ 00
=
2
{ 7T dx
J 1 x2
= - �Jx 21 = .'.'.:: .
2
(1)
If we use inscribed rectangles, and rotate them about the x-axis, then we get an
inscribed solid S�, with volume
2
n
vS� .2 7T ( 1 ) �x. =
i=l
-
X;
·
This is also a sample sum, of the same function g = 7T/x2• Therefore

2 7T
lim vS� =
n- cc i 1 X
2 dx. (2)
Therefore the volume vS of Sis squeezed between the volumes of the inscribed and
circumscribed solids:
for every n,
and so
2
vs = f '!!__ dx = '!!. . (3)
Ji x2 2
We shall now review this process and state the assumptions on which it is based.
Not all solids are measurable, in the sense that they have volumes; but the solids that
you are likely to encounter soon are measurable, and their volumes are governed by
the following laws.
By an elementary solid we mean a right parallelpiped, cylinder, or cylindrical
shell, as at the beginning of this section. We have been assuming that:
V.1. Elementary solids are measurable, and their volumes are given by the formulas
v = abc, v = 7Tr2h, v = 7Th(r2 - s2).
Two solids are nonoverlapping if they have no solid in common. (They may have
surfaces in common.)
V.2. If s1, s2, . • • , sn are nonoverlapping elementary solids, and Sn is their union,
then Sn is measurable, and
V.3. If S and S' are measurable, and S' lies in S, then vS' � vS.
V.4 (The squeeze principle). If (a) Si. S2, • • • are measurable solids containing S,
(b) S{, S�, ... are measurable solids lying in S, and (c) lim,H00 vSn = L = limn-•ro vS�,
then S is measurable, and
vs= L.
Using V.1 through V.4, we can show that the method of disks, which we have
'
used for the function
f (x) = I/x,
works for every function f which is � 0 and continuous.
y y
�
f
M; ---
-
'mi ---- I I
I I
I I
I I
I I
--lf---��x,�· -1 �-i
.�x ��---x
-
z
Given such a function f, on a closed interval

[a, b], let S be the solid of revolution
[a, b], with equal spacing. As usual, let
of R, about the x-axis. Take a net N over
mi and M; be the minimum and maximum values of f on the ith interval [x;_1, x;] .
If we rotate the inscribed rectangles about the x-axis, we get an inscribed solid S�,
of volume
n
vS� = I 7Tmi!lx.
i=l
If we rotate the circumscribed rectangles about the x-axis, we get a circumscribed

solid Sn, of volume
n
vSn = I 7TMi!lx.
i=l
V.3 says that
vS� � vSn for every n.
But vSn is an upper sum of the function
g(x) = 7Tj(x)2,
and vS� is a lower sum of the same g. As n-. oo, JNI-. O;
vSn _. f7rf(x)2 dx, vs� --f7Tj(x)2 dx.

By the squeeze principle it follows that S is measurable, and
vS = f7Tj(x)2 dx.
We also use this formula sidewise.
y
y
y=vx
Suppose that the region R on the left is rotated about the y-axis. Sidewise, R can be
regarded as the region under the graph of a function
x =f(y) =y2, O�y�l.
1 7T 1 7TY4 dy
Therefore the volume is
7T
1 0
[f(y)]2 dy =
10
=-
5
.
PROBLEM SET 7.3
1. Obviously a right circular cone can be regarded as the solid of revolution of a right
triangle about one of its legs. If we place the triangle in the xy-plane as shown in the
figure, then the hypotenuse becomes the graph of a function f Calculate/, and find the
volume of the cone by the methods of this section.
2. Similarly, a round ball of radius r can be regarded as the solid of revolution of a semi
circular region about its diameter. Find the volume, by the methods of this section.
y y
3. The region under the graph of /(x) = v':X (0 � x � 1) is rotated about the x-axis.
Find the volume of the resulting solid.
4. Same, forf(x) = sinx, 0 � x � .;,,
5. Same, for f(x) = x312, 0 � x � 1.
6. Same, forf(x) = cosx, -1Tj2 � x � 1T/2.
7. Let R = {(x,y) I 0 � x � 1,. x2 � y � 1} be rotated :>.!)out the y-axis. Find the
volume.
8. Same, for R = {(x, y) I 0 � x � 1, Sin-1x � y � 1T/2}.
9. Same, for R = {(x,y) I 0 � x � 1T/2, sinx � y � l}.
10. Same, for R = {(x,y) I 0 � x � 1, x3 � y � 1}.
11. Same, for R = {(x,y) I 0 � x � v'2/2, x � y � v'l - x2}.

7.4 The General Method of Cross Sections, and the Method oJ.' Shells 321
12. Find out whether the following is true:
Theorem (?). Let T and T' be triangles each of which has a side on the x-axis. If T
and T' have the same area, then when they are rotated about the x-axis, they give solids
with the same volume.
13. a) For each.x from 0 to 1, let T,,, be the triangle whose vertices are (0, O), (1, 0), and
(x, 1). What value or values of x give maximum volume, when Tx is rotated about
the x-axis?
b) Suppose that the triangles T,,, are rotated about the y-axis (instead of the x-axis).
Which value or values of x give maximum volume?
14. For each k from 0 to 1, let Tk be the triangle whose vertices are (0, 0), (k, 0), and
(0, Vl - k2). (Thus the hypotenuse of Tk has length 1.) Tk is rotated about the x-axis.
What value of k gives maximum volume? What is the maximum volume?
15. a) Given/(x) = 1/x. Let R be the region under the graph of/, from 1 to ro. Give a
reasonable definition of the area of R. Is this area finite?
b) The region R is rotated about the x-axis, giving a solid S. Give a reasonable definition
of the volume of S. Is this volume finite?
16. a) The region R under the graph of f(x) = 1/x2 from l to ro is rotated about the x-axis
giving a solid S. Does S have finite volume?
b) If the region R is rotated about the y-axis, do we obtain a solid with finite volume?
7.4 THE GENERAL METHOD OF CROSS

SECTIONS, AND THE METHOD OF SHELLS
The method of disks can be generalized in the following way. Given a solid S in
space. Suppose that we can calculate the areas of the cross sections perpendicular to
the x-axis.
y y
'\
\
I 'I
I , ,
I 11
I ,,
I; I
I
I
I
I
I
x
a x b
z
z
For each x from a to b we let A(x) be the area of the cross section. This gives a
function A which expresses the cross-sectional area in terms of x. (In our previous
examples, the cross sections were all circular.) As before, we divide the interval
[a, b] into n equal parts, and we approximate the volume by cylinders. In the figure
at the right, we show only the ith cylinder.
We then have
n
vn = L A(xi) D..x,
i=l
and the sum on the right-hand side is a sample sum of the function A. Therefore,
as the mesh goes to 0,
It is plausible to suppose that as the mesh approaches 0, vn approaches the

volume of S; and in fact this is true, for measurable solids, although we are not in a
position to prove it. Thus
vS = fA(x) dx.
By this method we can calculate volumes. For example, take the parabola
y = x�, for 0 � x � 1. For each y from 0 to 1, we take the horizontal segment
from (O,y) to-the point (x,y) of the parabola; and using this segment as an edge,
we construct a horizontal square. Thus we get a solid, as shown in the figure.
/
/
/
/
The cross-sectional areas perpendicular to the y-axis are given by the formula
A(y) = x2 = y.
V =
11A(y) dy 11y dy [y2] 1
= = - =
1
- .
0 0 2 0 2
The general method of cross sections applies, in a sense, to every volume problem.
That is, it is always true that
vS = fA(x) dx.
But often this formula leads to difficult calculations.
7.4 The General Method of Cross Sections, and the Method of Shells 323
x
,,. ,,.
I
-2 I
I
2
I
I
Consider, for example, the region R under the graph of
f(x) = cos x,
We rotate R about the y-axis, getting a solid of revolution, of which only the front
half is shown in the figure. We can find the volume by the cross-section method. We
have
A(y) = ?TX2 = ?T(Cos-1 y)2•
Therefore
V = J: A(y) dy = J: ?T(Cos-1 y)2 dy.
We can calculate this by integrating by parts twice, but there is a better way.
Instead of approximating the solid by thin cylinders, we approximate it by thin
cylindrical shells.
First we approximate the region R by rectangles with equal bases �x = ?T/2n, as

shown on the left. Then we rotate each of these rectangles about the y-axis, getting
cylindrical shells, as shown on the right. The altitude of the ith shell is
f(xi) = cos xi;
the outer radius is X;; and the inner radius is X;_1 = X; - �x . Therefore the volume
of the ith shell is
V
;
= ?TX� cos xi - ?T(X; - �x)2 cos X;
= ?T(x; - x� + 2x; �x - �x2) cos X;
= 27TX cos xi �x - 7T(COS X;)(�x)2.

Therefore the total volume of the inscribed solid Sn is
n n
= L 27TXi · cos xi · Llx - 7T Llx L cos xi Llx.
1 1
We need to find out what happens as the mesh goes to 0. The first sum is a sample
sum of the function 27TX cos x. Therefore
n
L 27TX; cos X; · Llx -+
i rr/2
27Tx cos x dx.
1 0
For the same reason,

n
L cos X; · Llx-+
i rr/2
cos x dx.
1 0
Therefore
[ r.12
7T Llx · L cos xi · Llx -+ 7T · 0 ·
J cos x dx.
o
Thus the entire second sum, in the expression for vSn, drops out when we pass to the
limit. Therefore
[ "12
vSn -+
J 27TX cos x dx,
o
and
[ "12
V =
J 27TX cos x dx
o = 27T[X sin x + cos x]�12
= 27T [i J + 0 - 277[1] = 7T2 - 27T.
The same method applies if we rotate a region lying to the right of the y-axis.
If the width of the region is given by a function h(x), then the volume of the ith
cylindrical shell is
vi = 7TX7h(x;) - 7T(Xi - Llx)2h(xi)
= 7Th(xi)(x7 - x7 + 2x; Llx - Llx2 )
= 27TX;h(x;) Llx - 7Th(xi) Llx2•
7.4 The General Method of Cross Sections, and the Method of Shells 325
Therefore
n n n
vn = L vi = 27T L X;h(x;) Llx - 7T Llx L h(x;) Llx
1 1 1
Therefore
V = 27T fxh(x) dx.

Here again the second part of the sum drops out when we pass to the limit.
Thus the sums behave as if V; were given by the formula
There is a simple reason why V; is well approximated by 27TX;h(x;) Llx.
2...x;
h(u){ �------------------]
\..,
D.x
If we make a vertical cut in the ith cylindrical shell, and :flatten it out, we get a rec
tangular prism. The length of the prism is the circumference of the outer circle in the
base of the shell. This is 27TX;. The altitude and the thickness of the prism are the
same as the altitude and the thickness of the shell; these are h(xJ and Llx. Therefore
the volume of the prism is exactly 27Txih(x;) Llx; and this ought to be a good approxi
mation to V; when Llx is small, because when the shell is thin, we can flatten it out
without distorting it very much. As we have seen, the error goes to zero as the mesh
goes to zero.
The method of shells applies to the problem that we were discussing above.
We know that the volume is
V =
irr/227TX
0 COS X dx.
Integrating by parts, we get
Jx cos x dx = {x sin x + cos x + C}.
Therefore
V = [27T(X sin x + cos x)]�12 = 7T2 - 27T.
The same method applies more generally. Consider the region
R = {(x,y) I 2 � x � 3, 0 � y � -x2 + 5x - 6}.

R
�-_Ll �� +- � ----'-
� -'-�-'-�.__--X
I
I
If R is rotated about the line x = -1, then by the shell method the volume of the
resulting solid is
f27T(x + 1)(-x� + 5x - 6) dx = f27T(-x3 + 4x2 - x -

6) dx
=
[ x4
27T - 4 +
4xa
3 -
x2
2 -
]a
6x. 2 =
t7T.
If the same region is rotated about the line x = 6, the volume is
f277(6 - x)(-x2 + 5x - 6) dx,
which is also equal to 77T/6. (Why?)

In some cases there is little to choose between the cross-section method and the
shell method. For example, suppose we take the region below the graph of y = x2,
0 ;;; x ;;; 1, and rotate it about the y-axis.
By the shell method,
V f127TX x2dx '!!..

Jo 2
= · =
The horizontal cross section at height y is the region between a circle of radius 1
and a circle of radius x = JY. Therefore
V = fA(y) dy f [7T = · 12 - 7TY] dy
7T
as before.
= 7T - 7T
fly dy
0 2
= - '
7.5 The Area of a Surface of Revolution 327
PROBLEM SET 7.4
1. Let R be the circular region with center at (5, 0) and radius 2. R is rotated abo1•t the
y-axis. Find the volume of the resulting solid.
2. A solid of the sort described in Problem 1 is called a solid torus. More generally, suppose
we have given a circular region of radius a, and a line L in the same plane, such that
the perpendicular distance from L to the center of R is b, with b � a. When R is rotated
about the line L, the result is a solid torus. Find its volume, in terms of a and b.
3. Let R be the square region with center at (4, 0) and sides of length 2, parallel to the
coordinate axes. R is rotated about the y-axis. Find the volume of the resulting solid.
4. Let T be the square region with center at (4, 0) and sides of length 2, with diagonals
parallel to the coordinate axes. Find the volume of the solid which results when T is
rotated about the y-axis.
5. a) The region under the graph of
y = lnx, 1 �x � e,
is rotated about the x-axis. Find the volume, by the method of disks.
b) Now solve this problem by the method of shells.
6. For eachx from 0 to 1, let Rx be the circular region perpendicular to thexy-plane, with
center at the point (x, x2) and radius 1. Let S be the solid formed by the regions Rx.
Find the volume of S.
7. a) The region described in Problem 5a is rotated about the y-axis. Find the volume, by
the shell method.
b) Now solve Problem 7a by the method of cross sections.
8. a) The region under the graph of y = e", 0 � x � 1, is rotated about the y-axis. Find
the volume by the method of shells.
b) Now solve Problem Sa by the cross-section method.
9. Let C be the cylinder with the y-axis as its axis of symmetry, and radius 1. Let S be the
sphere with center at the origin and radius 2. Find the volume of the solid which lies
inside the sphere and outside the cylinder.
10. Let C,, be the cylinder of radius 1, with the x-axis as its axis of symmetry; and let Cy
be the cylinder of radius 1 with the y-axis as its axis of symmetry. Find the volume of the
solid which lies in both Cx and Cy.
11. Let S be the sphere of radius v2 with center at the origin. Let C be the cone with vertex
at the origin, axis along the y-axis, and passing throug� the point (1, I). Find the
volume of the solid which lies inside the sphere and inside the cone.
7.5 THE AREA OF A SURFACE OF REVOLUTION
Given a line and a curve, lying in the same plane and lying on one side of the given
line. If the curve is rotated about the line, the resulting surface is called a surface of
revolution. The area of such a surface can be expressed as an integral. We begin with
the simplest case, in which a function-graph is rotated about the y-axis. Here the
functionfis defined on a closed interval [a, b] on the positive half of the x-axis. We
assume that f has a continuous derivative.
.. x
To calculate the area of the surface of revolution, we need the formula for the
lateral surface of a right circular cone. Let s be the slant height of the cone, and let
r be the radius of the base, so that the circumference of the base is 2Trr. We assert
that the lateral surface is the same as the area of a circular sector of radius s, with
boundary arc of length 2Trr. The reason is that we can make a straight cut in the cone,
starting at the vertex, and then flatten out the surface, without changing its area, so
that the resulting surface lies in a plane. The plane surface thus obtained is the sector
shown below.
But the area of a circular sector is half the product of its radius and the length of its
boundary arc. Therefore, for cones, we have
A = 'TrYS.
(Note that for a "cone of altitude O," that is, a disk, this formula gives the right
answer Trrs = Trr2. ) From this we can get a formula for the lateral area of a frustum
of a cone. If the larger cone (with slant height s2) has area A2, and the smaller cone
has area Ai, then the area of the frustum is

AA= A2 - Ai= 7Tr2s2 - 7Tris1•
Evidently
If Si = kri. s2= kr2, then
AA= 1Tkr� - 1Tkri= 1Tk(r2 - ri)(r2 + ri)

ri + r2 A
= 7T(S2 - s1)(r2 + r1) = 21T · --- us,
2
and so
AA= 2Ttf As,
where
i(r1 + r2).
f=
That is, the area of the frustum is equal to its "average circumference" 2Ttf times its
slant height D..s.
Consider now the surface of revolution obtained by rotating the graph off
about the y-axis. We take a net
N: x0, Xi, . • . , xi-l• xi, . . . Xn,

over the interval [a, b], with equal subdivisions, so that
b - a
Ax= --= INI
xi - xi-l =
n
for each i. For each i, let Pi be the point (xi, f (xi)) . These points determine a broken
line Bn which is an approximation of the graph off When Bn is rotated about the
y-axis, we get a surface Sn, with area An. By definition, the area of the surface of
revolution off is
A= lim An,
INl->O
if the limit exists. (This is like the definition of arc length.)
We shall now calculate An, and find its limit as INI -+ 0. Consider the ith segment,
from Pi-l to Pi. When this segment is rotated, it gives a frustum whose area is
ai= 21Txi · Pi-lpi;
.. x
Xi-1 Xi Xi
As in the calculation of arc length,
Pi-1Pi = .JLix2 + [f(xi) - f(x;_1) ]2
J f xi) xi )
= 1 + ( c �:c -l Y Llx
= .J1 + f'(i;)2Lix,
where X;_1 < i; < xi, as shown in the last figure.
We now have a formula for the area A n of the approximating surface:
An = I ai iI=l 27TiiPi-1Pi
i=l
= =
i=l
I 27Tii.J1 + f'(i;)2Lix.
Here i i is the midpoint of [xi_1, x i], and i; is somewhere on the same interval.
If it were true that i; = i i for each i, then A would be a sample sum of the function
n
g(x) = 27Tx.JI + f'(x)2,

and we would have no problem, because for
n n
I a; = I 27Ti;.J1 + f'(i;)2Lix,
i=l i=l
we know that
n ibg(x) dx ib27Tx.J1
lim I a; = = + f'(x)2 dx.
n�oo i=l a a
What we need to show, therefore, is that
It will then follow that

n n ib27Tx.J1
lim I ai = lim I a; = + f'(x)2 dx.
I NI ->O i=l INI ->O i=l a
Now
<Lix
xi __
I- x i,1 '
=
2
because i; lies on the interval [xi-1• ] whose midpoint is i i. Therefore
X; ,
lai - a;J = l27Tii.J1 + f'(i;)2Lix - 27Ti;.J1 + f'(i;)2Lixl

= 27T.J1 + f'(i;)2Lix Ii; - i;I
� 7T.J1 + f'(i;)2Lix2•
Therefore
I I �;�Ia;
n n n n
�a; - ;�a; - a;I �i� '"-J1 + f'(x;)2 .6.x2

;
= .6.x I '"-J1 + f'(x;)2 .6.x.
i=l
Now
b
lim .6.x i Tr-J1 + f'(x;)2 .6.x = 0 r Tr-J1 + f'(x)2 dx.
Ja
.
INJ-+O i=l
Therefore II a; - I a;I is squeezed to 0, which was to be proved.

Let us try this formula on some problems to which we already know the answers.
y
b
z
z
A cone is the surface of revolution of a segment. Here
f(x) = b - (b/a)x, f'(x) = -b/a,

and
-
a b2
A
J
r
J
= o 2Trx 1 + dx
a2
= 2Tr -Ja2 + b2 rax dx
a Jo
= Tra-Ja2 + b2,
which is the right answer. Consider next a quadrant of a circle of radius a, rotated
about the y-axis. In this case,
f(x) = -Ja2 - x2 (0 � x �a),

-x
f'(x) = I '
'\/ a2 - x2
x2 a2
1 + f'(x)2 = 1 + =
a 2 - x2 a2 - x2
---
•
Therefore the area is
It follows that the total area of a sphere of radius a is 47Ta2• This is the standard
formula.
It is harder to find the area when we rotate a function-graph about the x-axis
instead of the y-axis.
Given a nonnegative function f, on an interval [a, b]. As before, take a net N

over [a, b], with equal subdivisions, so that for each i,
b a
INI.
-
xi - xi-1 = D..x = -- =
As before, we approximate the graph by a broken line Bn- Then we rotate Bn about
the x-axis, getting a surface Sn, with area Aw We define the area of the surface of
revolution to be limlNl--?O An, if such a limit exists. We proceed to calculate:
where a; is the area of the ith frustum, shown in the figure. As before,
P;_1P; = -J 1 + f'(x�)2 D..x,

where
X;_1 < X� < Xi.
But when we rotate the chord from P ;_1 to P; about the x-axis, the "average circum
ference" is
Obviously i'; is between/(xi_1) andf(x;), because i'; is their average. By the no-jump
theorem of Section 5.7,
Therefore
An = I a; I 27Tj(.X;)-J1
i=l
=
i=l
+ f'(.X�)2 D..x.
If it were true that xi= x� for each i, then the sum on the right-hand side would be a
sample sum of the function
g(x)= 2rrf(x)J1 + j'(x)2.
As it stands, it is very close to being a sample sum. The idea is that
INI= �x � 0
for each i
for each i
i 2rrl(xi)-./1
i=l
+ f'(x;)2 �x � i 2rrl(x;)J1
i=l
+ f'(x;)2 �x
� f2rrl(x)-./1 + f'(x)2 dx.
At the end of the chapter, these ideas will be turned into a proof. Meanwhile let us
look at some applications of the formula
A= f2rrl(x)J1 + f'(x)2 dx.
1) If f(x) = k, on [a, b], then the surface of revolution is a cylinder. By the

integral formula,
A= f2rrkJT+02 dx= 2rrk(b - a , )

2) A sphere of radius a is the surface of revolution of a semicircle of radius a.
Here
l(x)= Ja2 - x2,

Hence
x
2x + 2lf'= 0, and l'=
l
Therefore
2 1 2 + x2 �2
1 + !'2= 1 + �= =
12 12 1 2'
and
A= f-aa 2rrl(x)J7
l(x)2
a
-- dx= f 2rra dx= 4rra2•
-a
PROBLEM SET 7.5
1. Let Ca be the circle with center at the origin and radius

lying above the interval
a, and let A be the arc of Ca
[ -a/2, a/2]. A is rotated about the x-axis. Find the area of the
resulting surface. What proportion is this, of the total area of the sphere?
2. The entire circle Ca is rotated about the x-axis, giving a sphere of radius a. Eb and Ee
are two planes, perpendicular to the x-axis, at x = b and x = c; and S is the part of
the sphere that lies between them. Find the area of S, in terms of a, b, and c. The
form of your answer ought to suggest a somewhat surprising theorem which can be
stated without the use of formulas. What is the theorem?
3. The circle with center at (b, 0) and radius a, a < b, is rotated about the y-axis. The
resulting surface is called a torus. Find its area.
4. The square with corners at the points (a, O), (a+k, k), (a+k, -k), and (a + 2k, 0)
is rotated about the y-axis. (Here 0 < k < a.) Find the area of the resulting surface.
5. Find the volume of the solid obtained when the corresponding square region is rotated
about the y-axis.
6. The same square is rotated about the line x = a+2k. Find the surface area.
7. The square region is rotated about the line x = a + 2k. Find the volume.
8. The square with center at (a, 0) and sides of length 2k parallel to the coordinate axes is
rotated about the line x = 2a. Find the area of the resulting surface. (Here 0 < k < a.)
9. Find the volume when the corresponding square region is rotated about the Jine y = k/2.
10. Consider the curve consisting of (a) the segment from (0, 0) to (a, 0), (b) the segment
from (0, 1) to(a, 1), and (c) the semicircle, pointing outward, with endpoints at (a, 0)
and (a, 1). This curve is to be rotated about the y-axis. For what value of a is it true
that the total area of the resulting surface is equal to 15?
11. For each a, let Sa be the area of the surface described in Problem 10, and let Va be the
volume of the solid that it encloses. What value of a maximizes the ratio VafSa?
12. The circle with center at (b, b) and radius a is rotated about the line
x+y=l .
Here a and b are both positive, and the Circle does not intersect the line. Find (a) the
area of the resulting surface, and (b) the volume of the solid that it encloses.
13. Same question, for the circle with center at (2, 10) and radius I and the line x+y = 2.
(The only natural solutions of this, on the basis of the theory that we have so far, are
rather clumsy. This suggests that some new ideas are needed.)
14. The graph of
y = 2x2,
from x = -1 to x = 1, is rotated about the line x = 5. Find the area of the resulting
surface.
15. If the same surface is rotated about the line x = 4, would the area of the resulting
surface be greater, or would it. be less, than the answer to Problem 14? Get a plausible
answer to this, and justify it as well as you can.
16. The graph of y = t(e"' + e-"'), 0 � x � 1, is rotated about the x-axis. Find the area of
the resulting surface.
17. The same graph is rotated about the y-axis. Find the area of the resulting surface.
18. Let G be the graph off (x)=sin x, from x = 0 to x = n/2. G is first rotated about the
line x+y = 4, and then about the line x +y = 5. Which of the resulting surfaces
has the larger area? Why? (A right answer, with a plausibility argument, is acceptable
7.6 Moments and Centroids. The Theorems of Pappus 335
as an answer to this one. It is possible, however, to give a proof of the right answer,
without calculating the area of either of the surfaces. That is, you can prove an in
equality of the form A < B, without calculating either A or B.)
7.6 MOMENTS AND CENTROIDS. THE THEOREMS OF PAPPUS
The ideas in this section are mathematical descriptions of physical ideas. Given a
finite set of "point masses" m;, at the points P; = (x;, y;) in a coordinate plane, the
moment (of the system) about the y-axis is
n
My= .L X;m;.
i=l
The left-hand figure below shows the general case.
P2 •
-2 -1
----�
--+-----____,� : --r---'---+--�---_...x
I
I
-1
• P3 · · . Pn I
I
I
m2=1�------ -2
In the example on the rig)1t, we have:
My = x1m1 + x2m2 = I · 2 + (-2) · I = 0.
Physically speaking, this means that if the plane is horizontal, resting on a knife-edge
along the y-axis, it will balance. The formula .L mix; for M v makes it plain that the
effect of each point mass depends only on the product m;X;; if we divide m; by 2,
and double xi, then the moment Mv is unchanged.
Similarly, the moment about the x-axis, of our finite system of point masses, is
defined to be
n
M
x =
L Yimi.
i=l
The total mass of alt the particles in the system is denoted by m. That is,
n
m = .L m;.
i=l
The centroid of the system is defined to be the point
P = (x, .Y)
such that
and Mx = ym.
Thus if we concentrate the entire mass of the system at P, the moments about the
x-axis and y-axis are unchanged.
For example, if we have m1 = 2 at P1 = (1, 2) and m2 = 3 at P2 = (2, 5), then
My= 2 + 6 = 8, MllJ = 4 + 15 = 19,
m = Lmi = 5,
8 = x. 5, 19 = ji . 5,
x = t. ji=1_/.
The above discussion does not prove that My, Mx, and P = (x, ji) have any
physical significance; only experiments can prove this. The fact, however, is that the
physical conditions for equilibrium are described by moments and centroids.
Let us now consider how these ideas can be applied to a region Rin the xy-plane.
We shall think of Ras a very thin sheet of homogeneous material, so that the mass per
unit area is constant, say, = I.
Suppose that we take a net over the interval [a, b], as in the figure; for each x ,
we let h(x) be the height of the cross section of Rat x, and we let
We use equal subdivisions, so that
xi - xi-l = b.xi = !:ix, for each i.

Then
is the area of the rectangle in the figure. The rectangle is narrow, and so its moment
about the y-axis should be approximately
If we approximate the region R by a finite set of such narrow rectangles, then the
moment of R about the y-axis ought to be approximately
n n
L xA !:ix = L xi h(x ) !:ix,

i=l i=l
i
and the approximation ought to get better as the mesh ,6.x decreases. This is the idea
of the following definition.
Definition. Let R be the region lying between the graphs of two continuous functions
/1 and/2, on an interval [a, b], with/1 � /2, and let
h(x)=fix) - /1(x).
Then the moment of R about the y-axis is
M,"= fxh(x) dx.

y y
The definition of M., is similar. Here (see the right-hand figure)
R= {(x,y) Jc� y � dg1( y) � x � g2(y)},

and
w (y)=g2(y) - gi(y),
and by definition,
M., = fyw(y) dy.

Since the total mass of R is its area
f
A= h(x) dx= w(y) dy, f
it is natural to define the centroid of R as the point P= (.X,ji) such that
Mv=.XA, Mx=jiA.
For example, consider a quadrant of a circle of radius a.
y=Va2-x2, O�x�a.
Here
Obviously
and so
Therefore
- 4
x =-a.
37T
By symmetry, interchanging x and y, we get:
- - 4
y = x = -a.
37T
The moment about the line x = x0 is defined to be
Mx=xo = f(x - x0)h(x)dx,

and the moment about the line y = y0 is
Mv=Yo = f(y - Yo)w(y)dy.

It is now easy to see that�
Theorem 1. Mx=:r = 0 = My=g·
Proof For any x0,
Mx=x =
o
fcx - x0)h(x)dx fxh(x)dx - x0fh(x)dx =My - x0A.
=
This is 0 for x0 = x. The proof of the other half of the theorem is the same. In fact,
the equation
M"'="o = My - x0A
shows that the converse of Theorem I is also true.
Theorem 2. If MX=Xo = 0, then Xo = x; and if MY=Yo = 0, then Yo =

ji.
Centroids are easy to find for regions which are symmetric, in the sense now to
be defined. Two points P and P' are symmetric across the line L if L is the perpen
dicular bisector of the segment between them. In the left-hand figure below, we say
that P' is the point symmetrically across L from P. A figure is symmetric about a line
L if for each point P of the figure, P' also lies in the figure. The figure may be either a
region or a curve. For example, a circle is symmetric about any line through its
center, and the interior of a circle has the same property.
p p
L
pt
P'
It is easy to see that if R is symmetric about the y-axis, then x = 0. In the figure
on the left, h(x) is an even function, with h(-x) = h(x). Therefore xh(x) is an odd
function, with (-x)h(-x) = - [xh(x)]. Therefore
M11 = faxh(x) = 0,
and x = 0.
y y
�--+-�-'-�x "'----'-�-'--�..L.._�
x0-k x0-t x0 x0+t x0+k
More generally, as in the right-hand figure, we have:
Theorem 3. If R is symmetric about the line x = x0, then x = x0.

Proof
rxo+k r xo+k
Mx�xo (x - X0)h(x) dx (x) dx.
J xo-k Jco-k efi
= =
By symmetry,
h(x0 - t) = h(x0 + t)
for every t; and so
efi(x0 - t) = [(x0 - t) - x0]h(x0 - t) = -th(x0 - t)
= -efi(xo + t).
Therefore the graph of efi must be like the graph shown below .
xo+k x
Therefore
r xo cp(x) dx '"o+k cp(x) dx,
Jxo-k Jxo
= -
and
xo+k
Mx=xo ixo-k cp(x) dx
= = 0.
It follows that x0 = x.
In this proof, all that we have used is the assumption that
h(x0 - t) = h(x0 + t).

This condition may hold for regions which are not symmetric, as below.
And interchanging x and y, we get the following theorem.
Theorem 4. If R is symmetric about the line y = y0, then ji = y0•

These ideas have the following geometric consequence:
Theorem 5 (Pappus' theorem, for volumes). If a region is rotated about a line not
intersecting it, then the volume of the resulting solid is equal to the area of the region
times the circumference of the circle described by the centroid.
That is, if the region below is rotated about the y-axis, then
V= 27T.XA.
Proof By the method of shells,
V = f27Txh(x) dx.
Therefore
V= 27TMv = 27T.XA,
because x was defined by the equation
Mv = xA.
Pappus' theorem can be applied in two ways. If we know x and A, we can
compute V = 27TxA; and if we know V and A, we can solve for x = V/2.;.A. For
example, consider a circular region R, of radius a, with center at the point (b, 0),
b > a. When R is rotated about the y-axis, we get a solid which is called a solid torus.
(The surface of the solid is called a torus.) By Pappus' theorem, we get
V = 27Tb · 7TG2 = 2TT2a2b.

y
y a
I
I
I
I
I
I
-�-+-�+----+�-- x
a pt=?
I
I
I
I
-a
We can use the theorem in reverse to find the centroid of a semicircular region.
If the region is rotated about the y-axis, we get a sphere of radius a, with volume
V = t7Ta3.
Obviously
Therefore
and
- 4
x =-·a.
37T
These ideas apply also to arcs. We shall think of an arc as a thin homogeneous
wire whose mass per unit length is constant, say, = 1. Suppose that the arc is the
graph of a function f, on an interval [a, b ]. As usual, we take a net over [a, b], with
equal subdivisions. The arc length over the interval [x;_1, X;] is
si f�/1
= + f'(x)2 dx.
y
Now
si R::! -./1 +f'(x;)2 f.,,x;

the moment of this little arc about the y-axis ought to be approximately x,si; and so
the moment of the whole graph ought to be
Mv R::! I xi-./1
i�l
+f'(x;)2 t>,.x.
Definitions. Given the function f, with a continuous derivative f', on [a, b], the
moment of the graph about the y-axis is
M,11 =
fx-./1 +f'(x)2dx;
and the moment about the line x = x0 is
M,x=xo =
fcx - x0)-./l +f'(x)2dx.
Similarly, we state the following:
Definitions. The moment of the graph off about the x-axis is
Mx =
ff(x)-./1 +f'(x)2dx,
Mv=110 = f(f(x) - Yo)-./1 +f'(x)2dx;
and the centroid of the graph is the point P = (x, ji) for which
M :XL ,
v
=
where L is the total arc length.
Our previous theorems for regions now have analogous forms for arcs, as follows.
Theorem 6. If the graph off is symmetric about a Ii ne x = x0 (or y = y0) then this
line contains the centroid.
Theorem 7. If the graph off is rotated about a horizontal or vertical line not inter
secting the graph, then the area of the resulting surface of revolution is equal to the
length of the arc times the circumference of the circle described by the centroid.
r-1
I
I
I
I
For example, if we rotate about the y-axis, then the area of the resulting surface is
S = f 27Tx�l + f'(x)2 dx
= 27TM" = 27TxL,
by definition of x. The proof of the theorem in the other cases is similar.
Throughout this section, we have used a fixed coordinate system to define and
investigate moments and centroids of regions and arcs. It is a fact, however, that
moments and centroids do not depend on the choice of a coordinate system; they
depend only on the regions and the arcs. ln particular, any line of symmetry (horizon
tal, vertical, or sloping) must contain the centroid. You may use this fact in the
problem set below.
PROBLEM SET 7.6
I. Let A, B, and C be the points (0, 0), (a, 0), and (b, c), a, c > 0. At each of these points
there is a particle of mass 1. Find the centroid of the resulting system.
2. Suppose at the points A, B, and C of Problem I there are particles of ma ss 1, 2, and 3

respectively. Find the centroid of the resulting system.
3. A median of a triangle is a segment between a vertex and the midpoint of the opposite
side. Show that every median of the triangle described in P roblem 1 passes through the
centroid.
4. Now consider the triangular region R determined by the same points A, B, and C.
Find the centroid of R.
5. The region R is rotated about the x-axis. Find the volume.
6. R is rotated about the y-axis. Find the volume.
7. The figure formed by the sloping sides of the triangle is rotated about the x-axis. Find
the area of the resulting surface.
8. Same question, for rotation about the y-axis, assuming b � 0.

9. A trapezoid has vertices (0, 0), (a, c), (b - a, c , ) (b, 0) (with b > a > 0 and c > 0).
Find the centroid of the region T bounded by this trapezoid.
10. The region Tis rotated about the x-axis. F!nd the volume.
11. The region Tis rotated about the y-axis. Find the volume.
12. The figure formed by the four sides of the trapezoid is rotated about the y-axis. Find the
surface area.
13. The circle with center at (b, 0) and radius a, with 0 < a < b, is rotated about the
y-axis. Find the area of the resulting torus.
14. Let the arc A be the portion of the circle with center at the origin and radius a which
lies in the first quadrant. Find the centroid of A.
15. The square with corners at (a, 0), (a + k, k), (a + k, -k), (a + 2k, 0) is rotated
about the y-axis. (Here 0 < k < a.) Find the area of the resulting surface.
16. Find the volume of the solid obtained if the corresponding square region is rotated.
17. The same square is rotated about the line x = a + 1k. Find the surface area.
18. The square region is rotated about the line x = a + 2k. Find the volume.
19. Consider the curve consisting of: (a) the segment from (0, 0) to (a, O); (b) the segment
from (0, I) to (a, I); and (c) a semicircle, pointing outward, with endpoints at (a, 0)
and (a, I). This curve is to be rotated about the y-axis. For what value of a is it true
that the total area of the resulting surface is equal to 15?
20. For each a let Sa be the area of the surface described in Problem 19, and let Va be the
volume of the solid that it encloses. What value of a maximizes the ratio VafSa?
7.7 IMPROPER INTEGRALS
The definite integral is defined as a limit of sample sums as the mesh of the net
approaches 0. This limit exists if the integrand f is continuous. But this definition
of the integral does not apply to the function
1
f(x)
Jx
=
on the half-open interval from 0 to 1.
It really is the half-open interval
(0, l] = {x j 0 < x � I}
that we are dealing with, because at x = 0 the function is not defined. On this
half-open interval the function is unbounded. Therefore, for every net over (0, l]
we can form a sample sum as large as we please, by taking the first sample point x�
close to 0. Thus,
the sample sum is large when x� is small, and so the sample sums do not approach a
limit as the mesh approaches 0.
Nevertheless, we can extend the definition of the integral in such a way that our
problem has an answer. The function f (x) = 1/J� is defined and continuous on
every closed interval [a, l], where a > 0. Therefore H (1/J�) dx is well defined.
7.7 Improper Integrals 345
f(x) 1
-
Vx
=
We define a new kind of integral by saying that
f
f 7x
=
!��+ �; '
if the indicated limit on the right exists. (We write a-+ O+, because a takes on only
positive values.) In the present case, the limit exists and is finite:
f �; \-112 dx [2x112J! 2 - 2.Ja.

L
= = =
f1 d
Therefore
lim [2 2.
a-+O J a '\/� a-+O+
lim
+
=
- 2.fa] =
There are similar-looking problems for which the limit is infinite. For example,
11 dx . 1-
dx
0 2 a->0+ 1a 2 '
-= hm
X X
if the limit exists. In this case,
1
-1 + - (a > 0),
a
r 1 dx
and so
a->0+ Ja x2
= oo
lim ,
in the sense defined in Section 5.3.

We abbreviate this by writing
f1 dx
Jo x2
= oo
.
The same test can be applied at any point where the function "blows up," as long
as there is only one such point, at an endpoint of the interval. For example, consider
/2sec x dx = sec x dx,

l"0 a.. rr 2 la0
lim
/
_
where the minus sign means that a-+ 7Tj2 through values less than 7T/2.
--+---rr,,.__----• X
2
Now
lasec xdx = [In !sec x + tan xJ]g

= In I sec a + tan aI - In 11 + 0I
= In I sec a + tan a j.
As a-+ "TT/2-,
sm
sec a= tan a
1 oo a oo.
-- -+ and = -+
cos a cos a
Therefore In [sec a + tan a] -+

oo
, and
a
a -"12- xdx =
Jim r sec
Jo
00.
We use the same method to define and evaluate such integrals as
ico dx2
x 1
•
Here the integration is supposed to be carried out all the way to the right, starting
at x = 1. Again our definition of the definite integral (as the limit of the sample sums)
does not apply, and so we define the improper integral as a limit:
lco dx = .m ladx •
2 h 2
1
a-coX 1 X
if the limit on the right exists. For the function
ladx
g(a) = 2'
the limit exists and is finite:
x 1
j'adx2 = [-.!]a=_ l + 1;
and so
xi x a i
lim r : = lim [ 1 .!] =

ad l.
-+oo
a x a-+oo a
J1
-
Here we have a second example of an unbounded figure with a finite area.
1
f(x) 2
x
=
For very similar-looking functions, the integral from 1 to oo may be infinite.

For example,
f 00 dx fa dx
1
=
Jim =
Jim [In x]� = lim ln a = oo
.
1 X a--+oo X a--+oo a--+oo
Integrals of the type that we have been discussing here are called improper
integrals, of the first and second kind respectively. The two kinds can occur in
combination. For example,
r oo dx
Jo ,jx(l + x)
is improper in two ways: the function blows up at the lower limit, and also the upper
limit is oo. Thus, for the integral to be finite, the two limits
.
hm
1
1 dx 1.
1m
fb dx
a-+o+ a,Jx(l + x) .jx(l + x)
_
,
b-+oo i
must both be finite; if they are, then the integral from 0 to oo is their sum. (Any
positive number k would have done equally well in place of 1.) In this case, both
limits are finite.
Improper integrals may appear in a disguised form, and so we need to be careful.
For example, a careless calculation gives
l d: [-1]1
J-1 x x -1 = = -1 - 1 = -2.
This is impossible, because the integrand is positive. The trouble here is that the
function blows up at x = 0, and so we need to make separate investigations of the
integrals
o dx
J
-1x 2 '
We find that both these limits are equal to oo. Therefore the original integral is equal
to oo.
Note that when we got the "answer" -2 we had no right to complain that the
theory was wrong, because the Fundamental Theorem of Integral Calculus does not
apply to functions whose domains have holes in them; the theorem applies only
to functions which are defined and continuous on the interval over which we are
integrating.
An even worse example of the same kind is
f-1 x
1 dx
= [In lxl]:.1 = ln 1 - ln 1 = 0 (?).
Splitting this into two parts, we get

o
r a
x
= li�[ln lxl]'.:1 = li� ln lal = -oo,
J-1 X a-+O a-+O
1 dx
( =lim[ln lxlJ! =lim[-lna] = oo.
Jo X a-+O
+
a-+o
+
The limits - oo and oo do not combine to give a well-defined limit, either finite or
infinite, and therefore the original integral is not defined at all. We also get no
answer for
"'
sin x dx. l
Here
l a
sin x dx =[-cos x]g = -cos a + 1,
which oscillates forever between 0 and 2, and therefore does not approach a limit.
Thus, when we investigate an improper integral, there are three situations that
we may encounter.
1) The integral may exist, as a finite limit. For example,
(
J1
"'
d
:
X
=Jim
a-+00 J1
a
r a:
X
= lim
a-+oo
[- l] X
a
1
= l.
2) The integral may exist, as an 'infinite limit oo or - oo. For example,
i"' dx
- =Jim l a
dx
- = lim lna =
1
oo.
X a-+ oo 1 X a-+ oo
3) Finally, the limit may not exist as ·

any limit, finite or infinite. This is what we
find for
(?) 1 dx
J -1 x
.
In the following problem set, when you are asked to "investigate" an improper
integral, you should find out which of the above three cases it represents. If it is
Case 1, you should find the limit, unless the contrary is stated.
PROBLEM SET 7.7
loo dx loo dx loo dx

Investigate:
1.
x2 2.
1 x2 x 3.
1 xs
oo dx f dx loo dx
1 + +
J x3
1
4. 5. 6.
-v:x -v:x
loo dx loo dx loo dx
o o 1
1 x1.0001 1 x0.9999 xlnx

f2 dx
7. 8. 9.
f dx f dx1 0001
10.
1x x 11.
0 x. 12.
0 xo.9999
loo
In
2 (x - l)s dx 1 (x - l)s dx fo00e-xdx

13.
1
14.
f 1
15.
J000 xe-xdx Loox2e-xdx Joo dx

16. 17. 18.
x)
loo xdx
0 v'x(l +
19.
1 x4 1 +
20. Show that f;' xn dx is never finite, for any value of n, positive, negative, or zero. (The
l/x, x
point is that something always goes wrong, either at 0 or at oo.)
21. Consider the graph of/(x) = 1 � < oo. Let R be the region under the graph;
let S be the solid of revolution (about the x-axis); and let Tbe the surface of revolution.
Investigate the improper integrals which represent (a) the area of R, (b) the volume of S,
and (c) the area of T.
Investigate the following for existence. (That is, find out whether the integral represents
a finite limit; but in the cases where it does, you need not calculate the liP)-it.)
loo-1--.,
dx dx loo dx Joo dx
22.
1 e +
23.
2 x x
1 + In
24.
o 1 + v' x(l x)+
loo--dx
x2e-x f-00oo e-x• dx f;" e-x• dx f=� e-"'2
x
(To show that this is finite, it will be suffi-
25. 26.
1 1 + cient to show that < oo. It will
dx
then follow by symmetry that
e-x•
< oo. And obviously
because is continuous on
f:_1 e-x•-dx1,
[
x
< oo,
1].)
J00 e-x•dx loo loo dx

x4 dx
1
27.
-00
28.
1 1 +
29.
2 v'xinx
J00 e-x• dx i00 G - x) dx /2tanxdx
30.
-00/2
31. Tan-1 32.
f0
i00 x2 xi dx Joo --dx xi
f0 xdx x2
!sin
33. csc 34. lsin 35.
Joo --dx x
1T
x2
sin
*36.
..
**37. a) Show that for each n,
IJv';;;(n+iJ.-. I 'Jv;;;
sm x2 dx < __ sin x2 dx .
I
f 00 v
v' (n-1>.-
b) Investigate sin x2 dx.

v'-
..
38. Let f be a decreasing function, with a continuous derivative, on the interval [a, oo ) .
The graph (and the region under it) are rotated about the x-axis. Show that if the
surface area is finite, then so also is the volume.
*7.8 THE INTEGRABILITY OF CONTINUOUS FUNCTIONS
Let/be a function which is continuous on an interval l. The function/is continuous

if for each x, and each E > 0, the graph has an Ec5-box at the point (x,f(x)). The
total height of an Ec5-box is h = 2E. The box is then called an h-box of every point of
the graph that lies in it.
The definition of continuity applies to the points x of the interval, one at a time.
It may appear, therefore, that iff is continuous at each point x of the interval, then we
have to use infinitely many boxes (one for each x) in order to exhibit the fact. But
if I is a closed interval, this is not so:
Theorem 1 (The finite covering theorem). Let f be a continuous function on the

closed interval [a, b], and let h be a positive number. Then there is a finite collection
of h-boxes, covering the entire graph off
y
y
~
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I I
I I I I
Proof Let [c, d] be a subinterval of [a, b]. If there is a finite collection of h-boxes,
covering the part of the graph determined by [c, d], then we shall say that [c, d] is good.
7.8 The Integrability of Continuous Functions 351
If no such finite collection exists, then we say that [c, d] is bad. We allow the case in
which [c, d] is all of [a, b]. Thus what we need to show is that [a, b] is good. Suppose,
then, that [a, b] is bad. We shall show that this leads to a contradiction.
Let [au b1] = [a, b]. If the left-hand half of [a1, b1] and the right-hand half of
[a1, b1] are both good, then it follows that [a1, b1] is good; we can fit together two
finite collections of boxes, getting another finite collection of boxes that covers the
whole graph. Therefore one or both of the halves of [a1, b1] must be bad. Let [a2, b2]
be a bad half of [a1, bi]. Similarly, one of the halves of [a2, b2] must be bad. Let
[a3, b3] be a bad half of [a2, b2].
Proceeding to infinity in this way, we get a nested sequence
[a1, b1J. [a2, b2], ..., [an, bn], . ..
of closed intervals, each of which is bad. By the nested interval postulate, there is a
number x which lies on all of these intervals.
Now f is continuous at .X. Therefore /has an h-box at (x,j(x)). Suppose that the
box is {(x, y) I x0 < x < x1, Yo < y < y1}. Since
b - a
bn - an = --1- '
2n-
the length of the nth interval [a,,, b,,] approaches 0. Therefore we must have
for some n.
This means that [an, bnl is good after all: one h-box covers the part of the graph that
lies above it; and 1 is finite.
We continue now at the point where we left off in Section 7.2. There we defined
net, mesh (of a net), upper sum S(N), lower sum s(N), and sample sum I (X). Theorem
3 of Section 7.2 was as follows:
Theorem. If f is continuous, and
lim [S(N) - s(N)] = 0,

[N[-+O
then the sample sums approach a limit, as INI ->- 0.
Now f� f (x) dx is defined to be limfNf-+O I (X). Therefore what we need, to

complete the proof that continuous functions are integrable, is the following theorem:
Theorem 2. If f is continuous on [a, b], then

lim [S(N) - s(N)] = 0.
INJ-+O
Proof Let E > 0 be given. We need to show that there is a o > 0 such that
INI <o => S(N) - s(N)<E.
We know by the finite covering theorem that for every h > 0 there is a finite
collection of h-boxes, covering the graph. (See the left-hand figure below. We have
not yet decided what h we want to use.) The x-coordinates of the vertical sides of the
boxes, together with a and b, form a net N0 over [a, b]. Let o be the length of the
shortest interval in N0• We assert that if N is any other net over [a, b], with IN! <o ,
then every little interval [x;_1, X;] in N lies under some one of our boxes. We illustrate
y
y
with the simpler figure on the right. If [x;_1, X;] contains no point of N0 (as on the
right) this is evident. If [xi_1, X;] contains a point Y; of N0 (as on the left), then Y;
lies on the open interval under one of our boxes, and so [x;_1, x;] lies under the same
box.
Now take a net N, with INI <o . The difference S(N) - s(N) is the sum of the
areas of a collection of rectangles, like this:
�
I
I
I
I
I I
Each of these rectangles lies in one of our h-boxes. (Why?) Therefore each of them
has height � h. Hence
S (N) - s(N) � h(b - a).
Thus we want
h(b - a)<E,
and this will hold if
€
h<--.
b - a
This is the way we should choose h at the beginning of the proof. The resulting o
is the o that we need.
Theorem 3. Every continuous function on a closed interval is integrable.
Proof By the preceding theorem, lim [S(N) - s(N)] = 0. By Theorem 3 of Section

7.2, f is integrable.
Some of the ideas in this proof are worth examining further. In Problem Set 7.2,
we gave the following definition.
Definition. Let f be a function on an interval /. Suppose that for every E > 0 there
is a o > 0 such that
Ix - x'I < o =>
If(x) - f(x')I < E,
where x and x' are any two points of I. Then f is uniformly continuous on I.
Note that while continuity is defined for one point x at a time, uniform continuity
is defined for the graph as a whole. The difference between these ideas may be clarified
by an analogy:
1) A man is literate if there is a language that he can read and write.

2) A group of men is literate if each of its members is literate.
3) A group of men is uniformly literate if there is one language that every member
of the group can read and write.
Thus, uniform literacy is a property not of the individuals in a group but of the
group as a whole; if each of the members of the group is literate, it follows that the
group is literate [see (2)], but it does not follow that the group is uniformly literate.
The difference between continuity of a function f on an interval I and uniform
continuity of f on / is analogous. For example,f(x) = l/x is continuous on the open
interval I= (0, 1), because/is continuous at every point x of I. But/is not uniformly
continuous on I. (For every E > 0, we can find two points x, x', as close together as
we please, such that lf(x) - f(x')I > E.)
But continuity implies uniform continuity when the domain of the function is a
closed interval.
Theorem 4 (the uniform continuity theorem). If f is continuous on [a, b], then f is

uniformly continuous on [a, b].
Proof Let E > 0 be given. By the finite covering theorem there is a finite collection
of boxes of height E, covering the graph. (We are using E as the h of Theorem 1.)
Let N0 be the corresponding net over [a, b], as in the proof of Theorem 2. As before,
let o be the length of the shortest interval in N0• It follows that if Ix - x'I < o, then
x. and x' lie under some one of our boxes. Therefore lf(x) - f(x')I < E, which was
to be proved.
This is the idea that we need, to complete the proof of the formula
A = f27Tf(x).J1 + f'(x)2 dx
for the area of a surface of revolution about the x-axis. In Section 7.5, we knew that
as INI --+ 0, the sample sum

n
I' = I 27Tf(x;)J1
i=l
+ f'(x;)2 ilx
approaches the integral. But the area of the approximating surface was
I= I 2TTj(i;)Jl + f'(.X';)2 Llx,

i=l
with two different sample points i;, x; used on each interval rx;_1, x;]. Thus we need
to show that
Iim II' -
JNJ->O
II = o.
We are assuming that / ' is continuous. Therefore so also is 2TTJ I + f'(x)2• Therefore
the latter function is bounded. Let M be such that
2TTJI + f'(x)2 < M, for every x.
Let E be any positive number. Then
€
> 0.
M(b - a)
By the uniform continuity theorem, there is a o > 0 such that
Ix - x'I <O (
lf x) - f(x')I <M(b E
- a)
=>
This is the o that we need.
Proof
INI <o => Ii; - .x;1 <o for each i
=> lf(x;) - f(x;)I <M(b €- a)

for each i.
Now
I - I' = I 2TT[f(i;) - f(x;)]J1 + f'(x;)2 ilx,

i=l
and so
n
II - I'I � I lf(.x;) - !c.x;) 1 27TJ1 + f'CxD2 ilx

i=l
·
<IM· lf(i;) - f(x;)I ilx.

n
i=l
Therefore
n
/I - I'/ <i�M. M(b - a) Llx = M. M(b - a);� Llx

n
INI <o
E E
=>
'(b - a) =
E
b ·-a = -- E,

'PROBLEM SET 7.8
Most of the questions below can be answered on the basis of a careful reexamination of
the theorems and proofs in Sections 7.2 and 7.8. Some of them, however, require independent
investigation. Naturally, all answers should be explained.
I. Suppose that/is known to be increasing on [a, b], but is not known to be continuous.
Does it follow that f is integrable?
2. Show that Tan-1 is uniformly continuous on ( - oo, oo ).

3. Same, for f(x) = x on ( -co, oo ).
4. Is it possible for a function to be uniformly continuous on an open interval (a, b)?

Why or why not?
5. If a function is uniformly continuous on an interval I, does it follow thatf is continuous

on I? Why or why not?
6. Suppose that/is (a) continuous at a, (b) continuous at b, and (c) uniformly continuous
on (a, b) . Does it follow that f is uniformly continuous on [a, b]?
*7. Suppose that f is bounded and integrable, but not necessarily continuous, on [a, b].
For each x of [a, b], let
F(x) = f' f (t) dt.
Show that Fis continuous. (The betweenness theorem for integrals, which is Theorem 5
of Section 7.2, may be useful here.)
*8. For the definition of Lipschitzian, see Problem 12a of Problem Set 7.2. Show that if f
is Lipschitzian on I, then/ is uniformly continuous on I. (Here I may be any interval,
open or closed, finite or infinite.)
9. Let a1, a2, • • . , an be any finite sequence of numbers. Show that
10. Let f be continuous on [a, b]. Show that
If f(x) dx
I f�
If(x)I dx.
8 The Conic Sections
8.1 TRANSLATION OF AXES
In Section 2.2 we stated the definition of a coordinate system on a line L.
x X2
A coordinate system on L is a one-to-one correspondence
L <--+ R, p <--+ x,
between the points P of L and the real numbers x, such that the distance between
any two points is the absolute value of the difference of the corresponding numbers.
That is,
If we subtract the same number from the coordinate of every point, we obtain
another coordinate system on the line. If we subtract h from every x, then
and so
'
Therefore the distance formula works, for the new coordinates x = x - h.
This process is called translation. The origin is moved to the point h, and all the
other number labels are moved with it.
x= 0 2 h h+l h+3
1
x = -h 1-h 2-h 0 3
Thus the old and new coordinates are related by the formulas
'
x = x - h, x = x' + h.
Consider now a plane with a coordinate system.
356
8.1 Translation of Axes 357
y y'
y=k x
'
x=h
Suppose that we translate the coordinate system on the x-axis, subtracting h
from every x-coordinate, and then translate the coordinate system on the y-axis,
subtractingk from every y-coordinate. The effect is to move the origin to the point
(h, k). Every point p now has a new pair of coordinates x', y', and these are related
to the old ones by the formulas
x = x' + h, x' = x - h,
y = y' + k, y' = y - k.
These formulas are easy to remember; the only way you are likely to go wrong
is to get them backwards ( by writing x' = x + k ( ?) or y' = y + k ( ?)). It is easy
to see, however, that the new origin must have old coordinates h, k and new coordi
nates 0, O; and from this we can tell which way the formulas ought to go.
As usual, (x,y) denotes the point whose old coordinates are x and y. Thus the
old origin is (0, 0), and the new origin is (h, k). When we write (a, b)' (with a prime
outside the parentheses) we mean the point whose new coordinates are a and b.
Thus the new origin is (0, O)', and the old origin is ( -h, -k)'. More examples are
given below:
y y'
P=(5,2)=(2, 1)1
2 ------- --�
I
�+--=-t��.._--,+�-'-�.._--- x
- 1, 0 2 3 4 5
/L--1
Q=(-1, -l)=(-4, -2)1
In the figure, h = 3 and k = I. Two points have been labeled both ways. At the
point P, we have
x = 5, y = 2, x' = 2, y' = 1,
so that the label
p = (5, 2) = (2, l)'
is correct. Similarly, at Q we have
x = -1, y = -1, x' = -4, y' = -2,

358 The Conic Sections 8.1
so that
Q = (-1, -1) = (- 4 , -2)'.
When we write an equation to describe a figure in the plane, the equation depends
on the choice of axes; and often one choice of axes gives a simpler equation than any
other. If we didn't start with the axes in the best position, then we can simplify the
equation by translation of axes. For example, consider the parabola with directrix
y = -1 and focus F = (3, 3).
y
F=(3, 3)
x, y)
---- P=(
"'-.__
--\
v
��-+-��--- x
0 1 2 3 4 5 16
D--- ________________ n_ __
-1 M
The parabola is the graph of the condition

FP =MP.
Algebraically, this says that
.Jex - 3)2 + (y - 3)2 = )(y + 1)2

x2 - 6x + 9 + y2 - 6y + 9 = y2 + 2 y + 1
-<::::>
<=>- 8y = x2 - 6x + 17
<=>- y = tx2 - !x + -1i-.
We know, however, that a parabola with a horizontal directrix and vertex at the
origin always has an equation of the form y = ax2• The vertex of our parabola is
halfway between the focus and the directrix, at the point V = (3, 1). This means
that we should translate the axes so that the new origin becomes the point
O' = (h, k) = (3, 1).

Relative to the new axes, the equation becomes
8(y' + 1) = (x' + 3)2 - 6(x' + 3) + 17;

here we have replaced x by x' + h = x' + 3 and y by y' + k = y' + I. This gives
8y' + 8 = x'2 + 6x' + 9 - 6x' - 18 + 17,

or
8y' = x'2, or y' = tx'2.
This is in the standard form y' = ax'2, where a = 1/2pand pis the distance between
the focus and the directrix.
Thus, by a translation of axes, we have eliminated the linear term in x and the
constant term. Here we knew in advance where the origin ought to be for the equation
8.1 Translation of Axes 359
to appear in a simple form. If we hadn't known this, we could still have investigated
algebraically, to find out what sort of simplifications a translation could accomplish.
To do this, we would regard hand k as unknown quantities, and make the substitution
x= x' + h, y = y' + k
in general form. This gives:
S(y' + k)= (x' + h)2 - 6(x' + h) + 17,

or
x'2 + (2h - 6)x' - Sy' + h2 - 6h + 17 - Sk= 0.
Certain facts are now obvious: (1) We can't get rid of the term x'2, by any choice
of h and k, because h and k do not appear in the coefficient of x'2. (2) For the same
reason, we can't get rid of the linear term in y'. (3) The total coefficient of x' is 2h - 6,
and the total constant term is
h2 - 6h + 17 - Sk.
We can therefore get rid of the x' term, by using h = 3. The constant term then
becomes
9 - 1 S + 17 - Sk' or s - Sk,
which is 0 when k= 1. Thus, translating the origin to the point (h, k)= (3, 1),
we get the equation in the form
Sy'= x'2 or y'= tx'z,
as before. This is the process that you follow if you don't know the answer in advance
PROBLEM SET 8.1
1. Find a translation which eliminates both of the linear terms from the equation
xy - 5y - 6x - 30 = 0.
Then sketch the graph, showing both sets of axes.
2. Is there a translation which eliminates the xy-term from the above equation? Why
or why not? How about the possibility of removing the constant term?
3. Find a translation which removes both linear terms from the equation
x2 + y2 + x + y - 2 = 0.
Then sketch the graph, showing both sets of axes.
2xy - x + 3y - 2 = 0.
x2 + y2 + 4x + 2y + 1 = 0.
x2 + xy - 3x + 2 = 0.
7. Find a translation that eliminates both linear terms in the equation
x2 +xy +y2 +x +y +5 = 0.
x2 +xy +y2 +x +y +1 = 0.
9. Show that there is no translation which removes both linear terms from the equation
x2 +2xy +y2 +x - y +1 = 0.
10. Show that there is no translation which removes both linear terms from the equation
4x2 +4xy +y2 +2x +y +8 = 0.
l 1. Consider the equation x2 +y2 +x +y - 2 = 0. Under what conditions for fl

and k does this equation take the form
x'2 +y'2 +Ax' +By' = 0,
with possible linear terms but no constant term? (You may be able to think of a way to
answer this question without doing any calculations at all.)
12. Show that if ad - be ;;f. 0, then the linear system
ah +bk = e, ch +dk = f
always has a solution. (Simply start solving it; at some point, you will need to assume
that ad - be ;;f. 0.)
13. Consider an equation of the form
Ax2 +Bxy +Cy2 +Dx +Ey +F = 0.
Show that if B2 - 4AC ;;f. 0, then there is always a translation that eliminates both of
the linear terms. (The converse is not true; there are simple examples of equations
where B2 - 4AC = 0, but where the linear terms are absent to begin with. Examples?)
14. Sketch the graph of the equation y2 = x(x +l)(x - 1).
15. Let C be the graph of an equation of the form

3
b2 y2 +h1Y +ho G3X +a x2 +G1X +ao.
2
=
Show that if the axes are translated, then C is the graph of an equation of the same form,
in the new coordinates x' and y'.
8.2 THE ELLIPSE
Let F and F' be two points, let c be half the distance between them, so that
FF'= 2c,
and let a be a number greater than c. Let C be the graph of the equation
FP + F'P = 2a.
The curve C is called the ellipse with foci F, F' and focal sum 2a.
8.2 The Ellipse 361
To draw an ellipse, we put two thumbtacks in a drawing board, at the foci F

and F'. We tie the ends of a string to the thumbtacks, in such a way that the length
of the string left free between the thumbtacks is 2a. Then we put a pencil in the loop
of string, placing the point so that the string is taut, and move the pencil around,
keeping the string taut all the way. (We need to do this in two steps, on the two sides
of the line through F and F'.)
In the definition of an ellipse, we really mean that F and F' are two points; that is,
F ;>6 F'. (Thus a circle is not an ellipse.) Also, we really mean a > c. (For a = c,
the graph of the condition FP + F'P = 2a is the segment from F to F'; and for
a < c, the graph is empty.)
Some things about ellipses are easily seen from the definition. For the definition
of symmetry of a figure, about a line or a point, see Section 7.6.
Theorem 1. An ellipse is symmetric about the line through its foci.
Proof In the left-hand figure below, P is on the ellipse, so that
FP + F'P = 2a.
And L is the perpendicular bisector of the segment from P to P'. By elementary

geometry, FP = FP', and F'P = F'P'. Therefore FP' + F'P' = 2a, and P' is on the
ellipse.
Theorem 2. An ellipse is symmetric about the perpendicular bisector of the segment

between its foci.
Proof? (This is not quite as simple as the preceding theorem. See the right-hand
figure above.)
Theorem 3. If a curve is symmetric about each of two perpendicular lines, then it is

symmetric about their point of intersection.
Proof? (We need to show that OP = OP".)
L'
P'
--,
I
I
I
I
I
I
For ellipses, this gives us:
Theorem 4. Every ellipse is symmetric about the point midway between its foci.
P0 is called the center of the ellipse. (See the right-hand figure above.)
These symmetry theorems convey nearly all that is easy to see about ellipses
merely from the definition. Our next step is to set up a coordinate system, and describe
our ellipses by equations. We take the origin at the center of the ellipse, and the foci
on the x-axis. The ellipse is then said to be in standard position, relative to the axes.
y
P(x, y)
As indicated in the figure above, let F and F' be the points ( -c, 0) and (c, 0). Then
FP + F'P = 2a
� J(x + c)2 + y2 + J(x - c)2 + y2 = 2a
� J(x + c)2 + y2 = 2 a - J(x - c)2 + y2
-----
=> x2 + 2cx + c2 + y2 = 4a2 - 4aJ(x - c) 2 + y2 + x2 - 2cx + c2 + y2
� aJ(x - c) 2 + y 2 a2 '-- ex
=
=> a2x2 - 2 a2cx + a2c2 + a2y 2 = a4 - 2 a2cx + c2x2

� (a2 - c2 )x2 + a2y2 = a2(a2 - c2)
x2 Y2 -
� + -- =l.
a2 az - c 2
8.2 The Ellipse 363
Thus every point on the ellipse satisfies the final equation. It is possible to show,
conversely, that every point (x, y) that satisfies the final equation lies on the ellipse.
(See Problem 22 below.) Thus we have:
Theorem 5. The ellipse with foci at (±c, 0) and focal sum 2a is the graph of
the equation
x2 y2
+
-�- = 1.
a2 a 2 - c2
For example, for a = 3 and c = 2 we get
x2 y2
-+-=l.
9 5
To sketch, we observe that for y = 0, x = ±3, and for x = 0, y= ±)5. We

then sketch an oval with these as its extreme points.
- 1
-2
Given an equation
x2 y2
-+- = 1 '
a2 b2
the graph is always an ellipse. Since a2 - b2 > 0, it follows that a2 - b2 = c2

for some c > 0. The graph is therefore the ellipse described in Theorem 5. Thus we
have proved half of the following theorem.
Theorem 6. Given the equation
x2 y2
+ /;2 1.
=
a2
For b2 < a2, the graph is the ellipse with focal sum 2a and foci at (±c, 0), where
c = )a2 - b2. For a2 < b2, the graph is the ellipse with focal sum 2b and foci
at (0, ±c), where c = )b2 - a2.
Proof of the second half of the theorem?

y
y
b
-a a x
-b
b<a, c=Va2-b2, a<b, c=vb2-a2,

FP+F'P=2a. FP+F'P=2b.
If the foci are not in either of the two positions shown above, then the equation
of the ellipse is more complicated. In some cases, when the equation is given, we can
simplify the equation by a translation of axes. Consider
4x2 + 9y2 - 8x + 18y - 23 = 0.
Making the substitutions x = x' + h, y =y' + k, we get
4x'2 + 9y'2 + (8h - 8)x' + (18k + 18)y' + 4h2 + 9k2 - Sh + 18k - 23 = 0.
Evidently we want h = 1, k = -1; and this gives the equation in the form
x2 y2
2 2 ' '
4x ' + 9y' - 36 = 0, or -+-=l.
9 4
The graph is the ellipse with foci at (±)5, O)' and focal sum 6; it intersects the x'
and y'-axes at the points (±3, O)', (0, ±2)'. We can now sketch, showing both sets of
axes
y '
y
-_
-3-+---�--+-_
- l-+-�-_._.,__+3---. x'
-1
-2
In doing such sketches, we start by drawing the new axes and the curve, in a con
venient position on the paper, and then draw the old axes, in the position where
they must have been.
8.2 The Ellipse 365
PROBLEM SET 8.2

Write equations for the ellipses described by the following conditions and sketch.
1. Foci at (±1, O); focal sum 4. 2. Foci at (0, ±1); focal sum 4.
3. Foci at (1, 2), (1, 4); focal sum 4. 4. Foci at (-1, -1), (1, 1); focal sum 4.
5. Foci at (-1, 1), (1, -1); focal sum 4.
6. Foci at (±2, O); focal sum 6. 7. Foci at (0, ±2); focal sum 6.
8. Foci at (-1, 1) and (I, -1); focal sum 6.
Find the foci and the focal sum, and sketch, showing both sets of axes, in cases where
more than one set is used.
9. x2/4 + y2 =1 10. x2 + 9y2 - 2x + 36y + 28 =0

11. 9x2 + y2 + 36x - 2y + 28 =0
12. x2 + 2y2 + 3x + 4y - 6 =0 (This one does not "come out even.")
y2 2y
13. 4x2 + y2 =1 14. x2 + x + - + - + 1 =0
9 3
15. Given an equation of the form
Ax2 + By2 + Cx + Ey + F = 0,
where A and Bare both positive, show that the graph is (a) an ellipse, (b) a point, or (c)
the empty set. (The same conclusion follows if A and B are both negative.)
16. A function f is odd if/(-x) = -f (x) for every x. Show that the graph of an odd
function is symmetric about the origin.
17. a) Let C be the graph of the sine function. Show that C is symmetric about the origin.
b) Now show that C is also symmetric about infinitely many other points. (Thus it
may happen that an unbounded figure has more than one "center." In fact, there
is a simpler example: a line is symmetric about each of its points, and so every point
of a line is "a center" of the line. For this reason, we ordinarily use the word
center only for bounded figures.)
18. a) Show that the graph of the cosine is symmetric about infinitely many points.
b) Show that the graph of the sine is symmetric about infinitely many lines.
19. Consider the infinite strip R between the lines y =1 and y = -1. That is,
R = {(x, y) I -oo < x < w, -1 2 y 2 l}.

Show that R is symmetric about infinitely many points, and find a simple description of
the set C consisting of all points which are "centers" of R.
20. Show that every cubic curve is symmetric about its point of inflection. Here by a cubic
. curve we mean the graph of an equation y = ax3 + bx2 + ex + d, with a yf 0.
21. Suppose that in Theorem 3 of this section we drop the hypothesis that the two lines of
symmetry are perpendicular. Would the resulting theorem be true? Why or why not?
*22. Given 0 < c < a, as in the definition of an ellipse. Let F = ( -c, O), F' = (c, 0).
Let P = (x, y) be a point satisfying the equation
x2 y2
a2 + a-2--2c -
= 1.
a) Show that
a2 - e2
y2 = -- - (a2 - x2).
a2
b) Show that
1 I
FP + F P '\/ (a2 + ex)2 + 1 '\/I (a2 - ex)2•
I
= - -
a a
c) Show that a2 + ex > 0. (There are two cases to consider, x � 0 and x � 0.

Remember that 0 < e < a, and use the fact that /x/ � a.)
d) Show that a2 - ex > 0.
e) Show that FP + MP = 2a.
(This completes the proof of Theorem 5.)
8.3 THE HYPERBOLA
Given 0 < a < c, and the points F and F', with FF' = 2c. Let C be the graph of the
condition
FP - F'P = ±2a (a < c).
The curve C is called the hyperbola with foci F, F' and focal difef rence 2a. The figure
shows what a hyperbola looks like, but the reasons for this appearance of the graph
are not obvious; the only thing that is easy to see, on the basis of the definition, is
that the hyperbola is symmetric about each of the two perpendicular lines. The first
step in our investigation of hyperbolas is to take the axes in a convenient position, as
shown above, with F = (-c, 0) and F' = (c, 0), and get an equation for the curve.
FP - F'P = ±2a
<=> FP = F'P ± 2a
J(x + c)2 + y J(x - c)2 + y2 ± 2a
2
<=> =
=> x2 + 2ex + c2 + y2 = x2 - 2cx + c2 + y2 ± 4aJ(x - c)2 + y2 + 4a2

<=> ex - a2 = ±aJ(x - c)2 + y2
8.3 The Hyperbola 367
� c2x2 - 2a2cx + a4 = a2x2 - 2a2cx + a2c2 + a 2y2

-¢:? (c2 _
a2)x2 _
a2y2 = a2(c2_
a2)
y2
--- = 1.
c 2 - a2
Thus every point P = (x, y) of the hyperbola satisfies the final equation. As in the
case of the ellipse, it can be shown conversely that every point on the graph of the
final equation is on the hyperbola. (See Problem 32 below.) Since c2 > a2, we may let
b2 = c2 - a2.
This substitution gives the standard form of the equation:
x2
---Y2 =l.
a2 b2
And we can sum up as follows:
x2 Y2
- - - = 1
a2 bz
is the hyperbola with foci at (±c, 0) (where c =.Ja2 + b2) and focal difference 2a.
We shall use our equation to justify the sketch which we gave at the outset.
1) No point of the hyperbola lies between the lines x = -a and x = a. The
reason is as follows. Solving for y, we get
Therefore the hyperbola is the union of the graphs of two functions
f(x) = "l}_.Jx2 - a2 and g(x) = - "!}_ .Jx 2 - a2,

a a
and neither of these functions is defined for -a < x < a.
2) The curve is symmetric about each of the coordinate axes. This is easy to see
algebraically. For each point (x, y), the symmetric point across the x-axis is (x, -y);
and if (x, y) is on the hyperbola, then so also is (x, -y). Similarly, if (x, y) is on the
curve, then so also is (-x, y); and so the hyperbola is symmetric about the y-axis.
3) The hyperbola is unbounded in both the x- and y-directions. Obviously f(x)
and g(x) are defined whenever !xi � a. As x-+ oo, f(x) -+ oo and g(x)-+ - oo.
And as x-+ oo,f(x) -+ oo and g(x) -+ - oo.
-
It remains to discuss the two lines which the curve seems to be getting close to
when both x and y become numerically large. The behavior of the hyperbola relative
to these lines seems to be similar to that of the curve
= f(x) = l,
x
Y
relative to the coordinate axes. The coordinate axes are called asymptotes of this
curve. By this we mean, roughly speaking, that points of the curve, far from the
origin, in the appropriate directions, are close to the axes. We want to extend this
idea to cases in which the asymptote is neither horizontal nor vertical.
y 1
y=f(x) = x' x;-<0.
As x-+ oo, the distance from the point P= (x, y) to the x-axis approaches 0.
(This distance is MP= IYI = 1 1/x j . ) We shall take this property as our definition
of an asymptote. That is, a line Lis an asymptote of a function-graph if the distance
from the line to the point P= (x,j(x)) approaches 0 as x-+ oo, or as x-+ oo.
=
-
It is evident that the x-axis is an asymptote of the graph off (x) I/x under this
definition. In fact, the x-axis is an asymptote in both the positive and negative
directions. We also say that a line L is an asymptote of a curve C if C contains a
function-graph which has L as an asymptote. In the case of y= l/x, we also have
x = g(y) = 1/y; thus the curve, looked at sidewise, is still a function-graph, and has
the y-axis as an asymptote, in both the positive and negative directions. This is shown
in the left-hand figure below.
y
x=g(y) = ly (y;o<O) y=f(x) = l
x (x;o<O)
M p
IimMP=O
y-.ro
lim MP=O
M Jim MP=O
y-.-ro
We return to our hyperbola. In the last figure on the preceding page, the slope
of the segment from the origin to the point P = (x, y) is
m(x) = ,!'.
x
= ..! £ .Jx2 - a2 = £ 1
x a
·
a
j - a2
x2
•
Obviously
Jim m(x) = bfa,
and this suggests that the line y= bx/a, or x/a - y/b = 0, is an asymptote of the
part of the curve that lies in the first quadrant. If we show this, then it will follow
by symmetry that the lines x/a ± y/b = 0 are asymptotes of the curve in all four
quadrants.
Thus we need to show that Jim., ... 00 MP = 0. Since MP < NP, it will be sufficient
to show that lim NP = 0. This can be done by an algebraic trick.
NP= £x _ E..Jx2 - a2 = E.(x - .Jx2 - a2)

a a a
x + .Jx2 - a2
£. (x - .Jx2 - a2).
+
=
a x .Jx2 - a2
b a2
-
a x + .Jx2 - a2 .
Obviously NP --+ 0 as x --+ oo. Therefore MP --+ 0, which was to be proved. This
gives the following theorem.
Theorem 2. The lines

x y
-±-=0
a b
are asymptotes of the hyperbola
x2 y2
- l.
a2 b2
You can sketch a hyperbola by drawing in ·the asymptotes and x-intercepts
exactly, and then filling in the curve freehand.
y
For example, consider

2 y2
x
- - - = 1.
9 4
The x-intercepts are at x = ±3, and the asymptotes are the Jines
y
x
-±- = 0 or ±Jx.
' Y =
3 2
A hyperbola whose asymptotes are perpendicular is called rectangular.
y =-x y y=x
' /
' /
' /
' /
' /
' /
' /
' /
' /
' /
' /
/ '
/ '
/ '
/
/
'
/ '
/ '
/ '
/ '
/ '
/ '
/ '
If such a hyperbola is in standard position, then the asymptotes must be the lines
x±y = 0, and the equation must have the form
2 y2
x
- - - = 2 2 2
2 2 1, or x -y =a .
a a
If the foci are on the y-axis, at the points (0,±c ) , then the equation of the hyper
bola takes the form
2
x
--- = 1.
2
c - a2
It follows that the graph of the condition
2 2
x y
= ±1
2 2
a b
is the union of two hyperbolas with the same asymptotes. These are called conjugate
hyperbolas.
PROBLEM SET 8.3
Sketch the graphs of the following equations.

l. x2 - 4y2 = 4 2. y2 - 4x2 = 4 3. x2 - 4y2 = -4
4. y2 - 4x2 =
-4 5. 9x2 - 4y2 = 36 6. 9x2 - 4y2 = -36
7. -x2 + 9y2 = 9 8. -9x2 + y2 = 9 9. 25x2 - 4y2 =
100
10. 25 y2 - 4x2 =
100
Derive equations for the hyperbolas determined by the following conditions, and sketch.
11. Foci at (±2, 0); focal difference 3. 12. Foci at (±2, 2); focal difference 3.
13. Foci at (0, 0) and (0, 4); focal difference 3. 14. Foci at (0, ±2); focal difference 3.
15. Foci at (±1, ±I); focal difference 2.
16. Foci at (±2, O); passing through the point (3, 4).
17. Foci at (±2, 0); focal difference 2.
18. Foci at (±3, 0); focal difference 4.
19. Foci at (0, ±3); focal difference 4.
20. Foci at (±3, O); passing through the point (5, 5).
21. Given F, F', and a, as for a hyperbola in standard position. What is the graph of the
condition FP - F'P = la? How about the graph of FP - F'P = -la?
22. Find a rectangular hyperbola in standard position (with asymptotes x + y = 0 and
x - y =
0) passing through the point (5, 3).
Investigate the graphs of the following equations. In each case, find all asymptotes.
23. (x2 - y2 - 1)2 =0 24. (x2 - y2)2 =

I
I
25. x2y2 - xy + ± =
0 26. y = ------
(x - l)(x - 2)
27. Let D be the line x -1, let F be the origin, and for each point P let DP be the
perpendicular distance between D and P. Let C be the graph of the condition
FP
2
DP= ·
What sort of curve is this? Sketch.
28. Let F and D be as in the preceding problem, and let C' be the graph of the condition
FP 1
=
DP 1.
What sort of curve is this? Sketch.
29. _Let G be the set of points P such that CP = 2DP, where C is the circle x2 + y2 = I and
D is the line x = 4. What sort of figme is G? Discuss and sketch.
30. The following passage occurs in the U.S.Internal Revenue Act of 1964.
" ...There shall be allowed as a deduction moving expenses paid ... in connection
with the commencement of work by the taxpayer ... at a new principal place of work ...
[However,] no deduction shall be allowed ... unless ... the taxpayer's new principal
place of work ... is at least 20 miles farther from his former residence than was his
former principal place of work ... "
Give a sketch, showing what this means. Your sketch should show (a) the former
residence, (b) the former place of work, and (c) the region in which the new place of
work must lie, for the moving expenses to be deductible. (The author is indebted,
for this problem, to Dr. Henry Pollak, of the Bell Telephone Laboratories.)
31. The region between two conjugate hyperbolas stretches out infinitely far, in each of four
directions. Find out whether the area of such a region is finite.
*32. Given 0 < a < e, as in the definition of a hyperbola. Let F = ( -e, 0), F' = (e, 0).
Let P (x, y) be a point satisfying the equation
=
x2 v2
- __,_ = I
a2 e2 - a2 .
a) Show that
e2 - a2
y2 = -- - (x2 - a2).
a2
b) Show that
I .I 1
FP - F P - v (ex + a2)2 - - '\I (ex - a2)2.
I
=
a a
c) Show that, if x � a, then
ex + a2 > 0, ex - a2 > 0, and FP - F'P = 2a.
d) Show that if x ;;;i -a, then
ex + a2 < 0, ex - a2 < 0, and FP - F'P = -2a.
(This completes the proof of Theorem 1.)
8.4 THE GENERAL EQUATION OF

THE SECOND DEGREE. ROTATION OF AXES
An equation of the second degree in x and y is an equation of the form
Ax2 + Bxy + Cy2 + Dx + Ey + F = 0,
where at least one of the coefficients A, B, and C is different from zero. The latter
condition is to guarantee that the degree of the equation really is 2, rather than 1 or 0.
We have found that all conic sections are graphs of equations of this type; and we shall
now investigate the converse. That is, we propose to find out what sort of figure can
be the graph of a second-degree equation. The possibilities that we have already
found are
a) a circle,
b) a p arabola,
c) an ellipse,
d) a hyperbola.
There are other possibilities, which we noted as exceptional cases when we were
studying the equation
x2 + y2 + Dx + Ey + F = 0,
8.4 The General Equation of the Second Degree. Rotation of Axes 373
in connection with the circle. The graph of
x2 + y2 = 0
is a point; and the graph of
x2 + y2 + 1 = 0
is empty. (See Theorem 2 of Section 2.3.) Our list of possible graphs of second
degree equations must therefore include
e) a point, and
f) the empty set.
And this is not all. The graph of
y2 = 0
is a line, namely, the x-axis. And the graph of
xy = 0
is the union of two lines, namely, the two axes. Similarly, the graph of
x2-y2=0
x2-y2 = (x
is the union of two lines. The reason is that y)(x + y). This is = 0
-
if and only if either x -y = 0 or xy = 0. Therefore a point P = (x, y) is on

+
the graph of x2 -y2 = 0 if and only if (i) P is on the line y = x or (ii) P is on the
line y = -x.
y
In this example, the lines intersect, but we may get the union of two parallel lines.
The equation
x2- x = 0
is equivalent to
x(x - 1) = O;
and the graph is therefore the union of the two parallel lines x = 0 and x = 1. Thus
the graphs, for the general equation of the second degree, include
g) a line, and
h) the union of two lines, either parallel or intersecting.
We shall show that the eight possibilities that we have just listed are the only possi
bilities. The method will be to reduce the equation to a recognizable form by moving
th� axes. In some cases, this cannot be done by translation; we may also have to use
rotation of the axes.
Suppose that we rotate the axes through an angle of measure 8, getting a new
pair of axes.
y' y
In the figure, r is the distance OP; P has coordinates x, y in the old coordinate system,
and coordinates x', y' in the new coordinate system. Evidently
x = r cos cp, y = r sin cp,
x' = r cos (y the formulas
x' = x cos 8 + y sin 8, y' = -x sin 8 + y cos 8.
If we rotate the new axes through an angle of measure -8 we are back where we
started. Therefore the old coordinates are expressed in terms of the new ones by the
formulas
x = x' cos (-8) + y' sin (-8), y = -x' sin (-8) + y' cos (-8).
These give
x = x' cos 8 - y' sin 8' y = x' sin 8 + y' cos 8.
Theorem 1. In any second-degree equation, the xy-term can be eliminated by a

rotation of axes.
Before going into the proof, let us try a simple example:
xy = l.
To rotate the axes through an angle 8, we should substitute
x = x' cos 8 - y' sin 8' y = x' sin 8 + y' cos 8. (1)
The equation then becomes
(x' cos 8 - y' sin 8)(x' sin 8 + y' cos 8) = 1,

or
x'2 sine cos 8 + x'y'(cos2 8 - sin2 8) - y'2 sine cos 8 = l.
We want the x'y'-term to vanish. Thus we want
cos2 8 - sin2 8 = 0, or cos 26 = O;

and this will happen when

'TT Tr nTr
W = - + nTr ' 8 = - + -.
2 4 2
One value of 8 is all we need, and so we take 8 = Tr/4, which gives
sin 8 = cos 8 = �
)2
and
sin 8 cos 8 = t.
Thus our new equation is
x' 2 y' 2 1.
=
2 2
This is the equation of a rectangular hyperbola.
Let us now return to our general equation
Ax2 + Bxy + Cy2 + Dx + Ey + F = 0.
Making the usual substitution, to rotate the axes through 8, we get
A(x' cos 8 - y' sin 8)2 + B(x' cos 8 - y' sin 8)(x' sin 8 + y' cos 8)
+ C(x' sin 8 + y' cos 8)2 + D(x' cos 8 - y' sin 8)
+ E(x' sin 8 + y' cos 8) + F = 0.
When we collect coefficients for the terms of various types, we get a new equation of
the same form, like this:
A'x'2 + B'x'y' + C'y'2 + D'x' + E'y' + F' = 0.

Algebraically,
A' = A cos2 8 + B sin 8 cos 8 + C sin2 8,

B' = -2A sin 8 cos 8 + B(cos2 8 sin2 8) + - 2 C sin 8 cos 8,
C' = A sin2 8 - B sin 8 cos 8 + C cos2 8,
D' = D cos 8 + E sin 8,
E' = - D sin 8 + E cos 8,
F'= F.
For future reference, we have written down all of these, but for the moment, all we
are interested in is B': we want to find a e that makes B' =0. Simplifying trigono
metrically, we get
B' = (C- A) sin2e + Bcos2e.
There are now two cases:
1) If A =C, then B' =Bcos 28. We must have B ¥- 0, or there wouldn't be any
xy-term in the original equation. Therefore
B' =0 when cos2e =o,

and cos28 =0 when
e = 27:.
4
Thus a rotation through TT/4 eliminates the xy-term whenever A =C.
2) If A ¥- C, then we can divide by A -C. Therefore B' = 0 when
B cos 28 = (A-C) sin 2e, or ---

A-C
B
=tan 28.
Thus, to get B' =0, we take

- B
e �Tan I --- .
=
-
A-C
This proves the theorem. (The theorem did not say that the coefficients in the new
equation were easy to compute.)
Theorem 2. The graph of a second-degree equation is (a) a circle, (b) a parabola,

(c) an ellipse, (d) a hyperbola, (e) a point, (f) the empty set, (g) a line, or (h) the
union of two lines (either parallel or intersecting).
Proof By the preceding theorem, we can assume that there is no xy-term. The
equation then has the form
Ax2 + Cy2 + Dx + Ey + F = 0.
We now need to discuss various cases.
1) Suppose that neither A nor C is = 0. We can then write
( �)
A x2 + x
2
+ C y + ( �) y = -F,
and complete the square twice to get
( �r c(y �r
A x + + +
2
= -F +
2
� :;2,
2
+
which has the form

Ax '2 + Cy'2 = F'.
Here we have translated the axes letting
D , E
x =x +- y =y+-.
I
2A' 2C
Since A =;r6 0, we can divide by A, getting
x'2 + C'y'2 = F" (C' = C/A =;r6 0).
There are six possibilities to be considered. For each of these cases, we have indicated
on the right what sort of figure the graph is.
(C' > 0, F" > 0 a circle or ellipse

C' > 0, F" = 0 a point
C' > 0, F"< 0 the empty set
(C'< 0, F" > 0 a hyperbola

C'< 0, F" = 0 two intersecting lines
C'< 0, F"< 0 a hyperbola (with foci on the new y-axis)
2) Suppose that C = 0. The equation then has the form
Ax2 + Dx + Ey + F = 0,
where A =;r6 0, because the degree is 2. We divide by A, getting
x2 + D'x + E'y + F' = 0;
and then we complete the square in x, so as to eliminate the linear term in x. This
gives an equation of the form
x'2 + F" = E'y.
For E' =;r6 0, this is a parabola. For E' = 0, the equation x'2 = - F" gives one line,
two lines, or the empty set.
3) Suppose that A = 0. This is exactly like Case 2; we interchange x and y, and
proceed as before. This completes the proof of the theorem.
It is easy to compute the new coefficients produced by a translation of axes.

