Professional Documents
Culture Documents
SECOND EDITION
EDWIN E. MOISE
Harvard University
London
Consulting Editor:
LYNN H. LOOMIS
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
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
bridges the gap between the modern concepts of function and derivative and the
"fractional" notation
du/dv
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"
The above remarks about the geometry of the ellipse also apply here.
9. The chapter on infinite series has been completely recast.
the idea of uniform convergence was built into the presentation, almost from the
outset;
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
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
10.
1 + x2)
subspaces has been omitted (from the text and therefore from the problems.)
11.
partial derivatives has been introduced, in parallel with the subscript notation
f.,,fv, . .
..
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
definition
of the definite
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.
taught, then the course is logically complete, and nothing that it includes needs to be
taught again later.
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
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.
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.
hard fact is that "practical" calculus courses are not practical.
The
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.
principle applies very widely.
The same
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.
E.E.M.
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.
the title
Elements of Calculus.)
Elements
are in the style of treatment; and the ideas underlying them may best be
1.
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.
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;
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.
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.
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.
4.9, is not given merely in order to make the student's life easier. The transition from
In
x =
n (dt/t)) is
on the definition
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;
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.
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 =
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).
1.
X-+0
then the technical difficulties are formidable, and workable problem material is hard
to come by.
linear
In Section 2.7 we
function.
to the limit, in this case, we merely plug the hole in a punctured line.
has no intrinsic significance.
To pass
This process
viii
Preface
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.
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.
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;
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.
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
3)
4)
5)
ix
Puzzle problems.
Sketching exercises, in which the student is asked to translate back and forth
between analytic ideas and visual images.
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;
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
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.
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.
For
and
thereafter the treatment of the two is closely tied together. The Schwarz inequality is
first introduced (on page
536, it is proved in the general case, and thereafter it is used in a great variety
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.
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
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.
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.
October 1971
E. E. M.
Contents
Chapter 1
Inequalities
1.1
Introduction
1.2
2
3
1.3
Order
1.4
Chapter 2
Analytic Geometry
16
2.1
Introduction
2.2
2.3
21
2.4
26
2.5
33
2.6
Parabolas .
38
2.7
Tangents
43
2.8
2.9
49
Chapter 3
16
51
57
3.1
63
3.2
69
3.3
75
3.4
Theorems on limits
82
3.5
89
3.6
97
3.7
3.8
109
3.9
119
Chapter 4
102
124
4.1
128
4.2
135
4.3
139
xi
xii
Contents
4.4
148
4.5
Composition of functions
154
159
4.6
4. 7
165
4.8
176
4.9
185
191
197
5.1
5.2
211
5.3
216
5.4
206
the use of
223
5.5
232
5.6
238
5.7
246
5.8
250
Chapter 6
6.1
Introduction
6.2
6.3
265
6.4
Integration by parts .
273
6.5
278
6.6
Integration by substitution .
284
6.7
Algebraic substitutions
291
6.8
297
Chapter 7
254
255
Algebraic
7.1
7.2
308
7.3
315
7.4
321
7.5
327
7.6
335
7.7
Improper integrals
344
350
*7.8
Chapter 8
303
8.1
Translation of axes
356
8.2
The ellipse
360
8.3
The hyperbola
366
8.4
372
Chapter 9
9.1
9.2
385
9.3
393
9.4
Polar coordinates
397
9.5
402
381
9.6
405
9.7
Vectors in a plane
409
9.8
Free vectors
9.9
415
422
430
Limits of sequences
10.2
437
10.3
445
10.4
Estimates of remainders.
448
10.5
453
10.6
457
10. 7
10.8
10.9
Taylor series
431
463
468
473
477
479
function
484
489
Differentiation of complex power series
493
499
508
11.2
512
11.3
518
11.4
526
11.5
Orthonormal bases
530
11.6
533
12.2
12.3
series
541
549
.
556
xiv
Contents
13.1
13.2
13.3
13.4
13.5
13.6
13.7
13.8
Linear transformations .
Composition of linear transformations and multiplication of matrices
Formal properties of the algebra of matrices. Groups and rings
The determinant function
Expansions by minors. Cramer's rule and inversion of matrices
Row and column operations. Linear independence of sets of functions
Linear differential equations
The dimension theorem for the space of solutions.
563
570
577
582
590
596
601
The nonhomo
607
geneous case .
14.1
14.2
14.3
14.4
14.5
14.6
14.7
14.8
Line integrals
651
660
666
674
680
687
690
695
697
700
702
705
707
711
713
717
721
725
727
733
759
Level curves
14.9
14.10
14.11
14.12
Appendix A
Appendix
Appendix
Appendix
Appendix
B
C
D
Appendix F
Appendix G
Appendix H
Appendix I
Appendix J
Appendix K
Appendix L
Appendix M
Appendix N
614
620
626
634
641
644
648
1.1
Inequalities
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.
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,+, ].
For every
Associativity.
and
For every
in R,
and
a+b
and
ab
are in R.
b,
a+(b+c)= (a+b)+c,
and
a(bc)= (ab)c.
Commutativity.
For every
and
b,
and
a+b=b+a
Distributive Law.
For every
a , b,
and
ab=ba.
c,
a(b+c)=ab+ac.
Existence of 0 and 1.
a+O=a
for every
and
a 1 =a
a.
Existence of Negatives.
Existence of Reciprocals.
For every
there is a number
a I /a=1.
1.2
Inequalities
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
contrary is stated.
We shall assume not only the field postulates but also the familiar laws based on
them.
1.2
(a - b)(a + b)
for every
a2
a 0
a.
.When we perform calculations, we shall not stop to justify them on the basis of the
field postulates.
If ab
0, then e ith er
0 or b
0.
Proof
1) If a
0,
2) If a 0,
=
-1 (ab)
a
1
=
1. b
0'
0,
and
b
Thus either
0 or b
0.
0.
(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
Ox=l.
Show that this equation has no root.
4. a) If
b) If
ab =ac,
ab =ac,
c?
c?
1.3
Order
5. a) Show that if
ab c
0, then
0 or b
0 or
0.
0.
b2, what can you conclude about the relation between a and b? Why?
1
=--
x +a
(a + b)2
a2 + b2?
(a + b)3
a3 + b3?
*9. Consider the "number system" which has only two elements 0 and 1, with addition and
----1
--0
(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
0
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
-V3
11'
-1
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,
+,
<],
0.1. (Trichotomy) For every a and b in R, one and only one of the following
conditions holds:
or
a<b,
or
a= b,
b<a.
We
c.
These four laws, in combination, tell the whole story: all of the elementary laws
of inequalities can be derived from them.
following problem set.
Theorem 1.
Theorem
2.
Theorem 3.
lf a
If a<0, then -a
> 0.
Theorem 4.
number.
That is, if a<band
Similarly,
Theorem 5.
number.
That is, if ac<be and c > 0, then a< b.
Theorem
number.
That is, if a<b and c<0, then ac > be.
1.3
Order
Theorem 7.
number.
That is, if be<ac, and c<0, then b
>
a.
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,
32+4<52+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)
4<2x + 7,
(2)
then
by AO. (We have added -3x to each side of the inequality.) Therefore
-3<2x,
(3)
x > -t,
(4)
by AO; and so
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)
-3<2x,
(3)
4 <2x + 7,
(2)
3x + 4<5x + 7,
(1)
then
by Theorem 4; therefore
by AO; and so
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.
two inequalities (or any two open sentences of any kind) we mean that whenever one
of them is satisfied, so is the other.
Inequalities
1.3
> 0
x2 > 0.
=>
b => a2
> 0
<=>
> 0 (?)
Similarly,
b2 is true, but
( ?)
is false, because if a ":/= 0 and b
<=>
a2
b2 (?)
does not.
The shorthand symbols<=> and => are worth learning and using.
The reason is
=>
(5)
a + e< b + e.
And given a + e< b + e, we can add -e to both sides, preserving the inequality.
Therefore
a
=>
+ e< b + e
a< b.
(6)
<=>
a + e< b + e.
We shall refer to this, for short, as ALO. Similarly, Theorem 4 says that
fore > 0,
a< b
fore > 0,
ae< be
ae< be.
=>
=>
Fore> 0, a<b
<=>
ae<be.
(7)
=>
b > a.
(8)
=>
a< b.
(8')
fore< 0,
a< b
fore< 0,
be<ae
ae > be
=>
and
And (8) can be rewritten in the form
fore< 0,
<=>
ae > be.
Order
Trich.
a<b,
or
b,
or
a<band b < e
a< e.
MO.
a>
ab>
AO.
a< b
Trans.
0 and
b>
1.
a>
-a< 0.
[Theorem
2.
a<O
-a>
[Theorem
3.
a<band e< d
a<b
0.
a+ e <b + e.
Theorem
ALO.
b <a.
<=>
0.
a+ e <b
d.
a+ e<b+ e.
MLO.
Fore> 0,
a<b
<=>
ae<be.
RO.
Fore< 0,
a<b
<=>
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
<=>
by ALO
<=>
-3 <2x
by ALO
<=>
-t < x
byMLO
x > -
<=>
by definition of >.
a.
5 - 3x
>
17+x
5x+ 3
>
17x+ 1
-3x - 7 <x +5
6x - 10
>
5x+ 3
2. 5x - 3<17x+ 1
4. 5+ 3x<17+x
6.
-4x -
8.
3 - 2x<4 - 3x
8 <2x+6
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.
Inequalities
1.3
you may assume that the results given in the preceding problems are known and you may
cite them as reasons.
=>
0 <a
-a< 0
=>
b) 0 <a
=>
d) a> 0
=>
- a + 0 < -a+ a
-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.
=>
=>
a+ c <b + c.
c <d
=>
b + c < b + d.
b) Similarly, for
c) Prove Theorem 3.
<=>
b - a> 0.
c> 0
b - a> 0
=>
(b - a)c > 0.
c) Prove Theorem 4.
f;; 0,
for every
x.
y2 - 2y + 1 f;; 0,
0, or
< 0.
for every y.
b) Show that
a> 0
=>
1
-
> 0.
[Hint: By Trich., it
0 and 1/a< 0 are impossible.
1/a
1.]
and
ac < be
=>
a<b.
1.4
21.
Absolute Values.
Give the reason for each step in the following proof of Theorem 6.
=>
=>
=>
=>
=>
22.
and
a <b
b - a >0
and
b - a >0
and
(b - a)(-e) >0
ae - be> 0
ae> be.
<
e <0
-e> 0
Give the reason for each step in the following proof of Theorem 7.
=>
be
be
<
ae
ae
and
=>
be
<
ae
and
-> 0
-e
=>
ae - be >0
and
- >0
-e
=>
=>
=>
b >a.
<
e <0
-e >0
and
23.
Is there a positive number which is smaller than all other positive numbers? Why or
24.
Is there a negative number which is larger than all other negative numbers?
why not?
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?)
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.
1.4
The absolute value lxl of a number xis defined by the following two conditions:
1)
If x 0, then lxl = x.
2)
-x.
(1)
we have
12 1=2,
and under Condition
(2)
we have
1-21 = -(-2) = 2,
Inequalities
10
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.
1.
x 0. Here lxl
I.
\x\2
x2.
x2
\xi=
4.
-J x2,
by definition of -J .
For every x,
1-x\
5.
-J.
-J (-x)2
I x!.
-J x2
Ix!.
-J (xy)2
jx\ \yj.
-J x2y2
-J x2 -J y2
Jxl IYI.
1,
1.4
Absolute Values.
Theorem 6.
For every
11
x,
xIx!.
0.
Here
0.
(Why?)
Thus
x <
0 <
-x =
The trouble with this theorem, if we try to prove it by brute force, is that there
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 ,
we have
yjy l,
x+ Ylxl+ IYI,
and so
Case 2.
and
x+ y <
0. Then
-x - y >
0. Therefore
Jxl < d
<::::>-
-d <
x < d.
lxl <d
line"; and the points that lie within a distance d of the origin are the numbers between
-d and d.
origin.
a instead of the
12
1.4
Inequalities
Theorem 9.
Ix - al <d
<=>
a- d< x<a+ d.
lx-aj<d
Proof In Theorem
8, substitute
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)
{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
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
Thus
[a, b]
{x I a x b}.
Absolute Values.
1.4
13
(a, oo)
{x I a
<
x}.
Similarly,
( - oo, a)
{x I x
<
a},
(a, oo)
a
(- oo, a)
0
This notation, in which
but it is convenient.
"numbers"
- oo
and
"oo"
[a, b)
{x I a
<
b}
( a, b]
and
{x I a
<
{x I x
a}.
b},
[a, b)
0
a
(a, b]
[a, oo)
{x I x a},
and
( - oo, a]
[a, oo)
a
(, oo, a]
a
14
1.4
Inequalities
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),
( - oo, a),
[a, oo),
(-oo, oo).
(- oo, a],
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
=>
=>
=>
=>
=>
IYl 2
Jxl2
x2 = y2
x2 - y2 =
=
(x - y)(x + y)
or
x y
x
=
0
=
-y.
a form
{xi 3x+4
<
2. {x I 3 - x > x - 3}
4. {x I Ix - 31 2}
6. {x I 2x+3 6x - 4}
8. {x I 12+xi < O}
9. a) Is it true that
Yx2
oo .
1.
3.
5.
7.
5x+7} = (-f,
x for every x?
.
{x I Yx2 = x},
{xi Y(x+1)2=x+1},
in the interval notation.
Find out for what numbers
case in which the solution set is an interval, the answer should be given in the interval
notation.
10.
12.
14.
16.
Yx2- 2x+1 = x - 1
lx2 - 5x+61=x2 - 5x+6
Ix+11 =11 - xi
Yx2-l=x
11.
13.
15 .
17.
Absolute Values.
1.4
19. ,12x - x2 = 1
21. Ix + ll +l2x +31
18. 1 7x +31 +1 3 - xi 6 Ix + ll
20. l2x - x21
22. Ix
x +2x2
>
15
Indicate graphically, on a number scale, the places where the following conditions hold;
describe the graphs in the interval notation if possible.
23. lxl
33.
24. Ix - 21 <1
26. Ix - 11 < -
28. Ix 21 < :l
30. l2x - 41 < 1
32. 1 2x - 1 1 1
<2
lxl -
and
a) Show that if
0, then
lI 11
=
b) Show that if
34.
l,bl
a
and
and
b,
b,
x 2
Ix - 21 ;;;; 1
l2x - 11 ;;;; 1
and
such that
lbl
y = I. Therefore
0, then
and
and (also)
lal
fbl.
la - bl lal - lbl.
la +bl lal - lbl.
35.
a/lal
defined?
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.
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
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.
(We say merely that PQ is nonnegative, rather than PQ > 0, because we are allowing
the case P
0.)
the points of a line. We choose a point 0 as the origin; it is given the label 0.
-OP2, which is
Thus we have a matching scheme, under which each point of the line is
-2
RS
I I I
1 V2 2
-1
For the points marked in the figure, the matching pairs are
P-2,
S
../2,
Q -1,
T 2,
Rl,
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
Coordinate Systems.
2.2
17
between the elements of one set and the elements of another, is called a one-to-one
Here P1 +--* x1 and P2 +--* x2 This distance formula holds no matter how the points
P1 and P2 are situated on the line:
0
lx2 - x11.
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 of P.
Every line has a coordinate system. And given any two points
0 and
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.
y
y N--------- -,p
Ml
I
L--------- y
p
N
I
I
M x
x
If M +--* x,
Analytic Geometry
18
2.2
+--+
y,
p +--+
(x, y)
(x, y)
order in which we write the numbers makes a difference. In the left-hand figure below,
and
(x, y),"
which is matched with (1, 2)" or "the point which is matched with
may write P
(x, y),
meaning P+--+
(x, y)."
Thus we
(x, y).
--
p
-,
I
N
y
I
I
Q
------,
I
I
I
I
I
I
p'
-----
tt
Obviously
and
and
I
I
I
I
I
are known, because the vertical line through M and the horizontal line
+--+
(x,y)
between the points of the plane and the ordered pairs of real numbers. Such a corre
spondence is called a
(x2, h)
(x1, y1)
and
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.
Coordinate Systems.
2.2
Theorem
19
1. If
and
then
PQ=
Proof
.J(x2 - X1)2
at the point
R.
Let
and
T be
+ (Y2 - Y1)2.
P,
meeting
and
respectively. Then
PR= ST,
because opposite sides of a rectangle have the same length. And
PR = lx2
Xii
theorem,
Therefore
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
2
2
IY2 - Y1l = (Y2 - Y1)
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.
minimum that must be indicated when graph paper is used for drawing pictures of
20
Analytic Geometry
2.2
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
y
1
II
III
IV
--+---X
-'--
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
c) (7, -5)
2.
a) (3, 7)
(3, 4)
and
3. Obviously, PQ
b) ( -2, -4)
and
(4, 2)
(5, -7)
d) (1, 0)
and
(0, 1 )
(-3, -7)
b ) ( 1, 3 )
and
(--2, 7)
and
does not depend on the order in which the points are named.
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.
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.
11.
Find a point on the x-axis which is collinear with the points (1, 2) and (0, 3).
remarks following Problem 6 also apply here.)
(The
2.3
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.
A
.,
l><i_.
That is, show that in the left-hand figure we have LACD > LA.
based on the figure on the right.
The proof is
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.)
..
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
Y2
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
22
Analytic Geometry
2.3
when the solution set is a set of points.) In this language, we say that the graph of the
condition OQ
r is
r.
y
r
-r
The
interior of the circle with center P and radius r (r > 0) is the set of all points
< r. Thus the interior is the graph of the inequality PQ < r. We
Q such that PQ
Sometimes the condition takes the form of an algebraic equation. For example,
if Q +--+
(x, y),
OQ
Jx2
+ y2.
OQ
(1)
Jx2
or
y2
x2 + y2
The point
r,
(2)
r2.
(3)
Thus the circle with center -at the origin and radius 2 is the graph of
J x2
y2
<=>
x2
y2
4;
J x2
y2 <
<=>
x2
y2 <
4.
x > 0
and
y > 0.
2.3
23
y
x>O
y>O
x>O
y<
O
IV
We found that the circle with center at the origin and radius
equation
2
x
y2
2
r .
(a, b)
and radius
r.
By
{PI QP
r .
.,.,....-
r,,....
,,
P(x, y)
Q(a, b)
QP algebraically, we get
QP
)(
. <=>
<=>
(x
- a)2
+ (y
- b)2
+ (y
- b)2 = r2
a)2
Thus:
Theorem 1.
(x - a)2
(a, b)
+
and radius
(y - b)2
r2
(x
2)2
(y - 5)2
4,
+ y2 + 4x - lOy +
25
0.
24
Analytic Geometry
2.3
5
4
3
2
To find out what the graph is, we first "unsimplify" by completing the square:
0,
x2+ y2
But
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
is never satisfied, for any
x and y.
{ }.
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:
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=
positive number
radius
r.
r,
r2 for some
(-D/2, - E/2) and
2.3
D/2
E/2)
25
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:
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
Thus (1) and (2) are entire
(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).
(1)
(4)
26
Analytic Geometry
2.4
Sketch the graphs of the following conditions, using cross-hatching to indicate regions.
11. x2
y2 = 1
14. x = 2
18. x2
and
12. x2
0 y 2
y2
<
15. x = - 3
y2 1
13. x2
1
16. y
19. x y 0
y2
>
17. y = x/lxl, x
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
an ellipse.]
not
(y - 1)2
v (x - 2)2
y2 = 1.
y3x - xy = 0.
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
Ax+ By+ C
where A and B are not both
0,
Every line is the perpendicular bisector of some segment. If Lis the perpendicular
bisector of the segment from
{P j PQ
PR}
then
Equations of Lines.
2.4
27
<::?-
2
2
2
2
.J(x - a 1) + (y - b1)
.J(x - a2) + (y - b2)
2
x2 - 2a1x + ai + y2 - 2b1y + b
x - 2a2x + a: + y2 - 2b2y + b
<::?-
0.
Ax + By + C
0,
with
and
Thus we have
and y.
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,
Analytic Geometry
28
2.4
+ k,
mx
where
A
m =
b2
bi
y-intercept
of the line.
If Pi
slope
of
The denominator x2
Xi
difference
in y.
difference
y
P1
P1
P,
t:.x
" <
y
" <
O
t:.x>O
P,
<
Slope
Y2-Y1
x2-x1
t:.y
t:.x
We shall show that all segments of the same line have the same slope, and that
this slope is the number
mx + b.
Given two points Pi (x1,y1) and P2 > (x2, Yz), on the line
y
then
Yz
x2
+ k
+ k,
mx
and
Therefore
and
Y2
Y1
X2 - X1
m.
2.4
Equations of Lines.
29
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=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
<=>
<=>
<=>
<=>
and so on.
3x - 3y=0
2
- y) =0
(x
(x - y)177=0,
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
m.
Therefore
X1 .
Analytic Geometry
30
2.4
Thus:
The graph of the equation y
Theorem 3.
slope
intercept form y
2x
y1
m(x - x1)
(x1, Yi).
3
-2(x + l)is the line which has
( -1, 3). Solving for y, we get the slope
-
+ 1.
Two nonvertical lines are parallel if and only if they have the same slope.
Theorem 4.
Given:
we need to prove two things: (1)If the slopes are the same, and the lines are different,
then the lines are parallel.
1)
If m1
m2,
then
k1
(2) Ifthe slopes are different, then the lines are not parallel.
--
k2,
parallel, because the two equations are inconsistent: they take the form
Since
2)
k1
--
k2,
If m1 --
m2,
If two nonvertical lines are perpendicular, then their slopes are negative
Equations of Lines.
2.4
31
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
(a2, b2).
Obviously
m2
This gives
-1/m1.
Theorem 6. Given two lines L1, L2, with slopes m1, m2. If
then
Li and L2
are perpendicular.
y
\
\
\
\
\
\
\
\
\
\
\
\
Lz
L?.
Proof
(Why?)
perpendicular to
L1.
Then
has slope
m2
Let
l/m1.
(Why?)
m2 cannot be
m1.
Therefore L is
L2,
and
L2
is
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.
Analytic Geometry
32
2.4
Find point-slope equations, and slope-intercept equations, for the Jines containing
the following pairs of points.
4.
25,
(3, 4).
Given thatP1
with
a2,
LetP2 be the point where the tangent atP1 crosses the x-axis.
-a
7.
9.
2x
2y
0.
x2
10.
2xy
+ 4y2 + I
4xy +
2x
4y
0.
a) y
lxl
b) y
-l2xl
c) y
- Ix - II
For this problem we offer a hint which applies equally well to a very large number of other
problems.
sketching the graph. This suggests that you should recall the definition of
2.5
11. Sketch the graph of
33
!xi + lyl = 1.
[Hint: As a first step, sketch the portion of the graph that lies in the first quadrant.]
12. Sketch the graph of the equation
y = x +!xi + 1.
13. Sketch the graph of the equation
!xi - lyl = 1.
14. Sketch the graph of the equation
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
1)2 +
(y -
1)2
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.
{(x,y) I (x -
1)2 +
(y -
1)2 > l }.
34
Analytic Geometry
2.5
y
halfplane.
Evidently H1
>
x.
The points lying below L form a half-plane H2; and H2 is the graph of the inequality
y< 1
x.
t< x < %.
The graph is an infinite vertical strip R1, lying between the lines
X
.l..
and
Ji
t<y< 1
is an infinite horizontal strip, as shown on the left below.
y
I
.I
I
I
y
1
___
--
__
----
--
I
I
I
I
I
t<y<l.
--- -
35
2.5
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
of the condition
y> x
and
y > -x
y=x
y >x
or
y > -x
y
S1
R
I-
R
I
I
'
I
I
l
2
S2
'
2
R
36
2.5
Analytic Geometry
Let us now see what sort of graph we get when we combine two inequalities by
"if ... then." Consider the condition
i<x<i
This says that
t<x< },
if
i<y<l.
=>
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
1)
If
(x, y)
is a point of R, and
t < x< !,
t and x
t <y< I.
There
not
between
t and i,
imposes no restriction on y at all. Therefore R contains all points to the left of the
line
t and
then
A person
with normal vision automatically obeys this law; its restrictive clause does not apply
to him. In the same way, the "law"
=>
(x, y)
t<y<l
for which
l <x< -;
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)
2)
t <x< !
x t,
and
t<y<
1,
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
region to the right of the line x
t;
i.
y
Yes
x<O
y>O
x>O
y>O
Yes
Yes
x<O
y<O
x>O
y<O
No
(3)
is the infinite
2.5
37
x>O
yO
=>-
x> 0
quadrant that
y < 0 s
holds and
y 0
x> 0,
y 0.
x> 0
=>-
In each of the following cases, the shaded region is the graph of the condition
appearing below it.
y
x;;;o
y
-_ +
1 -
r-
-t-.,.x
-1
Reasonably
neat freehand sketches, with cross-hatching used to indicate regions, are sufficient.
1. !
< x <
3. } g
5.
<
y < ii
Ix - 11 <
7. !y - 2i <
9.
2. Ix - 1 1 < l
4.
t and ly - 21
/0
=>-
fyl ;2 ix!
11. x2 + y2 ;2 1
< /0
fx - 11 <
ly - 21 <
6. lx - lj <
8.
'
110
( x - 1)2 + y2 ;2 1
ly - 21 < /0
;;;: lxl
10. x2 + y2 ;2 1
or
=>-
and
(x - 1)2 + y2 ;2 1
38
2.6
Analytic Geometry
12. x2+ y2 1
(x - 1)2+ y2 1
=>
x2+ y2 1
13. (x - 1)2+ 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
18. x2 + y2 1
and
y ;;;; x
=>
x2+ y2 4
and
x lyl
-x - y 1
25. !xi - ly l 1
21. Ix - yl 1
26. Ix+ yl 1
=
19. x2+ y2 1
23.
-x+ y 1
=>
21. x - y 1
20. x+ y 1
22.
17. x2+ y2 1
30. Ix - 31 < t
=>
29. Ix - 21 <Yo
=>
ly - 11 < t
l y - 21 < t
about Q?
aboutP?
aboutP?
=>
PARABOLAS
The
distance
from a point to a line is the length of the perpendicular from the point
and directrix
Dis the set of all points of the plane that are equidistant from Fand D.
Parabolas
2.6
39
axis of the parabola. The point where the axis crosses the
vertex. (There is only one such point, because any such point
::::-+-:+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
the directrix, in such a way that Dis below the x-axis and the focus is above it.
The
number
be a
Now let P
= Jx2 + (y r
FP
and
Therefore
FP
MP
Jx2 + (y - r = J (y + r
x2 + (y - )2 = (y + r
x2 + Y2 py + 2 y2 + py + 2
x2 = 2py
- 21p ..
_
a = l/2p,
and
p__
4
.,
p 2a1
=-.
p__
4
(x, y)
40
Analytic Geometry
2.6
y= ax2
is a parabola, with focus at
(0, 1/4a)
and directrix
y=-1/4a.
y
_ti_
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.
-1) and
FP=MP
<=>
<=>
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,
:;i: 0.
,t: 0.
x= Ay2 + By + C,
If the directrix is neither horizontal nor vertical, then the equation involves, in
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
Parabolas
2.6
1)
41
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
---------------
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.
If you rotate a parabola around its axis, you get a surface which is called a parab
oloid
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)
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.
42
Analytic Geometry
2.6
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.
y
-----
-:..:-...--
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
the motion stops.
= a
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.
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= 1.2,
x=0.2,
x = 1.4,
... ,
... '
x=0.9,
x = 1.8,
x=l,
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
y=ax2 , a>O
Xz
Y1 <yz
Tangents
2.7
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
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.
TANGENTS
is probably the one that you used to find the slopes of tangent lines to circles, in
Problem Set 2.4.)
z2
y2
+b2
1.
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.
44
Analytic Geometry
2.7
But for some curves, tangents cannot be described by the definition that we use
for circles.
(x1, y1)
The
a tangent.
y
We may try to get around this trouble by providing that the tangent line must
only touch the curve, without crossing it.
either. The graph of
y =
x3
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
y =x
examples indicate that the mathematical definition that works for circles does not
work in general.
x2,
(x,
x2
Tangents
2.7
45
x2
---
(x -:F- 1).
Here the restriction x -:F- 1 reflects the geometric fact that it takes two different
It also refers to the algebraic fact that fractions with
!/'m.
Lx
I
I
I
I
I
I
I
mx
mx
x+
(x -:F-
1). We have
(x -:F-
1).
The graph is a line from which one point has been deleted.
such thing as "the secant line through
such thing as the "fraction"
to see that
mx
m1
For x
0/0.
1, there is no
1, there is no
x->1
mx
2.
limit is clear, and so we use it in the definition of the tangent to the parabola.
Definition.
ax2 + bx +
c,
at a point (x0, y0) of the graph, is the line through (x0, y0) with slope
Sx0
where
mx
lim
xx0
111.,,
is the slope of the secant line passing through the points (x0, y0) and
(x, ax2 + bx +
c)
(x -:F- x0).
Analytic Geometry
46
2.7
Even in the general case, the slope is easy to calculate on the basis of this defini
tion. We have
ax2 + bx +
x0)
(x
x0)
= a(x + x0) + b
(x
x0)
= ax + (ax0 + b).
(x
x0)
X - Xo
The graph of y =
- ax - bx0 -
(x
m.,,=
X - Xo
m.,, is
a line with one point missing. The line from which the point
y = m.,, is
on the right.
y
I
I
I
-----
I
I
I
I
I
-4---+--- x
1ro
y=ax+(ax0+b)
Here again, the limit of
y=m.;
Let
(x, y)
y = ax2 +bx +
Then the slope of the tangent to the graph, at
S.,,
(0, 0) .
For each
(x, y),
is
2ax +b.
c.
0,
lxl - IOI
lxl
=m =
"'
x
0
x
-
Thus:
m.,, =
for
>
Ix!.)
0,
mx =
for
<
0.
2.7
Tangents
47
y
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
whenever
x is
one
(?)
x->O
mx
(?);
(?)
Jim
mx
(?)
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
tangent is I for
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
2. Given J = x2
2,
3.
4.
0, x = L and x = 1.
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.
5.
y = ax2
and a point
(x0, Jo) of the curve. Show that the tangent at (x0, Jo) is the only non
(x0, Jo) and has no other point in common with the
48
Analytic Geometry
2.7
(y - Yo)
intersects the parabola only at
(x - x0)
2ax0
ax2 + bx +
c.
6. a) Get a plausible answer for the slope of the tangent to the graph of
point
mx,
x3, at the
as well as you can why your value for the slope is plausible.
7.
a) Show that, if
of y
of y
b) Show that, if
8.
x lxl,
and describe this curve in terms of types of curve that we already know about.
which points does this graph have a tangent?
What is S2?
9.
What is S_2?
At
Give, if
y = x3 - 4x.
Where does this cross the x-axis? At which points is the tangent horizontal? What is
onlyfive points.
10.
y
11.
2x3 - 6x.
Show that every parabola has the reflecting property. In the figure, Tis the tangent at P,
and you need to show that
is a rhombus.
Cl
2.8
2.8
49
+ (a +
d) +
(a
2d) + + (a + [n
l]d).
Tn =
we write
n
Sn=
I a;.
i=l
Given a sum
n,
+ ar +
ar
+ + arn-1
can be written as
2 1
ar -
a 1
ar -
4 l
ar -
ar
ar
ar
n.
2)
2
n .
.o
Rn = '
L., z-.
i=l
3)
Sn =
L [a + (i
i=l
l)d].
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=
2.8
Analytic Geometry
50
1 to
n.
.::.., i 3
i=2
Here we take all values of i from i
fore
I i3
i=2
to i
23 + 33 + 43 + 53
In general, for m
n,
n
I a;
i=m
+ 21 + 64 + 125
am + a.,,+1 +
There
224.
+ a n.
Thus
4
I (af + 1)
i=2
a+ a+ a+ 3
Note that
I a; + 3.
1:=2
applies only to the expression immediately after it; in the last line, we
4)
indicate that
!t=2 (a + 1)
to i
1. >2
i=l
am + Gm+l +
3. I u2 - 1)
i=2
2. I u-1)2
i=l
+ an.
Lf=m a;.
s.
i=3
4. "2i2
i=l
s.
I u3
i=2
3
I (3b + d)
i=2
10. 12 + 22 + 32 +
12. k2 + (k
n2
1)2 + (k + 2)2 +
k-1
--
Is is true that
(n + 1)2
11. 32 + 42 + ... + k2
+ (n - 1)2
1
k
+-
n
I (a;
i=l
Why or why not?
1)
b;) = I a; + I b;?
i=l
i=l
9.
!i7
i=m
2.9
15.
51
Is it true that
n
L kai
i=l
k 2, ai?
i=l
Is it true that
n
L..
i=1
( -) hi 2
n
h3
-a
n
L..
iz?
i=l
()is the number of subsets with exactly k elements, in a given set with
<mis the number of possible 13-card bridge hands; (552) is
the number of possible 5-card draw poker hands. Show that ()
<D
Show that
GD G).
For 0
k b,
elements.
For example,
*18.
2.9
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,
in the figure,
2,
3, .
5.
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:
back on spindle A,
(2)
(3)
put disk
( 1) Put disk 1
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
disks.
n,
52
2.9
Analytic Geometry
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
Lemma 2. P 2 is true.
1 to spindle
C.
B.
Lemma 3. P3 is true.
+ I disks can
also be completed.
=>
P n+1
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,
a)
P1 is true, and
b)
Pn
=>
n,
are true.
The problem of the disks is probably the clearest illustration of what the induction
principle means.
In this section, we shall use it to get short formulas for certain sums.
2.9
Theorem 1.
For every n,
53
n
> = -(n+ 1).
i=l
Proof
n
Ii= -(n+ 1).
i=l
P1 is true, because
a)
Ii = 1 = tCl + 1).
i=l
b)
Pn
=?-
n, because
n
Ii= - en+ 1)
i=l
n
n
L i+ (n+ 1) = -(n+ 1)+ (n+ 1)
=?-
i=l
: i= G+ 1) en+ 1)
n+ 1
I i = -- (n+ 2).
=?-
n +l
=?-
i=l
In this chain of implications, the first equation is Pn and the last is Pn+i Therefore
Pn
=?-
= 1 + 2+ 3+
+ (n - I)+ n,
then
Sn
= n+ (n - I)+ (n - 2)+
+ 2+ I;
to n terms. Therefore
2Sn
= n(n+ 1)
and
Sn
n
= -(n+ 1) '
2
as before. This device is neat but very special. Consider now the problem of calculat
ing
Sn
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
D,
54
2.9
Analytic Geometry
n
Pn: L i2=Ana+ Bn2+ Cn
+ D.
i=l
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
A= t,
This gives
B = t(2 - 3A)= t,
C=l-A-B=i.
n
Pn: L i2=t na+ tn2+ i n+ D.
i=l
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
For some purposes, the following idea is easier to use thanthe Induction Principle.
least element.
(See, for example, Problems 10 and 12 below.) The Well-Ordering Principle and the
Induction Principle are equivalent.
2.9
55
1. Prove by any method that for every n, the sum of the first n odd numbers is n2 That is,
n
c2i - 1)
i=l
n2
This can be shown by induction, but there are at least two other ways.
2. Prove by induction that
1 + r + r2 +
rn+l - 1
+ rn
(r 1).
4.
C3i - 1).
I
i=l
I C4i
6. Find a formula for
Iu
7. Find a formula for
- 2) .
i=l
+ i + 1).
i=l
i (i2
- i).
i=l
A1A2 + AzA3 +
+ An_1An A1A,,.
Pn+l
2pn + 1.
p,,
(Since 210
moves.
2"
11,
1.
1024, this means that the game with 20 disks requires over a million
Thus, if you want to verify that P20 is true, the easiest way to do it is to
11,
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
56
2.9
Analytic Geometry
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.
*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.
?
As a check, the final value is T12
Thomas F. Banchoff.)
364.
* 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.}.
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
2.10
57
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 cut the closed interval [O, h] into n little intervals of equal length, using the
di.vision points
O, ,
11
11
11
11
l)h , h.
[
[(i
'
l
1
2:
[
0,
58
Analytic Geometry
2.10
With each of these intervals as base, we construct a rectangle, using as altitude the
height of the parabola at the right-hand endpoint.
ith interval is
ih/n.
(ih/n)2
Therefore
Let Rn be the union of all these rectangular regions. Then the area of Rn is
n
n h3i2
ai= .
An = .
i=l
i=l n3
h3
i2
.
n3 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).
6
i=l
Therefore
h3 n
An =- - (n + 1)(2n + 1)
n3 6
As
/1
h3
(1 + 1-) ( 1 + -1 )
/1
2n
--
so that
o,
and
1
1 + - -1,
n
1
-o'
2n
)(
h3
1
1
An =- 1 + - 1 + 3
n
2n
Therefore the area under the parabola, from 0 to
1
1 + --1,
2n
and
h,
h3
-.
3
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]
terval we have set up a rectangle whose altitude is the height of the parabola at the
2.10
59
ah
--...,
I
I
I
I
I
I
I
I
I
I
I
I
I
I
h/n]
val, with area 0; and thereafter the area of the ith rectangle is
a =
Let
[(i - l)h] 2
n
h3
n3
(i - 1)2.
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-
h3 n
(n + 1)(2n + 1)
A = 3
n 6
-
As
n2
Therefore
1. Therefore
=
An -
A---+ h3/3,
Let
R
Then the area of
{(x,
y) I 0
R is h3/3.
y
x h
and
h3 .
n
-
x2}.
2.10
Analytic Geometry
60
It is easy to extend this result to the case in which the parabola is the graph of
kx2, k > 0.
y =
y=kx2
.h
i
n
. - (ih)2h
Ui
When we multiply y by
by
k.
Thus, if A11
k,
,2;'.,.1 ai,
as before, and
n
Bn
.2 a;,
-i=l
we have
Bn
,2 ka;
h3/3,
i=l
i=l
Since An--+
k ,2 G;
kAn.
we have
Bn
kh3
3
-+-
2. Let
R =
with
k > 0.
{(x, y) I 0 x h
In general, for
<
0 y kx2},
and
kh3/3.
b let
kx2,
from
to
b. Then
3.
Akx2
(b3
3
a3).
<
b 0, a
<
0 b, 0 a
<
b.)
61
2.10
PROBLEM SET 2.10
Find the area under the graph of y
1.
From 0 to 4
2.
From 0 to 2
4.
From 2 to 4
5.
From -2 to 2
3.
From -2 to 0
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
10.
11.
x2 and y
x2 and y
x + 1.
x.
=
-x2 + 1.
2x2 - 1.
Solve, for the general case, the problem of Archimedes, stated at the beginning of
this section.
1
----=
v'n
Obviously An > 1 for every n.
1
An - l <
?
10 '000 '000
c) Let
An - 1 < E?
15.
a)
2n - 2
En= --.
3
n -1
Under what condition for n can you be sure that En < lo?
E,,. < E?
16.
E,
E,
show that Cn
4 <
62
Analytic Geometry
that Dn <
2.10
n2
3n
1
+ 2 .
102 ?
b) Given any positive number E, under what condition for n can you be sure that
Dn < E?
18.
-b
-
x2
y2
-+- = 1
a2
b2
This can be done by a method somewhat similar to one used in the preceding section of
the text.
A n - h3/n.
Verify this
Hint: Draw a figure showing both the inner and outer rectangles, and explain why
An - A
*20. a) Find a formula for
h
=
. 1z2.
""'
.., l 3
i=l
b) Find the area of the region under the graph of y = x3, from 0 to 1.
*21. a) Let
""'
Functions,
Derivatives, and Integrals
3.1
Consider
some examples.
I)
point P of
Then to each
P.
Thus
we have a function
E--+R
which matches points
2)
P of
E with elements x of R.
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
left-hand end of the arrow), this means that each point P is matched with its x
coordinate x.
3)
Thus we write
---+
between
sets and
between
f: R--+R,
2
X H X .
4)
Every nonnegative real number has one and only one nonnegative square root.
g: R+--+R,
: x H
1-Y
X,
64
3.1
h:
:XH
x
/ /
and
x for x 0
-x for x < 0.
In either case, if xis known, then /x/ is determined. Thus we have a function
i:
R-+R,
XH /x/.
Function
--
Domain
Range
R+
g
h
[2, co)
Rule
PH x
PHy
XH x2
XH V
xH Vx
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
x2
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
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
8)
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}.
kHA.
In Chapter 2 we got a formula for this function:
kH ik3
for every
9)
k 0.
x", for x 0.
y
To each k
region
is odd.)
66
3.1
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
/()-;;)
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.
y
y
r-Y!
I
I
I
I
I
I
In the case shown in the left-hand figure above, the domain is a closed interval
Consider, next, the function g in Example 4, which extracts nonnegative
square roots:
[a, b].
g:
:
R+-+ R,
x 1--4
)"'-;,
The graph of g (the right-hand figure above) is the graph of the equation
To see that this graph is approximately right, observe that
y = )
We get x
<=>
0,
0,
x=
y = )'-;.
y2.
3.1
of x
y
and focus
(t, 0).
67
y
I
I
I
y
I
No
Yes
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],
could be used as the range, but it is not obvious from the figure just what the smallest
2,
1, 2
We write f(O)
2,
f(l)
1,
J(2)
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
x H x ,
and 2 the image is all of R, and in the remaining cases the image is R+.
f: [O, 1]-+ R,
X H
../I
x2.
68
3.1
Here the graph is a quadrant of a circle, as shown on the left below, and the image
is the closed interval [O, 1].
y
{a, b,
for the set whose elements are
a,
b,
N
c,
c,
. ..}
.... Thus
{ l,
2, 3,
.}
{ ...
'
2 , -1, 0,
1,
2,
...}
{O, l}
is the set whose only elements are
and 1.
Givenf(x)
2. Givenf(x)
3.
5.
yn.
For
3.2
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.
*8.
Same question as
6, for y3 +
9. Is the graph of x =
Sketch.
ny = x.
Vy a function-graph?
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
In Section 2.7 we solved the tangent problem for parab olas. Given the graph of
y
x0,
ax2 + bx +
c,
(x0, y0)
of the graph
70
3.2
Obviously a parabola with its axis vertical is a function-graph; its equation expresses
in terms of x. Thus we have a function
f:
R-R
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
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
X2,
2X.
For each x, the value off' is the slope of the tangent to the graph off For example,
I the slope of the tangent tofis 2; and f'(l)
2 I 2.
at the point where x
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.
=
3.2
71
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/.
y
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.
3.2
72
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.
small.
point
is small, thenf'
(x)
It looks as iff'(2)
( 2,/(2)).
0, but
When
>
is
x >
1, f' (x)
y
y
f
f'
f'
-1
entiable.
differ
functions:
The Mean-Value Theorem.
Here by a
chord we mean a segment joining two points of the graph. The theorem
[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.
obvious.
It is geometrically
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.
3.2
73
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'(x)
f(b)-f(a)
b - a
for some x between a and b. In this style we can restate the theorem as follows:
f'(x)
f(b)-f(a) .
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
.... Q
,,,.,,,,, l
.,,.,,.
I
,,,,,.
I
I
I
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.)
3.2
74
y
1
1.
y
4
2.
3.
f,
-1
-2
-4
-2
y
4.
5.
y
4
f
2
-2
6.
f
f
-1
7.
-1
8.
10.
y
11.
12.
f
x
-1
f
x
-1
-1
13.
9.
-1
-1
14.
-1
15.
y
1
-1
-x
3.3
17.
16.
18.
75
-1
19.
-1
20.
21.
3.3
1.
J1
and in this book we will rarely deal with any other kind of function. But some very
simple functions are not continuous.
function defined in Problem 14 of Problem Set 3.1. The graph looks like this:
y
40
30
20
IO
76
3.3
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
x < 0
if
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:
3.3
1)
77
=>
f(x)
)
f(x0.
how
2)
We can makef(x)
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
4)
For each E > 0, there is a o > 0 such thatf(x) is -close tof(x0) whenever xis
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:
0-close to Xo
We can now draw a picture.
y
/(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.
positive number
We now restate
E,
(4)
/x - x01 < O
Thenf is
continuous at x0
=>
78
3.3
Given any
E/2.
2).
2x,
Then, algebraically,
Ix - II<
=>
=>
=>
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
would we do for f (x)
3x,
at any point
E/2.
(What
x0 ?)
I).
y
1-------
d'
x2,
at
79
3.3
Ix
11
< D
lx2
=>
(I, I),
so that
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
78.)
d'
<
f(d')
E;
d'2
1 +
E.
f (d)
<
<=>
d2 <
:>
<=>
lx2
X2
<
II <
< I
+ E
E.
Thus the dotted rectangle in the figure boxes in the graph, in the same way that an
ED-box does.
dd'-box.
good as an ED-box. And, in fact, given a dd'-box, we can always get an ED-box that
lies in it.
d'
d and
d'
1. (Jn
fact,
and so
Ix
11
< D
=>
Ix
< D
=>
lx2
<
<
d',
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.
E
Let
x < d'
Then/ is continuous at
f (x0)
=>
<
f (x)
<
f(x0) +
x0.
y
f(xo)+.- - - - -.---------
f(xo) - -
- - - -
- -
- -
--
- - - -
8
d
E.
80
Proof Let
o
d'
Ix
x0.)
Then
- x0I
< o
3.3
- d and d' -
f (x0)
=>
d <
<=>
< d'
x0
=>
x0.
f (xo)
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.
y
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
limxx 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:
=
0 <
Ix - x0I
< o
lm(x) - LI
=>
Then
Jim
For example, for f(x)
m(x)
x2, x0
x2
x
m(x)
l,
- 1
--
<
E.
L.
we have
+ 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,
=
3.3
81
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:
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
Definition.
0 <
Ix - x0I < b
If (x) - L I <
e.
Then
lim f(x)
x-+x0
L.
Note that here we have simply copied the preceding definition, using f for m: all
x0 was ruled out by the condition 0 < Ix - x01. The left-hand
along, the value x
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).
=
y
I
f(Xo)
L
/
-r?:
-- - ---
-+---<:---x
Xo
I
f(xo).
How close to 3 does x need to be, for 2x to be within 0.001 of 6? (Answer in the form
Ix - 31 <
.8-box
12x - 61 < 0.001. Sketch the graph of j(x) = 2x, and sketch your
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
82
3.
x2 is
3.4
x0
that was used in the text for the same function at the point
x0
d <x <
d' =>
32 -
3.
. ..
dd'-box.
Answer as in Problem
4.
j(x)
6.
8.
f(x)
f(x)
9.
f(x)
10.
2x2,
Xo
V':;, x0
xn, where
I.
8.
5.
j(x)
7.
f(x)
and x0
V,
Xo
x3, x0
x 0
4.
2.
is any number.
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.
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.
=>
If (x) - LI <
e;
Theorems on Limits
3.4
(2) for
o <o
O<lx-xl
Since
83
1-f(x)-(-L)I<"
=>
=>
(2),
theorem holds.
Theorem
2.
If limx-xJ(x)
L, then limx-xo
[f(x) - L]
0.
!I
J---1
I
I
51
rl
Proof
Given f, we get
f-L
(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 ]
L.
To prove this, merely use the previous proof in reverse; move the box along
with the graph.
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)
kf(x)
0 L.
84
2)
Suppose that
> 0.
For every
3.4
=>
3)
Proof for
and so lim,H,,0
(x0, L).
(x0, L). Thus
=>
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.
Jim [ (x)
x-+x f
0
g(x)]
L + I.:.
has a limit, and the limit of the sum is the sum of the limits.
Theorem 6.
If limx-xo
x-xo
Theorem 7.
If limxx
[f(x)g(x)]
L', then
= LI.:.
x-xo
f(x)
g(x)
!=._
I.:
0, then
Caution: The preceding theorem says nothing about what happens when L'
anything can happen, even in very simple cases. If
0.
f(x) = 2x,
then
f(x)
g(x)
2x
and
g(x) = x,
f(x)
x-o g(x)
Jim
2.
--+
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
Theorems on Limits
3.4
85
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.
f(xo).
By Theorem 4,
lim kf(x)
kf(x0),
and 6.)
>'6 0,
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:
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.
Theorem 11.
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.
=
3.4
86
Theorem 12.
continuous.
-- -- ----
Xo
Proof Given/(x)
every
"
>
0 there
Ix
Xol < o
=>
If (x)
kl <
E.
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.
2.
f(x)
kx2
Same, for f (x)
xn (n a positive integer).
Same, for f (x)
x3 - 3x2 + x - 4.
Apolynomial of degree n is a function of the form
3.
4.
5.
Same, for
6. Show that if
f(x)
xn
1 + x2'
then f is continuous.
A function f is
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.
f; and to
show that a function is unbounded, you have to show that no number Mis a bound off.
3.4
Theorems on Limits
87
Find out which of the following functions are bounded, on the given domains, and justify
your answers:
7. f(x)
9. f(x)
11. f(x)
13. f(x)
15. f(x)
17. f(x)
19. f(x)
1
=
--
1 + x2
10. f(x) =
12. f(x)
1 + x2
x
--
1 + x3
1 + x2
1
x + 1
8. f(x)
x
=
, - oo < x < oo
--
1 + x3
--2
1 + x
x
=
--
1 + x2'
x
, 1;;:; x < oo
14. f(x)
, - oo < x < oo
16. f(x)
, -1 < x < 1
18. f(x)
--
1 + x2
< x <
0 x 1
-
, - oo < x < -1
1
-- , 1 < x < oo
1 + x3
x4
- w
--
1 + x3
, 0 < x < oo
, -1 < x < 1
g(x)
0,
x-xo
then
Jim [f(x)g(x)]
0.
x-xo
then
Jim [f(x)sin x]
0.
x-xo
x-xo
0,
88
3.4
d) Show that if
Jim f(x)
0,
xo
then
Jim
xo
[t(x) ]
sin
0.
Jim x2 cos -
x-o
0.
-M
*24.
5 I
-
-----
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)
x-xo
0.
3.5
3.5
89
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
D(f + g) = h'.
D(x2 + 2x + 5)
(x2+2x+ 5).
ax2+bx+c
S,, =
2ax+b.
f(x) - f(x o)
rr;->x0
X - x0
!'(Xo) - 1.lm
_
- = 11m 0 = 0 .
k
= !.im --x->x0 X - X0
x->x0
90
3.5
Theorem 2.
f is
Jf
D(kj)
Proof
Take any
x0
kf for
every k, and
kDf
1.
Im
x-->x0
x0 ,
the derivative of
kf is
Dx2
2x,
D(kx2)
2kx,
We
and
as it should be.
Theorem 3.
f and g
If
D( f + g)
Proof.
Given an
x0
in the domain of
f and g,
f(x) - f(xo)
"'"""'o
x - x0
1. g(x)
Im
"''"'"'o
g, and
DJ+ Dg.
l im
and
f+
- g(x0)
we have
X - Xo
f '(xo)
'
g (x0).
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
For each
x,
3x2
let
f(x)
xa;
3.5
91
'
f (x0)
f(x) - f(xo)
x - x0
o:->o:0
Im
.
= IIm
(x - x0)(x2
= lim
(x 2
!t::t:o
x0x
x0,
x0x
x)
o:->o:0 X
- x0
Xo
xg)
(Why?) Therefore
for every
x3 - x
x -
.,_,.,o
r
= Im
and
x",
n,
we merely need
Proof.
Dx"
Let f(x)
.
IIm
nx"-1,
f(x) - f(x0)
x0
lffi
(x - x0)(x11-1
x"-2x0
X - Xo
z-+ :Z:o
= lim
Then
x" - xg
1.
1.
= Im
x,
n.
(x"-1
x-xo
11-2
x
x0
+
+
xxg-2
xg-1)
(to
Thus/' (x0)
nx-1
for every
x0,
and so
Dx" =
nx"-1,
terms)
92
3.5
because the derivative of the sum is the sum of the derivatives. This is
=
Theorem 5.
I--/'
I
I
I
I
---1I
I
I
x
---+---xc!--o---By definition, f'(x0) is the
x0.
But if f
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)
- f'(x0),
x-xo
X - Xo
.
Itm
then
hm [f(x) - f(x0)]
x-+x0
hm
f(x) - f(xo)
x-x0
f'(x0)
- x0
(x - x0)
0.
(Theorem needed, for the second step?) By Theorem 3 of Section 3.4, lim.,_,, f(x)
0
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)
Dx5
3x2
2x
5x4
6x3,
3.5
93
A correct formula can be derived as follows. Take any x0, and suppose, as usual,
'
f(x)-f(x0)
-f (X0)
.,_,.,o
X - Xo
(1)
g(x) - g(x0)
X - x0
.,_,.,0
(2)
1.
lffi
and
1.
tm
g '(Xo) .
.f(x)g(x)-f(x0)g(x0) .
X- x0
x->x0
(3)
lim
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
g(x ) - g(x0)
X - Xo
f(x)-f(xo)
X - Xo
(3),
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)
x-
=
f(x) - f(xo)
x - x0
1.
lill
x->x0
x ->-
g(x)
+ g(x) - g(xo)
f(xo).
X - x0
x0:
'
f(x)-f(xo)
-f (Xo,
)
X0
X
_
lim g(x)
g(x) - g(x0)
:z:->:z:o
x - Xo
1.
lffi
g(x0),
g'(Xo) .
f(x)g(x) - f(x0)g(x0)
-f'(x0)g(x0) + f(X0)g'(x0)
X - Xo
:z:->:z:o
!.
1m
94
3.5
In words:
Theorem 6.
derivative of the first, times the second, plus the first times the derivative of the
second.
More briefly:
D(fg) = f'g
f(x)
Now
/'(x)
Therefore
f'(x)g(x)
x3,
g(x)
3x2,
g'(x)
f(x)g'(x)
g'f
x2
3x2 x2
2x.
x3 2x
as it should be.
Next we want to find the derivative of the reciprocal
As usual, we take a fixed
have/ (x0) 0, or
x0;
'
1.
1m
xx0
x-->- x0,
l//
of a function f
not be defined.
g(x) - g(x0)
X - Xo
We must
Now
1.
I //(x)
Im
_
-
- 1 /J(x0)
,
X - Xo
xx0
Algebraically,
1/f(x) - l/f(x0)
x - X0
As
5x 4,
g(x0) would
g (x0) -
f(x0) - f(x)
x)f(x
f(
o)(x - x0)
- 1/ [/(x0)] 2 , because/(x0) 0.
f'(x0),
-1
f(x) - f(x0)
x
x - X0
f( )f(x0)
and the second fraction approaches
Therefore
,
g (x )
o
-f'(xo)
[f(xo)J2 '
and so
at every point
Theorem 7.
where f (x) 0.
In words:
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
(f);
( ;1)
D f
f'
fD
(1);
!'
=
;-
f.
-g' =
7
f'g - g'f
g2
In words:
Theorem 8.
the denominator times the derivative of the numerator, minus the numerator times
3.5
95
the derivative of the denominator, all divided by the square of the denominator
(wherever the denominator is not
0).
where
g(x)
0,
we have
() = gf' Jg'
f(x)
D
This is right, because
(-xx42) = x2
Dx2
g(x)
x4,
x2,
By Theorem 8 we get
2x5
==
4x3 - x4 2x
(x2)2
x4
2x.
2x.
(vii)
0,
Dk
kDf,
D(kf)
DJ + Dg,
D(f + g)
nxn-i
(n a positive
Dxn
D(fg) f' g + gf,
=
integer),
n() f[_2
D (J
) = gf' - Jg'
g2
(wherever f
(wherever
0),
g
0).
DJ is
When a function is
But when we describe functions by formulas, it is not always obvious what function
we mean. When we speak of "the function
x
H x2 - x
t H t2 + 2t
x2 -
t2 - tx2 +
x,
j: t
H !2 - tx2
X.
96
b)
3.5
g: x
f->
t2 - tx2 +
x.
(?)
(?),
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)
D.(ax3z + z2)
Da(ax3z + z2)
3ax2z,
ax3 + 2z,
x3z.
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.
2.
0
1. D(7x1 - x8)
4. D -x2 + 1
n
3. D --
x + 1
x + 1
y
5. D -y3 - 3
7.
D--
()
8.
x)
b) Da(3axy + x2 + a3)
D(?y4 _ y2 + 7T)
9. D(l + x)3
b) Dv (x3y + ay3 + xy 2)
x + 1
n 2
6.
1 3. D
x2 - x + 1
x2 + x + 1
14. D-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)
19. Givenf(x) = VJ - x2
3.6
20. Gi veng(x)
(x > 0).
97
24. Now solve Problem 19 by the methods of Chapter 2, without using limits or differentiation formulas.
Find out whether the following formulas are correct, and give your reasons.
25.
D(x2 + 1)2
2(x2 + 1) (?)
3(x2 + 1) (?)
2(x2 + 1)
3(x2 + 1)
2x (?)
2x (?)
-nx-n-1
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
we have
f(x)
f'(x)
Jx
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:
(viii)
D.jx
2,/x .
Dy'X+I
D J x2 + 1
DvI1
x2
2.Jx + 1
x ,
J x2 + 1
-X
--===
J1
x2
(x
> - 1),
(-1 <
< 1).
In each of these cases, we have the problem of finding the derivative of the positive
square root
98
3.6
n.J]=
?;
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 (Xo) =
'
By definition,
lm
lim
[.JI(X5 -- .Jf(XJ
.JTW
.J!!S?__
Xo
.jf(X5 Jf(xo)J
1
[f(x) - f(xo) .
JJ(x) JJ(xo)J
Xo
+
x->a:o
g'(x0).
lim
a:->a:o
= 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
and since the limit of the quotient is the quotient of the limits, we get:
Theorem 1.
DJ = f' .
] 2J]
Let us try this on the function x H Jx + 1.
for every x. Therefore
Here f(x)
=x
1
DJ.X+l = J.X+l '
2 x+ 1
which is the right answer.
DJx2
+ 1
For
x HJ x2
+ 1, we have
f(x) = x2 +
f'(x) = 2x,
x)
= DJJ(x) = f'( =
2JiW
1,
2Jx2
3.6
.JJ(x0),
(1)
x """' x0
Since
.J is continuous,
99
lim.JJ(x)
reasonable.
(1)
is
=>
we have
f(x) """'f(xo)
=>
x """' x0
.Jf(x)
=>
"""' .J f(x0),
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
g(x)
f"(x),
'
(Xo)
rCx) - rCxo)
X - X0
x-+x0
1.
Jill
Iim
x-+x0
g'(x0)
+ r-1cxo)l
nj"-1(x0)f'(x0),
(2)
xx0
fk(x)
(3)
xx0
it follows that
f2(x)
fk(x0)
lim
lim
f(x)
[f(x)f(x)]
f(x0),
f(x0)f(x0)
f2 (x0),
100
3.6
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.
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
f'(x) =
-Dxk
-kxk-l
=
x2k
(xk)2
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.
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'.
-
3.6
101
For convenience of reference, we list all the differentiation formulas that we have
so far:
(i)
(iii)
(v)
Dk = 0,
D(f
g) = DJ + Dg,
D( fg) = f' g
()
D L
(vii)
(ix)
(ii)
DJ]=
gf,
(vi)
()
J'
(x)
nr
(n :rt= 0),
- [_
!2'
D ! =
f
r=
...) DyX
(Vlll
gf' - Jg' ,
2
g
(iv)
D(kf) = kDf,
Dxn = nxn-i
2Jx
---=,
njn-1.f'
(n :rt= 0).
I.
2. D
8. DVx3 +2x +1
7. DVx(x - 2)
x2
1
10. D- -
x2 +1
-
11. D
DVx2 (Warning:
-J
1
x
(x2 +2x+1)2
x2 +1
6. D- -x2 - 1
9.
v'x - I
--
x2 + 1
12. DYv'
+x
3.
4. DVx4 +5x2 +2
13.
1
(x2 +x +1)2
>
16 Dx
0,
x3y2
(x2 +y2)2
[Hint:
_
.a,
By definition, v
.{;xv
xv is
( :X)1'
(x
>
o) .
x'l>.
Therefore you need to show that the qth power of the right-hand side of the above
equation is
xv.]
18.
definitions
Find
the answer in a form which brings out the analogy with the
x312 is
continuous.
102
3.7
*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'.
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.
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.
a - b
q is a positive integer.)
q_
q
b) Find DV x, assuming that V x is continuous.
(Here
*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
We found that
{(x,y) \ 0 x h, 0 y
Ah
kx2}.
h3.
y
2
y=kt
3.7
103
that h can be any positive number, and that when his named, Ah is determined. Thus
Ah can be regarded as a function.
F(l) =3
F( 3)
1 3 =- .
to
3;
this is
F(3) = 33 = 9k.
3
And so on.
the values of
F:
F(x) = x3
3
Here we have replaced h by
x in
3
A"=- h.
3
3
.
-x.
F.x1--+
3
For each
x,
the value of the new function is the area under the graph of the old one.
y
y=f(t)
n/l
I
I
I
I
I
I
I
I
I
I
I
I
104
3.7
y=t+l =f(t)
4
I
I
I
I
I
1
(x, x+I)
2-
I
I
I
I
I
I
I
/I
//
I
1 /
I
/
I
I
//
I
/
a=l
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).
=
t2,
t -2.
Here we have
f: [-2, oo)--+ [O,
co .
3.7
105
-2
For each x -2, F(x) is the area under this graph, from -2 to x. By our old formula,
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.
a to t
x.
y
fi J(t)
I
1x f(t) dt
I
I
and
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
106
3.7
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)
are positive numbers, representing the areas of the two shaded regions.
Ai positively; it is the area of a region above the t-axis.
it is the area of a region below the !-axis.
y
Thus we have
b)
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,
We count
We count A2 negatively;
3.7
107
y
(x, 2-x)
y
4
y=f(t)
x?
a
L:
f(t) dt
J(2 -
t) dt
t(x
-1
- 1 . Then, for
a =
x ;:;;; 2
2
+ 1)(3 +
we have
- x)
Hx
+ 1)(5
x),
x2
we have
J"'(2
- t) dt
-1
For
x;:;;;
-1
- t(x - 2)(x
- 2)
we have
J"'c2
t) dt
-1
r-\2
-[t(-1 x)(2 - x
Hx
- x).
