Sequences

and Series

J A Green

c
c

-.
CII CIS
.~
CIS G)
E

-.
E"tJ
G)
~..J
CIS G)
_~::
G)

CIS
o~

.-
..CIS ....0
~

.a .-
._ "tJ
..JW
SEQUENCES AND SERIES

BY

J. A. GREEN

LONDON: Routledge & Kegan Paul Ltd

NEW YORK: Dover Publications Inc
 First PUblished I958
in Great Britain by
Routledge 0- Kegan Paul Limited
Broadway House, 68-74 Carter Lane
London, E.G.4
and in the U.S.A. by
Dover Publications Inc.
I80 Vanck Street
New York, IOOI4
J. A. Green I958
Reprinted I959, I962 , I964, I966
No part of this book may be reproduced in any f01'm
without permission from the publisher, except f01'
the quotation of brief passages in criticism.
Library of Congress Catalog Card Number: 66-2I238

Printed in Great Britain

by Butler 0- Tanner Limited
Frome and London
 Preface

THIS book is intended primarily for students of science and

engineering. Its aims are, first, to present the fundamental
mathematical ideas which underlie the notion of a convergent
series, and secondly to develop, as far as the small space allows,
a body of technique and a familiarity with particular examples
sufficient to make the reader feel at home with such applica-
tions of infinite series as he is likely to meet in his scientific
studies.
I do not believe that these two aims are mutually antagon-
istic. It is true that a certain sophisticated skill is necessary
for the construction of proofs of even quite elementary theorems
involving, for example, the definition of the limit of a sequence,
and that the acquisition of such skill would take more time
than the non-specialist mathematician can spare. But this does
not mean that either the fundamental definitions or the state-
ments of the theorems cannot be clearly understood by the
non-specialist; on the contrary, it is essential that they should
be understood.
Accordingly it has been my policy to lay more emphasis on
the illustration of basic ideas by numerical examples, than on
formal proofs; the latter have often been relegated to small
print, or omitted (such omissions are noted in the text). In par-
ticular the idea of convergence itself is directly involved in the
practical problem of numerical calculation of the sum of a
series, and I have devoted some space to this topic, tradition-
ally neglected in elementary books on series.
It is a great pleasure to acknowledge my debt to my col-
leagues at Manchester, and especially to Dr. W. Ledermann,
for their constructive comments at every stage.
J. A. GREEN
The University,
Manchester
v
 Contents

PAGE
Preface V
CHAPTER
I. Sequences :r
I. Infinite sequences I

2. Sucressive approximations 2
3. Graphical representation of a sequence 3
4. The limit of a sequence 5
5. Other types of sequence 8
6. Rules for calculating limits 9
7. Some dangerous expressions 12
8. Subsequences 13
9. Monotone sequences and bounded sequences :r5
10. The functions x",n' and n'x" 17
II. Solution of equations by iteration 20

Exercises 22

2. Infinite series 24
I. Finite series 24
2. Infinite series 25
3. Convergent and divergent series 27
4. Some examples of infinite series 28
5. Some rules for convergent series 30
6. A test for divergence 32
7. The comparison test 33
vii
 CONTENTS
CHAPTER PAGE
8. The ratio test 38
9. The integral test 39
10. Series with positive and negative terms. Leib-
niz's test 43
II. Absolute convergence 45
12. Power series 47
13. Multiplication of series 51
14. Notes on the use of the convergence tests 54
Exercises 55

3. Further techniques and results 58

I. Numerical calculation of the sum of a series 58
2. Estimating the remainder of a power senes 61
3. Integration of power series 64
4. Differentiation of power series 70
5. Cauchy's convergence principle 72
6. Dirichlet's convergence test 73
Exercises 75

Answers to exercises 77
Index 78

viii
 CHAPTER ONE
Sequences

1. INFINITE SEQUENCES

A sequence is any succession of numbers al , a2, as, . ;

these numbers are called the terms of the sequence. A finite
sequence al , a 2 , , aN is one which has only a finite number
of terms, but we shall be interested mainly in infinite sequences
whose characteristic property is that they have no last term.
For example, the sequence I, 2, 3, .. of the positive integers
is an infinite sequence; so is the sequence I, -t, t, -1, ...
whose nth term is (-I)"+1/n.
If we want to describe an infinite sequence, it is obviously
impossible to write down all its terms. Instead we must give,
as in the last example, a rule for calculating the nth term a".
This rule for the 'general term' may take the simple form
an=f(n), wheref(n) is some easily evaluated function of n; in
the two sequences just mentioned, for instance, we had an =n,
and a,,=(-I)n+1/n , respectively.
It is often useful to write (an) as an abbreviation for the
sequence al, ai' as, . whose nth term is an'
Example 1. Take any fixed number x. Then (x") is the
sequence x, X2, xs, ....
Example 2. an=n' (s any fixed number) is the general term
of the sequence (n')=I', 2', 3', . For S=I this is just the
sequence (n) of positive integers.
Example 3. Take a,,=I, for all n. This defines the sequence
(1)=1, I, I, . . . all of whose terms are equal to I. (Notice that
it is not necessary that all the terms of a sequence should be
distinct.)
Example 4. Another example where the terms are not all
I
 SEQUENCES
distinct is the case %=-1 of Example I. This sequence con-
tains only the numbers I and -I, alternately, viz. -I, I,
- I , I, . .
On the other hand many of the sequences which occur in
practice are defined recursively; this means that a rule is given,
by which the nth term a.. can be computed when the earlier
terms are known.
Example 5. If a..+1=V2a.. , we can calculate a..+ t as soon as
a.. is known. The value of at must be given to start with, and
then any number of terms can be worked out in succession.
If at=I we have a2 =V2.I=v2", a3 =V 2v2, a,= V2V 2V2,
etc., or in decimal notation (a..) =1, 1.414, 1682, 1834,
1.91 5, . . .
Example 6. A similar 'recursive formula', but involving
two previous terms, is a..H =02a..+t -oIa... Here we need
to be given at and a. to start with, and then the subsequent
terms can be found. For instance, taking at=o, a.=I, we
get a3=02a.-oIat =02, a,=02aa-oIa a=-o06, a6=02a,
-0Ia3=-0032, etc.

2. SUCCESSIVE APPROXIMATIONS

One very important way in which sequences make their

appearance in practice, is where the numerical solution of some
problem is attempted by finding successive approximations.
These approximations form a sequence whose terms approach
the number which is being sought. To take an elementary ex-
ample, suppose that it is required to find the numerical value
of v'2; that is to say, we are looking for a positive number A
with the property that A 2=2. By the usual 'square root pro-
cess' of elementary arithmetic, we learn how to make a sequence
of (better and better) approximations to A. The first approxi-
mation given by this process, which we might call at, is I. At
the next stage we get a 2 =I4, then 113=1.-1-1, R,=I4I4, and so
on. None of the terms I, 14, 1.41, 1.-1-14, ... of this sequence is
equal to A. But they approach or 'tend to' A =V2, in the sense
2
 GRAPHICAL REPRESENTATION OF A SEQUENCE
that if we go sufficiently far along the sequence, we get numbers
which differ from A by as little as we like, It should also be
2' 1'414
I'

24 1'00
96
281 400
281
282 4 II9 00
II296
60 4
noticed that, in practice, this is all we require. For in any given
practical application, in which for some reason the value of V2
is required, all that is really necessary is its value correct to a
certain number of decimal places, and this we can secure by
working out afinite number of steps of the square root process.
It is this idea of a sequence which 'tends to' a limiting value,
which we shall discuss in the next paragraphs.

3. GRAPHICAL REPRESENTATION OF A SEQUENCE

It helps in understanding this notion if we can represent

sequences graphically. We shall do this in either of two ways;
first we can simply mark the values of al , as, as, ' , , as points
on a single axis or scale. It is advisable to write above each
point the name of the term to which it is equal (and a given
point may correspond to many terms), For example, the
sequence (1-;)=0, t, i, t, ' , ,is represented in Fig. I
0, 02 03 040506
! "

o
FIG. I

while the diagram for ((-1)")=-1, I, - I , I, . . . has only two

3
 SEQUENCES
points, each of which serves for infinitely many terms of the
sequence.

-I o +1
FIG. 2

Although this representation is very compact, it may be pre-

ferable to display the sequence more fully. Our second method
of depicting a sequence (a,,) is to draw a 'graph', regarding a" as
a function of n. However this graph is not a continuous curve,
but consists merely of a succession of isolated points, because a"
is supposed defined only for n=I, 2, 3, ... Fig. 3 shows the
graph of ((-1)"), for which a1 =-1, a2 =1, a3 =-1, a4 =118 ,
etc. For this sequence, it is not possible to give a real meaning
I
2

w S
1 4

FIG. 3

to the value of a" for general (non-integral) values of n-for

example we cannot define a3/1=(-1)3/2 without using complex
numbers. But even in a case, such as a,,=I-~ (Fig. 4), where
n
4
 THE LIMIT OF A SEQUENCE
it would be possible to 'fill in' the intermediate values, we re-
frain from doing so, on the grounds that it is only the integral
values of n which interest us at the moment.

x x x x x
x x
x
X

n
2 4 6 8 10
FIG. 4

4. THE LIMIT OF A SEQUENCE

Figures 3 and 4 display very clearly the fact that these

sequences are 'tending' to 'limits' or limiting values, as n
becomes large. It is clear, for example, that as n increases,
(_t}n becomes more and more nearly equal to 0, and that
I-~ becomes nearer and nearer to I. We express this by saying
n
that (_t}n tends to 0 as n tends to infinity, and that I-~
n
tends to I as n tends to infinity, respectively. The precise
definition of this kind of statement is as follows.
Definition. A sequence (an) tends to a limit A as n tends to
infinity, if, given any positive number h, however small, we can
find an integer N" such that all the terms a. of the sequence
after the N"th lie between A-h and A+h.
Notation. We write 'an--+-A as n--+-oo' (read 'an tends to A as n
tends to infinity'), or sometimes lim an=A. Occasionally the
11--+00
phrase 'as n tends to infinity' is omitted, for shortness; then we
should write simply an--+-A, or lim an=A. We can think of the
terms an as 'approximations' to A. If certain 'limits of tolerance'
hare allowed, i.e. if we are satisfied when an lies within h of A
5
 SEQUENCES
Ath I- - - - _ .Jfo_ - - ~- - - - - _______ _
A x Xx Xv

A-h ~ - - - - - - -x- - - - - - - - - ~ - - - - - .!
x
)(

n
Nh
FIG. 5

on either side, then (according to the Definition), all the terms

an after a certain one (which we have called the N"th) must lie
within these limits. Further, this must continue to be true,
whatever value h has-e.g. if we set a narrower tolerance k,
say (with k<h), it must again be true that all the terms after
a certain point are within the new limits (but of course it would
usually be necessary to go farther along the sequence; i.e. the
number N k such that all the terms after the N kth lie between
A-k and A+k, would usually be larger than N,,).
Example 1. To prove, directly from this definition, that
an=I-~---+-I as n---+-oo. The term an differs from I by ~.
n n
Take h=OOOI as an example; clearly an lies between 1-0001
and 1+0001, if n has any value >1000. Therefore N,,=IOOO
would satisfy the conditions in the definition, for the special
value h=OOOI. This is not enough: we must prove that an N"
can be found corresponding to any h. This however is quite
easy; take N,,=the first integer greater than I/h. Then it is
clear that an lies between I-h and I+h, whenever n>N".
Example 2. We prove next that xn---+-oas n---+-oo, if -I<X<I.
(This includes the case x=-t depicted in Fig. 3.) Let c be the
numerical value of x, disregarding the sign. Then the numerical
value of xn is cn (for example, the numerical value of (-l)n is
always mn, although (_t)n is negative if n is odd). Now cn<h
if n log c< log h. We must remember that log c is negative,
because C<I, and when we divide an inequality by a negative
number, its direction is reversed. Thus n log c< log h is equiva-
lent to n > log h/log c. Therefore if N" is the first integer
6
 THE LIMIT OF A SEQUENCE
greater than log hjlog c, then xn lies between o-h and o+h, for
every n>N". Since such an N" can be found for any positive h,
the conditions of the Definition are satisfied (with an=xn, and
A =o). (Another proof, not involving logarithms is given in
Example 6, p. 17).
In the case ~= -I, we must take c=l, and log c= -0'3010.
Thus N" may be taken to be the first integer greater than
log h/( -0'3010). For example, log (0'001) = -3, and so N,,=IO,
the first integer greater than - 3 / -0'3010. This shows that (-I)n
lies between +0'001 and -0'001, for all n> 10.
Example 3. Another useful limit is n'---+-o as n---+-oo, if s<o.
For example, the sequence (n- 1}=I, l, 1, 1, i, ... tends to 0
as n---+- 00.
Put s=-k, so that k>o. Then n'= ;k<h if nk>~, i.e. if

n> ;Ii. Therefore the conditions of the Definition will be satis-

fied by ta~g N,,=first integer greater than ;I~.

Example 4. The 'constant sequence' (a}=a, a, a, a, , all
of whose terms are equal to a fixed constant a, has a as limit,
according to our definition. For whatever value h may have,
a/J, the terms are between a-h and a+h.
Example 5. Any number can be written as a decimal, which
may of course be an infinite decimal. This is equivalent to say-
ing that the number is the limit of a sequence of finite decimals;
for example n=3'14IS9Z6S ... is the limit of the sequence
3, 3'1, 3'14, 3'141, ...
A number which can be expressed as the quotient of one integer
I -5 6
by another (such as -, - , -, etc.) is called a rational number. Not
34 1
every number is rational; for example it can be proved (this is not
easy) that:lt is not rational. However every number is the limit
of a sequence of rational numbers. For example, :It is the limit of
31 3 14 3 1 41
the sequence 3, 3'1 = 10' 3'14 = 100' 3'14 1 = 1000' whose
7
 SEQUENCES
terms are all rational. Obviously the same argument could be
applied to any number. (It should be remarked that there are
always many different sequences of rational numbers tending to
any given number.)

5. OTHER TYPES OF SEQUENCE

Not every sequence has a limit, as can be seen by considering

the following examples.
Example 1. (n)=I, 2, 3, 4, ....
Example 2. (-n 2 )=-I, -4, -9, -16, ..
Example 3. ((-2)")=-2,4, -8, 16, ... .
Example 4. ((-1)")=-1, I, -I, I, ... .
In none of these cases does a" tend to a limit A. In (I), for
example, the terms become larger and larger without limit; we
say that '(a,,) tends to +00 as n-+oo'-a statement which is
not covered by our definition on p. 5, because we assumed
there that A was an ordina(y number. The precise definition of
the phrase '(a,,) tends to +00 as n-+oo', or, as it is usually
written, 'a,,-++oo as n-+oo', is as follows:
A sequence (a,,) tends to + 00 as n-+oo, if, given any number
K, however large, we can find an integer N g such that all the
terms a" after the N gth are greater than K.

K~------ ________ ~~W~

" " "
_ _ _ _ _ ___

" n

FIG. 6

Example 5. n'-++oo as n-+oo, if s>o (cf. Example 3,

p. 7)
8
 RULES FOR CALCULATING LIMITS
For given any K (which we may assume is positive), we have
n' >K if n> TK. Although VK will not usually be an integer, we
just take N K=the first integer greater than n, and the condition
of our definition is satisfied.
Example 6. xn---+-+oo as n---+-oo, if X>I (cf. Example 2,
p.6).
For xR>K provided that n log :>log K, and this time log :>o,
because :>I, so that we may divide by log : without reversing
the direction of the inequality. Therefore :n>K if n>log K/log:,
and thus we may take N K to be the first integer greater than
log K/log:.
By analogy, we say that a sequence (an) tends to - 00 as
n-4oo (written an---+-- 00 as n---+-oo), if, given any number L,
we can find an integer N L such that all the terms an after the
N Lth are less than L. This is equivalent to saying that
-an---+-+ 00. For instance the sequence (-n 2) in Example z
tends to - 00.
We often write lim an =+ 00, as an abbreviation for
'an---+-+ 00 as n---+-oo', and similarly lim an=- 00. But it should
be remembered that '+00' and '-00' are not numbers, and
cannot be treated as if they were. Some reasons for this caution
will appear in 7.
To complete the classification of sequences, we say that a
sequence which does not (i) tend to a limit A (ii) tend to + 00,
nor (iii) tend to - 00, is an oscillating sequence. Such sequences
are, from our point of view, 'irregular', although they comprise
a very wide class. Examples (3) and (4) (an=(-z)n and
an=(-I)n) are both oscillating sequences.

6. RULES FOR CALCULATING LIMITS

If (an) and (b n) are two sequences, and if an---+-A as n---+- 00,

while bn---+-B, then the 'sum' sequence (an+bn)=a1+b 1 , a2 +b2 ,
... tends to A+B.
Naturally this statement requires rigorous proof, using thE
definition of 4. We shall not, for reasons of space, give this proof.
S.S.-B 9
 SEQUENCES
but it is really just a precise formulation of the following idea: a..
can be made as near A as we like, by making n large enough; simi-
larly bn can be made as near B as we like. When an is very near to
A, and also bn is very near to B, then an+b.. must be very near
A +B. That means that we can make a. +b,. as near A +B as we
like, by making n large enough.
In the same way we have rules for differences and products:
Rule 1. If an---+-A and bn---+-B as n---+-oo, then an+bn---+-A+B,
an-b,,---+-A-B, and anb,,---+-AB. Furthermore, if c is any con-
stant, then can---+-cA (this is the special case bn=c of anbn---+-AB).
Example 1. Find lim (1+ (1)n). Take an=1 (this defines a
constant sequence, see Example 4, p. 7) and bn =(l)". We know
that a,,---+-I, and bn---+-o (Example 2, p. 6); therefore a,.+b.
=1+ (l)n---+-I +0=1.