For a rotation, the new coefficients are expressed in terms of sine and cose, and e
is defined by the equation
e = 1.
2
Tan-1 _B
__
A-C
Thus we want to express sine and cose in terms of tan 28 ( = B/(A - C)) for the
case where
_!!<28<�.
2 2
When 28 is in the first or fourth quadrant, cos 28 > 0, and sin 28 has the same sign
as tan 28.
v1+k2
k?
k?
v1+k2
In the figure,
B
k = tan 2() = ___ .
A-C
Therefore
1
cos 2() =
I
v1 +k2
The half-angle formulas are
;
J J
x i+cos x . x l - os x .
cos-= ± ' sin-= ±
2 2 2
For the present case, these give
J J
i+cos 2e . i- cos 2()
cos e = ' sin e = ± '
2 2
where
1
cos 2() =
I +k2
v1
and where the sign in the formula for sine is the same as the sign of k =tan 2(),
3x2 +2xy +y2 =1.
Here
A=3, B =2 , c =1,
and
B 2
k= =--=1.
--
A-C 3-1
Therefore
1 1
cos 2e = = ----=
.
J1 + k2 J2
Hence
J J
i+1;/2 2 + /2
cos e = =
'
2 4
and
(In the second formula, sine > 0 because k > 0.) Therefore
2 + /i 2 - .J2 /i.
cos2e =
' sin2 e =
'
sine cose =
4 4 4
The new equation is

A'x'2 + C'y'2 =
1,
where
A' =
A cos2e + B sine cose + C sin2e
2 + .J2 . .J2 2 - .J2 2

+2 + .J + 2
3
.
=
4 4 4
and
C' =
A sin2e - B s in e cose + C cos2e
2 - .J2 - . .J2 2 + .J2 -.J2
2 + + 2.
3.
=
4 4 4
PROBLEM SET 8.4
In these problems, when you are asked to inrestigate an equation, you should find out
what sort of figure the graph is, and sketch. If the graph is a conic section, you should also
find the coefficients in the standard form.
1. Investigate
x - xy = I .
(Here it is easier to translate first and rotate second. Sketch, showing all three sets
of axes.)
2. Investigate
x2 - xy = 1.
3. Investigate
xy - x - 2y = 0.
4. Investigate
2xy - y2 + 2 = 0.
5 . . Investigate
x2 + 2xy + y2 + 2x + 2y + l = 0.
6. Investigate
x2 + 4xy + 4y2 + 4x + 8y + 3 = 0.
7. Show that, under a rotation of axes,
A' + C' = A + C and F' = F.
We express this by saying that A + C and Fare invariant under rotation.

8. a) Given the general equation of the second degree. Let A0, B0, C0, . • • be the new
coefficients, when the axes are rotated through an angle of measure 8. Thus A8, B8,
C8, • • • are the A', B', C', . . . of the text; and so
A8 = A cos2 8 + B sin 8 cos 8 + C sin2 8,

B8 = (C - A) sin 28 + B cos 28,
C0 = A sin2 8 - B sin 8 cos 8 + C cos2 8.
Show that the derivatives A�, B�, C� satisfy the differential equations
b) Show that the function

f(8) = B� - 4A8C8
is a constant. Thus we say that B 2 - 4AC is invariant under rotation of axes. It
may be of some interest to check this, in the cases where we have computed the new
coefficients.
9. Given x2- + 2xy + 3y2 + 4x + 5y + 6 = 0, the axes are rotated so as to eliminate
the xy-term. What are the possibilities for the coefficients of x2 and y2, in the new
equation?
10. Same question, for the equation
x2 + 2xy + 5y2 - 10 = 0.
11. Same question for

4x2 + v3 xy + y2 + 2x + 3 = 0.
12. a) Let D be a line, let F be a point not on D, and let e be a positive number. Let G
be the set of all points P such that
FP
= e
DP ,
where DP is the perpendicular distance from D to P. G is called the conic section
with directrix D, focus F, and eccentricity e. Show that G is (a) an ellipse if e < 1,
(b) a parabola if e = I, and (c) a hyperbola if e > 1.
b) Is a circle a conic section, in the sense defined in Problem 12(a)? Why or why not?
Paths and
9 Vectors in a Plane
9.1 MOTION OF A PARTICLE IN A PLANE
To describe the motion of a particle in a plane E, we need to explain where the

particle is, at each time t on a certain interval /. Thus to each time t on the given
interval I there corresponds a point P(t); and the motion is described by a function
P: I-E.
For the motion shown in the figure, I is the infinite interval [O, oo ), and the initial
point P(O) is the point (1, 1 ) .
In general:
Definition. A plane path is a function
P: I-E,
where I is an interval and E is a plane.
The same idea applies more generally: a path in space is a function P: I - S,

where I is an interval and S is space. In this chapter we shall be dealing only with
plane paths, and so we shall refer to them for short simply as paths.
The locus of a path is the curve which is traced out by the moving point. More
precisely:
Definition. Given a path

P: I-E,
the locus of P is the set of all points Q which are = P(t) for some t in I.
381
382 Paths and Vectors in a Plane 9.1
Briefly, the locus of P is the image of I under the function P. The locus is deter
mined when the path is named, but given a locus, the path is not determined: the
same curve can be traced out by a moving point in infinitely many ways.
We describe a path in a coordinate plane by defining two functions which give
the coordinates of the moving point for each time t. For example, we might take
x = f(t) 4t
= }
y = g(t) = 8t2
Here
/=(-00,00), and P(t) = (4t, 8t2).
At t = 0, P(t) = (0, 0). As t increases, starting from 0, both x and y increase, but y
increases faster. In fact, the locus of this path is a parabola. To see this, we observe
that from the first equation, t = x/4. Substituting in the second equation, we get
y= sGY =
tx2.
Thus every point of the path lies on the graph of the equation
And it is easy to check, conversely, that every point (x, y) of the parabola is on the
path.
y
When a path is described by a pair of functions x =f(t), y = g(t), the two

functions are called the coordinate functions, parameter. Sometimes
and tis called a
we can get a simple description of the locus of a path by writing an equation in x and y.
We then say that we have eliminated the parameter, getting a rectangular equation of
the locus. Often this process is useful: a path may trace out a simple figure, such as a
segment or a circle, in a complicated way; and when this happens we want to know it.
The process of getting rectangular equations for loci is often tricky. Consider,
for example, the path P described by the equations
x =f(t) = t2, y = g(t) = t4•

Every point of this path lies on the parabola
y = x2.
But the converse is not true. On the path, we always have x ?; 0, because t2 ?; 0.
Therefore the locus of P is only half of the parabola.
9.1 Motion of a Particle in a Plane 383
\
\
\
\
\
\
'
'
', P(O)
,
�� -- '-
� ,....-::;
- . ��-'---- x
1
We are always free to regard a parameter as representing time, and in many

physical problems, this is what the parameter means. But it often makes equally
good sense to regard the parameter as the measure of an angle. Consider, for example,
x =cost, y =sint.
These functions describe uniform motion around a circle. Here we may regard the
parameter as the measure of an angle, and write
x =cose, y =sin()
to describe the same path.
Somewhat similar looking paths have ellipses as their loci. For example, consider
x =a cose, y =b sine.
Here the locus of Pis an ellipse:
::
=cos() ' 2' = sin() '
a b
x2 y2
2 .
-2 + -2 =COS () + Sill2 () = 1.
a b
Investigating further, we see what values of () correspond to what points of the

elliptical locus. We draw circles of radii a and b, with centers at the origin, and
construct L() in standard position.
In the figure,
Q = (a cos e, a sin8),
R = (bcos8,bsin8).
Therefore
P = P(8) = (a cos8, bsin8).
Following the scheme of the above figure, using drawing instruments, you can plot
as many points of the ellipse as you want to, without making any numerical calcula
tions. The same idea is used in the construction of a drawing instrument called the
ellipsograph, which can be adjusted so as to draw the ellipse with any pair of semiaxes
a, b.
PROBLEM SET 9.1
Investigate the paths described by the following pairs of coordinate functions, sketch
the loci, and label a few points as P(O), P( 7T/4), and so on, so as to indicate the way in which
the moving point traverses the locus.
l. x = sec e, y = tan e 2. x = cos e, y = cos2 0 3. x = 2 cos e, y = sin 0
4. x = cos2 e, y = sin2 e 5. x = t3, y = lt31
(Check that not only /(t) = t3 but also g(t) = lt31 have continuous derivatives. Thus a
moving point can go smoothly around a sharp corner, if only it does so slowly enough.)
6. x = sec2 e, y = tan2 0 7. x = sec 0, y = cos 0 8. x = csc 0, y = cot ()

9. x = sin e, y = Jsin 01 10. x = t6, y = t4
y
y
9.2 The Parametric Mean-Value Theorem; L'Hopital's Rule 385
11. In the left-hand figure above, IJ ranges over the open interval (0, 7T) , OR b, and QP =
is a constant a. Find a parametric description of the path, and sketch the locus.
12. In the right-hand figure above, OR = b as before, and QP is a constant a. Find a
parametric description of the path, and sketch the loci, showing the three cases a < b,
a = b, a> b.
13. A circle of radius a rolls without slipping inside a circle of radius 2a. The initial position
is shown on the left below; a later position is shown on the right. Observe that RQ =
2a0, PQ = 2a0, PQ = aij>. Therefore 1> = 20. Let
S = (h, k) = (a cos IJ, a sin 0).

y y
y'
•
2a I
-2a
Then
x' = a cos (0 - if>), y' = a sin (0 - if>),
'
x = x' + h, y = y + k.
Complete the discussion to get a parametric description of the path, and find out what
the locus is. It will turn out that the figure on the right above is slightly misleading.
**14. lf you solved the preceding problem correctly, you found that some of the machinery
that you used was not necessary after all. But consider the case where the outer circle
has radius a and the inner circle has radius b = a/4. Find parametric equations for the
path, and eliminate the parameter to get the rectangular equation
iY� + iY.f = f/�.

This curve is called a four-cusped hypocycloid.
*15. Show that if f is differentiable for x � a, and limx-a ['(x) = k, then f is differentiable
at a, and[' (a) = k. [Hint: The theorem is conceptual, and the proof goes back to first
principles. Start by writing out the hypothesis and conclusion in terms of the basic
definitions of the statements (a) lim;;-a/' (x) = k and (b) f' (a) = k.]
9.2 THE PARAMETRIC MEAN-VALUE THEOREM; L'HOPITAL'S RULE

Given a path described parametrically by a pair of coordinate functions
x = f(t), y = g(t),
we may want to find the slope of the tangent line at the point corresponding to a
particular t.
y
In the figure, we see the path; we want to find the slope of the tangent at P, if such a
tangent exists. Suppose that P is the point corresponding to a certain t; and let Q
be the neighboring point corresponding to t + !it. Let
Lly= g(t + Llt) - g(t) and Llx=f (t + Llt) - f (t),

as indicated in the figure. Then the slope of the tangent at P is
Ll y
m= I.im-,
Ll.t-+O LlX
if such a limit exists. Suppose now that f and g are differentiable, and that f' (t) � 0.
Then we can write
Ll Ll /Llt [g(t + Llt) - g(t)]/Llt

m= lim y= lim y = lim
Ll.t-+OLlX Ll.t-+ollx/Llt Ll.t-+O [f(t + Llt) - f(t)]/Llt
limLl.t-+O {[g(t + Llt) - g(t)]/Llt} g'(t)
_
-
limLl.t-+O {[j(t + Llt) - f(t)]/Llt} j'(t)
Thus we get the formula
g'(t)
m=--.
f'(t)
This will be called the parametric slope formula. We have shown:
Theorem 1. Given a path defined by functions
x=j(t), y= g(t).
If f and g are differentiable at t, and f' (t) � 0, then the path has a tangent at the
corresponding point P, and the slope of the tangent is given by the formula
g'(t)
m= m(t) = .
f'(t)
An important case is the one in which f' (t) � 0 for a < t < b. Here x=f (t)
can never take on the same value twice, and so the locus of the path is the graph of a
function ef;.
Qt=b
p
t=a M=<t>'(x)
I
I
If P and Q are the endpoints of the graph, as in the figure, then the slope of the
secant line through P and Q is
g(b) - g(a)
f(b) - f(a)
By the mean-value theorem,there is a point i where the derivative cf' (i) is the slope of
the secant line. Thus
g(b) - g(a)
= rf'(x).
f(b) - f(a)
This number i must have come from somewhere. That is, there must be a i between
a and b such that
i =
f(i).
It follows that
g'(i)
f'(i) = .
f'(f)
Therefore
g(b) - g(a) g'(i)
= (a < i < b).
f(b) - f(a) f'(i)
What we have just proved is a parametric form of the mean-value theorem. The idea
is that, if a function-graph is presented parametrically, then we can rewrite the
mean-value theorem parametrically, expressing both the slope of the secant and the
slope of the tangent in terms of the parameter.
Theorem 2 (The parametric mean-value theorem). Given two continuous functions
Jandg,fora � t � b. If both functions are differentiable fora < t < b,andf'(t) ¥=
0 for a < t < b, then
g(b) - g(a) g'(i)
f(b) - f(a) f'(i)
for some i between a and b.
y
This theorem takes a simple form when
f (a) = g(a) = 0.
In this case, the theorem says that
g (b)
-- -
g'(i)
f(b) f'(i)'
for some f between a and b. This has the following consequence: if g'(t)/f'(t)
approaches a limit L, as t _,..a, then g(t)/f(t) approaches the same limit L. That is,
if
'(t)
f(a) a lim g L
g( )
0, and =
J'(t)
= = '
t�a
then
(t)
lim g = L.
t-a f(t)
This is called l'Hopital's rule. Roughly, the reason why it holds true is as follows.
Since f is between t and a, we know that t R:> a=> f R:> a. But
'(f)
fR:> a => g
- R:> L,
f'(i)
because g'(t) lf' (t) _,.. L. Therefore
fR:>a=:>fR:>a =>
g(t)
= g'(i) R::3 L.
f(t) f'(f)
Therefore
t
t R::3 a => g( ) R::3 L'
t
f( )
and so
(t)
lim g L.
J(t)
=
t-a
It is very easy to express this idea in the form of an E o proof; all we do is to for
- -
malize our statements involving "R::3" in the following way:
I) Hypothesis. For every E > 0 there is a o > 0 such that
t
O<lt-al<o => \g'( )_L\<E.
t
f'( )
2) Conclusion. For every E > 0 there is a o > 0 such that
t
0 < It-a I <o => \ g( )-L / < E.
f(t)
We need to show that ( 1) => (2). Given E > 0, as in (2), we take the o > 0
furnished by (I). For each t, let f be the f furnished by the parametric mean-value
theorem. This is the o that we need:
0 <It 0
-
al < o ==> < If - al < o
(f)
==>
I g
f''(f) - L i <
€
(t
==>
I g
f(t))
-
L I <
€.
These fit together to give the desired conclusion.

In the above discussion, we assumed that f(a) and g(a) were both defined, and
were = 0. It would have been sufficient, however, to suppose that
1 im f(t) 0 lim (t).

g
= =
t--+a t--+a
If these relations hold, and f and g are not defined at x = a, then we define f (a) and
g(a) to be O;f and g are then continuous, and the discussion of the limit of g/f goes
through exactly as before.
Using x instead of t, we get our theorem in the following form:
Theorem 3 (l'Hopital's rule,first form). If
L'
( )
I im f( x) ( ) 0 g' x
gx
Jim and Jim
f ' (x)
= = =
x--+a x--+a a:--+a
then
lim g(x) .
L
=
x--+a f(x)
Let us now look at some applications. Consider
1. sin x
1m -- .
x--+O X
This satisfies the conditions of l'Hopital's rule, with
g(x) = sin x, f(x) = x.

We investigate the quotient of the derivatives:
1.
cos x
1
Jim =
x--+O
Therefore
1.
sin x
Jim =
x--+O X
This discussion does not supersede the geometric proof of the same statement,
given in Section 4.2. The reason is that to apply l'Hopital's rule, we had to know the
derivative of the sine, and to find the derivative of the sine we needed to know that
lim.,�o [(sin x)/x] = 1. Moreover, if you know the derivative of the sine, you can
remind yourself of what lim [(sin x/x] is, without using l'Hopital's rule. The point
is that
. sin x 1. sinx - sin 0 . , 0,
l Im -- = Im = Sln
x-+O X x-+O X- 0
by definition of sin' 0. Since sin' = cos, and cos 0 = 1, we get the answer imme
diately.
It is not an accident that in applying the first form of l'Hopital's rule, we some
times find that we are merely solving a differentiation problem. The reason is that the
formula used in the definition of the derivative is always an instance of the rule,
whenever the function is differentiable. By definition,
F(x) - F(x0)
F'(x0) 1
- .Im
_ .
x-+x0 X - X0
The indicated limit on the right satisfies the conditions of Theorem 1, with
g(x) = Fx
( ) - F(x0) --+ 0, f (x) =x - x0 --+ 0,
asx --+ x0. Thus every differentiation problem is a problem of the sort that l'Hopital's
rule deals with. The rule, of course, applies in many other cases; and it is the other
cases that make it significant. For example,
2
1. sin x+x 1. 2sinxcosx+l
Im = 1m =l
. ,,
"' "'
x-+O e - 1 x-+O e
by the rule; and here the rule is needed.

Often, the application of l'Hopital's rule requires the use of some preliminary
device. For example, consider the possible limit
lim xcot x.
x-+O
Here we should start by writing

. xcosx
1im ---
x-+o sinx
and then use Theorem 3 ( unless we can think of something simpler ).
PROBLEM SET 9.2
Investigate the following indicated limits. (That is, calculate the ones which exi st.)
sin2 e sin3 e
1. limx cotx 2. lim 3. lim
82 82
x-o o-o o-o
cos2x - 1 lnx x- I
4. lim 5. Jim -- 6. lim -.,--
x -o x2 x-1
X - 1 x-1
e - 1
e"' - 1 . y2 - 2y + 4 cos3x - 1
7. Jim 8. ltm 9. Jim
.,_0 In (x + 1) y-2 y- 3 .,-,,/2 x3 - 1
1 x4 - 1
10. Jim x2 sec2 x 11. limx2 sin 2 12. Jim . 2X - 1
00-+rr/2 x-o x x-rr
Sill
ln2 x - 1 sin x cos x - tan x tan x

1 3. Jim 1 4. Jim 1 5. Jim --
x-e x2 - 1 x-o v:X x-o x
sine - cose x2 - 4x + 3
1 6. Jim 1 7. Jim x2 csc2 x 18. lim
e-rr/4 e - Tr/4 x-o x-+1 X
2 - 3x + 2
x2 - 1 3x
1 9. Jim .,-
20. Jim -- . -
X x-oe
---
x-l
In 1 -
(t)
21. Jim
t-e
In
t -
- 1
e
22. Jim
t-rr/2
[ f
sect
rr/2
Vl + sin3 t dt ]
23. Jim
x-1
[�· J,"'
X 1 1
e13 v( dt] 24. Jim
x-rr
[ l"'
csc x
rr/2
Vl + sin3 t dt ]
*25. Jim x In x 26. Jim x In sin x
CC-+O+ X-+0+
*27. Jim x In (sin x cos x) *28. Jim x In (cos2 x sin2 x)
a;_.o+ x-o+
29. A circle starts off tangent to the x-axis at the origin. The circle then rolls, without slip
ping, along the x-axis. The point P which started at the origin then traces out a path;
this path is called a cycloid. (The same term cycloid is applied also to the locus of the
path.) The parameter in the coordinate functions is thee indicated in the figure. Sketch
the locus, and calculate the coordinate functions of the path.
y y'
•
I
I
P=(x, y) =(x', y')'
As the figure suggests, the easiest method is to use a "moving coordinate system,"
as in Problem 13 of Problem Set 9. 1 ; we need to calculate the coordinates (h, k) of the
"moving origin" O', and calculate x' and y' as a cos </> and a sin ¢. (What is ¢ ?)
30. a) When a circle rolls on the inside of another circle, we get a hypocycloid. In the figure,
the fixed circle has radius a and the rolling circle has radius b.
y
Calculate the coordinate functions, using e as the parameter. The answer is
x = f(O)
b-a
= (a- b) cos e + b cos - - e,
b
y = g(O)
b - a
= (a - b) sin e + b sin - - e.
b
b) Get a rectangular equation for the locus, for the case b = a/4. Sketch.
31. When one circle rolls around the outside of another, the figure traced out is called an
epicycloid. Derive parametric equations for the epicycloid, using radius a for the fixed
circle and radius b for the moving circle. Use the same parameter e as in Problem 30(a).
32. Suppose that a railroad wheel rolls (without slipping) along a flat track. Find coordinate
functions for the path traced out by a point at the outer edge of the flange on the wheel.
In the figure below, the outer radius is b and the inner radius is a. Sketch the locus,
bearing in mind that it is not a function-graph; it has loops in it.
33. Make the same modification in the definition of the epicycloid, as suggested by the
figure below, and sketch the curve. The fixed circle has radius a; the rolling wheel
has inner radius b and outer radius c.
*34. A path is regular if the coordinate functions/, g are differentiable, and we never have
'
f' (t) = g (t) = 0 for the same t. Show that every chord of a regular path is parallel to
9.3 Other Forms of L'Hopital's Rule 393
the tangent at some intermediate point. Examples of this are as follows:

y y
(Evidently the locus need not be a function-graph, and the chord may be vertical.)
*35. Given a path, with differentiable coordinate functions/, g. Show that, if the axes are
rotated, then the coordinate functions F, G that work for the new set of axes are also
'
differentiable. Show that if f' and g never vanish simultaneously, then F' and G'
never vanish simultaneously.
9.3 OTHER FORMS OF L'HOPITAL'S RULE
The first f orm of l'Hopital's rule says that if

(x)
lim/(x) = lim (x) = 0, and lim g' L'
g
=
x-+a x-+a x->a j'(x)

then
(x)
Jim g L.
x->a f(x)
=
This can be generalized in three ways.
1) x--+ oo or x--+ - oo, instead of x--+ a, it doesn't matter;

If the rule still holds.
2 ) If g'(x)lf'(x)--+ oo or_,. - oo, as x _,.a, the rule still holds.

3) If f(x) _,. oo and g(x)--+ oo, instead of f(x)--+ 0, g(x)--+ 0, the rule still holds.
Similarly iff(x) --+ - oo and g(x) --+ - oo.
Thus, in the most general f orm ofl'Hopital's rule, we have: (1) x _,.a, x _,. oo, or
x--+ - oo; (2) g'(x)/f'(x)--+ L, g'(x)/f'(x) _,. oo, or g'(x)/j'(x)--+ - oo; and (3)
f(x), g(x) --+ 0, or--+ oo, or_,. - oo. Thus we have a grand total of 27 theorems, all of
which are true. One of these has already been proved, and the only hard one among
the others is the following.
Theorem I (The Northeast Theorem). If

(x)
lim f(x) = lim (x) = oo, and Jim g' L'
g x->oo j (x)
=
'
x-oo x-toc,
then
(x)
Jim g = L.
x->ro J(x)
This is proved in Appendix H. Meanwhile we shall use it.

Example I . To find
1. In x
im-,
X-+O'J X
we take derivatives and find
1/x
lim- = 0.
X-+00 1
By the Northeast theorem,
ln x
lim = 0.
X-+ 00 X
The case in which g' (x)Jf'(x) � oo causes no trouble.
Example 2. To find
"'
e
Jim - ,
X-t-00 X
we investigate
"'
e
lim - = oo.
X-+ 00 1
It follows, by one form of l'Hopital's rule, that

"'
e
Jim - = oo.
X-+00 X
The theorem being applied here is the following.
Theorem 2. If
. g'(x)
limf(x) =Jim g(x) = oo, and hm -- =co,
X-+00 X-+00 x-+oo f'(x)
then
g(x)
lim = oo.
x-+oof(x)
The proof is easy, on the basis of the Northeast theorem; we merely investigate
reciprocals. Since g'(x)Jf'(x) � co, we have
f'(x)
Jim = 0.
X-+00 g'(x)
By the Northeast theorem,
f(x)
lim = 0.
X-+00 g(x)
And f(x)/g(x) > 0 when x is large. Therefore
g(x)
lim =co.
x-+oof(x)
Example 3. Consider
. In x 1. - In x
I1m-=-1m -- .
x-+O+ X ., .... o+ x
9.3 Other Forms of L'Hopital's Rule 395
Here the limit on the right takes the form /

oo oo. Taking derivatives, we find
-1/x
lim -- =0 .
., ... o+ 1
By one form of l'Hopital's rule, it follows that
-In x = ,
I"Im+ -- O
X->0 X
and so the answer to the original problem is -0 = 0. The theorem being used here
is the following.
Theorem 3. If
(x)
lim f(x) = lim g(x) = oo, and lim g' = L'
x-+a+ +
x-+a x->a+ f'(x)
then
(x)
lim g = L.
x..,a+ f(x)
This is not quite as easy as Theorem 2.

Let
1 1
y= -- , x=a+-,
x - a y
so that y ---+ oo as x ---+ a+ and x ....... a+ as y ---+ oo. Then
lim
g(x) =lim g(a + l/y).
x-.a+ f(x) y-.oof(a+ 1/y)
Taking derivatives, we find
1.
Im g'(a+ l /y)(-l/y22) = Im
1. g'(a+ 1/y)
v->oo f'(a+ 1/y)(-1/y ) y-.oo f'(a+ 1/y)
. g'(x)
= hm -- = L.
x-.a+ f'(x)
The Northeast theorem now applies to
].
1m g(a+ 1/y)
'
Y->00 f(a
+ 1/y)
and tells us that this limit is L. Therefore
(x)
lim g = L.
x-.a+ f(x)
We have now discussed all the troublesome cases of l'Hopital's rule; once we
have gotten this far, the rest of the derivations are routine. Hereafter, we shall use
all forms of the rule without comment.
Sometimes we can apply the rule by taking logarithms. Consider
lim x"' = ?
.,_, 0
Let cf>(x) xx. Then
=x = 1/xx = f(x)
=
In g(x)
In cf>(x ) In x - -
.
Now
.
x-+o+
g'(x)
Ilffi --
f'(x)
=
1·
lffi
x-+O
--
1/x
-1/x 2
== 1·lffi ( -x )
x-+o+
= 0 .
..-. =
Therefore
Jim In cf>(x) = 0, and limxx e0 = 1.
x-o+ x o+
PROBLEM SET 9.3
2. Jim y (Tan-1 y - 7T/2) 3. Jim (In Jn x)/ v:X
5. Jim e-1/x' 6. Jim [(l/x)e-1/x2]

x-o+ x-o+
7. Jim [(sin x)(In x)/x] 8. Jim (1 + Tan-1 o:)Tau-ia 9. Jim ( l + csc x)•in' x
a--+ co x-o+
10. Jim (I + ax)1fx 11. Jim x3e-x 12. lim xe-1/x

x-o X-+00 X---+00
( � ))Tan-'
x
13. Jim (l - 2x)1fx 14. Jim 1 + Tan 1 (x 15. Jim x2e-11x
x-'"12 a:- o+
16. Jim (tan fJ)lano

o-o+
17. Jim
t-oo
(�)1 t
18. Jim x2 ln x
x-o+
19. Jim (e-lfx'/x2) 20. Jim (e-lfx' /x3) 21. Jim x Jn x

x---+O x-o+
22. Jim (1/x + In x) 23. Jim (l/x + n Jn x) 24. Jim ( l/x2 + /1 ln x)
x-o+ x-o+ x-�o+
25. Show that for every n, Jim (e-1/x' /xn) 0.
x)sin
=
x-o
26. Jim (sin x 27. Jim (1 - COS y)l-COS y

x--o+ v-o+
28. Jim (1 + COS y)l-COS Y 29. Jim (1 + sin kx)csc x

v-o+
30. Jim (I - sin

x-o
kx)csc x
x-o
31. Jim (v2)"

v-o+
'
32. Jim ww 33. Jim (I + tan 3 fJ)-CSC 9
Tf'-o+ o-o
{
*34. The nth derivative of a function .f is denoted by ['"1• Let
e-1/x2 for x ;e 0,
f( x ) =
0 for x =
0.
Show that for each n > O,fhas an nth derivative, for every x; and show that J<">(O) 0 =
for every n. [Hint: You are not likely to find a manageable general formula for J<n>(x).
But you ought to be able to show that for x ¥- O,J<">(x) is always given by a formula of
a certain form, involving certain constant coefficients; and you may be able to use thi5
form, to show that ['"1(0) 0, without needing to determine the coefficients.]
=
9.4 Polar Coordinates 397
9.4 POLAR COORDINATES
When we set up a rectangular coordinate system in a plane E, to every ordered pair

(x, y) of numbers there corresponds a point P of E. Thus we have a function
{(x, y)} � E, (x, y) f--+ P.
And the correspondence also works in reverse: when P is named, x and y are deter
mined. Thus a rectangular coordinate system gives us a one-to-one correspondence
{(x, y)} +--+ E between the ordered pairs of real numbers and the points of E.
We now consider another way of labeling points with pairs of numbers.
Given two numbers rand 8, we first draw the ray which starts at the origin and
has direction 8. On the line containing this ray we set up a coordinate system, with
the direction 8 as the positive direction; and we let P be the point with coordinate r.
(This is equivalent to saying that the directed distance OP is = r.) We then say that
P has polar coordinates (r, 8).
For example, in the left-hand figure it looks as if P1 has polar coordinates (2, 7r/3),
and P2 has polar coordinates (-2, 7r/3).
1r
y
2
��,f---__,___,_��-'---- x
2
I
I
I
I
I
I
I
I
P2(r,8)� (r< 0)
I
Thus to every pair (r, 8) of numbers there corresponds a point P. But the corre
spondence does not work uniquely the other way: every point P corresponds to
infinitely many number pairs (r, 8). Thus, in the right-hand figure, the point P with
rectangular coordinates (1, 1) has polar coordinates (._}2, 7r/4). But Palso has polar
coordinates ( -/2, 57T/4). And this is not all; the possible polar coordinates for Pare
cI2, 7r/4 + 2n7T), c-.J2, s7T/4 + 2n7T),

where n is any integer (positive, negative, or zero).
Thus, when we set up a polar coordinate system we have a function
{(r, 8)} � E,
but we do not have a one-to-one correspondence; the polar coordinates of a point

are not determined when the point is named. For this reason, graphs in polar co
ordinates can naturally be thought of as paths. Let us look at some examples.
1) Consider the graph of
r = cos 8 co� 8 � 27T).

Since the cosine is periodic, with period 27T, we can get all of the locus by restricting()
to the interval [O, 27T].
As an aid in sketching the polar graph, we first sketch the rectangular graph of the
equation r = cos e. We cut the curve into four parts, as indicated, and then sketch the
portion of the polar graph corresponding to each of them. As () increases from 0 to
7T/2, r decreases from I to 0. As () increases from 7T/2 to 7T, r decreases from 0 to -1.
Therefore the second part of the curve, in the fourth quadrant, comes from values of()
in the second quadrant. (See the figure on the left.)
,..
,.. 2
2
3
' 4
' 3,..
3,..
2
2
As() continues to increase, from 7T to 27T, we trace out a curve shown on the right.
This looks like the curve that we had already. And in fact it is exactly the same curve
as before, because
cos ce + 7T) = -cos e.

Further investigation shows that the graph is a circle.
If Phas polar coordinates (r, ()), then the rectanguar coordinates of Pare
x = r cos e, y = r sine.
(There are two cases to check. If r > 0, then these formulas follow from the defini
tions of the sine and cosine. Verification for r = 0 and r < O?) Therefore
x2 + y2 = r2 cos2 e + r2 sin2e = r2 .
This gives the three conversion formulas
x = r cose y = r sine, x2 + y2 = ,2,
We note that the equation

r = cose
does not involve any of the three expressions r cose, r sine, r2 which we know how
to convert into rectangular coordinates. But we can multiply by r, on both sides,
getting
r2 = r cose;
and this means that
x 2 + y2 = x.
This is the circle with center at the point (t, 0) and radius t.
2) Consider
r = sece.
Here we might sketch the graph without using rectangular coordinates.
It is easier, however, to multiply both sides of the equation by cose. This gives
r cose = I, or x = I.
Ase increases from 0 to 27T (skipping 7T/2 and 37T/2), this line is traversed twice. (It
is worthwhile to figure out how.)
3) In these two examples, it was easy to work back to a rectangular equation.

Consider, however,
r = 1 - sine co� e � 27T).
(As for r = cos(), the interval [O, 277] gives us the entire locus of the path.) First
we do a rectangular sketch as on the left. We then sketch the polar graph in four
parts:
11'
3,,.
2
This curve is called, for obvious reasons, a cardioid.

It is possible to write a rectangular equation for the cardioid. First we observe
that since
r = 1 - sin () ·;;; 0, (1)

we always have (on this particular curve)
r = J--;2 = Jx2 + y2.

We can therefore write
r2 = r - rsin(), (2)
x2 + y2 = Jx2 + y2 _ y. (3)
4) To get the equation of a line L, in polar coordinates, we proceed as follows.
Let N be the perpendicular to L through the origin. Rotate the axes through an angle
of measure cp, choosing cp so that N becomes the x'-axis.
!J
Then L is the graph of an equation

x' - p = 0,
where p is a constant. Since x

'
= x cos - p = 0.
Converting to polar form, we get

r cos e cos cf> + r sin e sin cf> - p = 0,
r cos ee - cf>) - p = 0.
This is the standard form of the equation of a line in polar coordinates.

Some geometric problems are most conveniently attacked by introducing polar
coordinates at the outset. To do this, we need a distance formula.
Theorem I. Let d be the distance between the points with polar coordinates (r1, e1)
and (r2, e2). Then
d2 ri + r� - 2r1r2 cos eel - e2).
=
Proof The rectangular coordinates of the two points are
for i = 1, 2. Therefore
2 2 2
d = (x1 - X2) + eY1 - Y2)
2 2
= erl cos el - '2 cos e2) + er1 sin el - '2 sin ez)
2 2 2 2
= riecos el + sin el) + r�ecos e2 + sin ez)
- 2r1r2ecos el cos e2 + sin el sin e2)
= ri + ,.� - 2r1r2 cos eel - e2),

For r1 > 0, r2 > 0, 0 < e1 - e2 < TT, this is simply the law of cosines.
But the polar distance formula applies for any values of r1, r2, e1, and e2•
PROBLEM SET 9.4
Sketch the following, and convert to rectangular coordinates if possible.
I. r=2 csc0 2. r = -2 sec 8 3. r cos (fJ - 7T/4)
4. r =1 +sin 8 5. r =1 - cos 8 6. r=sin 8 sec2 8
7. r2 =sin 8 sec3 8 8. r = 9. r=sin 28

sin 8 +cos 8
10. r = sin 30 11. r =sin 48 12. r =1 + r cos 8

36
13. r = 14. r2= 15. r = 1 +csc 8
1 +cos0 4 cos2 8 +9 sin2 8
16. r sin 0 17. r = 1 - sin0 18. r=cos 38
19. r= 20. r2 =sin2 8 21. r2 =sin0

sin 8 - cos 8
r
22. r = 2 +cos 8 23. = 2. 24. r =eBI�
1 +r cos 0
25. The figure given in the text suggests that at the origin, the two sides of the cardioid
have the same tangent, namely, the line 0 = 7Tj2. Show that this is correct.
Discuss, as in Problems 1 through 24.
26. r2 = cos2 0 27. r2 = cos 8 28. r2 = cos 28
29. r2 = a2 cos 28 (This curve is called a lemniscate.) 30. r2 =a2 sin 20
Find polar equations for the curves defined by the following conditions, and sketch.
Identify the curve if possible.
31. The set of all points which are equidistant from the origin and the line r = csc0.
32. The set of all points which are equidistant from the origin and the point (2 v2, 7T/4).
33. The set of all points P such that PA = 2PB, where A is the origin and B = (2v2, 7T/4).
34. The circle with center at (2, 7T/4) and radius 2.
Sketch:
35. r=2 - sin 8 36. r = 3 - 2 sin 0 37. r =3 + 2 cos 8

r 1
38. =- (What sort of curve is this, and why?)
1 +r cos 0 2
9.5 AREAS IN POLAR COORDINATES
Given
r = f(8) � 0,
where f is continuous, and the length of the interval [O'., /9] is ;;; 2rr. Let R be the
region between the polar graph and the origin.
9.5 Areas in Polar Coordinates 403
7r
2
That is,
R = { (r, &) I o: ;;; e ;;; f3 and 0 ;;; r ;;; f (&)}.
Consider a subinterval [&;_1, &i] of the interval [o:, {3]. Let mi be the minimum
value off on [&;_1, &;],let M; be the maximum value, and let flA; be the area of the
region between the origin and the part of the curve from e ei-1 to e e i. = =
The area of the inner circular sector, with radius m;, is
and the area of the outer sector, with radius M;, is
-�M; fl&;.
Therefore
Now take a net
over the interval [o:, {3]. The area of R is

n
A= .2flA;.
i=l
The above inequalities hold for every i; and so by addition we get

n n
I �-m; 6.8; � A � I iM; 6.0;.

i=l i=l
But the sum on the left is the lower sum s(N) of the function
F(8) = }/(B)2,
over the net N, and the sum on the right is the upper sum S(N), of the same function
F, over the net N. Thus
s(N) � A � S(N);
and so
Jim s(N) � A � Jim S(N).
I s1-o J.YJ-o
Since
lim s(N) (P U(8)2 dO lim S(N),

J,
= =
JXJ-o J.\·J-o
it follows that A is squeezed, and
A = fHC0)2 dO.
Thus we have:
Theorem 1. Let f be continuous and � 0 on [?:, (3], with (3 ?'. � 21T, and Jet - R be
the region between the origin and the polar graph off Then the area of R is
A = f §f(fJ)2 dO.
Let us try this in some simple cases. For the circular region with radius a and
center at the origin, the formula gives
A = i2"ta2 dO = -�a2 21T = 7Ta2,

·
which is the right answer. For

1
r = -----
' o � 8 � 7T/2,
cos 0 + sine
we get
A=
1"12 1
- ·
1
. dO
o 2 (cos 0 + sin 8)2
=
1
-
l;;/2 1
c1e
2 o 1 + sin 20
=
1
-
["12 l - sin 20
d6
2.o cos220
= Mt tan 2e - �- sec 28]�12
=
}(O + t) - �(O - i) =
�·
9.6 The Length of a Path 405
This is correct, because the region is a right triangle with legs of length 1.
rcosB+rsinB=l; x+y=l; A=t.
PROBLEM SET 9.5
Find the areas of the regions enclosed by the following curves, and sketch.
1. r = 1 - sin 0 2. r = 1 - cos() 3. r = cos()
4. r2 = cos2() 5. r2 = cos 20 6. r =sec() tan fJ, 0 � () � 1T/4

7. r = sec fJVtan fJ, 0 � () � 1T j4 8. r = sin() cos() 9. r2 = 4 cos 2()
10. Find the area of the inside loop of the graph of r = 1 - 2 sin fJ, and sketch.
1
11. r = 12. r = e0, 0 � 0 � 21T 13. r = e28, 0 � () � 21T
lcos fJI + lsin 01
14. Given a polar graph defined by a differentiable function r = f(O) ( <X � () � {3), derive
a formula for the slope of the tangent, at a point (r0, fJ 0) (r0 =f(fJ0)). Here we really
mean the slope, relative to a rectangular coordinate system superimposed on the polar
coordinate system.
9.6 THE LENGTH OF A PATH
Roughly speaking, the length of a path is the total distance traversed by the moving
point. For example, consider the path defined by the coordinate functions
j(t) = a cos t, g(t) = a sin t (0 � t � 47T).
The locus of this path is a circle with radius a and circumference 27Ta. But as t increases
from 0 to 47T, this locus is traversed twice. Therefore the length of the p ath is 2 27Ta
· =
41ra.
Lengths of paths are undirected; they are always positive (or zero, in trivial cas es) .
Thus the length of the path
x = f (t) = cos t' y = g(t) = 0
is four, not zero; the two halves of the path do not cancel each other out.
To be exact, path length is defined as follows. Given a path
x = f(t), y = g(t) (a� t � b),

let
be a net over [a, b]; for each i from 0 ton, let
Yi = g(ti),
Then
is the length of an inscribed broken line.
Po
The path length, by definition, is

n
s = Jim 2; P;_1P;.
IN1�0 i=O
We can express the path length as an integral. For each i, let
X; = f(t;), Y.; = g(t;),

.6.x; = X; - X;_1,
as indicated in the figure.
:I _____ _
Xi-1 Xi
Then
9.6 The Length of a Path 407
Since
and
we know by the mean-value theorem that
6.xi = f' (ii) 6.ti•

for some f; between ti-i and ti; and
for some r; between ti-l and ti. (We do not know that f; = i;, and this leads to trouble,
as we shall see.) Therefore
and so
i=l
i -J f'(ii)2 +
i pi-lpi i=l = g'(f/)2 6.ti.
This is almost, but not quite, a sample sum of the function
ix(t) = -Jf' (t)2 + g' (t)2;

it differs from a sample sum in that we have used two different sample points f;, i;
on each interval [ti_1, t;] of our net N. Since INI --+ 0, and g' is continuous, we ought
to have
i -Jf'(f;)2 +
i=l
g'(f;)2 6.ti � i -Jf'(i;)2 +
i=l
g ( f;)2 6.t;
' = i t1.(i;) 6.ti
i=l
� f-J f'(t)2 + g'(t)2 dt
when INJ � 0. For a proof of this, see Appendix I. Meanwhile we shall state the
following theorem and use it.
Theorem 1. If the coordinate functions f, g of a path have continuous derivatives

'
f', g , then the length of the path is
s = f-Jf'(t)2 + g'(t)2 dt.

�
This formula can be converted to polar coordinates in the following way. Suppose
that a polar path is described by a function
r = cp(6) (a � e � b).
The rectangular coordinate functions of the path are then
x = /(6) = c/J(6)cos e, y = g(6) = cp(6)sin6.

This gives
'
j'(6) = cp'(6)cos6 - cp(6)sin6, g (6) = f (6)sin6 + cp(6)cos6.
For short, let us write </>' for </>' (e), c for cos e, and s for sin e. This gives
and
Thus we have the following theorem.
Theorem 2. Given a path defined in polar coordinates by a function
(a� e � b),
where </>' is continuous, the length of the path is
PROBLEM SET 9.6
It is hard to propose reasonable problems in the calculation of path length; sometimes

the integral takes a troublesome but manageable form such as S V 1 + x2 dx, but most of
the time, path length problems are either easy or impossible. Therefore, if some of the
problems below look impossible, you should try to think of an approach that might make
them easy.
Find the lengths of the following paths.

I. r = cos(), 0 � () � n 2. r = eO, 0 � () � 2n
1
3. x = 0, y = cost, 0 � t � h/2 4. r = . , 0 � 0 � n/2
cos 0 + sm 0
2
5. r = 2 sin 8, 0 � () � r. 6. r = n/2 � 0 � n
cos() - sine '
1
8. ,. = 0 0 � () � 2r.
e
,
9. r = sec fJ tan 0, 0 � 0 � 7T/4 (Remember the above remarks.)
10. x = 1 + sint, y = 1 - cost, 0 � t � 7T/4

1 1. x = cos3 t, y = sin3t, 0 � t � 1Tj4 (What sort of curve is the locus of this path?)
12. x = t3, y = t
l 3I, -1 � t � 1 (Do these coordinate functions satisfy the conditions of
Theorem 1? That is, does g (t ) = lt31 have a continuous derivative?)
'-'13. The proof of Theorem 1 would have been much easier if we had been able to use the
following:
Theorem (?). For each i, there is a single point f;, betweent;_1 andt;, such that
pi_lpi = ..; f'(i;)2 + g'(i;)2 D.t;.
We could then have expressed P;_1P; as a sample sum, and passed to the limit, as
in Section 7.1. But the above theorem is false. Give an example of a path ( withf' andg'
continuous) for which the theorem fails. There is a very simple example of this kind.
9.7 Vectors in a Plane 409
9.7 VECTORS IN A PLANE
In Section 3.8, we found that the motion of a particle on a line could be described by
a single functionf, with real numbers as values, and that the velocity and acceleration
functions were the first and second derivatives
v =f' and a= v
I
= f" .
As we remarked at the time, these ideas are not adequate to describe the motion of a
particle in a plane (or in space). The motion of a particle in a plane Eis described
by a path, which is a function
P: I---+E
: t H P(t),
where I is an interval, and P(t) is the location of the moving particle at time t. Velocity
in this case is a "vector quantity," with both a magnitude and a direction, conven
iently pictured by an arrow. At each point P(t), the direction of the velocity vector is
the direction of the motion, so that the arrow always lies on the tangent line, pointing
in the appropriate direction on the tangent line; and the length of the velocity vector
is the speed.
y
This is the idea. We need to express it in a mathematical form in which it can be

used. The idea of a vector appears in a variety of forms. The simplest of these is as
follows.
--+
With each point P of the plane we associate the directed segment OP, starting
--+
at the origin and ending at P. Such a directed segment OP will be called a vector.
y
--+
We allow the "degenerate segment" 00; this is called the zero vector1 and may be
-+
denoted simply by 0. Moreover, since all our directed segments, in this section, are
--+
going to start at the origin, we can denote the directed segment OP by the shorter
-+
symbol P. Three operations can be performed, in this system:
-+ -+
Addition. Given P1, P2, with P1 = (x1, Yi) and P2 = (x2, y2), the sum is defined to be
where
Vector addition is governed by the same formal laws that govern addition of real
numbers, as follows.
-+ -+ -+ -+ -+ -+
A.1 Associativity. (P1 +P2) +Pa = P1 + (P2 +Pa).
- -+ -+ --+ -+ - -+
A.2 Existence of 0. There is a vector 0 such that 0 +P = P + 0 = P for
-+
every vector P.
-+ -
A.3 Existence of negatives. For each vector P there is a vector -P such that
- -+ -+ --+ -+
P + ( -P) = ( -P) +P = 0.
-+ -+ -+ -+
A.4 Commutativity. P1 +P2 = P2 +P1.
These follow from the corresponding laws for real numbers. For example, if
-+ - - -+ -+ -+ -+ -+
(P1 +P2) +P3 = Q, and P1 + (P2 +Pa) = Q',
then
Q = ((x1 + Xz) + X3, (Y1 +Y2) +Ya)
'
= (x1 + (x2 + Xa) , Y1 + (Y2 +Ya))= Q ,
-+ -+ -+ -+ � -
and so Q = Q'. The existence of 0 is obvious: 0 is 00. If P = (x, y), then -P =
-
Q, where Q = (-x, -y). Similarly for A.4.
Scalar multiplication. When we are discussing vectors, we refer to real numbers as

-
scalars. To multiply a vector P by a scalar a, we multiply the coordinates of P by a.
That is,
- -
aP = Q,
where
Q = (clX , ay).
We then have a kind of associative law.
- -
M.1. (a{J)P = a({J]>).
Because (a{J)x ==; a({Jx), and (a{J)y = a({Jy). Multiplication is connected with
vector addition by two distributive laws.
- - -
M.2. (a+ {J)P = aP+ {JP.
- -+ -+ -+
M.3. a(P1 + P2) = aP1 + r1.P2.
Zero and 1 work in the usual way:
- - -+
M.4. 0 · P = 0, for everyP.
- - -
M.5. 1 · P =P, for every P.
- -
M.6. a· 0 = 0, for every a.
-+
Let "f/' be the set of all vectors P. In "f/' we have defined two operations (addition
and scalar multiplication), and shown that they satisfy the laws A. l through A.4 and
M. l through M.6; "f/' is called a vector space (relative to these two operations). More
generally, any collection "f/' of objects is called a vector space if it is provided with two
operations satisfying the above formal laws. There are many important vector spaces
other than the one which we are now discussing. For example, we may consider the
- ---+
directed segments P = OP, starting from the origin in three-dimensional space, with
the two operations defined in an analogous way.
Finally, we introduce another kind of multiplication for vectors, called the
dot product or inner product. If P1 = (xi. y1) and P2 = (x2 , y2), as before, then the
inner product is a scalar, namely,
+ +
P1 . P2 = X1X2 + Y1Y2·
The following properties of this operation are easy to check:
-+ -+ -+ ->-
S.1. P1 P2 · = P2· P1.
- - - -+
S.2. (aP1)· P2 = a(P1 P2).
·
- - -+ -- -+ -+
S.3. P1 (P2 + Pa)

· P1 · P2 + P1 = · Pa.
- - -
S.4. P · P � 0, for every P.
-+ -+ -+ -+
S.5. If P · P = 0, then P = 0.
-+ -+
(The last condition rules out trivial "dot products" for which P1 P2· is always 0,
-+ -
for every P1, P2.)
Thus, "f/ is called an inner product space (relative to the three operations which
have now been defined). More generally, any collection 1/ is called an inner product
space if it is provided with three operations (addition, scalar multiplication, and
inner product) satisfying all the above laws.
As a matter of convenience, we have defined our three operations algebraically,
using the coordinates (x, y) of the terminal points P of the vectors. But it is important
to understand that all three of them have geometric meanings. We can add two
vectors, geometrically, by completing a parallelogram, as shown on the left.
To do this, we don't need to know the directions of the x- and y-axes. Therefore the
-+ -+
axes can be, and h�ve been, omitted from the figure. lf P1 and P2 are collinear, then
the parallelogram collapses, but the idea is the same.
-+ -+ -+
Geometrically, - P is the vector Q which has the same length as P, but has the
opposite direction.
-+
To multiply a vector P by a positive scalar IX, we draw a vector with the same
-+
direction as P, but multiply the length by x. If(/. < 0, we go in the opposite direction,
and multiply the length by j(/.j.
Q Q = -? Q = aP
The geometric meaning of the inner product is less obvious. Algebraically,
-+ -+
P1· P2 =
X1X2 + Y1Y2·
Under the conditions given in the figure,
Substituting cos el = X1/0P1, sin el = Y1/0P1, cos 82 = X2/0P2, sin 82 = Y1/0P2,

we get
so that
-+ -+
P1 · P2 = OP1 · OP2 cos 8.
Obviously cos 8 is independent of the directions of the axes, because () measures the
-+
angle between the two vectors. Note that the length of the vector P can be expressed
in terms of the dot product:
-+ +
p . p = x2 + y 2 = OP 2.
-+ +
The length of a vector P may also be denoted by IP!. Thus
IPI
� j+-+
= P·P.
-+-+ -+
By a linear combination of two vectors P1, P2 we mean a vector Q which can be
expressed in the form
where rx and f3 are scalars. In a coordinate plane, it is easy to find two vectors i and j
such that every vector is a linear combination of them. If the vectors i and j are as in
the left-hand figure below, and P = (x, y), then
-+
P =xi + yj (i = (1, 0), j = (0, 1)).
This is an equation between vectors, not numbers. On the right, we have multiplied
the vectors i and j by the scalars x and y, and added the resulting vectors.
y
y
0 1 x
This section contains no new information, but quite a lot of new language.
Learning a language takes practice. Therefore, while some of the problems below
are genuine problems, many of them are merely exercises in the process of translation
from the language of coordinate s y ste m s to the language of vectors and back again.
PROBLEM SET 9.7
Sketch the set of all points P satisfying the following conditions.
I. P = od ( - oo < C'I. < w) 2. P = r.tj ( - w < C'I. < w )

3. P = r.ti + r.tj (-w < C'I. < w) 4. P = r.ti·- r.tj ( - w <et. < w )
5. p . i 0 = 6. p . j = 0
.
7. p (i + j) = 1 8. p . (i + j) = 0
9. p . p 1 = 10. p. p = 0
11. J3 = r.ti + 2et.j co � C'I. � v3) 12. p. i = }
13. p. j =
t 14. P = cxi + r.t2j ( - O'J < C'I. < oo)
15. P = 2r.ti - C'l.j (-w < C'I. < w) 16. p = et.i + r.t3j ( - O'J < C'I. < O'J)
17. P(i + 2j) = 3 18. p = C'l.2i + r.t3j ( - Ct:) < C'I. < (/J)
19. P = r.tj + r.t2i (-w < r.t < co) 20. p . (i + 2j) = 0
21. p . (2i + j) = 0
22. Let c = i + j, d = i - j. Express i as a linear combination of c and d. (To do this
you will need to calculate with vectors, by the same processes that you use with real
numbers. This can be done; and this is why we stated and verified the laws A. I through
A.4 and M.1 through M.6.)
23. Express j as a linear combination of c and d.
24. Now show how any vector P can be expressed as a linear combination of c and d.
25. Let e = i + 2j, f = 2i - j.
a) Express i as a linear combination of e and f.
b) Express j as such a linear combination.
c) Show how every vector P can be so expressed.
26. Same problem, for e = i - 2j, f = 3i + 2j.
27. The vectors g and h span the vector space Y if every vector in Y is a linear combination
of g and h. (Thus in Problem 25 (c) you showed that e and f span Y.) Is it true that
every pair of vectors in Y span Y? Why or why not?
9.8 Free Vectors 415
28. Let P1 = (2, 1), P2 = (1, 2). Sketch the set of all points P such that
(0 ;;;i et: ;;;i I).

-+ -+
29. Let P1 and P2 be any two vectors (by which we mean two different vectors). Sketch the
set of all points P such that
- -+ -+
P = et:P1 + (1 - rx)P2 (0 ;;;i et: ;;;i I).
Sketch the set of all points P satisfying the following conditions:
30. p . i �0 31. p . j � 0
32. p . (i + j) � 0 33. p . (i - j) � 0
34. p = et:i + {i'j ( et: � 0, fi' � 0) 35. p = et:i + {i'(i + j) (o: � 0, /! � 0)
36. p . (i + 2j) �0 37. p . (i - 2j) < 0
*33. Let "ff/' be the set of all continuous functions on the interval [ -1, I]. State, for the
functions in 1/1, definitions of (a) addition, (b) scalar multiplication, and (c) inner
product, in such a way that 1f/' forms an inner product space. Verify that under your
definitions, the inner product space laws are all satisfied. (There is only one reasonable
definition for (a), and similarly for (b); but the "right" definition of the inner product is
less obvious. Hint: The""' operation is supposed to assign a numberf g to each pair ·
of functions j; g. Under what significant operation does a number correspond to one

function? As a check on your definition, it should turn out that if/ (x) = x3, g(x) = 1,
h(x) = x, then.f·g = O,g · h = 0, and/· h = %.)
This inner product space has important uses, later in the theory of functions.
9.8 FREE VECTORS

�
In the last section, we defined a vector to be a directed segment OP, starting at the
origin. We shall now introduce a different form of the vector concept, which for
some purposes is better.
By a translation of a coordinate plane we mean a correspondence of the form
x � x + h, y � )' + k,
where h and k are constants. This is different from the idea of translation of axes,
which we used in Chapter 8. Then, we were moving the axes, while now we are
moving the points (x, y), with (x, y) � (x + h, y + k).
� ----+
Suppose that we have given two directed segments PQ, P'Q', in a coordinate
plane. If there is a translation under which P � P' and Q �---+ Q', then we say that
"---+ ----+
PQ and P'Q' are equivalent.
y
Q'
I
Q
p
I p'
This idea is easy to describe in terms of coordinates. Let
Q' = (x�, y�).

We can always move P onto P' by a translation
x I-+ x + h, y I-+ y + k,
where
h = x{ - X1, k =
y{ - Yi-
If it is true that
y{ - Y1 = Y� - Y2,
-+ -----*
then this translation also moves Q onto Q', and PQ and P'Q' are equivalent.
-+
For each pair P, Q, the symbol PQ denotes the set of all directed segments
-----* -+
P'Q' that are equivalent to PQ. Such a set of equivalent directed segments is called
a free vector (or simply a vector, if the context makes it obvious what meaning is
intended). Thus the figure on the left below is a partial picture of exactly one free
vector. A free vector is called an equivalence class of directed segments; and any
directed segment which belongs to such an equivalence class is called a representative
of the class. Thus each of the arrows in the figure is a representative of the free
-+
vector PQ.
y
-+ -----*
If two directed segments PQ, P'Q' are equivalent, then they determine the same
-+ � -+ � -+ -----*
free vector, and PQ = P' Q'. And if PQ = P'Q', then the segments PQ and P'Q'
are equivalent. Therefore, when we write an equation of the form
-+ �
PQ =
P'Q',
��
we are saying that the segments PQ, P'Q' are equivalent under a translation.
It is now easy to define, for free vectors, the operations of addition, scalar
multiplication, and dot product. If
-+ ---+ -+
OP+ OQ =
OR ,
in the sense defined in the preceding section, then

--+ ---+ --+
OP+ OQ =OR,
�
by definition. This definition is complete, because every free vector ST has exactly
--+ --+
one representative segment which starts at the origin. Similarly, if rxOP = OQ, then
--+ -
rxOP = OQ,
by definition; and
--+ ---+ � -
OP OQ · = OP· OQ,
by definition. The form of these definitions makes it clear that all the vector laws
and inner product laws of the preceding section also hold true for free vectors. Since
� +
all we would need to do is rewrite them in the new notation (using OP for P), it is
not worth while to do so. The length of a free vector is the length of each of its
representatives. That is,
--+ --+
IOPI = IOPI =OP.
The free vector of length 0 is denoted by 0.
--+
It is convenient, in figures, to use the label PQ for any representative of the free
�
vector PQ. Thus different segments may have the same label, as in the left-hand
figure below; and when they do, this means that the segments are equivalent.
y y
It is easy to see that the right-hand figure above is correctly labeled. Therefore:
--+ --+
Theorem 1. PQ + QP = 0, for every P, Q.
Similarly, the labels are correct in the parallelogram below. Since
--+ � �
OP+ OR=OQ,
we have
--+ --+ -
OP+ PQ = OQ.
This has a geometric meaning: we can add free vectors by laying representative
--+ --+ -- --+
segments end to end. Solving for PQ, we get PQ = OQ - OP. And this gives:
� � �
Theorem 2. PQ + QR + RP = 0, for every P, Q, R.
Proof
� - � - � - �- -
PQ + QR + RP = OQ - OP + OR - OQ + OP - OR = 0.
y
R
I
y I
I
I
I
I
I
I
I
I
I p
I "
I/" ---
x
.-:.,. .......... x
0
As a matter of convenience, we have defined equivalence of directed segments in

terms of a coordinate system. But in fact this relation of equivalence is independent
� =-+
of the choice of the coordinate system. The directed segments PQ and P'Q' are
equivalent under translation if (a) their lengths PQ and P'Q' are the same and (b)
the directions e and e' are the same.
Note that while the directions e, e' depend on the directions of the axes, the equation
e = e' does not; if the equation holds, and the axes are rotated, then the equation
continues to hold.
Thus we say that the relation of equivalence between directed segments, used in
defining free vectors, is invariant under changes in the coordinate system.
It very often happens that we use coordinate systems in the study of things which
are invariant under changes of coordinates. Thus the distance between two points is
invariant, and so also is the question whether a given curve is a parabola. But we use
coordinate systems in the study of parabolas, and similarly we use coordinate systems
in the study of vectors. If P = (x, y), then x and y are called the x- and y-components
--+
of OP. In this case
--+ --+
P= OP = xi + yj,
--+
where P, i, and j are as in the preceding section. Corresponding to the vectors i, j
�
we have free vectors i, j; and OP is a linear combination of these free vectors:
�
OP= xi+ yj,
as shown on the left below. And of course pictures of the new i and j can be drawn
--+
starting at any point that we want. In the right-hand figure, PQ, i, and j are all
free vectors. In general, ifV and Tare any vectors, with T :;tf. 0, then the T-component
of V is the number
vT = 1v1 cos e,
where 8 measures the angle between the direction of T and the direction of V.
y
y Q
PQ=i+2j.
Thus, in the figure below VT is the directed distance PQ, relative to the given positive
�
direction on the line that contains PR.
Since
V T = IVI
· · ITI cos e,
it is easy to express the T-component in terms of the dot product:
V ·T
v -
T
--
IT!
PROBLEM SET 9.8
In the figures below, we use tick marks to indicate that segments have the same length.
Thus the tick marks in the figure below say that AB = AC.
-+ --+ --+
1. a) Calculate OS as a linear combination of OR and OP. (The figure is a parallelogram.)
0 0
--+ --+ --+

b) Calculate OT as a linear combination of OR and OP, shown on the right above.
(These two answers, in combination, give a vector proof that the diagonals of a
parallelogram bisect each other.)
--+ -+ --+ --+
2. a) Calculate SR and OT as linear combinations of OR and OS. (The figure is a
-+ --+
rhombus, so \OS\ = \OR\.)
--+ --+
b) Show that, in a rhombus, SR · OT = 0. (These two answers, in combination, give
a vector proof that the diagonals of a rhombus are perpendicular.)
� � -+
3. a) Calculate OS as a linear combination of OP and OR in the left-hand figure below.
R R
� � --+
b) Calculate OT as a linear combination of OP and OR, in the right-hand figure.
4. Do Problems 3a and 3b give a vector proof that the three medians of a triangle are
concurrent? Or do you need to carry out a third calculation of the same kind, to
complete the proof?
5. a) Show that
IV· Tl � IVI · ITI,
for every two free vectors V and T.
b) Show that for any real numbers a, b, x, y, we have
lax+ byl � Va2+ b2Vx2 + y2.

6. Show that if P, Q, R, and Sare any four points of the plane, then
---+ � ---+ ---+
PQ + QR + RS + SP = 0.
7. Let V0 be a fixed (free) vector. Show that if V0 • V = 0, for every V, then V0 = 0.
8. Suppose that V i 0, and [V[

· = 1. Is this information enough to determine V?
=
If so, what is V? If not, give a figure, showing the possibilities for V.

9. Given that V i · = 0 and V · j = 0, discuss as in Problem 8.
10. Given V · i = 1 and V · j = 1, discuss as in Problem 8.
11. a) A set of vectors Vv V2, , Vn are linearly dependent if there are scalars !Xv IX2,
• • • • • •,
1Xm not all 0, such that

=
Show that for any V, the vectors i, j, and V are linearly dependent.
b) Show that if one of the vectors Vi is 0, then the vectors V1, V2, ... , Vn are
=
linearly dependent.
c) Find a number a such that 2i + j and 7i + aj are linearly dependent.
12. a) A set of vectors V1, V2, ... , Vn are linearly independent if they are not linearly
dependent. Thus the V/s are linearly dependent if
n
L CliVi = 0 => IX 1 = IX2 = . . . = Cln = 0.

i�I
Show that i and j are linearly independent.
b) Are i and i + j linearly independent? Why or why not?
--+
c) Given that i and OP are linearly dependent, what are the possibilities for P?
13. Show that if
and
then
IV1 - V2I =
IW1 - W2I·
(Remember that IVl2 V V, for every V.) Then draw a figure, and restate the
= ·
theorem in the language of elementary geometry.

14. Explain how Problems 5a and 5b can be regarded as the same problem.
15. a) Consider the vector space which you were asked to define in the last problem of the
preceding problem set. Let 1 be the constant function which is = 1 for each x on
[ -1, 1 ]. Find ten nonconstant functions.f1,f2, ,/10 such that 1 f; = 0 for each i.
. • . ·
b) Show that in the same vector space, f f0 0 for every f => f0

· = 0. =
9.9 VELOCITY, ACCELERATION, AND CURVATURE
We return to the discussion of moving particles in a plane. Suppose that the motion is
described by a path
P: I�E
t H P(t),
where I is a time interval. Let the coordinate functions of the path be f and g, so that
P(t) = (f (t), g(t)) (ton I).

We now regard the path as a function whose values are the vectors
- ----+
pt= OPt,
-
where Pt is a vector in the sense of Section 9.7, and P(t) is denoted by Pt, to fit it into
the vector notation.
We then have
-+
pt= f (t)i + g(t)j.

We can now define the velocity and acceleration. These are the free vectors
Vt= f'(t)i + g'(t)j,

At= j"(t)i + g"(t)j,
9.9 Velocity, Acceleration, and Curvature 423
where i and j are the free vectors corresponding to i and j. Since Vt and At are free
vectors, we can draw pictures of them in any position we want; and so we picture
them by drawing arrows starting at the point P1•
The picture then says that at time t, the moving particle is at the point Pt and has the
indicated velocity and acceleration vectors Vt and At· Note that Vt lies along the
tangent line; and this is right. (This should be checked, for the various possible cases.
(a) If f'(t) and g'(t) are both 0, then V1 0, and there is nothing to prove. (b) If
=
f'(t) -:;!= 0, then V1 and the tangent line both have slope g'(t)/f'(t). (c) If f'(t) = 0
and g'(t) -:;!= 0, then V1 and the tangent line are both vertical.)
When we write Vt = f'(t)i + g'(t)j, A1 j"(t)i + g"(t)j, we are describing each
=
of the vectors Vt and At by a pair of numbers. Unfortunately, the numbersf'(t),

g'(t), f"(t), g"(t) have no physical meaning, because they depend on the coordinate
system. It is possible, however, to describe the acceleration by a pair of numbers
which do have physical meanings. This is done in the following way. First we take a
free vector T, with the same direction as V1, but with length I. T is called the unit
tangent vector at the point
Pt. (Here, and throughout the following discussion, we are
assuming that the speed IV11 is not zero. If the length of Vi is 0, then its direction is
not determined, and so T is not determined either.)
Next we take a free vector N, with length 1, perpendicular to T, and lying on the
same side of T as At· Then Ai must be expressible as a linear combination
Ai= o:T + (3N
of T and N. Here o: is the T-component of Au and (3 is the N-component. These

numbers are called the tangential and normal components of the acceleration. We
shall now compute them.
Y At
N��----
In the right-hand figure above, 8 is the direction of Vt. Since Vt = f'(t)i + g'(t)j,
we have
where
IVtl = -Jf'(t)2 + g'(t)2•

Similarly, </> is the direction of the acceleration, so that
f"(t) . (t)
cos = Sill</>= g"
-.
 - 8) = IA1I cos 4> cos 8 + IAtl sin 4> sin 8
f" (t)f'(t) � g"(t)g'(t)
= f"(t) cos 8 + g"(t) sin 8 = I
I
f' (t)f"(t) + g'(t)g"(t)
-Jf'(t)2 + g'(t)2
Theorem 1. The tangential component of acceleration is the derivative of the speed.

That is,
IX= Ar = .E:_ IVtl = IVtl'.

dt
Once this has been observed, it is easy to check it, by differentiating the function
IV tl = )j'(t)2 + g'(t)2.
The normal component fJ is computed as follows. If N is counterclockwise from
T, as in the figure on p. 423, then
fJ = I At I COS [(8 + rr/2) r/>]

-
= IAtl cos [(8 ef>) + rr/2] = -IA tl

- sin (8 - ef>).
If the direction of N is reversed, the sign of sin (8 - </>) is also reversed. In any case,
we want fJ � 0, because N is taken on the same side of the tangent as A1. Therefore
we must have
fJ = IAtl • !sin (8 - </>)I,
in all cases. Therefore
fJ = IAtl ·I sine cos</> - cose sin</>I
=
I IAtl sine cos</> - IA1I cose sin </> I = lf"(t) sine -
g"(t) cose1
lf"(t)g'(t) - g"(t) f'(t) I lf" (t)g'(t) - g"(t)f'(t) I
IVtl ,Jf'(t)2 + g'(t)2
This formula for fJ also has an interpretation, but its interpretation is harder to
see, and requires the idea of the curvature of a path at a point.
For the sake of simplicity, we start with the idea of the curvature of the graph of a
twice differentiable function at a point.
For each x, let s(x) be the length of the graph from t = a to t = x. Then
s(x) = f'J1 +f'(t)2dt,

and
s'(x) = ,J1 +f'(x)2•

For each x, let 8(x) be the direction of the tangent line, with -TT/2 < 8(x) < TT/2.
Since s is an increasing function, 8(x) is determined when s(x) is known. Thus there
is a function h such that
8(x) h(s(x)), =
ahd (in the language of Section 5.8)

ae d8/dx
= h'(s(x)) =
.
ds ds/d."
The curvature K is defined to be
=I �:I·
This is easy to calculate. Since
(}(x) = Tan-1 f'(x),

we have
d(} 1
{}'(x) f (x)
dx 1 + f'(x)2 "
= = ·
Therefore
K
= I d(}ds I I d(}/dx
=
ds/dx
I I f"(x)
•
1 f'(x)2 -J1
=
+
1
+ f'(x)2 I
f"(x)
=
I [1 + f'(x)2]3/2
.
I
For future reference:
Theorem 2. The curvature of the graph of a twice differentiable function is given

by the formula
f"(x)
K
=
I (1 + j'(x)2]3/2
I •
For paths, the idea is similar. Take a fixed t0, and for each t, let s(t) be the length
of the path, from t0 to t. Then
s(t) {t-Jf'(u)2 + g'(u)2 du,

Jto
=
and
s'(t) -Jj'(t)2 + g'(t)2•

=
For each t, let{}(t) be the direction of the velocity vector at time t. We are working
on a portion of the path where !Vtl -JJ'(t)2 + g'(t)2 ¥- 0. On such a portion of the
=
path, sis an increasing function, and so {}(t) is determined when s(t) is known. Thus
there is a function h such that
{}(t) = h(s(t)).
Therefore
d(} h'(s(t)) d(}/dt
ds ds/dt.
= =
But according to the definition of the curvature K of the path,
K
=
l��I·
In order to calculate K
we take first the case in which Vt is not vertical, so that
tan {}(t) g'(t) .

f'(t)
=
Taking the derivative, we get
2 f'(t)g"(t) - g'(t)f"(t)
[sec (}(t)](}'(t) =
.
j'(t)2
Now
g'(t)2 = f'(t)2 + g'(t)2

sec
2
e(t) = 1 + tan2 e(t) = 1 +
f'(t)2 f'(t)2
Therefore
f'(t)g"(t) - g'(t)f"(t)
010 = .
f'(t)2 + g'(t)2
This derivation works whenever the velocity vector is nonvertical. (Query: How
would you derive the same formula, in the case where the velocity vector is vertical?)
[ [
This gives
I
de d e/dt e'(t)
K= I I I = =
l
ds ds/dt s'(t)
= f'(t)g"(t) g'(t)f"(t)
I
_
1
f'(t)2 + g'(t)2 . .Jf'(t)2 + g'(t) 2

lf'(t)g"(t) - g'(t)f"(t)I
-
_
]3/
[f'(t)2 + g'(t)2 2
Thus we have:
Theorem 3. The curvature of a twice differentiable path, at any point where the
speed is not 0, is given by the formula
lf'(t)g"(t) - g'(t)f"(t)I
K= ]3/
[f'(t)2 + g'(t)2 2
Comparing this with the formula
lf"(t)g'(t) - g"(t)f'(t) I
/3 = '
.Jf'(t)2 + g'(t)2
we get:
Theorem 4. At any point where the speed is not zero, the normal component of
acceleration is given by the formula
/3 = AN = K IV1l2•
In our discussion, we used the notation f', g', ... for derivatives, most of the
time, in order to connect our work with the preceding theory. We used the notation
de/dt, de/ds, . . . only when we really needed to talk about the derivative of one
function with respect to another, in defining and calculating curvature. In the litera
'
ture of physics, however, the notation f', g, . . . is hardly used at all. The following
notations are far more common:
df. dg. dx dy
v = -1 + 1 - v = -i +--j, v =xi+ yj.
dt dt dt dt
,
In the last expression the dots over x and y indicate differentiation with respect to
time. Similarly,
d2j d2g . d2x d2y.
A= l + J= .
+ J = Xl + YJ·
. .. . ...
- - l -
dt2 dt2 d t2 dt2

-
In these notations,
and
li.Y Ji.XI
- l(dx/dt)(d2y/dt2) - (dy/dt)(d2x/dt2)1
K = =
[x2 + _y2]312 [(dx/dt)2 + (dy/dt)2]31 2

There is a good reason, in physics, for the use of the "fractional" notation df/dx,
dy/dt, ... for derivatives. Most of the time, physical problems involve a large number
of interrelated functions, and physicists need to talk about the derivative of one of
these with respect to another. Therefore, the rest of the time, they use the same
"fractional" notation df/dx for ordinary derivatives f'.
PROBLEM SET 9.9
1. Find the point of maximum curvature of the parabola y = x2, and find the maximum
value of K.
2. Find the point of maximum curvature of the parabola y =2 + x + x2, and find the
maximum value of K.
3. Find the points of maximum and minimum curvature of the graph of y = x3, and
calculate the values of K at these points.
4. Calculate the curvature of a circle of radius a.

5. Calculate the curvature at the points (a, 0) and (0, b) for the ellipse
x2 y2
�+ =1.
/;2
6. Sketch the path
Pt = i cost + j sint,
showing the velocity and acceleration vectors at several points.
7. Discuss (as in Problem 6)
- t t
Pt = i cos 2 + j sin 2 .
8. Discuss Pt =2i cost + j sin t.

9. Discuss Pt = it + jt2 .
10. Discuss Pt = ti + (t - t2)j (0 � t � 1).
11. Discuss Pt =i cos t2 + j sin t2. ·
12. Discuss Pt = it + jt3.
13. Discuss Pt = it3 + jt2•

14. Discuss Pt =i(l - t)2 + j(t3 - t).
15. Discuss Pt =(t cos ()()i + ( -!gt2 + t cos ()()j. In the sketch , show the velocity at
t = 0, and find the direction of this vector.
16. For a certain path, the velocity at time 0 has direction I)( and length 1. The initial
point P0 is the origin. For each t, At gj Express the path in the form Pt
= - . =
f(t )i + g(t)j.
17. Discuss as in Problems 6 through 15, and express the tangential and normal com
ponents of acceleration as functions of the time:
Pt = 3i cos 2t + 3j sin 2t.

18. Discuss and sketch
P0 = i cos2 8 + j sin 8 cos 8.
Describe this as a path in polar coordinates; find a rectangular equation for its locus,
and identify the locus.
Po = (cos 8 - cos2 8)i + (sin 8 - sin 8 cos 8)j.
Po = (3 cos 8 + cos 38)i + (3 sin 8 - sin 38)j.

P0 = i(cos2 8 - sin2 8)i + sin 8 cos 8j.
po = (cos 8 - cos 8 sin 8)i + (sin 8 - sin2 8)j.
23. Is the following statement true? (Why or why not?)

Theorem (?). Given a path with coordinate functions f and g, on an interval [a, b], such
that f and g are differentiable, and the velocity is nowhere = 0, then there is a time t
�
such that V1 has the same direction as PaPb.
The figure indicates that in some cases, at least, there is such a time t.
y
24. Given a path which has curvature K at time t0, suppose that the axes are rotated. Does
the curvature change? Why or why not? [Hint: This problem does not require a
calculation.]
25. Let a = i + j, b = i - j. Suppose we define an "inner product" V1 * V2, by agreeing
that for
V1 = x1a + y1b,
V2 =
x2a + y2b,
the * product is
V 1 * V2 = X1X2 + Y1Y 2·
a) Does *obey the same formal laws as the old inner product?
b) Is it true that V1 *V2 = V1 · V2 for every V1 and V2? Why or why not? In any
case, express the new operation * in terms of the old.
9.10 CONCLUDING REMARKS ON

VECTOR SPACES AND INNER PRODUCT SPACES
The treatment of vectors in this chapter has been brief, because so far we are working
in a plane, and the main advantages of a vector approach appear in three-dimensional
space, and in spaces of higher dimensions. Meanwhile we must bear in mind that
vector ideas appear in many different forms.
1) Free vectors. Velocity and acceleration are vectors in this sense, as in Sections
9.8 and 9.9.
2) Bound vectors. These have not only length and direction, but also position.
For example, if two forces act in opposite directions on the ends of a spring, then
they may be regarded as bound vectors.
�F_1_. � ---F�2-
In the figure, the two forces have the same length and opposite directions, but they
do not cancel each other out, as free vectors would; on the contrary, they compress
the spring.
3) Sequences of numbers, regarded as vectors. For example, ordered quadruplets

(w, x, y, z) can be regarded as vectors. We make the natural definitions
(w1, X1, Y1, z1) + (w2, Xz, J2, z2) = (w1 + W2, X1 + X2, Y1 + Y2, z1 + Zz) ,
<X(w, x, y, z) = (<Xw, <XX, <XJ, <Xz),
(w1, X1, Y1, z1) · (wz, X2, Yz, Z2) = (w1W2 + X1X2 + Y1Y2 + Z1Z2.
In fact, this is the usual way of describing a space of four dimensions.
4) Systems of other kinds, regarded as vector spaces and inner product spaces. Some
of these are unexpected, but turn out to be useful. See, for example, Problem 33 of
Problem Set 9.7, in which it appeared that a set of functions can be regarded as an
inner product space, although functions may not seem like vectors when we look at
them one at a time.
For this reason, when people speak of "vectors," we need to find out what kind
of vectors they are talking about.
10 Infinite Series
10.1 LIMITS OF SEQUENCES
Most of the time, so far, we have dealt with limits of functions, as x �a or as

x � oo. But often we have dealt with limits of sequences, as n � oo. For example, in
Section 2.10, we wanted to find the area A, under the graph of y = x2, from x =0
to a = h. We expressed A as the limit of a sequence A1, A2, • . . , where An is the area
of a circumscribed polygonal region Rn. We calculated
h3 ( r) ( 1 )
An = 3 1 + � 1
+ 2n '
and we found that
h3
limAn = -
n-+oo 3
We are now going to use limits of sequences more extensively, as a way of dealing
with infinite series. Given an infinite sum
. 00
L ai = a 1 + a 2 + ...
. '
i=l
we define
and we call the An's the partial sums of _2;:1 ai. Thus the An's form a sequence
Ai, A2, ...

If
limAn =A,
n-+oo
then we say that the infinite sum is convergent, and we write

00
2; ai =A.
i=l
We shall now examine limits of sequences more carefully, starting with the
definition of the limit, and building up the theory that is needed.
Definition. Given a sequence A1, A2, . • . of numbers, and a number L. Suppose that
for every E > 0 there is an integer N such that
n > N => IAn - LI < E.
431
432 Infinite Series 10.1
Then
Jim An= L.
n-+ oo
Note that this is like the definition of lim.,�00/(x). A sequence which has a limit
is called convergent. Here, as always, when we speak of a limit we mean a finite limit
(unless the contrary is stated.)
1
Theorem 1.
.
limn�oo -
n
= 0.
Proof Here L = 0, and IAn - LI = 11/n - OI = l/n. Thus we need to show that
for every E > 0 there is an N such that
n> N => - <E.

n
1
Now
1 1
-<E <:::> n>-.
n E
If l/E is an integer, let N = l/E. In any case, there is an integer N> l/E. Then
n> N => n > 1/E => I/n < E,
On the basis of the definition of limn_,00 A,,, we can prove the expected theorems
on sums, products, and quotients. These are much like the corresponding theorems
for limits of functions. In Appendix C they are listed in such an order that they became
easy to prove. Meanwhile we shall state the main results and use them.
Theorem 2. If lim,,�00 A,, = A and lim,,�00 B,, = B, then
and
If B ':;!: 0, and B,, ':;!: 0 for each n, then
These theorems justify the procedures that we have been using informally. For
example, they give a proof that
lim
n-+oo
h3 (i n.!) (i
3
+ + 1-
2n
) h3 3
10.1 Limits of Sequences 433
The steps are as follows:
lim .! =0
n-+ 00 n
lim (i .!) + = 1
n -+oo n
lim _!_ = .! lim .! = 0

n-+oo 2n 2 n-+oo n
lim ( i +) 1- =1
n-+oo 2n
lim ha 1 + ( .!) ( i + 1-) =ha.

n-+oo 3 n 2n 3
(Justification for each of these steps?)
If we start with convergent sequences AI> A2, . • • , B1, B2, • • • , and so on, then
Theorem 2 tells us that certain other sequences are convergent. But often we deal with
sequences which are not built up out of convergent sequences as in Theorem 2. We
then need the following ideas.
Definition. A sequence A1, A2, is increasing if An�An+i for every n. The

An+l for every n. (If An < An+i for every n, then the
. . •
sequence is decreasing if An �
sequence is strictly increasing; and if An+i < An for every n, then the sequence is
strictly decreasing.)
Definition. If there is a number M such that An� M for every n, then M is called
an upper bound of the sequence A1, A2, . . . , and we say that the sequence is bounded
above. If there is a number m such that m �An for every n, then m is called a
lower bound of the sequence, and we say that the sequence is bounded below. If there
is a K > 0 such that /An/ � K for every n, then the sequence is bounded.
Example: (1) If An =/;:,,for every n, then the sequence is increasing, and is

bounded below but not above. (2) If An = e- , then the sequence is decreasing, and is
n
bounded. (3) If An = sin n, then the sequence is neither increasing nor decreasing,
but is bounded, with /sin n/ � 1 for each n.
It is easy to see that if a sequence is bounded both above and below, then it is
bounded. Given m �An� M for every n, let K be the larger of the numbers /m/
and IM/.
Theorem 3. If a sequence is increasing, and is bounded above, then it is convergent.
That is, if
Ai�A2 �...�An�An+l �...� M,
then the sequence has a limit. The first application of this principle that you may
have seen is in geometry. Given a circle of diameter 1, we inscribe in it a regular
polygon of 2n sides. For each n, let An be the perimeter of our 2n-gon. (Note that
we had better start with n = 2.) It is a matter of elementary geometry to show that the
sequence A2, A3, . • • is increasing. Also, An < 4 for every n, because the perimeter of
every inscribed polygon is less than the perimeter of the circumscribed square.
(Draw a figure.) Therefore the sequence is convergent. Its limit, of course, is 'TT.
We proceed to the proof. Let S be the set of all numbers An. That is,
Then S has an upper bound. By the Least Upper Bound Postulate (LUBP), S has a
least upper bound. (See Section 5.6.) This is called the supremum of S, and is denoted
by sup S. Let
A= sups.
We shall show that
limAn = A.
n co
....
Let E be any positive number. Then A - E < A. Therefore A - E is not an

upper bound of S. Therefore AN > A - E for some N. Since the sequence is increas
ing, this means that
n > N => An > A - E.
Since A is an upper bound of S, and A + E > A, it follows that A + E is an upper

bound of S.
Therefore
for every n.
Therefore
n > N => IAn - Al < E,
and limn-co An = A, which was to be proved.

We have a similar theorem for decreasing sequences:
Theorem 4. If a sequence is decreasing, and is bounded below, then it is convergent.
That is, if A1, A2, • • • is decreasing, and An � K for every n, then the sequence
has a limit.
Proof For each n, let Bn = -An. Then B1, B2, • • • is increasing, and is bounded
above. Therefore it is convergent. Let limn-co Bn = B. Then limn-oo An = - B.
Some simple sequences converge for reas0ns which are not covered by the
preceding theorems. For example, given that
lim _! = 0'
n-.0 __ n
it is obvious that
lim
n-co
1.V"- = 0,
n + 2 n
because the second sequence is smaller, term by term. This is the idea of the following
theorem.
10.l Limits of Sequences 435
Theorem 5 (The squeeze principle). If lim n- co An= L, Iimn-•co en= L, and A n �

B11 � en for every n, then limn-oo Bn = L.
In many cases, it is easier to use this theorem than to do awkward calculations.
Similarly, it ought to be true that
cos n
lim = 0'
11-00 n
because lcos nl � I, and I/n -+ 0. But we can't get this result from Theorem 2,
because cos n does not approach a limit as n-+ oo. Hence we need the following:
Theorem 6 (The annihilation theorem). If limn-· co An= 0, and BI> B2, • • • is bounded,
then limn_,00 A nB n = 0.
Theorem 7. Every convergent sequence is bounded.
For increasing sequences, this is trivial: if Ai, A2, • • • is increasing, and

limn-oo An= A, then Ai� An� A for every n. Similarly for decreasing sequences.
For a proof in the general case, see Appendix C. This theorem has simple applica
tions: if we show that a sequence is not bounded, then it follows that the sequence
is not convergent.
The statements
lim An = oo, IimAn = -00
n...,co
mean what you would expect. You should be able to state your own definitions of
them, following, if you need to, the models of Section 5.3. Sequences like this are
not called convergent. If lim"_,00 An = oo, then we say that the sequence diverges
to infinity. And if limn-oo An = - oo, we say that the sequence diverges to minus
infinity. We have to be careful about this: if convergence allowed the limits oo and
- oo, then Theorem 7 would become false, and Theorem 2 would be meaningless
in many cases. (You can't perform algebraic operations on the "numbers" oo and
-oo. )
PROBLEM SET 10.1
Investigate the following indicated limits. That is, find out whether they exist, and find
out, if possible, what they are.
I
1. lim 2
n�oo
n
2 + 311
2. Jim --- (Try dividing the numerator and denominator by n.)
n-Cf) 3n
3. Jim 0 (Try using one of the last theorems in this section.)

n-oo n- + ll
1 ll
4. Jim 3 5. lim 3
n-oo 2 ll + n n->00 ll + n2 + 7T
. sin (2n + 1) . cos (n - 1)

6. hm 7. Jim
n n->ro n + 1
� n
)
8. Jim
n->OO
( 1 +
n
� -l/n
) [Hint: (1 x)1fx.
9. lim
n_,.oo
( 1 -
n (1 +
Surely you know limx-o
1/y)v,
+
and apply the result to the problem in hand.]
Now find liffiy_oo
10. lim,H"' Bn. where Bn is the perimeter of a regular 2n-gon circumscribed about a circle
of radius 1. (You need not prove that your answer to this one is right.)
1 1. Jim Inn 12. Jim In (n2)

n-oo n-oo
13. Jim In 1/n 14. Jim In ( �)

n-oo n---+CO ll
15. Jim i "dx

- 16. Jim
f"dx
-
n-c.o 1 x
n-+CO 1 X2 3
17. Jim
f " dx
312
n-oo 1 X
n 1
18. Jim .L � (Investigate existence only. A geometric interpretation is useful.)
n---+CO z=l l
"dx
19. Jim (
n--oo J1 X
n
1
20. Jim L :a (Investigate existence only.)
n--i-oo i=1 l
n
1
21. lim L .31 (Investigate existence only.)
n-+OO 1.=l l 2
n
I
22. Jim L -: (Investigate existence only.)
n->OO t=l l
23. Jim
n--co i=l
I ( hr)n n sin : (Geometric interpretation?)
n
I I)
_L --- (
!!__
24. Jim I ( 2n) n cos
hr
25. Jim
n�oo i=l 2 n-oo i=l I + (i/n) ;
y
10.2 Infinite Series. Convergence. Comparison Tests 437
26. lim I
" 1 (1)- 27. Jim -
1 n
L eifn
n-•co i=l 1 + (i/211) n n-co n i=l
28.
l n
lim - L e-i/n
n-co n i=l
29. Jim n -n1
sin
1
30. Jim n2 sin 2
n--+CO n
31. Jim 11 1
n-+CO
( - cos �
n
)
32. Jim n tan
1
33. Jim n
1-
sec
n-.oo n n
34. Jim
n--+CO
[I � - ]
1.=l l
In n
(In fact, this limit exists; if you can find a geometric interpretation of the problem, you
can prove it. The limit is known as Euler's constant. Nobody knows whether it is
rational.)
n 1 n 1
35. Jim
n->OO
I
i=l
.
(21 + 1) 2
(Investigate existence only.) 36. um I
n-.oo i=l (31.2
+ l
)2
10.2 INFINITE SERIES. CONVERGENCE. COMPARISON TESTS
By an infinite series we mean an indicated sum of the form

00
I ai = a1 + a2 + · · + an + ·
· · ·
i=l
We say "an indicated sum" because in many cases there is no such thing as the sum
of infinitely many terms. For example, the series
l+l+l+ ·" (to infinity)
has no sum; and neither does the series
1 - 1 +1 - 1 +1 (to infinity).
In many cases, however, the "sum of infinitely many terms" can be defined, by a
passage to a limit, in the following way.
Given the series
for each n, let

n
An = I ai = a1 + a2 + · · + an
i=I
·
Then An is called the nth partial sum of the series. If
limAn =A,
where A is a (finite) number, then we say that the series is convergent and that A
is its sum. We also say that the series converges to A. If the sequence A1, A2, • • •
has no limit, then we say that the series is divergent. If
limAn = oo,
n-+ oo
then the series diverges to infinity; and if
limAn = -oo,
n-+ oo
then the series diverges to minus infinity. We may write these statements briefly as
"' "' 00
L Qi = A, .z a;= co, .2;a; =-co .

i=l i=l i=l
Probably the first example that you have seen of a convergent series is the geo
metric series
n
1+ r + r2 + · · · + r + · · · (0 < r < 1).
Here
n
n 1- r +I 1 1
An= 1 + ,2 + ... + r = --- - , n_r_
1- r 1- r 1 - r 1-r 1- r
If we know that
Jim ,.n = 0 (0 < ,. < 1), (1)
-oo
n
then it follows that
1
limA = - - (0 < r < 1); (2)
-+oo n 1- r
n
and this means that
00
1
.2; ri = - (0 < r < 1). (3)
i=O 1 - I'
There are many ways of proving (1 ). The following proof is the easiest. Since
0 < r < I, we have
rn+i < rn for every n.
Therefore the sequence r, r2, r3, • . • is decreasing. And it has a lower bound, namely 0.
Therefore the sequence is convergent, to some limit L. Thus
lim rn = L, and Jim rn+i = L.

n-+oo n-+ oo
(Why? What happens to the limit of a sequence, if you omit the first term?) Therefore
l
L = Jim rn+ = r Jim rn = rL,
11-00 n-oo
and so
L = rL, and (1 - r)L = 0.
Since 1 - r � 0, it follows that L = 0. Therefore
Jim rn = 0 (0 < r < 1).

n-+oo
In fact, the same conclusion holds for -1 < r � 0. We get this from the following
observation:
Theorem 1. If limn-oo lanl= 0, then limn-oo an= 0, and conversely.
Proof? (If you rewrite these two statements, using the definitions of the statements
limn-oo lanl= 0 and limn-oo an= 0, they hardly even look different.) Thus we get
the following theorem.
Theorem 2. If -1 < r < 1, then
limrn = 0.
n-+oo
Algebraically, the formula
n n
1 + r + y2 + . . . + r = (1 - y +l)/(1 - r)
holds for every r ¥:- I. We therefore have a more general result for geometric series:

00
1
Lri = -- .
i=O 1 - r
n
This holds because lim n-oo [-r +l/(l - r)] = 0/(1 - r) = 0. If the first term
is any number a, rather than 1, then we have:

oo . a
Lar' = -- .
i=O 1 -r
The following theorem often makes it easy to see that a series diverges.
Theorem 5. If L:,1 ai is convergent, then lim n -oo an = 0.
Proof For each n, Jet
Let lim,._,00 An =A. Then limn-oo An-I = A, where n > I. Therefore
lim(An - An_i) =A - A= 0.
n->oo
But An - An-l = an- Therefore limn-oo an = 0, which was to be proved.
For example, the geometric series L:o ari is divergent for a ¥:- 0 and lrl � I.
In this case, lanl = lal · lrln � lal, and so an does not approach 0.
Warning. The converse of Theorem 5 is false. That is, the nth term of a series
may approach 0, and the series may still diverge. The simplest example of this is the
series
1 + t + t + ·� + i + i + t + t + t + t + · · ·
The next five terms are each equal to t; and so on. Here an-+ 0, but the series
diverges to infinity.
A more natural example of the same phenomenon is the harmonic series
00 1 1 1 1
I�=1+-+-+ . . ·+ -+
n
. . .
i=l i 2 3
In fact, this diverges. The easiest way to see this is to draw a picture:
y
1 2 3 n n+I
For each n, the area under the graph from x = 1 to x =n + 1 is less than the total
area of the circumscribed rectangles. Therefore
A
1 1 1
=1+-+-+ .. · +- >
ln+l dx -.
"
2 3 /1 1 X
But this integral is In (n + I); and
limln(n + 1) = oo.
Therefore the partial sums An form an unbounded sequence, and the series must
diverge to infinity. Briefly:
Theorem 6. 2;;:1 (l/i) = oo.
The same sort of comparison scheme can be used for other series, to show that
they converge. Consider, for example,
Here ai = l/i2, and so lim;�oo ai = 0. This does not, in itself, show that the series
converges. But the algebraic pattern suggests that the series is related to the improper
integral
100 dx la dx
2 = 1Im
[-l]a (-1 l)
.
2 =
1.
Im - =
1.Im - + = l.
x a-+ x a-+ x a-+
1 00 1 co 1 00 a
. Since 1 /x2 x, "

> 0 for every the integral approaches its limit from below, and
ld : < 1 for every n.