-
- t) dt
J.,
+ 3)]
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.
a) Consider f(t)
f
valid when
;;;;
0. Sketch.
ltldt,
108
3.7
one
d) Let
F(x)
Get a formula for F'(x), valid when
f'
t
l l dt.
> 0.
f'(t2
3.
+ 1) dt.
Algebraically,
when
< 0,
when
sigt
when
0,
> 0.
f'
sig t dt
i1
sig t dt
I;
1-i
13
sigtdt
sigtdt
3,
!)
1,
x.
3.8
f1 sig t dt
J_l
-1 + 1
O;
and so on.
sig t that you did forf (t)
Do the same six things for f (t)
Do the same six things, for f(t)
t It!.
a) Explain why
=
4.
5.
109
t2
+ 1.
( t3 dt
J_l
1
0.
b) Explain why
(1 t273 dt
J_l
0.
6.
I (-a)
Explain why
0
7.
Explain why
0
3.8
<
I (0)
fa f(t) cit
J3 -dt
<
I (a)
=
0.
o.
14
< 12
3 1---+t2 dt
J
1
<
.
3
5.
F(x) =
then
{xf(t) dt,
F'(x) = f(x)
for every
x.
y
y=f(t)
y=F(x)
110
Take a fixed
x0
By definition,
F'(x0 ) -
X -
o:->o:0
Now
F(x)
Lf(t) dt
F(x) - F(x0)
<
x,
Xo
F(x0)
and
Therefore
x0
F(x) - F(x0)
1.tm
For
3.8
f"f(t) dt.
a
"
C
f(t) dt - L f(t) dt.
.
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
x
- x0
x0
to
x,
F(x) - F(x0)
X - Xo
and the approximation gets better as
f(Xo'
)
gets closer to
x0
y=f(t)
For the situation shown in the figure, in which the graph rises to the right of
it is easy to see why the approximation is good.
F(x) - F(x0)
Here we have
(x - x0)f (x0)
+ E,
where E is the area of the little curvilinear triangle at the upper right.
E <
e(x - x0)
and
--E
X - X0
< e,
Now
x0,
3.8
where
111
F(x) - F(x0)
!"'3
(x
x0)f (xo),
F(x) - F(x0)
X
,,,..,
""
Xo
f(Xo) '
x0
If x < x0, the same approximation formula holds, although the reasons are
slightly different.
( x0 -
x)f (x0);
!"'3 (x -
x0)f(x0),
and
as before.
F(x)
- F(x0)
X - Xo
,,,_,
rv
f(Xo'
)
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
F'(x0)
f(xo).
y
F
112
3.8
D.,
at each point
f'f
(t) dt
a,
then
f(x)
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.
this chapter, and see, in the meantime, what the theorem is good for.
Consider the
following problem:
Problem I.
x4, from x =
0 to
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=f(t) =t4
y=x4
F(x) =
t4 dt.
It might seem that Problem 2 must be harder, but this is not true.
that, while information about the number
something about the function
The point is
1 tells us
F, namely,
F'(x) = x4
x4 as its derivative. We get powers
x, using the formula
by differentiating powers of
Thus
3.8
113
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) =
To be exact:
G'(x) =F '(x)
f'
t4 dt.
for every x,
(1)
G(O)=F(O).
(2)
G(O) = t05 = O
and that
F(O) =
Equations
(1)
t4 dt = 0.
G(x) =F(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.
I.
If
F(a) = G(a)
(3)
and
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) =
't4 dt,
G(x) = tx5
114
3.8
Suppose that F(b) G(b)for some bon the interval I. For each xon [a, b], let
H(x)
Then H(a)
0, but
H(b)
F(x) - G(x).
0.
y
The slope ofthe chord joining the endpoints ofthe graph of His
H(b) - H(a)
b-a
By MVT there is an x between
O.
H'(x)
H(b)-H(a)
b-a
Therefore H' (x) 0. But this is impossible: for every xwe have
H'(x)
F'(x)
G'(x)
0.
Obviouslythis area is
Let
F(x)
3.8
Then
F'(x)= I
by Theorem 1.
Now
115
x2,
F(-1)=0,
G(x) = x - tx3
and
G(-1) = -1 + t = -i.
(?)
G(x) = x - tx3 + i.
Then
G(-1)=0,
for every
x.
1:
Example
to
x = 2.
2.
G(x) = F(x)
Therefore
(1
t2) dt = F(l)
G(l) =
-} + i = f.
y = x2 + x + 2, from x =
-1
:
-1
1
1
Let
fy2 + t + 2) dt.
F(x) = Jy2 + t + 2) dt.
Then
F'(x) = x2 + x + 2.
so that
G'(x)
F'(x).
We find that
3.8
116
-t + t - 2
G(-1)
2x + ll, so that G( -1)
=
rt2 t
+
-1-l-.
+ 2) dt
G(x)
tx3 + t x2 +
0.
F(2)
G(2)
.
l 8 + t
+ 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.
F(x)
G'(x)
We then arrange for
G(a)
f'f (t)
G
Given
dt,
such that
F'(x)
f(x).
!Jf(t) dt
that
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.
Find
a)
fu2
2t
5)dt
b)
f'
tndt,
11.
cx2 + 2x + 5)dx
c)
(z2
+ 2z +
5)dz
3.8
117
5.
a) Find
(t3 + t
1) dt.
F(x) f + t - 1) dt.
(t3
6.
x
v' x2 + 1 '
y=
1 2.
x/ v'x2 + 1,
from
to
Dv'fb) Find
f'
2v'J'
-
Jl t2t + 1 dt.
__
-1
v'
__
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 +
=
Express
F'(x)
formula.
not
10.
dt.
v'()0
F(x)
by an elementary
118
3.8
1
_
1 __
x + 1
.
vz -x
--
Find
Find
15.
Find
16.
x2
(1
dx.
xa)2
1
-
1
never appears as the derivative of a power of x. If we allowed n = 0, then xn = xo
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
3.9
3.9
119
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.
y
4
3
2
1
-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,
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.
120
3.9
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)
F(t)
a(t) = k1-
(k1
const).
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)
where
3)
W(t) < 0
for every t on
[c, d],
W(t) is the weight at time t. W(t) < 0 because weight pulls things downward.
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.
F(t)
W(t)
Therefore
k2 < 0. Therefore
a(t) = k1
k2
for every
< 0
t;
and
a(t)
k3
< 0.
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
acceleration of every freely falling body, regardless of its mass. The number
k3
g= -m
is called the
then numerically
32,
measured in ft/sec2
3.9
121
a(t)
-g,
32 ft/sec2
We now consider the problem of finding the functions that satisfy the equation
a(t)
Problem.
J"(t)
-g.
The function
f:
R---+R
-g for every t,
form:
= -g for every t,
v(O)
v0,
c) f(O) =Yo
a)
a(t)
b)
Thus our data consist of (a) the constant acceleration -g, (b) the initial velocity
v0
a)
We know that
v'(t)
a(t)
-g
u(t)
-gt (?)
But this is
u(t)
-gt + v0
Our function u then has the same derivative as v, and has the same value at t
By the uniqueness theorem, u and v are the same function, and so
v(t)
b)
-gt + v0
f'(t)
v(t)
-gt + v0
0.
122
3.9
has
-gt + v0 as its derivative; the only trouble- is that z(O) is 0 instead of y0 But
The function z then has the same derivative as f, and has the same value at
By the uniqueness theorem, f and
f(t)
0.
t2 + v0t + Yo
2
Let f be a function R
__,,.
R.
If
a)
b)
then
d)
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)
-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
such that
a)
b)
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
123
3.9
v(t)
v,
- gt + gt0 + v0
z(t)
we observe that
(t0)
z(t)
(g/2)t + v 0t0
Then
has the same derivative as/, and has the same value at t0. By the uniqueness
f(t)
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)
Solution. Let v
f'.
3,/'(3)
Then v'(t)
1, and/(3)
2, what is/?
3.
3t.
Adding the
v(t)
3t - 8.
Now
f'(t)
This suggests f(t)
we have
f(t)
3t - 8.
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)
3. f"(t)
3t + 4,/(0) 4
-1,/'(0) 2,/(0)
=
2. f'(x)
4. f"(x)
x3 - 7x + 5,/(0)
3x2,f' (1) O,f(1)
-1
124
5. J"(t)
7. g (x) =
9.
I (t) =
3.10
1
,
6. f (x) = 2 ,/(1)
3
t ,f'(O) = 1,/(1) = 0
_
x
1
,g(O)
x2
t ,/(2) =
-1
10. f'(t) = t2(1 + t3)10,f(O) = 2 (By all means, do not use the binomial theorem on this
one.)
t2
(I
+ 3
t )2
,/(1)
2
=
14. g"(t) =
+ l 3,g(O) = l,g(I)= 1.
(t
)
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?
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 let
F(x)
we take a point
Let
x0;
ixf(t) dt;
(x0,f(x0)).
3.10
125
f(xo)+l-----------+-----,
x0-5
Thus
Ix - x0I < a
=>
<=>
Xo
f(x0)
- E
E.
We are going to use these inequalities to get information about the function
m(x)
Here
1
=
--
X - Xo
[F(x) - F(x0)].
F'(x0).
Evidently
F(x) - F(x0)
i
f
"'f(t) dt -
f(t) dt
!(t) dt,
:ro
and so
m(x)
If f is positive and
_
_
x - X0
[F(x) - F(x0)]
l_
X - Xo
f(t) dt.
"'o
shaded region.
Case 1.
Suppose that
f(xo)
Therefore
(x0 ;;;
;;; x).
126
3.10
and so
[f(x0) - E:](x - x0) <
ff(t) dt
<
'
"'
If (x0) - m(x)I
=>
(.1)
< E.
(x t x0),
1.
Therefore
"'0
[f(xo) - ]
"'0
[f(x0) + ]
dt J:f(t) dt i
<
<
dt.
I:f(t) dt
<
f(x0) - E <
--
J"'f(t) dt
X0 - X "'
<
f(x0) + .
When we interchange x and x0, this changes the sign of each of the factors in the
middle of this expression. Therefore
f(x0) - E <
To sum up:
X0 - o < x < X0
=>
X -1 Xo l"'f(t) dt
--
a:0
<
f(x0) + .
=>
E,
(2)
exactly as in Case
Therefore
1.
0 <
Ix - x01 < o
lim m(x)
a::-.a::o
lim
a::-t-a:o
=>
and 2, we get
Xo
X -
[F(x) - F(x0)]
3.10
127
f(x)
3.
h(x)
5.
g(x)
7.
f(x)
9.
h(x)
11.
g(x)
13.
15.
17.
f' (t3 1) dt
{"' dt
f(x)
Jo t
h(x)
i v t2 + 1dt
f (1 + t3)100dt
8.
10. f(x) J: v2 + tdt
2
12. h( )
1 t 4dt
"'
f
-x
14. J1x t3 dt+ 1
Jx -1t2++t110dt
2
18. {"' J 1 + t dt
Ja 1 + t4
Vl
v(
4.
6.
v!
l
+
16.
1T
D.,
f,
f /(t)dt
f(x)
f(x),
derivative of
*20.
2. g(x)
g(x)
dt
iJ,,x 4dt
2 t +1
J"' J 1 + tdt
t
-1
19.
r t2dt
J("'2,, (t4 - t)dt
r4 + t8dt
J("'4,, 1dt
{"'
dt
Jo 1 + t2
f (t2 + 1) dt
rJo2x
Vl
t8dt.
vl
t8 dt.
"'
(
Jo
'
Trigonometric and
Exponential Functions
4.1
DIRECTED ANGLES. TRIGONOMETRIC
FUNCTIONS OF ANGLES AND NUMBERS
If
--+
AB
and
-+
AC are rays
LBAC.
A,
word angle in such a way as to allow "zero angles" and "straight angles."
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.
Initial
Terminal
L.
0
OL.
A
LBOA A
LAOB
--+
OB
LAOB
Thus
128
-+
--+
(OA, OB);
LAOB is
Thus,
OA is
LBOA.
the ray
different from
-+
the
Directed Angles.
4.1
129
coordinate systems are marked R, and the left-handed ones are marked L.
y
Lx
"X
We can now define the trigonometric functions of an angle LAOB. The procedure
--+
is the positive half of the x-axis. On the terminal side OB we choose a point P 0.
P has coordinates
OP is a positive number
y
x
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
tan LAOB
sec LAOB
y/r,
yfx
rfx
cos LAOB
(for
(for
x
x
0),
0),
cot LAOB
csc LAOB
=
=
xfr,
x/y
r/y
(for
0),
(for y
0).
4.1
130
P,
-1
%
P0
-1
(1, 0),
1)
Given
fJ
fJ.
ends is P0
2)
Given
fJ
lfJI.
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.
rather than numbers.
are points
Po+2" ==Po,
for every e.
27T
to
fJ,
take another round trip around the circle, ending at the same point P0 where we
began.
Similarly,
and
fJ,
let
LO== LP00P0
Directed Angles.
4.1
The symbol
number e.
L()
is pronounced "angle () ;
"
L()
131
We now define
sin()= sin
L()=
sin
LP00P0,
cos() = cos
L()=
cos
LP00P0,
and so on.
y
L()
their geometric meaning. But for some purposes, it is simpler to forget about angles,
then
sin()
COS()=
y0,
Xo,
Yo
Xo
Xo
Yo
(whenever
x0
0),
(whenever
Yo
0),
sec()= -
(whenever
x0
0),
_!_
Yo
(whenever
Yo
0).
tan() =
cot() =
Xo
csc ()
Using these definitions, we can derive the usual formulas. Since P0 is on the unit
circle C, we know that
OP0=
and we have:
Theorem 1.
1.
Therefore
y=
1,
132
If the sign of
4.1
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
This gives:
Theorem 2.
For every
-Ye
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
Sln
7T
-
0,
cos 0
1,
cos
sin 7T = 0,
Theorem 4.
For each
'!!.
2
1,
= 0'
cos 7T = -1.
6,
sin (7T +
6) =
-sin
6,
cos (7T +
6) =
-cos
6.
Directed Angles.
4.1
133
Proof For each (), the points P9 and P1T+9 are symmetric across the origin. This
holds in all quadrants. Therefore
But if the angle is known, the number() is not determined. In the figure on the right,
!7T,
or
()
n,
21T + !7T
or
\1 7T,
()
t7T - 21T
-i1T.
If an angle
!7T + 2n1T.
LAOB
(We have seen that every angle
L().
For
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.
LP00Q
and
LP 0Os
3
L 7T.
4
134
4.1
x',
LQOS
'
y , we can also say that
L 7T
LQOT
and
LTT.
LSOP0
(- )
37T ,
4
and so on.
PROBLEM SET 4.1
be based on the definitions and theorems given in this section of the text.
1.
6.
sin
2.
7.
csc z
23. sin(7T - 6)
27. sec
( 7T - 6)
cos
16. cot
19. tan(7T + 8)
sin
8.
4.
tan x
cos
sin
12. cot2 8 + 1
11. 1 + tan2 8
1 5. tan( -8)
cos
3.
(-8)
9.
13.
20. cot(7T + 6)
24. cos(7T
8)
28. CSC(7T - 6)
cotx
sec
csc
secy
csc
sec
17. sec(-6)
21. sec(7T + 6)
22. csc(7T + 8)
2 5. tan(7T - 6)
sin
801;:;; 10 - 001.
--
14. sec2 8
sec x
10.
cscx
5.
26. cot(7T - 8)
='
4.2
4.2
135
Jn the figure on the left below, We have X9 =COS 6, andy9 =sin 6, by definition Of
the sine and cosine.
y
P(x, y)
Let
P be
any point of
---+
OPe,
and let
OP=a.
are
x=a cose,
Proof
y=a sine.
By similar triangles,
lxl =lxel
'
a
1
Therefore
and
lyl =a IYol
In these equations
xe also agree in
x=axe= acose,
and
y=aye=a sine,
y and y9
Therefore
136
4.2
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
= a2( cos2e
b2 - 2ab cose
sin2e)
b2
+ a2 sin2e
= a2 + b2 - 2ab cose'
which was to be proved.
Theorem 3.
For every
and cp,
cosce
Proof Let A = P0,
cp)
Then
A= 1
( ,0),
c =(cos
e
(
rp), since
rp)),
cp)
= cos2ce
rp) - 2 cos e
(
= 2 - 2 cosce
1]2
r/J).
sin2 e
(
+
rp)
cp)
+1 +
sin2 e
(
cp)
4.2
137
---+
We now set up a new coordinate system, with OP6 as the positive x'-axis.
y
cos2
8 -
= 2 - 2(cos
8 cos cf>
8 cos cf>
8 -
sin c/>)2
- sin
+ 2 sin
8 sin cf>
+ sin2 cf>
8 sin cf>).
(8
+ cf>) = 2 - 2(cos
8 cos cf>
- sin
8 sin cf>),
and
cos
(8
+ cf>) = cos
8 cos cf>
- sin
8 sin cf>,
For every
and cf>,
cos
Proof
(8
- cf>) = cos
8 cos cf>
- </>) =
and
(6 +
cf>).
4.2
138
For every 8,
Theorem 5.
cos
Proof.
( - )
e
and
=sine,
( )
sin
-e
=cose.
By Theorem 4,
cos
( )
-e
+ 1
( )
-e
sine,
= sine.
8 for 8, we get
cos
[ - ( - ) J
e
=sin
( )
-
e .
Therefore
sin
( )
-e
= cose.
complementi sinus,
meaning
Theorem 6.
(fJ +
Proof.
sin(fJ + cp) =cos
=cos
[ - (fJ + J
cp)
( - fJ)
cos cp + sin
( )
-
fJ
sin cp
3. cot (8 + </>)
5. sin 28
7.
cos 2li
tan A +tan B
=
l - tan A tan B
cote cot</> - 1
=
cote + cot</>
2 sine cose
1 - 2 sin2 IJ
cosine
2. tan (A - B)
4. cot (A - B)
6.
cos 28
8. cot (Ii
2 cos2 e
</>)
is from
4.3
31T
9. a) sin-=
2
31T
11. a) tan-=
2
12.
15.
18.
o
(}
2 sin - cos- =
2
2
+ cos(}
2
(}
tan 2
31T
10. a) cos2 =
b) sin (3 7T + o)
2
b) tan (3 7T + e
2
13.
16.
sin(}
+ cos(}
(} 1 - cos(}
19. tan - =
sin(}
2
2 cos2
0
2
=
14.
- 1
- cos 20
=
2
17.
l
J
b) cos (3 7T + o
2
139
+ cos 20 =
2
*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.
show that sin ((}/2) is not determined if only sin(} is known.
That is,
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.
If we try, in a straightforward way, to find the derivative of the sine, we get into
trouble.
By definition,
!'(
if the indicated limit exists.
x0)
For f (x)
Sln
, Xo
1.
f(x) - f(xo)
,
X - Xo
x->x0
1m
sin
x,
sin
1.
!ill
x->x0
- sin x0
X -
X0
4.3
140
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
f(x) - f(xo)
X - Xo
using various algebraic tricks.
It is as follows. Let
The symbol
Ax
Ax= x - x0
Ax
It is pronounced "delta
is the difference in
Similarly, let
Llf
Here
A/ is
x,"
Obviously, x
(x0,f(x0)).
and
A/ is
1J
I
I
I
I
x0
I
I
x=x0+t:.x
x->x0
lim
ax->O
+ Llx.
t:.f=f(x)-f(xo)
and
x0
x0 to x.
Ax
Jim
f (x) - f (xo).
x.
xo
Llf
= lim f(
Llx ax->O
x0
Ax) - f(xo)
= f'(xo)
Ax
are merely two different ways of describing the same limit f' (x0) .
4.3
141
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.
f(x0
=sin
Ax)
(x0
Ax)
=sin x0 cos
Ax
Thus we have
f'(Xo)
1.
sin
=Sill Xo = Jill
.
(x0
.
sin
=IIm
.
Ax
x0 cos Ax
Ax
.
cos Ax
Sill Xo
Ax
il.x->O
- 1]
x0
sin
+ cos x0 sin Ax
il.x->O
=I!ill
Ax)
il.x->O
1.
Jill
sin
cos Xo
x0
sin Ax .
--
Ax
il.x->O
cos Ax
and
=0
Ax
il.x->O
Jim
sin Ax
Ax
il.x->O
(1)
= l.
(2)
x0
=cos
x0,
and
D sin
=cos
x.
Since cos
= I, the first
(0
il.x->O
Ax)
- cos
Ax
=cos ,
0.
(3)
0.
(4)
(0
Ax)
- sin 0
Ax
=0 and sin'
=sm ,
= 1, then sin'
Theorem 1.
sine
Jim -8-
8->0
1.
x.
Ax,
142
4.3
Theorem 2. IIm
cose - 1
9-o
0.
cose - 1
e
o-+o
Jim
9-+o
cose - 1 . cose +
/:::i
cos2 e - 1
e(cose + 1)
cose +
1
9
-sin2 e
e(cose + l)
lim
cose - 1
e
-lim
8-+o
][
i_n_e
s_
e
][
1im sine
0
8-+
follows from
.
J
.
o-+o
cos e - 1
o-+o
(Query:
-1 . 0 . l
0.
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,
---+
1 as ()
-+
1Tj2
we have
sine e tane.
y
-1
4.3
and
QS
143
Since
=
tan(),
11
a1 = a2 =
= an.
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
and that
for each i.
Therefore
RP<A11< QS,
and so
sin()<A,, <tan 8.
As
n -+ oo,
A11 -+ () .
Therefore
sin()() tan().
144
4.3
<
<
A,. or A,,
An or A11 b is always
b is not necessarily preserved.
For example,
+ 1
for every
> 1
n,
11
(?)
lim
/1
+ 1
>
11
n-oo
1.
(?)
7r/2,
1
As 8-+ 0, cos 8-+ cos 0
1.
1_
- sine - cose
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
y=
cos e
e
----y = sine
1.1m
9 ... 0
e
--
sin e
"
and
1.
As we have seen,
4.3
145
Theorem 3.
D sin x =cos
x.
Once we know how to find D sin x, the derivatives of the other trigonometric
cos'
cos
Ll.x) - cos
(x0
X0
+
---'--"------'-----=
x0 =lim
Ll.x
Ax->O
1.
cos
=Im -
Ax->O
= cos
Xo
(cos
1.
cos
Im
Ao:->0
x0)
Ll.x -
Ll.x
(sin
x0)
Sill Xo
1.
sin Ll.x
Im -.6.x
Ao:->O
1 = -sin
x0.
Thus:
Theorem 4.
D cos
x = -sin x.
D tan
x =sec2 x,
D cot
x = -csc2 x,
4.3
146
y
g
(}
Lt------
(}
Two illustrations of the theorem are shown above. The theorem is geometrically
clear, and is also easy to prove.
any box for fat (x0,
at (x0,
L),
L) is
L).
L,
by definition of a limit.
y
f
g
-+-Ll--.,--.c.+-----,,.>.'-(} -.:-.-/ :
:
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.
(}
4.3
147
If
g(x) h(x) f(x),
and
limf(x)
lim g(x)
L,
x-+xo
L.
All of these ideas are very closely related, and we shall refer to all of them as the
squeeze principle.
PROBLEM SET 4.3
1. D tan x
5. DVl - sin2 x
6.
9.
:2
4. D csc x
DVl - cos2 x
7. D cos2 x
3. D sec x
ce>J"
D 2 sin x cos x
8. D(cos2 8 + sin2 8)
10. Dv'l + tan2 8
sin x
12. D
cos x
13. D l
.
+ smx
1 + cos x
*25. Make a plausible guess for D,, sin ax, and verify it if you can.
148
v'x.
4.4
f'
g,
x,
(b) g'
cos
then
=
-f,
(c)
and /2,
g2
f(O)
(d)
0,
g(O)
1.
Is it possible that there is another pair of functions satisfying the same four conditions?
function
g1
Consider the
Sill
4.4
I
=
cos'
COS,
sin 0
-sin,
cos 0
0,
1.
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
x
- x0,
so that
x0 + .6.x.
x0 of I.
For each
point
/.,
6}L
I
We let
IJ..f
f(x) - f(x0)
Llx
'
f ( Xo) _
by definition.
1.
lffi
x--+x0
f(x) - f(xo)
,
X - Xo
f'(
x0 )
1.
im
ax->O
1.
1m
ax->O
.6.f
.
ilx
-
of I, we let
.6.x
4.4
149
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.
Xo
In the figure, the line Tis the tangent to the graph off at the point
the slope of Tis
f' (x0). If
(x0,f(x0)).
Thus
X0
because the slope of the segment from P to S is the slope of the line T. This gives
Y
This quantity is called the
in the figure.) To repeat:
by definition. Since
x0
f(xo)
f'(x0) Llx.
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.
Llf::::::; df
when Llx::::::; 0.
Let us try this on some numerical examples, and see how good the approximation
looks.
150
Example
1.
4.4
Let
(x
f(x) = .J
and take
25,
Xo=
bx
O);
0.4.
y
5
4
3
2
---f---'
---'---'---'---'---'---''--'--'---'---'---!--,__ X
0
12
14
24 I 26
22
4
6
10
20
8
18
16
2
Xo=25
Then
and
f(x0)
f'(x)
.J25
1;-
2yx
df= lo bx
5,
(x
>
0),
lo (0.4)
0.04.
dfbf
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,
.-/25.4
Thus the error in our approximation is
approximation
Ax= 0.4 is
b f df wasn't
5.039841.
df
Ax= 0.1,
1
;- (0.1)
2y25
.J25.l f (5)
we get
dj
0.01;
5.01.
bx
is small; and
4.4
151
)25.1
so that our error is
0.00001,
5.00999,
Using t:.x
0.01,
we get
11-o(0.01) 0.001;
)25.01 f::::! 5.001.
dj
)25.01
f::::!
5.0010.
b.f f::::! df
f'(xo) b.x
f'(x0)
Multiplying by
b.x,
f::::!
we got
b.f
b.x
when
b.x
when
b.x
(1)
0.
f::::!
f::::!
(2)
0.
b.x
and
then the product
f::::!
f::::!
0,
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.
E(b.x)
f (x 0
Then
b.x) - f(x0)
b.x
lim E( b.x)
6.x-+O
because
1.
!ID
ax-+O
/(xo
j'(xo)
0,
(b.x
-:
0).
152
4.4
Thus the graph of the function Elooks like the figure on the left below.
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.
lim E(tu)
E(O)
We now have
0.
Ax-+O
x
---+- -xo- --a XoX-o+-
a -
is differentiable at
x0
such that
i)
6.J= f'(x0)6.x
ii)
limxo
(Proof
x0,
E(6.x)6.x,
(6.x)= E(O)
and
0.
d
f
j'(xo) 6.x
is a good approximation of
6.j = J(x0
6.x) - j(x0)
4.4
when !1x
153
0.
E(!1x)!1x.
11x.
df is
In many cases, it is possible to estimate the largest possible error that can result
when you use
df as
an approximation for
!1f
Appendix D.
Following is a partial table of the sine and cosine functions, for ready reference in solving
sin x
cos x
0.0814
0.0814
0.9967
0.2094
0.2079
0.9781
0.1222
0.3840
0.1219
0.9925
0.3746
0.9272
=
=
0.9964).
0.2108).
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.)
it is not to be expected that an approximation process based on the differential will give them
exactly.
8.
>'Y27.1
10.
{116.3
[Answer: 3.004]
V25.2
9.
11.
[Answer: 5.0200]
-V' -7.9
12. One of the standard approximation formulas used in mathematical physics says that
sin x
when x ""'0.
""'x
Explain how this formula is related to the ideas in this section of the text.
Consider the general approximation formula
b.f"" df
. What form does this take, for j(x)
13. Same question, for
cos (1 - x)
sin x, x0
""
O?]
when x ""1.
[Hint:
154
4.5
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.
when x""' 0.
""'1
.3; v
x
1 +
when x""' 0.
when x""' 0.
R:Jl-x
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
<f>(x)
(x2
3x
+ 5)5,
x2
3x
+ 5,
we let
g(x)
so that
5g4g',
</>'(x)
5(x2
3x
+ 5)4(2x + 3).
(x)
Jx2
we let
g(x)
x2
+ 1,
+ 1.
</>'(x)
2Jx2 + 1
x_
Jx2 + 1
__
Composition of Functions
4.5
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.
g: x
x2 + 3x + 5.
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 +
(x2 + 3x + 5)5
cp(x)
==
Diagrammatically:
g
x H x2 +
;v
1.
x2 +
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
We sum all this up in the following definition:
+ 1
( ?).
156
4.S
Definition.
g: A-+ B,
the composition
f: B-+C,
f(g): A-+ C
f(g)(x)=f(g(x)).
Here, for each
x, f(g)(x) denotes
x.
Diagrammatically:
f
AHBHC.
Let
f(u)=sin u,
Then
f(g(x))
A?
g(x)=2x+ 1.
sin g( x)
What are
sin
(2x+ 1).
B and C?)
Let
f(u)=u2+u+I,
g(x)
Then
A, B,
Example 3.
and
Let
Then
Example 4.
f (u)= sin
u,
g(x)=x2
I.
- 5.
Then
J; +
C?)
J;.
0 .
1)
1)
1)
5).
x2
- 1)
- 1)
dt.
dt.
are given a function , and in order to investigate the function , we express it as the
4.5
Composition of Functions
f2(t4
</>(x)
1)
</>.
157
For example,
dt,
</>(x)
(u(x)
Jo
where
g(x)
Thus
</>
where
f(u) =
lu(t4
(t4
dt,
1)
x2
f(g),
1)
dt,
g(x) = x2.
Similarly, in the preceding three examples, if</> is given by the final formula, we shall
f and g in,
</>=Jw.
f(u)
= ua,
g(x)
x.
=sin
Then
f'(u)
g'(x) =
= 3u2,
Therefore
f (g(x))
=sin3
f'(g(x)) =
x,
3 sin2
x,
cos
x.
f'(g(x))g'(x)
.3 sin2
x cos x.
Similarly,
g(f(u))
Df (g(x))
= sin u3,
sin3
x = f'(g(x))g'(x) =
g'(f(u))
= cos
u3,
3 sin2 x cos x.
which will turn out to be the derivative of cos u3. (Here cos u3 is the cosine of
not the cube of cos
u.)
f(u)
f(g(x))
u,
=cos
g(x) = .JX:,
x,
f'(g(x))g'(x)
g'(x) =
f'(g(x))
=(-sin
= -sin
-jx)
lr,
2"x
.J-;,
11_.
2, x
u3,
158
4.5
Theorem 1.
That is, if
limf(u) =f(u0),
u-+uo
then
Jim J(g(x))
= J(g(x0)).
x-+xo
R:o!
x0
g(x)
=>
R:o!
g(x0)
=>
f(g(x)),::::; f(g(xo)).
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.
4.
<f>(x) =sin 2x
5. <f>(x)
6. <f>(x) =cos 2x
7.
<f>(x) = Vl - x2
8.
10. <f>(x)
fin:r
tan 2x
<f>(x) =sin6 x
(t2 + 1) dt
9.
11. <f>(x) =
<f>(x) = f't
fos.r
(t2 + 1) dt
f'
(t2 + 1) dt
can be expressed without the use of integral signs; f can be calculated as a polynomial.)
12. a) Find Jim
U-4>1lo
sin u
sin u0
Ii
Uo
sin x2 - sin xg
b) Find Jim------.:.
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
<f>(x)
-=---
x2 - xfi
as the composition /(g) of two functions f andg, and then apply Theorem 1.)
13.
Find Jim
14.
G iv en
sin x3 - sin x8
----X
Xo
x-+xo
sin <X:
4.6
Given
17.
<f>(x)
fin
159
I + ,2 dt.
On the basis of the theory that you know so far, you are in no position to calculate
(u)
v'l+f2 dt.
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
<f>'(x)
some
[Hint:
[Hint: If you can figure out what the geometric meaning of this limit is, it will then be
easy to find its numerical value.]
19.
cos x + 1
Find lim ---x-;; x - 1T
21.
Find Jim
4.6
THE CHAIN
.x"/4
[Same hint.]
sin 2x - I
X
14
1T
tanx - I
20.
Find Jim
22.
secx - 1
Find Jim ---
:r-"14
:ro
14
1T
.X
RULE
You may have observed, in the preceding problem set, that the formula
Df(g)
held in a number of cases.
f'(g)g'
For example, if
f(u)
then
f' (u)
u",
nu"-1;
and
Df(g)
Dg"
ng"-1g'
f'(g)g'.
Similarly, if
f(u)
then
f' ( ll)
Df(g)
D,/g
Jli,
:'
----!,:
2,,.
ll
-1=
2 Jg
g'
f'(g)g'.
160
4.6
'
The formula
g'(x) = 2,
g(x) = 2x,
2 cos 2x.
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
This is a composite function
with
.\ ( ? \ '
cp(x) = sin (3x + 1).
cp(x) = f(g(x)),
f(u)= sin u,
g(x) = 3 x + 1,
f'(u) = cos u,
g'(x )= 3.
cp(x) = sin (k + x) .
4.6
Example
3.
Consider
cp(x)
f,kx
1
Here
cp(x)= f(g(x)),
f '(u) = 1:'
u
g(x)
Dcp
(k, x
dt
f(u) =
1
-
>
0).
lkx
1
1
-
dt
(u > 0),
dt
g'(x)
f'(g(x))= : '
x
kx,
"'
J
1
1
-
161
k,
_!.
x
dt.
D sing=
whatever g may be.
(cosg)g',
We can then apply the formula in cases where g' itself needs to
x)D sin x
(cos sin x) cos x.
=
Here sin sin x is the sine of the sine of
x,
Therefore
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.
l]
2(x3 + 1) 3x 2,
162
4.6
rp(x)
x0
f(g(x)),
we have
rp'(xo)
f'(g(xo))g'(xo).
b)
f has
has a derivative
a
lim
g(x)
Therefore
g(x0).
[g(x0
lim
t.x->O
x) - g(x0)]
= 0.
By definition,
rp'(x0)
rp(x) - rp(xo )
X - Xo
rp(x0 + x) - rp(x0)
= lim
t.x->O
x
. f( g(x o + x)) - f(g(xo))
= lim
x->x0
=hm
t.x->O
Let
u
so that
= g(x0 +
x) - g(x0)
g(x0
Then
"''
(Xo)
x)
1.
lm
f(uo
u0
t.x->O
g(x0
+
x) - u0,
u.
u) - f(u0)
.
x
f = f (u0
Liu) - f(u0)
f = f'(u0) u
Jim
E(u)
t.u->O
in place of
x,
E(u ) u,
= E(O) = 0.
4.6
163
Therefore
D.f
D.x
D.u
D.x
f'(u0)
f'(g(xo))
E(D.u)
g(xo
D.u
D.x
D.x) - g(xo)
D.x
E(D.u) g
(x0
x) - g(x0)
.
D.x
It is now easy to see what the limit is. By definition of g'(x0), we have
.
I1m
g(x0
<ix-o
As
D.x ---+
0,
D.u---+
D.x) - g(x0)
D.x
g'(x0)
0.
D.u,
This gives
D.f
ef/(x0) = Jim
<ix-o D.x
-
E(D.u)
= 0.
.
=
f'(g(x0))g'(x0)
+ 0
g'(x0).
We therefore have:
Theorem 1.
Let f and
g be
functions. Then
= f'(g)g',
Df(g)
at every point
x0
x0 and
(b)
f is
differentiable at g(x0).
x0
is differentiable at
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>(x)
sin (x2 +
f'(g)g'.
For example,
1),
f(u)
f'(u)
g'(x)
sin
u,
=COS!I,
=
2x,
g(x)
f'(g(x))
'
(x)
x2
cos (x2 +
1,
1),
[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
1)]2x
eliminate the brackets we have to change the order of the factors, as in the last expression
4.6
164
(?)
'(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.
1.
7.
12.
16.
20.
24.
27.
30.
34.
37.
2.
sinx2
8.
sin(x3 + x)
tanx -
sec 2x
sec
vx2
a) sin
(vx)2
b) tanx2
a) tan2 x
3-
21.
b) \sinx
(I
,i:x
[x[2
cos4x - sin4x
cos
vx2
sin sinx
cos:c
f,k:J; -1 dt
5.
tan(t2 +
10.
,1 cosx
15.
t3Jt
f,x 1
- dt
t
1)
I I.
a) sec2 x
18.
22.
6.
19.
x
tan a) '\!tanx
23.
-1
2
tan t2 +
x
tan --
b) secx2
cos 2x
26.
29.
32.
sin cosx
35. L
+ sinx)
cos3x
cos
14.
17.
25.
28.
31.
,,--
4.
9.
cos cosx
sin
cosx3
sinx3 +x
13.
3.
sin2 x
cos2
- sin2
sinx
+ cosx
b) tan ,;:;.:
33.
36.
ix'
f,k - dt.
0
cos f
dt
tan sinx
Let k be any positive number; and for each positive number x, let
(x)
Find the simplest possible formula for '(x). Then do the same, for the functions (x)
38.
f,x' - dt
1
(x >
.t
39.
0)
lx3dt-3 xdt
J t
t
sin 1
- dt (0 < <
44. J
t
45. For each x > 0, let
41.
J,"'' 1- dt - 2 Jx 1- dt
42. J\;dt
t
TT
f(x)
/(ab)
x3 1
40. l - dt
t
V;; 1
1 "' 1
43. J - dt - - dt
f
2J t
1
=f,x dt.
t
1
[Hint:
a and b,
f(a) + f (b).
we have
When we try to attack this problem by the methods of calculus, the obvious
trouble is that the problem does not appear to involve any functions.
first step should be to
Therefore our
Invertible Functions.
4.7
46.
165
ef>'(x).
*47. Given
D cos = -sin,
D sin =cos,
sin 0
cos 0
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)
for every k and
x. [Hint:
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.
find further confirmation of this.
*48. Let/be a function, defined
(a)
f"
-f ,
(b)
x for every x.
(a) g"
Show that g(x) = cos
4.7
-g,
Later we shall
for every
f (0)
x,
=0,
(c)
f' (0)
I.
such that
(b) g(O)
I,
x.
A function f is called invertible if its graph intersects every horizontal line in at most
x3
x2
y=f(x)=x3.
is not.
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,
=
4.7
166
1-1(x)
ifJ(y)
x.
f(x)
a
x .
a
x .
(I)
y = fl;: .
Thus
J-l (x) = -{Y;:,
as we would expect: the inverse of cubing is the extraction of cube roots.
y
Invertible Functions.
4.7
Theorem 1.
167
J(!-1(x)) = x,
for every
x.
Then/(y)
x,
We can use this idea to calculate the derivatives of inverse functions, assuming
that the inverse function has a derivative.
Example
Thus
1.
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 =
D\o/x = --1=
3\o/x2
1,
(x - 0).
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
r1<x>=vx
1)
2)
3)
(::Jx)q = x,
q(::Jx)q-l D/x =
-
Dx = ---==
q::}xq-1
1,
(x
>
0).
168
4.7
When we use this method, the equations that we write have the following general
form:
j(J-1(x))
1
f'(f- (x))Df-1(x)
1)
2)
3)
Df-1(x)
x,
1,
f'(r (x))
(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),
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
=
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.
D r-1( x)
1
f'(r1Cx))'
I(x + 27T)
I(x),
for every x for which the trigonometric function l(x) is defined at all. Therefore
Invertible Functions.
4.7
169
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
-1
y=f(x)=sinx
If we restrict
to the interval
[-7T/2, 7r/2],
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
() on the interval
sine.
[ -7T/2, 7r/2];
x.
170
4.7
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
(cos Sin-1
sin Sin-1
x,
x)D
Sin-1
1,
Sin-1
cos Sin-1
We want to simplify the expression cos Sin-1 x on the right, and, while we are at it,
Since
cos2 u + sin2 u
1,
For
u
.JI
.JI
=
sin2 u,
(1)
cos2 u.
(2)
Sin-1 x,
this gives
and so from
(I) we get
sin u
cos Sin-1 x =
Similarly, for
u
.J 1
x2
(3)
Cos-1 x
we have
cos
and so from
(2)
=cos Cos-1 x =x
,
we get
sin Cos-1 x
.JI
x2.
(4) are correct, but they are not good enough for
signs can be omitted, and the formulas still hold:
(4)
our purposes.
Invertible Functions.
4.7
171
Theorem 3.
cos Sin-1 x
==
sin Cos-1 x
.J 1
.J 1
- x2,
-
x2
Therefore, in
(3),
Similarly,
0 Cos-1 x 7T.
On this interval, the sine is 0.
applies.
We now substitute
D Sin-1 x.
Theorem 4.
.J l
- x2
This gives:
D Sin-1
,--
= 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
(-l<x<l).
x2
like this:
y
To get an invertible function Tan, we take the portion of the graph that lies between
-7T/2 and
x =
7T/2.
172
Theorem 6.
4.7
x in J, thenfis invertible.
f(a)
for some
and b in /.
f(b),
Tan' x
sec2
By MVT, this
1T/2).
On this interval,
0.
Thus
y
y
Tanx
7r
----------------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:
Invertible Functions.
4.7
173
y
I
I
I
I
I
I
3.,,.
-2
_,.
x =
:
I
'(\
I/cos
I
I
I
I
I
= secx
: 3.,,.
12
1r
:(\
-1
I
I
I
I
I
I
I
I
I
I
IY
I
I
I
I
I,,.
1
2
I
I
I
I
I
I
I
I
I
I
I
wherever cos
0.)
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
<
general agreement on what else we ought to use. To be safe, we define Sec x only for
<
y
Jy=Sec x
I
I
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) ?)
Sec-1
secy
(5)
and
7T
(6)
O:Sy
< -.
2
We have
Sec Sec-1
(7)
x = x,
and
Sec' u
for every
from
0 to TT/2.
Sec u Tan
u,
x =
1,
4.7
174
and
x=
(8)
1.
Therefore
1
x = -----
(9)
x Tan Sec-1 x
We now need a formula for Tan Sec-1 x, analogous to the formulas for sin Cos-1
and cos Sin-1 x. We know that
1 + tan2 u = sec2 u
D Sec-1
for every
u.
Therefore
tan
For
u=
) sec2
I.
Since
u=
x=
) sec2 Sec-1
x -
Tan Sec-1
0,
x = )x2
Therefore
1,
and we have:
Theorem 8.
D Sec-1
x = 1/x)x2
I.
D Sin-1
1 ,
x = --) 1 x2
D Cos -1x
D Tan-1
x=
1
1 +
x2
D Sec-1
=
)1 x2
x=
_1
x x2
_
_
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.
4.7
Invertible Functions.
175
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
9. Sin-1
1. Sin-1 (x
10. Cos-1
1)
Vl
- x2
13. Sec-1 x2
I
16.
Sec-1
\1 I
15. Tan-1
1
17. Cos-1
18. Sin-1x
x2
-
v1 - x2
J)
26. Tan-1 (I - x)
27. Sin-1 x2
See1 x
Cos-1 - ,
x
Sin-1 x + Cos-1 x
for every x on the interval [ I, 1 ].
uniqueness theorem of Section 3.8.)
-
30. Find
f -1 - ? dt.
+ rI
Sketch.
11'
2'
33. Find
-1
32. Find
f/2
0
VI
t2
dt.
'/
31. Find
2dt.
l-+ t
1v2 r
I
v1
t2
Sketch.
dt.
35. Given
0 x l,
find a formula for f-1(x). Then explain how your answer might have been predicted
without a calculation.
36. Find
(2
1
dt.
J 21v3 iVi2 1
11v3 t
dt.
1 --v t2 + 1
-
38. Find
37. Find
J xVx2
1
_
_
_
_
dx.
176
39.
4.8
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,
to Sin on
Tr/2, Tr/2]
or to Sec on
and so we
[O, Tr/2),
and 0.
Tr/2, Tr/2,
Tr/2),
following:
If f is differentiable on an interval I, and f'(x)
Theorem.
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).
F'(x)
G'(x)
the interval I.
Theorem
let
(?).
Let F and
be a point of I.
of I, then (iii)
F(x)
If (i)
=
3.8,
4.8
11
!(x) dx,
F(x)
Then F'(x)
f'f
(t) dt.
Then F and G have the same derivative f; and by adding a constant to G, we get
function, say H, such that H'
G'
by the uniqueness theorem that F(x)
=
f and H(a)
0.
Since F(a)
f(t) dt
H(b).
If f is continuous, and
G'(x)
(a x b),
f(x)
then
!(x) dx
G(b)
G(a).
f
a
0, we know
Simpsons' Rule.
4.8
Proof
For each
x,
The Computation of
TC
177
let
F(x)
=if(t) dt.
Then
F'(x)
f(x),
F(a)
and
0.
Let
H(x)
G(x) - G(a).
Then
H'(x)
F(x)
Therefore
H(x) for
G'(x)
every
x,
ff(t) dt
and so
H(b)
H(a)
and
f(x),
F(b)
=
G(b)
H(b).
-
0.
Therefore
G(a).
The proof reproduces the procedure that we have been using all along.
first
G that we try,
with
G'
G is
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'(x)
G such that
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.
val
[O, 1]
into
178
i=
xi=
4.8
ai =
Yi=
0.1
0.1
0.999
0.0999
0.2
0.992
0.0992
0.3
0.973
0.0973
0.4
0.936
0.0936
0.5
0.875
0.0875
0.6
0.784
0.0784
0.7
0.657
0.0657
0.8
0.488
0.0488
0.9
0.271
0.0271
0.7975
A
The approximation A
f(1 - x3) dx
0.7975
A1.
f(1 - x3) dx
We might also have used
0.7500.
inscribed rectangles.
0.7975 - 0.1000
A2
==
0.6975
A2
But their
average
is
considerably better.
Aa
The sum
A3 has a
t(A1
A2)
==
0.7475
0.7500.
zoids.
y
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.
A3
is
Simpson's Rule.
4.8
The Computation of
7t
179
I:::::!
another method which gives better results without any extra work. This method is
= Ax2 +
Bx + C
tion
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)
then
:i x3
3
G'
and so
fg(x)dx
!!. x 2
2
Cx '
g,
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]
[a;, ai+1]; we then use Simpson's rule on each
We shall now develop a shortcut formula for Simpson's rule, in a special case.
Theorem 2.
Let
Jk g(x)dx
-k
where Yo =
g(-k), y1
g(O), and y2
(Yo
3
g(k).
4y1 + Y2),
180
4.8
Before proving that this formula is true, let us first check it, in a simple case, to
2k = (1 + 4 + 1),
3
which is correct.
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 +
Let
G(x)
so that
A
B
- x3 + x2 + Cx'
3
2
-
G' = g. Then
fkg(x) dx
G(k) - G(-k)
y 0,
C.
iAk3 + 2Ck.
We need to express
To find
A, we use
Yo + y2
Y2
C,
Ak2 + Bk +
C,
Yo + y2 = 2Ak2 + 2Ji.
2A'k2 + 2C,
A:
A and
f g(x) dx
k
-k
C now give
= iAk3 + 2Ck
=
3 ( Yo
k
=
+ Y2),
4y1
f(x)
Here we have
C in terms of
g(O) = y1.
2
Yo = Ak - Bk +
A and
k =
1,
x + 2
--
Yo=
1,
-1 x 1.
Simpson's Rule.
4.8
The Computation of
7t
181
y
3
2
-- 21
I
I
I
-
- 1
f(x)
-+
x+2
--''--
'--2 --X
Jl-1
t(l
x
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.
-1
Let us now try
f(x) =
Here we have
k = 1,
1
,
1 + x2
Yo= i,
1 dx Ht
J-11
x2
+
---
Since
-1 x 1.
--
Y1 =
+ 4 + t)
-1 x =
D Tan
f1 --dx
=
-11 + x2
-1 1
Tan
. 1.57.
= !!...
2
1,
t 1.67.
1
,
1 + x2
---
Tan-1
( l)
-
7T
4.8
182
0.10,
2,
Let
ia+2k
g(x) dx
a
where
Yo = g(a),
k
=
Yi= g(a
(y0 + 4yi
Y2
k),
J2),
g(a
2k).
(a
k units
it is still a parabola (or a line); the integral does not change, and neither do the
numbers
k, y0, Ji,
2n
[a, b].
b-a
.
k=
2n
The division points are
x0,
Xi,
, x2n,
Yt
Yo
On the interval
2.
[a, b]
Simpson's Rule.
4.8
where
Yi =/(xi);
for each i.
On the interval
ia+4kf(x) dx
R:i
a+2k
[x2, x4]
The Computation of
[a + 2k, a + 4k]
7t
183
we get
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
better approximation of
11 d x
.
-1 x + 2
xi=
Yi=
0
1
2
3
4
5
6
7
8
9
10
-1.0
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1.0
1.0000
0.8333
0.7143
0.6250
0.5555
0.5000
0.4545
0.4167
0.3846
0.3571
0.3333
1
4
2
4
2
4
2
4
2
4
1
k = 0.2, to get a
1.0000
3.3332
1.4286
2.5000
1.1110
2.0000
0.9090
1.6668
0.7692
1.4285
0.3333
16.4795
This gives
1
f
J_1 x + 2
R:i
0 2
3
(16.4795)
R:i
1.0986.
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
11--dx = Tan-1 1
2
1 + x
o
Therefore
- Tan-1
7T .
0 =4
The interesting
184
4.8
Tr.
This is
Problem 1 below.
f(x)
with k = !.
(0 x 1),
1 + x2
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),
f(x)
x3 + x2 - 17
(-100 x 100),
Then calculate the integral exactly, and compute the error in the
with k = 100.
approximation.
f(x)
with k
5.
= x3
- 2x +
3,
(-1 x 1),
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
f(x)
on the interval
[O, 1],
=1
- x3,
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
y0, g(O) =Yi. and g(k)
y2 That is, write
quadratic function g such that g( -k)
=
g(x)
in which the coefficients
Yi andy2
A, B,
Ax2 +Bx +
C,
4.9
185
*9. Does the theorem that you proved in Problem 6 hold only on intervals of the type
As a check,
11'.
3.14159265,
11' =
In Appendix F, at the end of the book, you will find a theorem which enables us, under
4.9
For the case in which the exponents are positive integers, exponentials are part of
elementary algebra. We begin with:
Definition.
n,
(to
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
If
n =
:;if 0 we define
3 we have
x-
For
-n
(B)
= xmn.
1
--
x-C-3l
1
x3
:;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.
(a'IJfo)q
aP.
That is, aPfa must be the qth root of aP. Hence the following:
186
4.9
a'Pfa= ,:;a'P.
Here we cannot allow the case a < 0. For a =
( -1)1/3
( -1 )2/6
-1
( -1)2
-'-
we would get
1,
1 =
-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.
It is a fact that for a > 0, and x and y rational, the following laws hold:
2)
(A)
a xv,
(B)
aO = 1.
3)
(C)
The rational numbers on the x-axis do not fill up the x-axis, because every interval,
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.
y
a>l
\
\
\
\
'
'
'
'
I
I
I
I
I
I
f(x) =a",
............
.......... __
x in Q
a<l
-x
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)
ax> 0,
4)
I, we have f(x)
l'"
by definition.
(A),
>-
a'" (a
1).
(B), and
(C).
For
1,f is invertible.
That is,
av= x,
4.9
187
Therefore the domain of its inverse includes all positive numbers, and we have a
function
log,,:
(0,
co
__,..
R.
loga X + logay,
= X
loga 1 =
loga
(A')
(b > 0, b 1),
(B')
0.
(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=f(x)=ax, a>l
1.
loga (x0 + 1:1x) - loga x0
.
= lm
Jim - loga
t:.x-o 1:1x
1:1x
t:.x-o
x0 + 1:1x
x0
lim loga
t:.x-o
t:.x
!:1x 1/
+ -
= lim - loga
t:.x-o
Let
x0
Ax
h =
.
Xo
x0
f:1x "'o-'
+ -
) t:.x]
x0
188
Since
x0
4.9
This gives
_!_ lim
Xo h-+O
(1 + h)11h.
loga
If loga has a derivative, then the limit on the right-hand side exists, and conversely.
Suppose that
(1 + h)1'"
e.
Thus
(1 + h)11".
lim
h-+O
Suppose that loga is continuous, so that the limit of the logarithm is the logarithm of
the limit. Then
log
Thus
x0
Since
e1
e,
-x1
loga
l. loga e.
Xo
e.
we have
log.
and so for a =
e our differentiation
1;
D log.
6)
a
1,
f(x)
loga
x,
so that
J-1(x)
=a'".
J(j-1(x)) = x
thus takes the form
loga a'" =
x.
Since
Du loga
-1
loga
e,
Cx
loga
Da"
Therefore
Da" =
1
-e
loga
a'".
1.
1.
(For
4.9
In particular, for a =
189
we have
De"'= e"'.
A final simplification: we assert that
1
Proof.
-loga e
Let
Then
x = loga
e,
a"'=
e,
=log. a.
y =loge a.
(a"') 11
This holds when xy
= e11,
and
'"11
a = a.
1.
on the same value twice. Therefore the equation can hold only when xy = I. There
fore
-=
x
y,
Da'" =
'"
a log. a.
This is better, not just because it avoids a fraction, but also because
two bases for which tables of logarithms are published.
is one of the
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.
a)
c)
a0 =
e)
0)
1,
0, a
g)
i)
loga 1 = 0,
k)
loga a x= x,
1)
b)
(ax) v =
d)
a'" > 0
f)
e = l im1i-o (1 + h)1f1i.
h)
log0
j)
1)
bx
axv,
for every x,
= x loga b
D loga x =
(b >
l/(x log. a) ,
aloUa x = x.
x
e"' cos x
3.
5.
8.
[loge x]2
9. log. xsoo
2.
4.
7. log. x2
xe2"'
iex(sin x
xex
e'" sin x
1. x loge
6.
+cos x)
0),
190
4.9
10. esin a:
13. [log. e]"'
16. e "'
19. log. (1 - x)
22. el-X
!h
11. 10"'
14. elog, "'
2
17. xe"'
1 -x
1)
24. e"'
0,
log. x
30. log. (x
+ cot x)
v x2
1)
l
"'
i
-dt.
t
logb
+ tan x)
1
=
1-oga b
1, then
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 "'
theorem.
Theorem.
If g ' (x)
g(x), on
g(x)
Prove this.
f'(x)
f(O)
That is, show that
f(x)
( - oo <x < oo ),
1.
f'(x)
f(O)
That is, show that
e-"'
(1)
(2)
-f(x)
(2)
1.
(1)
4.10
4.10
191
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
If you review the formulas of the preceding section, you will see that after con
X = -
f,"' t dt.
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,
at
x =
1 (namely,
same function.
0);
ties, and then define all our other functions in terms of it.
we shall investigate
f: (1/t) dt
(Here In is suggested by
"' dt
Soon we shall show that every real number y is equal to In x for some x. For this
192
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
Proof.
D In x =
I/x.
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.
Proof.
kx
I/x.
Theorem 5.
Proof.
1
-
kx
1
k =-
The trouble with this theorem is that it does not appear to involve any fu1
every k, x > 0,
Inkx =Ink + In x.
----
Let
f(x) =Inkx,
g(x)
=Ink + In x.
Then
f'(x)
= _!
g'(x),
j(l)
=Ink,
and
g(l)
We now want to show that the graph of In looks approximately like the drawi
above.
4.10
know.
193
Other things suggested by the figure are conveyed by some of the following
theorems.
Theorem 6. In is invertible.
Proof
every
x.
and
x = 1/x;
every
Therefore In'
x.
x -:F- 0 for
xn = n In x.
Proof.
integer
n,
xn+i =
=
In
In
(x xn) = In x + In xn
x + n In x = (n + 1) In x.
n = 1.
n + 1.
xn
upper bound
f (x) M
n In x for every x,
for a function
for every
f if
x.
Proof
In Theorem 7, take
2.
Then
In 2 n
for every
n.
= n In 2 ,
have an upper bound; no number M is greater than or equal to all of the numbers
n In 2 ,
because
ln 2 > M
whenever
n >
M
-
In 2
194
Theorem 9. In
(1/x)
4.10
-In x.
Proof
In
+ In
In
e x)
In 1
0;
-n In 2.
By Theorems 7 and 9.
Proof
lower bound
sin x
unbounded below.
bounded below.
for every
x.)
x.
If
If no such number m
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.
Every real number is a value of the function In. That is, every number y
Theorem 12.
Proof
y lies
that y
R
Since In is unbounded both above and below, it follows that every number
between
=
twoYalues of In.
In x for se x.
If In
x1
< y < In
x2,
(-oo, oo).
Since In
ln
1.
e",
Theorem 13.
x.
x.
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
> 0
for every
x.
4.10
195
/
/
/
/
/
/
/
/
/
,,
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.
0.
1.
Proof.
(exp k)(ex
) =
x)
x.
x)
k +
x,
x)]
In exp k + In exp x
k +
x.
Since In never takes on the same value twice, the theorem follows.
Proof.
exp.
for every
x)
exp'
e"'
a)
In
f,.,
1
dt
-
Since In' u
-x
exp
exp
x = 1,
exp'
x =
x = e"'.
>
0),
e,
and
We shall do this in
(x
exp x,
Definitions
x.
x = 1,
Therefore exp'
x.
196
4.10
Laws for In
c) In 1
d) In xn
0,
e) In kx
Laws for
exp
g) exp 0
k) In exp x
f)
1,
exp x > 0
i)
In k + In x
In' x
In x
(x > 0),
(x > 0).
I/x
h) (exp k)(exp x)
j) exp In x
x
for every x,
x
= n
I)
for every x,
exp'
exp (k + x),
(x > 0),
exp.
4.9.
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;
should also.
1.
ln2
5. exp
x2
9.
exp sin
13.
In sin x
2. In In
3.
1)
4.
In
6.
x]2
7. exp (2 In
x)
8.
In (exp
(x In x)
1 2.
e"' log,
16.
(sin
[exp
10.
sin (exp
14.
sin In
x)
In
(x2
1 1.
exp
15.
X"'
(X >
0)
x2
+ 1
x2)
x)sin x
(sin x
>
0)
lx
1
dt
f(x)
Jx
dt
?
vt
18.
g(x)
19.
h(x)
.
lx
1
Given
find
l"'2
1
dt
--
vr
dt
2?
t
(0 <
<
Given
f(x)
findf'(x).
oo),
rsinx
Jo
vl+t2 dt,
h;
4.11
1
2 . Given
f(x)
fanx
0
find f'(x).
22. Given
g(x)
findg'(x).
23. Given
h(x)
=exp
find h'(x).
24. Find
lim
v1 +
The Existence of
197
t2dt,
fxdtt'
i
(f' dt),
sinx + 1
x3"/2 X-37TI2
(By far the easiest way to solve this problem is to think of a geometric meaning for it.)
25. Find
28\ Find
Jim
tanx-1
tanx + 1
x-rr/4 X +
I4 .
1T
--
--
X-+1if4 X-7T14 .
Jim
In x2
lnx
27. Find Jim
26. Find Jim
.
1.
x-..1X x1X - 1
exp ( 2x)- l
29. Find Jim
xo
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,
f (x)
lnxx-1.
< 1, the same inequality holds.
x-1, find a formula for 1-1(x), and sketch both functions on the
+ 1, for every x.
f (h)
h 0.
(1 + h)1'"
as
(1 + h)lfh,
1/h
The right definition is not hard to find. We know that if n is a positive integer, then
In
n
a =
n In a.
4.11
198
Therefore
a" = exp
(n ln a) .
"
a = exp (x ln a) .
We take this last equation as our definition of the exponential function a".
Definition. For
> 0,
a" =
Thus:
exp (x ln a) .
This gives:
Theorem 1. ln
a" = x ln
a.
(1 + h)1f1i
(1 +
f(x) =
Then
lnf(x)
1
= -
ln
approaches a limit, as
h ---+ 0.
Let
x)1 x.
(1 +
x).
ln f
, (x) =
!
x
ln
ln
(l +
(1 +
x)
x) - In
x
This last fraction is the fraction whose limit is ln' 1, by definition of the derivative.
Therefore
lim lnf(x)
Ax-o
and
ln'
1 = t = l,
Ax-o
Ax-o
explim lnf(x)
Ax-o
= exp 1 = ln -1 1.
Replacing x by
again, we have
lim
h-+O
(1 + h)111i =
exp 1 = ln-1
1.
Definition.
e = lim1io
(1 + h)1fh.
And we know:
Theorem
2.
e = exp 1 = ln-1 I.
4.11
The Existence of
199
That is,
le -1 dt= 1.
1
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
n.
The series on the right is infinite, but the terms diminish so rapidly as
increases
that we get good numerical approximations by using the first few terms.
We expected exp
by definition of
Theorem 3.
x.
y = loga x
Since
<=>
aY = x.
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),
4.11
200
by definition. Therefore
In
v
a
In exp
(y In a) ,
and
In
Since
a"'
av= y In a.
= x.
a=
In
x.
This gives:
Theorem 5.
For every
> 0,
Iogax =
x
-
In
In
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
Ioga
> 0, 1,
xy =
Ioga b"' =
Ioga
+ Ioga
(x, y
>
(g)
0),
loga b,
(h)
Ioga 1 = 0,
1
Ioga/ X =
log,
Ioga
a
(i)
a"' = x,
log."'
= x,
(j)
,
x,
for every
for
(k)
(I)
> 0.
Here the formula designations are those of the summary at the end of Section 4.9.
The proofs are as follows.
Proof
Joga =
For
a = e,
the first three formulas are known to hold, because in this case
a,
(1)
x=
D loga
x=
1
x
- - -In a
x In a
In
a"' =
exp
(.JC
In
a) =
"'
Ina
.
"'
"'
4.11
The Existence of
201
0,
(a)
(a"')v =a"'11,
(b)
aO = 1,
(c)
for every
a"' > 0
x,
(d)
(e)
Da" =a"'Ina.
The proofs are as follows.
a)
We have
In (a"
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
c)
0.
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.
n,
of Section
4.10.
Therefore
202
4.11
"'
a111q
a"
The proof is as follows.
( )
exp (x In a) = exp
In a .
( a) .
In
exp
Let
Then
and
q lny = p ln a.