Example 2. (I+~)(2-:2)---+-(1+0)(2-0)=2. (We use here

I 1
the fact that _=n- 1 and a=n- a both ---+-0 (Example 3, p. 7)).
n n
Rule 2. If a,,---+-A, and b,,---+-B, as n---+-oo, and if each term
bn;co, and also B;co, then an/bn---+-A/B.
We omit the proof of this rule. The conditions bn;co and B;co
are obviously essential, if the expressions a,.lb n and A IB are to
have any meaning.
Rule 3. Let (a,,). (b n) be two sequences.
(~1 If an---+-+ 00, or if an---+-- 00, then I/an---+-o.
(i~) If an---+-+ 00, and b,,---+-afinite limit B, then an+bn---+-+ 00.
(iit) If a,,---+-- 00, and b,,---+-afinitelimitB, then an+bn---+-- 00.
(iv) If an---+-+ 00, and bn---+-a positive limit B, then
a"b"---+- + 00.
(v) If a,,---+-+ 00, and bn---+-a negative limit B, then
a"b,,---+-- 00.
We shall not give formal proofs of these facts, but the reader
should try in each case to understand the general principles in-
volved. For example (i) if an---+- + 00, it means that an is very large,
for large values of n; hence lla,. is very near o. In (ti), the term b.
10
 RULES FOR CALCULATING LIMITS
is near to B, for large n, while an is very large. Even if B is a large
negative number, the sum an +bn can be made positive and indeed
as large as we like, by making n large enough. Therefore
an +bn---+- + 00.
Example 3. (n+~'=2' 2~, 31, 4i, ... tends to + 00. For
nl
an=n---+-+oo, while bn=~---+-o.
n
Example 4. 4-n+(lt---+--00. For this is the sum of a term
-n, which tends to -00, with a term 4+W n , which tends to
the finite limit 4.

Example 5. (n +1)/(n2+1)=n(I+~) / n2(1+ :2)

=(~)(I+~) / (1+ :2)---+-OXI/I=O, as 1---+-00.
Example 6. (n2+1)/(3n2+n+l)
=1(1+:2) / (1+ 31n + 3:2)---+-1.1/1=1, as n---+-oo.
Example 7. an=(I-n 2 )/(1+2n). Take a factor (-n 2 ) out of
the numerator, and a factor 2n out of the denominator; we get
an =(-;)(1-:2) / (1+ 2:). The factors in the two last
brackets both tend to 1as n---+- 00. Therefore, for large n, an will
be very nearly equal to ( -;) (in the sense that an /( - ; ) will

be very nearly I). Therefore, since -!! clearly ---+-- 00,

2
an---+-- 00 as well.
Example 8. Rational functions of n. A polynomial in n is a
function of the form c"n"+c"_ln"-l+ . +cln+cO' where
co, cl , ... ,c" are constants. If c,,~o, h is called the degree of the
polynomial, and c"n" is called the leading term. A rational func-
tion is the quotient of one polynomial by another; e.g. the
functions in Examples 5, 6, 7 are all rational functions of n.
:u
 SEQUENCES
Let (an) be a sequence whose nth term is a rational function of
c"n"+ c"_ln"-l+ ... + Co ( d ) W .
n, say an= d /e d d c", /er=o. e can Write
/en + k-I n
k-l
+ ... + 0
this

an = dc"n"/e' { 1 +C"-l!.+
kn c"n
+~c"n~}/
"
dk-l
{ 1 + dIe n1+ ... +dd,.on/eI}.
The terms in the brackets both tend to 1 as n---+- 00. Thus for
large n, an behaves like the quotient of its leading terms, in the
sense that the ratio of an to this quotient c"n"/dkn/e tends to
1c
1 as n---+-oo. Thus for example, if h<k, then c"n"/d/en
1
=(c,,/dk) 1c-1l---+-0 as n---+-oo, and therefore an also tends to o.
n
If h=k, then an---+-c,,/dk, and if h>k, then an---+-+ 00 or to - 00,
according as C,,/dk is positive or negative.
I t is sometimes useful to notice the following rule.
Rule 4. If an---+-A and bn---+-B as n---+-oo, and if an<bn, for aU
n, then A <.B.
This is almost obvious. But it should be remarked that if an <bn
for all n, it is not necessarily true that A <B-all we can be sure of
is that A <B. For example, 1- ~ < 1 for all n, but lim (1 -~) =1.

7. SOME DANGEROUS EXPRESSIONS

It is as well to remember that Rules 1 and 2 are not intended

to apply to sequences which tend to + 00 or - 00. Nor should
it be expected that expressions such as 00- 00,0/0, 00/00 have
any definite meaning. In general, we must accept the fact that
00' cannot be treated like an ordinary number, and that division by
o is impossible. The following examples make the position clear.
Example 1. If we take two sequences (an) and (b n), which both
tend to + 00, we might expect the quotient an/bn to have
00/00' as limit. Consider, however, the following three ex-
12
 SUBSEQUENCES
amples (i) an=nll, bn=n, (ii) an=n, bn=n, (iii) an=n, bn=nll.
In each case, an and bn both tend to + 00. But in (i) an/b n
=n---+-+ 00, in (ii) an/bn =1, which ---+-1 as n---+- 00, and in (iii)
an/bn=l/n---+-o. What value are we to give to 'oo/oo'? Our
answer is that we do not give any meaning at all to this expres-
sion; we simply avoid it. (It might be remarked that we could
construct examples like the three we have just mentioned, in
which an/b n has any given positive number we like as limit, or
again, in which an/bn has no limit at all, but oscillates.)
Example 2. The situation with '0/0' is the same. Consider the
following three examples: (i) an=l/n, bn=l/n B (ii) an=l/n,
bn=-2/n, (iii) an=l/n 2 , bn=l/n. In each case, an---+-o and also
bn---+-o, as H---+-OO. But for lim (an/b n) we find (i) + 00, (ii) -~,
(iii) o. Thus '0/0' cannot be given any meaning; it is another
expression which has to be a voided.
Example 3. '00-00.' Take (i) an=n, bn=n 2, (ii) a n=n+4
bn=n, (iii) an=(-i)n+n, bn=n. In each case an and bn---+-+oo
as n---+-oo. Thus if' 00- 00' has any definite meaning, we should
expect to find it by considering the limit of an-bn in these
examples. However in (i), lim (an-bn)=lim n(l-n)=-oo, in
(ii) lim (an-b n) =4, while in (iii), an=bn=(-i)n---+-o.
The reader should not imagine that, for example, an = n+1
n 2 +1
has no limit because it 'becomes 00-:1' on 'substituting n=oo'.
00 +1
What the discussion above does show is that we cannot calculate
limits by 'substituting n=oo'. We must work instead by apply-
ing Rules I, 2 and 3 to reduce the given expression to a com-
bination of functions whose limits are either known already, or
can be found from first principles, Le. from the definitions given
in 4 and 5.

8. SUBSEQUENCES

If (an) =a1 , a2 , a3 , is a sequence, then any infinite succes-

sion of its terms, picked out in any way (but preserving the
original order) is called a subsequence of (an).
13
 SEQUENCES
Example 1. ai' a" aI' as, . is a subsequence of (a,,). Its nth
term is al,,'
Example 2. ai' a" aD' all' ... is the subsequence (a,,2) of (a,,).
Example 3. An important type of subsequence is that ob-
tained by removing a finite number of terms from the beginning
of (a,,), but leaving the rest of the sequence unchanged. For ex-
ample a7, as, aD, is the subsequence obtained by omitting
the first six terms of (a,,). The nth term of this subsequence is
al +,,
The behaviour of a subsequence is very easy to describe, in
the cases where (a,,) tends to a (finite) limit, lor to + 00 or - 00.
We have the following rule.
Rule 5. If a,,-+A as n--+-oo, then any subsequence of (a,,) also
tends to A. Similarly, if aft--+-+ oo( - 00) then any subsequence of
(a,,) also tends to + oo( - 00).
For example, if a,,--+-A, it means that, given h, we can be sure
that all the terms a" after some point are within h of A. But if this
is so, then all the terms of the subsequence (which are just some of
the terms of (a,.)) are within h of A, after this point.
Example 4. The sequence (x"'+")=x l , Xl, xu, XIO, is a
subsequence of (x"). Therefore it tends to 0, if -I<X<I, since
x"--+-o in this case.
Example 5. (see Example 3). If (an)=a l , ai' as, .. tends to
A as n--+-oo, then any sequence obtained by removing a finite
number of terms from the beginning of (a,,) tends to the same
limit. For example, the sequence (an+!) =al , as, a" .. --+-A.
It is also clear that the introduction of a finite number of new
terms at the beginning does not alter the limit of a sequence.
For example, we know that (lin) =1, t, 1, t, ... tends to o.
Then the sequence 42, -4, 5011, 8, I, t, 1, t, I, ... , obtained
by putting in four extra terms at the start, still has limit o. In
1 When we say that (a,,) tends to a finite limit, this means that
a,,--+-some number A, as defined on p. 5. The word 'finite' is introduced
to distinguish this from 'an--+- + ex)' and 'a,,--+-- ex)', but it should be
remembered that these have quite dIfferent definitions from 'a,,--+-A'
(p.8).
14
 MONOTONE SEQUENCES AND BOUNDED SEQUENCES
fact, the behaviour of a sequence, of any type, is unaffected by
removing or inserting or altering a finite number of terms.
Example 6. An interesting exercise in the application of the
principles of this section is the following. Let a" =sin nx, x being
a fixed ang'le which is not an integral multiple of n radians (if
x is a multiple of n, then a,,=o for all n). We shall prove that
(a,,) does not tend to a limit. For suppose, to the contrary, that
a,,---+A as n---+oo. Let b,,=cos nx. Put ex=(n+I)x, (3=x in the
trigonometric identity sin (ex+(3)-sin (ex-(3) =2 cos ex sin (3;
this gives a"+2-a,,=2b"+1 sin x, hence b"+l = (a"H- a,,)/2 sin x.
Now if a,,---+A, it follows (Example 5) that a"+2---+A as well.
Therefore b"+l---+(A-A)/2 sin x=o, as n---+oo. Therefore
b,,---+o also (for (b,,) =b 1 , b2, b3 , is obtained by simply adding
one extra term b1 to the beginning of (b,,+!) =b2, b3 , ). Con-
sequently the subsequence (bin) =b2, b" be, has the same
limit o. But b2,,=cos 2nX=2 cos 2 nX-I=2b,,2_I. Take the
limits of both sides, and we get a contradiction 0=20 2 -r.
This can only mean that our original assumption that (a,,) tends
to a limit A, is false. (We have proved only that (a,,) does
not tend to a finite limit. However it is trivial that (a,,) does
not tend to + 00 or - 00 either, since - I < sin nx < I for
all n.)

9. MONOTONE SEQUENCES AND BOUNDED SEQUENCES

(a,,) is called an increasing sequence if a" <a,,+! for all n; that

is, if a1 <a2<a3 < ... Similarly, a decreasing sequence (an) is
one for which a" :>a,,+! for all n; thus a1 :>a2:>aa:> . . . A
sequence which is either increasing or decreasing is sometimes
called a monotone sequence.
Example 1. (n), (1-;) are increasing sequences.
Example 2. If (a,,) is increasing, then (-an) is decreasing, and
vice versa.
A sequence (a,,) is bounded above if there is a number H such
that a" <II for all n. It is bounded below if there is a number G
such that a,,:>G for all n.
IS
 SEQUENCES
Example 3. (n) is bounded below. but not above. (Every term
is >o).
Example 4. (I-~)=O. t. i. 1. t .... is bounded above and
1
below. For 0<;1--<;1. for all n.
n
We come now to an important general principle.
Fundamental Theorem on Monotone Sequences. (i) If an
increasing sequence (a,,) is bounded above. then it must tend to a
finite limit. (i.) If the increasing sequence (a,,) is not bounded
above. then a,,---+-+oo as n---+-oo.
Similar results hold for a decreasing sequence (b,,): if (b,,) is
bounded below. then b,,---+-a finite limit; if (b,,) is not bounded
below. then b,,---+-- 00. (This means. it is impossible for the in-
creasing sequence (an) to osciUate or tend to - 00. and it is impos-
sible for (b n) to osciUate or tend to + oo.)
Let us represent the terms a" by points on a line. Each a" is to
the right of all the preceding terms, because (a,,) is an increasing
sequence.
Case (ii) is illustrated by Fig. 7. For any given number K we
must have a,,>K for some n (otherwise a,,<K for all n, and that
would mean that (a,,) was bounded above). Then, because the
Q

, 'I ' , ' )

K
FIG. 7
sequence is increasing, all the subsequent terms are >K. This
shows that a,,---+- 00.
In case (i), there is some number H such that a,,<H for all n
(Fig. 8). A rigorous proof that a" tends to a limit is beyond the
scope of this book, but we can see how thIs comes about, as follows.

, Ii' , "'A3 Ii 4 '

i i
o
FIG. 8
16
 THE FUNCTIONS x", n' AND n'x"
All the points a" are to the left of H. There must be some greatest
integer m, which has points of the sequence to its right. In Fig. 8,
m=2, and all the a" after a, are between 2 and 3. Similarly, there
must be a greatest one among the numbers 2'0, 2'1, 2'2, , 2'9,

as a6 a 7 a8 a 9
, , , ,

2.'0
i i
i PI!
A 18
, i
30
)

FIG. 9

which has any points a" to its right. In Fig. 9, which is an 'enlarge-
ment' of the relevant part of Fig. 8, this 'greatest number' is 2'7.
All the a" after as lie between 2'7 and 28. In the same way, we
should find that all the a" after a certain one lie in a particular
interval of length 0'01; between 2'71 and 2'72, perhaps. We can go
on with this process as long as we like. The result is a certain
infinite decimal A (which in our case starts off 2'71 ), and A is
the limit of the sequence. For all the a" after a certain point are
within as small a distance as we like of A.

Example 5. a,,=I-~ is increasing, and bounded above. As we

n
have already seen (Example I, p. 6), a" does tend to a limit,
namely I.
Example 6. Let x be a number between 0 and I. Then (x") is
a decreasing sequence (since x"+l=x.x"<x", for all n), and it
is bounded below, because x">o for all n. Therefore it must
tend to a limit, X, say. Now (x"+l)=X2, x 3 , must have the
same limit X (Example 5, p. 14). But (x"+l) can also be ob-
tained by multiplying every term of (x") by the constant x. It
follows by Rule I (p. 10) that its limit is xX. Hence xX =X,
from which (X-I)X =0, and, since we know that X-I~O, it
follows that X =0. This is of course the result which we had
already found in 4 (Example 2, p. 6).

10. THE FUNCTIONS x", n' AND nix"

We first collect for reference some of the results obtained in
this chapter.
17
 SEQUENCES
(i) x"---+o as n-+oo, if -I<X<I (Example 2, p. 6),
x"---++ 00 as n-+oo, if X>I (Example 6, p. 9).
To complete the list, we remark that
x"---+I if X=I (for the sequence (xn) is then the constant
sequence (I)), and
xn oscillates if X<-I.
(ii) n'-+o as n-+oo, if s<o (Example 3, p. 7), and
n'-++ 00 as n-+oo, if s>o (Example 5, p. 8).
There is also the trivial case s =0, which gives the constant
sequence (nO) =(1); thus n~l.
Functions of the type n'xn. Consider the sequence whose nth
term is an =n3(!)n=n 3/2 n. This can be regarded as the quotient
of the two functions n 8 and 2n, which both tend to + 00 as
n-+oo. As we saw in 7, the expression '00/00' gives us no
indication of what the limit of (an) might be, or indeed of
whether it has a limit at all. On the other hand, it is always
worth while to consider some numerical values-these often
give an indication of the behaviour of a sequence, although any
guesses arrived at in this way have to be verified by a rigorous
argument based on our definitions.

