You are on page 1of 536

cr>

CM CM

BH

EnHI

UNIV.OF

TORONTO

BINDING

AX [NTRODUCTION To TIIK THEORY OF INFIMTK SKIMES

MACMI-LLAN AND
LONDON

CO.,

LIMITED

BOMBAY CALCUTTA MELBOURNE

THE MACMILLAN COMPANY


NEW YORK
ATLANTA
BOSTON CHICAGO SAN FRANCISCO
CO.

THE MACMILLAN

OF CANADA,

LTD.

TORONTO

AN INTRODUCTION
TO TIIK TIIKoRY OF

INFINITE SERI1>

T.

.1.
KKI.I

LA.

BROMWH'H.

M.A..

F.fc/8

FORMKRLY PROFESSOR OF M

.,

LIMITED M ACM ILLAN AN] ST MAR UN'S STREET, LONDON


1908

($>/?

GLASGOW: PRINTED AT THE UNIVERSITY PR1 hV KOHKHT MA< 1.KH08K AND CO. LTD.

PREFACE,
THIS
hook
is

based
;it

on

courses

of
(

l<

(^m-i-n's Coll.-^.-. Analysis given ialway. dm Hut additions have naturally LOUS I!MiM!M)7.
:

Mi"
ln-,-n

n.

in

preparing the manuvript fot press: in parti r hapter X I. ami th part of added, in tinbeen r, selecting subja
1

!ia\-.-

attempt
itli

t> include proofs

>!'

all
I\

tliron-n,

in

Priu^lM.-iin'-

Irrationalzahlen
thIn
i

////'/

wnendlicher
n-latin^- to coiitinu'l
is
j

-\<-rpti<.n

oi'

tli.-oi-rins

tVad
1'

i.

n-.
n-

Chapter I. limit ami

a
ol'

preliminary account
COTr
I

the
this

ha\e

not

in

clia-

ath-iiipt.-.l

the fun-i to supply arithmetic proof- ,,f thecrein^ (-..nctTnin-- the existence of limits, l.m bave allow-'-l
their truth
to
rest on an appeal to the reader's intuition the diseu^ion may thus lr mai-i,

the hope that


live
to

Ih-^inners.
J..

An

arithnietie treatment

will

!>

f"iin<l
;

in

where Dedekiml's definition ..i irratinnal num is the adapted as fundamental: this metliod lea<U at niom.tonic ]rim-ipl- of mce (Art. 14!h. I'roni \\hicli e\i breme limits t ia deduced L60);itist]

Appendix

<

easy

t"

.-taMish the ^-m-ral principle of


.1

In the remainder <,f the l,(...k fiv- n>- is made ,,f the n and principles of the >itl-ivntial and Integral Calculus: I for BOme time l>em c. mvince.l that b uld not attempt
1
1

to

study

Intinite

Series

in

any

detail

until

after

they

H.I.

I..

1'1>.

17 ;in.l

H'Jl).
litTf.

t Not only

Imt

in ni:tny <>th-:

>e

by a

systt nur

i.uin

have been and ininiinuin limits.

vi

PKEFACE.
of

mastered the differentiation and integration


functions,

the

simpler

and the geometrical meaning of these operations. The use of the Calculus has enabled me to shorten and

simplify the discussion of various theorems (for instance, Arts. 11, 61, 62), and to include other theorems which must have

been omitted otherwise (for instance, Arts. 45, 46, and the latter
part of 83).
It will be noticed that

from Art. 11 onwards,

free

use

is

made

of the equation

although the limit of (1 + 1/V)" (from which this equation is commonly deduced) is not obtained until Art. 57. To avoid the appearance of reasoning in a circle, I have given in Appendix II. a treatment of the theory of the logarithm of
a real number, starting from the equation

The use of this definition of a logarithm goes back to Napier, but in modern teaching its advantages have been overlooked
until comparatively recently. An arithmetic proof that the integral represents a definite number will be found in Art. 163, although this fact would naturally be treated as axiomatic when the subject is approached for the first time.

of double series,

In Chapter V. will be found an account of Pringsheim's theory which has not been easily accessible to English

readers hitherto.

uniform convergence usually presents diffifor this reason it has been explained at some length, and the definition has been illustrated by Osgood's The use of Abel's and Dirichlet's names graphical method. for the tests given in Art. 44 is not strictly historical, but i^
of
culties to beginners;

The notion

intended

to emphasise the similarity between the tests for uniform convergence and for simple convergence (Arts. 19, 20). In obtaining the fundamental power-series and products
is

constant reference

made

to the

principle of

uniform con;

vergence, and particularly to Tannery's theorems (Art. 49) the proofs are thus simplified and made more uniform than is

otherwise possible.

Considerable use

is

also

made

of

Abel's

ri;i:i

\ii

(Arta, 50

'-1

88) on
nl'

tip-

<-.ntinuity of power-series, a

the,,ivm which,

spit.importance has u-'ially not been diseased in text-book-. Mately --ral.lv contaii ('hapt.-r XI complete account of thof not, nt ami asymptotic iitly de\,-l,,p,'d theoriefl
iii

its

Beri<

iiP-nt

li

'-ontim-d
JMii-tlh-Mi-y

t,,

tli-

antlimth.-

applicatiMiis tM
tipM!'

i'linct

f \H'\w f

nutsid'-

sr

,,p,

iM,,k.
tip-

Afl

mi-'lit

IH-

exp
l.-.l

systematic
M!'

knnwn

r.-sults

has

IM -MUM- exteosiODfl
L21,
IL':;.

th--

th-

For

Instance, Arta L18


,,f
f

Tin-

inv.-Nt i-at SMUS

('haj.t.T
intinit.t'ir

and pa XI. imply an


ut

ar.juaiii'
inaiiii-

with

tip-

COnveTJ
I'l'Min

in!

wlp-n th-

repared
a\ailaM.I

pi-intini:
t

no

Kiiuli^h

ln>ok
(

was
an

which
lrl

tin-

Qe0688ary
writ*-

h-< IP-IMS

CMiiM
III.,

qnot
^ivin^r

was

tlp-n-fMiv

ti

out

App.-iplix

intnuliictiun to tip- throry ..!' int-_i:raN: IP-: HMH is (lir ct. ,1 to the ln-tv. of of and difference points similarity this tlp-ory ami that To emphasise tip- siinila961166,
!'

the tests of convrr-viic,- ami of unit'Mnn <-Mii\vrirence (Ai 171. 17'J. are c-alli-d l>y tip- saiip- nanp-s as in the case ol

ami
i^

Second Theorem of M.-an Valui-i-plat-i-il l,y iip-<|iialitirs (Art. Uiii) which are IMMI-.- obvioc Connected with Alu-l's Lcnnna (Art To illustrate the ral tlu'Mi-y. a ^h<.rt discu^siMii nt' IHrichlet'- inte^i-aN and
tip-

traditional 1'nnn of the

'

i.

"t'

will

the (ianmia integrals i^ given; it i- hoped that these pmnt's le t'niind Imth simple and ri-orou-.

ol

The examples (.,!' which there are over li(M>) include a nuinl.rr tlp-Mivms which CMiild not inserted n the fcert, and in
!
i

sm-h caaee referencefl are ^i\-en


ThrMU-'liMUt

t<>

SOD
it

I'urther inl'm-inatiMn.

the

l.M,,k

have made
nl'

my

aim

tn

keep
]ml.l.
i

in

view the practical applicationw..rk


in

the

tlp-Mreins to ev-

which present
Settled
'

analysis. ),,,p,. theinsel\.


I

that

most
in

dniiMe-limit

r<iU;/.
nl'

connexion with
-.

9,

dillt-rentiatiMii
l.y

and

BO I'Mrth.
h.
:

witliMUt ditliculty
l>ook has
l.t-fii

iisin-

the results -m-,,


books have appeared,

\VhiK-

my

in

tin-

press, three

<

\UiK-li

roiitums

MB
Integral* (ch. iv.),

-'ml

and Pierpont'a Theory

ntctiuns </

viii

PREFACE.

Lecturer of Trinity great help during the preparation of the book he has read all the proofs, and also the manuscript of Chapter XL and the Appendices. I am deeply conscious
College, has given
;

Mr. G. H. Hardy, M.A., Fellow and

me

that the value of the book has been

much increased by Mr. and Hardy's suggestions by his assistance in the selection and manufacture of examples. The proofs have also been read by Mr. J. E. Bowen, B.A., Senior Scholar of Queen's College, Galway, 1906-1907 and in part by Mr. J. E. Wright, M.A., Fellow of Trinity College, and Professor at Bryn Mawr College, Pennsylvania. The examples have been verified by Mr. G. N. Watson, B.A., Scholar of Trinity College, who also read the proofs of Chapter XL and Appendix III. To these three gentlemen my best thanks are due for
valuable
;

their careful work.


T.
J.
I'A.

BROMWICH.

CAMBRIDGE, December,

1907.

The following list comprises those books of use in arranging the material
:

which

have made most

Chrystal, Algebra, vol.

2.

Hobson, Trigonometry.
Osgood, Infinite Series. De la Vallee Poussin, Cours d* Analyse Infinitesimal?. Goursat, Cours d' Analyse Mathdmatique.

Tannery, The'orie des Fonctions d une Variable, t. Cesaro, Lehrbuch der Algebraischen Analyti*. Pringsheim, Mathematische Annalen, Bd. 35, pp.
y

1.

297-3!)

Reference has also been made to works on Analysis and Theorv .if Functions by Baire, Borel, Dini, Harkness and Morley, Hohs.m, Jordan, Lebesgue, Nielsen, Osgood, Picard, Runge, Schlomilch, Stolz, Vivanti, and various other authors, in addition to the sources iiu-nUnm-d alu.vf and
in

Chapter XI.

CONTENTS
(

'HAITI-:!!

I.

M"!l<'l"!
l'i
I

i!i''i]>:<

|'i"-i

ami

L.iu.-r

Lii:

Maximum and Minim am Limit*

Intiuir.

17

CHAPTEB
SI:KIK>

II.

01

I'OMTIVK TKK.MS.
'>
(

'.iiiil.-ii-

;.

('"iii|

;.

Logarithm:

Mrin.tk-i:

Aii"tli-:

EXAMPl

rilAITKK
i\\..

III.

Xwtl
i

Abel

and

l>irirhi.
i

Hati"
rniatiun.

Lfiiiin;t.

60

HAl'TKi;

IV

t.

Ajj)li.-ati"ii>.

i'

i-:.\

IMP

CONTENTS.

CHAPTER
DOUBLE SERIES, Sum of a Double
Positive Terms.
Series.

V.
PAGES
-

72-90

Repeated Summation.

Series of

vergence. heim's Theorems.

Tests for Convergence. Absolute ConMultiplication of Series. Mertens' and PringsSubstitution of a Power-Series in another

Power-Series.

Non-Absolute Convergence.

EXAMPLES,
Theta-Series (16)
;

90
other Elliptic Function-Series (19-24).

CHAPTER
INFINITE PRODUCTS,
Weierstrass's
either Sign.
Inequalities.

VI.
95-103

Positive

Absolute Convergence.

Terms Uamma-Product.
Terms.

of

EXAMPLES,
Theta-Products (14-20).

103

CHAPTER
SERIES OF VARIABLE TERMS, Uniform Convergence of Sequences.
of
Series.

VII.
-

108-125

Weierstrass's,

Abel's,

Uniform Convergence and Dirichlet's Tests.


of
Series.

Continuity, Integration, and Differentiation Products. Tannery's Theorems.

EXAMPLES,
Bendixson's Test of Uniform Convergence
(14).

125

CHAPTER
POWER
SERIES,
-

VIII.
-

128-142

Intervals of Convergence.
Differentiation, Integration.

Abel's Theorem.

Continuity,

Theorem

of Identical Equality

between two Power-Series.


Reversion of Series.

Multiplication and Division. Lagrange's Series.

SPECIAL

POWER

SERIES,
Series.

-143-160
Binomial
Series.

Exponential Series. Logarithmic Series,

Sine and Cosine Series.

arc sin and arc tan

EXAMPLES,
Series for
TT

161
(A. 48)
;

Abel's

Theorem

(B. 13-21)

Lagrange's

&

u-s

(B. 30, 31)

Differential Equations (B. 34-36).

.II:NTS.

\i

<

EAFTEE

IX.
-

TRIGONOMETRICAL INVK>HSin.a
'

IT,

Miiltip!-

188

CHAPTEB
<

X.
-

'o.MI'I.KX

\M>

I'!;

1'.'

Complex Numbers.
ami

le Moj\|.

in.

Ooir

lencec,

IV
\\

Alisolutr convergence T--' llnti.i l'i-i[i_'-li.-inr>

Oonvergenoe.
i

-el's

hii-i.-hk-t's
t

Tr>t
!'

rnifoiin
-

<

'iivk-

l-;\lt.iH-nti,i

Sim- ami

Cnsiii.-

Log U Binomial Srrirs.


Bine

S.

!!-.

it

In:

tnd

Differentiatioo "

imnM'tri.Mi sM-ir>.
-

and Conne ProdacU,


NuinliT>.
15fi ii'ulli;i!i

The OotanFiiu.-ti.

gent Kult-rs

I'.rrnonlli's

BZAMPl
Tri
('.
I

-41
an
An.irli-

(A.

12

Tli-.i

vm
1

<

15.

27.

9-11):

\\Vi.-i >u-ass's

I)..ul.k-

Series Tlu-.i VMI

(I

..

32).

CHAlTKIi
.-(

XI.

'uNVKK'iKNT AM'
I'lililiu^rapliy.

AsY.MI'T TI

'

SKIM
(n-m-ial

Historical

Intr>iln.-ti<.n.

Cn-

BORBL'8

MKMiMi'
MIS

UMATION,
I'liiulitimi
<-i

Borel's Integral.
lo a

Addition of
Suiniiiati
<

SuiuiMaM-

Al^oluti- Siiiiiinalulity.
-nt iati"ii,

Multiplication.
Ai.

'oininuit y. Di:

and Integration.

--in.

OTHKI;

MKTII->I.S

OF SCMMATIMN.
I{.y's

Boivl's

and Le
of
i

othT
Kulei-s
-

Minit
Series

ion>.

Q
1

of
,-al.

ition.

an

I'.oi

Examples

Kvil.

heorem.
'

Multiph'

\.iuiplea

xii

CONTENTS.
PAfiE.S

ASYMPTOTIC SERIES,
Eider's use of Asymptotic Series.

322-346

Remainder

in Euler's

Formula.

Logarithmic

Stirling's Series.

Fresnel's Integrals. Integral. Poincare's Theory of Asymptotic Series.

Summation of Asymptotic Stokes' Asymptotic Formula. Series. Applications to Differential Equations.

EXAMPLES,
Fejer's

347
la

and de

Vallee Poussin's Theorems

(5-7).

APPENDIX
Infinite Decimals.

I.

ARITHMETIC THEORY OF IRRATIONAL NUMBERS AND LIMITS, 357-389


Dedekind's Definition.
Algebraic Operations with

Irrational

Numbers.

Extreme Limits
vergence.

of a Sequence.
of-

Monotonic Sequences. General Principle of ConExtension


of

Limits

Quotients.

Abel's

Lemma.

Theorems on Limits.

EXAMPLES,
Infinite Sets (15-17)
;

390
Goursat's

Lemma (18-20)

Continuous

Functions (21-23).

APPENDIX
DEFINITIONS
OF

II.

THE

LOGARITHMIC

AND

EXPONENTIAL
-

FUNCTIONS,
Definition

396-410

and Fundamental Properties


Function.

Exponential

Exponential Series.
(Single and Double).

Logarithmic Arithmetic Definition of an Integral

of the Logarithm. Scale of Infinity.

KXAMIM.KS.

410

APPENDIX
FUNCTIONS,
-

III.

SOMK TllKOUKMS ON INFINITE INTEGRALS AND G. \.MMA414-466


Iviinitiim U Convergence, Divergence, and Oscillation. Limit of a Sum. Tests of Con \ri-rii.v. Analogue of Abt-l's

Lemma

(or

S>n<l Theorem

of

the

Mean).

Absolute

FAOE*
.ui.l

lMri.-lil.-t

T.--I-

Fi

ili.t

Int.
1

nif.-iin

'

--1'g

>iri<-hl<-t'>

T--t-

f-

hil'
,.->

M. ,in.l
.f

Intr-iati.

.n.

Sj)-.-i;il

Liiniti

Illtr<_ri:i

hit.-Jl

Inti-u'r.-il.

I:

Repeated
<

lnt-::i;i!-.
<

Tli--

;:iiniii;i

ii

.--nil.
i-'-i

Stii

hunma Panction
.

mn

Camma

Fun. -ti.

.us (35-47,

55-57).

EASI MISCELLANEOUS EXAMPLE


EXAMPLES,

MISCELLANEOUS EXAMPLI
EXAMPLES,
iiann's
I

i-506
)isi-.intiinnus
St-rirs

(18);

i.

(19,20):

\V.'i.-r>tras>'> \(,ii-|)illrivnti;il.lf

Kunrtimi (32-36)
Then-i-ius
li*-ta'1

Miami's

(-FnnctioD
S.-i-i.-s

(44-49);
(64-66);

Hifinann's
J.icolii's
(

n
:

.)U..nu-tri.-al

Kun.-i

(70-77):

Fiinrtii.ns \vith..,it

Analyti.-al

'..ntinuati..ns (78-

85);

Kumm.

;Mc ?ow{

87-100).

iNhKX

OF

SPECIAL

LNTBGEALS,

PBODUCT8

AND
7-508

I;AL

INII

509-Mi

ADDENDA AND CORRIGENDA.


p.

16,

l."17.
1

The

series is here
is

-2 + 3-4+...
:

supposed to oscillate

finitely

such a series as

excluded.

p.

29,

Art. 11, and p. 80, 1. 2. The integral test is commonly attributed to Cauchy it occurs in Maclaurin's Fluxions, 1742, Art. 350.

p.
p.

97.

Ex.1.

Thevalueof the second product


first

is

(sinh7r)/27r= 1-845... (Art. 91.)

101.

Art. 41. A proof of the Art. 77.

part can also be given on the lines of

p.

122.

The

discussion
limits
;

thus,

can be somewhat shortened by the use of extreme if is chosen so that

m
l
l

we
p.

get at once
of

141.

For methods

determining the region of convergence of Lagrange's series, see Goursat,. Cours d' Analyse Math., t. 2, p. 131, and Schlomilch, Kompendium der hoheren Analysis, Bd. 2, p. 100 ; the relation between these methods will be seen from theorems due to Macdonald, Proc. Lond. Math. Soc., vol. 29, p. 576.
be 4-8105...
instead of 4 '80,
1
.

p.

146, Ex.

The numerical results should 23-14... instead of 23 '00.


1.

and

p. p.

164. 190.

Ex. 25, Ex.


18.

4.

The index
is

- 1

is

omitted from

{(1

-xy)(l -xly)}~

This

taken from the Mathematical Tripos Papers, 1890.


v is

pp. 212, 213.


226.

It is

assumed that
2?r).

v(w)
p.

= v(u-

a function of w w ith a period


r

2?r,

so that

Professor Dixon's own version of his proof has just been published in the Quarterly Journal of Mathematics (vol. 39, p. 94, Oct. 1907).
1.

p.

227.

8.

The reference should be

to Arts. 44, 45

(1),

instead of Art. 49.


5,

p. 323.

1.

3.

should be 8 instead of the following four figures are 6060, according to Gauss.
last figure in Euler's constant

The

and

p..

410,

1.

4.

by using
division.

The proof that 2 > 8 n and a < S n can be made purely arithmetical 2', <r', the sums obtained by superposing the two modes of

We

2, and a' similarly


p.

is
<r

< a' < 2' < Sn

have then 2'rxr', while 2' is less than both 2' Thus 2 a' greater than both s n and <r.

>

> >

N,, *

and ami

471.

Exs.

Similar integrals occur in Electron-Theory Soinmerfeld, (iitttimji n Xachrichten, 11MI4, p. 117).


20, 21.

(compare

p.

490.

Other examples of a simple character Non-differentiable functions. have I.e. M -ixen recently by H. von Km-h. A<-tn Matin ninfifi. \'\. /hnt*>-l, n Math. 30, 1!K)7, p. 14",; and by Faber, Jahr, *l rirhf <1>
/
t

Verdi,,
p.

!',<!.

Hi,

1JI07,
1

p.
,

:>3S.
is

495,

Ex. 51.

The

function /(a,

;/i

Mippo.-etl monntniik-

with respect to

n.

AN INTRODUCTION TO THE THEORY OF


INFINITE SERIES

CHAPTEB
SEQUENCES AND
1.

I.

LIMITS.

convergence and divergence. we have agreed upon -MIL- rule, OT rul< which we can BSSOCiate a definite numher ,/, \\itli any .i-H^ned
Infinite sequences
that
:

SiipjM.s.-

positive inte-vr

,,

then the
'

set

of numl..-!^

"i-

arranged
1.

BO as to
I

c<

H-IV>JM m<

to

the

set

of

]nsit:

'!.

/'

will
\V.-

IM-

i-allr.l

an

infill

siiujily
!!tin-

shall

l'r.-i|ucnt ly

timl

it

convenienl

'I'h'lisequence, the word infinite simply means that every t.'i-ni iii the m-iju. is followed hy anothei- term.

notation (am) to

i--]ir-s.'nt

this

The

rule

detinin-(oi-

the >e|n,-iice

may

either

l>e

expl

ome

loi-iiiuln

me
Ex.
Ex.
1.
I'

BS an explicit fui; verbal Statement which indicates IK,\

formulae)

_d\ in-- a

rm
t.-nns.

be determined, either directly or

from the preceding


-Id

imnil.rrs

1.

2.

I'M.'

li;u iM"iii<

Ex.3.

Tlir

ni'_C

't'

thr

rati.-iial

positi

arm
of
;

to
i

ill't

>.!

ami
I--

ini>

h'T'-

'-an

'

ami

80

this

nu-tlr

.f

-\\

aiiuvm.Mit
|

.1-.--

n..t

It-ail

!"

anv OOITeBpOod
..'to

UK!

the
i

si-<|in-ii.

it

mil-

<>f

llx-ii

"hrir inn.

SEQUENCES AND LIMITS.

[(

H.

I,

The most important sequences in the applications of analysis are those which tend to a limit,
The
be
limit of a sequence (an ) is
to

said

to be

I,

found

correspond
i

to every positive

if an index m can number e, however

small, such that

< an <i +
to

provided

only that

n*>m.
contract these

It is generally

more convenient

two

in-

equalities into the single one

\l-a n \<6,
where the symbol
of x.
\x\
is

used to denote the numerical value

The following notations will be convenient abbreviations for the above property
:

I = Km a n

or

= lim an

or

a n -+l
is

>-oo

the two latter being only used when there to what variable tends to infinity.

no doubt as

Amongst sequences having no limit those with an infinite limit.

it is

useful to distinguish

A
the

sequence (a n ) has an

number

infinite limit, if,

may

be,

an index
an
771.

no matter how large


be

can

found such

that

>N,

provided only that

n>
;

This property is expressed by the equations or a.n ->>oc lim a n = oo or lim a n = oo


;

><x>

In like manner, we interpret the equations oo oo a w -> lim an = lim a n =


, ,

oo

n >oo

In case the sequence (a n ) has a


convergent and sequence has an
Ex.
'oo

finite

limit

I,

it

is

called
if

is

said

to

converge

to

as a limit',

the

infinite limit, it is called


1

divergent*
odd integers) we Imvj

1 big.
;

With a w = 2w

(the sequence

of

a diverge/it sequence.
2
ftfa.

Ex.

With a n =l/n

(the harmonic sequence)

we have

-*<);

convergent sequence.
but it <lir<i-<j< at us equivalent to non-conn r<j< nt distinguish l/<-t\veen sequences which tend to infinity as a limit and We shall fall the latter sequences <>.<<,-i//<tturit (Ai those which oscillate.
*

Some

writers regard

.oii\ cnit-nt to

1]

Ex. 3
ln>

It

til

ler
liiHttrl
h<

in-

liu
s.Mjii.-li'
-.

in

tin*

tli.-i.-

will

,i!.v;i\-

!:

unlimited nun.

\V- shall tind


:>hirally,
;iiul
iiid;.

it

convenient some! im

an OrdlE
'

MM

abfiG

'

<-<|iial

the
8

sr

|U.-n<-.-

may

'

pirtun-d hy joining
liin-.

tini

points with a l.ml

In
li.-

tli'

OOni

points

lii-

\\holly within a liuri/mital .strip <>!' \\i<lt! rtain valu-: if tin- sequence is divergent, fch< has passed al')\,wholly (or below) a certain i-taiii valur.
!

'I'in-

^rajhical i-rjirr^-ntat alr.-aly considered


i-

>n
i-

!'

tin-

initial

tmns

tin-

thr--

given

It

will

1)0

srrn

at

a ^lann- that

tin- t'.-w

tennfl r<-j'r-s,-ntrl
is

in

tintin-

iliM-T.-ini

sln-w

that

tin-

tirst

s.-.junin-

111

s.-cMinl

converge, and

tin-

third to <>scillat- (see Art

Tin- definition nf a limit


'/'/"

is
9

oft.-n ]oo>,-ly
/'//M'/
/.

th

'//

makt
<!

I-

Smdi
aa

a definition

id:

tin-

pos^jl.i;

may

l>-

Been tY-mi th
'

'

'

h
in

h
/

which
Hd-.-. l.y

n f 4)
n
\

if

is

,'\

en.

taking

large enough,
i-ut

we can

tin-l

berm

whirh

is

as .small

M W

tin-

and

r(,/ -f

4
(2) Infinity. It is to be
infinite,

SEQUENCES AND LIMITS.

[CH.

I.

remembered that the symbol oo and the terms infinity, infinitely great, etc., have purely conventional

in fact, anticipating the that infinity must be regarded q# an upper limit which cannot be attained. The statement that a set contains an infinite number of objects may be under-

meanings in the present theory;

definitions of Art. 4,

we may say

stood as implying that no

number

suffices to

count the

set.

Similarly, an equation such as lima n =oo is merely a conventional abbreviation for the definition on p. 2. In speaking of a divergent sequence (aw ), some writers use The numbers an become infinitely great, when phrases such as n increases ivithout limit. Of course this phrase is used as an
:

equivalent for tend to infinity


in the sequel.
(3) It
is

but

we

shall avoid the practice

evident that the alteration of a

finite

number

of

terms in a sequence will not alter the limit.


Ex,
4.

The two sequences


1, 1,

2, i,

3,

4,

J,

i,...

i, J,

i i

*, fc..

have the same limit

zero.

Further,

it

terms from
the limit;

evident that the omission of any number of a convergent or divergent sequence does not affect
is

but such omission

may
13,
2.

change

the character of o

oscillatory sequence.

Ex.

5.

Thus the sequences

1, 5, 9,

limits as those considered in Exs. t'ini> in the sequence

1,

... and 1, |, 1, ... have the same But the omission of the alternate
;\,

1, fc I, i,

!,*,*,...

changes

it

into a convergent sequence.

(4) In a convergent sequence, "//,

/////"//////.

ji i/ifc

mi mix

or none of the terms

may

be equal to the

limit.

Kxainplrs of these four possibilities (in order) aiv


1;
n

^ivm ly
5

= -sm(iw7r);
in

"N

=
W

-^ +

a-^J

the limits of which an-,

m-drr, 1,0,0,0.

1,

2|

IfONOTONK

-i

We
'

shall

usually employ
:

to

<!

ily

ive

iiuiiil..-!'It,

si

rid

i\
..!'
/

,.-

irord
'/
(

hOW

>//"'//.

whir!

added
\\.-

l.ut
i

861
.

!!.

is

/,

tliau

shall

al
b<

A'

(or

'')

to

denote an arl.itrarily
i"
:

tive
unii

numher:
Inn
variahl'
fco

are usually added


i

emphasise

that

In-

hail

N.

denot,-

any

positiv,- nunili-r.
t

we could
d
,
I

with

l.ut
it

it

avoids roMt'usioM
xnliu-t ilil.-s
c< Hi \
,

we two
H

II. \\.-\rJ-

is

-|

j,

\}'
)

\.
i

M.'ii
!)

ha]iM-iis

hat

ri-rtain

limit

can

lesa

than
<!'

+
iii'i

wln-n-

f,,,

can

mal.-

arl-it

rarily small

hy
>ull

choic.-

an

thru intVr that /_'. not ma'', 98 than

K'-r
/

ii'

wer

than
tin-

".

which cout ralicts


that
i

hypoth.
leSfl

A-

Special '\aui]lc
s.-ijii,
'lie. -s

ol'

this, supposi-

such that

can

I"-

mal.-

than

hy takinthditr.-rciM

,>

^>

n>

then

w.-

can

ch..,

j.i-ovidcd
///

that

//>///.

Thus

lini

/>

lini",,

<

only: hcnc.- we must hav<- lim/;


It

=
i(

lin

should
s

..... l).s,Tv.-l

that
it

ii'

Bequencee

U rh

that

,<i>,.
lim "

may
=
lim

easily ha].]M-n

that

K..r

the

.lit!

Ithou-'h constantly
imit.

j.osi

Thus
lim
<i,,

th-

'-.uiclusion

\'v\\,

hif'|Ualit\

'

<

:_

lim

2.

Monotonic
iu

sequences

and conditions
tor all

for

their

conailed
all

vergence.

which a
and
calle

vulu-

an
yah,

similar.;
1

and

ied

in

SEQUENCES AND LIMITS.


The
first

[CH.

I.

general theorem on convergence

may now

be stated

infinite;

monotonic sequence has always a limit, either finite or the sequence is convergent provided that \an is less than a number A independent of n; otherwise the sequence
\

diverges.

For the sake


that a n
is

detiniteness, suppose that tt+1 constantly less than the fixed number A.

of

and

however small the

positive

fraction

possible to find an index

such that a n <^a m -\-e, if n^>m', for, if not, it would be possible to select an unlimited sequence of indices p, q, r, s, ... such that ap >a^ + e, a q p + e,
,

may

be,

it

will

Then, be

>a

ar

>a +
g

e,

as ^>ar -\-,

etc.;

and consequently,
,

in the sequence p, q, r, s, ... we A contrary to hypothesis. index v such that a v if we the Thus, graphical representation described in employ the last article, we see that all the points to the right of the line x = will be within a strip of breadth e; and that the

enough*

after going far should arrive at an

>

breadth of

the
far

strip

can
to

be
the

made
right.

as

small

as

we

please

by going

enough
it

From

the

graphical

representation

approaches some
be
equal
to

appears intuitively obvious that the sequence limit, which cannot exceed A (but may

But inasmuch as intuition has this value). in mathematical reasoning, led to serious blunders occasionally it is desirable to give a proof depending entirely on arithmetical
grounds; such a proof will be found in the Appendix, Art.
Ex,
1.

14!>.

As an example

consider the increasing sequence

t
which
is
i

8.

i>

>>

represented by the diagram below.

1234567
FIG.
4.

In this case we

may

take

A = \, and

there
;

is

but that the limit of the sequence is equal to .1 have taki-n .1=2, in which case the limit would In-

no diHirulty in of oourse \u- might less than .1.

*The nmnher

of

terms to he taken

in

the

se^umce

/i.

</,

/,

v.

..

would he

equal to the integer next greater than (A -i)/c.

2|

KONOTON*
Ex.
2.

7
\

si-

the sequence

(1

iv

that

tli

that
\

limit

obtained
jn-.,ji.-i

is

th-

formal pr.mf
">s
:

!'

th-

ni..ii"t,,ni,-

||i-iuli\.
I'.ut

Ait.

ainl tin-

lit,

in

ease
1

n<>

numbrr such always an may In-, thnv


i.s

inl,
iinl

so that,

> .1.

then

it

is

plain

that
if
/'

tin-

+00,
in niodifyini:
^
I

>
Thwork
iva.lcr will havt;1
)

.1.

^m.
a sequence which
n-

no ditliculty
tr)
l ^

!''y
tt

ie

case

BO tliat

m+j

'/.
0,,

Ex.
If

3.

('niisiih-r

tin-

s.Mjufnce

= ^.
dftinitt*
t,

0<r<l,
:

the

se<|
/

ulily decreases but tin- ti-rm-

j..Mti\.1 /

and

r..nsfi|iuMitly
wt-

approadirs a
'

limit

such

that

> _ o.
II.

'I'hu-

can lind

//<

:".nd to
,

s.i

that

^<r"<^ +
1

if

n>r/i.

<r(/ +
itly

);
)

J<r" + <r(/ +
/(I

-;)</.
however
Miiall
e
;

-jiiality

is

inn-,

may

U-,

\vt-

havi-

= 0.
we can
:it

\Vhi-ii

/>

1,

it

follows

fi'.'in

th- last result that

Thus

\v- find /"

>

-aii

iaMi>h-d
MI
It

t'lMin tin- iimiiMtMnii- projii-i


1.

ty "f tin-

Mquenoe

"i

by

<i

K.\.

An
(-l)".|r|"
t'

tin-

sr.ju.'iH-e

can be drtfrniim

btained.
Sniinuinjr
It

"]>,

''"it

-Kr<l,

r"-*0;
1
:

8
In
If
all

SEQUENCES AND LIMITS.


other cases the sequence
oscillates,

[CH.
find
:

I.

and we

r<-l,
r=-l,

r2M ->oo,

r2n+

W-oo
=-l.

^=

1,

r-" +1

Ex.

4.

Take next

o n =r/w!.

p be its integral part. Then the sequence (a,,) decreases and since is positive it follows steadily, after n exceeds the value p that a n -+l ^0.
If r is positive, let
:

()

Now
Thus we can
and

an
find

n+ I

n+2

..

2n

Zn

so that

a-2n 3,

<a

>TO
)

Hence, as in Ex.

we

obtain

l<^(l + t) or
It follows that

= 0.
we
obtain the same result by writing

When

r is negative,

a
Thiis for all values of r

= (-l)". |r|"/!.

we have
lira

n\

=0

3.

General principle of convergence.


|

If a sequence is not monotonic, the condition that a constantly less than a fixed number is by no means sufficient to ensure convergence this may be seen at once from the
;

sequence given in
it

Example

3,

Art.

1,

for

which

0<a n <l.
is tin it

The necessary and

sufficient condition for convergence

may

be possible to

find an index m, corresponding

to any-

positive

number

e,

such that

for all values of

n greater than m. Interpreted graphically, this implies that sequence which are to the right of x = m,
of breadth
2e.

all
lie

points of the

within a strip

The statement

is

then almost intuitive, since

the breadth of the strip can be made as small as we pithy going far enough to the right; an arithmetical, proof will be found in the Appendix, Art. 151.

2,3]
Ex.

GENERAL PRINCIPLE
the
J>

"I

OONVJ

012345678
Pic

hieh
aiid
it

it

is

easily

>.-t-n

that tin- limit

is

1.

Tin- diairiain

i-

will

CAT
above
;i

der
test of
<-oii\ .-I'Li-'iicf

must

wunu-d

tin-

(even

as equivalent t<> a condition sometii in hooks which an- i;vnT;illy nann-ly:


>.

condition

f<rr

conv*

~
,

/////

This condition
/////.

is

certainly
f<>
/,.

necessary,

lnit

is

NT
<>j

siitli-

^,II>IH,S,<I

,i ,,

a/rbtirary function
//.
>'//

-l'-l

may
K-T

tend towards
i-\;ilii|iU',

/'///////'///

//-/'/A

*///

arbitral

suppose that
HIM
|=

then
/')=0,
\J

liui l..Lr(l -r/* W MO

it"

//

is

any

tixnl
tin-

iiuinln-r.

I'.ut

tinl."7.

B6qU61

be

seen t'mni

Appendix, Ait.
will

Tin-

ivjulrr

have

no

ditliculty
!//,

in

prnviii^-

that

ll>

den

ndea for calculating


Iim(all
linn

limits a
liin6

6ll)=limall

]>r<\ idi-d

that

tin-

sequencer
liu

nt.

]T"\-id-d

tli

and that
^-n. -rally
/
.

1:

And
where

that

liin/(" a
J

,.)=/(HmamJ Iim6w>
any
n>iiil linat

lim
l'<>ur

demurs

in

of

th-

-l-iM-jisj,,.ritii-d.

operations, MiKj.vt to conditions similar to those ah-.-ady

10

SEQUENCES AND LIMITS.

[CH.

I.

If the functional symbol contains other operations (such as extraction of roots), the equation above may be taken as a definition of the right-hand side, assuming that the left-hand

side

is

indices

found to converge. On this basis the theory of irrational and logarithms can be satisfactorily constructed.*
remembered that the
limits

It is to be

on the

left

may be
.

perfectly
illustrate

definite without implying the existence of liman and lini b n = (-!)", b = (- l)"-^! !/). this possibility, take
tl

To

Then

a n + b n = (-

n~l

l)

/n

and

(a n

+ 6,,)-+0,

a n .b n =-(l

+ l/n)

(.&)-> -1,

a n /b H =-n/(n
so that these three limits are quite definite, in spite of the non-existence of Km an and lini bn
.

If a n is convergent

without

6 n ->0, we cannot infer that n /& n ^oo that a n /b n has a fixed sign. proving If a n ->0, and 6 n ->0, the quotient ajb n may or may not have a

and

first

limit (see Appendix, Art. 152).


.

Thus with aw = l,
Again, with a
lt

bn

= (-l) n /n, we
oo

see that b n-*0,

but a n /b n = (-l)"n and


1

so a n /b n oscillates between

and

+ oo

= l/n,

bH

=(

1)"/X the value of a n /b n oscillates between

and +1.

When one of the sequences diverges (say n -*oo ) and the other converges (say to a positive limit) it is easy to see that
<

(a&n)^c
.sequences (an .b n )

a n .b n -+x>

n /6 n ->oo

6 n /a n

-0

the only case of exception arising

Again,

if

and then the and (a n/b n ) need special discussion. both a n and b n diverge to oo we have
6 n ->0,
,

when

but both (a n

bn )

and (a n /b n ) have to be examined specially


,

(see

Appendix, Art. 152). If a w ->oo and fr w ->oo

there are three distinct alternatives with

rc-pect to the sequence (a n/b n ),


(i)

*<nin,n/ tlmt
(iii)

it

aw /6 n ^0;

(ii)

ajbn ->k>0:
(a,,)

a.
J,

/>

TlniH \vc can define A-V^ a8 lima?"", where n \\hcn iunl / ate botli nionotonic,
(
;.

= l,

H.

-*\-.cd not con-

tlu- srijiirn,

.\|.|i.-ii(lix,

Art.

l.VJ,

Kx.

4).

3, 4j

rri-l;

\M- LOWER

UM
.

11

hi

C
////

in
i'

<'

than

>'<.

In

eastr

(ii)

eon\ enient

to

use the notation

wbn
Kill.

iven

in

the
in

mhiat ion of
in

Mm

<

Appendix a number of

Ait.

\:,'2\

for

the d
in.

cases which an-

practical work.
4.

Upper and lower


upp
l.rm
I.
it

limits of a sequence.

rm
tin-

//.

this
\\
i

.Mid
is

similarly.

l.-a^t

callcl
1

th- lower limit.

r.ut

quence

term,
he. th--rc
is is

it

i'oll.\\-

tliat
in-1-

n.

may
such
tli

always
an
e

lar^-r

I'ui-th.'i-. tln-rt-

inimhcr of such
fcerm
in
]

imlici-.s

!:

t.tln-r\vis.'

ii

the

tile Sllpp'sr j, I,, index sati^l'yine- the required condition. Hence t)i- terms the sequence which fall het ween am and Op are all
J)
\\
'

Iliake

definite

than

a,,

and

('lioose

now

Succession

<>f

values of

/>

such that

"^""r
/'!>'-

">_-

P*>Pl>
.

Pt>j
Then WG
:
i

and
a

for simplicity den..'


!
>

-truct.-d

moiiotonic sequence blt and this se.jUence limit (Art. 2). either a finite numl.er // or ao. If lim
eaii

has a
//.

find
e

///

BO that

bm

lies

hetweeii

and

//.

no m.r

how

small

may
//

he:
, f

and consc,|Uently we ha

e<-/
/

<

//,

prn\ ided that

II

is

//'

///.

>//>{

lim
//

if

of

fi-

ll

'Hi/ iiff.iinn/

Similarly,

if

lin.

\\-

can find

,i>

so that

>
no mailer
h,,\\

A',

provided that

lar-e

A'

ma;

12

SEQUENCES AND LIMITS.

[CH.

I.

If the upper limit of the sequence is H, whether attained or not, the sequence has the two following properties:

term of the sequence is greater than H. least one term of the sequence is greater than (ii) however small e may be* But if the upper limit of the sequence is oo, it has the
(i)

No

At

He,

property

An
how

infinity of terms of the sequence exceed

no matter

great

may

be.

It is easy to

modify these

definitions
oo
).

and

results so as to

refer to the lower limit (h or

8
7.

1O

11

12

FIG.

The diagram gives an indication of the mode of selecting the and h\ these are represented by dotted lines. sub-sequences for

Ex.

1.

(Art. 1)
&

a w = 2?i-l.

Here we have the least number


Ex.
2.

=
,

and

so the upper limit

is

oc

A = l, because

is

in the sequence.

Here

(Art. 1) J5T=1, because this


0,

an =
is

l/it.
;

seen to be

which

is

the greatest number in the sequence and A is not actually attained by any number of the sequence.
1,
is
,

Ex.

3.

(Art. 1)

-i,

1,
,

f, f,

i
f,

|,
...

f,

f,

....

Here the sequence (&)


h
is

|,

and gives
to be 0.
lines.

H=l

and similarly

found from the sequence


1

^, J, |, ^,

...

These sequences are

indicated in Fig. 3 of Art.


5.

by the dotted

Maximum and minimum limiting values of a sequence. We have seen in the last article that any infinite sequence
lias

upper and lower

limits.
,

Consider successively the sequeiuvs


,

a lt az a s a4> a
":;

fi

,...,
>

"r
";

"
".,>

",.

<V".
on.

and so
*

If //"is not attained,

there will be ;m infinite niimlirr of surh terms.

4,

5|

MAXIM
I
II
\\<-

\M. MINIMI M
i|>jM-r

1.

Mil v
.t

ami I0W6T
:

limit- 1- d>-n<

<-d

l.y

>
:

//,,

//

'
:

//,.

A,

an.

BO OIL
",

Thru

may ha\r //,-/,.


i

in

which 0886
iae

miM
//

!>

th--

trnii in
M all

in-

sequ<
//.
>

w-

^liall lia\ e //

//

cam
<

//
i-

Thus

//
.-UP!

//

//

the sequence

//.

jnonoh.nir

givee a limit
tliat
.'/

or

/:

<Ar
a
in
Jl
i

Similarly

limit
OftSe

gTOr-foo,
//!
:

It

may

be DOticed
3c:

//

ran only mily be

-f

'/:

//
...

//

... = +

and

ran

00,

ii'

= h-> = h i=
It

=
t<

is

imp.M-tant

notice

that

./

CCW1

/-

//"

14m
rithrr
//..

//

...

,<//

Ic'loll-<>

t> tllr >.M|llr]ir

'I

which
and

thr sub-sequence

for

i-

rr.incidmt
//

with
II

<

//
... ',.

II

thru
tVi.ni

//,

is

Itself

thr limit

<,['

a c-n'tain

Bub-Sequel
(i.

("). so that this samr

suh-sr.jii.-nc,- d.'tin.//.

An
have

similar

ar^uiwnt
o

ajiplirs to
//'
///

con

''

#ub-&

</"< n

it

in,

if

n'l,ii-/i
f

is

greater
///

tha/n

G
//

///"//

//.

!'

lim//
small

lt

= ^r we
,

ran find

BO that

&+,
//,,.

no

matt.-i-

is:

luit.

ly

tin-

drtinition of
//

\v-

have

so

hat
ii'

_
/

(j -f-e,

if

//_

///.

thr limit of any num-r^-rnt 8ub-8equenC nm-t ha\r /i.^'-j-e: and. rlutrarily small. this \\c In likr mann.-r 0. requiree prove that m can found to makr a, if and drdnrr tlia //-e,
is

Thus,

Thr two properties


'

just
;/

rMaMislird justify us
llf
//<>'// /'///

in

ralli-

ant/

MM
1
! :

in

symbols we
.

<

'
.

1 i

Thr symhol

Mm"
/'.

is

n^-d to denote rithrr

th./'

maximum

minimum
If
it

limit: thus an inequality

/< lim a <


//

impli.-s tliat

/<</ and d

<

hapjirns that ( = x \\r ha\.th<Tr must hr an intinity of terms


,'
.

s
irr

an-;

..-ntly

than an\

14

SEQUENCES AND LIMITS.


similarly

[CH.

I.

number N, however great;


!)

when y =
N.

x
;

there must
if

an

oc co it is easy to see that and similarly =4-oo we must have limaw =+oo. From what has been explained it is clear that every sequence has a maximum and a minimum limit; and these limits

liin if

infinity of
,

terms

less

than

On

the other hand,

Inn a n =

lim/< u

are equal, if, and only It is convenient to

if,

the sequence converges.

call

sequences
are

oscillatory

when
shall

the
call

maximum and minimum

unequal. these limits the extreme limits of the sequence, in case wish to refer to both maximum and minimum limits.
It will

limits

We

we

be evident that the

maximum

limit coincides with

the

upper limit, except ^vhen the latter is actually attained by one or more terms of the sequence; and similarly for the minimum and lower limits.

123456789
FIG.
8.

1O

Hn

The diagram gives an indication of the process. The points and h n are marked with O and are joined by dotted lines.
Ex.
1.

In the sequence (Ex.


i
A. 3.

3,

Art. 1)

we have //"=!, A B =

so that

i t, i i # = 1, g = Q.
f
to give

Here it is plain that convergent sub-sequences can be selected any limit between the extreme limits. Thus
i
f>

...gives the limit

and

f, ft,

fa ...gives the limit -L.

v*

Ex.2.

With

8,

-f,l,
2,
J

fc.-fc...

,,

ll

=
(-])-i(i+l)
//,

we
ami

get

J71 =
//
1
f

//,

//
;

|,

//,
4

= !!,...

=/t.

-if, /<3

//

=-^
o-(-l)
can
QO

80 that

6 = l,
3.

#=
//
/(

-1.
i.

Ex.
\vc
tiinl

With

1,

-2,

3,

5,
A,,

<:,...

and HO
In

= 00, 6' = 00,


tin-

=
IH>

</= -oc.
suit -M'i|inMH''s
In-

ErS,
!

J,

:5

it

will
<.tli-i-

he seen that
than

t'ouiul

to

limits

cxticnir limits.

5,

6|

OF AN

MINITI: SERIES

der will timl some ralee


limiis in
6.

reme
1

Exa

17

the

'-n-l

..i

thi- cha]

Sum
we

of an infinite series

addition of two series.


a second

that

il'-tlm-.-

lY-ni

tliis

.s.-.|M-iire

;iMlt

'hat
x

l=

= "l + ":i +
and
i

Thrn
tl

/'/'

th,

A">

/A*

lim

"+
It

+-- =

'

*!//,/
is,

//,.
'

||iiw.-\
is

!ii'l;nii-nt;il

iiii])(.i-t;inc.IM-

in

iniii'l

that

"

'in it:

unl acconlin-'ly care must


that
familialpropi-rtit's
i'or

tak'-n DOl
tinitr

nme without jrnt'


aiv
rily tnir
t1><
/

oi

MUMS

limits

sud
or oedtto
to
nsr',ll,ih

Similarly, if

/,/

div(
1

reap

Ex.
-

1.

-f

/'-

1'"

+
1
;

'//'///

-'

1.
i,t

\vh.-n

/=!,

,,

/)

and

\\h.

=w.

-1 <><!,
|r|
\\hrr.'
/

(1

+ </),
+

i>

jM.sitivf.

Tln-n

|rj"<
I

(1

iy tin- liiniiinial tln-oi

cm.
n

Thus
ill

liinr"=0,
*oo
I

\vhifh was

ulitaiiii-d

indf]'-nd-ntly in
1
-

'.

~2.

II.

5
it

lim*
that

Kr<l.
/-,

is

nli\ j.ius

\,

-_

and KOOOrdil
/
.

lim
sd that
ll;

When
and
II-

is

1.

1.

\vr
(

have

r
1

f 6),

-r)">

d.l.

H<

16
Thus

SEQUENCES AND LIMITS.


lim
sn

[CH.

=-*

lim
oo

s,,

= + cc
oo
.

arid so the series oscillates


If

between
if if

and +
ig

r=

1,

and
Thus the

= l, sn = 0,
sn

n
n

odd,
even.

is

series again oscillates.


justified all the statements of the enunciation.

We

have now

It follows at once

from the
1

results stated in Art. 3 that if


2
a.

and
then

=a + a + 3 +... to oo T=& + &2 + & 8 +... to GO,


1

fifT=(a1

51 )+(a2

62)+(a8

68)+....

rule for multiplication of S, readily (see Art. 34).


It

The

does not follow quite so


of

should be observed that the insertion


is

brackets in a

series

equivalent to the

selection of

a sub-sequence from

the sequence (s n )', and since an oscillatory sequence always contains at least two convergent sub-sequences (those giving the extreme limits), it is evident that an oscillatory series can

always

be

made

to

converge

by

grouping

the

terms

in

brackets; and, conversely, the removal of brackets a convergent series to oscillate.


Ex.
306...
2,

may

cause

The

series

and

1-306...;
'306...,

but the
while
1

l-J + -f + i-f + --. oscillates between series (1 - |) + (|-f) + (f-f) + ...


-(i

the values

converges
the

to the
1-306....

sum

[-306. ..

= l-log2 = - + -+...,

-)-(-)-

...

converges to

sum

see Arts. 21, 24, 63.]

It is

evident that

mining whether a

series is

any

finite

number

are only concerned with deterconvergent or not, we may neglect of terms of the series; this is often con-

when we

venient in order to avoid some irregularity of the terms, at


the beginning of the series. In particular, it is clear that the series

The sum of the latter ore simultaneously convergent. terms of the former. the remainder after

is

often

1.
1

17

Arts.
1.
It'

3.

+ /'.
i

here

'.

<

3.

inn. us

fin.

:.!

if

Ex.
4.
I

5.

I'

t" tin- limit (a

+ /.
6.
If
i

..

prove that

(<>,

+
an

/,<,)
!'

(!+/).
tl.-

In

.!,

if

'

tllf

and
in.-

::iiiniii

val

similarly

tin-

iken,
/

7.

If
:i"ii

limit.

If

:IM"M linn:
!>n
8.
tli--

limit

is

(sinli

;n nlniM-tic

ami

li;irm..n:

hen
N

tip;

'nil- ami

inmoii limit

are

9.

If

".,- Ac
ii.'ii,-t,,ni.-

tlif

aiithiiM-r

ami

and
i

la
TT

rl

ri"

La

18

SEQUENCES AND LIMITS.


ll

[CH.

10. In general, if a H+ i=f(a ), and if (a n ) converges to a limit I, then I must be a root of the equation x = f(x). But there are two limits which can be derived from (a n ) by means of sub-sequences, they must satisfy the equations x = f(y\ y =f(x). For illustrations, see Ex. 11.
11.

If

a n+ i = &",

>0, a number
loga,, +1

of alternatives arise;
if

we can write
of the

the condition in the form

= Aa,,,
we can

A = log&.

By means
;

curve y = log.r and the line y = \x


(i)

easily prove that

if

A>l/e, (a n )
1

is

a divergent monotonic sequence

(ii) if

the equation A# = log x has two real roots a, /? (say <A< that a < ft) then the sequence (a,,) is monotonic, and a n -* a
/e,
;

if

!</?; but

if

!>/?, a n

><x).

When A
difficulty
/

is

the sequence
\
)

()

negative, the equation log#=Ao; has one real root (a); but will be seen to be no longer monotonic. To meet this

we may
|

write log (log >


line

n + l/

= Iog(

A) + Aa,,_i and use the curve


It can then be proved that

y = log (log(iii) if

and the

y = log(- A) + \x.

the sequences (a2n ),


(iv) if

(a-2n+i)

are separately monotonic,


if

and
a.> n

-e<A<0,
A<-<9,
if
a-j
:

a w ->a.

+i-*,
-

a*,,

-*<>,

a 1 <a',

but

^u,

a.,)+1

^v,

>a

and

= a,

if

a:

a.

Here

w, v

are such that

u<a<v

and ^ M = v,

^*

= w.

This problem was discussed in the special case a^ = k by Seidel (Abhandlungen der k. Akad. der Wissensch. zil Miinchen, Bd. 11, 1870), who was the
first

to point out the possibility of oscillation, in case (iv). Previously, Eisenstein (CrelUs Journal fur Math., Bd. 28, 1844, p. 49) had obtained the this series is the same as root a as a series proceeding in powers of A
;

the one given in Art. 56, Ex.

4,

below.

Arts.
-

4,

5.

12.

limits,

The reader may find it instructive to determine the upper and lower and also the extreme limiting values of the following sequences.
(1) a M
(3)
a,,

Th<

relations of the terms a n to the limits should also be considered.

= (-l)"n/(2w+l). = + (-l) n (2H + l).


7i

(2)
(4)

M
CT,,

= (-l)"( + l)/(2w + l). = 2w + l+(-l) M w.

13. In an oscillatory sequence there may be a finite numlx-r of limits derived from sub-sequences, all, sinnc, or none of the limits bring attained, as may bo seen by considering
:

(1)

tl

= Hin(ttir),
1

which consists of

tlir

sevrn numbers

0,

-ti,

i x/3,

all

repeated infinitely often.


jsin(jb<7r)
tin*

(2)a,,=

M4-

has
is

the

same

seven

limits

as

in

ca

luit

only

value n

attained.

.'

J.I

Tli.

ba

:i

\V!IM!.

..f
;

limits (>

l.x.

1.

limits
:n

may

nil

tin-

lin.

rith
tilt-

.m-t

"t

ail.

tli.-|vt'..iv

raiiUMt

f.,MM

''

14.
It'

liiii-/,,

/,
/,

lim<>,, = A'

and
/

liin

/'

lim

/>

= A,

+ /^lim(^, + /O = A'+A

a n.l

/'-/.
tinrt-siilts
|

K-l
tplej
|fi

iiiiilti|ilii-atinn

\vt-

may
I"

>M|)|M.>.-

A"

and A

Mil

<></

I= A'A;
it'

<></<

A'

and /<i)<r

A,

w.-

1:

,-,-A-A
if

/<>< A"
win-re

and

/<0<A,
A

\v-

i:

_lim ((//>)__//,
nuin.-rically
:

is

tlu>

/-A

and

A'/.

//

,.f

//

and

A"/..

nlvalsii in the liniitiui;

-iscs /---A*,

",

A'^
in
-:ist-s

!-tandinu
also

tliat

tin-

ruli-s <-an

iriv,-

D"

inf.>rinati..n

when

It

is

t,,

In-

present, M ,,ti ..... that


1

i-..Mv-ri:'-i

that

tln-sr

in. -.jualit

i'-s

ma\
jiarii.-nlai-

iv

nnK'h

\\id.-r

lii

',,6 W )

than
15.
\

in

'

is

usually
an-1

!we
lind

multiplication ly
(1)
If

tin-

rM-i|ii-,,,-al.

(</1

/.

and
l.ut
if

/<()

A'.

in

i'iu-1 -i

all;
i

:n

thosi/,

I-

tin-

values of

Jjive

no

20
16.

SEQUENCES AND LIMITS.


To
illustrate the results
:

[CH.

of Exs.

14,

15,

the reader

may

consider

the following cases


(1)

(O=

1,

i,
,

i,

... ...

&,,-* 0,

an /b n

(&)=-> +
= (2)(a,,)
(6,,)=
(3)
1,

-f,
^,

lim
lim

(&/<>=-<*, Hm" (6n /a w ) = x


(&,,)=-!.

-0.^

*
i,

...

-2, +3, -4, +5,


},
1,

..

(O=l,
(&)=!,
is

2, i,

J, 2,

3, i,

4,

..

a n b n -*
lira

1.

3,

(a n /b n ) = 0,

Era"

(/&) = x

There
17.
bility,

no

difficulty in constructing further examples.

Verify the following table, and construct examples of each possiwhere (1) denotes convergence to a limit not zero, (2) to zero.
to

(3)

divergence

+x,

(3')

divergence to

-x,

(4)

finite

oscillation,

(5) infinite oscillation.


CL n

I.]

LMPLE8.

22.
1-f
':

s.uin-

lines ax tin-

'_'

'lu-t lira

prOgTM0i<

'.

6.

Miscellaneous.
23.
If

the

sr|M.-ii.-.
/'ili

-rue of
tern

h.it

tin-.-

Sequence* vary

iii

tip

:ise.

-...+/;),

24.

.i

sequence (l-a")/n

<a<l,
>1,
ami examine th- eztenaioi]
25.
hftlur,t'

tln-^-

iiuMjnaliii-

frni
it

tin-

iii<-([iialiti'
I >

i-.-iuai'

deduce

that

limit

f(a\
I"

.'in.!

that
,

26.

"" th-f-. *
1

nattiiv nf thIf
ti,

i'".,t-

nf tin-

i|ii:ili-:.-

,,iv

ili-iiMtnl

l.v

a, /3,

\vi-

can

J.K.V.-

that

if

/j

|iiali"n

i'

Himtly,
r

-a,
ft\
;

j.i-oviilnl

that

in

i..iial

case
;al

\vh.
t.. til--

'

thr

f..nil

\\h.

-i

t-asily

expressed
will

lent

that

timl

that
|

ami

jirin.i:

i" >!"'

ial

CHAPTER

II.

SEEIES OF POSITIVE TERMS.


7.

If all the terms (a lt

a2 a3
,

...)

of the series are positive, the

sequence (sn ) steadily increases and so (by Art. 2) the series 2 n must be either convergent or divergent; that is, oscillation is It is therefore clear (from the same article) that impossible.
; :

series converges if sn is less (1) all values for of n.

The

than some fixed number


be

(2)

The

series diverges if

a value of n can

sn is greater

than N, no matter how large

found

so that

is.

Ex.

1.

Consider the series given by a n = 1/nl, so that


8

= i + J_ + J_ + JL
i

1
T

Compare

sn

with the sum


<Tn=l

+ + 02 + 03+
,
,

+ o^i'
;

It

is

clear that

and

so on,

= 3.2>2 n\ = n. (n 1)
3!
2
;

4!
...

= 4.3.2>2 3
2>2"~
1
.

3.

Thus, from the third term onwards, every term in <r n is greater than the and the first and second terms in the sums corresponding term in s n
are equal.

Thus

cr,,>s,,.

so that
<

x,

<

<r n

< 2.
its sum cannot exceed we ran prove similarly that
-2.

'"usequently the series


1'
l

2a n
e

is
1,

convergent and
as usual,

In-

Mini

is

denoted by

By

direct

calculation
1^71*-'*

to

decimals

we

find

that

l-l-*;

lies

between
lies

and that l/{7(7!)} is less than 00003, 10 between 2'7182 and 2'7183. Further calculations ha\e shewn
-2-7182818!

and

that
that

81

POeiTn
liaiiiK.nir serieH

i:

EB
\O

Ex.

2.

<'<>n-i(!.-i

tin-

On^ln). (r

li

-r<>u|>

is

a power of

-2.

ainl

Sow

i-oiiii;iiT

.<

\\itli

the

sum

-fi

-DHi

-*(^-4
>;tni-

as

in

tl.

Imt
tli-

all
i]

tl,'

last

t.-rm

!'

tlu-

u i"iij.

Tlirn

tion.
cai'li

Hut
n.iita,1
1
'

'_MiMij

in

<r n

(after tlf !ii>t)


.-.|ii:il

i>

t-ijiin

t.
1

),

Sn

>)t(nt + \).
<-<>n-i'.|mMit ly
//-

Timall

L'.V-l
tin-

and

t0TM

<l'"r<)8.

ten

lifl

wmiM
Hate,

in

t.idt-r

It"

iinilai-lv

.f...

*G
t. a

l''"i-

it

i:i

million trniis

is

less

than

i'

\N'-

DOt

that

"
5

+f+

24

SERIES OF POSITIVE TERMS.


The method used here can
easily

[CH.

II.

be applied to discuss the two series

But the

discussions in Art. 11 are as easy

and have the advantage

of

being more easily remembered.

The method given

in

Ex. 2 can be put in the following


if

rule (often called Cauchy's test of condensation): The series 2a n converges or diverges with ^Na N
;

N=2 n
>

and an ^ an+l and it is easy to extend the proof given above so as to shew that we may take as the integral part of kn where k is any number greater than 1. (3) It is clear also, from the results of Art. 2, that if we can find n v so that s ni s n ^>h (where, h is a fixed positive constant), no matter how large n may be, then the series must

be divergent.

For we can then

select a succession of values

n l} n z n s>
,

n,

...

such that

h
Thus, on adding,

-Sh

etc.

we

find that

and therefore
above.

nf

can be made arbitrarily large by taking r

sufficiently great;

and so the

series diverges in virtue of (2)

As an example,
because
so,
s

consider Ex. 2 above


s
ni

we have then

-s n >(n l -n)ln^
!/(??

~s n contains (ni-n) terms ranging from ni


?i
1

+ 1)

to

/^

ami

by taking

= 2w, we

get
*Jn

- *n >

(4) If

is
(

the

sum

the sequence
reach,

n ) increases to the limit

of a convergent series of positive terms, 8 the value of sn cannot


;

than

fortiori, cannot exceed S. the sum of any number of terms,


<i

ami

Thus S must be ^ivntrr


tlu>

;ies;

for
all

can be chosen

lai-^v

taken arbitrarily, in enough to ensure th.-r

includes

these terms.

On

the other hand, any

has the property that sum exceeds N e.

number smaller than N. (say N e), we can liml terms in the 861168 whose

7,

81
It

i-

now
M

cl.-aiit'

tli.-i

an
i

mini
IMit

-i

pro\,-i|

fco
-

ijM-il

iii

l.rack.-tx,
/,,

will still

wh.-n

IP-

\,-.l.

that all tfk


c>f

8.
l!'

Comparison test for convergence


til.'
t-

positive

f,

'';;

-f
li.-r

Mt

ami

if

anot

series

"
l

<i.,-\-<t.,-\- ...

lias

valut-s

of
ent,

ber

than >>MM-

ti\i'ii

wli.-n

>

a'

we

1'

4-...-f",
if
'/'

is

the sun.

Thi
-

[&BB

than a fixed number


for

independent of

,,

}.

wi

establishes the convergence In cast- the ine<|itality holds


s
',,

all

valmT.

we

hav.-

<C

BO that
mlititm

the
tliat

sum cannot
<///

t-xc.M-d

TinIn1

tin- ti-i-ins

must
Inf

It.-

jH^itivi' in

^1"

and
a
tinit--

ill

ii.it

atl'i-.'tfil
tli--

liy

tli--

.mi--i..ii

,.f

nun

'

in>

at

tin-

lM"_rimiinur

taki

tlr

+ i_

with

in

tin-

>,.,-,,

ml m
;

'

-(. inline
"-

tri-ni

in
:

.uid tini

Hut

tin-

in

thi-

timl

s-.th.-u

In

26
9.

SERIES OF POSITIVE TEEMS.

[CH.

II.

The comparison
is

test leads at once to the following form,

which

often easier to
r

work with

Let the series 2(l/C n ) be convergent; then

^an

will converge,

provided that

is

For,

not infinite, both series containing only positive terms. when this condition is satisfied, we can find a constant

independent of n, such that

Hence a n
convergent

is less

than G/Cn which


,

is

the general term of a

series.

It is useful to

remark that

there is
;

no need

to

assume

the

existence of the limit lim (an Cn )

this is seen

by considering

the convergent series o

for

is alternately equal to 1 and 2. the test is sufficient only and is not necessary Further,

which a n Cn
see

as
t

we may

then anGn >n/2 by taking But Ea n converges (see Ex. 1, Art. 6). so that lim(a (7n ) = oo. The corresponding test for divergence runs
1

Cn = nl and

a.n

= l/2 n ~

ri

Let the series 2(l/Dn ) be divergent, then

2aw

itfiM

diverge,

provided that
lira( n Z)n )>0,
/>o///

series

The proof
tion,

containing only positive terms. is practically identical with the previous investiga-

when the signs of inequality are reversed. We note also that the limit lim(a n Dn ) need not exist; and that the test is not necessary.
follows immediately that necessary but not sufficient
It
:

the

folln-! ,/</

rttmlitnmx are

For convergence,
for divergence,
lint,

Mm Mm

(a n

Dn ) =
'

(",/')

in

^vn.-ral

tlicrc

is
////

no need for
condition,

tin-

limits

<>!'

(<>

1>

of (a n Cn ) to <-xist: and

[im(anDn )=Q, sometimes

given as necessary for convergence,

is incorrect,

9,

10
1

BS]

Ex.

If

\Mliiu

tilnl
i
.

iiat

liiu

(a
in

/'

will

lie

seen

Art.

1.

It

is

eaay bo see m lim (/'">


il

tliat

if

MK

t>

///-

<>

cmnlitinn
i

liin

is

.....

rergent,

we ran

-li-

-...+"<.
\M\V
i-acll
!'

tllrsr
(

t.TlllS
a/

is

i,,,t

1,

BO tliat
/'

)'',.<
iii

e,

it'

r.m
it'
//

choose
an(

>

> >

//'.

iiat

//"'<

6.

'.

Tims

/"/,

<

-2t\

if

//

>
.t

r.

cnn.s.-|U-ntly

liin

nOn)f..ll<.\vs f i.n.

Tliat tins ciiiiilition


</

is n.

snllii-irnt

\it. 11)
liiu (/"',,)

which -ivrv

di\
:

.iltlmuirh

= 0.

"li(liti"M
li'ini

\\hi.-h

("

such as lim(. //>..) (l ilv than N and r\ani]> DO ili-liiiitr limit will le found in
;

L'I

iiiLTshrin

HIM--

M-t inijih
/>).

10.

It'

C(,lii],;l!V.l

\\ill)

til.

iiil'.-r

Cauchj

wliicli

i-

tlj.M.n-tit-ally

of

__
If
It
i

liin

"">!.
:-i
.

ti"

it

inijM.rlaiK-,\
I

n-nn-inl-i-r that, in contrast

\vitii

tin-

rat;

J.

limit in^

\a

28

SERIES OF POSITIVE TERMS.

[CH.

II.
!_

n Further, to ensure divergence, it is not necessary that a n should be ultimately greater than unity, in spite of what is
i

sometimes stated in text-books; and if a nn oscillates between limits which include unity, the series diverges*
To prove
these rules, suppose
first
j_

that

lima n n = Gf
Take any number p between
_i_

<
;

1.

6'

and
1
,

then we can find

rn so that

an

<p<
is,

if

n>m.
2
ro

Hence, after the rath term, the terms of


convergent after p terms
series S/o
is less

are less than those of the

n
;

that

2a n

is

convergent.

And

the remainder

than p p /(lp) provided that


will

p>m.
n,

But
(say
n-t,

if

lima n w >l, there


, ,

be an

infinite

sequence of values of

n2 n3

...),

such that
i_

a n n >],

if if

n = np

and therefore

a w >l,

n = np
,

Thus the sum 2a, taken from 1 to np must be may be taken as large as we please, so that 2 n

greater than
diverges.

and

We know from Art.


and diverges
if

54 that lim a n n

lies

between the extreme


if

limits of (a n+1 /a n ); thus the series converges


limO',,
.

\im(a n+l /a.n )

<

1,

") >>

1.

This shews that d'Alembert's

test (Art. 12) is a deduction

from Cauchy's.
since

falls

But on the other hand, between the extreme

we only know
it

limits of (an+l /an ),


test

is

that linuf,," clear that

we cannot deduce Cauchy's


d'Alembert's.
If

in

its

full generality

from

we

consider a power-series 26 n aj n (in which b n and x are


for convergence,

supposed positive), Cauchy's test will give

#<!,
where
./

and se>l,

for divergence,

Z=lim/v.

Tims and divergent


This
.

= ]/

gives an exact boundary between


series,

com vrgrnt

supposing
at
\

to be diflrrcnt

from
niadIn-

/m

and

rni^lit
'Jncl

SIM-HI

to
<-li.

In-

ariaiicJj

\\itli

slat i-im-nl

\>\

('lirvMal

edition.

\\\l.
of
(

">,

I):

Imt

his

ivinaik must
in

umli-rst

cod as
.-

implying the
toi in
.,!'
t

insiiHii-ii-in-y
is

tin-

nictliod
5
1 1

ns.-d

that

atticlc.

Tin- roniplrt

In- t.-st

^ivi-n

iy

'hrystal in

of tin- sain.- c!ia].l.T.

10, ll

I\TI:;I;
I

\ i.

MM

li

it

It

'

= 0,
\

the

condition

all

p-.-ii
I

ive
\;dil>

aln

x-

tin-

rge

r.ui

if

\\.-apply d'Alembert's
t
!

test

to the pov.

mly

inf.-r

;id
\vli.

./

>

'

'iat

gr-Umi (i are when unequal


'/.

and

t;

Em(6
may
IM-!I;I\ i,,ur

(as they
tl,,.

I'htain

no

information

as

t,,

ra-dly of tl

.</<-'<
In

this

ti

'on.
th.C
it

d'Ali-nilf-rt^
f

t.

aMish
this
t.-^t

th-

of

its

simp!
\\id.-ly

(with

in

ordinary \\ork.

Second test for convergence; the logarithmic scale. :nv ,-in-an^i-il Suppose that tli.- tennfl of a p,
11.
ord.-i-

in

of ma-iiitudf. so thai

>

O.

may happm
is

that th- function

also di-linilr

for

\ai

which

aiv
if

not
./

ii

md
beti

tha-

ever ina
//

Th.-n.

n,. s

and

-f

I,

it

i->

plain that

>o.
Thus. iVoin
th,.

d,-tinitioii

of an

int-'-ral.

we

ha\--

r-fi
<>r

J n

=|
.ml

+r

we

find

Add'

.(iialit

ii-s

we

h:.

11.

!'

>().

30

SERIES OF POSITIVE TERMS.


Further

[CH.

IF.

(s n+l

- 7n+1 ) -

(.s

- 7J =

n+l

- fn+lf(x)dx ^ 0,
Jn

and therefore the sequence whose ??th term is s n increases', and since its terms are contained between the sequence must have a limit (Art. 2) and
a,

In never and a l>

Slim (-/.) SO.

Thus, the series Sctn converges or diverges with the integral*

f(x)dx ; if convergent, the sum of the series differs from the I" integral by less than a l ; if divergent the limit of (s n In )
,-

nevertheless exists
Ex.
If
3C

and
+ l),
is

lies

between
!),

and a r
and f*/{*)drlog.8.

a n = l/n(n
ci n

/(#) = !/#(# +

And 2
in

= l, which

contained between the values log 2 and ^ + log2,


result.

agreement with the general

A large number of very important special series are easily tested by this rule
:

(1)

Consider
if

-+

^4-+...,

where

ci

= n'P.
and gives

Here,

is

positive, the rule applies at once,


^.

p
thus the integral to
oo is

(^i-p-1);
if

convergent only

>

1.

Thus

the

given series converges only if p^>l: and the sum is tlifi) contained between the values l/(p 1) and p/(p l). If p = 1 the integral is equal to log x, and shews that the
,

harmonic

series is divergent (see Act. 7)

we

infer also that

n->cc
.xists

and

lies

between

and

1.

This

is

Euler's or

M*rln ',<>,/

/".s-

constant.

The convergence
L

of the st-rifs used in

Art. 9

+ gf+3

^,

+ ..^-^ +

^-1-...

can
*

in

inferred.

The

integral converges or <livrrm:s with the scqurnrr (/): for further details
iix

III.

LOG
an

\i:i

HMi'

9<

\u.

in-i'i.i.-.l

'+V4* 9* 16*
leM than

'

toi sum
'

ii

lonmder
\\

\Qogi)
./,

'.'

liidi

()

and

II.

f-aTHl
!

<!-/'>

Of

=
Thus
1 .

]<Mr(lo^r

if

/'=!
'^

tinit

^ivi-n
!

seriee
n..t.-<l

com
that
if

/'>! an
//=
!.-;
it'

'^'

-lioiiM

1.

th-

li\.

s|.\\.

th-

MUM
in
.t'

..I'

lillinn

t.-nns

I).-

pi-ov.-il

similarly that
t>

\\v
all

>mit
tin-

a
1

suti

nuinlx-r

tin-

rarly t-nns
it'

nisun- llmt

'^arit

Inns

aiv

jMi.sitiv.-.

ami

n n =(
or
if

i'

>

I.

<1.

i'

/'

=!
that
tli-

I
I

F'I

slog
/

[/'

tWO
<li\<

int.-^i-als

nnl

/'

:iv.3

lnit

deci
|>r..\

a limit, tin-

>am-

will

IM-

tn;
l.-ilnr.- tli.-

il-l

tl

Ifl

]sitivi- anl BO

we can

r.-nlt

1.

This
1.

iv.suit

'hat

th.

;n

in

binued
i.

\vith<-ut

For

nli

work. th- two types

are

32
The following
Let

SERIES OF POSITIVE TERMS.


results,

[CH.

II.

which are independent of the Calculus, have a


:

field of application

substantially equivalent to (4)

(Mn )

denote

an increasing sequence such


while
p~ ^ (Mn+l n )/'M H sum of (Mn+l -Mn)/Mn
l
l

that

limMn = cc
is

then

are divergent
For,
see
if

series,

Mn+1
as
n

convergent if

p>\.
to
r,

we take
its

the
is

n ranges from q

we

that

value

greater

than ~Z(Mn+l i

M )/Mr+l
^,

(Mr+l - Mg)/Mr+l
no matter how

because in the summation J/n

<J/r+1
sum
is

We
is

can choose r large enough to


(Art. 7 (3).)

make J/r+l ^2Mg


large q

may

be.

and so Thus the

this

greater than

series diverges.

Similarly, 2(J/n+1
If
/?

- M n )/Mn+l

divergent.

= 2,
;

the third series reduces to


if

^(jr--*/
is

:=
)

17

'

an ^ so

con "

vergent
series,

thus

p>

2,

the terms are less than those of a convergent

and

so the only case left for discussion


24, Ch.
I.,
( 1

given by

< p < 2.

From Ex.
Write now

we have the inequality - a m )/m > ( 1 - a n )/n, if m< n.


an
1
c,

m/n = k,

and then we have where k is a proper

c*>(l-c)

fraction.

To apply

this

lemma, write

and then we get

Mn+l

^ -M

< -L- [l -

(^
1

)" ']

1/1

From this it is plain that the given series has its terms less than those of a convergent series.
Ratio-tests for convergence. Rummer's test for the series 2a n runs thus:
12.

//

SAr

is

a divergent
if

series,

then
t/

2aM

is

(C) convergent,

lim (l) n
\
''

"
. i

L> /J+1

)>0, X

I,

if

"'"

lini(/>,, >
is

' '

>/

AH-I)<O. /
limit
IK-

For

in

tin-

first

case,

i!'

//

tin-

minimum
//,

;m<l

//

is

any
tl-.at

positive

numlxT

less

than //.an intr^vr


'

can

found

siu-h

^n)
-n-fl

i>

''

'*'

11, 12

i;
1

P8

Tl,

Mini:.

W(

II-

+<..+an <amDm /h
-inn
ft

nn

lli'-

ri-'lit

</<><>

SO

5]
tllfl.
Iii
lit.

fixed

number,

and

tin-

v,.r..nl

case
'

hat

/>
;

<<>,
I'n-i-

if

,/<,

if

II-

if

>
than
Phllfl
tl

ami BO
of

th.-

term
861166
<

aft-r

tin-

th'

diveruynt

!'

i-

also

_--nt.

Special cases of importance are


(
I

</'A/< n>l

/'
I :

tlu-n

tin1

condition^ are

i>
In-

(D)

In.

)<1.

e-ninpaivd with Caiu-hy's test of Art. 10.


>!'lli<-<l
t.t

Ex.
that
it

1.
..

If

th
i

rhf >-rirs

-...

we

see
if

.,

<

i,

,1.

>]
'lit-n

l,ut

the test jfives no result

ol.vimi>ly

!.!
/'.

wh.-n
;,]v

lini

(OM

'.,

1.

thrll

tll-

minlitinlis

hn.,

-!)]>!;
x

(D

IS

-!

EX.

2.

If

w.-

tak.-

^)
.1

and
'

if

ft>a+

1.

It"

o'

th.-

with
I

s.

34

SEEIES OF POSITIVE TERMS.


(3) If the limits used in (2) are both equal to 1, delicate tests, found by writing

[CH.

II.

we must

use

more

Dn = nlogn, n log n log (log n),


These functions are of the form
f(ri),

and so

on.
is

where f(x)

continuous

and f"(x) tends to zero as x tends to infinity. Then Rummer's test becomes (C) iim^ w >0; (D) lhn~/c n <0, where
For
f(n + 1) -f(n) -f(n} =
[f(n + #) -f(n}}dx =
if
v.
\

^=

ffaf*f(n

+ t)dt.

Now we
is

easily seen to

can find v so that. \f"(x) < e, be less than ^e, if n >

x > v, and Thus

so the last integral

f(n + I)-f(n)-f(n)-*0, as 91^00. Writing f(n + l) and f(ri) for Z) n+ i and Z)M in Rummer's at once to the form given above.

test,

we

are led

In
f

particular
1/X',

f' (x)

x log x, we find /'(#) = log ar+ 1/ thus we find de Morgan's and Bertrand's first test,
if

f(x)

(C) lim

/0n
"

>l;

(D)

KS /0n <l,

where
Their further
little

= 1 + - + ~^
etc.,

tests,

given by f(x) = x log x log (log a;),

are of

practical importance.

(4) It is

sometimes more convenient to replace the


(C) lim
<r n

last test

by the following:

>

(D)

!W <
7l

1,

where

w_^i-=1

After a certain stage, we have


also

<an /a n +i
if

<^-lo g (l+^)=[o r^^<^,


we
see that

thus

<p,,-(r,

< 2 (log %)/,


cases

and
in
in
'"

so p n -o-, -*0.
t

(5)

The most important

the quotient

ajan+l

being expressed
<*

practical the form

work admit

of

an+1
where
yU

=!

ff
,

?'

is

remains

less

a constant, X an indrx ^r'at-r than than a fixed nnmlMT .1 I'oi- all values of

1.
//.

and

12,

13
1

i:

\Tl'

'

It'

osion
I

ii

tort's

//<!.

T.

diaciiflfl

tl"-

ease

//.

I.

apply
<

tin1

t.

ider

tin-

limit

.i

~A

Hut, Since
\\-

<
1

.1.

ami lim

<

log

<.,>

= 0,
may
th-i
!'

ha\v
//

lim

UY

I..--

.=0.
\\V
rQfl

Thus
up
in
th.-.si-

ivMilt> in

the

working
tintlu>

rule (^s.-ntially

lu-

t<>

his
//'

in

ins

mi

Hypergeometric

quoh

>

i'

fl

^_
"
tl"
It'
-

>

//

>

1.

we

a). ply

tinit

ivsiilti>

"i'
i

Art. 39,
t>

Iv-

'i'-nt

imt
the

Milieu It

|ir..\-.-

thai
)

/'//

form' above,

tl"

condition lim(/' n
tin
<-'>ntra-t

=o

"nd

-suits

I'm-

sufficient far seriee in ^-iR-ral, AIM.


tin-

with

Ex.

3.

('uiisid.-i(/.( (/

Hy])fr^e')iiH-ti-i'l

S.-i

ojy

M.y*"
tfing
/

1.^. ;<y

+ l)^ (/j + + l).


i

-.3 7(7
.

-^(^ + 1X7+2)
family
^f-ii
t<>

d'Aleml

tliis
it'

snifIf
'

i-

converge

<

1,

and

r <' t" divei;u

J?>1,
"

!,

(>t

+ l)(y + /Q_ ~
/^,

+ i-a-0)coir,

\\hifh

'_

i\r^
if

/t

= 7 + 1 -a
/^.

so that

tin-

si-rii-

di\-iu
It

'--

=a+
1

will

aj|..-ai

from
that

Ait.
it

.".<

>

that

the
als..

M-!

if

-1<

and

fiMiM

Ait. ^

.-oiivcr^s

fop

./=-],

if

13
It

Notes on the ratio-tests.

'

tn

I.,-

imti-.l

that

'/'.I

n
lliat

will n>t

In-

jM.sMhlr to find a nuinlH-r


f,

^.-1>A>^
:

In

llivi-i
[

36
Secondly,
it is

SERIES OF POSITIVE TERMS.

[CH.

II.

not necessary for the convergence of the series


definite limit.

Sa n

that
it

an /an+l should have a

For

will be seen in Art. 26 that the order of the


;

terms does not

affect

the convergence of a series of positive terms the order may affect the value of lima n /a n+l

but

of course a

change in

Ex.

1.

The

series

a+

+ a + a + a + a + ...
3 2
5
4
,

is

a rearrangement of the

3 geometric series 1 +a + cr + a +... and so is convergent if 0<a<l. in this series the quotient an /a n+l is alternately a and I/a3
.

But

Ex.2.
is

Theseries
if

+ a + /3 2 + a 3 + /3 4 + a5 + /36 +...
;

convergent

<a<@<I

as

is

plain

by comparison with

In this series

we have
lim a n /ft n+1 = 0,
llm
n /3

/a

n+1

= oo

But even when


of the

the terms are

series does not

arranged in order of magnitude, the convergence imply the existence of the limit.

Ex.3.
has
its

Theseries

l+-ia + f a+|a 2 + ia 2 + a 3 + ^a3 +...


5
;

< a < 1 and terms arranged in order of magnitude, if 2 3 2 3 convergent, by comparison with l + a + a + a + a + u, + a +...
.

it is

then

But yet

lim(a n /a n+1 ) = l/a,

Km (an /an + = 1.
:)

Thirdly, if the quotient an /an+l has maximum and minimum limits which include unity, the whole scale of ratio-tests will
fail.
For,
if

\im(a n /a n+l

)O>

>g = \im(an/an+l we
),

can take

A",

k such that

and then a n /a n+l


while
If
it is less

for an infinite sequence of values of is greater than than k for a second infinite sequence of values. n belongs to the first set of values, we shall have

n,

n(a n /a n+l -l)>n(K-l);

but

if

it

belongs to the second set,

Hence
extends to
If

limw(a n /a n + - 1)=
1

+ 00,

lim n(a,,/a n+1


It
is

- 1)= -

co

and therefore Raabe's


all

test fails entirely.

easy to see that the failure

the following tests.


test to

we apply Raabe's
lira

Ex. 3 above, we iind

?*(/+ ,-!)+ oo,


we
get

Hn7i(a M /a M+1 -l)=l

and passing

to the next stage

imi (log n)[n(a n /a n +i

-!)-!]= + 00,

= lim(log70[/<(",, ",,-n-l)- 1] 0,

so that the ratio-tests can give no information.

13, 14]

BUM LKO1
he se,-n

It

will

from
i\

t:

mparat
it

fly
-I.

limited

ran^inl

may
I

why
<

due--

ln-iii

at

all.
I

and
Tin-

ii"t

In-

should w- tr.,uhl- bo with the n,


is

oompaiiBon-teftfl
<|iiotien!

answer
'i

to this

u,at. in
j

much simpler than


his

/.

and
I

l!

easier

t<>

086 the

ratio'

they apply) than


t.-st

Historical Note.

Kuniiner him-

in

the form

t'oi-

convergence, where
siil.jrrt
1..-

</><//>

is

an arhitrary Bequenoe
ion
liii;.

numler.s.

to

tin-

r ->t n'et

=(}.

whicli
i

was
to

pro\vd to
expresai
di\
.

siijH-riliious

hy
l'.

|)ini.

l)ini

aUo was

ohtain the condition

in

tin-

used
Kin-tli-

to

.rm ^iven above, test both for c-.


LOna ha\

where the i
hy I'rin^^heim.

14.

Ermakoff s
Mffiee
1

tests.!
in

The
of
Art.

^f{n\
(i
)

which f(n)
if

i-

suhject

to the conditions

I.

is

convei^vnt

Inn
X
'

<
'

1,

<

(ii)

divergent
i'

if

lim

'>!.
/
.

X>

'

I-'"]-,

in

tin-

tii

/>
4"

is

any nuiul.rr

!..-(

wmi

tli.-

in.ixiiiiuin

limit

and unity,

wt-

ran tind

s"

^ 1;it
if

<>.
-V

'

>$

Jug
\vli
.

tin-

ind.-prndt-nt
.

vari.-il.l.-

to

'

in

tin-

l.-ft-liand

int-gral,

jr

Th

(l- r
I

v.-iriatic;

ha\v

Ix-i-n

Lriven

liin'> an- uixloul.u-.lly tin- UK-:


-

bo
tlu-

u-e

iu

p-.Miu-triv.-al
f.>il-

represeiitatiou of f/(x)<fe as

aiva of

tin-

cui\.

\vlu-n

38

SEKIES OF POSITIVE TERMS.


Or, again, since the last term in the bracket
is

[CH.

II.

x is positive, because e

greater than X,

we have
"

'

As

this inequality is true for


/

any value

of
;

X greater

than

it is

clear

that the integral

f(x)dx must converge

and, therefore, so also does

the series 2/(w). In the second case,

can be found so that

As

above, this gives

7*

or

Jx

f(x)dx
.

f(x)dx,
f

if

how
is

great

X may be,

number X' = e x

/oo

f(x)dx

is

divergent, because, no matter


/

can be found such that

f(x}dx
(3).

greater than a certain constant

compare the argument of Art. 7

Thus the
These

series 2/(w) is divergent.

tests

include the whole of the logarithmic


p
),

scale.

For example, consider

f(x)=l/{x.\ogx. [log(log#)]
then

Thus
and so
or

ex

= e /{e .x. [log x]?}. f(e )/f(x) = [log(log x)y lim e xf(e x )/f(x) = Q, if
e x

f(e

,/'

^.

oo

= oo
That
It
is
is,

if if

1.

the series 2/(7i) converges

p>-l and

diverges

if

increases

easy to see that if </>(x) is a function which steadily with x> in such a way that (/>(x)>x, the proof
be generalised to give ErmakofFs tests:
.

above

may

(i)

convergence,

if

x->

p ^ lim 4/(x)f(A(x)) 7x T <1, J\ x )


'

(ii)

divergence,

if

lim xi^;

*'<)/(())> L
./('')

14,

15
1

LOG \i;n H.MK

39

15.

Another sequence of

tests.
;--s^

in

Although the follow or it tlin.-iry wurk,


Ait.
I'".

imp..:
,i

<*-tecU>

continuation of

('and,

seen
_T.-S

iii

An.
i

ii,

that

if

17-"M

is

divergent,
'8t
:

"iily

if

p>l,

;"

>o
!

wh.-i.

and
11

di verges if

-!,,.:[/ Inn
log[/X tt)j

:aJ

<l

II .ut tends steadily to


J

0.

th^u.-h

lirst
tli.it

CM6, as on previous occasions,

we

can fiiul/>>l and an

logK
-/"(/')[^( /)]",

-r

>-,

>
(

This

>h,.\\>
in

that

the second cast.

f .-<.ni|>:iri80n. Pgea, ly thr pnnriple ~u-h that

(/_
tlii>

slu-ws that
;

l'<r,

S|i.-.-ia

(1)

/'() = /j

i:i\''!i ly and the function to examine

iliv.Tges. .f this test arc-

is

logO
logn
mction
is

log(l/wa,)
log (logn)'
;iicti"H
is

and
in

>n.

D
\\hi.-li
It'

I,,-

tnuuformed into another


t>
(

-liape,

fii-st

given by Jamet,

the

i.-lati"ii

'a'h
).

h\
it

\v write A = log(l

,/.

is

ea>y

t-.

>ee tliat

it

A>
iii
i,

Thi;

;^=[
if

i^

(1

"
is

provided that lim (A /<)-<>: and,


Will
l.e

this Condition

tses of practical interest,

net satisfied, Cauchy's the test

40
Similarly
it

SERIES OF POSITIVE TERMS.


can be proved that test (2) can be replaced by

[CH.

II.

[n(1

- a "" } " log n

log flog *)

^> 1

'

0=^=1
16.

This form proves for example that the series but converges for x>\.
>

--lognj

diverges

if

General notes on series of positive terms. Although the rules which we have established are sufficient to test the convergence of all series which present themselves naturally in elementary analysis, yet it is impossible to frame

any

rule

which

will

give a decisive test for an artificially

In other words, whatever rule is given, a series can be invented for which the rule fails to give a decisive
constructed series.
result.

The following notes

(l)-(3)

and

(8)

show how

certain rules

first sight have been proved to be either incorrect or insufficient. Notes (4)-(7) shew that however slowly a series may diverge (or converge) we can

which appear plausible at

always construct series which diverge (or converge) still more slowly and thus no test of comparison can be sufficient for all
;

series.

sidered

Other interesting questions in this connexion have been conby Hadamard (Acta Mathematica, t. 18, 1894, p. 319,
t.

and

27,

1903,

p.

177).

has pointed out that there cannot be a positive function <j>(n) such that the two conditions
(1) Abel,
(i)

lim </>(n).an = Q,

(ii)

lim

</>()}).

an

>

are sufficient, the first for the convergence, the second for the divergence of any series Zf,,.
For,
l if so, 2J[<(wj)j~

would diverge

and
1

therefore,

if

Mn = wo)]-* + wa)]- +

. . .

+ ooor,
.

the sequence n would be an increasing sequence tending to oo Hence the series 2 (Mn n -\ )IMn would diverge also (Art. 11); but

so that

lim <l>(n)(MH first

)
,

-I/,,

=0,

contradicting the

condition.

(2)

function

Pringsheim has proved that there cannot be a positive c/r such that the condition <f>(n) tending to
,

15,

16
1

U
.

function
BeriOfl

'id
I"
1

for

any com
'

an

n-ar
lin,

fiat
'

=
II

J0

'

'

See

Mth. A
ini

\>.

:ni.

lias

pi-..

v.-l

that

th-r.-

i>e

a po-

function

uch that

tinlin.

condition

I'm-

tin-

dhvrirence of

>() 1"

In
;

function

r/>(//

ami

"//// di\<

DQ8 of tin- 861

can

I)--

BO arrani:v'l that

lim
i'l.-<l

=0,
the
!

that

tin-

t.Tin. .

!'

nl

to zero.

SetA1..-1

!.

8ft,

p,
il'

r.-iuarkt-.l
ii<l

that

1
also <Mv<T---iit
.

which

is

l>ut

BOCfa

lim (6^
write
i/ M = a +
>

+
-

..+//,

l,,

=
;

<i n

n.

MriM

-'

I.v

An.

11

and

(5) dii

1'iuis

IH-VIIIOII.!

shewed that
!/>

it'

^<>

i^

r ,n v.-r-vnt.
,

second

OOnvergeni

Beriee

can

1-

Tound

which

has

thi-

rty

MI.M

.!/,

-a,

J/,..,

.<

,/

_ 1+ ,,, i+ ._,4-

Then

J/ M ->x

and

<-..nsMjiu-iitly tin<

6M =
1'iovidfd
tliat
1.

7
it

is

jM.sitivi.-\

>

11).

idrnt

tliat

->/.
.

Sti.'ltjrs

sh.-w.-d
:

that
as
1'

il

...

i>

d*

ndin^

limit,

di

can If I'mind so that

L/m,; then

seriea

M + -M
n
}

1
''..

"~F^T ^fw-w..-*-!
BO that

i/~"' M

42

SERIES OF POSITIVE TERMS.


(7) Stieltjes also

[CH.

proved that

if

v lt v 2 v 3
,

... is

an increasing

sequence, tending to infinity as a limit, a convergent series 2c n can be found so that Zv n cn is divergent.
cn

i'

ll

For, write cM = l/vM -l/v,l+1 which makes 2c n a convergent series; then = (vn+ i-v n )/vn+ i, so that 2vn cn is divergent.
,

(8) Finally, even when the terms of the series 2a n steadily decrease, the following results have been found by Pringsheim

However

fast

the series Sc" 1


series
<j>(n)

may

converge, yet there are


0.

always divergent

2an such

that limcw a n =
oo

However slowly

may

increase to
,

with n, there are


<j)(n)
.

always convergent series Ean (although Iim7i.aw = by Art.

for
9).

which Kmti.

an =

<x>

See Math. Annalen, Bd. 35, pp. 347, 356.

EXAMPLES.
1.

Test the convergence of the series


:

2# M

where a n

is

given by the

following expressions

1
"'

(log)

10

n>

10

[log(logw)]

M'

a 1 /"2.