1 x
And since the function 1/x2 is decreasing,
(n > 1).
(Here the area of the rectangle is 1/n2.) Therefore

l_ l_
{ 2dx r s dx
< <
22 J1 x2 ' 32 J 2 x2 '
n dx
and
n
J1 x2
1
An 2 2 < 1 + < 2.
=
i=l n
-
Obviously the sequence A A 2,

1, • • • of partial sums is increasing; and we have just
seen that it is bounded above. Therefore:
Theorem 7. 2:1 (1/i2) is convergent.

Some of the ideas that we have been using to get these results are useful in so
many connections that they are worth recording as theorems.
Theorem 8 (The comparison theorem). Let 2:1 ai and 2:1 bi be series, with
0 � ai � bi for each i.
Then (1) if 2:1 bi is convergent, then so also is ,Li':1 ai; and (2) if ,L:1 ai is divergent,
then so also is .L:1 bi.
Proof For each n, let
Then
(Why?) And each of the sequences A1, A2, • • • and B1, B2, • • . is increasing. An
increasing sequence is convergent if it is bounded, and conversely. We can therefore
prove ( l) in the following steps:
00
i=l bi
2 is convergent
=> B1, B2, • • • is convergent
=> B1, B2, • • • is bounded
=> Ai, A2, • • • is bounded
=> A1, A2, • • • is convergent

00
2a; is convergent.
=> i=i
(Reason for each of these implication signs?)
following steps:
Similarly, we can prove (2) in the
00
i=l ai
2 is divergent
=> Ai, A2, • • • is divergent
=> A1, A2, ••• is unbounded
=> B1, B2, • • • is unbounded
=> B1, B2, • • • is divergent

00
=>
i2=l bi is divergent.
The comparison theorem gives us easy tests for some series. Consider for
example,
00 1 =1+-+-+-+···
1 1 1
i2-
=O i ! 1 2! 3!
Here
n! = l n (n · 2 · 3 · · ·
� 1)
and O! = 1, by definition. For each i, let
ai= �'1 bi= (2_ 1)i-l.

. l.
Then
a i � bi for each i;
1 erl (i = 0),
- < -
O! 2
1
1! Gr (i = 1),
-
and thereafter the strict inequality holds, with l/n ! < (l/2r-1 for n � 2. Therefore
our series is term by term less than the geometric series
which is known to converge. Therefore:
Theorem 9. 2;:0 (l/i!) is convergent.

In fact,
I �!
i=O l
=
e = ln-
1
1 = lim (1 + x)1/"'.
x-+0
But we won't be able to prove this until we have developed the theory much further.
The situation here is peculiar: the easiest way to get this special result is first to show
that
and then to set x = 1. (You have seen a situation like this before. The easiest way to
find H x4 dx is first to calculate the function frr t4 dt, and then to set x = I.) Consider
next
00
1
�
i n - /n
Since
1 > !
_
for every n,
n - �n n
and
it follows that the given series diverges:
ro
1
Cf).
i� n - )n
While the comparison theorem tells us, under some conditions, that a series
converges, it never tells us what the sum is. But such partial information may be
useful. In fact, some of the most important uses of series are in cases where a number
(or a function) can best be described by a series; in such cases, we use Lf=i a; (for
some large n) to get an approximation of ,L;:1 a;. For example, the approximation
is excellent, even for fairly small values of n; it gives by far the best way of computing
e; and in fact, the series approaches its infinite sum so fast that e is much easier to
compute than )2.

Therefore, when you are asked to show that a series has a sum, without finding
out what the sum is, you should not consider that the problem is artificial.
PROBLEM SET 10.2
Find out which of the following series are convergent. If the series is geometric, calculate
the sum.
I. -
oo
oo
1 1 00
. i� j 3 /2
2 Vi: 3. .2
= --
i=l ----: 2 i l Vi + 1
oo
1
6. i(
) � 2i
i � j3/2 + 2
4.
i=l 4
00 00
sin2 (2i - 1)
.2 (-l)i7T-i 8. L ( -2)ie2i I
.
7 9.
i=l i=l i =l j2
cos3 (2i)
i�
00 00
1
10. �l j2
11.
i=l j 3
12.
i�i'
oo 00
1 1 00
1
13. 14. 0 9 15. .2 -.-·
i�l j l-1 i� j. i=2 1 nl I
oo 00
1 1 00
1
i=2 �
2 l 1n I . 17. 2 _2
i=2 z:---1n2 z
16. . 18.
1n
i=2 1-:---
z z
t
•
00 co
00
1 1 1
i=2 l:a-n
1 .
19 20 .2 21.
· i�2 ln2 i i�O (i !)2
•
22.
� (i ! - 1)
£.,
i=O (i!)3
23.
oo
2 i(i+1)
i=l -
1
24. 2
co
-
1
i=2 i(i - 1)
--- i -._
i_
00 00
1 1
26. 2
2
(i + l)(i + 2) (i - l)(i)(i + 2) + 1
25. 27.
i=l i i=2 --- i=l l
+ 1
£.,
2 8.
� i
i=l 21_
30. i: �
i=l j2 - 1
31. If you think of Theorem 3 backwards, it says that
--1
1 -r
=1 +r+r 2+···
That is, 1/(1 - r) can be expressed as the sum for an infinite series. Express 1/(1+x)
as the sum of an infinite series. For what numbers x does your series converge?
32. Express 1/(1 +x2) as an infinite series. For what numbers x does the series converge?
33. Same question, for 1/(1 + x4).
*34. Suppose that :L:o aixi converges for every x. The series then defines a function
co
/(x) =
2 aixi.
i=O
It will turn out that functions which can be defined in this way are always differentiable,
and that their derivatives can be calculated by differentiating the series a term at a time.
That is,
co
f' (x) = 2 iaixi-I.

i=l
10.3 Absolute Convergence. Alternating Series 445
(Don't try to prove this; you haven't got a chance.) Granted that all this is true, what
must the a;'s be, if /(0) 1 and f'(x) /(x) for every x? Comment on your result.
= =
00
i 00 2i2
35. :L-.-
3 36. 1 I�1
i=l I + 1 i=l +
*37. For which numbers ct. is the series :L:1 (l/n") convergent?
*38. Prove the following.
Theorem A (The Integral Test). Let f be a positive decreasing continuous function,
on the interval [1, oo ) . If
ioo f(x) dx < oo, (1)
then
00
:L /Ci) < w, (2)

i=l
and conversely.
10.3 ABSOLUTE CONVERGENCE. ALTERNATING SERIES
Given a series _L!o ai (in which the terms may be positive, negative, or zero), we can
form a new series by taking the absolute value la;J of each term a;. For example, if
00 00
Lai = L( - l)iri = 1 - r + r2 - · · • ,
i=O i=O
then
00 00
_L la;I = L lrli = 1 + lrl + lrl2 + · · ·
i=O i=O
Given that La; converges, it does not follow that _L lai l converges. For example,
the series
1 a; =
1 - 1 + t- t + i - i + · · ·
is convergent, but the series
:L la;I =
1 + 1 + t+ t + i + t + · · ·
is not, because the harmonic series is not. The same sort of thing happens if we take
absolute values in the series
00 00
· 1 1 1 1
:L ai :L c-1)i+ 1--: 1
= = � - + - - - + ...
i=l i=l l 2 3 4
Here it is plain that L la;I diverges, but it is not quite so easy to see that La; is
convergent. This is worth proving, however, because the idea used in the proof is
useful in other connections.
Theorem 1. _L!1 ( - l)i+1(1/i) is convergent.

Proof Let
If n is even, with n = 2k, then
"k= 1 - I_ + I_ -
An = Ao.
2 3
· · ·
2k
- l_
= (1 -
l
2 )+( l
3 - )4 .+
1 ...
+ (2k 1- 1
-
1)
2k .
Therefore the sequence A2, A4, A6, ••• , A2k, • • • is increasing. And it has an upper
bound, because
A2k = 1 - G �) (� - �)
- - - · · · - ck� 2 - 2k� 1 ) 2k1 - < 1.
Therefore the sequence A2, A4, • • • has a limit. Let
A= lim A2k. (1)

k-+ 00
We shall show that A is the sum of the series. First we p.bserve that
so that
lim A2k+1= lim [A2k
le-> oo k-> oo
+ a2k+il = lim A2k
k-> oo
+ lim
le-> oo 2k + 1
,
lim A2k+i
k--+ 00
= A + 0= A. (2)
Thus we see that (I) as n--+ oo through even values, An --+A and (2) as n--+ oo through
odd values, An --+A. It follows that (3) limn�oo An= A.
Proof? (You need to show that for every E > 0 there is an N such that [An
- A[ <
n > N. Given such an i:, you know from (1) that there is an N1 such that
i:· for every
k
[A2k - A [ < i: for every > NI; and you know from that there is an N2 such that (2)
k
IA2k+i - Al < i: for every > N2• How can N be definecl in terms of NI and N2?)
The scheme that we used to prove Theorem 1 applies more generally. If you
reexamine the proof, you will see that the only facts ,about the series
that were used were the following:
1) The series is alternating. That is, successive terms a;, ai+I have opposite signs.
2) Limn�oo an = 0.
3) The sequence laII, la2I, ... is decreasing.
We have therefore proved the following theorem.

10.3 Absolute Convergence. Alternating Series 447
Theorem 2 (The alternating series test). Given an alternating series ,L:1 a;. If
l i mn_,00 an = 0, and the sequence la1i, la2i, . . . is decreasing, then the series converges.
(Strictly speaking, some of our formulas in the proof of Theorem 1 used the fact
that the first term was positive instead of negative. If you know that the theorem
holds in this case, how would you show that it also holds when a1 < 0 ?)
We have seen that if La; converges, it does not follow that L la;I converges.
But the reverse implication does hold:
Theorem 3. If _L;:1 la;I is convergent, then so also is ,L;:1 a;.

If .L la;I is convergent, then L
. a; is said to be absolutely convergent. In this
language, we can restate Theorem 3 as follows:
Every absolutely convergent series is convergent.

To prove this, we break up each partial sum
into a sum of positive terms and a sum of negative terms. To do this, we let
+_ a;
a;
{ if a; � 0,
- 0 if a; < 0,
{
and let
a; if a; � 0,
a; =
_
0 if a;> 0.
Let
n n
A+
n -
-
"" ai'
L- + A-;;-= _La;.
i=l i=l
Then
for each n, because
for each i. Obviously A�, A;, ... is an increasing sequence, and A-;, A;, ... is a
decreasing sequence. Let
ro
k = .L la;!.
i=l
Then
n
L ia; I :::::; k for every n.

i=l
Also
n
A!� _L la;I,
i=l
because A� is the sum of some (perhaps all) of the terms on the right-hand side.
Therefore At, A�, . . . is convergent. Let
A+= lim A;;.

Similarly,
n
A-;;- � I ( - lail),
i=l
because A: is the sum of some (perhaps all) of the terms on the right-hand side;
and if you omit negative terms, the sum becomes larger. Therefore the decreasing
sequence A�, A-;, ...is bounded below. Therefore it has a limit. Let A- = limn�co A:.
Then
lim An = Jim (A;; + A-;;) = A+ + A-,

n-+co n-+co
and I�1 a; is convergent, which was to be proved. In fact, we can say a little more:
Theorem 4. If I�1 /a;/ is convergent, then
Ii� a;I � ;� /a;J.

Proof We know that
By induction it follows that
Ii� a;I � ;� /a;/ for every n.
Passing to the limit, we get the inequality that we wanted.
PROBLEM SET 10.3
Find out which of the following series are alternating, which are convergent, and which
are absolutely convergent.
"' .
. l
3. '
£.. ( -1)' -. -
i=l i=l"' �·2 i=l I + I
"' 1 1-
4. I <-1)i -:{I s. I 2
i=l"' l + i
i=l
l
"'
i1 -
i=lI c ni1
I
9.
00
:L t)i ' (i 1)
1.
i=lI c-2)�)2i-:-i i=2 c
-
s. �
-
(- (
l. I
rri) � 00
10.
i=lf e
11.
i=lI ::.!_2
sin + cos
2 l
12. I
i=l
c-i)-i
10.4 ESTIMATES OF REMAINDERS
Given that a series converges, we often want to use a partial sum

10.4 Estimates of Remainders 449
as an approximation of the limit

ct)
A= lim An = L ai.
n-+co i=l
The approximation An � A is used in some of the most important applications, and
in all applications that use computers. As in all approximation processes, we are
better off if we can set a limit on the error. We shall now find ways to do this.
Given that 1:1 ai converges to a sum A, let"Rn= A - An. Then
ct)
Rn= L ai,
i=n+l
and obviously
limRn= 0.
n-+oo
For alternating series, of the type treated in Theorem 2 of the preceding section,
it is easy to get an estimate of R,,,. Let the series be
ct) ct)
"" ai = "" i+l =
£.., £.., ( -1) b; b1 - b2 + ba -
i=l i=l
where b; = Jail· Then
ct) ct)
R n= L a i= L (-l)i+l bi.
i=n+l i=n+l
If n is even, then
= (bn+l - hn+2) + (bn+3 - bn+4) + ' ' ' � 0.

But we can also write
Therefore
1) 0 � Rn � hn+l• when n is even.
If n is odd, then
Rn = -bn+l + hn+2 - hn+3 +

= -(bn+l - h,,,+z) - (bn+3 - b,,,+4) - · · · � O;
Rn = -bn+l + (bn+2 - hn+a) + (bn+4 - bn+5) +
Thus
2) -bn+l � Rn � 0, when n is odd.
Therefore
Since bn+l = Jan+il, we have proved the following theorem.

Theorem 1. Given Ii:1 ai. If (1) the series is alternating, (2) limn�co an = 0, and
(3) the sequence la11, la2I, . is decreasing, then
. .
IRnl � lan+il for every n.
That is, when you stop after a finite number of terms, the error is numerically no
larger than the first term that you omit. For example, take
i(-l)i+l_!2 = 1 - _!_2 + _!_2 - . . .
i=l i 2 3
By the alternating series test, this series converges. Let A be its sum. Then
1 1 1
A�l--+-- ··· +-·
2 2 2 ' 2 3 9
and the error in the approximation is � 1/102 = 0.01. This series does not converge
very fast. Next consider
C()
-1 1 1
I c lY-:-
i=O
-
l!
= 1 - 1 + -
2!
- - + ...
3!
This series converges to a sum A. (It will turn out that A = 1/e.) We have
1 1 1
ARJ--- + ··· +-·
2! 3! 10 ! '
and the error is less than 1/11 !. This series converges very rapidly:
11 ! = 39,916,800, and __!__ � 2.5052 · 10-s � 0.000000025052.

11 !
If you reexamine the proof of Theorem 1, you will see that the method that we
used to get an estimate of the error was very much like the method that we used to
establish convergence in the first place, in the proof of the alternating series test.
This happens most of the time: that is, a proof of convergence usually gives an
estimate of Rn- Consider, for example,
C() 1
I-:;.
i=l l
We let
and we observe that the sequence A1, A2, is increasing. To show that it is bounded
• • •
above, we draw a picture and observe that
An =
n 1
L -:2 < 1+
ln dx
2 ·
i=l l 1 x
10.4 Estimates of Remainders 451
The same sort of reasoning tells us that

oo
= � l_
J
< dx.
R £.,,,
i=n+l l·2 9
n
n x·
Since
we conclude that
< 1
Rn -
for every n.
n
This is nowhere nearly so small as the estimate of error for the corresponding alter
nating series. 1n fact, the positive series I (1/i2) converges very slowly.
Similarly, Theorem 4 of Section 10.3 gives an estimate of the error for series
which are absolutely convergent.
Theorem 2. Suppose that L lail is convergent. Let

00
Rn= L ai.
i=n+l
Then
00
IRnl � L lail·
i=n+l
That is, the error in Iai is numerically no greater than the error in I la;I- To
prove this, we apply Theorem 4 to the series
00
I ai,
i=n+l
If we use the comparison theorem of Section 10.2, to establish the convergence
of a positive series, then any estimate of the remainder of the larger series auto
matically is an estimate of the remainder of the smaller one. For example, we have
found that
00 1
I-:; < co,
i=l i
with Rn < l/n for every n. Since 0 < l/(i2 + 1) 2
< l/i for every i, it follows by the
comparison theorem that
00
1
I-. -- <co.
i=l i2 + 1
It also follows, for the remainder R� in the new series, that
and so
R� < l.
n
This scheme always works, whenever we establish convergence by means of the
comparison theorem.
PROBLEM SET 10.4
Each of the following series is convergent. In each case, get an estimate of the remainder
Rn, in the form \Rn\ � · · ·
i
I. i; (� ym
i=l 3}
2. i: (- �4)
i=l
i; co� 7Ti
00
3. I ( -l)i7T-i 4.
i=l i=l l
00
sin2 (2i - 1) 00 1
5. L 6.
i=l i2 i�l f3
00 ( - l)i+l
7. I 8.
i=l i-4
w ( - l)i w
1
9. L
i=l
- i0.9
10. L -:---
i=2 12.
1 n 1
10.5 Termwise Integration of Series. Power Series for Tan-1 and In 453
1 1.
co 1
12.
co
2
( --rl
i O (i!)2
� i=l e
co 1 co 1
13. 14.
�1 i(i + 1) i 2 i2(1 + i)
�
co 1 co
15. 16. I c -o-i

i l i(i + l)(i + 2)
� i= l
*17. Let
[a, b].
fv fz,
Let
..
f
. be a sequence of continuous functions defined on the same interval
be a function such that
Jim fn(x) f(x),

n-co
=
for each x on [a, b]. Questions: (1) Does it follow that/ is continuous? (2) If f is known
to be continuous, does it follow that
(?)Jim
n--+co
Jb fn(x) dx Jb f(x) dx?
a
=
i
* 18. Consider I!1 ( -l) +l (l/i).
Show that by writing the terms of this series in a different
order (using each term once and only once) you can get a series La; whose sum is 10.
19. Now reexamine your solutions of Problems 1 through 16. If you used any method other
than Theorem 1, in estimating the remainder in an alternating series, try using Theorem
1, and compare the new estimate with the old one. (The alternating series test usually
gives a good estimate, in the cases where it applies at all.)
10.5 TERMWISE INTEGRATION OF SERIES. POWER SERIES FOR Tan-1 AND In
A power series is a series of the form
(Here, as a matter of convenience, we are defining 0° = 1, so that a0x0 = a0 for every

x, including x = 0.) Thus every geometric series is a power series; writing x for r in
the old formula, we get
co
I
. -
x'=
1
(-1 < x< 1).
i=O 1- X
If a given series is convergent, for every x on an open interval (-r, r), then the series
defines a function/, on the same interval, and we write
co
f(x) = L a;xi (-r<x<r).
i=O
The following theorem is fundamental:
Theorem A. Given
co
f(x) = L G;Xi (-r<x<r).
i=O
Then/ is continuous and differentiable on (-r, r), and the derivative of the sum is
the sum of the derivatives. That is,
00
f'(x) =Ii aixi-l (-r<x<r).

i=l
The same idea applies to the integral.
Theorem B. Given
00
f(x) =I aixi (-r<x<r).

i=O
Then the integral of the sum is the sum of the integrals. That is,
x ["'
l f(t) dt =I ai dt =I -._ai_1 xi+1.
oo oo
0 i=O. 0 1=0 l +
As you might expect, the proofs are hard; they will be postponed until the end of
this chapter. But the theorems are easy to apply, and Theorem B gives the best method
of finding series for many functions. The method is as follows.
We know that
1
1 + x + x2 + ... + xn + ... = -- (-1 <x<1).
1 -x
Writing this backwards, we can express the function 1/(1 - x) as a power series:
1/(1 - x) = 1 + x + x2 + · · · (-1 <x < J).
Replacing x by -x, we get
l
__ = 1 - x + x 2 - . .. + (-lrx" + .. . (-l<x<l );
1 + x
and then, replacing x by x2, we get
_1_ = 1 - x2 + x4 - xs + ... + ( -1r x2n + ... (-1 <x<1).

1 + x2
Theorem B says that the series on the right can be integrated a term at a time. Thus
lx dt
= lo"'dt -lo"'t2 dt + ·
+ (-1)" l"'t2n dt + (-l<x<l ),
ol+t2
-- · · · · ·
. o
and this gives
lx dt
=x -
x
3
� + ...(-l)" 211 + 1
. x2n+1
+ ...
0 1 + t2
00
i 2i +l
=I C-l) -
�
- - (-1<x<1).
i=o 2i + 1
The integral on the left is equal to Tan-1 x. Thus we have:

10.5 Termwise Integration of Series. Power Series for Tan-1 and In 455
Theorem 1.
oo . x2i+1
Tan-1 x =I (-1)' ---
. (-l< x<l).
i=o 21 + 1
Granted that Theorem B is true, there is no need to test the convergence of the
series on the right; Theorem B tells us not only that the series has a sum, but also that
its sum is Tan-1 x. Note that the series includes only terms of odd degree. This
could have been predicted, because Tan-1 is an odd function, with Tan-1 ( -x) =
-Tan-1 x for every x.
The same method can be used to get a series for the natural logarithm.
Theorem 2.
x2 x3
In (1 + x) = x- - + -
2 3
(-1 < x< 1).
x
Proof We know that
dt
and we know that

In (1 + x) = -- ,
0 1 +t J
1
1 - t +t2 - t3 + ... + (-l)iti + (-l<t<l).
1 +t
-- =
f"'at - xtat +f"'t2 dt - "'

By Theorem B,
lx dt f
--
o l +t
=
o Jo o
· · · + c-1Y tiat +
o
x2 x3 xi
=x--+- l)i +
. 1
· · · + ( - - + · · ·
2 3 i
00
i
=IC-l)i+l� (-1 < x< 1).
i=l l
Note that this method cannot be used to calculate the integral from 0 to 2, because the
series for 1/(1 + t) converges only for !ti < 1.
The method that we have just been using can be applied so as to give answers,
in the form of series, for problems which up to now we could not have solved.
Consider
r -1/2 �
.
Jo 1 + x4
In Chapter 6, this would have been an impossible problem. But now we can solve it,
by expressing the integrand as a power series, and integrating a term at a time. In
the series for 1/(1 + x), we replace x by x4• This gives
1 .
1 - x 4 + x8 - + (-l)'x4'
.
+
1 + x4
--- = · · · · · ·
00
=I c-1)ix4i.
i=O
Therefore
i -112 dx oo 1 -112
(-l)'x4'dx
. .
4 I
1 + X i=O
=
--
0 o
I< -l)i 1< ) i+l 4

_ -t
--
.
i=O + 1
=
4z
-t - t(-t)5 + -H-W +
-! +_1 _ -
_
1 _ + ...
.
=
2 5 25 9 . 29
This is an alternating series; the terms diminish numerically, and approach 0 as
n--+ oo. Therefore, if we use the first three terms as an approximation of the integral,
the error is less than the fourth term. This is
which is quite small: 210 = 1024, 213 = 8192, and so E < I0-5•
PROBLEM SET 10.5
1. Calculate Tan-1 0.02 to six decimal places, and explain how you know that the error in
7
l/10
your approximation is less than 5 · 10- •
1
2. Calculate
J 0
---4 dx to five decimal places, and explain how you know that the
1 + x
error in your approximation is less than 5 · 10--G.
3. Using the first term only, in the series for Tan-1, we get the approximation formula
Tan-1 x ""' x for x ""' 0.
How might you explain and justify this approximation formula if you knew nothing
about infinite series?
4. Given
f(x) 1 + 6 •
x
a) Express f(x) as an infinite series.

b) Express
0.6�
as an infinite series.
Jo 1 + x
6
c) Calculate numerically the sum of the first three terms of your series.
d) Get (by any method) an estimate of the error in the resulting approximation of the
integral.
5. Do the same four things, starting with f(x ) = 1/(1 + vi), on the interval [O, 0.49}
(Your infinite series will use powers of v�, but the same methods will apply, for the
same reasons.)
10.6 The Ratio Test for Absolute Convergence 457
6. Do the same four things, starting with/(x) = 1/(1 + x), on the interval [O, 0.2].
1
7. Do the same, starting with f (x) 5 2 , on the interval [O, 0.25].
1 + x 1
=
1
8. Do the same, starting with/(x) = 3, on [O, 1/2].
1 + x
9. Express in the form of a series:
rk [ f--=:_J
Jo i=oi + 1
dx (0 < k < 1).
10. Using the first term only, in the series for In, we get the approximation formula
In (1 + x) R:J x for x R:J 0.
How might you explain and justify this formula, if you knew nothing about infinite
series?
11. Consider the function f(x) defined by the series
n
x2 x3 x
1 + x + + +· + +
2 3! · · n! . · ·
a) Express/'(x) as a series.
b) Express H f(t) dt as a series.
The results that you get ought to enable you to guess what the function is.
*12. For each n, let
(0 � x � 1).
a) Find limn-->oo f� fn(X) dx.

b) For each x on [O, 1], let f(x) = limn-->oo fn(x). Get a formula for the function f(x).
c) Find
rl
)o
f(x) dx = J1 0
[limfn(x)] dx.
n-co
*13. Your answers in Problem 12 suggest that the functions fn behave rather peculiarly.
Investigate as follows:
a) For each n, let .Xn be the point at whichfn takes on its maximum value. Get a formula
for Xn , and find limn_,00 Xn .
b) For each n, let Yn fn(.Xn ). Get a formula for Ym and find liffin_,00 Yn
·
=
c) Draw a sketch showing what the graph of fn looks like for n 1, n 2, and = =
n R:i oo. Your sketch will throw some light on the results that you got in Prob
lem 12.
10.6 THE RATIO TEST FOR ABSOLUTE

CONVERGENCE. APPLICATIONS TO POWER SERIES
Consider a series .Lai, in which the terms may be positive or negative, but not equal
to 0. For each i, let
so that
lai+II = lail ri.

An examination of the sequence r1, r2, • • • gives us a convergence test which works
very quickly, in the cases where it applies.
Theorem 1 (The ratio test). If

lim ri = r < 1,
i_..oo
then 2:0 ai is absolutely convergent.
Proof Let s be any number such that r < s < 1. Then there is an N such that
i� N => ri < s.
0 r t
r;
(i�N)
(In the defi11ition of limi�co ri, take € = s - r, so that r = s - €. ) It follows that
laN+II = laNI rN < laNI s,

laN+2I = [aN+1I rN+I < laNI ss = laNI s2;
and in general, given
it follows that
By induction,
laN+;I < laNI s1 for every j.
Therefore
"' "' . "' . 1
2 laN+1I �2 laN[ s' = laNl2s' = [aNI -- < oo.
i=O i=O i=O 1 - S
It follows that
and so
co N-1 co
2 fail = 2 [a;[ + 2 lail < oo,

i=O i=O i=]\T
What we are really using here is a comparison test between the series 2 la;I and
a geometric series; the comparison does not necessarily work for the first few terms,
but it does start working after a certain point; and this is good enough to tell us what
we want to know.
In Section 10.2, Theorem 9, we showed by a comparison test that I�o ( l/i ! ) is

convergent. The ratio test gives this result very quickly. We have
1
a;=�·
l.
Qi+l i! 1
r-=-=
' - -- '
a; (i + 1)! i + 1
and so lim;�co r; = 0. It follows that the series converges.
There are simple cases in which a series converges, but in which convergence
cannot be established by the ratio test. Consider
co
1
!-:;.
i=l I
which is known to converge. Here
a;+1 i2
r; = - = .
a; (i + 1)2
Therefore, while r; < I for each i, we have
1
I.1m r; = ]"1m = 1,
i-co i- co (1 + (l/i)]2
and so the ratio test does not apply. And Theorem 1 cannot be generalized to take
care of these cases, because if r; ---+ I it may easily happen that the series diverges.
This happens for
co 1 i
I -: oo, r;= -----+ 1.
i=l l i +l
An even simpler case is
co
I c-1)i= 1 - 1 + 1 - 1 + · · ·
i=O
Here
r; = 1(- l)i+l/(- l)il = I for every i,
so that ri ---+ I automatically, but the series diverges.

On the basis of the ratio test, we can derive a more general result for power series:
Theorem 2. Given the series
where a; � 0 for every i. Suppose that
a;+
Jim
t-+ro
I I
ai
i = L.
If L = 0, then the series is absolutely convergent for every x. If L > 0, then the
series is absolutely convergent for
lxl < 1/L.
Proof For x = 0, there is nothing to prove. For each x ¥- 0, we have
Therefore
lim ri = lxl · L.
If L = 0, then ri --+ 0, no matter what x may be. If L > 0, then limi�oo ri <
whenever lxl < I/L. In either case, the series converges absolutely.
By the first half of Theorem 2, we conclude that
converges absdlutely for every x. By the second half of Theorem 2, we see that
converges absolutely for lxl < 1. In each of these cases, the sum of the coefficients
forms a convergent series. But the theorem also applies in cases where the sum of the
coefficients diverges. Consider
CXl
I i11'iXi.
i=l
Here
ai ci + 1)11'i+l
1.Im
.
1.-+00
I I
+l
-
Qi
= Im
1.
n-+OO . i
l'TT
= 11'.
Therefore the series converges absolutely whenever lxl < 1/71'.

If the ratio rn approaches a limit which is greater than 1, then the series _L ai
always diverges. The reason is that in this case we have an N such that
i � N => r; > 1.
Therefore lai+11 > lail for i � N, and so after a certain point the sequence la1J,
Ja2J, . . . becomes an increasing sequence. Therefore ai cannot approach 0. This
observation enables us to add something to the conclusion of Theorem 2.
Theorem 3. Given the series _L�o a;xi, with
a;
Jim I I
i-+oo ai
+1 = L.
If L = 0, then the series converges absolutely for every x. If L > 0, then the series
converges absolutely for lxl < l/L and diverges for Jxl > 1/L.
This theorem can be adapted to take care of cases in which some terms of the
series are equal to 0. For example,
CXl
I c-1)ix2i.
i=O
Setting x2 = y, we get
00
.L (-1)y\
i=O
which converges absolutely for IYI < 1 and diverges for IYI > 1. Therefore the given
series converges absolutely for /xi < 1 and diverges for /xi > 1. Similarly,
oo
x2i+1 x2i
oo
= x_L--:-.
i=O 2
.L i
i=O 2'
-
Here
1
I I
. ai+l
lIm - =- .
i-+ oo a; 2
Therefore the series converges absolutely for x2 < 2 and diverges for x2 > 2.
Some more observations about Theorem 3 are in order.
1) The theorem applies only to the case in which /ai+1/a;/ approaches a limit. This
usually happens for series which are describable by simple formulas. But for series
in general it should be regarded as a remarkable accident. Suppose, for example, that
we start with
00
�
£.. x'
.
= 1 + x + x2 + · · ·
i=O
Here a; = 1 for every i, and so r; = 1 for every i. We now divide xi by i! for every
even i. This gives
oo
x2 4
.L b xi 1 + x+ + x3 + � +
i=O i
= -
2! 4!
The series still converges, for /xi < 1, but the ratio approaches no limit at all.
2) The theorem tells us that the series converges everywhere on the open interval
(-1/L, l/L), but it tells us nothing about what happens at the endpoints of the
interval. In fact, at the endpoints anything can happen. For example, _L;:1 (xi/i2)
converges on ( -1, 1), and converges at both the endpoints. The series .L:1 ixi
converges on the same interval, but converges at neither of the endpoints. The series
_L;:1 (xi/i) converges on ( -1, 1), and converges at x -1, but diverges at x
= 1. =
The series .L:1 (- l)i(xi/i) converges on (-1, 1), and converges at x 1, but =
diverges at x =-1. For this reason, to tell where the series converges, we have to
make separate tests at the endpoints.
3) Obviously every power series L a;xi converges for x = O; the sum is a0• But
sometimes 0 is the only value of x that gives convergence. Consider .L:o i!xi. For
every x ;tf 0, we have
r; = l(i + l)!xi+i/i!xil
= (i + 1) /xi --+ oo.
Therefore the series converges only for x = 0.
4) Finally, the results that we have been getting for power series suggest a conjecture.
In every case that we have investigated, the domain of convergence of .L aixi has
turned out to be of one of the following types:
i) The entire interval ( - oo, oo).

ii) An open interval (-a, a), plus, perhaps, one or both of the endpoints.
iii) The point 0 alone.
The question arises whether these are the only possibilities. For example, is there
a series .L a;xi whose domain of convergence is an interval whose midpoint is not the
origin? We shall see, as the theory develops, that the domain of convergence of
.L a;xi is always a set of one of the forms
( - oo, oo), (-a, a), [-a, a) , (-a, a], [-a, a], {O}.
PROBLEM SET 10.6
For each of the following series, find the domain of convergence, remembering, of course,
to test the endpoints.
ro ro ro
1. _Li2xi 2. _Li3xi 3. _Li2x2i-1
i=l i=1 i=l
ro oo xi x2i
ro
4. _L ix2i 5. .L� 6. .L�

i=l i=l v'l i=l VI
oo xi ro (-x)i co
7.
;� v2i + 1 8. I -= 9. .L(3i)3xi
i= Vi - I
2 i=l
00 ro co
10. .L (3i)x2i 11. L (3i)2x2i-1 12. L (3i)4x3i

i=l i=l i=l
ro x2i-1 oo x2 i ro
x 4i
13. .L (-l)i+l ( . - 14. IC-l)i 15. L ( - l)i (2 I')I
i=l 21 1)! t=O ( 2'I )' •
t=O ·
00 00 ro
16. _Lhi 17. _Li-ixi 18. ,L (Tan-1 i)xi

i=l i=l i=l
19. I (-l)i�
i=l i2 + 1
00
20. .L i(2x - l)i (Does the answer to this one contradict Theorem 3 ?)
i=l
oo ei
21. .L-:--- (x - 4)i (Same query as for Problem 20.)
i=ll3
oo (x - 2 )i
22. .L . (Same query as for Problem 20.)
i=l l
23. Show that ,L:1 (sin i)xi is absolutely convergent when Jxl < 1.
24. Prove the following theorem:
Theorem. If ,L;':1 a; is absolutely convergent, and b1b2, ... is a bounded sequence,

then ,L:1 a;b; is absolutely convergent.
10.7 Power Series for exp, sin, and cos 463
*25. Show that there are infinitely many integers i for which sin i > t.
*26. Show that ,L;:1 (sin i)xi is divergent when !xi > I.
(The ri;;:sults of Problems 23 and 26 show that for this very irregular series, the domain
of convergence is still of one of the types described by Theorem 3.)
*27. You may have noticed that the number 1 has come up very often as an endpoint of our
domains of convergence. The following theorem helps to account for this:
Theorem. Let p(i) and q(i) be polynomials in i, of any degree, with q(i) never equal to 0.
If a; = p(i)/q(i), then
converges absolutely for !xi < 1, and diverges for !xi > 1.
Prove this theorem.
10.7 POWER SERIES FOR exp, sin, AND cos
Theorem A of Section 10.5 asserts that power series can be differentiated a term at a
time. That is, if
co
f(x) = ,L a;xi ( -r < x < )

r ,
i=O
then
co
f'(x) = ,L ia;xi-l ( -r < x < r).

i=l
We shall use this to find a series for the exponential function. We start by assum
ing that ex can be expressed in some way as a power series, so that
co
f(x) = ex = ,L a;xi = a0 + a1x + · · · ,
i=O
for some sequence of coefficients a0, a1, . . . . On any open interval whf"re this works,
we have
co
;
f'(x) = ,L ia;xi-l = a1 + 2a 2x + · · · + ia;xi-l + (i + l)a;+1x + · · ·
i=l
It must be true thatj'(x) =/(x), and/(O) = 1; and so we want to find a sequence
of coefficients a0, a1, a2, • . . which gives these results for the series. This is easy: we
want
which gives f' (x) = f (x); and we want a0 = 1, which gives f (0) = 1. Thus
a0 = 1, a1 = a0/ 1 = I, a2 = a1/2 = t, a3 = a2/3 = 1/ (2 · 3) ;

and, in general,
a; = 1/i!.
This can be checked by induction. For i = 0, I, 2, the formula a; = I/i! holds true.
And
1 G; 1 1
- .
i+l - i + 1 - (i + l)i! - ( + 1)!
a -- => a - -
i i!
--
- i
This proceeding does not prove that
(1)
because we started off with an unproved assumption that ex had some power series
expansion. But now that we know what series to examine, it is very easy to show that
Eq. (1) holds. By the ratio test, the series on the right-hand side converges for every x.
It therefore defines a function g. Thus
oo i
x
g(x) =I-:- (-oo < x < oo).
i=O l !
a in such a way that g' = g and g(O) = l.

We chose the coefficients ; We need to
show that g(x) = e"' for every x. For each x, let ef;(x) = g(x)/e"'. Then
e"'g'(x) - g(x)e"' 1
ef;'(x) = = - [g'(x) - g(x)] 0.
2x
e e"'
=
Therefore ¢ is a constant, and ef;(x) = ef;(O) for every x. But ef;(O) = I. Therefore
g(x)/e"' = I, and g(x) = e"',

What makes this scheme work is the fact that the functionf(x) = e'" is completely
described by the conditions f' =f, f(0) = 1; no other function satisfies these con
ditions. Thus we have
Theorem 1. e"' = L�o (xi/i!).

Setting x = I we get
Theorem 2. e = L�o (1/i!).

This series converges so fast that some people enjoy using it to calculate e �
2.7182818, correct in the seventh decimal place.

We now want to get a series for the sine. As before, we start by assuming that our
problem has a solution, and then we try to find out what form the solution must take.
For f(x) = sin x we have
f'(x) = cos x, f"(x) = -sin x.
Therefore
f(O) = 0, f'(O) = I,
and
f"(x) = f(x) for every x.
-
Thus if
a;xi a0 + a1x + ·· · .
co
.
sm x= �
£..
i=O
=
we must have a0 = 0, and a1 = 1. Now
f'(x) = L ia;xi-1,
co
i=l
f"(x) = L i(i - l)a;xi-2

O')
i 2
= 2a2 + 3·2a3x + ··· + i(i - l)aix -
+ (i + l)ia;+lxi-l + (i + 2)(i + l)a;+2xi + · · ·
To getf" = -f, we want
(i 2)(i + l)ai+2 -a;, or a;+2 = -

(i + l)(i + 2)
+ =
Since a0 = 0, it folJows that every even-numbered coefficient a2i_is also equal to 0.

The odd-numbered coefficients are
a1 = 1,
a1 a1 1
as= ---=
(1 + 1)(1 + 2) 2·3 3!'

as 1 1
a5 = + -
(3 + 1)(3 + 2) 3! . 4. 5 5! '
and in general
i 1
- (-l) .
a2i+l -
(2i + 1)!
To check this by induction, we note that
i 1
a2i+l -
- (-l)
(2i + 1)!
a2i+l
ac2i+i>+2 -[
(2i + 1) 1]· [(2i 1) + 2]
=>- =
+ +
. 1 1
a2<i+i>+i = (-l)'( -l) .
(2i + 1)! . (2i + 2)(2i + 3)
1
- c l)i+l c -1Y+i
i
(2i 3)! [2(i + 1) + 1]!

-
Therefore, if there is a series for the sine, the series must have the form
co
x2i+1
.
(x) = ·
g i � (-l)' (2i + l)!
We need to show that g(x) =sin x for every x.

Let h(x) =g'(x). Then we know that
I) g' = h,
2) h' = -g,
3) g(O) = 0,
4) h(O) = I.
It ought to be true that
g(x) =sin x, h(x) =cos x.
If so, the function
cp(x) = [g(x) - sin x]2 + [h(x) - cos x]2
must be equal to 0 for every x. And conversely, if cp(x) =0 for every x, it follows
that g(x) =sin x and h(x) =cos x. Now
cp'(x) = 2[g(x) - sin x][g'(x) - cos x] + 2[h(x) - cos x][h'(x) + sin x]
=2[g(x) - sin x][h(x) - cos x] + 2[h(x) - cos x][-g(x) + sin x]
=0 for every x.
Therefore <P is a constant. But
cp(O) = [O - 0]2 + [l - 1]2 = 0.
Therefore cp(x) =0 for every x, which was to be proved. Thus we have:
Theorem 3.
oo
. x2i+1 x3 xs x7
sin x = I (-1)' =x -- + - - - + · · ·
i=O (2i + 1) ! 3! 5! 7!
By differentiation,
oo
i (2i + l)x2i oo
x2i
COS X = .? (-1)
i-0
(2.l + l) .I
= .? (-1)i (2 . ) I
t-0 l .
•
Thus:
Theorem 4.
oo . x2; x2 x4 xs
cos x =I c - 1) - = 1 - - + - - - +
' · · ·
i=O (2i)! 2! 4! 6!
Obviously, the series that we have been developing in this section can be used
for calculating the values of the corresponding functions. In fact, this is the way
people arrived at the values .that you find in the tables of exp, sin, and cos. And the
series can be adapted, in simple ways, to handle a variety of related problems. For
example, consider
o
r .5 e
'"2
dx.
Jo
If we could get a simple formula for a function F such that
"'°
F'(x) = e ,
then the integral could be expressed as F(0.5) - F(O). There is no such simple formula.
But we can express such an Fas an infinite series, in the following way. We know that
ro i 2
x x x
e = L� = 1 + X + -+ · · ·
i=D l ! 2!
ex'
Therefore
ro 2i 4
x 2 x
= L--;-- = 1 + x + - + ...
i=O l ! 2!
..
Integrating a term at a time, we get a function
a " ro 2i+1
x x x
F(x) = x + - + -- + · I --- .
-i=oi!(2i
3 5 2! + 1)
ex'.
·
Evidently
F(O) = 0, and F'(x) =
Therefore
f' e
t'
dt = F(x),
and so, using the series for F, we can calculate F(t) approximately, with an error as
small as we please.
PROBLEM SET 10.7
Find a series for each of the following functions. In each case, name the interval on
1. f(x) x In (x 2. f(x) x2 In (x2 + I).

which you know that your series converges to the given function.
3. f(x) 4. rxf(t) dt, where f is as in

= + I). =
2.
2
5. f(x) 2x. 6. f(x) x sin x.

x ln (x + I).
Jo
=
Problem
7. f(x) sin (x/2). 8. f(x) cos 2x.

= sin =
G). 10. f(x)

= =
11. f(x) 12. f(x)

9. /(x) = cos = sin x cos x.
'n-'x
;e 0
3 "'3 "'
13. f(x) dt. 14. f(x)

= x e • = xe - x.
C for x 0
l"' 3 t3
� for x
1
= t e =
15. F(x) lxf(t)dt, 14. 16. f(x) x ;e 0

=
(e"'
0
-
for x
x
wheref is as in Problem
t
= =
F(x) i"'f(t)dt, where/is as in Problem 16. 18. f(x) x ;e 0

for x =
r 0
17. = = ; for
for x =
19. F(x) = f'f(t)dt, where/ is as in Problem 18. 20. f(x) = cos2 x.
21. f(x) x3 cos2 x. 22. F(x) = { f(t ) dt where f i s

x ,
J as in Problem21.
o
=
23. f(x) = cos2 x - sin2 x 24. F(x) = x cos2 x + x sin2 x.

25. Find a series for a function f such that (1) f' tf and (2) f(0) I. Either before or
= =
after finding the series, find an elementary formula for such a function f
26. Find a series for a function f such that (1) /' (x) = 2/ (x)/x for every x � 0 and (2)
/(0) 0. =
27. Is there only one function satisfying the conditions of Problem26? Why or why not?
28. Get a formula for Dixi, where D1 denotes the ith derivative of the function f
29. Get a formula for Dix1, valid for i < j.
30. Do the same, for the case i > j.
*31. Given f(x) I:o a xi. Get a formula forJ(i>(O). (Here /( i ) denotes the ith derivative
i
=
of f )
*32. Is it possible that there are two different power series for the same function, valid on
the same open interval I? That is, given
<XJ <XJ
f(x) L aixi = L b;xi on I,
i�o i�o
=
does it follow that ai = b; for each i? Why or why not?

*33. A function f is called real-analytic on an interval I if f can be expressed as a power
series I:o a xi. Does there exist a real-analytic function /, on an interval (-a, a),
i
such that /(i) (0) (i !)2 for each i? Why or why not?
=
10.8 THE BINOMIAL SERIES
It is possible to show, by induction, that if n is a positive integer, then
n(n - 1) an-2 2 +
(a + bt = an + nan-lb + b
2
n(n 1) · · · (n
an-ibi + . . · + bn.
- - i + 1)
+
l·1.
Here the coefficient of an-ibi can be written more briefly as
(n) n! n(n - 1) · · · (n - i + 1)
·1
1.
=
i i !(n i)!
The induction proof of the binomial theorem depends on the identity
You may have seen this proved. In any case, we shall not stop to prove it now,
10.8 The Binomial Series 469
because the elementary form of the binomial theorem is a corollary of a more general
result which we shall prove presently.
We would like to generalize the familiar binomial formula
(a + br = i� G) an-ibt
to take care of the case in which n is not an integer. That is, we want a formula for
(a+ b)k, where k is any real number. The following observations are obvious:
1) Fork = 0, we have (a+ b)k = 1, and our problem is solved. We may therefore
assume that
k ¥ 0.
2) For the case of interest, in which k is not an integer, the exponential ck is defined
only for c> 0. (See Section 4.9.) Therefore we must assume that
a+ b> 0.
3) For a = b, the problem has an immediate solution: (a+ b)7' = (2a)k = 2kak.
We therefore may assume hereafter that
a-:;!:. b.
And we want to assume this, because the case a = b does not fit the pattern that is
going to emerge.
4) It is now a matter of notation to suppose that
a> b.
We let x = bfa, so that
a+ b = a(l + x), and
If we had \x\ = \bfa\ � 1, then either b � a or b � -a; and these possibilities are
ruled out by conditions (2) and (4). Therefore \x\ < I, and our problem takes the
following form:
Problem. Given k -:;!:. 0, and
f(x) = (I + x)k (\xi < I).

Find a formula for f (x), analogous to the binomial formula.
Our past experience with sin, cos, and exp suggests that we should investigate
the relation between f(x) = (1 + x)k and its derivatives, and use the results in the
investigation of the series. Now
j'(x) = k(l + x)7'-1.

Therefore
(1 + x)f'(x) = kf(x).
CQ CQ
f'(x) = L ia;xi-l = L ia;xi-1.

i=O i=l
Therefore
C()
xf'(x) = .L ia ixi.
We want to express (1 + x)J'(x) as a series, and so we need to express/'(x) in the

form .L bixi. For this purpose we use a trick. Let j = i - 1, so that i = j + 1.
This gives
co 00
f'(x) = .L (j + l)a;+1x1 = L (i + l)a;+ixi.

1=0
The equation (1 + x)j'(x) = kf(x) now takes the form
00 00
L [(i + l)ai+i + ia;]xi = L ka;xi.

i=O
Comparing coefficients of xi, we get
(i + l)ai+i + iai = kai <=> (i + l)ai+l = (k -i)ai

k-i
<=> a;+i = i ai.
+1
Obviously
a0 = /(0) = (1 + 0)1' = 1.
Therefore
a0 = 1;
k-
a1 = O a = k·'
0+1 0
k -1 k(k - 1)
a.= -- a = ·
- 1 + 1 1 2 '
k -2 k(k-l)(k - 2) - k(k- l)(k -2)
-
a ---a -
3·2
·
3 2+1 2- - 3! '
and in general, for i > 1,
k(k - 1) . .. (k -
i + 1)
ai = .
.,
I.
We denote the fraction on the right by the symbol (�),just as in the case where k is a
positive integer. The above formula then takes the form
ai.= (7) for each i � 0,
and the net result of the above discussion is that

00
k
'
.4..
i=O
aixi = (1 + x) => a; = (7) for each i.
That is, the series that we have found is the only series that might work. To know
that our series does work, we need the following two theorems.
10.8 The Binomial Series 471
Theorem 1. The series 2::o CDxi is convergent for lxl < 1.

Proof As in Section 10.6, let
Then
1)... (k
r
·
' = I k(k _ i + l)(k
(i+l)!
_ _ i). i!
k(k-l)···(k-i+l) .
xI
Evidently
=
I ��; x I·
·
Jim r
; = Ix!.
i-too
Therefore, by the ratio test, the series converges absolutely for lxl < 1, and diverges
for lx l > 1.
Theorem 2. For every k � 0, and everyx between -1 and 1,
i� (�)xi= (1 + x?.
Proof g Let be the function which is the sum of the series. We determined the
coefficients in such a way that
l)g(O)=l;
2) (I +x)g (x) = kg(x).
'
We need to prove that
g(x)=f(x)= (1+x)k (-1 < x < I).

For this purpose, we use the same device that worked for the exponential. Let
f(x) = (1 g(x)
+ x)k
Then
A. '
x) - (1 +x)kg'(x)(1-+ gx()xzk)k(l+x)k-l
't' C _
= (1 + (1x)g+'(x)-
)
kg(x)
x k+1
=
0
'
so that cf> is a constant. And the constant is 1, becauseg(O) = f (0) = 1. Therefore

g(x)=j(x) for every x, e .
which was to be p ro v d
PROBLEM SET 10.8
1. Write a series for v'�, and find out how many terms of the series you would need
to use, to calculate v'u, correct to three decimal places.
- -
2. .a;
Do the same, for v x + 1.
3. Do the same, for V1 x + 1.
4. Do the same, for V2x + 1.
5. Let n be a positive integer. Using the definition
n!
(n - i)!i!'
write formulas for cn;1) and (i.:'.1).

6. Using the apparatus of Problem 5, show that
7. Using the result of Problem 6, show that
<1 + xr = i ( � ) xi
i=O l
"""
(The first and last terms on the right-hand side require a separate discussion. But
note that (�) <nt1), because both are equal to 1 ; and similarly that (�)
= (�!}) = = 1.)
Since obviously (1 + x)1 = (�) + (�)x1, this gives an induction proof of the elementary
form of the binomial theorem.
Find a series for each of the following functions, and discuss for convergence. You need
not test for convergence at the endpoints.
x2
8. f(x) 9
j(X) =
v'I + x . .I
v 1 +x
x
10. f(x) = xVI + x 11. f(x) .a;
1 + x2
=
12. f(x) = Sa"' Vl + t10 dt 13. f(x) = (1 + x)312
14. f(x) = f'2 VI + tlo dt 15. f(x) = {/2 + x
16. f(x) = v'3 + x 17. f(x) =f' {/2 + t2 dt

18. f(x) = (2 + x2)" (k ;6-0) 19. f(x) = r (2 t2)k dt +
x t
20. f(x) 21. f(x) = ( "' dt
V 1 + x2 J v'1 + t2
= --
22. Find a function f such that (1) (1 + x2)[' (x) f(x) and (2) f(0) = 1. Then show that
=
· the function th�t you found is the only function satisfying conditions (1) and (2).
23. Same question, for the conditions (1) f' (x) sec x = f(x) and (2) f(0) = 1.
10.9 Taylor Series 473
10.9 TAYLOR SERIES
Obviously we cannot get a series of the type ! a;xi, converging to f(x) = l /x on an

open interval containing 0, because any such series is continuous at x = 0, while
lim:t-.o+ 1/x = oo. On the other hand, there is a series
1
f C-l)ixi __ (lxl <1).
1+X
=
i=O
If we let x' = 1 + x, x = x' - 1, then the above equation takes the form
(Ix' - 11< l);

and dropping the prime we get
1 00
f(x) !(-l)'(x - 1)' (Ix - 11<1).

• .
= - =
X i=O
The series on the right-hand side is called a Taylor series, oi a Taylor expansion of
the function f about the point 1. A power series ! a;xi, of the type that we have
x =
been discussing so far, is called a Maclaurin series. Thus every Maclaurin series is a
Taylor serie�:
00 00
! a;x; = ! a;(x - O)i,

i=O i=O
which is a Taylor series, with a = 0. In this language, we may say that f(x)
1/x =
has no Maclaurin series, but it does have a Taylor expansion about the point 1.
Similarly, f(x) = In x cannot have a Maclaurin series, because at x = 0 the
function approaches - oo. But we do have a series for
oo
xi
g(x) = ln (l + x) = !c-1r1--:- (l xl<1).
i=l l
Setting x' = 1 + x, x = x' - 1, we get

(x )i
Inx' = i(-l)i+l ' � l (Ix' - 11<1);
i=l l
and dropping the prime we get
(x l)i
lnx ic-l)i+i -:- (Ix - 11<1).
i=l
=
This is a Taylor expansion of In, about the point 1.

With the obvious modifications, all our theorems for Maclaurin series hold also
for Taylor series; and to prove them in the general case, we merely translate the axes
by the substitution x = x' + a, x' = x - a. For example, Theorems A and B
of Section 10.5 take the following forms:
Theorem A'. Given

00
f(x) = ! a;(x - aY (a - r<x< a + r),

i=O
thenf is continuous and differentiable on the interval (a - r, a + r), and the deriva
tive of the sum is the sum of the derivatives. That is
ro
f'(x)= L iab - aY-1 (a - r < x < a + r).

i= l
Theorem B'. Given
ro
f(x)= L a;(x - aY (a - r < x < a + r).

i=O
Then the integral of the sum is the sum of the integrals. That is,
("'f(t) dt I ("'a;(t - 4 dt I� (x - a)i+l.

Jo i =oJo i=Ol
= =
+ 1
Our other theorems can be generalized in the same style; we treat x - a in

exactly the same way that we used to treat x. Another example is as follows. You
found, in Problem 31 of Problem Set 10.7, that if f(x) = I:o a;xi, on an open
interval containing 0, then
so that
j<n>(o)
a = .
n --
n!
An analogous formula holds for Taylor series:
Theorem 1. If f(x)= L�o a;(x - a)', on an interval (a - r, a + r), then
J<n>(a)
a = for every n.
n --
n!
Proof The nth derivative of a function described by a formula rp(x) will be denoted
by Dncp(x). We observe that
nn(x - a)i= 0 for i < n,

nn(x - ar= n!,
nn(x - a)i = i(i - 1) . . . (i - n + l)(x - a)i-n for i > J'l.
Therefore, forf(x)= L a; (x - a)i, we have

ro
fCn>(x) a · n! + L b;(x - aY-n.

n i=n+l
=
We don't care what form the b/s have, because every term of the sum on the right
hand side has (x - a) raised to a positive power, and we are about to set x = a.
This gives
so that
fCn>(a)
a =
n --
n.1
'

10.9 Taylor Series 475
We have found that for some functions the use of Taylor series in place of
Maclaurin series is a necessity. For example, l/x and Jn xdon't have any Maclaurin
expansions. In other cases a Taylor series may be preferable, even though the
Maclaurin expansion exists. The point is that .Z a ix
i
usually converges rapidly when x
is close to 0, and more slowly when xis larger. To take an extreme case, we know
that sin 10,0007T = 0, because 10,000 is even. Therefore it must be true that
.I=O (-1)i (2i +1 1)! (10,0007T)2i+l o.

i
=
But in waiting for the partial sums to get close to 0, we had better not be impatient.
In general, if we want to use a series to calculate a function numerically, we should
choose the "base point" a as close as possible to the value of x that we want to
substitute. Suppose, for example, that we have calculated In 1.5 = 0.4055. One
way to calculate In 1.6 would be to take x = 1.6 in the series
x 1
ln l(-l)i+1C � y
=
x
i l
=
But the convergence of the series
.I c-1)i+l co"6)i
i=l l
is slow. We therefore use the base point 1.5. Thus

00
In x .2 a;(x - 1.5t
i=O
=
For f (x) = Jn x, we have

f'(x) = 1/x = x-1,
f"(x) = ( -l)x-2, • • • ,
fil(x) = (-l)i+l(i - l)!x-\

fi)(l.5) ( -l)i+l
ai = = . (1.5)-i (i > 0),
.,
L !
j(O)(l.5) = 0.4055.
Therefore
i
(- ) +l ( x - 1.5);·
Jn x 0.4055 + I �
1.5
=
.
t=l l
For x 1.6, this gives

� ( - i.y+i (1-)i,
=
In 1.6 0.4055 +
15
=
i=l l
which converges much more rapidly.

Note that the above derivation tacitly assumes that In has a Taylor expansion
about the point a = 1.5; if it has, then the coefficients must be given by the formula
f(a) .
ai = -,- .
L
It is a fact that if a function J has a Taylor expansion
<Xl
f(x) = ! ai(x - a)\

i=O
converging on an interval Ix - al < r, then any other point b of the interval can also
be used as a base point, giving an expansion
<Xl
f(x) = ! b;(x - b)i,

i=O
which converges on some interval containing b. But the proof would be hard, in the
present context, and should be postponed until we can use the theory of functions
of a complex variable.
PROBLEM SET 10.9
For some of the functions in the first twelve problems below, it is a practical proceeding
to derive a general formula for pn>(a), and use the formula to calculate the coefficients ai
in the series ! a;(x - a)i. In each such case, calculate the coefficients by this method. In
cases where the derivation of the general formula seems unreasonably difficult, merely cal
culate the first three terms of the series.
1. f(x) =sinx, a=0. 2. f(x) =tan x, a=0.

3. f(x) =tan x, a = 1T. 4. f(x) =cos x, a=21T.
5. f(x) =Tan-1 x, a = 0. 6. f(x) =Tan-1 x, a=1.
7. f(x) =ex, a=0. 8. f(x) =e"', a=1.
9. f(x) =In (2 + x), a=0. 10. f(x) =e"', a=any number.
11. f(x) =Jn (1 + x2), a=0. 12. f(x) = sin x, a = 1.
13. This is a separate problem, and it requires you to think of a trick. Given that In 1.4
0.3365, find a way, using series, to calculate In 2, correct to three decimal places. (To
four decimal places, In 2 = 0.6931.)
14. Let !%1 a; be any series. For each i, let
b;= i(a; + la;I),

ci=i(a; - la;I).
Show that if ! b; and ! c; both converge, then 2: a; converges absolutely.
15. Let 2: a;, 2: 2: c; be as in Problem 14. Show that if ! a; converges and 2: la;I
b;, oo,
then
and
LC; = - oo.
*16. Let n1, n2, • • • be a sequence of positive integers in which each positive integer appears
exactly once. That is, the numbers n1, n2, • • • are the integers 1, 2, 3, . . . arranged in
some order. For each series 2:%1 a;, we can then form a "rearranged series" !%1 an,,
in which the same terms appear in some order. The following theorem is a sort of
"commutative law of addition" for positive series.
10.10 Taylor's Theorem. Estimates of Remainders 477
Theorem. If ai > 0 for each i, and 2 ai = A, then every rearrangement of 2 a;

converges to the same sum.
Prove this.
*17. Show that
has a rearrangement which converges to 0. (Thus the "commutative Jaw for infinite
sums" does not hold in general.)
*18. Show that for every number k there is a rearrangement of the above series which
converges to k.
10.10 TAYLOR'S THEOREM. ESTIMATES OF REMAINDERS
In the preceding section, we showed that, if a function is expressible by a Taylor

series, with
co
f(x) = .2 a;(x - aY (a - r < x < a + r),
i=O
then the coefficients ai are given by the formula
i
J< >(a)
ai =
- .--
1 .
i.
Using the formula, we can write down a series. But there are three questions which
it is natural to ask:
1) For what values of x does the series converge? (We recall that Tan-1 x is defined
for every x, but its series converges only for -1 < x � 1.)
2) Does the series converge to the function f that we started with?
3) If we use a partial sum
n (i)(a)
J
Sn(x) = .2- . - (x - a)i
i=O l !
as an approximation ofj(x), what is the error? For this, we need an estimate of the
"remainder function"
Rn(x) =
/(x) - Sn(x)
n /(i)(a)
= f(x) - 2-. (x - aY, -
i=l l !
Partial answers to these questions are given by the following theorem.
Theorem 1 (Taylor's theorem). If f has n + 1 derivatives, on the interval [a, x] or

the interval [x, a], then
f(n+l)(.X)
Rn(x) = (x - ar+1'
(n + 1) !
for some x between a and x.
The proof is artificial, and hard to remember. We regard x as a constant; and for
each t we let
i
f (t)
f(x)- F(t) i
. (x - t) .
= f
i=O l !
Here we have simply replaced a by t in the formula for R (x). For t x we have
n
=
r<ol(x)
F(x) = f(x)- ·-- = f(x) - f(x) = 0.
O!
For t a we have F(a) R (x). Since
n
= =
f(t) f'(t) f"(t) J(n)(t)

F(t) = f(x) - -- - (x - t)- (x - t)2 - ·- • • • - -- (x - tf,
O! 1! 2! n!
we have
F '(t) = -f'(t) - [f" (t)(x- t)- f'(t)]
- [
f"'(t)
2!
(x - t)2-
f" (t)
2!
2(x- t) · J
- [ f(n+l)(t)
n!
J(n)(t)
(x- tr- -- . n(x - tr-1 .
n! J
Here all terms cancel out, telescopically, except the first term in the last bracket;
and so
f(n+l)(t)
F'(t) = - (x - ir.
n!
Now let
(x - tr+i
G(t)-
- �--
(n + 1)! '
so that
-(x - tr
G'(t) =
n!
To the functionsF and G, on the interval between a and x, we apply the parametric
mean-value theorem. (This is Theorem 2 of Section 9.2.) It gives
F(x)- F(a) F'(x)

-
_
G(x)- G(a) G'(x)'

for some x between a and x. Since F(x)
· = G(x) = 0, this means that
-F(a) F '(x)
=
-G(a) G'(x)
And
10.11 The Complex Number System 479
for every t. Therefore
F(a) = +1
f<n >(x).
G(a)
By definition of G(a) and F(a), we have
= = f(n+l)(-
x)
R n(x) F(a) = J<n+i>(x)G(a) (x - ar+l'
(n + 1)!
In some cases we can use this theorem to prove that a formal power series con
verges to the expected function. For example, we may be able to find a number M
such that
IJ<n+i>(x) I � M
for every n and every x between a and x. In such a case it follows that Rn(x) --+ O·
and f (x) is the sum of its formal Taylor series. Most of the time, however, estimates
of (n + l)st derivatives are hard to come by. For example, the calculation ofj <n> (x)
is unmanageable for the function
f(x) =
- 1 -,
1 + x3
even though we can easily see what the Maclaurin series is:
_1_ =I c-i)ix3i (lxl < 1).

X3 i=O
1 +
It follows, of course that
JC3i>(O) = ( -1)i(3i) ! ,
and thatJ<nl(O) = 0 if n is not divisible by 3. But this does not give us any information
about J<n>(x) for other values of x.
PROBLEM SET 10.10
1 through 6. In at least six of the first twelve problems in Problem Set 10.9, it is easy to
get an estimate of J<n>(x), and then show by Taylor's Theorem that the series converges to
the given function. Identify these six cases, and carry out the process.
10.11 THE COMPLEX NUMBER SYSTEM
Formally speaking, complex numbers are numbers of the type
z =a+ bi,
=
where a and bare real numbers, and where i is some sort of number such that i2 -1.
Granted that there is such a number system, and that it obeys the same manipulative
rules as the real number system, the equation i2 = -1 gives all that we need to
perform calculations. For example,
(a + bi)2 = a2 + 2abi + b2i2 =(a2 - b2) + 2abi;

(a + bi)(c+ di) =(ac - bd) + (ad + bc)i;
1 1.)2.
( + = z;
/2 i
J 2
,1 1 ·)2 .
= I;
(_;2 -
Jl
I -
i4 =(i2)2 =1;
i l0,001 = i;
G �3 + iy = i(l + _13 i)3

= i(l + 3J3 i + 3 . 3i2 + 3J3 i3)
= i(l - 9 + 3J3 i 3J3 i)
-
= -1.
Obviously 0 = 0+ 0i has no reciprocal. But if a + bi ¥- 0, then a+ bi has
a reciprocal in the complex number system. To see this, note that if a + bi ¥- 0,
·
then a and b are not both = 0. Therefore a bi ¥- 0, and -
1 1 a - bi a - bi
--- = --- . --- =
a + bi a + bi a - bi a2 - (bi)2
a - bi a + -b
= = i
a2 + b2 a2 + b2 a2 + b2
=A+ Bi.
This calculation begins with the assumption that a + bi and a - bi have reciprocals,
but once we know the answer, it is easy to check:
a -b )
(a+ bi)(A + Bi) =(a + bi) ( + i
a2 + b2 a 2+ b2
(a + bi)(a - bi) a2 + b2
= = =1.
a2 + b2 a2 + b2
Therefore A + Bi is the reciprocal of a+ bi.
There are several ways to define the set of complex numbers, as a mathematical
system, and check their properties. One such method is explained in Appendix J.
Meanwhile we shall regard the complex numbers as known, and calculate with them,
using the familiar laws of algebra and the fact that i2 = 1
conjugate of the complex number
- .
The
z =a+ bi
is the number
z =a - bi.
The ab solute value of z is
lzl = .Ja2 + b2•

By straightforward calculations, we get the following.
Theorem 1. For all complex numbers z, Zi. z2, we have:
z = z,
z + z is a real number,
z·z is a real number,

2
!zl = z z, ·
Proofs.
z = a + bi, z = a - bi, z = a - (-b)i = a + bi = z;
z + z = a + bi + a - bi = 2a;
z · z = (a + bi)(a - bi) = a2 + b2 = !zl2;
z1 + z2 = a1 + b1i + a2 + b2i = a1 - b1i + a2 - b2i
= (a1 + a2) - (b1 + b2)i = z1 + z2;
Z1Z2 = (a1 + b1i)(a2 + b2i) = a1a2 - b1b2 + (a1b2 + a2b1)i;
z1 z2 = (a1 - b1i)(a2 - b2i) = a1a2 - b1b2 - (a1b2 + a2b1)i = z1z2;
•
lz1z2I = .Jz1z2z1z2 = .Jz1z2z1z2 = .Jz1z1 .Jz2z2 = lz1I

· · lz2I·
(Note that in the last of these calculations all the radicands are real and � 0, as they
should be.)
So far, we have been treating all these ideas algebraically. We shall now interpret
them geometrically, plotting each complex number z = x + yi as a point (x, y) in a
coordinate plane.
y
Y ---------, z=x+yi
I
I
I
I
I
I
�+--��x,--�--x
Thus real numbers z = x fall on the x-axis; we shall think of this as the real axis.
And "pure imaginary numbers," of the form z = iy , fall on the y-axis; we shall think
of this as the imaginary axis. This explains the labels on the axes in the figure below.
Evidently z is the reflection of z across the x-axis. If you reflect twice, you get back
where you started; and this is the geometric meaning of the equation z = z. As the
z=x-yi.
figure suggests, lzl is the distance to z from the origin; the reason is that
lzl = ,Jx2 + y2,
which gives the distance. More generally, lz1 - z21 is the distance between z1 and z2.
Z2
I
I
I
____ _r:jI
I
For z1 = x1 + y1i and z2 = x2 + y2i, we have

Z1 - Z2 = (X1 - X2) + (Y1 - J2)i,
SO. that
2
lz1 - Z2I = .J (x1 - X2)2 + (Yi - J2) ,
which gives the distance.
If z = x + yi, then x is called the real part of z, and is denoted by Re z. The
number y is called the imaginary part of z, and is denoted by Im z. Thus
z =x + yi = Re z + i Im z.
It is easy to check that
Rez = t(z + z),
Im z = (1/2i)(z - z).
Since i2 = -1, we have
. 1 1
l = - - , -i, and Imz = £(z - z ).
i 2
These formulas enable us to connect complex numbers with the geometry of our
coordinate plane. lzl = 1 is the circle with
For example, the graph of the equation
center at the origin and radius 1; and the graph of the equation lz - z01 = a is the
circle with center at z0 and radius a. The vertical line through the point (1, 0) =
1 + 0 i is the graph of the equations
·
x= 1 <=> Re z = 1
¢;>- t(z + z) = 1
¢;>- z + z = 2.
In the following problem set, you will be asked to carry out a variety of such
processes. For short, we shall use the term C-equation to describe an equation in
which complex numbers are the only variables. Thus the vertical line discussed above
is the graph of the C-equation z + z = 2; and a certain circle is the graph of the
C-equation zz = 4.
PROBLEM SET 10.11
Reduce each of the following expressions to the form + yi.

(1 1 r
x
I. (1 + i)4 2. (1 - i)4 3. v2
+
v2i
4. (-1 - _1 r (1 v3
- +-1 6. 3
v2 v2
i 5.
2 2 J G- �
1
J
1
v3
- +- 8. (�3 - �iJ 9. ( _1 +
(2 21
J v2 v2 J
- _1
7. i
10. (v2 - v2 i)8 11. (1 + i)2 12. ( -1 + i)3
13. (2 + i)2 14. (1 - 2i)2 15. (v3 + i)4
16. (1 - v3 i)3 17. i +2

18. 2 - i
19. 1 2i +
20. 2i 1 1
2 .
I + 3i
1
-
22.
i - 3i
23.
2 + v3 i
24. - 2 v3 i
25.
v3 + 2i
26. - v3
1
27.
i2 + i + 1
1
1 29. i2 1
2i
28.
i3 + j2 + i -1 - i + 1
30. 1)3
(i +
---
31. 11 +i 3i 33. 1 -- 3i
2i 1 +
2 + 2
32.
- i 2i
34. is + i4 + i3 + i +I
35.
i4 + i3 + i2 + i + I
36. Show that lz"I = lzl", for every z.
37. Show that zn zn, for every 38. Show that l/z (l/z), for every 0.
Sketch the graphs of the following C-equations.
= z. = z �
39. Re z +Im z Re z - = I.
41. Re z =Im z. - 11 = I.
= I. 40. Im z
43. - -
42. I=
lz II < I. 44. lz II > I.
45. \z - i\ = t. 46. \z+i\ < t.

47. z+i = 0. 48. z+i>l.
49. \z\2 = 4. 50. \i\2 = 4.
51. z2 = 4. 52. z4 = 4.
53. z-i=l. 54. \z+i\ =
< l_
2•
55. \z - i\ = 1. 56. \i - z\ + \z+i\ � 2.
57 . Ii - zl + \z+ii = 2.
The following two problems require the theory developed in Appendix J.
*58. Find a polynomial q(x) such that (x5 + x4+ x3+ x2+ x + l)q(x) = 1 mod 1+ x2
*59. We found that if z =p(x) -¢ 0, then z has a reciprocal in C. In the language of
congruences, this says that if p(x) ¢ 0 mod 1 + x2, then there is a polynomial q(x)
such that
p(x)q(x) = 1 mod 1 + x2•
Similarly, if
p(x) - p'(x) = r(x)(x2+ x + 1),
then we write
p(x) = p'(x) mod x2+ x + I.
Show that if p(x) ¢ 0 mod x2+ x + 1, then there is a polynomial q(x) such that
p(x)q(x) = 1 mod x2+ x + 1. (In fact, the congruence classes of polynomials modulo
x2+ x + 1 form a field. All the conditions for a field can be checked, in the same way
as for 1 + x2, except for the existence of reciprocals.) Is it true that every polynomial
is congruent mod x2+ x + 1 to some linear polynomial ?
*60. In Section 10.8 we deduced the binomial formula
(a +hr =I (�)an-ibi
i=O I
from a more general result, using the methods of calculus. But the methods of Section
10.8 do not, as they stand, apply in the complex domain. Show, by induction, that
(1 + zr =i� G)zi = 1 + U)z+ (�)z2 + · · · + ( n � 1)zn-1 +z n,
z. Then show that for all complex

n
for every positive integer and every complex number
numbers u and v,
(u +vr = i (n·)un-ivi.
i=O }
10.12 SEQUENCES AND SERIES OF COMPLEX

NUMBERS. THE COMPLEX EXPONENTIAL FUNCTION
We know that for sequences x1, x

2
, • • • of real numbers,
lim Xn x � lim lxn - xi 0.

n_.co n-+oo
= =
The second of these conditions also has a meaning if the xn's and x are complex
numbers, because the absolute values lxn - xi are real in any case. We use this idea
to define limits for sequences of complex numbers.
10.12 Sequences and Series of Complex Numbers. The Complex Exponential Function 485
Definition. Let z1, z2, . • • and z be complex numbers. If
lim lzn - zl = 0,
n-+oo
then
limzn =z.
We can test for limits by examining the real and imaginary parts of the sequence
separately.
Theorem 1. For each n, let

Zn= Xn + Yni.
If the sequencesx1, x2, • • • and y1, y2, • • • are convergent, then z1, z2, ••• is convergent,
and
limZn =lim Xn + i lim Yn·
Proof Let
x = limxm y =limym
n-+oo n-+ oo
so that
lim(xn - x) = lim(yn - y) = 0.
We need to show that

lim(xn + Yni) = X + yi,
n-+ oo
which means, by definition, that
lim (J xn + Yni ) - (x + yi)I = 0.

n-+ oo
This is trivial:
2 2]112 = 02 2 1'2
lim lxn + Yni - X - yil = lim [(xn - x) + (Yn - y) ( + 0 ) = 0.
n-+ oo n-+ oo
The converse is also true.
Theorem 2. If limn_,.00 (xn + Yni) = x + yi, then limn_,.00 Xn=x and limn_,.oo Yn= y.
(Proof?) Once we have Theorems 1 and 2, it is a routine matter to verify that

the limit of a sum, product, or quotient is the sum, product, or quotient of the limits,
just as for real sequences. We shall use these rules without comment.
Just as for real numbers, we define convergence of infinite series in terms of
convergence of sequences.
Definition. Given a series ,L;:,0 z1. For each n, let
If
limSn= S,
n-+ oo
then we say that L�o z1 converges (to S), and we write

00
:L z1 S.
i=O
=
For real series, we found that if ! lxil converges, then ! xi also converges. The
same is true in the complex domain.
Theorem 3. If L�o lz11 converges, then !�o Z; converges.
Proof For eachj, let
Then
and similarly
for eachj. Therefore
and
Therefore ! x1 and L y1 converge, to su�s A and B. Therefore
! (x1 + y1i)
n n n
LX; + ilim LYi

n
:L z1
i=O i=O n-+ oo =O
lim
j=O
Jim = lim = =A +Bi.
j
n-+ oo n-+ co n-+ co
The simplicity of this theorem, and of its proof, are misleading: the theorem is
powerful. It gives immediately:
Theorem 4. L�o (zifj!) is convergent for every z.
This is so because
j
1 I �j! I 11�1
i=O j=O j !
=
< 00
}
for every real number lzl. This enables us to extend the domain of the exponential
ex = x to the entire complex plane:
exp
Definition. ez = exp z = L�o (zi/j!).
For the case in which z is a pure imaginary number i8, we can express e• in
another form:
ei6 = :L ( i8)j = :L (i8) ( i8)2k+l

00 2k 00 00
:L
i=O j! . k=O (2k)! k=O (2k + 1)!
- - +
Now
and
Therefore
. oo c 0ke2k . oo c 0ke2k+l
+ i:L��-
_ _
e'6=:L
k=O (2k)! k=O (2k + 1)!
10.12 Sequences and Series of Complex Numbers. The Complex Exponential Function 487
This gives:
Theorem 5. For every real number fJ,
ei8 = cose + i sine.

Settinge = Tr, we get a famous equation due to Leonard Euler:
It is easy to see that if z = ei0, then izl = I. The reason is that
lei81 = !cose + i sine1 = -Jcos2e + sin2 e.
Conversely, if izi = 1, then z = ei8 for some e. Here z = x + yi, x = cose, and
y = sine.
\
\
'\
'-
,_
More generally, every complex number can be expressed in the form
r = izl � 0.
To see this, we let
z
w=- '
izl
so that lwl = I. Therefore
w= ei8 = cose + i sin e

for some e' and
Z = izl W = izl eiB = reiO = r(COS fJ + i sin fJ).
The expression rew (or r(cose + i sin fJ)) is called the polar form for z, because it
describes z in terms of the polar coordinates r, e of the corresponding point.
z = 1 + 2i.
Here r = lzl = Is, and so

z 1 + 2i 1 2 .
w=-= =-+-i
-Js -J5 -Js .
---
izl
Let () be any number such that
1 . 2
cos() = .J5' sm () = .J .
S
Then
i8
w = cos() + i sin () = e
and
z = rw = r(cos () + i sin())
in polar form.
Theorem 6. For every () and IX,
w z w+z
(The same is true for all complex numbers w, z. That is, e · e = e . But we
are not yet in a position to prove it.)
Proof
i8 i
e • ea = (cos() + i sin ())(cos IX + i sin IX)
= COS() COS IX - Sin ()Sin IX + (sin 8 COS IX + COS()Sin 1X)i
= cos (() + IX) + i sin (8 + IX)
i(e+ l
e a. =
i8
In the polar form re , r is called the modulus and () is called the amplitude. It is
i8
a slight abuse of language to speak of () as the amplitude of re , because while the
modulus is determined when the number z is· named, the amplitude () is not deter
mined. In fact, when we apply the exponent i() to e , we get a periodic function/(8).
Theorem 7. For every integer n (positive, negative, or zero),
This is so because cos (8 + 2mr) = cos() and sin (() + 2mr) = si n e.
PROBLEM SET 10.12
Express the following complex numbers in polar form.

I. 1 + i 2. 3i 3. -7
v3
4. i s. v3 + i 6. -4 - 4i
2 + 2
7. In the complex domain, the sine and cosine are defined by the series
z2Hl
co
sinz = I (-1); . l)''

J=O (2} + .
co
z2;
cos z = I c-1); -. .
j=O
, (2j) ·
How do we know that these series converge for every z?

10.13 De Moivre's Theorem 489
8. In the text, we expressed ei8 in terms of sin fJ and cos 8. More generally, express ei•
in terms of sin z and cos z.
9. Show that for every z,
sin ( -z) = -sin z,

cos ( -z) = cos z.
10. Express e-i• in terms of sin z and cos z.
11. Express sin z and cos z in terms of the complex exponential function.
10.13 DE MOIVRE'S THEOREM
In Section 10.12 we found that every complex number could be expressed in polar
form, with
where
r = lzl.
And Theorem 6 said that for every e and a,
This gives us a rule for multiplying complex numbers in polar form: we multiply
the moduli and add the amplitudes. For
we have
To divide, we divide by the modulus of the divisor and subtract the amplitude.
z1/z2 = (r1/r2)ei<&1-02l (r2 ;of 0)

(This is easy to check by multiplication.) By induction based on the multiplication
rule, for z = rei8 we have
These ideas give us a method for extracting roots of any order. We shall now see
that the number 1 has three cube roots in the complex domain. If z3 = 1, then izl3 =
1, and so izl = 1. Therefore
for some 8. Therefore
and
38 = 0 + 2n7T
for some n. For three successive values of n we get:
n = 0, e = o, Z = Z1 = 1;
n = 1, e = f 7T, z = z2 = cos f7T + i sin f7T;
n = 2, e = f7T, z z cos t7T + i sin f7T.
3
= =
Using other values of n, we would get repetitions of the same cube roots. Thus the
roots are
z1 = 1,
1 J3
Z2 = -2 + li,
1 J3.
Z3 = -- - -z.
2 2
These cube roots could have been found by elementary methods, because
z3 - 1 = (z - l)(z2 + z + 1),
and the quadratic formula gives z2 and z3 as the roots of the equation z2 z + 1 0. + =
But for roots of higher order, and for numbers less simple than 1, the elementary
methods break down, and our new method still works. For example, i ei"f2• =
i
Therefore the fourth roots of are the numbers ei0 for which
1 + 4n
4() = :!!. + 2n1T' or () =
8
1T.
2
Four successive values of n give us
1T 51T - 91T - 137T

()o-
--, ()1=-, ()2 - , ()3- ,
8 8 8 8
from which the roots z; = eio; can be computed, by repeated applications of the half
angle formulas for the sine and cosine.
In general, every complex number z ¥- 0 has exactly n nth roots in the complex
domain, and the roots can be expressed in the form that we have been using.
Theorem 1 (De Moivre's theorem). Every complex number
z = reie ¥- 0
has exactly n nth roots. These are the numbers
.nr ·o
z.' = v r e'; (j = 0, 1, . . . , n - 1),
where
e; = (1/n)(() + 2j7T) (j = 0, 1, . . . , n - 1).
To prove this, we need to investigate two things.
10.13 De Moivre's Theorem 491
1) The modulus of the roots. If wn = z , then
lwln= lzl = r, and lwl= V";.

2) The amplitudes of the roots. If w = V"; eia., and w
n = z, then
ncx= () + 2j7T for some j,

and
ex = -+
()
n
-
j
n
· 27T.
Any n successive values of j give us n different values of e ia., but thereafter the values of
eia. repeat themselves.
z= 1 + 2i, n= 5.
Then
where
r = J5, e = Sin-1 (1/J5).
The fifth roots of z are the numbers
Z;= �r ei81
= M eie1= 5111oei8; (j= 0, 1, 2, 3, 4),
where
e; =tee + 2j7T)
= !!. + 2j
7T = 0, 1, 2, 3, 4).
5 5
(j
Thus
efs
Zo= 5111oei ,
rr) s
Zi = 51/1oe i(8+2 / ,
4rrl/s
z2 = 5111oei(9+ ,
and so on. Note that it is not easy to express these numbers in the form a + bi.
De Moivre's theorem shows that C not only contains roots of all orders for all
real numbers, but also roots of all orders for all complex numbers. This means, in
particular, that any quadratic equation with coefficients in C can be solved in C. The
method follows the derivation of the familiar quadratic formula.
2
az + bz + c = 0 (a � 0)
2 b c
z +-z=--
a a
2 2
2 b b c b
z + -z + - = - -+-
2 2
a 4a a 4a
b
2 2 )
b - 4ac
(z +- ·=
2a 4a
2 .
If b2 - 4ac = 0, then the equation takes the form
(z -)2 = 0
+
b
<=> z + -2a = 0,
b
2a
and
z = -b/2a
(
is the only root. If b2 - 4ac ¥- 0, then the complex number (b2 - 4ac)/4a2 has two
square roots z1, z , and
2
z + b/2a = z1'
az2 + bz + c = 0 ::::> or
Z + bj2a = Z .
2
Therefore the roots are the numbers
b b
- - + Z1 ' - - + Z 2.
2a 2a
In fact, a much more general result holds: every polynomial equation
with coefficients in C, has a root in C. We express this by saying that C is algebraically

closed. An easy proof appears eventually, in the theory of functions of a complex
variable.
PROBLEM SET 10.13
Solve the following equations, algebraically in terms of radicals.
1. z4 + 1 0. 2. z6 + i 0.
-
= =
3. z3 + 8 = 0. 4. z2 + 2z i + 1 = 0.
5. z3 + z2 + z + 1 = 0. 6. z2 + z + i + t = 0.
7. z7 + z6 + z5 + z4 + z3 + z
2
+ z + 1 = 0.
8. z5 + z
4
+ 2z3 + 2z2 + z + 1 = 0.
9. We know that for each n,-the number 1 has exactly n nth roots. Show that we can
always find one of these, say, z0, so that the complete set of nth roots are the powers
z0, z5 , zg , ... , z(; of z0•
10. List, in polar form, the fifth roots of i - 1.
11. If z0 is as in Problem 9, we say that z0 is a generator of the nth roots of 1. Can z0 be

chosen at random? That is, if z(; = 1, and z0 � 1, does it follow that z0 generates the
nth roots of 1 ? Why or why not?
12. For each n, let Zn be the set of all nth roots of 1. Show that (a) Zn is closed under
multiplication and (b) Zn contains the reciprocal of each of its elements.
*13. Let p be a prime, and let z0 be any element of Zp, other than 1. Show that the numbers
z0, zfi , ... , =�-1 are all different from 1. Then show that they are all different (from each
other). Explain how this result is related to one of the preceding problems.
10.14 The Radius of Convergence. Differentiation of Complex Power Series 493
*10.14 THE RADIUS OF CONVERGENCE.