1$
I I
--
Z -4 10 zo 50

1$1 I I 8 6-4 1000 8x lOa IZS X 10&

Z" Z -4 16 IOZ4 105 X 10' 11 X 10 15

- - - - - -I- -
1$a/z" 05 Z 10 7"6 X IO- a 11 X 10- 10
I -4
These rough values suggest that, starting with n =10, the values
of 2 n soon become very much larger than those of n 3 (although
they are smaller for some initial values of n) and that in fact
n 3/2R-+o as n-+oo. This is confirmed by the following general
result: if y>l, then n'/yR-+o as n-+oo, for any value of s.
First, if 5<0, then both n' and {I/y)~O, so that their product
n'/yn tends to o. If 5-0, then n'/yn_{I/y)n, and again this tends
18
 THE FUNCTIONS x", n' AND n'x"
to zero. We may assume now that s>o. Because Y>I, the 2s-th
~
root of y is also > I, and so we can write it in the form VY= 1 + t,
where t is a positive number. If a,,=n'/y", then, taking 2s-th roots
~ 21
of each side, .ya:= Vn/(Vy)" = Vn/(I +t)". By the binomial
n(n-I)
theorem (see p. 25) (I +t)"=1 +nt+ ___ t 2+ ... +t", which
2
is greater than nt (for all the terms in this expression are positive).
Therefore va:
~
< Yn/nt = 1 /tYn. Take 2S-th powers of both sides;
this gives a" < 1/(t2n)'. We can make 1 /(tln)' as small as we like by
making n large enough. Thus a" (which is always >0) can be made
as near 0 as we hke by making n large enough, and this shows that
a,,-+o.
This can be expressed in a slightly different way by saying:
n'~o as n-+oo, if O<X<1. For if O<X<I, thenY=I/x>I,
and we can apply the previous case, writing n'X"=n'/(I/x)"
=n'/y". Furthermore, if -I<X<O, then the numerical value
of n'X" (i.e. apart from sign) has the same as that of n'c", where
c is the numerical value of x. Thus n'X"-+o as n-+oo in this
case also. To summarize:
(iii) n'X"-+o as n-+oo, for any s, and any x such that
-I<X<I.
Example 1. Th result n'/y"-+o (in the case s>o and y>I)
can be interpreted as meaning that, although n' and y" both
tend to + 00, y" does so 'faster' than n'. This is true even if s is
very large andy is only slightly larger than I. For example, one
would at first sight imagine that n 1000 was very much larger
than (IOOOI)", and of course this is so for a finite number of
values of n. However eventually (IOOOI)" is much larger than
n 1000, and the quotient n100o/(IOOOI)"-+0 as n-+oo.
Example 2. Find lim (n ll - (IS)"). We can write a,,=nll- (IS)"
=(IS)"{nll/(IS)"-I}. The expression in the brackets {} tends
to O-I=-I, while (r-S)"-++oo. Therefore a,,-+-oo (Rule
3 (v), p. IO). (In a less precise way, one could say that (IS)"
tends to + 00 faster than nil, as in Example I, and that there-
fore for large n the term nil is negligible in comparison with
-(IS)")
 SEQUENCES
II. SOLUTION OF EQUATIONS BY ITERATION

There are several 'iterative methods' for solving equations;

that is to say, methods whereby the solution is obtained as the
limit of a sequence of successive approximations. We shall de-
scribe here the simplest of these methods, for details of others
the reader is referred to books on numerical methods (e.g.
Whittaker and Robinson, Calculus 0/ Observations, Blackie
& Co.).
Suppose that/(x) is any continuous l function of x. Take any
number Xl' and define the sequence (x,,) recursively by the rule
x"+1-/(x,,), for every n. Thus X2 =/(Xl ), X3 =/(X I ), etc. Suppose
that x" tends to a limit a, as n---+oo. Then /(x,,) must tend to
!(a), I but on the other hand, /(x,,) =X"+1' and (x"+1) has the
same limit as (x,,) (Example 5, p. 14). Therefore a=/(a). In
other words, if x" tends to a limit, then this limit is a root 0/ the
equation x-/(x).
The sequence (x,,) does depend on the starting value Xl' If we
take a different Xl' the resulting sequence might tend to a differ-
ent limit b (which would, by the same argument as before, be
another root of x=/(x)), or it might have no limit at all. It is
necessary to devise a working procedure which can guide us in
choosing a 'good' Xl'
It will usually be possible to fix the positions of the roots
approximately (e.g. by drawing a rough graph). Let/'(x) denote
the derivative of/(x). We have then the following
Criterion. Suppose (t) that there is known to be a root a 0/ x=/(x)
somewhere between X=OC and x=p, and (it) that there exists a posi-
tive number k<1 such that 0 </'(x) <k,for all X between oc and p.
Then the sequence (x,,) defined by Xl =OC, x"+1 =/(x,,), tends to a as
n---+oo. Furthermore, the difference between the nth term x" and
a is less than k"-lIP-ocl,a for any n.
1 See e.g. P. J. Hilton, Differential Calculus, in this series.
I We are using here the fact that 1(;,;) is a continuous function. The
characteristic property of a continuous function can be expressed in just
this way: that if a sequence (;';n) tends to a limit a, then the sequence
(1(;,;,,)) of function values must tend to I(a). All the ordinary functions
of elementary mathematics are continuous.
I We remmd the reader that I {J-Ilt I is the numerical value of the

20
 SOLUTION OF EQUATIONS BY ITERATION
Example 1. Let f(x)=V2X. The equation x=f(x) can be
solved easily in the ordinary way; its roots are x=o and x=2.
The derivative is !'(X)=!.2.X- i =I/V2X, so that o</'(x)
<1/v2, for all x between 1 and 3, say. We can apply the
Criterion (with a=2, CX=I, (J=3, k=I/V"Z). It shows that the
sequence defined by Xl =1, xn+l =V 2X" (some of the terms of
this sequence are given in Example 5, p. 2) tends to 2 as
n~C().

Proof of the Criterion. For the purposes of this proof, we shall take
a.<p. The case P>a. is easily treated on exactly the sanIe lines. We
assume that f(x) can be differentiated, and that its derivative /,(x)
is continuous. For such a function the 'mean value theorem' of the
calculusl avplies. This says that if x and a are any numbers, then
f(a) -1(x) = (a-x)/,(E), where E has some value between x and a.
If a and :& are both between a. and p, then so is E, and hence
o</,W <k. Therefore if x<a,
o<f(a) -f(x) a-x)k (I)
Put x=x1( = a.) in (I), remembering that f(x 1) =:&2 and that
f(a)=a. This gives 0<a-x.a-x1 )k. Next put :&=x. in (I),
remembering thatf(x.) =x.; we have o<a-xa < (a-x.)k. Since the
previous inequality tells us that a -x.< (a -x1 )k, this yields
0<a-xaa-x1 )k. Continuing in this way, we find
o<a -x,,< (a -x1)k n - 1,
for all n. Now a-x1 <p-a. (see Fig. 10), so that

i i

X.=oc '13

o<a -x" < (P _a.)k n - 1,

for all n. As n~oo. k"-1~O, because k<J. Therefore x" must
tend to a.
Working procedure. Suppose the equation to be solved is
given in the form F(x) =0. This can be put in the form x=/(x)
difference fJ- a., disregarding the sign. Notice that it is not assumed that
tII.<fJ
1 See e.g. P. J. Hilton, Differential Calcull/s, in this series.
21
 SEQUENCES
in many ways, e.g. by takingf(x)=x+cF(x), with any non-zero
constant for c. We must (I) locate the root a required as nearly
as possible, i.e. find ex and (J between which a lies, and (2) choose
c so that the Criterion applies tof(x) =x+cF(x). This technique
is illustrated below.
Example 2. Calculate {Ii The equation to be solved is
X8 -2=0, equivalent to x=f(x)=x+C(x3 -2). (I) To locate the
root. We soon find that a={I2lies between ex=12 and {J=13,
for (12)3=1"]28<2, while (13)8=2197>2. (2) Choose c so
that o <J'(x) =1+3CX2 <k(k<1), for all x between 12 and 13.
A suitable value is C=-02, for then f'(x)=1-o6x 2, which
decreases from 0.136 to 0026 as x varies from 12 to 13. Thus
o <.f'(x) <.k=0136, for all x between 12 and 13.
According to the Criterion, a={12 will be the limit of the
sequence defined by Xl =12, x"+1 =f(x,J =X"-02(X,,3_2). The
first four terms are Xl =12, X2 =12544, Xa=12596, X,=12599.
The accuracy of these approximations can be estimated from
the formula 0 <a-x" <({J-ex)k"-1=(01)(0136),,-1 (see p. 21).
In particular, x, differs from a={12 by less than (0.1)(0.136)3,
which a rough calculation shows to be less than 00003. In this
way we have proved that {12 is between X,=12599 and
x,+00003=12602, which guarantees that the first four signi-
ficant figures of {lz are 1260. Greater accuracy can be secured
by taking larger n.

EXERCISES ON CHAPTER I
I. Write down the first five terms of each of the sequence defined
below (use decimal notation).
(i) a" =1 -(02)". (ti) a" = 1 -( -02)". (iii) a" =(,,1 + 1)/(n+I).
(iv) a,,=Vn+1 -Vn. (v) a,,=( -I)"+ln. (vi) a.. = (sin nn/2)".
(viI) a"+1 =3/a", al = -I. (viIi) 2a,,+z = a"+l +a,,(al =I,a z =2).
1
(ix) a"+l=a,,+(i)", al=l. (x) a"+l=-(a1+a.+ +a.),
n
al=l.
z. Make graphs (see 3) for the sequences (i) to (vi) of Exercise I.
22
 EXERCISES ON CHAPTER I
3. State, in the cases (i) to (vi) of Exercise I, whether (a,,) tends
to a limit A, or to + 00, or to - 00, or if (a,,) oscillates.
4. How large must n be, for (t)" to be less than (i) 0'01, (h) 1O- 6?
5. How large must n be, for (1'1)" to be greater than (i) 108,
(ii) 106?
6. Find lim a", for each of the following a". (Note: lim a" may be
+ 00 or - 00). (i) 2-(0'2)". (ii) 2+( -0'2)". (hi) (n+I)/(n2+1).
(iv) (4+n)/(3n-2). (v) (2+n+n2)/(1 +4n+5n2).
(vi) (I +n3)/(2-3n3). (vii) (nl+I)/(n'+I).

(viii) (-I)"n/(nl+2). (ix)2"+( _~)"I. (x) { -2+;+(~)"r.

(xi) (3"+1)/(2"+1). (xii) (2"+1)/(2"+1+1). (xiii) 2"n2/5".
(xiv) (I +2")/(1 -n). (xv) (n-2")/(n'-3").
(xvi) (5"n l +I)/(5"+1n+I).
7. Find a number N such that n l /2"<0'OOI if n>N.
8. For what values of b is a,,=(n+b)/(n+l) increasing?
9. Show that (b,,) = (cos nx) does not tend to a limit as ~ 00,
unless x is an integral multiple of 2n radians. (See Example 6,
p. 15)
10. Let a" = V;(x > I). Show that (a,,) is (i) decreasing, and (ii)
bounded below, and hence prove that a,,---+a limit A as n---+ 00.
II. Find the limit A of Exercise 10, by considering the sequence
(a~n).
12. Apply the method of Example 2, p. 22 ( II) to find the (real)
root of the equation X 3 +X-I ~o.
 CHAPTER TWO
Infinite Series

I. FINITE SERIES

A 'series' is the sum of the terms of a sequence. For example,

if "I' "2' ... ,"Nisafinite sequence, the sum "1+"2+ ... +"N
is the corresponding finite series. A convenient notation for
N

"1+"2+ ... +"N L "", is

,,~1
For instance, we could write
N .
12+22+ ... +N2 L 2. as n (Notice that we could write it
,,=1
N N
L L
equally well as
,=1
1'2, or
.=1
S2; the 'summation variable' (n, or

or s, respectively) is sometimes called a 'dummy variable',

1',
because it does not appear when the series is written out in
1'1
full.} More generally, L ""
,,=M
denotes the sum "M+"M+1
+ ... +"N'
Some special finite series will be useful to us later on.
Example 1. If the general term "" of a finite series can be
expressed in the formf(n}-f(n+I}, for some suitable function
N

f(n}, then it is very easy to calculate L"",

,,=1
For "1+"2+ ...
+"N=(f(I}-f(2)) + (J(2)- f(3)) + (f(3)- f(4)) + '" +(f(N)
-f(N+I))~f(I}-f(N+I}.
For example, n=-ln(n-I} +In(n+I} =f(n}-f(n+I},
L n=I+2+ ...
N
where f(n}=-!n(n-I}. Therefore +N
,,-I
 INFINITE SERIES
N
Example 2. The sum of the series L: l/n(n+l} can also be
n=1
evaluated in this way. For "n=l/n(n+I}=(I/n)-(I/n+I}.
N
Therefore L: l/n(n+I}=I-_I-.
n=1 N+I
Example 3. The geometric series. The series I+X+X 2+ .
N
+xH= L: xn (notice that a term XO=I is included) is called
n=O
a (finite) geometric series.
Let 5=I+X+X 2+ . +xH- 1+xH,
then x5= X+X2+. . . +xH+XN+l,
and by subtraction, (l-x}5=I-xH+ 1. Therefore
5=(I-xH+l}/I-x
provided that X~I. If X=I, then 5=1+1+ ... +1 (there are
N+I terms here}=N+I.
Example 4. The binomial theorem of algebra is an 'expansion'
of the polynomial (x+y)Q in the form of a finite series. For the
theorem states that for any positive integer a,

(x+y}Q= i
n~O
(:)x nyo _n,

a(a-r) ... (a-n+I}f.....

-'----'------'-----'- 1 n .... 1, an
d (a) =1.
where (:) I. 2. n 0

2. INFINITE SERIES
If Ut, "2' "3' ... is an infinite sequence, with nth term Un'
then the 'sum' "1+"2+"3+ ... of all the terms is called an
CIO

infinite series, and we denote this series by L: Un' For ex-

n=1
CIO

ample, L:~=I+i+l+l+ ... Often it is convenient to

n=1
S.S.-c 25
 INFINITE SERIES
start a series with some term UM other than Ul, and we write

L:
CIO

then Un =UM+ulIf+l+ UM+2+ In particular, we may

n=M

include a term U o at the beginning, for example

,,=0
L: <0

xn=xO+xl

+X2+ =1+X+X 2 + ...

When it is clear from the context
where the series starts, the abbreviated notation LUn is often
used.
Evidently we cannot just 'add up' an infinite number of
terms in the ordinary way, and in fact it is not obvious that this
kind of sum has any meaning at all. The following two examples
show how an infinite 'sum' might arise.
Example 1. A recurring decimal is one of the most familiar
infinite series. For instance, the statement that 1=0'j=0'3333
... really means that 1 is the sum of the infinite number of
terms 0'3+0'03+0'003+0'0003+ .. , which we could also
express in our notation
n-l
L:
CIO

3/(IO)n.

Example 2. Achilles and the tortoise. In this famous problem,

it is supposed that a race is staged between a tortoise, on the
one hand, and Achilles on the other. Achilles can run m times as
last as the tortoise (m>I), so the tortoise is given a start, say
of one minute. Thus the distance between the positions Al and
TI of Achilles and the tortoise, respectively, at the moment
when Achilles begins to run, is the distance which the tortoise
can cover in one minute; we may as well take this as our unit of
length. It takes Achilles only I/mth of a minute to reach TI
(Achilles' position AJ, but the tortoise has by now advanced

A,
I

I
FIG. II
26
 CONVERGENT AND DIVERGENT SERIES
a further distance 11m, to T 2. Achilles covers the stretch T1T2
in I/m2th of a minute, but the tortoise is now at Ta, a distance
11m2 away. The ancient paradox was this: How can Achilles
ever overtake the tortoise, since he reaches each new position
T .. only to find his opponent leading, by the small-but positive
-distance 11m"?
The answer is that the whole infinite sequence of stages T oT 1 ,
T IT 2, T 2T a, .. takes place in afinite time, or equivalently, the
total distance ToT1+TIT2+T2Ta+ . is finite, even though
it is composed of an infinity of parts. At the end of this finite
length a point U is reached where the runners are level. After
that, Achilles takes the lead. The distance ToU is given by the
infinite sum 1+{l/m)+{l/m)2+ , and we shall see in 5
how to evaluate this.

3. CONVERGENT AND DIVERGENT SERIES

L:
<0

If we want to find the 'sum' of the infinite series tI.

,,=1
(assuming for the moment that such a thing exists), it is natural
to begin by working out the 'partial sums' SI =ul , S2=U1+u2,
Sa=fC]. +u2+ua, and so on. This gives a sequence of partial sums
SI, S2, sa, .. , whose nth term is S,,=U1+U2+ .. +u,,; we can
regard these partial sums as approximations to the 'full' infmite

L: u...
<X>

sum If this sequence of partial sums does tend to a limit

,,-=.1
S as n-+oo, then S is what we take as the sum of the infi"ite

L: u..=U1+U2+Ua+ ..
CIO

series If, however, s" does not tend

,,=1
to a limit, we must take it that the sum of the finite series

L:,,-1 u"
CIO

simply does not exist. Series of the first type are called

convergent series, and those of the second type, which have no

sum, are called divergent series. We repeat these definitions;
they are essential for all the theory which follows.
27
 INFINITE SERIES

L U,,=Ul +U2+U3+
00

I. The series is convergent if the

n-l
sequence 51' 52' 53' . of partial sums (the nth partial sum
being sn=ul +u2+ ... +u,,) tends to a limit S as n---+oo. The
number S is then called the sum of the series.
L UtI is divergent if the sequence of partial
00

II. The series

,,-I
sums does not tend to a limit as n---+oo (that is, if 5,,---++ 00, or
if 5,,---+- 00, or if 5" oscillates).
4. SOME EXAMPLES OF INFINITE SERIES
By far the most important, as well as one of the simplest
series, is the infinite geometric series (i) L X"=I+X+X 2+X
00

,,=0
8

+ ... (x is a constant). The nth partial sum (i.e. the sum of

the firstn terms) iss,,=I+x+x 2 + ... +xn- l =(I-xn)/(I-X)
(ifx~I), or s,,=n, if X=I (see Example 3, p. 25). If -I<X<I,
we know, from the last chapter (see ro, p. r8) that X"---+o as
n---+oo, and therefore s,,=(r-x")/(r-x)---+r/(r-x) as n---+oo.
Hence
L xn is convergent, and its
00

If -I<X<I, the geometric series

,,=0
sum is I/I-X.
If x has any other value, except X=I, then xn does not tend
to a limit as n---+oo. Therefore 5" does not tend to a finite limit
as n---+oo, which means that the series is divergent. It is also
divergent in the case X=I, for then s,,=n, which ---++ 00.
Hence
Lx" is divergent.
00

If X<-I, or if X>I, then

,,=U

Example 1. The recurring decimal 0'3=0'3+0'03+0'003+

... =0'3{1+ (0'1)+ (0r)2+ ... }. The series inside the brackets
converges and has sum r/(r-oI)=r/o9=lo/9. Therefore
0-3=(0'3) x)lIl!.=i
28
 SOME EXAMPLES OF INFINITE SERIES
Any recurring decimal can be expressed as a fraction by this
technique: e.g. 0i35=0I35I35I35 ... =0I35{I+(000I)+
(0.001)2+ ... }= 135 X I = 135 XIOOO=I35=~.
1000 1-0001 1000 999 999 37
Example 2. In the problem of Achilles and the tortoise, we

L(I/m)".
CIO

met the series I+(I/m)+(I/m)2+ . = This is

,,=0
convergent, because I/m<I, and its sum is I/(I-(I/m
=m/(m-I). This is the distance ToU from the start to the
point U where Achilles overtakes the tortoise.
CIO

(11.. ) ~
~ I/n (n+I )=-+-+-+
I I I ... IS. another senes
.
,,=1 12 23 34
whose partial sums can be easily evaluated. For we saw in
I I I
Example 2, p. 25, that s,,=-+-= +,----:----:-
12 23 n(n+I)
1 __
1_. Therefore S,,--+I as n--+oo, and it follows that the
n+I
L I/n(n+I) is convergent, and its sum is
CIO

infinite series I.
,,=1

L I/n=I+i+1+1+ ...
CIO

(iii) is sometimes called the

,,=1
'harmonic series'. There is no simple formula for the partial
sum s,,=I+l+1+ ... +I/n, nevertheless we can prove that
the harmonic series is divergent, as follows.
Consider ss=I+i+(1+1)+(!+I+t+1), which we bracket
in the manner shown. Clearly (l+1l+l=i, and (!+l+t
+11+1+1+1=1. Therefore ss>I+l+l+l=t. Similarly
s18=ss+(-t-+Io+A-+ + /8ss+(le+h+h+ ... +/8)
>!+l=~, and we find that S32>t, S8'>!. and, in general,
S2'> k+2. This shows that we can make S2" as large as we like,
2
by making k large enough; in other words the subsequence
S2' s" SS, ... of (s,,) tends to + 00. By Rule 5 (p. 14), (s,,)
29
 INFINITE SERIES
cannot tend to a finite limit, for if it did, any subsequence of

(s,,) would tend to the same limit. Therefore

n~l
L:
ex>

lin is

divergent.