(Art. 15)
1.

(Ex. 25, Ch.

I.,

and Art.

11)

Prove that

if

6-l>>0,

the series

a(a+l)(q+2)
converges to the

sum

(6

1)/(6

1).

Shew

also that the

sum

of

a-l)(6-a-2),
[If the first series is

if

&-

denoted by w

?i

+ M 2 +..., we

get

\\hidi

Hence (6-al.\

l)(x

(l

- V ) = a</,, -(< + /<)", hy

addition.

But

Ait.

The

Hence lim* w can be found. >ince the terms st.-adily decrease. second series can be expressed as the difference between two series
!,

of the fust

ii.
I

kiFLEfi

3.

I'iv.- that

tl

(20 +

1X30+2)

\xv*
4.

if

6>f>0
that
tin-

and dm-i-en
-Ties
'

if

|'|..\.-

'

+
l

1
I

3 5
-

irs.4.6

ulso that
3

+ \S74j
thrrwi-.- di\v:
'-1

V2.4.6

if

f>>-2,
that,
1

and
if

5.

I'l'-vi-

O<Y<!,
+

|i'-itiv-

tliat

)3

~ +"| +
.!

COOV-!

-.

>/3>l,
and

lint

that

tin-

ratio
6.

..nsrcutivr t-nn>
I'rnviI'
.i

i..-.-illati-

Between

10).

that

-n
in

oonvi
+ i.

d-<-reaaee t

i,,

7.

Sliew that
..f
1

Kunn
</>(/0
is

n-r i-(Hi\-.T^fn<-.-

\vt-

,-;IM

writt:

;.

pla<r al>u that

/>

whfiv
is
-

an
in

arliitrary

jx-xitivr

fun-ti<.n

Imt

\>\

tli.'ir

no advantage

making
aft

this .-liaiige.
in

convergent

if,

stage,

[Or
8.
ir
-

th-

jii"tirnt

"f

two
1'"
[I'i

|x.lyn..iuials in

degree
lii;h-st

X-,

whose

hiirln-t

b
div.-i^.-s

d
if

if

the

term
if

in

'/

POT

that

^>1,
us.]

rges
9.
1

^=1 and J>1.

10.

Kind limits fm

tli.-

sum

^\'(
in
tt-rins

of tin- inr

'^
s

and drdiu-f
Siniilarly

in<-iva--

in\

44
by means

SERIES OF POSITIVE TERMS.


of the integral
C
n

[CH.

ndx _

[i

dt
2
t'
'

Jo

x2 + ri2 ~Jol +

and deduce that


11.

cr,,

->

|TT.

Prove that

if

approaches zero through positive values,

and that

where

is

Euler's constant.
if

[Di RICH LET.]

[To prove the latter part, note that

f(v)

is positive but less than can be put in the form

l/v.

The desired
1

limit is that of

/"(I),

which

From this expression equal to that of

we can

easily see that the desired limit

must be

12.

More
n is

and

generally,
zero),

if

= an + bn

(where

bn

is less
|

than a fixed value,

never

and
If

lim

^2

~~
;J/~I+P

exists
)

and

is

finite.

[DIRICHLET.]

tends steadily to infinity with

n,

and

then

l{

p-+Q \

or =(1
or =0,

- l/o)/logc,

if
if

Mn+l /M -+c>l,
n

J/N+1 /J/,,-*oc.
37.]

Ex.
13.

= n*,

[PRINGSHEIM, Math. Annalen, Bd.


2", n!
II.

Utilise

Tln-i.ivm

of Art.

152 (Appendix) to shew that


is

if

(</)

decreases steadily, the condition lim(?m,,) =

necessary (Art. 9) for the

convergence of

2w

/( ,

by writing
,,

= (#/")

/,.

that
"

= S, =
,

r/
l

-,,,i

"- 1
I

,.

" ";~

=
!,

6', (

_1

[CESA.:,..]

"n

If

of

^l--log?ij 2X, ('nipniv Art.

w,,

prove that lim (//) =


!.").

and deduce the divergence

14.

,o also is

2(a N 6 N ),iustrat-d

!">(/,

dm-ip- and
:

\.-t

+ + -'
:: !

l.ut

iv

be seen

fnn
hand,

aft
if
<

implii
15.

[i'l::

Use

th.1

rhat

if

16.

If

that

')

=
N,
1

).

17.

tli",l

,,f

1
18.
S!H-\V

-l)]-'

= 21og2-l, 1
tln
n..tati..n

8-1).

that,

\\itli

:;

?
Deduce that

:,:'^l)=N
.

-+;>
<j/,,//,.

'

__=-:;

r,-;

i>(

1905.]

19.

l'i"\r -iniilarlv

that
"

1
and
20.
tliat

.'2v+l

1
tljat
j
~-

Slu-xv

21.
if

Kxamin-

th.-

OODV]
--

'i.-tilar,

<M n )=l
jr<
22.

+ + + 2 3

prove

tliat

tin- > -rifs

converges

if

Shrw
1

that
'

v
If

,i

'

'

4
;tlltl

-l~2'
2/7-7-7

'

n!

\ -

CHAPTEK

III.

SERIES IN GENERAL.
test of convergence is simply a transformation of the condition for convergence of the sequence sn (Art. 3); namely, that we must be able to find m, so that
17.
\s n

The only general

sm

|<e, provided only that

n>m.
n
,

If
:

we

express this

condition in terms of the series


"

we

get

It

must

be possible to
e,

find m, corresponding

to the

arbitrary

positive fraction

so that
.

a m+1 + a m+z +

+ a m+p |<

e,

no matter how
It is

large

p may

be.

an obvious consequence that in every convergent series*


lim an = 0,
n
-oo

lim (an+1 + a n+2 +


n >co

+ a n+p = 0.
)

sufficient unless p is allowed to forms of variation with n; and so they are not practically useful. However, it is sometimes possible to

But these conditions are not


all possible

take

non-convergence by using a special form for shewing that then the limit is not zero (as in Art. 7
infer

p and
(3)).

We

are

therefore

obliged

to

employ

special

tests,

which
are

suffice to

shew that a large number

of interesting series

convergent.
*

It is clear

from the examples


:

in

'liaptcr II.

not exclude divergent series perhaps might bo expect. .1.

Ua

it

does not

<

that the condition lima,, = does -\vn exclude oscillatory series, as

where lim

/,

= 0, iima n = l, and

\<t

In-

tn-ms

ti-inl

to /.cm.

17, 18

LB8OL1

II

OON1

El

17

18. A convergent series of positive terms remains convergent when each term an is multiplied by a factor vu whose numerical value does not exceed a constant can F..I- since 1 chow nvergent, th- indr.x
/.. ///
!

that

>/</,

Imw.-v.T small

may

be,

M+ "+/>
But

1 ^.
.

UJ
Thua
and therefore
TV.
i

s*
fche
ia

<.
ooni

.d

cases <f this

th.-.,i-rin
,>>.

deserve
tl

series
|a n
|

2a n
|

is cm\>->.r<i>

if

series

oj

values

is convergent.
1

For hciv r/ H = |a n and Such serifs are called


.1
'/<
/

' ;(

'<
/ ,

"
(l

=
.

1.

///,><,////,-/_//

///.

series
/A*///
/'
/

////'

<<,,','<

i><>n<l'

,j

i'

positive
Tliinfinity

/.

'...uld
I

observe
K\.

that

we cannot apply
i

this
kn<.\\

im-tlnxl

iii< li

i-

An

cxanij.!.'

is

atl'

-rdcd

l.y

L'

lu-!..\v.

Ex.
it'

1.
1.

It'

\\<-

tak-

we

Kii"-

b,

11

tlia'

.verges

f>
1

Th'

tlii-Mi-mi

t-nai'

thf

'

'

_1+1_1 4 4P 5P

+
that tin-1.
tir>t
.-f

It

\\ill a:
:

Ait-. 21,

thr-i- i61

'i'g*

luit

EX.
tli-

2.
".

!+*-*+'
tinfa.-i

:t,'8
.

tO so

sum
'

x
-

that

- -I.
-Jl.'2
1 + '
'

+ --!- '""

48
The sum
of the first

SERIES IN GENERAL.
2?z

[CH.

III.

terms

is

is that the original series contains an infinity of negative terms and that the series ceases to converge when these terms are made positive (Art. 11).
;

Thus, by Art. 11, Iims 2n =oo. But *2n-]>2n, and so also Kms2 -i = oo. Thus the new series is divergent. The reason for the failure of the theorem

It is

easy to see that the foregoing theorem can also be


:

stated in the form

remains convergent if each a whose numerical value does multiplied by factor not exceed a constant k.
absolutely convergent series

An

term

is

19. It is often convenient, however, to infer the convergence of a series from one which is not absolutely convergent. For this the following theorem may be used

A
un
is
,

purpose convergent series

2an (which need not converge


its

absolutely)

remains convergent if

terms are each multiplied by a factor the that provided sequence (un ) is monotonic, and that u n less than a constant k. (Abel's test.)
\

Under
vn = u

these conditions
is

u n when (un )
is

un converges to a limit u and write an increasing sequence, but v n = u n u


;

when (un )

decreasing.

Then

it is

clear that the sequence (v n )

never increases

and

converges to zero as a limit.

Now

so that

it

will suffice to prove the

convergence of 2ft n vn in order


.

to infer the convergence of Art. 23 below)