DIFFERENTIATION OF COMPLEX POWER SERIES
In dealing with real series, we observed the following phenomena.
1) The domain of convergence of ! a;xi was always symmetric about zero, except
perhaps for the endpoints of an interval. That is, the domain of convergence always
turned out to be (a) 0 alone, (b) ( - oo, oo), or (c) an interval of one of the types
(-r, r), [-r, r], [-r, r), (-r, r].
2
2) The function f(x) = 1/(1 + x ) is defined for every x, and has derivatives
of all orders. Nevertheless, its series !�o ( - l )ix2i converges only on the interval
( - 1 1) .
, Here the series goes bad for reasons which seem unrelated to the properties
of the function which it represents.
We shall now find out why these things happen. The next two theorems are
modeled on theorems which are known in the real domain.
Theorem 1. If ! Z; is convergent, then limn�"' zn = 0.
Proof For each n, let Sn = Li'=o z;, so that limn�oo Sn = S = !:o Z;. Then
limn�"' Zn = limn�oo (Sn - Sn_1) = S - S = 0. (Note that this is exactly like the
old proof.)
Theorem 2. Every convergent sequence of complex numbers is bounded.

Proof Let zn = xn + iYn· Since z1, z2, . • • converges, it follows that the sequences
x1, x2, • • . and y1, J2, . . . also converge. Therefore they are bounded, and we have
IYnl � b,
for every n. Therefore
Throughout this section, for each r > 0, D, denotes the interior of the circle
with center at the origin and radius r, in the complex plane. Here D stands for disk:
the interior of a circle is called an open disk. If we include the boundary circle, we
get the closed disk D, = {z 1 Jzl � r}.
Theorem 3. If a series !:o a;z; converges for z = z0, with z0 � 0, and 0 < s < lz0I,
then the series converges at every point of D,.
•z
The first step in the proof is to show that L Ja;J si is convergent. For this purpose
we use a comparison test. We have
i ;
I
!ail · ;0 Zo = !ail · ;0 · lzoli
!ail s; = ·
I I I
laiz&I ; ·
o
i
= · I I
But we know that L aizi is convergent. Therefore a;zi 0, and so the numbers aizi -'>-
form a bounded sequence, with
laizgJ � b for every j.

And
lzol· Therefore
I :J <
1,
<
because s
I Ja;I si
i=O
= I la;Z6J · I.!._
i=O
b I i
�b
_
z0 i=O z0
1
I I.!._
ls/z01 <
I i
= _ oo.
Now for each point z of Ds, we have Jzl � s. Therefore
L la z il � zla;I s1 <
"' "'
i=O ;
(z in Ds)·
i=O
00
Therefore .2; a1zi converges in D8, which was to be proved.

Suppose now that we have given a series L a;zi which converges for some z ;if 0.
It follows that the series converges on some disk D8• Let S be the set of all such
numbers s. That is,
S {
= s J _I a1z1
J=O
converges on I>s}·
If S is unbounded, then the series is convergent for every z. In this case, we say that
the radius of convergence is oo. If S is bounded, then S has a least upper bound
sup S. (See page 243.) Let
r =sup S.
In this case, we say that the radius of convergence is r.
Theorem 4. If a seriesL�o aizi has radius of convergence r

<
oo, then the series
converges for lzl < r and diverges for lzJ > r.
Proof
1) Given a number z0, with Jz0J < r. Since r is the least upper bound of the set S,
the number Jz01 is not an upper bound of S. Therefore there is an s > lz0J such that
the series converges on D8• Therefore the series converges at z z0• =
2) Given z0, with lz0J > r. Suppose that the series converges at z = z0, and let s be
such that
< s lz0J.
<
r
Then the series converges on Ds; and this is impossible, because r is an upper bound
for such numbers s.
Note that while this theorem tells us what happens inside the circle lzl = r, and
what happens outside the circle, it tells us nothing about what happens on the circle.
L a,xi. Suppose that the series
This theorem clarifies the situation for real series
converges for some x � 0. Then the complex series I,�0 aizi converges for some
z � 0 (namely, the same x). Let the radius of convergence be r. Then L a1zi con
verges for lzl < r and diverges for lzl > r. Therefore, for real values of z, L a1xi
converges for Jxl < r and diverges for lxl > r.
The circular domain of convergence for complex power series also accounts for
the behavior of the series
1
I< -1)ix2i = ___2 .
i=O 1+ X
If this series converged for some x for which lxl > 1, then for the complex series
1
�
""'(-l)'z'
. 2'
= --- 2,
i=O 1 + Z
we would have a radius of convergence r > 1. This is impossible, because the function
itself blows up at a point of the unit circle: Iii 1, and for z = i the denominator
=
on the right-hand side becomes 0. [Query: how do we know that the series is equal to
2
1/(1 + z ), for complex values of z?]
The derivative of a complex-valued function is defined by an obvious analogy
with the derivative of a real-valued function. That is,
f(z) - f(zo)
f'(zo) = lim .
z->zo Z - z0
This is a complicated limiting process, because z may approach z0 from any direction
in the complex plane.
G
8
To be exact, the indicated limit means that for every E > 0 there is a o > 0 such that
0 < lz - zol < o => l f(z)

Z
-
f(z o)
- Zo
- f'(zo) < I E.
Here the inequality lz - z01 < o allows z to lie anywhere in the interior of a circle
with center at z0 and radius o. lfj'(z0) exists, then we say that/ is differentiable at z0•
If/ is differentiable at every point of an open disk containing z0, then we say that/ is
analytic at z0• It is easy to see that if f(z) = zn, then f is differentiable everywhere,
and therefore analytic everywhere, with
j'(z) = nzn-1 •
n
The proof is exactly like the proof for f (x) = x . Similarly:
Theorem 5. If f is a polynomial, with
n
f(z) = 2;a1z1,
1=0
then/ is analytic everywhere, and
n
f'(z) = 2;ja1z1-1.
j=l
But the proof of the following theorem is hard.
Theorem 6. If f(z) has a power series 2;;:0 a;z1, converging in the open disk D.,
then f is analytic in D,., and
00
f'(z) = 2;ja1z1-1 (lzl < r).

j=l
Proof Take a fixed z0 in Dr. To computej'(z0), we need some preliminary results.

Let s be any real number such that
lzol < s < r
Then
00
L la11 s1 < oo (0 < s < r).

1=0
(See the proof of Theorem 4, where we used the same apparatus.)
Therefore the series defines a function g(s), and by Theorem A of Section 10.5
it follows that
00
g'(s) Lj la 1 sH,
1
j=l
=
where the series on the right is convergent. Therefore
00
Lj la 1 s1 < oo,
j=l 1
and so
00
lim L j Ja 1 s1 0.
n-+oo :i=n+l 1
=
. [--- ( La z ;)]
Now consider what we are trying to prove. The theorem says that
1 00 j 00 00
LJa;z0J-1,
•
hm ; - La z0
z - z0 J=O J=O 1 J=l
· =
z--+zo
for every z0 in Dr- Simplifying the expression in brackets, and�transposing the sum on
the right, we get the equivalent form
00
lim La;(z1-1 + z1-2z0 + z1-3z� + · · · + zzg-2 + zg-1 - jzg-1) 0.

z-+zo i=l
=
This is not as bad as it looks, because the limit of thejth term is 0 for eachj: each
of the firstj terms in the parenthesis approaches the limit zb-1, as z --+ z0, and so the
total expression in the parentheses approaches 0. Therefore if the sum were finite
our conclusion would follow: if
n
Sn(z) Lak )
i=O
= · · ,
then
lim Sn(z) = 0 for each n.
z-+zo
We need to find out what happens to the remainder
00
Rn(z) = L alzi-l + z1-2z0 + + zg-1 - jzg-1). · · ·
i=n+l
We know that lzol < s; and since we are discussing limh•o' we may assume that
lzl < s. Under these conditions,
la;(z1-1 + z1-2Zo + ... + zg-1 - jzg-1)1

� la;J . (Jzi-11 + Jz1-2zol + .. . + lzg-11 +I Jzg-11) < Ja;J . 2j s1-1.
Therefore, for lzl < s we have
00
IRn(z)J � L Ja J 2js1-1.
i=n+l 1
·
The sum on the right approaches 0 as n --+ oo.

Now let E > 0 be given. There is an n such that
(lzl < s).

There is a c5 > 0 such that
0 < lz - Zo l <0 => ISn(z)I < E/2.

Since
00
L ak ) Sn(z) + Rn(z),
1=1
· · =
we have
0 < lz - Zol < c5 => I Iai(· ··)I

�l
� ISn(z)I + IRn(z)I < � + � = E,
2 2
and so limZ-+•
o
L ak - ) · = 0, which was to be proved.
It is worthwhile to remember the device (used in this proof) of taking n large

enough to make IRn(z)I small, and then taking lz - z01 small enough to make
ISn(z)I small. This is not a special or isolated device, but a sample of a standard
method.
Once we know how to differentiate a series, we can apply in the complex domain a
variety of techniques which we have been using in the real domain. The generalizations
are not hard, as you will see. In the following problem set, D is used to indicate
differentiation. Thus
Df(z) = f'(z),
and
00 00
D L a1z1 = L ja1z1-1,
1=1 1=1
on any open disk Dr where the series on the left converges.
PROBLEM SET 10.14
1. Find De'. 2. Find De-•. 3. Find D sin z.
4. Find D cos z. 5. Find De•2• 6. Find D sin 2z.

7. Show that if/is analytic in Dri andg =Jn, theng is analytic in Dr, and
g'
n
=
Dt =
nr-1f'.
(The pattern of proof for the real domain works equally well here.)
8. Let f be analytic for every z , and let g(z) f(a + z). 3how that g'(z)
= = f' (a + z).
(Evidently this is another simple special case of the chain rule.)
9. Givenf(z) = L:o a1z1 in Dr. Show that/has not only a first derivative in D,., but also
derivativesf<2> "
f , f<3>, . . . . of all orders.
L:o a1z1
=
10. Given f(z) = = 0 for every z in Dr· Obviously a0 = 0, because f(O) = a0.
Show that a1 =0 for every j.
11. Given f(z) = L: o a1z1 in Dr.Show that if f'(z) = 0 for every z, then f is a constant
function and is equal to a0 for each z.
12. Given thatf(z) = L a1zi, g(z) = L b1zi in D r. Show that if f' = g', then/ - :; is a
constant.
10.15 Integration and Differentiation of Real Power Series 499
13. Let ¢(z) = cos2 z + sin2 z. Show that ¢(z) = 1 for every z.
14. Given that cos2 z + sin2 z = 1 for every z, does it follow that the complex functions
cos z and sin z are bounded? Why or why not?
15. Show that ea · e• = ea+z, for every a and z.

16. Does the addition formula for the cosine hold in the complex domain? That is, is it
true that
cos (a + z) = cos a cos z - sin a sin z,
for every a and z?
17. Answer the same question, for the proposed identity
sin (a + z) = sin a cos z + cos a sin z.

18. Show that sin 2z = 2 sin z cos z.
19. Show that cos 2z = cos2 z - sin2 z.
*10.15 INTEGRATION AND DIFFERENTIATION OF REAL POWER SERIES
In Section 10.5 we stated (in Theorems A and B) that a Maclaurin series could be
differentiated and integrated a term at a time, on any interval ( -r, r) on which the
series converges. We shall now prove these theorems. The ideas that are needed to
do this may be easier to understand if we first show how they apply to the geometric
series
00 .
1
f(x) Lx 1 < x < 1).
i=O '
=
--
= ( -
1 - X
We want to show that

rk 00 rk
J/(x) dx i�Jo xi dx (0 < k <
= 1). (1)
Let
n
Sn(x) Lxi, =
i=O
00
Rn(x) L x\
i=n+l
=
so that
f(x) Sn(x) + Rn(x), =
f(x) - Sn(x) Rn(x), =
and
f!(x) dx - fSn(x) dx fRn(x) dx. = (2)
Since S,,(x) is a finite sum, we know that
and
n ( " oo "
xi dx � J xi dx,
(
J
!�rr;, i� o i o
=
by definition of the infinite sum. We shall show that
lim (kR (x) dx = 0.

n-+ Jo n
(3)
ex:>
If (3) holds, then we can take the limit in formula (2), getting
[
11-+oo Jo Jo n-+co J
]
lim rf(x) dx - rksn(x) dx = lim rkf(x) dx [ o
- i
i=o J
(\i dx
o
]
lkxi dx = 0.
00
= kf(x) dx
This means that (1) holds.
l0
- !
i=O 0
The proof of (3) is as follows. We have
Rn(x) = L xi= xn+1 + xn+2 + ..

00
i=n+l
.
= xn+l ! xi =
xn+l 00
i=O
--
.
1- X
We are integrating from 0 to k. If 0 � x � k, then

1 - x � 1 - k,
and
Therefore
Therefore
r n+l kn+2
kRn(x) dx � k k
Therefore
lo J ol-k
dx -- = -- .
1-k
lim (k R (x) dx = 0,
n-+oo Jo n
What made this work was the fact that the functions Rv R2, • • • were squeezed
to 0 by a sequence of constants. We had
Mn =--
kn+l
1 k' -
0 � Rn(x) �Mn (0 � x � k),

and
IimMn=O.
n-->OO
It followed ihat
j.kRn(x) dx
0 k dx
� rM
Jo n
= kMn--+ 0.
This method applies in general, to prove Theorem B of Section 10.5. To do this, we

need some preliminaries.
Theorem 1. Every convergent sequence is bounded.
This is Theorem 7 of Section 10.1. It gives us the following result for series.
Theorem 2. If L a,xi is convergent on the interval (-r, r), and 0 < k < r, then
L a;ki is absolutely convergent.
Proof Let x1 be any number between k and r.
Then L a;x� is convergent, and so limi--.oo aix1 = 0. By Theorem 1, the sequence

a0, a1x1, a2x�, . .. is bounded. Let b be a bound for this sequence, so that
.
laixil � b for every i.
Now consider the series L aiki. Lets= k/lx11. Then 0 <s < 1, and so
.
00
1
L:s' = -- <co.
i�O 1 - S
But k = lx11 s, and so
Therefore
00 . .
00 . • 00
b
L la;k'I = L laixtl s' � L bs' = -- < co.
� � � 1-s
Therefore L aiki is absolutely convergent, which was to be proved.
Carrying the same ideas a little further, we get better information.
Theorem 3. If L aixi is convergent on the interval (-r, r), and 0 <k < r, then the
remainders
00
R n(x) = L a;xi
i�n+l
are squeezed to 0, on the interval [-k, k], by a sequence of constants. That is, there
is a sequence M1, M2, • • • of constants, such that
limMn = 0,
n-> oo
and
(-k � x � k).
Proof We know by the preceding theorem that L aiki is absolutely convergent.
For each n, let
L laikil.
00
Mn=
i=n+l
Then
limM n=O,
n-> oo
because
oo n
i
Mn= L lai kil - L la;k j.
i=O i=O
On the interval [-k, k], we have lxl � k . Therefore
and
Therefore
IRn(x)I � Mn for every n,
and so the remainders Rn(x) are squeezed to 0, on the interval [-k, k], by the con
stants M1, M2, • • • •
The ideas in this theorem are going to come up again, and so we need a briefer
language in which to describe them.
Definition. Let Ri. R2, • • . be a sequence of functions on the interval [a, b]. If there
is a sequence M1, M2, • • • of positive constants, approaching 0, such that
I n(x)I � Mn,
R
for every x on [a, b] and for every n, then we say that the functions R1, R2, • • •
approach 0 uniformly on [a, b], and we write
VlimR n(x) = 0 on [a, b].
Definition. Let Si. S2, ... , and S be functions on [a, b]. If
Ulim [S(x) - Sn(x)] = 0 on [a, b],

n-+ co
then we say that the functions S1, S2, • • • approach S uniformly on [a, b] , and we write
UJim Sn(x) = S(x) on [a, b ].

n-+co
In the case covered by Theorem 3, we had

[a, b] = [-k, k],
co
Rn(x) = L aixi,
i=n+l
n
Sn(x) = L a;x\
i=O
co
S(x) = L a;x;,
i=O
. Rn(x) = S(x) - Sn(x).
In our new terminology, Theorem 3 takes the following form:
Theorem 3'. If L a;x; is convergent on (-r, r), and 0 < k < r, then
n co
UJim L a;xi = L a;xi on [ - k, k].

n-+ co i=O i=O
At the beginning of this section, we used these ideas to integrate z xi a term at a

time. We want to use the same idea for all power series, but we have a new problem.
The series z xi represented a known function f(x) 1/(1 - x), which was con
=
tinuous. But whenf(x) is given only by a series z a xi , we first need to show that/is
i
continuous, in order to conclude that the sum has an integral. Thus we need the
following two theorems.
Theorem 4. Iffn is continuous for each n, and U limn-oofn(x) = f(x) on [a, b], then
f is continuous on [a, b].
Proof Take a fixed x0, and let E be any positive number. Let M1, M2, be as in • • •
the definition of U lim. Then there is an n such that Mn < E/3. Hereafter in the
proof, n is fixed. Sincefn is continuous at x0, there is a o > 0 such that
Ix - Xol l fn(x; - fn(Xo)I < E/3.

Since
{
l f(x) - f n(x)I < Mn for every x,
we have
If(x) - f n(x)I < E/3,
Ix - Xol < O => /fn(x) - fn(Xo)I < E/3,
/fn(Xo) - f(xo)I < E/3.
By the triangular inequality,
a
/a + b + cl � l / + /bl + /c/.
Therefore
/x - x01 < O :::::> If(x) - /(x0)1 < E/3 + E/3 + E/3 = E,
To conclude that f is continuous, it is not enough to know that limn-co f,,(x) =
f(x) for each x of [a, b]; we really need to know that U limn_00/(x) f(x) on [a, b].=
(See the following problem set, for an example showing this.) For power series, of
course, we know that Sn(x) is continuous for each n, because Sn(x) is a polynomial;
and we know that the sum of the series is not merely lim S,(, x) but also U lim S,,(x),
on every closed interval lying in the interval of convergence. This gives the following
theorem.
Theorem 5. If z a xi is convergent on (-r, r), then z a xi is a continuous function

i i
on_ (-r, r).
Therefore there is such a thing as the integral of 2 a;xi ; and we can show that it is
the sum of the integrals.
Theorem 6. If z�o a;xi converges on (-r, r), and lxl < r, then
{"'
Jo
[f al]
i=O
dt =I l"'0 ai dt.
i=O
Since f� ti dt = xi+I/(i + I), the theorem tells us that
(" [ii=Oal] dt = i=OI�

J0 + 1
xi+i.
l
To prove the theorem, we let Sn(t) = Lf=o aiti. Then for 0 < x < r we have
00
U lim Sn(t) =_Lai on [O, x].

n--+oo i=O
(For -r < x < 0, the same condition holds on [x, O], and the rest of the proof is
exactly the same.) Let
00 00
Rn(t) = L a/ = L aiti - Sn(t),

i=n+l i=O
and let M1, M2, • • . be as in the definition of U lim. Then
IRn(t)I = Ii�a;ti - Sn(t)I � Mn for every n.
Therefore
� l "Mn dt
=Mn Ix!.
Since limn�oo Mn lxl = 0, it follows that
l0x [Z=O� a/] dt = l"0 Sn(t) dt.

lim
n--+ oo
But Sn(t) is a finite sum, and can be integrated a term at a time. Therefore
l0x [i=O_La/J dt = l"'

and
oo n oo
lim
n�oo
_L a/dt = _L
i=O 0 i=O l"a/dt,
0

This theorem shows that if _L a;X; converges on (-r, r), then the integral series
_L [a;/(i + l)]xH1 also converges on ( -r, r) . The same is true for the derivative series
'co i
ki=l lQiX -1
·
•
Theorem 7. If _L;:0 aix i converges on (-r, r), then _L;:1 iaixi-i converges on (-r, r).
Proof This is going to be very similar to the proof of Theorem 2. Let x1 be any
number such that lxJ < x1 < r. Then _L a;xi is convergent; limn�oo aixi = O; and
there is a bound b such that laixil � b for every i. Lets = lxl/x1, so that
Jxl = X1S,
i
liaix -ll = liaixi-1si-ll = laixi-11 · isi-l � bii-1•
Therefore
C()
C()
i i
2 I iaix -ll � b 2 is -l.
i�l i=l
But s < 1, and so the series on the right-hand side converges, by the ratio test.
Therefore the series on the left-hand side converges. Therefore 2 iaix i-l converges.
It remains to show that the "derivative series" 2 ia;xi-l really gives the derivative,
but this is easy.
Theorem 8. If
C()
2 ai xi = f(x) on ( -r, r),

i=O
thenf is differentiable on (-r, r), and
C()
i
f'(x) = 2 iaix -1•
i=l
Proof Let
C()
g(x) = 2 ia;xi-l (-r < x < r).

i=l
Then
1"'g(t) dt 1"' [200

0
=
0 i=l
ia/-1 dt
J = 2
00 1"'
i=l 0
ia/-1 dt =
00
2 aix;
i=l
= f(x) - a0•
Therefore
i"'g(t) dt = f(x) - a0•
But the integral on the left-hand side is a differentiable function, and
Di"'g(t) dt = g(x).
Therefore fis differentiable, and f' = g, which was to be proved.
Obviously Theorem 8 can be applied again, to the derivative series, and so
f"(x) =
00
2 i(i - l)xi-2,
i=2
/<3\x) =
00
2
i =3
i(i - l)(i - 2)xi-a,
and so on. Thus ifj(x) is represented by a power series, thenfhas an nth derivative
for every n. In a way this is good; it means that functions given by series are in some
respects manageable. But it also means that if a function f does not have infinitely
many derivatives, then f cannot be represented by a power series. Later you will see
that many such "irregular" functions can be represented by series of other kinds,
notably by so-called Fourier series, of the form
00
a0 + L [ai cos ix + bi sin ix].

i=;ol
PROBLEM SET 10.15
1. Let
oo x 2i
J<x) .I <-1)i ' .
(2l )'
=
•=0 •
Calculate the series for j"(x), and verify that j"(x) = - f(x) . [This must be true,
because f(x) cos x.]
=
2. Find a simple formula for g" (x), given

°"
x2H1
(-l)i
g(x) =
�
i (2i + 1)! ·
3. Do the same for

oo
xi
h(x) L (-l)i-:-.
z!
==
i=O
n
4. For each n, let fn(x) x (0 � x � 1). Let /(x)
= lifin__.00f (x). Sketch the graph
=
off, and find out whether it is true that U liIDn ...... oof n(x) f(x) on [O, 1]. (This throws
=
some light on Theorem 4.)

5. Find, by any method, a power series for /(x) = xeX + e"'.
6. Same question, for g(x) = x cos x + sin x.
7. Same question, for h(x) = x sin x - cos x.
8. Find a series for a positive function f such that f'(x) 2xf(x) for every x, and f(O)
= = 1.
(There is a short-cut. Since f (x) > 0 for every x, f'(x) 2 xf (x) <=> f'(x)/f(x) = = 2x.
Now what?)
In each of the following cases, discuss limn__.oof n(x) and U limn__.00/ n(x).
9. fn(x) = nxn, - ! � X � t.
10. fn(x) = (1/n) sin nx, -1 � x � 1.
11. fn(x) = (1/n) cos 2nx, -1 � x � 1.
n
12. fn(X) = (1/n)e "', 0 � X � 1.
13. fn(X) = (1/n2)en"', 0 � x � 1.
14. fn(x) = n sin (x/n)2, 0 � x � 1.
*15. Find out whether the following is true:
Theorem (?). If U limn__.00/,:,(x) f(x) on [ -k, k], and each of the functions fn is
=
differentiable, then f is differentiable, and f'(x) liIDn...... cx,/�(x) for every x between
=
-k and k.
(If this is true, then it furnishes a straightforward proof of Theorem 8, replacing
the proof using integrals.)
**16. Here we return to complex power series, as in Section 10.14. It is evident that if
L aizi converges on Dr, then L I aizil converges on Dr; in fact, every time we have proved
convergence for a complex power series, we have first proved absolute convergence and
then used Theorem 3 of Section 10.12. It remains, however, to consider the question of
uniform convergence. Just as for sequences of real functions,
U lim/n(z) = f(z) on D8
n-co
if 1/n(z) - f(z)I is squeezed to 0 by a sequence of constants. That is, the above U Jim
relation holds if there is a sequence M1, M2, of positive constants such that
• • •
limMn = 0, and 1/n(z) - /(z)I � Mn, for every z in D5•

n-co
Prove the following:
Theorem. If .2 a1zi has r > 0 as its radius of convergence, and O < s < r, then
n co
U Jim .2 a1zi .2 a1zi on D5•
1=0
=
i=O
n-oo
11 Vector Spaces and Inner Products
11.1 CARTESIAN COORDINATE SYSTEMS IN THREE-DIMENSIONAL SPACE
To set up a coordinate system in three-dimensional space, we use the same scheme that
we used in a plane; the only difference is that we use three mutually perpendicular
lines instead of two. These are the x-, y-, and z-axes. On each of the axes we take a
coordinate system, in such a way that the origin 0 has coordinate 0. The plane con
taining the x- and y-axes is called the xy-plane. Similarly for the yz- and xz-planes.
/ -----
/
/ -:-'II
/ /
/ / I
f--- // I
I - -- P1 / I
I --- /
- I
I I I
I I
: ------ !
I
I
/7---- y
/
--
- -- I //
I
---....v/
/
x
These are called the coordinate planes. In the figure we have indicated the position of
the pointP1 by drawing the rectangular parallelepiped which has the origin 0 and the
point P1 as opposite corners, and sides parallel to the coordinate planes. We get
the coordinates of a pointPl> as before, by dropping perpendiculars to the coordinate
axes.
Here M1, M2, and M3 are the feet of the perpendiculars fromP1 to the three axes.
If these points have coordinates x1, Yi· Zi, on the respective axes, then Pi is matched
with the triplet (xi, Yi, zi), and we write
Pi +-+ (xi, Yi, Zi) .

As in the plane, we often fail to distinguish between a point and the number triplet
with which it is matched; and so we may write
Pi = (xi, Yi' z1) .
508
11.1 Cartesian Coordinate Systems in Three-Dimensional Space 509
In figures, we may label a point as P1(x1, y1, z1), to indicate that the given point has
the given coordinates.
The coordinate planes divide space into eight parts, called octants. The figure
above shows the fi rst octant, consisting of all points of space for which all three co
ordinates are � 0.
By two applications of the Pythagorean theorem, we see that each diagonal of a
rectangular parallelepiped has length Ja 2 2 2
+ b + c • This means that for each point
P1(x1, y1, z1) we have
0 P� = x� + y� + z�.
More generally, for any two points P1, P2, we have the distance formula given in the
following theorem.
b
i'\
a I \
I '-
'
'
'e
c '
J.-----'\--- ---·
/ ............. \
/ -....... \
/a d ', '
/ ......
...... _,
Theorem 1. If P1 � (x1, Yi. z1) and P2 � (x2, y2, z2), then
2 2 2
P1P2= J (x2 - X1) + (Y2 - Y1) + (z2 - z1) .
Proof Suppose first that x1 -:;!: x2, Yi -:;!: y2, and z1 -:;!: z2. Then P 1 and P2 are opposite
corners of a rectangular parallelepiped. In the figure on the left below,
a= lx2 - Xii, b= IY2 - Y1I, c = lz2 - z1I·

Therefore
510 Vector Spaces and Inner Products 11.1
If some of the inequalities x1 -:;!= x , y1 -:;!= y , and z1 -:;!= z do not hold, then our
2 2 2
parallelepiped reduces to a rectangle, a segment, or a point, and the same distance
formula holds for simpler reasons.
We shall use this result to describe planes by equations. (See the figure on the
right below.) Given a plane E, suppose first that E does not pass through the origin,
b
al
I
I P2
I
c I
- 1 ---- --- Y
/
�! __ ---
/
/
x x
Po <--> (a, b, c), P1 .--.. (2a,2b,2c)
and let P0 = (a, b, c) be the foot of the perpendicular from the origin to E. Let P1 be
the point (2a, 2b, 2c). Then OP0 = P0P1; P0 is the midpoint of the segment from 0
to P1; and Eis the perpendicular bisecting plane of the segment . Therefore Eis the set
of all points of space that are equidistant from 0 and P1. That is, Eis the graph of the
condition
<::::> x2+ y2+ z2 = (x - 2a)2+ (y - 2b)2+ (z - 2c)2
<::::> 0 = -4ax+ 4a2 - 4by+ 4b2 - 4cz+ 4c2
<=>ax+ by+ cz - (a2+ b2+ c2) = 0.
This has the form
Ax+ By+ Cz + D = 0, (A, B, C) -:;!= (0, 0, 0).
The condition on the right says that the numbers A, B, and Care not all equal to zero;
and this is correct, because the point P0 = (a, b, c) is not the origin. An equation of
the above type is called a linear equation in x, y, and z. Thus we have shown that every
plane that does not pass through the origin is the graph of a linear equation in x, y,
and z. For planes through the origin, the same result holds. Let P0(a, b, c) be any
point of the line perpendicular to Ethrough 0, other than 0 itself. Let P1 be the point
(-a, -b, -c). Then E is the perpendicular bisecting plane of the segment from
11.1 Cartesian Coordinate Systems in Three-Dimensional Space 511
Po(a,b,c), P1 (-a,-b,-c)
P1 to P0; and so E is the graph of the equation
P1P = P0P
��+�+�+W+�+�=�-�+�-W+�-�
� 2ax+a2+2by+b2+2cz+c2 = -2ax+a2- 2by+b2 - 2cz+c2
�ax+by+cz = 0.
This has the same form Ax+By+Cz+ D = 0, with D = 0, as it must be: the
origin lies in the plane E. In general:
Theorem 2. Every plane is the graph of a linear equation in x, y, and z.

This theorem can be applied in a variety of ways. For example, let Ebe the plane
through the points (3, 3, 4), (2, 4, 4), and (2, 3, 6). Now Ehas an equation of the form
Ax+By+Cz+ D = 0,
and the equation must be satisfied by the coordinates of the three given points.
Therefore the coefficients A, B, C, and D must satisfy the equations
3A+ 3B+4C+ D = 0, (1)

2A+4B+4C+ D = 0, (2)
2A+3B+ 6C+ D = 0. (3)
Subtraoting (2) from (1) we get
A -B = 0. (1)- (2)
Setting B =A in (2) and (3), we get
6A+4C+ D = 0, (2')
5A+ 6C+ D = 0. (3')
Subtracting (3') from (2') we get
A - 2C = 0, (2') - (3')
and so C = A/2. Therefore D = -8A. Now, to avoid fractions, we set A= 2.

This gives
A= B = 2, c = 1, D = - 16,
2x + 2y+z - 16 = 0.
This checks.
Note that any number different from 0 could have been used as A. There are some
cases, however, when this is not so. For example, the graph of the equation
y+z=l
is a plane, parallel to the x-axis. This plane is the graph of infinitely many different
equations, of the form
ky+ kz - k = 0 (k -:;t: O);
but x does not appear (with nonzero coefficient) in any of these equations.
PROBLEM SET 11.1
1. Find the equation of the plane containing all points equidistant from the origin and the
point P0 = (2,6, 4).
2. Find the equation of the plane containing all points equidistant from P0 = (1, 0,0) and
Pi = (0,2,3).
3. Find the equation of the plane containing all points equidistant from the planes
x +y + z = 2 and x +y + z = -1.
4. Find the equation of the plane through the points P0 = (1,0, 1), Pi = (1,1, 1), and
P2 = ( -1,2, 0).
5. Find the equation of the plane through the points P0 = (2,1,1),Pi = ( -1, -1,0), and
P2 = (0,0,3).
6. Find the equation of the plane through
P0 = (1,3,0) and P1 = (2, 2, 3),

which is parallel to the z-axis.
7. The point (1, -1, 2) is the foot of the perpendicular from the origin to a plane E.
Find the equation for this plane.
8. Find a plane throughP0 = (1,0,2),Pi = (2,2,1),andP2 = (3,4,0). Is there more

than one such plane? Why?
9. Let A = (1,0,0), let B = (4,0,0), and let K = {PI 2AP = BP}. Find an equation
whose graph is K. What sort of figure is this?
11.2 DIRECTION COSINES. THE DIRECTED NORMAL FORM
Linear equations for planes are more useful if we know the geometric significance of
the coefficients in the equations. For this purpose, we need the idea of directed
distance on a line. Given a line L with a coordinate system, and two points P and Q
of L, with coordinates x1 and x2, we define the directed distance from P to Q as
PQ = X2 - X1.
11.2 Direction Cosines. The Directed Normal Form 513
Since x1 - x2 = -(x2 - x1), we have
QP= -PQ for every P and Q.
{
PQ+ QR= PR for every P, Q, R.
And since the distance PQ is equal to lx2 - x1I, it follows that
PQ if Q is in the positive
direction from P,
PQ=
-PQ if Q is in the negative
direction from P.
This means that directed distances are determined if the positive direction on the line
L is known; we do not care where the origin is in the coordinate system.
P <--> (a,b,c),.o(0,0,0).
Consider now a directed line L, through the origin. Let P= (a, b, c) be a point
on the positive end of L, with OP= 1. Consider the angle between the positive end
of the x-axis and the positive end of L. In its own plane, this angle looks like this:
Note that the foot of the perpendicular really does have coordinate a on the x-axis;
in fact, this is the definition of the x-coordinate of P. If the angle has measure e<, then
cos CJ.. =a.

In the preceding three-dimensional figure, fJ and y are defined similarly. They are
called the direction angles of the directed line L. (This is an abbreviation: what we
really mean is that they are the measures of the angles between the positive end of L
and the positive ends of the axes.) The numbers
cos ex, cos {J, cosy
are called the direction cosines of L. Note that they determine not merely a line
through the origin but also a positive direction on the line. If the direction on the line
is reversed, then
ex-+ 7T - ex, {J-+ 7T - {J, y-+7T - y,
and each of the direction cosines changes in sign.
Suppose now that P is any point (xi, Yi, zi) of L, other than the origin. Let p be
the directed distance from 0 to P, relative to the given positive direction on L. If
p > 0, as above, then
XifP = cos ex,
by definition of the cosine. If p < 0, as on the facing page, then
COS (7T - ex) = Xif-p,

by definition of the cosine. Therefore
- COS ex = - Xifp,
and xifP = cos ex, as before. In the same way, we get
Xi = p cos ex, . Yi = p cos {J,

Zi = p COSy.
'
We now return to our equations for planes. First we choose a direction on the normal
line N, in such a way that the directed distance Of' = p is � 0. Then the plane is the
'
graph of the equation
ax + by +cz - (a2 + b2 + c2)·= 0.

.
Now p = .Ja2 + b2 + c2. The equation therefore has the form
a b c
-x + - y + - z - p = 0,
p p p
or
x cos ex + y cos {J + z cosy - p = 0.
If the plane passes through the origin, then p 0, but the same result holds: the =
equation ax + by + cz = 0 takes the form x cos

+ y cos {J + z cosy - 0 0. oc =
Here we have chosen the direction so as to make p � 0. If we choose the opposite

direction on the normal, then
oc - 1T - oc, fJ - 1T - {J, y - 1T - y,
cos oc - -cos oc, cos fJ - -cos {J, cosy - -cosy,
and p ___.. -p. This changes all the signs in the equation
x cos oc + y cos fJ + z cosy - p = 0.
Therefore the equation still holds; and we have the following:
Theorem 3. Let Ebe a plane, and letN be a directed normal to Ethrough the origin,
with direction angles oc, {J, and y. Then Eis the graph of the equation
x cos oc + y cos fJ + z cosy - p = 0,

where pis the directed distance from the origin to E, relative to the given direction
onN.
So far, we seem to have been talking about numbers which are hard to compute.
But it is easy to bring the discussion down to earth. Consider the following equation:
x + 2y + 3z + 4 = 0.
It is not reasonable to expect that an equation taken at random will be in the directed
normal form; after all, if a plane Eis the graph of the equation
Ax + By + Cz + D = 0,
then E is also the graph of the equation
kAx + kBy + kCz + kD = 0,

for every k :;i6 0. In fact, Eq. (1) cannot be in the directed normal form, because of the
following theorem:
Theorem 4. If oc, {J, y are the direction angles of a directed line L through the origin,
then
cos2 oc + cos2 fJ + cos2y = 1.
Proof. LetP be the point of L for which the directed distance OP is 1. IfP= (a, b, c)
then
a=COS IX, b=cos (3, c =cosy.
Since
OP=OP=1 =.Ja2 + b2 + c2,
we have a2 + b2 + c
2
= 1; and the theorem follows. We also have a converse:
Theorem 5. If a2 + b2 + =1, then there is a directed line whose direction cosines
c2
are a, b, and c.
Proof. Let L be the line from the origin through the point P=(a, b, c) , directed
positively from 0 to P. This does it.
This suggests that the equation
x + 2y + 3z + 4=0 (1)
has the form
xk cos + yk cos f3 + zk cosy - pk=0,
IX
for some IX, {J, y, p, and k. If so, we must have

k COS IX = 1, k cos {J=2, k cosy=3,
k2(cos2 IX + cos2 f3 + cos2y)=12 + 22 + 32=14,
and
Taking k=,J14, we get
1 2 3 4
x + + + = (2)
.)14 .}14 y .)14 z .}14 o.
Here
cos IX= I/./14, cos f3 = 2/./14,
cosy=3/./14, p= -4/./14.
y
We have sketched the graph by plotting the intercepts, on the axes, and then completed
a triangle just as we did in the case where the intercepts were in the first octant. The
graph of (2) is the plane whose normal has the given direction cosines 1/.J14, 2/.J14,
3/.J14, and which lies at a directed distance p = -4/.J14 from the origin.
Taking k = -.J 1 4 , we get
1 2 3 4
x - - (3)
- .J14 y
=
- .J14 .J14 z .J14 O,
which gives the opposite direction on the normal;
1
cos a'= - ==-cos a, cos /3'=
-
.J14
'
3 4
cosy'= - = -cosy, p = + - - -p.
.J14 .J14 -
The same scheme works for any linear equation in x, y, z. This gives a converse
of Theorem 2:
Theorem 6. The graph of every linear equation in x, y, and z is a plane.
Proof Given
Ax + By + Cz + D = 0, (1)
with A, B, and C not all= 0. Then A2 + B2 + C2 > 0. Let
k = .J A2 + B2 + C2,
and
A B c D
a= ' b= -, c=
' p=
k k k k
Then Eq. (1) is equivalent to
ax + by + cz - p = 0. (2)
The graph of Eq. (2) is a plane E: the direction cosines of a directed normal to E
are a, b, and c; and the directed distance from the origin to Eis p.
PROBLEM SET 11 2
.
l. Given x + y + z -1 = 0. Write the two directed normal forms for the same plane,
and sketch.
2. Do the same, for x + 2z - 3 = 0.

3. Do the same, for x + 2y + 2z - 3 = 0.
4. Do the same, for x + 2y + 4z - 4 = 0.

5. Do the same, for x + 2y + 4z + 4 = 0. (This is going to look awkward; the foot of
the normal does not lie in the first octant.)
6. The normal to E from the origin contains the point (2, 4, 6). The plane E contains the
point (1, 1, 1). Find the two directed normal forms of the equation of E. How far is E
from the origin?
7. Let K be the set of all points which are equidistant from A = (1, 0, 0), B = (0, l, 0),
and C = (0, 0, 1). That is,
K={PIAP =BP= CP}.
Prove that K is a line through the origin, and find a set of direction cosines for K (more
precisely, a set of direction cosines for a direction on K).
8. Let A, B, and C be three points which are all different, but collinear. Let
K={PIAP =BP= CP}.
Show that K is the empty set { }.

9. Let A, B, and C be any points, not necessarily different. Let K be as in Problem 6.
Show that K is (a) a line, (b) a plane, (c) all of space, or (d) the empty set { }. Give
examples to show that all of these possibilities can actually arise.
10. A plane E contains the points (a, 0, 0), (0, b, 0), (0, 0, c), a, b, c � 0. Find the two
directed normal forms of the equation, and sketch, showing the case a, b, c > 0.
11. The normal to E from the origin lies in the xy-plane; and E contains the points (1, 1, 1)
and (-1, 2, 1). Discuss as in Problem 10.
12. The normal to E lies in the yz-plane; and E contains the points (2, 2, 1) and (1, 1, 2).
Discuss as in Problem 10.
13. Let E be the plane z = -1, and let K be the set of all points wh i ch are equidistant
from E and the point A = (1, 0, 0). What sort of figure is this? Sketch.
14. Let E be the plane z = -2 and let K be the set of all points which are equidistant from
E and the x-axis. What sort of figure is this? Sketch.
11.3 THREE-DIMENSIONAL SPACE, REGARDED AS AN INNER-PRODUCT SPACE
Following the pattern of Section 9.7, we identify the point P = (x, y, z) with the
� �
directed segment OP from the origin to P; we denote the resulting vector as P; and
for P1 = (x1, y1, z 1), P2 = (x2, y2, z2), we define addition, scalar multiplication, and
inner product by the formulas
- -
P1 + P2 = (x1 + X2,Y1 + Y2, Z1 + Zz),
-
CJ.Pi·= (()(X1, ()(Y1, )

()(Z1 ,
- -
P1 P2 = X1X2 + Y1Y2 + Z1Z2.
•
�
To simplify the notation, however, we drop the arrows, and write P for P. Thus we
have an inner-product space
V = {P} = {(x, y, z)},

11.3 Three-Dimensional Space, Regarded as an Inner Product Space 519
with addition, scalar multiplication, and inner product defined by the formulas
P1+P2= (x1 +X2,Y1 + Y2, Z1 + z2),

rxP1 = (rxxi. rxy1 , rxz1),
P1. P2=X1X2+Y1 Y2+ Z1Z2.
The resulting system satisfies all the vector and inner-product laws of Section 9.7.
In the new notation (without the arrows) these are as follows.
A.2. "f/' contains a vector 0 such that
O+P=P+O=P
for every P.
A.3. For every P in"f/' there is a vector -P in"f/' such that
P+ ( -P)= ( - P)+P= 0.
M.1. (rx{J)P= rx({JP).
M.2. (rx + {J)P= rxP+ {JP.
M.3. rx(P1 + P2)= rxP1 +rxP2.
M.4. 0 · P= 0, for every P.
M.5. 1 · P=P, for every P.
M.6. rxO= 0, for every rx.
In M.4 and M.5, 0 is the real number zero, and 0 is the zero vector. Thus M.4
says that the scalar product of the number 0 and any vector P is the zero vector; and
M.6 says that the scalar product of any number rx and the zero vector is the zero vector.
It is understood that all sums and scalar products are also vectors, that is, elements
of"f/'. But we had better make this explicit:
CA (Closure under addition). For every V1, V2 in"f/', Vi + V2 belongs to"f/'.
CSM (Closure under scalar multiplication). For every Vin "f/' and every real number
rx, rxV belongs to "f/'.
As in Section 9. 7, any set"f/', with operations satisfying the above laws, is called a
vector space. The space that we are dealing with at the moment, in which the vectors
are the triplets P= (x,y,z) of real numbers, is denoted by R3,and is called Cartesian
three-space. Thus
R3= {(x,y, z) I x,y,z in R}.
The inner product that we have defined in R3 has the following properties:
S.1. Pi·P2 = P2·Pi.

S.2. (Cl.Pi) · P2 = C1.(Pi ·PJ.
S.3. Pi · (P2 + P3) = Pi · P2 + Pi· Pa.
S.4. P ·PE;;; 0, for every P.
S.5. If P·P = 0, then P = 0.
As in Section 9.7, we define an inner-product space to be a vector space in which an

inner product is defined, satisfying S.1 through S.5.
The distance from 0 = (0, 0, 0) to P = (x, y, z) is
.jx2 + y2 + z2 = .jp . P.
Hereafter this number will be called the norm of P, and will be denoted by [[ P\[. (The
double bars are a reminder that we are performing an operation on a vector rather
than a number.) For inner-product spaces in general, the formula ../x2 + y2 + z2
may not apply, but the expression ../P ·P always has a meaning, and so we use it as
our definition of the norm.
Definition. In any inner-product space,

[[P [[ = .jp · P.
\IP [[ is called the norm of the vector P.
Just as in the plane, the inner product has a geometric interpretation in R3•
Theorem 1. In R3,
� --+
where() is the measure of the angle between the directed segments OP1 and OP2•
The proof is by definition of the. norm, together with the law of cosines:
2
(Pi P2) OPi + OP� - 2· OPi · OP2
= cos 0;
2 2 2
(X1 - X2) + (Yi - Y2) + (zi - Zz)
= x� + Yi + zi + x� + Y� + z� - 2[\P1\\ · \\P2\I cos O;
-2(X1X2 + Y1Y2 + ZiZ2) = -2[IP1\I . l\P2ll cos();
Pi · P2 = \\Pill ·l \P21\ cos 0.

From this we get immediately:

Theorem 2 (The Schwarz inequality). In R3,
2 2
(P1 P2) � JIP1ll2 ll P2ll •
·
•
2
This is true because cos () � 1.
Following the pattern of Section 9.7, we let
i = (1,0,0), j = (0, 1, 0), k = (0, 0, 1),
so that for each P = (x, y, z) we have
P =xi + yj + zk.
In general, if V = ix1V1 + ix2V2 + + ixnVn, for some scalars ix1, ix2, ..., ixn,
· · ·
then Vis a linear combination of the vectors Vi, V2, ..., Vn. Thus every vector in R3
is a linear combination of i, j, and k.
Definition. A set {V1, Vz, . .., Vn} of vectors spans the vector f' if every Vin f'
is a linear combination of the Vi's. (Thus {i, j, k} spans R3.)
A set {V1, V2, ..., Vn} is linearly dependent if there are scalars ix1, ix2, • • • , an,
not all equal to zero, such that
ix1V1 + ix2V2 + ... + anVn = 0.
Thus, in R3, every-set of vectors of the form {P, i, j, k} is linearly dependent, because
for P = (x, y, z) , we have P =xi + yj + zk, and so
P - xi - yj - zk = 0.
Here ix1 = 1, ix2 = -x, ix3 = -y, and ix4 = -z; and the numbers ixi are not all
=0, because ix1 = 1.
A set of vectors is linearly independent if it is not linearly dependent. Thus
{V1, V2, ..., Vn} is linearly independent if
For example, {i, j, k} is linearly independent. The reason is that in R3,

and 0 = (0,0,0).
Therefore
=> IX 1 = IXz = IX3 =0.

A set of vectors {V1, V2, •, Vn} forms a basis for a vector space f' if ( 1) the set
• •
spans f' and (2) the set is linearly independent. Thus we have:
Theorem 3. {i, j, k} is a basis for R3•
Obviously the points of the xy-plane form a vector space in themselves; in fact,
this is the vector space that was discussed in Section 9.7. In fact, all three of the
coordinate planes
Exv = {xi + yj + 0 k},

· Exz ={xi + 0 j + zk}, · Evz ={O i + yj + zk},
·
form vector spaces. Such sets are called subspaces of R3• More generally:
Definition. Given a vector space"/'" and a subset"/'"'. If"/'"' also forms a vector space
( under the same definitions of addition and scalar multiplication) then "/'"' is called
a subspace of"/'".
Thus a subspace must satisfy all of the vector laws. But this is not as tedious to
check as one might think, because of the following theorem.
Theorem 4. Let"/'"' be a subset of the vector space"/'". If"/'"' is closed under addition
and scalar multiplication, then "/'"' is a subspace of"/'".
Proof Many of the laws can be checked all at once. Since A.1 and A.4 hold for all
vectors in"/'", they automatically hold in"/'"'. The same is true for M.1 through M.6
and S.1 through S.5. Therefore the only things remaining to verify are A.2 and A.3.
1) By M.4, 0 · P= 0, for every Pin"/'"'; and by CA, 0 P belongs to"/'"'. Therefore

·
0 belongs to "/'"', and so A.2 holds.
2) Given P in"/'"'. By M.2,
[1+(-l)]P=1 P +(-l)P. ·
By M.4 and M.5, this gives

0 =P +(-l)P.
Therefore (-1)P= -P, and - P belongs to "/'"'.

On the basis of Theorem 4, it is easy to see that each of the three coordinate
planes is a subspace of R3• The point is that if
then
P1 + P2 = (x1 + X2)i + (Y1 + Y2)j,
so that the set E,,11 of all linear combinations of i and j is closed under addition.
Similarly, CJ..P1 = CJ..X1i + CJ..y1j, and so E,,11 is closed under scalar multiplication.
Similarly for Ev• and E,,,. In fact, a more general result holds:
Theorem 5. Every plane through 0 forms a subspace of R3•
Proof Let E be such a plane. Then E is the graph of an equation of the form
Ax+ By+ Cz= 0.
If P 1 = (x1, Ji, z1) and P2 = (x2, y2, z2) belong to E, then
Ax1 + By1 + Cz1 ::::: 0 and
By addition,
A(x1 + X2) + B(Y1 + Y2) + C(z1 + z2) = O;
and this means that P 1 + P2 is in E. Similarly, for every real number CJ..,
ACJ..X1 + BCJ..Y1 + CCJ..Z1 = CJ.. • 0 = 0,

and rxP1 is in E. Therefore Eis closed under scalar multiplication. By Theorem 4,

Eforms a subspace .
There is, however, a much better way to get this result, using vector-space
methods instead of using the results of the preceding section. Given the plane E
through 0, let
P0 = (A , B, C)
be any vector such that the line through 0 and P0 is perpendicular to E. Then for
� �
each P-:;!:. 0 in E, OP and OP0 are perpendicular, and so
Po · P =
llPoll · llPll cos (7T/2) = 0.
For P = 0 the same equation holds. Conversely, if P0 • P= 0, then Plies in E.

Therefore
E= {P I P0 • P = O}.
In terms of coordinates, this tells us that Eis the graph of the equation
Ax + By + Cz = 0,
which we already knew. But when we describe Eby a vector equation, using the
inner product, this suggests the following theorem:
Theorem 6. Let "I"" be any inner-product space; let V0 be any vector in "I"", different
from 0; and let
"I""' = {V I V0 • V = O}.
Then "I""' is a subspace of "I"".
Proof We need to show that "I""' satisfies CA and CSM. If Vi and V2 are in "I""', then
V0 • Vi = 0 = V0 • V2•
Therefore
Vo · (Vi + V2) = Vo · Vi + Vo V2, ·
by S.
3 . Therefore V0 • (Vi + V2) = 0, and Vi + V2 is in "I""'. Similarly,
V0 • a.Vi = a.Vi· V0 = rx(Vi V0)·

= rx(V0 • Vi) = rx · 0 = 0.
(Reasons for these steps?)

If you rewrite these formulas, in the forms that they take when V0, Vi, and V 2
are vectors in R3, with
V0 = (A, B, C),
you will find that you are simply copying the proof of Theorem 5. (This is worth
going through, to see how it works.) Thus it may seem that nothing is new in Theorem
6 except the notation. But this is not true, because Theorem 6 and its proof work in
every vector space, including spaces of four dimensions, spaces of functions, and so on.
Thus when we proved Theorem 6, we found that the method used in proving Theorem
5 had nothing to do with any special properties of R3; it depended only on the inner
product space laws. From now on, easy generalizations of this kind will occur often.
We shall treat the vector laws (or the inner-product space laws) as basic assumptions,
like postulates in geometry, and any theorems that we derive from them will be
known to hold in every vector space (or any inner-product space.)
If a plane E does not contain the origin, then it never forms a subspace of R3 ,
because it does not contain 0. But we can still write a vector equation for E, because
Ax+ By+ Cz + D = 0 <=> P0 P • = -D,
where P0 = (A, B, C). Thus we have:
Theorem 7. Every plane in R3 is the graph of an equation of the form
P0 • P =a.
PROBLEM SET 11.3
Each of the following is the equation of a plane. Convert each of them to the form
P0 • P = a, giving the value of a that you are using and the coordinates A, B, C of P0•
1. z=x+y 2. z=x - y
3. z= -x - y 4. x= 3y - 4z
5. y = 4z - 3x 6. z= 4x+ 3y
7. z= 1 8. x= 4 -
x y z x y z
+ + =4 lO. 2 - 4 + 3 = 2
9· 1 2 3
11. Let V1 = i +j, V2 =j + k, Va k. = Show that each of the basis vectors i, j, k is a
linear combination of V1, V2, V3•
12. Now show that {V1, V2, Va} spans R3.
13. Now show that {V1, V2, Va} is linearly independent. (By definition, this means that
cx1 V1 + ix2 V2 + ix3V 3= 0 =:-- ix1 = ix2 ix3 0. Problems 12 and 13, in combination,
= =
tell us that {V1, V2, V3} is a basis for R3.)

14. Let V1 = i +j, V2 = j + k, V3 = i + k. Proceed as in Problems 11 through 13.
15. Let V1= i - j +k, V2 = i +j - k, V3 = -i +j + k. Proceed as in Problems 11
through 13.
Show that the following hold, in any inner-product space. (Each of them should be
derived from the inner-product space laws, with a reason given for each step .)
16. V1 •(1XV2) = 1X(V1 V2) ·
17. (1XV1) ({JV2)

·
1X{J(V1 V2)
=
·
18. 1X(V1 + V2 + V3) =1XV1 + 1XV2 + cxV3

19. l\1XVll= 10(1 llVll
·
20. l\-1XVll =llXI llVll

·
21. (P+ Q) (R + S)
·
= P R + Q R + P S + Q S (Here P, Q, R,
· ·
· ·
S are vectors,
of course .)
22 . (P + Q)(P - R) P P+ Q P - R · P - Q R
= · · ·
23 . (P+ Q) (P - Q) = P · P - Q · Q
·
24. (P + Q) · (P + Q) P · P + 2(P Q) + Q Q
= · ·
25. (P - Q) (P - Q) = P· P - 2(P Q) + Q · Q
· ·
26. (P - Q) · (Q - P) = -P P + 2(P Q) - Q · Q
·
·
27. Is it true that in R3, (P · Q)R (Q · R)P? (Each side of this equation has a meaning,
=
because P · Q and Q R are scalars.)

·
28. Is the following true in any inner-product space? Theorem ( ?) If P Q = 0 for every ·
Q, then P = 0. Why or why not?

29 . Let the plane Ebe the graph of the equation z = 2x + 3y. Show that Econtains the
vectors V1 = i + 2k and V2 = j + 3k.
30 . Show that Econtains every vector Vof the form xV1 + yV2. Then show, conversely,
that every vector of this form lies in E. [Hint: Express Vin the form ( )i + ( )j +
( )k.]
31. Show that the V1 and V2 of Problem 29 are linearly independent.
32 . Show that Vi and V2 span E. (Problems 30 through 32 tell us that {V1, V2} is a basis
for E.)
33. Let the plane Ebe the graph of the equation x + y + 2z 0, and let V1 = i - j and =
V2 = 2j - k . Proceed as in Problems 29 through 32.

34. Let V1 = i + j, V2 = j + k; and let Ebe the set of all vectors of the form V = xV1 +
y V2• Write an equation for E, in the form Ax + By + Cz 0. =
35. Let V1 = i + 2j, V2 = j + 2k. Proceed as in Problem 34 .

36. Two vectors V, V' are orthogonal if V V' 0. A set { V1, V2, •••Vn} of vectors is
· =
orthogonal if every two (different) vectors in the set are orthogonal. Verify that {i, j, k }
i s an orthogonal set.
37. A set { V1, V2, •• . , Vn} is orthonormal if (1) the set is orthogonal, and (2) II Vi II 1 for =
each i. (Thus the basis {i, j, k} for R3 is orthonormal.) Given that { V1, V2, V3} is
orthonormal, express the inner product
��+��+��·��+��+��
in the simplest possible form.
*38 . Let R4 be the set of all quadruplets P = (w, x, y, z) of real numbers. The set R4 forms
an inner-product space, under the obvious definitions of sum, scalar product, and inner
2 2 2
product. As always, llPll vp · P. Show that (P1 • P2) � llP1ll • llP2ll •
=
*39. Let P be the set of all polynomials with real coefficients. For
n n
V =�a.x i W �b.xi
£., • ' =
£., • '
i=O i=O
we define
n
V+ W i
= L (ai + bi) x ,
i=O
n
i
�v = L �aix ,
i =O
n
V· W = Laibi.
i=O
Show that
2 2 2
(V· W) � llVll • llW ll •
40. Does the vector space P of Problem 39 have a finite basis? If so, describe such a basis.
If not, explain why no such basis exists.
11.4 THE DIMENSION OF A VECTOR SPACE.

VARIOUS WAYS TO FORM A BASIS
For each positive integer n, let Rn be the set of all n-tuples of real numbers. Thus
Rn = {(x1, X2, ••• 'Xn) I xi ER},
and Rn forms an inner-product space, under the obvious definitions of sum, scalar
product, and inner product. Rn is called Cartesian n-space. Let
Bn = {E1. E2, ... ' E n},
where
E1 = (1, 0, 0, ..., 0), E2 = (0, 1, 0, ..., 0), ... ,
E n-i = (0, 0, ..., 1, 0), En = (0, 0, ... , 0, 1).
(In general, Ei has 1 in the ith position, and O's everywhere else.) The vectors Ei span
the space Rn, with
n
(xi. x2, ••• , x n) = L xiEi.
i=l
They are linearly independent, since
n
L xiEi = 0 => (x 1, x2, ••• , x n) = (0, 0, ..., 0)
i=l
Therefore Bn forms a basis.

Every subset of Bn gives a subspace of Rn. For example,
"f/1 = {ixE1 \ ix ER}
is a subspace, and forms a line through the origin (0, 0, . . . , O);

"Y2 = {ixE1 + fJE2 \ ix, (J ER}
11.4 The Dimension of a Vector Space. Various Ways to Form a Basis 527
is a subspace, and forms a plane, and so on, for any subset of Bn. But Rn has
many subspaces which are not obtainable in this way. For example, we have found
that in R3, any line or plane through the origin forms a subspace. To investigate
these other subspaces, we need to use bases other than the obvious basis Bn. Our
investigation of other bases will also be useful in other connections.
The key to the theory of bases is the following theorem.
Theorem 1. Let "f/ be a vector space. Let A be a set of m vectors, and let B be a set
of n vectors, such that (1) A is linearly independent, and (2) B spans '"f/. Then "f/ is
spanned by a set C consisting of (a) all the elements of A, and (b) exactly n m-
of the elements of B.
"f/ =R ,
a
A = {E1 + E2, E2 + Ea},

B = B3 = {E1, E2, E }.
3
Here we can take
C = {E1 + E2, E2 + E3, Ea}.
This has the desired form, using all elements of A and 3 - 2 = 1 element of B. To
see that C spans R3, we observe that
Therefore every basis element E1, E2, E is a linear combination of the elements of C.
3
Therefore every vector in R3 is a linear combination of the elements of C, and C
spans R3•
We proceed to the general proof. First we list the elements of A in such a way
that the vectors which also belong to B come first. Thus
where A1, A2, • • • , Ai belong also to B, but Ai+l• ... , Am do not. Then B can be
described in the form
One of the possibilities, of course, is that i

= 0, as in the above example. Another
possibility is that i
= m; in this case there is nothing to prove. We shall show that
given a set B, as above, with 0 � i < m, we can always delete one of the vectors B1,
and replace it by Ai+I• getting a new set which also spans '"f/. In a finite number of
such steps, we get a set C of the sort that we want.
Since B spans V, it follows that Ai+l is a linear combination of the form
Ai+l = ct.1A1 + ct.2A2 + ... + ct.;A; + fliB1 + f32B2 + ... + fJn-iBn-i•
Here it cannot be true that all the numbers {J1 are equal to zero, because if so it
would follow that A is linearly dependent. It is a matter of notation, therefore, to
suppose that {31 :;rt: 0. We can therefore solve for B1, in the above equation, getting
1
B1 = f31 (Ai+l - Cl1A1 - Cl2A2 - ... - Cl;Ai - f32 B2 - ... - /Jn-iBn-i).
Now let
B' = {A1, A2, • • • , A;, Ai+i• B2, • • • , Bn-i}·
Every element of B (including B1) is a linear combination of the elements of B'; and
B spans "f/. Therefore B' spans "f/. In m - i steps of this type, we get the desired
set C.
Let us now check to see how this general scheme of proof applies to the above
example. We had
A = {A1, A2} = {Ei +E2, E2 +Ea},
B = { B1, B2, Ba} = {E1, E2, Ea}·
Here m = 2, n = 3, and at the outset, i = 0. Also A1 is a linear combination of the
elements of B, with
A1 = E1 +E2•
This equation can be solved for Ei. giving Ei as a linear combination of Ai and E2.
Therefore we can replace Ei by Ai in B, getting
B' = {Ai, B2, Ba} = { Ei +E2, E2, Ea}.
This completes step 1. Next we express A2 as a linear combination
E2 +Ea. A2 =
This equation can be solved for E2, giving E2 as a linear combination of A2 and E 3•
Therefore we can replace E2 by A2 in B', getting
B" = C = { Ei +E2, E2 + Ea, Ea};

and now we are done.
We shall now pursue some of the consequences of Theorem 1.
Theorem 2. Let "f/ be a vector space. Let A be a set of m vectors, and let B be a
set of n vectors, such that (1) A is linearly independent, and (2) B spans "f/. Then
m � n.
This follows from Theorem 1, because Chas n elements, and contains all of A.
Theorem 3. If a vector space "f/ has a basis with n elements, then every basis for "f/
has exactly n elements.
Proof Let B be a basis with n elements, and let A be any other basis, with m elements.
Then A is linearly independent, and B spans "f/. By Theorem 2, m � n. But we also
know that Bis linearly independent, and A spans "f/. By Theorem 2, n � m. There
fore n = m, which was to be proved.
Thus the number of elements in a basis is independent of the choice of basis.

This justifies the following definitions.
11.4 The Dimension of a Vector Space. Various Ways to Form a Basis 529
Definitions. A vector space Y is finite-dimensional if it has a finite basis. If Y is

finite-dimensional, then the dimension of Y is the number n which is the number of
elements in every basis. The dimension of Y is denoted by dim Y.
If you review the conditions for a vector space, you will see that they are all
satisfied in the trivial case where Y contains a zero vector 0 and nothing else. In
this case we define dim Y = 0. That is,
dim {O} = 0.
(Here the empty set is being regarded as a "basis" for {O}.) In a way it is a nuisance
to allow this case, but to rule it out would lead to worse nuisances in the long run.
Theorem 4. In an n-dimensional vector space, every set of more than n vectors is

linearly dependent.
Proof Let B be a basis, with n elements, and let A be any set of vectors, with m
elements, with m > n. If A were linearly independent, this would contradict Theorem
2. Therefore A is linearly dependent, which was to be proved.
Theorem 5. In an n-dimensional vector space, every linearly independent set with n

elements is a basis.
Proof Given a linearly independent set B {V1, V2,

= , Vn}. If B spans Y, then
• • •
B is a basis. If not, there is a vector Vn+i which is not a linear combination of elements
of B. It follows that the larger set
B' = {V1, V2, • • • , Vn, V n+i}

is linearly independent; and this contradicts Theorem 4.
Theorem 6. Let Y be a vector space, and let B = {V1, V2, . . • , Vm} be a set which
spans Y. Then B contains a basis for Y.
Proof If B is linearly independent, there is nothing to prove. If not, some V; is a

linear combination of the others. Suppose that this is V1• Then {V2, • • • , Vm} also
spans Y. Repeating this process, removing superfluous vectors one at a time, we get
a basis.
Theorem 7. If dim Y = n, then no set of fewer than n elements spans Y.
Proof Any set which spans Y contains a basis, and every basis has n elements.
Theorem 8. Let Y be an n-dimensional vector space, and let "f/"' be a subspace off.
Then "f/"' is finite-dimensional, and dim f' � n.
P;oof Let m be the largest number for which it is true that "f/"' contains a linearly
independent set of m vectors. (By Theorem 4, there is such a largest number m, and
m � n.) Let
B = {V1, V2, • • • , Vm}
be a linearly independent set in "f/"'. We assert that B spans "f/"'. (Proof If not, there
is a vector Vm-'-' in "f/"' which is not a linear combination of elements of B, and it
follows, as in the proof of Theorem 5, that the larger set
B' ={Vi, V2, • • • , Vm, Vm+i}

is linearly independent.) Therefore B is a basis for "/"', and dim"/"' =m � n.
Thus every subspace "/" of Rn has a basis.
PROBLEM SET 11.4
1. Given Vi = Ei +2E2, V2 = 2Ei + 3E2, Va = 3Ei +4E2, in R2• Theorem 4 predicts

that. {Vi. V2, Va} is linearly dependent. Exhibit the linear dependence, by finding
numbers °'i• °'2, °'a• not all = 0, such that z °'iVi = 0.
a
2. Given Vi =Ei +2E2, V2 = 2E2 + 3Ea, Va = 3E2 +4Ea, Vi =4Ei - 5E2, in R ,
proceed as in Problem 1.
a
3. Given Vi =Ei +E2, V2 =E2 +Ea, Va = E2 + 3Ea, V4 =4Ei +E2 +Ea, in R ,
proceed as in Problem 1.
4. In Ra, let'"/!'" ={VI V · (Ei +E2)= 0}. Find a basis for '"fl'".
5. Same question for'"/!'" ={VI V · (Ei +E2 - 2Ea)= O}.
6. In R4, let'"/!'" ={VI V· (Ei +E2 +Ea)= O}. Find a basis for '"fl'".
7.
4
Find a basis for R , using Ei +E2 and Ea +E4 as basis elements.
8.
a
Find a basis for R from among the vectors of
B ={Ei +E2 +2Ea, E1 +Ea, E a +E2, E1 - E2, Ei +E2 +Ea}.

9. In Ra, let'"/!'" ={VI V· (E1 +E2)=0, V · (£1 +2Ea)= 0}. Find a basis for '"fl'".
10.
4
In R , let '"fl'"={Vj V· (E1 +E2 +Ea)= 0, V· (E2 +2Ea +E4)= O}. Find a
basis for'"/!'".
11.5 ORTHONORMAL BASES
In the preceding section, we found for the space Rn a basis
Bn = {E1. E2, ... ' E n},

where the ith coordinate of Ei is 1 and the other coordinates of Ei are 0. Thus
n
V = (x i x2, , xn) = ,L xiEi.
i=l
. • • •
If
n
W (Yi. Y2, · · ·' Yn) =
L Y;E;,
i=l
=
then
n
.
v w =
.L XiYi·
i=l
Thus, for linear combinations of the E/s, we have a simple formula for the inner
product:
n
=
L X;Y;·
i=l
11.5 Orthonormal Bases 531
This formula does not hold for all bases. For example, the set
{ Vi, V2} = {E1, E1 + £2}

forms a basis for R2, but it is not true that
( ?) (cx1Vi + IX2V2) •
(f31 Vi + f32V2) = rx1f31 + rx2f32 ( ?).
(Try taking rxi = fJi = {32 = 1.) But the above formula for the inner product does
hold for a certain kind of basis, now to be defined.
Vi and V2 are called orthogonal Vi · V2 = 0. Ei £2-
Rn.) More generally, a set
Two vectors if (Thus and
are orthogonal in
B= {Vi, V2, • • • , V n}
is orthogonal if
for i � j.
Thus Bn = {Ei, £2, • • • , En} is an orthogonal set, but {Ei, Ei + £2} is not, because
Ei · (Ei + £2) = 1 + 0 = 1 � 0.
If
ll V;ll = 1 for each i,
then B= {Vi, V2, , V n} is normal. If Bis both orthogonal and normal, then B
2
orthonormal. Thus nn is orthonormal. Since II Vill = vi . vi, we note that Bis
• • •
is
orthonormal if and only if
V;·V.=
' o { 1
for i � j
for i = j.
Theorem 1. Every finite-dimensional inner-product space has an orthonormal basis.
Proof We shall show, by induction, that every n-dimensional inner-product space

has an orthonormal basis. For dim"/" = 1, this is obvious: Given a basis {Vi}, we
let W1 = Vi/llVill. Then {W1} is an orthonormal basis. We suppose, then, that every
n-dimensional inner-product space has an orthonormal basis; and we need to show
that every (n + 1)-dimensional space has the same property.
Given dim "/" = n + 1, let
B= {Vi, Vi, ... 'Vn, Vn+i}

be a basis for V. Let "/"' be the subspace spanned by {Vi, V2, , V11}. Then
dim"/"' =
• • •
n, and so"/"' has an orthogonal basis

W = {Wi, W2, • • • , W11}.
Then the set C = {Wi, W2, , Wn, Vn+i} is a basis for"/". (Check that C is linearly
V�+ i such that the set
• • •
independent, and spans"/".) We shall now find a vector
D = {Wi. W2, ...' wn> v�+l}

spans"/" and is orthogonal. For each i, let a i = Wi · V n+i· Let
n
v�+l = vn+l - .L ai JV;.
i�i
Then for each k from 1 to n we have

n
wk· v�+i = wk Vn+1 - 2 aiwk
· · a:.
i=l
In the sum on the right, the only nonzero term is Wk Wk, because Wis an orthogonal ·
set; and Wk Wk · 1, because W is orthonormal. Therefore

=
W k V�+1
· = Wk· Vn+i -
ak = ak - ak = 0.
Therefore D is an orthogonal set. The last step is trivial. Let

IX= llV�+1ll,
and let
Then
and
Therefore the set { Wi. W2, • • • , Wn, Wn+i} forms an orthonormal basis.
Note that the pattern of this proof supplies us with a method of actually finding
an orthonormal basis, starting with a basis which is not necessarily orthonormal.
The proof gives a scheme for "orthonormalizing" a given basis, a step at a time.
For example, in R3 let V be the subspace spanned by
B = {V1, Vi} = {E1 + E2, E2 + £3}.
Then Bis a basis for V, but is neither orthogonal nor normal. We can get an ortho
normal basis for V by following the pattern of the proof of Theorem 1.
1) Let
Then II Will = 1.
2) We shall now treat V2 as the Vn+i of the above proof. Let
Let
v; = V2 - a1W1 = (£ 2 + E3) - t(E1 + E2)
=
-
tE1 + !E2 + £3.
The theory predicts that
11.5 Orthonormal Bases 533
and this checks, because
W 1 V� =
•
:2. (E1 +E2) (-tE1 +fE2 +Ea)
•
1
= ./2 (-t +t +0) = 0.
3) Now we normalize V �, by letting
W2 = V�/11V�ll-
Since
we have
llV�ll
2
= V2 ·
V� = ! +t + 1 = -f,
J-3
1 2
=
11v;11
and
1 1
W2 = - ._j(, E1 +.Jf, E2 +yr
i £3,
so that
II W2ll2 = i +i +i = 1
,
and II W2II = 1 , as it should be. Now { W1, W2} is an orthonormal basis.
Orthonormal bases are what we need to get a simple formula for the inner
product:
Theorem 2. If {V1, V2, • • • , Vn} is orthonormal, then
= 2 ;{J
i=l
ct. ;.
Proof We know that

(i ¥= j),
and
Therefore
ct.iV; . L f3;V; =
n
i=l
ct.;{3; for each i,
and so
(i: v:) ( i f3 ivi) = i

•=l
Cl.;
·
J=l i=l
ct.; {3;.
Of course, for inner products of the form V V = II V l!2,
· we have
R2 and Ra, it is easy to

- Q II.
In see by the distance formula that the distance between
any two points P and Q is llP For inner-product spaces in general, we use the
latter formula as the definition of distance. Often we shall think of inner-product

spaces geometrically, and so we may refer to their elements as points rather than
vectors, as in the following definition.
Definition. In any inner-product space, the distance between two points P and Q is
llP- Qll.
The distance between P and Q may be denoted by d( P, Q), or simply PQ. An
orthonormal basis gives us a distance formula.
Theorem 3. Let B = {V1, V 2, • . • , Vn} be an orthonormal basis for the inner-product

space V, and let
n n
P = .L aiV;, Q = .L Pit:·
i=l i=l
Then
Proof We have
n
d(P, Q) = llP - Qii , P- Q = .L (a; - p;)Vi,
i=l
and
llP - Qll2 =
(P- Q) · (P - Q).
Since the basis is orthonormal,
2 n
llP - Qll L (a;- P;)2,

i=l
=
and the theorem follows.
PROBLEM SET 11.5

3
1. Let V be the subspace of R spanned by the basis B ={V1, V2} ={E1 +E2 + E3,
E1 +E2}. Find an orthonormal basis for V.
2. Same question for B ={V1, V2} ={E1 +E2 +E3, E1 - E2}.
3. Use the orthonormalization scheme described in the proof of Theorem 1 to find an
orthonormal basis for the subspace spanned by the basis
B ={E1 +E2 +Ea, E2 +Ea, Ea} ={V v Vi, Va}·

4. Same question for B ={E2 +E3, E3, E1 +E2 +E3} ={V1, V2, V3} .
5. Same question for B ={E3, E1 +E2 +E3, E2 +E3} ={V1, V2, V3}.
3
6. In R , find an ={V / V · (E1 +E2 +E3) = O} .
orthonormal basis for 'i'
7. Same question for 'i' ={V / V (E1 +2E2) = O}.

·
8. Let V1 = 2E2 +E3, V2 = 4E1 +E4. Find vectors V3, V4 so that {Vv V2, V3, V4} is
4
an orthogonal basis of R •
9. Given V1 = E1 - E3 +E4, V2 = E1 +E2 +E3, proceed as in Problem 8.
4
10 . In R, find an orthonormal basis of the subspace
'i' ={VJ V· (E1 +E2) = 0, V· (E2 +E4 ) = 0}.

11.6 The Schwarz Inequality. More General Concepts of Norm and Distance 535
11. Same question forr ={VI V · (£1 + £2) = 0, V· E3 = 0, V · E4 = O}.

12. Show that if V1 is a fixed nonzero vector in an n-dimensional vector space, then
= {VI V · V1 = O} is an (n - 1)-dimensional subspace.
"//'
13. Suppose that {V1, V2, , Vn} is orthogonal, but not necessarily orthonormal.
• • • Find
a formula for
14. Let {V1, V2, • • • , Vn} be an orthogonal set of nonzero vectors. Let V = L ixi V; be any
linear combination of them. Show that
V· V; = 0 for every i =<>- V = 0.

15. Let {Vv V2, , Vn} be as in Problem 14. Show that the set is linearly independent.
• • •
(This means that in an n-dimensional vector space, no orthogonal set of nonzero vectors
can have more than n elements. Thus, for example, in R3 there is no set of four con
current lines, every two of which are perpendicular.)
16. Let ii'" and f1E be subspaces of a vector spacer. If every vector in ii'" is orthogonal to
every vector in f1E, then ii'" and f1E are orthogonal subspaces, and we write ii'" J_ fE. Show
that if ii'" J_ fE, then dim ii'" + dim q; ;;:;; dim r. Give an example to show that the
equality does not necessarily hold.
17. The following is a converse of Theorem 2.
Theorem (?) Let B ={Vi. Vi, . .. , Vn} be a basis for i"". If for all vectors V = L ix; V;,
W = ! /3;V;, we have
then B is orthonormal.
Is this true? Why or why not?
18. Show that if {V1, V2, • • • , Vn} is an orthonormal basis for r, then for every Vinr,
n
v = L (V· V;)V;.
i�I
That is, for V=L IX; V;, we always have IX; = V V;.
·
11.6 THE SCHWARZ INEQUALITY.

MORE GENERAL CONCEPTS OF NORM AND DISTANCE
The "triangular inequality" for points in a plane (or in three-dimensional space)

asserts that for any points P, Q, R we have
PR� PQ +QR.
The equality holds if the points are collinear, and Q is between P and R; and in every
other case, the strict inequality holds.
We propose to show that in any inner-product space, the same inequality holds
for distances. That is,
d(P, R)� d(P, Q) +d(Q, R), (1)
for every P, Q, and R. Since distance was defined by the formula
d(P, Q) = llP - Q ll.

the proposed inequality means that
llP - R\\ � ll P - Q\\ + \\Q - RI\.

This has the form
ll A + Bii � ll A ll + llBI\, (2)
where A = P - Q and B = Q- R. Note that this is analogous to the inequality
\x + YI � \x\ + IYI,
which is known to hold for both real and complex numbers. Obviously any general
proof of (2) , for all inner-product spaces, must appeal to the definition of the norm:
ll A ll =
..{;0A, \I A ll2 = A · A. (3)
Therefore the natural first step, in proving (2) , is to restate it in terms of the definition
given in (3). In these terms,
ll A + Bii � ll A ll + llBll (2)
<::;, l\ A + Bll2 � \I A \\2 + 2 l\ A \I · llBll + llBll2

<::;,( A + B) · ( A + B) � A · A + 2 ll A ll · llBll + B B ·
<::;, A · A + 2A · B + B· B � A · A + 2 ll A ll · llBll + B B ·
<::;, A · B � \I A I\ · llBll . (4)
Here l4) automatically holds whenever A · B < 0. But if ( 4) always holds, then
we must have
I A · B\ � l\ A I\ · ll Bll. (5)
(5) were false, for some A , B, then (4) would also be false, either for A , B or
(If
- , B.) And Eq. (5) is obviously equivalent to
A
( A · B)2 � ll A ll2 • llBll2• (6)
Formula (6) is called the Schwarz inequality. We shall now prove it.
Theorem 1 (The Schwarz inequality). In any inner-product space,
( A · B)2 � ll A ll2 • llBll2, (6)

for every A and B.
Any proof must use, at some stage, the fact that P· P � 0 for every P, because
this is the only inequality that is given by the inner-product space laws. But to use
this law to prove the theorem, we first need to reduce the theorem to a manageable
special case. This is done as follows.
If A = 0 or B = 0, then the inequality (6) takes the form 0 � 0, which is true.
We may therefore suppose that A -:;if 0 and B -:;if 0. For the same reason, we may
suppose that A · B -:;if 0.
Suppose now that we replace A by ex. A and B by (JB, with a -:;if 0 and (J -:;if 0.
The inequality (6) then takes the form
(l't.A (JB)2 � lll't.All2 · llfJB ll2

·
(7)
<=> i't.2(J2(A. B)2 � l't.2 llAll2. (32 llBll2
<=>(A· B)2 � llAll2 • llBll2• (6)

Since (7) <=> (6), for every l't. , (J =;I= 0, we are free to choose l't. and (J as we please.
We take
1
l't.=- so that lll't.A ll = 1.
llAll'
We then choose
1
(J= so that l't.A (JB = I.
·
l't.A B
-
We let P = l't.A, Q = (JB. Our theorem now takes the form
llPll = 1 and P · Q= 1 => 1 � llQll2•

This is easy to prove: for llPll = 1, P · Q = 1, we have
(P - Q) · (P - Q) � 0
=> P · P - 2P Q + Q · Q � 0·
=> llPll2 - 2P Q + llQll2 � 0

·
=> 1 - 2 + 11Q11 2 � o
=> 1 � llQll2•
The theorem follows. In the light of the discussion which led to the Schwarz inequality,
we also have the following:
Theorem 2. In any inner-product space,
llP + Qll � llP ll + llQll· (2')

( We sometimes express this by saying that the norm is subadditive. In general, a
real-valued function is subadditive if f (x + y) � f (x) + f (y), for every x and y.)
Theorem 3. In any inner-product space, distance is triangular. That is,
d(P, R) � d(P, Q) + d(Q, R) .

Equivalently,
llP - Rll � llP - Qll + llQ - Rll .

Let us now review what we know about norms and distance. For norms, we have
N.1. llAll � 0 for every A.

N.2. llAll = 0 => A = 0.
N.3. lll't.All
= ll't.I · llAll . (Homogeneity)
N.4. llA + Bii � llAll + llBll. (Subadditivity)
All these are easy to check, on the basis of the definition llA ll =.JA· A, the vector
laws, and the Schwarz inequality. For distance, we have
D.l. d(P, Q) � 0.
D.2. d(P, Q) =0 => P = Q.
D.3. d(P, Q) =d(Q , P ).
D.4. d(P, R) � d(P, Q) + d(Q , R).
On this basis, we shall define various types of mathematical systems which are
more general than inner-product spaces.
Definition. A normed vector space is a vector space in which a norm is defined,

satisfying N.1 through N.4.
Thus a normed vector space is a quadruplet
[r, +, sm, II 111.

where [r, +, sm] is a vector space, and the "norm operation" P � ll P ll satisfies
N.l through N.4.
Example 1. Let 21 be the set of all continuous functions f, on the interval [ - 1 1 ].
,
Addition and scalar multiplication are defined in the obvious way. We define
llfllu =max lfl,

where max lfl is the largest of the numbers If(x)I ( -1 � x � 1). It is easy to check
that N.l, N.2, and N.3 hold. To verify N.4, we observe that for each x,
If(x) + g(x)I � If(x)I + lg(x)I

� llfll + llgll.
Therefore max {lf(x) + g(x)I} � llfll + llgll. Thus we have a normed vector space.
The norm that we have just defined is called the uniform norm. (Hence the notation
II II.,.)
Example 2. Let 21 be as Example 1; and let + and sm be as before. We can define
another norm by the formula
llfll1 =f lf(x) I dx.

1
As before, the verifications of N.l, N.2, and N.3 are easy. To verify N.4, we observe
that
II!+ gll 1 = f1lf(x) + g(x)I dx � f1[1f(x) I + lg(x)I] dx
(x) I dx + f 1lg(x) I dx
1lf
=f
= 11!111 + II gll1·
It should be emphasized that a normed vector space is not merely a vector space
in which a norm can be defined, but rather a linear space in which a norm has been
defined. Thus, in Examples 1 and 2 we defined two different norms II II,, and II 111
in the same vector space [21, +, sm]; and this gave us two different normed vector
spaces
[21, +, sm, 11 II,,], [21, +, sm, 11 Iii].
In any normed vector space, we can define distance by means of the formula
d(A, B) = llA - Bii.

It then follows that the distanced satisfies D.1 through D.4. (This should be checked.)
More generally, we state the following.
Definition. A metric space is a set S which is provided with a distance d, satisfying
D.1 through D.4. The distanced is called the metric.
Thus a metric space is a pair [S,d ], where d is a metric for S. It is evident that
metric spaces can arise in ways that have very little to do with vector spaces or with
norms. For example, S may be the surface of a sphere in R3; that is,
S = {(x, y, z) I x2 + y2 + z2 = 1},
and for each pair of points P, Q on S, the distanced(P, Q) may be the length of the
shortest arc on S, joining P and Q. It is not hard to see that this system forms a
metric space, that is,d satisfies D.l through D.4. In fact, this is the metric space used
in navigation on the open sea, with arc length measured in nautical miles.
PROBLEM SET 11.6
1. Show, by any method, that for any pair of pairs of real numbers (x1, x2) and (y1, y2)
we have (x� + x;)(y� + y�) ;;;;; (x1y1 + X2J2)2 •
2. Show, by any method, that for every pair of finite sequences xl> x2, ••• , Xn, Yi. y2, ••• ,
Yn of real numbers,
3. Let E be a coordinate plane and, for each P = (x, y) and Q = (a, b), let
d(P, Q) = (x - a)2 + (y - b)2.
Thus d is the square of the usual distance. Does [E,d] form a metric space?
4. Same question,for d(P, Q) Ix - al + ly - bl.
=
5. ·Same question,for d(P, Q) maxmium of Ix - al and ly - bl.

=
6. If [S,d] is a metric space and k > 0, does it follow that [S, kd] is a metric space?
7. The real number systemR clearlyforms a vector space. For eachxinR,let llxll = Vlxl.
Does this give a normed vector space?
8. Let S be the set of all airline passenger terminals in the world; and for each P, Q in S
let d(P, Q) be the minimum number of hours required to get from P to Q by a combination
of regularly scheduled flights. Is [S, d] a metric space?
9. Let .!l'1 be the set of all continuous functions on the interval [ -1, 1], as in Example 1
of the text. For each/, g in .!l'1, let
d (/,g )
f 1 l/(x) - g(x)I
dx.
l/(x) + g(x)I
=
-1 1 +
Does this give a metric space?

*10. For the definition of U Jim, see p. 502. Let .!£'1 be the normed vector space, with
the "uniform norm," defined in Example 1 of the text. Show that for any sequence
/1,f2, ... of functions in .!l'1,
lim 11/n llu = 0 <=> U lim/n = 0.
(This is why the norm defined in Example 1 is called the uniform norm.)
11. Let .!l'1 be the same as in Problem 10, with the norm
11/111 =
f 1
l/(x)I dx,
as in Example 2 of the text. Is it true that
Jim llfnll1 =
0 => U lim/n = O?
n-co n-oo
Is it true that
U lim fn = 0 => Jim llfnll O?
1 =
12. Let C0[-7T, 7T] be the set of all continuous functions/ on the interval [-7T, 7T], with +
and sm defined as usual. Set f g (f"..." f(x)g(x) dx), and verify that C0[ - TT,
· = ] with
TT ,
this inner product, forms as inner-product space.

13. Find a polynomial f(x) a + a1x + x2 which is orthogonal to g(x) 1 and
0
= =
h(x) = x, in the inner-product space defined in Problem 12 .

14. Find an orthonormal basis for the subspace of C0[-7T, 7T] spanned by the elements
{l,x,x2}.
*15. For each n, let Tn be the set of all functions of the form
n
f(x) a + I [ai cos ix + bi sin ix],
0
=
i=l
on the interval [-7T, 7T]. (Such functions are called trigonometric polynomials.)
Evid�nt!y Tn forms a subspace of C0[-7T, 7T], and the set
B = { 1 ; cos x, cos 2x, . . . , cos nx ; sin x, sin 2x, . . . , sin nx}
spans Tn. Show that (a) B is orthogonal, and (b) B is a basis for Tn. Then find an
orthonormal basis for Tn. [Warning: This one is long. It is easier if you note that in
(a) you need not necessarily compute indefinite integrals; what you need, in each case,
is the definite integral, from - 7T to 7T. The identities
cos (A + B) - cos (A - B) =
-2 sin A sin B,
cos (A + B) + cos (A - B) = 2 cos A cos B
are also useful.] Problem 15 of Problem Set 11.5 is useful at one stage.
12 Fourier Series
12.1 PROJECTIONS INTO A SUBSPACE.

TRIGONOMETRIC POLYNOMIALS AND FOURIER SERIES
The idea of a projection is taken from elementary geometry. Let Ebe a plane in R3,
and let P be a point . To suit the terms of our later discussion, suppose that Epasses
through the origin. Then the projection of P into Eis the point Q which is the foot of
the perpendicular from Pto E. (If Pis in E, then the projection of Pis P.) The follow
ing facts are well known:
+-�
1) The line PQ, through Pand Q, is perpendicular to every line in Ethat contains Q.
++
(In fact, this is the definition of the statement that PQ J_ £.)
2) If P is not in£, then there is one and only one point Q in£, satisfying (1).
3) If R is in£, then
(PQ)2 + (Q R)2 = (PR)2•
It follows immediately that:
4) The distance from P to Q is the minimum distance from P to E.
We shall now regard R3 as an inner-product space , regard the points P, Q, . ..

as vectors, and restate these ideas in vectorial form.
In the figure below , we have completed a rectangle, by inserting the point P - Q.
(Check that in the figure, (P - Q) + Q = P, as it should be.) In elementary geo
metric terms, 0 is the foot of the perpendicular from P - Q to 0. Therefore (1)
541
542 Fourier Series 12.1
through (4) take the following forms:
1) (P - Q)· S = 0, for every Sin E.
2) There is one and only one Q in E, satisfying (1).
3) If R is in E, then
\IP - Q\12 + llQ - R\12 = \IP - R\12.
Therefore
4) ll P - Qll is the minimum distance from P to E.