5. SOME RULES FOR CONVERGENT SERIES

We give next some simple rules.

L: L:
ex> ex>

Rule 6. If u", v" are two convergent series, with sums

,,~1 ,,=1

L:
ex>

S, T respectively, then (u,,+v,,) converges and its sum is S+T.

,,-1
L:
<0

,,-1
(u,,-v,,) converges and its sum is S-T, and if c is any

constant, then
"~1
L:
CIO

cu" converges and its sum is cS.

Let S" =U1 +u. + ... +tI". and let t" =v1 +va + ... +v". Then,
for example. s" +t" is the nth partial sum of L""
,.=1
tI" +v", Since

s,,-+5 and t,,-+T as n-+oo, it follows that s,,+t,,-+5+T, and

L:
ex>

therefore (u" +v,,) converges. and has sum 5 + T. The other

n=1
cases can be established in a similar manner.

L:
CIO

Example 1. The series x"=x+x 2 +x3+ . can be ob-

,,=1

tained by multiplymg the series L:

CIO

,,=0
x"=I+X+X 2 + by

the constant x. Therefore its sum is x.(_I_), if -1<X<I.

I-X
30
 SOME RULES FOR CONVERGENT SERIES

2>" from Lx", we find, in the

CIO CIO

Example 2. By subtracting
,,=0 ,,=0

case that x, y arc both between -I and I, that L (x"-y")

CIO

"=0
is

convergent and has sum

(I/I-X)-(I/I-Y) =(X-y)/(I-X)(I-y).

L L
CIO CIO

Rule 7. If u,,' v" are two convergent series, with sums

,,=1 "=1
S, T respectively, and if u" <v" for all n, then S <T.
For it follows that s"<t,,, all n. Therefore 5=lim s,,<lim t" =T.

Rule 8. Let L
,,=1
<0

u" be a convergent series, with sum s.

(t) If a finite number of terms are removed from the series, then
the resulting series is still convergent, and its sum is S-P, where
P is the sum of the terms removed.

(it) If a finite number of terms are added to L U",

n=1
CIO

then the

resulting series is still convergent, and its sum is S+P, where

P is the sum of the terms added.

A sketch of the proof of (i) is as follows: let Lu,.'

CIO

,,=1
be the new

series, and let s,,' be its nth partial sum. It is clear that s; + (sum
of the removed terms) =a partial sum, say s,,+lI:' of the original
series. Since (s,,+1:) ='.,Hlo s,Hlo . is a subsequence of (s,,), it has
the same limit 5 as (s,,), That means that s,,' +P-+5, whence
s,,'-+5 -Po (il) can be proved similarly.
Example 3. The series X+X2+ . of Example I can be ob-

tained by removing the first term of the series L

71=0
<0

x"=I+X

+X2+ ... This gives an alternative proof that the sum is

(I/I-X)-I=X/I-X.
31
 INFINITE SERIES
Remainder of a series. It is useful to have a name for the infinite

L:
CIO

series u.. =uN+1+ UN+2+, which starts after the

.. =N+l

L: u,,; we call it the remainder after N

CIO

Nth term of the series

,,~l

terms. (An alternative notation, which is quite equivalent,

L:
CIO

would be UN+,,') Rule 8 tells us that if the original series

,,=1

L: u" L: u"
CIO CIO

is convergent, then the remainder is also con-

,,=1 ,,~.V+l

vergent, for it is obtained by removing the first N terms of

L: U", Furthermore, the sum RN=UN+l+UN+2+ ... is equal

<0

,,~l

to 5-(U1 +U2 + ... +uN)=S-sN' We have therefore the

formula
S=sN+R N,
which holds for all N. It is often possible to find an 'estimate'
for RN even when we cannot calculate its value exactly, and
this is essential if we wish to compute S numerically; for ex-
ample if we can prove that R6 lies between 0 and 0'001, then
we know that S is between S6 and S5+0001. In other words, we
can get to within 0'001 of the full sum S, by using the partial
sum S6' This question will be discussed in Chapter 3.

Example 4. If L: u..
<0

,,=1
is convergent, then Ry-+o as N-+oo.

For we have RN=S-SN for every N, and sy-+S as N-+oo

by the definition of S. Therefore Ry-+S-S=o.

6. A TEST FOR DIVERGENCE

The following very simple test can sometimes be used to

show that a series is divergent.
32
 THE COMPARISON TEST

Test 1. A series
n=1
L
<X>

Un is divergent if its nth term Un does not tend

to zero as n~oo.
For if sn=u1 +u2 + . +Un, then we have un=sn-sn_l'
L
ex>

If the series Un were convergent, then (sn) would tend to a

n=1
limit S, hence (sn-1) would also tend to S. (Note: sn-l is not
defined for n=l; we have to think of this sequence as starting
with n=2.) Therefore un=Sn-sn_l~S-S=O. It follows that
if tin does not tend to 0, then LUn is not convergent, i.e. it is
divergent.

L
ex>

Example 1. n/n+l=i+i+l+ .. diverges, because

n=1
un=n/n+I~I.

Example 2. Lex>

n=O
(_I)n diverges, because Un=(-I)" does not

tend to a limit at all.

Example 3. Lex>

n=1
2 n /n18 diverges, because Un =2n /n 18 does not

tend to zero as n~oo (it tends to +00, see p. 18).

Warning. Test I can never prove that a series is convergent.
It is possible that un~o even for a divergent series, for ex-

L
ex>

ample we saw ( 4 (iii), p. 29) that I/n is divergent,

n=1
although the nth term I/n~o as n~oo.

7. THE COMPARISON TEST

One of our main tasks is to find 'convergence tests'; that is,

tests which enable us to decide whether a given series Lex>

n=1
Un is

convergent or not (without necessarily finding its sum). Test I

33
 INFINITE SERIES
is a very easy preliminaiy test, but it is of no use for proving
that a series is convergent; it can only tell us that certain series
are divergent.
In this paragraph, and in 8, 9, we shall be mainly con-
cerned with series whose terms are positive (or at any rate non-
negative, i.e. positive or possibly zero); we reserve the notations
p", q" for the terms of such series. The nth partial sum SrI =Pl
L: p,. of positive terms is an
00

+P2+ ... +p" of a series

,,=1
increasing sequence. For S,,+1 =S,,+P"+l >s", all
n. Therefore,
by the Fundamental Theorem on p. 16, (s,,) either tends to
a finite limit or to + 00. This fact makes series of positive terms
easier to deal with.

L: p", L: q" be two

00 00

Test 2. Comparison Test (first form). Let

,,=1 .. =1
series of non-negative terms (i.e. p,,>o, and q,.>o, for all n).
Then
(i) If p" <.q"for all n, and ij"Eq" converges, then "Ep,. converges
also.
(it) If p" >q"for all n, and ij"Eq" diverges, then "Ep,. diverges
also.

As we have said, the sequences (S,.) and (t,,) of partial sums

srI =Pl +Pa + ... +p", ttl =ql +qa + ... +q" are both increasing.
(i) If I:q" is convergent, it means that t,.---+ a limit T. All the ttl
are < T. But p,,<q,. for all n, hence s,,<t,,<T, for all n, i.e. (5,,) is
bounded. Therefore (5,,), being a bounded increasing sequence, must
tend to a limit, S, say (see p. 16), and so I:p" converges. Inci-
dentally the sum S of 'I:p" is < the sum T of I:q", by Rule 4, p. 12.
(ii) If I:q" is divergent, the sequence (t,.) must tend to + 00.
However we are given that p,,> q", for all n, hence 5,,> ttl for all n,
and thus srI must tend to + 00 as well. Therefore I:p,. is divergent.

L: llYn. We 'compare' with L: lin, which is

00 00

Example 1.
,,=1 .. =1
divergent. Since I/Yn>lln, for all n (because Yn<.n), part
34
 THE COMPARISON TEST

L: llYn diverges as
00

(ii) of the Comparison Test shows that

n-l
well.
L: lin' diverges, if s -<;1. For
00

Example 2. In general, the series

n~l

if S-<;l, then n'<n, for all n, hence Iln'>Iln, for all n. Thus
Llln' diverges, as before.

L: IIn 2 is convergent. The idea is to compare this

00

Example 3.
n-l

with the series L: Iln(n+I}, which we know ( 4

00

p. 29) (ii),
n=l
is convergent. A direct comparison of nth terms is unsuccess-
ful (for Iln 2>Iln{n+I)}, but on the other hand I/{n+I)2<
I/n{n+I}, for all n (because (n+I)2={n+I)(n+I) >n{n+I)),
and thus by part (i) of the Comparison Test, the series
00

L:I/(n+I}2=I2+';+ 12+ ...... is convergent. Our series

n=l 2 3 4
L: Iln =I+ +.;+..;
00

.. =1
2

3 4
+. . . is
2
12 obtained from this
adding the single extra term I, and so it converges also (Rule 8).
00

This does not, of course, tell us what the sum S of L: I/nl is,
n=l
but only that it exists. \Ve can however notice that
00 00 00

S= L:I/nl=1 + L:I/(n+I)2<1 + L:I/n(n+l) =1 +1==2.

ft=l ,,,=1 11==1
See also I, Chapter 3 (p. 60).

L: lin' is convergent, ifs>2. For if S>2, then

00

Example 4.
,,=1
n' >n 2 for all n, hence lin' <Iln2, for all n. By the Comparison
35
 INFINITE SERIES
Test, since Lr/n2 is convergent, then Lr/n' must converge as
well.

L: r/(3"+r).
00

Example 5. It is clear that r/(3"+r) <..r/3"

,,=0

L: (l)" converges,
00

= (l)", for all n. Since the geometric series

n-O
L: r/(3"+r) .
00

so does
.. =0
The following modification of the comparison test is often
easier to use than the first form, although it is less general.

L: P.., L: q.. are

00 00

Test 3. Comparison Test (second form). If

.. =1 .. =1
two series of positive terms, and if P../q..-+- a non-zero (finite)
limit C as n-+-oo, then either LP.. and Lq.. both converge, or both
diverge.
By the definition of a limit (p. 5), we know that, given any
positive h, there exists an integer N /I such that P.. /q .. lies between
C-h and C+h, if n>N/I. Take for h any number less than C
(e.g. iC), and write N/I=N, for short. Suppose that l:q.. is con-
vergent. We have P.. /q .. <C +h, hence p"C+h)q,,, for n=N +1,

L: q..
00

N +2, Now is convergent (Rule 8), therefore

,,=N+l

L:
00

(C +h)q" is convergent, and so, by the Comparison Test (1st

.. -N+l

L: P..
00

form), is also convergent. It follows (by Rule 8(ii)) that

.. -N+l

L: p"
00

itself is convergent. Similarly, if I:q.. is divergent, we can

t/=1
prove thatI:p" is divergent as well, by comparing itwithI:(C -h)q".

Example 6. This test makes the convergence of Lr/n2 much

easier to prove. Take p,,=r/n 2 , and q,,=r/n(n+I). Then
36
 THE COMPARISON TEST

p.Jq,,=n(n+r)/n2=r+~-H as n----+oo. Since 1:q" converges,

n
so does "i:.P".
Series "i:.P", where p,,=(cllnll + +c1n+co)/(dknk+
+d1n+do) is a rational function of n, are easily handled by
Test 3. We have only to take q,,=nll-k; then P.Jq.----+;: (see
p. 12).
00

Example 7. L (n 3-r)/(2n4+1l+r). Take q,,=n3-'=n-1,

,,=1
which we know gives the divergent series "i:.n- 1 ="i:.r/n. Then
p,,/q,,=n(n 3-1)/(2n'+n+l)----+! as n----+oo. Therefore 1:p..
diverges also.
00

Example 8. L (3n2+n+l)/(n6+r). Take q.. =n 2 - II =r/n 3.

n~1

From Example 4 above, 1:q" converges. We find that

P,,/q,,=(3n5+n'+n3)/n6+r)----+3 as n----+oo. Hence by Test 3,
"i:.P" is convergent.
It will be noticed that Test 3 could not be used, for example for
00

the series L(4-n)/(n8+I), in which all the terms after the

,,=1
fourth are negative. However it is possible to mOdify Test 3 by
allowing one of the two series involved to have negative terms, as
follows:
Test 3'. If :Eu" is any series, and if :Eq" is a series of positive terms,
and if u"lq,,----+ a non-zero limit C (which might be negative) as
n----+oo, then either :Eu" and :Eq" both converge, or both diverge.
The proof remains unchanged if C>O. If C<O, we simply con-
sider :E( -u,,) instead of :Eu,,; this does not affect the convergence.
Using Test 3' we may discuss any series whose nth term is a rational
function of n, even if its terms are not all positive. For the 'known'
series :Eq" can always be taken to be of the form :En', which has only
positive terms. Thus if u,,=(4-n)/(n8+I), take q,,=n l - 3 =I/n l
Then:Eq" converges, u,,/q,,----+ - I as n----+ 00, and therefore :Eu. also
converges.

37
 INFINITE SERIES
8. THE RATIO TEST

The next test depends, essentially, on a comparison of the

given series with a geometric series. We shall describe it first
for series of positive terms, and give later (p. 47) a vcrsipn
suitable for series which contain negative terms.

L p" be a series of
00

Test 4. Ratio Test (positwe terms). Let

.. =1
positive terms, and suppose that p,,+t!P,,~ a limit L as n~oo.
Then (t) If L<r, the series "i:.P" is convergent, and (ii) If L>r,
the series is divergent. If L=r, this test fails, and the question of
the convergence of"i:.P" must be investigated by some other method.
(i) Suppose L< I. By the definition of the limit (p. 5), it must
be possible, given any h, to find an integer N,. such that P"+tlP,,
lies between L-h and L+h, for every n>N,.. Take h=!(1 -L),
80 that L+h=t(1 +L) <I, and write N,.=N, for short. We have
PN+tlPN<L+h, PN+zIPN+l <L+h, etc. Thus
PN+n PN+3 PN+I
PN+n=p--" 'p-'p-.PN+l< L+h) PN+l'
( "-1
N+ ..- l N+I N+l
for all n. The series
00

LPN+l(L+h)"-I=P N+l{I+(L+h)+(L+h)I+ }
.. - I
is convergent, because L +h < I. Therefore by the Comparison Test
00

(1st form), L PN+.=PN+l +PN+z+ ... is also convergent. By

,,-1
L
00

Rule 8, the full series Pta is itself convergent.

,,-1
(il) Suppose that L> I, then we can find N such that
p,,+tlP. >L -h for all n >N; we take h to be !(L - I) this time, so
that L-h=!(L+I) >1. Now P"+lIP" > I, for n>N, implies that
P"+1 >P.. for n> N, i.e. (P .. ) is an increasing sequence. Since the
P.. are all positive, it is impossible for them to tend to zero.
Therefore ~P .. diverges, by Test I.

L
00

Example 1. rlnl is convergent. (n!=I.2.3 . (n-r).n

,,-0
 THE INTEGRAL TEST
if n is a positive integer, and O! =1, by definition.) Take
p,,=Iln!, then P,,+lIP,,=nl/(n+I)!=Iln+I----+o as n----+oo.
Since this limit L=o is less than I, the series converges. (Note:
p" is actually the (n+I)th term of this series, since we start
with a term Po. This makes no difference, for the limit of
(P"+lIP,,) is the same as that of (p,,+t!P"+l))'

I
00

Example 2. n 2x" (x>o). Here P,,+I/P,,=x(n+I)2In 2

,,=1
=x(I+~) 2--+ x, as n----+oo. Therefore Ln 2x" converges if X<I,
If X=I, the ratio test fails. However, for X=I the series be-
comes Ln2, whose nth term tends to + 00, and which therefore
diverges (Test I).