Therefore
In y =!!.in a,
and
y =
exp
( )
In a ,
Dxk
holds true in certain cases.
kx"-1
x >
0, it says that
Dx112
k was a positive
k was a n7gative integer. For k = t,
/
ix112-1
2
ix-112
2
1_
__
2.Jx,
Proof
By definition,
x"
Therefore
Dx"
= exp
(k In x).
kxk-1
In this section we have presented no new results, except for Theorem 8; we have
4.ll
The Existence of
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
smhx
e"' - e-"'
= --2
coshx
e"' + e-"'
= ---
tanhx
co thx
sechx
cschx
e"' -
-"'
e
e"' + e-"'
e"' + e-"'
e"' - e-"'
2
e"' + e-"'
2
e"' - e-"'
sinhx .
= -coshx
coshx
= -- .
sinhx
1
- --
cosh x
1
-sinh x
Identities
sinh
( -x,) = -sinhx.
cosh (-x)
tanh
(1)
= coshx.
(2)
( - x) = -tanhx.
cosh2 x- sinh2 x
(3)
=:===,,(.
(4)
1 - tanh2x = sech2 x.
(5)
coth2 x-
(6)
1 = csch2 x.
sinh (x
(7)
cosh (x
(8)
tanh (x
+ y) =
sinh 2x
= 2 sinhx coshx.
(10)
cosh 2x
= cosh2 x + sinh2 x.
(11)
"'
tanhx + tanh
1
+ tanhx tanh y
= coshx + sinh x.
e-"
cosh x - sinhx.
sinh
y.
(9)
(12)
(13)
204
4.11
Derivatives
sinh' x = cosh x.
(14)
cosh' x = sinh x.
(15)
tanh' x = sech2 x.
(16)
coth' x = -csch2 x.
(17)
(18)
csch' x =
(19)
csch x coth x.
1. (12)
2. (13)
3. (1)
4. (2)
5. (3)
6. (14)
7. (15)
8. (16)
9. (17)
10. (18)
11. (19)
F(x)
cosh2 x
sinh2 x.
Show that A + B
0.
18. (7)
20. (9)
19. (8)
21. (10)
22. (11)
x > 0
=>
sinh x > 0.
x < 0
=>
sinh x < 0.
) cosh is increasing.
oo
cosh x 1
for every 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.
4.11
31.
32.
Find D sinh-1 x.
33.
The Existence of
205
35.
Let
(0 x).
Cosh x = cosh x
(0 x
invertible.
7T .
Sillh X
37.
Show that
38.
Find D Cosh-1 x.
39.
Firid D Cosh-1 x2
{v
cosh2 x
for x
1,
for x <
-Vcosh2x - 1,
sinh Cosh-1 x =
0,
x2
0.
1.
42.
Find D tanh-1 x.
43.
Solve for x:
e2"'
e"'
- 6 =
0,
e"'
e"'
+ y - .6y2e-"'
\
I
2 - 35e-"'
0.
0.
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)
i(e11 - e-i-).
(2)
Here we have reflected the graph across the line y = x, by interchanging x and y in
Eq.
Then
= (
"
sinh-1 x =
47.
The Variation of
5
5.1
Continuous Functions
x < x'
=>
f(x) <f(x').
x < x'
=>
Similarly, f is decreasing if
y
y
Here
and
x'
For example,
f(x)
x2
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
This function is
[x0, x1]
and
f
J
207
is increasing on
I
I
I
I
I
I
I
: I
I
--+---x
X2
X
X1 X3
Similarly,f is decreasing on
[x3, x4].
We recall that an
If we had
a< b,
(?)a,b in l,
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
because such an
f'(x)
<
0 for some
between
and
b.
This is impossible,
(?) c,
din/,
c< d,
as on the right, then the chord would be horizontal, and we would have f' (x)= 0
at an interior point
of I.
f ( x)=
xz,
I=
f'(x) = 2x,
[O, co).
Consider
5.1
208
hairs; if we required that/'(x) be >O everywhere in I, then the theorem would not
cos x, I=
apply in the simple case f(x)
x2, I= [O, oo), or to the case f(x)
=
[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
Here
sign.
f is
f(x)
Here
f'(x)
2x
x2
f'(t)
- 1,
- x
0,
(0
2).
and
f'(x)
>
on
( t, 2].
In the left-hand figure below, we have used this information, and have plotted
f(t)
and/(2).
Obviously /(1)
0,
y
2
-1
-1
(Proof
0 at
I. Therefore f is decreasing on
I.)
5.1
(0, !],
209
x2 - x, on the interval
2x
<
for
< x < t.
x3 + 2x2
3x
- 4,
-2
2.
This is not a put-up job; it is a "real-life" problem, and nothing is going to come out
even.
3x2 + 4x -
0.
Now
3,
so that
f'(x) = 0
Since
J13
3.6,
when
-2
X=
JTI
3
=
are
-6
Since the graph off' is a parabola opening upward, it must look like the drawing
on the left above.
Thus
f'(x) >
when
x < x2,
f'(x) <
when
f'(x) >
when
x > x1.
210
5.1
2,
::::::; -
f(x2)::::::; 2.1,
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
0. Similarly, f'(x) > 0 for -2 < x < x2, because on the interval
value f'(x1)
[-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].
y
__/L_
I
{x I x1 < x x2}.
0.
Iff is decreasing on [x1, x2], then f(x) < f(x1) for every x on (x1, x2].
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
1.
f(x)
2. f(x)
3.
f(x)
4.
f(x)
5.
f(x)
6.
f(x)
7.
f(x)
8. f(x)
9.
f(x)
sin x,
-1;:;:;
Sin-1 x,
1
- --2
1
+x
-2;:;:; x;:;:; 2
'
x3 - 3x,
'
x 2
x ;:;:; 3
-1 ;:;:;
x;:;:; 2
1,
x3 + 6x2 + 9x + 3,
ew
2;:;:;
-1;:;:;
x3 + 3x2 - 2,
x;:;:; 1
-2;:;:; x;:;:; 2
'
-1 --2
+x
0;:;:; x ;:;:; 1
1 0. f(x) = x In x,
1 ;:;:; x;:;:; 5
11 .
f(x) =cos x,
12. f(x)
13.
f(x)
14.
f(x)
15.
f(x)
6
1 .
f(x)
7
1 .
f(x)
18. f(x)
19.
f(x)
211
0 ;:;:; x ;:;:;
sin 2x
-x
+2x2
1
=
-1 +x 4
-2;:;:; x;:;:; 2
-1;:;:;
xe-w
---4
1 +x
1T
x;:;:; 2
-1
;;;; x ;;;; 1
-1
;;;; x ;:;:; 1
x cos x - sin x
x/2 + sin x,
e"'
2x,
0 ;;=;x;:;:; h
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).
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
=
212
5.2
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:
=
(fV
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
X1-0XJ X1+0
I
I
I
I
I
I
I
I
I
I
X1 X1+0
m
I
I
X1 -o
I
I
I
I
Xj
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
I
I
Xi-5
I
I
I
I
I
Xi xi+5
!\):
0
I
I
I
I
I
I
X1 xi+5
213
I
I
I
I
I
I
I
I
X1-5 Xi
ILMax.
There are simple conditions under which a function has an ILMax or an ILMin
at a given point.
Theorem
Theorem
If f' > 0 on
1.
(xi
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
-+----x-X1
f'(x1) =0
Theorem 3.
If f has an ILMax at
m(x)
f(x) - f(xi)
,
X
Xi
-
so that
Jim
m(x)
f'(x1).
Let
0.
0.
214
1)
5.2
Suppose thatf'(xi) > 0. Then the function m(x) must be >0 when x Xi.
y
Take x
!::::!
x Xi
and
Then
x > Xi
=>
m(x) > 0
=>
=>
and
x - Xi > 0
=>
2)
Suppose thatf'(xi) < 0. Then the function m(x) must be <0 when x Xi.
!::::!
Xi
and
which is impossible.
Since
proved.
(1)
and
(2)
x < Xi
and
=>
m(x) < 0
=>
=>
x - Xi < 0
=>
0, which was to be
215
5.2
This is the standard method for finding an ILMax. Given a differentiable function
x wheref'( x)
0. Usually there are only a finite number of
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
such points.
x3, -1 x 1,
thenf'(O)
0, it does
not
For example, if
[-1, 1]
y
Proof
x1,
Let
andf is differentiable at
g(x)
x1.
x1,
then f'(x1)
0.
-f(x).
Thereforeg'(x1)
0.
Thereforef'(x1)
-g'(x1)
I
I
I
I
I
I
[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
the right,
An
Hence:
inflection point
ILMax or an ILMin.
f'
has either an
216
5.3
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.
x2/3 (0 x
<
oo
oo ,
jx-1/3
x = 0.
The
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
f(x)
= x sin
/(0)
for 0 <
'TT
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
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,
and so on, and also intervals with holes in them.
tangent is
D
{x I x
( - oo,
oo
TT/2
oo),
[O,
oo ),
nTT};
5.3
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.
!"'---
_)
-1
1l
I
I
I
I
I
I
-1
f(x)
1
=
(xO)
I
I
f(x)
1
x2-1
xI
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)
lim
f(x)
0,
(2)
lim
f(x)
= oo,
x--+ oo
x-+-co
(3)
x-+o+
lim f(x) =
x-+O-
(4)
- oo.
The
function whose graph is shown on the right above has the following properties:
i)
lif(x)
0, x2
>
0, x2 - 1
0. (At x = 0,
-1, and 1/(x2
>
1)
<
-1.)
-oo.
o:--+1
iii)
iv)
v)
lim
f(x)
- oo.
x-+-1
lim
x-+-1-
f(x)
oo.
Here statements (ii) through (v) mean the things that the figure suggests.
An
> 1,
For example, if
218
5.3
limx00f(x)
> M
=>
E.
L-<
--- --------
---------
Roughly,
L.
means
x x0
=>
f(x) L,
means
=>
f(x) L.
lim f(x)
and
limf(x)
oo
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).
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
x
This is trivial: take M
Definition.
limx-00/(x)
1/E. When x
=
x <
whenever x > M.
>
- E </(x) <
>
0 there is an
L +
E.
such that
5.3
219
L-
limxx0+ f(x)
oo
x0 <
x < x0 + o
M there
f (x) > M.
=>
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
of x0 and very close to x0 (i.e., between
x0
and
+ b.)
x0
x x
or
x x)
x to
the right
because these
often turn out to be different. In some cases, however, the one-sided limits have the
same value. In such cases, limxx.
common value.
Thus
1.im 2
1
x-+O X
= 00.
The following two theorems justify the remarks that were made above about
f(x)
1/(x2 - 1).
Theorem 1.
If limxx. f(x)
0, then
1
= 00.
j(x)
Proof
x0
<
Given
f ( x) <
whenever
x0
< x <
x0
+ b.
5.3
220
x0 < x < x0 + o,
f(x) <
we have
1,
and hence
1
__
> M.
f(x)
(Remember that M >
0,
0 for
the values of
Similarly, we have:
Theorem 2.
Suppose that
Proof
Given
-E
When
Let E =
M < 0.
f(x) >
1
-
f(x)
0, then
= -oo.
-1/M > 0.
x0 < x < x0 + o.
whenever
<f(x)
x0 < x < x0 + o,
If lim,,_,.,0+ f(x)
we have
1
-E,
< - '
f(x)
-- >
E
1
-
f(x)
-<M.
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:
lim
f(x)
oo,
oo.
x--oo
Jim f(x) =
lim
- oo,
f(x)
X4-00
(?)
lim
x-+oc:
(1 !)x (?)
+
We use question marks, because it is not obvious that the indicated limit exists at all:
as
x---->-
of the
oo,
Jim
h-+0
This was also of the form
f(u)
(1 + h)lf1i
"1 a:i."
=
= e =
111-1
l.
(1 + u)11u
and
g(x)
1
=
'
5.3
221
then
( r
1 +
f(g(x)),
x-+ co
If lim., xo g(x)
= u0 = g(x0)
(u o ) .
limf(g(x))
For the case in which
x---+
ro,
= j(u0), then
=f
If lim.,_.00
g(x) = u0
and limu_.,,0
f(u) =
limf(g(x))
L.
x-+ oo
L, then
ro
g(x)
u0
If lim.,00
g(x)
and limu oo
ro
f(g(x))
lim f(g(x))
x oo
ro.
f(u)
L.
L, then
L.
f(u) = (1 + u)llu,
g(x)
1
-
f(g(x))
(1 ;r
+
we get immediately:
Theorem 5.
lima:-.oo
(1 + 1/x)"' =
e.
e,
theorem.
e =
limho
(1 + h)1fh
appears as a
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. f(x)
1
=
x(x
- 2)
(x >'6 0, x >'6 2)
222
2 f(x)
3.
f(x)
(x - l)(x -
4. f(x)
8. f(x)
IO. f(x)
x-+oo
3)
(x -2, x
x
-2
x + 1
1
x + 1
3
x -x
)"'2
( -1 x)
1
1 + 2
X
( Ir
( r
3. 3r2
(
13.
16.
lim
x-,,12
0)
xz
7. f(x)
9. f(x)
-2
x + 1
x + 1
-3
x +
1-
x3
-3
+x
(-1 x)
(-1
x)
x-o+
Jim (x)1f(x-ll
x-1
1 +-
18. f(x)
1 +
21. /(x)
2x
(x
+ cosx)secx
(1
11 . f (x)
(x 0, x 1, x -1)
20. f(x)
5. f(x)
17. f(x)
3)
-2
x + 1
Investigate:
12. lim
( x ;= l, x ;=
2
x -x -6
6. I <x)
3)
5.3
(1
2
1 +x
+
(1 L)"'
( -rx
19. f(x)
22.
e-xy'
l +
ln x
1 + x
11
=
(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
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
(x a),
and
Jim f (x)
x--oo
Jim
x-
h(x)
L,
5.4
223
(1 + t)2
$2.25.
Suppose now that interest is compounded continuously: the bank passes to the limit,
as
increases without limit, and at the end of the year they charge you the limit. How
30. Suppose that the basic interest rate is 6%, but interest is compounded continuously,
as in Problem 29. How much do you owe?
you will need to use one of the tables at the end of the book.)
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
from
o to
4
x ,
1.
!I
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)
and then set
x ==
t.
(ab)
==
In
a +
In
b,
224
5.4
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
(x > 0),
(x > 0).
f(x) =In kx
g(x) = Ink + In x
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
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).
.! x + .J1 - x2
2 .Jr x2
-
1 2x2 - 1
2 .Ji - x2
.! -x2 x2)
2
.J 1 - x2
(0 x 1).
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.
5.4
225
Theorem 1
y
y
--(1\1
I
I
I
I
I
--
I
b=x
If the
y =sine,
x =cose,
and
A(O)
226
5.4
Therefore
The only point 0 on the interval [O, TT/2] where A' (0)
= !!. .
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?
=
y
3
2
Solution. Let
f(x) =AP+ PB
= .J12 + x2 + ./(3 - x)2 + 22
=.Ji + x2 + .Jx2 - 6x + 13.
5.4
Then
f'(x)
x
J +
i
xJx2
x2
x -3
Jx2
6x
227
+ 13
6x + 13 + (x - 3)-,/
J 1 + x2 Jx2 - 6x + 13
-
Thereforef'(x)
0 when
x2(x2 - 6x + 13)
or
x4
or
or
6x3
+ 13x2
(x2
6x
-9
x2 +
2x
- 3 = 0,
or
(x
+ 3)(x -
I)
6x
+ 9)(x2 + 1),
x4 - 6x3 + 9x2 + x2
3x2 +
6x
+ 9,
0,
0.
I)
x decreases past 0, AP increases, and so does PB. The same is true when x
3. Therefore, in searching for a minimum, we can restrict the search to,
interval [ -1, 4].
When
increases past
say, the
2)
Suppose that we know that the function has a minimum, somewhere on the
interval (-1,
4].
x =
I.
1.
0. There
Theorem 2 (Existence
value on [a, b].
ofm in ima).
If/is continuous on
1 is true.
-M
:
b l -- x
-- x
-+--+'a'--I
-t- ---I
-f
I
I
228
5.4
Here again, once the problem is solved, you may be able to think of a simpler
attack on it.
Problem 3.
of radius 1.
y
-1
(0 x 1).
This gives
V'(x)
2TT 2x
Ji -
x2 + x2
x2
Ji -
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 =
Now V has a maximum, because V is continuous on [O, 1].
ILMax, because V(O) = 0
fore V'(x)
Ji
J'i.
Hence
There is another function that we might have used to solve the same problem.
might have written
V(y)
TTr2h
TTX2 2y
TT(l - y2)2y
2TT
(y - y3).
We
5.4
so that
V' (y)
<=>
3y2
<=>
Jt
or
229
-/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
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
x0
x0
x0
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
a.
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.
If the area of
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
5.4
230
is the thickness, and k is a constant depending on the wood (and on the units of
a,
in such a
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.
square pieces are to be thrown away.) If the sheet has edges of length
a,
(The
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
10.
128
10
possible.
11.
12.
Solve the same problem, given that the cylinder is open at one end.
Solve the same problem, given that the cylinder is open at both ends.
13.
n feet long, with open top, open ends, and triangular cross sections. What is the greatest
possible cross sectional area?
14.
100
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 =
y =
17.
18.
Find the rectangle of maximum area contained in the region above the line
to the right of the line x
19.
x4
1,
{(x,
y) I ixl
= 1/x.
[y\ 1},
y = t,
in such a way as to
and j ustify your answers. You need not find the maximum values of the functions.
20. f(x)
22.
h(x)
x
=
+ x2
J"'
Problems
-1
Sin-1 t dt
24 through
21.
23.
g(x)
</>(X)
=
=
x
--
1 + x4
finx
V1
+ t8 dt
5.4
231
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
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.
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.
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.
that LBAC has measure
and
ix.
ix,
Given
232
5.5
5.5
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
What position for the segment gives maximum area for the resulting triangle?
y
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
of the other endpoint.
0 to 1, letf(x) be they-coordinate
[j(x)]2 = l2,
[ f(x)]2 is the square of the length of the segment. Therefore the function
(0
x2+f2=l
The area of the triangle is
A(x)
Now in
(1),
1).
(1)
tx f(x).
(2)
(0 <
+ff' = 0
< 1).
to
1.
(l')
0 <
< 1,
but this
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.
We now know:
!' =
!' = _i
x
on
(0, 1),
at the maximum.
(2")
5.S
233
x 2 + f2 =
1,
x + 2 ff'= 0,
f' =
J;
A(x) = tx f(x),
and hence
A'= 0
f'= - j
x
<::?-
x
- -= _J
x
f
x = f(x).
and
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=
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.
in Section 5.4.
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,
and
f' =
x
f
x + 2 ff' = 0,
(0
<
<
a).
(3)
s.s
234
so that
0, and so
f' =
- 2xf= - 2!
x
x2
(4)
(at Max).
Therefore, at the maximum, both our formulas for f' must hold, and so
2
=- f
x
f
-
and
./2
For a
(5)
f=-x.
2
- '\/1;,_
3
-
'
as before.
Note, however, that in a way the most natural answer to a problem like this is a
gives the answer in the form of a shape, that is, in the form of a ratio between two
measurements.
mines the
size
a,
which deter
x2 + f2 = a2.
(5)
2f(x) = /2 x,
-a
5.5
235
The answer is also a shape when the problem is to find the rectangle of maximum
area in a given circle:
x2 + f2
f'
2x + 2ff'
(0
-
f
A(x)
a2,
0,
x < a);
<
(2x) 2f(x) = 4
x f,
<=>
f'
-f
x
because
f '=
and
x = f,
The constant
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
=>
Eq. (7).
(6)
2. x + 2 . ff' = 0,
(7)
It often happens that there is more than one such function/ For example,
consider
Here
f{(x)=
and
f(x)
-x
,
fi( x)
-x
..Ja2 - x2
x
..J a2 - x2
-x
--;===
-..Ja2 - x2
Therefore
Therefore
fi(x)f{(x)
-x,
and
{2. x + 2. fd{ = 0,
2
-x
f2(x)
. x + 2. ! !
2
0.
(8)
236
5.5
That is, both/1 and/2 satisfy (7). A figure makes it obvious what is going on here.
y
fi(x)=
-x
=
f;(x)/x f;(x) '
-1
y= x3 - x
(9)
x= ya-y
(10)
We have interchanged
and y in Eq. (9), and reflected the graph across the line
5.5
y =
x.
237
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
3 Pf'
!'.
This is what we are claiming when we differentiate the functional equation, and write
x
=/3
=>
1 = 3 . !21'
f'.
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.
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.
238
5.6
x +ff'= 0.
(You need not show that the solutions that you describe are the only ones.)
F(x) =
Calculate
F(x),
f'f
(t)/1(t) dt.
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
=
f (x) = (x - a)2
Differentiating, we get
and
[j'(x)]2 = 4
Evidently (1)
/(x)
(2)
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
In Section 5.4 and later, we have used the fact that, if f is continuous on
then f has a maximum value o:n
[a, b].
[a, b],
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.
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
numbers.
239
to A nor to B. Then
1)
1--"x'---x-- 1=----
(1) holds.
Similarly,
2)
-,...2
l X--:':-X-+-1 ---
-
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
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).
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.
B
Xo
Given a sequence
240
5.6
nested.
For example, if
( - ! !)
[ai' b]
i
'
i.
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.
scribed regular
(n +
2)-gon. Evidently
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
have just discussed.
Similarly, if
./2
that we
Using the nested interval postulate (NIP), we shall prove the following theorem:
Theorem
1. If/is continuous on
[a, b],
[a, b].
[a, b].
Lemma. If/is unbounded above on an interval [c, d], then/is unbounded above on
at least one of the halves of
By the halves of
[c, d]
[c, d].
[c, (c + d)/2]
and
then/has an upper
5.6
241
I
I
I
I
I
--1------1
I
I
I
I
+-
c
c+d-d,,...__
._ x
2
We proceed to prove the theorem.
good
[a, b]
[a, b]
bad if
Let
it is not
If
[a, b] is bad, then it follows that at least one of the halves of [a, b] must be bad.
[a1, b1] be a bad half of [a, b]. For the same reason, [a1, b1] must have a bad half.
[a2, b2] be a bad half of [a1, b1]. Continuing this process to infinity, we get a
sequence
and so
b; - a;
By NIP, there is an x such that
But f is continuous at
-: (b
2'
a; x b; for
a).
each i.
x.
f has
(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,
i-+ 00
(b; - ai)
0,
(x -o, x + o).
But since
242
5.6
we have
bi - ai
<
for some i.
(x
- a,
x-o
a;
(x - a, x + a)
[ai, b;]. This contradiction completes
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
(bi
a;)
0.
i-+ 00
(x - a, x
(x - a,.\'
Tan-1.
y
---------------
7r
5.6
243
to
smallest.
Here
sup
is
2
pronounced supremum.
To be exact:
k =sup B.
(a, b).
Every number
Here
Therefore
Consider now
B_
=sup B.
{1 '3 '4'
... f'
[ 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
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:
5.6
244
Theorem 3
supf
g(x)
1
=
(a
k - f(x)
for some x.
b).
g(x) M
Then
1
:$
k - f(x) -
and
f(x) :$
-
M,
- _!_
M
k - f(x),
for a
b.
Theorem 4
of Section 5.4.)
2.
Let B be the set of all rational numbers p/q for which p2/q2 < 2. What is sup B?
Consider a circle of radius 1. For each polygon
3.
245
5.6
Consider the graph off
(x)
sin
x, 0
1T.
[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?
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
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.
then k is a strict upper bound of B.
If
in B,
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
of B by its negative
-x.
That is,
-B
For example, if B
(-
oo,
{ -x Ix EB}.
[ -2, -1]; if B
[ -1,
oo
) , then -B
B And
.
{ava2,
}, B
{b1,b2,
}.
x bi for every i.
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.
246
* 11.
5.7
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.
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.
=
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.
=
f(b
l-
----------
5.7
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)
g(b)
f(a),
f(b).
g '(x)
f(b) - f(a)
,
b - a
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
c/>'(x)
Since cf>(a)
c/>(b)
some x. Thus
=
f'(x) - g'(x)
f(b) - f(a)
f'(x) b - a
f '( x)
and
f' (x)
f(b) - f(a)
b - a
=
Therefore cf>'(x)
0 for
o,
f(b) - f(a)
a
b
-
x0 -
O <
<
x0
+ c'J
=>
f(xo).
f (x)
> 0.
If
248
x0 -
< x < X0
because f(x0)
+ o
=>
E =
f(x0)
f(x0) -
0. Therefore the
E =
5.7
>
0.
There is a
<f(x) <f(x0) +
o >
E
0 such that
=>
0 <f(x),
that we wanted.
o >
0 such that
x0 -
Proof?
< x < x0
=>
+ o
f(x) < 0.
But
1.)
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)
>
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)
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
b; - a;
i(
b - a),
5.7
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
x0 + o). By Theorem 2 of Section 5.6, there is an i for which [a;, b;] lies in (x0
Xo + 0), as indicated in the figure.
o,
0,
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
and let
g(x)
f(x)
k.
5.8
250
Nobody expects that a doctor will write down a definition of the word
and then write a few assumptions about
men,
man
can be derived by logical reasoning from the definition and from the assumptions.
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.
As late as 1800,
mathematical ideas which require careful logical analysis have become more
5.8
Let
fandg
Af= f(x0
A g =g(x0
Ax) - f(x0),
Ax) - g( x0) .
If
off
!if= df
6.x-+O A g
dg,
Jim
'
df
f
-=-,
dg g'
whereverg'(x) "16- 0.
Proof
!if
6.x-+O fig
lim
Jim
6.x-+O
!if/fix
fig/fix
x,
x,
f'(xo)
.
g'(x0)
then
df df
=
dg dx
f'(x).
Obviously,
df
=
dg
for each x0
Jim
6.x-+O
!if
fix
f'(x0),
5.8
251
d sin x
=
---
d cos x
dsin x
--- =
dx
cos x
---
-sin x
-cot x,
cos x,
de"'
e"'
dx2
2x
(wherever x ":/= 0)
We often write
d
-f(x)
dx
df
for
dx
f'(x)
'
f
d
-f(x)
dx
df.
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
fx 'X=Wo'
df. dg
dg dh
df
dh
l:lj
!:lg
df
----
as l:lx
--+
dg '
!:lg
dg
l:lh
dh '
_ _,,_ _
0. This requires
g'(x0) ":/=
0,
Sometimes
252
5.8
Theorem 3.
dg
df
dg. dh
'
df
dh'
dg 1_
df
df/dg'
_
wherever df/dg :- 0.
du2
du
This is like
2u.
df
dx
f'(x)
2x.
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.
Let f and g be functions.
then we say that f is a function of g.
Definition.
</>(g),
d sin2 x
d sin x
---
du2
=
2u
2 sm x'
d(u2 - 2u)
d(cos2 x - 2 cos x)
d cos x
du
du
2u
2 cos x - 2.
Let/be a function of g,
Then
wherever g' :- 0.
</> ( )
' g,
5.8
253
Proof
df
dg
f'
g'
gg
</>'( ) ' = f (g).
g'
2 tanx,
dtanx
instead of using Theorem 1 and writing
dtan2x
2 tanx sec2x
dtanx
sec2x
= 2 tanx'
4. f(x) =exsinz,g(x) =x
5. f(x) =exsinx,g(x) = x3
__
</> such
.) Then calculate ' (u) =
. Finally,
calculate <f>'(g),and compare it with your previous formula for df/dg. (Or, if you worked
that f = </>( g). (Answer in the form
</>(u) =
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
26.
Given /3 + t3 = 1, find df/dt. Then calculate f = f(t), find /' (t), and compare the
27.
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
6.1
of Integration
INTRODUCTION
to
b,
To :find the
function
f f(t) dt
A=
rYf
I
We know that
F'(x)
f'f(t) dt
f(x)
for every
G'
G'
If it happens that
G(a)
0, then we have
H(x)
Then
H'(x)
Therefore
G'(x)
H(x)
x.
G such that
F '.
G(x)
F(x)
for every
G(x) - G(a).
F'(x)
F(x)
and
for every
H(a)
x,
and so
ff(t) dt
F(b)
H(b)
254
G(b) - G(a).
0.
x.
If not, we let
6.2
255
To sum up:
G'
The notations
and
=>
G(b) - G(a).
F'
More formally:
Theorem 1
ff(t) dt
ff(x) dx
=>
F(b) - F(a).
f,
Iff is continuous on
then
ff(x) dx
F(b) - F(a).
To apply the theorem, ofcourse, we need to find F when/is given. This process
antidijferentiation. We shall see later that the method of antidifferentiation
is called
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
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),
is called the
g(x)
..}
(x
0).
independent variable.
It is simply a
Logically
For example, we
f (t)
t2
- 1
g(t)
g(t)
Ii
(t 0).
cos2
x.
=
the uniqueness theorem, that iftwo functions have the same derivative, on an interval,
256
6.2
C,
C) =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
C},
C},
and so on. Any other dummy letter would have done as well:
and
f(3x3
2x4) dx =
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.
257
6.2
When the problem is understood in this sense, it is plain that the answer is
3-
(i)
In (ii),
(iii)
(iv)
In (iii),
x,
=>
1)
f xndx= { n 1 xn+1 + .c}
2-y
=>
J)xdx=
sin
cos
=>
f x dx=
{sin
cos
-sin
=>
f xdx=
{-cos
D(-cos
sin
{2Jx + c},
cos
x + C},
x,
sin
x + C},
(n
-1),
258
In
.! (x
x
>
0)
6.2
=>
De"'=e"
=>
J dx=
fe"' dx=
{In
x + C} (x
{e"'
+ C}.
>
0),
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
f (u )
is a composite function.
Df(u)=j'(u)u'.
It follows that
ff'(u)u'(x) dx
{f(u(x)) + C}.
For example, if
u(x)= x2 + 1,
f(u)= sin u,
then
(cos
u)2x
[cos ( x2
More generally,
sin
u(x)
u(x)]u'(x),
[cos
and so
[cos
u(x)]u'(x) dx
{sin
u(x) + C}.
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
u'(x) dx by the
cos
symbol
du.
Thus we write
6.2
259
[cos
u(x)]u'(x) dx
u(x),
we have
Similarly, we write
fe"du
which means that if
u is any
{ e" + C}
fe""">u'(x)dx
This is true, because
Deu<x>
{eu!x> + C}.
e"x>u'(x).
Using different dummy letters, we can convert the above formula to any of the forms
feu<t>u'(t)dt,
or
and so on.
fex2+12x dx
has the form
where
u(x)
x2 + 1.
Therefore
fex2+12xdx fe"du
=
Similarly,
[sin
(t2 + 1)]2t dt
sin
u du
{ e" + C}
form
{ ex2+1 + C}.
f sin u du.
Therefore
{ -cos u + C}
Note that the solution is not finished in the third formula above, because
function.
t.
To sum up:
F'
=>
D[F(u)]
= f (u)u'.
Therefore
ff(x)dx
{F(x) + C}
ff(x)dx
=>
ff(u(x))u'(x)dx
{F(x) + C}
=>
{F(u) + C}.
we have
ff(u) du
{F(u) + C}.
is a
in terms of
260
6.2
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
cos
J J:.udu={
ln
u + C} (u
>
sin
0),
u
e
because
D[f+ g] =DJ+ Dg
D (kf) = kDf
and
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
We therefore have
This checks:
Example
2.
Consider
-Jx
1_ dx
"\ x
cos
(x
>
The only form that might fit this integral is the form
u(x) =.jX,
0).
f cos udu.
1
du = u'(x)dx =--dx.
2.jX
The only difference between what we have and what we want is a multiplicative
6.2
261
constant. Therefore
JX
dx=
Jx
cos
Example 3.
(cos v x)
;-
1
x dx= 2
J
;(cos v x)
1
x dx
2J
C}= {2 sin Jx
C}.
Con:.ider
C}.
u(x)
=cos
x,
du
u'(x) dx=
-sin
x dx.
J{-
-e008"'(-'-sin
eu
x) dx= - eu du
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
u,
u(x)
u(x)
>
0 for
every
x,
f (lju) du.
Given a function
we know that
1
D ln u(x)= - Du(x).
u(x)
>
J du=
But even where
u(x)
<
0,
{In
C}
(u
D[ln
This gives us
u(x)
<
0).
J du;
answer is easy: if
>
0,
then
-u(x)
>
0.
Therefore
J du= {ln(-u)
C}
(u
<
0).
262
6.2
below.
Ikf(x) dx k ff(x) dx (k O)
f [f(x) g(x)] dx ff(x) dx f g(x) dx
=
(n -1)
f du
{ In u
f du
{In (-u)
C}
(u
>
C}
(u
sin
sin
0)
(4)
<
0)
(5)
-cos
sec2
csc2
tan
-cot
sec
tan
sec
csc
cot
-csc
"
"
e
f
{
1
u
I u J
u2 - 1
=
>
Sin-1
Tan-1
{ Sec-1 u
.
<
C}
+
C}
(2)
(3)
f u du { u C}
I u du { u C}
f u du { u C}
f u du { u C}
I u u du { u C}
I u u du { u C}
f du { C}
Ja" du {1:"a c} (a 0, a 1)
I
J1 - u2 { u C} (lul 1)
cos
(1)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
(13)
(14)
(15)
(u
>
1)
(16)
6.2
263
To solve the following problems, you will start by expressing the given integral
in the form
f f(u) du.
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
J(x3
+ 1)10 x2dx
Jt<x3 1)13x2dx
fiu10 du {t TI-u11
C}
u{x)
du
x3 + 1
3x2 dx
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}
(?)
f(x3
u(x)
1)10x2dx
x3 +
1, du
(?)
J<x2
+ 1)2dx
(?)
J<x2
+ 1)2dx
du
x2 + 1
dx(?!)
J<x2
+ 1)2 dx
J<x'
+ 2x2 +
1) dx
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.
4.
f
I
(1 + x2)3x dx
2.
(t4 + l)t8dt
5.
f
f (x2
(1 + t 3)t2dt
3.
t2)3 dx
6.
f
f (x2
(2 + u2)3u du
+
t2}8tx dx
264
7.
10.
13 .
16.
19.
21.
22.
24.
27.
30 .
33.
3 6.
39.
42.
45.
48.
51.
54.
57.
f
J
f
f
f
f
fl
f
J
J
J
J
J
J
J
J
J
J
f
(rs/2
l)rs/2dt
11.
14.
(e"'
x2
(I
+ x3)3
f
f
f
8.
e-"')dx
6.2
(l
1
+ y1x)3-=dx
v'x
(I
+ 2)4 e"' dx
(e"'
17.
x2
(e"'
12.
15.
f :
20.
dx
dx
1 + xs
9.
f
f
f
e-"')3 dx
x2
(t3f2 + 5)10.,/(dt
(e"' - 2)3e-"' dx
f :
18.
(1
dx
x2)2
dx
23. f
in xdx
25.
28.
31.
(cot2 0 + 1) dO
34.
cos 0
-- dO
37.
40.
cos2 0dO
43.
sin2 0
sin2 0dO
46.
49.
cos (0/2) dO
52.
x e-"'2 dx
55.
e2"'dx
58.
J
J
J
J
J
J
J
J
J
fJl
J
J
In (x2)dx
(Same comment.)
26.
cos2 x sin x dx
29.
(1 + tan2 0)dO
3 2.
cot2 0dO
3 5.
sin 20d0
38.
41.
(1 - 2 sin2 0)dO
44.
sin2 20dO
47.
sin 0
50.
(1 - sin2 0)dO
- cos 0
dO
2
5 3.
t2et3 dt
56.
e5t dt
59.
J
J
J
J
J
J
J
J
J
J
J
J
tan2 OdO
sin 0
--do
cos2 0
cos 20dO
(2 cos2 0 - 1)dO
(2 sin2 0 - 1)dO
sin3 0dO
xe"'2 dx
e1
tdt
6.3
60.
63.
66.
69.
72.
75.
78.
J dt
J sin t dt
I dx
J (2
dt
dt
I
et'+3t
61.
e008 t
64.
(10x)2
67.
+i-312)
70.
( .Y1
I
I
t3
1 +t
73.
12)s
7 6.
4 dt
ex
Yl - e2x
dx
osx
c--dx
I x
2 I x x x xdx
f dx
J2x+idx
dt
I (2
ein sec' x
62.
65.
+ o-3/2
68.
tdt
I
dt
I
I exdx
71.
.Y1 - t2
t2
74.
\o/1 +t3
ex
77.
1 +
xdx
-x
I 2
csc x x xdx
I x x
Algebraic Devices
f cosxdx
J xdx
J
dt
dt dt
I
dt
I - (2t)2
I -=dx
esin x
10x
t(2 +t2)-312
79.
80.
sin
sec 2
8 .
+sec
tan
83.
secx +tan
.y 1 - t2
.y 4
ex
Y] _ex
84.)
sin
cos
+ csc
csc
265
cot
+cot
81.
84.
J xdx
J xdx
tan
sec
J du/u,
(0, oo)
(-co, 0).
u
d
-;
;
I
du
- =
I
=
Since
lul = u
when
>
and
{In u +
{ln
( - u) + C}
luJ = -u
0)
(4)
(u < 0).
(5)
(u
C}
when
>
u < 0,
r- =
du
Ll
{ln
Ju J + C}
(on
(0,
oo
or
( - oo, 0)).
(17)
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
"1/u"
0.
When
0,
u < 0;
it cannot be used
on the left or the "In lul" on the right. Thus, whenever we apply formula
r -l d
J-2
(17),
(17) is that it is easier to use.
6.3
266
l-1d =
-2
2
F(-1) - F(-2)
F(x)
In !xi.
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):
- ln2.
Hereafter, we shall use the following shorthand for this kind of calculation:
r-i dx
L2 x
In general
by definition. Sometimes, where no confusion could result, we may omit the opening
bracket on the left. Thus
3I= -
tan u du =
we have
cos u
du.
v' v u, dv
u du.
J
dv
(v
v
J --;- vl
u
d:
u du =
du = J
u
J
J
=cos
Since
sin u
--
tan
{ln l
+ C}
= - sin
> 0 or
<
0),
- sin
--
cos
tan u du
{Jn !sec ui + C}
<
0).
(18)
6.3
Algebraic Devices
167
Similarly,
This gives
cot u du =
cos u
du.
-.
smu
-
(19)
secxdx.
Since
secxdx =
D
and
tan x = sec2 x,
where
u = secx+ tan x,
Therefore
du
secxdx = {In u
J l + C}
As always, the chain rule gives us a more general formula for J sec u du:
(sec u+ tan
Similarly,
cscxdx =
cscx(cscx+cot x)
csc x+cot x
(csc u +cot
Sec-1x =
1
xJx2
(x > 1).
(21)
268
6.3
7r
(See Section
4. 7.)
oo
).
vertical; and so the differentiation formula holds only for x > 1. It gives
dx
.
.J
x x2 -
+ C}
(x
>
1),
du
= {See1 u
U'\I U 2 -
+ C}
(u
>
1).
{See1
(16)
-2
1
J-ax.Jx2-l dx
___
makes sense.
Ix.Jx21 - 1
dx
__
where
-1).
On this domain,
1
dx=
(-1)
I -x.J( -x)2 - 1
f .J - 1 du,
u
u(x)=-x,
Therefore, for
<
x'./ x
-x
du=(-l)dx.
-1,
-1
= {See1 u
=
because !xi
u2
{See1 lxl + C}
when x < 0.
du
fu.Ju2 - 1
+ C} = {S.ee1 (-x) + C}
(x
< -1),
instead of
{Sec-1 lul + C}
x),
we get
(u
>
or
u < - 1)
(22)
6.3
Algebraic Devices
269
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.
D In x = .!. .
x
{u
ln3 x
dx=
=
I
{:
(ln3 x)
du= - dx
x
x dx
(1 + x2)7 .
Looking for functions which are derivatives of other functions, we observe that
x dx
=
(1 + x2)7
2
where
u(x) =
2x dx
! u-7 du,
J
=
(1 + x2)7
2
+ x2,
!.
Example 3.
1
-7 + 1
u-7+l
} {-l
+ c =
12
} {-l (1
u-6 + c =
x dx
)1 - x4
12
= l x
u3 du
+ c = {t ln4x + C}.
Example 2.
dx=
+ x2r6 + c .
270
6.3
2x dx
du
1
- u2
Example 4.
=J(1 + -) dx.
x-1
Jx-1
(Here we have divided the denominator into the numerator, getting a quotient and a
remainder.) Therefore
Example 5.
xdx
= {x + In Ix-11 + C}.
x-1
as this:
dx
J1 + e-x
else. But
dx
du
=
=.
1 + e-x
ex+ 1
u
u=ex+ 1
r --dx
-
e" + e-x
---
du
e" dx
e" dx
1 + e 2" - 1 + (e")2
1 + u2
(u=ex, du=e"dx)
_E_= !
J + x2 J
4
Here
dx
t dx
= !.2.
1 + (x/2)2
4
1 + (x/2)2
{t
Tan-1
+ c}.
6.3
Algebraic Devices
271
x
(1/J3) dx
I J3d-x x2 I/31 Ji -d(x/J3)
I
2
Jt (x/J3)2
{ ;3 }
Similarly,
sin-1
+ c .
Ia2 dx x2 {.!a
IJa2dx- x2 {
+
Tan-1 + c
u2
IJa2du- u2
{a
{
-Tan-
(/3)2
In the same
(a > 0).
we get two more standard formulas:
+ C
22 and 3
(a > 0),
sin-1 + C)
Passin
sin-1 + c
(a > 0),
(23)
(a > 0).
(24)
1. I
4. J
7. I
10. I
13. I
J
J : e
J e"'e"' - e-x
e25. J (e"' e-')(e2"'
v' 1
x4
v'4 - y4
dx
v'1 -
dy
5.
9 :sx4 dx
16.
19.
22.
zs
zs
dt
dt
2zs
dz
dz
1 :55zs
dz
xs
dx
dt
dz
,.dx
t4
11.
dz
t2
x
dx
1 + 9x 4
x
3
dx
Vl - x8
2. J 4
J
8. J
J 1
14. J
17. J 1
20. J 1 :t e2t
23. I e"'e"' - e-x
e-x
J
e-2"') dx 26.
dx
2
dx
v'2 - x3
3. J
J
J
12. J 1
J
18. J
21. J e"' e-x
24. J (e"' e- )(
2 7. J
4y4
v'1
6.
15.
:39 4 dx
x
z6
dz
6dz
+ z
x
7
dx
Vl - x8
xs
dx
dx
"'
e"' - e-"') dx
2
2
dx
v'2 - x6
272
2 dx
28. J v1 x-2x3
J v1Sin-1x
-x2dx
34. J xe"' dx
37. J (xcosx
sinx) dx+
J x sinxdx
43. J (3x+2 Inxx2
) dx
1
46. J x( l + 1n2x) dx
49. I x2 V; - 1dx
J V 1 x2dx
J Cos-Ixdx
f
1 + e4u du
6 .1 I 1 :x2dx
64. We know that
31.
40.
52.
55.
58.
Consider
6.3
2 dx
29. J v1x-2x6
32. J Tan-Ix
-1-+x2 dx
35. J In e"'' dx.
38. fxcosxdx
J (2x Inx+ x) dx
44. J x21nxdx
47. J (x + x1)2In+(x2x2 +2x) dx
f Ve2x dx
f Sin-Ixdx
J Cos-I (2x) dx
f du
59. l :2u
62. f Tan-1 xdx
41.
50.
1
30. fxln-2xdx
33. f (xe"' + e"') dx
36. J ln2 e"' dx
f (x sinx -cosx) dx
42. J xlnxdx
J ln3x
--dx
x
48. I tVt21 dt
J (sin-Ix+ Vl x2) dx
J (Cos-Ix - Vl x_ x2) dx
f du
1 + e2u
f(Tan-Ix+ 1 :x2) dx
Jv1 -z2dz
39.
45.
51 .
53.
54.
56.
57.
60.
63.
x2
6.4
Integration by Parts
273
f(x)
Then
F'
F(x)
x2'
1
x
f Therefore
fl -dx
1
-1
x2
F(l) - F(-1)
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.
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
6.4
0 to
1.
Show
INTEGRATION BY PARTS
By differentiation, we get
D[x sin x]
x cos x
x.
+ sin
Since
D cos x
sin
x,
+ sin
we have
D[x sin x
+ cos
x]
x cos x
sin
x cos x.
Therefore
Jx
cos
x dx
{x sin x
+ cos
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.
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
T herefo re
J [u(x)v'(x)
u'(x)v(x)] dx
Ju(x)v'(x) dx
u(x)v(x)
{u(x)v(x)
C}.
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.
short notation
du
u'(x) dx,
dv
v'(x) dx,
Using the
6.4
274
Ju dv= uv - Jv du.
This is the formula for integration by parts; the word parts refers to the functions
x cos
x dx.
Let
dv
=x,
x dx,
=cos
so that
du= dx
and
= Sill
X.
will work. We will return to this point in a moment.) By the basic formula, we get
x cos
x dx = u dv= uv
Jv du= x
= {x sin x +cosx
sin x
sin
x dx
+ C}.
=sin
+ c,
we would have
Judv =
x(sinx + e) -
(sinx + ;)dx
= {x sinx + ex + cosx
- ex
+ C}
= {x sin x
+ cos
+ C},
u(v
+ c)
f<v
+ c)du
= uv
+ uc
Jv du - uc
= uv
- v du.
In applying the basic formula, we made what may seem to be an arbitrary choice
of
u and dv.
=cos
x,
dv = x dx
du
= -sin x dx,
x2
v = -.
2
Integration by Parts
6.4
udv = uv -
v du
cos x +
x cos x,
du= cosx
du= dx,
which gives
xcosxdx = udv = uv -
x2sinxdx.
f x cos x dx,
275
x sin x,
vdtt
Here again the new integral is harder than the old one. We remember also that no
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
Practice is a help,
but there are general rules which help us to decide what choices are promising:
1)
2)
We want
simpler than
3)
f vdu to be
u.
f u du.
J xe"'
r
u = e
u=
J xe"'dx = J u
Note that
(3)
x,
du=
i:
dx.
Therefore
(1)
gives us no guidance.
"
e ,
Dx
Rule (2)
u = x is to be preferred.
du= dx,
uv
This gives
u.
(De"'=
Therefore we want du to be
use
v = e"'.
C}.
dx,
6.4
276
won't work.
Consider next
f
Rule
x2e'" dx.
du = e'"dx. We therefore take
du= 2xdx,
u= x 2,
This looks good under rule (2), and acceptable under rule
x2e'" dx =
du=
uv
(3).
du = x2e" - 2
= e'".
We get
xe'" dx,
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;
integral that we started with, but an equation that can be solved for, this integral.
Consider
We take
u
= e'",
= -cos x,
u= e'",
du= e'"dx,
=sin
x.
I=
We then have
I = -e'" cos x +
Here the last integral is simply the one we started with. Therefore
2I = {e"' sin
and
- e"'cos x +
C},
in x dx.
C}.
is the whole integrand
6.4
Integration by Parts
Here we use
u= lnx,
du= - dx,
x
dv = dx,
277
v = 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:
inx dx
uv
J
--J1
v du
= xlnx
x Inx
J
x
dx
dx = {xlnx - x +
C}.
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
you ought to be able to solve each of the problems below on the first try.
Each answer should be checked by differentiation.
1.
4.
7.
10.
13.
16.
19.
22.
J ln2xdx
J
J
J
J
J
J Tan1xdx
2.
xe axdx
5.
ea'" sin x dx
8.
11.
x3e"'dx
14.
x3 ln x dx
17.
20.
J
J
J
J
J
J
J xTan-1xdx
In (x2) dx
3.
x sin axdx
6.
eax
9.
COS
xdx
12 .
x2sin xdx
J
J xcosaxdx
J
J
J2
J
J
(ax)e"'dx
X " COS X dx
9
x2 ln xdx
15.
Sin-1xdx
18.
21.
'
x In2 xdx
Sin-1(2x) dx
xex sin x dx
In" xdx
= x
In" x
lnn-l xdx.
23.
Find
24.
ln3 xdx.
By n
1 applications of the
6.5
278
J sin In x dx. (Here you should survey the situation, decide on the most promising
procedure, and then proceed with faith.)
J
where
and m are any integers, positive, negative, or zero, and integrals of the forms
and
where n
0 and
secnx tanmxdx,
cscnx cotmxdx,
difficulty.
(1)
and
+ sin2x
sin2x cos3 x dx
J
J
Since
cos2x = 1 - sin2x,
1,
{t sin3x
sin4 x cosxdx
t sin5x + C}.
sin"x cos2k+lx dx
f
J
sinn.x(cos2x?cosxdx
sinnx(l - sin2x?cosxdx .
6.5
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).
f sin11
cosm x dx,
(2)
J sin" x cosm
and n
x,
and even.
(3)
sin2 x = cos2
(1 - cos2 x)
2 cos2 x
1.
+ cos 2x
2
Similarly,
cos 2x = (1 - sin2 x)
and
sm 2 x =
sin2 x
cos 2x
2 sin2 x,
Making these substitutions in the integrand, we get a form in which the exponents
are divided by 2. For example,
sin2 x cos2 x dx
cos 2x
1 +
cos 2x
2
dx
6.5
280
1 - cos 4x
dx
4J
tx - t cos 4xdx
When the exponents are large, this method is tedious, but at least we know that it
will work.
(n positive).
For
1,
we know that
Jtan xdx
For n
(4)
{-In jcos xi + C}
(4a)
2,
Jtan2xdx
{tan x - x + C}.
(4b)
and so
Jtann xdx
1
-- 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
This is like (4). For n
cos x
-.-dx
Jcot x J sm
x
=
For
(n positive).
(5)
1,
.
{In Ism x I + C}
(5a)
2,
csc2x. ) Thus
Jcot2xdx
{-cot x - x + C}.
(5b)
6.5
For
281
> 2,
1) dx;
and so
Icotn x dx
1
-
(5c)
By repeated applications of (5c), we can reduce our integral to one of the forms
(5a) and (5b).
fsec" x dx,
For
n =
2, J sec2 x dx
n =
2k, k >
(6)
1,
When we expand (1 + tan2 x)k-l by the binomial formula, we get a sum of integrals
of the form
We integrate each of these by the power formula and add the results. For example,
/1 =
n =
2k, k >
2,
fcsc2 x dx
For
fcscn x dx
This is like (6). For
{-cot x + C}.
1,
(7)
282
6.5
f
f
For
n = 1,
cot' x( -csc2
seen
xdx,
x) dx =
J uidu.
(8)
we found that
secxd x
secx(sec
sec
+tan
+tan
x)
dx =
fdu-u
where
sec
u =
+tan
du =
x,
(sec
x tan x
+sec2
x) dx.
Therefore
fsecxdx =
For
x dx
= (1
+tan2
x)
sec
xdx,
because the second term fits no standard form. The solution is obtained by integrating
by parts. We have
Let
seen
u =
xdx =
secn-2
x,
secn-2 x sec2
xdx.
sec2
x dx,
dv =
=tan
x.
This gives
seen
xdx =
=
f udu = uv fvdu
-
secn-2
:; secn-2 x tan
secn-2
f
x - (n - 2) f
x - (n - 2) f
tan x -
x tan
(n - 2)
secn-2
x tan2 xdx
secn-2 x(sec2
x - 1) dx
seen
(n - 2)
xdx
sec"-2
xdx.
6.5
283
secn-2 x tan x
and
seen
x dx
1
=
--
n-1
secn-2 x tan x +
n-2
--
n-1
secn-2 x
dx.
There is a similar reduction formula which works for odd powers of the cosecant:
cscn
x dx
1
=
--
n-1
cscn-2 x cot
x +
--
n-1
csc"-2
n-2
cscn-2 x
x, dv
csc2
dx.
(9)
x, and proceed
dx.
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.
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.
4.
7.
10.
J
J
J
J
2.
cos2xdx
5.
cot4xdx;
8.
sec4 xdx
11.
J
J
J
J
3.
6.
tan5 xdx
9.
csc4xdx
12.
J
J
J
J
sin2 x dx
tan4xdx
cot5 xdx
sec3 xdx
13.
x dx
6. x x dx
x 2x dx
19.
22. 2x 2x dx
25. --dx
x
2 . J x dx
30.
J sin
sec
J sin2
cos
sin2
sec
f sec5
17.
J cos
csc
f csc
sec
tan
14.
J cos3
sin
f tan
x dx
x x dx
21.
2x dx
24. I --dx
x
1
27. I --dx
x
x dx
x x dx
20.
2x dx
23. x x dx
1
26. I --dx
x
29. x1 x dx
f csc3
6.6
284
x dx
J sin
sec3
cos2
sin4
cot
A sinn-l
18.
sinx
tan
J csc5
f cos2
cos4
csc
15.
x x
cos
+BJ sinn-2
x dx,
where A and Bare constants, expressed, of course, in terms of n. Derive such a formula.
31.
[Hint: It is no use trying to do this merely by the use of the elementary trigonometric
Derive it.
6.6
J cos"
x dx
A cos"-1
x x
sin
+ BJ cos"-2
x dx.
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
(We are using a different dummy letter in the second problem, for reasons which will
{tx3 +
xsino
C}
,,_,sin o
C}.
Integration by Substitution
6.6
The
( !)
285
at the bottom indicates that the equation in the bottom line is the final
= {F(x)
ff(x) dx
l
l
ff(u) du =ff(u(O))u'(O) d() = {F(u(O))
X-+U(O)
X-+U(O)
(!)
Jf(x) dx,
C}
C}.
Jf(u) du.
ff(u(e))u'(e) d(),
Jf(x) dx.
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
consider
.J 1
x2
x = Sine,
dx
becomes Sin'()
de,
We then
dx
Sine, then
x.
the square
by the function
following diagram:
J
J
dx
.Ji
cos
.J1
=?
x2
X-+inO
e de
Sin2 e
(Query:
J.J1
e de
Sin20
_
_
cos
J .J
e de
cos2e
1 d()
= {O
C}.
do we need to consider?)
286
6.6
fJldx1
f
X2
(I)
x-Sino
cos e de
J1
- Sin2e
C}
{Sin-1 x +
o-Sin-1x
{0 +
c}
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
ff(u) du ff(u(O))u'(O) dO
x-u(8)
{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.
dx
x2J1 - x2
_?
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
dx --)-Sin' e dO
This gives
f J dx
x2 1 - x2
(Throughout,
-Tr/2
--)- f
cos e dO
< e<
Tr/2;
2 '
cos e de.
Sin2 0 cos e
csc2 0
dO
{-cot 0 +
fh
x2 1 - x2
{-cotSin-1x +
C}.
C}.
6.6
Integration by Substitution
287
Here
Sine= x,
.
cot Sm 1
_
Therefore
x=
e=
cot
dx
=
x2J1-x2
Sin-1
x;
1 - x2
e= -k = J
-'----x
x
{- J
x
+ c
Note that all the trigonometry has cancelled out of the problem. Our answer checks:
-x
DJ = l. x
- J1 - x2
x
x2
J1-x2
=
-1
1
.
[-x2-cJ1-x2)2J=
x2)1 - x2
x2)1 - x2
dx
Jx2J1-x2=(
l
J e de =
-.. e
xSinO
csc2
- x2
The substitution
integrand is
Example 2.
J1
Consider
Sin
{- }
x
I
+ c
oSin-1 :v
{-cot
e+
C}
In other cases,
x-...
Tan
288
6.6
(- 2
x -+ Tane
This gives
!!)
dx -+ sec2ede.
2 '
Therefore
JVl
Jsec3ede= fsecnede
(n = 3)
1
= -- secn-2etan e+
n-1
n
2
n-1
- Jsecn-2ede
t secetane+ t secede
x.
This gives
x.
Integration by Substitution
6.6
In the figure, -7T/2 < e < 7r/2, bute may be positive or negative. We take
This gives
x =Tane,
e = Tan-1 x,
r = sece = secTan-1 x,
so that
secTan-1x
OP
289
1.
=.Ji+x2
f.J1 +x2dx =
.J 1 +x2+ x
> 0
As before, we sum up in a diagram the process by which the problem was solved:
I.JI+x2dx
l
J 8d 8 =
x-Tane
sec3
e-Tan-1 x
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
l
0) {G(u-1(x)) + C}
x- 1<(6)
Jf(u(8))u'(8) de =
e- u-1(x)
{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:
290
6.6
G'
f(u)u'
G'
Therefore
x.
G'(u-I) Du-1.
f(u)u'. Therefore
G'(x)
for every
=f
D[G(u-I)]
By hypothesis,
D[G(u-I)]
=>
G'(u-I)
f(u(x))u'(x)
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))
because
u(u-I(x))
x for every
f(u(u-1))u'(u-I) Du-I
f,
Therefore
x.
u'(u-1) Du-I
u'(u-1)
--
u'(u-1)
D[G(u-1)]
Now
u' (u-1)
_,.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
1.
4.
7.
10.
13.
16.
19.
22.
J
J
J
J
J
J
J
J
(1 - x2)-3!2dx
2.
dx
dx
x(l + x2)
5.
dx
x2(1 + x2)
8.
dx
x2v' x2 - 1
x2dx
14.
Vl - x2
x2v'l
xv'l
11.
x2 dx
x2dx
x2v' x2 - 1dx
17.
20.
dx
v' x2
J
J
J
J
J
dx
dx
xv' x2 - 1
xdx
+ x2
xdx
x3dx
v'l - x2dx
+
12.
15.
v'1 - x2
x3 v' 1
6.
9.
v' x2 - 1
3.
x2dx
18.
21.
J
J
J
J
J
J
J
dx
v' x2 - 1
x(l - x2)-3f2dx
Vl - x2dx
x2v'1 - x2dx
x2dx
x2
1 +
(1
: x2)3dx
x3
dx
x2
Algebraic Substitutions
6.7
Jb f(u(O))u' (0)dO
iu(b)
ida>
whether
u is
b) Show that if
291
/(x) dx,
invertible or not.
u is invertible, then
u-l(c)
24.
G'
6.7
f(u)u'
D[G(u-1)]
=>
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
with.
J-::: =x2= == d x.
.J2x + 1
This does not fit any of the standard forms that we know about. There is no reason to
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
tx
- u(t)
x2 dx
J.J2x +
Ht - 1 ) .
Ht - 1),
dx-+ u'(t) dt
t dt.
-+ J t(t -
1)
.Ji
dt
! J t2 - 2 +
8
.jt
dt
+ C}.
292
6.7
To get the answer to the problem which we started with, we use the inverse substitution
x2 dx
-J2x+ 1
x2
J-J2x+1 dx
lx1i(t)
J
t2 - 2t+ 1
r
8v t
= {G(u-1(x))+ C}
<1>
dt
{G(t)+C}
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
x2 3
y2=2x+l,
Thus we use
x = t(y2 - 1).
-J2x+ 1
; ay =i cy4
J
_,..
y,
2y2+ 1) dy
Algebraic Substitutions
6.7
293
{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
x---+
Tan
x2
This works:
J-Jx2dx
+ 1
---+
e de
sec e
sec2
sec
e de ,
2 of Problem
Set
6.6.
z2
x
---+ u(z) =
dx
,/ z2
\ z
This gives
dx
-Jx2 +
__..
J!
z
,Iz2
+ 1,
dz.
1
x2
dz
1
,
J-J__:!:___
z2
6.6.
dz
1 + ./:'
294
6.7
1 +;-; t,
we need
t.
To get
z (t - 1)2
This gives
J;=t- 1,
1 +;;=t,
z = (t- 1)2
- Ji :=J:Z J
2(t - 1) dt,
2(t- o dt = (2 - ) dt
t
t u-l(z) = 1 +)-;,
gives the final answer
(Query:
{2(1 +,Jzj
2 In (1 +-Jzj + C}.
"1"
This is probably the most efficient solution. If we hadn't thought of it, we might
have tried
which gives the substitution
,J; = t,
z u(t) =t2,
dz u'(t) dt = 2t dt,
l
Dividing
1 +t into t,
Therefore
dz
2t dt
J +Jz J1+t.
we get
t
1
--=1---.
l+t
l+t
6.7
Algebraic Substitutions
295
u-1
(z)
.jz,
getting
I 1 +dz.J
--- =
r + C}.
r - 2 ln (1 + vz)
{2vz
(Is this really the same as the previous answer? Why or why not?)
We have used the substitutions x---+ Sin8;
Tan8, and
x---+
x---+
Sec() to handle
and
Slight variations enable us to take care of more general cases, involving
For example, to find
.Ja2
- x2,
.Jx2
Ja2 - x2dx
(a >
- a2.
0),
a2(1 - Sin28),
Ja2 - x2
a cos().
Sin8, so that
() = Sin-1 .
a
Thus
Ja 2 - x2dx
J
J
{:
a2 cos 2fJdfJ
a2 t ( cos 28 +
(a cos8)a cos8dfJ
---+
\in2fJ +
Now
sin 28
2 sin()cos8
This gives
.Ja2 - x2dx
tx.Ja2 - x2
x---+ a Tan 8,
! Ja2 - x2
a a
Sin-1
+ c} .
1) dfJ
e
+ c .
296
to get. rid of
6.7
to get rid of
Sec
e,
-J x2 - a2.
dx
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
x ---+ u(t)
1
=
-1
dx ---+ - dt.
t2
This gives
fx2(x +
1)
---+
---+
J(
{-
-1+
+ C}
Tan-1
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}
1x11(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
297
7.
10.
13.
16.
19.
22.
25.
28.
31.
J
J
J
J
J
J
J
J
J
J
J
dx
2.
V:x-)3
(I +
(a2 + x2)-af2dx
5.
dx
8.
l + x
z3dz
11.
vz + I
dx
14.
l + vx
(I +
vx)3vxdx
17.
dx
20.
Yl + e2"'
x Sin-1xdx
23.
l
dx
(1 + vx)2
26.
29.
dx
v1 +fix
dx
32.
x3(1 - x)
J
J
J
J
vx
(1 + Vx)3
3.
dx
dx
6.
v1 + e"'
dx
9.
(1 - x)2
z3dz
Vz2
12.
dx
J
J
J
J
J
J :
J
x4(x
15.
I)
dx
18.
Sin-1xdx
21.