We shall now show that these ideas apply under far more general conditions, as
follows .
Theorem 1. Let "I"' be any vector space, let "fr be any finite-dimensional subspace,
and let P be any vector in "I"'. Then there is one and only one vector Q in "fr such that
(P - Q) · S = 0 for every Sin "Ir.
Proof The easy part is to show that there is only one such Q. If
(P - Q) · S = 0 and (P - Q') · S = 0,
for every S, then
[(P - Q) - (P - Q')] · S = 0,
for every S. Therefore
(Q' - Q) . s = 0,
for every S. In particular, for S = Q' - Q, we have (Q' - Q) (Q' - Q) · = 0, and

so Q' = Q.
To show that there is one such Q, we need to find one; and as a guide, we look at
a simple case. For
"I"' = R3 = {x1E1 + x2E2 + xsEs},

"fr = R2 = {x1E1 + X2E2},
P = 1:t..1E1 + 1:t..2E2 + (/.3E3,
we ought to have
12.1 Projections into a Subspace. Trigonometric Polynomials and Fourier Series 543
here Q is the projection of Pinto the xy-plane. This works:

P- Q = 0 · E1 + 0 E2 + rxaE3, ·
(P - Q) E1 · = (P - Q) E2 · = 0,
and so
(P - Q) S · = (P - Q) · (x1E1 + x2E2) = 0
for every Sin "fl/. In our formula for Q, we have

Q . E1 = rx1E1 . E1 + rx2E2 . E1 = rx1,
Q · E2 = rx1E1 E2 + rx2E2 E2
•
• = rx2;
and so
Q =
(Q · E1)E1 + (Q · E2)E2.
We shall see that this pattern carries over to the general case. Let
B = {W1, W 2, • • • , Wn}
be an orthonormal basis for "fl/; for each i from 1 to n, let
rx; = P W;,
·
and let
n
Q = ! rxiW i.
i=l
Then
n 2
Q · W; = ! rxiW,: · W; = rx; llW;lt = rx3,
i=l
and so
(P - Q) ·
W; = P W; - Q
· · W; = rx; - rx; = 0,
for each j. Therefore for every
n
S = ! /J3 W; E "fl/,
j=l
we have
n
(P - Q) · S = ! /J;(P - Q) · W; = 0,
j=l
The point Q is called the projection of Pinto "fl/, and is denoted by Pr P (or by
Prif" P, if there is any doubt about the subspace into which we are projecting). To
repeat:
Definition. Let "f/ be any vector space , and let "fl/ be any finite-dimensional sub
space of "f/. For each P in "f/, Pr P (or Prif" P) is the point Q of "fl/ such that
(P - Q) S 0 for each Sin "fl/.
· =
Theorem 1 tells us that this definition defines something. And one of the ideas
in the proof of Theorem 1 is worth noting for future reference:
Theorem 2. If Pr P is the projection of P into "fl/, and {W1, W 2, • • • , Wn} is an
orthonormal basis for "fl/, then
n
Pr P ! (P W,:)W,:. = ·
i=l
(In the proof of Theorem 1, we found that this sum satisfied the conditions for Q,
and that there is only one such Q.) It remains to show that conditions (3) and (4),
stated at the beginning of this section, hold on the basis of our general definition.
Theorem 3. If Q = Pr P, and R is in if/, then
llP - Qi12 llQ - Ri12 = llP - Rll2

+
Proof Obviously (P - Q) + (Q - R) P R; and (P Q) · (Q R) 0,

in if/. Therefore Theorem 3 is a consequence of
=
because Q R is
= - - -
- the following.
Theorem 4 (The Pythagorean theorem). In any inner-product space,
A B · = 0 => llAll2 + llBll2 = llA + Bll2•

Proof
llA + Bll2 = (A + B) · (A + B)
=A·A+2A·B+B·B
= llAIJ2 + 0 + llBll2•
As in the special case of R3, this immediately gives:
Theorem 5. The projection of a point Q into a finite-dimensional subspace if/ is the

point of if/ which is closest to P.
This idea has the following unexpected application. Let
"Y = C0[-TT, TT],

where C0[-TT, TT] is the set of all continuous functions on the closed interval [-TT, TT];
and consider the inner-product space
["Y, +, sm, · ],
where + and sm are defined as usual for spaces of functions, and the inner product is
defined by the formula
· g /(x)g(x) dx.
f = f:
For each positive integer n, let Tn be the set of all trigonometric polynomials of order
n, that is, the set of all functions of the form
n
g(x) = a0+ I [a; cos ix + b; sin ix].
i=l
Obviously Tn forms a subspace of"f/. Consider the set
{1; cos x, cos 2x,·. .. , cos nx; sin x, sin 2x, ... , sin nx}.
This set spans Tn. To verify that the set is orthogonal, we need to show that
f: cos ix dx = f: sin ix dx = 0 for every i,

" "
J: cos ix sinjx dx = 0 for every i, j,

"
and
J:11
sin ix sinjx dx =
,,
J:
cos ix cosjx dx = 0 for i ;;i!: j.
All of these answers can be calculated by brute force, but there are tricks that help.
By more straightforward calculations, we get
1/1112 = f,,12dx = 27T, \11\I = .J21T;
llcos ixll2 = 7T = \\sin ixl\2•

To get an orthonormal basis, we divide each basis element by its norm. This gives an
orthonormal basis of the form B = {C0, C1, Cn; S1, S2, , Sn}, where . . • , • • •
1
Co=
.J21T'
.
Ci 1 COS lX (i > 0),
.J;
=
1 .
S. - s m i .x (i > 0) .
.J;
=
'
Now let/ be a function in C0[-7T, 7T]. By Theorem 2, the projection off into the
finite-dimensional subspace "fl/" Tn is the vector
=
n n
P r nf = I (f Ci)Ci + I (f S,)S;. ·
i=O i=l
·
f" f(x) dx,

Here
f · C0 = J f(x)
_,,
"
· --=
1
.J21T
dx =
1
.J21T
·_
_,,
f· C; =
1 f" f(x)
1-
'\/ 7T -11
cos ix dx (i > 0),
f S;· = �7T- J_1T{" f(x)

'\/
sin ix dx.
Therefore
n
Pr nf = a0 + I [a; cos ix + b; sin ix],
i=l
where
1
a0 =
27T
-
f"" f(x) dx,

_,,
a; =
- f f(x) ix dx
1
7T -1T
cos (i > 0)
b; =
1 f" f(x)
-
7T -1T
sin ix dx.
546 Fourier Series 12.l
[Check that (j C0)C0 a0, (j Ci)Ci

· ai cos ix, and (f S;)S;
= · b; sin ix
= · =
(i ).
> O ]
It now seems reasonable to hope that Prnf is in some sense an approximation of/
when n is large. That is,
( ?) n � co � f � Prnf ( ?).
If we judge the approximation by observing II/ - Prnf II , it is clear, at least, that
we have done our best: Theorem 5 tells us that Prnf is the element of Tn which
minimizes II/ - Prn/11. If our best was good enough, then we should have
(?) lim I Prnfll (?). (1)

n-+oo I! -
= 0
A stronger conjecture is that the approximation is good even in the uniform norm:
lim - Prnf llu

(?) 0 (?). (2)
n-+oo llf
=
It may be disturbing, at this stage, to observe that (2) cannot be true as stated:
Prnf (-Tr) Prnf (Tr), because all trigonometric polynomials have period 2Tr.
=
Therefore (2) cannot be true unless/has the same property. We shall see, however,
that this is the only way that (2) can fail to hold:
Theorem A. If/ has period 2Tr, andf' is continuous, then limn�oo II/- Prnfll,. = 0.
This implies, of course, that
f (x) = !�°!, [a0 + i� (ai cos ix + bi sin x)J

00
= a0 + I (ai cos ix + bi sin ix),

i=l
where the limit is a pointwise limit in the elementary sense.
The last of these expressions is called the Fourier series for f, and the numbers
a0, a1, • b1, b2,
• • , are called the Fourier coefficients off Obviously every con
• • •
tinuous function has a set of Fourier coefficients, and therefore has a Fourier series.
The question is under what conditions we can conclude that the Fourier series con
verges to the function. This question is complicated, and the situation is not yet
thoroughly understood by anybody; Theorem A, above, is the best of the simple
results.
Meanwhile, the successive projections Pr1/, Pr2/, have two encouraging • • •
properties. First, each projection Prn+if is simply a continuation of the preceding one
Prnf; to get Prn+if from Prnf we merely add a term of the form
an+l cos (n + l)x + bn+l sin (n + l)x,
leaving the preceding terms unchanged. Second, the error in the approximation
Prnf � f, as measured by II/ - Prn/11, is nonincreasing.
Theorem 6. Iff is continuous on [-Tr, Tr], then
II/ - Prn+dll � II/ - Pr;Jll.
(Proof Since T,,+1 contains T,,, the minimum distance from/to T,, cannot be less
than the minimum distance from/ to T,,+i.)
Note that the Fourier series for a function f depends only on the values off on
the interval [-7T, 7T]. Therefore, when we set up the series forf(x) x, what we are
=
really dealing with is a discontinuous function, with period 27T, whose graph looks
like this:
y
Similarly, when we set up the series forf(x) x2, the series turns out to represent a
=
periodic function whose graph is obtained by fitting together infinitely many parabolic
arcs, like this:
y
__,'---"'""'-�....._'--""""'---'-�-'"4""-�L----"..£.�-'-��_,___.� x
-511" - 3,,. -.,,. .,,. 3,,. 5,,.
Of course the periodic function that we get from f(x) x is not continuous.
=
But it turns out that this doesn't matter: if the graph ofjis obtained by fitting together
a finite number of continuous functions with continuous derivatives, then the Fourier
series always converges to a function F; and F(x) =f(x) at every point where f is
continuous.
y
At points where the "continuous pieces" of the function fail to fit together, as at
x = 0 in the figure above, the series makes a compromise, and converges to the
average of the lefthand value and the righthand value. Similarly, ifj(-7T) =;C j(7T),
then
F(-7T) F(7T)
= ![f(-7T) + f(7r)].
=
Thus, for the function fin the preceding figure, the graph of the function F given by
the series looks like the figure below. Here
F(-rr) = F(O) = F(rr) = 0.
PROBLEM SET 12.1
Throughout this problem set, it should be understood that a0, av . . , b1, b2, are
. • • •
the Fourier coefficients of the function f, and that F is the function to which the series
converges. In each case, the graph offshould be sketched; and Fshould be sketched also,
in those cases in which Fis different from/
Compute the Fourier coefficients for each of the following functions.
1. f(x) = x 2. f(x) = x2 3. /(x) = x + x2
4. /(x) = x3 5. f(x) = x - x3/rr2 6. /(x) =!xi

7. /(x) = x on [O, rr], /(x) = 0 on [ -rr, O]
8. /(x) = -x on [ -rr, O], /(x) = 0 on [O, rr]
9. /(x) = x on [O, rr], /(x) = 1 on [-rr, O]
10. /(x) = 1 on [O, rr], f(x) = -1 on [-rr, O]
11. f(x) = 2x on [O, rr], /(x) =x on [-TT, 0]
12. For the odd functions x and x3, you found that the series used only sines, with ai 0 =
for each i. Show that this happens for every odd function (/is odd if/( -x) = -/(x)
for every x).
13. Similarly, show that if/is even (with/(- x) =/(x)), then the series uses cosines only.
14. Show that for each/in C0[-rr, TT], 11/112 � 2rr 11/11�.
15. In Theorem A,fand/' are continuous, and since
(1) JI/ - Pr,,/llu � 0,

we have
(2) U Jim Prnf =f on [ -rr, rr].
n-oo
(See Problem 10 of Problem Set 11.6) Can (2) hold iffis not continuous?
12.2 Uniform Approximations by Trigonometric Polynomials 549
16. Let f be as in Theorem A. Assuming that Theorem A is true, show that
Jim II/ - Pr,./11 = 0.
17. Working merely on the basis of the formulas for the Fourier coefficients of a continuous
functionf, give a geometric plausibility argument for the statement
Iima,. =Jim bn = 0.
n-oo n-co
(What does the graph of y = cos nx look like, when n is very large? How about the
graph of y =f(x) cos nx?)
18. In Theorem 6, under what simple conditions does the equality hold?
12.2 UNIFORM APPROXIMATIONS BY TRIGONOMETRIC POLYNOMIALS
The purpose of this section is to show that for every continuous function f, with
period 27T, and every E > 0, there is a trigonometric polynomial
n
</J(x) = a0 + I (a, cos ix + b; sin ix)

i=l
such that
II! - <Pllu < €.
That is,
lf(x) - </J(x)I < E for every x.
Here we are not claiming that the coefficients in <P are Fourier coefficients. In
fact, if we have a <P which makes !If - <Pllu < E, and we want to improve the approxi
'
mation, using an E < E, we cannot always do this merely by adding new terms to the
old t/J; we may have to start afresh, with new coefficients even in the first few terms.
f+•
f
¢
f-· x
The first clue to this situation is that the trigonometric polynomials form a bigger
system than one might think:
Theorem 1. The set of all trigonometric polynomials is closed under multiplication.
From this it follows immediately that sin2 x, cos2 x, . . . sinn x, cosn x are trig-
onometric polynomials. For example,
+ .l
1 + cos 2x _ 1
Cos2 x = cos 2x.
2
- 2
2
The general proof depends on the trigonometric identities

cos A cos B = t[cos (A + B) + cos (A - B)],
sin A sin B = -![cos (A + B) - cos (A - B)],
sin A cos B = t[sin (A + B) + sin (A - B)].
These are easy consequences of the addition formulas for the sine and cosine.
Consider now two trigonometric polynomials
n m
a0 = 1(a; cos ix + bi sin ix), A0+1(A1cosjx + B1sinjx).

i=l i=l
Every term of the product has one of the forms

a;A0 cos ix,
a;A1cos ix cosjx, a;B1cos ix sinjx, b;A0 sin ix,
b;A1sin ix cosjx, b;B1sin ix sinjx.
Each such term is a trigonometric polynomial. Therefore so also is their sum.
n
Consider now the function g(t) = cos2 t, which we now know to be a trigono
metric polynomial. If n is large, then cos2n t R:1 1 only when t R:1 O; everywhere else,
n
cos2 t R:i 0. Thus the graph looks something like the figure above, on the interval
[ -7T, 7T]. Let o be any number between 0 and TT, and let
In= L: cos2n t dt,
-0
i"0
n
Jn=
J -IT
cos2 t dt = cos2n t dt,
K ==
n
f 0
-o
n
·cos2 t dt.
We shall show that when n is large, In R:1 Kn. That is:
Theorem 2. For each O > 0, limn ...... ro Jn/In= 0.

This does not follow from the fact thatJn-+ 0, because In-+ 0 also. To prove the
theorem, we need to get good estimates ofJn and In. The first of these is easy:
Jn= l " n
cos2 t dt < (7T - o) cos2
n
0.
To estimate In, we use the reduction formula
n
f cosn x dx = � cosn-l x sin x + : 1 fcosn-2 x dx .
(See Problem 31 of Problem Set 6.5) The formula can be derived by integration by
parts. This gives a recursion formula for the definite integral:
i" n - li"
n
cosn x dx = -- cosn-2 x dx .
Using 2n for n, we get

_,, -11
i" 2n - li" 2n - 1
2n 2n
In = cos2n x dx = cos2<n-ll x dx =
In-l•
_,, _,,
Therefore
2n - 2n - 3 2n - 5 111
2n 2n - 2 2n - 4
1 1
I,, = ·
·
· · · - cos0 x dx
2 -11
3 (2n - 3)(2n - 1) . 17
2
2 4 6 (2n - 2)2n
1 · · 5 · · ·
=
3.-
5 .- . . . 2n - 1 . -
1 .
· · · · ·
- 27T > - .
2 4 6 2n - 2 2n n
7 'TT
=
Therefore
('TT - ()) cos2" ()
- < -- n cos2n u.
Jn 'TT - ()
TT/n
Ji
= ·
In
Since 0 < cos c5 < 1, it follows that Jn/In--+- 0. [In fact, L!1 (J;/I;) converges;
'TT
the easiest way to see this is to recall that L ki(x2)i converges for 0 < x < l.]
Now let/be any continuous function with period 2TT, and for each n let
J'!..,,f(x + t) cos2" t dt
..,, ..(x)
,1.
f'!..,. cos2n t dt
.
=
It is plausible to suppose that
n oo =:;> c/>,,(x) f(x).

Roughly speaking, the reason is that for each c5 > 0,
� �
1
n
In i{J
� c/>,,(x) :::::: - f(x + t) cos2" t dt;
oo =:;>
if c5 :::::: 0, then/(x + t) � f(x) for -c5 � t � c5, and so

-!J
1
n
i{J
:::::: oo =:;> cf>n(x) :::::: - f(x) cos2n t dt
In -!J
1
:::::: - f(x)I n = f(x).
In
As we shall see, these ideas can be built up into a proof that
lim II/ - </>nllu = 0.

n-+OO
First, however, we need the following:
Theorem 3. For each n, </>n is a trigonometric polynomial.
Proof First we observe that
</>n(x) =
1
-
f1l f(x t) + cos2n t dt
Jn -1!
J -tr
1
cos2n (t -
=
- f" f(t) x) dt.
n
(The integrand has period 27T, and so f.'.'." is unchanged if we slide the graph back and
forth horizontally.) We know that cos2n tis a trigonometric polynomial; say,
m
cos2n t = a0 + !
i=l
(a; it
cos + bi sin it).
Therefore
m
cos2n (t - x) a0 = + ! (ai cos

i=l
it cos ix + ai it
sin sin ix
Therefore
it ix
+ b; sin cos - b; cos it ix).
sin
J:/(t) cos2n (t - x) dt J:"a0f(t) dt

=
i� [f:}a; it
+ cos + b; sin it)f(t) dt] ix cos
;� [f'"(a + i sin it - b; cos it)f(t) dt] ix. sin
The coefficients here are complicated, but they are constants, and so the indicated
integral is a trigonometric polynomial.
Theorem 4. Iff is continuous, and has period 27T, and
.J..
�nX - () _
f':_" f (x t)+ cos2n t dt
then
f7:_1l cos2n t dt '
lim II/ - </>nllu = 0.

n-.oo
Proof Given E > 0, we need an n such that
n � N=;> II!- </>nllu < E.
By definition of the uniform norm, this means that
n � N -=;> lf(x) - </>n(x)I < E for -7T � x� 7T.

Step 1. Since/is continuous on [-TT, TT],fis bounded on [-TT, TT]. Also/is periodic.
Therefore there is an Msuch that
If(t)I < M for every t. (1)
Step 2. Since f is continuous on [-2TT, 2TT], it follows that f is uniformly continuous
on [-2TT, 2TT]. And f is periodic. Therefore there is a (J > 0 such that
This means that

Ix -x'I lf(x) (x')I
 -f < E/2.
ltl /j(x) f(x

 t )/ - + < E/2. (2)
We are now ready to calculate. For Mas in Step 1, and c5 as in Step 2, we have
lf(x)- ef>nCx)I \ f(x) s�,,fJ�,, (x t) t dt + cos2n

= -
t dt I cos2n
L JI(x)J: t dt -J:/(x t) t dt I
=
7
cos2n + cos2n
;n If,, [f(x)-f(x t)] t dt /

= + cos2n
� - J " lf(x) -f(x

1 t)/ t dt + cos2n
I _,,
[J_,,-/j/f(x)-f(x t)lcos2ntdt+ J_)f(x)-f(x

n
=
I
1
n
/j + + t)/cos2ntdt
f/f(x) - f(x t)/ t dt] + +

,
cos2n
� 1. [ f - t dt f-o t dt Jo( t dt] 0 "

1M 0 cos2n +� cos2n + 2M cos2n
I -tr 2
n
_!_ [
=
I
4MJ + �
n 2
f-o t dt] 0
cos2n
n
J
< 4M · n+ l . .=J" tdt cos2n
I I 2 -1I
n n
J €
= 4M- n+-.
In 2
Therefore there is an N such that
J
n � N => 4M n < .: .
I 2
We now have n

n � N => I !- ef>nllu < E,
This theorem has an immediate consequence for Fourier series.

Theorem 5. Iffis continuous, and has period 27T, and

n
Prnf = a0 + 2 (a, cos ix + b, sin ix)
i=l
is the nth partial sum of the Fourier series off, then
limllf - Prn/11 = 0.
Proof Let E > 0 be given, and let

N
cp(x) = A0 + 2 (A; cos ix + B, sin ix)
i=l
be a trigonometric polynomial of order N. Then
II! - PrN fll � II! - <foll ,

because PrNJ is the element of T N which is closest to f And
II! - <'1112 � 2 7T II! - <foll!.

(See Problem 14 of Problem Set 12.1) Therefore
We take <P so that
27T II! - <Pl

l: < €2•
(This will hold whenever II/ - <Pl

lu < E/.,/277.) It follows that
II! - PrNfll < €,
Since the norm of this difference is nonincreasing (Theorem 6 of Section 12.1), it

follows that lJ - Prnfll < E for every n � N. Therefore limn-.co II/ - Prnf 11 0. =
Finally, we observe that for each n there is a point xn at which the error in the
proposed approximationf R:> Prnf is actually equal to 0.
Theorem 6. If f is continuous, then for each n there is an xn such that f (xn) =
Prnf(xn).
Proof By definition of the projection, f - Prnf must be orthogonal to every vector

in Tn. In particular, the inner product (f - Prnf) 1 must be 0, because 1 is in Tn.
·
Therefore
J:"[f(x) - Prnf(x)] • 1 dx = 0.
Here the integrand must vanish at some point xn.

12.2 Uniform Approximations by Trigonometric Polynomials SSS
At first, this theorem may seem almost like a joke, but it isn't. See the following
section.
PROBLEM SET 12.2
Theorem 1 implies that each of the following functions is a trigonometric polynomial.

Compute these functions in the form of trigonometric. polynomials.
1. sin3 x 2. sin2 x 3. cos3 x
4. cos2 (2x) 5. sin2 x cos x 6. cos2 x sin x
7. cos x sin 2x 8. cos4 x 9. sin4 x
10. sin2 x cos2 x

1 1. Suppose that in Theorem 3 we had used
s�1T f(x + t) cosn t d t

</>n(x)
s�1T cosn t dt
=
Would Theorem 3 still have been true? (Either prove the theorem in the more general
form, with odd exponents allowed, or give an example to show that the more general
theorem is false.)
12. Show, by any method, that
[f/ r <x) dx � 21T f} f(x)]2 dx.
[Hint: There is a quick method, on the basis of what you know now.]
13. Show that if/is as in Theorem 5, then
:� fJ f(x) - Prn/(x)I dx = 0.
14. Show that if/is as in Theorem 5, then
:�n: f,,Prn/(x) dx = f/ (x) dx.
15. Now show that the same result holds on every interval [a, b].
* 16. In Section 10.8 we proved the binomial theorem
(1 + xr = � C)
i
x
i
by the methods of calculus, in the real domain. Thus the proof in Section 10.8 does not
·show that
for every complex number z. Prove the latter theorem, by induction.

* 17. Now use the result of Problem 16 to get an explicit formula for cos4 x in the form of a
trigonometric polynomial, with coefficients given numerically.
*18. Now get a general formula for cosn x as a trigonometric polynomial.

[Note that in the text, we did not need the full force of Theorem 1; all we needed
was the result of Problem 18; and the proof of the special result is neater. But Problem
18 is misleadingly special; and Theorem 1 ought to be regarded as the "real reason"
why cos2n x is a trigonometric polynomial.]
19. In the proof of Theorem 3, we used the function
f'!...ir f(x + t) cos2n t dt f'!...ir f(t) cos2n (t - x) dt

</>,,.(x)
f'!...ir cos2"' t dt f'!...ir cos2n t dt
= =
Let f(x) = x2, on the interval [-1T, 1T]; and extend the graph so as to get a function
of period 21T. (See the figure on p. 547.) For this function f, compute the function
<f> (x) in the form of a trigonometric polynomial, using definite integrals as coefficients.
2
You need not compute the integrals numerically.
20. Given a trigonometric polynomial
n
<f>(x) = a0 + L (a; cos ix + b; sin x).

i=l
Is it always true that 0 there is a poly
nomialp(x) L�=o a; x; such that
=
lf(x) -
p(x)I < E (0 � x� 1).
(This is a celebrated theorem due to Karl Weierstrass.)

*22. Given that the theorem proposed in Problem 21 is true, show that if f is continuous
on [a, b], and e > 0, then there is a polynomialp(x) such that
lf(x) - p(x)I < E for a� x � b.
12.3 INTEGRATION OF FOURIER SERIES.

THE UNIFORM CONVERGENCE THEOREM
We defined the Fourier series

00
a0 + !(a; cos ix + b; sin ix)

i=l
by using the projections
n
Prnf = a0 +!(a; cos ix + b; sin ix)
i=l
off into the subspace Tn of trigonometric polynomials of order n. We now want to
show that iff has period 2?T, and f' is continuous, then:
1) the Fourier series off converges;
2) the sum of the Fourier series is the function/;
3) the series can be integrated a term at a time.
12.3 Integration of Fourier Series. The Uniform Convergence Theorem 557
The ideas suggested by (3) are the key to the situation: to prove convergence, we
first need to find out how the operations Prn are related to differentiation and
integration.
Theorem 1. Iffhas period 27T, andf' is continuous, then
Prnf' = (Prnf)'.
That is, the projection of the derivative is the derivative of the projection.
Proof Let
n
Prnf= a0 + L (ai cosix + bi sinix),
i-=1
n
Prn f'= A0
iL
�l
(Ai cosix + Bi sinix).
+
We need to show thatA0 = 0, Ai= ibi, Bi= iAi . The Fourier coefficients forfare
-
given by the formulas
f" fx( ) dx,

1
ao =
27T -1T
f" fx( ) cosix dx

-
1
ai = (i > 0),
1T -1T
-
bi = .!. [" fx
( ) sinix dx.
J_1T
1T
Similarly,
f"f'(x) dx,
1
A0 =
27T
-
-1T
Ai = .!. {" f'(x) cosix dx i( > 0),

1TJ_1T
B,. = .!. {" f'(x) sinix dx.
1T J_1T
By the fundamental theorem of integral calculus,
Ao = _L [f7r ( 7r )] =
( )- f- 0.
27T
In �i fori > 0, we integrate by parts, using

u = cosix, dv = f'x( ) dx, du = -i sinix, v = f(x).
This gives
A,. = .!. {"if(x) sinix dx = 0 + ibi = ibi.

.!. [f(x) cosix]'.'.." +
1T J_1T 1T
Similarly, using u = sinix, dv = f'(x) dx, du = i cosix dx, v = f(x), we get

" if(x) cosix dx=
J
1 1
Bi = - [fx
( ) sinix]:'.." - - 0 - iai = -iai.
1T' 1T' -ir
Note that Theorem 1, as it stands, does not tell us that
D[ i�a
0
+ (ai cos ix + bi sin ix) ] i�
= (-iai sin ix + ibi cos ix).
In fact, at this stage, we don't know that either of the,indicated'series converges to any
function at all.
We now propose to find out how Prn is related to integration. Theorem 5 of
Section 12.2 says that
Jim II! - Prnf 11
n-+CO
= o,
which means that
What we want is
J: lf(x) - Prn f(x)I dx 0.

,,
�
For this purpose we need the following:
Theorem 2. If g is continuous on [a, b], then
b
[f g(x) dx r� (b - a) i [g(x)]2 dx.
Proof Let C0[a, b] be the set of all continuous functions on [a, b]. Under the usual
definitions of+, sm, and·, C0[a, b] forms an inner-product space, and so the Schwarz
inequality holds. In the inequality
(A B)2 � llAll2
· • llB ll2,
we take A = g, B = I. This gives
b
[f g(x) .
1 dx r � [f g(x) . g(x) dx ] [i (l . 1) dx J
= ( b - a) f [g(x)]2 dx.
This tells us that the integral of a function is small if the norm of the function is
small. Applying this principle to the function If - Prn/I, we obtain the following
theorem.
Theorem 3. Let f be a continuous function, of period 27T, and let
Then
lim Mn 0.
'7t-+CO
=
Proof
M�= [f,,1 1cx) - Prnf(x)I axJ 2

� 27TJ_') J(x) - Prnf(x)}2 dx
= 2 7T II! - Pr nfll2•
By Theorem 5 of Section 12.2, the last of these expressions approaches 0. Therefore
M!-O, and Mn- 0.
We are now ready to prove a convergence theorem. If/is as in Theorem 3, and/'
is continuous, then f satisfies the conditions of Theorem 3. Let
Mn= J:,)f'(x) - Prnf (' x)I dx.

Then
limMn= 0.
n->OO
Let xn and x be any points of [-7T, 77]. Since Prnf' = (Prnf)' , we have
L:[f'(t) - Prnf'(t)] dt= [f(t) - Prnf(t)]�

=f(x) - Prnf(x) - [f(xn) - Prnf(xn)].
By Theorem 6 of Section 12.2, we can choose each Xn so that f(xn)= Pr1.f(xn).
This gives
lf(x) - Prnf(x)I= IJ� [f'(t) - Prnf'(t)] dt � I f� lf'(t) - Prnf'(t)I dt � Mn-
Therefore II/ - Prnflu � Mn. This gives:

Theorem 4. Iff has period 27T, and f' is continuous, then
l J - Prnfllu= 0 .
lim l
In the language of Section 10.15, this takes the form:

TbeQl'em 4'. If/has period 27T, and/' is continuous, then
Ulim Prnf =f on(-00,00).
In each case, the reason is that the differences/ (x) - Prnf (x) are squeezed to 0
by a sequence of positive constants. Finally, all this can be restated in terms of the
formula
co
Prnf(x)= a0 + I (ai cos ix + bi sin ix),

i=l
and the formulas for the Fourier coefficients ai and bi. This gives a third form of the
theorem:
Theorem 4". Let f be a function with period 27T and a continuous derivative. Let
a0
Then
=-
1 J"f(x) dx, ai = -1 f"f(x)
27T _,, 'TT -1T
cos ix dx, bi = -
1 f"f(x)
'TT -1T
sin ix dx.
1;!"�� [a0+ i� ( i ix+ bi

a cos sin ix) ] =f(x) on ( -co, oo) .
Just as we found for power series, in Chapter 10, uniform convergence enables us
to integrate a term at a time. In general:
Theorem 5. If the functions fn are continuous, and U limfn

n-+ro
a
=f on [ , b], then
b b
rfn(x) dx = rf(x) dx.
n-+oo Ja
lim
Ja
This gives:
Theorem 6. Iffhas period 27T andf' is continuous , then the Fourier series forf can
be integrated a term at a time; on any interval.
We found, at the beginning of this section, that P r nf' = (Pr nf) ' . If
f' (x) = lim P r nf' ,
it follows that the series for f can be differentiated a term at a time. But we have
proved Theorem 4 only for functions with continuous derivatives; and so our con•
vergence theorem applies tof' only whenf"is continuous. Hence the heavy hypotheses
in the following theorem:
Theorem 7. Iffhas a period 27T andf' andf" are continuous, then the Fourier series
for f can be differentiated a term at a time.
That is, forf(x) = limn..... oo P r nf, we have

f'(x) = lim Prnf'(x) = lim [Prnf(x)]'.
n-+ro n-+ro
(We get the first of these equations from Theorem 4, and the second from Theorem 1.)
In the further development of the theory, in which we allow discontinuous
functions, the status of Theorem 7 is very different from that of Theorems 4 and 6,
in that the latter two theorems hav� quite satisfactory generalizations, but Theorem 7
does not.
y
Suppose that we form a function of period 27T by fitting together the graphs of a
finite set of continuously differentiable functions, end to end. We reconcile the values
at common endpoints by defining
f (x1) = Hf1(xJ + f2(X1)],

and so on; and we reconcile the values at -?T and 7T by defining
f(7T) = f(-7T) = Hf1(-7T) + fa(7T)].

A function obtained in this way will be called a function of the Fourier type. Evidently
functions of this type are integrable, and so every such f has a Fourier series, with
coefficients given by the same integral formulas as for continuous functions. The
following turns out to be true:
Theorem B. Let f be a function of the Fourier type, with Fourier series

<X)
I (x) = a0 + I (ai cos ix + bi sin ix).

i=l
Then
I) For every x, L (x) converges to f ( x).
2) The convergence is uniform on every closed interval on which f is continuous.
3) The series can be integrated a term at a time on any closed interval (even if the
interval contains points of discontinuity).
4) lim II! - Prnf 11 = O.

n->oo
But it is not necessarily true that the series for a function of the Fourier type can
be differentiated a term at a time, even on an interval on which f, f', and f" are
continuous. This will be brought out in the problem set below. In working on these
problems, you should regard Theorem B as given.
PROBLEM SET 12.3
1. Show that iffis a function of the Fourier type , then
[f/ (x)dx J 2 � 27T f,, ] dx.

[/(x)2
(A simple proof is possible, if you use the right ideas.)

2. ·Let/be the function defined by the conditions
(1) f(x) =1 on (0, 7T), (2) /(0) =/(7T) =0,
(3) /(x) = -1 on (-7T, 0), (4) /(x + 2n1T) = /(x).
Calculate the series for/, and discuss the series obtained by termwise differentiation.
3. Let/be the function defined by the conditions
(1) /(x) =ex on (-7T, 7T),

(2) /(-7T) =/(7T) =t(e-" + e"),
(3) f(x + 2n7T) =/(x), for every x.
f
Evidently has a Fourier series
a0 + L (a; cos ix + b; sin ix).

00
i=l
Show that termwise differentiation of the series forf cannot give the Fourier series for/'.
4. Now calculate the Fourier series off (You did not need to do this, to solve Problem 3.)
5. Now verify, by a calculation, that for this series,
Jo('" [ao + i a; cos it + b; sin it] dt =Jor '"ao dt + i Jo("'(a; cos it + b; sin it) dt.
i=l i=l
Problems 6 through 8. Let / be the function defined by the conditions

(1) /(x) = e"' on (0, 7T), (2) /(0) = 0, /(7T) = ! (e" - e-"),
(3) /(x) = -e'" on ( -7T, 0), (4) f(x + 2n7T) f(x),
= for every x.
Proceed as in Problems 3 through 5.
*9. Let f and g be functions of the Fourier type, and let L (x) be the series for f, as in the
text. Show that
f f(x)g(x) dx J�"a0g(x) dx + �1 J:" [a; cos ix + b; sin ix]g(x) dx.

.. =
*10. Show that if / is as in Problem 1, then
=Ia�+ Ib�.
00 00
111112
i=O i=l
Linear Transformations,
13 Matrices, and Determinants
13.1 LINEAR TRANSFORMATIONS
Let Rn and Rm be Cartesian spaces of any dimension. A function

j; Rn-'>- Rm
is called a linear function, or a linear transformation, if it preserves sums and scalar

products. That is,f is linear if
f(P + Q) = f(P) + f(Q) for every P, Q, (1)
and
f(rxP) = rxf(P) for every rx, P. (2)
If f is linear, then
f(rxP + {JQ) = f(rxP) + f({JQ) = rxf(P) + {Jf(Q); (3)
and conversely, if (3) holds, then (1) and (2) both hold.
To see how linear transformations work, let us examine some special cases in
low dimensions. Suppose that
f: Ra_,,.. Ra
is linear. In R3 we use the "standard basis"
where
E1 = (1, 0, 0), E2 = (0, 1, 0), Ea = (0, 0, 1).
Now f(E1) must be a vector in R3; and B3 is a basis for R3• Therefore we have
f(E1) = a11E1 + a21E2 + aa1Ea,
for some set of scalars a11, a21, aa1. [Here we are using double subscripts; ai1 is the
coefficient of Ei inthe expression forf(E1) .] Similarly,f(E2) and/(£3) have the forms
f(E2) = a12E1 + a22E2 + aa2Ea, f(Ea) = a13E1 + a2aE2 + aaaEa.
Iff(E1),f(E2), andf(E3) are known, then this determinesf(P) for every P in Ra. The
reason is that for
we must have
563
564 Linear Transformations, Matrices, and Determinants 13.1
This enables us to write a formula forf(P) in terms ofthe numbers ai1:
f(x1E1 + X2E2 + xaEa) = a11X1E1 + a21X1E2 + aa1X1Ea
+ a12X2E1 + a22X2E2 + aa2X2Ea
+ a13X3E1 + a23X3E2 + aa3X3£3.
Collecting the coefficients of E1, E2, and E3, we get:
(a11X1 + a12X2 + a13X3)E1
+ (a21X1 + a22X2 + a23X3)E2
+ (aa1X1 + aa2X2 + aaaXa)Ea
Y1E1 + Y 2E2 + YaEa.
There is a special apparatus which enables us to write such formulas more

compactly, and deal with them more conveniently. Evidently/is completely described
by the 3 · 3 = 9 numbers ai1. We write these in a rectangular array, like this:
Such an array M is called a 3 by 3 matrix. (In general, an m by n matrix is a rectan

gular array of numbers, with m rows and n columns.)
3
At the beginning of Section 11.3, before we introduced a basis in R , we repre
sented the vector P = x1E1 + x2E2 + x3E3 by the ordered triplet (x1 , x2, x3). We
now write this triplet vertically instead of horizontally, in the form of a "3 by 1
matrix"
When Pis described in this notation, we call Pa column vector. Similarly, for
f(P) =Y 1E1 + Y2E2 + YaEa = (y1, Y2• Ya) ,

we write
f(P) = [��] [:�� ! :�:�: ! :�:�:] .

Ya
=
aa1X1 + aa2X2 + aaaXa
Note that the array on the right i; a column vector; once the indicated additions are
performed, there is only one entry in each row.
Now we define the operation of multiplication, of the column vector P by the
matrix M, in such a way thatf(P) =MP. That is:
Definition
13.1 Linear Transformations 565
Under this definition of the "product" MP, we have
MP � [f:] � f (P)
There is a usable pattern in this multiplication: to get the entry y1 in the first row of
the product, we regard the first row of M as a vector, and form its inner product
with the column vector P. Similarly for the other rows.
Let us examine an example. The matrix
[ 1 2
-2
1
1
-1 �]
f(xi. X2, Xa) =
[
describes a linear transformation f, with
1
-2
2
1
1] [X1]
2 X2 =
[ X1 + 2x2 + X3]
-2X1 + X2 + 2x3 =
[Y1]
Y2 .
1 -1 2 Xa Xi - x2 + 2x3 Ya
Two questions arise naturally here.
Problem 1. Given a particular point (Ji, y2, Ya), for what points P, if any, do we have 1
f(P) = (Yi. Y2, Ya)? For example, for what points P is it true that
f(P) (1, 2, 3)?
{
=
To answer this question, we need to solve the system
Xi + 2X2 + X3 = 1, (1)
-2xi + x2 + 2x3 = 2, (2)
Xi - x2 + 2xa = 3 (3)
of linear equations in the unknowns Xi. x2, x3• Almost any method will do. We shall
use a method which will be of theoretical importance later.
Step 1. Eliminate Xi from (2) and (3), by adding twice (1) to (2) and subtracting (1)
from (3). This gives a new system which is equivalent to the original system, in the
sense that it has exactly the same solutions:
Xi + 2X2 + Xa = 1, (1)
5x2 + 4X3 = 4, (2') = (2) + 2(1)
- 3 x2 + Xa = 2. (3') = (2) - (1)
(The notations on the right indicate where the new equations come from.)
Step 2. Eliminate x2 from (3'), by adding t of (2') to (3'). This gives the equivalent
system
Xi + 2X2 + Xa = 1, (1)
5x2 + 4xa = 4, (2')

(3") = (3') + !(2')
We now say that the system is in triangular form. In general, a system of n

linear equations in n unknowns is in triangular form if the nth equation involves
only xn, the (n
-l)st equation involves only Xn-l and Xn, and so on. (This means
that in the matrix of coefficients, a;1 = 0 for j < i.)
Step 3. Solve, by successive substitutions, working from bottom to top, getting
1] [ ] [1]
X = H.
3
1
This means that
13 7
H -1 ; -r� ;
2
.
This should be checked.
Problem 2. Find, if possible, a set of formulas expressing P = (xv x2, Xa) in terms of
(Y1. Y2, Ya ).
To do this, we need to get a "general solution" of the equation
-11
2
in which the x/s are expressed in terms of the y /s. This is only slightly more trouble
some than Problem 1; we treat (y1, y2, Ya) in exactly the same way as we treated
(1, 2, 3) in Problem 1. The solution is
X1 = 147Yl - 151Y2 + 137Ya·
X2 = 161Y1 + l\Y2 - 14-iYa,
']
Xa = l\Y1 + 131Y2 + 151Ya·
The coefficients in these equations give us a new matrix
[':
- rr T7
M-1 = 17
1
1
T7 :
-1 7 .
T7
3
T7 17
We call this matrix M-1 because it is the matrix of the transformation

1-1: Ra� Ra.
That is, M-1 reverses the action of M, in the sense that
,
M-1(MP) = P, for every P = (x1, x2, Xa) .
It may easily happen, however, that a linear transformation
/: R"�Rm
does not have an inverse

(?) 1-1: Rm� Rn. (?)
There are two things that may go wrong: (1) some points Q of Rm may not be values
(2)
1
of the function at all. In such cases, there is no such thing as J- (Q). Some points
Q of Rm may be equal to f(P) for more than one point P. In such cases, f is not
[1 1
invertible. An example of both these phenomena is furnished by the matrix
J
0
M= 0 0 .
0 0 0
Here
of the function, andf(P) =

This function projects R3 into the xy-plane; no point outside the xy-plane is a value
f(P') whenever P and P' lie on the same vertical line.
To find out how a given linear transformation behaves, we try to compute its
inverse, by the method used in the above example. If there is an inverse, the method
gives it to us; and if there isn't, the method still tells us what is going on. Consider,
[l21 25 38.7]
for example, the transformation f described by the matrix
M=
z) (a,
4
To simplify the notation, we-use (x, y, for (x1, x2, x3) and b, c) for (y1, J2, Ya)·
We then have
MP = [; ; 7�][;z] [2� !+ ;� !+ ��i7z [�]·

1 4
=
x 4y
=
c
This gives the linear system

+ 2y + 3z =a,
2x + + 8z =
x
+ + 7z
Sy b,
x 4y = c.
+ 2y 3z
Reducing this to triangular form, we get the equivalent system
+ 2z b - 2a, (2)
x + =a, (l)
2b + (3)
y =
= 0 3a - c.
Thus the equation
3a - 2b +
f(P) = MP = (a, b, c)
{(a, 2b +
cannot have a solution unless c = 0. Let
E= b, c) I 3a - c = O}.
Then Eis a plane, and every point of Eis = f(P) for some P. The reason is that as
568 Linear Transform;itions, Matrices, and Determinants 13.1
long as (3) is satisfied, we can solve (2) and (1) in the forms
y = -2z +b - 2a, (2')
x = -2y - 3z +a
= z - 2b +Sa. (1')
Here we can choose any z; if
P = (z - 2b +Sa, -2z +b - 2a, z),
then
f(P) = (a, b, c) .
As in Section 3.1, the image of a function is defined to be the set of all values of
the function. If f is a function
A � B, then the image is denoted by f(A). More
generally, if A' is any subset of A, then
f(A') = {bI b = f(a') for some a' in A'}.

Thus, in the example above, the image is
f(R3) = E= {(a, b, c) I 3a - 2b +c = O}.

The kernel of a linear transformationfis
Kerf= {PI f(P) = O}.
Obviously the kernel contains 0, no matter whatf may be, because
f(O) = 0.
(Proof f(O) = f(P - P) = f(P) - f(P) = 0.)
If f is not one-to-one, however, then the kernel Ker f contains vectors other than 0
(Proof If f(P) = f(Q), for some P ¥: Q, then f(P - Q) = f(P) -f(Q) = 0,
and so P - Q belongs to Ker f)
In the above example, the kernel is the solution set of the equation
To get the solution, we set a = b = 0in equations (2') and (l'). Therefore Pis in
the kernel if P has the form
.P = (z, -2z, z).
Using (J.. for z (to fit the usual notation of scalar multiplication), we get
Kerf= {(J..V0}, V0 = ( l , -2, 1).
[
Therefore Kerf is a line. In other cases, the image may be an even smaller set, and the
kernel even larger. For
1 0O
M= 0 0 0,
000 J
the image is the x-axis, because
rn � �J[�J m
The kernel is the yz-plane, because the equation
�
0, 0 0, 0 0,
gives the system
x = = =
whose solution set is obvious.

Finally, a few remarks on questions which the above discussion may suggest.
R3.
1) As far as the ideas in this section are concerned, there is nothing special about
linear transformations Rn_,. Rn. We have discussed R3 merely to avoid tedious

Exactly the same methods apply, in exactly the same ways, when we deal with
notation, large matrices, and pointlessly long computations.
2) In the examples that we have discussed, the image and kernel of a linear trans
is easy, on the basis of Theorem 4 of Section

formation have turned out to be subspaces. This is what always happens. (The proof
11.3. You will find it hardly more
trouble to write it yourself than to consult the following section.)
3) If you are acquainted with determinants, and with the process of solving linear
systems by Cramer's rule, then you may suspect that the method used above, by which
we convert the system to a triangular system, is naive or inefficient or both. But this
is not true. In computation, triangularization is about as good a method as any.
PROBLEM SET 13.1
In each of the following problems, you are given a matrix M, describing a linear transfor
mation f Iff turns out to be invertible, compute 1-1. If not, find the image and the kernel.
Thus the answer to each of the first ten problems below should be in one of the forms
(a) M-1 = [ · · · ]
or
(b) /(R3) ={(a, b, c) I ·}, and Ker/= {(x,y, z) I···}.
[[ !]
· ·
, [H �] ]
I
I
.[H �] 2
. [�HJ 5
[: ; ] -3
-
+2
1
[-� -� -; 6.
4 2 -2
[h i] [! g �]
_ _
7 [H :J · [H i] 9
10.
ll. Suppose that the matrix of/has the form
[
a, ll 2
M=
"•]
aa1 Olll 2 Olll3
f3a1 f3a2 flaa
What sort of subspace is f(R3)? How about Ker/?

•
12. Same question, for matrices of the form

•u
[T ]
ll12
ll22 ll23 ,
0 ll33
["" g l
where a11a22a33 � 0.
0
M= a21 ll22
ll31 ll32 ll33
with a11a22a33 � 0.
[g
]
ll12 "
u
M� 0 aa
J ,
ll31 0
with a12a23a31 � 0.
]
ll12
rn
"
u
M� 0 aa
0
� •
13.2 COMPOSITION OF LINEAR TRANSFORMATIONS

AND MULTIPLICATION OF MATRICES
The preceding section was devoted almost entirely to investigation of examples of

linear transformations. We shall now develop some of the theory.
Theorem 1. Let J be a linear transformation Rm--+Rn. Then the image f(Rm) is a

subspace ofRn, and the kernel Ker f is a subspace ofRrn.
Proof By Theorem 4 of Section 11.3, we merely need to show that these sets are
closed under addition and scalar nrnltiplication. For eachf(P),f(Q), we have
f(P) + f(Q) = fP
( + Q);
and for eachfP ( ),ocfP
( ) = f(ocP). Thereforef(Rm) is a subspace. IfP andQ belong
to Ker f, then
f(P) = f(Q) = 0,
and sof(P + Q) = f(P) + f(Q) = 0 + 0 = 0, and

foc
( P) = ocf(P) = oc • 0 = 0.
Therefore Ker f is a subspace.

13.2 Composition of Linear Transformations and Multiplication of Matrices 571
Theorem 2. If f and g are linear transformations Rm-+ Rn, then so also are f + g
andrx.f
Here the sum and scalar product are defined by the obvious conditions
(f + g)(P) = f(P) + g(P),

(rx.f)(P) = rx.(f(P)).
The verification of linearity is trivial:
(f + g)(P + Q) = f(P + Q) + g(P + Q)

=
f (P) + f(Q) + g(P) + g(Q)
= (f + g)(P) + (f + g)(Q);
(f + g)(rx.P) =
f(rx.P) + g(rx.P)
=
rx.f(P) + rx.g(P) = rx.(f(P) + g(P))
= rx.(f + g)(P).
This gives:
Theorem 3. For each m and n, the linear transformations f: Rm -+ Rn form a vector

space.
Theorem 4. If g: Rm-+ Rn and f: Rn-+ RP are linear, then the composite function
f(g): R"'-+ RP
is also linear:
f(g)
The verification is straightforward:
f[g(P + Q)] = f [g(P) + g(Q)] =

f [g(P)] + f[g(Q)]
=
f(g)(P) + f(g)(Q );
f[g(rx.P)] =
f[ixg(P)] = rx.f [g(P)].
We fol!nd, in the preceding section, that the action of a linear transformation
R3 -+ R3 could be described by a 3 by 3 matrix. Similarly, the action of a linear
transformation g: R"'-+ Rn can be described by an n by m matrix. The scheme is as
follows. If
then
Therefore, for P = (x1, x2, • • • , Xm) in Rm, we have
g(P) =
Adding by columns, to get the total coefficient of each E; on the right, we get
Describing P and f(P) as column vectors, and representing g by the matrix with a; 1 in
the ith row and jth column, we get
bn b12 b1m X1 Y1
b21 b22 b2m X2 Y2
=
b,.1 b,.2 . . .
bnm Xm Yn
which has the form

X1 Y1
X2 Y2
Mg =
Xm Yn
Here the pattern of the operation is the same as in the case m n 3: to get y; in= =
the column vector on the right, we regard the ith row of the matrix Mg as a vector, and
form its inner product with the column vector.
Now if g and fare as in the preceding theorem, then each of the transformations
g,f, andf(g) can be described by a matrix. Let these matrices be
M0 = [b;1] (n by m),
M1 = [a1k] (p by n),
M1<g> = [cik1 (p by m).
Here [a,:;] is a shorthand for the matrix with the number a;; in the ith row and the
jth column, and similarly for [b1k] and [cik]. We define the product of two matrices,
in this case, to be the matrix of the composite function. That is,
Definition. Given linear transformations

g: Rm-+ Rn,
J: Rn-+ RP,
J(g): Rm-+ RP,
with associated matrices M1, Mg, Mt<a>· By definition,
M1Ma = Mt<g>·
We shall now get a formula for the product M1M11 of two matrices. The general
formula looks complicated, but its pattern is easy to see by an examination of the
case m
= n = p = 2. Let the matrices of the transformationsf, g, andf(g) be
M, = [an
a21
a12
a22
J
, Mg= [ bn
h21
h12
h22 'J Mt<g> = [en J
C21
C
12 .
C22
[:�] [ J[ J [ J [�:].
Then
= bu b12 X1 = bnX1 + h12X2
Mu =
b21 b22 X2 b21X1 + b22X2
{:�J [J [J
and so
Mt<a = M M X1 = M
Yi
g
f X2 f Y2
[ J[ ] [
J
= an a12 Yi = a11Yi + a12Y2
a21 a22 Y2 a21Yi + a22Y2
= [au(b11X1 + b12xJ + a12(b21X1 + h22X2)
a21(b11X1 + b12X2) + a22(b21X1 + h22X2)
J
= [ (aubn + ai2h21)X1 + (a11b12 + ai2h22)X2
(a21b11 + a22b21)X1 + (a21b12 + az2h22)X2 J
[ aubu + a12h21
a21b11 + a22b21
auh12 + a12b22
a21b12 + a22b22 J[ JX1
X2 •
Therefore M1Mg is the 2 by 2 matrix in the last formula. The pattern of the operation
is clear: to get the number C;;, in the ith row andjth column of the product, we regard
the ith row of M1 and the jth column of M a as vectors, and compute their inner
product. That is,
C;; = a; 1b1; + a;2b2;·
This is called the row by column rule of matrix multiplication. The same rule applies
in the general case, and the only problem is that the formulas are complicated to write
down. We have
X1 bu h12 b lm X1 _L;'!,1 bi; X; Y1

X2 h21 b22 b2m Xz L7!.1 b2;X; Y2
= =
Mu -
Xm bn2 b nm x _L;'!,1 bn;X; Yn

bnl
. . .
Here M0(P) = (y1, y2, • • •

, Yn), where
Yi 2: bi;X; (i = 1, 2, .. , n).
J=l
= .
Therefore
Yn ap1 ap2 apn Yn
2:f=1 aliyi L�1 ali(Lf:1 b;1x,) Z1

Lf=1 a2iYi Lf=1 a2;(2:f!=1 b;1x1) Z2
= - =
2:f=1 ap;{Lr;'=1 bi1x1) z
Here, for each k from I top, we have
m
= LCk;X; (k 1, 2, 'p),
J=l
= . . .
where
n
ckJ· = L akibii"
i=l
Thus
But the above, formula for ck1 says that c k; is the inner product of the kth row of M1
and the jth column of M0; these are the row vector and column vector
and their inner product is ck; = z�1 akibi;· For example, in the product
]
c22 is the inner product of (-2, 0, 2) and (3, 2, 1). Therefore c22 = -4. The com
plete calculation gives the answer
10.
u
-4
4 .
-2
(This should be checked.)

Similarly, we define the sum of two matrices to be the matrix of the sum:
Definition. Given the linear transformations /: Rm->- Rn and g: Rm->- Rn, with
matrices M1 and M9• Then
M1 +Mg= Mt+u·
The sum is easy to calculate; the simplest possible idea works. Let
au ll12 a ln bu b12
ll21 ll22 a2n b21 b22
M, = Mg=
Then for P = (x1, x2, . • • • , xm), we have
L�=l aux;
L�=l ll 2;X;
f(P) =
g(P) =
Therefore
z7=1 (a1; + b1;)X;

L7=1 (a21 + b21 )x1
(f + g)(P) = f(P) + g(P) =
The matrix which gives this result in one step is
[ci;] = [ai; +bi;] = [ai;] + [bi;];

au a12 aln bu b12 bin
a21 a22 a2n b21 b22 b2n
+
am1 am2 ... a mn bml bm2 bmn

a u+b11 012 +b12 a1n +bin
a21 +b21 a22 +b22 a2n +b2n
aml +bml am2 +bm2 ... amn +bmn
PROBLEM SET 13.2
Carry out the indicated operations, expressing each answer as a matrix (which may, of
course, turn out to be an n by 1 matrix, that is, a column vector).
[� iJ[: :J [� m: �] [; �][� �]
2 0
1. 2. 5 1 3.
8 0
[� m� !] [! m� i] [� m� !J
2 0 10 2 0
4. 5 1 5. 1 5 6. 1
8 0 0 8 1
7.
rn
0
2
0 urn 8. [3
f�]
0 9.
[! !J[l]
2
5
[! !][�] [� mi i] [� m�f �J
1 4 1
10. 0 11. 0 5 12. 0
f
0 0 6 0
n m� !] ff f G f
4 1 ·1 l
13. 5 0 14. 0 15. 0
6 0 0 0
[� n [! ff"" [l �m �]
1 1 0 1
16. 0 17. 0 18. 4 3
1 0
['� i][i �]
0 0
[l m� �] [� m: �]
1 0 0 1 1 0
19. 3 1 20. 1 3 21. 15 1 1
0 0 0 0 18 8 14 1 0
13.3 Formal Properties of the Algebra of Matrices. Groups and Rings 577
0 1
[ _l ][ -2
]
22.
[l mf i l]
4
1
3
0
23.
0
-6
2
0 O
-6
2
I -1
I
-1
4
4
[� J
1 0 0
24.
[� f 1J1
0
0 0
25.
0
0
0
1
0
O
0
1
26.
[� 0
1
0
1
0
0
0
0
0
27.
[� T 0
1
0
1
0
0
0
0
0
For each of the following four matrices, find the i nverse M-1 if there is an inverse; if
[ I 2 3
OJ [�
not, give the simplest reason that you can for concluding that no inverse exists.
0 0 2 3
[� �
7
4]
2 0 4 5 6 0 6 8
28. 29. 30. 31. [� �]
0 2 7 8 9 0 10 11 12
0 0 10 11 12 0 0 0 0
*32. Let
12 =
[� �l
For how many 2 by 2 matrices
M = [� �]
is it true that MM = 12? (It is easy to find two such "square roots" of 12• The question
is whether there are others, and if so, what they are.)
33. Let
02 =
[� �].
This matrix acts like 0, in that for every 2 by 2 matrix M, we have
02 + M M + 02 = M.
Question: If A and B are 2 by 2 matrices, and AB 02, does it follow that A 02

=
or B 02?
= =
34. If A and B are 2 2 matrices and AB 02, does it follow that BA 02?
=
x = =
.
13.3 FORMAL PROPERTIES OF THE
ALGEBRA OF MATRICES. GROUPS AND RINGS
In the preceding problem set, you found that addition and multiplication of matrices
were analogous in some ways, but not in others, to addition and multiplication of
real and complex numbers. We shall now investigate the algebra of matrices syste
matically, and find out how far the analogy goes.
Throughout this section we shall be concerned only with square matrices. The
set of all n by n matrices is denoted by ,An. In our investigation of the formal prop
erties of vltn, under addition and multiplication, it will not be very useful to think
about square arrays of numbers; the ideas are much easier to see if we work with the
linear transformations f that the matrices represent.
Definition. !l'n is the set of all linear transformations f: Rn ---* Rn.
We found, in Theorem 3 of Section 13.2, that !l'n forms a vector space, under
addition and scalar multiplication. Therefore, in particular, we have:
C.1. !l'n is closed under addition.
A.1. Addition in !l'n is associative.
A.2 (Existence of zero). There is an element/0 of !l'n such that fo + g = g + fo = g

for every g.
(Obviously/0 is the linear transformation such that
f0(P) = 0 = (0, 0, ... , 0) for every Pin Rn.)
A.3 (Existence of negatives). For each/ in !l'n there is a -f such that
f + ( - f) ( -f) + f fo.
= =
A.4. f + g g + f, for every f and g.

=
A pair [!l', +] is called a group if the operation + satisfies C. I, A.I, A.2, and
A.3. If A.4 is also satisfied, then [!l', +] is called a commutative group. We can
therefore sum up as follows:
Theorem 1. For each n, [!l'n, +] is a commutative group.
We defined multiplication for matrices by composition of functions. That is,
M1Mu = M1tu>,
by definition. We therefore need to investigate composition of functions in !l'".
For the sake of convenience, we shall denote the composite function f (g) by the
notation f g . Theorem 4 of Section 13.2 tells us that iff and g are linear: R" ---* Rn,
o
then so also is f o g. This gives

C.2. The set !l'n is closed under the operation o.
For each/, g, h in !l'n,

f 0 (g 0 h)
is the composition off and g(h); and
(f 0 g) h 0
is the composition off(g) and h. T� ese give the same answer:

f 0 (g 0 h) = (f g) 0 0 h.
The reason is shown by a diagram:
fog
13.3 Formal Properties of the Algebra of Matrices. Groups and Rings 579
Starting at any point w, we get to the same point z, no matter how the functions
f, g, hare grouped. Therefore we have:
M.1. In 2n, the operation o is associative.
(Obviously this has nothing to do with linearity; compos1t10n of functions is

always associative, regardless of the domains, or the ranges, or the nature of the
functions.)
M.2 (Existence of unity). There is anf1 in 2n such that
for every g.
This f1 acts like the number 1, under our "multiplication." Obviously f1 is the
"identity" function, such that/(P) = P for every P. We call/1 the unit element.
So far, M.l and M.2 are precisely analogous to C.l, A.I, and A.2; these conditions
say the same things, about addition in one case and "multiplication" in the other.
1
But the analogy now breaks down: not every linear function f has an inverse J- ,
and composition of functions is not, in general, commutative. (We have seen many
examples of both of these.) But we do have:
DL (The distributive law). In 2n,f 0 (g+ h) =fog+f 0 h.
The reason is thatf 0 g isf(g); andf(g+ h) = f(g)+f(h) because f is linear.
A system satisfying all the conditions that we have mentioned so far is called a
ring. More precisely:
Definition. Given a set 2, with two operations+ and 0• The system
[2,+, o]
is a ring if the following conditions are satisfied:
R.1. The pair [2,+]is a commutative group.

R.2. The set 2 is closed under 0; a is associative; arid 2 contains a unit element.
R.3. The operation a is distributive over+. (That is,/ a (g+ h) =fag+fa h.)
All this discussion carries over immediately to the set ._,«n of n by n matrices,
since M1 + Mu = Mt+u and M1Mu = M1.u. This gives:
TheQrem 2. For each n, the system [Jin,+, 0]is a ring.
Here 0 is used to denote matrix multiplication. It is easy to see that the zero
eleinent of ._,«n is
0 0 0
0 0 0
0 0 ... 0
and that the "unit element" of ..,1tn is the matrix
1 0 0 0
0 1 0 0
1
In =
with l's on the main diagonal and O's everywhere else. There is a shorthand for this:
{
we define
/J .. = 1 for i = j,
Z?
0 for i -:;6 j.
We then have
In = [o;;].
If you try proving Theorem 2 directly, carrying our calculations with n by n
arrays of numbers, you will see that the use of the system !t' of linear transformations
offered great advantages; most of the proofs were easier to write down than even
one n by n matrix. In particular, a direct verification of the associativity of matrix
multiplication, using the formula for the product of two matrices, would be extremely
tedious.
A final remark, on the notation used in describing a group. In this section, the
group operation is denoted by +. This is partly because addition was what we meant,
in the case that we were discussing. Also it is customary to use the symbol + when the
operation is commutative. More generally, however, we can state the conditions for a
group as follows:
Definition. Given a set G and an operation*· The system
[G,*]
is a group if the following conditions are satisfied:
Cl (Closure). G is closed under*·
A (Associativity). a* (b *c) = (a*b) *c, always.
(Existence of Unity). e e *a a*e a,

=
EU There is an element of G such that =
for every a.
EI (Existence of Inverses). For each a in G there is an element a-1 of G such that

a a-1 a-1*a e.
=
* =
As before, the group is commutative if it satisfies:
Com (Commutativity). a*b = b*a, always .
A field is a system [F, +, ·] which satisfies all the conditions which were stated
for the real number system in Section 1.1. Thus [F, +, ·]is a field if (1) [F, +, ·]is.a
ring, (2) multiplication is commutative, and (3) every x -:;6 0 has an inverse x-1, such
that x · x-1 = 1.
13.3 Formal Properties of the Algebra of Matrices 581
Obviously the real-number system furnishes examples of all the ideas that we
have been talking about in this section: [R, +, ·] is both a ring and a field, and
[R, +] is a group. But if the real-number system were the only algebraic system that
we were concerned with, there would be no advantage in using the terms group, ring,
and field. The advantage is in other conne
. ctions: already we have been dealing with
vector spaces, which form groups (under addition), but do not form rings or fields;
and from now on, we shall be dealing with (a) groups which are not rings, (b) rings
which are not commutative, (c) commutative rings which are not fields, and so on.
To find our way around in this variety of algebraic systems, we need a language in
which we can explain briefly and clearly what sort of system we are dealing with at a
given moment.
PROBLEM SET 13.3
1. Let G be the set of all complex numbers of the form

z=cosO+isinO (OinR.)
Which, if any, of the following statements are true, and why, in each case?
a) [G, +]is a group.
b) [G, ·]is a group.
c) [G, +,·]is a field.
2. Let G be the set of all "pure imaginary" numbers, of the form z = iy (y inR.) Discuss
as in Problem 1.
3. Let G be the set of all complex numbers of the form m + in, where m and n are integers.
*4. Same problem, where m and n are rational numbers.
5. Show that if a and bare rational, and a + bv2 = 0, then a = b = 0.
6. Let G be the set of all real numbers of the form a + bVZ, where a and bare rational
7. In this problem, you may regard it as known that 1T is not a root of any linear or quadratic
equation with rational coefficients. Let G be the set of all real numbers of the form
a + b1T, where a and bare rational. Is [G, +,·]a ring?
8. A permutation matrix is a square matrix with exactly one 1 in each row, exactly one 1
in each column, and O's everywhere else. The set of all n by n permutation matrices is
denoted by pn. Show that [P2 , o] is a group, and write a multiplication table for the
group, in the form
0 A
12
A
What is the effect of a matrix in P2 on a vector P = (x1, x2)?

9. Now carry out the same process for [P3, o). (You need not write a complete multi
plication table, but you should find out whether the group is commutative.)
* 10. An upper triangular matrix is a matrix [ai1] with a;1 0 for j < i. Denote the
=
set of all n x n upper triangular matrices by vn. Show [ vn, +] is a group, but [ vn, o]
is not.
11. Show that if f and g are in zn,

and f and g are invertible, then fog is also invertible.
(The proof should be direct: you should produce a function which is the inverse of
Jog.)
12. . Let GL(n) be the set of all invertible transformations in zn. Show that [GL(n), ] is a o
group. (This is called the general linear group.) Then show that [GL(n), +, ] is not a
o
ring.
*13. Let GLU(n) be the set of all upper triangular n x n matrices [ai1], with a11a22 • • · ann ¢
0. Show that [GLU(n), ] is a group, but [GLU(n), +, o] is not a ring.
o
13.4 THE DETERMINANT FUNCTION
The determinant function assigns, to every square matrix, a real number. The
definition of this function begins as follows.
a11 a 12 ll1n
ll21 ll22 G2n
Given an n by n matrix [a;1], we take all possible products of the form
using exactly one element from each row and exactly one element from each column.
Thus the numbers
are all different. To each of these products we attach a + or -'-- sign, according to a
rule which will be stated presently. We then take the sum
of all terms which can be formed according to the above rules. This sum is called the
determinant of the matrix, and is denoted by det M, or det [ a;1 ]. Thus, when we have
explained how the sign is to be chosen for each term, we shall have a function
n
det: ,A ---+ R.
The rule for the signs takes time, to explain and justify. With each term
there is associated a function
where
In= {l, 2, 3, ... , n}.
Here p{i)= j;; that is, p(i) is the column number of the element a;1, that we chose
from the ith row. We can describe such a function p by a diagram in the following
13.4 The Determinant Function 583
(1 2 3 . )
form:
n
p
· · ·
= jl j2 ja . • jfl ,
with the numbers i in the top line and the numbers p(i) below them. For example,
(13 12 43 24)
is the function under whose action
1H3, 2H1, 3 H4, 4H2.
A one-to-one function p: In---+ In is called a permutation. When permutations are
described in the two-line notation, the order of the columns does not matter; all that
matters is what is under what. For example,
(13 12 43 24) = (24 34 12 13) = (31 34 12 24) '

and so on.
If p interchanges two integers a and b, and leaves every other integer in In fixed,
then p is called a transposition. For example,
p
=
(13 22 13 44)
1 3.
(13).
is a transposition; it interchanges and We denote this permutation by the short
hand In general, (ab) is the permutation which interchanges a and b.
Theorem 1. Every permutation can be expressed as a product of transpositions.
Here the word product is used in the sense of composition of functions. For
example, for
p
= G i ! �).
we can use the following transpositions:
) (13) ) = (21) � �) (24).

G ; � : = G i ; : =
We now have
(24)(21)(13). p =
As always, for composition of functions, the operations are performed in the order
from right to left. Thus
1 (13) 3 (21) 3 (24) 3,

2 613) 2 (21)
I- �b"ti:-
(24) 1,
3 r(13) (21) 2 (24) 4,
)
41 (13) (21) 4 (24) 2.

\
4 ----+-
This checks, with
1H3, 2H1, 3H4, 4H2.
The scheme that we used on this example always works. Given
(1 2 3 n)
··
· · ·
p= ji j2 ja · jn '
we first take the transposition (1 j1); this puts ji under 1, where we want it to be. To
the resulting sequence, we apply a transposition which puts j 2 in the second position,
and so on. A further example:
�)·
2 3 4 5 6
p= G 5 4 1 7 3
Here we could use the following stages:
1 2 3 4 5 6 7
1
(12)
2 3 4 5 6 7
(15)
2 5 3 4 1 6 7
(34)
2 5 4 3 1 6 7
(31)
2 5 4 1 3 6 7
(37)
2 5 4 1 7 6 3
(36)
2 5 4 1 7 3 6
It is not claimed, in Theorem 1, that every p can be expressed in only one way as a
product of transpositions; and in fact this is not true. For example, the above
diagram gives
p= G ; ! 1 � � �) =
(36)(37)(31)(34)(15)(12).
But it is also true that
(36)(73)(16)(43)(57)(24)(16)(27)(14)(57),
p=
which looks different, and uses ten transpositions instead of six. Nevertheless all such
expressions for a given have a common property, now to be described.
p
A permutation is called even if it can be expre&0� : as the product of an even
p
number of transpositions; is odd if is the product o an odd number of trans
p p
positions.
Theorem 2. No permutation is both odd and even.

Proof The alternating function on n variables is defined by the formula
i<j
where the expression on the right is the product of all differences xi - X; for which
i < j. (Analogously, we might use
to denote _2;=1 xi). For example,
f(x1, X2, X3) = (x1 - X2)(x1 - X3)(x2 - X3),

and
f(xu x2, x3, x4) = (x1 - x2)(x1 - xa)(x1 - xJ(x2 - xa)(x2 - X4)(xa - xJ.
Now consider what happens to f when we apply the transposition (i j), thus inter
changing xi and X;. The factors of f are of the following types:
1) (xr - xi), (xr - X;) (r < i),

2) (xi - x8), (X8 - X;) (i < s < j),
3) (xi - xt), (x; - xt) (j < t),
4) (Xi - X;).
When we apply the transposition (ij), interchanging xi and X;, the effect is
1)
( <Xr - Xi) H (Xr - X;),
(x, - X;) H (xr - xi) ,.
2)
{(xi - x,) H (x; - x,) -(X8 - X;),
=
(x, - x;) H (x, - xi) = -(xi - x,);
3)
{(xi - xt) H (x1 - xt),
(x; - Xt) H (xi - Xt);
4) (xi - x;) H (x; - xi) = -(xi - X;)·

Thus the factors of the first three types fit together in pairs, and in each case, the
products of the pairs are left unchanged. But the sign of xi - x1 is changed, and so
the sign of f is changed. Briefly,
(ij)f(x1, X2, ... 'Xn) = -/(x1, X2, ... 'Xn)

or; more briefly still,
(ij)f =
-f
It follows that pf= f if p is an even permutation, and pf= -f if p is an odd per
mutation. No permutation can have both these effects, and so the theorem follows.
On this basis, we can finally define the determinant function:
det [ai;] = ,2 ±a1;, a212 • • • anin'

where we use + if the permutation
= (1. 2 3 . �) . .
P
j2 ja
Ji ' '
·
}n
is even, and - -if pis odd.
In practice, when we have developed some of the theory, we shall never have to
make direct use of the above definition; and this is fortunate, because the definition
is even more tedious to handle than one might think. In order to form a term of
det [a;;], for an n by n matrix, we have to choose an element from each row, in such a
way as never to use the same column twice. Thus we have n possibilities to choose
from in the first row; there are then -1 n possibilities in the second row; and so on.
Therefore the total number of terms is
n(n - l)(n - 2) 3 21 · · · · · = n!
Therefore the determinant of an n by n matrix is the sum of n! terms. In particular,
for n = 20, the number of terms of det Mis
=
20! 2,432,902,008,176,640,000 .
The number of seconds in a year is only 31,526,000. This is why nobody asks even an
electronic computer to calculate the determinants of large matrices by brute force.
Nevertheless, the definition of the function det is usable conceptually, as the
basis of a theory which leads quickly to efficient techniques. In the rest of this section,
we shall begin to develop the portion of the theory which makes direct use of the idea
of odd and even permutations.
Theorem 3. Every permutation is invertible.
Obviously, since every permutation is one-to-one.
Theorem 4. For each n, let S,. be the set of all permutations p: I,.--+ I,.. Then
[Sn, ] is a group.
o
Proof Sn is closed under because the product p o q of two one-to-one functions

(1) o,
In--+ In is another such function.

2) The operation o is associative; composition of functions always is.
3) There is an identity
= (1, 2, 3, ... , n )
e
1, 2, 3, ... , n ·
4) By Theorem 3, every pin Sn has an inverse.
Theorem 5. In any group, the inverse of a product is the product of the inverses,
in reverse order.
Thus
because
(p o q) o (q-1 o p-1) = p o (q o q-1) o p-1 = p o e o p-1 = p o p-1 = e.

Hereafter, we shall omit the operation sign o. For products of n factors, the theorem
says that
(P1P2 · · · Pk-1Pk)-1 = p/;1p/;!.1. · · P21P1\

and this is true, because in the product
P1P2 ... Pk-1PkP.i/P"k!.1 ... P21P1\

all the factors cancel each other in pairs, starting in the middle.
Note that if the group is commutative, then the order doesn't matter, and the
theorem takes a simpler form. But the group that we are working with at the moment
is not commutative.
Theorem 6. For each p in Sm p and p-1 are either both even or both odd.
Proof We express p as a product of transpositions:
p =
P1P2 . .
.
h-1h·
where each p; is a transposition. Every transposition is its own inverse. Therefore
p -l =
hh-1 . .
.
P 2Pi·
If k is even, then p and p -1 are both even. If not, p and p -1 are both odd.
The transpose of a matrix .!It is the matrix obtained by reflecting .!It across its
main diagonal. The transpose is denoted by .Jtt. Thus
and in general
Theorem 7. For each Min .,1tn,
det M = det M.t
Proof The terms of det Mare of the form
with + or - according as the permutation
p
(1 2 3 .
.
. �)
= ji j2 ja ' ' · }n
is even or odd. The corresponding term of det Mt is
with + or - according as the permutation
q
=
Ct 1 j; �n)
is even or odd. Obviously p-1, and sop and are both even or both odd. There
=
q q
fore the terms of det Mand det Mt have the same signs, and det M det M.t =
Theorem 8. If two rows of Mare interchanged, then the determinant of the resulting
matrix is - det M.
For example,
The reason is that when two rows are interchanged, this contributes exactly one
transposition to the permutation
p
(1 2 3 ··· n ) .
= A jz ja ··· jn
For example, if the first and third rows are interchanged, then the sign of the term
a a • • • a in the new determinant is determined by the permutation
(1 )
1ii 212 nJn
2 3 ··· n . .
q =
=
()1 ]a) 0 p.
h j2 ji ··· n
PROBLEM SET 13.4
1. Working directly from the definition of <let, get an explicit formula for
[anll21 .
J
<let
ll12
ll22
2. Similarly, get a formula for
<let
["" ""] au
a21 a22 a23 •
ll31 ll32 ll33

3. Similarly, get a formula for
a,, ""
[001 010 ]
<let
ll23 ll24
ll33 ll34 •
[� 000 007 400 !]

000 ll44
Calculate the following.
[1
0 0
1 0 OJ [ 0 0 2
0 00 00
4. 00 00 01 010
<let
0 5. � 0 001 �]
0
d� 6. det
7. [� 001 00
�·
0 4 �] 2
8. [� 000 007 400 !]
0
det
00 01 2
9. det
[! 00 07 00 �]
00 000 401
2
10. det [i � i � i] 0
1
� �
0
1
3
0
5
[H H �] �]
2 -6 1
0 0 0
12 det 0 0 4
[! : ]
0 -14 0
0 0 0
0 0 0 0 0
o o 6 1
0 -2 0 0 0 0
� o 4 o � 3 0 0 0 0 0
14 . det 0 15. det
0 0 0 0 -4 0
l
0 0 0 0 0 5
7
0 0 -6 0 0 0
1
6
�
2
5
3
4
4
3
5
2
6
1
0
1
1
0
0
0
0
0
0
0
0
0
1 2 3 4 5 6 0 0 0 1 0 0
16. det 17 . det
2 3 4 5 6 7 0 0 1 0 0 0
3 4 5 6 7 8 0 0 0 0 2 3
4 5 6 7 8 9 0 0 0 0 4 5
18. Let An be the set of all even permutations in Sw Is [An, 0 ] a group?
19. Let Bn be the set of all odd permutations in Sn. Is [Bn, 0] a group?
20. Let Cn denote the set of permutations in Sn, for whichji = 1. Is [Cm a] a group?
21. Let Dn denote the set of permutations in Sn such that either Ji = 1 and h = 2 or
h = 2 and iz = 1. Is [Dm a] a group?
22. Can any general statement be made about the evenness or oddness of the following
(
permutation?
;)
_ 1 2 3 .·
. n-2 n- 1
p- n n- 1 n- 2 ... 3 2
23. What can you say about the sign of the following permutation?
=
( 1 2 3 . . · n- 2 n- 1 n
)
q n 1 2 ... n- 3 n-2 n- 1
24. Suppose that [G, a] satisfies all the conditions for a group, except that some elements
of G may not have inverses. Let H be the set of all elements of G that have inverses.
Does it follow that [H, a] is a group?
25. Find the roots of the equation
det [�: f: fJ � 0.
Express the left-hand member as a quadratic expression in factored form.


�
a
x x xa x3 ]
2 3
x x2 x2 x2
det
X1
� 2 3 = 0.
X2 X3 X
[ 1 1 1 1
Then express the lefthand member as a cubic equation in factored form.

*27. Find a matrix M whose determinant is the alternating function / (xi. x ,
2 • • • , Xn).
13.5 EXPANSIONS BY MINORS.

CRAMER'S RULE AND INVERSION OF MATRICES
In the preceding section we showed that
detMt = detM
for every square matrix M; and we showed that if two rows ofM are interchanged,
the effect is to change the sign of detM. These statements in combination give us the
following:
Theorem 1. If two rows of M are interchanged, or two columns are interchanged,

then the determinant of the resulting matrix is -detM.
To get the second half of this theorem, we take the transpose, perform the
appropriate interchange of two rows, and take the transpose of the resulting matrix.
The first and third of these operations leave the determinant unchanged, and the
second one reverses the sign.
In fact, since detMt = <let M, every theorem about rows automatically gives us
a theorem about columns.
Theorem 2. IfM has two identical rows, then <letM = 0. Similarly for columns.
The reason is that when the two identical rows are interchanged, nothing happens
to the determinant (or even to the matrix). Therefore detM = -detM, and
detM = 0.
The minor of an element a;1, in a square matrix M, is the matrix that we get by
deleting the ith row and the )th column ofM. The minor is denoted by M;1, and its
determinant detM;1 is denoted by D;1.
It is easy to see that the sum of all terms of detM that include au is auDu:
au G12 G1
3
a ln
...... ------ ----- ----------
a 21 G22 a 2a a2 n
aa1 aa2 aaa aa n
M =
13.5 Expansions by Minors 591
Every term that involves a11 has the form
here a21 a31 an is (except possibly for sign) a term of Du = det M11• And
in
• •
2 3
•
these two corresponding terms of det Mand det Mu have the same sign, because the
) (� )
permutations
(l
1
� �
)2 )a
.
•
.. � '
· · }n )2 h
� .
"· ·
.. �
Jn
are either both even or both odd.

This leads to a more general result:
1
Theorem 3. The sum of all terms of det Mthat involve a;j is (-l)i+ ai1Dw
Proof This is known for the case i = j = 1. We shall reduce the theorem to this case.
By a simple row transposition we mean an operation which interchanges two
consecutive rows of a matrix. Similarly for simple column transpositions. We assert
that ai; can be moved into the first column by j - 1 simple column transpositions:
[c1, c2, C3, C4, C5, . , Cn).

.._.. ._,.. �
• •
'l<-.A
Here c1 denotes the jth column. For j = 5, the transpositions are (c4c5), (c3c5),
(c2c5), (c1c5). The new order of columns is
Thus the fifth column becomes the first, and the other columns are in the same order,
among themselves, as they were before.
Similarly, we can then move a1
; into the first row, by i - 1 simple row trans
positions. Let the new matrix be M'. Then
det M' = ( -1 )<i-l>+!i-1> det M

i
= (-l) +i det M,
and so
det M = (-1)-H det M' = (-l)i+i det M'.
But the sum of all terms of det M' that involve a;1 is a;1 det M/1, where M1
/ is the
minor of ai; in M'; and M1
/ is M1
; , because our total operations on the rows and
columns of M did not disturb the order of the rows and columns of Mi;· Therefore
the sum of all terms of det M that involve ai; is
. . det Mi3··
(-l)i+ia1.J = (-l)i+ia i..3 Di1'
..
Theorem 4 (Expansion about the minors of a row). For each i,

n
i+i
det M = ""
£.. ( - 1 ) aiJD;J·
i=l
This is true because every term of det Minvolves exactly one element in the ith
row. Thus the above formula separates the n ! terms of det M into n classes, with
(n - 1)! terms in each class. Similarly:
Theorem 5 (Expansion about the minors of a column). For eachj,
det M = L(-1/+1a;1DiJ·
i=l
(At this stage you should check to see how these formulas apply to a 3 by 3 matrix,
using, say, the second row and the second column.)
If we multiply the elements of one column by the determinants of the minors of
some other column, with the appropriate signs, and add, we get 0:
Theorem 6. If k ;;C j, then
The reason is this. Let M' be the matrix obtained by changing the jth column
so as to make it identical with the kth column of the given matrix M. Then the above
sum is the expansion of M' about the minors of itsj th column. Therefore the sum is
det M'. But det M' is 0, because M' has two identical columns. Similarly for rows:
Theorem 7. If k ;;e i, then
Theorems 6 and 7 may seem, at first glance, to be merely descriptions of what

happens if somebody makes a mistake, but in fact they are very pointed statements.
They enable us to write explicit formulas for the solution of a set of n linear equations
inn unknowns, in the case in which the solution exists and is unique. To avoid tedious
notation, we show how the method applies in the case n = 3. (The general case is
exactly the same in principle.) Given the system
auX1 + a12x2 + aiaXa = b1,

a21X1 + a22X2 + a2axa = b2,
aa1X1 + aa2X2 + aaaXa = ba.
Let M be the matrix [ai11 of the system; let D det M, and suppose that D ;;if 0.
=
In the first equation, we multiply by D11, in the second by -D21, and in the third
by D31• Then we add:
auD11X1 + a12D11X2 + a13Duxa =

b1Du
-a21D21X1 - a22D21X2 - a2aD21Xa =
-b2D21
aa1Da1X1 + G32Da1X2 + G33Da1Xa = baDa1
i
Dx1 + 0 · x2 + 0 · x3 = L:=1(-l)i+ b;Di1.
On the lefthand side, the coefficient of x1 is D, by Theorem 5, and the coefficients of
x2 and x3 are 0, by Theorem 6. Thus the use of the minors as multipliers has given us
an equation in which Xi is the only unknown. The sum on the right-hand side, in the
last equation, is easy to describe: it is the determinant Di of the matrix
obtained by replacing the first column of M by the b/s.

To solve for x2, we multiply in the three equations by - D 12, D 22, and - D 32
respectively, and add. This gives
3
2 i
0 ·Xi + Dx2 + 0 ·X3 = ! (-1) + biDi z·
i=i
Here the sum on the righthand side is the determinant D2 of the matrix obtained by
replacing the second column of M by the b/s. The same scheme works for x3• There
fore, if D � 0, the system has one and only one solution, namely,
Xi = Di/D, X2 = D2/D, X3 = D3/D.
Obviously none of the above discussion depended on the condition n = 3. In general,
we have:
Theorem 8 (Cramer's rule). Given a linear system of the form
X1 bi
Xz b2
M =
Let D = det M. If D � 0, then the system has one and only one solution; and the
solution is given by the formula
X; = D ;f D,
where D 1 is the determinant of the matrix M1 obtained by replacing the jth column
of M by the vector ( bi, b2, , bn).
• • •
Cramer's rule has the following consequence. A square matrix M is called non
singular if M has an inverse.
Theorem 9. If M is a square matrix, and det M � 0, then M is nonsingular, and
its inverse is given by the formula
where for each i and j,

1
ci i = -(-1)'+·'D 1i·
D
i
That is, M- is the transpose of the matrix
[� (-l)i+iD;;J ·
To see why this is true, consider the matrix equation
M =
Here
n
Dxi =
I (-I)i+iyiDii•
i�l
by Cramer's rule. Therefore
xi I ( - ly+i(l/D)Di;Yi·
i=l
=
What we want is a matrix M' = M-1, such that
Here the y's form a column vector, and if M' = [c;;l, then
n
X; = L ciiYJ·
i�l
To convert our previous formula for xi to this form, we interchange i and j on the
right, getting
i
X; = t
i l
(-ly+ (-ii) Diih
The value of the sum on the right is unchanged when we usej as an index of summation.
Therefore
M-1 = M1 = [cii] =
[� (-l)i+1D1l
PROBLEM SET 13.5
1. Find multipliers which eliminate y and z from the system
x + 2y + 3z = 4,
2x + 3y + 4z = 5,
3x + 4y + 5z = 6.
Carry out the multiplications, add, and solve for x. (Here you are not supposed to use
Cramer's rule; you should use the scheme used in deriving Cramer's rule.)
2. Similarly, solve for y in the system
x - 2 y + 3z = -4,
2x + 3y + 4z = -5,
3x - 4y + 5z = -6.
3. Similarly, solve for z in the system
x - 2y - 3z = 4,
2x + 3y - 4z 0= -5,
-3x + 4y - 5z = 6.
Find the inverses of the following matrices, by a direct application of Theorem 9, and
check your answers by matrix multiplication.
4.
[� �]. 5.
[� �]. 6.
[! ]-2
4 .
G �]. [� !l [! �l
1 1
7. 8. 1 9. 0
0 0
1 0 0
[� �l [� �l
0 2 0
10. 11.
0 1 0 1
0 0 4 0
Find, by any method, the inverses of the following matrices. (You need not calculate
the determinants unless you need to, as a step in finding the inverse.)
0 0 0 0 0
[� �l [� !l [� !l
0 1 0 0 1 0
12. 13. 14.
1 0 0 0 0 0
0 0 0 1 0 1
0 0 -1 0 0 0
[� �] [� �]
-2 0 0 0 0 1
15. 0 0 0 16. 0 1 0
0 0 0 1 0 0
0 -5 0 0 0 0
·
17. Suppose we form a 4 by 4 matrix by fitting together four 2 by 2 matrices, like this:
Let D det M, and let D;; det M;1 for each i,j. Is it true that
[ ]
= =
Dn D12 ?
D = det
D21 D22
If so, verify it. If not, give an example of a case in which it fails.

18. Similarly, discuss the case in which a 2n by 2n matrix is formed by fitting together
four n by n matrices.
19. Similarly, discuss the case in which a 6 by 6 matrix is formed by fitting together nine
2 by 2 matrices.
13.6 ROW AND COLUMN OPERATIONS.

LINEAR INDEPENDENCE OF SETS OF FUNCTIONS
We shall now show that when we apply to a square matrix the "triangularization"
process that we applied to systems of linear equations in Section 13.1, the determinant
of the matrix is unchanged.
Theorem 1. If one row of a square matrix is multiplied by a scalar, and the resulting
vector added to another row, the determinant of the matrix is unchanged. Similarly
for columns.
Proof Suppose that the kth row of the matrix M = [ai;] is multiplied by a and added
to the ith row, giving a matrix M'. Expanding M' about the minors of the ith row,
we get
n
;
det M' = ! (- lY+ (a;; +aak 1)D;;
i=l
n n
; ;
= ! (-1y+ a;;D;; +a! (-l)H ak;D;; = detM +a· 0,
i=l i=l
by Theorems 4 and 7 of Section 13.5. We get the other half of the theorem by taking
transposes, as in the proof of Theorem 1 of Section 13.5.
Iterations of this procedure constitute the most efficient scheme for computing
determinants; by appropriate row (or column) operations, we can introduce O's
into a particular row (or column), so that when we use an expansion by minors, only
one of the minors needs to be computed. Note that without these preliminaries, an
expansion by minors is not a short cut in computation, but merely a device for
systematizing our work; in an expansion by minors, the same number of terms appear
as under the original definition of the determinant; they have merely been sorted into
n sets of (n - 1) ! terms each.
Tbeorem 2. If the rows of a matrix M form a linearly dependent set, then det M = 0.
Similarly for the columns.
Proof Let M = [a;1]; let the rows be r; = (a;1, a;2, • • • , a;n); suppose that
n
! a;r; = 0,
i=l
for some set of numbers a;, not all equal to 0. Then some r;, say, r1, is a linear
combination of the others:
n
r1 = L {Jiri
i=2
13.6 Row and Column Operations 597
By i - 1 row operations as in Theorem 1, we get a new matrix M' in which r1 is

replaced by a row of O's. Therefore <letM' 0, and so <letM <letM'
= 0. = =
·
The converse is a little harder.
Theorem 3. If detM = 0, then the rows of Mform a linearly dependent set (and so
also do the columns).
Proof The proof is by induction. Obviously the theorem holds for 1 by 1 matrices.
We need to show that if it holds for n - 1 by n 1 matrices, then it also holds for
-
n by n matrices.
Given an n by n matrix M [ai;], with rows r1, r2,
= , r n· If any row ri is the • • •
zero vector, then the linear dependence of the set
R = {r1, r2, • • • , r n}
is obvious. Therefore we may assume that r 1 -¥- 0. We may also assume that a11 -¥- 0,
since the linear dependence or independence of the rows is unaffected by permutations
of the columns.
Now consider the matrix M' whose rows form the set
R' = {r ,{ r�, ... , r�} = {r1, r2 - Ol2r1, . • • , r n - Olnr1 }.