I
00

Example 3. x"ln2 (x>o). Here P,,+1IP,,=xn 2/(n+I)1

n-l

=X/(I+~r----+x as before. Therefore the series converges if

if X<I, and diverges if X>I. When X=I, the ratio test fails.
But in this case the series becomes Llln 2 which converges, as
we have seen (p. 35).

9. THE INTEGRAL TEST

Suppose that it is possible to express the nth term of a series

I
00

p" in the form p,,-4(n), wheref(x) is a continuous function

,,=1
defined for all X>I (and not just for integral values x=n), and
satisfying the following conditions:
(a) f(x) >0, for x >1, and
(b) f(x) is decreasing as x increases from I.

I
00

Example 1. The series lin' (5)0) is in this class. For

,,=1
p"=f(n), where f(x) is the function IIx'. Conditions (a) and
(b) are satisfied.
39
 INFINITE SERIES
The graph y =/(x) is a curve of the form suggested in Fig. 12.

Let F(t) = J :/(X)dX; this is the area under the curve between

2 3 n-I n )C

FIG. 12

X=I and x=t. Let S,,=Pl+P2+ ... +P.. , the nth partial sum

I
00

of the series P... The area of the shaded part of Fig. 12 is

.. -1
P2+Pa+ +P.. =S,,-Pl (for the width of each shaded
rectangle is I, and their heights are /(2)=P2' /(3)=Pa, etc.).
This area is less than the area of the curvilinear region F(n),
and F(n) is, in turn, less than the sum of the larger rectangles
in Fig. 12. Their heights are PI' P2' ... , P.. -l respectively.
Thus
S,,-Pl <F(n) <S"_I'
The sequence (F(n)), like the sequence (s ..), is an ncreasing
sequence. It is possible that F(n) tends to a finite limit A, as
n---+ 00. This limit A would represent the total area under the
curve y f(x), to the right of the line X=I.
" I
Example 2. In the case f(X)=I/X 2, F(n)=
As n---+oo, this ---+1. Therefore the area A
J 1
dX/X 2 =1--.
in this case is I.
n

4 0
 THE INTEGRAL TEST

On the other hand, if we takef(x) =I/X, we get F(n) = J: dx/x

= log.n, which ----++ 00 as n----+oo. This means that the area
under the curve Y=I/X to the right of X=I is infinite.
When this area A is finite, we have s,,-Pl<F(n) <A, for all
n, so that s"<Pl+A or all n, i.e. the sequence (S,.) is bounded
above. Since it is also increasing, it must tend to a limit 5, and
L: p,. is convergent. Incidentally it follows by
00

that means
n~l

taking limits of the terms in the inequality sn-Pl<F(n) <sn-l

(see above) that 5-Pl <A <5, i.e. 5 lies between A and A+Pl'
If, on the other hand, F(n)----++ 00 as n----+oo, we see from the
inequality sn_l>F(n), or equivalently s,.>F(n+I), that

L: p,. diverges.
00

s,.----++ 00 also. In this case, therefore,

,.=1
We have now established the following

L: Pn can
00

Test 5. (Integral Test). If the nth term P.. of the series

n=l
be expressed in the form p,.=f(n), where f(x) is a continuous
fttnction of x satisfying conditions (a) and (b) (p. 39) then
(t) 2: p,. converges if Inf(x)dx tends to a finite limit A as n----+
00

.. =1 1
00,

and the sum 5 of the series lies between A and A +Pl' while
(it) L: p,. diverges if Inf(x)dx tends to +
,.~1
00

1
00 as n----+ 00.

Example 3. Consider the two functions I/X 2 and I/X (Ex-

ample 2, above). Since J:dX/X~I as n----+oo, we deduce that

L: I/n? converges and that its sum 5 lies between

00

I and
,.=1

I+Pl =2. On the other hand, J: dx/x----++ 00 as n----+oo, and

5.5.-D 41
 INFINITE SERIES

2:,,-I I/n diverges. (We have of course already proved

00

therefore

these facts by independent methods.)

2: I/n' (S>O).
00

Example 4. More generally, take the series

,,=1
Put f(X) =I/x', which satisfies conditions (a) and (b). The
integral f"
dx/x' has the value ~I __I_} (we are assuming
1 s=Il.
n'-l
here that S~I; the case S=I has been dealt with in the previous
example). Now I/n'-~o as n~O(), if S-I>O (see p. 7),
while if S-I<O, then I/n'-~+O(). Therefore "EI/n' con-
verges if S>I, and diverges if S<I.
To summarize our knowledge of this type of series (see also
Example 2, p. 35)
L: I/n'
00

converges if S>I, and diverges if S <I.

,,=1

Example 5. L:
00

Vn 3 +2n 2 +n/-'1+7nll . Let p" be the nth

,,=1
tenn of this series. We can use the principle employed in
Examples 6, 7, 8 (p. 37). The quotient of the leading terms of
the numerator and denominator is vnaI-'7nll = ;7 n-sI&.
Therefore if q,,=n- SI&=I/n 51&, we have P,,/q,,~I/-'7 as
n~O(). The series "Eq" is convergent, by what we have just
proved, and hence "Ep. is also convergent, by Test 3.
Example 6. Euler's constant. Sometimes the method used in
proving the integral test can give useful information even about
divergent series. For example. consider the sequence (a,,) whose
nth term is a,,=s,,-F(n). From the inequality s"_l>F(n)
(p. 40) we see that a,,=s"_l+p,,-F(np,,>o. for all n; i.e.
(a,,) is bounded below. We shall prove next that (a,,) is a de-
creasing sequence. For a,,-a"+l=(s,,-s"+l)-(F(n)-F(n+r))
=-Pn+l+(the area under y=f(x) between x=n and x=n+I).
In Fig. 13. the shaded rectangle has area P"+l' and it is clear
42
 SERIES WITH POSITIVE AND NEGATIVE TERMS

n n+1
FIG,I3

that this is less than the corresponding area under the curve.
Therefore a,,-a"+l>o, i.e. a,,>a"+l' for all n. It follows from
the Fundamental Theorem that (a,,) tends to a finite limit.
In the casef(x) =r/x, this limit has the value 0'5772 ... and
is denoted by i' (Euler's constant). Thus
(r+1+1+ .. +r/n)-log.n~=o'5772 ... , as n----+oo
This means that, although we can give no simple formula for
the finite sum r+1+1+ ' .. +r/n, it is approximately equal
to log.n+o'5772 ... , for large values of n.

ro. SERIES WITH POSITIVE AND NEGATIVE TERMS.

LEIBNIZ'S TEST

We have dealt so far mainly with series "Lp", whose terms p"
are all positive. For such a series, the sequence (s,,) of partial
sums is an increasing sequence, If, however, we take a series
00

such as r-1+1-1+ = I(-r)"+I/n, which has both

,,=1
positive and negative terms, then the sequence (s,,) will not be
increasing. The partial sums of the series mentioned start off
5 1 =r, s2=0'5, 5 3 =0'833 ... , s.=o'5 833 .. , s5=O'7 833 ' , .
In fact, this series is convergent, as we can prove with the help
of the following test.
43
 INFINITE SERIES
Test 6. (Leibniz's Test, or Alternating Series Test.) The series
co
L: (-I)n+la n=a1 -a9 +a3 -a,+ is convergent provided
"-1
(t) (an) is a decreasing sequence, and (it) an~O as n~oo. (A
series whose terms are alternately positive and negative is
called an alternating series.)
Because (an) is decreasing, an -an+1 > 0 for all n. Consider the
sequence (s2n-l) =SI' sa, S5' ... of the odd partial sums of
co
L: (-I)n+l an This is decreasing, since
"=1
S.n+1 -S2n-1 - -a 2n +a Zn- 1< o.
for all n, and bounded below, because
s.n-l = (a 1-a.) + (aa -ac) + ... + (a.n-a -aan-.) +a2n-1> 0
for all n. Therefore (S.n-1) tends to a linIit, 5', say. Similarly, the
sequence (s.,.) =s., sc, s, ... is increasing. since
s."+I-s.n =aZn+1 -aan+I>0,
for all n, and also bounded above, for
s.n =a1- (a.-aa) - ... - (a 2n-a -a 2n- 1) -a.n< ai'
for all n. Therefore (s.n) tends to a limit, say 5". Now
s.n-l -S2n = a.n,
and (a. n), being a subsequence of (an). tends to o. Thus
5' =lim s2n-1 = lim (s.n +a.n) =5" +0,
i.e. 5' =5". Denote by 5 the common value 5' =5". Then both the
even and the odd terms of the sequence (sn) =SI' S2. sa. sc, ... tend
to the same limit 5, and so (sn) itself tends to this limit. Therefore
the series converges.
52 S4 S5 S,
, , S.
i
', I )
i
0 S I
FIG. 14
co
Example 1. The series L: (-I)n+l/n=I-l+l-!+ ... is
n=1
convergent. For this is the case an=I/n, which certainly satis-
fies the two conditions (i) and (ii) of Leibniz's test. shall me
44
 ABSOLUTE CONVERGENCE
prove in the next chapter (Example I, p. 67) that the sum is
log. 2=0'6931 ....)

Example 2. 2 (-I)..+I/(2n-I)=I-l+i-t+ ...

co

.. =1
is con-

vergent, since (a ..) =(I/(2n-I)) also satisfies both the condi-

tions (i) and (ii). (we shall see in the next chapter (Example 3,

p. 68) that its sum is ~.)

II. ABSOLUTE CONVERGENCE

It is convenient at this point to recall the following defini-

tion: if x is any number, then Ixl (the 'modulus' or 'absolute
value' of x) is the numerical value of x, disregarding its sign.
For instance, 1-31 =3, 141 =4, 101 =0.
This can be expressed more formally by saying Ixl =x, if x> 0,
while Ixl = -x, if x< o.
Example 1. If R is a positive number, Ixl <R means the same
as-R<x<R.Similarlylx-al<Rmeansthata-R<x<a+R.
Example 2. IxYI=lxIIYI, for any two numbers x and y. This
generalizes to n factors, Ix1 x2 x ..1=lx111x21 ... Ix..l. In par-
ticular, putting X1 =X2 = ... =X.. =X, we have 1%"1=lxl".
Example3. It is sometimes useful to notice that Ix+yl <Ixl + Iyl,
for any x and y. In fact, Ix+yl =Ixl +Iyl if x and y both have
the same sign, but if they have opposite signs (and neither is
zero) then Ix+yl<lxl+lyl. The reader should check this by
trying a few cases. In general Ix1 +x2 + .. +x..1 <lx11+lx21
+ ... +lx..l, for any n numbers Xl' x2, , x...
2
co
Suppose now that Ufo is a series, some of whose terms Ufo
.. =1
co
are negative. We can make the series of absolute values 2Iu. l;
,.=1
the terms of this series are all positive or zero. Now it is possible
45
 INFINITE SERIES
that ~u.. is convergent, but that ~lu ..1 is divergent. An example
of this is provided by the series r-1+1- ... , which was
shown to be convergent, by Leibniz's test. However the series
of absolute values r+1+1+ ... is divergent (p. 29).
On the other hand, the following statement is true: If the

L: lu.. is convergent, then the series L:

00 00

series 1 u.. is also conver-

.. =1 ..=1
gent. A series ~u.. , for which ~lu..1 is convergent, is called an
absolutely convergent series. Our statement can then be phrased
as follows: Every absolutely convergent series is convergent. The
proof is given below. Sometimes a series which, like 1-1+1
- ... , is convergent, but not absolutely convergent, is called
a 'conditionally' convergent series.

L: L:
00 00

Suppose that u .. is a series, such that lu..1is convergent.

.. =1 .. =1
Write a.. =l{u.. +lu.. i), b.. =l{lu.. l-u .. ). If u .. >o, then
a.. = l{u.. +u.. ) =u..'
while b.. =l{u.. -u..) =0. If u .. <o, then a.. =l{u.. -u..) =0, and
b.. =l{ -u .. -u..) = -u ...
So in either case a.. and b.. are> 0. Further, Iu ..1 = a.. +b.. and
u .. =a.. -b... Since l::lu..1 is convergent, and a.. <a.. +b.. =lu.. I,
it follows from the comparison test (first form) that l::a.. is con-
vergent. Similarly l::b.. is convergent. Hence l::u.. =l::(a.. -b..) is also
convergent.

L: {-r)"+l/n2=r- r2+~-
00

Example 1. r2+ .. is abso-

.. =1 3 4 2
lutely convergent, since the series of absolute values is ~r/n2,
L: {-r)..+l/n
00

which is convergent. Hence 2 is convergent .

.. =1
(This could also have been proved using Leibniz's test.)
00

'" sm
Example 2. ~ ~ nx.IS convergent (for any fixed value of x)
.. =1

For let u.. ={sin nx)/n2, then lu..1=Isi., nxl/n2.

46
 POWER SERIES
Now lsin nxl <I, for any values of n and x, hence lu ..1 <I/n i ,
for all n. Hence ~Iu..lconverges, by the comparison test. There-
fore ~un is convergent as well. We have proved in fact that
~(sin nx)/n 2 is absolutely convergent.
co
Example 3. If a series 2:
n-1
Un is absolutely convergent, and

co co
if 2: lunl has sum
..=1
T, then the sum 5 of 2:
.. =1
u .. has absolute

value less than or equal to T. For let A, B be the sums of the

two series ~a .., ~b.. used in the proof above. The sum T
of ~ lu..I=~(a ..+bn) isA+B, and the sum 5 of~u .. =~(a .. -bn)
is A-B. Now A and B are both positive (or zero), hence
ISI=IA-BI=IA+(-B)I <A+I-BI=A+B=T.
co
Test 7. Ratio Test (general form). Let 2:
n=1
u .. be a series none of

whose terms u .. is zero, and suppose that IUn+l/u ..l~a limit Las
n~oo. Then (t) If L<I, then ~u.. is absolutely convergent (and
hence is convergent), while (iJ) If L>I, then ~u .. is divergent. If
L=I, the test gives no information about the convergence of~un'
If L < I, we have by Test 4 that l;lu..1is convergent. Hence l;Ufl
is absolutely convergent. If L> I, we must argue rather differently,
for the fact that l;lu..1 is divergent does not imply that l;u.. is
divergent (take e.g. Un=(-I)fI+l/n). However, if L>I, we can
apply the argument on p. 38 (with P.. = lu.. D to show that
lu..I~+ 00 as ~OO. Therefore un cannot tend to zero, and so
l;u.. is divergent by Test I.
Some examples of the use of this test are given in the next
section; its application raises no new difficulties.

12. POWER SERIES

co

A series of the form 2:

.. =0
c..x"=cO+C1X+CsX 2 + .. , involv-

ing the powers of a variable x, is called a 'power series'. The

47
 INFINITE SERIES
constants co, C1' C2, . are called the coefficients of the series;
they can be positive, negative or zero.
co
Example 1. 2 X"=I+X+X 2+ .. is a power series.
,,-0
So is

co x2 x8 x'
'" x"ln!=I+X+-!+,+-!+
~ 2 3. 4
.
,,-0
Let 2 c"x" be a given power series. A number R such that
co

n=O
the series is absolutely convergent when Ixi <R, and divergent
co
if Ixi >R, is called the radius of convergence of 2 c"x". For
,,=0
example, the radius of convergence of the geometric series
I+X+X2+ ... is I, for it converges if Ixi <I, and diverges if
Ixl >1 (p. 28). In a great many cases the radius of convergence
can be found by using the ratio test, as in the examples which
follow.

2 x"ln!. To apply the ratio test, take

co
Example 2. Consider
,,=0
u" =x" In! (we must remember that this will be negative, if
x<o and n is odd). Then IU"+llu"I=lxl/(n+I)~o as n~oo.
Therefore the series converges absolutely, for all x. In such a
case we say the radius of convergence is infinite. (This series is
called the exponential series, and its sum, written exp(x) or e",
is a function of x called the exponential function. See also
Example 2, p. 52.)
co

Example 3.2 n!X"=I+X+2!x +3!x + .. Here lu,,+t!u,,1

,,=0
2 8

=(n+I)lxl, which tends to +00 as n~oo, unless X=O.

Although this case 'L=+ 00' has not been covered in our state-
ment of the ratio test 7, it is easy to adapt the argument which
was used for the case L>I to prove that a series for which
48
 POWER SERIES

2: 1t!xn is di-
ct)

lim IUn+llunl =+ 00 is divergent. In particular,

.. =1
vergent for all x, except X=O, and therefore R=o for this
series. Naturally a power series with zero radius of convergence
has little interest for practical purposes.