Tan-1 xdx
24.
1
dx
vx(l + vx)2
27.
dx
e"')2
30.
(l
33.
x2 ln xdx
J
J
J Yvx
J
J
J
J
J
J
J
J
(a2
x2)-af2dx
dx
v1 - e"'
dx
+ 1
dx
YI + 'x
(l - x2)4dx
dx
(1 + e"')4
xln xdx
xTan-1 xdx
1
l + fix
dx
dx
v 1 + e3X
x2Tan-1xdx
Ja2
x2 and
Jx2
x- + x +
9
Jax2 + bx
x- + x + i + !
?
c.
For example,
(x + t)- +
?
(.)3)2
2
Therefore
dx
v x2 + x +
dx
'
(x + t)2
JJuzdu+
a2
+ (./3/2)2
'
6.8
298
x2 +x = x2 +x +t - t = ex +t)2 - ew,
and so
dx
.Jx2 +x
dx
J
J du
=
.Ju2 - a 2
=?
x2 - 1
We try
x--+ Sec()
so that
ex> 1)
()
d(),
giving
csc ()
Now
1
x+ 1
x
x+ 1
--+
=
=
x -1
.Jx2 - 1
.Jx2 - 1
.Jex + 1)ex - 1)
-
1 c}
J { 1
x2 - 1
= t 1n +
x +1
ex> 1).
ex> 1).
We check by differentiation:
( l : I )
D tin
--
x -1
1
x2 - 1
1
1
1 ex + 1) - ex - 1)
=-
-
2 x+1
2
ex - l)ex + 1)
--
6.8
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
_l.
2_
_
2_ +
x - 1
x +1
__
x2 - 1
Proof
_
_
x - a
_B_
+
x - b
(x a, 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
A=
ae+d
be+d.
B=
b - a
a - b'
A+B
x - a
x - a
x - a
-- +-- =--.
1. If a b, then
ex +d
(x - a)(x - b)
ae+d
a -b
be+d . _1_
+
.
x - a
b - a
x -b
1
_
_
fcx 2x
-
- 5)
=
?
(x
------ = -- +--
- 2)(x -
5)
x - 2
300
6.8
The only problem is to find out what they are, numerically. We first write
x =2 and for x = 5.
Therefore
1 =B 3;
(-3),
B =t.
A= -t,
(x - 2)(x - 5)
=-
1,
l
_!_ .__
l __
1
+ .
3 x-2
3 x-5
would not have been valid. To see this, consider the following analogous procedure:
"Problem."
and
sin
b such that
x =ax+ b.
Therefore
b =0, and
sin
Letting
x =7r/2,
a =2/7r.
and
x =ax.
we get
l=a2:
2'
Therefore
.
(?)
Sill
X
.
2x
=-
for every
(?)
This is wrong: in fact, our formula is correct for only three values of
0 and
x = 7r/2.
x,
namely,
=>
The statement
(1)
=>
( 2)
sin
is a linear function
sin
(1)
2x/7r.
(2)
(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+ 1
x+ 2
(x+ l)(x+ 2)(x+ 3)
x+ 3
--
9 = -B,
19 =2C,
B= -9;
C=:!ll-.
301
6.8
Theorem 2.
p(x)
(x - a)(x - b)(x - c)
_
_
x - a
_
_
_
_
(3)
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:
1.
4.
7.
8.
9.
12.
15.
J
J
J
J
J
J
J
dx
2.
x2+2x+5
dx
5.
x2+x- 4
J
J
dx
v'x2+2x+5
dx
Yx2+x -
J
J
3.
6.
dx
x2 - 4
dx
v2 - x2
dx
>12 - 2x - x2
dx
v -2x- x2
dx
10.
x2+ 6x + IO
dx
(x - l)(x -
13.
2)
J
J
dx
v-x2-
6x +10
x dx
(x - l)(x
11.
14
- 2)
J
J
d
x2
6 +10
dx
x(x - l)(x - 2)
x dx
x(x- l)(x-
2)
18.
20.
1
(x - I)2 (x - 2)
1
x2 (x
23.
Find
25.
Find
(x - 1)2
B
C
+ -+
x- 1
x - 2
--
A
x+ I
--
dx
(x+ l)(xdx
x(xz + I)
dx
x(xz+1)2
2)3
B
(x-
22.
24.
2)3
17.
Find
19. Find
--
J
J
J
A
B
C
D
2 +x
- + (x - J)2 +x- I
l )2 = x
21. Find
c
(x -
2)2
J
J
(x
1x
x2(x
1)2
D
(x - 2)
---
A
Bx+ C
=-+
2
x2+ I
x
x(x + I)
I
x(x2+1)2
26. Find
Dx+ E
Bx+ C
A
=-+
+
x
x2+1
(x2+ 1)2
sine
d8
1 + cos 8
2)
302
27.
Given 8
28.
6.8
->-
u(x)
2 Tan-1 x,
sin 8
_,.
d8
?
_,.
u'(x) dx
cos 8
->-
2 dx
1 + x 2'
But when you see the answer, you may be able to think of a simpler method of solution.)
Find:
29.
32.
35.
f
f
f
d8
1 +sin 8
d8
sin 8 +cos 8
d8
1
sin 8
30.
33.
36.
f
f
J
d8
31.
cos 8
d8
sec 8 +tan 8
d8
2 +cos 8
34.
37.
f
J
J
dx
x2 +6x +9
sec3 8 cot 8 d8
d8
cos 8
sin 8
The
7
7.1
Definite Integral
i,
= x0 < x1 <
< xi <
X;+i
<
<
Xn
b.
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.
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
7.1
304
Definition. Let
be a net over
p(N)
INl->O
L.
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.
interval
Let f be a function, on an
Definition. Let f be a
function, on an
[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
INI < 0
0 <Ix - x01 <a
Then
=>
lf(x) - LI < E .
limf(x)
=>
lp(N) - LI <
E.
Then
Jim
L.
p(N)
INl->O
x--+xo
L.
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)
(See the figure on p .
I P;-1P;.
i=l
P;_1 to Pi looks
Yi=f(xi)
Yi-I= f(Xi-1)
-+---'---.L__-'---x
Xi-I Xi
Xi
like this:
7.1
305
Thus
[i (:YJ Lix;.
+
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
Lixi
= f'(xi)
=1
p(N)
ton, we get
n
=l
=l
=I
;-1 =I J1
i P P;
i
+ [f'(xi)]2 Lixi.
The problem is to find out what happens to the sum on the right as INI
-+
0. We can
For each
g(x)
J1
In the figure,
g(x),
for
n.
+ [f'(x)]2.
g(xi) Lixi.
The sum of these areas is
i=l
g;
+ [f'(i;)]2 Lixi
and I g(xi)
p(N).
306
7.1
under the graph of g, when the mesh of the net N is small. That is, we ought to have
INI
=>
= .J1
lbg(x) dx.
a
[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.
Thenj'(x)
Let
f(x) =
= 0, and
Oxl.
= Jf1o .J1
02 dx
1,
f(x) = kx,
= 1,
Oxl.
Thenf'(x)
= k, and
L
= .J1
k2 dx
= .J1
k2,
= .J1
-x
-x2,
Ox.J2.
-
Then
f'(x)
1 +
.J1
[f'(x)]2
[f'(x)]2
=1
- x2 + x2
1 -x2
x2
x2
'
-x2
1
_'
_
x2
7.1
307
and
L
i,1;10
"-J1 + [f'(x)]2 dx
i\!212
o
Sin-1
-Ji
dx
-J2_
x2
!!. .
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)
/'(x)
2x
x2, 0 x
1 + [j'(x)]2
1.
f-J1
Now
Here
=
+ 4x2 dx.
1 + 4x2,
This is
2x.
ls+
2
t In c2 +
-Js).
The answer suggests that no method would have made the problem look easy.
I.
f(x) = x312,
x, 0 x 7T/4
x, 7r/4 x 7r/2
*4. f (x) = Jn x, 1 ;;; x ;;; 3
5. f (x)
1 + tx3i2, 0 ;;; x 4
2.
3.
f(x) =
f(x)
x ;;; 2
In cos
In sin
*6. f (x)
7. a)
x
= e ,
/(x)
t(ex
e-
"'
}Ce"
e-x .
He"' -
e-" ,
But the
) cosh x
For the definitions of the hyperbolic functions sinh and cosh, and the
4.11.)
308
8.
7.2
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.
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.
p
10.
[a, b], and let x be any point of(a, b). Let mv m2, and mbe the
[a, x], [x, b], and [a, b]. Show that mis between
m2
f(x) - f(a)
x-a
m.)
f(b) - f(x)
b-x
and
m
f(b)-/(a)
b - a
7.2
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
bg(x) dx
309
Lim
2 g(.Xi) xi.
JNJ...,Oi=l
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
y
f(b)
[a, b].
y
------
f(b)
x0=a x1
I
iI -II
-
f(a)
--+-x
Xn=b
x3
a=xo xi xz
(a)
f(a)
On the interval
[a, b] we
N: a
I
I
I
---
_____
IX3
_J
Xn=b X
_J
I
I
I
_
.J
(b)
form a net
x0 < x1 <
< Xn
b.
N cut up the interval [a, b] into little intervals [xi_1, X;]. For
m; be the minimum value off on the ith interval [x;_1, x;],
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
from 1 to
n,
let
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
310
7.2
(1 i 11) .
The sequence
is called a sample of the net
N.
L (X)
L f(xi) 6.xi.
i=l
INI
Let
max {6.x;}.
be the region between the graph off and the x-axis, from a to b.
y
Theorem 1.
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,
and
B ?;;_ D,
-B -D,
- B C - D,
Theorem 2.
1s1-o
0.
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.
7.2
y
311
--
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
j(a) --,
I
I
I
I
I
This difference is
n
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
and so
S(N) - s(N)
Theorem 3.
-+
0 as
IN/
-+
0.
If f is continuous, and
Jim
IXl->O
.L (X)
[S(N) - s(N)] =
approach a limit, as
0,
INI
-+
0.
l1Yl->O
Proof
L (X) =
k.
bounded above.
The numbers
s(N) are
sums.) Let
k
Consider an interval
sup
{s(N)}.
i,
m; f(i;) M;.
312
Since
7.2
That is, every sample sum lies between the lower sum and the upper sum.
We are now almost done. Given
INI < 0
By Theorem 3 there is a
=>
Thus when
E.
INI < o, the interval from s(N) to S(N) has length less than
s(N) 2.::(X)
E.
S(N)
fa f(x) dx
lim1N1-?0
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
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
(a x b),
7.2
313
then
m(b
Proof
[a, b],
and let
X be any sample of N.
m f (xi) M
Then
for every i.
Therefore
we get
i=l
But
i=l
'L m D.xi
i=l
m 'L D.xi;
i=l
'L M D.xi
i=l
M 'L D.xi,
i=l
and
n
'L D.xi
i=l
(Why?) Therefore
m(b
a.
a);
and this holds for every sample sum, over every net N. Therefore the same inequalities
hold for lim
'L (X),
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.
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.
314
3.
7.2
Theorem A
4.
ff
j(x)(b - a).
*5.
(x) dx
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.
L [F(x;) - F(xi_1)]
i=l
As INI
along.
*9.
___,.
0,
___,.
L j(x;) D.xi.
i=l
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
cannot be the derivative of any other function f
7.3
*11. Theorem.
315
xl-
x:-a+
More precisely, for every
" > 0
a *
12.
a)
A function
c5 <
there is a
x1
<
c5 > 0
<
x2
such that
<
+ c5
<
"
> 0
ct:> .
0.
IN/o
13.
a)
[x - x'[
< c5
I/(x) - f(x')[
"
> 0
there is a
c5 > O
< "
(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 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
I
I
I
) ---- ---/
/
V =abc.
V = 7rh(r2 - s2).
316
7.3
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.
y
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
Let
net
vS
vS
over the interval [1, 2]. For convenience, we use equally spaced points, so that
xi
xi-l
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
=
7.3
y
317
7T (-1-)2 x
xi-1
and so the total volume of the circumscribed solid Sn is
n
n
1 2
vSn .2 vi .2 7T
x.
i=l
i=l X;-1
V
7Tr2 x
'
(-)
g(x)
1
7T
x2'
(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
and so IN!
as
INI
co.
{ 7T dx
J 1 x2
.! ,
n
Therefore
lim vSn
n-+ 00
- Jx 21
.'.'.:: .
(1)
If we use inscribed rectangles, and rotate them about the x-axis, then we get an
inscribed solid S, with volume
n
2
vS .2 7T ( 1 ) x.
=
i=l
X;
n- cc
2 7T
2 dx.
1 X
7T/x2 Therefore
(2)
Therefore the volume vS of Sis squeezed between the volumes of the inscribed and
circumscribed solids:
for every n,
318
7.3
and so
vs =
f
Ji
'!!__ dx = '!!. .
2
x2
(3)
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
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.3. If S and S' are measurable, and S' lies in S, then vS' vS.
V.4 (The squeeze principle). If (a) Si. S2,
(b) S{, S, ... are measurable solids lying in S, and (c) lim,H00
vSn
L = limn-ro
vS,
vs=
L.
Using V.1 through V.4, we can show that the method of disks, which we have
'
f (x) = I/x,
works for every function
f which is
0 and continuous.
f
M;
'mi
----
I
I
I
I
I
I
I
I
I
I
.x
--lf---x, -1 -i
-
---x
---
7.3
319
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
n
vSn = I 7TMi!lx.
i=l
vS vSn
for every
n.
g(x) = 7Tj(x)2,
and vS is a lower sum of the same g. As
vSn _.
f7rf(x)2 dx,
n-. oo,
vs
JNI-. O;
--f7Tj(x)2 dx.
vS
f7Tj(x)2 dx.
y=vx
R
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,
Therefore the volume is
1 7T
[f(y)]2 dy
Oyl.
1 7TY4 dy
1
0
7T
=5
320
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.
y
3.
4.
Same, forf(x)
5.
6.
Same, forf(x)
7.
Let R
volume.
8.
Same, for R
9.
Same, for R
v':X
sinx, 0 x .;,,
x312, 0 x 1.
cosx, -1Tj2 x 1T/2.
Find the
321
The General Method of Cross Sections, and the Method oJ.' Shells
7.4
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)
ro.
Give a
1/x2 from l to
ro
Given a solid S in
space. Suppose that we can calculate the areas of the cross sections perpendicular to
the x-axis.
y
'\
'I
, ,
11
,,
I;
\
I
I
I
I
For each
from
I
I
I
I
I
I
x.
This gives a
322
7.4
We then have
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,
fA(x) dx.
x, for 0 x 1.
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.
y
/
/
The cross-sectional areas perpendicular to the y-axis are given by the formula
A(y)
Therefore the volume is
V
x2
y.
1
=
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.
7.4
323
,,.
,,.
I
I
I
I
I
-2
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)
Therefore
V
J:
?TX2
A(y) dy
?T(Cos-1 y)2
J:
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.
y
?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;
2x; x
- x2) cos X;
324
7.4
L 27TXi
cos xi
n
Llx - 7T Llx L cos xi Llx.
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
L 27TX; cos X;
For the same reason,
Llx
L cos X;
-+
7T Llx
L cos xi
rr/2
27Tx cos x dx.
Llx-+
Therefore
rr/2
cos x dx.
Llx
-+
7T
[ r.12
J cos x dx.
o
Thus the entire second sum, in the expression for vSn, drops out when we pass to the
limit. Therefore
vSn
and
V
-+
[ "12
J 27TX cos x dx,
o
[ "12
J 27TX cos x dx
o
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.
y
If the width of the region is given by a function h(x), then the volume of the ith
cylindrical shell is
vi =
=
=
7.4
Therefore
n
vn
L vi
325
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;
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;.
same as the altitude and the thickness of the shell; these are
the volume of the prism is exactly
mation to
V;
when
Llx
27Txih(x;) Llx;
h(xJ
and
Llx.
Therefore
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.
Jx
Therefore
V
cos
x dx
{x sin x +
[27T(X sin x +
cos
x)]12
cos
x +
C}.
7T2 - 27T.
326
7.4
y
-_Ll
I
I
+-
resulting solid is
f27T(x
-'--'-.__--X
----'-
+ 1)(-x + 5x - 6) dx
f27T(-x3
[ x4
27T - 4
f277(6 - x)(-x2
77T/6.
+ 4x2 - x
+
4xa
6,
x2
6) dx
]a
6x. 2
t7T.
the volume is
+ 5x - 6) dx,
(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
0
;;; x ;;;
x2,
f127TX x2dx
Jo
'!!..
The horizontal cross section at height y is the region between a circle of radius 1
and a circle of radius
x =
JY.
as before.
Therefore
fA(y) dy f [7T
=
7T - 7T
fly dy
0
7T
2
12 - 7TY] dy
= - '
7.5
327
1. Let R be the circular region with center at (5, 0) and radius 2. R is rotated abo1t 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
a,
a.
When R is rotated
about the line L, the result is a solid torus. Find its volume, in terms of
3.
and b.
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.
rotated about the y-axis.
1 x
lnx,
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
7. a) The region described in Problem 5a is rotated about the y-axis. Find the volume, by
the shell method.
Let C be the cylinder with the y-axis as its axis of symmetry, and radius 1. Let S be the
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
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
revolution.
surface of
[a, b]
Here the
328
7.5
.. 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
= Trr2. ) From this we can get a formula for the lateral area of a frustum
answer Trrs
of a cone. If the larger cone (with slant height s2) has area A2, and the smaller cone
7.5
329
7Tr2s2 - 7Tris1
Evidently
If Si
then
7T(S2
s1)(r2 + r1) =
and so
21T
+ ri)
ri + r2 A
us,
--2
where
i(r1 + r2).
That is, the area of the frustum is equal to its "average circumference"
slant height D..s.
f=
2Ttf
times its
Consider now the surface of revolution obtained by rotating the graph off
about the y-axis. We take a net
N: x0, Xi,
[a, b],
xi-l xi,
. . . Xn,
b - a
Ax= --= INI
n
for each i. For each i, let Pi be the point (xi, f (xi)) . These points determine a broken
xi - xi-l
line Bn which is an approximation of the graph off When Bn is rotated about the
revolution off is
A= lim An,
INl->O
if the limit exists.
INI -+
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
7.5
330
Pi-1Pi
where
X;_1
xi )
f xi)
1 + ( c :c -l Y Llx
.J1 + f'(i;)2Lix,
Here
i i is
I ai iI=l 27TiiPi-1Pi
i=l
the midpoint of
i;
[xi_1, x i],
and
i i for each i,
g(x)
I 27Tii.J1
i=l
+ f'(i;)2Lix.
i; is somewhere
27Tx.JI + f'(x)2,
I a;
i=l
we know that
n
I a;
lim
noo i=l
I 27Ti;.J1 + f'(i;)2Lix,
i=l
ibg(x) dx ib27Tx.J1
=
+ f'(x)2 dx.
lim
I NI ->O
Now
because
i; lies
I ai
i=l
lim
INI ->O
I a;
i=l
xi __
x i,1
I-
on the interval
lai - a;J
[xi-1
ib27Tx.J1
a
+ f'(x)2 dx.
<Lix
'
2
X; ,
7T.J1 + f'(i;)2Lix2
7.5
331
Therefore
a; - ;a;
;
I ;Ia;
n
- a;I
+ f'(x;)2 .6.x2
= .6.x I '"-J1
f'(x;)2 .6.x.
i=l
Now
lim
.6.x i Tr-J1
INJ-+O
Therefore
i '"-J1
i=l
+ f'(x;)2 .6.x
b
= 0 r Tr-J1
Ja
.
+ f'(x)2
dx.
Let us try this formula on some problems to which we already know the answers.
y
y
b
f(x) = b - (b/a)x,
A
f'(x) = -b/a,
a
b2
r
= o 2Trx 1 + dx
J
a2
f(x) = -Ja2 - x2
(0 x a),
-x
f'(x) = I
'
'\/ a2 - x2
a2
x2
1 + f'(x)2 = 1 +
=
a 2 - x2 a2 - x2
---
a,
rotated
332
7.5
is
47Ta2
formula.
It is harder to find the area when we rotate a function-graph about the x-axis
instead of the y-axis.
y
[a, b],
[a, b].
D..x
b
=
--
INI.
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
where
a;
where
-J 1
X;_1 <
f'(x)2 D..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.
7.5
333
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
for each i
for each i
i 2rrl(xi)-./1
i=l
f'(x;)2 x
i 2rrl(x;)J1
f'(x;)2 x
i=l
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=
1)
If f(x)
f2rrl(x)J1
f'(x)2 dx.
By the
integral formula,
A=
- a ,
2)
A sphere of radius
Here
2x + 2lf'= 0,
Therefore
1
and
A=
!'2=
and
2
12
1
+ =
x
l
l'=
2
2 + x2
=
12
1 2'
a
f-aa 2rrl(x)J7
-- dx= f 2rra dx= 4rra2
-a
l(x)2
Let
Ca
a,
[ -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?
334
7.5
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
<
5. Find the volume of the solid obtained when the corresponding square
region is rotated
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
For each
a, let Sa be the area of the surface described in Problem 10, and let Va be the
a maximizes the ratio VafSa?
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
x
+y
7.6
as an answer to this one.
7.6
335
B.)
Given a
My= .L X;m;.
i=l
y
P2
-2
-1
--r---'---+-----_...x
I
I
-1
I
I
I
m2=1------2
----
--+-----____, :
P3
Pn
x1m1 + x2m2
2 +
(-2)
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
M
L Yimi.
i=l
.L m;.
i=l
(x, .Y)
such that
and
Mx
ym.
m.
That is,
336
7.6
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,
19 = ji . 5,
8 = x. 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.
y
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
Then
xi -
for each i.
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
7.6
,6.x decreases.
337
Let R be the region lying between the graphs of two continuous functions
/1 and/2, on an interval
h(x)=fix) - /1(x).
Then the
moment of
fxh(x) dx.
M,"=
and by definition,
Jc y
g1( y) x g2(y)},
w (y)=g2(y) - gi(y),
M.,
and
fyw(y) dy.
Mx=jiA.
y=Va2-x2,
Oxa.
a.
338
7.6
Here
Obviously
and so
Therefore
x =-a.
37T
By symmetry, interchanging
and y, we get:
y =
=
= -a.
37T
x0 is
defined to be
f(x - x0)h(x)dx,
Mx=xo
y0
Mv=Yo =
is
f(y - Yo)w(y)dy.
Theorem 1.
Proof
For any
Mx=x =
o
This is 0 for
x0
0 = My=g
x0,
x. The proof of the other half of the theorem is the same. In fact,
the equation
M"'="o = My
- x0A
2. If MX=Xo
0, then Xo
x; and if MY=Yo =
0, then Yo
ji.
Two points
P'
region or a curve.
7.6
p
pt
P'
h(x) is an even
with (-x)h(-x)
on the left,
function,
function, with
- [xh(x)].
M11
and x
h(-x)
h(x).
Therefore
0. In the figure
xh(x)
is an
Therefore
faxh(x)
0,
0.
y
--+--'-x
"'----'--'--..L.._
x0-k x0-t x0 x0+t x0+k
Proof
Mxxo
By symmetry,
rxo+k
(x
J xo-k
h(x0
- X0)h(x) dx
t)
h(x0
x0,
then x
x0.
r xo+k
(x) dx.
Jco-k efi
+ t)
efi(x0
- t)
339
- t)
[(x0
-efi(xo
efi must
- x0]h(x0 -
t)
-th(x0
+ t).
xo+k
t)
odd
7.6
340
Therefore
and
It follows that
x0
x.
0.
h(x0
t)
h(x0 + t).
This condition may hold for regions which are not symmetric, as below.
y
And interchanging
y0, then ji
y0
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
f27Txh(x) dx.
Therefore
V=
27TMv
27T.XA,
7.6
341
xA.
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.)
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.
graph of a function f, on an interval [a, b ]. As usual, we take a net over [a, b], with
equal subdivisions.
f/1
+ f'(x)2 dx.
342
7.6
Now
si
R::!
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
Definitions.
R::!
I xi-./1
il
+f'(x;)2 t>,.x.
Given the function f, with a continuous derivative f', on [a, b], the
fx-./1 +f'(x)2dx;
x0 is
Mv=110
ff(x)-./1 +f'(x)2dx,
f(f(x) - Yo)-./1 +f'(x)2dx;
:XL ,
x0 (or y =
y0)
then this
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.
y
r-1
I
I
I
I
7.6
343
For example, if we rotate about the y-axis, then the area of the resulting surface is
S
27Txl + f'(x)2 dx
27TM"
27TxL,
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.
(0, 0), (a, 0), and (b, c), a, c > 0. At each of these points
3.
ma
ss 1, 2, and 3
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.
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.
9.
b 0.
- a, c ,
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.
14.
a,
with
Let the arc A be the portion of the circle with center at the origin and radius
lies in the first quadrant.
15.
a which
(Here
16.
17.
= a
region is rotated.
344
7.7
x =
from (0,
For each
let Sa be the area of the surface described in Problem 19, and let Va be the
7.7
IMPROPER INTEGRALS
The definite integral is defined as a limit of sample sums as the mesh of the net
approaches
0.
f(x)
on the half-open interval from
Jx
to 1.
(0, l]
{x j 0
x
<
I}
On this
(0, l]
we can form a sample sum as large as we please, by taking the first sample point
close to
0.
Thus,
0.
Nevertheless, we can extend the definition of the integral in such a way that our
problem has an answer.
every closed interval
The function f
[a, l],
where
a > 0.
(x)
1/J
Therefore
H (1/J) dx is
well defined.
Improper Integrals
7.7
345
f(x)
1
-
Vx
f 7x
!+
; '
positive values.) In the present case, the limit exists and is finite:
f ;
Therefore
\-112 dx
[2x112J!
f1 d
lim [2
a-+O J a '\/ a-+O+
=
lim
+
2.fa]
2 - 2.Ja.
=
2.
There are similar-looking problems for which the limit is infinite. For example,
11 dx
.
1dx
-= hm
0 2 a->0+ 1a 2 '
X
1
-
-1 +
r 1 dx
a->0+ Ja x2
and so
lim
(a > 0),
oo
,
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 =
lim
sec x dx,
a.. rr 2 la0
/
where the minus sign means that a-+ 7Tj2 through values less than 7T/2.
l"0
346
7.7
--+---rr,,.__---- X
2
Now
As a-+ "TT/2-,
sec a=
1
--
cos
-+
+ tan
+ tan
In I sec a + tan
oo
oo
xJ]g
aI
0I
a j.
tan a
and
In 11 +
sm
=
cos
a
a
-+
oo.
and
a
-"12- xdx =
a
We use the same method to define and evaluate such integrals as
ico dx2
x
Here the integration is supposed to be carried out all the way to the right, starting
at x =
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
a-co
if the limit on the right exists. For the function
ladx
g(a) = 2'
x
the limit exists and is finite:
j'adx2 = [-.!]a=_ l + 1;
x
a
x
and so
ad
.!] =
lim r : = lim [ 1
J1
-+oo
a x a-+oo a
Jim
r sec
Jo
00.
1.
2
X
2
X
l.
Improper Integrals
7.7
347
f(x)
1
2
00
dx
Jim
a--+oo
fa dx
1
Jim [In
a--+oo
x] = lim ln
a =
a--+oo
may be infinite.
oo
oo
Integrals of the type that we have been discussing here are called improper
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.
1 dx
.
a-+o+ a,Jx(l + x)
hm
1.
1m
b-+oo
fb
i
dx
.jx(l + x)
oo
is their sum.
(Any
l d: [-1]1
J-1 x x -1
=
-1
x =
o dx
x
-1
'
to
-2.
oo.
348
7.7
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
1 dx
= [In lxl]:.1 = ln 1
-1 x
ln 1 = 0
(?).
J-1
(
J
a-+O
1 dx
a-+O
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.
Here
+ 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.
J1
"'
=Jim
a-+00
r a:
J
1
= lim
a-+oo
i"'
1
dx
- =Jim
X
a-+ oo
(?)
oo
[- l]
or
= l.
- oo.
dx
- = lim lna =
X
For example,
oo.
a-+ oo
any
1 dx
.
-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.
Improper Integrals
7.7
349
loo dx
x2
oo
dx
J x3
loo dx
1 x1.0001
f dx
1x x
loo
2 (x - l)s dx
J000 xe-xdx
loo xdx
1 x4
f;' xn dx
loo dx
1 x2 x
f dx
loo dx
1 x0.9999
f dx1 0001
0 x.
f
1 (x - l)s dx
Loox2e-xdx
Investigate:
1.
4.
2.
1 +
5.
7.
10.
13.
16.
19.
20.
8.
11.
In
3.
6.
-v:x
9.
12.
14.
17.
15.
18.
-v:x
v'x(l
x)
1 +
Show that
l/x,
loo dx
1 xs
loo dx
loo dx
xlnx
f2 dx
0 xo.9999
fo00e-xdx
Joo dx
oo.)
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
represents
a finite limit; but in the cases where it does, you need not calculate the liP)-it.)
22.
25.
27.
30.
33.
*36.
loo-1--.,
dx dx
1 e
loo--dx
x2e-x
x
+
1 1 +
J00 e-xdx
-00
J00 e-x dx
-00/2
f0 xdx
x
Joo --dx
x2
csc
sin
..
23.
26.
28.
31.
34.
loo dx
2 x x
f-00oo e-x dx
1 +
loo
24.
In
1 +
v'
dx
1 +
Tan-1
e-x
because
lsin
dx
x(l x)
f;" e-x dx f= e-"'2
f:_1 e-x-dx1,
loo dx
2 v'xinx
/2tanxdx
f0
xi
Joo --dx
x2
x4 dx
i00 G - x) dx
i00 x2 xi dx
1
Joo
is continuous on
29.
32.
!sin
35.
1T
< oo,
1].)
7.8
350
n,
IJv';;;(n+iJ.-.
f 00 v
I 'Jv;;;
sm x2 dx <
b) Investigate
38.
v'..
__
sin x2 dx .
v' (n-1>.-
sin x2 dx.
oo ) .
(x,f(x)).
The
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
have to use infinitely many boxes (one for each
x)
But
closed interval
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
351
which
finite collections of boxes, getting another finite collection of boxes that covers the
whole graph. Therefore one or both of the halves of
be a bad half of
[a3, b3]
be a bad
..
Yo <
<
y1}.
bn - an
the length of the nth interval
b
=
[an, bnl
- a
2n-
--1- '
0.
for some
This means that
(x,j(x)).
Since
one
n.
net, mesh (of a net), upper sum S(N), lower sum s(N), and sample sum I (X).
Theorem
If f is continuous, and
lim [S(N)
[N[-+O
- s(N)]
f f (x) dx
INI
is defined to be limfNf-+O
->-
0,
0.
I (X).
complete the proof that continuous functions are integrable, is the following theorem:
352
7.8
Theorem 2.
INJ-+O
Proof Let
>
0.
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:
y
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.
7.8
Theorem 3.
353
0. By Theorem 3 of Section
Some of the ideas in this proof are worth examining further. In Problem Set 7.2,
we gave the following definition.
Let f be a function on an interval /. Suppose that for every E > 0 there
is a o > 0 such that
Definition.
Ix - x'I
<
=>
I.
If(x) - f(x')I
<
E,
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)
2)
3)
of
A
A
A
the
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
354
as
INI
--+
7.8
I' = I 27Tf(x;)J1
+ f'(x;)2 ilx
i=l
x;
to show that
Iim II' -
JNJ->O
II
2TTJI + f'(x)2
E
Thus we need
o.
Let
rx;_1, x;].
< M,
for every
x.
M(b
a)
>
0.
Ix - x'I
<O
=>
lf x) - f(x')I
<M(b
- a)
Proof
INI
<o
<o
=>
Ii; - .x;1
=>
lf(x;) - f(x;)I
for each i
<M(b
Now
I - I'
for each i.
a)
i=l
and so
II
i=l
n
i=l
Therefore
INI
<o
=>
--
E,
7.8
355
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.
[a, b],
oo
).
3.
4.
oo,
on
( -co,
oo
).
(a,
b)?
6.
*7.
(a, b) .
a,
[a, b]?
of
[a, b],
let
F(x)
[a, b].
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.
Lipschitzian,
10.
Let
a1, a2,
an
Let f be continuous on
[a, b].
Show that
If
f(x) dx
I f
If(x)I dx.
Show that if f
8.1
TRANSLATION OF AXES
X2
<--+
p <--+ x,
R,
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.
and so
'
h.
This process is called translation. The origin is moved to the point h, and all the
other number labels are moved with it.
0
x=
1
x =
-h 1-h 2-h
h+l
h+3
3
Thus the old and new coordinates are related by the formulas
'
h,
x' + h.
Translation of Axes
8.1
357
y'
y
y=k
'
x=h
Suppose that we translate the coordinate system on the x-axis, subtracting
from every x-coordinate, and then translate the coordinate system on the y-axis,
subtracting
k 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
= 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
( by writing x'
to see, however, that the new origin must have old coordinates
nates
0, O; and from this we can tell which way the formulas ought to go.
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
As usual, (x,
given below:
y'
P=(5,2)=(2, 1)1
--
2 -------
+--=-t.._--,+-'-.._--- x
- 1,
0
2 3
4
5
/L--1
Q=(-1, -l)=(-4, -2)1
In the figure,
h =
3 and
point P, we have
x
= 5,
y = 2,
x'
= 2,
y'
= 1,
is correct.
Similarly, at Q we have
x
= -1,
= -1,
x'
= -4,
y' = -2,
358
8.1
so that
= (-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.
x, y)
---- P=(
--\
"'-.__
-+---- x
0
1
2
4
5
16
3
D---1
________________
n_
M
__
FP =MP.
Algebraically, this says that
We know, however, that a parabola with a horizontal directrix and vertex at the
origin always has an equation of the form
y = ax2
= (3, 1).
This means
that we should translate the axes so that the new origin becomes the point
O'
This gives
or
x and the
constant term. Here we knew in advance where the origin ought to be for the equation
Translation of Axes
8.1
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
x= x' + h,
in general form. This gives:
0.
Certain facts are now obvious: (1) We can't get rid of the term x'2, by any choice
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,
of
h2 - 6h + 17 - Sk.
We can therefore get rid of the
becomes
9
which is 0 when
- 1 S + 17 - Sk'
k= 1.
or
- Sk,
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.
Find a translation which removes both linear terms from the equation
x2 + y2 + x + y - 2
0.
Find a translation which removes both linear terms from the equation
2xy - x + 3y - 2
5.
0.
Find a translation which removes both linear terms from the equation
x2 + y2 + 4x + 2y + 1
0.
6. Find a translation which removes both linear terms from the equation
x2 + xy - 3x + 2
0.
Why
360
7.
8.2
8.
0.
Find a translation which removes both linear terms from the equation
x2 +xy +y2 +x +y +1
9.
0.
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
l 1.
0.
0.
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.
ch +dk
e,
always has a solution. (Simply start solving it; at some point, you will need to assume
that ad - be ;;f. 0.)
13.
0.
Show that if B2 - 4AC ;;f. 0, then there is always a translation that eliminates both of
the linear terms.
where B2 - 4AC
0, but where the linear terms are absent to begin with. Examples?)
14.
15.
b2 y2 +h1Y +ho
3
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.
The Ellipse
8.2
361
We tie the ends of a string to the thumbtacks, in such a way that the length
F'.
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'.)
;>6
F'.
c,
FP
F'P
(For a
to
F';
= c,
and for
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.
Proof
FP
P is
+
F'P
2a.
FP
FP',
and
F'P
F'P'.
Therefore
P to P'.
FP' + F'P' = 2a,
By elementary
and
P' is on
the
ellipse.
Theorem 2.
362
8.2
Proof?
P'
--,
I
I
I
I
I
I
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)
FP + F'P
=>
x2 + 2cx + c2 + y2
=>
and
(c, 0).
Then
2a
( -c, 0)
-----
4a2 - 4aJ(x
aJ(x - c) 2 + y 2
a2 '-- ex
a2x2 - 2 a2cx + a2c2 + a2y 2
c) 2 +
y2
+ x2 - 2cx + c2 + y2
a4
a2(a2 - c2)
- 2 a2cx + c2x2
The Ellipse
8.2
It is possible to show,
22
363
(x, y)
the equation
x2
a2
For example, for
3 and c
y2
--
+
=
(c, 0)
a 2 - c2
2
2a
is the graph of
1.
we get
x2
y2
-+-=l.
9
5
0,
y=
)5.
We
y
2
-2
Given an equation
x2
y2
a2
b2
-+- = 1 '
>
0.
Since
a2 - b2
a2 - b2
c2
a2
For
<
a2,
y2
/;2
1.
c =
at
b2
2a
and foci at
(c, 0),
2b
where
and foci
8.2
364
-a
-b
b<a,
c=Va2-b2,
a<b,
FP+F'P=2a.
c=vb2-a2,
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
4x'2 + 9y'2 + (8h - 8)x' + (18k + 18)y' + 4h2 + 9k2 - Sh + 18k Evidently we want
23
0.
4x ' + 9y' - 36 = 0,
or
x2
y2
'
'
-+-=l.
9
4
'
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
O);
focal sum 4.
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
13. 4x2 + y2 =1
15.
14. x2 + x +
y2
-
2y
-
+ 1 =0
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.)
f (x)
for every x.
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.
may happen that an unbounded figure has more than one "center."
(Thus it
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.
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)
w,
-1
2 y 2 l}.
Show that R is symmetric about infinitely many points, and find a simple description of
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?
1.
Let F = ( -c,
366
8.3
a) Show that
y2
a2 - e2
-- - (a2 - x2).
a2
b) Show that
FP + F P
I
a,
e) Show that FP + MP
I
'\/ (a2 + ex)2 + 1 '\/I (a2 - ex)2
a
-
2a.
8.3
THE HYPERBOLA
Given
condition
FP - F'P =
2a
(a < c).
F, F'
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
<=>
F'P
(-c, 0) and
2a
=
J(x - c)2 + y2 2a
x2 + 2ex + c2 + y2
- a2
2a
2
J(x + c)2 + y
ex
F'
aJ(x - c)2 + y2
8.3
The Hyperbola
c2x2 - 2a2cx
-:?
(c2
a2)x2
a4 = a2x2 - 2a2cx
a2y2 = a2(c2
a2)
a2c2
367
a 2y2
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.
x2
-
a2
bz
We shall use our equation to justify the sketch which we gave at the outset.
and
g(x) =
<
<
a
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.
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
368
8.3
= f(x) = l,
x
asymptotes
of this
By this we mean, roughly speaking, that points of the curve, far from the
As x-+
oo,
(This distance is
P=
x;-<0.
1
y=f(x) = x'
asymptote
(x,j(x)) approaches 0 as
x-+
oo,
or as x-+
oo.
definition.
directions.
g(y) = 1/y;
In the case of
y=
contains
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)
M
IimMP=O
y=f(x) = l
x (x;o<O)
y-.ro
lim MP=O
Jim MP=O
y-.-ro
The Hyperbola
8.3
m(x) = ,!'.
x
369
y) is
..! .Jx2 - a2 = 1
x a
a
a2
x2
Obviously
Jim m(x)
and this suggests that the line
y=
bfa,
bx/a, or x/a
y/b =
quadrants.
Thus we need to show that Jim.,
... 00
- y/b =
0, is an asymptote of the
NP=
Obviously NP
--+
0 as x
x
a
x
. (x - .Jx2 - a2).
a
x
b
a2
a
--+ oo.
+
+
.Jx2 - a2
.Jx2 - a2
.Jx2 - a2 .
Therefore MP
--+
This
The lines
x
y
--=0
a
b
are asymptotes of the hyperbola
x2
a2
y2
b2
l.
370
8.3
2
x
y2
= 1.
or
Jx.
Y =
y=x
'
/
/
'
'
'
'
'
/
/
/
/
rectangular.
'
'
'
'
'
'
'
'
'
'
If such a hyperbola is in standard position, then the asymptotes must be the lines
xy
or
2
2
2
x -y =a .
If the foci are on the y-axis, at the points (0,c ) , then the equation of the hyper
2
x
--2
c - a2
1.
2
y
= 1
2
b
is the union of two hyperbolas with the same asymptotes. These are called
hyperbolas.
conjugate
8.3
The Hyperbola
371
x2 - 4y2
4. y2 - 4x2
7. -x2 + 9y2
2. y2 - 4x2
-4
10. 25 y2 - 4x2
5. 9x2 - 4y2
8.
6. 9x2 - 4y2
36
-9x2 + y2
3. x2 - 4y2
9. 25x2 - 4y2
-4
-36
100
100
Derive equations for the hyperbolas determined by the following conditions, and sketch.
11.
Foci at (2,
13. Foci at
14. Foci at
19. Foci at
21. Given F, F', and a, as for a hyperbola in standard position. What is the graph of the
condition FP - F'P
-la?
0 and
Investigate the graphs of the following equations. In each case, find all asymptotes.
23. (x2 - y2 - 1)2
25.
x2y2 - xy +
=0
=
I
------
(x - l)(x - 2)
-1, let F be the origin, and for each point P let DP be the
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
DP
What sort of curve is this?
Sketch.
=
1.
=
I and
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 ...
"
372
8.4
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.
Let F
(e, 0).
x2
-
a2
v2
__
,_ =
e2 - a2
a) Show that
y2
e2 - a2
-- - (x2 - a2).
a2
b) Show that
FP - F P
I
c) Show that, if
ex
a,
then
+ a2 > 0,
1
I .I
- v (ex + a2)2 - - '\I (ex - a2)2.
a
a
-a,
ex
- a2 > 0,
FP - F'P
and
2a.
then
ex + a2 < 0,
ex -
a2 < 0,
and
FP - F'P = -2a.
0,
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.
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
Rotation of Axes
373
y2 = 0
x2
is empty.
y2
1 = 0
point, and
the empty set.
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
y)(x + y). This is = 0
y = 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.
= 0 or x
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)
h)
line, and
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.
374
8.4
Suppose that we rotate the axes through an angle of measure 8, getting a new
pair of axes.
y'
In the figure,
= r
cos cp,
x'
= r
cos (<P - 8)
y'
= r
sin (cp - 8)
sin cp,
= r
= r
= r
Evidently
r
r
Therefore the new coordinates are given in terms of the old ones l:>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
These give
x
Theorem 1.
rotation of axes.
Before going into the proof, let us try a simple example:
xy
l.
(1)
1,
or
x'2 sine cos 8 + x'y'(cos2 8 - sin2 8) - y'2 sine cos 8
We want the x'y'-term to vanish. Thus we want
cos2 8 - sin2 8
0,
or
cos 26
O;
l.
8.4
Rotation of Axes
375
'TT
=
nTr
Tr
'
cos 8
)2
and
sin 8 cos 8
nTr
+ -.
t.
x' 2
y' 2
1.
0.
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'
B'
C'
D'
E'
A cos2
F'= F.
0.
2 C sin 8 cos 8,
376
8.4
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)
B' =0
cos2e =o,
--B
or
A-C
=tan 28.
Tan
I
-
---
A-C
(The theorem did not say that the coefficients in the new
(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
1)
( )
A x2 +
0.
( )
2
+ C y +
= -F,
( r c(y r
A x +
-F +
:;2,
2
The
8.4
Rotation of Axes
377
E
,
y =y+-.
2C
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
a point
a hyperbola
C'< 0, F" = 0
C'< 0, F"< 0
2)
This
For E' =;r6 0, this is a parabola. For E' = 0, the equation x'2 =
3)
Suppose that A = 0.
A-C
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.
8.4
378
v1+k2
k?
x
k?
v1+k2
In the figure,
B
Therefore
1
cos 2() =
I
v1 +k2
The half-angle formulas are
x
cos-=
2
i+cos x
2
'
sin-=
2
l - os
x .
cos e =
where
i+cos 2e
2
.
sin e =
'
i- cos 2()
2
'
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(),
For example, consider
Here
and
B =2 ,
c =1,
B
2
=--=1.
k=
3-1
A-C
--
Therefore
1
1
=
cos 2e =
J1 + k2
J2
----=
Hence
cos e =
i+1;/2
2
2 + /2
4
'
8.4
Rotation of Axes
379
and
(In the second formula, sine > 0 because k > 0.) Therefore
cos2e
2 + /i
'
sin2 e
A'x'2
A'
A cos2e
3
2 + .J2
4
and
C'
C'y'2
+2
. .J2
/i.
4
1,
B sine cose
sine cose
'
2 - .J2
C sin2e
2 - .J2
2
.J + 2
- B s in e cose + C cos2e
2 - .J2 - . .J2 2 + .J2 -.J2
+ 2.
2
+
A sin2e
3.
inrestigate
what sort of figure the graph is, and sketch. If the graph is a conic section, you should also
1. Investigate
x - xy
of axes.)
2.
Investigate
3.
Investigate
x2 - xy
4.
xy -
1.
- 2y
0.
Investigate
5 . . Investigate
6. Investigate
7.
2xy - y2 + 2
0.
x2 + 2xy + y2 + 2x + 2y + l = 0.
x2 + 4xy + 4y2 + 4x + 8y + 3
0.
A' + C'
A + C
and
F' = F.
380
8.
8.4
Let
be the new
coefficients, when the axes are rotated through an angle of measure 8. Thus
C8,
are the
A8, B8,
B - 4A8C8
B 2 - 4AC is invariant
It
may be of some interest to check this, in the cases where we have computed the new
coefficients.
9.
4x
+ 5y + 6
What are the possibilities for the coefficients of x2 and y2, in the new
equation?
10.
11.
10
0.
v3 xy
+ y2 + 2x + 3
0.
e
,
with directrix
D, focus F,
(b) a parabola if
and eccentricity e.
G is called
> 1.
b) Is a circle a conic section, in the sense defined in Problem 12(a)? Why or why not?
9.1
Paths and
Vectors in a Plane
I there
corresponds a point
P(t);
E,
P: I-E.
For the motion shown in the figure,
point
is the point
P(O)
(1, 1 )
I is
[O,
oo
),
In general:
Definition.
P: I-E,
where
I is
an interval and
E is
a plane.
P: I - S,
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.
precisely:
P: I-E,
the locus of P is the set of all points
Q which are
381
P(t)
for some
in I.
More
382
P is
the
9.1
image
P.
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.
x = f(t)
4t
y = g(t) = 8t2
=
Here
At
/=(-00,00),
As
t increases,
P(t)
and
starting from
0,
(4t, 8t2).
both
and
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.
y=
sGY
tx2.
Thus every point of the path lies on the graph of the equation
(x, y)
coordinate functions,
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.
for example, the path
P described by
Consider,
the equations
x =f(t) = t2,
y = g(t)
t4
y = x2.
But the converse is not true.
Therefore the locus of
P is
?; 0, because t2 ?; 0.
9.1
383
y
\
\
\
\
\
\
'
'
', P(O)
,
'
,....-::;
. -'---- x
good sense to regard the parameter as the measure of an angle. Consider, for example,
x =cost,
y =sint.
x =cose,
y =sin()
Somewhat similar looking paths have ellipses as their loci. For example, consider
x =a cose,
y =b sine.
=cos() '
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.
384
9.1
y
In the figure,
Q
R
(a cos e, a sin8),
(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
4.
sec e, y = tan e
2. x = cos e, y = cos2 0
cos2 e, y = sin2 e
5.
x = t3, y
3.
x = 2 cos e, y = sin 0
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
9. x
tan2 0
sin e, y = Jsin 01
y
7. x = sec 0, y = cos 0
10. x = t6, y
8. x
t4
y
csc 0, y = cot ()
9.2
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.
=
parametric description of the path, and sketch the loci, showing the three cases
Find a
a
<
b,
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
(h, k)
20. Let
2a
y'
-2a
Then
x'
a cos (0
x = x' +
y'
y
if>),
h,
a sin (0
'
if>),
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
iY
This curve is called a four-cusped
principles.
k.
iY.f
f/.
k, then f is differentiable
Start by writing out the hypothesis and conclusion in terms of the basic
9.2
hypocycloid.
at a, and[' (a)
f(t),
y =
g(t),
386
9.2
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.
t + !it.
Llx=f (t + Llt)
and
t;
and let Q
Let
-
f (t),
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
g'(t)
m=--.
f'(t)
parametric slope formula.
We have shown:
t,
y= g(t).
corresponding point P, and the slope of the tangent is given by the formula
m= m(t)
g'(t)
.
f'(t)
can never take on the same value twice, and so the locus of the path is the graph of a
function
ef;.
9.2
L'Hopital's Rule
387
t=a
Qt=b
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)
f(b) - f(a)
rf'(x).
This number i must have come from somewhere. That is, there must be a i between
a and b such that
It follows that
f'(i)
Therefore
g(b) - g(a)
f(b) - f(a)
f(i).
g'(i)
.
f'(f)
g'(i)
f'(i)
(a <
< b).
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
for some
between a and b.
g(b) - g(a)
g'(i)
f(b) - f(a)
f'(i)
9.2
388
f (a)
g(a)
0.
g'(i)
(b)
g
-- f(b)
f'(i)'
for some f between a and b.
approaches a limit
L,
as
_,..a, then
g(t)/f(t)
L.
g'(t)/f'(t)
That is,
if
f(a)
a
g( )
ta
then
(t)
f(t)
lim g
t-a
L.
'
L.
'(t)
J'(t)
lim g
and
0,
=>
But
'(f)
g
- R:> L,
f'(i)
Therefore
fR:>a=:>fR:>a
=>
g(t)
f(t)
g'(i) R::3 L.
f'(f)
Therefore
t
t R::3 a => g( )
t
f( )
and so
(t)
J(t)
lim g
t-a
R::3
L'
L.
It is very easy to express this idea in the form of an E o proof; all we do is to for
-
I)
Hypothesis.
For every
"R::3"
E > 0
O<lt-al<o
2)
Conclusion.
For every
E > 0
(I).
For each
t,
=>
t
\g'( )_L\<E.
t
f'( )
furnished by
=>
t
\ g( )-L / < E.
f(t)
Given
9.2
theorem.
<It
al < o
==>
(f)
g
f''(f)
I
I
==>
==>
< If
(t
g
f(t))
f(a)
0.
lim
t--+a
and
<
<
.
g(a)
1 im f(t)
g(a)
389
al < o
L'Hopital's Rule
O;f and g
f and g
t--+a
(t).
a,
then we
Theorem 3
Jim
x--+a
( )
gx
and
Jim
a:--+a
( )
g' x
f ' (x)
L'
then
lim
x--+a
g(x)
f(x)
L
Consider
1.
sin
1m
--
x--+O
g(x)
f(x)
sin x,
x.
cos
1
sin
1.
1.
Therefore
Jim
x--+O
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
1.
x/x]
The point
390
9.2
is that
. sin x
1. sinx - sin 0
. , 0,
= Sln
l Im -- = Im
x-+O
X
x-+O
X- 0
by definition of sin' 0.
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.
F'(x0)
By definition,
1
- .Im
F(x) - F(x0)
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.
1.
Im
x-+O
For example,
2
sin x+x
1. 2sinxcosx+l
=l
= 1m
. ,,
"'
"'
e
- 1
x-+O
e
. xcosx
1im
--x-+o
sinx
Investigate the following indicated limits. (That is, calculate the ones which exi st.)
1. limx cotx
x-o
cos2x - 1
4. lim
x2
x -o
7.
10.
e"'
- 1
Jim
.,_0 In (x + 1)
Jim x2 sec2 x
00-+rr/2
2.
sin2 e
lim
82
3.
o-o
o-o
5.
Jim
x-1
lnx
6.
-X -
. y2 - 2y + 4
8. ltm
y-2
y- 3
1
sin3 e
lim
82
9.
x- I
lim -.,-e
- 1
x-1
Jim
.,-,,/2
12. Jim
x-rr
cos3x
x3 - 1
x4 - 1
. 2X - 1
Sill
9.2
1 3.
1 6.
1 9.
21.
23.
*25.
ln2 x - 1
Jim
x2
x-e
Jim
e-rr/4
Jim
x-l
Jim
t-e
Jim
x-1
1 4.
sine - cose
1 7.
e - Tr/4
Jim
Jim x2 csc2 x
x-o
(t)
t
22.
e13
v( dt
24.
Jim x In x
CC-+O+
26.
*28.
a;_.o+
29.
lim
x2 - 4x + 3
x-+1 X
2 - 3x + 2
- 1
[ J,"'
X
3x
.,. 20. Jim -1
x-oe
---
In
--
x-o
18.
391
tan x
Jim
1 5.
v:X
x-o
x2 - 1
In
L'Hopital's Rule
[ f
[ l"'
sect
Jim
t-rr/2
Jim
csc x
x-rr
rr/2
]
]
Vl + sin3 t dt
rr/2
Vl + sin3 t dt
Jim x In sin x
X-+0+
Jim x In (cos2 x sin2 x)
x-o+
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'
I
I
As the figure suggests, the easiest method is to use a "moving coordinate system,"
(h, k) of the
. (What is ?)
y' as
cos
</> and
sin
392
9.2
f(O)
b-a
(a- b) cos e + b cos - - e,
b
g(O)
b - a
(a - b) sin e + b sin - - e.
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.
y
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.
y
*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
393
(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.
A
9.3
lim
x-+a
x-+a
(x) =
0,
and
(x)
g'
x->a j'(x)
lim
L'
then
(x)
g
x->a f(x)
Jim
L.
If
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
x-oo
lim
x-toc,
then
(x) =
oo,
and
(x)
=
g
x->ro J(x)
Jim
If
(x)
g'
'
x->oo j (x)
Jim
L.
L'
394
Example I .
9.3
To find
1. In x
im-,
X-+O'J X
1/x
lim- = 0.
X-+00 1
ln x
= 0.
lim
X-+ 00 X
The case in which g' (x)Jf'(x)
Example 2.
oo
causes no trouble.
To find
e
"'
Jim - ,
X-t-00 X
we investigate
"'
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.
then
If
limf(x) =Jim g(x) = oo,
X-+00
X-+00
and
.
g'(x)
hm -- =co,
x-+oo f'(x)
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)
f(x)
lim
= 0.
X-+00 g(x)
Example 3.
Consider
g(x)
lim
=co.
x-+oof(x)
.
In x
1.
- In x
I1m-=-1m -- .
x-+O+ X
., .... o+
x
9.3
oo oo.
-1/x
--
395
=0 .
X->0
and so the answer to the original problem is -0 = 0. The theorem being used here
is the following.
Theorem 3.
then
If
lim f(x) = lim g(x) =
+
x-+a+
x-+a
(x)
lim g'
= L'
x->a+ f'(x)
and
oo,
(x)
lim g
= L.
x..,a+ f(x)
Let
1
x=a+-,
y
y= -- ,
x - a
so that y
---+ oo
as
x ---+ a+
and
x .......
a+
as
Then
---+ oo.
Im
v->oo
g'(a+ l /y)(-l/y22)
f'(a+ 1/y)(-1/y )
1.
= Im
y-.oo
=
The Northeast theorem now applies to
g'(a+ 1/y)
f'(a+ 1/y)
g'(x)
hm -x-.a+ f'(x)
g(a+ 1/y)
Y->00 f(a
+ 1/y)
].
1m
L.
'
(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.
lim
.,_, 0
x"'
= ?
Consider
9.3
396
Let cf>(x)
xx. Then
=x
In cf>(x )
Now
.
g'(x)
Ilffi --
x-+o+
f'(x)
g(x)
= 1/xx = f(x)
In
In x
lffi
x-+O
1/x
--1/x
==
Jim In cf>(x)
0,
1lffi ( -x )
limxx
x o+
and
x-+o+
..-. =
Therefore
x-o+
e0
1.
5. Jim e-1/x'
6. Jim [(l/x)e-1/x2]
x-o+
x-o+
7.
a--+ co
14. Jim
x-'"12
17.
X-+00
o-o+
Jim
t-oo
))Tan-'
(
()1
x--o+
x)sin
x-o
kx)csc
Tf'-o+
18.
Jim x2 ln x
x-o+
21. Jim x Jn x
x-o+
24. Jim ( l/x2 +
x-o+
/1
ln x)
0.
27.
v-o+
x-o
32. Jim
X---+00
1 + Tan 1 (x
x-o+
'
ww
f( x )
e-1/x2
for x ;e 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.]
=
Polar Coordinates
9.4
9.4
397
POLAR COORDINATES
(x, y)
E,
E.
f--+ P.
And the correspondence also works in reverse: when P is named, x and y are deter
mined.
{(x, y)}
the direction 8 as the positive direction; and we let P be the point with coordinate
(This is equivalent to saying that the directed distance OP is
r.
,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).
coordinates ( -/2, 57T/4). And this is not all; the possible polar coordinates for Pare
cI2, 7r/4
c-.J2, s7T/4
+ 2n7T),
+ 2n7T),
{(r, 8)}
E,
1)
cos 8
co 8 27T).
398
9.4
Since the cosine is periodic, with period 27T, we can get all of the locus by restricting()
to the interval [O,
27T].
r
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
7T/2, 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.
,..
'
3,..
'
3,..
2
As() continues to increase, from
7T
to
27T,
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.
(r, ()),
x
= r
cos e,
y = r sine.
Polar Coordinates
9.4
399
(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
x2 + y2
0 and
r2 cos2 e + r2 sin2e
r2 .
r cose
x2 + y2
r sine,
,2,
cose
does not involve any of the three expressions r cose, r sine, r2 which we know how
to convert into rectangular coordinates.
getting
r2
r cose;
x.
(t, 0)
and radius t.
Consider
r
sece.
It is easier, however, to multiply both sides of the equation by cose. This gives
r cose
Ase increases from 0 to
27T
I,
or
I.
(It
Consider, however,
r
sine
co e 27T).
9.4
400
(As for r
cos(), the interval [O, 277] gives us the entire locus of the path.)
First
parts:
11'
2
2
r
3,,.
2
This curve is called, for obvious reasons, a cardioid.
First we observe
that since
1 -
sin () ;;; 0,
(1)
r
We can therefore write
r2
J--;2
=
x2 + y2
4)
Jx2 + y2.
r - rsin(),
Jx2 + y2
(2)
_
y.
(3)
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
x' - p = 0,
Polar Coordinates
9.4
'
= x
cos <P +
cos <P +
401
sin cf> - p
0.
0,
0.
1, 2.
Therefore
d
2
2
(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) + recos e2 + sin ez)
- 2r1r2ecos el cos e2 + sin el sin e2)
ri
+ ,. -
TT,
But the polar distance formula applies for any values of r1, r2, e1, and e2
9.5
402
5.
7. r2 =sin 8 sec3 8
8. r =
13. r =
=1 - cos 8
11.
sin 30
36
1 - sin0
19.
r=
sin 8 - cos 8
20. r2 =sin2 8
22.
23.
2 +cos 8
15.
4 cos2 8 +9 sin2 8
17.
r
1 +r cos 0
r=sin 28
12. r =1 + r cos 8
=sin 48
16. r sin 0
9.
sin 8 +cos 8
14. r2=
1 +cos0
6. r=sin 8 sec2 8
4. r =1 +sin 8
10. r
2. r = -2 sec 8
r=2 csc0
= 1 +csc 8
18. r=cos 38
= 2.
21.
r2
24.
=sin0
=eBI
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
27. r2
cos2 0
cos 8
28. r2 = cos 28
29. r2 = a2 cos 28
r2
30.
=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
= 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).
9.5
r
1 +r cos 0
36.
= 3 - 2 sin 0
37.
=3 + 2 cos 8
1
=- (What sort of curve is this, and why?)
Given
r
f(8) 0,
is
;;;
2rr.
Let
be the
9.5
403
7r
That is,
R
and
;;;
;;; 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.
=
m;,
-M; fl&;.
A= .2flA;.
i=l
is
404
9.5
i;
I -m; 6.8;
i=l
I iM; 6.0;.
i=l
But the sum on the left is the lower sum s(N) of the function
F(8)
over the net
N,
}/(B)2,
and the sum on the right is the upper sum S(N), of the same function
N. Thus
s(N) A S(N);
and so
I
Since
Jim s(N)
s1-o
lim s(N)
JXJ-o
is squeezed, and
it follows that
Jim S(N).
J.YJ-o
(P U(8)2 dO
J,
Thus we have:
fHC0)2
lim S(N),
J.\J-o
dO.
Theorem 1.
f f(fJ)2
i2"ta2
dO
r = -----
'
cos 0 + sine
we get
A=
1"12
o 2
l;;/2
2 o
["12
1
2.o
8 7T/2,
+ sin 20
- sin 20
cos220
dO
c1e
d6
}(O + t)
be
dO.
Let us try this in some simple cases. For the circular region with radius
center at the origin, the formula gives
A
(O - i)
and
9.6
405
This is correct, because the region is a right triangle with legs of length 1.
r =
4.
r2 =
7.
r =
- sin 0
r = 1 - cos()
r2 = cos 20
2.
cos2()
5.
()
j4
1T
8. r
11.
r =
1
lcos
fJI +
lsin
12.
01
= e0, 0
r =
6.
9.
r2
= sin() cos()
10.
r =
0 21T
() 1T/4
4 cos 2()
cos()
3.
(r0, fJ 0)
= e28, 0
() 21T
mean the slope, relative to a rectangular coordinate system superimposed on the polar
coordinate system.
9.6
Roughly speaking, the length of a path is the total distance traversed by the moving
point.
a cos
t,
g(t)
0 to 47T,
sin t
(0
47T).
But as t increases
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'
g(t)
is four, not zero; the two halves of the path do not cancel each other out.
9.6
406
x = f(t),
g(t)
y =
(a t b),
let
be a net over
[a, b];
Yi
g(ti),
Then
Po
s = Jim
IN10
2; P;_1P;.
i=O
.6.x; = X; - X;_1,
as indicated in the figure.
y
:I
_____ _
Xi-1
Then
Xi
g(t;),
9.6
407
Since
and
we know by the mean-value theorem that
6.xi
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 pi-lpi i=l
i -J f'(ii)2 +
i=l
g'(f/)2 6.ti.
ix(t)
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
f-J f'(t)2
g ( f;)2 6.t;
'
i t1.(i;) 6.ti
i=l
+ 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
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
cp(6)
(a e b).
y =
cp'(6)cos6 - cp(6)sin6,
'
g (6)
x
This gives
j'(6)
/(6)
g(6)
=
cp(6)sin6.
f (6)sin6 + cp(6)cos6.