2a n un

But by Abel's lemma

m+1

P V ni+\

where p

is

greater than any of the sums

m-ni
|

'"

~~~H

-'

-j

18, 19, 20]

01

Ai'.l

.1.

\M>
,

I'llMdll

m
fchtU

can be chosen BO that

p=se

'\.

no matt.-r ho\\ small f

18]

^''
.

'

ll

!|

f.

and

c..ns,.ijufiitly

1
thai
ill.-

at

mgent
.

conditions as

the factors are sul

lit

coiidil

Ex.

1.

If

uv takr

th.-

MriM

\
'

+\- +
}

...

Oiln-ady used
n-tors

and

rnij.li.y

oiiic

sequ

0,

w-

ol.tain

tin-

\\liich

niii>t

th.-r.-foi

.-

---ni.

'!"

\-iify

that

this

tin-

OM6,

hat

_/l_l\_/l_2\
""

,i-l

1
i'-

'"

1_

3-

Thus

li

in ,<,._!
F1

t-xi-ts (liy

Ait.

1)
is,

(1)),

and
B61

sine.-

.<...

=%,_i -

(/<+i

liin*. _,

= liin.N-

,1

That

the

Ex.
In

2.

It is

easy to sec why, fn.ni our present point of


18,

series
it

Ex.

2,

Art.

converge; the sequence of factors employed

in
a.

any

\s

important intViviic-r is that it' tluu depend the conditinn of t'ori. ay on a varial >lr <suljrct
i

./

iin-nntMiiic

sMju-nc<-).
-

tin-

r.-iiiaind.-r

after in terms in

^'-,|u-ntly

tin/'//-/-

nunirrirally choice of ni, \vhicli


;

less

than i>('\+
niak.-s
r^ -f
1

this
/^

and ): ivmaindrr
that

than

e.

is

o/OJj

so lon^ as

is tinit,-.

This ].r.']Tty
!

may

In-

.-xpv-s.-d *-,,, witf

>y

^ayin-'

41.

below.)

-prcial ca.M- of thi>,

whi--h

wul.

th.I'i

origin*]

ran

1-y

takin.i:

".,--

^1.

20.

//
lvm,\

1
D

minimwm
l.S.

50

SERIES IN GENERAL.

[CH.

III.

to zero

terms are multiplied by a decreasing sequence (v n ) which tends as a limit. (Dirichlet's test.) *
Abel's

lemma

gives the inequality

m+p
I

^^ /

(K

1)

^^

nt)

771+ 1

where p

is

any number not

less

than the greatest of the sums

a m +i>
than each of the differences
>
|

It is sufficient to suppose p not less

S m+p

- Sm

.
\

Now,
find

if

the extreme limits of sn are both

finite,

we can
for
21.

some constant! I, such that sn is not greater than I, any value of n. Thus sn sm = 21, and we may take p = We can now choose so that vm+l e/Z, and then
\

<

m+p

|2wJ<2e, m+1
proving that the series 2a ?l f n converges.
Ex.
Of
27T.

The

series 2-yn cos nO, *2v n sin

nB converge

if

is

not

or a multiple

For

l>;

m+l

cos?i0 = sin(te9) cos


.

[m + ^(p + 1)]<9.

cosec

m+p

and
so that

m+l

2 an0=sm(J0).8m[m+i(p+l)]0.eoeecJ0;
\

could here take p = cosec J0|. the first series may be convergent or divergent according to but the second series, being 0+0 + + ..., converges to the the form of vn

we

When = 0,
0.

sum

special case of the last article is the result: terms of a series '2( I) n 1 v n are alternately positive If and negative, and never increase in numerical value, the series
21.

the

will converge, provided that the terms tend to zero as


*

limit.

Abel knew of this test the history is sketched by Pringsheim .(Math. Annul ,-n, I'.d. 25, p. 423, footnote). But to distinguish it clearly from the test of Art. 19, it seems better to use Dirichlet's name, following Jordan (Cours d' Analyse, t. 1, 299).
It is practically certain that
:

briefly

fThis constant I will be either_the greatest value of greatest value) the greater of Ilim,*,,! and |lim.<*,,|.

|,

or

(if

there

is

no

20, 21

AI.I I;I;N
1

\n: 9ER]
1

.".I

|-'<>i

if

we take
la>t
til'

-f

...

as the oscillat*

of

the

article,

\v need only

.suppose

that p

is

1;

and

<<|Uently

v 1 ->'.>+>',-v4 + v & -Vt+...


it.

<)n

account of the

frequent
it.

use of
It

this

theon-m.
that

\v.

now

give another proof of

is

plain

%i-(i-^)-K^-4)+...f(^|. 1 -i
and each of
th-s.- l.i-ack.-ts
i>

]><it ive (or at least

n<-'

ive),

so that, as n increases,

th.- s.Mjuciice
(

of terin-

<>

>

never decreases.

Also

=
l

,1

-(,> 2 -i; 3 )>

-...-(t)

^K

and so the sequence


Further

..,,.;>

Qever increases.

=-)

and
Hence, by Art. L the sequence (> v lias a limit not greater than >\ and (fi^+i) Has a limit not less tlian 0. But these two
)
.

limits

must be equal

since

lini

c.,

n ..^

=0, so that

Hence the
Ex.
1.

series
M-rii-s

converges to a sum lying between


already mentioned
in

and v r

The

Art.

1*,

K.\.

1,

is

n->\\

>een to converge, i>n>\idfd that

0<jol-l.

In the special case

p=l we

get

tin-

\\hirh

is

easily seen t> le cvjual

t<

l>^-2.

F.i

'i

jo

+.r

.'

1 H-

f1

.'-'"

,.
1! " 1
J,','i

,,,,1,,,

if-*L(i-J+J-J+.

52

SERIES IN GENERAL.

[CH.

III.

The diagram indicates the first eight terms in the sequence (sn ) obtained from this series by addition the dotted lines indicate the monotonic convergence of the two sequences
;

5
9.

FIG.

It is obvious that if the sequence (v n ) never increases, but n ~ lv n approaches a limit I, not equal to zero, the series 2( l)

will oscillate between two values whose difference is equal to I; in fact we have by the previous argument lim s2n+1 = lim s 2n 1.

given by the following test special which is similar to that of Art. 12


case
of
is
:

interest

If v n /v n+1

can

be expressed

in the form

J^l +
the series 2(
\)
0,

2+

FX>1
^5,
{j

n ~ lv

n is

convergent if p

>
we

0,

For

if /x

>

after a certain stage

oscillatory if shall have

/m

= 0.

x~ n *^~ n*

>n

'

and further (by Art. 39, Ex. 3) limv n = 0. so that v w n+1 But, on the other hand, if /x = 0, it is clear (from Art. 39) that

>v

limt> n
/JL

is

not zero, and so the series must

oscillate.

And,

if

<

0,

after a certain stage

we

shall

have vn

<v

n+l) so that

limi; w cannot be zero, leading to oscillation again.


Ex.
2.

Take the
a.
ft

series

l.y

1.2.y( 7 -fl)

1.2.3.7(7 + 1X7 + 2)

So the series converges if The same condition .ipplies

to the series (Art. 20)

and

to the corresponding series of sines.

21,

22
1

\Ui:i;\

I.

9EB]
tin',-,!/. 1
{>
1/

hmiM
i.

l>e

observed

that

if

>M<1

nag

THIN

in

ihr

nor

sequences

tli.

fcmej ami

ill

reason In supjMi^- that tin- tln-nivm i- -till m-r.-ssarily fact r CODStrUCJ '\ani]>l<-s nf tin- failmv,
]
L>

snch as

+ i^| + j^i 4
:i:.

j^i
-f

-ily
Ii-ss

rooogDied

than

1111 l+
S
-2

M divergent
(Art

tlie first

p-

t-6.-

-- +
the

5i

>

aii-1

ii

therefore leu than

lint

sum

of the first n negative

tlll!

1/11 -iK+5 +
ntly th.
first
its

1\
'

2n terms of the given

series tends

to

-OD as
22.

limit.

A ruriou^ thc(,rcm,
i^

bo

some

L-xtt-nt

kin<l

of con

of th- laM.

'In-'

tn

'rsaro:

If a
<K,

farms

<> arranged
'/,.

ng order of
//<//////
n

tude,th(
n a
//

of P*
>r/,

ca/nnoi

owy

other Ivmti flam


tl>'

it

i/

"

jli-xt

terms

<>/

fh<>

series.

mi

>f

th-

fi n

jM.sitivf

tiTin-

=
I

/'

'-h.-.-hy
..f

:-''

fence th-

sum

th-

tir-t

n
1

t<Tiii

(Pi-9
Suppose now,
limit
/;
if
/,

^+(/',
it'

-i-v
i

i-vj+dp,-^
mis to a poiml<

p-^sililr. that

tlu-n.

<

/,

\v.

can tinl an
n

;ch

that

>/,,

if

//

>

//'.

Hence

^i/
1
1

[n^
But,
sin.

>+'"'
given series
;

>
not

,C. + 'Wi +

---

+
^ont.
tlu-

i->

alsolut<-l\

ami

ntly

>'

+ >\

...

54

SEKIES IN GENERAL.

[CH.

III.

can be made greater that NJ^ by taking n greater .than (say) n n can be Hence, no matter how large is, a value t found so that
.

//;

if

?i>7i

hence

sn

must tend
if

to oo with n, contrary to hypothesis.

It follows similarly that (p n

qn)/n cannot approach a negative

limit; so that

lim (p n
if

q n )/n exists its

value must be
value must be

0. 1.

Now

n=pn+qn, and so
This proof
Ex.
is

lim(p n /q n ) exists, its

substantially the

same as one given by Bagnera.*


sum
of

The

series 1

+ ^- + i + - + ^ + - + ...
we
note that the

As
than

a verification,

cannot converge. 3n terms is certainly greater

so that the series

is

divergent.

23.

Abel's

Lemma.

If

the sequence (v n ) of positive terms never increases, the


n ties between

sum

2 anV

Hv and
l

hv : where H, h are
,

the

upper and

lower limits of the sums

For, with the usual notation,

al = s 1

a 2 = sz

sl

Thus

Now

the factors (fj

v 2 ), (v2

v s ),

...

(vp . l

vp ), vp are never

negative, and consequently


8 i( v i

- v z> < H v - v 8*( V - vs) <H(v - vs -V9 )<H(vp . ^V9 8f Vp <H> 8p . (v9
(
i

s).

),

)t

Bagnera, Rend. (4), t.

Hull.
4, p.

Sci.
\\\'.\.

Mnth.

(2),

t.

12,

p.

227:

Ces&ro,

Rom. Ace.

Lincei,

22, 23, 24

kBEL'fi

Ui.MM

Hence
or
In
likr

<

Xa
manner
l't
i

<//'-,.
[fl

lli-

sum
I

ilian

",,'',,

<

//

where H'
lil
it

IB

the greater of
!

17 and |&|; that


1

is,

//

i-

the upper

"
is

-.., |*p|.
limitfl
lo\v.-r
.

It

Bometimefl <l'^irallc to obtain


that
//
//. A

for

se

mote
aiv
tl

th<-

u]|nT and

limits of 8mt

while

Th. -n exactly

<

Mi

tin-

ntli.-r

hand,

whm

.!/

i^

an

iiu-n-a^i

w.-

find

24.

Transformation of slowly convergent


</

series.

Let us writ

r/

l(

=
,

and

aw +2olH l -hh^=
.

tc.

Thru
(
i

\\v lia\r

and
whI.'

rous,',,urntly

^J",/"

="
lt

'/-f
1

/
.

//

= .'(
we

+.'>

|"-atin^ this op.-rat

>n.

tind

<<

=o y-KjRo )y+...4
l l

-f.

56
It

SERIES IN GENERAL.

[CH.

III.

can be proved* that in all cases converges, the remainder term

when

the original series

tends to zero as
positive.

increases to infinity, at

least

when x

is

Consequently,

The

cases of chief interest arise

when x = l, and

the terms a

are alternately positive and negative.

Write then
,

a l =+v 1
and we have

a 2 =-v 2

aB =+v 3

a= -v,

...,

where we write

Dvl = vl

v2

L^v l = Dv l

Dv 2 = v l

2v 2 + 1> 3

etc.

We
if

v n =f(n), where f(x)


for
all

can write down a simple expression for the remainder, is a function such that/* (as) has a fixed
positive

sign

values of

x,

and steadily decreases in

numerical value as x increases.

For we have
so

Dv n =
l

Jo
l

f'(x l

Jo

dx l

Jo

( Xl

and generally
ri ri

J)Pv n

I)P

Jo

dx l
Jo
l

dx

Thus the

series

D pv

D pv + D pv
2

...

consists of a succession

Its sum is therefore of decreasing terms, of alternate signs. p in numerical less than value; and consequently v^

where
*For the
p. 188.

cose

x=l,

see L. D.

Ames, Annals of Mathematics,

series

'2,

\.>1.

24
1

ION
This ivsult
,

lea

of th- type

Here
d
tin-

II

*>
n

ti,,

anl
tin- ac--

Teases as
\\ni-k, in

inn
r;i]riil;iti.ilis.

arithln.-tiral
tii^t

tO

apply

th.-

!';i

tion twice, starting


tin'

aay

at

and

s,-<-..inlly

at j

r ~^\\ i )

iv.Milts

arc suhstaiitially the sain.- \v-

may
U
\ve
I

be ^ati-li.-d that

the \\orl
Ex.
1.

T;ikf

and
in

\\v shall
tal'lf

r=\ if \v- work a=l -70711 apply tin- UMH


;
:

li\

get

-',

tin-

l).-l.>\v

58

SERIES IN GENERAL.

[CH.
is

III.

Now

'40825 --19568 = -21 257, so that

s'

certainly

contained between

0-21256 and 0-21258.

But s = 0-81746-5', so that $ = 0'6049 to four decimal places. If we used the original series, it would need over a hundred million terms to get this
result.

Ex. 2. Similarly we may sum the series lTo 6 decimals, the first 8 terms give 0'634524 and from the next terms we get the table
:

V.

24
1

l.i

l.i.l:

59
will

A
a
n.
I

reference
;i

t<

hew

that

we have
.-ry

series of

terms v

-?.,

for

\\hi.-i

small

IP-v*

has always the same sign.

We

ha\.

tin -n

/>,',

+ ...).
iti.l

Now

if

Dh n

is

positive

we ha\\

/>/-,

/>>,

-/

lim

/>-

-o.

because the series

in tin-

l-ia.-l.--t

^et

Dvj.
Similarly
tlu*
if
//-'
.

have

ir,

>*> ^(r, -f Dv^.


ii_:li

series ca

degree of

'I'll.-

trans!'.. rniatidi <l.-rriU-<l

above WBS
i>

.en

by K
Kuininer
the

and ami

tin- tirst

proof of
lia\.-

il

acniracy
oth.-r

<lu.-

to

Ponoelet
!'.,r

Mai'kotl'

i'<un.l

transl'nnnatioiis

th- latt.-r>

nn-ili<.l
\\

includes Kult-r's aa a

an

t-\ain]il-

<i'

Mark.tf>.

may

-2)1
Lnvr the

sum mnvrtly

to 20

<1-

TM

ajj)ly Killer's

method

t> this
l

example the ivader may

n<

-'
\\\->\
;

V& -U
r
1

4 3 *'")
:

trn

term>

..f
i

the series in th-

|.].ly

Kulei's

m.-th.-d

the

next

>i.\.

\\

.e -9011165, and if for the value '0004262 get

<-f

the remainder: thus

^057 to six
1
-

'I-ofi.
The >um
mrtli.nl ,,f

4 +
lira.-k.-;

j'

the

tiist
'1

teli

t.-rms in the
reiiiaiii-.

f-r the
1
'

Thus

1'

1=2 (0-82 2469)=

l^v.
p. 178.

j..

i:u

/'

i./Jg,

1896),

F.-i-oth.:

.-MiiMilt

l'rin::vli.-ii!.

60

SERIES IN GENERAL.

[CH.

EXAMPLES.
1.

Prove that the

series

i_J_ J__L.
x

x+\

converges for any value of but that both the series

x which makes none

of the

denominators zero

1, x x+l
,

J__L_
1_
1

J_
+
1
1

__ ~~
^na n
is

are divergent.
2. 3.

Prove that
If

if

convergent, so also

is

n.
(J/,,)

the

series
co

increases to

with

is convergent and then (see Art. 23) lim (a^J/i 4- 2 ^2 + + an

2a w
n,

the

sequence
0.

steadily

Mn }IMn =
. . .

[KRONECKER.]

4.

Prove that the

series

a - a? + a? - a? + a* - a* +
oscillates,

but can be made to converge to either of

its

two extreme

limits

by
is

inserting brackets.

On

the other

hand the

series

convergent.
5. is

Shew that if a series converges, it number of brackets are inserted, grouping


the converse
6. is

still

convergent when any

the terms.

And shew

also that

true, if all the terms in the brackets are positive.

Calculate, correctly to 20 decimals, the

sum

of the series

for

x=
sum

t\y>

the
7.

for

x=

*&

How many
1

terms would have to be taken, to calculate

fy to 3 decimals?

Shew

that the series


~
;

-a 2 + a 3 -a 4 + ...

diverges

if

an =

C_iy'-i

-j-+

n 1 ] a=l/[-x/w + (-l) although the terms are alternately positive tend to zero as a limit. and and negative

or

if

8.

If

|#|>1, prove that


'

_2

.r+1
converges to the
9.

*2 +l

__L_ +
A-*+l
if

sum

-.

If

a,,

-> a and b n ->

6,

verify that

the series

converges to the sum


A,

>S,

then the series


3)

(01

- a 2 ) + 6 2 ('/ 2 -

+ 6 3 (r 3 /

</

4)

-f ...

converges to the

sum S-ab.

111.
1

LMFLE8
1

10.

-!

if

the

>...
.

ei j^ent

.so

al

+^(
\\

rt

+ ^3) + ^ (

(/

^
t

converse always
i-se

that

is
ti

certainly tin.-

h-n

11.

Ni.sruss

cS

4-...

_I_ + _L_- +
./-Cj
C,

+_L_rl +
ijucin-r

tending to

oo

12.

Verifv that
/
1

?)
ru'i-iit
if
<

to

x and x
,

is

not equal

I the
i

vak
'

If

is

fixed,

verify that

al.-olutely

convergent
that

if

r ia tin-

int^ia! part "f

/.

13.

She*

T
in

_!_ = _ A
'

inti'^er

and

th-

1'

in.

-in-

th..'

>>e

frmii

the -unnnatioM.
>UIM ran
'>

In
|

fad

tin-

writti-n

M-\ +
*fl_,)
-

14.

With the

>anie notation as

\-

~
i //<

if

)i<

Find an

62
15.

SERIES IN GENERAL.
Discuss the convergence of the series whose general term

[CH.
is

III.

and

also that of the series

with cosw$ in place of sinnB.

[Art. 20.]

[Math. Trip., 1899.}


16.

cos(w + !)

Apply Art. 20 to the series whose sums 2 #, and deduce that

to

n terms are sin(w


sin

^^sin nd
are convergent
17.
if

cos

n2 0,

2v n

sin

nO

n2
[HARDY.]

vn steadily -O.
if

vn tends steadily to zero, in such a not convergent, then the series

Shew

that

way

that 2v n

is

r=D

converges
18.

if

(and only
(

if)

a l + a 2 + ...+ak = 0.
convergent, prove that liniw(aM+1
)

If the sequence

n) is

must

either oscillate or converge to zero.


19.

If

2an

converges, and a n steadily decreases to


If,

0,

2?i(an -an+ i) is

convergent.

in addition, aw

2an+ i + a n+2 >0, prove that

n 2 (a n - n+ i>->0.
20.

[HARDY.]

Apply

Euler's transformation to

shew that

21.

Utilise the result of Ex. 3, p. 58, to


1
~*\

shew that the sum

of the series

x
T
I

.v
~T~ T
'.

2
*j

x3
T
".

-j

~T~

tends to the limit \ as x-*\.


It
is

easy to see that


lies

(if

<#<!), Dvn
i(^i

is

positive
1
4

the

sum

between ^v

=J and

+ ^i)= ^j

and deceases

thus.

"1

x-

22.
nl

By taking ?> ;l = log( + H), shew, as on p. 56, that steadily decreases; deduce that

D p vn

is

negative

CHAITKK

IV.
I.

ABSOLUTE CONVERGENT
25.
Il

is

i'ainiliar
tin-

fact

that
.|'

a
tli,-

Unit--

BUID

ha- the

value.

IK.

matter how

termfl

MUM
'

projn-rty, series as
;

however,

is

ly no means uni\

.-r-ally

an

illustration. consider th-

which we know
|i'itivf
this

is

c-on\ ci'-vnt

(Ait.
L-t

K\.

L),

and has a
tin-

value N
so
:

^rt-attT

than

.1.

u^ arrange
is

t-niis of

srrics
ii\

that

each

positive

t-nii

i'ol]o\\

v trnns
t

th- 96068 th-n beOOl

. .'-V 4^:;
1

1
-

-J

i;

>_

K>

i-j

Now we

liav-

i_Ui_i 4 ^^ 2 -ri;

'

J_

-2

=
Thus
and
the
it

.,

is

easily
oi'

>een that
/

lim/.

..,

lim/.

..,

lim

sum

the series

i-

64

ABSOLUTE CONVERGENCE.

[CH. IV.

Consequently, this derangement of the terms in the series


alters the

sum

of the series.

In view of the foregoing example we naturally ask under what conditions may we derange the terms of a series without altering its value? It is to be observed that in the derangement we make a one-to-one correspondence between the terms
of two series; so that every term in the first series occupies a perfectly definite place in the second series, and conversely.

Thus, corresponding to any number (n) of terms in the first series, we can find a number (n) in the second series, such that the n terms contain all the n terms (and some others); and
conversely.

For instance,
terms of
s

in

the derangement considered above, the


first (871

first

(2?
first

are all contained in the


t

+ 1)

terms of
s.

t\

and the

3p terms of
26.

are

all

contained in the

first

4p terms of

series of positive terms, if convergent, has a sum independent of the order of its terms; but if divergent it

remains divergent, however


;

its

terms are deranged.

As above, denote the original series by s and the deranged series by t and suppose first that s converges to the sum S. Then we can choose n, so that the sum sn exceeds S e, however
small

Now, t contains all the terms of s (and if any term happens to be repeated in s, t contains it equally often); we can therefore find an index p such that tp contains all the terms sn Thus we have found p so that tp exceeds 8 e,
e

may

be.

the terms in ip s n are positive or zero. Now t contains no terms which are not present in s, so that however great r may be, t r cannot exceed 8} and, combining these two

because

all

conclusions,

we

get

S^t r >S-e,
Consequently the series
Secondly,
if s is
t

if

r^p.
sum
;

converges to the
t t

S.

divergent,
if

cannot converge
s

argument shews that


Consequently
If
t

converges,

for the foregoing must also converge.

is

divergent.
t<>

we attempt

to apply this ai ^unicnt

the

two

series considered in

Art. 25,

..1-1+I-.J+...,
terms
in
,,

1-1-1-1 + 1-1-1+,..,
////<////<.

we

find that the


tf

arc partly
of fact

Thus we cannot
frmn
Art.
:.'">

pi

that

>S

and as a matter

we

sec

that

this

25,

26
1

M:I; \\'.I.M
>iu.il;ul\,
til--

65
argument ueed above ti h.-i-r>\.
I,,,.it"

fails

to

that
It

-1

(r

although thi> happ.-nto

(,,

is

now easy
its

prove that
is

if

a series

"

is

absolutely

convergent,

sum X

not altered by derangement.

write

0Zaw
:-j

n |].

'2|aJ, ami th.-n introduce the new contains no negative terms, The aeriea
-;

ami
term

no
in

I.

TIM

in

i<

than
is

twic,.-;

a; BO, since not -renter than 2.1,


SUJ.JM.S.'

convergent,
th./;

the corresponding _jes to a sum

\vl
.1
.

sum

of a;

and so
th.-

B B = S+A.
;

now

that N.
(

are

th.- MIIII^ ot'

three series

al'ti-r

any
H'

.l.-rani;-i-im-nt

siij.|ns,.,l

tlie

saim-

for each ser

Tli.-n
/'

= S'+A';

hut

hy what has been proved A'


tern

and

ntain only positive


II,:

N'=/;'-.|
tin-

ll-A

}T..\ in-'

Ex.

1.

ii-iil-r

thf

>.

1-.
.trly

convergent
tla-

l.y

An.

11;
after

;m.l

therefore

ih-

series

rgent, ;nnl lias


t!

same sum

any deran^fin.-nt.

is

t<>

-f

'

__L_JL. ^
1Q2

12

Ex.
betw<

2.

From our present

point

ot'

Hew, WQ observe
i

that

tlu- iiu-jiiality

,.1.1

.^-i+j-j_^ + ...

-.ulain.-d
11).

l.y

th.-

tact

that
o'

The

wriefl a,

these serifs are not absolutely OO1 li\.-i^.-iit, and ..f couise we havt-

to infer thaTli.-

last

n-sult

should

U-

c. .nt

rastt-d
:

with

thf

-tate

of

affairs

explained
article,

at tin- h.---innino- of Art. l^s

u^ino- th- nota'

.it

we

lind

lini(r,-//,
so that

)-N,

lim

<

./
,

-f

,'/

\A
.<

lim

(8+A
(.,'
;

).

limy -

-8+A
-.verges

whereas there the se|Uencex


^ej.arately. althouo-h
thrii-

).

^-11

dill'-!

>

>

to a

sum. whose value depends en

th-

66
27.

ABSOLUTE CONVERGENCE.

[CH. IV.

Applications of absolute convergence. Consider first the multiplication of two absolutely convergent series A =2a n> B = ^b n Write the terms of the product so as to
.

form a table of double entry

/ t / t / t / a6 ajb^aj)} a b / / / t aj)^aj)^aj} a t / /
2 3
2

a^

afiz

a-J) 3

a^
4

...

...

3 4

Z>

...

a4 b l -+a4 bz-+a 4 b 3 -+a 4 b

...
4:

It is easy to

prove that

AB

is

the

sum

of the series

arrows in the
is

where the order of the terms is the same as is indicated by the table. For the sum to n terms of this series (1)

A nBn

if

Now A = 'Z
/

an and B' = ^
\

bn

are convergent

by hypothesis.

Thus the
(2)

series

obtained

by removing the brackets from

(1),

convergent, because the

sum

of the

absolute

is absolutely values of any

number

has the same


convergent,

of terms in (2) cannot exceed A'B'. Accordingly, (2) sum as the series (1). Since (2) is absolutely

AB

we can arrange

without changing the sum.


(3)

in any order (by Art. 26) it Thus we may replace (2) by


1

a 1 6 1 +a2 6 1 +a 1 6 2 4-a36 1 + a 2 & 2 +"

/>,

+ "/> + ...,
1

following the order of the diagonals indicated in the diagram. Hence we find, on inserting brackets in (3),

where

cl

= a b 1)
1

= c + c 2 +c8 +... to C2 = a2 6 + a 6 2 c3 = a 8 6
-A.S
1 1 1
,

oc,
1

H-a2 6 2 +a 1 6 3

and
For other results on the multiplication of series thr should refer to Arts. 34, 35.
iv;ul<-r

27
MM.!
us-|'ul

AI'I'LH

i<

application of
of

tin.-

theorem of Alt. 26
in

19

to justify thf

step

arrang
in
./
:

lowers
\vliere

wh.-iv
It

;/

is

polynomial
sutli-i.-nt

say

//

=
v

'>

-f

>

r/:

...

-f l*^.

is

hriv

to

have !<

,<=
and from

0.1,

u-A+ftf+.-.+Af, A.-IU f-||:


1<>.

Art.

w.-

966 that
A,
if

this
1

iv.|iiires

,;<
'I'ln-

\-

= liin
shall

1668

condition iv.juiivs that j8d <X, and that than BOOI6 ti\-d value; and thm the n.-ct-ssary
last

will

In

certainly not alter th- sum most of the ordinary 68864

th-

BOTJ
//

\=l. and

is

of th-

1'orm

the condition

is

th.-n

In particular,
if

.^__2.

it

is

to tak-

<

/ x 2

1.

which

ainly
-

satisfied
,,

\vln-n

uri,

,,,,.,.

be
is

sutlii-ient

lut

iav ). t.-mpt.',! t,. think th;it the coiulitiuii \y\<\ w..uUl tliis is m.t the case. K"i- we have to ensure the
i>

\viitt-n cut at length,

and

V
\vhi--h

inaih

pusitivr

in
>f

/<>f/

form.

AB an
has ih.
ird

illustration
[1

tliis
1

j.oint, r..n>i.l'-r
-'.

>um
l.y

-(ir-.r-')]-

= (!-.)

\V!MMI
1)

jurii'-iilar liy

any vahux }, bed


get

>f

\toe-a?\<l. This conditi iad lyin^r Between 1 -r s ;

i'.nt

if

th-

in

i"
)

-X*

68

ABSOLUTE CONVERGENCE.
;

[CH. IV.

This equation is true if both series converge although the proof does not follow from our present line of argument. It may be guessed that, in general, the condition found for f in the text is unnecessarily narrow
;

and
ever,

this is certainly the case in a

we we wish
is

of special applications. are not here concerned with finding the widest limits for x

number

Howwhat when x
;

to

shew

is

that the transformation

is

certainly legitimate

properly restricted.

In view of Riemann's theorem (Art. 28) it may seem surprising that the condition of absolute convergence gives an unnecessarily small value for f. However, a little consideration will shew that Riemann's theorem does not imply that any derangement of a non series
-absolutely convergent

be made to have any value which may easily be of a far more n sweeping character than the derangement implied in arranging ^a^ according to powers of x.
;

will alter its

sum

but that such a

series can

by means

of a special derangement,

Riemann's Theorem. a // series converges, but not absolutely, its sum can be made to have any arbitrary value by a suitable derangement of the it can also be made divergent or oscillatory. series; Let xp denote the sum of the first p positive terms and yn
28.

the

sum

of the first

negative terms

then we are given that


oo
,

= lim(a?p-2/ n ) s,
where p,

lim (xp + y n ) =

tend to

oo

according to some definite relation.


oo
,

Hence

lim xp =
p
>oo

n >oo

lim y n =

oc

Suppose now that the sum of the series is to be made equal o-; since xp -*vo we can choose p l so that x er, and so that Pi p l is the smallest index which satisfies this condition. Similarly we can find n^ so that 2/ %1 2^ 0", and again suppose that 71 is the least index consistent with the inequality.
to

>

>

Then, in the deranged series, we place first a group of p l positive terms, second a group of n^ negative terms, keeping the terms in each group in their original order. Thus, if >S, is
the

sum
f

of

terms,
if

it

is

plain tint

<S

O,

v<p v

but

><r,

if

p^ v<p
;)

+ "r
pi)
//,^-ho-;

We now continue
positive terms,

where

the process, placing third a group of (p 2 p 2 is the least index such that r

>

and fourth, a group of (n^


the least index such that
//

n^) negative

terms, where

n8

ifl

>

,/,,.

<r.

27, 28]

IMKMANN
im-thon!

BEORE14
evidently be carried
/*,

'!

Tin-

of
it

construction
is

can
if

indefinitely,
is

and

d.-ar that

/*,.+
/>,

>^=p
1

^ r
tin-

on *

positive,
if
<!

I'M:
.

oannoi exceed th- (/>,+


/> r

.>^

t-rm of
Imt

MC
these

\vhil.-

i>,

,-f-

>' = />,.+
/', -)ili

//,-.

T
i'<.r

N,

is

jMiMtiv.-,

does not
.-it

tin-

</',

t'-riu:

N.-T
tferiea

ehangefl sign

terms.

Thus, since
increases,
It

tin-

ti-rm^ "f

th'-

mu-t

t.-nl to

zero as v

we ha\-

limN

=&

easy to mo. lily tin- t'oiv^oin^ m.-thoil so as to get a <li\fi-L;Tiit or o.scillat'ry 901166, by starting from a sequ<-:
is

which
in

is

ritht-r

<li\

vruvnt or oscillatory and takin.


tir-i

.....

turn to be

tin-

inlic-s
!

which satisfy thr


l

int-'jualii
t

>^-^i s>y
howe\ work and
;

--<rii

> xP -

;ml so

on.
of fact,

As a matter

fur pra'-tical

\vr

tnn> prooeM i- 'juite out of the have to adopt an entirely dift'-

!i-ti..n.

>t.-adiiy dr<T.-asing to zero as


i>\

increases

ami
\-

-i)"-y(/0, \vhii-h converges,


jiusiihv
tei-ni
s.-ri,->,

Art. 21.
:

is

f<.ll.i\vrd
first
/

itiv.-

tt-nn
/-

that, in

tla-

dfraniTfd

tin-

t-nns i-,ntain

-r r

ji--iliv,-

and suppose to n

negative terma (so that

-In

+ r = r).

Then the sum

of these r tern:

win-re the second l>rark.-t contains r trims,

and

s..

lios

between

and
Suj.jiosr
first

r*r-

that

/-/'(/')
1

tcnd>

intinity

with

n,

then

/'(-")

''<>f

iM-tw.-iMi

and

/(*+?),

Tin;-

llOOSe v to

be

inction
tin-

hange

in

tlu-

sum
th,-

of th. rr-ult
:

/,

l.t-.-au-.-

th-n

\Vr ha\- thus

!ii->t

''

?'*

*M^/'

=/.

taking the series ^(-ly-'/r*,


tlu-

we

see
1 !

that v
v

may be
bi-

integer neireet
i

.r

again with 1( -i)--

^,

may

integt

!ogn.

i.

-J-J,

j..
1

to take

always the nearest


.:--nl.

order

to xitisiy tin- i-.iulitiiin.

Hut

tl

70
Next,
value of
if
c,

ABSOLUTE CONVERGENCE.
lim nf(n)
is finite,

[CH.

however

small, a value

say equal to g, it follows that, for any positive nQ can be found such that

Let v be chosen so that k = \im(n + v)/n.


It is easy to see,
i

by an argument similar
i
i

to that of Art. 11, that


rn+v

dx

Hence the

alteration

is

contained between the two values

(#c)log.
Thus, since
e is

arbitrarily small,

we must have

Hence, if lim nf(n) =g, and if k is the limit of the ratio of the number of positive to the number of negative terms, the alteration I is given by l=%glogk. In particular, since 1 -* + ... log 2 (Art. 21), we see that when

=
;

this series is

arranged so that
is

Ten

terms

its

sum

log 2-f

log

k=%
we

positive terms correspond to n negative and so, if there are two positive log 4

terms to each negative term,

get

While,

if

there are two negative terms to each positive term,

we have

a result which has been proved already (Art. 25). Finally, if there are four negative terms to each positive term,

we

find

To save space, we refer to V. of Pringsheim's paper for the discussion of the rather more difficult case when liranf(n)=Q.

EXAMPLES.
1.

Criticise the following

paradox

+ 1+*+*+*+*+...
2

+J

+ 1 + ...)

= 0.
2.

If a transformation similar to that of

Ex.

1 is

applied to the series

1-i+l.lo.I. 2* 3* 4 6
Hhew that
(if

p<\) we

series is negative, But, if the sum of the given series in

obtain the paradoxical result that the sum of the 1, the result obtained is correct and rx presses

p>

terms of the corresponding series of positive

terms.

iv.)

MI'LES. wt writ- /(.)-!/*, and =i/(n), shew that

Tl

3.

If

t/W
")
[It
4.
is

+ ...H'-!* 1 -

proved

in

An.

71

that i

I'K.VC that

5.

Apply

a transfoi mation similar

tliat

of Ex.

to the series

i-i+j-H~,
ami piovr that
tini

suit iiiur

----.
>!
ies is

which converges to
6.

-um ItH

tlian

that of

th'-

-iven series.

absolutely coiiveru'ent

convergent series may be converted into an wriM ly tin- insertion i.f brackets. [See Art. 5.] Any oscillating s.-rii-s may W converted into a convergent sei and the bracket* may be arranged so that the in>.-rti..n ..f l.ra.-ki-ts

Any

n>n-alis..lutrly

la7.

sum

i|ual

to

any of the
.f

limits of

In

Mi-iK-i-

that

tin-

valu.-

n.n-al>>"lutt-ly
in

convergent series
tei

may

it-main

unaltered

aftt-r

a <-.-rtain chain:'-

the order of the

sutli'-it-nt

that the product of the

greatest subsequent term

may tend
G
s
Il(lt

displacement <>f the //th term by the to zero as n increases to oo.


\

[Iiui:Ki..
30

V"tL

(2),

t.

14,

1890, p.

1*7.]

8.

Prove that

1
'

dfierminate numl>er,

luit

that

is

p.-rfe.-tly

definite.
//

Here

./

is

supposed not to U> an integer, and


f

implies that

=0
lim

is

to le

>nii

Shew

that

wheie p and
9.

./

tend to x

in

siidi

way
sei

that lim

(</;<)

Find the product of the two

+...+-K., and
10.

1-..-

+
,

'-. ..+(-1)-

Shew

that

if

.<, 1

= ^ + (/,+a s +...-l-a (1

then
.

CHAPTER
DOUBLE

V.

SERIES.

29. Suppose an infinite number of terms arranged so as to form a network (or lattice) which is bounded on the left and above, but extends .to infinity to the right and below, as

indicated in the diagram

The first suffix refers to the row, the second suffix to the column in which the term stands. Suppose next that a rectangle is drawn across the network
so as to include the first in

rows and the

first

n columns
.

of

the array of terms; and denote the sum of the terms contained within this rectangle by the symbol sm |M If # /M>M s t a limit as n and tend to definite approaches infinity

<

<

the

same time

(but independently), then

s is called the sun*

the double series represented by the array.* In more precise form, this statement requires that it shall be

possible

to

find
e,

an index
such that
!-V

//,

corresponding to an arbitrary
if

positive fraction

-*!<*>

>,n>p.

the last inequality is implied that m, n are subject to no other restriction than the condition of being greater than /z.

By

*ThiH definition

is

framed

in

arrordaixT with the om- adopted


p.

l.y
1

(Miinchener Sitzunytberichte, Bd. 27, 1897,

101

see particularly pp. K:i.

UK

29]

Sl'.M
:

OF

DO1
I

BLE 3EBU
I

pn.p.-rty

-y

th- e
\

|iiati<

HIM
->
tinis

s
tii
_

=
ifl

s-.

,r

lini

symhol
as to tlif
t.)
u.x,.

not MiMid-iit
1

unless sum.- indi'


1 ]

.-iM.-<|

mode
nthei-

of MI n

ia

i<

ii

.!

I'mi

it

eoii\ .-ni.-nt

method
tl>

:;1

whidi

i,

give values dilli-ivm

from

s
it

''

l+*m-\,
that
\\-IK-H

t'..lln\v.s

o
that
n.

>

converges,
in

we can
//
;i

find

/z

so

|tt w>H
iliis

<e,
of

pruvidi-d
doefl
//',
//

/"*///
(

and

ber

than
will

fi

eOUIM

n.-t

arily
I

ini])ly
/"/'///.

that

t-nd

t> /..T., wh.-n


'I'll--

t-nd t"

filiations
lini
.sr

y:

or lim x iHi?J
nuinlH-r

=
fr.

v:

imply that, ^ivm any can tind //. Midi that

jH.vitiv.-

lio\v-\

.-r

l.i

8m,n>G,
and
It

it'

'"
'

>f*:
\V<-

tin-

d<>uM-

Berieci 18

then Baid
-/:
.

d. tint-

similarly divergence to
is

aNi

jM.^il,!,.

that
in

tin-

d(uM-

Beriefl

m.

and
l'oi-

th.-r.'

is

littlf ditliculty
tli.

modifying the ni'-thod of Art.

to

establish
i

any

d..ull<-

sr.ju.

these

may
lii"
(m, w)

I).-

d'Mot'd

l.y

8m

and lim
i-

-imply that
/<

tli-

of

between two numerically less than e. ii


tli.-

t.Tin-

r.-ctan^lfs

//'.

and
/*
;

/*.

'/

must be

than

m- in

wh.-iv of
dition
d.-not,'
is
i

(-"iii-s,-

the value of

//

will

depend on
it

e.

Tl
i>

.,nd to see that


l,y

tli,-

\alue of

>;;
,

when
if

,n

i>

(so that th- rectangle


yi.-ld-

replaced ly

Then OUT condition


!o-9 -<r M

|<
limil

</>

>

At.

Heiu-e T

approaches

and so we can
if

^, such
"The

that

IS.^K^
in
full
l>y

lt>
1900,

rnnir-i

Wl,

74

DOUBLE

SERIES.

[CH. V.

Now
and

the general condition gives also


|

p,,-

<r

<K
S
>

if

so, if

// 3

is

the greater of /^ and


S
\

p,q>n>[jL 2 and n > // 3 we // 2


',

find

P,- \<
If

if

^ 2>^satisfied.

Ex.

1. 2.

Convergence:

m>w =l/m+l/,

Ex. Ex.
30.

Divergence: If s w> M
Oscillation
:

3.

If *,,

=0 and /x^2/c. = m + w, the condition of divergence is = ( l) m+n the extreme limits are -1
,

and +1.

Repeated

series.

In addition to the mode of summation just defined it is often necessary to use the method of repeated summation; then we
first

form the sum of a row of terms in the diagram, and


oo oo

obtain b m =

a miM

after

which we sum

n=l

2
1

6m

This process gives a value which

we

denote by

(JH)

(n)

this

is

called the

In like

sum by rows of the double series. manner we define the repeated sum
\ ^J n=i x m=l
j

^m

>

or

2 2

which is called the sum by columns of the double series. Each of these sums may also be defined as a repeated
limit, thus":

2 2a m M =
,

(m) (n)

lim (lim m->m n-+<x>

s m>n ) or

lim s m<ll
(m)(n)

with a similar interpretation for the second repeated sum. In dealing with a finite number of terms it is obvious that
8

M / N \TM x^

\
\

*-SSv^Hi;gg<v.;necessarily true that


if

N M VM v*
/

\
\

But
lias

it is

by no means
s in

a double series

the

sum

the sense of Art. 29 then also

a)

s=i m/

indeed the single series formed by the rows and columns of the louble series need not converge <>//, I ml may oscilla

29, 30)

REPEA1 ED
th.
L

81

MM Mlhe example

That

oolomnt need not oonrergc


'A

hi.'h
.

1-0
lim
ii

lini

m
exists at
all.

Vn
f

I'l-iii^h.-im

has
.//.

proved. 1. <i,l if th,

/Ae

rows
/;/' //'.
I

</,, >i/,/,

9eri& is conr*

quatiow
In
I'aet

<

o&ot*

we have

MI that

lim

^e,

if

//<>//;

since,
'

l>y

hypothesis, this limit


lim
u->

dim
n-*>oo

>(/1
/t

= x.
<

In

like

manner we can prove the other


i^

\\'hen the douh''

not

half of equation convergent, the equation

).

g(fr*.)-|(!
i^

mt

n.M-r.^arily

valid

win-never the two repeated 961166

convergent
1

h.

no reason \\hau-ver for assuming that the equation


lim (lim *,.)

= lim

(lim

v)
= !.
tin-

the repeated limits


/>),

we

tiiul

lim

(lim

x fllill

= n,
it

lim (lim
is

.<,

Knm
not

1'rin^shrim's theorem

clear that

.lnMe
oi

series

(mvr^e

(the rows and columns


(

ln-in--

^upposed con-

vergent) unless e(|uation no reason for avximiin^


.

'2

is

valid:

Init
[

the truth
the

tli-

douMe

with

<

.....

= IHI<
(lim
Mi

(;/j

+ /o-, we

find

lim

.,

,,)

= = lim

(lim -V,
.

lint yi-t tht- d..ul.le >eiies ca;

w,i =

FT
true,

some purp.^ex it without tnmMin^

is

nvrt'ul

to

know
th-

thai

equation c2)

is

to

con^id.-r

76

DOUBLE

SERIES.

[CH.

v.

convergence of the double series. In such cases, conditions may be used which will be found in the Proceedings of the

London Mathematical Society,* would go somewhat beyond our

the discussion of them here


limits.
(2)

A
may

further example, due to Arnclt, of the possible failure of equation be added


:

Thus
but

lira

m*-x> n

(lim

s m>

= - J,

lim (liin* mt w)=+i-

Other examples of points in the general theory will be found at the end
of the chapter.

(See Exs. 1-6 and

9.)

31.

Double series of positive terms.


if

anticipate that
s

been proved in Art. 26, we may a series of positive terms converges to the sum in any way, it will have the same sum if summed in any
In view of what has

other

way which

includes

all

the terms.

For,

however many

terms are taken, we cannot get a larger sum than s, but we can get as near to s as we please, by taking a sufficient number of
terms.
shall now apply this general principle to the most useful special cases. (1) It is sufficient to consider squares only in testing a

We

double series of positive terms for convergence. = n; then Write for brevity s mn = o-n when

m
.

plainly
if

crn
<r n

must converge

to the limit

s,

if

sw

iH

does

so.

Further,

converges to a limit s, so also will s m<n so that <TM lies between s and s e; /j.
greater than
/z,

For then we can find but if and n are

we have

so that

= ,>*

e.

Hence s m<n converges to the limit 8. The reader will find no difficulty in extending the argument
to cases of divergence.

*Bromwich, Proc. Lond. Math.

Soc., seri-

-J,

\..l.

I.

I'.MH, p.

ITti

30,31]

SERIES
If

mem

for

puri

ion,

we
'>nd

iccession of curves*

plainly,
a

when
two

tin-

mi therefore

th-<'
it

-jiiares)

sum
'

tained

V particular CUTVe of tin- square-, and that tl


if

to be con-

aqua

p,

thus

is

the

sum

for

tti<

,-ive,

in (1)
l-'urtli.-r.

^cr^a.
to
//

tt-iid

tn iiitinity in all <lin--//

can

makr
\vt-

i>

greater than

1)V
rr
t
-N-

taking
,

>
it'

Thus, since
h

> <

x
fi

e,

>
two
!'

and SO
In
liknianii.-i-.

lint

N aft
;|

ly cnclo-in-'
that
if

s.juar.-

ln-t\vM-n

the

cm".
tin-

can

>lu-\v
t

tin-

cur
rect
boo, in

M|U;UVN (and

h.-iv!', ,r,.

tin-

virtue of (1)

particular da-s

<>|'

th,-

curv.-s

u^-d
tin.-

in

-J

i-

t'orim-d

ly

drawing
<>t'

diagonals, equally indiiu-d to the nrt\v<>rk as indicated in

Imri/nntal and v.-rn'cal


riidit-hand

tin-

10.

Tin-

summation ly
t
i

>.|Uar-s

is

indicated on the

l-l't.

It

should

he noticed

Thus, hy sjuaivs.

we

are suniiiiin^ the


)

r(.2l

+ ",, + ",, +
\vhlly IT
in
i

-f-

that each curvr enclose

78

DOUBLE

SERIES.

[CH. V.

and by the diagonals we get a n + (a a + a 12 ) + (a 3l +

22

+a

13 ) -f-

. .

Of course the equality between these two series is now seen to be a consequence of Art. 26 but we could not, without further proof, infer theorem (1) from that article since Art. 26
;

refers only to single and not to double series. By combining Art. 26 with (1) above, it will be seen that
(3)

No derangement

can

alter the

terms of a double series sum, nor change divergence into convergence.


of the (positive)

important to note that: the terms of the double series are positive, its convergence implies the convergence of all the rows and columns, and its sum is equal to the sums of the two repeated series. For, when the double series has the sum s, it is clear that
It is also (4)

When

8 m> n

cannot exceed

and consequently the sum

of

any number

of terms in a single row cannot be greater than s. Also, for lims any fixed value of m, m>n exists and is not greater than s.

Now we
than
JUL.

(n)

can find

JUL

so that s mt n

>

e,

if

m, n are greater

Consequently

silimsTO|W >s-e,
(n)

if

m>

//.

Hence

Km [lim sm
(HI)

= s, ]

(n)

In a similar way,

we

see that each

column converges and that

As a converse
(5)

to (4), we have The terms being always positive, if either repeated


:

series

is convergent, so also is the other

and

also the double series;

and

the three

sums are

the same.

For, suppose that


lim [lim*,,,
]

= *,

Art. 2 the sequence (a-m) converges to a limit then it follows from (4) above that x = rr. and that other repeated series has the s.-mir sum.
;ui<l

Hence by

y.j

tin-

31, 32]

PS

FOB
littlr

"N\ BRGEN4
ditlirnlty
in

79

'I'll.-

ivnd'-r will

find

ni'.difyi:
-i

in

(4)
32.

and

(."o

so as to cover the case of di\.

Tests for convergence of a double series of positive

terms.
It'

we compare An.
>

s
,

with

<

>

.f

the last article,

we

see

<

If

t;

rms of a
///,;,
/,

<l<>i>/<

series wi less il>i<


to converge, th*
'

mother doub
Similarly for
Tinin.ist

;*

known
'

d;

with

greater
.-r^'-nt

in

plai
1

iiiipnrtaiil

\y\- d' c)ii\


1

series'

i^

amtH = (<
that
tliis

wln-iv ^^',"
.inn!.],-

is

convergenl singl
t

-i\-n by to
.

seriee

i>
.i'

convergent we note
sj.1.l

-\\u\

ntain-l in a

s.|iiar:

n
-

i^
'

plainly r.jUa

+ ^o- +
limit.
last

and therefore T
Converges, ly
(
(

ha)

r<>ii^.,|u,-iitly
articl.-.
i'<

tin-

<l>ul.-.

of tin-

>n
/

tin-

tlirr
-

liand.
1
.

th- chirf ivp-i,

ly

=(y/>
aeries;

\vli<-i-.-

,n

;/

,i

and

divergence' " is a ^/>


1
1

di

-in^lf

t>
in

diagonals,
ill-'

as

c2)

recognise of tin-

tin-

iliv.-r^.-ncf.

take
thi^

tin-

Bum

last

artirlr.

In

way

Beriea

which

i-

aeon

In-

divci--vnt.

.n

coin paring

it

with

'+..
Ex. Ex.
divei
1.

?.)n-*n-fi coir
l'(//j

a>l, /^>1.

2.

+ /)~

ft

di\

^v=rK
Ex.
3.
I

- A
,

opavergea

if

X>1,
thL:I>

div-i
-

\\h--iv

i>

uid
"t'

b\.
i

Thus thf and


-J.

roiidit in->

dix'i

iiu.

.1.'

80

DOUBLE

SERIES.

[CH. V.

The reader will have no difficulty in seeing that the following generalisation of Cauchy's test (Art. 11) is correct: (2) If the function f(x, y) is positive and steadily decreases
to zero

increase to infinity* then the double series S/(m, n) converges or diverges with the double integral
\
\

as x

and y

f(x,y)dxdy.

test (2)

However, nearly all cases of interest which come under the can be as easily tested by the following method, which depends only on a single integral (3) // the positive function f(x, y) has a lower limit g() and an upper limit G() when y = x and x varies from to and if G(), g() tend steadily to zero as ->x
:

then
//>
I

the

double

series

22/(m, n) converges

if

the

integral

&()$;
00

converges;

but the series diverges if the integral

f
1

0()^

diverges^

For then the sum of the terms on the diagonal x-{-y = n lies between (n-l)g(n) and (n l)6r(?i); thus the series con30

verges with Z(?i

1) G(n), that

is,

with the integral


l)</(?i),

f
I

G()d\
with the

but the series diverges with 2(?i


integral

that

is,

y ()(d(.

Ex. 4. A particular case of (3) which has some interest is given by double series f(am* + 2bmn+o*,*), where /(#) is a function which 2 steadily decreases as its argument increases, and am* + %bnin+cn is subject to the same conditions as in Ex. 3 above. If A is the greatest of a, |6|, c, it is evident that
the

is

less

than A[a? + 2.r(

-.r)
is

+( -x)

2 ]

=A
</,

2
.

When

is

positive,
is

we

see in

the same
/?
2
.

way But if b

that
is

if

the least of

b, c,

the expression

uivutrr than

negative,

we can put
</tf)

the expression in the form


,

[{(

+ c-26).r + (6-c)^} 2 -f(ac-^)P]/(a + c-26)^^ 2

if

4Jfl=ac-&

Hence
*That
if
I

=(Ap)
is,

and

we suppose
f
Ml'

/(f,

i?)
'

=/(.''.
7
?

//).

=
II.

-{

""''

=='!/

Tin- us.-

singlr int.'-ral
lS7'i,
|i.

|',.r

testing mull ipl-

s.-rirs

srrniS to be
i-at ion
s.-cins
is

due to

Cii-m.-mii

(</.M.

H'o-Ar,

l-VJi;
p.

an
s:ti.

'

all

I'liat

in\

t-st

given by

Hut-wit/ (Math. Ann'i/m,

I',. I.

Tin-

.-ilmv.-

tonn

to lr imvi-1.

32, 33]

/.
il.trlv
tl

seen to diverge

..

:i

al.o\e;
.'i

and

it

-li.-wa also

that

\\

'
t

>0)
tin-

tin-

>eries

converges;

..ii

the other hand,


p)

181161

divergei

if

33.

Absolutely convergent double series.


as in tin- tlifry
,
.

lust
!

<>!'

sing
,,
,-,

I,

w-

call

the series

iit<l ;l

if

^
Ar
1

,n\ .-r^-nt.

Tlit-

in. -ill.

.,1

u^.-.l

in

m
in

be

applii-.l
:il
!'.

,-it

once to shew

that

all

tin-

ivMilts

pn.v.-il
still

Art.

-r

l..uM.-

I'..sitivr
il..nl.lc

ten
aeriea
tin-

true

for

any
tlie

In

tliN

oonnezioD
of
tin.I/"'/'.

aK^>lutely convergent reader \vlm has advanced

l-.-ycnd

.l.-in.'iits

sul.j.-.-t

slmuld

t-misult
p.

pajM-r

by Hardy

(Proc, A'-"/.

N''. c2)

voL

l.

I'.'o:;.

Tliat tlu-se results are

not necessjirily trm- for n>n-al.>'.lutely i-<nivergent


t\v..
1

may
(1)

In-

sc-n
first

liy

taking

siiupk-

examples:

Consider

+ 1 + 1 + 1 + ... + 1-1-1-1-...

+1
\vli.Tc all
ll'-i,I'

+0 + 0+ ...
\.\\

tli--

tTiuif
?/J,

jit

in tin- tirst
tin-

rows and
has
tinit

c.lui.

^,, = 2
plheim'l

tt>\, sn that l-linition. Hut if


.iniuiiij,'

s.-ri-<

>um
:il
),

i\

a--.-..

riling

the d<uil>l- series into a


Alt.
\v,-

tlie

(1:

get

1+2 + 1+0 + + .. .=4.


nlivii.iisly,

too,

tip-

convergence

--f

tlii-

series does not


(.})

imjily the con-

tWO fowl ami columns (compare


(2)
'

Art. 31).
:

the (loul.K- series suggested


1

by Cesaro

1,1
j

1,1
7

2~
1

^8~16 + l6~7

1,1

+ 21-,

:i

:i

L6

16

73

153
16*

+ r6'

15s

l.s,

82
Here the sums
of the

DOUBLE
rows
i
2'

SERIES.

[CH. V.

in order are

L I
22
'

I
'

23

24

'

*"

'

and so the sum of all the rows is 1. But the sums of the columns are

+ 1,

-1, +1, -1, +1. -1, +1,

...,

proving that (5) of Art. 31 does not apply. This result is specially striking because each
(the

row converges absolutely


secondly, the series
also converges

terms being
,
,

less

than

t+i+i+J+J-K..)j and
rows
is

formed by the sums


,

of the

a
*2i

+ 53
2i

~*~

>

which

absolutely.

But the justification for applying (5) of Art. 31 is that the double series still converges when all the terms are made positive, which is not the case here ; since the sum of the first n columns then becomes equal to n.

The
series

fact that the

sum

of a non-absolutely convergent double

may have different values according to the mode of summation has led Jordan* to frame a definition which admits Such restriction seems, however, only absolute convergence.
series is used,

unnecessary, provided that, when a non-absolutely convergent we do not attempt to employ theorems (1) to (5)

of Art. 31 without special justification.

For example

in

Lord Kelvin's discussion


"

of

the force between

two

electrified spheres in contact, the repeated series

r*

m=iUt'i
is
if

mn~\ (-l) (m + nY J (m+nf

m+n

used.t

This series has the sum


first

(log2-), and

it

has the same value

with respect to m. However, Pringsheim's sum does not exist but oscillates between limits^ J(log2-f) and (log2 + |); while the

we sum

diagonal series oscillates between


34.

oo

and

+ QO

A special example of deranging a double series is gi\vn rule for multiplying single series given in Art. 27 the by above.
Suppose we take the two single
construct from
*

series J.

= 2ttn 5 = 26n
,
.

and

them the double


t. 1,

series

P = 2a m bn

Contra cTAiiaJye,
1,
i>.

p.

;>!>_';

<-..mp;nv <."iii>ufs Aiiu/i/xi* (translation

l>y

H.-.lrick), vol.

:r,7.

t-K.-lvin,
:|.

/.'

/*///// >,f
;iiul

MiTfricn/

MM.
l.,i<l.

Bromvvich
9).

Hardy, /Vof.

Mnf/i.

So,-.,

sri i-s

-J,

>1.

'_',

l!Ol,

p.

lt!l

(see

33, 34]

MILTI1M.H \TI"\ OF
ii.it

-I.I:

It

/'

eonvi-r;r,.s
i-

in

lYini^h.-im

--use, provided

ha:

.1

/>

ponvi
*m.n

ire

lu

= ("i + ",+
J
I!.

+ ",></>,+/>,+ ... +

'

BO that
lint

liin

>,

-*
f"i-

to convert

praeiiral work in analysis it ilouhle aeriea /' into a single series, tli'i

ssary the one

usually ehoflen

hein;:
i^>

tin-

sum

l>y

diagonals (see (2) A

This

siii-'l,.

senee

-'. where
tVoin

It

follows

at

Once

Art.

>>

that

//'

//

ivo

serif*
t<>

their
^_

product

is

eqwil

,!},',,},

Jut,
tht's.-

if

cnu>'<

,'!<' i'f.

und.-r
al^oluti-ly

cinMiinstanr.-s
m,

tin-

lciil)l:

!' -au.se 2|am |.|6.| converges to oonvergentj


it

<-

'.(S^,,!);
<t'

and 2|cn converges because the sum


|

of

terms from ~\c n cannot exceed


\

tin-

pi'oduct

If.

however,
/->.

one
;/>

or

both of
f<>

A,

B
=
l>i

should

not converge
f/lt ,f

absolutely,

we have

Abel's theorem:
Is
'i.

/V<',V,>,/

the aeries

^
if

8wn
l

/<ii

n
ti n

/>,<///</ A.B.*

For
\ve

tl

we

write

An =n +

-}-...^-'i nt

'.>+.

..

+ !>,

tind

cl +ct +...+cn

=alBn +atBn . l +...+amBv

Hence

Ol +Ct+.,.+On -A lB1>+AJBn . l +...+A

Now

(App., Art.

r>4>.

when \\m(' n =

r.

,.

have also

lim

((',

+ (',+
i

...

'

and again

<AJ.I...

An.
1.1
./.'
L

t54

lim

-*-.!.//
II
'

"
r

4-

...

+ 4 HI )=
A
,/y.

.1 />.

Hence

<'

= AH.
ha>

(For an
;i

alt.'i-nativ.- |'ioof,

see

Art 54 be

pr.ivi-tl.

i.y
t->

siinil.u

nn-th<Kl.
tin-

oonvt
.-onvTL.'-.
provitlrtl

jual

thi?

sum

.!

diagonal series, v
\

that

i-oluinii.

84

DOUBLE

SERIES.

[CH. V.

It should be observed that the series

2an 26W
,

2c n

is

2c n cannot diverge (if are convergent), although it may oscillate. For, if = oo, and therefore also divergent, we should have lim Cn

IV

2c n oscillates, it is clear from the article quoted that AB lies between the extreme limits of 2c n that in some cases 2c n does oo and -foe) oscillate (and that its extreme limits may be is evident from Ex. 3 below; but in all cases the oscillation is of such a character* that
;

by

Art.

154;

whereas

this

limit

must be equal

to

AB.

If

IV

Ex.

1.

Undoubtedly the
Thus,
if

of power-series.

the

cases of chief interest arise in the multiplication two series

are both absolutely convergent for \x g iven bv + +

<r

(see Art. 50), their

product

is

0^ 02^ + ...,
bn

which

is

also absolutely convergent for


c

=a

cn

=a

|#|<r; where we have written + ^i&n-i + + an bQ


. . .

the notation being slightly changed from that used in the text.

Ex.

2.

If

we apply

the rule to square the series

we have no

reason (so far) to anticipate a convergent series


,

but we

find the series

-Q +
is (

i
1

in

which the general term

l)"" ;^,

where

so that

Ces&ro (to whom this result is due) calls such serii-s xhnply ind>(> miniate the degree of Indeterminacy being measured l>\ the number of means \vhioh have to be taken before a definite value is obtained. (See Art. 122 below.
;

34,

35
1

Mi l.THM.K LTIOH OF SEE]

li.nfu
9
liin "
lini

logn)='
w-f

LI),

!>gn)
1

,,'-i)-;r7T

theorem,

10-1
this

-S('+4
agrees

with

l'i

in^lx-ini's

general

th.-c.r.-i:

[MEx.
3.

1900.]

Hut

if

w.-

-qnare the

s.

^i-I^....,
"litain
l'(
1

)"

'". where
-r[2(/l
if

'.at

[./'(

-.r)]"

and
it'

">/''
/>

b, th<> .<'r/f.<

-(~ 0"~ ljr n

?J{

de for

iniilri

(>n th- I'tln-r


tlu-

hand,

if

/'

>

A,

I'l-in^h-

ml.-

35.
.! /;.

M.-rt.-n-

lia^

|.i..\,-d

tliat

tit.

'h<it
\

one
t,

Suppose that

and thru writr


...-f-,-...
t'

f->r In-

.-|a|,
/

r^^-f-r.-f
,
,

In.

/.

4- />+...

...

/!,

it

/?=/
(/

/;

('.'nsr.juently

= </,/* + /,/;.. = (//_r,,)+


,/
I

,-f...-

,(^-r,)

.1

/;

//

#
so that

.-i

-f-.-.-fc
...,-i-H. ..4-

|,|i

86
Now,
can find
since the aeries

DOUBLE
a limit
so that
it
,

SERIES.
>

[CH. v.

to zero as

B is convergent, the sequence p 1? p2 Pa* tends has therefore a finite upper limit H. Also we
/o

/o

m pm+ i,

m+2

...

are

all less

than

thus,

if

p=n

m,

we have

Take the

limit of the last inequality as

n tends to

oo

and we

find

because
so that

tends to

oo

with

n.

Now

is

arbitrarily small

and

is fixed,

we must have
HnT|fln
|

= 0,

(see

Note

(6), p.

5),

that

is,

lim
lirn(A n

/f n

= 0.
)

Hence
or
It

B-Cn = 0, MmCn =AB.


is

must

not, however,

be supposed that the condition of Mertens

necessary for the convergence of 2cw ; in fact Pringsheim has established a large number of results on the multiplication of two series neither of

which converges absolutely. of interest) is as follows


:

The simplest
n~ l l)

of these (including

most cases

If

7=2(

)**"%,

F=2(

vn are convergent in virtue

of the conditions

their

product

is

given by 2(

-!)""%,
}

where
1 l
,

wn = u v n + u2 vn _ + ...+ u nv
provided that the series 2(wn y n ) is convergent. Since lim^w = 0, we can write
2

7= Ul - (u 2 - UJ + (u 3 - u 2 ) - (HI -

where

8l

=u

l ,

S2
,

= w2 ~ w n
6"

^3

= W3~ W25
all

Now, by

hypothesis, 8 2

83 ,

...

are

negative, and accordingly

convergent, and therefore the series 8 1 -8 2 + 8 3 B 4 + ... we can therefore Mertens' and theorem absolutely convergent apply obtain

Hence, 2|8|

is

is

where

7i

= Vi = ^i
l

= &ii>n + 82*^?j) -f (w3

Th UH

2 U F= w - (w2 -

- w2) ~ ( W4 ~

and therefore we have


/'!'-=

Wj

- w, +

<r
:,

provided that lim

^',,

= 0.

35,36]

M:I:

\v,

i.

Mi
(in
n.-.-d

iir*-ut
ii

virtu...|,1\

..f

that

limy,,

<>,

M
i

that

\\'

M.-rteiw' theorem), for

vx
less

and
\M-

>iiji|.o--

that

i>

:i".v

index

th:in/>; as
--cause

in
:

7+1

ha\e

n-v^r
1

ti.

lolll

to

,/,

\\e

have

,',.

+
Thus we
find
i

f,

"/!-$)

If,

as

is

8iippc-

tind 7 so

tl.

is

leas

than Ac; 7 having been fixed, take the limit of the last inequality as n tend.- to infinity, and we find lini Thus, since t is ail'itraiily

"_.
1

small,

-0, and it follows that lim "-<. other n-sults are due to Voss and C'ajoii, in addition to those found of 1'ringsheim's aiti'le in the K"i iirjsheim.
lini
'/
1

'i

Ikl.

I.;
/,

two
vol.

nii-re
-2,

i-.-.-mt
j,j,.

]>;i])ers
-J.-)

will

l.e

found

in

IJHM,

and

4<l.

36.

ThU
,j

Substitution of a power-series in another power-series. operation -ivr^ anotlicr t-xaniplc of deranging a double and -r 4-.. oaider tin- aeries
.

t**f(

/._,//

l,
t}

l>

i'

/'.,./-

4- ...:

if
./'

convergent

at

all.

th-y

con\.

al.solut.-ly for

//

<N.

<

thru
Beries

arises
in
tlif

\vln-tluT
tir^l
if
:

tin-

9&y (see Art. of ^u >^t ivMilt


/

50)
i

Tin- <|U-ti..n

nt n i

th-

BeOOnd

gent,

and

ami arran^ino; in jo\vrrs of X is ev. It valuBO, t'>r what apj-.-arthat

from

thr

powers

of

//

i-an

lr

calculat.-d

usino;
-ijual

tin-

rule
///

for

tinl>;i

to

8WM

and tln-n multiplication of series, fOWS of the doulde s


:

+ "/' + "A
4-..
:;

"
:

-f-M
-f...
4-...

88
If this

DOUBLE
double series
it

SERIES.

[CH. V.
x,

is

arranged according to powers of


these

we

are
if

summing

by columns

two sums are certainly


is

equal

made

still converges, after every term 31 positive (Art. (5) and Art. 33).

the double series

Write
|a|

= an

\b n

=/3 n

x\

= g,

and then the new

series is

not greater than

--(2)

Now

this series,

summed by

rows, gives

which converges, provided that Take now any positive number less than r, say p, then the series 2/3 m /o m is convergent, and consequently the terms m have a finite Thus our condition is /3m p upper limit M.
satisfied if

or

if

Hence

if

/3

<s, and

g<( s p

)p

/(M+s-/3o), the

series

2)

of positive terms will converge. Consequently the derangement of the series (1) will not alter its sum. Thus the trans-

formation

is

permissible if the two conditions


|6

(i)

|<,

(ii)

|*|<(8-|6

|) /

,/(af+-|6

|)

are satisfied (where p<^r, b n p n ^M). In particular, = 0, the conditions may be replaced by the one
>

if

n // the series converges for all values of y it is evident that the condition x is sufficient to justify fit r
y

derangement.

The
series,

case 6

is

of special interest in practice;


is

the coefficient of x n in the final series

not

itself

and then an infinite

but terminates; a few of the coefficients are

and generally,

if

n>

2,

cn will

contain the terms

36, 37

N"\ LBSOL1
!

Ti.

<

Ex.

.1
y=i"
iation
all
is

hyH
'>'-',,

tlu-n

tin-

verges for
a

vainin
z
//,

all'.waMf, pn.vid'-d that |#|<1, since z con- ! I/1. Tin- i- -ult is .lvi..M-l\ ./;", where
-

polynomial
Lining that
;

sn<-h
t>/

that

tin-

term

of

hi_;li'--t

degree

ifl

"

= e* and
A ill

= /zlog(l
/x

+.r) (see

A
...,

vanish for
l.rf<.i-'-

= 0,

1,

2,

/'-I, l--ans- in

'

caeOfi

th-

neiiefl

t.-nninat.-s

i-.-a-hinj

II.

cH

=
/'
.

fj.(fj.

an.l

we

'il.tain

the l>in'inial Bttiei (Ait-.

;|

ami

37.

Non-absolutely convergent double series.


tlic

AliiK.st
tiv.-ly

dily Lrnirral tyj ..... f su.-h


l.y

>.-ri.^
t..

has

l..--n

iriv.-n.

.,,ni|,aiM-

r.-mitly.
i:. L ".
(

Hardy;*
series
'

it

n.n-,-sji,,n,ls

tinf
'

t\|i-

discussed '<<

in

Art*

Th.- th.-.-n-ni
''

ia

th- aztenrioD
(/<>
.<//;/i

niii.-h..

^, n
-

m>

/"///"'/- '"^//

'

con>'^

"//

//// '/.<

/-

^^*

.-'.*!.,
In

;<
,.

<,

'

,,+i

'/*'//

'''"'^ ^o -2^0

as either

or n
i*:'.

ta.-t.

just
tin-

as in
Lrivt'ii

th> proof of Abel's leninia (Art.


i-onditions
f"r
//

that

iiinh-i

\vc

22<Vt>
win re
//
is

an upprr limit to

M
''"

''

11
it'

- | fci-fc*-l-*-H'H
i

s..

that

rithiM-

//

or

LB

I.

//
.

-''.
1

and otherwise
A
i-

//

./

.i/

.v

N'-

S2-Z 2 1111
(V
!.,

S
1
,

2-1-2
M

2-f2
1

S
,

that

v V'v;
lM ad<-

+, V(1 ),
//.

whit h .an

as small as w.;

|>l,>as.-

l.y

]I-MJ.T

.-h'-i.-.-

.f

r,

because
''ges.

and P^ | both tend to


AII apjilii-atinn
is
r

and
th.'y

>..

th'

diMilil.-

iv-n

Iiy

wc/)),

<,..

\vhi-rr

o,

'-.',

p
'.

an
.l/'/'/K

SOC., srri.-s

_'.

vol.

1.

UNI:?.

},.

I-.M

roL

J.

l'.U.

]..

I'.MI.

90
For then

DOUBLE
\s mttl

SERIES.

[CH.

and

-.'

\<4\cosec$0cosec$<f>\

while lim v m>n =0.

EXAMPLES.
1.

As examples
n

of double sequences

we

take the following

(1) *m

=
in
0.

I--;

here the double and repeated limits exist and are

all

equal to
(2) s m n
,

m n =(-l) +

(!+!); \m n)
limits

here the double limit


exist,

is

again

0,

but the

single

and repeated

do not
s m<n \

although we have
,

lim /lim

= =

m
(3) Sm,w (4) s m n
,

>-oo

\n
5

-ac

lim / lim sm n >-3o \ m wo

= wm/(w 3 + = m/(m + n)

here the double and repeated limits are again all 0. here the double limit does not exist, but we have <*,<!, and limsm n =0, lim *,. = !,
tt

because, however large p may be, we can find values of m, n greater than such that s m<n and other values of m, n for which * TO ,n>l e. But /x,

<

the repeated limits exist and are such that


lim
/

lim
>cr>

MO \n

s m> ,A J

=0,

lim
n
>-oo

lim
>

s m<n \

= 1.

\n

Similar features present themselves in the sequences


sm .n = mn/(m* + n?) and s m n = l/[1 +(m-n)*]. m 2 3 s G = sm n we have (-l) m n /(m + n
,

(5) If

),

lim /limsw
(m)

\=0=lim /lims m
J

n\
/

(n)

()

\ (m)
,

bat yet
us
to

limsmtn =-ao,
be seen by taking

lim sm n =

+ oc

may

m=

ri

2
.

Here

it

should be noticed that the

limit of the single sequence given by putting w- = n exists and is equal. ; although the double limit does not exist. [PRINGSHEIM.]
2.

The double

series given

by
</.,

+ bn + (, - 6 ) + 2 + f a4 -f ... (-a +!) )+(-a -!> )-a 2 -a3 -a 4 -... b. +0 + 0+0+... 2 +0 + 0+0+... A,
(a
)
()

!>.,

/>.,

.^ives

the

sum

in Pringlheim'f sense,

whatever may be the values

.f

But the sum by rows is only convergent if 2a,, converges; and the sum by columns converges only if 17>,, is convergent. The sum by diagonal is lim (n + />), if this limit exists; and is otherwise oscillatory.

",

/>.

lt

v.|
3.

IfFLES
In
tin(1.

.ul. I.-

neriea

1+2+4+*-i-1-2-4-i-J-1-2-I

j
l'i

ci.luiiin

converges

t<>

o,
:

Inn

row
tin-

li\.

!_"-.
is

<

iii^.-ht-iin's

sum cannot

t-xi-t

and

sum

l.v

diagonals

divergent.

4.

The

series

gran

l.v

0+1+0+0+0+...

-1+0+1+0 + 0+..
+ 0-1+0+1 + + + 0-1+0+..
..

ha> the
tlu<>

sum
H<

l.v

rows; -1

l.v

columns; u ly diagonals; and naturally


Pringsheim's sense.
if

doul>

'iiverge in

In

fact,

if

m
'I

and
i

is

-1

if

m>n,

or

+1

m<n.

5.

given by

-2 + 1+0 + + 0+... + 1-2 + 1+0 + +


..

0+1-2+1+0 + + 0+1-2 + + + + 0+1-2 +


1

..
..

..

has

tin-

sum -

Imth

scillates

ln-twet-n

-2 and
if

-1
6.

and th- diagOIM] ly r.\\s and columns 0. There is no sum in I'rin^hfiin's m=/>, and is ..tlu-rwise 0.
;

The

dnul.l.

2 + 0-1+0 + + 0+.. 0+2 + 0-1+0 + 0+.. -1+0 + 2 + 0-1+0+..

1+0+2+0-1 + ...
+ 0-1+0 + 2 + 0+.. 0+0 + 0-1+0 + 2 +
ha>
the
tin.

..

sum
l.v

l.v

K.US

sum

diaLTonals

+ +0 + 0+ ... =2, and is 2 r<> + + + ... = 2.


1
<

tin-

<>

samr sum l.v -..lui. Thus t/iee three sums are


/we, since
</ w
,

= 2.

7.
l.y

I'l-nvt- that tinmultiplication rule summiiiL: the dmil.

fr

",/',.

+ 'ij>.*i-+ ...
+...

+ </,'

first

liy

ru\\s

and

secoiully l>y

92
8.

DOUBLE
Discuss the following paradox
:

SERIES.

[CH.

If

we sum the double

1111

series of positive terras

first

by rows and secondly by columns, we obtain

+ 1=5

or
[J.

1=0, where
BERNOULLI.]

If the double series 22a m>n is 9. convergent in Pringsheim's sense, it does not follow (in contract to the case of single series) that a constant C can be found such that sm C for all values of m, n this is seen by
| ,
|

<

and supposing 2a,,, 26 n to be divergent. In like manner we cannot infer s m> n < C from the convergence of the sum by columns or by rows (see for instance Ex. 3).
considering the series of Ex.
2,
\

m+n in which a min ( /mn does not converge l) sums by rows, columns and diagonals are equal to one another and to Pringsheim's sum. The common value is, in fact, (log 2) 2 m + nu v Exactly similar results apply to the series in which a mtn = (-l) m n where the sequences (u m \ (vn ) steadily decrease to zero.
10.

The double
;

series
its

absolutely

but yet

11.

Consider the double series in which

and

am
find

= 0.

Here we

<)

2 2 = -|2< IS = + P(W.

[M^Pall

[HARDY.]

12.

Prove that
V V

m?-, n?
tends to
/.<!'
.

when

v tends to oo

pnvilnl

tliat

are omitted from the summation.


13.

trrms for which m = n Moth. Trip. 1895.]


[

If

mam
,

and
then

= 2~ m
/

= - 2~",
If

= 0,

m=0 \i-0

Yi *)--

w=0

m=0

V.|
14.
It

!.\ \.MTi

.u.

,(-ir*"m/(>wm

1ml

tin-

-inn

i.

illatt->
/
,'rt

lirtwi-rn
/

-x
e

and

'

and

sum

oscillates l.rt\\.-rn
ti,,.

and
I!

+ jV
Iin.inwi.-h

[F.-i

and Hardy, Proc.

M.I,

j,.

17.-,.]

15.

'rove that the product

<>f

the

two

series
.r

x
,il

x*
'

^
^
2 (3!) .6!
"

to
*

16.

It

<

,. V

)=i+ y (.,.+
1

.i/^) + ^
-

)4....

iy:<i,
r
i

and
th.-n

J?)-^*"*/(*/^^)-25ffi
-Axing).</)-."('.

v=-i,
.? ).5r(y ,9
2 2

2
),

17.

Verify that
1

L!(n-l)!+J

!':</'

1
-2).
:

./-- 2

K*-)!
+

4?+8

use Arts. 33, 57 to infer I'rym's identity,

1.1
'

./

x(x 4-

!)".'(.' +!)(.'

+ 2)

+< "

Lr

1!

.r+1

2! .r-f-2

3!

18.

8h6H

that

_.

_._
j

....

into

;i

dulle
.

M-rirs.

and transform

it

..
,
,

Taki- /r

lit

and 8O

calcnlatt-

I'l

/,-

t> 7 d.-rimal

places.

[STII.

19.

Ckm^

_. + _^
it

into a dnul'le Beriee, and drdu.r that

i>

c<jual

t<>

94
20.

DOUBLE
Shew
that
(if

SERIES.
series

[CH. V.

|#|<1) Lambert's X1 3? X r^"*"!^? 4"!-

and deduce that

this series is equal to Clausen's series,

l+x
Hence evaluate Lambert's
series to five decimal places, for is also equal to

Shew

that each of these series

the coefficient of
21.

xn being the number


19, or directly,
.

of divisors of
if
|

(1

and n

included).

From Ex.
x
T

prove that,
.

x <1,
.

3?

3?

&
+a

&

'^l+^6 ^l+^ 10 ^"*~l-^2

l-j^^l-^ 10

~\+x
22.

Shew

that, if

\x\<l]
3.r*

2x*

(l+xf
"*

[For the connexion


functions, see Jacobi,
23.
If \x\

between

the

series

in

Exs.

19-22

and

elliptic

Fundamenta Nova,

40.]

<1, shew that

where
that,
if

<^>

.is

the

sum
,

of the divisors of

n (including

and

n).

Deduce

<^_ 1

= 0=<#>

[Math. Trip. 1899.]


24.

If \x

<1, prove

that

where 0(w) denotes the sum 1 and n).


2

X/(rf)
*

for

all

the divisors of n (including


6 6

'

In
[LAGUEHKK.
25.
]

Shew

that in the special series

<>f

Art. 37, the repeated series also


sfiies;

converge to the saim- sum as tinmay oscillate, for instance u = /? =


series

(loulilil,

but the diagonal series


gives for the diagonal
[H,\i:i.v.]

= <=7r, p = l, _ 3 + _ + ....

CHAPTER

VI.

INTIMTK PRODU(
38.

Weierstrass
and
leee
1

inequalities.
ej,
..

In

this artirlf the nunilx-rs

an- suppn

])(>N]'ti\c

than

this In-in^r the case,


1

we

(!+",)(

+ ",)= f(ol -fo1)+olat > +(",+


1

Eeuoe
(

+0,

)(

+",>(

+" >[
S

+('/! +",)](
see

+";;>

>

+<"!

+ ",+

ami
(l)

(-(.ntiiiuin-

this proc,

tliat

(1 4-0^(1
liU(

H-a^l^at)...(l+aw)>l-h(a1 -fa1 +a1 -f... 4


\vc

In

nianinT
1

ha\.1 1

-OiX -",)= -("


1
1

-h",,)

+ " ",>l
1

-('/!

Thus, since

Oj

i^

|<sitiv-.

\\r

liave

(l-al)(l-o^(l-at)>[l-(al 4-a1)](l-at)>l-(al -|-a1 4-fl


ami so
\\v
1

nt-rally,

>>\-> +, +
l

,+

...-

and

tip;
tin-

/,

+</.,+

+
1

1.

\v-

have,

>y

aid nf c2).
(

r.-Milt
1

1+,/jH

+ ./.)...< 4-OnXP -(",+">-...^".)]_!t-utrr tluiii


1.
tin.-

1
.

siiu'i

it

_+Oj+...would tluMi

iiu-|U,ilit

96
Similarly,
(4)

INFINITE PRODUCTS.

[CH. VI.

we
1

find
1
.

(l_a )(l-a2 )...(l-an )<[l+(a 1 + a 2 +... + a n )]By combining these four inequalities, we find the results
(l-2a)(l+2a)all the letters
is less
1 1

(5) (6)

>n(l+a)>l + 2a, >n(l-a)>l-2a,


and
and
1,

where
that

a denote numbers between


1.

such

2a

than
,

39. If

a lt a z

a3

...

are numbers between

1,

the

convergence of the series 2a n is necessary and sufficient for the convergence of the products n Qn to positive limits

P,

as

n increases

to co,

where

For clearly

Pn
is
3.

increases as

increases,

and Q n decreases.

Now,
that

if

2an

convergent,

we can

find a

number

such

= am+i + an+s + am+s + ...

to oo

<L
we have
'

Then by the

inequalities (5), (6) of the last article,

...

+aB ) ^ P m

and

-a
n
is

Hence, provided that

greater than m,
ra

we have

Pn <P
and
Thus, by Art. such that
2,

/(l-<r),

.>Q.(l-<r).

PH and Q H approach
we can

definite finite limits P, Q,

-P^Pm /(l-<r),
But,
if

QSQ.(l-<r).
find in so that

Sa M

is

divergent,

no matter how large Hence, by the same


and
It

N may be.
inequalities,
I

P n >l + N, QW <1/(1+JV) = lim Q n = 0. consequently lim P n


,

if

w >m,

should be observed that if a product tends to zero ^mi^, without any of its factors being zero, the product /*

38, 39|

POSH

97

rallrlism

with
_'>,

tli.-

tli. -MI

Ex.

1.

ili'-

I'i'xl

will

approach a

limit

because

thr |.i"diu-t
I)(M

+ 1)

+l

so

tll.it

lllu '/

Similarly

+^

\(

+--

<

It

)(

hough iU value

is

'ulaU-il so iradily.

Ex.

2.

8
(i

+ *)(i 5

:-i)(i-S)0-i)...

\\ill

(livt-rirf
,,

M+l
liin

n+

'

3 = T-' V "- ^3'4 l


1

-1

^
= 0.
K-h that

/'==

-f. ,

lim
MI

V'.

= 0.
f,

Ex.

3.

/
c

I.,
;

iiii(/,,

iuulfi tin- u'ivfii

if. nil,

timl

an

l>,,>lb>0
have

if

n^tn.

and

:.:Mso that

Dlim

x,
lim
</

().

It

is

easy to 896 that the ur^mm-nt "f Ai


"

iiti-,l

>rove that
<</' //
.

= (!+.,,

+-,.(!+.

98

INFINITE PRODUCTS.

[CH. VI.

For if, in any new arrangement (represented by accents), we have to take p factors to include the first n factors of P and Q, then

Now n
to

can be taken large enough to bring


Q, respectively, as

Pn

and Q n as

close

and

we

please.

Consequently
In like manner,
diverges to
0,

lim
r >oo

P/ = P,

lim
r ><*>

Q r = Q.
'

if

diverges to

oo

so does P'

and

if

so does

Q'.

taking logarithms we can see at once that the present theorem is deducible from the theorem of Art. 26.

By

40.

Convergence of

infinite

products in general.
infinite

It is quite possible that in

an

product

the numbers

loss of generality it

u lt uz UB) ... may have may be supposed


,

both signs.

But without
all

cally less

than

that they are for there can only be a finite


1

numeriof

number

them greater than

(otherwise the product would certainly or and the corresponding factors can be oscillate), diverge omitted without affecting the convergence.

Now we

have*
log(l

<u
Thus,
if

+ u)< ?u
or

if
if

is positive,

<$u*/(l+u)
l

0>M>-1.
...,

is

the lower limit of the numbers


1,

1+Ui,

+ u.

...,

l+u nt

we have

fu

,/..

For

log(l

)=Jo|

"

Hence,
but,
if

if

is

positive

p~ < u - log
xdx < u
log

+ u) <
+ n)<

is

negative

\
.'ii

"f^l
1

JQ

+U

39,

40
1

TERMS
<f

99

Consequently,
can
mad-'

teries

-ujtaeo/

th-- di:

M-1-HWrK..+tf >-log[(l+tWi)(l+*Wi)---(l+*O]
!>

arhitrarily
is.

bet

how

large W

small by properly choosing m, no Tim- we have tin- theorem


:

ij

1
i

/<*

to oo

if
:

2u w
!

diverges to

'/" i/ 2u,,

kites if

in*

\vi//

h;i\.-

!"-

-4-

//

>
>

in'
'

2
(

4-

//

>

if

"

i^>

poeitn
/

..r

if

0> > -1,


!+",.,). ..(!+"

so

tliat

..+

...

//

>-ln.i;

>
wluT'I

H-...-H4) A
1,

L
is

is

the upper limit of

(1 +%),(!+
9O fl"lf
'

jrli

/'

"//'

not

+x)
vcd/UA

//////'

1"

/'>,

///'

d-h^xi + ^jd + r
to

d"

o.
tl;

.nly cases not covrr-l ly in wliich Su^1 diverges and ^i//

Tin-

in--

in.-thod are those


t

.-itln-r
!'

divergee
(

-f

or has

,i^
It

its inaxiiiiiiiii

limit (in

i-as.-

<.v

-illatioii

).

is.
v

perhaps, a littl.Loth dix.TLiv


it

prrph-xin^
t

at

tirM

si^-lit

that

\vh-n

Converge; hut
typ.\
pOSJ

is
)

Fr. Id
MI-.

^'-,

th- jn-mlurt may n.-\-.-rth-l,this juit t-asy to const rurt a pi-nluct s of a COnvei a li\
s.
.

>t'

and

i'nrni

the product

of

which

th<

Mth

and

-J/'th

tenii^ aiv given

hy

,=!+/.,
//

Then, provided that lim and BO olvimisly C


diverge
if

(),

the pi-odiu-t
Ar;
:'.'.'
i.

-f "

>-

ll<

+Cm )
will

(by
)

Kurthi-r

1"
;

-''/.-l

J^'/

=-

'

isdivergeni

and

us page.

M not

abso,

LOO

INFINITE PRODUCTS.

[CH. VI.

2unz must also diverge.* This condition can be satisfied in many ways; one simple method is to take c n = d^, and then we must suppose 2dn ^Zdn2 both divergent, and 2c n3 convergent; for instance, we may take d n = n~ p where %^.p>\. The u2n = nproduct is then given by u 2n . = n~i
,

Ex.

1.

Since the series

1111
also.

are both convergent, the two infinite products

0+(l-i)(l+i)(l-*)... and (lconverge

In fact the

first is

obviously equal to

and the second

to}.
Ex.
2.

Cesaro's theorem (Art. 22)

Since the series ^ + |- + ^ + -^+... diverges in virtue of it follows that the infinite products

a+jxi+i)a-}xi+ixi+#i-..*^
are both divergent. In fact the first diverges to for they are equivalent to the products
oo

and the second

to

Ex.

3.

Since the series

J__JL _,_!__!,
is

convergent,

V2 v/3 but ^ + ^ + 4 + 5 + --.

v/4
is

V&
it is

divergent,

clear that the

two

products

and
hoth diverge to the value In fact
<>.

1+

j_)[,

-^3-1 -^^^[i-^-.j,
is

so that this product

(l-a

w ).

is always less than 1, and can be put in the form Further lirn(waw ) = l, so that our two products diverge to by

Art. 39.

Ex.

4.
;

diverges

If M n = (-l)"i, it thus the product

evident that

2?*,,

oscillates,

while

2?/,,

must diverge
*If

to 0,

which may be

verified

1>\

inspection.

Su n2 were

convergent, the divergence of

2n n would imply the divergence of

the product.

40, 41

&B8OL1 TE

<

"N

101

41.

Absolute convergence of an infinite product.


<//s,,/,//,/
(

y convert ! ui,
//// /'//

tl>

II*

,-./>
f

/"

WMW

pendent of
i:

'>T8.

Writ.last

then -"

converges; and, as explained

articl.-.

\v can suppose a w

<

BO

that

"'<",.

II-

tlso

convergent (Art. 8), thai


it

--"ii

verges; and
11(1

Zt/,,

.ind th-ivf(.nin virtin>f

-MO

a tli-!vni
i-

proved

in tin-

last arti<

Su|ij...v,. n.-xt

that

tin-

861166 Stt,
i

l-ranged SO as to b"
lr-vity

IK!

writ.- fur

...

(14-14
'
'

r.-a+v^a+tg-O
>

'...0

'

Then suppase
(and multiplying out
l.Tiu in
I'

j)

chosen

(>
.4

//

BO that t/p contains th-

B.-(i+6i)a+^)...(i+W.

whole

it

nmtly p contains the whole of B n ); on is CMiitains every } vi.l.-m that A p /B n

Up/Vn

\.

hut with the si^ns

made
n

p.

.sit:

H.-nce

\Up/Vn -l

^A p 'B

-l,

and
s.i

Fn ^//
l^-^l^.l,
:>!>.

that

-/;
/^,,

N"\v. as rxplaim-.l in Art.


s,M|u-ntly
/',,

lim

=P

lim

.-!

=A

say.

Con-

can

!>

t'..uinl

BO -T--at that
t

A
ii'-i-

>A9 >Bn >A-ie lfjp>>


^i p

-/y,,<

if if

//

>
:

and
Hut
E7^

\UV -V <

//>
and
th.-i

if

approaches a limit ^7 as y trnds t<> y //>//> /^ Wr ha\V 9 is the Thus, if greater .if n and /^,. \v.-

UU

<

//

/,

li

Iff-r.Ke,
that
is

&n>i
'hiite

Hn,

V=U.
product by

TIuluit
in ay
l>y

IllX'MIIH'nt

us.

'.I

I..

r-t.tl)lish

Uii'lii;iMir-

tie
..

little

change to shew that

nn-al>solutely
\

averment

be

made

to converge to
-f

any

to

jmdiverge, or to oscillate,

ititinit>

altering the ..nl.T

tin-

fi.-tors.

102

INFINITE PRODUCTS.
infinite

[CH. vi.

Suppose that the value of the product is P when the positive and negative terms occur when the limit of the ratio of the and let its value be alternately number of positive to the number of negative terms is k.

II[l+(-l)

Perhaps the case of chief interest is that afforded by the n~1 is positive and a M ], where \im(na n )=g.

product

Then
where

Km (1 +,n+1 )(l + a an+8) ... (1 + X/P= n \im(v/n) = k.


-oo

2v-i),

Now
article,

it

is

it is

plain that clear that

2a M 2

is

convergent, and therefore, by the last

can be made arbitrarily small by taking n large enough. Pringsheim's method (Art. 28), it is clear that
lim (
2w+1

Further, by

+ a2n+3 +

+ a-i) = \g log k,
k9
.

and therefore
or
42.
It

X/P=Jk

The Gamma-product.
is

evident from the foregoing articles (39, 40) that the product

is

= divergent except for # 0.


I
.

But we have

so that the expression

increases with n.

Also,

as.

in Art. 40,

we have

[Art11
where A
Art.
2,
is

(1)] >

either

1, if

is

positive, or

Sn approaches
(Art. 11)

a definite limit

l+#, negative. as n increases to oo


if

is

Hence, by

Now

l+||4-...+-logJi
(7,

approaches a definite limit

and therefore

x log n log n log (n*IPn \ Now so that n*IPn has also a definite limit; this limit Gauss's notation.
THUS

P =

is

denoted by II (.r)

in

n *-"
= e cx - s =e cx
lim e~ Sn = e
H
*-ao

which, again, can be written in Weierstrass's form


(
'

U
r=l

When x

is

a positive integer, Gauss's form gives

II (.r) =.<!,

because

(l+)(2 +

)..

.(*+)

n/\

41, 42]

G IMMA-PEODl OT,
convenient
\

Although we have
"t necetiftai

rentrict (1

-f .f)

to be positive,

nvergem-e
Converge
liy
if
.>

prod an intrjjer.*
It i-

Mill

easy to see th,r has any n-u'ativr valm- wh


;

and

it

is

rasy to

\.-rifv

integration ly parts that

Kuler's integral

rd
has the property
,.f

1,,-in^
tin-

,.

j,,

t<

./

when x

is

an integer; and we

mtii-ipatr

(|uati.!i

r(i+./-)^n.
\\lii.-h
It

will

be proved to be correct
-f
1

in

Art
<>f

I7fi

of

bhe
tind

\\i-

ebl

in

IK-

definition

!!(.'),
'

wr

w*"^

"

+
",!'-

4-l)
It

= (.r+l)II

f,,ll,,ws

that

(1+*X* +*)...(*+*)- H(

+*)/!!(*),

and

f.iiM.-juentlv tin- definition leads to the e<juatim

or
It is

1=

,. hni

often convenient to write th-

last

iMjuation in the

form

I'.y

the notation t-xjilained in Art. 3. reversing the foregoing argument

we

see that

tfo function

I'.

hi/

(I"

II

-4-l)

= (r4-l)Il(.'-)

<

II(w+./-)ru/<'i;

AMPLES.

1.

Discuss the convergent

/"(/()

is

a polynomial
II
1

in

/*.

2.

Piovt- that

~
of
/
.r,

[(

(w

= l,

2, 3, ...)

absolutely

f,.i-

any valur

provided th;r
r

integer

ami that

r
if

\i-i

^"VfTflJ
j converm-nt
./

<

1.

Tlu- ronvfi'iifiu't- j)Tsists also lur i-oinpU-x

values of

(set-

104
*
if

INFINITE PEODUCTS.
Wn ~
diverges to
if

[CH.

_
;

then
if

Hwn

m>1
if

to

oo

if

m<l.
to
oo
if

If 7n

= l, Hwn
!

a = c.
4.

diverges to

a<c, and

a>c; and
,

converges

[STIRLING.]

If

(),

M 2 = 0,
2
,

then 2w n 2wn $ Verify that the same is true

= P>k

u 2n -i=-n~p u 2n = n~p + n~ 2p are both divergent, but 11(1


,

where

n>l

and

+u n )

is

convergent.

if

Ma-i = - W--P, w 2n = w~* +


5.

?i~ 2p

+ w- 3^,

(n

>1

).

[Jfo^. rnjo. 1906.]

Verify the identity

~~ .,of
6.

-~~
.

Shew that as n tends to infinity, the product x except 0, but the series converges, provided
Prove that (1+^)(1+^ 2 )(1+^4 )(1+^8 )..
-n-

diverges for that #>0.


if

all

values

= !/(! -x\
-^-,

\x\<\.

it

7.

Verify that
i

.*.*,

cos -

x
.

cos -^

cos

x 3

...

= sn x
...

and that
8.

tan|

+i

tan

J + gi tan^ +

= i-cot^?.

[EULER.]

Determine the value of

in terms of the
9.

Gamma

functions

F(l+c) and T(l+x + c).

Shew

that

j
.

frm. The

product

is

ir(^+2w) r(w) n f f^TT(x + n) F(27i) J

10.

Prove

that, if

is

an integer,

lim

verify that

and that

(w

l)

wa M -(w + l)a ll+ i = to + (<- l)a,, +1 w6 M _, -7ia ( + l)6 n = < 2 [(2- l)a,, +1 -f n6J.
ll

Shew

that limtta n = 0, Iimw 3 6 n = 0,

if

^>^, and deduce

that

Hence prove that

6,,

= 2 ('It - 1 )

or,,

3/*.

vi.
I

LMPLE8
I

L05
hree hyperge>i
'

12.

tin-

^eneral

'

erief

fry,
111

i),

i'>

/'<-

-l,Ar.
that

'-fty +

i-

\vlii.-h

y>u +

/;f.

Then prove

(y

-)(</-&) = /:?'/-, + ("


liin

)/.,

(/m,,)=0.
-

Deduceth.u
and that

/;-( y

_/^r, (y- tt )(.l -/*)=.,

J= (y-)(yK\.

13.

Kr.-in

1)

--_
:)

(X

r(y-a-/J)-

'/^7 +

i-

and

slu-\v

that

tin-

last

t-xprrssimi tt-nls t> the limit

as H-*-o(

'-

duce

.14.
if

,/= IK IK.

-,/->,

?1
y;i

+v),
9ctf s

=n

!-!^F=n(i+ n
? \
(i

-,/if

the four iii-udiirU air als..lut-ly (-..nverp-nt

<

1.

= II(l

-y"), yi7 2 =n(l-iy,Y.'/ 3

and
Thu>
15.
.

=l:t

[<1- 7 )(1
i>

-7^(1 -7-')...] = (l+y)(l+v-)(l+y


convergent,
./
;

)....

[K
.lately

It

1"

al>s<liitfly
;u ,y

tht>
it

]i<>duct
!

11(14-.in

r,. Ilt .-iivri u

f,,,.

vain.- of

and

can

expanded
r,

an

al>s<>lutel\

c..nvei-.;,. Ml

.,

+ r ./-+r^+...,
1

ui.c,,.

Shew

aU,, that
11(1 +.*//)(! +/'

r)=

'+

r i(-''+

K.16.
I.

iiavc at

.nee (!--/

and

l.y

the
...

la>t

example

Thus we find
</+:
,

which give
=

<i-f)<i-y',
and
u'enerally
/'
('
/

-^)(i/'
.

106
17.

INFINITE PRODUCTS.
If

[CH.

we

see,

from Ex.

15, that

we can write
so

But qxF(q*x) = F(x), and


yielding

we

find

F2 = V^, F3 =

Thus

^)=F [l+ ? Cr+l/^) +


F we may
use the results of Exs. 15, 16, from which

To determine we find

F" =
2

Thus
because

pn 7 -l<
<Io

+
<Io

+...,

where g
r

= U(l Q
-

I<?I<1 and

Hence
so that

Pn V

-1

Pn+r/Pn >qQ, P >q < q2n/q (l - q


2
2

"),

lim(Pn F ) = l, or
16,

= l/?owe have

Thus, using the notation of Ex.


/(#, q)

Ch. V.,

= q U(l+ f-\v}(l + q^lx\


[.TACOBI.]
14,

from which a number of interesting results follow.


18.

From

Ex. 17
/(I,

we

find,
,

with the notation of Ex.

= <M 22 ?)

/(-I, ?)=<M 3

2
,

/(?, y)

Or, writing these equations at length,

we have

where the indices in the third

series are of the type


<?)/(!

n(n+l).

Again, by taking the limit of /(a?,

+<?/#) as

# approaches -g, we have


[Compare Art.
46.]

the indices being the same as in the third series.


19.

Again, from Ex.

17,

we get

so that

tfo/?!

= 1 - 2? 2 4- 2? 8
2?i 2 .
.

the indices being of the form

Also
so that

f(Jq,

n Jq) = 2H(l-q )

11(1

W?3
[GAUSS.]
l

the indices being of the form $n(w+l).


Similarly,

/(-<?*,

^)=n(i- ?

").
2

no -f- w -9 '- )=n(i -r/)


3 2

or

M3 = l-(? + 9 )+(7

12 16 +9 )-(9 + 9 )+-->
7

the indices being alternately

tt(3?il).

[EULEK.]

vi.]
20.

LMFLEa
Writ*

107
.

ml
b
\

|>'it

'.

N 'y

iii

j.la-

in

tli.-

Thru
x s

ire

x
s

!!

D
N

-*),
2 :i(i-9 -',

-that

Thus.

<>n

niultiplirati.'ii,

we

tind
/>]-'

<.//

Wi*
ami
si
i

we have the identity


'/,'!
/'I"'.

In

particular.

if

we

\\in.-

I.

\v-

find tin- intf-n-stin.-; ivsiilt


8

7i

whii-h leads a^ain

t<

the identity

(l+iy +
the
an-

V + *V+---)
-ivt-n

-(l -2y

+ 2y 4
,-,1,,

:.'</'
4
,

-K..)

"+.-.)
\\!i-iT
sei'ir>

tin-

in

\-]\.

!s

[.TACOBI.]

21.
-..ii

,,,,..
if

if

-.-<-.
l\/

:;;;>U,e
\\"li-n

product,,
8ati>ti.-.l,

verge the product

= / and
in
tin-

i!/>.

thcs- d.nditiona are

-.\|

form

In

parlii-ular.

prove thai

(/t+a + 6)

= r(l+a)r(l4-6)

*?(+)(.
22.

I'lovr that

Tak- the trims


I

in

pairs

and use the

last

example.]
(>-,

23.

.l-nntes
,.ij

r(#)/T.'),
( '

can write

,,_,V

m !,,_] __J-- s-1--... --?-Y T 1 + r 2-fjw+jr/


||
1

Thru wr
24.
It
\

tind

_e.
n]
of
infinite
01

..
f

is

easy to drdiu-r
11.
di-r
hi),

fi-.-m
I'./

th-

thr..i-y
I

product* Abel's
diverge togct
:

that

and

!<

OODVWge
Il.<
.

In
if
.

the pr.-diK't
..n

II

..

..nst

also di\rii:r

wilidl d:\n-gea to re a m >0.]


I

Otlirr o\ani]tlr>

pi..ilu.-t> will lie

found

at

the emls of Chapters

CHAPTER

VII.

SERIES OF VARIABLE TERMS.


Uniform convergence of a sequence. may happen that the terms of a sequence depend on some variable x in addition to the index n; and this is indicated by using the notation Sn (x). We assume that the sequence is convergent for all values of x within a certain interval (a, 6), and then the limit ii m 8n (x)
43.

It

n->ao

defines a certain function of x, say F(x), in the interval (a, b). The condition of convergence (Art. 1) implies that, given an
arbitrarily small positive

such that
|

number

e,

we can determine &n


if 71

integer

_F

<
m

e>

> m.
we can

Obviously the definition of

is

not yet precise, but

make
in

it

precise

by

agreeing to

always

select the least integer

the prescribed inequality. When this is done, it is natural to expect that the value of in will depend on x, and so we are led to consider a new function ra(#), which

which

satisfies

depends on

We
is

e and on the nature of the sequence. note incidentally that, regarded as a function of e, in(x) monotonic since (for any assigned value of x) it cannot
e

decrease as
Ex.
1.

diminishes.

If

Sn (x) = \j(x + n\ where .r^O, we have

= 0. F(x)= nlim,,(.r) +ao


Then the condition
of convergence gives

so that

m(.r)

or

*+*>!/> = the integral part of (l/c)-.r, when .<!/, m(x) = 0, when xzZlfc.

43
1

[FORM
Ex.
2.
It

>.

<

'

>

'

< -

v*
1.

MI

kb
i

all

values

,,f

R,

\v,-

nm-t

tak--

Ex.

3.

It

uli.-re

-r^O, we have
-n
;

ami

It

is

easily >-rn tint


il

part

"I

(>,

and
Ex.
4.

'=0.
[f

d+n** 1

),

j?

bt-i
>'
.'

ted,

ire

!..<

/'(./)-

liui

-0.

-to Exs.
Tin- .i.nditiMii
..f
.

n|,
tinltlu'ii.

if<i
intr-i-al

part of

[1

+(1 - 4

)*]/2

x\

if

|j?|>0,

= 0.

It

will
tluin

U
1

st-cii

that

in -

Kx. 1

the functi'in
-

///(

./

>

i^
1-

alway-

e:

lut in K\.

/:

as .f-*l

ami

in

lining
tfl

that
,!'

<

>.

Thix

consideratioD

i'urth.-r

sul<li\ ision

cnn\.

Bquenoes, which

will pr-

inqmrtaiKM' in
distinction.

the

s.-ijUcl.

and

it

n luces a

UK. iv

snl.tl.-

\V- ^hall
flu-

vay that
(/.
/').
J

'

uniformly in
I

ink

/'<"/

//"'/

'

98

tl>

i>

-//; l.nt

ln-i

me
ami
/>).

iii\
,-

may vary with


variea tV.m a to
t> intinii

e.

nni^t

1..

Thus, alixt-il

6,

u].j.-r limit

si.

canmH t-ml

any point Thus in


values of
OQD

in

th- interval (a, Ex,


1

x
it

sin
:i"t

ni

tak--

/<(e)=l

Uniterm t'..r Hut ii. e.


r

all

}>tin-

nnil'nrm in an interval
in
tin-

-aching
o)t

u]>

/x

uniform
th.-n
t,

interval

(0.

it'

'><'<!.

we can

110

SERIES OF VARIABLE TERMS.

[CH. VII.

Hence in Ex. 2, x = l cannot be included in any interval of uniform convergence such a point will be called a point <>f
:

non-uniform convergence. Similarly in Exs. 3, 4 the point x= must be excluded to ensure uniform convergence. This distinction may be made more tangible by means of a graphical method suggested by Osgood.* The curves y = Sn(x) are drawn for a succession of values of n in the same diagram
;

this is
if

done in Figs. 11-14 for the sequences of Exs. 1-4. Then, Sn(x)-*F(x) uniformly in the interval (a, 6), the whole of
F(f)=f

Ffo)=o
FIG. 13.
FIG. 14.

the curves for which

?,>//()

will lie in the strip

bounded

glance at Fig. 11 will shew that this does occur in Ex. 1. But in Ex. 2, as we see from Fig. 12, every curve y = Sn (x) finally rises above 2/ = e; and the larger n is I thus x taken, the nearer to x = 1 is the point of crossing
;

by y = F(x) + e.

is

In the same way, a point of non-uniform convergence. = () is of non-uniform cona 14 shew # that Ki-^s. 13, point
3,
4.

vergence for each of the sequences in Exs.


i

'-.tin

of the American Math. Society

(_>),

vol.

:\,

1S'.)7,

p.

:,!).

43
1

\ll'o|;M
a

OOin
..i'

Ki:

In

ordei
in>t

d.-linition
l.-t.

uniform G
of
that
/

Involve the actual


follov.

niiinatic.il

due.C0ll\
,

tin-

to

of

An
>rm con-

/respond n>>!

number

e.

tile

fiil am

>n<!ex

m, which

is

independent of x,

and

/>

#m-h

//"'/

')
,///

<,
and for
')

vafa

/A"//

///.

all

points

tl-

It

will l>e
jx

..ii<litinn
.i'

that

seen on coni|>ari.u with Art. h- iinl.-jM-n<lcnt >f /// is t<


x.
\

that the nnly


\vh-r.-a-> tin- t.-nn>

,/,

M'<|Ui-nc- an- functions of That tlic cnmlitimi


tini

id.-nt, for
1

if

N,(r) t.-n-U

to

F(x\

>'

''//-

^- can
>

\\-i-it.j

///

IJL(

:.e).

so that

\Sn
II. -nc.-

<H
uniform
-

''

n>m-l.
<-r^vnc-

the

i-oiMliti.ui

for

c >n\

leads to

tin-

The condition
must
an<1

is
t,,

al-

///

for

if

it

is

satMi.-d.

convrr-v

some

limit.

\\

in

virtue of

****
|

lim

N (.0=/
/(
|

we bave
llenc-.-

F(. O-N,,U)

^e,
if

[Art.

\F(a-)-S n '.'-) <2e.


noticed that
that
is.

>

//'.

and BO the condition of uniform convergence


iould
lie
//

to

'i^tied.

closed:

if

*/<.</'. that .r =
l#,

it
>>

will R!BO

converge uniformly
discontinuities oi

unifnnly ^aason

f.-r

<i.

are not

found so that

,(^)|<
a

je.

if

<<6 and
tf

>

A U.i.
tind
o

si]

continuous function

90 that
l^'
1

and

<J
%

-a<4
J
1!

where

may defend

on

\\.

112

SERIES OF VARIABLE TERMS.

[CH. vn.

Hence
so that

Sn (a)-Sm (a)\<e,

Sn (a)

= converges, and x a can therefore be included in


Similarly for x
b.

the interval of uniform convergence.


Ex. 5. The sequence because lim*S'n (.r) = and
interval
is

Sn (x)=x"(l Sn (x)<.l/n,

#),

O^r-r^l, converges

unifor/it!//,

since the

maximum

of

S n (x)

in the

with
of n.

Ex.

given by # = w/(w + l). The reader should contrast this result should draw the curves y = S n (x] for a few values 2, and

44.

Uniform convergence of a

series.

If, in Art. 43, we suppose the sequence to be derived from a series of variable terms

by writing

Sn (x)=fo(x')+fl (x)+ -"+fn(x),


test

we

obtain the

for uniform convergence of a series in

an
x,

interval

(a, b) in the form: It must be possible to find

a number

independent of

so as to satisfy the condition

\fn+i(x)+f+*(x)+...+fm+P (x)\<e,
at all points of the interval (a,
b).

where p =

l,

2, 3, ...

Each
series

of

the examples given in Art.

43 can be used to construct a

by writing
is

but a more natural type of non-uniform convergence

the following

* Here we
so that
find

H^ + (TT^
\

x2

x"-

S n (x) =
F(x]

and
There
will see
is

/X0) = 0.
a point of non-uniform convergence at
.r

= 0,

as

the reader

by considering the condition

But, just as the general replaced by narrower


tests

test

for

convergence
III.)

is

(compare Chap.
practice,

\\hicli

usually ,nv

more
tests
:

convenient
the
test

in

ordinary

so

here

we

usually

replace

above

by

one

of

the

three

following

43,

44
1

NVI

ND

M'.KI. -

I:>TS.

i:t

(1)

Weierstrass's M-test for uniform convergence.


<.f

The majontv
proved
Ifl8

861166 im-t

with

in
\>y

to

<

uniformly
de--rrih"d
l.rietly

means
.!/-

elementary analysis can of a test due to


fl"

and
flu if
,)

as the
'

/Hist-

for nil

/"
i>r<>/>''rty
,

fa

has

'!

.!/

re

Mn
-,,,/,,

?>

'irf constant,

independent of x:
Tin n
vn
f/

<///'/

XH/>]HM
is

tin-

series

~M

vergent.

fl"'

series

2/n (a;)
Art.
'

and
the

Tlie
to

ahsolute

absolutely convergent con\ci-^ence f(llows

interval (a

at

once

from

18;
to

realisr

uniform

niher that

convergence, it is only for any integral \alue <f p,


t~<

if

^r^ if

IS
('nsfijii-ntly,
in
if

\\v
e.

choos,f

///

90
is

Jis

to

make

th'
e:

rt'inainder

1M/,,

lss than
[g

'(./)

also

Less

tlian

and

this

cli"i.

nlivinllsly

\\\(

IcjiriK lent

of X, 80 that the

of

uniform
Phys.

i-onvrr^.-nci!'>/

ii.

satistinl.
],.

[CoiMpan- Stokes,

and
(2)

rst

\>l.

-2sl.]

moiv

Abel's test for uniform convergence. ilrlicjititi-M for uniform converg


to
A1..-1.

due.
in Art.

in

and has
2

1.,.,-n

aln-ady nu-nt ion-l


nt;
>>
fl'
f

l!:

xeries *2a n vn (x)


'i'ir

and
>;,i
,!
-

tlif

//////

'

fined

'//

fl"

vnti /''//.

K"i-.

in

virtue of

tli--

COBV<
,

we can
f

find

in.

8O

tliat,

whatc\-r
+!

]>ositi\v inte-vr
.-.-

may
.

l>e,

", .!+",

!-!-

B H2-|-----f

"+|.
of

all

nuniei-icallx
6

Then,
that

in

\irtue

lemma

i-y

hyjM.tht11

<AT.

L8.

114
.

SERIES OF VARIABLE TERMS.

[CH. VII.

Further, since a n and K are independent of x, the choice of is independent of x also and therefore 2a n v n (x) converges uniformly in the interval.

m
of

It is obvious that the


x,

terms a n

may

be themselves functions
tl
j

provided

that

2a n

is

uniformly convergent in

interval.
(3) Dirichlet's test for

uniform convergence.

This
1,

is

also

more

delicate than the I/- test.

(Compare Exs.

4 below.)

The
(a,
b),

series *2a n v n (x) is

uniformly convergent in an interval


oscillates

provided that 2a n

between finite limits; that

for

any particular value of x in the interval vn (x) is positive and never increases with n ; and that, as n tends to oo vn (x)
,

tends uniformly to zero for all values of x in the interval. For then the expressions

a m +i \am+l are less than a fixed


,

+ am+2

|,

-..,

number

K\

Kn+i + <Vf2+ ...-ftf m +j,| and we can find an index

such that
for all values of

v m (x)</K

x in the

interval.

Thus, using Abel's

lemma

as before,

we

see that

m+l for all points in the interval.

Again, the terms a n may be changed to functions of x, provided that the maximum limit of 2a w remains less than a
|

fixed
Ex.

number throughout the


1.

interval.

Weierstrass's M-test.

Consider

2 ^7^ 2
OOHW
np

~=i
and
i

(fl);
.r,

these converge uniformly for all real values of

because then

Sinn* ., 1 np np
Ex.
2.

<

np

(1/ ,(P) is

con

nt

yl&eJ'*

^^.
case

Consider

the

vn (.r)=l/7i x
(0,

(O^I./'^l);
~c,,

then

^(ajn*)

converges

uniformly in the interval

1)

if

c-onverges.

[DlRIOHLKT.]

Ex.
If

3. is

Abel's

test.

2an

convergent,

converge uniformly

in

the interval

(0,

1).

[HARDY.]

44,

45
1

Ml\l MY
4
.

i:,

Ex.

'

nf

2r,

(p> 0)
in

then WTii

we

see

that

lioth
\

series
i\
i-

converge
anirlr.

nnit'orinlv

an

in

~a

\vhc:

posit

45.

Fundamental properties of uniformly convergent


earlier analysts
<

series.

Caiichy and the

\\ith the

lined that the continuity of fr.im that


.

/'(.'')

exception of Abel)
1>.-

lim >'(./) could


n-*-<n

deduced

immediately from K
that
a

that this assumption is no-f Art. 4:i. Kurtln-r. tin


/'(./)

,ples

discontinuity in

imj.li.-s

point

of

nondis-

uniform convergence; although K\. -l. An. 4:1. indicai DOn-Unifomi convergence does not nee, ssarily invohe the
continuity of
\i;ain,
if
/';

w- wish

to integrate

/'(./).

the

r|lini
is

>'(./ )|</./-= lin,

"N,ir,,/./'

,tf
n..t

n.-ci'ssarily
1

tni- <-ithT. as will be seen from th- .-xanipl.'


.,

on

}.

US Ix-low.

In

IS47 Stok.-s* published his

lisciv-ry of

tin-

distinction

In-tw.-.-n

uniform
in

'ivm (1)
Beriefl

and non-uniform OOnveTgenoe, and below, which <-stallishcs tin- continuity of


Ifinrntai-y

iv.junvd
is

due to
(-xc-pt
in

Tln-nn-m c2\ on analysis. \\Yii-rM ra-,. and smns not to ha\'l.'t-tur.-s)

int.--

until

1^7<>

//

fl,r
(,t.
/,).

'Hivergei'
\f

tl,>

',

,,t, ,-,-,,/

amd

each

of
''-

th>

cinifiniiiin.fi

in

tin

iiif'Tcnf,

90

'/"

//"

9Wm
c<

/'

Kr.
nuiiilu-r

in
///

virtue of the definition of uniform

can

chosen
such
a

independently

of

mver^ence. X (provided

tln

>nlv

that

ti^Lx

= ln,

in

way

that

ilix-civrry \\as
\\.t-

made
IK

alsu

l.y \\Yi<-r>t ra>-

and

s,

,,),

],, lt

xtoljgg'g

paper

pulilislird

yrar
i^

auarr
(see

of

tin-

distinction

pulilished. pp.
i>7

riu>
si
,

some the years at I'-.i-t bd ''s's paper WAS dear from paper- in tin- tii-t \olunieof \V ci erst Pass's \\
ivinaiiifd
iui])ul>lisiuHl

\vljiL-h

for

aU.ut

.Vi

\ears.

116

SERIES OF VARIABLE TERMS.


e

[CH. VII.

no matter how small the constant

may

be.

Now

write

and

it is

then clear that

where
of x.

it'

\F(x)-Sm (x)\<&, must be remembered that 771,


c is

(a^x^b),
e are quite

independent

Thus
so that

if

any value of x within the

interval,

we have

\F(c)-F(x)\<$e + \Sm (c)-Sm (x)\.


77i

Now

being

fixed,

and therefore we can

8m (x) is a tind a value

continuous function of x, S, such that


if

\8m (c)-Sm (x)\<fr,

\c|c-o;|
(a, b).
;

Hence

\F(c)-F(x)\<e,

if

which proves the continuity of F(x) within the interval


It
is

and

it is

not unusual for beginners to miss the point of the foregoing proof therefore advisable to show how the argument fails when applied

to such a series as

(l-x) + (.v-j(?) + (x*-a*)+...,

(Ex.

2,

Art. 43)

when we take
Here
and

= l.
to
co

/m(#)+/m+i (#)+...

=xm
=0.

if

0<.r<],

/(l)+/+i(l)+-

to oo

Thus, if we wish to make both these remainders less than choose m, if we can, so that vm <i e

e,

we must
(.1)

but to make

we must
or

take

\Sm (l)-S,n (x)\<$ l-,vm <$.

^ m >]--^,
inequalities (A)

................................. (B)

and the two


that

and (B) are mutually contradictory (supposing


in the general

<1). Consequently the two steps used


;

aimiment are incom-

patible here

and the reason

for this difference lies in the fact that the

inequality (A) does not lead to a determination of when x can approach as near to 1 as we please.

independent of

The assumption

that

the series converges uniformly enables us to avoid the difficulty involved in such a condition as (A).

(2)

//

the series

F(x) = 'Zfn (x)

i#

interval

in

the

fl/f fu m-titm* /(./) (a, b), and if each of we write interval, may

uniformly nmrergent in (/' continuous


/'x

45
1

ENTEOBATK
K<-r,

117

in
id
///

virtu.-

ol'

tin-

uniform

c>n\ .-r^-nce

of 2/n(#),

we

can

til

BO that
if

l/(*)+/. + ,(.'-.)+...-f/;jr) <*,


however
vndept
i"
I-

p>m
<>f

-mall
at

may

!>

ami

tin-

value

will

be

of x, as before.
li

Hence We

ami since
that

this

is

true,

no matter how lar^e


,

/>

may

be,

we

see

OOnV<

:il

that

i| m
At
th.-

=e'--

J'i

saim- timr \\v have


|ji(

).L()|;

ami oonaeqnently
l
I

J
i^

c, e,

m (a>)da> F()da>-\"s J
c, Cl

last

inequality can be written


':

so that, coiiihinini: the

two

ine.|U;.lities,

we

find

\vhTf
Since
inent

fc-

may
is

In-

a^ small as \\c plra^*-.


<>n

l>y

]iri|nT choice
>ile, the

"!'

//-.

///

no l<m-vr juvsmt
in

the let't-haml

-i\-en

N,
"

p,

:,.

shews that

Jc

This

o].el-;iti..ll

ix

.t'tell

lrsf|'il

'

as

/,/',/,-/,//-/,///

\vill
,,f

|,i-,,l,;il,ly
.

tin.l

!.->> Ml

ilitVu-Mlty
!

here

in

rea!
.!"
./'.

tilt-

onditioi)

tlial

>houl(l

indrprlldrlil
.f

It

n>(,

hnwrviT.

r;isy
in

to

i-onvi-rirt-nt

>erirs

give a really simple example \\hi.-li irrni-l>\ -term inte^rat i.-n

n<>M-uir.:
!

l-.ul>

nroneous

n-sults.

Tin- f'-llMU-ini:.

ahh

\<A

as good as any.

118
Let
so that

SERIES OF VARIABLE TERMS.


Sn (x) = nxe-*\ F(x) = lim Sn (x) = 0,
n
*co

[CH. VII.

if

x > 0,

and obviously

S n (0)=0,

so that F(0) = 0.

Hence
But on the other hand

so that

2^

fn (.<v)dx= lim w >


I

(1 -<?-')

which

is

obviously not equal to

F(x)dx.

= and gives some figure shews the non-uniform convergence at .iindication as to the reason why the area under y = Sn (x) does not tend
The
to zero.

FIG. 15.

Of course the argument above assumes that the range of the conditions under which an infinite is finite to oo say, belong more properly series can be integrated from to the Integral Calculus; but some special cases are given in
integration
; ,

Art. 176 of the Appendix.


46. Differentiation of

an

infinite series.

If

we

consider Ex. 4 of Art. 43, for which

we

see that

z->0

z-j

Thus

lim
n->co

*Sf

M'(0)== oo,

although F () =
/
J

It follows that the equation

''

[lim
/(

,(*)]- lim
-i

Sn'W

*.

45,46]

I'll!

II!)

uln-ii y tru.!'.

!!<>n-iinii'<.rin

,-,

nee pres
i:

Hut

it

-hniiM
the

IM-

untie. -<1

that

it

is

,im

n.n-.
i'ailurr.

of

\\hirh
t'n.in

is

the era*

as will

he apparent

tin-

^-n.
tin-

h,

-..

Th<-

rradcr

may

nuisidi-r similarly
:.H
,/./-..

Pi

hri-

Bfl
fi,,-

unitnrmK
,

\/(#) oscillates.
\

I/

sert<
,,t

rential
<///
'

r*mly
I
,

///'////'//

(,/.

/,

,y//,//

/,,

F'(x),
/"'

of
/-

/''

///////

//^'

/'//

f/

',

,,t,

ri-<if*

Write
">.

\v.-

have
i

'I'hus.

ly

tin-

t'uii.lani'-ntal
/

ju'ujn-rly

<t'

an

iiit-_;ral.

\\-

hav-

F'(x)=G(x)
'

("

.'-i

jirnoi'

ol'

tin-

l'(prr-(.in^

ihcoi-ciii

BOme usr
i1

nf

thr Intr^ral Calculus;


tlms,J.PM.I'S

hut

is IK rasy witlnmt the following infthn.!


it
-

easier

than

which
on,-

depend
ivsult

<.n
is

tin-

)itlrn-ntial
_:!-alitu.
in
IM-

Calculus.

an<l

only

Ml,tain-d

ly

chosen so as to

make

()
Thus
|>r..\id.-d
I

<,

U^X;

\^
i

K(x)d*
fco

<((
th.-

that

<.,

,-.,

hrl,,,,-

int.-rval

(.

/>

>.

This

^
and
sn

\^
i/i

iufrr the

this

wou.

An

al(Hti<>nal

assumption

\\lurli

\\

.).

n..t

I]

120

SERIES OF VARIABLE TERMS.

[CH. VI L

Or, changing the notation,

we have
-

?x {/ to
-

^1

-/(*)}

'^e,
(a, 6).

provided that both x and x + h belong to the interval

Now we
^

have

4
=x,.
t

+ Fro, say.
that

We

have

proved

=^

and

in

virtue

of

our

original choice of m,

But we have
h

identically

_
and so we
^(05

J J

find

+ ^-^(05)
77i

-Q(x)
/ij

Since

has been fixed without reference to

we can allow

the right-hand side of the last inequality then approaches the limit 2e, because each term under the summation sign tends to zero. Consequently,
the

to tend to zero, without changing

m;

maximum
;

limiting value of the left-hand side


this

is

not greater
limit

than 2e

and since e is arbitrarily small, must be zero (see Note (6), p. 5).

maximum

Thus we have
Jim

F(x+h)-F(x)

-G(x) = 0,

or

which
47.

in

the required theorem.

important to bear in mind that the condition of is merely Niijlirirnf for the truth of tlu theorems in Arts. 45, 46 but it is by no means a necessary condition. In other words, this condition is too narrow but in spite of this, no other condition of equal simplieity has been
It is

uniform

com rr^vnr.-

46,47,48]

PRODUCTS,
ami
ra
\\v

discover
Mii
:

shall

not

t'urtln-r

into

tin-

That
tilt\Vii

unit''.

mi
,.\;i||,|.

not

D6CC

following

(i)

hat
<lis.-,,iitinuity.

Don-uniform convergence

do*

imply

-f

,-'

.,-+...

,1

+ .r),

(0

<.<!).

:ind

\<^-2

is

also equal to the SIM

i-t+J-J+found hy integrating tenn-Ky-t'-mi. Nevertheless r=l is a point of non-uniform ...UMT^-IH-O of the series
in
./
;

l>e-au-f
to a
is

tin-

ivmaindi-!
of
/,,

:.l
wlii.-li

the condition ^jr"<


of
./

l.M.ls
1

(It-t.-nnination
in
tin-

OO**Oi

ind.-jx.nd.-nt

(v.

indudnl

intt-rval

48.

Uniform convergence of an
l'tinition
oi'

infinite product.
at
tin-

Tinone.'
I.,

unii'orm
])rnlucl
:

convergence can be extonl-l


l>ut

an

infinit.-

,M])]lic;itins

jrim-ijl.}>r<

occur

Irss I'rtMjuriitly in
tin-

rlcinrntary analysis, and for our


In-

purpos.-

following theorem will


values
<>/

sntlicient

If far
>)

"//

im

'!>'

interval (0,6) the fwn*


/,,,<

has th /"/>< rty \fn (x)\^M,,. slant (independent <>f x), tl >/ *!>
flu

isapoeMix

>

aeries

-M
i

is

convenj

i>r<xl

art

/v-)=fi +./;,(y)][i + /;(.')]! 1+u.i-)]... continuous fumctio^ of x im th* >l. continuous in the fun
/

l" lf

' !

For writ-

I
[

-K, (,0][

1+ /-,(./)]. ..[H-/
I

Ml

..,(r)]

LI

+/(')][
'
i

]...

=P tooo-<

B1

then

ii:

<

><n
t

I-/-

where

^Hil-l/J}; fl;i+/, <1 + !/.+,(*) +


'

-/' (I-/to

'

/,'

...

Thus
|

Q.(x)-l|< RJ(l-R m ).
I..HM-I
P.I
; :

in.ulr U>
s.-1-ii-s
'J.

bg
_
:

tin..
:

author (JVofc
-'

/. ..... /.

.V'C

vol.

1.

!'.(! I.

|..

IS7]

;.

t-nn-l>y-t*Tin
i

int.-uiation
in

li.i\,-

l...n

jivBO
IT-''.

bj

thr Apjx nli\. Ai

tt,

122

SERIES OF VARIABLE TERMS.

[CH. VII.
,

Now
where
all

R m < Mm + Mm+l + Mm+ + ...


,

to oo

Hence ?/i can be are independent of #. chosen (independently of x) to make 72 TO <-J-e, for all values of # within the interval (a, b).
the
JJ/'s

Hence, assuming

e to

be

less

than

we have
ifa^o^fc,

\Q m (x)-l

<e/(4-e)<ie,

where
of x.

(as already explained)

is

chosen quite independently


interval,

Thus, if c is the inequalities

any value

of

x within the

we have

and

i-i^Q^
Q(x) ^e l+^eQ
m (c)

Or

1-fr P..(aO^P(s)
'

Now,

since

is

fixed and independent .of x, and since

are continuous functions of

x,

so also

is

the product

/',(,'):

thus

we can

find 3 so that
i

.,

< -P-() <

*
,

provided that \x

c\<^8.
T
T

*IMM

"^ n/^v "\

71

\9

'

and so*
provided that
|a?

c|<[<5.

Hence
b).
//

P()

is

a continuous function

of x within the interval (a,

If the function /(;/)

///*

derivate, snch

In

it
\

fn '(x)

< M,,,

<md

if

*For

(l

so that

(l-)(H-iO<(l-iO s

48,

49
1

\\.\l.

l;\

>

III

///,/,
,i

//'

d
Pi
tieee

-|

i-"iiditi<iis
1

ire

''

<-

'

!+./
v,,

ihai

An.

h;

ran

!..

a|.].li.-,l;

and

\v.-

tin<l.

ac -,,i.lin-ly.
(

1
49.
i.s.

!elv connected with

tin-

theory of unit'iTn,
.jiu-iit

tin- I'dllou ini:

theorem 4 whirl
"

u-- in sob

OpO84

ilnit

mi

in

\^vl (n)+vt (n)^...^vp

and
/

tlntt

we wcwt
7

to

fond
./;/_//

fl- limit
/,,

lim
it
j

l-\,<\.

ii

/////

iiijin

with

/*.

'/'A',

limvr(n)wf
////

(r lu-in^ fixed),

//////

lini

/*//

/'(//)=

*/

+ <'-+ ^''o-f ...


<'/"/'
.U,

to 00

a= IT.

vided
///r

//"'/

',(/')

.I/,.

independent of
th.-nn-m
\\'.

00TM8 l.U,
reader
will

At
note
that
;

'rh.-

for
.l/-t-st

th<-

is

Milistantially
(Art.

th-

sainis

bhe

due
sain.-.

t>

H
i.

).

'I'll''

irMl'. tOO,

alnmst thr

Kirst
<!'
//

choose a nuinlnT
Mich that

'/

(which

.f

DOOT86

is

iii.l.-]--inl,-nt

M
and
h-t
,i

+ ...

to

he taken
.-

lari;.-

r rm.u_u h to

x <e, make />></:


.;.

then

+ 'Vfi-f -.. +

'|.

=^/ + J/,
1/

+ ...

to

x<e

e ot

in tlu-

tot of Art,

1.

124
Also

SERIES OF VARIABLE TERMS.


\w q + w q+l + w q+2 +... to
<z>\

[CH. VII,

= M + Mq+l +...
q oo)|
1

to

oo

<

Thus

\F(n)-(wQ + w l + w 2 +...

to

<|(% + ^i+--. + ^-i)--( w


/

+w

+...

+ wq-i)\ + 2e

>

to be remembered that so far n has only been by the condition p^>q. Now, since q is fixed and independent of n, we can allow n to tend to infinity in the last inequality, and then we find

and

it

is

restricted

because

lirn t;r (?i) n><x,

= wr

Hence, since

e is arbitrarily small,

we

find, as

on

p. 120,

or

HmFn=W=W+w

+ w+...

to oo

The following example will serve to shew the danger of trying to use the foregoing theorem when the J/-test does not apply. Consider the sum

so that

= tv()=logf 1 + ^J and p n.
wr = }imv r (n) =
n
*oo
t

Then obviously
and
so the

sum

of the series

w + 20+w + ...

is 0.

But
so that

vr (n) lies

between r/n2 and rj(n^+n\ and

2> =

and hence

lim F(n) = \.

Another theorem of importance


analogous result for products Suppose that
:

in

this

connexion

is

the

where

Then if \imv r (n) = wr


n->oo

tends steadily to infinity with n.


,

and

if

\vr

(n)\^Mr where
we have
...

is

independent of

n and I,Mr

is convergent,

the
.

equation

to oo

49]

MI'LES.
iv.-ul.-r
<>f

Tin-

sl,,,uM

liavr

littl<li:

ditlicnlt

ii

proof
ill.-

this tln-.TiMii
(,!'

on

tin-

be
limits
l

foregoing,
I'm-

'-inj.!'.;

ivsnlts

Art-.
I

bo

timl

tin]

t-V<l4
I

Mini
in

+"V< +

"'.,.,)...
(

to 00

terms Of
b6W the

tin-

n-ni;mil.-r
Die

J/

+ J/,. + ...

to oo

condition >n<-h as

tin- .l/-t-

tllf

X.-MMjilr

in

which
that
I'.Ml

j-,

= /, = ... = !/,
...=0.

tllf

,-,,M;itinl|

IJui

)'

=1
<.f

niM-i-ssarily
/<),

true.

In

fa<-t

tin-

valm-

the limit

value of liiu(yy

because

<
1

.,

/'

I.X \M!'I.I>.
1.

Shi-w

that

if

^.(i) = .i"
it>

(\

+.'"),
1
>

r=l
1

is

j)int

of

noii-u;

OOnvei
2.

limit.

raw graphs
is
t

of

&(.) and

lim >

Sh-\v that

tlir

,)=^-.'ii

uniformly o.iivrn:.
i

all

Ity

t.-ini-l.y-t.

in

dnt'.'i .-nt iat i.-n.

If /'(.) is nr\. 3. i-ontinunus fuiirtioii f

.,-

iii

th- interval
interval,

(",

/o.

and
-

if

r in tin-

same

shew
w.-

that tin-

uniformly
i

in

th- int.-r\al.
|

iMxi.]

Lfl

any

jM.int

of the

int,-ival.

can timl

//?

siu-h

0^
l-'uithd.
:

(C)<j.
continuous,
>.,(.r)-5

we can

find &

\K
pi-vidcd that

r-c| <8.

Senoe
at

= F(.r)-Sm (.r)<
.f

all

]>>ints

the inteival (c-8, C + S).

12()

SERIES OF VARIABLE TERMS.


since

[CIL

Now,
Thus

Sn (x)

never decreases (as n increases), we have


(.*)

^Sm

(.r),

if if

F(x)-S

n (.v)<,

n>m. n>m,
;

at all points of the interval (c-8, c + S). Consequently there can be no point of non-uniform convergence in the neighbourhood of c and therefore

there

no such point in the whole interval (a, b). The reader should observe that this argument fails in cases such as those illustrated in Figs. 14, 15, because then Sn (x) may be further from F(x) than Sm(x) is. In the cases considered here, no two of the approximation
is

curves can intersect.]


4.

Shew

that the series

is

continuous for

all

values of

.i\

and deduce from Ex. 3 that it converges uniformly. = is the only possible point of discontinuity, [It is easy to prove that ^ by means of the J/-test. Now if we take v as equal to the greatest integer
contained in
since
2

l/.#

we have
r

<.r(l +log
filld

f),

2<l/#v,
v +i

if

0<#<1. Hence

=< r'
Shew
that f(x) = 2^ L
;

we
is

the limit of which as x-*0


5.

-~p
//"
j

zero, so that the series is

continuous at #=0.]
all

for % 2 ~f~ 71 r~x converges uniformly


i?/'"*

values

of

examine whether f(0) can be found by term-by-term


1

differentiation.

6.

Shew

that

2-1
n\
for

l+x*a 2n
all

_^_

converge uniformly

values of

and that

if

<1

and

they are respectively equal to the series


e~ a
x.

and

obtained by expanding each fraction in powers of


7.

[PRINGSHEIM.]

If

/(*)=*''(! -.r);

then

we have

but
8.

2/n (#)=#/(l -.r2 ), 2/', (l) = 0, although


1

if

Shew
it

that the series

2 l/(w + ^)
J/,,

converges uniformly
.

if

.v^O; but
[OsoooD.]
I'.l/,,

that
9.

cannot be integrated from


t> If \fn (x)\<.Mn lx ,

to oc
is
;i

where
if

positive mnstant such th,U

converges and jo>l,

then,

a>0,

Apply
to Ex.
8.

this to the series

2 !/(+#)*

and shew that

it

does not apply

[OsaooD.]

L27
10.
If
1'",,

oscillates

fmit.-U

!!

the series 2(ajn*)

is

[IMKK MLKT.]
11.

.,,,..., ,,

|.J.
I

Km-

.'-"

12.

It'

Itely

00]
;u id

illts.

>ll-

11(1+

erges absolutely
t

uniformly

in
1
'

int--r\al.
1
i

fcbsolntely)

;md

1",,- '-ii\ orgec,

uniformly
13.

(l>ut

my
tin-

tinili-

intei val.

Shew

that

produ.-ts
D".'-/'J,

II[l+(OOQVerge unifm-mly

II|.l+(-l.
linitc

)] t

II.-..>.

in

any

int.-rval,

and

tliat

tin-

tliinl

f"ii\erges

1
-o

'han a
\

lix.-tl

nuinlM-r

at

;ill

j).in;
1

and

for

all

ii-n
001

it'

-s at

;ill

poin

int.-rval

(./.

/),

it

[For, divide

tli-

interval

,nifniily. into \- snl.-intfrvals

[HKM-IX>N.]
,-sich
>f

Imu'th

win -re

6<^,
'hat
at
tip.-

lriii'4

any assi^iu-d small


ra--li

jM.sitivc

nunil-r.

N-xt

tinJ

t-nds of

suit-interval

<t>(>>'r)="'fn(-'-r\
,,,-t-l

(^-1,

2,

^ ...)
-fl).

is

numerically

less

than

o.
./////

Thi//';/

i-

possil.le

le.-;m>e th

h of thr-M- points,

Now

if
-

any point

of tin- interval the nt-aiest


A/
;

end of

a sul>-intt-rval

not

further distant than

hem-.-

\<P(*)\<

and so the
15.

te>t

|<K*)|<I<M.V)| + S<2S<, of uniform


<

'led.]

Apply li'ndi\*Mifs
^-

ti-st

to th.-

16.

It

-'
tin-

OODVergM M
'

hi.-h

t.-nls steadily to

with
their

n,
it,

//
,

OOHYergM
:
<

unit'..rni!y

if

'-.^0.

Deduce
that the
.'
;

'

in

-.!!, Tal,

>oiue nuinl.er ;: sii-h thai


if
i

>(*
is

and

do,-s

not

OODVergQ

<>f
.

OOOFM

it

|M>ssil>le

l.i

nil

val

values of
-

exai.

n.

[CA

17.

Shew

that

CHAPTER
POWER
50.

VIII.

SERIES.
is

The power-series 2a wo; n


of

one of the most important


10, that if

types

uniformly convergent

series.

We

recall the result-

proved in Art.
Inn >w

the power-series converges absolutely


series

when x

<
!

I /I

but the

cannot converge

if

x
\

>

I /I, for

then lim an

xn

>

and

so there will be an infinity of terms in the series whose absolute values are greater than 1.

Thus any power-series has an interval ( 1/, +!/) within which it converges absolutely, and outside which convergence is impossible. By writing x in place of Ix, we can reduce this interval to the special one ( 1, +1): and we shall suppose this done in what follows (we exclude for the moment the
cases
I

or

oo

).

Thus suppose that we have a power-series which is absolutely 1 and + 1 so that if convergent for values of x between n / is 1 number between and the series any ^|^,,|& is con:

vergent.
series

Then, by Weierstrass's

Jf-test,

it

is

clear
(

that the
k,

2a n#n converges uniformly


|

in
| |

the interval
.

+/>),

because in that interval


n -suit that

a nxn

a power

series

Hence we have the converges uniformly in an intercut


|
.

an kn

which falls entirely within its interval of absolute converge It sometimes happens lint further tests (such as those given in Art. 11) shew that the scries is .il>solm<-ly convergent for = l; and then the interval of uniform <-<>n\ i-r^viice extends |oj|

50]
:

vi

it

U)

-f

:npare
\\

tin-

ith

ami apply

,in.
tli,;'
.

But
\rt.

may

al>

happ.-n
nt:
in

not

this

ca -

m
variahle
r

11),

lerau*e

tli.-

leqUi
U

apply Abel's test factors X n D<

inn-eases with //.ami

is

than

(if

O^u

1
I

i.

Conn
2a,,a;
1

rmly
nei-tl
IP
i

in

an interval which
1

includes
it'

./

^
\

(hut

as far as ./=

).

Similarly

~(

nt ili- interval nf

uniform C0nv<
;

1.

Ex.

1.

The
<>

ft

uniformly
l"t\v-cii

in

thetlit-

int-rval
1,

(-,
-4-1

+/)? where
.1,,

is

any

niunl.-i-

and

lnit

jH.ints

n,,t

l-l.nig to the region of

uniform

Ex.

2.

Tinin

+...

i-.ifi's

uniformly

the interval (-1, +1).

Ex.

3.

The

ft

8
<-"ii\erL. es
f

-i

uniformly
1

in

the interval
1

(-,

li.-tv.

luit

the p-'int -

il>es n>t lifhuig to

+1). whore / is any urn the region of uniform

\\V

HM\\
k-t'l

ivturn
.n

to

th.-

cases
It'

()

or

oo

which wv havr

hitherto

one

side.

it

happen^ that

O,
tin-

"=0

seriea

1"

SB"

will
>i'

and
(-.1.

the

interval

converge ai.sMhurly tor any value of x\ uniform convi may be taken as


can
)>

-f-^1).

Where

.1

arhitrai-ily

la:

Tims the

si-i

1-f

.y foi-1,

any value

of

* ami

is

uniformly omver^ent

ii

+.1

).

On

the other luunl.it'


t'or

lini
\

a n |" =

the

Sei

"

eannot

any value of x other


of
tli

tha.

An example

led 1-y

i:

130

POWEE
is

SERIES.

[CH. VIII.

There

one important distinction between the intervals of

absolute and of uniform convergence; the interval of uniform convergence must include its end-points, but the interval of

convergence need not. terminology, the former interval


absolute

Or,
is

to

use

convenient
latter

closed',

the

may

be unclosed.

That the interval of absolute convergence of a power-series is evident from Ex. 1 above, in which the series is absolutely convergent for any value of x numerically less than 1, but the series diverges for x = \ and oscillates for 1 x= On the other hand, Ex. 2 gives an illustration of a
need not be closed
.

closed interval of absolute convergence. But we proved (at the end of Art. 43) that the interval of

uniform convergence must be closed, whenever the function

Sn(x) is a continuous function of x. Now for 2aw # n we have Sn (x) = a + a :x + a^x? + -f anxn
,
. .

a power-series

which is obviously continuous for all values of x. Consequently the interval of uniform convergence of a power-series is certainly
closed.
p.

This fact
top,

is

129,

or Art.

not deducible from Abel's theorem (see 51), for it does not appear impossible
at

a priori that a power-series might diverge 1. be uniformly convergent for x


|

x=l

<

and yet

51. Abel's

theorem

is

expressed by the equation

Km
provided that 2a n is convergent; this of course follows from the fact (pointed out in Art. 50) that x = l belongs to the
region of uniform convergence and from the theorem (1) in
Art. 45.

Abel

also

shewed that when 2a M diverges, say


1.
;

tends to +oc as x approaches


appeal to uniform convergence theorems.

to + oo then 2a n # w also This theorem cannot be proved by any but the following method applies to both
,

Write

A =a
since
1,

A^ciQ+a^
,

...,

J,,

rt

+ a 2 +... +

'

Then

x,

#2

...

is

a decreasing sequence,

we have by

the second

form of Abel's

Lemma

(Art. 23)

50, 51

ABEL'fl THfiOBl
//,
//

\\hnr
//
,
'

ar-

tin-

IIJIJMT

ami

1" wt-r
.

limit-

.1,,,

.1,

...... m _i, and


l

are those
ill,-,-

Sin..-.-

limit- an:

ind-|M-nIi-nt

<!'

y,

we have
i
'

(1)

-O-H /'.-''" =Sa^^ //


I

//,,*".

*MWI
I

*,

then

we can

h:il

however small

may

be.

ami
the
indi-x
oi
///

di

]..

-ntls

only on the series

2# n and

is

therefore

independent

Thus

\vt-

Me frin

(1),

that

Hut

is

arbitrarily small,
valu.

and so

(see

Note

(6), p.

5) both th.lal

maximum

ami minimum limiting

to

s.

Hence

lim (l'/,,r M ) =
x-*l
.

=we
can
timl
>//

say

to

oo,

s<>

that

//. n

h" \\.-v.-r

lar.u't-

.V

may !.

Thus

so that from (1)

lim
x
*!

I'./,, ./"

_:

.V.

Thus,

l.y

an aixumrnt similar t> that

ust-d

alx.ve,

we prove

that

/-'////,

// So,, otcillatesi

we can

see

by the same argument

that

lim
c.uin.M-trd

J n |fjniafrr n
tin-

ly

with

f..icur

inur

results

is

the theorem of comtl.-

parison for

two divergent
that
I'./,,,
I'/-,

series.

Suppose
ility

are

t\\"

ilm-r^-m

seiief,

ive

an

similar t>

(1),

\\hnv

/,

A'

ami
/..

from

Z?,,

=^

-i-^i

+ -" + ^.

IM(>

way

as

//

and
//,

Thus, provided that


(2)

/>,

//v

yc<

tind

".

v=
it

.")+*-

-"
is

In

applications of the theorem

is

generally the case that a*

always

132
.

POWEK
lim
Xf-l
</>

SERIES.
X>I
3> (x)

[CH. VIII.

Obviously

(# ) = km jHm ,

and lim

= Km/h m
I

and now suppose that

Bn /A n approaches a limit then possible to choose m, so that


may
lim X>I
be.
<f>

as

n increases

it

is

however small

Then we have

at once
<I?

(.r)

^l
if

and lim
X --l

(#)

Hence from

(2)

we

have,

/(^)
X-+1

X-+\

Repeating the former argument, we find that


!

In a similar way we can prove that


that
if

if

Bn /A n-*&,

Bn IA n

oscillates,

f(x)

may
.

oscillate

then f(x) -+cc and between limits which cannot


;

be wider than those of B n /A n Thus we have the theorem, due to Cesaro + b n ) / ( + !+...+ a M ) approaches a If BJA n = (6 + 6 X +
:

. . .

definite

limit,

Jmite or

infinite,

then
lira [(26 B ay)/(2o H ac^l
)]

= lim Bn /A n
>oc

a result which can be obtained also from Art. 153 of the Appendix. It should be noticed that if b n /a n approaches a limit, Bn/A n approaches the same limit (Appendix, Art. 152) thus a particular case of the theorem is
;
:

If b n /a n approaches

definite limit, finite or infinite, then

lim (bn /a n ).

Ex.

1.

It

is

possible to obtain the first

form

of Abel's

theorem from

the last theorem, by comparing the two series

Ex.

2.

Similarly,

by comparing the
.J
2

series

4;r +
we
see that
if

+ .4 + .4# + ... and +A


x >!
l

lim -(A

then

lim(2,,.r

w
)

=
is

Z.

[FROBENIUS.]

Ex. 3. Again, if the limit in Ex. 2 a further mean. Suppose that


,.

not definite,

we may

consider

n
the series

tlien

we can compare
y|

+ (2^ + ^

)^+(3/l

+^J
lim

-i-.l

i)^-f-...

and
[C'ompare Art. 123.]

and prove that


\\'n.
.t<-

(!'</,.../")

=
2,

tliat eacli

of tin- <-.\ainik's

1,

3 includes the preceding one.

51, 52]

CONTIM
4.

ITY.

Ex.

AH

tin-

n-a<ler

in,.

'hat

<'

+ *'+...) =

(ii)
'.r

+ 3'-V+...)=I
.

*!
'

'-Kr' -.!>

+ ...)=

In rase

(i),

the

sei

14.

^4.... gi vtZfH

while the series for (l-.r)~* gives


'ml
.l
f|

= 3.
\\hile
tin-

1)

-2

<)...2.

cx>logtt/log,
tli;c

log(l-x)
-f 1).

M0

(iii)

\ve

DM

th- fa-t

!'

Kinally, in case (iv)


1

we

li:ive
liav.-

J,,+

.-!,

+ ... -f

.l

rl

roA.
great generality on quote th- f-ll-.v.

1'rin.u'slHMui*

pi-..\,-l

the,, reins of
ire

\vhieh diverge at
It'

r=l.
of

A> an
A(./-) .*]

examj.],-

A(

i.-

a
r,

funetiin

sl,,wly

than

that
(1

liin
[

<.
<>f

th.-n
./

2A'(n)
1.

approximately ly A[l
52.
'I'ln-

-.r)] for values

neaf t<

Properties of a power-series.
LT'-ii'-i'al

thcoiviii^

proved
\v.-

in Arts. 43,

40 of course apply
i'ullnwin--

to a ptwvr-srrii^.
iiu-nts
:

BO

that

can inakr thr

state-

(1)
,

power-series
'/

2an#* is u
/'/>

<;,,ili

n>iu.-< fu

<>

any

,,f,

rval

"/////'//

region

/
f/

<

<Val ii'itliin

r>

!'",>

<>f

\
-

*(Saj> xfc-2

+1
')"
(<-;

-'-r

+t+I
).

//'

is

any
!/,

uithin

fl

reg
:

noe

\\'r

nutf

that

tin-

interval
alt.'ivd
i'act

<t'

altsolut,-

convei
>

>f

p^
80 that

nit
I'rnin

l,y
;

ditrcrcntiatiun
lini
//
1

thr

that

1,

lit,

,Mf
i(

"

= lini 1^1" = lim


-.

M,
will he found.

f:

linn.

p.

i.

\vi,,:i-

full

references

>r

lim(n +!)/!=

1.

so that liui/c

1-y

Art.

l.Vt in

.i\.

134

POWER

SERIES.

[CH. VIII.

By applying we see that in


of
the

Abel's theorem (Art. 51) to the integrated series (2) the point c z may be taken at the boundary

interval

of absolute

convergence, provided

that

the

integrated series converges there, series does so or not.

no matter whether

the original

An
(4)

example of

this has occurred already in Art. 47.

If a power-series f(x) =
(

^a x
n
o
t

converges

within

an

interval

k,

-f-/c),

there is

an

has no root except, perhaps, x = 0.


For suppose that a m then the series
is
is

interval within which f(x) =


f(x) which does not vanish

the

first coefficient in

a continuous function of x, in virtue of (1) above, and fm (x) = Q for #=0. find c, so that \fm (x} is less than ^ a m if x c k consequently in the interval ( c, c),

Thus we can

< <

Hence /(#) =
and
if
(

has no root other than # =

in the interval (-c,

c)

is
c,

different
c).

from

zero (so that ra

= 0),

/(#)

has no root in the

interval

(5) It is
series f(x)

an immediate deduction from


oo oo

(4) that:

If two power-

= ^anxn

g(x)

n ^b nx are both convergent in the

interval ( k, +/c), and if, however small 8 find a non-zero value x l in the interval ( 8,

may
S),

be,

we can
satisfies

which

then

a =b
the

= &!,

and
f(x) that

two

series are identical.

In practice,
is
|

we hardly

known x |< &

ever need this theorem except wlu.-n be to equal to g(x} for all values of x, such

have hitherto discussed the continuity of the po\\< Tfrom the point of view of the variable x\ but it sometimes happens that we wish to discuss a series 2/n (y) xn
53.

We

series

regarded

as

function of

the variable

y.

The following

theorem (due to Pringsheim) throws some light on this question:*


*

1006, p.

Further results have been established by Haitn^s (Math. Annalen, 9), Using more elalmiatr aiiaK

K<1.

CrJ,

52, 53)

PEINGSHEIM'e THEOREM,
\'

caTi be

found

8u</< ////

trl,f>re

A,

are fixed

and

'//'/

?i /ias
',

any value.
interval
(

Then
<

2/,<
th.it

>ntlnuou8 fuiirfiun
la
./

if

,i

f/tc

continuous for all finite mines ofn, and

<

A'.

To prov
with -/l?i
;>

tli-

theorem,
,

we
t

in-.-.l

<.nly

compare the
ami
i>

series

(-pJ-J

which

is ino!

-p.-ii<l.-nt

of y

OQQVefg

when |05|<A': tlms the s.-rics ^/' (^).a5* (by cm vi-r^-s uniformly with iv^unl to // in
(1
i

Weierstrass's
thof
//

int.-rval

'),

ainl

is

tlu-r.-1'ore

a continuous

function

in

that

Interval.
It

was erroneously ^uppo^nl


i

li\l>)

Aln-1

that

1/

'/

A'

in

tin-

continuity
tinuous).

of

intn-val (a, n ^/ ,(iy)a; for


(

was
./

sullici-nt

the convergence of to ciivun- th-

< <A

But

rriii^shrini
is

lias cotistruct-<|

(a-<umin^ /',(//) an .-xanipl.- s


(5)

con-

that this c.nlitimi

not sutlicient (see

Example

The

follnwitiLT
>

rxainpk-s are due to AK-1,


l'jc+-2 '.c-+...
!

\\itli

the i-xccptimi of (5):

(1) Tlu-

+ ,<\c" + ...

(\jc

<1)

represents a continuous function of y.

ffiiny+i
MtiiiMini.-

'->in ^// -.',' sin 3//


./

+ ...
tinst-rit-s

\vlii-n
\,
it

<1

Imt altln-uirh
2?r,

still

COnvergea

it"

i-

ili>

-.

ntiiiu.us at

v=0,

4;r, ...

(see Art. 65).

miniums

fniH-ti.>ii

<.f

>i

if

X\<\\ and
+
'

thus

liin

/'(y)

= U.

Hut

if

'

the 9eries

+
/

+
."'

(see

Art

u)

from
tiian
tlutii>t

"'-'

tan"

//

t.-nn y(l-f//-).

Thus

it

th.it

(,

I)

Thr

...nv.TLrt-iu-c

of tlu-

sn

Spim

136

POWER

SERIES.
w

[CH. VIII.

does not follow from that of


series
8in.y

2/M (#)#
y

for all values of


,

y>0.

Thus the

Bin2y

sin2y-

n2^
y
1

y
converges
diverges
if
if

#<1, when y>0; but


Example
:

the series

x>\.
J/'n+1 /J/ B

(5) Pringsheim's

Let Jfn tend steadily to oo with n in such a way that Km and let J/ = 0. [For example J/" = 0, J/ =wn .] Then write
7l

and
of

it

is

evident that the series


for

y and

any value*
real
,y

of x.

2/n (y)#2n converges for all real values Further, the functions / /j, ... are con,

tinuous for

all

values of

?/.

But

if

#^1,

the

series

2/n (y)^r2n

is

discontinuous at

For

and

so

if

|y|>0,
o

Now

the terms

rt

(y) are positive, so that

^i
From
these facts
if
it is

if

^1
is

clear that the series


is

Of course

\x

positive but less

<J, the series than 1, and so

continuous at

discontinuous at # = 0, if x^\. = 0, because /,(y) is;/

and thus the Weierstrass-test


54.

applies.

Multiplication and division of power-series.


power-series, the
results

of Art. 34

As regards multiplication of two shew that if both series

converge absolutely in the interval! is given by 9

/,

+/), their product

* Because
ami

^x2n /M n

we have taken y not


and
UK-

Of course converges for any value of x, since lim Afn+1 / J/ M = ao to be zero; if y = 0, all the terms of the series are zero,
.

S/M (0).x2 = 0.
two
/'

f If the
-///'///

series

have different regions of convergence,

this

iuUTval

\\ill

be

of tin- two.

53,54]

MUI/ni'l.K \||M\
al.-olut<-ly
Co
ill

.\Nh I.IVI>ION.
int.-r\al.
\\
:

which OOnTflVgei

the same

H'/V
~>\)

If

\v-

apply Abel'd theorem (An.

to the equate

**-(So,

we can deduce
pn>\ id-d that
ditr.-ivnt
</;/;>;<;,/,

at

oner
tin---"-

hiBeriefi

tln-nivin

(Art.

:{-!

that

CAE,
term
1.
i>

all

convei
tirst

we

assuni,-

that

tin-

constant
it

from zero; and for simplicity


tir-t

\v- tak.-

08

Thus

>nsilrr

where

//

Now
and
liy

Art.

.')U.

th

may

1..-

arran-v,l in

pw.-rs

<>:

pi-i>vil-l

that

In-ill^',,.'-",

any and

nuiiiln-r
th.-

Irss

than

th.-

radius of coir.
h.'iv

16

of

upper

limit of \b n \p n (of cour>-

>=!.

\\'-

ohtain

<!+//)-

1=1-^., + ,^ -:_/,

,.,-:

_,/,,.-:_ O/^/,^.

This
in
./:,

seriefl

may

th.-n

l>e

multiplird
-

by any other

j.

and we

ol.tain

p
:

..v

for tlie (jiioti.-nt

^r" r ''-h" ,'-+...)(l-f^ r ''-f^


in.,

.f

th,.

initial
l.c

trim-

in tin- ilriic-minat

tli.-

.lu.'tinit

may
( ,

still

found as
tliat

powei
n..t

Th

= 0,

l^=(l

1'iit

/-_,

is

BWO)

th.-n

w-

h.,

wben

/
/>',

Tlu-n, as

al,.

:u l

'',

Thus

138
In practice
it is

POWER
coefficients

SERIES.

[CH. vin.

often better to use the

method

of undeter-

mined

thus

we

should write

multiply up, and obtain, in virtue of Art. 52

(5),

from which we get successively


^0
'

'

2'

* ' '

'

A
is

more exact determination

of the interval of convergence

given in Art. 84 below.


55.

Reversion of a power-series.
series

Suppose that the

y = a^x + a 2x
required to express x, Let us try to solve
power-series
If
if

+a x
B

-i- ...

converges absolutely in the interval

k,

+ k),

and that

it is

possible, as a power-series in y.

the equation, formally, *by inserting a

2b n y n

is

convergent for any value of y other than zero

(by Art. 36), the resulting series may certainly be re-arranged in powers of y without altering its value, at any rate for some values of y leaving the question of what these values are
;

for subsequent examination,

we
2

have, in a certain interval,


2

y=(
or

A) y + 0,^ = 1,

A+

V)2/

a 1 6 2 4-a 2 6 1 2

+ OA + 2 & A + 3 V)2/ 3 + = 0, a^3 + 2a b b + a^ = 0,


3
z
1

and so Thus

on, in virtue of Art. 52 (5).

we can
6X

determine,

step-by-step,

the

succession

of

coefficients,

= l/a v
0,

b.

= -a

z /<**,

bs

= 2a22/a 6 - aja^,
1

....

It is evident

different

from
case

from these results or the assumed solution


aj

that a t must be supposed


will certainly fail.*

*For the
K.xs.

when

and a a

is

not zero, the reader

may

refer

to

B, 31-33, at the end of the chapter.

54, 55]
\\'e

REVERSION OF
then
1

9E1
loss of generality,

may
Q

without

for the

he writ! equation ran

and

BO,

with a slight
//

change of notation, we can start from

= + ",.''' + 'V
.,-

:1

+....

Then the equations

i'or

l>
{

.-.juations sh.-w tliat \b n \^/3n ,

where

given

])l'o\

id.-d

that
the (Mjuations
l.y

Now
ol)taine(l

for

the

,-J's

are

those

which

wouM

l.e

in.srrtin^ the series

in

the equation

Hut

any positive nuniher less than /. the aeriefl 1 iver-vnt, and BO we can timl* a nuiuher .17 such that
it'

ft

is

/'

fol all

values of

//.

Thus we ran put

_,_
/.'/

this equation

-'i

Of

iiiii.itic.n

i.f

M.

Mfl

below.

140

POWEE

SERIES.

[CH. VIII.

the negative sign being taken for the square-root, because and tj vanish together.

But
where

(p

+ tf-^M+p^X-Mft-n), X = 2M+ p + 2 [M(M+ p)]


*,

Thus we have

and

can be expanded* in a thus, since X>yu, the value of convergent series of powers of rj, provided that 0<i;</x. But
this series is clearly the
if

same

verges absolutely convergent in the interval ( /x, +/x). Consequently the formal solution proves to be a real one, in the sense that
it

0<>7</u.

Now

as 2/3n >7 n which therefore conw is so that finally 26 n i/ j8nS|&|<


;

is

certainly convergent for sufficiently small values of y.

It is perhaps advisable to point out that the interval ( /*, +//-) has not been proved to be the extreme range of convergence of the series 26,,y we only know that the region of convergence is not less than the interval
;

For

instance, with the series


$ X1 ^ = r + 2! + 3!
'

we

could take as the comparison-series

This

is

found to give

X=
and
so

(_^

the method above gives an interval only slightly greater than +). But actually (see Arts. 58, 62 below)
series for

and the
56.

y = e*-l, so that ar = log(l+^), ^ converges absolutely in the interval

1,

+1).

Lagrange's Series. In books on the Differential Calculus, an investigationt is commonly given for the expansion of x in powers of y, when an equation holds of tin.- form

We

fSee

= anticipate here the binomial expansion of Art. 61, for the case v lt. -ntinl 7: Khvanl>, for instance \Villiain.-on, ('d/nt/it*, />//// chap.

Differential Colai'm, dia|). 18.

55, 56)

I.

LOB
anal

\\'-i

process 'gives an

ratioo
1

f-r the coefficient* in

th<-

it

gives no

inf-.i

hi-h
tinl'-d
tli.-ii

an

-\|.i:
-t

expansi>:

'ally u.

jH.^il.K-

m
,-,

in a

nvergent powerpOMible to write the


<

"-"

""

''"""

MIC -"it

tin-

dt..

mis ^grange's prol>h-m


.f

i-

n>w seen
^a,^,
i.il

to be,
in

in

ivality,

t-<|iiivalrnt

tli-

iw.-i >i<.n

the power-series
-)i,-\\>
ti

the

I^agrange's

iii\-r>ti^aii"ii

to the
'

M
in
tin-

nf

[/'(-')]"

"!',

ng

r\|aii.

\v.-

cstalilislj
iiidr<l

tin-

inert-

ur 'n-ral

form

i.f
00

Lagrange's series in
"
-

\vhirh ;H
p'\\\\'.-

in

^_
t

is

another
i

knew

in fa.
|.y|,

(lV"in

Art-.
\\

''><>

and

."."))

that

ntly

small values of \x\ and

we can

wlu-ninti-rval

/,----<,
..f

f,

anil

tlie

otln-r

c..i-Hi.'ii-nts

have

still

to be found.

The

convergence cannot In- f.iuinl. ly flrnit-ntary ni'tli"ls, until the coftli.-ifiits havr IMM-II (l.-trnniiifd.
Mtiaic tin
.'1

we

find

-|(*.;
hividc in.w
liv
r
i/
,

\\1,

uiv \vliolr nuinlier,

and wr get

Suppose both
:

thifl
riu'lit.

i-|iiatiin
is

t<>

IK-

-\pandi-d in asn-iulin^

tln-ii.

on the

thriv

..nly mi.- t.-nn+

containing *

'

and

this

= r.
It

is

n,\v

,-1,-ar

tli

th.-

r,,-Mi.-i-ut

..f

./--

in

the rxpansion of

'.lie

finl of tlu- cha]

\cept for

7i

= r,
1.,

= Jo*-"' +
hut
in

this

t-\i

sen

if

is

loss

than
;

'
>

r,
,.f

term

in

because x

nt

any pow-r

On

tlu-

othfcr h:inl.

if

r, \vr

1,

g+^
= + J5,^
J

142
r

POWER
in ascending
;

SERIES.

[CH. VIII.

g'(x}jy
It

formula, although
is

first

and this is equivalent to Lagrange's powers of x put in the above form by Jacobi.*
if

to be observed that

the equation in #,

is

solved by some algebraic (or other) process, there will additional solutions as well as the series found by reversion.

usually

be

This series

gives the solution which tends to zero with y.

Ex.
of

1.

If

y=x-ax?, and g(x}=x, we have


(x-ax*)r

to

find the

coefficient

x~ l

in the expansion of

=x- r (\-axY r

Thus we get
and
so
if
|

rdr =

x =y + a# 2 + 2a 2y 3 + 5a3?/ 4 +
ay
\

. . .

which converges Of course this

< J.
namely

series gives only one root of the quadratic in x,

Ex.

2.

In like manner,

if

y=x-axm+l we
,

find for one root

Ex.

3.

The reader

will find similarly that

if

y=.r(l+#)
y
4
,

n
,

then

37i37i+l

+ ~-

method of expanding g(x\ we take the followbx ax e in powers of y = xe To expand ing example r (a ~ rb)x r and so the coefficient of x~ is easily seen to be Here g'(x)ly =ax~ e
Ex.
4.

To
:

illustrate the

a(a-rbyThus we have
S* = l

f(r~l)l.
2

+ ay+ a(a-26) 2
,

a(a-36) V+y +
,

. s

.,

which converges if |y|<l/e|6|. In particular, with a = l, 6=-l, ^ = e*, we obtain Eisenstein's solution of the equation log=,y (see Ex. 11, p. 18), in the form of the series

Qes.

Werke, Bd.

6, p.

37.

56, 57

KXroNKNTI
SPECIAL
limit.*

\|.

SKHIER

CI:I;TAI.\
57.

i'<>\\ 1:1;

SERE

The exponential

\\v pn.r.-.-d tu prove that

win-!-.-

05

= liliH.
-*<>

Consider

first

tin- sin-rial
//.

6886 wh.-n

t-nds to intinity th;

intr-Tal values

and

writ.-

n*=*X.
"

Thru we

tiiiil.'-

>n .-xpaiKlii

i-

-3(i-J)...(i-^
Now.
can
u^i-

this

ezpressioi] satisti-s tin- cnmlitions

ni'

Art. 4!>:

as

tin- roinj.arisoii-series,

-,,...

.,..

where
F<
>r

A',,

is

the <nvatrst value} of |X| for any value

ol

we have
r

wln-n-

A',,

is

(,!'

coarse iinl<'|enlem
Hill
>'

Also

=J
Finallv the
infinity.
in-i

n
i

use

lini

n->.\
-|iial t<> n.

-)=

and

liinA'=.'-.

and so of com

Thus
"Tinfurther.

|iml+*l+s+.l_^+..
Mitik'ndnl
;i\

to

x.

II.

before proo

tFor

tht-

L'<

-iitTiil ti-rni

in the liiiunnial

H expansion of (1+^)

is

That

tluTitht-

is

;i

jji-i-ati-.st

val

;,-iit.

livcause it

is

supposed

-lu-s

limit

x,

us

in.n

,ist-s

t<>

intinity.

144
If

SPECIAL POWER SERIES.

[CH. vill.
v will,

v tends to infinity in any other way, be contained between the two integers stage, say; and of course n will tend to infinity as

now

at

any

n and
v
)

(n + I)

(!

)"

vg will

between* (l + ) n and (l + be contained between ng and (n + l)g, so that


will be contained

does. n+1
;

Thus and

n>x(ng)

lim

= Km (n+l)=x.
n >oo

Thus, from what has already been proved,

we

see that

since (! follows that

and

)" is

contained between (1
2

+
s

M
)

and

(l

n+1 it
,

If

we

write for brevity

(*+)"it

r=0

2/r(*,*),

will

be seen that we have used the theorem


lira 2/r(*, n >-oor=0

n) =

r=0

r (#, [lim/ >)

)].

That is, we have replaced a single limit by a double (repeated) limit and of course such a step needs justification (see the examples in Art. 49).
;

Special cases.
If

= l/i/, we

have the equation


l

+ d! i-,+ + l+i z!
2-7182....

...

tooo=&
and proved to be

The value
If

of e has been calculated in Art. 7

=l(i/-l), we have
lim (1
v->ao
v

so that

l/v)~

e.

These two results

may

be combined into the single equation

where X approaches
*It
is

from either

side.

of course

then this value


rational.

On

tin- ih-tinitioii

understood that the positive value of (!+)" is taken; and H+1 if " is obviously fnntaiiu-d between (!+{)" and (l+f) a is if is the statement other h.md, the irrational, consequence of <>f un irrational ]>o\ver.
is
>

57,

58
58.
\V.

CES,

The exponential function. may denote hy tli.- synil."!


'

/'<

B)

ill.-

xjMuiential series

x*

a*

a*

'I'd-

-ii

(1

tliat
X,

K(x)

is

a contin
i^

t'uneti..n

..f

and that

its

ditl.-ivntial

coefficient

-i\-)i

by

term

ditl'-ivntiation, go

that

i)-i++lJ+jP+...-jrM
.

.=
rl

(/--I).
">7

TT| .

we

see

!'r.ni

Art.

that

"x

This rrsult
I-'A
>

i-an

also
tin-

prn\-r(l

dir.-ctly

I'rmn
r

tin-

>

rios
I

for

rule

(An

.lut.-ly

..|uati.iis

(in

oonvergeni serie& Tli.- i-.-ult which re positive int.-^ers)

inultiplyin^ leads din-ctly tc

+-[(iyf-E(n)-[E( Thus W B6 that


/

(-

;/-

'I'

valu- nf

t\.r
i'

rational valu.- of
in-atinnal

B,

Hut
j

tl..it'

relation must also be tru.ts

>r

values
l'

ot'

i'or

Sequence

i-ati"iial

nuinlx-i-^

wli..x,.

limit

is

./,

we

h;.

lim //(/ i=

A'.

the

]:ist

step In

alid IMMMU^w,-

'

i>n

(Art 46)
E(x)\
it

I'utuiv.

shall

p-nerally
in
l.

write

,-

instra

hut

when

the

e\]M.ii-iit

mplicate

is
!'

sometimes clearer

to use

ezpx, as
K

the c-hapt.T.

146

SPECIAL
will

POWER

SERIES.

[CH. VIII.

The reader

find

an independent proof (depending on

the Integral Calculus) of the equation

in the

Appendix
As

(Art. 162).

Ex.

a numerical example, the reader

may shew

that

e^=
and hence that
[Reference
results

4-800..., 6^ = 23-0 ____

may be made to the example of Art. 59 for some of the needed in the calculation.]
sine

59.

The

and cosine power-series.

[3 Hn+gr-"*and

Then it is plain functions of x and

that

Sn (x), Cn (x)

are

both continuous

as

may

be seen by differentiation.
(x)
;

Now C
is

= cosx
but

1 is negative,

and consequently
}

-i-

negative

(x) vanishes
is

with x and consequently

SQ (x)

is

negative when x

positive.*
(x) is positive

Thus -r-[Cl (x)]=


is positive.

when x
C^x)
is

is

positive; but

GI(X) vanishes with x, and therefore

positive

when x
and

Hence

-T-[Sl (x)]

=C
x,

(x) is positive

when x

is

positive;

S^x) vanishes with


is positive.

so that

S^x) must be
is

positive

when x

That

is,

-r-[C2 (x)]=

S^xy

negative

when x
x,

is is

positive;

and

therefore, since
is

C2 (x)

vanishes with

C2 (x)

negative

when x

positive.
use throughout
this
article

*We

make

of

the
(if

obvious fact that

if

is

increasing with x and is zero for for positive values of x.

x=Q, then y

continuous) must be positive

58, 59]

SINK AM'
t

us INK SKKIKS.
.

17

\Ve can cnnt iniie

his ai'innm-nt

and

l.y

doing SO
j

W6

find that
\\

(7 (ar),

Cz (x), C4 (x), CQ (x), ...


fi

are negativ.

h -n

(aj), J Dispositive, while the expressions with odd Mitlix.- are positive, This .sh.-ws that sin,/' lirs between th- tw<.

...

+
,.,,1

'

.v

+(
-

+ i>:
1

-...+<-

-I

3-2/1+3
' ,

(2iT~

Hence, since

=
t

(see Ex. 4, p. 8),

we u have
In like

sin2
manner we
1

2n+3)l Xs X6 X 7 KT+T? &-:+


t

to

ao,

find
...
i

>.4

7.6
i

These results have been established only for positive values of x; but it is evident that since and its series both change sign while cos./' and its series do not change sign, so that with
./',

the results are valid for negative values of x also. The figure lelw will s.-rve to sli'W the relation between

sin./-

and the

first

two or three terms

in

the infinite ser

Ku:.

If,.

Ex.
\\Y

Lot us calculate c<.s(irr) and


li,t\,- .r

= ^7r=r.")7<s

vci v
'

This

-0

&**=
07W,
,

= -00003.
tlu-

=
L-S546

-0209,

i-j:>;

0,

wror

1-oing less
leea

than -00003.

Also

r-1, the error

IKMH.I:

than -00003.

148
60.

SPECIAL POWEK SERIES.

[CH. VIII.

Other methods of establishing the sine and cosine

power-series.
(1)
is

to

Probably the most rapid method of recalling the series to memory assume that sin x and cos x may be represented by power-series.
if

Thus

sinx=
cos x
-j-

we have
and
so

(sin x)

+ 3a 3# 2 +

. . .

4.

Further, a

= 0,

because sin.r
1
.

is

and
=0,

COS.T is

1,

for

#=0.

Hence we get

2 a2 =
.

2. 3. o 3 =

-!=-],
-a2 = 0,
3,

or

=-

3.4.a-4 =
4.5.<x 5

=-a

or && =

and

so on.

But of course we have no a priori reason for supposing that sin x and cos# can be expressed as power-series and therefore this method is not
;

logically complete.
(2)

We may

start

from the

series,

and

call

them,

'say,

S(x\

C(x).

Then

multiplication of the series gives

C(x) C(y)

Hence
and

in particular

[C
C(Zx) = \C(xff-\S(xff.

S(*c) = ZS(x)C(x\
these formulae

From

we

can shew that S(x) and C(x) satisfy the ordinary

results of elementary trigonometry.

Further, that C(x) =

it

can be seen without

much

trouble that (7(2)

is

negative,

so

has at least one root between

and

2.

But

and S(x)

is

always positive! for any value of x between

and

2.

Thus

C(x) can have only one root between


in that interval.

and

2,

because

it

steadily decreases

liccaiiHt-

\\i'

[ES

~
(

'all

this r,,ot

thru

and ao

l>e

positive.

Hence

ami
<

W
MI

>

C(X+T)--C
can be based.
the

these facts
[t

tin- \vh..lr
ditliciilt

r of Analytical Tri.u "ii"iiirtry

is

nut

to piove,

liy

imliK-tioii

or

l.y

methods given

in

<

'hap.

IX.

l>rlo\v,

that

n(n-l)
2!

\vhrre

= t&n0, and

both

aeries trnninatr after An or 4(

+ l)

<>r

i(/>

Thus we have, on putting


\

# = .r,

n>

To theae azpnemona
u>inu as
r

i:

-an

apply the

the.-:

tin-

coinjtarison-series
...

;!

=|tano:|^|f |.
\\itli

that rinpl..yed

f..r

the

exponential

limit

in

Ait.
/

.~7

ami

\ve

get

r+

1\/
1
1

2\^*

iLM

"*)!

*-\
1
-f-

-]= COS -, ^

we have
that

lim n /=0,
li,n

an.l

thus

eoe*-

lim

(H

M=l

(see Art. 58),

150
(4)

SPECIAL POWER SERIES.


Another instructive method
x

[CH. VIII.

is

to apply the process of integration

by

parts to the two equations

sin# =
If

c
/

cos(x-t)dt,

co.s#-l=-/
we obtain

rx

s\u(x-t)dt.

Jo

Jo

we

integrate twice

by

parts,

2
~\

c
/

x 2
t

(x-t)--2jsin(x-t)\

co$(x-t)

r-

and

cos#=l-|

sin(#-) +

f-2

f-x

cos(#-)J

fi

-^m(x-

and so on. Thus we

find that, in the notation of Art. 59,

Hence we

find

and
61. The binomial series. Let us examine the series

-l)-j

+ K^-l)(^-2)+...

to oo.

from elementary algebra that if v is a positive now this series terminates and represents (l+x) v integer for values to examine the theorem other proceed corresponding
.

We know

We

of

v.

By
by
x
|

Art.
|

12,
;

the

region

of

absolute

convergence
is

<

and so

vergent in

any
f( a; ) =

(Art. interval (

50)
A:,

the
/.,),

series

given uniformly con1


.

is

where

< <
/

Now

^l +

( l/

-l)a;4.( l;-l)(^2)^+... to

x]
v.

= vg(x)
where g(x)
differs

say,
(v
1
)

from f(x) by having

in place of

60, 61 J

BINOMIAL
(1
-f

Also

\+( v -l)x + (i>-

1)(.

x -2)jj+...

BO that

Hence we
or

see that
/(aj)=.-l(l

+
x.

where

a constant in<l. JM n<l. lit of v\<- choose the po*>' consequently,


.4
is
it'

But/(0)=l: and
>lue

for (l+x)",

have

.1

that

18,

This result has, of course, been proved only for the im '.\trnl-l t. inrhnl.(-/, + /.); K't u- now see if it can
!

th.-

series

points ly the

1,
(

+1.

Th.>th
(??

<jU<'ti.-nt

<>t'

the

/>th

term

in

th--

,/

is
i^

//

1).?:,

if

/>>i' +

l,

and so the
an.l

series con\< r-vs for ic=


if
">
1
)

1 if v is positive (Art.
).

for

./=+!
(Art.

(r+1)
tile

is

thenl-elll
i-

sillll

positive (Art. '2\ of thf Sel'leS for

Thus by Abel's
iC

1
is

18

0,

if

is

positive: and

for

.r=+l
f,,r

the

sum

is

'2'

if

i'-fl

positive.

l'hB

ni">t

rapitl

nictli-ul

r.-calling

the series to m-

th- ditlrivntial

M|uatinii
(1

iy

assuming a

sei

f(.r)=\
\vr

+a r v + <i.,.r-+
1

On

sul.stitMti..ii.

lind that
-'
'..,

+ </i =

etc.
..nU-i
t..

tliis

mu>t

l>e

supplemented as abov.- in

o>ni)>l-t-

tin-

|i"..t".

in
r,

multijilv l'_:-!li.[
\<
:

t\\,,
)

writ

\vitli

tlitlVrmt values of f (say


:

ami

r.)
r

and

\it. \rt.

L'T

that
1

tlu-ir

j.l.xlu.t

whii-h

s;i.

.iiimt

as was
lu-

must

tlu-

rth

|H,

\\n

,,f

its

valur

f.-i

= l.

152

SPECIAL POWEE SERIES.

[CH. vin.

(3)

We

have

where we obtain the

last line

by integrating by

parts.

Continuing thus,

we

get

Now,

if

is

positive,

(l+x-t)

lies

between

and (l+#),

so that

provided that

> v - 1.
if

On

the other hand,

is

negative,
x}

we can only say that


v

IJ

-,

if

r>v-l.

Thus, in either case, the difference

tends to zero as r increases to infinity. If

x=

1 it is
,

series to a finite

interesting to note that we can sum the binomial number of terms. Thus we have

and so on. Hence the sum

to (r-fl) terms

is

Therefore
is positive,

it is

clear

from Art. 39 that


if v is

this

sum tends

to

if v

to infinity

negative.*

62.

The logarithmic

series.
II.) of

take as our definition (see Appendix logarithm the equation

We

the natural

Because the series

+i+ +

...

to GO

is

divergent.

81, 62]
v

I'

GABITHM*
'i

\vln-ii

./'

<
\\-

wriit-

ml-t+t
tin(

...tOQO,

l.y

Art.

uniformly ha\- th- -\|>ai,


:!

from
4

to

=/.

log(l
llo\\.-v.-r.
it

+ .!) = ./not
861160

A./--+^./- -J.i-

+.
of
tinunit''.

i^

(oiivrr^rnrc

"1

tin-

in

ry <>nl-r

i"

make use
to

mi
;

inlt-ji-al.-

t.-i-m-ly-i -Tin

Can

\\rib-

Thus

in-

i-f.'-)

= ,--^,^-f!.' -...-H(-l
;

oF+
It'

*
i*

is

positive, tin- la^t intf--ral

cl.-arly

l.-ss

than

P<
J
/'

+1
i>n.vMMl even for
that

which
)n

tends
1.

to

/,-ro
//M

<s=
(

Thus

as // tends to infinity. Ini/m-it/iiHtc series if >"//''/*

x\.

the otht-r han<l.

when

./

is

-,-111

only say that

jf;
98

than
1
tl.

+#
Ssion
'

Jo
it

anl from

woull

!.

cxi..-ct-il that

./=

be '-xrlnd.-.l iVoin
9;

of
tin-

th<-

loirarili

ami, as a matter of fact,

96li6fl

been

|'ro\

.<!

to

li\

-2

..

tin-

i.l-

(1-f J-X = tf

-i

Tli.it

is,

.-re

be

ext-

of

unitnim

OCH

tegrated series (compare

Art. 47 UM-1 Ar-

154

SPECIAL POWER SERIES.

[CH. vni.

It is now necessary to consider whether the first of these series can be re-arranged in powers of y without changing its value. By Art. 26, this derangement will be permissible if the series

is

if

convergent, where f = |# |, ^ = \y\. But the last series is the expanded form of (1 -)~ T!, which is convergent <1 that is, if |#|<1. Thus the derangement will not alter the sum.
;

Hence we get (from the

coefficients of
...

?/

and y 2 ) the equations

to oo,

^(l + HHi)+-..
Similar (but less
of log (1+0?).

to oo.

simple) series

may

be deduced for

higher powers
9.]

[Compare ChrystaPs Algebra, Ch. XXVIII.,

63.
is

For purposes of numerical computation of logarithms

it

better to use the series

x which can be found from the previous series by writing 2 x ). for x and then subtracting; or directly, by integrating 1/(1
In either than

way

the remainder after

terms

is

seen to be less

iX.

-I

* ~T~

~*

J-

~1

The

details of calculation for log f

may

be arranged as follows 00 00 00

I 5

= -20 00

00 00

-20

-^3=
-J3-=

800000 +3

2666

67

32000 +5
12 80
51

6400
1

-^r=

+7

83

+9
2

06
20 27 32 56

= -40 log f
Tin-

54 65
l.cvoiul
tln>

error involvrd

in

m^li-ct in^

terms

tifth

is

loss

than

-i4L_-.
result
is

rff))

which cannot

affect the eighth

<lt<imal.

Hence the

correct to the sixth decimal.

62, 63, 64]

LBH

>l\

\M

AJEf

get

again

l-ss

th;ui

;i

unit in

tin-

ri^hth

!.

Also

log}
th.-><-

= :.'[-imilll+-OOCn.=a'2L'.'Ji
1

'KXXXKX)3]

t" -ix
tlnil

decim
the natural logarithms of all integer7,

ivMilts \\-

t..

In (with tin- rM-i'ptinn of

whi.-h .-an

1..-

t'--iind

simOarlj

r'i"in

l"g).

In

particular,

\vi-

havr

Iog2=
log:.
:

-698147,

Iog3

= 1-09861:.'.
\vlii.-h
<

= log 10 ^:
tin-

han

febove hav- I>MI f"iin<l.


to

-on verge

more
:

rapitlly,

and so enable the logaritlnns


>f
.")

nmiilM-r
l"ic

placrs.
in

T<.
..f

illu>t iat<-.

be easily calculated to a the reader may find formulae for


r \\ritiii'_ ./--!

log 2, log^i.
1

terms

the three series obtained by


--

31,

lt;

'"

+ ir* + ...).
and
,n

64.
II'

The power >in ^. we


~

series for arc sin z


b;i\

and

BO,

l>y

tlit-

principle of

reversion of
b-ast

Beriefl

(Art

~>~>),
r

we

can expreefl th- Dumerically


in
.>'.

value of
<!'

as a

_ ent

tin-

first

t\v>

t.-nn*

\\bicb arc

\\-f\-rr.

it

i^

ii"'

i-asy

tx>

obtain
\v<-

tin-

^'iirral

la\\"

<!'

tlic

coetlici.-nts in this

manner: hut

can

me

th>-

ditliculty

by nsin^ the Calcn! \V.- have in i'act

^c^il-^r
wlr
i!

+ :r +

.:;
(

,g

'

:>. 4''

^.l.,;

to

lie

between

-ATT and -fl-. sn that

positive
B6VJ

The
writing

r-

i^

nhtain.-d
:

lV..in

the

binomial

Beri

I'mit will th. -f.r: --^. ami unil'nrmly in any interval (/,-. +/,) if ()</,<!. We may int.- urrat.- fcenn-hy-term and so obt-iin
i

./

arcrina

= ,- + ,
i

i.

156

SPECIAL POWER SERIES.


converges

[CH. VIIL

which
(

1,

+1), as
l,

may

absolutely and uniformly in the interval be seen from the test of Art. 12 (5). Thus,

we have a series for |TT. Although we have not found* a series for tana?, we can we have For, writing x = tan easily find one for arc tan x.
<j>,

writing x =

dx
or
d>

'o

where

</>

is

supposed to

lie

between

|TT

and

+ JTT.

The

integral last written is less, in


aj

numerical value, than

2n+l

/2n,7/__!
;

lo

and this tends to as n tends Hence we ha vet

to cc

provided that x
|

= 1.

...

to

'oo

where

==+!,
we have

jTr^

In particular

Of course this series converges very slowly, but by the aid method given in Art. 24, the reader will find no great difficulty in calculating JTT to five decimals, from the first 13 or 14 terms. The result is JTT = '78540.
of the
*

We

can, of course, form such a series

by dividing

sin.c

by cos a- (compare

Art.

54),

but there

is

no simple general law for the

coefficients.

The

first

three terms are


anil

in

more coefficients are given in Chrystal's Alyebra, vol. II., p. 344. The method used in Art. 54 shews that this series will certainly be convergent XZ yA y& < 1 &"il a short calculation will shew any interval for which ol + T-, + ^j +

that this is satisfied in the interval (-1*3, -f I'.'i); but by means of a tlu-ornn given in Art. 84, we can shew that the region of convergence is ( ^ir, + ^TT). fOf course the term-by-term integration could also have been justified by

making use

of the

uniform

coim-riM-in-i-

<>f

the series

-x'2

+ x*-

...

64,

65
1

ntlGONOMETRK
rfoai

\i.

t"

uvergence of the series


i\.

kii'.un
Tl.
I

,'\,

tan

[en

7r )

= 2.H
\
I*-.
1

.tli. -r

m
iVi'in
1

1.

65.
Il

Various trigonometrical power-series.


is

clear

Art.

'21

that

tin-

BZpCU
i

l-2rco80+r*)- =l+(2rco80-r )+ without may In- arraii-vd in ]M\V.-I^ |mvi<lrl that* <r^-|.
i

f+... to
alt-riiii:
its

oo

..!'

\alu\

Tin- srijih-nre of coeflirimts


t'..r

is.

h..\v\

er,

MHI-- -a^ily l--ti-niiiinMl

tli.-

fraction

(l-rcoa0)

(12

.1

.1

.1

....

are of cuur^.' functiona

<>i'

H.

have thf Mi-ntity


1

-rooed- L-h^r

+J

:i

4-

-l
:;

/-

+...

-^00802^^00802
ami
h.-ncr

80+...

we

get,

u-in--I

Art.

52

-0-1
.1
,

39,

and BO <>n. Thus w,

ha\.-

tli.-

r_
It'

-l+rco80-Hrt coe20-frI coe30-h...


sul.tract
1

to

\\v

and

ilivi.lt-

l.y

r,

\\

J^+y-co8 an.l whn. r | ..atfl+r^^airi+r a <l,as.. Thiii|2roM0


:

;f9-f ...

tn

Jr|+^ = f|<l.

158

SPECIAL
these,

POWER

SERIES.

[CH. VIII.

Combining
1

we

get also
...

r2
to
oo
.

An

exactly similar argument will give the result


to oo
.

By
led to

inspection

we

see

that or

all

these series converge

when

r + r 2 + r3 +...
the interval

converges, enquire whether the equations are not also true for
(0,
1).

when

0<r<l.

Thus we are

Now we

find, identically,

Hence, as for the geometrical progression (Art. 6), we see that lim.R n = 0, if and accordingly the first equation holds for the interval (0, 1). And the other equations can be

0<r<l,

extended similarly. Again we have

= - 2(cos 0+r cos 20+r2 cos 30+


by what has been proved.
Hence, integrating,* 2r cos + r2 ) = log(l

.)

we have
2(r cos

+ Jr

cos 20

+ ^r3 cos 30 +...),


0.

no constant being needed because both Also we have


dt
>

sides are zero for r =

(*'

dt
7

==
$7\

~1

5 ^ r^ ^ an

Thus we

find
sin
.

<ft

Jo

= r sin + ^r2 8^20 + 1^ sin 30+..


*

may

Term-by-terni integration is purinissihU- hrcause, if 0^r^fc< 1, tlie series be compared with 1 + k + k- + k?+ ..., ami Wt-ierstrass's J^-test can be u]])lied.

65

TBIGONOMETRK
\\
,

\i.

SEED
tin1,

li.-iv.-

only

.-stal.lMi.-d

ill--

-,,

nations above on

</<!:

hut

we knew from Art 20 thd


>-f ...

ih-

two

series

rin0+j MB
lepi
tin-

!ir>t

20+in30+-= or i'or
~>i
>.

Thus, hy AbeTfl theorem (Art

we have
I"-'*
1

C080+
and similarly,

''+..--

Mm "J

-l-Vn.sfl + r2 )

-!Tjlog(4Bin

cos0-4cos20+Jcos:W-...=
a
in
in

li

-r2 )

1'iv^li
lit-

tioo
iV.iiu
tin-

i^

unnecessary,
1,\-

the la-t
clian-iiiL:l'r
:

driluo-il

]>ivcc(lin--

In

likf

inaniii-r

\v.-

have
v

hisin30-|-...=limarctaD
^I
\ii\\
is
tli.-

t'roin

tinr/>,

ti^un-

it

is

-vi<l-iit

that
to
J.

tin-

an^l.- in
drtinitioii

ijii-

anu.-lc

wliich

(acc..nlin^
lit-

the
TT

of

th--

tan

runctioii)

must

l>-t\\v<'ii

and +J7r; so

that

17.

Thu>
I

Jut

Bm0+ism20+i8m30-|v .. = J(ir-0) = 0, nr "2~. th- valu.- of th- Beriefl i> if


>

if
(>

0<d<
l-rc-au'

t.Tiii

in

it

vanishes: thu^

tin-

.>ntinu':

and
If
lies

Ix'Uveen
aegatiye,

'2L--

and 2 (Jb 4-1) IT,


ha\.

wli.-iv

/,-

is

an

iir

w.-

= j[-series (A:

N].

MM these are points

of n.-n un.

160
It
is

SPECIAL

POWER

SERIES.

[CH.

not difficult to discuss the series

by a
/(0),

direct method.

In

fact, if

Sn (0)

is

the

sum

of the first n terms in

we

have,

by

differentiating,

Thus

n(#)

Now, by

Art.. 174, Ex. 2 (Appendix), the limit of this integral


lies

is

TT,

provided that ^0

between

and

TT

and consequently

/(0) = i(7T-0),

if

0<0<27T.

But

/(0) = 0=/(27r).

Thus the curve y=f(6) consists of a line making an angle arc tan ^ with the horizontal and two points on the horizontal axis. glance at Figs. 12 and 13 of Art. 43 suggests the conjecture that the limiting form of the curve y = Sn (0) consists of the slanting line and

two

below).

vertical lines, joining the slanting line to the axis (see the figure But as a matter of 'fact this is not quite correct, and the vertical
clearly the point

lines really project

For

above and below the slanting $_ w^ ? ,_ g


(X.lri)

line.

belongs to the curve y = Sn (B), whatever the positive number Now, as tt-*oo, this point approaches the limiting position
/i

A may

be.

f J

A sin
t

in virtue of Ex.

3,

Art. 174.

Similarly the point

u = Sir.

y=

/"*
\

sin
t

dt,

Jo

belongs to the limiting form of the curve.

Now

the integral

(sin t/t) dt

can have any value from


i

to its

maximum

TTX (1-1790), which occurs for A = TT so that in the limiting form of the curve y = Sn (B), the two vertical lines have lengths T85194, instead
1'851 94 =
;

.'o

of

^7r=r57080, as conjectured.
85
--

-J7T
-1-85-

FIG. 18.

Some
of

of the approximation curves

y = S n(0) are drawn


<

for various values

in Byerly's Fourier Series, etc., p. 63 (No. II.),

,sv-/vV, etc., p.
/'////.

49

and the curve w = 80


45, ls)s,
I'].

is

and in arslau s /',.,///>/.< given by Michelson and Stratton,


5.

Mag.

(5),

vol.

XII., Fig.

vin.
I

A.

161

Differentiation
1.

and Integration.

.Ju-tify

tin-

i-iju.-itinii

5-a4s + J^-J:i/- + -~A'n:? rf


Thus the
De<1 ......
series can

'

<a>

4>0

>-

be found
~

in

Unit.

..nal.

[GAUSS.]

"""

+ -=
1

1111
that

ll^'-sba+-= -Jl
[Math. Trip. 1896.]
.'\v
if

and

i-

are positive integers

3.

Utilise
1

Ex. 2 to prove that

(l-.r) r^73 + 273T4 + 3T4T5 + --^~2^ + ^^~


<<

x*

and

find a

fonnula

fr

the

sum

of the

Obtain the former result also by


4.

111
i:

the logarithmic

se:

2r4-4-rr6 + 6TY78--=i('r
5.

- 3 )-

.i''

Shew

that

with

rrt:iin

restri-tions

on

a,

j8,

y,

and dedu
Ft

\:>,.

rh.

r (y)r(y-q-)

7-)r<y-!sy
*In
a

number

of these examples, the

is

used aa equiva

me
are implied.
I.s.

cases the words /or sufficiently small values of


I

x
^

162
6.

POWER
The complete

SERIES.

[CH.

elliptic integrals are

7.

Prove that

8.

From Ex.

13,

Ch. VI., or Ex. 5 above, prove that


-

r(2)

-4

,2.4.6.8
[Write
9.

a=

J,

/3=-, 7 = !.]

1 Multiply the expansions of (1-x)' and log(l-.r), and deduce by

integration that

[Compare Art.
10.

62.]

If

n y = (l+^)- log(l+^), shew that

Deduce the expansion

11.

Prove that

[By direct multiplication, or by expanding the

differential coefficient

(l-^)
12.

Prove that

[Use

the equation
similarly that

Shew

13.

Prove

that,

if

\x\<\,

Is the result true for

x=\

vin.
I

LMPLE8
Sln-w that,
if

A.

14.

\jc.

< 1,
i
.

:j

:>,

log[i{lW(l-*)}]=15.

Prove that

K^,,^.,^.'^,
[It
is

easy to see

th.it

(1-

,(.

+ J* + J** + ...)

+ (l+*%r-J^ + J*-. = 2{*-(J ,116.


tf

j/=(seco:-ftanar)
.

m=

2a^,

prove that
,

-IK-i + jrfa-SK.s -... =


Of
(\
//.]

[Hen
17.

I6C,

[.V

.1896.]

Verify that

J(arctan.r).log(l-

where
and dciluoe that

J log 2
[Here
(1

=
.1

"

!>-....

[Math.T

+)$-<
= S...
fi

.S)^+(*6

For the second

]>art,

use Al.rls thftOT

Derangement of Expansions.
18.
/

Prove that the

I'lu-fli.-ii-iit

f .;-"

in th.-

i-\ji:uiin

>

iy .w(m

+ l)...(iM+M-l)
'
71
!

,.r

><(w
]

+ /0

/>

//(/<-!)(/
4!
-,

2!

whei
Sh.-w that

7
it

is

inultiplr

.f

the o..tlicient <f

in

the expansion of

--7V
(1+-.

- -- -i+qt
j
3-;;

(1-

(1-

164
19.

POWER
Expand exp (arc tan .r) up

SERIES.
term which contains
x>.

[CH.

to the

[Math. Trip. 1899.]


[This function satisfies

(l+^2 )=y, and

so

we

find that

if

/nl ) then

^ = 1,
=-l,

an+l = a n -n(n-l)a n -i.


4

This gives

=1,

=-7,

....

The
20.

possibility of the expansion follows

from Art.

36.]

Shew

that,

if

|#|<1,

>. 1902.]

[This
21.

is

the expansion of (2+y)a where


,

#=2y+# 2
first

use Lagrange's series.]


in the

If

y=a

+ a 1 ^+a2 -^2 + .-.


1

>

obtain the
1
*

and second terms

power-series for
-

y22.

and

I Ids -----

a^x

y-a

x dy

=-.

Apply the

last

example to prove that


!

if
1
,

- [(x - a) cos f(x) = (sin x sin a)~ a]" J


then

/"(a)

= lim/(.r),
x

^a
[

j- [/(a)]

- lim/(#) = | sec 3 a y
n

A sec

a.

Math. Trip. 1896.]

23.

Prove that the

coefficient of

in the expansion of (1

-1

n\
the

r L

7i(7i-l)

^-1)^-2)^-3)
"*"

2271-1 2(271-1)

2.42w-l2rc-3

"'J'

number

of terms being ^(^ + 1) or


first

24.

Determine the

three terms in the expansion of

2 2 1 25. If [(l-#3/)(l-tf,y )(l-#/.y)(l -tf/^ )]" is expanded in powers of x, the part of the expansion which is independent of y is equal to s - *2 2 (1 +s )/(l -^)(1 ) [Math. Trip, 1903.]
.

[If

we expand {(\-xy)(l

-x/y)},

we

obtain

It

is

then easy to pick out the specified terms in the form

26.

Expand
first

(l+.r)* = exp(l
in x*.

2 3 -^x + ^x - J.r -}-...)

up

to

and including the term


three terms are e(l

[The

-J^ + JJ^

2 )
;

the possibility of the expansion

follows from Art. 36.]

VIM.]
27.

165
th.

1.3

1.3.5
3!
(1
..-"

x*

(T^?
in

powers of

*,

she win* that

th- ooefl

ia

that

tlu-

iii->t

MI

,-ial

to

N '(

-./-'),

and hence find an

28.

\\li-

HARDY.]
i

29.

""-rr?

-<*<!.

in
ami extend to any nunil>-r
30.
-f

!p'tin.-

indices of 8iiiuniati"n.

[.I/"'

:K)3.]

Shf\v

that

following series are alislutely


IH-\\

-(inverg/nt,

and by

summing with
SB

tliat
T
J

tlu

are as

gh

=
,.

-M7

^TT
-

r.r
X X i

_j_

E
[STERN.*]

Special Series.

i-tsii*
sln-w that
also
[If
'(1
i

l-.-^T^'^y+1
i-

th-

|t..>itivi-:
-

and
is

if

that

tli.-

x-i-i.-s

i>

njual to
l,\-

1).
\~>

V>0,
-tyjt
;

wt'

ran

j.iov,.

Art.

that

'(

/-,

y)

equal to

and

this .-an

I..-

extended

ise

0>y> -1,

by

the
32.
If

\aiial.lr

from

to

\-f
l

in

iri-t

th-

~'ilt.]

(1-r

.+ ftr +
s

-(i-^o+^y1900.]

*See

Dirichlet's

.-,te

InttgraU

(c

( i

166
33,

POWER
Shew
that the
is

SERIES.

[CH.

sum

of the squares of the coefficients in the binomial

power-series

>-*-

[Math. Trip,

[Put a = /3=-i/, y = l in Ex.


34.

13,

Ch. VI., or in Ex. 5 above.]

Prove that

Discuss the validity of this equation for


35.

x=\.

Shew
its

times

that at r per cent., compound interest, a capital will increase to original value in n years, approximately, where

In particular, the capital will be doubled in the time given by the

approximation

n=

-69.3
r

h'35.

36.

If

prove that/,(#)

is

a polynomial of degree

in

x which

satisfies

the equation
l

x ~tt

**

/2 =(.r + a) 2 +#, /3 = (# + a) 3 + 3#(. + )+#, and that if a is positive all the roots of /,(#) = are real and negative, and that they are = [HARDY and Math. Trip. 1902.] separated by the roots of /,_i(#) 0.
Shew
that /j =x + a,
;

37.

If

^a nxn = (l-x)-%, prove

that

and deduce that


1

n-

n-...
2.]

[For the latter part, use Ex. B,

[Math. Trip. 1890.]

Trigonometrical Series.
38.

By

writing

r=-#sin0

in the power-series for

arc tan {r sin @/(l

r cos

#)},

shew that
arc tan (a -f x)

- arc tan a = (x sin

0) sin

6-

(.r

sin 0)2 sin

20 + $(x sin

3 0) sin

36
...
,

where
39.

=cot#.

[EULER.]

Prove that the

eerie s

is

convergent and is equal to $ir-6, when 6 lies between -eot0 in lv\. :is or [Put r = cos0 in Art. (;:
:
. |

and

TT.

vin.
I

167
Sh-\\
that
li.-iiiflrr
<

40.

qual to

le

equal

tin

i,.-i_-lit

of a regular h-jtagon inscribed in a circle IB nearly il triangle whose Bide is equa.

41.

If

tan./'
|

-A

42.

<l,tain the f..ll.\vin;

d.-li

rals

Iog(l-2rcos0 + r-').

'(i

Vl -rc<-

z)*'

~>

oosne

^
1

/"'

sin>^si

Jo
ItiOD
1
1

l-2rcosi
i
'.

'-f'<;.')

in

an

intt-_'

and tTiii-ly-terni integration (Art. and 1.


(Art.
'

45).

43.

Pv int"jratinu thr

|Mati'ii

*-...

= (Awe can
infer the results
2
,

and
1
.

./

lies

between

<>

and

1,

C080 + ;,

:J0+...

')7T

nn

r3

cos^-f^cos
[\.t- that
tin-

30+...=
[!'.

I.M.]

.iixtrmts of
\aiiiil-.

intrifiati..!!
i->

can

!<

d-l.-i inin.-d
.

thus:
:

The

unifcim
'

anil thei

tinnuus.

f,,i

all

values of

0.

Th>;

+
r et u

-f ...

J,

= ATT*, we

tind th.

-JA**,
whi,
I

168
44.

POWER
Shew
that,
if

SERIES.

[CH.

|r|<l,

arc tan
45.

Prove that
cos
sin

6+ cos 3(9 + 1 cos 50+ ... = J log cot 2 J0, 0+sin30 + sin50+...= +7r, (0<0<7r) - 1 sin 30 + ^ sin 50 - ... = log (sec 0+tan 0) 2 sin
series
;

The second

these values of

= 7r, STT, ..., and changes sign at while the fourth changes sign and

is is

discontinuous for
discontinuous at

0=^5
46.

IT*

cos

Prove that 0cosa + ^cos2#cos2a + cos3#cos3a+...= - J log [4 (cos 0- cos a) 2 ],

cos
sin

0cosa-cos20cos2a-f ^cos30cos3acos a +

...
. . .

= Jlog[4(cos + cosa)2],

cos

20 cos 2a + sin 30 cos 3a + sin a + J cos 20 sin 2a + J cos 30 sin 3a +


sin

=/(0),

. . .

=g(0\

where, assuming

<a<

TT,

i(ir-a),

ifO<0<a,
ifa<0<7r.

-^a,
47.

Prove

that,

if

0^0^7r,

and

find the

sums

of

" cos 710


'

sin

?i

48.

Series for

TT.

Sin ce

- ?r = arc tan - + arc tan 5 o 2> 4

2 2

,2.4/2\

3F

,2.4/1

both of these results are due to Euler.

vni.|

LMPLEfi
'.sit
\

i;

169

'1

which leads to
6 fl-i* 1+

Hn
l

4"fi>L
i

3 io

-L YJ. la. "flJ. + OV!oJ f -J 4TooL +


'

2 4

2
:

/'

a.

'

/ l Vj. 5(100;

+-}

also

gave the result


l

-7r

l
:<

'$

11=5,

\\lii.-ll

TT

-^^
lead

lu'lilv

COB

*-*
-|

7r,*4f.

l++
1

2.4/144V
below
'--in

+l>e

In writing these serif* il"\\n, \\in


tin-

n..ti-

that

..

2,

put

f'-rni

r,

2
-

\
1

-J-

EXAMPLES
1.

1',

Euler s Transformation.
Sl,.'\\

hy the

saint-

nirtli.xl as
...

in

'hat
...1

N'(

!+//-)[
01

il

-/r,.

Similarly. jr'V- that


;

^-

.r'+...^
x

s
N

>'/

+...],

If

[KrLKR.]
1.

2.

By takin_

',

= J,
'

cr

=l,

...

in

K\.

ihew that

if

:/

< 1,

logty

K\
Dedu*-c that

V3 log V~7r/~~

J\J*\_

,2.41

2.4.6

v:

h"
=1
3.

in the first
if

Ml

[Ei

Shr\\

ftlM that

'

V:

4-...=

vV)-(Mr,,),-^ + (/>-,,)^
,

^;'

170
4.

POWER
In particular,
if
,,

SERIES.

[CH.

=w

3
,

we
/

find

a =0,

DaQ =-l,

Thus from Ex.

3,

^^
2 ^ #" = (a? -f 7#2 + 6-r3 + .z4 ) e*.
oo

Similarly,

,,r

5.

If

*=2|L,
e,

is

an integral multiple of

and

in particular

[WOLSTENHOLME.]
6.

By

integrating the formulae of Ex. 2 above, prove that

if

|y|<l,

24 l|+|:
2 w4 2 y4 2 4
.

Are these equations


7.

valid for

y = l?
if

[J/a^A.

^n>. 1897 and 1905.]

From Ex.

4 above, prove that

oo

n
i
:

xn n
I

4
n- l

and that

^(-l)
1

Sn
71
!

= 0.

[Math. Trip. 1904.]

Shew

similarly that

Obtain these results also by differentiating the exponential


series for (1 -4?)~*.
8.

series

and the

Apply

Euler's

method

to prove that

m m+l m + 2 f+3
"*"+ UH-W
9.

__(_*\+
~
f

'

2
/"

"(7tt

+ l)(m + 2)\l+#/

Vi
"

Apply

Euler's

method

to

prove that
a
.

F(a, ft 7, ^) =

(T^-a^(
!

7- A

7'^)
x* +
...
.

where

/(,

= ft 7 ,)
,

x+

[GAUSS.]

\MI'I. l>

I',

171

Miscellaneous.
10.
If
//

(!+.'-)

ail.

l...th

less

tha:

that

[#=;
11.
If

',-_'!'
'

Also

1.4
12.

From
in

the fxpan-i-.n

.f

(1-^.,)"
tint-iiln-

ih-t-i niin.-

tli--

drrinial i-lan-s.
(>l>tain

[Til

Uirr.ilL'.-JTS.]
\\-iv

the

saint-

root

of

from

tin-

-.\]>ansi.>!
:

KK -l

Theorems of Abel and Frobenius.


13.

With

thr notation of Kx.

!>.

sh-\v

tliat

tf

14.

If

n.i

if

then

OUL]
that
< 1

ivea

apply AWl's
15.
If

tlit-..|

-iii.
]

PJ
if

ITS and
I'/
lin,
^.'iil

liui t

51 to

pr'\c (hat

oi-

di\

.-1

L.

'-nt
I

that

if

/''

.1

tenda to

lim{2
X *1
Hi;.'

J-^,
ntiv

and

that

.1

tendl to

172
,

POWER

SEEIES.

[CH,

16. If the coefficients b n satisfy the conditions of Ex. 15, and fn (x) n decreases as n increases (but is always positive), prove that ^b n fn (x) will converge provided that 2a,,/n (.r) does, if lim A n fn = Q. Deduce that when

/n (l)=0,

and lim

2, ^(.r) = rr>0,

then

\i

[Apply the lemma


17.

of Art. 153.]
:

If

Ayn = y -v n+ i, & 2 v = v n - 2v n+ i + v n +2, and


ra
ll

lim(?iv n )

= 0,

shew that

Writing

fn(x) = &?vn A 2v

in

Ex.

16,

shew that

if

y n is positive,

and

if

then the series


are equal.

-f

2 A 2 ?> 2 + 3A 2 y 4 +

. .

v
1

2,v l

+ 3v 2 - 4y 3 +
x tends
to

. .

If further v n has the limit

as

1,

converge and then the last

series has the limit | as

x tends

to

1.

Use the method \imn&v n = and limv


18.
n
x vl

of Ex.

17 to
x *!

shew that

if

&2 v n

is

positive,

and

if

rt

= l,

then lim(i'
3,

-^4- y 2 -y 3 +...) = |.

[For another method see Ex.


19.

Art. 24.]

From Ex.
lim
x-+i
i

15,

prove that
1

= 1 log 2 = lim |(- 1 IX-l)- -^^-u -f>. )-^^ 1-.^' *_i n(l+x ) 2
i

20.

Establish the asymptotic formulae (as^-*!),


, and $n(-x)

~-1
i

the second, multiply by


21.

[The difference between the two sides of the \-x and use Ex. 18.]

first is

less

than

in

On

the lines
:

of

Exs.

17-20 establish the following asymptotic

formulae (as /r-*l)

Tf

--'~4irp
where

is

Euler's constant.

[CEsXno.]

Lagrange's Series.
22.
(t

Shew
t

that,

if

+ a) n = n + wi(t +
is

a positive integer, ~ n~l + 4- n r a(a - rh) r l>)


is
. . .

+ //>)" ~ r + ... + a(a

nb)"-

where w r

the ordinary

binomial coefficient
[ABEL.]

vni.
I

B.
.". ;ind

multiply
i.il

I'.v

equate

coetl

authors have con-id- n-d th- \aliditv

i.f

tin-

equation, also due to

A1...I.

4(t + a)=<t>(t) +
h.-ii
i.

<i>'((

+ b) + a(a
IM-M-.
\\'.

-* b)
<!>"((+.
M.-iy

--ults raniK't
if

I..-

..'ivi-ii

remark, ho\v-v.-r. tha'


Tiniiio.-t

</>(f)
i

is

Ini:
1:<1.

recent

28,
tin-

").]

23.
is
(

Prove that the


1

i-ot-Hiri,

in \\
[
'

BZpl
<.f

-1)]"

l'i"\,- tliat tin- roctVirirnt

'f.,"

in tin- <-\j>ansi<.n
2 "- 1

(l+.r)

(2

+
[Mtl,.
Trij..

1906.]

[Tse Lagrange's series

(1) for

y = e*-

and

(L

-j/), wli-

y
24.

']
l>u
r

Kxjtaml

t*

and

>

in

)"

t-f*-t-

= (a1

and

drtri-iiiint-

tin-

interval

[Write
25.

(*-^l+y and apply


is
1

J.a^ran.u'*

'-

-e.]
\v

If /'(.')

pc \vt-r-st-rics
in

in

./,

wh.
/('')]", in a>.-,-nding

that the
of
:-e to

the r.\jiaii.in <>f [1 the '"tli. -it-lit ..f .'" in th- Sip
.'

powers

iiiiin.-

tlie

coefficient

for tlie f.-llowini;


:

f>

n.r;

n,

(3) log(l
[\\'<

MK.]

[The ivsnlts
:

0)'"" ^
<4)

;;':;

(8)0or(-l^
Oor(-l

(3)1/( W -1):

(6)0
\

on.

Tin- values
26.

occur

\vh-

.-n.]

that

.i

N
27.

Lagrange's

s.

Use Lagrange's

series to establish the eqi:

where

/=./(! -.r) and

174
28.

POWER
Prove that
4

SERIES.

[CH.

6.7# 5
and
1

8.9.10*7
2
1

where
29.

t=x(\

+x2)
u-3
,

< fa.
.

Shew

that
,

on-l

nt *n-S

tt(M+l)
,

*(*+ 1)
,

(2tt

- 2)

=
n

^
TVzp. 1903.]

[M^A.
is

[This is the coefficient of l/# in the expansion of therefore equal to na n if

\,v(l

-z)\-

(l

-Zx)-\ and

2o^" =
Hence

TT

log(l

2.9?),

where y = x{ 1 - .r).
or a n = 4"- 1 /??.]
first

2a ny n = - J log (1 - 4;y),

An alternative way of stating the result is to say that the sum of the terms in the binomial series for (1 -|)~ M is equal to the remainder.
30.

Extend the method

of Art. 55 to
4

prove that
,r

if

y = a*>x*'+ a$y? +
there are two expansions for

. . .

of the

form

Shew
where
31.

also that

if

W
As

<7Oi

n is

the coefficient of

l/.r

in the expansion of
if

a particular case of the. last example, shew that

then
32.

.?!

+ ^2 = a,y + (ab + c)?y 2 +


find

. . .

If

y^x^+x}-,

we

# + ff2 = 2a,y,
1

where

a 1 = m, a 2 = wl (2^-l)(2m-2)/3

!,

33.

l)

- 2) (3m - 3) (3m - 4) (3 w

etc.

It

is

easy to write
:

down

the general forms for the expansion of

in the following cases

Differential Equations.
34. If

P
,/--,,

are two power-series which converge for |.rj</i', prove that the differential

equation

has solutions of the type y A + Ai. + J.y-+... \x\<R. Here A and A\ are arbitnuy, while conil>iri;itions of J and J,.
(

whirh also converge for


.!._,,

.1..,

^J 4

,...

are linear

vni.|

LMPLES

r,

17
.'e

the

coii

,,-t

-f ...

+ .!,'/.
.1,
.

i-t-J,//

we
i,/;

that

/;

.i,,j,

if

#,=

/;

.!:

and

*B

..v+c*-;
that
/;,,|<J//-",
//,,

J^

where
\\,

r<R
ti,,.,,

and
tind

u mob

lin:

-r.

Thn-

tmvergM

if

.-

</; and

\v.-

lind that

1.

.verges

35.
th-'

Witli

til-

differential

<juati(n

tli.-

type
-...),

ie

t(t-l) = tpQ + q
A
.

and

J,,

is

arl-itiai-y,

the

other

erti'-i-nt8

being

inultiilr- ot
[If
/'

is

thr nth-r rM.t

..f'

th.-

<|ua<lrati<-,

we

lind
.

-{(*-l+t)pl +qJA n-l +...+(tpn +qn)A


wliidi

may

!*

i-mnjiai -d

with

K-l+T)^+...+(l-fT)^},
re S
If

= \t-t'\,
tak./>

r = \t\,

\p,,

u>8.
shall

we

= |^ol
o.

Bp =

A,

l.-n-r

;us

J0^8, we

have

Fur-

that
6
:

lin,

=r.

that

1'.!

.-

COnV<
nicditicai

<

/,',

as in

il

in-tli-..i
;

with

tip

tht-

"(uadiat
\\i!l
].

-MI

an integer. wh-n d( (+t-t') vanishes


in
!'

\\

i|tinn of tin- i-haiiL'es

f-und

36.

Sup|iose (hat

'hat

|*-bl^a, \y-y \^b,


and
that

y=y,,+

= 1.

176
Then y n approaches
satisfies

POWER
dv

SERIES.

[CH. VIII.]
77,

as its limit a continuous function of x, say

which

dx=f(x

>

^
x

for sufficiently small values of \x-x \. [It is assumed that for these values of x,

[PICARD.]

y
is

\f\<M,
where
function of y.

M and A are constants


|yw

\f(x,y')-f(x,y)\<A\y'-y\, and so in particular f(x, y)


;

a continuous

Then

values of

and so, if c is the smaller of a and b/M, the \, within the prescribed limits, provided that X-XQ <c. Suppose now, for brevity of statement, that x>x& then

-y |< M\x xQ
fall

yn

ly-n-yl = r\f(x,yn )-f(x,y n-i}\dx<


Jxo

and
I

#1

- .yo
find

< M(x ~ ^o) < Me.

Thus, by integration

we

and generally

yn

yn ^ \<^MA

n~l

(x

- x^ n <
(

MA n ~ cn
1

Thus, by Weierstrass's rule, y


their limits
77,
.

and (similarly)

-^

converge uniformly to

dx

Also,

g = lim^
is

because f(x, y)

continuous as regards

yJ]

CHAPTBB
Tl;I<;<>N<>.MKTl;irAL IN
66.

IX.

VKSTH IATK
tf>

)XS.

Expressions for

and

<-in /<M sin

as polynomials

in c

\\V have seen (Art. 65) that


cos

+ } r^cos 20 + ^cos 30 +
2
/>2
'

Bui
f/ .-) = _[(
wln-n.

y _r-)+^/-y-

<

,<'

7-^)3 +...]
may
be
re-

y = 2cos0; ami,
j.o\vi-rs

t'urtlu-r.
r,

the latter series


2

arr.in-vd in
that

of
is
tin-

without alteration of

valu.-. jn-ovided
is

()</-:^|.
/'

It
in
J

therefore
r\j session

evident that -cos//#

the

coefficient of
1

(/-y-'-'
/ /

)"

+ 7;-^T 7(

n />2

+ ---+<''yr*.

becan^'

"/.
"
,

i"

+2
,

...

contain no terms in

Thus
//

//I
(f/

(71-8).

*!

innnlxT of tt-nns
1

lciiiLT

ritlu-r

i(//

Of

<

/'

1.

= y"
4f

2
/?

y""

2
*

+
l.s.

..

178
Similarly,

TRIGONOMETRICAL INVESTIGATIONS.
we have
rsinO
seen that

[CH. IX,

1-ZrcosO+r2

= rsin#+r sin
2
-

Hence we deduce that


series
1
2 )

sin $

7r

is

the coefficient of rn
2 3
.

~l

in the

+ (ry - r + (ry - r + (ry - r +


2 2 )

Thus

where the number

of terms

is

either

$n

or $(n

+ l). We

note

that this formula can be 'deduced from the last


It is therefore evident that

by

differentiation.

both cosnO and sin7i$/sin$ are in of cos$, polynomials degrees n and (n 1) respectively. But
for

some purposes

it

is
0.

more useful
This

of

nO

in terms of sin

we

shall

to express the functions do in the following

article.

we

Before leaving the formulae above, it is worth while to notice that if write y = t + Ift, instead of 2cos0, then 1 -ry + r2 = (l -rt)(l -r/t).

Hence
and
so,

log (1

ry + r

2 )

= log (1 - rt) + log (1 - rjt) = ~

rn (t n

+ t~ n \

from the foregoing argument, we get the algebraic identity


t

+ t~ n = y n ny n z + ...

as above.

Similai ., y ,

and

so

we

find

~r. =

n~ l

-(n-2
cosTifl

n -3

+ ...

as above.

67.

Further forms for

and
to (J^r
0)
;

In the formulae of the last article change

then

y = 2 sin 6, and we find - 1 )m 2 cos 2ra0 = y 2m - 2my' (