Example 4. The logarithmic series. 2: {-r)n+lxnln=x-:

n=O
ct) 3

x 8 X'
+---+ .. IUn+llunl=l-xnln+rl=lxl{nln+r)~lxl, as
3 4
n~oo. Therefore the series converges if Ixl<r, and diverges
if Ixl >r. Thus the radius of convergence is 1. {We shall see
(p. 67) that the sum of this series is log.{r+x).)
Example 5. The binomial series. By putting y=r in the bi-
nomial theorem (p. 25) we obtain the finite series {r+x)1I

= ~ (=)xn. Now the 'binomial coefficient' (:) can be de-

fined, even when a is not a positive integer, by the usual for-

mu Iae ( a) a{a-r) ... (a-n+r) (if n>o, ) (a) =r. H ow
n 1.2 . . . n 0

ever, in this case the series I

n-O
(:)xn will be infinite, because

the coefficients (=) are all non-zero, unless a is a positive integer

or zero. This infinite series which we get by using a value of a
which is not a positive integer or zero is called the binomial
series. To investigate its convergence, we observe that IUn+llunl
=I(n~r)x / (n)I=I:~:IIXI~lxl, as n~oo. Therefore the

binomial series i
n=O
(:)xn=r+ax+ a{;~r)X2+ .. converges
absolutely if Ixl <r, and diverges if Ixl >r; in other words its
radius of convergence is r. {The sum of this series is in fact
(r+x)II, for any value of a. See p. 71.)
49
 INFINITE SERIES
Example 6. The sine and cosine series. The two series
"" X2 x' x8
(i)
2
n=O
{-I)nx2n/{2n)!=I-_+ ___ + ... , and
2! 41 61

~ {-I)n+Ix2n+I/{2n+I)!=x-_+
(ii) ~ x 8 x___
5
x' + ..
,,=0 3! 51 7!
are closely related to the exponential series. For (i) we find
IUn+Ilunl =x2/(2n+I)(2n+2)~0 as n~oo, and again, for (ii),
IUn+Ilunl =x2/2n(2n+I)~o as n~oo. Therefore each of these
series converges absolutely for all values of x, i.e. they have
infinite radius of convergence. (It is shown in text-books on the
calculus that the sum of the series (i) is cos x, and the sum of
(ii) is sin x; in each case x being in radian measure.)
'"
Example 7. It is possible to prove that every power series 2c"xn
n=O
has a radius of convergence R (which may be infinite). However it
may not be possible to find R by the ratio test. For example, the

series ~cnxn = 1+ (;) +x. + (;) 8+x' + (;)' + . .. would glVe

IUn+1/unl = mn+llxl, if n is even, and Iun+dunl = 2 nl xl, if n is

odd. It is clear that lun+dunl does not tend to a limit as n~ 00,
and therefore the ratio test cannot be applied. On the other hand,
the series converges absolutely if Ixl < I, by the companson test
(Test 2) (for Iuni < Ixl n , all n, and l:lxl n is convergent if Ixl <I),
while if Ixl > I, the nth term utI does not tend to zero and therefore
l:c"xn diverges (Test I). Thus this power series has radius of
convergence I.
Interval of Convergence. If R is the radius of convergence of the

power series 2""

n=O
c"x", then the interval -R<x<R is called

the interval of convergence of the series. The values x=R and

x=-R are called the end-points of the interval of convergence.
In general the ratio test gives us no information about the con-
vergence of ~cnx" at these end-points, and there is no universal
50
 MULTIPLICATION OF SERIES

4-converges absolutely...
diverges diverges
, : , )'
-R 0 R x
FIG. 15
rule; some power series converge at one end-point only, some at
both, and some at neither.
Example 8. The end-points of the interval of convergence of
the logarithmic series (Example 4) are %=1 and %=-1. At
%=1, the series becomes 1-!+1-1+ .. , which is conver-
gent (p. 44). Atx=-I we get -1-1-1-1- ... , that is, the
harmonic series with the sign changed throughout. This is
divergent.
13. MULTIPLICATION OF SERIES

2: u..=uO+U1+U2+ ... and 2: v.. =vo+v1+v.+

00 00

Let 0

n=O .. =1
be two infinite series. If we try to multiply these together we
get an infinite number of products of the form U,VI' which we
can arrange in an 'infinite square' as below.
"OVo "oV, "oVI "oVa "oV, o ,

",Vo ",V, ",V, "IV. ",V

"IV. ".v, "IV. ! ".v, ".v

r
/
/

"aV V1 ",VI "aV, o

",vyu,v, u,V. ",Va u,v,

FIG 16

We should expect the product {UO+Ul+U~+ . .)(VO+v1+v.

51
 INFINITE SERIES
+ ...) to be the sum of all these terms. To make this sum
appear as an ordinary infinite series, we arrange the terms as
follows: uOVO+(UIVO+UOV1)+(U2VO+ulvl+UOV9)+(UaVo+U2Vl

2: W", where w,,=u..vO+u,,_lvl+un_2VS

ct)

+U 1V2+uoVa)+ =
,,=0
+ ... +uov". The terms wo, Wl' Wa, ... of the 'product series'
~Wn are the sums of the terms on the successive 'diagonals' in
Fig. 16.

2: 2: v" are both abso-

ct) ct)

The Multiplication Theorem. If U,,'

,,=0 ,,=0
lutely convergent series, with sums S, T respectively, then the pro-
2: w.. defined above is absolutely convergent, and its
ct)

duct series
n-O
sum is ST.

2: x", 2: y" are both absolutely con-

ct) ct)

Example 1. The series

11=0 .. =0
vergent if Ixl and Iyl are less than I, and their sums are I/I-X

2: W,,' where
ct)

and I/I-y, respectively. The product series is

n~O

w,,=x1jI+x"-ly l+x"-IyS+ +xY". Thus I/(I-X)(I-Y)

=1+ (x+y) + (X S+Xy+y2) + (X 3+x2y+xyS+y3) + ... The
special case x-y gives I/(I-X)2=I+2X+3X2+4X3+
00

= ~ nx"-l.
,,=1
Example 2. Let exp(x) stand for the sum of the series

2: x"/nl (see p. 48). We have seen that this series converges

ct)

,,=0

absolutely for all x. Therefore exp(x} exp(y}=(~ x"/nl)

(~Y"In!) = ~ w", where
52
 MULTIPLICATION OF SERIES
xn xn-1y xn-Iys y"
W.. =,+( _ )1
n. n I.I 1+( _
n 2.2)1 1+ +-1
n
Remembering that the binomial coefficient (~) equals

nt;r!(n-r)!, we have w.. =~(x.. +(~)xn-ty+(:)X"-Iy2+ ...

+y") =~(x+ Y)". Therefore exp(x) exp(y) = ~ (x+ Y)"

n. ~ n!
.. =0
=exp(x+y). From this fundamental 'functional equation'
exp(x) exp(y) = exp(x+y) it is possible to deduce the properties
of this very important function exp(x) (d. P. J. Hilton,
Differential Calculus, in this series).
To prove the Multiplication Theorem, consider the nth partial
sums Sn' t,.. q.. of the three series l:u.., l:v.., l:w.., respectively. For
example, q,=WO+w1 +w.+ws+we is the sum of all the terms in
the 'triangle' above the diagonal line in Fig. 16. This triangle can
be divided into a square and two smaller triangles, as shown. The
sum of the terms in the square works out to be
(u o+lll +u 2)(VO+Vl +va) =ssts.
Therefore q,=ssts+A +B, where A =uOv3+uove+ulvs, and
B =uavo +uevo +uav 1.
We know that l:lu.. 1and l:lv.. 1are convergent, because we have
assumed that l:u .. and l:v .. are absolutely convergent. Suppose that
5* and T* are the sums of l:lu.. l, l:lv.. 1respectively, and let L .. = lu ..1
+ lu..+1 1 + ... be the remainder after n. terms of l:lu.. l, while
M n = Iv ..1+ Iv ..+1 1 + ... is the corresponding remainder of l:lv.. l.
Now (see Example 3, p. 45) IA I< IUollvsl + IUolivel + lu1l!vsI <
(111 01 + IU1ll(lvsI + IVel) < S*M s, and IBI < IUslivol + IUe!lvol
+ luSl!v11 < (ivol + Iv1l)(Iusi + luel) < T* Ls. Therefore Iqs -satal-
IA +BI < IA I + IBI <S*Ms +T*L s. In general, we find in exactly the
same way that Iq .. -smtml<S*Mm+T*L.... where m=!n, if n is
even, and m=!(n-I), if n is odd.
Now the remainders L m, M m tend to 0 as m-+- 00 (see Example 4,
p. 32). Therefore Iq .. -smtml-+-o as n-+- 00 (for naturally m-+- 00
as n-+- 00). i.e. q.. -smtm-+-o as n-+- 00. However, by definition
sm-+-S and tm-+-T as m-+- 00, and hence q.. =smtm + (q .. -smtml-+-
53
 INFINITE SERIES
ST +0 =ST as n~ 00. This proves that I:wn is convergent and
that its sum if ST, as required.
To prove that I:wn is absolutely convergent, we must show that
I:lwnl is convergent. Now we may apply what we have just proved,
to the product of the series I:lunl, I:lvnl; the resulting product series
I:W" is convergent, and W n = IUnllvol + lun-1llv11 + ... + IUollvnl.
But IWnl =lu"vo+Un-1V1+ ... +uovnl< W n , for all n. Therefore
I:lwnl is convergent, by the comparison test.

14. NOTES ON THE USE OF THE CONVERGENCE TESTS

The first thing we want to know about a series ~un is whether

it is convergent or not. The following notes are intended to give
some advice on choosing the convergence test most suitable to
the series on hand.
(1) As a rough guide, the ratio test should be tried whenever
un contains either (i) factors with n in the exponent, e.g.
un=nx"+l or un=(2n+n)/(5"+4n3) or (ii) factors containing
factorials of functions of n, e.g. I/n!, n2/(2n+I)1. Case (i) in-
cludes all power series. (This is not to say that the ratio test will
work in all these cases, but it is worth trying first.) Only the
general form (Test 7) need be remembered, because the earlier
version (Test 4) comes to exactly the same thing, when the
series ~un has only positive terms.
Warning. It is essential to find the limit L of IUn+l/unl first,
and then see whether L<I, >1, or =1. It is no use saying
'lun+lJ'Unl<I for all n'-this is not the same as 'L=lim IUn+l/unl
co
is <I'. For example, the series L I/n is divergent, yet
n~l

un+l/u,,=n/n+I<I for all n. In fact L=lim n/n+I=I, so

that the ratio test cannot be applied in this case.
(2) If Un is a rational function of n (p. 37) the ratio test is
always useless (it will give L=I), but, in compensation, the
comparison test (Test 3, or its modified form Test 3') will
always be successful. The same applies to any series ~Pn where
Pn is a quotient of roots of polynomials, e.g. Pn={'I+n+n 2/
v'I+n'+gn7. Such series 'behave like' the series whose nth
54
 EXERCISES ON CHAPTER II
term is the quotient of the leading terms in the numerator and
denominator of Pn; this quotient is (apart from a constant factor)
a power n', for some fixed s. So we take ~qn to be ~n' (this is a
'known' series; we found that ~n' converges if S<-I, and di-
verges if S:>-I, see p. 42). For example in the case mentioned
above, the quotient of the leading terms is {InilIV9n7 =In- 17' 8
Take qn=n-17/ 8 We can verify that Pnlqn-+l as n-+oo, and
therefore, since ~qn converges, so does ".pn (Test 3). See also
Examples 7, 8 p. 37, and Example 5, p. 4 2
(3) Although the ratio test will very often determine the
co
interval of convergence of a power series L cnx", it will not be
n=O
sufficiently sensitive to determine whether the series is conver-
gent at the end-points. Often Leibniz's test can decide this ques-
tion at one of the end-points.
(4) We have only used the integral test to investigate the
series ~lln' (see also Exercise I4 at the end of this chapter).
However the idea of comparing an infinite sum with an integral
is one which is greatly exploited in the more advanced theory
of series. See also Example 6, p. 42, and Example 2, p. 59.

EXERCISES ON CHAPTER II
I. Show that 1+3+5+ ... +(2N-I}=NI.
N

2. Find L I/n(n+I)(n+2).
n=1
3. Prove that sin nx sin ix=icos (n-i}x-icos(n+i}x. Hence
N

calculate L sin nx.

n=1
co
4 Prove that
n-l
L I/n(n+I)(n+z) converges, and find its sum.

(See Exercise 2.)

55
 INFINITE SERIES
N
,. Calculate sN= 2: nxt'-1. (Differentiate the formula
,,-1
1 +x+x l + ... +xN =(1 _X N+1 )/(1 -x) with respect to x.)
co

6. Using Exercise 5, prove that 2: nxt'-1 converges if -I <X<I,

,,-1
and find its sum. (Alternative methods are given in Example I,
p. 52, and Example 2, p. 71.)
7. Express as fractions (i) 0i48, (ii) o'oi, (iii) 072i:i.
co co

8. Calculate the first 6 partial sums of (i)2: (0'2)", (ii)2:( -0'2)".

,,~O ,,-0
Find also the sums of these senes, using the formula
S=I/I -x.
9. Find whether the followmg senes are convergent or divergent
(all sums are from n=1 to 00). (i) ~n. (ii) ~I/(2n+I).
(iii) ~2"j(n6+1). (iv) ~(n-I)/(nl+n-3). (v) ~{(!)" +~}-
(vi) ~(n+2")/(n2"). (vii) ~(2nl+I)/(2n-I).
(viii) ~(n +I)/(n +2)2". (ix) ~(2" +n)/(3" +nl).
(x) ~-{YI +n+n2/VI +n'+n7. (xi) ~n(n+I)/Vn-Z:-:-(n-+-2"-).
(xii) ~{(I -2n)/-{YI +8n'}'. (xiii) ~(n!)2/(2n)1.
(xiv) ~(I +2")/(1 +n2"). (xv) ~(_I)"+1/n3.
(xvi) ~(-I),,+l/vn. (xvii) ~(-I),,+lvn.
(xviii) ~(-I)"+ln!/n". (xix) ~{1-(-2)"}/{1+2"}.
(xx) ~{1-(-2)"}/{1+3"}. (XXI) ~{(-9)"+n}/{n!+I}.
(xxii) ~(-n)-".
10. Find the radius of convergence of each of the following power

series. (i) iX"/n. (ii) i(-I)"x"/(n+I). (nil iG)" .

.. -1 ,,=0 71=0
co co co

(iv) 2: x"'. (v) 2: (n+I)x"/(2"+n). (vi) 2: (2n)!x"/(nl)l.

n=O n=O n-
co co

(vii) 2: (n t +2")X". (viii) 2: x"j(n!+l).

,,-0 ,,=0
56
 EXERCISES ON CHAPTER 11
co
II. Investigate the convergence of L n'x", for all values of $

"=1
and x.
12. By multiplying the series for 1/1 -x and 1/(1 _Xli (see
Example I, p. 52), find a power series for 1/(1 _Xli. What is
the radius of convergence of your series?
10g.(1 -x) .
13. Show that ---=X+(I +1)x l +(1 +1+1)x 3 + .. ,11
X-I
Ixl<l.
L
co
14. Prove that I/(n+I) log (n+l) is divergent. (Integral test.)
,.~1

S.S.-E 57
 CHAPTER THREE
Further Techniques and Results

I. NUMERICAL CALCULATION OF THE SUM OF A SERIES

L
co
After a series Un has been proved convergent, there still
,,-1
remains the problem of calculating the sum 5. We know that
the partial sum SN of the first N terms of the series is an
approximation to 5, for 5 is, by definition, the limit of SN as
N-+oo. However, different convergent series may differ very
much in their 'rapidity of convergence', we may find in one case
that S6 differs from 5 by less than 0'0001, for example (so that
we get to within 0'0001 of the true sum 5 by adding up only the
first five terms of the series), while to obtain the same accuracy
in another case might require thousands of terms. It is essential
to have some way of estimating by how much each partial sum
S N differs from 5; then we can tell in advance how many terms
of the series must be added up, to get an answer within a pre-
scribed amount of the full sum 5.
We have seen (P.32) that 5=sN+R N, where RN is the sum
co

of the 'remainder after N terms' L

n=N+l
Un=uN+l+UN+II+

If we could calculate RN exactly, then of course we could find

5 exactly. But usually we can only 'estimate' R N , which means
that we prove (for a given N), that RN lies between two num-
bers a and b. This tells us that 5=SN+R N lies between sN+a
and sN+b. If a and b are very close together, we obtain in this
way a good estimate for 5.
The three examples which follow illustrate some of the ways
in which this estimation of RN can be carried out. It is also
interesting to notice how the three series differ in rapidity of
convergence.
58
NUMERICAL CALCULATION OF THE SUM OF A SERIES
Example 1. The number e is defined to be exp(I), i.e. the sum
<X)

of the series'" lin! =1+ .:.+.:.+.:.+ ... The remainder after

~ I! 21 31
tl=O
N terms is
I I I I{ I
R N =N!+(N+I)I+(N+2)!+'" =N! I+N+I

+(N+I;(N+2}+" }<~!{I+(N:I)+(N~I)II+ }-
The series in the brackets is geometric, and its sum is
I/(I _ _I_)=N+I. It is obvious that RN is positive,
N+I N
and so we have the estimate o<RN<N+I. For example,
N.NI
9 I " I I
0<Rs <-=--<0"0002. The parbal sum Ss=I+-+-
8.8! 7Ib8 II 21
I
+ .. +- works out to be 2'71825 , and therefore
7!
e=ss+Rslies between ss+o and ss+0"0002, i.e. e lies between
2"7182 ... and 2"7184 ... This proves that the value of e
correct to four significant figures is 2"718, and we could obvi-
ously get any higher degree of accuracy required, by taking N
large enough.
L Pn can be proved convergent by the
<X)

Example 2. If a series
"~1
integral test, it is always possible to derive an estimate for RN
by comparing this with an integral. Suppose that Pn j(n),
wheref(x} is the function described on p. 39. Let AN denote the
area under the curve y=f(x} to the right of the line x=N; this
area is finite, since we are assuming that J:f(X)dX~ a finite
limit A, as n~oo (p.41). In fact, AN is the limit of J:f(x)dx,
as n~oo. By considering the sum of the shaded rectangles in
Fig. 17, we see that R N=PN+l+PN+2+PN+3+ . is less than
59
 FURTHER TECHNIQUES AND RESULTS

)(

FIG. 17

AN, while the sum PN+PN+l+PN+2+ ... =PN+R N of the

larger rectangles is greater than AN' From these two remarks
we get the estimate AN-PN<RN<A N.