408
9.6
and
(a e b),
where </>' is continuous, the length of the path is
2. r = eO, 0 () 2n
1
, 0 0 n/2
4. r =
.
cos 0 + sm 0
r = cos(), 0 () n
3. x = 0, y = cost, 0 t h/2
5.
9.
r = 2 sin 8, 0 () r.
r =
8.
,.
2
cos() - sine '
1
= 0
n/2 0 n
0 () 2r.
10. x = 1 + sint, y
1 1.
6.
x =
cost, 0 t 7T/4
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 )
'-'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.
Vectors in a Plane
9.7
9.7
409
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).
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
--+
P.
--+
OP
OP,
starting
will be called a
vector.
9.7
410
--+
We allow the "degenerate segment" 00; this is called the zero vector1 and may be
-+
denoted simply by 0.
-+
P.
Addition.
--+
OP
by the shorter
Given P1,
-+
P2,
with P1 =
(x2, y2),
where
Vector addition is governed by the same formal laws that govern addition of real
numbers, as follows.
-+
A.1
Associativity.
A.2
Existence of 0.
every vector
-+
-+
-+
-+
-+
-+
-+
-+
-+
--+
P + ( -P) = ( -P) +P =
Commutativity.
A.4
--+
-+
-+
+P = P + 0 = P
for
P.
Existence of negatives.
A.3
-+
-+
-+
-+
such that
-+
0.
-+
P1 +P2 = P2 +P1.
-P
-+
=
Q,
and
-+
-+
-+
For example, if
-+
then
-+
-+
Q = Q'.
Q, where Q
-+
-+
(-x, -y).
Similarly for
A.4.
(x, y),
then
-P =
9.7
Vectors in a Plane
411
- aP = Q,
where
(clX , ay).
M.1. (a{J)P
a({J]>).
Because
(a{J)x
==;
a({Jy).
-+
aP+ {JP.
-+
-+
aP1 + r1.P2.
M.4. 0
M.5.
M.6. a
P
P
-
-+
0, for every
P.
P, for every P.
= 0, for every a.
-+
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
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
P1 . P2
X1X2 + Y1Y2
-+
-+
S.1. P1 P2
S.2. (aP1) P2
-
->-
P2 P1.
-+
a(P1 P2).
--
-+
-+
P1 P2 + P1
(P2 + Pa)
- S.4. P P 0, for every P.
S.3. P1
-+
Pa.
412
If P
S.5.
-+
0, then
-+
-+
9.7
0.
(The last condition rules out trivial "dot products" for which
for every
-+
-+
-+
P1 P2
is always 0,
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.
x-+
lf P1 and
P2
-+
-+
opposite direction.
-+
P,
-+
To multiply a vector
-+
by a positive scalar
x.
IX,
-?
aP
-+
P1 P2
X1X2 + Y1Y2
Algebraically,
Vectors in a Plane
9.7
413
Substituting cos
we get
so that
-+
-+
P1
Obviously cos
P2 = OP1
OP2 cos 8.
angle between the two vectors. Note that the length of the vector
-+
P can be expressed
p . p
= x2 + y 2 = OP 2.
-+
IP!. Thus
j+-+
= PP.
-+-+
-+
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
-+
P =xi + yj
(i = (1, 0), j
(0,
1)).
414
9.7
y
This section contains no new information, but quite a lot of new language.
Learning a language takes practice.
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.
3. P
od ( -
<
oo
5. p . i
0
.
7. p (i + j)
9. p . p
1
2. P
< w)
C'I.
<
4. P
< w)
C'I.
10. p. p
r.ti + 2et.j
13. p. j
15. P
19. P
co
C'I.
14. P
C'I.
<
w)
16. p
18. p
r.tj + r.t2i
21. p . (2i + j)
22. Let c
12. p. i
v3)
(-w
<
r.t
20. p .
< co)
(i
C'I.
<
)
)
8. p . (i + j)
<
6. p . j
11. J3
r.tj ( -
cxi + r.t2j
( - O'J <
C'I.
< oo)
et.i + r.t3j
( - O'J <
C'I.
< O'J)
C'l.2i + r.t3j
+ 2j)
( - Ct:) <
C'I.
< (/J)
i + j, 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
25. Let e
i + 2j, f
2i - j.
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?
Free Vectors
9.8
28.
Let P1
(2,
1), P2
(1, 2).
(0 ;;;i
29.
-+
415
et:
-+
;;;i I).
different vectors).
Sketch the
et:P1 + (1
-+
rx)P2
(0 ;;;i
;;;i I).
et:
30. p
. i
32. p .
34. p
36. p
*33.
(i +
=
31. p . j 0
33. p .
j) 0
et:i +
. (i + 2j)
35. p
37. p
j) 0
(i -
et:i + {i'(i
. (i -
2j)
<
(o: 0, /! 0)
+ j)
0
1/1,
[ -1, I].
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)
h(x)
x,
then.fg
O,g
0,
and/
x3, g(x)
1,
%.)
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
)'
k,
+ h,
(x + h, y + k).
----+
Suppose that we have given two directed segments PQ, P'Q', in a coordinate
plane.
"---+
P' and
Q'
Q
p'
---+
9.8
416
Q'
Let
(x, y).
+ h,
x I-+ x
I-+
where
h
x{ -
X1,
y + 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
intended).
vector.
(or simply a
vector,
Thus the figure on the left below is a partial picture of exactly one free
equivalence class
representative
vector PQ.
y
y
-+
-----*
If two directed segments PQ, P'Q' are equivalent, then they determine the same
-+
P' Q'.
-+
And if PQ
-+
-----*
-+
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
Free Vectors
9.8
417
--+
---+
OP+ OQ =OR,
by definition.
by definition; and
by definition.
---+
--+
OP OQ
--+
OQ, then
--+
rxOP
--+
rxOP
OQ,
OP OQ,
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
not worth while to do so.
representatives. That is,
--+
--+
IOPI =OP.
IOPI
OP for P), it is
--+
Thus different segments may have the same label, as in the left-hand
PQ.
figure below; and when they do, this means that the segments are equivalent.
y
y
u
Theorem 1.
--+
PQ + QP
0, for every
P, Q.
Therefore:
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
--+
9.8
418
Theorem 2.
Proof
PQ + QR + RP
-
PQ + QR + RP
0, for every P, Q, R.
OQ - OP + OR - OQ + OP - OR
0.
R
I
I
I
I
I
I
I
I
I
I
I
I
"
PQ
and
P'Q'
PQ
and
=-+
P'Q'
are
Note that while the directions e, e' depend on the directions of the axes, the equation
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
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
Free Vectors
9.8
of
--+
OP.
where
In this case
--+
P,
--+
--+
P= OP =
i, and
xi +
yj,
i, j;
and
OP is
419
PQ, i,
and
--+
j can
be drawn
and
are all
free vectors. In general, ifV and Tare any vectors, with T :;tf. 0, then the T-component
of
is the number
vT
1v1
cos e,
where 8 measures the angle between the direction of T and the direction of
y
V.
Q
PQ=i+2j.
Thus, in the figure below
VT is
PR.
R
Since
V T = IVI
ITI
cos e,
v T
V T
--
IT!
420
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.
-+
--+
--+
0
--+
--+
--+
--+
-+
--+
--+
(The figure is a
--+
-+
\OR\.)
T
0
--+
--+
OT
0.
Free Vectors
9.8
421
-+
--+
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,
Show that if P, Q, R, and Sare any four points of the plane, then
---+
---+
---+
PQ + QR + RS + SP
Let V0 be a fixed (free) vector.
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.
7.
Show that if V0
0 and V
0, discuss as in Problem 8.
11.
Given V
1 and V
10.
0.
1, discuss as in Problem 8.
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
iI
CliVi
=>
IX 1
IX2
Cln
0.
c) Given that i and OP are linearly dependent, what are the possibilities for P?
422
13.
9.9
Show that if
and
then
IV1 - V2I
IW1 - W2I
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
We return to the discussion of moving particles in a plane. Suppose that the motion is
described by a path
P: IE
t
where
I is a
P(t),
(ton
g,
so that
I).
We now regard the path as a function whose values are the vectors
-
----+
pt= OPt,
-
P(t) is
denoted by
Pt,
to fit it into
We then have
-+
423
9.9
where i and
Since
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
t,
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.
which do have physical meanings. This is done in the following way. First we take a
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
Ai=
of T and N.
o:T + (3N
Au
These
numbers are called the tangential and normal components of the acceleration.
shall now compute them.
We
424
9.9
At
is the direction of
we have
g'(t)j,
where
<P
A1I cos(</>
.
(t)
Sill</>= g"
-.
f"(t)
IAtl '
IX of
IA1I
At, we have
8) = IA1I cos
4> cos
f"
f"(t) cos 8 + g"(t) sin 8 =
8 + IAtl sin
4> sin
(t)f'(t) g"(t)g'(t)
I
I
-Jf'(t)2
Theorem 1.
+ g'(t)2
That is,
IX=
Ar
Once this has been observed, it is easy to check it, by differentiating the function
IV tl =
)j'(t)2
+ g'(t)2.
fJ is computed as follows.
423, then
T, as in the figure on p.
fJ
r/>]
I At I COS [(8 + rr/2)
ef>) + rr/2] = -IA tl
IAtl cos [(8
-
If N is counterclockwise from
(8
sin
(8
- ef>).
A1. Therefore
425
9.9
we must have
fJ
IAtl
(8
!sin
</>)I,
fJ
IAtl
lf"(t) sine
g"(t) cose1
,Jf'(t)2 +
IVtl
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.
y
For each
x,
let
For each
Since
x,
let
8(x)
to t
x.
Then
f'J1 +f'(t)2dt,
and
s'(x)
,J1
+f'(x)2
-TT/2
<
is a function
5.8)
ae
ds
The curvature
h'(s(x))
d8/dx
.
ds/d."
is defined to be
K
=I :I
8(x)
<
TT/2.
Thus there
426
9.9
(}(x)
we have
{}'(x)
Therefore
K
d(}
dx
Tan-1 f'(x),
1
f (x)
1 + f'(x)2 "
f"(x)
I
I d(}ds I I d(}/dx
I
ds/dx
1 f'(x)2 -J1
=
f"(x)
[1 + f'(x)2]3/2
1
+ f'(x)2
by the formula
K
f"(x)
(1 + j'(x)2]3/2
t0 to t.
Then
s(t)
and
{t-Jf'(u)2
Jto
g'(u)2 du,
{}(t)
d(}
ds
h(s(t)).
h'(s(t))
d(}/dt
ds/dt.
In order to calculate
of the path,
lI
{}(t)
[sec
(}(t)](}'(t)
g'(t) .
f'(t)
f'(t)g"(t) - g'(t)f"(t)
.
j'(t)2
9.9
427
Now
2
sec
Therefore
f'(t)g"(t) - g'(t)f"(t)
.
f'(t)2 + g'(t)2
(Query: How
would you derive the same formula, in the case where the velocity vector is vertical?)
This gives
K=
de
ds
I I I
d e/dt
ds/dt
[ [
e'(t)
s'(t)
g'(t)f"(t)
= f'(t)g"(t)
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.
speed is not
lf'(t)g"(t) - g'(t)f"(t)I
]3/
[f'(t)2 + g'(t)2 2
K=
/3 =
lf"(t)g'(t) - g"(t)f'(t) I
.Jf'(t)2
we get:
Theorem 4.
+ g'(t)2
'
At any point where the speed is not zero, the normal component of
/3 = AN = K IV1l2
In our discussion, we used the notation f',
g',
time, in order to connect our work with the preceding theory. We used the notation
de/dt, de/ds,
function with respect to another, in defining and calculating curvature. In the litera
ture of physics, however, the notation f',
'
g,
dg.
df.
v = -1 +
1
dt
dt
-
dx
dy
v = -i +--j,
dt
dt
x
over
and y indicate
time. Similarly,
A=
d2g .
d2x
d2j
J=
l +
d t2
dt2
dt2
.
.
l
v =xi+ yj.
differentiation with respect to
d2y.
J = Xl + YJ
dt2
.. .
...
428
9.9
In these notations,
and
K
Ji.XI
[x2 + _y2]312
li.Y
l(dx/dt)(d2y/dt2) - (dy/dt)(d2x/dt2)1
[(dx/dt)2 + (dy/dt)2]31 2
There is a good reason, in physics, for the use of the "fractional" notation
dy/dt,
df/dx,
... 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.
"fractional" notation
df/dx
f'.
2.
3.
x2,
K.
K.
x3,
and
at these points.
4.
5.
a.
x2
y2
+
=1.
/;2
6.
Pt = i cost + j sint,
showing the velocity and acceleration vectors at several points.
7.
6)
t
Pt = i cos 2 + j sin 2 .
8. Discuss
9. Discuss
10. Discuss
Pt = it + jt2 .
Pt
ti +
(t - t2)j (0 t
11. Discuss
12.
Pt
Discuss
it + jt3.
13. Discuss
Pt = it3 + jt2
14. Discuss
15.
Discuss
1).
9.9
429
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
P0
Describe this as a path in polar coordinates; find a rectangular equation for its locus,
and identify the locus.
19. Discuss and sketch
Po
Po
P0
Given a path with coordinate functions f and g, on an interval [a, b], such
some
calculation.]
25. Let a
i + j, b
that for
V1
V2
x1a
+ y1b,
x2a
+ y2b,
V2, by agreeing
430
9.10
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?
case, express the new operation * in terms of the old.
In any
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.
1)
Free vectors.
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_.
---F2-
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)
4)
Systems of other kinds, regarded as vector spaces and inner product spaces.
Some
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.
Infinite Series
10
10.1
LIMITS OF SEQUENCES
oo.
oo.
a or as
For example, in
Section 2.10, we wanted to find the area A, under the graph of y = x2, from
to
=0
h3
1 +
r) (
1
+ 2n '
h3
limAn = n-+oo
3
We are now going to use limits of sequences more extensively, as a way of dealing
with infinite series.
L ai
i=l
= a 1 + a 2 + ...
.
'
we define
sums
of
_2;:1 ai.
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.
for every
n > N
=>
IAn
431
LI
<
E.
L.
Suppose that
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.
Proof
limnoo
1
=
n
0.
Here
for every
- <E.
1
=>
Now
1
-<E
n
If
l/E
is an integer, let N =
l/E.
n> N
<:::>
n>-.
E
n > 1/E
=>
=>
l/E.
Then
I/n < E,
A,,,
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.
Theorem 2.
If lim,,00
A,, = A
and lim,,00
B,, = B,
then
and
If
':;!: 0, and
B,,
n,
then
These theorems justify the procedures that we have been using informally.
example, they give a proof that
lim
n-+oo
h3 (i n.!) (i
3
) h3
12n
For
Limits of Sequences
10.1
433
lim
.!
n -+oo
lim
(i .!)
lim
_!_
lim
n-+ 00 n
n-+oo 2n
n-+oo
= 1
= .! lim .! = 0
2 n-+oo n
i +
1-
lim ha 1 +
n-+oo 3
=1
)
( .!) (
2n
i +
1-
=ha.
2n
, B1, B2,
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,
sequence is decreasing if An
The
An+l for every n. (If An < An+i for every n, then the
.
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
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)
bounded.
(3)
bounded below but not above. (2) If An = e- , then the sequence is decreasing, and is
n
and IM/.
Theorem 3. If a sequence is increasing, and is bounded above, then it is convergent.
That is, if
AiA2 ...AnAn+l ... M,
1,
we inscribe in it a regular
(Note that
we had better start with n = 2.) It is a matter of elementary geometry to show that the
Infinite Series
434
10.1
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
n co
A.
....
Let
Then A
< A. Therefore A -
is not an
n > N
An > A
=>
E.
is an upper
bound of S.
Therefore
for every n.
Therefore
E,
has a limit.
Proof
-An.
B.
Some simple sequences converge for reas0ns which are not covered by the
preceding theorems. For example, given that
lim
n-.0 __
_!
0'
it is obvious that
lim
1.V"n-co
n + 2 n
0,
because the second sequence is smaller, term by term. This is the idea of the following
theorem.
10.l
Limits of Sequences
Theorem 5
435
cos
11-00
= 0'
because lcos
because cos
nl I, and I/n -+ 0.
n does not approach a
Theorem 6
limit as
n-+ oo.
is bounded,
then limn_,00 A nB n = 0.
Theorem 7.
For
increasing
sequences,
this is trivial:
n.
not
if Ai, A2,
is increasing,
and
is not convergent.
The statements
lim An =
mean what you would expect.
IimAn = -00
n...,co
oo,
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,
in many cases.
-oo.
oo
and
1.
lim 2
noo
2.
Jim
2 + 311
3.
4.
6.
Jim
3n
n-oo n- +
Jim
---
n-Cf)
n-oo 2 ll
ll
5.
+ n
.
sin (2n +
hm
n.)
1)
7.
lim
n->00
3
ll
ll
+ n2 +
.
cos (n
Jim
n->ro
7T
- 1)
n + 1
436
8.
9.
Infinite Series
Jim
n->OO
lim
n_,.oo
(
(
n
)
1 +
10.1
-l/n
)
[Hint:
(1 +
(1
1/y)v,
x)1fx.
10. lim,H"' Bn. where Bn is the perimeter of a regular 2n-gon circumscribed about a circle
1.
of radius
1.
13.
15.
17.
Jim Inn
n-oo
1/n
Jim In
n-oo
"dx
n-oo
i
f
Jim
.L
Jim
n-+CO
Jim
n---+CO z=l
19.
Jim
21.
14.
n---+CO
(n2)
Jim In (
Jim
n-c.o
)
ll
"dx
3
1 x
"dx
n--oo J1
Jim
n-oo
dx
312
"
Jim In
12.
16.
X2
18.
20.
(You need not prove that your answer to this one is right.)
n--i-oo i=1
X
1
:a
l
n
1
L .31
n-+OO 1.=l l 2
lim
Jim
Jim
22.
n->OO t=l l
23.
n--co i=l
sin
24.
Jim
cos
-:
( hr)n n
I ( 2n) n
:
hr
noo i=l
(Geometric interpretation?)
n
!!__
25.
I)
_L --- (
n-oo i=l I + (i/n) ;
Jim
10.2
26.
Infinite Series.
lim
n-co
"
i=l
1 + (i/211)
28.
l n
lim - L e-i/n
n-co n i=l
30.
Jim n2 sin 2
n
n--+CO
32.
Jim
n-.oo
34.
Jim
n--+CO
(1)-
27.
31.
437
L eifn
n-co n i=l
-n1
( 1n
Jim
sin
Jim 11 1
n-+CO
Jim
33.
tan
Comparison Tests
1 n
Jim -
29.
Convergence.
sec
cos
[I - ]
In
1.=l l
(In fact, this limit exists; if you can find a geometric interpretation of the problem, you
The limit is known as Euler's constant. Nobody knows whether it is
rational.)
35.
Jim
n->OO
10.2
i=l
36.
(21 + 1)
n
1
um I
.2
)2
i=l
(31
+ l
n-.oo
By an
infinite series
I ai
i=l
= a1 + a2 +
+ an +
We say "an indicated sum" because in many cases there is no such thing as the sum
of infinitely many terms.
l+l+l+ "
has no sum; and neither does the series
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
An =
I ai
i=I
a1
+ a2 +
of the series.
+ an
If
limAn =A,
where A is a (finite) number, then we say that the series is
is its
sum.
converges to A.
that the series is divergent. If
limAn =
n-+ oo
oo,
convergent
and that A
438
Infinite Series
10.2
"'
L Qi =
i=l
00
.z a;=
A,
-oo,
co,
i=l
=-co .
.2;a;
i=l
Probably the first example that you have seen of a convergent series is the geo
metric series
1+
Here
=
An
1 +
,2
r + r2 +
1- r
Jim
-oo
,.n
00
.2; ri
i=O
0 <
< I,
we have
(0 <
(0 <
I'
(1 ).
<
(0 <
< 1).
rn+i
1
1-r
- , n_r_
1- r
--1- r
= -
1
- 1- r
-+oo
limA =
n
+ r
n
+I
If we know that
then it follows that
1- r
+ ... + r =
rn
< 1);
< 1).
(1)
(2)
(3)
Since
for every n.
and
Jim
n-+
rn+i = L.
oo
(Why? What happens to the limit of a sequence, if you omit the first term?) Therefore
L = Jim
11-00
and so
Since
rn+
L = rL,
- r
0,
n-oo
observation:
If limn-oo lan
l=
0,
rn
= rL,
(1 - r)L = 0.
and
Proof?
= r Jim
it follows that L = 0.
Jim rn =
n-+oo
Theorem 1.
Therefore
(0 <
-1 <
< 1).
0.
then limn-oo an
=
0,
(If you rewrite these two statements, using the definitions of the statements
limn-oo lan
l=
and limn-oo an
=
0,
Infinite Series.
10.2
Theorem 2.
If -1 <
Convergence.
Comparison Tests
439
< 1, then
= 0.
limrn
n-+oo
Algebraically, the formula
1 + r + y2 + .
holds for every
Theorem 3.
n
+ r = (1
00
i=O
Theorem 4.
r)
< 1, then
Lri
a,
If -1 <
is any number
n
- y +l)/(1
-- .
= 0/(1
r) = 0.
If -1 <
< 1, then
oo
Lar'
i=O
= -- .
1 -r
Proof
If
L:,1 ai
For each
n,
Jet
= A,
lim(An - An_i) =A -
an-
Therefore limn-oo
an
where
n->oo
But An - An-l =
diverges.
> I. Therefore
A= 0.
L:o ari is divergent for a :lrln lal, and so an does not approach 0.
lanl = lal
Warning.
0 and
lrl
I.
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 +
Here
an-+
diverges to infinity.
A more natural example of the same phenomenon is the
00
i=l
I=1+-+-+ .
harmonic
+ -+
n
series
10.2
Infinite Series
440
For each
n,
Briefly:
2;;:1 (l/i) =
=n +
ln+l dx
-.
and
(n + I);
Theorem 6.
/1
limln(n
diverge to infinity.
1 to
1
1
1
=1+-+-+ .. +- >
"
n+I
An
+ 1)
oo.
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
converges. But the algebraic pattern suggests that the series is related to the improper
integral
100 dx la dx
[-l]a (-1 l)
x a-+ x a-+ x a-+
/x2
x, "
ld
.
. Since 1
2 = 1Im
00
1.
Im
co
1.Im
00
l.
< 1
for every n.
Infinite Series.
10.2
Convergence.
Comparison Tests
441
1/x2 is decreasing,
(n
>
1).
1/n2.)
{ 2dx
J1 x2 '
l_
22
and
<
An
A A 2,
1,
2
i=l
Therefore
l_
32
2 < 1 +
n
<
r s dx
J 2 x2 '
n dx
J1 x2
< 2.
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
Then
(1)
if
Then
ai bi
with
for each i.
2:1 bi is convergent, then so also is ,Li':1 ai; and (2) if ,L:1 ai is divergent,
.L:1 bi.
then so also is
Proof
Let
For each
n,
let
442
10.2
Infinite Series
(Why?)
is increasing.
An
( l)
i=l bi
2
is convergent
is convergent
is bounded
is bounded
is convergent
00
=>
2a;
is convergent.
i=i
(2)
in the
00
i=l ai
2
is divergent
is divergent
is unbounded
is divergent
00
=>
i2=l bi
is divergent.
1 =1+-+-+-+
1 1 1
i2=O i !
1 2! 3!
n (n
n! =
00
Here
and
O!
Then
i, let
1)
l.
for each i;
- <
Consider for
10.2
Infinite Series.
Convergence.
Comparison Tests
and thereafter the strict inequality holds, with l/n ! < (l/2r-1 for
443
2. Therefore
9.
Therefore:
In fact,
I !
i=O
ln-
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
1. (You have seen a situation like this before. The easiest way to
find
I.)
Consider
next
00
/n
Since
1
n
>
for every
n,
and
1
n
)n
Cf).
While the comparison theorem tells us, under some conditions, that a series
converges, it never tells us what the sum is.
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
some large
n)
to get an approximation of
,L;:1 a;.
Lf=i a;
(for
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.
Infinite Series
444
10.2
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
2 Vi:
i=l ----:
oo
3.
6.
1
4.
i j3/2 + 2
oo
00
.2 (-l)i7T-i
i=l
10.
00
cos3
(2i)
j2
oo
13.
16.
19
14.
il j l-1
2 l 1n I .
i=2
17.
.,
i=O
00
25.
2 8.
(i!)3
---
2
i=l i
.,
i=l
(i + l)(i + 2)
+ 1
20
i l Vi + 1
2i
i=l
sin2
0 9
i j.
1
1
i=2 l:a-n
18.
_2
z
i=2 1-:--1n
co
1
21.
iO (i !)2
co
1
24. 2
i=2 i(i - 1)
2 i(i+1)
i=l -
00
.2 -.-
i=2 1 n
00
2
.
i=2 z:---1n2 z
.2
ii'
00
15.
(2i - 1)
j2
i =l
00
12.
oo
23.
9.
i
i=l j
00
i2 ln2 i
(i ! - 1)
22.
L ( -2)ie2i
i=l
00
00
11.
.2
= --
00
00
oo
8.
)
i(
00
i j 3 /2
26. 2
(i - l)(i)(i + 2)
i=2 ---
27.
21_
30.
i_
i -._
+ 1
i=l
i:
i=l j2 - 1
--
1
=1 +r+r 2+
1 -r
That is,
1/(1 - r) can be expressed as the sum for an infinite series. Express 1/(1+x)
32. Express 1/(1 +x2) as an infinite series. For what numbers x does the series converge?
*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
.3
:Li=l I + 1
35.
*37.
00
ct.
is the series
2i2
1 I1
i=l
+
36.
oo
) . If
ioo
(1)
then
00
:L /Ci)
i=l
< w,
(2)
and conversely.
10.3
Given a series
_L!o ai
00
Lai
i=O
then
L( - l)iri
i=O
=
L lrli
i=O
of each term
- r + r2 -
00
00
_L la;I
i=O
la;J
1 +
lrl + lrl2
1 a;
a;.
For example, if
lai l converges.
1 - 1 + t- t + i - i +
For example,
: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
Here it is plain
convergent.
:L ai :L c-1)i+ 1--: 1
i=l
i=l
l
that L la;I diverges, but it
=
Proof
Let
1
-
- - + ...
4
La;
is
This is worth proving, however, because the idea used in the proof is
446
10.3
Infinite Series
If n is even, with n =
2k,
then
2
)+(
An = Ao.
"k= 1 - I_ + I_ =
(1
)4 .+
1
2k
- l_
...
(2k 1-
)
2k .
1
bound, because
G ) ( - )
A2k = 1 -
has a limit.
ck 2 - 2k 1 ) 2k1
-
<
1.
Let
A= lim A2k.
k-+ 00
(1)
We shall show that A is the sum of the series. First we p.bserve that
lim A2k+1= lim [A2k
so that
le-> oo
k-> oo
+ a2k+il
lim A2k+i
k--+ 00
Thus we see that
lim A2k
k->
oo
lim
le-> oo
2k
1
+
(2)
A + 0= A.
(2)
as n--+
oo
through
i: for every
(2)
If you
reexamine the proof, you will see that the only facts ,about the series
1)
The series is alternating. That is, successive terms a;, ai+I have opposite signs.
3)
2)
Limnoo an
0.
Absolute Convergence.
10.3
447
Theorem 2
l i mn_,00 an
Alternating Series
(Strictly speaking, some of our formulas in the proof of Theorem 1 used the fact
that the first term was positive instead of negative.
holds in this case, how would you show that it also holds when a1 < 0 ?)
We have seen that if
La;
L la;I
converges.
If
If
.L la;I
_L;:1 la;I
is convergent, then
L
. a;
is said to be
,L;:1 a;.
absolutely convergent.
In this
into a sum of positive terms and a sum of negative terms. To do this, we let
if a; 0,
+_ a;
a;
-
and let
if
if a; 0,
a;
a; =
0
_
a; < 0,
if a;> 0.
Let
n
A+
n
A-;;-= _La;.
i=l
"" ai'
+
L-
i=l
Then
for each
n,
for each
i.
because
Obviously
A, A;, ... is
.L la;!.
i=l
Then
n
L ia; I
:::::; k
for every
n.
i=l
Also
n
A! _L la;I,
i=l
because
is the sum of some (perhaps all) of the terms on the right-hand side.
Infinite Series
448
Therefore
10.4
At, A, . . .
is convergent.
Let
A+=
lim
A-;;-
I ( - lail),
Similarly,
because
A:
A;;.
i=l
is the sum of some (perhaps all) of the terms on the right-hand side;
Then
lim An = Jim (A;;
n-+co
n-+co
and
I1 a; is convergent,
Theorem 4.
If I1
/a;/
Let A-
limnco A:.
A-;;) = A+ + A-,
is convergent, then
Ii a;I ; /a;J.
Proof
We know that
Ii a;I ; /a;/
for every
n.
Find out which of the following series are alternating, which are convergent, and which
are absolutely convergent.
"'
i=l
4.
1.
10.
10.4
"'
1
<-1)i -:{I
i=l"'
i1
i=lI c-2))2i-:-i
(i=lf
l
l.
s.
i=l"' 2
1
I 2
i=lI c ni1
(
i=lI ::.!_2
-
i=l
11.
'
..
i=l
( -1)' -. + I
+ i
"'
s.
3.
9.
sin
cos
rri)
2
l
12.
ESTIMATES OF REMAINDERS
00
:L
i=2 c
t)i ' (i
00
i=l
c-i)-i
1)
Estimates of Remainders
10.4
449
A= lim An = L ai.
n-+co
i=l
The approximation An
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
Rn=
and obviously
L ai,
i=n+l
limRn=
n-+oo
0.
For alternating series, of the type treated in Theorem 2 of the preceding section,
it is easy to get an estimate of R,,,.
ct)
ct)
i+l =
"" ai = ""
b;
b1 - b2 + ba .., ( -1)
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
'
' ' 0.
Therefore
1)
0 Rn hn+l
when
is even.
If n is odd, then
Rn =
=
Rn =
Thus
2)
-bn+l Rn 0,
when
is odd.
Therefore
Since
450
10.4
Infinite Series
Theorem 1.
0, and
for every
IRnl lan+il
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
i
1 - _!_2 + _!_2 - . .
2
3
.
i=l
1
1
1
+-
Al--+-2 '
2
2
2
C()
I c lY-:i=O
l!
-
sum
1 - 1 +
A.
1
-
2!
- - + ...
3!
2!
3!
10 !
1/e.) We have
ARJ--- + +-
'
and the error is less than 1/11 !. This series converges very rapidly:
11 !
39,916,800,
and
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,
1
C()
I-:;.
i=l l
We let
An
n 1
-:2 <
i=l l
ln
1+
dx
2
x
Estimates of Remainders
10.4
451
= l_
R
.,,,
i=n+l l2
<
oo
dx.
9
Since
we conclude that
Rn
< 1
-
for every
n.
This is nowhere nearly so small as the estimate of error for the corresponding alter
nating series.
(1/i2)
Similarly, Theorem 4 of Section 10.3 gives an estimate of the error for series
Suppose that
L lail
is convergent.
00
Rn= L ai.
i=n+l
Let
10.4
Infinite Series
452
Then
00
IRnl
Iai
lail
i=n+l
I la;I-
To
I ai,
i=n+l
If we use the comparison theorem of Section
10.2,
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.
found that
00 1
2
< l/i for every i, it follows by the
I-. -- <co.
i=l i2 + 1
It also follows, for the remainder
and so
R < l.
Each of the following series is convergent. In each case, get an estimate of the remainder
Rn, in the form \Rn\
I.
i; ( ym
i=l 3}
2.
00
3. I ( -l)i7T-i
i=l
00
sin2 (2i - 1)
5. L
i2
i=l
00 ( - l)i+l
7. I
i=l i-4
w
9.
L
i=l
( - l)i
i0.9
4.
6.
i: (- 4)
i=l
i; co 7Ti
l
i=l
00
il f3
8.
w
10.
L -:--12.
1 n 1
i=2
10.5
1 1.
co
12.
i O (i!)2
13.
15.
co
1 i(i
14.
+ 1)
co
16.
[a, b].
Let
( --r
co
2
i=l
co
i 2 i2(1 + i)
co
fv fz,
453
i l i(i + l)(i + 2)
*17. Let
I c -o-i
i=
..
f
n-co
fn(x)
f(x),
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
* 18. Consider
order
19.
10.5
i
I!1 ( -l) +l (l/i).
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.)
TERMWISE INTEGRATION OF SERIES.
A power
series is a
x, including x = 0.)
1, so that
a0x0
a0
. -
co
1
x'=
I
i=O
1-
for every
x for r in
(-r, r),
co
f(x) = L a;xi
i=O
(-r<x<r).
Given
co
f(x) = L G;Xi
i=O
(-r<x<r).
(-r, r),
10.5
Infinite Series
454
(-r<x<r).
Given
00
f(x) =I aixi
i=O
(-r<x<r).
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
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 <x<1).
- x) = 1 + x + x2 +
- x) as a power series:
x by -x, we get
l
__ = 1 - x + x 2 - . .. + (-lrx" + ..
1 + x
and then, replacing
(-l<x<l );
x by x2, we get
(-1 <x<1).
Theorem B says that the series on the right can be integrated a term at a time. Thus
lx
dt
ol+t2
--
lo"'dt -lo"'t2 dt
lx
0
dt
1 + t2
=x -
+ (-1)"
3
x
l"'t2n dt
o
x2n+1
+ ...(-l)" 211 + 1
00
i 2i +l
=I C-l)
i=o
2i + 1
(-l<x<l ),
+ ...
(-1<x<1).
x. Thus we have:
10.5
Theorem 1.
Tan-1
. x2i+1
oo
x =I (-1)' --.
21 + 1
i=o
455
(-l< x<l).
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.
could have been predicted, because Tan-1 is an odd function, with Tan-1
x.
This
( -x) =
The same method can be used to get a series for the natural logarithm.
Theorem 2.
In
(1 + x)
x2
x3
x- - + 2
We know that
In
1
1 +t
--
By Theorem B,
lx
dt
(1 + x) = -- ,
0 1 +t
dt
--
o l +t
x2
x3
=x--+2
00
=IC-l)i+l
l
i=l
l)i +
.
(-l<t<l).
"'
f
+ c-1Y tiat +
o
xi
-
Note that this method cannot be used to calculate the integral from 0 to 2, because the
series for
1/(1 + t)
!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.
the series for
1/(1 + x),
we replace
1 + x4
---
1 00
x by x4 This gives
4 + x8 -
=I c-1)ix4i.
i=O
.
+ (-l)'x4'
In
456
10.5
Infinite Series
Therefore
-112
dx
4
1 + X
--
oo
=
I
i=O
-112
(-l)'x4'dx
I< -l)i
< ) i+l
_ -t
i=O
+ 1
4z
-t - t(-t)5 + -H-W +
1
-! +_
.
25
+ ...
9 . 29
1024, 213
8192, and so
E < I0-5
2.
Calculate Tan-1
l/10
0.02 to six decimal places, and explain how you know that the error in
Calculate
7
10-
1
---4 dx to five decimal places, and explain how you know that the
1 + x
10--G.
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 +
0.6
J
o
as an infinite series.
1 +
6
x
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.
same reasons.)
v,
1/(1 +
vi),
on the interval
[O, 0.49}
10.6
457
1
5 2 , on the interval [O, 0.25].
1 + x 1
1
3, on [O, 1/2].
1 + x
rk [ f--=:_J
Jo i=oi +
dx
10. Using the first term only, in the series for In, we get the approximation formula
In (1 + x)
R:J
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
1 + x +
x2
x3
+
3!
n
x
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)
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 .
fn(.Xn ). Get a formula for Ym and find liffin_,00 Yn
b) For each n, let Yn
Consider a series .Lai, in which the terms may be positive or negative, but not equal
to 0. For each i, let
458
Infinite Series
10.6
so that
lai+II
An examination of the sequence
r1, r2,
lail ri.
i_..oo
then
2:0 ai is
ri
r <
1,
absolutely convergent.
Proof Let s be any number such that r < s < 1. Then there is an N such that
i N
ri <
=>
s.
r;
(iN)
ri, take
laN+II
laN+2I
s - r, so that r
s -
) It follows that
laNI s2;
By induction,
Therefore
"'
for every j.
"'
"'
i=O
i=O
laNl2s'
i=O
It follows that
and so
co
2 fail
i=O
N-1
=
2 [a;[
i=O
[aNI -- <
1 - S
oo.
co
2 lail
<
oo,
i=]\T
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.
10.6
459
Io ( l/i ! ) is
convergent. The ratio test gives this result very quickly. We have
1
a;=
l.
1
i!
Qi+l
- -- '
r-=-=
'
(i + 1)!
i + 1
a;
and so lim;co
There are simple cases in which a series converges, but in which convergence
cannot be established by the ratio test.
Consider
co
!-:;.
i=l I
which is known to converge.
Here
i2
a;+1
r; = - =
.
(i + 1)2
a;
Therefore, while
r; <
I.1m r;
i-co
]"1m
i- co
1
(1
+ (l/i)]2
= 1,
r; ---+
i
r;= -----+ 1.
i +l
oo,
I c-1)i= 1 i=O
Here
+ 1 - 1 +
so that
ri
---+
for every i,
On the basis of the ratio test, we can derive a more general result for power series:
Theorem
where
0 for every
a;
i.
Suppose that
Jim
t-+ro
a;+
i = L.
ai
I I
x.
Infinite Series
460
Proof
10.6
For each
- 0, we have
Therefore
lim ri = lxl
If L = 0, then
ri
--+
L.
whenever lxl < I/L. In either case, the series converges absolutely.
ri <
converges absdlutely for every x. By the second half of Theorem 2, we see that
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
1.Im
.
1.-+00
ai
I I
+l
Qi
ci + 1)11'i+l
= 11'.
. i
n-+OO
l'TT
1.
= Im
_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,
This
_Lo a;xi,
with
a;
+1
i-+oo ai
Jim
If L
I I
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
10.6
x2
Setting
461
y, we get
00
.L (-1)y\
i=O
which converges absolutely for
series converges absolutely for
IYI <
/xi <
IYI
/xi
> 1.
x2i+1
oo
.L i
i=O 2
Here
.
lIm
i-+ oo
Similarly,
x2i
x_L--:-.
i=O 2'
oo
I I
ai+l
a;
1
=-
x2 < 2
x2
2.
>
/ai+1/a;/
..
i=O
Here
a;
even i.
1 for
i,
every
x'
and so
r;
1 + x + x2 +
1 for every
i.
We now divide
xi by i!
for every
This gives
oo
.L b xi
i=O i
The series still converges, for
2)
x+
/xi < 1,
x2
2!
x3 +
4!
The theorem tells us that the series converges everywhere on the open interval
(-1/L, l/L),
interval.
converges on
( -1, 1),
The series
(xi/i2)
.L:1 ixi
_L;:1 (xi/i)
The series
diverges at
we have to
3)
L a;xi
converges for
;tf 0, we have
r;
O; the sum is
l(i + l)!xi+i/i!xil
(i + 1) /xi --+ oo.
0.
a0
.L:o i!xi.
But
For
462
4)
Infinite Series
10.6
Finally, the results that we have been getting for power series suggest a conjecture.
.L aixi
has
( - oo, oo).
(-a, a), plus, perhaps, one or both of the endpoints.
.L a;xi
origin?
.L a;xi
( - oo, oo),
(-a, a),
[-a, a) ,
(-a, a],
[-a, a],
{O}.
ro
1.
i=1
ro
4.
_L ix2i
i=l
oo
7.
10.
13.
oo
5.
xi
3. _Li2x2i-1
i=l
x2i
xi
ro
6. .L
i=l VI
.L
i=l v'l
(-x)i
co
ro
; v2i + 1
8.
00
I -=
i= Vi - I
9.
11. L (3i)2x2i-1
i=l
x2 i
14. IC-l)i
(
2'I )'
t=O
.L (3i)x2i
i=l
x2i-1
ro
.L (-l)i+l ( . 21
i=l
ro
15.
17. _Li-ixi
i=l
x 4i
L ( - l)i (2 I')I
t=O
ro
00
16. _Lhi
i=l
12. L (3i)4x3i
i=l
oo
1)!
.L(3i)3xi
co
ro
00
19.
ro
2. _Li3xi
_Li2xi
i=l
18.
i=l
,L (Tan-1 i)xi
i=l
I (-l)i
i2 + 1
i=l
00
ei
oo
i=ll3
oo
(x - 2 )i
22. .L
i=l
Show that
24.
,L:1
Theorem. If
,L;':1 a;
20.)
20.)
23.
3 ?)
Jxl
< 1.
is absolutely convergent.
10.7
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
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 <
i=O
then
x <
r ,
co
f'(x) = ,L ia;xi-l
i=l
( -r < x
r).
<
We shall use this to find a series for the exponential function. We start by assum
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 +
i=l
ia;xi-l
;
(i + l)a;+1x +
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,
and, in general,
a; = 1/i!.
a3
= a2/3 = 1/ (2
3) ;
10.7
Infinite Series
464
1
-i!
=>
2,
the formula
a;
1
1
G;
.
i+l - i + 1 - (i + l)i! - ( + 1)!
i
--
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.
i
x
g(x) =I-:-
i=O l !
(-oo <
< oo).
show that g(x) = e"' for every x. For each x, let ef;(x) = g(x)/e"'. Then
ef;'(x) =
e"'g'(x) - g(x)e"'
e
2x
1
= - [g'(x) - g(x)]
e"'
We need to
0.
But ef;(O) = I.
Therefore
g(x) = e"',
and
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"' =
Lo (xi/i!).
Setting x = I we get
Theorem 2.
e =
Lo (1/i!).
This series converges so fast that some people enjoy using it to calculate e
2.7182818,
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,
Therefore
and
f(O) = 0,
f"(x) =
f(x)
-
f"(x) = -sin x.
f'(O) = I,
for every x.
10.7
465
Thus if
.
sm
we must have
a0 =
0,
x=
co
..
i=O
and a1 = 1.
a;xi
a0 + a1x + .
Now
f'(x) = L ia;xi-1,
i=l
co
i 2
= 2a2 + 32a3x + + i(i - l)aix + (i + l)ia;+lxi-l + (i + 2)(i + l)a;+2xi +
To getf"
(i
Since
a0
-f, we want
2)(i + l)ai+2
-a;,
a;+2 = -
or
(i + l)(i + 2)
a1 = 1,
as=
(1
a1
1)(1
2)
23
2)
as
(3 + 1)(3
a5 =
a1
---=
and in general
3!'
1
3! .
4. 5
1
5! '
i
a2i+l - (-l) .
(2i + 1)!
i
a2i+l - (-l)
=>-
ac2i+i>+2
(2i + 1)!
a2i+l
-[
(2i + 1)
- c
1] [(2i
1) + 2]
1
1
.
(2i
1
(2i
3)!
1)! . (2i
-
c -1Y+i
2)(2i
3)
[2(i + 1) + 1]!
Therefore, if there is a series for the sine, the series must have the form
(x) =
co
x2i+1
466
10.7
Infinite Series
I) g'
h,
2) h'
-g,
3) g(O)
= 0,
4) h(O) = I.
It ought to be true that
h(x) =cos x.
g(x) =sin x,
If so, the function
cp'(x)
=2[g(x)
=0
Therefore
<P
sin
for every
is a constant. But
0.
Theorem 3.
oo
sin
I (-1)'
i=O
By differentiation,
x2i+1
(2i + 1) !
oo
COS X =
.? (-1)
i-0
=x
x3
(2i + l)x2i
(2.l
Thus:
xs
x7
-- + - - - +
3!
5!
7!
l) .I
oo
x2i
t-0
.? (-1)i (2 . ) I
.
Theorem 4.
cos
oo
. x2;
x2
xs
x4
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.
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.
example, consider
o
r .5
Jo
'"2
e
dx.
F'(x)
= e
"'
For
10.7
467
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
x
e =
ex'
Therefore
ro
ro
=
L
i=D l !
2i
x
L--;-i=O l !
a
x
"
x
2!
..
ro
F(O)
and
0,
f'
t'
F'(x)
dt =
2i+1
I --- .
-i=oi!(2i
+ 1)
Evidently
Therefore
2!
- + -- +
3
1 + x + - + ...
F(x)
+ -+
2!
1 + X
ex'.
F(x),
F(t)
small as we please.
1. f(x)
x In (x
which you know that your series converges to the given function.
3. f(x)
5. f(x)
7. f(x)
11. f(x)
9. /(x)
13. f(x)
15. F(x)
17.
F(x)
2
x ln (x
I).
2x.
sin (x/2).
sin
cos
Jo
10. f(x)
G).
3 "'3
x e
l"'
6. f(x)
8. f(x)
+ I).
3 t3
t e
12. f(x)
dt.
lxf(t)dt,
wheref is as in Problem
14.
14. f(x)
16. f(x)
18. f(x)
x sin x.
cos 2x.
2.
sin x cos x.
'n-'x
"'
xe - x.
(e"'
Problem
for x
for x
for x
;e 0
0
;e 0
0
=
x ;e 0
for x
0
for x
for
468
Infinite Series
19.
F(x)
21.
f(x)
23. f(x)
10.8
f'f(t)dt,
x3 cos2 x.
20.
f(x)
22. F(x)
cos2 x - sin2 x
24. F(x)
cos2 x.
{ f(t ) dt where f i s
J as in Problem21.
o
x cos2 x + x sin2 x.
26. Find a series for a function f such that (1) /' (x)
/(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.
>
j.
I:o a xi. Get a formula forJ(i>(O). (Here /( i ) denotes the ith derivative
i
*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
f(x)
L aixi
io
<XJ
L b;xi
on I,
io
*33. A
10.8
(a + bt
an +
nan-lb +
n(n
n(n - 1) an-2 2 +
b
2
1) (n
l1.
- i
+ 1)
an-ibi + . . + bn.
(n)
i
n(n - 1) (n
n!
=
!(n
i)!
1
1.
- i
+ 1)
10.8
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
to take care of the case in which
i G) an-ibt
(a+ b)k, where k is any real number. The following observations are obvious:
1)
Fork
0, we have
(a+ b)k
assume that
k
2)
c>
only for
0.
(See Section
4.9.)
k is
0.
a+ b>
3)
For
ck is defined
(2a)k
2kak.
a-:;!:. b.
And we want to assume this, because the case
going to emerge.
4)
We let
a+ b
If we had
a> b.
bfa, so that
\x\
a(l + x),
and
Given
k -:;!:.
0, and
f(x)
(I + x)k
f(x)
(1 + x)k
j'(x)
Therefore
k(l + x)7'-1.
(1 + x)f'(x)
f'(x)
CQ
L ia;xi-l
i=O
kf(x).
CQ
L ia;xi-1.
i=l
470
10.8
Infinite Series
Therefore
xf'(x)
C()
.L ia ixi.
00
.L (j + l)a;+1x1 = L (i + l)a;+ixi.
f'(x) =
1=0
00
i=O
Obviously
a0
Therefore
/(0)
<=>
(i + l)ai+l
<=>
a;+i =
(1 + 0)1'
(k -i)ai
k-i
ai.
i
+1
1.
a0 = 1;
a1 =
O a = k'
0+1 0
k(k - 1)
k -1
a.= -- a =
1 + 1 1
2
'
k -2
k(k-l)(k - 2) - k(k- l)(k -2)
a ---a
32
3!
'
3 2+1 2
>
1,
ai =
k(k - 1) . .. (k
.,
I.
i + 1)
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
0,
'
.4..
i=O
k
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
Theorem 1.
Proof
10.6,
As in Section
Then
r
'
= I k(k
=
Evidently
2::o CDxi is
The series
convergent for
lxl < 1.
let
1)... (k
i + l)(k
(i+l)!
_
i).
i!
k(k-l)(k-i+l)
r
; =
Ix!.
Theorem 2.
For every
0,
lxl <
Let
1, and diverges
x
i ()xi= (1 + x?.
and every
between -1 and 1,
g
l)g(O)=l;
2) (I +x)g (x) = kg(x).
Proof
xI
Jim
lx l
I ; x I
i-too
for
471
We determined the
'
g(x)=f(x)= (1+x)k
(-1 <
x < I).
For this purpose, we use the same device that worked for the exponential.
f(x)
Then
= (1 g(x)
+ x)k
x) - (1 +x)kg'(x)(1-+ gx()xzk)k(l+x)k-l
kg(x)
= (1 + (1x)g+'(x))
g(O) = f (0) =
x,
e .
A. '
't' C
x k+1
g(x)=j(x)
for every
Let
'
1.
Therefore
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.
10.8
Infinite Series
472
2.
.a;
x +
Do the same, for v
3.
4.
5. Let
1.
+ 1.
V2x + 1.
n!
(n - i)!i!'
write formulas for
6.
7.
<1 + xr =
6,
show that
i ( ) xi
i=O
"""
(The first and last terms on the right-hand side require a separate discussion.
note that ()
<nt1), because both are equal to 1 ; and similarly that ()
(!})
=
Since obviously (1
+ x)1
() + ()x1,
But
= 1.)
f(x)
8.
v'I + x
14. f(x)
Sa"' Vl + t10 dt
f'2 VI + tlo dt
f(x) = (2 + x2)"
20. f(x)
22.
x
=
15.
V 1 + x2
--
19.
x2
.I
v
1 +x
.a;
v
x
1 + x2
(1
+ x)312
f(x) = {/2 + x
f' {/2 + t2 dt
f(x) = r (2 t2)k dt
=
J v'1 + t2
0
dt
23.
(k ;6-0)
13. f(x)
17. f(x)
j(X) =
11. f(x)
1.
Taylor Series
10.9
10.9
473
TAYLOR SERIES
1/x
= oo.
f C-l)ixi
1
__
1+X
i=O
If we let
x'
1 + x, x
(lxl <1).
f(x)
1
=
00
!(-l)'(x - 1)'
(Ix - 11<1).
i=O
00
! a;x;
00
i=O
i=O
1/x
1.
In x cannot have a Maclaurin series, because at x = 0 the
! a;(x - O)i,
has no Maclaurin series, but it does have a Taylor expansion about the point
Similarly,
f(x)
function approaches
- oo.
g(x)
Setting
x'
1 + x, x
x)
oo
xi
!c-1r1--:-
i=l
(l xl<1).
x' - 1, we get
Inx' =
)i
(x
i(-l)i+l ' l
i=l
(Ix' - 11<1);
(x l)i
ic-l)i+i -:l
i=l
(Ix - 11<1).
x = x' + a, x' = x - a.
10.5 take the following forms:
by the substitution
of Section
Theorem A'.
Given
f(x)
00
! a;(x - aY
i=O
(a - r<x< a + r),
10.9
Infinite Series
474
(a - r, a + r),
(a - r
<
<
a + r).
(a - r
<
<
a + r).
Given
ro
f(x)= L a;(x - aY
i=O
Then the integral of the sum is the sum of the integrals. That is,
("'f(t) dt
Jo
("'a;(t - 4 dt
i =oJo
(x - a)i+l.
I
+ 1
i=Ol
x.
a =
n
x - a
in
j<n>(o)
--
n!
I:o a;xi,
on an open
If f(x)=
Lo a;(x - a)',
a =
n
Proof
by
on an interval
J<n>(a)
(a - r, a
r),
then
for every n.
--
n!
Dncp(x).
We observe that
nn(x - a)i=
nn(x - ar=
n!,
for
Therefore,
<
forf(x)= L a; (x - a)i,
fCn>(x)
a
n
n,
(i - n + l)(x - a)i-n
for
>
J'l.
we have
ro
n!
+ L b;(x - aY-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)
This gives
so that
a =
n
which was to be proved.
fCn>(a)
--
n.1
'
a.
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,
expansions.
.Z
i
a ix
0, and more slowly when xis larger. To take an extreme case, we know
10,0007T = 0, because 10,000 is even. Therefore it must be true that
is close to
that sin
o.
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"
substitute.
way to calculate In
1.6
would be to take
ln
1.6
x =
Jn
x,
we have
f'(x)
f"(x)
fil(x)
ai
Jn
=
1.6,
fi)(l.5)
.,
L
0.4055.
One
.I c-1)i+l co"6)i
l
00
.2 a;(x - 1.5t
i=O
1/x
1.6
x-1,
( -l)x-2,
(-l)i+l(i - l)!x-\
( -l)i+l
.
(1.5)-i
(i > 0),
0.4055.
0.4055 +
this gives
In
j(O)(l.5)
Therefore
For x
For f (x)
1.5
x
1
l(-l)i+1C y
l
=
i l
i=l
is slow.
in the series
(- ) +l ( x - 1.5);
I
.
1.5
0.4055 +
t=l
( - i.y+i (1-)i,
i=l
15
1.5;
ai
f(a) .
.
= -,L
476
Infinite Series
10.9
f(x)
converging on an interval
Ix - al
<
! ai(x - a)\
i=O
then any other point b of the interval can also
r,
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.
cases where the derivation of the general formula seems unreasonably difficult, merely cal
culate the first three terms of the series.
a=0.
a = 1T.
3. f(x) =tan x,
a = 0.
5. f(x) =Tan-1 x,
a=0.
7. f(x) =ex,
a=0.
9. f(x) =In (2 + x),
a=0.
11. f(x) =Jn (1 + x2),
2. f(x) =tan x,
a=0.
a=21T.
4. f(x) =cos x,
6. f(x) =Tan-1 x,
a=1.
a=1.
8. f(x) =e"',
a=any number.
10. f(x) =e"',
1. f(x) =sinx,
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
= i(a; + la;I),
ci=i(a; - la;I).
b;
15.
b;,
oo,
then
and
LC;
*16. Let
n1, n2,
exactly once.
- oo.
n1, n2,
2:%1 a;,
1, 2,
3,
. . . arranged in
!%1 an,,
Taylor's Theorem.
10.10
Theorem. If
ai
2 ai
and
Estimates of Remainders
477
2 a;
Prove this.
*17. Show that
has
a rearrangement
which converges to 0.
a rearrangement
converges to k.
10.10
f(x)
co
.2 a;(x - aY
i=O
ai
i
J< >(a)
.-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)
(i)(a)
J
.2.
i=O
(x - a)i
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 !
-
Rn(x)
for some x between a and x.
f(n+l)(.X)
=
(n + 1) !
(x - ar+1'
10.10
Infinite Series
478
t we let
i
f (t)
i
f(x). (x - t) .
i=O l !
Here we have simply replaced a by t in the formula for R (x).
n
F(t)
F(x)
For
a we have F(a)
F(t)
f(x)-
r<ol(x)
--
O!
f(x) - f(x)
R (x). Since
n
f"(t)
f(t)
f'(t)
(x - t)2 f(x) - -- - (x - t)O!
1!
2!
F '(t)
x we have
0.
we have
For
J(n)(t)
- -- (x - tf,
n!
J(n)(t)
f(n+l)(t)
(x- tr- -- . n(x - tr-1 .
n!
n!
Here all terms cancel out, telescopically, except the first term in the last bracket;
and so
F'(t)
Now let
f(n+l)(t)
(x - ir.
n!
(x - tr+i
G(t)- -(n + 1)! '
so that
G'(t)
-(x - tr
=
n!
F 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
To the functions
F(x)- F(a)
F'(x)
G(x)- G(a)
G'(x)'
_
for some
between
-F(a)
And
-G(a)
G(x)
F '(x)
G'(x)
10.11
479
F(a)
G(a)
+1
f<n >(x).
F(a) = J<n+i>(x)G(a)
x)
f(n+l)((n + 1)!
(x
ar+l'
IJ<n+i>(x) I
for every
and
x.
Rn(x)
--+
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)
f(x)
1 -,
+
x3
1
even though we can easily see what the Maclaurin series is:
_1_ =I c-i)ix3i
X3 i=O
1 +
(lxl
<
1).
JC3i>(O) = ( -1)i(3i) ! ,
and thatJ<nl(O)
0 if n is not divisible by
x.
3.
10.11
=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
480
Infinite Series
10.11
perform calculations.
For example,
/2
(_;2
,1
Jl
I;
i4 =(i2)2 =1;
l0,001
i
= i;
G 3
+
)2
iy = i(l +
= i(l +
= i(l = -1.
= 0+ 0
_13 i)3
1
1
a - bi
a - bi
--- = --- . --- =
a + bi
a + bi a - bi
a2 - (bi)2
=
a - bi
a2 + b2
=A+ Bi.
a2 + b2
-b
+ bi and
a
(a+ bi)(A + Bi) =(a + bi) (
+
a2 + b2
=
Therefore A +
(a + bi)(a - bi)
a2 + b2
a2 + b2
-b
2+ b2
a2 + b2
a2 + b2
bi have reciprocals,
i
=1.
There are several ways to define the set of complex numbers, as a mathematical
Meanwhile we shall regard the complex numbers as known, and calculate with them,
using the familiar laws of algebra and the fact that
The
=a+ bi
=a - bi.
is the number
i2 =
10.11
The
ab solute value
of
481
is
lzl = .Ja2 + b2
By straightforward calculations, we get the following.
Zi.
z2,
we have:
z = z,
z + z
is a real number,
is a real
zz
2
!zl = z z,
number,
Proofs.
z = a - (-b)i = a + bi = z;
z = a - bi,
z = a + bi,
z + z = a + bi + a - bi = 2a;
z z = (a + bi)(a - bi) = a2 + b2 = !zl2;
lz1z2I = .Jz1z2z1z2
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 =
z = iy ,
real axis.
of this as the imaginary axis. This explains the labels on the axes in the figure below.
Evidently
is the reflection of
where you started; and this is the geometric meaning of the equation z
= z.
As the
482
Infinite Series
10.11
z=x-yi.
z
figure suggests,
I
I
I
____
_r:jI
I
For
z1
= x1
+ y1i and z2
= x2
Z1 - Z2
+ y2i, we have
SO. that
lz1 - Z2I
.J (x1
- X2)2 + (Yi
2
J2) ,
number
=x
+ yi
Re
z + i Im z.
i2 =
=
=
t(z + z),
(1/2i)(z
z).
-1, we have
.
l =
-i,
Imz
and
= (z
2
- z ).
These formulas enable us to connect complex numbers with the geometry of our
x=
<=>
Re
;>-
t(z + z)
;>-
z + z
2.
10.11
483
In the following problem set, you will be asked to carry out a variety of such
processes.
which complex numbers are the only variables. Thus the vertical line discussed above
is the graph of the C-equation
C-equation zz
+ z
4.
I. (1
4. (-1 - _1 r
1
J
+ i)4
v2
7.
10.
v2
v3
- +21
(2
2
8. (3 - iJ
11. (1
14. (1
.
20. 2i 1
+ i)2
(v2 - v2 i)8
(2 + i)2
2i +
25.
v3 + 2i
28.
i3 + j2 + i
31. 11
34.
+i
v3
-1
32.
+I
Show that zn
37.
39.
41.
43. -
27.
2i
Re z +Im z
Re z =Im z.
lz
II < I.
I.
z.
40.
+ i)3
v2
_1
I + 3i
v3 i
i2 + i + 1
1
(i +
2i
---
1 +
36.
38.
v2
30. 1)3
33. 1 -- 3i
2 +
i4 + i3 + i2 + i + I
1 r
v2i
- i + 1
- i
is + i4 + i3 +
35.
26. 29. i2 1
3i
2i
G( _1
( -1
2 .
2 + v3 i
23.
v2
2 - i
3i
(1
18.
1
24. -
i +2
17
3.
6.
9.
12.
- 2i)2
- v3 i)3
- i)4
v3
- +-1
2
J
5.
13.
16. (1
19. 1
1
22.
(1
(1
2.
+ yi.
Re z
42.
I=
44.
lz
11 = I.
Im
II > I.
= I.
0.
484
45.
\z - i\ = t.
49.
\z\2 = 4.
47.
10.12
Infinite Series
z+i = 0.
50. \i\2 = 4.
51. z2 = 4.
52. z4 = 4.
55. \z - i\ = 1.
56. \i - z\ + \z+i\ 2.
53. z-i=l.
57
< l_
\z+i\ =
2
54.
Ii - zl + \z+ii = 2.
*58. Find a polynomial q(x) such that (x5 + x4+ x3+ x2+ x + l)q(x)
1 mod 1+ x2
such that
p(x)q(x)
Similarly, if
1 mod 1 + x2
then we write
p(x)
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 ?
=I ()an-ibi
i=O
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 +
u and v,
(u +vr =
1)zn-1 +z n,
i (n)un-ivi.
i=O
n_.co
2
x
of real numbers,
lim lxn - xi
n-+oo
0.
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
Definition.
and
485
z be complex numbers. If
lim lzn
n-+oo
- zl = 0,
then
limzn =z.
We can test for limits by examining the real and imaginary parts of the sequence
separately.
Theorem 1.
For each
let
n,
Zn= Xn + Yni.
If the sequencesx1, x2,
and
limZn =lim Xn
Proof
+ i lim Yn
Let
x
= limxm
n-+oo
so that
y =limym
n-+ oo
lim(xn - x) = lim(yn - y) = 0.
We need to show that
lim(xn + Yni)
n-+
= X
+ yi,
oo
This is trivial:
lim lxn + Yni
n-+
- X
- yil
oo
lim
n-+ oo
+ yi)I = 0.
- (x
2
[(xn - x)
2]112 = 02
(
(Yn - y)
2 1'2
+ 0 ) = 0.
convergence of sequences.
Definition.
Given a series
n,
If
limSn= S,
n-+ oo
let
486
10.12
Infinite Series
Lo z1 converges
S),
(to
and we write
00
:L z1
i=O
! lxil
Proof
If
Lo lz11
converges, then
S.
converges, then
!o Z;
! xi also converges.
The
converges.
Then
and similarly
for eachj. Therefore
and
Therefore
Jim
n-+ oo
:L z1
i=O
i=O
lim
n-+ co
Therefore
lim
=A +Bi.
co
The simplicity of this theorem, and of its proof, are misleading: the theorem is
powerful.
It gives immediately:
Theorem 4.