2 "'

--+... to

=
in case

(m + 1 )

terms,

is

even and equal to 2m.

66,67]

BINES

AND 006INE8 OF Mil.Tiru:


njiial in
2 'M
l
<

\\<;i,E8.

I7:

Bir
(

M.l.l aii-1

-2n> -f

-1

<

iNm<2,,,+l)0 = y

+ -<:>/,<

+ l)y 2m -' + ...to(m+ l)terms,


>

(-l)

M(2m + 1 )d
c<

asy

-(2m-l)y

->

+... to

(m+1)

terms.

llo\vr\.-r.

tli.-sr

f.

nnul;u' take a

arm
it

wording
not
sji.M-ially
lnit
it

more ele<: pe \vli-n nding powers of sin0; of course


tn

i-

litlicult

rearrange

ilnill.-

l.i-.iically.

is

instnirt i\v to obtain

ezpreeeiona al. n-ults in another

way,
in
//rt

or cos

//H,

\v-

ha\-

\vrit-

.'

sin H.

anl w- have

it'

\\-.-

<(

n^ilT
//

tin-

fxjiri-svioii
'

^ivrii alx>ve for cos

2m6,

966

that

/'//*

//

M
//

polynomial of degree
in \vritr

in X,

can be expressed as a containing only even powers; thu^

cos
tin-

?>

=1+J
I.

,.'-'

-1

* 4 4-

4-

A nx n

ronxtaiit t.Tin
\\.'

l.'iii;_;-

h.-caiisc

0=0 gives CO8 1101, 050.


in
th<*
diti'.-r.-ntial

Il'

vul.stitutr this

expression
4.

equation,

ire

tin.l

.2.

.l..-f:{.

/--'+.

..+<,/-

and

so

.1,

L__
r-

__ZV

42 )

H
\\ht-n

ii!!<.

oo6ft0>l~
i-

+
'

k
. t

//

Siinilarl;.

tin<l

tliat

sin

nO
.r

is
:

a polynomial
tl.

of degree

/'.

which contain^ milyodd powers of


sin 710=
4-...-I-J

180
the

TRIGONOMETRICAL INVESTIGATIONS.
first coefficient

[CH. IX.

being determined by considering that for


~

= 0,

dx

d
find
.

Hence, on substitution,
.

we
.

A x+5
3 3

x*

+n(n- l)A nxn

+n
Thus
giving

(nx + A 3x

+A

x5 +

+ A nxn ) = 0.

3.2.^ 3 +

-l)?i = 0,

5.4.^ 5

to J(7i
?i

+ l) terms,

being

oc?<i.

To verify the algebraic identity between these results and those of Art. 66, consider the case n = Q. Then Art. 66 gives
or
cos

60 = 32 cos 6 0- 48 cos 4 0+ 18 cos 2 -

1.

Change from 9
cos

to (TT

0),

and we get

66= 1 - 18 sin 2 + 48 sin 4 6- 32 sin 6 6 2 2 -2 2 )(6 2 -4 2 ) 62 6 2 (6 2 -2 2 ) 2 = ^sm 4 -- (6 sm ^ l-^sm + ~6!


.
.

-^

in

agreement with the formula for cosnO.


Again, consider the case n

=l

we

have, from Art. 66,

or

cos70 = 64cos 7 0- 112 cos 6


Hence, changing
to
TT

+ 56cos 3 0- 7 cos d.

0,
3

we have
6'

sin7<9=7sm0-56sin 0+112sm 6 0-64sin 7


-l v - 7(7
2 2 )
.

7(7

-l

)(7

-3 2 )

differentiating the formulae just obtained for cosnO, we find

By

sinnO and

to A(?i

+ l) terms,

when n

is odd',

and

67,68]
BID

-SINES
is

AND 008INE8 OF

Ml

I.Tll'l.i:

AV-LE8.

1-1

3T
to JTI term-.
that

when n
Til'-

f
in;iy
>1|. \V

iv;ulrl-

co70/coB0^
an.lth.it

+ 80 sin 4 0-64 sin 6


::i'>iii
::

i(>0/co80

//-T-:^>in 5 ^f

;unl '"iiiuif tli.-sr formula--

\\ith

til"-,-

,,f

An

68.

'I'd'-

tone obtained in

the last article are resti


//.

ly

c.-i-tain

conditions on

tin-

tonn of
in

L.-t

us no\v

st-e

il'

these

ran

J,

. (

i-,.ino\-<-i

any

\vuy.

.iniplr

the &
-

which was proved to terminate and to represent cos??$, win -n


//

is
\\

,-.

/'

ifl

au odd integer, or
It

is

ninatc.
,

is

natural

not an integer, the series does to consider whether it is con-

to investigate its >inn. The test (5) of Art. 12 ahewa at once that the seri .Tges X =1; and so, as we have explained in al>solut'-ly when
it'

and

so,

Art.

-")(),

the

series

converges
is

,i1,..,/

nt>

! >/

and

U/nifan

It

l'llnws that
j>.

continuous for

all real

values of

6,

and

as on

17!,

it

satisfies
it

the equation + ~7^


follows^ that
//

?j2

# = 0form

From

this equation

is

of the

J.cos
\

6+

7>

sin

= 0, we have y = l, Thus .1 = 1, 7^ = 0, and so


w
for

= 0.
-,')(

cos
for

nO=l

-**+%!^s*-...to
.juiitiiMi
:

oo

any valu
This
ol)\

;m.l

since
l>e

tin-

solution contains tiro

:ul)itr;uy

constants, Equations, Art. 8).

no other

solution

can

found

(Forsyth,

Differential

182

TRIGONOMETRICAL INVESTIGATIONS.

[CH. IX.

Similarly, we find that all the other formulae of Art. 67 are valid for any value of n, provided that x 1 and that the series are continued to infinity.
\

<

however be noticed that the third and fourth series are not convergent when x = 1 but they converge absolutely for x x ^k 1. 1, and uniformly for
It should
| \

<

<

69.

Various deductions from Art.


that

67.

We know
sin

we can

write

nO _

sin0
r

(n odd)
sin"- 0,
2

sinflcos

= ^ + ^2 sin2 + "- +nan

(n even)

where the

coefficients

are the same as those

Art. 67, but are not needed in

Now

explicit the left-hand side vanishes for

worked out in form at present.

0=Tr/n, + 27T/71,

37i>,

...

so that the right-hand side (regarded as a polynomial in sin 0) must have roots
sin

=
is

sin

(TT/TI),

+ sin (Zir/n), + sin (3-Tr/n), ____

When n
are
.

odd, there

all different;
~

must be (n 1) of these roots which and these can be taken as


.

sm0=+sin-,
,

TT

m.

2-7T

,...

But

if

is
.

even, there are only


.

(?i

2) different roots,

namely

^ sm0=sm-,

TT

2?r
,...,

Thus we have the formulae


(n odd)
sin 710 sin
sin 2 0\/

=
=

sin 2

sinVV
sin 2 0\/

sin 2 2a/

'" [

X^
sin 2

(n even)
sin 710

~n/

sin 2
2

sinVV

sin 2a/

"' [

where, in each case, a = 7r/7i.

68, 69]
It

SINKS

\\|,

<>siNi:>

.r

\I

I.TITI.

\ve

compare these with the


ii

explicit

I'-TIM*

<'>7,

we can

^T
6
In
a

Ja

+-

_
8ini(n-2)a'
tin-

8m-.

-imilar

way w- pn.v-

identities

,,v

cos nfl

-^y
4n<
where

= /. 1
l

sin

0\/ ~ sin 2 \
1

f
I

-ii.-'
I

"si,^A
.

",iii(n^in 2

/. ^m'0\ ,>nO = (\- sm-0\/^ B (l--v-.7~b

f^
...

sm-/3/\

an*3/8/
<"/'/

l--^-o7 sm-(/'L
/8

P = w/2n and

only thih tin-

multiples of
7,

appear.

fonn>
---

we

see that

^
Again,
if

=
iI^8-

^3^ +
ri

+
sinV-2))8'

'

f
;

-+

8in"(-l)^
Art.
()(i,

\\v cMHi>itl-r tin)

I'm-muK-u- of

it

is

that (cos 710 c-.s ,,\ may -x]>r^sed as a polynomial of degree ;/ in cos^, tin- term of hi^hrst tindegree bein^i \ cos iv if zero (cos ?i0
!

,,

-2-//\, 47T/'\,....
in

tin<lit}-r.'iit

factors of

the polynomial

(jurstion

will

be

of theXprefisiODfl
v
,.,

form

H_,.,, S
i-

(\2a), C080
f<>r

cos(X4a
1>-

.....

wli.-rr
It

as befon-.
s.-.-n

u^--.l

is

.-asily

that the n ditirn-nt factors ean

tak-n as

ooe(X+2a
because ooe (X 2 H.-ii,-.- \v,- ha\- the
cos/If

>.<'-

''

cosO-cos(\-f -2,

(A+2(n
i'l.-ntity

n-l
.:

r=0
\\r
\\i-itf

x(X+2

in

this
2

-X]IV>-MIM wt- ha\r


II

-'.

184
or,

TRIGONOMETRICAL INVESTIGATIONS.
with a change of notation,
sin

[CH. IX

nO=

2^11
r-o
,

sin (0

+ TV*).

But the sign must be a, all the factors; because, if are positive, and it is easily seen that both sides change sign together (as passes through any multiple of a).
70.

< <

We

as an infinite product. Expression of sin have seen in the last article that, if n is an odd integer,
sin
71

\
/

where

a = 7r/n.

Thus

if

we

write

= 6 we n<j)

have

T<7

r=l

To

this equation

we apply

the general theorem of Art. 49

we

have, in fact,*
sin 2 (0/7i) sin 2 (r7r/7i)

^4r
2

_02_
2
'

because rTr/n

is

less

than
series

JTT.
00

Now
2

this expression is inde-

pendent of n, and the

r=l

^0

/4<r

is

convergent; consequently

the theorem applies. But we have lim n sin(0/7i) = 0,

r hm

'
.

= r hm
=
,

sin

Consequently,

The

special value

-^O = %TT

II (I

Theorem t

leads at once to Wallis's

7T

TT

224466
sin

We

see,

by differentiation or otherwise, that


:

xfx decreases as x increases,


jr.

from

to ^ TT

thus
1

> (sin x)fx >

2/Tr,

if
if

Consequently,

sin (rw/n) "> 2r/n,

<x< < i n.

Also |sin0/n|< |0/w|, for any value of 6; and so the inequality follows.

69, 70, 71]

THE SINE
thl'"nnula

PRO]
i'.r

I'.y

combining
;

MM

fJ

witb

the results of

Art.

ee that

''--""'
,

:,.-ijis

\v..i-tli

\\hile

to

ifffr

Kin-fly to

an
>ii\

in-

ompl.-tf

M
]
i

given in Bonn

for

'

0=nr,
-*0

1,

it

is

ur^-d that

fj

m
thi-

the

i\v

hut .-\artly th-

s.r
v

would apply equally to


ival

th-

'

number,

so that

* /

/M \
{jossibly

only suggest* that


:iii"t

11

(l-^-rJ may

be of the form a 6 sin 0/0;


ii"t.-l
tl.

pi>
//-

we

separate
iian

l-^-j
n-L
r
i

In this (.(niH-xioi). it luav be / 6\ / f) \ ), and th into factors (l (l-\

--

more

is (say f>
is

--ry

q negative factors),
41,

^
\\

the

th.-

pr..(bi,-t

~i

ff) /**]?). Vf/ \ ^ /

fsee L

Art.

an.l

0^=-.]
j.i-.M.f
i

A sat
_u">"il

i. it

thf..- lin.-s inv..lv-s

some knowledge
M<>il.-y
>

of fin.

pp'iif

is

Lri\-i!
I

by Haiku.-- and
..ut

1^).

have Already piint-tl


to cases

when
.>)

tin-

J/-t<-.-t

the (lan^i-r dees n..t

'f

jipplyinj;

the thi-Mi-em of

h>kl ^ood.

An

additional

his risk

may be m
it
I

= sin</>,
ti.

'/'-f-1)
1

follows that the values of sin(rr tlmse Nvhen r ranges

t:

t<

A(/<--l),

but

in

thf iwris" "idt-r.


8in 2

-'r
h'

pa
and
if

SL^a
:.
\\-

sin^^y

\v-

applv the theorem

appeal to get

whi.-h eonti;idi-t> thf


i-

n-siilt
'

..btaim-d

before.
is

<

x
'_,

that

thf i!if<|uality
;

__"
i-.

<

n,,

lon^.-r
-ible t"

true,

since

nr/n

may

be greater than A-

and

it

in fact,

imp.

<>nstnHt a n divergent

oomp
71.

Expressions for cosO,


l'nlln\\ ilio<

cotfl.
in I'xjT.-^ing
thi'

Tin- reader ^b<.uhl tin<l n<> dilliculty


illtillitr

cosO as an

pl-odlK't,

tb<'

lilli-s

nl'

p!-fCC(li]i_

\\'r

hive

in

fed

Art.

cos

*= "ill.- .T
i^ "'/'/.

wb.

ami

186

TRIGONOMETRICAL INVESTIGATIONS
is

[CH. IX.
is

Here the comparison-series

20 2 /r2 and
,

the result

Alternative methods are to write cos


to the product of Art. 70 rearrange the factors.
;

0=

S1 "

/~S~?

an<^ appeal directly

or to write $7r-@ for

in that product

and

To express

cot$,

we know
n~
i
.

that

0^ sm = n sin- II r, 1 n r=
i

1_

\\, sm 2 (r7r/^)J'
.

sin 2 (0/7i)-]
.,,
,

(n being odd)

and

so

if

we take
-

logarithms and then differentiate,

we have

the identity
1 4.0 -=cot

^ V

1}

2
.

n smo, r7rn /
2

sin(0/n)cos6n
.

si

assuming that

is

not a multiple of

TT.

this identity we apply the theorem of Art. 49, first step is to obtain a r comparison-series

To

and our

^M

Now
.also

(see footnote, Art. 70, p. 184)

sm2 (T7r/n) > 4r2 /^2


2

provided that r

< \n
2

sin (0/7i)

<

/7i

Thus,

sin 2 (r7r/%)

- sin2 (0/7i) >


n>
2r

(4r

)/

for all values of r such that


sin(0/7i)cos(fl/?i) sin 2 (r7r/7i) sin 2 ($/?i)

>

and consequently
if

Since 6 is fixed, there may be a fixed number of terms (i.e. a number independent of n) at the beginning of the expression for cot 6 for which the last inequality is not valid. But for
the rest of the terms

we can

use the comparison-series

which

is

To

these terms

obviously convergent and independent of n. we can apply the theorem of Art. 49

for

the fixed

number

of terms preceding

them no

special test is

necessary.*

*No test of this character is requisite for any fixed number of terms; the object of Art. 49 is to enable us to handle a sum the numher of whose terms tends to oo with n.

71]

in:

<'T\v.i:vr SEB1
"
'

187

'

li,,,

ocri

I!,,

'-/tan J)-J n/ 9 n\

i
.

8in(0/7i)co8(0/n)

e
*

sinVir/n)-8m*(0/
1

'I'

-0i
This ivsult can also
>in(>.
l.e

derived

at

01106 lYoin the

product for

ty differentiating theorem at the end <!' Art. 4s


In exactly the saint-

u>e the Logarithmically; we ^hould then


to
jn^til'y

the operation.

way th-

identitir- (p.

1*3),

^IJ'S"^^^
and
.11

Cod
(r

-a-^s="S 2 nsin 2 (ri

dd> n

w*

apply intr An.

I'.',

the

two

s<-:

a~TrV
or
r
-f-

l_y_l_

2~7'

'

4.

and
uis- tlin-c
i

f+:
is

>-..=
the second of th--c
fr'iii tlu

no

dilliciilty in d-lucin.i;

use

+ ... =
(
j

-p+-

-J)(i'
th.-

4*
l.v

traiisf.,nnati.ns

being

ju<tili.-il

Ait.

i'h.

IV.)
r];il,o|-at<,,r

1'urtluT

dflails

of

(altliouM-h

greal

interest)

w.ull
ch.

analytical l-al UN too i'ar


a
i

ati.-ll.
<

\\'.-

content
('liry.stal's

oursi'lvi'S
2.

with making
in

to

Algebra,
ol'

vol.

XXX..

which
ident;

will

l.c-

i'ouml

lar^c minihcr

us.-l'ul

and interesting

188

TRIGONOMETEICAL INVESTIGATIONS.

[CH,

EXAMPLES.
1.

From

the formula (of Art. 68)

sin(mO)=msin 9
obtain a power-series for (arc sin xf
ai
:

-sin 3

+ ...

to

oo

namely
o2 /

==

6T
2.
'

2^\l

3 2 /2. 4"5"^\1 2 "^32

2.4. 6

From
I

the formula for sin

n9

given at the

end

of Art. 69, prove that r

cosec 2 6 + cosec 2 (0 + a) +
If

. . .

+ cosec 2

+ (n - 1) a = n 2 cosec 2 nO.
}

[Math. Trip. 1900.]


3.

is

odd and equal to 2m + 1, shew that

2
l

tan 4 (r7r/7i) = ^(7i-l)(7i2 +^-3).

[Math. Trip. 1903.]

4.

If

is

odd, shew that


r=l

'^cosec^nr/Ti) = J(%

- 1).

If
If

n = abc...k, where
to

extended only

a, 6, c, ..., & are primes, shew that the above sum, values of r which are prime to n, is equal to

(a -l)(6 -l)(c
5.

-l)...(F-l).

[Math. Trip. 1902.]


are given

Shew

that the roots of the equation 3?

#2 + if =
2 sin

by

AV7 = 2sin(^r),
-l) To distinguish between the two
7(3/

2sin(f7r),

(fir).

[Write ^ = 2sin0 in the formula of Art. 67 for sin

70.

Then we get
that 2 sin (f TT),
0.]

/=

or

^=^7(^-1).
lies

sets of three roots note

2sin(f7r) are both greater than


6.

1,

while 2sinf;r

between -1 and

Shew

that the roots of the equation


2 cos

are

2cos(y7r),

2cos(f?r),

(fir).

[The formula of Art. 66 for sin 70 shews that

are the roots of

f - bf + 6y y=x+Z
if

= 0.

The

substitution

leads at once to the required result.]


1,

7.

Shew

that,

\x\ <

and deduce

that, if

y=2

cos 20,

IX.]
.in

iMl'LKS.
the series
;i

,'

all

terra* for

\vhi--h

is

multiplr

.f

fn.iii

tin-

series

Shew

that

if

0<*<*,

the greatest

int.

_'

-r

IN

tl

to

').
9.

RISKS.-

Sum

the series

-i(

-)+^( Kin
'

!si

[~'l!u-

MUD

is

tin-

smalK-i-

>f

anl
~'^

if

<>__.,<
7T.J

in

it

"'(- + /',)(/Mtl
II. -li..
,

sin

(q^ain (a*r) ... sin (a^)^


if

'~sin(/>J 7r)8in(^7r)...8in(
It'

S6-2a^

>h-\v that,

tin-

;io-rnt

iinj.lirs

that

:
.

= li m

II

/,,

/'.

:lu-r, if

lim(// r)

= /\

\\- hav.-

wh.-r.-

i>

rxcliuled

and KM

n. ,.,..

hl ,t

_x

-C
**
~i

-J

8 " iL =1 -

8 -fi?^-

[K
fa
I

,
'

Why
12.

caniict tin- first ]n-olu.

l.r

written without an i-xpciu-ntial

Sh.-w that
(l-.r)(l
tin-

-fr)...
in
j.air-

= co8(|MX. Jl.
(

terms

ami
1

aj]>l\

'h.

VI.. ..Staining

r(-)
4.1

\viit.'

OU<

tin-

i>r..lm-t

f.-rm

-(

-.r)}
l
:

sin(|7r).]

13.

Prove that

_r

n fi L

-**- 1= (nTT+jr^J

[Ki

190
14.

TRIGONOMETRICAL INVESTIGATIONS.
Determine the limit
of the product

[CH,

where
15.

/x,

tend to

oo

in such a

way

that

Km (/x/v) = k.
sin

Shew

directly

from the products for

sin (x

= cos x, TT)

cos (x + ^TT) =

x and cos x that - sin #,


and
cosine.

and deduce the periodic properties


16.

of the sine

Deduce the

infinite

product for sin.r


fi*
Jo
/

from the equation

ran (&)*=* 2#

oos(2f)c&

by means
17.

of the series for cos (2xt) in

powers of
.

sin

(Art. 68).

Shew

that
sin (TTX)

" irx II _ QQ

F/ - #\ - )e*/ n H 1
( L\

%/

= -+ 2' =2

7r[cot( 7r^)-cot( v 7ra)]= ^J

-n a-n

---

7T

cot

(TTJ?)

cosec 2 (7r^) =

oo

2
-co

-j

(.1?

W)

ri 3

cosec ^^) - ? o

(TT^)"] _l

=2

oo^^-'

^.4y
is

[EULER.]
excluded,

In the
18.

first

and second equations the accent implies that n =


last

Deduce from the

example that
n=-V X
7
(a?
ll

lim
I/->ao

and
where
19.

v _^o

limf 2 2 L m= _v = - ^
summation

- m)(x - n),1=-^ J
,

in the double

all

values

m = n are

excluded.

Shew

that

and

sec ^

r^2-CTZ^ + yjl a*"'"'


cosec x = cot
( Ja?)

[EULER.]

[We

can get the expansion for cosec# from the identity

- cot #.
first)

The second series can be derived from the series (equivalent to the

by writing x+\ir

for

x and grouping the trims

differently.]

i\.
I

iMFLES
Shew
that

20.

I
Deduce the value
of

-lfc+A+A+.~
l)-'
tin-

n=0
21.

being excluded from


l'lo\e
tli:it

Miiimiation.

[J/.-

1896.]

?("
22.
If

-6-7^7 + "can be divided into partial

the general
in
tlic

term

<

form

1\1=0,
tli. -11

-B
\\li.-ic

2"

1.1-'->t(</7r),
,i

all

the ntunbers

a are supposed
sci

ditl.-r-ni

from

xero.

23.

Find the sum of the

-0
[\ H< L
24.

being 57i-2

iv\

'hided.

1\

6^^57?00

"m/+
2
i
.
.

Find

tlu-

vain-

w ;tnd dftt-nniiif 5,

A
-I)
3

-1A_
(271)3

C so

that

(i'/'

may be
if

quadratic

i:

25.

We

have seen <rh.

V!..

Kx. 11) that

rr(nr(QT + rr^noT rr(3)r(/n-' f/)J Lr(i+oJ"*"Lr(3+o'J


th-n
M.I

=
J{L
=

if^>A.
l.v

Now

^l=

--

Ait. 71,

We

lind

^ = ^(^-9), ^

^fcW+fcW^-S

CHAPTER

X.

COMPLEX SERIES AND PRODUCTS.


of complex numbers. assume that the reader 'has already become acquainted with the leading features of the algebra of complex numbers. The fundamental laws of operation are

72

The algebra

We

If

x = g-\-irj, x + x' =
x
x'

x'

= g' +

irj',

then

+' +

i(>i

*)'),

(addition)
(subtraction)

xx

= j/), g'-\-i(r) = gg' w 4- i (gri + g'rj),


.

(multiplication)

En

'ti

laws include those of real numbers and that these four operations can be carried out without exception (excluding division by zero). Further, these laws are consistent with the relations with which we
It is easily seen that these
;

as a special case

are familiar for real numbers, such as

= (xy)z.
Thus any of the ordinary algebraic identities, which are established in the first instance for real numbers, are still true if the letters are supposed to represent complex numbers.
It
is

which would be equally satisfactory.


suggest for multiplication

natural to ask whether other assumptions might not be made Thus the analogy for addition might
##' = {f' + i7;i/'.

But

this

is

inconsistent

with

tin-

relation

x+x = 2x,

since #' = 2

would

then give

2.r=2

+ i(0) = 2f,
t'

whereas

x+ x= + +

[72, 73

H.sXUIIIpl
MI

Mill

tin-

tive

laws an

to fix the

Uw of ranlttp
by
It

iuulti|>li<-Hti'.ii,
tliil'

a8iii:
rail
'

MIlllllMT*.
(t*
r.l.

be shewn

II..

\>\>.

*-\-2) that

we

jitjons

mad*-

i-lse

we

are forced

'he produ\\

ithout

t-it

ht-r

./-,

73.

Argand s diagram.
is

Tin- reader
tati"ii d' tin-

douhti*

I'amiliar

with the usual represen-

;nt

wit):

(^,

O
r.ut
it

f
19.

may

!>

i-nnvi-ni-iit

t<

inv.-

ln-i.-t'

^uinmary of
can
writ.-

th.-

tod
If

we intnxlucf polar
./

coordin.-,
/((.-< J
i

\v./'

^-f

si:
<>l'

\\.-

shall call r

>/

33

(it

is

someshall

times also callcfl the

mo
\x\.
'I'his.
.

</*w of a*):
i-

and wo

denote

it

l,y tli.-

symbol
n^,-,!
./.

(juitc ccn^i^tcnt

with the notation


her
\V"
call

+./

or

previously: for if x is real, \x\ will according as x is pov-itiv.- or negat


<>f
./
:

H the

wrgwneni

it

is

snm- times
:-

called

the

From

tlu-

diagram the nieani

:.

ident.

194
If

COMPLEX SEKIES AND PKODUCTS.


we draw through A, AB, CA equal and C are respectively x+x' and x x.

[CH. X.

then B,

', parallel to The fact that

OA

x +x

= x'+ x

is

represented
parallel to

by the geometrical theorem that

A'B

is

equal and

OA
or
\

(Euclid

I.

33).

Since

OB<OA + AB
the relation
|

OB < OA + OA,
x
|
\

we have
and

x +x' x
x'

similarly,

< <
or

x
| \

+ +

x'
1

|,

x'
1
\

Again, supposing

OA < OA,
# 4-^
r
|

we have

OB + AB>OA
Thus,

OB>OA-OA'.
x
|

>

x'
\ \

and so

also
|

x'

^>\x

x'

It is easy to

prove similarly that

and generally, that


and write

2x
1 \

<E
/2

x
|

.
\

These facts can be proved algebraically, thus, consider the

first

inequality

R = \x + af\,
J

so that IP = (f +
/

f ) + (^ + T?') 2
2

Then we have
Hence
and
this
is

= r2 H-r + 2(^ 2 2 /2 = 2 (r/ (r + /)


ft 2

certainly positive

if

^' + t]tj'

is

zero

or negative.

But

if

s positive,

we have

so that

the sign of equality only occurring

if

Thus
or

in all cases

and the sign


which
74.
is

and + >;>/> r represented geometrically by supposing that 0^1' falls along OA.
of equality can only appear
if

>?'-'?7

'

Multiplication; de Moivre's theorem.

If

we

multiply together the two numbers

x = r (cos
the product
is

+ i sin 0),

x'

= r'(cos 0' + i sin 0'),


sin 0'

found to be

xx' = rr'(cos 6 cos

& - sin 6 sin 0') + irr'(cos

+ sin

cos 0')

= rr

[cos (0 + 0') + i sin (0 + 0')].

73, 74]

Dl KOIVBE'S THEOREM,
tlio

195

Thus

absolute valu- of

/'

N
i.

/T';

or

I-I*M*'
and
tli-

ai'innm-nt

of

././'

is

O + tf,

tli-

sum
f)

of

the arguments
/

In pariicular,
to

if

/'=!,

tin-

product ,/<>->

4-

-in H

is

equal

r[oo

h* sin (0+0')],

o
It

is

therefore clear that

if

=cas0 + -isin 0,
/- l

1-

\^

pial

t<>

20+ % six
ami so
on, as indicat.-d
in

hisin30;
th-

to

cos(-0)+z sin(-0);

22.

if

i^

any

\vli<.l.-

nuinl>T. ]M>siti\)"

<

r n. -Dative. \\v

have

(cos 6

+
.

''

si

= T = cos
///

>?0

sin

/?

0.

To

intrrj.r.-t

\vh.-iv

and
(f

//

arm
f

wliol,still

nun
true.

agree that th- law of

indi<-.->

")"

is

Qenoe

if

./<-^''sin</>),

wr

p (cos ??<+/
I.

sin

,><;>

)r-r..x //-f)+ 'sin


^

thus,

and ty

//<t)4--2/"7r.

whciv
'rh;i-

/',

is

any whole
-cos
|

nunil>.T.

j(m0+2fcr)l

+ifflu

[^(m0+l
l.y

thus
I.

has

/,

litlrn-nt
/

Th'

casr

()

values, uiv.-n constitutes d-

taking fc0,

I.

196

COMPLEX SERIES AND PRODUCTS.


large

[CH. X.

number
of

Morley, some
to the
1.

which

of elegant geometrical applications have been given by will be found in Harkness and Morley's Introduction
II.

Theory of Analytic Functions, chapter


triangles x, y, z

We

note a few samples


if

The

and

x', y',

z'

are directly similar

x', y',

z'

1,

1,

1
1

If the triangles x, y, z and x', y', z are directly similar, any three points dividing xx', yy', zz in the same ratio form a third similar triangle.
2.

3.

The conditions

=
a,
1, 1,

and
i

zx-\-yw,

7
1
.

1,1

y
x, y, z

(where

a, ft,

y are

real

should be collinear, and that


75.

numbers) are respectively the conditions that x, y, z, w should be concyclic.

Sn = X n +iYnt we say that the sequence (Sn ) converges, when both (Xn ) and (Fn ) are convergent; and if Xn -*X, and Fn HF, we write Sn -*X + iY. The necessary and sufficient test for the convergence of the
If

General principle of convergence for complex sequences,

sequence (Sn )
e,

is that,

however

small,

we can

corresponding to any real positive number find an index such that

|#n-m|<e,
To prove
so that
this statement,

if

>m.

we observe

that

\Sn -Sm \^\X,n

-Xm

and

Sn -Sm \^\Xn -Xm


if

\,

+ \Yn -Ym -S m \^\Yn -Ym \Sn


\

\.

Now,
find

(Sn ) converges,

(Xn ) and (Yn )

are also convergent, so that

we can

to satisfy

and consequently
so that the condition
is

Sn - S,n

<

e,

if

>m

necessary.

On

the other hand,

when

this inequality is satisfied,

we have

also

Xn -Xm \<,
;

\Yn -Ym \<,

if

n>m,
sufficient.

and therefore the sequences (Xn ), ( F,,) are convergent. Thus (S) converges and therefore the condition is

application of this principle we consider the interpretai sin 6 and a is irrational. note tion of t* where t = cos

As an

We

as a preliminary that
|

(cos

$ + i sin

$)

(cos

tjs

4- i sin

\js)

=2

sin

|(0

74,

75

<

i:

OF

<

OMPL1
numl.-i^ which
lias

can lind

in

so
.

-,,,|<e/|tf
'I'llUS

11

>m.
|

|{cos(an 0)-f

9)+* 00(0*0)} 0!<e, <l /"

>

//<

and so the sequence


It

9)-j-isini
1

<-on-;

gent

IN

thT
rfiii.-nilM-i-.--l

'iiral

to delin.<///

/"

as lini/'

but

of course to be

that
*)

tin-

limits*

limlcosajtf +

l.'

+ :>/,-)]

(/.'=+!,

2,
/

3,
a
.

to

Thus equally well In- regarded a> incIulMl in tin- symbol tak.-n -I th.m.-anincr must be be attach. care to to ^{..-cit'y special

may

to<*; for most purposes


r.-diir, - U)
I

it

i->

>utlici'-iit
h;it

r.-tain tin- vulin-

which

\vln-ii

t)

tin-

valur ^i\'-n

l>y

Convergence and divergence of a series of complex terms.


It'

"n

iti n

w' bave

....
+
.,

Th. u

il'

fj

-f

to

* and
s,
t

//

-|-/y 2

4-

...

to

are separately

nt

tn tin-

sums

resptrtivi'ly, \vc say tliat


...

+ '/.,+
Z// (1

to oo

OOIH

tli-

sum

N-(-/V.

"/

div.T^es (or oscillates),

we say

that
It

^L"
is

<li\
t

o^-illates).

easy

<m this drtinitinn


///.

and

tin-

l'>n-^i.in_

COSfflOI] l'r

BtMjurncrs that

inn

e,

\vt-

LB Dimply th can tind /// such that

-ji.mdin^

tin- n-al

no mattci- lm\v
In
like

i^r.-at

j>

may

be.
th.-

manner
]>r<><i

\\v

..Main

<lrtiniti.>n
.

.!'

n\

<

rgence of

an
\\e

infinite
n.-.-.l

ordinary \\ only absolutely convergent }>roducts, which we shall


oniplr\
t

l,

in

in
*

Art.

77.

All

:ui

iitt<

these values are uiuMjujil. bOMMi ( must not be allowed

makes la equal to
t

;h

198
76.

COMPLEX SERIES AND PRODUCTS.

[CH. X.

Absolute convergence of a series of complex terms. a n = gn -\-iri n and if 2|a n is convergent, we shall say that 2an is absolutely convergent. It is evident in this case, from Art. 18, that the separate series 2 n 2^ are both absolutely
If
|

convergent, because
|fi.|^
It follows therefore that
|

and Ifcl^KI.
is

2an

convergent in virtue of our


,

definition (Art. 75); and by Art. 26 the sums 2 n 2^/ n are independent of the order of the terms. Hence also, the sum of

an

absolutely convergent series is independent of the order of

arrangement.
It
is

perhaps worth while to give a graphical illustration of the method

of inferriDg the convergence of

2a n from

Let

= a + a 2 +. ovHail +
sn
1

and represent the numbers

in Argand's diagram, as

below

O
By
definition

ff,

0*

ff

a <7f

O-

FIG. 23.

we have
Os 1 = 0(T li
S 1 S2

= <T (T2
1

S2S3

= 0-2 0-3,

etc

Thus
But we have
where
<r

*n*p
<r H <rp

= =

0"n<7>
o- no-,

by elementary geometry.
if

p>n,
.
|

represents the sum 2 an = lima w Now, by the definition of convergence, we can choose n to correspond to so that <r Mor = cr-<r w <e.
,

Consequently, n can be found, so that

**,<,
or, geometrically, all the points 8P

if

p>n

centre is s n

Since

can

!<

within a circle of radiiis irJnw taken as small as we please, the last result
lie
,

must

shews

(see Art. 75) that


f,

tintin-

p-tints

*,,

must duster alumt

s.unc
it.

limiting

\\hi--li

may

In-

<>

ciivlc
riivli-s

Imt cannot be outside

(li;r_iT;un

indicates the
ffiii
>f

with centres
lias

2,

3,

.-.,.

An

alti-rnaiivr

diagram
\.,1.
:.,

been proposed by G. H.

A.*ndU nf Mathematics

(2),

76.77|
It

AlisoLI
ni'

II.

(X)NVJD

199

is

CMIU-S,-

obviMiis n-Miu
alisMluh-

of ival

-hut

what has been said as to series is not oonvei necessary


i>l.-x

series whirl

although

M<

-*+.
For both
s.-ries

ival

and imaginary parts cunvi-r^, by Art. 21; but


is

of absolute values

which

div.-rjres

by

Art.

<

K\. >)
I\'.

W
l.y

Ai-t.

11.
M!'

evident I'mm chap.

that

the

sum

a non-absolut.-ly

oonvergeni series
77.

may

altrr.-d

derangement.

Absolute convergence of an infinite product of complex


intinitr proiluct
ii'

factors.

Thr
at

11(1 -f"

i-

"ail
i-

th.-

j.rMluct

11(1+

to be absolutely conIt foil convergent

onc-f
-

that

'v

convergent,
so

so

also

is

lid
t\\-M

F.-r

\\.-

know
have

that

S|a n converges: and


|

by

Art. 39

th..-

pi-M-lucts

Jl(l4-|a n |)
still
<

Hut

\\v

to

11(1 |a |) are convergent. sht-w that (in the case of complex


tt

and

pi-Miiu.-ts)

wmpU
^

am
if

deck

tonv*

\\\

</,<!. we have
"

1and SO
m+p
II

^11 + a.l
i

'

(lII(l-f a
>

II
IR-fl

(1+rO
it

since
that

converges,

follows

1
:

and thnvfoiv
thus we can tind
//>.

No

1( I

from Art. 39 a n \)
|

gent

s. that

H
\

<

"

>>l-e
be.
i

and

"ll'\l

",

)<l+e,

m+l
ei Mnall

may

11.

MII tind
11
-,,4-1

hat

L+a

>

<

1+6,
/'

-mall

may

le.

and however lar^e

may

be.

200

COMPLEX SERIES AND PRODUCTS.


is,

[CH. X.

That

II(l-fan )

satisfies

the general condition for con-

vergence; and so the product is convergent. It is easy to modify the proof of Art. 41 to shew that if *an is absolutely convergent the value o/II(lH-an ) is unaltered

by changing the order of the factors.


78.

Pringsheim's tests for absolute convergence of a com-

plex series.*
of chap. II. can be applied at once but since to the series of positive terms 2 a n
1 \

Of course the conditions

= V(&2 + ?n2 |on|


it is

)..

if

On = fn + ^n,

Of course the

evident that the square -root complicates some of the tests. tests of Art. 9 can be at once changed to
lim
(7n
2
2

,\a n
.

<

oo,

(convergence)
(divergence);
rule,

limDn

an

2
|

>0,

but the same transformation cannot, as a


the ratio-tests of Art. 12.

be applied to

Thus, the condition

n+l

by no means sufficient to ensure the convergence of 2 an ; because whenever lim Dn oo (which is usually the case), the above condition does not exclude the possibility
is
1 \

n+l

which may occur with a divergent


For instance, with l/a n = n\ogn and
2

series.

Dn = n

(n>2),

we have

because

Thus,

but yet
*

2|a|

diverges.

[See Art. 11

(2).]

Archivfiir Math.

u.

Phys.

(3),

Bd.

4,

1902, pp.

1-19.

77,78,79]
It

TESTS FOB

LB8OL1

can be

therefor*
li.ul />

ii/

tin-

conditions

,.ee)

\\ni\D
\\-\i\

uliMantially <-<juival.-nt
if

t<i

tin-

conditions
lind
/i

..l'

Art.

"2.

the

first

condition

is

s;it

:nid

/"

*!;!'
"
.
i

%
/;

"

"'

--s;
it

Now
limit to

in all cases of ju-n-ti.-al intt-rost,*


I

is

possible to assign an

-2s-

and 9st, say


1

th.-u

l) n

J?S-

-//

and so the

r'nditi.ni "f Art.

;(!.

But

if

the

sfc..n<l

.-..nditinn

hld>. we

C0

tind

si.

that

if

that

tli-

lir-t

fa'-l.r

must

nr-m iv.

79.
It

Applications.
i^

easy
/'
)

t<>

UMii^l'orii!
\-2

ri-iii^ii.-iin'.s

rnnlitinns ly writing
tind

/'

/(

in Ail.
'

(3),

and

th.-n

\vc

lim/v-,,>

win -re
only

fash point

to

notice

i-

that

p.
in

ii,J""'i-' =li ,,,

*-'
>!'

jr-*

\irtur
-o.

L'l I -

j.ital's

nili.

(Art.

l.">2):

no\v

w-

that

and

|/<
us take

also tends to /.era

In particular, In

-n,
ailiuitx slightly

(8)/(n)nlog
-itions.

m.-n-

202

COMPLEX SERIES AND PRODUCTS.


obtain,
after

[CH. X.

We

few

transformations,

the

following

conditions:*
(1)

lim
[

'n+l

(2)

lim

an
7i

2,
2,

(convergence)
(divergence).
> 2, ' 2,

(3)
''n+i

(convergence)
(divergence).

The most interesting case in practice (5) of Art. 12, where we can write
(4)
^n+l

is

one corresponding to

'""

(I

w
in
(4),

It is easy to see that -if


i2

^=a

(5)

n+l

from (2) we see that the series 2|a n converges if and diverges if a in case a = l, we apply (3), and 1 find divergence. Thus we have the rule:t

and

so

.|

>1

<

When
a.

n
real part

the series

than

1,

2|aw converges if the and otherwise diverges.


|

of

jm

is

greater

On
holds,

account of the importance of the series


proceed to consider application of (1) above

2an 03 M when

(4)

some further results. An shews that the series converges if absolutely |os|<l, diverges if \x >1; and we have just examined the case when \x =1 and a> 1, proving that then 2an # w is absolutely convergent. But we do not know yet whether the series may converge
(although not absolutely) even when a^l. will increase with n, negative, the sequence \a n ut least after a certain sta^v, in virtue of (o) above; and
for
|fl5|

we

=l
is

If

*The

inequalities

are here used as equivalent to the two


t Weierstrass, Gea. Werke, Bd.
1,

HmP>l, HtnP<l.
185.

p.

79]

consequent
termfl
-i

nee

of

2(t n x*
/<

is

impassible,

sin,-,-

th.-

M
if

ineivases.

Kurth.-r.

0,

it

i-

OftSJ

bo

see (by an
a

argument ^imilar
1

to

linite

limit,

BO

that

On
or.
\\'e

the other haml.


ti'iiU
|

if

0>0,

it

follows lY"m
;

Kx.

:;.

Ar

that |a n

to /..TO as a

limit

so that convergence

may

prove next that


2
first

tradition (4) holds

and

< a =1,
h-

In the
vali;

place

ii

is

evident
|rt n
|

that

(at

least
//

after a cei-tain

jiience

decreases as
istfl

im-n-ases:

the

finally

.f

uly.
(5),

and

it

is

obviously convergent
J0n_
I'.

Kurth.-r.

we bave, from

|q-a,.

(<

n
i

1_I^ +

"A

|J|

^n

*n+l

!/lk +

/'a. -n+l

^Ik-H,

Thus
I

and

it

follows, as

in

Art.

IN.

that

the series 2|

a w+1

is
|

convergent
v,

\\v

have identically.

ami

ami eon-e.juently,
00

if

|a?|=

ami X
'

\^

litlen-nt

t'n

-m

1.

the
fl

1
o

-"

conxer^es ami linn/


l>irirhl>

= 0.

This result
\\'e

may aU
(

.leri\e.l

t'rm

have "Mtaine.l no information as to the K-haviour nf 1 as a mat a :_,


>

<

it.

Lilt

the

proof

is

little

tedious,

ami we simply
limit

li

i-

to
r

unless

= 0;
:lty

should

in

proving

tlr.~

204

COMPLEX SERIES AND PKODUCTS.

[CH. X.

refer to Pringsheim for the details.* Of course we have already proved that the series diverges in the corresponding case when

an

is

The
Tf

real (see Art. 12). final conclusion is thus

_^n_

aH+1

= l_l_/

^_ VS*
.

>!) (\\<A)
(

absolutely for x cannot converge for as 1. a + i^, there are three cases: If \x\ = \ and
|
|

then the series ^a nx n converges

>

<

1,

and

fji

(i)
(ii)

when

>

1,

the series converges absolutely.


l,
\.

when when
As

0<a =

except for x =
(iii)

the series converges (but not absolutely),

^ a, the series does not converge.


we may
take the

(Weierstrass, Lc.)

Ex.

a special example

hypergeometric series

(Art. 12), in

which
an a n+l

so that

u
see that, for

Then we
(i) if

x
|
\

=1

the real part of


this real part is for .r=l
;

(y-a-fi)
;

is

positive,

the hypergeometric

series converges absolutely


(ii)

if

between -

and

0,

the series converges except


series does not converge.

(iii) if

the real part

is

or less than

- 1, the

Special cases of this have been found already in Arts. 12

and

21.

Further tests for convergence. will find no difficulty of Art. 18 to establish the theorem:
80.

The reader
// 2a n
|

in

modifying the proof

(1)

is absolutely convergent, so also is

2anvn provided
,

that
|

v n never exceeds a fixed

number

k.

L.c. pp.

easily seen to diverge) the behaviour of 2a,,

13-17; the proof shews that (except for /&=!, when the series is the same as that of

is

1-A*

"I"
And
the

(1-AQ(2-/Q 1.2

(1- M )(2-/0(8-AO

1.2.3
is

sum

of

?i

terms of this series

L^\
1.2. 3. ,.(7i-l)
2
?i,

)'"\

L + (i

Lz/A n-l)

It can easily be seen that the absolute value of this expression tends to oo with and that its argument has no definite limit.

79, 80]

IBEL'8 TES1
is,

<>i;

<

It

ho\\,

that
in

Ai

'.ill
*

need a
\\

little

alt t-rat

ion
ti,

to include cases

which

i-

complex.

shall

prove

9,

/////

nsider the
Writ-for
i.re\
;

modilied form of
note

Ahel'fl

I.-ii.

the
1

-um
to

of

tl

vn

vn+l

-f

t' n

+j

i',j+ 2

-f

>at

V -

l'n+i.

\Vc

1,

;
i

>+...+([T;
it'

-v

SO thai

Vw+1
is
1

<

:,

/'

//

>

//<.

N.\v (see Art.

Til) tli'

series -('..

2|vn
SO that

COM v !;] it.

And

>-}-...4-('\
'

'

,)='',-'
:

lias

<lctiiiit'-

limit
//

;/

MS

R-QC
.

and

h-t

n-

\vrit.-

G:n.
\v

=Hm|t; n

-'
.

iiat

if

we take

the limit of this

ine.jiialit
[.[
.

\v- tind

Q
Thus,
if

\' \^ we write }' = (!+!' It i- aNo obi lOOfl that the se.jUriir, injual to Beqne&Oe, and that !'- I'

al\\
i

than
.

I'

'..+i|
:>:>.

if

-"j' !-f
''.,)+

".,''.,+
^

+",''

^'e tind. as in Art.

that

Zn
Thus,

=
if

l(>\-

)+...+
than the upper limit of
-^
|,

S,VW
|*t
|,

// is not

Ie86

...

|w|,

we
v

tind

ITRFj-F^-f
is

which

the form of
if
//

A.bel'fl

n-...+(F-i-F Lemma for complex


\ijp-r limit
|

//r
:

term-.

Similarly,
1

is

m-t !*< than

tin-

ti

/,/

1,

ami

i/

than
.

tin-

..f * r -cr|, as r ranges ujijwr limit as r ranges fron.

from

:iul

i_

,;( I",

r,,,)-f

>/

206
If

COMPLEX SERIES AND PRODUCTS.


the sequence

[CH. X.

G = \imv n

(#)
,

and

Vn = vn

is real, positive and decreasing, we find that so that these inequalities are almost the same as

those of Art. 23.

Now
It
is

apply the

Lemma

to the

sum
in

then evident that this


is

is

less,

absolute value, than

H'Vm+i, where H'

the upper limit of

Now

since 2a,t converges,


e,

m
e

make H' <


|

however small
. . .

can be chosen so large as to may be and then we have


;

a m+l v m+1 +

+ a m+p vm+p
2a Bv n

|< e Vm+1 < e Vv


convergent. complex series.

which proves that the


This

series

is

may

be called Abel's test

for

a function of #, Abel's test also enables us to ensure the uniform convergence of 2a M v n at all points x within an area for which 1^1 and C/i remain less than a fixed value. For
If vn is

F = +*7
1

arid

0<

vj

tfi.

Similarly,
(3)

we can
vn

establish Dirichlet's test for complex series.

0, then Ztt n t> n will Vn+i\ converges and limi> n limits. between oscillates 2a n finite converge if In fact is here zero, and (just as in Art. 20) we can find

If 2

a constant

(independent of
m+l

m)

so that

H'

< 2.
of

Thus

which can be made


r,H.o.

less

than

by proper choice

because

in Art. 79. special case of this test has been already used

Uniform convergence. After what has been explained in Art. 43, there will be nouniform convergence for difficulty in appreciating the idea of a series 2/n (se), when x is complex the only essential point of
81.
;

novelty being that the region of uniform convergence now = +irj) q) plane (if x usually consists of an area in the ( It is also instead of an interval (or segment) of the real axis.

sometimes convenient to use the idea of uniform, convergence along a line, which should present no fresh difficulty to the
reader.

80,81]
Weierstrass's

\inl;M

<

ONI

l.i

'

M-test for uniform convergence can be


tl,

ret;i

almost

unaltered,

-f+o
)

irifliin
t/nif

>'

oerttwn

wren

.1,

'/"

/,,

is

//--

/</"/"

/'///

when

M
.1.

<><'/"

Constant,
''.^nltitiln

///'/

ift)

^.M
'

co
i*

a/nd uniformly

Abel's and Dirichlet's tests for uniform convergence (for a

com-

plex

are <>htained at
of Art.
4~>

Once from

tin-

last

art;

The proof I/ i') u

(1) can he easily

iiioliti.-d
n-'ithlii
tin-

to sh.-\v that
f}
,-'

IWU8 fnixt'"!'
I

"/
fluif

./

9epa/ra&

><ma

contfawHU
Tli.-

vn

f/

lis<MiwiiiM

<.f

(liU'.-ivntiatiun

and

int.-^i-ation witli respect

tn tin-

\arial)l<is

falls Ix-yoiid tint<>

lnt

it

ini'iital

plannotions ha\;' hccn made clear) the results of Arts.


|>raet it-ally

not

out

<>i'

im-utiou

tliat

BOqpe of this IxH.k: (wlu-n tin- fun-la4.">-47

remain
It

unchanged.
Art.
1^
i-

is

evident a No that
variahle.
i-

-till

correct,

when X N

c<.mj.le\ Tliere
lit

no ditliculty
valid,

in

remain

wlien
to

seeing that the two theor.M, re the functionc<m}>lex.


integrate a complex function with

It

is

often important

rasped to a real variahle: in particular it i- usrful to consider the mean value of a continuous function /(./) along a circle |.r = p,

which

is

defined hy the equation


i

where
of a definite
'
.

' ,

The existence
fr-'in

mean value

i-

inferred at
:

and

the o.ntinuity of /( JUM afl in Art. lti-"> of the Appendix the following condu>i -n^ are immediate consequences of
:

the definition
i

\\r i<>

ti.

if

.ii-tant,
.1.

(iii

<
'+
.

if

f(
(>.

<J
/

on the

eircle.
/.ero),

(iiii

\Vi'

if

i.

an int.-vr (not

because

'

sin /HI

and
!

Jo

00,

=(.

208

COMPLEX SERIES AND PRODUCTS.

[CH. X.

Further, from Art. (45 2), we deduce that if converges uniformly on the circle \x\ = p, then

We can define the mean value without using the Integral Calculus, by supposing the circumference divided into v equal parts at #t #2 ...av, and writing
, ,

Cauchy's inequalities
82. Circle of

This method leads to the same results as those just proved and thus can be established without the Calculus. (p. 209)
;

n convergence of a power-series ^a nx

From
vergent

Art. 10
if

it is

evident that the series

is

absolutely con-

ni
I

.x-i

and the

series certainly

cannot converge

if

because then a nx n cannot tend to zero as a limit. Hence, if we write (as in Art. 50)

(where, of course,
verges absolutely

is

real

if

<

and positive), the power-series con1/L l/; and cannot converge if \x

>

To interpret this geometrically, let a circle of radius \jl be drawn in Argand's diagram then the series is absolutely con;

vergent at

any point within the

circle,

and cannot converge at

Fio. 24.

convergence

any point outside the circle. The circle is called the and it will be seen that, if a n is real, the
;

circle

of

interval
circle.

of convergence obtained in Art. 50

is

a diameter of the

81, 82

<

U;i

|.|

Of

<

MI
<!!<

'h-

in

wh

with respect i,-i n case to which


t

we can
[nation b
in

\\VI.TM

raae'e

mlthat

diacu
tin-

Art. 7!.

It

nniM
OO

nut.

li

npjx.s.-il
ilitifs
B

three

tli.-

circle;

in<l<-c<|.

I'rui^h'-iin*
-

has
circle.

<

*in

n hidi convei

<>n

th.-

l.ut

not ahsolutcly.
Tinan.
I

ice,
-

\\

hen - "

rgent
the ciro\

i.

jx ti H

,//,,,/,'
\

of

th.- cirrlin-oiii

x =1
.17

i.l.-nt

th.-

6 c;in

at

|.IVS,-MI
"//
h-

only

-;i\

ti

iinii'onnly within

and

any

cirri.

=/,-,

\vhn'int

between
iiinr.-

<

ami
Art.

I.

\\V -hall,

fully Tin* iva.lrr will tind

in

lit

1.-

difficulty in

that the theorems


Certain

of

Art-

h<

>!'

for

cninplcx

|><

>\Vfr-series,

Mnall

\.-i-lal

;ilt-rati(inx

h.-in^

made.
-

uniformly
crm.

<>n

every

ciivlr

,,,

!'..i-

which'

j>

<

/.

we ran
t

ivnlily ohtain
>y-t

its

mean-value

de hy
.

intc-'i-atiiiL:-

.-i-jn-1

if

\^

\,

,-

60

tli.i!

kl = /.i/0
tto2 /" '/x >;ihi

nf

//..

Similarlv.

wis

lind

8[/<
Thus,
if
.!/

-u.;

the niaxiinuin \:ilu-

nn the

circle

./

=/>.

we

lia\.

<
.-ii.l

.V.

"

<M
ergence,

\!l

belongs

t<>

we may

of

>

La

210

COMPLEX SERIES AND PRODUCTS.

[CH. X.

Again, since the series

converges uniformly on the circle

03
1
1

= p, we

find

w^'
a/

Similarly,

we

find

shall now consider the question Can a power-series be determined so as to have given values along a definite circle, say \x\-\ 1 Let us write the coefficients a n in the form a n + ifi n where a n /3 H are real;
:

We

then write 2o n # n =/1 (#), ^iftn xn =f2 (x\ so that f(x) = 2an xn =fl (x)+f2 (x). Now suppose that when x = coa0 + i sin 6, we have f^x) = u l -\- iv lt f2(x) =u2 + iv2 = u + w where u lt v^ etc., are all functions of 6 such that U = U I + H.,, a,TLdf(x)
t

v = v l + v2

if |c|<l, we find as above (assuming the uniform conn = vergence of 2a n # on the circle |#j l)
.

Then,

and

In the second integral put \Jx for x so we have far

this will
c

change u l + iv 1 to

u-^

iv^

= ^-1

27T-/0

(u l

-w C-X de.
) l/ 1

If

we

subtract the last result from the formula for

/i(c),

we

obtain

/ ( C)=J^ STT^'O
t

Similarly,

by addition we get

In the same way we can find integrals for


nly essential change in the
.

/2 (c)

in

terms of
.*

w.2 ,

v2

the
1

changed to .', u 2 + iv 2 becomes -u 2 + iv 2 This, however, does not alter the final formulae /v.sand so by addition we see that these formulae remain true when the x are omitted throughout. Thus /(c) is completely determined (save for a constant) by a knowledge of either u or v. But, given an arbitrary continuous function for u (or v\ we do not yet know that it is actually possible to determine /(c) so that its real (or imaginary) part does assume the given values on the circle; this problem will be discussed in Art. 83.
is
;

argument being that when

/////'.

83. Abel's

theorem and

allied theorems.
Sa,,o;" ant
I

Suppose that |#|


tliat

=1

is

the circle of convergence for


imt

2a

rt

is

known

to converge, altli<>u-li

absolutely.

82,831
It'

ABKl.s
in

THEOREM
tind that
'/
|

'2\\

we take vn = j- n
hat

An

= lim|v n = 0, and

V = 2\vn -vn+l = \l\

Thus,

sine.in

'it,

the series Sxt^" will converge


h,<

unit'nniily

any an

for

which we

I-'
wh.

Ml-M),
1,

any assigned mimher greater than


i.

and of course
be NM

\x\<
'I'-,

(See &beJ

ted OB

p.

interpret this iin-,,uality wre

observe thai

it

may
i

p<\d-/->
where
1

or

(X-p)^g>
(see
(

./

=
|

/* (CMS 0-f

Thus

_-_)\ /J

__
/

32>X

n
In
this minlitinn.
.;,

-i)/,
li.-s

2(X

(l--2/,'-( s,/>-h f

oo8^t-

between
,;>-X),

~,

and the equation

-l)p-8
Bfl

(-^7T<0<i
(witli

tinr.

innrr
.u-'hly

arc
in

of

;i

liuiunni
'2~>

n>(l-

at

/>

ti^un-

I'm-

tin-

case X

-i.

It

is

easy to see that the anto the circle |#|

nt'

tintin-

lima;.m
lai

ajpn);ii'lu'v

thr

nearly

1,

taken.
1

Thus
//>/,

/A

Il'any rc^ulai- c-urvc i> drawn fmin a point inside the circle 1, then, provided the curve ruts the circle at a tc tlie ]>'im
./

/'unbioncnthtori
i

use a

wliii-h

in

our

n<>i.itii>n

wouUl be representetl by p=2oos<f>

212
finite angle,

COMPLEX SERIES AND PRODUCTS.


we can draw
is,

[CH. X.

curve
curve.

that

a lima^on to enclose the whole of the the series will converge uniformly along the
,

Hence lim 2a n xn = 2a n where x approaches 1 along any regular curve which cuts the circle at a finite angle*
This
is

The theorems
so

the extension of Abel's theorem to complex variables. in Art. 51 relating to the divergence of 2an

cannot be extended so as to hold for complex variables quite easily, because the lemma of Art. 80 gives less precise

and it is necessary to assume that the possesses some further property in addition to the divergence of 2aw For as a matter of fact, even if an is real and positive, Pringsheim has shewn that the
information than the
of Art. 23,
.

lemma n series 2a nx

divergence of

2an

does not ensure!


.

lim
|

2anxn = oo
|

for all paths defined as above.

The condition introduced by Pringsheim


divergence, which implies

is

that of

uniform

where an

and the point x lies within the limacon. from Art. 51 that lim *La n xn = oo The reader will find no great difficulty in modifying the
It is then obvious
.
|

>

proofs given in Art. 51 so as to apply for complex variables when Pringsheim 's condition is satisfied.

We proceed now to find

the limiting values of the integrals, given at the

end of Art. 82. For example, suppose that v is an arbitrary real continuous function then the second formula gives a value for /(c) which can be expanded as
;

a power-series in c, convergent if |c|<l. We shall now prove that if this function is denoted by U+i V, then V tends to v as c moves up to any point on the circle so that we have determined a power-series whose imaginary
;

= l. part has an assigned continuous value v along the circle |c| Clearly it is sufficient to establish the result for any point on the circle
so
*

we

shall calculate the limit of

as c

moves up

to

1.

Picard, Traiti d'Analyte,

t.

2,

p.

73.

t For consider

Af
7

= A'P-^ (cos 20 -

sin 20)

J,

where E(x) =

*.

There

is

no

difficulty in seeing that,

the coefficients a n are positive and that


cos

when this series is put in the form 2a n diverges. However, if $* <0 <
-,

^ir,

20

is

negative,

and so lim E\ p->o

L(l -#)

J|=0.

83]
It

POI
will

UL

be seen that
j-

+ c _ 1 - 1*+ 2ir sin (0 - o>)


<u

where now

.r=co8<o + t sin
'

and

,'

from Art. 65,

->+
and we note further
value of the
ti
:

K'-H-H-i,
is

int.-jrati..n

positive,
less

the
1.

(km over any


f
'

x//<.
>

////

ran^e must be
find
.

than

f>r
>

<>,

we

(,'I

)(l-rV<o -2r 008(0-0))+^'


u>,

and sinee

is

a eontinii"u

Min<-tiou of

we can determine a

so that

Th*
.I

27T\Jo

Iff*.
j
-

= 2a to w = 2rr-2a; here, next to consider the integral from // that is not i.>) greater than cos a, and so
1

>)

while
;

+ /- ^1 - iV e, ,s L-fJ <2(l-r>
1

-f

/-'

2 jT sin a

he upper limit to the values of \V\ on the eiivle,


n

we have

(v-v )(l|l-2rcos(0I

'juently
1

^ftr-fc 27rj2a
It
i->

.-^[(1-r
1

-in*a
t<

'

th.-i -i-t".

p(.--iMe

find
if

a,

<-..

so that
1

|V-P |<,
that
is

tf'<u, and

r<8j
'0

to sa\

moves up towards
tegral

tO and
still

-;i\rs

id

tlie

preceding

\\

did as c
1,

T-> deal with appioa-he-, any point on the eiivle exOSpl 1. <> ha> the limit /. when that tlii'-'iiirh p.iti\e \a!
.*

suppose
limit

tin-

ne^al

i\

e
I in ^.
-1
\
.

it"

We

\\

V=V
it

iit

evident fioin
the
\

becomes continuous at
.=0.
I"

o>

= 0,

if

we

assign

to

Furth-r.

is

indicated in the diagram on


p.

wluMe

<:>

rej.ie-ent- the

same angle as

:M1.

214

COMPLEX SERIES AND PKODUCTS.


from the
first

[CH. X.

Now

case considered

we have

lim V'
(r, e)

limv' (w )
7T
<f>

so that

lim
(r, 6)
</>

V
as
(r,
Tito,

where

In have m=-l,

6) approaches 1. the particular case when v is given in the form 2an sin and then the result is sin

is

the limiting value of

we

shall

nO=
;

2a n cannot be convergent were convergent we should have lim 2 nrw (cos n + i sin n 0) = 2an (M) in virtue of the extension of Abel's theorem (at the beginning of this
It will be noted that in this case the series
if it

for

article);

that

is,

lim
(r, 0)

2a^sin*0 = 0,

which
84.

is

not the case.

Taylor's theorem for a power-series. have seen that a power-series ^a nx represents a continuous function of x, say f(x\ within its circle of convergence

We
|

11

x
|

=R

let

us

in h.

Draw

now attempt to express f(x + h) as a power-series the circle of convergence, and mark a point x

FIG. 26.

inside

such that |oj| = r; draw a second circle (of radius with centre x, to touch the first, and mark a point x + h r\ within the second circle. We shall now see that f(x + h) can be expressed as a powerit,

series in h.

In fact f(x + h)

is

the sum,
l

by columns,
2

of the double series

+a x+ a^ + + aj

ag? +..,

83, 84]

TAYLOR'S THEOREM.
is

215
ln-raus.-.
ii'

Hut this series

term by

its

absolutely convergent, absolute value, we ^.


t

r.-pla<-<-

tl.

\vh.-iv
liv
v
tli.-

\**\+\*i\(r+p)+\**\(r+pP+\**\(r+i = 'Muse Now tlii i>


fi
|.

mnstructinii
Tliat

ami
\vr

ti

'Intin-

<loull<- series

converges
l.y

lately.

i-,

can SIMM
(

loiible

series

rows,

willmut
H.'licr

altri

\alu-

Art.

/(/

+ /,)=/!
')

\\h

ft
l

+ '*i.,.rh3.4a4 o^+.,
f...,

so that these series

may
t<

be obtained

inmi

/<

l.y

applying

tin-

l'>nnal nil.- for SO


att.-nti..n
in
//

LOU, will

any
Series
It

th- nifanin- of the process.


!><

may
t..

rallfl

'/''//''

>^8.

in.

iy

ln>

usrful

ivuiaik that thr

.-ir.-N/,'
;

..f

h-

.-iivl.-

./

\v kiinw
it

.-..nvergence for th that it

f\-idcii.-f
.

that
*

may

not rxtcixl further.

it

is

<-a>y

t-

that

it'

\\

writ.-

...,

th,n

-M
mgea
it'

J
|A|<|1
\/t\<
-

Thus the Taylor's

series

in tini

_.!!<..

w.- !M\.'

tli-;-

>)mded area ii >wei wi


i

i<

rh

II

\M>
iln

I'liohl

CTS.
M

|.

inn.

ii.

.n
i

/(

<

I..-\MIM|
I.

I.

he area of
111

nri^inn.1

dnfinil
"I
i

niM-.il.il
.i.|r

Wrmi-MtMWl'H
!!M
../
.

Ilifiv
I

KlIlirMnllM,

l"il
1.

in

IM

.l.l.i

IM

.Mil

.nil
/,,

province
(Ii,'

(If

MI

mil

nil
I'M

(ill

IP
I

.'Hid

Morlry'H
,nt
..
I

Illtrixilli'd'nii

'/'//<-,<///

.\inlliiti,'

l-'niK'tioii*

(In

\\V rnn
I

now

ol.t.nin .in

rxlrndrd
Mir
-..

1'nnn

<>f

'nnrli\
n|'

':.

Ait.
i

S'J).
i.s

nnt

Suppose known,
ill'
,>'

Mint,
but.

id

l.ldill',

r. ,n

nr.j u.dil 1<


< i

i.-,
..I'

Mint

|aj|

r
h

in

Known
Mint
I'ol'

to
nil

be witlmi
pninln
I

Mir riirlr
tlir

rin\ cri-rnrr
I

.Hid

I'tlillirl
)

nil

cirrlc
I

/',

tlir
/!/

//

srric

/(.'

roiix rn;r;,
/(,/
|

inn

>t

in

<n

|/(.|BBM.
Hiicli

Tlirii, if

is

t.ln'

inn xiiiiiini nf

It)

for nil

points
in

that |u3|"r,

|A|M|

wi^

lui.\'(^

by applying Oaucliy

r.|n.dity

to tlir /^-series

,M_ M
H

HnN
srrirs for
,

\ppl\in
"'
ll

.
:

Mir
\

,;imr
I

inr<|iiiilit
;/

y I" Mi"
I

.'

/;,<

.-,-

(m

n)(m

!)...(///

I),

I/

,17

'/'////

mi

<>/'

"

(r

f-N)'

|(^

<(

/(
|

or/-//

term

ia

lew

Until

,!,,<{

Un'i'ifon*

(r
It

N)

|,/.

|< (/ ,

,.

n
<>/'

I/.

I'olloU',

tli.it

UK- r,l</ills of

,;>i,r, I</<-IK-<

i]/,,."'

tO
'I'll!

(f-f>
.

lr;ld

.it

to tlir Mn'O|VHl
cii'i'lr

MlJlt-

////'/r

ix

(if

Intxt
r
-;,

<K'
n,
s
.

liifiilur
is,

[mini on UK'
point
in

of

<-on i',T</f
i,f

lift

oj'n

fioir,

Mir nrrdd,,, iirlio.>d

which
N|IC\\
t'H

Taylor'', Minirrin

' i

li.i

ftlsO
/

ir.rd

this
'

rcMiilt.
(i

(,(>

Ili.ii

/In'

<-ir</<

<-/

con
tin

i'i

i'</<

iifi
../'

of UK
///.

of
is

i<
/<>

Is ritlirr
/>
/,

UK

siinir
"/'

Unit

oi-K/tiKil
irliirli

UK

:<T<>

,/,,,
In
l.i'
I I

Iffifl
IM

iH'itrrxt
'I
il

UK' orK/iii.
I

.n

......

Mill

..I

Ii. -\\

Ii

il

il

i,

I..

m.
/.'

in.

HIM

.HI.
I

iiiiinmmii
|

\.ih
'

Inn
i

mid

oil
if

.in\

n.

I.-

A',

tin n

In//
/

\\

ill

,,u\ t'l'^c

I
( ,

Mild

ll.n.l

.....

,|

//,'

(/

I.

II

I.

M.

.111.1

M,.|

II

-'
.

\l
i

|H

'

'

/'..

Lond M">
IK
III
I

oL

M
I

.....
.

i-

186
IM

to
l.-Mil.
.1
I

II.

\\

VlirillllloM, IIIUI <>f

M|||

,,

.Ml

..

,lll\

II

Mill

I..

84, 85|

r,

IK

'

rf //
<
,

//

O6M,

ill-

i.nlniH in NIMMI

I"

I"-

IK.

ill. 111

\sln-M-

/<

86.
'I'lp
-

The exponential power-series.


ilillimlt
\

in

MiMilir\ in"

In
|

.1

In.. (I

)'

X
!
I

J?

^iin.
I
i

l'.\

iniill

ijilir.it

i.

.11

o
liii'l

th.M

Oi

An
In

IT

A in.

ii

hiii'i.nii. ni

.1

equal

.1

Li

A kind of

li

pM

irith

//

the

"""

1,'iiuly

ln.i

tin

'"lulu KII
in
1

require
1 1

in'.

equal

ion

bo

ho

i!

.'I"'

1.

<

/,

\*

ll

IM.||,

I,,

,,,//(,

218
then
or

COMPLEX SERIES AND PRODUCTS.


/()x/(0)=/(a)
so

[CH. X.

Hence a n = l, and

Again,

and

if

we take

the limit of both sides as y tends to zero,


f'(x)
(3),

we

get at once Or, applying Art. 52

= a f(x).
l

we have

and

since this equation


(

must hold

for all values of

in the

interval

^R, +%R\ we must have

a 2 = a:2

a3 = a 1 a 2

a4 = a 1 a3
a

....

See Art. 52
. . .

(5).

That and so

is,

a2 = ax2
f(x) = 1

a3

a-f,

= a^
.

a n = a^\

+o
this
;

xz

X3
satisfies all

We

do not

know from
it

argument that f(x)

the

conditions of the problem

but

we

see that if there is such a

power- series,

can be no other than

E(alx).

Now

E(a^x) does

satisfy the relation

E(a,x) x
for

E(a1 y) = E [^ (x + y)]

any real or complex values of x, y. Consequently our problem has been solved;* and
is

where a l

the coefficient of x in the power-series for f(x). and in many respects convenient, to write e* for even when x is complex. But it must be remembered E(x) that this is merely a convention and that in such an equation
It is usual,
;

as

e^ = i

(see below, Art. 86)

the index does not denote an

ordinary power.
*

It does not follow from the foregoing that

relation f(x)

But, if f(x) has the exponential form.

no other function can satisfy the xf(y)=f(x + y), because we have assumed f(x) to be a power-series. we assume that f'(x) is continuous, there is no difficulty in shewing that

85,86]
86.

SINK
Connexion
tin-

\M'

<

219
the

between
./'

exponential

and

circular

functions.
in tinntial B6f n.mplex \arial>le variable on a mil /, we may differentiate term-by t-rm depends \\ itli respect to /, and obtain the same formula as if x w<
1
1'

-i

that

"rr->jM. mlin^ in a

.lian.u'*'

''

in

',

<

hangea

A.,,1
..],

ls.
r,

M 8<-*0, &r/& approaches the lim


be la-t iiM-|iialit v
li,n
( i
:

.<laoit

foil..

;-,

!jA',,
t!

In pai-ticular.
/'//.

.siip|.ns,.
:

puiv iina-iiiary and equal to

wli.-n-

//

is

ival

thru \\c

or,

if

/:<

w;)r(oo80-f

'-i

we have

euj

-*-L
tn|
.,1'
//
;

Thus and ft q are and so r=l, = o.


/

ind.-j.endent

l.ut

t'.r

//

Hi-lie.- L^-M. -rally

r=

1.

;/,

and SO which

j-

'sin

;y.

is

continued

l.y tlu-

ivinark that
-;;/)
/
' /

A'"V
Another
>l.scr\in^ that
J

)=^(0)=1.
r.-sult
i-

m.-tlidd

of establishing

the

last

^iven ly

sin

;;

= (cos^>-|-/
lim

it'

N"\v write

cos^-f isin#=

-f

and we

see that

mra ;/,

220
because

COMPLEX SERIES AND PRODUCTS.


lim (n sin <) =
r\

[CH. X.

lim (sin </<) = q,


lim
[(1
</>)/</>]

and

lim n(l
cos
r\

- cos 0) =

Hence

+ i sin

r\

- cos = 0. w = lim (1 + n ) = E(irj)


r\

/c

by

Art. 85.
another method, analogous to that of Art. 59, can be used.

Still

In

fact,

write

Then
But,
if

^
y = r(cos

{yn _^

an d Ji

_ i( GOS + i sin 77).


77

we have
or

^| =
Hence

Lrii

+ r'(^)

^ ~
drj
'

drj

and

3/0,

#1,

...

are all zero for

77

= 0.

Hence we

find, if

77

is

positive, the

sequence of equations

<
Thus
If

<
lim
71

?/

M =0.

><K

we

substitute i^ in the exponential series,

we

find

and

so

we have now a new method

of finding the sine

and

cosine power-series (Art. 59). If we write r\ = \ir and TT,

we get the equations E(Tri) = i, E(7ri)=-l.


in Art. 85,
ir*
t

Using the notation explained


cos
r\

we may

write

+ i sin = e
r\

and changing the sign of


cos

rj,

we

find

r\

i sin q

= e~
r\

ll>
;

thus

cos

f]

= (e^+e~ '*),

sin

-.

(e^

-e~ ^\

86, 87]
\\'e

WX3 MMTIIMir
have
at

I'l

VI

'2-21

pivsriit

IK.

d-finiti
;l

<-O80;
<-,,n\ .-nieiil
1

and
real.

-in

'

wh'-n

and

to detin,- tli.-m
is

by
the

tin-

JM.V.

already
I

est ahl ish.-

when X
i

Then

e.juai

COS

''

('"
.,

are
It

tni.-

I'm-

comj.lrx valurs of X as well as


also
that

i-.-al

inll(,\\s

any

ri-<

un nn-t rical

i'<ii-iiiul;n-

which
Arts.

depend only on tin- addition-theorema complex \ariahh-s: thus in particular


re

r.-inain
tin.-

unalt<-r-il

formula*-

of

main vmte x
-in

tru-.
:

///.

it

will
/

Ix-

seen that

COS X = COS
,'

cosh

;/

sin

^sinh
,h

;y.

--in

fcoah

;y

-f

;y.

wh\\'-

cosh/; = A(^4-r

l),

sinh

//

=
<!'

\(<

"-,

shall not elaborate tin- <l-tails

inl

cosh

functions:

th.-

results can

\\n> analysis of tin- sinh he found in many t.-xt

V instance,

C'lirystal's

is

to be noticed that
sin
|

Algebra, ch. XMX.i. n-hen x is complex, fl"35


1

>

//"/

/////// '.<<

as

|<

1,

|cosic|<l

are no longer

vulii/.

\\V can. liowever, replace them hy "there,


*
.sii.l,

= x

and

so, if

|#|<

1,

we have

|rfn|<|
Similarly,

we have
^ eoab r|;

and.

if

\x\

<

1,

we

find

87.
\\'.'

The logarithm.
have already seen that
/.'i

if

//

is

a real
s i"
; /-

ai

'/
y

008 If -f
(

'

Henee

if

is

any

inte^.-r

j.o.sit

i\

..]

n.-^ative),

AVJ, --;>=!,

222

COMPLEX SERIES AND PRODUCTS.

[CH. X.

and

since E(g+ir}) the equation

other than
It follows

=0,
that

ri

= e(cosr]+i8inr)) there E((+irf) = \ = 2mr.


t

are no solutions of

if

we wish

to solve the equation

E(y) = x,

so as to obtain the function inverse to the exponential function, the value obtained is not single- valued, but is of the form

y=y
where y
is

+ 2mri,

(n = 0,

1,

2,...),

one solution of the equation.

Q*
If

O
FIG. 28.

we

represent x geometrically in Argand's diagram,

we have

x = r (cos

+ i sin 0) = rE(i6).

logr is the logarithm of the real number r, defined as in Art. 157 of the Appendix, we have r = E(log r), x = E(logr+iO). and consequently Thus we can take y = logr+ i0, and then the general solution is 2,...). (n = 0, + 1, y = \ogx = \ogr+i(0 + 2n7r),

But

if

We

define the logarithmic function as consisting of all the


;

and we can specify a exponential function one-valued branch of the logarithm by supposing a cut made
inverses of the

along the negative part of the real axis, and regarding x as prevented from crossing the cut. Then we shall have

= log x log r + i9,


With
is

where

TT

<C Q^TT-

this determination,

logx

is real

when x

is real,

which
should

generally the most convenient assumption. be observed that then such formulae as
\og(xx')

Hut

it

= log x + log x'

can only be employed with caution, since it may easily happen that (0 + 9') is greater than TT, in which case we ought to write = log x -f- log x' 2?ri. log (.''./'')

The reader

will note that for

two points such as P, Q

in the

diagram (Q being the reflexion of


of the real axis),
1

in

the negative

half

i 1 1

(log x p

log X Q )

= 2-Tri.

87, 88]
at

LOG \UTH.Mh
repi

the

rut.

ill'-

I.

ranch

sejrcted

for

log X IB

oh\

jollsly

continuous over

tlir

whole

plillie

of X.

88.
\\'r

The logarithmic power-series. know 1'n.ni .ind lii', that


v
>,

if

is

real

and

\X

<

(1)

//

= .i--

A.i-'+'.r'-...

represents the function

inverse I" the exponential

function

In other words.
(

if

we substitute
;icc. .I'din-t

tin-

series (1) in the

*J

.and then arrange


1-j-.'.
is

must le

and. as such,

powers of X, the result* Hut thi- ran-fonnation is merely al^el>r;i equally true whether X \^ real or complex.
to
.

Since the series c2\ c]\

i.solutely

f >r

all

values of y.

the d'-ran^eim-nt implied in this transformation is legitimate (see is Art. :>ii). provided that the series al>olute]y COB
(

Hence,
ipiat
i-

if
'ii
<

|o^|<[l, equation (1) ^ives one


'J
i

value of
real

>/

satisiyinu
./

ami
the

furthei'.

from
the

(1).

//

i-

when
in

is

real.

Thus, usine
artic!--.

hranrh

of

In^arithm

defined

the

we have
,

+.,)

= .,_

^+., .:;_...
p>

(if

|a.|

<

i).

-1

From
Wh-

the ti-ure.

it

is

evident

that

this equation

--j\es

This result can he

at

-7T <</>= +~ once continued


(if
<

^ tM

'

Al

l,y

reference t>

Ar

where we proved

that

< <
)'
' J

'2tf+

Jr'cosSH:;>

arctan
j

^'

^=
r(oOfl H-f-

,-sin

Hf))

}.

r-sin ^(9+^/ J sin

If

we

\\

me

X
:;

sin

in

the poi
^

./

-...-,
-f

+ |rcos:{H-...
up
to,

/v/'sin^-^. /-sin

l^+^sin.Stf-.
i-Diuputation

Ml

is

a gooii

e\>

i.y nuiiu-i -ir.-il

say, a*.

224

COMPLEX SERIES AND PRODUCTS.


p tan
2

[OH. X.

and obviously

= 1 4- '2r cos 6 + r = r sin 0/(l + r cos 0).


2
,
<

Thus our
except that

results

are in agreement with those of Art. 65,


see that

we now

actually lies between


TT.

\-K

and

+ JTT (because r < 1) instead of TT and We shall obtain an independent proof of (if log(I+x) = x-x + x*-...
2

the equation

a?|<l)

in the following article.

The

series for arc since

and
1

arc tan

a?.

Again, by

Art. 64, the series

y =x +2
,

xs

+ 2T4y + -3"
, ,

3 x5

represents the function inverse to the sine-function (Art. 59).

is

1. for real values of x, y, such that x for all values of absolutely convergent y,
|

<

Since the series (4)

and the

series (3) for


is

aj|<l, the algebraic relation between these series to persist for complex values of x, and we can

now

seen

accordingly

write
1

xs

3 x5

..

since the series (4) is taken as defining the sine for complex values of the variable (Art. 86). Similarly the pair of functions
(5)

y=x

are inverse to one another for real values of

x,

such that x
| \

<

and we may therefore write


arc tan
a?

for
b

complex values of x
...

a?

\x*

+ \x

(if

|#|<1).

In these equations the values of the inverse functions are determined by the conditions

88, 89]

BINOMIAL SERIES.
The binomial power-series.

225

89.

/(, x).
\vheiv hoth
.lit
j

+ ,* + ,< ,

C+
.11
1

.,

,.

2)+
is

;nil

in.'iy

r,

>1*
j

Th- -rieS
hav,tin-

absolr/

when |#|<1, ami thus we


/(,.

il.-ntity

r>x/<.',*)=/(, + ,>)
tin-

(|*|<
in

i!'

we

pick nut

meHiri.-nt

of

&

th- product

/<**)x/(,',a>>
it

is

seen by the ordinary

of degree r in

lth

and

rule (Art. 54) to be a polynomial of .r r \\\ /'; thus tin- co-tlici-nt

,X)-f(+l
-o a

polynomial
,.
j

of the

same de^
because
,.

Hut.

when

are
.

(my two
(

int.---.-^

and
is

cniis ..ju,.ntly ~\

must be identically
.V,.
'2.

M186,

wh.'ii

j'

any a^i-nr<l
values of
i-

int-^-r.
I.

is

zero for an
to oo
).

infinity of <liti'<Mvnt

(namely.

8,...

Thus, identically,

x)xf(v',x) =/(r+i

'..'),

(|

*|<1).
in

From
til

this

relatin we can apply

the

method indicated

c2) to

prove that
,.<)

= (!

-ho'

when
I'.ut

i/

is

a rational number.

to deal with the case of


ditl'erently.
in
i
:

complex values
place /(v,
a;)

somewhat
as a
1-y

In the
for

lirst

columns

power-eeriee .f the

t'd.

<.f j. w.proceed can be expressed x) can be regarded as the

doul.l,-

961166

1-f.

fj

''

-f

...

11

<

-f ....

ua

226

COMPLEX SERIES AND PRODUCTS.

[CH. X.

|i/|

Now this double series is absolutely convergent because, if = and |a;| = the sum of the absolute values of the terms
/

a;

in the (p

+ l)th

column

is

~~x/,
which
is

the (p + I)kh term of the series


if

/(i/

a?

);

and

/(i/

o?

by converges The double series being absolutely convergent its sum is not altered (see Art. 33) by changing the mode of summation to
rows, which gives
f(v, x)

x <^

Art. 12.

where

= l + vX = x ^x
x),

Thus, since /(j/, #)x/(V, x)=f(vd-v, above, and deduce that*


/(,, x)

we can apply

Art. 85

= E( vX

\
i/

In order to determine

lt

let

us write

= l,

which gives

Thus
x

is

is real,

it is

a value of log(l+x); and since the value defined in Art. 87.

is

real

when

We

have thus a

new
(see

investigation of the logarithmic series.

Hence
fig.

X = log p +
l

i(f>

29, Art. 88), so that

sn
where
a result which
v = a + i/3,
is due to Abel. The investigation above is based Goursat the on (Cours d' Analyse Mathtfmatique, proof given by

275).

indices

The method given in the example of Art. 36 (p. 89) applies and the following method was suggested in 1903 by
;
:

to

complex

Prof. A. C.

JDixon

The

relation
f(v,

/(i>, .r)

x/(i/, x) =f(v + v, x)

gives at once

*?-[/(i

*)J-0 +f )"
of that article

Sa7

where n
*

is

a positive integer.
v

Of course

corresponds here to

jc

and

A'j

corresponds to a^

89]

BIKOM1

\i.

('

>(-a(-*)*
-=)('

-]
its

the aeries in square bracket* has each of


lll.lli

terms

less, in

absolute

tllf

'.n-t^|)Ml|ili||^

tf|

+
;i

convergent

series,

to oo can be

independent of n and consequently th- limit found ly taking tin- limit of each term
;

Hence

)-v(*-

But

(Art. 86)

lim(l

+ f)" = /;(./'),

if

.r'

= lim ().
[

Thus,

/(is #)

= lim(l +)" =

log(l

+.i-)].

Tin- di>eu-M<n -i\,-n alM>\i- appli--. only to points within tin*


o5|

= l.
c-an
>t'

To
i-.-i'.-r

exaniiiif

tin-

ire

back to

WetarafcasB'fl nilf (Art. 79);


s.-i-irs,

convergence at points on tinbut on


\\<-

tinimportance of the binomial an indt-pciHh-nt treatment of case (ii). give \\\ have at once
tint
i

shall

<-

^-*
i
i

-.
i

-"I
\vhere
i<

I~

T~~9

[(*-o?+/fi

= u4;

-;

and

|o)|

<

.1.

Thus the
a
is positive

aeries

converges absolutely

the circumi

and SO 2a
'Ilm-.

up
ff[vlog(l+flj)]

to
i-

and including the


also oontinaoi

imference.
nt

aince

that
C)
,if

>0)

at

th- cii-cuinl'crrnce |a;|=l. points (>n th.thT hand, ii' a+l=0, after a certain stago the absolute \alues of the terms of the series never decrease, and
till
///

so

fd
1
;

Lverge \vhen
liut
tlu-r.is

|a;|

= l.
tli.U
(it
:

no

ilitliculty in MTJIIJ;

liin

A'[

For the absolute value

of this exponential is

p*~^, where

228
Again,
n->*>
if

COMPLEX SERIES AND PRODUCTS.

[CH. X.

-1
also

< a^

lim|an =0;

0, we see from Ex. we have /(I, x) = I+x, so

3,

Art.

39,

that

that

(l+aO/fo x)=f( v +l,If

x).

we

take only the terms up to xn in this identity,

we

see that

where Sn Sn are the sums of the and /(i/H-1, x) respectively.


,

'

first

(n+I) terms

in f(v, x)

and so

the real part of j/H-1 is a + 1, and is accordingly positive ' follows from the previous argument that Sn tends to a definite limit as n increases to oo
it
.

Now

And

since

lim
n\
n

=
is

lim a n

= 0,

it

follows
;

that lim(l+oj)
%->oo

definite

and equal to E[(v+I)\og(I+x)]


f(v,
n-><*>

thus, unless

x=

1,

we have

x)= lim 8n = E[vlog( ! + #)],


1.

and the series converges on the circle, except at x For points on the circumference, it is evident that
,0

= 200810,

= ^0,

(-7T<0<7r)

and so we have
f(v,
a?)

= (2 cos J0) ae-^[cos{ Ja0 + /31og(2 cos

1. provided that a For the special value

>

1,

we have

the identity (see

p.

152)

It follows, as in Arts.

42 and 61, that


|^n

but that
90.

Sn ->0, ifa>0, Sn oscillates if a = 0.

Koo,

ifa<0,

Differentiation of Trigonometrical Series. In some cases of interest, it is found that although the series

f(x) = 2tt ne
is

inx

= 2an (cos nx + i sin nx)


coefficients

uniformly convergent, yet the series of differential

If this occurs, it is often possible to obtain ceases to converge. the value oif'(x) by differentiating the series

89,90]

MITI:I;I:M

\Tiov

This gives

pn>\ ided that

the

s.-ries

^/H/
n(o

")'

+8(a ^ uniformly
interval
1'rnin

-...],

c.,iver-

linly
fc

the 6886 (in


, ,

any
is

\vhirh

= 0, 2-7T

are excluded

if

/,.

>

a r.-al positive
,

Lence, or in

any

int.-rval

ii'

l/<

"

an

is
|

decreasing convergent (see

Arts,
It'

n
\ve substitute

the value of f(x) as a

seri-

re liml

th,-

eqaation
,,!-<
In
i

1.%^
it

-(",-!

'.>'

-...].
6

].i-actic.-

will usually
first,

found best to

litl

the

for f(x)

and obtain thf /"/////

(Mjuatinii

which

and 4 l>y (1 as if the differentiated of JMW.-I^ rrangemeni accinlin^ This rule will be seen in ('haj.ter XI. Be were convergent
is

to

int.Tj.i-rtcd
t

l>y

multiplication
.

>

to'

be more than an accidental coincidence [Ar


It

md

UO(2)J
is

easy to see that exactly the same method can be


-'
<

if

necessary, to .-staMi-di a similar rule to interpret the 91 oiwhere + +' ly IIMH- the 1'artoi>

i.

is

;iu

integ

Another process, which in practice U almost the xiine as the L 1, foregoing, is due to Stok.-x <.!/.////. i///// /'////>-. /'"
pp.
t

D);

the

tirst

case of Stokes's rule

may
-1>

rcdue,-d

the form

where

.1

i-

Constant

determined

liy

the

e<iidition

that

i- e,,i,

and another case can be


i
.-

writt.-n

-i.

B]<

-/>'

liy t-.|uival-nt
.

tliis
:>:.!

luive

Item

pven by Lerch (Ann.

dt,

(S),

t.

!_'.

lv:.. p.
t
i

,m.l r.n-nk
,,g

<

8,

I'.'M;.

p,

^7).

H, lt

process seems simpler, Uotli

in

}u-:icti.-c

;unl

in

principle.

230
if

COMPLEX SERIES AND PRODUCTS.


is

[CH. X.

a constant which makes the series on the right convergent. These results of course depend on the fact that

in virtue of Art. 65 (see also Ex.


Ex.
1.

1,

below).

Consider

f(x) = e +
ix

This gives the formal equation

/()-t'(*
which leads to the
real equation

(\-<&)f(x) = i(**
or

/(#)=-(*+ cot
Thus
f(x)
for X=TT,

#),

0<.r<27r.

- log (sin J#) - \ix + const.

But
so that in

we

find

/(7r)=-log2,
sin J#)

f(x) =

- log (2
65.

+ %i(ir-x\
that

< x < 27T,

agreement with Art.


Ex.
2.

Similarly
ix

we can prove
. . .

+ ^tx + fypx +
if

==

J log ( tan i^) + i?,

< <
^P

TT,

and

ete

These give,

cos
sin

x + } cos 5^ + ^ cos 9^4-

. . .

=|TT

- j log(tan

Jo?),

# + J5

sin

5# +

sin

9^ +

. . .

= |TT + J log(sec # + tan #).


ginx

oo

Ex.

3.

Again consider

2i f( x ) = _,t

-7i

Here we get the formal equation


f'(x)-itf(x}=-i^e**,
00

which gives the

real equation

[/(^)-i(/(.r)](l-6^) = 0.

Thus

where A does not involve we obtain the result #=7T Putting


itx
,

f(x) = Ae

#,

and x ranges from

to

2?r.

or

^e

ip

=a

__. + _._..

i=s7r cosec(7rO (see

Art.

!:>).

90,91, 92]

BINE
:

\M>
"',

'

06IKE PBQ01

<

Th

\ilim:

l-y

we get

'

-coeff-iOf ~ 2< > 7 _


sin

--

'
/

i -

n(t-n)x

91.

The

infinite

products for
Art.
in
a;
/

and rosa.
complex values of
05,

The identitie- "!' and we deduce, as

(i!

ivinain true for

Art. 70,

sin
?i

.^'i
KT

sin 2 (a/n) 1

sn

Now.

BU

to t.-ml

-an

always

t-nsuiv tliat

greater than

\x\ t

and so Art. 86 gives

and, since

<
sin

sin(r7r/?i)> 2r/n

(see footnote, p. 184>. : n 2/ r n ,. I iC/ TV tUU II J


v

Hence

^ ^ N

Q
'

~
-2.")

L r2

l/r */ I*
'

and consequently we can

in

tin-

thron-in

.1'

Art.

P.'.

Hence, as

in

Art. 70,

we

h'ml

r =i

In

tli.-

^anif

way
C( )S

\\-

find

./'

11

-Br=
-/
,

92.

The

series of fractions for cot


r
'

tan
for

./
,

cosec

'.
)>

The investigation u v -\t. uded \sith'Ut difficulty


i

i"

Art.

71

real

angles, can

to a

complex argument. ly making

the

t'nl

lowing nnditicati"!
e,

\Ve ha\

of ooune, the idi-ntity

x
lie

n)
identity
is

merely an al^el.raieal deduction i'n>m


li-

trigonometrical addit ion-t


.nients are real
r

.\liieh

ai

\vhether the

232

COMPLEX SERIES AND PRODUCTS.


here, as in Art. 91,
sin (r7r/ri)

[CH. X.

Now

we have
\

> 2r/n,
.

sin (x/ri)

\<%\x\/n,

assuming that
Also

n> x
|

we have

(Art. 86)

Hence

2n sin
1

(x/ri) cos (x/ri)

and

ri

2
1

sin 2 (r7r/ri)

- siu
|

(x/ri)

< > 4r
\

"-\

-f

Thus, provided that 5r

>3

\,

we have
30|a?

2 sin (x/ri) cos

Consequently

we can

use the comparison-series

^
and so the theorem
further,
(
'

30|*|

of Art. 49 can be applied just as in Art. 71;

X T hm /n sin - \
\

n/

= x,

X V 1 lim / cos \ = 1, \ n/
(

exactly as

if a;

were real instead of complex.*


lim
(

Hence

Vn

- cot - = n/
)

and

lim

*(/n)<x*(x/n)
sin 2 (rTr/n)
sin 2 (x/ri)

%x

rV

'

x2
complex (except

Thus we have,
multiples of
TT),

for all values of x, real or

cot

# = - + Zv -Q X X2
,

^ T

2x
ri 7T
o 2 2

/ 1 =+ ^v 7 ~
,

1
TlTT

IN
)>

VaJ

7?7r/
r.

where

TI,

is

now

the variable of summation, instead of


:t

Now

the following identities hold

tan x = cot x
cosec x = cot J#
*

2 cot 2#,

cot x.
for the sine

These statements follow at once from the power-series


identities are familiar results

and

cosine.

fThe

when

.i

is

real; for other values, they

follow from the formulae obtained in Art. 86.

92, 93J

ill.

OTANQENT

8KB]

Hence we

find,

on

su)itra<-ii..n.

&c

^2n-l)

_*
.

^-o^'
4-1

2x

811

'

= cos* =
.

.e
<

to
.

+e
-r
'

-j ,'-"

=-,
1

ly writing

y2i0,

\v- M'- tliat

'

5^"i
93.

^tl.iVi-i v
for x

-j

The power-series
Beri<

The exponential

and consequently (as in Art. :>4) can expanded in powen "i'


!
'

th<- iv<-iprocal

function x
.*
|
|

''-

>

i-r'-vidi-d

that

<

p,

\\\.

2!"*"3!

+ "*= 1

'

This last

con<liti<'ii is

certainly satisti-l

l>y

taking

OK

l.y tai

p=
can
\\
i

Thus

\\v

,./'+...,
1

if

||<1-L>.

'

ii'.

At

'

n'nd

l-'- =1+ 2'


1

It

\vr

sulttract th.- lir^t of these ezpansioiifl


tin-

bom

the aa

we
s<-

obtain

idmtity r=

^3=-

-l-.

(>

234

COMPLEX SERIES AND PRODUCTS.

[CH. X.

Consequently we can write

*where
JBj,

""

?J_7?^_ +
J

_L R ^2+J
7?

5 5
2,

3,

...

are Bernoulli's numbers.

It is easy to verify

by

direct division that


8

A=

i>

= *V>

= A,

#4 =

#>=F

but the higher numbers become very complicated.* Again, from the last Article we see that

Now
x,

if

x <C

2-7T,

each fraction can be expanded in powers of

giving

ZnW
And
the resulting double series
is

the series of absolute values

is

absolutely convergent, since obtained by expanding the

convergent series

2laj|

It

is

powers
e*

of

therefore permissible to arrange the double series in x and then we obtain


t

L
is

which

now

seen to be valid for x


|

<

2-jr.J

By

comparison with the former expression,


T-k

11
J-

we

see that
1 -1-

V~>

T-

Q J X~>

1 A

T-

J,^ *"

K^>

and generally
*The numbers
<;i;ii.sher

(as decimals) and tlu-ir logarithms have been tabulated by to fiw are givi-n (Trans. Camb. Phil. Soc., vol. 12, p. 384); and l

by

Adams

(Scientific

0.) Qhryitel'i .\/,,<bra, tThat 'lir is the radius of conver^riirf may be smi tVoiu Kaker's theorem x (Art. 84); for the zeros of e -\ are y = 2niri, and the least distance of any of

Paperx, vol. Ch. XXVIII.

1,

pp. 453

and

455).

(For more details, see

these from the origin

is

L'TT.

93, 94]

l:l

I:\MII.I.I -

\i

MI;|.;RS.

Wo

S-i
Alt
6
It

ol.tain thus tin- n-siilts

(l)
,

\^

-^ v V
\\

'

Z^^c
hen x

^ V i
is real,

1 -

^~9450*
liav-

is

instructive to notic,. that.

we

any

value of
**
2
'

^+

(2.2 4nVJ o
.

l.y

ail.litiMH.

we
tintli-

see that
srrirs

i-

c^ ~~

n-].ivs-nt.-l
is

\>y

th-

(r
tin-

+ 3)

terms of
ti-rin

with an error which

l-ss

than

following

of

For instance,

for

any

ival po^itiv.- \alu-- of

,/\

W6

h.-.

720'
an.l so on.

Ex.

By HUM

identity
1

WO

ca

liat

md
94.

Bernoullian functions.
r.-'i-nonllian

function of
in

the

th-

D of

'',-",'
whicli.
/

lv the

in

powers of

<

236

COMPLEX SERIES AND PRODUCTS.

[CH. X.

Thus we have

so that
.

<p n (x)

rn

~2

-l

^Q-l) D T n-2 _
~2~!
x

" 1 )( W " 2 X^ " 3 >P


41
.

*** x

~n-4

where the polynomial terminates with either x or x 2

From
the

this formula, or

by
:

direct multiplication,

we

find that

first six

polynomials are

= yz,

where
Again,
n (o?

y = x(x

\)

\-.T-'

+ l)

(j>

n (x)

is

the

coefficient

of

/n\

in

the

expansion of

so that
If

^ n (33 + 1 )

n (x)

= nx n ~ 1

results,

we write 03 we see that,

= 1,

2,

3, ...

in the last equation


integer,

and add the

if x is

any positive

n
which gives one application of the polynomials. Further, by differentiation we see that n'(x)
(j>

is

the coefficient

of

jn\ in the expansion of


t*e _,t(e*<-\) e'-l~
t

Hence we

find

94]

BERNO1
0',m(z)

LI.

\\

II

\< TI<

and generally
1
1

= 2m^,,
ami
/

-I.

+ <-l

1..

Change x

to

a;

to -/.

hat

$*aThna
Sim-.' 0,
i
i

='*,':;

i
<//>u
njuatinn that B
1

-,> = <-!
t'nllnws 1'p.ni tin- last
ia
.

<i

)--=<>.

it

i-

root of
tin-

'

(/>

>and thai
,

root of

<j,

and

a glance at

ti\- i'uncti"!

f/

v lraK to
lias

tin-

ronj.-i-nnv that

^,n (^)
-J-.
1'.

no root

"x
that
tliis

""///

'/"'

^oo<

Suppose
valu-

rouj.-cturr

IM-.-M

i-sialilisli-l

up

to.

then

since
<-c),

:(#)

is
<>

innnri-ically

at

./-=A,

and cannot
ran
I

\a:

en

and

1.

Consequently ^

('2/jL

:\

d
i

between havr no xrro lM-t\v'-n X -<> and '..and thcr.-1'mv non<- l>etween
/

Imn- ^

,.

Mid

I.

Thus
and
it

tinll

tlu'on-ni

has

IMM-II

rxti-n<h-d
tru-.

to tin-

\alu-

a-fl:

iin^ly
'

always

Tin-

indiratrs th- ivlations


illustrates tin- general

diagram and so

ar^p.n

^j''

It

will
1).

lr

s.M-n
<f>
,

thai

)-'

(in

>

l'";in#e

m (.'')=
,

<.
/

"/"/ t/tt

tin'

/x

// M/

O y

i).

238
95. Euler's

COMPLEX SERIES AND PRODUCTS.


summation formula.
if

[CH. X.

We

have seen in Art. 94 that

x and

are positive integers,

li

the polynomial containing 1(71 + 2) or J(ti + 3) terms. It is obvious that if f(x) is a polynomial in x, we can obtain the value of the sum

/(l)+/(2)+. ..+/(*-!)

by

multiples of the Bernoullian functions of proper degrees. But to obtain a single formula, we must utilise the Calculus and so we observe that we

the

addition

of

suitable

can write the foregoing -expression in the form

Hence when f(x)

is

a polynomial, we have Euler's formula*

/(l)+/(2)+. ..+/(*-!)

where there is no term on the right-hand side (in its final form) which is not divisible by x. However, the most interesting applications of this formula
arise

when

f(x)

is

rational,

algebraic,

or

transcendental

function, and then of course the foregoing method of proof cannot be used; and the right-hand side becomes an infinite series which may not converge.

To obtain the formula


integration

by

parts.*

in these cases, we shall apply the method of Consider in fact the integral

where

<

KViniMiilmriiig

denotes the Bernoullian polynomial. that </*_.(/) vanishes for /


is

<)

and

/-I,

\vi

tincl

that

this integral

equal to

*Seliwanoff, J}i/ereuzenrechnun</,

38, 39.

'

95
1

M.MATH'N

|,,|;MI

|. \

Int.-.natiiix

ly parts again, w-

mil.-u-ly

+0[^-a(0 + (-l)"^Hence
if

we

v,

.V

we have the

-i=(-l)'

')],

(:

Let us now examine the case n =


A"

we have
,//.

tli -n

Thus, rinoa

<MO = '*-'> we

have

-l

Th.r

:r'

Thus

*f

r'(.r)

//(o^4-.;L/'o--Hi)-r

!)-/].

and

fl-

uke

a succepi:itii.u>.
Tlifii
\\i-

/*-!,
A'_.

and add the

liavr.

if

\\

M->J. at

f>r instance,

1 )

+/(&) - ( /(')<ft +

* L'X'O +/"(")]

240

COMPLEX SERIES AND PRODUCTS.

[CH.

And obviously we can introduce as many terras as we please on the right-hand side. For a further discussion of Euler's formula, see Chapter XI. Art. 131.
Ex.
1.

If tyn (x) is the coefficient of

\n\ in the expansion of

prove that

and

if

is

a positive integer,

- -^n(O)
Deduce that
if

^n(x) = l n - 2 n + 3" a polynomial in x,

. . .

f(x)

is

/(I)

^Ex.
2.

...

+ const.

As

particular cases of Ex.

1,

we

find

These give, when x

is

a positive integer,
or

1-2 +3 -4 +...+(_i)*-i# =-U 2 2 2 1 - 2 + 3 - 4 + ... + ( - \- a? = (- \


l

and

so on.

In the
is

first

and third

cases the alternatives are to be chosen

according as x

even or odd.

Ex.

3.

It

is

easy to see that

From
Shew
Ex.
4.

the foregoing equations and from those of Ex. 3 prove that

also that

(x-%)
if

is

a factor of the odd polynomials, and

x(x\)

of the even polynomials.

Prove that

is

odd and k

is

an integer,

r=0

and obtain the corresponding

result

when n

is

even.

x.|

LMPLBfi

A.

\.

Complex Numbers.
1.
It

th.-

inn

an-

l".tli
ii.

'mpl.-x,
li,

>h-w
;m

tl.

;-.ints

tli.-v
-I,

..ii

ri|iii;u

od n"i
d cases

when x

is

al.
if

and

;liat

!>--=a',

|&+]/&:= > /[:i!,-,,>i>(


\\ln-i
.

I/--M-..

_rer.

2.

'

y are complex, prove f l*+yl +


;

that

pivt

thi^

-|uati>n
!-'

in

Ai-iranil's ilia-^raiii.

'hat

l*+yl +
3.
It'

[HAKKXKS.S and
I,

/!

are the points


.

in
.

.m

whirh represei.
'=0,
t.
tlif

uid
"<

.1

slu-w

that

tin-

.iindittMii

.N-nt

condi

o.i-

OA ./:.
..f

.:

OA
0.]

.\<>i:.
[.!/<

mid point

Ml.

1901.]

[Tian-fT
4.
if
It'

t"

.1-

"I'iirin,

\\hi.-l,

Q08J

I.

wh(
:i

inh //=

sin

and that

we write
in

^
'

\\li.-i

0=

sii

mi the roots of
tliat
6.

th.-

-.|

nation \\, :

(*+l

)\

and prove

tln-\
It"

the

|iiati"ii

ha(t \s<'

lii-nrj

real

and

t \\

-"ni|>l. s ,-as = 0.

7.

If

/]

a,

t!

tin-

point

't.

5+y

M
8.
t

a rouii- m- straight

lim-.

\vln-n

takes

all

n-al

\a!

onnliti'!is for tin1

[STU./ und
-Miplrx
iinniUtM
-u.

<

ti

IW

tliat

as

\arir-. tin- point

"
1.

at + h

t-c when it

,'ht

lin.-.

LB,

242
9.

COMPLEX SERIES AND PRODUCTS.


If
t

[CH.

varies so that |j

= l,

shew that the point


|a]

in general describes an ellipse

whose axes are

+ |6| and

|a|-

6|,

and

whose
line

foci are

If |o|

= |6|,
is
t

which
If

given by x = ab. prove that the point x traces out the portion of a straight terminated by the two points x2 =ab.
2

10.

varies so that

|"|

= 1,

prove that the point

in general describes a limagon, whose focus is c b 2/a. Find the node ; b , shew that the limagon reduces to a cardioid. and if a [MORLEY.]

11.

If

varies so that

|tf|

= l, shew
a

that the point


b

*-iZI + *Fl+
in general describes a hyperbola,

conditions

is

and find its asymptotes. Under what the origin (1) the centre, (2) a focus of the curve ?
x=
-,

Prove

also that the point

t
3

(t

z^ H
*

-^

=
^-

+c

describes a parabola.
12.

Constructions for trisecting an angle.


a = cosa + *sin
a,
t,

If

the determination of ga

is

equivalent to the solution

of the equation in

= a.

To

effect
1
1

the circle

this geometrically we use the intersections of a conic with I the form of the conic is largely arbitrary, but we shall

give three typical constructions, the first and second of which, at any rate, were known to the later Greek geometers (e.g. Pappus).
(1)

rectangular hyperbola.
in the

If

we write our equation

form

t that the points trisecting the + irj, 2 are + r/ 2 =l of given by three of the intersections with the circle angle the two rectangular hyperbolas

and then put

*-o/4 = -;, we find l/t

-?/

-(cosa + 77sin

a)

= 0, 2^-^sina + ?/cosa = 0.
of the hyperbolas
is

Of course the fourth intersection


is

the origin and so


is

not on the

circle.

Either of these hyperbolas solves the problem, but the second

the

easier to construct; its asymptotes are parallel to the axes (tin- one axis being an arm of the angle to be trisected), its centre is the point

(-Jcosa, Jsina), and

it passes through the centre of the circle (that is, the vertex of the angle to be trisected). Since a hyperbola is determined by its asymptotes and a point on the curve, we can now construct the

hyperbola.
(2)

The

first

as the hyperbola f 1 - <ty* - 2( 00*0+17 sin

hyperbola of eccentricity 2. hyperbola in (1) cuts the circle

-fr/

=l

in

the same points

a)+ 10.

\.|

LMPLE8
li\

\.

Tin-

pei

I,,,

la

baa
tin-

ecr.

>

),

and

IT-|M.M.|
oth.-r
I.

the
tin-

vertex on the

ran. h

,,f

pres.-n;
(1
;

naturally tli.m th'.-e ^iven in


ad
tii'

hut the

I'.v

that a*

tin-

iir-t

(1) cute

the circle

2 *+/ = l

in

tin-

same point-

,/sin

n -

1=0.
through the points

Thi- paral>ola has

irallel

to ?;=0, passes

ami touches
13.
If

tin-

line

u f

//

sin u -

-----)

at

tin-

point (sec./.

and
(i)
/

.V

l'.'

shew
\
-

ti

= = 11
:

x /7,

...

,11.
V
:

(iii)
1

13
at

s
1

'i:'..

kin^

Oftfll

A'

s -f -'

N \ li'

-./O-K'-'-v

easily

|.i-.,v,-,l

that

,V-f.S"=-l,

.)-=n.
-

and

=2.
_

Thus

.v

is

a root of

,S-'

+ + = o,
.S'

:J

\\hi.-h

X*
It is

:.

or

.V

easily

proved
niii-t
,.

;.

\\\^

tor
tli >t

7)-

un(&r
ease

7)

sini
|..

irr

7)

the >i-n

Le

oompa
,th
(ii).

188,

In
In

like inann.'i

CM6

(iii)

now
s=.,
*

+ 3* + J>
::
1

II.

ig

1.

l,,,t

13.
i>

That

.s

(and the!

.mst

l.e
|.

iilivimis

ly
the
first

eonsid.-i iu^

;.

4-cos(67r
in
\\hi.-h
,,

term
vain-

i>

jh-

,nd

thai

lew

than the
14.

(in

nunn-rieal
n

With

tin-

same

in

the

last
-

example, she\\

that

= 4 gives A'=(l '= 8 V (l


(iii)

a- 18

...

[In the
In
tlu-

we ha second case, wfirst

case

In the third QMO,

).]

244
The value
to a, can

COMPLEX SERIES AND PRODUCTS.


a-l

[CH.

15.

of the

more general sum

bn y=^x =o