The r/s are linear combinations of the r;'s, and vice versa. Therefore R and R'
span the same subspace of Rn. This gives:
1) If R' is linearly dependent, then R is linearly dependent, and conversely.

But the Ol/S can be chosen so that a11 is the only nonzero element in the first
column ofM'. Therefore <letM' a11D�1, where D�1 is the determinant of the minor
=
M1{ of a11 in M'. Since

<letM' = detM = 0 and
we have
2) <letM{1 = 0.
ButM{1 is an n 1 by n - 1 matrix. Therefore the rows ofM{1 form a linearly

-
dependent set. Therefore R' is linearly dependent. Therefore R is linearly dependent,

This theorem easily gives:
Theorem 4. If the rows of a matrix M are linearly dependent, then so also are the
columns, and conversely.
Proof By the preceding two theorems, each of the following conditions is equivalent
to the next:
1) The rows of Mare linearly dependent.

2) detM 0. =
3) The columns of Mare linearly dependent.

Here, and throughout this chapter so far, we have been talking about matrices of
numbers. We shall now discuss linear independence of sets of functions; for this
purpose, we shall use matrices of/unctions; and the first thing that we need to under
stand is that for matrices of functions,the analogue of Theorem 4 is false. This can be
shown by a very simple example, as follows:
[l 2
M(x) = x 2x
2 2
x 2x
Here the columns are linearly dependent, obviously; but the rows are not: if
for every x,
for some real numbers 1)(1, 1)(2, 1)(3, then 1)(1 = 1)(2 = 1)(3 = 0, because the only poly
nomial that vanishes for every x is the zero polynomial.
The simplest general test for linear independence of functions uses the determinant
of a matrix of functions. Let /1, /2, • • • Jn be functions on an interval I. The
Wronskian of the sequence /1,/2, ,fn • • • is the function
!1 f{ f�
!2 f� f�
n l
fn J� f� · · · f� - )
Thus
W(x) = det
[JJ1-1>];
the Wronskian matrix has the (j - !)-derivative off; in the ith row andjth column.
Note that the Wronskian really depends on a sequence of functions, and not merely
on a set of functions; if the same functions are taken in a different order, the sign of W
may change. The notation W(x) is meant to emphasize that the Wronskian is a
function and not a number. The following is easy:
Theorem 5. If {fi,/2, • • • , fn} is linearly dependent, then W(/1,/2, • • • ,fn) = 0.

That is, W(x) = 0 for every x.
The reason is that for each x, the rows of the Wronskian matrix are linearly
dependent: if
n
2, a,f�(x) = 0 for every x,
i=l
for some numbers et.i which are not all = 0, then automatically
n
D1 2, et.Jlx) = 0 (j = 1, 2, . . . , n - 1).
i=l
(Here Di[· · ·] denotes thejth derivative.) Therefore
n
2, l)(;f�1>(x) = 0 (j = 1, 2, . . . ,n - 1),
i=l
for every x, and the rows are linearly dependent.
13.6 Row and Column Operations 599
Ordinarily, we apply this theorem backwards, using the following equivalent

form:
Theorem 5'. If W(x0) ¥:- 0 for some x0, then the set {fi,/ 2, • • • Jn} is linearly
independent.
j;(x) =cos x.
[ ]
Here
sin x c ?s x
W(x) det
cos x
=
-sm x
= -sin2 x - cos2 x = -1 for every x.

Here W(x) ¥:- 0 for every x, and it follows that sin and cos are linearly independent.
But sometimes we have to choose a particular x0• For example, consider
flx) = sin 2x.
[� .
Here
J
snx cos x
W(x) =det
sm 2x 2 cos 2x
Here W(O) = 0, but
w(�) =det [� �J- = -2.
By Theorem 5', {sin x, sin 2x} is linearly independent. Similarly for
[ xe"' xe"' [ xe"'

Here
J J
+ e"' e"'
W(x) =det 2 ., + 2xe ., =det
2 .,
xe xe x2e ., 2xe"'
[1x 2x
J
1 =x2e2"''
== xe2"'det
and so W(x) ¥:- 0 for every x ,= 0.

Elaborations of these techniques will build up gradually, in the following problem
set. Meanwhile, it is natural to inquire about the converse of Theorem 5.
(?)Theorem(?). If W(x) =0 for every x, then {fi,/ 2, • • • Jn} is linearly dependent.
.As it stands, this is false; and strong hypotheses need to be added to make it a
true theorem. For example, let
fi(x) =x2,
{
2x2 for x � 0
!2(x) = 3x2 for x � o'.
Then W(x) = 0 on the interval [O, co), because/1 and/2 are linearly dependent on
[O, oo); and for the same reason, W(x) =0 on ( oo, O]. Therefore W(x) =0 for
-
every x. Nevertheless, {/t>/2} is linearly independent, because there is no one pair

of nonzero constants oc1, oc for which

2
for every x.
If the above equation holds for every x, then setting x = -1 and x = 1 we get
{ oc1 + 2oc
2
= 0,
oc1 + 3oc = 0.
2
By subtraction, oc = 0. It follows that oc·1 = 0.
2
This indicates that the Wronskian can give proofs of linear dependence only for
special types of functions. Fortunately, these functions are of special interest and
importance, as we shall see.
In the following problem set, you will be writing long strings of equations between
determinants, and it will be convenient to use I · · · I as an abbreviation for det [· · ·].
For example,
I� �I = det [� �J = ad - be.
PROBLEM SET 13.6
Investigate the following sets of functions for linear dependence.
1. {e"', e2"'} 2. { e"', e2"', e3"'} 3. {e-"', e"'}

4. {e"', e2"', xe"'} 5. {sin x, sin2 x} 6. {cos 2x, sin2 x}
7. {sin x, cos x, sin2 x} 8. {x, sin x} 9. {ea"', eb"'} (a -,= b)
10. {ea"'' eb"'' xea"'} (a -,= b) 11. {eax' ebx xeax' xeb "'}' (a -,= b)
12. {e"', xe"', x2e"'} 13. {sin x, sin 2x, sin 3x}
14. {cos x, cos 2x, cos 3x} 15. {cos x, x cos x}
16. {cos x, cos 2x, cos2 x} 17. {cos2 x, sin2 x}
4
18. {cos x, sin4 x, sin2 (2x)} 19. {e"', sin x, cos x, e"' sin x}
20. {e"', e"' sin x, e"' cos x} 21. {x, e"', xe"'}
22. Since a polynomial equation of degree n has at most n roots, it follows that for each n,
{ 1, x, x2, , xn} is linearly independent. Get an alternative proof of this statement
• • •
by calculating the Wronskian.

X 1 x12 x13
X2 x22 x3
D4 2 0.
X3 x32 x33
=
x x2 x3
Here the numbers x1, x , and x are all different. How do you know that D4 is a poly
2 3
nomial of degree three? Express D4 as a product of linear factors, not involving
determinants.
13.7 Linear Differential -Equations 601
*24. Express the following determinant as a product of linear factors:
2
x
1
x2
2
1 Xn 1
-
1 Xn
*25. Investigate for linear dependence: {ea1"', ea•"', eas'", . a '"
. . , e n , } where the a/s are all
different.
13.7 LINEAR DIFFERENTIAL EQUATIONS
Consider a differential equation of the form
f" +bf'+ cf= 0, (1)
where b and c are constants. As always, when we study a differential equation, we

want to answer the following three questions:
i) Does the equation have any solutions?

ii) If so, what are they?
iii) How do we know that the solutions that we found are the only ones?
For Eq. (1), and for many others like it, we can give complete answers to all these
questions. (In fact, the only hard one is the third.) As a guide to what to try, we look
first at cases in which some of the solutions are obvious. The equation
f" -f=O
has the solutions
because D2e"' = e'" and D2e-x = e-'". And it is easy to see that if f1 and f2 are solutions,
then so also is rxif1 + rx2f2, for every pair ofscalars rx.1 and rx. •
2
We had better make a
note of this, more generally:
Theorem 1. Given a differential equation
pn> + f <n-1)
an_i + .
..+ aif' + aof =
0,
where a/s are constants. Let "// be the set of all solutions of the equation. Then "//
f orms a vector space.
That is, "// is closed under addition and scalar multiplication. This is trivial to
check. Note, however, that ifthe zero on the right is replaced by a nonzero function,
or even a nonzero constant k, the solutions of the resulting equation never form a
vector space. (If the sum of two solutions is a solution, then 2k = k, and k = 0.)
The above example suggests that we try solutions of the form
f(x) = em'", f'(x) = memx,

f"(x) = m2emx.
If such a function is a solution, then
emx
and since -:/:- 0 for every x, this is equivalent to the equation
m2 + bm + c = 0. (2)
Equation (2) is called the auxiliary equation. There are three possibilities for its
solutions.
I. If b2 - 4c > 0, then there are two root:>
-b + )b 2 - 4c -b - )b 2 - 4c
2 2
and both these roots are real.
II. If b2 - 4c = 0, then there is only one root
m1 = -b/2,
and this is a real root of multiplicity 2, with
m2 + bm + c = (m - m1)2.
III. If b2 - 4c < 0, then the roots are two conjugate complex numbers
where(/.. = -b/2 and (3 = )4c - b2•

In case I, the functions
are solutions, and so also is every linear combination
Also {f1,f2} is linearly independent, because
W(/1,/2) = det [:::: ::::::] =

em1xem2x
det [�
em'xem2x m (
= 2 - m1) -:/:- 0.
Therefore the solutions that we have found for case I include all the solutions, if
the following theorem is true:
Theorem 2. Let Y be the solution space of the equation
f" + bfI + cf = 0.
Then dim "Y = 2.
13.7 Linear Differential Equations 603
This is true, and will be proved in the following section. Meanwhile we shall use it.
In case II, we seem to have only one solution
/1(x) = em•"';
and on the basis of Theorem 2, we need to find another one, h, such that {/1,/2} is
linearly independent. We do not know how somebody first thought of trying
h(x) = xem1"',
but at any rate, it works:
j�(x) = m1Xem1x + em1x,

f�(x) = m�xem1x + m1em1x + m1em1x
= mixem1x + 2m1 em'"';
f�(x) + bn(x) + cf2(x) = xem1"'[mi + bm1 + c] + em1"'[2m1 + b] =
0
,
0
because both the expressions in the brackets are equal to .
Also, {/1,/2} is linearly independent, because
m1
= e2m1x det [1x m1x + 1l_J
= e2m1 x � 0.
Case III looks peculiar. Taken at face value, the roots of the auxiliary equation
give us
f 1(x) e<a+Pilx ea"'(cos f3x + i sin f3x),
= =
f2(x) =
e<a-Pilx =
ea"'(cos{Jx - i sin {Jx) .
At the outset, we did not intend to get into the complex domain; but in the complex
domain, our formulas still make sense: if mis complex, then the function/(x) em"' =
is well defined; in fact, we have a function

cp(z) = entz'
with
cp'(z) = mem z,
In particular,when z is real, = x, we have
rp(z) = J (x) = em"'' rf>'(z) = f'(x) = mem"', rf>"(z) = j"(x) = m2emx.
Therefore /1 and /2 really are solutions. But at the moment we are interested only in
real solutions (in another sense), and so we take the real and imaginary parts separately,
getting
(Check that if a complex-valued function f is a solution, then its real and imaginary
parts are also solutions. This is easier than checking g1 and g2 by a brute-force
calculation.) It remains to verify that g1 and g2 are linearly independent. We have
g{(x) = rxe""' cos {Jx - {Je""' sin {Jx, g�(x ) = rxe""' sin {Jx + {Je""' cos {Jx.
Therefore
rxe""' cos {Jx - {Je""' sin {Jx ]
rxe""' sin {Jx + f3e""' cos {Jx
- e2ax det
-
[ . {Jx rx
cos cos {Jx - f3 sin {Jx ]
{Jx rx
sm sin {Jx + {J cos {Jx
= e2""' <let
[ ? {Jx
c s - /3 sin {Jx ] = {Je2""'
¥-
0
· '
¥-
sm {Jx {J cos {Jx
because {J 0.
In case III, the linear combinations
f(x) = k1g1(x) + k2g2(x) = k1e""' cos {Jx + k2e""' sin {Jx

can be described in a better form. We have
f(x) = e""'(k1 cos f3x + k2 sin {Jx)
= .Jk21 +
k22 e""' (.Jkiki +
k�
cos {Jx +
.Jk�
k2
+
k�
sin {Jx )
= ke""' cos {J(x - x0),
where
k = l 2
'\/ k 1
+
k22•
and x0 is any number such that cos f3x0 = k1/k and sin f3x0 = k 2/k. Using t for x,
we get
f(t) = ke"t cos {J(t - t0),
which describes the motion of a particle along a line, with the position given as a
function of the time. This kind of motion is called damped oscillation. To get the
graph of f, we start with the "simple oscillating function" cos {Jt; we move the graph
{Jt0 units to the right, so that t0 acts like O; and then we damp the function by multiply
ing each value by ke"t. (For rx < 0, this damps the oscillations as t -+ oo; for rx > 0,
the oscillations are damped as t -+ - oo.)
Note that in our formula for f, the constants rx and {J play a very different part
from k t0: rx and {J are determined by the coefficients
and b and c in the differential
equation, while k and t0 range arbitrarily.
It is a fact that a solution f of the equation
+ +
f" bf I cf = 0
is completely determined if f(x0) and f'(x0) are known, for some x0. This can be
verified by a calculation, for the three types of solutions that we have found, but the
theorem is best postponed until the next section, where we can give the "right proof."
Meanwhile, in the following problem set, you will find that such initial conditions
always determine an answer.
13.7 Linear Differential Equations 605
Case III, in which b2 - 4c < 0, and we get real solutions by making a detour
into complex variables, may seem peculiar, but it is case III that has the most
elementary application in physics: it describes the behavior of a vibrating spring.
This problem is as follows. Suppose that you hang a coiled steel spring from a rigid
support, like this:
The spring has a certain natural length L. If you hang an object of weight w to the
bottom end, the spring will be stretched by a distance s. It turns out experimentally
that if the weight w is not too great, then the ratio w/s is a constant k; that is, s = w/k;
the stretch is proportional to the weight. This statement is called Hooke's law.
The proportionality constant k depends on the physical properties of the spring; the
thicker and stiffer the spring, the larger k will be. This law, of course, applies only
within certain limits: if you hang a brick on the hairspring of a watch, the result will
not be an illustration of the law. Note, however, that the validity of the law for a
given spring and a given range of weights is capable of being tested by static experi
ments; and this is important, because we are about to deduce from Hooke's law first a
differential equation and then a law of motion.
If the spring is in equilibrium, when stretched to a length L + s, with a weight w
at the bottom, then the spring must be exerting a force of magnitude w = ks, upward,
to balance the force w exerted downward by gravity. Let us now set up a coordinate
system on the line which is the axis of the spring, in such a way that the origin is at
the equilibrium point for the given weight. In the figure below, we omit the spring
itself, to clarify the labeling.
.'V
x x
Suppose that the spring has been stretched to a point with coordinate x. Then two
forces are acting:
1) The force Fi exerted by the spring. This is
Fi = -k(x + )
s ,
because x + s is the total stretch. We use the minus sign because the x-axis is directed
downward.
2) The weight w. This counts positively, because weight acts downward, in the posi
tive direction on the x-axis. Therefore the total force is
F = -k(x + s) + w= -kx.
Now suppose that the weight is pulled down to a certain point x0 and then released.
Then the weight will bob up and down, with its position given as a function of the time.
For x =f(t), the velocity and acceleration are
v(t) =J'(t),
a(t) = v' (t) =f" (t).

Newton's second law says that
F(t) = ma(t),
where m is the mass. The force represented by the weight is equal to the mass m times
the acceleration g of gravity. Thus w =mg, and m = w/g. This gives
w
F(t) = - a(t).
g
But we know that
F(t) = -kx.
Therefore the function f which describes the motion must satisfy the differential
equation
w
- a(t) = -kf(t),
g
which can be written in the form
kg
f"(t) + f(t) = 0.
w
Since k, g, and ware all positive, this has the form
f" + Of' + cf = 0 (c > 0),

where bf - 4c = -4c < 0.
PROBLEM SET 13.7
In each of the following eight problems, find the solution space of the given differential
equatiOn, and then find the scalars which give the solution satisfying the initial conditions
on the right. In each of these cases, you should use .the methods but not the results of this
13.8 The Dimension Theorem for the Space of Solutions 607
section of the text. That is, set up the auxiliary equation, solve it, and then use the root(s)
to get two solutions which form a linearly independent set.
1. f" - 5f' + 6/ = O; f(O) = 1, j'(O) = 2.

2. f" - f' = O; f(O) = 2, f'(O) = 3.
3. f" - 6f' + 9f = O; f(O) = 1, f'(O) = 1.
4. f" + 4[ = O.; f(l) = 1, J'(l) = 2.
5. f" + f' + f = O; f(O) = f'(O) = 0.
6. f" - 7/' + 6f = O; f(O) = 0; f'(O) = 1.
7. f" - 7/ = O; /(1) 0, /'(1)= = 2.
8. 4f" - 4f' + f = O; f(O) = 1, f'(O) = 0.
Solve by any method:
9. A spring is such that an 8-lb weight stretches it 6 in. A 4-Jb weight is attached, allowed
to reach equilibrium, then pulled 2 in. below the equilibrium point and released. What
happens? What is the period?
10. A spring is such that a 10-lb weight stretches it 18 in. A 1-lb weight is attached, allowed
to reach equilibrium, pushed 6 in. above the equilibrium point, and released. What
happens? What is the period?
11. You found, in Problem 27 of Problem Set 4.3, that the sine and cosine are the only
functionsf and g for which it is true that
f' =g , g' = -f, /(0) = 0, and g(O) = 1.

Show that there is only one function f for which it is true that
f" = -f, /(0) = 0, and j'(O) = 1.

*12. We know that the set of all infinite sequences of real numbers forms a vector space.
Letf1,f2,f3 be solutions of the differential equation
f" + bf' + cf = 0;
and for i = 1, 2, 3, and every j, let
This gives three "vectors"
Y2 = (J21. Y22• · ·
. ), Ya = (y31, Ys2, · · . ),
which form a "3 by infinity matrix." Show that the rows of this matrix form a linearly
dependent set.
13.8 THE DIMENSION THEOREM FOR THE SPACE

OF SOLUTIONS. THE NONHOMOGENEOUS CASE
In the preceding section, we found that for every equation of the form
f" +bf'+ cf= 0, (1)

the solutions formed a linear space "Y. In each case, we found solutions /1, h such
that {j1,j;} is linearly independent. We shall now show that the linear combinations
f= rx i f1 + rxd2
are the only solutions.
Lemma. Every solution of (1) has derivatives of all orders.

Proof Obviously every solution of (1) has a first and a second derivative, and
f" = -bf' - cf
Here the righthand side is differentiable, and so also is the lefthand side. Therefore
j < a> = -bp2> - cf(l>.

Similarly,
p4> = -bpa> _ cp2>;
and by induction,
pn> =
-bpn-1) - cpn-2)'
for every n, which proves the lemma.
It follows that for every solution of (1) we can write a "formal Taylor series"
oo . oo
f(i\a) .
I a;(x - a)' =
I-.- (x - a)'.
i=O i=O l !
We shall now show that f is real-analytic; that is, the Taylor series converges, for
every x, and its sum is the function/that we started with.
Theorem 1. If/is a solution of (I), defined in a neighborhood of a point a, then
oo j<il(a)
f(x)= I -.- (x - a)i for every x.
i=O l !
In the proof, we shall use Taylor's theorem. (This is Theorem 1 of Section 10.10.
Note that we are now using it for the first time.) The theorem says that for each x,
the remainder
n f(i)(a) .
Rn(x) = f(x) - L -. - (x - a)'
i=O l !
is given by the formula
f(n+l)(x
-
)
R
n
(x) = (x - a)n+l'
(n + 1)!
for some x between a and x. We want to conclude that Rn(x) � O; and to do this,
we need to show that the numberspn+i>(x) cannot increase fast enough to overcome
the effect of the (n + 1) ! in the denominator. It should be understood that x is fixed,
throughout the following discussion.
Let k be any number such that

k � lbl, k � !cl, and k � 1;
and let M be a number which is an upper bound for both lf(t)I and l/'(t)i, on the
interval
I = {t I It - al � Ix - al}.
I
�
0 x? t a x?
For each t on I, we then have
lf"(t)I = 1-bf'(t) - cf(t)I

� lbl lf'(t)I + !cl lf(t)I
· ·
� kM + kM = 2kM.
Similarly,
l/<3\t)I = I-bf"(t) - cf'(t)I
� k · 2kM + kM
< (2k)2M.
We now claim that
for every n.
This is known for n = 1 and n = 2. And if it holds for two successive integers
n - 2 and n - 1, then it holds for the next integer n:
{1pn-1>(t)i < (2kt-2M
IJ<n+i>(t)I 1-bpn>(t) - cpn-ll(t)I
=
=> =
IJ<nl(t)I � (2kt-1M
� k IJ<n>(t)I + k IJ<n-ll(t)i
� k(2kt-1M + k(2kt-2M
= (2k)n-2(2k2 + k)M < (2k)nM ,

because k � 1. From this the theorem follows, because it gives
t
(2k M
IR (x)I ::5 Ix - aln+l'
n (n + 1)!
-
which obviously approaches O; in fact, the expression on the right is the (n + l)st
term of the series for
M e2klx al.
-
2k
It is not an accident that the series for f converges with the rapidity of an exponential
series: the solutions that we have found, so far, for our differential equation have been
combinations of exponentials, sines, and cosines; and we are about to find that these
are the only solutions.
Theorem 2. If/"+ bf'+ cf= 0, andf(a) =f'(a) = 0 for some a, thenf(x) = 0
for every x.
Proof Since pn> = - b pn 1> - cpn-2), it follows by induction that pnl(a) = 0
-
for every n, and so all the coefficients in the Taylor series are equal to 0. This gives
the result which was used without proof in the last problem set:
Theorem 3. If Ji and/2 are solutions of the equationf" + bf'+ cf= 0, and
fi(a) = f2(a) and f {(a) = f �(a) for some a,

then/1(x) =fix) for every x.
This is because the difference functionf =/1 - j; is also a solution, and
f(a) = f'(a) = 0.
The dimension theorem is now easy:
Theorem 4 (The dimension theorem). Let "f/' be the space of solutions of the equation
f" + bf'+ cf= 0. Then dim "f/' = 2.
Proof We found, in the preceding section, that every equation of this form has
two linearly independent solutions. Therefore dim "f/' � 2. It remains to show that
[ ]
every three solutions fi,f2,f 3 form a linearly dependent set. Consider the matrix
f 1(0 ) f {(O) f'{(O)

M = f2(0) f �(O) f �(O) .
fa(O) f �(O) f�(O)
This is not the Wronskian; it is not a matrix of functions but a matrix of numbers.
Therefore Theorem 4 of Section 13.6 can be applied to it: since the columns of Mare
linearly dependent, so also are the rows. Thus
for some scalars IX1, IX2, 1X3, not all equal to 0. Let
/ = 1Xi/1 + 1Xd2 + 1Xaf3.

Thenf(O) =f'(O) = 0, and sof(x) = 0 for every x. Therefore {/1,/2,/ 3} is linearly
dependent, which was to be proved.
All the results that we have been getting for equations of order 2 can be general
ized, in a straightforward way, tont. h-order equations, of the form
n-1
1<n) + I bJw = o.
i=O
As before, we tryf(x) = em"', and form the auxiliary equation

n-1
mn + I bimi = o.
i�O
If m1 is a root of the equation, with multiplicity k1 (so that (m - mi)k1 divides the
lefthand member), then the functions
are solutions, and form a linearly independent set. If IX + {Ji, IX - {Ji are a pair of
conjugate complex roots, with multiplicity k2, then the functions
e"-"' cos {Jx, xe"-" cos {Jx, ... , xk•-1e"-" cos {Jx
and
e"-"' sin {Jx, xe"-"' sin {Jx, ... , xk.-1e"-" sin {Jx
are solutions, and are linearly independent. Moreover, the total set of functions
obtained in this way forms a linearly independent set, and the number of elements
in the set is n, because n is the sum of the multiplicities of the roots of the auxiliary
equation. Therefore the dimension of the solution space is at least equal to n. As for
equations of order 2, it can be shown that all solutions of the equation are real
analytic; and matrix theory then furnishes a proof that every set of n + 1 solutions
forms a linearly dependent set. It follows that the dimension of the solution space is
exactly n.
Thus the results follow the pattern that we found for equations of order 2.
However, to derive them, in a reasonably efficient and natural way, requires new
theoretical ideas, and, in particular, a new kind of algebraic formalism. This theory
is best postponed to a systematic course in differential equations.
Meanwhile we consider what happens, in a linear differential equation with
constant coefficients, when the 0 on the right is replaced by a function. Consider,
for example,
f"(x) + 5f'(x) + 6f ( x) = e". (I)
We know how to find all solutions of the equation
f" + Sf' + 6f = O; (2)

the solution space is
"f/ = { IX1 e-2x + 1X2e-3'"}.
Equation (2) is called the reduced equation. In general, a linear differential equation
with 0 as its righthand member is called homogeneous.
Iff andf0 are solutions of (l), then
f"(x) + 5f'(x) + 6f(x) = e"' and f�(x) + 5f�(x) + 6f0(x) = e"',

and so by subtraction we get
Cf -
fo)" + (f - fo)' + 6(f -
fo) = 0.
Therefore the function f - Jo is a solution of the reduced equation (2). This means
that if we can find one solution of (1), then we can express all solutions of (1) in the
form
f(x) = 1X1e-2'" + 1X2e-3'" + fo(x);
every solution of (1) has this form, because f - f0 is of the form 1X1e-2"' + 1X2e-3"'.
If the function on the right is real-analytic, then there is a systematic scheme for
looking for solutions of a nonhomogeneous equation: we assume that
co
I(x) = L aixi,
i=O
and solve for the coefficients ai one at a time. But if the function on the right is simple,
then the method of trial and error may work faster and lead to a simpler formula. In
the example above, we try
l(x) = Ae"', J'(x) = Ae"', l"(x) = Ae"',
and try to find A so as to make I a solution:
Ae"' + 5Ae"' + 6Ae"' = e"',

12A = 1, A = 1\, l(x) =
/2e"'.
Therefore the solutions of (1) form a set
The set H does not form a subspace. A set of this kind is called a hyperplane. In
general, if if/ is a subspace of a linear space "f/", and lo is any point of"f/", then the set
H =
U +lo I I in if/}
is called a hyperplane. Note that every subspace is automatically a hyperplane,
because lo may be zero. The term hyperplane is suggested by the language of geometry
in Cartesian 3-space. If E is a plane through the origin, and P0 is any point of R3, then
the set
H = {P +P 0 I Pin E}
is a plane. (We use the prefix hyper because in vector spaces of higher dimension, the
dimension of a hyperplane may easily be greater than 2. The set H may, of course,
be of dimension 1 or 0; every line in R3 forms a hyperplane, under the above definition,
because every line through the origin forms a subspace. The same applies for a point,
although we rarely have any occasion to say so.)
Similar devices work for various other functions on the right in a nonhomogeneous
equation. For example, consider
J"(x) + 5l'(x) + 61(x) = sin x.

First we try
(?) l(x) = A sin x (?).
This gives
(?) - A sin x + 5A cos x + 6A sin x = sin x (?),
which is impossible, because {sin, cos} is linearly independent. Next we try
(?) l(x) = A sin x + B cos x (?),

which gives
-A sin x - B cos x + 5(A cos x - B sin x)
+ 6A sin x + 6B cos x = sin x
<=>(-A - 5B + 6A) sinx
+ ( -B + 5A + 6B) cos x sin x
{
=
5A - 5B 1 =
<=> 5A + 5B 0 =
This gives
fo(x) /o(sin x - cos x),
=
so that the hyperplane of solutions is

H {cx1e-2"' + cx2e-3x + /0(sin x - cos x)}.
=
PROBLEM SET 13.8
For each of the following equations, find the space ii" of solutions. Answers should be
in the form
11' = {aif1 + a2f2 + + an fn} · · ·
·
1. JC4l +4/<3> +6f" +4f' +f=0. 2. [<4> +J<3l - 3f" - Sf' - 2/ = 0.
3. 1<4> +2/<3l - 3[" - 4[' +4/ = 0. 4. 1<4> +2f" +f = 0.
5. pal - f" + f' - f= 0. 6. /(4) +J<3l - f' - f=0.
7. /(5) - 2/(3) +f' = 0.
For each of the following, find (a) the space 11' of solutions of the reduced equation, in
the same form as in the preceding problems, and (b) the hyperplane Hof solutions of the
given equation, in the form
H= {ai f1 + azfz + · · + anfn +/o}·

·
8. f" +f = e"' 9. f" +f=sin x 10. /" +f = 1

11. f" +f = cos x 12. f" +f=x + 1 13. f" +f=x
14. f" +f = x3 15. f" +f=e"' sin x 16. f" +f = x2 +1
17. f" +f=e"'cos x 18. f" +f=xe"' 19. /" +f=xe"' +e"'
20. f" +f = x sin x 21. f" +f = x cos x
22. f" +f = sin x +2x cos x
14 Functions of Several Variables
14.1 SURFACES AND SOLIDS IN R3
We recall, froll)._elementary geometry, the definition of a right cylinder. Given a

set Bin a plane E, the right cylinder with base Bis the union of all lines that intersect
B and are perpendicular to E.
Bz
\
I
I
,, I
,.--i
/ I
I I
IB
I
Hereafter in this chapter, when we speak of a cylinder we shall always mean a

right cylinder. Some other remarks are in order here.
1) The figures above suggest that a cylinder is a bounded figure, with a lower
base B1 and an upper base B•2 But this is merely a device for clarifying the meaning
of the pictures; according to our definition, cylinders are of infinite extent, in each of
two directions. In the same way, planes are unbounded, although we indicate them
in pictures by drawing parallelograms.
2) The base may be any set of points in a plane. If the base is a curve, as on the left
above, then the cylinder is a surface., If the base is a region, as on the right above,
then the cylinder is a solid. The definition applies to each of these cases in exactly the
same way. In each case, if the plane of the base is regarded as horizontal, then the
cylinder is the union of all vertical lines that intersect the base.
To avoid possible confusion, we may distinguish these cases by speaking of cylindrical

surfaces and cylindrical solids.
14.1 · Surfaces and Solids in R3 615
If the base is in the xy-plane, and is described by an equation in x and y, then the
same equation can be regarded as a description of the cylinder.
y
z
-1
/
)r -----
I
I
y I
I
/)- __
7c-- - y
__ _
/
x --- /
,,
- 1
x
On the left above we show the unit circle in the xy-plane; this is the graph of
the equation x2 + y2 = 1. In the center above, we see the same figure in perspective,
as it appears when we are about to draw in a z-axis. In the three-dimensional figure
on the right above we see the portion of the cylinder that lies iri the first octant. The
cylinder is the graph of the same equation x2 + y2 = 1 ; since the fquation imposes
no restriction on z, the graph includes the vertical line through each of its points.
To be more precise, the circle is
{(x,y) I x2 + y2 = I},
and the cylinder is
{(x,y,z) I x2 + y2 =I}.
The relations among these figures deserve careful examination. At the left above,
the tangents to the circle at the y-intercepts are horizontal,that is,parallel to the ,x-axis.
This should be true also in the perspective drawings at the center and right. Hence
the dotted guide lines. Similarly, the tangents to the circle at the x-intercepts are
vertical, that is, parallel to the y-axis. This should also be true in the perspective
drawings; it is indicated by the dotted guide lines. Often a correctly drawn figure
looks peculiar, unless you analyze it in this way. For example:
z
y
1 ----- - - 1 /
-71
/ /i
I
I
I / y I
I /
I
I /
I
I
/
I
-L
I
/--L
::..._
__
__
I y
_
x /
--t-< ___ __:_I _.
__ x /
----- /
-
x
y=x2, O�x�l}. {(x,y,z)iy=x2, O�x�l}.
To get a circular disk, instead of a circle, we use the inequality x2 + y2 � 1

instead of the corresponding equation. Using this as base we get the solid cylinder,
616 Functions of Several Variables 14.1
which is
{ cx,y,z) I x2 + y2 � 1}.
If we use the xz-plane or the yz-plane as the plane of the base, then the same
scheme works, in a similar way. For example:
z
z
z
x
x+z=l, O�x;;;l. {(x,y,z)lx+z=l, O;;;x;;;l}.
As usual, the figure is cut off at the ends, to clarify it in a pictorial sense. In its own
plane, the cylinder is an infinite strip, of width J2.
v'2
�->--'-��l��'----� '
If we had used the entire line x + z = 1, y = 0 as base, then the cylinder would
have been the entire plane
{(x,y,z) ix + z = l}.
This plane is parallel to the y-axis. Thus any plane parallel to one of the coordinate
axes can be described as a cylinder. In fact, for appropriate choice of the base plane,
any plane whatever can be regarded as a cylinder.
We have seen that cylindrical surfaces with their bases in the coordinate planes
are easy to describe by equations (if their bases are so describable). The next simplest
surfaces are the surfaces of revolution, whose areas we learned to compute in Section
7.5. Given a curve in, say, the yz-plane, we may rotate the curve about !he y-axis.
This generates a surface.
z
x
14.1 Surfaces and Solids in R3 617
The cross sections of the surface, in planes parallel to the xz-plane, are all circles,
with their centers on the y-axis. If the generating curve is described by a function, say'
z = f(y) G: 0,
then for each y0, the cross section in the plane y = y0 is the circle with center at
(0, y0, 0) and radius f (y0). Thus the cross section is the graph of the condition
Y =Yo,
and the surface of revolution is the graph of the equation
x2 + z2 = [f (y))2.
Two important special cases are as follows:
-a a
1) Consider the generating curve
z = -J a2 - y2, x = 0.
This is a semicircle. We rotate about the y-axis. The surface of revolution is the graph
of the equation
x 2 + z2 = [-}a 2 _ y 2) 2 = a2 _
y2
<=> x2 + y2 + z2 = a2.
This is as it should be, because the surface of revolution is the sphere with center
at the origin and radius a; it is easy to see by the distance formula that the sphere
must be the graph of the equation
-J(x - 0)2 + (y - 0)2 + (z - 0)2 = a
<=> x2 + y2 + z2 = a2.
2) Consider the line

z =my, x = 0,
in the yz-plane. When we rotate about the y-axis, we get a cone (that is, a conical
surface).
As usual, the figure shows only the first octant. The conical surface is the graph of
the equation
x2 +z2 = (my)2
<=> x2 - m2y2+z2 = 0.
If we had taken a line through the origin and rotated it about one of the other
coordinate axes, we would have gotten an equation of one of the forms
(a) -m2x2+y2+z2 = 0,
(b) x2+y2 - m2z2 = 0.
z z
/
/
t------
y
:t x
(a) (b)
Each of the surfaces that we have investigated so far has been the graph of an
equation of the second degree in x, y, and z, that is, an equation of the form
Ax2 +By2 +Cz2+Dxy+Exz +Fyz+Gx+Hy+Iz+J = 0,
where the first six coefficients are not all equal to 0. Using the method of rotation of
axes in a plane, as in Section 8.4, we can find out what the plane cross sections of such
surfaces are like. Let £0 be any plane, and let N be the normal line to E0 through the
origin. Let F0 be the plane which contains N and the z-axis, and let L be the line in
which F0 intersects the xy-plane. By a rotation of axes in the xy-plane, we can make
14.1 Surfaces and Solids in R3 619
L the x'-axis. The equations of the rotation are of the form
'
x = x' cos () - y' sin () and y = x' sin () + y cos ().
In the new coordinate system, the equation of the surface that we started with has the
form
A' x'2 + B' y'2 + C'z2 + D'x'y' + E' x'z + F'y'z + Gx' + Hy' + I'z + J = 0.
(Query: How do we know that the constant term is unchanged? And how do we
know that the first six coefficients are not all equal to 0?) In the x'z-plane, we now
perform another rotation of axes, in such a way that N becomes the new x-axis. The
equations for this rotation are of the form
• ,/. ,/.
x' x" cos cf> - z' sin cf>,
ff I
= Z = X Slll 'I' + Z COS 'I';
and in the x"-, y'-, z'-coordinate system, the equation of our surface is still of the
second degree, for the same reason as before. The plane F0 is the graph of an equation
of the form
x" = k,
where k is the distance between the origin and F0• To get the equation of the inter
section of the surface with F0, we should set x" = k in the equation of the surface.
This gives an equation of the second degree in y' and z'. By Theorem 2 of Section 8.4,
this means that every plane cross section of a second-degree surface is (a) a circle,
(b) a parabola, (c) an ellipse, (d) a hyperbola, (e) a point, (f) the empty set, (g) a line,
or (h) the union of two lines (either parallel or intersecting).
In particular, every plane cross section of a cone is a "conic section" of the sort
that we investigated in Chapter 8.
PROBLEM SET 14.1
Sketch the graphs of the following, in the first octant only. All the equations are to be
regarded as equations in (x, y, z). For example, x+y=1 is the equation of a plane,
x2 + y2 - 1 =O is the equation of a cylindrical surface,and so on.
1. x+y =1 2. x+z=l 3. y - z=1

4. x2+z2=1 5. x2+y2 = 1 6. y2+z2=4
7. z=y2, 0�y �1 8. x =z2, O�z�l
9. x =4y2, 0�y � 1 10. (x2/4)+y2=1, x,y � 0
11. (,y2/4)+z2=1, y, z � 0 12. (x2/4)+ (z2/9)= 1, x, z � 0
13. x2+ (z2/4)=1, x, z � 0 1 4. x = ly - II
15. lxl=y + 1
Find equations for the surfaces described as follows,and sketch in the first octant.
16. The graph of z=sin y, 0�y � TT is rotated about the y-axis.
17. The graph of y+z=1 is rotated about the y-axis.
18. The same graph is rotated about the z-axis.
19. The line which passes through the origin and the point (1, 3, 1) is rotated about the
y-axis.
20. The same line is rotated about the z-axis.
21. The same line is rotated about the x-axis.
22. The graph of y =e"', 0�x � 1 is rotated about the y-axis.
23. The same graph is rotated about the x-axis.
24. The graph of y =lxl+ 1 is rotated about the x-axis.
25. The graph of y =cos z,0�z� ;;2, is rotated about the z-axis.
26. The same graph is rotated about the y-axis.
14.2 THE QUADRIC SURFACES
A quadric surface is a surface which is the graph of an equation of the second

degree in x,y, and z. Thus all the surfaces discussed in the preceding section are
quadric surfaces. We now consider less simple cases.
1) Spheres and ellipsoids. Obviously the graph of the equation
x2 + y2 + z2 = a2
is a sphere of radius a. More generally, the graph of the equation
x2 Y2 z2
-+-+-=1
a2 b2 c2
is called an ellipsoid.
The sketch on the left shows the entire surface. On the right we show only the
part of the surface that lies in the first octant. Such partial sketches are much easier
to draw, and sometimes they are actually easier to interpret and to use.
14.2 The Quadric Surfaces 621
It is easy to see, algebraically, that every cross section of an ellipsoid, parallel to

one of the coordinate axes, is an ellipse, a circle, a point, or the empty set. For
example, setting x = x0 we get
Y2 z2 x2
-+-=1-_Q-k.
b2 c 2 a2 -
Fork > 0 the cross section is an ellipse or a circle; for k =0 the graph is the point
(x0, 0, 0); and fork < 0 the graph is empty. Similarly for y = y0 or z = z0•
2) Elliptic and circular cones. We found in the last section that the graph of the
equation
x2 - m2 y2 + z2 =0
is a circular cone. More generally, the graph of
x2 z2
--y2 +-=0
a2 b2
is a cone, either elliptic or circular.

z
/yo
/
--y
/
x /
x /
/
The figure on the left shows the entire cone, and the one on the right shows only the
portion that lies in the first octant. The cross section in the yz-plane is obviously a
pair of lines, because it is the graph of
z2
x =0, y2 =-.
b2
Similarly, the cross section in the xy-plane is a pair of lines, because it is the graph of
x2
z = 0, y2 =
2·
a
The cross section in the plane y = Yo is the graph of
x2 z2
Y =Yo, + = Y2o
·
a2 b2
This is a point for Yo = 0, and is a circle or an ellipse for Yo � 0. Note that the cross
section in the plane x = x0 (x0 � 0) is the graph of
which is a hyperbola.
3) The hyperboloid of one sheet. Consider the equation
x2 y2 z2
-+- --=l.
a2 b2 c2
Setting z = z0 and transposing, we get
x2 y2 z2
-+-=l+_Q=k2:::
- 1.
a2 b2 c2
Therefore all horizontal cross sections of the graph are ellipses. Rewriting in the form
x2 y2
=
a2k + b2k 1'
we see that as lzol increases the ellipses get bigger, but their shape does not change.
I
I
I
y I
I
)- - --,_
__
__
y
x x
14.2
The Quadric Surfaces 623
The cross sections in the other coordinate planes are hyperbolas; they are the graphs
of the conditions
y2 z2
x = 0, - -- = 1'
b2 c2
x2 z2
y= 0, --- =l.
a2 c2
4) The hyperboloid of two sheets. Consider the equation
x2 y2 z2 v2 z2 x2
-- - - - = 1 '--+ = - - 1. -
a2 b2 c2 b2 c2 a2
Again we investigate cross sections. For \x0\ < a, the cross section in the plane
x = x0 is empty; for x = ±a, the graph is a point; and for x� > a, the graph is an
ellipse (or a circle), being the graph of
y2 z2 x2
X = X0, -+ -
=
_Q - 1 > 0.
b2 c2 a2
The cross sections in the xz-plane and the xy-plane are obviously hyperbolas.
5) The hyperbolic paraboloid. This one is hard to visualize and hard to sketch. It is
the graph of the equation
Y2 x2
CZ= - - - (c '=r6 0).
b2 a2
We give the sketch for the case a = b= c = 1. Thus the equation of the surface
becomes
To sketch, we use the cross sections in the planes z = -1, z = 1, x = 0, andy = 0.

(Rather oddly, it is a bad idea to draw the cross section in the plane z = O; this is a
pair of intersecting lines, and when we indicate it correctly, the result is very hard to
interpret pictorially.)
z z
x=O, z=y2. y=O, z=-x2.
Using these cross sections in a perspective drawing, we get the result shown
below.
For other values of a and b, we get hyperbolas of different shapes in the hori
zontal cross sections. And when the sign of c is changed, the effect is to reflect the
surface across the xy-plane.
PROBLEM SET 14.2
Sketch the graphs of the following equations, and identify the surfaces.
�
1. x2+-+-= l
� �
2. x2--+ - = 1
�
4 9 4 9
2 z2 2 z2
y y
3. x2+--- =1 4. x2----= 1
4 9 4 9
2 z2 2 z2
y y
5. x2 - - - - =0 6. x2+---=0
4 9 4 9
2 z2 2 z2
y y
7. -x2+---=0 8. x2 ____ -1
4 9 4 9
2
10. z2 =x2+t
y
9. z=x2--
4 4
14.2 The Quadric Surfaces 625
y2 y2
11. z =- - x2 12. z2 = - - x
4 4
13. z = xy [Hint: This is easier to sketch if you first identify it.]

14. z2 = x2 + 2xy + y2
15. Consider the hyperbolic paraboloid which is the graph of the equation
y2 x2
cz = - (c ¥: 0).
b2 a2
Let (x0, y0) be any point of the xy-plane. For each cc, consider the path whose coordinate
functions are
X = Xo + t COS cc, y = y0 + t sin cc.
(In effect, we have set up a coordinate system on a line L through (x0, y0), in such a
way that L becomes the t-axis.) Now consider the path in space defined by the coordinate
functions
X = Xo + t COS cc,
y = y0 + t sin cc,
1 y2( x2
z =- - - -
) 1 1
= - ( y0 + t sin cc)2 - - (x0 + t cos cc)2•
c b2 a2 b2c a2c
What kinds of curves can be the loci of such paths?

16. A surface S is said to be ruled if for each point P of S there is a line L which contains P
and lies entirely in S. (Thus every cone is automatically a ruled surface, and so also
is every cylinder.) Show that every hyperbolic paraboloid is a ruled surface.
* 17. Show that any hyperboloid of one sheet is a ruled surface (see Problem 16 for definition).
18. Find the volume of the region inside the graph of the equation
x2Ja2 + y2/b2 + z2 /c2 = 1 .
19. Find the volume of the solid defined by the inequalities
0 � y � 1, -y � x � y,
[This is the solid which lies (a) above the xy-plane, (b) below the hyperbolic paraboloid
z = y2 - x2, and (c) between the planes y = 0 and y = 1.)
20. Find the volume of the solid which lies between the planes z = 0 and z = 1, and
inside the one-sheeted hyperboloid
2
2 y 2
x + 4 = z + 1.

2 2
x y 2
4 + 9 = z + 1.
22. Find the volume of the solid which lies between the planes z = 1 and z = 2 and the
two-sheeted hyperboloid
2 2 2
z = x +y + 1.
14.3 FUNCTIONS OF TWO VARIABLES.

SLICE FUNCTIONS AND PARTIAL DERIVATIVES
So far, most of the functions that we have been studying .1ave been of the following
types.
1) Functions whose domains are sets of real numbers. In these cases, the' domain
Dwas usually an interval.
2) Functions of one vector space into another. These were always linear, and were
referred to as linear transformations.
3) Sequences. (Every sequence a1, a2, • • . can be regarded as a function, with

f(i) a;, and sometimes it is useful to think
= of sequences in this way. You may be
able to recall such an occasion.)
We shall now investigate functions of the type
f: D-+ R,
where Dis a region in an inner-product space, andfis not necessarily linear.
'
I
I
,--1=- ---- .... ,
. I I
I I
I
c=0
I
l �I I
1
l: : I
-- y
•
x
p D
z=f(P).
14.3 Functions of Two Variables 627
Suppose that a rule is given under which to each point P of D there corresponds a
real number z. We then say that we have a function
f: D-+ R.
As usual, the number corresponding to the point P of D is denoted by f(P). If D is

a region in R2, and P = (x, y), then we may writef(x, y) forf(P); in this case,fis
called a function of two variables, by which we mean two real variables. In this case,
the graph offis
{(x, y, z) I (x, y) in D and z = f(x, y)},
or, equivalently,
{(P, z) J P in D and z = f(P)},
as in the figure above.

Roughly speaking,fis continuous at a point P0 of D if
P ::::,; P0 => f(P)::::,; f(P0).
As for a function defined on an interval, the limit off is defined, more generally,
for the case in which P0 does not necessarily lie in D. In this case,
lim f(P) = L
P-+Po
means that
and P ¥- P0 => f (P) ::::,; L.
To make these ideas precise, we interpret P ::::,; P0 to mean that llP - P0ll is small,
and we interpretf(P)::::,; L to mean that lf(P) L I is small. This gives the following
-
definition :
Definition. Let D be a region in an inner-product space Y, let fbe a function D-+ R,
let P0 be a point of"f/ (not necessarily lying in D), and let L be a real number. Suppose
that for every E > 0 there is a o > 0 such that
0 < llP - Poll < 0 => lf(P) - L I < E.

Then
lim f(P) = L.
P-+Po
Note that this is a quite straightforward generalization of the definition of the

statement limx-.xJ(x) = L, with Ix - x0\ replaced by llP - P01\. Following, in the
same way, the pattern of our earlier work, we state:
Definition. Let fbe a function D-+ R, and let P 0 be a point of D. If
lim f(P) = f(Po),

P-+Po
then/ is continuous at P0•

The elementary theory of limits can now be generalized very easily.
Theorem 1. Let D be a region in a vector space "/', and let f and g be functions
D-+R.If
lim f(P) = L and lim g( P) = L',
P-+Po P-+Po
then
lim [f(P) + g(P)] = L + L' and lim [f(P)g(P)] =LL'.
P-+Po P-+Po
If L' � 0, then
f (P)
lim = !::__ •
P-+Po g(P) L'

The proofs of these results are exactly the same as in Appendix B; we merely
need to translate the old proofs into the new language in the same way that we have
just translated the definitions. In this process, there is no advantage in using coordi
nates, writingf(x, y) forf(P), g(x, y) for g(P), and so on, even if domains in R2 are
the only ones that we are concerned with. In fact, the use of coordinates in these
particular proofs merely complicates the notation and obscures the ideas. But for
other purposes, we need to know how continuity is related to coordinates, as in the
following:
Theorem 2. Let D be a region in R2, let/be a function D---+ R, and let P0 = (x0, y0)
be a point of R2• Suppose that for every E > 0 there is a o > 0 such that
Ix - x01 < o and IY - Yol < o => lf(x,y) - LI < E.
Then
lim f(P) = L.
P-+Po
To see this, we merely need to examine the geometric meaning of the inequalities
on the left (preceding the =>).
y
m
�
The inequalities Ix - x01 < o, IY- - Yol < o hold inside the square, and the m
equality llP - P0ll < o holds inside the inscribed circle. Therefore
llP - Poll < o => Ix - x01 < o and IY - Yol < o
=>
lf(x,y) - LI < E;
and so
lim f(P) = L.
P-+Po
From this it follows, as for functions of one variable, that for functions of two
variables, continuity is preserved under composition of functions:
Theorem 3. Let f, g, and h be continuous functions of two variables. For each
P = (x, y), let
<f>(x, y) f[g(x, y), h(x, y)].
=
(Equivalently, <f>(P) f[g(P), h(P)].) Then</> is continuous.

=
We give the proof in outline:

(1)
=> x � x0 and Y �Yo (2)
=> g(x, y) � g(x0, Yo) and h(x, y) � h(x0, Yo) (3)
=> f[g(x, y), h(x, y)] �f[g(x0, y0), h(x0, y0)]. (4)
The reasons for these implications are as follows:
1) P = (x, y) and P0 =(x0, Yo).
2) Any circle llP - P01\ < o lies in its circumscribed square, as in the figure above.
3) g and h are continuous.
4) f is continuous, and Theorem 2 holds.
To verify that various functions defined by algebraic formulas are continuous,
we need the following:
Theorem 4. Let </> be a function of one variable; and for each x and y (in a domain D)
let
f(x, y) = </>(x).
If <f> is continuous, thenf is continuous.
Note that the graph of such anf is always a portion of a cylinder, like this:
I
)-. __
/
/ / y
/
/
/ /
,------k:___j,/
x
Geometrically it is obvious that the cylinder must be a continuous surface, and an

algebraic proof is almost equally easy. Given (x0, y0) in D, and E > 0. We know
that there is a o > 0 such that
Ix - Xol < O => l<f>(x) - <f>(xo)I < €.
Therefore
Ix - x0\ < o and IY - Yo\ < o => lcf>(x) - cf>(xo)I < E

<=> lf(x,y) - f(x0,Yo)\ < E.
Therefore, by Theorem 2,f is continuous.

These theorems enable us to infer that various simple functions, which obviously
ought to be continuous, really are. Consider, for example,
x2y2 _
xa y
f(x, y) cos •
1 + x 2 + y2
=
By Theorem 4 the functions
x2, y2, xa, y, 1,

are all continuous, considered as functions of two variables. By Theorem 1 it follows
that x2, y2 , x2y2, and x3y are continuous. Therefore x2y2 - x3y and 1 + x2 + y2
are continuous (Theorem 1 again). By Theorem 1, (x2y2 - x3y)/(1 + x2 + y2 ) is
continuous. By Theorem 3, so also is f
We investigated the quadric surfaces by examining cross sections. We do the same
for the graphs of functions of two variables. Given a region D in R2, and a function
/: D-+R,
take a fixed y0,and consider the intersection of the graph with the plane y = y0•
x
/
The cross section is a curve, and is the graph of a function cf>v.: for each x for which
(x,y0) is in the domain of/, we have
ef>v0(x) = f(x, Yo)·

Such a function is called a slice function off If the slice function is differentiable at
x0, then its derivative is denoted by
Geometrically, f,,(x0, y0) is the slope of the surface in the x-direction at the point
(xo, Yo).
Naturally, we can restate this in purely analytic terms, without any reference to
the geometry, and without mentioning a slice function. We can state:
Definition
1. f(x, Yo) - f(xo, Yo)
(
f"'Xo, Yo) - 1m ,
_
x-+xo x - x0
if such a limit exists.
The function f,, is called the partial derivative off with respect to x.
Standard differentiation formulas give us partial derivatives very easily. For
example, given
f(x, y) = x3 + xy2 + x2y + y4,

we sety = y0 , to get
This gives
cp�0(x) = 3x2 + yg + 2xy0 = f,,(x, Y0 ).

Dropping the subscript 0, we get
f,,(x, y) = 3x2 + y2 + 2xy.
The formal rule is simple: regard y as a constant, regard x as the dummy letter
defining the function, and differentiate.
We can equally well consider slice functions in the y-direction, setting x = x0•
The derivative of the slice function is now the partial derivative off with respect toy.
More precisely:
Definition
) 1. f(xo, y) - f(Xo, Yo)
f,v(Xo, Yo - im '
_
v-+'llo Y - Yo
if such a limit exists.
The partial derivatives f., and .fv, once we get them, are also functions; and their
partial derivatives are defined in the same way. For example, consider
f (x, y) = x3 + xy2 + x2y + y4.

Here
f.,(x, y) = 3x2 + y2 + 2xy,
fv(x, y) = 2xy + x2 + 4y3•
The partial derivative of f., with respect to x is denoted by .fc.,. Thus
f,,,,(x, y) = 6x + 2y.
Similarly
h v(x, y) = 2x + l2y2.
Now !rev is the partial derivative off., with respect toy. We have
fxv =
Dv(3x2 + y2 + 2xy) = 2y + 2x.
And similarly
hx = D,,(2xy + x2 + 4y3) = 2y + 2x.
Note that while fxv and hx turned out to be the same function, they were not defined
in the same way, and they were not arrived at by the same process. Therefore the fact
that.fcv fy,,, for this particular function f, must be due either to an accident or to a
=
nontrivial theorem.
Finally, some remarks on notation. Often, people write
of of
for f,,, for f11,
ox oy
o2f o2f
for fxx• for fvv•
ox2 oy2
02! o2f
for fxv• for fvx•
oy ox ox oy
and so on. Note, in the last line, that in the symbols.fcv and.fv.,, the letters indicating
partial differentiation accumulate on the right; while in the symbols
o2f 02/
and
oy ox ox oy
the letters indicating partial differentiation accumulate on the left. Thus
oY oY
'.l
'.l '.l f vx and '.l '.l fxyy·
uxuyuy 0yuyux
= v =
·
Note that in the CJ-notation, the symbols for higher derivatives look like "prod
ucts" of "factors" of the types of ox, o/oy. Thus
a 0 a o3f
fm f (x y) =
oy oy ox , oy oy ox
=
This is why the symbols accumulate on the left instead of the right.
PROBLEM SET 14.3
Citing the theorems of this section, at the points where you need them, show that each
of the following functions is continuous.
1. f(x,y) = vx2 + y2 2. f(x,y) = Vx4 + y4 + 1 3. f(x,y) = -v'x2 + y2
xy
4. f(x,y) [(x,y) � (0, O)] 5. f(x, y) sin (x2y + y2x)
x 2 + y2
=
=
x2 _
y2 xy
6. f(x,y) [(x,y) � (0, O)] 7. f(x,y)
x + y2
2 x + y2 + 1
2
= =
sin x cosy - 1
8. f(x,y) - (y � 0) 9. f(x,y) (x � 0)
y x
= =
sin xy cosy
10. f(x,y) -2- [(x,y) � (0, O)] 11. f(x,y) [(x,y) � (0, 0)]
x -
+ y2
=
x-2--
+ y
2
=
Problems 12 through 22. For each of the functions f given in Problems 1 through 11,
find fx, fv, fxy, and fvx·
23. Obviously the definition off, in Problem 4, is valid only for (x,y) � (0, 0). Is it
possible to give a separate definition off (0, 0), in such a way that the resulting function
is continuous ? That is, is there any such thing as
. xy
( ?) lim ---
2 + y2 ?
(x,vJ-<o,OJ X
Why or why not ?

24. Same question, for the function defined in Problem 10.
25. Same question, for the function defined in Problem 11.
26. By a polynomial in x and y we mean a function f which is the sum of a finite number
of terms of the form aiixiyi, where the a;;'s are constants. Thus
n n
f(x, y) = .2 .2 a;;xiy i.
i=O i=O
Show that if f is a polynomial, then fxy = fvx·

27. Let us say, for short, that a function/is regular if/x v = fv:v Show that if/is regular
then so also is f2.
28. Show that if f is regular and positive, then vJ is regular.

29.· Show that if/is regular, then so also is/3•
30. Show that if/is regular and never zero, then (1//) is regular.
31. Show that if f and g are regular, then so also is f/g, at every point (x,y) where
g(x,y) � 0.
32. Given a function f, and a point (x0,y0). Let t::.f
. = f(x,y) - f(x0,y0). For the function
f(x,y) = x2y + y2x,

show that 11/ can be expressed in the form

11/ =
f,,(x0,y0) 11x + fy(x0,y0) 11y + E(11x,11y) 11x + F(11x, 11y) 11y,
where Eand Fare functions such that
Jim E(l1x,11y) = Jim F(11x,11y) = 0.
(Ll.x,Ll.v)---+(0,0) (Ll.x,Ll.ul---+(O,O)
*33. Write a complete proof of Theorem 3, showing that for every£ > 0 there is a c5 > 0
such that ...
14.4 DIRECTIONAL DERIVATIVES AND DIFFERENTIABLE FUNCTIONS
The partial derivatives of a function f: D---+ R, where D is a region in R2, were

defined as the derivatives of the slicefunctions z =f(x,y0) and z =f(x0,y); and we
got the slice functions by taking cross sections of the graph off, parallel to the yz
plane and the xz-plane. We proceed to consider more general slice functions, obtained
by taking cross sections in any vertical plane whatever.
z
y
Yo
/
y
/
XQ /
QI_
x t(x, y)
x
Given a function
f: D---+ R
with continuous partial derivatives.fx and.fv. Let (x0,y0) be a point of D, let L be a

directed line through (x0, y0), and let a be the angle between the positive direction on
L and the positive direction on the x-axis. Then L is the locus of the path
X = Xo + t COS CJ.,
y =Yo+ t sin a.
We now form the composite function

cp,,(t) =f (x0 + t cos a, Yo+ t sin a) .
We call <Pa the slice function in the direction a. If <Pa has a derivative at t = 0, then
cp�(O) is called the derivative of f in the direction a, and is denoted by f,,(x0,y 0).
That is,
1. f(x0 + t cos a, Yo + t sin a) - f(x0, Yo)
J+a(Xo, Yo) = lffi .
t-o t
14.4 Directional Derivatives and Differentiable Functions 635
For ex = 0, the slice is parallel to the x-axis, and so it ought to be true that /0 =fz.
And this is true: for ex = 0 we have cos ex = 1, sin ex = 0, and
. f(x o+ t,Yo) - f(xo, Yo)

+ (Xo,Yo) = 1im
Jo
t-+O t
. f(x0 + �x,Yo) - f(xo, Yo)
= lim =f.,(Xo,Yo) ·
lix-+O �X
Similarly, for ex = 7r/2 we have
cos iX = 0, sin ex = 1,
and
f,,12(Xo,Yo) = fv(Xo,Yo).
We now want a general formula forfr,. As a guide to what we should be aiming at,
we consider first the simplest case, in which/is linear, with
f(x,y) =Ax+ By+ C.

For each point (x0,y0) and each ex, we have
</>a.(t) =f(x0+ t cos ex, Yo+ t sin ex)

= A(x0+ t cos ex) + B(y0+ t sin ex) + C,
</>�(t) = A cos ex+ B sin ex.
Since
fx(x,y) =A, fv(x,y) = B ,
for each x and y , it follows that these equations hold at the particular point (x0, y0);
and so we have the following:
Theorem A. If f is a linear function of two variables, then

fa.(xo,Yo) = fx(Xo,Yo) cos ex+ fv(xo,Yo) sin ex,
for each ex.
We shall now see that this formula holds under much more general conditions,
when/is not necessarily linear, but is "approximately linear near (x0,y0)," in a sense
which we shall define presently.
We recall that if f is a differentiable function of one variable, the difference
�f = f(x0 + �x) - f(x0)
can be expressed in the form
�f = f'(x0) �x+ E(�x) �x,

where
lim E(�x) = 0.
t.x-o
We want to get an analogous expression for the difference
�f = f(xo+ �x,Yo+ �y) - f(xo,Yo).

It is fairly easy to find out what form the formula has to take if it exists at all. If f is
linear, with
f(x,y) =Ax+ By+ C,
then
!J.f=Ax+ By+ C - ( Ax0+ By0+ C)
=A(x - x0)+ B( y - Yo)
=A !J.x+ B !J.y.
Here, as before,
A=f,,(x,y), B=fu (x,y);
for a linear function, the partial derivatives are simply the coefficients of x and y.
This suggests that our expression for /J..f ought to take the form
/J..f =fx(Xo,Yo) /J..x+ fv(xo,Yo) /J..y+ [- - -],

where [- - -] is, we hope, a function which approaches 0 very rapidly as (/J.x,/J..y) _,..
(0, 0). Consider, for example,
f(x,y)=x2+ xy + y2.
Here
/J..f=(Xo+ /J..x)2+ (xo+ /J..x)(yo+ /J..y)+ (Yo+ /J..y)2 - xg - XoYo - yg

= 2x0 /J..x+ !J.x2+ x0 /J..y+ Yo /J..x+ /J..x /J..y+ 2y0 /J..y+ /J..y2
=(2xo+ Yo) /J..x+ (x0+ 2y0) /J..y+ [/J..x2+ /J..x !J.y+ /J..y2]
= fxCxo,Yo) /J..x+ fu(xo,Yo) /J..y+ [- - -],
where [- - -] _,.. 0 rapidly, as /J.x, /J.y _,.. 0.
We now attack the general problem, for functions with continuous partial
derivatives fx,fv. We can get from (x0,Yo) to (x0+ /J..x,Yo+ /J..y) by moving first
vertically and then horizontally. Algebraically,
!J.j =f(xo+ /J..x,Yo+ /J.. y) - f(xo,Yo+ /J..y)+ f(xo,Yo+ /J..y) - f(xo,Yo).

y
Yo+Ay ---- -.- --- -- -- °1
+ II
I
Yo ___ __. I
I I
I I
��--+-�x��-1��---
x0 x0+Ax
We apply the mean-value theorem MVT to the function
cp(x)=f(x,Yo+ /J..y),
on the interval from x0 to x0+ /J..x.
x0 x! x0+t.x
z= q,(x)
The mean-value theorem says that there is an .X, between x0 and x0 +Lix, such
that
This means that
f(x0 + Lix,Yo + Liy) - f(x0,Yo + Liy) =

f,Jx,Yo + Liy)Lix.
��+----''-----'-��---'-�y
Yo Y Yo+t.y
z=Y,(y)
Similarly, we apply MVT to the function
7P(Y) =
f(xo, y),
on the interval from y 0 toYo + Liy. Thus
7P(Yo + Liy) - 7P(Yo) = 7JJ'( ji)Liy,

for some ji between y 0 and y0 + Liy; and so
f(xo,Yo + Liy) - f(xo,Yo) =

f/xo, ji) Liy.
Fitting these two results together, we get
Lif =
f.,(x,Yo + Liy)Lix + fu(x0, ji) Liy,
where xis between x0 and x0 + 6.x, and ji is between Yo and y0 + Liy.
We are now almost finished. For each 6.x, Liy, let
E1(Lix, 6.y) = f.,(x,Yo + 6.y) - fx(xo,Yo),

E2(Lix, Liy) = fu(xo, ji) - fu(xo,Yo).
Then
and
Therefore
b.J = f.,(xo, Yo) b.x + fixo, Yo) b.y + Ei(b.x, b.y) b.x + E2(b.x, b.y) b.y.
Note that
lim E1(b.x, b.y) = 0,
(4x,4·y)-+ (0,0)
lim E2(b.x, b.y) = 0,

(4x,4y)-+(0,0)
because fx and fy are continuous. Thus, if the partial derivatives off are continuous,
then b.f is well approximated by the linear function f.,(x0, y0) b.x + fu(x0, y0) b.y.
For functions of two or more variables, the idea of approximation by linear functions
is used as the definition of differentiability. To be exact:
Definition. Let D be a domain in R2, let/be a function D --+ R, and let P 0 = (x 0,y0)
be a point of D. Suppose that there is a linear function
L(b.x, b.y) = A b.x + B b.y,

and functions E1 and E2, defined in a neighborhood of (0, 0), such that
b.f = A �x + B �y + E1(�x, �y)�x + E2(�x, b.y), (1)

and
lim E1(b.x, b.y) = lim E2(b.x, b.y) = 0. (2)
(4x,4y)-+ (0,0) (4x,4y)-+(0,0)
Then f is said to be differentiable at (x0, y0).
This definition was modeled on the preceding discussion, and so we have already
proved the following theorem.
Theorem 1. If/., andfv are continuous at (x 0, y0), then/is differentiable at (x0,y0),

with
b.J R:! A b.x + B b.y = f.,(xo, Yo) b.x + fu(xo, Yo) b. y.
For functions of one variable, we defined the differential to be the linear function
df = df(b.x) = J'(x0) b.x,

which gives good approximations of b.f = f (x0 + b.x) - f (x0). For functions of
two variables, the differential is defined analogously:
Definition. If f is differentiable at (x0, y0) , then
df = df (�x, b.y) = f.,(xo, Yo) b.x + fu(xo, Yo) b.y.

The definition of differentiability is complicated to state, but it is easy to use.
It gives us the formula that we wanted, for directional derivatives:
Theorem 2. If f is differentiable at (x 0, y 0) , then for every direction rx,
f,,,(xo, Yo) = f.,(x0, Yo) cos rx + fv(x0, y0) sin rx.

Proof By definition,
-r _ 1. f(x0 + t cos ix, Yo + t sin ix) - f(x0, y0)

Ja(Xo, Yo) - 1m
t-+0 t
This has the form
. f(x 0 + Ax, Yo + Ay) - f(xo, Yo) 1. Af

l 1m = 1m-,
t-+0 t t-+O t
where
Ax = t cos ix, Ay = t sin ix.
Since f is differentiable,
Af = A Ax + B 11y + £1 Ax + E2 11y
= f,,(x0 , y0) 11x + fu(x0, y0) 11y + E1 11x + E2 11y
= f,,(x0, y0)t cos ix + fu(x0,y0)t sin IX + E1t cos IX + E2t sin ix.
Therefore
+ Ei(t cos ix, t sin ix) cos ix + E2(t cos ix, t sin�) sin ix,
and so
!1f
faCxo,Yo) = l im
t-+0 t
= f,,(x0, Yo) cos IX + fy(x0,Yo) sin ix + 0 · cos ix + 0 · sin ix,
In the first five problems in the following problem set, you are asked to
"verify directly" that certain functions are differentiable at certain points. In each
of these cases, you should go through an elementary calculation to express 11/ in the
form
For example, given

2
f(x ,y) = x y (xo.Yo) = (1, 1),
you would proceed as follows:
11/ = f(x0 + 11x, Yo + 11y) - f(xo, Yo)

= (1 + /1x)2(1 + !1y) - 1
2
= 1 + 2 11x + /1x + 11y + 2 11x 11y + /1x2 11y - 1
2
= 2 11x + 11y + (11x + 2 11y) 11x + (/1x ) 11y.
The answer can now be written in the form

Note that other choices of E1 and E2 would have worked just as well. For example,
fif = 2 fix + fiy + (fix) fix + (fix2 + 2 fix) fiy.

Therefore each of the first nine problems below has more than one right answer.
PROBLEM SET 14.4
Verify directly that each of the following nine functions is differentiable at the indicated
point.
1. f(x, y) =xy, (2, 1) 2. f(x, y) =x2y2, (-1, -1)

3. f(x, y) =xy2, (1,1) 4. f(x, y) =x3, (0,0)
5. f(x, y) =x2 - y2, (1,1) 6. f(x, y) =y4, ( - 1 1)
,
7. f(x, y) =x2 + y2, ( -1, 1) 8. f(x, y) =4x2 + y2, (1, 1)

9. f(x, y) =x2 - 4y2, (1, - 1 )
10. Given f (x, y) = v x2 + y2, (x0, y0) = (1, 1), get a general formula for fa(x0, y0).
For which ex does f.. (x0, y0) take on its maximum value? For which ex do we get the
minimum value?
11. Same question, forf(x, y) =x2 - y2, (x0, y0) =(1, 1).
12. Same question forf(x, y) = xy, (x0, y0) =(1, 1).
13. Suppose that f has a directional derivative/o: in every direction ex, at a point (x0, y0).
Is it possible that f.. (x0, y0) > 0 for every ex? Why or why not? (Try to answer this
one merely on the basis of the definition of fa, without appealing to Theorem 2.)
14. Show that if fa(x0, y0) =0 for every ex, then fx(x0, y0) =fy(x0, y0) =0.
15. Give an example to show that the following "Theorem" is false:
Theorem(?) "Given/: D-+ R. Iff,,(x, y) =fy(x, y) =0, for every (x, y) in D, then
f is a constant."
16. Show that the following theorem is true:
Theorem A. Given
z = f(x, y) (a < x < b, c < y < d).
If fx(x, y) =fy(x, y) =0, for every (x, y) in the given domain, then f is a constant.
Here we are requiring that the domain be a rectangular region with sides parallel to
the x- and y-axes.
d ---- r ----- ,
i I
I J
c ----+------ �
I I
I I
I I
I I
a b
14.5 The Chain Rule for Paths 641
17. Theorem A, stated in Problem 16, is artificially special; it does not apply, as it stands,
to domains like the following:
y y y
Find a way of describing the property of D that is really needed in the proof of Theorem
A, and prove a theorem which uses your more general hypothesis.
14.5 THE CHAIN RULE FOR PATHS
In the preceding section, we defined the directional derivative fa as the derivative off
along a linear path
X = g(t) = Xo + t COS IX,
y = h(t) = Yo + t sin IX,
and we found that if f is differentiable, then
f a(Xo, Yo) =
fx(Xo, Yo) cos IX + fixo, Yo) sin IX.
This result can be generalized, so as to apply to derivatives along paths which are not
necessarily linear. Suppose that a path Pis defined by a pair of coordinate functions.
Strictly speaking, we should write
x = g(t), y = h(t), a< t< b.
But it is easier to keep track if we use the letters x and y as the names of the co
ordinate functions. Thus we write
x = x(t), y = y(t), a< t< b.
�p
Let F: D---+ R be a differentiable function, and suppose that the locus of the
path Plies in D. We can then form the composite function
cf>(t) = F(x(t),y(t)),
and we have the following theorem:
Theorem 1 (The chain rule for paths). Let cf>(t) =

F(x(t),y(t)). If F, x(t), and y(t) are
differentiable, then cf> is differentiable, and
cf>(t) = F.,(x(t),y(t))x'(t) + Fu(x(t),y(t))y'(t).
F(x,y) = x2y + y2x,

x(t) = cost,
y(t) = sin t.
Here the locus of the path P is a circle. And
cf>(t) = cos2 t sin t + sin2 t cos t,

cf>'(t) =
-2 cost sin2 t + cos3t + 2 sin t cos2 t - sin3 t.
Since
F'"(x,y) = 2xy + y2 and F u(x,y) = x2 + 2xy,
Theorem 3 gives us
cf>'(t) =
(2 cost sin t + sin2t)(-sin t) + (cos2 t + 2 sin t cost ) cost,
We proceed to the proof. Take a fixed t0, and let
Xo =
x(to), Yo =
YUo),
Lix =
x(t0 + Lit) - x(t0),
Liy = y(t0 + Lit) - YUo)·
In this notation,
"-' 1. F(x0 + Lix,Yo + Liy) - F(x0,y0)

't' (t0) - Im .
_
M�o Lit
(Re-examine the definition of cf>.) Now
F(x0 + Lix) - F(x0, Yo) =

LiF
=
F,ix0, y0) Lix + Fy(x0, y0) Liy + E1 Lix + E Liy,
2
where
as (Lix, Liy)---+ (0, 0).
Therefore
14.5 The Chain Rule for Paths 643
Therefore
�F
cp'(t0) = lim
M-+O �t
= F,,(x0, y0)x'(t0) + Fu�x0, y0)y'(t0) + 0 · x'(t0) + 0 · y'(t0)
= Fixo, Yo)x'(to) + Fu(xo, Yo)y'(to).
This holds for every t0• Therefore
cp'(t) =
F,,,(x(t),y(t))x'(t) + Flx(t),y(t)),
Briefly,
In the "fractional" notation,
d<fo oFdx oFdy

+
=
.
dt oxdt o ydt
Both of these short formulas should be regarded as abbreviations of the longer
formula preceding them. The last of these three formulas is the easiest to remember,
especially if we give it an intuitive interpretation, as follows. The derivative dcp/dt
measures the rate of change of cp, when t changes slightly. The change in <Pis due to
(1) the change in x and (2) the change in y. These two effects are measured by the
quantities
oF ox oFdy
ox o t ' o ydt
The formula says that to combine these two effects, we simply add them.
PROBLEM SET 14.5
1. Given F(x,y) =cosxy,x(t) =t2 + 1, y(t) =t3, and <f>(t) = F(x(t),y(t)), find ¢'.
Did you need to use Theorem 3?
2. Same question, for F(x,y) =sinxy,x(t) =t2 + 1, y(t) =t2 - 1, <f>(t) =F(x(t),y(t)).
3. Same question, for F(x, y) =2xy , x(t) =cost, y(t) =sint, <f>(t) =F(x(t), y(t)).
4. Same question, for F(x, y) =x2 + y2,x(t) =cost, y(t) =sint, <f>(t) =F(x(t),y(t)).
5. Same question, for F(x,y) =xy,x(t) =t2,y(t) =t3, <f>(t) =F(x(t),y(t)).

6. Same question,for F(x,y) =x2 + y2,x(t) =cost, y(t) =sint, <f>(t) =F(x(t),y(t)).
7. Given F(x,y) =xy, x =x(t), y =y(t), and <f>(t) =F(x(t),y(t)), find ¢', using
Theorem 3. ( This will give you a circuitous derivation of the formula for the derivative
of a product.)
8. Same question, for F(x,y) =x/y. ( This will give you an equally circuitous derivation
of the formula for the derivative of a quotient.)
9. Same question, for F(x,y) =xu,x > 0. This will give you the formula
D[gh] = hgh-1g' + (gh ln g )h '.

This formula is easy to remember: first we differentiate as though the exponent were
a constant, then we differentiate as though the base were a constant, and then we add
the results.
10. Now derive the same differentiation formula, without using the theory developed in
this chapter, appealing only to the basic definition
(a > 0).
14.6 DIFFERENTIABLE FUNCTIONS OF MANY VARIABLES. THE CHAIN RULE
All the ideas which we developed in the last section, for functions of two variables,
can be generalized immediately for functions of any number of variables. Limits
and continuity have already been defined in the general case, in Section 14.3.
Following the pattern of Section 14.4, we say that a function of n variables is differ
entiable at a point if the difference function is well approximated by a linear function,
in small neighborhoods of the point. The definition is as follows:
Definition. Let D be a region in Rn, let/be a function D--+ R, and let
Po = (a1, az, ... , an)
be a point of D. For each point P = (xi. x2, • • • , xn) of D, let
let
and let
l:,. f = f(P) - f(Po)·
Suppose that there is a linear function
L(/:,.P) = L(l:,.xi. l:,.x2, • • • , l:,.xn) = A1 /:,.x1 + A2 l:,.x2 + · · · + An l:,.xn,
and a set of n functions E1, E2, • • • , En, defined in a neighborhood of 0, such that
!:,.f = L(l:,.x1, !:,.xz, ... , !:,.xn) + E1(/:,.P) !:,.x1 + E2(/:,.P) l:,.x2 + · · · + En(l:,.P) l:,.xn,
(1)
and
lim E;(l:,.P) = 0 (2)
!J.P->O
for each i. Then f is differentiable at P0•

Partial derivatives are defined in exactly the same way as for functions of two
variables. For example,
14.6 Differentiable Functions of Many Variables 645
and so on. Sometimes it is convenient to write f1(P0) for f,,1(P0), and in general
f;(P0) for f,,/P0); that is, f; is the derivative of f with respect to the ith coordinate
in Rn. Thus, for Rn= R2,P= (x,y), we may writef1 forf,, andf2 forJ..
Just as in the preceding section, if a function is differentiable, then it has all its
first partial derivatives, and these are the coefficients in the linear approximation
L(b.P):
Theorem 1. If f is differentiable at P 0, with tlf !"=:! L(b.P), then f;(P 0) is defined for
each i, and
The proof is just the same as for two variables: we take a fixed integer k, and set
tlx; = 0 for i :;zf k, so that b.P = b.xk. Then
tlf = L(b.P) + Ek(b.P) tlxk,

because all other error terms on the right get multiplied by 0. Since b.P = b.xk,
we have L(b.P)= Ak b.xk. Therefore
and
As before, the differential of f at P0 is defined to be the linear function which gives

good approximations of tlf Thus
df = f1(P0) b.x1 + j;(P0) b.x 2 + · · · + f n(P0) b.x n-

By now it should be clear that we are in much the same situation as we were
when dealing with n by n matrices, in Chapter 13: the ideas are adequately conveyed
by the case n = 3, but the notation of the general case is tedious. For this reason,
we shall often deal hereafter with only three variables, using
P = (w,x,y), b.P= (b.w,b.x, b.y) .

You will probably find it easier to generalize the ideas for yourself, in your head, than
to read a generalized version.
Theorem 2. Let D be a domain in R3, and let f be a function D � R. If f has all

three of its first partial derivatives in D, and these are continuous at the point P0 ,
then f is differentiable at P0•
Proof Let
tlf = f(P) - f(P0) = f(w,x,y) - f(w0,x 0,y0).
By a slight extension of the device that we used in the proof of the same theorem for
two variables, we write
tlf = [f(w,x ,y) - f(Wo,x,y)] + [f(w0,x,y) - f(w0,x 0,y)]

+ [f(wo,Xo,y) - f(wo,Xo,Yo)].
In the first bracket, we regard x and y as constants, and apply the mean-value theorem
to the function
<f>(w) =
f(w,x, y).
Then
</>(w) - <f>(w0) = </>'(w) !::i.w,
where w is between w0 and w. Since </>'(w) = fw(w,x, y), we have
f(w,x, y) - f(w0,x, y) =
fw(w, x,y) !::i.w.
By two more such applications of the mean-value theorem, we get
!::i.J =
fw(w,x, y) !::i.w + fx(w0,x, y) !::i.x + fy(w0,x0,y) !::i.y. (1)
Let
E1(!::i.P) =
fw(w, x,y) - fw(w0,Xo,Yo),
E2(!::i.P) = f,,(w0,.X, y) - fx(w0,Xo,Yo),
E3(!::i.P) =
f/wo,Xo,y) - jy(wo,Xo,Yo).
Then
!::i.J =
fw(P0) !::i.w + fx(P0) !::i.x + jy(P0) !::i.y
+ E1(!::i.P) !::i.w + E2(!::i.P) !::i.x + E3(!::i.P) !::i.y,
as in the definition of differentiability.
The chain rule for paths takes the same form as for two variables, and has the
same proof.
Theorem 3 (The chain rule for paths). Let P be a path, with coordinate functions
w(t), x( ), y(t), and with locus lying in the domain D in R3.
t Let f be a function
D-+ R, and for each t, let
</>(t) f(w(t),x(t),y(t)). =
Iff and the three coordinate functions are differentiable, then</> is differentiable, and
</>'(t) =
fw(w,x,y)w' + f,,(w,x,y)x' + fy(w,x, y)y'.
That is,
</>(t) = fw(w(t),x(t),y(t))w'(t)
+ Jx(w(t),x(t),y(t))x'(t)
+ jy(w(t),x(t),y(t)) y'(t)
for every t.
This automatically gives us a chain rule for composite functions in which w, x,
and y are functions of several variables. As in Theorem 3, let D be a domain inR3,
D-+ R. But now let w,x,andy be functions of three variables,
and let/be a function
defined in a domain D'. Suppose that for each point (t,u,v) of D', the point
(w(t,u,v), x(t,u,v),y(t,u,v)) lies in D. We then have a composite function
cp: D'-+R,
defined by the formula
</>(t,u,v) =
f(w(t,u,v), x(t,u,v ),y(t,u,v)),
14.6 Differentiable Functions of Many Variables 647
and we want to find the partial derivatives cp1, <f>u, and <Pv· But this is not a new
problem, really: in calculating cp1, we regard u and v as constants, and this means that
we can calculate <Pt by means of the chain rule for paths. The only difference is that the
derivatives w'(t), x'(t),y'(t) in Theorem 3 are now the partial derivatives wt(t, u, v),
x1(t, u, v) , y1(t, u, v) , and the final answer cp'(t) in Theorem 3 now becomes cp1(t, u, v).
This gives the formula
cp1(t, u, v) = fw (w, x, y)w1(t, u, v)
+f,,(w, x,y)xv(t, u, v)
+fv(w, x,y)y/t, u, v).
This may be easier to remember in the o-notation. In this notation,
o<f> o f aw of ax of oy
= + + .
ai aw ai ax at oy at
Similarly for <Pu and <Pv· Thus we have
Theorem 4 (The chain rule). Letfbe a differentiable function of w, x, and y; and let
w, x, and y be differentiable functions of t, u, and v. Let
cp(t, u, v) = f( w(t, u, v), x(t, u, v), y(t, u, v)).

Then the partial derivatives of o f aw of ax of oy
-=--+--+--,
a1 aw a1 ax at oy at
o<P of aw of ax of oy
- = -- + -- + - - ,
au aw au ax au oy au
o<P ofaw of ax of oy
+ +
=
.
ov ow av OX ov oy OV
f(w, x,y) w2+ x2+ y 2,

=
w t+ 2u+ 3v, x 2t+ 3u+ 4v, y = 3t+ 4u+ Sv.

=
=
Here
of
- = 2w = 2(t+ 2u + 3v),
OW
of
2x 2(2t+ 3u + 4v),
=
- =
OX
of
- = 2y = 2(3t+ 4u + Sv),
oy
ow OX oy
2 3, 4.
=
OU - ,
=
OU OU
Therefore, by the chain rule,
o<f> = 2(t + 2u + 3v)2 + 2(2t + 3u + 4v)3 + 2(3t + 4u + 5v)4

OU
= 40t + 58u + 76v.
This is the right answer; by a direct calculation, we get
<f>(t, u, v) = 14t2 + 29u2 + 50v2 + 40tu + 76uv + 52tv,

so that
o<f>
- = <f>11(t, u, v) = 58u + 40t + 76v,
OU
as before.
PROBLEM SET 14.6
1. Given/(t,u) = t 2u,g(t,u) = t +u2,and <f>(t,u) = J2 +g2,find <f>t and <f>u·

2. Given/(t,u) =t3 - u,g(t,u) = t - u2,and <f>(t,11) =/2g2,find <f>t and <f>u-
3. Given/(s,t,u,v) = s2 +t3 +u4 +v5,g(s,t,u,v) = s +t +u +v,and </>(s,t,u, v) =
sin f cos g; find ¢1 and <f>u-

4. Under the conditions of Problem 3, find </>sand <f>v·
5. Given f(t,u) =t cos u, g(t,u) = t sin u,and <f>(t,u) = /2 +g2,find </>1 and </>,,.
6. Given f(t,u) cos u cost, g(t,u)
= = cos u sin t, h(t,u) =sin u, and </>(t,u) = f+
g + h, find </>1 and ¢u-
7. Given f(t,u) cos u cost, g(t,u)
= = cos u sin t, h(t,u) =sin u, and </>(t,u) = /2 +
g2 + h2, find ¢1 and ¢u-
8. Given f(t,u,v) t cos u cos v, g(t,u,v)
= t sin u cos v, h(t,u,v)
= = t sin v, and
<f>(t,u,v) /2 + g2 - h,find </>t, </>,., and <f>v·
=
9. G iven f(s,t,u,v,w) = stuvw, verify by a direct calculation that f is differentiable at

(1, 0, 0, 1, 0). That is, find error functions Ei(D.P), ... , E5(D.P) as in the definition
of differentiability of a function at a point.
10. Same question, for/(s,t,u,v) = s2 +tu +v3,at (0, 1, 1, -

1 ).
14.7 DIRECTIONAL DERIVATIVES AND GRADIENTS
We shall now generalize the idea of the directional derivative f,a in such a way that it
applies to functions of any number of variables. D in Rn, and a
Given a region
differentiable functionf: D-+ R, let P0 be any point of D, and let V be any vector in
Rn, with Ii VII = l. (A vector of norm 1 will be called a direction in Rn.) The derivative
offin the direction V, at the point P,0 is defined to be
f(P0 + tV) - f(P0)

JV(P)
r _ 1.
o - lID .
t-+O t
14.7 Directional Derivatives and Gradients 649
To compute fv(P0), we let
Po = (a1, a2, ... 'an),

V= (C1, C2, , Cn)·• • •
We now have a path, defined by the coordinate functions
xi(t) = ai + cit (i= 1,2, . . ,n), .
and a composite function
rp(t) = j(X1 (t), X2(t), ... , Xn(t));

and
fv(Po) = rp'(O).
We can now calculate rp' (0) by the chain rule for paths:
Thus we have
Theorem 1. If f is differentiable in D, then f has a directional derivative at every
point of D, in every direction. If Vis a unit vector (c1, c2, , en), then • . •
fv(Po) = fx1(Po)C1 +fx2(Po)C2 + +fxJPo)cn · · ·
= f1(Po)C1 +f2(Po)C2 + · +fn(Po)cn. .
(You should check that for the case n = 2, our definition of the directional
derivative, and the formula given in Theorem 1, agree with the definition and formula
given in Section 14.4.)
The gradient of a differentiable function, at a point P0, is the vector whose
components are the partial derivatives off at P0. The gradient vector is denoted by
grad/ Thus ifjis a differentiable function D R, with real numbers as its values,
__.,.
then grad f is a vector-valued function D Rn, with__.,.
where the f;,'s are the first partial derivatives off That is,
gradf(P ) = (fx1(P),fx2(P), ... , fx"(P)),
for each Pin D. For example, if
f(P) = f(x, y) = x2 + xy +ya,

then
f1( x , y) = fx(x, y) = 2x +y,
and
gradf(x, y) = (2x +y, x + 3y2),
which is a vector in R2, as it should be, for each point P = (x, y).
The definition of the gradient may seem arbitrary, but it is not; the gradient has
a geometric meaning, now to be explained. First we observe that for each unit vector
V, with
V= (c1, C , , en),
2
. . •
llVll2 = c� + c� + + c; = ,1
· .· .
the directional derivative fv can be expressed as an inner product:
fv =/1C1 + f2C2 + · · · + fncn,

where the fi's are the first partial derivatives off, and so
fv = (grad f) · V.
We shall prove the following:
Theorem 2. If/is differentiable at P, then (1) the direction of gradf(P) is the direction
which gives the maximum value of the directional derivative fv(P), and (2) the norm
of grad f(P) is the maximum value offv(P).
Proof Let G= gradf(P), so that

G = (/i(P),fiP), . . . . f n(P)) ,
and let V be the unit vector with the same direction as G; that is,
V= -- 1 G
llGI! '
so that
where
f;(P)
C· = --
i llGll .
Then
fv(P ) = (gradf(P) ) V= G V
· ·
=- G · (_Q_)
llGll
l
= _ (G
llGll
· G) = llGll = llgradf(P)ll.
Thus the directional derivative, in the direction of the gradient, is the norm of the
gradient. And this is the direction which maximizes the directional derivative: if W
is any unit vector, then we know by the Schwarz inequality (Theorem 1 of Section 11.6)
that
(G. W)2 � llGll2 II w112 = llG ll2 ,
and so
fw(P) = G W � llGll = fv(P ).
·
This completes the proof.
A continuous function D--+ 1/, where Dis a region in a Cartesian space Rn and
1/ is a vector space, is called a vector field. We ordinarily draw the graphs of
14.8 Interior Local Maxima and Minima 651
vector fields by using free vectors. For example, for
f(x, y) = x2 + y2,
gradf(x, y) = (2x , 2y),
we can indicate the vector field grad f by drawing sample vectors in the xy-plane,
like this:
y
x
-3 3
-3
At each point P = (x, y), the direction of gradf(P) is the direction of the ray from
the origin through P, and the length l[gradf(P)ll is twice the distance from the origin
to P. At the origin , the gradient vector vanishes. Such a point is called a singularity
of a vector field.
PROBLEM SET 14.7
For each of the following functions f, sketch the graph of grad f

1. f(x,y) =x + y 2. f(x,y)
Vx2 + y2
= 3. f(x,y) = x - y
4. f(x, y) = y2 5. f(x,y) = xy 6. f(x, y) = x3
7. f(x,y) = x2 - y2 8. f(x,y) =
Vl - x2 - y2
x_
2__
J
__
y2
9. f(x,y) = 1 4
- 4 - 10. f(x,y) = x2 + y
1
11. f(x,y) = (x2 + y2)2 12. f(x,y) = 13. f(x,y) = 4y2 - x2
x2 + y2 + 1
14.8 INTERIOR LOCAL MAXIMA AND MINIMA,

FOR FUNCTIONS OF TWO VARIABLES. LEVEL CURVES
For functions of one variable, defined on a closed interval, we had two kinds of
maxima. In the figure on the left below , the maximum occurs at the endpoint b; at
x1 the function has a local maximum, but not a maximum, because/(x1) < f(b).
In the figure on the right, the function has a maximum at x1; this is an interior maxi
mum, and so/'(x1) must be 0.
y y
One of the simplest theorems for functions of one variable was the following:
Theorem 1. Suppose that f, f', and/" are continuous, in a neighborhood of x0•

If/'(x0) = 0 andf"(x0) < 0, then/ has a local maximum at x1•
f'(xo) = 0, f11(x0) < 0, f has a Ll\Iax at x0.
The proof is simple. Sincef"(x0) < 0, andf" is continuous, it follows that there
is a neighborhood (x0 O, X0 + 0) Of X0 such that
-
f"(x) < 0 for x0 - o < x < x0 + o,
as indicated in the figure. Therefore f' is decreasing on the interval
(Xo - O, Xo + 0).
Therefore
J'(x) > 0 for x0 - o < x < x0
and
f'(x) < 0 for x0 < x < x0 + o.
Thereforef is increasing, from x0 - o to x0, and f is decreasing, from x0 to x0 + o.