Lo (zifj!) is
This is so because
for every
1
i=O
I j! I
z.
j
111
j=O j !
< 00
exp
ei6
2k
= :L ( i8)j = :L (i8)
00
i=O j! .
-
Now
00
k=O (2k)!
oo
e'6=:L
c 0ke2k
_
k=O (2k)!
we can express
( i8)2k+l
k=O (2k + 1)!
00
:L
and
Therefore
i8,
. oo c 0ke2k+l
+ i:Lk=O (2k + 1)!
_
in
10.12
487
This gives:
Theorem 5.
= Tr,
\
\
'\
'-
,_
r =
izl 0.
z
w=- '
izl
w= ei8
= izl
cose + i sin e
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.
For example, consider
z
Here
= lzl
Is,
1 +
2i.
and so
z
w=-=
izl
2i
-Js
---
=-+-i
-J5
-Js .
488
10.12
Infinite Series
.J5'
sm
()
Then
w =
and
+ i sin ()
cos()
r(cos
z = rw =
.J
i8
() + i sin())
in polar form.
Theorem 6.
IX,
w, z.
That is, e
w+z
.
But we
Proof
e
i8
i
ea
(cos()
COS() COS IX
i(e+ l
a.
e
i8
In the polar form re , r is called the
=
mined. In fact, when we apply the exponent i() to e , we get a periodic function/(8).
Theorem
+ 2mr)
+ 2mr)
si n e.
1 + i
4.
v3
2.
i
s.
3i
v3
+ i
3.
-7
6.
-4 - 4i
7. In the complex domain, the sine and cosine are defined by the series
co
sinz =
J=O
co
cos z =
z2Hl
.
l)''
(2} +
.
I (-1);
z2;
I c-1); . .
,
j=O
(2j)
De Moivre's Theorem
10.13
ei8
489
ei
9.
z,
sin ( -z)
cos ( -z)
e-i
10. Express
-sin z,
cos z.
z.
11.
10.13
DE MOIVRE'S THEOREM
In Section
10.12
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.
(r2 ;of 0)
z1/z2 = (r1/r2)ei<&1-02l
(This is easy to check by multiplication.)
rule, for
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,
and so
izl
1.
Therefore
38
for some
n.
0 + 2n7T
n =
= 1;
0,
e = o,
1,
e = f 7T,
e = f7T,
z = z2 =
cos
n = 2,
Z1
f7T +
i sin
f7T;
1,
then
izl3
Infinite Series
490
10.13
Using other values of n, we would get repetitions of the same cube roots.
Thus the
roots are
z1 =
1,
Z2 =
1
J3
-2 + li,
Z3
--
J3.
- -z.
These cube roots could have been found by elementary methods, because
z3
(z -
l)(z2 +
+ 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
ei"f2
methods break down, and our new method still works. For example, i
Therefore the fourth roots of are the numbers ei0 for which
=
4() = :!!.
2n1T'
or
() =
4n
1T.
1T
51T
()2 -
()1=-,
8
()o--,
8
=
91T
- 137T
()3-
reie - 0
where
e;
.nr o
v r e';
(1/n)(()
2j7T)
(j = 0, 1,
(j = 0,
. . . , n
1,
. . .
1),
, n
1).
10.13
1)
De Moivre's Theorem
wn
= z , then
lwln= lzl = r,
2)
and
Any
lwl= V";.
and
=
V"; eia.,
+ 2j7T
ex
491
and
= z, then
for some j,
()
= -+
27T.
successive values of j give us n different values of e ia., but thereafter the values of
n= 5.
z= 1 + 2i,
Then
where
r =
J5,
Sin-1
(1/J5).
r ei81
= M eie1= 5111oei8;
Z;=
where
and so on.
1, 2, 3, 4),
e; =tee +
2j7T)
7T
= !!. + 2j
Thus
(j= 0,
(j
= 0, 1, 2, 3, 4).
efs
Zo= 5111oei ,
rr) s
Zi = 51/1oe i(8+2 / ,
4rrl/s
z2 = 5111oei(9+
,
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)
c
b
2
z +-z=-a
a
2
2
c
b
b
b
2
z + -z + - = - -+2
2
a
a
4a
4a
2
2
b
b - 4ac
z +- =
.
2
2a
4a
492
Infinite Series
If b2
10.13
(z -)2 = 0
2a
b
<=>
z +
-2a = 0,
b
and
z = -b/2a
b
- - + Z1 '
2a
b
- - + Z 2.
2a
z4 + 1
3. z3 + 8
5. z3 +
z2
0.
2. z6 + i
0.
+ z + 1
4. z2 + 2z
=
8. z5 +
9.
0.
i + 1
6. z2 + z + i +
0.
7. z7 + z6 + z5 + z4 + z3 +
4
+ 2z3 + 2z2 + z + 1
+ 1
0.
0.
0.
0.
We know that for each n,-the number 1 has exactly n nth roots.
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.
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.
other). Explain how this result is related to one of the preceding problems.
10.14
493
1)
! a;xi
perhaps for the endpoints of an interval. That is, the domain of convergence always
turned out to be (a) 0 alone, (b) ( - oo,
oo),
The function
of all orders.
( - 1 1) .
f(x)
1/(1 +
!o ( - l )ix2i
Here the series goes bad for reasons which seem unrelated to the properties
Proof
limn"' Zn
n,
let Sn
Li'=o z;,
0.
so that limnoo Sn
S - S = 0.
!:o Z;.
Then
old proof.)
Proof
x1, x2,
. also converge.
closed disk D,
Theorem 3. If a series
{z
Jzl
= z0, with z0
Infinite Series
494
10.14
We have
L Ja;J si is convergent.
I I
i
;
!ail ;0 Zo = !ail ;0 lzoli
i
laiz&I ;
o
0, and so the numbers aizi
But we know that L aizi is convergent. Therefore a;zi
!ail s;
I I
-'>-
laizgJ
And
for every j.
I :J
i
i
I
I.!._
=
I
I la;Z6J I.!._
I
z0 i=O z0
i=O
<
1,
lzol Therefore
<
b
_
b
I Ja;I si
1
ls/z01 <
i=O
Now for each point z of Ds, we have Jzl s. Therefore
L la z il zla;I s1 <
(z in Ds)
i=O ;
i=O
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
because
"'
oo.
"'
00
s.
That is,
S
= s J _I a1z1
J=O
D8
converges on
radius of convergence is
S. (See page 243.) Let
oo.
Theorem 4.
z.
=sup S.
radius of convergence
is
r.
If a series
converges for
I>s}
r
In this case, we say that the
;if 0.
<
oo,
Proof
z0, with Jz0J < r. Since r is the least upper bound of the set S,
Jz01 is not an upper bound of S. Therefore there is an s > lz0J such that
z0
the series converges on D8 Therefore the series converges at z
2) Given z0, with lz0J > r. Suppose that the series converges at z = z0, and let s be
1)
Given a number
the number
such that
Then the series converges on
for such numbers
s.
Ds;
< s
lz0J.
<
and this is impossible, because
is an upper bound
10.14
Note that while this theorem tells us what happens inside the circle
lzl
= r,
495
and
what happens outside the circle, it tells us nothing about what happens on the circle.
The circular domain of convergence for complex power series also accounts for
the behavior of the series
I< -1)ix2i
i=O
If this series converged for some
1
___2 .
+ X
""'(-l)'z'
i=O
.
2'
1 +
1/(1 + z ),
2,
---
z?]
f'(zo)
lim
z->zo
f(z) - f(zo)
Z
z0
may approach
G
8
Infinite Series
496
10.14
=>
>
0 there is a o
f(z)
f(z o)
- f'(zo) <
- Zo
>
0 such that
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)
If f is a polynomial, with
f(z)
n
= x .
Similarly:
f'(z)
nzn-1
2;a1z1,
1=0
n
2;ja1z1-1.
j=l
Theorem 6.
D.,
00
f'(z)
2;ja1z1-1
j=l
lzol <
< r
Then
00
L la11 s1 < oo
1=0
(See the proof of Theorem
4,
(0 <
< r).
10.14
g(s),
it follows that
g'(s)
497
00
Lj la 1 sH,
1
j=l
Lj la 1 s1 < oo,
1
j=l
and so
00
L j Ja 1 s1
n-+oo :i=n+l 1
lim
. [--- ( La z
0.
;)]
Now consider what we are trying to prove. The theorem says that
hm
z--+zo
for every
z0 in
00
z - z0
J=O
00
- La z0
J=O 1
00
LJa;z0J-1,
J=l
lim La;(z1-1 +
z-+zo i=l
z1-2z0 + z1-3z +
0.
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
our conclusion would follow: if
Sn(z)
n
Lak
i=O
zb-1,
as
--+
z0,
and so the
then
lim Sn(z)
z-+zo
for each n.
Rn(z)
We know that
lzl < s.
00
+ zg-1 - jzg-1).
L alzi-l + z1-2z0 +
i=n+l
lzol < s; and since we are discussing limho' we may
=
assume that
lzl < s
we have
00
IRn(z)J L Ja J 2js1-1.
i=n+l 1
n --+ oo.
498
10.14
Infinite Series
Now let
E > 0 be
given. There is an
There is a c5
> 0 such
that
Zo l
00
L ak
1=1
we have
and so limZ-+
o
L ak - )
<0
)
Sn(z) + Rn(z),
0,
=>
I Iai( )I
=>
such that
0 < lz -
Since
IRn(z)I
enough to make
ISn(z)I
small.
lz - z01
large
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.
differentiation. Thus
Df(z)
and
00
f'(z),
00
L a1z1 = L ja1z1-1,
1=1
1=1
2.
Find De'.
5. Find De2
4. Find D cos z.
7.
Find De-.
=Jn,
n
Dt
3.
Find D sin
6.
z.
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)
9. Givenf(z)
10.
11.
12.
f' (a + z).
L:o a1z1
=
0 for every z in Dr
0 for every j.
L: o a1z1 in Dr.
Obviously a0
0, because f(O)
a0.
Given f(z)
Show that if f'(z)
function and is equal to a0 for each z.
=
derivativesf<2>
Given f(z) =
Show that a1
Given thatf(z)
constant.
L a1zi, g(z)
L b1zi in
D r.
0 for every
z,
then f is a constant
g', then/ - :; is a
10.15
13. Let
cos2
+ sin2
z.
+ sin2
cos
15.
(z)
z
and sin
Show that
ea
Show that
=
(z)
1 for every
1 for every
499
z.
z,
ea+z,
for every
and
z.
16. Does the addition formula for the cosine hold in the complex domain? That is, is it
true that
cos
for every
(a
z)
cos
cos
sin
a sin z,
+ cos
a sin z.
and z?
2z
2z
*10.15
(a
z)
sin
a cos z
2 sin z cos z.
cos2
sin2
z.
In Section
10.5
) on which
-r, r
the
do this may be easier to understand if we first show how they apply to the geometric
series
f(x)
00
1
=
--
- X
rk
Let
00
Lx
i=O '
1 <
< 1).
rk
1).
(1)
(2)
00
so that
and
Since
and
oo
"
500
10.15
Infinite Series
(3)
ex:>
(3) holds, then we can take the limit in formula (2), getting
= kf(x) dx
(\i dx
J
o
lkxi dx = 0.
00
i=O
i
i=o
00
--
1 - x 1 - k,
and
Therefore
Therefore
Therefore
r n+l
kRn(x) dx k k
dx
o
J ol-k
--
kn+2
1-k
= -- .
lim (k R (x) dx = 0,
n-+oo Jo n
Rv R2,
were
squeezed
kn+l
Mn =-k'
1
-
0 Rn(x) Mn
and
(0 x k),
IimMn=O.
It followed ihat
j.kRn(x) dx
0
n-->OO
k dx
rM
Jo n
= kMn--+
0.
of Section
10.5.
To do this, we
10.15
Theorem 1.
L a;ki
Proof
If
L a,xi
(-r, r),
is absolutely convergent.
Let
x1
and
r.
k = lx11
s,
iO
and so
sequence
1
L:s' = -00
But
501
<co.
Therefore
.
00
L la;k'I =
Therefore
L aiki
00
00
b
1-s
<
co.
If
remainders
L aixi is
R n(x) =
M1, M2,
(-r, r),
00
in+l
[-k, k],
a;xi
limMn = 0,
n-> oo
and
Proof
For each
n,
(-k x k).
theorem that L aiki is
let
L laikil.
00
Mn=
i=n+l
Then
limM n=O,
because
n-> oo
oo
i
lai kil - L la;k j.
i=O
i=O
Mn= L
absolutely convergent.
502
Infinite Series
10.15
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.
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)
Definition.
on [a, b].
UJim Sn(x)
n-+co
on [a, b],
S(x)
on [a, b ].
[a, b]
Rn(x)
Sn(x)
S(x)
. Rn(x)
[-k, k],
co
L aixi,
i=n+l
L a;x\
i=O
co
L a;x;,
i=O
S(x) - Sn(x).
If
L a;x;
L a;xi
i=O
co
L a;xi
i=O
on [
k, k].
r,
then
10.15
503
n,
and U limn-oofn(x)
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 < b
we have
Ix - Xol < O
=>
=>
Since
for every x,
/a + b + cl l / + /bl + /c/.
Therefore
/x - x01 < O
:::::>
E,
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 zo a;xi converges on (-r, r), and lxl < r, then
{"'
Jo
[f al]
i=O
dt
=I l"'0 ai dt.
i=O
Infinite Series
504
Since
10.15
f ti dt = xi+I/(i + I),
J0
00
U lim
n--+oo
(For
-r
< x <
Sn(t) =_Lai
i=O
on
x < r
we have
[O, x].
[x, O],
00
IRn(t)I =
Iia;ti - Sn(t)I
Mn
for every
n.
Therefore
"Mn dt
=Mn Ix!.
Since limnoo Mn
lxl = 0,
it follows that
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 =
and
oo
lim
noo
l"'
n
oo
_L a/dt = _L
i=O
i=O 0
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
Then
_L a;xi
is convergent; limnoo
x1 be
aixi = O;
Let
any
and
10.15
Jxl
i
liaix -ll
X1S,
liaixi-1si-ll
Therefore
laixi-11
C()
C()
il
i=l
505
lxl/x1, so that
isi-l bii-1
i
i
2 I iaix -ll b 2 is -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),
f'(x)
Proof
and
C()
2 iaix -1
i=l
Let
g(x)
Then
Therefore
C()
2 ia;xi-l
i=l
ia/-1 dt
i=l
00 1"'
i=l 0
i"'g(t) dt
ia/-1 dt
00
2 aix;
i=l
f(x) - a0
f(x) - a0
Di"'g(t) dt
g(x).
f"(x)
/<3\x)
00
2
00
2 i(i - l)xi-2,
i=2
i =3
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.
many derivatives, then f cannot be represented by a power series. Later you will see
506
Infinite Series
10.15
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
i=;ol
+ bi sin
ix].
1. Let
J<x)
oo
.I
=0
<-1)i
x 2i
' .
(2l )'
f(x)
g(x)
x2H1
(-l)i
(2i + 1)!
xi
(-l)i-:-.
z!
i=O
oo
h(x)
==
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.)
=
xeX + e"'.
x cos x + sin x.
6.
7.
8.
x sin x - cos x.
=
1.
2x.
In each of the following cases, discuss limn__.oof n(x) and U limn__.00/ n(x).
9.
fn(x)
10. fn(x)
11. fn(x)
12. fn(X)
13. fn(X)
14. fn(x)
nxn,
t.
-1 x 1.
n sin (x/n)2,
0 x 1.
0 x 1.
L aizi converges on Dr, then L I aizil converges on Dr; in fact, every time we have proved
10.15
507
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,
lim/n(z)
n-co
f(z)
on D8
limMn
n-co
and
0,
1/n(z) - /(z)I
Mn,
for every z in D5
If
.2 a1zi
has
> 0
U
n-oo
1=0
i=O
O <
<
r,
then
Jim
11
11.1
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.
z
/
/
----/
//
f--I
- -P1 /
I
--- /
-
-:-'II
I
I
I
I
I
I
: ------ !
/7---- y
I
--
- --
//
I
/
---....v/
I
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
Yi
Zi,
and we write
Pi +-+
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'
508
z1) .
11.1
509
octant,
octants.
The figure
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
have
0 P
x + y + z.
P1, P2,
following theorem.
b
a
i'\
I \
I
''
'
'e
'
J.-----'\--- ---
/ .............
\
-.......
/
\
d ',
'
/a
......
/
...... _,
Theorem
P1P2=
Proof
J (x2
then
2
2
2
- X1) + (Y2 - Y1) + (z2 - z1) .
x2, Yi -:;!: y2, and z1 -:;!: z2.
a= lx2 - Xii,
Therefore
b= IY2 - Y1I,
c = lz2 - z1I
11.1
510
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.
right below.) Given a plane E, suppose first that E does not pass through the origin,
al
I
P2
I
I
- 1 ----
!
/
__
--- Y
---
/
/
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
0.
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.
point of the line perpendicular to Ethrough 0, other than 0 itself. Let P1 be the point
(-a, -b, -c).
11.1
511
Po(a,b,c), P1 (-a,-b,-c)
+++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 =
z.
This theorem can be applied in a variety of ways. For example, let Ebe the plane
Ax+By+Cz+ D =
0,
and the equation must be satisfied by the coordinates of the three given points.
Therefore the coefficients
Subtraoting
(2)
from
(1)
A, B, C,
3A+ 3B+4C+ D =
0,
(1)
2A+4B+4C+ D =
0,
(2)
2A+3B+ 6C+ D =
0.
(3)
we get
A -B =
Setting
B =A
Subtracting
in
(3')
(2)
from
and
(2')
(3),
(1)- (2)
0.
we get
6A+4C+ D = 0,
(2')
5A+ 6C+ D
(3')
0.
we get
A - 2C
0,
(2')
(3')
512
11.2
and so C = A/2.
Therefore D =
-8A.
A= 2.
This gives
A= B = 2,
D = - 16,
1,
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
(k -:;t: O);
but x does not appear (with nonzero coefficient) in any of these equations.
PROBLEM SET
1.
point P0
2.
11.1
Find the equation of the plane containing all points equidistant from the origin and the
=
(2,6, 4).
Find the equation of the plane containing all points equidistant from P0
Pi
(0,2,3).
3.
Find the equation of the plane containing all points equidistant from the planes
4.
P2
5.
z =
and
+y +
( -1,2, 0).
P2
6.
+y +
P0
=
(0,0,3).
z =
=
-1.
(1,0, 1), Pi
(2,1,1),Pi
P0
(1,3,0)
P1
and
(2, 2, 3),
7.
The point
(1, -1, 2) is the foot of the perpendicular from the origin to a plane E.
8.
(1,0,2),Pi
(2,2,1),andP2
9.
Let
(1,0,0), let B
{PI 2AP
11.2
DIRECTION COSINES.
Linear equations for planes are more useful if we know the geometric significance of
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
the coefficients in the equations.
PQ = X2
- X1.
11.2
Direction Cosines.
Since x1 - x2
513
QP= -PQ
for every
PQ+ QR= PR
And since the distance
PQ is
and
Q.
P, Q, R.
PQ=
if
PQ
Q is
in the positive
direction from
if
-PQ
Q is
P,
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.
z
<-->
(a,b,c),.o(0,0,0).
OP= 1.
P= (a,
b,
c)
be a point
of the x-axis and the positive end of L. In its own plane, this angle looks like this:
L
Note that the foot of the perpendicular really does have coordinate
in fact, this is the definition of the x-coordinate of
cos
CJ..
P.
on the x-axis;
direction angles
then
=a.
e<,
They are
514
11.2
measures
cos ex,
are called the
direction cosines
cosy
of L.
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,
p > 0,
If
as above, then
cos ex,
XifP =
= Xif-p,
and
xifP =
= - Xifp,
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
graph of the equation
Of' = p is 0.
'
.Ja2
+ b2 + c2.
.
The equation therefore has the form
b
a
-x +
p
p
-
or
c
-
x cos ex + y cos {J +
- p = 0,
cosy
- p = 0.
11.2
Direction Cosines.
515
oc
cos
and p
___..
oc
1T
- -cos
oc,
fJ
cos
oc,
fJ
- {J,
1T
y -
-cos {J,
1T
- y,
cosy - -cosy,
x cos
+ y cos fJ +
oc
cosy - p
0.
Theorem 3. Let Ebe a plane, and letN be a directed normal to Ethrough the origin,
with direction angles
oc,
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.
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,
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.
11.2
516
Proof. LetP be the point of L for which the directed distance OP is 1. IfP= (a, b, c)
then
b=cos (3,
a=COS IX,
c =cosy.
Since
OP=OP=1 =.Ja2 + b2 + c2,
we have a2 + b2 +
2
=
Theorem 5.
If a2 + b2 +
are a, b, and c.
c2
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
has the form
for some
IX,
x + 2y + 3z + 4=0
k COS
k2(cos2
and
(1)
IX
IX
IX
= 1,
k cos {J=2,
k cosy=3,
Here
.)14
x +
2
3
4
+
+
=
.}14 y
.}14
o.
.)14 z
cos f3 = 2/./14,
cosy=3/./14,
p= -4/./14.
(2)
11.2
Direction Cosines.
517
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
- .J14
- .J14 y
3
.J14
.J14
(3)
O,
cosy'= -
1
==-cos a,
.J14
3
.J14
cos /3'=
'
p =
= -cosy,
4
-
- -p.
.J14 -
x, y, z.
of Theorem 2:
Theorem 6. The graph of every linear equation in
Proof
Given
Ax + By + Cz +
with
A, B,
and
C not
all= 0. Then
x,
is a plane.
D = 0,
A2 + B2 + C2
k = .J A2
y, and
(1)
> 0. Let
+ B2 + C2,
and
B
b= -,
A
a= '
k
c
c=
'
k
p=
D
k
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,
PROBLEM SET
l. Given
11 2
.
+ y + z -1
and sketch.
0. Write the two directed normal forms for the same plane,
+ 2z - 3
+ 2y + 2z
0.
+ 2y + 4z - 4
+ 2y + 4z + 4
0.
0.
518
11.3
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
and
K (more
direction on K).
9.
Let
K be
as in Problem
{ }.
6.
Give
10. A plane E contains the points (a, 0, 0), (0, b, 0), (0, 0, c),
b,
a,
directed normal forms of the equation, and sketch, showing the case
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
P = (x, y, z)
with the
and
and
()(Z1 ,
for
P.
Thus we
11.3
519
with addition, scalar multiplication, and inner product defined by the formulas
A.2.
O+P=P+O=P
for every
A.3.
P.
For every
M.1.
(rx{J)P= rx({JP).
M.2.
M.3.
M.4.
M.5.
M.6.
P= 0, for every P.
P=P, for every P.
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 scalar multiplication). For every Vin "f/' and every real number
rx, rxV belongs to "f/'.
CSM
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}.
520
11.3
The inner product that we have defined in R3 has the following properties:
S.1.
S.2.
PiP2
P2Pi.
(Cl.Pi) P2
C1.(Pi PJ.
Pi P2 + Pi Pa.
S.3.
Pi (P2 + P3)
S.4.
P PE;;; 0,
for every
P.
S.5.
If
PP
0, then P
0.
(0, 0, 0)
to
.jx2 + y2 + z2
(x, y, z)
is
.jp . P.
P,
[[ P\[.
(The
double bars are a reminder that we are performing an operation on a vector rather
than a number.)
../x2
+ y2 + z2
and so we use it as
[[P [[
\IP [[
.jp P.
P.
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
OPi + OP - 2 OPi OP2
(Pi P2)
2
2
2
(X1 - X2) + (Yi - Y2) + (zi - Zz)
=
cos 0;
x + Yi + zi + x + Y + z - 2[\P1\\ \\P2\I
-2[IP1\I
l\P2ll
cos();
cos O;
521
11.3
j = (0, 1, 0),
i = (1,0,0),
k = (0, 0, 1),
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,
not all equal to zero, such that
ix1V1 + ix2V2 + ... + anVn = 0.
, an,
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
Therefore
=>
IX
IXz
0 = (0,0,0).
= IX3 =0.
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},
522
11.3
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
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
2)
[1+(-l)]P=1 P +(-l)P.
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
:::::
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,
11.3
rxP1
and
523
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.
through 0, let
P0
(A , B, C)
P-:;!:.
0 in E,
For
P0
is perpendicular to E
. Then for
and so
Conversely, if
P0
0.
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 +
+ Cz
By
0,
which we already knew. But when we describe Eby a vector equation, using the
inner product, this suggests the following theorem:
Theorem 6.
V0 be
{V I V0 V
O}.
Proof
V0 Vi
V0 V2
Therefore
Vo (Vi + V2)
by S.
3
. Therefore
V0 (Vi + V2)
V0 a.Vi
a.Vi V0
Vo Vi + Vo V2,
0, and
Vi + V2 is
rx(Vi V0)
in "I""'. Similarly,
rx(V0 Vi)
rx
0.
524
11.3
If you rewrite these formulas, in the forms that they take when
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
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 +
where
P0
Theorem 7.
(A, B, C).
= 0 <=>
11.3
giving the value of a that you are using and the coordinates A, B, C of P0
2.
z=x+y
3. z= -x - y
5. y = 4z - 3x
7. z= 1
x y z
+ + =4
9 1
2
3
11. Let V1 = i +j, V2 =j + k, Va
k.
linear combination of V1, V2, V3
12. Now show that {V1, V2, Va} spans R3.
=
13.
-D,
=a.
1.
P0 P = a,
Thus we have:
Every plane in
PROBLEM SET
P0 P
z=x - y
4. x= 3y - 4z
6. z= 4x+ 3y
8. x= 4
z
x y
lO. 2 - 4 + 3 = 2
-
j,
k is a
{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
15. Let
V1
i +j,
V2
V1= i - j +k, V2
through 13.
+ k,
=
V3 = i + k.
+j
k,
V3 = -i +j
13.
+ k. Proceed as in Problems 11
11.3
525
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)
1X{J(V1 V2)
17. (1XV1) ({JV2)
18. 1X(V1 + V2 + V3)
1XV1 + 1XV2 + cxV3
19. l\1XVll
10(1 llVll
20. l\-1XVll
llXI llVll
21. (P+ Q) (R + S)
of course .)
22 . (P + Q)(P - R)
P R + Q R + P S + Q S (Here P, Q, R,
are vectors,
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
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.
=
526
11.4
*39. Let P be the set of all polynomials with real coefficients. For
n
i
V =a.x
., '
i=O
we define
V+ W
Show that
n
b.xi
., '
i=O
n
i
L (ai + bi) x ,
i=O
n
i
L aix ,
i =O
n
Laibi.
i=O
V W
2
2
2
(V W) llVll llW ll
a finite basis?
11.4
For each positive integer n, let Rn be the set of all n-tuples of real numbers. Thus
Rn
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
where
E1
E n-i
E2
En
(In general, Ei has 1 in the ith position, and O's everywhere else.) The vectors Ei span
the space Rn, with
(xi. x2, , x n)
n
L xiEi.
i=l
=>
(x 1, x2, , x n)
(0, 0, ..., 0)
{ixE1
ix ER}
{ixE1 + fJE2
ix, (J ER}
11.4
is a subspace, and forms a plane, and so on, for any subset of Bn.
But
527
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
Theorem 1. Let "f/ be a vector space. Let A be a set of m vectors, and let B be a set
of
(1) A
B.
"f/ =R ,
A and 3
= 1 element of B. To
spans R3
We proceed to the general proof.
that the vectors which also belong to
where
A1, A2,
A in such a way
possibility is that i
B spans
V, it follows that
528
{31 :;rt: 0. We can therefore solve for B1, in the above equation, getting
suppose that
B1
11.4
Now let
B'
{A1, A2,
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
Here
m =
2, n
elements of
{A1, A2}
B, with
A1
B'
E1 +E2
A2
A2 as a linear combination
E2 +Ea.
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
and now we are done.
B"
Theorem 2.
set of
m
Let
be a set of
n.
(2) B
spans "f/.
B be
Then
A.
Theorem 3.
has exactly
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,
11.4
529
In
That is,
0.
dim
{O}
0.
{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.
vectors is
linearly dependent.
Proof
n.
elements is a basis.
Proof
{V1, V2,
, Vn}. If B spans Y, then
Vn+i which is not a linear combination of elements
=
{V1, V2,
Vn, V n+i}
{V1, V2,
be a set which
Vm}
Proof
V1
If not, some V; is a
Then
{V2,
Vm}
also
spans Y. Repeating this process, removing superfluous vectors one at a time, we get
a
basis.
Theorem 7.
Proof
If dim Y
n,
Let
elements.
n.
be the largest number for which it is true that "f/"' contains a linearly
independent set of
m
elements spans Y.
Any set which spans Y contains a basis, and every basis has
Theorem 8.
P;oof
m,
and
n.) Let
B
{V1, V2,
Vm}
(Proof
If not, there
530
11.5
={Vi, V2,
Vm, Vm+i}
is linearly independent.)
=m
n.
11.4
1. Given Vi = Ei
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
7.
8.
R4, let'"/!'" ={VI V (Ei +E2 +Ea)= O}. Find a basis for '"fl'".
4
Find a basis for R , using Ei +E2 and Ea +E4 as basis elements.
a
Find a basis for R from among the vectors of
B ={Ei
Ra, let'"/!'" ={VI V (E1 +E2)=0, V (1 +2Ea)= 0}. Find a basis for '"fl'".
4
In R , let '"fl'"={Vj V (E1 +E2 +Ea)= 0, V (E2 +2Ea +E4)= O}.
Find a
9. In
10.
basis for'"/!'".
11.5
ORTHONORMAL BASES
Rn
a basis
Ei
V = (x i x2,
.
then
v
are 0. Thus
i=l
If
W
Ei
n
, xn) = ,L xiEi.
n
L Y;E;,
i=l
n
.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
531
Orthonormal Bases
11.5
This formula does not hold for all bases. For example, the set
( ?)
(Try taking
R2,
but it is
(cx1Vi + IX2V2)
1.) But the above formula for the inner product does
{Vi, V2,
if
Vi V2 = 0.
(Thus
Ei
and
2-
V n}
orthogonal if
Thus
Bn = {Ei, 2,
for i j.
En}
Ei
(Ei + 2) =
If
ll V;ll =
then B=
is
( ?).
{Ei, Ei + 2} is
not, because
1 + 0 = 1 0.
for each i,
{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
for i j
V;V.=
o
'
Theorem
1.
for i
j.
B=
be a basis for
dim"/"' =
W = {Wi, W2,
{Wi, W2,
W11}.
V11}.
Then
a i = Wi V n+i Let
n
v+l = vn+l - .L ai JV;.
ii
532
11.5
from 1 to
we have
wk v+i
wk Vn+1 - 2 aiwk
i=l
a:.
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
Therefore
Wk Vn+i
ak = ak - ak
0.
and let
llV+1ll,
Then
and
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}
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.
Let
v;
V2 - a1W1
( 2 + E3)
-
t(E1 + E2)
tE1 + !E2 + 3.
Orthonormal Bases
11.5
:2.
W 1 V =
./2
3)
Now we normalize
V ,
(-t +t +0) = 0.
by letting
W2 = V/11Vll-
Since
llVll
= V2
we have
V = ! +t + 1 = -f,
1
11v;11
and
W2 = so that
and
533
._j(, E1
J-3
2
i 3,
+.Jf, E2 +yr
II W2ll2 = i +i +i =
II W2II
W2}
1
,
is an orthonormal basis.
Orthonormal bases are what we need to get a simple formula for the inner
product:
Theorem 2. If
{V1, V2,
Proof
2 ;{J
i=l
ct.
;.
We know that
(i = j),
and
Therefore
. L f3;V; =
n
ct.iV;
and so
for each
J=l
{3;
ct.;
i=l
- Q II.
i=l
{3;.
ct.;
V V = II V l!2,
i,
we have
534
11.5
spaces geometrically, and so we may refer to their elements as points rather than
vectors, as in the following definition.
distance
and
Q is
PQ.
An
llP- Qll.
and
Q may
be denoted by
d( P, Q),
or simply
Theorem 3. Let B
space
V,
{V1, V 2,
and let
, Vn}
.L aiV;,
i=l
Then
Proof
.L Pit:
i=l
We have
d(P, Q)
P- Q
llP - Qii ,
and
llP - Qll2
(P- Q)
.L (a; - p;)Vi,
i=l
(P - Q).
2
llP - Qll
L (a;- P;)2,
i=l
11.5
3
V be the subspace of R spanned by the
E1 +E2}. Find an orthonormal basis for V.
1. Let
2.
basis B
B
4. Same question for
6. In
3
R , find an
an
9. Given
10 .
In
8. Let
4
R,
Find vectors
V3, V4
so that {Vv
proceed as in Problem 8.
is
11.6
={VI V (1
2) =
0,
E3
0,
535
V E4 = O}.
then
"//'
13.
Find
a formula for
V2,
Let
V = L ixi V; be any
V V; = 0 for every i
=<>-
V = 0.
can have more than n elements. Thus, for example, in R3 there is no set of four con
;;:;; dim r.
Theorem (?) Let B ={Vi. Vi, . .. , Vn} be a basis for i"". If for all vectors V =
W
L ix; V;,
= ! /3;V;, we have
then B is orthonormal.
Is this true? Why or why not?
18. Show that if {V1,
V2,
Vn}
n
= L (V V;)V;.
iI
11.6
V=L
IX;
IX;
= V V;.
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
P, Q,
and
R.
d(P, Q)
llP - Q ll.
(1)
536
11.6
where
P - Q and B
(2)
ll A + Bii ll A ll + llBI\,
=
Q-
\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.
(3)
Therefore the natural first step, in proving (2) , is to restate it in terms of the definition
given in
In these terms,
(3).
(2)
ll A + Bii ll A ll + llBll
<::;, 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\
Here
l4)
llBll .
(4)
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
A
- , B.) And Eq. (5) is obviously equivalent to
(If
(6)
Theorem 1
for every
is called the
Schwarz inequality.
(6)
A and B.
(6)
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
A B -:;if 0.
A -:;if
0 and
11.6
The inequality
the form
(7)
(6)
Since (7)
We take
<=>
(6),
for every
l't. , (J =;I= 0,
1
We then choose
537
l't.=llAll'
(J=
l't.A B
-
l't.
so that
lll't.A ll = 1.
so that
l't.A (JB = I.
and
(J
as we please.
We let P = l't.A,
Q = (JB.
llPll =
This is easy to prove: for
1 and
P Q=
llPll = 1, P Q =
=>
llQll2
1, we have
(P - Q) (P - Q) 0
=>
P P - 2P Q + Q Q 0
=>
llPll2 - 2P Q + llQll2 0
=>
- 2 + 11Q11 2 o
=>
llQll2
The theorem follows. In the light of the discussion which led to the Schwarz inequality,
we also have the following:
(2')
N.1.
N.2.
N.3.
N.4.
538
11.6
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.
where
Example
ll P ll satisfies
1 1 ].
,
Addition and scalar multiplication are defined in the obvious way. We define
where max
that
Therefore max
II II.,.)
Example
II!+ gll 1
=f
(x) I dx + f 1lg(x) I dx
= 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
11.6
539
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 Iii].
[21, +, sm, 11 II,,],
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
1.
2.
3.
11.6
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
Show, by any method, that for every pair of finite sequences xl> x2, , Xn, Yi. y2, ,
Yn of real numbers,
(x - a)2
(x, y) and Q
+
(y - b)2.
Thus d is the square of the usual distance. Does [E,d] form a metric space?
4.
Ix - al + ly - bl.
maxmium of Ix - al and ly - bl.
If [S,d] is a metric space and k > 0, does it follow that [S, kd] is a metric space?
5. Same
6.
question,for d(P, Q)
Vlxl.
7.
The real number systemR clearlyforms a vector space. For eachxinR,let llxll
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?
540
11.6
9. Let .!l'1 be the set of all continuous functions on the interval [ -1,
d (/,g )
1], as in Example 1
l/(x) - g(x)I
dx.
l/(x) + g(x)I
-1
1 +
the "uniform norm," defined in Example 1 of the text. Show that for any sequence
<=>
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
l/(x)I dx,
1
n-co
Is it true that
U lim fn = 0
=>
=>
U lim/n = O?
n-oo
Jim llfnll
O?
12. Let C0[-7T, 7T] be the set of all continuous functions/ on the interval [-7T, 7T], with +
h(x)
f(x)
a + a1x + x2 which is orthogonal to g(x)
0
x, in the inner-product space defined in Problem 12 .
=
] with
TT ,
1 and
14. Find an orthonormal basis for the subspace of C0[-7T, 7T] spanned by the elements
{l,x,x2}.
*15. For each
i=l
on the interval [-7T, 7T]. (Such functions are called trigonometric polynomials.)
Evidnt!y Tn forms a subspace of C0[-7T, 7T], and the set
B
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)
cos (A + B) + cos (A
B)
-2 sin
A sin B,
2 cos A cos B
are also useful.] Problem 15 of Problem Set 11.5 is useful at one stage.
12
12.1
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 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.
++
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
4)
Therefore (1)
542
12.1
Fourier Series
p
(P - Q) S
\IP - R\12.
Therefore
4)
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
Proof The easy part is to show that there is only one such Q. If
(P - Q)
and
(P - Q') S
0,
[(P - Q) - (P - Q')]
0,
(Q' - Q) . s
for every S. In particular, for S
so Q'
0,
0, and
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
"fr
R2
P
we ought to have
12.1
543
E1 + 0 E2 + rxaE3,
(P - Q) E1
(P - Q) E2
0,
and so
(P - Q) S
(P - Q)
(x1E1 + x2E2)
rx1E1 . E1 + rx2E2 . E1
rx1E1 E2 + rx2E2 E2
rx1,
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,
and let
Q
Then
Q
W;
and so
(P - Q)
W;
n
! rxiW,:
i=l
to n, let
n
! rxiW i.
i=l
W;
n
! /J3 W;
j=l
we have
2
rx; llW;lt
Wn}
P W;,
P W; - Q
(P - Q)
W;
rx3,
rx; - rx;
0,
E "fl/,
n
! /J;(P
j=l
Q)
W;
0,
Theorem 2.
, Wn} is an
12.1
Fourier Series
544
(In the proof of Theorem 1, we found that this sum satisfied the conditions for
and that there is only one such
Q.)
(3)
Q,
and (4),
stated at the beginning of this section, hold on the basis of our general definition.
Theorem 3.
If
= Pr P, and
Proof Obviously
because Q
R is
-
is in if/, then
Proof
llA
= 0
+ Bll2 =
0,
(Q
R)
the following.
-
llAll2
=>
(A
llBll2
+ B)
(A
llA
+ Bll2
+ B)
=AA+2AB+BB
=
llAIJ2
+ 0 +
llBll2
Theorem 5.
P.
"Y
where
C0[-TT, TT],
C0[-TT, TT] is the set of all continuous functions on the closed interval [-TT, TT];
["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
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].
= f:
Obviously
Tn forms
i=l
x, cos 2x,. .. , cos nx; sin x, sin 2x, ... , sin nx}.
{1; cos
This set spans
Tn.
f:
"
cos
J:
ix dx
cos
"
= f:
"
sin
ix sinjx dx
ix dx
=
for every i,
for every
i, j,
12.1
and
J:
545
J:
sin ix sinjx dx =
cos ix cosjx dx = 0
for i ;;i!: j.
,,
11
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,
llcos ixll2
7T
\11\I
.J21T;
\\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,
, Sn}, where
Cn; S1, S2,
.
.J21T'
(i > 0),
1 .
s
(i > 0) .
.x
m i
.J;
'
COS lX
.J;
1
S.
Co=
Ci
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
=
P r nf
Here
f C0
n
n
I (f Ci)Ci + I (f S,)S;.
i=O
i=l
J f(x)
"
_,,
C;
f S;
Pr nf
a;
-11
dx
.J21T
cos ix dx
(i > 0),
sin ix dx.
n
a0 + I [a; cos ix + b; sin ix],
i=l
27T
1
=
7T
b;
.J21T
1 f" f(x)
7T- J_1T{" f(x)
1-
'\/ 7T
where
a0
--=
'\/
Therefore
_,,
-1T
1 f" f(x)
7T
-1T
cos
sin ix dx.
(i > 0)
546
Fourier Series
12.l
b; sin ix
[Check that (j C0)C0
a0, (j Ci)Ci
ai cos ix, and (f S;)S;
).
> O ]
It now seems reasonable to hope that Prnf is in some sense an approximation of/
when n is large. That is,
(i
( ?) n
co
Prnf
( ?).
(?)
A
lim I
n-+oo I!
Prnfll
(?).
(1)
stronger conjecture is that the approximation is good even in the uniform norm:
0 (?).
lim
(2)
- Prnf llu
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 limnoo II/- Prnfll,.
0.
00
I (ai cos ix
i=l
bi sin ix),
an+l cos (n
leaving the preceding terms unchanged. Second, the error in the approximation
Prnf f, as measured by II/ - Prn/11, is nonincreasing.
Theorem 6.
12.1
547
(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:
=
__,'---"'""'-....._'--""""'---'--'"4""-L----"...-'-_,___. x
- 3,,.
-.,,.
.,,.
3,,.
5,,.
-511"
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)].
=
548
12.1
Fourier Series
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.
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
4. /(x) = x3
7. /(x) = x
on [O, rr],
8. /(x) = -x
2. f(x) = x2
3. /(x) = x + x2
5. f(x) = x - x3/rr2
6. /(x) =!xi
/(x) = 0
on [O, rr]
on [O, rr],
/(x) = 1
10. /(x) = 1
on [O, rr],
f(x) = -1
on [-rr, O]
/(x) =x
on [-TT, 0]
on [O, rr],
on [-rr, O]
9. /(x) = x
11. f(x) = 2x
on [ -rr, O]
/(x) = 0
on [ -rr, O],
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
549
n-oo
cos
nx
n-co
18. In Theorem 6, under what simple conditions does the equality hold?
12.2
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
</J(x)
such that
That is,
lf(x) - </J(x)I
<
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.
y
f+
f
f-
The first clue to this situation is that the trigonometric polynomials form a bigger
system than one might think:
Theorem 1.
+ cos 2x
2
- 2
+ .l
cos 2x.
550
12.2
Fourier Series
i=l
a;B1cos ix sinjx,
b;A1sin ix cosjx,
b;B1sin ix sinjx.
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=
Jn=
L:
-0
J
f
cos2n t dt,
n
cos2 t dt =
-IT
K ==
n
We shall show that when
Theorem 2.
0
n
cos2 t dt.
i"0
cos2n t dt,
-o
is large, In
......
ro
R:1
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=
"
n
cos2 t dt < (7T
o) cos2
0.
12.2
cosn x dx
n
cosn-l x sin x +
551
: 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"
i"
cosn x dx
2n 2n
_,,
I,,
2n 2n
1
=
Therefore
li"
li"
cosn-2 x dx .
-11
cos2<n-ll x dx
2n - 1
_,,
2n - 3 2n - 5
2n - 2 2n - 4
111
2n
In-l
cos0 x dx
-11
(2n - 3)(2n - 1) . 17
2
2 4 6
(2n - 2)2n
3.5 .7
1 .
'TT
. . . 2n - 1 . 27T > - .
2 4 6
2n - 2 2n
n
n -
--
_,,
cos2n x dx
Therefore
Jn
,1.
..,, ..(x)
J'!..,,f(x + t) cos2" t dt
f'!..,. cos2n t dt
oo =:;>
c/>,,(x)
f(x).
1
c/>,,(x) :::::: - f(x + t) cos2" t dt;
In i{J
-!J
if c5 :::::: 0, then/(x + t) f(x) for -c5 t c5, and so
n
oo =:;>
::::::
oo =:;>
1
cf>n(x) :::::: - f(x) cos2n t dt
i{J
-!J
In
1
:::::: - f(x)I n
In
f(x).
552
12.2
Fourier Series
n-+OO
II/ - </>nllu
0.
Theorem 3.
Proof
For each
n,
</>n is a
trigonometric polynomial.
</>n(x)
f1l f(x t)
Jn
f" f(t)
1
cos2n
-1!
t dt
x) dt.
cos2n (t J -tr
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,
-
cos2n
Therefore
cos2n
(t
i=l
cos2n
(t
(a; it
cos
x) a0
! (ai cos
i=l
it
cos
it).
+ bi sin
cos
it ix
x) dt J:"a0f(t) dt
i [f:}a; it
; [f'"(a
+ b; sin
Therefore
J:/(t)
a0
ix
- b; cos
ai it
it ix).
sin
sin
ix
sin
cos
+ b; sin
sin it - b; cos
it)f(t) dt] ix
it)f(t) dt] ix.
cos
sin
The coefficients here are complicated, but they are constants, and so the indicated
()
nX then
f':_" f
lim
Given
> 0, we need an
n
cos2n
f7:_1l cos2n
n-.oo
Proof
(x t)
II/ - </>nllu
t dt
t dt
'
0.
such that
<
E.
-=;>
lf(x)
</>n(x)I
<
for -7T
7T.
12.2
553
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
t.
(1)
Ix -x'I
ltl
lf(x) (x')I
t )/
/j(x) f(x
1,
(x t) t dt
lf(x)- ef>nCx)I \ f(x) s,,fJ,,
t dt I
L JI(x)J: t dt -J:/(x t) t dt I
;n If,, [f(x)-f(x t)] t dt /
t)/ t dt
- J " lf(x) -f(x
/j
[J_,,-/j/f(x)-f(x t)lcos2ntdt+ J_)f(x)-f(x
f/f(x) - f(x t)/ t dt]
t dt f-o t dt Jo( t dt]
1. [ f f-o t dt]
_!_ [
l . .=J"
tdt
This means that
< b
< b
< E/2.
-f
=>
=>
< E/2.
cos2n
cos2n
cos2n
cos2n
_,,
cos2n
cos2n
1
I
n
1
=
I
n
(2)
+ t)/cos2ntdt
1M
=
I
-tr
cos2n
4MJ +
n
2
J
< 4M n+
I
I
n
n
J
n+-.
4MIn
2
+ 2M
cos2n
"
cos2n
cos2n
-1I
cos2n
Therefore there is an
n
We now have
which was to be proved.
N
=>
N =>
such that
J
4M n < .:
I
2
n
I !-
ef>nllu < E,
cos2n
554
12.2
Fourier Series
Prnf
a0
i=l
Proof Let
0.
cp(x)
Prnf(xn).
Therefore
J:"[f(x) - Prnf(x)]
1 dx
0.
12.2
SSS
At first, this theorem may seem almost like a joke, but it isn't. See the following
section.
PROBLEM SET 12.2
cos2 (2x)
7. cos x sin 2x
2. sin2 x
3. cos3 x
5. sin2 x cos x
6. cos2 x sin x
8. cos4 x
9. sin4 x
</>n(x)
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.
:n: f,,Prn/(x) dx
f/
(x) dx.
15. Now show that the same result holds on every interval
[a, b].
(1 + xr
C)
i
x
by the methods of calculus, in the real domain. Thus the proof in Section 10.8 does not
show that
z.
* 17. Now use the result of Problem 16 to get an explicit formula for cos4 x in the form of a
12.3
Fourier Series
556
</>,,.(x)
x) 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.
=
a0 +
L (a; cos ix
i=l
+ b; sin
x).
Is it always true that <P is its own Fourier series? That is, do we have
form n?
*21. Let/be a continuous function on [O, 1]. Show that for every E > 0 there is a poly
nomialp(x)
L=o a; x; such that
=
lf(x)
p(x)I < E
(0
1).
12.3
p(x)I
< E
for a x b.
a0
+ !(a; cos
i=l
ix
+ b; sin
a0 +!(a; cos ix
i=l
ix)
+ b; sin
ix)
12.3
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.
That is, the projection of the derivative is the derivative of the projection.
Proof Let
Prnf=
a0
i-=1
Prn f'=
A0
ao =
27T
-1T
ai =
1
1T
-
(i >
0),
i( >
0),
-1T
( ) sinix dx.
bi = .!. [" fx
J_1T
1T
Similarly,
A0 =
1
27T
f"f'(x) dx,
-1T
Ao =
( )- f( 7r )] =
_L [f7r
27T
0.
In i fori >
-i sinix,
v = f(x).
This gives
1T
Bi =
1
-
1T'
1T
( ) sinix]:'.." - [fx
1T'
iai = -iai.
Fourier Series
558
12.3
D[ i
a
(ai cos ix
] i
bi sin ix)
(-iai sin ix
In fact, at this stage, we don't know that either of the,indicated'series converges to any
function at all.
Jim
n-+CO
Prn
is related to integration.
II! - Prnf 11
Theorem 5 of
o,
What we want is
0.
Let
Proof
C0[a, b]
[f
b],
g(x) dx
then
b
[g(x)]2 dx.
(b - a)
(A B)2 llAll2
we take
g,
[f
llB ll2,
I. This gives
g(x)
r [f
1 dx
] [i
g(x) . g(x) dx
( b - a)
(l
1) dx
[g(x)]2 dx.
This tells us that the integral of a function is small if the norm of the function is
small.
theorem.
Theorem 3.
Then
If - Prn/I,
lim Mn
'7t-+CO
0.
12.3
Proof
M=
559
27T
= 2 7T II! - Pr nfll2
By Theorem
Mn=
Then
limMn= 0.
n->OO
Let xn and x be any points of [-7T, 77]. Since Prnf' = (Prnf)' , we have
IJ
I f
[f'(t) - Prnf'(t)] dt
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
Fourier Series
560
12.3
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
=-
27T
Then
'TT
_,,
cos
ix dx,
1 f"f(x)
bi = -
-1T
a cos
sin
'TT
] =f(x)
ix)
on
sin
ix dx.
-1T
( -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:
=f on [ , b], then
n-+ro
b
b
rfn(x) dx = rf(x) dx.
n-+oo Ja
Ja
lim
This gives:
Theorem 6.
Iffhas period 27T andf' is continuous , then the Fourier series forf can
= lim P r nf' ,
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)
(We get the first of these equations from Theorem 4, and the second from Theorem
1.)
12.3
561
27T
f (x1)
Hf1(xJ
f(7T)
f(-7T)
-?T
f2(X1)],
and
7T
by defining
Hf1(-7T) + fa(7T)].
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
I (x)
<X)
Then
I)
For every
2)
3)
The series can be integrated a term at a time on any closed interval (even if the
x, L (x)
converges to f ( x).
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
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
1.
12.3
[f/
(x)dx
27T
f,,
] dx.
[/(x)2
on
(0, 7T),
(-7T, 0),
on
(4) /(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.
562
12.3
Fourier Series
i=l
4. Now calculate the Fourier series off (You did not need to do this, to solve Problem 3.)
5.
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
/(x)
/(x)
e"'
Proceed as in Problems
*9. Let
(2) /(0)
on (0, 7T),
on ( -7T, 0),
(4)
-e'"
0, /(7T)
! (e"
- e-"),
for every
x.
through 5.
f and g be functions of the Fourier type, and let L (x) be the series for f, as in the
111112
=Ia+ Ib.
00
00
i=O
i=l
Linear Transformations,
Matrices, and Determinants
13
13.1
LINEAR TRANSFORMATIONS
j;
Rn-'>- Rm
and
f(rxP)
If f is linear, then
f(rxP + {JQ)
f(P) + f(Q)
rxf(P)
for every P, Q,
f(rxP) + f({JQ)
(1)
(2)
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
E2
(1, 0, 0),
(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)
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)
f(Ea)
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
13.1
3 = 9 numbers ai1.
When Pis described in this notation, we call Pa column vector. Similarly, for
we write
f(P) =
[] [: ! ::: ! :::] .
=
Ya
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
Linear Transformations
13.1
565
[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.
1
-2
1
2
1
-1
1
-2
1
1] [X1]
2 X2
2 Xa
2
1
-1
]
[
X1 + 2x2 + X3]
-2X1 + X2 + 2x3
Xi - x2 + 2x3
[Y1]
Y2 .
Ya
Problem 1.
f(P)
f(P)
(1, 2, 3)?
Xi + 2X2
X3
1,
(1)
-2xi + x2 + 2x3
2,
(2)
Xi - x2 + 2xa
(3)
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
5x2 + 4X3
- 3 x2 + Xa
1,
(1)
4,
(2')
2.
(3')
(2) + 2(1)
(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
5x2 + 4xa
1,
(1)
4,
(2')
(3")
(3') + !(2')
566
13.1
X
3
1] [ ] [1]
1
H -1 ; -r ;
= H.
13 7
Xa)
in terms of
-11
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
']
M-1 =
[':
17
1
T7
- rr
5
T7
3
T7
T7
-1 7
17
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) .
R"Rm
(?) 1-1:
Rm Rn.
(?)
There are two things that may go wrong: (1) some points Q of Rm may not be values
13.1
(2)
Linear Transformations
1
of the function at all. In such cases, there is no such thing as J- (Q).
Q of Rm may be equal to f(P) for more than one point P.
567
Some points
[1 1
J
M=
0
0
Here
0 .
This function projects R3 into the xy-plane; no point outside the xy-plane is a value
of the function, andf(P)
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]
z)
(a,
[; ; 7][;z] [2 !+ ; !+ i7z []
+ 2y + 3z =a,
2x + + 8z =
+ + 7z
+ 2y 3z
+ 2z b - 2a,
=
2b +
3a - 2b +
2b +
{(a,
M=
4y
Sy
b,
4y
= c.
=a,
3a -
f(P) = MP
c.
(2)
(3)
(l)
(a, b, c)
c = 0. Let
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
13.1
568
long as
y = -2z +b - 2a,
x = -2y - 3z +a
Here we can choose any
= z - 2b +Sa.
z; if
=
(1')
then
f(P) = (a, b, c) .
the function.
If f is a function
f(A')
{bI b = f(a')
for some
a' in A'}.
f(R3) =
The
E=
{(a, b, c) I 3a - 2b +c = O}.
In the above example, the kernel is the solution set of the equation
(J..
for
{(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
M=
1 0O
0 0 0,
000
Linear Transformations
13.1
569
rn J[J m
0,
0,
0,
1)
As far as the ideas in this section are concerned, there is nothing special about
Exactly the same methods apply, in exactly the same ways, when we deal with
2)
In the examples that we have discussed, the image and kernel of a linear trans
formation have turned out to be subspaces. This is what always happens. (The proof
11.3.
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
M-1 = [
(a)
or
(b) /(R3) ={(a, b, c)
[H ]
. [HJ
[H :J
I
.
},
and
[[ !]
[: ; ]
[H i]
-3
+2
I}.
, [H ] ]
[- - -;
[h i] [! g ]
Ker/= {(x,y, z)
6.
10.
-2
570
ll.
12.
ll 2
a,
aa1 Olll 2 Olll3
f3a1 f3a2 flaa
"]
u
[T ]
["" g l
where a11a22a33
13.
13.2
ll12
ll22 ll23 ,
ll33
0
0.
M=
14.
with a11a22a33 0.
Same question, for
M
15.
a21 ll22
ll31 ll32 ll33
[g
ll31
with a12a23a31 0.
Same question, for
M
13.2
ll12
0
ll12
rn
"
aa
J
"
aa
Theorem 1.
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)
and for eachfP
( ),ocfP
( )
to Ker f, then
and sof(P + Q)
( + Q);
fP
f(P) + f(Q)
foc
( P)
f(Q)
0 + 0
ocf(P)
oc
0,
0,
0
and
=
0.
13.2
Theorem 2. If f and
andrx.f
571
Here the sum and scalar product are defined by the obvious conditions
(f + g)(P)
f(P)
(rx.f)(P)
g(P),
rx.(f(P)).
(f + g)(P
Q)
f(P
f (P)
Q)
g(P
+ f(Q) +
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.
n,
space.
Theorem 4. If
g:
f(g):
R"'-+ RP
is also linear:
f(g)
f[g(P
Q)]
f[g(rx.P)]
f [g(P)
g(Q)]
f(g)(P)
f[ixg(P)]
f [g(P)]
f[g(Q)]
f(g)(Q );
rx.f [g(P)].
g:
572
13.2
Therefore, for P
(x1, x2,
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
b12
b21 b22
b1m
b2m
bn
X1
X2
Y1
Y2
=
b,.1 b,.2 .
bnm
Xm
Yn
Y1
Y2
=
Xm
Yn
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,
13.2
573
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
Then
Mu
and so
an
a21
a12
,
a22
[:] [
{:J
=
Mt<a
Mg=
=
=
bn
h21
h12
h22 '
J[ J [
[J [J
[ J[ ] [
[
[
[
bu
b21
an
a21
a12
a22
= M
[en J
C
Mt<g> =
12 .
C22
C21
J [:].
bnX1 + h12X2
b21X1 + b22X2
X1 =
X2
b12
b22
= M M X1
g
f
X2
Yi
Y2
a11Yi + a12Y2
a21Yi + a22Y2
Yi =
Y2
J
J
J[ J
auh12 + a12b22
a21b12 + a22b22
X1
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
bu
h21
X1
X2
b lm
b2m
h12
b22
Mu
Xm
bnl
bn2
b nm
_L;'!,1 bi; X;
L7!.1 b2;X;
X1
Xz
Y1
Y2
-
_L;'!,1 bn;X;
Yn
574
13.2
Here M0(P)
(y1, y2,
, Yn), where
m
Yi
2: bi;X;
J=l
(i
1, 2,
.. , n).
Therefore
ap1 ap2
Yn
Yn
Z1
Z2
L1 ali(Lf:1 b;1x,)
Lf=1 a2;(2:f!=1 b;1x1)
2:f=1 aliyi
Lf=1 a2iYi
=
apn
top, we have
where
LCk;X;
J=l
(k
1, 2,
'p),
Thus
ckJ = L akibii"
i=l
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
13.2
ck; = z1 akibi;
c22
(-2, 0, 2)
and
(3, 2, 1).
575
Therefore
c22 =
10.
-4
4 .
-2
Similarly, we define the sum of two matrices to be the matrix of the sum:
Definition.
M1 +Mg= Mt+u
The sum is easy to calculate; the simplest possible idea works. Let
au
ll21
a ln
a2n
ll12
ll22
b12
b22
Mg=
M, =
Then for
bu
b21
P = (x1, x2,
, xm),
we have
L=l aux;
L=l ll 2;X;
f(P)
g(P) =
Therefore
z7=1
L7=1
(f + g)(P) = f(P)
g(P)
(a1; + b1;)X;
(a21 + b21 )x1
13.2
576
[ci;]
au
a21
aln
a2n
a12
a22
[ai; +bi;]
[ai;] + [bi;];
bu
b21
b12
b22
bin
b2n
bml
bm2
bmn
+
am1
am2 ... a mn
a u+b11
a21 +b21
012 +b12
a22 +b22
aml +bml
a1n +bin
a2n +b2n
Carry out the indicated operations, expressing each answer as a matrix (which may, of
1.
4.
7.
10.
13.
16.
19.
[ iJ[: :J
[
rn
[!
n
[
[l
2
5
8
0
2
0
0
0
4
5
6
1
0
0
1
3
0
[ m: ]
[! m i]
f]
[ mi i]
f
ff f
[! ff""
[ m: ]
2
2.
m !]
urn
!][]
m !]
n
m ]
0
1
5.
8.
11.
14.
10
3.
[ m !J
[! !J[l]
[ mf J
G f
[l m ]
[' i][i ]
1
1
9.
12.
0
0
l
15.
18.
20.
[; ][ ]
0
6.
17.
0
0
[3
21.
15
18
14
13.3
22.
24.
26.
[l mf i l]
[ f 1J
4
23.
1
0
25.
1
0
27.
[ _l
][
2
I
-6
-6
577
-2
4
-1
I
-1
[ J
[ T
0
1
0
For each of the following four matrices, find the i nverse M-1 if there is an inverse; if
OJ [
not, give the simplest reason that you can for concluding that no inverse exists.
28.
[
2
*32. Let
29.
10
11
is it true that MM
12
30.
12
10
11
6
0
[ l
4]
8
12
0
31.
[ ]
[ ]
12? (It is easy to find two such "square roots" of 12 The question
33. Let
02
[ ].
M + 02
or B
02?
2 matrices and AB
M.
02
02?
.
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.
set of all
by
The
erties of vltn, under addition and multiplication, it will not be very useful to think
578
13.3
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
g + fo
(0, 0, ... , 0)
A.4.
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
M1tu>,
o.
(g
h)
(g
h)
(f g)
0
fog
h.
13.3
w,
579
z,
2n, the
operation
is associative.
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.
Definition.
2, with
[2,+,
The system
o]
R.1.
The pair
R.2.
The
R.3.
arid
2 contains
a unit element.
All this discussion carries over immediately to the set ._,n of n by n matrices,
since M1 + Mu = Mt+u and M1Mu
TheQrem 2.
Here
[Jin,+,
0]is a ring.
eleinent of ._,n is
0
...
580
..,1tn
In
13.3
is the matrix
1
=
0
with l's on the main diagonal and O's everywhere else. There is a shorthand for this:
we define
/J
..
Z?
for
for
i = j,
i -:;6 j.
We then have
In
[o;;].
by
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
by
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:
(Closure).
Cl
G is closed under*
(Associativity). a* (b *c)
EU
for
EI
group
(Existence of Unity).
every a.
(a*b) *c,
always.
There is an element
of G such that
(Existence of Inverses).
a-1
a-1*a
e.
commutative
e *a
a-1
a*e
a,
of G such that
Com
(Commutativity). a*b
b*a,
if it satisfies:
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
ring, (2) multiplication is commutative, and (3) every
that x
x-1
1.
x -:;6 0
(1)
[F, +, ]is.a
13.3
581
Obviously the real-number system furnishes examples of all the ideas that we
have been talking about in this section:
[R, +] is a group.
[R, +, ]
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)
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
z=cosO+isinO
Which, if any, of the following statements are true, and why, in each case?
a) [G, +]is a group.
b) [G, ]is a group.
2. Let
3. Let
G be the set of all complex numbers of the form m + in, where m and n are integers.
z =
as in Problem 1.
iy (y inR.) Discuss
Discuss as in Problem 1.
+ bv2
=
a
Discuss as in Problem 1.
a +
b1T,
1T
0, then
bVZ,
b = 0.
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
12
A
[P3, o).