~,

where

b is

an integer

prime

now be

inferred in these special cases.


if

We

find, in fact,

if

if
if

if if
if

= 1, 6 = 3, 6 = 1, 6 = 2, 6 = 1,
6

2, 5,

4,
6.

3,
6,

4,
7,

5, 9,
8, 9,
7,

10, 10,
8,

3, 4,
5, 6,

12,

(-l+;)2v/2,

if
if

= 2, = i, 6 = 3,
6
fc

11.

(-l-z)2 v/2,
(1-1)2^2,
16.

6 6

if

= 5, = 7.

from a consideration of the special cases discussed ar n * may, or may not, be equivalent n set x In a is of the form 4/- + 1, where k is an to the the former case, and the of k sum consists 8 integer pairs of terms, whose indices are complementary (that is, of the form v, a v). On multiplying SS' out, it is easily seen to be the same as k(S+S'). or JT 2 = 4& + l=a. Thus we find S'2 + S - k = Similarly, if a is of the form 4& + 3, we find that the terms x~ belong
It will be seen
in Exs. 13, 14, that the set of values
*.
;

n<1

to

',

and then

Thus

we find 88' = (2k+\) + k(S+S') = k+\. S'2 + S+k + I=0 or X 2 = -(4 + 3)= -a.

[Mnth. Trip. 1895.] (and indeed a complete general determination of the sign of discussion of the distribution of indices between S and *S") belongs to the problem of quadratic residues in the Theory of Numbers.*

17.

When a
a=

is

an even integer a = Zk, where k

is

odd,

we note

that

affi+W^-aF*, so that

When
18.

complete proof

identically zero. 4&, the results of Ex. 14 suggest that JT=(l+) N /, but of this requires some further discussion. t
is

Deduce from Ex. 13 that


tan
-

r- 1895.]

[In fact,

since

#" = !.

Thus, in the notation of Ex.


i

13,

tan

= S-S'-2(A-

*For example, see Gauss, Disq. Arithm., Art. 356; Wcrke, Bd. 1, p. 441; G. B. Mathews, Theory of Numbers, pt. 1, pp. 200 iil-_>: Werke, Bd. 2, p. 11 111-117. H. Weber, Algebra, Bd. 1, 179 Dirichlet, Zahlenlheorie,
; ;

t Gauss, Werke, Bd.

2, pp. IM

15,

x.|
19.
tin-

l.\

LMFLE8
to
.in

Tin? ivsnli

in lead
:

easy geometrical
.;

r.Mjular

heptagon
:io

gee at ODCe

'

roots of th-

s7 + l)-l
s<

tint

),

-,

2
.i--

x4 are

root-

If

\\>

\\nt--

'

+ //. l/<"4
,-

>

/N '7-l=0,
tlir-ni^i,
tintin'
:

\\hirh

represents
tlii-r

ular
r

IIVJH-I-IM-:

ticea

n-^iilai- lu'|it;i.u iMi

insnil.nl in

f..!i>tru-ti..n
:{

bj rith.-r
".
i

/r

.-c

/N 7

20.

If

that
.

(r=0,

1.

_.

an.l

l.-liici-

l.y

.litl.-i-riitiati.'ii

that

1
'

duce
t-

that,

if

rw) cosec u = J<u = ?r/-l/'.


!

58

^ -f rw)

'

osec^^.
...

5a

to

i>

.Mjiuil

[J/
is

1901.]
/,,

21.

If

/'

an

...1.1

r,. r int- u an ,l ^

an y

intt'irn-

jirinn- t.

shew that

||f-ll

2ain(S
wlu-re

^) i^-A,
inn-iM-r
\\hfii

6 = irqlp, ami A
MiiiM-

is

any

fr.ni

t.

(l...th
/-.

inch:.;

the \aii ..... f the

sum

than

[K:
I

:^)7.]
partial
fra.-ti.-ns

Writ., r-.-'". thni

fn.m the th-,rv

..f

;,
8

K4
and
\v

tin*

limit

-f

In.th

sides as J>--l,

get

-A-2<
1

1
;.]

-se

A</>,

-A
22.
I'ri'vi'

similarly that, \\ith


\

tin-

same notation
v

as in the last example,

(a + H#)

f,

\vhnv

is

mid and

n.-i

1,

l>ut

net-d

not be odd.

[\Yriu-

A=

i(X-

+ ]),

.r

= e-' '*
2

in

tin-

partial fra.-tions nsed in

Kx. ^l.J

246

COMPLEX SERIES AND PRODUCTS.


Convergence of complex sequences.
If (a n )
is

[CH.

23.
limit,

and

if

a sequence of complex numbers, which converges to a as a b is another complex number, shew that values of b n can
is

be selected so as to form a convergent sequence, whose limit


the values of ba
24.
.

one of

If

is real,

l prove that any value of x oscillates

finitely

both as x

tends to

and

to co.

25. straight line can be so that the series


!

drawn
])

in the plane of the

complex variable

#,

^|^2!
line.

*(*-!)(*- 2).
3!
;

converges to

on one side

of the line

and

its

modulus tends to

infinity

on the other side of the


26.

[Math. Trip. 1905.]

least
is

If an infinite set of points is taken within a square, the set has at one limiting point (that is, a point in whose neighbourhood there
set).

an infinity of points of the

[For if the square is subdivided into four by bisecting the sides, at least one of the four contains an repeating infinity of points of the set this argument, there is an infinity within at least one square whose side is a/2 n where a is the side of the original square, and n is any integer. It is then not difficult to see that we can select a sequence of squares,
; ,

each within the preceding, and each containing an infinity of points of the set the centres of these squares then define a sequence of points which can be proved to have a limiting point. Finally, we can shew that within any square whose centre is at this limiting point, there is an
;

infinity of points of the set.]

of

Sn (x)=f (x)+fl (x) + f2 (.v) + ...+fn (x), and let the roots be marked in Argand's diagram for all values of n if these roots have x = a as a limiting point, the series
27.

Suppose that

(#)

has # = a as a zero, provided that the series converges uniformly within an area including ,v = a. [HURWITZ.]

EXAMPLES
1.

B.

Power-series.
If H,
:

K are
is

the radii of coin er^enre of

^/,,.r"

and

-!>.<" respectively,

then

(1) HI!'

(2) If 11 is less than

but

if

the radius of convergence of ^Jt .'". n is tin- radius of /.' /.' convergence of 2(a n + bn )x ; 11= It' the radius is at least equal to A' and may l>e greater.
.

[Apply the method


2.

of

Art. 82.]

If a power-series is /.en. at all points of a set \\hirli lias the origin Art. .">_'. .as a limiting point, then the series is |< '<>in|>aie identically /.en.

x.|
\ power-series cannot within

IMPLEfi
!

a
ary) at
'in.

purely real

all

jM.ints

tin-

la-t

example.)
in

4.

NV- have seen that


i

many

ra-.-s

ill.-

radiu-

.f

Convergence of

2ar*

can be determined

lun(o

a
/
,

and
that

Kal'iv
if

ha-

pr..\ed
is

(te
t..

lal:tni;irl.

pp.

!:

this

limit

e.|iial

A.

th<-n

{.WIT ->.-i
t

!-.

Si.nirtiin.-s

wh.-n thr limit

a singular point "f the A does not exist, we can deter\B

=A

n-luti.>n

such as

\\ln-i.

'

ml

t.i

drtinit'-

limits

/>,

>/

as 7i-*oo.
t

ThiMi

tin-

radi
3

tin-

nioilulus of the least root

of

-jt.r-qx = Q. [For details, si-i- Van1

Vl.-.-k.

M-itL Soc^ roL

1.

1<>00, p.

293.]

5.

An

illustration of tin- last

example

is

afforded

by the

series for

log[(H
in

**+i*-K..>
(;j

which
tin-

4-^)'/,,+-j=
1

and

IMOCiatod

|iiadrati.-

is

Similarly for the series

f^ + |^+5.r+...
oUaincd from
-l<.u (l
(/
r -

.'-)

+ ar<i

tan r,
,

we + (/

get
-f l)// M +

+3K,
ciil.

=0,

and
l.-it

tht-

associated
if
-

=0.

:ally.
.litl'en-nt
:IH-

have the same radiuiy

singular |.int>. th under Van VK-ck's rule.

rgence be expected

6.

If

-"

'

'

converges within
value
>>'
i

tin-

circle

.r|

= fl(>0), shew

that

Com

all

examine the relation between the regions

and 1
value of
l'a,,r"
(

7.

If.

-'/

is

the

maximum
ilian

,n

a circle

.<

/,

is

also

grejiter than the valu-

any
the radiu> of

p.-int

within

th-

com
within
ti,

Sliew tiMtlier that


vvln-r..1

[Use
8.

auchy's ine<|ualit

ie-

"i".
]

I'

/.'.

tlien (>e,
:

"JV ",

wh.

>educe Caudiy's

ineijiialit;

248
[For we have

COMPLEX SERIES AND PRODUCTS.


<Sfor"f(x)
,

[CH.

= ant
to /(#).

and

if

an

'

is

the conjugate to
n n '2a n'r* /x
1

Thus and

=f (r2lx) is the conjugate f(xW = f(x}fi(r*lx} = 2<C^n ^#- n/(#) 2R|/(tf)|


\

(Art. 82)

9.

If /(a?)

= 2a H#n

converges for #
|

< R,

then

where

D is

the

maximum
.a l

of !/(#)-/(

-#)| on the
l

circle

\x\=r<R.

[LANDAU and TOEPLITZ.]


[In fact,
so that

= morif(x\ -a^m.v- f(-x\


result.]
,

2a l =m[x-i{f(x)-f(-x)}~},
that

which gives the desired


10.

n the series the radius of convergence of 2an# ^dnX"* will converge absolutely, provided that the argument of x is greater

Shew

if

is

than log (!//>).


11.

Obtain from the binomial

series, or otherwise,

the equation

where

v is real

and greater than

1.

What

restrictions are required as

to the value of 61

Shew
12.

that the equation ceases to be true for

i>

= 7r, and

explain why.

Find the sum of


cos 30+...

and

of

[Apply Art.
13.

89,

putting

v=-J,

x-e

i9
.'\

If

is

positive,

shew that

and examine the special form [Take x = i in the expansion


14.

of the result of

when

w = l/10.

(l+#)

m
.J

Examine the convergence

of the power-series

[Apply Weierstraas's
15.

rule, Art. 79.]

Discuss the convergence of the power-series

[In

Lin-

third,

tin-

...rllirim!

,,f ./"

steadily decreases; see Ex.

-2,

Art.

.]

16.

d of

f/017.

Determine

tip-

expan>i..n of
*

-*
s

it)

in po\\er-

;,,/

Jo
[I'u;
in

-1

Jo

the exponential series.]


lv\.

18.

Shew

tliat

(compare

17)
-

*
\\here there are
(/<

...},

!)

b(n

+ -2)

tei-ms in
Q
/./-.

the luackets.

Determine
19.

a siini!
if

Sh.-w that
cos

0|<l/e,
1

0=1 -6

22

3s
'/*

Bin 6-* 33
_!

cos

40-

3!

[JA/M.

r/-/

1891.]

[Write

<t

=b=i

in

the formula

.f

K.\.

I.

Ait. 56.
lie

The

intr.nlneti,.;

complex numlters in the place of i-eal oii.-s may of the same type as that used in Art. 88.]
20.
It'
'
'

justified

hy an

ariju:

'.

sheii

that

Lagrang*

H+^l
ay
I
|

ami that the


21.
'

< 2.
shew
that
.-...

It"

OM^H
1
,

0<r<l.

/'**

(1

+ ft)"t

and

SF/
ted
;

^H
lie

-,i.?-i*nf-'
c,,etlicieiits
iii

the liinoinial series.

fiom

.\IM-I,

theorem

that,

if

>o,

+--3><
.neiMr).
[Note that the ai'mimen; -1) when ril
tirst

jiproaches the limit 4-Jnir


itive,
if

and

summation

is

valid

L-- when did tli- M-.-..ml


!

if

>

lj

lut

th-

me

rather IIUMV ditlicult

\\

'ive.]

'250

COMPLEX SERIES AND PRODUCTS.


.r = cos# + isin#,
|

[CH.

22. If is convergent, prove n -*0 and 2|an -a n+1 Dirichlet's test (Art. 80) that 2a,Xl is convergent if 0<0<2?r. If 26,Xl is convergent, under similar conditions, then shew that

by

cn

=a

bn

+ ^ &_, + ...+
\

provided that

2^ n Z?n converges, where 4 n = \a n -a n+1 + \an+ i-a n+3 + ...,


\

Bn = \b n |
|

b n+l

6 M+1

6 n+2

+ ...

[Note that the sequences (A n \ (B n ) are decreasing sequences and that A n \bn On Then apply the method given in Art. 35 n (as in Art. 80). for establishing Pringsheim's theorem on the multiplication of series.]

<

<B

23.

If

f = l/n log n when n>\, and/ = l,


x

the series

convergent, but not absolutely, while 2 v n -v n+ i\ is convergent. that the signs of v n are the same in groups of 1, 2, 4, 8, ... terms.
is

Note

[For we have /5</4<4/2 or /4+/5</2 and /7</6<i/s or /6+/7</3Thus /2+/3</4+/6 +/o+/7 and so on for each group of 2* terms with the same sign. The convergence then follows from Art. 21. 2 v n - v n+l = + Again, 2) + )+
:
1 \

(t\

+/

(/2 +/4

(/4 +/8 )

. . .

so that 2|v M -v n+1


24.

converges.]
series

[PRINGSHEIM.]

It follows

from the previous example that the

=
converges, but not absolutely, at every point of the circle
|.r|

= l.

[The convergence of 2|-y n -v,l+I |, combined with limv n = 0, establishes the n and the last example convergence of ^v nx at all points except x=\
;

enables us to include x=\.


that
25.

As regards
divergent]
,

the absolute convergence,

we note

2|tn

*|
l

2|t>1

,|2/n
l

is

[PRINGSHEIM.]
are the upper limits of 1.5,, as it follows from

If H,

= b + b + ... + b n and
to ?>i-l,

if

A", A',,,

n rant^s from
Arts. 81, 83 that
is

and from

to GO, respectively,

'J'li

us,

if

*2,a n

a series of positive terms satisfying the condition of


83),

uniform divergence (Art.


x
I

we

h;ivr
0,
if
li

In-

llrii'-<:

path of approach lying within the lima^on of Art. 83. ^ciM-rallv, for such path-,

the right-haml limit Iming supposed to

e.\i>t.

[PRINGSHKIM.]

x.|

LMPLEfi
:

K
>

r\ten<i

to tin- coinpN-.x.

variaMe
tO

;in<l

in

part

i-nl.i

theorem
lyinu'
:

M
It'

to

apply
\\
,

-h

\\ithin

tin-

In

note Blao
-_

tin-

foil..

ilt

-IM-H. ;is

.r-l,

-\

t'i!ili.-i

'-xtelision

is

quoted on

|.

I:'!',

al"

27.
If

Converse of Abel's theorem.


liin(l
Mt- to
ii

ial

t<

;i

finite

nuinl't-r
fui'tlit-r
tlii->
[\
i

.1.

it

\\itlmut

'--t

i'-t

iotj

.n

liiri.-iit^.

In

two

siiiijilr

c
:

in
all
JM.^

ni.'ikr

(1)

\Vln-n the coeffici'

\Vlu-ii lini /"/,

<>,

and

.>

aj.pr-.a.-li.-

l.y

any path within


\\e

tin-

Ol

Ar
[Sincr in ras
of
(I)./lini !'</. /"

[T.\-

can apjirnafli that !'/ cannot


1

l>y

IM! valurs.
;

can

infi-r

fi.m

tin-

livrr^.is

t'ui lli.-r, ^.n,,

cann-.i
l>y

llrncf, in case (1)

^i-f,,

convives, and
,f
r[

then-fore eijual to J,
find

Abel's

theorem.
In
<
;

it,,

i,

,. then we

1(1

r)|

= |l+^ +
.',.
.
.

>

./--

+...+.r"U)
f.
.

|^
1.

the ilppel- limit

to

;.

...

We

then

./

as a
in

ha\e

th.-n.

point Ait. M3,

<m

tin-

iriven

path Mich that


i.

\\

|l-j:|<A
1
on th--

and 80

T|^
-x,'a<'h of the

1
liirht

//,.

ti-rnis

t<-nd- t>

<Mt)

lt

\\

i,,

and SO
lini
,

,i
^<i n
o
\\

lim

1
e

.1.]

28.

In
I'lom
th.-

ii"

d'-fmite limit.
lini z

can infer
h

th.

nee of

of

within the lima

and from

tlie

Condition
lini

(i

'2'i.
i

4-

&

/
;:

-f

Thoe
[Tau

coiiditinns are

Loth neecssary fop the convergence of


vutl,
.

2^,

and,

taken together, they are

!'.d.

^.

i;d.

:n,

UMI.

252
29.

COMPLEX SERIES AND PRODUCTS.


Applications of Art.
84.

[CH.

it

is

easily verified that,

with the notation of Art.

84,

and so

on.

Thus,

we

obtain the transformation

The two
|
1

series

on the right-hand are both convergent


the latter condition
circle
| |

#!-#!< 1+#|, and

is

satisfied for

which are outside the


of the binomial series.

= 1 we have thus obtained a continuation #j Repeating the process, we obtain


;

if .v < 1 and some points x^


|
\

where we assume that the broken


\xr+l -xr \<\l+xr
\

line

from x to xn
I,...,

is

drawn

so that

= 0, (r

n-1).

For example, by taking

so that

where

if we take a broken line passing below the real axis, we find /(-2) = e^ 7r [cos(a7r)- <'sin(a7r)] we thus obtain two different values for/(~2) by approaching -2 along different jf>aths. This
;

Thus we are led But it should be noticed that

indicates (what
(3

we know
wimilai

=
30.

to be the case) that f(x)

is

manyvfelued

unit's

and a

is

an integer.
to the last

method

example can be applied to

for

which we

find

and
tl
<

same

|Hiints

.1-,

.r,,

./'

.1-.,,

.r

us in the

last

example, we get
ival

<(-2) = Z7r.

And

with

lnkni

lim-

passing below the

axis,

\\<-

get <^>(-2)= -/V.

x
I

31.

If

tin-

OCX

aiv
<_'.,
tin-i
i

all

positivr
ai.d
if

(fl

\i>*

has a

BJ

i-oint

iln-

radr.
l:..d
,,f

VIVANTI

pi

Su|,|,(^.-,

possible,
--on-

that

(t'..i

0<r</0
than
A'
/

the
;

HIT;-

\v

can

l!

1111111!..

f>

that

tli.

84)

M
it

a d<>ul>;
kin

will

thm
X
It

OOnvergen(
>in^iilai-

\\ln-n

>iimiiiMl
!" tin-

as
.-i

I'//.,

{/-

That

\\

ill

000
In-

i-inal hyjM.tli-

and

si

nin-t

point.]

32.

Weierstrass's double-series theorem.*


that

thr

1
anall

-.)
and
nnlli.'i

OOnvergenl

f.-i-

A',

that th-

OODY<
(1) thr

'/////// a
90
i

1>

in

L:

r\.-ry circle
fin-

\\1.

--ss

than

A',

'i

r\.-ry

valur

..f

1
tn=0

-I

<lt.

n=0
i

tin-

niran IMMII^ taken aloii^ an


is

\..w on this

uniformly
r.jual
!-

i-mi\ rru'nit.

and

so

th-

-t

be

-ti!ivri'_u'-nt

and
//

in,

if

is

any integer and


'

f '.,

\\.

ha\r similarly
/;
. .

,.d

/;

if

M^

is

thr niaxinnini
ii

ol

=r,.

Eenoe,

N=0

and

1.

LIT.

Q(

<M^

so that
-r.

li.l.

11..

p.

206.

254

COMPLEX SERIES AND PRODUCTS.


identically

[CH.

Now, -we have

F(x)-G(x) =

n=0

2 (^

-#,)'"

because this equation contains only a finite

number
L

we

find

we

n=0

(/z)

of series: am'

< ^(Sr, - r)/(r


p.
oo

- r),

if
|

*
|

= r.

But, since F(x) converges uniformly on the circle #|=r,


as small as

please
/x,

independent of

by proper choice of we must have F(x) =

we can make ^fl n Thus, since F(x) and ~2A nx are


J\

71

n ^A nx =

EXAMPLES

C.

Miscellaneous Series.

the series

*La n
is

where f
of
all

F(x, n) converges absolutely, provided that 2 a n \n ^does so r n the real part of x. Thus, in particular, if ^a nx has a radius

convergence greater than 1, "Za n F(x, n) is absolutely convergent for But if the radius of values of x (other than real negative integers). convergence is less than 1, ^La n F(x,ri) cannot converge.
Finally,
if

the radius of convergence

is

equal to

1,

suppose that

where A >
-a.

and
|

a> n

<A

then

F(x, n)
x

is

absolutely convergent if

[KLUYVER,

see also NIELSEN,

Gammafunktion,

1)3, J.)4.]

[Note that
2.

F(.r,

n)cuT(x)n~

.']

Shew that, with the notation of the 2 a n F(x, n) converge for the same
/

last

values of x.
;

[Apply Art. 80

l/vn
3.

= n*F(x, n) (2), taking v n then are n easily proved to be convergent.] \jv fi\
that in the notation of Ex.
/'(.''-I,
1,

example, the two series [LANDAU.] the series 2|i',,-v n+ i| and

Shew

n)-F(x-\, n + l) = F(*, n), and deduce that S/'X-r, n) converges only when f >1 that is, -F(.i; Shew also that 2( l) n /X#, n) convi can only converge absolutely.
;

u)

if

^>0;
4.

and apply Ex.


series (see

to

deduce

the

corresponding results

for

The

Ex.

19,

Ch.

1.)
''
]

*
represents the function
5.

#
if

1_^41_.^
.r/(l -.r),

|.r|<l, and

1 /(I

-.<), if

|.r|>l.
[J.

TANNKUV.]

Slievv that tlu; scries

are botli convergent for all values of ./, except Kor applications, sec N IKI.SKN,
|

0,

-1, -2, -3,....


.^33.

ti'-iiniii<ifnnktion<

x.|
6.

LMPLE8
1

:i-e

of

comph-\

nuiiilM-rs

-n.-h

th.it

iily

to

>0, >ln-u

that

ho series

erge* abftolut.-ly

f..r all

uniformly
within
I

within
'

th.p
ill-t

dly
Cw
|

l.y

the circle

isilly tin.-it
.\

liy
/,',

= r,

\\liich

an:ll

muni"-!-

enough to

any

vrlappmu'

"f
t

7.

'I

.-vampi.
>M.-h

implith-d in CftM th- p"

iionur a .-traiu'ht

lin-.

and ancirciii!

that

- CH

^.k>

'

a constant
\Vltll

undt-r thrsr
to

|Tx|,,-.-t

nfi v
distril.ut.-d

8.

A-jain.

if

tin-

pointin*

-v,

although not
tli. -in

alon- a

linr.

tliat

t\\n

of

lilar statfini'iits

can

mad.- with
.

i-

n(iA
fuii'
t

-iin|)l.

set

of thi- \\.
tl,

bheory
n<-t\\i'ik of
JM.;I

>f

elliptic

ions, the

j-oints

r,

In-ill'.;

jiaia!.
fall

notr

that

imt

nnm- than onwe


:n.t

witliin

<ir.r.
niol'i-

than

^,,t

]n>int^

.-all

li-

liMv

th- two Bqnarea (m :

)/,

(ni

;
.

Eence 2

ron\-fl\
a st-ijiu-ncr of
is

9.

I'

i.-al

numli.-rs

\vhich
i>

'

and

if

conijih-x

numl"

part

jto^itiM-,

tin-

B81

lit.

.:/

--+''>! an<l

<.!/

.)/+,/

=A

(I ,

tin-

ratio

,,f

th,.

-,-n.-ral

t-

'" /f

)/(l

-A
.

if

and BO
<>n

if

MC
othrr hand,
if

that

//

^3.
i

tin-

A
A,,)1
.
,

\vhi-h l.-ads to tin- n-ult


1

/,'

s
!

'i

use
2
'
'

+
i>

'

In either rase tlu-iv

tinit.-

up|).-r

limit

//
.--nt.]

i-onvrrirt^ beOtjUae the eoni]

256
10.

COMPLEX SERIES AND PRODUCTS.


If
is

[CH.
if

2a M

is

convergent, the series

7* is n 3/f

convergent

the real part

of

is Thus, in general the region of convergence of *2a n positive. bounded by a line parallel to the imaginary axis. Further, in case 2a n is convergent, the series converges uniformly in a

M^

sector of the plane

bounded by the

lines rj=

-i

K, where

is

any assigned

number.

[CAHEN.] [For then we can use Abel's theorem (Arts. 80, 81), merely noting that we may write Vn = HM~_ in virtue of the last example.]
11.

It

is

the same type as in the last example) by any path between the two lines ?;= K (where K is an assigned number), the series has a limit /I, and if lim {a n f(n)/f'(n)} = Q, then 2a w converges to the sum A.
is
:

to series of the type ^Zan e~ xf(n \ steadily increases to infinity (this


If as

not hard to modify the result of Ex. 27 (p. 251) so as to apply where f(n} is a positive function which
to

x tends

[Landau, Monatshefte /. Math.


12.

u. Phys.,

Bd.

18, 1907, p. 19.]

Prove that

if

O<#<^TT,
*

cosx
13.

cos 3^ cos 5x + ) ^ o

\
...

)=cos.r +
if

cos 3.r

o3

cos

5x

R2O

+ ...

From Ex.

2,

Art. 90, deduce that,

O^^^TT,
sin

_ sin x -X(TT-X)TT
,
v

sin
I

3#

bx
+...,

^p
if

^3

and obtain the sum


14.

of the series for all values of x.

[EULER.]

From Ex.

13,

shew that
2

-^=v=%Tr,
siii3#
sin 5^7

7r.r/7r

#2 \_

and (by taking

8"VT~3";~ # = j7r) shew that


llTT 4

~ar

~5*~

1536
15.

v/2
2,

Deduce from Ex.


cos

Art. 90, that,


5.r

if

-\TT^X^.
t
1

3x

cos

cos

*ix

1. 3.

5~3. 5.75.7.9"

_TT = CC
8

~3

and

find the

sum
that

of the series for all values of x.

16.

Shew

where n=0 is excluded from the summation and Deduce that

0<y<l.

etc [L'A.MiK and ScHlAMlLcil.] and apply Ait. !U.|

[Put

= 0/27Ti,

^=27r//

iii

Ivx.

:i.

Art.

!>n.

x.|
17.
r'n.i.1
:

i:\

\\n
it'

i-duce that

li

between th

two even

integt

f
aild rXHUlillf the
r.i

[Writ,- *
18.

(A
that

8r)r,

-f

-.]

Shrw

and divide
19.
\\li.-r-

this r<|iiati>n

int..

ival

and imaginary
_.;-<' 0'r

|.

Sh.-w that

_.
//

rintar
9

is
.'

tli,-

ditr.-ivn.T
r

between
tineerifl

and

tin-

int-^-!
90,

to

v.

-d
is

-L'>/)

iii

and ..W-rve that

20.

If

"

an int-grr and
1
'

<

><./<!, we
4n*
th<<>

V'
,~i
\vhri'
-

m* - ri*
i-xdudt-d from
.-,
00
I

~~n
[Kx.
tlic
if
if if
</,,( / /)

'

summation.

3.

Art. 90.]
r-

21.

If

1"'th

y an-

lirt

\v-rn

and

j.,

w- ha\-

following

-7r(./--i),

-r>v,
.'

or

--('

-.{),

or
i'.-v)

= 7r[</>,(,-) +
T-'[</,,(,-)

+ // -

J],

if

.r>y, .r<y,
2.

or
<.V,o>
,-'-.,-

+ </>,(,/) 4- .i-H,

if

+J

is

the IVrnoullian function of order

22.

By writing

Mt. 90, or otht-i-wise. sh-w that

sllll:

Sllll;

|
!i)
t.

0<<>r.
of
MI

Use

Stok<-s'>

mil- (Art.

dcduc> cadi

the otln-r
'/'

l>y

ditlrivntiatin.i:.

[JVM.
i-t

23.

rxainjilf.
2
-

that

77

C08ha.r_l = - -f- ^" \\'t~ a


.
!

>

7T

'

_sini
.

-1

1 -I)"" '**"

<..<...,

Slllll

|c=
and obtain
l.v
a
.

+(-!)"

258
24.

COMPLEX SERIES AND PRODUCTS.


Shew
that the product

[CH.

converges absolutely except [If the general term is


25.

when x

is

!-,

lim(iAi) *

a positive integer.

Evaluate

#2 [(H-.r>4) and
i i

vol. [Glaisher, Proc. ZOT^. 3/a*A. Soc. (1),

7,

p.

23.]

26.

From Ex.

10,

Ch. IX., find the values of

27.

Prove that

where

log H c
of logc being taken.

any determination
28.

[HARDY.]
[GRAM.]

Shew
i

that
so that
t

n^l=| +
2
*i

</J

[If

= J(-l+iV3),

= l, we

have, as in Ex. 21, Ch. VI.,

Now

write

^=1, and observe

that

29.

Shew

that

"1
_^
w^-r

TT

sinh (?r.2\/ 2 ) + sin (

[Math. Trip. 1888.]

[We
L

have
^

= 4

where

$f4?\i+to

J_Y_1-- i),
/
is

1.

Thus, as in Ex. 22, Ch. IX., the given sum

and

this gives the required result.]

30.

Apply a method

similar to Ex. 29 to find

x.|
31.

\MI-LIProve
ti

259

that

til.-

lr;i-t

Vtlllf

/j_v '
[.I//'/.

7 rip. 1892.]

*)+/
and so the
32.

%U+(* + i>)~n+<*-iy)r

sun.

Shew

that

-l

4/1-

-1
,

..

7J\/3

,7T V

^ech-^-.
[M,,-

1898.]

+ ,T^ 2^^"2M^rV'
\\li.-iv
<,>

= ^(-l4-'x
easily reduced
tliat

ii

tli*

.iriv-n

seriei

A7r{cosec(i7rto)
t<

whirh
33.

is

tin-

-jivi-n

fm-ni.]

Sh.-w

aivtan,
1

.,

-arc tan.,

.\

tan

...

......

=arc tan

L
1891.]

[.!/

[It

is

<>asy

t<>

prove that

aivtan-5 "-

-.'-

- an- tan f

*
N

..I -fare tan

X
x

..I,

\ 3/

and

it

follows that

th.-

^ivt-n

il

to

\,

\vh-i.

:;.

.\i'pl_\

iil

the ^ivcii foiniula.]

34.

Slu-w that

'=
"

L
In partii-ulai- \ve
tintl
\\

\n + .r/

ith

H3
v/ 1
l<s
i

0-1=

T --

+ are

tan

rsinh(Tr.r)~l
.

-.')

260

COMPLEX SERIES AND PRODUCTS.

[CH. X.]

[We have

and

log

tan(^-)

In each of these, change x tox + iy, and equate the imaginary parts on
the two sides.]
35. The points P, Q have coordinates (p, q\ ( the point with coordinates (na, 0). Shew that

- p,
if

q) respectively

is

then
[If

we write p + iq=x, -p + iq=y,


x

[Math. Trip. 1894.] tan0=tan(7rp/a)coth(7r?/a). it will be found that

id = log sin (n-xja} - log sin (iry /a).]


36.

Verify that,

if

is

a positive integer,

where we are to put EP [Apply Art. 94.]


37.

= -B BP+ = Q
l

after expansion.

[Math. Trip. 1897.]

Shew

that

fcotf-1-^fj-B^-.B.f;-...,
and that

x
g 2 sin

_Bx^ + B^x*_ + BS^ +


2~2!
4 4! 6 6!
84),

(Jar)""

38. In virtue of Baker's theorem (Art. powers of x if |#|<JTT, and if we write

we can expand sec#

in

the coefficients

En

are called Euler's numbers.

Prove that
=1385,

111 l-3^
1

# = 1, ^ = 5,
2

^ = 61,
XXX.,

+ 5^-7^+... = 2^,

TT''"

+I

^ )f
.'-"

[See also Chrystal's Algebra, Chap.


39.

3.]

Shew from Ex.

37 that
cos x

cosh x

iog--^40.

-^ n+lcos
,<,_

,,

Bn

^-)^(2^r
[Math. Trip. 1890.]

Assuming
is large.

Stirling's

formula (Art. 179), shew that


J

when n

CHAPTER
\

XI.

CONVERGENT AND ASYMPTOTIC

SKIM

96.

Short bibliography.
follow in-

Tin-

are

th-:

principal

BOOTOeS

l'r<.ni

which this

chapter has been d'Ti\<-d


r.'i;K!.,
/.5

is,

1901.
2,

1896, p. 103.
1899, p.
]>.
!.

Normal
<

10,

1.

n. ISDO,
;.

IN.
H',

II.

H.MM'Y,

'///
'.<> i.
ij'.iT

'am

,,.

(published 1902).

^ //,///.-////
I-:.

./" '/>////

nf

M,iti.

01.36,1908,1).!

LI.

BtoT,

I'.HH.
II.
I'.
I
I

p.

317.
1

l^"'
1

..

I'.

..

\
/

\.-\\

Yolk.

l!n;.,

j..

<

i'/.ig,

1906.

97.

Historical introduction.
tiniiu'tlii..ls
<!'

r.rfniv

Analysis
before
tin(

had

IM-.-II

put

on

sinv

footing,

and

in jiarticular
l>y

tln-ory of

convergence had
little
t

Alirl

and

'audiy. niatlicinat iciaiis liad


ln-

in u-in-- in ii-r

n\ -i-r^-nl aeriea in Imtli

n-tical

and

investigations, In nunirrical Work, however,

which
tin-

art-

n..w rallc.l
ln---in
If

naturally used i.nly asymptotic (Alt, l.'JO below); in such


tli.-y

s<

trnii.s

to decrease,
tak.- the

and

r.-ach

minimum, afterwards

incr

we

sum

to a stage at which the terms


th<

*T
listi;

rs

on these serifs use


part of
ti

trgtnt aa including

in
viotis

quotations.

a<i

262

NON-CONVERGENT AND ASYMPTOTIC

SERIES. [CH.

XI.

we may hope to obtain an approximation with a degree of accuracy represented by the last term retained; and it can be proved that this is the case with many series which are convenient for numerical calculations (see Art. 130 for examples).
are sufficiently small,

An

important class of such series consists of the series used


:

by astronomers to calculate the planetary positions it has been proved by Poincare * that these series do not converge, but yet the results of the calculations are confirmed by observation. The explanation of this "fact may be inferred from
Poincare's theory of asymptotic series (Art. 133). But mathematicians have often been led to employ series of a
different character, in

which the terms never

decrease,

and
:

may

increase to infinity.
(1)

Typical examples of such series are

(2)
(3)
(4)

i_i + i_i + i_i + ... 1-2 + 3-4 + 5i^2 + 2 -2 3 +


2

Euler considered the


finite

1-2! +3! -4! +5! -6! + .... " sum " of a non-convergent

series as the

numerical value of the arithmetical expression from the expansion of which the series was derived. Thus he defined
the
"

sums

"

of the series (l)-(3) as follows


(

~1 + 1~2'
and
of Art. 98 (see
p.

J__l.

^_

_1.

-(1 + 1)2-4'

,^_

-1 + 2-3'

_L

his discussion of the series (4) will be


267).

found at the end

In principle, Kuler's definition depends on the inversion of two limits, which, taken in one order, give a definite value, and taken in the reverse order give a non-convergent series. Thus series
(!) is

lim l-limoj + limic 2 -lim^ 3 +...


;

as x tends to 1

Euler's definition replaces this

by

So, generally, " sum " as lim


X-+-C

2/w (c) is not convergent, Euler defines the 2/n (#), when this limit is definite a definition
if
;

which should be compared with the principle


the beginning of Art.
*Acta Mathematica,
t.

L'ormuluti'd

.it

!M).

13,

1890; in particular

i:t.

97|
'I'--

fflSTOR*
tlii-

\i.

[KTEODUi
<

i<

definition
IM-

('all.-t

raided the
in;.;

>lj-cti>n
in
t

tliat

the series

can

aUo

ohtained

l,y writ

he series

(5)
,

+*-+...,
istead of $.
n,,-i
I

whereas the left-hand side th.-n This ol.j.-ct ion ..!' 'all.-t 's \\as
(

,\

vrho suggeated that

ti

-h.ul<l

rk of Lagrange's, written as

ami

that th.-n

tin- <l-ri\-.-.l

..tiM

!>

1+0-1+1+0-1+140-1+.
Tin- la-;
I

imfl
....

to

!.<>.

|(i

BO
.

that
|,

tin- "

',

which
th-

i-

value
aeriee
<

^i\,. n
1

},y

1(

.ft
1.

hand
<>.

^\>\<-

.f

.">

In

nri^inal

).

the aroma are


cin^r
u-'ith
t-

i,

o.

i,

o, ...,

of which the

Kulci-'s

sum.

lla\inu rc-'ard

the fact that


l>y

KulT and
M-i-ii-^

made numerous discoveries


\\e

may

ftgrOe

with

Dnn-l

in

usin^ the

other mat heinat which do not converge,


:

statement

that

the

ollcr

n:athMiiaticians
that

had
in

sutlicient ly
if

^(.o<l

the use of such series as


results t

experimental they \veiv cnnverevnt led to

-vil-

COrrecl

the majority of easefl

when they presented


Bi t"
..litaiii

themselves
i]l.1

,/,/////,/////.
.f

rxumplr
1>\-

tli-

is

atl"nl.-il
i.

pa-au

'

in

1.

p.

rier

is

ol)t;iiniiiu

\vliat
'

\v<-

should
}

IK>\V

call

K"iui.
la

inctk>O

f\

ih

and

finds that tlu-

*Thi
tluMMvm
of
in

more satisfactory
l;,l.
x'.i,

basis
_'.,

Fiolx-nius c'
Ait.
."!

ls>u,

],.

iiat
liln

*-!
:iu
'

sum

,(

-f-

1)

lim

J.
with a few
al-.

MB

|>t

264
Tims the

NON-CONVERGENT AND ASYMPTOTIC


...,

SERIES. [CH. XL
and may therefore

coefficient of sin x appears as 1-1+1-1 + be expected to be if we adopt Euler's principle. As a matter of fact this is correct, since
,

sinh

x sin x dx = | (cosh x sin x - sinh x cos x \


2 f
I

so that

f(x)*v&xdx=^.

Abel and Cauchy, however, pointed out that the use of nontheir anxiety to

convergent series had sometimes led to gross errors; and, in place mathematical analysis on the firmest
foundations, they felt obliged to banish non-convergent series from their work. But this was not done without hesitation
;

thus

Abel

writes

to .his

former

teacher

Holmboe

in

1826

(Oeuvres d'Abel, 2me. ed. t. 2, p. 256): "Les series divergentes sont, en general, quelque chose de bien fatal, et c'est une honte qu'on ose y fonder aucune demonstration ... la partie la plus

Pour la des Mathematiques est sans fondement. il est vrai, mais les resultats sont justes, plus grande partie, c'est la une chose bien etrange. Je m'occupe a en chercher
essentielle
la raison,

probleme tres interessant."


in

And Cauchy,

the

preface

of

his

Analyse Algebrique
par exemple, qu'une

(1821), writes: "J'ai ete force' d'admettre diverses propositions

qui paraitront peut-etre

un peu dures
"

serie divergente n'a pas de somnie."

series (see Art.

the asymptotic property of Stirling's 132 below), by means of a method which can be applied to a large class of power-series. But the possibility of obtaining other useful asymptotic series was generally over-

Cauchy

established

in

looked by later analysts and after the time of Cauchy, workers the regions of analysis for the most part abandoned all
;

In England, attempts at utilising non-convergent series. however, Stokes published three remarkable papers! (dated 1850, 1857, 1868), in which Cauchy's method for dealing with
Stirling's series was applied to a number of other problems, such as the calculation of Bessel's functions for large values

of the variable.

the sense in which he

*Of course no one would now regard made it.


fSee the references
of Arts.
1.'!::,
i:i.~>

('auehy's statement as unusual,

Ix-low.

97, 98|

M.I:

\l.

ON8IDBB

Hut

no

-eii.-ral

tin
.1

when papers I'oinc.i written hy Stiel' of researches has.- heeii puMi-hed


OOming
until

ISM!,

aO8 that time a nun


nil
tli.-

theory. 9OQ16 of which


the

<|Uot-d
In

in

Ait
in- articles
tin-

the

t'oll.\\
t

we
i)l'

^hall

r.

.r

mod

par!

e..ntine
1

our exposil ion


1>\

theory

non
al

)i

1'nuvl

in

MMI. although

we

<!> not

ay> adopt hisim-th


series.

98.

General considerations on non convergent


\ii-\v

In

of

th'-

r.-sults

ol.taiiK-d
ins

in

th-

]>ast

l.y

th-

use of
a

non
I'.-rlVrtly
j.ivcis,-

prnhaM.- that
juirpo-.^

\ve

can attach

niranii

sud
uiili-r |M

may
cnal'i'

1-

M---I

!'<

.r

ivergenl of formal

calculation.

jtrictioDB,

Tim- we attempt

to formui.-i'

which

n a

tvergenl
a
call

not).

t<

as-ociatf
f

\\ith

it

jMTfcctly the -.sum" of the Beriea


is

dclinitc

function

....
It

numhcr. which and which

i-:

i-

of course
nl

olvins
hut

tliat
it

the

(letinition cho-.-n

to a

l;i

arl.itrary;

>lnull

he such that
lie,
(

the n -suiting laws of


that the
to
!,u

c-alciilatioii

with those of t-onver-eiii &


s

M'

COUT
aeriea

i'leiit

BUBD

'i

with

a noii-

convergenl

i-

not

conf-'unded
.

with the sum of a


a'

in the sense of \rt.iii: hut it will Convergent confusion if the deiinit i<ui is such that tin- same operation,

when

applied
.'In
i..
I..-

Qvergenl
t.-il

yields thtliat

sum
rate,

in

tli.-

ordin

|i..ii,

..in
hi.-,

Kuii-r.
"
't'

;it

;.n\

wai perfectly

of the di>t inrti'in

ln-t \\t-f

"

sum
lu-

nmi

.nvcr'_:-ii;

Of I
hat tin-

Tliu>

says

(jj<

los-l
i

U
1

d.\
ii>u.-ly raiiiu't
1;

I-:
doesS.

mi
1

.'.

in tin/

>nlin:r
i>,

th- larp-i-

tin- iinira

i-th tlu-

lull

account of this

IM-.T;.

rory see

Ar

-w).

266

NON-CONVERGENT AND ASYMPTOTIC

SERIES. [CH.

XI.

And he adds, after referring to various difficulties, that contradictions can be avoided by attributing a somewhat different meaning to the word sum. "Let
us say, therefore, that the sum of any infinite series
the expansion of which the series
infinite series 1
is
is

the finite expression, by

generated.

In this sense the sum of the


because the series arises from

-x+x

-x* + ...

will be

--

the expansion of the fraction, whatever number is put in place of x. If this is agreed, the new definition of the word sum coincides with the ordinary meaning when a series converges and since divergent series have no
;

sum,

proper sense of the word, no inconvenience can arise from this new terminology. Finally, by means of this definition, we can preserve the utility of divergent series and defend their use from all objections."
in the

In writing to N. Bernoulli (L. Euleri Opera Posthuma, t. 1, p. 536), Euler adds that he had had grave doubts as to the use of divergent series, but that he had never been led into error by using his definition of "sum." To this

more than one expression, and that


-could be given.

Bernoulli replies that the saine series might arise from the expansion of " if so, the sum " would not be definite
;

to which Euler rejoins that he does not believe that any example of this

However, Pringsheim (Encyklopddie, Bd. I., A. 3, 39) has examples to shew that Euler fell into error here but in his definition almost exclusively in the form Euler used practice
given a

number

of

and

if

which do not bottom, from those held by modern workers on this subject although of course his attempted definition leaves something to be desired,
It will be seen
differ greatly, at
;

restricted to this case, Euler's statement is correct. from these passages that Euler had views

in the light of

modern

analysis.

It is to be carefully

borne in mind that the legitimate use


is
;

of non-convergent series

merely

always symbolic the operations being convenient abbreviations of more complicated transin

" shorthand Naturally this background. labour us to the of not enable avoid representation" does justifying the various steps employed but when general rules

formations

the

have been
It

laid

down and
in

firmly

established

we may apply

them with confidence

may
work

the

particular case. that we might just as well writr be urged very likely in full, and so avoid all risk of misinterpretation.

any

But experience shews that the use of the series frequently suggests profitable transformations which otherwise might never
be thought
of.

An

c.\ainj)k;

of

this
(//.

may be taUm fr>m


Euleri ()//'! /W/tuma,

KuK-r's
t.

ronvspmuli'iu-.' with
;

Nicholas JWimiilli

1,

p. f>47)

uhrir

ilu> ivul

98,99]
olij.

BBL'fl

WTBORA1
Caning
ti-

ll.-

prOTM

lir-t

that

th- M

'!'

which

In-

nlitain-

tin-

intrirral

if

|-;h''

"

(l

in

tin-

"th.T hand,

liv

n>inu' tin- rul-->

which

In-

li:nl

olitain-d

f-r tin-

transformation
j?

x
i-h

2.r

&r

3.r

3oi-t-

1+
aiid
it
1:

1+

rr
l.v
I.

TT~ r+~

|.r..v.-(l

"~Ldt = 1 1+AY 1+
N"\v this n-latiun
\v..rk
l.t-.>

__-__
1+
1

1+

1+
1

1-f
ill

n.it

iUggi
i

naturally \vithut
is

tlie

alit-ady

nnai -ki-d,

it

r\ idnit

tli..'

was rntiirly guided


.r1,

1>\

thr aim of evaluating the


fi-.-ni

Writing

Kult-r cl-tains

th<- c'.ntinu.-d

fia.-tion tlic

nuivfrgente

1>
and
value
Ity

usini;

tli-

l-'Uh

and

1-lth

In-

int"-i-

tin-

numerical

Hr then

sulti-a-ts this drrimal fi'dii


J

and

infd-s that th-- Value

-f

tin-

99. Borel's integral;


Tl).-

summable
<>!'

series.
la->t

general considerations

th.-

m-ticl.-

lta\.t

n\\ to be
In-

Specialised ly adMjitin-" as Insum of a n


t

aome

mivrntinnal im-jmin^

tak.-n

Hardy,

in

tli:

[.MJM-I-S
//'

i|ii"ti-.i

rniulatt'd

tlu-

inllowin-- jn-incijilr

definib ordt

vdku

A'.

Imt.

268

NON-CONVERGENT AND ASYMPTOTIC


Y,

SERIES. [CH.

XI.

meaningless expression

we may agree

to interpret

as

meaning

X.
~=o

For example, the equation


J
is it

f(x,

n)dx =

[lim/(#, n)]
n-><x>

dx

JO

But true under certain restrictions* on the function f(x, n). definite may happen that the right-hand side is perfectly

be

while the left-hand oscillates between certain values (which may oo and + x ). In such cases, Hardy's principle is that we

may
limit.

take the right-hand integral as defining the left-hand

In particular, suppose that f(x, n) is the of a series of functions of x; say that

sum

to

n+1

terms

and that

The equation then becomes


o

<&(x)dx
Jco o

is

con-

vergent,

when

the series on the left

is

not

if

so,

we have an
of a non-

integral ivhich

may

be taken as defining the

"

sum "

convergent

series.
<j>

special type of function interesting results is


<t>

The

n (x)

which has led to the most

n (x)

= e-*un x n/nl,
x.

where

un

is

independent of
too o

n (a?)

dx = un

(see Art. 178)

arid
let

$(x) = e~*(u Q + up + u 2 |j + u 3 |j +...)

= e~ x u(x)
un

us say. We must assume t that the coefficients that the seriefl u(x) converges for all values of x.

are such

*A
fAt

few simple cases will be found in the Appendix, Art. 172. least for the present; but see Art. 130 below.

99, 100J

I'll

>\

>l

Then provided

//</

//

integral

conve,

we may
tl>

agree to associ
/':

<

//-/'/A

/A*-

*grie*

^
e

i*

series

this int.-^ral
i;ty

ma\
IK- rall'-d

called

tin-

MMIMIU1

Tin-

sum may

!>

d.-n,t,-d

l.y

tin-

syml)ol

This definition
tlir

Borel, who deduced it. 1, definition u-ivni hrlnNV ill Al' III: liUt BUICe the
is

due to

i'r..m

lias

].ro\.-.i

more serviceable
rd
tlii>

in
tin-

suls-.jurni

in\-<-sti-j-ati<Hi-.

it

as

i'uinlaiinMital

d.-tinit

.n.

100.
It

Condition of consistency.
oltvinus that
tin-

is

<l-finit

in ^ivm

in tin-

last

artirl'-

will

lead to diilirulti.'s- unl.-ss tin- .sum


\vli-in-vt-r
1
:

^U

a^r.--s with

tin-

sum

shall
"

imw pm\
that
if

oditiOD

i'l'

consistency

is

satistinl.

Fr.

ffrgent,VfG rnn tind a nunil).-:

<
Then
and consequently
all

e,

fi.

St.
tins.

',

/j,

rics

\J
.,

//

r
''

-e~ x
;

converges uniformly for


(0.
,\
>.

\alucs

df
1,-r

,/

within an interval

\\ln-iv

may

be

arliitrarily

f\ o
by A
rx
(

S^
u
|

|*A

''~ z

,c?a;

Jo

fA

where

,\

A_ te(_iT+7 V(// + D: (n+!


\
\

.+...toa

New,
Alu-l's

since

,\

is

jm.siti\r.

tin-

aence; ami

r..nM-.|in-ntIy \vc
L':!

sequence (X,) ia a can apply tin- rxtriidt-d form of


li
|

Icimna (Ait.

t> tin-

-um

\,,.

and we sec that


/(

A(\,-\

;
'

<//(\,

-,\,

)-f //

270

NON-CONVERGENT AND ASYMPTOTIC

SERIES.

[CH.

XL

where H, h are the upper and lower limits of UQ+U^ ... + ur as r varies from 1 and to m hm are the limits as r varies

from in to

oo

Now
hm = s
to oo
,

if
,

^un

Hm =

converges to the sum s + e, however small


1,

s,

we can

choose

so that

X and X m tend to the limit


lim
)

maybe; and as X tends because is fixed.* Thus


e

[h(\ Q \ m + h m \ m = h m ^s-e
]

and

lim [#(X
A-*- oo

- X m ) + H m \ m = Hm ^
]
00

Hence we

see that

A->oo

But
to
s,

is

maximum and minimum


so that

arbitrarily small, and so (see limits of 2un\ n


00

Note (6), p. 5) the must both be equal

lim
A-^co

2u
o

n\i

=8

>

or,

what
But
if

is

the same thing,

_ -^v

2un

diverges (say to +00),

we can

find in so that

however great
Thus, since

may

be.

lim [h(\
X->co

- \ m ) + h m \ m = h m > N,
]
00

we

see that

lim
A->->3

^ un^n = ^
o

r
-

Since

is

arbitrarily

great,

this
X.

inequality
is,

implies

that

tends to infinity
n
o

with

That

the integral for

is

divergent,
it

v when 2*** ^
o
;

divergent.

Of course

this

is

only as

should be

otherwise there might be a risk of

inconsistencies arising.

*For

=l-e-A,

Xm = l -

+ i\2 +

...

m
;X
)e-*>

and each

of

these

expressions tends to the limit

1.

100,101]

ADDITION 0V TERMS TO
/

SI

MMAHLE SEWER
00

101.
It

Relation between

'/

//

and

>
i

.loes

not follow immediately tY'-m their definitions

tliat

the

suma
are related
tin-

8= <*UH and
,

<J/
i

in

such a
<>f

way

thai

tin-

existence
definition

>!'

either

ini]

existence
relation

the

ther.

The

tin-

9l Jo
1186

r-'y/t. /)'/.',
////
'

.s'=| e-*u'(i''i Jo
a;

2
,

/.''

+";:.).+
are -m-h that u(x)
<

the coefficient
I'oi-

any valniii'4"

of

./

an-l

o MISC. jn.-nt ly
for
8*

Art.

")0

can be
tin.:

OSed to justify trnu-liy-tenii <litr.'r-ntiation of u(x).


(

)n

int.'-i-at

tin-

expression
r

ny

\>

tind

relation

x'^

g-w(a;)
I

+x.
exist, tlie limit

This shews that when


X
fcfl

>,

>'

l>oth
r
//

lini
|

(./)]

><*>

also:

and the \aine of

this

limit

must

he

zero,

Miiri

otlierwise the inte-'ral for

cnld

not conver^.-.

(S.-c

Art

Appendix.)
Hein-i-.
/,,,,-,.
i>i-<,

>>,/></
:

flnif

Jx>t/i

integrals st
aeriei

s'=/i u
in

thus the t\\n

j)i'n|ierti'>.

Beiiea

<

M'

respect, ana In--, .us to course any failure of the analogy


this
;

bere would
"

li

and w.u the definition which we have


.lvious ditlicnltie^

ly
sel<

reduce the
r

the

sum
of

utility "

of

sunuuahl'
I'.nt
it

must be rememhnvd
Jo 'o
I

that

the oonv<

l>y

no mean- enSUTQfi

ti;

delinit- value

lini

X >ao

as

is

cli-ai-

fn>m Art.

Dili

!'

the

Apii.-nilix.

It follows that

<>/ ./////

definite

valw

272

NON-CONVERGENT AND ASYMPTOTIC


example of by taking
this has

SERIES. [CH.

XI.

An
paper),

been given by Hardy

(p.

30 of his second

Then
but the integral for
s'

=
is

which oscillates. Thus 5' does not converge, although

does converge.

The reader

will find little difficulty in seeing that here

the algebraical transformation will be found in Hardy's paper.


00

Although the convergence of


that
yet
e x

u(x) tends to a
that

e"*u(a/)daj does not ensure Jo definite limit as x increases to oo

we can infer

lim e~ x u(x) = 0,

when

the

integral

Jo
r

e~ x ii(x)dx is convergent;
also.

and consequently

that the integral

e~*u(x)dx converges
For, write

y= Jo
/

e~ xu(x)da
,,

Then
and consequently

-*

d?y _ _ x

The

last expression

tends to a finite limit

?,

as

increases to

oc

so write

and then

(TJ

T;')

tends to the limit


e

0.

Thus we can choose


if

A",

so that

If

YI
;

these

has an unlimited number of extreme values, //' vanishes at each of Thus hence at any extreme value beyond A', we have rj <
.
| \

|T?|<

if*>.\', hence Iiui7/ = 0.


?/

or
if
r;

Iini7;

= 0, and

has not an unlimited number of extreme values,

must be

finally of constant sign,

say a; thus

r)'

must approach a definite 77 approaches the limit -a. Now, from Art.
and so

limit (Art. 2),


152, it follows

101, 102|
that
if

ADDITION OF
:ij,|,r..:i.-li.--

si

\]

\ i.

-/

tin-

limit

-o,
",

tli.-n

//

'.r

approaches the same

limit.

Tim* Wl

hav
liiu
if

Inn

(,/ .r)

-a,
be zero.

l.ut

ii

and

,,

hu\v

tli--

Thus we must always


liiu

h
?)

= 0,

lim
X-^oo

;/'

= 0.

X->oo

or

Jo

"ttto+l e-*u'(x)dx t Jo

and

Mm
x->>
that
tin-

'-'//(. /')

= 0,
vergent.

]>r<\ idcd

integral
tin-

Jo
n-sult
:

Accordingly W6 have

//' f/te

series

vmmdbUj

//"

//

00 "/>"

the series
..
:

//'

r' I

1.

ion between their

?////>

mode

of stating these

i-.-sults
/>c

is

the following:

///<"/
,

prefixed to

remain swwmdbl* //" .sums will be related as if (/' tm> series were convery Hut fhi- removal of even 6 singl* term from /// tin' of tl ////// destroy j>n>/r(;/ of swmrm
v.
/

and

/"

oortefl

"'iff

102.
It

Some obvious theorems.


ao

follows immediately tY<m the d-tinitin> that


oo
'

if

(*u n

and

oo

are BUmm&ble, tli-n

/\n n +''n) and


o

/
o

are also

suniiuahl.-.

and

Kurtht-r.

ii'

''

is

any

1'act.u-

independent <f
o

/'.

l.s.

274

NON-CONVERGENT AND ASYMPTOTIC

SERIES. [CH. XL

We

should note further that the addition of a constant to


series is equivalent to adding the this of course is a special case of the addition of summable series. For the series
to its

any term of a summable


same constant
the theorem on

sum

are

all

summable and equal


CO

to

k,

Thus
o

although, as we have just proved (Art. 101), is not necessarily summable when + u^ + u^ + us +
103.

is

so.

Examples of summable
u(x) = \-x +
/oo

series.

Ex.l.

i_i + i_i + i_....


x2
x3
...

Here

^-^+
dx = L

=e~ x

foo o

e~ x u(x)dx=

e~ 2x

Jo

[Compare

Art. 97.1

Assuming

this series to be

summable, and to have a sum


1

s,

we can

evaluate s by prefixing a term

at the beginning

of the series (Art. 101): this process leads to

s-l=- + 1-1 + ...= _ s


1

(by Art. 102)

so that

= J.
,

Ex.2.

Here, just as in Ex.

l-t + t--t*+.... (t>I) tx = we find 1, u(x) e~ and then we


Jo

get

Here again, assuming the series summable, we obtain 1 at the beginning of ts, giving value s by prefixing
}

its

or

Ex.3.

= !/(!+/). 1-2 + 3-4 + 5-....


^(oO=l -2u; + 3-4+... =er'\l e-*u(x)dx=\
e'^(\ -.e)rfa? =
'>

Here

and so
Jo

Jo

i-J = J.

102, 103]

LMPLES OF 9U1CB LT
tli.-

Assuming

siimiiialiilit y.

!!

us write

!UI ,1

so
til,-

9.1-2+8-4+5-64 0+0-0+1-2+8-4+J
corresponding
i.-rm-.

MI. 102).

Adding

1-14-114-1-1 + ...-}
<

(by Kx. l)

T
Ex.
4.

J.

(7=l+co80+ cos20+cos30 + ...,

0+8H10+ sin 20+n30+....


Thus
C+i/?=l-;
i'unrtion
i-

ami the associatol

)=1+^,
Of
Hi-jir,-.
pi-.
,\
;

+...=t'X 1 )(^)
e

.)

= er(co+i8in)
that
1)

id.-d

is

not zero or a

inulti}

tind

thi-

Mini

Jn

6U= T
1

COSO

S1110

-n

= s(H2\
tin-

Of coon
In-

iiiin^-

the
of

suimuablr property,
sann- drvici- as
in

MHH can
'2.

nlitaiiu-d

l>y

nn-aiis

tin-

Kx.

which

[nation

e(('+;N)+l=
and there!
Henl
(7-fifl

l/(l-<
fmind.

with

tli-

rrsnlt just

Thus wr
and
In

havt.

1+costf + n. () + sinH + siii2rt + .sii,:^+...,


likt-

nianin-r

w'+

find

t!

'

')(-

+
|

msf) + rMs:>f) + (.'0830+...=


sin
I'.y

,__

0+ sin 26
t-

>+...-Jc
H.

rhan^-in^ I'nnu

re

l'tain
.
,

.=

rin0-am20+ain30-...

-..)

276

NON-CONVERGENT AND ASYMPTOTIC


C=cos0 + cos30 + cos50+... $=sin0+sin3$ + sin 50+...
(7+ iS = e
4.
ie

SERIES. [CH.

XI.

Ex.5. If

and

we have
using Ex.

(l

+e +e

4ie

+...) = "/(!

-e

<')>

Hence
It

will

6 (7=0, TT). (0 =Jcosec0 be noticed that these two values are the same as

< <

are given (Ex. 4)


cos

by summing

and
but there
is

sin

no general theorem which would justify our


to

equating these

new series to By changing from


cos0

G and
0,

S.

JTT

we

obtain

cos30 + cos50

...=

^a^\
if

-sm30 + sin50-...=.
Ex.
to
6.

It is easy to

deduce from Ex. 4 that

is

not equal

+ 0, we
cos
sin

have

cos
sin
sin

cos
104.

+ cos 20 cos 20+.-. = + sin 20 sin 20+ =0, + cos 20 sin 20 + = ^sin 0/(cos
-g-,

...

cos 0).

Absolutely summable series. Following Borel, we say that a series

is absolutely

summable
Joo o

if the integrals
/

e~ x u(x)dx

and
Jo
\

e~ x u^(x)dx

where \ represents any index of differentiation, are


lutely convergent. It is at once evident that
if

all abso-

the series i^ + 1^ + 1& 2 +

. . .

is

absolutely summable, so also is the series obtained by removing any number of terms from the beginning of the series for the integral function associated with the series
:

by hypothesis

Jn

e~ x u'(x)dx is absolutely convergent.

103, 104|
It

OL1
in

-I

MM .Mill.

follows

\irtu-

of

what was proved


-0,

in

Art.

1"1

lim
X->!

liine-*wA (a;) = 0.

nimbly.

'/

*
f...

is

al*l tii'i >j swm/mable,

90 also art

"/ ;

/A

series

<//,i/

their

sums

<n->

And, conversely,
Bumrndble, BO also
'I'h.-it

'/
is

//"' th<>

M'rii'*

as '///"'// were convert)' n -f "._,+ " 3 + ... '-s


'

'

series

formed by prefixing u
<>!'

is,

\vt-

can

ini'.-i-

th- convergence
[

tli.-

e-*\u(x)\dx

Jo '0
IV.
.111

that of
I
i

"

o'cte.

Jn '0
T.

irvr

this,

h/t

us writ,'

Jn

Thus
BO
li

')^
\U(.'

Id -nc.Jo

'' ,

//(./)

J./'

urrtainly OOD
.!--.

Jo
I'.ut

&(x)d&

is

(Miivrr^vnt.

the r..n\
i

-nee of tin/ /.

last

iiit'_Ljral

can be

inl'.-nvd lV"in that of

l>y
1..-

the ar^tinu-nt of Art.


simplitird
h.-rc.

HM

Jo

above.

Tin- ]r<t', Imwrxcr, can of tlu- inonotniiic nature of


,
,

in

\'irtue

For we

liave tin-

identity

fJT o

/'- -' -*A(Z)-f


Jo
"

and consequently, because

/>(

and

ive,

\
o

.1
Jo

-I
Jo

278

NON-CONVERGENT AND ASYMPTOTIC

SERIES. [CH.

XI.

x Hence, since the integrand e~ <^(x)

is positive,

the integral

is

Art. 166 of the Appendix. It will be noted that this argument cannot be reversed

convergent,

by

and

that the convergence of

e~ x

(j>'(x)dx is

not deducible from the

Jo

Jao o

e- x $(x)dx.
e-*<$>(x)dx

is

x convergent, but lime- (f>(x) does not

exist,

may

be obtained by taking

so that
It follows

from a general result due


I

to

du Bois Reymond that

2 - 3 exp ( x sin .^) dx

is

convergent (see Ex.

3,

Art. 166).
in a power-series for

The function $(x) can obviously be expanded


value of x, although the law of the coefficients
is

any

not simple.

absolutely summable series, we may refer to those given already in Art. 103. Thus, for instance, take Ex. 2; here it will be seen that

As examples

of

and
so that

dx*
the

dx

is

obviously convergent, since 1


cases

>

0.

Similarly

other

are

proved

to

be

absolutely

summable.
105.

Absolutely convergent series are absolutely summable.

For we have

so that
|

u(x)

^ v(x),
5> n
Jo
is

if

vn =

un

\.

Now,

since the series

convergent, the integral

e~*v(x)dx

converges (Art. 100), and thrn-l'mv the integral

e-*u(x)dx
{oo

is

absolutely convergent.

104, 105|

Mill.!

-279

Further.

tli.'I

m
.same
I

^_it A+1 H-v/ A ^_.-i-

...

is

absolutely conthis

tin-

argument applied
\dx

to

series

shews

that

tin-

int.

Jo

dec absolutely convergent


absolutely
1'iit
I!..)-.

That

is,

an absolutely n.h
-

suiiiinable.

in. -lit.
is

that

any
lias

<-<>n\

utely

Minmiablr,
const

not correct. as
llar.ly
(in
h

b.-.-n
I

proved by an example
.a
]

n id.

-.1

l.y
is

per.

pp.

-~>

2&).

This

fxaiiijil.-

^ivrn hy

takin.u'

in

\\liii-h

K,

-I/'

i-

\\li.-n

/-

r-,

and
ill.-

i-

"

\\ li'-ii

i-

lid

M]
;

Hanly then shews


and
in
all

that
'lie

lai-<->t

saint-

.id.i

integer contain. -d .itmle as the


is
It

s
t

\\hich

:
i-

and

that

'*
<
\

u(x)\>
L.-inu'

K r,

\\ln-i.- K
r

intn -\

any int-u

-i.

a constant, throughout is then easy to prove

that

^'-nt.

It

is

hut not absolutely:

easy to construct simple series which aiv suiimial>l<\ an example is ^i\-n by the series

For here
and so
I

')=

x ['

(c.s./-+

sin

./)

!],

v , f*cosic (j-)<Lr=\

c-'-f/sinaj -

_ ^ ==
,
fl

But
SO
i

ucc

"i
|*\

[Art. 173 (DJ.

,.-*

Jo

U(x)\dx^\ Jo
.

lit.

{
In iVLranl
t

this srrii-s.

\v-

m>t<-

that

-O/^
\vhirli
'

tends

t<

B6TO, a
i

in

thu>. as

in

Art.

1"1,

we

B66 that

"'
I

ges.

280

NON -CONVERGENT AND ASYMPTOTIC


of this integral
x
is

SERIES. [CH.

XI.

But the convergence


so that
/oo

not absolute, because

\er

u'(x)\^j2lx-2la?,
is

if

Thus the

integral

JQ

e~ x u'(x)
\

dx
|

divergent.
Vac
I

Similarly,

we

establish the convergence of

e~xv>(x)dx,
/oo
I

A.

being any

index of differentiation, and the divergence of

e~ x \u x (x)\dx.
all

Thus

the

series

^A +

%+ + %+
i

+ "-

summable for

values of

A.,

although not absolutely.

If the

106. Multiplication of absolutely two series

summable

series.

are both absolutely summable, and

if

wn = u
then the series
is

also absolutely

summable, and

W = UV;
For then we have
I

Jo

e- x u(x)clx\
- (x+v)

e~

Jo

= lim
where the
square of side X.

\\ A->oo J J

u(x)v(y)dxdy,
taken over the area of a

latter double integral is

Now, in case u(x) and v(y) are both positive, the integral taken over a triangle such as OA'C' lies between the integrals taken over the squares OABC and OA'B'C'] so that, since the
latter integrals both approach the limit uv when OA tends to oo, the integral over OA'C' will also approach the limit uv. 15nt if u(x) and v(y) are not always positive, the difference

105, 106|

Mi I.TIN. h
tinini

\TIN
n.\ /;'
I.

between
tin-

OA BO and
'

integral

"'WaOI.
1

'"

taken over the area


the integrals
f
J

.I//'

''//.!

an<l

this

is

the

dil

A
-

C
'

A
"

'(

'i)\dy

Jo
'

and
Jo

'|tt(j")|

tLc
\

e"V\v(

which teniU

to
r
I

BerO, since the


30

ii,

e~*

and

r
<

-'

Jo

Jo

are in.th convergept we can writ=

lim
A

'

>/:Jj

<lxdy,
t.
!

when-

tin-

integral

is

now

tak-n
i-

"\-.-r

th- ai-ea of

triangle such as O.1V",


Del

wl.^-

sil-77

_\.

us write
tli-

= .,-+,/
Ix-conif-

= ^,

and th-n

intr-Tal

Jn
|

l-iy.].
Jo

.-(

Owino-td the
^I'-.'nt
I'.-)-

fact

that tin- scries

l'>r
\\

//

./

).

-(//)

aiv

al>s,.

lately

any \ah

can obtain the product

fti-)]-*(dr)
hy
the
rule
t'or

inultijdicat

i.

.11

of

Beriefi

(Ari

uid

the

product-Seriee

will
iliat

obvi

be

uniformly

convergent

with

resped

we may then
in

integrate terin-hy-terni.

The general term

the

|>r

and

['

,,

tl.al
I

282

NON-CONVERGENT AND ASYMPTOTIC


integral over the triangle
is

SERIES. [CH.

XI.

Thus the

where
-and accordingly

we have

the equation

uv =
'0

But we have not yet completed our


the last integral
is

the

sum

investigation, because of the series

and
is

(Art. 101) it does not

immediately follow that the series

summable.

It

should be observed, however, that

we can

infer the absolute convergence of

because

|TP()|
if

=
Jo

u {(l

*i)} I-

v (&)\ d>j.

Hence,

we

reverse the argument used above,


rx
r\

we

find

e~%
JA o

e~ W(g)\dg<\ Jo

\u(x)\dx.

Jo

e~y\v(y)\dy

Joo o

e-*\u(x)\dx.\ e~y\v(y)\dy.
Jo

Thus, since the integrand is positive, the integral

is

convergent (see Art. 166).


let

To complete the demonstration of the theorem, the two absolutely summable series
Their product
is

us multiply

given (in virtue of the foregoing) by the absolutely convergent integral

where

106|
'I'll.'!'

Ml UII'LK

\TI<

)="',,+

"-,V

...-,/'

"|

anil

so

ah
|

JO

--'Itrtf)

ami therefore
ix

<

Jo
ahsnint.-iy

W((

gent
:

No*
i^

HT()

i*()

-,4-^;-

-r

<<>...

tin-

t'unction assu-iatrl

with the scries


//-

w=
whicli
is

4- /r t 4- /'._>+...
its

therefore suniinablr: ami

sum

is

r|ual to that of

0+ "'+"'!+ "',4-...
If

(Art

101).

we multiply

similarly the ahx.lutely

Mimmable
'-+'"'

ser;

//
;{

+"

+...

'
,

'i

we can

establish th- oonvergenoe of

<

'out

inning the

].r..cess.

VT6

can

pn\v

that

Jo
|

e.n

for

any

int.

ue

.,f

\;

an-1

aceuplin^ly

the

!'<)hltely

slJininahle.

By

conihinin- the foregoing results with the obvious


>lutely

nn the addition
t'nllnwili.
//'
/'
.

suniinal-i

the

''

p
Ill

IIKllll'

ng

fl
/

si

'/'//

i4

un#

6c

284
107.
series.

NON-CONVERGENT AND ASYMPTOTIC

SERIES. [CH.

XI.

Examples of multiplication of absolutely summable


1.

Ex.

(1-1 + 1-1 + .. .)2 = i


results agree

Of course these
Ex.2. Write

with the examples of Art. 103.


,

(7=l+cos
'

+ cos 20 + cos30+... =0 + sin + sin 20 + sin 30+


...
l

then

C7+i#=l+e + e K * + e s ''+....

Hence (0+ iS) 2 = 1 + 2e ie + 3e 2 2 or (7 ->Sf = l + 2c

+ 4e ^ +
3

By

using the values of

C and S

found in Art. 103,


..

it is

easy

to deduce that

+ cos 0+2 cos 20 + 3 cos 30+ = + sin 0+2 sin 20 + 3 sin 30+.. = 0,
.

cosec 2 (|0),

agreeing with the results found in Art. 110 below.


Ex.
3.

Another exercise in multiplication


directly
2
;

is

given by squaring

C and S

this yields

above. These 2 We can also find C + S* by multiplying (C+iS)x(C-iS); and we can find CS directly.
108. Multiplication of non-absolutely summable series. Hardy (p. 43 of his second paper) has given two theorems, \vhirh are the extensions to summable series of the theorems of Mertens (Art. 35) and

= Jcosec0(sin0+ sin20+ sin 30+...) + JO + 2 cos + 3 cos 20 + ...), = S + sin 20+ sin 30 + .) | cosec 0(siri -1(1 +2 cos + 8 cos 20+...). 2 S 2 found obviously agree with the value of C
2
. .

The second of these extensions will be found in Art. 111. theorem is that it is sufficient to suppose one of the series u v then w will be summable and its sum will be equal absolutely summable We suppose that u is absolutely summable, while v is summabK' to uv.
Abel (Art.
34).

The

first

but not absolutely.

On
(luubl

reference to Art. 106

it

is

plain

that the ditfeience between


is

tin-

integrals over the areas

OA<\ <>M>C

107, 108

Mi LTIPLU

\i

i"V

writ,-

ami

th. MI

/'

'

tin-

integral
limit

c~ y

<(>/}

<(>/

is

roppoted
<.f

tt

<

.n

vergent,

^(

finite
;

upper

//

f<T all

values

* ami A

further we can timl

hat
if

X-jrg
from
in

Now
exceed

arbitrarily small. divide tin- integral (1) int>

may

t\v.

parts,

to
],v

A-// and

fim

\-fji to A;
//.

in

tin-

t'oimei

>>(
i

is

l-xs
.

than

t,

the

Tims, sim-o

</>(./)

we

timl

\D\<
.'0

,(

Now
i<-

/"*
/

<:

so

if

we take

the limit

last

incijuality as A trmls t> infinity.

\v.-

timl

lim

l>

A->
Hut
!

i-

arl'itrarily small,
/>

ami
i>

M
tin-

p.

.")

tin- la-l

in.Mjuality

IT trm- Miih->- lim


I)

Thus lim
A-+-OO

0; and

tin-

intr^ral OTef
I

<>.(!',<'

may again

II'

!ila--l ly th- intr-ial ..VIM- tin\\,m.i\ \\ nt-|iu;. Q Art.

'.!

'
.

lim

/
|

A-*
the integral l.ein- taken over the triangle

OAO.
the same lines

After tin- Stage Thus tli>- M /' \ "-,,-f


r

\.i.-lly
"', -f

tlint
It

one
can
)>e

shewn

that

any

s-

+ ...

is

summahh' umler

the above hyp.thM at

the end <f Art.


r-

H.'I

is

that

"a

-4-

"'.,,

-i

"a

.-

...

may

l.i-

summal>le

an example of the possibility 'hout


t

286

NON-CONVERGENT AND ASYMPTOTIC

SERIES. [CH.

XI.

mable

109. Continuity, differentiation series.

and integration of a sumseries are functions

Suppose that the terms of the summable


of a variable
a,

so that

we may
a\ =

write
( a \?L_.

u (x

^u
o
oo

then, in agreement with the previous use of the

word uniform,

we

shall say that


to

Tlie series

Q^un (a)
o

is

uniformly summable

with respect
e~ u(x,
x

a in

an

interval

(/3,

y),

provided that the integral


interval.

a)dx converges uniformly in that

Thus, in particular (Art. 171, Appendix), the series is certainly uniformly summable, if we can find a positive function M(x} >

independent of

a,

such that

\u(x,a)\<M(x\
while
j.90
I

(/3^a^y)

e~ x M(x)dx is convergent.

Then the following theorems


(i)

are true

If

all

the

terms

un (a)

are

continuous and the series


is

n (a) is
o

n uniformly summable, and ^J un (a)x /n\


o

uni-

formly convergent for any


the interval
(/3,

finite value of x,
00

when

a lies in

y), the

sum o^un (a)

is

a continuous function

of a in the same interval. For then, if a is any particular value of a in the interval
oo

r
1

Km (?un (a) = lim


a >a

e~ x u(x,

a >a Jo

a)dx= ^ e^'lim

|<

[u(x, a)]dx,

->o

this transformation being justified by the of the integral (Appendix, Art. 172).
00

uniform convergence
n

Next lim
a^oo

[u(x, a)]
is

= lim
a-^-oo

~ =u x 2 ^() x = 2 <(o) x
U

*
o

71

IT,

>

o).

because

u n (a)

a continuous function of
45).
r JO
is

a,

and the

series

converges uniformly (Art.


oo

Thus

a->a

lim Q^<u n ()=

which sliews that the sum

a continuous function of u

.it

any

point of the interval

(/3,

y).

109, iio|

hii ii.

IMA n

TI\ \M>

i\

raoB

di

/',/./, r

tht

9am
T[
Jft

<* in

(i

"
\

Jft

it-Hi

be cor/
f..ll..ws l,y

This
tiii)

an ar^uinenl
conditions
of

similar to that of
"/"'/'/////'///
*/////''

(i).

//

//"

saint

'*

^
easy deductiOD from A

00

This thenivm
110.

is

also

aii

\~~2.

Applications of Art. 109.


00

Ex.

1.

If

we consider

tl.

where x
o

is

& complex
that

number

of absolute value not


,

less
f

than
j

1.

we

find

Jn

135*
;

where
It
.

me

that

may be observed
and
e.'nse.jurntly.

the ival pari that |ex '| = >p


if
1

lesa
ft

than

i.

f
,

wlieiv

...

\\i:

fi

the ival part arbitrarily small


is

but positive. thwith rrsprct


is
/5

inti-^i-al
:

for

.'

will
:

conv'i-^f

unifoi
oo

t..

./-,

thus
P6gi
tin-

tl,.-

sum
:

^xn
1,\-

a
1

r..ntiniion.s
<i,

fuin-ti..n

.f

x within

tin-

ti.-l

as
If

is

nhvinus
(
I

f|-,,m

thr valu- of
lind

Ex.

2.

in

we

writ.-

that

and 80
Thus, the integral functi..n
00

bed
,/;,

with

288
because

NON-CONVERGENT AND ASYMPTOTIC


we have
the algebraical identity

SERIES. [CH.

XI.

7^ = 7i(?i- l)...(?i-_p + l)+^l 1 ?iO- I)... (n-p + ty + ^. + Ap^n, where A 1} A 2 ... A p . are certain coefficients which depend on p but not on n (compare Exs. 4, 5, p. 170).
,
,

Hence

\up (xt)\^e*

(t*

+ \A
t

\tP-

+...

+ \A p _

1 \t),

and therefore the

integral

re~
.

up (xt)dt
a.

converges uniformly

if

cos

=1

Thus we may
00 00
,

differentiate the series


'

found in Art. 103 for


is

<25^ sin 710, as


i

often as

we

please, provided that

not a multiple of

2?r.

Hence we

find, for instance,

J^n* cos 7i0 = 0,


i

J?VS+1 sin n6 = 0.
i

Taking O = TT in the we find the results:


1* _ 2* + 3* Again,
.
. .

first

equation and J?r in the second,


1
26 +1
'

= 0,

- 3* +1 + 5 2s+1 -

= 0.

provided that Thus, since

is

not an odd multiple of

TT.

we

find

_ <>>+i)
92+2
I

[Art. 126, Ex.

5].

Thus

in particular*

l_2 + 3-. =
..

(Art. 103),

*The

similar to those used in Art. 103.

second and third of these may also be obtained by means of devices See also Art. 126, Exs. 2-4.

110] In
lik.-

Di;

NATION
sn

\\i,

289

manmT
COB

ih,-

O-CQB 30+cos 50- ... =

sec

l.-a.U

to

tli.'

n-iilt

where E

Buler'fl

3+5-... = i(-l)'#., number (Ch, V Ex, C


I

[Ci

Hi-nee in particular
._:;

+ ;,-_...= _
.

Ex.3.

As examples

m we may tak-

the series

-n
whi<-li (as
in
-2)

are unilMi-inly suininalil- in an interval


ATT.

when- <><o< Thus \\v

s fi &

Now
f^flq-MC* 6
Ji i

2n+l
scries

and

iherel'on' tin-

mi th- ri^ht
J,

is

convergent;
,;),

also,
4, p.

2cos(2?i + l)^/(2?i + l)=

1'

(see Ex.
2

290)

ami
<-..ii

28inc2/<
verges uniformly
l.y

l)c)

(2//

+ 1)

\V.-i.-rstraas's test.
"

--',,+rami therefore
o
./ft-. -

,,-

<-')'

1
In like

manner we
f J

pr-.ve that

dOOtdcW=2

^f *Hsin
i

ami

hei
I *fl,.
(

,t

d,/rt=9 is

V/ _1 V82<-i
T

g -:;io 2

2.

290
Ex.
4.

NON-CONVERGENT AND ASYMPTOTIC


As other examples
first series in

SERIES. [CH.

XI.

integrate the
cos

Ex. 3 from
. . .

of integration, we note that we may to JTT, and so obtain

0+icos 30+icos 50+

= Jlog cot J0,

(0

< <
is

TT)

where the resulting

series converges.

This series

summed

independently in Art. 90, Ex.

2, and Ex. A, 43, Ch. VIII. series the from Similarly cos0 + cos20 + cos30+... = -J,

sin

+ sin 20 + sin 30+


to
.

...

=J

we

find,

by

integration from

TT,
. .

the results

sin

+ 1 sin 20 + J sin 30+

= JO - 0)
(0<0<27r)
to the simpler

and _(i+cos0) + JrO-cos20)-J(l+cos30)+...

= log sin 1 0.
These form
series converge,

and the second one leads


...

cos

+ i cos 20 + J cos 30 +

= - Jlog(4 sin 2 i #).


Similarly

we can

All these series agree with the results of Art. 65. establish the results
sin

0- J sin 30 + i sin 50-

= ilog(sec + tan 0)
=i?r
(0

and
Ex.

sin
.

+ J sin 30 + isin 50+

...

< <

TT).

5.

T If

^=

<S cos x (a 4- nx)

we

get

Thus

The values
Ch. X.
(p.

of

for

cc

= 0,
it

cc

=2

are found in Ex. C, 31 r

259); from these

follows that
.

u=T
6

1 /
(

cosh bx sinh --26 --- 8mn iTol o 'cos 2a \cosh26


,
i

- -

For other

applications, leading to the values of

some

interest-

ing definite integrals, and for the investigation of conditions under which the equation

refer to the latter part of the references given in Art. 96). paper (see
is

true,

we must

Hardy's

first

110, ill

AUK i.- THEOREM,


(An. 5L)
tin-

111.

Analogue of Abel's theorem.


that
oo

<*Un
//,/
,

is

summahlf, ami consider


n-al
ion.

possihilitv

of

summing

/
o

\v

i.

ami

lies

between

ami

1.

\\V have th.-n,

l,y

li-tinit

<&u n =\ e-*u(xt)dx, Jo
t

ami
(

BO,

ehaninn^
fold

tin-

iml'-]>'ml-nt

variable

from

to

xt

=//, sa\

Thus,

it'

17 = 1+0,

su that

is

n-al

ami \>

ire

have

Jo

Now, by

hyp'ttlM-si^. the integral

eJo
|

mvergent; and consequently (from Art. 171. Aj|M-mlix) the integral


Jo
f

Y.\.

~2.

>t

e-9< l +*>u(y)dy
6,

converges unit'nnnly with n-spect to win-re a i^ any positive lumber.


i

in

any interval

(0

oo

Ih-nce

^*u o
(

t"

can le
1).

summed uniformly with


havi- the result
l

resp.

the interval

<>,

Thus

in

partieiilar

w-

')='

Further,

pi-Mvi.lr.l

that

the series

ummaMe

for every integral value of X, the series


f->r all

(^u n
o

n
t

is
1

also absolutely summable for tlu'ii we have limi

positi\.-

\aluesof

less

than

)]

= 0,

(X^O).

292

NON-CONVERGENT AND ASYMPTOTIC


find a constant A, such that

SERIES. [CH.

XI.

Thus we can

Now
so that

[*e-*^[u(xt)]dx

= P-

\e

Thus, since the integrand


to
oo

is

positive, the integral


00

taken up

must converge

(Art.

166);

and so the conditions of

absolute summability are satisfied

by the

series
o

Thus we have the theorem


//
n o^u o
is

*
:

summable, then
t

o^un n is
t o

uniformly summable
1);

for all values of


oo

in the interval

(0,

and

if,

further,

is
o
oo

summable for

all positive integral values of A, then

(^un n
t o

is absolutely

summable for any value of


is

between

and

1.

immediate consequence from this result theorem on multiplication of series


:

An

Hardy's second
the

// neither

u nor

is

absolutely

summable

equation

uv = w
that

is still correct is

(using the notation of Art. 106),


00 00 CO
y

provided

summable.
series

For then the three

Q7*u n t
o

n
,

<^vntn
o

Q^wn n
t

are

absolutely summable, and therefore (Art. 106)

000
t

(0

< <
t

1).

Hence, taking the limit as

approaches
/TV

1,

we have

*Due
Hardy

(p.

in part to Phragmen (Complex 44 of his second paper).

Rendw,

t.

132,

1901, p.

1396)

and

to

111,112]
112.

!'<>V,

An important

class of

summable power-series.
|

Suppose that

where /()
int.-Tal
I'm-

i.s

sueh
is

tliat
j'

f(g)\dg
<

i-

convergent, so that

//

aUo absolutely
A. Art.

>n\ ,-r^ent.

Then, appl\

17").

\ve

have

beam-.'

tli.-

aeri
'/;-

converges uniformly with respect to


is

and

H^I

Jo

convergent
I'm-

Thus

r.oivl's

int.-ral
'

P
[

"..//
it

c-'cfl

P
Jo

Jo
Fr<.in

Jo
t'>ll>ws

**/()<*

be inverted
i-^

imt

order of integration can integral, provided that the real part oi !'<>r then the than a fixed nuniher / i<l); ^reati-r is seen to be absolutely C(n\ rr^vnt ly coniparini: it
Art.
177.

that

tin*

in

this

with

tht-

"
o

Jo

Thus

IJiiri-l's

integral
I'
J

reduces to
C
(
-

/
1

V'/^
.''(

(real

part of
will

x^k <

It

is

-vident
t

that

the

last

integral

he con\ -i-i^eut for


I
;

all

values of x
last

except

real

values which are greater than

thus the

gives a larger region nl' suiiinialility \\V cnuld, of COUTSe, have adopted this than IJmvl'.x integral. " as a definition nl' the sum 1" making another l.y
,|'
,

int'-rai

\rt. 99 above); but it applicatinn of Hardy's prineip. would not have been evident to what extent the new definition

Could he deduce.
L.-

tVoin

Corel's.

in

.!' applicati< ms ni' this integral the paper ijunted above (Art. !ti). and the method has 1" nded )>y Hardy.^ ly th.- aid of contour-integrals,

liny has -iv.-n a nuiul.er

*/'/

.oi.

;?,

ini:,.

p.

294
Ex.
1.

NON-CONVERGENT AND ASYMPTOTIC


We have
1

SERIES. [CH. XL
1

*^L /"V-V- dt= f f "-'[loga/ft?- d%, n


J

Jo

(p>0),

and accordingly the

series
1

+ 2* + 3* + 4P+
is
>

3,

^2

can be

summed

and

its

sum
L

expressed by the integral

lo/dt =
r1
\

rff

[HADAMARD.]

Ex.

2.

Again,

-ji ^ +

=
1

r
/

00

e -* sin t

n~ l

sin [log ( 1 /)]

d,

JQ

-'o

so that the series


i

I_l-l2
is

J.1.92'

**%'

i_

summed by means

of the integral

Ex.

3.

b>a>0, we have a(a + l)...(q + 7i-l) T(b) [ 6(6 + 1) ...(6 + -l)~r(a)r(6-a) J


If

so that the series

can be

summed by

the integral

d-ft*-'- ^.
In particular,
if

[HADAMARD.]

6=1, the integral can be evaluated by writing

which gives
If
is

x is complex the last transformation requires a beyond our range, as it depends on the theory
4.

little justification,

which

of contour integration.

Ex.

Consider similarly the general binomial series

where

can apply the nirihixl

(m + l)

contained between two integers and m + 1 then, if w^O, we ..f Ex. 3 to the series obtained by omitting the first terms, and so pn>\r th.it tla- original series is summable by applying
is
;

Art. 101.

112, 113|

H
of

!'"\N

113.

Application
a
first

Borels

process

to

power-series

in

general.

example, ennsider the series

1+r + .r- + .r +
;

\\'e

Jind

by Bond's

m.

-i

hod

so that

<

^x
?

!'

=\
Jo

t'~

('<(((=:I
'

and
is

this

integral convei--.-^.
1.

pn, \id--d that

the real part of

less

than

Similarly, ly considering the 861166

we

find

tin-

HI.
'

Cc
h
'

a?

a;

<

.,

j.i-n\

id.-.l

that

tin,,r

n-al parts of x/a, x/b, x/c

are

all less

than

1.

Tin-

i.^ioiis

examples are as In-low: thr n-^ion to be excluded is shaded and the


suininaljility
in

thrv,-

t\v

in<licat- the t-ircles of coin'tT^t'iict-

of

th<-

power-series.

snmmahle
I'ln

Similarly, the -jM-cial 861168 e\amin-d in the in the region imlicated in li-airr
3

last

article are

1,-ad

to

the

r.-njeeture

t!

Tin

m/mabili
,-<></
i
i

from

th<

origin

to

of

i/

il'|>"M- ;iny

jMMut

/' tak.-ii

in tln> rr_i:i"ii

tlu-i.

drawn on
-n

-iniMil.ii 'it;.

\m within or
le
.,11-iti--.-.

tlu-

-iivunr
hi.
}i

It

i> tli.-j.

of radius

endooee

".

/'

and

>till

296
Let
is

NON-CONVERGENT AND ASYMPTOTIC


/(.r)

SERIES. [CH.

XI.
if

= 2an#n

be the series; then

the mid-point of OP, and

is

it follows from Art. 82 that any point within the second circle,

where the mean value

is

taken round the

circle
|

k\

= r.

Thus and so we get

an

where the interchange of summation and integration is permissible because the series converges uniformly at all points on the circle since

Hence the
part of

integral
I

f
\

f u*(x, t)\e~ dt converges, provided that the real

on the circle; now positive at all points at P, because OP subtends an acute angle 6 at all points of the circle, and so (f #)/*= pe**, where p is real and positive. Thus the series is absolutely summable at any point within the polygon

(-*)/

x/(

is

this is satisfied

when x

is

specified.

By the aid of complex integrals and a slight modification of the method of Art. Ill, it may be proved that if a power-series is absolutely summable at P, it can have no singularity within the circle described on as diameter.* This is the converse of the last theorem.

OP

It will be evident that, whenever the number of singular points is finite, the method of summation enables the value of the power-series to be found at points belonging to certain and we have thus regions outside the circle of convergence
;

a process for finding an analytical continuation for the powerseries.

summable within any area inside

from Art. Ill that the power-series is uniformly the polygon of suinmdbility. This property completes the analogy with the circle of conIt follows

vergence^ Borel has also proved in a later paper t that if F(xt) is the integral function associated with the given power-series, then
the straight lines (drawn as above) determine on each radius through 0, the limit of the points x for which

Borel, Le$om p. 108. t Borel, Math. Annaltn, Bd. 55, 1902, p. 74.
t

113,_114]

B0i:i:i.>

OWI

i-i.i

IMTIMV

OTHKi: MI:TIIOI
114.

OF BUMMATK

Borels primary definition of the sum of an oscillatory


*,

series.

Writ.-

= "+"! +
6X]
<

".,+ ..+U,,
.

and thru consider

tin-

-*u^=e-u'(X
|
<

).

Hence

(1)

^
inte^n-ul

Thus, vimv the


series

''//</),/./

is

associated with the

Jo

we

see that

if

the

s.-ri-s

<^Un
i

is

summahle, then the

limit

--

;
:

)-

ami

is

eijuul

t" tin-

sum
if

But we have proved that


90

(*U
I

can be summed, so also can


00

(^u

(see Art.
fl

KM

):

and since

ire see that

when

wnmable,

i<><ition

limfx-^xL
rue; andccn
<> ,-'

V-V
'/

=
'"'.)] ./ /*

when
;

^s
i

^u
i
i

and &s
o

bath

www nw
existence
<>f

^
th-

quati&n
limit
l.-adth.l.y

For
o

tin-

{and
l>y IJoivl:*
.'iirsr
tliis
;it

implies

thr

e(.n\-r--.-nce
_

of

this a^-iin

t> t-.juatin

Th.- limit just nl.tain-'d

was
,,ut

<i-i-inal d.-tinitinn

BOggeeted
d.-tin:

hut, as pointed
inctlioil
11.
'

Hardy,
to

tin-

integral

r.m

!,

]<|>|iliril

nuni.noal

calctllni

-ii*

the

least

54
1

s
.i

has

i-aliMilju.

^
|

that

this

expression gives

to thrrc

.li-.-iinals

it

is

possible to use th-

pr.uvss to ol.tain

iiiiiui-ri.-al

loddenbl
j.

1896,

p,

119

i-Ji.

298

NON-CONVERGENT AND ASYMPTOTIC

SERIES. [CH.

XI.

(deduced by Borel from this limit) can be applied to cases in which the limiting process gives no definite value (see for
instance the example of Art. 101). It is almost evident from Art. 100 that the limit
is

the sum,

if

the original series

is

convergent;

but
s,

equal to a direct

proof
<jhoose

is

easy.

For

if

converges to a

sum

we can

so that
S
>

Hence we

find

and

therefore

m-l

where

is

the upper limit of

\s n

s as

ranges from

to

and

so

we

see that

lim

But

is

maximum

arbitrarily small, limit must be zero,

and so (see Note and therefore

(6), p.

5) this

It is

easy to modify the argument so as to prove that


divergent,

when

un

is

lim
Ex.
If

-- =00.

we have
Thus

and
giving

The reader

will

find

it

exercise to discuss similarly the other

eries given in Art.

103.

114, 115, 116J

i.i:

BO1

DBJ

IM

'

299
definition.

115.
I.--

Le Roy's extension of Borel's Roy has r\\ 'n an extension


>

integral

by

modifying

tin-

seri.-s

in

tin-

form

t^+0+0+... +04^+0+0+. ..+0+14+0+.,


where (p 1) zero terms anThe analytical formula i'nr
']!.

n
Mini
is

un and un+l

tin-

tht-n

where
'I'll.-

W =l
ol.j.-et
nt'

iisin-- this .U-tinitinii is


th.-

id

Ilon-I's int

asrs in wliich

Beriee
tli.-

()

-nul.l

oever coim-r^-:

such

cases are illustrated

by

Of course
tinth<-

iimditird

is equal t<> that of nsiii nlivinusly condition of consist. -ncy (in virtu.- of Art. 100).

if

!>/,

is

convergent,
1

its

value

s.-ri.-s.

and BO L-

116.
'I'll.-

Le Roy

independent
h.-r....

definition.
for
tlir

expression

taken
is

sum

of

an

oscillatory

Beriee

"+// +

*/.,+

the limit

Nn\v.

assuming

that

the

series

u^-d

in

this

definition

is

bsolutely

convergent, we can write

T(nt + l)

('

rt^nj

Jo

v
-',':

rh^lan.
I

:,':

thri-rt'..!.-

tin-

1'

Jo

iivrr^nit. -

that

tin-

justify tin- in\ri>i.'i

j>|.i-mlix, can be applied to ml summation.

n.

300

NON-CONVERGENT AND ASYMPTOTIC


is

SERIES. [CH.

XI.

This integral

the same as
00

Jo

e~ x u(x t )dx

and so Le Roy's

definition

becomes

limt->i
t

1 f
I

x )u(x)dx
ll/t

____ Jo o

x l+a )u(x) dx.

If Borel's limit exists, this definition gives the

same value

for the function

x a Qxp(x

x l+a )

decreases* as x increases, if x 1 and a 0, and is equal to 1 for x = 1 and we can therefore apply Abel's test for uniform
;

>

>

convergence given in Art. 171 of the Appendix, which gives


x

r*

x a exp(x
i

x l+a ).e~ x u(x)dx=

Ji

e~ x u(x)dx,

because the latter integral converges. Further,


lim
I

a^-OJO

# a exp(#

x l+a ). e~ x u(x)dx =
is

e~ x u(x)dx,

JQ
finite.

because the range of integration


t

Hence Le Roy's
is

definition coincides

with Borel's, whenever

the latter

convergent.

117. Borel's third

method.
differs

This method of

summing

from the

first

method

in

the fact that the terms are arranged in groups of k before applying the method. It is however of less importance than
the other methods, at any rate from the arithmetical point
of view.
CO

Thus we obtain from

^un
x

the integral

Ie~ o
Because the logarithmic derivate
a

uk (x)dx,

is

\vlii--h

is

plainly

a1) l-(l+a).e + a/a:= - (x and when a>0. #>1 negative

116, 117, 118]


\\

AN

Of

HI.

Ml IMI

[<

here
')

-0
If
its
<

2u n

sun,,

is convcr^ nt, tliis ],r. .cuss of grouping eann.t alter and e.>ns,',|U<-nt ly (Art. M'l) this integral will also

-"ii

verge to the sum. Hut in -vii.-i-al tin- \aluc obtained dep.-inls on / for example, - 4- - -f ... -iv-s <) + <) + <)+...(=<)) if / is 2 or any
;

number, lut

-f

...(

if

is

odd.
\\lii.li

As another example, take foregoing methods.


With

-2 + 3-4
,

-H...

has

tl

-y

the

=2 we get - 1 - 1 - 1 - 1 - ... -'-5 + 8- ..., \vhi< h is the same as


3(1
..

\\

hi.-h

divi-r^s tu
+...) = !

- r.

while

k=3

-2 + 3 -4 + .)-(!-! + 1-1

~i = twhich steadily
that
all

118.

A
-s

further extension of the


that
to
<j)(x)
./'

method of summation.
-

is

a
t<

pnsiti\30,

function,

as

tends

in

such a

way

tin-

integrals

''=
ar.-

po
1

V'

Jo

convergent
f Jo

Then

if

we

consid.-r th-

#(*)(S; N o cn
to be

=
'

it

is

easily
nt.

s.-.-n

correct

when

absolutely

K'-r

then

f Ui

un

\,

because

</>(/)

and

./"

an- jm>itiv.t

and

,i.-,-,,nliiii:ly
..f

the test
tlu-

Aj.jirndix, can be aj)]ili-d and intr uqati.n. Tims

justify tlu- iiivi-i>in

onK-r

.f

sunnn

-^).'/
.

1
u

\- o

')**'dx) cn I

=
(

if

1"

is

oonvergent within the interval

1.

-f-1),

we ha\-

ag

\.,

302

NON-CONVERGENT AND ASYMPTOTIC

SERIES. [CH. XL
as

Now, because <t>(x/t) steadily decreases we have, by applying Abel's theorem for
(Appendix, Art. 171),

increases,

definite

integrals

provided that the latter integral converges. Thus, provided that ^un t n converges absolutely within the interval ( 1, +1), we have

Km 2
provided that the

Jo

f <j>{x)( V^ x n] dx, c
\

latter integral is convergent. Thus, if for two different functions 0, \}s we can prove that the corresponding integrals are convergent, we can infer that their values are equal. In particular, if we can shew that BoreUs

integral is convergent, we can obtain its value (when convenient} by means -of any integral of the form
n

more

,0

^~^
3,

rn
)

rlv
'

which also converges.


Ex.
Consider the series of Ex.
Art. 112 (with 6 =
1).

We know
real part of

that this series


t

is

summable by means
so to find its value

of Borel's integral,

if

the

is less

than

we

could take

Then

119. Euler's

method

for

summing

oscillatory series.

Euler (Inst. Gale. Diff., Pars II. cap. I.) employed his transformation (already given in Art. 24) for summing oscillaThis method is in many cases the most rapid tory series.
in practice ; and Art. 103, before

apply it to some of the examples of examining its theoretical foundation.


shall

we

118,119,120)
Ex.
1.

i:i

u:i;>
\

TRANSFORMATION.

Ki"iu the coefficients

I,

l.

l.

I,...

we

get the aeries of differences


s

0, 0, 0,

simply

-L.
Ex.
2.
in tin<...

[Kxs.

-,

4,

Art

1-2+3-4
1,

2,

3,

4,

5,

...
...

we
and

get the differences

-1, -1, -1, -1,


0, 0,

0,

the

ran>f<.i mati"ii

gj

i-J = i
Ex.
3.

[Ex.

."3.J

l-2 2 +3 2 -4 2 + 5 2 the
il

11. l.-

-i.-mYi-nt8 are

1,

4,

9,

16,

-'

ami

tin-

are

-3,
0, 0,

-9,
...

Eenoe the sum

is

i-| + i=0.
pra-ti.-.:

2.

Ait.

lio.J

Ex

4. As easy examples for Kuler f>r tin.- ni">t pail

\v-

may

give the following, taken

= 1-3 + 5-7 + 9-.. 1-3 + 6-10+15-.. =


.

l-2 4 + 3 4 -4 + 5 -...=
4 4

O/

L [Ex.

2,

Art.

It

will be seen that in all thrse cases tin- n-sults

fmnul

integral

with

those

obtained by
tint

usino-

Ku
that

transformation.
ral

This
can
!

fact

suu-.-sts

ivlation
shall

ohtaincil

lM-t\v- -n

th- t\vn

methods;

and we
120.

investi^atr this point in the following art;

Connexion between Euler's transformation and Borel's


that
[688
>i

integral.

\^

Supposr ival and

~^,<.t
1.

wh'Tt- a

i^

real

and
'24

positive, whih\\<-

than

Tln-n. as in Art.

introduc.

= ".-,

I'

h-

-''
::

rr -

/^''M.

etc.

304

NON-CONVERGENT AND ASYMPTOTIC


these equations,

SERIES. [CH.

XI.

From

we

find at once

a3 = 6 -36 1 + 36 2 -63 = D3 6
Thus, in Borel's integral

etc.

we have

or, expressed in

terms of 6

b lt 6 2

...,

u(x)=

bjxt

can obtain the value of u(x) by summing this series according to columns, provided that the series converges
absolutely.

We

Thus, provided that the series

converges,

we

find the equation

Hence, Borel's integral


e~ x u(x)dx=

is
- t)

e- x(l
Jo
\

[b

-b

Jo

(xt)+...]dx.

If

we

integrate this series term-by-term,

we

obtain Euler's

series

(l-O4
a43

still

equivalent to Borel's integral although of course to consider the validity of the transformations.
;

we have

120]
il'

i.i

EUE8
f

\M- BOREU6

IN

BOB

\i.

Now. Kult*r 8 \n ion (with

series oonvei
;i

can repeat

tip-

argument
tl

ivioufl

changes) to prove that


(

-bpt+l
is

ahsnlut.dy
til--

C0nv<
;

.'

ttd

that
i

BoreTfl

intend

is

to

MUM

have obtain'l
xeries
\

th-

//

Kiders

series is cow<
;/'
t

'!

2ant*
1.

is

l/

BorePs

///////

is

real

//////

summable amd l"


f

"il*
It

is

natural
.

t.
it

iKjuiiv
<1<>^
tin-

it'

th

lls

for

complex

\ah;
iii.-th<><l.
i

lut

not

S.M-HI

\vin_n'

to

fact

that

the same jM.svil,!,. to apply Irimna Art. the NO leads


!'

to a

litliculty

here.
h.-ive
t

Thus we should

tl

~
(

irb>t

(T=Tr^

^=r
lY M A
fl

and

t..

coiisidrr the continuity of

as

A^oc.
and
.

Tin.- kMiiin;i !-<

juirrs tin- conv.-rua-nro

>f ^/-,,

)f

^; A,,-A n + ij; no

we

ll

.,

hy

diif't
if

integration.
t

Now
thus

= r + is,

|c-A(i-0|

= (\ = e-Mi-r) and 2An -Aw+1 = e-Mi-'0fA i-r _]


}

8-|l-< -(!-/)
-

K-a-t
A.
I

\\hcn

.<

i-

nut wro,

and consequently the sum


thfl

1 A,,- A M + 1
continuity

j-*\

\vitli

lenini,.

the

On
We

the other

hand,

if

\\

assume that the


/)
ITti.
(/

sei

convergent,
<-an

\vh.-i-e

p = \t\/(l
I',

IK-HILT

tin-

apply
;,

te>t
it

of Art.

Appendix.

real part of Km- since 2 lt n p*

t'nllows that

2(
*

-D
is

It
'

should
uile.
i

l>e

m>t.

.I

th.it

if

on

ita

s.

306

NON-CONVERGENT AND ASYMPTOTIC

SERIES. [CH.

XI.

converges absolutely and uniformly with respect to x in any finite interval. We can now infer from our test that

converges to the same

sum

as

bn t n

2(

1)"
(l

.v

n+1

t)

We

have in fact
n\

Jo

Jo

n\

and the

n last expression reduces to \b n \p /(l -r).

Thus the
is

series
o

n convergent, because E \b n p converges. article quoted, we can write


\

Consequently, in virtue of the

Jo

From
p<^l,
\t\<^l(l
is

Art. 50,

we
is

see that the series

2
1

bn

n
\

converges
are

if

say

thus
that

Borel's

and

Euler's

r),
I,

eccentricity

whose focus

to say at all is at the origin

equal if points* within a conic of

sums

and whose

directrix

the line

rl.
bounded by the
circle

Euler's series will converge in an area

the auxiliary circle of the conic; if I 1, |, |tf| the area is within the circle because t = is inside the circle
1
is
|

which

<

if

If

>1, the area is = 1 1, the boundary


is

outside
is

the

circle

because

=
;

is

so.

to the left

a straight line (V = J) and the area is of the line in the ordinary form of diagram and then a parabola.

of course the conic

By appealing to the Theory of Functions, we can now see that Borel' 8 integral and Euler's series must be equal of allpoints where both converge. But there is no obvious means
of determining the region of convergence of one from that of the other, as will appear from the two simple examples

given below.
* In case/>l, the conic is a hyperbola, and uc must take only points within one branch; that is, the branch for which r< 1, and this is tin- /iff-lnim/ brunch to the right of the origin). 1 in the ordinary way of drawing tin- diagram (with
/

120
1

KM
1.

Ex.

Tfckeo,

this gives /fo N


I.

-a N -a H + 1

M/ =J, ZPoo^t,
t

etc.,

dor's series is
I