As an example, take the series i: III'

n=1 n
HereA N = lim
n-+CO
fn dxlx
N
ll

. (I - -I) =N'
I- and PN= -I . Therefore ---<R
I I I
_co
=hm -
N n Nil N Nil N<-
N'
and consequently 5=SN+R N lies between SN+~-~II and

SN+ ~. The difference between these two numbers is I/NII; thus

if we want a value which is guaranteed to be within 0'01 of the
I I
true sum 5, we may take N =10. Now SI0=1+-+ ... + -
211 lOll
-=1'549 ... , therefore 5 lies between S10+0'I-O'01 =1639 ...
and SI0+01=1649 ... (The value correct to three decimal
places is 1'645, but we should have to take more than 30 terms
to obtain this accuracy, by this method. A much better esti-
mate for RN can be found by approximating to the curve in
Fig. 17 by trapezoidal figures, instead of by rectangles.)

Example 3. Let L'"

=1
(-I)n+la n=a1-a ll +aa-ac!- . be an

alternating series, such that (an) is decreasing and tends to zero

60
ESTIMATING THE REMAINDER OF A POWER SERIES
as n-+oo. It follows from Leibniz's test (p. 44) that the series
is convergent. It is also clear from the discussion on p. 44, and
from Fig. 14, that 5<SN, for all odd N, and 5>SN' for all even
N. For every N, 5 lies between SN-l and sN, and consequently
IRNI <ISN-SN-Il =aN' Thus -aN<RN<o if N is odd, and
o<RN<aN, if N is even, i.e. the Nth partial sum SN is within
aN of 5, for every N. For example, the Nth partial sum of the
series I-!+l-!+ ... is within liN of the sum of the series.
This estimate compares unfavourably with those of the last two
examples; to be sure of a value within 0'01 of the true sum, for
example, we should have to add together the first 100 terms.

2. ESTIMATING THE REMAINDER OF A POWER SERIES

Many important functions such as exp(x), sin x, cos x,

(I+X)G, 10g.(I+x) can be expressed as the sums of power series
(see 12, ch. II). The general problem of 'expanding' a given
tunctionf(x) in a power series (that is, of finding a power series
Lco

n=O
cnx" such that f(x) is the sum of this series for all values of

x inside the interval of convergence) is dealt with in books on

the theory of functions. We shall adopt a different point of
view, in that we shall suppose here that the power series

Lco
cnx" is given, and see what information can be derived
"=0
concerning its sum f(x).
For a number x within the interval of convergence, the series

Lco
cnx" is absolutely convergent (by definition of the interval
"=0
of convergence). The remainder after N terms is RN(x)=CNXN
+CN+1XN+l+CN+sXN+2+ . This might he either positive or
negative, but we consider its absolute value IRN(x)I, and we
know that IRN(X)I <ICNXI+ICN+1XN+l1 + 1CN+sXN + 21 + ... (see
Example 3, p. 45). It is important to have an 'estimate' for
RN(X) for two reasons: (i) for numerical computation-very
61
 FURTHER TECHNIQUES AND RESULTS
often a power series provides the most convenient way of cal-
culating the values of a function f(x}, and (ii) for theoretical
purposes; for example in deriving a new series by integration
(see 3). The technique for estimating the remainder of a power
series is to compare the remainder senes with a suitable geo-
metric series. The following examples illustrate this.
Example 1. Geometric series. This is one of the rare cases where
the remainder can be calculated exactly. For RN(X) =xN+XN+ 1
+ ... =XN(I+X+X Il + )=XN/I-X, provided that Ixl<I.
In fact this also follows from the formula SN=I+X+X Il +
+XN-1=(I-XN)/(I-X) =(I/I-X) _xN/(I-x)(see Example 3,
p. 25). This gives the identity (which holds for all x except X=I)
II N-l xN
I
-=I+X+X
I-X
+ ...
+X
I-X
+-.
Example 2. Exponential series. The remainder after N terms
. ~ x" . xN XN+1 XN+II
of the senes L n! IS RN(X) =NI+(N+I)!+(N+2)t+ ...
n-O
Therefore
IxlN IxlN+l IxlNH
IRN(X)l';;: Nt + (N+I)!+ (N+2)t+ .

=1~~{I+~~I+(N+~~~~+2)+ -}
ll
IXIN{ Ixi Ixl
<m }
I+N+I+(N+I)II+ .
The series in the last brackets is geometric, and provided that
Ixl <N +1, it converges and has the sum I/(I-~). There-
N+I
fore we have the estimate

IRN(x)l<I~~/ {I-~1I}
(If x>o, we can see at once that RN(X) >0. So in this case we
have a rather better estimate O<RN(X)<~/ {I-N~I})
62
ESTIMATING THE REMAINDER OF A POWER SERIES
Suppose, for example, that we want to calculate exp( -i) to
4decimal places. Our estimate gives IR.(-!)I!~' / {I-~}.
which a rough calculation shows to be less than 0'00003. On
the other hand S.=I-!+(!)1I/2!-(!)813!+(!)'/4!-(!)1/5!
=0.60651 ... Therefore the full sum is between s,-0'00003
=0.60648 and s.+000003=060654. It follows that exp(-l)
=0.6065, correct to 4 decimal places.
Example 3. Binomial series. The remainder after N terms of the

series ~(:)xn isRN(x) =(~ )XN+(N ~I )XN+l+ (N ~2 )xNH

..
+ .. , N obcmg h (a) a-N(a)
t at N+I =N+I N'
(N+2
a )=N+I
a-N
tI-N-I(a)
N +2 N' etc., we have

IRN(X)I<\(~)xNI{I+I~~llxl+I~~~lla~:~IIIXIII+ .. }-
We shall consider first the case a>-I. If this is so, then
la-NI <N+I, la-N-II <N+2, etc., and therefore IRN(X)I
<1(~)llxIN{I+IXI+lxllI+ ... }. Thus for Ixl<I, and if a>-I,
we have the estimate
IRN(X)I <1(~)llxIN/(I-lxD.
If a<-I, then la-NI>I.
N+I However we find in this case that

a-NI la-N-II
N+I> N+2 >la-N-
2
N+3 1>.... and thus IRN(X)I
l

<\(~)llxIN{I+I~~~IIXI+I~~rIXIII+ .. -} Therefore if
IXII~~~I<I' we have that IRN(x)I<I(~)IIXIN/{I-I~~~IIXI}.
As an example, let us calculate {/2, correct to 6 decimal places.
63
 FURTHER TECHNIQUES AND RESULTS
Divide the equation 2 X 8 8=1024=10 8 X 10024 by 88 and then
take cube roots; one finds ~ = X (I +0024)1/80 Now

(1+00024)1/8 is the sum of the binomial series ~(!)(00024)n.

"m:O

Our estimate above gives IR,(00024)1< 1(!)1(00024) '/{I - (00024)}

=11( -iH -%H -%)1(00024)'/00976 1o 2 5 8(00008)'/00976

1020304 1020304
0 6
= 4 9 X 10- 11 <2 X 10-eo The partial sum s,=1+1(00024)
2 928
-i(00024ra+h(00024)8=loo0793685 ... Therefore (10024)1/3
lies between s,-0000000002=loo0793683 .. and s,+
.3/-
000000002=100793687 .. It follows that v 2=1 025X
(10024)1/3 lies between 1025992103 and 10259921090 This proves
that its value correct to 6 places of decimals is 102599210

3. INTEGRATION OF POWER SERIES

>
Suppose that 2: Cnxn=CO+CIX+c~2+ . is a power series
n-O
with radius of convergence R, and thatf(x) denotes its sum, for
Ixl<R. If RN(x) is the remainder CNXN+CN+1XN+1+ , we
have the equation
f(x) =cO+C1X+CaX 2+ +CN_1XN-1+RN(x).
Replacing, for convenience, the variable x by I, and then
integrating this formula from 1=0 to I=x, we find
8 N
f '"
0
X2 C2-+
f(t)dt=CoX+ Cl-+
2
x
3
. +CN-I x
N 0
+ f'"
RN(t)dl (I)

If it is possible to prove that J: RN(t)dt-40 as N-4oo, then we

x2 x8 2:> xn+ 1
can show that the series CoX+Cl-+CZ-+ .. = Cn - - is
2 3 .. -0 n+1
64
 INTEGRATION OF POWER SERIES

convergent, and has sum [f(t)dt. For equation (I) shows that

f:f(t)dt=the Nth partial sum of this series + f: RN(t)dt. There-

fore if f: RN(t)dt-,;o as N-,;oo, it means that the Nth partial

sum of ~ cn-
~
n-O
x"+l
- tends to
n+I
f"
0
f(t)dt as N-,;oo.

x. x8
The process of getting the series co%+c1-+c.-+ .. from
2 3
cO+C1X+C 2X2 + ... is called 'integrating term by term'. There
is a general theorem that

2: cnxn is a power series with radius of convergence Rand

IX)

If
n=O

sum f(x), then the series 2: Cn - - obtained by integrating it

IX) x"+1

n+I
n=O
term by term has the same radius of convergence, and its sum is
f:f(t)dt.
We shall not attempt to prove this result here. However it is
possible in many important particular cases to prove quite easily
that the 'integrated remainder' f: RN(t)dt does tend to zero as
N-,;oo (provided that Ixl <R). This, by what we have said
above, gives a proof of the theorem just stated, for the par-
ticular series on hand.
(i) The series for log (I+X).1 Replace x by -t in the formula at
the end of Example I, p. 62. We shall also replace N by n. This
gives
I ~
-=I-t+e:a+ +(_I)n-l~-I+(_I)n_.
1+1 I+t
1 Unless the contrary is explicitly indicated, 'log' stands for 'loga-
rithm to the base e'.
S.S.-E* 65
 FURTHER TECHNIQUES AND RESULTS
Integrate both sides from t=o to t=x, giving

f-~=log
o I+t
(I+X)
x' x . +(_1),,-1 X"
c=x--+-- -+(-I)"L,,(x) (2)
2 3 n
where L,,(x) = f:t"dt/(I +t). Any attempt to evaluate this in-
tegral directly would lead back to the starting point. However
we can make an estimate for L,,(x), as follows.

o X t
FIG. IS

Take first the case 0 <x <I. If t is between 0 and x, then

l+t>l, so that t"/I+t<t" (see Fig. 18). Therefore L,,(x) (re-
presented by the shaded area in Fig. 18) is < [t"dt=X"+1 /n+ I.
Now suppose that -I<X<O. Introduce a new variable
u=-t, so that L..(x) = f:t"dt/I+t becomes (_1)"+1 f: u"du/
I-U, where y=-x (=Ixl). If u is between 0 and y, then
I-U>I-Y, and hence U"/I-U<'U"/I-Y. Consequently
f: u"du/(I-u) <f:u"du/(I-Y) =Y"+1/(n+I)(I-Y) There-
fore IL..(x)1 <lxl"+1/(n+I)(I-lxl), if -I<X<O.
66
 INTEGRATION OF POWER SERIES
It is now clear that L,,(x}--+-o as n--+-oo, if -I<X<I. For
x,,+l/(n+I}--+-o as n--+-oo, if 0 <x <I, and Ixl,,+l/(n+I}{I-lxl)
--+-0, if -I<X<O. Therefore

(a) The series x--+-- ... =

2 3
x2 x8
2: (-I)-xnn
00

,,-0
converges if

-I<X<I, and its sum is log (I+X).

Going back to the original formula (2) we have also
X2 Xl xt'
(b) log (I+X}=X--+-- . +(-I},,-I_+(-I)"L,,(x),
2 3 n
where o <L,,(x) <;xn+l/(n+I}, if O<X<I, and IL ..(x} I<lxl,,+lj
(n+I)(I-lxl) if -I<X<O.
This second statement gives estimates for the remainder of
the logarithmic series, which can be used when calculating the
sum for particular values of x.
Example 1. Put X=I, and we have 1-1+1-1+
=log(I+I}=log,2. The case X=-I gives the divergent series
-I-i-l- .
(ii) The series for tan-Ix. Starting again with the formula at the
end of Example I, p. 62, we replace x this time by - t 2 The
resulting identity
I t~
--=I-t2+t'- .. +(-I)"-lt~-2+(-I}" __
I+t2 I+t2
is integrated from t=o to t=x. This gives

f "'~=tan-IX
oI+t2
X8 X5 X2,,-1
:=:x--+-- ... +(-I),,-I_ _ +(-I}"T,,(x), (3)
3 5 2n-I
where T,,(x) = [t 2n
dt/I+t 2
T,,(x) is the area under the curvey=t~/I+tl, between t=o
and t=x. Because this curve is symmetrical about the y-axis,
T,,(-x)=T,,(x), for any x, so it is enough to examine the case
x>o. Suppose in fact that 0 <X<I. For any value oft, I+t2>I,
67
 FURTHER TECHNIQUES AND RESULTS

therefore t2"/I+t2<t2". It follows that T,,(x) = f:t2"dt/(I+t2)

<f:t2"dt=xs"H/(2n+I). This tends to zero as n~oo, since

o <x <I. By the symmetry which was mentioned above, it fol-
lows also that T,,(x)~o as n~oo, for -I <x <0. Therefore
x8 x5 ~ xlnH
(a) The series x--+-- = ~
(-I)"-- converges
3 5 n=O 2n+I
when -I <x <I, and its sum is tan-Ix ('Gregory's Series').l
Going back to the equation (3), we have also the formula
which exhibits the remainder, namely
x8 x5 X 2,,-1
(b) tan-1x=x--+-- +(-I),,-l _ _ +(-I)"T,,(x),
3 5 2n-I
2
where o <T,,(x) <x "H/(2n+I), if x>o, and o>T,,(x)
>-lxI2"+1/(2n+I) if x<o.
Example 2. Leibniz's series. Put X=I, and we get I-l+l-t
+ :!:. In theory, this provides a means for calculating n.
4
However this series, although convergent, is so slowly conver-
gent that the amount of work required to secure even a modest
number of decimal places of n would be prohibitive.
Example 3. A more rapidly convergent series for n is

~6=tan-l(I/V3)=.113
~ {I-2..+~-~8+
33 5.3 73
..}. 6

From (b), the remainder after n terms is (-I)"T,,(I/v3),

and 0 <T.. (I/ v3) I/ V3)2"+1 /(2n+I) =I/(2n+I)3"v3.
For example, T1(I/v3) <-h/31 V3 = I/32085v3 < I/50,000
=000002. This means that ~ lies between S1 and s1-0oo002
(we must remember that the remainder is (-I)"T,,(I/V3), i.e.
I The value of tan -IX referred to here is the 'principal value', defined
as follows: 8=tan- 1x is the angle between -~ and +~ such that
z z
tan 9-K. (Radian measure must be used.)
68
 INTEGRATION OF POWER SERIES
for n=7,it is -T7(1/v'3)). The sum S7 ofthe first 7 terms ofthe
series works out to be 0'523612 ... ,and therefore 0'523592 ..
~<0'523612 .. ,i.e. 3'14r55 .. <n<314167 .
Example 4. The integral f e-t' dt is one which cannot be ex-
pressed in terms of the 'elementary' functions (log, exp, sin,
cos). However it is possible to work out the value of f: r'l dt,
~ (_t 2)11
for any x, by integrating the series L --nr-=r-fa+t'/21
11=0
-t 8/3!+ .. , which has sum exp(-t2) or r". We find that
f'"
6 7 211 1
X8
r"dt=x--+---+ x x = 2:'" (-r)" X + .
o 3 5. 21 7.31 ,,~O
(2n+r).nl
It can be verified, using the ratio test, that this series converges
for all x. It has been used to make tables of values of f:rt'dt,
which are used extensively by statisticians.

Example 5. The binomial series for a=-l is i: C:!)XII,

11=0
and

its sum is (r+x)-t, if Ixl <r (see Example 3, p. 7r). Replace

x by _t2; this gives (r-t2)-I=r/v'(r-t2) =r+lt 2 +
(-!)( -I) (_t2)2+ (-!)( -i)( -t) (_t2)3+ , the coefficient
r. 2 r. 2. 3
of t 211 b'
emg r35 ... r MI'
(2n-r) .-. u tIpIy numerator an d d e-
1.2.3 . n 2"
nominator by 2.4.6 ... (2n) =nI2", and this becomes
(2n)!j(nl)22 211 Integrating the series from t=o to t=x, we find

f'"
o
dt/v'(r-t 2)=sin- 1x=
2:'" -
(2n) 1 x2n+1
1)2(
n. 2 2n+r
,,=0
211-(- - ) (for Ixl<r). This

series can also be used to computen; for example putting x=r/2

we have Sin-1(~)=~= ~ (2n)!
2 L (nl)2(2n+r)2"'+1
6 11-0
69
 FURTHER TECHNIQUES AND RESULTS

4. DIFFERENTIATION OF POWER SERIES

2: CnXn=CO+CIX+C2X2+CaX8+ is
IX)

If the power series

n-O
differentiated 'term by term', the new series Cl+2CI%+3c3X8
2: nc x
IX)

+ ... = n
n- 1 results. There is the following general
n-O
theorem, analogous to the theorem on integration:
2: cnxn has radius of convergence R and sum f(x) , then the
IX)

If
n=O
co
series 2: ncn obtained by differentiating it term by term has
x"-1
n~U

the same radius of convergence, and its sum is /' (x) = !f(X).
We shall not give the proof of this result. It is important
however to notice that a proof is necessary; the theorem is by
no means obvious, as might appear at first sight. There are
cases where term by term differentiation of a series (other than
a power series) destroys its convergence.
IX)

smnx.
Example 1. We saw (Example
.. =1
n2
con-
2, p. 46) that
2: --15

vergent, for all x. Differentiating term by term, we get

IX)

"
~ cos
-n-'1tX wh'lC h'IS d'Ivergent f or x=o. D'ff" .
1 erentlatmg agam
n-1

2: (-sin nx).
IX)

gives When x is an integral multiple of n

n-l
radians, all the terms of this series are zero, but for all other
values of x, it is divergent, because the nth term un=-sin nx
does not tend to zero as ~oo (see Example 6, p. IS).
Even for a power series, we cannot be sure that the differ-
entiated series will be convergent at the end-points of the inter-
val of convergence, even if the original series was. For example,
70
 DIFFERENTIATION OF POWER SERIES
X2 X8
theloganthmlcsencsx--+-- ... convergeslf -I<X<I.
2 3
Differentiating this gives the geometric series r-x+x 2- ...
(this is not surprising, since we originally obtained the logarith-
mic series by integrating this geometric series!), which does not
converge for X=I.
Example 2. The series I+X+X 2+X I+ ... has sum I/(I-X}
and radius of convergence I. Therefore the series I+2X+3x 2
d
+ . = L
00

n=l
nxn- 1 has sum dX(I/(I-X))=I/(I-X)2, for
Ixi <I. We have already proved this in a different way (Ex-
ample I, p. 52).
Example 3. The Binomial Theorem for arbitrary exponent.
Writef(x) for the sum of the binomial series I (:)xn
n-O
(lxl<I).