Thereforef(x0) is the maximum value off on the interval from x0 o to x0 + o. -
14.8 Interior Local Maxima and Minima, for Functions of Two Variables. Level Curves 653
The same proof proves the following theorem, which is going to be more useful:
Theorem 2. Givenf,f', andf", on an interval (x0 - o, x0 + o). Ifj'(x0) = 0, and
f"(x) < 0 for x0 - O < x < x0 + O,
thenf(x0) is the maximum value off on the interval (x0 - o, x0 + o).

We return now to functions of two variables. Let P0 = (x0, y 0)be a point of the
xy-plane. For each o > 0, the o-neighborhood of P0 is the interior of the circle with
center at P0 and radius o. This is denoted by N(P0, o). Thus
N(P o, o) = { P I Po P < o} = {(x, y) (x - I Xo 2

) + (y - Yo)2 < 02}.
®
Let D be a set of points in the xy-plane. If P0 is a point of D, and D contains a
neighborhood of P0 (for some o), then P0 is called an interior point of D.
Thus Pis an interior point of D if P0 lies in D, with at least a little room to spare.
Consider, for example,
D = {(x,y) J x2 + y 2 � l}.
Here D consists of the unit circle, plus its interior. If OP0 < 1, as in the figure, then
P0 is an interior point; if we let
o = 1 - OP0,
then N(P0, o) lies in D.
This works, no matter how close P0 may be to the circle, as long as P0 isn't actually on
the circle; no matter how small the positive number 1 - OP0 may be, we can use it
as our positive o. On the other hand, if OP1 = 1, so that P1 is on the circle, then P1
is not an interior point of D; no matter how small we take o, the neighborhood
NP
( 1, o) contains points outside of D.
. ---,
-...
____
..._ __,., --y
f has an !Ll\fax at P0.
Consider now a function of two variables
f: D---+ R,
and let P0 be a point of D. lffP

( ) �fP( 0 ) for every Pin D, then we say that/ has a
maximum at P0. If P0 is an interior point of D, and P0 has a neighborhood NP ( 0, o)
such that
fP
( ) �fP
( o) for every P in N(P0, o),
then we say that/ has an interior local maximum (ILMax ) at P0•

This discussion has been rather lengthy, but if you review the figures which have
been given in this section so far, you will find that they convey, by themselves, most
of the ideas that we have been talking about.
Our purpose at this stage is to find conditions under which we can conclude that a
function of two variables has an ILMax at a given point. At an ILMax, we must have
because an ILMax offmust be an .ILMax of both the slice functions
<Po(t) = /(xo + t, Yo),

<P 1 2 ( t )
,, = f(xo, Yo + t).
(At this point you may want to review the definition of slice functions, at the
beginning of Section 14.3.) Obviously, however, the vanishing of the partial deriv
ativesfx and fu is not enough to guarantee an ILMax; we might have a minimum or
a saddle point. For example, for
f(x,y) = x2 - y2,
we have
fx(O,0) = fv(O,0) = 0,
/
but the point (0, 0) is a saddle point; the slice function
</>0(t) = f(t, 0) = t2
has a minimum at 0, and the slice function
</>"12(t) = f(O,t) = -t2

has a maximum at 0. One way to see the difference between the behavior of a function
at an ILMax or ILMin and its behavior at a saddle point is to consider the so-called
level curves, in the xy-plane, on which the function takes on various constant values.
For the function

f(x, y) = x2 + y2,
the level curves are circles with center at the origin, as shown above. For each
k > 0, the level curve on whichf(x,y) = k is the circle with center at the origin and
radius .J'k. The origin is a singular point of this family of curves; and this is the point
at which the function takes on its obvious minimum value 0.
For the function
f(x,y) = x2 - y2,
there is no maximum and no minimum. Since/is defined for every x and y,any Max
or Min would have to be an ILMax or ILMin; at any such point, both the partial
derivatives .fc and fv would have to vanish; f,, and fv vanish simultaneously only at
656 Functions of SE .eral Variables 14.8
(0, 0), and at (0, 0) the function has a saddle point. The level curves for this function
look like this :
y y
y=x
y= -x
y= -x
Fork > 0, the level curve on whichf(x,y) = k is a hyperbola in standard position,

and the level curve on whichf(x,y) -k < 0 is conjugate to it. The level curve
=
on which f(x,y) = 0 is the union of two lines, which intersect each other at the
origin,where f has a saddle point. These examples are typical of the way level curves
behave in simple cases.
Even if each of the slice functions in the x- and y- directions has an ILMax at a
point (x0, y0), we may still have a saddle point, on which a man in the saddle would
be facing in some third direction. Consider
f(x,y) = -xy - tx2 - tf.

Here
c/>o(t) = -!t2,
so that </;0 has an ILMax at 0; and similarly for
</>rr12(f) = -!f2•
But for rx = 37T/4 we have
1 . 1
cos rx = .J2. , sm rx = .J2. ,
<f;.(t) = f(t cos rx, t sin rx)
1 t2 1t2
42 42
t2 --
t2 1 2
- =-t ,
2 4 4
which has a minimum at 0.
Thus, if we want to infer that f has an ILMax at (x0, y0), we need to consider
every direction rx, and examine all the slice functions
<fa(t) =f (x0 + t cos rx, Yo + t sin rx) .

This is the basis on which we shall attack the problem. Now
cp�(t). =fx(x0 + t cos rx, Yo + t sin rx) cos rx + fv(x0 + t cos rx, Yo + t sin rx) sin rx;
here we are using the chain rule. Applying the chain rule again, to each term, we get
cp�(t) = fxx(x0 + t cos rx, Yo + t sin rx) cos2 rx

+ fxv(x + t cos rx, Yo + t sin rx) cos rx sin rx
+ fvxCx0 + t cos rx, Yo + t sin rx) sin rx cos rx
+ fvv(x0 + t cos rx, Yo + t sin rx) sin2 a
= c 2fx:. + 2 csf xv + s 2jYY'
Here we are using the abbreviations
c =cos rx, s =sin rx,
!xx =fxx(Xo + t cos rx, Yo + t sin rx) ,

and so on. We are also assuming that all our derivatives are continuous; in this case
it is a fact thatfvx = fxv· (See Appendix K, where this is proved.)
The following is easy to see:
Theorem A. Suppose that

<f;�(O) =0
for every rx. Suppose also that there is a number o > 0 such that for !ti < a we have
cp�(t) < 0 for every rx.

Then/has an ILMax at (x0, y0).
The reason is that for every rx, <fa(O) is the maximum value of <Pa on the interval
(-o, o). It follows thatf (x0,y0) is the maximum value of/in the a-neighborhood
of (x0,y0).
We therefore need to find conditions under which such a c5 exists. We have
r/>�(t) = c2fxx + 2csfxv + s2fyv·

We now use an "ingenious device." We write
lr
r/>;(t) =fxx C2 + 2
csf
·
fxx
�+
s2 r
frxx J
--.:!.Y!!.
s�;Y
=fxx [( c + s . )
fxv
fxx
2 _
f xx
+
s2fvv
fxx
]
=fxx [( c + ) s2]
fxu 2 fxxfvv - f;v .
S • +
fxx J2xx
Suppose now that at the point (x0, y0) we have
fxx < 0,
We are assuming that all the partial derivatives that we are dealing with are con
tinuous. It follows that the same inequalities hold in the c5-neighborhood of (x0, y0),
for some c5 > 0. Thus for !ti < c5 we have
r/>�(t) < 0 for every o:.
If we also know that
then we have
rf>�(O) = 0 for every o:.
Therefore, by Theorem A, f has an ILMax at (x0, y0). We sum all this up in the
following theorem:
Theorem 3. Suppose that f has continuous second partial derivatives in a neighbor

hood of (x0, y0). If
f.(xo, Yo) = fv(xo, Yo) = 0, (I)
fxx(Xo, Yo) < 0, (2)

and
J;ixo, Yo) - fxx(xo, Yo)f,viXo, Yo) < 0, (3)

then/has an ILMax at (x0, y0) .
Not only the proof of this theorem, but also the theorem itself, are hard to read
and hard to remember. This is typical of what you can expect from now on: when we
pass from one variable to two or more, the calculus takes on a higher order of
·
difficulty.
The following is a corollary of Theorem 3:
Theorem 4. Suppose that f has continuous second partial derivatives in a neighbor
hood of (x0, y0). If
f,,(xo, Yo) =fv(xo, Yo) = 0, (1)
fxx(Xo, Yo) > 0, (2)
and
J!vCxo, Yo) - fxxCxo, Yo)fvvCxo, Yo) < 0, (3)

then/has an ILMin at (x0 ,y0).
Proof If f satisfies the hypothesis of Theorem 4, then -f satisfies the hypothesis

of Theorem 3. Therefore -f has an ILMax at (x0,y0). Therefore f has an ILMin
at (xo, Yo).
When we were studying functions of one variable, we found simple cases in
which a function had an ILMax or an ILMin, but the second derivative test failed to
reveal the fact. For example, f(x) = x4 has an ILMin at x = 0, but f"(O) = 0.
Here the trouble seems to be that the function approaches its ILMin value "very
flatly." The same sort of thing can happen for functions of two variables. For
example, the function f(x, y) = x4 + y4 has an ILMin at (0, 0), but Theorem 3
does not apply, because at the point (0, 0), all the second partial derivatives fxx• fv11,
and fxv are equal to 0.
Sometimes, however, we can get negative information by examining the quantity
J!vCxo, Yo) - fx,cCxo. Yo)fvvCxo, Yo)·

If this quantity is positive, then we can infer that f has neither an ILMax nor an
ILMin at (x0 ,y0). The proof is as follows. We found that
</>;(t) = c2fxx + 2 csfx11 + s 2 f1111•

We sett = 0. Let
Then
</>:(o) = A cos2 IX + 2B sin IX cos IX + C sin2 IX,

and we are assuming that
B2 - AC> 0.
Consider the function
1jJ(u) = A + 2Bu + Cu2•
The graph is a parabola; the discriminant
(2B)2 - 4AC = 4(B2 - AC) > 0,

and so 1P(u1) > 0 for some u1 and 1P(u2) < 0 for some u2• But for cos IX � 0 we have
</>�(O) = cos2 1X[A + 2B tan IX + C tan2 IX];
for tan IX = u1, we have </>�(O) > O; and for tan IX = u , we have </>;(O) < 0. Therefore
2
the direction of concavity of the slice functions </>a is different for different values of IX,
and we cannot have an ILMax or an ILMin. To sum up:
Theorem 5. If f ;v -
f xxfvv > 0 at P0, then f has neither an ILMax nor an ILMin
at P0•
PROBLEM SET 14.8
Investigate the following functions for interior local maxima and minima. Not all of
these problems can be worked by straightforward applications of the theorems in Section
14.8; you may need to examine slice functions, or use other elementary methods.
1. f (x,y) = xy 2. f (x,y) =x2 + xy

3. f(x,y) = x2 - y2 4. f(x,y) = (x + y + 1)2 + (x - y + 1)2
5. f(x,y) =x2 + y2 + x + y + 1 6. f (x,y) =x2 + y2 + 2x + 1
7. f(x,y) = x2 + xy + y2 + x + y + 1 8. f (x,y) =x2 + 2xy + y2

9. f(x,y) = 1 - x4 - y2 10. f(x,y) =x4 - 2x2 - y2
11. At what point does the function f (x,y) =x2 ( 1 - x2 - y2) take on its maximum
value? What is the maximum value of the function?
12. At what point or points does the function
f(x,y) = Y(x - 1)2 + (y - 2)2 + .Yx2 + y2
take on its minimum value? What is the minimum value of the function?
13. Consider the ellipsoid
x2 + y2/4 + z2/9 = 1 .
In this surface we are to inscribe a rectangular parallelepiped, with sides parallel to

the coordinate planes. What is the maximum possible volume of such a parallelepiped?
Give the coordinates of its corner in the first octant.
14. Same question, for the ellipsoid
y2 z2
x2 + + -;;=l.
4
15. Same question, for the ellipsoid
y2 z2
x2 + 4 + 4 = 1.
16. Let A1 = (0, 0), A = (1, 2), and A3 = (2, 1). For each P (x, y), let
2
=
f(P) = (A1P)2 + (A2P)2 + (A3P)2•
At what point P0 does f(P) take on its minimum value?

17. Same question, for Ai = (ai, b;) (i = 1, 2,. . . , n), and f (P) =.2f=1 (A;P)2•
14.9 DOUBLE INTEGRALS, INTUITIVELY CONSIDERED
You recall that in Section 3.7 we gave a preliminary intuitive definition of the definite
integral of a continuous function over a closed interval. Here the A;'s are areas, in
the elementary geometric sense, so that A; � 0 for every i. To get the integral, we
count areas above the x-axis positively, and areas below the x-axis negatively.
14.9 Double Integrals, Intuitively Considered 661
Later, in Section 7.2, we gave a new definition of the integral, as the limit of the
sample sums of the function as the mesh of the net approaches 0:
provided, of course, that such a limit exists. The new definition was necessary for
two reasons. First, we needed it to clarify the underlying theory. Second, we wanted
to use the definite integral to solve problems which did not, at the outset, look like
area problems at all. For example, to calculate arc lengths, surface areas, volumes,
and moments, we regarded them as limits of sample sums, as the mesh approaches
zero. Thus our second definition of the definite integral was not only more exact but
also more widely applicable.
We shall follow the same scheme with multiple integrals, first giving an intuitive
definition, and then reformulating it when the need arises (which will be soon).
Suppose that we have given a nonnegative continuous function
f: D--+ R,
defined in a domain Din the xy-plane. (See figure on the left below.)
,0,
� I
y
____,-
I - - 1 ---y
'1
--
1......
' ./
---- x
x
a
x
The expression
fff(P)
D
dA
denotes the volume of the region lying above the xy-plane and below the graph off
This is called the integral off over D. Thus the integral is the volume of the solid
S = {(x,y,z) \ (x,y) ED and 0 � z �f(x,y)}.

In "reasonable" cases, double integrals can be calculated by the method of cross
sections, developed in Section 7.4. The scheme is shown on the right above.
Given a solid S, in space,lying between the planes x a and x b. Suppose that
= =
for each x0 from a to b we can compute, somehow, the area of the cross section in
the plane x =x0. If for each such x0 we let A(x0) be the area of this cross section,
then the volume of our solid S is
vS = f A(x) dx.
This method works for many solids whose volumes are not given by standard
formulas. Consider the following.
y z
y=vx. y
We start with the region
R = {(x,y) \ 0 � x � 1, 0 � y � )-;,},
in the xy-plane. For each x, we join the point (x,0) to the point (x, y2), by a segment.
On each such segment we set up an isosceles right triangle, as shown above on the
right. Let S be the union of all these triangles (including, of course, their interiors).
For each x, the area of the triangle _at x is
A(x) = t)-;; )� = tx.
f A(x) dx = t f x dx = t[tx2]� =
-!.
Here A(x) was computable by an elementary formula, because the cross sections
for constant x were triangular. But no matter what method you use to compute
A(x), you can still find the volume by integrating A(x) between the appropriate
limits. In particular, you can use the method when A(x) is itself computed as a definite
integral.
y
Consider the following. Let
D = {(x,y) \ 0 � x � 1, 0 � y � 1 - x}.
For each (x, y) in D, let
f(x, y) = x2 + y3.
We want to find the volume of the solid lying above D and below the graph off
Now for each x0, the cross section in the plane x = x0 looks like the drawing on the
left below. Therefore the area of the cross section at x0 is
A(x0) =
f -"'0 (x � + y3) dy = [x�y + !y4]�-xo
= xW - Xo) + i(l - X0)4
= x� - xg + !( 1 - x0)4•
Dropping the subscript we get
A(x) = x2 - x3 + Hl - x)4•
ff!(P) dA fA(x) dx
= = [tx3 - !x4 - ! i-(1 · - x)5]�
D
= (t - t] - (-io] = T\.
The method works more generally. Suppose that we have a region D in the
xy-plane, lying between the graphs of two functions, as on the right below.
z
y
x2
0
1
We have given F(x, y) � 0 on D, and we want to find
ff
D
F(x, y)dA.
This is the volume of the solid lying above D and below the graph of F. For each x,
the cross-sectional area is
A(x) =
lgf(x)(x)F(x, y)dy.
Here x is being held constant, and we are integrating from f (x) to g(x). But A(x),
once you get it, is a function. Therefore the total volume is
ff F(x, y)dA = f A(x) dx = f [i:�: >

J
F(x, y)dy dx.
D
This takes a very simple form when D is a rectangular region defined by inequalities of
the form
a� x � b, c � y � d.
Here
ff F(x, y)dA = ff F(x, y)dy dx.

D
Here it is to be understood that the "inside integration" is to be performed first,

giving the cross-sectional area
A(x) = ia F(x, y)dy,
and the resulting function is to be integrated from a to b.
:��] I
I
I
I I
I I
I I
I I
a x b
Of course, we could equally well have used cross sections for constant y. This
would give a different cross-sectional area function
B(y) = f F(x,y)dx;
and we would have
ff F(x, y)dA = f B(y)dy = ff F(x, y)dx dy.

D
The expressions
ffF(x, y) dy dx, ffF(x, y) dx dy

are called iterated integrals. If the general assumptions that we are making in this
section are correct, and in fact they are, then it follows that the two iterated integrals
are equal; that is, the order of integration does not matter. The reason is that each
of the two iterated integrals is equal to the double integral.
The same phenomenon occurs, in a less simple form, if the domain is not
rectangular. In- the figure above, the domain D can equally well be described by the
inequalities
0 � � 1 2
0 � x � 1, y - x , (1)
or
0 � y � 1, (2)
Thus, for any continuous nonnegative function/: D--+ R we have
JJ F(x, y) dA f f-"' F(x, y) dy dx,

2
=
and we also have
f1 r-vl-Y
ff
D
F(x, y) dA
=Jo Jo
F(x, y) dx dy.
Therefore the two iterated integrals must have the same value.
PROBLEM SET 14.9
In each problem below, we have a domain D described by a pair of inequalities, and a

function defined by a formula. In each case, express the double integral
JJF(x,y)dA
D
as an iterated integral in two different ways, evaluate both of your iterated integrals, and
check by observing that they ought to have the same value.
1. D: 0 �x �2, 0 �y �x3; F(x, y) = x + y

x2
2. D: 0 �x �2,
-
0:5y:5-·
- 4 ' F(x,y) = x - y
3. D: 0 �x �2, x3 �y � 8; F(x,y) = x2 + y
4. D: 0 �x � 1, x2 � y �x; F(x,y) = x + y
5. D: 0 �x �1, x �y � 1; F(x,y) = x3y3
6. D: 0 �y �1, y �x � 1; F(x,y) = x2 + y2
7. D: -1 �x � 1, 0 �y � 1 - x2; F(x,y) = xy
8. D: 0 �x �1, -Vl - x2 �y � V l - x2; F(x,y) = (x2 + y2)2
9. D: 0 �x � 1, 0 �y �x2; F(x,y) = .Yxy
10. Let
Find / ( 0<:) .
'
11. Let </> be positive and continuous, and let
/(0<:) =
ff </>(x,y)dxdy.
Find the simplest formula that you can for[' (0<:).

/(<X) =
rld </>(x, y) dxdy.
13. Let </> be a positive function, with continuous first and second partial derivatives. Get
the simplest formula that you can for
fld z
tl
</>xy(X,y) dxdy.
_
____
/
/
I
I
I
I
I
I
I
)-_____ 7r
// .7c---
1 - 2
14.10 CYLINDRICAL COORDINATES IN SPACE. 0

THE DEFINITION OF THE INTEGRAL
To set up a system of cylindrical coordinates in space, we use polar coordinates in

the xy-plane, and leave the z-coordinate unchanged.
For some solids and surfaces, this leads to a considerable simplification. For
example, the cylindrical surface of radius 1, with the z-axis as its axis of symmetry,
is the graph of the equation r = 1. (See the figure above.)
14.10 Cylindrical Coordinates in Space. The Definition of the Integral 667
The unit sphere with center at the origin is the graph of the equation r2 + z2 = 1.
(See the figure on the left below.)
Recalling the familiar formulas giving x and yin terms of rand e, we see that
rectangular and cylindrical coordinates are related by the formulas
x = r cose,
y = r sine,
z = z,
x2 + y2 =
,2.
z
z
'
'
'
'
I '· (r,0,z)
I
I
l..
I ',
I
I 1/ ,'1I
I/ IZ
--�7r
/
/��---- -r---- 2 7r
0
r
--2
0 0
As for polar coordinates in the plane, these formulas work in only one direction:
when rande are named, x and y are determined, but when x and yare named, there
are two possibilities for rand infinitely many possibilities for 8. (See figure on the
right above.)
Suppose now that we have given a domain D in the plane z = 0. The plane
z = 0 may be regarded as the xy-plane or the r8-plane; sometimes we shall refer to
it simply as the base plane. Suppose that we have given a continuous function
/: D --+ R. If we describe a point P of D by its polar coordinates (r, 8), then we have
z = f(P) = f(r, 8).
We now want to compute
ff!(P)
D
dA,
and we want to do this without transforming to rectangular coordinates. In some

cases we might not be able to transform; and in other cases we wouldn't want to,
because the rectangular form would turn out to be unmanageable. Therefore we.
need to know how to deal with cylindrical coordinates in their own terms. This can
be done as follows.
Q
I
J
I
I
Given a domain D in the base plane. By a net over D we mean a finite collection
N: D1, D2, , Dn • • •
of regions such that (1) D is the union of the D/s, (2) each Di has an area (i.e., is
measurable, in the sense defined in Appendix G), and (3) if Di intersects D1, then the
area of the intersection is 0. The sets Di are called the cells of the net. The figure
indicates, at long last, why we use the word net in integration theory.
By the diameter of a set Di we mean the supremum of the distances between its
points. The diameter is denoted by oDi. Thus
oDi = sup {PQ I P, Qin Di}.
Note that if Di is a circular region, then oDi is the diameter of D; in the elementary
sense. The mesh of the net N is the greatest of the diameters of the cells of the net.
The mesh is denoted by JNJ. Thus JNJ Max {aDi}.
=
?
A sample of the net
is a sequence
of points, where P; belongs to D; for each i (see figure at the right above).
For each i, let flA; be the area of D;. A sample sum off over the net N is a sum
of the form
n
"J.J(Pi) LlA;.
i=l
We are now finally ready to give our definition of the double integral. By definition,
JJJ(P) dA = lim _if(P;) flA;,

INl-+o i=l
D
if such a limit exists. If the limit exists, thenf is said to be integrable on D, or simply
integrable. In this definition, liml.ivl-o means the same thing that it meant in the
definition of the integral for functions of one variable; when we write
n
lim "J.J(P;) flA; = L,

INl-+Oi=l
this means that for every e > 0 there is a c5 > 0 such that
!NI < c5 => I i�f(P;) fl A; - L < I €.
We recall that if/ is continuous on the closed interval [a, b], then/ is integrable
on [a, b]. We want to state an analogous theorem for functions of two variables.
It would hardly do to restrict ourselves to "two-dimensional closed intervals"
a � x � b, c � y � d. On the other hand, we cannot allow all sets D in the xy
plane as domains, because continuous functions on some domains may not even be
bounded. (Examples?) What is needed here is the following:
Definition. A point P is a limit point of a set D if every neighborhood U(P, c5) of P
contains a point of D other than P.
Definition. A set D is closed if it contains all its limit points.
Thus a closed interval is closed, but an open interval is not; the region
D = {(x, y) I x2 + y 2 � 1}
is closed, but the region
D' {(x, y) I x2 + y2 < 1}
=
is not.
We recall that a set D in a plane is bounded if it lies in the interior of some circle
(or, equivalently, if it lies in the interior of some rectangle). We can now finally
state our theorem:
Theorem 1. Let D be a closed, bounded, measurable set in the xy-plane, and let f be
a function which is continuous on D. Then/ is integrable on D.
You may be able to convince yourself of this, for positive functions, by thinking of
the integral as a volume, and thinking of the sample sums as approximations of the
volume; the idea is that we can approximate the volume as closely as we please,
by cutting up the base domain into sufficiently small pieces. If the function is negative
somewhere, then we need to use volumes with signs attached, but the idea is much
the same. But a mathematical proof that all this works is far beyond the scope of
this book, and we make no attempt to present one.
Meanwhile we assume that the theorem is true, and return to the problem of
integration in cylindrical coordinates. For the sake of simplicity, we consider first a
domain of the type
D = {(r, 6) I a � r � b, Cl � e � {J}.
This is the polar equivalent of a rectangular region.
7r
2 z
The first step in the calculation of the integral is to set up a net N, on the interval
[a, b] and a net N8 on the interval [cc, {J]. Thus we have
No: 60, 61, . . . 'em.

The circles r = ri and the rays 6 = 61 now cut up the domain D into nm little pieces,
like this:
7r
2
Let Di; be the shaded region in the figure. Thus
D;1 = {(r, 0) I r ;_1;;; r ;;; r;, 01_1;;; 0 ;;; 01}.

Evidently the D;/s form a net N over D. For each i, j, the area of D;1 is
Then
and
r; - r;:1 = r; - (r; - 2r; Llr; + Llr;2)
= 2r; Llr; - Llri.
Therefore
Ll A;1 = i Ll01[ri - rLJ = i Ll01[2r; Llr; - Llri]
= r; Llr; LlO1 - i Llri LlO1•
In each cell D;1 of the net we pick the sample point P;1 = (r ;, O;). We now form the
sample sum
n m
L = L Lf(r;, O;) LlA; ;

i=l i=l
n m
= 2 2 f(r;, 01)[r; Llr; LlO1 - t Llr;o 1].

i=l i=l
Evidently � is a sample sum off over D; and our problem is to find
lim
INl->o
L= fff(P) d A.
D
But this is much easier than it looks:

n m n m
2 = 2 2 f(r;, 01)r; Llr; LlO1 - }Llri 2 2 f(r ;, 01) Llr; LlO1•

i=li=l i=li=l
We interpret each of these double sums as a sample sum in rectangular coordinates.
On=/3
oi - - -1-- +-----+- -.---<
pij
- --
oj-1 - - - 1--+-----1---1
00=a
Let D' be the rectangular region shown in the figure. That is,
D' = {(r' e) Ia �r� b' ex �e� {J}.
The limits of our sums are now known:
Iim I � f(ri, e;)ri !lri Ile; = JJ !(r, e)rdrde,

JNJ -+O i=l i=l
D'
and
i � f(ri, e;) !1ri 11e, =

lim
JNJ -+O i=l i=l ff1(r, e) drde.
D'
Since IimlNl-o !1ri = 0, we have
JJ f(r, e) dA = JJ f(r, e)rdrde + JJ f(r, e) drde = JIf(r, e)rdrde.

0 ·
D D' D' D'
Thus the second integral has dropped out.

We usually evaluate double integrals by converting them into iterated integrals.
For the special type of domain that we have been discussing, we can sum up our
results in the following theorem:
Theorem 2. Let
D = {(r, e) Ia �r� b, ex �e� (J},
in polar coordinates, and let f be a function which is continuous on D. Then
JJ f(r, e) dA = rfj(r, e)rdrde =fJ:f(r, e)rdedr.

D
Let us try this out in a simple case in which we know the answer. Consider the
hemisphere under the graph of
z =f(x,y) = )1 _ x2 _
y2.
In cylindrical coordinates,
z =f(r, e) = )1 - r2•
Let
D = {(x ,y) I x2 + y2 � l} = {(r, e) I r2 � l}.
Then the volume of the hemisphere is
JJ-!1 - r2dA =f"f_;1 - r2rdrde.

D
Now
f J1 -
r2rdr = {-t·�(l - r2)312+ C}.
Therefore
Therefore
ffJ1 - r2 dA = f" !dfJ 2;. =
This is right, because the volume of the whole sphere is 477/3 13 · = 477/3.
PROBLEM SET 14.10
1. Let D = {(x, y) [ x2 + y2 � l}. Find
JJcx2 + y2)712dydx.
D
2. Let D = {(x,y) I x2 + y2 � l}. Find
JJ v 1 + x2 + y2dydx.
D
3. Find the volume of the solid which lies under the paraboloid z = x2 + y2 and over
the interior of the cardioid r = 1 - sin e.
4. Find the volume of the solid lying inside the cylinder x2 + y2 � 1 and inside the
sphere x2 + y2 + z2 = 4.
5. Find the volume of the solid lying inside the cylinder x2 + y2 � 1 and inside the
ellipsoid
x2 y2
4 + 4 + z2 = 1.
6. Find
il Jyl-x• _(x2 + y2)10dydx.
-1 -v1-x2
7. Let D be the circular region with center at (0, 1) and radius 1 , in the xy-plane. Find
x
ffv 2
D
x + y2
dydx.
8. Let D be as in Problem 7. Find
y
ffvx2
D
+ y2
dydx.
9. Let S be the part of the disk D lying in the half-plane {(x, y) Ix ;;; O}. Find
xy
ff s
x2 + y2
dydx.
10. Find
f1 i'h-x• sin7 cos7 x y
-1 - y l-x• (X2 + 2)3/2 dydx.
Y
11. Find
12 iy4-x• 2 x2 - Y.
- dydx.
0 0
- x2 + y2
-
14.11 MOMENTS AND CENTROIDS OF NONHOMOGENEOUS BODIES
We recall, from Section 7.6, the definitions of moments and centroids for finite
systems of point masses in a coordinate plane. Suppose that we have given a set of
particles Pv P2, Pn, with masses m1, m 2,
• • • , , m n, at the points (x1, y1), • • •
(x2, y2), , (x n, y,,). The moment of the system about the y-axis is defined to be
• • •
M11 = .L m;X;,
i=l
and the moment about the x-axis is

n
M., = _L m;Y;.
i=l
More generally, the moment about the line x = x0 is

n
Mx=xo L m;(x; - Xo), =
i=l
and the moment about the line y = Yo is

n
Mu=vo = L m;(y; - Yo) ·
i=l
If
then the point (.X, ji) is called the centroid of the system. By easy calculations we get
n
1 m; · I
ji = _L m;Y; m = - ( )
m ;=1 i=l
It is easy to see that if the axes are translated, the centroid is unchanged: for
x = x' + h, y = y' + k, x' = x - h, y' = y - k,

the coordinates of the centroid in the new coordinate system are given by the formulas
x
_,
= -1 ,,;;;.,� m;x;, -m1 �,,;;;., m;(X; - h)

m;=1
=
i=l
1
= x- - - ,,;;;., m; = x- - h,
h�
m i=l
_,
Y = -m1�;,,;;;.,=1 m;Yi
,
= Y- - k,
14.11 Moments and Centroids of Nonhomogeneous Bodies 675
so that in the new coordinate system we get the same centroid as before. Similarly,
if we reverse the direction of the x-axis, or the y-axis, or both, we get the same
centroid as before. Finally, we observe that the centroid is unchanged if we rotate
the axes through an angle of measure(), We have
x = x' cos() - y' sin(),

y = x' sin() + y' cos(),
Therefore
- 1 � . ()
x = - £., mi(xi' cos() - Yi' sm )
m i=l
= .!m ( ii=l mix�) cos() - .!

m i=l
( i miy�) sin()
= x' cos() - ji' sin(),

where x' and ji' are the new coordinates of the centroid. A similar calculation
gives
ji = x' sin () + ji' cos e.
Thus the old coordinate system and the new one give us the same point as centroid.
Suppose now that we have a thin rod, lying on an interval [a, b] on the x-axis.
We do not suppose that its mass per unit length is constant. But in any case there
is a function f which gives, for each x, the mass of the part of the rod that lies on
the interval [a, x]. If f has a continuous derivative f', then .
if'(t) dt = f(x) - f(a ) = f(x),
becausef(a) = 0. More generally,
A function p which behaves in the way that we have just observed for f' is called a
density function for the rod. That is:
Definition. Given a rod on [a, b]. For a � x1 < x2 � b, let m(x1, x2) be the mass
of the part of the rod that lies on [x1, x2]. A densityfunction for the rod is a function
p such that
It follows, of course, that the total mass m of the rod is m(a, b) g p (x) dx.=
And the definition agrees with our intuitive notion of what density at a point ought
to mean. If p(x0) is the density at x0, then it ought to be true that
when <5 � 0.
Here the lefthand side is the average mass per unit length on the interval [x0, x0 + o],
and this ought to be approximately p(x0) when o R::> 0. And this is true, at any point
xo+o
where p is continuous:
o-+O o Jxo. p(x) dx = p(x0),

1 1
Jim - m (Xo, Xo + 0) = lim -
o-+o O
by the general formula for the derivative of the integral. We assume hereafter that p
is a continuous function. Let us take a net N: x0, x1, • • • , xn over [a, b], and form
the sum
n
,L xip(xi) 6.xi.
i=l
This sum is the moment, about the origin, of a finite system of particles of mass
p(x1) 6.x1, ... , p(xn) 6.x n ,

at the points x1, x2, • , xn. The limit of the sum, as the mesh of the net approaches
• •
0, is f� xp(x) dx. This integral is defined to be the moment of the rod about the origin.
Thus
b
M0 = a xp(x) dx. J
More generally, the moment about the point x = k is
Mk= f(x - k)p(x) dx.

We now define the centroid as the point x such that
M33 = 0.
It is easy to calculate that
_ J! xp(x) dx
x= .
J! p(x) dx
By the definition of the density function, the integral in the denominator is the total
mass m of the system. Thus, briefly,
x = -
1 iabxp(x) dx.
m
Suppose, for example, that the rod lies on the interval [O, 2], and that the density
is proportional to the distance from the origin. Here we have
p(x) = kx,
m = fkx dx = [tkx2]g = 2k,
"
x= _!_ [ x kx dx _!_ k[lx3]� = ! l 8 f.
Jo
· = · · · =
m 2k
This is greater than one, as it should be.
Consider next a thin plate, occupying a region D in the xy-plane. Again we do

not suppose that the density per unit area is constant.
For each subregion D; of D let m(D;) be the mass of the portion of the plate that
lies in D;. Following the analogy of the rod, we define a density function for the
plate to be a function psuch that
m(D;) =ff ( ) dP
pP
D;
for every D; lying in D. In particular, D; may be all of D; and in this case the total
mass of the plate is
m = m(D)
=ff D
p(P) dA.
Hereafter, we assume that pis continuous. We take a net
N= D1, D2, • • • , Dn
over D; we take a sample P1, P2, • • • ,Pn of N, with Pi= (xi, y;); and we form the
sum
n
( i, yJ �Ai,
xipx
iL
=l
where �A; is the area of Di. This sum is the moment about the y-axis of a system of
particles of mass p(x i, Yi) �Ai, with x-coordinates x i . As the mesh of the net
approaches zero, these sums approach the limit
Jf xp(x, y) dA.
D
By definition, the moment of the plate about the y-axis is this integral. More generally,
the moment about the line x =k is
Mx=k =fJ cx - k)p(x, y) dA;

D
and similarly,
M11=k = ff<y - k)p(x, y) dA.

D
The centroid is the point (x, ji) such that
Since the total mass is
m =ffp(x, y) dA,
D
an easy calculation gives
x = ; ffxp(x, y) dA, ji = ; ffYP(x, y) dA.

D D
In the preceding discussion, we have assumed for the sake of simplicity that the
density is continuous, so that we don't need to worry about whether our integrals
exist. In some very simple cases, however, the density is not continuous. Suppose,
for example, that we take a rod of unit length, with constant density 1, and another
rod of unit length, with constant density 2, and lay them end to end.
p=l p=2
0 2
Thus p(x) = 1 for 0 � x < 1, and p(x) =
2 for 1 < x � 2. At the midpoint 1,
we split the difference, and take p(l) t· We now have a discontinuous density
=
function whose graph looks like this:
-+------+--x
2
y=(x)
Note, however, that pis integrable, and that
fp(x) dx = 1 + 2 = 3,
which is equal to the mass, as it should be. The function xp(x) looks like this:
y
2 I
•
I
I
I
I
I
I
I
x
2
and
M0 = fxp(x)dx = i + i(2 + 4) = t.
Therefore
x = _!_ Mo = l t · = t.
m
This is the right answer. Ifwe assume that the masses of the two halves of the rod are
concentrated at their centroids, then we get two particles, of masses 1 and 2, at the
points i and t.
• x
0 2
Here
Mo = t ·
1 + t 2 · = {, m = 3,
and
x = t. t = i,
as before.
This illustrates the way in which our formulas work, for discontinuous density
functions. The general theory, however, is hard, and we make no attempt to discuss
it here. Meanwhile the above example shows that some very simple physical situations
lead naturally to discontinuous functions.
PROBLEM SET 14.11
1. A thin rod occupies the interval [2, 4]. Its density is proportional to the distance from
the origin. Find the centroid.
2. A thin rod occupies the interval [I, 2]. Its density is proportional to the square root
of the distance from the origin. Find the centroid.
3. A thin plate occupies the unit disk with center at the origin. Its density is proportional
to e-<•"+Y'>'. Find the centroid.
4. A thin plate occupies the unit disk with center at the origin. Its density is proportional
to Vl + x2 + y2• Find the centroid.
5. A thin plate occupies the righthand half (x � 0) of the unit disk with center at the
origin. Its density is proportional to the distance from the origin. Find the centroid.
6. Same question, where the density is proportional to the square of the distance from the
origin.
7. A thin plate occupies the interior of the cardioid r = 1 - sin IJ. Its density is constant,
say, = 1. Find the centroid. (The computation is Jong, even if the appropriate
short-cuts are used.)
8. A function f is defined by the conditions
f(x) =
{ x2 for 0 � x � 1 ,
(x - 2) 2 for 1 � x � 2.
Find S�f(x)dx.
9. A function/ is defined by the condition
for 0 � x � 1 ,
f(x) =
{; = �2 for 1 � x � 2.
Find S�f(x)dx.
10. A thin plate occupies the square region whose corners are (0, 0), (1, 1 ), (2, 0), and
(1, - 1 ). Its density is proportional to the distance from the y-axis. Find the centroid.
11. A thin plate occupies a triangular region with vertices (0, 0), ( 1 , 1 ), and (1, -1). Its
density is proportional to the distance from the x-axis. Find the centroid.
1 2. Given a thin plate, occupying a region D, with density function p. The moment of
inertia of the plate about the point P0 = (x0, y0) is defined to be
lp0 = JJ (P0P)2p(P)dA JJ [(x - x0)2

= + (y - y0)2]p(x, y)dA.
D D
Suppose that the plate occupies the unit circle with center at the origin, and that the
density is constant. Find the moment of inertia about the origin.
13. Under the conditions of Problem 12, find out which point P0 gives the minimum value
of the moment of inertia.
1 4. The moment of inertia of a thin plate about the line x = x0 is
lx=xo =ff cx -x0)2p(x,y)dA.

D
The moment of inertia IY=Yo about the line y y0 is defined similarly. A thin plate
=
occupies the righthand half of the unit disk with center at the origin, and its density is
proportional to the distance from the origin. Find lx=o and ly=o·
1 5. Given a thin plate, with density function p, on a domain D. For what point P0 does the
moment of inertia Ip take on its minimum value?
14.12 LINE INTEGRALS
Suppose that we have given a path P: I--+ D, where I is a closed interval [a, b] and
D is a region in a coordinate plane.
14.12 Line Integrals 681
�
P(a) P(b)
Let f and g be the coordinate functions of the path P, so that

P(t) (j(t), g(t)) =
for each t, and suppose that f and g have continuous derivatives. Let F and G be
continuous functions defined on D. The line integral Sr F dx + G dy, of F and G
over the path P, is defined as follows.
Let
be any net over I= [a, b]. For each i, let
xi = f(ti), Yi g(ti),
=
ti.xi =
X; - xi-1• L'l.yi = Yi - Yi-1·
Then
J{P F dx + G dy = Jim
JNJ..,O
i [F(x;, Yi) L'l.x; + G(x;, Y;) L'l.y;].
i=l
y
We need, of course, to show that the limit exists. We shall do this by deriving a
formula for the limit, as follows:
Theorem 1. If F and G are continuous, and the coordinate functions f and g of the
path P have continuous derivatives, then
�
tF dx + G dy JrbF(J(t), g(t) )f'(t) dt + rbG(f(t), g(t) )g'(t) dt.

Ja
a
=
Proof On each little interval [ ti 1 , t;]

_ of the net, we take a i; such that
!:ix; = f' (i;) !:it;
n n
.L F(x;, Y;) !:ix;= .L F (f(t;), g(t;))J'(i;) !:it;.

i=l i=l
This is almost, but not quite, a sample sum of the function
</>(t) = F(j(t),g(t))j'(t)
over the net N; the only trouble is that we have substituted two different sample
points in two different places in the formula for if>. But in the limit, this does not
matter. (See Appendix I, where a very similar case is discussed in detail.) Therefore
n
lim
iNi-Oi=l
_LF(x;, y;) !:ixi=
lb</>(t) dt lbF(j(t),g(t))f'(t) dt.
a
=
In exactly the same way, we get
lb
n
Jim .L G(xi, Yi) !:iyi = G(f(t), g(t) )g'(t) dt ;
INl-+O i=l a
and from this the theorem follows.
P(t) = (j(t),g(t)) = (t + 1, t 2) (0 ;;; t ;;; 1),

F(x,y) = x + y, G(x,y) = x + y2•
Then
LF dx + G dy = r [(t + 1 + t2) + (t + 1 + t4)2t] dt

= f(2t5 + 3t2 + 3t + 1) dt= _g_l.
Line integrals have the following quite natural physical interpretation. We regard
the path P: I-. Das a description of the motion of a particle in the plane, during
the time interval a ;;; t ;;; b. Suppose that at each point (x,y) of Dthere is a resisting
force R(x,y) = F(x,y)i + G (x , y)j .
F
Here R is a vector, with components F and G in the x- and y-directions, and the indi
cated addition is vector addition. As the particle moves from P(t;_1) to P(t;), the work
should be approximately
W; = F(x;, Y;) Llxi + G(xi , Yi ) Ll yi ,

where the first term is the "work in the x-direction" and the second term is the "work
in the y-direction." Therefore the total work W should be
n
W R:j L [F(xi, Y;) Llx; + G(x;, Y;) Lly;].

i=l
Passing to the limit, as !NI� 0, we get
W= L F dx + G dy,
which is the formula ordinarily used as the definition of work.

In the above discussion we have described the path P and the resistance R in
terms of the coordinate functions f, g (for P) and F, G (for R). Thus, in effect, we
have been using a particular basis {i, j} for the base plane R2, with
P(t) = f(t)i + g(t)j,

R(Q) = F(Q)i + G(Q)j.
Note, however, that
so that the sum

n
L [F(x;, Y;) Llx; + G(x;, Y;) Lly;],

i=l
whose limit is the line integral, is really a sum of inner products:
n
_L R(Q;) LlPi,
·
i=l
where Q ; P(t;)
= (x;, y;). Therefore the line integral depends merely on the
=
vector-valued functions P and R; it is independent of the coordinate system in the

base plane. Of course this must be true, for any mathematical concept which has a
physical meaning. In vector notation, the line integral is denoted by
LR·dP.
A very important special case is the one in which the function

R: D�R,
: (x, y) � F(x, y)i + G(x, y)j
is an exact differential. This is the case in which there is a function x = F, cf>11= G,
so that the line integral takes the form
LR· dP =LF dx dy L <Px dx ef>v dy

+ G = +
=f [ef>.,(f(t), g(t))f'(t) <Py(f(t), g(t))g'(t)] dt.

+
Here, by the chain rule for paths, the integrand is the derivative of the function
(t ) = 1>(/(t), g(t)).

It follows that
LR· dP = ef>(f(b), g(b))- ef>(j(a), g(a)).

The answer here depends only on the endpoints
P(b) = (j(b), g(b)), P(a) = (j(a), g(a))
of the path; it is independent of the way in which the path proceeds from the initial
point P(a) to the terminal point P(b).
In such a case, we may describe the line integral merely by using the endpoints of
the path as limits of integration, writing
r<c',d')<Px dx + <Pv dy
Jcc,a>
for
L <f>x dx + <f>.v dy.

In these formulas, we use ef>.,, �v for F and G to emphasize that the notation fi�:g;>
can be used only in the case where the integrand is an exact differential.
PROBLEM SET 14.12
Calculate the line integral fp R · dP, by any method, for the following functions R
andP.
1. R(x,y) =x2i + y2j; P(t) = i sint + j cost (0 � t � 2rr)

2. R(x,y) = x2 i + y2j; P(t) = it + jt2 (0 � t � I)
3. R(x,y) = yi + xj; P(t) = i sint + j cost (0 � t � )
1T
4. R(x,y) = yi + xj ; P (t) = i sin11t + j cos11t (0 � t � )

1T
5. R(x,y) =xyj; P(t) = it + jta (0 � t � l)

6. R(x,y) = i y e"'Y + j xexy; P(t) = it3 + (j l - ts) (-1 � t � l)
7. R(x,y) = i y exy + jxexy; P(t) = it (-1 � t � l)
8. R(x,y) =xi - yj , P(t) = i sin2t + j cos2t (0 � t � 21T)
9. R(x,y) =xi - yj , P(t) = it 2 + jt3 (0 � t � l)
10. R(x,y) =xi - yj , P (t) = i sint + j cost (0 � t � 1T/2 )
11. R(x,y) = yi - xi, P(t) = i sint + j cost (4 � t � 7)
12. R(x,y) =(x2 - y2)i + 2xyj , P(t) = i cost + j sint (0 � t � )
1T
The Shorthand of
Appendix A Logic and Set Theory
In this book, the use of logical symbols is held to a minimum, on the ground that
words are usually easier to read. But the symbolism explained below will at some
points be convenient in the text, and is even more useful in notebooks and on black
boards.
We explained in Chapter 1 that <=> means "is equivalent to." Thus
x < 1 <=? 1 > x.

And => means "implies." Thus
x >2 => x2 > 4.

Occasionally we write this symbol backwards:
x2 > 4 <= x > 2,

which means the same thing. We also recall from Chapter 1 that
{x I P(x)}
x such that P(x) is true. That is, {x I P(x)} is the solution
denotes the set of all obj ects
set of the open sentence P(x). Thus the closed interval from 0 to 4 is
[O, 4] = {x I 0 � x � 4},
and the open interval from 1 to 2 is
(1, 2) = {x I 1 < x < 2}.

If A and B are sets, of any kind whatever, then
AcB
means that A is a subset of B; that is, every element of A is also an element of B.

For example, if
A = (1, 2) and B = [O, 4],
then
Ac B.
We allow the possibility that A = B, so that A c A for every set A. Thus c is

like �, not like <. If A c B, we can also write
B ::i A.
688 Appendix A
If x is an element of the set A, then we write
x EA.
This is read "x belongs to A." The denial of this statement is indicated by a diagonal
stroke. That is,
x¢A
means that x does not belong to A. The union of A and B is denoted by
A UB.
Thus
A UB={x I x EA or x EB}.
The formula A u B is read "A cup B." The intersection of A and B is denoted by
AnB.
Thus
AnB={x I x EA and x EB}.
The formula AnBis read "A cap B." The difference
A-B
of A and Bis the set of all elements of A that are not elements of B. To write A-B,
we need not suppose that B c A. For example, if A= [O, 2] and B = [1, 4], then
A-B= [O, 1).
These sets are intervals, of course, described in the notation of Chapter 1. The
empty set is denoted by { }. Thus
AnB={}
means that A and B have no element in common. And
A-B={}
means that A c B. When we write
(x-1)2= x2-2x + 1 y.,,
we mean that the equation on the left holds true for every x. Similarly,
x2-y2=(x-y)(x + y) v.,,11
means that the equation holds true for every x and y. The symbol "Y" is read "for
every." More informally, we may write
xs-ys=(x-y)(x2 + xy + y2) y'

where the symbol "Y" means that the preceding equation holds true for all values
of all the variables that appear in it. When "V" stands alone, it may be pronounced
always.
The symbol 3 stands for "there exists," and the symbol 3 stands for "such that."
The Shorthand of Logic and Set Theory 689
Used in moderation, this symbolism is a convenience. But its use can be over
done; and it takes practice to read formulas like
(x ER and y ER and x < y) => 3z 3 z E R and x < z < y.
This says that between any two real numbers there is a third.
This symbolism is introduced merely as a scheme of abbreviations of English
words and phrases. This book makes no attempt to deal with symbolic logic; and its
use of the "theory of sets" is entirely intuitive. But the shorthand of logic is useful
simply as a shorthand; the point is that we are more likely to say what we mean if we
have a quick and easy way to do so.
Algebraic Operations
Appendix B with Limits of Functions
In Section 3.4 a number of theorems on limits were stated without proofs. Here we
give the proofs. First we recall some of the results of Section 3.4.
Theorem 1. If limX---+.,.f(x) =L, then lim.,-.,.[f(x) - L] = 0.
Theorem 2. If lim.,-.,0 [f(x) - L] = 0, then lim.,-.,J(x) =L.
(These were Theorems 2 and 3 of Section 3.4.)
Theorem 3. If lim.,-.,J(x) = 0 and limo:---+xo g(x) = 0, then
lim [f(x) + g(x)] = 0.

x-+xo
Proof Let " > 0 be given. There is a 01 > 0 such that

0 < Ix - 0
x 1 < 01 => x I < E/2.
If ()
(Here we are using i;:/2 for " in the definition of the statement limo:---+., J(x) = 0.)
Similarly, there is a 02 > 0 such that
0 < Ix - Xol < 02 => lg() x I < i;:/2.
Leto be the smaller of the numbers 01 and 02• Brieflyo
, = min (01, 02). Then
0 < Ix - x01 < O => If (x)I < i;:/2 and lg(x)I < i;:/2.
Since
If (x) + g(x)I � If (x)I + lg()
x I,
we have
0 < Ix - x0I < O => If (x) + g(x)I � If (x)I + lg(x)I
< E/2 + E/2 = E,
Theorem 4. If lim.,-.,J(x) =Land Jim.,-.,0g(x) =L', then
Jim [f(x) + g(x)] = L + I.:.
Proof By Theorem 1 we have
Jim [f(x) - L] = 0, lim [g(x) - I.:] = 0.

X--+CCO
690
Algebraic Operations with Limits of Functions 691
By Theorem 3,
lim [f(x) + g(x) - (L + I:)] = 0.
By Theorem 2,
lim [f(x) + g(x)] = L + I:.
X--tXo
A function f is locally bounded at a point x0 if there are positive numbers Mand o

such that
0 If(x)I < M.
Theorem 5. If f approaches a limit, as x � x0, then f is locally bounded at x0•
Proof Let
limf(x) = L.
Let Ebe any positive number, and let o be as in the definition of a limit.
f
L
I I
I I
L-• --
�---4---�
I I I
I I I
�-+--��-x
x0-o x0 Xo+o
Thus
0 < Ix - x01 < o => L - E< f(x) < L + E.
The o that we now have is the o that we wanted. We let M be the larger of the
numbers IL+ E l , IL - E l .
Theorem 6. If lim,,_,,J(x) = 0, and g is locally bounded at x0, then
lim [f(x)g(x)] = 0.
692 Appendix B
Proof Let a positive number E be given. Take 01 > 0 and M > 0 such that
0 < Ix - x0I < 01 (1)

=> lg(x)I < M. (2)
Next, using e/ M in place of e in the definiton of a limit, take c52 > 0 such that
0 < Ix - Xol < 02 (3)

E
=>
If(x)I < - . (4)
M
0 < Ix - x0I < O, (5)
(
then (1) and (3) both hold, and so (2) and (4) both hold. Therefore
g(x) < M,
Ix - x01
0 < < O =>
f(x) < �.
Since
lf(x)g(x)I = lf(x)I lg(x)I,

·
we have
0 < Ix - x0I < o => lf(x)g(x)I < !.... · M = e.

M
Thus, given an E > 0 we have found a o > 0 such that
0 < Ix - x01 < o => If (x)g(x)I < e.

This means that
lim [f(x)g(x)] = 0,
:z:-:z:o
Theorem 7. If lim.,._.xJ(x) =Land limx�x. g(x) = L', then

lim [f(x)g(x)] = LL'.
Proof By Theorem 2, we need to show that
lim [f(x)g(x) - LL'] = O;
Theorem 7 will then follow. Now
f(x)g(x) - LL' = f(x)g(x) - L g(x) Lg(x) - LL'

+
= [(j(x) - L)g(x)] [(g(x) - L')L].
+
In each of these brackets, the first factor approaches 0 as x � x0, and the second
factor is locally bounded at x0• Therefore
lim [(f(x) - L)g(x)] = 0,

:z:-+a::o
Algebraic Operations with Limits of Functions 693
and
lim [(g(x) - L')L] = 0.
a::-+xo
Therefore the sum of the two bracketed expressions approaches 0, which was to be
proved.
Roughly speaking, a function f is locally bounded away from 0 at x0 iff (x) is not
very close to 0 when x is close to x0 and different from x0.
Definition. Suppose that there are numbers € > 0 and c5 > 0 such that
0 < Ix -
Xol < c5 => If(x)I > €.
Then f is locally bounded away from 0 at x = x0•
y y
�-t-��----.,___,�1 -Xo��
1 1
� Xo-+�0,,-----x
st- -t-i--
-x;;--
1 I I
I I
Note that, if/(x) is never 0 when x ¥= x0 and xis close to x0, it does not follow
that/ is locally bounded away from 0 at x0• The situation shown in the figure on the
right above can easily occur. Here f is undefined at x,0 f (x) ¥= 0 for x ¥= x0, and
lim.,_,.., f(x) = 0. In this case f is not locally bounded away from 0 at x•0
0
Theorem 8. If l i m.,_,.x0 f(x) L, and L ¥= 0, then f is locally bounded away from
=
0 at x0•
Proof Suppose first that L > 0. Let € = L/2, and let c5 be as in the definition of a
limit.
y
Then
694 Appendix B
Thus
0 < Ix - x01 f(x) > E
=> lf(x)I > e.
Suppose now that L < 0. -L > 0, and limx-x0 [-f(x)]
Then = -L. There
fore -f is locally bounded away from 0 at x0• Therefore so also is f (Look again at
the definition: it is a statement about I f!.)
Theorem 9. Iff is locally bounded away from 0 at x0, then l!fis locally bounded at x0•
Proof We know that there are positive numbers e and b such that
0 < Ix - x0I I f(x)I > E.

Therefore
=>
0 < Ix - x01 - 1- <
lf(x)I
.!
e J i
--
1
f(x)
<
.!e .
Therefore 1 /f is locally bounded at x0; we use the b that was given for f, and we use
the bound M = l/e.
Theorem 10. If lim.,_.,0 f(x) = L, and L :;.6 0, then
lim -1- = !_.

x-+xof(x) L
Proof We need to show that
Jim
x-+xo
[f(x) - !.]L
-
1
= 0.
Now
1 1 L - f(x)
1
[L - f(x)].
L
= =
f(x) Lf(x) Lf(x)

·
-
The bracket on the left approaches 0. The fraction on the right is locally bounded at
x0• Therefore the product approaches 0.
Theorem 11. If lim.,_.,0 f(x) = L, lim.,_.,0 g(x) = L', and L' :;.6 0, then
f(x) L
Jim .
I.:
=
x-+xo g(x)
Proof by Theorems 7 and 10.
Algebraic Operations
Appendix C with Limits of Sequences
Nearly everything in this appendix is analogous to something in the preceding one.

The analogy begins with the definition of the limit of a sequence.
Definition. Given a sequence a1, a2, • • • and a number L. Suppose that for every
e > 0 there is an integer N such that
n > N => Jan - LJ < E.
Then
lim an= L.
n.-+oo
The following theorems are modeled on the theorems of Appendix B. We there

fore give only a few of the proofs; you ought to be able to supply the rest of the
proofs yourself.
Theorem 1. If limn--+OO an= L, then limn--+OO (an - L) = 0.
Proof Let e > 0 be given. There is an N such that

n > N => Ja11 - LJ < e.
Therefore
n > N => J(an - L) - OJ < e.
Since for every e > 0 there is such an N, it follows that limn--+oo (an - L) = 0.
Note that statements of the form n > N are playing exactly the same part as
statements of the form 0 < Ix - x0J < o.
Theorem 2. If limn--+oo (an - L) = 0, then lim11--+00 an= L.
Theorem 3. If limn--+oo an= 0 and limn--+oo bn= 0, then
n-+OO
Theorem 4. If limn--+OO an= L and limn--->OO bn= L' then
Definition. A sequence a1, a2, • • • is bounded if there is a number M such that

Jani � M for every n.
(For sequences, the question of local boundedness does not arise.)
696 Appendix C
Theorem 5. If limn-«> an =L, then a1, a2, • • • is bounded.
Proof Take any E > 0. There is an N such that
n > N � L - E < an < L + E.
Therefore the sequence aN+i. aN +2, • • • is bounded. (The larger of the numbers
IL - l IL + E l is a bound.)
E ,
And the finite sequence a1, a2, • • • , aN is bounded. We now get a bound M for the
entire sequence a1, a2, • • • : let M be the largest of the numbers IL - l
E , IL + l
E ,
la1I, la2 I • . . .
, laNI·
Theorem 6. If limn-«> an = 0, and bi b2, . • • • is bounded, then
lim anbn = 0.
n-+oo
lim anbn =LL'.

n-+«>
Definition. Suppose that there are numbers E > 0 and N such that
n > N � lan l > E.
Then a1, a2, • • • is bounded away from 0.
Theorem 8. If limn-«> an =L ¢ 0. then ai. a2, • • • is bounded away from 0.
Theorem 9. If a1, a2, • • • is bounded away from 0, and an ¢ 0 for each n, then the
sequence 1/a1, 1/a2, • • • is bounded.
Theorem 10. If limn-«> an =L, and L ¢ 0, and an ¢ 0 for each n, then
Jim l. = !. .
n-+oo an L
Note that we must require that an ¢ 0 for each n; otherwise the sequence of
reciprocals is not defined.
Theorem 11. If limn-«> an =L, limn-«> bn =L' ¢ 0, and bn ¢ 0 for every n, then
an L
lim =-.
n-+oo bn
-
L'
Theorem 12 (The squeeze principle). If limn-«> an =L, limn-«> en =L, and an ;£

bn ;£ en for every n, then limn-«> bn =L.
The Error in the
Appendix D Approximation a1� df
At the beginning of Section 4.4, we gave numerical examples of the use of the approxi
mation 6.f � df, and we found that when we checked our approximate answers
against the exact answers, the approximations looked good. But numerical approxima
tion methods are important precisely in those cases where their accuracy cannot be
checked in this way: if you can find the exact answer, then you use it; you don't get
an inexact answer to compare with it. This brings up the problem of setting a limit
on the error that results when you use df in place of 6.f The solution of this problem is
as follows. We have
6.f = f(xo + 6.x) - f(xo), df = f'(x0) 6.x.

y y
I
f I f
I
I
I
I I
I
------ r -- -1
I
p p
I I
I I
I I
I I
I I
x x
XQ x0+�x XQ x x0+�x
Applying the mean-value theorem (MVT) to the function f, on the interval from x0
to x0 + 6.x, we conclude that
f(xo + �� - f(xo) =
j'(x),
for some x between x0 and x0 + 6.x. (For x0 < x0 + 6.x, as in the figure, this result
follows. You should check that it also follows when x0 + 6.x < x0• Hereafter we
shall assume that 6.x > 0. The case 6.x < 0 needs to be checked separately.)
Therefore
6.f = f ' (x) !ix, x0 < x < x0 + !ix,
and
'
!if - df = f (x) 6.x - f'(x0) 6.x = [f'(x) - f'(x0)] !ix.
697
698 Appendix D
We now apply MVT to the function/', on the interval [x0, x] . MVT tells us that
�
f'(x - f'(xo)
= j"(x'),
X - X0
for some x' between x0 and x. Therefore
f'(x) - f'(x0) =
f"(x')(x - x0),
and
b,.f - df = f"(x')(x - x0) 6.x.
Now Ix - Xol � 16.xl. Therefore 16.f - df I � lf"(x')I b,.x2, where x' is between
x0 and x0 +b,.x.
y
It often happens that we can find a bound M for the numbers If" (x)I, for x
between x0 and x0 + 6.x. If so, we can conclude that
lt:if - dfl � M b,.x2.
For example, if f(x) = sin x, then j"(x) = -sin x, and lf"(x)I � 1, no matter
what x may be. Before giving further examples, let us write down the theorem that
we have proved:
Theorem 1. Suppose that f has a second derivative f", and that lf"(x)I � M, for
every x between x0 and x0 + b,.x. Then
lb,.f - dfl � M b,.x2.
Let us see how this applied to Example 1 of Section 4.4. Here we had
f(x) =
..J�, Xo = 25, b,.x = 0.4.
Now
f(x) = ·x1f2, f'(x) = tx-1/2,
f" (x) = -:lx-a/2 = __=..!.._ .

4R
We want to find a bound for this function, on the interval [25, 25.4]. As x increases,
so also does x3, and therefore so also does 4)x3. Hence the maximum value of
The Error in the Approximation !::if R> df 699
lf"(x)I on the interval [25, w), is 1/"(25)1. Now

1
1!"(25)1 = = -1 .
4.J253 500
We can therefore take M = -d-0. This gives
l�f - dfl � sh �x2 = -sh(0.4)2 = 0.00032.

In fact, we had
�f = 0.039841, df = 0.04, l�f- dfl = 0.000159.

Thus the error was considerably smaller than Theorem 1 predicted that it had to be.
The Continuity of
Appendix E Composite Functions
In Section 4.5 we stated the following theorem, with only rough indications of proof.
Theorem. The composition of two continuous functions is continuous. That is, if
lim g(x) = g(x0) = u0, (1)
x-xo
and
lim f(u) = f(u0), (2)
then
limf(g(x)) = f(g(xo )). (3)
Proof Let E be any positive number. By (2), there is a number 01 such that
lu - Uol < 01 => If (u) - f (uo)I < E.
Now take o1 as E, in the definition of hypothesis (1). By (1), there is a o > 0 such that
Ix - x01 < O => Jg(x) - g(x0)J < 01.
This is the o that we wanted: we have
Ix - Xol < O => Jg(x) - g(x0)J < 01 => lf(g(x)) - f(g(xo))I < E,
and so (3) holds.

Limits of composite functions are trickier than one might think. For example,
the following "theorem" is false.
Theorem (?). If lim.,__,.,0g(x) = u0, and lim,,_.,,J(u) = L, then lim.,_..,J(g(x)) = L.
To see that this is false, consider the following example. Let

for x ;tf.
f(x)_ =
g for x =
1,
1.
I
I
2 •
I
I
x
2
700
The Continuity of Continuous Functions 701
Let g be the same as/, and let x0 = u0 = 1. Then
lim g(x) = lim g(x) = 1 = u0,

and
lim f(u) = Iimf(u) = 1,
u-uo u-+1
but it does not follow that
(?) limf(g(x)) = 1.
x�xo
In fact,
x :;z6 1 => g(x) = 1 => f(g(x)) =
2,
and so
limf(g(x)) = 2.
x-+1
Appendix F The Error 1n Simpson1s Rule
The results of our calculations in Section 4.8 and in later sections suggest two questions:
1) Why does Simpson's rule give such good approximations?
2) In a particular computation, how can we tell how good the approximation is?
That is, how can we determine a bound for the error?
These questions have the following answer. In the theorem below, f(4) denotes
the fourth derivative off Thus j<ll is f', j<2l = f", j<3l = DJ" = J'", and J<4l =
Dj<3l. As usual,
Yo= f(-k), Y1 = f(O), Y2 = f(k).
Theorem 1. If f has a fourth derivative, on the interval [-k, k], then the error in
Simpson's rule is equal to
E(k) = :�j<4l(.X),
where xis some number between -k and k. That is
0
l
k k
l(x) dx - - (Yo + 4Yi + Y2) =
- 1<4l(.X) (-k < x< k).
- 3 90
k
It follows, of course, that if Jj<4l(x)J � M ( -k < x < k), then
JE(k)J � 9�k5M.
The latter is the statement which is most convenient to apply. Before proceeding to
the proof of the theorem, let us look at an application of it.
Example. Suppose that we want to compute
to five decimal places. Here
l(x) = .!. = x-1,

x
1111(x) = -6x-4,
702
The Error in Simpson's Rule 703
Since/(4) decreases as x increases, its maximum value on the interval [1, 2] isj<4>(1) =
24. Therefore /j<4>(x)/ � 24 (1 � x � 2). Therefore, if we cut up the interval
[1, 2] into 2n parts, each of length k, we have
/E(k)[ � 910k5 24. •
We want
/E(k)/ < 5 10-s, ·
for the fifth decimal place in our approximation to be correct. Thus we want to take
k such that
9�k5 24 < 5 10-s,
• ·
or
ks < H . 5 . 10-s.
Arithmetically, this reduces to
ks<\"-. 10-s,
which surely holds if k = 0.05. Therefore E(0.05) < 5 10-s. ·
This example was selected for its simplicity. For most functions, the calculation
of fourth derivatives is tedious.
We proceed to the proof of Theorem 1. Let F be any function such that
F' =f
(How do we know that there is such a function?) Then
f/(x) dx = F(k) - F(-k),

so that
E(k) =
i-k k
f(x) dx - - (Yo 4y1 +
Y2) +
k
3
= F(k) - F(-k) - � [f(-k) 4f(O) + + f(k)].

3
Therefore
E'(k) = f(k) + f(-k) - � [-j'(-k) + f'(k)] - Hf(-k) + 4f(O) + f(k)]

3
= if(k) + if( -k) - tf(O) + � j'(-k) - � j'(k),

3 3
E"(k) = if'(k) - if'(-k) -� f"(-k) H'(-k) - tf'(k) - � f"(k)

+
= H'(k) - tf'(-k) - � f"(k) - � f"(-k),
E"'(k) = tf"(k) tf"(-k) - � f"'(k) - tf"(k)

+ � f"'(-k) - tf"(-k) +
= - � [f"'(k) - f'"(-k)].
3
704 Appendix F
We need all of these formulas, not just the last. It is easy to check that
E(O) = E'(O) = E"(O) = E'"(O) = 0.

From now on, k is going to be regarded as a constant. For each ton [-k, kJ, let
E(k)
G(t) E(t) - t5•
k5
=
It is easy to check that

G(O) = G(k) = 0,
and
G'(O) = G"(O) = G111(0) = 0.
y y
On the interval [O, k] we apply the mean-value theorem (MVT) to the function G.
This gives
0<X1<k.
We next apply MVT to the function G', on the interval [O, x1]. This gives
G"(x2) = 0, 0<X2<X1.
Applying MVT to G", on [O, x2], we get
G"'(x3) = 0,
By a straightforward calculation,
E(k)
G"'(t) E"'(t) - (60t2)
k5
• =
E(k) 2
- !._ [f"'(t) - j"'(-t)] - 60 t•
k5
=
Setting t = x3, and solving for E(k), we get
k5 k5 f'"(xa) - f"'(-xa)
E(k)
f"'(xa) - f"'( -xa)
90
= - - · = - · ·
180 . X3 2X3
We now apply MVT for the last time. By MVT there is an .X, between -x3 and x3,
such that the second fraction on the right is equal to j<4>(x). This gives
k
E(k) s 1<4>(x) (-k<x<k),
90
=

The Idea of a
Appendix G Measurable Set
If you reexamine Section 2.10, you will see that at the end of the section we were in a
peculiar position: we had gotten an answer for the area under the graph of y = kx2,
from x = a to x = b, but we were not in a position to prove it, because we had no
definition of area. The trouble, however, is easy to remedy.For the sake of simplicity,
consider first the case in which R is the region under the graph of y = x2, from
x = 0 to x = h. In Section 2.10, we proved the following two things:
1) There is a sequence R1, R2, of polygonal regions containing R, with areas
A1, A2,
• • •
, such that
limAn =
• • •
L.
n-+co
(Here Rn was the union of the outer rectangles, An was (h3/3)(1 + l / n) ( l + 2/n),
and Lwas h3/3.)
2) There is a sequence R�, R�, ...of polygonal regions lying in R, with areas
A�, A�, ... , such that limn�co A� is the same number L.
(Here R� was the union of the inner rectangles, A� was (h3/3) zf�i (i - 1)2, and Lwas
h3/3, as in condition 1.)
These ideas can be used to give a definition of area, in the following way.
Definition. Let R be a region in the plane. If R satisfies conditions (1) and (2),
then R is said to be measurable, and the number Lis called its area.
Under this definition, the plane regions discussed in Chapter 2 are measurable,
and their areas are the numbers that we computed. The same conclusion follows
whenever we compute an area by means of a definite integral. In Section 7.8 we
showed that every continuous function is integrable.This gives the following:
Theorem. Let f be continuous and nonnegative on [a, b], and let R be the region
under the graph off Then R is measurable, and the area of R is
A =ff(x) dx.
Proof Take a sequence of nets
over [a, b], with IN;I - 0. For each i, let A; be the upper sum S(N;) and let A� be the
705
706 Appendix G
lower sum s(Ni). Then A; is the area of a polygonal region containing R, and A; is
the area of a polygonal region lying in R, as in the definition of a measurable set.
And
Ai � ff(x) dx � A;.
The sequences A1, A2, • • • and A�, A�, ... have the same limit, namely, the integral.
Therefore R is measurable, and its area is the integral.
This theorem can be extended so as to apply to the region between the graphs of
two continuous functions.
It might seem that we could simplify the preceding discussion by defining the area
to be the integral, in the first place. But this will not work. The point is that some
regions can be represented in many different ways as the regions between the graphs
of two continuous functions. Different directions for the axes give different limits of
integration, and also different integrands, even for so simple a figure as an ellipse.
In the theory that we have just developed, we know that all the resulting integrals
give the same answer, because they all give the right answer for the area of the region.
But if we defined the area to be the integral, we would have the problem of showing,
by the methods of calculus, that all the integrals have the same value, and this would
be hard.
Proof of the
Appendix H Northeast Theorem
The Northeast theorem asserts that if
limf (t) = lim (t) = oo, (1)

t-+oo t-cc g
and
t
lim g'( ) = L, (2)
t->CO
f'(t)
then
( t)
lim g
j t)
= L.
(
t�co
To start the proof, we first observe that since g'(t)/f'(t)->- Las t ->- oo, we must
have f' ( t ) � 0 when t is sufficiently large, say, for
t � t0• Since we are taking the
limit as t ->- oo, we mayregard t0 as the initial point of the path. Since g' (t)lf' (t) ->- L,
the function g'If' must be bounded on some interval [t1, oo) (t1 � t0). The reason is
that for every E > 0, we have
(t)
L - € < g' < L + €'
f'(t)
fort � a certain t1• Therefore g'lf' is bounded on the interval [t1, oo). We now take
t1 as the initial point of the path.
As a further simplification, we translate the point (j(t1), g(t1)) to the origin,
replacing f(t) and g(t) by F(t) = f(t) - f(t0), G(t) = g(t) - g(t0), where c � t0•
ObviouslyG'/F' is bounded, and
t)
lim G'( = L'
t->oo F'(t)
because F' = f' and G' = g'. And if we can prove that
lim G(t) = L'

t-> co F(t)
then it will follow immediatelythat

t)
lim g( L.
J( t )
=
t->oo
707
708 Appendix H
The reason is that
g(t) = G(t) + g(t1) =

G(t)/F(t) g(t1)/F(t)
+
f(t) F(t) + f(t1) 1 + f(t1)/F(t)
Therefore it will be sufficient to prove the theorem in the following special form.
Theorem A. Let F and G be differentiable functions on the interval [ti. oo ), such that
lim F(t) lim G(t) (3)

=
= oo,
t-+ 00 t-+ 00
G'(t)
lim
=
L' (4)
t-+oo F'(t )
F'(t) � 0 fort � t 1, (5)
G'/F' is bounded, (6)

and
F(t 1) G(t 1) 0. (7)
= =
Then
G(t)
lim L. (8)
=
t-+ oo F(t)
We now make a final simplification: under the conditions of Theorem A, the

locus of the path must be the graph of a function <f, defined on the interval [O, oo),
with
'( G'(t) (x
0, and <f x) F(t)).
=
<f(O) = =
F'(t)
m y=<J>(x)
x=F(t)
-
- G1(t)
. m-</> (x) -
I
F'(t)
Evidently
G(t) <f(x) (x F(t)).
= =
F(t) x
Therefore, rewriting Theorem A in terms of <f, we get the following:

Proof of the Northeast Theorem 709
Theorem B (The Northeast theorem, rectangular form). Let <P be a function on the
interval [O, oo ) such that
,
¢(0) = 0, (9)
<P' is bounded, (10)
lim ¢(x) = oo, (11)
X-+<Xl
and
lim </J'(x) = L. (12)
x-+oo
Then
(x)
lim </J = L. (13)
X-+ co X
The proof is in several steps.
Step 1.
. <f>(x2) - </>(x)
l lffi -
- L.
X-><Xl X 2 - X
Proof By the mean-value theorem (MVT), for each x there is an x, between x and x 2,
such that
�
</>(x2 - </>(x)
= <f>'(x).
x - x
As x -+ oo, x-+ oo, and so <f>'(x)-+ L. Therefore the fraction on the left also-+ L
y y
Step 2.
lim
X-+<Xl
[<f>(x2� - <f>(x) - </>(�2)]
X - X X
= 0.
Proof
<f>(x2) - <f>(x) <f>(x2) x<f>(x2) - xrf>(x) - x<f>(x2) + <f>(x2)
---=
x2 - x x2 x2(x 1) -
=
__l _ <f>(x)
x 1 x -
[ +
x2
]
</>(x2)
·
710 Appendix H
By MVT, there is an .X between 0 and x such that
</>(x) = </>'(.X).
x
Since </>'is bounded, it follows that <f>(x)/x is bounded. Therefore </>(x2)/x2 is bounded.
Since -1/(x - I)-+ 0, it follows that
_ _
x
1_ </> ( x)
[
- 1 x
+ </>(x2)
x2
] -+ O.
Step 3.
(x2)
lim </> = L,
X-+CO X2
by Steps I and 2. From Step 3 it follows immediately that </> (x)/x -+ L, which was
to be proved.
Proof of the
Appendix I Formula for Path Length
Here we complete the proof of Theorem 1 of Section 9.6, which asserts that the
length of a path is given by the formula
s = f.fJ '(t)2 + g'(t)2 dt,
wheref and g are the coordinate functions andf' and g' are continuous. The notation
is that of Section 9.6. By definition,
n
INl-+Oi=l
s = lim L P;_1P;;
we know that
i=l
I Pi-1Pi iI=l .JJ '(ii)2 + g'(f;)2 �ti,
=
where i i and i; are between t;_1 and t;; and
lim .JJ'(i;)2 g'(i;)2 �ti= fb.JJ'(t)2 + g'(t)2 dt,

INl-+O
+
Ja
by definition of the integral. Therefore what remains to be proved is that
['f .JJ'(ii)2 g'(fD2 �ti I .JJ'(ii)2 + g'(ii)2 �ti] = o.

INl-+O i=l
lim
i=l
+ -
In terms of the definition of limw1�0, this can be restated as follows:
Lemma. For every c: > 0 there is a o > 0 such that
Proof 1) Sincef' and g' are continuous on [a, b], so also is the functionf'2 + g'2•
Let M be such that
f' (t)2 + g'(t)2 � M for a � t � b.

2) The function z f-4 .J; is continuous on the interval [O, M]. Therefore this
function is uniformly continuous on [O, M] (Section 7.8, Theorem 4). Therefore,
711
712 Appendix I
given any e > 0, there is a () > 0 such that

- - €
iz - z'I < () => IJz -Jz'I < --.
b - a
3) The function g'2 is continuous, and therefore uniformly continuous, on [a, b].
Therefore, given () > 0 there is a o > 0 such that
It: - t/ < o => lg'(i/)2 - g'(idl < o.

4) Theo given by t3) is the o that we need:
INI < O => If[ - iii < O (for every i)
=> lg'(n)2 - g'(fi)21 < () (for every i)
=> IJj'(fi)2 + g'(ff)2 - Jf'(fi)2 + g'(fi)2 < 1 b

�a
=> I[ Jf'(fi)2 + g'(ti)2 - Jf'(fi)2 + g'(fi)2 Llti < J l b � a Llti
=> � I[ Jf'(fi)2 + g'(fD2 - Jf'(ii)2 + g'(fi)2] Llt;I
i
n
< -- L Llti -- (b - a)
E E
= = e.
b - ai�1 b - a
Since the absolute value of the sum is less than or equal to the sum of the absolute
values, it follows that o satisfies the conditions of the lemma.
A Method for
Constructing the
Appendix J Complex Numbers
In Section 10.11 the complex numbers were presented as a formal system of symbols
a + bi, with i2 = -1. We shall now define a mathematical system of this kind, and
show that it has the properties that we want. There are various ways to do this.
The following method has the advantage of copying the pattern of the manipulative
processes that we would be using anyway. It has the further advantage of introducing
ideas that will be useful later, in modern algebra.
Let P(x) be the set of all polynomials p(x) = .2f=0 aixi. In P(x) we can add and
multiply. We know that in P(x), these operations obey the CAD laws, that is, they
are commutative, associative, and distributive:
pq = qp' p +q = q + p'
p(qr) = (pq)r, p + (q + r) = (p + q) + r,
p(q + r) = pq +pr.
These follow immediately from the corresponding laws for the real numbers p(x)
which are the values of our polynomial functions.
Two polynomials p, q will be called congruent modulo 1 + x2 if their difference
is a multiple of 1 + x2• We then write
p(x) = q(x) mod 1 + x2,

or briefly p = q. Thus
p(x) = q(x) if p(x) - q(x) = r(x)(l + x2),
for some polynomial r(x). For example, x3 = -x, because x3 - (-x) x3 + x = =
x(l + x2); and x2 -1, because x2 - (-1) = x2 + 1 =I· (1 + x2).

=
In the following theorem, the primes do not indicate differentiation; p, q, p',

and q ' are supposed to be any polynomials.
Theorem 1. If p = p' and q = q' , then p + q = p' + q', and pq = p'q'.
This is a straightforward calculation. Given
p' = p + r · (x2 + l), q ' = q + s · (x2 + 1),

we get
p' + q ' = p + q + (r + s) · (x2 + 1),
713
714 Appendix J
so that p' + q' = p + q. And

p'q' = pq + (ps +qr)(x2 + 1) + rs(x2 + 1)2
= pq + ( ps +qr + rsx2 + rs)(x2 + 1),

so that p'q' = pq.
For eachp= p(x) in P(x), let p be the set of all polynomials that are congruent
top. That is,
p= {p' IP' =p}.

These sets are called congruence classes. Let C be the set of all such congruence
classes p . In C we define addition and multiplication by the conditions
p +q = p +q, and p . q = pq.

These definitions make sense, because the congruence class that contains p +q
does not depend on the choice of p and q; it depends only on the congruence classes
p and q. The same applies for the product.
Theorem 2. The CAD laws hold in C.
Proof We know that these laws hold in P( x). Therefore, under our definitions of
addition and multiplication in C, we have
p. q = pq= qp= qp;
p + q = p +q= q + p = q +p;
p (q +F ) = p q + r= p(q + r)= p q +pr= pq +pr= Pi] +pr;

·
and similarly for the other laws.

We now observe that
and
Therefore every power of x is congruent to a linear polynomial. Therefore:
Theorem 3. Every polynomial p(x) is congruent to a linear polynomial a + bx.
For example,
p(x)= 7x7 - 5x3 + 6x2 - 3
= 7(x2)3 • x - 5x2 • x + 6x2 - 3

= -7x + Sx - 6 - 3
= -2x - 9.
In fact, the system C that we have just defined has all the properties of the number
system that we wanted. To describe it in the familiar notation, we denote each
congruence class p(x) by the formal expressionp(i), in which xis replaced by i. Thus,
A Method for Constructing the Complex Numbers 715
if p(x) is as in the preceding example, we have
p(x) =p(i) =7i7 - 5i3 + 6i2 - 3
=7(i2) 3 • i - 5i2 • i+ 6i2 - 3
-7i+ 5i - 6 - 3
-9 - 2i.
Here we have simplified by substituting -1 for i2, and this is right; since
x2 = -1,
we have
x2 =i2 =-1;
any congruence between two polynomials p(x) and q(x) gives an equation between
their congruence classes p(i) and q(i) And our number system satisfies the conditions
.
for a field, given in Chapter 1: the CAD laws hold; there are numbers 0 and 1,
such that if
z =a+ bi,
then
0. z = 0, and 1 · z =z;
for each z there is a number

-z =-a - bi,
such that
z + (-z) =0.
Finally, every z -:;tf:. 0 has a reciprocal. To prove this, we first observe that
a + bi =0 => a =b =0.
The reason is that
a + bi =0 <=> a + bx = 0 mod 1 + x2
<=> a+ bx = r(x)(l + x2).
Here r(x) must be 0, because otherwise r(x)(l + x2) would be of degree � 2. Since
r =0, a + bx is the zero polynomial, and so a =b =0. Similarly,
a - bi =0 => a =b =0,
and so
a + bi -:;tf:. 0 => a - bi -:;tf:. 0, and a2+ b2 > 0.
We shall now find a reciprocal for any a+ bi -:;tf:. 0, by assuming, experimentally,

fractions with complex denominators make sense, and then checking that our answer
works:
1 1 a - bi a bi
--- =--- . =
.
--- ---
a + bi a + bi a - bi a 2 + b2 a 2 + b2
716 Appendix J
The last expression really is the reciprocal of a + bi, because
(a + bi) [a2 a b2 - a2 bi b2]

+ +
2
1
a + b
. .
2 (a + bi)(a - bi) =
a2 + b2
a2 + b 2
. = 1.
To sum up:
Theorem 4. C is a field.
Note that when we passed from P(x) to C, by forming congruence classes modulo
+ x2, the algebraic character of the system changed: in P(x), only the constant
polynomials p(x) = a ¥- 0 have reciprocals; but every congruence class p(i) =
p(x) ¥- 0 has a reciprocal.

To set up this number system, and check its properties, we need to use equivalence
classes of polynomials (or some equivalent device). But now that we have such a
system, we need not remember where we got it. In the future, we shall never have
occasion to refer to the fact that a + bi was defined to be
{p(x) I p(x) = a + bx}.

Iterated Limits.
APPENDIX K Mixed Partial Derivatives
In discussing double limits of the type
lim f(x, y),

(x,y)-+(xo.Yo>
for functions of two variables, we assume that f is defined in a neighborhood of
(x0, y0), except, perhaps, at the point (x0, y0) itself. Under the same conditions,
we can discuss limits of the type
lim lim f(x, y) and lim lim f(x, y).

a::-+xo 11-+110 'Y-+Yo X-+Xo
These are called iterated limits.

There are simple examples in which the iterated limits both exist but have different
values. That is, the order in which we take the limits may make a difference. Consider
2 2
x Y
_
f(x, y) =
2 + 2
x y
Here
. . x 2 - y2 . x
2
bm hm hm 2 = 1,
2 2
=
x-+O 11-+0 X + y x-+O X
and
2
2 - - 2
x y y
lim Jim = lim -- = -1.
2 + 2 2
y-+0 x-+O X y y-+0 y
This sort of thing cannot happen, however, if/is continuous in D and the double
limit exists. That is, we have the following theorem:
Theorem 1. If/is continuous in D, and
lim f(x, y) = L,
(x,v)-+(xo.vo>
then
lim limf(x, y) = lim limf(x, y) = L.
Proof Iff(x0, y0) is defined at all, thenf(x0, y0) must be L. Iff(x0, y0) is not defined,
we define it to be L. Thus we may assume that f is continuous in a neighborhood of
(x0, y0), including (x0, Yo)·
717
718 Appendix K
The proof is now trivial: For each x, we have
limf(x, y) = f(x, Yo), (1)
because f is continuous. Therefore, for the same reason we have
lim limf(x, y) = limf(x, Yo)= f(xo, Yo) · (2)

Similarly,
Jim Jim f(x, y) = limf(x0, Y) = f(xo, Yo)· (3)

11-+Yo X-+Xo Y-+Yo
In (1) , all we are saying is that ifjis continuous (as a function of two variables)
then the slice functions, for each fixed x, are also continuous. This may be easier
to keep straight if we rewrite (1) in the form
Iimf(a, y) =
f(a, Yo), (1')
11-+Yo
which reminds us that xis fixed as y ---+ y0• Equations (2) and (3) follow by repeated
applications of the same principle.
Since in some cases the two iterated limits are different, we always have to
investigate, in the cases where we need to know that they are the same. One such
case comes up when we consider the "mixed partial derivatives" fxy and hx· If we
write in full the definitions of fx,u(x0, y0) and fux(x 0, y0), we see that they are iterated
limits of the same function:
1. fxCxo, Yo + 6.y) fx(Xo, Yo)

fxY(Xo, Yo) - im
_ -
Ay->O 6.y
. 1 f(x o + 6.x, Yo + 6.y) - f(x o, Yo + 6.y)
=Im
I Im
-
[i·
Ay -> 06.y A ->O
x 6.x
. f(xo + 6.x, Yo) - f(xo, Yo)
- 1Im
]
Ax-+O 6.x
= J im Jim -1- {[f(xo + 6.x, Yo + 6.y) - f(xo + 6.x, Yo)]

Ay-+OAx->06.y 6.x
- [f(xo, Yo + 6.y) - f(xo, Yo)]}.
Note that in the last step we have changed the order of two of the terms. The reason
for this will soon be clear. Thus we have
fxvCxo, Yo) :::;: Jim Jim -1- F(6.x, 6.y),

Ay-+0 Ax->O6.y 6.x
where F(6.x, 6.y) is the function defined by the expression in the braces. An entirely
analogous calculation tells us that fux(x0, y0) is the iterated limit, in reverse order, of
exactly the same function:
fvx(xo, Yo)= lim J im -1- F(6.x, 6.y).