*
tli.it

r _ oa if

|*|<2|l-f;

i>,

in

the region outside the circle


|

tin-

tw.i p..int-

lint

Borel's integral gives

ami

tin- int.-^ral

wh.-n

tli-

n-.-il

I.--.*

tli;m 2.

In

tin-

diagram,

tin-

if_ri"ii
<>f

"t

convergence of
I'.nn-l's

Killer's series

is

\\.-

rdej and that


the
linr

integral

is

the space
i.

t<

the left of
}>y

Ml.
^ivcii

Tli.al..iv.-

ai--a
i-

in
tin-

\\liidi

they air pruvt-d


tinl-ft
-f

].

t-.jnal

the

in-tli.(l

ar-a

t-

tin-

hyj,.-;

-r).

(Tea in

which the twn


i<

ly

o|ual
In

is

the area which

the the

l'-ft

if

All and
i

..nt-idr
;

this case

one or
'

in>
-

1"-

applied

=
_'.

'2.

Tlie "i -i-inal

thin the

\\hi.-li

drawn. Ex.
2.

Consider

n.-\t

tli-

case
hl

a*-L>X
:

Th.-n

:.

Thus
I

_L. \-r(\
which
Hut
.-..

**
ihat
in
A k ,

nver^es
is

if

,'

i-

the aiea in-ide the circle

whose

diameter

tin-

liir

the points

-1.

intcuM-a;

and

si.

the integral oonverges wli

>

than

J.

308

NON-CONVERGENT AND ASYMPTOTIC

SERIES. [CH.

XI.

In the diagram, the interior of the circle represents the area of conand Borel's integral converges in the space vergence for Euler's series to the left of AB. The two expressions have been proved to be equal in
;

FIG. 36.

the area within the ellipse

(\t\

<(! -r))

but they are actually equal


|=J, which
is

at all points within the circle.

The
Ex.

original series converges within the circle


3.

not drawn.

The reader may examine


is

similarly any example such as

where k
t

In particular,
:

a positive constant. if a n = I or n, Euler's series terminates for any value of Borel's integral converges only if the real part of t is less than 1.

121.

Numerical examples of Euler's transformation.


transformation lends
itself

Euler's

very

naturally to

the

numerical evaluation of non-convergent


Ex.
1.

series.

As an example
* 2

let

us take the series


is

-(* +

+ ^ 3 +...)
.
.

[which
24 +

equal to log(l-),

if

|*|<1]

and write t= -2.

We obtain
the form

2 2 -f J 2 3 ;

. . .

and we
the

shall utilise the calculations

thus we sum already made in Art. 24 which give - 19*314286, to six decimals.

first

eight terms separately,


in

The remainder can be put

where 6 6 n b 2l ... are the differences given in Ex. 2, Art. 24. But we can obtain a general formula for these differences, and so establish
,

the convergence of the transformed series.

For

D an =Dan -Da n
z

1.2
.

and so on

leading to

120, 121)
<

M
tin.

Mi:i:i<

\l.

I.

\ \\IIM.I
less

LEB
than
|

'I.

-aily

expct
,

ft

1,

and so the tranBformed


the values

series certain

1\

Mince |tf|/|l
v

= |.
liiul

With

tin-

ditl.-i'-ne.-s

f.iiin.!

\\e

7407

898
81
7

_
II
1

.-'

It

\\e

;i|)|ly

tli-

foinmla given
2

that the remainder


.

lies

between J4 and

7
,

of the last

t<

-i

r>

-t

.un.-.l,

that

we can take

4-.

..

= -119607.
should be

Now

^'(-1 !!); .7)

i;n-llL'!>,

an<l BO tin- Mliefl

L129

I'.Kil

:i

1-0986,

whii-h agrees with log 3 to the last figure.

Ex.
.*

2.

As a second example,
US
!

for

comp;iri-"ii

\\itli

Borel's

numerical

1ft

'-i'-o'
ami
au'ain

nrl'l-o-^
tin.-.-

if

IKII
shall

write
nf

-2.

Tin-

-HIM

tin-

lirst

^-iiiis

i>

1*5

and we

apply the

transformation to
\\ '.-

tin-

following it-mis.

liave
1

.3...(2n-3) 1.3...(2-1)_ ~2. 4. ..(2 + i>)~ 2.4...2n

3)

2.4.

and

priM'frdini: thus

we
,

find
1

2.4.6...(2wH-2jD)

Thus, jnitting

= 3,
.10...(2;^

:.iws that

/
.

decreases

asp

increase^ so that Kuler's


t

erges;

and fonsdjiu-nt ly the integral of \\\- '- niftli mention because Boivl himself does not
-\

\M-a-nt
s.-.-m

this

to

have succeeded

in pi-oviiiL: this dir.-.-tly

\Ve have,
1

d ready established

the convergence

An.

2,

i),

2,

ivn;.

p,

ll'i.

310
In our

NON-CONVERGENT AND ASYMPTOTIC


case,

SERIES. [CH.

XI.

we have

and

so on.
.

Then the series is l'5 + 2 3 (c + C! + c2 + ...). The numerical values given by these formulae have been checked by we have then direct calculation of 3 a 4 6 ... and their differences
,

= '062500 c, c2 = 15625 c3 = 5208 2025 C4 = c5 = 868 c6 = 398


086624 378
087002

c 7 = -000192 96 c8 = 50 c9 = 26 c 10 = 14 cu =

-000378

rough estimate

of the

remainder gives
.

c n , so

that

we get

= '087021,

the approximation being probably in excess of the true value. Thus we find for the sum of the original series
1-5

Now

+ '232056 = 1-732056. ^3 = 1-7320508...,


;

it is closer than Borel's, so that our approximation is a very good one although he works to 7 figures and uses 34 terms of the series.

Ex.

3.

Euler calculates in this way the value of the series 3 + log 10 4 - ... 210
Iog 10
Iog
,

starting to take differences at Iog 10 10.

He
!
!

obtains 0'0980601.

(Compare

Ex. 10, p. 351.) Euler also attempts to evaluate 1 - 2 + 3 - 4 + by this method, and he obtains '4008..., but although the first and second figures agree with those found in Arts. 98 and 132 (1), yet it does not appear that his method rests on a satisfactory basis here.
!

122. Cesaro's
It

method

of summation.

has been proved (Art. 34) that when two convergent series are multiplied together, the product-series is at most simphi
indeterminate.
the limit
(1)

That

is, if

denotes the

sum

to

(?i

+ l) terms,

Km

121,122]
is

\.\\.

mi
that

and
in

is

Unit.-.

\\'<-

have al-o
.!'

^.,-11
(

thU nn-an-value

the theorem

l-'rohenius
lt

lim (i' x*).


Cesi\ro has proposed to adopt the limit d'-tinition nf the sum of a non-conv
exists, as the
-ries
;

or,

more

generally, he defines

tin-

^\\m as
|P

Hi"
n->-oo

{.S.

'M.

W },

wh<

8
an.l
<

.!

>i

course the limit (1)


i

is

tin-

special

case of (2) which

is

liy

putting

= !.
it

With

tln-s.-

drtinitions,

is

cvid.-nt

that*

V,
an.l

:=(\-X):

1^
(1

NMW
and hy
'

-x)- (r+l} = (l+x + a*+...)il-x)


^,-e

juatin^ Cddlic:

that

Thus
in

tlu-

drnoniinator

is

equal to the
,,'
;

sum
in

of the coeffic
if

the ininicrat.r of N,

and
.
. .

BO,
-

particular,

=
">',,

=
>'j

>'.,

'

/(

we

see that
i

they are
'"'

all

e<jual

to

Sn /A n
(r)

(r}
.

Furtlu

= (l+-'.-+
-x)- (r+11 <:

we
and

h;

c<.nstM|U.-ntlv

.+(H

J" +l

(l-,)-'=l + r, +
iiiiiltiplii-il
l.\

^"
:

the

coeffit

tion.

312

NON-CONVERGENT AND ASYMPTOTIC


also

SERIES. [CH.

XI.

Thus we can write

it is evident that the effect of r is to more increasing give weight to the comparatively early terms of the series 2uw it is therefore obvious that by increasing r we may hope to counteract the oscillatory character of the
;

From

the last expression

series

2un and
,

so to replace the series

by a convergent

limit.

Now,

since

we have
and
similarly,

2Sn(r+1}xn = (l+x+x

By

n equating coefficients of x

we have

if

Consequently, from Stolz's theorem (Appendix, Art. 152, II.), lim [Sn(r) /A n(r)] exists, so also does lim [Sn(r+l]/A n(r+1) ] and
;

n >oo

jj,_^.oo

the second limit


exists

for

any
is

is equal to the first. Thus if the limit (2) value of r, it exists also for any higher value

of

r.

If

Cesaro
123.

calls

the least value of r for which the limit (2) exists, the series "Zu n k-ply indeterminate.

Extension of Frobenius's theorem. We have tacitly assumed that the power-series used in the last article would converge This is, however, absolutely. (r) tends to a finite capable of easy proof whenever 8n /A n limit I', for then we can fix an upper limit C to the absolute value of this quotient. Thus 2$n(r) #n will converge absolutely when 2M n(r) #n = (l x)~ (r+1) does so; that is to say, for values of x such that \x 1. Since 2u n x n and 2sn #n can be derived from 2Sn(r) xn by multiplying this series by (1 x) r+l and r n n x) respectively, we infer that I,u n x and 2sn x are also (1 1. convergent for x
(r]

<

<

122, 123, 124

Ml.
1

AN

Al.

818

in,

ly Art.

-"'I

\\-

B66 that

25;
11111

=shm

=
so that
liia

-.
'

=/

pl

\\hirh

is

tin

<

'tension of Frobe //' //>'* //,//,/, .r thi> theorem are air<>r.l,.<i i,\- M\-.
t.
!

:;

i.-sult

aj]).-;us

iK.vrl

lnt a closely

ivhit.-.l

th .....

h;i.s

been

=:

( J b+'i

+ -+-) - 5
!

ro
;

=71 -f- 1

7'

'"

= --r [7
~

"-'4-ri '^-f...

7'

tlu-n

if

lini T,, n >oo

/.

II.'ld.T has
<

|,i-..vr,l

that

lin: x

:.

It

stains

1;

that
!"

tin-

in. -ans

found ly

.tlmd and ly
t..
/

tli,-

>aiii.-,

and
|ti'<.

this lias lu-.-n j.n.vi-d uj.

J.
t

Ht.lder's process must r l-'or hiu h-i vallMI

Kn'])p* has
It

vrd that \vh-:u?\vr H"ld<-r's

m hod
tlian

gives a limit, Cesaro's


H.'.ld-r'>
'

will j^ive the saint- limit.


i>

usually l.cttn- to apply Ceskro's


an-oiiiit
!'

method

a given series on
124.
It'

it>

-_

ri-at-r >implirity.

Inferences from Cesaro


\^

definition.

^LH
it
,

/'-ply

iiil<'trniiinat'

ami

lias

tlic

sum

/,

ii,

EolloWS t'n.m thr

.l.'tinition

that

Thus

\v-

can

writ,-

SflJ

-^.i
-h

\\hri

that

(1)

lim
tli.-

=0.
i.l.-ntiti--

Benoe, remembering
tin-

of Art.

rj-j.

we have

n->\v

identity

(2)
i,

,1-0
/

'(i-x

314

NON-CONVEKGENT AND ASYMPTOTIC

SERIES. [CH.

XI.

where the
&bove.

coefficients in p(x) are subject to condition (1) given And, conversely, if (2) holds, the series n is at 'most

Su

r-ply indeterminate and has the


It
is

sum

I.

to be noticed that (2) does not allow us to write

where
For we have
<r n =
/o n

lim{o n /-d^~

1)

-p n _i and A
|

*~ l}

= 0. = A * - A^,
(
|

and

so

if
|

/_!

< eljjtj
|

and

/>

< *A ^\
(

we can
or

only deduce that

|er n
(

<e{
}

A^ + A^}
}.

*~*

cr n

/A

)<{!+ (2n/r)

Thus, owing to the presence of n in the factor multiplying r ~ l) be inferred that limJ IA^ \=Q
if

e,

it

cannot

we know nothing more than

that the condition (1)

is

satisfied.

simple example

is

given by taking
<r

r=2 and
. .

Then
so that (Tn/A^ oscillates

(x)

=I2

between

and +2.
identity

In particular,
(3)

if
(r+l)

we have an

(l-x)-

2unx n = l
series

where P(x) is a polynomial not divisible by 1 x, the 2un is at most r-ply indeterminate, and has the sum I.
For
if

there are
r

p terms
than

in

values of the coefficients of

is the greatest of the absolute P(x\ and if P(x\ it is clear that the coefficient of .v" in

P(x)(\-x)-

is less

MpA
by A n
(

(r

~ l)

and

the quotient of this

is

Mpr/(r + n\

which tends

to zero as

n tends

to GO

In practice the most common form of identity - x)- (r + ^unxn = 1(1- x)~ (r+l} + P^x) (1 (1 (4)
l

is
2 r

)-

from which we can deduce the same results as from

(3).

theorem on the multiplication of The following lemma will be needed:


125. Cesaro's

series.

IfZu n
upper

2t> n

are

wo

series
(

limit, while v n /A n

such that \u n \/A ? has a finite tends to zero, then the quotient
( . . .

(u v n + u.Vn^ +
tends to zero.
It

+ unv )/A ; +8+1}


(

may

be noticed that either (or both) of

?,

may

be zero here.

124, 125|

!.l

ll'l.h

[(

we can
\

find
II

aini
-

Um

Viilu.-s

ami

l^l-

>m.
;.:,*"

Then
ami

|2*,.,*|<r

|2*l<tt
rl
A',,

.ayH)

Now
ami
10

i> th,- ,.,,,is

'x(l

= (l

/;

equal bo

Hence

vK^l'

-",

Consequent lv,
and, since

lira)

-*^

2u
,

v \/A (r+*+"~
it

can

1>

tak

a> small a> \\- please,

follows that
==0.

--i
\\'e

-..~

" +1>

proceed

HoW

to the pFOOf of C'


aii'l

Suppose

that 1"..

^''

respectively, and that their MUMS are


I

are /'-ply and *-ply V BO that I'.


.

i:

iliate

we can write

-^- /)

.1
wliei-.-

' +1)

+
=

o-(

p|
liin
-I

= ^/r,,.'[ft

.cli

that

f=0,

HIM

[o-,,/^!^}

Ti,

,1

>x(2
i

S
,.,-}-...-

wl"
e

(l-)
=

rn
in
in
/,'</>
/"(
l

where R(x)

r<

-.c)-

(r+

V-'')+ I'd-.'')
\\
,-

'

To each
n alo\.i->
/

of
:

the
i',

terms

can

apply the

lemma

.r

instance,
if

BO

/'(1-r)-"

SI
(

we have
Similarly for the

lin. ;A',,

A ?'
Ri

+l}

}=().
in
;,

two

..th,-r

termJ
.1

we have

lin. ;/,'

=0.

316

NON-CONVERGENT AND ASYMPTOTIC


see

SERIES. [CH.

XI.

Thus we
and that
its

from
is

(1)

and

(2) of the last article that the

degree of indeterminacy of

sum

equal to

2wn is UV*
:

not higher than (r+s +

1),

The product of a series (whose degree of indeterminacy by a series (whose degree is s) can be formed as if the series were convergent. The degree of indeterminacy of the resulting series cannot exceed (r
This leads to the theorem
is r)

126.

Ex. 1. The series has the sum \. Ex.


if

Examples of Cesaro's method. is 1-1 + 1-1 + 1-1 +


..
.

simply indeterminate and

2.

The

series
.

1-2 + 3-4 + 5-6 +


find
s

..

is

doubly indeterminate and


>
;

has the

sum

For here we

=1

>

we

write r = 2 in the formula of Art. 122,

i=-l> S 2 = 2 *3= 2, ... we have to evaluate

and

(n+ 1)50 + 715! = 1, (i-l)*2 + (ii-2)*3 = 2, = Thus, when n is even ( 2m), we have

Now

etc.

+ ns + ...+= 1 + 2 + 3 + ... +m + (in + l) = $(m + I)(m + 2). When n is odd ( = 2m + l), we have again (n + I)8 + ns + ...+*=! + 2 + 3 +
(n + I)s
l
.

..

Thus the

fraction has the limiting value

It will be observed that

- 1 + 1 - 1 + ...) 2 and, according to Cesaro's theorem of the last article, (1 should be at most trebly indeterminate ; and this of course is verified here.
Ex.
3.

The

series 1

2 2 + 32

-4 2 +...

is trebly

indeterminate, for

we have

Hence

(1

x}~*(\

-2 2^ + 3% 2 -4 2.r +. ..) = (! -#V,


}

which obviously
Ex.
4.

satisfies

the condition (3) of Art. 124 and gives l-2 2 + 3 2 -4 2 + ...=0.

In like manner, we find

Now flz^L

has the value

-i

for

# = 1,

so that

we have

_
:

3.

where P(x)

is

a polynomial of degree

* The reader will probably find it instructive to use this method to establish the theorem for r = 0, = 0, already established in Art. 34. Cesaro's treatment of the general case is on the same lines as the proof given in Art. 34.

125, 126, 127|

EXAMPLES OF

Efi

\l:-

Ml

BOD,

we

find
... + i)=/'(.r)(l--*V, r...= -J,
t*
t

(11

I
.d<

ries

being q
i

minate.
il

Ex.

5.

diili'-nlt

t<>

MM- thai

the series

1-ir
i'ly

*-...

indeterminate, because

:;

we

find

<"+[(l_,

-..)-']=

WOO--'

is

a polynomial of degree (r-l); and

-ftGI
-0,
if
/

i*

//(A/.

Bei

lie >MIII

of th- Beriei

i-

>'

-ARO.]

Ex.

6.

The reader
(1-1

will

have

little

dihVulty

in

vnifyin 4 K
:

inultij.licati'.n.

F.r in-tan.

we

lind

+1-1
;-'

or

+...)(! -i + 3 -4 + ...)= 1-3 + 6 -10+... 1-3 + 6-10 + .. + ...--i(l- 3 + 0-10+.. .)-(!- 2 + 3 -4 + ...),
.

and so
Ex.
7.

|=0.

The

follou

!+
sin
(y

''

+ sin 2^+sin3^+...
3^ + cos
~>0

+ ...=J, ^oot^,
. . .

cos ^ + cos

= 0,

-.-

n-^ults agree with th.s- f..und liy


d'.\

I'-uirl's

ni.-th..d

(Ait. 103).

evaluated

l.y

1.

-ml

M-I

ii-in-

tin-

ni'-an-

valu- jn

'

>'-o.>).

tputcula

127.

Limitations of Cesaro's method.


process
let
i,

Tlu- iiiran-vulu.'
6

li;i>

ratln-r
,iinc

narrow limits of
type of
ha
,

01
sc;

this,

tli.

which can

l.r

Miiiiinc.1

in

this

way.

\\".-

that

^^
so that

aI

(i

,n

which givee
3

--+l>N:,"

+ "'t

-'terms.

318

NON-CONVERGENT AND ASYMPTOTIC


if
1

SERIES. [CH. XL

Now
value

approaches a definite limit /, we can find a such that when i/>> r, all the terms /A ? lie
S^V-d-iT

between

(l

e)

and

(l

+ e).

Thus

The

coefficient of

in this expression

is

easily seen to be zero

in virtue of the fact that

so that

Thus
\

U
un

That

is,

< e.2
^;

r+1

A
(

(n >p),
().

and

so

limuA =
p

But
so that

Iim4 n
il-D

/7i

=]/r!.

Thus,

we have

the result

which gives a necessary condition that


r-j9^2/

the series

2^n

??ia2/

te

indeterminate.

There are many series surnmable by the former methods which do not satisfy this condition in fact the simplest of all,
;

does not satisfy


sense.

it,

and

is

therefore not

summable

in Cesaro's

128. Borel's original

We may
s
,

regard the most general form of


as given

method contrasted with Cesaro's. mean amongst

s v 82

...

by

127, 128, 129]

HO|;I:I.>

BUM \M
i

V
tin-

wh<

never

sei

*w

are

<

II.

-I-.-

we may

'-itln-r

<h
<>\

\\\

every

r all,

or

(iii

function

\ariall!'..

./

which

made to

bend to Borne definite


alternative

limit.
nil.

may

h<-

Dili

and
-

ad
Cesaro's thM ami second mean dou Mr indeterminacy come ondei
i

of

BU

i)

by taki
it'
i<

xn ^
or
wlii-iv

ii'

O.
0,

#,

= ./+ 1

1'.

ii

if

/'

> >
i

is

al't.-r\\ai-U
tin-

m;i|.-

t.

t.-n.l

mvMi^h
'

valu-o.

Ami
(

<thcr im-ans are given 1'V -imilar, lut In these rases thr osoilla t'oriiiulju-.
1

M-i|ll.-nre

>'

>

^
th.

rnMVerteil

into
.

,y

th--'n-.l

rharartrr of

that

little

wei

he t'Tins with hi-'h

inilirrs.

which an- the

ms that
'

are of importancr in studying tin- Beriea To meet this <litlicu!ty. IJorel chooses the factors
increase
,-n

/(

so as to
\\itl:
<1'

firsi

maximnm (whose
steadily
./

pi

-sit

ion \ari--

ami

ai'trrwanls

to

The
part
in

|K.-itin

the

maximum

recedefl as

increases,

which ensure^ that

tin-

terms

with hi^h indice^ play an important mean.

determii:

ilitions

Thr mo-t natural type ..f Function which inciva.se> so Ion-:

-i'yin^
afl

t!

/'<''. and

tinally

deCP This

givefl

ry rapidly. ihr m.-an afl


.'f

-:+..

r/

2TT^r
;.

n+->
M 4.

which
129.

is

the definition

Connexion between Borels sum and Cesaros mean


L*:J)

for a given series 1 \Ve have pro\rd Art.


(

tliat

i:

limit

/.

thru
liiii-1

-/.

320

NON-CONVERGENT AND ASYMPTOTIC


we have proved
t^>i

SERIES. [CH.

XI.

Further,

(Art. Ill) that


*

lim (&u,nt*) =

Jo

e~ x u(x)

doc,

provided that the latter converges. Thus, when Cesaro's mean exists, it gives the value of BoreUs integral, provided that the latter is convergent; the integral may, however, oscillate

between extreme limits which include Cesaro's mean.* Obviously the same result is true for any mode of summation given

by the

integral of Art. 118,

Hardy has proved in his second paper (see also the small type below) that if r=l, and if a further condition is satisfied,
the
for

convergence
of
r,

of

Borel's

integral

will

follow

from

the

In general, with higher values Cesaro's mean. seems likely that the same condition will suffice to deduce the convergence of the integral from the existence of the mean. However, the algebraical difficulties involved
existence
it

seem at present too formidable to make

it

worth while to
l.

write out a rigorous proof, except for the case r =

We

have seen (Art. 114) that the existence of the limit

implies the existence of Borel's integral, and that the two expressions are equal and we shall prove now that if
;

lim
n
*-oo

(*

+s +
l

. . .

+ s n )l(n + 1 ) =
=
l,

Z,

then
subject to Hardy's
Borel's integral

\\n\e-* L V ^oo
condition.
It

(^sn -\\ nl/J


is

then obvious from Art. 114 that

must

also be equal to L

Now
as

suppose that

//,

h are the upper and lower limits of


"n

= (*0 + *1 +
while //,,

+ *)/( W + ] )
/<,

n ranges from

to

oo

are those of

<r

<rm+1

cr m+1

...

then, from Art. 153 of the Appendix,

we

see that

It

is

of course understood that Cesaio

>

mean

exists.

If

we

are dealing with


(see

series

in general,
127).

the

mean may

oscillate

though the integral converges

Art.

129)

BOBEU8 BOM AND


'

LN,

Letween

//

in(1

,<//
rr

//

)//!..

+ (//.-*....
f.-im.

whentii^

./'/:
'

and

<,
:

i-

tin-

maximum
.n

H
oo,
t

be equal
!%,

to

mi.

Lrral
',

and

making n tend to

+ *, + ... +

baa

th.-

limit

and so we get
(/.

and
No\\

>*.-.-.[(//-//
any
valiu- of
///.

+ (//,-/

].

M.inlinx t"

we ran write
'-re.,
i>

*>/-..
wh.-r- <,.,->0 as

//

increases.

And

tlu-iv

a constant

.1

>urh that

/,,-/,--.!,

//-//

tlm>

\\c

find

k
.

^+'-'D'(7,;^
Stirlin.:
-

}
we
see that

if

we use

formula

'

Ap^-ndix. Ait. 179),

''

l'Mr-'

-(/'-

wln-n-

/'

tt

-nds to
/ Klo g ^

as

Now
\vhi-i-

<
log [ e

1.

'

"(7^):]-

A"

d'-in-t.-

liuiiil'.-i
1..

n..;

lnit

ahvay>

<-.,ntain-d

rily tin- saint- in all ini'iua! itain ti\-<l limits, such as "0001 and 10000.
I

Similarly,

wo can

ju-ovr that

if

a> the intr^ial part

!"

^p,

^-W<
!'

ll>

i-

th- integral

j.ait

^
s

n
lilll <e

322

NON-CONVERGENT AND ASYMPTOTIC


that the condition lim
;

SERIES,

[en. XI.

Hardy remarks

wN/m) =

is

for the truth of the theorem

in fact, it is not satisfied

convergent series (which has virtue of Art. 100).

the

sum

and

is

by no means necessary by the following therefore summable in

Let

*,,

= Ji-,
-a
)/(
i

and then

n+ l)

Thus by Art

11,

we

see that

so that

lim

nx/w)

= oo

although lime n = 0.
(I.e.

But Hardy shews by another example


lini(e,iv/tt)

p.

41) that

if

the condition
;

=
is

is

broken, the series 2w,

may

not be summable

so that the

condition

not merely a consequence of the special presentation of the

argument.

ASYMPTOTIC
130. Euler's

SERIES.

use of asymptotic series. most instructive examples of the application of non-convergent series was given by Euler in using his formula

One

of the

of

summation

(Art.

95)

for the calculation

of

certain

finite

sums.*

Thus, taking f(x) = !/#, a =


1

l,

n, Euler finds

+v+ +

..

+ i=log7i + -^+ -j^ i-^L+.-.+const. ATI 2 4?? o??


/

^ti

A
7?

Now
because

this series, if continued to infinity, doe^ not converge,

we have

(Art. 92)

2r(2r

1)

^n?''

~'
and 2

Z
but,
if

r>3,

2_<llBr

(see Art. 7),

Llliit

B,^

150-

hence the terms in the series steadily increase in numerical value after a certain value of r (depending on n). It does not appear whether Euler realised that the series could never converge
*
;

but he was certainly aware of the fact that


17V> (Pars Posterior), cap.
vi.

it

does not

Innt. Cole. Difl.,

129, 130|
6

r:i

L8YMPTO]
the

i-

for
tli.

/<

1.

He

rinploy-<i

series

~1<

to

ralnilatr

\\

liidi

h.-

Hoarded m

th- get

V
The reason why
hat
///.

ih

ran

In-

used, although

u.,t

con-

MaAntd
is

/'//

thorn

t>

6ft

Tin- trutli
thiM.r.-iu

<>f

ti

UK-HI
h,-l<,\v

follows at
Art.
S

in once from
>

tip-

U''HTal

|ir<vi-|

(see

1M1)

'

thai

This

fact

i-nalil-N

gn and lias tinus to see at on<-'- that


a
;

all

of Kul-r'>

are correct, after


:

making
Kul<-!-'>

f.-w

uuimpoi-taut changes.

Ex.

1.

-i-

...
/'

.17.

324
Ex.
4.

NON-CONVERGENT AND ASYMPTOTIC


Taking f(x) =1/^
2 2
,

SERIES. [CH.

XI.

prove similarly that


2+... to

00=
o

-0951663357,

and deduce that


Ex.
5.

^- = 1-6449340668. 6
similarly that

Shew

l+^s

+ Js + p+... = 1-2020569032.

Euler obtained in this manner the numerical values of 2l/wr from r = 2 to 16, each calculated to 18 decimals (I.e. p. 456); Stieltjes has carried on the
calculations to 32 decimals

The values

from r 2 to 70 (Acta Math., Bd. 10, p. 299). to 10 decimals (for r = 2 to 9) are given in Chrystal's Algebra, vol. 2, p. 367.
Ex.
6.

If /(.r)=l/(

+ .r 2

),

prove that

sin 2(9

2U

P + ri2 /

Z(e

^-l)

2 + -r 4

sin 4 0sin4#
:

^.

where tan0=/w; the constant is determined by allowing n and using the series found at the end of Art. 92.
Ex.
7.

to tend to GO

In particular, by writing

=n

(in
l

Ex.
^

6),
'

we

find

, =4

.(-,l T+

_L

4-

By
Ex.

= writing n 5, Euler calculates the value of


8.

TT

to 15 decimals.

If /(#)

= log x, we

obtain Stirling's series

which
131.

is

found differently

in Art.

132 below.

The remainder

in Euler's formula.

In virtue of the results of Art. 95 (small type), we can write


(1)

<

''

>

-*

where

130.131,132]

i:\A\l!

/,

R
:

in virtu./,'

..!'

tin-

..-mid in that artid.-:


,',.
<

and
re

in th- U

he

pol\ nomials

'>"th

of comst

u
(/,.>'

th.-ii
i

Thus if we 088"
"/<
'/..

-urne

awl

////

>/

.///

i;i/n,

thr

/;.,

<u
vm
//"
//*./:/

Thus
(I)

//M

u wwerricatty
ii.it

!<*

///"//

is,
.

til--

s'-ries
lea
<>!'

so

ol)tain.-il

/rm, a-n^ /ta l,a^ th,- same

the

same

prop-

as a conv'-i

ilfci-.-.-^iiiL:

trnn> which
conv.-r-vnt

luivt- altt-rnate

Theoretically,
]ushr.l
IM

however, an arbitrary degree


in

tin-

seriea

can be
as

of

appnxiin.-itioii, (1)

while <l>
[iiitc

cannot:

Init

pranio-

tin'
i-

s.-rirs

usuallx

good an approximation as
132.
(

necessary for

ordinary cak-ulati
series.

Further examples of asymptotic


~<

Th,- intr-ral*
oftrll

<U
t

dfllot.Ml

l.y

thr M'lllhol
it

li(

In

many
u

proUK-ms.
for
/i

ini.'-ral

///
<if
<

important to calculatr tin- value of vahn-s of x. Nan \vritet


is
i

'H/
i-

(1

-'

<if
<

dt

/-J,,
-

57-J,
i
i

\vh.-i-,.

Baler's constani
.

(see

Appendix. Art.

178).

tin- priii.
i'

-i

siiil.ol

"

"
li

pal value of tlu- integral is to


:

be taken (see Art


of
ti.i

tin-

meaning

'logy

i.

-nt

on

unt!ii'_'

//

',

;/

<

',

.in.l

then

li(y)=|

du/logu.
except
/
,

utegrals are convergent


\vhicli

of

we

inus;

326

NON-CONVERGENT AND ASYMPTOTIC

SERIES. [CH.

XI.

is

this expansion, although convergent for all values of x, unsuitable for calculation when \x\ is large, just as the exponential series is not convenient for calculating high powers

But

of

e.

To meet
x
is c x e -t

this difficulty

we

If

positive,

we

write

proceed as follows: x(l-\-u), and then we have

dt = e-

Jx

l+U

du

where
|

Rn |< n

x~ (n+l)

This result can also be found by integration by parts. When x is large, the terms of this series at first decrease

very rapidly. Thus, up to a certain degree of accuracy, this series is very convenient for numerical work when x is large; but we cannot get beyond a certain approximation, because the terms finally increase beyond all limits.
For example, with # = 10, the estimated limits for jR9 R 10 are equal and are less than any other remainder. And the ratio of their common value
,

term in the accuracy from the first


to the first

series is series

about

2500.

To get
terms.
:

this degree of

we should need 35
term
is less

# = 20, the

ratio of

RIQ to

the

first

than

K> G

Again, with but 80 terms of

the ascending series do not suffice to obtain this degree of approximation.

When
then

is

negative,

we

x^
/-

we

write
t

and

= x(l

u) = (u

1);

find

J
Q
where

1-^ du
c

')/

n p~ u

n4.<u,+v?+...+u*- )e-**dM+P\

Jo

-du

~\
\,

denotes the principal value of the integral (Appendix,

Art. 164).

Thus

where

R* = P

du

132|

AST MI-HU

i<

i.y

an elaborate
to

li--us-i>n.
;ise

which

is

too

l.-n-thy to be

also

we
1

approximation
aii-l

-y

taking

.|iial

tin-

get the best part of


2'7r/)*.
('

tliat

tin-

valur
inteie-4
it.

tli.-n

of

tin-

OPd
j;

:i't

\\ithoiit

Dotetha!
se

to Lacroix

two expansions were


tin-

utilised

by

t. liiul

Kul.-r's

O
nnl in
"

AiP'tlHT

apji!:

Miinniat imi

"

of

taking thr value


If

<.f

('

a-

km-un
r.juat*tli.-

ire

\\rite

#=1 and

>eii.-s

MI,

pp.

we have

-C-M
i:

-..)-,

,n.....
tli.-

Lacroix

giv.-s

\a!u.
tin-

o-;j)65996 as

tlie

value of

round

brackets, which yields


witli tin(1.6.

agree in.i;
ill

"sum" U-2!+8!-4! + ...-'4 n->;ik fund fn-in Killer's coiitiii'


'-

'.

98.

p.

.'1

.u'ivi-s

aiii.tlp-i-

cal'-ulatinii

'f

tliis

><! ..-..illati.i-y

the

method

.if

ajipi "\iinate
'

quadrature
I

to evaluate tip-

integral

i/y

rf^d
\vlii
'

...),

r.

l.ut

\vitli"
.

and he
integral

also suggests the applica

appi-.'ximate

<|ii:,

tlie

dv
1

-log.

whirh

is

f..und

'

ly writing

no nun.

'/>.

.-il-r tin-

two

/-I
I

rr

f* 8"3
)

x
witli in tin-

x
l

\vhich anin
tin-

iii.-t

lh-ory of
\\;r-

liysi-al

Optii-s,

and also

tlirory of

l.M-p-wat-r

*A*x
illy

ISM;,

p.

-Jul.

the

i.

-.1

have oc

propel ties of the

Hit
s

with

this prol.l.

paper).

328

NON-CONVERGENT AND ASYMPTOTIC


have
if t

SERIES.

[CH.

XL

We

U+iV= f
= x(l + u) we
have

Jx \/t

^j-dt,

so that

thus,

du =

+
by

du
parts.

by applying the process

of integration

Continuing

we have
e

~~'
j

l.S.i
'

1.3.5
'

1.3.5.

ixu 7 f" e

du

and the process can be continued as far as we please. A moment's consideration shews that the remainder integral at any stage is less in absolute value than the last term of the series and thus for any value of x we can determine the stage at which it is best to stop in numerical work.
;

We

are thus led to the asymptotic equation (see Art. 133)

*>-^(X-Y),
The
series for

say.

iY

can be

"summed" by
00

observing that

Thus

1.3. 5. ..(271-1) L=
applying Art. 118,
in Borel's

V
we
way,
its

j7T

e~

v n -*dv,

Jo

and
be

so,

see that if the series

iY

can

summed

sum

is

X - i Y= -%my,
i

"f
v

~
_ v~

given by
(see Art. 136).

y_ "

e~

dv

00

e-"v

dv

and

U=

.-(-Xsinx+Ycosx),

V=-j- (^Tcosa;-h F since).

132]
01
,..,-,.

>iii:u\
these expressions the
\

qua] to
(lill'.-lrntl.l!

tl

.j>lete

Th
Benoe we

Id
i'm.1

Lav
\

\
<-.iuliti..n

/}')!= _^1.

tli-

r)H
</}'
)-

-JT--1.
"<1

..

It

1-

M
for
.\\
)'

It

th.-sr
tl.-

l.v

tin-

l.i-

integralB

;m.l

that
:.!

---IHI

to

1,0

i.--j.r.-tiv.-iy
'

as

that

f\

I'

an<l

.\',

)'

man:

-t.-il

l.\

'I'lif

an- actually int


i.sson (see

to
I

be

du-

t>
i.lv

'aiK-liv,

and

tin-

asympt,,;

It

is

}M-rh;i}

\\ni-tli

wliil.A'.

to

niakr
aii-i

tinI'

additi
-t

mark
naturally

that

tluI

relations

between

I'

hy th- qse 6f
[ing's

tin-

asymptotic
1*0) that

*
(Appendix, Art

in

!><

j.rovt-d

wliere

\fs(.r
i

= (.'+
.>)-

.1

1<>^.

./-

Now

we have
>'/

aivtaiK

X ) = (v

-...

+ (-1.
whew
(

,/_,(''>

Appendix, Art.

ITi'-.

Bx.

:>>'.

We

li;i

wh.-iv

//

ix

numerically

less

than the

tirst

term omitted

the

961

330

NON-CONVERGENT AND ASYMPTOTIC

SERIES.

[CH. XI.

If we take the quotient of two consecutive terms and remark that (compare Art. 130)

where Q

is

a factor slightly less than


is

1,

we

see that the least

value for the remainder


integral part of TTX; of accuracy which is
(4)

given by taking n equal to the but the first two terms give a degree ample for ordinary calculations.*
to those

of (1)

The reader may discuss, by methods analogous and (2) above, the following integrals:
7
/
I

OJ.AJ.

i/

'

~r

the

first of

which

is

related

to

the error-function, while the

second and third are the cosine- and sine-integrals.


133. Poincare's

theory of asymptotic series. All the investigations of Arts. 130-132 resemble one another
Starting from some function J(x),

to the following extent:

we

develop

it

formally

in a series
a,
.

ct f

is not convergent, but yet the sum of the first terms {tt+1) gives an approximation to J(x) which differs from n+l where than n /x n depends only on n and J(x) by less not on x.

This series

Thus,

if

Sn

denotes the

sum

of the first (n

+ 1)

terms,

we have

In
i

In

that the series is asymptotic to function', and the relation may be denoted by the symbol
all

such cases,

we say

Such
*An

series

were called aemiconvergent by the older writers.

when x

elementary treatment of this approximation will be found (for tin ra.Mis an integer) in a paper by the author (MeHsenyer of Moths., vol. .'>(>,

J90C, p. 81).

132,133]
It

II

is

to be noticed.
t.,

ho\
.'xampl.
lin.

more

thai
-0,

tinIt

vuinr

s.-rir.s

\\ ill

ivpi

fi.ll..\\s

from
tlf
of

tin-

definition

tl

..

lit.

JC6

pr-'dlict

Illptotic
appli.-s.
\V.-

.Illill^

that

the

nil.-

Art.

ril-

still

then

liml.
'

if

)^" +
(l

"

XX
1

-i

^/'o +
!

V
X

.'

tin-

formal product
',

ll(.')^'',,+

-f

J-f

where

= ",/' + "/'

,-h--

+"

r/jT,
\\liii-li

t-nl ttip- pi

by definition 1
;!
ins
in
"
1

<-Min,-i(lr->
fi
'/'

\\ it li

mlurt

>'

7'.,

nj
MI

t<>
1

ami
'

iip-lu<lin;_'
t..
1

..

thui

.''""

\\lu-iv /'
'

is

polynomial
-

in
/

...

whose
/1

lii-li--

mi

if

1).

'

us

[ /( /v -, "][
.'

1-2.+'.
A
i

p-0,

<r->0,

ami accordin.
lini

.*"[./(,

A
JT

'=0.
**>

OB

Thus
l.V
II'

tin-

].rMiuct

/{

\.

A'
In

d
II"

asyiujit.

>t

iid

in
i

p-v.

a-ymptotir
conoid.-!tin-

,y

particular W6 CU ohtain any the ordinary ru


,,t'

Lft

us

imw

p-.ssiliility

sul'stitutin^
first

an

In

tlir

p
-',,-f ./

may
./

c\id-ntly writ.- /{
th<

^ul^titut<-

in

-f'V 7

'

*Of

332

NON-CONVERGENT AND ASYMPTOTIC


,

SERIES.

[CH.

XL

the

and rearrange in powers of J 1 provided that |a is less than radius of convergence (Art. 84); because lim /1 = 0, and we can therefore take x large enough to satisfy the restriction that
|

tt
l

ol~H</il is to be less

than the radius of convergence.

This having been done, we the asymptotic series

may

consider the substitution of

for

/,_

in the series

*Vi) = t' + C.J, Let us make a formal substitution, as if were convergent; then we obtain some new

the series for


series

where

*-*4+^ +
Sn and 2 n
2,/

:::.

Let us denote by

the

sum

of the

terms up to \jxn in J^ and

respectively.

Now,
'

if

= C + C.Sn + C,Sn* +
in
n
,

. . .

+ CJSS,
'

2n and 2 n
polynomial
(1)

agree up to terms
l/.r,

in

and consequently 2 W l/# n2 n+1 to thus (l/^) ranging from terms in (l/#)
;

is

Next,

if

rn = CQ
Sn
represents J^ asymptotically,

we

have, since

\imxn (J1 r

Snr )=0, and

therefore
(2)

Finally,
thus, since ^("/i)
is

^- Tn = Cn+
convergent,

where Jf

is

a constant.
find

Hence, we
(3)

because

By combining

(1), (2)

and

(3),

we

see

now

that

lim

;\

3BR1

Thus
pou
t /,,

the
!;.

sequent

kan
Further, a

tl

tin-

cor
in

in

wo

places <>niy.

tir-t

in

"i-deitin-

rearrange

in
j

i'

./,.

;iti(l

secondly

bo

estal

iii.-,,uality

/'-'/'

<
i

M-l
-1
il'

his

in.-ijualitv .'./ bo /'(/,):

th--

i^

a
in

and thru we must


'

>UJJM,>.-

tl.

ord<
ill.-

-/
l

Mini

mtir.-ly M\.il
"//

r-^tridinn
j

that

/(./ is

whose
-

first

term

zero
-i'/f

mo
a,,
i

..fitufii-

nt.

An

application

.f

th<-

t'nnm-r n^nlt
i->

is

t.

rstal)li>li

tli-

(assuming that OQ
"'

\\<>(

x-rn).

Fr

\\-

can

\\

-".(14wht-re
A'-x,

A').

Th.n
aini
liy
:

J-i

,,

-i
o
(

i_A'4_Aan asvni])t"t
it'

thus r.instnirt
sain*-

ic

Beriefl

1
i

ij"
t

.-xacily tin-

rule

as

th-

w.-iv '}\-

Thus, applying the rule for multiplication,


,,/,,

we

see

divide

Finally,
;

l.-t

which

'i

= 0).

j&l
\\,I

J-8, <f
././..] J

..".

if

x>
if
,

>.:.

<
(

./ '/./

is

-,{

asympintically

l.y

.r

334

NON-CONVERGENT AND ASYMPTOTIC


an asymptotic

SERIES. [CH. XL

But, on the other hand,


be

series

cannot safely

differentiated without additional investigation, for the existence of an asymptotic series for J(x) does not imply the existence of one for J'(x).

Thus

e~*sin(e*) has an asymptotic series

0+2++.x a?
But
its differential coefficient is
;

sin (e?)

+ cos (e*),

which

oscillates as x-

tends to GO
expansion.

and consequently the

differential coefficient has

no asymptotic

On

expansion,

the other hand, if we know that J'(x) has an asymptotic it must be the series obtained by the ordinary rule

for term-by-term differentiation.


This follows by applying the theorem on integration to J'(x)
quite as simple, and perhaps more instructive. theorem that if <f>(x) has a definite finite limit as x tends either oscillates or tends to zero as a limit*
;

proof

is

We

make
to QC
,

but a direct use of the


then
<'(- r )

If J(x) <x

+ ajx + a2 Ar2 4-

. . . ,

we have

Thus the
if

differential coefficient

^+
it

[t

/'(^)-^'(

has a definite limit, must tend to zero. But xn [J(x)-Sn (x)] does tend to zero, so that \imxn+l [J'(x)-Sn'($y], if it exists, is zero. That is, if /'(#) nas an asymptotic series, it is

It is

instructive to contrast the rules for transforming

and

combining asymptotic series with those previously established for convergent series. Thus any two asymptotic series can be whereas the product of two convergent multiplied together series need not give a convergent series (see Arts. 34, 35).
:

any asymptotic series may be integrated term-by-tern i, not although every convergent series can be integrated (Art. 45).
Similarly
*
111

fact

if <f>(x)

tends to a definite limit

we can
if

find

r
Q
.

so that

"
Thus, since

= 0'(a
I

>X where x > * >


x
'

a-

we

find
I

*'() </(*-*())

but the last cannot approach any definite limit other than zero may not take a// values greater inequality does not exclude oxcillatioH, since than .r as x tends to 00,
<f>'(x)
;
:

So

i.

133)

Til
I!M.

i:

l:Y

<r kSYMPTl

'

On

other hand
unless

ditl'eivnt

we
<l-ri\

in-

depend'-nt
'nptotir
aerii

ivasonin
.

-mlin^
li,

in

dealing with a COir


-in

in

apply
i.-ii-

tli-

<-on\

to

the

iliH'eivni
li;i^

that

th--

l'-i

an

contrasts,
In
wi-

bo*
tin-

l'-l

as sm-pi
to

convei

parameter with
d-tinition
pies,

ivsp.-et

\\

ilitl'.-n-n!

and

in

no on

way
tin-

enters
Ml in

in'" thasyni]'-

of

tin-

an
'

the

l.-jH

-niU

i>arani.

Tin- contract

may

l>c

ilii.

in

an eveo

iiioi-i-

l'tinilam-ntal

way.
1

"-llicii-ir

may drlin.- a aM in])!<>!ic Berii


It

jM-rt'rrt ly

^
a

asymptotic coinpl.-t.'(l whole


to .-xt.-ml
l>y

lnl-i
in
tin-

>au

ies.
i./

sniii.-tiiiifs
i-

convrni.-nt

our
tli.

l-linitin and -ay

that

represented asymptotically

when

J
i

'I*

is

represented by

"+ X
.

''

!+..., where

<!.

l'

two suitahly chosen functions


Thus, for example, asymptotic formula
f

we can deduce from

Stirlii,_

the

n~.

i/V^...),
'

r1C
-2

/!
-2

12'

r =-B
,

11

-)

Hitherto

./

i,

sn]]ios,-d

to

ten<l

to

-s.

through
lin-rtion.
l;

\aluex
coiiij.lex
i

l.ut

the
/:

theory
in

remains
..tln-r

unaltei
Mt

and tends

to

any

definite

,-

dued analyi

336
In

NON-CONVERGENT AND ASYMPTOTIC


fact, if

SERIES. [CH.

XI.

we can determine r_ a
"

constants M, R, such that


2'

_i <M_ x
^l*!

HI C|> !->/?

it

that J(x)

follows from elementary theorems in the theory of functions is a regular function of l/x, and consequently the
is

asymptotic series

convergent.

For different ranges of variation of the argument of x, we may have different asymptotic representations of the same function which between them give complete information as to its nature.

good

illustration of this

phenomenon

is

afforded

by the

P>essel

functions which have been discussed at length by Stokes.*


134.

Applications of Poincare's theory.


to

important application of Poincare's theory is solution of differential equations.! The method may be in the up following steps
:

An

the

summed

1st.

formal solution

is

obtained by means of a non-consolution

vergent
2nd.
exists

series.

It is shewn,

by independent reasoning, that a

capable of asymptotic representation. Thus we has been done in Arts. 131, 132 above, deduce a as may either, definite ^.epral from the series first calculated; or we may

which

is

find a solution as a definite integral directly,


it

and then identify

with the
3rd.

series.
is

The region

determined in

which the asymptotic

representation is valid. Poincare has in fact proved { that every linear differential equation which has polynomial coefficients may be solved by

asymptotic series; but his work is restricted to the case in which the independent variable tends to oo along n specified
*Camb.
p.

Phil.

41-J;

Mti/h.

Tram., vol. 9, 1850, vol. 10, and Phy*. Papers, vol. 2, p.

also
r.

Mathe.matica, vol. 26, 1902, p. :>!>:}, that in the asymptotic series examined by him, the change in representation occurs at a value of the argument, which gives the same sign to all the

Ada

1857, p. 105, and vol. 11, See 350, vol. 4, pp. 77. -Js'i. and Papers, vol. 5, p. 283. Stokes

marks

terms of the divergent series. Some interesting remarks on the sense in which an asymptotic series gives a Million of a differential equation, have been made by Stokes (Papers, vol. 2,
I

]..

:W7).

Acta Mathematica,

t.

8,

1S86, p. 808,

133, 134, 135|

AlTLh LTK
tli.i

M
.).-

PTl

-I

( i

BEE!

and
hy
tO
l.y

not

determined.
-perial
,

This gap has


cases;* and
different
in

tilled

Horn
a

iii

iiiiin

we
ial

ma;.

detailed

diM-u^i,
tin-

,\

equation

.lacnl.Mhnl in

paper
intin-

|ii.t'-l

l.elnw

applieat ion- "f

Poincare?fl

ami Manly
given
with
very
l.y
i'u',1

iii

.nst

met

theory have been asy mptot ie re]


Hai
oi
this
\\'.-

made

l>y

ly power-Beriee referencea i^ ui\--ii


i^>

nmnary
l.y

{he

inei

a|..j.tMl

I'.ani.^

IM-VMIK!
<>!'

tin-

liinit^
in!

book, a
give,
fa

lr|M-nls
in

MM th- tlnM.ry
thl'nll\\ in-\\^-<\
l.c

c< >nt< .111-

-rat inn.
iiH-th'.,l

art id.<l-aliirj;

^imjil.-

due to Stol

which can

in

with c-rtain

typ--

135. Stokes'

asymptotic expression for the series

v !'(/'+', + !). ..rfri + -'r ( , l+ 6 1 -fi)...r(7i-f


X
\Vi-itf
is
>

ar+D^,
consider
th-

ival

and

>
"
_p

I--=/UL,

-/'

= \, and

term

wln-rr
\\'.-

is

luro',

and

is

not of lii^hcr order than v //.

tind.

(see Art. ls<>. Appendix), is ronvrni.-nt to It suppose that X


/

is

of the
IMis

form

/"

whnv

is

an

nioiv

integer (a elaborate methods);


-

n-Mrirtion which can

and then

A'

usin^ the greatest term


l.y

renmvrd

because logi

u log

'.

an<
lM iir

log Z
or

= M /-Aul.-(^/)-\ r+;(
-

Z
This
ivefl

/
:.

/-J. u/

A
,

c\p<

\uf>*/t).

the

See a series

ot
i

Is,

and
I.

U
:.
-.

'!! in

HIM

\iisohnitt \'l
;

see also

//,

vol.

i-p.

ins, in;.
fi

c<ut>. rhit.

ua

338

NON-CONVERGENT AND ASYMPTOTIC


/t
.

SERIES. [CH.

XI.

where q = e~^

Making use

of Art. 51, Ex. 3,

we

see that (since


is

q approaches the limit 1) the series in brackets approximately by 7r*(l q)~*, or by (2-jr//tiJ*.

represented

Thus the asymptotic expression


e'H-*
T
.

is

rr,

where

= x-.
same way, that
if

M*(arO**" Hardy* has proved, somewhat

in the

f(x)

is

represented asymptotically by
+...

Aex/(27rx)\ where
and

= e~*.
SERIES.

SUMMATION OF ASYMPTOTIC
136.

Extension of Borel's definition of summable series. In the foregoing work (Arts. 130-135) we have shewn how to obtain asymptotic expansions of certain given functions, and we have established rules of calculation for these expansions.

We

are

now

led to consider the converse


;

a given asymptotic series


Borel's integral.
is

not sufficiently series, such as (Arts. 98, 132)

problem of summing method is to apply But on trial, it appears that Borel's method powerful to sum even the simplest asymptotic
arid the natural

If

we apply

Borel's

method

of

summation

to this series

we

are led to use the associated function

which converges only if t<x. Thus the results of Arts. 99113 no longer apply, but we shall now proceed to modify the
definition so as to obtain corresponding results for such series.

We
and

have

u(t, x)

= x/(x + 1),
which
is

if

<

x,
t

this defines a function


.

regular, as

ranges from

to oo

Thus we can agree

to take the integral

*Proc. Lwul. Mntli. sV.

(-2),

vol.

i>,

191)4, p.

339.

135, 136]

1M \Tl'\ OF
Hi.
i

U3YMPTO]

i<

339

as

d-tiniii^

the series above.

writ-

in

this

intfi-al,

But we can also and so obtain a second definition

Tin- reatlrr uiii


\vlii.-h

i,

integral agrees with Art. 08.


e.\

I.

\\a-

'_

i\'-n

in

The
l>ut
it'

t\vc integrals are


/

vah-nt so long an

is

real

and
./

p<>

c,mple\ th,- f,,nu.-r i- convergent e\,-,-pt \\ln-n negative. whereas the latter converges only if tin- ival part <>f
is
It

and
i->
j

i>

nut
in

ditli.'iilt

hat
(

tin-

I'MMIM-I-

in'

fiinrtimi

tin-

part

tin-

plain-

for

\vhi<-h
/.

th-

an analytic ir-'in the

negative real
I

ny

p..-iti\.lint-s

numl..-!;

Thi- p;.

iy

two

paiall-l

juiin-il

as in

tl

lilish thi>

that
"te

<4<.

and
(

that

th.-

int.'-ial
!>

that

tlu-

test

..f

Ait.

174

Aj)pcti(li.\)

can

In likr niannri

tin-

function in
/
;

\ part of the plain' for which tin- r-al part this n-^iitii is l.oumlc.l l.y a line parallel to the iina^in.i!

tin-

s.-mml integral can be proved to define an analytic J is greater than

nirtlio,!

in this

example suggests a general process

for Minmiin.o;

an

that

is

an asymptotic 96riee Mich that the associated


'.-

seri-'-

+ 'V + ", L + "::i: +);

r-

OOB
i

tl.-tiiuMl

certain jiositivr valm-s by the series i'r these

oi'

'-.and tliat thr I'unction

\aln-s of

is

i-ontinuous

i'roni

=
\vt-

to
di'tiiu'

tlu>

sum

..t'

tin-

srrirs

l>y

tin-

integral

'.
|

340

NON-CONVERGENT AND ASYMPTOTIC

SEEIES. [CH.

XI.

ever,

To make any practical use of this definition we must, howassume further that a positive number I can be found
v-><n

such that

lime- lv fn (v) = 0,
differentiation.
is

where
It
is

is

any index of

evident that then the integral defines a function which within the region indicated in the figure.

analytic

137.

Rules for calculation.

It is easy to see that the series of the last article represents the integral asymptotically, x being real and positive.

For,

if

we

integrate

by parts we have

= aU +-l
/

I .

e-'f(-\dt, \ /
/

because /(0) = a

We

continue this process, and

we

obtain

Now, by
so that

hypothesis,

we can
w+1

find the positive constants

|/
'

and therefore
Consequently,

"

(^)l<-^^
/

'

which establishes the asymptotic property. It will be remembered (see Art. 133) that the differentiation of an asymptotic series requires special consideration; but
this

special

type of asymptotic series

may

be

differentiated

any number

of times.

For

the differentiation under the


(if

integral
(x

sign being permissible


- l}t
,

x>l), because
t

\te-

f'(t/x)\<Atedt
is

and the

integral

te~

~ (x l)t

convergent.

Jo

136, 137]

SUM.M VTION
tli.-

n|

AS1 KPIU1
!'

I'

Now, within

int.-r\al

OQHV6fgenoe

"i"

/<

>;)

we have

so that

/'(g) is

ptotically

l>y

the series

-' x\x

a;

or

...), /

which
formal

is

tin-

Beriefl

d.-ri\rd

i'mm

tinlie

orij

for /(#) liy


pat
.--I

diir.-iviitiation.

as

could

ant

ici

from

A
of times.

rly tin-

opn-atioM may be ivp-at-d any number


.

:'-al-i- \vl.

.intril

with

tli.-

oliMiients <f funrt ion-theory will

lia\r

IK-

litliriilty in

pruvin^ that th-se n-snlts arc valid within the part of


l.y

|ilarn-

indii-atfil

th-

i:i;.

L--t
i

IH

c.

.iixidnv

M.-xt

th.'

algebraic

operations;
ie Beriefl

it

follows

I'-Miicaiv

and multiplied as
the product can considered above.

theory that two asymptnt may be added if Uut w- must aUo pn>\v that convergent,
!.-

ivpi-.-^.-nt-'d

l>y

an

integral

of the

type

\Ve have proved that

is

tin-

asympt'itie Beriee

-"nvspondin^ to

tin-

product,

it'

Now

if

both

AO-Xa

oonverge und.-r the con. lit ion || such that |a w |r"<J/./!


f

= r,
1

we can
/

find a constant
/

6n |r"
(

< M.

Thus
|

Cn r-|<3f [/.:-h(//-i):

:J

---j.:.-j:-h...-f-'

<
Art.

.!/-(

and oonsequently 1

will

oonveige

it'

<

/.

Next, as in

106 above, we have

'"

..i,

ai^K'

342

NON-CONVERGENT AND ASYMPTOTIC


differentiation

SEEIES. [CH.

XI.

and consequently by

we

find

Thus

Borel's associated function for the product-series is


V

F(v) = atf(v)+

\ Jo

f(v-t)g(t)dt

and
But,
I

\F(v)\\a

\.\g(v)\+\"\f(v-t)\.\g(t)\dt. Jo
find the positive constants

by hypothesis, we can

and

such that

\f(t)\<A#,
Hence

\g(t)\<A#.
lv

F(v}\< a \Ae

+A

ve lv

and consequently we find


2lv

v>ao

lim[ethat

F(v)]

= Q.

Similarly,

we can prove
v-><

lim [e-* vF*(v)]

= 0.

Thus F(v)

the necessary conditions. these we have the following rule results, By combining a certain number contains polynomial expression
satisfies all

If
of

asymptotic series* (of the present type) and their derivates, the value of the polynomial is expressible asymptotically by a
series (of the

by applying the ordinary rules of calculation, as if the series were all convergent. And the result cannot be identically zero, unless the terms of the resulting series are zero.
The reader who
difficulty in
is familiar with elementary function-theory will have no extending these results to the complex variable, in a suitably

same

type), obtained

restricted area.

*0f course some


in
I/a;.

of the functions

may have

ordinary convergent expansions

137,138]
138.

SUMMATION OF AflYMFIOTH
series.

Another integral for an asymptotic


numlM-r of
ini_
i:

Tin-re are a

\ampl.-s
nn-thod,
in

come
IT

which do not but can be


tin- <l-tinition.

asily ly niaki:

we suppose

Art.

MS
I'm-

it

i-

a function Mich as is specified in 0(.r) to be that th.seen results are still pn-vious rea.lily
.

true

tin-

int.

/(a
provid.-il that
06
/>

|
haa
i

f.
ni'

t!

DOH-ZerO ra<liuMItOCe

anl that crrtain n-M ri-t of increase of |/"(v)| with


particular,
ii'

-ati-li.-<l

as to

In

\\c

can tind a

i'un-ti<.n

\!s(g)

such that

(-D
we
shall

Jo
|

have

tin-

-jnation

For

this

i-

t-a^ily

take this integral as

}>roved to be true when |v|</>, ami we can tin- d.-tinition of f(r) for larger values
.

Then,
an,l

!/()!<

-)<f>' Jo
tin-

That

intt-^ral

i-

in
l.y

fact

r-])i-i-^-uted

asymptotically by

tin-

series follows at
}

one.-

Ml.s.-rvin^ that

Me) Jo

fi

r
!

~
-

-...+(-1 r'+...+2

'"

+,,.

win
1

<
|.

"'

+...+

344

NON-CONVEEGENT AND ASYMPTOTIC


J'(x)=

SERIES. [CH.

XI.

Again*

and, as previously, this can be shewn to give the asymptotic series obtained by the ordinary rule for differentiation.

special type is given

by taking

and then Thus our

cn

= T(n-i-\).

definition

becomes

=
It
is

easy to

repeat

the

previous

argument (with small

changes) to establish the corresponding theorems for operating with this integral.
Art. 132,

As an example of thewe could infer

last

method, we may point out that in Ex. 2, the differential equations for X, by simply

operating on the asymptotic series. Another example is given by the series

which leads
the integral

to f(v)

= I/(l+v),

so that the series represents asymptotically

139. Differential equations.

We

shall conclude

by giving a few examples


;

of the

way

in

which asymptotic
Arts.
1.

series present themselves in the solution of and we shall illustrate the methods of differential equations

136-138 by summing these series. Let us try to solve the differential equationf

dy _a
dx~~ x
*

Differentiation under the integral sign

is justified,

because

we suppose

and

t<f>(t)dt

Jo

converges.

Thus the test of Art. 172 (3) applies. fThis is the simplest case of a general type of equations examined by Borel (Annalea de VEcole Normale Suptrienre (3), t. 16, 1899, p. 95).

138, 139]

IHKl
!'

1.1:1

\i.

BQ1

VI i"N'S.

l>y

meanfl

.-MI

asyinp'

On

substitution.

we

find

'I'll

is

g\:

-4

= 0, A = l

-I

..=

i*

"

-2A,_
' :

1.2a ~~

34 8 _1.2.3.a
'

I,

I,

/,'

Thus we

lino!

the formal solution


1

'/=-/' ox
and by
Art.
l.'lti

1-,

-.

ox
the equivalent

)'+...]. J
integral

bhif

-"
and
tinit
i

-,
I

7-T-|
.

Lriven

diivetly that this inte-cal e.jiiation, as of course must be the


bo vei'ify

Of

Art.

137,
tin-

n>i<l-r

iiK'tlitied

Bessel's

matiin

-a
Writ.- 'n
.

ami then

\vi-

tind

If

ire

sul.st.tui,7;

=i

ire
...

+ ... +
Hence we
and
so find

...)=o.

--(fife*

s45

1*.3.5 2

3> T

346
Thus
>/

NON-CONVEEGENT AND ASYMPTOTIC


is

SERIES.

[CH.

represented by the series


1 1 1
2

.3 2

__

.3 2

.5'2

2.4.6
That
is,

*=(- D"
*

---l
2.4.6...2w

l}

1
2"'

and

so

if

we put

cn

r(f)

we have

v^ vn =
A~i

This leads to the integral (Art. 138)

TT

Hence

(/

s>&
-y-

,,00
/

y=
we

ptfjf C Ct'O
t
,

which can be proved directly to be a solution


Similarly,

of

the given equation.

can obtain the solution

for the equation

d2y
3.

J + 5s
In like manner the differential equation

du

*s+<. + /~*>i-<*-*
is satisfied formally

by the asymptotic power-series

y ~x\}~

.IT,

a(l-^) l.x

a(q+l)(l-^)(2-^)
1.2.^2

which leads to the integral

This equation has been very fully discussed by Annalen, Bd. 56, 1903, p. 129), to whose paper we
as to the various solutions.

W. Jacobsthal may refer for

(Math.
details

have now completed a tolerably full account of the theory of non-convergent series, so far as it has been yet developed on the "arithmetic" side. Its applications to the Theory of Functions lie beyond the scope of this book, and we have made no attempt to give more than a few theorems whose proofs can be readily supplied by any reader interested
in such developments.

We

XI.
I

LMPLE&
Integration of
1.

Summable
B

Series.

Wi- have seen

bhatth<

'he series

.-

series

to be

*a).

This suggests*
\vhi.-h
is

c 87U'-' /
f'
b
.>
(
.

OS 4?

COS a

(0<a<7r),
tity

-a>ily

vi-i ili'-d

.os nx - cos na

_ ___
MM(i

,-^ n(l +

{>>

ai

cos ./--rns a

2.

.)u-t

a-

in

lli-

and so

integratiiii:

\v-

tind

log

the constant being foond


'

bj ufling the value

of the

sej

-.65).

log(f"fi.rin
..f

An'th-r

tht-

first

intrirral

is

-Trlogma -^)}.
e

that

if

,-

i.

0..+HX
3.

Sinn-

\\r

':

it

i>

sii^^f-t-tl

* that

if

//

is

a multii>

HI

=ir.

This

i-i-siilt

is

easily

vt-rit'u-tl

indr|MMid-Mtly.
intririiit
.e

*Thr n-M
\,

x=

nor

>vi-r

ti

348
4.

NON-CONVERGENT AND ASYMFTOTIC


From
Art. 103(4)

SERIES.

[CH.

we

see that

Thus
and
Hence, integrating with respect to
0,

we

find

2J

sin
is

^
/?(#), p.

The

series

2( o

l) /(x

+ n)

sometimes denoted by

and can be

expressed by means of the ^-function (see Exs. 42-44,

475), since

=\

Thus, ^y x is rational, the series can be summed in finite terms the case has occurred in Arts. 65, 90, so take now x = % as a further example.
;

We get

where

< 6<

TT.

If 0/7T is rational the series

may be

expressed by means of "^-functions

and

so the integrals are then expressible in the same form. If we allow to tend to in the sine-series, we get

[For

by Weierstrass's

Jlf-test,

and
(.

^ I ?^(^) =

_ ^) COS

^_J

S in

^6 log(4

si

Cesaro's Mean-value Process.


5.

A specially
Feje>,

by

who

applied

interesting example of Ceskro's process was given recently it to the Fourier-series for a function f(x) which
174,

does not satisfy Dirichlet's conditions (Art. series need not converge.

App.

III.), so

that the

In

fact, if

a n = - { f(0)<x*n$d6.
TTjO

6,,

= - (*f(0)fa.nO dO.
'

TTJO

and

and we write

xi.
I

JfPLEfi
find
tli.it

\\-

"

and

that

tin-

aritlinn-t

i-

im-an

"t"

.<,

x,

...

wh
ial
int..

tWO,
baa a

W6
iln-

'

I'm

' 1~ :>

(A|.|.

fnin-ti..!!

The extension

to

caws
IK*

\\h-

Unit*l.ut
i-

mind"

n.ntinuitiea present.)

fn-li
!

(litli.-ulty
inti'^r.-ilili-

tin-

pi.-.f

und*-i-

tli-

>ingle

ni!i>t

L.-vnud our r;m_


us.-d
in
th.-

6.

L.-t
it

u> writ.is
'

,r,,

f.,r

tin'

arithiiM-ti'.

iiM-an

la-t
r
fl

thru
it>

ra.-ily

Men

that

"lit iniK-us

the
in

.nvrr--

to

limit

to

L':

r
(l

-^,1-''

'

Since
\\f

o- w _i

=a

1
r
1

^l

tind

that (|-ayin-

at ti-nt i'ii

t. tin- il-tiniti-

''

l/l'-)J-^--

^+
rfl

A
(

v
(

^+/v)(i-^)<^ /;V
|

/-

i-

any

nuinl.ri

K->s

than

,<

and

BO,

taking

tin-

limit a> N

tends

to oo, \vc find

I
r=l

=}
x

l"'[

becanae

li.M./, l= o.

Thti-

:it

(Alt.
givi

7).

and

wr may

a].jly

1*^1

Tannri

y's

thf-i'-ni

(Ait.

1'.')

to

./

\vhirh

1
[Thi.->

>;-'

+ V).
al>. a jnji-r

ivsnlt

is
i

due

IM

dt-

la

Vall'--

I'mis-in

MT

lv ilnrwit/

{Mutt7.

Wr

~MiVr th-

'

Lft,

Ch,
1

N.\

T6

MT
'

that

- ,

and

I
n

}<

\.

;,
.

1<.

.'!

Lebesgue, Strie*

I'.Hlti.

j,j,.

(_>-

350

NON-CONVERGENT AND ASYMPTOTIC


series

SEEIES.

[CH.

Hence the

a n sin nx -f bn ( 1 - cos nx) }


therefore equal to the

is

convergent

and

its

sum
But

is

sum found by taking

the arithmetic mean.

this is equal to
<r n

lim

n-^Jo
because
<rn

dx=

{ f(x)dx,
-

[This result
8.

converges uniformly to the value f(x). is also due to de la Vallee Poussin.]

Consider the application of Frobenius's theorem to the series

where and x

t is

a complex

number

of absolute value

1,

but

is

not equal to

1,

is

real.

It is easily
sl

proved that
S3

S2

_l-t' ~" ~

i~ So~
if

'

S8
2

_l+t

3
*>

i>

and generally

sn

= {l-(-t)}/(I+t),
is
2

(v-l)

^
),

Hence the arithmetic mean


and thus
or
x>-l

+ = lim (1 -^ + .^ -.^ +. ..) !/(! +


found to be
3

1 /(!

lim (1
X
>1

- .r cos + #4 cos 20 -.r9 cos 3# +...) = 55

lim (x sin
* >1

6-x* sin W+x

sin

30 -...) = i tan

[See also Appendix, Art. 155.]


9.

Apply Cesaro's mean-value process +


/I
(/I

to the aeries
(/I

+/*)

(/I

+/2 +/S>

+/2 +/3 +/4 ) +


zero,

where fn is positive and decreases steadily to The sum is


Kni [/i
a:

but 2fn diverges.

- (/i +/2)-^ + (/i +/2 +/3 )^ 2 --].

[HARDY.]

[We
*1

have, in fact,

=/!

*2

~/2>

*2n-l

=
of
.s-,,

and so the arithmetic nu-an

*2 ,

...,

%,

is

Apply

Stolz's

theorem (Art. 152,


is

II.),

and we

find that the limit of the

arithmetic

mean

equal to

lim^.,,.

tends to the same limit

that lim(^,, + i/w) = lim/o, (+1 = 0, -as that of


l.y

Again, from Stolz's theorem we see and so the arithmetic mean of 1} *a ..., %,+!
,

s,,

.,,

,.., *_,.

found

applying Art. 124

(<

tin-

pi <.<luct (1

The result can also be + 1 - ...)(/ -/i+/ - ...).J


2

10.

Illn

'

unple are L'iv-n


I

liv

t.ti

(I

:.,-M

r>
[in
wllieh
11.
tin-

seeond
tli

m
base
10,

to

iU"

log ($TT)
il

= -098060

to 6 d
\rt.
1 -J 1
.

ices,
]

V.-lili'--

Killer's .-alt-illation <|U..ted


II

that

O"
decreases,

and

C.

Thus the

series

2(-l)r-[(n
tli-

-n]
nu isseen to be

nvergent and, fi.>m


value nf
is

iC-filog
Tlif
tin-

->uin

can

l>y

i.l.-n

in^ that

th-

sum

t<>

rms

f-i)

+
(.

-...+('0*,;:,
Series.

Asymptotic
12.

Ksialilish tlu- asvni]titir


in./.

furnmla
L

Vn=

JM
tin-

(-l)'T, 1 **-

-;

v ,+ (/'-)i-

-- ]

I.
.

...<
4

-...

(WTT)

whi.-h It-ads very easily t>


|

nunu-ri<-al

0787,

'

-M;:ii,

<

and so on.
[C(.in[.air
13.

Art.

132.]

[(

t'se the integral of l0 ^


,

An.
")

^>

t<>

sli.-w

that

/'T

-/ 1

-" 'M<
I

/'

JT

and th-ducf thf

asvni|.t"ti.- .-\iKHisii.n

Where
14.

-:

(a)

i-

th.-

Hernnullian

fun.-ti..n

..f

An
1
<>

/* *-**

-/,^}
olitaiiu-il liy
.t.-iijtin^

the

I-IT..I-

at

any >tage being


identity
/'

less

than the
v

f.-lb

t'Tiu in the

sn
first

and iMr.im:

[Km- the

in;.
1

tin-

- 1

,~

K.T the MOOOd, integral.- l.y jur ..Cgested by the series and tan le established dirertly.]

integral*

352

NON-CONVERGENT AND ASYMPTOTIC

SERIES.

[CH.

the error being again less than the following term. [Apply the same methods again.]

By

writing

f" e~ tz dt=
J^k
latter

P
Jo

e~ t2 dt-

and integrating the

by

parts,

we

find that the first integral is equal to

16.

Generally,
r<

if

0<s<l, we

find

And

similar expressions can be given for

17.

Apply Art. 131

to the

function /(#) = #-(*+*),

A>0, and

deduce

in particular the formula

72xl03
Hence evaluate the sum

'"

l+i+i+^ 2* 31
to five decimal places.
18.

If

^
^-^.
^o

/oo2i.a2
JO
1

l-a:2

shew that and that

Hence prove that

r^dt-'
J*

1.3. ..(271-3)
2

and that the value


In
particular,

of the

remainder

is

approximately (a

- 71
infer
tluit

for

a = 4, by taking 16 terms we can

the

value of the integral

L-fd*

lies

between 1149400-6 and 1149400-8.


[STIKLTJES,

A eta Math., Bd.

9, p. 167.]

XI.)
19.
<>!'
i:Jl

tin-

asymptotic expansion

^1( r
and
d<

(
:

6!

approaches

1, l>y

writing
-\!ii.. ii.

= log(l
1

Com}

1..

p.

238.]

20.

From

tin-

<(rb)"

-1 1.2...(r-l) r(w+l)(^4-2)...(n + r)J

Thai
//-'

n+1

21.

Similarly |ii>\^ that

-H[See
Si
-ii I..".MI i.i -ii,
'

'7-100;
.

1.,

pp.

I'.L'

!:>.]

22.

I'lovr

al>..

that

icuiiiu'

numerators

I>MIILT

-'.

-I.

>
-

354
23.

NON-CONVERGENT AND ASYMPTOTIC


More
generally prove that

SERIES.

[CH,

where

t(t-

1)

...

(t-

and

24.

It

must not be assumed that

vergent for all values of x, same asymptotic representations.

and 2a n #n is conn that the two functions 2a #n ?b n x have the


if

lim (/&)=!

rt

For example, consider the two series x2 x* 3*

co

^ + __ = ^ +T j-_^i_ g
.-11111
tAS
I

A.

*^

"^

_^

*^/T + .... + _j-_^_-j


i

Miscellaneous Applications.
25.

Shew

that

if

we

.attempt to find a Fourier sine-series for


series
.

from x = Q to ^ = TT, we obtain the

2 (sin 2x + sin 4# -f sin Qx -f

. .

),

which can be proved to have the sum cot# by using either Borel's integral
or Cesaro's
26.

mean value

(Arts. 103, 126).


:

Hardy has extended the result of the last example as follows If f(x) satisfies the conditions of Fourier's expansion from #=0 to

2?r,

except that near


conditions
;

+ x=a, f(x}- i_# -7^ a 7


at
if

then f(x) can be

a) expanded as

(x

~~\2

+ '~ + ?
a

\x

^
a)

satisfies these
is

Fourier series which

summable except
In particular,

x=a.

7n

= l, we

have the simple result


00

f(x) =

+ <2^ ( a
i

cos

nx + ^n sin w#)j
t

where

= ^-P r/(x)dx, ZTT Jo


bn

a n = -P f f(a;)co&nxdx TT Jo
f(x) sin nx dx
;

= -P
7T

f JO

and

this is easily extended to the case in which /(.r) has any finite numln-r of such singular points. [HARDY, Messenger of Mat/is., vol. 33, 1903, p. 137.]
27.

Applying a method similar to that of Ex. 1.2 J_


A

2,

p. 42,

prove that

^ "T" 2wi 27n(2//<-l)


\

or

m+
(2w + l)!-(2)!.l! + (2w-l)!.2!-...=0.

while

xi.
I

BXAMPLE&
i

355

li.-it

tin-

;IM mptotK- series


'

J+p

+
,..

+ ->

'

is

represented

by

As

->|M-.-ti\rlv,

a verification, thru trim-l.y trim


-

two
(li tl'rivii t i;it

series are

ion gives
1

denoted (compare Ait. 137)

,/,-

--

a,

a.,.1

s,,

\vliich

agrees with the result deduced above fr'in

direi.-t

multij*!.

28.

Le Roy's method desril..(l

in

Art.

115 can be app!

ding's

&
taking
;>

=2

in

Le Roy's formula.
fact

L\V, lind

in

series

which converges absolutely

if

||<2ir|jr|.
.

Also (Art. 176, Ex. 2)

.-*+1)J
-

^T"
tli.-

intfivluinge
17''',

of

suinniatiiiii

and

A,

long as
i:i;,

||<:
we can agree
it

int^ratioii being justifiable Thus, in a-,vrni.-nt with tin- j.rir.


to define

l.y

cxplain.-il
tli.-

in

Art.

^()

as always

givm

l.y

la-t

Th.-n it" LIttegni with rt-sjK-ct t> ^ ai

is

p.-nui>Ml.l.- to invt-rt tin- Drd

-gra-

bare
f

*'**
Md
l>ut

*}.

we get the
this last

rain
Jo

<r"

iiiv-rsi..ii

i-

n,.t

p n-tirularly easy to justify.]

356

NON-CONVERGENT AND ASYMPTOTIC

SERIES. [CH.

XI.]

29. As examples of series which can be readily summed by means of various integrals similar to those of Art. 136, we give the following simple

cases

30.

Determine a formal solution

of the equation

in the

form
\

and express the

result as a definite integral.


is

[The
31.

integral

^/

V'(l +;)"'-<*,

if

> 0.]

Obtain a formal solution of the differential equation

in the

form

_n(n-l) *-**Vfl 4^2 L

n(n-l)(n-2)(n-8)_
4 8#*
.

1 J

and express the sum

of this series as a definite integral.

References.

Hardy, Proc. Lond. Math. Soc. (1), vol. 34, p. 55, and For a justification of the integration used in Ex. 1, see Hardy, Trans. Camb. Phil. Soc., 1904, vol. 19, p. 310. For Exs. 14-16, see Diriehlet's Bestimmte Integrate (ed. Arendt), pp. 208-215 Nielsen, Mathematische Annale?i, Bd. 59, p. 94; Hardy, Proc. Lond. Math. Soc. (2), vol. 3, p. 453. The equality of the integrals of Ex. 14 was proved
1,

For Exs.
35,

2, see

vol.

p.

81.

by Cauchy, (Euvres,

t.

1,

p.

377.

In Exs. 20-23 the series are not asymptotic but convergent ; they may be used instead of some of the asymptotic series found in Arts. 130, 132. But the law of the coefficients is much simpler in the asymptotic series and for this reason they are usually preferable.
;

AITKNMX
AUTIIMKTic
Tlll-:ni;v <1-

I.

[RRATIONAL ffUMBE

AND
140. Infinite decimals.
II'

LIMIT-.

\v-

rath.nal

apply tin- Mi-ilinary prO060a fraction a/6 to a <lrriinal. it

!'

livisi..n .-\il.-ni

t>

convert a
.-ith.-r
.

ix

thai

tin-

process must terminate >r else tin- <|Utii-ntx imM n-eiir (//-I) divi-ion^ at most; t'T in li\ilinu l.y l>, tln-n- are not more than (/>-!) ditl'.-ivnt iviiiaiinL-rs pussiM.-. nann-ly
(
1

L'

3,. ../>-!>.
nninatin.ir
i

aft.-i
-.

tin.-.-

di\

71

!:

= 7-

1) divi-

-.

frcurriiiur after s.-\cii divisions.


i))
<livi-.
,

And
If tin-

drcimal part
'

is

ran

pur-ly prrimlir and .-..ntainPm /''l"-"-l), ly im-ans of


'').

'In-

d--iiua!
f..r

tlio

formula
'"HV'M

sunn
:

Tli';
1>V ritllfl'
'2

ii;il

t".

''

1
1

and
"i-

BO

/'

\t

ll"t
-f

<li\'isil
I

)!

"I'

.").

<

it

t""ll"\vs

fr-'in
1'v

K".
-1
.").

!i-i.n

di\i-iM-

,-itlM-r

an
.if

index

/>

theorem that can be found BO


/'
(

\\li.-ii
t'

di\MlhI'.ut

liv/-: thus a I

is

tin-

f..rm
in

In'

1)

and
i

so can be expand- d
i"dic

i..-ridi.if

decimal with
t

liirurcs

the pel

the drrimal pai

i> nii\.-.l.

e,,ntainin_
:

and

/>
;

j-

n\MH'
tin-

Miu-t
i

contain eith.

because
relation

decimal

tie
1

Tic-

hrt\\
it

the
in

.itlier pi th,'I'!,,

Mi..t

l.e

discussed so simply.
USB, Ditq.

is

pi..\.'d

that

if

lt

= V&r**"t* ....

where

r,

t,

r, ...

are

prin
.

is

a factor of

It'

IKINV

we
I

inated (say
2t
i

-\-l'}\

l>y

an

intinitr <l.'cinial

will

).

.-vidt-nt bhal

fracfa
.rdin.K to tlie n.
it

is

more

o.nvi-niriit

;inl

\\

shall .i.llu-re to

'l

358

IRRATIONAL NUMBERS.
easily

[AP.

I.

But we can
exhaust

see

that the rational numbers do not

all infinite decimals.

Thus consider the decimal


1010010001000010...,
consists of unities separated by zeros, the number of zeros increasing by one at each stage. Clearly this decimal neither terminates nor recurs: and it is therefore not rational.

which

Similarly,

we may

take a decimal

11101010001010001...

formed by writing unity when the order of the decimal place is prime (1, 2, 3, 5, 7, ...), and zero when it is composite (4, 6, 8, 9, This cannot be rational, since the primes do not form a 10, ...). sequence which recurs (in rank), and their number is infinite, as appears from the Theory of Numbers.
If the

primes recurred in rank after a certain stage,

it

would be possible

to find the integers a, b} such that all the

numbers

would be prime. Now this is impossible, since a + ab is divisible by a; and therefore the primes do not recur. If the primes were finite in number, we could denote them by p lt p 2 ...,>; and then the number jt? 3
, ,

would not be
of primes

divisible
;

is infinite

this

by any prime, which is absurd. Thus the number theorem and proof are Euclid's (Bk. ix, Prop. 20).

As an example of a different type, consider the infinite decimal obtained by applying the regular arithmetic process for extracting the square-root to a non-square such as 2 this process gives a sequence of digits*
;

1-414213562373....

This decimal has the property that, of the first n digits after the point,

if

An

denotes the value

which gives
*
h.

2-^ n

< 3/10",

since

2^1 n +l/10 w

<

3.

rapid VIII.

way

of finding the decimal

is

to use the series of Euler, Ex. B. 12,

140, 141

IM

l\l

i.

DEI IM

UA
<>r

To

see that

this decimal

cannoJ
i"

terminate

recur,

we have

only to pn.\r that there is

>on whose aqua/re is

\ecangivpoesilil*
,

tli;it

(<//6)

= 2 we may
;

are mutually

j.iim.-.
i

ami
U-

'

j.i-iM.f thug Suppose, if are .positive integers which Now at least one of them is odd.
:

;'

<-an!i"t
,

must
/*

I..-

<,<1(1.

lint

if

we write

a = 2c, we get 2c2 = 68


contradiction.

SO that

.ann.t

l.-

..tl.l

w.-

thu> arriv.- at a

141.
It

is

The order of the system of jM.ssiM... ;iinl in many way\vhnlr


tli.

infinite decimals.
it

i>

.list in<-t

ly best, to l)iiill

up

tin-

.,1V

"!'

rational nuinlH-rs

,,n

tin-

basis of order,

be nuinlMTs as

marks distinguishing certain objects


1

arran-v<l in a <l-tinit- nnli-r.

1'.

numlM-rs to
shall
thfii

tin-

ri^lit
tin-

ni'

tin-

smaller.

u-ual. \v- ]>lac- tin- lai aln^ a slrai^lit lin',

we
<'

regard
tin/>'.

iiiMjualit i-s
.1

.!>/). ]{>(.' simply as


of

m.aninto the

that
of

mark

is

t<

th.- ri-'ht

an<l the
<\

mark

lefl

BO that

/;

falls

between
j
l>f<n<>

and

/;

\\'.'
I,, I

shall

now pn.vr

that KM cam

t lie

same

inent

directly.

that

w<-

fiml

and that
say that

tin- in'

I,,,

an-

tin-

same up

to a certain sta_

we

tind

.1.-...,

+;,
me,
n
.1

...

with

nin-fspoinlin^ interpretation for

/>',.

Then

\ve

1.

> 1/10".
would
in.

..it

a,,

may

be negative, but that <r,

'i...

,i.d

lew

tba;

IRRATIONAL NUMBERS.
Also lO n (B
virtue of the

[AP.
is less

I.

Bn

is

a rational number, and


of finding

than

1,

in

method

Bn

from B.
n
,

Thus
while

B<B

+l/lO

and Hence

A = An A n ^Bn +l/W n A > B.

Thus, in order to determine the relative position of two infinite decimals (derived from rational fractions), we need only compare their digits, until we arrive at a stage where the

corresponding digits are different; the relative value of these digits determines the relative position of the two decimals. By extending this rule to all infinite decimals (whether
derived from rational numbers or not) we can assign a perfectly for example, the decimal definite order to the whole system 1010010001 ... given in Art. 140 would be placed between the
:

two decimals
1010000
...

(zeros)

and '1011000

...

(zeros),

and and

also

between
101001000... (zeros) and '101001100
...

(zeros),

so on.

Similarly,

we may shew

that the infinite decimal derived from extracting


.

the square root of 2 must be placed between f f and ff

For,

by

division,

we

find

ff = 1-411...,

ff = 1-4146 ...

so that, in

agreement with the


ff

rule,

< T41421

...<{f.

It must not be forgotten that at present the new infinite decimals are purely formal expressions, although, as we have

explained, they fall in perfectly definite order into the scheme of infinite decimals derived from rational fractions.
142.

Additional arithmetical examples of infinite decimals


rational.
first

which are not


Consider

the sequence of fractions (an ), where

If

is

any

integer,

and n
1

> m,
1

we

find

.1

141, 142)

IM Mil.
I

DEI IM

LLa

which

is

l.-.ss

than
'

nt'.l

'Iin

-'

+
1
(

+
//
I
i

+
!,

l<
iliH-iniiil

V
'

'

V(iV ra+1/
Ix-tw.-.-n
tin-

Tim

>//<

the

for

lies

by
?lll

+l

= <'.+
hl)l
su.Ti->siv-ly
tini

In this

\\.i\

w,.

tiiut

lii:.

and
1708
i

Mid

w=3

:!;;

and TT TIHC, and


a.,,1
!

L71888
and

'H.

AJB

increases, these

two
t

<l.-<-imaU
1...1
t
<

hi-arly i-i|ual:

and

\v- an-

Inis

n^t

become nmn- and m<nv rud an infinite decimal


i

(] '71 8281828... X

Which T
1

-1

as.Mjuival.-nt

(1)
It

+ -+' +
c<

+
to
t<>

will
tin-

IIMW
<>nr

IM-

pi-MYrd that this intinitr di-ciinal can'


)rr^j)<)ii'U

with

which

any

i-atinnal

MUMih'T.

I-'MI-

tin-

dfcinial corresponding

111
31
lM-iinal

21

4!

J_ wl
il.-riv.-d
I'I-MMI

than th-

+
-2
'2
.

+
:{
-2
.

:>-

+ " + 2.3W - 2
'

'

Ami

the last ex;

i-

SW
Bence,
n<>

'-8-!(-^>
many
th.-ir

matt.-]- IIMW
d.'i'i\.'d
i'r..in

t.-nn^

we take from
l>c

]
.

^,-f

+ ...

the drcimal
75,
I'.ui
1

sinn will

less

than th- d.-cimal

'lian

1.

1
1

1_

21^0-H

ir^t.-fix

>

362
-and

IRRATIONAL NUMBERS.
so
on.

[AP.

I.

Thus the decimal

representing any

number

of

terms from
O\ O\

"I

\/ ^

\ /

must be

less

than

'75.

that (1) could lead to an infinite decimal agreeing with the decimal derived from a/c, where a and c are positive integers. Multiply by c\, and (1) becomes

Suppose now,

if possible,

The terms
iind that

in brackets give

some integer

/, say,

and so we

that

is,

an integer equal
absurd.

to a

decimal which

is

less

than

'75,

which

is

Thus no

fraction such as a/c can give the

same

decimal as (1) does. Consider next the continued fractions


infinite

&

=!+

jx"'

to

terms

Here,

we

recall the facts that if

bm

=p/q,
\p8

b nl+1

= r/s,

then
while b n
lies

qr\

= l,

between

6 TO

and

b m+l if

?i>w + l.
...,

Thus we

find for the successive values of b n ,

1,2,
so,

|,

I,

converting to decimals,
1

we

see that 6n lies

between

2
1-5

1-67
1-6

1-625 1-615

1-6191

142, 143

OBOME1

l:i<

.\l.

\\iri.l.lliniv
!

6
-ijiiaI.

;in.l

DXM

.-mil

lead

i"

an

iiitinih-

decimal

10...,
I

which can

he considered IB equivalent

bo th.-

intii.

inued n

L+ITiriTTTThis
fraction
infinite
r

'

decimal
ii'

caimoi

d"-n\.-d

lV.m

rational

it

were we -houM

have the inequality


'

_"
'/

<

/.'_'
'/

SO that
>inc..

\PC

~ a<J\ <
1.-

'

(pcaq)
is
j

IB

an integer,
,

tin-

la-i

.,!,.

litimi

Inn

this

nlivinusly alsurd.

iiiinator
C
(

gives r>>-: of

rth cniiN.T^.nt

than

il'

>
'

Similarly,

tin-

cnit inin-l

lVac-tin

14-L-L.L
2+
'2+

can
'I

IM-

]u-nvrl

t<>

1,-ad

to

an inlinitc decimal which


utiiiin-d
;

is

nt

rational.

lit-

rt-.-nicr \\lii. is

fainilar with tin- th>

fi

that the aqtuire of (3) con vergea to the value 2


in

whil<

leted
143.

..iiMfxii.n

\vith

tin-

-\aiiij>lc
it

of Art.

A-

SMiM.-what ditl'nvnt

.-xani].!.-.

is

easy to 866 that the

intinihi

dfcimal
i
!

I)

-30H
it'

J]

...

cann..t

rational.
.

For

it

\\viv .-<pial

"

C,

w- should
endfl

hut

10"
;

must

,-nd
is

with

0.

wh<

with

r,

or

Thus
\v

l<>'

&

Similarly.
lo^ai'.it

impossible, that 3,5,6,7, 11.... cannot

have nr

Inn-.

143.

Geometrical examples.
fche .-.\aiii]>lfs

I-YMIM
tinfulfil

--ivcn

in

Ar
is

14:2

it

i-

evident that

system
all

of

rational
,!'

nuinhers

l,y

no means sullicient to

the n-. ds
that
it

al^ehra.

\\'e

shall

now
r*

jrive

an ezao

to

-hew
L'-t

does not Mitlice for geometry.


line
.!/;
II.

^ht
Euclid,

be divid.-.l at
11).

in

"golden section"
:

(M

in

Bonk

.prop.

so that

AC \CB = AB

AC.

r ~Y~b

7"

"i

364
It is
less

IRRATIONAL NUMBERS.
then easy to see that

[AP.

I.

AC
off

must be greater than CB, but


equal to

than twice CB.*

Cut

AD

CB;

it

follows at

once that
at C.

AC

is

divided at

in the

same

ratio as

AB is

divided

For we have

AC:AD = AC :CB = AB:AC,


and consequently

AD:DC=AD:(AC-AD) = AC:(AB-AC) = AC:CB.


Also

AD
if

is less

than half of AB.

Thus we AN, which is

Now
rational

repeat this process 2n times, we arrive at a line n less than the 2 th part of AB. suppose, if possible, that AC/AB can be expressed as a
fraction
is

r/s;

then

DC/AB
is

(2r-s)/s.

Hence

AD/AB = CBIAB is AE/AB is (2r-s)/s


;

(s

r)/s,

and

and

ED/AB

Continuing this argument, we see that AN/AB must be some multiple of 1/s and so cannot be less than 1/s.
(2s
3r)/s.

than l/2 so that we are led to a contradiction, because we can choose n so that 2 exceeds s. Thus the ratio AC AB cannot be rational.

But we have seen than

AN/AB
:

is less

re

71

prove similarly that the ratio of the side to the diagonal of a square is not expressible as a rational In fact, let ABC in the figure represent half a square fraction.
It is

not

difficult to

of

which

greater than

AB is a diagonal it is at once evident that AB is AC and less than 2 AC. Cut off BD = BC, and erect
;

B
FIG. 38.

= and perpendicular to AB at D; then we have ED DA, = for the line of symmetry EC ED, because BE is a quadrilateral BCED. Thus EC=DA. If we repeat the same construction on the triangle ADE, we see in the same way that AF=FG =

DE

*The
less

first

follows from the definition

and so we see that


it

than twice AC.

Now,

since AC'.

CB=AB :AC,

follows that

AB = AC+Cli is AC is less

than twice CB.

143, 144]

\l.

365

Timarrive
at

.1

D(

than half
Tim-.
i.y

AO\ and

-imilarly,

/'

ii

than hair Ah.

continuing the eond

we
is less

an isosceh

lo

ANP,*\u-\i that

AN

than

the

L'

\\

Ar

A/I isa rational fraction

4 then
J

-1 1>

.!/>'

is(*-r)/,
\B

hat

that

AF/ABis 8r 2a)/a; AN/AJi is in lea than


.1
1

ami continuing the process, we see


I

I,

Of that
i

.\'

40

ad

less

than
"Ex..

r,

\\hieh

leads to a emit radict


m;i\
sli.'\\
I
.

n.

as !!'
th.- coiitiiiMfd
i

Tin-

ivadn
cle

My
tli.-

tli.it
'

of tin

oonrerget
ratiM

in

M
144.

to tin-

.f

di.i-"ii;il

to tin-

the golden iection side of a 8quai*e.

whfla

special classification of rational numbers.


I

imli-at.- tin- iif.-l I'mM> Th<- rxain]l-v ,.f Art-. I'.ut before smut- thirrational numbers. proceeding to
-'i

f'Tinal

l-liniti<n.

which

will

!><

found

in

tin-

n-xt

arti-l-,

a we

shall jrive

some

cnnsil-iMti<.ns
t<

which -h-w how


1

infinite .i.-cimals

which do not ivcur h-ad un


Tin- intinit-- d-ciinal
'

)cd. -kind's

dctinition.
in
c

1'4I42I

...

disj-ii-M-d

Art. 140 mal>l-s

divid- "// rational nmnl'.Ts into tw<


/"////

(A) Thi

r//xx,

which contains
than or
i-ijual

all

rational

fractions
of
tin-

(Mich as

jM

l.-.-s

to

smm- tmii

sequence of t-niiinat<'d decimals


1-4.
1

1-41.
t

1-414,

1-414-J.

etc,

(I

,)

T/

upper do88

\\hich contains

all

rational
>f

fraci
s-.ju,

It

is

ih'-n

(-uch as ;;> ^-i-.-atcr than every trrni drar that

tin-

(i)

Any

numl>-r
in

in

tin-

n]>p.-i-

rla^-

i-

-Train- than every


I..W.T class;

numltcr
(ii)

the lower
~t

dasa
nuinl-cr in the

Thnv
no
tin-

i-

ami

least

numlier
..f

in

the upper da--.


-t :it.-iiM-iit
.

'!'

M6

truth

tin-

<!.. n.l

\v.-

m.iv

obMTVe

that,

if

we have
bean;

/-/--i'(i
latiiinal
:

L'-/-'
iiiiin

>(S

MB,

we have
I

/(,
:

and,

t'^r tli.-

sam.g

that

\\ill

^ to

tin- l(.\v-r rlasx.


It'

'1'liriv is tlii-n-fiiiv n..

.mlirr in tin- l.\v-r class.


tin- u|>p<-r class,

MOW w. suppose

t<i

lf a iMtii>nal MUMilx-r of
/

we prove
-

l.v a

similar aix'uin.-nt that

i-

aU,, a innn';

less

tlian /.

366
Ex.
1.

IRRATIONAL NUMBERS.
Prove similarly that
if

[AP.
1

I.

F</V,

then l>k,

<N,
1,

where

l=(Na + bk)l(b + ak) and b*> No*.


Ex.
2.

Establish inequalities similar to those of Ex.

taking

Ex.

3.

The formula, corresponding


Utilise the last

to Ex.

2,

for the nth root of

N
first

Ex.

4.

example to find approximations to 2*

the

two may be taken

as

1,

f.

The

classification

of rational

numbers which has been just

described can, however, be obtained by a different process. the arithmetical process of extracting the square-root of

From
2, it is

evident that
2

2
,

2
,

(1-4)

(1-41)*,
;

(1-414)

(1-4142)

,...

but the sequence contains numbers which are as close to 2 as we please. Thus the lower class contains
are
all less

than 2

it

every positive rational number whose square is less than 2 and also contains all negative rational numbers. Since the two
;

classes together the upper class

contain all rational numbers, it follows that must contain every positive rational number

whose square is greater than 2. Thus the same classification is made by putting, (A) In the lower class, all negative numbers and all positive numbers whose square is less than 2. In the upper class, all positive numbers whose square (B)
is

greater than

2.

145. Dedekind's definition of irrational numbers. Suppose that some rule has been chosen which separates all rational numbers into two classes, such that any number in the upper class is greater than every number in the lower class. Thus, if a number k belongs to the upper class, so also does every rational number greater than /,. There are then three mutually exclusive possibilities (1) There may be ;i number (/ in the lower class which is greater than every other number in that class. (2) There may be a number I in the upper class which is less than every other number in that class. (3) Neither (j nor I may exist.
:

144, 145]
'I'll.-

hl.M.KINh- M.I IMTi>

i>\

repreeenl

nai niiiui.>

/'

represent thunit of !!

r.

the

upper class

c<

.11

rational points to the right of

on

tin-

lint'

and

tin-

lower class consists of v and


It

all

rational p-.inta to the

Similarly, \v- .MM illustrate case


thai
i/

(2).

mi^ht be thouglit
is

and /m:

>uslv; l.ut this

.-.vim

i.-d i,\-

at tiist sight the hypothesis

b
that
falls
-ial

nuiiil"
i

to

betWi

/;

and

this

be classed. Now $(g~ would .-...ajM- .'la ilirat imi.


<j

>nal

and

TliMt thri-c an- cast's in whic-li neither


1'nnii

nor/ can
wln-ntinit

.'\i

lh-

-\aiujil' ^ivt-ii

in

the

last

articl.-.

\va- prov.-.l

th<T' could lc no greftiesi nunilx-r in no least nuniluT in the u]]H-r class.


that

lo\\vr class,

and

\vlii.-h

ar.-

\ani|lr, h-t us illustrate mi a straiirlit liiu- tin- ajijro\iniati-.iidt-ri\'d from th; te to the rontintifd fraction
i

+ _L ^2+
'

LJ_ "
'1+ -2+

onvergenta of
while
an- tho>- of
It

tin-

lower class are Bf


1-

tin-

uiM-r
'

niav

..... l.srrvyil

con\rru ciit of the same


r

the other class


Tlien
ii_:uie
!..-

falls

l.et\\.

_J

J_ttl

_ _ _
i-

that

if

rant

of

.-itlu-r

clttM,

the-

next

'

',

\vliil-

>>

the intermedia!.- OOnVBTgei

hewn

in

tl

Je reproduction of the segment of

ti.

!ii.-h

L'

II

HI

It

nnu

ti

368

IKRATIONAL NUMBERS.

[AP.

I.

in the rational

It is clear that in case (3) the rule gives a cleavage or section numbers and to fill up the gap so caused in our
;

number-system, we

defining a new of irrational numbers*


said that these

agree to regard every such section as number. This constitutes Dedekinds definition

For

it

is

clear

from what has been


is

new numbers cannot be

rational.

On
so no

the other hand, in cases (1), (2) there new number is introduced.
;

no

section,

and

146. Definitions of equal, greater, less

deductions.
:

a,

For the present we use the following notation An irrational number is denoted by a Greek letter, such as the numbers of the corresponding lower class by small ft
;

italics,

as

a,

those of the upper class

by

capital italics, as

A,

B.

The

classes

themselves

may

be denoted

by adding

brackets, as (a), (A).


These definitions may be indicated graphically thus
i i i

a
It is

an obvious extension of the ordinary use of the symbols


to write

<, >,

a<a<A, b<@<R

In particular, we say that a is positive wT hen belongs to (a)', a is negative when belongs to (A). Two irrational numbers are equal, if their classes are the

same; in symbols we write a j3 if (a) = (6) and (A) = (B). The reader who is acquainted with Euclid's theory of ratio will recognise
in

that this definition of equality is exactly the same as that which he adopts Euclid in fact says that A:B=C:D, provided that his Elements.

the inequalities mA^-nB are accompanied by mC^-nD, for any values of m, n whatever. In Dedekind's theory, the inequality mA > nB implies that thus Euclid's definition implies njm is in the lower class defining A B that A B and C D have the same lower class and the same upper class.
:

the other hand, the number a is less than the number /3, when part of the upper class (A) belongs to the lower class {&), so that at least one rational number r belongs both to (A)

On

and

to

(b).

This definition of inequality also coincides with Euclid's.


*

Other definitions have been framed by Meray, Weierstrass and G. Cantor.

145, 146, 147

M.l'i
1

<

[QNa
it

It

follou

im the definition thai

</3and $<y,
{lumber.*
iste
in
tin-

then u

< y.
<'<i<ft,
.1

">f
I'-.-i^t

and
u

/i

one rational num:


i-

sii'-li

tli;it

</<,*.
that
i

NOTR
find

there

DO
li-.nal
/'.

numher
>

we ran
U

an

nuiiiln-i

hrloi,

and

than

Thus

a<r<<
Thru
if

iv

any two

j.i.sitivr

intr^ri-s.

w-

hi

that
147.

all

thr-r rational nunihrrs

lir

hrtwrrn

<i

and

ft.

Deductions from the definitions.


expressed as an

,'<!/.

B*OT

th.Tr
/',,

\\ill

hr
to
(

solllr illtr-vi'
tilt'

//
,

|ji.siti\r

<

U'

llr^ltixr) sllch
!

that

hrluiii;^

lower rla*s (a) and

/',,+

!>'

the

u]>}M-i-

cla-N

).

Thru consider thr


,-hi^T,
>'

rational

t'racti!

+
in

fl

......
t

"o+iV "M+
to class
30
i

"ln^
that

to

i.

MIJ
i

//+

//!

1<>

is

tin

dfl

that

\\v

'out

inning this j.rorrw.

\\

ari'i\r

at

the result
-

"

If

we

rail
!

these

is two decimal tVacti.n- " .1 plain that made [688 than any proi-rihed rational
. .

i;

i'racti<.n

merely

l.y

taking

l.e

sutlicieiitly
\\

great;
that
((

a:
is

We (-..IliinUe the p numher detined l.y the The ariji.ment


A
\\

indefinitely
infinite decim;il

see

the

JflO
t,,

that
t'

fix

f/,,,f

i^

u-et'ul
;tll\

i..

not,-

further that
r/'
<

can

l>e

chosen so as to
el,:

pl-rviTiled ilUmher

>f

the loWel"

370

IRRATIONAL NUMBERS.
is

[AP.

I.

a" be another number of the lower class which a''; and then choose r so that 10 r a'). l/(a"

greater than

>

Then
a".

Ar
But 4 r >a",

ar

<a"

a',
a'.

or

ar

a'> A r

so that a r

>

Modified form of Dedekind's definition.

Suppose that a
(1) if

classification of the rational


:

numbers has the

following properties

a belongs to the lower number less than a;

class

so does every rational

(2) if

belongs to the upper


:

class, so

does every rational

(3) (4)

number greater than A every number a is less than any number A numbers A, a can be found in the two classes such
;

that

Such a
irrational.

than an arbitrary rational fraction. classification defines a single number, rational or


is less

For any rational number r which does not belong to either must lie between the two classes, since any number less than a number of the lower class must also belong to the lower class and therefore r must exceed every number of the lower class similarly, r must be less than every number of the upper class. Hence, if a, A are any two numbers of r A the two classes a
class
;
:

< <
s is