(x1, x2)?
plication table, but you should find out whether the group is commutative.)
set of all n
is not.
582
13.4
zn,
12.
Jog.)
Let GL(n) be
*13. Let
GLU(n) be the set of all upper triangular n x n matrices [ai1], with a11a22
[GLU(n), ] is a group, but [GLU(n), +, o] is not a ring.
0. Show that
13.4
ann
The determinant function assigns, to every square matrix, a real number. The
definition of this function begins as follows.
a11
ll21
Given an
by
matrix
[a;1],
a 12
ll22
ll1n
G2n
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
explained how the sign is to be chosen for each term, we shall have a function
det:
,A
---+
R.
The rule for the signs takes time, to explain and justify. With each term
In= {l,
Here
p{i)= j;;
that is,
p(i)
2, 3,
... , n}.
13.4
(1 2 3 .
jl j2 ja
form:
p
with the numbers
583
n
jfl ,
(13 12 43 24)
1H3, 2H1, 3 H4, 4H2.
For example,
A one-to-one function
p:
In---+ In is called a
permutation.
described in the two-line notation, the order of the columns does not matter; all that
matters is what is under what. For example,
and so on.
If
then
is called a
transposition.
For example,
(13 22 13 44)
1 3.
p
=
(13).
is a transposition; it interchanges
hand
Theorem 1.
In general,
(ab)
and
and b.
product
example, for
For
G i ! ).
) = (21)
)
(13)
=
G i ; :
G ; :
(24)(21)(13).
) (24).
=
We now have
p =
As always, for composition of functions, the operations are performed in the order
from right to left. Thus
----+-
584
13.4
4H2.
3H4,
2H1,
1H3,
p=
we first take the transposition
(1
2 3
ji j2 ja
(1 j1);
n)
jn '
1,
the resulting sequence, we apply a transposition which puts j 2 in the second position,
p=
2 3
5 4
4 5
1 7
6
3
5 4
5 4
(12)
(15)
(34)
(31)
(37)
(36)
1,
diagram gives
p=
G ; ! 1 )
(36)(37)(31)(34)(15)(12).
p=
(36)(73)(16)(43)(57)(24)(16)(27)(14)(57),
which looks different, and uses ten transpositions instead of six. Nevertheless all such
expressions for a given
A permutation
p
number of transpositions;
positions.
is odd if
is the product o
13.4
Proof
The
585
i<j
where the expression on the right is the product of all differences
i < j.
xi - X;
for which
to denote
_2;=1 xi).
For example,
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
changing
xi
and
X;.
1)
(xr - xi),
(xr - X;)
(r < i),
2)
(xi - x8),
(X8 - X;)
3)
(xi - xt),
(x; - xt)
(j < t),
4)
(Xi - X;).
(ij),
3)
( <Xr - Xi)
(x, - X;)
{(xi - x,)
(x, - x;)
{(xi - xt)
(x; - Xt)
(x1 - xt),
(xi - Xt);
4)
(xi - x;)
(x; - xi)
1)
2)
H
H
H
H
H
xi
interchanging
and
X;,
(i j),
thus inter
the effect is
(Xr - X;),
(xr - xi) ,.
(x; - x,)
-(X8 - X;),
(x, - xi) = -(xi - x,);
=
-(xi - X;)
Thus the factors of the first three types fit together in pairs, and in each case, the
It follows that
(ij)f
pf= f if p
xi - x1
is changed, and so
-f
-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
anin'
586
13.4
. .
(1. 2 3 . )
j2 ja
Ji
'
'
}n
possibilities to choose
-1
n(n - l)(n - 2)
3 21
20,
n by n matrix is
n!
the sum of
n!
terms. In particular,
20! 2,432,902,008,176,640,000 .
31,526,000.
For each
n,
Then
[Sn, ] is a group.
o
Proof
2)
3)
4)
(1)
o,
The operation
There is an identity
e
By Theorem
Theorem 5.
3,
(1, 2, 3,
1, 2, 3,
... ,
... ,
n
n
in reverse order.
Thus
because
(p
q) o (q-1 o p-1)
p o (q
q-1) o p-1
p o e o p-1
p o p-1
e.
13.4
587
o.
says that
(P1P2
Pk-1Pk)-1
p/;1p/;!.1.
P21P1\
Proof
We express
in
Sm p
p;
P1P2 .
h-1h
p -l
If k is even, then
The
as a product of transpositions:
p
where each
p-1
and
transpose
and
p -1
hh-1 .
P 2Pi
and in general
Proof
p -1
det M.t
(1
p
=
ji j2 ja
'
.
'
}n
Ct 1
j;
n)
588
13.4
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
A jz ja
jn
For example, if the first and third rows are interchanged, then the sign of the term
a
a
1ii 212
a
nJn
(1
q =
h j2 ji
()1 ]a)
p.
1.
Working directly from the definition of <let, get an explicit formula for
<let
[anll21
ll12
ll22
["" ""]
au
<let
a,, ""
ll23 ll24
ll33 ll34
ll44
[001 010 ]
000
0
0
[
0
5. 0 001 ]
00 01
0
[ 000 007 400 !]
<let
0
OJ
[1
0
0
1
0
4. 00 00 01 010
7. [ 001 00
0 4 ]
<let
6. det
8.
det
00 00
0
[ 000 007 400 !]
00 000 401
[! 00 07 00 ]
2
9. det
13.4
10. det
12
det
[i i i]
[H H ]
[! : ]
o
14 . det
-14
-2
15. det
17 . det
-6
0
-4
0
0
0
3
0
6
16. det
-6
0
0
589
0
0
0
0
1 and h
2 or
22. Can any general statement be made about the evenness or oddness of the following
permutation?
_
p-
n n- 1
n- 2
n-2
n- 1
...
;)
23. What can you say about the sign of the following permutation?
=
q
n 1
. .
n- 2
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?
det
[: f: fJ
0.
590
det
X1
a
2
x2
2
x2
3
x2
X2
X3
xa
3
13.5
x3
0.
13.5
x ,
2
, Xn).
EXPANSIONS BY MINORS.
CRAMER'S RULE AND INVERSION OF MATRICES
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.
0.
The reason is that when the two identical rows are interchanged, nothing happens
to the determinant (or even to the matrix).
detM
Therefore detM
-detM, and
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
a ln
G12
G1
a 21
G22
a 2a
a2
aa1
aa2
aaa
aa n
Expansions by Minors
13.5
591
permutations
l
1
)2
)a
.. '
)2
}n
..
"
Jn
1
The sum of all terms of det Mthat involve a;j is (-l)i+ ai1Dw
1.
.._.. ._,..
For j
- 1
We assert
, Cn).
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 )<i-l>+!i-1>
and so
det M
det M
i
(-l) +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'
For each i,
n
i+i
.. ( - 1 )
det M = ""
aiJD;J
=l
i
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
592
(n - 1)!
13.5
Theorem 5
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,
Theorem 6.
If k ;;C j, then
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
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
3.
[ai11
auX1
a12x2
aiaXa
a21X1
a22X2
a2axa
aa1X1
aa2X2
aaaXa
ba.
b1Du
-b2D21
aa1Da1X1
Dx1
a12D11X2
G32Da1X2
x2
x2
b2,
a13Duxa
auD11X1
and
b1,
D
det M, and suppose that D ;;if 0.
D11, in the second by -D21, and in the third
x3 are 0, by Theorem 6.
G33Da1Xa
0
x3
baDa1
i
L:=1(-l)i+ b;Di1.
x1 is D, by Theorem
Expansions by Minors
13.5
593
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
b/s.
x2, we multiply in the three equations by - D 12, D 22, and - D 32
0 Xi
Dx2
0 X3
3
2 i
! (-1) + biDi z
i=i
replacing the second column of M by the b/s. The same scheme works for
fore, if
X2
Di/D,
D2/D,
X3
D3/D.
3. In general,
we have:
Theorem 8
bi
b2
X1
Xz
M
Let
det M. If
D 0, then the system has one and only one solution; and the
X;
D ;f D,
where
Theorem 9.
ci i
1
-(-1)'+'D 1i
D
i
That is, M- is the transpose of the matrix
[ (-l)i+iD;;J
594
13.5
Here
Dxi
I (-I)i+iyiDii
il
xi
What we want is a matrix M'
I ( - ly+i(l/D)Di;Yi
i=l
[c;;l, then
X; =
xi
L ciiYJ
il
to this form, we interchange i and j on the
(-ly+
i l
(-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
x + 2y + 3z
4,
2x + 3y + 4z
5,
3x + 4y + 5z
6.
13.5
Expansions by Minors
595
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
2x + 3y + 4z
3x - 4y + 5z
-4,
-5,
-6.
5z
4,
-5,
0=
6.
Find the inverses of the following matrices, by a direct application of Theorem 9, and
check your answers by matrix multiplication.
4.
7.
[ ].
G ].
5.
[ !l
9.
[ l
1
11.
[! ]
-2
[! l
1
6.
8.
10.
[ ].
0
0
[ l
the determinants unless you need to, as a step in finding the inverse.)
12.
15.
[ l
13.
-1
-2
-5
[ !l
0
14.
16.
[ !l
0
[ ]
1
17. Suppose we form a 4 by 4 matrix by fitting together four 2 by 2 matrices, like this:
Let D
det
Dn
D12
D21
D22
13.6
596
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
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
to the ith row, giving a matrix M'. Expanding M' about the minors of the ith row,
we get
det M'
n
;
! (- lY+ (a;; +aak 1)D;;
i=l
n
n
;
;
! (-1y+ a;;D;; +a! (-l)H ak;D;;
i=l
i=l
detM
+a
0,
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
sets of
(n
1) !
terms each.
Tbeorem 2. If the rows of a matrix M form a linearly dependent set, then det M
0.
Proof
Let M
[a;1];
r;
n
! a;r;
i=l
for some set of numbers
a;,
L {Jiri
i=2
a;n);
suppose that
0,
r1
(a;1, a;2,
Then some
r;,
13.6
597
If detM
also do the columns).
Theorem 3.
0,
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
-
{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.
span the same subspace of Rn. This gives:
}.
<letM'
detM
and
we have
2) <letM{1
0.
Theorem 4.
Proof By the preceding two theorems, each of the following conditions is equivalent
to the next:
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
598
13.6
M(x)
2
2x
2
2x
[l
x
2
x
Here the columns are linearly dependent, obviously; but the rows are not: if
for every
for some real numbers
1)(1,
1)(2,
then
1)(3,
1)(1
1)(2
1)(3
x,
The simplest general test for linear independence of functions uses the determinant
of a matrix of functions.
Wronskian
of the sequence
Jn
be functions on an interval I.
The
is the function
!1 f{ f
!2 f f
fn J f
Thus
W(x)
det
n l
f - )
[JJ1-1>];
the Wronskian matrix has the (j - !)-derivative off; in the ith row andjth column.
Note that the Wronskian really depends on a
on a
may
That is,
If
W(x)
{fi,/2,
0 for every
fn}
x.
x, the
n
2, a,f(x)
i=l
et.i
W(/1,/2,
(Here
Di[ ]
for every
for every
x,
0, then automatically
(j
1, 2,
. . . , n
1).
x, and
n
2, l)(;f1>(x)
i=l
,fn)
0.
D1 2, et.Jlx)
i=l
(j
1, 2,
,n
1),
13.6
599
Jn} is linearly
independent.
j;(x) =cos x.
Here
W(x)
det
sin x
cos x
c ?s 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
Here
W(x) =det
Here W(O)
By Theorem
0, but
5',
w()
=det
flx)
sin 2x.
snx
cos x
sm 2x 2 cos 2x
[ J
-
-2.
Here
[ xe"' xe"'
xe xe
[1x 2x
xe2"'
J
W(x) =det
==
2 .,
det
+ e"'
2 ., + 2xe .,
1
=det
[ xe"'
x2e .,
e"'
2xe"'
=x2e2"''
fi(x) =x2,
2x2 for x 0
!2(x) = 3x2 for x o'.
600
13.6
for every x.
If the above equation holds for every x, then setting x = -1 and x = 1 we get
By subtraction, oc
oc1 + 2oc
2
oc1 + 3oc
2
= 0. It follows that
oc1
= 0,
0.
= 0.
This indicates that the Wronskian can give proofs of linear dependence only for
special types of functions.
I I
= det
[ J
ad
be.
1. {e"', e2"'}
3. {e-"', e"'}
5. {sin x, sin2 x}
8. {x, sin x}
9. {ea"', eb"'}
(a -,= b)
17. {cos2
(a -,= b)
x, sin2 x}
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
D4
0.
x1, x , and x are all different. How do you know that D4 is a poly
2
3
X 1 x12 x13
X2 x22 x3
2
X3 x32 x33
x x2 x3
13.7
*24.
601
x2
2
1
1
*25. Investigate
Xn 1
-
Xn
a '"
. . , e n
,
different.
13.7
(1)
f" -f=O
because
D2e"'
then so also is
e'"
and
D2e-x
f1 + rx2f2,
rxi
e-'".
rx.1
and
rx.
2
pn>
f <n-1)
an_i
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.
f(x) = em'",
f'(x) =
memx,
f"(x) =
m2emx.
602
13.7
and since
emx
m2
Equation
bm
0.
(2)
solutions.
I.
If b2 -
4c
-b
)b 2 - 4c
-b -
)b 2
4c
4c
m1
and this is a real root of multiplicity
m2
III. If b2
where(/..
- 4c
2, with
bm
-b/2,
(m - m1)2.
-b/2 and (3
)4c
b2
Also
{f1,f2}
W(/1,/2)
det
[:::: ::::::]
em'xem2x m
2
em1xem2x
det
- m1) -:/:- 0.
Therefore the solutions that we have found for case I include all the solutions, if
the following theorem is true:
Theorem 2.
f"
Then dim "Y
2.
bfI
cf
0.
603
13.7
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"',
f(x)
f(x)
bn(x)
cf2(x)
m1Xem1x
em1x,
mxem1x
m1em1x
mixem1x
2m1 em'"';
xem1"'[mi
bm1
m1em1x
c]
e2m1x det
m1
m1x +
[1x
1l
_J
+
0
em1"'[2m1
b]
0
,
e2m1 x
0.
Case III looks peculiar. Taken at face value, the roots of the auxiliary equation
give us
ea"'(cos f3x + i sin f3x),
e<a+Pilx
f 1(x)
=
f2(x)
e<a-Pilx
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)
with
cp'(z)
In particular,when z is real,
rp(z)
J (x)
em"''
entz'
mem z,
x, we have
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
604
g{(x)
g1
13.7
g(x )
Therefore
cos
{Jx.
- e2ax det
=
because
{J
e2""' <let
[ . {Jx rx
{Jx rx
[ ? {Jx
cos
cos
sm
sin
c s
sm
0.
{Jx
{Jx
- /3 sin
{Jx
{J
cos
{Je2""'
'
f(x)
k1g1(x) + k2g2(x)
f(x)
where
+ k2 sin
.Jk21
(.Jkiki
k22 e""'
x0 is
cos
{Jx +
k2
.Jk
sin
{Jx
and
{Jx)
l 2
'\/ k 1
f3x0
k22
k1/k
and sin
f3x0
we get
f(t)
k 2/k.
Using
t for x,
which describes the motion of a particle along a line, with the position given as a
function of the time.
graph of
f,
{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
f"
is completely determined if
bf
cf
b and
c in
the differential
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
605
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
is not too great, then the ratio w/s is a constant k; that is, s = w/k;
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.
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
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.
0
.'V
606
13.7
Suppose that the spring has been stretched to a point with coordinate x. Then two
forces are acting:
1)
because x +
-k(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
+ s)
F = -k(x
-kx.
+ w=
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)
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
the acceleration g of gravity. Thus
m times
F(t) = -
a(t).
-kx.
Therefore the function f which describes the motion must satisfy the differential
equation
w
- a(t) = -kf(t),
g
kg
w
f(t)
= 0.
f"
where bf
- 4c
= -4c <
Of' + cf = 0
(c
>
0),
0.
on the right. In each of these cases, you should use .the methods but not the results of this
13.8
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.
2. f" - f'
j'(O)
= 3.
f'(O)
= 2.
= 0.
f'(O)
= 2.
f'(O)
= 2.
= 1.
= 1.
= 0.
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,
j'(O) = 1.
and
*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"
and for
bf'
cf
= 0;
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
In the preceding section, we found that for every equation of the form
(1)
13.8
608
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
Proof
(1)
f" = -bf' - cf
Here the righthand side is differentiable, and so also is the lefthand side. Therefore
p4> = -bpa>
and by induction,
pn>
cp2>;
-bpn-1) - cpn-2)'
(1)
oo
I a;(x - a)'
i=O
oo
f(i\a)
i=O
I-.- (x - a)'.
!
We shall now show that f is real-analytic; that is, the Taylor series converges, for
every
x,
(I),
j<il(a)
for every
x.
of Section
10.10.
Note that we are now using it for the first time.) The theorem says that for each
x,
the remainder
Rn(x)
f(i)(a)
i=O
= f(x) - L -. - (x - a)'
!
(x) =
n
f(n+l)(x
)
(n
1)!
(x - a)n+l'
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.
x is fixed,
13.8
Let
and let
interval
k !cl,
k lbl,
609
k 1;
and
M be a number which is an upper bound for both lf(t)I and l/'(t)i, on the
I
{t I It - al Ix - al}.
t on
For each
x?
x?
I, we then have
lf"(t)I
1-bf'(t) - cf(t)I
kM + kM
Similarly,
l/<3\t)I
2kM.
I-bf"(t) - cf'(t)I
2kM + kM
< (2k)2M.
for every
This is known for
n
- 2 and n
{1pn-1>(t)i
1 and n
n.
(2kt-2M
IJ<nl(t)I (2kt-1M
=>
IJ<n+i>(t)I
1-bpn>(t) - cpn-ll(t)I
k IJ<n>(t)I + k IJ<n-ll(t)i
k(2kt-1M + k(2kt-2M
=
because
(n + l)st
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
610
13.8
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:
-
and
fi(a) = f2(a)
f {(a) = f (a)
for some a,
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
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"',
n-1
mn + I bimi =
iO
o.
13.8
If
k1 (so that
611
If
IX
{Ji,
IX
{Ji
are a pair of
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.
obtained in this way forms a linearly independent set, and the number of elements
in the set is
2,
analytic; and matrix theory then furnishes a proof that every set of
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
f"(x)
5f'(x)
6f ( x)
Consider,
(I)
e".
f"
the solution space is
Equation
(2)
"f/
is called the
Sf'
{ IX1 e-2x
6f
+
(2)
O;
1X2e-3'"}.
(l),
f"(x)
5f'(x)
6f(x)
then
e"'
and
f(x)
6(f
fo)
5f(x)
6f0(x)
e"',
Cf
Therefore the function f
fo)"
- Jo
+ (f
f(x)
(1) has
0.
form
every solution of
- fo)'
1X1e-2'"
1X2e-3'"
- f0
(2).
This means
all solutions of
(1)
in the
+ fo(x);
is of the form
1X1e-2"'
1X2e-3"'.
612
13.8
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
I(x)
and solve for the coefficients
co
L aixi,
i=O
ai one at a time.
then the method of trial and error may work faster and lead to a simpler formula. In
the example above, we try
l(x)
and try to find
J'(x)
Ae"',
l"(x)
Ae"',
Ae"',
so as to make I a solution:
12A
Therefore the solutions of
(1)
1,
1\,
l(x)
/2e"'.
hyperplane.
e"',
form a set
is called a
lo is any point
hyperplane.
In
U +lo I I in if/}
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
hyper
{P +P 0 I Pin E}
because in vector spaces of higher dimension, the
2.
First we try
(?) l(x)
sin
This gives
(?)
- A sin x + 5A
cos
x + 6A
sin
x.
(?).
sin
sin
(?),
(?) l(x)
sin
x +
B cos
(?),
13.8
613
which gives
-A sin x - B cos x
+
6A sin x
6B cos x
sin x
<=>(-A - 5B
6A) sinx
( -B + 5A + 6B) cos x
5A - 5B
1
5A + 5B
0
<=>
sin x
This gives
For each of the following equations, find the space ii" of solutions. Answers should be
in the form
+ an fn}
11' = {aif1 + a2f2 +
1.
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=
f" +f=sin x
10. /" +f = 1
8. f" +f = e"'
11. f" +f = cos x
14. f" +f = x3
16. f" +f = x2 +1
9.
14
14.1
Given a
right cylinder with base Bis the union of all lines that intersect
Bz
I
I
,,
/
I
I
I
IB
,.--i
The figures above suggest that a cylinder is a bounded figure, with a lower
base B1 and an upper base B2 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.
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
surfaces
and
cylindrical solids.
cylindrical
14.1
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.
-1
__
__
-----
I
I
I
I
/)-
)r
---
7c-- - y
,,
On the left above we show the unit circle in the xy-plane; this is the graph of
the equation x2 + y2
no restriction on z, the graph includes the vertical line through each of its points.
To be more precise, the circle is
{(x,y)
x2 + y2
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.
the dotted guide lines.
Hence
y
1 -----
::..._
--t-<
-71
1
I
I
I
I
I
I
__:_I
___
y
I
/--L
x
_.
__
y=x2, Oxl}.
/ /i
I
/
I
I
I
I
__
__
_
-L
y
/
/
x
x
-----
{(x,y,z)iy=x2, Oxl}.
y2
instead of the corresponding equation. Using this as base we get the solid cylinder,
14.1
616
which is
{ cx,y,z)
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
y
x
x
{(x,y,z)lx+z=l, O;;;x;;;l}.
x+z=l, Ox;;;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
1,
ix
y =
+ 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
surfaces are the
7.5.
Given a curve in, say, the yz-plane, we may rotate the curve about !he y-axis.
14.1
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,
[f (y))2.
-a
1)
-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
<=>
2)
x2 + y2 + z2 = a2.
in the yz-plane.
surface).
x = 0,
cone
618
14.1
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
(b)
x2+y2 - m2z2
0,
0.
t------
y
:t
(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,
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
619
and
'
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
(Query: How do we know that the constant term is unchanged?
0.
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'
Z =
,/.
ff
X Slll 'I' +
,/.
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
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.
620
14.2
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
4. x2+z2=1
5. x2+y2
7. z=y2,
0y 1
3. y - z=1
6. y2+z2=4
8. x =z2,
9. x =4y2, 0y 1
Ozl
10. (x2/4)+y2=1,
x,y 0
11. (,y2/4)+z2=1,
y, z 0
x, z 0
1 4. x = ly
x, z 0
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, 0y 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.
14.2
is a sphere of radius
a.
Y2
z2
a2
b2
c2
-+-+-=1
is called an ellipsoid.
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
621
For
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
a2
b2
--y2 +-=0
/yo
x
x
--y
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
x
=0,
z2
y2 =-.
b2
622
14.2
Similarly, the cross section in the xy-plane is a pair of lines, because it is the graph of
y2 =
z = 0,
x2
x2
a2
z2
b2
= Y2o
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)
y2
x2
z2
-+- --=l.
a2
b2
c2
Therefore all horizontal cross sections of the graph are ellipses. Rewriting in the form
y2
x2
a2k + b2k
1'
we see that as lzol increases the ellipses get bigger, but their shape does not change.
I
I
I
)- - --,_
__
_
_
14.2
623
The cross sections in the other coordinate planes are hyperbolas; they are the graphs
of the conditions
x
y=
z2
b2
c2
- -- =
0,
z2
x2
--=l.
a2
c2
1'
4)
x2
y2
--
a2
z2
-
b2
c2
x0 is empty; for x
z2
v2
x2
=
'--+
- 1.
b2
c2
a2
y2
0,
z2
y2
-+
b2
c2
X0,
x2
_Q
a2
1 >
0.
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
CZ= -
b2
x2
(c '=r6
a2
b= c
1.
0).
624
14.2
-1,
1, x
(Rather oddly, it is a bad idea to draw the cross section in the plane
0, andy
0.
O; this is a
pair of intersecting lines, and when we indicate it correctly, the result is very hard to
interpret pictorially.)
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
PROBLEM SET
14.2
Sketch the graphs of the following equations, and identify the surfaces.
1. x2+-+-= l
4
9
2. x2--+ - = 1
4
9
2
z2
y
3. x2+--- =1
9
4
2
z2
y
4. x2----= 1
4
9
2
z2
y
5. x2 - - - - =0
9
4
2
z2
y
6. x2+---=0
9
4
z2
2
y
7. -x2+---=0
9
4
8. x2
2
y
9. z=x2-4
z2
____
10. z2 =x2+t
4
-1
14.2
625
y2
y2
12. z2 = - - x
11. z =- - x2
4
y2
b2
x2
a2
(c : 0).
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 =- - - c b2
a2
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.
-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.)
626
14.3
Find the volume of the solid which lies between the planes z = 0 and z = 1, and
inside the one-sheeted hyperboloid
20.
2
y
2
= z + 1.
21.
2
x
= z
+ 1.
Find the volume of the solid which lies between the planes z = 1 and z = 2 and the
two-sheeted hyperboloid
22.
z
14.3
= x
+y
+ 1.
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
2)
Functions of one vector space into another. These were always linear, and were
f:
D-+ R,
'
I
I
,--1=- ---- .... ,
I
.
I
I
I
I
I
I
l I
1
-l
I
p
D
:
:
c=0
x
z=f(P).
14.3
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.
{(x, y, z)
and
(x, y) in D
z =
f(x, y)},
or, equivalently,
{(P, z) J P in D
and
z =
f(P)},
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)
P-+Po
means that
P - P0
and
=>
f (P)
::::,;
L.
To make these ideas precise, we interpret P ::::,; P0 to mean that llP - P0ll is small,
L I is small. This gives the following
and we interpretf(P)::::,; L to mean that lf(P)
definition :
-
Then
lf(P)
=>
lim f(P)
L I < E.
L.
P-+Po
R,
lim f(P)
f(Po),
P-+Po
then/ is continuous at P0
The elementary theory of limits can now be generalized very easily.
628
14.3
Theorem 1.
D-+R.If
lim f(P)
P-+Po
then
lim g( P) = L',
and
P-+Po
P-+Po
and
P-+Po
If L' 0, then
f (P)
= !::__
P-+Po g(P)
L'
lim
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:
Let D be a region in R2, let/be a function D---+ R, and let P0
be a point of R2 Suppose that for every E > 0 there is a o > 0 such that
Theorem 2.
Then
Ix - x01 < o
and
IY - Yol
< o
=>
lf(x,y) - LI
<
(x0, y0)
E.
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
The inequalities Ix - x01 < o, IY- - Yol < o hold inside the square, and the
equality llP - P0ll < o holds inside the inscribed circle. Therefore
llP - Poll
and so
< o
=>
Ix - x01 < o
=>
lf(x,y) - LI
and
< E;
lim f(P) = L.
P-+Po
IY - Yol
< o
14.3
629
From this it follows, as for functions of one variable, that for functions of two
variables, continuity is preserved under composition of functions:
3. Let f, g, and h be continuous functions of two variables.
(x, y), let
<f>(x, y) f[g(x, y), h(x, y)].
Theorem
For each
(1)
=>
x x0
=>
(2)
Y Yo
and
and
(3)
(4)
P
(x, y) and P0
(x0, Yo).
Any circle llP - P01\ < o lies in its circumscribed square, as in the figure above.
g and h are continuous.
f is continuous, and Theorem 2 holds.
=
Let </> be a function of one variable; and for each x and y (in a domain D)
let
f(x, y)
</>(x).
)-.
/
__
,------k:___j,/
x
Xol < O
=>
630
14.3
Therefore
Ix - x0\
IY - Yo\
and
< o
< o
=>
<=>
Therefore, by Theorem
2,f is
<
E.
continuous.
These theorems enable us to infer that various simple functions, which obviously
ought to be continuous, really are. Consider, for example,
f(x, y)
cos
x2y2
xa y
1 + x 2 + y2
are continuous.
again).
Therefore
By Theorem
y0,and
D in R2,
and a function
D-+R,
y0
(x,y0) is
ef>v0(x)
Such a function is called a
x0,
cf>v.:
for each
x for
which
slice function
f(x, Yo)
14.3
Geometrically,
(xo, Yo).
f,,(x0, y0)
631
Naturally, we can restate this in purely analytic terms, without any reference to
Definition
1.
1m
(
f"'Xo,
Yo) -
x-+xo
example, given
we sety
= y0 ,
f(x,
to get
y)
= x3 +
For
This gives
cp0(x) = 3x2 +
yg + 2xy0
= f,,(x, Y0 ).
f,,(x,
y)
= 3x2 + y2 +
2xy.
as a constant, regard
x = x0
The derivative of the slice function is now the partial derivative off with respect toy.
More precisely:
Definition
1.
im
v-+'llo
f(xo,
y)
- f(Xo, Yo)
Y - Yo
'
632
14.3
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,
Here
y)
x3
f.,(x, y)
fv(x, y)
xy2
x2y
y4.
3x2
y2
2xy,
2xy
x2
4y3
f,,,,(x, y)
Similarly
h v(x, y)
6x
2x
2y.
+
+
l2y2.
Now !rev is the partial derivative off., with respect toy. We have
fxv
And similarly
hx
Dv(3x2
y2
2xy)
D,,(2xy
x2
4y3)
2y
2x.
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.
of
ox
o2f
ox2
02!
of
oy
o2f
oy2
for f,,,
oy ox
for
fxx
for
fxv
o2f
ox oy
for
f11,
for
fvv
for
fvx
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
oy ox
02/
ox oy
and
oY
'.l '.l
uxuyuy
'.l
f vx
v
oY
'.l '.l
0yuyux
and
fxyy
Note that in the CJ-notation, the symbols for higher derivatives look like "prod
ucts" of "factors" of the types
fm
of ox, o/oy.
Thus
0 a
f (x y)
oy oy ox
,
a
o3f
oy oy ox
This is why the symbols accumulate on the left instead of the right.
14.3
PROBLEM SET
633
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)
4. f(x,y)
vx2 + y2
xy
x 2 + y2
y2
x2
2
x + y2
_
6. f(x,y)
8.
f(x,y)
10. f(x,y)
sin x
=
(y
sin xy
-2x
+ y2
2. f(x,y)
Vx4 + y4 + 1
5. f(x, y)
7.
f(x,y)
0)
9.
f(x,y)
11. f(x,y)
3. f(x,y)
-v'x2 + y2
xy
2
x + y2 + 1
cosy - 1
cosy
2
x-2-+ y
(x 0)
[(x,y) (0, 0)]
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,
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
( ?)
lim
(x,vJ-<o,OJ
xy
--2 + y2
X
f(x, y)
Show that if f is a polynomial, then fxy
.2 .2 a;;xiy i.
i=O i=O
fvx
and g are regular, then so also is f/g, at every point (x,y) where
f(x,y)
t::.f
.
f(x,y)
x2y + y2x,
634
14.4
f,,(x0,y0) 11x
+ fy(x0,y0)
11y
E(11x,11y) 11x
14.4
3,
Jim
F(11x,11y)
(Ll.x,Ll.ul---+(O,O)
0.
> 0
there is a
c5 > 0
Yo
XQ
QI_
t(x, y)
Given a function
f:
D---+ R
= Xo + t COS
y =Yo+ t sin
CJ.,
a.
a,
Yo+
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,
J+a(Xo,
Yo)
1.
lffi
t-o
14.4
For
ex
635
= 0, the slice is parallel to the x-axis, and so it ought to be true that /0 =fz.
ex = 0 we have cos ex = 1, sin ex = 0, and
Jo
+ (Xo,Yo) = 1im
t-+O
= lim
lix-+O
Similarly, for
ex = 7r/2
=f.,(Xo,Yo)
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
Since
for each
fx(x,y) =A,
fv(x,y) = B ,
x and y , it follows that these equations hold at the particular point (x0, y0);
If
for each
ex,
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
f = f'(x0) x+ E(x) x,
where
lim E(x) = 0.
t.x-o
14.4
636
It is fairly easy to find out what form the formula has to take if it exists at all. If f is
linear, with
then
!J.f=Ax+ By+ C
( Ax0+ By0+ C)
A=f,,(x,y),
B=fu (x,y);
and
y.
(/J.x,/J..y)
_,..
for a linear function, the partial derivatives are simply the coefficients of
This suggests that our expression for
/J..f ought
[- - -]
f(x,y)=x2+ xy + y2.
Here
where
We now attack the general problem, for functions with continuous partial
derivatives
fx,fv.
(x0,Yo)
to
by moving first
----
-.- --- -- -- 1
+
Yo
___
__.
I
I
II
I
I
I
--+-x-1--x0
x0+Ax
cp(x)=f(x,Yo+ /J..y),
on the interval from
x0
to
x0+ /J..x.
14.4
637
x0 x!
x0+t.x
z= q,(x)
+Lix,
that
f,Jx,Yo + Liy)Lix.
+----''-----'----'-y
Y
Yo+t.y
Yo
z=Y,(y)
7P(Y)
on the interval from y 0 toYo
+ Liy.
f(xo, y),
Thus
+ Liy;
7JJ'( ji)Liy,
and so
Lif
+ 6.x,
For each
E1(Lix, 6.y)
E2(Lix, Liy)
6.x, Liy,
let
fu(xo,Yo).
+ Liy.
such
638
14.4
Then
and
Therefore
b.J
Note that
lim
E1(b.x, b.y)
lim
E2(b.x, b.y)
(4x,4y)-+ (0,0)
(4x,4y)-+(0,0)
fy are
because fx and
then
b.f
0,
0,
For functions of two or more variables, the idea of approximation by linear functions
is used as the
definition of differentiability.
To be exact:
--+
R,
and let P 0
(x 0,y0)
L(b.x, b.y)
E1 and E2,
and functions
b.f
and
lim
is said to be
b.x
+ B
b.y,
defined in a neighborhood of
(4x,4y)-+ (0,0)
Then
+ B
E1(b.x, b.y)
differentiable
at
E1(x, y)x
lim
(4x,4y)-+(0,0)
(0, 0),
such that
E2(x, b.y),
E2(b.x, b.y)
0.
(1)
(2)
(x0, y0).
This definition was modeled on the preceding discussion, and so we have already
proved the following theorem.
(x 0, y0),
then/is differentiable at
(x0,y0),
with
b.J R:!
b.x
+ B
b.y
+ fu(xo, Yo) b. y.
For functions of one variable, we defined the differential to be the linear function
df
df(b.x)
b.f
J'(x0) b.x,
f (x0
b.x) - f (x0).
For functions of
is differentiable at
df
(x0, y0) ,
df (x, b.y)
then
+ fu(xo, Yo)
b.y.
is differentiable at
f,,,(xo, Yo)
(x 0, y 0) ,
rx
+ fv(x0,
rx,
14.4
Proof
639
By definition,
-r
Ja(Xo,
Yo)
1.
- 1m
.
l 1m
where
Ax
Since
t cos ix,
Ay
1.
Af
1m-,
t-+O
t sin ix.
is differentiable,
Af
= A
=
=
Ax +
11y + 1 Ax + E2 11y
Therefore
+ Ei(t cos ix, t sin ix) cos ix + E2(t cos ix, t sin) sin
and so
faCxo,Yo) =
ix,
!1f
t-+0 t
l im
cos ix + 0
sin
ix,
2
f(x ,y) = x y
(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.
=
11/ in the
14.4
640
Note that other choices of E1 and E2 would have worked just as well. For example,
fif
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,
3.
f(x, y) =xy2,
(1,1)
4. f(x, y) =x3,
5.
(1,1)
6. f(x, y) =y4,
7.
f(x, y) =x2
( -1, 1)
8.
9.
y2,
f(x, y) =4x2
(-1, -1)
(0,0)
(
+
y2,
1 1)
,
(1, 1)
(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)
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?
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.
D, then
(a < x < b,
f(x, y)
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
c
----
r -----
i
I
I
J
----+------
I
I
I
I
I
I
I
I
a
14.5
641
17. Theorem A, stated in Problem 16, is artificially special; it does not apply, as it stands,
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
In the preceding section, we defined the directional derivative fa as the derivative off
along a linear path
X
y
g(t)
h(t)
Xo
t COS IX,
Yo
t sin IX,
f a(Xo, Yo)
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
g(t),
h(t),
a<
t<
and
b.
as the names of the co
x(t),
y(t),
a<
t<
b.
642
Let
F: D---+
D.
path Plies in
14.5
cf>(t)
F(x(t),y(t)),
F(x(t),y(t)).
If F,
are
differentiable, then
cf>(t)
F.,(x(t),y(t))x'(t)
Fu(x(t),y(t))y'(t).
F(x,y)
x(t)
y(t)
x2y
y2x,
cost,
sin
t.
cf>(t)
cf>'(t)
Since
cos2
t sin t
-2 cost sin2 t
F'"(x,y)
2xy
t cos t,
+ sin2
+ cos3t +
y2
2 sin t cos2 t -
F u(x,y)
and
x2
sin3 t.
2xy,
Theorem 3 gives us
cf>'(t)
(2 cost sin t
+ sin2t)(-sin
Xo
Lix
Liy
x(to),
t)
+ (cos2
YUo),
x(t0
Lit) - x(t0),
y(t0
Lit) - YUo)
In this notation,
"-'
't' (t0)
1.
Im
Mo
F(x0
F(x0
cf>.)
=
Lix,Yo
Now
LiF
F,ix0, y0) Lix
where
as
Therefore
Liy) - F(x0,y0)
Lit
E1 Lix
E Liy,
2
14.5
643
Therefore
cp'(t0)
lim
M-+O
F
t
cp'(t)
F,,,(x(t),y(t))x'(t)
x'(t0)
+ 0
y'(t0)
Flx(t),y(t)),
d<fo
dt
oFdx
oxdt
oFdy
.
o ydt
(1)
the change in
and
(2)
the change
quantities
oF ox
ox o t
'
oFdy
o ydt
The formula says that to combine these two effects, we simply add them.
PROBLEM SET
14.5
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)).
8. Same question, for F(x,y) =x/y. ( This will give you an equally circuitous derivation
D[gh]
hgh-1g'
(gh ln g )h
'.
14.6
644
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
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,
(xi. x2,
f(P) - f(Po)
xn) of D, let
let
and let
l:,. f
L(/:,.P)
and a set of
!:,.f
L(l:,.xi. l:,.x2,
l:,.xn)
A1 /:,.x1 + A2 l:,.x2 +
+ An l:,.xn,
+ En(l:,.P) l:,.xn,
(1)
and
lim E;(l:,.P)
!J.P->O
(2)
14.6
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
each i, and
f is
differentiable at
is defined for
The proof is just the same as for two variables: we take a fixed integer k, and set
tlx; =
b.P = b.xk.
Then
tlf = L(b.P)
Ek(b.P) tlxk,
L(b.P)= Ak b.xk.
Since
b.P = b.xk,
Therefore
and
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
by the case
by
P = (w,x,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
three of its first partial derivatives in D, and these are continuous at the point
then
f is
Proof
differentiable at
P0
all
P0 ,
Let
By a slight extension of the device that we used in the proof of the same theorem for
two variables, we write
14.6
646
to the function
<f>(w)
Then
where
f(w,x, y).
</>(w) - <f>(w0)
w
w0
is between
and
w.
Since
</>'(w)
f(w,x, y) - f(w0,x, y)
</>'(w) !::i.w,
fw(w,x, y), we
=
have
!::i.J
(1)
Let
E1(!::i.P)
E2(!::i.P)
E3(!::i.P)
Then
!::i.J
Theorem 3
coordinate functions
Let f be a function
Iff and the three coordinate functions are differentiable, then</> is differentiable, and
</>'(t)
That is,
</>(t)
for every
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)
y are functions
D be a
w, x,
R3,
domain in
cp: D'-+R,
defined by the formula
</>(t,u,v)
14.6
647
new
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 aw
o<f>
ai
aw ai
ax
of oy
+
.
ax at
oy at
of
w, x, and y
Let
v.
o<P
au
aw
of
ov
--
aw au
of
ax
+ -- +
ax au
of oy
-
oy au
of ax
of oy
aw
+
+
.
ow av
OX ov
oy OV
of
w2+ x2+ y 2,
f(w, x,y)
t+ 2u+ 3v,
x
2t+ 3u+ 4v,
y = 3t+ 4u+ Sv.
=
Here
of
-
of
OW
OX
2w
2x
2(2t+ 3u + 4v),
2(3t+ 4u + Sv),
of
- = 2y
oy
ow
2
OU - ,
2(t+ 2u + 3v),
OX
OU
3,
oy
OU
4.
648
14.7
o<f>
OU
<f>(t, u, v)
so that
o<f>
<f>11(t, u, v)
- =
OU
as before.
PROBLEM SET
1. Given/(t,u)
14.6
t 2u,g(t,u)
t +u2,and <f>(t,u)
and
f(t,u,v)
t cos u cos v, g(t,u,v)
t sin u cos v, h(t,u,v)
/2 + g2 - h,find </>t, </>,., and <f>v
<f>(t,u,v)
8. Given
<f>v
7. Given f(t,u)
s +t +u +v,and </>(s,t,u, v)
<f>u-
f+
/2 +
t sin v, and
9. G iven f(s,t,u,v,w)
(1, 0, 0,
1,
0).
as in the definition
10.
14.7
1 ).
D-+ R,
Given a region
f,a
D in Rn, and a
D, and let V be any vector in
will be called a direction in Rn.) The derivative
1.
lID
t-+O
defined to be
14.7
649
+ cit
and
fv(Po) = rp'(O).
We can now calculate rp' (0) by the chain rule for paths:
Thus we have
Theorem 1.
point of
D,
(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
R, with real numbers as its values,
grad/ Thus ifjis a differentiable function D
Rn, with
then grad f is a vector-valued function D
__.,.
__.,.
where the f;,'s are the first partial derivatives off That is,
gradf(P ) = (fx1(P),fx2(P), ... , fx"(P)),
for each Pin
then
and
D.
For example, if
which is a vector in
650
14.7
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
.
. .
fv =/1C1 + f2C2 +
+ fncn,
where the fi's are the first partial derivatives off, and so
fv = (grad f)
V.
. . f n(P)) ,
1 G
V= -llGI! '
so that
where
f;(P)
C = -llGll .
i
Then
fv(P ) = (gradf(P) ) V= G V
=- G
(_Q_)
= _ (G
llGll
llGll
l
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 ).
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
651
x2
gradf(x, y)
y2,
(2x , 2y),
we can indicate the vector field grad f by drawing sample vectors in the xy-plane,
like this:
y
-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
1.
4. f(x, y)
f(x,y)
7.
9.
f(x,y)
11.
f(x,y)
14.8
y2
x2 - y2
y2
x_
2__
1
4
4
5.
f(x,y)
8.
f(x,y)
10.
f(x,y)
12.
f(x,y)
xy
Vl - x2 - y2
6.
f(x, y)
13.
f(x,y)
x
x3
__
(x2
y2)2
x2
y
1
x2
y2
+ 1
4y2 - x2
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
652
14.8
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
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)
f'(xo)
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,
for
x0 - o < x < x0
f'(x) < 0
for
x0 < x < x0 + o.
and
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.
f"(x) < 0
x0
x0 +
o,
> 0, the
o)
{ P
Po P <
o}
{(x, y) (x -
0, and
o,
x0 +
o).
o.
o-neighborhood of
Ifj'(x0)
x < x0 + O,
O <
thenf(x0) is the maximum value off on the interval (x0 xy-plane. For each
o).
o).
Xo 2
Thus
+ (y - Yo)2 < 02}.
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)
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
then N(P0,
o) lies
OP0,
in D.
y
This works, no matter how close P0 may be to the circle, as long as P0 isn't actually
the circle; no matter how small the positive number 1
on
14.8
654
as our positive
o.
of D.
o,
the neighborhood
. ---,
__,., --y
-...
____
..._
f:
D---+ R,
such that
( ) fP
( o)
fP
o),
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
/(xo + t, Yo),
f(xo, Yo + t).
(At this point you may want to review the definition of slice functions, at the
beginning of Section
14.3.)
14.8
Level Curves
655
f(x,y)
we have
fx(O,0)
/
(0, 0)
x2 - y2,
fv(O,0)
0,
</>0(t)
has a minimum at
0,
f(t, 0)
t2
</>"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.
f(x, y)
x2
+ y2,
the level curves are circles with center at the origin, as shown above.
k >
0,
For each
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
For the function
f(x,y)
0.
x2 - y2,
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,,
14.8
656
y=x
y=
Fork >
0,
y= -x
-x
origin,where f has a saddle point. These examples are typical of the way level curves
behave in simple cases.
x-
point (x0, y0), we may still have a saddle point, on which a man in the saddle would
be facing in some third direction.
f(x,y)
Consider
-xy
- tx2 - tf.
Here
c/>o(t) = -!t2,
so that </;0 has an ILMax at
But for
rx = 37T/4 we
0;
</>rr12(f) = -!f2
have
cos
rx =
.J2. ,
sm
rx = .J2. ,
t2
42
0.
t2 -t2
1 2
=-t
,
4
2
4
t2
42
14.8
Level Curves
657
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) .
cp(t). =fx(x0
t cos rx, Yo
+ t
rx
+ fv(x0 + t
cos rx, Yo
here we are using the chain rule. Applying the chain rule again, to each term, we get
fxx(x0
t cos rx, Yo
fxv(x
fvxCx0
+ fvv(x0
+
+
t cos rx, Yo
t cos rx, Yo
cos rx, Yo
cp(t) =
= c
2fx:.
+ t
+ 2 csf xv +
2jYY'
=cos rx,
!xx =fxx(Xo
=sin rx,
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.
< 0
> 0
<
a we have
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).
658
14.8
csf
s2 r
+
l
fxx
frxx J
s;Y
s2fvv
2
fxv
]
+
=fxx [( c + s . )
fxx
fxx
f xx
fxu 2
fxxfvv - f;v
.
s2]
)
+
=fxx [( c +
fxx
J2xx
r
r/>;(t) =fxx C2 + 2
--.:!.Y!!.
(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
for some c5 > 0. Thus for !ti < c5 we have
If we also know that
r/>(t) < 0
for every
(x0, y0),
o:.
then we have
rf>(O) = 0
for every
o:.
(x0, y0).
following theorem:
Theorem 3. Suppose
hood of (x0, y0). If
0,
(I)
(2)
(3)
f.(xo, Yo)
fv(xo, Yo)
and
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.
Theorem
hood of
4.
3:
(x0, y0).
If
(1)
(2)
14.8
Level Curves
659
and
(3)
Proof
If
of Theorem
at
(xo, Yo).
3.
Therefore
-f
has an ILMax at
4,
then
-f
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.
f(x, y)
For
(0, 0),
0.
</>;(t)
We sett
0.
Let
Then
</>:(o)
A cos2 IX + 2B sin
cos
IX
IX
+ C sin2 IX,
B2
AC> 0.
+ 2Bu + Cu2
(2B)2 - 4AC
and so 1P(u1)
1P(u2) <
cos2 1X[A +
0 for
0 we have
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,
for tan
at P0
If f ;v
f xxfvv > 0
at
P0,
then
14.8
660
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.
2. f (x,y) =x2 + xy
1. f (x,y) = xy
3.
f(x,y) = x2 - y2
5.
f(x,y) =x2 + y2 + x + y +
7.
f(x,y) = x2 + xy + y2 + x + y + 1
4.
6.
f (x,y) =x2 + y2 + 2x +
f (x,y) =x2 + 2xy + y2
8.
9. f(x,y) = 1 - x4 - y2
11. At what point does the function f (x,y) =x2 ( 1 - x2 - y2) take on its maximum
take on its minimum value? What is the minimum value of the function?
Consider the ellipsoid
x2 + y2/4 + z2/9 = 1 .
15.
z2
y2
+ -;;=l.
4
16. Let A1 = (0, 0), A = (1, 2), and A3 = (2, 1). For each P
2
f(P) = (A1P)2 + (A2P)2
(A3P)2
. .
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
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:
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.
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
'1
1......
--
'
----
./
x
a
x
662
14.9
The expression
fff(P)
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)
0 z f(x,y)}.
and
ED
vS
A(x) dx.
This method works for many solids whose volumes are not given by standard
formulas. Consider the following.
y
y=vx.
{(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)
A(x) dx
t)-;; )
x dx
tx.
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
14.9
integral.
663
For each
(x, y)
{(x,y) \
in D, let
f(x, y)
1, 0 y 1
x2
x}.
+ y3.
We want to find the volume of the solid lying above D and below the graph off
Now for each
x0,
A(x0)
-"'0
(x
y3) dy
[xy
!y4]-xo
A(x)
x2 - x3
Hl
x)4
ff!(P) dA fA(x) dx
=
(t - t] - (-io]
x)5]
T\.
xy-plane, lying between the graphs of two functions, as on the right below.
z
y
x2
0
664
We have given
F(x, y) 0
14.9
ff
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)
Here
lgf(x)(x)
F(x, y)dy.
ff
F(x, y)dA
A(x) dx
f [i::
>
g(x).
But
A(x),
This takes a very simple form when D is a rectangular region defined by inequalities of
the form
a x b,
Here
ff
F(x, y)dA
y d.
ff
A(x)
ia
F(x, y)dy,
:]
I
I
I
I
I
I
I
I
I
I
b
Of course, we could equally well have used cross sections for constant
would give a different cross-sectional area function
B(y)
and we would have
ff
F(x, y)dA
F(x,y)dx;
B(y)dy
ff
y.
This
14.9
665
The expressions
iterated integrals.
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.
y
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
1,
2
x ,
(1)
or
(2)
1,
ff
F(x, y) dA
f1 r-vl-Y
F(x, y) dx dy.
=Jo Jo
Therefore the two iterated integrals must have the same value.
PROBLEM SET
14.9
JJF(x,y)dA
D
666
14.9
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;
2. D: 0 x 2,
x2
0:5y:5-
'
-
3. D: 0 x 2,
4. D: 0 x 1,
5. D: 0 x 1,
6. D: 0 y 1,
7. D: -1 x 1,
F(x, y)
F(x,y)
- 4
x3 y 8;
F(x,y)
x2 y x;
F(x,y)
x y 1;
F(x,y)
y x 1;
F(x,y)
0 y 1 - x2;
8. D: 0 x 1,
-Vl
9. D: 0 x 1,
0 y x2;
x + y
x - y
x2 + y
x + y
x3y3
x2 + y2
F(x,y)
x2 y V l
F(x,y)
x2;
xy
F(x,y)
(x2 + y2)2
.Yxy
10. Let
Find / ( 0<:) .
'
/(0<:)
ff
</>(x,y)dxdy.
/(<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
tl
</>xy(X,y) dxdy.
/
____
I
I
I
I
I
I
I
)-_____
//
14.10
7r
.7c--1
- 2
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.
14.10
The unit sphere with center at the origin is the graph of the equation r2 +
z2 =
667
1.
cose,
sine,
z = z,
x2 + y2
,2.
z
'
'
'
I
I
I
l..
I ',
I
I 1/ ,'1I
I/
'
'
IZ
/---- -r---- 2
--7r
7r
(r,0,z)
--2
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
right above.)
Suppose now that we have given a domain D in the plane
z =
z =
0.
The plane
f(P)
f(r, 8).
ff!(P)
dA,
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.
668
14.10
Q
I
I
Given a domain
(1) D
Di
The figure
indicates, at long last, why we use the word net in integration theory.
y
Di
oDi
Note that if
Di is
oDi.
Thus
Di}.
14.10
669
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
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
INl-+Oi=l
L,
!NI < c5
=>
if(P;) fl A; - L <
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.
Thus a closed interval is closed, but an open interval is not; the region
is closed, but the region
{(x, y) I x2 + y 2 1}
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.
670
14.10
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
D = {(r, 6) I a r b,
Cl
e {J}.
N, on the interval
r = ri and the rays 6 = 61 now cut up the domain D into nm little pieces,
like this:
7r
14.10
671
i, j,
the area of
D;1
is
Then
and
Therefore
Llr;2)
sample sum
n
i=l i=l
L=
INl->o
fff(P) d A.
D
i=li=l
+-----+-
oj-1
00=a
- -
-.---<
pij
--
1--+-----1---1
672
14.10
Let D' be the rectangular region shown in the figure. That is,
D'
= {(r' e)
Ia r
b'
ex
{J}.
Iim
JNJ -+O i=l
i=l
D'
and
lim
JNJ -+O
Since IimlNl-o
!1ri =
i=l
i=l
ff1(r, e) drde.
D'
0, we have
D'
D'
D'
Let
D
= {(r, e)
Ia r
b,
ex
(J},
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
Let
D
=f(r, e) = )1
= {(x ,y) I x2 + y2
l}
r2
= {(r, e) I r2
l}.
Now
f J1
14.10
673
Therefore
Therefore
ffJ1
r2 dA
PROBLEM SET
1. Let D
477/3.
14.10
{(x,
y) [ x2
+ y2
l}. Find
JJcx2
y2)712dydx.
2. Let D
{(x,y) I
x2
y2 l}. Find
JJ v
y2dydx.
1 + x2 +
3. Find the volume of the solid which lies under the paraboloid
r =
sin e.
sphere x2 +
y2
z2
4.
ellipsoid
x2
6. Find
y2
4
il Jyl-x
z2
ffv 2
x +
x2 +
y2 and over
y2
y2
1.
y2)10dydx.
_(x2 +
-1 -v1-x2
x2
y2
dydx.
ffvx2
y
+
dydx.
y2
ff
s
x2
xy
+ y2
dydx.
674
10.
11.
Find
x
y
(X2 + 2)3/2 dydx.
Y
Find
14.11
14.10
x2 - Y.
- dydx.
x2 + y2
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
Pn, with masses m1, m 2,
particles Pv P2,
, 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
n
= _L m;Y;.
i=l
Mu=vo
If
Yo
i=l
is
n
L m;(y; - Yo)
i=l
then the point (.X, ji) is called the centroid of the system. By easy calculations we get
n
1
m;
m =
ji =
_L m;Y;
i=l
m ;=1
It is easy to see that if the axes are translated, the centroid is unchanged: for
x = x' + h,
x' = x - h,
y = y' + k,
y' = y
k,
the coordinates of the centroid in the new coordinate system are given by the formulas
x
_,
=
m;=1
_,
i=l
1
x- - h
,,;;;., m; = x- - h,
m i=l
-m1;,,;;;.,=1 m;Yi
,
=
Y- - k,
14.11
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 =
1
-
.,
m i=l
()
x'
cos() -
.!
( i miy)
m i=l
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),
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:
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
Definition.
when <5
0.
676
14.11
Here the lefthand side is the average mass per unit length on the interval
p(x0)
when
where p is continuous:
o-+O o
1
Jim - m (Xo, Xo
+ 0) =
lim
o-+o O
R::>
[x0, x0 + o],
xo+o
Jxo. p(x) dx = p(x0),
by the general formula for the derivative of the integral. We assume hereafter that
is a continuous function. Let us take a net N:
the sum
x0, x1,
, xn
over
[a, b],
and form
,L xip(xi) 6.xi.
i=l
This sum is the moment, about the origin, of a finite system of particles of mass
at the points
0, is f
Thus
b
M0 = a xp(x) dx.
Mk=
is
0.
x=
J! xp(x) dx
.
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.
p(x)
x=
kx,
fkx dx
[tkx2]g = 2k,
"
_!_ [ x kx dx
m
Jo
_!_ k[lx3]
2k
! l
f.
14.11
677
D;.
=ff
m(D;)
pP
( ) dP
D;
for every
D; lying in D.
D; may be all of D;
In particular,
m = m(D)
=ff
p(P)
dA.
over
D;
we take a sample
P1, P2,
D1, D2,
,Pn of N,
Dn
with
Pi=
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
Ai, with x-coordinates x i . As the mesh of the net
Jf
xp(x,
y) dA.
By definition, the moment of the plate about the y-axis is this integral. More generally,
the moment about the line x
=k
Mx=k
is
=fJ
cx
k)p(x, y) dA;
678
14.11
and similarly,
= ff<y
M11=k
k)p(x,
y) dA.
=ffp(x, y) dA,
D
; ffxp(x, y) dA,
ji
; ffYP(x, y) dA.
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
Thus
p(x)
1 for 0
y
2
-+------+--x
2
y=(x)
Note, however, that pis integrable, and that
fp(x) dx =
1 +
3,
14.11
679
3
2
I
I
I
I
I
I
I
and
M0
fxp(x)dx
i + i(2 + 4)
t.
Therefore
x =
_!_
l t
Mo =
t.
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
and t.
Here
Mo =
+ 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
2.
14.11
680
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)
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
say,
1.
f(x)
Find
for 0 x 1 ,
x2
(x -
2) 2
for 1 x
2.
Sf(x)dx.
f(x)
Find
Sf(x)dx.
{; = 2
for 0 x 1 ,
for 1
x 2.
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
lp0
P0
(x0, y0)
is defined to be
(y - y0)2]p(x, y)dA.
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.
x0 is
=ff cx -x0)2p(x,y)dA.
D
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
P,
is defined as follows.
Let
[a, b].
xi
ti.xi
Then
J{P F dx
f(ti),
+ G
dy
X;
Yi g(ti),
L'l.yi = Yi - Yi-1
=
xi-1
Jim
JNJ..,O
i=l
path
P have
If F and
G are continuous,
tF dx
G dy
g of the
682
14.12
Proof
[ ti 1 , t;]
_
i;
such that
</>(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>.
matter. (See Appendix I, where a very similar case is discussed in detail.) Therefore
n
lim
iNi-Oi=l
INl-+O i=l
lb
a
P(t)
(j(t),g(t)) = (t + 1, t 2)
F(x,y)
Then
LF
dx + G dy
x + y,
G(x,y)
x + y2
Line integrals have the following quite natural physical interpretation. We regard
the path
P:
R(x,y)
Line Integrals
14.12
683
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;
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
R:j
i=l
F dx + G dy,
P(t)
R(Q)
f(t)i + g(t)j,
F(Q)i + G(Q)j.
i=l
_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
=
RdP.
exact differential. This is the case in which there is a function <P such that
cf>x = F,
cf>11= G,
14.12
684
Here, by the chain rule for paths, the integrand is the derivative of the function
<I>(t )
It follows that
= 1>(/(t), g(t)).
P(b)
= (j(a), g(a))
of the path; it is independent of the way in which the path proceeds from the initial
point
P(a)
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
Jcc,a>
for
L <f>x dx
<Pv dy
<f>.v dy.
can be used only in the case where the integrand is an exact differential.
PROBLEM SET
14.12
Line Integrals
14.12
1. R(x,y) =x2i
2. R(x,y) = x2 i
y2j;
y2j;
P(t) = it + jt2
(0 t I)
(0 t
1T
(0 t
1T
P(t) = it + jta
(0 t l)
3. R(x,y) = yi
xj;
4. R(x,y) = yi
xj ;
5.
R(x,y) =xyj;
6.
R(x,y) = i y e"'Y
7. R(x,y) = i y exy
P(t) = it3
j xexy;
jxexy;
(j l - ts)
(0 t 2rr)
)
)
(-1 t l)
P(t) = it
(-1 t l)
8.
R(x,y) =xi - yj ,
(0 t 21T)
9.
R(x,y) =xi - yj ,
P(t) = it 2 + jt3
(0 t l)
(0 t 1T/2 )
(4 t 7)
(0 t
2xyj ,
1T
685
The Shorthand of
Logic and Set Theory
Appendix A
In this book, the use of logical symbols is held to a minimum, on the ground that
words are usually easier to read.
points be convenient in the text, and is even more useful in notebooks and on black
boards.
We explained in Chapter
that
<=>
x < 1
And
=>
1 > x.
<=?
x >2
=>
x2 > 4.
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
P(x). Thus the closed interval from 0 to 4 is
[O, 4]
to
{x I 0 x 4},
is
(1, 2)
AcB
means that A is a subset of B; that is, every element of A is also an element of B.
For example, if
(1, 2)
and
[O, 4],
then
Ac B.
We allow the possibility that A
like
::i
A.
Thus c is
Appendix A
688
If
A,
then we write
x EA.
This is read
"x
belongs to
A."
xA
x
means that
A.
The union of
and
is denoted by
A UB.
Thus
A UB={x
The formula
is read
"A
x EA
B."
cup
or
x EB}.
The intersection of
and
is denoted by
AnB.
Thus
AnB={x
The formula
AnBis
read
"A
B."
cap
x EA
and
x EB}.
The difference
A-B
of
B c A.
For example, if
A= [O, 2]
B.
and
A-B,
[1, 4], then
To write
}.
1.
The
Thus
AnB={}
means that
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
x2y2=(xy)(x + y)
means that the equation holds true for every
x.
Similarly,
v.,,11
xsys=(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."
689
y)
=>
3z 3 z E R and x <
< 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.
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
Here we
Theorem 2.
Theorem 3.
x-+xo
Proof Let
"
>
=>
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
we have
and
x I,
If (x) + g(x)I If (x)I + lg()
=>
=>
Proof By Theorem
we have
lim [g(x) - I.:] = 0.
Jim [f(x) - L] = 0,
X--+CCO
690
691
By Theorem 3,
0.
By Theorem 2,
lim
X--tXo
< I x - x0 I < o
If(x)I <
=>
M.
Proof
Let
limf(x)
Let Ebe any positive number, and let
L.
f
L
I
I
I
I
---4---
I
I
I
I
I
I
-+---x
x0-o
x0
Xo+o
L-
--
Thus
0
The
=>
- E<
f(x) < L +
E.
numbers
Theorem 6. If lim,,_,,J(x)
0, and
lim
[f(x)g(x)]
0.
692
Appendix B
Proof
be given. Take
01
> 0 and
Ix - x0I < 01
lg(x)I < M.
(1)
(2)
0 <
=>
Next, using e/ M in place of e in the definiton of a limit, take c52 > 0 such that
Ix - Xol
0 <
0 <
then
(1)
and
(3)
0 <
Ix - x01
< - .
(4)
Ix - x0I
(2)
(3)
If(x)I
=>
02
<
and
< O
(4)
(5)
< O,
=>
g(x)
< M,
f(x)
<
Since
lf(x)g(x)I
lf(x)I lg(x)I,
we have
0 <
Thus, given an
Ix - x0I
< o
=>
lf(x)g(x)I
< !....