Differentiating, we get

f'(X) = ~ n(:)xn-l=(~)+2(:)X+3(~)X2+ ... , hence

xf'(x) =(~)X+2(:)X2+ ..
Adding, we have (I t-x}f'(x}=a+ %J(n+I}("~I)
+n(:)}xn. Now (n~I)=(:):~:' and so (n+I)(n~I)
+ n(=) = (:) {a - n+ n} = a(:} Therefore (I + x)f'(x)

=a,~ (:)xn=af(X), i.e. f'(x)/f(x} =a/(I+x}, or :x {logf(x)}

=a/(I+x}. Integrating both sides, we find that logf(x)
=a 10g(I+X)+k, where k is some constant. By putting x=o
(which reduces the binomial series to the term I) we get k=o,
therefore logf(x)=a 10g(I+X), that is,J(X)=(I+X)".
71
 FURTHER TECHNIQUES AND RESULTS
X2 X3
Example 4. The exponential series I+X+-+-+ ... repro-
21 31
duces itself on differentiation. This establishes the well-known
property of the exponential function, that ~(exp(x))=exp(x).
dx

5. CAUCHY'S CONVERGENCE PRINCIPLE

2: u.. is a convergent
IX)

If series, then the partial sum

.. =1
s.. =U1 +u2 + .. +u.. tends to a limit S, the sum of the series,
as ~oo. By the definition of a limit (p. 5), this means that,
if any positive number h is given, it is possible to find an integer
N" such that s.. is between S-h and S+h, for all n>N,,; or,
equivalently, that IS-s..1( =ls.. -SI)<h, for all n>N". There-
fore if m, n are two integers both >N", then ISm-s ..1=I{S-s..)
+(sm-S)I <IS-s.. 1+Ism-SI <h+h=2h. If, in addition, we
suppose that m>n, then sm-s.. is the sum u ..+l+u.. +2+ ...
+Um , a sort of 'excerpt' from the series.
It is convenient to make slight changes of notation: put
k=2h, and write Mk for N", then we can say:

2:
IX)

If u .. is a convergent series, then for any positive number

.. -1
k, however small, we can find an integer Mk such that lu.. +!
+u.. +2+ ... +uml<k, for any m, n which are both >Mk.
The converse of this theorem is of fundamental theoretical
importance; it was discovered by A. Cauchy (1789-1857).
2: u.. s any
IX)

Cauchy's Convergence Principle. Suppose that

n=1
series, with the property that, given any positive number k, however
small, it is possible to find an integer M k such that
IU"+l +U.. +2 + +uml <k,

2: u. is convergent
IX)

whenever m, n are both >Mk. Then

.. -1
 DIRICHLET'S CONVERGENCE TEST
This is not normally used as a test of convergence, in the way
that for example the ratio and comparison tests are used, be-
cause it is usually difficult to estimate the sum u"+l +U"+2
+ ... +um However it is used to derive other, more special
tests (such as Dirichlet's test, in the next section) which are
easier to apply.
The proof of Cauchy's principle is beyond the scope of this book.
However it can be roughly explained as follows: we know that
U"+I+U"+I+ . +Um=Sm-S" lies between -k and +k, for all
m, n>M,.. Therefore all the partial sums Sm after s" (we can take
n =M,,+ I if we like, for definiteness) lie between s" -k and s.. +k.
If we take k to be very small in relation to the accuracy required,
this means that the partial sums Sm for m >n are 'practically' (e.g.
up to a certain number of decimal places) indistinguishable from
s ... Thus s .. would, up to this number of decimal places, represent
the full sum. In this sense it would be possible to calculate the sum
S to any reqUired degree of accuracy.

6. DIRICHLET'S CONVERGENCE TEST

Test 8 (Dirichlet's Test). Let (a ..), (b ..) be two sequences such that
(t) The sums t.. =b1 +b s+ .. +b.. are bounded, i.e. there exists
a number H>o such that It..1<H,Jor all n, (ii) The sequence (a ..)
2: a..b..
IX)

is decreasing, and (iit) a..-+-o as n-+-oo. Then the series

.. =1
converges.
Proof. Take any positive number k. For any m, n(m >n) write
R m, .. =a ..+1bn+1 +a"+2b"+1+ . +ambm. Our ainI is to find an
integer M" such that IR.n, ..1<k, whenever m, n >M k' If we can do
this (for an arbitrary positive k), then it will follow from Cauchy'S

2:
IX)

convergence principle that anb .. converges.

.. -1
By (iii), a,,-+-o as n-+- 00. We take M" to be an integer such that
an <k/2H, for all n>M". Such an integer M" must exist, by the
definition of the statement a..-+-o (p. 5); the reason for the
curious number k/2H will appear shortly.
Remembering that b ..+l =t..+ l - t.., we have
73
 FURTHER TECHNIQUES AND RESULTS
Rln,n =an+1(tn+1 -tn) +an+l(tn+l-tn+l ) + . + am_l (tm-l -tm-I)
+am(tm -tm-l) = -tnan+1 +tn+l(a n+l -an+l) +tn+l(a n +l-an+3)
+ ... +tm-l(am-l- am) +amtm. Now an+1-an+l' an+l-a n+s, ...
am- l -am are all > 0, because (an) is a decreasing sequence (condi-
tion (ii)). Incidentally, conditions (ii) and (iii) together imply that
all the an are> o. Therefore IRm,nl < Itnla n+1 + Itn+ll(an+1-anH)
+ltnHI(anH-an+a)+ . +ltm+ll(am-l-am)+ltmlam, and since
each Itnl<H (by (i)), we have IRm,nl<H {a n+1 +(an+1-anH)
+(anH-an+a) + ... + (am-I-am) +am}=zHa n +l. If now
m, n>M", it follows that an+1 <k/zH, and therefore
k
IRm,nl < zHa n+1 <zH. zH~k.

Example 1. Leibniz's test (p. 44) is a special case of Test 8. For

if we suppose that (an) is a decreasing sequence which tends to
zero, and that bn=(-I)n-1, then the conditions (i), (ii) and (iii)
are all satisfied. For tn=I-I+I- . +(_I)n-1=0 if n is
even, and is 1 if n is odd. Thus Itnl <I, for all n. Therefore the
L
00

series (-I)n+1a n converges, whenever (an) is a decreasing

n-1
sequence which tends to zero-and this is just what Leibniz's
test states.
Example 2. Suppose that bn=cos nx, where x is any number
which is not of the form 21m; (k integer). It is easy to verify the
trigonometric identity 2 cos nx sin iX=(2 sin ix)bn=sin (n+i)x
-sin (n-i)x. From this, (2 sin iX)tn=(2 sin ix)(b1+b z+ .
+bn) =-sin ix+sin (n+t)x. Now Isin 01 <I, for any 0, and so
we deduce that 12 sin ixlltnl =I-sin ix+sin (n+i)xl <I-sin ixl
+Isin (n+i)xl <1+1=2, and hence (since sin ix~o), Itnl
<I cosec ixl, for all n. Thus tn is bounded. Consequently any
L
co
series of the form an cos nx is convergent, if (an) is a decreas-
n=O
iltg sequence which tends to zero, except possibly when x is an
integral multiple of 2n.
Example 3. We can prove similarly that Isin x+sin 2X+ ..
+sin nxl <I cosec ixl, when x is not of the form 21m; (k integer),
by starting with the identity 2 sin 1tX sin ix=cos (n-l)x
74
 EXERCISES ON CHAPTER III
co
-cos (n+!)x. Therefore the series L an sin nx is convergent,
n=l
provided that (an) is a decreasing sequence which tends to zero.
(This holds for any value of x, because when x is an integral
multiple of 211:, the terms of the series are all zero.)
Example 4. As an example, we can see that the series

L
ao
cosnx
- -, L -smnx- are both convergent .
ao.
tf o<s <I, except,
n=l
n' .. =1
n'
in the case of the first series, when x is an integral multiple of
211:. For (;,) is clearly a decreasing sequence which tends to
zero. (Both series converge if S>I; for in that case they both
converge absolutely. For e.g.lcos nxl <lin', all n, and ~Iln'
tL'

converges since S>I, hence ~ Ico::x\ converges, by the com-

panson test. When S <I this argument fails, and the more
sensitive test is necessary.)
ao
Series of the form L (an cos nx+bn sin nx), the coefficients
n-O
an, btl being given constants, are called trigonometric series.

EXERCISES ON CHAPTER III

ao
I. Show that the remainder after N terms of Ln=l
I/n' lies he-

X I X
tween - --a and - (see Example 2, p. 60). Hence provCl
3N3 N 3N3
that the sum of this series lies between x08x and x083
(Take N -=5.)
75
 FURTHER TECHNIQUES AND RESULTS
00

2 Show that the remainder after N terms of ' " ~ lies be-
. ~ n"+a"
n=l
tween ~(~-tan-l ~ ~(~-tan-l ~ ___I_(a>o).
a 2 a J and a 2 a lJ NS+a

Prove that the sum of L00

n=l
I/{nl +64) lies between 0' 183 ...

and 0'193 ... (Take N =6.)

3. Show that exp{o'l) lies between 1'1050 and 1'1052, using the
estimate for the remainder of the exponential series given on
p. 62. (Take N = 3.)
4. (i) Show that e' = I +x, with an error which is less than 0'0001
if Ixi < 0'01.
(ii) Show that ~ I +x = I -l-tx, with an error which is less than
0'0003, if Ixl < 0'05 (see p. 63).
5. VerifythatlOx5'=8'-1518,whence{lIO=- 8(1 -1518)1/S
--
5 32 768
Use 4 (ii) above to prove that ~IO lies between 1'5846 ...
and 1'5856 ...
Xl Xl
6. (i) Show that 10g(1 -x) = - x - - - - - ... if -1<X<1.
2 3
.. 11
(n)Showthat][og -+X) (I
xli xS .. l'f -1<X<1.
- =x+-+-+
I-X 3 5
(iii) Find a power series expansion for 10g(1 +x +X2), valid
when -1<X<1.
7. Use the estimate for the remainder of the series for tan- 1 x
which is given on p. 68 to prove that tan- 1 0'2 lies between
0'197395 ... and 0'197397 ... (radians). (Take n=3.)
8. Find the radius of convergence of the series for sin- 1 x (po 69).

L{-
00

9. Show that I)n an cos nx is convergent, If (an) is a decreas-

n=l
ing sequence which tends to zero, and if x is not an odd
multiple of n. {Replace x by x+n in Example 2, p. 74.}
 ANSWERS TO EXERCISES
Chapter I:
I. (i) 08, 0'96, 0'992, 0'9984, 0'99968. (ii) 1'2, 0'96, 1'008, 0'9984,
1'00032. (iii) I, 1'6667,2'5,3'4,4'6667. (iv) 0'4142, 0'3179, 0'2679,
0'2361, 0'2134. (v) I, - 2, 3, - 4, 5. (vi) I, 0, - I, 0, I. (vii) - I,
- 3, - I, - 3. - I. (viii) 1'2, 1'5, 1'75, 1.625. (ix) I, 1'5, 1'75, 1875,
I 9375. (x) I, I, I, I, 1.
3. (i) a,.-+-I. (ii) 1ln~I. (iii) an~+ 00. (iv) a,,~o. (Write an in the
form (V n+ 1- vn)(V n+ I + v;;j/(v n+ I + vn) = ((n+ 1)- n)/
Vn+I+Vn=I/(Vn+I+Vn). The denominator ~-roo.) (v)
osc. (vi) OSC. 4. (i) n> 5. (ii) n> 13. 5. (i) n>73- (ii) n> 145.
6. (i) 2. (ii) 2. (iii) o. (iv) 1. (v) 1. (vi) -1. (vii) I. (viii) o. (ix) + 00
(x) - 8. (xi) + 00. (xii) 1. (xiii) o. (xiv) - 00. (xv) o. (xvi) + 00.
7. N>I9. 8. b<I. 11.1.
12. Using the recursive formula xn+l=Xn-04(.~n'+~n- I), and
taking x t =06, one has xs=06736, x.=06819I, x,=0'6823I,
x,=0 682 34 '
Chapter II:

2.1-1(N+I)(N+2) 3 lcoseclx{coslx-cos(N-J-l)x}
4 1 5. {I-(N+I)xN +NxN + 1 }/(I-X)I.
6. I/(I-X)I. 7. (i) /,. (ii) ie. (iii)m.
8. (i) I, 1'2, 1'24, 1'248, 1'2496, 1'24992. S= 1'25. (ii) I, 08, 084,
0'83 2,0'833 6,0'833 28. 5=t=0'83333 ..
9. (i) div. (ii) div. (iii) div. (iv) div. (v) cony. (vi) div. (vii) cony.
(viii) cony. (ix) cony. (x) cony. (xi) mv. (xii) cony. (xiii) cony.
(xiv) div. (xv) cony. (xvi) cony. (xvii) div. (xviii) cony. (Test 6).
(xix) div. (xx) cony. (xxi) cony. (xxii) couv.
10. (i) I. (ii) I. (iii) 3. (iv) I. (v) 2. (vi) 1. (vii) t. (viii) 00.
II. conv.lxl < I. mv.lxl > I, for all 5. For X= I, cony. 5<- I, div.
5>- I. For x=- I, cony. 5<0 (Test 6), div. 5>0.
co

12. I+3X+ 6xl -\ IOX'+ ... = L

n=O
!(n+I)(n+2)xR. RadiuscollV.~I.
ChaPte~ Ill:
Xl 2X' X' x' 2X
6. (iii) x+---+-+--- + . =
2 3 4 5 6
L 00
I
cnX", where Cn=- un-
n
n=1
less n is a multiple of 3. while Cn=- nif n is a multiple of 3.
2

(Express I+X+X I as (I-X')/(I-X).)

I.

77
 Index
Absolute convergt'uce, 46 Rational function, I I
value, ..., RecuITmg decimal. 28
Achilles, 06 Remainder of a series, 32
estunation of, S8
Binouual sene3. -49, lA}, 71
theorem, 25
Boundrd sequence, IS
Sequencc~, 1
bounded, I,
conc;;tant,1
Cauchy's CODvcrgf'nc-e pnnriple, 72 fmite, t
Comparison test, 34. 36 ge:lt"ral term of, t
Convergence, 27 graphIcal rcprcS('ntation of, 3
absolute', 46 increasing, IS
mtcrval ')f, 50 inf..rute, I
principle, 72 limit of,-;
See also Tests of convtrg~nce monotone. J.5
Cosine series, 50 of approximations, 2
oacIllatmg,9
Decunal, rN"llrnnc, 28 recursively defined, 2
DIfferentIation of series, 70 SUb-,13
Divergf"nr.f'. 27 Series
absolutely convergent, 46
Equations, solution by iteration, 20 alternating, 44
EstlIDation of renlcunder, 58 binomial. 49, 6<), 71
Euler's COOS1.l0t, 42 conditlonJ.!ly convergent, 46
Exponr-ntiai !tCncs, 48, 52, 62, 7a convergent, 27
cosine, So
Firute sequt..nce, 1 divergent, 21
Sf'neS,24 exponential, 48, ,2, 62, 72
firute,24
Gt:'Ometric senes, 25. 28 grometric, 2.5, 28
Gregory's series, 68 Gregory's, 68
hannoruc, 29
Increasmg sequence, 15 infmite,2,
Infinite Sf"qucnce, I Leibruz's, 68
senes,25
Integral tcst, 41
Interval of convergence, 50
logaritlumc, 49, 6,
multip!Jcation of,
of absolute values,
,I4'
ItpratlOD, 20 power (see Power senes)
remainder of, 32
Lrading term, I I sine, .50
LeJbniz's senes, 68 sum of, 27
test, tngonometnc, 75
Limit, , Sine series, SO
Logaritlumc senes, 49, 6, Subsequence, 13

Monotone sequen~, 15 Term of a sequence, t

Multiplication of series, 51 Tests of convergence,
alternating series (see Leibmz's test),
Oscillating sequence, 9 companson (first fonn), 34
companson (second form), 36
Parhal sum, 27 Dmchlet's, 73
Power series, 47 Integral, 41
dIfferentiation. 70 Leibmz's, 44
integration, 64 ratio (positive tenus), 38
remamder, 61 ratio (general form), 47
Product sene., 51 Tortoise, 26

Ratio test, 38, 47

Trigonometnc senes, 7'