a second rational number which belongs to neither class; then a<^s<^A. Hence \r s\ must be less than A a; but this is impossible, since by hypothesis A, a can be chosen so that A a is less than any assigned rational fraction.
that
classification

Suppose now

Consequently, not more than one rational number can escape if there is one such number, the classification may
;

be regarded as defining that number; but

if

there

is

no rational
a Dedekind

number which
section,
148.

escapes classification,

we have obtained
irrational

and have therefore defined an

number.

Algebraic operations with irrational numbers. The negative of an irrational number a is defined by means of the lower class A and the upper class a; and it is denoted -a. by
Tin reciprocal of
1

an

irrational

number
first

is

defined most

easily

by

restricting the classes at

to contain only terms

147,

H8|
i

\l.<;i:i;i:

\n

OPEB kTIO
>-,il

li.-n

L/a

by the lower

.1

ami
t

tli.-

upp-r
tin-

class

ft.

Thus
,,f

if

th--

number
lower

'

ive,

In-

compl.-t.-

s|MM-ili,-;it

ion

th.I

will

he

-i\'-n
ih.-r

by putting with all negative numbers, while


t! i-

who),-

of

.1

in
t

th-

class,

188 will

contain

only

definition

framed
t>

l'ir

when
al

tl

is

m
i

nurru
as
ti

always
or

i\i

anl
:

i^

i-.jual

<.r

i^

],..siiiv-

ive

it

is

(l.-nt-l

by

|o|.
'/X.

B|
ii

,-i

fco

be
/>'.

the

two

given
tin-

irrationals,

so

that

< <
Art.

.1,

making
the up],
o!'
1

Tlh-n dassily a IIH-HI!MT <f tin+/* typical


b
o'

< <

rational
cl.i
<-,

nuin

lo\v.-r

nil.-

nl-\
it

iou.sly satisti.-s
s.-u i.sii.-s

,ndit i,,us

17.

To prove
///'

that

ronditinn

(4)

ami 8O

l.'tii

nuiiil.-r.
.

w-

not./

that

14. />>>_(, /
at

,..

u i_ (M _ H(
i/

/;_
147.
<-ach

ami, as .'XplaiiH-d
!

thr

iM'^innin^ of

Art.
A'
//

\\v
l.-ss

can tind

and

/>.

II
.1

BO &8 to
(r/

mak- J
-f/o
is
Lefifl

and

than

.',e:

and then
nunil
It

+/>')

than
I..-

t.

II.

Mice

our

classifica:

tion dctin.-s a iinnilxT \\hich


.il.-d

may

rational or irrational

this

tin-

MUM
typ.-

<t+/3.

i'ollo\

DO thai
tin-

o+(
'

is

represented ly
is.

a)=0; for here the lower dan J.and th.- upper das- i,y A
<t.

That

(lie

lo\v-i- rlass oonsistfl


all

and
is

tin-

upper class of
t

"fall negative rational numbers positiv rational numl.ers; he;


not
class,-.!,

the only rational numlier defined l.y

numher

and

tln-r-

iticatioii.

In

virtue of the relation


jual

,'*

/8)

= 0, w- may

detine ,<->'

t. tlie suni
>!><</

,+(

ft).

M
l-'or

't

i/

simplicity of statement,

We

<>mit

the

numl..-rs

from the lou.-r olasses; and then \v,- detin.- the jro(lu'.-t a/3 by usin- the tyj>e ,if, for tin- lower elass and A Ii for the &
spondine- upper claSS,
let
t-

To prove

that this

.1.

-tines

-.in-l,.
l.-sx

nun
tha

denote an arbitrary positive rational fraction

372

IRRATIONAL NUMBERS.

[AP.

1.

then choose any rational number P which is greater than a+/3+l. Next find numbers A, a and B, b such that A a<^6 l ,B b<i l

where

el

= e/P.
}

The determination
Art. 147.

Now we

of A, a and B, b have

is

possible in virtue of

or

That

is,

AB-ab<6.
;

Thus the classification by means of ab and AB defines a single number which may be rational or irrational and this number is called a/3.
In particular, if /3=l/a, the product is equal to 1 for the lower class is represented by a/ A and the upper class by A/a.That is, the lower class contains all rational numbers less than 1
;

and the upper


which escapes

class all rational


is

numbers greater than


1,

1.

Con-

sequently the product

equal to

the single rational

number

classification.

Multiplication of negative irrational numbers is reduced at once to that of positive numbers by agreeing to accept the "rule of signs" as established for rational numbers.
Division.

In consequence of the relation (l//3)x/3 = l, we may define the quotient a//3 as equal to. the product ax(l//3). It is at once evident that any of the fundamental laws of algebra which have been established for rational numbers

remain true for irrational numbers. Thus, we have the following laws

For example,

let

us prove the theorem a + )8 =

By
and

definition

we have

But a + b = b + a and A

+n = fl+Aj

so that a + j3
r<|ii;il.

and

/2

+a

are dHinrd ly

the same two classes and are accordingly

148, 149]

KONOION*

NUM
in a ^imii
intial iiili<--s

t"

and

of |M-^;H

if

Inn-,, ..n tlith.-ii

t.

li.-

i-itli.

149. The principle of convergence for monotonic sequences whose terms may be either rational or irrational.

A monotonic sequenc
iiinnal

-n

in

tl.

niimi
defiuiteneSfl
tin-

Suppose I'min which


;
'

that

the ^fiju-ncf
le8G

i^>

an

-main

than

lixed

nni

;lt

...<A.
i.-il

numl.er, one of
in

two
(a,

alt
IK-

HUM
to
Ol

occur;

eitJ

term

sequence ry term
;

tin-

will

t-ijual

!'

will l^e

[066

than

/,-.

\\\> d.-linc tin-

afl
-.

tin-

<-la>s

nf

all

rational
(

//

\vliich
all

sati>,t'y

In-

tir-t

,iiliti<n.
<-.

tin*

class

tt)

as

<>t'

\\hirh

>atis|'y

lln-

96COnd

,mliti.n.

Ty]ical
!>'

!'

the C

are

tin-

rational nunilM-rs

which

the
it'T
It

Beqv

than
rl<-ar

.1.

\\liilc (II) contains .-vry i-ational nuinh.-r and possiMy sonic rational nuinhrrs ].-ss tlian J.

i^

that

tin-

d
\V.-

Aether
/-

<-.

.ntain

///

ra'

HUIII|MT>.

and

th-r-

which
call
i

dctini

nuinl.cr ^. rational or irratinnal.

may

/,'

t'othc

\\]

and lower
\..\\

clfl
i-a
;

every
i--

ju-ctivdy. defined ly the se.jiie; i"ital numlter -Teater than >1 I'clon-'s
;

t<>

the

ujijier
th--

oil

same

tru-

And ter than any term .' nnmlter in-atimal y ^ivater than ft ry
,.
i

there will he rational nuinl>errational nnndn-i


ter
r

:id
-',
:
:

.i.

and these
also

than any term


<

th'.

than

any

tequentlj
rcecd
ft.
o'

the other liand. every rational nuinhcr lc>s than helon-- to the lower Now if t i- any
<

)n

inu^t

tiiei-e

will

l.e

rational iiuinl-

these nunil"
i.

than
i>,
f/,*
:

.-},

they nr.
}
e.

That

374
Hence, since

IRRATIONAL NUMBERS.
a n = am,

[AP.

I.

we have
That
is,

@ = a n = /3

e,

^ 7i>m, if n > m.

lim a n = ft.

such a sequence is afforded by the terminated decimals derived from an infinite decimal and it will be seen at once that the section described here is an obvious extension of the method used in
;

A good example of

Art. 144 above.

Suppose next that the terms of the sequence, while still It increasing, do not remain less than any fixed number A. is then evident that if a m ^> A, we have a n A if 7i>m.

>

Thus

lim a n = oo
n >oo

Exactly similar arguments can be applied to a decreasing


sequence.

As an example we shall give a proof of the theorem that any continuous monotonic function attains just once every value between its greatest and least values. Suppose that /(#) steadily increases from x=a to x=c, so that

b<d,
Then
is

if

f(d) = b and/(c) = rf.


d,

also

if I is any number between b and between b and d suppose that


;

we

consider /{^(

+ e)},
less

which
,

this is

found to be

than

write then
<*!

= ( + = c, Cl

<!),

b 1 =f(a l )<l,

d =f(c )>l.
l l

On

the other hand,

when /{J(a + c)}

is

greater than

I,

we

write

d =f(c )>l.
1 1

Continuing the process we construct two sequences (), (c,,), the first M never decreasing and the second never increasing; and c -a n = (c-a)/2 Also by the method of conso that (an ), (CM ) have a common limit k. unless it happens that at struction it is evident that f(an )<l<f(cn ) some stage we find /() = I, in which case the theorem requires no
;l

further discussion.

Now

since f{x)

is

continuous

we can

find
I

f(c n )-f(a n

)<,

if

n>v,

and both f(k) and


find v so that

an integer v such are contained between

that
f(
,)

and /(CM ).

Thus we can

\f(k)-l\<

if

n> v

and therefore, as
it is

in Art.
is

clear that there

= f(k) l. From the method of construction and this is also evident only one value such as /
1

(6),

from the monotonic nature of

#.

149, 150]

LIMITS

150.

Maximum and minimum


i

limiting values of a sequence


are less
M.
1

of rational or irrational terms.


tli.it

all
il

tli.-

than SOUL..mil

iiiiinl

l.-t

p be a

smaller
ater

inn
'I'll. -ii
i!'

icfa

that
i

an

intiiiity <f

tennfl

tlian
it

p,

we

is

evideni

an

intinit
id
/'
:

y or
iii

a finite
tl,

number
r

lall

beta
'

case

we

\N

in

the

latr-

\Y<

bains

an infinity "f term


gee are
i

.-at

the

in tin-

diagram.

P J

Then

the
lees
'/
(

sequence />. /',. than /': and so


<!'

/>.,.

/>....

never
1

1-

anl
t'-nniiK'S
]

th.-

sequence

which may
article
/'
i.

COUTSC K- cithfi rational

irratinnal,

ID

tin-

la>t

Thru the

limit
i!'
-

a.
i-

because an arhitrarily small


ff,

p*(Pp)
:

-'

sequ< * n

U M
n

the

>aiin'!'

increasea

]Miti\e

numli.T.

we can

tin.l

//

/'

G'-fe.

Conse.|uent ly am

vn

"i"

Thus we can .letennine ami we can its limit


:

the

bermfl

<>f

the

Bub-aeqtience 1'p'iii (" ..) which has G a certain stage after which all are less I'f + e: than thus no >e,juenn.
a

find

convergent

sab-sequence
s/H-tr

can

have
;>
//

Inn

ber

thai.

tlmt

a
/'

A qut
It'

DO

such

numl.ei-

afl

can
of
80
if
/

be

l'"Un<l

in

ar-'uinent.

th.-r.

;iiml)er8

the
('

sequence

ater

than
flu a

at

naMe numher.

JO.

On

the other hand,


\\ill
IK-

no such numl>er as

can be
_^r.

found, there

only

rm^
1

than

.V.

however
lim "
B

consequeni

376

IRRATIONAL NUMBERS.

[AP.

I.

sequence has a
All the

For the sake of uniformity we may say even then that the maximum limit, which is, of course, oo
.

foregoing discussion can be at once modified cc ). establish the existence of a minimum limit (g or
151. The general principle of convergence stated in Art. 3
is

to

both

necessary and sufficient

n >oo

the terms being rational or irrational. In the first place, the condition is obviously necessary for if lim an = I, we know that an index in can be found to correspond
;

to

e,

in such a

way
|

that
l
\

Thus

a n < Je, if n^m. -am |<e, if n>m. -a l-a n \+ m |= |an


is

In the second place, the condition


fixed so that

sufficient

for let

be

or

n > m, am -Je<a n <am + Je, if n > m.


a n -a m
Je,
if

<

Then
has a
index

it

finite

follows from the last article that the sequence (a n ) maximum limit G so that an infinity of terms fall
;

between

G
is

Choose one of these, say ap whose than m. Thus we have greater


Je
6r-fje.
,

and

Also

ap

Je

Thus and since


it

follows that
;

+f a m \e<^a n < + Je, G e < an <C (r +


e,
a,,,

- f e < aw <

<

<x

< ap +

Je,

since

>

>
7?i.

w..

if ?i
if TI

> m,
>>

e,

Thus a n -+G and consequently the sequence is convergent. Of course in this case g G, the extreme limits being equal
in a convergent sequence.
Various proofs of this general theorem have been published, sonic icing much shorter than the foregoing series of articles. But on
1

apparently

of the shorter investigations it will be seen that. in all cases the apparent brevity is obtained by avoiding the definition of an This virtually implies a shirking of the whole difficulty irrational number. for this difficulty consists essentially* in proving that (under the condition

examining the foundations

of Art. 3) a sequence
*

may

be used to define a "number."

I. A. 3, 14) says: "As the truth of this theorem and exclusively on an exact definition of imitituml inn.', naturally the first accurate proofs are connected with the arithmetical theories of irrational numbers, and with the associated revision and improvement of the

Pringsheiin (Encyklopiidie,

rests essentially

older yeometrical views."

150, 151, 152

LIMIT1

"I

MI MIII

152.
I.

Theorems on limits of quotients. and lii // Mm",,


<>
i

th>-

steadily decreases,

Inn

,.

<',,
.

--lim

,.

."

second

</

SUJ.J.M-.:i

tirM

that tin- limit

i>

linit.-

ami

-'jUal

to /: th.-n
foiiinl

if
t

arl.itrarily small j.o.xim,. Eractl

AD
u

so

i/t

---

</+*

inoe

(''

'',.

i'-e><^-^..,<'',. -''-,<<' + *
Chaii^tllfll
/'

to

/'

!.

/*

+'2. .../'+/'

ni'l

ali

tin-

\Vr

tillil

(/-cHl*,,-!*, .,,)<",<-".,,

<(/ + f
as
/>
~ s.
;

..

p ).

N>w

takf the limit of this

n-.Niilt

w-

ol.tain

()

ly hypoth-

Hi-nci-.

sine

positive,

we,

h.-r.

-/
or

if

//

>
\\-

..

liini",,

'')

/.

On

the oth.-r hainl,

if

tin*

limit

i^

can timl
,

so that

>
how-\,-ibefore,

.V.

if

>

lar^r
re
timl

!i

ctly the same aixr um-nt as

..+*).

which
or

1.

A'.

if

/>

>

m,

Thus
is

linn,/,,
in*

AJ-

x
that, \\ith tht-

ditliciilty in
D
"11

'\ti'inliii.

llir

iii

t>n

'

;"-""-'.

378

IRRATIONAL NUMBERS.

[AP.

I.

This theorem should be compared with the theorem (L'Hospital's) of the Differential Calculus, that:

//
then

lim0(X) = 0,
X>x>

limbec) = 0,
<f>'(x)/\{s'(x),

lim 0(aj)/^r(aj)=lim

X->oo

provided that the second limit exists and that \^'(x) has a constant sign for values of x greater than some, fixed value. II. If b n steadily increases to oo then
,

n+l~n

provided that the second limit exists* For if the second limit is finite and equal to I, as in I. above, we see by a similar argument that it is possible to choose

so that

(l-e)(b n

bm )
is

< an - am <
positive,

(I

+ e) (b n - b m

),

if

n>m.

Thus, since b n

Now,

since b n -+ x

we have

and

lim
n^-

(i

+ e)

= +
i

e.

And

so

we

find

But these extreme limits are independent of m, and therefore and so the inequalities can only be true if each of also of e the extreme limits is equal to I. Hence
;

lim (a n /b n) =
Similarly,
if

1.

the limit

is

oo

we can
),

find

m, so that

a n -a m

>N(b n -b m
be.

if

n > m,

however great

may

Thus
*

i
for the case b n =n.

Extended by Stolz from a theorem given by Cauchy

152
1

LIMITS
tinliat
-

an.l

limit

of

t!

on

lh-

is

N, OS

Jim

'

nn\v this miiiiiiniiii limi?


-I---'

and

lli.

Thus

Inn
imilarU
tli:it

with

tin-

'
I

| *

Cf M

'

'/

"~~
1

(.It.

-o*
IVIM
(

-o w
tinI'nllov

IL should be compared with


1

<>rem

LI

I'

>ill'T. -Altai's) of tinincreases steadily


)

f<>

s_

tt-ifl,

./,

/A

'^1:
/,</

tin

it

d,,>

second
..ut
i.f
|il,

//'///

It

|tr<'l>,illv

n-t

nl mi

tliis

iiiijxirtant

theorem,
>of .*

he extemU'tl

f"i-in

Ban-value

tln-'-i-.-ni

\\-.-

have

and

>>(

')-'>(,,)

is

j.ositiv.-.

Tin:

adfl

to a limit

-lin,.se

a so

tl

')

Ami

fr'lii

hn
if

.....

\vai(U

tin-

;n -iiniriit

])i-

Similarly,

that
</>(a-)-</>(a)

>()],

Ex.1.

If

,-r+...+

^=
iiin.laiiirntal

(^+ i ) p
+i_ n p+i-

A
limit
..f
tl

ilus.

H.'

lin.

JL,
liiu" 6.

ami 80

380
Ex.
2.

IRRATIONAL NUMBERS.
If
.

[AP.

I.

a n = lo7?
n + i-

=n
\

we have
so that
if

bn+ i-b n

n
161.)

lim

=0.
,

(Compare Art.
find

Similarly,

a n = (logw) 2

b n = n, we

^
which tends to

= [log n + log ( + 1)] log (l


result.

+1)

< | log (

+ !),

by the previous
that

Thus
Similarly

\irn[(\ogn)*/n]

= Q.

we can prove

lim [(log #)*/] =0.

The reader may


Ex.
3.

also verify this result

by using L'Hospital's theorem.

If

a n =p n
a n+l
n

bn

= n,

we have
and hence

lini(a n+1

Thus

-a n =p (p~l\ & JI+1 -6,, = 1, -a w ) = 0, if/>ll, = x if p > 1 or lim(p'-0, ifjo^l,


,
.

or

=ao,

ifjo>l,
2.

(Art. 161).

Of course Ex. 3
Ex.
exist.
4.

is

only another form of Ex.


(a n )

Even when

In this

case, the

and (6 W ) are both monotonic, lim n /6 n need not theorem shews that lim (a n+l - a n )/(bn+l - b n ) does
given by

not

exist.

An

example
o

is

Here

i-a=/ [^+p(-l) ll+1 -g'-( -1)"]


fl

and so steadily increases Then we have

if

q>(p + l)'/(pbn+1

1).

^.iiz. o
n+ i-o w

jo-1
1,

= j-l. -b
n

p-l

while
Since

lim(aw /& w ) = 2-

lim (/&) = ?+ 1. + (jo l)/(p-l)>l,

these results agree with the extended form of

Theorem

II.

Ex.

5.

If

ow

we

see that

o w +i-a n = 3 + 2(-l) ll+1


1 /(

Thus

(a /l+ i-a, )/(6 H+ i-6

oscillates

between

and

5,

although

Again,

if

we

find

152, 153

\l-.l

!--

1. 1.

MM

\.

l/2(-f 1); and 00


I,,,,;

tt-s

mill,:-

a drlimtr limit.

Ex.
.saiilx

6.

It'

!">

ii. .t

steadily in.n-aae, the Theon

II

neces-

Tim*

aw =

ii

&.-<
ntlv
!

=
)

luit

.-

= 1.
\\ln-n \fr'(x)

In

lie

same M
"ii-id.-i
-

.inn

may

fail

chang*

intiiiitrU
'riiu>
.

/+! +>

^('
for
\\hi.-li
.

\v,-

tin.

i
|

-(),

\vliii-

Ms.-illat.-s

bet

and
Ex.

;i>

7.

<

'"ii-i<tfi-

tin-

and

|>i"\f that

Mm

= l,
.,-)=!+<,.

wl.iK-

Mm.

153.
y,,

An

extension of Abel
'
'

Lemma.
r -{-,.,
i

,/, /,

ran '*"

.\

,,

/,

rn

-f- //

-\-

Ii

''.,4-

'V

~^~

~^~

"

,
'

\\i'itf

.1,,

'',,,

A =
,

'',,

''j

.-1

((

ami

/'

/'
.
,

=
/>',

!> {t

-}-/',.

/>'
,

= '/,,-f-Cj = + +
/'

-f~

~f

'>!

4-

'

Then
\

^(Vo-^) 4-4
(

/*,(
'",

'-,

4-

-f />,

.i(v n .i~
% 1

'>-',)-

-...

+ ^^

('

H-i-v n )+^
re all

//'

]n.>iti\ 6,

We
//

'MII
.1

nl.tain

;in

n]|n-r

lin,

in
,

and

//.I

in

jla--

..f

/;

0,

382

IRRATIONAL NUMBERS.

[AP.

I.

where H,

Hm
B
l

are the upper limits of


^ ->, -pJi + n/

(B l-2-> -j-, \-L ^Ij


respectively.
. . .

Bn

ana

f or

/Bm Bin+l
,
,

...,

\-A-m

-"-?n+i

Bn\ "'
V^

i,

This follows because the factors


are
all positive.

(VQ

(t^

v 2 ),

(f ri-i

vn ) and f

Thus

and,

we

replace J. r by its value in terms of obtain,* since m by definition,


if

we

H^H

+a +
1

...

+a r>

"

i^ 'i

~T

In like manner

we prove

that

where

h m are the corresponding lower limits of Br/A r Secondly, suppose that the sequence (v n ) steadily increases. In the numerator of v x ), (v x f 2 ), n the factors (V Q
A,
.

( v n-i~ vn) are a ^ negative, while v n is positive, so that the value of the numerator is increased by writing

hA r
and
/i

in place of
in place of

w -4r

Br Br

= 0, 1, ... m 1) = m,m + l, ...,n (r


(r
,

l),

while in the last term

we must put

Hm A n

in place

of

Bn

This changes the numerator to

= k m (a
and, since h m

+ .+a n v n + (Hm - h m )A n v n + (h - h)(A Vm - " r - - am vm


. .

tn

llt

),

^h,

the

negative

this again will not be decreased by omitting terms in the last bracket. Hence, since the

denominator

is positive,

(Hw - h m } A n vn 4- (h m - h) A

in

v in

*The numerator
which
is

is

actually
+' 0,1?!

av
less

+ ... + a m r m - (a

-f-

^+

..

+ a m )v

than the value given

153,154)

Till

1.

Mil-

Similarly,

w-

timl

n _
'/'A*

II,,,

1'

m )A n vn

+(H-H m )A Mvm
.(
is

ilir

We may sequence, and


'')

xuppn.sr that the trim Vf

tli-

-iv a t.-t in the


i

than

/-.

ll.-i-,-

K-tors

while
!

vw are

Thu^

iln-

numerator of

Xn

is

greater
/'| .!.,(

tlian
I

'-',)+...-!-.

+/',,,M.

'

,)+...

+ //
= y/
-...+"

^4)+

+4i
-...-"

/'

x.i,,.'-,,.-'/.

'-...th<-

Hriir-

we deduce,
V

l-y

an ar--um-nt similar to

<
we

//

Similarly,

find

av +a
Ex.
l'i-..v
(

v l +...+a n v n
divi-r.Lr/

iha;

be

provided that
154.
It

-. ....
...

+ 6,,

[.Ins-

Other theorems on
t'n>ni

limits.

follows at onre

Theoivm

II.

of

Art.
to

1">2.

that

if

->
Thus
l.y

ha-

definite
8m

limit.

fcendfl
6
/

the same
\f a
-

limit.

writing
<i

=a

+Oj+...+a
I

that

as

iin

if,

fh>-

fi>

lim

(r/

-|-

/.,+ ...-!*"

I' 1

t!

;ivj,

monotonic

l\it it

saves

troul)K' to i-xjiiniiu-

it

lirr-.

384

IRRATIONAL NUMBERS.
limit

[AP.
first

I.

Of course the second


with
..._!

= 1,

a-j,,

= 0,

may

exist,

when the
,

does not

thus

the second limit becomes

although the sequence

{a n ) oscillates.
III. If all the terms of a sequence (an ) are positive, lim if (an+l /a n ) is definite, so also is lim an ; and the two limits are the same. [CAUCHY.]

THEOREM

and

For,

if

lim (a?l+1 /a w )
,

is

finite

and equal

to

(not zero),

we

can write a n = c n l n so that


lim(c n+1 /c n )=l,

Thus we can

find
1

m, so that

< cn+1 n <


/c

e,

if

n ^ m.

By

multiplication,

we

obtain

so that

c w (l-e) w
c

<

cn cw

<c m (l + e)
<c, n
(l

tt
.

Hence

m (l-e)<
n >oo

+ e).

Now
so that
1 e
i

limcm
lim c n

=l

(see Ex. 4, p. 17),

^ lim cn ^ 1 +

e.

of e;

these extreme limits are independent of and therefore e is arbitrarily small, so that the inequalities can only be true if each of these limits is equal to 1, or if

Now

and

lim<v = l.
Thus
But
if

lim

n
oo

= I = lim (an+l /an


,

).

lim (a n+l /an ) =


<-i>

we can
t

find
if

m,

so that
*

n +il<*>n>N
be.
a> n (f

n^m,
m

however great
or

may

Hence, as above,

in>

>l

ai>^[
But
n

w /JT]i.

lim K,/JV']

=1

(Ex. 4,

p.

17),

so that

lim a n

J\r.

This

minimum
is

limit

is

and

independent of

m and therefore

oi

arbitrarily great, so that

wr must have

Lim On*

a oo.

154]
h.-n

HEOREMfi
1

LIMIT-

Mm
tli.-

can be reduced to

th.

|,rt-\

ioni

:n _"!in.-Mt

to

proTi

r:il

lim
.ilnl

/.,".

');
',+iKX-,

alsi.

it

'
.

tod

reaaes to

Ex.

1.

lit.:

ft

n*,
,:,t

'

Thus

lit.,

result can

rig's

fonnuhi

f(.|'

(SIT

\.

Ex.2.

To

liiul

HIM

[(//<

l)(// <

+ 2)...(

!*

tixed,

ami

tin,,

/'+
i.-u

tin-

limit

ajruin
\\t-

is

Ex.

3.

Similarly

find that
l

HIM

[(/<

-<,

I\".

O.
]

90

that

Thm

p n ->0,

'/

tin-

-iv.'ii

ezpreaaion tuk.^

ill.-

f..nn

"^^'/i + .-.-f
Now, My
li"

Pi+

4
"'

-f...

+r
that

the r.-ult at

tin- iM-^iniiinir

'lii^ artid.-. \\,- SIM-

/,-fV,, -f ...-I-.

386

IRRATIONAL NUMBERS.
to

[AP.

I.

and consequently we have only

prove that
-

lim -(Piq+piq*-i + tv

+Mi) =
we can
and

Now,
which
JUL

since

pn

q n both tend to

0,

find a constant

is

greater than \p r
\p r

and

\q r \;

further,

we can

find

so that

\<e/A,

qr\<e/A,

ifr>//.

Then,

if

>

2/z,

we have
\<e,
if

\p r qn-r + i\<A\q n - r+l

r^$n,

n-r^$n> p,
p..

and

\p rq n - r +i

\<A\pr \<e,

if

r^$n >
if

Consequently

-|(Mn+Mn-l+-..+Ml)l< e
IV

>

>2/*,
is

and so tends to
established. 155.

zero, as

n tends

to

oo

thus the theorem

THEOREM

V.

If 26 n

2c
.

ri

are two divergent series of

positive terms, then

Co + ^4-.. .+C

_ - 11111
i
.

provided that the second limit


condition*

exists,

and

that either (1) c n /b n

the steadily decreases, or (2) c n /6n steadily increases subject to

is fixed. where Let us write for brevity

Let us also write

c n /bn

= vn
b v

',

then

gn = (^QO)^O +

Cn

+...+b n v n

The second part


the
t.

of the

1907, p. 269;
(2),

first is

13,

1889,

p.

61.

theorem is due to Hardy, Quarterly Journal, vol. 38, given by Cesaro, Bulletin des Sciences Matlnmatiques Of course A'>1 in case (2), in virtue of the t';u-t

that

c n /b n increases.

154, 155

BEOREM8 OH LDA
tin-

we can apply
tie-'

An

and
(1)

ire

tind.

in

the

CM6, when
i{

-(hm
//,
/,

-h&<$<Hm +(H-B
'

^n
x

whento
f
It'
.

an//
.

tin'

iipp.-r

and lower limit-.


!'

.1

:md
I',,

are

Pm/Bm
BO that

P
A,,,
/

lini
si
i

&,-, We
1'n -in

can tind
<

01

<:

II

+ e.

and

we have

>

(/-,)-,/-/,,'/;<;;<(/+.)+(//-/.
BO
in-.|uality,
\v.
tli.-it
it'

u.-

tak-

th-

linii'

last

tind

/- t
I!*
-ii
a

inn/; ^n
/

I...,;!"

/-he.
<t'

;|

arliitrarily small,
t<
:

each

tin- .-xt r-ni<-

limits

in

HUM

!)

.Mjual

In likr

manner,

it'

l n
'

>

inn(^C'n ) = /. x \v<> can H,,


,

tind

so that

\'.

""'

"""
'.'

>.V-(.V-A
i'ruin

Sine.-

*'

"

-s_

\v-

tind

the last in-i|uality tliat

\in(QJCn )^
and
linKy,,
<

In tin- s-o>ml case, wln-n c n

!>

in-retises,

we have

//,-,//

-,//-/,
( fi

<^<*.+
N<.\v.
li\-

hvpoth
a
//

',"

< A'

0,,

/.

\vhnv

A"

is

QODSU

fcei

than unity.

II.-nc.-

A ',//

-/,
(

_
-A
tind

< /;'</',-f A'i//


then,
jn.st

A'
<

-/-^
II
>

as c2)

is

d.-dur,-d

fmn,
in,,/;'
i'.

,.

we

from

that

/-C-Vv'-lk
I'rniii

/-frJA'as

which the same

result

.ll-.vv-s

388

IRRATIONAL NUMBERS.

[AP.

I.

in the first case the series 2c n diverges more slowly than 26 W while in the second case 2c M diverges more rapidly than S6n but the final condition excludes
It is instructive to note that
, ,

series

which diverge

too fast.

It should be noticed that if sn tends to a definite limit, Theorem V. is an immediate corollary from Theorem II. of Art. 152 for then both fractions have the same limit as sn
;
.

The applications

of
'

most interest
b

arise

when

= \=...=b n = l,

and then we have the result: // 2c n is a divergent series of positive terms, then
Co

+ Ci+.-.+c*
second limit exists

n+ 1
and

provided that

either (1) that c n c or that increases, decreases, n steadily subject to (2) steadily the restriction
the

nc n
where

<K(c + c

+...+c n ),
arises

is

a fixed number.
from applying the

Ex. 1. A specially interesting application theorem of Frobenius (Art. 51) to the series

where

form an increasing c c lt c2 ... satisfying the condition just given.


,

sequence of positive integers,

Here
so that

it is

evident that the series must be written in the form

if

= ao> -4q>-i ^o = o> ^i = o> A =a -}-a A +i = aQ + a ..., and so on. A v =a + a + ...-}- ant Generally we have c + c + ... + c, _i^v<c + c + ...+c n we have Thus, if 8n = aQ +a + ...+a A + A! + ...+A V =s co+ 5 c + + *n-icw_i + s(v - c - c - ...-c,,_
>

Co

Co

1l1

--

and therefore Frobenius's mean,

if it exists,

is

given by

which we have proved to be the same as


provided that the last limit
exists.
1

Ex.

2.

Interesting special cases of Ex.


6-0

are given by taking

+ 0! +

..

c,,

= (w + l)

;2

(w

+ 1) 3

etc.,

for

which

may be taken

as

2,

3 respectively.

155

THEOBEMfi "\ LIMITS


re.--.

Th us we have the
lin.
x
1

<*+...) = K">
n
ae

*+*> +

+*
"*"**

an-

tr* + 0^r

r +...) =

I'

*'j*

Ex.

3.

Hut

if

wt \M

\e c

= 2,
our
-

="ixlit ion is l.ioki-n. so

Thai

i>,

that we have no right

<

lint

'

(*o
;

+ .+ *)
...

""I

;ls

matter of

fact

particular
t.-iul^

SMJ.
I

can

roved to oscillate as j.
[H.\

to
4.

(oompti

!>.

489).
tO 8
,,f
tl.

Ex.
liv

We

can n>1.

tlii>

tin-., i, -in

taking x,,=

Then,

in

tli-

notation

rt

=jy,

'1
i

oonvergea,
:niot

\\\-.
i

.res.

Thn-

aj)]ioa<-h

anv

limit

"tln-i

tlian xero.

Ex.5,
an-i

w
/

/;('' \6
;

;^')=/ i*i/
...

+ /;
that

fl

Irt

A,,

/',.

/;..

th. MI

].!-o\,-

)'
i,

in

CMC
1

x, lut that

/'

'

ly applying Art.

that

*1:

pi

worth whiK-

bo

;i

ce

does actually incriMse

1.

v.

Iii
.

fa.-t
i:;:;i

the
in

loth
the

an<l

11

'"Is in

the

first

sequence, and

"n.l.
is

t This

is

tlic

only point in the proof at \\ln.h care

necessary.

390

IKRATIONAL NUMBERS.

[AP.

EXAMPLES.
Irrational Numbers.
1.

(1) If

is

a rational

number lying between a 2 and A-

(a + 1)

2
,

prove that

(2) If

where

a,

p n qn
,

are rational numbers, prove that

and that
Thus
if

p,

-Aqn2 = (a2 -A)

n
.

an approximation to *JA, p n/q n is a closer approximation. [The approximation p s /q s is the same as that used by Dedekind (see
is

Art. 145).]
2.

(1) If a, b, x,

?/

are rational

numbers such that

prove that either rational numbers.

(i)

x = a and

?/

= &,

or

(ii)

<J(l-ab) and ^/(l-.r?/) are


[Math. Trip. 1903.]

(2) If the equations in x, y^

ax + %bxy + cy = 1 = la have only rational solutions, then 2 J[(b-m) -(a-l)(c-n)] and ^(<^
2
1

are both rational.


3.

[Math. Trip. 1899.]

If a

is

irrational

and

a,

b,

c,

are rational (but such that

ad

is

not equal to bc\ then

aa + b and
-are irrational
4.

Any

numbers, except when a = in the former. irrational number a can be expressed in the form

a
where a
is

a
,
,

any assigned positive integer and c 1? c 2 c 3 ... are positive integers less than a. Thus, in the scale of notation to base a, we may write a as a decimal For example, with a = 2, that
in the

is,

binary

scale,

we

find

^2 = 1-01 10101000001....
If %, a 2 a 3 is an infinite sequence of positive integers siuli ... 7 n can be found to make (a^^... a,,) divisible by JV whatever the integer N may be, then any number a ran bo e\|n vssrd in the form
5.
, ,

that

''
!

+...,

cH

<a

1.
1

Iii

ordri

(hat

.-

Th.thii-.

rettliotiOO

th

mu>t
find

hi-

.i

r
L
->~>T^7

~1

t9J-^\ --fl -r 1 -J )
dtln.u^h

'

and
e|ual
6.

so
'

thr

sum

to

intini!

If

we can determine a

d;

such that

linn

^0,
'/..a.
;

"
i

/,

is

thr

int.-i;,.|

to
K.\.

th-'ii

,i.

niu-t

!<

irrational.

Aj.ply

sjM-cial

cast-

of

S)

to

thai \\hi-n

il

an tind an
nun:
[

intr^i-i

\ mcb
M
\v-

thai

.V-

.)/

ii

.,

any assigned
intt-^ral

pl.-as.-.

v.li.-i,-

.I/

j^

tlu-

pait of ^o,

Fr

tin-

first

part.

n..i-

that

'

'

<

For th- sfi-ond part, express a as a conti:


7.

The sums of

!
\\lifr-/-.
'/

t'or

an- any po-hmtin- |i odn< 'II


anil

inti-_rcrs,

aiv irrational

numl.r!
[Ki-

[For simi
p.

-.\t-nsions, ser (Jlaisln-r,

/'

->\.

1."-.

1SU.J
.1

8.

If

i-

th-:

illation

of

!_

-urh that

>
\

~K^'
Thai
if

podtive integera.
h

can

find

s-i|i;.

that

when
oi

/'

ifl

iln-

n.-an-st

integer to

eu/ M

//-

n\bcr

degree

i:

""
wln-i.
.

"'

than

It'.

ly taking

it-ncc

9,[

.in

pio\ r that

Li

392
9.

IRRATIONAL NUMBERS.
Suppose that a
is

[AP.
is

an irrational number which

converted into a

continued fraction

and p n /q n

is

the convergent which precedes the quotient a n

write further

Qn+ i=g n A n + q*-i,


where

A n = an +
sin
|

a n+ i
\

+
a, i+2

. . .

Then shew that


if

man

> Kj Qn +\
n +i;

qn
|

<m<q

and also that where en tends

sin q n air

= sin (ir/Qn+i) = (1 + e n )7r/$, i+


to oc.

i ,

to zero as

n tends

[HARDY, Proc. Lond. Math.

Soc. (2), vol. 3, p. 444.]

Monotonic Sequences.
lies between the two preceding monotonic sequences. two compounded (2) If a sequence of positive numbers (a n ) is monotonic, prove that the sequence (bn ) of its geometric means is also monotonic, where

10.

(1) If in a sequence (an ) each


it is

term

terms, shew that

of

(3) If

<?!,

c2 ,

...,

cp

are real positive numbers, and

if

prove that the sequence


the same
11.
is

(ju,

n +i//^n)

steadily

increases;

and deduce that

true of //,.

If

Sn =

'

where ar
its

is positive and independent of ^, shew that if 2a r is convergent, gives the value of lim n (see Art. 49). Conversely, if lim n exists, shew that 2a r converges, and that its sum

sum

is

Apply
12.

equal to the limit of Sn . to Ex. 12, taking k = l

and

to Ex.

6,

Ch.

XL,

taking & = 2.

Apply Cauchy's theorem (Art. 1 [~n n- 1 n-2


,

152, II.) to prove that


1
,

where C is Euler's constant (Art. Prove also that for all values of

11).

n,

the expression

lies

between

and

1.

[Maf/t. Tri[>. 1!H>7.]


13.

Apply

Stolz's

theorem (Art. 152,


(

II.) to

prove that

if

Inn

+!-

+ A -"J=,
H

where A.>-1,
-

then

+A

-T-.

1.
1

IIPLE8
[If
/'
(]

th.-n

th.-

L'iven

expression can

!"

\\

\\here

/'..

Th. -ii

/-,

increases with

//,

if

1-fA

i>

p..

-in

!L'

-li.-u

in

which we

iven

ret

Quasi-monotonic Sequences.
<iiven a seq'
,

N.-t

tin-

^reateM

index

for

which

are all less than


nil.-:'M
/,
)

m-ti>n of
iiii>niit<iiiii-

is

nut

ami dm--

m.t

t.-inl

t"

inl'ir

tii.n

f"i-

th'-

decrwuing case.

14.
\

it

-I'lasi-moii., toni.iiiuit
.

si-ijin-iin-

OOHV61

wliirli

if

si-(|MMK

itli.-i divf-ri^ to oo or than any trrni nf th- >c|uence. f diverged t" / or converges to a limit

(in.i

great i-r than any

t.-rni <.f \}\<

tln-n tin- s<-ou<'n<--

is

quasi-m
.jua-i-nioiiotonic

ami

in<Tf;i>in'_r
\

s'i|iiciic.'-

,-an ny .on\n-^rin liaving a comnion limit.

IK-

sul i-dividt-d into

two
Q

Illustrations of i|iiasi-iiionotoni<- seqaenoet


4 (-!)"/>,
\vh.-n-.
in
tinfirst

ly

a n = n 2 + (-l)";^,
,nt.

,,

=!

-{2 + (("H \i:i.v.)

two.

Infinite Sets of
15.
irlti<-li

Numbers.
need to

For some
<<>

]>ur|>"-,,^

,t'

analysi-

\\.-

Beqilrli'

't'tcli

i-all>11

Prove that th
countable,

r.-al

nund><

and

[For
-a-h

if

/,,

Of,

'

...

i--

an\

B6C]

write

t.-i'in

as an

intinit.-

decimal
'0*
iff*.
<'..

Tlicn considci- the

infinit--

d-'cimal

7=
in
It
\\

hich
clear th.r
all
/

-lit i,.n

,t"

i<

to tin-

exhaust

the u-al nuinl<-

.nd

1.]

394
16.

IRRATIONAL NUMBERS.

[AP.

Given any infinite set of numbers (&) we can construct a Dedekind by placing in the upper class all rational numbers greater than any number &, and in the lower class all rational numbers less than some
section

number

k.
;

This section defines the upper limit of the set prove that this upper limit has the properties stated on p. 12 for the upper limit of a sequence.

Frame also a corresponding definition for the lower limit of the set k and define both upper and lower limits by using the method of continued
;

bisection (as in Art. 150).


17. The limiting values of an infinite set of numbers consist of numbers A such that an infinity of terms of the set fall between A-c and A + e, however small e may be.

Given an
section

infinite

set

of

numbers
of the

(k)

we can
k,

by placing
but a

in the

upper

class all rational

than

all

finite

number
than an
the

terms
limit

construct a Dedekind numbers which are greater and in the lower class all
of terms
set
;

rational

numbers
limit

less

infinite

number

k.

This section

defines
is

maximum

of the

prove that the

a limiting value of the set, in accordance with the definition given above and further that no limiting value of the set can exceed the maximum limit (compare Art. 5, p. 13). Frame a corre;

maximum

sponding definition for

the

minimum

limit

and

state the analogous properties.

Goursat's

Lemma.

18. Suppose that an interval has the property that round every point of the interval we can mark off a sub-interval such that a certain inequality

Then is satisfied for every point Q of the sub-interval. divide the whole interval into a finite number of parts, such that each part contains at least one point (P) for which the inequality \Q', P} is
denoted by {$, P}

we can

satisfied at

lies. every point Q' of the part in which not if either half does the interval satisfy the condition, original [Bisect If we continue the process, one of two and so on. bisect it again
;

alternatives

must occur

either

we

shall obtain a set of sub-divisions

which

satisfy the condition, or else, however small the divisions may be, there is always at least one part which does not satisfy the condition. in the latter, we have an In the former case, the lemma is proved
;

infinite

sequence of intervals
b H+ i-a n +i

(, &),

each half the preceding, so that

= $(b H -a H ) and

aw

^a

ll+1

<6,1+ i^6 n

and the condition

not satisfied in any interval (a,,, 6,,). Now the sequences (), (&) have a common limit / mark off round / Then we the interval (/, g) within which the inequality {Q, 1} holds.
is
;

have an

= = &n

an ^ we can choose n so that bn - d

lt

<l-f=g-l.
(/,

Thus
tlii-refore

">/+(& -0=/
is
is

and l>n<g + (l-<*)^ff,


g)
;

so that the interval (a n , />)

contained within the interval

and
is.

the inequality {^,

l\

satisfied at all points of (<(, /;).

That

'I
,ii..n
/.

-ati-lied in (/./*,.):

and
'

-o

we

ai.

led

to

.11

l.y

asumii

ue.]*

111.
..t

nat
i

an
-MII

in:

ih-

mark ,ny two

'

point

di\

id.-

(I,

Miml.er of

parts
\\liieli li

mci|iialit.
.a
I

-mis

t.

I'loVe a
[

IS.]

20.

Slate
;

and prove

K\-v
;

Is.

\\>

j,,

to

ie-ion>

in

plane

..r

I..

dOB.

Continuous Functions.
21.
'

:nuoii>

in

the interval

the intei-val,
)

the tipper and lower limits


(Apj.lv the method
In
i<
.

//

and

It)

of /(*) in the interval.


that //

of continued

1'iseetion.

i:et

an

infinite sM|iienec of
in

intervals

\>-\\

the upper limit of


limit
/.

/'(/')

the inte:

<-ommon

Then

if

/(/)<//, choose 8

that
8.

Th-n choose n so
at
all

that
I,

A.,

~a n =&; and we

tind

tl.

W
an
int.

+S( f )}

p"int>
I
i

contrary
it'

to hypothesis.]
in

22.

s to

then

\\i>-

int.-i\al '-an lie


'ii
)

prove that divided into a


that

finite

numliei of

j>art- (the

mn

;idin-

>ni-h

iny
23.
a

two points
said t"
in

in

the same subdivision.


I

INK.]

fun<-tion

i-

b
if
-

it-

absolute vahu-

finite

upper

limit
_'

the interval.
in

that

an

intei \al.

it

finite

in

the interval:

and

a!s..

that

that

l/(f ^
Tin'in-sat
I

f* in

<&
proof is of

form

suitat.l.-

will he
|s,

found
\\)

in

his C<

M-ein to ha\

APPENDIX

II.

DEFINITIONS OF THE LOGARITHMIC FUNCTIONS.


156. In the text
it

AND EXPONENTIAL

has been assumed that

d
and a number of
used.
to
It
is

allied properties of the logarithm have been customary in English books on the Calculus
coefficient of log x from the exor else from the exponential series

deduce the differential


57)
It

ponential limit (Art.


(Art. 58).

would, therefore, seem illogical to assume these of properties logarithms in the earlier part of the theory; although, no doubt, we could have obtained these limits quite
book. But from the point of view seemed more natural to place all special limits after the general theorems on convergence. It is, therefore, desirable to indicate an independent treatment of the logarithmicfunction and it seems desirable to use this way of introducing the function in a first course on the Calculus.*
at the beginning of the

adopted

it

157. Definition of the logarithmic function. There appears to be no real need for the logarithm at the beginning of the Differential Calculus, but we require it in the Integral Calculus as soon as fractions have to be integral' M!.

At

first it is

probably best to denote \dx/x by L(x), and post/,(.')

pone the discussion of the nature of the function


*See a paper by Bradshaw (Annal* of Matkematic*
sunl
(-2),

until

vol.

4,

190,'i,

j>.

.~>l>

Osgood's Lehrbitch der Funktionentheorie, Bd.

1,

pp. 487-500.

157|

in-:i

i\i

[0
ball

the
tin05

present,
a

th-

that

presents
nli

an-a

between
Ai

curve, the;
Of
t!i.-

nates

= a,
n

x=l
below

tln-Mivm

will

lan-ular
shall
tli-

hyp.-i-l.ola

= l/x

be drawn,
l.y

th.-n

w-

denote
!i\<-d

l>y

A(./-i

tin-

area

HTM

hound. -d

th-

ordinair
:

AH
in

(fl5=l),

tin-

axis of

x and

tin-

variable

ordinal^ I'M
(
I

or.

th<-

notation of the Calculus,

/.</>=
as
will
h.-\

1/j/x,
th<-

wht-rr.

id<-nt

from

ti^ur.-.

is

siij.p

It

\B

..hxioiis

fi-

,iii

tin-

drtinition that
/.,!,

().

Furthrr.

if

paralh-i Mill

raun throii-h two rectang

//

and

7'

to

tin-

HTM
Thus

and the Other nitiivly within AHl'M.


w<h:,

/-I
or.

>
of
1

/.(./)>(./-!
nutation.

with

slight

rhan--.-

./>
(3)

At

+.'>>.>

-f.o.
it

has
In-

is

easy to ahew
\\lim
1.

...ld
v.

g,,Kl

///

-/-'

//////.

.<!.
i.J),

must
itive,

t.-ikm
all
tlu-

t"
in-

ni'ti.-.-

tliat

ten than
[iialitii-s
..i

in

or

in

u'-').

the nuiurin-al \alnr.s the inniualit

irv

\vmiUl havr

IM-

r.-\cj-sed.

398

LOGARITHMIC FUNCTION.
(3),

[AI>. N.

For instance, we get from

But

in numerical value
if

<

Z(|)|<1.

Again,

we take an

ordinate

QN, such that


.

ON=2.0M,

we have
L('2x) -L(x) =

PMNQ > rect. MN


<rect.
t

NQ,

and
That
is,

MN.MP.
or

x(^)>L(2a)-L(x)>x(^)
get

Thus,

we

1>

Z(2)>J,

since

and so

on.

It follows

by addition that
it is

Now,
Hence
if

if

X>X
(n +
n
,

evident from the figure that


+1
)

1)> Z(2
and Note
it is

> L(x) > Z(2) > K

2 n+1

>a;>2
1,

with x (Art.
(5)

2),

evident that L(x) tends to infinity or

Again,

if

we

write x =

l/t,

we have
dt

cZa?_

~^~
so that
(6)

"T
=
- ^(
or

Z(^) =

a;c

=-

Hence, as a? approaches zero, since Ifx tends to L(x) tends towards negative infinity, or
(7)

infinity,

limZ(aj)=-oo.
a;->0

Again, the function L(x) is continuous for all povifiw values of x. For we see at once that \L(x + h)-L(x)\ lies between two rectangles,' one of which is equal to \h\/x and
the other to \h\/(x + h).

Thus
if
\

\L(x+h)-L(x)\<c,
\\

Inch proves the continuity of L(x).

157
1

S'lTION
folloWB lYnin
;iii'l

It

tin*

t'mi'lani'-nlai
tliat

between d>

tinn

int'-jiMt inn

.A,-''-'

hut

without

aj.jM'alin:i;it
|

in

this

fact
L(. '))/'
is

WC can

nhtain
1,,-lv.

/.(./

+ //)

rniitain.-d

I/a;

and

l/(cr

A).

Thus
l

cia;

= ii..i'[(+A) T i()].i,
^_,

ire

two

onli

,ch

that

f>

> >
'/
t

\v-

><(6-a
l>y

exactly

tin-

same ar^uin'iit (.'>>. we have

as

\vc

usnl

^
or

Tims
or
Consf(|u<-ntly.
'.NVX;

the functior(

L(x)/(xl)
n.-arly olivioiis
.!/>'
/

as x

which cm-ivspniuls to the that //" mean ordinaic


increases.
i

-///'/

/'.-.

cw

a;

.\aiii])|f, tin-

ifadri

may jn'Vr
g

tin-

la^t

result
<!'

ly

ditlci-fiitiatioii.

Tin-

timuv

ln-lnw

will

n.-i-al

id. -a

tin-

course of

the logarithmic I'unctinn.

41.
'I'lic

dotted

liiu-8

rcprt'si-nt

(lu-

rurves
I

y = (x-l)-i(a-

400
158.

LOGARITHMIC FUNCTION.
Fundamental properties of the logarithmic u f dx

[AP.

II.

function.

In the formula

change the independent variable from x to g by x = g/v; we find then

writing

or,

going back to the definition,

= L(uv)-L(v).
Thus
(1)

L
equation (1)
is
it

From
(2)

follows at once that


n L(u ) = nL(u),

where n

any

rational number.*

Suppose now

that

is

number such that L(e) = I;


is

the

existence of e follows from the fact that L(x) x function, which steadily increases from varies from to oo (see Art. 149).

a continuous

to

+x

as

a?

Then equation
(3)

(2) gives, for rational values of n,

L(e") = n,
to

which proves that L(x) must agree with the logarithm


base
e,

as

ordinarily defined;

we

shall,

therefore,

write for

the future logx in place of L(x). We can obtain an approximation

to

the value of

by

observing that,
writing x =
l

when n

is

positive, (3) of Art. 157 gives, on

+ l/n,

(4)

n/
l

n+1
tends to
1

+ (lV
1
)

as its limit;

and

so,

since the logarithmic function

is

continuous and monotonic,

must tend
*

to

e.

Equation (2) maybe used (see Bradshaw's paper, 4) to establish the existence of roots which are not evident on geometrical grounds for example, th;

Of course, from the point of view adopted in this book, it is rnoiv natural to establish the existence of such roots by using Dedekind's section.
lifth

root.

158|

li

M>

\Ml-.N

\l.

PROPER]

Similarly.

WG

1.

Thus, we lind
lim( 1i

'

".
)

It

is

easy to give a dn
I'-i

hat

the
i

<

a definite

limit
./

in

bi

that (logjr)/(jr-i) decreases aa

increases.

Thuth-ief-'!-i(

-i-e
1

that
tlie

lo_ir(l

it)"

iiHTrast-x

ur
that
(1

i-

/)"

does

so.

In

saint-

\\ay

we
less

~*
1

prov.

I'.ut

MC
(1

that
-2.

than (1-1

/<)-",

and

is

tl.

Thus
limit

//)"<4, and c(nse|iiently

(1

-fl/0" converges to a definite

A-

matlrr
i'r

!'

fact.

In
c<

>\\ <-\

-r,

tlu-su

limits

ar-

imt

v.-rv

Their ^Mnn-tri< .inj.utatinn. Letter approximation: for it will be seen that


nuiiit-rical

II

nice

Tlius

Wi1
(

it

will

I..-

found tint
( V
1

]
-

\
tin-

V = 2-7048,
is

V
:nit

\n -\
in

V =2
I

third of whi-h

only WTOB

the last place.

It i> -alcu{>erhaps .if hist..rical int.-r.-^t to note that N lated the tir^t tal>le <-f h i-ai'it lim^ l.y mean-> an appn -ximate
!'

formnla aili-l

to

i.

nam,/-

La

402

LOGARITHMIC FUNCTION.

[AP.

ii.

The error in the approximation (8) can easily be shewn to be about b[(a,b)/af, and in Napier's work (a b)/a does not so that Napier's approximation is right to the exceed* 5/10 5 13th decimal place.
;

Napier's definition of a logarithm is exactly equivalent to the definite integral which we have employed he supposes that the velocity of a moving point is proportional to the distance
;

from a fixed point 0, and is directed towards 0; so that of the time represents the logarithm of the distance OP, if the initial distance OP is taken as unit.
159. The exponential function. Since the logarithmic function logy steadily increases as y to + oo it follows from Art. 149 that, correincreases from
,

sponding to any assigned real value of x


positive solution of the equation

there

is

a real

y the exponential function when x is the indeand write y = exp x the graph of the pendent function can be obtained by turning over Fig. 41, p. 399 and interchanging x and y. The figure obtained is shewn below
call

We

variable

-2-1

01
FIG. 42.

Dotted curve, y =

It is evident that the exponential


*

function

is single-vain

interval for

Generally, the function inverse to a given function is single-valued in any which the given function steadily increases (or steadily decreases).

158, 159J

I:\I-M\I.\

\i

mot

OCM

value of X in
.dlled
fmicti,,|l
<

'

'

ion

Suppose now
80
tlial
//

that
.

.'-}-/',

Thus.
(1)

fi-MMi

Arl
f

e-have

+*)<*<*/*
+ /).
Mini
if

/<//</</'<//
H-m-,-.
///,

|A|</(yH
see that

|A-|<e, and

>

//<

expont'iit'nil /"i><-fion

ilUOUS.

\Ve have proved in Art. i:,s (3) thai. \vli-n x is iiu*- of e*. -\]Mii-ntial t'unrtii.n i*> thin-atinnal, <lrtmr.l
ra

ro/^'

If

now

a;

is

l>y tli-

u]j>-r and

ln\v.-r

we

= t'.\jM/
' '

< cxp <


./:

'\p

.1

=eA

kuse tin- cxpoin-iitial


.-\j..r
fa

t'unctinn
"

increases with

./.

Al

i-iMtinumis.

can
p.

In-

mad.- a^ small as

we

please;

ami

<-'>i!MM|ii,-ntly

(rompaiv
(e*),
''.
o!'

o7<

expos
is

is

tin-

sinM,-

number
with
"//

d.'tin.-d

hy
,

tin-

cla^s.-s

A
(e
).

Thus rxpr
irrational;
/'x

coincil->

D.-drkind's ilrtinition
/V//,
///i
/

wh-n X
that

and so for
/v////*

.i-y,/,
it

/M

,///'.//

fil.,,rt',u,,

///r

y;ox/'/;/V

/g*.
iV..m

Sine,-

log 1

= 0,

i'..Il..ws

>"=!. and 80 w.

hav,-.

th- continuity of

tin-

exponential function.
Hn,
.r->0
,-->

= ,"=

1.

ii.

because

Logy+logy'log(yy^ we ha\r
l

et+.>

Of course

(8) and (4) agree witli tin- ordinary laws of inaM^ln-d for rational nundtcrs in honks ..n ah'vhra. Ki-.-ni the dctinition^ of the logarithm and exp)iiential functions
1
i

it

follows at once that


/if

</>/

'/

//

Thus
that
is.

flu-

wop

'i

derivuf

'(self,

j
This result can also
1

deduced

at

<>n

404

LOGARITHMIC FUNCTION.
inequalities are often useful:* for any value of x.

[AP.

II.

The following
(6) (7)

e*>l+#,

e*<l/(l-aj), These follow from (3a) of Art. 157.


160.

ifOO<l.

Since,
(1)
it

Some miscellaneous inequalities. when a;> 1, we have from (3)


\OgX<^X
if

of Art. 157,

<05,

follows that
(2)

is

any

positive index,

\vgx Again, from the same


positive,

<xn

or

log x

< x /n.
n

article,

we

see that if

x and n are

x/(n -f x)

< log (1 + x/ri) < x/n.

Thus,
(3)

we

find

'
n->w

Since
function,
(4)

lim=#, and
it

since the

exponential

is

a continuous

follows from (3) that


*

71

+ |)". lim(l >t X


't''

Similarly,

we can prove

that

if

n > x > 0,

_
When n
is

a positive integer,

we have

K)"='+<'('-:>and
since all the terms are positive, this gives,

from

(3),

and consequently, by taking the


(5)

limit as ?i->oo,
if

e*^l+a: + ia

2
,

>

0.

Similarly,

we can prove

that
if

./>().

*The

geometrical meaning of

(6)

is

simply that the exponential cutvo

entirely above

any

of its tangents.

159,160,161]
161.

IMJ.l ALII
limits
;

II.

>

AM*

LIMITS.

Some

the logarithmic scale of infinity.


..!'

\\V have seen in c2)

th.//,

last
if
I

article tliat
1,

<./"
and BO
I

s>

>0,

I.

-IlCe

limllo^.r
x *co

fl

mply
(

>

lii.

<!<>-./

if

//

>0.
is
tli.-

X'-^OO

Since

loga=

log (I/a:),

tin-

last

equation
it'

same as

= 0, liui(.'-"loga:)
(3) of th- last
article

>0.

we
if

SCt-

that

(x/n)<e*>
or
U,

<

,.

>0, /?>0.
,,-aj^
f
)

/j-n^

and therefore
or
1)\

liin(./-'v---

= o,

writing x in place of 2x,


lim(./"v-')

we have
if

= 0,
(.'*)

Of course w
(4)
Tin- results (3)
is

can also write


lim
x ><a
(/*.,

in the
if
//

form

>0.
if

and

course true

//_(); lut no proof

thru

We
Art.

can also obtain


;ilx)ve.

(I

;>

\>y

aii)ealing

to

L'Hospital's

rule,

l.'n'

HUM
and
l'\
<

,^=0,
-hanging
or

to o^

we
?*

get

(1).

Si,,,.:

Inn
JT

0^*
A

"

>0.
power,

g
It

OD

**

>oo

\ve

write

</

m-l

raise the last to the 7ith

we

get

(4).

Th<-

limits

(!) (4)
It

form

tin-

basifl
l^-./-

"*^
tends to
^mall
e

CO/

follows from (1) thai


of
'.

oo
its

more slowly
index

than

any positive power


on.

li\v.-\vr

may
tends

still

more slowly,
e*

ami so

On

th- other hand,

we

see

from (4) that

406
to
oo

LOGARITHMIC FUNCTION.

[AP.

II.

be

faster than any power of x, however large its index may and hence a fortiori e? tends to oo still faster, and so on. Thus we can construct a succession of functions, all tending
;

to oo

say,
. . .

< log (log x) < log x < x < < ef <


ex

. .

and each function tends


the

to

preceding function, of the following function.

but

oo faster than any power of more slowly than any power

It is easy to see, however, that this scale by no means exhausts types of increase to infinity. Thus, for instance, the function
e (loga:)2_ ^loga;

all

tends to
of x.

oo

more slowly than

x
,

but more rapidly than any


a* = e* l **

(fixed)

power

Similarly,

tends to infinity more rapidly than e*, but more slowly than e? or than ef. Other examples will be found at the end of this Appendix (Exs. 11, 12,
p. 412).

162. If

The exponential
write

series.

we

we have

Thus
and
so,

.(l-e-*X n ) = e-*(Xn -Xn _,) = e-*


since 1
e~ x

Xn

is

zero for

cc

= 0, we

have

a result which can also be easily obtained by applying the

method

of repeated integration

by

parts to the integral

i>Multiplying the last equation by


ex ,

we

find

161, 162, 163]

ONEN

II \l.

407
last

I.

But when X and 8O


e
'

in

th<-

integral

i>

losa

than

Iori
\\'li.

"J
(=
say),

=
-'

3.11+1

(^+i?
is

i~

less

than

e*,

and so

ir IJo
w.- have bherel

'
>

l-f-^-fC-r.-.-h'^-rA',.

A'.i

<

r^-r.,

when
wh-n

./

^>

0,

./

<

0.

In

t-i

lini

/.'

<)

(Art.

-2,

Kx.

4),

and so
^

.,.

...

to

163.

The existence
)>

of an area for the rectangular hyperbola.


tin-

\Yc inv.- here a proof thai

ivrt

angular hypi'rl>ola has an


this

area which can


,ual.
arithin.'t
ic

found
l'c\v

l>y

ddinitf limiting process;


curxv.

sincr

lisciission

of

tin-

Ku^lish aiva of a

hooks ^i\- any ad'juate

The method

applies at once to any curve which can he divided into a // numhcr of parts, in each of which the ordinate steadily increases
a

or sti-adily deereasee; although the at-tual proof refers only to curve lik-' the rectangular liyp-rl)ola in wliicli the ordinate
<

constantly

Increases.

408
associate

LOGARITHMIC FUNCTION.

[AP.

II.

with this strip an outer rectangle PMNP' and an inner rectangle Q'MNQ\ now bisect at and draw the l ordinate R^M^ This gives two outer and two inner rectangles outside the curve, R^, QM1 inside. But namely, ly the sum R is less than the outer + 1 1 obviously original is greater than the rectangle; and R 1 1 original inner

MN

PM R^ N PM M+QM

rectangle.
If we again bisect four inner rectangles

MM
;

and M^N, we obtain four outer and l and the sum of the outer rectangles has

again been diminished while the increased by the bisection.

sum

of the inner has been

When
sum

divided into 2 n equal parts, let us denote the of the outer rectangles by Sn and the sum of the inner
is

MN

rectangles

by

sn

Then

S
and
AISO
s

>S >S >...>Sn >..., < < <...< <....


1

sl

82

8n >Sn

(71

= 0,1,2,3,...).
sl
is

Now, from the figure, we see that the difference S1 the sum of the two rectangles PR l} R : Q, which is equal

to

Similarly,

and generally

Sn - sn = J (Sn _ x - sn . x ) =

^(S
and

-s
is

).

It is therefore clear

from Art. 149 that


;

'Sn

sn

approach a
called the

common

limit as

increases

this limit,

say A,

area of the figure

PMNQ.

But it is essential to prove that we find the same area A in whatever way the base is supposed divided to form

MN

the rectangles.

163|
I. .-i

Al;i: L

OF
U

i:\

I.

4<M

i denote
up
<!'

the
n

turn

<>\'

wh.-n
Vt <>r irr.

MN
1.

is

ili\ id.'d

in
i

any
*

nd

l.-t

/r

denote
glanoo

tin-

sum
t

in-

oorresponding inner
will

Th-n
w.-

at

li.-

li-iin-

shew

tliat
l

!'<

S>
wli.
'i'liu

uite

<

independent

"!'

n.

lini

Nn a

.-1

lini

We

1.

2=^,
1
,-)'

<T_ J.

llut

T
>!'

v/M/
tin-

-QjV),
_!

\\hnv
llr:

is

tin-

linsultli

wil-st

.main.-.l

in

the SHIMS

^l. rj.

MII ohOQf

LCb
if
j 1

that

2-(r<e,
ami
tln-r't'. .!-.

S<

since

/!=T. w-

Thus.

Hiii }L

=J =

liin T.

^^o
That
lias,//

^^.0
*<i ,n>

is.

wis

olitain
ilividol.

the

aiva
f/,,,f

.1.
//,/

in
/

whut-\vr way

tin-

.17

A'
:'

!>,<;,/.,/

//'/>

/"

/-a

Extensions of the definition of integration.


If tin-

function

t<>

lr

iutr'_r i"itr.l
s.-.-

i>

ii"t
i':i,

mOOOtonic, hut
::

tinitf

in

tin-

int<-i\al

(./,

A) (f>r (Ictinitinii,
/'
.

K\.

p.
i.f

taking

//,-.

tin-

upj-r ami

l<>\v.-i

limits

tin-

fum-timi

in

th-

int-'i\;il

r,

an.l

Writing
is
it

$ n ^//I

\vlin-.-

y,.

..l.tainril
is

iy

divi.linur (".

ii;

easy to see that limits as i< trmls t> intinit

iiion.t,ni.-,
iict-d

ami
!

BO liavr

limit>
;

n-a

f|ual.

have gone
Supjxisi' that in
\v.-

at

prrst-nt

hut

we m\v

intr<du.-f

//// sMl-iiivi-j,,n

Qto -ul-int<-i-\
n thf iippn-

aU
c>

,j

denoi
// ...

and

b-w.-r
-

liin.-

function in th- intn \al

then

it
i

nn;

il.lr

t. tind

a-

1
ill

//

W r <,
that
//,.

modes

of <livisic.ii

..t'

th.-

int.

leea

thai
luivo
iiulio;it-.l

only

the

rc.-t.ui-;.

latin

hciiiu'

<lttel
an.

th.

for

IT

410
Under

LOGARITHMIC FUNCTION.
this condition

[AP.

we have obviously

so that

= liin Km Sn = lira sn = A

(-)

(),

sa}*.

Then, just as above, we prove that for any mode of division

and by Riemann's
Thus
It
is

condition,

S-o- <e,

if

yj r

<8.

Km2 = J=limcr.
App.
for (see easy to shew that a continuous function is integrable - a), if r <S. Thus we find I.) we can find 8 so that <o r c/(b rj
;

Ex.

22,

<

2i7 r a> r

<,
is

because

^r) r

= b-a.
;

It

two

also easy to extend the definition of integration to a function of let us consider the meaning of variables, say .r, y

///(.r,

y)dxdy,
a' to
6'.

where x ranges from a to


If

6,

we

divide

(a, 6)

into 2 m equal parts


== 2jtl IL, vifj., v i

and y from

and

(a', '&')

into 2 n equal parts,

we

obtain two sums


*^m, n

&m, n

^'^fJ.,

/u,

v>

are the upper and lower limits in a sub-rectangle or n is Then, just as above, we see that Sm n decreases if either 'TV,*. and s m>n have each a limit when thus increased, while s m<n increases ,,,,

where

H^

and A M

wi, T&

Further,

tend to infinity in any manner (see Ch. V., Art. 31). if f(x, y) is continuous we prove as above that

Km S
Now we
/

nlt

= lira s Mi n
\

F, say.

have, from the definition of single integrals,


Cb
I

m and

Km
*oo

lira \ M >-oo

Sm>n )=
/

dx
dy

fb'
\

f(x,

y)dy

Km ( Km SW) B ) =
>

\m^ao

(* ''

'

(* /(a?,

so that these two repeated integrals are each equal to F,


>each other.

and

therefore to

EXAMPLES.
1.

Prove directly from the integral for log# that

,[For
If

we have

log(2) =

J*^,
1

we take

these integrals from

to 1J,

from \\ to

1J, etc.,

we

find that

Il.'l

111 W

2.

t.-h

th.-ii

gra]

3.

If

>i, t

-*l as

tends to

prove tint
'.

Dcdll.T

th;,t
l

...+o
\\hi.'h
(

,f

tin-

t\v.

<>na
1>r

(*'

'

l.-j.-n

ithin-

ami

ii"t.-

that

181.]
5.

If
if

/'

is

nmnrrirallv

l-ss

than a lixed number A,

iiuU'|H-i.

and

.og

(l

<

then
1 log( \

lim.r

+ i + _L_
/'
//

l"u

/'

/'

/'

log?i

=
Art
6.
1-2

(1).]
t.

Tsc the

last

t-xain.!!-

>h.-\v

that

if

'|
.

I-

tormi

1" converges
,

if

/x>l,and

..thorwisr diverges.

iiat

tin-

_] )-<"+!)
is

div.-i^riit.
7.

Compare
".

Art
that .-

IS,

If

>

pfov
'2,

-(l+./-)!lg(l

+
1906.]

[Write log (!+./) =


8.

and

UM

th.-

'-e'f
to

> 2^

if

is

Provt- that

as

rank's fr..m
1

-1

x,

tlu-

funrti..ii

_1
-

log(l-fj-)
iitinuous

and steadily decreases

fr'in

to 0.

412
[From the
continuity
is

LOGARITHMIC FUNCTION.
;

[AP.

dislast example, we see that the derivate is negative only possible at # = 0, and when \x\ is small we find that

_J_ -l_l-.a
log(l-Kr)
the series converging
9.
if

12"*"

|.r|<l.]

Shew how

to determine

so that
,

e*>Mx y

\tx>X,

where M, are any assigned large numbers. [We have to make


which
(since

x > log M+ Nlog x>


satisfied

log x < &Jx) can be

But

as a rule these determinations of

by taking #>21ogJ/' andlG.V-.


are unnecessarily large.]

10.

The logarithmic function logx


161.]

is

not a rational function of x.

[Apply Art.
11.

by means

when

is the largest number which can be expressed algebraically of three 9's ? Estimate the number of digits in this number written in the ordinary system of numeration.

What

12. Arrange the following functions in the order which they tend to infinity with x
:

of the rapidity with

x*,

# l0 **,
the

(log

a?)*,

(logo?)'

10 **) 8
,

(logtf)

10 * 10

10

**,

(log log .r)^

**.

Indicate

position

of

each

of

these

functions

in

between

the

members
13.

of the standard logarithmic scale.

If

we assume
to be

the binomial series for any integral exponent, and

suppose n

an integer greater than

x
| \

we

find

Deduce

that,

if

is

positive,

and so obtain the exponential


14.

series.

Shew

that

(-1W
i

/
i

1
2
v,

1\*
a/ ill
get

(nl)

[If the

product on the

left is called

we
a?
,

^- V = -^- =
Taking
v

dv

e*-l

+ 2! + 3! +
,

= 0,

= 'Za nxn /nl, we

get at once, since o

ii.
I

LMPLE8

vre find

tin-

idmtit

whi.-h
15.

i-

rly.J

Sh.-u

that.

.iH

*-*<),

G^-r!
16.
If

v
-'""

-,

=
(l
' .

11).

|.|-..\i-

<x
[It

<

'-

will

be seen that

Xwhirli give*
17.
tin-

X-J

_ [

_Mc_
-.'

\
}_

0/4 ( W

result.]

good

a|)ii-.'.\iin:iti.n

t. the

fuii<-ti.. n
1

..f

K\.

16

l>y

taking

the

ermr

in

wliieli

is

<.f

tlie
.

'il.-i
1

(I""

[A.

L"i><;K.]

[Aj.j.ly
18.

Kuler's

:)).]

I'I-MV.

that

(/<

):

I=^Y^_ J !\fl
.

n-f2

T 8!

[Dill'eiviitiutr,

and

l...tli

-id.- irdu-,- t.

19.

Shew

that the

Mqn<
= e-,
3

= e'\

<-

Ifl

t<

infinity
tliat

nmrr rapidly than any


tl

ni.-nil.er of

tli-

ial

scale.

20.

I'mve

2(loj

convergee

it'

-i

or

it"

ami

and

<.tli-i\\i>r

APPENDIX

III.

SOME THEOREMS ON INFINITE INTEGRALS AND


GAMMA-FUNCTIONS.
164.

Infinite integrals: definitions.

range is infinite or the subject of integration tends to infinity at some point of the range, an integral maybe conveniently called infinite* as differing from an ordinary
If either the

integral
differs

very much in the same way as an infinite series from a finite series. In the case of an infinite integral, the method commonly used to establish the existence of a finite integral will not apply,

as will be seen

We

if we attempt to modify the proof must accordingly frame a new definition


:

..

Secondly, if the integrand tends to infinity at either limit

ff
f(x)dx when this limit
a
a

of Art. 163.

f(x)dx

{00

exists.

fb f(x)dx
as equal to the limit
lim
6->0 J (t+8
[

f(x)dx

(S

> 0)

when
*

this limit exists.

Following German writers (who use uneiyentlich), some English authors have used the adjective improper to distinguish such integrals as we propose to call The term used here was introduced by Hardy (Proc. Lond. Math. ^ infinite.
(ser.
1),
is

vol.

which

34, p. 16, footnote), and has .several advantages, not the least of the implied analogy with the theory of injinite series.

[164]
If

DEF1NITIO]
tinoi'

int'"_;!-an'l

tendfl

t"

intii.

point
li\il.- ral
i

within

tin-

ran-.-

in

to
tli-

i^ral

into fcwo,

ami
I

.-uM
liml
i-

il-tim- tin- int

ly

liml
,-"'.
il,,.
t

>

But
a
tin/

in

n-rtain proLL-niv
intinitf.
liin
\vliil.|

\\

limit^ in tintli-

la*-:

ln.th
tinitr

two
limit

int.'-raU tends to
:

tinit-

\v.-

then

r\ /,-)^=ii.It

' :

-i-

['/(./) //*].
bend
tinu^..!'

is

at

one,-

.-viilmt

tliat
aii-1

tin-

temuj
definitiona
Exs.
1

OA

apply t> these

>

%-

ftr

^=

liiM

IA

;='

JO

A-.
,

n/.
/"fr

"/./

di-

,^

"
I

-T+lim
pi
!

K)-'-5

i-*o 2

n J-a X
=
\\hfic
in
tinla>t
i\\..

5_o\.

i("
int.

(s
>u])jisf
/

and

/<

shnuhl

ik.-d

that

in

tin-

la-t

CMQ ^- ihoold
8

ml to a
Maik,
<

tlffuiiu-

limit

BI

doea

so.

1.

1.

j..

-.Ml.

416
Exs.
(of

INFINITE INTEGEALS.
divergence and oscillation).

[AP. in.

dx
diverges
l

and

/*
I

sm.vdx
.

oscillates,

dx x

,.

diverges

and

Jo x*

am

\xj

oscillates.

It

must not be supposed that the two types

of

infinite

integrals are fundamentally different.

An

infinite integral of

one type can always be transformed so as to belong to the


other type; thus, if /(#)-> oo as x->b, but where in the interval (a, b), we can write
.

is

continuous

else-

=7 b

a x

or

a-\-b -

and the integrand


1

in

is

everywhere

finite.'

Ex

f
Jo

d-^

-*-/2 )

-^

reversing this transformation it may happen that an integral to oo can be expressed as a finite integral.

By

rao

Ex.
if

When

= !/,

r\

ar'd-x
still

becomes
J^
if

%'- d%,

which

is

a finite integral

5^2

(both integrals

converge

2>s>l).

It is also possible in many cases to express a convergent infinite integral of the second type as a finite integral by a change of variable. Thus we have

o(l

x 2 y*
latter integral is finite if f(x) is (0, 1) (for definition see Ex. 23, p. 395).

by writing x = sin
finite in

0,

and the

the interval

Kronecker in his lectures on a transformation is always


theoretically true,
*

definite integrals states that such

it

is

possible, but although this is not effectively practicable! in all cases.

of f(x)

Care must be taken in applying this kind of transformation when the infinity is inxide the range of integration. Here it is usually safer to divide

the integral into two, as already explained.


t If

/()-*

oo

as

-a,

we can

write
;

J'(x)dx = iin the


</;.

and introduce

as a

new

Jm

variable.

Similarly in other cases

and
/

same sense we can always

express a divergent integral in the form

164,165]
165.

m-:n \rrioNa

Special case of monotonic functions. Although. as we have pointed out in tin- last
i

art

definition of a definite integral iv-jinn^ in gen.-ral a


for
ilif

case of

an
,,f

infinite

integral.

can

obtain

direet

definition

tin-

as tin- limit of a sum,

when

integrand steadily increases or steadily 'Increases.

f(x)dx
steadily
tli.-n

tin-

function f(x)
c;

lrrivasr> to o for values of

x greater than

w-

cMnsi.lri-

only
i'alls

tin-

iiit.-^ral
j

/
in-

'Because the integral


rul.-s.
'I'ln-n
let
/:
:

un.l.-r

tin-

ordinary
fcluefl

.'',,(='),

be a seju-!
1.

to

liave,

(*+,-'
Thus,
ii'

>'
I

/ -'

>

Jx*
tin-

integral

/<.')</./

to

the value

/.

:,)^I^"
<

-'

)f

th>f

two
tin-

MM-ii-s

in

(I),

tin-

srconil

vii-tut-

COnVi

the

certainly converges, in integral and the fact


"

the

contains
it'

only
-J-"

j>ositi\e

tei-ms.
< '

The
i--

tii-st

need

not

conver-v,

tlm rate of increase of


./

instance, \\ith

and

/()=]
l.e

./-.

it

sufficiently rapid: will l>e found that

every

t'-nn
-\-i-r.

in
l.y

ti

Hw.

taking Xn to
l>

a
,<

}.i-operly
.

chosen function of
easily en^un- the
|.n>vr that
i^

some parameter
Convergence
the
t

(as

well as of
io

we can
.

of

lli
have
a

(1); and we can also


limit

\\

common
this

made
he

to

tend t"
to
/.

ZrO ly varying//:
Miple.
si

common
(1).

limit

must

in

virtue of the int-.|ualiti'^

a^

is

independent of

//

and

eijiial

to A, say:

then

.r n

= r+nlt and we have


>

-A[/(c)+/(+A)4
i

+ ...].

,-,

418

INFINITE INTEGRALS.

[AP.

III.

It follows that the difference

so that both are convergent,

between the two sums is and their difference tends

hf(c),

to

with h

hence

q,

In like manner, if xn+1 /xn is independent of = cq n we have say, so that xn


,

n and

equal to

and

Thus we can again infer the convergence of the from that of the second, and we see that

first

series

/-co

Ex.

1.

Consider

xe~ x dx, with


h X)

xn =nh.

We

have then

7=

a value which can be verified by integration by parts.


x~*dx,

/oo

(where s>l).

Here write xn =cq

n
,

and we get

by applying one
Ex.
r
/

of the fundamental limits of the differential calculus.

3.

It can

be proved
oo

if f(x) steadily decreases to


sin

as

by rather more elaborate reasoning that x tends to oo then


,

r<*>
I

<t

xf(x) dx= lim h ^f(nh} sin nh,

cos xf(x)dx =

liiu

h ^f(nh) cos nh.

Let us consider the simple example


r ),

sin

**

the

sum

is

then

//

+ (sin h + $ sin 2A + ...)

165
1

DEFINITIO
,
I

-H'.i

Sin,
is

I,---,

than
tinv.

-27T),

tic-

UTT

M
givei
tin-

limit

Ar
l.y

th.it
<

tlii-

i^ivi-

th--

correct
I.

the
I

i.il

-an

be verified

Art.

Ex.

4.

The reader may

verify in th- SHIM-

\\.-iy

that

Jo

-?
./'
|
-

Ex.

5.

liy

means

..f

tin-

int.-^i;il
<

''/r,

w- ran
"./-'

|r-.vi-

that

lim/, v

v-o

(<--*

+ 2*-

e- * + 3*-'e-

;i

+ ...) =
Art.
-<>.

r
.It

whi.-li

has already been found


if
/'(./)

in

"1

ly

an-.th.-i

inrtho.!.

In

lik-

iiiaiiin-i-.

oe

flB

Inu

st-alilv decreases
ft \vli.-n

as

a;

varies
erges,
I

fr.in

to

6,

we can

ji-<vt:

that

f(x)dx

we have
i

f(

JO

Ex.

6.

Take [\ogrdx', we have lim l>(\ -

t..

find

= 6 (lo- />)-/>,
av.

int-'^i.tt inn.

In

tin-

piwimis
i-anu'.

\\-Mi-k

\\.>

have SMMI

how

to

evaluate an

infinite
\vln-n

intr^Ta!

the

ly calculating the limit of an infinite 9BTJ is tinit-- WG can also "l.tain the result as the
series;

limit
l.y

of n /in

iff

that

is.

\\e

can replace a

loulle limit

single limit.

(See ah
tlie
<

Thus, suppose that in


i

nt

inte--ral

f(x)d&
1

tlie

nt -^ra
-

nl
t<>

'

an<l

that

/'(./)

steadily

from "
to that

/'.

Then
II

\\i-it.

>>

>i

.- ////.

and an argument similar


i/./-

of Art.

will

shew

that

lies

liet\\-en

the

l<

{/("

-f

/'

)+/> + 1'A )+ ... +/</>-/'

)]

and

420

INFINITE INTEGEALS.
as
;

[AP.

III.

7^0, the integral tends to a definite limit and the between the two sums is Ji{f(a + h) /(&)], which tends to zero with h in virtue of the monotonic property of f(x) (see pp. 423, 424 below). That is,
difference

Now,

P
Ja

f(x)dx = limh[f(a+h)+f(a+2k)+...+f(b)],
7t->0

which gives the value of the integral as a single


ft

limit.

Ex.

7.

Consider

J^x-'dx,

where
to find

0<s<l.

Write A = -, and we have

(by Ex.

1,

Art. 152, above).

Ex.

8.

In the same way

\Qgxdx
*

is

found as
Art. 154).

liml 2log(-) = limlog( 6

YT =-1

(Ex.

],

Ex. 9. If we divide the last equation of Art. 69 by sin tend to zero, we find, if a = 7r/ n,
/

0,

and

let

n = 2"- 1 sin a sin 2a

. . .

sin (n
;

- 1 ) a. we
get, pairing the terms,

Now

change from n to 2n and write h for a

Thus, extracting the square root,


sin A sin 2A... sin
(?i

-!)/$

A
+

-",

(if

A = 7r/2w),

and from
For

this

we can

find

"log sin #.

this integral is equal to

lim h [log sin h + log sin


M)

(2/<)

. . .

+ log sin (nh)]


2]

=
Ji!." 27i

f* log
2.

n ~( n ~

x ) io

=166.

TT

log

Tests

of convergence

for

infinite

integrals

with a

positive integrand. If the function f(x)

is positive,

at least for sufficiently large


rA

values of
increases

x,

it

is

clear

that the integral

f(x)dx steadily
lor

with X;

thus in virtue of the monotonic test

165, 166]
t

integral

;,,/(

.1

\.

In

practic.-

tin-

usual

ni.-t!i."l
<

!'

applying
is

thlfl

ted

nppi-al to tin!'

positive

principle of terms; in i'a-t.

in the case of

WM

ii'

positive function also


[

which

't(x)dx COIIN.T^,--,

tlii-n

/(./)</./

converges
than

if

any
tixrd
nuiiili'-!

rat--

for

valu.-s

of

iC

M>me

(Le
I

<[
is

ami

tliis

la.st

rxpn-ssion

in<l-j.'nl.-nt

of \.

Thus, suppose
[">;'

we ooond
;

'.\\hereaispoArt.
'

_Mii\.\v.-

fr.iin
'

I'll

:il'.v-. \v.- lind that

./^e'l^-^O a JT-^OO,

>n that

eu

I'-t.-riniiH'

A-**
iind tlion
i-=.r^e"
-,
1

<l,

if
if

./->c,

ar

<e~i a
/

-',

r>c.
is

1.

".

Ait. i;i)

e~^*dx

convergent, and conse-

/
If

.^e~

in write A'=e*, we

fin.l

tliat
l

\-fA'-(
th
(a
*

\t.\'.

n '/./-

i<

.-(.nvcr^nit.

be niultijilifd
iiinic

tc.

;u

l.y

th.-

;ii<l

of

scale of

inlii,:

Timfluif

",/

a <ni>/ a

rU

Ii

one

"/'

///

'l
/
fi'dl
J

,<>o

->,

J(
I'm<l

Tin- coinparisoii
h'

t.-st

I'un-

as follov
iln-n

G
is
j

Iwaya

.:nl
]

<;<

80 also

at

any

rate

at't.-r

crrtain

valur

422

INFINITE INTEGRALS.
(see the small

[AP.
if

III.

We have proved found so that

type above) that

is

positive, c can

be

whatever the index

/?

may
Jc
/

be.

Now
and

this expression tends to oo

with

A, so that the integral to oo is divergent.

Thus
I

y&e^dx
it is

also diverges.

If

a=0,

By changing

the variable,

easily seen that this integral diverges if we deduce as before that

fi=

I.

diverges under the same conditions.

Joe a

f(x)dx diverges if we can find

an index

a^iO, such that one of the conditions - a) (1 a (i) f(x) J5(loga;y a-

>

(ii)

/()>

4
J

>

BxPer",

or a

= 0, /3^-l,

is satisfied.

These conditions are analogous to those of Art. 11 for testing the convergence of a series of positive terms; and, as there remarked, closer tests can be obtained by making use of other terms in the logarithmic scale (although such conditions are not of importance for our present purpose). But one striking
feature presents itself in the theory of infinite integrals which has no counterpart in the theory of series. integral

An

f(x)dx
limit

may

converge even though f(x) does not tend to the


Naturally,

zero.

we must then have an


is

function, for lim /(&)

()

obviously necessary in
.

all

oscillatory cases of

convergence; but

general way, that this

To see, in a we may even have lim/(#) = oo is possible, we may use a graphical

method.

1
l-i...

IS,

Consider a curve which has an infinite series of

/>'/,

x,

of

to suppose steadily increasing height; then, it is quite possible that their widths are correspondingly decreased in such a AV.-IV

166
1

TESTS OF
the
(|ii,

'

ONI

Bl

that

areas of the peak-

t'.nn

oonvergeni

series;

ntly
i

f(x)dx may con


in

pai ti'-ular tin- fun.-tKui

-r* (l+.r-sin 2 .r),


1
i

(a>/3>0).
.c
ft

~ a
,

itli

It'll

its

Xiaj.l
..f

to

carve // i^, at every point In tin- intnval fi<


tin-

fr
.

\\hi<-h

./

is

.-i

multiple

r,

we have

(n+1)

thai

f '
this
it

is

evident that
|

/'(./)</./

.'..iivrrL't-s

.,r

diverges with the

';th;r

/> 2(/?+l)
</>(#),

or
to

a^2(/?+l).
oc

And

^eiu-rally,

if

( A% )

steadily

with

*, tin-

integral

r
J
l

44>

+v
if

^rs
diverges.

if l\/i(/< -f 1)

[VK*)P converges and divrr^-s

-00'

Ex.

/"*

o^dor

1.
-/0

converges

"i

divergee according as

l+j<r\*inx\

a>(j+l
Ex.
With
2.

or

a^j8+l.

[HAUI.Y.*]

/"</,(

.nrergM with

-^|^^]

and diverges
[IIAK..Y.]

-''
irff)

Ex.
with
I'

3.

/"<(*)-Jo
M/
'

ih

1
^

jfefi

diverges

()

:r
.

[1M-

Hois RBTMOSI..]

111

spit.-

!'

tin-

last

rrsuh,
''I'/

\v,

can
tl>

i'1-ov.-

(a.s

in

Art.
i

86) that

decrease*,
<jence of
I

l)

Accessary for

t/<>

.Messenger of Math>

mm iX

April

1JX-.

111.

424
For here we have

INFINITE INTEGRALS.

[AP.

III.

is less

thus for convergence it is necessary to be able to find A so that than e for any value of p greater than A.

(/x

- A)/(/x)

Hence lim #/(#)=() is necessary for convergence. But even so, no such condition as lim {x log x)f(x) = is necessary in general (compare Art. 9); but it is easy to shew that if (for instance) xf(x) is monotonic, then
x\ogxf(x) must tend to 0. More generally, if <f>(x) tends steadily to x and f(x}l<f>(x) is monotonic, then f(x)<j>(x)l<f>(x') must tend to zero thismay be proved by changing the variable from x to <(#). [PRINGSHEIM.]
;

It is perfectly easy to as to apply to integrals in

* so modify all the foregoing work which the integrand tends to infinity,

say at x = 0.

The
than

results are

The integral

Jo

f(x)dx converges (if b

is less

1),

provided that we can satisfy one of the conditions

where either

(i)

a>0
hand,

or

(ii)

a=

0,

/3<

1.

On

the other

the integral diverges

when

where
167.

(i)

a<0

or

(ii)

(),

/3^-l. we
consider

Examples.
illustrate the last article,

To

two simple

cases.

It is easy to see that the integral converges, so far as concerns the upper limit, by applying the tests of the last article. There is an apparent difficulty at the lower limit,

because of the factor

I/a);

but since

the difficulty
*

is

apparent only.
results directly by writing
l/.r

Or we may obtain the

for

.r

in the integral.

166, 167

425
f
'

tin-

intt-'n-al

i.s

liui

(<

-<

l-*J

dx x
X

nid

"
J
l>y
ri
\

J, 5

{"

the

Benoe

nr

int.-

Inn

* <-*

But
l>y

CM
I

e~*

lirs

between
"

tin-

\alu.-s

i'..unl

1\-

r-j,i,i

'

and ly

'"'
:

tli->,.

valn-s an-

g-'lno-/

and

both
H.

ul'

wliicli

tend to lo^
liin

as o kendfl to ^- z

0.

=
=

1(.

->oJ

and aceordin.
f

(e~

e~ tx )

1.

iider
c

(Ae-

J*<-

^ + (v--

///*
,

/.

//,

are positive and

A
It

may

sh.-wn as above

tliat

tin-

integral converges

cnnditions aiv

NOW
fi

J3
!'

VJl

OJ*

-rJ6
,-\

In \irtu--

the r(,nditi<'n

1J./=O,

it

is

now

idrnt

that

B
f
I (l

*
'

:=
,1.1.
|

Thu>
and BO

= "'J 2(^
I
l

.;

1.1-

Jo

426
3.

INFINITE INTEGRALS.
The reader can prove
similarly that,
if

[A P.

III.

p > n,

where

2^A ra r = 0.
L
'

(k = Q,

1, 2,

...,

1)

168. Analogue of Abel's Lemma. If the function f(x) steadily decreases, but is always positive, in an interval (a, b), and if \<f>(x)\ is less than a fixed number A in the interval, then *
}

f(x)<p(x)dx<Hf(a),
i

where H, h are

the

upper and lower limits of


r*

the integral

as

ranges from a

to

b.

For, assuming that f(x)

is

differentiate,

we have

J=
Now,

we
in

since /(&) is positive and f'(x) is everywhere negative, obtain a value greater than J by replacing x(&) an d x^) the last expression by H, and a value less than J by
h.

replacing

them by Thus we find

f(x)dx <J<Hf(b)-H\ f(x)dx -h\ Ja Ja


or
Similarly, if interval (a, c)

hf(a)<J<Hf(a).

Hv /^ are and H h
2,

the limits of the integral x() in the interval (c, b), we find 2

n the

- H, ( f(x)da> - H2
Ja

[f(<e)da> Jc

or
*If f(x) should be discontinuous at x = a, /(a) denotes the limit of/(#) as r tends to a through larger values.

167, 168]
\\

AN

\l.(M,i

i;

OF

Mil

IMA
is

li

that

if

y
iiinii

nunilM-r.

unction of a real variable, it n-al <T minpl.-x. air,

an- th- upjM-r

as

ranges from

,/

to

and from

to

>>.

ivsjM-rtm-ly. th.-n

|7
\Vln-u
/'./)
is

Uf()

<

/yj

/'('')-/('>

+ >).>f(c).
fco

complex,
p.

formula.ol>tainr<l

ronvs|,oiilin(gee results
/'

of
1.

Art.
i,

80 can be
()":

li>07,

hut

th.-sr

an-

not

u.-.-j.-.l

foi-

oiu-

j)i-f>i-nt

p
on
p.
i^<;
is

<t

ineijuality

iMjui\'alent
first
Itet

to

the
is

Of

Mean
-I.

Value.
p.
:V.t~>
i

To
,i--i;n

see thi>,

note
\alue

that

\(^>

ri.htiniK.us.

Second Theorem and 80


in

inter\al (o,

''o.

Thus the

ine,,uality

leads

ween A, // at least oner t<. Bonnet's theorem


'

the

^/('Oxtfo)' ul "' iv
i

= =
,

''

this

du Bois Reymond's theorem, which


liy

i>

true f..ran\

monotonic

writing

!/('')- ff(b)

f"
\

^hu-

we

find

the

t'oini

comnK.nly quoted

V
BOM
value of
,,

<}>(

>(l>)l^

cannot

le determined,

Mee

equa

Although the restriction


importance here.
as th
\.-t
it

t!

l.e
.-t

ditl'erential.le

i>

of

little

i-

h---i

ic.illy

de.-iral.le

to establish

-udi results

\Ve shall tln-ivfore ni; with the greatest generality possible. a second proof, based on one due to Pringsheim*, in whuh we me nothing aln.ut the existei
in;
:

Divide the interxal


.

|]

part-

l.y

inserting

j...

and wri

then we ha\e
=

where

'<K),

p.

'209;

this

paper contains a

inure general form of th.

equality on
vol.

j>.

4-t'..

Another
i>.

}>roof

baa been given

i.\

H.ir.iy,

MtMtny

.SO,

HHMi,

1".

428
Hence,*
(1)
if

INFINITE INTEGRALS.

[AP.

III.

/r+1 =/(.rr+]
Jr-r

),

but in virtue of the decreasing property of f(x\ [/(#)-/r +i] the last integral and is less than (fr fr+i), so that
(2)

is

positive in

<(fr-fr+i)A(b-a)/n,
<j>(x)\<A and xr +i-xr =(b-a)ln. the equations (1), bearing in mind the inequality
n _l
rxr+l

because
see that
(3)

By adding up

(2),

we

J-Zff+il J XT r=0

$(x)dx = Rn

where
because
If

Rn < 4 (6 \

)(/

-/) < ~ (b -

)/

fn =f(b)

is

positive.

now we apply

Abel's

Lemma
r=0

(Art. 23) to the


r+l

sum

we

obtain the limits

hf-

and
1

.fi^ for

it,

because
f
x

T^r+l

and the sequence /i, /2 ...,/ Thus, from (3), we find


,

is

decreasing.

(4)

A/i-^-)/P<^<^/i+^-)/f
/! =/[

where
If

+ (5 -

)/]

now we

take the limit of (4) as n tends to infinity,

we

obtain the

desired result, t

In exactly the same way we can make the further inference that if
lies

between a,

b,

and
to

if
c,

H^

A x are the upper and lower limits of I


,

^ (.r )././

as

ranges from a

while ZT2

h 2 are those as f ranges from

c to 6, then

In case

/(a;)

should be discontinuous at xr +\,

we

define

fr +\
;

as the limit

of f(x) when x approaches ar r+ i through smaller values of exist in virtue of the monotonic property of /(.<)

this limit will

fit
that
/*
,
-.

will be seen that the condition

|0(#)|<-4

is

by no means
t't>

essential,

and

it
;

may
for

be broken at an infinity of points, provided that

\<j>(x)\dx con-

verges

we can then make a

division into sub-intervals, for each of whirh

than any assigned number. But Pringsheiin has proved / that it is only necessary to assume that (f>(x) and f(x) x (x) are integrablc in the interval (a, l>) compare Proc. Loud. Math. Soc. vol. 6, 1907, p. tii?.
\<f>(x)\dx is less
;

168, 169

429
1

169.

Tests of convergence in general.


in- the genera]
be
I

AppK

<,////
fral
J

and
-!n,t

for

tl

Convergence

<>f

tin-

/i

we oam find

nieh

j
T
'//.
f

/in

"//'/ e 18 ni'ftifi'"

HoW( vet
which can
tin- t'ollo\\
1.
iii

'
.

I'm-

inlinit.

.--ral

test for con-

usually r-|laci-l
l>c

in

iff j.i'act

a]>plic<l

more

<|uickly.

ly BOm The thn-.-

DMTO
chief
t'-sts

are

Absolute convergence.*
i

The

literal
.'

will

certainly converge

it'

I/O/

'

converges,

.<'

<!>'
I

Jf
i'.iit

naturally
in

tin-

analogy h.-twccn Midi


is

intc^i-als

lut.-ly
/

<|uitc convergeni aeriea the values of a function.


if

not

c(,ni])l-tr.

since

th.-r.-

and absois no

In particular,

|/(#)|<

//'(.').

wli--

Readily decreases
will coii\>

:icivascs. tin- integral


J

/'(')'/./

2.

Abel's test.
Infinite

integral wh
'

''hinnjii

m,t absolutely)

and
Suppose
tonic
\lr(.i'}

//N>-

>e).
/)'/./

that
J

,/,(.

;id

that
i>

inonotit

function, such

that

*fr(x)

<.l:

it

th-

that

tt-nds t. sonic limit


if
if
i

Thus, i-/
*

we

wri
-.-.

thai

tnd

'I'lit

ahsohitil.y B(
-.

is

clearly

pointed out

ami

/'A

430
decreases to
;

INFINITE INTEGRALS.
and
it is

[AP.

III.

obviously sufficient

to

prove the

Joo a

f(x)<j>(x)dx.

Now,

by the analogue of Abel's


IJ1

Lemma

(Art. 168),

we

see that

< Hf(() < Hf(a),


<j>(x)dx

where

is

the upper limit to x

when
of
I

X ranges from
<j)(x)dx,

to

'.

Now,

in virtue of the convergence


so as to

we can determine
f/<

make

and then

so that the integral

f(x)<f)(x)dx converges.

Hence
3.

also

f
I

Ja

\}r(x)<t>(x)dx converges.

Dirichlet's test.

infinite integral which oscillates finitely becomes convergent after the insertion of a monotonic factor which tends to zero as a limit.

An

Here again we have


I

f^'
J

f(x)$(x)dx

< Hf(),
constant independent of ^; * so that e, and con-

and

H will be less than some fixed


we can
/QO
I

thus, since f(x)-> 0,

find

#/()<

sequently the integral


Although the

Ja

f(x)(h(x)dx

is

convergent.

tests (2), (3) are

almost immediately suggested by the tests

of Arts. 19, 20, yet it is not clear that they were ever given, in a complete form, until recently. Stokes (in 1847) was certainly aware of the theorem (3) in the case <(#) = sin .r (Math, and Phys. Papers, vol. 1, p. 275), but he makes no reference to any extension, nor does he indicate his method of

The first general statements and proofs seem to be due to Hardy proof. his argument is somewhat (Messenger of Maths., vol. 30, 1901, p. 187) diflerent from the foregoing, and is on the lines of the following treatment of the special case <(.?;) = sin ./.
;

Since the integral

<f>(x)dx oscillates finitely it


;

remains

less

than some fixed

number G

for all values of

thus

we have

169
1

\\i.
.

MI;K

III.L;

In tin-

i>.

--sci

Hates between the two


the figure.

i.ly

in

almost

intuitively

-\

id.-nt

that

tin-

aivas

.,f

tin-

\\aves steadily

decrease in value, and have


r

alt-

in

and

.sim-e

-in
;

positive in tin- integral,


it

tliis

lies
s

U-t
oo
.

fl

TT)

so that

tends to zero as

to

Further,

vdx--Jf(x+**+lir)KD
It

which

i>

obviously negative and nuinerically less than the area of the


wave.
follows that
f*
/

luiis

sin

is

c.n\ t-rgent,

by applying

th- th.-MUMii of Art. 21.

In general,

if

<f>(.r)

method.

usiiiL'

i>iriehlet'>

changes sign infinitely often we can apply a similar t->t (An. ^>) to establisli th.- ronvergence of

If

tin-

intro-ranil

tends
tin-

JO,
.!'

lor

convergence and
tin'/

tr-.t

-ay as .r-*a, the general t-M. als(lut- convergence run as


/'"/

Th>

'!/

8UJj
././

flu

<'<>n>; r< jence


><!,

is

fl.-

fix

if

Ib

ivherc

///.-

any

;>ox//?>v

*/
I

|/

,ii

/'.>

/A.

'//.

It

is

i>M>vil,l,.

to

write

nut
l>ut

con-.^jioinliii^-

nioliticat inns
ci

of

Altai's

in

and Dirichlet's tests ; practice and an- (oetl

such

t.-sts

are not

to tin-

ingenuity

<>!'

tlu

reader.

432
Ex.
1.
/
!

INFINITE INTEGRALS.
__ctr,
1(
/

[AP.

III.

.-dx

converge absolutely

if

p>

and so does

( Jo
/oo

^<foif 0<<1
x^
/*oo

because |sin#|^l, |cos#|^l.

Ex.

2.

*-^.(lx,

-^^ d# con verge


</>(#)

if

0<p<l
cos tfcfo?

and generally

sin .rcr,
test,

</>(#)

converge in virtue of Dirichlet's


I

if
I

<j>(x)

tends steadily to zero.


I

Cb

For
I

7a
3.

sin x dx

cos

cos b

fb
/

2,
|

cos x

dx

sin b

- sin
|

2.

Ja

Ex.

Further examples of Dirichlet's


/(*) =
-"*, (logj?)-*
;

test are given

by
[HARDY,
I.e.]

= (sina?)-*, <^(d7)

log(4cos%).

170. Frullani's integrals.

As a simple and

interesting example

of the

tests

of the

last article, let us consider the value of

x
pco

where

<f>(x) is

such that

<j>(x)dx oscillates

between

finite limits

{or converges).

Then, by applying Dirichlet's or Abel's

test,

we

see that

x
is

convergent and

is

equal to

'

if

a "5

>as x tends to
0,

THUS
and
if
<f>(x)

X
tends to a definite finite limit
,

the last integral has a finite range and a thus we have

finite

integrand;

In the same

way we

can prove that

if

For examples, take 0(.) = coso5 or

sina;.

170, 171

Tin-

foniH-r

in!'

<'\aliiat'-<|

IT

,j,(

>

tends to

M x tend

is

X
\\hi<-h
th.\.-ilu.-

J.
.f

03

by

means

of

l-Yullani's

integral

may
1

be

Tli.'

integral
a
i

f.

.iiml
<>f
1

in

AM.

|ii7

ill

i>

particular case of

formula

i<l

a U..
i.r
1

thai

>n8

and

l.y

Hai'

171.

Uniform convergence of an

infinite integral

i
tin-

lue

!'

^-\

lor

wliidi

tin-

in.-(|iiality

^^
JA
Lfi

<6,
as

(\>$),
.f

function
!:).

w.-ll
iliat

e.

In

-nee

witli

Art.

we Bay
it'

tin-

intf^r.al
.f
//

converges uniformly in

interval (a, 3),

fr
if
t!

all

vahu-s

in tlir intn-val

r-iii,

than
pendent of
interval

a
//.

i'linction
T.iit.
.

A' if),

wliicli

d.-pt-nds

on
tin

but

is

inde-

which

]>art

icuK-n-

vahif //,
f

:.

to be a point of non-uniform convergence


Ex.
1.

tin-

int-^ral.

(if

-"

tin-

intc^i-.-il

i)

unit'
1

kc

-\'u

t.

Ex.
0,

^=
II flirt'
>/

=0.
.rnJ.

is

/via/,
v,.l.
:;:?.

I'.nu.

,,.

11:;.

3e

ilM
*.

t.

Papers, vd.

I.

j..

l.s.

434

INFINITE INTEGRALS.

[AP.

III.

we have usually in practice uniform convergence which is similar to the general test for convergence (Art. 169): The necessary and sufficient condition for the uniform convergence of the integral
But, just as in the case of series,
to introduce a test for
r
I

Ja

f(x, y)dx,

in an interval of valuer of
,

y, is that

we can find

a value of

independent of

y,

such that
e,

f(x,

y)dx

where
small.

'

has any value greater than f

and

is arbitrarily

The only fresh point introduced is seen to be the fact that must be independent of y. The proof that this condition is both necessary and sufficient follows precisely on the lines of Art. 43, with mere verbal alterations. But in practical work we need more special tests which can be applied more quickly the three most useful of these tests are
:

1.

Weierstrass's test.
(a, /3),

Sup2)ose that for all values of y in the interval

the

function

f(x, y) satisfies the condition

where M(x)
the integral

is
I

\f(x,y)\<M(x), a positive function, independent

of y.
I

Then, if

Ja

M(x)dx

converges, the integral

Ja

f(x,

y)dx

is

absolutely and uniformly convergent for all values of y the interval (a, /3).

For then we can choose

independently of

y,

so

that

M(x)dx
Joe
I

is less

than

e;

and therefore

\ J

f(x,

y)dx

< M (x)dx <


f J

f J

M(x)dx
and

<

e.

integra