M = e.
Ix - x01
< o
=>
[f(x)g(x)]
If (x)g(x)I
=
< e.
0,
:z:-:z:o
Proof
By Theorem
2,
[f(x)g(x)]
L', then
LL'.
[f(x)g(x) - LL']
= O;
f(x)g(x) - LL'
f(x)g(x) - L g(x)
[(j(x) - L)g(x)]
Lg(x) - LL'
[(g(x) - L')L].
approaches 0 as x x0, and
x0
Therefore
lim
[(f(x) - L)g(x)]
:z:-+a::o
= 0,
the second
693
and
lim [(g(x)
L')L]
a::-+xo
0.
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.
Ix
Xol < c5
> 0
=>
=
If(x)I
>
x0
y
-t-----.,___,1 -Xo
1 1
Xo-+0,,-----x
st- -t-i--
-x;;--
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 x0
0
Theorem 8. If l i m.,_,.x0 f(x)
L, and L = 0, then f is locally bounded away from
0 at x0
=
> 0.
limit.
y
Then
Let
Appendix B
694
Thus
the definition: it is a
Theorem 9.
Proof
=>
Then
away from 0 at
the bound M
Theorem 10.
l/e.
If lim.,_.,0 f(x)
=>
lf(x)I
L, and L :;.6
lim -1-
f(x)
There
0.
x-+xo
10.
=>
.!
J i
1
-f(x)
<
.!e .
!_.
L
0.
1
[L - f(x)].
Lf(x)
0.
L, lim.,_.,0
Jim
then
[f(x) - !.]L
Lf(x)
If lim.,_.,0 f(x)
0,
L - f(x)
=
x-+xo
Theorem 11.
-L.
(Look again at
x0;
x-+xof(x)
Now
I f(x)I > E.
1- <
=>
x0
lf(x)I > e.
Iff is locally bounded away from 0 at x0, then l!fis locally bounded at x0
Therefore
Proof
f(x) > E
-f is locally bounded
0.
=>
g(x)
f(x)
g(x)
L
.
I.:
L',
0,
then
Algebraic Operations
with Limits of Sequences
Appendix C
=>
and a number L.
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
Proof
1.
Let
>
0 be given. There is an
n
> N
=>
N such that
Ja11 - LJ <
e.
Therefore
n
>
> N
J(an - L) - OJ <
=>
0 there is such an
Theorem
2.
0.
Ix - x0J < o.
If limn--+oo (an - L)
e.
n-+OO
Theorem
4.
Appendix C
696
Proof
Take any
> N
is bounded.
n
IL
< an < L +
is bounded.
E.
l IL + E l is a bound.)
E ,
la1I, la2 I
aN is bounded.
IL
E ,
IL +
E ,
, laNI
is bounded, then
lim anbn = 0.
n-+oo
n
Then a1, a2,
is
> N
lan l >
0.
E.
n,
then the
is bounded.
!. .
L
n;
n,
then
an
L
lim
=-.
n-+oo bn
L'
-
Theorem 12
bn ;
Appendix D
At the beginning of Section 4.4, we gave numerical examples of the use of the approxi
6.f df,
mation
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.
on the error that results when you use
as follows. We have
6.f
f(xo
df in place of 6.f
6.x) - f(xo),
df
f'(x0) 6.x.
I
I
I
I
I
I
I
I
------ r -- -1
x0+x
XQ
XQ
I
I
I
I
I
I
I
I
I
I
x0+x
Applying the mean-value theorem (MVT) to the function f, on the interval from
to
x0
6.x,
we conclude that
f(xo
for some
follows.
x between x0 and x0
- f(xo)
6.x.
(For
x0
<
j'(x),
x0
6.x
> 0.
6.f
!if - df
The case
6.x
x0
'
f (x) 6.x - f'(x0) 6.x
697
x0
x
=
<
x0
!ix,
Appendix D
698
[x0, x]
f'(x - f'(xo)
for some x' between
x0
and x. Therefore
f'(x) - f'(x0)
and
b,.f - df
Now
x0
j"(x'),
X - X0
Ix - Xol 16.xl.
b,.x.
Therefore
and x0 +
f"(x')(x - x0),
x'
where
is between
between x0 and x0 +
6.x.
If" (x)I,
for
may be.
f(x)
sin x, then
j"(x)
-sin x, and
lf"(x)I 1,
no matter
Before giving further examples, let us write down the theorem that
we have proved:
Theorem 1.
Suppose that
f has a second
b,.x. Then
derivative
f",
and that
lf"(x)I M,
for
f(x)
..J,
Xo
25,
b,.x
4.4.
Here we had
0.4.
Now
f(x)
f" (x)
x1f2,
=
f'(x)
-:lx-a/2
tx-1/2,
__=..!.._ .
4R
x3,
4)x3.
[25, 25.4].
As
x increases,
!::if R> df
699
4.J253
0.039841,
-1
500
lf - dfl sh x2
In fact, we had
f
1
=
df
0.04,
-sh(0.4)2
0.00032.
lf- dfl
0.000159.
The Continuity of
Composite Functions
Appendix E
In Section 4.5 we stated the following theorem, with only rough indications of proof.
Theorem.
x-xo
and
lim f(u)
g(x0)
u0,
(1)
f(u0),
(2)
f(g(xo )).
(3)
then
limf(g(x))
Proof Let
Now take o1 as
E,
=>
E.
=>
such that
=>
=>
E,
(?). If lim.,__,.,0g(x)
L,
then lim.,_..,J(g(x))
for x ;tf.
for x
1,
1.
I
I
I
I
2
700
L.
lim g(x)
u0
1.
Then
lim g(x)
u0,
and
lim f(u)
but it does not follow that
In fact,
x :;z6 1
=>
Iimf(u)
1,
u-+1
(?)
and so
u-uo
limf(g(x))
1.
xxo
g(x)
limf(g(x))
x-+1
=>
f(g(x))
2.
2,
701
Appendix F
The results of our calculations in Section 4.8 and in later sections suggest two questions:
1)
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?
In the theorem below, f(4) denotes
j<ll is f', j<2l = f", j<3l = DJ" = J'", and J<4l =
Dj<3l.
As usual,
Y1 = f(O),
Yo= f(-k),
Y2 = f(k).
[-k, k],
(-k <
x<
E(k)
where xis some number between
-k
:j<4l(.X),
and
k
l(x) dx - - (Yo + 4Yi + Y2)
3
k.
That is
0
=
90
1<4l(.X)
k).
then
JE(k)J 9k5M.
The latter is the statement which is most convenient to apply.
the proof of the theorem, let us look at an application of it.
Example.
-6x-4,
702
Before proceeding to
703
We want
for the fifth decimal place in our approximation to be correct. Thus we want to take
k such that
or
ks < H . 5 . 10-s.
ks<\"-. 10-s,
which surely holds if
k = 0.05.
Therefore
F be
F' =f
(How do we know that there is such a function?) Then
so that
E(k)
k
f(x) dx - - (Yo
4y1
Y2)
k
ik
4f(O)
= F(k) - F(-k) - [f(-k)
+
f(k)].
Therefore
E'(k) = f(k)
= if(k)
f(-k) - [-j'(-k)
3
f'(k)] - Hf(-k)
j'(-k)
3
4f(O)
f(k)]
j'(k),
3
= - [f"'(k) - f'"(-k)].
3
Appendix F
704
We need all of these formulas, not just the last. It is easy to check that
E(O)
From now on,
E'(O)
E"(O)
k is going to be regarded as
G(t)
G(O)
and
G'(O)
E(t) -
k5
0.
t5
0,
G111(0)
0.
On the interval
E(k)
G(k)
G"(O)
E'"(O)
This gives
0<X1<k.
G"(x2)
Applying MVT to
0<X2<X1.
0,
0,
By a straightforward calculation,
G"'(t)
Setting
E"'(t) -
E(k)
k5
(60t2)
E(k)
k5
=
180
k5 f'"(xa) - f"'(-xa)
X3
E(k) 2
t
k5
90
2X3
such that the second fraction on the right is equal to j<4>(x). This gives
E(k)
which was to be proved.
k
s 1<4>(x)
90
(-k<x<k),
Appendix G
The Idea of a
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:
, such that
limAn =
n-+co
L.
(Here Rn was the union of the outer rectangles, An was (h3/3)(1 + l / n) ( l + 2/n),
and Lwas h3/3.)
zfi (i
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.
i,
Appendix G
706
A;
A;
is
Ai
The sequences A1, A2,
and
ff(x) dx A;.
A, A, ... have
regions can be represented in many different ways as the regions between the graphs
of two continuous functions.
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
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
t-+oo
and
lim
t->CO
then
lim
t-cc
(t)
g
oo,
t
g'( ) = L,
(2)
f'(t)
( t)
j t)
(
lim g
tco
(1)
L.
< g'
(t)
f'(t)
< L +
'
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)
because F'
lim G'(
t->oo F'(t)
=
g'.
L'
G(t)
L'
t)
lim g(
L.
lim
t-> co
t->oo
F(t)
J( t )
707
Appendix H
708
g(t)
f(t)
G(t)
g(t1)
F(t)
f(t1)
G(t)/F(t)
g(t1)/F(t)
+ f(t1)/F(t)
+
Therefore it will be sufficient to prove the theorem in the following special form.
Theorem A.
lim G(t)
t-+ 00
lim
t-+oo
oo
), such that
(3)
oo,
t-+ 00
G'(t)
L'
F'(t )
F'(t)
(4)
(5)
fort t 1,
G'/F' is bounded,
(6)
G(t 1)
(7)
and
F(t 1)
0.
Then
lim
t-+ oo
G(t)
F(t)
L.
(8)
y
y=<J>(x)
x=F(t)
- G1(t)
. m-</> (x) F'(t)
-
Evidently
G(t)
F(t)
<f(x)
x
(x
F(t)).
interval [O,
709
oo
(0)
0,
(9)
(10)
<P' is bounded,
lim (x)
oo,
(11)
L.
(12)
L.
(13)
X-+<Xl
and
lim </J'(x)
x-+oo
Then
lim </J
X-+ co
(x)
X-><Xl
L.
- X
Proof By the mean-value theorem (MVT), for each x there is an x, between x and x 2,
such that
</>(x2 - </>(x)
x - x
As x -+
oo,
x-+
oo,
and so <f>'(x)-+
L.
<f>'(x).
Step 2.
lim
X-+<Xl
- X
0.
Proof
<f>(x2) - <f>(x)
x2 - x
<f>(x2)
x2
---=
__l _ <f>(x)
x
1 x
-
</>(x2)
x2
<f>(x2)
710
Appendix H
</>'(.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
_
Step
1_ </> ( x)
[
x
- 1
</>(x2)
x2
-+
O.
3.
lim </>
X-+CO
by Steps I and
to be proved.
2.
From Step
(x2)
X2
L,
-+
L,
which was
Proof of the
Formula for Path Length
Appendix I
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
s =
INl-+Oi=l
lim L P;_1P;;
we know that
i=l
where
and
+
t;;
and
Ja
g'(t)2 dt,
['f
g'(fD2 ti
i=l
For every
Proof 1)
Let M be
2)
c:
Sincef' and
such that
The function
z f-4
g'
are continuous on
f' (t)2
.J;
g'(t)2 M
[a, b],
for
a t b.
[O, M]
711
g'2
[O, M].
Therefore this
4).
Therefore,
712
Appendix I
given any
iz - z'I < ()
3)
The function
=>
It: - t/ < o
=>
O =>
=>
=>
=>
=>
(for every
i)
(for every i)
I
I[
I[
J l b a Llti
g'(fi)2] Llt;I
< -- L Llti
b - ai1
E
-- (b - a)
E
b - a
= e.
Since the absolute value of the sum is less than or equal to the sum of the absolute
values, it follows that
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
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.
multiply. We know that in P(x), these operations obey the CAD laws, that is, they
are commutative, associative, and distributive:
pq = qp'
p(qr) = (pq)r,
p +q
q + p'
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)
or briefly p
q. Thus
p(x)
if
q(x)
p(x) - q(x)
r(x)(l + x2),
Theorem 1.
If p
p' and q
q' , then p + q
p' = p + r
(x2 + l),
we get
p'q'.
714
Appendix J
p + q. And
p'q' = pq + (ps +qr)(x2 + 1) + rs(x2 + 1)2
= pq + ( ps +qr + rsx2 + rs)(x2 + 1),
p= {p'
IP' =p}.
p . In
p +q = p +q,
p . q = pq.
and
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.
Proof We know that these laws hold in P( x). Therefore, under our definitions of
addition and multiplication in C, we have
For example,
x - 5x2
-7x + Sx - 6
x + 6x2 - 3
= -2x - 9.
In fact, the system C that we have just defined has all the properties of the number
system that we wanted.
congruence class p(x) by the formal expressionp(i), in which xis replaced by i. Thus,
715
i - 5i2
i+ 6i2 -
-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
then
for each
0. z
z there
=a+ bi,
and
0,
=z;
is a number
-z
=-a - bi,
such that
z
Finally, every
z -:;tf:. 0 has
(-z)
=0.
a + bi =0
=>
a =b =0.
a + bi =0
<=>
a + bx
<=>
a+ bx = r(x)(l + x2).
0 mod 1 + x2
2.
Since
=>
a =b =0,
and so
a + bi -:;tf:. 0
=>
and
a - bi -:;tf:. 0,
a2+ b2 > 0.
fractions with complex denominators make sense, and then checking that our answer
works:
1
1
a - bi
a
=
--- =--- .
a + bi a - bi
a 2 + b2
a + bi
---
---
bi
a 2 + b2
716
Appendix J
(a + bi)
[a2 a b2 - a2 bi b2]
+
2
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
a - 0 have reciprocals; but every congruence class p(i)
polynomials p(x)
{p(x) I p(x)
a + bx}.
Iterated Limits.
APPENDIX K
f(x, y),
(x,y)-+(xo.Yo>
and
a::-+xo 11-+110
'Y-+Yo X-+Xo
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
f(x, y)
Here
2
2
x
Y
2 + 2
x
y
_
.
. x 2 - y2
bm hm
2
2
x-+O 11-+0 X + y
and
lim Jim
y-+0 x-+O
2
.
x
hm 2 = 1,
x-+O
- 2
2
y
y
=
lim -- = -1.
2 + 2
2
X
y
y-+0 y
2 -
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.
lim
L,
(x,v)-+(xo.vo>
then
lim limf(x, y)
Proof
lim limf(x, y)
L.
Iff(x0, y0) is defined at all, thenf(x0, y0) must be L. Iff(x0, y0) is not defined,
718
Appendix K
x,
we have
(1)
lim limf(x, y)
(2)
(3)
Similarly,
Jim Jim f(x,
y)
Y-+Yo
11-+Yo X-+Xo
In
(1) ,
all we are saying is that ifjis continuous (as a function of two variables)
(1)
x,
in the form
Iimf(a, y)
11-+Yo
which reminds us that xis fixed as y
---+
f(a, Yo),
(1')
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
fxY(Xo, Yo) - im
_
fxCxo, Yo
6.y)
Ay->O
fx(Xo, Yo)
6.y
1
f(x o
.
Im
=Im
I
x
Ay -> 06.y A ->O
[i
6.x, Yo
6.y)
f(x o, Yo
6.x
6.y)
J im Jim
1-
{[f(xo
Ay-+OAx->06.y 6.x
- [f(xo, Yo
6.y)
6.x, Yo
6.y)
f(xo
6.x, Yo)]
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),
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-
Ax->OAv->06.y 6.x
F(6.x, 6.y).
Iterated Limits.
F.
719
Applying the mean-value theorem MVT to the function <fa, on the interval from
x0 to x0 + x, we get
F(x,y)
where xis between
ip(y) =f.,(x,y);
we have
By MVT,
where
(x0,y0) as
1
F(x, y)
A
A
Ll
Ll
X
y
t.yo t.x-o
(by definition).
2)
Jim
(ilx,t.y)(o,o)
x y
--
F(x, y)
exists, and is equal to fxy(x0,y0). This is what we have just proved. Suppose now
lim lim
1
F(x,y).
fy.,(Xo, Yo) =6.x->O
6.y->OX y
3)
Therefore fxy(x0,y0)
Theorem 2.
D, thenfxv = f11x-
720
Appendix K
Proof
Therefore
(hua)v
andfxvxv
(hx11)11,
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),
=
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
Some of the definitions that we have used in Chapter 14, and the hypotheses of some
of our theorems, may seem needlessly strong.
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.
1)
2)
all the slice functions f(x, y0) (with y held constant) are continuous, and
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
.... 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
(0, 0).
f(x, y0)
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
----
-+---....._x
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)
(J2' Jl)
is not continuous. Its graph is shown below. But even if a function f has slice functions
<Pa(t)
a, Yo
f (x0 + t cos
+ t sin
a,
a)
continuous.
z
Example
1)
2)
2.
<Pa
is continuous, but
f is not continuous.
Proof.
approaching the origin as a limit. On each disk, the graph off is a blister of height
as in Example
1;
0.
1,
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
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
f (x0 +
y
I
Yo
-- - - -
tJ.,
-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
The following plausible-looking variation on Theorem A is false.
tJ..
Appendix L
724
</>it)
If
(1)
each function
rp,,
f(x0 +
t cos ex, Yo
ex,
let
t sin ex) .
(1)
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)
is equal to 0 in a
rf>a(t)
0 for
!ti
neighborhood of 0.
f(t cos
ex,
t sin ex)
ex
< o". Therefore every <Pa has an ILMax at 0. But obviously f does
ex
there is no
inf {o"}
0. If you reread the proof of Theorem 3, Section 14.8, you will see how
ex.
If
ex
> 0 and
ex !::::;
0, then o"
!::::;
O; and so
this trouble was avoided: using the continuity of fxv fxx, and fvv, we found a single
o > 0 which worked for every
ex .
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.
APPENDIX M
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
in a Cartesian space Rn
Theorem 1.
a ;;a
;;a b
R, where D is a domain
and c
bounded above.
Proof
---+
(n > 1).
;;a
y ;;a d.
Let
be a continuous function D
---+
R.
Then f is
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],
same way.
as defined in Section 5.6; and they are going to be used in exactly the
'! =
=1----+----1
C---,
I
I
a
:I
di
J,[
Ci
I
a+b
2
-
I
I
b
--
a;
Di
I
I
I
Ii
bi
Following the pattern of Section 5.6, we say that a closed rectangular region D' is
that the giyen D is bad. It follows that one of the quarters of D must be bad. (Why?
See Lemma
on page
240.)
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
Appendix M
726
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.
=
PP < 0
=>
=>
PP < o
E,
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.
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) -
- 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
R1
and
R2,
An Exact Definition of
the Idea of a Function
APPENDIX N
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.
f: R-+ R
: x
1-->
2
x
for every x.
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.
f
P=(x,x2)
f
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)
2)
f ={(x, x2)}.
B=R
f=Sin-1= cx, y) \ -1 x 1,
, 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
729
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).
thing as a fifteen-word rule.
function.
different from his name; and in the same way, a function is different from the phrases
and formulas that we use to describe it.
such a way as to be mathematical objects.
730
Appendix N
Table 1
Natural Trigonometric Functions
Angle
fu
Angle
46
47
0.803
0.719
0.695
0.820
0.731
0 682
1.072
0.838
0.743
0 669
1.111
ful
n t_1
. l _g_e_
0
0.000
0.000
1.000
1
2
0.017
0.017
1.000
0. 035
0.035
0 999
0.035
3
4
0.052
0. 052
0 999
0.052
0.070
0. 070
0.998
0. 070
48
49
0.855
0.755
0.656
1.150
0.087
0 087
0.996
0.087
50
0.873
0.766
0.643
1.192
0.105
0.105
0.995
0.105
0.890
0.777
0 ()29
l.2:l5
7
8
0.122
0.122
o.993
o.12:i
51
52
0.908
0.616
1. 280
0.140
0.139
0.990
O.Hl
53
0.788
0.925
0.799
0.602
1.327
9
10
0.157
0.156
0.988
0.158
0.942
0.809
0 588
1.376
0.175
0.174
0.985
0.176
54
55
0.960
0.819
0 574
1.428
11
12
0
i:i
0
14
15
0.000
0.017
0.192
0. 1\Jl
0.982
0.194
0.209
0.208
0.978
0 2J:l
0.227
0.225
0 974
0.231
0.244
0.242
0.970
0.262
0.259
0.279
0.297
0
5()
57
1.o:l6
0.977
0.829
0.559
1.483
0.995
0.839
0.545
1.540
1.012
0.848
0 530
I.GOO
0.249
58
59
1.030
0.857
0.515
l.GG4
0.966
0.268
60
1.047
0.866
0.500
1.732
0.276
0.961
0 287
0.875
0.485
1.804
0.956
0.306
61
62
1.065
0.292
0.883
0.469
1.881
rno
0.314
0.309
0.951
0.325
1 100
0.891
0.454
1.963
0 332
0.326
0.946
0.344
0
()3
64
1.082
1 117
0 899
0.438
2.050
20
0.349
0.342
0.940
0.364
65
1 134
0.906
0 423
21
22
0.367
0.358
0.934
0.384
1 152
0.914
0.407
0.384
0.375
0.927
0.404
1 169
0.921
0 391
2.356
23
0.401
0.391
0.921
0.424
1 187
0.927
0.375
2.475
24
25
0.419
0.407
0.914
0445
66
0
()7
0
()8
69
l 204
0.934
0 358
2.605
0.436
0.423
0.906
0 466
70
1.222
0.940
0.342
2.748
71
72
1.239
0.946
0.326
2.904
1.257
0.951
0 309
3.078
1.274
0.956
0.292
3.271
1()0
o
n
18
0.454
0.438
0.899
0.488
0.471
0 454
0.891
0.510
28
0.48\J
0.469
0.883
0.5:32
29
30
0.506
0485
0.875
0.554
7:3
740
1.292
0.961
0.276
3.487
0.524
0.500
0.866
0.577
75
1.309
0.966
0.259
3.732
31
32
0.541
0.515
0.857
0.601
1.326
0.970
0.242
4. 011
0.559
0.530
0.848
0.625
76
77
1.344
0.974
0.225
4.332
33
0.576
0.545
0.839
0.649
1. 361
0 978
0.208
4.705
5.145
34
0.593
0.559
0 829
0.675
78
79
1.379
0.982
0.191
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
1.414
0.988
0.156
6.314
37
38
0 646
0.602
0.799
0.754
81
82
1. 431
0 990
0.139
7.115
0.663
0.616
0.788
0.781
1.449
0 993
0.122
8.144
39
40
0.681
0.629
0.777
0.810
83
84
1.466
0.995
0.105
9.514
0.698
0.643
0.766
o.s:io
85
1.484
0.996
0.087
41
0.716
0.656
0.755
0.86!)
1.501
0.998
0 070
14.30
42
0.733
0.669
0.743
0. 900
86
87
1.518
0.999
0.052
19.0
43
0.750
0.682
0.731
0.933
1.536
0 999
0.035
28.64
44
0.768
0.695
0.719
0.966
57.29
45
0.785
0. 707
0.707
i. ooo
I
I
ggo
0
39
90
1 553
1.000
0.017
1.571
1 t. ooo
0 000
: :: I
2G0
27
1143
i"
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
e"
e-x
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
148.41
0.0067
1.6
4.9530
0.2019
403.43
0.0025
1.7
5.4739
0.1827
1096.6
0.0009
1.8
6.0496
0.1653
2981.0
0.0003
1.9
6.6859
0.1496
8103.1
0.0001
2.0
7.3891
0.1353
10
2.1
8.1662
0.1225
2.2
9.0250
0.1108
2.3
9.9742
0.1003
2.4
11.023
0.0907
22026
0.00005
732
Appendix N
Table 3
Natural Logarithms of Numbers
n
log,n
*
log,n
log,n
---
4.5
1.5041
9.0
2.1972
4.6
1.5261
9.1
2.2083
4 7
1 .5476
9.2
2 2192
4.8
1 .5686
9 3
2.2300
4.9
I 5892
9.4
2.2407
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. 3
I. 4
0.2624
5.8
I.7579
13
0.3365
5.9
1 7i50
14
I. 5
0.4055
6.0
1 7918
15
1.6
0.4700
6. I
l. 8083
16
2.7726
I. 7
0.5306
6.2
1.8245
17
2 8332
I. 8
I. 9
0.5878
6.3
1.8405
18
2.8904
0.6419
6.4
1.8563
19
2.9444
2.0
0.6931
6.5
I. 8718
20
2.9957
0.0
0.1
7.6974
0.2
8.3906
0.3
8.7960
0 4
9.0837
0.5
2.5649
2.6391
2.7081
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
70
4.2485
3 0
1 0986
7.5
2.0149
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
8. l
2.0919
JOO
4.6052
3 7
I. 2809
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
I. 4110
8.5
2.1401
4. 1
8.6
2. 1518
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. x
7. x
<
-3
>
13
1.3
3. x
9. x
PROBLEM SET
1. (
5. (0, 10)
-a:J,
9.
a) (0,
<
<
t
2
) b) [-1, a:J)
a:J
25. [t. tJ
[-1, 3)
29.
PROBLEM SET
c) 2v'2
v3)
2.3
3. y
4, x 0
7. y 2x, x - 1
21. b) y x
PROBLEM SET
1. y - 1
5. 4y
11. (3, 0)
1. x 3
5. x2 + y2 + 4x - 4y + 4
0
9. x2 - 2x + y2 - 3
=
a:J
d) v2
23. x
-3
2.2
b) 6v'2
PROBLEM SET
-3
<
3. (-1, 1)
7. (- a:J, -1) and [3,
11. ( -a:J, a:J)
15. { }
19. {1}
23. (-2, 2)
27. [f, tJ
31. [t. 3)
13. {-2, t}
17. ( -a:J, a:J)
21. (- a:J, -2) and(-!-, a:J)
5. (1
>
1.4
-1)
1. a) 2v'2
5. x
11. x
3x
2.4
l
-6(x
- 2), y
25
x
5
7
S
3. y
f(x - 1), y
fx - f
Selected Answers
734
PROBLEM SET
2.6
7. F= (0, f),D:y= t
11. F= (-1, t),D:y= -t
9. F= (2,f),D:y=f
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 =
(0, O) =
-4;
PROBLEM SET
y > 0 if -2
<
<
0 or 2
<
x; y
3. 50
5. 33
7. b3
+
(m
1)7
PROBLEM SET
n7
7. t
15. a) n 4 b) n 2 1010 c)
>
<
2c4
2c5
<
slope at
b4
b5
11. >2
i=3
l)(n
- 1)
5
_4,Q
3
9. t
-2 or 0
2.10
>
2c3
2.9
itXQ
17. a)
7. in(n
(n
<
v3
5. 2n2
PROBLEM SET
0 if x
----2 --:
2.8
1. 14
9. m7
<
> -2 +
. - 3)
!lQ
3
11. t
1 I
v E(8
2
also works. Note that you were not asked to find the smallest possible n.
98
b) n
PROBLEM SET
>
1
-2
3.4
7. bounded with M = 1
9. bounded with M = 1
17. unbounded
M= 1
Selected Answers
PROBLEM SET
3.5
1. 70x9 - 8x7
3.
+ 1)2
-1
7
. (x - 1)2
11. a) 3ay + 2x b) 3xy
-2y3 - 3
5.
(y3 - 3)2
9.
13.
3(1
x)2
2(x2 - 1)
(x2 + x + 1)2
23.
x2 - x
-3x2
4x
----
2x
l)(x3
3x
---:====
v(1 - x2)s
-3x + 1
3
. (x + l)s
7)711(3x2
2x - 1)
3.6
5. 712(x3
6x2
3a 2
-x
2x + 3
1. --;=====
2Y(x + l)(x + 2)
23. f(x2
1 9. ---===
Vl - x2
PROBLEM SET
19. f(3x2
(x
15. 4x3
1
17. --2.Yx + 1
-1
21.
2.Y
9.
7.
11.
<
+ x)l/2
1)312(2x
x -1
-:===
.Yx(x - 2)
-1
----:===-----:===
v1 - x Y(l + x)s
3)
27. f. x<PfqJ-1
q
PROBLEM SET
1. a)
x2
2
3. a) x
x2
2
b) - b) -x
PROBLEM SET
1. t
bll
3.7
c) tx lxl
c) JxJ
d) x
d) 1
e) -x
e) -1
f) Ix!
f) sig x
3.8
1
11
3. a) U
b) - (b11 - a11)
5. a) -!
b)-
x4
4
x2
- -x
2
c)
1 b101
(
101
.
_
alOl)
d) -- (x n+l - an+l)
n
+
7. a) vs - v2
b) 0
735
736
Selected Answers
9. (1+ xlO)lOO
13.
11. 2(V2 - 1)
15.
PROBLEM SET
674
3.9
t2
-2 +2t+3
5.
1. ft2+4t+4
3.
7. -Vl- x2
9. 2v/ - 2v2+ 5
7
-1
l3. 3(1 + t3) +
6
17.
Vo
15.
-100+g
PROBLEM SET
20
g
, a(t)
19. gL
9.
I3.
I7.
--.
I5.
I
' -;::==-;:==:::::;
-4x3
' (x4+I)2
I9. 2VI+(2x)8
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
9. a) -I
I I. a)
ro
0,
20
g
PROBLEM SET
[ ]
I
5
II. -vx(x2+I), -2x
vx 2vx
(1 +x2)2
x x
-- I+ x
I- x
7. v:;,
-2x
--
t3
2
1i.3+t+3
3. x4 - x, 4x3 - I
VI+ x8
2 1
- 20
3.10
4x7
I + x2'
+t
6 ft/sec
I. x2, 2x
5. VI+x8,
t5
20
4.2
b) -cos 0
b) -cot O
13. cos 0
27. 4 cos3 0 - 3 cos 0
Selected Answers
PROBLEM SET
I.
4.3
5. -sinx
3. sec x tan x
sec2 x
11.
9. 2 cos 2x
PROBLEM SET
13.
-
cos x
lcos xi
737
7. -2 sin x cos x
-1
.
+smx
4.4
7. 1
PROBLEM SET
4.5
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
,
2vl- x2
2vu
-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
cos x,f(u) =Jg (t2 +1) dt,f'(u )=u2 +1,g'(x)
11. g(x)
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.
19.
21.
PROBLEM SET
1.
4.6
2x cos x2
3. -3x2 sin x3
-1
sin vx
9.
2v:X
x- 1
-2
13. 1
15. a) 2 sec2 x tan x
19.
-2 sin 2x
25.
-x
sin Vx2 +1
vx2 +1
--
b) 2x sec x2 tan x2
17. -2 sin 2x
21. cos x
23.
27.
b)
---
1 +cos x
!(sin x)-2!3 cos x
31.
39.
41.
43.
29. cos x
0
738
Selected Answers
PROBLEM SET
1.
4.7
3.
----
Y2x - x2
9.
7. 2x
13.
19.
25.
2
xYx4 - 1
x
21 .
.Y1 +x2
1
27.
Y -x - x2
33. 1
37.
15.
--===
1
-
1
-x
11.
lxl Yl - x2
-1
17.
.Y1 - x2
23.
(1 + x2)a/2
2x
1.
.Y1 - x4
35. J-1(x)
--=
.Y2
5. 1
2+2x+x 2
2x
2 +2x2+x4
1
x.Yx2 - 1
1
(1
x2)a;2
77"
77"
PROBLEM
SET
1. log.x+ 1,
4.9
1
x
3. (I +2x)e2"', (4+4x)e2"'
2e"' cos x
500 -500
9. -,- x2
x
7. 2/x, -2/x2
13. 0, 0
19.
-1
1
-1
x ' (1 - x)2
PROBLEM SET
4.10
2
1. - In x
x
2x.
3. - -x2+ 1
7. 2x
5. 2x exp x2
11. (In x+ 1) exp (x In x)
13. cot x
19. 2
21. sec3 x
23. 1
25. 2
27. 2
29. 2
Selected Answers
PROBLEM SET
4.11
2x
1. increasing on
31. v'1
x2
33
43. x = In 2
Vx4-1
PROBLEM SET
739
2
v'1
4x 2
45. x = In (2y)
v'x2 - 1 )
5.1
0.,
y
-2
9. increasing on [O, 1]
-1
[0, tr]
.. x
Selected Answers
740
y
y
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
concave upward
concave
downward
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
[-2, -v'l] and [v'i, 2]
points
15. local maximum O; local minima -1, 1; maximum O; minima -1, l; inflection points
{! %
[-1,
{ff]
and
[{If, l];
concave
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
Selected Answers
PROBLEM SET
741
5.3
1. maxima none; minima none; local maximum 1; concave upward (- oo, 0) and (2, oo) ;
concave downward (0, 2); inflection points none;
Iim f(x) = oo,
Iim f(x) = oo,
lim f(x) = -oo,
limf(x) = 0,
limf(x) = 0,
X-+oo
lim f(x)
x--2-
x--..- oo
x-...O+
x-o-
x--2+
= - oo
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
lim f(x)
X-+3+
X--..-2+
x---oo
X-+3-
X-+-2-
- oo
5. maxima none; minima none; local maxima none; concave upward (-oo, 0) and (0,
concave downward { }; inflection points none;
limf(x) = 0, Jim f(x) = 0, Jim f(x) = oo, Jim f(x) = oo
X--+00
a'-+0+
x--co
oo) ;
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--oo
X-+00
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
13.
19.
25.
x_.co
e
e112
0
PROBLEM SET
1. a
2
7 x =-a
v3
19.
35.
17. e
23. e213
29. $e
5.4
2v'2
= -a 9 128 in.3
v3
k3
15
1728
1
21. {1 3
w2
8
2
25. -
x--..-1 +
3. 2a2
13.
X-+-1-
15. e
21. e
27. 0
{13
4 '17'r3
3v'3
39. (3 +2v'2)r2
17'
742
Selected Answers
PROBLEM SET
5.5
1. t
5. 1
3. t
11. v2
9. h/d
PROBLEM SET
9.
13.
15.
5.8
2 5. cp (u)
v' cp'(u)
27. cp(u)
33. df/dt
35. df/dt
/ (!)
2
5.
13.
17.
19.
21.
23.
25.
31.
37.
, cp'(u)
, cp'(g)
v'1 - u
cos x
=
sin x
1
23. xe"'
-u
'
cp (g)
-tan t cos3 t
-sin t cos2 t
(1 + u2)3/2
6
6t5
-- (Tan-1 u)5 , cp' (g)
(Tan-1 u)6, cp' (u)
+ u2
1 + (tan t)5
-t!Jfor any function/(!) such that/ 2 + t2 + 1 0
= -t3/f3,J1 (t) (1 - t4)114,j(t) !(-4t3)(1 - t4)-3/4
-t3!f'f,
-(1 - 14)114,f;(t) !:(4t4) (1 - t4)-3/4
-t3/f{
v1
+ u2
----
PROBLEM SET
9.
e"' cos2 x)
21. 2t3
1.
1
3. exp Tan-1 g -- (
1 + g2
3
7.
e"'
4x
17. cp(u)
3. 2
PROBLEM SET
11.
l/v'3
5.6
1. v'2
5.
7.
6.2
{i(l + x2)4 + C}
3. {!(2 + u2)2 + C}
6
{x7/7 + !-x5t2 + x3t4 + xt + C}
7. {(x/8)(x2 + t2)4 + C}
{a2a(t3/2 + 5)11 + C}
11. {t(l + sin x)3 + C}
{-{(cos x)312 + C}
15. {te2"' - 6e"' + 12x + se-x + C}
{te3"' + 3e"' - 3e-x - !e-3"' + C}
{ -W + x3)-2 + C1}, x > -1; { -l(l + x3)-2 + C },x < -1
2
{t In (1 + x3) + C1}, x > -1; {t In (-1 - x3) + C },x < -1
2
{ln2(x2) + C1},x > O; {ln2(x2) + C2},x < 0
27 . h-h- sin102 x + C}
H sin3 x + C}
29. { -i cos4 x + C}
35. {sec 8 + C}
{tan 8 + C}
33. {-cot 8 + C}
39. {t sin 28 + C}
41. {t sin 28 + C}
{ -t cos 28 + C}
Selected Answers
45.
{ - isin2fJ + c}
{: - a\ sin40 +c}
47.
55. {tet3 + C}
57. {te2"' + C}
59. {tet + C}
61. {tan x + c}
63. { -ecost + c}
65.
49. { -l cos3
+ C}
fJ
+ t)-112 +
.
67. { -2(2
C}
743
{21
10
2
10"' + c
C}
6.3
PROBLEM SET
1. {t Sin-1 (x2)
3. {! Sin-1 (2y2) + C}
C}
5. {i Tan-1 (3x2) + C}
+
7. {i In (9 +x4) + C}
9. {t Tan-1 (z3) + C}
13. {in (5 +z6)
15. { -tv'I - x8
C}
c}
17. {l In (1 +x8) + C}
33. {xe"'
35.
C}
C}
C}
{ c}
+
37. {x sin x + C}
39. { - x cos x
41.
43. {x3 ln x + c}
{ In x + c}
oo
45. {tln4 x + C}
51. {x Sin-1 x + C}
55.
57. {t In (1 + e2u) + C}
- v'l - x2 + c}
59. {t (e2u - In (1 + e2u)) + C}
61. {t In (1 + x2) + C}
PROBLEM SET
6.4
1. {x ln2 x - 2x In x + 2x
v
{x Cos-1 x
C}
3. {a(x - l)e" + C}
7. {[1/(1 + a2)]ea"'( -cos x + a sin x) + C}
744
Selected Answers
19. {x Tan-1 x -i
17. {x Sin-1x+v1 -x2+c}
21. {t(x sin xe"+e" cos x -xe" cos x)+ C}
In (1+x2)+ C}
6.5
cscx
25. {-cscx+ C}
31.
PROBLEM SET
1.
>= 0
1
=
-,
n
=
-1
--
6.6
{x/v1 -x2+c}
5. {Sec-1 x+ C}
7. {-1/x -Tan-1 x+ C}
c}
1i.{vx2-1+c}
15. {x -Tan-1 x+ C}
19. { t(l +x2)312+ C}
6.7
{ x.
{( -1 -2v) + C
3.
a2va2
(1+v x)2
}
5. {x-2 In (ve"+l +1)+c}
7. {t(1+ \o/)2 -6 d+ \o/)+3 In Jl+ \o/;I+c}
1.
9. {t(v'vx+1) 3 -4Vv1x+1+c}
.;._
+c
x2
Selected Answers
1
- -In I 1-xi +c}
31. {1nlxl-x - 2x2
33. {tx3Tan-1x
- +iln(l+x2)+c}
PROBLEM SET
1. GTan-1
6.8
() +c}
3. {In
1: :I+c}
x 1
x-4 2(x+
5. {In 12 v'x2+
v'17 + 17 + C} 7. {sin-1 ( ;3 ) +c}
11. {Tan-1(x- 3)+ C}
9. {Tan-1(x+ 3)+ C}
15. {In = +c}
13. {1In1: +c}
17. {xl -lnlx-11 +lnlx-21 +c}
1
19. {-x+2lnx-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}
2x ,case=-1 -x2
27. sine= 1-l+x2
+x2
29. {-sec e+tane+ C}
3i. {-x+ 3+c}
33. {In (1+sinx)+ C}, sinx -1
35. {tanx+sec x+ C}
37. { 2.In Isec (e+ 77/4)+ tan (e+ 77/4)1+c}
_
.i
f)I
1: I
=I
1-
716
745
Selected Answers
746
PROBLEM SET
7.1
3.
1
2a
7. a) i-(e - 1/e)
b) -(ea - e-a)
PROBLEM SET
1. f(x)
5. j(5312 - 1)
ln(v'2+1)
7.3
volume
(r/h)x,
trr2h/3
3. tr/2
5. tr/4
7. tr/2
13.
(b)
ii.
x =
PROBLEM SET
7.4
7.
5.
3. 32tr
1. 40tr2
1T
a) 2: (e2
2tr (
3 1-
9. 2 v31T
+ 1)
PROBLEM SET
11.
a) tr(e - 2)
4tr ( v'2 - 1)
T
7.5
5. 4tr(a
+ k)k2
13. 10 v2 1T2
PROBLEM SET
1. (x' y)
(a
=
7. trc(Vb2
11.
b
; '
c2
7.6
5.
v'(a - b)2
c2)
19. a
1
}-tr
4
PROBLEM SET
1. tr/4
7. 10,000
13. t
19. tr/8
27. finite
33. not finite
ac2
9 (x, y)
_
13. 4tr2ab
trCb(b - a)
1T
(b c (3b - 4a))
2' 6 (b - a)
b)
120tr)
7.7
3. t
9. 00
15. 1
not finite
not finite
35. finite
23.
29.
5. 2
11.
17. 2
25. finite
31. not finite
00
1 )
v2
Selected Answers
PROBLEM SET
747
8.1
1. x' =x - 5, y' =y - 6
5. x' = x +2, y' = y +1
PROBLEM SET
8.2
1.x2/4 +y2/3=1
7. x2/5 + y2/9
vs),
(-2, 1
vs);
focal sum
8.3
15. xy=t
17. x2 - y2/3 =1
23. x +y =1, x - y =1
8.4
1. hyperbola
1.
A' =2
3.hyperbola
v2,
PROBLEM SET
11.
c'
=2
:r:
v2
9. A' =t
13. x =2acos
3. 1
5.1
7. 1
11. 0
13.0
15.1
17. 1
1
21. -
23.
25. 0
27. 0
29.x = ae +asin
8,
y =a - acos
31.x=(a+b)coslJ+bcos
PROBLEM SET
1. 0
13.
23.
-2
- co
c' =t :r:
v3
8,
y=0
9.2
1. 1
v3,
9.1
PROBLEM SET
5. straightline
9. -8/ (7r3
19. 2
a+b
a+b
- - e,y=(a+b)sin!J+bsin - - e
b
b
9.3
3. 0
5. 0
9. 1
11. 0
15. 0
17.e
19. 0
21. 0
27. 1
29.
31. 1
33.
3
e-
8)
Selected Answers
748
PROBLEM SET
9.4
3. x +y = v'2
9. (x 2 +y2)2 = 2xy
1. y = 2
7. y = .x3
13. 2x +y2 -1 = 0
29. (x2 +y 2)2 = a2(x2 _ y2)
33. 3r2 -16rcos () - 16r sin () + 32 = 0
PROBLEM SET
9.5
3. rr/4
1. 3rr/2
1.
7.
5. t
11. 2
9. 2
PROBLEM SET
+ i-)3/2 - 1)
PROBLEM SET
23. j = ic
1. a) OS
-
5. 2rr
1t. !
9.7
25. a) i
9.8
-+
1)
3. 3
id
PROBLEM SET
-+
13. i(e8"
9.6
7T
i((l
19. x - y = 1
31. r = 1/(1 + sin 0)
---+
-+
-+
fOR + tOP b) OT
-+
-+
-+
OP + f(OR -OP)
-
-+
-+
-+
-+
tOP + toR
-+
-+
---+
9.9
1. maximum at x = 0, K = 2
3. maximum at x = (45)-114, K = 5312 (45)-114 6-112; minimum at x
_53/2 (45)-1/4 6-1/2
K =
-(45)-114,
1 1
5 a'
'b
PROBLEM SET
1. 0
7. 0
13.
19.
10.1
3. 0
5. 0
11.
9. -
- 00
15. 1
00
00
17. 2
21. converges
23. 2
25. In 2
27.
29. 1
31.
33.
00
35. converges
Selected Answers
PROBLEM SET
749
10.2
3. not convergent
1. not convergent
1T
7. -1 + 1T
11. convergent
9. convergent
13. convergent
21. convergent
17. convergent
23. convergent
25. convergent
00
33. I <-1)ix4i, -1
<
35. convergent
x 1
i=O
PROBLEM SET
10.3
l.
10.4
1
5. IRnl n
1
7. IR..I
<n
11. IRnl
1)4
9. !Rn!
(n
1
+
1)9
possible.)
1
13. !Rn! n
1
15. IR..I
2n2
PROBLEM SET
1.
0.019997
5.
a) /(x)
7. a)/(x)
10.5
00
(-lY( v)i
oo
L (-l)i x5if2
i=O
PROBLEM SET
+)
2 /2
il (-1) 2(0 49)<i
i + 2
00
c)
ro
2(0.25)<5i+2)/2
51 + 2
c) L (-1)' -.
.
i=l
--
10.6
1. (-1, 1)
3. (-1, 1)
5. [-1, 1)
7. [-1, 1)
9. (-1, 1)
11. (-1, 1)
17. ( -oo, ex:>)
13. ( -oo,
oo
19. [-1,1]
15. ( -oo,
oo
21. [4 - 1/e, 4
1/e]
Selected Answers
750
PROBLEM SET
oo
I (-1Y
i.
i =O
(-l)
oo
5.
9.
oo
10.7
xi+ 3
. - , c-1, 11
l +1
i=O
oo
x2i+l
.
, (- oo, oo
-l)i 2i+1
7
(
2
(2i +1)!
i
co x3i +3
11. I-.- c- oo , oo
i=O ,I.
x2i+1
co
15. Ic-1)i .
[-1, 11
c 2l + n2,
i=O
x2i+l
(-l)i
19.
+
(2i
1)(2i + l)! ' - oo, oo)
i
xi+ 2
. - , ( -1, 1 1
l +1
22i+ix2H1
co
3.
I c-1)i
oo
21.
25.
i .I (-l)i
co
i=O
f(x) = e"'l2
22ix2i+3 x3
. 1 + -, (- oo,
2
(2I )
9.
13.
n+ 1
17. 2113
21.
oo
J.
00
i=O
(n + 1)!
=(n+l- i)!i!'
xi
.
l
on
on
( )
i- 1
I c-1Y
xi+2
+2
-:-- on
l
on
( - v2, v2)
(-1, 1)
19. 2k
co
I.
an = 0 if n is
3.
a0 = 0, a1 = 1, a2 = 0
5.
an = 0 if n is
even,
an
(-l)m
if n= 2m + 1
(2m + i) !
( -l)m .
1f n= 2m +1
a11 = (
Zm + l)
on
(-1, 1)
on
(-2, 2)
x2i +l
on -v'2, v'2)
2i
2 (2i +1)
10.9
even,
x2i+I
( -:t)
;o (;) (J
()
i=O
15. 211a
(-1, 1)
x2i+1
n!
(-1, 1)
2 i(2i + 1)
PROBLEM SET
1
7. an= f
n.
22ix2i
. ,, c- oo , oo
c2l )
i=O
29. j(j- 1)
(j- i +l)xi-1
23.
10.8
-:i xi+ 2
i=O
co
( )
( )
i (;)
(; )
-i
_2 ( )
i
oo
PROBLEM SET
Selected Answers
9.
11.
a0 =In 2, a1 = !, a2 = -!
an
0 1f n
IS
PROBLEM SET
1.
IRn(x)j
odd,
an =
(-l)m-1.
if
1m
10.10
lxln+l
(n + l ) !
PROBLEM SET
10.11
5. -1
3. -1
1. -4
11. 2i
9. 1
15. -9 +sv'3;
13. 3 +4i
19.
2i
1
5 -5
25.
v'3
21.
2i
1
10
3i
10
27. -i
33.
31. i
PROBLEM SET
23.
3.
z =
5.
z=
z=
1
--=
v'2
v'3 i
- - 7
29. i
i
+5
35. 1
1ei"16
10.13
----='
----=
v'2 v'2
i
--
v'2
-2, 1 + v3 i, 1
-1, i, -i
-
10.12
PROBLEM SET
z=
5-5
5.
1.
17.
l , i, -i,
v2
1
'
- v3
- ----=
v'2
--
v'2
1
'
----=
v'2
i
----=
v'2
i
1
v'2
v'2 v2 - v2'
'
v2'
1
v'2
- v2
751
152
Selected Answers
PROBLEM SET
1.
3.
PROBLEM SET
5.
9.
10.14
GO (i + l)Xi
.L
7.
.,
I.
i=O
PROBLEM SET
if 0 <
n-+co
5. 5x - Sy + z = 3
x 2 + y2 +
z2 = 4.
11.2
+L +_:_
3
__l_
(2i +l)x2i
(2 I")I
'+l
( -1)'
3. x + y +z = t
7. x - y +2z = 6
2ze"2
11.1
1. x + 3y + 2z = 14
PROBLEM SET
.-
oo
n--+ co
i=O
lim/n(x) = 0 = U lim/n(X)
n--+ oo
i.
5.
10.15
n-+co
13.
cos
2z
-
'
v'3
2y
_l'_
1
_:_ + - -
2z
3. - + - +- - 1 = 0 ' - - - - - - + 1 = 0
3
3
3
3
3
3
2
2
5 __!__ + Y + + = 0 - __!__ - Y - - = 0
l
1
1
1
1
l
'
l
7
PROBLEM SET
-x
2y
3
-- +- =0
Vs
Vs
11.3
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,!, !)
13.
=>
=>
=>
15.
11. i = Vi
i = t(V1
Selected Answers
=:>
xi
z
(V2+ V3)+2. (Vi
753
+ V3)
'1
=> a:1
=:>
'1
=1,
'2
11.4
= -2,
'a
=1
PROBLEM SET
11.5
3.
9.
11.
v'3
Va= E1 - E2 - E4, V4
=-E2+Ea+E4
{ (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.2
D.3
D.4
d(P, Q)
d(P, Q)
d(P, Q)
d(P, R)
=max
=0 =>max
754
Selected Answers
1T2
13. /(x) = x2 - 3
PROBLEM SET
12.1
7T2
2
.
1. a0= 0, ai =0, bi= -: (-1)'+1
5
(i37T'2)
'
(-l)i4
j2
--
'
2
bi = - (-l)i+l
E_ (-l)Hl
a=
Ob=
i
'
i
1T
9.
3 . a0 = - ai =
3
a0= -4 + -2, ai
1T
11. a0=-4
'
.
1
((-l)i - 1), bi=-: (-l)i+1
l 1T
l1T
1
.
3
1
+
a=i
i
j ((-1)'- 1) ' b=-(-1)'
21T
PROBLEM SET
12.2
3. ! cos 3x +!cos x
7. -i cos 3x + t cos x
1. -! sin 3x + ! sin x
5. -! cos 3x + ! cos x
9. i - t cos 2x + i cos 4x
PROBLEM SET
12.3
1
2
1
. (-2 + (e" + e-")(-l)i)
7r(l + 12)
i
( +2 + (e" + e-")(-I)i+l)
b=
'
7r(l + i2)
[l -u
PROBLEM SET
13.1
-4
3
-2
1.
M-
5.
Ker/={(x,y, z) I 2x + y
3.
M-
[ ]
0
3
-1
z = O}
9. M-
-3 -I
3 -1
2
0
756
!
15
15
LP ]
PROBLEM SET
13.2
3. [ ;]
7 UJ
11. [l 64 !]
15. [ 001 ]
19. [ 130 ]
23. [l J
00 10
27. [! 10 00
5. [4741 5825 ]
9. LJ
13. [i 231 ]
17. [ 00 !]
21. [21327 158 2i]32
01 00
25. [ 00 10
]
29.
31.
no inverse
PROBLEM SET
7.13. 112-2.
I. D 11
-1,
2, D31
-1
l
[; :1
17. -2-2
5. 1
G33 G44
13.5
-D21
M-'
3.9. 56
15. -720
13.4
1. G11G22 - G12G21
PROBLEM SET
Selected Answers
11.
3.
z =
5047
755
Selected Answers
756
5.
[ :]
7.
[! H]
11.
[ _!]
[i !]
0
5
2
0
PROBLEM SET
1
0
0
15.
-1
-1
0
1
1
5
0
0
13.6
1. independent
3. independent
5. independent
7. independent
9. independent
11. independent
13. independent
15. independent
19. independent
21. independent
PROBLEM SET
1. {e
2"'
, e
} , f(x)
PROBLEM SET
1. "fl/'= {
5. "fl/' = { IX1e"'
9. H
13. H
15. H
17. H
19. H
21. H
dependent
13.7
3"'
"'
1X1e-
17.
2"'
3. {e3"',
xe
} , f(x)
3"'
(1
2x)e3"'
13.8
+
1X2
{ IX1 sin x
{ IX1 sin x
{ 1X1 sin x
{ 1X1 sin x
{ 1X1 sin x
{ IX1 sin x
2
x
IX3 X e-
X
IX2Xe-
+
+
+
+
+
+
cos
+
x +
IX2
cos
IX2
cos
IX2
cos
IX2
cos
IX2
cos
IX2
cos
3
"'
IX4X e-
Selected Answers
PROBLEM SET
14.1
23. y2 + z2
21. z2 + y 2 = 10x2
25. x2 + y2 = cos2 z, 0
PROBLEM SET
7T/2
14.2
1. ellipsoid
3. 9ne-sheeted hyperboloid
7. elliptic cone
9. hyperbolic paraboloid
19.
PROBLEM SET
21. 87T
2y3
fx=
15.
/.,
17.
fx= (x2
19.
21
f"'
/v =
1 ,
v'x 4 + y4 +
_
x2)
(x2 + y2)2
x(x2
.fv = (x2
+ y2)2
'fv= (x2
/xv=
y2)
+ y2)2
+ y2)2
.fxv=
'/xv=
2x sin xy
(x2 + y2)2
+ y 4 + 1 )312
.,...y4 + 6x2y2
x4
(x2 + y2) 3
8xy(x2 - y2)
(x2 + y 2)3
f'
>Jv
=/vx
=fvx
=f vx
PROBLEM SET
-4x3y3
,/xv= (x4
1
-cos x
-sin x
cos x
v'x 4 + y 4 +
-4yx 2
4xy2
14.3
2x 3
y(y2
5. elliptic cone
13.
e2'"
=fyx
14.4
= 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 =
1. A
PROBLEM SET
1.
3.
14.5
rp'(t)= -(5t4 +
rp'(t)= 2 cos 2t
rp'(t)
5t4
757
758
Selected Answers
PROBLEM SET
l.
<Pt= 4t3u2
14.6
+ 2t +
4ut + 4u3
<Pu= 0
7. <Pt= 0, <Pu= 0
PROBLEM SET
14.8
1. no ILMax, no ILMin
3. no ILMax, no ILMin
5. no ILMax, ILMin at ( !
9
. ILMax at (0, 0), no ILMin
-
!)
16
v3
PROBLEM SET
'coordinates
v3' v3 ' v3
32-V3
9- 'coordinates=
v3
1376
3.
21
7. 0
544
27
14.10
1. 27r/l 1
3. 151T/16
5. ct
7. 0
14.11
28
1. 9
x = 3/27r,
9.
-11/6
11. 0
PROBLEM SET
5.
2
2
,
v3 v3
14.9
1. 35
PROBLEM SET
3.
ji = 0
PROBLEM SET
x = 0,
ji = 0
7. x = 0, ji = - t
3
11. x =- y-=o
4'
13. (0, 0)
14.12
1. 0
3.
7. 0
5. 3/7
11. 3
v3)27r
Index
cardioid,400
Cartesian n-space, 526
acceleration, 119,422
of gravity, 120
addition formula,for the cosine,136
centroid, 335 ff
of nonhomogeneous bodies, 674
chain rule, J 59 ff
proof of, 162
for paths, 642, 646
for functions of many variables, 647
alternating series,446
chord, of a function,72
circle,23
annihilation theorem,435
antidifferentiation, 255
A0,4
apocryphal anecdote,120
approximations by trigonometric polynomials,
549
arc length,303 ff
formula for, 306
disk, 493
set, 669
commutative group, 578
commutative law, for real numbers, 1
for positive series,476
comparison theorem,for positive series,441
completeness, of the real number system,238 ff
Archimedes, 57
complex numbers,479 ff
of a surface,327 ff
areas in polar coordinates, 402 ff
arithmetic series,49
associativity,1
asymptote,368 ff
axis, of a parabola, 38
cone, 41
conical surface, 618
Banchoff,Thomas F., 56
continuity, of a function,75 ff
binomial series,468 ff
binomial theorem, 468
induction proof of,472, 484
bound, of a function, 86
bounded below,194
continuous,definition of,77
continuous function,75
elementary theorems on, 85
continuous interest, 223
convergent, 431
bounded function, 86
coordinate functions,382
sequence,517
bounded sequence, 433,695
set,669
box, for a function, 77
coordinate system,16 ff
Cos, Cos-', 170
cash,203
cosine, series for, 466
759
760
Index
coth, 203
Cra:mer's'rule, 593
cross section, 321
csch, 203
curvature, 425 ff
cycloid, 391
cylinder, 316, 614
cylindrical coordinates, 666
cylindrical shell, 316
solid, 614
surface, 614
D, 95
Dx, 96
damped oscillation, 604
DCP, 239
dd'-box, 79
decreasing function, 206
sequence, 433
Dedekind cut postulate, 239
definite integral, definition of, 308 ff, 312
De Moivre's theorem, 489 ff
density function, 675, 677
derivative, of a function, 69 ff
of an arbitrary power of x, 202
of one function with respect to another, .250 ff
of the integral, 109 ff
of the integral, proof of formula for, 124 ff
determinant function, 582
df/dg, 250
diameter, of a plane set, 668
differences, Ax and !:if, 140
differentiability, for functions of two variables,
638
differentiable function, 89
differentials, 148 ff
differentiation, 89 ff
of complex power series, 496
of series, 453, 505
for functions of many variables, 644
dimension, of a vector space, 529
dimension theorem, 610
directed angles, 128
distance, 512
normal form, 512
segments, 415
direction angles, 513
cosines, 514
directional derivatives, 634, 648
directrix, of a parabola, 38
disk, 493
distance formula, 19
from a point to a line, 38
in an inner-product space, 534.
in polar coordinates, 401
distributive law, 1
diverges to infinity, 435
divisor, 55
domain, of a function, 64
of convergence, 462
e,
factor, 55
falling body, 120
field postulates, 1, 580
finite covering theorem, 350
finite dimensional vector space, 529
first octant, 509
focal difference, 366
sum, 360
focus, of a hyperbola, 366
of a parabola, 38
of an ellipse, 360
force, 119
Fourier coefficients, 546
type, 561
fractional exponents, 101
free vectors, 415 ff
frustum, 329
function, 63
of g, 252
of x, 255
of two variables, 627
function-graph, 67
functional equations, 232 ff
differentiation of, 235
fundamental theorem of integral calculus, 255
Index
Galileo,120
general equation of the second degree, 372 ff
geometric series, 49
gradient, 649
graph, of a condition, 21 ff
of a function, 66
of an inequality,33 ff
group,578
half,of an interval, 241
half-open interval,13
half-plane, 34
harmonic series,439
Heaviside function,76
homogeneous differential equation, 611
Hooke's law,605
hyperbola, 366 ff
hyperbolic functions, 203 ff
hyperboloid of one sheet, 622
of two sheets, 623
hypocycloid,385, 391
ILMax, 213
ILMin, 212
Im z, 482
image, of a function, 67
implies,6, 507
improper integrals, 344 ff
increasing function, 206
sequence, 433
indefinite integral, 256
integration,257
independent variable, 255
induction principle, 51 ff
inequalities,4 ff
infinite interval,13
infinity,13
inflection point,215
initial side, 128
inner product,411
space, 412, 518
inscribed broken line,303
inside function, 155
integers modulo 2, 3
modulo 4, 3
integrability of continuous functions, 353
integrable function, 312
integral,of a nonnegative function, 102 ff
integral test for infinite series, 445
integrand,106
integration by parts, 273 ff
integration, of power series, 454,503
of Fourier series, 556
interior local maximum,213
local minimum, 212
interior of a circle,22
interior point,of an interval,176, 207
of a set in a plane,653
interval, 12
inverse, of an invertible function, 165
761
762
Index
permutation, 583
perpendicular, 30
plane path, 381
p(N), 303
point of contact, 43
point-slope form, 30
polar coordinates, 397 ff
distance formula in, 401
polar form, of a complex number, 487
polygonal inequality, 55
polynomial, 86
in x and y, 633
power series, 453
powers of functioi1s, 99
prime number, 55
projection into a sub space, 541
Pythagorean theorem, 19, 22
quadrant, 20
quadratic function, 179
quadric surface, 620
radius of convergence, 494
range, of a function, 64
ratio test, 457 ff
ray, 25, 128
real-analytic, 468
rectangular equation of a locus, 382
rectangular hyperbola, 370
rectifiable, 304
reduction formula, 277,284
reflecting property, of a parabola, 48
regular path, 392
Re z, 482
right cylinder, 614
right-handed coordinate system, 129
ring, 579
RO, 7
Rolle's theorem, 246
roots of functions, 97
rotation of axes, 374 ff
ruler postulate, 17
saddle point, 655
sample, of a net, 310
of a net over a plane region, 669
sample sum, 310
scalar, 411
Schwarz inequality, 521, 536
Sec, Sec-1, 173
secant line, 45
sech, 203
second derivative, 119
segment, 25
signum, 108
Simpson's rule, 176 ff
error in, 524
Sin, Sin-1, 169
sine, series for, 466
sinh, 203
slice function, 630, 634
Index
slope, of a segment, 28
of a line, 29
slope-intercept form, 29
S(N), 309
s(N), 309
solid torus, 327
solution set, 12
span, 414, 521
sphere, 620
square root, 10
squeeze principle, for functions, 145, 222
for sequences, 435, 518, 696
for volumes of solids, 318
standard position, for an angle, 133
for an ellipse, 362
for a parabola, 40
subadditive function, 537
subspace, 522
substitution, integration by, 284 ff, 291 ff
justification of, 290
summation notation, 49
sup, 243
supremum, 243
surface of revolution, 327, 616
symmetric, 338
value, of a function, 67
vector space, 411
vectors, in a plane, 409 ff
operations on, 410 ff
velocity, 119, 422
vertex, of a parabola, 39
A BCDEFGH'/987 65432
U Jim, 502
unbounded, 193-194
uniform convergence, 502
norm, 538
uniformly accelerated motion, 119
uniformly continuous, 315, 353
uniqueness theorem, 113
unit element, 579
unit tangent vector, 423
upper bound, of a function, 193
of a set, 243
upper limit of integration, 105
upper sum, 309
weight, 120
well-ordering principle, 54
winding function, 130
763