Ax->OAv->06.y 6.x
Iterated Limits. Mixed Partial Derivatives 719
Let us now investigate the function F. This function can be regarded as the
difference of two values of the function
cp(x) =f(x,Yo + �y) - f(x, Yo)·

That is,
F(�x, �y) = cp(x0 + �x) - cp(x0).
Applying the mean-value theorem MVT to the function <fa, on the interval from
x0 to x0 + �x, we get
F(�x,�y) = <f'(x) �x = [fx(x,y0 + �y) - fx(x,y0)] �x,

where xis between x0 and x0 + �x. Now the quantity in brackets can be regarded
as the difference of two values of the function
ip(y) =f.,(x,y);
we have
F(�x, �y) = [VJ(Yo + �y) - VJ(Yo)J �x.
By MVT,
VJ(Yo + �y) - VJ(Yo) = ip'(y) �y,
where y is between Yo and y0 + �y. This gives
F(�x,�y) =fxy(x, y) �Y �x.
Suppose now that fxy is continuous. We then have
[Since Ix - x01 < �x and IY - Yol < �y, we must have (x,y) � (x0,y0) as
(� , �y) � (0, 0).)
x
Let us now take stock. We had
1
1) fxy(X0,y0) = lim lim F(�x, �y) (by definition).
t.y�o t.x-�o Ll X Ll y
A A
2) Iffxy is continuous at (x0 ,y0), then the double limit

1
Jim F(�x, �y)
(ilx,t.y)�(o,o) �x �y
--
exists, and is equal to fxy(x0,y0). This is what we have just proved. Suppose now
that fy x is also defined in a neighborhood of (x0, y0), and is continuous. Then
.
3) fy.,(Xo, Yo) =6.x->O

lim lim 1 F(�x,�y).
6.y->O�X �y
--
Since the double limit in (2) exists, the iterated limit in (3) must be equal to it.
Therefore fxy(x0,y0) =fvx,(x0,y0). Thus we have proved the following theorem:
Theorem 2. Iffx v and fvx exist and are continuous, in a domain D, thenfxv = f11x-
720 Appendix K
By repeated applications of this theorem, we can draw a more general conclusion

about partial derivatives of higher order: if these exist and are continuous, then all
that matters is the number of times we differentiate with respect to each variable.
Thus, for example,
Proof
Therefore
(hua)v = (hx11)11,
andfxvxv = fxxvv• which was to be proved.
Warning: It is not true that if/xv exists and is continuous, thenJ;.."' also exists and
is the same. To see this, let
f(x, y) <f>(y),
=
where </> is any function which is not differentiable. Then
fx( x,y) = 0 for every x, y,

because the slice functions for constanty are constant. Therefore, trivially,fx11 exists,
and fxy( x,y) = 0 for every x, y, so that fxu is continuous. But /11 is not defined,
because </>'(y) is not defined. Therefore fvx is not defined either.
Possible Peculiarities of
APPENDIX L Functions of Two Variables
Some of the definitions that we have used in Chapter 14, and the hypotheses of some
of our theorems, may seem needlessly strong. In fact, they are not. The theory of
functions of two variables includes some rather odd and unexpected phenomena;
and if we want to draw simple conclusions, we need to use hypotheses sufficiently
strong to rule out the oddities. Some of these are as follows.
Example 1. There is a function f such that
1) all the slice functions f(x, y0) (with y held constant) are continuous, and
2) all the slice functions f (x0, y) (with x held constant) are continuous, but
3) f is not continuous.
Proof In the first quadrant of the xy-plane we take an infinite sequence Di. D2, • • •
of circular disks, not intersecting each other, with radii approaching 0, and approaching
the origin as a limit.
y z
.... y
As indicated in the figure on the left, we take these disks with their centers on the
line y = x, in such a way that no horizontal or vertical line intersects more than one
of them. This is easy to arrange, because we can make the disks as small as we want.
If (x, y) lies in none of the disks D;, then we definef(x, y) to be 0. Over each disk
D;, the graph off is a "blister" of height 1, shown on the right.
Obviously f is not continuous at (0, 0). But all the slice functions ef>(x) = f(x, y0)
are continuous. Since no horizontal line intersects more than one of the disks D;,
it follows that the graph of cf> looks, at worst, like the graph shown below.
721
722 Appendix L
z z
1 ----
-+--�-....._x
On the left we see what happens if the line y = y0 passes through the center of a disk.
If this doesn't happen, then the maximum of cp is smaller; and of course cp(x) may
be 0 for every x. Similarly for the slice functions for constant x.
Here, of course, the slice function
<Prr14(t) = J
(J2' Jl)
is not continuous. Its graph is shown below. But even if a function f has slice functions
<Pa(t) = f (x0 + t cos a, Yo + t sin a)

which are continuous, for every (x0, y0) and every a, we still cannot conclude that/ is
continuous.
z
1 -
Example 2. There is a function/ such that
1) Every slice function <Pa is continuous, but

2) f is not continuous.
Proof. Consider the parabola y = x2, in the xy-plane. Between the parabola and the
x-axis we take a sequence D1, D2, • • • of circular disks, with radii approaching 0,
approaching the origin as a limit. On each disk, the graph off is a blister of height 1,
as in Example 1; everywhere else,f (x, y) = 0.
y
Possible Peculiarities of Functions of Two Variables 723
As before, f is not continuous. But all the slice functions are. The reason is that
no line L intersects more than a finite number of the disks D;:
y y y
L L
If L does not pass through the origin (as on the left above) or if L passes through the
origin and has negative slope (as in the center), this conclusion is trivial. The interest
ing case is shown on the right. Here L passes through (0, 0) and has positive slope.
Near the origin, in the first quadrant, the line lies above the parabola and the disks
lie below it. Therefore L cannot intersect infinitely many disks. Therefore the slice
functions defined along any line L are continuous.
Our next peculiar function is going to be continuous. We recall that in Section
14.8 we proved the following theorem:
Theorem A. Given a function f, defined in a neighborhood of (x0, y0) . For each tJ., let
cf>a = cos tJ., Yo + t sin tJ.) .
f (x0 + t
Suppose that c/>�(O) =0 for every tJ.; and suppose that there is a number o > 0
such that for !ti < o we have c/>�(t) < 0 for every tJ.. Then/has an ILMax at (x0, y0) .
y
Yo -- - - -
The reason is that for each tJ., cf>a(O) is the maximum value of cf> on the interval
( -o, a). See the figure below.
z
-iJ
<t>&(O)=O, <t>&'(t)<O for- iJ<t<iJ
Here it is essential that there be a single number a > 0 which works for every tJ..
The following plausible-looking variation on Theorem A is false.

724 Appendix L
? Theorem B? Given a functionf, defined in a neighborhood of (x0,y0). For each ex, let
</>it) = f(x0 + t cos ex, Yo + t sin ex) .

If (1) each function rp,, has an ILMax at 0, then (2) f has an ILMax at (x0, y0).
The falsity of this "theorem" is demonstrated by:
Example 3. There is a continuous function f such that (1) every slice function
through the origin has an ILMax at the origin, but (2) f does not have an ILMax
at the origin.
This is similar to Example 2. As before, we take a sequence of disks lying under a
parabola. We define f(x, y) to be 0 everywhere except on the disks. But this time,
we take the blister over the ith disk D; in such a way that its height is 1/i. Now our
function f is continuous.
y
As before, no line Lin the xy-plane intersects more than a finite number of the disks.
Therefore every slice function
rp,,(t) = f(t cos ex, t sin ex)
is equal to 0 in a neighborhood of 0. That is, for each ex there is a o,, > 0 such that
rf>a(t) = 0 for !ti < o". Therefore every <Pa has an ILMax at 0. But obviously f does
not have an ILMax at (0, O);f(O, 0) = 0, but every neighborhood of (0, 0) contains
a disk D;, on which f(x, y) > 0.
z
The trouble here is that while for every ex there is a o,, with the desired property,
there is no one o which works for every ex. If ex > 0 and ex !::::; 0, then o" !::::; O; and so
inf {o"} = 0. If you reread the proof of Theorem 3, Section 14.8, you will see how
this trouble was avoided: using the continuity of fxv• fxx, and fvv, we found a single
o > 0 which worked for every ex . Thus the proof of Theorem 3 was not merely
complicated in a technical way but was also subtle, in a way which is not likely to be
understood unless we re-examine the proof in the light of Example 3.
Maxima and Minima, for
APPENDIX M Functions of Two Variables
Here we give a brief sample of the way the theory of continuous functions of one
variable can be extended so as to apply to functions f: D ---+ R, where D is a domain
in a Cartesian space Rn (n > 1).
Theorem 1. Let D be a closed rectangular region in R2, defined by the inequalities
a ;;a x ;;a b and c ;;a y ;;a d. Let f be a continuous function D ---+ R. Then f is
bounded above.
Proof Suppose that f is not bounded above. We shall show that this leads to a
contradiction.
The region D is the union of four closed rectangular regions, shown in the figure
on the left below. These will be called quarters of D. These are like the "halves" of an
interval [a, b], as defined in Section 5.6; and they are going to be used in exactly the
same way.
y y
--
di
'!� = C---, =�1----+----1

I
I I
:I I
I
I x
J,[
Ci
- -
I
I
I
I
Di
x
a a+b b a; bi
-2
-
Ii
Following the pattern of Section 5.6, we say that a closed rectangular region D' is
good ifjis bounded on D'; and D' is bad if/is unbounded on D'. We are assuming
that the giyen D is bad. It follows that one of the quarters of D must be bad. (Why?
See Lemma 1 on page 240.) Let D1 be a bad quarter of D. Similarly, let D2 be a bad
quarter of D1• Proceeding in this way, we get a sequence
of closed rectangular regions, each of which is bad, such that for each i, D;+1 is a
quarter of D;. As indicated in the figure on the right above, let/; and l; be the closed
intervals which are the projections of I; and l; onto the x- and y-axes. Then fi, 12, •••
is a nested sequence. By the Nested Interval Postulate (NIP) there is an x which lies
725
726 Appendix M
on each interval !;. Similarly, there is a y which lies on every interval Ji. It follows
that the point P (x, y) lies in every region Di.
=
But f is continuous at P. Thus for every E > 0 there is a o > 0 such that
PP < 0 => If(P) - f(P)I < E,
and this means that

PP < o => f(P) < f(P) + E;
that is, f is bounded on the circular disk with center at P and radius o. But this
circular disk contains some Di, because P lies in all the D/s, and the height and width
of D; both ___,.. 0 as i ___,.. oo. Therefore Di must be good for some i, which contradicts
our hypothesis.
Theorem 2. If/is continuous on a closed rectangular region D, then/has a maximum
value on D.
The proof is exactly like the proof of Theorem 3 of Section 5.6. Let k sup f =
If k =f(P) for some P, then/has its maximum value at P. If f(P) < k for every P
in D, let
(P) - 1
g - k -f(P)
Then g is continuous on D, but is not bounded above; and this contradicts Theorem 1.
As before, the existence of maxima gives, as a corollary, the existence of minima:
Theorem 3. Iffis continuous on a closed rectangular region D, thenfhas a minimum
value on D.
(Proof Any maximum value of -f is a minimum value off)
The same scheme works for continuous functions defined on an "n-dimension
interval"
D {(x1, x2,
= , Xn) I a;� X; � b; for i
• • • 1, 2, ... , n}. =
We use a subdivision process just as in R1 and R2, dividing our "interval," at each
stage, into 2n parts.
An Exact Definition of
APPENDIX N the Idea of a Function
In Chapter 3 we explained that a function f: A ---+ B is defined if a rule is given under

which for each element of the set A, there exists one and only one corresponding
element of the set B.
This formulation of the idea of a function is adequate for the purposes of elemen
tary calculus, and so, in the text, we have let it stand. But eventually we need a more
exact definition, now to be explained. To see what we are driving at, in the new
definition, consider the function
f: R-+ R
2 for every x.
: x 1--> x
The graph is a parabola.
We shall now approach our new definition in the following two steps.
Step 1. We regard the function as being indistinguishable from its graph, so that the
function/becomes a set of points P, in a coordinate plane. (In this case, the function
is a parabola.)
Step 2. We regard a point P, in a coordinate plane, as indistinguishable from the

ordered pair (x, y) which gives its coordinates.
P=(x,x2)
727
728 Appendix N
"·
The graph now becomes a collection ofordered pairs ofreal numbers, namely, \
f={(x, x2)}.
This collection ofordered pairs has the property that each real number x is the first
term ofexactly one ordered pair (x, y) in the set. (This is because the graph intersects
every vertical line in exactly one point.)
This final description off, as a collection of ordered pairs {(x, x2)}, can be
generalized to apply to functions ofany kind, on any domain. The final definition is
as follows.
Definition. Let A and B be sets. Let f be a collection ofordered pairs(a, b). Suppose
that
1) if (a, b) belongs to f, then a belongs to A and b belongs to B, and
2) every element a ofA is the first term ofexactly one pair belonging to f
Then f is a function of A into B, and we write
f: A-+B.
For each a in A,f(a) denotes the second term ofthe ordered pair whose first term is a.
For the function whose graph was the parabola, we had

A=B=R, f ={(x, x2)}.
Similarly, for the function Sin-1, we have
A= [-1, 1], B=R (or B= [-rr/2, TT/2]),
{
f=Sin-1= cx, y) \ -1 � x � 1, -� �y � �, x=siny}.
Here again the idea is that the function is defined to be its graph, and the graph is
regarded as a set ofordered pairs ofreal numbers. Note, however, that our general
An Exact Definition of the Idea of a Function 729
definition of a function applies in a variety of other contexts, in which A and B

may be sets of quite different kinds. For example, A may be a vector space, or a
region in a vector space, or the set of all positive integers. In this book, B has usually,
but not always, been a set of numbers.
Eventually, the exact definition of a function becomes useful, as a matter of
technique. A little reflection will convince us, however, that it cannot be quite right,
at any stage, to define a function to be a rule. The point is that rules are formed with
words (or with a combination of words and symbols). Therefore there is such a
thing as a fifteen-word rule. But surely there is no such thing as a fifteen-word
function. Analogously, there is such a thing as a three-syllable name, but there is
no such thing as a three-syllable man. As a matter of common sense, a man is
different from his name; and in the same way, a function is different from the phrases
and formulas that we use to describe it. Therefore functions ought to be defined in
such a way as to be mathematical objects.
730 Appendix N
Table 1
Natural Trigonometric Functions
�
Angle
� fu
\
�
Angle
�
/ I
0° 0.000
n t_1
��.� � � � �l� _g_e_
0.000 1.000
�
0.000
ful�
1°
2°
0.017
0. 035
0.017
0.035
1.000
0 999
0.017
0.035
46°
47°
0.803
0.820
0.719
0.731
0.695
0 682
1.o:l6
1.072
I
3° 0.052 0. 052 0 999 0.052 48° 0.838 0.743 0 669 1.111
4° 0.070 0. 070 0.998 0. 070 49° 0.855 0.755 0.656 1.150
5° 0.087 0 087 0.996 0.087 50 ° 0.873 0.766 0.643 1.192
6° 0.105 0.105 0.995 0.105 51° 0.890 0.777 0 ()29 l.2:l5
I
7° 0.122 0.122 o.993 o.12:i 52° 0.908 0.788 0.616 1. 280
8° 0.140 0.139 0.990 O.Hl 53 ° 0.925 0.799 0.602 1.327
9° 0.157 0.156 0.988 0.158 54° 0.942 0.809 0 588 1.376
10 ° 0.175 0.174 0.985 0.176 55° 0.960 0.819 0 574 1.428
11 ° 0
0.192 0. 1\Jl 0.982 0.194 5() 0.977 0.829 0.559 1.483
12° 0.209 0.208 0.978 0 2J:l 57° 0.995 0.839 0.545 1.540
0 58°
i:i 0.227 0.225 0 974 0.231 1.012 0.848 0 530 I.GOO
0
14 0.244 0.242 0.970 0.249 59° 1.030 0.857 0.515 l.GG4
15° 0.262 0.259 0.966 0.268 60 ° 1.047 0.866 0.500 1.732
1()0 0.279 0.276 0.961 0 287 61° 1.065 1.804
0.875 0.485
o
n 0.297 0.292 0.956 0.306 62° 1.082 0.883 0.469 1.881
18° 0.314 0.309 0.951 0.325 0 1 100
()3 0.891 0.454 1.963
rno 0 332 0.326 0.946 0.344 64° 1 117 0 899 0.438 2.050
: �:: I
20° 0.349 0.342 0.940 0.364 65° 1 134 0.906 0 423
21° 0.367 0.358 0.934 0.384 66° 1 152 0.914 0.407

0
22° 0.384 0.375 0.927 0.404 ()7 1 169 0.921 0 391 2.356
0
23° 0.401 0.391 0.921 0.424 ()8 1 187 0.927 0.375 2.475
24° 0.419 0.407 0.914 0445 69 ° l 204 0.934 0 358 2.605
25° 0.436 0.423 0.906 0 466 70 ° 1.222 0.940 0.342 2.748
2G0 0.454 0.438 0.899 0.488 71° 1.239 0.946 0.326 2.904
I
27° 0.471 0 454 0.891 0.510 72° 1.257 0.951 0 309 3.078
28° 0.48\J 0.469 0.883 0.5:32 7:3 ° 1.274 0.956 0.292 3.271
29° 0.506 0485 0.875 0.554 740 1.292 0.961 0.276 3.487
30° 0.524 0.500 0.866 0.577 75° 1.309 0.966 0.259 3.732
31° 0.541 0.515 0.857 0.601 76° 1.326 0.970 0.242 4. 011
32° 0.559 0.530 0.848 0.625 77° 1.344 0.974 0.225 4.332
33° 0.576 0.545 0.839 0.649 78° 1. 361 0 978 0.208 4.705
34° 0.593 0.559 0 829 0.675 79 ° 1.379 0.982 0.191 5.145
35 ° 0 611 0.574 0.819 0 700 80° 1.:396 0.985 0.174 5.671
:�50 0.628 0.588 0.809 0.727 81° 1.414 0.988 0.156 6.314
37° 0 646 0.602 0.799 0.754 82° 1. 431 0 990 0.139 7.115
38° 0.663 0.616 0.788 0.781 83° 1.449 0 993 0.122 8.144
39° 0.681 0.629 0.777 0.810 84 ° 1.466 0.995 0.105 9.514
40 ° 0.698 0.643 0.766 o.s:io 85° 1.484 0.996 0.087 1143
41° 0.716 0.656 0.755 0.86!) 86° 1.501 0.998 0 070 14.30
42° 0.733 0.669 0.743 0. 900 ! 87° 1.518 0.999 0.052 19.0 �i"
ggo
I
43° 0.750 0.682 0.731 0.933 1.536 0 999 0.035 28.64
44° 0.695 0.719 0.966 0 1 553 1.000 0.017 57.29
0.768 39
45° 0.785 0. 707 0.707 i. ooo I 90° 1.571 1 t. ooo 0 000
An Exact Definition of the Idea of a Function
Table 2
Exponential Functions
x e" e-x x e" e-x
0.00 1.0000 1.0000 2.5 12.182 0.0821

0.05 1.0513 0.9512 2.6 13.464 0.0743
0.10 1.1052 0.9048 2.7 14.880 0.0672
0.15 1.1618 0.8607 2.8 16.445 0.0608
0.20 1. 2214 0.8187 2.9 18.174 0.0550
0.25 1.2840 0.7788 3.0 20.086 0.0498

0.30 1.3499 0.7408 3.1 22.198 0.0450
0.35 1.4191 0.7047 3.2 24.533 0.0408
0.40 1.4918 0.6703 3.3 27.113 0.0369
0.45 1.5683 0.6376 3.4 29.964 0.0334
0.50 1.6487 0.6065 3.5 33.115 0.0302

0.55 1.7333 0.5769 3.6 36.598 0.0273
0.60 1.8221 0.5488 3.7 40.447 0.0247
0.65 1.9155 0.5220 3.8 44.701 0.0224
0.70 2.0138 0.4966 3.9 49.402 0.0202
0.75 2.1170 0.4724 4.0 54.598 0.0183

0.80 2.2255 0.4493 4.1 60.340 0.0166
0.85 2.3396 0.4274 4.2 66.686 0.0150
0.90 2.4596 0.4066 4.3 73.700 0.0136
0.95 2.5857 0.3867 4.4 81.451 0.0123
1.0 2.7183 0.3679 4.5 90.017 0.0111

1.1 3.0042 0.3329 4.6 99.484 0.0101
1.2 3.3201 0.3012 4.7 109.95 0.0091
1. 3 3.6693 0.2725 4.8 121.51 0.0082
1.4 4.0552 0.2466 4.9 134.29 0.0074
1.5 4.4817 0.2231 5 148.41 0.0067

1.6 4.9530 0.2019 6 403.43 0.0025
1.7 5.4739 0.1827 7 1096.6 0.0009
1.8 6.0496 0.1653 8 2981.0 0.0003
1.9 6.6859 0.1496 9 8103.1 0.0001
2.0 7.3891 0.1353 10 22026 0.00005

2.1 8.1662 0.1225
2.2 9.0250 0.1108
2.3 9.9742 0.1003
2.4 11.023 0.0907
732 Appendix N
Table 3
Natural Logarithms of Numbers
I
n log,n n log,n n log,n
---
*
0.0 -
4.5 1.5041 9.0 2.1972
0.1 7.6974 4.6 1.5261 9.1 2.2083
0.2 8.3906 4 7 1 .5476 9.2 2 2192
0.3 8.7960 4.8 1 .5686 9 3 2.2300
0 4 9.0837 4.9 I 5892 9.4 2.2407
0.5 9.3069 5.0 1.6094 9.5 2.2513

0 6 9.4892 .5.1 l .6292 9.6 2.2618
0.7 9.6433 5.2 I 6487 9.7 2.2721
0.8 9.7769 5.3 I 6677 9.8 2.2824
0.9 9.8946 5.4 1 .6864 9.9 2.2925
-
JO 0 0000 5.5 I. 7047 10 2.3026

11 0.0953 5.6 1.7228 11 2.3979
12 0.1823 5.7 1.7405 12 2.4849
I
I. 3 0.2624 5.8 I.7579 13 2.5649
I. 4 0.3365 5.9 1 7i50 14 2.6391
I. 5 0.4055 6.0 1 7918 15 2.7081

1.6 0.4700 6. I l. 8083 16 2.7726
I. 7 0.5306 6.2 1.8245 17 2 8332
I. 8 0.5878 6.3 1.8405 18 2.8904
I. 9 0.6419 6.4 1.8563 19 2.9444
2.0 0.6931 6.5 I. 8718 20 2.9957

2.1 0.7419 6.6 1.8871 25 3.2189
2.2 0.7885 6.7 I 9021 30 3.4012
2.3 0.8329 6.8 1.9169 35 3.5553
2..4 0.8755 6.9 1.9315 40 3.6889
2.5 0.9163 7.0 1 .9459 45 3.8067

2.6 0.9555 7. I 1 .9601 50 3.9120
2.7 0.9933 7.2 I 9741 55 4.0073
2 8 1.0296 7.3 1.9879 60 4.0943
2.9 I. 0647 7.4 2.0015 (i5 4.1744
3 0 1 0986 7.5 2.0149 70 4.2485

3. l 1.1314 7.6 2.0281 75 4.3175
3.2 I 1632 7.7 2.0412 80 4.3820
3.3 I. 1939 7.8 2.0541 85 4.4427
3.4 1 2238 7.9 2.0669 90 4.4998
3.5 1.2528 8.0 2.0794 95 4.5539

3.6 I. 2809 8. l 2.0919 JOO 4.6052
3 7 I. 3083 8.2 2.1041
3.8 1.3350 8.3 2.1163
3.9 !. 3610 8.4 2.1282
4.0 I. 3863 8.5 2.1401

4. 1 I. 4110 8.6 2. 1518
I
4.2 1.43.51 8.7 2.1633
4.3 1.4586 8.8 2. 1748
4.4 14816 8.9 2.1861
Selected Answers
PROBLEM SET 1.3
1. x < -3 3. x < t 5. x > -3

7. x > 13 9. x < 2 11. x < -3
PROBLEM SET 1.4
1. ( -a:J, -1) 3. (-1, 1)

5. (0, 10) 7. (- a:J, -1) and [3, )
a:J
9. a) (0, ) b) [-1, a:J)

a:J 11. ( -a:J, a:J)
13. {-2, t} 15. { }
17. ( -a:J, a:J) 19. {1}
21. (- a:J, -2) and(-!-, a:J) 23. (-2, 2)
25. [t. tJ 27. [f, tJ
29. [-1, 3) 31. [t. 3)
PROBLEM SET 2.2
1. a) 2v'2 b) 6v'2 c) 2v'2 d) v2

5. (1 + v3, 1 - v3) and (1 - v'3, 1 + v3)
7. (i!, 154), 154 v'26
9. a) lyl b) lxl 11. (3, 0)
PROBLEM SET 2.3
1. x 3 = 3. y 4, x � 0
=
5. x2 + y2 + 4x - 4y + 4 = 0 7. y 2x, x � - 1
=
9. x2 - 2x + y2 - 3 0 = 21. b) y x =
23. x = t
PROBLEM SET 2.4

x 7
1. y - 1 -6(x
l
- 2), y 5 + 3. y f(x - 1), y fx - f
S
= =
= = -
5. 4y + 3x = 25 7. (1, 1) and (1, -1)
733
734 Selected Answers
PROBLEM SET 2.6
5. F= (0, -12), D:y= -/2 7. F= (0, f),D:y= t

9. F= (2,f),D:y=f 11. F= (-1, t),D:y= -t
13. F= (-t, 11s-), D:y = --16
PROBLEM SET 2.7
9. crosses x-axis at (0, 0), (2, 0), ( -2, 0); tangent horizontal where x = ± ----2 --:
v3
; slope at
(0, O) = -4; y > 0 if -2 < x < 0 or 2 < x; y < 0 if x < -2 or 0 < x < 2
PROBLEM SET 2.8
1. 14 3. 50
5. 33 7. b3 + 2c3 + b4 + 2c4 + b5 + 2c5
k
9. m7 + (m + 1)7 + + n7 11. �>2
i=3
· · ·
PROBLEM SET 2.9
5. 2n2 7. in(n + l)(n - 1)

PROBLEM SET 2.10
1 •
itXQ
3 3 •
_4,Q
3 5 •
!lQ
3
7. t 9. t 11. t
15. a) n � 4 b) n � 2 1010 c) · n > -2 +

1
1 I
2€
. - 3£)
v E(8
(n � > also works. Note that you were not asked to find the smallest possible n. )
17. a) n > 98 b) n > -
1
€
-2
PROBLEM SET 3.4
7. bounded with M = 1 9. bounded with M = 1

11. bounded with M = 8 13. bounded with M = t
15. bounded with M = t 17. unbounded
19. bounded with M= 1
Selected Answers 735
PROBLEM SET 3.5
1
1. 70x9 - 8x7 3.
+ 1)2
(x
-2y3 - 3 -1
5. 7
(y3 - 3)2 . (x - 1)2
9. 3(1 + x)2 11. a) 3ay + 2x b) 3xy + 3a 2
2(x2 - 1)
13. 15. 4x3 + 6x2 + 2x
(x2 + x + 1)2
1 -x
17. --- 1 9. ---===
2.Yx + 1 Vl - x2
-1 1
21. 23. ---:====
2.Y v(1 - x2)s
�
PROBLEM SET 3.6
2x + 3 -3x + 1
1. --;===== 3
2Y(x + l)(x + 2) . (x + l)s
x -1
5. 712(x3 + x2 - x + 7)711(3x2 + 2x - 1) 7. -:===
.Yx(x - 2)
-3x2 + 4x + 1 -1
9. ---- 11. ----:===-----:===
2Vx - l(x2 + 1)2 v1 - x Y(l + x)s
13. 1 if x > 0, -1 if x < 0 15. 3(2x3y - 3x2_y2)(x3y2 - x2y3)2
19. f(3x2 + l)(x3 + x)l/2 21. y11211
23. f(x2 + 3x + 1)312(2x + 3) 25. b) tx-213
27. f. x<PfqJ-1
q
PROBLEM SET 3.7
x2 x2
1. a) b) - - c) tx lxl d) x e) -x f) Ix!
2 2
3. a) x b) -x c) JxJ d) 1 e) -1 f) sig x
PROBLEM SET 3.8
1. t
bll 1 1 b101 1
3. a) U b) - (b11 - a11) c) ( _ alOl) d) -- (x n+l - an+l)
101 1
.
11
_
n +
x4 x2
5. a) -! b)- + - -x 7. a) vs - v2 b) 0
4 2
9. (1+ xlO)lOO 11. 2(V2 - 1)

13. i� 15. 674
PROBLEM SET 3.9
t2 t5 2 1
1. ft2+4t+4 3. -2 +2t+3 5. +t - 20
20
t3 2
7. -Vl- x2 9. 2v/ - 2v2+ 5 1i.3+t+3
-1
l3. 3(1 + t3) +
7
15.
20
, a(t) -gin the time interval
[ ]
0,
20
6
=
g g
17. Vo = -100+g 19. gL = 6 ft/sec
PROBLEM SET 3.10
I. x2, 2x 3. x4 - x, 4x3 - I
4x7
5. VI+x8, 7. v:;,
VI+ x8
I -2x -
5
I
9. II. -vx(x2+I), -2xvx -
-
I + x2' (1 +x2)2 2vx

I I I -4x3
I3. --. 2 I5. � ' (x4+I)2
x x
-- -
J
I+ x I
I7. -- ' -;::==-;:==:::::; I9. 2VI+(2x)8
I- x VI+x V(I - x)3
PROBLEM SET 4.1
I. csc 0 3. cot x 5. cosy

7. tan 0 9. tan 0 I1. sec2 O
13. tan x I5. -tan 0 I7. sec O
I9. tan O 21. -sec O 23. sin 0
25. -tan 0 27. -sec 0
PROBLEM SET 4.2
9. a) -I b) -cos 0
I I. a) ro b) -cot O 13. cos 0
27. 4 cos3 0 - 3 cos 0

PROBLEM SET 4.3
I. sec2 x 3. sec x tan x 5. -sinx -

cos x
lcos xi
7. -2 sin x cos x
-1
9. 2 cos 2x 11. 0 13.
1 +smx
.
PROBLEM SET 4.4
7. 1
PROBLEM SET 4.5
I. g(x) =sin x,f(u) =u2,f'(u) =2u,g'(x) =cos x,f'(g) =2 sinx, q/(x) 2 sin x cos x =
3. g(x) =sin x +cos x,f(u) = u2,j'(u) =2u, g'(x) =cos x - sin x,j'(g) =2(sin x +
cos x), cp' (x) =2 cos 2x
5. g(x) =2x,f(u) =tan u,f'(u) =sec2 u,g'(x) =2,f'(g) =sec2 2x, cp'(x) =2 sec2 2x
/- 1 1
7. g(x) = 1 - x2,f(u) =vu,f (u) = r ,g (x) = -2x,f (g) = . i ,
• I I I
2vu 2vl- x2
-x
cp'(x) =
v1 - x2
9. g(x) =1 +x, f(u) =u113, f'(u) =tu-213, g'(x) =1, j'(g) =t(l +x)-213, cp'(x) =
!(1 +x)-213
11. g(x) =cos x,f(u) =Jg (t2 +1) dt,f'(u )=u2 +1,g'(x) = -sinx,f'(g) =cos2 x +1,
cp'(x) = -sin x(cos2 x + 1)
13. -3x5 cos (x&)
15. g(x) =x3, f(u) =sin u, f'(u) =cos u, g'(x) =3x2,f'(g) =cos x3, cp'(x) =3x2 cos x3
17. cp'(x) =cos xVl +sin2 x
19. 0 21. 0
PROBLEM SET 4.6
2x cos x2 3. -3x2 sin x3 5. 2t sec2 (t2 +1)

( )
1.
-1 x- 1
7. (3x2 +1) cos (x3 +x) 9. sin vx 11. t sec2 --
2v:X 2
13. 1
15. a) 2 sec2 x tan x b) 2x sec x2 tan x2 17. -2 sin 2x
19. -2 sin 2x 21. cos x 23. ---

25.
-x
sin Vx2 +1
-- 27. a) tx-213 cos x113 b)
1 +cos x
!(sin x)-2!3 cos x
vx2 +1
29. cos x 31. -sin x cos cos x 33. cos x sec2 sin x
35. -sin x cos3 x 31. 0 39. 0
41. 0 43. 0
PROBLEM SET 4.7
1 1
1. ---- 3. 5. 1
Y2x - x2 2+2x+x 2
-x 2x
7. 2x 9. 11.
lxl Yl - x2 2 +2x2+x4
2 -1 1
13. --=== 15. 17.
xYx4 - 1 .Y1 - x2 x.Yx2 - 1
x 1
19. 21 . 23.
.Y1 +x2
(1 + x2)a/2 (1 _
x2)a;2
1 2x 77"
25. 27. 1. 3
Y -x - x2 .Y1 - x4
1
33. 1 - --= 35. J-1(x) = .Y 1 - x2, 0 ;2; x ;2; 1
.Y2
77"
37. 3
PROBLEM SET 4.9
1
1. log.x+ 1, - 3. (I +2x)e2"', (4+4x)e2"'
x
5. (sin x+cos x)e"', 2e"' cos x 7. 2/x, -2/x2
500 -500
9. -,- - 11. (log. 10) 10"', (log.10)2 10"'
x x2
13. 0, 0 15. (2x + x2)e"', (2+ 4x+x2)e"'
-1 -1
17. (2x2+ l)e"'2, (4x3+ 6x)e"'2 19.
1 - x ' (1 - x)2
21. e"'-1, e"'-1 23. tan x, sec2 x
25. cot x, -csc2 x 27. sec x, sec x tan x
29. -csc x, csc x cot x
PROBLEM SET 4.10
2 2x.
1. - In x 3. - -- 5. 2x exp x2
x x2+ 1
7. 2x 9. cos x exp sin x 11. (In x+ 1) exp (x In x)
13. cot x 15. (In x+ l)x"' 19. 2
21. sec3 x 23. 1 25. 2
27. 2 29. 2
PROBLEM SET 4.11
2
23. 4 cosh3 x - 3 cash x 31. v'1 + x2 33
· v'1 + 4x 2
2x
39. 43. x = In 2 45. x = In (2y)
Vx4-1
47. Cosh-1(x) =In (x + v'x2 - 1 )
PROBLEM SET 5.1
1. increasing on [ - tr/2, tr/2], decreasing 3. increasing on [-2, OJ, decreasing on

on [- tr, -tr/2] and [tr/2, tr] [O, 2]
0.,
y
y
-2 -1 1 2
5. increasing on [-2, -1] and [1, 2], 7. increasing on [- tr, -3tr/4], [-tr/4, tr/4],
decreasing on [-1, 1] and [h/4, 5tr/4], decreasing on
[-3tr/4, -tr/4] and [tr/4, 37T/4]
9. increasing on [O, 1] 11. increasing on [tr, 27T], decreasing on

[0, tr]
.. x
13. increasing on [-2, -1] and [O, 1], 15. increasing on [-1, O], decreasing on
decreasing on [-1, O] and [1, 2] [0, 1]
y y
17. increasing on[-Tr, -Tr/2J and ['"/2, '"J, 19. increasing on [In 2, 2J, decreasing on
decreasing on[-'"/2, '"/2J [O, In 2J
PROBLEM SET 5.2
1. local maxima -Tr, '"/2; local mm1ma -'"/2, '"; maximum '"/2; mm1mum -Tr/2;
inflection point O; image[-1, lJ; concave upward[-'"· OJ; concave downward [O, '"J
3. local maximum 1; local minima -2, 2; maximum 1; minima -2, 2; inflection points
-l/v'3, l/v'3; image [i-, lJ; concave upward [-2, -l/v'3] and [l/v'3, 2]; concave
downward [ -l/v'3, l/v'3]
5. local maxima -1, 2; local minima -2, l ; maxima -1, 2; minima -2, l; inflection
point O; image[-2, 2J; concave upward [O, 2J; concave downward [-2, OJ
7. local maxima -3Tr/4, '"/4, '"; local minima -Tr, -Tr/4, 3Tr/4; maxima -3Tr/4, '"/4;
minima -Tr/4, 3'"/4; inflection points -Tr/2, 0, '"/2; image [-1, lJ; concave upward
[-Tr/2, OJ and [7T/2, '"J; concave downward[ -1T, -7T/2] and [0, 7T/2]
9. local maximum 1; local minimum O; maximum 1; minimum O; inflection points none;
image[l, e - 2]; concave upward [O, 1]; concave downward {}
11. local maxima 0, 2'"; local minimum 1T; maxima 0, 21T; minimum '"; inflection points
7r/2, 37T/2; image [-1, 1]; concave upward [7r/2, 37T/2]; concave downward [O, 7r/2]
and[3'"/2, 27T]
13. local maxima -1, 1; local minima -2, 0, 2; maxima -1, 1; minima -2, 2; inflection
-v'i, + v'k; image[-8, 1]; concave upward [-v'i, v'l]; concave downward
points
[-2, -v'l] and [v'i, 2]
15. local maximum O; local minima -1, 1; maximum O; minima -1, l; inflection points
_
{fl; image [t,
{! % ,
l]; concave upward [-1, _
{ff] and [{If, l]; concave
downward [- v-'i, {Ii]

17. local maxima -7T/2, '"; local minima -Tr, 7r/2; maximum '"; minimum -1T; inflection
point O; image[-7T, 7r]; convex upward[O, 7r]; concave downward[-7T, OJ
19. local maxima 0, 2; local minimum In 2; maximum 2; minimum In 2; inflection points
none; image [2 - 2 In 2, e2 - 4]; concave upward [l, 2]; concave downward {}
21. No
PROBLEM SET 5.3

1. maxima none; minima none; local maximum 1; concave upward (- oo, 0) and (2, oo) ;
concave downward (0, 2); inflection points none;
limf(x) = 0, limf(x) = 0, lim f(x) = -oo, Iim f(x) = oo, Iim f(x) = oo,
X-+oo x--..- oo x-...O+ x-o- x--2+
lim f(x) = - oo
x--2-
3. maxima none; minima none; local maximum t; concave upward (-oo, -2) and
(3, oo); concave downward (-2, 3); inflection points none;
limf(x) = 0, lim f(x) = 0, Jim f(x) = -oo, Jim f(x) = oo,
:2:-+00 x---oo X--..-2+ X-+-2-
lim f(x) = oo, Iim f(x) = - oo

X-+3+ X-+3-
5. maxima none; minima none; local maxima none; concave upward (-oo, 0) and (0, oo) ;
concave downward { }; inflection points none;
limf(x) = 0, Jim f(x) = 0, Jim f(x) = oo, Jim f(x) = oo
X--+00 x--co a'-+0+ X--+0-
7. maxima none; minimum O; local maxima none; concave upward [-1/v'3, l/v'J];
concave downward (-oo, -l/v'3] and [l/v'3, oo); inflection points -l/v'3, l /v'3;
lim f(x) = 1, Jim f(x) = 1
X-+00 x--oo
9. maxima t; minima none; local maxima t; no local minima; concave upward (-oo, O]
and [l, oo); concave downward [O, 1]; inflection points 0, 1;
Iim f(x) = 0, lim/(x) = 0, Jim f(x) = !; Jim f(x) = !
X---+--00 X-+CO X--+-1- x.-.-1+
11. no maxima; no minima; no local maxima; no local minima; concave upward ( - oo, -1)
and [o, -fit]; concave downward ( -1, O] and [-fit, oo); inflection points 0, -fit;
lim /(x) = 1, Iim f(x) = 1, Jim f(x) = oo, Jim f(x) = - oo
X-+-co x_.co X-+-1- x--..-1 +
13. e 15. e 17. e

19. e112 21. e 23. e213
25. 0 27. 0 29. $e
PROBLEM SET 5.4
1. a 2
3. 2a2
2 2v'2
7 x =-a
· v3 •
y = --
v3
a 9 128 in.3
·
w2 k3
13. 15
8 • 1728
1 17'
19. 2 21. - 23. 2 +217'k, k an integer
{1 3
25. -
{13
4 '17'r3
35. 39. (3 +2v'2)r2
3v'3
PROBLEM SET 5.5
1. t 3. t 5. 1 7. l/v'3
9. h/d 11. v2 13. 4x3 + 4.ff' = 0
PROBLEM SET 5.6
1. v'2 3. 2
PROBLEM SET 5.8

1
3. exp Tan-1 g -- ( e"' cos2 x)
1 + g2
=
1 . 3 •
5. (sin x + x cos x)e"' sm"' 7. e"'
3x2 4x
9. 3x2 cos2 x
11. cp(u) eu214, cp'(u) (u/2)eu2/4, cp' (g) xe"'2
= = =
'
13. cp(�) eu sin u, cp' (u) (sin u + u cos u)eu sin u, cp (g)
=
= = (sin x + x cos x)e"' sin'"
15. cp(u) = e"312, cp'(u) fu112eu312, cp'(g)
= i x e"'3 =
cos x
17. cp(u) = v'� cp'(u) = - �, cp'(g)
v'1 - u
=
sin x
1
19. cos (t)/e1 21. 2t3 23. -
xe"'
1
2 5. cp (u) = _ / , cp'(u) =
'
cp (g)
-u
,
-tan t cos3 t -sin t cos2 t
= =
+ u2 (1 + u2)3/2
·
v1
6 6t5
27. cp(u) (Tan-1 u)6, cp' (u) -- (Tan-1 u)5 , cp' (g)
+ u2 1 + (tan t)5
= = = ----
33. df/dt = -t!Jfor any function/(!) such that/ 2 + t2 + 1 0 =
35. df/dt = -t3/f3,J1 (t) (1 - t4)114,j�(t) !(-4t3)(1 - t4)-3/4

= = -t3!f'f, =
/ (!)
2
= -(1 - 14)114,f;(t) !:(4t4) (1 - t4)-3/4
=
-t3/f{ =
PROBLEM SET 6.2
1. {i(l + x2)4 + C} 3. {!(2 + u2)2 + C}

5. 6
{x7/7 + !-x5t2 + x3t4 + xt + C} 7. {(x/8)(x2 + t2)4 + C}
9. {a2a(t3/2 + 5)11 + C} 11. {t(l + sin x)3 + C}
13. {-{(cos x)312 + C} 15. {te2"' - 6e"' + 12x + se-x + C}
17. {te3"' + 3e"' - 3e-x - !e-3"' + C}
19. { -W + x3)-2 + C1}, x > -1; { -l(l + x3)-2 + C },x < -1
2
21. {t In (1 + x3) + C1}, x > -1; {t In (-1 - x3) + C },x < -1
2
23. {ln2(x2) + C1},x > O; {ln2(x2) + C2},x < 0
25. H sin3 x + C} 27 . h-h- sin102 x + C} 29. { -i cos4 x + C}
31. {tan 8 + C} 33. {-cot 8 + C} 35. {sec 8 + C}
37. { -t cos 28 + C} 39. {t sin 28 + C} 41. {t sin 28 + C}
43. {t sin 2fJ + C} 45. {� - isin2fJ + c} 47. {: - a\ sin40 +c}

49. { -l cos3 fJ + C} 51. {2 sin fJ/2 + C} 53. {i(l - cos 0)3/2 + C}
55. {tet3 + C} 57. {te2"' + C} 59. {tet• + C}
61. {tan x + c} 63. { -ecost + c} 65. {21� 10

2
10"' + c }
67. { -2(2 + t)-112 + C} 69. {2t - 2t-112 + C}
.
71. {Sin-1 (t) + C}, ltl < 1 73. {t(l + r3)2/3 + C}
75. {iIn (l +t4) + C} 77. { -2v'1 - e"' + c}
79. {In I sin xi + C} on any interval where sin x ¥- 0
81. {In I sec xi + C} on any interval where cos x ¥- 0
83. { -In Iese x +cot xi + C} on any interval where csc x +cot x ¥- 0
PROBLEM SET 6.3
1. {t Sin-1 (x2) C} + 3. {! Sin-1 (2y2) + C}

5. {i Tan-1 (3x2) + C} 7. {i In (9 +x4) + C}
9. {t Tan-1 (z3) + C} 11. { (1/3 v'z) Tan-1 ( v'z z 3) + c}

13. {i�n (5 +z6) + C} 15. { -tv'I - x8 + c}
17. {l In (1 +x8) + C} 19. {In (1 + ez) + C}
21. {Tan-1 e"' + C} 23. {In le"' + e-"'I + C} on ( - oo, 0) or (0, )
oo
25. {(e"' - e-"')3 - e"' + e-x + C} 27. { (1/3) Sin-1 (x3/ v'z) + c}
29. { (1/3 v'z) Sin-1 ( v'z x3) + c} 31. {t(Sin-1 x)2 + C}
33. {xe"' + C} 35. {� c}

+
37. {x sin x + C} 39. { - x cos x + C}
41. {� In x + c} 43. {x3 ln x + c}
45. {tln4 x + C}
47. {i ln2 lx2 + 2xl + C} on any interval where x2 + 2x ¥- 0
49. {t Sec-1 (x2) + C} 51. {x Sin-1 x + C}
53. {x Sin-1 x + v'1 - x2 + c} 55. {x Cos-1 x- v'l - x2 + c}
57. {t In (1 + e2u) + C} 59. {t (e2u - In (1 + e2u)) + C}
61. {t In (1 + x2) + C} 63. {t (Sin-1 (z) +zv'l - z2) + c}
PROBLEM SET 6.4
1. {x ln2 x - 2x In x + 2x + C} 3. {a(x - l)e" + C}

v v
5. { -(x/a) cos ax + (l/a2) sin ax + C} 7. {[1/(1 + a2)]ea"'( -cos x + a sin x) + C}
9. {[l/(a2 +b2)]ea"'(a sin bx - b cos bx) + C}
11. {-x2 cos x+ 1x sin x+2 cos x+ C}

13. {x 3e" -3x2e"+6xe" -6e"+ C} 15. {(x 3/3)(ln2 x - f In x+ �)+ C}
17. {x Sin-1x+v1 -x2+c} 19. {x Tan-1 x -i In (1+x2)+ C}
21. {t(x sin xe"+e" cos x -xe" cos x)+ C}

23. {x ln 3 x -3x ln2 x+ 6x In x -6x+ C}
25. fxne" dx xne" - n fxn-le" dx
=
27. {(x/2)( sin In x -cos In x)+ C}
PROBLEM SET 6.5
1. { t sin 3 x - t sin5 x+ C} 3. {ix - l sin 2x+ C}

5. {tx -i2sin 4x+ C} 7. {x+cotx - t cot3 x+ C}
9. { -! cot4 x+t cot2 x -In Icos xi+ C}, cos x ,= 0
11. {-t cot3 x =- cotx+ C}
13. {-1/2(cscx cotx+In lcscx+cotxJ)+ C}, cotx+cscx ,= 0
15. { -! csc3 x cotx+ i cscx cotx+ i In Iesex+cotxi+ C}, cotx + cscx >= 0
17. {In I sin xi+ C} 19. {t sin 4 x+ C}
21. {i+i sin (4x)+c} 23. {InJ secx+ tan xi+ C}
25. {-cscx+ C} 27. {-cotx -i cot3 x+ C}

1 -1
29. {+ In I sec x+ tan xi - sin x+ C} 31. A = -,
n
B =
n
--
PROBLEM SET 6.6
1. {x/v1 -x2+c} 3. {ln Jx+ v1x2-11+c}

5. {Sec-1 x+ C} 7. {-1/x -Tan-1 x+ C}
9. {t(Sin-1 x+xvl -x2)+c} 1i.{vx2-1+c}
13. {t(Sin-1 x -xvl -x2) + c} 15. {x -Tan-1 x+ C}
17. { -i(l � x2)a12+ C} 19. { t(l +x2)312+ C}
21. {tx2 -t In (1+x2)+ C}
PROBLEM SET 6.7
1.
{( -1 -2v�) + C 3.
{ x. +c
}
(1+v x)2 } a2va2 .;._ x2
5. {x-2 In (ve"+l +1)+c}
7. {t(1+ \o/�)2 -6 d+ \o/�)+3 In Jl+ \o/;I+c}
9. {t(v'vx+1) 3 -4Vv1x+1+c} 11. {vz2 -1+l(vz2-1)3+c}
13. {!('711+ v'x)5-3(\o/1+ v'x)2+c} 15. {x-tx3+ tx" fx7+ tx9+ C}

-
17. {x-In(Yl+ e2"'+I)+ C} 19. {2x -In ( v'1+e4"'+ 1)+c}
21. {tx2 lnlxl -tx2+ C} -
2 3. {x Tan-1x ! In(I+x2)+ C}
25. {2 In1(1+ v'�)+2(1+ v'�)-1+c}
27. {4(tx3'4 - tx1'2+x1'4 - ln (1+x1'4))+c}
29. {x-In (1+e"')+I � e"'+ c}
- -In I 1-xi +c}
31. {1nlxl-x� - 2x2
1
33. {tx3Tan-1x - �+iln(l+x2)+c}

PROBLEM SET 6.8
1. GTan-1 (�) +c} 3. {�In 1: :�I+c}

5. {In 12 v'x2+ x-4 2(x+ f)I x 1
_
v'17 + 17 + C} 7. {sin-1 ( ;3 ) +c}

v
.i
9. {Tan-1(x+ 3)+ C} 11. {Tan-1(x- 3)+ C}

13. {1In1: +c}
=�I 15. {In = +c}
1: �I
17. {x�l -lnlx-11 +lnlx-21 +c}
19. {-x�+2lnx-- 1
x--1-2Inlx-ll+C}
21. {111n Ix-21 -27lnIx+II-6(x � 2)2+9(x 1_ 2)+c}
23. {In lxl-tIn(x2+1)+ C}
25. {1nlxl+2(x/+ 1) -tin(x2+ 1)+c}
27. sine= 1-- 2x ,case=--1 -x2
+x2 l+x2
29. {-sec e+tane+ C}
3i. {-x- + 3+c}
1-
33. {In (1+sinx)+ C}, sinx -1 716
35. {tanx+sec x+ C}
37. { �2.In Isec (e+ 77/4)+ tan (e+ 77/4)1+c}
PROBLEM SET 7.1
3. ln(v'2+1) 5. j(5312 - 1)
1
7. a) i-(e - 1/e) b) -(ea - e-a)
2a
PROBLEM SET 7.3
1. f(x) = (r/h)x, volume = trr2h/3 3. tr/2 5. tr/4

2tr ( 1 )
7. tr/2 ii. 3 1-
v2
13. (b) x = 1
PROBLEM SET 7.4
1. 40tr2 3. 32tr 5. a) tr(e - 2)

4tr ( -
v'2 - 1)
1T
7. a) 2: (e2 + 1) 9. 2 v31T 11.
T
PROBLEM SET 7.5
5. 4tr(a + k)k2
13. 10 v2 1T2
PROBLEM SET 7.6
(a b
1. (x' y) = ; ' n 5.
1T
J
ac2
(b c (3b - 4a))
7. trc(Vb2 c2 v'(a - b)2
_
+ c2) 9· (x, y)
_
+ +
2' 6 (b - a)
=
11. trCb(b - a) 13. 4tr2ab

15. sv2 trk(a + b) 11. sv2 trk2
1
19. a = }-tr + v' tr4 - 8tr2 + 120tr)
4
PROBLEM SET 7.7
1. tr/4 3. t 5. 2
7. 10,000 9. 00 11. 00
13. t 15. 1 17. 2

19. tr/8 23. not finite 25. finite
27. finite 29. not finite 31. not finite
33. not finite 35. finite
PROBLEM SET 8.1
1. x' =x - 5, y' =y - 6 3. x' =x +t, y' = y + t

5. x' = x +2, y' = y +1 7. x' = x +t, y' = y +t
PROBLEM SET 8.2
1.x2/4 +y2/3=1 3. (x - 1)2/3 +(y - 3)2/4 =1

5. 3x2 +2xy +3y2 =8 7. x2/5 + y2/9 = 1
9.foci ( ± v3, 0); focal sum 4
11. foci (-2, 1 + vs), (-2, 1 - vs); focal sum 6
13. foci (0, ±VJ/2); focal sum 2
PROBLEM SET 8.3
11. 4x2/9 - 4y2/7 =1 13. -4x2/7 +4(y - 2)2/9 =1

15. xy=t 17. x2 - y2/3 =1
19. -x2/4 +y2/5 =1 23. x +y =1, x - y =1
25. x =0, y = 0
PROBLEM SET 8.4
1. hyperbola 3.hyperbola 5. straightline

1. A' =2 ± v2, c' =2 :r: v2 9. A' =t ± v3, c' =t :r: v3
PROBLEM SET 9.1
11. x = bcot e +acos e, y= b +asin 8 13. x =2acos 8, y=0
PROBLEM SET 9.2
1. 1 3. 1 5.1 7. 1 9. -8/ (7r3 - 8)

11. 0 13.0 15.1 17. 1 19. 2
1
21. - 23. e 25. 0 27. 0
e
29.x = ae +asin y =a - acos

) )
8, 8
31.x=(a+b)coslJ+bcos
( a+b a+b
- - e,y=(a+b)sin!J+bsin - - e
(
b b
PROBLEM SET 9.3
1. 0 3. 0 5. 0 9. 1 11. 0
13. e -2 15. 0 17.e 19. 0 21. 0
k 3
23. - co 27. 1 29. e 31. 1 33. e-
PROBLEM SET 9.4
1. y = 2 3. x +y = v'2
7. y = .x3 9. (x 2 +y2)2 = 2xy
13. 2x +y2 -1 = 0 19. x - y = 1
29. (x2 +y 2)2 = a2(x2 _ y2) 31. r = 1/(1 + sin 0)
33. 3r2 -16rcos () - 16r sin () + 32 = 0
PROBLEM SET 9.5
1. 3rr/2 3. rr/4 5. t
9. 2 11. 2 13. i(e8" - 1)
PROBLEM SET 9.6
1. 7T 3. 3 5. 2rr
7. i((l + i-)3/2 - 1) 9. tv's +t1n(2 +1v's) 1t. !
PROBLEM SET 9.7
23. j = ic - id 25. a) i = te +if b) j = -!e - -tf
PROBLEM SET 9.8

-+ -+ ---+ -+ -+ -+ -+ -+ -+
1. a) OS = fOR + tOP b) OT = OP + f(OR -OP) = tOP + toR
- -+ - -+ - -+ -+ -+ ---+
3. a) OS = !OR + !OP b) OT = !OP + !{OR -!OP) = !OP + !OR
11.c)a=f
PROBLEM SET 9.9
1. maximum at x = 0, K = 2
3. maximum at x = (45)-114, K = 5312 (45)-114 6-112; minimum at x = -(45)-114,
K =
_53/2 (45)-1/4 6-1/2
1 1
5· a'
'b
PROBLEM SET 10.1
1. 0 3. 0 5. 0
1
7. 0 9. - 11. 00
e
13. - 00 15. 1 17. 2
19. 00 21. converges 23. 2
25. In 2 27. e - 1 29. 1
31. 0 33. 00 35. converges
PROBLEM SET 10.2
1. not convergent 3. not convergent
7. --
1T
9. convergent 11. convergent
1 + 1T
13. convergent 15. not convergent 17. convergent
19. not convergent 21. convergent 23. convergent
25. convergent 27. not convergent 29. not convergent
00
33. I <-1)ix4i, -1 < x �1 35. convergent

i=O
PROBLEM SET 10.3
l. alternating, absolutely convergent 3. alternating, absolutely convergent

5. alternating, absolutely convergent 7. alternating, absolutely convergent
9. not alternating, not absolutely convergent
PROBLEM SET 10.4
1
5. IRnl � -
n
1 1
7. IR..I � 9. !Rn! �
<n + 1)4 (n + 1)·9
1
11. IRnl � -
(This is the estimate which is easiest to derive. Much better estimates are
n
possible.)
1 1
13. !Rn! � - 15. IR..I �
n 2n2
PROBLEM SET 10.5
1. 0.019997
.
2 /2 +)
i�l (-1)• 2(0 49)/2 c) L (-1)' --
.
. --
i=l 51 + 2
=
i=O
PROBLEM SET 10.6
1. (-1, 1) 3. (-1, 1) 5. [-1, 1)

7. [-1, 1) 9. (-1, 1) 11. (-1, 1)
13. ( -oo, oo ) 15. ( -oo, )
oo 17. ( -oo, ex:>)
19. [-1,1] 21. [4 - 1/e, 4 + 1/e]
PROBLEM SET 10.7
oo xi+ 2 co xi+ 3
I (-1Y . - , ( -1, 1 1 I c-1)i - . - , c-1, 11
l +1 l +1
3.
i=O
i.
i =O
-
22i+ix2H1 oo x2i+l
i -l)i 2i+1
� (-l) . , (- oo, oo )
oo
7 ( 2
(2i + 1) ! ' - oo' oo) ( (2i +1)!
5.
i i�
x2i co x3i +3
9. ( 1) 11. I-.- c- oo , oo
oo
i � - 3)2i(2i)!'
(- oo, oo
\ ) i=O ,I.
, )
x2i+1
15. Ic-1)i .
co
[-1, 11
i=O c 2l + n2,
x2i+l
19. (-l)i
oo
�
i (2i + 1)(2i + l)! ' - oo, oo) (
22ix2i+3 x3 co 22ix2i
i .I (-l)i
co
21. . 1 + -, (- oo, ) 23. I c-1Y . ,, c- oo , oo )
(2I ) 2 c2l )
- oo
i=O i=O
f(x) = e"'l2
• •
25. 29. j(j- 1) (j- i +l)xi-1

· · ·
PROBLEM SET 10.8
n+ 1 (n + 1)! n n!
5
·
( ) i =(n+l- i)!i!' ( )
i- 1 (n- i +1)! ( i - 1)!
9. �( ) -:i xi+ 2 11. x2i+I

i=O l
on (-1, 1) I
i=O
( -:t)l
on (-1, 1)
i� (;) xi (-1, 1) 15. 211a (-2, 2)

13. on ;�o (;) (�J on
x2i+1 x2i +l
17. 2113
oo J.
�()
co k
19. 2k on -v'2, v'2)
� (; ) ( - v2, v2)
_ _
i 2 i(2i + 1)
on
i i 2i
2 (2i +1) (
-i xi+2
23. f(x) = e•in(x)
00
21. _2 ( ) (-1, 1)
+2
. -:-- on
i=O l l
PROBLEM SET 10.9
(-l)m
an = 0 if n is an if n= 2m + 1
(2m + i) !
I. even,
=
3. a0 = 0, a1 = 1, a2 = 0
( -l)m .
an = 0 if n is a11 = ( 1f n= 2m +1
Zm + l)
5. even,
1
7. an= f
n.
9. a0 =In 2, a1 = !, a2 = -!
. . (-l)m-1.
11. an = 0 1f n IS odd, an = if n = 1m, m > 0 (a0 = 0)
1m
PROBLEM SET 10.10
lxln+l
1. IRn(x)j �
(n + l ) !
PROBLEM SET 10.11
1. -4 3. -1 5. -1
9. 1 11. 2i
2 i
13. 3 +4i 15. -9 +sv'3; 17. 5-5
1 2i 1 3i 2 v'3 i
19. 5 -5 21. - 23. 7 - - -
10 10 7
v'3 2i
25. 7 27. -i 29. i
7
8 i
31. i 33. 5 +5 35. 1
PROBLEM SET 10.12
5. 1ei"16
PROBLEM SET 10.13
1 i 1 i
-- 1 i 1 - i
1. z= --= + ----=' ----= - - ----= + -- ' - ----= ----=
v'2 v'2 v'2 v'2 v'2 v'2 v'2 v'2

'
3. z = -2, 1 + v3 i, 1 - v3 i
5. z= -1, i, -i
1 i 1 i 1 1 1 i
7· z= l , i, -i, + - + -
v2 v'2 v2 - v2' v'2 v2' v'2 - v2
- '
PROBLEM SET 10.14
1. e• 3. cos z 5. 2ze"2
PROBLEM SET 10.15
(2i +l)x2i
5.
GO (i + l)Xi
.L
i=O
.,
I.
7. .�
£.-
i=O
'+l
( -1)'
(2 I")I•
9. lim/n(x) = 0 = U lim/n(X) 11. Iim fn(x) = 0, U Jim does not exist

n-+co n-+co
13. Jim /n(O) = 1,Iim fn(x) = oo if 0 < x � 1,U lim does not exist
n--+ oo n--+ co
PROBLEM SET 11.1
1. x + 3y + 2z = 14 3. x + y +z = t 5. 5x - Sy + z = 3
7. x - y +2z = 6 9. The figure is a sphere with equation x 2 + y2 +z2 = 4.
PROBLEM SET 11.2
1
i. � +L +_:_ __l_ _:_ + - -
v'3 v' v'3 v'3=0
3 '
-�
v'3
_ _l'_
v'3 v'3 v'3=0
_
x 2y 2z x 2y 2z
3. - + - +- - 1 = 0 ' - - - - - - + 1 = 0
3 3 3 3 3 3
2 2
5 __!__ + Y + � +� = 0 - __!__ - Y - � - � = 0
· �l � �l �1 ' �1 �1 �1 �l
7
·
(:3' :3' :3)
x 2y -x 2y 3
11. - 3 =0 -
- +- - -- +- =0
v'5 v'5 v'5 ' Vs Vs Vs
PROBLEM SET 11.3
1. a= 0,P0 = (1,1, -1) 3. a= 0, P0= (1, 1 , 1)

5. a=0,P0=(-3,-1,4) 7. a= 1 , P0=(0,0, 1)
9. a= 4,P0= (1,!, !) 11. i = Vi - V2 + V3, j = V2 - V3, k =V3
13. cx1V1 + cx2V2 + cx3V3 = 0
=> CXl (i +j) + <X2(j +k) +CX3(k) = 0
=> cx1i +(cx1 + cx2)j +(cx1 + cx2 + cx3)k = 0
=> CX1 = 0, CX1 +CX 2 = 0 so CX2 = 0,and
CXl + CX 2 + CX3 = 0 SO CX3 = 0
15. i = t(V1 +V2), j = t(V2 +Va), k =!(Vi +Vs)

x y z
xi +yj+zk =2 (V1 +V2) + 2 (V2+ V3)+2. (Vi + V3)
=!(x+z)V1 +!(x +y)V2+t(y +z)V3,

°'1V1 + °'2 Vi + °'3V3 = 0
=:> °'1(i - j +k)+ °'2(i+j - k)+ <X3(-i +j+k) = 0
=:>°'1 + °'2 - <X3 =0, -<Xi+ °'2+ °'a = 0, °'1 - °'2+ <X3 = 0
=> a:1 =0, a:2 =0, °'a =0 by elimination
33. Vi =i - j = (1, -1, 0) satisfies x+y+ 2z =0 as does V2 = 2j - k = (0, 2, -1).
Any vector of the form aV1+bV2 can be written a(i - j)+b(2j - k) =
ai+ (2b - a)j - bk and this satisfies the equationx+y +2z =0. Ifa:1V1 +a:2V2 =0,
then a:1(i - j)+ a2(2j - k) =O; therefore a:1i+ (2a:2 - a:1)j - a:2k= O; therefore
a:1 =0, and a2 =0 since i, j, and k are linearly independent. Any vector ( -y - 2z, y, z)
in E can be written as (-y - 2z)V1 - z V2, and so V1 and V2 span E.

35. 4x - 2y+ z =0
PROBLEM SET 11.4
1. °'1 =1, °'2 = -2, °'a =1 3. °'1 =4, °'2 = -5, °'a = 2, <X4=· -1
5. {E1 - E2, 2E2+ Ea} 7. {E1+E2, E2, Ea+E4, E4}
9. {2E1 - 2E2 - Ea}
PROBLEM SET 11.5
1. {�
v'3
(E1+E2+E3), -1- (-E1 - E2+2E3)}
v'6
3. { 1 (E1 E2+Ea), �
-- + (2E1 -E2 -Ea), � (E2 - Ea)}
v'3 v'6 v'2
9. Va= E1 - E2 - E4, V4 =-E2+Ea+E4
11. {� (E1 - EJ}

PROBLEM SET 11.6
2
1. [Hint: Expand (x1y2 - x2y1) .]
. 3. No. D.4 fails for an obtuse triangle.
5. D.1 d(P, Q) =max {Ix - al, /y - bl} � 0 since.both are � 0.
D.2 d(P, Q) =0 =>max {Ix - al, /y - bl} =0 =>Ix - al = 0, /y - bl = 0, =>P =Q
D.3 d(P, Q) =max {/x - a/, ly - b/} =max {ly - b/, Ix - al}= d(Q,P)
D.4 d(P, R) =max {Ix - u/, /y - vi} =max {/x - a+a - u/, /y - b +b - v/}
�max {Ix - a/ +/a - ul, /y - b/+lb - v/}
�max {Ix - al, /y - bl} +max {/a - ul, lb - vi}
= d(P, Q)+d(Q, R)
1T2
13. /(x) = x2 - 3
PROBLEM SET 12.1
2 . 7T2 (-l)i4 2
1. a0= 0, ai =0, bi= -: (-1)'+1 3 . a0 = - a·i = j2 bi = - (-l)i+l
l i
--
3 ' '
5• a·=
i Ob·=
' i E_ (-l)Hl (i37T'2)
.
1T 1 1
7. ao =-4, ai= -;z- ((-l)i - 1), bi=-: (-l)i+1
l 1T l
9.
1T 1
a0= -4 + -2, ai
( 1 ) ((-1)'. - 1), bi=-:1 (-1)•+. 1 +,1 ((-1)'.-1)
-;z-
l 1T l l1T
=
1T 1 . 3 · 1
11. a0=-4 j ((-1)'- 1) ' b·=-(-1)'
a·=- +
' i 21T i
i
PROBLEM SET 12.2
1. -! sin 3x + ! sin x 3. ! cos 3x +!cos x

5. -! cos 3x + ! cos x 7. -i cos 3x + t cos x
9. i - t cos 2x + i cos 4x
PROBLEM SET 12.3
1
7. a0= 1T ( -2 + e" + e-")
2
1
ai= . (-2 + (e" + e-")(-l)i)
7r(l + 12)
i
b·= ( +2 + (e" + e-")(-I)i+l)
' 7r(l + i2)
·[l -u ·[� �]
PROBLEM SET 13.1
-4 0
1. M-• � 3 3. M-• �
3
-2 -1
5. /(R3) ={(a, b, c) I b =3a, c = -2a}
]
Ker/={(x,y, z) I 2x + y - z = O}
7. /(R3) = {(a, b, c) I 3b = 2c}
�H
-3 -I
Ker f = {(x , y, z) I x + 2y = 0, z = o} 9. M-• � 3 -1
0 2
756 Selected Answers 755
!
PROBLEM SET 13.2
3. [� ;]
LP �]
15 15
5. [4741 5825 ·�] 7 UJ
9. L�J 11. [l 64 !]5
13. [i 231 �] 15. [� 001 �]

17. [� 00 !] 1
19. [� 130 �]
21. [21327 158 2i]32
I
23. [l J
01 00 00 10
25. [ 00 10 27. [! 10 00
� �] �l
29. no inverse 31. M-'
�
[; :1
PROBLEM SET 13.4
3.9. 56
7.13. 112-2. 17. -2-2
1. G11G22 - G12G21 G33 G44 5. 1
15. -720 11.
PROBLEM SET
I. D 11 = -1,
13.5
-D21 = 2, D31 = -1 3. z =
-
5047
5. [� :] 7.
[� _!]
[i !]
0 0
5
[! H] 11.
0
-
2 +
•. 0 1
2 0 -1
+
0 0 0 0
3
0 0 0 0
2
15. 1
0 0 0 0
5
-1 0 0 0 0
1
0 0 0 0
4
PROBLEM SET 13.6
1. independent 3. independent 5. independent

7. independent 9. independent 11. independent
13. independent 15. independent 17. dependent
19. independent 21. independent
PROBLEM SET 13.7

2"' 2"'
1. {e , e } , f(x)
3"'
= e 3. {e3"', xe
3"'
} , f(x) = (1 - 2x)e3"'
PROBLEM SET 13.8

2
1. "fl/'= { 1X1e-
"'
+ IX2Xe-
X
IX3 X e-
x
+ } 3. "fl/'= { 1X1e"' 1X2Xe"' + 1X3e-2"' + IX4Xe-2"'}
3
IX4X e-
"'
+ +
5. "fl/' = { IX1e"' 1X2 cos x + IX3 sin x} 7. "fl/' = { IX1 + 1X2e"' + IX3Xe"' + 1X4e-X + IX5Xe- "'}
+
9. H = { IX1 sin x + IX2 cos x - tx cos x} 11. H = { IX1 sin x + IX2 cos x + ix sin x}
13. H = { IX1 sin x + IX2 cos x x}
+
15. H = { 1X1 sin x + IX2 cos x + ie"' sin x - -}e"' cos x}
17. H = { 1X1 sin x IX2 cos x -}e"' sin x ie"' cos x}

+ + +
19. H = { 1X1 sin x + IX2 cos x !xe"')
+ 2
21. H = { IX1 sin x IX2 cos x ix cos x + !x sin x}
+ +
PROBLEM SET 14.1
17. x2 + z2= (1 - y)2 19. 9(x2 + z2) = 2y2
21. z2 + y 2 = 10x2 23. y2 + z2 = e2'"
25. x2 + y2 = cos2 z, 0 � z � 7T/2
PROBLEM SET 14.2
1. ellipsoid 3. 9ne-sheeted hyperboloid 5. elliptic cone
7. elliptic cone 9. hyperbolic paraboloid 11. hyperbolic paraboloid
1
13. hyperbolic paraboloid 19. 3 21. 87T
PROBLEM SET 14.3
2x 3 2y3 -4x3y3
13. fx= v'x 4 + y4 + /v = ,/xv= (x4 =/vx
1 , v'x 4 + y 4 + 1 + y 4 + 1 )312
y(y2 _
x2) x(x2 _
y2) .,...y4 + 6x2y2 x4
.fv = (x2 .fxv= =fvx
_
15. /., (x2 + y2)2 + y2)2 (x2 + y2) 3

=
4xy2 -4yx 2 8xy(x2 - y2)

17. fx= (x2 + y2)2 'fv= (x2 + y2)2 '/xv= (x2 + y 2)3
=f vx
cos x -sin x -cos x
19. fx = -- ,/y= --2 - ,/xy = --2- = fyx
y y y
_ (x2 + y2)y cos xy - 2x sin xy f' _ (x2 + y2)x cos xy - 2y sin xy

21 ·
f"' -
(x2 + y2)2
>Jv -
(x2 + y 2)2
(-cos xy - xy sin xy)(x2 + y2)2 + Sxy sin xy

/xv= (x2 + y2)3
=fyx
PROBLEM SET 14.4
1. A = 1, B = 2, E1 = Ay, E2= 0
3. A = 1, B 2, E1 = Ay2, E2 = Ay + 2Ax
=
5. A = 2, B = -2, E1 = Ax, E2 = -�y
7. A = 2, B 2, E1 = Ax, E2 = Ay
=
. 9. A = 2, B -8, E1 = Ax, E2 = 4Ay

=
11. /a(l, 1) = 2 cos cz - 2 sin cz, maximum at cz = -7T/4, minimum at 3 TT/4
PROBLEM SET 14.5
1. rp'(t)= -(5t4 + 3t2) sin (t5 + t3 )
3. rp'(t)= 2 cos 2t 5. rp'(t) = 5t4

PROBLEM SET 14.6
l. <Pt= 4t3u2 + 2t + 2u2, <Pu= 2t4u + 4ut + 4u3

3. <Pt= 3t2 cos/cosg - sin/sing, <Pu= 4u3 cos/cosg - sin/ sing
5. <Pt= 2t, <Pu= 0 7. <Pt= 0, <Pu= 0
9
. E1 t:it
= fiu fiw(l + fiv), E2= fiu tiw(l + fiv), E3 = 0, E4 = 0, E5= 0
PROBLEM SET 14.8
1. no ILMax, no ILMin· 3. no ILMax, no ILMin

5. no ILMax, ILMin at ( ! - ,
-
!) 7. no ILMax, ILMin at(-!, -!)
9
. ILMax at (0, 0), no ILMin
11. maximum value at x= (1/v2, 0) and x = (-1/v2, 0), maximum value= t
13. maximum volume

16
v3
'coordinates ( 1 2
v3' v3 ' v3
3
)
15. maximum volume=
32-V3
-
9- 'coordinates= ( 1
v3
,
2
,
v3 v3
2
)
PROBLEM SET 14.9
544 1376
1. 35 3.
21
9 4
7. 0 ·
27
PROBLEM SET 14.10
1. 27r/l 1 3. 151T/16 5. ct - v3)27r
7. 0 11. 0
PROBLEM SET 14.11
28
1. 9 3. x = 0, ji = 0
5. x = 3/27r, ji = 0 7. x = 0, ji = - t
3
9. -11/6 11. x =- y-=o 13. (0, 0)
4'
PROBLEM SET 14.12
1. 0 3. 0 5. 3/7
9
7. 0 . 0 11. 3
Index
absolute convergence, 447 cardioid,400

absolute value, of a real number, 9 ff Cartesian n-space, 526
of a complex number,481 center, of an ellipse, 362
acceleration, 119,422 of any point-symmetric set,365
of gravity, 120 centroid, 335 ff
addition formula,for the cosine,136 of nonhomogeneous bodies, 674
for the sine,138 chain rule, J 59 ff
algebraic substitutions,291 ff proof of, 162
ALO,7 for paths, 642, 646
alternating function, 585 for functions of many variables, 647
alternating series,446 chord, of a function,72
alternating series test,447 circle,23
amplitude,of a complex number,488 circular cone, 621
annihilation theorem,435 closed interval,12
antidifferentiation, 255 disk, 493
A0,4 set, 669
apocryphal anecdote,120 commutative group, 578
approximations by trigonometric polynomials, commutative law, for real numbers, 1
549 for positive series,476
arc length,303 ff comparison theorem,for positive series,441
formula for, 306 completeness, of the real number system,238 ff
Archimedes, 57 completing the square, in integration, 297 ff
area, of a parabolic sector, 57 ff complex numbers,479 ff
of a surface,327 ff construction of, 535, 713
areas in polar coordinates, 402 ff components of a vector,418
arithmetic series,49 composition, of functions, 154 ff
associativity,1 for functions of two variables, 629
asymptote,368 ff concave,215
axis, of a parabola, 38 cone, 41
conical surface, 618
Banchoff,Thomas F., 56 conjugate, of a complex number, 480
base plane, 667 conjugate hyperbolas, 370
basis,for a vector space,521 constant function, 89
betweenness theorem for integrals, 312 continuity, of a function,75 ff
binomial series,468 ff of a function of two variables, 627
binomial theorem, 468 of composite functions, 520
induction proof of,472, 484 continuous,definition of,77
bound, of a function, 86 continuous function,75
bounded above, 193 elementary theorems on, 85
bounded away from 0, 696 continuous interest, 223
bounded below,194 convergent, 431
bounded function, 86 coordinate functions,382
sequence,517 coordinate system,16 ff
bounded sequence, 433,695 Cos, Cos-', 170
set,669 cash,203
box, for a function, 77 cosine, series for, 466
759
760 Index
coth, 203 dot product, 411

Cra:mer's'rule, 593 double integrals, 660, 666
cross section, 321
csch, 203
e, definition of, 198
curvature, 425 ff
existence of, 198
cycloid, 391
as a limit as x -->- oo, 221
cylinder, 316, 614
series for, 464
cylindrical coordinates, 666
eccentricity of a conic section, 380
cylindrical shell, 316
ellipse, 360 ff
solid, 614
area of, 62
surface, 614
ellipsoid, 620
elliptic cone, 621
D, 95
epicycloid, 392
Dx, 96
epsilon-delta box, 77
damped oscillation, 604
equivalence, 5
DCP, 239
of directed segments, 415
dd'-box, 79
equivalence class, of directed segments, 416
decreasing function, 206
equivalent to, 509
sequence, 433
error in Simpson's rule, 702
Dedekind cut postulate, 239
estimates of remainders, 448 ff
definite integral, definition of, 308 ff, 312
even function, 339
De Moivre's theorem, 489 ff
permutation, 584
density function, 675, 677
exact differential, 683
derivative, of a function, 69 ff
existence, of maxima, 225, 244
of an arbitrary power of x, 202
of minima, 227, 244
of one function with respect to another, .250 ff
existence theorems, use in finding maxima and
of the integral, 109 ff
minima, 223 ff
of the integral, proof of formula for, 124 ff
exp, 194
determinant function, 582
Jaws for, 196
df/dg, 250
series for, 464
diameter, of a plane set, 668
exponential function, in the complex domain,
differences, Ax and !:if, 140
484 ff
differentiability, for functions of two variables,
exponentials and logarithms, 185 ff
638
differentiable function, 89
differentials, 148 ff factor, 55
differentiation, 89 ff falling body, 120
of complex power series, 496 field postulates, 1, 580
of series, 453, 505 finite covering theorem, 350
for functions of many variables, 644 finite dimensional vector space, 529
dimension, of a vector space, 529 first octant, 509
dimension theorem, 610 focal difference, 366
directed angles, 128 sum, 360
distance, 512 focus, of a hyperbola, 366
normal form, 512 of a parabola, 38
segments, 415 of an ellipse, 360
direction angles, 513 force, 119
cosines, 514 Fourier coefficients, 546
directional derivatives, 634, 648 type, 561
directrix, of a parabola, 38 fractional exponents, 101
disk, 493 free vectors, 415 ff
distance formula, 19 frustum, 329
from a point to a line, 38 function, 63
in an inner-product space, 534. of g, 252
in polar coordinates, 401 of x, 255
distributive law, 1 of two variables, 627
diverges to infinity, 435 function-graph, 67
divisor, 55 functional equations, 232 ff
domain, of a function, 64 differentiation of, 235
of convergence, 462 fundamental theorem of integral calculus, 255
Index 761
Galileo,120 invertible function, 165

general equation of the second degree, 372 ff derivative of, 168
geometric series, 49 iterated integral,665
gradient, 649 limits, 717
graph, of a condition, 21 ff
of a function, 66 kernel, 568
of an inequality,33 ff
group,578 law of cosines, 135
Least Upper Bound Postulate,243
half,of an interval, 241 left-handed coordinate system,129
half-open interval,13 length of a path,405 ff
half-plane, 34 formula for,407
harmonic series,439 level curves, 655
Heaviside function,76 l'Hoptal's rule,388, 389, 393 ff
homogeneous differential equation, 611 limx-oo, limx--oo, limx-a+,limx-a-,
Hooke's law,605 217 ff
hyperbola, 366 ff limit,45
hyperbolic functions, 203 ff definition of,80,81
hyperboloid of one sheet, 622 of a function of two variables,627
of two sheets, 623 of a sequence,definition,431
hypocycloid,385, 391 of integration, 105
limit point of a set,669
ILMax, 213 limits,theorems on,82 ff
ILMin, 212 line integrals, 680
Im z, 482 linear combination,413,521
image, of a function, 67 differential equation,601
implies,6, 507 equation in x, y, and z, 510
improper integrals, 344 ff function,563
increasing function, 206 transformation, 563
sequence, 433 linearly dependent, 421,521
indefinite integral, 256 independent,421, 521,596
integration,257 Lipschitzian function,82, 315, 355
independent variable, 255 LMax, 212
induction principle, 51 ff LMin,212
inequalities,4 ff In,191 ff
infinite interval,13 laws for,196
infinity,13 series for,455
inflection point,215 local maximum,212
initial side, 128 minimum,212
inner product,411 locally bounded function, 88, 511,691
space, 412, 518 locally bounded away from 0, 693
inscribed broken line,303 locus of a path, 381
inside function, 155 logarithms, 186 ff
integers modulo 2, 3 laws of, 189
modulo 4, 3 logic and set theory,687
integrability of continuous functions, 353 lower bound, 194
integrable function, 312 lower limit of integration, 105
integral,of a nonnegative function, 102 ff lower sum, 309
integral test for infinite series, 445 LUBP, 243
integrand,106
integration by parts, 273 ff Maclaurin series, 473
integration, of power series, 454,503 matrix, 564
of Fourier series, 556 maxima, 211 ff
interior local maximum,213 for functions of several variables,651
local minimum, 212 mean-value theorem, 72 ff,246
interior of a circle,22 measurable set, in a plane,527, 668, 705
interior point,of an interval,176, 207 measurable solid,318
of a set in a plane,653 "measure," of a directed angle,133
interval, 12 mesh of a net over an interval, 304
inverse, of an invertible function, 165 over a plane region,668
762 Index
metric space, 539 permutation, 583

minima, 211 ff perpendicular, 30
minor, of an element of a matrix, 590 plane path, 381
mixed partial derivatives, 717 p(N), 303
MLO, 7 point of contact, 43
M0,4 point-slope form, 30
modulus, of a complex number, 488 polar coordinates, 397 ff
moments, 335 ff distance formula in, 401
of nonhomogeneous bodies, 674 polar form, of a complex number, 487
multiplication of matrices, 573 polygonal inequality, 55
MVT, 73, 246 polynomial, 86
in x and y, 633
natural logarithm, 191 power series, 453
neighborhood of a point, on a line, 152 powers of functioi1s, 99
in a plane, 653 prime number, 55
Nested Interval Postulate, 240 projection into a sub space, 541
net, over an interval, 303 Pythagorean theorem, 19, 22
over a plane region, 668
quadrant, 20
Newton's second law, 120
NIP, 240 quadratic function, 179
no-jump theorem, 191, 249 quadric surface, 620
for derivatives, 314
nonoverlapping solids, 318 radius of convergence, 494
nonsingular matrix, 593 range, of a function, 64
norm, 520 ratio test, 457 ff
normal component, 423 ray, 25, 128
normal set of vectors, 531 real-analytic, 468
normal vector space, 538 rectangular equation of a locus, 382
Northeast theorem, 393 rectangular hyperbola, 370
proof of, 527, 707 rectifiable, 304
reduction formula, 277,284
reflecting property, of a parabola, 48
octant, 509 regular path, 392
odd function, 339 Re z, 482
odd permutation, 584 right cylinder, 614
open disk, 493 right-handed coordinate system, 129
interval, 12 ring, 579
sentence, 5 RO, 7
order, 3 Rolle's theorem, 246
orthogonal vectors, 531 roots of functions, 97
orthonormal basis, 531 rotation of axes, 374 ff
outside function, 155 ruler postulate, 17
saddle point, 655
Pappus' theorem, for surfaces, 342 sample, of a net, 310
for volumes, 340 of a net over a plane region, 669
parabola, 38 ff sample sum, 310
paraboloid of revolution, 41 scalar, 411
parallel, 30 Schwarz inequality, 521, 536
parallelepiped, 316 Sec, Sec-1, 173
parallelogram law, 412 secant line, 45
parameter, 382 sech, 203
parametric mean-value theorem, 387 second derivative, 119
parametric slope formula, 386 segment, 25
partial derivatives, 631 signum, 108
partial fractions, 298 ff Simpson's rule, 176 ff
partial sum, 431 error in, 524
path, 381 Sin, Sin-1, 169
path length, formulas for, 407, 408 sine, series for, 466
proof of formula for, 531, 711 sinh, 203
patfl of motion, 41 slice function, 630, 634
Index 763
slope, of a segment, 28 term-wise integration of series, 453

of a line, 29 transitivity, 4
slope-intercept form, 29 translation, of axes, 356 ff
S(N), 309 of a coordinate plane, 415
s(N), 309 transpose, of a matrix, 587
solid torus, 327 transposition, 583
solution set, 12 triangular inequality, 11
span, 414, 521 trichotomy, 4
sphere, 620 trigonometric functions, of angles, 129
square root, 10 of numbers, 131
squeeze principle, for functions, 145, 222
for sequences, 435, 518, 696 U Jim, 502
for volumes of solids, 318 unbounded, 193-194
standard position, for an angle, 133 uniform convergence, 502
for an ellipse, 362 norm, 538
for a parabola, 40 uniformly accelerated motion, 119
subadditive function, 537 uniformly continuous, 315, 353
subspace, 522 uniqueness theorem, 113
substitution, integration by, 284 ff, 291 ff
unit element, 579
justification of, 290 unit tangent vector, 423
summation notation, 49 upper bound, of a function, 193
sup, 243 of a set, 243
supremum, 243
upper limit of integration, 105
surface of revolution, 327, 616 upper sum, 309
symmetric, 338
Tan, Tan-1, 171-172 value, of a function, 67

Tan-1, series for, 455 vector space, 411
tangent, 43 ff vectors, in a plane, 409 ff
operations on, 410 ff
to a parabola, 45
velocity, 119, 422
tangent vector, 423
tangential component, 423 vertex, of a parabola, 39
tanh, 203
Taylor series, 473 ff weight, 120
Taylor's theorem, 477 ff well-ordering principle, 54
terminal side, 128 winding function, 130
A BCDEFGH'/987 65432

Edwin E Moise Calculus PDF

Uploaded by

Copyright:

Available Formats

Edwin E Moise Calculus PDF

Uploaded by

Document Information

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

Edwin E Moise Calculus PDF

Uploaded by

Copyright:

Available Formats

CALCULUS

ADDISON-WESLEY PUBLISHING COMPANY

Consulting Editor: LYNN H. LOOMIS

Copyright© 197.2 by Addison-Wesley Publishing Company, Inc. Philippines copyright 1972

3) Some are a combination of types 1 and 2, with exact theoretical material

New York City, N. Y. E.E.M.

1. THE SPIRAL PROCESS

significance, because it is used to solve a nontrivial problem, namely, the problem

New York City, N. Y. E. E. M.

Chapter 2 Analytic Geometry

Chapter 3 Functions, Derivatives, and Integrals

3.1 The idea of a function . 63

Chapter 4 Trigonometric and Exponential Functions

4.1 Directed angles. Trigonometric functions of angles and numbers . 128

4.4 The approximation of differences by differentials 148

Chapter 5 The Variation of Continuous Functions

5.1 Intervals on which a function increases, or decreases 206

Chapter 6 The Technique of Integration

6.1 Introduction 254

Chapter 7 The Definite Integral

7.1 The problem of arc length 303

Chapter 8 The Conic Sections

8.1 Translation of axes 356

9.1 Motion of a particle in a plane 381

Chapter 10 Infinite Series

10.l Limits of sequences 431

Chapter 11 Vector Spaces and Inner Products

11.1 Cartesian coordinate systems in three-dimensional space 508

Chapter 12 Fourier Series

12.1 Projections into a subspace, trigonometric polynomials and Fourier

Chapter 13 Linear Transformations, Matrices, and Determinants

13.1 Linear transformations . 563

Chapter 14 Functions of Several Variables

14.1 Surfaces and solids in R3 614

Appendix A The Shorthand of Logic and Set Theory . . 687

Closure. For every a and b in R, a+b and ab are in R.

Associativity. For every a and b,

a+b=b+a and ab=ba.

1.2 PRODUCTS WHICH ARE EQUAL TO ZERO

Theorem 1. If ab = 0, then e ith er a = 0 or b = 0.

PROBLEM SET 1.2

1. Show that if x2 =0, then x=0.

How do you know that no other number is a root of the equation?

3. If 0 had a reciprocal, then its reciprocal would be a root of the equation

Show that this equation has no root.

5. a) Show that if ab c = 0, then a = 0 or b = 0 or c = 0.

6. a) If a2 = b2, does it follow that a = b? Why or why not?

8. a) Under what conditions (if any) is it true that

b) Under what conditions (if any) is it true that

We think of the real numbers as being arranged on a line, like this:

where<is a relation having the following properties:

0.2. (Transitil'ity) If a<b and b< c, then a<c.

a is positive if a > O; a is negative if a<0. Zero is neither positive nor negative.

MO. If a > 0 and b > 0, then ab > 0.

AO. If a<b, then a +c <b+c for every c.

Theorem 1. lf a > 0, then -a<0.

Theorem 2. If a<0, then -a > 0.

Theorem 3. If a<b, and c<d, then

That is, if a<band c > 0, then ac<be.

That is, if ac<be and c > 0, then a< b.